ISTUDY

Advanced Mechanics of Solids

Build on the foundations of elementary mechanics of materials texts with this modern
textbook on the analysis of stresses and strains in elastic bodies.
Key features include the following.

• Presentation of advanced strength of materials through an integrated framework that

focuses on four key components: computational tools, a step-by-step methodology for
problem solving, treatment of the work–energy concept, and solving advanced strength of
materials problems.
• A force-based finite element method alongside the conventional displacement-based (stiff-
ness) finite element method.
• Detailed description of both uniform and nonuniform torsion problems, including the
nonuniform torsion of members with general cross-sections.
• Consideration of three-dimensional stress, strain, and stress–strain relations in detail with
matrix–vector relations.
• Extensive integration of MATLAB® throughout.
• A complete online teaching package that includes slides, a solutions manual, and
MATLAB® code.
Based on classroom-proven material, this valuable resource provides a unified approach useful
for advanced undergraduate and graduate students, practicing engineers, and researchers.

Lester W. Schmerr Jr. is Emeritus Professor of Aerospace Engineering at Iowa State

University.
Advanced Mechanics of Solids
Analytical and Numerical Solutions with MATLAB®

Lester W. Schmerr, Jr.

Iowa State University
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781108843317
DOI: 10.1017/9781108910132
© Lester W. Schmerr Jr. 2021
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2021
Printed in the United Kingdom by TJ Books Limited, Padstow Cornwall
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Schmerr, Lester W., author.
Title: Advanced mechanics of solids : analytical and numerical solutions with Matlab® / Lester W. Schmerr, Jr.,
Iowa State University.
Description: New York, NY : Cambridge University Press, [2021] | Includes bibliographical references
and index.
Identifiers: LCCN 2020037696 (print) | LCCN 2020037697 (ebook) | ISBN 9781108843317 (hardback) |
ISBN 9781108910132 (epub)
Subjects: LCSH: Engineering mathematics.
Classification: LCC TA350 .S362 2021 (print) | LCC TA350 (ebook) | DDC 620.1/1292–dc23
LC record available at https://lccn.loc.gov/2020037696
LC ebook record available at https://lccn.loc.gov/2020037697

ISBN 978-1-108-84331-7 Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents

Preface page xi

1 Introduction 1
1.1 Axially Loaded Members 1
1.2 Beam Bending 6
1.2.1 Pure Bending 6
1.2.2 Engineering Beam Theory 9
1.3 Torsion of a Circular Shaft 11
1.4 Limitations and Extensions of the Theories 16
1.5 The Foundations of Deformable Body Problems 17
1.6 About This Book 19
1.7 Problems 20
Reference 27

2 Stress 28
2.1 The Stress Vector 28
2.1.1 Stress Vector on an Arbitrary Plane 30
2.2 Normal and Shear Stresses on an Oblique Plane 32
2.2.1 Stress Components on Oblique Planes 33
2.2.2 Stress Transformation Equations 34
2.3 A Relationship between the Normal Stress and Total Shear Stress 41
2.4 Principal Stresses and Principal Stress Directions 42
2.4.1 Special Case when I 3 ¼ 0, Plane Stress 50
2.4.2 Special Case when Principal Stresses are Equal 51
2.5 Mohr’s Circle for Three-Dimensional States of Stress 53
2.6 Stresses on the Octahedral Plane 60
2.7 Stress Notations and the Concept of Stress – A Historical Note 61
2.8 Problems 63
References 68

3 Equilibrium 69
3.1 Equations of Equilibrium – Cartesian Coordinates 69
3.1.1 Plane Stress 72

v
vi Contents

3.2 Strength of Materials Solutions 73

3.3 Force and Moment Equilibrium 76
3.4 Equations of Equilibrium – Cylindrical and Spherical Coordinates 80
3.4.1 Plane Stress 84
3.5 Problems 84

4 Strain 88
4.1 Definitions of Strains 88
4.1.1 Normal Strain 88
4.1.2 Shear Strain 89
4.2 Strain–Displacement Relations and Strain Transformations 91
4.2.1 Dilatation 98
4.2.2 Plane-Strain Principal Strains 98
4.3 Strain Compatibility Equations 99
4.3.1 Plane-Strain and Plane-Stress Compatibility Equations 102
4.3.2 Multiply Connected Bodies 103
4.4 Satisfaction of Compatibility – Strength of Materials Solutions 104
4.5 Strains in Cylindrical Coordinates 112
4.6 Problems 114
References 118

5 Stress–Strain Relations 119

5.1 Linear Elastic Materials 119
5.1.1 Linear Elastic Isotropic Materials 119
5.1.2 General Linear Elastic Stress–Strain Relations 122
5.2 Plane Stress and Plane Strain 129
5.3 Transformation of Elastic Constants 131
5.4 States of Stress and Strain on a Surface from Strain Gage Measurements 139
5.5 Problems 142

6 Governing Equations and Boundary Conditions 146

6.1 Governing Equations in Three Dimensions 146
6.2 Governing Equations in Two Dimensions 150
6.2.1 Plane Stress 150
6.2.2 Plane Strain 154
6.3 Boundary Conditions 158
6.3.1 Saint-Venant’s Principle 160
6.4 Governing Equations in Matrix–Vector Form 163
6.5 Equivalent Algebraic Matrix–Vector Governing Equations for Structures 166
6.5.1 Displacement Formulation 170
6.5.2 Stress (Force) Formulation 172
 Contents vii

6.6 Structural Analysis – A Brief Preview 174

6.7 Problems 176
References 179

7 Analytical Solutions 180

7.1 Displacement-Based Solutions 180
7.1.1 Axisymmetric Solutions in Cylindrical Geometries 180
7.1.2 Thick-Wall Pressure Vessel 183
7.1.3 Shrink-Fits 187
7.2 Airy Stress Function Solutions in Cartesian Coordinates 190
7.2.1 Solutions of the Biharmonic Equation in Cartesian Coordinates 191
7.2.2 A Simply Supported Beam 191
7.3 Airy Stress Function Solutions in Polar Coordinates 197
7.3.1 Michell’s Solutions 197
7.3.2 Stress Concentration at a Circular Hole in a Large Plate 198
7.3.3 Pure Bending of a Curved Beam 202
7.3.4 Comparison with a Curved Beam Strength of Materials Solution 205
7.3.5 Concentrated Force on a Wedge 208
7.3.6 Concentrated Force on a Planar Surface 209
7.3.7 Elastic Bodies in Contact 212
7.4 Stress Singularities 214
7.5 Problems 217
References 223

8 Work–Energy Concepts 224

8.1 Work Concepts 224
8.1.1 Work for a Deformable Body 224
8.2 Work–Strain Energy 226
8.2.1 Linear Elastic Material 227
8.2.2 Isotropic Case 230
8.2.3 Distortional Strain Energy 232
8.3 Complementary Strain Energy 233
8.3.1 Linear Elastic Material 234
8.4 Strain Energy for Strength of Materials Problems 234
8.4.1 Axial Loads 235
8.4.2 Bending 236
8.4.3 Torsion 238
8.5 Principle of Virtual Work and Minimum Potential Energy 239
8.5.1 General Deformable Body 239
viii Contents

8.6 Principle of Complementary Virtual Work and Minimum Complementary

Potential Energy 249
8.6.1 General Deformable Body 249
8.7 Work–Energy Principles and Discrete Forces and Moments 254
8.7.1 Virtual Work and Potential Energy 255
8.7.2 Complementary Virtual Work and Complementary Potential Energy 259
8.7.3 Principle of Least Work 263
8.8 Reciprocity 267
8.9 Problems 270
References 275

9 Computational Mechanics of Deformable Bodies 276

9.1 Numerical Solutions – Axial Loads 277
9.1.1 Principles of Virtual Work and Complementary Virtual Work 277
9.2 Stiffness-Based Finite Elements for Axial-Load Problems 281
9.3 Force-Based Finite Elements for Axial-Load Problems 295
9.3.1 A Summary and Discussion 307
9.4 Generation of the Compatibility Equations 312
9.5 Numerical Solutions – Beam Bending 315
9.5.1 Principles of Virtual Work and Complementary Virtual Work 316
9.6 Stiffness-Based Finite Elements for Beam Bending 320
9.7 Force-Based Finite Elements for Beam Bending 329
9.8 Some Extensions of a Force-Based Approach 337
9.9 Finite Elements for General Deformable Bodies 356
9.9.1 Stiffness-Based Finite Elements 356
9.9.2 Force-Based Finite Elements 358
9.10 The Boundary Element Method 361
9.11 Problems 366
References 371

10 Unsymmetrical Beam Bending 372

10.1 Multiple Axis Bending of Nonsymmetrical Beams 372
10.2 Shear Stresses in Thin, Open Cross-Section Beams 384
10.3 The Shear Center for Thin, Open Cross-Section Beams 388
10.4 Thin, Closed Cross-Section Beams 402
10.5 Problems 404

11 Uniform and Nonuniform Torsion 409

11.1 Torsion of Circular Cross-Sections – A Summary 409
11.2 Uniform Torsion of Noncircular Cross-Sections – Warping Function 411
 Contents ix

11.3 Uniform Torsion of Noncircular Cross-Sections – Prandtl Stress Function 417

11.4 Nonuniform Torsion 426
11.5 General Solution of the Nonuniform Torsion Problem 440
11.6 Torsion of Thin, Open Cross-Sections 442
11.7 Torsion of Thin, Closed Cross-Sections 454
11.8 Problems 467
References 474

12 Combined Deformations 475

12.1 Bending and Torsion of Thin, Open Cross-Sections 475
12.2 Bending and Torsion of Thin, Closed Cross-Sections 477
12.2.1 Single-Cell Cross-Sections 477
12.2.2 Multiple-Cell Cross-Sections 479
12.3 Twisting Induced by Axial Stresses 488
12.4 Problems 494
References 495

13 Material Failure and Stability 496

13.1 Theories of Static Failure 496
13.1.1 Maximum Normal-Stress Theory 496
13.1.2 Maximum Shearing-Stress Theory 497
13.1.3 Maximum Distortional Strain Energy Theory 499
13.2 Fatigue Failure 502
13.3 Fracture Mechanics 508
13.4 Stability 515
13.5 Problems 520
References 525

Appendix A Cross-Section Properties 527

A.1 Parallel Axis Theorem 529
A.2 Area Moments in Rotated Coordinates and Principal Axes 531
A.3 Calculating Centroids and Area Moments in MATLAB® 533
Appendix B The Beltrami–Michell Compatibility Equations 537
B.1 Compatibility Equations for Stresses 537
Appendix C The Sectorial Area Function 540
Appendix D MATLAB® Files 557
Index 565
Preface

Mechanics of deformable solids is a highly developed discipline that is rooted in the desire to
analyze and design safe and reliable structures and components. Engineers first encounter this
subject in a strength of materials course that covers the extension, bending, and torsion of
simple members, and presents the fundamental concepts of stress and strain. Traditionally, an
intermediate or advanced strength of materials course builds on that elementary course and
often introduces a wide variety of topics. These include bending of beams with nonsymme-
trical cross-sections and curved beams, torsion of noncircular shafts, stability of columns,
energy methods, thermal stresses, inelastic behavior, elements of elasticity theory, numerical
methods, bending of thin plates, and the membrane theory of thin shells, elements of fatigue
and fracture analysis, as well as others. Covering such a wide range of topics often makes an
advanced strength text seem like the objective is simply to describe as many of the important
results of the past as possible. Having taught elementary, intermediate, and advanced strength
courses, as well as courses on elasticity, for over 40 years from different texts and notes, I have
developed a deep appreciation of the role that these intermediate and advanced courses play in
providing a sound foundation for present-day engineers. This book, Advanced Mechanics of
Solids, is designed to communicate the course content I have developed to a wider audience
and to enhance that content through a new, integrated framework. There are five key elements
to the approach found in this text.
(1) A detailed description of stress and strain concepts using modern computational tools.
The treatment of stress and strain in introductory strength of materials courses is very
cursory by necessity. These topics are of fundamental importance to understanding the
mechanics of deformable bodies, so a higher-level course should cover these concepts in
more depth. In this book the discussions of stress, strain, and stress–strain relations are
conducted with matrix–vector algebra, and MATLAB® is used to obtain numerical
solutions. One benefit of this approach is that it allows students to treat the stress–strain
relations for anisotropic materials with little additional effort – a topic that is normally left
to elasticity courses or courses on composite materials. In fact, matrices and MATLAB®
are used throughout the book, in contrast to the solution-by-hand emphasis found in most
other texts.
(2) The formulation and solution of problems in terms of the four fundamental principles
(called the four “pillars”) of equilibrium, compatibility, stress–strain relations, and bound-
ary conditions.

xi
xii Preface

In elementary strength of materials courses, the problems considered are based on

physically and experimentally motivated assumptions on the deformations and/or stresses
expected. While this method is very effective for the problems considered in an introduc-
tory course, second-level and advanced courses must provide a more fundamental basis
that is suitable for treating the complex applications found in modern engineering. The
four “pillars” of equilibrium, compatibility, stress–strain relations, and boundary condi-
tions provide this fundamental framework so an entire chapter is devoted to examining the
most common ways that stress analyses are structured around these four principles.
Important canonical problems such as the stress concentration around a hole, deformation
of a thick-walled cylindrical pressure vessel, pure bending of a curved beam, etc., are
solved to illustrate the use of these four pillars. I should note that, unlike some other texts,
I have not used the Einstein summation notation (also called indicial notation) to describe
equations and relations. Although that notation makes expressions more compact and
“elegant” by avoiding multiple summation signs, it adds an extra layer of abstraction that
I have tried to avoid. Matrix–vector notation, however, is freely used throughout the book
because of the frequent use of MATLAB®, and that notation also makes complex relations
more compact.
(3) A detailed treatment of work–energy concepts and their connection to numerical
solutions with either displacement-based methods or force-based methods.
Since the 1960s, computational mechanics has arisen as the primary tool for analyzing
forces and deformations in deformable bodies. As a consequence, most recent advanced
texts have included sections on numerical stress analysis, primarily the finite element
method. Normally, a finite element discussion centers around a stiffness-based approach
using the principle of virtual work. We will also examine this approach. However, we will
also discuss a more recently developed force-based finite element method based on both the
principle of virtual work and the principle of complementary virtual work. Unlike the
stiffness-based finite element method, the force-based method will be shown to be a discrete
model of the four “pillars” of stress analysis, thus tying the finite element method more
directly to traditional analytical analyses. This force-based method is not described in any
current advanced mechanics of materials text. We will also briefly describe a method called
the boundary element method that provides an important alternative numerical approach to
that of finite elements.
(4) The solution of advanced strength of materials problems such as the multiaxis bending
of beams with nonsymmetrical cross-sections and the torsion of noncircular shafts.
Introductory strength of materials courses consider only simple geometries and loadings.
There are, however, important extensions of those elementary analyses possible that provide
a more complete description. For example, one can examine the bending of beams of
nonsymmetrical cross-sections and the torsion of noncircular shafts. The torsion problems
are covered in a new, unified way that emphasizes the distinction between uniform and
nonuniform torsion problems. In particular, the governing equations for the nonuniform
 Preface xiii

torsion of general cross-sections are developed that describe both the primary and secondary
warping deformations and stresses. This formulation, like the force-based finite element
method, is also not found in current texts.
(5) A discussion of theories of failure and instability.
The classical static theories for describing brittle and ductile material failure are examined
as well as some of the key aspects of fatigue failure. The fracture mechanics of cracks and
crack growth are also considered. The role of nondestructive evaluation (NDE) inspections
in evaluating the remaining life of structures is briefly described. Finally, bifurcation
instabilities (buckling) and other forms of instability – such as limit-load instabilities and
snap-through buckling – are examined.
There are, of course, many other topics, as described previously, not covered in this book.
The basic philosophy behind the topics that were chosen is to provide a bridge between an
elementary strength of materials course and higher-level courses such as elasticity, finite
elements, fracture mechanics, advanced structural analysis, and others. To build this bridge,
one must emphasize both the fundamentals and the foundations for numerical analyses. The
book, however, is not only a textbook, as I have included many derivations and background
material that allow the book to be also used as a valuable reference and resource. In a course
setting, these materials can be covered selectively. MATLAB® has been used throughout the
book as a modern computational tool to replace otherwise tedious calculations. I have titled
this book “Advanced Mechanics of Solids” rather than “Advanced Strength of Materials”
to emphasize that it does not follow a traditional path in both content and approach. There
are significant additional materials available with the book including a solutions manual,
PowerPoints of the text figures, PowerPoint lectures, and MATLAB® m-files associated with
both the text (see Appendix D) and the solutions. The solutions manual is available in both
PDF and PowerPoint forms that give an instructor great flexibility in their use. All of these
additional materials are available through the publisher at www.cambridge.org/schmerr.
MATLAB® and Simulink® are registered trademarks of The MathWorks, Inc.
1 Introduction

Mechanics of materials (also called strength of materials) is an area of engineering that

describes the deformation of bodies and the distribution of forces within those bodies.
Introductory courses in this discipline normally consider three types of simple deform-
ations – the stretching or compression of axially loaded bars (Fig. 1.1a), the bending
(flexure) of laterally loaded beams (Fig. 1.1b), and the torsion (twisting) of circular shafts
(Fig. 1.1c). Many complex structures are composed of members that have these simple
types of behavior. Advanced mechanics of deformable solids, the subject of this book, will
examine the behavior of the structures of Fig. 1.1 in more detail and extend those
examinations to more complex configurations. We will describe the fundamentals under-
lying the description of deformation and forces in general, and cover numerical methods
that can be used for obtaining solutions. We will make extensive use of vectors and
matrices in this book, since quantities such as stress and strain are best described in those
terms, and matrix-based numerical methods, such as finite elements, are now in wide-
spread use for solving complex engineering problems. MATLAB ® will be used frequently
in our problem solutions since this software package is specifically designed to efficiently
make matrix–vector calculations.
This chapter will outline the three types of problems seen in Fig. 1.1 to set the stage for our
later discussions and to review the basic assumptions made so that those assumptions can be
later examined and, in some cases, relaxed.

1.1 Axially Loaded Members

Figure 1.2 shows a deformable bar that is being extended by an axial load, P. Although the
bar is shown having a rectangular cross-section, a bar of any constant cross-sectional area
can be considered in exactly the same fashion. As we will see in later chapters, this problem
can be treated by applying basic principles of equilibrium and deformation. Those prin-
ciples lead to partial differential equation boundary value problems, which require a fairly
comprehensive background in applied mathematics. In an introductory mechanics of
materials approach, however, we can learn most of the essentials of this problem by
making some strong assumptions on the deformation and load distributions we expect to
see. In this case, for example, we assume that the primary deformation of the body is a
uniform extension of the body in the x-direction, which is along the axis of the bar.
Consider a bar of length L and cross-sectional area, A, shown in Fig. 1.3a. The bar is
fixed at x ¼ 0 and experiences an x-displacement, Δ, at the end x ¼ L where the load, P, is

1
2 Introduction

Figure 1.1 (a) An axially loaded member, (b) a

laterally loaded beam, and (c) torsion of a
circular shaft.

(a) (b)

(c)

Figure 1.2 An axially loaded member. z

y
P P x

z wu x
ux  dx
ux wx

y
x
dxc

x dx A
L ' dx
(a) (b)

Figure 1.3 (a) The deformation of an axially loaded bar. (b) The local deformation of a small element.

applied. Since the x-displacement, ux , is zero at x ¼ 0 and Δ at x ¼ L, we will assume that

the displacement at any location, x, just varies linearly between those two limits, i.e.,

x
ux ðxÞ ¼ Δ (1. 1)
L
 1.1 Axially Loaded Members 3

It then follows that the displacement gradient, ∂ux =∂x, is just a constant throughout the bar
given by
∂ux Δ
¼ (1.2)
∂x L
[Note: since the displacement here is a function of only the one variable, x, strictly speaking
we should write the displacement gradient as a total derivative, i.e., dux =dx. However, in
more general deformations ux can be a function of all three coordinates ðx; y; zÞ so we will
use the partial derivative here in anticipation of that fact.] To see what this displacement
gradient means in terms of the physical deformation, consider a small element of the bar of
length dx at point x along the initially unloaded bar, as seen in Fig. 1.3a. When the bar is
loaded, the left-hand side of this element experiences a displacement, ux , while the right-
∂ux
hand side, which is located at x þ dx, sees a slightly different displacement, ux þ dx.
∂x
From Fig. 1.3b we see that
∂ux
ux þ dx0 ¼ ux þ dx þ dx (1.3a)
∂x
or
∂ux dx0  dx
¼ (1.3b)
∂x dx
So from Eq. (1.3 b) we see that the displacement gradient is just the elongation of the element
divided by its original length, a quantity that is also called the normal strain in the
x-direction, exx , i.e.,
dx0  dx ∂ux
exx ¼ ¼ (1.4)
dx ∂x
From Eq. (1.2), therefore, it follows that the normal strain is also a constant in the bar.
These results, of course, are a consequence of our assumptions on the deformation
in the bar. We can verify that the normal strain exx ¼ Δ=L by measuring this strain
(at least on the surfaces of the bar) with resistance strain gages. (See Chapter 5 for a
discussion of strain gages.) However, such gages will reveal that there are also normal
strains eyy , ezz in the y- and z-directions, respectively. For a material that is isotropic
and linearly elastic (see generalized Hooke’s law for an isotropic material in
Chapter 5) we have
eyy ¼ ezz ¼ vexx (1.5)
Here v is a material constant called Poisson’s ratio. Thus, an axially loaded bar has a
 
deformation characterized by the three normal strains exx , eyy , ezz and the picture of the
deformations in Fig. 1.3 is actually incorrect and should also show the bar cross-sectional
area reducing in the y- and z-directions (because of the minus sign in Eq. (1.5)) when the bar
is being stretched in the x-direction.
4 Introduction

Figure 1.4 (a) The bar separated into two parts,

P Fx Fx P
showing the internal normal force, F x , acting in
the bar. (b) The force/unit area (stress) acting
across the cross-section, where it is assumed this
stress is uniformly distributed.
x
(a)

P
V xx Fx / A

(b)

Now, let’s turn our attention to the forces acting in the bar. If we imagine passing an
imaginary cutting plane through the bar whose normal is along the x-axis and that separates
the bar into two parts (Fig 1.4a), by equilibrium we see that there must be an internal normal
force, F x , acting on the imaginary cross-section and that we must have F x ¼ P. This result is
true regardless of the x-location of the cutting plane so that the internal normal force must
be a constant in the bar. However, F x is actually the resultant of forces that are distributed
over the entire cross-sectional area A at the cutting plane. Since the bar, according to our
deformation assumptions, has a uniform strain deformation throughout the bar, it reason-
able to assume that the internal force, F x , is also distributed uniformly across the cross-
sectional area. Under that assumption, therefore, we have a uniformly distributed force/unit
area acting across A in the x-direction, called the normal stress, σ xx , where (see Fig. 1.4b)
Fx P
σ xx ¼ ¼ (1.6)
A A
We can describe the stress acting within the bar by cutting out a very small element at any
location and showing the normal stress, σ xx , acting on this element (Fig. 1.5a). Since there is
only one stress acting on the element this is called a uniaxial state of stress.
There is a restriction on how the load P is applied if our assumption of a uniform stress σ xx
distribution is to be valid. This can be seen by relating the stress distribution to the total
internal forces and moments in the bar that σ xx can produce. If we assume that the internal
force, F x ¼ P, acts at a point C in the cross-section (see Fig. 1.5a), then by integrating over
the cross-sectional area A of a cut, as shown in Fig. 1.5b, we must have
ð
F x ¼ σ xx dA ¼ P
A
ð
M y ¼ zσ xx dA ¼ 0 (1.7)
A
ð
M z ¼  yσ xx dA ¼ 0
A
 1.1 Axially Loaded Members 5

z Figure 1.5 (a) The stress, σ xx , acting on an element

within an axially loaded bar. (b) The stress, σ xx ,
y acting on a cross-sectional cut.
P P x
C V xx

(a)
z
y
V xx
y
z
x
C

(b)

 
where F x is the total internal force in the x-direction and M y , M z are the internal moments
about point C that F x can produce in the y- and z-directions (the moments are zero since we
have assumed F x acts through C). We see that the force equation is satisfied by a uniform
stress σ xx ¼ P=A but the moment equations can only be satisfied if
ð ð
ydA ¼ zdA ¼ 0 (1.8)
A A

Equation (1.8) requires that the origin of the (x, y, z)-coordinate system, point C, through which
the applied load P acts, must be the centroid of the cross-sectional area, A. If P acts at a
 
noncentroidal point then moments M y , M z may be nonzero. These moments will produce
nonuniform strains across the cross-section, thus rendering our deformation assumptions invalid.
For a body that deforms in a linear elastic manner the normal stress, σ xx , and the normal
strain, exx , are related through Hooke’s law:
σ xx ¼ Eexx (1.9)

where E is a material constant called Young’s modulus. Relating the stress to the applied load
P and the strain to the elongation Δ of the bar, we then have equivalently the load–
elongation relation for the bar
 
EA
P¼ Δ (1.10)
L

where the quantity K ¼ EA=L acts like a linear spring constant.

6 Introduction

1.2 Beam Bending

1.2.1 Pure Bending
Lateral loads acting on a beam, as shown in Fig. 1.1b, will cause that beam to bend. To
describe the deformation of the beam, let us first examine the special case of so-called pure
bending, where the beam is subjected to applied moments, M, as shown in Fig. 1.6a. In this
case, internally in the beam there is constant internal moment, M y ¼ M, as can be seen from
equilibrium of any section of the beam, as shown in Fig. 1.6b. Because any element of the beam
is subjected to this same moment, we expect all elements to deform identically. Thus, consider
any small element of the beam when the beam is undeformed, as shown in Fig. 1.7a. We will

Figure 1.6 (a) Pure bending of a beam. z

y
(b) The internal moment of the beam,
showing it is a constant throughout the
beam and equal to the applied M
M B x
bending moment.
C
A

(a)

M M M M

x
(b)

Figure 1.7 (a) An undeformed element

of a beam. (b) The deformed element
r z
when a pure bending moment is
applied. These deformations are
greatly exaggerated in order to show B
B
them clearly. M y M

x dw x
dx

neutral axis A
A

undeformed deformed
(a) (b)
 1.2 Beam Bending 7

ux Figure 1.8 The deformed geometry of the beam, showing the x-displacement of the
B cross-sectional plane AB.
dw z
dx
assume that the bending action will cause plane cross-sections of the beam
such as AB to rotate through a small angle ψ about a line in the cross-
section, called the neutral axis, without distortion, i.e., the plane section AB
A is assumed to remain plane after the deformation. This deformation is
shown in Fig. 1.7b. [Note: all the deformations shown in Fig. 1.7 are
exaggerated so that they can be more easily displayed.] We see that the neutral axis itself
will rotate through a small angle dw=dx (since the tangent of a small angle is approximately
equal to the angle itself ), where uz ¼ wðxÞ is the displacement of the neutral axis in the
z-direction. If we also assume that cross-sectional planes initially at right angles to the neutral
axis remain at right angles after the deformation, then we also have ψ ¼ dw=dx. Under these
deformation assumptions the beam will experience a displacement in the x-direction, ux , given
by (see Fig. 1.8):
dw
ux ¼ z (1.11)
dx
This displacement in turn will produce an axial normal strain given by
∂ux d 2w
exx ¼ ¼ z 2 (1.12)
∂x dx
From analytical geometry we know that if wðxÞ describes the z-displacement of the neutral
axis and the slope of the neutral axis, dw=dx, is small then d 2 w=dx2 is just the reciprocal of
the radius of curvature of the neutral axis, ρ (see Fig. 1.7b), i.e.,
d 2w 1
¼ (1.13)
dx2 ρ
In addition to our assumptions on the deformation, we will also assume that the only
significant stress in the beam is the normal stress, σ xx , and, like the case of the axially loaded
bar, this stress will be related to the strain by Hooke’s law, i.e., σ xx ¼ Eexx . Thus,
d 2 w Ez
σ xx ¼ Ez ¼ (1.14)
dx2 ρ
Equations (1.12) and (1.14) show that there is no axial strain or stress at z ¼ 0, which is the
reason that this axis is called the neutral axis. Although there is only a single normal stress
present in the beam, as in the axial load case there will be strains in the y- and z-directions,
which we will write in terms of the axial strain and Poisson’s ratio:
eyy ¼ ezz ¼ vexx (1.15)

Also, as in the axial load case, the stress of Eq. (1.14) must produce the appropriate
internal forces and moments, which gives (see Fig. 1.5b):
8 Introduction

ð
σ xx dA ¼ F x ¼ 0
A
ð
yσ xx dA ¼ M z ¼ 0 (1.16)
A
ð
zσ xx dA ¼ M y ¼ M
A
 
where F x is the internal force in the x-direction and M y , M z are internal moments acting
in the y- and z-directions, respectively. Placing Eq. (1.14) into Eq. (1.16) we find
ð
zdA ¼ 0
A

I yz ¼ 0 (1.17)
EI yy
¼ M y ¼ M
ρ
where
ð
I yz ¼ yzdA
A
ð (1.18)
I yy ¼ z dA2

The quantities I yz , I zz are geometrical properties of the cross-sectional area A called the mixed
second area moment and second area moment, respectively (see Appendix A). The first
condition in Eq. (1.17) says that the neutral axis must go through the centroid of the beam
cross-sectional area, A. The second condition requires that the mixed second area moment,
I yz , must vanish. This can be satisfied as long as either the y- or z-axis is an axis of symmetry
for the cross-section. Finally, the last condition in Eq. (1.17) is called the moment–curvature
relationship since it relates the internal bending moment in the beam to the radius of curvature
of the neutral axis of the deformed beam. Because the bending moment is a constant
throughout the beam, this relationship shows that the radius of curvature is also a constant
so that the shape of the deformed neutral axis is a circular arc. The moment–curvature
relationship can also be written in terms of the z-displacement of the neutral axis as
d 2w My
2
¼ (1.19)
dx EI yy

which, when placed into Eq. (1.14), leads to the bending (flexure) stress expression
M yz Mz
σ xx ¼ ¼ (1.20)
I yy I yy
 1.2 Beam Bending 9

1.2.2 Engineering Beam Theory

When beams are used in practice, they normally support lateral loads (see Fig. 1.9a) rather
than just the applied moment considered previously. Physically, this has two consequences.
First, the internal bending moment in the beam is no longer a constant. This varying internal
moment can be found by applying equilibrium to an arbitrary section of the beam, as shown in
Fig. 1.9b. Second, from the moment–curvature relationship the curvature of the neutral axis
will also not be a constant. In engineering beam theory, it is assumed that at each location x
the beam acts as if it was in pure bending but where the internal moment and curvature are
allowed to vary with the location x in the beam. Thus, the governing engineering beam theory
equations for the displacement, strain, and stress are assumed to be the same as in the pure
bending case but where now M y ¼ M y ðxÞ and 1=ρ ¼ 1=ρðxÞ ¼ d 2 wðxÞ=dx2 , giving
dwðxÞ
ux ðx; zÞ ¼ z
dx
d 2 wð x Þ
exx ðx; zÞ ¼ z , eyy ¼ ezz ¼ vexx
dx2
(1.21)
d 2 wðxÞ M y ðxÞ
2
¼
dx EI yy
M y ðxÞz
σ xx ðx; zÞ ¼ Eexx ¼
I yy

However, as seen in Fig. 1.9b, equilibrium conditions will also require that an internal force
acting in the z-direction, V z ðxÞ, must exist in addition to the internal bending moment,
M y ðxÞ. This force is called a shear force and it will be produced by a distribution of shear
stresses, σ xz , acting tangent to a cross-sectional cut in addition to the bending normal-stress
distribution, as shown in Fig. 1.10b. When we examine a general state of stress in a body in
Chapter 2, we will see that these shear stresses will act on both horizontal and vertical cutting

z Figure 1.9 (a) A simply supported beam carrying

lateral loads. (b) A free-body diagram of a section of
x the beam showing the internal shear force and
bending moments, which are both functions of the
location, x, in the beam.
(a)

Vz ( x )
M y ( x)

x
(b)
10 Introduction

Figure 1.10 (a) The internal shear B B

force and bending moment acting on Vz
an AB cutting plane in a beam and
the corresponding state of stress V xz V xz V xx
present in the beam. (b) The shear- V xx
stress and normal-stress V xz
distributions on plane AB that My neutral axis
produce the shear force and
bending moment.
A A
(a) (b)

Figure 1.11 A small element of a qz ( x )

beam showing the shear force,
bending moment and applied
load/length acting on it.
My
M y + dM y
O
Vz
Vz + dVz
dx

planes, so that a small element within the beam will have the normal and shear stresses
shown in Fig. 1.10a.
The variation of the shear force in the beam is related to the bending moment. We can
obtain this relationship by looking at the equilibrium of a small element of the beam as
shown in Fig. 1.11. The varying shear force and bending moment are shown, as is any
applied lateral load/unit length distributed along the beam. Summing moments about the
center point O of the element, and noting that for a small element the load/unit length is
nearly a constant so gives no net moment about O, we have
X   dx dx
M Oy ¼ M y þ dM y  M y  V z  ðV z þ dV z Þ ¼ 0 (1.22)
2 2
which reduces to
dM y
¼ V z ðxÞ (1.23)
dx
To obtain the shear-stress distribution we can imagine passing a horizontal cutting plane
through a small element at a height z0 above the neutral axis to obtain a section of area Aðz0 Þ
of the beam as shown in Fig. 1.12, where the thickness of the beam at the bottom of this area
is tðz0 Þ. In the x-direction there will be a varying normal stress acting on the front and back
faces of the section as well as an average shear stress, σ xz , acting on the bottom face area tdx.
From equilibrium in the x-direction we find
 1.3 Torsion of a Circular Shaft 11

Figure 1.12 Geometry of a section of the beam for

V xx
z obtaining the shear stress.

V xz dA

A zc V xx  dV xx
dx t zc y
z
zc

ð ð
ðσ xx þ dσ xx ÞdA  σ xx dA  σ xz tdx ¼ 0 (1.24)
Aðz0 Þ Aðz0 Þ

which gives
ð
1 dσ xx
σ xz ¼ dA
tðz0 Þ dx
Aðz0 Þ
ð
1 dM y
¼ zdA (1.25)
I yy tðz0 Þ dx
Aðz0 Þ

V z ðxÞQðz0 Þ
¼
I yy tðz0 Þ

where Qðz0 Þ is a first area moment of the area Aðz0 Þ. This is the formula for the shear stress
usually derived in elementary mechanics of materials texts. For a rectangular cross-section
where the thickness, t, is a constant, it is easy to show that the shear stress has a quadratic
variation in the distance z0 , being zero at the top and bottom of the beam and a maximum at
the neutral axis.

1.3 Torsion of a Circular Shaft

Now consider a circular shaft of radius c and length L that is fixed to a rigid plane at
x ¼ 0 and is subjected to a torque, T 0 , that produces a total twist, ϕ0 , at x ¼ L where the
torque is applied (Fig. 1.13). In the deformation we will assume that cross-sectional planes
whose normals are along the x-axis rotate about the bar axis. As seen in Fig. 1.14a, if two
cross-sectional planes A and B, for example, both rotate through the same amount then the
12 Introduction

Figure 1.13 A circular shaft z

subjected to a torque.

I0

T0 c
L x

Figure 1.14 (a) The case where two cross-sectional

planes A and B of a shaft have the same rotation,
causing the shaft to have a rigid body rotation (no
deformation). (b) The case where the plane at A is A
A B B
fixed and plane B rotates, causing the shaft to twist.

(a) (b)

bar simply has a rigid body rotation and there is no deformation. However, if cross-section
A remains fixed but plane B rotates, then the bar between A and B will experience a twist, as
shown in Fig. 1.14b. Thus, we will assume that the applied torque causes the cross-sectional
planes of the shaft to rotate through a small angle, ϕðxÞ, which varies from plane to plane,
i.e., it is a function of x. For the bar of Fig. 1.13 ϕ varies from zero at the fixed end x ¼ 0 to
the angle (twist) ϕ0 at x ¼ L where the torque T 0 is applied. We will assume this variation is
linear so ϕðxÞ ¼ ϕ0 x=L.
When a cross-sectional plane rotates through a small angle, ϕðxÞ, points in that cross-
section will have a displacement in the θ-direction, uθ , that varies linearly in the radial
distance r, as shown in Fig. 1.15a. Thus, the assumed deformation of the shaft can be
described by the displacements
ux ¼ 0
(1.26)
uθ ¼ rϕðxÞ

We can decompose the displacement uθ into y- and z-components, as seen in Fig. 1.15b, so
we have finally the assumed displacements of the shaft in Cartesian coordinates given by
ux ¼ 0
uy ¼ uθ sin θ ¼ rϕðxÞ sin θ
¼ zϕðxÞ (1.27)
uz ¼ uθ cos θ ¼ rϕðxÞ cos θ
¼ yϕðxÞ
 1.3 Torsion of a Circular Shaft 13

z uT Figure 1.15 (a) The displacement of a cross-section due to

a rotation about the x-axis. (b) The decomposition of that
uT
uz displacement into y- and z-components.

−u y
y
O
T r z
c T
O y

(a) (b)

Figure 1.16 (a) A small element in an

undeformed body whose sides are parallel
to a set of Cartesian axes. (b) When the
T xz body is loaded this element is transformed
into the deformed shape seen, where the
T yz lengths of the sides are changed as are the
dz
angles between the sides.
dy ds y
dsz T xy

dx dsx

undeformed body deformed body

(a) (b)

Before we continue with the torsion problem, we need to say a few words about strains in a
more general setting. Figure 1.16a shows a small element in a body before any loads are
applied so that it is undeformed. When the body is deformed this element will change to the
element shown in Fig. 1.16b. The lengths of the sides of the element will have changed from
 
ðdx; dy; dzÞ to dsx , dsy , dsz and the angles between the sides will have changed from π=2
 
(90) to the angles θxy , θxz , θyz . If these changes are all small, then the element will
 
experience small normal strains exx , eyy , ezz defined by
dsx  dx
exx ¼
dx
dsy  dy
eyy ¼ (1.28)
dy
dsz  dz
ezz ¼
dz
which can be also written in terms of displacement gradients (see Chapter 4)
14 Introduction

∂ux
exx ¼
∂x
∂uy
eyy ¼ (1.29)
∂y
∂uz
ezz ¼
∂z
These normal strains characterize completely the changes of lengths of the small element.
We can also define strains that characterize the small changes in the angles between the sides
of the element. These are called engineering shear strains (γxy , γxz , γyz ) and are given by
π
γxy ¼  θxy
2
π
γxz ¼  θxz (1.30)
2
π
γyz ¼  θyz
2
In Chapter 4 we will see that these strains can also be written in terms of displacement
gradients as
∂ux ∂uy
γxy ¼ þ
∂y ∂x
∂ux ∂uz
γxz ¼ þ (1.31)
∂z ∂x
∂uy ∂uz
γyz ¼ þ
∂z ∂y

From the assumed displacements in our torsion problem, Eq. (1.27), we obtain the strains
exx ¼ eyy ¼ ezz ¼ 0
dϕ
γxy ¼  z
dx
(1.32)
dϕ
yxz ¼ y
dx
γyz ¼ 0
 
We also assume that the shear stresses σ xy , σ xz , σ yz are just proportional to these engineer-
ing shear stresses, so in the shaft we have
σ xy ¼ Gγxy ¼ Gϕ0 z
σ xz ¼ Gγxz ¼ Gϕ0 y (1.33)
σ yz ¼ Gγyz ¼ 0

where ϕ0 ¼ dϕ=dx is the twist per unit length in the shaft and G is the shear modulus.
 1.3 Torsion of a Circular Shaft 15

Figure 1.17 (a) Passing a cutting plane through the shaft,

showing the internal torque, T, acting on a cross-section.
(b) The shear stresses acting on a cross-section.
T0
T T
x (a)
z
y
V xz
V xy y
z

A x

(b)

By equilibrium the internal torque, T, acting across any cross-section of the shaft, as
shown in Fig. 1.17a, must be a constant equal to the applied torque, T 0 . This internal torque
is produced by the shear stresses acting on the cross-section (Fig. 1.17b) so that
ð
 
T ¼ yσ xz  zσ xy dA
A
ð
0
  (1.34)
¼ Gϕ y2 þ z2 dA
A
0
¼ Gϕ J

where
ð ð
 
J¼ y2 þ z2 dA ¼ r2 dA (1.35)
A A

is the polar area moment of the cross-sectional area, A. Since T ¼ T 0 is a constant through-
out the shaft and the twist/unit length, ϕ0 ¼ ϕ0 =L, it follows from Eq. (1.34) that
ϕ0 ¼ T 0 L=GJ (1.36)

which is the moment–twist relation for the bar. Using Eq. (1.34) in Eq. (1.33) it follows that
the shear stresses can also be written as
Tz
σ xy ¼
J
(1.37)
Ty
σ xz ¼
J
16 Introduction

Figure 1.18 Combining the shear z

stresses acting on a cross-section
to form the total shear stress, τ. W
V xz
T
V xy
c r z y
T
O y

The total shear stress, τ, in the cross-section (Fig. 1.18) acts perpendicular to a radial
direction whose origin is at the center of the circular cross-section and is given by
τ ¼ σ xy sin θ þ σ xz cos θ
z y
¼ σ xy þ σ xz
r r (1.38)
Tr
¼
J
which is a familiar result seen in elementary mechanics of materials texts and shows that the
maximum total shear stress occurs on the outer boundary of the cross-section where r ¼ c.

1.4 Limitations and Extensions of the Theories

The elementary theories we have presented of axial extension, bending, and torsion have
varying degrees of restrictions on the geometries and loads involved. The axial-load case has
some of the fewest restrictions since the bar can have any cross-sectional area and still have a
uniform stress distribution σ xx ¼ P=A acting throughout the bar (as long as the applied load
acts through the centroid of the cross-section as the theory requires). The actual distribution
of the applied loads at the ends of the bar, however, can locally produce highly nonuniform
stresses near those ends. Thus, a bar must be long enough for the elementary theory to be
valid over the most of the bar length. The distribution of the end axial loads can also have
significant effects on a thin structure, producing twisting of the structure in additional to an
axial extension. This will cause variations of the normal stress in the cross-section along the
entire length of the structure and introduce a new load quantity call a bimoment. We will
discuss this case in Chapter 12. For the elementary theory to be valid, strictly speaking the
bar must have a constant cross-section, but the theory can still be applied when the cross-
sectional area is slowly varying and the theory can also accommodate having axial loads
 1.5 The Foundations of Deformable Body Problems 17

distributed along the length of the bar. In both cases the stress will no longer be a constant,
i.e., we will have σ xx ðxÞ ¼ F x ðxÞ=AðxÞ. Such extensions of the theory are often found in
elementary mechanics of materials texts.
The elementary theory of beam bending requires that the mixed second area moment
I yz ¼ 0 so that theory is typically valid only for beams having symmetrical cross-sections.
If a bending moment is applied about the y-axis to a beam with an unsymmetrical cross-
section, bending deformations will typically be produced about both the y- and z-axes, not
just about the y-axis as the elementary theory assumes. Thus, by generalizing the assumed
displacements to account for these multiaxis bending effects, one can develop a beam
theory that describes the normal stresses and deflections of an unsymmetrical beam (see
Chapter 10). The shear-stress formula, σ xz ¼ V z Q=I yy t, is more problematic since – except
for cross-sections with vertical sides, such as rectangular cross-sections – this shear – stress
distribution does not properly satisfy the boundary conditions (see Chapter 6) for the
cross-section. However, for general thin cross-sections (which can be symmetrical or
unsymmetrical), it is possible to generalize the shear stress formula (which is often written
in terms of a shear flow (force/length) rather than the shear stress) for those thin sections
(see Chapter 10).
The theory of torsion presented in Section 1.3 is valid only for shafts having solid or
hollow circular cross-sections. This strong geometry restriction is essential since, if we apply
a torque to a noncircular shaft, out-of-plane (axial) warping displacements will be present,
rendering our displacement assumptions invalid. The elementary theory, however, can be
modified to include warping effects using either a warping function approach or a Prandtl
stress function approach (Chapter 11). In either of those approaches one must solve partial
differential equation boundary value problems so that computer-based solutions are often
employed. For cross-sections consisting of thin closed or open sections, more analytically
based formulas can be developed, as shown in Chapter 11.

1.5 The Foundations of Deformable Body Problems

As can be seen from the discussions in the previous sections, the usual approach to problems
in elementary mechanics of materials is to make some strong assumptions about the
displacement deformations in a particular case and to relate those displacements to the
corresponding local deformations as described by the strains. The stress–strain relations of a
body then give the corresponding stresses. When we relate the stresses to the applied forces
or moments, we then have an explicit relationship between loads and deformations that we
can use to describe how a structure will deform and to evaluate the magnitude of the stresses
produced by the loads. This elementary approach requires that each new problem be based a
different set of assumptions, and the approach becomes more difficult to apply as the
geometries and loadings of the structure become more complex. In this book we will
continue to apply this elementary approach to extensions of the problems found in elemen-
tary texts. However, we will also emphasize a set of basic principles that are based on a
18 Introduction

Figure 1.19 The four foundational

“pillars” that underlie all deformable
body problems.
equilibrium compatibility

boundary stress–strain
conditions relations

general theory of deformable bodies (also called the theory of elasticity) and provide the
foundation for solving even the most difficult problems. These four principles we have called
the four pillars of all stress analyses (Fig. 1.19). First, there is equilibrium. As we will see in
Chapter 3, stresses must locally satisfy equilibrium equations at every point within a body.
In Chapter 3 we will examine whether or not the stresses seen in the elementary solutions of
this chapter satisfy local equilibrium and, if not, what must be modified in the analysis.
Second, compatibility conditions are likewise equations that the strains must satisfy locally
at every point in the body. As shown in Chapter 4, these conditions guarantee that the
strains will produce physically meaningful displacements of the body, and we will examine in
that chapter if the elementary strength of materials solutions satisfy the compatibility
relations. In analyzing more complex problems such as the torsion of noncircular shafts,
for example, we will see that compatibility plays a significant role in obtaining a solution.
Stress, which is a concept discussed extensively in Chapter 2, is a quantity that characterizes
how forces are distributed within a body. Similarly, strains, as described in detail in
Chapter 4, are measures of the deformation distributions in the body. A third pillar,
stress–strain relations, therefore, is a key element that relates these internal force and
deformation distributions. In Chapter 5 we will examine stress–strain relations for both
isotropic and anisotropic elastic materials. Finally, the fourth pillar involves the conditions
that must be satisfied on either displacements or stresses (or combinations of stresses and
displacements) at the boundaries of a body. These boundary conditions are essentially a
detailed description of how a body is supported or loaded. It is difficult in practice to know
precisely the boundary conditions on the displacements and/or stresses so that boundary
conditions are one pillar of an analysis that may be difficult to specify. Fortunately, as
discussed in Chapter 6, Saint-Venant’s principle gives us a practical way to often relax the
boundary conditions so that one can deal with any uncertainty that is present.
A solution that satisfies the four pillars of local stress equilibrium, strain compatibility
equations, the stress–strain relations, and boundary conditions is an exact solution to a
deformable body problem. In most cases one can find exact, analytical solutions to only a
few relatively simple problems. In Chapter 7 we will obtain some important exact solutions.
 1.6 About This Book 19

We will also examine numerical solutions (see Chapter 9) that can allow us to solve much
more complex problems using methods such as the finite element method. We will see that
energy methods (Chapter 8) are often used to replace the pillars of equilibrium and com-
patibility when formulating such numerical solutions.

1.6 About This Book

Mechanics of materials is a highly developed area of engineering that has a long and rich
history with many important developments. In an advanced-level book on this subject,
therefore, one has a wide range of topics one can choose to cover. I have chosen not to
follow the path of many “Advanced Strength” or “Advanced Mechanics of Materials” texts
that provide a very broad coverage of many topics. Instead, I have structured this book as a
bridge between elementary strength of materials and higher-level courses where the text
stresses the fundamentals underlying the field. Thus, Chapters 2 through 6 cover in some
detail the four foundational principles described in the previous section. Chapter 2 gives an
in-depth look at the concept of stress while Chapter 4 covers the concept of strain and the
important associated concept of compatibility. Both of these concepts are at the heart of
how we describe the distributions of forces and deformations in a body. Chapter 3 covers the
local equations of equilibrium and, as mentioned previously, reexamines the elementary
strength of materials problems described in this chapter in light of those equilibrium
conditions. Chapter 5 examines stress–strain relations, including the relations for anisotropic
materials such as composite materials. In all these introductory chapters, MATLAB® is used
extensively to solve problems since this is a modern engineering tool that is a natural
successor to slide rule and calculator solutions that formed the basis of previous generations
of such texts. All of these foundational principles are combined in Chapter 6 to describe the
governing equations and boundary conditions that must be satisfied to solve deformable
body problems. As might be expected, there are different choices one can make in these sets
of governing equations. Those choices have important consequences, so in Chapter 6 we
describe both displacement-based and stress-based approaches as well as the use of auxiliary
functions such as the Airy stress function to formulate problems. Although there are
relatively few exact solutions to the governing equations and boundary conditions of
deformable bodies, there are some important “canonical” problems that are solvable, such
as the thick wall pressure vessel, stress concentration around a hole, and others that can help
to show the behavior of the stresses and strains explicitly. Thus, Chapter 7 is devoted to
examining a selection of those important problems. In Chapter 8 we consider work–energy
concepts – concepts that underlie many of the approaches used for solving problems
numerically. It is shown how the principles of virtual work and complementary virtual work
can serve as replacements for the equilibrium and compatibility equations, respectively,
and so provide an alternative path for formulating and solving problems. In Chapter 9
these work–energy principles are shown to be the natural foundations for finite element
methods, which are the computational tools most frequently used to solve complex
20 Introduction

engineering problems. Problems such as axial-load problems and beam-bending problems

are described in Chapter 9 so that all of the fundamental aspects of finite elements can be
described in a simple context. A unique aspect of Chapter 9 is that it covers both
displacement-based (stiffness) finite elements as well as a force-based finite element method.
While the stiffness-based finite element approach is the method normally described, force-
based finite elements is an important alternative that has not been adequately covered in
texts at any level and is in many respects closer to the fundamental principles of this field
(equilibrium, compatibility, etc.) so that it deserves to have a wider audience. Another
alternative to stiffness-based finite elements as a computational tool is the boundary element
method, so Chapter 9 provides a brief overview of that important method as well. Three
chapters of the book, Chapters 10 through 12, reexamine the problems of bending and
torsion and axial loads for cases that cannot be treated with the assumptions found in
elementary strength of materials. These include the bending of beams with nonsymmetrical
cross-sections, the torsion of noncircular shafts, and twisting induced by axial loads. The
final chapter, Chapter13, examines various types of material failure and instabilities.

1.7 PROBLEMS

P1.1 In elementary strength of materials courses, one learns how to obtain shear force
distributions, V z ðxÞ, and bending moment distributions, M y ðxÞ, in a beam. Normally,
this is done by using force and moment equilibrium, considering each section of
the beam where the free-body diagram is different. The resulting shear force and
bending moment functions are then described by functions that change from section
to section in the beam. It would be nice, however, if one could express the shear
force and bending moment in terms of functions that are valid for the entire
beam. Such functions are called singularity functions. For example, consider a long
beam which has a concentrated force, P, acting at x ¼ a in the beam as shown in
Fig. P1.1a. If we examine equilibrium of the beam at a section taken before and after
the load, as shown in Fig. P1.1b, then the shear force and bending moment can be
written as
(
0 x<a
V z ðxÞ ¼
P x>a
( (P1.1)
0 x<a
M y ðxÞ ¼
Pðx  aÞ x > a

But we can express these functions in terms of singularity functions as

V z ðxÞ ¼ Phx  ai0

(P1.2)
M y ðxÞ ¼ Phx  ai1
 1.7 Problems 21

P Figure P1.1 (a) A concentrated

force on an infinitely long beam.
(b) The shear force and bending
moments in the beam both
x=a before and after where the force
is applied.
(a)
P
x−a
My = 0
M y = P ( x − a)
Vz = 0
Vz = P
for x < a for x > a
(b)

where the singularity functions are defined as


n 0 x<a
hx  ai ¼ n0 (P1.3)
ðx  aÞn x>a

Note that singularity functions can be differentiated and integrated like ordinary func-
tions since
d
hx  ain ¼ nhx  ain1 n1
dx
ð (P1.4)
n hx  ainþ1
hx  ai dx ¼ þC n0
nþ1

(a) Determine the shear force V z ðxÞ and bending moment M y ðxÞ for the load distri-
butions shown in Fig. P1.2 in terms of singularity functions and use the appropriate
singularity functions to describe the loaded beam shown in Fig. P1.3. Sketch the
shear force and bending moment distributions.
(b) Write a MATLAB® function that implements these singularity functions. The
most commonly occurring first four of these distributions are shown in Fig. P1.4.
Call this function s_func. The general MATLAB® function call should be
y ¼ s_func(x, a, n);

where x is a vector of position values that span the length of the entire beam, a is
the x-location where the singularity function begins, and n is an integer that is
greater than or equal to zero that defines the power of the function. Note that for
n ¼ 0 the function has a discontinuity at x ¼ a so let this function ¼ 1.0 for x  a
and zero for x < 0. The other functions are all continuous at x ¼ a.
22 Introduction

Figure P1.2 Load distributions w (lb/ length)

C (ft-lb)
for (a) a concentrated couple,
(b) a concentrated force, (c) a
constant force/unit length that
starts at x ¼ a, and (d) a linearly x=a x=a
increasing force distribution (a) (c)
that starts at x ¼ a.
s (lb/ length2)
P (lb)
1

x=a x=a

(b) (d)

Figure P1.3 A beam in equilibrium under the z 20 lb/ft

given loads.

x
10 ft 10 ft
50 lb 150 lb

Figure P1.4 Four singularity functions, which are

(a) a step function, (b) a linear function, (c) a x −a
0
2
quadratic function, and (d) a cubic function, all 1.0 x −a
starting at x ¼ a and zero for x < a.
x=a x=a
(a) (c)
1
x −a 3
x −a

x=a x=a

(b) (d)

Use these singularity functions to express the shear force and bending moment in
MATLAB® for the beam problem of part (a) and plot the shear force and bending
moment distributions for the entire beam.
P1.2 Consider a beam with the rather complex load distribution shown in Fig. P1.5. Note
that the origin of the x-coordinate is not at the left-hand end of the beam. Also note
 1.7 Problems 23

z Figure P1.5 A beam in equilibrium under the given loads.

5000 lb 4000 lb/ft
1000 lb/ft
9000 ft-lb
x

3' 3' 3' 3' 3'

8000 lb 9000 lb

that the singularity functions of Fig. P1.2 for distributed loads start at x ¼ a but they
never stop, so distributions such as the 1000 lb/ft load in Fig. P1.5 must be described by
the superposition of several singularity functions.
(a) Use the singularity functions obtained in problem P1.1(a) to express the shear force
and bending moment for the entire beam.
(b) Express the shear force and bending moment in terms of the MATLAB® singular-
ity function developed in problem P1.1(b) and plot the shear force and bending
moment for the entire beam.
Note that once we have the shear force, V z , we can get the bending moment, M y ,
directly from integration since
ð
M y ðxÞ ¼ V z ðxÞdx þ C (P1.5)

where C is a constant of integration. We can use Eq. (P1.5) directly since from
Eq. (P1.4) we know how to integrate the singularity functions. The net constant
term after the integrations of all the singularity functions are done will be zero if
the bending moment is zero for x-values less than the starting point for the
beam (which is x ¼ 3 in the present example) and this is always the case so we
can set C ¼ 0.
(c) Show that your results for the bending moment of part (a) are consistent with using
Eq. (P1.5) to integrate the shear force expression.
You may wonder why these functions are called singularity functions, because all the
functions we have defined do not have any singular behavior. But if, for example, we
differentiate the function hx  ai0 we will get a singular function called a delta function
that is zero everywhere except x ¼ a where the delta function is infinite. Further
differentiations will produce even more singular functions [1]. Such singular functions
arise if we try to express concentrated loads or concentrated couples as special types of
distributed loads but when we work directly with the shear force and bending moment
expressions, as done here, there is no need to introduce truly singular functions since
the shear force and bending moment are always relatively well-behaved functions
(although they have “jumps” at concentrated forces or moments).
24 Introduction

P1.3 Since we know how to integrate singularity functions (see problem P1.2), if we obtain
the bending moment in terms of singularity functions, we can then integrate the
moment–curvature relationship of Eq. (1.19) given by

d 2w My
2
¼ (P1.6)
dx EI yy
to obtain the beam deflection. There will be constants of integration in this process,
which can be found from the displacement or rotation boundary conditions. For the
beam shown in Fig. P1.6:
(a) Obtain the reaction forces at the supports A and B and write the bending moment
for the beam in terms of singularity functions, using the results of problem P1.1.
(b) Integrate the moment–curvature relationship of Eq. (P1.6) to obtain the deflection,
wðxÞ, in terms of singularity functions. Plot the deflection of the beam, using the
MATLAB® function of problem P1.1.
P1.4 A cantilever beam (Fig. P1.7a) is a simple structure often considered in elementary
strength of materials texts. Consider the case when there is a small gap, Δ, between the

Figure P1.6 A simply supported beam. 1320 lb

200 lb/ft

A B

4 ft 6 ft 5 ft 5 ft

Figure P1.7 (a) A cantilever beam q0

where there is a small gap, Δ,
between the beam and a rigid
floor. (b) The case where the D L
distributed load is large enough to
cause a portion of the beam to be
(a)
in contact with the floor.
dw
=0
dx

d
L–d R
(b)
 1.7 Problems 25

beam and a rigid floor, where the distributed load,q0 , is large enough to close the gap.
As a consequence, a reaction force, R, exists at the left side of the beam.
(a) Determine an expression for the reaction force, R, in terms of Δ and q0 . Similarly,
determine an expression for the slope of the beam at the reaction force.
(b) What is the smallest value of q0 , q0  q1 , that will just close the gap? What is the
value of q0 , q0  q2 , that will cause the slope of the beam at the reaction force to go
to zero?
(c) The slope of the beam at the floor cannot be less than zero since we have assumed
that the floor is rigid. Thus, for q0 > q2 a portion of the beam must come in contact
with the floor, as shown schematically in Fig. P1.7b. In this case, determine
the reaction force, R, in terms of Δ and q0 . Plot the reaction force, R, versus q0 from
q0 ¼ 0 to values q0 > q2 . What is the difference between the cases when q0 < q2 and
q0 > q2 ?
Let Young’s modulus of the beam be E and let the area moment I yy ¼ I .
P1.5 The moment–curvature relationship for beams (Eq. (1.19)) allows us to obtain the
beam deflection from a knowledge of the internal bending moment. This relationship
also allows us to solve statically indeterminate problems where the internal moment
cannot be obtained from equilibrium alone. Consider, for example, the beam shown in
Fig. P1.8 where the beam is loaded by a constant distributed load/unit length, q0 , over
its entire length, L, and is rigidly fixed at both ends A and B. Use the moment–
curvature relationship to obtain the reactions at A and B.
Problems P1.6 to P1.9 are axial-load problems (see Fig. P1.9) where you are asked to obtain
analytical solutions for the displacement and internal force and plot your results for some specific
problem values. These problems are also used in Chapter 9, where you are asked to obtain these
solutions numerically via a number of different finite element methods. Thus, these analytical
solutions can serve as a check on those numerical finite element results.

P1.6 The bar shown in Fig. P1.9a is fixed between two rigid walls and carries a uniform load
qx ¼ q0 . The cross-sectional area of the bar is A, and its Young’s modulus is E. Determine
the internal axial force F x ðxÞ acting in the bar and its displacement, ux ðxÞ. Plot the
displacement and force for q0 ¼ 10 N=mm, E ¼ 104 N=mm2 , A ¼ 9 mm2 , L ¼ 3000 mm.
P1.7 The bar shown in Fig. P1.9b is fixed between two rigid walls. A concentrated axial
force, P, acts at its center. The cross-sectional area of the bar is A, and its Young’s
modulus is E. Determine the internal axial force F x ðxÞ acting in the bar and its displace-
ment, ux ðxÞ. Plot the displacement and force for P ¼ 20 000 N, E ¼ 104 N=mm2 ,
A ¼ 9 mm2 , L ¼ 3000 mm.

q0

Figure P1.8 A statically indeterminate beam problem where the beam is rigidly
A B
L fixed at both supports A and B.
26 Introduction

Figure P1.9 (a) A bar fixed at both ends x

qx = q0 x
with a uniform distributed force qx ¼ q0
P
(force/unit length) acting over its entire
length. (b) A bar fixed at both ends with a
concentrated force, P, at its center. (c) L L/2 L/2
A bar fixed at both ends with a uniform
(a) (b)
distributed force qx ¼ q0 (force/unit x
length) acting over its right half. (d) A bar x qx = q0 qx = q0
x
L
fixed at both ends with a linearly
increasing distributed force qx ¼ q0 x=L
(force/unit length) acting over its
L/2 L/2 L
entire length.
(c) (d)

P1.8 The bar shown in Fig. P1.9c is fixed between two rigid walls. A constant distributed
axial force, q0 , (force/unit length) acts over its right half. The cross-sectional area of
the bar is A, and its Young’s modulus is E. Determine the internal axial force F x ðxÞ
acting in the bar and its displacement, ux ðxÞ. Plot the displacement and force for
q0 ¼ 10 N=mm, E ¼ 104 N=mm2 , A ¼ 9 mm2 , L ¼ 3000 mm.
P1.9 The bar shown in Fig. P1.9d is fixed between two rigid walls. A linearly varying
distributed force, qx ¼ q0 x=L (force/unit length) acts over its entire length. The cross-
sectional area of the bar is A, and its Young’s modulus is E. Determine the internal
axial force F x ðxÞ acting in the bar and its displacement, ux ðxÞ. Plot the displacement
and force for q0 ¼ 10 N=mm, E ¼ 104 N=mm2 , A ¼ 9 mm2 , L ¼ 3000 mm.
Problems P1.10 to P1.13 are beam problems (see Fig. P1.10) where you are asked to obtain
analytical solutions for the displacement and internal shear force and bending moment and
plot your results for some specific problem values. These problems are also used in Chapter 9,
where you are asked to obtain these solutions numerically via a number of different finite
element methods. Thus, these analytical solutions can serve as a check on those numerical finite
element results.

P1.10 The beam shown in Fig. P1.10a is fixed at both ends. A force, P, acts at its middle. The
Young’s modulus of the beam is E and its area moment about the y-axis is I yy ¼ I .
Determine the internal moment, M ðxÞ ¼ M y ðxÞ, and shear force, V z ðxÞ, acting in the
beam and the vertical deflection, wðxÞ, of the neutral axis. Plot the moment, shear force,
and displacement in the beam for P ¼ 20,000 N, E ¼ 2  105 N=mm2 ð200 GPaÞ,
I ¼ 3  106 mm4 , L ¼ 3000 mm.
P1.11 The beam shown in Fig. P1.10b is fixed at both ends. A constant distributed force, q0 ,
acts over its entire length. The Young’s modulus of the beam is E and its area moment
about the y-axis is I yy ¼ I . Determine the internal moment, M ðxÞ ¼ M y ðxÞ, and
shear force, V z ðxÞ, acting in the beam and the vertical deflection, wðxÞ, of the neutral
axis. Plot the moment, shear force, and displacement in the beam for q10 ¼ 10 N=mm,
E ¼ 2  105 N=mm2 ð200 GPaÞ, I ¼ 3  106 mm4 , L ¼ 3000 mm.
 Reference 27

z z Figure P1.10 (a) A beam fixed at both

ends and acted upon by a concentrated
P q0
force, P, at its center. (b) A beam fixed
at both ends and acted upon by a
x constant distributed force, q0 , over its
x
length. (c) A beam fixed at both ends
L/2 L/2 L and acted upon by a constant
(a) (b) distributed force, q0 , over its right half.
z
z (d) A concentrated force, P, acting on a
q0 P beam fixed at its left end and sitting on a
roller at its right end.

x x

L/2 L/2
L/2 L/2
(c) (d)

P1.12 The beam shown in Fig. 1.10c is fixed at both ends and carries a constant distributed
force, q0 , over its right half. The Young’s modulus of the beam is E and its area
moment about the y-axis is I yy ¼ I . Determine the internal moment, M ðxÞ ¼ M y ðxÞ,
and shear force, V z ðxÞ, acting in the beam and the vertical deflection, wðxÞ, of the
neutral axis. Plot the moment, shear force, and displacement in the beam for
q0 ¼ 10 N=mm, E ¼ 2  105 N=mm2 ð200 GPaÞ, I ¼ 3  106 mm4 , L ¼ 3000 mm.
P1.13 The beam shown in Fig. 1.10d is fixed at the left end and sits on a roller at the right end.
It carries a concentrated force, P, at its center. The Young’s modulus of the beam is E
and its area moment about the y-axis is I yy ¼ I . Determine the internal moment,
M ðxÞ ¼ M y ðxÞ, and shear force, V z ðxÞ, acting in the beam and the vertical deflection,
wðxÞ, of the neutral axis. Plot the moment, shear force, and displacement in the beam for
P ¼ 20 000 N, E ¼ 2  105 N=mm2 ð200 GPaÞ, I ¼ 3  106 mm4 , L ¼ 3000 mm.

Reference
1. E. Volterra and J. H Gaines, Advanced Strength of Materials (Englewood Cliffs NJ: Prentice-Hall,
1971)
2 Stress

Applied forces acting on a body affect both the internal and external deformation of that
body. The internal deformations in particular depend on how the forces are distributed
throughout the body. Stress is a key concept that gives us a way to characterize those
internal force distributions. This chapter will discuss in depth the stress concept.

2.1 The Stress Vector

The basic quantity that we use to describe internal force distributions in a deformable body
is called the stress vector or, equivalently, the traction vector. To obtain the stress vector, we
imagine that we pass a cutting plane through a point P in a body that separates it into two
pieces as seen in Fig. 2.1. The unit normal to this cutting plane (for one of the pieces) is called
n, as shown. Let the internal force, acting over a small element of area, ΔA, at P be ΔF. Then
the stress vector, TðnÞ ðPÞ, at point P is defined as

ΔF
TðnÞ ðPÞ ¼ lim (2.1)
ΔA!0 ΔA

which gives a measure of the force/unit area acting at point P. The superscript (n) is present
in the definition since the stress vector depends on the orientation of this imaginary cutting
plane as well as location of the point P. For economy of notation, however, we will
normally omit showing the explicit dependency of the stress vector on point P and write
it simply as TðnÞ . Since the internal forces always occur in equal and opposite pairs, it
follows that

TðnÞ ¼ TðnÞ (2.2)

Because the stress vector depends on the orientation (n) of the cutting plane through the
point P it might appear that, in order to use this concept, we will need to know values of the
stress vector for an infinite number of possible cutting planes through point P. However, as
we will see shortly, this is not the case and to completely determine the stress vector on any
cutting plane we need only to determine the stress vectors
on three orthogonal  cutting
ðex Þ
planes. Figure 2.2 shows, for example, the stress vectors T , T , T ð ey Þ ðez Þ
acting on the
 
three cutting planes whose unit normals are along the ðx; y; zÞ axes, where ex , ey , ez are unit
vectors for this Cartesian coordinate system (the cutting planes are actually all taken
through point P but they are shown as being the front faces of the small cube seen in

28
 2.1 The Stress Vector 29

Figure 2.1 The separation of a body into two pieces by an

imaginary cutting plane, showing the internal force, ΔF,
−DF acting at a point P over a small area, ΔA, of the left-hand
DA piece, as well as the equal and opposite internal force
P acting on the other piece.
n
−n P
DA DF

y (e y )
T Figure 2.2 The stress vectors acting on three cutting planes at P that are
along a set of Cartesian (x, y, z) axes.

T(
ex )
ey Fig. 2.2 (for ease of display). The Cartesian components of these
ex three stress vectors are what we call the stresses at point P with
ez
P x respect to the ðx; y; zÞaxes. For example, σ xx , σ xy , σ xz are the
components of the stress vector, Tðex Þ , so that we have
z
Tðex Þ ¼ σ xx ex þ σ xy ey þ σ xz ez (2.3)
T(
ez )

These stresses are shown in Fig. 2.3. If we similarly express the

other stress vectors acting along the y- and z-planes into their components, we have
altogether

Tðex Þ ¼ σ xx ex þ σ xy ey þ σ xz ez
Tðey Þ ¼ σ yx ex þ σ yy ey þ σ yz ez (2.4)
Tðez Þ ¼ σ zx ex þ σ zy ey þ σ zz ez

which, letting ðx; y; zÞ ! ð1, 2, 3Þ, we can write more compactly as

y X
3
Tðei Þ ¼ σ ij ej ði ¼ 1; 2, 3Þ (2.5)
j¼1

V xy These nine stresses form up what is called the state of stress

V xz at any point in a body (Fig. 2.4). The directions of the stresses
V xx x
z
Figure 2.3 The Cartesian stress components acting on a cutting plane with a
normal along the x-axis.
30 Stress

y Figure 2.4 Complete state of stress at a point in a deformable body.

V yy
 
V yz σ xx , σ yy , σ zz are normal to the planes on which they act so they are
V yx  
V xy called normal stresses, while the stresses σ xy , σ xz , σ yx , σ yz , σ zx , σ zy
V zy are called shear stresses. There are actually only six independ-
V zx V xx
V xz ent stresses since it can be shown that the shear stresses are
symmetric, i.e.,
V zz
x
σ xy ¼ σ yx
σ yz ¼ σ zy (2.6)
z σ xz ¼ σ zx

This result is a direct consequence of the moment equation for a small element such as the
one shown in Fig. 2.4 but we will not derive Eq. (2.6) here (see problem P2.1). An easy way
to remember the meaning of the subscripts associated with these stresses is to note that the
first subscript indicates the direction of the normal to the cutting plane on which the stress
acts while the second subscript indicates the direction of the stress on the cutting plane.
Thus, σ yz , for example, is a stress that acts on a cutting plane, the normal of which is in the y-
direction and the stress acts in the z-direction on this plane (see Fig. 2.4).

2.1.1 Stress Vector on an Arbitrary Plane

If we know the stress vectors acting on the ðx; y; zÞ planes (and their corresponding stress
components) at point P then we can find the stress vector at P for any other cutting plane as
well. Thus, the state of stress at a point completely determines stresses on all planes at that
point. To see this, consider cutting out a small tetrahedron at point P whose back faces are
parallel to the x-, y-, and z-axes and whose front face is along an oblique plane whose unit
normal is n (Fig. 2.5).
The area of the oblique face of the tetrahedron is ΔAn and the
 
areas of the back faces are ΔAx , ΔAy , ΔAz , as shown in Fig. 2.5.
y
From Newton’s second law we have
DAz
−T(
ez )

TðnÞ ΔAn  Tðex Þ ΔAx  Tðey Þ ΔAy  Tðez Þ ΔAz þ ρgΔV ¼ ρaΔV
(n )
−T (ex ) T (2.7)
n

ey DAn where ρ is the density (mass/unit volume) of the body, a is the

ex acceleration (if the body is in motion), and ΔV is the volume of the
DAx
x
ez P
DAy
Figure 2.5 Stress vectors acting on the faces of a small tetrahedron at a point
(e y )
z −T P in a body.
 2.1 The Stress Vector 31

y

Figure 2.6 The Cartesian components T ðxnÞ ; T ðynÞ ; T ðznÞ of the stress vector TðnÞ .

Ty(n )
tetrahedron. However, one can show (see problem P2.2) that the
ratios of the areas of the faces of the tetrahedron are just the compon-
T( n )
ents of the unit normal, n, i.e.,
Tx(
n) x
Tz(n )
ΔAx ΔAy ΔAz
nx ¼ , ny ¼ , nz ¼ (2.8)
z ΔAn ΔAn ΔAn
so that we can rewrite Eq. (2.7) as
ΔV ΔV
TðnÞ  Tðex Þ nx  Tðey Þ ny  Tðez Þ nz þ ρg ¼ ρa (2.9)
ΔAn ΔAn
Taking the limit as we shrink the tetrahedron to the point P and using the fact that volume
goes to zero faster than the area of the oblique face, we find

TðnÞ ¼ Tðex Þ nx þ Tðey Þ ny þ Tðez Þ nz (2.10)


which shows that indeed a knowledge of the three stress vectors Tðex Þ , Tðey Þ , Tðez Þ or, equiva-
lently, the state of stress in terms of the stress components given previously allows us to determine
the stress vector for any cutting plane taken through the body at point P. If we decompose the
stress vector TðnÞ into its three components along the (x, y, z) axes (see Fig. 2.6), i.e.,

TðnÞ ¼ T ðxnÞ ex , T ðynÞ ey , T ðznÞ ez (2.11)

and use Eq. (2.4) together with Eq. (2.10), then equating components we find

T ðxnÞ ¼ σ xx nx þ σ yx ny þ σ zx nz
T ðynÞ ¼ σ xy nx þ σ yy ny þ σ zy nz (2.12)

T ðznÞ ¼ σ xz nx þ σ yz ny þ σ zz nz

or, more compactly,

ðnÞ
X
3
Ti ¼ σ ji nj ði ¼ 1; 2, 3Þ (2.13)
j¼1

We can also write Eq. (2.13) in vector–matrix notation by writing the stress vector and unit
normal as column vectors and the state of stress as a 3  3 matrix:
8 ðnÞ 9 8 9 2 3
>
> T1 > > > n1 > σ 11 σ 12 σ 13
n o < > >
= >
< = >
6 7
TðnÞ ¼ T ðnÞ ¼ T ð2nÞ , n ¼ fng ¼ n2 , ½σ  ¼ 6 4 σ 21 σ 22 σ 23 7
5 (2.14)
>
> >
> >
> >
>
>
: ðnÞ ; > : ;
T n3 σ 31 σ 32 σ 33
3
32 Stress

which gives
n o
T ðnÞ ¼ ½σ T fng (2.15)

where ½σ T is the transpose of the stress matrix ½σ . However, since the shear stresses are
symmetric ½σ T ¼ ½σ  so that we can also write Eq. (2.15) as
n o
T ðnÞ ¼ ½σ fng (2.16)

2.2 Normal and Shear Stresses on an Oblique Plane

We can always break the stress vector on an arbitrary cutting plane into two components –
one acting along the unit normal, n, to the plane, and the other component in the plane on
which the stress vector acts. We will call these two components the normal stress, σ nn , and the
total shear stress, τ s , respectively, and we will let s be a unit vector that defines the direction
of the total shear-stress component in the plane (see Fig. 2.7). Since σ nn is just the Cartesian
component of TðnÞ in the n direction we have

σ nn ¼ TðnÞ n
(2.17)
¼ T ðxnÞ nx þ T ðynÞ ny þ T ðznÞ nz

From the expressions for the stress vector components in terms of the stresses, Eq. (2.12), we
find
σ nn ¼ σ xx n2x þ σ yy n2y þ σ zz n2z
(2.18)
þ 2σ xy nx ny þ 2σ xz nx nz þ 2σ yz ny nz

or, in matrix–vector notation

σ nn ¼ fngT ½σ fng (2.19)

To obtain the total shear stress we note that its magnitude can be found since
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jτ s j ¼ jT ðnÞ j2  σ 2nn
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi



 (2.20)
ðnÞ
2 ðnÞ
2 ðnÞ
2
T (n )
n ¼ Tx þ Ty þ Tz  σ 2nn

To find the direction of the total shear stress we let

s nn
s ¼ sx ex þ sy ey þ sz ez (2.21)
s
ts
Figure 2.7 The stress vector, TðnÞ , acting on an arbitrary cutting plane and its
normal-stress component, σ nn , and total shear-stress component, τ s .
 2.2 Normal Stress and Shear Stresses on an Oblique Plane 33

and note that

jτ s js ¼ TðnÞ  σ nn n (2.22)
 
which, in component form, leads to three equations that can be solved for sx , sy , sz

jτ s jsx ¼ T ðxnÞ  σ nn nx
jτ s jsy ¼ T ðynÞ  σ nn ny (2.23)
jτ s jsz ¼ T ðznÞ  σ nn nz

2.2.1 Stress Components on Oblique Planes

The results of the previous section gave us the equations needed to solve for the normal stress
and total shear stress on a general oblique cutting plane. In some cases, however, we need to
know instead the individual shear-stress components acting on an oblique plane along two
mutually orthogonal directions in that plane that we will call t and v (see Fig. 2.8). These
shear stress components are σ nt , σ nv . If we let t and v be unit vectors in the t and v directions,
respectively, then we can write them in terms of their Cartesian components as
t ¼ tx ex þ ty ey þ tz ez
(2.24)
v ¼ vx ex þ vy ey þ vz ez

Just as we obtained the normal stress from the stress vector, Eq. (2.17), we can obtain these
shear stress components:

σ nt ¼ TðnÞ t
(2.25)
σ nv ¼ TðnÞ v

or, in component form

σ nt ¼ T ðxnÞ tx þ T ðynÞ ty þ T ðznÞ tz

n (2.26)
σ nv ¼ T ðxnÞ vx þ T ðynÞ vy þ T ðznÞ vz
t
V nt which can be expressed in terms of the stresses as
Ws
σ nt ¼ σ xx nx tx þ σ yy ny ty þ σ zz nz tz
V nv  
þ σ xy ny tx þ nx ty
y (2.27a)
þ σ xz ðnz tx þ nx tz Þ
 
ey v þ σ yz nz ty þ ny tz

ex x
ez
z Figure 2.8 Stress components acting on a general oblique plane.
34 Stress

and
σ nv ¼ σ xx nx vx þ σ yy ny vy þ σ zz nz vz
 
þ σ xy ny vx þ nx vy
(2.27b)
þ σ xz ðnz vx þ nx vz Þ
 
þ σ yz nz vy þ ny vz

Both of these equations can be written more compactly in the same form as we obtained for
σ nn . Summarizing all these results we have
X
3 X
3
σ nn ¼ σ ij ni nj
i¼1 j¼1

X
3 X
3
σ nt ¼ σ ij ni tj (2.28)
i¼1 j¼1

X
3 X
3
σ nv ¼ σ ij ni vj
i¼1 j¼1

or, in matrix–vector form

σ nn ¼ fngT ½σ fng

σ nt ¼ fngT ½σ ftg (2.29)

σ nv ¼ fngT ½σ fvg

2.2.2 Stress Transformation Equations

We have found the results for the stress components acting on a single plane whose unit
normal was n. However, since n is arbitrary, we can use those results, appropriately
modified, to obtain the stresses acting along three mutually orthogonal oblique planes in
terms of the stresses acting along the ðx; y; zÞ planes. These relationships are called the
stress transformation equations. To obtain the stress transformation equations, we note
that we can identify n, t, and v as unit vectors along three mutually orthogonal axes
 0 0 0
x1 ; x2 ; x3 that are rotated relative to the ðx; y; zÞ axis, as shown in Fig. 2.9. Note also that
all these unit vectors can play dual roles as either unit normal vectors to various planes as
well as unit tangent vectors within a given plane. Thus, for example, n can be a unit normal
to the cutting plane whose normal is along the n-axis but it can also play the role as a unit
vector in the n-direction that is tangent to the planes where the normal is t or v. If we let
 0 0 0  
e1 ; e2 ; e3 be unit vectors along the x01 ; x02 ; x03 axes and ðe1 , e2 , e3 Þ unit vectors along the
ðx1 , x2 , x3 Þ, respectively, then we have
 2.2 Normal Stress and Shear Stresses on an Oblique Plane 35

y, x2 t , x2c Figure 2.9 (a) The stress acting at a

point along the Cartesian axes
c
V 22 n, x1c ðx1 , x2 , x3 Þ. (b) The stresses at the
V 22 c
V 21 V 12c same point acting along the rotated
V 21 c
V 23 V 11c Cartesian axes ðx1 , x2 , x3 Þ.
V 23 V 12 V 32c V 13c
V 32 V 31c
V 11 x, x1
V 31 V 13 V 33c
V 33

z , x3 v, x3c

(a) (b)

e01 ¼ n ¼ n1 e1 þ n2 e2 þ n3 e3
e02 ¼ t ¼ t1 e1 þ t2 e2 þ t3 e3 (2.30)
e01 ¼ n ¼ v1 e1 þ v2 e2 þ v3 e3

It is more convenient to write these components in a more uniform way, namely

ni ¼ l i1 , ti ¼ l i2 , vi ¼ l i3 ði ¼ 1; 2, 3Þ (2.31)

in terms of the nine quantities l ij ði ¼ 1; 2, 3Þ, ðj ¼ 1; 2, 3Þ . Then Eq. (2.30) can be written as
X
3
e0i ¼ l ki ek ði ¼ 1; 2, 3Þ (2.32)
k¼1

From Eq. (2.32) we see that

X
3   X3
e0i ej ¼ l ki ek ej ¼ l ki δkj ¼ l ji (2.33)
k¼1 k¼1

where ek ej ¼ δkj since the unit vectors are orthogonal to each other, where δkj is called the
Kronecker delta, whose definition can be written as

1 for i ¼ j
δij ¼ (2.34)
0 otherwise

If we interchange the i and j subscript labels in Eq. (2.33) and use the properties of the dot
product it follows from Eq. (2.33) that



l ij ¼ ei e0j ¼ cos xi ; x0j (2.35)

i.e., l ij is the cosine of the angle between the xi axis and the x0j axis. Thus, the [l] matrix is
called a direction cosine matrix. Note that, from Eq. (2.31), the n, t, and v unit vectors are
36 Stress

just the columns of this direction cosine matrix. Thus, we can write the [l] matrix in the two
equivalent forms
2 3 2 3
l 11 l 12 l 13 n1 t1 v1
½l  ¼ 4 l 21 l 22 l 23 5 ¼ 4 n2 t2 v2 5 (2.36)
l 31 l 32 l 33 n3 t3 v3

Now, consider the stress components ðσ nn , σ nt , σ nv Þ we obtained previously in Eq. (2.29).

These are just the stresses on the x01 ¼ constant plane so we have
X
3 X
3
σ 011 ¼ σ nn ¼ l i1 l j1 σ ij
i¼1 j¼1

X
3 X
3
σ 012 ¼ σ nt ¼ l i1 l j2 σ ij (2.37a)
i¼1 j¼1

3 X
X 3
σ 013 ¼ σ nv ¼ l i1 l j3 σ ij
i¼1 j¼1

However, if the normal to the cutting plane was in the x02 direction instead we would have
X
3 X
3
σ 022 ¼ σ tt ¼ l i2 l j2 σ ij
i¼1 j¼1

3 X
X 3
σ 021 ¼ σ tn ¼ l i2 l j1 σ ij (2.37b)
i¼1 j¼1

3 X
X 3
σ 023 ¼ σ tv ¼ l i2 l j3 σ ij
i¼1 j¼1

and, similarly, for a cutting plane with a normal in the x03 direction
X
3 X
3
σ 033 ¼ σ vv ¼ l i3 l j3 σ ij
i¼1 j¼1

X
3 X
3
σ 031 ¼ σ vn ¼ l i3 l j1 σ ij (2.37c)
i¼1 j¼1

X
3 X
3
σ 032 ¼ σ vt ¼ l i3 l j2 σ ij
i¼1 j¼1

Because the results in Eqs. (2.37a, b, c) are in a uniform notation, we can write them all
together as
X
3 X
3
σ 0mn ¼ l im l jn σ ij ðm ¼ 1; 2, 3Þ, ðn ¼ 1; 2, 3Þ (2.38)
i¼1 j¼1
 2.2 Normal Stress and Shear Stresses on an Oblique Plane 37

or, in matrix form, as

½σ 0  ¼ ½l T ½σ ½l  (2.39)

where
2 3 2 3
σ 011 σ 012 σ 013 σ 11 σ 12 σ 13
6 0 7 6 7
½σ 0  ¼ 6
4 σ 21 σ 022 σ 023 7
5, ½σ  ¼ 6
4 σ 21 σ 22 σ 23 7
5 (2.40)
σ 031 σ 032 σ 033 σ 31 σ 32 σ 33
 
are the stress states associated with the x01 ; x02 ; x03 and ðx1 , x2 , x3 Þ axes, respectively.
Equation (2.39), therefore, is the general transformation of the state of stress from
 
the ðx1 , x2 , x3 Þ axes to the x01 ; x02 ; x03 axes. It is easy to show that we can also write
the stresses with respect to the ðx1 , x2 , x3 Þ axes in terms of the stresses for the
 0 0 0
x1 ; x2 ; x3 axes as

½σ  ¼ ½l ½σ 0 ½l T (2.41)

since the direction cosine matrix ½l  is an orthogonal matrix which satisfies

½l ½l T ¼ ½l T ½l  ¼ ½I  (2.42)

It is called a proper orthogonal matrix if its determinant is equal to one, i.e.,

det ð½l Þ ¼ 1 (2.43)

If we dot both sides of Eq. (2.32) with a general position vector, x, we have
X
3
xe0i ¼ l ki xek ði ¼ 1; 2, 3Þ (2.44)
k¼1

or, equivalently, in component or matrix–vector form

X
3
x0i ¼ l ki xk
k¼1 (2.45)
fx0 g ¼ ½l T fxg

Note that some authors define the direction cosine matrix components instead as
   
Qij ¼ l ji ¼ e0i ej ¼ cos x0i ; xj (2.46)

Thus, ½Q ¼ ½l T and we have

X
3
x0i ¼ Qik xk
k¼1 (2.47)
fx0 g ¼ ½Qfxg
38 Stress

You should be careful to determine which definition of the direction cosine matrix is being
used when encountering expressions for transforming position vectors, stresses, etc. If we
take the transpose of Eq. (2.36), we see that the n, t, and v unit vectors are the rows of the ½Q
direction cosine matrix.
 
If the coordinates x01 ; x02 ; x03 and ðx1 , x2 , x3 Þ are both right-handed coordinates, then
both ½l  and its transpose will be proper orthogonal matrices. It is important to only use
right-handed coordinates so that there are no reflections of coordinates present that can have
unintended effects such as, for example, changing the directions of vector cross products
from their usual definitions. A right-handed coordinate system ðx1 , x2 , x3 Þ with unit vectors
ðe1 , e2 , e3 Þ along those coordinates will satisfy e3 ¼ e1  e2 .

Example 2.1 The Stress Vector, Stresses and Stress Transformations

As stated before, a knowledge of the state of stress at a point allows us to calculate the stress
vector on any cutting plane and stress components in any direction at that point. As seen in the
previous sections, most stress and stress-vector relations can be written in a matrix–vector
form. Thus, MATLAB®, which is designed specifically to handle such matrix–vector manipu-
lations, is an ideal software package for performing all the necessary algebra. In this example
we will use MATLAB® to explore a number of the matrix–vector relationships associated with
stresses and stress vectors. We will begin by forming a general three-dimensional (3-D) state of
stress with respect to a set of ðx; y; zÞ axes as a 3  3 matrix in MATLAB®:
stress = [ 20 40 30; 40 -10 5; 30 5 20]
stress = 20 40 30
40 -10 5
30 5 20

where the stress values are assumed to be given in MPa. Note that the stress matrix is
symmetrical, as it should be since the shear stresses in the off-diagonal terms across the main
diagonal of the matrix must be equal. If we are given this state of stress, we can calculate the
stress vector on any cutting plane with Eq. (2.16). Thus, consider defining a cutting plane
with a unit normal n, which we will define in MATLAB®:
n = [3; -4; 5]/sqrt(9+16+25)
n = 0.4243
-0.5657
0.7071

Then Eq. (2.16) in MATLAB® form is

Tn = stress*n
Tn = 7.0711
26.1630
24.0416
 2.2 Normal Stress and Shear Stresses on an Oblique Plane 39


which are just the stress-vector Cartesian components T ðxnÞ ; T ðynÞ ; T ðznÞ written as a column
vector. If we want to find the normal stress on this plane, we can use Eq. (2.17) where we
compute the component of the stress vector in the normal direction via a dot product:
snn = Tn’*n
snn = 5.2000

so the normal stress is 5.2 MPa. [Note: there is also a built-in MATLAB® function dot we
can use instead.] Alternatively, as shown with Eq. (2.19), we can calculate this normal stress
directly from the state of stress:
snn = n’*stress*n
snn = 5.2000

If we want to find the total shear stress, tots, on the plane then from Eq. (2.20) we have
tots = sqrt(norm(Tn)^2 -snn^2)
tots = 35.8533

and the total shear-stress direction, s, is from Eq. (2.22) given by

s = (Tn-snn*n)/tots
s = 0.1357
0.8118
0.5680

We can verify that in this direction on the plane we obtain the total shear stress by
calculating the component of the stress vector in the s direction, namely
tots2 = Tn’*s
tots2 = 35.8533

which agrees with our previous total shear-stress result.

To examine the full stress transformation equations, we need to have three mutually
orthogonal unit vectors ðn; t; vÞ that form up a right-handed set of coordinate axes. We
already have a vector, n, so let’s define a vector, t, that is orthogonal to n by forming the
cross vector of n with the unit vector, ez , in the z-direction and then dividing by the
magnitude of that cross product to get a unit vector:
t = cross(n, [0;0;1])
t = -0.5657
-0.4243
0
t = t/norm(t)
t = -0.8000
-0.6000
0
40 Stress

Then we can define a third unit vector, v, that is orthogonal to both n and t by again using
the cross product:
v = cross(n, t)
v = 0.4243
-0.5657
-0.7071

We said that we needed a set of mutually orthogonal directions ðn; t; vÞ that are right-
handed. The unit vectors ðn; t; vÞ form a right-handed system if v ¼ n  t. However, we
obtained the vector v via precisely this relation so we do have a right-handed system along
the ðn; t; vÞ coordinate directions.
We now are in a position to calculate many quantities associated with these coordinates.
For example, we can use Eq. (2.15) to write the stress vector associated with the t = constant
and v = constant planes by simply replacing the n vector in that equation with either t or v, i.e.,
n o
T ðtÞ ¼ ½σ ftg
n o
T ðvÞ ¼ ½σ fvg

which in MATLAB® gives

Tt = stress*t
Tt = -40
-26
-27
Tv = stress*v
Tv = -35.3553
19.0919
-4.2426

To obtain the full state of stress with respect to the ðn; t; vÞcoordinates, we need the direction
cosine matrix formed up with n, t, v unit vectors in its columns (see Eq. (2.36)):
l = [n t v]
l = 0.4243 -0.8000 0.4243
-0.5657 -0.6000 -0.5657
0.7071 0 -0.7071

Then the state of stress associated with those coordinates, ntv_stresses, is given as (see Eq. (2.39))
ntv_stresses = l’*stress*l
ntv_stresses = 5.2000 -21.3546 -28.8000
-21.3546 47.6000 16.8291
-28.8000 16.8291 -22.8000
 2.3 Relationship between Normal Stress and Total Shear Stress 41

2.3 A Relationship between the Normal Stress and Total Shear Stress
The stress vector has some interesting properties that can be used to develop a relationship
between normal stresses and total shear stresses. For example, let nð1Þ and nð2Þ be unit
normals for two different planes passing through the same point. Then the stress vectors
associated with those planes at that point satisfy

Tðn Þ nð2Þ ¼ Tðn Þ nð1Þ

ð1Þ ð2Þ
(2.48)

i.e., the component in the nð2Þ direction of the stress vector acting on the plane whose
normal is nð1Þ is equal to the component in the nð1Þ direction of the stress vector acting on
the plane whose normal is nð2Þ . This is easily shown by writing both sides in Eq. (2.48) in
terms of the stress state components associated with the ðx; y; zÞ axes, giving the same form
in both cases:

Tðn Þ nð2Þ ¼ Tðn Þ nð1Þ

ð1Þ ð2Þ

ð1Þ ð2Þ ð1Þ ð2Þ

¼ σ xx nx nx þ σ yy ny ny þ σ zz nðz1Þ nðz2Þ


ð1Þ ð2Þ ð1Þ ð2Þ
þ σ xy nx ny þ ny nx
(2.49)


ð1Þ ð2Þ
þ σ xz nx nðz2Þ þ nðz1Þ nx


ð1Þ ð2Þ
þ σ yz ny nðz2Þ þ nðz1Þ ny

We can use Eq. (2.48) as follows. Let nð1Þ and sð1Þ be unit vectors that define the normal to
a plane on which the normal stress, σ nn , acts and a tangential direction in that plane
along which the total shear stress, τ s , acts (Fig. 2.10a). Now, consider rotating by a
 
very small angle, dθ, in the plane defined by nð1Þ , sð1Þ to a new plane whose normal is
nð2Þ and tangential direction sð2Þ (Fig. 2.10b). Since the angle dθ is small the normal stress
and total shear stress on the rotated plane will be to first order σ nn þ ð∂σ nn =∂θÞdθ and
   
τ s þ ð∂τ s =∂θÞdθ, as seen in Fig. 2.10b. Expressing nð2Þ , sð2Þ in terms of nð1Þ , sð1Þ to first
order we have

Figure 2.10 (a) The normal stress and

n
2
n
1 total shear stress on an arbitrary cutting
wW s plane. (b) The normal stress and total
dT Ws  dT shear stress on a cutting plane that is
V nn wT
wV nn s
2 rotated by a small angle dθ with respect
V nn  dT
Ws wT to that original plane.
dT
s 1

(a) (b)
42 Stress

nð2Þ ¼  sin ðdθÞsð1Þ þ cos ðdθÞnð1Þ

ffi dθsð1Þ þ nð1Þ
(2.50)
sð2Þ ¼ cos ðdθÞsð1Þ þ sin ðdθÞnð1Þ
ffi sð1Þ þ dθnð1Þ

so that, again to first order

  
Tðn Þ nð2Þ ffi τ s sð1Þ þ σ nn nð1Þ  dθsð1Þ þ nð1Þ
ð1Þ

(2.51)
¼ τ s dθ þ σ nn

and, similarly,

∂σ nn  
Tðn Þ nð1Þ ffi
ð2Þ
σ nn þ dθ dθsð1Þ þ nð1Þ nð1Þ
∂θ

∂τ s  
þ τs þ dθ sð1Þ þ dθnð1Þ nð1Þ (2.52)
∂θ
∂σ nn
¼ σ nn þ dθ þ τ s dθ
∂θ
Equating the expressions in Eq. (2.51) and Eq. (2.52) and dividing by dθ we find
∂σ nn
¼ 2τ s (2.53)
∂θ
Thus, whenever the normal stress on a plane is an extremum we have ∂σ nn =∂θ ¼ 0 and Eq.
(2.53) shows that on any such plane we must have τ s ¼ 0 also. Planes of extreme normal
stress are called principal planes so we have shown that principal planes must also be free of
shear stress. In the next section we will see how we can use this fact to determine the explicit
values of the extreme normal stresses.

2.4 Principal Stresses and Principal Stress Directions

In the previous sections we have learned how to find the stress vector or stress components
acting on any cutting plane. Often, however, we are interested in the stresses acting on those
particular cutting planes where the normal stresses take their extreme values since those
stresses play an important role in material failure theories (see Chapter 13). These extreme
normal stresses are called principal stresses and the planes they act on are called the principal
planes. The unit normals to the principal planes are similarly called the principal directions.
As shown in the previous section, on the principal planes there are no shear stresses so one of
the easiest ways to obtain the principal planes is to simply find the planes on which the stress
vector consists of a purely normal stress, i.e.,

TðnÞ ¼ σ p n (2.54)
 2.4 Principal Stresses and Principal Stress Directions 43

where σ p is the unknown (at this stage) principal stress on a principal plane. Using the
expression for the stress vector in terms of the stress components, Eq. (2.13), and the
 
symmetry of the shear stresses σ ij ¼ σ ji , we have

ðnÞ
X
3
Ti ¼ σ ij nj ¼ σ p ni ði ¼ 1; 2, 3Þ (2.55)
j¼1

Expanding these three conditions out, we find a homogeneous system of equations for the
 
unit normal components nx , ny , nz :
 
σ xx  σ p nx þ σ xy ny þ σ xz nz ¼ 0
 
σ yx nx þ σ yy  σ p ny þ σ yz nz ¼ 0 (2.56)
 
σ zx nx þ σ zy ny þ σ zz  σ p nz ¼ 0

The only way that such a homogeneous system can have a nontrivial solution is for the
3  3 determinant of the coefficients to be zero, i.e.,
 
σ xx  σ p σ xy σ xz
 
σ yx σ yy  σ p σ yz ¼0 (2.57)
 
σ zx σ zy σ zz  σ p

When expanded out this condition leads to a cubic equation which has three real
 
roots that are the three principal stresses σ p1 , σ p2 , σ p3 . This cubic equation can be
written as
σ 3p  I 1 σ 2p þ I 2 σ p  I 3 ¼ 0 (2.58)

where ðI 1 , I 2 , I 3 Þ are stress invariants given by

I 1 ¼ σ xx þ σ yy þ σ zz
I 2 ¼ σ xx σ yy þ σ xx σ zz þ σ yy σ zz  σ 2xy  σ 2xz  σ 2yz (2.59)
I 3 ¼ σ xx σ yy σ zz þ 2σ xy σ xz σ yz  σ xx σ 2zy  σ yy σ 2xz  σ zz σ 2xy

The last two invariants can also be written in a more compact form using determinants as

σ xx σ xy σ yy σ yz σ xx σ xz
I2 ¼ þ þ
σ xy σ yy σ yz σ zz σ xz σ zz
σ xx σ xy σ xz (2.60)
I 3 ¼ σ xy σ yy σ yz
σ xz σ yz σ zz

These quantities are called invariants because they can be shown to have the same values
when expressed in terms of stresses as measured in any coordinate system. Thus, in particu-
lar, in terms of the principal stresses we must have
44 Stress

I 1 ¼ σ p1 þ σ p2 þ σ p3
I 2 ¼ σ p1 σ p2 þ σ p1 σ p3 þ σ p2 σ p3 (2.61)
I 3 ¼ σ p1 σ p2 σ p3
Once the principal stresses are obtained, we can place one of those principal stresses into
 
Eq. (2.56) and try to solve for the corresponding principal direction as defined by nx , ny , nz .
However, because the determinant of the homogeneous system of Eq. (2.56) is zero, this
means that the three equations are not all independent. Thus, the most we can do is, say,
solve two of these equations for two of the unit normal components in terms of the other
third component. However, we can then find this third component (to within a plus or
minus sign) by using the fact that the components of the unit normal must be components
of a unit vector. We can repeat this whole process for each of the other principal stresses.
Thus, a procedure for finding the principal stresses and principal directions is: (1) find the
 
three principal stresses σ p1 , σ p2 , σ p3 as roots of the cubic equation of Eq. (2.58), and (2) for
each of those principal stresses, determine the corresponding principal direction vectors
 ð1Þ ð2Þ ð3Þ 
n , n , n , using the fact that these normals must be unit vectors whose components
with respect to the ðx; y; zÞ axis satisfy

2
2
2
nðx1Þ þ nðy1Þ þ nðz1Þ ¼ 1

2
2
2
nðx2Þ þ nðy2Þ þ nðz2Þ ¼ 1 (2.62)

2
2
2
nðx3Þ þ nðy3Þ þ nðz3Þ ¼ 1

This procedure, however, is rather tedious and must be modified in some special cases, as we
will see shortly. Solving the cubic equation of Eq. (2.58) can be done in a number of ways.
Many calculators and various software packages have cubic equation solvers. Also, there is
an explicit solution for the roots of a cubic. However, the easiest way to solve for the
principal stresses and directions when using software packages such as MATLAB®,
MathCad®, Mathematica®, Maple®, etc., is to recognize that this is a matrix eigenvalue
problem. If we write Eq. (2.55) and Eq. (2.62) in matrix–vector form, then we have the
problem of finding the solution of the equations
½σ fng ¼ σ p fng
(2.63)
fngT fng ¼ 1

The σ p are called the eigenvalues of the matrix ½σ , which here is just the matrix defining the
state of stress with respect to the ðx; y; zÞ axes. The corresponding solutions for the normal-
ized vectors fng are called the eigenvectors. For a real, symmetrical matrix ½σ  it can be
shown that the eigenvalues are real and the eigenvectors are orthogonal to each other. Thus,
the three principal directions will form three unit vectors of a coordinate system whose axes
are along the principal directions. However, as we will see, the axes of these principal
 2.4 Principal Stresses and Principal Stress Directions 45

coordinates may not always form a right-handed system, so we may need to choose the
eigenvectors so that this right-handedness is satisfied.

Example 2.2 Eigenvalue Problem Solution for Principal Stresses and Directions
Let us write a state of stress in MATLAB® by first defining a 3  3 state of stress matrix,
stress, as
stress = [ 3 5 10; 5 6 4; 10 4 1]
stress = 3 5 10
5 6 4
10 4 1

where the stresses are measured in MPa. We can obtain the principal stresses in a matrix
called pvals and the principal directions in a matrix pdirs by simply providing the stress
matrix as an argument to the built-in MATLAB® function eig which solves the eigenvalue
problem of Eq. (2.63):
[pdirs, pvals] = eig(stress)
pdirs = -0.6782 0.3747 0.6322
0.0323 -0.8442 0.5350
0.7341 0.3833 0.5605
pvals = -8.0625 0 0
0 1.9648 0
0 0 16.0977

The three principal stresses are approximately σ p1 ¼ 8:06 MPa, σ p2 ¼ 1:96 MPa, and
σ p3 ¼ 16:10 MPa. The matrix pvals is actually the state of stress associated with coordinates
along the principal directions since the shear stresses are zero, as we have already shown.
   
The matrix pdirs contain the three unit vectors nð1Þ , nð2Þ , nð3Þ associated with σ p1 , σ p2 , σ p3 ,
respectively, in the columns of the pdirs matrix, i.e., it has the elements
2 ð2Þ ð3Þ
3
nðx1Þ nx nx
6 7
pdirs ¼ 6
4 ny
ð1Þ ð2Þ
ny ny 7
ð3Þ
5 (2.64)
nðz1Þ nðz2Þ nðz3Þ

These three unit vectors form the unit vectors of a coordinate system along the principal
 
directions, which we will call ðp1 , p2 , p3 Þ just as ex , ey , ez are unit vectors along the ðx; y; zÞ
axes. Are the ðp1 , p2 , p3 Þ axes a right-handed system? We can check to see by extracting the
unit vectors from the pdirs matrix:
n1 = pdirs(:,1);
n2 = pdirs(:,2);
n3 = pdirs(:,3);
46 Stress

(where the MATLAB® vectors nj ¼ nðj Þ ðj ¼ 1; 2, 3Þ) and perform the cross product with the
built-in MATLAB® function cross, as we have done before. We should find nð3Þ ¼ nð1Þ  nð2Þ :
cross(n1, n2)
ans = 0.6322
0.5350
0.5605

which, when we examine the pdirs matrix shows that we do obtain the MATLAB® vector
n3  nð3Þ .
[Note: if e is an eigenvector then so is – e. The MATLAB® eig function does not always
guarantee that the signs chosen for the three eigenvectors form a right-handed system so it
is important to check and, if necessary, to change the sign on an eigenvector or reorder
the numbering on the eigenvectors so that the eigenvectors do form a right-handed
system. A simpler check than the one given above is to compute the determinant of pdirs
with the MATLAB® function det. A right-handed system must have det(pdirs) return a
value of þ1.]
In this case the eigenvectors do form a right-handed system so we can view the pdirs
matrix as the matrix of direction cosines that transforms the stresses along a right-handed
ðx; y; zÞ axes to the stresses along a set of right-handed principal ðp1 , p2 , p3 Þ axes. We can
verify this by letting the MATLAB® matrix pdirs = [l] in Eq. (2.39) and computing that
equation in MATLAB® as:
Stress p = pdirs’*stress*pdirs
Stress p = -8.0625 -0.0000 0.0000
-0.0000 1.9648 0.0000
0.0000 0.0000 16.0977

which shows that the MATLAB® stress_p matrix contains the same stresses along the
principal directions that was given by the MATLAB® matrix pvals.

Using the MATLAB® eig function makes the calculation of principal stresses and
principal directions very easy. Other software packages can solve eigenvalue problems but
you should check to make sure that the eigenvectors are normalized by the package so that
they are unit vectors (the MATLAB® eig function does do this normalization) and that they
form up a right-handed system.

Example 2.3 Direct Solution for Principal Stresses and Principal Directions
Although an eigenvalue solver such as eig is a very effective way to find principal stress and
principal stresses, we do not see the details of the solution process. In this example we will
examine those details with the more direct approach we outlined previously where we first
 2.4 Principal Stresses and Principal Stress Directions 47

find the roots of a cubic equation and then solve for the principal directions explicitly.
Consider, for example, the following state of stress (in MPa):
stress = [120 -55 -75; -55 55 33; -75 33 -85]
stress = 120 -55 -75
-55 55 33
-75 33 -85

To solve the cubic equation in Eq. (2.58) for this state of stress we need the coefficients of the
cubic, which are in terms of the invariants ðI 1 , I 2 , I 3 Þ. A MATLAB® function has been
written, stress_invs (see Appendix D, which has listings of all the MATLAB® functions
developed specifically for this book) which can be called as:
[I1, I2, I3] = stress_invs(stress)
I1 = 90
I2 = -18014
I3 = -4.7168e+05

where I 1 has the dimensions of MPa, I 2 has the dimensions (MPa)2 and I 3 has the
dimensions (MPa)3. Having these invariants, we can then use the MATLAB® roots func-
tion, which takes a vector argument containing the coefficients of the cubic (the invariants
together with the appropriate signs as seen in Eq. (2.58)) and returns the three roots that are
the principal stresses:
p_stresses = roots( [1,-I1,I2,-I3] )
p_stresses = 176.7995
-110.8640
24.0644

We can, of course, also obtain these principal stresses with eig:

[pdirs, pvals] = eig(stress)
pdirs = 0.2872 0.4654 0.8372
-0.0944 0.8836 -0.4587
0.9532 -0.0527 -0.2977
pvals = -110.8640 0 0
0 24.0644 0
0 0 176.7995

which returns the same values (but in a different order).

ð3Þ
Now, consider finding the principal directions. Let us choose first σ p ¼ 176:8 MPa and


ð3Þ ð3Þ
try to obtain the associated principal direction components nx ; ny ; nðz3Þ . From Eq. (2.56)
we have
48 Stress


120  σ ðp3Þ nðx3Þ  55nðy3Þ  75nðz3Þ ¼ 0


55nðx3Þ þ 55  σ ðp3Þ nðy3Þ þ 33nðz3Þ ¼ 0


75nðx3Þ þ 33nðy3Þ þ 85  σ ðp3Þ nðz3Þ ¼ 0

Since the determinant of this homogeneous system of equations is zero, these three equations
are not all independent so we cannot solve the system directly. However, we can try to find a
subset of these equations that are independent and solve that system for a ratio of the
principal direction components. For example, let’s divide the first two equations by nðz3Þ and
try to solve that system of equations:

 nð3Þ ð3Þ
ny
120  σ ðp3Þ
x
ð3Þ
 55 ð3Þ
¼ 75
nz nz
nx
ð3Þ
 nð3Þ
55  σ ðp3Þ ð3Þ ¼ 33
y
55 ð3Þ
þ
nz nz

This is easy to do by hand but we can use MATLAB® also by forming up the coefficients of
these equations and the right-hand side and then using a built-in MATLAB® solver for a
linear system of equations. The steps are:
% choose a principal stresss
sp3 = pvals(3,3);
% form up the matrix of coefficients of a pair of equations for this
stress
C(1,1) = 120 - sp3; C(1,2) = -55; C(2,1) =-55; C(2,2)=55- sp3
C = -56.7995 -55.0000
-55.0000 -121.7995

% form up the right hand side as a column vector

b = [ 75;-33]
b = 75
-33
% solve the system of equations [C]{r} ={b} for the ratios of unit
% vector components
r = C\b
r= -2.8126
1.5410
 2.4 Principal Stresses and Principal Stress Directions 49

so that in our case we have approximately

ð3Þ ð3Þ
nx ny
ð3Þ
¼ 2:81, ð3Þ ¼ 1:54
nz nz

2
2   2
ð3Þ ð3Þ
However, from the normalization nx þ ny þ nðz3Þ ¼ 1 we have

1
nðz3Þ ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi



ð3Þ ð3Þ 2 ð3Þ ð3Þ 2
1 þ nx =nz þ ny =nz

so that we find from MATLAB®, choosing the plus sign

nz3 = 1/sqrt(1 + r(1)^2 + r(2)^2)
nz3 = 0.2977
nx3 = r(1)*nz3
nx3 = -0.8372
ny3 = r(2)*nz3
ny3 = 0.4587

We can repeat these calculations choosing the minus sign so we obtain, finally, the two
choices
ð3Þ
nðx3Þ ¼ 0:8372, nx ¼ þ0:8372
ð3Þ
nðy3Þ ¼ þ0:4586, ny ¼ 0:4586
nðz3Þ ¼ þ0:2977, nðz3Þ ¼ 0:2977
Either of the two sign choices will give legitimate principal stress directions. If we examine
the pdirs output of eig we see the eigenvalue solver made the second choice listed above, but
if we compute the determinant of the the pdirs matrix we have
det(pdirs)
ans = -1.0000

so in order for the pdirs matrix to form up the matrix of unit vectors for a right-handed
system if we keep the first two columns of pdirs we should change the last column to be the
unit vector

nðx3Þ ¼ 0:8372
nðy3Þ ¼ þ0:4586
nðz3Þ ¼ þ0:2977
We now have to repeat this entire process twice to find the other two eigenvectors. This
makes the direct process of solving the system of equations for the principal directions rather
50 Stress

tedious. Also, there are some issues with this solution process. For example, how did we
know that the two equations we chose were independent equations? We simply assumed they
were independent and tried to find a solution under that assumption. If the two equations we
chose were not independent, our solution would fail and we would have to make another
choice. Also, what if only one of these was an independent equation? We will examine that
possibility shortly.

2.4.1 Special Case when I3 ¼ 0, Plane Stress

If the stress state is such that the third invariant, I 3 , is zero, then we can actually find the
principal stresses in an explicit form. Thus, consider setting up the following state of stress
(in MPa):
stress = [ 0 20 0; 20 0 10; 0 10 0]
stress = 0 20 0
20 0 10
0 10 0

[I1, I2, I3] = stress invs(stress)

I1 = 0
I2 = -500
I3 = 0

so this state of stress does satisfy I 3 ¼ 0. (The invariant I 1 ¼ 0 also in this case, but that is
not important for our discussion.) We can use eig to find the principal stresses and principal
directions in the normal fashion to find
[pdirs, pvals] = eig(stress)
pdirs = 0.6325 -0.4472 0.6325
-0.7071 0.0000 0.7071
0.3162 0.8944 0.3162

pvals = -22.3607 0 0
0 -0.0000 0
0 0 22.3607

We see that one of the principal stresses is zero. If we examine the cubic equation, Eq. (2.58),
when I 3 ¼ 0, that equation can be rewritten as


σ p σ 2p  I 1 σ p þ I 2 ¼ 0 (2.65)

Using the quadratic formula, we can explicitly find three roots of Eq. (2.65) as
 2.4 Principal Stresses and Principal Stress Directions 51

V yy
Figure 2.11 A plane state of stress (also called a biaxial state of stress).

V yx
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
I1  I 21  4I 2
V xy V xx σ p1 , σ p2 ¼
y 2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (2.66)
I1
¼  I 21 =4  I 2
x 2
z σ p3 ¼ 0

which you can verify gives the same values as found from eig. Thus, whenever I 3 ¼ 0 we can
find the principal stresses directly from Eq. (2.66). An important special state of stress that is
often examined in elementary mechanics of materials courses is the case of a state of plane
stress (also called a state of biaxial stress) where σ zz ¼ σ yz ¼ σ xz ¼ 0 (see Fig. 2.11). In this
case I 3 ¼ 0 and if we examine the expressions for I 1 , I 2 it is easy to show that the principal
stresses from Eq. (2.66) are
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

σ  σ 2 ffi
σ xx þ σ yy xx yy
σ p1 , σ p2 ¼  þ σ xy 2
(2.67)
2 2
σ p3 ¼ 0

which is a formula found in most elementary texts.

2.4.2 Special Case when Principal Stresses are Equal

Let us define a state of stress, given in MPa, as:
stress = [ 0 100 100; 100 0 100; 100 100 0]
stress = 0 100 100
100 0 100
100 100 0

If we use eig to find the principal stresses and principal directions, then we find
[pdirs, pvals] = eig(stress)
pdirs = 0.4082 0.7071 0.5774
0.4082 -0.7071 0.5774
-0.8165 0 0.5774

pvals = -100.0000 0 0
0 -100.0000 0
0 0 200.0000

which all seems normal, but we notice that two of the principal stresses are equal so that
there are more details to this problem that the use of eig does not expose. To see what is
52 Stress

happening, consider the equations for the principal directions (Eq. (2.56)) for the
ð3Þ
σ p ¼ 200 MPa value, which are

200nðx3Þ þ 100nðy3Þ þ 100nðz3Þ ¼ 0

100nðx3Þ  200nðy3Þ þ 100nðz3Þ ¼ 0 (2.68)

100nðx3Þ þ 100nðy3Þ  200nðz3Þ ¼ 0

We can, as done previously, solve two equations from this system and use the normalization
to find

nðx3Þ ¼ 0:5774
nðy3Þ ¼ 0:5774 (2.69)
nðz3Þ ¼ 0:5774

which we recognize as the result found with eig if we choose the plus signs. However, if we
ð1Þ, ð2Þ
consider instead the two repeated principal stresses σ p ¼ 100 MPa, then we obtain:

100nðxiÞ þ 100nðyiÞ þ 100nðziÞ ¼ 0

100nðxiÞ þ 100nðyiÞ þ 100nðziÞ ¼ 0 ði ¼ 1; 2Þ (2.70)

100nðxiÞ þ 100nðyiÞ þ 100nðziÞ ¼ 0

These are all the same equation so even with the normalization equation, we have an
infinite number of solutions. The eigenvectors associated with different eigenvalues are
perpendicular to each other so this means that every direction in a plane P perpendicular
to nð3Þ must be a principal direction where the principal stress is –100 MPa. This is shown in
Fig. 2.12. To find a particular direction in plane P we could, for example, set nðz2Þ ¼ 0. Then
ð2Þ ð2Þ
the solution of Eq. (2.70) is nx =ny ¼ 1 and normalizing the vector to make it a unit
ð2Þ
vector gives ny ¼ 0:7071 so that choosing the minus sign we have

nðx2Þ ¼ 0:7071
nðy2Þ ¼ 0:7071 (2.71)
n(1)
nðz2Þ ¼ 0
100

100 n ( 3)
Figure 2.12 The case being considered when two principal stresses are equal.
200
( 2) We can find a principal direction, nð3Þ , for the principal stress of 200 MPa (to
n
within a sign), and particular set of orthogonal directions in a plane P that
 
are perpendicular to nð3Þ , nð1Þ , nð2Þ , that define directions for the repeated
principal stresses of –100 MPa, but any other set of directions in the plane,
P such as the dashed lines shown, for example, are also principal directions with
the same principal stress.
 2.5 Mohr’s Circle for Three-Dimensional States of Stress 53

which is identical to the result found for nð2Þ with eig. To find another unit vector in the plane
P so that we have a right-handed system, we can choose the plus sign in Eq. (2.69) for the
components of nð3Þ and then compute nð1Þ ¼ nð2Þ  nð3Þ , giving

nðx1Þ ¼ 0:4082
nðy1Þ ¼ 0:4082 (2.72)

nðz1Þ ¼ 0:8165

which is the negative of the result obtained with eig since, as you can easily check, eig does
not produce eigenvectors that form a right-handed system.
What happens if all the principal stresses are equal? In that case every direction is a
principal direction so that the state of stress is only a normal-stress component, which is the
same in all directions. This is similar to what we see in an ideal fluid at rest where the normal
stress is negative, i.e., we have a pressure which is the same in all directions at any point in
the fluid (although the pressure, of course, can vary from point to point).

2.5 Mohr’s Circle for Three-Dimensional States of Stress

In elementary mechanics of materials courses, a graphical construction called Mohr’s circle
is often used in place of the transformation equations we have described in this chapter to
analyze problems when there is a state of plane stress. Since software such as MATLAB®
can effortlessly solve even complex stress relations, there is now less need to use such
graphical methods to perform detailed stress calculations. However, one can gain some
insight into the nature of stresses by examining the Mohr’s circle construction for a general
state of stress, which is what we will now do.
If we write the stresses on an arbitrary cutting plane whose normal is n in terms of the
principal stresses and the components of n as measured in the principal coordinates
ðp1 , p2 , p3 Þ, as shown in Fig. 2.13, then the stress-vector components along the principal
directions are (see Eq. (2.12)):
ðnÞ
T 1 ¼ n1 σ p1
ðnÞ
T 2 ¼ n2 σ p2 (2.73)
ðnÞ
T 3 ¼ n3 σ p3

The normal stress and total shear stress on the arbitrary plane and the condition that the
normal n is a unit vector are given by
N ¼ σ nn ¼ σ 2p1 n21 þ σ p2 n22 þ σ p3 n23




ðnÞ 2 ðnÞ 2 ðnÞ 2
S 2 ¼ ðτ s Þ2 ¼ T 1 þ T2 þ T3  N2 (2.74)

n21 þ n22 þ n23 ¼ 1

54 Stress

Figure 2.13 The normal stress, N, and total T( n )

shear stress, S, on an arbitrary cutting plane
p2 n = ( n1 , n2 , n3 )
with a normal, n, as expressed in terms of the
principal coordinate directions. N = s nn
S =ts
s p2

s p1
s p3 p1
p3 principal
directions

Equation (2.74) can be considered to be three equations for the unit normal components
(squared) in terms of N, S, and the principal stresses. We can solve these three equations for
 
these positive unknowns, n21 ; n22 ; n23 , where
  
S 2 þ N  σ p2 N  σ p3
n1 ¼ 
2   0
σ p1  σ p2 σ p1  σ p3
  
S 2 þ N  σ p1 N  σ p3
n2 ¼ 
2   0 (2.75)
σ p2  σ p3 σ p2  σ p1
  
S 2 þ N  σ p1 N  σ p2
n3 ¼ 
2   0
σ p3  σ p1 σ p3  σ p2

If we order the stresses such that σ p1 > σ p2 > σ p3 then the inequalities of Eq. (2.75) imply
that
  
S 2 þ N  σ p2 N  σ p3 0
   
S 2 þ N  σ p3 N  σ p1 0 (2.76)
   
S 2 þ N  σ p1 N  σ p2 0

which can also be rewritten equivalently as


σ
σ p2 þ σ p3 2
p2  σ p3 2
S þ N
2
2 2

σ
σ p1 þ σ p3 2
p3  σ p1 2
S2 þ N  (2.77)
2 2

σ
σ p1 þ σ p2 2
p1  σ p2 2
S2 þ N 
2 2

If we plot the two quantities S and N, the three inequalities in Eq. (2.77) can be interpreted as
the regions exterior or interior to three circles, shown as the shaded region in Fig. 2.14,
whose centers are at
 2.5 Mohr’s Circle for Three-Dimensional States of Stress 55

S Figure 2.14 Mohr’s circle construction showing all the

normal stresses and total shear stresses possible at a given
point in terms of the principal stresses as the shaded region
between circles.

τ2
τ1
V p3 c1 V p 2 c2 c3 V p1 N
τ3

σ p2 þ σ p3
c1 ¼
2
σ p1 þ σ p3
c2 ¼ (2.78)
2
σ p1 þ σ p2
c3 ¼
2
and whose radii are
σ p2  σ p3
τ1 ¼
2
σ p3  σ p1
τ2 ¼ (2.79)
2
σ p1  σ p2
τ3 ¼
2
which are also the magnitudes of the three extreme values of the shear stress, as seen in
Fig. 2.14.
For plane-stress problems, many books use the Mohr’s circle construction to analyze the
stresses on cutting planes that correspond to one of the circles in Fig. 2.14. In that case, the
Mohr’s circle construction offers a graphical alternative to using the transformation equa-
tions discussed in this chapter. We will not discuss this type of use of Mohr’s circle here but
you can find the details in elementary mechanics of materials or strength of materials books.
For a general 3-D state of stress, we may be interested in stresses (more precisely, N and S
values) which lie anywhere within the shaded region of possible stresses shown in Fig. 2.14.
In this case the Mohr’s circle graphical construction is not particularly useful for finding
those stresses but we can easily obtain any stresses needed with the transformation equations
previously discussed. However, the Mohr’s circle construction of Fig. 2.14 is useful to help
56 Stress

locate the planes on which the extreme shear stresses act. For example, we see that there are
planes of extreme shear stress when
σ p1 þ σ p2
N¼
2

σ  σ 2
p1 p2
S2 ¼
2
so that from Eq. (2.75) we find
1 1
n21 ¼ , n22 ¼ , n23 ¼ 0
2 2
Similarly, when
σ p1 þ σ p3
N¼
2

σ  σ 2
p1 p3
S2 ¼
2
we have
1 1
n21 ¼ , n23 ¼ , n23 ¼ 0
2 2
and, finally, for
σ p2 þ σ p3
N¼
2

σ  σ 2
p2 p3
S2 ¼
2
we have
1 1
n22 ¼ , n23 ¼ , n21 ¼ 0
2 2
These results are all summarized inpTable
ffiffiffi 2.1.
Since sin ð45 Þ ¼ cos ð45 Þ ¼ 1= 2 and the components of the unit normal in Table 2.1
are measured with respect to the principal directions, we see that the planes of extreme shear

Table 2.1 The planes of extreme shear and the shear stress (squared) and
normal stress on those planes

n1 n2 n3 S2 N
pffiffiffi pffiffiffi  2  
0 1= 2 1= 2 σ p2  σ p3 =4 σ p2 þ σ p3 =2
pffiffiffi pffiffiffi  2  
1= 2 0 1= 2 σ p1  σ p3 =4 σ p1 þ σ p3 =2
pffiffiffi pffiffiffi  2  
1= 2 1= 2 0 σ p1  σ p2 =4 σ p1 þ σ p2 =2
 2.5 Mohr’s Circle for Three-Dimensional States of Stress 57

p3 S V p1  V p 3 / 2 Figure 2.15 Stresses on a plane of extreme shear stress. Note that

V p1  V p 3 actual direction of the total shear stress may be opposite to that
N
2 V p1 shown, depending on the values for the principal stresses.
n
S all lie at 45 from the principal directions. One such
extreme shear stress plane is shown in Fig. 2.15.
45o Generally, however, we are interested only in the magni-
p2
tude of the largest extreme shear stress.
p1
Since we ordered the principal stresses such that
V p3 σ p1 > σ p2 > σ p3 , the magnitude of this largest shear stress
will be σ p1  σ p3 =2 and one of the planes on which it acts
is shown in Fig. 2.15. If we want to find a plane of extreme shear and the direction of the total
shear stress on that plane explicitly, then this can easily be done with MATLAB® as we will
now show with an example.

Example 2.4 Finding the Plane of an Extreme Shear Stress and its Direction
Consider a state of stress (in MPa) given in MATLAB® by
stress = [ 10 -3 0; -3 -7 2; 0 2 5]
stress = 10 -3 0
-3 -7 2
0 2 5

and use eig to calculate the principal stresses and principal directions
[pdirs, pvals] = eig(stress)
pdirs = 0.1641 0.0884 0.9825
0.9746 0.1390 -0.1753
-0.1521 0.9863 -0.0633
pvals = -7.8172 0 0
0 5.2819 0
0 0 10.5353

These principal directions do form a right-handed coordinate system since the determinant is
plus one:
det(pdirs)
ans = 1.0000

(If the system were not right-handed then we would change the signs on one of the unit vectors
in pdirs.) The magnitude of the largest shear stress on any cutting plane is one half the
magnitude of the largest principal stress minus the smallest principal stress given here by
58 Stress

tau_max = abs(pvals(3,3) -pvals(1,1))/2

tau_max = 9.1763

and the normal stress on this plane is

norm_stress = (pvals (1,1) +pvals(3,3))/2
norm_stress = 1.3591

These values are easily seen in the Mohr’s circle construction for this case, which is shown in
Fig. 2.16. The principal stresses, ordered from the largest to the smallest, are approximately
σ p1 ¼ 10:54 MPa, σ p2 ¼ 5:28 MPa, σ p3 ¼ 7:82 MPa. Now, let the unit vectors in the
columns of the pdirs matrix that are associated with these principal stresses be ep1 , ep2 , ep3 ,
respectively. These unit vectors are shown schematically in Fig. 2.17a. We can extract these
unit vectors from the pdirs matrix:
ep1 = pdirs(:,1)
ep1 = 0.1641
0.9746
-0.1521
ep2 = pdirs(:,2)
ep2 = 0.0884
0.1390
0.9863
ep3 = pdirs(:,3)
ep3 = 0.9825
-0.1753
-0.0633

The planes of largest shear associated with the principal

 
stresses σ p1 , σ p3 will be at 45 from the principal direc-
S  
(1.36,9.18) tions ep1 , ep3 containing either lines a–a or b–b as shown
in Fig. 2.17a. Either of those planes could be chosen and we
could also choose to look at the element on either side of
the cutting planes through a–a or b–b so there are actually
four choices we could make for planes having the largest
– 7.82 5.28 10.54 shear stress. It is not important which of these choices we
N
make, so let’s use the cutting plane through a–a and

Figure 2.16 Mohr’s circle construction for Example 2.4

showing the principal stresses and the normal stress and largest
total shear values on the plane of maximum shear stress.
 2.5 Mohr’s Circle for Three-Dimensional States of Stress 59

e p2 Figure 2.17 (a) The principal stresses and

principal stress directions. (b) The unit
b normal, ne , acting on plane a–a, which is a
10.54
a 5.28 plane of extreme shear. (c) Plane a–a, as seen
7.82
looking in the ep2 direction.
a
45o
e p3 a
a
7.82 10.54 S = 9.18
b N = 1.36
e p1
(a) (b)
−e p1
−e p 3
a
45o
e p1 + e p 3 a
ne =
2
(c)

examine the element shown in Fig. 2.17b. If we look down the ep2 axis, this element
appears as shown in Fig. 2.17c. From the geometry of Fig. 2.17c, it is easy to see that
the normal to the plane containing the largest shear stress is given in terms of the
  pffiffiffi
principal stress directions as ne ¼ ep1 þ ep3 = 2 . In MATLAB® we can calculate this
unit normal easily:
ne = (ep1+ep3)/sqrt(2)
ne = 0.8108
0.5652
-0.1523

Since we now know the normal to this cutting plane and the state of stress, it is
easy to use the transformation equations discussed previously to find any infor-
mation we want about the stresses on this plane. For the stress vector on the plane
(see Eq. (2.16)):
stress_vec = stress*ne
stress_vec = 6.4119
-6.6935
0.3688

For the normal stress (Eq. (2.19)):

norm_stress2 = ne’*stress*ne
norm_stress2 = 1.3591
60 Stress

which agrees with our previous result. For the magnitude of total shear stress (Eq. (2.20)) we
have:
Smax = sqrt(norm(stress_vec)^2 - norm_stress2^2)
Smax = 9.1763

which is obviously the largest shearing stress. The direction of this stress is just (see Eq. (2.22))
s = (stress_vec -norm_stress2*ne)/Smax
s = 0.5787
-0.8131
0.0628

Most books do not go into this explicit detail about determining the plane of the largest
shear stress and its direction since it is only the magnitude of the maximum shearing stress
that appears in the commonly discussed theories of failure [1] (see Chapter 13). However, we
have given a more extensive discussion here since it also shows the importance of the
transformation equations we have derived earlier.

2.6 Stresses on the Octahedral Plane

In addition to the principal planes and the planes of maximum shearing stress a particular
cutting plane of interest in some material failure theories is the octahedral plane. The
octahedral plane is by definition a plane that makes equal angles with respect to the three
principal directions. Thus, if we let nj be the component of the unit normal to the octahedral
plane as measured with respect to the to the principal directions, then we see

1
n1 ¼ n2 ¼ n3 ¼ pffiffiffi (2.80)
3

defines an octahedral plane. The three components of the stress vector on the octahedral
plane (as measured also with respect to the principal directions) and the normal stress on the
octahedral plane are
ðnÞ
T 1 ¼ σ p1 n1
ðnÞ
T 2 ¼ σ p2 n2
(2.81)
ðnÞ
T 3 ¼ σ p3 n3
σ nn ¼ σ p1 n21 þ σ p2 n22 þ σ p3 n23

so that we find
 2.7 Stress Notations and the Concept of Stress – A Historical Note 61

σ p1 þ σ p2 þ σ p3 σ xx þ σ yy þ σ zz I 1
ðσ nn Þoct ¼ ¼ ¼
3 3 3
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2  2  2
jτ s joct ¼ σ p1 n1 þ σ p2 n2 þ σ p3 n3  σ p1 n21 þ σ p2 n22 þ σ p3 n23
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (2.82)
1  2  2  2
¼ σ p1  σ p2 þ σ p2  σ p3 þ σ p1  σ p3
3
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
¼ 2I 21  6I 2
3
Since the last expression for the total shear stress on the octahedral plane is in terms of the
stress invariants only, we can also write this total shear stress in terms of stresses associated
with the (x, y, z) axes. We find
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

2
1  2  2
jτ s joct ¼ σ xx  σ yy þ ðσ xx  σ zz Þ2 þ σ yy  σ zz þ 6 σ 2xy þ σ 2yz þ σ 2xz (2.83)
3
A closely related normal stress is called the von Mises stress, σ v , given in terms of the
octahedral stress as
3
σ v ¼ pffiffiffi jτ s joct (2.84)
2
which is a stress that appears in theories of yielding of materials [1] (also see Chapter 13).
This stress is also sometimes called an equivalent stress or the effective stress. For complex
3-D states of stress, the von Mises stress is often used as the stress of choice to plot to show
the internal distribution of stresses in a body.

2.7 Stress Notations and the Concept of Stress – A Historical Note

The concept of stress was developed in 1821 by Augustin-Louis Cauchy (1789–1857), so it is
200 years old. Over the years, various notations have been used to characterize the state of
stress. Some of the most common of these notations are listed in Fig. 2.18 along with the
names of the scientists who used them. [For a text that is notable for having pictures of many
of the individuals who contributed to the field of mechanics over the years, see [2].] As you
have seen, in this book we will generally write stresses as σ xx , σ xy , etc., or use numerical
subscripts to write them as σ 11 , σ 12 , etc. The use of numerical subscripts is important since it
facilitates writing stress relationships in matrix–vector forms. Occasionally, we will distin-
guish normal and shear stresses by writing normal stresses using the Greek symbol sigma ðσ Þ
and shear stresses with the Greek symbol tau ðτ Þ. The normal stress and total shear stress are
examples where we wrote them as ðσ nn , τ s Þ.
The concept of stress is not an intuitive one, so before the work of Cauchy many
researchers struggled with how to describe internal force distributions. Also, the concept
62 Stress

Stress notations

Cauchy (early work) Poisson Coriolis, Saint-Venant,

Maxwell, Cauchy (later work)
A F E P3 Q3 R3 pxx pxy pxz
F B D P2 Q2 R2 p yx p yy p yz
E D C P1 Q1 R1 pzx pzy pzz

Neuman, Kirchhoff Kelvin Clebsch

Rieman, Love

Xx Xy Xz P V T t11 t12 t13

Yx Yy Yz V Q S t21 t22 t23
Zx Zy Zz T S R t31 t32 t33

Pearson Current notation

ˆ ˆ ˆ
xx xy xz
V xx V xy V xz V 11 V 12 V 13
ˆ ˆ ˆ
yx yy yz V yx V yy V yz or V 21 V 22 V 23
ˆ ˆ ˆ V zx V zy V zz V 31 V 32 V 33
zx zy zz

Figure 2.18 Stress notations used by various authors to describe the state of stress.

of stress we have presented is in a sense incomplete since we know from statics that a general
distribution of forces is statically equivalent to both a force and a moment. Thus, strictly
speaking, in Fig. 2.1 we should have drawn a small net force, ΔF, and a small net moment,
ΔM, acting on an area ΔA of a cutting plane. In principle, therefore, we could then also
define a couple stress, CðnÞ , as
ΔM
CðnÞ ¼ lim (2.85)
ΔA!0 ΔA

Studies of “ordinary” structural materials have not found instances where the concept
of couple stresses is needed, and in this book we do not include them. However,
modern materials with complex microstructures may require such generalizations of
the stress concept to describe local distributions of forces and moments in
such materials.
 2.8 Problems 63

2.8 PROBLEMS

P2.1 Consider a small element such as the one shown in Fig. 2.4. The moment equation for
this element can be written as
ð ð X6 ð
ρr  adV ¼ ρr  gdV þ r  Tðni Þ dA (P2.1)
i¼1
ΔV ΔV ΔAi

where ΔV is the volume of the element, ΔAi are the six faces of the element
(including the back faces, which are not shown in Fig. 2.4) on which the stress
vectors Tðni Þ act, ρ is the mass/volume (density) of the element, and a is the
acceleration of the element. The quantity g is the acceleration of gravity. The
position vector r is measured from the center of the element. Show that in the limit
as ΔV ! 0, ΔAi ! 0 that Eq. (P2.1) implies that the shear-stress components are
symmetric. Note that in the limit the volume integrals go to zero faster than the
surface integral so that in fact those volume integrals can be neglected.
P2.2 Prove Eq. (2.8), i.e., show that the ratio of areas of the faces of the tetrahedron
shown in Fig. 2.5 are components of the unit normal of the front inclined face.
(Hint: use properties of the cross product of vectors along the sides of the
tetrahedron.)
P2.3 Consider the state of stress with respect to the ðx; y; zÞ axes given in MPa by
2 3
40 40 30
½σ  ¼ 4 40 20 0 5
30 0 20
(a) Determine the normal stress and total shear stress acting on a plane which is oriented
at angles of 40 , 75 , and 54 approximately with respect to the ðx; y; zÞ axes,
respectively.
(b) Determine the direction of the total shear stress on the plane of part (a).
P2.4 For the state of stress with respect to the ðx; y; zÞ axes given in MPa by
2 3
12 6 9
½σ  ¼ 4 6 10 3 5
9 3 14
determine the normal stress and shear stress acting on the octahedral plane.
P2.5 For the state of stress with respect to the ðx; y; zÞ axes given in MPa by
2 3
60 40 40
½σ  ¼ 4 40 0 20 5
40 20 20
determine the state of stress in the ðx0 , y0 , z0 Þ system shown in Fig. P2.1 obtained by
making a 30 counterclockwise rotation about the x-axis as shown.
64 Stress

y' y Figure P2.1 A coordinate rotation.

º
30

P2.6 Consider the state of stress with respect to the ðx; y; zÞ axes given in
MPa by
z 2 3
20 12 0
30º
½σ  ¼ 4 12 17 22 5
z' 0 22 15
x, x'

(a) Determine the principal stresses and principal stress directions.

(b) Determine the absolute maximum shear stress in magnitude and the plane on
which it acts as well as the normal stress on this plane of extreme shear stress.
P2.7 Consider the state of stress with respect to the ðx; y; zÞ axes given in MPa by
2 3
200 40 30
½σ  ¼ 4 40 50 20 5
30 20 60
(a) Determine the ðx; y; zÞcomponents of the stress vector, TðnÞ , acting on a plane
P whose unit normal is
3 3 1
n ¼ pffiffiffiffiffi ex þ pffiffiffiffiffi ey þ pffiffiffiffiffi ez
19 19 19
 
where ex , ey , ez are unit vectors along the ðx; y; zÞ axes, respectively.
(b) Determine the normal stress and total shear stress acting on plane P and the
direction of the total shear stress on this plane.
(c) Two orthogonal unit vectors that lie in plane P are given by
1 1 6
t ¼ pffiffiffiffiffi ex þ pffiffiffiffiffi ey  pffiffiffiffiffi ez
38 38 38
1 1
v ¼  pffiffiffi ex þ pffiffiffi ey þ 0ez
2 2
Determine the stresses σ nt , σ nv acting on plane P. Show that these two shear stresses
combine to produce the total shear stress (in both magnitude and direction)
calculated in (b).
(d) Determine the principal stresses and principal directions and the magnitude of the
maximum shearing stress.
(e) Let the ðx0 , y0 , z0 Þ axes be oriented along the ðn; t; vÞ directions, respectively.
Determine the complete state of stress with respect to these ðx0 , y0 , z0 Þ axes. How
are your results related to the results of parts (b) and (c)?
 2.8 Problems 65

B 20 ksi Figure P2.2 An angular plate subjected to uniform shear stresses on

C its edges.
o
20 ksi 40

20 ksi P2.8 An angular plate in a state of plane stress carries stresses

40
o of 20 ksi on its sides as shown in Fig. P2.2. If the state of
stress is uniform throughout the plate, determine
A
D (a) the stress vector components T ðxnÞ and T ðynÞ exerted
20 ksi on planes AB and BC;
 
(b) the stresses σ xx , σ yy , σ xy in the plate.
P2.9 An angular plate in a state of plane stress carries stresses of 20 ksi on its sides as shown
in Fig. P2.3. If the state of stress is uniform throughout the plate, determine
(a) the stress vector components T ðxnÞ and T ðynÞ exerted on planes AB and BC;
 
(b) the stresses σ xx , σ yy , σ xy in the plate.
P2.10 The state of stress, in MPa, with respect to a set of ðx; y; zÞ axes, is given by
2 3
10 15 0
σ ¼ 4 15 20 5 5
0 5 30
(a) Using MATLAB® determine the principal stresses and principal directions. Does
the matrix of principal directions form a right-handed system?
(b) Using Eq. (2.19) show that the normal stress on the plane with a normal in the
direction of the largest principal stress agrees with the results of part (a).
(c) Using Eq. (2.16) determine the (x, y, z) components of the stress vector on the
plane with a normal in the direction of the largest principal stress. Show that the
component of this stress vector along the largest principal stress direction agrees
with the results of part (a).
P2.11 The state of stress, in MPa, with respect to a set of ðx; y; zÞ axes, is given by
2 3
12 35 5
σ ¼ 4 35 40 0 5
5 0 10
20 ksi
(a) Using MATLAB® determine the principal
B
C stresses and principal directions. Does the
o
40 matrix of principal directions form a right-
20 ksi
handed system? If not, make it right-handed.
20 ksi (b) Determine the magnitude of the extreme
o
40 shear stress, τ max , and unit vectors n, t that
A
D
Figure P2.3 An angular plate subjected to uniform stresses
20 ksi parallel to its edges.
66 Stress

are normal to the τ max plane and parallel to the extreme shear stress vector,
respectively. Verify, using the stress transformation equations, Eq. (2.29), that
you obtain the correct normal stress, σ nn , and shear stress, σ nt , on the plane of
maximum shear stress.
P2.12 The state of stress, in MPa, with respect to a set of ðx; y; zÞ axes, is given by
2 3
50 10 10
σ ¼ 4 10 30 5 5
10 5 20
(a) Using MATLAB® determine the principal stresses and principal directions. Does
the matrix of principal directions form a right-handed system? If not, make it
right-handed.
(b) Determine the magnitude of the extreme shear stress,τ max , and unit vectors n, t
that are normal to the τ max plane and parallel to the extreme shear stress vector,
respectively. Verify, using the stress transformation equations, Eq. (2.29), that
you obtain the correct normal stress, σ nn , and shear stress, σ nt , on the plane of
maximum shear stress.
P2.13 At a point in a body the state of stress, as measured in ksi, with respect to a ðx; y; zÞ
coordinate system is given by
2 3
10 4 6
½σ  ¼ 4 4 3 2 5
6 2 8
Consider three lines that extend from the origin of this coordinate system given by
the vectors ðOX; OY; OZÞ, where
OX ¼ i þ j þ k
OY ¼ i þ j  2k
OZ ¼ i þ j
where the vectors ði; j; kÞ are unit vectors along the ðx; y; zÞ axes respectively. Note
that these vectors are all perpendicular to each other and act along a right-handed
ðX ; Y ; Z Þ coordinate system. Determine:
(a) the ðx; y; zÞ stress vector components for the two planes whose normals are along
the OX and OZ lines, respectively;
(b) the normal stresses acting on the two planes of part (a);
(c) the shear stresses associated with the pair of lines ðOX; OYÞ and the lines
ðOY; OZÞ.
(d) For the ðX ; Y ; Z Þ axes along the ðOX; OY; OZÞ directions, determine the state of
stress with respect to these ðX ; Y ; Z Þ axes. How are your results related to the
stresses you calculated in parts (b) and (c)?
 2.8 Problems 67

(e) Show that both the state of stress you found in part (d) and the original state of
stress produce the same cubic equation for determining the principal stresses.
Find both the principal stresses and principal stress directions.
P2.14 The state of stress, as measured in MPa, with respect to an ðx; y; zÞ coordinate system,
is given by
2 3
100 50 50
½σ  ¼ 4 50 30 20 5
50 20 70
(a) Determine the ðx; y; zÞ components of the stress vector acting on a plane P whose
unit normal, n, makes equal angles with respect to the ðx; y; zÞ axes. [Note: this is
not the octahedral plane, which makes equal angles with respect to the principal
axes.]
(b) Find the normal stress and total shear stress acting on plane P and the direction of
the total shear stress on this plane.
(c) Two orthogonal unit vectors ðt; vÞ that lie in the plane P are given by:
1 1 2
t ¼ pffiffiffi ex þ pffiffiffi ey  pffiffiffi ez
6 6 6
1 1
v ¼ pffiffiffi ex þ pffiffiffi ey
2 2
Determine the shear stresses ðσ nt , σ nv Þ acting on the plane P. Show that these two
shear stress components combine to produce the total shear stress (in both
magnitude and direction) calculated in part (b).
(d) Determine the principal stresses and principal stress directions and the magnitude
of the maximum shear stress.
(e) Let the ðx0 , y0 , z0 Þ axes be oriented in the ðn; t; vÞ directions, respectively. Determine
the complete state of stress with respect to the ðx0 , y0 , z0 Þ axes.
P2.15 The deviatoric stresses, sij , are related to the stresses, σ ij , by subtracting off a mean
 
normal stress component, σ m ¼ σ xx þ σ yy þ σ zz =3, from all the normal stresses, i.e.,
2 3 2 3
sxx sxy sxz σ xx  σ m σ xy σ xz
4 syx syy syz 5 ¼ 4 σ yx σ yy  σ m σ yz 5
szx szy szz σ zx σ zy σ zz  σ m

This deviatoric stress state is used in theories of failure with respect to slip (yielding)
since it is a measure of those stresses that produce only distortional changes in a body.
Consider the state of stress (in MPa):
2 3
55 0 24
½σ  ¼ 4 0 46 0 5
24 0 43
68 Stress

(a) Determine the deviatoric stresses.

(b) Compare the cubic equation that defines the principal deviatoric stresses with the
cubic equation for the principal stresses. How are they related?
(c) Compare the principal deviatoric stresses with the principal stresses. How do they
differ?
(d) Compare the principal deviatoric stress directions with the principal directions of
the stresses. How are they related?
(e) Draw the 3-D Mohr’s circles for both the deviatoric stresses and the stresses. How
is the magnitude of the maximum deviatoric shear stress related to the magnitude
of the maximum shear stress?

References
1. J. R. Barber, Intermediate Mechanics of Materials, 2nd edn (Switzerland: Springer, 2011)
2. E. Volterra and J. H. Gaines, Advanced Strength of Materials (Englewood Cliffs NJ: Prentice-Hall,
1971)
3 Equilibrium

The previous chapter focused on the behavior of the stress vector and stresses at any fixed
point in a body. However, stresses will also vary from point to point within a deformable
body so that we need to describe those spatial variations. As we will see in this chapter, local
equations of equilibrium involving the stresses must be satisfied everywhere within a body.
Those governing equations of equilibrium, as described in Chapter 1, form one of the
fundamental “pillars” of any analysis of stresses.

3.1 Equations of Equilibrium – Cartesian Coordinates

From statics we know that for a rigid body to be in equilibrium, there must be no net vector
force, F, or moment, M0 , acting about any point, O, on the body; i.e., we have
F¼0
(3.1)
M0 ¼ 0

If we break a body into parts, then Eqs. (3.1) must hold for any part as well. For
deformable bodies, these equilibrium conditions must also be satisfied, both for the entire
body and for any part we choose. Let us consider a part of the deformable body that has a
volume, V, contained within a closed surface, S (see Fig. 3.1). On S the force distribution
will described by the stress vector, TðnÞ (whose outward unit normal at any point is n), and
within the part we will allow body forces, f (such as gravity), to act throughout the
volume, V, where f has the dimensions of force/unit volume. The force and moment
equilibrium of the part requires
ð ð
ðnÞ
T dS þ fdV ¼ 0
S V
ð ð (3.2)
ðnÞ
xs  T dS þ x  fdV ¼ 0
S V

where dS and dV are surface and volume elements, respectively, and ðxs ; xÞ are position
vectors from point O to a surface point or point within the volume, respectively, where either
TðnÞ or f acts. If we use Eq. (2.10) we can write the stress vector, TðnÞ , in terms of the stress
vectors, Tðei Þ , along the ðx; y; zÞ ¼ ðx1 , x2 , x3 Þ Cartesian axes (and where ei are unit vectors
along those axes) and the components of the unit normal to S to obtain

69
70 Equilibrium

Figure 3.1 A deformed body under the action of loads (shown as

concentrated forces only for illustration) and the distributed surface
forces (stress vector TðnÞ ) and body forces, f, acting on an arbitrary
volume V of a part of the body, where S is the closed surface
n
surrounding V and n is the unit normal on S.
T(
n)
S
V
y, x2 xs f
ðX
3 ð
x
Tðei Þ ni dS þ fdV ¼ 0
i¼1
S V
ðX ð (3.3)
x, x1 3
O ðei Þ
xs  T ni dS þ x  fdV ¼ 0
i¼1
z , x3 S V

Gauss’s theorem says that for any function gðxs Þ (which could be a scalar, a vector, or a
matrix) acting on S we have
ð ð
∂gðxÞ
gðxs Þni ðxs ÞdS ¼ dV (3.4)
∂xi
S V

so applying this theorem to the terms in Eq. (3.3) we have

ð X !
3
∂Tðei Þ
þ f dV ¼ 0
i¼1
∂xi
V
ð ! (3.5)
X3
∂ h ðei Þ
i
xT þ x  f dV ¼ 0
i¼1
∂xi
V

But since the part defined by the volume V is arbitrary, the integrands of Eq. (3.5) themselves
must vanish so that at any point in a deformable body we must have
X
3
∂Tðei Þ
þf ¼0
i¼1
∂xi
(3.6)
X3
∂ h i
x þ Tðei Þ þ x  f ¼ 0
i¼1
∂xi

If we expand the derivatives in the second equation of Eq. (3.6) we have

" #
X3
∂x ∂T ðei Þ
 Tðei Þ þ x  þf ¼0 (3.7)
i¼1
∂x i ∂xi
 3.1 Equations of Equilibrium – Cartesian Coordinates 71

Since ∂x=∂xi ¼ ei and the term in square brackets in Eq. (3.7) vanishes because of the first
equation in Eq. (3.6) we obtain
X
3
∂Tðei Þ
þf ¼0
i¼1
∂xi
(3.8)
X
3
ei  Tðei Þ ¼ 0
i¼1

The first equation in Eq. (3.8) are the conditions for local equilibrium in the deformable
body in terms of the stress vectors acting on the Cartesian coordinate planes. We can write
these stress vectors in terms of the stress components and likewise write f in terms of its
components as
X
3
Tðei Þ ¼ σ ij ej ði ¼ 1; 2, 3Þ
j¼1
(3.9)
X
3
f¼ f j ej
j¼1

to obtain the equations of equilibrium for the stresses as

X
3
∂σ ij
þ f j ¼ 0 ðj ¼ 1; 2, 3Þ (3.10)
i¼1
∂xi

For a general three-dimensional (3-D) state of stress, these equilibrium equations are,
therefore, explicitly, in terms of ðx; y; zÞ components
∂σ xx ∂σ yx ∂σ zx
þ þ þfx ¼ 0
∂x ∂y ∂z
∂σ xy ∂σ yy ∂σ zy
þ þ þfy ¼ 0 (3.11)
∂x ∂y ∂z
∂σ xz ∂σ yz ∂σ zz
þ þ þfz ¼ 0
∂x ∂y ∂z

Now, let us consider the second equation in Eq. (3.8). If we again write the stress vectors in
terms of the stresses this equation becomes
X
3 X
3  
σ ij ei  ej ¼ 0 (3.12)
i¼1 j¼1

We can consider the cross-product terms as the components of a vector-valued 3  3 matrix,

U, with components, Uij ¼ ei  ej , where
72 Equilibrium

2 3
0 e3 e2
U ¼ 4 e3 0 e1 5 (3.13)
e2 e1 0

which you can easily verify by carrying out the various cross products such as
U12 ¼ e1  e2 ¼ e3 , etc. Then Eq. (3.12) can be written in matrix form as
 
trace ½σ T ½U ¼ 0 (3.14)

We can manually carry out all the matrix operations in Eq. (3.14) but we can also let the
symbolic toolkit of MATLAB® do this for us with symbolic algebra. First, we define the
unit vectors and stresses as symbolic values and form up the symbolic stress and U
matrices:
syms e1 e2 e3 s11 s12 s13 s21 s22 s23 s31 s32 s33
U = [0 e3 -e2; -e3 0 e1; e2 -e1 0]
U = [0, e3, -e2]
[-e3, 0, e1]
[e2, -e1, 0]
S = [s11 s12 s13; s21 s22 s23; s31 s32 s33]
S = [s11, s12, s13]
[s21, s22, s23]
[s31, s32, s33]

where el ¼ e1 , s11 ¼ σ 11 , etc. Then we simply carry out the matrix operations in
Eq. (3.14):
trace(S.’*U)
ans = e3*s12 - e2*s13 + e1*s23 - e3*s21 - e1*s32 + e2*s31

We can rearrange this answer and set it equal to zero, obtaining

e1 ðσ 23  σ 32 Þ þ e2 ðσ 31  σ 13 Þ þ e3 ðσ 12  σ 21 Þ ¼ 0 (3.15)
which gives
σ 12 ¼ σ 21 σ xy ¼ σ yx
σ 31 ¼ σ 13 or σ zx ¼ σ xz (3.16)
σ 23 ¼ σ 32 σ yz ¼ σ zy

i.e., the shear stresses are symmetric as a result of moment equilibrium, as we stated without
proof in Chapter 2.

3.1.1 Plane Stress

Consider a thin plate with a hole as shown in Fig. 3.2 where the direction of the loading lies
in the x–y plane of the plate and the loading is uniform across the plate thickness. If there are
 3.2 Strength of Materials Solutions 73

t Figure 3.2 A thin plate (thickness, t, is small) with a hole under a uniaxial
tensile load.
y

no body forces in the z-direction, and the front and back of the
x
z plate are stress-free (which means that σ zz ¼ σ xz ¼ σ yz ¼ 0 on
those faces) then since the plate is thin it is reasonable to assume
that σ zz ¼ σ xz ¼ σ yz ¼ 0 throughout the plate. This is an example
of a state of plane stress where the equations of equilibrium, Eq.
(3.11), reduce to
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
(3.17)
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
and where (σxx, σxy, fx, fy) are all functions of ðx; yÞ only. This is an important special case
since, as we will see in Chapter 7, we can solve exactly a number of useful plane-stress
problems (a large plate with a hole, as seen in Fig. 3.2, is one of them).

3.2 Strength of Materials Solutions

The axial loads, bending, and torsion stresses described in Chapter 1 were developed with a
number of assumptions on the deformations and the stresses. Although these stresses were
shown to produce overall force and moment equilibrium, satisfaction of the stress equilib-
rium equations was not addressed. Do those problems satisfy the local equilibrium equations
of Eq. (3.11)? For the axial-load case (Fig. 3.3a) we had only the single stress σ xx ¼ P=A and
that stress was a constant, so it is obvious that the equilibrium equations are satisfied. For
the torsion of a circular rod (Fig. 3.3b) we had shear stresses (Eq. (1.37)):
Tz
σ xy ¼
J
(3.18)
Ty
σ xz ¼
J
Placing these into Eq. (3.11) we see all the derivatives are zero so these torsion stresses also
satisfy local equilibrium. Now, consider the case of pure bending of a beam (Fig. 3.3c),
where the only stress was the bending (flexure) stress given by (Eq. (1.20)):
Mz
σ xx ¼  (3.19)
I yy
Placing this stress into Eq. (3.11) also shows that local equilibrium is satisfied. Finally,
consider the case of engineering beam theory (Fig. 3.4a) where there were both bending and
shear stresses given by (see Eq. (1.21) and Eq. (1.25)):
74 Equilibrium

Figure 3.3 Stresses present for (a) axial loading, z

(b) torsion, and (c) pure bending. Only the y
stresses acting on the front x ¼ constant face of
the element are shown. V xx x
P
A
(a)
z

V xz y

V xy
x
T
(b)
z
y

V xx x
M
(c)

z
z h/2
y
s xz y
x s xx

(a) (b)

Figure 3.4 (a) Stresses in beam bending for engineering beam theory. Only the stresses acting on the x ¼ constant
face of the element are shown. (b) The beam cross-section, showing the quantities needed to compute Q(z).

M y ðxÞz
σ xx ðx; zÞ ¼
I yy
(3.20)
V z ðxÞQðzÞ
σ xz ðx; zÞ ¼
I yy tðzÞ

All of the other stresses were ignored. Let’s examine a rectangular beam where the thickness
t ¼ constant. The beam is unloaded on the front and back faces where σ yx ¼ σ yy ¼ σ yz ¼ 0
 3.2 Strength of Materials Solutions 75

so it is reasonable to assume that these stresses are zero throughout the beam. But the beam
is loaded on the top face (z ¼ constant) in the negative z-direction so that we must keep σ zz .
Thus, the equilibrium equations become:
   
∂ M y ðxÞz ∂ V z ðxÞQðzÞ
þ0 þ ¼0
∂x I yy ∂z I yy t
0 þ0 þ0 ¼0 (3.21)
 
∂ V z ðxÞQðzÞ ∂σ zz
þ0 þ ¼0
∂x I yy t ∂z

The first equation is satisfied because we have dM y =dx ¼ V z ðxÞ and for the rectangular
cross-section (see Fig. 3.4b):
ð
h=2 !
t h2
Qð zÞ ¼ z0 tdz0 ¼  z2
2 4
z (3.22)
dQðzÞ
¼ zt
dz
Thus, we are left with the third equation, which gives
" #
∂σ zz dV z =dx h2
¼ z 2
(3.23)
∂z 2I yy 4

At the top of the beam since we assume the loading is uniform across the thickness of the
beam
σ zz ðx; z ¼ h=2Þ ¼ qðxÞ=t (3.24)

where qðxÞ is the applied downward force/unit length acting along the beam. Consider now
the force equilibrium of a small element of the beam of length dx at a location x in the beam,
as shown in Fig. 3.5. We find
dV z
¼ qðxÞ (3.25)
dx
Placing Eq. (3.25) into Eq. (3.23) and using the fact that
the area moment for the rectangular cross-section is
q ( x)
I yy ¼ th3 =12 we obtain
" #
dVz z ∂σ zz 6qðxÞ h2
Vz ( x ) Vz ( x ) + dx ¼ z 2
(3.26)
dx ∂z th3 4
x

Figure 3.5 A small element of the beam showing the changing

dx shear force and applied force/unit length acting along the beam.
76 Equilibrium

z Figure 3.6 The variation of the normal stress, σ zz , across the cross-section of
h/2 a rectangular beam.

−q ( x ) / t Integrating this equation and using Eq. (3.24) to evaluate the con-
0 stant of integration, we have an explicit expression for the stress σ zz :
s zz " #
6qðxÞ h2 z z3 h3
σ zz ðx; zÞ ¼  þ (3.27)
th3 4 3 12

−h/2 Figure 3.6 shows a plot of the variation of this stress across the cross-
section. Normally, this stress is not discussed
 in a strength
 of materials
 
solution. To see why, note that the maximum value of this stress is ðσ zz Þmax ¼ qmax =t, where
qmax is the
 largestvalue of the applied force/unit length. The maximum bending stress from Eq.
(3.19) is ðσ xx Þmax  ¼ 6M max =th2 , where M max is the largest bending moment in the beam. Thus,
the ratio of these stresses can be written as
 
 ðσ zz Þmax  6qmax L2 h 2
 
ðσ Þ  ¼ M L
(3.28)
xx max max

where L is the total length of the beam. If the coefficient of ðh=LÞ2 is not large, then for
long, slender beams where h=L  1 the bending stress will be much more significant than
the σ zz stress and it is justified to neglect the effects of σ zz in applying engineering
beam theory.

3.3 Force and Moment Equilibrium

When analyzing problems with a strength of materials approach, one often must use force
and moment equilibrium of a small section of the body (see Fig 1.11 or Fig. 3.5 for bending
problems, for example) to describe the changes in the internal forces and moments (see Eq.
(1.23) or Eq. (3.25)). The equilibrium equations of Eq. (3.11) instead involve the changes of
the stresses locally at every point in a body. Clearly, there must be some relationship between
these two different types of equilibrium relations. In this section we will examine that
relationship for axial load, bending, and torsion problems. Let’s start with the uniaxial load
problem where the significant stress is σ xx , which is the stress that appears in the first
equilibrium equation of Eq. (3.11). If we integrate that equation across the cross-sectional
area, A, of the bar (see Fig. 3.7), we find (assuming there are no body forces)
ð ð
d ∂σ yx ∂σ zx
σ xx dA þ þ dA ¼ 0 (3.29)
dx ∂y ∂z
A A

We recognize the first integral in Eq. (3.29) as just the internal force, F x ðxÞ, acting in
the x-direction in the bar. The second integral can be converted into a closed line
 3.3 Force and Moment Equilibrium 77

z Figure 3.7 The geometry of a bar of arbitrary cross-sectional area,

A. The contour, C, encloses A. Also shown is the x-component of
Tx(
n)
n the stress vector, T ðnÞ
x , acting on the surface of the bar, as well as

y the unit normal, n, to that surface.

x
C
integral around the contour, C, of the cross-sectional
area using Gauss’s theorem in two dimensions, which
A gives for any function, g,
ð þ ð þ
∂g ∂g
dA ¼ gny ds, dA ¼ gnz ds (3.30)
∂y ∂z
A C A C

so that Eq. (3.29) becomes

þ
dF x  
þ σ yx ny þ σ zx nz ds ¼ 0 (3.31)
dx
C

The integrand in Eq. (3.31) can be recognized as just the x-component of the stress vector,
T ðxnÞ , acting on the surface of the bar. Let us define a quantity, qx , as
þ
qx ¼ T ðxnÞ ds (3.32)
C

Physically, qx ¼ qx ðxÞ is a force/unit length acting in the x-direction generated by applied

stresses acting on the bar surface. Thus, our integrated equilibrium equation is

dF x
¼ qx (3.33)
dx

If the force/unit length is zero, then the internal axial force is just a constant, as we assumed in
Chapter 1 when describing axial loads. If this force/unit length is not zero then we can use
Eq. (3.33) to find the varying internal force, and , if we let F x ¼ σ xx A, σ xx ¼ Eexx ¼ E∂ux =∂x,
Eq. (3.33) becomes
d ∂ux
EA ¼ qx (3.34)
dx ∂x

which is an equation we can use to find the axial displacement, ux , in the bar.
Now, consider the case of engineering beam theory. In that theory the two most signifi-
cant stresses were a bending stress, σ xx , and a shear stress, σ xz . Since the bending stress again
appears in the first equilibrium equation in Eq. (3.11) let us multiply that equation by z and
integrate over the cross-sectional area of the beam. We obtain (again, with no body forces)
78 Equilibrium

ð ð
d ∂σ yx ∂σ zx
zσ xx dA þ z þz dA ¼ 0 (3.35)
dx ∂y ∂z
A A

and we can recognize the first integral in Eq. (3.35) as the internal bending moment, M y ðxÞ.
We can rewrite the second integral, giving
ð  ð
dM y ∂   ∂
þ zσ yx þ ðzσ zx Þ dA  σ zx dA ¼ 0 (3.36)
dx ∂y ∂z
A A

Since σ zx ¼ σ xz we recognize the last integral in Eq. (3.36) as the internal shear force, V z ðxÞ,
acting in the z-direction in the beam. The other integral can again be transformed into an
integral over the contour, C, involving the stress vector component, T ðxnÞ , so we find
þ
dM y
þ zT ðxnÞ ds  V z ¼ 0 (3.37)
dx
C

However, in engineering beam theory there are usually no distributed loads acting on the
beam in the x-direction so T ðxnÞ ¼ 0 and we have, finally,
dM y
¼ Vz (3.38)
dx
which is the moment–shear force relationship normally seen in beam theory. Now let us
consider the shear force itself by examining the third equilibrium equation in Eq. (3.11) and
integrating it over the cross-sectional area, A. In that case, we obtain
ð ð
d ∂σ yz ∂σ zz
σ xz dA þ þ dA ¼ 0 (3.39)
dx ∂y ∂z
A A

The first integral in Eq. (3.39) is just the shear force, V z ðxÞ, and again Gauss’s theorem can
be used to change the second integral into one over the contour, C, giving
þ
dV z  
þ σ yz ny þ σ zz nz ds ¼ 0 (3.40)
dx
C

where now the integrand is just the stress vector component, T ðznÞ , acting in the z-direction
and we can define the force/unit length in the z-direction, qz ðxÞ, acting along the beam as
þ
T ðznÞ ds ¼ qz (3.41)
C

giving, finally,
dV z
¼ qz (3.42)
dx
 3.3 Force and Moment Equilibrium 79

which is the same as Eq. (3.25) with qz ¼ q since q acts in the negative z-direction.
From the moment–curvature relationship we have M y ¼ EI yy d 2 w=dx2 (see Eq. (1.19))
so taking a x-derivative of Eq. (3.38) and using Eq. (3.42) we find
!
d2 d 2w
EI yy 2 ¼ qz (3.43)
dx2 dx

which can be integrated to find the deflection, w, of the neutral axis. Normally, however,
in elementary mechanics of materials courses the neutral axis deflection is obtained by
first finding explicitly M y ¼ M y ðxÞ in the beam and then integrating the moment–
curvature relationship to find w. In that case only two integrations are required instead
of the four integrations needed with Eq. (3.43) and one can more easily handle applied
concentrated forces and couples acting on the beam since qz cannot be described by
regular functions for those types of applied loads and one must introduce generalized
functions, called singularity functions, in order to use Eq. (3.43). Most elementary
strength of materials texts describe these issues in the determination of beam deflections
so we will not discuss them further here. In the problems of Chapter 1, however, singularity
functions are used to describe the internal shear force and bending moment distributions in
compact forms.
Now, consider the case of torsion. The internal twisting moment, M x ðxÞ, (called the
torque, T, in Chapter 1) is given by
ð
 
M x ¼ yσ xz  zσ xy dA (3.44)
A

Let’s multiply the second equation in Eq. (3.11) (with no body forces) by –z and the third
equation by y, and integrate over the cross-sectional area, A. If we sum the two resulting
equations, we obtain
ð ð 
d   ∂σ yz ∂σ zz ∂σ yy ∂σ zy
yσ xz  zσ xy dA þ y þ z þ dA ¼ 0 (3.45)
dx ∂y ∂z ∂y ∂z
A A

which can be rewritten as

ð 
dM x ∂   ∂ 
þ yσ yz  zσ yy þ yσ zz  zσ zy dA
dx ∂y ∂z
ð A
(3.46)
 
 σ yz  σ zy dA ¼ 0
A

But using the symmetry of the shearing stress and Gauss’s theorem, this becomes
þ þ
dM x    
þ y σ yz ny þ σ zz nz ds  z σ yy ny þ σ zy nz ds ¼ 0 (3.47)
dx
C C
80 Equilibrium

or, equivalently in terms of stress vectors (Eq. (2.12))

þ 
dM x
þ yT ðznÞ  zT ðynÞ ds ¼ 0 (3.48)
dx
C

where the line integral represents a torque/unit length, tx ðxÞ, acting in the x-direction along
the surface of the shaft, so that, finally,
dM x
¼ tx (3.49)
dx
If tx is zero then the internal torque is a constant in the shaft, as seen in Chapter 1. If tx is not
zero then the internal torque and the twist/unit length, dϕ=dx, will be functions of x. From
the relationship between the torque and rate of twist, (see Eq. (1.34)), M x ¼ GJdϕ=dx so that
d dϕ
GJ ¼ tx (3.50)
dx dx

which is an equation that can be used to find the twist, ϕðxÞ, in the shaft.
All of the force and moment equations considered in this section were obtained by simply
integrating the local equations of equilibrium appropriately. All of these equations can be
obtained alternatively by examining equilibrium of small sections such as the one shown in
Fig. 1.11 for bending problems. However, we have wanted to highlight here the connection
between force and moment equilibrium relations and the equations of equilibrium for the
stresses.

3.4 Equations of Equilibrium – Cylindrical and Spherical Coordinates

The equilibrium equations in terms of the Cartesian components of the stress vector and
relationship between those stress vector components and the stresses along cutting planes
whose normals are in the Cartesian coordinate directions were
X
3
∂Tðei Þ
þf ¼0 (3.51)
i¼1
∂xi

and
X
3
Tðei Þ ¼ σ ij ej (3.52)
i¼1

In this section we would like to generalize these equations to other coordinates such as
cylindrical and spherical coordinates. We can do this if we write these equations in a
coordinate-invariant form. This can be done with the aid of the vector identity
rðϕAÞ ¼ ϕrA þ Arϕ (3.53)
 3.4 Equations of Equilibrium – Cylindrical and Spherical Coordinates 81

where ϕ is any scalar and A is any vector. First, consider the divergence of a vector
given by
X
3
∂Bi X
3
¼ rB ¼ r B i ei (3.54)
i¼1
∂xi i¼1

If we let ϕ ¼ Bi and A ¼ ei , where the unit vectors ei are not necessarily constants, then
X
3
∂Bi X
3
¼ ½Bi ðrei Þ þ ðei rÞBi 
i¼1
∂xi i¼1
(3.55)
X
3
¼ ½ðrei Þ þ ðei rÞBi
i¼1

which is a coordinate-invariant form of the divergence for a vector B written in other than
Cartesian coordinates. Now, consider
!
X3
∂Bi X 3
∂ X 3
¼ Bij ej
i¼1
∂xi i¼1
∂xi j¼1
X3 X 3
∂Bij ∂ej
¼ ej þ Bij (3.56)
i¼1 j¼1
∂x i ∂x i

X3 X  
3
∂Bij  
¼ ej þ Bij ej r ej
i¼1 j¼1
∂xi

But from our previous result on the divergence we can write

X
3
∂Bij X
3
¼ ½ðrei Þ þ ðei rÞBij (3.57)
i¼1
∂xi i¼1

so that we have
X
3
∂Bi X
3 X
3      
¼ Bij ej rej þ ej ej r Bij þ Bij ej r ej
i¼1
∂xi i¼1 j¼1

X
3
¼ ½Bi ðrei Þ þ ðei rÞBi  (3.58)
i¼1
X
3
¼ ðrei þ ei rÞBi
i¼1

which is a coordinate-invariant form. If we let Bi ¼ Tðei Þ then the equilibrium equations in

coordinate-invariant form are
X
3
ðrei þ ei rÞTðei Þ þ f ¼ 0 (3.59)
i¼1
82 Equilibrium

Figure 3.8 (a) Cylindrical coordinates. z

(b) Spherical coordinates.
z

r z y y
T T
x x

(a) (b)

We can use Eq. (3.59) for cylindrical coordinates ðr; θ; zÞ (see Fig. 3.8a) by letting
e1  er ¼ cos θex þ sin θey
e2  eθ ¼  sin θex þ cos θey (3.60)
e3  ez

and

Tðe1 Þ ¼ σ rr er þ σ rθ eθ þ σ rz ez
Tðe2 Þ ¼ σ θr er þ σ θθ eθ þ σ θz ez
(3.61)
Tðe3 Þ ¼ σ zr er þ σ zθ eθ þ σ zz ez
f ¼ f r er þ f θ eθ þ f z ez

Using these results and the expression for the gradient in cylindrical coordinates given by
∂ 1 ∂ ∂
r ¼ er þ eθ þ ez (3.62)
∂r r ∂θ ∂z
we find
1 ∂
re1 þ e1 r ¼ þ
r ∂r
1 ∂ (3.63)
re2 þ e2 r ¼ þ
r ∂θ
∂
re3 þ e3 r ¼
∂z
so that the equilibrium equations in terms of the stress vectors are

∂Tðer Þ Tðer Þ 1 ∂Tðeθ Þ ∂Tðez Þ

þ þ þ þf ¼0 (3.64)
∂r r r ∂θ ∂z
 3.4 Equations of Equilibrium – Cylindrical and Spherical Coordinates 83

or, in terms of the stresses

∂σ rr ∂σ rθ ∂σ rz σ σ rθ σ rz 
rr
er þ eθ þ ez þ er þ eθ þ ez
∂r ∂r ∂r r r r

1 ∂σ θr ∂σ θθ ∂σ θz ∂er ∂eθ
þ er þ eθ þ ez þ σ θr þσ θθ (3.65)
r ∂θ ∂θ ∂θ ∂θ ∂θ

∂σ zr ∂σ zθ ∂σ zz
þ er þ eθ þ ez þ ð f r er þ f θ eθ þ f z ez Þ ¼ 0
∂z ∂z ∂z

which gives
∂σ rr σ rr  σ θθ 1 ∂σ θr ∂σ zr
þ þ þ þfr ¼ 0
∂r r r ∂θ ∂z
∂σ rθ 2σ rθ 1 ∂σ θθ ∂σ zθ
þ þ þ þfθ ¼ 0 (3.66)
∂r r r ∂θ ∂z
∂σ rz σ rz 1 ∂σ θz ∂σ zz
þ þ þ þfz ¼ 0
∂r r r ∂θ ∂z
and, as with the Cartesian components of the stress, from moment equilibrium the shear
stresses are symmetric, i.e.,
σ rθ ¼ σ θr , σ zθ ¼ σ θz , σ zr ¼ σ rz (3.67)

We can consider spherical coordinates in exactly the same fashion (see Fig. 3.8b). In that
case we have
e1  er ¼ sin ϕ cos θex þ sin ϕ sin θey þ cos ϕez
e2  eϕ ¼ cos ϕ cos θex þ cos ϕ sin θey  sin ϕez (3.68)
e3  eθ ¼  sin θex þ cos θey

with
∂ 1 ∂ 1 ∂
r ¼ er þ eϕ þ eθ (3.69)
∂r r ∂ϕ r sin ϕ ∂θ

and

Tðe1 Þ ¼ σ rr er þ σ rθ eθ þ σ rϕ eϕ

Tðe2 Þ ¼ σ ϕr er þ σ ϕθ eθ þ σ ϕϕ eϕ
(3.70)
Tðe3 Þ ¼ σ θr er þ σ θθ eθ þ σ θϕ eϕ
f ¼ f r er þ f θ eθ þ f ϕ eϕ

so that the equilibrium equations are of the form

84 Equilibrium

V TT V rT Figure 3.9 The state of stress in polar coordinates for plane stress.
VTr V rr
y

∂Tðer Þ 2Tðer Þ 1 ∂Tðeθ Þ 1 ∂Tðeϕ Þ cos ϕ ðeϕ Þ

þ þ þ þ T þf ¼0
∂r r r sin ϕ ∂θ r ∂ϕ r sin ϕ
r (3.71)
T
x Placing Eq. (3.70) into Eq. (3.71) and using Eq. (3.68), after
some algebra we find

∂σ rr 1 ∂σ rϕ 1 ∂σ rθ 2σ rr  σ ϕϕ  σ θθ þ σ rϕ cot ϕ
þ þ þ þfr ¼ 0
∂r r ∂ϕ r sin ϕ ∂θ r
∂σ rθ 1 ∂σ ϕθ 1 ∂σ θθ 3σ rθ þ 2σ θϕ cot ϕ
þ þ þ þfθ ¼ 0 (3.72)
∂r r ∂ϕ r sin ϕ ∂θ r
 
∂σ rϕ 1 ∂σ ϕϕ 1 ∂σ θϕ 3σ rϕ þ σ ϕϕ  σ θθ cot ϕ
þ þ þ þfϕ ¼ 0
∂r r ∂ϕ r sin ϕ ∂θ r

3.4.1 Plane Stress

A special case of the cylindrical coordinate expressions is when we have a state of
plane stress where σ rz ¼ σ θz ¼ σ zz ¼ 0 and f z ¼ 0. Then the equations of equilibrium
reduce to

∂σ rr σ rr  σ θθ 1 ∂σ θr
þ þ þfr ¼ 0
∂r r r ∂θ
(3.73)
∂σ rθ 2σ rθ 1 ∂σ θθ
þ þ þfθ ¼ 0
∂r r r ∂θ

A state of plane stress is shown in Fig. 3.9. We see that these stresses are described in this
case in terms of the polar coordinates ðr; θÞ only.

3.5 PROBLEMS

P3.1 Consider a long, thick cylinder that is subjected to a constant internal pressure, p0 , and
where there is no axial stress developed, i.e., σ zz ¼ 0 (Fig. P3.1). Under these condi-
tions, because of the symmetry of the loading and the geometry it is reasonable to
assume that there are no shear stresses developed and that the radial and hoop stresses
ðσ rr , σ θθ Þ are functions of r only.
 3.5 Problems 85

(a) If we let
B
σ rr ¼ A 
r2
B
σ θθ ¼ A þ 2
r
σ zz ¼ 0
σ rz ¼ σ rθ ¼ σ θz ¼ 0
where ðA; BÞ are constants, show that the equations of equilibrium in cylindrical
coordinates are satisfied identically.
(b) Determine the constants ðA; BÞ in terms of p0 and the radii ða; bÞ from the known
values of the radial stress at r ¼ a and r ¼ b and the corresponding values of the
radial and hoop stresses.
(c) Where do the largest stresses (in magnitude) occur for the pressurized thick
cylinder? Obtain the largest stresses to first order when the thickness of the
cylinder, t ¼ ðb  aÞ is small. What can you conclude about the relative size of
σ rr and σ θθ for a thin cylinder?
P3.2 Consider a closed thin cylindrical pressure vessel that carries a constant internal
pressure, p0 , as shown in Fig. P3.2a.
(a) Assume the hoop stress, σ θθ , and axial stress, σ zz , are the only two significant
stresses in the cylinder and because the cylinder is thin further assume these two
stresses are constants across the thickness. Use force equilib-
b rium and the free body diagrams of Fig. P3.2b, c to determine
a p0 those stresses. Give an argument for why these are the only
two significant stresses in the cylinder.
p0

Figure P3.1 A thick cylinder carrying a uniform internal pressure, p0 .

Figure P3.2 (a) A closed thin-wall

cylindrical pressure vessel carrying
t r an internal pressure, p0 . (b, c) Free
body diagrams for determining the
p0 V TT V TT hoop stress, σ θθ , and the axial stress,
p0
σ zz .
(a) (b)
t V zz

r
p0

V zz
(c)
86 Equilibrium

(b) How does the hoop stress expression here compare with the result of part (c) in
problem P3.1?
P3.3 Consider a thick wall spherical pressure vessel carrying an internal pressure, p0 , as
shown in Fig. P3.3. Based on the symmetry of the problem and the loading, we expect
that there are no shear stresses developed in the vessel and that σ θθ ¼ σ ϕϕ (why?).
 
Because of the symmetry we also expect that σ rr , σ θθ , σ ϕϕ are all functions of the
radial distance, r, only in spherical coordinates.
(a) If we let
B
σ rr ¼ A 
r3
B
σ θθ ¼ σ ϕϕ ¼ A þ
2r3
σ rθ ¼ σ rϕ ¼ σ θϕ ¼0
where ðA; BÞ are constants, show that the equations of equilibrium in spherical
coordinates, Eq. (3.72), are satisfied identically.
(b) Determine the constants ðA; BÞ in terms of p0 and the radii ða; bÞ from the
known values of the radial stress at r ¼ a and r ¼ b and the corresponding
stresses.
(c) Where do the largest stresses (in magnitude) occur in the sphere? Obtain these
stresses to first order when the thickness of the sphere, t ¼ ðb  aÞ is small. What
can you conclude about the relative size of σ rr , σ θθ ¼ σ ϕϕ for a thin sphere?
P3.4 Consider a thin spherical pressure vessel that carries a constant internal pressure, p0 , as
shown in Fig. P3.4a.
(a) Assume the hoop stress, σ h ¼ σ θθ ¼ σ ϕϕ , is the only
significant stress in the sphere and because the sphere
b p0 is thin further assume this stress is a constant across the
thickness. Use force equilibrium and the free body dia-
a gram of Fig. P3.4b to determine that stress.

Figure P3.3 A thick-wall spherical pressure vessel.

Figure P3.4 (a) A thin-wall spherical pressure

t
vessel. (b) A free body diagram for determining t
the hoop stress, σ h . R R

p0
p0
Vh Vh
(a) (b)
 3.5 Problems 87

z Figure P3.5 A cantilever beam.

z

2h P
y
x
L
t

(b) How does the hoop stress expression here compare with the result of part (c) in
problem P3.3?
P3.5 Consider the cantilever beam with a rectangular cross-section as shown in Fig. P3.5.
Assume the stresses in the beam are given by
σ xx ¼ a xz
σ xz ¼ b þ cz2
σ xy ¼ σ yz ¼ σ yy ¼ σ zz ¼ 0

(a) Determine the relationship between the constants ða; b; cÞ so that the equations of
equilibrium are satisfied.
(b) Determine the relationship between the constants such that the top and bottom and
front and back faces of the beam are stress-free.
(c) If the shear stress acting on the left face of the beam produces the load P, determine
all the stresses in terms of this load and show that these stresses are consistent with
engineering beam theory.
4 Strain

Just as the concept of stress gives us a measure of force distributions in a deform-

able body, the concept of strain describes the deformations locally at every point
within the body. In this chapter we will define strains and describe how strains
change with directions and with the choice of coordinates, as we have done with
stresses. Strains will also be related to the displacements of the deformable body.
Since the strains often found in practice are quite small, we will also consider the
behavior of small strains and their relationship to displacements in this important
case. In fact, this book will consider problems only for small strains. It will be
shown that strains must satisfy a set of compatibility equations to ensure that they
represent a well-behaved deformation of a body. As discussed in Chapter 1, these
compatibility equations form one of the four “pillars” of analyzing deformable
bodies. Finally, we will examine strains and strain–displacement relations in cylin-
drical coordinates.

4.1 Definitions of Strains

As briefly discussed in Chapter 1, locally we can describe the deformations of a body in
terms of changes of length and changes of angle for a small element (see Fig. 1.16). Here, we
want to analyze this local deformation in some detail and define the strains that characterize
those changes.

4.1.1 Normal Strain

When a body is loaded and deforms, elements of that body elongate or contract and are
distorted. The quantity that we use to measure the changes of length occurring in the body is
called the normal strain. Consider, for example, a small directed line segment, drn , between
two very close points P and Q in an undeformed body (Fig. 4.1a). Let the length of this line
segment be dsn and its direction, n, where n is a unit vector. When the body deforms, point P
will move to a new point, P∗ , and Q will move to Q∗ and the original line segment will
change to a new segment, dRn , of length dS n (see Fig. 4.1b). The normal strain, E nn , at point
P is then defined as

88
 4.1 Definitions of Strains 89

Figure 4.1 (a) A line element PQ in the

undeformed deformed direction n in an undeformed body. (b) The
body body corresponding line element PQ in the
n = drn / dsn deformed body.

dS n
dsn Q P*
drn dR n Q*
P

(a) (b)

1 dS 2n  ds2n
E nn ¼
2 ds2n
1 dRn dRn  drn drn
¼ (4.1)
2 ds2n


!
1

dRn

2
¼ 1
2
ds

This definition can be used regardless of the size of the strain so that it is a general definition.
However, if the changes of length occurring in the body are quite small, which often occurs
in practice, then this normal strain reduces to the normal small strain, enn , defined as the
relative elongation of the element, i.e.,
1 ðdS n  dsn ÞðdS n þ dsn Þ
E nn ¼
2 ds2n
(4.2)
dS n  dsn
ffi enn ¼
dsn
which is the usual definition of strain seen in elementary strength of materials texts.

4.1.2 Shear Strain

In addition to changes of length occurring in a deformable body, there are distortions
(corresponding to changes in angles) which also must be characterized. The shear strain is
the quantity we use to define these changes. To obtain the shear strain, we consider two
small line segments ðdrn , drt Þ in the undeformed body along the n and t unit vector
directions which are initially at right angles to each other (Fig. 4.2a). In the deformed
body, these line segments, ðdRn , dRt Þ, will no longer be orthogonal. We will let θnt be the
new angle between the two distorted line segments. The shear strain, E nt , will then be
defined as
90 Strain

Figure 4.2 (a) Two small line PQ undeformed

and PR segments that are body deformed
orthogonal in an undeformed body. body
(b) The corresponding line segments n = drn / dsn R*
P*Q* and P*R* in a deformed body
t = drt / dst dR t
dS n
that are no longer orthogonal. dst dsn
Q P * nt dSt
R dR n
drt drn Q*
P

(a) (b)

1 dRn dRt
E nt ¼  (4.3)
2 dsn dst
This shear strain is related to the angular distortions occurring in the body since the cosine of
the angle between any two unit vectors is just the dot product of those unit vectors. Thus,
dRn dRt
π  

dsn

dst

cos θnt ¼ sin  θnt ¼



(4.4)
2
dRn

dRt


ds

ds

n t

Using the definitions of the normal and shear strains for the n and t directions, then
π  2E nt
sin  θnt ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (4.5)
2 1 þ 2E nn 1 þ 2E tt
which shows that if θnt ¼ π=2 (so that there is no angular distortion of these two lines) then
E nt ¼ 0. When the changes of lengths and changes of angles are small then
π  π 
sin  θnt ffi  θnt
2 2 (4.6)
E nn  1, E tt  1
so that
1 π 
E nt ffi ent ¼  θnt (4.7)
2 2
where ent is the small shear strain. It is common in engineering studies to use the engineering
shear strain, γnt , in place of ent , where
π
γnt ¼ 2ent ¼  θnt (4.8)
2
The engineering shear strain is just equal to the change of angle so that it has a direct
physical meaning. However, we will see that in strain transformations it is more convenient
to use ent , which is also called the tensor shear strain.
 4.2 Strain–Displacement Relations and Strain Transformations 91

4.2 Strain–Displacement Relations and Strain Transformations

Although the normal strains and shear strains define the deformation locally everywhere in a
body, even knowing these strains it is not easy to directly determine what shape the
deformed body actually takes. A more convenient way to define the deformation is to
describe the displacement vector, uðx1 , x2 , x3 Þ, for every point P ¼ ðx1 , x2 , x3 Þ in the
undeformed body. Obviously, the strains are related to these displacements so we need to
determine these strain–displacement relations. From the geometry (Fig. 4.3) we see that

Rðx1 , x2 , x3 Þ ¼ r þ uðx1 , x2 , x3 Þ
X
3 (4.9)
¼ xi ei þ uðx1 , x2 , x3 Þ
i¼1

so that if r changes by a small amount drn in the n direction, we have

dRn ¼ drn þ du (4.10)

or, dividing by the length of drn , dsn , we find

dRn drn du
¼ þ
dsn dsn dsn
(4.11)
du
¼nþ
dsn

Using this result in the definition of the normal strain, Eq. (4.1), gives
 
1 dR dR
E nn ¼  1
2 dsn dsn
(4.12)
du 1 du du
¼ n þ 
dsn 2 dsn dsn

undeformed Figure 4.3 Deformations of a body defined by the

body u ( x1 , x2 , x3 ) displacement vector, u, at any point in the body.
( 1 , x2 , x3 )
x deformed
body
x2
P P*
r
e2
R ( x1 , x2 , x3 )
e1 x1
e3
x3
92 Strain

Similarly, if we let r change by a small amount drt in the t direction

dRt drt du
¼ þ
dst dst dst
(4.13)
du
¼tþ
dst
so, from Eq. (4.3),
1 dRn dRt
E nt ¼ 
2 dsn dst
  (4.14)
1 du du du du
¼ t þ n þ 
2 dsn dst dsn dst

If the strains are small enough so that we can neglect the products of the displacement
gradients in these definitions, then we see that for the small normal and shear strains we have
du
enn ¼ n
dsn
  (4.15)
1 du du
ent ¼ n þ t
2 dst dsn

Since Eq. (4.14) and Eq. (4.15) are valid for any directions n and t, we can compute both the
finite and small strains for line segments along the Cartesian ðx1 , x2 , x3 Þ ¼ ðx; y; zÞ axes by
simply making particular choices for n and t. For example, consider letting n be along the
x1 -axis direction and let t be along the x2 -axis direction. Then n ¼ e1 , t ¼ e2 so that
∂u 1 ∂u ∂u
E xx ¼ E 11 ¼ e1  þ 
∂x1 2 ∂x1 ∂x1
∂u
exx ¼ e11 ¼ e1 
∂x1
  (4.16)
1 ∂u ∂u ∂u ∂u
E xy ¼ E 12 ¼ e1  þ e2  þ 
2 ∂x2 ∂x1 ∂x1 ∂x2
 
1 ∂u ∂u
exy ¼ e12 ¼ e1  þ e2 
2 ∂x2 ∂x1

In exactly the same fashion, we can obtain the other normal strains and shear strains. We
can, in fact, write all the normal and shear strains in the same forms as
 
1 ∂u ∂u ∂u ∂u
E ij ¼ ei  þ ej  þ  ði; j ¼ 1; 2, 3Þ
2 ∂xj ∂xi ∂xi ∂xj
  (4.17)
1 ∂u ∂u
eij ¼ ei  þ ej  ði; j ¼ 1; 2, 3Þ
2 ∂xj ∂xi
 4.2 Strain–Displacement Relations and Strain Transformations 93

These results are still in vector notation. However, if we write the displacement vector u and
its gradients in terms of scalar components along ðx1 , x2 , x3 Þ axes, i.e.,
X
3
u¼ ui ei
i¼1
(4.18)
∂u X 3
∂ui
¼ ei ðj ¼ 1; 2, 3Þ
∂xj i¼1
∂xj

then the strains can be written directly in terms of these components as

!
1 ∂ui ∂uj X 3
∂uk ∂uk
E ij ¼ þ þ  ði; j ¼ 1; 2, 3Þ
2 ∂xj ∂xi k¼1 ∂xi ∂xj
(4.19)
 
1 ∂ui ∂uj
eij ¼ þ ði; j ¼ 1; 2, 3Þ
2 ∂xj ∂xi

These nine strains associated with the Cartesian coordinate directions are actually
only six independent strains since the strains are inherently symmetric by their defin-
itions, i.e., E 12 ¼ E 21 , e12 ¼ e21 , etc. These strains define the state of strain at any point
in the body since they can be used to obtain the normal and shear strains in any
directions at a point, just as the Cartesian components of stress allow us to obtain any
normal or shear stresses at a point. We can write the states of finite or small strains in
matrix form as
2 3 2 3
E 11 E 12 E 13 e11 e12 e13
½E  ¼ 4 E 21 E 22 E 23 5 ½e ¼ 4 e21 e22 e23 5 (4.20)
E 31 E 32 E 33 e31 e32 e33

Consider finding a finite normal strain E nn from the state of strain ½E . Since
X
3
drn X 3
dxi
n¼ ni ei ¼ ¼ ei
i¼1
dsn i¼1
dsn
! (4.21)
du X 3
dui X3 X3
∂ui dxj X3 X 3
∂ui
¼ ei ¼ ei ¼ nj ei
dsn i¼1
dsn i¼1 j¼1
∂xj dsn i¼1 j¼1
∂xj

we find (see Eq. (4.12))

du 1 du du
E nn ¼ n þ 
dsn 2 dsn dsn
! (4.22)
X3 X 3
∂ui 1 X3
∂uk ∂uk
¼ þ ni nj
i¼1 j¼1
∂xj 2 k¼1 ∂xi ∂xj
94 Strain

Note that we can write

   
∂ui 1 ∂ui ∂uj 1 ∂ui ∂uj
¼ þ þ 
∂xj 2 ∂xj ∂xi 2 ∂xj ∂xi (4.23)
¼ eij þ ωij

where eij are the small strain components and ωij can be shown to be local average
rotations in the body [1]. But the matrix of ωij terms is antisymmetric, i.e., ωij ¼ ωji so
that
3 X
X 3
ωij ni nj ¼ 0 (4.24)
i¼1 j¼1

giving
X
3 X
3
∂ui X
3 X
3
ni nj ¼ eij ni nj (4.25)
i¼1 j¼1
∂xj i¼1 j¼1

This means that we can rewrite Eq. (4.22) as

!
X3 X 3
1 ∂ui ∂uj X3
∂uk ∂uk
E nn ¼ þ þ ni nj
i¼1 j¼1
2 ∂xj ∂xi k¼1 ∂xi ∂xj
(4.26)
X
3 X
3
¼ E ij ni nj
i¼1 j¼1

which shows that we can find the normal strain in any direction from a knowledge of the
state of strain. In matrix–vector form we have, equivalently,

E nn ¼ fngT ½E fng (4.27)

We can follow the same steps for the finite shear strain, E nt , to find
!
X3 X 3
1 ∂ui ∂uj X 3
∂uk ∂uk
E nt ¼ þ þ ni nj
i¼1 j¼1
2 ∂xj ∂xi k¼1 ∂xi ∂xj
(4.28)
X
3 X
3
¼ E ij ni tj
i¼1 j¼1

or

E nt ¼ fngT ½E ftg (4.29)

When the strains are small, we can simply neglect the product of the displacement deriva-
tives and find, analogously
 4.2 Strain–Displacement Relations and Strain Transformations 95

3 X
X 3  
1 ∂ui ∂uj
enn ¼ þ ni nj
i¼1 j¼1
2 ∂xj ∂xi
X
3 X
3
¼ eij ni nj
i¼1 j¼1
  (4.30)
X
3 X
3
1 ∂ui ∂uj
ent ¼ þ ni tj
i¼1 j¼1
2 ∂xj ∂xi
X
3 X
3
¼ eij ni tj
i¼1 j¼1

or, in matrix–vector notation again,

enn ¼ fngT ½efng
(4.31)
ent ¼ fngT ½eftg
If we compare Eq. (4.31) with our previous results for stresses (Eq. (2.29)) we see that they
are identical in form if we simply make the replacements σ nn , σ nt ! enn , ent , and σ ij ! eij (or
equivalently for finite strains σ nn , σ nt ! E nn , E nt and σ ij ! E ij ). Thus, our previous equations
involving stress transformations, principal stresses, Mohr’s circle, etc., remain valid for
strains as well. For example, using these replacements we can write the strain transformation

equations relating the small strains e0ij along a set of x01 ; x02 ; x03 axes to the small strains eij
along a set of ðx1 , x2 , x3 Þ axes through the direction cosines as
3 X
X 3
e0mn ¼ l im l jn eij (4.32)
i¼1 j¼1

or, in matrix–vector notation,

½ e0  ¼ ½ l  T ½ e ½ l  (4.33)

Similarly, the principal strains, ep , satisfy the cubic equation (see Eq. (2.58) for the stresses)
e3p  J 1 e2p þ J 2 ep  J 3 ¼ 0 (4.34)

in terms of the strain invariants

J 1 ¼ exx þ eyy þ ezz
J 2 ¼ exx eyy þ exx ezz þ eyy ezz  e2xy  e2xz  e2yz (4.35)
J 3 ¼ exx eyy ezz þ 2exy exz eyz  exx e2zy  eyy e2xz  ezz e2xy

which can be expressed in terms of the principal strains as

J 1 ¼ ep1 þ ep2 þ ezp3
J 2 ¼ ep1 ep2 þ ep1 ep3 þ ep2 ep3 (4.36)
J 3 ¼ ep1 ep2 ep3
96 Strain

Example 4.1 Strain Calculations

Consider the following state of small strains with respect to the ðx; y; zÞ axes as defined in
MATLAB®:
strain = [2670 2670 -330; 2670 2000 -1500; -330 -1500 -2000]
strain = 2670 2670 -330
2670 2000 -1500
-330 -1500 -2000

The strains here are given in microstrain (μstrain, or μ), i.e., e11 ¼ 2670 μ ¼ 2760  106 ,
etc. If this state of strain is uniform throughout a region of a body such as the parallelepiped
shown in Fig. 4.4, then we can analyze the strains in different directions by examining the
behavior of finite lines in the body rather than the infinitesimally small lines considered
previously in defining the strains since every portion of the finite lines will experience the
same strains. First, calculate the normal strain experienced by the line EA in Fig. 4.4. Thus,
we need to define a unit normal in the EA direction. In MATLAB® we can form up this unit
normal as a column vector, n, in two steps:
EA = [0 -1 -2];
n = EA'/norm(EA)
n = 0
-0.4472
-0.8944

Then we can calculate the normal strain, enn, in this normal direction with Eq. (4.31) as
enn = n'*strain*n
enn = -2400

which we can write as enn ¼ 2400 μ. Now suppose we want

z
to find the engineering shear strain experienced by lines EA
G F and EF in Fig. 4.4. These two lines are orthogonal to each
other, as required by our definition of shear strain. Define a
D E column unit vector t along EF, which is very simple in this
t
2.0 case, and then use it in conjunction with the unit vector n in
n Eq. (4.31):

C t = [-1 0 0]';
O
y
ent = n'*strain*t
1.5 ent = 898.8993
A B
1.0
Figure 4.4 A parallelepiped in a region of a body where there is a
x
uniform (constant) state of strain.
 4.2 Strain–Displacement Relations and Strain Transformations 97

The engineering shear strain for this pair of lines, therefore, is

engstrain = 2*ent
engstrain = 1.7978e+03

which we can write as γnt ¼ 1798 μ. If we want to find the principal strains and principal
directions for the state of strain we have been considering, then we can simply use the
MATLAB® function eig as done with the stresses:
[pdirs, pvals] = eig(strain)
pdirs = 0.1356 0.6807 0.7199
-0.3794 -0.6356 0.6724
-0.9152 0.3643 -0.1721
pvals = 1.0e+03 *
-2.5729 0 0
0 0.0003 0
0 0 5.2426

where the principal direction unit vectors are in the columns of the matrix pdirs and the
principal strains ep1 ffi 2573 μ, ep2 ffi 0:3 μ, ep3 ffi 5243 μ are the diagonal values in pvals.
Recall that the principal directions must be checked to see if they form a right-handed
system by evaluating the determinant of the pdirs matrix. (In this case the determinant is –1
so the system is not right-handed.) One of the principal strains is almost zero so Mohr’s
circles for this case is as shown in Fig. 4.5. The largest tensor shear strain is just equal to the
radius of the largest circle so the maximum engineering shear strain is just the diameter of
this circle, which is the difference of the largest and the smallest principal strains:
maxshear = pvals(3,3) - pvals(1,1)
maxshear = 7.8155e+03

so we have jγmax j ffi 7816 μ.

ent Figure 4.5 The Mohr’s circle construction for the strains in
Example 4.1.
γ max / 2

–2573 0.3 5243 enn

98 Strain

Figure 4.6 Changes of a small volumetric element in terms of Dx'

the small strains present. Dy'

ezz Dz Dz'

Dz
Dy eyy Dy
Dx

exx Dx

4.2.1 Dilatation
Since small normal strains are measures of changes of length, they are also related to the
change of volume occurring in a deformable body. We have, from the definitions of the
strains (see Fig. 4.6):
Δx0 ffi Δx þ exx Δx
Δy0 ffi Δy þ eyy Δy (4.37)
0
Δz ffi Δ þ ezz Δz

so that to first order (neglecting products of the small strains)


Δx0 Δy0 Δz0 ¼ ΔxΔyΔz þ exx þ eyy þ ezz ΔxΔyΔz (4.38)

For a small element V ¼ ΔxΔyΔz the relative change of volume of that element is defined as
the dilatation, Δ, given by
ΔV
Δ  ¼ exx þ eyy þ ezz
V
(4.39)
∂ux ∂uy ∂uz
¼ þ þ
∂x ∂y ∂z

Since the sum of the normal strains in Eq. (4.39) is an invariant, the dilatation is
independent of the orientation of the Cartesian coordinate system used to describe
the deformation.

4.2.2 Plane-Strain Principal Strains

The special case of plane strain is where the displacements are given as ux ¼ ux ðx; yÞ,
uy ¼ uy ðx; yÞ, uz ¼ 0, which produces a state of strain where exx ¼ exx ðx; yÞ, eyy ¼ eyy ðx; yÞ,
exy ¼ exy ðx; yÞ, exz ¼ eyz ¼ ezz ¼ 0. In this case the third strain invariant J 3 ¼ 0 and the
principal strains are given explicitly from Eq. (4.34) as
 4.3 Strain Compatibility Equations 99

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
e  e 
exx þ eyy xx yy
ep1 , ep2 ¼ þ e2xy
2 2 (4.40)
ep3 ¼ 0

which are relations for the strains directly analogous to those for the stresses in the case of
plane stress, as shown in Chapter 2.

4.3 Strain Compatibility Equations

We know that stresses in a deformable body must satisfy local equilibrium equations
everywhere in the body. Likewise, we will find in this section that strains also must satisfy
a set of compatibility equations within the body. To see this, consider a body with displace-
ments at points P1 and P2 given by u1 and u2 , respectively (see Fig. 4.7). Let Δu ¼ u2  u1 ,
which is the relative displacement of P2 with respect to P1. Breaking that relative displace-
ment into components, we have
X
3
Δu ¼ Δui ei (4.41)
i¼1

Then these components can be written in terms of integrals over some path C from P1 to P2
(Fig. 4.7) in terms of the local small strains and rotations as
Pð2
3 ð 3 ð 
P2 P2
X ∂ui X
Δui ¼ dui ¼ dxj ¼ eij þ ωij dxj (4.42)
j¼1
∂xj j¼1
P1 P1 P1

But we have

 X3
∂ωij
ωij dxj ¼ d ωij xj  xj dxk (4.43)
k¼1
∂xk

so that Eq. (4.42) becomes

!
3 ð
P2 P2
X X X
jþ
3 3
∂ωij
Δui ¼ ωij xj eik  xj dxk (4.44)
j¼1 P1 j¼1 j¼1
∂xk
P1

P2
C u2

P1 Figure 4.7 Two points, P1 and P2, in a deformable body where the
u1 displacement vector at those points are u1 and u2 , respectively, and an
integration path, C, between those two points.
100 Strain

Through differentiation of the strain–displacement relations, it can be verified that these

derivatives of the local rotation can also be written in terms of derivatives of the strains, i.e.,
 
∂eik ∂ejk 1 ∂ ∂ui ∂uj ∂ωij
 ¼  ¼ (4.45)
∂xj ∂xi 2 ∂xk ∂xj ∂xi ∂xk

giving

3 ð
P2 P2
X X
jþ
3
Δui ¼ ωij xj ξ ik dxk
j¼1 P1 j¼1
P1
(4.46)
X
3  
∂eik ∂ejk
where ξ ik ¼ eik  xj 
j¼1
∂xj ∂xi

The first term in the above Δui expression only depends on the points P1 and P2. The
second term is independent of the path C between these two points (and hence also
only depends on P1 and P2) if, for a simply connected region (a region with no holes),
we have
∂ξ ik ∂ξ il
¼ (4.47)
∂xl ∂xk
(see, for example, Wylie, C. R., Advanced Engineering Mathematics, 4th edn, McGraw-
Hill, p. 684; see also[2]). Thus, if Eq. (4.47) is satisfied everywhere in a simply con-
nected region, the displacement will have a single value everywhere in that region. (The
displacement is single-valued but not unique since we can always add a rigid body
displacement that does not change the strains.) The derivatives contained in Eq. (4.47)
can be written as
  X 3  2 
∂ξ ik ∂eik ∂eik ∂elk ∂ eik ∂2 ejk
¼    xj 
∂xl ∂xl ∂xl ∂xi j¼1
∂xj ∂xl ∂xi ∂xl
  X  2  (4.48)
3
∂ξ il ∂eil ∂eil ∂ekl ∂ eil ∂2 ejl
¼    xj 
∂xk ∂xk ∂xk ∂xi j¼1
∂xj ∂xk ∂xi ∂xk

However, the first two terms on the right-hand side of Eq. (4.48) cancel and the third terms,
when placed back into Eq. (4.47), also cancel, so Eq. (4.47) reduces to
X3  2 
∂ eik ∂2 ejl ∂2 ejk ∂2 eil
xj þ   ¼0 (4.49)
j¼1
∂xj ∂xl ∂xi ∂xk ∂xi ∂xl ∂xj ∂xk

Since the xj are independent, the quantity in brackets in Eq. (4.49) must vanish, and

∂2 eik ∂2 ejl ∂2 ejk ∂2 eil

þ   ¼0 (4.50)
∂xj ∂xl ∂xi ∂xk ∂xi ∂xl ∂xj ∂xk
 4.3 Strain Compatibility Equations 101

We will write Eq. (4.50) symbolically as

Rijkl ¼ 0 (4.51)
Since i, j, k, l can all have values ranging from 1 to 3, Eq. (4.51) looks like a total of 81
equations. However, because of the following symmetries and antisymmetries:
Rijkl ¼ Rklij
(4.52)
Rijkl ¼ Rjikl ¼ Rijlk
we can choose only six distinct terms in Eq. (4.51) given by
S 11  R2323 ¼ 0
S 22  R3131 ¼ 0
S 33  R1212 ¼ 0
(4.53)
S 21 ¼ S 12  R2331 ¼ 0
S 23 ¼ S 32  R3112 ¼ 0
S 13 ¼ S 31  R1223 ¼ 0
which we have written in terms of the components of a symmetric S matrix:
2 3
S 11 S 12 S 13
4 S 21 S 22 S 23 5 (4.54)
S 31 S 32 S 33
All of the equations in Eq. (4.53) are not independent since the components of the S matrix
can be shown to satisfy the three equations:
X
3
∂S ij
¼ 0 ði ¼ 1; 2, 3Þ (4.55)
j¼1
∂xj

The six equations in Eq. (4.53) are called the strain compatibility equations. Because of Eq.
(4.55) these compatibility equations represent only three independent conditions that the
strains must satisfy. Explicitly, these six equations are
∂2 e23 ∂2 e22 ∂2 e33
2   ¼0
∂x2 ∂x3 ∂x23 ∂x22
∂2 e31 ∂2 e33 ∂2 e11
2   ¼0
∂x3 ∂x1 ∂x21 ∂x23
∂2 e12 ∂2 e11 ∂2 e22
2   ¼0
∂x1 ∂x2 ∂x22 ∂x21
(4.56)
∂2 e33 ∂2 e12 ∂2 e23 ∂2 e31
þ   ¼0
∂x1 ∂x2 ∂x23 ∂x3 ∂x1 ∂x3 ∂x2
∂2 e11 ∂2 e23 ∂2 e31 ∂2 e12
þ   ¼0
∂x2 ∂x3 ∂x21 ∂x1 ∂x2 ∂x1 ∂x3
∂2 e22 ∂2 e31 ∂2 e12 ∂2 e23
þ   ¼0
∂x3 ∂x1 ∂x22 ∂x2 ∂x3 ∂x2 ∂x1
102 Strain

4.3.1 Plane-Strain and Plane-Stress Compatibility Equations

Because the six compatibility equations are not independent, it is difficult in general 3-D
problems to use these equations directly. However, for plane-strain problems, recall we have
only two nonzero displacements because plane-strain conditions require
u1 ¼ u1 ðx1 , x2 Þ
u2 ¼ u2 ðx1 , x2 Þ (4.57)
u3 ¼ 0
which implies that
e33 ¼ e13 ¼ e23 ¼ 0
e11 ¼ e11 ðx1 , x2 Þ
(4.58)
e22 ¼ e22 ðx1 , x2 Þ
e12 ¼ e12 ðx1 , x2 Þ
so there is only one compatibility equation that is not identically zero, given by
∂2 e12 ∂2 e11 ∂2 e22
2   ¼0 (4.59)
∂x1 ∂x2 ∂x22 ∂x21
Recall that, for plane-stress problems, we assume
σ 13 ¼ σ 23 ¼ σ 33 ¼ 0
σ 11 ¼ σ 11 ðx1 , x2 Þ
(4.60)
σ 22 ¼ σ 22 ðx1 , x2 Þ
σ 12 ¼ σ 12 ðx1 , x2 Þ
In the next chapter when we discuss stress–strain relations, we will see that for a linearly
elastic isotropic material the condition of plane stress implies that the corresponding condi-
tions on the strains are:
e13 ¼ e23 ¼ 0 but e33 ¼
6 0
e11 ¼ e11 ðx1 , x2 Þ
(4.61)
e22 ¼ e22 ðx1 , x2 Þ
e12 ¼ e12 ðx1 , x2 Þ
so that the compatibility equations reduce to
∂2 e33
¼0
∂x22
∂2 e33
¼0
∂x21
(4.62)
∂2 e33
¼0
∂x1 ∂x3
∂2 e12 ∂2 e11 ∂2 e22
2   ¼0
∂x1 ∂x2 ∂x22 ∂x21
 4.3 Strain Compatibility Equations 103

Plane-stress problems usually involve thin bodies such as plates where the x3 -direction is
normal to the plate and where the e33 strains across the thickness of the plate are controlled
by the in-plane normal strains, ðe11 , e22 Þ, as we will see in the next chapter. Thus, the first
three equations in Eq. (4.62) typically cannot be satisfied and plane-stress conditions can
only be approximately true in such a body. However, the behavior of the out-of-plane
displacement and strain of a thin body such as a plate are usually not of great interest
anyway so that in solving plane-stress problems, one usually ignores the first three equations
in Eq. (4.62), leading to the single plane-stress compatibility equation for the strains:
∂2 e12 ∂2 e11 ∂2 e22
2   ¼0 (4.63)
∂x1 ∂x2 ∂x22 ∂x21

which is identical to the plane-strain compatibility equation, Eq. (4.59). If this single
compatibility equation is satisfied, we expect that in both plane-stress and plane-strain
problems we should be able to integrate the in-plane strains ðe11 , e22 , e12 Þ to find the in-
plane displacements ðu1 , u2 Þ.

4.3.2 Multiply Connected Bodies

For bodies with holes the compatibility equations are not sufficient by themselves to
guarantee that the strains can be integrated to find a single-valued displacement field. In
fact, in nonsimply connected bodies there may be some cases where we want displacements
that are not single-valued. The split ring shown in Fig. 4.8, for example, is an example where
the point P on one side of the split is fixed (displacement ¼ 0) while the displacement of the
same point P on the other side of the slit has a value of D. In this case we see that integration
paths like C 1 that do not enclose the hole yield single-valued displacements while paths such
as C 2 do not, i.e.,

Figure 4.8 A multiply connected body where the

C1 displacement at point P has multiple values.

x2

P P x1
D

C2
104 Strain

ð
duj ¼ 0 ðj ¼ 1; 2, 3Þ
C1
ð (4.64)
duj ¼ Dδj1 ðj ¼ 1; 2, 3Þ
C2

in terms of the Kronecker delta symbol, δji , (see Eq. (2.34)). To ensure that a multiply
connected body such as the ring cannot split in this manner, we must supplement the
compatibility equations by additional conditions. For a body with m holes such as that
shown in Fig. 4.9 if, in addition to satisfying the compatibility equations, we also satisfy the
m subsidiary conditions
ð
duj ¼ 0 ðj ¼ 1; 2, 3Þ ði ¼ 1; 2; . . . ; mÞ (4.65)
Ci

where the integrals are taken around each hole, then the displacements will be single-valued.
Note that, in solving any problem, if we end up obtaining a displacement field that
represents the deformation of the body, then compatibility is not an issue since we can
always obtain a set of strains that are compatible with those displacements simply by taking
the appropriate displacement derivatives. However, when we directly solve a problem only
for the strains (or corresponding stresses), it is not automatically guaranteed that a well-
behaved single-valued displacement field can be obtained by integrating those strains. The
compatibility equations (and any subsidiary conditions needed for multiply connected
bodies) provide the guarantee that a well-behaved, single-valued displacement field can be
obtained from the strain field. Later in this book we see examples of both types of problems.

4.4 Satisfaction of Compatibility – Strength of Materials Solutions

In Chapter 3 we examined if the elementary strength of materials problems of extension,
bending, and torsion satisfied the local equilibrium equations. In this section we want to
examine the strains in those problems and see if the strain compatibility equations are satisfied.

Axially Loaded Members

The strain field described in Chapter 1 for axially loaded
members subjected to an end load P acting through the
C1 C2 Cm
centroid of the cross-sectional area had the constant strain
components

Figure 4.9 A multiply connected body with m holes.

 4.4 Satisfaction of Compatibility – Strength of Materials Solutions 105

P
exx ¼
AE
eyy ¼ ezz ¼ νexx (4.66)

exy ¼ exz ¼ eyz ¼ 0

so that compatibility is satisfied for these strains. If we use this theory and apply it to bars of
varying geometry or Young’s modulus and/or distributed axial loads acting on the bar
surface, then we have instead
F x ðxÞ
exx ¼
AðxÞE ðxÞ
eyy ¼ ezz ¼ νexx (4.67)

exy ¼ exz ¼ eyz ¼ 0

where F x ðxÞ is the varying internal axial force in the bar. In this case the compatibility
equations give

d 2 eyy
¼0
dx2
(4.68)
d 2 ezz
¼0
dx2
which generally are not satisfied exactly except in very special cases. However, if the
derivative of exx is only a slowly varying function of x then the derivative of the lateral
strains will also be slowly varying in x and these compatibility equations can be
approximately satisfied.

Torsion of Circular Shafts

In the torsion problem discussed in Chapter 1, where a circular shaft is subjected to an end-
twisting moment, T 0 , the strains present, based on the deformation assumptions, were (see
Eq. (1.32))
exx ¼ eyy ¼ ezz ¼ 0
1
exy ¼  ϕ0 z
2 (4.69)
1 0
exz ¼ ϕ y
2
eyz ¼ 0

in terms of the constant twist/unit length dϕ=dx ¼ ϕ0 ¼ T 0 =GJ. Since these strains are linear
functions of the coordinates and the compatibility equations involve only second derivatives,
compatibility is satisfied. However, if we attempt to use this theory for twisting of a circular
106 Strain

bar with varying cross-section or shear modulus and/or distributed torques acting along the
length of the shaft, we will have ϕ0 ¼ T ðxÞ=G ðxÞJ ðxÞ, where T ðxÞ is the internal torque.
Then the compatibility equations yield

∂2 exy 1 00
¼ ϕ ¼0
∂x∂z 2
(4.70)
∂2 exz 1 00
¼ ϕ ¼0
∂x∂y 2
00
where ϕ ¼ d 2 ϕ=dx2 . Equation (4.70) can be satisfied approximately if the twist/unit length is
slowly varying along the length of the shaft.

Beam Bending
Now, consider engineering beam theory, which includes pure bending as a special case. In
this theory the strains obtained in Chapter 1 were

M y ðxÞz
exx ¼
EI yy
eyy ¼ ezz ¼ νexx
(4.71)
exy ¼ eyz ¼ 0
1 V z ðxÞQðzÞ
exz ¼
2 GI yy tðzÞ

In discussing such problems, however, we want to change the coordinates (from those
used in Chapter 1) so that the vertical axis is the y-axis instead of the z-axis (see Fig
4.10a) and the internal moment on a right-hand cut is taken along the positive z-axis
while the internal shear force is positive on a right-hand cut in the positive y-direction, as
shown in Fig. 4.10b. This coordinate change will facilitate our later comparisons with
more detailed plane-stress solutions to such problems. With these changes the strains can
be written as

Figure 4.10 (a) Coordinates used for y

beam problems in this section. (b) The
positive directions taken for the internal
shear force and bending moment on a Mz
right-hand cut taken in the beam. x
A A Vy
B
(a) (b)
 4.4 Satisfaction of Compatibility – Strength of Materials Solutions 107

M z ðxÞy
exx ¼ 
EI zz
eyy ¼ ezz ¼ νexx
(4.72)
exz ¼ eyz ¼ 0
1 V y ðxÞQðyÞ
exy ¼
2 GI zz tðyÞ
In this case the compatibility equations reduce to
∂2 ezz
¼0
∂x2
∂2 exy ∂2 eyy
2  ¼0 (4.73)
∂x∂y ∂x2
∂2 ezz
¼0
∂x∂y
or, equivalently,
00
νM z ðxÞy
¼0
EI zz
 
V 0y ðxÞ d Q
00
νM z ðxÞy
 ¼0 (4.74)
GI zz dy t EI zz
νM 0z ðxÞ
¼0
EI zz
where ðÞ0 ¼ dðÞ=dx and ðÞ00 ¼ d 2 ðÞ=dx2 . If the internal bending moment, M z , is a constant
and V y ¼ 0 (which is true for the pure bending case) then the compatibility equations will be
satisfied exactly. Otherwise, we see that in general compatibility will not be satisfied. In
engineering beam theory this incompatibility is normally not addressed since one considers
only the y-displacement of the neutral axis, which we will call va ðxÞ here, which is obtained
directly by integrating the moment–curvature relationship d 2 va =dx2 ¼ M z ðxÞ=EI zz , not by
integrating the strains.
Let us examine a case where we can have compatibility satisfied at least approximately so that
we can integrate the strains to find the displacements throughout the beam. Note that if
dM z =dx ¼ V y is a constant then the first two equations in Eq. (4.74) will be satisfied, but
not the third. Consider a cantilever beam (Fig. 4.11a) where the beam has a rectangular cross-
section (Fig. 4.11b). For this beam, the shear force V y ¼ P throughout the beam. Thus, the
compatibility equation for the strains exx ðx; yÞ, eyy ðx; yÞ, exy ðx; yÞ in the x–y plane, which is
the second equation in Eq. (4.74), will be satisfied exactly, but the third equation in Eq. (4.74) for
the out-of-plane strain, ezz , namely ∂2 ezz =∂x∂y ¼ 0, cannot be satisfied. This is similar to the
plane stress case so we expect that we should be able to at least integrate the strains in the x–y
plane to find the in-plane displacements ux ðx; yÞ, uy ðx; yÞ, which we will now show is indeed true.
108 Strain

Figure 4.11 (a) A cantilever beam of length L. (b) The y

rectangular cross-section of width b and height, 2h. y

P z
2h
O
x
L
b
(a) (b)

First, let us obtain the y-displacement of the neutral axis, va using the moment–curvature
relationship. In the beam the internal force and moment are V y ¼ P and M z ¼ Px. For

the rectangular cross-section, we have QðyÞ ¼ b h2  y2 =2 and I zz ¼ 2bh3 =3. Integrating
the moment–curvature relationship twice gives

d 2 va Px
2
¼
dx EI zz
dva Px2
¼ þ C1 (4.75)
dx 2EI zz
Px3
va ¼  þ C1x þ C2
6EI zz
in terms of the constants of integrations C 1 , C 2 , which can be found from the boundary
conditions of zero deflection and slope at the fixed wall:
dva ðLÞ PL2
¼0 ! C1 ¼
dx 2EI zz
(4.76)
PL3
va ðLÞ ¼ 0 ! C2 ¼ 
3EI zz
which gives the displacement of the neutral axis:
P x 3 L2 x L3
va ¼  þ  (4.77)
EI zz 6 2 3

Now, consider integrating both the normal strains exx , eyy . From Eq. (4.72) these integrals are
∂ux Pxy
¼ exx ¼
∂x EI zz
Px2 y
ux ðx; yÞ ¼ þ gðyÞ
2EI zz
(4.78)
∂uy νPxy
¼ eyy ¼ νexx ¼ 
∂y EI zz
νPxy2
uy ðx; yÞ ¼  þ hðxÞ
2EI zz
 4.4 Satisfaction of Compatibility – Strength of Materials Solutions 109

in terms of two arbitrary functions gðyÞ, hðxÞ. But from the shear strain we find
∂ux ∂uy 3 P  2
þ ¼ 2exy ¼ h  y2 (4.79)
∂y ∂x 4 bh3 G

which yields, using Eq. (4.78) and rearranging the terms,

Px2 dh 3 P  2 νPy2 dg
þ ¼ h  y 2
þ  (4.80)
2EI zz dx 4 bh3 G 2EI zz dy

Equation (4.80) is of the form H ðxÞ ¼ G ðyÞ and since x and y are independent variables, this
equation can only be true for all x and y if H ðxÞ ¼ G ðyÞ ¼ C, where C is a constant. Thus,
integrating the two sides of Eq. (4.80) on x or y, respectively, and using the value of I zz we find
Px3
hðxÞ ¼ Cx  þ v0
6EI zz
  (4.81)
P y3 νPy3
gðyÞ ¼ Cy þ h y
2
þ þ u0
2GI zz 3 6EI zz

in terms of two constants of integration u0 , v0 . Placing these values into the displacement
expressions in Eq. (4.78) gives
 
Px2 y P y3 νPy3
ux ðx; yÞ ¼ þ h2 y  þ  Cy þ u0
2EI zz 2GI zz 3 6EI zz
(4.82)
Px3 νPxy2
uy ðx; yÞ ¼   þ Cx þ v0
6EI zz 2EI zz
The constants all have a physical meaning. In particular,
ux ð0, 0Þ ¼ u0
uy ð0, 0Þ ¼ v0

(4.83)
∂uy


¼C
∂x
x¼0, y¼0

so that ðu0 , v0 Þ are the x- and y-displacements at the load P and C is the slope of the neutral
axis at the load. If, at the wall, we set
ux ðL; 0Þ ¼ uy ðL; 0Þ ¼ 0
∂uy ðL; 0Þ (4.84)
¼0
∂x
we find
PL3
u0 ¼ 0, v0 ¼ 
3EI zz
(4.85)
PL2
C¼
2EI zz
110 Strain

Using the relationship between the shear modulus and Young’s modulus, namely,
G ¼ E=2ð1 þ νÞ, the displacements then become finally
 
Px2 y Pð1 þ νÞ 2 y3 νPy3 PL2 y
ux ðx; yÞ ¼ þ h y þ 
2EI zz EI zz 3 6EI zz 2EI zz
(4.86)
Px3 νPxy2 PL2 x PL3
uy ðx; yÞ ¼   þ 
6EI zz 2EI zz 2EI zz 3EI zz
If we examine the displacements of the neutral axis from these expressions, we have
ua ¼ ux ðx; 0Þ ¼ 0
Px3 PL2 x PL3 (4.87)
va ¼ uy ðx; 0Þ ¼  þ 
6EI zz 2EI zz 3EI zz
which agrees with the strength of materials solution for the vertical displacement, Eq. (4.78).
This displacement is shown (greatly exaggerated) in Fig. 4.12a. According to the assump-
tions of the beam theory presented in Chapter 1, plane sections initially at right angles to the
neutral axis remain plane and at right angles after the loads are applied which would imply
that a small pair of horizontal and vertical lines at the wall would remain horizontal and
vertical, as shown in Fig. 4.12a. However, the displacements of Eq. (4.86) require that a line
in the beam that is originally vertical at the wall becomes sloped, while a horizontal line will
remain horizontal, as shown in Fig. 4.12b. Thus, there is an inconsistency between the
original bending deformation assumptions and the displacements predicted by the theory.
This is not surprising since in engineering beam theory we know that there are shear stresses
generated when the bending normal stress, σ xx , is not constant along the beam so that there
must be associated shear strains developed as well and these shear stresses will lead to
additional shearing displacements. One can remove this inconsistency and generate a
modified beam theory that accounts for beam deflections due to shear that is called
Timoshenko beam theory [3]. However, for long, slender beams (where h/L << 1), these
shear deflections are not significant and engineering beam theory is adequate. We can see

Figure 4.12 (a) The neutral axis y x=L

deflections of a cantilever beam (greatly neutral axis
exaggerated). According to the
elementary engineering beam theory x neutral axis
deformation assumptions a small pair
of horizontal and vertical lines at the v0
neutral axis and at the wall location
L
should remain horizontal and vertical.
(b) The displacements generated by the (a) (b)
theory, however, predict that the two
lines are no longer at a right angle.
 4.4 Satisfaction of Compatibility – Strength of Materials Solutions 111

Figure 4.13 Displacement of the neutral axis of a cantilever beam in

neutral axis elementary engineering beam theory (greatly exaggerated) when a small
vertical line along the neutral axis at the wall is held vertical. This requires
the neutral axis to have a slope at the wall.
v0

this type of behavior within the context of the present theory if we simply change the
boundary conditions at the wall so that Eq. (4.84) is replaced by

ux ðL; 0Þ ¼ uy ðL; 0Þ ¼ 0
∂uy ðL; 0Þ (4.88)
¼0
∂x

This means we are requiring a small vertical line at x ¼ L, y ¼ 0 to remain vertical but a
small line at the neutral axis can now have a slope at the wall as shown in Fig. 4.13. These
boundary conditions give
u0 ¼ 0
" #
PL3 h2
v0 ¼  1 þ 3ð1 þ νÞ 2
3EI zz L (4.89)
" #
PL2 h2
C¼ 1 þ 2ð1 þ νÞ 2
2EI zz L

so that there is a larger deflection and slope predicted at the load (see Eq. (4.85))
with correction terms that are of order h2 =L2 (so that they are small for long, slender
beams).
Since we have obtained the displacements throughout the beam from engineering beam
theory, we can plot these displacements to examine this displacement field. It is convenient
to write these equations in normalized form as
!
EI zz ux x 2 y h2 y3 νy3 y
3
¼ þ ð1 þ νÞ 2 y  þ 
PL 2 L 3 6 2
(4.90)
EI zz uy x 3
x y
ν 2  1
x
¼  þ 
PL3 6 2 2 3

where x ¼ x=L, y ¼ y=L. A MATLAB® script cantilever_d (see Appendix D) was written
that computes these normalized displacements for h/L ¼ 0.1, ν ¼ 1=3, and plots them as a
“quiver” plot of arrows representing the displacements. The result is shown in Fig. 4.14. It is
seen that, while the neutral axis only undergoes a vertical displacement, this is not true of
other points in the beam.
112 Strain

Figure 4.14 Displacements of a cantilever beam.

ez

ez eθ
ey er
ex
r z y

T
x Figure 4.15 Cylindrical coordinates.

4.5 Strains in Cylindrical Coordinates

Consider the small strains along an orthogonal set of coordinates ðn; t; sÞ. From Eq. (4.15)
the normal strains and shear strains associated with these coordinates can be written as
∂u ∂u ∂u
enn ¼ n , ett ¼ t , ess ¼ s
∂sn ∂st ∂ss
      (4.91)
1 ∂u ∂u 1 ∂u ∂u 1 ∂u ∂u
ent ¼ n þ t , ens ¼ n þ s , ets ¼ t þ s
2 ∂st ∂sn 2 ∂ss ∂sn 2 ∂ss ∂st

Before, we let ðn; t; sÞ be along a set of ðx; y; zÞ axes to obtain the strains in Cartesian
coordinates. Similarly, we can use these expressions to find the strains in any other set of
orthogonal coordinates. Consider a set of cylindrical coordinates, for example, as shown in
Fig. 4.15. In that case we can take the lengths of small elements in the ðr; θ; zÞ directions as
∂sn ¼ ∂r
∂st ¼ r∂θ (4.92)
∂sz ¼ ∂z

and the unit vectors ðn; t; sÞ as

n ¼ er ¼ cos θex þ sin θey
t ¼ eθ ¼  sin θex þ cos θey (4.93)
s ¼ ez
 4.5 Strains in Cylindrical Coordinates 113

The displacement vector, u, is given by

u ¼ ur er þ uθ eθ þ uz ez (4.94)
Consider now the normal strain in the r-direction. We have
∂u
err ¼ er 
∂r
 
∂ur ∂uθ ∂uz
¼ er  er þ eθ þ ez (4.95)
∂r ∂r ∂r
∂ur
¼
∂r
In the θ-direction, however,
 
1 ∂u ∂ur ∂uθ ∂uz ∂er ∂eθ
eθθ ¼ eθ  ¼ eθ  er þ eθ þ ez þ u r þ uθ
r ∂θ ∂θ ∂θ ∂θ ∂θ ∂θ
(4.96)
1 ∂uθ ur
¼ þ
r ∂θ r
where we use the fact that
∂er ∂eθ
¼ eθ , ¼ er (4.97)
∂θ ∂θ
The additional term coming from the radial displacement can be explained if we consider
how such a displacement affects the elongation of an element initially in the θ-direction, as
shown in Fig. 4.16. We see from that figure that a radial displacement causes a strain in the
θ-direction for an element initially in that direction given by
ðr þ ur Þdθ  rdθ ur
ðeθθ Þur ¼ ¼ (4.98)
rdθ r
Now consider the strain in the z-direction. We find
 
∂u ∂ur ∂uθ ∂uz
ezz ¼ ez  ¼ ez  er þ eθ þ ez
∂z ∂z ∂z ∂z
(4.99)
∂uz
¼
∂z
y ( r + u r ) dq
rdq

dq ur
r
q Figure 4.16 Effects of a radial displacement, ur , on the length of an element (bold line)
x in the θ -direction.
114 Strain

Figure 4.17 (a) Two small lines along the r- and θ - directions E
in an undeformed body. (b) Those same two lines in the uT
deformed body when there is a constant displacement, uθ , of
uT
the radial line. r E
T
r

(a) (b)

Next, consider one of the shear strains, namely

 
1 1 ∂u ∂u
erθ ¼ er  þ eθ 
2 r ∂θ ∂r
   
1 1 ∂ur ∂uθ ∂er ∂eθ 1 ∂ur ∂uθ ∂uz
¼ er  er þ eθ þ u r þ uθ þ eθ  er þ eθ þ ez (4.100)
2 r ∂θ ∂θ ∂θ ∂θ 2 ∂r ∂r ∂r
 
1 1 ∂ur ∂uθ uθ
¼ þ 
2 r ∂θ ∂r r

The first two terms in Eq. (4.100) are analogous to the terms that appear in Cartesian
coordinates. The last terms can be explained by the fact that, even if the displacement in the
θ-direction is not changing in r, it can still produce a shear strain as shown in Fig. 4.17, where
β 1 uθ
ðerθ Þuθ ¼  ¼  (4.101)
2 2 r
(the minus sign exists because a constant displacement in the θ-direction causes the angle between
two lines initially along the r- and θ-directions to be greater than 90 , as shown in Fig. 4.17b).
Finally, in exactly the same fashion we can obtain the other two shear strains, which are
given as
 
1 ∂uz ∂ur
erz ¼ þ
2 ∂r ∂z
  (4.102)
1 ∂uθ 1 ∂uz
eθz ¼ þ
2 ∂z r ∂θ

4.6 PROBLEMS

P4.1 Consider the state of strain of Example 4.1.


(a) If we let ep1 , ep2 , ep3 be unit vectors along the three principal strain directions
obtained in that example, show for the lines at 45 from the largest and smallest
principal strains defined by the unit vectors
 4.6 Problems 115

 pffiffiffi
d1 ¼ ep1 þ ep3 = 2
 pffiffiffi
d2 ¼ ep1  ep3 = 2

we obtain from the strain transformation equations the largest tensor shear strain
(in magnitude) along these directions and that twice this largest strain is the largest
engineering shear strain.
(b) Define the unit vector v ¼ n  t and use the strain transformation equation from
the ðx; y; zÞ coordinates to the ðn; t; vÞ coordinates to obtain the state of strain with
respect to the ðn; t; vÞ coordinates. Identify the strains ðenn , ent Þ previously calcu-
lated in this example for this state of strain.
P4.2 Show that the compatibility equation for a state of plane strain, Eq. (4.59), can be
found directly from the displacements obtained by integrating the strain–displacement
relations.
P4.3 The state of strain (in μstrain) at a point P in a deformed body is given for a set of (x, y, z)
coordinates by
2 3
200 300 0
½e ¼ 4 300 100 200 5
0 200 100

(a) Determine the principal strains at point P.

(b) For the largest positive principal strain, determine the corresponding principal
strain direction explicitly, i.e., DO NOT USE MATLAB® or any other software
package, but instead obtain the principal direction components by hand. Show all
your work. (You can use MATLAB® or any other package to verify your answer.)
P4.4 Consider the same state of strain as given in problem P4.3 and assume these strains are
constant throughout the deformed body. Figure P4.1 shows a set of lines in this region
of constant strain that have the lengths and directions as shown in the undeformed
body (|PA| ¼ |PB| ¼ |PC| ¼ a)
(a) Determine the length of line AC in the deformed body in terms of a.
(b) Determine the angle, in degrees, between lines DB and DA in the deformed body
(point D is at the center of AC).
P4.5 The state of strain in a region about a point P in a stressed body is
z given by
B 2 3
0:003 0 0:003
½e ¼ 4 0 0:004 0:005 5
0:003 0:005 0:006
a
a A
P
a y
C D
Figure P4.1 A set of lines (as seen in a body when it is undeformed) that are in a
x region of uniform strain in the deformed body.
116 Strain

and these strains are constant in this region. Consider the lines in this region shown
in Fig. P4.2.
(a) Determine the change in angle (in degrees) between lines PA and PB, which are
initially have a length, a, and are at a right angle to each other in the deformed
body, as shown.
(b) Determine p the
ffiffiffi new length of line PC in the deformed body, which originally has a
length of a 2.
(c) Determine the state of strain with respect to the ðx0 , y0 , z0 Þ axes shown in Fig. P4.2.
How is this state of strain related to any answers in parts (a) and (b)?
P4.6 The state of strain (measured in μstrain) with respect to an (x, y, z) coordinate system
is:
2 3
100 30 10
½e ¼ 4 30 50 0 5
10 0 250
(a) Determine the principal strains and the principal strain directions.
(b) Determine the magnitude of the largest engineering shear stress and the directions
(unit vectors) of the lines in the body along which this largest shear stress occurs
(see problem P4.1). Use the strain transformation equations to verify that one
obtains this largest shear stress along the directions you obtained.
P4.7 The state of strain (measured in μstrain) with respect to an (x, y, z) coordinate system is:
2 3
200 25 0
½e ¼ 4 25 100 0 5
0 0 200

(a) Determine the principal strains and the principal strain directions.
(b) Determine the magnitude of the largest engineering shear stress and the directions
(unit vectors) of the lines in the body along which this largest shear stress occurs
(see problem P4.1).
Use the strain transformation equations to verify that one obtains
y, y' this largest shear stress along the directions you obtained.
P4.8 The state of strain (measured in μstrain) with respect to
B an (x, y, z) coordinate system is:
2 3
300 50 30
a
C ½e ¼ 4 50 100 0 5
a P 30 0 100
D a x
o
45
z' 45o
A
Figure P4.2 A set of lines (as seen in a body when it is undeformed) that
z x' are in a region of uniform strain in the deformed body.
 4.6 Problems 117

y' y y Figure P4.3 (a) A coordinate rotation of 40

y'
about the z-axis. (b) A coordinate rotation
x' 30o of 30 about the x-axis.

40o
O x O
z

z, z' x, x'
z'
(a) (b)

Determine the state of strain with respect to the ðx0 , y0 , z0 Þ axes that are rotated 40
about the z-axis (see Fig. P4.3a).
P4.9 The state of strain (measured in μstrain) with respect to an (x, y, z) coordinate
system is:
2 3
400 25 30
½e ¼ 4 25 200 50 5
30 50 100
Determine the state of strain with respect to the ðx0 , y0 , z0 Þ axes that are rotated 30
about the x-axis (see Fig. P4.3b).
P4.10 When the strains are uniform in a region, we can analyze the changes of lengths and
angles of finite lines in that region, as considered in a number of previous problems.
However, if the strains vary from point to point then this is no longer possible.
Consider, for example a finite line drawn from point A to point B in an undeformed
body, as shown in Fig. P4.4a. Suppose this line becomes the line drawn from A0 to B0
in the deformed body (whose deformation, assumed to be small, is greatly exagger-
ated). If at a point P on the line AB we draw a small line segment of length ds (which
acts in the n direction), and if that point moves to point P0 where the line segment now
has a length ds0 , then the small normal strain at P along the line is defined as
enn ¼ ðds0  dsÞ=ds. It follows that
ð0
B ðB ðB
0
ds  ds ¼ enn ds
A0 A A
ðB
! L0  L ¼ enn ds
A

where ðL0 ; LÞ are the lengths of the lines A0 B0 and AB in the deformed and
undeformed bodies, respectively. Another way to write this relation is
118 Strain

Figure P4.4 (a) A finite line in a undeformed body. (b) The line
A'
A P P' after a small deformation (which is greatly exaggerated).
ds
ds'
n
B B'

(a) (b)

y Figure P4.5 A plate geometry where two lines AB and AC are shown
when the plate is undeformed. Each line is 2 in. long.

C
ðB
4 L0  L 1
enn ¼ ¼ enn ds
B L L
A A
3
where enn is the average normal strain for the finite
x
line AB.
Consider a pair of lines AB and AC in a plate as shown in Fig. P4.5. Each of these lines is
two inches long. If the plate has a 2-D displacement field given by ux ¼ cð2x þ yÞ,

uy ¼ b x  3y2 , where x and y are measured in inches and b and c are both very small
constants, determine the lengths of these lines in the deformed body in terms of b and c.
What can you say about the change in angles?

References
1. I. S. Sokolnikoff, Mathematical Theory of Elasticity (New York, NY: McGraw-Hill, 1956)
2. C. R. Wylie and L. C. Barrettt, Advanced Engineering Mathematics, 6th edn (New York, NY:
McGraw-Hill, 1995)
3. H. Reismann and P. S. Pawlik, Elasticity-Theory and Applications (New York, NY: John Wiley and
Sons, 1980)
5 Stress–Strain Relations

Stresses describe the local distributions of forces within a deformable body, and strains
describe the local deformations. In this chapter we want to describe the relations between
stresses and strains as these are the key relationships that allow us to connect the loads
applied to a body to its changes in shape. In this book, we will consider only linear elastic
materials, where the stresses are proportional to the small strains present. Both isotropic and
anisotropic linear elastic materials will be discussed. As described in Chapter 1 these stress–
strain relations are one of the key “pillars” of stress analyses.

5.1 Linear Elastic Materials

5.1.1 Linear Elastic Isotropic Materials
The relationships between stresses and strains seen in the mechanics of materials
problems discussed in Chapter 1 were simple since there were very few significant
stresses and strains present in those elementary problems. In the axial loading and pure
bending cases, for example, there was only one significant axial stress, σ, and we related
this stress to the small axial strain, e, through Hooke’s law σ ¼ Ee, but we noted that
there were also small strains, el ¼ eyy or ezz , generated by this axial stress in those lateral
directions where el ¼ νe. In the case of torsion we had shear stresses, τ ¼ σ xy or σ xz ,
generated that we again related to shear strains, γ ¼ γxy or γxz , through a similar Hooke’s
law τ ¼ Gγ. These three material constants, the elastic constant E (Young’s modulus),
the constant G (the shear modulus), and the constant ν (Poisson’s ratio), while appearing
separately in those problems, are in fact the constants that help to define the general
relationship between stresses and strains for small deformations of an isotropic, linearly
elastic solid. The term “isotropic” means that the stress–strain relations for the material
are the same in all directions at any point in a body at that point (although the stress–
strain relations may vary from point to point if the body is inhomogeneous). The term
“linearly elastic” means that that there is a linear relationship between the stresses and
the strains. Hooke’s law is obviously such a linear relationship. Remarkably, we can
generate the general stress–strain equations for an isotropic, linear elastic material (also
called generalized Hooke’s law) by using the simple results of the elementary problems
discussed in Chapter 1 and the principle of superposition. We start by applying to an
element a uniaxial stress σ xx , as seen in Fig. 5.1a, which produces both a direct small


strain, exx , and lateral small strains eyy , ezz . For an isotropic material, these lateral

119
120 Stress–Strain Relations

Figure 5.1 (a) Applying a normal V yy V yy

stress to an element acting in the
x-direction, followed by adding
(b) a normal stress in the y-direction,
(c) a normal stress in the z-direction, V xx V xx V xx
and (d) a shear stresses σ xy and
σ yx ¼ σ xy . (e) A complete state of V zz
y
stress where there are three normal (a) (b) (c)

 V yy V yy
stresses σ xx , σ yy , σ zz and three shear


stresses σ xy , σ yz , σ zx and their
symmetric counterparts. Note that x V xy V yz V xy
only the stresses acting on the front z V zx
V xx V xx
faces of the element are shown.
V zz V zz
(d) (e)

strains are equal since the behavior of the material is the same in both the y- and z-
directions, giving
σ xx
exx ¼
E
νσ xx
eyy ¼  (5.1)
E
νσ xx
ezz ¼ 
E
Now, we hold the stress σ xx constant and apply a normal stress in the y-direction, as seen in
Fig. 5.1b. This causes both a direct strain, eyy , and lateral strains ðexx , ezz Þ. For an isotropic
material these strains are related to the stresses with the same elastic constants used in
Eq. (5.1) so for the state of stress in Fig. 5.1b we have
σ xx νσ yy
exx ¼ 
E E
σ yy νσ xx
eyy ¼  (5.2)
E E
νσ xx νσ yy
ezz ¼  
E E
Keeping the σ xx and σ yy stresses constant, we then add the normal stress, σ zz , as shown in


Fig, 5.1c, which produces an additional direct strain, ezz , as well as lateral strains, exx , eyy ,
giving the state of stress shown in Fig. 5.1c where:
σ xx νσ yy νσ zz
exx ¼  
E E E
σ yy νσ xx νσ zz
eyy ¼   (5.3)
E E E
σ zz νσ xx νσ yy
ezz ¼  
E E E
 5.1 Linear Elastic Materials 121

Now, we add the shear stress, σ xy (and its equal counterpart σ yx ), as shown in Fig. 5.1d. This
generates an engineering shear strain, γxy , given by γxy ¼ σ xy =G in terms of the shear
modulus. Similarly, when we apply the shear stresses σ yz , σ zx we will generate additional
engineering shear strains γyz ¼ σ yz =G, γzx ¼ σ zx =G, producing the stress–strain relations for
the complete three-dimensional state of stress shown in Fig. 5.1e where
σ xx ν

exx ¼  σ yy þ σ zz
E E
σ yy ν
eyy ¼  ðσ xx þ σ zz Þ
E E
(5.4)
σ zz ν

ezz ¼  σ xx þ σ yy
E E
σ xy σ yz σ zx
γxy ¼ , γyz ¼ , γzx ¼
G G G

In obtaining this result the fact that the material was assumed to be isotropic allowed us to
use the same elastic constants ðE; G; νÞ in relating stresses to strains in the different
directions. Because we also assumed linear relations between the stresses and strains, we
could add the strains together through superposition. Equation (5.4) gives us the strains in
terms of the stresses. We can invert these relations and instead write the stresses in terms of
the strains as
E 

σ xx ¼ ð1  νÞexx þ ν eyy þ ezz
ð1 þ νÞð1  2νÞ
E  
σ yy ¼ ð1  νÞeyy þ νðexx þ ezz Þ
ð1 þ νÞð1  2νÞ (5.5)
E 

σ zz ¼ ð1  νÞezz þ ν exx þ eyy
ð1 þ νÞð1  2νÞ
σ xy ¼ Gγxy , σ yz ¼ Gγyz , σ zy ¼ Gγzx

While these general stress–strain relations are written in terms of the three material con-
stants, ðE; G; νÞ, an isotropic, linearly elastic solid is actually characterized by only two
independent constants since the condition of isotropy can be shown to require that
G ¼ E=2ð1 þ νÞ.
Equations (5.4) and (5.5) are generalized Hooke’s law for general 3-D states of stress and
strain in a linear elastic, isotropic material. Many common structural materials such as steel
and aluminum alloys, for example, can be characterized by these relations. Because Eq. (5.5)
is so commonly used, we have written a MATLAB® function stress_strain that has the
calling sequence
stress = stress_strain(E, nu, strain);

which takes a 3  3 strain matrix, strain, where the strains are measured in μstrain, and
returns the 3  3 stress matrix, stress, with the stresses measured in MPa. Here, Young’s
122 Stress–Strain Relations

modulus, E, is given in GPa, and nu is Poisson’s ratio. In the same fashion we have a
MATLAB® function strain_stress that has the calling sequence
strain = strain_stress(E, nu, stress);

which takes a 3  3 stress matrix, stress, where the stresses are in MPa, and returns the 3  3
matrix, strain, with the strains measured in μstrain. Again, Young’s modulus, E, is given in
GPa, and nu is Poisson’s ratio.

5.1.2 General Linear Elastic Stress–Strain Relations

In the most general case when the stresses are linearly related to the small strains, the stress–
strain relationship can be written in matrix form as
X
3 X
3
σ ij ¼ C ijkl ekl (5.6)
k¼1 l¼1

in terms of the elastic constants C ijkl . Since there are nine stresses and nine strains, there are a
total of 81 constants. However, if we use the fact that the stress matrix and strain matrix are
symmetric, we have
C ijkl ¼ C jikl
(5.7)
C ijkl ¼ C ijlk

so that we are only really relating six stresses to six strains and there is a total of 36 constants.
We can write these stress–strain relations out explicitly as
8 9 2 38 9
>
> σ 11 >
> C 1111 C 1122 C 1133 C 1123 þ C 1132 C 1113 þ C 1131 C 1112 þ C 1121 >
> e 11 >
>
>
> > 6 7> >
>
> σ 22 >
>
> C C C C þ C C þ C C þ C >
>
> e >
>
>
= 6 7<
2211 2222 2233 2233 2232 2213 2231 2212 2221 22
< 6 7 =
σ 33 C C C C þ C C þ C C þ C e
¼66
3311 3322 3333 3323 3332 3313 3331 3312 3321 7 33
7 e23 >
> σ 23 >
>
> > 6 C 2311 C 2322 C 2333 C 2323 þ C 2332 C 2313 þ C 2331 C 2312 þ C 2331 7>
> >
> >
>
> > 4 C 1311 C 1322 C 1333 C 1323 þ C 1332 C 1313 þ C 1331 C 1312 þ C 1321 5> >
> σ 13 >
>
: >
; > e13 >
>
: >
;
σ 12 C 1211 C 1222 C 1233 C 1223 þ C 1232 C 1213 þ C 1231 C 1212 þ C 1221 e12
(5.8a)

and using the symmetry properties of the elastic constants

8 9 2 38 9
>
> σ 11 >
> C 1111 C 1122 C 1133 2C 1123 2C 1113 2C 1112 >> e11 >
>
>
> >
> 6 C 2211 C 2222 > e22 >
> >
>
> σ >
> C 2233 2C 2223 2C 2213 2C 2212 7
7>
> >
>
< 22
= 66 C 3311 C 3322 7< =
σ 33 C 3333 2C 3323 2C 3313 2C 3312 7 e33
¼6
6 (5.8b)
> σ 23 >
> > 6 C 2311 C 2322 C 2333 2C 2323 2C 2313 2C 2312 7> e23 >
7> >
>
> >
> > >
> σ 13 >
> >
4 C 1311 C 1322 C 1333 2C 1323 2C 1313 2C 1312 5>
> e13 >
> >
>
: ; : ;
σ 12 C 1211 C 1222 C 1233 2C 1223 2C 1213 2C 1212 e12

where the factors of two comes from combining terms for the symmetric shear strains. We
can eliminate those factors by writing the stress–strain relations in terms of the engineering
shear strains, giving
 5.1 Linear Elastic Materials 123

8 9 2 38 9
>
> σ 11 >
> C 1111 C 1122 C 1133 C 1123 C 1113 C 1112 >> e11 >
>
>
> >
> 6 C 2211 > e22 >
> >
>
> σ >
> C 2222 C 2233 C 2223 C 2213 C 2212 7
7>
> >
>
< 22
= 66 C 3311 7< =
σ 33 C 3322 C 3333 C 3323 C 3313 C 3312 7 e33
¼6
6 C 2311 (5.9)
>
> σ 23 >
> C 2322 C 2333 C 2323 C 2313 C 2312 7
7>
> γ >
>
> >
>
6 > 23 >>
>
> σ 13 >
>
4 C 1311 C 1322 C 1333 C 1323 C 1313 C 1312 5>
>
> γ > >
: ; : 13 >;
σ 12 C 1211 C 1222 C 1233 C 1223 C 1213 C 1212 γ12

We will see in Chapter 8 that, if we also assume these stress–strain relations come from an
internal energy function, the elastic constants have the additional symmetry relations
C ijkl ¼ C klij (5.10)

If you examine the relations of Eq. (5.10) for the coefficient matrix in Eq. (5.9), you will see
that Eq. (5.10) means that the coefficient matrix is symmetric, which reduces the total
number of constants to 21 (six diagonal terms and 15 off-diagonal terms). For computa-
tional purposes, it is advantageous to rewrite Eq. (5.9), in terms of a vector of six stresses
σ I ðI ¼ 1, 2 . . . ; 6Þ and six strains eJ ðJ ¼ 1, 2 . . . ; 6Þ related by a 6  6 symmetric matrix C IJ
where
(
i for i ¼ j
I¼
9  ði þ j Þ for i ¼ 6 j
( (5.11)
k for k ¼ l
J¼
9  ðk þ l Þ for k ¼ 6 l

so that we have in this contracted notation

X
6
σI ¼ C IJ eJ ðI ¼ 1, 2 . . . ; 6Þ (5.12)
J¼1

or, in matrix–vector notation,

fσ g ¼ ½C feg (5.13)

where

fσ g ¼ ½σ 11 , σ 22 , σ 33 , σ 23 , σ 13 , σ 12 T
(5.14)
feg ¼ ½e11 , e22 , e33 , γ23 , γ13 , γ12 T

or, equivalently, in terms of the (x, y, z) Cartesian stress and strain components
8 9 2 38 9
>
> σ xx >
> C 11 C 12 C 13 C 14 C 15 C 16 > > exx >>
>
> >
> 6 > eyy >
> >
>
> σ yy >
> C 21 C 22 C 23 C 24 C 25 C 26 7 7>> >
< = 6 6 C 31 <e > =
σ zz C 32 C 33 C 34 C 35 C 36 7
¼66 C 41
7 zz (5.15)
>
> σ yz >
> 6 C 42 C 43 C 44 C 45 C 46 7 7>> γyz >>
>
> >
> >γ >
>
> σ xz >
>
4 C 51 C 52 C 53 C 54 C 55 C 56 > 5 >
>
>
xz >
>
: ; : ;
σ xy C 61 C 62 C 63 C 64 C 65 C 66 γ xy
124 Stress–Strain Relations

where the 6  6 elastic constants matrix, [C ], is symmetric, i.e., C 12 ¼ C 21 , C 23 ¼ C 32 , etc.

Equation (5.15) is the general linear stress–strain relationship for an anisotropic material,
written in what is called Voigt notation. Most anisotropic materials used in engineering
practice, however, have certain directional symmetries that considerably reduce the number
of independent constants. We will not derive in detail how such symmetries affect the
constants matrix (for such details see theory of elasticity texts) but will simply list below
some of the important special cases.

Orthotropic Material (Nine Constants)

Materials such as wood or some composite materials have an orthotropic behavior (see
Fig. 5.2) where there are different constants along three orthogonal directions. Although
these materials are inhomogeneous on a microscopic level, engineers usually treat them in
terms of an equivalent homogeneous continuum model where these local properties are
“smeared out” and averaged. Along the (x, y, z) material directions (see Fig. 5.2) the stress–
strain relations reduce to
8 9 2 38 9
>
> σ xx >
> C 11 C 12 C 13 0 0 0 > > exx >
>
>
> >
> 6 C 12 > eyy >
> >
>
> σ >
> C 22 C 23 0 0 0 7 7>> >
>
= 6
yy
< 6 C 13 7 < =
σ zz C 23 C 33 0 0 0 7 ezz
¼6
6 (5.16)
> σ yz >
> > 6 0 0 0 C 44 0 0 7 7>> γ >
>
> >
> > yz >>
> σ xz >
> >
4 0 0 0 0 C 55 0 5> >
> γ > >
: ; : xz >;
σ xy 0 0 0 0 0 C 66 γxy

which can also be inverted to write the strains in terms of the stresses in terms of a
compliance matrix, [D], where

x
Figure 5.2 (a) A composite material
consisting of elliptical-shaped bundles
of fibers embedded in a matrix
material. This material has different y
y
constants values along the (x, y, z)
directions and can be treated as
z
orthotropic. (b) Wood also has a
fibrous structure, which produces
different constants values along the fiber x
wood fibers, in the radial ring growth bundles
direction, and tangential to the growth matrix
z
rings, leading to orthotropic behavior. (a) (b)
 5.1 Linear Elastic Materials 125

8 9 2 38 9
>
> exx > D11 D12 D13 0 0 0 > σ xx >
>e >
> > > >
> σ yy >
>
>
> yy >
> 6 D12 D22 D23 0 0 0 77>
>
>
<e > = 6
6 D13 7
>
< >
=
D23 D33 0 0 0 7 σ zz
¼6
zz
6 0 7 (5.17)
>
> γyz >
> 0 0 D44 0 0 7> > σ >
>
> >
>
6 > yz > >
>
> γxz >
>
4 0 0 0 0 D55 0 5> >
> σ > >
: ; : xz > ;
γxy 0 0 0 0 0 D66 σ xy

Note that the choice of these constants or compliance values depends on our definition of the
directions of the ðx; y; zÞ axes. For the fiber-reinforced composite of Fig. 5.2a, for example,
we have taken the z-axis to be along the fiber bundle direction, while for the wood case we
have taken the wood fibers to be along the x-axis instead.
In engineering practice, the compliance is often written in a form similar to that of
generalized Hooke’s law for an isotropic material, namely
1 νyx νzx
exx ¼ σ xx  σ yy  σ zz
Ex Ey Ez
1 νxy νzy
eyy ¼ σ yy  σ xx  σ zz
Ey Ex Ez
(5.18)
1 νxz νyz
ezz ¼ σ zz  σ xx  σ yy
Ez Ex Ey
1 1 1
γyz ¼ σ yz , γxz ¼ σ xz , γxy ¼ σ xy
G yz G xz G xy

where
νyz νzy νxz νzx νxy νyx
¼ , ¼ , ¼ (5.19)
Ey Ez Ex Ez Ex Ey

so that effectively we have three different Young’s moduli, three shear moduli, and three
Poisson’s ratios for this material. In terms of the full compliance matrix:
8 9 2 38 9
>
> exx >
> 1=E x νyx =E y νzx =E z 0 0 0 >
> σ xx >
>
>
> eyy >
> 6 νxy =E x νzy =E z 0 7 >
> σ yy >
>
>
> >
> 6 1=E y 0 0 7 >
> >
>
<e = 6 7 < =
ν =E ν =E 1=E 0 0 0 σ
¼6 7
zz xz x yz y z zz
6 (5.20)
> γyz >
> > 6 0 0 0 1=G yz 0 0 7 7>> σ yz >
>
>
> > > >
> γxz >
> > 4
> 0 0 0 0 1=G xz 0 5> >
> σ > >
: ; : xz > ;
γxy 0 0 0 0 0 1=G xy σ xy

Note that unlike the general anisotropic materials, where all the stresses and strains
are related to each other, in an orthotropic material normal strains only produce
normal stresses and shear strains only produce shear stresses along the material
directions (although, unlike the isotropic case, this is not true in general for other
directions).
126 Stress–Strain Relations


For plane stress σ zz ¼ σ zx ¼ σ zy ¼ 0 the stress–strain relations for an orthotropic solid
reduce to
Ex

σ xx ¼ exx þ νyx eyy
1  νxy νyx
Ex
 (5.21)
σ yy ¼ eyy þ νxy eyy
1  νxy νyx
σ xy ¼ G xy γxy

Transversely Isotropic Material (Five Constants)

A transversely isotropic material gets its name from the fact that, in a plane
orthogonal to a particular material axis, the material behaves as if it were isotropic.
An example is the case where circular glass or graphite fibers are embedded in a
polymer matrix (Fig. 5.3). In this case, assuming the z-axis is along the fiber
direction, the material has isotropic behavior in the x–y “transverse” plane and the
constants matrix is:
8 9 2 38 9
>
> σ xx >
> C 11 C 12 C 13 0 0 0 >
> exx >
>
>
> σ yy >
> 6 C 12 C 11 C 13 7>
> eyy >
>
>
> >
> 0 0 0 >
> >
< = 66 7
7<e > =
σ zz C C C 0 0 0
¼6 13 13 33 7 zz (5.22)
> > 6
> σ yz > 6 0 0 0 C 44 0 0 7> γyz >
7> >
>
> > > >
>
> σ > > 4 0 0 0 0 C 55 0 5>
> γ > >
: xz > ; : xz >
> ;
σ xy 0 0 0 0 0 ðC 11  C 12 Þ=2 γxy

and the corresponding compliance matrix can be written as

8 9 2 38 9
>
> exx > 1=E x νxy =E x νzx =E z 0 0 0 > σ xx >
>e >
> >
> 6 νxy =E x ν =E 0 7
> >
> σ yy >
> >
>
> > 1=E 0 0 7> >
<e > = 6 > >
yy x zx z
6 νxz =E x νxz =E x 7 < =
1=E 0 0 0 7 σ zz
¼6
zz z
6 (5.23)
>
> γ > 0 0 0 1=G 0 0 7 7> σ yz >
> yz >
> > 6 xz >
> >
>
>
> γxz >>
>
4 0 0 0 0 1=G xz 0 5> >
> σ > >
: ; : xz > ;
γxy 0 0 0 0 0 1=G xy σ xy


where νxz =E x ¼ νzx =E z and G xy ¼ E x =2 1 þ νxy . To make it
easier to connect these compliance values to the fiber and trans-
y verse plane directions, an f subscript is often used to denote “fiber”
and a t subscript used to denote “transverse”, giving

fibers x
Figure 5.3 A material where fibers of circular cross-section are embedded in
a material matrix. This material acts macroscopically like a transversely
z matrix isotropic material.
 5.1 Linear Elastic Materials 127

8 9 2 38 9
> exx > 1=E t νt =E t νf =E f 0 0 0 > σ xx >
>
> > > >
>e >
> >
> 6
0 7> >σ >
7> >
>
>
> >
> 6 νt =E t 1=E t νf =E f 0 0 > >
7> >
yy yy
>
> > 6 > >
< ezz >
= 6 νf =E f νf =E f 1=E f 0 0 7>
0 7 σ zz =
< >
6
¼6 7 (5.24)
>
> γ >
> 6 0 7 > σ yz >
>
>
yz >
> 6 0 0 0 1=G f 0 7>
> >
>
>
> >
>
γxz > 6 7>
>
>
>
>
>
>
> > 4 0 0 0 0 1=G f 0 5> > σ xz >
>
>
: >
; >
: >
;
γxy 0 0 0 0 0 1=G t σ xy

Cubic Material (Three Constants)

8 9 2 38 9
> σ xx > C 11 C 12 C 12 0 0 0 > exx >
>
> >
> >
> >
>
>
> >
> 6 0 7 > >
>
> σ yy >
> 6 C 12 C 11 C 12 0 0 7>
> e >
>
7> >
yy
>
> > 6 > >
< σ zz >
= 6C C 12 C 11 0 0 7>
0 7 ezz =
< >
6 12
¼6 7 (5.25)
> > 6 0
> σ yz > 0 7 > γyz >
>
> >
> 6 0 0 C 44 0 7>
>
>
>
>
>
>
> >
> 6 7>
> >
>
>
> σ xz >
> 4 0 0 0 0 C 44 0 5> > γ >
>
>
: >
; >
:
xz
>
;
σ xy 0 0 0 0 0 C 44 γxy

Materials of pure aluminum crystals (as opposed to aluminum alloys) have this cubic
material behavior.

Isotropic Material (Two Constants)

We have already given the stress–strain relations in the isotropic case. In terms of the elastic
constants matrix we have
8 9 2 38 9
> σ xx > C 11 C 12 C 12 0 0 0 > exx >
>
> >
> >
> >
>
>
> σ yy >
> 6 7> eyy >
>
> >
> 6 C 12 C 11 C 12 0 0 0 7>
>
>
>
>
>
>
> >
> 6 7>
> >
>
< σ zz = 6 C 12 C 12 C 11 0 0 0 7 ezz =
<
6 7
¼6 7
> > 6 0 7> >
> σ yz >
>
> > 6
>
0 0 ðC 11  C 12 Þ=2 0 0 > γyz >
7>
> >
>
>
> > 6 7> >
> σ xz >
> > 4 0
> 0 0 0 ðC 11  C 12 Þ=2 0 > γxz >
5>
> >
>
>
: >
; >
: >
;
σ xy 0 0 0 0 0 ðC 11  C 12 Þ=2 γxy
(5.26)

where
E ð 1  νÞ Eν
C 11 ¼ , C 12 ¼
ð1 þ νÞð1  2νÞ ð1 þ νÞð1  2νÞ
(5.27)
ðC 11  C 12 Þ E
¼ ¼G
2 2ð1 þ νÞ
128 Stress–Strain Relations

These stress–strain relations are identical to Eq. (5.5), namely

E 

σ xx ¼ ð1  νÞexx þ ν eyy þ ezz
ð1 þ νÞð1  2νÞ
E  
σ yy ¼ ð1  νÞeyy þ νðexx þ ezz Þ
ð1 þ νÞð1  2νÞ (5.28)
E 

σ zz ¼ ð1  νÞezz þ ν exx þ eyy
ð1 þ νÞð1  2νÞ
σ xy ¼ Gγxy , σ yz ¼ Gγyz , σ zx ¼ Gγzx

By rearranging these terms, we can write the stress–strain relationship in matrix terms. We
have, for example, equivalently
E 

σ xx ¼ ð1  2νÞexx þ ν exx þ eyy þ ezz
ð1 þ νÞð1  2νÞ
E 

σ yy ¼ ð1  2νÞeyy þ ν exx þ eyy þ ezz
ð1 þ νÞð1  2νÞ
(5.29)
E 

σ zz ¼ ð1  2νÞezz þ ν exx þ eyy þ ezz
ð1 þ νÞð1  2νÞ
E E E
σ xy ¼ exy , σ yz ¼ γyz , σ zx ¼ ezx
ð1 þ νÞ ð 1 þ νÞ ð1 þ νÞ
which now can be written, finally in the compact form
! !
E ν X3
σ ij ¼ eij þ ekk δij ði; j ¼ 1; 2, 3Þ (5.30)
ð1 þ νÞ ð1  2νÞ i¼1

where δij is the Kronecker delta (see Eq. (2.34)).

Now, consider the compliance matrix where we have
8 9 2 38 9
>
> exx >> 1=E ν=E ν=E 0 0 0 > > σ xx >
>
>
> >
> 6 > σ yy >
> >
>
> e yy > ν=E 1=E ν=E 0 0 0 77> >
<e > = 66
>
< >
=
ν=E ν=E 1=E 0 0 0 7 σ
¼6 7
zz zz
6 (5.31)
> γ
> yz > > 6 0 0 0 1=G 0 0 77>
> σ yz >
>
>
> > 4 > >
> γxz >
> >
> 0 0 0 0 1=G 0 >5>
> σ xz >
>
>
: ; : ;
γxy 0 0 0 0 0 1=G σ xy
and G ¼ E=2ð1 þ νÞ. This gives Eq. (5.4):
σ xx ν

exx ¼  σ yy þ σ zz
E E
σ yy ν
eyy ¼  ðσ xx þ σ zz Þ
E E
(5.32)
σ zz ν

ezz ¼  σ xx þ σ yy
E E
σ xy σ yz σ zx
γxy ¼ , γyz ¼ , γzx ¼
G G G
 5.2 Plane Stress and Plane Strain 129

Again, we can rearrange these equations to write them in matrix terms. Equivalently, we
have
ð1 þ νÞσ xx ν

exx ¼  σ xx þ σ yy þ σ zz
E E
ð1 þ νÞσ yy ν

eyy ¼  σ xx þ σ yy þ σ zz
E E
(5.33)
ð1 þ νÞσ zz ν

ezz ¼  σ xx þ σ yy þ σ zz
E E
ð1 þ νÞσ xy ð1 þ νÞσ yz ð1 þ νÞσ zx
exy ¼ , eyz ¼ , ezx ¼
E E E
which can be written, finally, as
! !
1 X3
eij ¼ ð1 þ νÞσ ij  ν σ ij δij ði; j ¼ 1; 2, 3Þ (5.34)
E i¼1

One important difference between the isotropic case and the others listed above is that
while the matrix of constants for the isotropic case is the same for any orientation of the
ðx; y; zÞ axes associated with the stresses and strains, this is not true in general for other
materials and the expressions given previously are valid only for a particular set of mater-
ials axes associated with the symmetries present in the materials. We will give the trans-
formation equations shortly that will allow us to obtain the matrix of constants as
measured in any coordinate system. These transformations, like the stress and strain
transformations, involve the direction cosines relating a set of ðx0 , y0 , z0 Þ axes to the
ðx; y; zÞ axes in which the above stress–strain relations are written. Another important
difference between isotropic and anisotropic media is that the principal stress and principal
strain directions do not coincide in general for anisotropic materials so that we need to
calculate those directions (and the corresponding principal stress and principal strain
values) separately for the stress and strain.

5.2 Plane Stress and Plane Strain

Two important special cases for isotropic materials we have mentioned previously are the
cases of plane stress and plane strain. In the plane-stress case, it is assumed that there are no
normal and shear stresses in, say, the z-direction, while the stresses acting in the x- and
y-directions are functions only of ðx; yÞ, i.e.,
σ zx ¼ σ zy ¼ σ zz ¼ 0
σ xx ¼ σ xx ðx; yÞ
(5.35)
σ yy ¼ σ yy ðx; yÞ
σ xy ¼ σ xy ðx; yÞ
130 Stress–Strain Relations

Figure 5.4 (a) A plane-stress

problem where a thin plate lying in
y
y
the x–y plane is pressurized at an
interior hole. (b) A plane-strain x
problem where a long cylinder is z
pressurized at an interior hole and is z x
constrained by rigid end containing
walls (not shown) that prevent
deformation in the z-direction.

(a) (b)

The case where a thin plate is acted upon by pressure at an interior hole (see Fig. 5.4a) is an
example where we might expect to see a state of plane stress. In this case, generalized
Hooke’s law for an isotropic material reduces to equations only for the in-plane stresses


σ xx , σ yy , σ zz given by
E  
σ xx ¼ exx þ νeyy
1ν 2

E   (5.36)
σ yy ¼ eyy þ νexx
1ν 2

σ xy ¼ Gγxy

but note that, in addition to the in-plane strains (exx, exy, γxy), there is also a normal strain,
ezz , since
ν 
ezz ¼  σ xx þ σ yy
E
ν   (5.37)
¼ exx þ eyy
ð 1  νÞ

Physically this makes sense since in-plane loads such as the pressure in Fig. 5.4b can cause
changes in the plate thickness, resulting in strains in the thickness direction.
The case of plane strain, in contrast, occurs when there is no displacement in, say, the
z-direction, and the displacements in the x- and y-directions are only functions of ðx; yÞ, i.e.,
uz ¼ 0
ux ¼ ux ðx; yÞ (5.38)
uy ¼ uy ðx; yÞ

or, equivalently, for the strains

ezz ¼ γzx ¼ γzy ¼ 0
exx ¼ exx ðx; yÞ
(5.39)
eyy ¼ eyy ðx; yÞ
γxy ¼ γxy ðx; yÞ
 5.3 Transformation of Elastic Constants 131

which gives the stress–strain relations for an isotropic material as

E  
σ xx ¼ ð1  νÞexx þ νeyy
ð1 þ νÞð1  2νÞ
E   (5.40)
σ yy ¼ ð1  νÞeyy þ νexx
ð1 þ νÞð1  2νÞ
σ xy ¼ Gγxy

But in this case, there is also an out-of-plane normal stress, σ zz , given by

Eν  
σ zz ¼ exx þ eyy (5.41)
ð1 þ νÞð1  2νÞ

This normal stress is present in order to enforce the condition that there can be no displacement
in the z-direction. An example of a plane-strain condition is where a hollow cylinder is
pressurized at its inner radius and is restrained from moving in the z-direction at its ends by
rigid supports (see Fig, 5.4b). One should note, however, that the rigid supports are necessary to
enforce the plane-strain conditions since, as we will show in Chapter 7 (where we consider an
exact solution for a pressurized thick cylinder), if the ends of the cylinder are unloaded and free
to expand the cylinder is in a state of plane stress even though it is not thin in the z-direction.
Comparing the plane-stress and plane-strain cases, we see that the stress–strain relations
are very similar. In the next chapter, when we use these stress–strain relations as part of
complete sets of governing equations for solving problems, we will see that through a simple
transformation of the elastic constants we can turn a plane-stress problem solution for a case
such as the one shown in Fig. 5.4a into a the plane-strain problem solution for a case such as
shown in Fig. 5.4b, and vice versa.

5.3 Transformation of Elastic Constants

 
In two different sets of coordinates fxi g and x0i the stress–strain relations in terms of the
elastic constants are (see Eq. (5.6))
X
3 X
3
σ ij ¼ C ijkl ekl
k¼1 l¼1
(5.42)
3 X
X 3
σ 0ij ¼ C 0ijpq e0pq
p¼1 q¼1

But we have the stress and strain in these coordinates also related through
X
3 X
3
σ 0mn ¼ l im l jn σ ij
i¼1 j¼1
(5.43)
X
3 X
3 X
3 X
3
¼ l im l jn C ijkl ekl
i¼1 j¼1 k¼1 l¼1
132 Stress–Strain Relations

and
X
3 X
3 X
3 X
3
σ 0mn ¼ C 0mnpq l kp l lq ekl (5.44)
p¼1 q¼1 k¼1 l¼1

Equating Eq. (5.43) and Eq. (5.44) and using the fact that this equality must be true for all of
the common strain components, ekl , in this expression we find
X
3 X
3 X
3 X
3
l kp l lq C 0mnpq ¼ l im l jn C ijkl (5.45)
p¼1 q¼1 i¼1 j¼1

If we multiply Eq. (5.45) by l kr l ls and sum over (k, l) we find

X
3 X
3 X
3 X
3
C 0mnrs ¼ l kr l ls l im l jn C ijkl (5.46)
k¼1 l¼1 i¼1 j¼1

Equation (5.46) is the transformation relationship for the elastic constants when we change
 
our coordinate system from the fxi g to the x0i coordinates. However, in practice it is
usually more convenient to use the Voigt notation where the stress–strain relations are
written in terms of a reduced 6  6 constants matrix, C IJ , (I, J ¼ 1, 2,. . ., 6). Thus, we need
to find this transformation for the reduced constants matrix. To do this let us define
M ijkl ¼ l ki l lj (5.47)

Then we can write Eq. (5.46) as

X
3 X
3 X
3 X
3
C 0mnrs ¼ M rskl M mnij C ijkl (5.48)
k¼1 l¼1 i¼1 j¼1

To obtain the reduced form of this equation we can follow the same procedure used in
obtaining the reduced constants matrix. First, we consider the terms in Eq. (5.48) involving
M mnij only. Omitting the subscripts and sums that are not involved with this matrix, we have
3 X
X 3
C 0mn ¼ M mnij C ij M ...: (5.49)
i¼1 j¼1

which is now a relationship between two symmetrical matrices C 0m n and C ij analogous to the
relationship between the stress and strain matrices. Thus, like the stress–strain relations, we
can express Eq. (5.49) in terms of a 6  6 matrix M MI , where
X
6
C M ¼ M MI C I  M  ðM ¼ 1, 2 . . . ; 6Þ (5.50)
I ¼1

In the same fashion we can write Eq. (5.48) as

X
3 X
3
C 0rs ¼ M rskl C kl M  (5.51)
k¼1 l¼1
 5.3 Transformation of Elastic Constants 133

which is again a relationship between two symmetric matrices that can be written in terms of
the reduced matrix M RK , where
X
6
C 0R ¼ M RK C K M  ðR ¼ 1, 2 . . . ; 6Þ (5.52)
K¼1

Combining Eq. (5.50) and Eq. (5.52) we obtain

6 X
X 6
C 0MR ¼ M MI C IK M RK ðM; R ¼ 1, 2 . . . ; 6Þ (5.53)
I ¼1 K¼1

which is the transformation relationship for the reduced constants matrix. The form of the
6  6 matrix ½M  in terms of the M ijkl components is the same as found earlier for the form
of the elastic constants matrix (see Eq. (5.8a)), i.e.,
2 3
M 1111 M 1122 M 1133 M 1123 þ M 1132 M 1113 þ M 1131 M 1112 þ M 1121
6 M 2211 M 2222 M 2233 M 2223 þ M 2232 M 2213 þ M 2231 M 2212 þ M 2221 7
6 7
6 M 3311 M 3322 M 3333 M 3323 þ M 3332 M 3313 þ M 3331 M 3312 þ M 3321 7
6
½M  ¼ 6 7
7
6 M 2311 M 2322 M 2333 M 2323 þ M 2332 M 2313 þ M 2331 M 2312 þ M 3321 7
4 M 1311 M 1322 M 1333 M 1323 þ M 1332 M 1313 þ M 1331 M 1312 þ M 1321 5
M 1211 M 1222 M 1233 M 1223 þ M 1232 M 1213 þ M 1231 M 1212 þ M 1221
(5.54)

which can be written in terms of the direction cosines as

2 2 3
l 11 l 221 l 231 2l 21 l 31 2l 31 l 11 2l 21 l 11
6 2 7
6 l 2
l 22 2
l 32 2l 22 l 32 2l 12 l 31 2l 12 l 22 7
6 12 7
6 2 7
6 l 13 l 223 l 233 2l 23 l 33 2l 13 l 33 2l 13 l 23 7
½M  ¼ 6
6
7
7 (5.55)
6 l 12 l 13 l 22 l 23 l 32 l 33 l 22 l 33 þ l 32 l 23 l 12 l 33 þ l 32 l 13 l 12 l 23 þ l 22 l 13 7
6 7
6l l l 11 l 23 þ l 21 l 13 7
4 11 13 l 21 l 23 l 31 l 33 l 21 l 33 þ l 31 l 23 l 11 l 33 þ l 31 l 13 5
l 11 l 12 l 21 l 22 l 31 l 32 l 21 l 32 þ l 31 l 22 l 11 l 32 þ l 31 l 12 l 11 l 22 þ l 21 l 12

Note that in terms of matrix multiplication we can write the transformation of Eq. (5.53)
more compactly as

½C 0  ¼ ½M ½C ½M T (5.56)

Since the 6  6 matrix ½M  is rather complex, we have written a MATLAB® function

M_Matrix that has the calling sequence
M = M_Matrix(l);

where l is the matrix of direction cosines and the function returns the 6  6 matrix ½M . Note
that the direction cosine matrix ½l  is associated with the coordinate transformation from a
set of fxg coordinates to a set of fx0 g coordinates given by

fx0 g ¼ ½l T fxg (5.57)

134 Stress–Strain Relations

where (see Eq. (2.35))

 
l ij ¼ ei e0j ¼ cos xi ; x0j (5.58)

and some texts define the direction cosines matrix instead as the transpose of ½l . Thus, one
must be careful to use the direction cosines defined by Eq. (5.58) as the input argument in the
M_Matrix function. In the use of the reduced set of 6  6 elastic constants we often need to
transform from 3  3 stress or strain matrices to 6  1 stress or strain vectors, and vice versa.
Thus, we have also written a pair of MATLAB® functions m_to_v and v_to_m that are
called as:
vector = m_to_v(matrix, ‘type’);
matrix = v_to_m(vector, ‘type’);

where matrix can be either a 3  3 matrix of stresses or tensor strains and vector can be either a
6  1 stress or strain vector. The ‘type’ input argument is a string having the value of ‘stress’ or
‘strain’ to indicate the type of matrix and vector quantities we are considering. The m_to_v
function takes a 3  3 stress or strain matrix and outputs the corresponding stress or strain
vector. If ‘type’ is equal to ‘strain’ then the function transforms the tensor strains in the input
matrix to the engineering shear strains in the output strain vector. Similarly, the function
v_to_m takes a 6  1 stress or strain vector and transforms it to the corresponding 3  3 stress
or strain matrix. If ‘type’ is equal to ‘strain’, then the engineering shear strains present in the
input strain vector are transformed to the tensor strain components in the output strain matrix.

Example 5.1 Stress, Strain, and Stress–Strain Transformations

This example will consider transformations of stress, strain, and stress–strain relations
using MATLAB® to perform all the calculations. Specifically, we will consider an ortho-
tropic material where the only nonzero elastic constants (in GPa) in a set of material
coordinates are
C 11 ¼ 103; C 22 ¼ 50; C 33 ¼ 75;
C12 ¼ C 21 ¼ 55; C 13 ¼ C 31 ¼ 25; , C23 ¼ C 32 ¼ 40;
C 44 ¼ 45; C 55 ¼ 10; C 66 ¼ 27:6

Normally the strains are given in terms of μstrain (multiples of 106 ) and the stresses given in
MPa. In that case we must multiply the elastic constants in GPa by a factor of 103 so that
these constants are in units of MPa/μstrain. Thus, in MATLAB® we have
C = zeros(6,6); % Set up empty 6x6 matrix
C(1,1) = 103; C(2,2) = 50; C(3,3) = 75; % define constants
C(1,2) = 55; C(1,3) = 25; C(2,3) = 40;
C(4,4) = 45; C(5,5) = 10; C(6,6) = 27.6;
 5.3 Transformation of Elastic Constants 135

C(2,1) = C(1,2); C(3,1) = C(1,3); C(3,2) = C(2,3);

C = C*10^-3 % put C in units of
% MPa/microstrain
C =
0.1030 0.0550 0.0250 0 0 0
0.0550 0.0500 0.0400 0 0 0
0.0250 0.0400 0.0750 0 0 0
0 0 0 0.0450 0 0
0 0 0 0 0.0100 0
0 0 0 0 0 0.0276

Now, let us consider the following state of strain (in μstrain) as given in the material
coordinates:
strain = [300 50 20; 50 200 30; 20 30 100] % state of
% strain matrix
strain = 300 50 20
50 200 30
20 30 100

We can change this strain matrix into a strain vector, e:

e = m_to_v(strain, ‘strain’) % change strain matrix to strain
% vector
e = 300
200
100
60
40
100

where we see that the tensor shear strains have been transformed to engineering shear
strains. We can multiply these strains by the reduced elastic constants matrix to obtain the
stress in vector form and change it into a stress matrix:
s = C*e % calculate the stress vector
s = 44.4000
30.5000
23.0000
2.7000
0.4000
2.7600
136 Stress–Strain Relations

stress = v_to_m(s, ‘stress’) % change stress vector to stress

% matrix
stress = 44.4000 2.7600 0.4000
2.7600 30.5000 2.7000
0.4000 2.7000 23.0000

Having the state of stress in the material coordinates, we can calculate the principal stresses
and principal stress directions with the MATLAB® eig function
[pdirs, pstress] = eig(stress) % calculate principal stress
% directions and principal stresses
pdirs = 0.0217 -0.1980 -0.9800
-0.3130 0.9296 -0.1948
0.9495 0.3110 -0.0418
pstress = 22.1190 0 0
0 30.8154 0
0 0 44.9656

If we check the determinant of pdirs, we see it is plus one:

det(pdirs)
ans = 1.0000

so the principal directions matrix, pdirs, is just the matrix of direction cosines, ½l , that relate
the principal coordinates to the material coordinates (otherwise we need to change the sign
on one of the columns of pdirs, as discussed Chapter 2). Thus, we can calculate the strains in
the principal stress coordinates, strain_new, by using the transformation of the state of strain
from material to principal stress coordinates (see Eq. (4.33)):
l = pdirs; % rename direction cosines matrix
strain_new = l'*strain*l % calculate strains in principal
% stress coordinates
strain_new = 92.2052 -5.9178 -6.8183
-5.9178 190.7245 -31.8258
-6.8183 -31.8258 317.0703

However, these are not principal strains (obviously, since the shear strains are not zero) because
the principal strain directions are not the same as the principal stress directions for this
orthotropic material. We can calculate the principal stresses from these strains in the principal
stress coordinates directly from the stress–strain relations if we know the reduced constants
matrix in the principal stress coordinates. But we know how to transform this constants matrix.
We need the M-transformation matrix, which we can get from the M_Matrix function
M = M_Matrix(l) % calculate M matrix needed for
% transforming the constants matrix
 5.3 Transformation of Elastic Constants 137

% to the principal stress coordinates

M = 0.0005 0.0980 0.9015 -0.5945 0.0413 -0.0136
0.0392 0.8641 0.0967 0.5782 -0.1232 -0.3681
0.9603 0.0379 0.0017 0.0163 0.0819 0.3817
0.1940 -0.1811 -0.0130 -0.0994 -0.2965 -0.8724
-0.0213 0.0610 -0.0397 -0.1719 -0.9314 0.3025
-0.0043 -0.2910 0.2953 0.7852 -0.1813 0.0822

and then the new constants matrix is just

C_new = M*C*M' % find the constants matrix in principal

% stress coordinates
C_new = 0.0845 0.0281 0.0280 -0.0001 0.0032 -0.0120
0.0281 0.0677 0.0502 0.0076 -0.0062 0.0179
0.0280 0.0502 0.1033 -0.0002 0.0025 -0.0075
-0.0001 0.0076 -0.0002 0.0241 -0.0038 -0.0063
0.0032 -0.0062 0.0025 -0.0038 0.0126 -0.0041
-0.0120 0.0179 -0.0075 -0.0063 -0.0041 0.0322

which we see looks nothing like our original constants matrix. Changing the strain matrix
strain_new in these principal stress coordinates to a corresponding strain vector, e_new, we have

e_new = m_to_v(strain_new,‘strain’) % change strain matrix to

% strain vector
e_new = 92.2052
190.7245
317.0703
-63.6516
-13.6365
-11.8356

which we then can use to calculate the stress in vector form in these principal stress
coordinates as

s_new = C_new*e_new % use stress-strain relations in

% principal stress coordinates
% to calculate stress in vector form
S_new = 22.1190
30.8154
44.9656
-0.0000
-0.0000
0.0000
138 Stress–Strain Relations

and changing this vector to a stress matrix we have, finally

stress_new = v_to_m(S_new, ‘stress’) % obtain stress matrix
% in principal stress coordinates
stress_new = 22.1190 0.0000 -0.0000
0.0000 30.8154 -0.0000
-0.0000 -0.0000 44.9656

which is identical with the principal stress matrix, pstress, calculated previously from eig. This
example should give you a good feel for how we can use the stress and strain transformations
and the transformation of the elastic constants matrix to solve problems. Now, let us go back
to the question of the principal strains and the principal strain directions. We had the state of
strain matrix in the original material coordinates given by the matrix strain so let’s use eig to
calculate the principal strains and principal strain directions:

[strain_dirs, pstrain] = eig(strain) % calculate principal

% strain directions and principal strains
strain_dirs = -0.0322 -0.4169 -0.9084
-0.2525 0.8827 -0.3962
0.9670 0.2166 -0.1337
pstrain = 91.4995 0 0
0 183.7468 0
0 0 324.7537

and we see the matrix of principal strain directions, strain_dirs, is very different from the
principal stress directions matrix, pdirs, calculated previously.
Before leaving this example let us consider the case of an isotropic material where, in
material coordinates (in GPa), we have

C 11 ¼ C 22 ¼ C 33 ¼ 50;
C 12 ¼ C 21 ¼ C 13 ¼ C 31 ¼ C 23 ¼ C 32 ¼ 20;
C 44 ¼ C 55 ¼ C 66 ¼ ðC 11  C 12 Þ=2 ¼ 15

Thus, the elastic constants matrix, C2, in MPa/μstrain can be calculated in MATLAB® as

C2 = zeros(6,6); % form up constants matrix for

% isotropic material
C2(1,1) = 50; C2(2,2) = C2(1,1); C2(3,3) = C2(1,1);
C2(1,2) = 20; C2(2,1) = C2(1,2); C2(1,3) = C2(1,2); C2(3,1) = C2(1,2);
C2(2,3) = C2(1,2); C2(3,2) = C2(1,2);
C2(4,4) = 15; C2(5,5) = C2(4,4); C2(6,6) = C2(4,4);
C2 = C2*10^-3
 5.4 States of Stress and Strain on a Surface 139

C2 = 0.0500 0.0200 0.0200 0 0 0

0.0200 0.0500 0.0200 0 0 0
0.0200 0.0200 0.0500 0 0 0
0 0 0 0.0150 0 0
0 0 0 0 0.0150 0
0 0 0 0 0 0.0150

If we use the M matrix for the principal stress directions calculated previously to examine the
elastic constants matrix in those coordinates, we find
C2_new = M*C2*M' % constants matrix in principal stress
% coordinates
C2_new = 0.0500 0.0200 0.0200 0.0000 -0.0000 0.0000
0.0200 0.0500 0.0200 0.0000 -0.0000 0.0000
0.0200 0.0200 0.0500 0.0000 -0.0000 0.0000
0.0000 0.0000 0.0000 0.0150 0.0000 -0.0000
-0.0000 -0.0000 -0.0000 0.0000 0.0150 0.0000
0.0000 0.0000 0.0000 -0.0000 0.0000 0.0150

which has not changed from our original constants matrix, showing that the stress–strain
relations are indeed the same in both of these coordinate systems (and, in fact, in any
coordinate system).

5.4 States of Stress and Strain on a Surface from Strain

Gage Measurements
It is normally not possible to measure the states of stress and strain inside a deformable
body. However, if one can perform a series of normal strain measurements on the free
surface of a body one can deduce the states of stress and strain on that surface using the
strain transformations and the stress–strain relations. In this section we will demonstrate this
process for an isotropic material. Consider, for example, a free surface, S, of an isotropic
elastic body, as shown in Fig. 5.5. The x- and y-coordinates are tangent to the surface at
P and the z-axis is normal to the surface. The problem is to find the complete 3-D states of
stress and strain at point P given by
2 3 2 3
exx exy exz σ xx σ xy σ xz
½e ¼ 4 eyx eyy eyz 5, ½σ  ¼ 4 σ yx σ yy σ yz 5 (5.59)
ezx ezy ezz σ zx σ zy σ zz

Since the surface is assumed to be free, as shown in Fig. 5.5, we have σ zz ¼ σ zx ¼ σ zy ¼ 0. It

follows from the stress–strain law for an isotropic material that
140 Stress–Strain Relations

Figure 5.5 A free surface of an s zz = s zx = s zy = 0 z

elastic body. y

P
x
S

Figure 5.6 (a) A resistance strain gage. c b

(b) A strain gage rosette.

a
Tc T
b
Ta
x

(a) (b)

E 

σ xx ¼ ð1  νÞexx þ ν eyy þ ezz
ð1 þ νÞð1  2νÞ
E  
σ yy ¼ ð1  νÞeyy þ νðexx þ ezz Þ
ð1 þ νÞð1  2νÞ
E 
 (5.60)
σ zz ¼ ð1  νÞezz þ ν exx þ eyy ¼ 0
ð1 þ νÞð1  2νÞ
σ xy ¼ 2Gexy ¼ 0
σ xz ¼ 2Gexz ¼ 0
σ yz ¼ 2Geyz ¼ 0

and from the condition σ zz ¼ 0 one finds

ν

ezz ¼  exx þ eyy (5.61)
ð 1  νÞ

so, using the symmetry of the shear strains and shear stresses, the states of strain and stress at
P are
2 3 2 3
exx exy 0 σ xx σ xy 0
½e ¼ 4 exy eyy 0 5, ½σ  ¼ 4 σ xy σ xy 0 5 (5.62)
0 0 ezz 0 0 0

Normal strains on the free surface of a body can be measured with resistance strain gages
such as the one shown in Fig. 5.6a. A gage consists of a grid of wires or thin foil that are
 5.4 States of Stress and Strain on a Surface 141

placed on a thin paper or plastic backing, which can then be cemented to the surface of a
body. The change of length of the wires or foil causes the electrical resistance of the gage to
change and this resistance change can be related to the normal strain along the gage. A set of
three gages placed at a location on the surface is called a strain gage rosette. Figure 5.6b
shows a general rosette configuration where the three gages are oriented at angles ðθa , θb , θc Þ
with respect to an x-axis.
Now, suppose we place a strain gage rosette on the surface where the gages are along the
x- , b-, and y-axes, as shown in Fig. 5.7 , and that these gages measure the normal strains ea ,
eb , ec , respectively. The a and c gages of the rosette, therefore, give us the three normal
strains, exx ¼ ea , eyy ¼ ec , and ezz ¼ νðea þ ec Þ=ð1  νÞ, leaving only the shear strain exy as
unknown. However, from the strain transformation equation (see Eq. (4.32)) we have that
the normal strain in a x01 direction is
e011 ¼ l 11 l 11 e11 þ l 21 l 21 e22 þ l 31 l 31 e33 þ 2l 11 l 21 e12 þ 2l 11 l 31 e13 þ 2l 21 l 31 e23 (5.63)

which, if we take the b-axis along this x01 axis, and the ðx; yÞ axes as the ðx1 , x2 Þ axes, we can
rewrite as
eb ¼ ea cos 2 θ þ ec sin 2 θ þ 2exy sin θ cos θ (5.64)


since (see Eq. (2.35)) l i1 ¼ cos xi ; x01 ði ¼ 1; 2, 3Þ. Equation (5.64) shows that a knowledge of
the strains ðea , eb , ec Þ and the angle θ are all that is needed to calculate the remaining unknown
shear strain, exy , thus giving the complete state of strain on the surface in Eq. (5.62). The
stress–strain law, therefore, also gives us the stresses in the state of stress in Eq. (5.62). Note
that while the state of stress on the surface is a two-dimensional (2-D) state of stress it is not a
state of plane stress since the stresses may be functions of all three coordinatesðx; y; zÞ and not
justðx; yÞ, as assumed for plane stress. To make it easier to do analysis with a rosette, we have
written the MATLAB® function rosette that has the calling sequence
[exx eyy exy] = rosette(anga, angb, angc, ea, eb, ec);

which takes the measured strains ðea , eb , ec Þ and the angles ðθa , θb , θc Þ ¼ ðanga; angb; angcÞ,
as measured in degrees from the x-axis for gagesða; b; cÞ, respectively, and returns the strains


exx , eyy , exy by solving the normal strain equations for the three gages, Eq. (5.63), written as:

ea ¼ exx cos 2 θa þ eyy sin 2 θa þ 2exy sin θa cos θa

z eb ¼ exx cos 2 θb þ eyy sin 2 θb þ 2exy sin θb cos θb (5.65)
y
eb ec ¼ exx cos 2 θc þ eyy sin 2 θc þ 2exy sin θc cos θc
ec
Accurate measurements with strain gages require that one corrects the
b
T gage readings for temperature changes and for sensitivity to transverse
P
x strains (i.e., strains not along the axis for which the strain is to be measured).
S ea

Figure 5.7 A strain gage rosette placed on a free surface.

142 Stress–Strain Relations

We will not discuss those issues here, but details can be found in a number of advanced
strength of materials texts.

5.5 PROBLEMS

P5.1 Although aluminum alloys are often treated as if they are isotropic (E ¼ 69 GPa,
ν ¼ 0:33 are often used as typical values) pure aluminum is in fact a cubic material with
C 11 ¼ 103 Gpa, C 12 ¼ 55 GPa, C 44 ¼ 27:6 GPa. If the state of strain at a point in a
pure aluminum material with respect to a set of (x, y, z) axes in μstrain is
2 3
300 50 0
½e ¼ 4 50 200 0 5
0 0 100
determine:
(a) the principal strains and principal strain directions;
(b) the stress components with respect to the (x, y, z) axes;
(c) the principal stresses and principal directions;
(d) repeat part (b) assuming that the aluminum is isotropic (i.e., use the E, ν values
given above). Determine the principal stresses and principal stress directions. What
are the differences between the principal values in the two cases?
P5.2 Along the orthotropic axes of a piece of birch-wood material, the stress–strain relations
are
8 9 2 38 9
>
> exx >> 72:5 36:25 36:25 0 0 0 >
> σ xx >
>
> eyy >
> > 6 36:25 942:5 652:5 > σ yy >
>
>
> >
> 6 0 0 0 7 7 >
>
>
>
>
<e = 6 7 < =
zz 6 36:25 652:5 1450 0 0 0 7 σ zz
¼6
> γyz >
> > 6 0 0 0 4350 0 0 7 7>> σ yz >
>
>
> >
> >
> >
>
> γxz >>
4 0 0 0 0 1087:6 0 5 >
> σ xz >
>
>
: ; : ;
γxy 0 0 0 0 0 1015 σ xy

where the strains are in μstrain and the stresses are in MPa. The x-axis here is along the
wood grain, the y-axis is radial to the tree, and the z-axis is tangent to the growth rings in
the tree. If the stresses (in MPa) at a point in the wood with respect to the (x, y, z) axes are
2 3
7 1:4 0
½σ  ¼ 4 1:4 2:1 0 5
0 0 2:8
determine the following:
(a) the principal stresses and principal stress directions;
(b) the state of strain with respect to the (x, y, z) axes;
(c) the principal strains and principal strain directions.
 5.5 Problems 143

100 mm Figure P5.1 Measurement of strains on the surface of a

y composite plate.

boron fiber

c b
strain gage 45° 45° a
rosette
30°

x
z

P5.3 A 100 mm-thick composite plate is fabricated from boron-fiber reinforced epoxy, as
shown in Fig. P5.1. A 0-45-90 strain gage rosette is placed on the stress-free x–y surface
of this plate, as shown in Fig. P5.1. The nonzero elastic constants (all in GPa) for this
material as measured in the (x, y, z) coordinates are given by
C 11 ¼ 209
C 22 ¼ C 33 ¼ 29:4
C 12 ¼ C 13 ¼ 24:6
C 23 ¼ 13:7
C 44 ¼ 5:9
C 55 ¼ C 66 ¼ 8:14
The strains (in μstrain) measured by the strain gages are
ea ¼ 385
eb ¼ 70
ec ¼ 285

(a) Based on these strain measurements, what are the exx , eyy , and exy strains?
(b) What is the complete 3-D state of strain at the location of the rosette gage in (x, y, z)
coordinates?
(c) What is the complete 3-D state of stress at the location of the rosette gage in the
(x, y, z) coordinates?
(d) What is the change of thickness of the plate, assuming the strains are uniform
throughout the plate?
(e) What are the principal strains and principal strain directions at the location of the
rosette gage, and the corresponding principal stresses and principal stress
directions?
P5.4 A rubber cylinder with Young’s modulus, E R , and Poisson’s ratio, νR , is compressed
inside a steel tube by an axial stress, σ 0 , as shown in Fig. P5.2. The Young’s modulus of
144 Stress–Strain Relations

t Figure P5.2 A rubber cylinder being compressed inside a thin steel tube.
r

the steel is E S and its Poisson’s ratio is νS . Determine an

expression for the pressure exerted between the rubber and
steel
the steel tube if:
rubber (a) the elasticity of the steel is considered, and
(b) the steel is rigid ðE S ¼ ∞Þ.
(c) Take E R ¼ 0:02 GPa, νR ¼ 0:5, E S ¼ 200 GPa, νS ¼ 0:3,
V0
t ¼ 4 mm, r ¼ 100 mm. Calculate the pressure in terms of
σ 0 for cases (a) and (b). How good is the assumption of
rigidity? You can assume the tube is sufficiently thin so that the hoop stress in the tube
can be considered to be uniform across the thickness.
P5.5 Consider the following 3-D state of strain (in μstrain) with respect to a set of (x, y, z)
axes:
2 3
200 300 0
½e ¼ 4 300 100 200 5
0 200 100

(a) If this state of strain exists in a component made of stainless steel with E = 200 GPa,
ν ¼ 0:3, determine the principal stresses and principal stress directions.
(b) Alternatively, consider stainless steel as a transversely anisotropic material (which
it is) where the stress–strain relationship with respect to a set of (x, y, z) material
axes is given by
8 9 2 38 9
>
> σ xx >
> C 11 C 12 C 13 0 0 0 >
> exx >>
>
> >
σ yy > 6 C 12 C 11 C 13 7>> eyy >>
>
> >
> 0 0 0 7>> >
>
< = 6 6 7 < =
σ zz C C C 0 0 0 e
¼6 7
13 13 33 zz
6 7> γyz >
> σ yz >
>
> > 6 0
>
0 0 C 44 0 0 7>> >
>
> > 4 5> >
> σ xz >
>
: >
;
0 0 0 0 C 44 0 > γxz >
>
: >
;
σ xy 0 0 0 0 0 ðC 11  C 12 Þ=2 γxy

where
C 11 ¼ 262:7 GPa, C 33 ¼ 216 GPa, C 44 ¼ 129 GPa
C 12 ¼ 98:2 GPa, C 13 ¼ 145 GPa
For the same state of strain as in part (a), determine the principal stresses and
principal stress directions.
P5.6 A 0-45-90 strain gage rosette is placed on the free surface of a linear elastic isotropic
component whose Young’s modulus and Poisson’s ratio are E ¼ 150 GPa, ν ¼ 0:3. The
angle the gage makes with respect to the longitudinal (x-axis) of the component is 25
so the gage angles from the x-axis are ðθa , θb , θc Þ ¼ ð25, 70, 115Þ , respectively. The
measured strains are ðea , eb , ec Þ ¼ ð77:5, 51:8, 82:5Þ μstrain, respectively.
 5.5 Problems 145

(a) Determine the state of strain at the gage location and from the stress–strain
relations the corresponding state of stress. Using this state of stress, determine
the principal stresses and principal stress directions.
(b) From the state of strain at the gage location, determine the principal strains and
principal strain directions directly. Use the stress–strain relations to determine the
corresponding principal stresses.
P5.7 The strains (in μstrain) measured on a free surface of a component with a rosette are
ea ¼ 400, eb ¼ 300, ec ¼ 50 for θa ¼ 30 , θb ¼ 30 , θc ¼ 90 . Determine the magni-
tude of the maximum engineering shear strain at the gage. Let Poisson’s ratio ν ¼ 1=3.
6 Governing Equations and Boundary Conditions

Previous chapters have examined a wide variety of topics including equilibrium, compatibility,
strain–displacement relations, and stress–strain relations. When these elements are combined,
we can form up different complete sets of governing differential and algebraic equations. In
order to solve those sets of equations we must also specify the conditions that arise from
having known loads or geometric constraints. These are called the boundary conditions for the
problem. As discussed in Chapter 1, boundary conditions are one of the four pillars of stress
analyses. In this chapter we will examine some of the choices we have for formulating
complete sets of the governing equations and how those governing equations can be combined
with appropriate boundary conditions to solve for the stresses and deformations. We will also
discuss the principle of Saint-Venant, which gives us some flexibility in how we specify the
boundary conditions.
In general, we can use either the displacements or the stresses as the fundamental
unknowns. Solving for the stresses is often done with the use of auxiliary potential functions
to help simplify the governing equations. We will see an example of that simplification for
solving stress-based 2-D problems with the use of an Airy stress function.
The governing equations for deformable bodies are combinations of partial differential
and algebraic equations so that obtaining solutions of complex problems is difficult and not
formulated in a way that is well suited to numerical computation. However, we will see that
problems can alternatively be expressed in terms of algebraic matrix–vector governing
equations that can be easily solved with computers. Both displacement-based and stress-
based approaches will be outlined that can solved with linear algebra. These approaches will
be shown to be the counterparts of the governing differential/algebraic equations discussed
originally. A classical deformable body problem, Navier’s table problem, will be used as an
example of these purely algebraic methods.

6.1 Governing Equations in Three Dimensions

In order to have a complete set of governing equations we must (1) identify the unknown
variables, and (2) produce a set of differential/algebraic equations that are equal in number
to the number of unknowns present. In this section we will list some of the common
combinations for three-dimensional (3-D) problems.
The most general way to develop a complete set of equations is by taking all the
unknowns present (the stresses, the strains, and the displacements) as the unknowns. In that
case we can form up the following set of governing equations.

146
 6.1 Governing Equations in Three Dimensions 147

3-D Governing Set 1

No. of Equations No. of Unknowns

∂σ yx
Equilibrium ∂σ∂xxx þ þ  þ fx ¼ 0
∂y 3 6 stresses
Stress–strain σ xx ¼ ð1þνÞEð12νÞ ½ð1  νÞexx þ    6 6 strains
Strain–displacement exx ¼ ∂u
∂x   
x
6 3 displacements
TOTAL 15 15

The compatibility equations do not appear because if the strain–displacement relations

are appropriately differentiated, the displacements can be eliminated, resulting in the com-
patibility equations. Thus, the compatibility equations are still implicitly present. Here there
are a very large number of unknowns. However, we can place the strain–displacement
relations into the stress–strain equations, which would express the stresses directly in terms
of the displacement gradients, and then place those stress–displacement gradient relations
into the equilibrium equations to arrive at three equations for the three displacements only.
To see this reduction, consider a general linear anisotropic material, where the equations of
this governing set can be written as
X
3
∂σ ij
þfj ¼ 0 ðj ¼ 1; 2, 3Þ
i¼1
∂xi

X
3 X
3
σ ij ¼ C ijkl ekl ði; j ¼ 1; 2, 3Þ (6.1)
k¼1 l¼1
 
1 ∂uk ∂ul
ekl ¼ þ ðk; l ¼ 1; 2, 3Þ
2 ∂xl ∂xk

Placing the strain–displacement relations into the stress–strain relations and using the
 
symmetry of the elastic constants C ijkl ¼ C ijlk gives
X
3 X
3
∂uk
σ ij ¼ C ijkl (6.2)
k¼1 l¼1
∂xl

Placing Eq. (6.2) into the equilibrium equations then gives three equations for the three
unknown displacements forming an alternative set of equations:
X
3 X
3 X
2
∂2 uk
C ijkl þ f j ¼ 0 ðj ¼ 1; 2, 3Þ (6.3)
i¼1 k¼1 l¼1
∂xi ∂xl

When the material is a linear isotropic material, these equations become Navier’s equations
for the displacements:
148 Governing Equations and Boundary Conditions

X 3
∂2 uj X 3
∂2 uk
μ þ ðλ þ μÞ þ f j ¼ 0 ðj ¼ 1; 2, 3Þ (6.4)
k¼1
∂xk ∂xk k¼1
∂xk ∂xj

where the Lamé constants ðλ; μÞ are μ ¼ G, λ ¼ Eν=ð1 þ νÞð1  2νÞ in terms of the shear
modulus, Young’s modulus, and Poisson’s ratio constants ðG; E; νÞ, respectively, where we
recall G ¼ E=2ð1 þ νÞ. These displacement equations, therefore, provide an alternate com-
plete set of governing equations. For the general anisotropic case of Eq. (6.3) we have
the following.

3-D Governing Set 2

No. of Equations No. of Unknowns

Equilibrium 3 3 displacements
P
3 P 3 P 2
C ijkl ∂x∂ i u∂xk l þ f j ¼ 0 ðj ¼ 1; 2, 3Þ
2

i¼1 k¼1 l¼1

TOTAL 3 3

A displacement formulation of the governing equations is very attractive since there are
very few unknowns (the three displacements), and these unknowns can be obtained directly
from the three equilibrium equations of Eq. (6.4). A displacement formulation is also a
commonly used approach for solving complex problems numerically. Historically, the finite
differences method was one of the first numerical methods to be used. In this case, differential
operators such as those appearing in Eq. (6.4) are approximated by finite differences of
displacements at discrete points, resulting in a matrix system of algebraic difference equations
that are then solved under appropriate boundary conditions. An alternate numerical method
is to use a set of fundamental solutions to Navier’s equations to recast the differential
equations of Eq. (6.4) into a set of integral equations involving the displacements on the
boundary of the deformable body and then to approximate those integral equations as a
system of linear equations which are solved numerically. Displacements and stresses through-
out the body can then be obtained from the known boundary displacements. This approach is
called the boundary integral equation method or the boundary element method. A third
approach uses work–energy methods (see Chapter 8) as an alternative way to satisfy the
equilibrium equations. In this case the equilibrium equations (written in terms of the displace-
ments) are replaced by a set of equivalent matrix–vector equations ½K fug ¼ fPg, where fug is
a vector of displacements at discrete locations (called nodes) in a body and fPg is a vector of
known discrete loads. Once these nodal displacements are known, they can be used to obtain
the stresses in the body. The matrix ½K  is known as the stiffness matrix. This method, which is
called the stiffness-based finite element method, is by far the most commonly used numerical
method for solving for the displacements and stresses in deformable bodies.
 6.1 Governing Equations in Three Dimensions 149

In Chapter 9 we will outline the finite element method and the boundary element method
in more detail. We will, however, not give a comprehensive treatment of those displacement-
based methods. There are many books available that are excellent sources ([1], [2]). We will
not describe the finite difference method in this book as this approach is now used infre-
quently but you can find more details in some advanced strength of materials or elasticity
texts (see [3], for example).
Another alternative for generating a complete set of governing equations is to use the
stress–strain relations in the compatibility equations to write those equations in terms of
the stresses. The details are rather lengthy so they are given in detail in Appendix B for the
case of an isotropic linear elastic material. The compatibility equations in this form are
called the Beltrami–Michell compatibility equations given by
 
1 X 3
∂2 σ kk ν X
3
∂f j ∂f i ∂f l
r2 σ il þ ¼ δil  þ ði; l ¼ 1; 2, 3Þ (6.5)
1 þ ν k¼1 ∂xi ∂xl 1  ν j¼1 ∂xi ∂xl ∂xi

where ν is Poisson’s ratio, fi are the components of the body force, and δil is the Kronecker
delta symbol (see Eq. (2.34)). Because the stresses are symmetric, Eq. (6.5) represents six
equations, but again only three of them are independent. Thus, we can form a complete set
of governing equations as follows.

3-D Governing Set 3

No. of Equations No. of Unknowns
∂σ xx ∂σ yx 3 6 stresses
Equilibrium þ þ  þ fx ¼ 0
∂x ∂y
∂2 σ xy ∂2 σ xy
Compatibility þ þ  ¼ 0 3 (independent) –
∂x2 ∂y2
TOTAL 6 6

In this formulation, one can introduce a set of auxiliary functions (also called stress
potential functions), which automatically satisfy the equilibrium equations, thus reducing
the number of equations and unknowns. The Maxwell stress function and Morera stress
function are examples that are often discussed in texts on elasticity theory. In this book we
will only see examples of such functions in simpler contexts, including the use of the Airy stress
function for planar problems (which will be discussed later in this chapter and in Chapter 7)
and the Prandtl stress function for torsion problems (see Chapter 11). Later in this chapter we
will see that we can also develop a discrete stress-based formulation that combines the
equilibrium and compatibility equations similar to the 3-D Governing Set 3. As discussed
briefly in Section 6.6, this discrete stress formulation can also be developed as a force-based
finite element method, a method that will be discussed in more detail in Chapter 9.
150 Governing Equations and Boundary Conditions

6.2 Governing Equations in Two Dimensions

6.2.1 Plane Stress
In two-dimensional (2-D) problems, there are very similar choices that can be made for the
sets of governing equations. For example, for plane-stress problems we can take the stresses,
strains, and displacements as the unknowns for the following set.

2-D Governing Set 1 (Plane Stress)

No. of Equations No. of Unknowns

Equilibrium 2 3 stresses
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
Stress–strain 3 3 strains
E  
σ xx ¼ exx þ νeyy
1  ν2
E  
σ yy ¼ eyy þ νexx
1  ν2
σ xy ¼ Gγxy
Strain–displacement 3 2 displacements
∂ux
exx ¼
∂x
∂uy
eyy ¼
∂y
∂ux ∂yy
γxy ¼ þ
∂y ∂x
TOTAL 8 8

The plane-stress stress–strain relations can be written compactly as

Eν X 2
σ mn ¼ ekk δmn þ 2Gemn ðm; n ¼ 1; 2Þ (6.6)
1  ν2 k¼1

The in-plane strain–displacement relations and the equilibrium equations are likewise
 
1 ∂um ∂un
emn ¼ þ ðm; n ¼ 1; 2Þ (6.7)
2 ∂xn ∂xm

and
X
2
∂σ mn
þ f n ¼ 0 ðn ¼ 1; 2Þ (6.8)
m¼1
∂xm
 6.2 Governing Equations in Two Dimensions 151

Combining the strain–displacement and stress–strain relations gives

 
Eν X 2
∂uk ∂um ∂un
σ mn ¼ δmn þ G þ ðm; n ¼ 1; 2Þ (6.9)
1  ν2 k¼1 ∂xk ∂xn ∂xm

and placing this equation into the equilibrium equations, we find, after some algebra,
X2
∂2 un E X2
∂2 uk
G þ þfn ¼ 0 ðn ¼ 1; 2Þ (6.10)
k¼1
∂xk ∂xk 2ð1  νÞ k¼1 ∂xn ∂xk

which are Navier’s equations for plane stress. These equations form a second set of governing
equations.

2-D Governing Set 2 (Plane Stress)

No. of Equations No. of Unknowns

Equilibrium 2 2 displacements
P
2 ∂ un
2
E X2
∂2 uk
G þ þfn ¼ 0
k¼1 ∂xk ∂xk 2ð1  νÞ k¼1 ∂xn ∂xk
TOTAL 2 2

Finally, consider the plane-stress case where we use the stresses as the basic unknowns. In
this case there is only one compatibility equation
∂2 exy ∂2 exx ∂2 eyy
2   ¼0 (6.11)
∂x∂y ∂y2 ∂x2

and the two equilibrium equations (using the symmetry of the shear stress)
∂σ xx ∂σ xy
þ þfx ¼ 0
∂x ∂y
(6.12)
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y

The stress–strain relations are (for the strains written in terms of the stresses)
1 
exx ¼ σ xx  νσ yy
E
1 
eyy ¼ σ yy  νσ xx (6.13)
E
1 ð1 þ νÞ
exy ¼ σ xy ¼ σ xy
2G E
152 Governing Equations and Boundary Conditions

so that placing these stress–strain relations into the compatibility equation gives

∂2 σ xy ∂2 σ xx ∂2 σ yy ∂2 σ yy ∂2 σ xx
2ð1 þ νÞ  þν  þν ¼0 (6.14)
∂x∂y ∂y 2 ∂y 2 ∂x 2 ∂x2
But if we take ð1 þ νÞ times the x-derivative of the first equilibrium equation and ð1 þ νÞ
times the y-derivative of the second equilibrium equation and add the two resulting equa-
tions together, we find
 2 
∂2 σ xy ∂ σ xx ∂2 σ yy ∂f ∂f y
2ð1 þ νÞ ¼ ð 1  νÞ þ  ð 1 þ νÞ x  ð 1 þ νÞ (6.15)
∂x∂y ∂x2 ∂y2 ∂x ∂y

If we place Eq. (6.15) into Eq. (6.14) we obtain, finally, the plane-stress equivalent of the
Beltrami–Michell equations:
     
∂2 σ xx þ σ yy ∂2 σ xx þ σ yy ∂f x ∂f y
þ ¼ ð1 þ νÞ þ (6.16)
∂x2 ∂y2 ∂x ∂y

which is often written in the more compact form

 
r2 σ xx þ σ yy ¼ ð1 þ νÞrf (6.17)

where r2 ¼ ∂2 =∂x2 þ ∂2 =∂y2 is the Laplacian operator and r ¼ ex ∂=∂x þ ey ∂=∂y is the
gradient operator, with f ¼ f x ex þ f y ey being the vector body force. Thus, with the com-
patibility equation written in terms of the stresses, we have the governing set of equations.

2-D Governing Set 3 (Plane Stress)

No. of Equations No. of Unknowns

Equilibrium 2 3 stresses
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
Compatibility 1 –
 
r2 σ xx þ σ yy ¼ ð1 þ νÞrf
TOTAL 3 3

The use of a stress potential function in this case is particularly effective to simplify these
equations and reduce their number. Consider, for example, if we assume that the body forces
can be written in terms of a potential function, V, i.e.,
∂V ∂V
fx ¼  , fy ¼  (6.18)
∂x ∂y
 6.2 Governing Equations in Two Dimensions 153

Common forces such as constant forces or the force of gravity, for example, can be
represented in this manner. Also, let the stresses be given in terms of V and a potential
function, ϕ, called the Airy stress function, where

∂2 ϕ
σ xx ¼ þV
∂y2
∂2 ϕ
σ yy ¼ þV (6.19)
∂x2
∂2 ϕ
σ xy ¼ 
∂x∂y

Then the equilibrium equations are satisfied identically and the compatibility equation
becomes
r4 ϕ ¼ ð1  νÞr2 V (6.20)

where r4 ¼ ∂4 =∂x4 þ 2∂4 =∂x2 ∂y2 þ ∂4 =∂y4 is called the biharmonic operator, which we could
  
also write as r4 ¼ ∂2 =∂x2 þ ∂2 =∂y2 ∂2 =∂x2 þ ∂2 =∂y2 ¼ r2 r2 , i.e., it is the Laplacian
operator applied twice. So far, we have been using Cartesian coordinates, but this is not
essential. For example, in polar coordinates ðr; θÞ the equilibrium equations are (see Eq. (3.73))
∂σ rr σ rr  σ θθ 1 ∂σ θr
þ þ þfr ¼ 0
∂r r r ∂θ
(6.21)
∂σ rθ 2σ rθ 1 ∂σ θθ
þ þ þfθ ¼ 0
∂r r r ∂θ

If we express the body force in terms of a potential, V ðr; θÞ, we have

∂V 1 ∂V
fr ¼  , fθ ¼  (6.22)
∂r r ∂θ

so that if we write the stresses in terms of the Airy stress function, ϕðr; θÞ, as

1 ∂ϕ 1 ∂2 ϕ
σ rr ¼ þ þV
r ∂r r2 ∂θ2
∂2 ϕ
σ θθ ¼ þV (6.23)
∂r2
 
∂ 1 ∂ϕ
σ rθ ¼ 
∂r r ∂θ

the equilibrium equations will again be satisfied identically. Now consider the compatibility
equation. We can write that equation in a completely coordinate invariant form as

r2 I 1 ¼ ð1 þ νÞr2 V (6.24)
154 Governing Equations and Boundary Conditions

where I 1 ¼ σ xx þ σ yy ¼ σ rr þ σ θθ is the first stress invariant for plane stress. But

∂2 ϕ 1 ∂ϕ 1 ∂2 ϕ
σ rr þ σ θθ ¼ þ þ þ 2V
∂r2 r ∂r r2 ∂θ2 (6.25)
¼ r ϕ þ 2V
2

so that from Eq. (6.24) we have again

r2 ϕ ¼ ð1  νÞr2 V (6.26)

and the biharmonic operator is once more the Laplacian operator in polar coordinates
applied twice:
 2  2 
∂ 1∂ 1 ∂2 ∂ 1∂ 1 ∂2
r2 ¼ þ þ þ þ
∂r2 r ∂r r2 ∂θ2 ∂r2 r ∂r r2 ∂θ2 (6.27)
¼ r2 r2

As we will see in Chapter 7, we can obtain the solution to a number of important canonical
problems with the use of the Airy stress function. Thus, this is an important choice of
governing equations.

2-D Governing Set 4 (Plane Stress)

No. of Equations No. of Unknowns

Compatibility r4 ϕ ¼ ð1  νÞr2 V 1 1

TOTAL 1 1

6.2.2 Plane Strain

Recall that the case of plane strain is where the displacements are given as ux ¼ ux ðx; yÞ,
uy ¼ uy ðx; yÞ, uz ¼ 0, which produces a state of strain where exx ¼ ∂ux =∂x ¼ exx ðx; yÞ,
 
eyy ¼ ∂uy =∂x ¼ eyy ðx; yÞ, γxy ¼ ∂ux =∂y þ ∂uy =∂x ¼ γxy ðx; yÞ, γxz ¼ γyz ¼ γzz ¼ 0. From
the 3-D stress–strain relations (see Eq. (5.5)) it follows that we also have
σ xx ¼ σ xx ðx; yÞ, σ yy ¼ σ yy ðx; yÞ, σ zz ¼ σ zz ðx; yÞ, σ xy ¼ σ xy ðx; yÞ so that the equilibrium
equations reduce to
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
(6.28)
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
 6.2 Governing Equations in Two Dimensions 155

which are identical to the plane-stress case. In this case the stress–strain relations are
E  
σ xx ¼ ð1  νÞexx þ νeyy
ð1 þ νÞð1  2νÞ
E   (6.29)
σ yy ¼ ð1  νÞeyy þ νexx
ð1 þ νÞð1  2νÞ
σ xy ¼ Gγxy

so that in terms of the stresses, strains, and displacements the governing equations are as follows.

2-D Governing Set 1 (Plane Strain)

No. of Equations No. of Unknowns

Equilibrium 2 3 stresses
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
Stress–strain 3 3 strains
E  
σ xx ¼ ð1  νÞexx þ νeyy
ð1 þ νÞð1  2νÞ
E  
σ yy ¼ ð1  νÞeyy þ νexx
ð1 þ νÞð1  2νÞ
σ xy ¼ Gγxy
Strain–displacement 3 2 displacements
∂ux
exx ¼
∂x
∂uy
eyy ¼
∂y
∂ux ∂uy
γxy ¼ þ
∂y ∂x
TOTAL 8 8

If we compare this plane-strain set of equations to the plane-stress case, we see that they are
identical in form, but the elastic constants that appear are different. This means that we can
transform a plane-stress to a plane-strain problem, and vice versa, simply by redefining the
elastic constants. For example, if we have a plane-stress solution valid for ðE; ν; G Þ and set
G0 ¼ G
ν0 ν
ν¼ ! ν0 ¼
1  ν0 1þν (6.30)
E E 0 ð1  ν0 Þ E ð1 þ 2νÞ
¼ ! E0 ¼
ð1  ν2 Þ ð1 þ ν0 Þð1  2ν0 Þ ð1 þ νÞ2
156 Governing Equations and Boundary Conditions

we will obtain a plane-strain solution valid forðE 0 , ν0 , G 0 Þ. Similarly, if we have a plane-strain

solution valid for ðE; ν; G Þand set
G0 ¼ G
ν
ν0 ¼
1ν
(6.31)
E ð1  νÞ E0 E
¼h i ! E0 ¼
ð1 þ νÞð1  2νÞ 1  ð ν0 Þ2 ð1  ν2 Þ

we will have a corresponding plane-stress solution valid for ðE 0 , ν0 , G 0 Þ. Table 6.1 summar-
izes these relationships. [Note: the roles of ðE 0 , ν0 , G 0 Þ and ðE; ν; G Þ have been reversed in the
table in order that the solution being generated always has the values ðE; ν; G Þ.] For
example, consider Navier’s equations for plane stress (Eq. (6.10)), which we will rewrite as
X2
∂2 un E0 X2
∂2 uk
G0 þ 0
þ f n ¼ 0 ðn ¼ 1; 2Þ (6.32)
k¼1
∂xk ∂xk 2ð1  ν Þ k¼1 ∂xn ∂xk

Then using Table 6.1 we have

G0 ¼ G
 
E0 E= 1  ν2 E (6.33)
¼ ¼
2ð1  ν0 Þ 2ð1  ν=ð1  νÞÞ 2ð1 þ νÞð1  2νÞ

so that Navier’s equations for plane strain become

X2
∂2 un E X2
∂2 uk
G þ þ f n ¼ 0ðn ¼ 1; 2Þ (6.34)
k¼1
∂xk ∂xk 2ð1 þ νÞð1  2νÞ k¼1 ∂xn ∂xk

We could have also obtained this result directly from the 3-D Navier’s equation (Eq. (6.4))
by noting that plane-strain assumption u3 ¼ 0 eliminates all the terms involving u3 and its
derivatives, so that 3-D form remains the same but the summations only run from one to two.

Table 6.1 Transforming from a plane-stress solution to a plane-strain solution, and vice versa

Given a plane-stress solution for substituting will give a plane-strain solution for
ðE 0 , ν0 , G0 Þ E ðE; ν; GÞ
E0 ¼
ð1  ν2 Þ
ν
ν0 ¼
ð1  νÞ
G0 ¼ G
Given a plane-strain solution for substituting will give a plane-stress solution for
ðE 0 , ν0 , G0 Þ E ð1 þ 2νÞ ðE; ν; GÞ
E0 ¼
ð1 þ νÞ2
ν
ν0 ¼
ð1 þ νÞ
G0 ¼ G
 6.2 Governing Equations in Two Dimensions 157

In terms of the Lamé constants, λ þ μ ¼ E=2ð1 þ νÞð1  2νÞ and μ ¼ G so we see we again
arrive at Eq. (6.34) for plane strain and the governing set of equations are as follows.

2-D Governing Set 2 (Plane Strain)

No. of No. of
Equations Unknowns

Equilibrium 2 2 displacements
X 2
∂2 uj E X2
∂2 uk
G þ þfj ¼ 0
k¼1
∂xk ∂xk 2ð1 þ νÞð1  2νÞ k¼1 ∂xn ∂xj
TOTAL 2 2

Now consider the governing set 3 for plane stress, which consists of the equilibrium and
compatibility equations. In that case the only elastic constant term appearing for the plane-
stress case is ð1 þ ν0 Þ for a plane-stress Poisson ratio value ν0 so that, from Table 6.1,
ν 1
ð1 þ ν0 Þ ¼ 1 þ ¼ (6.35)
ð1  νÞ ð1  νÞ

so that the governing set for plane strain is as follows.

2-D Governing Set 3 (Plane Strain)

No. of Equations No. of Unknowns

Equilibrium 2 3 stresses
∂σ xx ∂σ yx
þ þfx ¼ 0
∂x ∂y
∂σ xy ∂σ yy
þ þfy ¼ 0
∂x ∂y
Compatibility 1 -
  1
r2 σ xx þ σ yy ¼  rf
ð1  ν Þ
TOTAL 3 3

Finally, for the governing set using the Airy stress function, the only elastic constant term
is ð1  ν0 Þ for a plane-stress Poisson ratio value ν0 so
ν ð1  2νÞ
ð1  ν0 Þ ¼ 1  ¼ (6.36)
ð 1  νÞ ð 1  νÞ

and the plane-strain governing set is as follows.

158 Governing Equations and Boundary Conditions

2-D Governing Set 4 (Plane Strain)

No. of Equations No. of Unknowns

ð1  2νÞ 2 1 1
Compatibility r4 ϕ ¼  r V
ð1  ν Þ
TOTAL 1 1

Note that when there are no body forces or when r2 V ¼ 0 the compatibility equation
for either plane stress or plane strain becomes simply r2 ϕ ¼ 0, which is called the biharmo-
nic equation. Since the elastic constants in this case are absent, if the boundary conditions
only involve the stresses we can solve for the stresses and the stress solutions will be
independent of the elastic constants. This will not be true, however, for the strains and
the displacements.
We can transform solutions for the in-plane stresses and displacements from plane stress
to plane strain, and vice versa, but remember that the out-of-plane stresses and strains are
inherently different in the two cases. In the plane-stress case, we have σ zz ¼ 0, but the
corresponding normal strain is
ν  
ezz ¼ exx þ eyy (6.37)
ð1  νÞ

In contrast, for plane strain ezz ¼ 0, but the normal stress is

 
σ zz ¼ ν σ xx þ σ yy (6.38)

Since these out-of-plane stresses or strains are obtained from the behavior of the in-plane
quantities this difference does not affect how we solve either a plane-stress or plane-
strain problem.

6.3 Boundary Conditions

The governing sets of equations are combinations of differential and algebraic equations in
general that must be satisfied throughout the deformable body. In order to solve these sets,
we must also specify conditions on the boundary of the body. Generally, these boundary
conditions arise because the body is being supported in some fashion and because there are
loads being applied to the surface of the body. The applied loads are usually characterized
by specifying the stress vector acting on a portion, S t , of the total surface, S, of the body, i.e.,
TðnÞ ðxs Þ ¼ T∗ ðxs Þ, when the point xs is on S t , where T∗ is a known quantity. Often the
supports are considered to rigidly fix the body over the portion of the surface, S u , where the
body is being supported, i.e., the displacement uðxs Þ ¼ 0 when point xs is on S u . It is
assumed that the total surface of the body, S, is S ¼ S t þ S u so that these stress vector
 6.3 Boundary Conditions 159

C Figure 6.1 (a) A deformable body where

St applied loads are described by the stress
St
T(n ) = T* P
vector acting over a portion S t of the surface
of the body and the displacement is zero
Su fl over the remaining portion, S u . (b) Cases of
Su
T( ) = T* applied loads such as a concentrated force,
n

u=0 P, a concentrated couple, C, and a force/unit

u=0
length, f l , distributed along a line on the
surface. All of these loads can be treated as
(a) (b) special distributions of the stress vector, TðnÞ .

and displacement values characterize the behavior of the solution for every point on the
surface (see Fig. 6.1a). By making the stress vector be very large over a small area of the
surface S t we can simulate a concentrated force, P, or a concentrated couple, C. Similarly,
we can concentrate the stress vector along a line on the surface to simulate a line load (force/
unit length), f l . These cases are illustrated in Fig. 6.1b together with a stress vector distribu-
tion, T∗ . Later, we will examine deformations of a body under some of these different types
of loads, but for the present we can consider all of them as simply special cases of specifying
a known stress vector distribution, T∗ ðxs Þ, on S t . The zero-displacement (rigid) boundary
condition is a highly idealized support condition that may not be satisfied in practice, so one
could relax such a boundary condition by providing some elasticity to the support through
some “spring-like” conditions where the boundary displacement and stresses are related to
each other. Likewise, one could relax the rigid specification by allowing some known
displacement, u∗ ðxs Þ, at the support, i.e., setting uðxs Þ ¼ u∗ ðxs Þ on S u . Finally, we should
note that, in some cases, the boundary conditions may involve combinations of the displace-
ment and stresses at the boundary. However, to have a well-posed boundary value problem
it is not permissible to specify any combination. For example, consider breaking the stress
vector at the surface into its normal- and tangential-stress components by letting

TðnÞ ¼ σ nn n þ σ nt t þ σ ns s (6.39)

where σ nn is the normal stress acting in the unit normal direction, n, and ðσ nt , σ ns Þ are the
shear stresses acting in the tangential directions ðt; sÞ, respectively, along the surface.
Likewise, let the displacement, u, at the surface be represented by its normal and tangential
components, ðun , ut , us Þ (see Fig. 6.2):
u ¼ un n þ ut t þ us s (6.40)

We can specify combinations of these stress and displacement components on the boundary as
long as we do not specify a pair of displacements and stresses in the same direction. For
example, we could set σ nt ¼ σ ns ¼ 0 and un ¼ 0 and this would be acceptable. Such a
boundary condition is called a rigid-smooth boundary condition. Physically, this condition
might occur, for example, where a relatively flexible body is in contact with a much more rigid
160 Governing Equations and Boundary Conditions

Figure 6.2 (a) Normal- and tangential-stress n n

components of the stress vector acting on a V nn un
surface. (b) The corresponding normal- and V nt ut
tangential-displacement components. V ns s us s
t t

(a) (b)

and smooth surface. In the rigid-smooth case, there is nothing to prevent slipping of the body
in the tangential directions (ut and us are unknown) and so no tangential shear stresses are
developed at the surface, but the rigid surface does prevent displacement in the normal
direction, which will produce an unknown normal stress σ nn at the boundary. To see why
specifying pairs of displacements and stresses in the same direction is not acceptable,
consider, for example, setting us ¼ 0, σ ns ¼ 0. If the displacement component in the s-direction
is specified to be zero at a point on the surface, a corresponding unknown shear stress, σ ns ,
must be developed at that point to counteract the displacement in the s-direction produced by
other loads acting on the body. However, the shear stress required is known only after we
solve the governing equations for all the displacements and stresses in the body and we cannot
specify that shear stress to be zero (or any other value), since this would, in general, not be
consistent with the value actually required by the solution. We have used physical reasoning to
show why only particular combinations of stress and displacement components are permis-
sible, but one can also use a mathematical argument and show that it is only under these
permissible combinations that we can guarantee a unique solution (see Chapter 8). Similarly,
specifying displacements (having either zero or nonzero values) on one part of the boundary
and the stress vector on the remaining part or specifying spring-like boundary conditions will
be boundary conditions for which the solution to the governing equations will be unique [4].
Although there are many possibilities for specifying boundary conditions, in this book we
will typically only use special cases of the following set of boundary conditions described
originally, namely
uðxs Þ ¼ u∗ ðxs Þ ¼ 0 on S u
(6.41)
TðnÞ ðxs Þ ¼ T∗ ðxs Þ on S t

when obtaining solutions and discussing solution procedures. Figure 6.3 shows an example
of these types of boundary conditions for a cantilever beam where the components of the
stress vector and displacement are given explicitly.

6.3.1 Saint-Venant’s Principle

In many cases, boundary conditions cannot be known exactly or the modeled boundary
conditions might be different from the actual values. These uncertainties might appear to
 6.3 Boundary Conditions 161

−50 psi on A Figure 6.3 Boundary conditions for a

s zz = cantilever beam.
z 0 otherwise
y s zx = s zy = 0
A
s xx = s xy = s xz = 0

s yy = s yx = s yz = 0 s zz = s zx = s zy = 0
ux = u y = uz = 0
(front and back faces)

Figure 6.4 A simply-supported

beam supported either by (a)
smooth rollers, or (b) by pinned
supports. In case (a) there are
contact stresses over the base of
the beam from the roller, while
in case (b) there are contact
(a) stresses acting on the top of the
hole from the inserted pin.

(b)

call in question many of the solutions we obtain. In elementary strength of materials courses,
such issues are present but usually ignored. For example, consider the two simply supported
beams shown in Figs. 6.4a, b. The applied loading is identical for both beams, but the beam
in Fig. 6.4a sits on roller supports while the beam in Fig. 6.4b is supported by smooth pins.
The stress boundary conditions are obviously quite different for these two cases (as seen in
the insets shown in Figs. 6.4a, b) but those differences are ignored in a strength of materials
solution for the bending and shear stresses since the support end forces and moments are the
same for either case. Thus, in these problems it is implicitly assumed that the stresses in the
beam are the same for both of these cases, at least as long as we are not too close to the ends
where the detailed stresses reflect the boundary conditions acting on the local geometry at
the supports. A “principle” originally due to Adhémar Jean Claude Barré de Saint-Venant
(who will henceforth be called simply Saint-Venant for obvious reasons) can often be used to
help justify such assumptions. Saint-Venant’s principle states:
162 Governing Equations and Boundary Conditions

Statically equivalent systems of surface tractions (stress vectors) applied to a small portion of the surface
enclosing an elastic body are elastically equivalent (i.e., they produce the same stress and strain distribu-
tions) except in the immediate neighborhood of the region where these surface tractions are applied.

Statically equivalent systems of tractions are two different stress vector distributions (which
we will label I and II) that produce the same net force and moment with respect to any point,
P, i.e.,
X X
F ¼ F
I II
X X (6.42)
MP ¼ MP
I II

At the end supports in Figs. 6.4a, b the net vertical forces and moments are the same
so that, according to Saint-Venant’s principle, we could expect that the bending and
shear stresses throughout most of the beam (i.e., not too close to the supports) will be
unaffected.
Saint-Venant’s principle can also be easily illustrated by considering the case of uniaxial
loading of a bar, which is one of the first problems considered in strength of materials
courses. If the applied axial force is P (acting through the centroid of the cross-section) and
the cross-sectional area of the bar is A, the strength of materials stress solution is just
σ xx ¼ P=A, a constant value throughout the bar. If, at the ends of the bar, the applied load
is uniformly distributed across the area A, then the strength of materials stress solution is, in
fact, an exact solution to the governing equations, satisfying both equilibrium, compatibility
and the boundary conditions. This case is shown in Fig. 6.5a (note that the stresses and
applied loads are compressive). If we make an imaginary cut through the bar of Fig. 6.5a at
any section, regardless of how close the cut may be to the ends we obtain the same constant
stress distribution, as shown by the dashed cut seen in Fig. 6.5a. However, now consider the
case where the stress distribution at the ends is concentrated over a very small area at the
centroid (Fig. 6.5b) but where the total force produced by this stress distribution is still P.
Then for cross-sectional cuts near the ends, the stresses will not be uniform and will be larger

Figure 6.5 A bar under uniaxial loading

where (a) the end boundary stresses are
applied uniformly over the ends. (b, c)
A case where the boundary stresses are (a)
concentrated over a small area near the
centroid of the cross-section, showing the
interior stresses at two different locations.

(b)

(c)
 6.4 Governing Equations in Matrix–Vector Form 163

than P/A near where the stresses are applied, as shown in Fig. 6.5b. However, this non-
uniformity of the stress distribution becomes smaller as we go further from the ends, and we
approach a uniform stress distribution having magnitude P/A as shown in Fig.6.5c when we
are only about a diameter away from the ends.
The "principle" of Saint-Venant is not a fundamental principle such as equilibrium, and it
can be shown to be violated in some cases. Also, this principle, as previously stated, is not
quantitative since it includes nonprecise terms such as “immediate neighborhood.”
Nevertheless, it can often be used successfully to relax how precisely we must specify the
boundary conditions in many problems.

6.4 Governing Equations in Matrix–Vector Form

We separately developed the equations of equilibrium for the stresses and the compatibility
equations for the strains, but there is a close connection between the differential operators
appearing in those equations that we want to highlight here. Similarly, there is a relationship
between the differential operators of the equilibrium equations and the operators appearing
in the strain–displacement relations. In this section we will use those relationships to write
the governing equations in equivalent matrix–vector forms. At first glance these may seem to
be simply formal mathematical relationships but bear with me, as their importance will
become evident later in Chapter 9 where we will see that these relationships between matrix
differential operators can be transformed into equivalent algebraic matrix relationships –
relationships that can be used to describe the solution of the governing equations and
boundary conditions solely in linear matrix–vector algebra terms. Such linear algebra
formulations are ideally suited to obtaining numerical computer solutions of complex
deformable body problems and complex structures such as trusses, frames, plates, shells,
etc. We will see a simple example of such a formulation in the next section.
Consider the first set of governing equations for 3-D problems that we considered (3-D
Governing Set 1). The equations of equilibrium can be rewritten in a matrix–vector form as
8 9
> σ xx >
>
> >
>
>
> >
2 3>> σ yy >
> 8 9
>
∂=∂x 0 0 0 ∂=∂z ∂=∂y > > >
>
>
< > <
= >fx = >
6 7 σ zz
4 0 ∂=∂y 0 ∂=∂z 0 ∂=∂x 5 þ fy ¼ 0 (6.43)
>
> σ yz >
> >
: > ;
∂=∂z ∂=∂y ∂=∂x >
> >
>
0 0 0 >
> >
> fz
>
> σ >
>
>
>
xz >
>
: ;
σ xy
or, more succinctly
½ℒE fσ g þ ff g ¼ 0 (6.44)

where ½ℒE  is the 3  6 matrix of derivatives appearing in Eq. (6.43), fσ gis the 6  1 row
vector of stresses, and ff g is the 3  1 body force vector. We have already placed the stress–
strain relations in a matrix–vector form as
164 Governing Equations and Boundary Conditions

8 9 2 38 9
>
> σ xx > C 11 C 12 C 13 C 14 C 15 C 16 > exx >
>σ >
> >
> 6 C 21
>
> eyy >
> >
>
>
> >
> C 22 C 23 C 24 C 25 C 26 7
7>> >
= 6 <e >
yy
< 6 C 31 7 =
σ zz C 32 C 33 C 34 C 35 C 36 7 zz
¼6
6 C 41 7 (6.45)
>
> σ yz >
> C 42 C 43 C 44 C 45 C 46 7>> γ >
>
> >
>
6 > yz >>
>
> σ xz >
>
4 C 51 C 52 C 53 C 54 C 55 C 56 5>>
> γ > >
: ; : xz >;
σ xy C 61 C 62 C 63 C 64 C 65 C 66 γxy

Writing the strains in terms of the stresses

8 9 2 38 9
>
> e xx >
> D11 D12 D13 D14 D15 D16 >> σ xx >
>
>
> eyy >> 6 D21 >
> >
>
>
> >
> D22 D23 D24 D25 D26 7>
> σ >
>
<e = 6 7< yy
6 D31 7 =
D32 D33 D34 D35 D36 7 σ zz
¼6
zz
6 D41 (6.46)
>
> γyz >> 6 D42 D43 D44 D45 D46 7> σ yz >
7> >
>
> >
> > >
>
> γ > 4 D51 D52 D53 D54 D55 D56 5>
> >
> σ xz >
: xz > ; : >
;
γxy D61 D62 D63 D64 D65 D66 σ xy

In matrix–vector form
fσ g ¼ ½C feg or feg ¼ ½Dfσ g (6.47)

in terms of the 6  6 elastic constants and compliance matrices discussed in Chapter 5 and
the 6  1 strain vector, feg, written in terms of the normal strains and the engineering shear
strains. The strain–displacement relationship in matrix–vector form becomes
8 9 2 3
>
> exx >> ∂=∂x 0 0
>e >
> > 6 8 9
>
> yy > 0 ∂=∂y 0 7
<e > = 6 6
7< ux =
7
0 0 ∂=∂z 7
¼6
zz
> γ > 6 0 ∂=∂z ∂=∂y 7: uy ; (6.48)
> yz >
> > 6 7 uz
>
> >
γ > 4 ∂=∂z 0 ∂=∂x 5
: xz >
> ;
γxy ∂=∂y ∂=∂x 0

However, if we compare Eq. (6.48) with the equilibrium equations, Eq. (6.43), we see that
the transpose of the equilibrium matrix of derivatives is just the matrix appearing here, i.e.,
we have

feg ¼ ½ℒE T fug (6.49)

Thus, the governing equations for the 3-D Set 1 become simply the following.

Matrix–Vector 3-D Governing Set 1

½ℒE fσ g þ ff g ¼ 0
fσ g ¼ ½C feg or feg ¼ ½Dfσ g (6.50)
feg ¼ ½ℒE T ½fug
 6.4 Governing Equations in Matrix–Vector Form 165

Note that, in this form, it is very easy to express the 3-D governing equations in terms of
the displacements since we can place the strain–displacement relationship into the stress–
strain equation and then place that result into the equilibrium equations to arrive at the
following.

Matrix–Vector 3-D Governing Set 2

½ℒE ½C ½ℒE T fug þ ff g ¼ 0 (6.51)

These are the equivalent to Navier’s equations for a general anisotropic media.
Now, consider the compatibility equations, which we can also write in a matrix–vector
form where
2 38 9
0 ∂2 =∂z2 ∂2 =∂y2 ∂2 =∂y∂z 0 0 >
>
exx >
>
6 7>> eyy >
> >
>
6 ∂2 =∂z2 0 ∂2 =∂x2 0 ∂2 =∂x∂z 0 7>> >
>
6 7>> >
>
6 ∂2 =∂y2 ∂2 =∂x2 ∂ 2
=∂x∂y 7 < e =
6 0 0 0 7 zz
6 7 ¼0
6 2 7> γ >
6 0 0 2∂ 2
=∂x∂y ∂ 2
=∂x∂z ∂ 2
=∂y∂z ∂ 2
=∂z 7>>
> yz >
>
6 2 7> γ > >
4 2∂ =∂y∂z 0 0 ∂2 =∂x2 ∂2 =∂x∂y ∂2 =∂x∂z 5> >
>
>
>
xz >
>
>
: ;
0 2∂ =∂x∂z
2
0 ∂ =∂x∂y
2
∂ =∂y
2 2
∂ =∂y∂z
2 γ xy
(6.52)

or, more succinctly

½ℒS feg ¼ 0 (6.53)

The 6  6 compatibility matrix is also related to the transpose of the 3  6 matrix of

equilibrium derivatives. By simple matrix multiplication it is easy (but rather lengthy) to
show that

½ℒS ½ℒE  T ¼ 0 (6.54)

where the “zero” is actually a 6  3 matrix of zeros in Eq. (6.54). Taking the transpose of Eq.
(6.54) it is also true that

½ℒE  ½ℒS T ¼ 0 (6.55)

where the “zero” is now a 3  6 matrix of zeros.

In the 3-D Governing Set 1, the compatibility equations did not appear since it was stated
that the strain–displacement relations implicitly satisfy compatibility. With our present
results it is easy to see this since, if we multiply the strain in the strain–displacement
relationship of Eq. (6.49) by the matrix compatibility operator, we have, using Eq. (6.54),

½ℒS feg ¼ ½ℒS ½ℒE T fug ¼ 0 (6.56)

166 Governing Equations and Boundary Conditions

We can form up a set of governing equations similar to 3-D Governing Set 3 involving only
the stresses by using the equilibrium equations and placing the stress–strain relations into
these compatibility equations to arrive at the following.

Matrix–Vector 3-D Governing Set 3

½ℒE fσ g þ ff g ¼ 0
(6.57)
½ℒS ½Dfσ g ¼ 0

But note that these are not the Beltrami–Michell compatibility equations since those
equations, as derived in Appendix B, use a set of six compatibility equations different from
Eq. (6.52) and also use the equilibrium equations to write the Beltrami–Michell equation
in its final form, as seen in Eq. (6.5). In writing the compatibility equations, there is always
some flexibility since we can use different sets of six equations formed from the original
81 equations (see Eq. (4.50)). However, regardless of the set we use, there are still only
three independent compatibility relations contained in the six equations. In 2-D problems,
the Beltrami–Michell equations are typically the equations of choice since, when combined
with the Airy stress potential, the biharmonic equation appears, which is a well-studied
equation.
We can form similar matrix–vector forms for 2-D plane-stress or plane-strain problems,
but we will not cover those details here.

6.5 Equivalent Algebraic Matrix–Vector Governing Equations

for Structures
The matrix–vector differential and algebraic governing equations and boundary conditions
associated with deformable bodies described above have equivalent purely algebraic coun-
terparts for structural problems that have long been studied. By making that equivalence
more explicit we can learn about two general ways to solve problems. A simple problem that
has been used for many years to illustrate a deformable structure is called Navier’s table
problem, as shown in Fig. 6.6a. A table with four legs (labeled 1, 2, 3, 4 in Fig. 6.6a) is a
statically indeterminate problem since, if we make a cut through the legs, we see that there
are four unknown internal forces ðF 1 , F 2 , F 3 , F 4 Þ and corresponding compressive stresses
ðσ 1 , σ 2 , σ 3 , σ 4 Þ acting (where F i ¼ σ i Aði ¼ 1; 2, 3, 4Þ, and A is the common cross-sectional
area of the legs) but only three equilibrium equations. Thus, Navier’s table problem must be
treated as a deformable body in order to be solved. We will assume we can neglect the weight
of the table top, which is assumed to be rigid, so that the only applied force is the load P,
 
which is applied eccentrically at distances ex , ey from the center C of the table, as shown in
Fig. 6.6a. The equilibrium equations are
 6.5 Equivalent Algebraic Matrix–Vector Governing Equations 167

z P
y P
ex ey
C x

4 3
F4 = s 4 A F3 = s 3 A
1 2 F1 = s 1 A F2 = s 2 A
2a
2b

(a) (b)

Figure 6.6 (a) An eccentrically loaded four-legged table (Navier’s table problem). (b) The forces and
corresponding stresses acting in the legs in equilibrium with the applied load, P.

X
Fz ¼ 0 F1 þ F2 þ F3 þ F4  P ¼ 0
X
M Cx ¼ 0 ðF 3 þ F 4 Þa  ðF 1 þ F 2 Þa  Pey ¼ 0 (6.58)
X
M Cy ¼ 0 ðF 1 þ F 4 Þb  ðF 2 þ F 3 Þb þ Pex ¼ 0

which can be written in matrix–vector form in terms of the stresses as

8 9
2 3> σ 1 > 8 P=A 9
1 1 1 1 >< > = > < >
=
4 a a a 5 σ2
a ¼ Pey =A (6.59)
> σ3 > > >
b b b b > : > ; : Pe =A ;
σ4 x

or, more symbolically,

 
~ fσ g ¼ fσ ∗ g
E (6.60)
 
~ is the 3  4 equilibrium matrix in Eq. (6.59) and
where E
8 9 8 9
>
> σ1 >
< > = < P=A >
> =
σ2
fσ g ¼ , fσ ∗ g ¼ Pey =A (6.61)
>
> σ > > >
: 3> ; :
Pex =A
;
σ4

where fσ g is a stress vector and fσ ∗ g can be considered to be a generalized applied “stress”

vector (note, however, that some terms in fσ ∗ g are not dimensionally stresses, but the
equilibrium matrix consistent with these “stresses” is also not dimensionally homogeneous).
Now, let’s consider the four strains in the legs which we call ðe1 , e2 , e3 , e4 Þ. Since the forces
ðF 1 , F 2 , F 3 , F 4 Þ were assumed to be compressive forces in the legs, the stresses also will be
compressive, so that to be consistent we will also take the strains to be compressive strains
168 Governing Equations and Boundary Conditions

Figure 6.7 Deformations caused z

by the rotations of the table top, (2,3)
which is assumed to be rigid. The b b x
  Ty
angles θx , θy are the rotations (1,4)
(a)
about the (x, y) axes, respectively.
(a) Looking down the negative
y-axis. (b) Looking down the
positive x-axis. z

(1,2) Tx y
a a (3,4)
(b)

corresponding to shortening of the legs. Since we assumed that the table top was rigid, the
top can only experience a translation in the plus or minus z-direction as well as rotations
about the x- and y-axes (see Fig. 6.7). If we let w be the translation of the table in the
 
negative z-direction and the angles θx , θy be the rotations about the (x, y) axes, respect-
ively, then the total compressive strains in the legs are
 
e1 ¼ w  bθy þ aθx =L
 
e2 ¼ w þ bθy þ aθx =L
  (6.62)
e3 ¼ w þ bθy  aθx =L
 
e4 ¼ w  bθy  aθx =L

which can be written in matrix–vector form as

8 9 2 3
> e1 > 1 a b 8 9
< >
> = 6 < w=L =
e2 1 a b 7
¼64
7 θ =L (6.63)
>
> e > 1 a b 5: x ;
: 3> ; θy =L
e4 1 a b

or, more symbolically,

 T
~ fqg
f eg ¼ E (6.64)

where
8 9
> e1 > 8 9
< >
> = < w=L =
e2
feg ¼ , fqg ¼ θx =L (6.65)
>
> e > : ;
: 2> ; θy =L
e4

Equation (6.64) is the relationship between the strain vector, feg, and a generalized
“displacement” vector, fqg. Again, we see the elements of this generalized displacement
 6.5 Equivalent Algebraic Matrix–Vector Governing Equations 169

vector are not dimensionally homogeneous. The transpose of the equilibrium matrix
appears in the generalized strain–displacement relationship of Eq. (6.64), which is consist-
ent with what we saw earlier in the governing differential equations. Now consider the
analogous stress–strain relations for this problem. The table legs are just members under
uniaxial loads so, from Hooke’s law, σ i ¼ Eei ði ¼ 1; 2, 3, 4Þ, where E is Young’s modulus.
In matrix–vector form we have
fσ g ¼ ½C feg (6.66)

where the generalized elastic constants matrix, ½C , is

2 3
E 0 0 0
60 E 0 0 7
½C  ¼ 6
40 0 E 0 5
7 (6.67)
0 0 0 E

In summary, we have the equilibrium, strain–displacement, and stress–strain equations for

the table to be
 
E~ fσ g ¼ fσ ∗ g
 T
feg ¼ E ~ fqg (6.68)
fσ g ¼ ½C feg

which are very similar in form to the 3-D Matrix–Vector Governing Set 1, except now the
 
equilibrium matrix operator ½ℒE  is replaced by algebraic equilibrium matrix E ~ . You
might wonder about the boundary conditions. They have already been implicitly satisfied as
part of the formulation of the governing equations in Eq. (6.68). There are displacement
boundary u1 ¼ u2 ¼ u3 ¼ u4 ¼ 0 conditions at the floor (see Fig. 6.8) but these have already
been satisfied by writing the strains in terms of the displacements of the table top only,
instead of the difference between the table top displacements and the floor displacements.
Similarly, the stress boundary conditions have also been accounted for by directly writing

z P Figure 6.8 Navier’s table problem, showing the

y displacements at the floor, all of which are assumed to
be zero.
ex ey
C x
u3
u4

u2
u1 2a
2b
170 Governing Equations and Boundary Conditions

the unknown stresses in terms of the applied stresses fσ ∗ g. Thus, Eq. (6.68) contains a
complete set of governing equations and boundary conditions for this problem.

6.5.1 Displacement Formulation

There are two ways to combine these governing equations to solve this problem. The first, a
displacement formulation, is to place the stress–strain and strain–displacement relations into
the equilibrium equations, yielding
   T
E~ ½C  E
~ fqg ¼ fσ ∗ g (6.69)

which is the purely algebraic equivalent of Navier’s equation for the table problem. We can
easily solve this problem numerically by placing explicit values into Eq. (6.69), but the
problem is simple enough that we can also obtain the solution symbolically. MATLAB® can
help us to do the necessary symbolic manipulations.
First, we define as symbolic the variables P, A, E, ex, ey, a, and b, which are the applied
 
force, the cross-sectional area, Young’s modulus, the eccentricities of the load ex , ey , and
 
the dimensions ða; bÞ. We then form up the equilibrium matrix, Em ¼ E ~ , a known vector
∗
of stresses, pv ¼ fσ g, and the elastic constants matrix Cm ¼ ½C :
syms P A E ex ey a b
Em = [1 1 1 1; -a -a a a; b -b -b b];
pv = [P/A ; P*ey/A ; -P*ex/A];
Cm = [E 0 0 0;0 E 0 0;0 0 E 0;0 0 0 E];
   T
~ C
Next, we form the matrix of coefficients in the displacement formulation, M ¼ E ~ E~ ,
solve the system of linear equations, and simplify the final expression:
M = Em*Cm*Em.' ;
q = inv(M)*pv ; % we can also write this as q = M\pv;
q = simplify(q)

Note the use of the MATLAB expression .’ for the pure matrix transpose. The generalized
displacement vector is
q = P/(4*A*E)
(P*ey)/(4*A*E*a^2)
-(P*ex)/(4*A*E*b^2)

which we can multiply by L and rewrite as

PL
w¼
4AE
PLey
θx ¼ (6.70)
4AEa2
Plex
θy ¼ 
4AEb2
 6.5 Equivalent Algebraic Matrix–Vector Governing Equations 171

If ex ¼ ey ¼ 0 then w ¼ PL=4AE and θx ¼ θy ¼ 0, so the table top does not tilt, as expected.
To obtain the vector of compressive stresses in the legs sv ¼ fσ g from these displacements,
we must write the stress–displacement relations:
 T
fσ g ¼ ½C  E~ fqg (6.71)

which in MATLAB® is
sv = Cm*Em.'*q
sv = P/(4*A) - (P*ey)/(4*A*a) - (P*ex)/(4*A*b)
P/(4*A) - (P*ey)/(4*A*a) + (P*ex)/(4*A*b)
P/(4*A) + (P*ey)/(4*A*a) + (P*ex)/(4*A*b)
P/(4*A) + (P*ey)/(4*A*a) - (P*ex)/(4*A*b)

This result can be rewritten as

P  
σ1 ¼ 1  ry  rx
4A
P  
σ2 ¼ 1  ry þ rx
4A
(6.72)
P  
σ3 ¼ 1 þ ry þ rx
4A
P  
σ1 ¼ 1 þ ry  rx
4A
where rx ¼ ex =b, ry ¼ ey =a. Equivalently, multiplying by A we obtain the forces in the legs.
If we have ex ¼ ey ¼ 0 so the load is applied at the center of the table, then we have simply
σ 1 ¼ σ 2 ¼ σ 3 ¼ σ 4 ¼ P=4A.
We can summarize the displacement formulation as a five-step process.
(1) Form up the equations of equilibrium
 
~ fσ g ¼ fσ ∗ g
E

(2) Write the strain–displacement relationship

 T
feg ¼ E ~ fqg

(3) Use the stress–strain relations to write the stress–displacement relationship

 T
fσ g ¼ ½C feg ! fσ g ¼ ½C  E~ fqg

(4) Place the stress–displacement relations into equilibrium and solve for the displacements
   T
~ ½C  E
E ~ fqg ¼ fσ ∗ g ! fqg

(5) Calculate the stresses from the stress–displacement relations

 T
~ fqg
fσ g ¼ ½C  E
172 Governing Equations and Boundary Conditions

6.5.2 Stress (Force) Formulation

A displacement formulation is a commonly used approach to solve structural problems.
However, we can also formulate the table problem in terms of only the stresses (or,
equivalently, the forces) by combining the equilibrium equations with the compatibility
equations analogous to the approach used in the Matrix–Vector Governing Set 3. We have
already seen that the compatibility equation can be obtained from the strain–displacement
relations by eliminating the displacements to obtain relationships between the strains
themselves. Since we have algebraic relations here, that elimination is easy to perform.
From Eq. (6.63) we see by subtracting the first two relations and then adding the difference
of the last two relations, we find
e1  e2 þ e3  e4 ¼ 0 (6.73)

which we can write in matrix–vector form as

8 9
>
> e1 >
< > =
e
½ 1 1 1 1  2 ¼ 0 (6.74)
> e3 >
> >
: ;
e4

or, more succinctly,

 
~ feg ¼ 0
S (6.75)
 
where here S ~ is a 1  4 compatibility matrix. We can verify that the compatibility and the
equilibrium matrices here satisfy the same relationship we found for their matrix operators,
  T   T
namely S ~ E~ ¼ E ~ S~ ¼ 0, by writing the latter of these two relationships, which
yields
8 9
2 3> 1 > 8 9
1 1 1 1 <> >
= <0=
4 a a a 5 1
a ¼ 0 (6.76)
> 1> : ;
b b b b >: >
; 0
1

To combine the compatibility equations with the equilibrium equations we must write the
compatibility equations in terms of the stresses, using the stress–strain relations written as
feg ¼ ½Dfσ g in terms of the compliance matrix, ½D, where here
2 3
1=E 0 0 0
6 0 1=E 0 0 7
½D ¼ 6
4 0
7 (6.77)
0 1=E 0 5
0 0 0 1=E

But in this case the compatibility equations just become

 
~ ½Dfσ g ¼ ½ 1=E 1=E 1=E 1=E fσ g ¼ 0
S (6.78)
 6.5 Equivalent Algebraic Matrix–Vector Governing Equations 173

and the common factor of 1/E could be eliminated if we so desire. If we combine this
compatibility equation with the equilibrium equations, we have
2 38 9 8 9
1 1 1 1 >
> σ1 >
> >
> P=A >>
6 >
7> > > > > >
>
6 a a a a 7< σ 2 = < Pey =A =
6 7 ¼ (6.79)
6 b b 7
4 b b 5>>
> σ3 >
>
>
>
>
> Pex =A >
>
>
: >
> ; > : >
;
1=E 1=E 1=E 1=E σ4 0

Using the same symbolic variables described previously, we can solve these equations in
MATLAB® as:
Hm = [ 1 1 1 1; -a -a a a; b -b -b b; 1/E -1/E 1/E -1/E];
% equilibrium matrix
sv2 = [P/A; P*ey/A; -P*ex/A; 0]; % right hand side
% of equilibrium equations
st2 = inv(Hm)*sv2; % solve for stress vector st2.
% Can also use st2 = Hm\sv2;
simplify(st2)
ans = -(P*(a*ex - a*b + b*ey))/(4*A*a*b)
(P*(a*b + a*ex - b*ey))/(4*A*a*b)
(P*(a*b + a*ex + b*ey))/(4*A*a*b)
(P*(a*b - a*ex + b*ey))/(4*A*a*b)

which gives the same stresses again organized in a slightly different form. Note that
Young’s modulus is missing from these results, as we would expect, since it could be
eliminated before solving. Since the stresses are known we can calculate the strains
from the stress–strain relations and then use the strain–displacement relation to back-
calculate the generalized displacements, if necessary. The stress formulation is nice in
that, once the compatibility equations are known, we can solve for the stresses directly,
which are often the quantities of most interest. We could summarize the stress formula-
tion as a six-step process.

(1) Form up the equilibrium equations

 
~ fσ g ¼ fσ ∗ g
E

(2) Write the strain–displacement relationship

 T
feg ¼ E ~ fqg

(3) Eliminate the generalized displacements to obtain the compatibility equations

 T  
feg ¼ E ~ fqg ! S ~ feg ¼ 0
174 Governing Equations and Boundary Conditions

(4) Use the stress–strain relations to write the compatibility equations in terms of the
stresses
 
feg ¼ ½Dfσ g ! S~ ½Dfσ g ¼ 0

(5) Append these compatibility equations to the equilibrium equations and solve for the
stresses
2 3 8 ∗9
~
E <σ =
4  5fσ g ¼  ! fσ g
  : ;
~ ½D
S 0

(6) Back-calculate the displacements, if needed, from the stress–displacement relations. This
process will be described in more detail in Chapter 9, but here we can simply describe the
process symbolically as:
 T
~ fqg ! fqg
fσ g ¼ ½C  E

6.6 Structural Analysis – A Brief Preview

In our discussion of Navier’s table problem, we used the stresses and strains to describe the
problem so that we could make a close connection to the governing equations covered in the
first sections of this chapter. Note that we did have to use generalized displacements, fqg,
since the problem was best described in terms of both displacements and rotations. Similarly,
engineers in structural problems often prefer to use generalized forces, fF g (internal forces
or moments or torques), instead of stresses. Generalized strains (deformations), fΔg, are also
often used instead of the ordinary strains. For example, in Navier’s table problem we could
have replaced the strains in the strain-generalized displacement relations by the shortening
of the legs Δi ¼ ei Lð1 ¼ 1; 2, 3, 4Þ instead of the strains themselves. The basic building blocks
for a structural analysis, however, remain the same: the equilibrium equations for the
generalized forces, the deformation (generalized strain)–generalized displacement relations
and the generalized force–deformation relations, i.e.,
 
~ fF g ¼ fP∗ g
E
 T
fΔg ¼ E ~ fqg (6.80)

fΔg ¼ ½G fF g or fF g ¼ ½G 1 fΔg

where fP∗ g are known generalized forces applied to the structure and ½G  is a flexibility matrix
for the structure which plays the same role for the generalized forces and generalized strains
that the compliance matrix does for the stresses and strains. These equations can be used to
form and solve for the generalized displacements directly as done in the table problem
  1  T
~ ½G  E
E ~ fqg ¼ fP∗ g (6.81)
 6.6 Structural Analysis – A Brief Preview 175

or, for the generalized forces, if the compatibility equations

 
~ f Δg ¼ 0
S (6.82)

are obtained by eliminating the generalized displacements in the deformation-generalized

displacements and then combining these compatibility equations, written in terms of the
generalized forces, with the equilibrium equations to solve for the generalized forces:
2 3 8 ∗9
~
E <P =
4  5fF g ¼  (6.83)
  : ;
~ ½G 
S 0
Historically, however, the displacement approach has not been implemented with
Eq. (6.81) but with a closely related formulation based on the “stiffness” method of finite
elements, as discussed briefly in Section 6.1. In terms of our present discussion, in a stiffness
finite element approach the generalized forces would be expressed in terms of generalized
displacements as
fF g ¼ ½K fqg (6.84)

where ½K  is a stiffness matrix for the structure. Typically, work–energy methods (see
Chapter 8 and Chapter 9 for more details) are used to relate the generalized displacements
to known generalized applied forces:
½K fqg ¼ fP∗ g (6.85)
In this approach the stiffness matrix is developed directly and is not formed from a
combination of equilibrium and flexibility matrices although we see we could consider the
matrix appearing in Eq. (6.81) as a pseudo stiffness matrix, K pseudo , where
    1  T
K pseudo ¼ E~ ½G  E~ (6.86)

In some problems the stiffness matrix and the pseudo stiffness matrix are identical but in
general they are different. The stiffness-based finite element method, as mentioned earlier,
has become the de facto method of choice for solving deformable body problems. In this
method a deformable structure is broken into small elements (which is where the term “finite
element” comes from) and the stiffness matrices are defined for those elements. The stiffness
matrix for the entire structure can then be assembled from those elemental matrices, and
Eq. (6.85) solved for the unknown generalized displacements once the known loads are
expressed as known generalized forces, fP∗ g. Generalized forces or stresses can then be
calculated from the displacements. In contrast, when problems are solved with Eq. (6.83),
where the generalized forces are the fundamental unknowns, this is called a force-based
method. In this force-based method we can also assemble the equilibrium equations for an
entire structure from the equilibrium equations for small elements, so that it also can be
treated as a force-based finite element method.
There are, of course, many details involved in implementing stiffness-based or force-based
finite elements that we have not discussed. We will get into the weeds with some of those
176 Governing Equations and Boundary Conditions

details later in the book in Chapter 9. In this chapter we have simply wanted to lay out the
framework of governing differential equations for solving stress-analysis problems using
either displacement-based or stress-based “elasticity” approaches, and the corresponding
governing algebraic equations for structural problems using generalized displacement-based
or generalized force-based methods. In the next chapter, we will obtain some elasticity
solutions based on both displacement or stress-based methods.

6.7 PROBLEMS

P6.1 The following state of stress

2 3
Ay þ Bz Cz Cy
½σ  ¼ 4 Cz 0 0 5
Cy 0 0
is supposed to exist in a cylindrical bar of length L and radius r, as shown in Fig. P6.1,
where the bar is loaded only at its ends. Here, A, B, and C are constants and the
cylinder is made from an isotropic elastic material. Show that this can be a legitimate
state of stress by
(a) showing that it satisfies the equations of equilibrium everywhere in the cylinder,
(b) showing that it satisfies stress-free boundary conditions on the surface of the
cylinder,
(c) showing that it produces strains that are compatible.
(d) What other conditions would have to be met before we could say that these stresses
were an exact solution to a particular problem?
(e) If the cylinder is relatively long (L  r) use the idea of Saint-Venant and assume it
is only important that these stresses generate the appropriate end forces and
moments. What must those end forces and moments be in terms of A, B, C, and r?
P6.2 Consider three elastic rods that are all attached to a rigid floor and connected to a
rigid bar, which is initially horizontal before being eccentrically loaded with a

Figure P6.1 A cylindrical z

geometry.

r y

x
 6.7 Problems 177

d P Fig. P6.2 Three elastic rods connected to an eccentrically loaded rigid bar.
rigid
A C
force, P (see Fig. P6.2). Neglect the weight of all the members. The
L
cross-sectional area of each rod is A, and they all have the same
E1 E2 E3
length, L, but different Young’s moduli ðE 1 , E 2 , E 3 Þ, as shown. This
is a statically indeterminate problem that we can solve for the
tensile forces ðF 1 , F 2 , F 3 Þ in the rods by combining equilibrium,
compatibility, and the force–deformation (generalized stress–
a a
strain) relations.
(a) Summing forces vertically and taking moments about the center point C on the
rigid bar, obtain the equilibrium equations for this problem in the form
 
E~ fF g ¼ fP∗ g

(b) If we let ðΔ1 , Δ2 , Δ3 Þ be the generalized strains (elongations) of the rods, the
generalized strain–displacement relations can be written as
 T
fΔg ¼ E ~ fqg

where fqg contains two unknown generalized displacements ðq1 , q2 Þ. From these
strain–displacement relations, determine the physical meaning of these generalized
displacements.
(c) By eliminating the displacements from the strain–displacement relations obtain the
    T
compatibility equation S ~ fΔg ¼ 0. Verify that S~ E ~ ¼ 0.
(d) Relate the generalized strains to the forces through the relations fΔg ¼ ½G fF g,
where here the flexibility matrix ½G  contains the flexibilities of the rods
gi ¼ L=AE i ði ¼ 1; 2, 3Þ along its diagonal. Placing these relations into the compati-
bility equation, and then appending the resulting equation to the equilibrium
equations, solve for the forces ðF 1 , F 2 , F 3 Þ in the bars in terms of ðgi ; P; a; d Þ.
(e) Repeat parts (a)–(d) if our equilibrium equations are instead obtained by summing
forces vertically and taking moments about point A on the rigid bar. Verify that you
obtain the same solution for the forces and show how the point about which the moment
is taken is related to the displacements present in the strain–displacement relations.
P6.3 Consider the torsion of a circular rod that composed of two sections, AB and BC,
that are welded together at B and are fixed between two rigid supports at A and
C (see Fig. P6.3a). The shear moduli, polar area moments, and lengths of sections
AB and BC are ðμi , J i , Li Þði ¼ 1; 2Þ, respectively, as shown in Fig. P6.3a. [Note: we
have used μ instead of G for the shear modulus since G has been used to denote the
flexibility matrix.] A torque, T, acts at the junction B between AB and BC and there
are two unknown end reaction torques ðT A , T C Þ as shown in Fig. P6.3b. The twist at
the applied torque is θB . This is a statically indeterminate problem that we can solve
for the internal torques ðM 1 , M 2 Þ acting in sections AB and BC as shown in
Fig. P6.3c by combining equilibrium, compatibility and the torque–twist relations.
178 Governing Equations and Boundary Conditions

Figure P6.3 (a) Two bars m2 , J 2 T

welded together at B and fixed m1 , J1 TA
at sections A and C. (b) The T
two bar assembly showing the
applied torque T, the end
reaction torques ðT A , T C Þ A B C TC
and the angle of twist, θB , at qB
L1 L2
the applied torque. (c) The
internal moments ðM 1 , M 2 Þ (a) (b)
in the bars (assumed to act in TA M1
the directions shown).

M2
TC
(c)

(a) Write down the equilibrium equation relating ðM 1 , M 2 Þto the applied torque, T in
the form
 
E~ fM g ¼ fT ∗ g

(b) If we let ðΔ1 , Δ2 Þ be the generalized strains (which in this case are taken as the
twists) of sections AB and BC the generalized strain–displacement relation can be
written as
 T
fΔg ¼ E ~ fqg

From this strain–displacement relation determine the physical meaning of the

generalized displacement fqg.
(c) By eliminating the generalized displacement from the strain–displacement relation
    T
~ fΔg ¼ 0. Verify that S
obtain the compatibility equation in the form S ~ E~ ¼ 0.
(d) Relate the generalized strains to the internal torques through the relations
fΔg ¼ ½G fM g, where the flexibility matrix ½G  contains the torsional flexibilities
of the sections AB and BC gi ¼ Li =μi J i ði ¼ 1; 2Þ along its diagonal. Placing these

A E

B
V0 C D
Figure P6.4 A thin plate with a central hole under V0 x
uniaxial tension.
 References 179

relations into the compatibility equation, and then appending the resulting equation
to the equilibrium equations, solve for the internal torques ðM 1 , M 2 Þ in terms
of ðgi ; T Þ.
P6.4 Consider a thin plate with a central hole and loaded by a uniform normal stress on its
ends, as shown in Fig. P6.4. If we analyze only the first quarter of this plate, region
ABCDE, what are the boundary conditions we must satisfy on the boundaries of this
region? Assume plane-stress conditions.
P6.5 Show that on a stress-free surface, the boundary conditions can be expressed in terms
of the Airy stress function as the conditions, ϕ ¼ 0 and ∂ϕ=∂n ¼ 0, where n is in a
direction normal to the surface.

References
1. J. N. Reddy, Introduction to the Finite Element Method, 4th edn (New York, NY: McGraw-Hill,
2018)
2. M. Bonnet, Boundary Integral Equation Methods for Solids and Fluids (New York, NY: John
Wiley, 1999)
3. A. C. Ugural and S. K. Fenster, Advanced Mechanics of Materials and Applied Elasticity, 5th edn
(Boston, MA: Prentice Hall, 2012)
4. C. E. Pearson, Theoretical Elasticity (Cambridge, MA: Harvard University Press, 1959)
7 Analytical Solutions

Most problems involving complex-shaped deformable bodies require a numerical solution of

the governing equations and boundary conditions. However, there are a number of import-
ant simple problems that can be solved analytically and that reveal the nature of the stresses
and deformations in geometries of practical significance. In this chapter we will examine a
number of such problems, using some of the various sets of governing equations discussed in
the previous chapter.

7.1 Displacement-Based Solutions

Reducing the governing equations to Navier’s equations is one of the important choices
outlined in the previous chapter. In this section we will examine cylindrical geometries under
radial pressure loads in configurations where explicit solutions to Navier’s equations can
be obtained.

7.1.1 Axisymmetric Solutions in Cylindrical Geometries

One case where we can solve the governing equations analytically is the case of axisymmetric
solutions in cylindrical coordinates where there is no dependence of the solution on the angle
θ and where we can assume that the stresses developed satisfy
σ rr ¼ σ rr ðrÞ, σ θθ ¼ σ θθ ðrÞ
(7.1)
σ rθ ¼ σ rz ¼ σ zθ ¼ σ zz ¼ 0

In this case the equilibrium equations which in cylindrical coordinates with no body forces are
∂σ rr σ rr  σ θθ 1 ∂σ rθ ∂σ rz
þ þ þ ¼0
∂r r r ∂θ ∂z
∂σ rθ 2σ rθ 1 ∂σ θθ ∂σ θz
þ þ þ ¼0 (7.2)
∂r r r ∂θ ∂z
∂σ zr σ zr 1 ∂σ zθ ∂σ zz
þ þ þ ¼0
∂r r r ∂θ ∂z
reduce to only one equation
∂σ rr σ rr  σ θθ
þ ¼0 (7.3)
∂r r

180
 7.1 Displacement-Based Solutions 181

The stress state given by Eq. (7.1) is one of plane stress so that the stress–strain relations are
the plane-stress equations
E
σ rr ¼ ðεrr þ vεθθ Þ
1  v2
E
σ θθ ¼ ðεθθ þ vεrr Þ (7.4)
1  v2
v
εzz ¼ ðσ rr þ σ θθ Þ
E
To guarantee that we automatically satisfy compatibility, we will work directly with the
displacements as the fundamental unknowns for this problem. Because of the assumed radial
symmetry, we expect that uθ ¼ 0, i.e., the θ-displacement is zero. Since the stresses are
uniform in the z-direction, we also expect that the radial displacement will be uniform so
that ur ¼ ur ðrÞ. The strains in cylindrical coordinates are
∂ur 1 ∂uθ ur ∂uz
εrr ¼ , εθθ ¼ þ , εzz ¼
∂r r ∂θ r ∂z


1 1 ∂uz ∂uθ uθ
εrθ ¼ þ 
2 r ∂θ ∂r r

 (7.5)
1 ∂uz ∂ur
εrz ¼ þ
2 ∂r ∂z


1 1 ∂uz ∂uθ
εθz ¼ þ
2 r ∂θ ∂z

which, under our assumptions, reduce to

dur ur ∂uz
εrr ¼ , εθθ ¼ , εzz ¼
dr r ∂z (7.6)
εrθ ¼ εθz ¼ εrz ¼ 0

Placing these strain–displacement relations into the stress–strain relations of Eq. (7.4) gives


E dur ur
σ rr ¼ þ v
1  v2 dr r


E ur dur
σ θθ ¼ þ v (7.7)
1  v2 r dr
∂uz v
εzz ¼ ¼ ðσ rr þ σ θθ Þ
∂z E
so that when the stresses of Eq. (7.7) are placed into the equilibrium equation, Eq. (7.3), we
find

d 2 ur 1 dur ur
þ  ¼0 (7.8)
dr2 r dr r2
182 Analytical Solutions

This is Navier’s equation for this radially symmetric problem. You may wonder why we did
not obtain this equation directly from the plane-stress Navier’s equations outlined in the
previous chapter. The reason is that, in this problem, some of our assumptions are based on
stresses and some are based on displacements so that it is more straightforward to obtain the
stresses as a function of the reduced displacements, as done in Eq. (7.7), and then use those
equations in the equilibrium equations to deduce Navier’s equations. It is easy to solve
Eq. (7.8) for the displacement since we can write it equivalently in the form
 
d 1d
ðrur Þ ¼ 0 (7.9)
dr r dr

Integrating once on r gives

d
ðrur Þ ¼ C 01 r (7.10)
dr
where C 01 is a constant of integration. Integrating Eq. (7.10) once more then yields
C 01 C2 C2
ur ¼ rþ ¼ C1r þ (7.11)
2 r r
where C 1 ¼ C 01 =2, C 2 are both constants of integration. If Eq. (7.11) is placed back into the
stress–displacement relations, Eq. (7.7), the stresses and axial strain are
E E C2
σ rr ¼ C1 
1v 1 þ v r2
E E C2
σ θθ ¼ C1 þ (7.12)
1v 1 þ v r2
2v
εzz ¼ C 1 ¼ constant
1v
If we define two new constants (A, B) as
E E
A¼ C1, B¼ C2 (7.13)
1v 1þv
then we can rewrite the stresses and axial strain simply as
B
σ rr ¼ A 
r2
B
σ θθ ¼ A þ 2 (7.14)
r
∂uz 2vA
εzz ¼ ¼ ¼ constant
∂z E
Equation (7.11) and Eq. (7.14) yield the displacements and stresses for our axisymmetric
problem. They satisfy exactly the governing equations of equilibrium, compatibility, and the
stress–strain relations. If we can also exactly satisfy the boundary conditions, then we will
have an exact “elasticity” solution.
 7.1 Displacement-Based Solutions 183

7.1.2 Thick-Wall Pressure Vessel

Consider the problem shown in Fig. 7.1, where a cylindrical vessel is loaded by uniform
pressures ðpi , pe Þ on the inside and outside of the cylinder, respectively. We will also assume
the ends of the cylinder are stress free. A real pressure vessel will have its ends closed in some
fashion, but we will deal with that case shortly. For the case shown in Fig. 7.1, the boundary
conditions are: (1) for 0  θ  2π, 0  z  L on the inner and outer surfaces we have
σ rr ðr; θ; zÞjr¼ri ¼ pi
σ rθ ðr; θ; zÞjr¼ri ¼ σ rz ðr; θ; zÞjr¼ri ¼ 0
(7.15)
σ rr ðr; θ; zÞjr¼re ¼ pe
σ rθ ðr; θ; zÞjr¼re ¼ σ rz ðr; θ; zÞjr¼re ¼ 0

and (2) for 0  θ  2π, ri  r  re on the cylinder ends

σ zz ðr; θ; zÞjz¼0 ¼ σ zr ðr; θ; zÞjz¼0 ¼ σ zθ ðr; θ; zÞjz¼0 ¼ 0
(7.16)
σ zz ðr; θ; zÞjz¼L ¼ σ zr ðr; θ; zÞjz¼L ¼ σ zθ ðr; θ; zÞjz¼L ¼ 0

But from our original assumptions on the stresses, Eq. (7.1), all of these boundary conditions
are satisfied identically except the two conditions
σ rr ðrÞjr¼ri ¼ pi
r (7.17)
σ rr ðrÞjr¼re ¼ pe
V rr
pe
V rz V zr which, from Eq. (7.14), gives
V zz D A C V zz B
B A ¼ pi
V zr V rz r2i
pi (7.18)
V rr z B
pi A  2 ¼ pe
re

Solving for (A, B) then we have

pe
r2i pi  r2e pe r2 r2 ðp  p Þ
L A¼ , B ¼ i e2 i 2 e (7.19)
re  ri
2 2 re  ri
(a)
pe V rT V rr These constants then yield complete solutions for the
V zr stresses and displacement of the cylinder:
A
V zT C
V rr
B Figure 7.1 A thick-wall cylindrical pressure vessel subjected to
pi V rT
ri internal and external pressures along its length (only parts of the
pressure distributions are shown) showing the stresses acting (a)
re in a longitudinal plane, and (b) in a cross-sectional plane.
Generic points where these stresses act are shown as points (A,
(b) B, C, D).
184 Analytical Solutions



pi r2i r2e pe r2e r2i
σ rr ¼ 2 1 2  2 1 2
re  r2i r re  r2i r



pi r2i r2e pe r2e r2i
σ θθ ¼ 2 1þ 2  2 1 2
re  r2i r re  r2i r

2 
 (7.20)
duz 2v ri pi  r2e pe 2v r2i pi  r2e pe
εzz ¼ ¼ ! u z ¼ zþC
dz E r2e  r2i E r2e  r2i


ð1  vÞ r2i pi  r2e pe ð1 þ vÞ r2e r2i ðpi  pe Þ 1
ur ¼ r þ
E r2e  ri 2 E r2e  r2i r

where we have integrated the strain, ezz , once on z to obtain the z-displacement in terms of a
constant of integration, C.
The solution of Eq. (7.20) is an exact solution to our pressurized cylinder problem since it
satisfies all the governing equations and boundary conditions. (Note that the homework
problems P3.1 and P3.2 in Chapter 3 considered only satisfaction of the equilibrium
equations and boundary conditions.) One can show that for a pressurized cylinder subject
to an internal pressure only, the highest stresses occur on the inner wall where we have


pi r2i r2e
σ rr jr¼ri ¼ 2 1  2 ¼ pi
re  r2i ri

 (7.21)
pi r2i r2e
σ θθ jr¼ri ¼ 2 1þ 2
re  r2i ri

For a thin-wall cylinder these stresses become

σ rr jr¼ri ¼ pi




pi r2i r2e pi r2e þ r2i pi re þ ri t2
σ θθ jr¼ri ¼ 2 1þ 2 ¼ ¼ þ
re  r2i ri t re þ ri t 2 2ðre þ ri Þ (7.22)

p re þ ri  p rm
 i ¼ i
t 2 t
in terms of the thickness t ¼ re  ri and the mean radius rm ¼ ðre þ ri Þ=2. This agrees with
the result for the thin-wall cylindrical pressure vessel hoop stress usually obtained in strength
of materials courses (where the small radial stress component is normally ignored).
Now, let’s come back to the question of the boundary conditions at the ends, since the
“open ends” conditions that our exact solution satisfies are not those usually found in
practice. Normally the ends of a pressure vessel are sealed, as shown in Fig. 7.2, by placing
end caps on the cylinder. Figure 7.2a shows where a flat plate is welded to the cylinder, while
Fig. 7.2b shows spherical end caps. In either case the end caps have a different radial stiffness
than the cylinder so that the tendency of the cylinder to expand is not identical to the end
caps. This will cause shear stresses and bending stresses to develop in the region near where
they are connected. However, the pressure acting on the end caps also will cause the cylinder
 7.1 Displacement-Based Solutions 185

to be stretched in the axial z-direction throughout its length under an axial stress and
corresponding axial load. Using the principle of Saint-Venant, we might expect that if we
are not too close to the end connection (dashed lines in Fig. 7.2), the cylinder will experience
a uniform axial stress, σ zz , across its thickness. Thus, consider the case of a flat end cap, as
shown in Fig 7.3. Summing forces in the axial direction gives

σ zz π r2e  r2i ¼ pi πr2i  pe πr2i (7.23)

which gives
pi r2i  pe r2e
σ zz ¼ (7.24)
r2e  r2i

This can also be shown to be the axial stress for other types of end caps. From generalized
Hooke’s law
ur σ θθ vðσ rr  σ zz Þ
¼ eθθ ¼  (7.25)
r E E
But the solution for the open-ends case satisfies plane-stress conditions where
σ θθ vσ rr
eθθ ¼  (7.26)
E E
so we see that adding end caps produces an additional
axial stress term, which in turn affects the radial displace-
ment in the cylinder by adding the term
vσ zz r
ður Þσzz ¼ (7.27)
E
(a)
with σ zz given by Eq. (7.24). The stresses σ rr , σ θθ , are
unaffected. The axial strain, ezz , is also changed so the
total axial strain is now, by superposition with the previous
strains,

Figure 7.2 A cylindrical pressure vessel closed at its ends by (a) a

(b) flat plate, and (b) a half-sphere end cap.

V zz
ri re
pe
pi

V zz Figure 7.3 Stress and pressures acting in the axial direction of a

closed cylinder.
186 Analytical Solutions

ezz ¼ ðezz Þσzz þ ðezz Þσττ , σθθ


σ zz 2v pi r2i  pe r2e
¼ 
E E r2e  r2i (7.28)



1  2v pi r2i  pe r2e
¼
E r2e  r2i

In summary, the stresses and displacements for the closed pressure vessel are



pi r2i r2e pe r2e r2i
σ rr ¼ 2 1 2  2 1 2
re  r2i r re  r2i r



pi r2i r2e pe r2e r2i
σ θθ ¼ 2 1þ 2  2 1þ 2
re  r2i r re  r2i r

pi r2i  pe r2e
σ zz ¼ (7.29)
r2e  r2i


ð1  2vÞ r2i pi  r2e pe ð1 þ vÞ r2e r2i ðpi  pe Þ 1
ur ¼ r þ
E r2e  ri
2 E r2e  r2i r



1  2v r2i pi  r2e pe
uz ¼ zþC
E r2e  r2i

Note that these are no longer exact solutions for the stresses in the cylinder since we have
used the Saint-Venant principle to arrive at these results. Stresses and deformations in the
end caps are obviously much more difficult to describe.
We should make a few remarks about the assumptions we made originally in this problem
which led to the solution for the open cylinder being a plane-stress solution. Plane stress is
normally thought to apply to a very thin body, not a long cylinder under internal loads.
However, we showed that our solution was in fact an exact solution to the governing
equations and boundary conditions regardless of the length of the cylinder. Physically, in
the open cylinder case the cylinder is free to expand uniformly in the z-direction regardless of
how long the cylinder is and the ends are stress-free so a σ zz is not needed anywhere in the
cylinder. Thus, there is no contradiction with having a state of plane stress for the long,
open cylinder.

Example 7.1 Thick-Wall Pressure Vessel

A closed cylindrical vessel has an inner radius ri ¼ 10 mm and an outer radius re ¼ 20 mm
and is subjected to an inside pressure pi only. The properties of the vessel are E ¼ 210 GPa,
v ¼ 0:29. A strain gage placed on the outside of the vessel measures a circumferential strain
eθθ ¼ 400 μ when the inside pressure is applied. What is the internal pressure pi ?
 7.1 Displacement-Based Solutions 187

The stresses at outside of the vessel are

σ rr ¼ 0
2pi r2i ð2Þð100Þ
σ θθ ¼ ¼ p ¼ 0:667pi
re  ri
2 2 400  100 i
pi r2i 100
σ zz ¼ ¼ p ¼ 0:333pi
r2e  r2i 400  100 i

The measured hoop strain is

1
eθθ ¼ ðσ θθ  vσ rr  vσ zz Þ
E
1
! 400  106 ¼ pi ½0:667  ð0:29Þð0:333Þ
210  103
so that solving for pi we find pi ¼ 147 MPa.

7.1.3 Shrink-Fits
Another problem where we can use the axisymmetric solutions of a cylinder with free ends
subjected to inner or outer pressures is where two cylinders are pressed together by a
“shrink-fit” process where, say, the outer cylinder is heated and allowed to expand, and
then fitted over an inner cylinder (having a slightly larger outer radius than that of the inner
radius of the outer cylinder at room temperature) and allowed to cool, creating a final state
where there is a large residual pressure between the two cylinders. This final state of loading
is shown in Fig. 7.4, where the outer cylinder is initially assumed to have an outer radius, re ,
and an inner radius, ri , and the inner cylinder initially has an outer radius, ri þ Δ, and an
inner radius, ru , all as measured at room temperature. The quantity Δ is called the

outer cylinder Figure 7.4 A shrink-fit problem where an outer cylinder,

shown in (a), has a slightly smaller inner radius, ri , than the
inner cylinder outer radius of the inner cylinder, ri þ Δ, shown in (b), and
where Δ is the interference and where p is the interference
p
ri pressure when they are pressed together in the shrink-
p ru fit process.

re ri + D

(a) (b)
188 Analytical Solutions

interference. The mutual pressure between the two cylinders, p, is called the interference
pressure. For either cylinder the stresses and the radial displacements are given by the forms
derived previously, namely the stresses from Eq. (7.14) and the radial displacement from
Eq. (7.11), using the relationship between the constants in Eq. (7.13):
B
σ rr ¼ A 
r2
B
σ θθ ¼ A þ 2 (7.30)
r
 
1 B
ur ¼ ð1 þ vÞ þ ð1  vÞAr
E r

We will let the outer cylinder have a Young’s modulus and Poisson’s ratio of ðE 1 , v1 Þ,
respectively, and the inner cylinder will have the corresponding properties ðE 2 , v2 Þ. Then the
stresses in the outer cylinder will be given as
B
σ rr ¼ A 
r2
(7.31)
B
σ θθ ¼Aþ 2
r
and, for the inner cylinder,
B0
σ rr ¼ A0 
r2
(7.32)
B0
σ θθ ¼ A0 þ 2
r
Solutions for both the inner and outer cylinders subjected to the interference pressure, p, can
be obtained from our general solution to the thick-wall cylinder problem with free ends (see
Eq. (7.20)), giving
Pr2i
A¼
r2e  r2i
pr2i r2e
B¼
r2e  r2i
(7.33)
0pr2
A ¼ 2 i2
ri  ru
pr2i r2u
B0 ¼
r2i  r2u

Thus, the stresses are known once we find the interference pressure, p. These stresses do not
ðoÞ
involve the elastic constants but the displacements do. If we let ur be the displacement of
ði Þ
the outer cylinder and ur be the displacement of the inner cylinder, then
 7.1 Displacement-Based Solutions 189

 
1 B
uðroÞ ¼ ð1 þ v1 Þ þ ð1  v1 ÞAr
E1 r
  (7.34)
1 B0
uðriÞ ¼ ð1 þ v2 Þ þ ð1  v2 ÞA0 r
E2 r

The final state of the interface between the two cylinders after the shrink-fit process is shown
ðoÞ
in Fig. 7.5. The outer cylinder experiences a stretching corresponding to ur and the inner
ði Þ
cylinder experiences a shortening, ur . The interface between the two cylinders after the
shrink-fit lies somewhere between ri and ri þ Δ, but the interference is usually quite small so
we can assume from Fig. 7.5 that

uðroÞ  uðriÞ ¼ Δ at r ffi ri (7.35)

If we place Eq. (7.34) into Eq. (7.35), we can obtain the relationship between the pressure, p,
and the interference, Δ. The result is algebraically rather complex, so consider the important
special case where E 1 ¼ E 2 ¼ E, v1 ¼ v2 ¼ v, i.e., the materials are the same for both
cylinders. In that case we find from Eq. (7.35)
   
p ð1 þ vÞr2e ri þ ð1  vÞr3i p ð1 þ vÞr2u ri  ð1  vÞr3i
 ¼Δ (7.36)
E r2e  r2i E r2i  r2u

Solving for Δ in terms of p:

"  #
p 2r3i r2e  r2u
Δ¼   (7.37)
E r2e  r2i r2i  r2u

or, solving for p,

"  #
r2e  r2i r2i  r2u
p ¼ EΔ  (7.38)
2r3i r2e  r2u

In these shrink-fit solutions, we again used the plane-stress solutions found for open
cylinders. One might question the plane-stress conditions in this case, however, because in
the shrink-fit process there may be axial frictional forces developed that restrain the axial
deformations of the cylinders. If, in fact, we assume that there
is no axial deformation allowed during the shrink-fit, then we
have a plane-strain situation instead. From the results of
(i )
ur( o ) −ur Chapter 6, we can obtain the plane-strain result corresponding

r to Eq. (7.38) by making the replacement E ! E= 1  v2 .
i
D

Figure 7.5 The displacements experienced at the interface between the

final position of the interface outer cylinder and inner cylinder in a shrink-fit.
190 Analytical Solutions

Example 7.2 Shrink-Fit

Consider the shrink-fit between a cylindrical jacket and a solid cylinder ðru ¼ 0Þ.
(a) What is the relationship between the interference pressure, p, and the interfer-
ence, Δ? (b) If the jacket and the solid cylinder both have the same length L before
the shrink fit, what is the expression for their difference in lengths afterwards?
Assume that there is no axial stress developed between the two bodies during the
shrink-fit.

(a) From Eq. (7.38) we have, with ru ¼ 0

 2 
re  r2i
p ¼ EΔ
2ri r2e

(b) The stresses in the solid cylinder are (see Eq. (7.32) and Eq. (7.33)) σ srr ¼ σ sθθ ¼ p, σ zz ¼ 0
and the axial strain is
1 2vp
eszz ¼
ðσ zz  vσ rr  vσ θθ Þ ¼
E E
 
1   v σ j
rr þ σ j
θθ
For the jacket we have ejzz ¼ 0  vσ jrr  vσ jθθ ¼ but from Eq. (7.32) and
E E
Eq. (7.33):
2r2i
σ jrr þ σ jθθ ¼ 2A ¼ p
r2e  r2i

2v pr2
so that we obtain ejzz ¼  2 i2 . Thus, for their differences in length we have
E re  ri
 2v pr2 L
ΔL ¼ L eszz  ejzz ¼  2 e 2
E re  ri

7.2 Airy Stress Function Solutions in Cartesian Coordinates

One of the other choices of governing equations described in Chapter 6 treated the stresses as
the fundamental unknowns, which required satisfying the equilibrium and compatibility
equations. As discussed in that chapter, the Airy stress function is often used to simplify
these equations, resulting in a single equation for plane-stress and plane-strain problems. In
this section we will discuss obtaining solutions in 2-D problems using an Airy stress
function approach.
 7.2 Airy Stress Function Solutions in Cartesian Coordinates 191

7.2.1 Solutions of the Biharmonic Equation in Cartesian Coordinates

Plane-stress or plane-strain problems where there are no body forces both satisfy the
biharmonic equation for the Airy stress function. In Cartesian coordinates we have
∂4 ϕ ∂4 ϕ ∂4 ϕ
r4 ϕ ¼ þ2 2 2þ 4 ¼0 (7.39)
∂x 4 ∂x ∂y ∂y
One can obtain some useful solutions by representing the Airy stress function as the sum of
polynomials where
X
M X
N
ϕ¼ Amn xm yn (7.40)
m¼1 n¼1

and Amn are constants. Polynomials up to third order (cubic polynomials) satisfy the
biharmonic equation without any restriction on the constants. Fourth-order and higher
polynomials require the coefficients to be related.
There are many interesting problems that can be solved with such polynomials, but in this book
our coverage will be limited to a single example. In using these functions, we have three objectives.
First, we want to show a nontrivial example of a problem solved with such polynomials. Second,
we want to highlight the importance of the boundary conditions. Third, and finally, we want to
compare a solution with a corresponding solution from strength of materials. The simply-
supported beam problem discussed in the following section will meet those three objectives.

7.2.2 A Simply Supported Beam

Consider a rectangular beam of height, 2h, length, 2L, and width, b, that has a uniform
distributed pressure, p0 , acting on it top surface and is simply supported at its two ends, as
shown in Fig. 7.6. This is a geometry suited to the use of Cartesian coordinates and we
expect from the strength of materials solution for this problem that the stresses can be
described with relatively few polynomial terms. The polynomial we will use for this problem
is a fifth-order polynomial in terms of six coefficients:
ϕ ¼ ax2 þ bx2 y þ cy3 þ dx4 y þ ex2 y3 þ fy5 (7.41)

y Figure 7.6 A simply supported beam

p0 of height 2h, length 2L, and width b
subjected to a uniform pressure, p0 ,
acting on its top face. The reaction
h
width = b forces at the ends are also shown.
x
h

L L

p0bL p0bL
192 Analytical Solutions

which has terms that are only even in x and odd in y. The stresses generated by these
polynomials are
σ xx ðx; yÞ ¼ 6cy þ 6ex2 y þ 20f y3
σ yy ðx; yÞ ¼ 2a þ 2by þ 12dx2 y þ 2ey3 (7.42)

σ xy ðx; yÞ ¼  2bx þ 4dx þ 6exy
3 2


where the stress functions σ xx , σ yy are even in x and odd in y while the shear stress, σ xy , is
odd in x and even in y. This behavior is what we see in a strength of materials solution for
this problem so it might be expected that the functions in Eq. (7.41) are a reasonable choice.
There are more detailed ways in which we can try to choose such functions (see [1], for
example) but we will not go into them here. Since we have used a fifth-order polynomial,
there will be relations that have to be satisfied among these six coefficients. If we place the
polynomial of Eq. (7.41) into the biharmonic equation, we find ð24d þ 24e þ 120f Þy ¼ 0,
which must be satisfied for all y, so we have
24d þ 24e þ 120f ¼ 0 (7.43)

Thus, there are actually five free constants in our solution which we must use to satisfy the
boundary conditions. On the top and bottom of the beam we have the conditions
σ yy ðx; hÞ ¼ p0
σ xy ðx; hÞ ¼ 0
(7.44)
σ yy ðx; hÞ ¼ 0
σ xy ðx; hÞ ¼ 0

The boundary conditions on the ends are, however, more problematic because, as we
discussed in Chapter 6, different simple support systems produce different stresses. Thus,
we will use the idea contained in the principle of Saint-Venant that static equivalency is the
essential condition that must be satisfied and simply require the stresses to produce the
correct support forces and moments. Thus, at the ends we will specify so-called Saint-Venant
boundary conditions, namely at x ¼ L the shear stress must produce the end reaction force
(see Fig. 7.6), and zero axial force and moment
ð
þh

σ xy bdy ¼ p0 bL
h

ð
þh

σ xx bdy ¼ 0 (7.45)
h

ð
þh

σ xx ybdy ¼ 0
h
 7.2 Airy Stress Function Solutions in Cartesian Coordinates 193

where the minus sign is present since σ xy is taken positive acting in the negative y-direction
on the face of the beam at x ¼ L. Similarly, at x ¼ L
ð
þh

σ xy bdy ¼ Py ¼ p0 bL
h
ð
þh

σ xx bdy ¼ 0 (7.46)
h
ð
þh

σ xx ybdy ¼ 0
h

However, since we are using an Airy stress function that satisfies equilibrium throughout the
beam and the boundary conditions are satisfied on the top and bottom of the beam exactly, if we
satisfy these Saint-Venant boundary conditions on just one end of the beam they will be
automatically satisfied at the other end. Thus, a complete set of boundary conditions for this
problem is given by Eq. (7.44) and Eq. (7.46). Consider applying the boundary conditions of Eq.
(7.44). We find from the expressions for the stresses, Eq. (7.42) and the first equation in Eq. (7.44)
σ yy ðx; hÞ ¼ p0
(7.47)
! 2a þ 2bh þ 12dx2 h þ 2eh3 ¼ p0

which must be satisfied for all x. This can only be the case if we have
d¼0
(7.48)
2a þ 2bh þ 2eh ¼ p0

From the other three boundary conditions in Eq. (7.44) we find

σ xy ðx; hÞ ¼ 0 ! b ¼ 3eh2
σ yy ðx; hÞ ¼ 0 ! 2a  2bh  2eh3 ¼ 0 (7.49)
σ xy ðx; hÞ ¼ 0 ! b ¼ 3eh2

where two of these boundary conditions lead to identical equations. We can solve the five
independent equations contained in Eq. (7.43), Eq. (7.48), and Eq. (7.49) for the five
constants ða; b; d; e; f Þ as
p0 3p0 p0 p0
a¼ , b¼ , d ¼ 0, e¼ , f ¼ (7.50)
4 8h 8h3 40h3
The only remaining constant to find is the constant c. The boundary conditions left to satisfy are
the Saint-Venant boundary conditions, where we will use those conditions at x ¼ L. First, we have
ð
þh

σ xy ðL; yÞbdy ¼ p0 bL !  4bh þ 4eh3 bL ¼ p0 bL (7.51)
h
194 Analytical Solutions

which is already satisfied by the constants in Eq. (7.50). Next, we have

ð
þh

σ xx ðL; yÞbdy ¼ 0 (7.52)

h

which is automatically satisfied since the stress σ xx is odd in y. Thus, we are left with the last
boundary condition
ð
þh

σ xx ðL; yÞybdy ¼ 0 ! 4ch3 þ 2eL2 h3 þ 8fh5 ¼ 0 (7.53)

h

which gives the last coefficient as


p0 2 L2
c¼  (7.54)
8h 5 h2

You can verify that these constants also satisfy the Saint-Venant boundary conditions at
x ¼ L, as we stated they should.
The stresses for the beam are now


p 6 3L2 3p p
σ xx ðx; yÞ ¼ 0  2 y þ 03 x2 y  03 y3
h 20 4h 4h 2h
p0 3p0 p
σ yy ðx; yÞ ¼   y þ 03 y3 (7.55a)
2 4h 4h
3p0 3p
σ xy ðx; yÞ ¼ x  03 xy2
4h 4h
which we will rewrite and group as


3p0  2 3y y3
σ xx ðx; yÞ ¼ 3 x  L y þ p0
2

4h 10h 2h3


1 3y y3
σ yy ðx; yÞ ¼ p0 þ  3 (7.55b)
2 4h 4h


3p y2
σ xy ðx; yÞ ¼ 0 x 1  2
4h h
This is in a form that we can now directly compare to a strength of materials solution where
the bending stress and shear stress are given as
M ðx0 Þy
σ strength
xx ¼
I
(7.56)
V ðx0 ÞQðyÞ
σ strength
xy ¼
Ib
 7.2 Airy Stress Function Solutions in Cartesian Coordinates 195

Here M is the bending moment and V is the shear force acting in the beam, both acting
as shown in Fig. 7.7, and the distance x0 ¼ x þ L is measured from the left-hand end of

the beam. For our rectangular beam we have I ¼ 2bh3 =3 and QðyÞ ¼ bh2 1  y2 =h2 =2.
From equilibrium of the element shown in Fig. 7.7, we have V ðx0 Þ ¼ p0 bx,

M ðx0 Þ ¼ p0 b x2  L2 , as expressed in terms of the x-coordinate, so that Eq. (7.55b) can
be written finally as


M ðx0 Þy 3y y3
σ xx ðx; yÞ ¼ þ p0 
I 10h 2h3


1 3y y3
σ yy ðx; yÞ ¼ p0 þ  3 (7.57)
2 4h 4h
V ðx0 ÞQðyÞ
σ xy ðx; yÞ ¼ 
Ib
Our solution has two additional stresses not contained in the strength of materials solution.
They are


3y y3
σ xx ¼ p0
add

10h 2h3

 (7.58)
1 3y y3
σ yy ¼ p0 þ  3
add
2 4h 4h

We have seen the additional normal stress, σ add yy , before in Chapter 3, where we showed that
the equilibrium equations required such a stress to be present and we derived an expression for
this stress in Eq. (3.27) that is identical to the current expression if we make the following
changes from the solution in Chapter 3: z ! y, t ! b,h ! 2h, qðxÞ ! p0 b. Our current
expression for σ add
yy , which is based on satisfying both equilibrium and compatibility, is plotted
in Fig. 7.8. The other additional stress, σ add
xx , is entirely new. This term is plotted in Fig. 7.9.
This stress is called a self-equilibrated stress since it does not produce any net axial force or
moment. The axial force is obviously zero since the stress is an odd function of y, so its integral
over the cross-section of the beam is automatically zero (also see Fig. 7.8). The moment is also
zero, which is easy to verify by direct integration. If Saint-Venant’s principle was satisfied, we
would expect this self-equilibrated stress would produce zero stress outside
p0 a small region at the ends, but this is clearly not the case as this stress is
present throughout the beam. Thus, this is an example where Saint-
M ( x' ) Venant’s principle is violated. However, the maximum strength of materials
bending stress, which occurs at top/bottom faces at the center of the beam,

V ( x' ) Figure 7.7 Element of the simply supported beam in Fig. 7.6, showing the
p0bL internal shear force and bending moment acting in the beam. The coordinate x0
x' is measured from the left end of the beam.
196 Analytical Solutions

Figure 7.8 The distribution of the normalized normal 1

stress, σ yy , across the height of the simply supported 0.8
beam of Fig. 7.6. 0.6
0.4
0.2
y/h
0
–0.2
–0.4
–0.6
–0.8
–1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
−V yyadd / p0

Figure 7.9 The distribution of the normalized normal 1

stress, σ xx , across the height of the simply supported 0.8
beam of Fig. 7.6, showing the regions of positive and 0.6
negative values. 0.4 +
0.2
y/h 0
–0.2
–0.4
–0.6
–0.8
–1
+
– 0.2 –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2

V xxadd / p0

is still the dominant stress in this solution since the maximum strength of materials bending
stress and maximum additional stresses in the beam are
 3p0 L2  p0  
max σ strength
xx ¼ , max σ add
xx ¼ , max σ add
yy ¼ p0 (7.59)
4h2 5
so that we have the ratios
 

max σ add 4 h2 max σ add
yy 4 h2
 xx  ¼ ,  ¼ (7.60)
max σ strength
xx
15 L2 max σ strength
xx
3 L2

For long, slender beams where h=L  1, both of these additional stresses can be neglected.
 7.3 Airy Stress Function Solutions in Polar Coordinates 197

7.3 Airy Stress Function Solutions in Polar Coordinates

There are a number of important problems that are best described in polar coordinates
whose solutions can be obtained with the use of the Airy stress functions. In this section we
will examine two cases of this type – the stress concentration around a hole in a plate, and
the stresses in a curved beam. The Airy stress function solution for the curved beam will be
compared to an equivalent strength of materials solution.

7.3.1 Michell’s Solutions

For zero body forces the Airy stress function satisfies the biharmonic equation. If one seeks
solutions to the biharmonic equation in polar coordinates of the form ϕ ¼ f ðrÞgðθÞ, then
there are a number of such solutions, called separation of variables solutions, that one can
obtain. We will not go through the details of the separation of variables solution process.
Instead, we will simply write down a number of the important solutions of this type due to
Michell [1], given as
ϕ ¼ a0 ln r þ b0 r2 þ c0 r2 ln r þ d 0 r2 θ þ e0 θ
 c1 
þ a1 r þ b1 r3 þ þ d 1 r ln r cos θ þ j 1 rθ cos θ
r
 g 
þ e1 r þ f 1 r3 þ 1 þ h1 r ln r sin θ þ k1 rθ sin θ
r
(7.61)
X ∞ 
þ an rn þ bn rnþ2 þ cn rn þ d n rnþ2 cos nθ
n¼2

X
∞ 
þ en rn þ f n rnþ2 þ gn rn þ hn rnþ2 sin nθ
n¼2

Most of these solutions depend on both r and θ, but we see there is a subset of axisymmetric
solutions where θ is absent given by
ϕ ¼ a0 ln r þ b0 r2 þ c0 r2 ln r (7.62)

These solutions produce the stresses

a0
σ rr ¼ þ 2b0 þ c0 ð2 ln r þ 1Þ
r2
a0 (7.63)
σ θθ ¼  2 þ 2b0 þ c0 ð2 ln r þ 3Þ
r
σ rθ ¼ 0

Note that the terms involving the coefficients ða0 , b0 Þ are of the same form that we used in
Eq. (7.14) to solve the pressurized vessel and shrink-fit problems, so those problems could
also be solved with the use of these Airy stress functions. Since these stresses satisfy the
compatibility equation, we can relate the stresses to the strains and then integrate those
198 Analytical Solutions

strains to find the displacements. The details are rather lengthy so we will just give the final
results. For plane stress, the displacements ður , uθ Þ are

1 ð1 þ vÞa0
ur ¼  þ 2ð1  vÞb0 r  ð1 þ vÞc0 r
E r


þ2ð1  vÞc0 r ln r þ C 1 sin θ þ C 2 cos θ

4c0 rθ
uθ ¼ þ C 1 sin θ  C 2 cos θ þ C 3 r (7.64)
E
while for the case of plane strain (which we can obtain from the plane-stress solution as
described in Chapter 6):

1 ð1 þ vÞa0
ur ¼  þ 2ð1  2vÞð1 þ vÞb0 r  ð1 þ vÞc0 r
E r


þ2ð1  2vÞð1 þ vÞc0 r ln r þ C 1 sin θ þ C 2 cos θ

4 1  v2 c0 rθ
uθ ¼ þ C 1 sin θ  C 2 cos θ þ C 3 r (7.65)
E
where the constants ðC 1 , C 2 Þ are constants of integration that describe a rigid body translation,
and the constant C 3 is a constant of integration that describes a rigid body rotation. These rigid
body terms produce no stresses or strains but may be needed to satisfy the boundary conditions
of a problem. In the pressure vessel and shrink-fit problems, we had assumed ur ¼ ur ðrÞ, uθ ¼ 0
so that such rigid body terms were absent. If our body includes the origin, the stresses and
displacements are unbounded at r ¼ 0 unless we set a0 ¼ c0 ¼ 0. If the body does include a hole
centered at r ¼ 0 (i.e., it is multiply connected) then solutions for the displacements involving
both a0 and c0 are acceptable since the point r ¼ 0 is not a part of the body but the solutions are
not single-valued since uθ ðr; 0Þ ¼6 uθ ðr; 2π Þ. This should not be surprising since we saw in our
discussions of compatibility in Chapter 4 that for multiply connected bodies, the compatibility
equations by themselves are not sufficient to ensure we have single-valued displacements. If we
provide an additional criterion here such as
2ðπ 2ðπ
∂uθ
duθ ¼ dθ ¼ 0 (7.66)
∂θ
0 0

then c0 ¼ 0 and the displacements are single-valued.

7.3.2 Stress Concentration at a Circular Hole in a Large Plate

One problem that can be described with Michell’s solutions is the case of an infinitely large
thin plate containing a hole of radius a (Fig. 7.10) and under a uniform stress σ xx ¼ σ 0 at
infinity. This is a plane-stress problem where the only nonzero stresses are ðσ rr , σ θθ , σ rθ Þ.
 7.3 Airy Stress Function Solutions in Polar Coordinates 199

Figure 7.10 An infinitely large plate containing a

hole of radius a and subjected to a uniform axial
y stress, σ 0 , acting in the x-direction.
r
V0 T V0
a
x

dA V rT V rr
Figure 7.11 (a) An element in the plate at a
V rr large distance from the hole where the
V rT T
T stresses σ yy and σ xy are zero. (b) An element
V xx V 0 T T V xx dAcosT at an arbitrary point in the plate, where there
V rT is a full state of stress ðσ rr , σ θθ , σ rθ Þ.
V xy
dAcosT
dA
V TT
dAsinT
(a) (b)

To solve this problem, we need to decide which parts of the Michell solutions will be needed.
Consider a small element at infinity (see Fig. 7.11a) which has the stresses σ xx ¼ σ 0 , σ rr , and
σ rθ acting on its faces (σ yy ¼ 0 at infinity so it is not shown). From the force equilibrium of this
element in the r- and θ-directions it follows that
σ0
σ rr ¼ σ 0 cos 2 θ ¼ ð1 þ cos 2θÞ
2
(7.67)
σ0
σ rθ ¼ σ 0 sin θ cos θ ¼  sin 2θ
2
where we have also used double angle formulas of trigonometry. But from the stress
invariant I 1 ¼ σ xx þ σ yy ¼ σ 0 ¼ σ rr þ σ θθ we also find
σ0
σ θθ ¼ ð1  cos 2θÞ (7.68)
2
We want the stresses at infinity to have these forms in polar coordinates and because of the
symmetry of the circular hole, we expect that there will also be axisymmetric stresses
generated from the presence of the hole, so we will try an Airy stress function of the form
ϕ ¼ a0 ln r þ b0 r2 þ c0 r2 ln r
 (7.69)
þ a2 r2 þ b2 r4 þ a02 r2 þ b02 cos 2θ
200 Analytical Solutions

but we must set c0 ¼ 0 to have single-valued displacements. The stresses generated are
a0 
σ rr ¼ 2 þ 2b0  2a2 þ 6a02 r4 þ 4b02 r2 cos 2θ
r

σ rθ ¼ 2 a2 þ 3b2 r2  3a02 r4  b02 r2 sin 2θ (7.70)
a0 
σ θθ ¼ 2 þ 2b0 þ 2a2 þ 12b2 r2 þ 6a02 r4 cos 2θ
r
We must set b2 ¼ 0 to have stresses that are only functions of θ as r ! ∞, as demanded by
Eq. (7.67) and Eq. (7.68). We then can match the stresses in Eq. (7.70) as r ! ∞ with the
remaining stresses in Eq. (7.67) and Eq. (7.68) to obtain
σ0 σ0
a2 ¼  , b0 ¼ (7.71a)
4 4
From the boundary conditions σ rr ¼ σ rθ ¼ 0 at r ¼ a we find

a0 =a2 þ σ 0 =2  σ 0 =2 þ 6a02 =a2 þ 4b02 =a2 cos 2θ ¼ 0

σ 0 =2  6a02 =a4  2b02 =a2 sin 2θ ¼ 0

which, equating the constant term and terms involving θ separately equal to zero, gives the
remaining constants as
a0 ¼ σ 0 a2 =2
b02 ¼ σ 0 a2 =2 (7.71b)
a02 ¼ σ 0 a4 =4

and the stresses in the plate are, therefore, explicitly



σ0 a2 σ0 3a4
σ θθ ðr; θÞ ¼ 1þ 2  1 þ 4 cos 2θ
2 r 2 r


σ0 3a 4
2a 2
σ rθ ðr; θÞ ¼ 1  4 þ 2 sin 2θ (7.72)
2 r r



σ0 a 2
σ0 3a4 4a2
σ rr ðr; θÞ ¼ 1 2 þ 1 þ 4  2 cos 2θ
2 r 2 r r

It is useful to consider the stress σ xx =σ 0 as this stress goes to a constant value of one far from
the hole so that one can see the effect that the hole has in modifying this uniform stress at
infinity. This stress can be calculated from the stresses in Eq. (7.72) by using the element
shown in Fig. 7.11b and examining the force equilibrium in the x-direction. We find
σ xx ¼ σ rr cos 2 θ  2σ rθ sin θ cos θ þ σ θθ sin 2 θ (7.73)

A 2-D contour plot of this stress is shown in Fig. 7.12a. On the top and bottom points of the
hole ðx ¼ 0, y aÞ a maximum stress of σ xx ¼ 3σ 0 is present (Fig. 7.12b). Along the x-axis at
the hole surface, the stress σ xx ¼ 0 by virtue of the boundary conditions and there are small
 7.3 Airy Stress Function Solutions in Polar Coordinates 201

Figure 7.12 (a) A contour plot of σ xx =σ 0 in the plate.

(b) The largest and smallest values of σ xx =σ 0 and where
high stress they occur on the periphery of the hole.

low stress low stress

V0 V0

high stress

(a)
y

3V 0

0 0 x

3V 0
(b)

stresses near those points. The value σ xx ¼ 3σ 0 is the largest stress anywhere in the plate so
the stress concentration factor of the hole (ratio of the maximum stress to the stress at
infinity) is 3.0. Similarly, notches, fillets, and noncircular holes produce stress concentrations
so the stress may be increased significantly over its nominal value away from the geometry
changes. Analytical and numerical evaluations of stress concentration factors for many
standard geometries have been calculated and tabulated in handbooks. Note that, in this
case, the stress concentration factor is independent of the size of the hole. However, if an
edge of a more realistic finite-sized plate is within about four radii from the hole, then the
influence of the edge may start to affect the stress concentration for an infinite plate
calculated here. Another plot of interest is that of the hoop stress, σ θθ , along the periphery
of the hole at r ¼ a, since large tensile values of this stress can initiate cracking at the hole.
From Eq. (7.72) we have (see Fig. 7.13):
σ θθ ða; θÞ ¼ σ 0 ð1  2 cos 2θÞ (7.74)

The extreme values of σ θθ =σ 0 and their location along the edge of the hole are shown in
Fig. 7.13. At the x-axis, the hoop stress has its largest compressive values ðσ 0 Þ while at the
202 Analytical Solutions

y Figure 7.13 The largest and smallest values of the hoop stress and
3V 0 their locations.

y-axis, the hoop stress has its largest tensile values ð3σ 0 Þ, which is
x the most significant stress since cracks are initiated by large
−V 0 −V 0
tensile, not compressive stresses. The 3σ 0 values also appear in
Fig. 7.12b since σ θθ ¼ σ xx at these locations.
3V 0

7.3.3 Pure Bending of a Curved Beam

The axisymmetric Airy stress functions and resulting stresses of Eq. (7.62) and Eq. (7.63) can
also be used to solve the problem of pure bending of a curved beam of rectangular cross-
section, as shown in Fig. 7.14. It is convenient to define a new set of constants c3 ¼ a0 ,
2b0 þ c0 ¼ c1  c2 ln a, and c2 ¼ 2c0 and rewrite the stresses in Eq. (7.63) as
r c3
σ rr ¼ c1 þ c2 ln þ 2
a r
 r  c
σ θθ ¼ c1 þ c2 ln þ 1  2
3 (7.75)
a r
σ rθ ¼ 0

Since the shear stress σ rθ ¼ 0, the boundary conditions at both r ¼ a and r ¼ b are σ rr ¼ 0,
which gives the two equations
c3 ¼ a2 c1

2  (7.76)
a b
c1 2  1 ¼ c2 ln
b a

On the ends of the beam, we use the Saint-Venant boundary conditions that the stresses must
produce the appropriate forces and moments, giving
ðb
σ rθ tdr ¼ 0
a

ðb
σ θθ tdr ¼ 0 (7.77)
a

ðb
σ θθ rtdr ¼ M
a

where t is the thickness of the beam. The first equation in Eq. (7.77) is satisfied identically.
The second equation gives
 7.3 Airy Stress Function Solutions in Polar Coordinates 203

M
M
r
a b a r (a + b)/2

(a) (b)

Figure 7.14 (a) Pure bending of a rectangular cross-section curved beam. (b) The beam cross-section.


hr
1 1 r r ib
ðc1 þ c2 Þðb  aÞ þ c3  þ c2 a ln  ¼0
b a a a a a


 (7.78)
1 1 b b b
! ðc1 þ c2 Þðb  aÞ þ c3  þ c2 a ln  þ 1 ¼ 0
b a a a a

But placing Eq. (7.76) into Eq. (7.78), we find

a c1
c1 ðb  aÞ  c1 ða  bÞ  ðb  aÞðb þ aÞ ¼ 0
b b
(7.79)
a 1
! 1 þ  ða þ bÞ ¼ 0
b b
so that the second equation in Eq. (7.77) is also satisfied. From Eq. (7.76) we can write both
c1 and c3 in terms of c2 as

b2 ln ðb=aÞ
c1 ¼ c2
a2  b2
(7.80)
a2 b2 ln ðb=aÞ
c3 ¼ c2
b2  a2
and we can find c2 from the third equation in Eq. (7.77). The three integral terms in that
equation are
ðb
tc2 b2
tc1 rdr ¼  ln ðb=aÞ
2
a

ðb 
tc2 b2 tc2 b2  a2
tc2 rð ln ðr=aÞ þ 1Þdr ¼ ln ðb=aÞ þ (7.81)
2 4
a

ðb
tc2 a2 b2 ½ ln ðb=aÞ2
tc3 dr=r ¼
a2  b2
a
204 Analytical Solutions

so the third equation in Eq. (7.77) becomes

2 h i3
2
b a
2 2 a 2 2
b ln ðb=a Þ
tc2 4   2 5¼M (7.82)
4 b  a2

which can be solved for c2 in the form


a2
4M 1  2
b
c2 ¼ 2
(7.83)
tb N
where

2

2
a2 a2 b
N ¼ 1  2  4 2 ln (7.84)
b b a

Using the known constants in Eq. (7.75) gives finally


  


4M a2 r a2 b
σ rr ¼ 2 1  2 ln  1  2 ln
tb N b a r a



 (7.85)
4M a2 n  r o a2 b
σ θθ ¼ 2 1  2 1 þ ln  1 þ 2 ln
tb N b a r a

Figure 7.15 plots both of these stresses across the beam cross-section for the case b=a ¼ 2.
Unlike with the straight beam, we see that the bending stress, σ θθ , has a nonlinear distribution
across the cross-section. The radial stress, σ rr , is small and can typically be neglected, as is the
case for the straight beam. If we simply ignore the curvature and treat the beam as straight, we
can use the standard flexure expression and write the bending stress, σ b , as (see Fig. 7.14b)
" #
12Ma r=a  ð1 þ b=aÞ=2
σb ¼ 3 (7.86)
bt ð1  a=bÞ3

Figure 7.15 Normal stresses (normalized) in a curved 20

beam under pure bending for b=a ¼ 2.
10 sqq
s 0
( M / tb2 )
– 10
s rr

– 20

– 30

– 40
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
r/a
 7.3 Airy Stress Function Solutions in Polar Coordinates 205

30 Figure 7.16 Comparison of the normalized bending

stress for a curved beam (solid line) and a straight
20
beam (dashed line) for the case b=a ¼ 2.
10
s
( M / tb ) 0
2

–10 centroid
sb
– 20
sqq
– 30

– 40
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
r/a

A comparison of σ b with the curved beam σ θθ is shown in Fig. 7.16. It can be seen that
although the stress distributions are similar, the neutral axis of the curved beam is no longer
at the centroid of the cross-section, which is another difference between the straight and
curved beam cases.

7.3.4 Comparison with a Curved Beam Strength of Materials Solution

One can use assumptions on the deformation of a curved beam similar to those used for a
straight beam to obtain a strength of materials solution. In this section we will obtain the
strength of materials solution and compare it to the solution found in the previous section. We
will again examine pure bending of a curved beam and assume the cross-section is symmetric
but, unlike the Airy stress function solution, the beam need not be rectangular. We will also
assume plane cross-sections remain plane after deformation and that the only significant stress in
the beam is σ θθ . The geometry and geometrical parameters to be used are shown in Fig. 7.17a.
The radius to the centroid is R and the radius to the neutral axis is Rn . The segment of the beam
shown subtends an angle Δθ. We will again let r be the radius to a general fiber located at P in the
beam. The hoop strain in the beam for the fiber located at P, assuming that plane cross-sections
remain plane so that they rotate through a small angle, Δϕ (see Fig. 7.17b), is


Δl ðRn  rÞΔϕ Rn
eθθ ¼ ¼ ¼ω 1 (7.87)
l rΔθ r

where
Δϕ
ω¼ (7.88)
Δθ
From Hooke’s law we have


Rn
σ θθ ¼ Eeθθ ¼ Eω 1 (7.89)
r
206 Analytical Solutions

centroid
neutral axis
B
B

R Rn − r
Rn
r P
M after
A P deformation
A
'I
'T
(a) (b)

Figure 7.17 (a) An element of a curved beam in pure bending. (b) The deformation, assuming plane
sections remain plane.

Requiring that this stress does not produce an axial force, N, we have
2 3
ð ð ð
dA
N ¼ σ θθ dA ¼ Eω4Rn  dA5
r (7.90)
A A A

¼ EωðRn Am  AÞ ¼ 0

where Am , which has the dimensions of a length, is defined as

ð
dA
Am ¼ (7.91)
r
A

It follows from Eq. (7.90) that the radius of the neutral axis is
A
Rn ¼ (7.92)
Am
which is, in general, not at the centroid. For the bending moment we have
ð ð

Rn
M ¼  σ θθ ðR  rÞdA ¼ Eω  1 ðR  rÞdA
r
A A
2 3
ðð ð ð (7.93)
dA
¼ Eω4Rn R  R dA  Rn dA þ rdA5
r
A A A A

! M ¼ EωRn ðA  RAm Þ
 7.3 Airy Stress Function Solutions in Polar Coordinates 207

where the final result in Eq. (7.93) is obtained since the second and fourth integrals in
Eq. (7.93) cancel from the definition of the location of the centroid. Solving Eq. (7.93) for
Eω and placing that result back into Eq. (7.89), the flexure stress is given by
M ðA  rAm Þ
σ θθ ¼ (7.94)
AðA  RAm Þr

To compare this result with the Airy stress function solution, we have for the rectangular
beam


b
Am ¼ t ln , A ¼ tðb  aÞ, R ¼ ða þ bÞ=2 (7.95)
a

so that the strength of materials solution becomes, after some rearrangement,

!


b2 b r b
2 2  1  ln
M a a a a
σ θθ ¼ 2



  (7.96)
tb b b b b r
1 2 1  þ 1 ln
a a a a a

We can compare this result and the corresponding straight beam solution with the Airy
stress solution obtained previously. It is convenient to use the slenderness ratio R/h to make
that comparison, where h is the thickness of the beam in the r-direction and R is the distance
to the centroid (Fig. 7.18a). Table 7.1 shows the ratios of the maximum bending stresses in
the beam calculated (a) with the curved beam strength of materials and Airy stress function
solutions and (b) with the straight beam strength of materials solution and the Airy stress
function solution. It can be seen for large slenderness ratios (R/h > 5 approximately) one can
neglect curvature effects and use the ordinary straight beam flexural stress result. For smaller

Figure 7.18 (a) Definition of the terms in

C2 A2
the slenderness ratio, R/h, for a
R2 rectangular beam. (b) The areas and radii
h = (b − a ) C1 A1 to the centroids (C1, C2) of a
composite area.
R1

R = (a + b) / 2

(a) (b)
208 Analytical Solutions

Table 7.1 Comparison of the ratios of the maximum bending stresses for (a) the curved beam
strength of materials solution divided by the Airy stress function solution, and (b) for the
straight beam strength of materials solution divided by the Airy stress function solution

R/h (a) Curved strength/Airy stress solutions (b) Straight strength/Airy stress solutions

0.65 1.0455 0.4390

0.75 1.0124 0.5262
1.0 0.9970 0.6545
1.5 0.9961 0.7737
2.0 0.9973 0.8313
3.0 0.9986 0.8881
5.0 0.9994 0.9331

slenderness ratios we see that there is little difference between the curved beam strength of
materials solution and the Airy stress function solution as long as the R/h value is not too
small. Thus, generally we can use the straight beam strength formula for R/h > 5 and the
curved beam strength formula for R/h > 0.5.
The curved beam strength of materials solution can be used for more complex cross-
sections composed of composite areas, Ai (see Fig. 7.18b), where
X
A¼ Ai
X
Am ¼ Ami (7.97)
P
Ri Ai
R¼ P
Ai

It should be noted that normal forces and shear forces are also present in many curved beam
problems, and curved beam strength of materials solutions can account for those forces as
well (see [2], for example). However, we will not give those details here.

7.3.5 Concentrated Force on a Wedge

Another case that we will examine with a Michell solution is the case of a concentrated force
P acting on a wedge of thickness t and opening angle 2α (Fig. 7.19a). The boundary
conditions for r > 0, θ ¼ α are σ θθ ¼ σ rθ ¼ 0. If we try an Airy stress function
ϕ ¼ Crθ sin θ, where C is a constant, the stresses are given as
2C
σ rr ¼ sin θ
r (7.98)
σ θθ ¼ σ rθ ¼ 0

so the boundary conditions on the faces of the wedge are satisfied. To find the constant C,
consider the radial stress acting on a circular cut-out at the wedge tip (Fig. 7.19b).
 7.3 Airy Stress Function Solutions in Polar Coordinates 209

P
y
V rr
y

P q
x
e
q=− q =D
x
(a) (b)

Figure 7.19 (a) Concentrated force on a wedge. (b) Geometry near the wedge tip.

This radial stress will not produce any force in the y-direction or moment about the wedge
tip, but it should produce the force P in the x-direction, so we must have, at r ¼ ε,
ð
þα

Fx ¼ ðσ rr cos θÞtεdθ ¼ P (7.99)

α

which gives
ð
þα
 þα
θ sin 2θ
2Ct cos 2 θdθ ¼ 2Ct þ ¼P
2 4 α (7.100)
α
P=t
!C¼
2α þ sin 2α
The radial stress is then given explicitly as
2P cos θ
σ rr ¼ (7.101)
tð2α þ sin 2αÞr

This stress is singular near r ¼ 0 because the applied force is idealized as a concentrated
force.

7.3.6 Concentrated Force on a Planar Surface

An important special case of the wedge problem is when α ¼ π=2 so that the concentrated
force acts on a plane surface (Fig. 7.20a) and where
2P cos θ
σ rr ¼ (7.102)
πt r
210 Analytical Solutions

P
P
P
t t

y
r
T

(a) (b)

Figure 7.20 (a) Concentrated force on a planar surface. (b) Plane-stress and plane-strain problems.

Since this solution is independent of the elastic constants it can be used both for a plane-
stress problem where the load is distributed over the thickness of a large thin plate or a
plane-strain problem where the load is distributed on the surface of a long body (see
Fig. 7.20b). Of course, the displacements in the plane-stress and plane-strain problems
are different.
The equation for a circle of diameter d can be written as r ¼ d cos θ so stress distribution
of Eq. (7.102) has the property that the radial stress is a constant on such circles as shown in
Fig. 7.21. If we take a set of axes ðx0 , y0 Þ in the ðr; θÞ directions, we see that the state of stress
is just a uniaxial stress σ x0 x0 ¼ σ rr . Thus, the magnitude of the maximum in-plane shear stress
is just jσ rr j=2. If we plot the contours of this maximum shear stress in MATLAB® we see
these circles clearly in Fig. 7.22, where the circles become more and more dense near the load
because the stresses are singular there.
We can use the stress transformation relations to obtain the stresses along the x- and
y-directions as
2P cos 3 θ
σ xx ¼ σ rr cos 2 θ ¼ 
πt r
2P sin 2 θ cos θ
σ yy ¼ σ rr sin 2 θ ¼  (7.103)
πt r
2P sin θ cos 2 θ
σ xy ¼ σ rr sin θ cos θ ¼ 
πt r
or, since
cos θ ¼ x=r
sin θ ¼ y=r (7.104)
r ¼x þy
2 2 2
 7.3 Airy Stress Function Solutions in Polar Coordinates 211

we have equivalently
2P x3
σ xx ¼ 
πt ðx2 þ y2 Þ2

2P y2 x
σ yy ¼  (7.105)
πt ðx2 þ y2 Þ2

2P yx2
σ xy ¼ 
πt ðx2 þ y2 Þ2

This concentrated load solution is very useful as a building block for obtaining the stresses
due to an arbitrary vertical force distribution acting on the plane surface. To see this, let us
displace the load so it acts at a distance y1 along the y-axis and let the force P be a force
dP ¼ pðy1 Þtdy1 due to a pressure pðy1 Þ acting on a small area t dy1 (Fig. 7.23). Then if we
superimpose such solutions due to an arbitrary pressure acting on the entire surface we have
from Eq. (7.105) with the replacement y ! y  y1
ð
2 x3 pðy1 Þdy1
σ xx ðx; yÞ ¼   2
π
x2 þ ðy  y1 Þ2
P ð
2 xðy  y1 Þ2 pðy1 Þdy1
σ yy ðx; yÞ ¼   2 (7.106)
π
x2 þ ðy  y1 Þ2
d s rr = constant ð
2 x2 ðy  y1 Þpðy1 Þdy1
σ xy ðx; yÞ ¼   2
π
x2 þ ðy  y1 Þ2

Figure 7.21 Circles of constant radial stress.

Figure 7.22 A contour plot of the maximum shear stress due to the
concentrated load.
212 Analytical Solutions

p ( y1 ) Figure 7.23 Geometry for a concentrated load when the load is due to
a pressure distribution pðy1 Þ acting over and small area on
dy1 the surface.

O q r y
y1
7.3.7 Elastic Bodies in Contact
One problem where Eq. (7.106) finds application is the prob-
x lem of two bodies in contact, such as the two cylinders shown
in Fig. 7.24a. When these cylinders are pressed together with
a force P a high pressure between the bodies is developed over a very small area. This
pressure can cause the materials to yield and fail so it is important to know the maximum
pressure exerted over the contact area and the size of the contact region. As shown in
Fig. 7.24a the two cylinders can have different shear moduli and Poisson’s ratios given by
ðG 1 , ν1 Þ and ðG 2 , ν2 Þ as well as different radii ðR1 , R2 Þ. The length of both cylinders is
assumed to be t. The equal and opposite pressure distributions are shown in Fig. 7.24b as
well as the half-length, a, of the contact region. The solution to this problem can be
found by using a set of assumptions called Hertz contact theory [3]. We will not go over
the details of such a solution but will present some of the major results. Even though the
surfaces in contact are curved, the contact area is typically very small so that the surfaces
can be considered to be locally plane (see Fig. 7.24b), thus allowing us to use Eq. (7.106)
to calculate the stresses. The pressure pðy1 Þ is found to be the distribution
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p0
pðy1 Þ ¼ a2  y21 (7.107)
a
where p0 is the maximum pressure. This maximum pressure and the contact half-distance, a,
can be found explicitly as
  1=2
2P ðR1 þ R2 Þ
p0 ¼
πt ðk1 þ k 2 ÞR1 R2
(7.108)
 
2P ðk1 þ k2 ÞR1 R2 1=2
a¼
πt ðR1 þ R2 Þ

where
ð 1  ν1 Þ
k1 ¼
G1
(7.109)
ð 1  ν2 Þ
k2 ¼
G2
The stresses in a cylinder are thus given from Eq. (7.106) as
 7.3 Airy Stress Function Solutions in Polar Coordinates 213

P Figure 7.24 (a) Two elastic cylinders pressed

together. (b) The surfaces in contact where a
pressure p is exerted over the contact area. Locally
a the two surfaces are considered to be planar.
R1 O
G1 ,Q 1

O p

R2 p
y
G2 ,Q 2
O
a
P
x

(a) (b)

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð
þa
x 3
a2  y21 dy1
2p0
σ xx ðx; yÞ ¼   2
πa 2
a x þ ðy  y1 Þ
2

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
xð
þa
ð y  y Þ 2
a2  y21 dy1
2p 1
σ yy ðx; yÞ ¼  0  2 (7.110)
πa
a x2 þ ðy  y1 Þ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð 2
þa
x ðy  y1 Þ a2  y21 dy1
2p
σ xy ðx; yÞ ¼  0  2
πa
a x2 þ ðy  y1 Þ2

These expressions can be integrated numerically to find the normalized maximum in-plane
shearing stress, τ max =p0 , where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
τ max ¼ σ xx  σ yy =4 þ σ 2xy (7.111)

The results are shown in the 2-D contour plot of

high-stress
region Fig. 7.25. Away from the surface, the plot looks
much like that of a point source. However, here a

Figure 7.25 Contour plot of the normalized maximum

shearing stress, τ max =p0 , due to the contact
pressure distribution.
214 Analytical Solutions

local maximum normalized shear stress in the body occurs at a distance approximately
0.78a below the surface (within the region indicated below the distributed load) and has a
value of about 0:3p0 . In practice this means that failure will also likely occur first below
the surface and result in a chunk of the material being spalled off, a type of failure that is
commonly seen in such contact problems.

Example 7.1 Steel Cylinder Pressed on a Steel Plate

To illustrate the type of numerical values that can be found in contact problems, consider the
case of a steel cylinder of radius R1 ¼ 0:25 in. being pressed by a force/unit length of P=t ¼
50 lb=in: while contact with a steel plate (Fig. 7.26). Let G ¼ 12  106 psi, ν ¼ 0:3 for both
the cylinder and the plate. Note that in this case for the plate we have R2 ¼ ∞. The half-
length contact distance is therefore
 1=2
4P ð1  νÞ
a ¼ R1
πt G
 1=2
ð4Þð50 lb=in:Þð0:7Þ (7.112)
¼ ð0:25 in: Þ
πð12  10 lb=in: Þ
2

¼ 9:64  104 in:

while the maximum pressure is

 1=2 "  #1=2
P G ð50 lb=in:Þ 12  106 lb=in:2
p0 ¼ ¼
πtR1 ð1  νÞ πð0:25Þ ð0:7Þ (7.113)
¼ 33 036 psi

showing that the contact area is indeed quite small and that it is easy to develop very high
stresses in such problems.

7.4 Stress Singularities

Stress singularities appear in concentrated load problems because of the assumption that
the load is idealized as acting at a point. In the 2-D problems just discussed, we saw that
the stresses had a 1/r singular behavior near r ¼ 0. Stress singu-
P/t = 50 lb/in. larities can also appear if we assume the geometry has sharp
changes. Thus, let us consider the two-dimensional problem of
a sharp corner (notch) in a plate that is being loaded in some
0.25 in.
Figure 7.26 A steel cylinder being pressed in contact with a steel plate.
 7.4 Stress Singularities 215

Figure 7.27 (a) Notch in a loaded plate. (b) The

polar coordinates ðr; θÞ at the notch tip.
r
E
D T

(a) (b)

fashion, as shown in Fig. 7.27a. We want to examine the stresses in the neighborhood of
the notch tip in terms of the polar coordinates ðr; θÞ as seen in Fig. 7.27b. We could use an
Airy stress function approach (see [4], for example), but instead we will describe the
problem in terms of displacements, as done by Karp and Karal ([5], [6]). The details are
rather lengthy so we will just outline the steps here. Near the notch tip the displacements
are assumed to be given as
ur ¼ rm f ðθÞ
(7.114)
uθ ¼ rm gðθÞ

where f and g are arbitrary functions. The corresponding stresses will then behave like
rm1 near the notch tip so that we will seek solutions where 0 < m < 1, in which case
the stresses will go to infinity at the notch tip but will not be as singular as in a
concentrated load problem. Placing Eq. (7.114) into the equations of equilibrium
written in terms of the displacements (Navier’s equations in polar coordinates) we find
solutions of the form
ur ¼ rm fA1 cos ½ð1 þ mÞθ þ A2 sin ½ð1 þ mÞθ þ A3 cos ½ð1  mÞθ þ A4 sin ½ð1  mÞθg
uθ ¼ rm fA2 cos ½ð1 þ mÞθ  A1 sin ½ð1 þ mÞθ þ γA4 cos ½ð1  mÞθ  γA3 sin ½ð1  mÞθg
(7.115)

where
3ð1  νÞ
γ¼ (7.116)
3  5ν
in terms of Poisson’s ratio ν. These displacement expressions are then used to calculate the
stresses, which are placed into the boundary conditions:
σ rr ¼ σ θθ ¼ 0 θ ¼ þβ
(7.117)
σ rr ¼ σ θθ ¼ 0 θ ¼ β
216 Analytical Solutions

These are a set of homogeneous equations that have a nontrivial solution only if the
determinant of their coefficients is zero, a condition that gives the two conditions
sin ð2mβÞ sin ð2βÞ
¼∓ (7.118)
2mβ 2β

which are equations for the power m. However, an examination of the roots of these two
equations shows that the equation with the minus sign in Eq. (7.118) gives the most singular
stresses for 0 < m < 1so its roots will dominate the stress behavior for r small. Thus, we will
examine the roots of
sin ð2mβÞ sin ð2βÞ
¼ (7.119)
2mβ 2β

which we will write in terms of the opening angle, α, of the notch, where α ¼ 2π  2β, giving
sin ðm½2π  αÞ ¼ m sin ðαÞ (7.120)

A MATLAB® function stress_singularity was written to find the roots of Eq. (7.120) and
plot the results, which are given in Fig. 7.28. It can be seen that a sharp crack (α ¼ 0) has a

Figure 7.28 The behavior of the stress 1

singularity as a function of the opening angle 0.9
of a notch. 0.8
0.7

smallest 0.6
root, m 0.5
0.4
0.3
0.2
0.1
0
0 20 40 60 80 100 120 140 160 180
D, degrees

D =0 D = p/2
D=p
(crack) (right angle
reentrant corner) (nonsingular)

stresses ~ r − 0.5 stresses ~ r − 0.45 stresses ~ r 0

 7.5 Problems 217

singular behavior of r1=2 for the stresses and other cases also have a singular behavior as
long as α < π , i.e., they represent reentrant corners.
It may appear that such singular behavior is purely a mathematical consequence of the
highly idealized geometry, but such behavior also has an important physical significance for
the behavior of cracks. In fact, the field of fracture mechanics, which describes cracks and
their relationship to failure, has arisen from analysis of the stresses associated with such
singularities. We will describe some basic elements of fracture mechanics in Chapter 13.

7.5 PROBLEMS

P7.1 Figure P7.1 shows a hydraulic conduit composed of a steel liner in a concrete pipe. The
material properties of the steel and concrete are:

E steel ¼ 30  106 psi, E concrete ¼ 4  106 psi

νsteel ¼ 0:3, νconcrete ¼ 0:2
The inner radius of the concrete pipe is a and its outer radius is b. The thickness of the
steel liner is t, as shown. The pipe carries an internal pressure, p.
(a) If a ¼ 3 in., b ¼ 8 in., t ¼ 0:2 in:, p ¼ 100 psi, determine and plot the radial and
hoop stress in the steel liner and the concrete pipe, using the thick-wall pressure
vessel expressions.
(b) How much do your answers to part (a) change if you use the strength of materials
expressions for a thin-wall pressure vessel for the steel liner, where we neglect the
radial stress and assume the hoop stress is uniform across the liner?
P7.2 A large plate made of aluminum ðE Al ¼ 72 GPa, νAl ¼ 0:33Þ contains a circular hole of
radius, a (Fig. P7.2). A steel plug ðE s ¼ 194 GPa, νs ¼ 0:30Þ of radius a þ Δ is shrunk-
fit into this plate.
(a) Determine the stresses σ rr , σ θθ at the contact surface in terms of ða; Δ; E Al , νAl , E s , νs Þ.
[Hint: you can use the thick-wall pressure vessel expressions for either the inner
radius ¼ 0 or the outer radius ¼ ∞ to consider the plug or plate, respectively.]
(b) Determine the maximum allowable interference, Δ, if
the maximum allowable shear stress in the aluminum
plate is 50 MPa. Take a ¼ 40 mm.
P7.3 A cantilever rectangular beam of length, L, height, 2H, and
width, b, carries a uniform shear stress, τ 0 , on its top surface,
p as shown in Fig. P7.3.
a
b

t Figure P7.1 A steel liner of a concrete conduit where a uniform pressure, p,

acts within the liner.
218 Analytical Solutions

Figure P7.2 A large aluminum plate and an oversized

steel plug.
steel plug

a
a+'
aluminum plate

Figure P7.3 A cantilever beam where a uniform shear y

stress, τ 0 , acts on its top surface.
2H
τ0

beam width = b

(a) Determine the internal axial force, shear force, and bending moment in this beam
and the axial stress and shear stress as determined from a strength of materials
approach.
(b) Using the Airy stress function given by

ϕ ¼ axy þ bxy2 þ cxy3 þ dy2 þ ey3

determine the constants ða; b; c; d; eÞ so that the appropriate boundary conditions are
satisfied. How do your results differ from the strength of materials solution?
P7.4 Consider a triangular retaining wall subject to a hydrostatic pressure, p, from the
surrounding soil that varies linearly with depth, as shown in Fig. P7.4. Neglecting the
weight of the wall, consider an Airy stress function
a b c d
ϕ ¼ x3 þ x2 y þ xy2 þ y3
6 2 2 6
where ða; b; c; d Þ are constants.
(a) What are the exact boundary conditions on ϕ that must be satisfied on the
vertical wall?
(b) What are the exact boundary conditions on ϕ that must be satisfied on the inclined
wall? [Hint: consider the x- and y-components of the traction vector.]
(c) What are the Saint-Venant boundary conditions on the horizontal base?
(d) Determine the constants ða; b; c; d Þ that satisfy this problem and the corresponding
stresses.
 7.5 Problems 219

(e) Sketch out the normal and shear stresses at the base (y ¼ L). Without solving
explicitly for the strength of materials solution for this problem, comment on
whether the stresses you have sketched appear to be consistent with a strength of
materials solution.
P7.5 The rectangular concrete wall shown in Fig. P7.5 is subjected to a linearly varying
pressure, p ¼ ky on both of its faces. The weight of the concrete can be taken into
account by including the body force due to gravity, which can be written in terms of
potential function, V, given by V ¼ γy, where γ is the weight/unit volume of the
concrete. Using an Airy stress function

ϕ ¼ ax2 þ by2 þ cy3

(a) write down the exact boundary conditions that must be satisfied at the top and on
the sides of the wall.
(b) Using the boundary conditions from part (a), determine the constants ða; b; cÞin
terms of γ and k. What are the stresses at the base (at y ¼ L)?
(c) Because of symmetry, we expect that ux ¼ 0 at x ¼ 0. Also, at y ¼ L we expect
uy ¼ 0. Using these displacement conditions, determine the displacement compon-
ents in the wall in terms of ðγ; k; E; ν; LÞ, where E
x and ν are the Young’s modulus and Poisson’s ratio,
respectively, for the concrete. Assume the wall is in
b a state of plane strain.
x = y tan b p =g y
P7.6 Consider the Airy stress function ϕ ¼ Ar2 ð1 þ cos ð2θÞÞ.
L
What is the state of stress in Cartesian coordinates gener-
ated by this function? How does your answer change if
ϕ ¼ Ar2 ð1  cos ð2θÞÞ?
Figure P7.4 A retaining wall subjected to a linearly varying
y soil pressure.

p=k y p=k y
L

Figure P7.5 A concrete wall subjected to its own weight and linearly
y
varying pressures on its sides.
220 Analytical Solutions

P7.7 Consider a rigid disk of radius a that is bonded to a large elastic plate, as shown in
Fig. P7.6a, where the disk is subjected to a torque, T. The thickness of the plate is t.
(a) Using a free body diagram of a cut section of the plate of radius r, as shown
in Fig. P7.6b, and assuming the shear stress, τ, is uniform on the cut section,
determine the shear stress as a function of ðT; t; rÞ.
(b) From your results in part (a), what Michell type of Airy stress function (see
Eq. (7.61)) would be a solution for this problem?
P7.8 Consider a circular hole in a large plate, where the remote stress field is the biaxial field
shown in Fig. P7.7a.
(a) On a diagram similar to that seen in Fig. 7.13, show the extreme values of the
normalized hoop stress on the hole.
(b) Repeat part (a) for the biaxial stress field shown in Fig. P7.7b.
(c) Repeat part (a) for the biaxial stress field shown in Fig. P7.7c. What more can you
say about this case?
P7.9 Consider a circular hole in a large plate where the remote stress field is the pure shear
stress shown in Fig. P7.8a.
(a) On a diagram similar to that seen in Fig. 7.13, show the extreme values of the
normalized hoop stress and where they occur on the hole.
(b) Repeat part (a) when the remote stress field is the more general state of stress
shown in Fig. P7.8b.

Figure P7.6 (a) Rigid disk of

radius a embedded in a large
plate and subjected to a torque,
T. (b) A free body diagram of a T T
circular section of the plate of a a τ
radius r. r

(a) (b)

Figure P7.7 A hole in a plate under the V0 V0 V0

remote stress fields shown in (a), (b), and (c).
2V 0 2V 0 V0 V0 V0 V0

V0 V0 V0

(a) (b) (c)

 7.5 Problems 221

P7.10 The strength of materials flexure stress expression of Eq. (7.94) was for a curved beam
under a pure moment. In practice, curved beams will also have axial forces and shear
forces present. Consider, for example the curved member under a pair of forces, P, in
Fig. P7.9. The largest moment in the member will occur at section A–A, but at that
section there will also be an axial force, N, acting at the centroid of the cross-section.
It is easy to show that in this case the flexure stress in the curved beam is given by the
sum of extensional and bending contributions where
N M ðA  rAm Þ
σ θθ ¼ þ
A AðA  RAm Þr
If the curved member of Fig. P7.9 has a square 50  50 mm cross-section, find the
maximum tensile and compressive stresses acting on section A–A if P ¼ 5.9 kN. Plot
the total stress across the cross-section.
P7.11 Consider the beam in Fig. P7.10a where the applied load and the reaction forces are
all applied through rollers. In strength of materials it is normally assumed that these
forces are concentrated forces so that the shear force diagram experiences “jumps” at
the supports and the applied load as shown by the solid line in Fig. P7.10b. In reality,
these loads are spread over contact areas and the force changes are, therefore, spread
out over finite lengths.
(a) If the applied force P ¼ 4 kN, the width of the beam (and rollers) is 100 mm, the
diameter of the each of the two support rollers is 50 mm, and the diameter of the

s yy = 2s 0 Figure P7.8 (a) Hole in a plate under a remote

t stress that is a pure shear stress. (b) The case where
s yx = 3s 0
the remote stress is a more general 2-D state
t t s xy = 3s 0 of stress.
s xx = 10s 0

t
(a) (b)

P
A A
P

100 mm 30 mm Figure P7.9 A pair of equal and opposite forces applied to a curved member.
222 Analytical Solutions

roller under the load is 25 mm, estimate the finite lengths over which the shear
force diagram changes at the applied load and supports. Assume both rollers and
the beam are made out of steel (E ¼ 210 GPa, ν ¼ 0:29).
(b) In Fig. P7.10b, the dashed lines indicate the shear forces varying at the loads. Plot
the actual distribution of the shear force within the contact area at the left
support. [Hint: first find the load, P, in terms of the length a and the maximum
pressure, p0 , in the stress distribution.]
P7.12 Consider one of the Michell solutions for the Airy stress function, ϕ ¼ Crθ cos θ,
where C is a constant.
(a) Show that this function is the solution of a wedge-shaped plate of thickness, t,
with a concentrated force P at the tip of the wedge, as shown in Fig. P7.11, and
find explicitly the radial stress in the wedge.
(b) Determine the normal stress, σ xx , and the shear stress, σ xy , in polar coordinates
and then obtain approximate values for these stresses, written in terms of the
Cartesian coordinates ðx; yÞ, when the wedge half-angle, α, is very small.
Compare these stresses with the strength of materials expressions obtained by
considering this wedge as a beam whose height varies linearly in x. Where there
are discrepancies, comment on which expression you would trust more, and why.

(a) (b)

Figure P7.10 (a) Simply supported beam loaded and supported through rollers. (b) The idealized shear
force diagram for concentrated loads (solid line) and for loads distributed over the contact area between
the beam and the rollers.

y
Figure P7.11 A wedge-shaped T=α
plate loaded by a vertical force.
P r

T
x

T = -D
 References 223

References
1. J. R. Barber, Elasticity, 3rd edn (Switzerland: Springer, 2010)
2. A. P. Boresi and R. J. Schmidt, Advanced Mechanics of Materials, 6th edn (New York, NY: John
Wiley, 2002)
3. A. C. Ugural and S. K. Fenster, Advanced Mechanics of Materials and Applied Elasticity, 6th edn
(Boston, MA: Pearson, 2020)
4. M. H. Sadd, Elasticity – Theory, Applications, and Numerics (Burlington, MA: Elsevier, 2005)
5. S. N. Karp and F. C. Karal, “Elastic field behavior in the neighborhood of a crack of arbitrary
angle,” Comm. Pure and Appl. Math, 15, (1962), 413–421
6. F. C. Karal and S. N. Karp, “Stress behavior in the neighborhood of sharp corners,” Geophysics,
29, (1964), 360–369
8 Work–Energy Concepts

Work–energy concepts are important for two reasons. First, they provide an alternative way
to guarantee equilibrium and compatibility, which are two key elements in all stress
analyses. Second, energy methods have become the basis of formulating numerical methods,
so they are at the heart of the field of computational mechanics, which will be discussed in
Chapter 9. In this chapter we will discuss two types of internal energy in deformable bodies –
strain energy and complementary strain energy. Although we will show that these internal
energies are equal for linear elastic bodies, we will see that they play distinct roles in terms of
work–energy relations.

8.1 Work Concepts

You are likely familiar with the concept of work involving forces where the differential
work, dW, done by a force, F, taken through a small displacement, du, is just dW ¼ Fdu.
The work concept for a deformable body, which is discussed in the following section, follows
this same form except the forces can be distributed over areas in the form of stresses, or over
volumes in terms of body forces. These distributions can be expressed in various ways, which
make the work concept for a deformable body appear rather complex. This complexity,
however, will also lead to very useful and powerful tools, as we will see, for analyzing the
deformations and stresses present.

8.1.1 Work for a Deformable Body

Consider a deformable body occupying a volume V whose surface is S. If this body experi-
ences a small displacement, duðxÞ, the work, dW , done by the stress vector, TðnÞ ðxs Þ, acting
on the body surface and a body force, f ðxÞ, is given by (see Fig. 8.1):
ð ð
ðnÞ
dW ¼ T ðxs Þduðxs ÞdS þ f ðxÞduðxÞdV (8.1)
S V

As in Chapter 6, this stress vector may represent surface loads distributed over the surface as
well as line loads and concentrated forces and/or moments as special cases (see Fig. 6.1).
Concentrated forces and moments in particular play an important role in certain work–
energy theorems, and these will be discussed later. If we express the stress vector in terms of
the Cartesian stress components and correspondingly the displacement vector and body
force in terms of their Cartesian components, i.e.,

224
 8.1 Work Concepts 225

3 X
X 3
TðnÞ ¼ σ ij ni ej
i¼1 j¼1
X
3
f¼ f j ej (8.2)
j¼1
X3
du ¼ duj ej
j¼1

then we have
3 X
X 3
TðnÞ du ¼ σ ij ni duj
i¼1 j¼1
(8.3)
X
3
fdu ¼ f j duj
j¼1

which, when placed into Eq. (8.1) gives

X 3 X 3 ð  3 ð
X
dW ¼ σ ij duj ni dS þ f j duj dS (8.4)
i¼1 j¼1 j¼1
S V

However, the surface integral can be transformed into an integral over the volume, V, since
by Gauss’s theorem (for any well-behaved g, where g can be a scalar vector, or matrix)
ð ð
∂g
g ni dS ¼ dV (8.5)
∂xi
S V

so that we can rewrite Eq. (8.4) as

!
X3 ð X 3
∂  
dW ¼ σ ij duj þ f j duj dV
j¼1 i¼1
∂xi
V !
3 ð X 3 ð
(8.6)
X 3
∂σ ij X3 X
∂  
¼ þ f j duj dV þ σ ij duj dV
j¼1 i¼1
∂xi i¼1 j¼1
∂xi
V V

The first volume integral in Eq. (8.6) vanishes because the

x du ( x ) equilibrium equations must be satisfied, and the inte-
grand of the second integral can be written as
T(
n)
f (x) xs
( xs )
x2
V
e2
du ( x s )
e3 x1
e1
Figure 8.1 A deformable body acted upon by a body force and
x3 stress vector and small displacements generated by those forces.
226 Work–Energy Concepts

X3 X  
3
∂   X 3 X 3
∂uj
σ ij duj ¼ σ ij d
i¼1 j¼1
∂xi i¼1 j¼1
∂xi
X
3 X
3  
¼ σ ij deij þ dωij (8.7)
i¼1 j¼1
3 X
X 3
¼ σ ij deij
i¼1 j¼1

where the sum of the terms involving the change of the local rotation, dωij , vanishes because
the stresses are symmetric while the rotations are antisymmetric. Thus, Eq. (8.6) becomes
ð X !
3 X 3
dW ¼ σ ij deij dV (8.8)
i¼1 j¼1
V

8.2 Work–Strain Energy

The work done by all the external forces acting on an elastic body is equal to the change of
the internal potential energy, U, of the body, an energy of stored deformation, also called the
strain energy. We can write this work–energy principle in differential form as
ð ð
dW ¼ dU ¼ d u0 dV ¼ ðdu0 ÞdV (8.9)
V V

where u0 is the strain energy per unit volume, i.e., the strain energy density. Comparing Eq.
(8.9) and Eq. (8.8), we see that the change in this strain energy density is given as
X
3 X
3
du0 ¼ σ ij deij (8.10)
i¼1 j¼1

If we write the stresses in terms of the strains, then we can express the strain energy density as
a function of the strains only, which we write symbolically as u0 ¼ u0 ðeÞ, and from which it
follows that
X
3 X
3
∂u0 ðeÞ
du0 ðeÞ ¼ deij (8.11)
i¼1 j¼1
∂eij

so that comparing Eq. (8.11) and Eq. (8.10) we find

∂u0 ðeÞ
σ ij ¼ ði; j ¼ 1; 2, 3Þ (8.12)
∂eij

Thus, just as a conservative force can be written in terms of the spatial gradient of a potential
energy, the stresses in an elastic body that has a strain energy density function can be written
 8.2 Work–Strain Energy 227

as the strain gradient of that strain energy density. It is sometimes also convenient to write
the stresses as components of a stress vector, σ I ðI ¼ 1; 2; . . . , 6Þ, and the strains as com-
ponents of a strain vector, eI ðI ¼ 1; 2; . . . , 6Þ. In that case the increase in the strain energy
density can be written as
X
6
du0 ¼ σ I deI (8.13)
I ¼1

and the stress vector as

∂u0 ðeÞ
σI ¼ ðI ¼ 1; 2; . . . , 6Þ (8.14)
∂eI

8.2.1 Linear Elastic Material

Equations (8.12) and (8.14) show that if we can obtain the strain energy density for a
material, then we can use that strain energy density to obtain the stress–strain relations. In
general, those stress–strain relations will be nonlinear. However, if we assume that the
strains in the material are small, it makes sense to expand the strain energy density in a
power series. Keeping only quadratic terms at most we have
X
6
1X 6 X 6
u0 ðeÞ ¼ A þ BI e I þ C IJ eI eJ (8.15)
I ¼1
2 I ¼1 J¼1

If, when the strains are all zero the strain energy density is zero and the stresses are also zero,
then A ¼ BI ¼ 0 and the strain energy density is expressible entirely in terms of the 36 elastic
constants, C IJ . However, from the second derivatives we find

∂2 u0 ðeÞ ∂2 u0 ðeÞ
CI J ¼ ¼ ¼ C JI (8.16)
∂eI ∂eJ ∂eJ ∂eI
Thus, the matrix of elastic constants is symmetric and there are only 21 constants for the
most general linear elastic material where
X
6
σI ¼ C IJ eJ ðI ¼ 1; 2; . . . , 6Þ (8.17)
J¼1

which, we recall, can also be written in terms the C ijkl constants as

X
3 X
3
σ ij ¼ C ijkl ekl ði; j ¼ 1; 2, 3Þ (8.18)
k¼1 l¼1

Note that from Eq. (8.15) the strain energy density for a linear elastic material can be
written as
228 Work–Energy Concepts

1X 6 X 6
u0 ¼ C IJ eI eJ (8.19a)
2 I ¼1 J¼1

or, using Eq. (8.17)

1X 6
u0 ¼ σ I eI (8.19b)
2 I ¼1

If we write the strain in Eq. (8.19b) in terms of the stress through the compliance matrix, DIJ ,
then we also have

1X 6 X 6
u0 ¼ DIJ σ I σ J (8.19c)
2 I ¼1 J¼1

In the first form in Eq. (8.19a) we see that we have written the strain energy density in terms
of the strains only, i.e., u0 ¼ u0 ðeÞ, while in the second form in Eq. (8.19b) the strain energy
density is being written as a function of both stresses and strains, i.e., u0 ¼ u0 ðσ; eÞ. In the
third form of Eq. (8.19c) only stresses are involved so u0 ¼ u0 ðσ Þ. We can also write these
forms in terms of C ijkl elastic constants and Dijkl elastic compliances as

1X 3 X 3 X 3 X 3
u0 ¼ C ijkl eij ekl (8.20a)
2 i¼1 j¼1 k¼1 l¼1

1X 3 X 3
u0 ¼ σ ij eij (8.20b)
2 i¼1 j¼1

1X 3 X 3 X 3 X 3
u0 ¼ Dijkl σ ij σ kl (8.20c)
2 i¼1 j¼1 k¼1 l¼1

It is useful to graphically illustrate these strain energy concepts for the simple case of a
state of uniaxial stress. In this case, Fig. 8.2a shows the stress–strain curve for a general
elastic material where the stress σ xx is plotted versus the strain exx . The differential of the
strain energy density is shown in that figure as well as the total strain energy density,
which is just the area under the stress–strain curve. Figure 8.2b shows the special case of
a linear elastic material where, in the uniaxial case, σ xx ¼ E exx and E is Young’s
modulus. In that case, as shown, the total strain energy density is the shaded triangular
area under the stress–strain curve, which can be written in the equivalent forms
u0 ¼ Ee2xx =2 ¼ σ 2xx =2E ¼ σ xx exx =2 depending on whether the strain energy density is
written in terms of the strain, the stress, or in terms of both stress and strain. The factor
of one half in these expressions arises from integrating the differential of the strain energy
density from zero to the final stress or strain state when the stress–strain law is linear. For
Ð Ð Ð
example, we have u0 ¼ du0 ¼ σ 0xx de0xx ¼ E e0xx de0xx ¼ Ee2xx =2, where exx is the final
strain. In this case u0 ¼ u0 ðeÞ but for u0 ¼ u0 ðσ Þ a similar integration yields u0 ¼ σ 2xx =2E.
 8.2 Work–Strain Energy 229

Figure 8.2 (a) A one-dimensional (1-D) stress–strain

curve, showing the graphical interpretation of the
V xx V xx differential of the strain energy density, du0 . (b) The
V xxc special case of a linear stress–strain law. In both cases
E the total strain energy density, u0 , is the shaded area
decxx 1 under the stress–strain curve from zero to a final state
of stress and strain of ðσ xx , exx Þ.
exx exx
du0 V xxc decxx

(a) (b)

The mixed form of u0 ðσ; eÞ ¼ σ xx exx =2 also arises directly from the area formula for the
shaded triangular area in Fig. 8.2b.
So far, we have considered the work–strain energy principle in terms of differential
changes of the strain energy and differential work (see Eq. (8.1)). For a linearly elastic
material we can write a similar work–energy relationship in terms of total strain energy and
total work terms. To see this, use Eq. (8.20b) and write the total strain energy as
ð 3 X
1 X 3
U¼ σ ij eij dV (8.21)
2 i¼1 j¼1
V

Using the symmetry of the stress matrix, Gauss’s theorem, and the equilibrium equations we
can rewrite the strain energy as
ð 3 X
1 X 3
∂uj
U¼ σ ij dV
2 i¼1 j¼1 ∂xi
V
ð 3 X ð 3 X
1 X 3
∂   1 X 3
∂σ ij
¼ σ ij uj dV  uj dV
2 i¼1 j¼1 ∂xi 2 i¼1 j¼1 ∂xi
V V
ðX
3 X
3 ðX
3
1 1
¼ σ ij ni uj dS þ f j uj dV (8.22)
2 i¼1 j¼1
2 j¼1
S V
ðX
3 ðX
3
1 ðnÞ 1
¼ T j uj dS þ f j uj dV
2 j¼1
2 j¼1
S V
ð ð
1 1
¼ TðnÞ u dS þ fu dV
2 2
S V

which says that the total strain energy stored in the material is equal to the total work done
by the stress vector and the body force acting on the body. This result is called Clapeyron’s
theorem. The factor of one half is present again because in loading the elastic body from its
230 Work–Energy Concepts

undeformed state to one where the displacement vector is u, the stresses are proportional to
the strains in the loading process.

8.2.2 Isotropic Case

As seen in Chapter 5, if a linear elastic material is also isotropic then the material coefficients
can be expressed in terms of only two elastic constants such as Young’s modulus, E, and
Poisson’s ratio, ν, where
E ð1  νÞ
C 11 ¼ C 22 ¼ C 33 ¼
ð1 þ νÞð1  2νÞ
Eν
C 12 ¼ C 21 ¼ C 13 ¼ C 31 ¼ C 23 ¼ C 32 ¼
ð1 þ νÞð1  2νÞ (8.23)
1 E
C 44 ¼ C 55 ¼ C 66 ¼ ðC 11  C 12 Þ ¼ G
2 2ð1 þ νÞ
all others ¼ 0

Placing these values into Eq. (8.19a)) gives

1 E h    i
u0 ð eÞ ¼ ð1  νÞ e2xx þ e2yy þ e2zz þ 2ν exx eyy þ exx ezz þ eyy ezz
2 ð1 þ νÞð1  2νÞ
(8.24)
1  
þ G γ2xy þ γ2xz þ γ2yz
2
or, in terms of principal strains,
1 E h    i
u0 ðeÞ ¼ ð1  νÞ e2p1 þ e2p2 þ e2p3 þ 2ν ep1 ep2 þ ep1 ep3 þ ep2 ep3 (8.25)
2 ð1 þ νÞð1  2νÞ
which can also be written in terms of the strain invariants as
1 E ð1  νÞ
u0 ðeÞ ¼ J 2  2GJ 2 (8.26)
2 ð1 þ νÞð1  2νÞ 1

where (see Eq. (4.36))

J 1 ¼ ep1 þ ep2 þ ep3
(8.27)
J 2 ¼ ep1 ep2 þ ep1 ep3 þ ep2 ep3

Another way to express the strain energy density is in terms of the strain matrix components,
eij . First rewrite Eq. (8.24) as
G h    
u0 ðeÞ ¼ ð1  νÞ e2xx þ e2yy þ e2zz þ 2ν exx eyy þ exx ezz þ eyy ezz
ð1  2νÞ
 i (8.28a)
þ 2ð1  2νÞ exy þ exz þ eyz
2 2 2
 8.2 Work–Strain Energy 231

and rearrange the terms as

G h  
u0 ðeÞ ¼ ð1  2νÞ e2xx þ e2yy þ e2zz þ 2e2xy þ 2e2xz þ 2e2yz
ð1  2νÞ
 i (8.28b)
þν e2xx þ e2yy þ e2zz þ 2exx eyy þ 2exx ezz þ 2eyy ezz

which then can be expressed finally in the compact form

2 !2 3
X 3 X 3
ν X3
u 0 ð eÞ ¼ G 4 eij eij þ eii 5 (8.29)
i¼1 j¼1
ð1  2νÞ i¼1

If we write the strains in terms of the stresses, then we can also express the strain energy
density solely in terms of the stresses, i.e., u0 ¼ u0 ðσ Þ. The end result is
1 h 2  i
u0 ðσ Þ ¼ σ xx þ σ 2yy þ σ 2zz  2ν σ xx σ yy þ σ xx σ zz þ σ yy σ zz
2E
(8.30)
1  2 
þ σ þ σ xz þ σ yz
2 2
2G xy
In terms of the principal stresses we have
1 h 2  i
u0 ðσ Þ ¼ σ p1 þ σ 2p2 þ σ 2p3  2ν σ p1 σ p2 þ σ p1 σ p3 þ σ p2 σ p3 (8.31)
2E
or, for the stress invariants I 1 , I 2 ,
1  2 1 2 1
u0 ðσ Þ ¼ I 1  2ð1 þ νÞI 2 ¼ I1  I2 (8.32)
2E 2E 2G
where (see Eq. (2.61))
I 1 ¼ σ p1 þ σ p2 þ σ p3
(8.33)
I 2 ¼ σ p1 σ p2 þ σ p1 σ p3 þ σ p2 σ p3

We can also express the strain energy in terms of the stress matrix components, σ ij . Starting
from Eq. (8.30), we can rewrite that equation as
1 h 2  
u0 ðσ Þ ¼ σ xx þ σ 2yy þ σ 2zz þ 2ð1 þ νÞ σ 2xy þ σ 2xz þ σ 2yz
2E
(8.34)
 i
2ν σ xx σ yy þ σ xx σ zz þ σ yy σ zz

and then rearrange it to obtain

1 h  
u0 ðσ Þ ¼ ð1 þ νÞ σ 2xx þ σ 2yy þ σ 2zz þ 2σ 2xy þ 2σ 2xz þ 2σ 2yz
2E
 i (8.35)
 ν σ 2xx þ σ 2yy þ σ 2zz þ 2σ xx σ yy þ 2σ xx σ zz þ 2σ yy σ zz
232 Work–Energy Concepts

which then can be written finally in matrix form as

2 !2 3
1 4 X3 X 3 X3
u0 ðσ Þ ¼ ð1 þ νÞ σ ij σ ij  ν σ ii 5 (8.36)
2E i¼1 j¼1 i¼1

8.2.3 Distortional Strain Energy

Strain energy is a very useful quantity when discussing failure concepts since intuitively we
expect a material will fail if we require it to store too much energy. (See Chapter 13 for a
discussion of failure.) For failure with respect to slip (yielding), it has been observed experi-
mentally that such failure cannot be produced by a hydrostatic state of stress, i.e., where we
apply a pure pressure to the material. Since the strain energy in general contains a hydro-
static energy component, we first need to remove that hydrostatic component of the strain
energy before we can use that energy in a failure theory for slip. To accomplish this removal,
note that, for a state of stress consisting of a pure pressure, p, we have σ p1 ¼ σ p2 ¼ σ p3 ¼ p
which produces from Eq. (8.31) a strain energy density
3ð1  2νÞ 2
u0 ¼ up ¼ p (8.37)
2E
For a general state of stress, if we replace the pressure, p, by the average of the three
principal stresses, i.e.,
σ p1 þ σ p2 þ σ p3
p¼ (8.38)
3
then we can define the hydrostatic component as
ð1  2νÞ  2
up ¼ σ p1 þ σ p2 þ σ p3 (8.39)
6E
and we can also define the distortional strain energy density, ud , as
ud ¼ u0  up (8.40)

Using Eq. (8.31) and Eq. (8.39), after some algebra we find
ð1 þ νÞ h 2  2  2 i
ud ¼ σ p1  σ p2 þ σ p1  σ p3 þ σ p2  σ p3 (8.41)
6E
If you recall, in Chapter 2 we discussed the stresses acting on a particular plane called the
octahedral plane. There we found an explicit expression for the magnitude of the total shear
stress acting on the octahedral plane, ðτ s Þoct , Eq. (2.82), which shows that the distortional
strain energy density seen in Eq. (8.41) is just
3ð1 þ νÞ 2
ud ¼ ðτ s Þoct (8.42)
2E
 8.3 Complementary Strain Energy 233

so that a failure theory based on the distortional strain energy density becoming too large
can also be viewed as a theory where the total shear stress on the octahedral plane has to be
greater than a particular stress value for the material to fail [1] (see Chapter 13).

8.3 Complementary Strain Energy

In applying work–energy principles, we will see that a quantity called the complementary
strain energy density, uc0 , also appears, where this density is defined as
X
6
uc0 ¼ σ I eI  u0 (8.43a)
I ¼1

or, equivalently,
3 X
X 3
uc0 ¼ σ ij eij  u0 (8.43b)
i¼1 j¼1

If we consider the one-dimensional (1-D) state of stress again then, as Fig. 8.3a shows, the
complementary strain energy density is the shaded area shown above the stress–strain curve
relating σ xx to exx . A small change of this complementary strain energy density is therefore
given by the chain rule of differentiation as
X
6 X
6
duc0 ¼ eI dσ I þ σ I deI  du0
I ¼1 I ¼1
(8.44a)
X
6
¼ eI dσ I
I ¼1

where we have used Eq. (8.13). This change is also shown graphically in Fig. 8.3a for the 1-D
state of stress case. We can also write this relationship in terms of the stress and strain
matrices as

Figure 8.3 (a) A 1-D stress–strain curve, showing the

graphical interpretation of the differential of the
s xx s xx complementary strain energy density, duc0 . (b) The
ds xx' special case of a linear stress–strain law. In both cases
E the total complementary strain energy density, uc0 , is
the shaded area above the stress–strain curve from
1
zero to a final state of stress and strain of ðσ xx , exx Þ.
e'xx exx exx
du0c = e'xx ds 'xx

(a) (b)
234 Work–Energy Concepts

3 X
X 3
duc0 ¼ eij dσ ij (8.44b)
i¼1 j¼1

If we now express the strains as a function of the stresses so that the complementary strain
energy density is a function of the stresses only, we have
X
6
∂uc
duc0 ðσ Þ ¼ 0
dσ I (8.45)
I ¼1
∂σ I

Comparing Eq. (8.45) with Eq. (8.44a) we find

∂uc0 ðσ Þ
eI ¼ ðI ¼ 1; 2; . . . , 6Þ (8.46a)
∂σ I
or, equivalently, we can obtain
∂uc0 ðσ Þ
eij ¼ (8.46b)
∂σ ij

Comparing Eq. (8.46a) with Eq. (8.14) or comparing Eq. (8.46 b) with Eq. (8.12), it follows
that u0 and uc0 play complementary roles where the stresses and strains are concerned – hence
the name complementary energy density for uc0 .

8.3.1 Linear Elastic Material

If we have a linear elastic material, then it follows from Eq. (8.19b) and Eq. (8.43a) that
uc0 ¼ u0 , where in the “mixed” form for these densities we have

1X 6
u0 ðσ; eÞ ¼ uc0 ðσ; eÞ ¼ σ I eI (8.47)
2 I ¼1

All the other expressions obtained earlier for the strain energy density are also valid for the
complementary strain energy density as well. In the 1-D case (see Fig. 8.3b) it is easy to see
that these densities are identical since the area below the linear stress–strain “curve” is equal
to the area above that “curve.”

8.4 Strain Energy for Strength of Materials Problems

In the previous sections we have discussed the concept of strain energy for the general case of
a deformable body. In this section, we will obtain explicit total strain energy expressions for
the simple problems of axial extension/compression, bending, and torsion treated in an
introductory strength of materials course. In all these cases the material is assumed to have
linear elastic behavior.
 8.4 Strain Energy for Strength of Materials Problems 235

8.4.1 Axial Loads

In the case of the uniaxial loading of a bar (see Fig. 8.4a), the stress, strain, and stress–strain
relations are
F x ðxÞ
σ xx ¼
A
dux (8.48)
exx ¼
dx
σ xx ¼ Eexx

where F x is the internal axial load, ux is the axial displacement, and A and E are the cross-
sectional area and Young’s modulus, respectively, of the bar. As discussed in Chapter 3, the
equilibrium equation for the bar is
dF x
¼ qx (8.49)
dx
where qx ðxÞ is a distributed load/unit length acting on the bar. In terms of the displacement,
the equilibrium equation becomes

qx Figure 8.4 (a) An axial-load

geometry showing a distributed
A
force/unit length, qx ðxÞ, acting on
the bar and the internal force, F x .
(b) Bending of a beam showing the
x Fx ( x ) distributed vertical force/unit
length, qz ðxÞ acting along the beam
L x and the internal bending moment,
(a) M ¼ M y , and shear force,
qz V ¼ V z , acting in the beam. (c)
z
y Torsion of a circular shaft showing
M ( x ) = −M y ( x ) a distributed torque/unit length,
tx ðxÞ, acting along the shaft and
x the internal twisting moment
V ( x ) = Vz ( x ) (torque), M x .
L
x
tx (b)
A

x
M x ( x)
L x
(c)
236 Work–Energy Concepts

 
d dux
EA ¼ qx (8.50)
dx dx

Since there is only one stress the total strain energy of the bar, U, is
ðL
1 σ 2xx
U ¼ Adx
2 E
0
(8.51)
ðL 2
1 ðF x Þ
¼ dx
2 AE
0

where L is the length of the bar. Similarly, in terms of the strain or displacement
ðL
1
U ¼ Ee2xx Adx
2
0
(8.52)
ðL 2
1 dux
¼ EA dx
2 dx
0

which also follows directly from Eq. (8.51) and the force–displacement relation F x ¼
EAdux =dx for the bar.

8.4.2 Bending
In engineering beam theory, the normal stress, normal strain, and normal stress–strain
relation for a beam are (see Fig. 8.4b):
M ðxÞz
σ xx ¼ 
I
M ðxÞz d 2w (8.53)
exx ¼  ¼ 2z
EI dx
σ xx ¼ Eexx

where M ¼ M y is the internal bending moment about the negative y-axis, I ¼ I yy is

the area moment of the cross-section about the y-axis, and w is the displacement of the
neutral axis of the beam in the z-direction. In Eq. (8.53) we have also used the moment–
curvature relationship

d 2w
M ðxÞ ¼ EI κðxÞ ¼ EI (8.54)
dx2
 8.4 Strain Energy for Strength of Materials Problems 237

where the curvature of the beam, κ, is just the second derivative of the displacement, w, when
the slope of the beam is small, as assumed in engineering beam theory. Force and moment
equilibrium equations for the beam are
dM
¼ V
dx
(8.55)
dV
¼ qz
dx
where V ¼ V z is the internal shear force in the beam (see Fig. 8.4b) and qz ðxÞ is a distributed
load/unit length acting in the z-direction. The equilibrium equation for the moment can be
written directly in terms of the distributed load by eliminating the shear force, giving

d 2M
¼ qz (8.56)
dx2
or, using the moment–curvature relation,
!
d2 d 2w
EI 2 ¼ qz (8.57)
dx2 dx

The total strain energy due to bending for a bar of length, L, is

ðL ð
1 σ 2xx
U¼ dAdx
2 E
0 A
8 9
ð
ðL
1 < 2 = M2
¼ z dA dx (8.58)
2 : ; EI 2
0 A

ðL
1 M2
¼ dx
2 EI
0

or, using the moment–curvature relation in Eq. (8.58), the strain energy in terms of the
displacement of the neutral axis is
ðL !2
1 d 2w
U¼ EI dx (8.59)
2 dx2
0

There is also internal strain energy associated with the shear stress in the beam. In engineer-
ing beam theory, this shear stress is τ ¼ σ xz given by
V ðxÞQðzÞ
τ¼ (8.60)
I tðzÞ
238 Work–Energy Concepts

so that
ðL ð
1 τ2
U ¼ dAdx
2 G
0 A
8 9
ðL< ð 2 = V2
1 A Q
¼ dA dx (8.61)
2 :I 2 t2 ; GA
0 A

ðL
1 V2
¼ k dx
2 GA
0

where the dimensionless constant, k, is

ð
A Q2
k¼ dA (8.62)
I 2 t2
A

This constant is a function of the geometry of the cross-section only. For example, for a
rectangular cross-section k ¼ 1.2. In bending problems involving long and slender beams,
the strain energy associated with the bending (flexural) stress is considerably larger than this
strain energy due to the shear stress. Thus, in those cases the strain energy of the shear
stresses can be neglected.

8.4.3 Torsion
The shear stress, shear strain, and stress–strain relations for the torsion of a circular shaft are
given by
M x ðxÞr
τ¼
J
dϕðxÞ (8.63)
γ¼ r
dx
τ ¼ Gγ

where τ is the total shear stress in the shaft (see Eq. (1.38)) and M x is the internal moment
about the x-axis (torque) (see Fig. 8.4c). The total shear strain, γ, given in Eq. (8.63) follows
from the torque–twist relation given by (see Eq. (1.34))
dϕðxÞ
M x ðxÞ ¼ GJ (8.64)
dx
and Hooke’s law, where ϕðxÞ is the twist of the cross-section about the x-axis of the shaft,
and J and G are the polar area moment of the cross-section and the shear modulus. The
equation of equilibrium for the shaft is (see Eq. (3.49))
 8.5 Virtual Work and Minimum Potential Energy 239

dM x
¼ tx (8.65)
dx
where tx ðxÞ is the distributed torque/unit length acting along the shaft. In terms of the twist,
the equilibrium equation becomes
 
d dϕ
GJ ¼ tx (8.66)
dx dx

Using these results, the strain energy of the shaft in terms of the torque is
ðL ð
1 τ2
U ¼ dAdx
2 G
0 A
8 9
ððL
1 < 2 = M 2x
¼ r dA dx (8.67)
2 : ; GJ 2
0 A

ðL
1 M 2x
¼ dx
2 GJ
0

or, in terms of the twist, using the torque–twist relation in Eq. (8.67)
ðL  2
1 dϕ
U¼ GJ dx (8.68)
2 dx
0

8.5 Principle of Virtual Work and Minimum Potential Energy

Satisfying the governing differential equilibrium equations is only one way in which we can
ensure equilibrium is satisfied in a deformable body. In this section we will see that an
alternative approach is through work–energy principles and the introduction of the concept
of virtual work and potential energy.

8.5.1 General Deformable Body

Assume the equilibrium equations for a deformable body are satisfied. If we multiply those
equations by a set of weighting functions, δuj ðj ¼ 1; 2, 3Þ, which represent small changes of
“virtual” displacements (i.e., not the actual displacements), uj ðj ¼ 1; 2, 3Þ, of the body and
integrate over the volume, V, of the body we obtain
ðX !
3 X 3
∂σ ij
þ f j δuj dV ¼ 0 (8.69)
i¼1 j¼1
∂xi
V
240 Work–Energy Concepts

which we can rewrite as

ðX 3 X ðX3 X ðX
3
∂   3
∂δuj 3
σ ij δuj dV  σ ij dV þ f j δuj dV ¼ 0 (8.70)
i¼1 j¼1
∂xi i¼1 j¼1
∂xi i¼1
V V V

We can use Gauss’s theorem on the first term in Eq. (8.70) and interchange the real
 
differential changes with the virtual changes, i.e., ∂ðδui Þ=∂xj ¼ δ ∂ui =∂xj , to obtain
ðX 3 X 3 ðX3 X 3   ðX3
∂uj
σ ij ni δuj dS  σ ij δ dV þ f j δuj dV ¼ 0 (8.71a)
i¼1 j¼1 i¼1 j¼1
∂xi j¼1
S V V

which we can also write as

ðX 3 X 3 ðX
3 ðX
3
ðnÞ
σ ij δeij dV ¼ T j δuj dS þ f j δuj dV (8.71b)
i¼1 j¼1 j¼1 j¼1
V S V

where we have introduced the stress vector and used the symmetry of the stress matrix to
write the left-hand side of Eq. (8.71b) in terms of virtual strain changes. However, recall
from Eq. (8.12) when we write the strain energy density in terms of the strains only, we have
∂u0 ðeÞ
σ ij ðeÞ ¼ ði; j ¼ 1; 2, 3Þ (8.72)
∂eij

and so
ðX3 X 3 ðX3 X 3
∂u0 ðeÞ
σ ij ðeÞδeij dV ¼ δeij dV
i¼1 j¼1 i¼1 j¼1
∂eij
V V
ð (8.73)
¼ δu0 ðeÞdV ¼ δU ðeÞ
V

where U ðeÞ is the total strain energy when written as a function of the strains only and δU ðeÞ
is the virtual change of this strain energy due to the virtual displacements δuj (and corres-
ponding virtual strains). This leads to the principle of virtual work, which is
ðX 3 ðX 3
ðnÞ
δU ðeÞ ¼ T j δuj dV þ f j δuj dV
S
j¼1
V
j¼1 (8.74)
¼ δW v

where the right-hand side of Eq. (8.74) is the virtual work, δW v , done by the stress vector
and body force during the virtual displacements, δuj (compare to Eq. (8.1) where we have
actual displacements and work terms).Thus, we have shown that, if equilibrium is satisfied
throughout the body, then the principle of virtual work is satisfied. This virtual work
principle can also be written compactly as
 8.5 Virtual Work and Minimum Potential Energy 241

ð ð
δU ðeÞ ¼ TðnÞ δu dV þ fδu dV (8.75)
S V

We can simply reverse the steps just outlined and go from Eq. (8.75) to Eq. (8.69). If we
assume the principle virtual work is satisfied for all virtual displacements then the δu
appearing in Eq. (8.69) is arbitrary, which in turn requires that the integrand of Eq. (8.69)
itself must be zero, i.e., the equilibrium equations must be satisfied at every point in the
body. In summary, we can say that satisfying equilibrium for all points in a deformable body
and satisfying the principle of virtual work for all virtual displacements are equivalent, i.e.,

equilibrium ) principle of virtual work

principle of virtual work ) equilibrium

As we will see, using this equivalence can give us an alternate way to satisfy the governing
equations of equilibrium either exactly, or approximately.
One can make an even stronger connection between the principal of virtual work and the
solution to the problem where the displacements are specified as known values uj ¼ u∗ j on a
∗
portion of the surface, S u , and the stresses are specified as known values σ ji nj ¼ T i on the
remaining portion, S t , where the total surface S ¼ S u þ S t . Namely, one can show the following.

If a displacement field, uj , is found such that (1) it satisfies the prescribed boundary
displacement values uj ¼ uj ∗ on S u and where δuj ¼ 0 on S u , and (2) if the principle of
virtual work in the form
ð ð
δU ðeÞ ¼ T∗ δu dV þ f ∗ δu dV (8.76)
St V

is satisfied for all δuj which vanish on S u then uj satisfies both equilibrium and the
P3 ∗
stress boundary conditions j¼1 σ ji nj ¼ T i on S t , i.e., it is the solution to the
governing equations and boundary conditions.

In Eq. (8.76), ðT∗ , f ∗ Þ are the prescribed stress-vector values and body force values in the
problem. We will not prove the result stated above, but the proof follows the same steps
needed when using Eq. (8.75) to show that equilibrium is satisfied, or you can find the details
in [2]. Virtual displacements that satisfy δuj ¼ 0 on S u are said to satisfy the "essential"
boundary conditions. In contrast, the boundary conditions σ ji nj ¼ T i ∗ on S t are said to be
the “natural” boundary conditions. These terms come from considering these problems with
the techniques of variational calculus [3].
In using the principle of virtual work, Eq. (8.75), it is often useful to also consider virtual
displacements where δuj ¼ 0 on S u , since otherwise Eq. (8.75) will include an unknown
242 Work–Energy Concepts

“reaction” stress vector, RðnÞ , on S u corresponding to the stresses that must act to enforce the
displacement boundary conditions there. By choosing virtual displacements that vanish on
S u , this unknown stress vector can be eliminated when using Eq. (8.75). However, these
reaction stresses can be found once the problem is solved for the stresses in the body.
Now, let us introduce the potential energy of the deformable body, Π, defined as
ðX 3 ðX 3
ðnÞ
Π ¼ U ðeÞ  T j uj dS  f j uj dV (8.77)
j¼1 j¼1
S V

ðnÞ
where T j and f j are the actual stress vector and body force while uj is an arbitrary
displacement field and the strains in U ðeÞ are obtained from that displacement. For a linear
elastic material, we will show the following.

 
The minimum value of the potential energy, Π uj , is attained for the true displace-
ment field, uj .

This is called the theorem of minimum potential energy. To prove this theorem let uj be
the true displacement field and uj þ δuj be any other displacement. Also, let ðeI ; δeI Þ be the
vectors of strains associated with uj , δuj , respectively. Then, using the integral of the strain
energy density for a linear elastic material given in Eq. (8.19a) to represent the total strain
energy, we have
ðX 6 X ðX ðX
    6 3
ðnÞ
3
Π uj þ δuj  Π uj ¼ σ I δeI dV  T j δuj dS  f j δuj dV
I ¼1 J¼1 j¼1 j¼1
V S V
ð 6 X (8.78)
1 X 6
þ C IJ δeI δeJ dV
2 I ¼1 J¼1
V

where σ I is the vector of stresses due to the true displacement, uj . But from the principal of
virtual work the sum of the first three terms on the right-hand side of Eq. (8.78) vanish.
Thus, we find
ð 6 X
    1 X 6
Π uj þ δuj  Π uj ¼ C IJ δeI δeJ dV (8.79)
2 I ¼1 J¼1
V

This difference is positive and Π is indeed a minimum if the matrix of elastic constants, C IJ ,
are positive definite so that the right-hand side of Eq. (8.79) is the corresponding positive
definite strain energy due to the displacement δuj . From the expression we obtained earlier
for the total strain energy of an isotropic linear elastic material written as a function of the
strains (see Eq. (8.29)) we see that the total strain energy is indeed a positive definite quantity
in that case, i.e., it is a positive quantity and zero only if all the strains and/or stresses are
 8.5 Virtual Work and Minimum Potential Energy 243

identically zero. If the material is not linearly elastic then the theorem of minimum potential
energy is still true as long as the matrix of second derivatives of the strain energy,
∂2 U=∂eI ∂eJ , is positive definite but we will not show that generalization here.
The positive definiteness of the strain energy also allows us to show that solution to the
typical boundary value problem for linear elastic deformable bodies (consisting of the
governing equations and boundary conditions) is unique. From Eq. (8.22) we found that
as long as equilibrium is satisfied for a symmetrical stress matrix
ð ð
1 1
U ð eÞ ¼ TðnÞ u dS þ fu dV (8.80)
2 2
S V
 
for the displacement, u. Now, assume there are two solutions uð1Þ , uð2Þ to the same problem
 
ð1Þ ð2Þ
and the corresponding stresses are σ ij ; σ ij . Then for a linear elastic material the dis-
ð1Þ ð2Þ
placement u ¼ uð1Þ  uð2Þ and stress σ ij ¼ σ ij  σ ij are also solutions to the governing
equations with zero body forces. Using those solutions in Eq. (8.80) gives
ð
1
U ðeÞ ¼ TðnÞ u dS (8.81)
2
S

If the stress vector is specified over the entire surface S then, since both solutions must satisfy
the same stress boundary conditions, we have TðnÞ ¼ 0 so the total strain energy must be zero.
Similarly, if the stress vector was specified over part of the surface, S t , and the displacement
specified over the remainder of the surface, S u , we would have TðnÞ ¼ 0 on S t and u ¼ 0 on S u ,
so that the strain energy would be zero again. Obviously, if the displacement is specified over
the entire surface then the strain energy is also zero. In the more general case, which we
discussed in Chapter 6, as long as one of the components of either the stress vector or the
displacement in the product TðnÞ u are specified on all or part of the surface, the surface
integral in Eq. (8.80) will be zero so the strain energy will be zero. For all these typical
boundary conditions U ðeÞ ¼ 0 so if the total strain energy is positive definite the strains must
ð1Þ ð2Þ
satisfy eij ¼ eij  eij ¼ 0 and the strains in the two solutions will be identical. The stress–
strain relations then also require that the stresses in the two solutions be identical. If there are
no displacement boundary conditions, the vanishing of the strains means that the displace-
ments in the two solutions can differ at most by a rigid body displacement. If there are
displacement boundary conditions, then typically these rigid body displacements must vanish
and so the displacements in the two solutions are also identical.

Example 8.1 Virtual Work and a Truss

As a simple example of how the virtual work principle can be used as a replacement for
satisfying equilibrium directly, consider the two-bar truss shown in Fig. 8.5a subjected to
 
the loads Px , Py . The two bars, labeled (1) and (2), are assumed to have the same cross-
sectional areas, A, the same lengths, L, and the same Young’s modulus, E, and are pinned at
244 Work–Energy Concepts

Figure 8.5 (a) A two-bar truss. (b) The free body x

diagram of the forces at pin C.

A B
y
T T
A, E, L
V1 A V2A
(1) (2) T T
C Px
Py
Px
Py
(a) (b)

points A, B, and C. Since each of the bars is under uniaxial loads, the stresses ðσ 1 , σ 2 Þ in bars (1)
and (2), respectively, are constants, as are the corresponding axial strains, ðe1 , e2 Þ. This is a
statically determinant problem that we can solve for the stresses by simply examining force
equilibrium at pin C, as shown in Fig. 8.5b. From equilibrium in the x- and y-directions we find
þ X
! Fx ¼ 0

σ 1 A cos θ þ σ 2 A cos θ þ Px ¼ 0
þ X
(8.82)
# Fy ¼ 0

σ 1 A sin θ  σ 2 A sin θ þ Py ¼ 0

which are two equations for the two unknown stresses ðσ 1 , σ 2 Þ. We can write these equilib-
rium equations in matrix form as
" #
A cos θ A cos θ σ1 Px
¼ (8.83)
A sin θ A sin θ σ 2 Py

Now, let us consider the use of virtual work. We can view the three pins A, B, and C as the
“boundary” of the structure, where pins A and B satisfy zero displacements and pin C has
the loads applied, so that pins A and B act as the surface S u of a continuous body where the
displacements are specified and pin C acts as the surface S t where the stress vector is
applied. The two bars themselves act as the volume V for the continuous body. The
displacements of the two bars can be described by the pin displacements as shown in
Fig. 8.6a, and the forces acting on the structure can be described by the applied loads
     
Px , Py and the reaction forces Ax , Ay and Bx , By . Since the strains are constants for
 8.5 Virtual Work and Minimum Potential Energy 245

Figure 8.6 (a) Displacements of the truss defined at

A u1 B u3 A Ax B Bx
pins A, B, and C. (b) The applied forces and
v1 T T v3
Ay T T By reaction forces.

u2 Px
C C
v2 Py

(a) (b)

the two bars, they are just equal to the bar elongations divided by their lengths. Thus, the
strains can be obtained directly from the pin displacements in Fig. 8.6a as
ðu2  u1 Þ cos θ þ ðv2  v1 Þ sin θ
e1 ¼
L
(8.84)
ðu3  u2 Þ cos θ þ ðv2  v3 Þ sin θ
e2 ¼
L
The real displacements, of course, satisfy the displacement boundary conditions u1 ¼ v1 ¼ 0
and u3 ¼ v3 ¼ 0 but the virtual displacements do not have to satisfy those conditions. If we
choose virtual displacements that do not satisfy the displacement boundary conditions, then
the reaction forces at A and B will do virtual work as well as the applied loads. If our focus is
on obtaining the unknown stresses ðσ 1 , σ 2 Þ, these reaction forces are not needed so we can
eliminate them in the principle of virtual work by choosing virtual displacements ðδu2 , δv2 Þ
with δu1 ¼ δv1 ¼ 0 and δu3 ¼ δv3 ¼ 0. The corresponding virtual strains will then be
δu2 cos θ þ δv2 sin θ
δe1 ¼
L
(8.85)
δu2 cos θ þ δv2 sin θ
δe2
L
We can place these strains into the principle of virtual work but it is useful to examine the
separate contributions in Eq. (8.85) from δu2 and δv2 . Consider the virtual displacement δv2
as shown in Fig. 8.7a. This will produce an elongation δv2 sin θ in bar (2) as shown in
Fig. 8.7b so the strain in that bar will be δe2 ¼ δv2 sin θ=L and by symmetry this will also be
the strain in bar (1). From the principle of virtual work, we have
δU ¼ δW v
ð ð
σ 1 δe1 dV þ σ 2 δe2 dV ¼ Py δv2
barð1Þ barð2Þ
    (8.86)
δv2 sin θ δv2 sin θ
σ1 AL þ σ 2 AL ¼ Py δv2
L L
 
σ 1 A sin θ þ σ 2 A sin θ  Py δv2 ¼ 0
246 Work–Energy Concepts

Figure 8.7 (a) Virtual displacement of A B B

the truss that satisfies the displacement q q
boundary conditions at A and B. (b)
The deformation of bar (2), showing the (1) (2) d v2 sin d
elongation and corresponding strain. By
L d e2 =
L
symmetry bar (1) has the same strain. d e1 = d e2
q

Px d v2 d v2
d v2 sin q

Py

(a) (b)

which, since the virtual displacement δv2 is arbitrary, just gives the second equilibrium
equation in Eq. (8.83). In entirely the same fashion for the virtual displacement δu2 as
shown in Fig. 8.8a we have (see Figs. 8.8b, c)
δU ¼ δW v
ð ð
σ 1 δe1 dV þ σ 2 δe2 dV ¼ Px δu2
barð1Þ barð2Þ
    (8.87)
δu2 cos θ δu2 cos θ
σ1 AL þ σ 2 AL ¼ Px δu2
L L
ðσ 1 A cos θ  σ 2 A cos θ  Px Þδu2 ¼ 0

which, since δu2 is arbitrary, gives the first equilibrium equation in Eq. (8.83).
Now consider what happens when the virtual displacements do not satisfy the displace-
ment boundary conditions. In that case the virtual strains and displacements are related
through the strain–displacement relations of Eq. (8.84), and the principle of virtual work,
following similar steps as before, gives
 
ðδu2  δu1 Þ cos θ þ ðδv2  δv1 Þ sin θ
σ1 AL
L
 
ðδu3  δu2 Þ cos θ þ ðδv2  δv3 Þ sin θ (8.88)
þσ 2 AL
L
¼ Ax δu1 þ Ay δv1 þ Px δu2 þ Py δv2 þ Bx δu3 þ By δv3

Collecting terms that multiply the same virtual displacements and using the fact that the
virtual displacements are arbitrary so the coefficients of those virtual displacements must
vanish, we arrive at a set of equations which can be placed in matrix–vector form for the
unknown reactions and stresses in the bars as
 8.5 Virtual Work and Minimum Potential Energy 247

A B A Figure 8.8 (a) Virtual displacement that

T T satisfies the displacement boundary
conditions at A and B. (b) The
G u2 cos T
(2) G e1 deformation of bar (1) showing the
(1) L
elongation and corresponding strain. (c)
L
G u2 cos T The deformation of bar (2) showing the
Px shortening and corresponding strain.
G u2
Py T
G u2

(a) (b)

B
T

L G u2 cos T
G e2
G u2 cos T T L

G u2

(c)

2 38 9 8 9
1 0 A cos θ 0 0 0 > > Ax >> >
> 0 >>
>
> >
> >
> >
6 0 1
6 A sin θ 0 0 0 7>
7> A y>> >
> 0 >>
>
6 0 < = < =
6 0 A cos θ A cos θ 0 0 7
7 σ 1 Px
6 0 ¼ (8.89)
0 A sin θ A sin θ 0 0 7
7>
> σ2 > > Py >
6 > > > >
> >
>
4 0 0 0 A cos θ 1 0 5>>
> B >
> >
> 0 >>
: ; : >
x> > ;
0 0 0 A sin θ 0 1 By 0

The first equation in Eq. (8.89) is just the equilibrium equation in the x-direction at pin
A while the second equation is the equilibrium equation at pin A in the y-direction.
Similarly, the last two equations in Eq. (8.89) are the equilibrium equations at pin B in
the x- and y-directions, respectively. The middle two equations in Eq. (8.89) are the same
equilibrium equations derived previously at pin C in terms of the applied loads. We have
left these middle equations in terms of the stresses to make the connection more explicit
to the principle of virtual work as originally derived, but in practice it is better to write
these equations in terms of the internal forces, which will also make the equilibrium
matrix in Eq. (8.89) dimensionally homogeneous. Equation (8.89) represents six equa-
tions in six unknowns, which can be solved so that the problem remains statically
determinant even when we include the unknown reaction forces. But if we solve for the
stresses from the equilibrium equations at pin C, we can always go back and find the
248 Work–Energy Concepts

reaction forces at pins A and B from those known stresses so we see that choosing virtual
displacements that satisfy the displacement boundary conditions simplifies the problem
greatly. In problem P8.1, you are asked to use the symbolic algebra capabilities of
MATLAB® to solve the system of equations in Eq. (8.89). You should find that the
solution can be written compactly as:
1  1 
Ax ¼  Px þ Py cot θ , Bx ¼ Py cot θ  Px
2 2
1  1 
Ay ¼  Py þ Px tan θ , By ¼ Px tan θ  Py (8.90)
2 2
   
1 Px Py 1 Py Px
σ1 ¼ þ , σ2 ¼ 
2A cos θ sin θ 2A sin θ cos θ

We could have also written the strain energy entirely in terms of the displacements through
the use of the stress–strain and strain–displacement relations. Since the displacements satisfy
the displacement boundary conditions, the stresses then are
u2 cos θ þ v2 sin θ
σ1 ¼ E
L
(8.91)
u2 cos θ þ v2 sin θ
σ2 ¼ E
L
so that the principle of virtual work, when it is written entirely in terms of displacements,
gives the equilibrium equations as (compare with the stress equations of Eq. (8.83)):
2 3
2EA
6 L cos θ
2
0 7 u2 Px
4 2EA 5 ¼ (8.92)
v2 Py
0 sin θ
2
L
where the matrix of coefficients is called a stiffness matrix for the truss. Because this
matrix is diagonal, the displacements are obtained directly and then the stresses can be
obtained from Eq. (8.91). This truss problem is statically determinant so there is little
difference if we solve for the stresses as the fundamental unknowns or the displacements.
But in statically indeterminate problems, the equilibrium equations for the stresses are
insufficient by themselves to solve for the stresses and we must combine the equilibrium
and compatibility equations to reach a solution, as seen in Navier’s table problem in
Chapter 6. However, if we use the principle of virtual work written directly in terms of
the displacements, even for statically indeterminate problems we can generate a stiffness
matrix for the structure that defines a set of equations for the displacements that can be
solved for those displacements and then the stresses back-calculated. We will examine
both these approaches in more detail in Chapter 9.
 8.6 Complementary Virtual Work & Minimum Complementary Potential Energy 249

8.6 Principle of Complementary Virtual Work and Minimum

Complementary Potential Energy
The complementary strain energy also plays an important role for deformable bodies. Just
as the principle of virtual work can act as a substitute to the equilibrium equations, the
principle of complementary virtual work can replace the strain–displacement relations and
hence also acts a substitute for the compatibility equations.

8.6.1 General Deformable Body

Assume that the strain–displacement relations are satisfied. If we multiply those relations by
a symmetrical weighting matrix, δσ ij ði; j ¼ 1; 2, 3Þ, which represents small virtual changes of
the stresses σ ij , and integrates over the volume V of a deformable body we have
ðX 3 X 3   
1 ∂ui ∂uj
eij  þ δσ ij dV ¼ 0 (8.93)
i¼1 j¼1
2 ∂xj ∂xi
V

which, by virtue of the symmetry of the virtual stresses (δσ ij ¼ δσ ji ), we can write as
ðX 3 X 3  
∂ui
eij  δσ ij dV ¼ 0 (8.94)
i¼1 j¼1
∂xj
V

and which can be rearranged as

ðX 3 X ðX3 X ðX3 X
3 3
∂   3
∂δσ ij
eij δσ ij dV  ui δσ ij dV þ ui dV ¼ 0 (8.95)
i¼1 j¼1 i¼1 j¼1
∂xj i¼1 j¼1
∂xj
V V V

Using Gauss’s theorem, the second terms in Eq. (8.95) can be placed on the boundary and
that equation can be expressed as
ðX 3 X 3 ðX
3 ðX
3
ðnÞ
eij δσ ij dV  δT i ui dS  δf i ui dV ¼ 0 (8.96)
i¼1 j¼1 i¼1 i¼1
V S V

where

ðnÞ
X
3
δT i ¼ δσ ij nj
j¼1
(8.97)
X
3
∂δσ ij
δf i ¼ 
j¼1
∂xj
250 Work–Energy Concepts

are the virtual stress vector and virtual body force generated by the virtual stresses.
However, recall the relationship between the strain and the complementary strain energy,
Eq. (8.46 b), from which it follows
ðX 3 X ðX 3 X
3   3
∂uc0
eij σ ij δσ ij dV ¼ δσ ij dV
i¼1 j¼1 i¼1 j¼1
∂σ ij
V V
ð (8.98)
¼ δuc0 ðσ ÞdV ¼ δU c ðσ Þ
V

where U c is the total complementary strain energy of the body. Thus, we obtain the principle
of complementary virtual work:
ðX 3 ðX 3
ðnÞ
δU ¼
c
δT i ui dS þ δf i ui dV
S
i¼1
V
i¼1 (8.99)

¼ δW cv

where δW cv is the differential complementary virtual work. This result can also be written
more compactly as
ð ð
δU c ¼ δTðnÞ udS þ δfudV (8.100)
S V

We have shown that if the strain–displacement relationship (and hence compatibility) is

satisfied throughout the body, then the principle of complementary virtual work must be
satisfied. We can also reverse the steps in going from the strain–displacement relation to
the principle of complementary virtual work and show that if the principle of comple-
mentary virtual work is satisfied for all stresses σ ij whose virtual changes, δσ ij , satisfy
Eq. (8.97) then the strain that is generated by those stresses will satisfy the strain–
displacement relations and hence be compatible strains. Again, we can summarize these
results by saying that satisfying compatibility and satisfying the principle of complemen-
tary virtual work are equivalent, i.e.,

compatibility ) principle of complementary virtual work

principle of complementary virtual work ) compatibility

As in the virtual work case, one can make a stronger statement about what the principle of
complementary virtual work implies for the problem when the stress vector is specified on S t
and the displacement is specified on S u . That statement is as follows.
 8.6 Complementary Virtual Work & Minimum Complementary Potential Energy 251

Let a symmetric state of stress, σ ij , be found such that

X
3
σ ij ni ¼ T ∗
j on S t
i¼1

X
3
∂σ ij
þf∗
j ¼ 0 in V
i¼1
∂xi
∗
(where T ∗
j , f j are specified stress vector and body force components) and such that for
all symmetrical δσ ij satisfying
X
3
δσ ij nj ¼ 0 on S t
i¼1

X
3
∂δσ ij
¼0 in V
i¼1
∂xi

the principle of complementary virtual work holds, which in this case is

ð
δU ðσ Þ ¼ δTðnÞ udS
c

Su

(where, since the stress vector is specified on S t , the integral over that part of the surface
vanishes and, since the body force is also specified, we have δf j ¼ 0). Then the stress σ ij
satisfies the displacement boundary conditions ui ¼ u∗ i on S u and the strain–displacement
relations, i.e., it is the solution to the governing equations and boundary conditions.

If we define the complementary potential energy, Πc , as

ð ð
Πc ¼ U c ðσ Þ  TðnÞ udS  fudV (8.101)
S V

(where u is the true displacement and TðnÞ , f are the stress vector and body force correspond-
ing to any stress field, σ ij ) and follow similar steps shown for the minimum potential energy
theorem we can also prove that, for a linear elastic material the principle of minimum
complementary potential energy is true, as follows.

 
The minimum value of the complementary potential energy, Πc σ ij , is attained for the
true stress field, σ ij .
252 Work–Energy Concepts

Again, if the material is not linearly elastic but the matrix of second derivatives of the
complementary strain energy, ∂2 U c =∂σ I ∂σ J , is positive definite the theorem of minimum
complementary energy is still true. We will not prove the two complementary energy
theorems described above, but you can find details in [2] and [4].

Example 8.2 Complementary Strain Energy and a Truss

To illustrate the application of the principle of complementary strain energy consider a truss
again, but where now the truss has three members so that it is statically indeterminate
(Fig. 8.9a). There are still only two equilibrium equations at pin C given from the free body
diagram of Fig. 8.9b by
þ X
! Fx ¼ 0

σ 1 A cos θ þ σ 2 A cos θ þ Px ¼ 0
þ X (8.102)
# Fy ¼ 0

σ 3 A  σ 1 A sin θ  σ 2 A sin θ þ Py ¼ 0
which we also could get from the principle of virtual work. In matrix–vector form these
equations in terms of stresses are
8 9
< σ 1 =
cos θ  cos θ 0 Px =A
σ ¼ (8.103)
sin θ sin θ 1 : 2 ; Py =A
σ3
Choosing virtual stresses δσ j that also satisfy these equilibrium equations, we have
8 9
< δσ 1 =
cos θ  cos θ 0 δPx =A
δσ ¼ (8.104)
sin θ sin θ 1 : 2 ; δPy =A
δσ 3

Figure 8.9 (a) Statically indeterminate truss where x

all the members have the same cross-sectional area,
A, and Young’s modulus, E, and bars (1) and (2)
have lengths L (not shown) while bar (3) has a length y
L sin θ. (b) The free body diagram of the forces at A D B
pin C.
T T
V3A
(3) V1 A V2A
(1) (2) T T
Px
C Py
Px
Py
(a) (b)
 8.6 Complementary Virtual Work & Minimum Complementary Potential Energy 253

and the principle of complementary virtual work gives

δU c ¼ δW cv
ð ð ð
e1 δσ 1 dV þ e2 δσ 2 dV þ e3 δσ 3 dV ¼ u2 δPx þ v2 δPy
barð1Þ barð2Þ barð3Þ (8.105)
e1 δσ 1 AL þ e2 δσ 2 AL þ e3 δσ 3 AL sin θ ¼ u2 δPx þ v2 δPy
¼ u2 ðA cos θδσ 1  A cos θδσ 2 Þ þ v2 ðA sin θδσ 1 þ A sin θδσ 2 þ Aδσ 3 Þ
where we have used Eq. (8.104). Since the δσ j are arbitrary, we have
e1 L ¼ Δ1 ¼ u2 cos θ þ v2 sin θ
e2 L ¼ Δ2 ¼ u2 cos θ þ v2 sin θ (8.106)
e3 L sin θ ¼ Δ3 ¼ v2
in terms of the strains, ei , or, equivalently, the elongations, Δi , which we can also write in
matrix–vector form as
8 9 2 3
< Δ1 = cos θ sin θ
u
Δ ¼ 4  cos θ sin θ 5 2 (8.107)
: 2; v2
Δ3 0 1

Thus, we have shown that the principle of complementary virtual work has given us the
generalized strain–displacement relations of Eq. (8.107). We recognize the matrix in
Eq. (8.107) as just the transpose of the equilibrium matrix in Eq. (8.104), following the
same relationships seen in Chapter 6 for Navier’s table problem.
Because there are three elongations (or strains) written in terms of two displacements, these
elongations are related by a compatibility equation which is easy to obtain in this case since
adding the first two equations in Eq. (8.106) and comparing the result to the third equation gives
Δ1 þ Δ2 ¼ 2Δ3 sin θ (8.108)
or, in terms of the strains,
e1 þ e2 ¼ 2e3 sin 2 θ (8.109)
which we can multiply by Young’s modulus to obtain the compatibility equation in terms of
the stresses as
σ 1 þ σ 2  2σ 3 sin 2 θ ¼ 0 (8.110)
Combining the equilibrium and compatibility equations (Eq. (8.103) and Eq. (8.110)), we
then have three equations for the three stresses given as
2 38 9 8 9
cos θ  cos θ 0 < σ 1 = < Px =A =
4 sin θ sin θ 1 5 σ 1 ¼ Py =A (8.111)
: ; : ;
1 1 2 sin θ
2
σ3 0
254 Work–Energy Concepts

which can be solved directly. This is a stress-based (or force-based) solution to the truss
problem. To obtain a displacement-based solution, we will first write the equilibrium equa-
tions in terms of the internal forces F j ¼ σ j A as
8 9
< F 1 =
cos θ  cos θ 0 Px
F ¼ (8.112)
sin θ sin θ 1 : 2 ; Py
F3

which we will write symbolically as


E~ fF g ¼ f P g (8.113)

in terms of the equilibrium matrix E ~ , the internal force vector fF g ¼ ½F 1 , F 2 , F 3 T , and the
 T
applied load vector fPg ¼ Px , Py . Then we can express the strain–displacement equa-
tions in terms of the forces and relate them to the displacements as
 T
fF g ¼ ½C fΔg ¼ ½C  E ~ f ug (8.114)

where fug ¼ ½u2 , v2 T and

2 3
EA=L 0 0
½C  ¼ 4 0 EA=L 0 5 (8.115)
0 0 EA=L sin θ

Placing Eq. (8.114) into the equilibrium equations, Eq. (8.113), we then find the displace-
ment equations
½K fug ¼ fPg (8.116)
  T
in terms of the pseudostiffness matrix ½K  ¼ E ~ ½C  E
~ . We can symbolically solve either
the combination of Eq. (8.116) and Eq. (8.114) or Eq. (8.111) with MATLAB® (see problem
P8.2) for the stresses or forces in the truss. The result for the internal forces is
Px Py sin 2 θ
F1 ¼ þ
2 cos θ 2 sin 3 θ þ 1
Py sin 2 θ Px
F2 ¼  (8.117)
2 sin 3 θ þ 1 2 cos θ
Py
F3 ¼
2 sin 3 θ þ 1

8.7 Work–Energy Principles and Discrete Forces and Moments

Consider an elastic body which carries a set of surface loads, some of which are distributed
and some of which are discrete (concentrated forces and moments), and also body forces.
 8.7 Work–Energy Principles and Discrete Forces and Moments 255

Mi Figure 8.10 A deformable body acted upon by concentrated forces

Pi and moments. There may also be distributed forces or body forces
(body forces not shown) acting on the body.

Ti
ui Let the displacement at the concentrated load, Pi (i ¼ 1,2,. . .,
M1 N), in the direction of that force be given by ui . Similarly, let
the rotation at the concentrated moment, M i (i ¼ 1, 2,. . ., N), in
P1 the direction determined by that moment from the right-hand
rule be given by θi (see Fig. 8.10). We will assume that the strain
energy of this body is expressed in terms of these displacements
and rotations as U ¼ U ðu1 ; . . . , uN , θ1 , . . . , θN Þ. Note that the
strain energy may also depend on other variables but, unlike
the displacements and rotations at the concentrated loads and moments, these other variables
will not be changed so we will not show them explicitly.

8.7.1 Virtual Work and Potential Energy

Now, imagine that we fix the concentrated forces and moments and make small changes to
the displacements and rotations, δui , δθi . These changes, in turn, will cause the strain energy
to change by a small amount, δU. Then the principle of virtual work says that the work done
by the forces and moments taken through these virtual displacements must be equal to the
resulting virtual change in the strain energy, i.e.,
X
N X
M
δU ¼ Pδui þ M i δθi (8.118)
i¼1 i¼1

Similarly, in terms of the principle of minimum potential energy, where the potential energy
is defined here as
X
N X
M
Π¼U Pi u i  M i θi (8.119)
i¼1 i¼1

since the potential energy must be stationary, we have

X
N X
M
δΠ ¼ δU  Pi δui  M i δθi ¼ 0 (8.120)
i¼1 i¼1

which again gives the principle of virtual work.

Castigliano’s First Theorem

Since the strain energy has been assumed to be written as a function of the displacements and
rotations at the concentrated loads and moments, it follows that
256 Work–Energy Concepts

X
N
∂U X
M
∂U
δU ¼ δui þ δθi (8.121)
i¼1
∂ui i¼1
∂θi

so that placing Eq. (8.121) into either Eq. (8.118) or Eq. (8.120) gives
XN   XM  
∂U ∂U
 Pi δui þ  M i δθi ¼ 0 (8.122)
i¼1
∂ui i¼1
∂θi

which can only be satisfied for all possible virtual displacements and rotations if
∂U
¼ Pi ði ¼ 1; . . . ; N Þ
∂ui
(8.123)
∂U
¼ Mi ði ¼ 1; . . . ; M Þ
∂θi
which are a set of relations called Castigliano’s first theorem.

Example 8.3 Castigliano’s First Theorem

As a simple example of the use of Castigliano’s first theorem, consider an axial-load problem
where two bars are welded together between two rigid walls and are subjected to a concen-
trated force, P, at their connection (see Fig. 8.11). Let uP be the displacement at the load P.
The internal forces, stresses, and strains in the bars AB and BC are all constants so we
assume the tensile strain in bar AB is e1 ¼ uP =L1 and the compressive strain in bar BC is
e2 ¼ uP =L2 . Thus, from Eq. (8.52) the strain energy of the entire system is
1 A1 E 1 2 1 A2 E 2 2
U¼ u þ u (8.124)
2 L1 P 2 L 2 P
Castigliano’s first theorem gives
 
∂U A1 E 1 A2 E 2
P¼ ¼ þ uP (8.125)
∂uP L1 L2

which can be solved for the displacement as

P
uP ¼   (8.126)
A1 E 1 A2 E 2
þ
L1 L2

from which we can obtain the strains, stresses, and internal forces in both sections of the bar.
It can also be verified by solving this problem directly that Eq. (8.126) is the exact result.

uP
A1 , E1 A2 , E2 Figure 8.11 An assembly of two bars acted upon by a concentrated load
P P at B where the two bars are welded together. The ends A and C are
A B C rigidly fixed. The cross-sectional area, Young's modulus and length of
bar AB and bar BC are ðA1 , E 1 , L1 Þ and ðA2 , E 2 , L2 Þ, respectively. The
L1 L2
displacement at B in the direction of the load P is uP .
 8.7 Work–Energy Principles and Discrete Forces and Moments 257

The Rayleigh–Ritz Method

In the previous example, we obtained an exact solution because the assumed deformation field
that we used to calculate the strain energy coincided with the exact values. In more compli-
cated problems, it may not be a simple matter to obtain exact deformation expressions.
Nevertheless, if we can make a “reasonable” guess at the form of the deformations in terms
of some simple functions and unknown coefficients, then the principle of minimum potential
energy (or virtual work) gives us a method to determine those unknown coefficients in such a
manner that equilibrium will be approximately satisfied. This method is called the Rayleigh–
Ritz method. The procedure of the method is to assume that the displacements of the body can
be represented parametrically in terms of a set of basis functions and n unknown parameters
ða1 , a2 ; . . . , an Þ. For example, in a 1-D problem we might take the displacement in the form
ux ¼ a0 þ a1 x þ    þ an1 xn1 (8.127)

In the Rayleigh–Ritz method, it is assumed that the functions and constants are chosen so
that any boundary conditions involving the displacements or rotations of the body (which
are called the “essential” boundary conditions) are satisfied. Then the functions and coeffi-
cients are placed into the expression for the potential energy of the body so that this potential
energy can be expressed in the form
Π ¼ Πða0 ; . . . , an1 Þ (8.128)

The potential energy is made stationary by requiring that

X
n1
∂Π
δΠ ¼ δai ¼ 0 (8.129)
i¼0
∂ai

which can only be satisfied for arbitrary changes of these coefficients if

∂Π
¼0 ði ¼ 0; . . . , an1 Þ (8.130)
∂ai
which are n equations to be solved for the n coefficients.

Example 8.4 The Rayleigh–Ritz Method

As an example of the Rayleigh–Ritz method, consider a cantilever beam loaded by an end
moment, as shown in Fig. 8.12. The exact solution for the vertical deflection, wðxÞ, of the
neutral axis of the beam is
M 0 x2
wðxÞ ¼ (8.131)
2EI
so that the end deflection is given by
M 0 L2
wð LÞ ¼ (8.132)
2EI
258 Work–Energy Concepts

Figure 8.12 A cantilever beam z

acted upon by an end moment, w(L)
M 0 , and the displacement of the M0
end at x ¼ L, w(L).
x
L

To obtain an approximate solution to this problem, let us assume

h  πx i
wðxÞ ¼ a 1  cos (8.133)
2L
This function meets the requirements stated previously; i.e., it satisfies the essential condi-
tions of zero displacement and slope (where, for small slopes, the slope is just the rotation of
the neutral axis) at x ¼ 0. The total potential energy is:
Π ¼ U  M 0 θðLÞ
ðL 2 !
EI d w dwðLÞ (8.134a)
¼ 2
dx  M 0
2 dx dx
0

Placing the approximate deflection expression into Eq. (8.134a) and carrying out the
indicated differentiations and integrations, we find
π2 EIa2 πM 0 a
Π¼  (8.134b)
64L3 2L
Requiring ∂Π=∂a ¼ 0 yields a ¼ 16M 0 L2 =π 3 EI and, since a ¼ wðLÞ we obtain
0:52M 0 L2
wðLÞ ¼ (8.135)
EI
which is very close to the exact solution of Eq. (8.132).

Although the Rayleigh–Ritz method is a very useful tool, for complicated 3-D problems it
is impossible to make good guesses for what the deformations might be for the entire body,
i.e., to choose global functions of approximations. However, suppose the body is broken into
small elements over which locally the deformations can be reasonably assumed to have
simple variations, and where these variations are written in terms of unknown parameters
(called nodal variables). Then the principle of minimum potential energy (or virtual work)
can be used to form up a set of linear equations for all these nodal variables (in the example
just shown there was one nodal variable a ¼ wðLÞ). Solving this linear system yields an
approximate solution for the deformation of the entire body. This is the basic idea behind
the finite element method, which will be discussed in more depth in Chapter 9.
 8.7 Work–Energy Principles and Discrete Forces and Moments 259

8.7.2 Complementary Virtual Work and Complementary Potential Energy

Consider the body shown in Fig. 8.10 again, where the concentrated forces and moments
were shown together with their corresponding displacements and rotations. Now, imagine
these displacements and rotations are held fixed while we vary the applied concentrated
forces and moments by small virtual amounts ðδPi , δM i Þ, where these virtual changes
satisfy the equations of equilibrium. For a statically determinate elastic system, we can
write the complementary strain energy solely in terms of these applied loads as U c ¼
U c ðP1 ; . . . , PN , M 1 , . . . , M M Þ so that virtual changes in the applied loads will cause the
complementary strain energy to change by an amount δU c . The principle of comple-
mentary virtual work states that
X
N X
M
δU c ¼ ui δPi þ θi δM i (8.136)
i¼1 i¼1

For example, Fig. 8.13 shows a specific example of a statically determinate truss structure
loaded in two dimensions where the uniaxial forces in all the truss members can be found
from equilibrium in terms of P1 so that the complementary strain energy can be written
explicitly in terms of P1 , and the principle of complementary virtual work gives
δU c ¼ u1 δP1 . The reaction forces at supports A and B are also present but they do no
complementary virtual work since the corresponding displacements in the directions of these
forces are all zero. If we define a complementary energy, Πc , for the system as
X
N X
M
Πc ¼ U c  ui Pi  θi M i (8.137)
i¼1 i¼1

then the principle of minimum complementary potential energy requires that the comple-
mentary energy be stationary, i.e.,
X
N X
M
δΠc ¼ δU c  ui δPi  θi δM i ¼ 0 (8.138)
i¼1 i¼1

Engesser’s First Theorem and Castigliano’s

D C u1 Second Theorem
Since, as stated earlier, the deformable body is assumed to be a
P1 statically determinate problem where the complementary
strain energy can be written explicitly in terms of the applied
forces and moments, it follows that the change in complemen-
tary strain energy can also be expressed as
A B

Ax Figure 8.13 A statically determinate truss where the forces in the bars of
Ay By
the truss can all be written in terms of the applied force, P1 .
260 Work–Energy Concepts

X
N
∂U c X
M
∂U c
δU c ¼ δPi þ δM i (8.139)
i¼1
∂Pi i¼1
∂M i

so that Eq. (8.136) or Eq. (8.138) give

XN   XM  
∂U c ∂U c
 ui δPi þ  θi δM i ¼ 0 (8.140)
i¼1
∂Pi i¼1
∂M i

If the virtual changes in the applied loads are varied independently, the only way for Eq.
(8.140) to be satisfied is if
∂U c
¼ ui ði ¼ 1; . . . ; N Þ
∂Pi
(8.141)
∂U c
¼ θi ði ¼ 1; . . . ; M Þ
∂M i
a result that is called Engesser’s first theorem. For a linear elastic material U c ¼ U, so that
∂U
¼ ui ði ¼ 1; . . . ; N Þ
∂Pi
(8.142)
∂U
¼ θi ði ¼ 1; . . . ; M Þ
∂M i
which is usually called Castigliano’s second theorem. Although we have assumed that the
system under question is statically determinate when deriving these results, they are also true
for statically indeterminate structures as we will show at the end of Section 8.7.3.

Example 8.5 Castigliano’s Second Theorem

As a simple example of the use of Castigliano’s second theorem, consider a cantilever beam
subjected to an end force, P, as shown in Fig. 8.14. The bending moment in the beam is
given by M ðxÞ ¼ Px so that the strain energy is
ðL
1 ½M ðxÞ2
U ¼ dx
2 EI
0 (8.143)
2 3
1P L
¼
6 EI
Castigliano’s second theorem says that the deflection at the load P in the direction of P, uP ,
is given by
∂U 1 PL3
uP ¼ ¼ (8.144)
∂P 3 EI
which can be verified independently by integrating the moment–curvature relation.
 8.7 Work–Energy Principles and Discrete Forces and Moments 261

As this example shows, we can obtain the deflection (or rotation) at the location of any
applied force (or moment). However, we can use the concept of a dummy load to obtain the
deflection (or rotation) at any point in the body. To see this, consider the cantilever beam of
Fig. 8.14 again where now we want to obtain the deflection at the center of the beam. In
order to use Castigliano’s second theorem we place a dummy force, Q, at the center, as
shown in Fig. 8.15. The bending moment in the beam is now
Px ð0 < x < L=2Þ
M ðxÞ ¼ (8.145)
Px  Qðx  L=2Þ ðL=2 < x < LÞ

so that the strain energy is

U ¼ U ðP; QÞ
8 9
>
< ð
L=2 ðL >
=
1 2
(8.146)
¼ P x dx þ ½Px þ Qðx  L=2Þ dx
2 2
2EI >
: >
;
0 L=2

From Castigliano’s second theorem we have

ðL
∂U ðP; QÞ 1
uQ ¼ ¼ ½Px þ Qðx  L=2Þðx  L=2Þdx (8.147)
∂Q EI
L=2

This is the deflection at Q due to both P and Q but we want the deflection at Q due to P only,
which we can obtain from Eq. (8.147) by simply setting Q ¼ 0 to find
ðL
∂U ðP; QÞ 1
uQ Q¼0 ¼ ¼ ðPxÞðx  L=2Þdx
∂Q Q¼0 EI
L=2 (8.148)
5 PL3
¼
48 EI
P which is the desired answer. With the use of a dummy moment
x (couple) instead of a dummy force, the same procedure would have
allowed us to obtain the local rotation at any point in the beam.

DP
L Figure 8.14 A cantilever beam acted upon by an end force, P, and the
deflection at P in the direction of the force, ΔP .

Q P
x

Figure 8.15 Use of a dummy load, Q, to obtain the deflection at the center of a
L/2 L/2 cantilever beam.
262 Work–Energy Concepts

There is an alternate way of viewing the dummy load method which leads to what is called
the unit-load method. Consider, for example, a bending problem of the type we have been
discussing where the beam is acted upon by a set of concentrated forces, Pi , and moments,
M i . Following the dummy load procedure, we can calculate
∂U ðQ; Pi , M i Þ
uQ ¼ (8.149)
Q¼0 ∂Q Q¼0

where the strain energy, U, is given by

ðL
1
U¼ M 2 dx (8.150)
2EI
0

so that
ðL
1 ∂M
uQ ¼ MjQ¼0 dx (8.151)
Q¼0 EI ∂Q Q¼0
0

But MjQ¼0 ¼ M ðPi , M i ; xÞ is just the moment distribution without the dummy load present
and, by superposition, we have the total moment distribution with the dummy load present
given as
M ðQ; Pi , M i ; xÞ ¼ M ðPi , M i ; xÞ þ M 1 ðQ; xÞ (8.152)

where M 1 ðQ; xÞ is the moment distribution due to only Q. Since we are dealing with a linear
system
M 1 ðQ; xÞ ¼ QM 1 ðQ ¼ 1; xÞ (8.153)

where M 1 ðQ ¼ 1; xÞ is the moment distribution due only to a unit force at Q. Using

Eq. (8.152) and Eq. (8.153) it follows that
∂M ðQ; Pi , M i ; xÞ
¼ M 1 ðQ ¼ 1; xÞ (8.154)
∂Q Q¼0

so that Eq. (8.151) becomes

ðL
1
uQ Q¼0
¼ M ðPi , M i ; xÞM 1 ðQ ¼ 1; xÞdx (8.155)
EI
0

which is an expression for desired displacement using the unit-load method. The advantage
of using Eq. (8.155) over the original dummy load procedure is that in the unit-load method,
we need calculate only (1) the original bending moment distribution (without Q) and (2) the
bending moment distribution from the unit load itself, rather than having to compute a
bending moment distribution when both the original loads and Q are present. As you can
verify yourself, having the original loads and Q present leads to more complicated moment
 8.7 Work–Energy Principles and Discrete Forces and Moments 263

distribution expressions, but where much of the complexity ultimately disappears through
the differentiation process and setting Q ¼ 0 in the original dummy load method. Note that
we can use a unit moment in exactly the same manner as done here with a unit force to
calculate the rotation at any point in the beam instead.
Consider again the previous cantilever beam problem of Fig. 8.14 where we calculate the
deflection at the center of the beam with the unit-load method. The required moment
distributions are
M ðP; xÞ ¼ Px ð 0 < x < LÞ
(
0 ð0 < x < L=2Þ (8.156)
M 1 ðQ ¼ 1; xÞ ¼
ðx  L=2Þ ðL=2 < x < LÞ

which you can verify leads to Eq. (8.148) for the center deflection.

8.7.3 Principle of Least Work

All of the previous uses of the principle of complementary virtual work have been for
statically determinate problems where we could write the complementary strain energy in
terms of the applied loads. For statically indeterminate problems, this is not possible, and we
can only write the complementary strain energy in terms of the applied loads and a set of
unknown forces and/or moments (called redundants) which arise from the overconstrained
nature of statically indeterminate problems. For example, Fig. 8.16a shows a truss structure
similar to the one shown in Fig. 8.13 but where there is an extra internal member BD.
Similarly, Fig. 8.16b shows the same truss as in Fig. 8.13 but where there are now two pin
supports instead of a pin and a roller. In both cases we cannot write the complementary
strain energy in terms of the load P. However, we can use complementary energy and work
concepts to solve for the redundants present in such problems. To illustrate the procedure,

B C B C Figure 8.16 (a) A truss where the

external reaction forces
P  
P Ax , Ay , Dy can be obtained from
equilibrium of the entire truss but
where the truss is statically
indeterminate because of too
A many internal truss members,
A D D
leading to a redundant internal
force, such as the force in member
Ax Ay Ax Ay Dx BD, for example. (b) A truss that
Dy Dy
is statically indeterminate because
(a) (b) of being overly constrained at its
supports, leading to a redundant
support force such as the force Dy ,
for example.
264 Work–Energy Concepts

we will use the specific example of Fig. 8.16b. There are four reaction force components
 
Ax , Ay , Dx , Dy at A and D for the truss and only three equilibrium equations, so there is
one excess unknown which we will take as a redundant force, Dy . We can then write the
 
complementary strain energy as U c ¼ U c P; Dy . Now, by the principle of complementary
virtual work, if we vary the force Dy , we have
∂U c
δU c ¼ δDy ¼ δW cv ¼ 0 (8.157)
∂Dy

since there is no vertical displacement at the reaction force so that there is no corresponding
complementary virtual work. Since this result must be true for any variation, δDy , we have
 
∂U c P; Dy
¼0 (8.158)
∂Dy

which is an equation that we can use to solve for Dy .

This same process works in more general cases where we have multiple known applied
forces and moments ðPi , M i Þ and similarly multiple redundant forces and moments
ðRi , M Ri Þ at locations where the generalized displacements or rotations are zero. In this
case we can write
U c ¼ U c ðRi , M Ri , Pi , M i Þ (8.159)

If we vary the redundant forces and moments, the principle of complementary virtual work gives
X
n
∂U c Xm
∂U c
δU c ¼ δRi þ δM Ri ¼ δW cv ¼ 0 (8.160)
i¼1
∂Ri i¼1
∂M Ri

where we have assumed that there are n redundant forces and m redundant moments. If we
can vary these redundant forces and moments independently, then
∂U c
¼ 0 ði ¼ 1; . . . ; nÞ
∂Ri
(8.161)
∂U c
¼ 0 ði ¼ 1; . . . ; mÞ
∂M Ri
are n þ m independent equations that can be used to solve for the n þ m unknown
redundants. Equation (8.161) is called the theorem (or principle) of least work (or
Engesser’s second theorem). Stated explicitly, this theorem says:

Of all the possible values of the independent redundants ðRi , M Ri Þ that satisfy
equilibrium for a statically indeterminate elastic system, the correct values of the
redundants are those that make the complementary strain energy stationary and
hence satisfy Eq. (8.161).
 8.7 Work–Energy Principles and Discrete Forces and Moments 265

The theorem is called the theorem of least work because δW cv ¼ 0 (although, strictly
speaking, this means only that the complementary virtual work has a stationary value). We
have illustrated the principle of least work for the case where the redundants arise from an
excess number of external constraints on a system but the principle is also valid when the
redundancy is due to having too many internal forces, as is the case for the truss shown in
Fig. 8.16a (see problem P8.3).

Example 8.6 The Principle of Least Work

As an example of the use of the principle of least work, consider the statically indeterminate
beam shown in Fig. 8.17a. This problem is statically indeterminate to the first degree, i.e.,
there is one more unknown reaction forces than equations of equilibrium. We can take the
unknown reaction force, R, at the roller support A as the single redundant. The bending
moment in terms of the coordinate x shown in Fig. 8.17b is
q x2
M ðxÞ ¼ Rx  (8.162)
2
so that the complementary strain energy is
ðL  2
1 qx2
U ðRÞ ¼ U ðRÞ ¼
c
Rx  dx (8.163)
2EI 2
0

and from the principle of least work

ðL  
∂U c 1 qx2
¼ Rx  x dx ¼ 0 (8.164)
∂R EI 2
0

which yields R ¼ 3qL=8. This can be verified to be the correct answer by integrating the
moment–curvature relationship twice to obtain the displacement, wðxÞ, of the neutral axis
and then using the boundary conditions wð0Þ ¼ wðLÞ ¼ 0 and dwðLÞ=dx ¼ 0 to solve for the
constants of integration and the reaction force, R.

x Figure 8.17 (a) Statically indeterminate beam

q q subjected to a uniform distributed load, q. (b) The
reaction force, R, at the roller support A, which
A B A B can be taken as the redundant force for
R the structure.

L L

(a) (b)
266 Work–Energy Concepts

Figure 8.18 (a) Statically P P

indeterminate beam between two MA MB
fixed supports. (b) The free body
diagram of the beam showing the A B
reaction forces and moments at
L/2 L/2 RA RB
the supports.

(a) (b)

In choosing the redundants, one must ensure that we can vary those redundants independ-
ently without violating equilibrium, which is the condition we used to arrive at Eq. (8.161).
For the beam between two fixed supports A and B, as shown in Fig. 8.18a, there are four
unknown reaction forces and moments (Fig. 8.18b) and only two equations of equilibrium
so that the problem is statically indeterminate to the second degree and we must choose two
redundants. We cannot choose the reaction forces RA and RB as those redundants since force
equilibrium requires that RA þ RB ¼ P so that the virtual changes of these forces are related,
i.e., δRA ¼ δRB . However, we could choose, for example, RB , M B as the redundants.
Earlier, we stated that Engesser’s first theorem and Castigliano’s second theorem are also
true for statically indeterminate systems. The principle of least work shows why this is the
case. In a statically indeterminate case, the complementary energy can be described only in
terms of the known loads and moments and the redundants, i.e., U c ¼ U c ðPi , M i , Ri , M Ri Þ.
The principle of complementary virtual work gives
!
XN
∂U c XM
∂U c X m
∂U c Xn
∂U c
δU ¼
c
δPi þ δM i þ δRi þ δM Ri
i¼1
∂Pi i¼1
∂M i i¼1
∂Ri i¼1
∂M Ri
(8.165)
XN XM
¼ ui δPi þ θi δM i
i¼1 i¼1

where the displacements and rotations at the redundants are zero so they do no comple-
mentary virtual work. By the theorem of least work, the term in parentheses in Eq. (8.165)
vanishes so we are again left with Eq. (8.140) which leads to Engesser’s first theorem and
Castigliano’s second theorem. Although the redundants themselves are functions of the
applied loads, i.e., Ri ¼ Ri ðPi , M i Þ, M Ri ¼ M Ri ðPi , M i Þ, it is not necessary to take deriva-
tives of the applied loads and moments appearing in these redundants in computing partial
derivatives of the complementary energy since by the chain rule we have, for example,
∂U c ðPi , M i ; Ri ðPi , M i Þ, MRi ðPi , M i ÞÞ ∂U c
¼
∂Pi ∂Pi M i , Ri , M Ri ¼constants
!
Xm
∂U c ∂Rk
þ (8.166)
k¼1
∂Rk Pi , M i , M Ri ¼constants ∂Pi
!
Xn
∂U c ∂M Rk
þ
k¼1
∂M Rk Pi , M i , Ri ¼constants ∂Pi
 8.8 Reciprocity 267

But by Eq. (8.161) all the partials with respect to the redundants must vanish, leaving
∂U c ðPi , M i , Ri , MRi Þ ∂U c
¼ (8.167)
∂Pi ∂Pi M i , Ri , M Ri ¼constants

and, similarly,
∂U c ðPi , M i , Ri , MRi Þ ∂U c
¼ (8.168)
∂M i ∂M i Pi , Ri , M Ri ¼constants

Thus, in applying Engesser’s first theorem or Castigliano’s second theorem we can treat the
redundants as constants in those theorems.

8.8 Reciprocity
A linear elastic body possesses the important property of reciprocity that can be used to
advantage in many cases to find solutions to problems. Consider a linear elastic body that
occupies a volume V whose surface is S, and let x be an arbitrary point in V and let xs be an
ð1Þ ð1Þ ð1Þ
arbitrary point on S. Also, let σ ij , eij , uj be the stresses, strains, and displacements in V for
ð1Þ ð1Þ ð1Þ
this body due to T j ðxs Þ, uj ðxs Þ, f j ðxÞ, which are the surface stress components, surface
displacement components, and body force components (we can call this case (1)). Similarly,
ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ
let σ ij , eij , uj be the stresses, strains, and displacements due to T j ðxs Þ, uj ðxs Þ, f j ðxÞfor
the same body (we can call this case (2)). Then
X 3 X 3 ð X3 X3 ð
ð1Þ ð2Þ ð2Þ ð1Þ
σ ij eij dV ¼ σ ij eij dV (8.169)
i¼1 j¼1 i¼1 j¼1
V V

which follows directly from placing the linear stress–strain relations

ð1Þ
3 X
X 3
ð1Þ
σ ij ¼ C ijkl ekl
i¼1 j¼1
(8.170)
ð2Þ
X
3 X
3
ð2Þ
σ ij ¼ C ijkl ekl
i¼1 j¼1

into Eq. (8.169) and using the symmetry of the elastic constants C ijkl ¼ C klij .
Because of the symmetry of the stresses, for either case (1) or case (2) we have
X 3 X   X
3
1X 3 X 3
∂ui ∂uj 3 X 3
∂uj
σ ij eij ¼ σ ij þ ¼ σ ij (8.171)
i¼1 j¼1
2 i¼1 j¼1
∂x j ∂x i i¼1 j¼1
∂x i
268 Work–Energy Concepts

so that

X 3 ð
3 X ð2Þ
∂uj X3 X 3 ð ð1Þ
ð1Þ ð2Þ ∂uj
σ ij dV ¼ σ ij dV (8.172)
i¼1 j¼1
∂xi i¼1 j¼1
∂xi
V V

From the use of the chain rule of calculus and the equations of equilibrium, for either case
we have
3 X
X 3
∂uj X 3 X 3
∂   X 3
σ ij ¼ σ ij uj þ f j uj (8.173)
i¼1 j¼1
∂xi i¼1 j¼1
∂xi j¼1

so that Eq. (8.172) can be written as

3 ð 3 ð
X3 X
∂  ð1Þ ð2Þ  X ð1Þ ð2Þ
σ ij uj dV þ f j uj dV
i¼1 j¼1
∂x i j¼1
V V
(8.174)
3 ð 3 ð
X
3 X
∂  ð2Þ ð1Þ  X ð2Þ ð1Þ
¼ σ ij uj dV þ f j uj dV
i¼1 j¼1
∂xi j¼1
V V

But, by the divergence theorem, the volume integrals in Eq. (8.174) can be transformed into
integrals over the surface of the body, giving
X3 ð X3 ð X 3 ð X3 ð
ð1Þ ð2Þ ð1Þ ð2Þ ð2Þ ð1Þ ð2Þ ð1Þ
T j uj dS þ f j uj dV ¼ T j uj dS þ f j uj dV (8.175)
j¼1 j¼1 j¼1 j¼1
S V S V

ðm Þ P3 ðmÞ
where T j ¼ i¼1 σ ij ni (m ¼ 1,2) are the surface stress vector components for the two
cases. [Note: the notation in Eq. (8.175) is different than used previously where we expressed
ðnÞ
the stress vector components on a surface as T j (see Eq. (2.13)). Here we will drop the (n)
ðnÞ
superscript, where n is the unit normal to the surface, so that T j  T j . The superscript (m)
ðmÞ
in T j is now simply an indication of the case number.] Equation (8.175) is a statement of
the reciprocal theorem of Betti–Rayleigh. Stated explicitly in words, this theorem says that
the work done by the surface stresses and body forces of case (1) acting through the
displacements of case (2) is equal to the work done by the surface stresses and body forces
of case (2) acting through the displacements of case (1). For the special case when all the
surface forces and moments ðPi , M i Þ are concentrated and the body forces are absent, this
theorem can be written as
X
N
ð1Þ ð2Þ
X
3
ð1Þ ð2Þ
X
N
ð2Þ ð1Þ
X
3
ð2Þ ð1Þ
Pj uj þ M j θj ¼ Pj uj þ M j θj (8.176)
j¼1 j¼1 j¼1 j¼1
 8.8 Reciprocity 269

Example 8.7 Reciprocity

To illustrate a use of reciprocity, consider a cantilever beam acted upon by a distributed
load, qðxÞ, as shown in Fig. 8.19a. Also, let the same beam be loaded by a concentrated load,
P, at a distance, d, from the support, as shown in Fig. 8.19b. Then from the reciprocal
theorem we have
ðL
 qðxÞwð2Þ ðP; x; d Þdx ¼ Pwð1Þ ðw; d Þ (8.177)
0

where wð2Þ ðP; x; d Þ is the z-displacement of the beam at x due to a load P acting at d and
wð1Þ ðq; d Þ is the displacement at d due to the distributed load qðxÞ. The minus sign exists on
the left side of Eq. (8.177) because the distributed load acts in the negative z-direction. It
is relatively easy to solve the concentrated load problem of Fig. 8.19b and obtain the
z-displacement of the beam for that problem as
8
>
> Px2
>
< ð3d  xÞ ð0 < x < d Þ
6EI
wð2Þ ðP; x; d Þ ¼ (8.178)
>
> Pd 2
>
: ð3x  d Þ ðd < x < LÞ
6EI
which gives
8d 9
1 <ð ðL =
wð1Þ ðq; d Þ ¼ qðxÞx2 ð3d  xÞdx þ qðxÞd 2 ð3x  d Þdx (8.179)
6EI : ;
0 d

Equation (8.179) gives the z-displacement of the beam at any point d ð0 < d < l Þ due to the
arbitrary distributed load qðxÞ.

This example shows that the usefulness of the reciprocal theorem is based on its ability to
relate the effects of two solutions on each other, a property that can be used to extract
information about one of those solutions. The boundary element method, which is one of the

z z Figure 8.19 (a) Cantilever beam loaded by an

q ( x) P
arbitrary distributed load, qðxÞ. (b) The same
d beam loaded by a concentrated force, P,
located at an arbitrary distance, d, from the
x x fixed end.
L L

(a) (b)
270 Work–Energy Concepts

important numerical methods described briefly in Chapter 6, is based on reciprocity and uses
a fundamental solution (which is a solution to point sources acting in the body) to derive an
integral equation for the solution of a stress analysis problem. The solution derived in
Eq. (8.179) also is based on a point-source solution, Eq. (8.178), so it shares some of the
same characteristics found in the boundary element method, which will be discussed in more
detail in Chapter. 9.

8.9 PROBLEMS

P8.1 For the two-bar truss problem of Fig. 8.5, solve Eq. (8.89) for the bar forces and
reactions using the symbolic algebra capabilities of MATLAB®. Verify that you obtain
the solution given in Eq. (8.90).
P8.2 For the three-bar truss problem of Fig. 8.9, use the symbolic algebra capabilities of
MATLAB® to solve for the bar forces using both a displacement-based approach and
a force-based approach. Verify that you obtain the solution given in Eq. (8.117).
P8.3 Solve the statically indeterminate truss of Fig 8.16a for the redundant force, R, in bar
BD using the principle of least work. Assume all bars have the same cross-sectional
area, A, and Young’s modulus, E.
P8.4 In statics courses one usually examines the equilibrium of rigid bodies where the
assumption of rigidity means that there can be no elastic strain energy stored in the
bodies. In this case the principle of virtual work gives simply δW v ¼ 0. As in the case of
elastic bodies, the virtual work principle can be used in place of the equations of
equilibrium. For example, consider the two rigid bars shown in Fig. P8.1. These bars,
each of length L and weight W, are pinned to each other and supported by a pin at
point A. At C, where the bar rides on a smooth surface, a couple, M, acts.
(a) Using the equations of equilibrium, determine the angle, θ, of the system in terms
of M, W, and L.
(b) Using the principle of virtual work, determine the angle, θ, of the system in terms of
M, W, and L.
P8.5 The “wishbone” frame shown in Fig. P8.2 is subjected to a horizontal force, P, at
point B. The frame is pinned at point A and on rollers at point B. Determine
expressions for the horizontal deflection and the rotation at B using Castigliano’s
second theorem and the unit load method, respectively. Assume that the bending
stiffness, EI, is constant throughout the beam and
M consider the strain energy contributions for both
axial extension and bending.
A L L C

T T
W W Figure P8.1 Two rigid bars in equilibrium under the action of
B their own weights and an applied couple, M.
 8.9 Problems 271

P8.6 The bending stiffness, EI, is uniform throughout the frame shown in Fig P8.3. The
frame is fixed (built-in) at point B and supported on rollers at point A. Using
Castigliano’s second theorem and the dummy load method, determine the horizontal
displacement at point A in terms of the load, F, the stiffness, and the length, L.
Consider the strain energy due to bending only.
P8.7 A symmetrical L-shaped bar of circular cross-section lies in a horizontal plane and
is rigidly fixed at ends A and B to a vertical wall. The two legs of the bar are of equal
length, L, and at right angles to each other. The stiffnesses EI and GJ are constant
throughout. If a vertical force, F, acts on the bar as shown in Fig. P8.4 determine
the support forces and moments at the wall supports and show them on a sketch.
Assume the strain energy is due to bending and torsion only. For simplicity, also
assume EI ¼ GJ. [Hint: by symmetry, the vertical forces
L at A and B are equal.]
P8.8 A structure in the form of a three-sided rectangular frame of
constant circular cross-section lies in a horizontal plane and
is rigidly fixed at points A and B, as shown in Fig. P8.5.
L L A force, P, is applied to the frame.

A B
P Figure P8.2 A frame pinned at point A and on rollers at point B, subjected
to horizontal force, P.

F
L L

L
2L A

Figure P8.3 A frame that is clamped (built-in) at point B and on rollers at

B
point A.

Figure P8.4 An L-shaped bar of circular cross-

A B section that lies in a horizontal plane and is rigidly
fixed at A and B and subjected to a vertical force, F.
L L

F
272 Work–Energy Concepts

(a) Using symmetry, determine explicitly the reaction forces and bending moments at
points A and B. What does symmetry imply about the reaction torques at A and B?
(b) Using energy methods, determine the reaction torques at points A and B in terms
of P, a, b, EI, and GJ. Consider only the strain energy due to bending and torsion.
P8.9 The frame shown in Fig. P8.6 is of constant cross-section and is subjected to a
uniform distributed load of intensity q0 lb/unit length. Using energy methods, deter-
mine the horizontal displacement of the frame at roller D in terms of q0 ,H, L, and EI.
Neglect the strain energy due to shear.
P8.10 A sign of weight W is hung from a bent bar of circular cross-section, which is built
into a wall at point C (see Fig. P8.7) and has a constant EI and GJ along its entire
length, where, for simplicity, take GJ ¼ 2EI /3. Determine – in terms of W, L, and
EI – the vertical sag of the bar at A (i.e., the relative vertical displacement at
A relative to that at B). Consider only the strain energy due to bending and torsion.
P8.11 A sign of weight W is hung from a frame which has a constant EI along its entire
length (see Fig. P8.8). Determine at the end point P:
(a) the downward deflection, Δ, in terms of W, L, and EI
(b) the clockwise rotation, θ, in terms of W, L, and EI.
Use the unit-load method and consider only the strain energy
due to bending.
B
P8.12 A pair of forces of magnitude P each act on a rect-
y
P angular frame which has a constant EI, as shown in
x Fig. P8.9. Determine the total horizontal elongation of
b the frame between the points at which the loads act.
z Consider only the strain energy due to bending.
b
A
a
Figure P8.5 A rectangular frame that lies in the x–z plane and
subjected to a force, P, in the negative y-direction.

q0

B C

H H

A D

L Figure P8.6 A frame subjected to a distributed load.

 8.9 Problems 273

P8.13 The rectangular beam shown in Fig. P8.10 carries a uniform distributed force/unit
length, q0 , acting at the bottom of the beam over its entire length. Its Young’s
modulus is E. Determine the horizontal and vertical deflections of the beam at point
O in terms of E, q0 , L, b, and h. Point O is at the centroid of the cross-section at the
free end. Consider the strain energy due to the bending moment
and the axial force in the beam only.
P8.14 The L-shaped frame shown in Fig. P8.11 has a uniform EI
C L
value and is pinned at both ends so that it represents a
statically indeterminate problem. Determine the horizontal
L B
component of the reaction force at point C in terms of the
uniform distributed load intensity,q0 , and the lengths d, and l.
A Consider the strain energy due to the internal bending
W moment only.

Figure P8.7 A sign of weight W hung from a bent bar at points A and B.

L/2 L/2

4L
W

Figure P8.8 A sign of weight W hung from a frame.

Figure P8.9 A rectangular frame acted upon by a pair of

b horizontal forces.
P P
b

2a

z Figure P8.10 (a) Rectangular cantilever beam where a

uniform axial distributed force/unit length acts on the bottom
O of the beam over its entire length. (b) The dimensions of the
L h
cross-section.

q0 x
b
(a) (b)
274 Work–Energy Concepts

P8.15 A circular thin-wall tube is bent to form the symmetrical frame shown in Fig. P8.12
and is loaded by a pair of forces, each of magnitude P, acting at the centers of the top
two horizontal sections. Let a ¼ b ¼ c and E ¼ 3G, where E is Young’s modulus and
G is the shear modulus. Using energy methods, determine:
(a) the internal bending moment in the frame at the applied loads in terms of P and a;
(b) the relative displacement of the frame along the line AB between the two loads in
terms of P, a, E, and I, where I is the area moment of the thin circular tube;
(c) the rotation of the frame about the x-axis at B in terms of P, a, E, and I.
Neglect the strain energy in the frame due to the internal
shear force and axial load.
d P8.16 An end force, P, is applied to a thin cantilever
q0
beam which is bent into a circular shape, as shown
A B in Fig. P8.13. Determine the radial displacement
of the beam as a function of the angle, ϕ, the
force, P, and the EI of the cross-section, where E
is Young’s modulus and I is the area moment. Use
l
ordinary engineering beam theory as developed
y

C Figure P8.11 An L-shaped frame pinned at both ends and loaded by a

x
uniform distributed force/unit length, q0 .

y
Figure P8.12 A bent frame acted upon by a pair of forces.

B
z x P

A b

2c
2a

y
Figure P8.13 A thin beam bent into a circular shape and loaded by a
force, P.

R
I
x
P
 References 275

Figure P8.14 A thin beam bent into a circular shape. A uniform load/unit length, q0 , acts
q0
along the beam.

I
R

for a straight beam and consider the strain energy due to the internal bending
moment only.
P8.17 A thin cantilever beam is bent into a circular shape as shown in Fig. P8.14. A uniform
distributed load/ unit length, q0 , acts along its entire length.
(a) Determine the horizontal and vertical deflections of the beam at its free end,
ϕ ¼ 0, in terms of q0 , R, and the EI of the cross-section, where E is Young’s
modulus and I is the area moment. Use ordinary engineering beam theory as
developed for a straight beam and consider the strain energy due to the internal
bending moment only.
(b) Repeat part (a) but compute the horizontal and vertical deflections of the beam at
its free end, ϕ ¼ 0, in terms of q0 , R, and the EA of the cross-section, where E is
Young’s modulus and A is the cross-sectional area. Consider the strain energy due
to the internal axial force only.

References
1. R. G Budynas, Advanced Strength and Applied Stress Analysis, 2nd edn (Boston, MA: McGraw-
Hill, 1999)
2. C. E. Pearson, Theoretical Elasticity (Cambridge, MA: Harvard Univ. Press, 1959)
3. W. B. Bickford, Advanced Mechanics of Materials (Menlo Park, CA: Addison Wesley, 1998)
4. H. Reismann and P. S Pawlick, Elasticity – Theory and Applications (New York, NY: John Wiley,
1980)
9 Computational Mechanics of Deformable Bodies

Analytical solutions of deformable body problems are possible in cases involving simple
geometries and loads. The complex problems commonly found in practice, however, can
only be solved with approximate numerical methods. The method of finite elements has
been highly developed over the past 60 years to become the numerical method of choice
for analyzing deformable bodies as well as many other engineering problems. More
precisely, the stiffness-based finite element method is the engineering tool used almost
universally for stress and deformation analyses. In this chapter we will describe the basic
elements of both stiffness-based and force-based finite elements. [Note: what we are
calling here force-based finite elements is also called the Integrated Force Method to
distinguish it from earlier force methods [1], [2].] This introduction is intended to pave the
way for more in-depth studies of finite elements in higher-level courses and to show the
importance of the energy methods discussed in the previous chapter. Force-based finite
elements are not discussed in current advanced strength of materials texts but they have
been presented in an elementary strength of materials setting [3]. The lack of discussion
of force-based methods is partially due to the almost exclusive use of the stiffness
approach in powerful commercial software codes but it is also rooted in the fact that
force-based methods rely on appropriately combining equilibrium and compatibility
while the stiffness method relies solely on equilibrium. Thus, it has taken longer to
appreciate how that combination can be used effectively as a general computational
tool. Force-based methods, however, are more closely related to the governing equations
described in Chapter 6 than are stiffness-based methods, so there is much to be gained by
understanding how force-based methods work.
We will first examine both stiffness-based and force-based finite elements for axial-load
and bending problems (the discussion of torsion problems is omitted since they are identical
to axial-load problems in their structure). These problems will be described using a matrix–
vector approach that will then be generalized to general deformable body problems. Simple
examples will be used that allow us to work through the application of these finite element
methods in detail and to compare results with analytical solutions.
Another method that has proven to be an important alternative to the stiffness-based
finite element approach is the boundary element method. This method, like force-based
finite elements, has been largely overshadowed by stiffness-based finite elements, but it is an
important method and a computational tool that has a useful niche in solving specialized
problems such as cracked bodies. We will not provide a detailed description of boundary
elements, but will outline the basic foundations of the method and contrast it to
finite elements.

276
 9.1 Numerical Solutions – Axial Loads 277

9.1 Numerical Solutions – Axial Loads

Many of the axial-load problems described in elementary strength of materials courses are
simple enough that they can be solved analytically. Here, however, we want to lay the
foundations for solving more complex problems numerically so we will use axial load
problems as part of that foundation. Consider the problem shown in Fig. 9.1a where a bar
of varying cross-sectional area, AðxÞ, is subjected to concentrated axial forces (such as P∗ j ) and
distributed forces, qx ðxÞ. The forces ðR1 , R2 Þ could be either reaction forces at the ends where
displacements are fixed or they could be applied end forces. We will also let the Young’s
modulus, E ðxÞ, of the bar be a continuously varying function. We will examine both stiffness-
based and forced-based finite element methods for solving this problem numerically.

9.1.1 Principles of Virtual Work and Complementary Virtual Work

In the finite element method, the bar is divided into M elements as shown in Fig. 9.1b. If we
consider the ( j)th element of the bar, as shown in Fig. 9.2a, this element will have end forces


Pj , Pjþ1 acting in the x-direction, a distributed force/unit length, qx ðξ Þ, acting in the
element, as well as an internal axial force, F x ðξ Þ ¼ σ xx ðξ ÞAðξ Þ, where σ xx ðξ Þ is the axial
stress and Aðξ Þ is the cross-sectional area of the bar. The coordinate ξ is measured from the
start of the element as shown in Fig. 9.1b. There are also end displacements of the element,


U j , U jþ1 which are called the nodal displacements. Note there are M + 1 nodal displace-
ments for the entire bar if there are M elements. The length of the element is l. We will

y Figure 9.1 (a) A general axial-load problem

qx ( x ) where there are concentrated forces and a
R2 body force distribution acting on the bar. The
R1 x
*
bar can be of varying cross-sectional area and
P j have a varying Young’s modulus. (b) The bar
divided into M finite elements.
A( x), E ( x)

(a)

xj x

(1) (j) (M)

(j)th element
l

(b)
278 Computational Mechanics of Deformable Bodies

Figure 9.2 (a) The ( j)th element of Uj

U j +1
the bar, showing the distributed
ðj Þ
force/unit length, qx ðξ Þ, and the


displacements U j , U jþ1 and forces qx


Pj , Pjþ1 acting at the end nodes. Pj Pj +1
(b) The same element showing the
internal axial force, F x ðξ Þ (assumed
x
to be tensile), in the element and its
values at the end nodes. l

(a)

Fx ( 0 ) = s xx ( 0 ) A ( 0 ) Fx ( l ) = s xx ( l ) A ( l )
Fx ( x )

(b)

assume that these elements are chosen so that any concentrated force such as the P∗ j shown
in Fig. 9.1a lies at the start or end of an element, i.e., at a node.
The principle of virtual work applied to the ( j)th element says that the virtual change of
the strain energy of the element, δU, is equal to the virtual work, δW v , of the external forces
acting on the element, i.e.,
δU ¼ δW v (9.1)

From the general results developed in Chapter 8 for a deformable body (see Eq. (8.71b), for
example) specialized for this uniaxial-load problem we have
ðl ðl X
jþ1
σ xx ðξ Þδexx ðξ ÞAðξ Þdξ ¼ f x ðξ Þδux ðξ ÞAðξ Þdξ þ Pm δU m (9.2)
m¼j
0 0

where f x is the body force (force/unit volume) and Pm ðm ¼ j, j þ 1Þ are the forces acting at
nodes j and jþ1 (Fig. 9.2a). The quantities ðδexx , δux , δU m Þ are virtual changes of the
strain,exx , and x-displacement,ux , in the element and virtual changes of the nodal dis-
placements,U m ðm ¼ j, j þ 1Þ, at the ends of the element, respectively. The internal force
acting in the element is just F x ðξ Þ ¼ σ xx ðξ ÞAðξ Þ and we can define the distributed force/unit
length, qx ðξ Þ, in terms of the body force term in the element as qx ðξ Þ ¼ f x ðξ ÞAðξ Þ so that the
principle of virtual work for the element becomes
ðl ðl X
jþ1
F x ðξ Þδexx ðξ Þdξ ¼ qx ðξ Þδux ðξ Þdξ þ Pm δU m (9.3)
m¼j
0 0
 9.1 Numerical Solutions – Axial Loads 279

[ A note on the notation: various properties will vary from element to element so at times it is
important to distinguish the element associated with that property. We will do this by using a
superscript with a number or symbol written in parentheses, e.g. ( j), to indicate the element.
Numbers or symbols written in subscripts without parentheses, e.g., j, jþ1, will indicate nodal
values. Some quantities such as the area written as Aðj Þ ðξ Þ are only functions of the element
number, while some other quantities, such as the nodal displacements, U j , have only nodal
values and are not associated with a particular element. An end force such as Pjþ1 , however,
might be the force at the right end of element ( j), as seen in Fig. 9.2a, or it might be the end
ðj Þ
force at the start of element ( jþ1). Thus, the Pjþ1 seen in Fig. 9.2a can be written as Pjþ1 to
indicate it is associated with the element ( j) and acts at node jþ1 and to distinguish it from
ðjþ1Þ
Pjþ1 , which is an end force for element ( jþ1) at the same node jþ1. This notation is very
explicit but it is not used except where needed for clarity. Also note that, for convenience, we
have taken the start and end nodes of element ( j) to be labeled as j and jþ1, but such
sequential numbering is not essential. We should also note that we have denoted the strain
energy by U and the nodal displacements by U m . This should not cause confusion since the
strain energy is always a scalar, while the displacements are always vectors written with either
an explicit subscript or written as a vector in matrix–vector notation such as fU g.]
Equation (9.3) also follows directly from the equation of equilibrium for the internal force
in the bar following similar steps used in Chapter 8, starting from the local equilibrium
equations for the stresses. Recall from Eq. (8.49) we have in the bar
dF x
þ qx ¼ 0 (9.4)
dξ
If we multiply this equilibrium equation by a virtual displacement δux ðξ Þ and integrate over
the length of the element we have
ðl  
dF x
þ qx δux dξ ¼ 0 (9.5)
dξ
0

But we have
dF x d d
δux ¼ ðF x δux Þ  F x ðδux Þ
dξ dξ dξ
 
d dux
¼ ðF x δux ÞF x δ (9.6)
dξ dξ
d
¼ ðF x δux Þ  F x δexx
dξ
so using Eq. (9.6) in Eq. (9.5) and carrying out the integration of the perfect differential we find
ðl ðl
F x ðξ Þδexx ðξ Þdξ ¼ qx ðξ Þδux ðξ Þdξ þ F x ðl Þδux ðl Þ  F x ð0Þδux ð0Þ (9.7)
0 0
280 Computational Mechanics of Deformable Bodies

We also have δux ðl Þ ¼ δU jþ1 , δux ð0Þ ¼ δU j and F x ðl Þ ¼ Pjþ1 , F x ð0Þ ¼ Pj (see Figs. 9.2a, b)
so that the principle of virtual work becomes
ðl ðl
F x ðξ Þδexx ðξ Þdξ ¼ qx ðξ Þδux ðξ Þdξ þ Pjþ1 δU jþ1 þ Pj δU j (9.8)
0 0

which is identical with Eq. (9.3).

Now let us turn our attention to the principle of complementary virtual work, which states
that the virtual change of the complementary strain energy, δU c , for an element is equal to the
complementary virtual work done by the external forces, δW cv , acting on that element, i.e.,
δU c ¼ δW cv (9.9)

In applying this principle, we normally consider virtual changes of the stresses that satisfy
the homogeneous equations of equilibrium, i.e., the virtual changes δσ ij satisfy Eq. (8.97)
with δf i ¼ 0. In that case the principle applied to the element of Fig. 9.2a gives (see Eq.
(8.96)):
ðl X
jþ1
exx ðξ Þδσ xx ðξ ÞAðξ Þdξ ¼ U m δPm (9.10)
m¼j
0

or, in terms of the internal force, the principle of complementary virtual work is:
ðl X
jþ1
exx ðξ ÞδF x ðξ Þdξ ¼ U m δPm (9.11)
m¼j
0

We can also develop Eq. (9.11) directly from the 1-D strain–displacement relationship for
the element following similar steps carried out in Chapter 8 for a general state of strain. We
have
dux ðξ Þ
exx ðξ Þ ¼ (9.12)
dξ
Multiplying this relationship by a virtual change of the internal force, δF x ðξ Þ, and integrat-
ing over the element gives
ðl  
dux
exx  δF x dξ ¼ 0 (9.13)
dξ
0

and we have
dux d d
δF x ¼ ðux δF x Þ  ux ðδF x Þ
dξ dξ dξ (9.14)
d
¼ ðux δF x Þ
dξ
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 281

where we have used the fact that the virtual change of the internal force satisfies the
homogeneous equilibrium equation
d
ðδF x Þ ¼ 0 (9.15)
dξ
(see Eq. (9.4) with qx ¼ 0). Using Eq. (9.14) in Eq. (9.13) and carrying out the integration of
the perfect differential gives
ðl
exx ðξ ÞδF x ðξ Þdξ ¼ ux ðl ÞδF x ðl Þ  ux ð0ÞδF x ð0Þ
(9.16)
0
¼ U jþ1 δPjþ1 þ U j δPj

which is identical to Eq. (9.11).

Having the principles of virtual work and complementary virtual work, we now will
examine the use of these principles to develop stiffness-based and force-based finite
element methods for solving axial-load problems numerically. We will start with the
stiffness-based method, which is, as mentioned previously, the one in predominant
use today.

9.2 Stiffness-Based Finite Elements for Axial-Load Problems

We can express the principle of virtual work, Eq. (9.3), solely in terms of the displacement of
the element (and derivatives of that displacement) as
ðl ðl X
jþ1
Aðξ ÞE ðξ Þexx ðξ Þδexx ðξ Þdξ ¼ qx ðξ Þδux ðξ Þdξ þ Pm δU m (9.17)
m¼j
0 0

In the stiffness-based finite element method, we write the displacement in the element in
terms of known functions, H m ðξ Þ, and a set of unknown nodal displacements, U m . By
differentiating the displacement, we can also write the strain in the element in terms of the
known functions Bm ðξ Þ ¼ dH m ðξ Þ=dξ and the nodal displacements, i.e.,
X
ux ðξ Þ ¼ H m ðξ ÞU m
m
X (9.18)
exx ðξ Þ ¼ Bm ðξ ÞU m
m

The functions H m ðξ Þ are called the shape functions that define the approximation of the
displacement we wish to use for the element. If the element is very small or the loading is
very simple, a linear variation of displacements (and, hence, a constant strain in the element)
might be sufficient, a choice that we will make to keep the algebra simple in our discussions.
Thus, in the element we will let
282 Computational Mechanics of Deformable Bodies


U jþ1  U j ξ
ux ðξ Þ ¼ U j þ
l
  ε (9.19)
ξ
¼ Uj 1  þ U jþ1 ð0 < ξ < l Þ
l l

where the shape functions H j ðξ Þ, H jþ1 ðξ Þ are

ξ ξ
H j ðξ Þ ¼ 1  , H jþ1 ¼ (9.20)
l l
or, in matrix–vector forms,

ux ðξ Þ ¼ ½H fU g ¼ fU gT ½H T (9.21)

where

Uj
fU g ¼ , ½H  ¼ ½1  ξ=l ξ=l  (9.22)
U jþ1

For this linear variation of displacement, the strain in the element can be written similarly in
terms of strain shape functions Bj ðξ Þ, Bjþ1 ðξ Þ as
∂ux U jþ1  U j
exx ¼ ¼
∂ξ l
   
1 1
¼ Uj  þ U jþ1 (9.23)
l l
X
jþ1
¼ Bm ð ξ Þ U m
m¼j

In this case, the strain shape functions are just constants given by
1 1
Bj ðξ Þ ¼  , Bjþ1 ðξ Þ ¼ (9.24)
l l
and in matrix–vector forms we have

exx ¼ ½BfU g ¼ fU gT ½BT (9.25)

where
½B ¼ ½ 1=l 1=l  (9.26)

We are going to use these approximate expressions for the displacements in the principle of
virtual work so that we also need corresponding forms for virtual displacements and virtual
strains for this element. The matrix–vector forms are

δux ¼ ½H fδU g ¼ fδU gT ½H T

(9.27)
δexx ¼ ½BfδU g ¼ fδU gT ½BT
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 283

where

δU j
fδU g ¼ (9.28)
δU jþ1

Using the matrix–vector forms involving either fU g or fδU gT in Eq. (9.21), Eq. (9.25), and
Eq. (9.27) then allows us to write the virtual work expression of Eq. (9.17) as
ðl
fδU g T
E ðξ ÞAðξ Þ½BT ½Bdξ fU g
0
(9.29)
ðl
¼ fδU gT ½H T qx dξ þ fδU gT fPg
0

where

Pj
fPg ¼ (9.30)
Pjþ1

Since Eq. (9.29) must be satisfied for all choices of fδU gT we must have
ðl ðl
E ðξ ÞAðξ Þ½B ½Bdξ fU g ¼ qx ðξ Þ½H T dξ þ fPg
T
(9.31)
0 0

which is just a set of linear equations of the form

½K fU g ¼ fQg þ fPg (9.32a)

or, equivalently, using the more explicit notation where a superscript ( j) denotes the ( j)th
element
h in o n o n o
K ðjÞ U ðjÞ ¼ QðjÞ þ Pðj Þ (9.32b)
n o
where K ðj Þ is called the stiffness matrix for the element and Qðj Þ is a set of nodal forces
produced by the distributed force/length acting in the element. For our specific choice of
linear displacement and hence constant strain we have
ðj Þ " #
h i lð 1=l ðjÞ
ðj Þ ðj Þ ðj Þ
K ¼ E ðξ ÞA ðξ Þ 1=l ðjÞ 1=l ðjÞ dξ
ðj Þ
0
1=l
(9.33a)
ðj Þ
lð " #
1 1 1
¼ 2 E ðjÞ ðξ ÞAðjÞ ðξ Þdξ
l ðj Þ 1 1
0
284 Computational Mechanics of Deformable Bodies

If the area and Young’s modulus are constants, and all elements have the same length l, then
the stiffness matrix for every element is simply
h i EA
1 1 EA=l EA=l
K ðj Þ ¼ ¼ (9.33b)
l 1 1 EA=l EA=l
n o
For the QðjÞ vector associated with element ( j) we have
8 ðj Þ 9
>
> lð
  >
>
>
> >
>
8 ðj Þ 9 > >
> q ðj Þ
ð ξ Þ 1  ξ=l ðj Þ
dξ >
>
>
n o < Qj = < > x
>
=
ðj Þ 0
Q ¼ ¼ (9.34a)
: Qðj Þ ; >>
ðj Þ
lð >
>
>
>   >
>
jþ1
>
> qðxjÞ ðξ Þ ξ=l ðj Þ dξ > >
>
> >
>
: ;
0

However, normally the distributed load is not given as a function of ξ but instead is given as
ðj Þ ðj Þ
qx ¼ qx ðxÞ, i.e., it is a function of position as measured from a fixed origin, whereas we
recall ξ is measured from the beginning of the element (Fig. 9.1b). Assuming we have
n o
ðj Þ ðj Þ
qx ¼ qx ðxÞ, then we must write QðjÞ instead as
8 ðj Þ 9
> lð >
>
>
   >
>
>
> ðj Þ
þ ξ  ξ=l ðj Þ >
>
8 9 >
> q x j 1 dξ >
>
n o < Qðj j Þ = >
<
x
>
=
0
Qð j Þ ¼ ¼ (9.34b)
: Qð j Þ ; > > ðj Þ
lð >
>
jþ1 >
>
  >
>
>
> >
>
>
> qðxj Þ xj þ ξ ξ=l ðjÞ dξ >
>
: ;
0

where xj is the location of the start of the element in the x-coordinate system (Fig. 9.1b). For
the special case of a constant distributed force/unit length q0 acting in an element of length l,
it does not matter if we use Eq. (9.34a) or Eq. (9.34b). From either equation we find
8 9 ( )
n o < Qðj jÞ = q l=2
Qð j Þ ¼
0
¼ (9.35)
: Qð j Þ ; q0 l=2
jþ1

Equation (9.32b) is a stiffness-based expression for equilibrium of an element based on the

use of the principle of virtual work. To obtain the equilibrium equations for the entire bar,
we must assemble the elements, i.e., combine these equations appropriately. First, let us
examine the nodal force terms on the right-hand side of Eq. (9.32b). Figure 9.3a shows the
nodal forces generated by the distributed load acting on two adjacent elements individually,
and Fig. 9.3b shows the combined nodal forces when these two elements are combined
(using our more explicit notation where the element involved is indicated). For example, if
qx ¼ q0 ¼ constant over both elements and the elements have the same length (Fig. 9.4a),
then adding the nodal contributions found from Eq. (9.35) we have nodal forces
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 285

Q (j j ) Q (j +1)
j
Q (j +1 )
j +1
Q (j + 2 )
j +1 Figure 9.3 (a) The equivalent
nodal forces on two adjacent
elements that are produced by a
distributed load. (b) The
(j) (j+1)
(a) equivalent forces when these
elements are combined.
Q (j +j 1) + Q (j +j 1+1)
Q (j
j)
Q (j + 2 )
j +1

(j) (j+1)

(b)

(j) (j+1) Figure 9.4 (a) A bar with a uniform distributed

q0 force/unit length, q0 , acting over two adjacent
q0
elements. (b) The equivalent nodal forces.

l l
(a)

q0l q0l
q0l
2 2

(b)

contributions shown in Fig. 9.4b when the two elements are combined. Now, consider the
   
nodal forces contained in the end forces Pðj Þ and Pðjþ1Þ . Figure 9.5a again shows these
forces for two adjacent elements ( j) and ( j+1). If an applied concentrated force, P∗ jþ1 is
present at node j+1 then Fig. 9.5b shows a free body diagram of a small section of the bar at
that node. From the equilibrium of that small section, we have
ðj Þ ðjþ1Þ
Pjþ1 þ Pjþ1 ¼ P∗
jþ1 (9.36)

so that the nodal forces simply combine to the value of that concentrated force. This also
shows that if no concentrated force is present, the sum of the end forces must vanish. This
result also follows from the fact that these end forces arise from the internal axial forces
present in the bar, as shown in Fig. 9.5c. At node j+1, these internal forces (assumed tensile)
satisfy, from Fig. 9.5c,
 
F ðxjÞ l ðjÞ  F ðxjþ1Þ ð0Þ ¼ P∗
jþ1 (9.37)
286 Computational Mechanics of Deformable Bodies

Figure 9.5 (a) The nodal forces Pj( j ) Pj(+1)

j
Pj(+1
j +1)
Pj(+ 2
j +1)

produced by the internal forces in

two adjacent elements. (b) The
free body diagram of a small
(j) (j+1)
section of the bar when a (a)
concentrated force, P∗jþ1 , is
present at node j+1. (c) The same
( j +1)
Pj(+1) Pj*+1 Pj +1
j
free body diagram expressed in
terms of the internal forces in
the bar.

(b)

Fx(
j)
(l ( ) )
j
Pj*+1 Fx(
j +1)
( 0)

(c)

so that whenever a concentrated force is not present, the internal force must be continuous.
When a concentrated force is present at a node, then the internal force must exhibit a jump
in value at that node as found from Eq. (9.37).
Now consider the combination of the elements of the stiffness matrix as well as the nodal
forces. The entire bar will involve all the nodal displacements so, if we have M elements,
we will end up with a set of M + 1 linear equations for the nodal displacements
ðU 1 , U 2 ; . . . , U Mþ1 Þ of the form
2

38 9 8 ð1Þ
9 8 ð1Þ 9
K 1, 1 K 1, Mþ1 > U > >
> Q >
> > P1 >
> >
7>> . > > > >
> > > . >
1 1
6 .. .. > > >
> . > > >
>
6 7 > . > > . > > . >
6 . . 7>< . > = >< . >
= > < . > =
6 .. 7 U ðj Þ ðjþ1Þ
6 K 1, jþ1 . K 7 jþ1 ¼ Qjþ1 þ Qjþ1 þ P∗ (9.38)
6 jþ1 , Mþ1 7>> . >
> >
> >
> >
>
jþ1 >
>
6 .. .. 7> .. > > .. > > .. >
4 . . 5>>
>
>
>
>
>
> .
>
> >
> . >
>
: ; >>
: ðM Þ
>
>
;
>
>
: ð M Þ
>
>
;
K Mþ1, 1  K Mþ1, Mþ1 U Mþ1 QMþ1 PMþ1

All the body force nodal terms are known values as well as the applied forces, P∗
j , at
the nodes j ¼ 2, . . ., M (which may be zero or nonzero). The end forces are
ð1Þ ðM Þ
P1 ¼ R1 , PMþ1 ¼ R2 (see Fig. 9.1). These may be unknown reaction forces if the end
nodes are restrained. If there is an applied force at an end, then the end value will just be
that force. We will discuss later how to deal with any unknown reaction forces when
solving the linear system of Eq. (9.38). To show how the stiffness matrix for the entire
bar is assembled, it is convenient to consider a specific case such as a uniform bar (A, E,
and l are all constants) with three elements (and four nodes). Then for each element, the
equilibrium equations in Eq. (9.32b) contribute terms to the entire set of equations given
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 287

EA / l −EA / l 0 0 U1 Q1(1) P1(1) Figure 9.6 The equilibrium equations for a bar
(1) (1) composed of three elements.
−EA / l EA / l 0 0 U2 Q2 P2
element 1 = +
0 0 0 0 U3 0 0
0 0 0 0 U4 0 0

0 0 0 0 U1 0 0
( 2)
0 EA / l −EA / l 0 U2 Q P ( 2)
element 2 = 22 + 2 2
0 −EA / l EA / l 0 U3 Q3( ) P3( )
0 0 0 0 U4 0 0

0 0 0 0 U1 0 0
0 0 0 0 U2 0 0
element 3 = ( 3) +
0 0 EA / l −EA / l U3 Q3 P3(3)
0 0 −EA / l EA / l U4 Q4(3) P4(3)

in Eq. (9.38) that are shown in Fig. 9.6. When we add these three elemental contribu-
tions, we arrive at
2 38 9 8 ð1Þ 9 8 9
EA=l EA=l 0 0 > U1 > >
> Q1 >
> > R1 >
> > > > > >
6 7>>
<
>
>
=
>
>
<
> >
ð2Þ >
= >
< ∗>
>
P2 =
ð1Þ
6 EA=l 2EA=l EA=l 0 7 U2 Q2 þ Q 2
6 7 ¼ þ (9.39)
6 0 EA=l 2EA=l EA=l 7
4 5>>
> U3 >
>
>
>
>
> Q3 þ Q 3 >
ð2Þ ð3Þ
>
>
>
>
> P∗ >
3 >
>
>
: >
; > > >
> > >
0 0 EA=l EA=l U4 : ð3Þ ; : R2 ;
Q4

which shows the diagonal matrix stiffness terms of the individual elements adding when they
overlap in the system stiffness matrix of Eq. (9.38). Although we have used three elements to
illustrate this process, it is the same regardless of the number of elements present. For seven
elements, each of length l, for example, we would find the system stiffness matrix
2 3
1 1 0 0 0 0 0 0
6 1 2 1 0 0 0 0 0 7
6 7
6 0 1 2 1 0 0 0 0 7
6 7
EA 6 6 0 0 1 2 1 0 0 0 7
7
l 66 0 0 0 1 2 1 0 0 7
7
6 0 0 0 0 1 2 1 0 7
6 7
4 0 0 0 0 0 1 2 1 5
0 0 0 0 0 0 1 1

which shows that for a large number of elements, the system stiffness matrix is sparsely
populated with nonzero values only on or near the matrix diagonal. When solving large
systems of linear equations defined by such a matrix, we can take advantage of this behavior
to make the storage of the matrix and the solution process more efficient.
288 Computational Mechanics of Deformable Bodies

Example 9.1 Statically Determinate Bar – Stiffness Method

We have shown how to generate a system of linear equations using finite elements. To
see how to solve such a system, consider the problem shown in Fig. 9.7a where a
uniform bar (A and E are constants) carries both a uniform distributed force q0 over
the middle third of its length and an end force P ¼ q0 L. The bar is rigidly fixed to the
wall at x ¼ 0. This is a statically determinate problem that we can easily solve
analytically. The solution is
8    
> x x
>
> 2 0 1
>
> l l
>
< x  1 x 2 1  
AE x
u x ðx Þ ¼ 3   1   2 (9.40)
q0 l 2 >
> l 2 l 2 l
>
> x 3  
>
> x
: þ 2 3
l 2 l
To solve this problem with finite elements, we will use three elements (see Fig. 9.7b) so that
we can use Eq. (9.39). In this case we have the system of equations
2 38 9 8 9 8 9
1 1 0 0 > > U1 >
> >
> 0 >> >
> R1 > >
6 7>> > >
> > > >
> > >
>
EA 6 1 2 1 0 7< U 2 = < q0 l=2 = < 0 =
6 7
l 6 7 U > ¼ > q l=2 > þ > 0 > (9.41)
4 0 1 2 1 5> >
> 3>
> >
> 0 >
> >
> >
>
>
: >
; > : >
; > : >
;
0 0 1 1 U4 0 q0 l
To correspond to the form of the exact solution let us define normalized nodal displacements
~ j ¼ AE U j so that Eq. (9.41) becomes simply
U
q0 l 2

Figure 9.7 (a) A bar loaded by a uniform

distributed force/unit length, q0 , over the middle q0
third of its length and a concentrated end force
P ¼ q0 l. (b) A finite element model of the bar
using three elements showing the P = q0l
l l l
nodal displacements.
(a)
U1 U2 U3 U4
R1

q0l
q0

(b)
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 289

2 38 9 8 9
1 1 0 0 > ~ 1 > > R1 =q0 l >
U
>
<~ = <> > >
=
6 1 2 1 0 7 1=2
6 7 U2 ¼ (9.42)
4 0 1 5
2 1 > ~ > >
U 1=2 >
: ~3>
> ; >
: >
;
0 0 1 1 U4 1
Because the problem is statically determinate, applying equilibrium to the entire bar gives
R1 =q0 l ¼ 2, so it is tempting to place this value into Eq. (9.42) and try to solve for the
displacements. However, this will fail since the determinant of the stiffness matrix is zero so
the matrix does not have an inverse. For example, if we evaluate the determinant in
MATLAB® we find:
Kg ¼ [1 -1 0 0; -1 2 -1 0; 0 -1 2 -1; 0 0 -1 1]
Kg ¼ 1 -1 0 0
-1 2 -1 0
0 -1 2 -1
0 0 -1 1
det(Kg)
ans ¼ 0

We cannot solve for the displacements from Eq. (9.42) because we have not used the
boundary condition of zero displacement at x ¼ 0. Without this condition, the bar is free
to have an arbitrary rigid body displacement and we cannot hope to uniquely solve for the
displacements. We can also see this in MATLAB® by letting the nodal displacements all
have some arbitrary constant value and multiplying the system stiffness matrix by those
constants:
u ¼ [2 ; 2; 2; 2];
Kg*u
ans ¼ 0
0
0
0

These constant displacements produce zero forces so they could always be added to the
actual solution without affecting the conditions of equilibrium. We can satisfy the boundary
~ 1 ¼ 0, by simply eliminating the first row
condition of zero displacement at the wall, i.e., U
and column of the stiffness matrix in Eq. (9.42). The remaining equations are then
2 38 9 8 9
2 1 0 < U ~ 2 = < 1=2 =
4 1 2 1 5 U ~ ¼ 1=2 (9.43)
: ~3; : ;
0 1 1 U4 1
which we can solve for the remaining displacements. In MATLAB®, for example, we can
obtain the solution as follows:
290 Computational Mechanics of Deformable Bodies

K ¼ Kg; % define a copy of the system stiffness

% matrix that we will modify
K(1,:) ¼ [ ] % eliminate the first row
K ¼ -1 2 -1 0
0 -1 2 -1
0 0 -1 1
K(:,1) ¼ [ ] % eliminate the first column
K ¼ 2 -1 0
-1 2 -1
0 -1 1
P= [1/2 1/2 1]’ % define a force column vector {P}
P ¼ 0.5000
0.5000
1.0000
u ¼ K\P % solve the system of equations [K]{u} ¼ {P}
u ¼ 2.0000
3.5000
4.5000

so that the normalized displacements are U ~ 2 ¼ 2, U

~ 3 ¼ 3:5, U
~ 4 ¼ 4:5: If we multiply the
original system stiffness matrix by these displacements together with the displacement
U~ 1 ¼ 0, we can obtain the normalized reaction force at x ¼ 0 (as well as the original nodal
forces):
ug ¼ [0 ; u] % make a column vector of all the nodal
% displacements
ug ¼ 0
2.0000
3.5000
4.5000
Kg*ug % multiply these displacements by the system
% stiffness matrix
ans ¼ -2.0000
0.5000
0.5000
1.0000

which indeed gives the correct normalized reaction force. The exact displacement of the bar
together with the finite element solution are plotted in Fig. 9.8. It can be seen that the linear
displacement approximation for each element captures the solution quite well for this
problem even with only three elements. This was possible since the exact solution of
Eq. (9.40) has linear behavior in the first and third elements.
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 291

4.5 Figure 9.8 The displacement of the bar shown in

4 Fig. 9.7a. Dashed line – exact solution, solid line –
finite element solution.
3.5
3
AE
u 2.5
q0l 2
2
1.5
1
0.5
0
0 0.5 1 1.5 2 2.5 3

x/l

To obtain the internal axial forces and stresses in the element in the general case we have
X
jþ1
F x ðξ Þ ¼ E ðξ ÞAðξ Þexx ðξ Þ ¼ E ðξ ÞAðξ Þ Bm ðξ Þ U m
m¼j
(9.44)
X
jþ1
σ xx ðξ Þ ¼ E ðξ Þexx ðξ Þ ¼ E ðξ Þ Bm ðξ Þ U m
m¼j

but where in our previous example all these terms are constants. We will leave the determin-
ation of the stresses to the reader.

In cases where the body forces are zero so that all the applied loads are due only to
concentrated forces and the product EA is piecewise constant, the elements can be chosen
such that the exact internal forces are constants in the elements and hence the strains will be
constants (as assumed in our model with a linear shape functions for the displacement).
Thus, a stiffness-based finite element model can reproduce an exact solution for problems
such as the stepped bar shown in Fig. 9.9a. [Note: to be more precise, by “exact” we mean
the solution will be identical to a solution that satisfies the strength of materials governing
equations. However, remember those equations themselves are approximations to the
underlying equations of elasticity.] As shown in Chapter 8 (Section 8.4), the torsion of
circular shafts satisfy the same basic equations as axial loads if we make the replacements
F x ! M x , EA ! GJ, u ! ϕ, qx ! tx , where M x is the internal axial torque, GJ is the shear
modulus times the polar area moment of the cross-section, ϕ is the angle of twist, and tx
is the distributed torque/unit length. Thus, for tx ¼ 0 and when GJ has piecewise constant
values for the shaft, using linear shape functions for the twist can also produce
exact solutions. An example is shown in Fig. 9.9b. We will show in Section 9.6 that
stiffness-based finite elements can also produce exact solutions for beams containing only
292 Computational Mechanics of Deformable Bodies

concentrated forces (or moments) and where EI is piecewise constant such as the beam of
Fig. 9.9c. All the examples shown in Fig. 9.9 are statically determinate problems, but this is
not necessary and similar statically indeterminate structures also can have exact stiffness-
based finite element solutions. Even more complex structures may have exact finite element
solutions. A 2-D truss is a set of axial-load members (whose own weights are neglected)
connected by pins and where there are only concentrated forces applied at those pins (see
Fig. 9.10a). The internal force in each truss member is a constant so that each member can
be treated as a single constant-strain element in a stiffness-based finite element solution.
Similarly, a frame structure consisting of beam and truss parts and carrying only concen-
trated forces (see, for example, Fig. 9.10b) can be decomposed into finite elements and
solved exactly. We have pointed out these examples because, although the real power of the
stiffness-based finite element method is to find approximate numerical solutions to very
complex 3-D geometries and loading situations, the method can also find exact solutions to
many of the problems considered in strength of materials, a fact that may not be appreci-
ated. There may also be some instances when exact solutions can be produced in cases with
distributed loads.

Figure 9.9 Cases where a stiffness-based finite element

solution can be formulated to produce an “exact”
solution. (a) An axial-load problem. (b) A torsion
problem. (c) A beam problem.

(a) (b)

(c)

Figure 9.10 (a) A truss structure. (b) A frame

structure. Both of these structures can be solved
exactly by the stiffness-based finite
element method.

(a) (b)
 9.2 Stiffness-Based Finite Elements for Axial-Load Problems 293

Example 9.2 Statically Indeterminate Bar – Stiffness Method

Consider now the statically indeterminate problem of Fig. 9.11a, where a uniform bar is
fixed at both ends and subjected to a uniform distributed force/unit length, q0 , over its entire
length. We will use three constant-strain elements to solve this problem (see Fig. 9.11b) so
that we can use Eq. (9.39), which in this case becomes
2 38 9 8 9 8 9
1 1 0 0 > > U1 >
> >
> q0 l=2 >
> > > R1 > >
EA 6 1 2 1 0 7< U 2 = < q0 l = < 0 =
6 7 ¼ þ (9.45)
l 4 0 1 2 1 5> > U > > ql > > 0 >
: 3> ; > : 0 > ; > : > ;
0 0 1 1 U4 q0 l=2 R2


which in terms of the normalized displacement U ~ j ¼ EA=q0 l 2 U j becomes
2 38 9 8 9 8 9 8 9
1 1 0 0 > U~ 1 > > 1=2 > > R1 =q0 l > > R1 =q0 l þ 1=2 >
>
< >
= >
< >
= >
< >
= >
< >
=
6 1 2 1 0 7 U ~2 1 0 1
6 7 ¼ þ ¼ (9.46)
4 0 1 2 1 5> U ~ > > 1 > > 0 > > 1 >
: ~3>
> ; > : >
; > : >
; > : >
;
0 0 1 1 U4 1=2 R2 =q0 l R2 =q0 l þ 1=2

To solve these equations, we must apply the displacement boundary conditions U ~1 ¼ U ~4 ¼ 0

by eliminating the first and fourth rows and columns of the stiffness matrix to arrive at
 
2 1 U~2 1
~3 ¼ (9.47)
1 2 U 1

which has the solution U ~2 ¼ U~ 3 ¼ 1 or, equivalently, U 2 ¼ U 3 ¼ q0 L2 =9EA, where L is the

total length of the bar. Multiplying the original stiffness matrix by these known nodal displace-
ments and the boundary values U ~1 ¼ U ~ 4 ¼ 0 we then obtain Rj =q0 l þ 1=2 ¼ 1 ðj ¼ 1; 2Þ.
®
These steps in MATLAB are:
Kg ¼ [ 1 -1 0 0 ; -1 2 -1 0; 0 -1 2 -1; 0 0 -1 1];
% stiffness matrix
K ¼ Kg; % copy of the stiffness matrix
% which we will modify
K(4,:) ¼ [ ]; K(:,4) ¼ [ ]; % delete largest column and row
% first so that when we apply the next boundary condition
K(1,:) ¼ [ ]; K(:,1) ¼ [ ]; % its row and column number is
% not changed
P= [1; 1]; % define a force column vector {P}
u =K\P % solve the system of equations [K]{u} ¼ {P}
u ¼ 1.0000
1.0000
ug ¼ [0; 1; 1; 0]; % define a column vector of all
% the nodal displacements
294 Computational Mechanics of Deformable Bodies

Figure 9.11 (a) A uniform q0

distributed force/unit length, q0 ,
acting on a uniform bar that is L
fixed at both ends. (b) A finite
element model for the bar (a)
using three elements and
U1 U2 U3 U4
showing the nodal displacements
U j ( j ¼ 1, 2, 3, 4) and end q0
R1 R2
reactions Rj ( j ¼ 1, 2).

l l l
(b)

Figure 9.12 Displacement of the bar for the axial-load 1.2

problem shown in Fig. 9.11a. Exact solution – dashed
line, constant-strain finite element solution for three 1
elements – solid line.
0.8

EA 0.6
u
2 x
q0l
0.4

0.2

0
0 0.5 1 1.5 2 2.5 3

x/l

Kg*ug % multiply these displacements by

% the original stiffness matrix
ans ¼ -1
1
1
-1

Solving for the reaction forces explicitly, we find R1 ¼ R2 ¼ 3q0 l=2 ¼ q0 L=2. These
reaction forces obviously are in equilibrium with the total force, q0 L, generated on the bar
by the distributed load. We can compare these results with the exact solution for the
displacement and stress, which are given by
 9.3 Force-Based Finite Elements for Axial-Load Problems 295

1.5 Figure 9.13 Stress in the bar for the axial-load

problem shown in Fig. 9.11a. Exact solution – dashed
1 line, constant-strain finite element solution for three
elements – solid line.
0.5
A
xx
q0l 0

– 0.5

–1

– 1.5
0 0.5 1 1.5 2 2.5 3

x/l

EA 3 x 1 x 2
u ðxÞ ¼
2 x

q0 l 2l 2 l
EA ~2 ¼ 1
ux ðl Þ ¼ U
q0 l 2
(9.48)
EA ~3 ¼ 1
ux ð2l Þ ¼ U
q0 l 2
A 3 x
σ xx ðxÞ ¼ 
q0 l 2 l

showing that the nodal displacement values obtained are in fact exact. Figure 9.12 shows
both the exact and finite element solutions for the displacements while Fig. 9.13 com-
pares the stresses. Typically, the stiffness-based finite element approximation for the
displacements is better than the approximation for stresses. Obviously, it will require
many more constant strain elements than used here to obtain an adequate approximation
for the stress.

9.3 Force-Based Finite Elements for Axial-Load Problems

The stiffness-based finite element method has been highly developed and is widely used.
However, there is a stress-based or force-based finite element approach that is an important
alternative method we will consider in this section. Again, we will use axial-load problems in
order to discuss the details. This method will employ both the principle of virtual work and
296 Computational Mechanics of Deformable Bodies

the principle of complementary virtual work. We will write the virtual work principle (see
Eq. (9.8), for example) for a bar element as
ðl ðl
F x ðξ Þδð∂ux ðξ Þ=∂xÞdξ ¼ qx ðξ Þδux ðξ Þdξ þ Pjþ1 δU jþ1 þ Pj δU j (9.49)
0 0

and the principle of complementary virtual work as (see Eq. (9.16))

ðl
1
F x ðξ ÞδF x ðξ Þdξ ¼ U jþ1 δF x ðl Þ  U j δF x ð0Þ
E ðξ ÞAðξ Þ (9.50)
0
¼ U jþ1 δPjþ1 þ U j δPj

Also, recall in deriving this form of the principle of complementary virtual work, the virtual
variation of the internal force was required to satisfy the homogeneous equilibrium equa-
tion, Eq. (9.15), i.e.,
d
ðδF x Þ ¼ 0 (9.51)
dξ
so that any approximation we choose for this variation must also satisfy Eq. (9.51). The
exact internal force itself satisfies
dF x ðξ Þ
¼ qx ðξ Þ (9.52)
dξ
Both virtual work principles have been written in terms of the internal force in the element
and the displacement only. Although we will call this a force-based method since the internal
forces will be the primary quantities we will solve for, Eq. (9.49) and Eq. (9.50) show that the
method will also need to consider the displacements. In order to link the displacements and
forces, strains and stress–strain relations will also be involved. We can satisfy Eq. (9.51) if we
assume that the approximation used for the internal force itself also satisfies the homoge-
neous equilibrium equation. For this problem we need only assume that the internal force is
a constant in each element, i.e., F x ðξ Þ ¼ F j , since a constant force satisfies the homogeneous
equilibrium equation identically. For the displacement we will assume the same linear
variation used in the stiffness-based method so the displacement and strain will again be
ux ðξ Þ ¼ ½H fU g
(9.53)
exx ðξ Þ ¼ ½BfU g

where the shape functions are again given by Eq. (9.19) and Eq. (9.24).
In explaining the force-based method, we will want to use forms that will correspond to the
more general problems we will discuss later. In this axial-load case, the force, displacement,
and strain in the bar are all scalars, whereas they will be vectors in more general problems.
Similarly, while we have simply a scalar relationship between internal strain and force,
 9.3 Force-Based Finite Elements for Axial-Load Problems 297

exx ðξ Þ ¼ F x ðξ Þ=E ðξ ÞAðξ Þ, in the more general case we will have a matrix relating the vector
strains and displacements. In addition, the scalar distributed load in the general case will
become a vector. Thus, in all our following discussions we will use the following expressions:
fF x ðξ Þg  F x ðξ Þ ¼ ½N ðξ ÞfF g
fux ðξ Þg  ux ðξ Þ ¼ ½H ðξ ÞfU g
fexx ðξ Þg  exx ðξ Þ ¼ ½Bðξ ÞfU g (9.54)
fqx ðξ Þg  qx ðξ Þ
fexx ðξ Þg ¼ ½1=EAfF x ðξ Þg

where fF g is a vector of force coefficients while fU g is a vector of nodal displacements. We

will use these matrix–vector expressions with the understanding that, in the axial-load case
with the choice of linear and constant shape functions for the displacement and force, we will
evaluate these terms explicitly as
½N ðξ Þ ¼ 1
fF g ¼ F j
½H ðξ Þ ¼ ½1  ξ=l ξ=l  (9.55)
½Bðξ Þ ¼ ½1=l 1=l 
½1=EA ¼ 1=E ðξ ÞAðξ Þ

The notation of Eq. (9.54) will then carry over to more general problems with other choices
of shape functions and will also allow us to see more clearly the general structure of the
terms involved in a problem. In more general cases, the requirement that the variation of the
force expressions satisfy homogeneous equilibrium equations will be satisfied (as done here
in the axial-load problems) by choosing a matrix of force shape functions, ½N ðξ Þ, that satisfy
the homogeneous equations of those more general cases.
[Note: the vector of forces, fF g, that define the internal forces, unlike the displacements,
fU g, are not necessarily values defined at nodes, so our previous convention that subscripts
define nodal values is only true for the displacements. For the axial-load case, for example,
there is only one internal force, F j , for the ( j)th element (see Fig. 9.14) and two nodal


displacements U j , U jþ1 . In other problems there can be more
Uj than one of these internal forces (or stresses) and more than
U j +1
two displacements at a node. The number, K, of F k values for
an element can be considered to be the number of assumed
Fj Fj force degrees of freedom for that element, just as the number,
Pj Pj +1 M, of U m values is the number of nodal displacement degrees

x
Figure 9.14 An element showing the internal force, F j , acting in
l the element.
298 Computational Mechanics of Deformable Bodies

of freedom. For the axial-load case we are considering, K ¼ 1, M ¼ 2.]

In these generalized matrix–vector forms, then, the principles of virtual work and comple-
mentary virtual work for an element can be written as
ðl ðl
fδexx ðξ ÞgT fF x ðξ Þgdξ ¼ fδux ðξ ÞgT fqx ðξ Þgdξ þ fδU gT fPg
0 0
(9.56)
ðl
fδF x ðξ ÞgT ½1=EAfF x ðξ Þgdξ ¼ fδPgT fU g
0

where explicitly in the axial-load case we have

( ) ( )
Uj Pj
fU g ¼ , fPg ¼
U jþ1 Pjþ1
( ) ( ) (9.57)
δU j δPj
fδU g ¼ , fδPg ¼
δU jþ1 δPjþ1

Placing the matrix–vector expressions of Eq. (9.54) into Eq. (9.56) then gives
ðl ðl
T T T
fδU g ½B ½N dξ fF g ¼ fδU g ½H T fqx gdξ þ fδU gT fPg (9.58)
0 0

and
ðl
fδF g T
½N T ½1=EA½N dξ fF g ¼ fδPgT fU g (9.59)
0

Consider first the virtual work principle, Eq. (9.58). Since the fδU gT in Eq. (9.58) is
arbitrary, we must have the force vector fF g satisfy the equilibrium equations
~ fF g ¼ fQg þ fPg
E (9.60)
~ for the element is defined as
where the equilibrium matrix, E
ðl
~ ¼ ½BT ½N dξ
E (9.61)
0

and the distributed force term, which is the same as in the stiffness-based method, is
ðl
fQg ¼ ½H ðξ ÞT fqx ðξ Þgdξ (9.62)
0
 9.3 Force-Based Finite Elements for Axial-Load Problems 299

If we use the specific values for our axial-load problem, the equilibrium matrix
becomes

~ ¼ 1
E (9.63)
1

and the equilibrium equation for the element, Eq. (9.60), is

 
1 Qj Pj
Fj ¼ þ (9.64)
1 Qjþ1 Pjþ1

where (recall Eq. (9.34b)) for qx given in terms of a global x-coordinate

8 9
>
> ðl >
>
>
>
 >
>
>
> q x þ ξ ð 1  ξ=l Þdξ >
>
 >
<
x j >
=
Qj
fQg ¼ ¼ 0
(9.65)
Qjþ1 >
> ðl >
>
>
>
 >
>
>
> q x þ ξ ðξ=l Þdξ >
>
>
: x j >
;
0

Now, consider the complementary virtual work principle, Eq. (9.59). The virtual vari-
ations of the internal force must satisfy the homogeneous equilibrium equations (no distrib-
uted force term) so from Eq. (9.60) we must have
~ fδF g ¼ fδPg
E (9.66)

Placing the transpose of this relationship into Eq. (9.59) and using the fact that the resulting
equation must be satisfied for all variations fδF gT , we obtain
~
½G fF g ¼ E
T
fU g (9.67)

in terms of a flexibility matrix, ½G , given by

ðl
½G  ¼ ½N T ½1=EA½N dξ (9.68)
0

If we define a set of discrete generalized strains as fΔg ¼ ½G fF g, then we see the virtual
work and complementary virtual work principles yield
~ fF g ¼ fQg þ fPg
E
(9.69)
~
fΔg ¼ ½G fF g ¼ E
T
fU g

which are just the discrete equilibrium and strain–displacement relations for an element in
the force-based finite element method. Explicitly, for the axial-load case and a uniform
element, where E and A are constants, we have
300 Computational Mechanics of Deformable Bodies

" # ( ) ( )
1 Qj Pj
Fj ¼ þ
1 Qjþ1 Pjþ1
  ( ) (9.70)
l Uj
Δj ¼ F j ¼ ½1 1 
EA U jþ1

where the generalized strain in the bar, Δj , is just the elongation of the element.
Having these results for a single element, we now want to assemble a set of elements
and solve axial-load problems. For the equilibrium equations we find, using our
previous example of three elements and again letting the superscript ( j) denote the
( j)th element:
8 9 8 9 8 9 8R 9
2 3 > Q1
ð1Þ
> >
ð1Þ
P1 > > Q1
ð1Þ
> >
> 1>>
> > > > > > > >
1 0 0 8 9 > >
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> ∗>
>
6 1 1 0 7 < F 1 = < ð1 Þ
Q2 þ Q1
ð 2Þ = < ð 1Þ
P2 þ P 1
ð 2Þ = < ð 1Þ
Q2 þ Q1
ð2 Þ = < P2 >
=
6 7 F2 ¼ þ ¼ þ
4 0 1 1 5: ; > ð2Þ
F3 > Q2 þ Qð13Þ >
> > >
> > Pð22Þ þ Pð13Þ >
> > >
> > Qð22Þ þ Qð13Þ >
> > >
> > P∗
>
>
>
3 >
>
> >
> >
> >
> >
> >
> >
> >
0 0 1 : ð3Þ ; : ð3Þ ; : ð3Þ ; : > > >
;
Q2 P2 Q2 R2
(9.71)

where ðR1 , R2 Þ are again the end forces acting on the bar and P∗ j are known concentrated
forces acting at the inner nodes. In this case the assembly process is very simple since we are
just adding columns, suitably displaced, to the complete equilibrium matrix. The generalized
strain–displacement equations similarly become
8 9
8 9 2 3> U 1 >
< Δ1 = 1 1 0 0 < > >
=
4 5 U2
Δ2 ¼ 0 1 1 0 (9.72)
: ; > U3 >
Δ3 0 0 1 1 > : >
;
U4

where now we are adding rows to form the matrix of Eq. (9.72), which is just the transpose of
the complete equilibrium matrix for the entire bar. In matrix–vector form we have the same
form as Eq. (9.69), which was for a single element, namely
~ fF g ¼ fPg
E
(9.73)
~
fΔg ¼ E
T
fU g
~ is the complete equilibrium matrix for the body.
where now E

Example 9.3 Statically Determinate Bar – Force Method

As a very simple example of the use of the force-based method for a statically determinate
problem, consider the bar shown in Fig. 9.15a. In this case we only need the equilibrium
equations to solve for the forces. Breaking the bar up into three elements of length l each, we
can use the assembled system of Eq. (9.71) for this case and obtain
 9.3 Force-Based Finite Elements for Axial-Load Problems 301

Figure 9.15 (a) A statically determinate uniform bar

3P 2P 4P
carrying a set of concentrated forces. (b) A statically
indeterminate uniform bar carrying a set of
concentrated forces.
l l l
(a) 2 3 8 9
1 0 0 8 9 > > R1 >
>
6 1 1 0 7< F 1 = < 3P =
6 7 F2 ¼ (9.74)
3P 2P 4 0 1 1 5: ; > 2P >
F3 >
: >
;
0 0 1 4P

l l where R1 is the reaction at the fixed support. The

l
internal forces in the bars themselves must be in
(b) equilibrium with the known applied forces so we
can delete the first equation in Eq. (9.74) and solve
the remaining system:
2 38 9 8 9
1 1 0 < F 1 = < 3P =
40 1 1 5 F 2 ¼ 2P (9.75)
: ; : ;
0 0 1 F3 4P

We can symbolically solve this system easily in MATLAB®:

E ¼ [-1 0 0;1 -1 0;0 1 -1; 0 0 1]; % define the full
% equilibrium matrix
Er ¼ E;
Er(1,:) ¼ [ ]; % eliminate the first row in E to obtain a
% reduced matrix
syms P
Pr ¼ [3*P;-2*P; 4*P]; % the symbolic force vector for the
% reduced system
F ¼ inv(Er)*Pr % solve for the internal forces (can
% also use F ¼ Er\Pr)
F ¼ 5*P
2*P
4*P

So we see F 1 ¼ 5P, F 2 ¼ 2P, F 3 ¼ 4P. If we multiply the original equilibrium matrix by

these internal forces, we find
E*F % obtain reaction force and original loads
ans ¼ -5*P
3*P
-2*P
4*P
302 Computational Mechanics of Deformable Bodies

Figure 9.16 Internal force distribution for the bar of 5P

Fig. 9.15a.
4P

internal force
3P

2P

0
0 1 2 3
x/l

which shows that R1 ¼ 5P. We can also obtain this same value for the reaction force
directly from equilibrium of the entire bar since this is a statically determinate problem.
Figure 9.16 shows the internal force distribution in the bar. We also see that there are
downward jumps in the internal force at the locations of the applied forces that are equal to
the value of those forces in the x-direction. This is simply a consequence of applying
equilibrium in a small region about those concentrated forces, as is often discussed in
elementary strength of materials courses and which we have also used previously in the
stiffness-based approach to relate the behavior of the internal forces to the applied concen-
trated forces (see Fig. 9.5c and Eq. (9.37)).

Statically determinate bar problems are, of course, very simple and can be solved directly
by application of the equilibrium equations. Statically indeterminate bar problems are also
simple but require more than equilibrium equations for their solution, as we will see in the
next example.

Example 9.4 Statically Indeterminate Bar – Force Method and Force-Based

Stiffness Method
To see the application of a force-based finite element approach to a statically indeterminate
problem, consider the bar shown in Fig. 9.15b. Since we again have three elements, our
equilibrium equations become
2 3 8 9
1 0 0 8 9 > > R1 >>
6 1 1 0 7< F 1 = < 3P =
6 7 F2 ¼ (9.76)
4 0 1 1 5: ; > 2P >
F3 >
: >
;
0 0 1 R2

where there are now two unknown reaction forces ðR1 , R2 Þ. There are a total of four
equations and five unknowns, and the problem is statically indeterminate. Even if we
 9.3 Force-Based Finite Elements for Axial-Load Problems 303

eliminate the two equations for the unknown end reactions, we have two equations in three
unknowns:
8 9
< F1 = 
1 1 0 3P
F ¼ (9.77)
0 1 1 : 2 ; 2P
F3
~ f fF g ¼ fPg to indicate that we are working with a reduced “free”
which we will write as E
equilibrium matrix associated with free nodes, i.e., those without constraint forces. Thus,
regardless of the unknown forces we choose, there is always one more unknown force than
equilibrium equations, so we say the problem is indeterminate of degree one, requiring one
more equation in order to solve. The additional equation we need we can get from the
strain–displacement relation, fΔg ¼ E~ T fU g: which in this case is again (see Eq. (9.72)):
8 9
8 9 2 3> U 1 >
< Δ1 = 1 1 0 0 < > >
=
4 5 U2
Δ2 ¼ 0 1 1 0 (9.78)
: ; > U3 >
Δ3 0 0 1 1 > : >
;
U4

However, before we can use this relationship, we must apply the displacement boundary
conditions U 1 ¼ U 4 ¼ 0, which we can do by simply eliminating the first and fourth rows of
the matrix in Eq. (9.78), to obtain
8 9 2 3
< Δ1 = 1 0 
U2
Δ ¼ 4 1 1 5 (9.79)
: 2; U3
Δ3 0 1
T 
~f
which is still in the form fΔg ¼ E U f but where now the unknown reaction forces have
been eliminated from the equilibrium equations and the corresponding specified displacements
have been eliminated from the strain–displacement relations so that only the free displace-
ments are present. In our example we have the three equations
Δ1 ¼ U 2
Δ2 ¼ U 3  U 2 (9.80)
Δ3 ¼ U 3

These are nothing more than the relationships between the elongations of the elements and their
end displacements when the displacement boundary conditions are satisfied, a result we could
have written down by inspection. Since there are three elongations written in terms of only two
displacements, there must be one compatibility equation that relates those elongations. We can
use this compatibility equation to find the additional force equation(s) we need in a statically
indeterminate problem in the following fashion. Because there are fewer free displacements than
strains, we can always take combinations of the strain–displacement relations and eliminate the
displacements to arrive at a set of compatibility equations, which we write as
~ fΔg ¼ 0
S (9.81)
304 Computational Mechanics of Deformable Bodies

where S ~ is a compatibility matrix relating the generalized strains fΔg. Applying this compati-
 
bility matrix to the strain–displacement relations fΔg ¼ E ~ f T U f we have
 
~ E
S ~ f T Uf ¼ 0 (9.82)
 
but here U f are all free (unconstrained) displacements so that the only way we can satisfy
Eq. (9.82) is if
~ E
S ~f T
¼ E ~
~f S T
¼0 (9.83)

where we have taken the transpose to obtain the second relation seen in Eq. (9.83). Later, we
will see how Eq. (9.83) is part of the determination of the compatibility matrix in general. In
the present case being considered, we can obtain the compatibility matrix by direct elimin-
ation. For example, simply by adding the three equations in Eq. (9.80) we find the compati-
bility equation
Δ1 þ Δ 2 þ Δ3 ¼ 0 (9.84)
~ fΔg ¼ 0, where here
which is our compatibility equation S
~ ¼ ½1 1 1
S (9.85)

It is easy to verify that

2 3
1 0
~ E
S ~f T
¼ ½ 1 1 1 4 1 1 5 ¼ ½ 0 0  (9.86)
0 1

Physically, the compatibility equation, Eq. (9.84), says that the total elongation of the bar is
zero, a result we could have written down by inspection.
To turn the compatibility equation into the additional equation(s) we need to solve the
problem, we relate the unknown forces to the generalized strains through the flexibility
  h i 
matrices. Recall, for each element we had ΔðjÞ ¼ G ðj Þ F ðj Þ so we can combine these
elements into a similar relationship for the entire system as
fΔg ¼ ½G fF g (9.87)

where, for example, for M elements we have

2 h ð1Þ i 3
G  0 0
6 h i 7
6 .. 7
6 . G ð2Þ
0 7
½G  ¼ 6
6
7
7 (9.88)
6 .. 7
4 0 . h 0 i5
0 0  G ðM Þ

which is simply the concatenation of the element flexibility matrices along the main diagonal
of the ½G  matrix. In the case at hand, the element flexibility matrices are just identical scalars
 9.3 Force-Based Finite Elements for Axial-Load Problems 305

h i
for each element, namely G ðjÞ ¼ l ðj Þ =E ðjÞ AðjÞ ¼ l=EA, and there are three elements so we
have
2 3
1 0 0
l 4
½G  ¼ 0 1 05 (9.89)
EA
0 0 1

Placing the strain–force relations of Eq. (9.87) into the compatibility equations, Eq. (9.81) we find
~ ½G fF g ¼ 0
S (9.90)

which, in general, are the additional force equations required and which can be added to the
equilibrium equations E ~ f fF g ¼ fPg to obtain a system of equations that can be solved for
the forces, namely,
2 3 8 9
~f
E < fPg =
4  5fF g ¼  (9.91)
~ ½G  : ;
S f0g

In our case

~ ½G fF g ¼ l ðF 1 þ F 2 þ F 3 Þ ¼ 0
S (9.92)
EA
and we can eliminate the common flexibility term. The augmented equilibrium equations
then become the solvable system:
2 38 9 8 9
1 1 0 < F 1 = < 3P =
4 0 1 1 5 F 2 ¼ 2P (9.93)
: ; : ;
1 1 1 F3 0

The solution is F 1 ¼ 4P=3, F 2 ¼ 5P=3, F 3 ¼ P=3. We can go back to Eq. (9.80) and find
the displacements as U 2 ¼ 4Pl=3AE , U 3 ¼ Pl=3AE.
We can also combine the equilibrium and strain–displacement equations of the force-
based method to solve directly for the displacements instead of the forces and produce a
displacement-based finite element method. From Eq. (9.73) and Eq. (9.87), for the entire
bar, for example, we have
~ fF g ¼ fPg
E
(9.94)
~ T fU g
½G fF g ¼ E

[Note: here the equilibrium matrix is the original equilibrium matrix seen in Eq. (9.76).]
Solving for the force vector and placing it into the equilibrium equation then gives

E ~
~ ½G 1 E T
fU g ¼ fPg (9.95)

where E ~ is given in Eq. (9.76), ½G 1 ¼ ðEA=l Þ½I , and where ½I  is the identity matrix so that
~ ½G 1 E
E ~ T ¼ ðEA=l Þ E ~ E ~ T . Carrying out this matrix multiplication in MATLAB® we
find (omitting the EA=l term):
306 Computational Mechanics of Deformable Bodies

E ¼ [ -1 0 0; 1 -1 0; 0 1 -1; 0 0 1];
E*E’
ans ¼ 1 -1 0 0
-1 2 -1 0
0 -1 2 -1
0 0 -1 1

so we recognize that we have obtained the stiffness matrix for the system, i.e.,
2 3
1 1 0 0
EA 6 1 2 1 0 7
~ ½G 1 E
E ~ ¼ ½K  ¼
T
6 7 (9.96)
l 4 0 1 2 1 5
0 0 1 1

and our system of equations is explicitly

2 38 9 8 9
1 1 0 0 >
> U1 >
> >
> R1 >
>
EA 6 7< = < =
6 1 2 1 0 7 U 2 ¼ 3P
(9.97)
l 4 0 1 2 1 > 5 U > > 2P >
: 3>
> ; >
: >
;
0 0 1 1 U4 R2

which is identical to the equations found in the stiffness-based finite element method.
Applying the displacement boundary conditions and solving the remaining system, as done
before, we find the same values previously obtained with the force-based method, namely
U 2 ¼ 4Pl=3AE and U 3 ¼ Pl=3AE.

Let us finish this discussion of the force-based method by considering an example where
distributed forces are present

Example 9.5 Bar with Distributed Loads – Force Method

Specifically, let us examine the problem of Fig. 9.11a, again using three elements. In this case
we can use Eq. (9.71). From our discussion of the stiffness-based method, we have already
evaluated the distributed force terms so we find
2 3 8 9 8 9
1 0 0 8 9 > > q l=2 >
> > > R1 >
>
6 1 1 0 7< F 1 = < q0 l = < 0 =
0
6 7 F2 ¼ þ (9.98)
4 0 1 1 5: ; > ql > > 0 >
F3 : 0 >
> ; > : > ;
0 0 1 q0 l=2 R2

and we can eliminate the first and last equations to obtain

8 9
< F1 = 
1 1 0 q0 l
F ¼ (9.99)
0 1 1 : 2 ; q0 l
F3
 9.3 Force-Based Finite Elements for Axial-Load Problems 307

The compatibility equation written in terms of the forces for three uniform elements was given
in Eq. (9.92) so we can use that equation here to augment Eq. (9.99) and arrive at the system
2 38 9 8 9
1 1 0 < F 1 = < q0 l =
4 0 1 1 5 F 2 ¼ q0 l (9.100)
: ; : ;
1 1 1 F3 0

which has the solution F 1 ¼ q0 l, F 2 ¼ 0, F 3 ¼ q0 l. It is easy to see that the nodal
displacements are just U 1 ¼ U 4 ¼ 0, U 2 ¼ U 3 ¼ q0 l 2 =EA ¼ q0 L2 =9EA. All of these results
are identical to those obtained with the stiffness-based method. This should not be surprising
because, in both cases, the forces and displacements assumed were the same (constant force
and strain, linear variation of the displacements in each element).

9.3.1 A Summary and Discussion

The force-based finite element method can be summarized as follows: For statically indeter-
minate problems, using a combination of the principles of virtual work and complementary
virtual work we can obtain discrete versions of the equilibrium and strain–displacement
equations for the entire system written as
E~ f fF g ¼ fQg þ fPg
  (9.101)
f Δg ¼ E ~ f T Uf

where E ~ f is a “free” equilibrium matrix (i.e., one that does not involve unknown reaction
forces) and where a set of generalized strains are related to the free (i.e., unknown) nodal
 
displacements U f through the transpose of the same “free” equilibrium matrix. The
internal forces are related to these generalized strains through a flexibility matrix, ½G :
fΔg ¼ ½G fF g (9.102)

By eliminating the displacements from the strain–displacement equations (either by hand or

by a more automatic procedure to be discussed shortly), we can find a compatibility matrix
~ fΔg ¼ 0 and, placing Eq. (9.102) into these compatibil-
for the generalized strains where S
ity equations, we obtain
~ ½G fF g ¼ 0
S (9.103)

which can be appended to the equilibrium equations to arrive at a system that can be solved
for the unknown forces:
2 3 8 9
~f
E < fQg þ fPg =
4  5fF g ¼  (9.104)
~ ½G  : ;
S f0g
308 Computational Mechanics of Deformable Bodies

Equation (9.104) describes the final equations that must be solved in the force-based
method. It requires the “free” equilibrium matrix, E ~ f , the compatibility matrix,
~ , the flexibility matrix, ½G , and the equivalent distributed force vector, fQg, as
S
well as the known concentrated forces, fPg. These equations are the discrete equiva-
lents of the Matrix–Vector 3-D Governing Set 3 equations described in Chapter 6
(see Eq. (6.57)):
½ℒE fσ g þ ff g ¼ 0
(9.105)
½ℒS ½Dfσ g ¼ 0

which were obtained by combining the equilibrium equations for the stresses with the
compatibility equations and the stress–strain relations:
½ℒE fσ g þ ff g ¼ 0
½ℒS feg ¼ 0 (9.106)
feg ¼ ½Dfσ g

while the final equations, Eq. (9.104), of the discrete force-based method came from
combining
~ f fF g ¼ fQg þ fPg
E
~ fΔg ¼ 0
S (9.107)
fΔg ¼ ½G fF g

We see in the discrete approach that the matrices of differential operators have been
replaced by equivalent matrices, i.e., ½ℒE  ! E ~ . We have placed a tilde
~ f , ½ℒS  ! S
over these matrices to emphasize that they represent discrete matrix replacements
for operators.
Since we solve for the forces in a force-based approach it would be useful to have an
explicit expression for calculating the nodal displacements. We can do this by noting that the
forces are obtained by the solution of Eq. (9.104) which we will write as
8 9
< fQg þ fPg =
~ system fF g ¼
E  (9.108)
: ;
f0g

Now, recall from the generalized strain–displacement relations we had

 
fΔg ¼ E ~ f T Uf (9.109)

where E ~ f is the “free” equilibrium matrix. Here we have a nonsquare matrix relating the
strains and displacements, but we can make the matrix square without changing Eq. (9.109)
by adding additional columns to the matrix and placing corresponding additional zeros in
the displacement vector. Specifically, we extend Eq. (9.109) as:
 9.3 Force-Based Finite Elements for Axial-Load Problems 309

8 9
> Uf >
h i>
< >
=
f Δg ¼ E~ T j ½G  S T ~ T

f >
> >
>
: ;
f0g
8 9 (9.110)
>
> Uf >>
< =
¼ E~ system T

>
> >
>
: ;
f0g

This relationship can be inverted to yield

8 9
>
> Uf > >
< =  1
T
 ¼ E system fΔg
>
> >
>
: ; (9.111)
f0g
 1
T
¼ E system ½G fF g

which gives us the free displacements (as well as some zeros) directly from a knowledge of
the internal forces and other quantities we have already used in determining those forces.

Example 9.6 Statically Indeterminate Bar – Force Method and Calculation

of Displacements
As an example of the use of Eq. (9.111), consider the bar between two fixed supports in Fig.
9.11a. Let us solve this problem symbolically in MATLAB® where we have let q0 ¼ q, l ¼ L:
syms q L A E % define variables
Esys ¼ [ 1 -1 0; 0 1 -1; 1 1 1] % system matrix derived
% previously
Esys ¼ 1 -1 0
0 1 -1
1 1 1
Q ¼ [ q*L; q*L; 0] % nodal forces due to distributed load
% vector {Q}
Q ¼ L*q
L*q
0
F ¼ inv(Esys)*Q % solve for internal force vector {F} (can
% also use F ¼ Esys\Q)
F ¼ L*q
0
-L*q
310 Computational Mechanics of Deformable Bodies

G ¼ (L/(E*A))*eye(3,3) % flexibility matrix, [G]

G ¼ [L/(A*E), 0, 0]
[ 0, L/(A*E), 0]
[ 0, 0, L/(A*E)]
U ¼ inv(Esys.’)*G*F % solve for the two free displacements
% and a zero (can also use U ¼ Esys.’\(G*F))
U ¼ (L^2*q)/(A*E)
(L^2*q)/(A*E)
0

which agrees with our previous results for the internal forces F 1 ¼ q0 l, F 2 ¼ 0, F 3 ¼ q0 l
and the two displacements U 2 ¼ U 3 ¼ q0 l 2 =EA.

We have seen previously that we could combine the equilibrium equations with the
generalized strain–displacement equations and the force-generalized strain equations to
arrive at a set of equations to solve for the displacements. Let us examine that case further
here. Recall, we can use
~ f fF g ¼ fQg þ fPg
E
 
fΔg ¼ E ~ f T Uf (9.112)

fΔg ¼ ½G fF g

to arrive at
T 
~ f ½G 1 E
E ~f Uf ¼ fQg þ fPg (9.113)

which is the discrete version of the Matrix–Vector 3-D Governing Set 2 of equations of
Chapter 6 (see Eq. (6.51)) given by

½ℒE ½C ½ℒE T fug þ ff g ¼ 0 (9.114)

which, as mentioned previously, are just Navier’s equations for the displacements. Thus,
Eq. (9.113) is just a discrete form equivalent to Navier’s equations.
[Note that we can also write a set of equations similar to those in Eq. (9.112) in terms of
the original equilibrium matrix, E ~ , and all the displacements (free or otherwise), fU g.
Then Eq. (9.113) will also involve all the equilibrium equations and displacements but there
will be unknown reactions on the right-hand side of Eq. (9.113) that must be eliminated
before we can arrive at a solution.]
We also obtained very similar looking expressions in the stiffness-based method, where
we found
½K fU g ¼ fQg þ fPg (9.115)
 9.3 Force-Based Finite Elements for Axial-Load Problems 311

and where ½K  is a stiffness matrix. However, Eq. (9.113) and Eq. (9.115) are not always
identical. One can see this by looking at the expressions at the element level. An element
stiffness, for example, can be written in a complete matrix–vector form as (see, for example,
the stiffness matrix appearing in Eq. (9.31)):
ðj Þ
lð
h i h iT h i
ðj Þ
K ¼ Bðj Þ ½EA Bðj Þ dξ (9.116)
0

where the matrix ½EA ¼ E ðj Þ AðjÞ , i.e., it is just a scalar for axial load problems. In contrast,
h i h i1
ðj Þ
at the element level the pseudostiffness matrix, K pseudo ¼ E ~ ðjÞ G ðj Þ ~ ðj Þ T is composed
E
of the terms:
ðj Þ
lð
h i h iT h i
~ ðj Þ ¼
E Bð j Þ N ðjÞ dξ
0
ðj Þ
lð
h iT h iT h i
~ ðj Þ
E ¼ N ðj Þ BðjÞ dξ (9.117)
0
ðj Þ
lð
h i h iT h i
ðj Þ
G ¼ N ðj Þ ½1=EA N ðjÞ dξ
0

In the stiffness method, the stiffness matrix is formed from shape function matrices involving
the strain only. Shape functions involving both the force and the strain are present in the
pseudostiffness matrix, so there is no guarantee that these two stiffness matrices are the
same. For a uniform axial-load element, however, we saw that when we use a linear strain
shape function, the elemental stiffness for a uniform element was given by (see Eq. (9.33b)):
h i E ðjÞ AðjÞ
1 1
K ðj Þ ¼ (9.118)
l ðj Þ 1 1

In a force-based method, if we also use a linear shape function for the strain in an element
and a force shape function of unity in that same element, however, a uniform element has
the terms in the pseudostiffness matrix given by
" #
h i 1
~
E ðj Þ
¼
1
h iT
~ ðj Þ ¼ ½ 1 1 
E (9.119)

h i l ðj Þ
G ðj Þ ¼
E ðjÞ AðjÞ
312 Computational Mechanics of Deformable Bodies

and we have
h i E ðj Þ A ðj Þ E ðjÞ AðjÞ 1 1
ðj Þ 1
K pseudo ¼ ½ 1 1  ¼ (9.120)
l ðj Þ 1 l ðj Þ 1 1

We see in this case the stiffness and pseudostiffness matrices for each element are identical.
When we assemble these elements, the system stiffness and pseudostiffness matrices
will also be identical, as we found in a previous example. In strength of materials torsion
and bending problems, we will also find equality between the stiffness and pseudostiffness
matrices but in other problems such as 2-D and 3-D elasticity problems, they will be
different. As a consequence of these differences, the stiffness-based approach, which is the
method commonly used today in finite element analyses, is not in general a discrete form of
Navier’s equations.

9.4 Generation of the Compatibility Equations

When solving statically indeterminate problems with a force-based finite element method,
we saw a key step was the determination of the compatibility matrix, S ~ . Since compatibil-
ity equations are relations between the strains that guarantee we can relate those strains to a
set of displacements, the strain–displacement equations are the natural starting point for
obtaining the compatibility matrix. In the discrete form we have the generalized strains, fΔg,
 
and the free nodal displacements, U f , which are related through the transpose of the
“free” equilibrium matrix, E ~ f , i.e.,
 
~ f T Uf
fΔg ¼ E (9.121)

The compatibility equations S ~ fΔg ¼ 0 define the relationship(s) we seek between the
strains. One way to obtain these compatibility equations, which we have used previously
in all our examples, is simply to eliminate the displacements by taking appropriate combin-
ations of the equations in Eq. (9.121). For simple problems this elimination can be done by
hand but we would like a more automated process. Recall, we also mentioned that applying
the compatibility matrix to Eq. (9.121) gives
T 
~ fΔg ¼ S
S ~ E
~f Uf ¼0 (9.122)

which must be true for all displacements so that we find

~ E
S ~f T
¼ E ~
~f S T
¼0 (9.123)

[Note: in this section we will describe how to obtain compatibility equations for the “free”
equilibrium matrix, E ~ f , but the same steps can be used to find compatibility equations for
the original equilibrium matrix, E ~ .]
 9.4 Generation of the Compatibility Equations 313

We can also determine the compatibility matrix if we can find a matrix S ~ that satisfies
Eq. (9.123). To see the steps needed, it is useful to choose a specific example, which we will
solve with MATLAB®. Suppose we have an equilibrium matrix defined as
E ¼ [1 1/2 0 0 0 1; 0 1/2 1 0 0 0; 0 0 0 1/2 1 0;
0 0 -1 -1/2 0 0]
E ¼ 1.0000 0.5000 0 0 0 1.0000
0 0.5000 1.0000 0 0 0
0 0 0 0.5000 1.0000 0
0 0 -1.0000 -0.5000 0 0

We see that there are four rows and six columns, so this represents four equilibrium
equations for six forces and is statically indeterminate to the second degree. Let us consider
finding all the solutions to the homogeneous equilibrium equation
~ f fF g ¼ f0g
E (9.124)

The first step is to perform a row reduction on these equations and produce what is called a
row-reduced echelon form of the equilibrium equations. This form can be obtained in
MATLAB® with the function rref:
rref(E)
ans ¼ 1 0 0 0 -1 1
0 1 0 0 2 0
0 0 1 0 -1 0
0 0 0 1 2 0

In this form we see that leading nonzero coefficients in columns one through four (corresponding
to the forces ðF 1 , F 2 , F 3 , F 4 Þ ) have a value of one. This means that the forces ðF 5 , F 6 Þ are free
forces in the sense that we can write all the forces in terms of them. To see this, let us write down
the homogeneous equilibrium equations in this row-reduced echelon form:
F1  F5 þ F6 ¼ 0
F 2 þ 2F 5 ¼ 0
(9.125)
F3  F5 ¼ 0
F 4 þ 2F 5 ¼ 0

which we can use to group and write all the forces as

8 9 8 9 8 9
>
> F1 >> >
> 1 >> >
> 1 >
>
>
> > > > > >
>
> F2 >> >
> 2 >> >
> 0 >
>
< > = >
< >
= >
< >
=
F3 1 0
¼ F5 þ F6 (9.126)
>
> F4 >> >
> 2 >> >
> 0 >
>
>
> >
> >
> >
> >
> >
>
> F > > 1 > > 0 >
>
: > 5
; >
: >
; >
: >
;
F6 0 1
314 Computational Mechanics of Deformable Bodies

but since F 5 and F 6 are arbitrary, we could call them constants c1 and c2 , respectively, and
write
8 9 8 9 8 9
>
> F1 >> >
> 1 >> >
> 1 >
>
>
> > > > > >
>
> F2 >> >
> 2 >> >
> 0 >
>
< > = >
< >
= >
< >
=
F3 1 0
¼ c1 þ c2 (9.127)
>
> F4 >> >
> 2 >> >
> 0 >
>
>
> > > > > >
>
> F > > >
> 1 >> >
> 0 >
>
: 5> ; >
: >
; >
: >
;
F6 0 1

Equation (9.127) represents the general solution to the homogeneous equilibrium equations,
showing that any combinations of the two column vectors appearing in Eq. (9.127) are solutions
to those equations. Thus, we can consider these two column vectors as basis vectors for all the
homogeneous solutions. If we form a matrix of these two column vectors and call it S ~ T , i.e.,
2 3
1 1
6 2 0 7
6 7
6 1 0 7
~ T
S ¼6 6 7 (9.128)
7
6 2 0 7
4 1 0 5
0 1

then Eq. (9.123) will be satisfied and we will have found the compatibility equations matrix.
We can verify this statement in MATLAB®:
ST ¼ [ 1 -1; -2 0; 1 0; -2 0; 1 0; 0 1];
E*ST
ans ¼ 0 0
0 0
0 0
0 0

All the force vectors given by Eq. (9.127) are said to lie in the null space of the equilibrium
matrix E ~ f , i.e., they are all sets of forces that generate zero (null) applied loads. The basis
vectors, or equivalently the matrix S ~ T of these basis vectors, can be said to describe the null
space since we can write any vector in this space in terms of them through Eq. (9.127). We
can generate the matrix S ~ T directly in MATLAB® without going through all the steps
outlined above since there is a function null that does all the work for us:
ST =null(E, ‘r’)
ST ¼ 1 -1
-2 0
1 0
-2 0
1 0
0 1
 9.5 Numerical Solutions – Beam Bending 315

We see that the function null outputs the transpose of the compatibility matrix S ~ , The
‘r’ argument in the null function produces a “rational” set of basis vectors for the null
space obtained from the row reduced echelon form. What this means is that if the matrix
E~ is a small matrix with integer elements, the elements of the basis vectors generated by
null(E, ‘r’) will be ratios of small integers, generating results similar to what we would
obtain by hand elimination. Consider now if we call the null function without this
rational option. We find:
ST2 ¼ null(E)
ST2 ¼ -0.2302 -0.6862
0.6136 -0.0670
-0.3068 0.0335
0.6136 -0.0670
-0.3068 0.0335
-0.0766 0.7197

This looks nothing like the previous transpose of the compatibility matrix but it is also a
legitimate solution. Recall, we can use combinations of the basis vectors multiplied by
arbitrary constants and still have equivalent compatibility equations. In a number of
previous examples, we used this flexibility in writing the compatibility equations to produce
a more consistent appearance with the equilibrium equations. If we check the null property
of these alternative compatibility equations, we find
E*ST2
ans ¼ 1.0e-15 *
0.0278 -0.2220
-0.1665 -0.0208
0.0555 -0.0069
-0.1665 0.0139

which is not identically zero (due to numerical round off errors) but is indeed very small. The
null function called without the rational ‘r’ argument present computes the null space from a
singular value decomposition method and produces orthonormal basis vectors. More details
can be found in linear algebra texts and by looking at the MATLAB® reference page for
singular value decomposition (svd), but we will not discuss them here. [Note: when using the
null function with a symbolic matrix, as we will do shortly, one should use the function
without the ‘r’ argument.]

9.5 Numerical Solutions – Beam Bending

Axial-load problems provide a simple context in which to examine the method of finite
elements. As mentioned previously, the torsion of circular shafts will satisfy the same basic
316 Computational Mechanics of Deformable Bodies

equations as axial loads if we make the replacements F x ! M x , EA ! GJ, u ! ϕ, qx ! tx

where M x is the internal axial torque, GJ is the shear modulus times the polar area
moment of the cross-section, ϕ is the angle of twist, and tx is the distributed torque/unit
length. Thus, we will not discuss separately finite elements for torsion problems. Beam-
bending problems, however, have a more general structure than axial-load and torsion
problems, so we will consider those problems here.

9.5.1 Principles of Virtual Work and Complementary Virtual Work

Figure 9.17a shows an element of a beam and the forces acting on the element. The strain
energy for bending was given in Eq. (8.59), which can be written for the element as
ðl
1
U¼ E ðξ ÞI ðξ Þ½κðξ Þ2 dξ (9.129)
2
0

where κ ¼ d 2 w=dξ 2 is the curvature of the beam in the element, wðξ Þ is the z-displacement
of the neutral axis, and ðE; I Þ are Young’s modulus and the area moment of the cross-
section about the y-axis (both of which can be functions of ξ). As is customary, the strain
energy due to shear stresses is ignored as it is usually much smaller than the bending strain
energy of Eq. (9.129). The principle of virtual work says that the virtual change of this
strain energy is equal to the virtual work done by the forces acting on the element, so from
Figs. 9.17a, b we have

z q z (x )

Mj M j +1
x M (x )

V j +1 Vz (x )
Vj
l x
x
(a) (b)
Wj W j +1

θ j +1
θj

(c)

Figure 9.17 (a) A beam element for use in finite element analysis. (b) The internal force and moment in the
element. (c) The nodal displacements and slopes at the ends of the element.
 9.5 Numerical Solutions – Beam Bending 317

ðl
δU ¼ E ðξ ÞI ðξ Þκðξ Þδκðξ Þdξ
0
(9.130)
ðl
¼ δW v ¼ qz ðξ Þδwðξ Þdξ þ V j δW j þ V jþ1 δW jþ1 þ M j δθj þ M jþ1 δθjþ1
0


where qz is the force/unit length acting in the z-direction, V j , V jþ1 are the shear forces


acting at the end nodes, and M j , M jþ1 are the end moments. The displacements



W j , W jþ1 are the end nodal displacements of the beam in the z-direction, and θj , θjþ1
are the nodal angular displacements. For small slopes, as assumed in engineering beam
theory, we have
 
dw dw
θj ¼  , θjþ1 ¼  (9.131)
dξ ξ¼0 dξ ξ¼l

The positive directions of all the external forces and moments are shown in Fig. 9.17a. The
corresponding positive directions of the nodal displacements and nodal angular displace-
ments are given in Fig. 9.17c. Figure 9.17b shows the positive directions assumed for the
internal shear force and bending moment within the beam.
The principle of virtual work can also be obtained directly from Eq. (8.56) for the
equilibrium of a beam element. Multiplying that equilibrium equation by a virtual displace-
ment δwðξ Þ and integrating over an element, we have
ðj Þ
lð !
d 2M
 qz δwðξ Þdξ ¼ 0 (9.132)
dξ 2
0

which we can rewrite as

ðl   ðl   ðl
d dM dM d
δw dξ  ðδwÞ dξ ¼ qz ðξ Þδwðξ Þdξ (9.133)
dξ dξ dξ dξ
0 0 0

Carrying out the integration of the perfect differential and using dM=dξ ¼ V ðξ Þ and
dw=dξ ¼ θðξ Þ, Eq. (9.133) can be shown to become
ðl   ðl
dM
 δθ dξ ¼ qz ðξ Þδwðξ Þdξ þ V ðl Þδwðl Þ  V ð0Þδwð0Þ (9.134)
dξ
0 0

which can again be rewritten as

ðl ðl   ðl
d d
 ðMδθÞ þ M ðδθÞ dξ ¼ qz ðξ Þδwðξ Þdξ þ V ðl Þδwðl Þ  V ð0Þδwð0Þ (9.135)
dξ dξ
0 0 0
318 Computational Mechanics of Deformable Bodies

We have dθ=dξ ¼ d 2 w=dξ 2 ¼ κðξ Þ, where κðξ Þ is the curvature of the neutral axis and the
perfect differential in Eq. (9.135) can be performed, giving
ðl ðl
ðMδκÞ dξ ¼ qz ðξ Þδwðξ Þdξ
(9.136)
0 0
þV ðl Þδwðl Þ  V ð0Þδwð0Þ þ M ðl Þδθðl Þ  M ð0Þδθð0Þ

However, we can recognize the relations (see Figs. 9.18a, b, which show the forces and
moments at the ends):
V ðl Þ ¼ V jþ1 , V ð0Þ ¼ V j
M ðl Þ ¼ M jþ1 , M ð0Þ ¼ M j
(9.137)
δwðl Þ ¼ δW jþ1 , δwð0Þ ¼ δW j
δθðl Þ ¼ δθjþ1 , δθð0Þ ¼ δθj

so that by using Eq. (9.137) in Eq. (9.136) as well as the moment–curvature relation

d 2 wð ξ Þ
M ðξ Þ ¼ EI κðξ Þ ¼ EI (9.138)
dξ 2
we obtain
ðl ðl
ðE ðξ ÞI ðξ Þκðξ Þδκðξ ÞÞdξ ¼ qz ðξ Þδwðξ Þdξ
(9.139)
0 0
þ V jþ1 δW jþ1 þ V j δW j þ M jþ1 δθjþ1 þ M j δθj

which is identical to Eq. (9.130).

Figure 9.18 (a) A beam element showing the end Vj V j +1

loads. (b) Small portions of the element at the
M j +1
ends. These portions must be in equilibrium, Mj (j)
which gives the relationship between the end
loads and the internal force and moment in the x
beam at the ends. (a)

Vj M ( 0) M (l ) V j +1
M j +1
Mj

Vz ( 0 ) Vz ( l )

(b)
 9.5 Numerical Solutions – Beam Bending 319

Now consider the principle of complementary virtual work for a beam. For linear
problems the complementary strain energy is the same as the strain energy, which written
in terms of the internal moment for an element is
ðl
1 1
U¼ ½M ðξ Þ2 dξ (9.140)
2 E ðξ ÞI ðξ Þ
0

If we consider the principle of complementary virtual work for variations of the internal
moment, δM, that satisfy the homogeneous equilibrium equation (δqz ¼ 0) then that
principle for an element states that the virtual change of the (complementary) strain energy
is equal to the complementary work of the external loads, i.e. (see Fig. 9.17)
ðl
1
δU ¼ M ðξ ÞδM ðξ Þdξ
E ðξ ÞI ðξ Þ
0 (9.141)
¼ δW cv
¼ W jþ1 δV jþ1 þ θjþ1 δM jþ1 þ W j δV j þ θδM j

Alternatively, we can use the generalized strain–displacement relation

d 2 wð ξ Þ
κðξ Þ ¼ (9.142)
dξ 2
and multiply it by a virtual change of the internal moment that satisfies the homogeneous
equilibrium equation to give
ðl !
d 2 wð ξ Þ
κðξ Þ  δM ðξ Þdξ ¼ 0 (9.143)
dξ 2
0

which we can rewrite as

ðl ðl   ðl
d dwðξ Þ dwðξ Þ
κðξ ÞδM ðξ Þdξ  δM ðξ Þdξ þ δV ðξ Þdξ ¼ 0 (9.144)
dξ dξ dξ
0 0 0

where we have used the relationship between the bending moment and the shear force.
Carrying out the integral of the perfect differential and rearranging the last integral in Eq.
(9.144) as the sum of a perfect differential and an additional integral, we have
ðl ðl
d
κðξ ÞδM ðξ Þdξ  ðwðξ ÞδV ðξ ÞÞ
dξ
0 0 (9.145)
dwðl Þ dwð0Þ
¼ δM ðl Þ  δM ð0Þ
dξ dξ
320 Computational Mechanics of Deformable Bodies

where no additional integral is present in Eq. (9.145) because the virtual change of the shear
force also satisfies a homogeneous equilibrium equation, i.e., d ðδV ðξ ÞÞ=dξ ¼ 0. Performing
the integration of the perfect differential in Eq. (9.145) gives
ðl
dwðl Þ dwð0Þ
κðξ ÞδM ðξ Þdξ ¼ wðl ÞδV ðl Þ  wð0ÞδV ð0Þ þ δM ðl Þ  δM ð0Þ (9.146)
dξ dξ
0

which we can write in terms of the nodal displacements and angular displacements of
Fig. 9.17 and the variations of the end forces and moments as:
ðl
κðξ ÞδM ðξ Þdξ ¼ W jþ1 δV jþ1 þ W j δV j þ θjþ1 δM jþ1 þ θj δM j (9.147)
0

which is identical to the principle of complementary virtual work expression of Eq. (9.141) if
we use the moment–curvature relation in Eq. (9.142).

9.6 Stiffness-Based Finite Elements for Beam Bending

To use the principle of virtual work in a stiffness-based finite element approach, we need to
define the shape functions that we will use to approximate the displacement in an element.


Since there are four nodal variables W j , θj , W jþ1 , θjþ1 that we use as discrete unknowns,
we write the displacement in an element in terms of four constants in a cubic polynomial:

wðξ Þ ¼ a1 þ a2 ξ þ a3 ξ 2 þ a4 ξ 3 (9.148)

We must relate these constants to the nodal unknowns by using the conditions
wð0Þ ¼ W j , wðl Þ ¼ W jþ1
 
dwðξ Þ dwðξ Þ (9.149)
¼ θ , ¼ θjþ1
dξ ξ¼0 dξ ξ¼l
j

which leads to
8 9 2 38 9
>
> Wj > > 1 0 0 0 > > a1 >
< = 61 1 < > =
θj 0 0 7 a2
¼6
41 l
7 (9.150)
>
> W jþ1 >
> l2 l 3 5>
> a >
: ; : 3> ;
θjþ1 1 1 2l 3l 2 a4

Inverting Eq. (9.150) gives

8 9 2 3 38 9
>
> a1 >
> l 0 0 0 > > Wj > >
< = 16 < =
a2 6 0 l3 0 0 7 7 θj
¼ (9.151)
> > l 3 4 3l
> a3 > 2l 2 3l l 2 5>
> W jþ1 >
>
: ; : ;
a4 2 l 2 l θjþ1
 9.6 Stiffness-Based Finite Elements for Beam Bending 321

which, when placed into Eq. (9.148) gives

   
ξ2 ξ3 ξ2 ξ3
wð ξ Þ ¼ 1  3 2 þ 2 3 W j þ ξ  2 þ 2 θ j
l l l l
 2    (9.152)
ξ ξ3 ξ2 ξ3
þ 3 2  2 3 W jþ1 þ  þ 2 θjþ1
l l l l
and which we can write in matrix–vector form as
8 9
>
> Wj > >
< =
θj
fwg  wðξ Þ ¼ ½H fU g, fU g ¼ (9.153)
>
> W >
: jþ1 >;
θjþ1

where fU g is a vector of generalized nodal displacements. Based on this approximation we

have the curvature given by
   
d 2w 6 12ξ 4 6ξ
κðξ Þ ¼ 2 ¼  2 þ 3 W j þ  þ 2 θj
dξ l l l l
    (9.154)
6 12ξ 2 6ξ
þ 2  3 W jþ1 þ  þ 2 θjþ1
l l l l

which in matrix–vector form is

fκðξ Þg  κðξ Þ ¼ ½BfU g (9.155)

Now, consider the principle of virtual work, Eq. (9.139), which we write in matrix–vector form as
ðl   ðl
E ðξ ÞI ðξ Þfδκðξ Þg fκðξ Þg dξ ¼ fδwðξ ÞgT fqz ðξ Þgdξ þ fδU gT fPg
T
(9.156)
0 0

where fPg is a vector of generalized loads, i.e.,

8 9
>
> Vj >>
< =
Mj
fPg ¼ (9.157)
>
> V >
: jþ1 > ;
M jþ1

Placing the matrix–vector expressions for the displacement and curvature and their vari-
ations into Eq. (9.156), we have
ðl
T
fδU g E ðξ ÞI ðξ Þ½Bðξ ÞT ½Bðξ Þdξ fU g
0
(9.158)
ðl
¼ fδU gT ½H ðξ ÞT qz ðξ Þdξ þ fδU gT fPg
0
322 Computational Mechanics of Deformable Bodies

which must be true for all variations of the generalized nodal displacements so that we find a
set of linear equilibrium equations for the generalized displacements:
½K fU g ¼ fQg þ fPg (9.159)

where ½K  is the stiffness matrix for an element:

ðl
½K  ¼ E ðξ ÞI ðξ Þ½Bðξ ÞT ½Bðξ Þdξ (9.160)
0

and fQg are the equivalent nodal forces from the distributed force qz :
ðl
fQg ¼ ½H ðξ ÞT qz ðξ Þdξ
0
(9.161)
ðl


¼ ½H ðξ ÞT qz xj þ ξ dξ
0

where the second form in Eq. (9.161) should be used when the distributed force expression is
given in terms of global x-coordinates instead of the local ξ-coordinate, as discussed for axial
load problems. For a constant distributed force acting in the negative z-direction, qz ¼ q0 ,
and an element of length l either form can be used, giving
8 9
>
> ðl >
>
>

> q 1  3ξ 2 =l 2 þ 2ξ 3 =l 3 dξ >
> >
>
>
> 0 >
>
>
> > 8
> 9
>
> 0 >
> q0 l >
>
> >
> >
> >
8 q 9 > >
> ðl >
>
>
>
>
> 2 > >
>
> Vj > >
>
 >
> >
> >
>
>
> >
> >
> q ξ  2
=l þ ξ 3
=l 2 >
> >
> >
>
>
> >
> >
> 0 2ξ dξ >
> >
> q l 2
>
>
> q >
< Mj = < > >
= >
<
0 >
=
0 12
f Qg  ¼ ¼ (9.162)
>
> V q >
> >
> ðl >
> >
> q l >
>
>
> jþ1 > >
> >
 > >
> > 0 >
>
>
> > > q0 3ξ 2 =l 2  2ξ 3 =l 3 dξ > > 2 >
: Mq > ; > >
>
>
>
>
>
>
>
>
>
>
>
> >
> >
> >
>
jþ1
>
> >
> >
> q l 2 >
>
>
>
0
>
> : 0 ;
>
> >
>
>
> ðl >
> 12
>
>
 >
>
>
> q0 ξ 2 =l þ ξ 3 =l 2 dξ > >
>
: >
;
0


where V qm ; M m (m ¼ j, jþ1) are the nodal forces and moments equivalent to the distributed
q

force. Figure 9.19 shows an element of length l carrying a uniform distributed force and the
equivalent nodal forces and moments consistent with Eq. (9.162).
The element stiffness matrix ½K  is a 4  4 matrix with terms K mn ðm; n ¼ 1; 2, 3, 4Þ. We
will not give the details of the calculations for this matrix but instead simply show one of the
element terms explicitly for a uniform element (EI ¼ constant), namely
 9.6 Stiffness-Based Finite Elements for Beam Bending 323

q0 Figure 9.19 (a) A beam element with a uniform distributed

force. (b) The equivalent nodal forces and moments.

ðl  2 ð
2
6=l
l 12ξ 6 EI l 3
K 11 ¼ EI  2 dξ ¼ v2 dv
(a) l3 l 12
0 6=l 2
q0l / 2 q0l / 2 ¼ 12EI =l 3

Carrying out the other integrations similarly, we have

2 3
12 6l 12 6l
q0l 2 /12 q0l 2 /12 EI 6 6l 4l 2
6l 2l 2 7
½K  ¼ 3 6
4
7 (9.163)
(b) l 12 6l 12 6l 5
6l 2l 2 6l 4l 2

which is, as expected, a symmetric matrix.

When the elements of the beam are assembled, the sum of the forces and moments
acting at the ends of adjacent elements must be in equilibrium with any applied concen-
trated applied forces and moments at a node, as shown in Fig, 9.20b, or equivalently the
internal forces and moments from the adjacent elements must be in equilibrium with these
applied forces and moments, as shown in Fig. 9.20c. At node j+1 between element ( j) and
( jþ1) as shown in Fig. 9.20 we have, therefore,
ðj Þ ðjþ1Þ
V jþ1 þ V jþ1 ¼ V ∗
jþ1
ðj Þ ðj Þ
M jþ1 þ M jþ1 ¼ M ∗ jþ1
  (9.164)
V ðzjÞ l ðjÞ  V ðzjþ1Þ ð0Þ ¼ V ∗
jþ1
 
M ðj Þ l ðjÞ  M ðjþ1Þ ð0Þ ¼ M ∗ jþ1

where V ∗ ∗
jþ1 and M jþ1 are the applied vertical force and moment at node jþ1 as seen in
ðj Þ ðj Þ
Fig. 9.20 and we have used the more explicit notation of M jþ1 , V jþ1 , etc. for the end
moments and forces. These are very similar to the same relations we obtained in axial-
load problems.

Example 9.7 Statically Determinate Beam – Stiffness Method

As a very simple example of a statically determinate problem, consider the cantilever beam
of Fig. 9.21, which is acted upon by a uniform force along its entire length. In this case we
can use a single element as shown in Fig. 9.21b and from Eq. (9.163) our equilibrium
equation is
324 Computational Mechanics of Deformable Bodies

Figure 9.20 (a) The end moments and forces M (j +j 1) M (j +j 1+1)

at node j + 1 between element ( j) and
element ( j+1). (b) Forces and moments at
node j + 1 where an applied force, V ∗ jþ1 , and
applied moment, M ∗ jþ1 , are present. (c) The
V j(+1) V j(+j1+1)
j
(j) (j+1)
same node j + 1 but where the internal
forces and moments in elements ( j) and (a)
( j+1) are shown instead.
V j*+1
M *j +1

M (j +1) M (j +j 1+1)
j

V j(+j1) V j(+j1+1)

(b)

V j*+1
M *j +1

M ( j) (l ) M ( j +1) ( 0 )

Vz( j ) ( l ) Vz( j +1) ( 0 )

(c)

2 38 9 8 q; ð1Þ 9 8 ð1Þ 9
12 6L 12 6L > W > >
> V1 >
> >
> V1 > >
>
>
1 >
> >
> >
> >
> >
>
6 7
2 7<
>
> >
> >
> >
> >
> >
>
EI 6
2
6L 2L 7 θ1 = < M1
q ; ð1 Þ = < M1 =
ð 1Þ
6 6L 4L
6 7 ¼ þ
36
L 4 12 6L 12 6L 7 > W2 > > V q; ð1Þ > > V2 >
ð1Þ >
5>>
>
>
>
>
>
>
>
>
>
>
>
>
> >
>
>
: >
; >
>
2
>
> >
> >
>
6L 2L2 6L 4L 2
θ2 : q; ð1Þ ; : ð1Þ ;
M2 M2 (9.165)
8 9 8 9
>
> q0 L=2 > > >
> V R1 >
>
>
> >
> > > >
>
< q0 L2 =12 >
> = > < M R1 > =
¼ þ
>
> q0 L=2 > > > 0 >
> >
>
> >
> > > >
>
>
: >
; > : >
;
q0 L =12
2
0

where ðV R1 , M R1 Þ are the reactions at the fixed support as shown in Fig. 9.21b. To solve this
problem we must apply the boundary conditions at the wall which are W 1 ¼ θ1 ¼ 0
 9.6 Stiffness-Based Finite Elements for Beam Bending 325

z Figure 9.21 (a) A cantilever beam carrying a uniform

distributed load. (b) The single finite element used for this
q0
problem, showing the nodal displacements and angular
displacements and the reactions at the wall.

L x
(no deflection and slope) and which we can do so by
(a) eliminating the first two rows in Eq. (9.165) as well as
W1 W2 the first two columns of the stiffness matrix. We find
T1 T2  
M R1 EI 12 6L W2 q0 L=2
¼ (9.166)
L3 6L 4L θ2 q0 L2 =12
2
VR1
(b)
which we solve symbolically in MATLAB®:

syms E I L q
P ¼ [-q*L/2; q*L^2/12]; % the known loads
Kr ¼ (E*I/L^3)*[12 -6*L; -6*L 4*L^2]; % the reduced
% stiffness matrix
inv(Kr)*P % solve for the generalized displacements (can
% also use Kr\P)
ans ¼ -(L^4*q)/(8*E*I)
-(L^3*q)/(6*E*I)

This gives the free end deflection and slope W 2 ¼ q0 L4 =8EI , θ2 ¼ q0 L3 =6EI . You can
verify these are actually the exact results. To find the end reactions, we can go back to the
original stiffness matrix in Eq. (9.165) and multiply by the known generalized displacements:
syms E I L q
K ¼ [ 12, 6*L , -12, 6*L; % the original stiffness matrix
6*L , 4*L^2, -6*L, 2*L^2; % without the EI/L^3 factor
-12, -6*L 12 , -6*L;
6*L, 2*L^2, -6*L, 4*L^2]
K ¼ [ 12, 6*L, -12, 6*L]
[ 6*L, 4*L^2, -6*L, 2*L^2]
[ -12, -6*L, 12, -6*L]
[ 6*L, 2*L^2, -6*L, 4*L^2]
U ¼ [ 0 ; 0 ; -(L^4*q)/(8*E*I); -(L^3*q)/(6*E*I)] % all
% the generalized displacements
U ¼ 0
0
326 Computational Mechanics of Deformable Bodies

-(L^4*q)/(8*E*I)
-(L^3*q)/(6*E*I)
(E*I/L^3)*K*U % find the loads, including
% reactions
ans ¼ (L*q)/2
(5*L^2*q)/12
-(L*q)/2
(L^2*q)/12

From these results and from Eq. (9.165) we find V R1  q0 L=2 ¼ q0 L=2, M R1  q0 L2 =12 ¼
5q0 L2 =12 so that V R1 ¼ q0 L and M R1 ¼ q0 L2 =2, which are obviously the correct reactions
by considering equilibrium of the entire beam.

As a slightly more complex statically indeterminate problem, consider for the next example
the beam shown in Fig. 9.22a.

Figure 9.22 (a) A statically z

indeterminate beam. (b) The beam
P
broken into two finite elements, showing
the nodal displacements and angular
displacements, the reaction force and
moment at the fixed wall, and the x
reaction force at the smooth roller. (c) L L
The end forces and moments acting on
each element. (a)

W1 W2 W2 W3
(1) (2)
M R1 q1 q2 q2 q3
L L VR 3
VR1
(b)

(1) (2)

L L
( 2)
M 1( ) V1( ) V2(1) M 2( ) M 2 V2( ) V3( ) M 3( 2)
1 1 1 2 2

(c)
 9.6 Stiffness-Based Finite Elements for Beam Bending 327

Example 9.8 Statically Indeterminate Beam – Stiffness Method

If we use two finite elements to describe the problem of Fig. 9.22a, the nodal displacements
and angular displacements are shown in Fig. 9.22b as well as the end reaction force, F R1 ,
and moment, M R1 . The end forces and moments acting on these elements are shown in
ð1Þ ð1Þ ð2Þ ð2Þ
Fig. 9.22c, where we have V 1 ¼ V R1 , M 1 ¼ M R1 , V 3 ¼ V R3 , M 3 ¼ 0. Here, when we
assemble the elements, the system equilibrium equations will be:
8 9
ð1Þ
2 38 9 > > V >
> 8 9
12 6L 12 6L 0 0 > W1 >
> > > >
>
1 >
> >
> > V R1 >
>
6 7> > > > > >
0 7> > θ1 >> > >
> > > M R1 >
ð1Þ
6 6L 4L2 6L 2L2 0 > > >
> M1 > > >
>
6 >
7> >
> >
> >
> > > >
>
>
< >
= >
< ð2Þ >= >
< >
6 7 P =
ð1Þ
EI 6 12 6L 24 0 12 6L 7 W 2 V2 þ V2
6 7 ¼ ¼
L3 6 2
8L2 6L 2L2 7 > θ2 > > M2 þ M2 >
ð1Þ ð2Þ > 0 >
6 6L 2L 0 7>
> >
> >
> >
> >
> >
>
6 7>
> >
> >
> >
> >
> >
>
4 0 0 12 6L 12 6L 5> >
> W 3>
>
> >
>
> ð 2Þ >
>
>
>
>
> V >
>
>
>
: >
; > > V >
> >
:
R3
>
;
2
6L 4L 2
θ3 >
>
3
>
>
0 0 6L 2L : ð2Þ ; 0
M3
(9.167)

where V R3 is the vertical reaction force at the roller. The assembled stiffness matrix comes
from the two element contributions shown in Fig. 9.23 which can be seen to add to produce
the matrix of Eq. (9.167). The element end forces
element (1) combine during the assembly to produce any existing
external forces or moments, as outlined in Fig. 9.20.
12 6 L −12 6 L 0 0 To solve this system, we apply the boundary condi-
6 L 4 L2 −6 L 2 L2 0 0 tions W 1 ¼ θ1 ¼ 0 at the wall and W 2 ¼ 0 at the
EI −12 −6 L 12 −6 L 0 0 roller by eliminating the first and second rows and
L3 6 L 2 L2 −6 L 4 L2 0 0 columns of the stiffness matrix, and the fifth row and
0 0 0 0 0 0 column leaving
2 38 9 8 9
0 0 0 0 0 0 24 0 6L > > θ2 >
> >
> P >>
EI 6 7< = < =
6 0 8L2 2L2 7 W 3 ¼ 0 (9.168)
L3 4 5>
>
:
>
>
;
>
>
:
>
>
;
element (2) 6L 2L2 4L2 θ3 0

0 0 0 0 0 0
0 0 0 0 0 0
EI 0 0 12 6 L −12 6L
L3 0 0 6 L 4 L2 −6 L 2 L2
0 0 −12 −6 L 12 −6 L Figure 9.23 The stiffness matrices for the two elements used
0 0 6L 2 L2 −6 L 4 L2 in the beam problem of Fig. 9.22a.
328 Computational Mechanics of Deformable Bodies

which we will solve symbolically in MATLAB®:

syms E I L P
K ¼ [12 6*L -12 6*L 0 0; % form up stiffness matrix
% (without EI/L^3)
6*L 4*L^2 -6*L 2*L^2 0 0;
-12 -6*L 24 0 -12 6*L;
6*L 2*L^2 0 8*L^2 -6*L 2*L^2;
0 0 -12 -6*L 12 -6*L;
0 0 6*L 2*L^2 -6*L 4*L^2];
Kr ¼ K; % define reduced matrix where fifth row
% and column are eliminated
Kr(:, 5) ¼ [ ]; % and first and second rows and columns
% are eliminated
Kr(5,:) ¼ [ ];
Kr(:, 1:2) ¼ [ ];
Kr(1:2, :) ¼ [ ] % (remember to go from highest to lowest
% rows and columns)
Kr ¼ [ 24, 0, 6*L] % show reduced stiffness matrix
[ 0, 8*L^2, 2*L^2]
[ 6*L, 2*L^2, 4*L^2]

F ¼ [-P; 0; 0]; % define generalized force vector

% of applied loads
U ¼ (L^3/(E*I))*inv(Kr)*F % solve for generalized
% displacements, putting EI/L^3 factor back (can also use U =
% (L^3/(E*I))*(Kr\F))
U ¼ -(7*L^3*P)/(96*E*I)
-(L^2*P)/(32*E*I)
(L^2*P)/(8*E*I)
U2 ¼ [ 0; 0; U(1);U(2);0; U(3)] % define vector of all
% displacements
U2 ¼ 0
0
-(7*L^3*P)/(96*E*I)
-(L^2*P)/(32*E*I)
0
(L^2*P)/(8*E*I)
 9.7 Force-Based Finite Elements for Beam Bending 329

(E*I/L^3)*K*U2 % multiply all displacements by original

% stiffness matrix, putting the EI/L^3 factor back
ans ¼ (11*P)/16
(3*L*P)/8
-P
0
(5*P)/16
0

These results show that the unknown generalized displacements are W 2 ¼ 7PL3 =96EI ,
θ2 ¼ PL2 =32EI , θ3 ¼ PL2 =8EI . Also obtained were the unknown reactions V R1 ¼ 11P=16,
M R1 ¼ 3PL=8, and V R3 ¼ 5P=16 which you can verify are in force and moment equilibrium
with the applied force P for the entire beam.

9.7 Force-Based Finite Elements for Beam Bending

To apply a force-based finite element approach to our bending problems, we write the
principle of virtual work and complementary virtual work in matrix–vector notation for
an element as
ðl ðl
fδκðξ Þg fM ðξ Þgdξ ¼ fδwðξ ÞgT fqz ðξ Þgdξ þ fδU gT fPg
T
(9.169)
0 0

and
ðl
1
fδM ðξ ÞgT fM ðξ Þgdξ ¼ fδPgT fU g (9.170)
E ðξ ÞI ðξ Þ
0

where
8 9 8 9 8 9 8 9
>
> Wj > > >
> Vj > > >
> δW j >> >
> δV j >>
< = < = < = < =
θj Mj δθj δM j
fU g ¼ , fPg ¼ , fδU g ¼ , fδPg ¼ (9.171)
>
> W > > V > > δW jþ1 > > δV jþ1 >
: jþ1 >; : jþ1 >
> ; >
: >
; >
: >
;
θjþ1 M jþ1 δθjþ1 δM jþ1

We will use the same shape functions described previously to write the displacement and
curvature as
fwðξ Þg  wðξ Þ ¼ ½H fU g
(9.172)
fκðξ Þg  κðξ Þ ¼ ½BfU g
330 Computational Mechanics of Deformable Bodies

and their variations as

fδwðξ Þg  δwðξ Þ ¼ ½H fδU g
(9.173)
fδκðξ Þg  δκðξ Þ ¼ ½BfδU g

Here, we will also use shape functions to define the internal moment and its
variation as
fM ðξ Þg  M ðξ Þ ¼ ½N fF g
(9.174)
fδM ðξ Þg  δM ðξ Þ ¼ ½N fδF g

where fF g is a set of generalized force parameters that represent the internal moment. The
nodal displacements and angular displacements are similarly parameters that define the
components of the generalized displacement vector fU g. In the principle of complementary
virtual work the variations of the moment must satisfy the homogeneous equilibrium
equations so we will choose shape functions such that

d2
ðδM ðξ ÞÞ ¼ 0 (9.175)
dξ 2
which can be satisfied if

d2
½N  ¼ 0 (9.176)
dξ 2
so that the shape functions must be linear functions. For the ( j)th element we will take
n o M 
ξ

ξ
ðj Þ 2j1
F ¼ , ½N  ¼ 1 (9.177)
M 2j l l

so that the internal bending moment for the ( j)th element, M ðjÞ ðξ Þ, is

ξ
M ðjÞ ðξ Þ ¼ M 2j1 þ M 2j  M 2j1 (9.178)
l


and M 2j1 , M 2j can be seen to be just the values of the internal moment at the beginning
and end of the element, i.e., M 2j1 ¼ M ðj Þ ð0Þ and M 2j ¼ M ðjÞ ðl Þ. Thus, there are two
internal moment values for each element. Note, however, that these internal moment values
are different from the end moments acting on an element that were sometimes previously


labeled as M j , M jþ1 (see Fig. 9.18a). To be more precise, those end moments appearing in
Fig. 9.18a should be labeled, using our more explicit notation for an element ( j), as
 
ðj Þ ðj Þ
M j ; M jþ1 . We previously omitted the ( j) superscripts to keep the expressions for the
end moments and forces less complex-looking during derivations, but now it is necessary to
be more precise. Those end moments and these internal moments are related through
ðj Þ ðj Þ
M 2j1 ¼ M j and M 2j ¼ M jþ1 .
If we place all of these shape functions into the principle of virtual work, we have
 9.7 Force-Based Finite Elements for Beam Bending 331

ðl
fδU g T
½Bðξ ÞT ½N ðξ Þdξ fF g
0
(9.179)
ðl
¼ fδU gT ½H T fqz ðξ Þgdξ þ fδU gT fPg
0

which must be satisfied for all variations of the generalized displacements so we arrive at the
discrete equilibrium equations
~ fF g ¼ fQg þ fPg
E (9.180)

where
ðl
~ ¼ ½Bðξ ÞT ½N ðξ Þdξ
E
0
(9.181)
ðl
fQg ¼ ½H ðξ ÞT fqz ðξ Þgdξ
0

These equations are identical in form to those we obtained for axial-load problems (see Eq.
(9.61) and Eq. (9.62)), although the shape functions used here are different.
Now, consider the principle of complementary virtual work. Placing the approximations
in terms of the shape functions into this principle, we have
ðl
1
fδF g T
½N ðξ ÞT ½N ðξ Þdξ fF g ¼ fδPgT fU g (9.182)
E ðξ ÞI ðξ Þ
0

If we now define a set of discrete generalized strains fΔg as

fΔg ¼ ½G fF g (9.183)

where we have the generalized flexibility matrix

ðl
1
½G  ¼ ½N ðξ ÞT ½N ðξ Þdξ (9.184)
E ðξ ÞI ðξ Þ
0

then Eq. (9.182) becomes

fδF gT fΔg ¼ fδPgT fU g (9.185)

However, since the virtual changes of the internal moments must satisfy homogeneous
equilibrium equations, we have
~ fδF g ¼ fδPg
E (9.186)
332 Computational Mechanics of Deformable Bodies

so placing the transpose of Eq. (9.186) into Eq. (9.185) and requiring the result be satisfied
for all variations of the generalized forces, we obtain
~
fΔg ¼ E
T
fU g (9.187)

which is the generalized strain–displacement relationship, again in the same form as found
for axial loads.
To see what the various matrices are explicitly, we need to carry out the evaluations of the
integrals. For the generalized flexibility matrix, for example, if we assume a uniform element
(EI ¼ constant) of length l we have
2 l 3
ð 2 ðl  
6 ξ ξ ξ 2 3
6 1 dξ 1 dξ 7
7 1 1
l l l
1 66 0 0
7
7 l 6
63 67
7
½G  ¼ 6l 7¼ (9.188)
EI 6 ð   ð 2
l 7 EI 1 1 5
4
6 7
4 1  ξ ξ dξ ξ
dξ 5 6 3
l l l2
0 0

The equilibrium matrix given in Eq. (9.181) is a 4  2 matrix. If we write the shape functions
in this matrix as
½N ðξ Þ ¼ ½N 1 ðξ Þ, N 2 ðξ Þ
(9.189)
½Bðξ Þ ¼ ½B1 ðξ Þ, B2 ðξ Þ, B3 ðξ Þ, B4 ðξ Þ

where
 
ξ ξ
N 1 ðξ Þ ¼ 1  , N 2 ðξ Þ ¼
l l
   
6 12ξ 4 6ξ
B1 ð ξ Þ ¼  2 þ 3 , B 2 ð ξ Þ ¼  þ 2 (9.190)
l l l l
   
6 12ξ 2 6ξ
B3 ðξ Þ ¼ 2  3 , B4 ðξ Þ ¼  þ 2
l l l l
~ has components given by
then the equilibrium matrix E
ðl
~
E mn ¼ Bm ðξ ÞN n ðξ Þdξ ðm ¼ 1; 2, 3, 4Þ ðn ¼ 1; 2Þ (9.191)
0

If we carry out all the integrations of these products of shape functions, we find
2 3
1 1

6 l l 7
6 7
6 1 7
~ ¼6 6 0 7
E 7 (9.192)
6 1 17
6  7
4 l l5
0 1
 9.7 Force-Based Finite Elements for Beam Bending 333

We can use these matrices to find force-based beam solutions. However, recall we can also
define a pseudostiffness matrix and solve for the displacements. For an element this pseu-
~ ½G 1 E
dostiffness matrix, we recall, is given by K pseudo ¼ E ~ T . Let us use MATLAB®
to obtain the pseudostiffness for a beam element explicitly:
syms L E I
Em ¼ [-1/L , 1/L; -1 , 0; 1/L , -1/L; 0 , 1]; % define the
% equilibrium matrix
G ¼ (L/(E*I))*[ 1/3 , 1/6; 1/6, 1/3]; % define the
% flexibility matrix
Kp ¼ Em*inv(G)*Em.’ % compute the
% pseudo-stiffness matrix
Kp ¼ [ (12*E*I)/L^3, (6*E*I)/L^2, -(12*E*I)/L^3, (6*E*I)/L^2]
[ (6*E*I)/L^2, (4*E*I)/L, -(6*E*I)/L^2, (2*E*I)/L]
[ -(12*E*I)/L^3, -(6*E*I)/L^2, (12*E*I)/L^3, -(6*E*I)/L^2]
[ (6*E*I)/L^2, (2*E*I)/L, -(6*E*I)/L^2, (4*E*I)/L]
(L^3/(E*I))*Kp % remove the EI/L^3 factor to make
% it easier to read

ans ¼ [ 12, 6*L, -12, 6*L]

[ 6*L, 4*L^2, -6*L, 2*L^2]
[ -12, -6*L, 12, -6*L]
[ 6*L, 2*L^2, -6*L, 4*L^2]

Comparing these MATLAB® results with Eq. (9.163), which we write here again:
2 3
12 6l 12 6l
EI 6 6l 4l 2 6l 2l 2 7
½K  ¼ 3 6 7
l 4 12 6l 12 6l 5
6l 2l 2 6l 4l 2

we see that the pseudostiffness matrix is indeed the same as the stiffness matrix
obtained previously. However, we should emphasize again that the pseudostiffness
matrix is not in general always identical to the stiffness matrix used in stiffness-based
finite elements.

Example 9.9 Statically Determinate Beam – Force Method

Consider now the statically determinate problem of Fig. 9.21a (a cantilever beam carrying a
distributed force) with the force-based method. Since there is only one element, the equilib-
rium equations are just
334 Computational Mechanics of Deformable Bodies

2 3 8 q; ð1Þ 9 8 ð1Þ 9 8
1 1 9 8 9
> > > V1 >
6L > V1 > > > > q0 L=2 > > V R1 >
L 7 ) >
>
> >
> >
> >
> >
> >
> >
> >
>
6
6 1
7(
7 >
< q; ð1Þ >
>
=
>
>
< ð1Þ >
>
=
>
>
<
>
>
=
>
>
<
>
>
=
6 0 7 M 1 M 1 M 1 q 0 L 2
=12 M R1
6 7 ¼ þ ¼ þ (9.193)
6 1 1 7 M2 >
> q; ð1Þ >
> >
> ð1Þ >
> >
> q L=2 >
> >
> 0 >
>
6  7 >
> V >
> > 2 >
V > > > >
> > >
> > >
> > >
0
4 L L5 >
>
2
> > > >
: q; ð1Þ > ; > : ð1Þ > ; : q L2 =12 ; : 0 ;
M2 M2 0
0 1

where ðV R1 , M R1 Þ are the reactions at the wall. Applying the boundary conditions to this
problem we can eliminate the first two rows of these equations and find
" #( ) ( )
1=L 1=L M1 q0 L=2
¼ (9.194)
0 1 M2 q0 L2 =12

Solving this in MATLAB® we find:

syms L q
Eg ¼ [ -1/L , 1/L; -1 , 0; 1/L , -1/L; 0 , 1]; % the
% original equilibrium matrix
Er ¼ Eg;
Er(1:2, :) ¼ [ ] % generate reduced equilibrium
% matrix and show it
Er ¼ [ 1/L, -1/L]
[ 0, 1]
P ¼ [ -q*L/2 ; q*L^2/12]; % the known loads
F ¼ inv(Er)*P % solve [Er]{F} ={P} for the
% generalized forces (actually
F ¼ -(5*L^2*q)/12 % moments. Can also use F ¼ Er\P)
(L^2*q)/12
Eg*F - [ -q*L/2 ; -q*L^2/12; -q*L/2 ; q*L^2/12] % compute the
% end reactions at the wall (and zeros)
ans ¼ L*q
(L^2*q)/2
0
0

We see the solution is M 1 ¼ 5q0 L2 =12, M 2 ¼ q0 L2 =12. The reactions at the wall are also
obtained by using the original equilibrium matrix and subtracting off the distributed force
contributions. These reactions are V R1 ¼ q0 L, M R1 ¼ q0 L2 =2, which agrees with our previ-
ous results. The terms in the final MATLAB® result given above are the applied force and
moment at the free end of the beam, which are both zero.
 9.7 Force-Based Finite Elements for Beam Bending 335

Consider now the statically indeterminate problem of Fig. 9.22a with the force-
based method.

Example 9.10 Statically Indeterminate Beam – Force Method

To solve the problem of Fig. 9.22a, we use two elements as shown in Fig. 9.24 and four
unknown generalized internal forces ðM 1 , M 2 , M 3 , M 4 Þ. Assembling these two elements the
principle of virtual work gives
2 3 8 9
1 1 ð1Þ 8 9
 0 0 >
> V1 >
>
6 L L 7 >
> >
> >
> V R1 >
>
6 78 > > > >
6 1 7 M 9 > >
>
ð1Þ >
>
>
>
>
>
>
>
R1 >
6 0 0 0 7> 1> > M > > M >
6 >
7> >
> >
>
1 >
> >
> >
>
6 1 1 1 1 7>>
< > >
>
= > ð1Þ
< ð2Þ >> >
= >
<
>
>
6   7 M2 V2 þ V2 P =
6 L L L L 7 ¼ ¼ (9.195)
6 7> > > ð1Þ ð2Þ > > >
6 0 1 7
0 7>>
> M 3 >
> >
> M þ M >
> >
> 0 >
>
6 1 > > 2 2 > > >
6 7>
: >
; > >
>
>
>
>
>
>
>
>
>
>
6 1 1 7 M4 >
> ð 2Þ >
> >
> V >
>
6 0 0  7 >
> V >
> >
>
R3
>
>
4 L L 5 >
>
3
>
> : ;
: ð2Þ ; 0
0 0 0 1 M3

where again ðV R1 , M R1 Þ are the fixed end reactions and V R3 is the reaction force at the roller
support. If we generate a reduced set of “free” equilibrium equations by eliminating the first
two equations and the fifth equation in Eq. (9.195), we have
8 9
2 3 M1 > 8 9
1 1 1 1 > >
> >
>
  >
> > >
> > P > >
6L L L L7 < = < =
6 7 M2
60 1 0 5> 7 ¼ 0 (9.196)
4 1 >
> M3 >
>
>
>
>
:
>
>
;
> >
0 0 0 1 >:
M
>
; 0
4

which are three equations for the four unknown moments. Thus, we need one compatibility
equation. Note that the third equation in Eq. (9.196) gives M 4 ¼ 0 so we could eliminate
that unknown, but we will still have two equations in three unknowns so that the problem
remains statically indeterminate to one degree. We will, however, for generality keep all four
equations here, which we will write as E ~ f fF g ¼ fPg. We need to define the compatibility
equation S ~ fΔg ¼ 0 associated with these equilibrium equations and write that compatibil-
ity equation in terms of the generalized forces by using the generalized flexibility matrix,½G ,

M1, M 2 M3, M 4 Figure 9.24 The two force-based elements

used for the problem of Fig. 9.22a.
L L
M 1( ) V1(1)
1
V2( )
1
M 2(1) M 2( 2) V2( 2) V3( 2) M 3( 2)
(1) (2)
336 Computational Mechanics of Deformable Bodies

i.e., fΔg ¼ ½G fF g, where here the flexibility matrix is the diagonal concatenation of the
element flexibility matrices:
2 3
1=3 1=6 0 0
L 66 1=6 1=3 0 0 77
½G  ¼ (9.197)
EI 4 0 0 1=3 1=6 5
0 0 1=6 1=3

Appending the compatibility relation S ~ ½G fF g ¼ 0 to the free equilibrium equations, we

can then solve for the unknown generalized forces. As we have done before, placing these
forces back into the original equilibrium equations we can find the unknown reactions in
Eq. (9.197). These steps in MATLAB® are:
syms E I L P
Eg ¼ [ -1/L 1/L 0 0; % define equilibrium matrix
-1 0 0 0;
1/L -1/L -1/L 1/L;
0 1 -1 0;
0 0 1/L -1/L;
0 0 0 1];
Er ¼ Eg; % define reduced, “free” equilibrium matrix
Er(5, :) ¼ [ ]; % and show it
Er(1:2, :) ¼ [ ]
Er ¼ [ 1/L, -1/L, -1/L, 1/L]
[ 0, 1, -1, 0]
[ 0, 0, 0, 1]
null(Er) % obtain the compatibility equation for
% this statically indeterminate system
ans ¼ 2
1
1
0
Gn ¼ [ 1/3 1/6 0 0; % define the flexibility matrix with the
1/6 1/3 0 0; % L/EI factor removed since it cancels
0 0 1/3 1/6; % in the subsequent compatibility term
0 0 1/6 1/3];
SG ¼ (null(Er)).’*Gn % form up compatibility times the
% flexibility term
SG ¼ [ 5/6, 2/3, 1/3, 1/6]
Es ¼ [Er ; SG] % append this term to the free equilibrium
 9.8 Some Extensions of a Force-Based Approach 337

% equations
Es ¼ [ 1/L, -1/L, -1/L, 1/L]
[ 0, 1, -1, 0]
[ 0, 0, 0, 1]
[ 5/6, 2/3, 1/3, 1/6]
Pv ¼ [ -P; 0; 0; 0]; % define the vector of known loads
F ¼ inv(Es)*Pv % solve for the generalized forces (can
% also use F ¼ Es\Pv)
F ¼ -(3*L*P)/8
(5*L*P)/16
(5*L*P)/16
0
Eg*F % multiply generalized forces by the original
% equilibrium matrix to get reactions as well
% as original loads
ans ¼ (11*P)/16
(3*L*P)/8
-P
0
(5*P)/16
0

These results show that the reactions are V R1 ¼ 11P=16, M R1 ¼ 3PL=8, and V R3 ¼ 5P=16,
which are same as the previous values obtained with the stiffness-based method.

9.8 Some Extensions of a Force-Based Approach

We have used strength of materials problems as examples of the generation of both stiffness-
based and force-based finite elements. In the application of the finite element approach we
have solved problems using very few elements so that all the steps and intermediate results
are able to be examined. Such solutions, however, may not be possible in all cases. For
example, consider the axial-load problem of Fig. 9.25a, where a uniformly loaded bar is
fixed at both ends to rigid supports. If we try to use a single element in either a stiffness-based
approach or a force-based approach, there are no free displacements to use in the stiffness-
based method and no free equilibrium equations in the force-based method since the element
is highly constrained. Similarly, in the beam of Fig. 9.25b we cannot use a single element to
solve the problem.
338 Computational Mechanics of Deformable Bodies

q0 U1 U2

L R1 R2

(a)
q0
W1 W2
T1 T2
M R1 M R2

L VR1 VR 2
(b)
q0
W1 W2
T1 T2
M R1
L VR1 VR 2
(c)

Figure 9.25 (a, b) Two statically indeterminate problems where the use of a single element will not lead to
a solution but which can be solved with a force-based method by including the reactions. (c) A statically
indeterminate beam problem that will be used to illustrate how reactions are incorporated into the
solution. The unknown reactions and the generalized nodal displacements are also shown for
these problems.

These difficulties, however, are easily eliminated by simply choosing more elements so
that there are sufficient degrees of freedom present in each element to solve the problem. If
we want to solve such problems without using more elements there are other options. We
will see that one way is to bring the reactions into the problem explicitly as additional
unknowns. In fact, this approach will lead to a new way to solve all strength of materials
problems, not just the problems such seen in Figs. 9.25a, b. We will demonstrate this new
approach with the problem shown in Fig. 9.25c (which can be solved with a single element
and an “ordinary” force-based finite element method as well). This statically indeterminate
problem is very similar to the statically determinate beam that was solved previously. The
equations are, therefore, the same as those in Eq. (9.193) except there is now an unknown
reaction force, V R2 , at the roller. Thus, in place of Eq. (9.193) we have
2 3 8 q; ð1Þ 9 8 ð1Þ 9 8
1 1 9 8 9
 >
> V1 >
> >
> V1 > > > q0 L=2 > > V R1 >
6 L L 7( ) > > >
> > > >
> > > >
> >
> >
>
6 7 >
< q; ð1Þ >
= >
< ð1Þ >
= >
< >
= >
< >
=
6 1 0 7 M M M q L 2
=12 M
6 7 1
¼ 1
þ 1
¼
0
þ
R1
(9.198)
6 1 17 > q; ð1Þ > > ð1Þ > > q > > >
6 7 M >
> >
> > > > L=2 > > V >
4 L L5 V > > > 2 >
V > > >
> > >
2 0 R2
>
> 2 > > >
: q; ð1Þ > ; > : ð1Þ > ; : q L2 =12 ; : 0 ;
M2 M2 0
0 1
 9.8 Some Extensions of a Force-Based Approach 339

We could write these equations in the more general matrix–vector form

~ fF e g ¼ fQg þ fPg þ fRg
E (9.199)

where fF e g is the vector of generalized internal elastic forces, which in this case has as
components the moments ðM 1 , M 2 Þ, fQg, is the vector of discrete distributed loads, and fRg
is the vector of generalized reactions, which in this case has the components
ðV R1 , M R1 , V R2 ; 0Þ. We have also for generality included a set of discrete applied loads
fPg, which is of course absent in the present problem.
To bring the reactions into the problems as unknowns, we place them on the left side of
the equations in Eq. (9.198), giving
2 38 9
1 1 > M1 > 8 9
 1 0 0 >
> >
> > q0 L=2 >
6 L L 7> > > >
6 7>< M2 >
> > >
= > < q L2 =12 >
>
=
6 1 0 0 1 0 7
6 7 V R1 ¼ 0
(9.200)
6 1 7> >
6 
1
0 0 1 7>> > >
> > q0 L=2 >
> >
>
4 L 5>> M > > >
L > R1 >
: > : q L2 =12 ;
; 0
0 1 0 0 0 V R2

In the general matrix–vector form, we could write Eq. (9.199) as

8 9
  < Fe =
^ F^  E
E ~ jE
~a  ¼ fQg þ fPg (9.201)
: ;
R

where in the present problem E ^ is the equilibrium matrix appearing in Eq. (9.200) and
 
^
F is now a vector of generalized forces with components consisting of both generalized
^ is our original equi-
internal forces and generalized reactions. The equilibrium matrix E
~ ~
librium matrix E of Eq. (9.198) augmented by a matrix E a , which in this case is
2 3
1 0 0
6 7
~ a ¼ 6 0 1 0 7
E (9.202)
4 0 0 1 5
0 0 0

Now, consider how this inclusion of the reactions into the equilibrium equations affects the
principle of complementary virtual work. The complementary strain energy can be written
 
as a function of the generalized forces F^ and we have


δU c F^ ¼ δW cv ¼ fδQgT fU g þ fδPgT fU g (9.203)

The displacements fU g contain the generalized displacements associated with the displace-
ment degrees of freedom present in the problem. In the present case there are four displace-
ments ðU 1 , U 2 , U 3 , U 4 Þ in fU g, where U i is the ith generalized displacement associated
with the ith equilibrium equation at a node. The first equilibrium equation, for example, was
summation of forces in the vertical direction at the wall so that U 1 ¼ W 1 is the vertical
340 Computational Mechanics of Deformable Bodies

displacement at the wall (see Fig. (9.25c). Similarly, U 2 ¼ θ1 , the slope at the wall, and in the
same fashion U 3 ¼ W 2 , U 4 ¼ θ2 (Fig. 9.25c). The variation of the complementary strain
energy can be written as

 X ∂U c
δU c F^ ¼ δF^ i (9.204)
∂F^ i
We could also write Eq. (9.204) as


  T ∂U c
δU c F^ ¼ δF^ (9.205)
∂F^
 
where δF^ now contains variations of both the internal elastic forces as well as the
n co
reactions, and ∂U^ contains both derivatives with respect to the internal elastic forces as
∂F
well as the reactions, i.e.,
8 9
8 c 9 > ∂U c =∂M 1 >
> ∂U > >
> >
>
>
> >
> >
> ∂U c
=∂M >
>
 c >
< ∂F e >
= >
< 2 >
=
∂U
¼   ¼ ∂U =∂V R1
c
(9.206)
∂F^ >
>
> ∂U c
>
>
>
>
> ∂U c =∂M >
> >
>
>
: >
; >> R1 >
>
∂R >
: >
;
∂U c =∂V R2

where the last column vector on the right-hand side of Eq. (9.206) is the explicit expression
for the present problem. From the equations of equilibrium, Eq. (9.201), we have
 
^ δF^ ¼ fδQg þ fδPg
E (9.207)

From the complementary strain energy principle, Eq. (9.203) and from Eq. (9.205) and
Eq. (9.207), it follows

 T ∂U c  T
δF^ ¼ δF^ ^ T fU g
E (9.208)
∂F^
 
which can be satisfied for all δF^ when
 c
∂U ^ T fU g
¼ E (9.209)
∂F^

Equation (9.209) is in the form of a strain–displacement relation if we define generalized

 
^ associated with both the internal forces and reactions as
strains Δ
 c
  ∂U
^
Δ ¼ (9.210)
∂F^
or, more explicitly in component form
∂U c ðF e ; RÞ ∂U c ðF e ; RÞ
Δei ¼ , ΔRi ¼ (9.211)
∂F ei ∂Ri
 9.8 Some Extensions of a Force-Based Approach 341

Because the internal forces, reactions, and applied loads are related to each other through
the equilibrium equations, we can write the complementary strain energy U c ðF e ; RÞ in terms
of different combinations of the internal forces and reactions. Thus, in a given problem the
complementary strain energy will not depend on all the elastic internal forces and reactions,
and in Eq. (9.211) values of the subscript i will run only over those generalized forces
actually present. For the present problem we have a uniform beam element where
ðl
1
U ¼
c
½M ðξ Þ2 dξ (9.212)
2EI
0

We can write the internal moment, for example, solely in terms of the internal elastic
moments as
M ðξ Þ ¼ M 1 þ ðM 2  M 1 Þξ=l (9.213)

so that U c ¼ U c ðM 1 , M 2 Þ. The internal moment terms ðM 1 , M 2 Þ are the same types of

generalized internal forces we have considered before and we can relate those moments to
corresponding generalized internal strains ðΔM 1 , ΔM 2 Þ through a flexibility matrix, ½G .
Previously, we calculated this flexibility matrix through its definition in terms of the shape
functions (see Eq. (9.68)). However, we can also compute this matrix with the use of Eq.
(9.211) which now becomes explicitly
∂U c ∂U c
ΔM 1 ¼ , ΔM 2 ¼ , ΔV R1 ¼ ΔM R1 ¼ ΔV R2 ¼ 0 (9.214)
∂M 1 ∂M 2
We can place Eq. (9.213) into Eq. (9.212) and then use Eq. (9.214). Carrying out the
integration, it is easy to show that we find the same flexibility matrix as before (see Eq.
(9.188)), namely
 
ΔM 1 l 1=3 1=6 M1
¼ (9.215)
ΔM 2 EI 1=6 1=3 M2

for the generalized internal elastic strains ðΔM 1 , ΔM 2 Þ associated with the generalized internal
forces ðM 1 , M 2 Þ. Note that the relationships between the elastic generalized strains and the
derivatives of the complementary strain energy terms in Eq. (9.214) are similar to the results
we obtained in Chapter 8 for the relationship between the internal strains and the derivative
of the complementary strain energy density with respect to the stresses (Eq. (8.46a)) which
we rewrite here:
∂uc0 ðσ Þ
eI ¼ ðI ¼ 1; 2; . . . , 6Þ (9.216)
∂σ I
so that we could consider Eq. (9.214) to be analogous to Eq. (9.216) in terms of generalized
strains and generalized forces.
Writing the complementary strain energy in terms of ðM 1 , M 2 Þis only one possibility. We
could instead write the complementary strain energy in terms of one or more of the
342 Computational Mechanics of Deformable Bodies

Figure 9.26 (a) Free body diagram q0 q0

involving the reaction force at the
roller of Fig. 9.25c. (b) Free body
diagram involving the reaction M(x) M R1
M(x)
force and moment at the fixed wall
of Fig. 9.25c. V(x) V(x)
x VR 2 x
VR1

(a) (b)

reactions. For example, we can write the internal moment in terms of the reaction force at
the roller as
M ðxÞ ¼ V R2 x  q0 x2 =2 (9.217)

(see Fig. 9.26a). In this case we have the complementary strain energy given as
U c ¼ U c ðV R2 , q0 Þ. Alternatively, we could write the internal moment in terms of the reac-
tions at the fixed wall:
M ðxÞ ¼ M R1 þ V R1 x  q0 x2 =2 (9.218)

(see Fig. 9.26b) in which case U c ¼ U c ðM R1 , V R1 , q0 Þ. Thus, Eq. (9.211) gives us a very
   
general way to relate the generalized strains Δ^ to the generalized F^ .
We can determine the compatibility equations for Eq. (9.200) of our present problem by first
computing the null space matrix S ~ for the equilibrium matrix E ^ . In MATLAB® we find:

syms L q E I
Eg ¼ [ -1/L 1/L -1 0 0; % obtain the equilibrium matrix
-1 0 0 -1 0;
1/L -1/L 0 0 -1;
0 1 0 0 0];
S ¼ (null (Eg)).’ % generate the null space matrix

S ¼ [ L, 0, -1, -L, 1]

which shows that S ~ ¼ ½L; 0; 1,  L, 1. With this null space matrix and Eq. (9.210), we
then can write the compatibility equation(s) in terms of the generalized internal forces.
To summarize: if we form up a set of equilibrium equations E ^ that includes internal
forces and/or reactions and determine the compatibility matrix, S ~ from those equilibrium
equations, we can obtain the generalized forces by solving a combination of the equilibrium
equations and compatibility equations in the form:
 
^ F^ ¼ fPg þ fQg
E
 
~ Δ
S ^ ¼0
 c (9.219)
  ∂U
^ ¼
Δ
∂F^
 9.8 Some Extensions of a Force-Based Approach 343

 
where now the generalized forces F^ can include internal generalized forces, fF e g, and the
external reactions, fRg.
Let us now use this formulation to solve the problem of Fig. 9.25c, making several
different choices for how we write the complementary strain energy.

Example 9.11 Statically Indeterminate Beam – Use of Complementary

Strain Energy
As a first choice for solving the beam of Fig. 9.25c, let us express the internal bending
moment in terms of ðM 1 , M 2 Þ. This is the same choice as found in the “ordinary” force-
based finite element method. In this case, the first two generalized strains are given by Eq.
(9.215) in terms of the flexibility matrix, ½G , and the remaining generalized strains are all
zero (Eq. (9.214). Thus, the compatibility equation matrix in terms of the generalized forces
becomes:
G ¼ (L/(E*I))*[ 1/3 1/6; 1/6 1/3] % define the flexibility
% matrix
Se ¼ [ [L ,0]*[G] , 0, 0, 0] % obtain the compatibility
% equation matrix in terms of the generalized forces
Se ¼ [ L^2/(3*E*I), L^2/(6*E*I), 0, 0, 0]

If we then append this compatibility equation to the original equilibrium equations and also
modify the vector of known applied forces appropriately, we can then solve the problem for
all the generalized forces:
Es ¼ [Eg; Se] % add compatibility equation to the
% equilibrium matrix to obtain the system equilibrium matrix
Es ¼ [ -1/L, 1/L, -1, 0, 0]
[ -1, 0, 0, -1, 0]
[ 1/L, -1/L, 0, 0, -1]
[ 0, 1, 0, 0, 0]
[ L^2/(3*E*I), L^2/(6*E*I), 0, 0, 0]
Pv ¼ [-q*L/2; -q*L^2/12; -q*L/2; q*L^2/12; 0] % extend the
% applied loads
Pv ¼ -(L*q)/2
-(L^2*q)/12
-(L*q)/2
(L^2*q)/12
0
F ¼ inv(Es)*Pv % solve for the generalized forces
344 Computational Mechanics of Deformable Bodies

Figure 9.27 (a) The reactions, the applied q0 L / 2 V = ( M 1 − M 2 ) / L = −q0 L / 8

nodal force and moment, and the internal
force, V, and moment, M 1 , in the beam of
Fig. 9.25c at the wall. (b) The reactions, the
applied nodal force and moment, and the M R1 = q0 L2 / 8
internal force, V, and moment, M 2 , of the
beam of Fig. 9.25c at the roller support. M 1 = −q0 L2 / 24
VR1 = 5q0 L / 8
q0 L2 /12
(a)

q0 L / 2

M 2 = q0 L2 /12 VR 2 = 3q0 L / 8

( M R2 = 0)
q0 L2 /12

V = ( M 1 − M 2 ) / L = −q0 L / 8

(b)

F ¼ -(L^2*q)/24
(L^2*q)/12
(5*L*q)/8
(L^2*q)/8
(3*L*q)/8

These results give the moments M 1 ¼ q0 L2 =24, M 2 ¼ q0 L2 =12 with the reactions given by
V R1 ¼ 5q0 L=8, M R1 ¼ q0 L2 =8, and V R2 ¼ 3q0 L=8. There are quite a few unknowns here so
we have explicitly shown these results together with the applied nodal forces and moments from
the distributed force and the internal shear force, V, in Fig. 9.27. Results at both ends of the
beam are given in that figure, where it can be seen that all the forces and moments are in
equilibrium. Problem P9.1 asks you to also use the strain–displacement equations for this
problem to find the slope of the beam at the roller support, which is one of the generalized
displacements present. However, you should realize that one cannot always solve the generalized
 
strain–displacement relations for the displacements because the generalized forces in F^ are not
all independent and the displacements may in some cases not have a definite meaning. We will
say more about this shortly. However, here the generalized strains ðΔM 1 , ΔM 2 Þ are independent
(and all the others are zero) so, as seen in problem P9.1, we can solve the strain–displacement
relations for the correct unknown generalized displacement (rotation) in this case.
If instead of using the moments ðM 1 , M 2 Þ, we chose to write the internal moment in terms
of the reaction force at the roller, using the free body diagram seen in Fig. 9.26a, we would
have (see Eq. (9.217)):
 9.8 Some Extensions of a Force-Based Approach 345

ðL
∂U c 1

ΔV R2 ¼ ¼ V R2 x  q0 x2 =2 xdx
∂V R2 EI
0
(9.220)
3
V R2 L q L4
¼  0
3EI 8EI
ΔM 1 ¼ ΔM 2 ¼ ΔV R1 ¼ ΔM R1 ¼ 0

Since all the other derivatives of the complementary strain energy are zero, the compatibility
equation in terms of the generalized forces simply becomes ∂U c =∂V R2 ¼ 0 or, equivalently,
V R2 ¼ 3q0 L=8, which is the value of the reaction force we obtained previously. If we append
this compatibility equation (which we note is now an inhomogeneous equation, so there will
be a component of 3q0 L=8 added to the vector of loads) to the equilibrium matrix and solve
we find
Es2 ¼ [ Eg; 0 0 0 0 1] % add compatibility equation to the
% equilibrium matrix
Es2 ¼ [ -1/L, 1/L, -1, 0, 0]
[ -1, 0, 0, -1, 0]
[ 1/L, -1/L, 0, 0, -1]
[ 0, 1, 0, 0, 0]
[ 0, 0, 0, 0, 1]
Pv2 ¼ [ -q*L/2; -q*L^2/12; -q*L/2 ; q*L^2/12; 3*q*L/8]
% extend the vector of loads
Pv2 ¼ -(L*q)/2
-(L^2*q)/12
-(L*q)/2
(L^2*q)/12
(3*L*q)/8
F2 ¼ inv(Es2)*Pv2 % solve for the generalized forces

F2 ¼ -(L^2*q)/24
(L^2*q)/12
(5*L*q)/8
(L^2*q)/8
(3*L*q)/8

which is identical to the solution we obtained before.

346 Computational Mechanics of Deformable Bodies

Note that the use here of the compatibility equation leads to ∂U c =∂V R2 ¼ 0, which is
identical to simply using the principle of least work to solve for the unknown redundant
reaction force at the roller, as we did in Chapter 8 (see Fig. 8.17 and Eq. (8.164)). Thus, to
solve problems where we want to include unknown redundant reactions explicitly, we could
alternatively use the principle of least work directly, as done in Chapter 8, to augment the
equations of equilibrium and produce a determinate set of equations we can solve. However,
the use of the compatibility equation is more general than using the principle of least work in
that we can use unknowns that do not have to be independent redundants. For example, see
problem P.9.2 where you are asked to write the complementary strain energy in terms of the
reactions ðM R1 , V R1 Þto solve for the generalized forces in the current problem.
Both in the present case, where the complementary strain energy is written in terms of
V R2 , and when we write the complementary strain energy in terms of ðM R1 , V R1 Þ, if we try
to solve the strain–displacement relations for the displacements present we may find incor-
rect values for those displacements since the generalized strains, as defined here, are not all
independent. See, for example, problem P9.2 for the case involving ðM R1 , V R1 Þ. This,
however, does not affect the solution of Eq. (9.219) for the generalized forces. Also, in some
cases the displacements appearing in the strain–displacement relations may not have a clear
meaning. For example, consider the axial-load problem of Fig. 9.25a. The equilibrium
equation in matrix–vector form is

R1
½1 1 ¼ q0 L
R2

and the corresponding generalized strain–displacement relation is


ΔR 1 1
¼ U
ΔR 2 1

However, what is the displacement, U, here? All that we can say is that U is an
x-displacement degree of freedom associated with the force equation in the x-direction.
If we write the complementary strain energy in terms of either R1 or R2 , then these strain–
displacement relations will simply yield U ¼ 0. Thus, in general we can solve Eq. (9.219)
for the generalized forces but we should normally determine the displacements separ-
ately by other means.
Equation (9.219) is an extension of the formulation used previously in force-based finite
elements where all the unknown generalized forces were internal generalized elastic forces
fF e g that could be related to internal generalized strains through a flexibility matrix,½G , i.e.,
where Eq. (9.219) becomes
~ fF e g ¼ fPg þ fQg
E
~ fΔe g ¼ 0
S (9.221)
fΔe g ¼ ½G fF e g
 9.8 Some Extensions of a Force-Based Approach 347

where S ~ is the compatibility matrix associated with the equilibrium matrix E ~ and fΔe g
are the elastic strains associated with the elastic forces. Since the strains play only an
intermediate role in either Eq. (9.219) or Eq. (9.221) and can be eliminated, either of these
equations can be considered to be purely force-based methods, similar to the Matrix–Vector
3-D Governing Set 3 described in Chapter 6. The strain–displacement relations also do not
play a direct role in either of these formulations since the compatibility matrix S~ can be
calculated directly from the equilibrium matrix, so successful use of either Eq. (9.219) or
Eq. (9.221), as mentioned previously, does not depend on our ability to solve those strain–
displacement relations for the displacements. In fact, all of the statically indeterminate
problems that commonly appear in strength of materials texts can be solved for the
generalized forces present by the purely forced-based approaches of either Eq. (9.219) or
Eq. (9.221). Once all the unknown generalized forces are obtained in this fashion, one way
we can obtain the generalized displacements is through integration, as is normally done in
elementary strength of materials courses. In bending problems, for example, one integrates
the moment–curvature relationship twice and applies boundary conditions to obtain the
beam slope and vertical deflection. For a comparison of the use of Eq. (9.219) with other
methods commonly used in advanced or elementary strength of materials courses, see
problem P9.3.

Example 9.12 Specified Displacements at Supports – Force Method

Another situation where it is useful to bring the reactions into the problem formulation is in
an ordinary force-based finite element solution when there are nonzero displacements
specified at the supports of a body. As a simple example, consider the bar shown in
Fig. 9.28a where a force P acts at the center of the bar and the ends of the bar have known
displacements U 1 ¼ U  1 ,U 3 ¼ U
 3 . The unknown displacement at the force P is U 2 . If we let
the reaction forces at those supports be ðRA , RB Þ and consider two elements as shown in
Fig. 9.28b, then the equilibrium equations are
2 3 8 9
1 0  < RA =
4 1 1 5 F 1 ¼ P (9.222)
F2 : ;
0 1 RB

where ðF 1 , F 2 Þ are the internal forces in the two elements. We can incorporate the reactions
into the equilibrium matrix by rewriting Eq. (9.222) as
8 9
2 3> RA > 8 9
1 1 0 0 < > >
= <0=
4 0 5 F1
1 1 0 ¼ P (9.223)
> F2 > : ;
0 0 1 1 > : >
; 0
RB

which we can again write symbolically as

 
^
E F^ ¼ fPg (9.224)
348 Computational Mechanics of Deformable Bodies

Figure 9.28 (a) A bar acted upon by a load P at its center and U1
 1, U
 3 Þ. (b) U3
where the ends have specified displacements ðU
The reaction forces, internal forces and nodal displacements.
A B
P

L L

(a)

U1 = U1 U2 U3 = U3

F1 F2
RA RB
P
(b)

The generalized strains ðΔA , Δ1 , Δ2 , ΔB Þ associated with the forces ðRA , F 1 , F 2 , RB Þ can then
be related to the nodal displacements ðU 1 , U 2 , U 3 Þ through the transpose of the equilibrium
matrix in Eq. (9.223) to give the strain–displacement relations as
8 9 2 3
>
> ΔA >> 1 0 0 8 9
< = 6 7< U 1 =
Δ1 1 1 0
¼6 7 U (9.225)
> Δ2 >
> > 4 0 1 1 5: 2 ;
: ; U3
ΔB 0 0 1

which we can also write symbolically as

 
Δ^ ¼ E
^ T
fU g (9.226)

Now, consider the second and third equations of these strain–displacement relations, which
we write as
 
Δ1 1 0 1
U 1
¼  3 þ 1 fU 2 g (9.227)
Δ2 0 1 U

We will express Eq. (9.227) symbolically as

T 
~R
fΔe g ¼ E
T
gþ E
fU ~f Uf (9.228)
 g is a vector
where fΔe g is the vector of generalized internal “elastic” strains for the body, fU
 
of the given displacements, and U f are the free displacements, which in this case are just
U 2 . The matrix E~ f ¼ ½ 1 1  is the “free” equilibrium matrix (obtained by eliminating the
equilibrium equations associated with the reactions from Eq. (9.222)), as seen previously in
other problems. We will discuss the meaning of the matrix E ~ R T shortly. Thus, in general, by
first including the reactions (and their associated known displacements), we can always
generate a problem with specified nodal displacements having the form
 9.8 Some Extensions of a Force-Based Approach 349

~ f fF e g ¼ fQg þ fPg
E
T 
~R
fΔe g ¼ E
T
gþ E
fU ~f Uf (9.229)

fΔe g ¼ ½G fF e g

where now fF e g ¼ ½F 1 , F 2 T is the vector of the internal forces only, ½G  is the compliance
matrix that relates those internal forces to the elastic strains, and fPg is a vector of the
known applied generalized forces only, which is the vector obtained by eliminating the
reaction forces from the right-hand side of Eq. (9.222). For generality, we have also included
nodal forces fQg due to any distributed forces that may be present. By introducing the
compatibility matrix S ~ associated with the free equilibrium matrix E ~ f we can eliminate
the free displacements and obtain a set of inhomogeneous compatibility equations (which is
a single equation in the present case)

S ~ E
~ ½G fF e g ¼ S ~R T
g
fU (9.230)

which then leads to the solvable set of equations

2 3 8 9
~f
E < fQg þ fPg =
E~ system fF e g  4  5fF e g ¼  (9.231)
~ ½G  : ~ ~ T  ;
S S E R fU g

In our case S ~ ¼ ½ 1 1 . Note that the compatibility equation, Eq. (9.230), written in terms
~ fΔe g ¼ S
of the elastic strains is S ~ E~ R T fU
 g, which gives, explicitly for our problem,
3 U
Δ1 þ Δ2 ¼ ðU  1Þ (9.232)

This makes sense since Eq. (9.232) says that the sum of the generalized elastic strains
(elongations) of the two elements is equal to the overall elongation of the entire bar. If we
express Eq. (9.231) for this problem in MATLAB® we can solve it symbolically:
syms E A L P U1 U3
Es ¼ [1 -1 ; 1 1]; % system equilibrium matrix(note
% that the compatibility equation is multiplied by EA/L)
Pv ¼ [ P; E*A*(U3-U1)/L]; % right hand side of known loads
% and displacements
inv(Es)*Pv % solve for F1, F2
ans ¼ P/2 - (A*E*(U1 - U3))/(2*L)
- P/2 - (A*E*(U1 - U3))/(2*L)

In this case we see that the internal forces in the two elements are
P AE   1Þ
F1 ¼ þ ðU 3  U
2 2L
(9.233)
P AE   1Þ
F2 ¼  þ ðU 3  U
2 2L
350 Computational Mechanics of Deformable Bodies

We could also set up and solve the extended equilibrium equations of Eq. (9.223) by
appending the compatibility equation to those equations:
Es2 ¼ [-1 -1 0 0; 0 1 -1 0; 0 0 1 -1; 0 1 1 0]; % use system
% equilibrium matrix with reactions
Pv2= [0; P; 0; E*A*(U3-U1)/L] % new right-hand side with
% known loads and displacements where the compatibility
% equation is again multiplied by EA/L
inv(Es2)*Pv2 % solve for RA, F1, F2, RB
ans ¼ (A*E*(U1 - U3))/(2*L) - P/2
P/2 - (A*E*(U1 - U3))/(2*L)
- P/2 - (A*E*(U1 - U3))/(2*L)
- P/2 - (A*E*(U1 - U3))/(2*L)

which again leads to the internal forces of Eq. (9.233) as well as the reactions:
P EA   1Þ
RA ¼   ðU 3  U
2 2L
(9.234)
P EA   1Þ
RB ¼  þ ðU 3  U
2 2L
We can also solve for the free displacements (here the unknown displacement at the force P)
by the same process as used previously. Combining the last two equations in Eq. (9.229), we
have
 
~ R T fU
½G fF e g ¼ E gþ E ~ f T Uf (9.235)

which can be extended as

8 9
> Uf >
h i>
< >
=
~R T
 ~ T T ~ T
½G fF e g ¼ E f U g þ E f j ½G  S 
>
> >
>
: ;
f0g
8 9 (9.236)
>
> Uf > >
< =
~R
¼ E
T
 ~
fU g þ E system
T

>
> >
>
: ;
f0g

and then solved as

8 9
< Uf = h i  
T 1
 ~ system
¼ E ~R
½G fF e g  E
T
g
fU (9.237)
: ;
f0g
 9.8 Some Extensions of a Force-Based Approach 351

In our present problem it is easy to show that Eq. (9.237) gives

8 9
#1 > L
(
U2
) "
1 1
>
< F1 þ U1 >
 >
=
EA
¼
1 1 > L
> >
0 : 3 >
F2  U ;
EA (9.238)
8 9
" > L F þU
#>  > 8 PL ðU 1 þU 3Þ 9
1=2 1=2 < EA 1 1>
= < =
þ
¼ ¼ 2EA 2
1=2 1=2 > > L > : ;
: 3 >
F2  U ; 0
EA
so that we see that the displacement at the applied force P given by
PL ðU1 þU 3Þ
U2 ¼ þ (9.239)
2EA 2
This displacement is composed of two terms. The first term is the displacement at the force P
when the ends of the bar are held rigidly fixed. The second term is the displacement at the
center of the bar when the ends are displaced and the force P is absent. This center
displacement is just the average of the two end displacements.

 g at the
To summarize, we see that in a problem where there are known displacements fU
support nodes present, the unknown internal forces and free displacements are obtained
from solving Eq. (9.231) and Eq. (9.237):
2 3 8 9
~f
E < fQg þ fPg =
~ system fF e g  4  5fF e g ¼
E  (9.240)
~ ½G  : ~ ~ T  ;
S S E R fU g

8 9
< Uf = h i  
T 1
 ~ system
¼ E ~R
½G fF e g  E
T
g
fU (9.241)
: ;
f0g

which is identical in form to the cases considered earlier when the displacements at the
supports are zero (see Eq. (9.104) and Eq. (9.111)). Although we obtained Eq. (9.240) and
Eq. (9.241) for an axial-load problem, these equations can also be used in a general force-
based finite element analysis.
The effect of the support displacements is described by the presence of the E g
~ R T fU
~ T
~ T
term, so now let us consider the meaning of E R . We saw that E R appears as the matrix
multiplying the known displacements when the strain–displacement relations are reduced to
those involving only the free displacements on the right-hand side of those relations. This
procedure can always be used to obtain the E ~ R T matrix but this is a “process” type of
352 Computational Mechanics of Deformable Bodies

definition that does not reveal what this matrix represents. Consider the strain–displacement
relations, Eq. (9.226), involving all the generalized strains and displacements. We can write
this equation in matrix–vector form as
X
Δk ¼ E ik U i (9.242)

where k takes on values for all the forces present (internal and reaction forces) and i takes on
values for all the all displacements (free and known). We will drop the use of the “ ^ ”
superscripts in Eq. (9.242) and the following equations, and will use different subscript
symbols to represent specific parts of the matrix or vector terms. For example, let K
represent the values of k for all the internal forces and let I represent the values of i for all
the known displacements. Then, if we examine only those values of k ¼ K for the internal
strains corresponding to the internal forces, we can write Eq. (9.242) for those strains as
P
ΔK ¼ E iK U i
P X
¼ E IK U I þ E iK U i (9.243)
i6¼ I

which is nothing more than Eq. (9.228), since the displacements U i ði ¼6 I Þ are just the free
displacements and E iK is the transpose of the free equilibrium matrix. However, now let us
consider the equilibrium equations involving both the internal forces and reaction forces,
Eq. (9.224), which we write as
X
E ik F k ¼ Pi (9.244)

(For simplicity we will omit the distributed load terms.) Now, let us consider only those
equilibrium equations that involve the known displacement degrees of freedom (i ¼ I). They
are:
X
E Ik F k ¼ PI ¼ 0 (9.245)

since there are no applied loads at the reaction force (known displacement) nodes. We can
rewrite Eq. (9.245) as
X X
E IK F K þ E Ik F k ¼ 0 (9.246)
k6¼ K

However, we have
X
E Ik F k ¼ RI (9.247)
k6¼ K

since these terms arise simply from including the negative of the reactions RI in the equilib-
rium equations for i ¼ I so we have
X
RI ¼ E IK F K (9.248)
 9.8 Some Extensions of a Force-Based Approach 353

which we can write as

~ R fF e g
fRg ¼ E (9.249)

where we have again labeled the internal forces (due to elastic deformations) as fF e gto
explicitly distinguish them from our original set of forces fF g that included the reactions.
Thus, we see that E ~ R is just an equilibrium matrix that relates the reactions to the internal
forces. (If nodal forces from the distributed loads are present at the reactions, then those
forces, which have been neglected in our discussion, will also be present in Eq. (9.249) but
will not alter the E ~ R matrix.) Identifying the equilibrium equations from Eq. (9.249) is
therefore another way to define the matrix E ~ R . The strain–displacement relation of Eq.
(9.243) can similarly be written as
 
fΔe g ¼ E ~ R T fU
gþ E ~ f T Uf (9.250)

Equation (9.250) is of course identical to the previous result seen in Eq. (9.228).
One can obtain a different viewpoint of these strain–displacement and compatibility
equations by bringing the term involving the known displacements over to the left side
of Eq. (9.250) and defining a set of generalized strains, fΔ  g, associated with those
displacements as
g ¼  E
fΔ ~R T
g
fU (9.251)
 g are not the generalized strains ðΔA , ΔB Þ seen earlier.] Then the
[Note: these strains fΔ
strain–displacement equation can be written as
 
g ¼ E
fΔt g ¼ fΔe g þ fΔ ~ f T Uf (9.252)

where fΔt g is the total generalized strains produced by both the specified displacements as
well as the elastic behavior of the body. The compatibility equations then are simply
~ f Δt g ¼ 0
S
(9.253)
! ~ fΔe g ¼  S
S ~ fΔ
g

where S ~ is the compatibility matrix associated with the free equilibrium matrix E ~ f . Equation
(9.253) shows that with this definition of the initial strains fΔ  g the compatibility equations are
just the ordinary compatibility conditions for the total strains fΔt g. In terms of these total strains,
the compatibility equations are again homogeneous equations but when expressed in terms of the
elastic strains the equations are now inhomogeneous as seen in Eq. (9.253).
To close our discussion of the case where there are nonzero displacements at the supports,
we will also further connect the strains fΔ  g to the strain energy concepts discussed in
Chapter 8. First, consider a 1-D stress problem like the axial-load case just considered where
there is an linear elastic strain, ee , as well as a specified strain, e, due to some specified
354 Computational Mechanics of Deformable Bodies

displacements at the supports. The total strain, et , is et ¼ e þ ee . Then the total strain energy
density, u0 , is given by
ðet ðet
u0 ¼ σdet ¼ E ðet  eÞdet
e e
E
¼ ðet  eÞ2 (9.254)
2
which is just the triangular area shown in Fig. 9.29. Similarly, the complementary strain
energy density,uc0 , is given by
ðσ ðσ  
σ
uc0 ¼ et dσ ¼ þ e dσ
E
0 0 (9.255)
σ 2
¼ þ eσ
2E
which is the other area shown in Fig. 9.29. In terms of generalized strain and the internal
force, the total complementary strain energy for a constant area bar under uniaxial load is
therefore
ðL ð
L
1 Δ
U ¼
c
F dx þ
2
Fdx (9.256)
2EA L
0 0

which for a constant internal force, F, gives

F 2L 
Uc ¼ þ ΔF (9.257)
2EA
We can extend these 1-D results to a general finite element setting by noting that, if we
examine the equations of equilibrium for all the “free” nodes present in a problem, we have
~ f fF e g ¼ fPg þ fQg
E (9.258)

where E ~ f is the free equilibrium matrix and fF e g is a vector of all the internal generalized
elastic forces. The principle of complementary strain energy can also be written as

δU c ðF e Þ ¼ fδF e gT fΔe g
   
u c
0 ¼ fδRgT fU  g þ fδQgT U f þ fδPgT U f (9.259)
σ
u0
Figure 9.29 The 1-D stresses and strains for an linear elastic solid when a
E specified strain, e, is present in addition to the elastic strains, ee , due to the
applied loads. The corresponding strain energy density, u0 , and
1
complementary strain energy density, uc0 , are shown as the two areas
e ee indicated in the figure.
 9.8 Some Extensions of a Force-Based Approach 355

 g are the
where fΔe g are the internal elastic strains associated with the internal forces, fU
 
known displacements at the support nodes (where the reactions fRg are present), and U f
is the vector of free nodal displacements. Now, suppose we relate the reactions to the
internal elastic forces (see Eq. (9.249)) and define the strains fΔ  g associated with the

displacements fU g, as done in Eq. (9.251), then Eq. (9.259) becomes
   
 g ¼ fδQgT U f þ fδPgT U f
fδF e gT fΔe g þ fδF e gT fΔ (9.260)

which we could rewrite as

 Þ ¼ δW c
δU c ðFe ; Δ (9.261)
v

where δW cv is the virtual change of the complementary work done by the applied loads and
 Þ is the virtual change of the total complementary strain energy due to both the
δU c ðFe ; Δ
g
internal elastic forces and the strains due to the specified displacements. Since the strains fΔ
are fixed and hence do not vary, the total complementary strain energy, therefore, is
 Þ ¼ U c ðF e Þ  fRgT fΔ
U c ðFe ; Δ g (9.262)

which is the generalization of Eq. (9.257) that we sought.

 in terms of the known displacements (see Eq. (9.251),
If we use the definition of the strain Δ
and the relationship between the reactions and the elastic internal forces, Eq. (9.249), then we
can also write the complementary strain energy of Eq. (9.262) in the form

U c ðF e ; R; U g
 Þ ¼ U c ðF e Þ  fRgT fU (9.263)

But we can also solve for the elastic internal forces in terms of the applied loads and the
known displacements (see, for example, Eq. (9.239)) and use that solution in Eq. (9.263) to
write the complementary strain energy as
 Þ ¼ U c ðP; Q; U
U c ðP; Q; R; U  Þ  fRgT fU
g (9.264)

We can use Eq. (9.257) and Engesser’s first theorem of Chapter 8 to determine the displace-
ment at the force P for the axial load problem we have just considered in Example 9.12,
where
F 21 L  F 2L 
Uc ¼ þ Δ1 F 1 þ 2 þ Δ 2F 2 (9.265)
2AE 2AE
and ðF 1 , F 2 Þ were found in terms of P and the known displacements in Eq. (9.233), which we
rewrite here as
P AE   1Þ
F1 ¼ þ ðU 3  U
2 2L
(9.266)
P AE   1Þ
F2 ¼  þ ðU 3  U
2 2L
356 Computational Mechanics of Deformable Bodies

1 ¼ U
The generalized initial strains due to the specified displacements are simply Δ  1,
 
Δ2 ¼ U 3 (see Eq. (9.227)). Thus, from Engesser’s first theorem we have
∂U c 
U2 ¼ ¼
L P AE 
þ  1 Þ þ Δ1
ðU 3  U
∂P 2AE 2 2L 2
(9.267)


L P AE 
 þ  1 Þ  Δ2
ðU 3  U
2AE 2 2L 2

Canceling common terms, we find

PL 1 þU
U 3
U2 ¼ þ (9.268)
2AE 2
which is identical to the results obtained previously (see Eq. (9.239)). Note that if we had
used Eq. (9.264) instead to calculate this displacement, we must consider the reactions to be
functions of the load P, and we cannot treat the reactions as constants in applying Engesser’s
first theorem, as was done in Chapter 8 when discussing the principle of least work for the
case where the generalized displacements at the supports are zero.

9.9 Finite Elements for General Deformable Bodies

We have seen the use of finite elements for axial loads and bending problems. In this section
we want to outline briefly the finite element method more generally for deformable bodies.
Both stiffness-based and force-based approaches will be given.

9.9.1 Stiffness-Based Finite Elements

For a general 3-D body, the principle of virtual work gives (see Eq. (8.71b))
ðX 3 X 3 ðX
3 ðX
3
ðnÞ
σ ij δeij dV ¼ f j δuj dV þ T j δuj dS (9.269)
i¼1 j¼1 j¼1 j¼1
V V S

which can be written in matrix–vector form as

ð ð ð n o
fδegT fσ gdV ¼ fδugT ff gdV þ fδugT T ðnÞ dS (9.270)
V V S

where fσ g is a column vector of the six stresses σ I ðI ¼ 1; 2; . . . , 6Þ and feg is a similar

column vector of the six strains eI ðI ¼ 1; 2; . . . , 6Þ, as discussed previously. The vector fug
has three components ui ði ¼ 1; 2, 3Þ, and the body force vector ff g has the three components
  ðnÞ
f i ð1 ¼ 1; 2, 3Þ. Similarly, the stress vector T ðnÞ has three components T i ði ¼ 1; 2, 3Þ.
The displacements are represented in terms of a matrix of shape functions ½H  and a vector of
nodal displacements fU g as fug ¼ ½H fU g. Recall from Chapter 6 that the strains and
 9.9 Finite Elements for General Deformable Bodies 357

displacements are related through a matrix of derivatives, i.e., feg ¼ ½ℒE T fug(see Eq. (6.49)),
so we have feg ¼ ½ℒE T fug ¼ ½ℒE T ½H fU g ¼ ½BfU g, where ½B is the matrix of differenti-
ated displacement shape functions. From the stress–strain relations we have fσ g ¼ ½C feg,
where ½C  is the symmetrical 6  6 matrix of elastic constants discussed in Chapter 5. Placing
these expressions into Eq. (9.270) gives
ð ð ð n o
fδU gT ½BT ½C ½BdV fU g ¼ fδU gT ½H T ff gdV þ fδU gT ½H T T ðnÞ dS (9.271)
V V S

which must be satisfied for all variations of the nodal displacements so that we find a set of
linear equations of the form
½K fU g ¼ fQg þ fPg (9.272)

where
ð
½K  ¼ ½BT ½C ½BdV
V
ð
fQg ¼ ½H T ff gdV (9.273)
S
ð n o
fPg ¼ ½H T T ðnÞ dS
S

The matrix ½K  is the stiffness matrix, fQg is a vector of equivalent nodal body force terms,
and fPg is a set of nodal forces representing the surface stress vector. If we consider the body
here to be an element in a larger body, then we see that Eq. (9.272) is identical to the form we
obtained previously for strength of materials problems such as, for example, the axial-load
problem. Similarly, the definitions of Eq. (9.273) have analogous terms in those simpler
problems. In most cases the terms in Eq. (9.273) must be found through numerical integra-
tions so that, in a general finite element solution process, those integrations become an
important part of that process. Needless to say, the assembly process for these general
problems is more complex than the simple problems we have considered in this book. For
the stiffness matrix, the assembly process is basically placing the element stiffness matrix
terms in the appropriate positions in the overall system stiffness matrix for the body. The
stress-vector terms are integrations over the surface(s) of the element. Most of those surfaces
will be adjacent to other interior element surfaces where the stress-vector terms will cancel,
so that the only surface integrations needed for the entire body are for those surface parts of
the elements that are at the outer surface of the entire body, where either the stress vectors
are known applied values or represent unknown stress vectors corresponding to reactions
where the displacements are specified. As with the strength of materials problems we have
previously examined, if we take Eq. (9.272) as representing the entire body, the equations
involving unknown reactions can be eliminated by applying the displacement boundary
358 Computational Mechanics of Deformable Bodies

conditions and the remaining equations solved for the “free” nodal displacements. The
strains and hence the stresses then are also expressible in terms of these nodal displacements.
This, in a nutshell, is the stiffness-based finite element method as applied to general 3-D
stress analysis problems.

9.9.2 Force-Based Finite Elements

The force-based approach relies on both the principle of virtual work and the principle of
complementary virtual work. Thus, in addition to Eq. (9.270) we must examine the comple-
mentary virtual work principle. From Eq. (8.96) we found
ðX3 X 3 ðX
3 ðX
3
ðnÞ
eij δσ ij dV  δT i ui dS  δf i ui dV ¼ 0 (9.274)
i¼1 j¼1 i¼1 i¼1
V S V

As in the strength of materials problems we previously examined, we will consider this

principle for variations of the stresses that satisfy homogeneous equilibrium equations
(δf i ¼ 0), so when we write this principle in matrix–vector form we have
ð ðn oT
T
fδσ g fegdV ¼ δT ðnÞ fugdS (9.275)
V S

Thus, when we write the stresses in terms of a vector of generalized “force” parameters, fF g
(which are not values defined at nodes, as mentioned previously), as
fσ g ¼ ½N fF g (9.276)

we will use shape functions ½N  that satisfy the homogeneous equilibrium equations, i.e.,
½ℒE ½N  ¼ 0. The principle of virtual work now becomes, with the use of the displacement
and strain expressions defined previously,
ð ð ð n o
fδU g ½B ½N dV fF g ¼ fδU g ½H  ff gdV þ fδU g ½H T T ðnÞ dS
T T T T T
(9.277)
V V S

which for arbitrary variations of the nodal displacements gives the set of equilibrium
equations for the generalized forces
ð ð ð n o
½BT ½N dV fF g ¼ ½H T ff gdV þ ½H T T ðnÞ dS (9.278)
V V S

which we write more succinctly as

~ fF g ¼ fQg þ fPg
E (9.279)
~ ,
where fQg and fPg have already been defined and we also have the equilibrium matrix, E
given by
 9.9 Finite Elements for General Deformable Bodies 359

ð
~ ¼
E ½BT ½N dV (9.280)
V

Now, consider the principle of complementary virtual work, Eq. (9.275). Using the stress–
strain law written as feg ¼ ½Dfσ g in terms of the 6  6 symmetric compliance matrix, ½D,
we find
ð ðn oT
T T
fδF g ½N  ½D½N dV fF g ¼ δT ðnÞ ½H dS fU g (9.281)
V S

However, from Eq. (9.278) since the variations of the stresses must satisfy the homogeneous
equilibrium equations, we have
ð n o
~ fδF g ¼ ½H T δT ðnÞ dS
E (9.282)
S

or, taking the transpose of Eq. (9.282),

ðn oT
~
fδF gT E
T
¼ δT ðnÞ ½H dS (9.283)
S

Using Eq. (9.283) in Eq. (9.281) and the fact that the resulting equation must be satisfied for
all variations of the unknown forces, we obtain
~
½G fF g ¼ E
T
fU g (9.284)

where ½G  is the flexibility matrix defined as

ð
½G  ¼ ½N T ½D½N dV (9.285)
V

We can also define a vector of generalized strains, fΔg, associated with the generalized
forces, fF g, through
fΔg ¼ ½G fF g (9.286)

in which case we obtain the generalized strain–displacement relations as

~
fΔg ¼ E
T
fU g (9.287)

Thus, the three basic relations we have developed for the force-based method are:
~ fF g ¼ fQg þ fPg
E
~
f Δg ¼ E
T
fU g (9.288)

fΔg ¼ ½G fF g
360 Computational Mechanics of Deformable Bodies

We can consider Eq. (9.288) as the equations for a single element or, if we assemble all those
elements, Eq. (9.288) can represent the relations for the entire body where the flexibility
matrix will be the concatenation (along the diagonal of the of the flexibility matrix for the
entire body) of the element flexibility matrices, as we have seen in the strength of materials
problems. The assembly of the equilibrium matrices of the elements, as we have seen in the
strength of materials problems, is different in detail from the assembly process for the
stiffness-based method but again it is basically the placing of the element equilibrium
matrices in the proper places of the overall equilibrium matrix for the entire body. If we
consider Eq. (9.288) as the assembled equations for the entire body that includes reactions as
well as the elastic internal generalized forces, then we can eliminate the unknown reaction
forces from those equilibrium equations, producing a “free” equilibrium matrix E ~ and
 f
similarly eliminate the known displacements, leaving the “free” displacements U f . We
then obtain these equations in the form
~ f fF e g ¼ fQg þ fPg
E
 
fΔg ¼ E ~ f T Uf (9.289)

fΔg ¼ ½G fF e g

where fPg now will contain only known applied loads and fF g contains only the internal
forces, which we have previously called fF e g. To solve these equations, we obtain the
compatibility equations by eliminating the free displacements from the strain–displacement
relations to produce the compatibility equations
~ fΔg ¼ S
S ~ ½G fF e g ¼ 0 (9.290)

which we append to the “free” equilibrium equations to obtain the final system of equations
2 3 8 9 8 9
~f
E <Q= < P=
~ system fF e g  4  5fF e g ¼  þ 
E (9.291)
~ ½G  : ; : ;
S 0 0

These equations can then be solved for the unknown generalized forces. As demonstrated
earlier, we can also obtain the unknown free nodal displacements from these forces through
(see Eq. (9.111))
8 9
< Uf =  1
 ¼ E ~ system T ½G fF e g (9.292)
: ;
f0g

Note that the inverse and the transpose in Eq. (9.292) can be interchanged without changing
the end result. Also note that we can easily include specified displacements at nodes by
including terms similar to those discussed in the previous section but we will not show that
generalization here.
 9.10 The Boundary Element Method 361

We have also shown previously that we can develop a pseudostiffness type of approach
with the force-based method. From the equations in Eq. (9.289) we can express the forces in
terms of the displacements and arrive at the equations
 
~ f ½G 1 E
E ~ f T U f ¼ fQg þ fPg (9.293)

which can be written as

 
K pseudo Uf ¼ fQg þ fPg (9.294)

where the pseudostiffness matrix, K pseudo , is

~ f ½G 1 E
K pseudo ¼ E ~f T
(9.295)

Thus, all of the equations for a general deformable body using the force-based finite element
approach are identical in form to the equations seen in the strength of materials problems.
One of the crucial parts of the force-based method is to obtain a set of shape functions for
the stresses that satisfy the homogeneous equilibrium equations, i.e., ½ℒE ½N  ¼ 0. For the
strength of materials problems, this satisfaction was very easy but it is less so for more
general problems. Fortunately, one can use stress functions to help define the needed shape
functions. For 2-D problems, for example, the Airy stress function can be employed. The
details are beyond the scope of this book but the reader can find further information in [1]
about the conditions that must be met for acceptable shape functions. Determining the
compatibility matrix, S ~ is also an important step. As we have seen, this entails determining
the null space of the equilibrium matrix, E ~ f , or explicitly eliminating the free displace-
 
ments from the generalized strain–displacement relations fΔg ¼ E ~ T U f to produce the
compatibility equations S ~ fΔg ¼ 0. Neither the shape function requirements nor the need
to generate a compatibility matrix prevent the force-based finite element method from being
a very general computational tool capable of solving complex engineering problems (see [4],
[5], [6]). It is hoped that the coverage of the force-based method in this book, which is a
departure from the numerical stress analysis discussions found in most texts, can help to
raise the awareness of this method.

9.10 The Boundary Element Method

Navier’s equations are partial differential equations for the displacements that, together with
appropriate boundary conditions, form the governing equations of a deformable body. The
algebraic equations of stiffness-based or force-based finite elements together with boundary
conditions provide an alternate set of (approximate) governing equations. There is, however,
another formulation of the governing equations based on integral equations defined over the
surface of a deformable body that can be used effectively. If those surface integral equations are
solved approximately by breaking the surface of a body into elements and replacing the integral
equations by a set of algebraic equations, this method is called the boundary element method.
362 Computational Mechanics of Deformable Bodies

The starting point for formulating the boundary element method is usually the reciprocity
theorem for a linear elastic body developed in Chapter 8 (see Eq. (8.175) where here we have
ðmÞ ðmÞ
let T j ¼ tj ), which is a relationship between two different solutions (labeled (1) and (2))
in that body, namely
3 ð
X X3 ð X3 ð X3 ð
ð1Þ ð2Þ ð1Þ ð2Þ ð2Þ ð1Þ ð2Þ ð1Þ
tj uj dS þ f j uj dV ¼ tj uj dS þ f j uj dV (9.296)
j¼1 j¼1 j¼1 j¼1
S V S V

Now consider taking solution (2) to be a set of body forces given by

ð2Þ
fj ¼ ej δðx  yÞ ðj ¼ 1; 2, 3Þ (9.297)

where ej is a unit vector acting in the xj direction and δðx  yÞ is a generalized “function”
representing a concentrated body force acting at a point y in the body. This generalized
“function” has the property that, for x¼ 6 y δðx  yÞ ¼ 0 and for any function gðyÞ,
ð 
gðxÞ y inside V
gðyÞδðx  yÞdV ðyÞ ¼ (9.298)
0 y outside V
V

The three body forces will each generate three displacements in the body, producing a 3  3
matrix of displacements, U ij ðx; yÞ (see Fig. 9.30), where the j subscript represents the xj direction
of the displacement at a point x in the body, and the i subscript represents the direction (along
the xi -axis) of the concentrated body force acting at point y. Similarly, these body forces will
produce a 3  3 matrix of stress vectors, T ij , as shown in Fig. 9.31. The total displacement
ð2Þ P ð2Þ P
generated by these unit forces is uj ¼ 3i¼1 U ij ei and the total stress vector is tj ¼ 3i¼1 T ij ei .
For an isotropic linear elastic solid, one can obtain explicit expressions for these point-source
solutions (which are also called fundamental solutions) given by [7]:
1 ∂r ∂r
U ij ðx; yÞ ¼ ð3  4νÞδij þ
16πG ð1  νÞr ∂xi ∂xj
  (9.299)
1 ∂r ∂r ∂r ∂r ∂r
T ij ðx; yÞ ¼ ð 1  2ν Þδij þ 3 þ ð 1  2ν Þ n i  nj
8πð1  νÞr2 ∂n ∂xi ∂xj ∂xj ∂xi

where ðG; νÞ are the shear modulus and Poisson’s ratio, respectively, δij is the Kronecker
delta, ni are components of a unit normal, and r ¼ jx  yj is the distance between points x
and y (Figs. 9.30 and 9.31). Placing these fundamental solutions into Eq. (9.296) as solution
(2) we find
0 1 0 1
X3 X 3 ð 3 X
X 3 ð
@ U ij ðxs ; yÞtj ðxs ÞdS ðxs ÞAei þ @ U ij ðx; yÞf j ðxÞdV ðxÞA ei
i¼1 j¼1 i¼1 j¼1
S V
0 0 1 1 (9.300)
X
3 X
3 ð X
3 ð
¼ @ T ij ðxs ; yÞuj ðxs ÞdS ðxs ÞA ei þ @ ui ðxÞδðx  yÞdV ðxÞA ei
i¼1 j¼1 i¼1
S V
 9.10 The Boundary Element Method 363

U12 Figure 9.30 The components of

x = ( x1 , x2 , x3 ) the displacements,U ij , in a
deformable body at point x, when
r a concentrated body force at point
1 U11
U13 U 22
y in the body acts (a) in the x1 -
x2 , y2 y = ( y1 , y2 , y3 ) x = ( x1 , x2 , x3 ) direction, (b) in the x2 -direction,
r and (c) in the x3 -direction.
(a) 1 U 21
U 23
y = ( y1 , y2 , y3 )
x1 , y1
x3 , y3 U 32 (b)
x = ( x1 , x2 , x3 )
r U 31
1 U 33
y = ( y1 , y2 , y3 )

(c)

T12 Figure 9.31 The components of

x = ( x1 , x2 , x3 ) the stress vector,T ij , in a
deformable body at point x, when
r a concentrated body force at point
1 T11
T13 T22
y in the body acts (a) in the x1 -
x2 , y2 y = ( y1 , y2 , y3 ) x = ( x1 , x2 , x3 ) direction, (b) in the x2 -direction,
r and (c) in the x3 -direction.
(a) 1 T21
T23
y = ( y1 , y2 , y3 )
x1 , y1
x3 , y3 T32 (b)
x = ( x1 , x2 , x3 )
r T31
1 T33
y = ( y1 , y2 , y3 )

(c)

where xs is an arbitrary point on the surface, S, and x is an arbitrary point in the volume, V.
We have dropped the (1) superscript on solution (1), which we will take as the problem we
wish to solve for the deformable body. Using the sampling properties of the generalized delta
function δðx  yÞ given by Eq. (9.298) and the fact that Eq. (9.300) must be true for each
component,ei ði ¼ 1; 2, 3Þ separately, we find
364 Computational Mechanics of Deformable Bodies

3 ð
X 3 ð
X
U ij ðxs ; yÞtj ðxs ÞdS ðxs Þ þ U ij ðx; yÞf j ðxÞdV ðxÞ
j¼1 j¼1
S V
(9.301)
3 ð
X
¼ T ij ðxs ; yÞuj ðxs ÞdS ðxs Þ þ ui ðyÞ ði ¼ 1; 2, 3Þ
j¼1
S

Equation (9.301) is called Somigliana’s identity. It can be written more compactly in matrix–
vector notation as
ð ð
fuðyÞg ¼ ½U ðxs ; yÞftðxs ÞgdS ðxs Þ  ½T ðxs ; yÞfuðxs ÞgdS ðxs Þ
S S
ð (9.302)
þ ½U ðx; yÞff ðxÞgdV ðxÞ
V

If the stress vector, ftg, and the displacements, fug, are known on the surface (as well as the
body force, ff g, acting in the volume) then Eq. (9.302) is an integral representation that
gives the displacement at any point, y, in the body. Of course, in general we do not know
both the surface displacements and stress vector on the surface, so before we can use
Eq. (9.302) we must solve for whatever displacements of stress-vector components are not
known. As it stands, Eq. (9.302) is a mixture of surface and volume integrals. For many
body forces, however, we can transform the volume integral into an equivalent surface
integral (see [8], for example, for the details) in which case Eq. (9.302) will be a purely
surface integral relationship. This will obviously also be the case if there are no body forces.
In either case, the body force term will be a known integral so it does not play a major role in
the remainder of the solution process. Thus, we will henceforth drop this term to keep our
discussion focused on the other important terms.
We can turn the integral relationship of Eq. (9.302) into an integral equation by taking the
limit as the point y in the volume, V, goes to a point, ys , on the surface. However, this limit
must be done with care since the fundamental solution displacements and stress-vector terms
are singular when xs ¼ ys (see the expressions in Eq. (9.299) where the distance, r, appears in
the denominator). We will not go into the details of this process but simply write down the
end result, namely
ð ð
½C fuðys Þg ¼ ½U ðxs , ys Þftðxs ÞgdS ðxs Þ  ½T ðxs , ys Þfuðxs ÞgdS ðxs Þ (9.303)
S S

where there is a known constant 3 3 matrix, ½C , that depends on the geometry of the
surface at point ys . If the surface is smooth at that point ½C  is just the scalar value of 1/2.
Both integrals in Eq. (9.303) are singular integrals where the integral involving ½U  is weakly
singular while the integral containing ½T  is highly singular and must be interpreted as a
Cauchy principal value integral. Again, we will not go into the details of how those integrals
 9.10 The Boundary Element Method 365

are calculated but will refer you to some of the many texts (for example, [7], [8]) that describe
the integration procedures explicitly.
Equation (9.303) is a surface (boundary) integral equation that we can use to find
whatever displacement or stress-vector components are unknown. For example, if the stress
vector is known for the entire surface, then Eq. (9.303) is an integral equation (called an
integral equation of the second kind [9]) for the unknown surface displacements. If, instead,
the displacements were given for the entire surface, Eq. (9.303) is an integral equation for the
unknown stress-vector components (called an integral equation of the first kind [9]). Other
combinations of known displacement and stress-vector components similarly lead to integral
equations for whatever are the unknown displacements or stress-vector components. To
solve for the unknowns in these integral equations, we can break the surface (boundary) up
into elements where we assume the displacement and stress-vector components have a
simple, known behavior within each element. This is obviously similar to what we do in
finite elements so this type of solution is called the boundary element method. For example,
we could take both the displacement and stress vectors to be simply constants within a
surface element, S j , that are equal to their values at the centroid of the element. In that
simple case then Eq. (9.303) becomes a system of linear equations
h in o X N ð h  i n o X N ð h  i n o
C ðiÞ uðiÞ ¼ U xs ; yðiÞ dS ðxs Þ tðj Þ  T xs ; yðiÞ dS ðxs Þ uðjÞ (9.304)
j¼1 j¼1
Sj Sj

for the known and unknown centroidal (nodal) values of displacements and stress-vector
components of the N elements. These known and unknown values can be rearranged to
produce a set of simultaneous linear equations for the unknowns that can then be solved.
Once this solution is available, then all the displacements and stress-vector components in
Eq. (9.302) are known and can be used to obtain the displacements at any point within the
body. If the displacement expression of Eq. (9.302) is differentiated, then a similar integral
relationship for the stresses can be obtained and used to find those interior stress values at
any point as well.
This is a very abbreviated description of the boundary element method. Although we have
examined only the case when the shape functions were constants for the elements, other
shape functions can be used to make the method more efficient. As we have seen, the method
basically relies on the representation of the unknowns as a superposition of known solutions
(the fundamental solutions) to generate the requisite equations for those unknowns. Thus,
the method requires that we know those fundamental solutions and, since it relies on
superposition, it typically means that we must be dealing with linear problems. In contrast,
the finite element method can be generalized to nonlinear as well as linear problems. One of
the main attractive features of the boundary element method is that it requires that we break
only the surface of the body into elements, not the entire volume. This usually means that the
discretization of the geometry of the body is easier with boundary elements than with finite
elements. This feature also makes it easier to handle problems with boundary elements
where the geometry is highly irregular, such as in cracked bodies. As we have seen in this
366 Computational Mechanics of Deformable Bodies

short description, the integrations involved in boundary elements are a very important part
of the method since one must deal with fundamental solutions and their derivatives, which
are singular. Thus, the details of these integrations make the boundary element method
inherently more “mathematical” than finite elements, where there are also integrals involved
but where those integrals are normally well-behaved.

9.11 PROBLEMS

P9.1 Consider the statically indeterminate beam problem of Fig. 9.25c, which was solved for
the generalized internal forces and reactions using the internal moments ðM 1 , M 2 Þ as
the fundamental unknowns in the complementary strain energy expression in Example
9.11 in Section 9.8.
(a) Obtain the unknown slope of the beam at the roller (which is one of the generalized
displacements present in this problem) by extending the strain–displacement rela-
tions, as done previously, and solving in MATLAB® the set of equations
8 9
  < fU g =
^ ¼ E
Δ ^ system T  (P9.1)
: ;
f0g

for the generalized displacements fU gwhere E ^ system is the matrix of equilibrium

equations that include the internal moments and reactions together with the
compatibility equation.
(b) Use the solution for the generalized forces obtained in Section 9.8 for this problem
and the unit-load method of Chapter 8 to obtain the slope of the beam at the roller
and verify that it agrees with the solution of Eq. (P9.1).
P9.2 Consider the statically indeterminate beam problem of Fig. 9.25c which was solved for
the generalized internal forces and reactions using either the internal moments
ðM 1 , M 2 Þor the reaction force, V R2 , as the fundamental unknowns in the comple-
mentary strain energy expression in Example 9.11 in Section 9.8.
(a) Write the complementary strain energy instead in terms of the reactions ðV R1 , M R1 Þ
at the wall and use it to obtain the relationship between the generalized strains and
generalized forces. Solve the combination of the equilibrium equations obtained in
Section 9.8 and the resulting compatibility equation for all the unknown generalized
forces in this problem using MATLAB®.
(b) Use the strain–displacement relations to try to solve for the displacements in this
problem. Do you obtain a correct solution? Give a physical interpretation of those
displacements that are nonzero.
P9.3 Consider the statically indeterminate beam of Fig. P9.1. Write the equations of
equilibrium for the entire beam and solve for all the reactions by
(a) using the principle of least work,
 9.11 Problems 367

(b) integrating the moment–curvature expression M ðxÞ ¼ EId 2 w=dx2 , where w is

the vertical displacement, and applying the slope and displacement boundary
conditions,
(c) using the force-based approach of Eq. (9.219).
Applying the principle of least work to solve statically indeterminate problems is
often found in intermediate and advanced strength of materials texts, while the use of
the integration method is normally discussed in elementary texts. The force-based
approach of Eq. (9.219) is an alternative that is not commonly covered.
The problems shown in Fig. P9.2 and Fig. P9.3 are axial-load and beam problems considered in the
problems at the end of Chapter 1. There, you were asked to obtain the exact analytical solutions.
Here, we want to develop MATLAB® finite element solutions of various types. These are relatively
lengthy problems that are probably best treated as class projects. For each case you should write a
MATLAB® script that will let you obtain a finite element solution where the number of elements
can be varied (but the number of elements must be an even number to match the boundary
conditions seen in some of the problems). Note that the displacement boundary conditions are the
same in all of these problems (fixed–fixed restraints) except for the problem in Fig. P9.3d so that
you only need to change the force boundary conditions in those problems with common restraints.
Write these scripts, treating these problems as numerical problems, not symbolic algebra problems.

P9.4 The axial-load problem shown in Fig. P9.2a is the fixed-fixed bar with a distributed load
discussed with the stiffness-based method in Example 9.2 and with the force-based
method in Example 9.5. Write a MATLAB® script that solves this
z problem via a stiffness-based finite element solution for any number
of elements. (Note, however, you should use an even number of
P
elements, as this is required by the loading in other cases we will
A B C
consider.) Plot the nodal displacements in the bar. Plot the internal
x forces (which are constants), as discrete values at the centroids of

L L Figure P9.1 A statically indeterminate beam problem.

x Figure P9.2 Axial-load problems whose

qx = q0 x
P solutions are to be obtained by various
finite element methods.

L L/2 L/2

(a) (b)
qx = q0 x
x x qx = q0
L

L/2 L/2 L

(c) (d)
368 Computational Mechanics of Deformable Bodies

Figure P9.3 Beam problems whose z z

solutions are to be obtained by various
P q0
finite element methods.

x x
L/2 L/2 L
(a) z (b)
z
q0 P

x x

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

(c) (d)

the elements. Obtain the solution and plots for 20 elements. Compare these plots to
the exact solution obtained in Chapter 1 for this problem. Let L ¼ 3000 mm, A ¼ 9 mm2,
E ¼ 10 000 N/mm2 (10 GPa), q0 ¼ 10 N=mm.
P9.5 Repeat problem P9.4, but use a force-based finite element solution instead.
P9.6 Repeat problem P9.4, but use a pseudostiffness finite element solution instead.
P9.7 Consider the axial load problem in Fig. P9.2b. Solve this problem using 20 elements
and a stiffness-based finite element method. Obtain the same types of plots as in
problem P9.4 and compare with the exact solution obtained in Chapter 1 for this
problem. Let L ¼ 3000 mm, A ¼ 9 mm2, E ¼ 10 000 N/mm2 (10 GPa), P ¼ 20 000 N.
P9.8 Repeat problem P9.7, but use a force-based finite element method instead.
P9.9 Repeat problem P9.7, but use a pseudostiffness finite element method instead.
P9.10 Consider the axial load problem in Fig. P9.2c. Solve this problem using 20 elements and a
stiffness-based finite element method. Obtain the same types of plots as in problem P9.4
and compare with the exact solution obtained in Chapter 1 for this problem. Let L ¼
3000 mm, A ¼ 9 mm2, E ¼ 10 000 N/mm2 (10 GPa), q0 ¼ 10 N=mm.
P9.11 Repeat problem P9.10, but use a force-based finite element solution instead.
P9.12 Repeat problem P9.10, but use a pseudostiffness finite element solution instead.
P9.13 Consider the axial-load problem in Fig. P9.2d. Solve this problem using 20 elements and
a stiffness-based finite element method. Obtain the same types of plots as in problem P9.4
and compare with the exact solution obtained in Chapter 1 for this problem. Let L ¼
3000 mm, A ¼ 9 mm2, E ¼ 10 000 N/mm2 (10 GPa), q0 ¼ 10 N=mm.
P9.14 Repeat problem P9.13, but use a force-based finite element solution instead.
P9.15 Repeat problem P9.13, but use a pseudostiffness finite element solution instead.
P9.16 Consider the fixed–fixed beam with a central concentrated load as shown in Fig.
P9.3a. Solve this problem using 20 elements and the stiffness-based finite element
method. Obtain solutions and plots for the internal shear force, bending moment, and
 9.11 Problems 369

displacement, and compare with the exact solution obtained in Chapter 1 for this
problem. Let L ¼ 3000 mm, E ¼ 2  105 N/mm2 (200 GPa), I ¼ 3  106 mm4 , P ¼
20 000 N.
P9.17 Repeat problem P9.16, but use a force-based finite element solution instead.
P9.18 Repeat problem P9.16, but use a pseudostiffness finite element solution instead.
P9.19 Consider the fixed–fixed beam with a uniform distributed load shown in Fig. P9.3b.
Solve this problem using 20 elements and the stiffness-based finite element method.
Obtain solutions and plots for the internal shear force, bending moment, and dis-
placement and compare with the exact solution obtained in Chapter 1 for this
problem. Let L ¼ 3000 mm, E ¼ 2  105 N/mm2 (200 GPa), I ¼ 3  106 mm4 ,
q0 ¼ 10 N=mm.
P9.20 Repeat problem P9.19, but use a force-based finite element solution instead.
P9.21 Repeat problem P9.19, but use a pseudostiffness finite element solution instead.
P9.22 Consider the fixed–fixed beam with a uniform distributed load acting over its right
half as shown in Fig. P9.3c. Solve this problem using 20 elements and the stiffness-
based finite element method. Obtain solutions and plots for the internal shear force,
bending moment, and displacement, and compare with the exact solution obtained in
Chapter 1 for this problem. Let L ¼ 3000 mm, E ¼ 2  105 N/mm2 (200 GPa),
I ¼ 3  106 mm4 , q0 ¼ 10 N=mm.
P9.23 Repeat problem P9.22, but use a force-based finite element solution instead.
P9.24 Repeat problem P9.22, but use a pseudostiffness finite element solution instead.
P9.25 Consider a beam that is clamped at one end and simply supported at the other, and
carries a concentrated load at its center, as shown in Fig. P9.3d. Solve this problem
using 20 elements and the stiffness-based finite element method. Obtain solutions and
plots for the internal shear force, bending moment, and displacement, and compare
with the exact solution obtained in Chapter 1 for this problem. Let L ¼ 3000 mm,
E ¼ 2  105 N/mm2 (200 GPa), I ¼ 3  106 mm4 , P ¼ 20 000 N. Note that the
displacement boundary conditions are different from the other beam cases considered
in Fig. P9.2.
P9.26 Repeat problem P9.25, but use a force-based finite element solution instead.
P9.27 Repeat problem P9.25, but use a pseudostiffness finite element solution instead.
P9.28 Consider the simply supported beam shown in Fig. P9.4 where a beam of length 2L
with three supports carries a uniform distributed force q0 (force/unit length). Solve
this statically indeterminate problem using the force-based method of Eq. (9.219)
where we combine equilibrium and compatibility and use the complementary strain
energy to obtain the compatibility equation. Take a single
z
q0 “element” (i.e., the entire beam) and obtain the unknown
reactions in terms of q0 and L.
x

L L Figure P9.4 A simply supported beam with three supports.

370 Computational Mechanics of Deformable Bodies

P9.29 Consider the statically indeterminate beam problem of Fig. 9.25c where a beam is
fixed at one end and simply supported by a roller at the other end and subjected to a
uniform distributed force, q0 , over its entire length, L. In this case, assume that the
simply supported end is given a known z-displacement, wR . Show explicitly the
following steps.
(a) Using a single element (i.e., the entire beam), write down the finite element
equilibrium equations for the internal moments ðM 1 , M 2 Þin terms of the reactions
and the equivalent nodal forces and moments coming from the distributed load.
(b) Eliminate those equations that involve the unknown reactions and write the
remaining equilibrium equation in terms of a free equilibrium matrix, E ~f .
From the equilibrium equations of part (a), identify the E ~ R matrix associated
with the known displacement and obtain (1) the free equilibrium equation, (2)
the strain–displacement relation (in terms of E ~ f T and E ~ R T ) that relates the
generalized strains ðΔ1 , Δ2 Þ associated with ðM 1 , M 2 Þ to the free generalized
displacement, and (3) the relationship between the generalized strains and the
generalized internal forces (moments). These are the three sets of equations seen
in Eq. (9.229).
(c) Obtain the compatibility matrix associated with the free equilibrium matrix and
use it to write the compatibility equation in terms of the internal generalized
forces (see Eq. (9.230)). Solve the combination of the free equilibrium equation
and the compatibility equation for the internal generalized forces and obtain the
value of the reaction force at the roller in terms of q0 , L, EI, and wR .
(d) Finally, determine the unknown slope at the roller in terms of q0 , L, EI, and wR
using Eq. (9.237). You might want to use the symbolic algebra capabilities of
MATLAB® for this part. Can you give a simple explanation for the terms
appearing in the expressions for the reaction force and slope at the roller due to
the known displacement?
P9.30 In this chapter we showed how known support displacements can be incorporated
into the force-based finite element method. Incorporation of known displacements
with the stiffness-based method is much easier since we are already working with the
displacements. Consider for example, the very simple problem of Fig. 9.28 where a
bar of length 2L has a concentrated axial load P at its center and the end supports
have displacements ðU  3 Þ. Using two elements:
 1, U
(a) Write down the stiffness-based equations of equilibrium for the bar,
(b) Eliminate the equations involving the end reactions from the equilibrium matrix
and move the columns of the matrix associated with the known displacements to
the right side with the known load(s). Solve for the free displacement(s) in terms of
the known load(s) and displacements. This same process will work for much more
general problems.
(c) Determine the internal forces in the two elements. Show all your results agree with
the force-based solution obtained previously.
 References 371

References
1. S. N. Patnaik, D. A. Hopkins, and G. R. Halford, Integrated Force Method Solution to
Indeterminate Structural Mechanics Problems (NASA/TP-2004-207430, Washington, D.C., 2004)
2. I. Kaljević, S. N. Patnaik, and D. A. Hopkins, “Three-dimensional structural analysis by the
Integrated Force Method,” Comp. Struct., 58 (1996), 869–886
3. S. N. Patnaik and D. A. Hopkins, Strength of Materials: A Unified Theory (Oxford, UK: Elsevier,
2004)
4. I. Kaljević, S. N. Patnaik, and D. A. Hopkins, Development of Finite Elements for Two-Dimensional
Structural Analysis using the Integrated Force Method (NASA Technical Memorandum 4655, 1996)
5. I. Kaljević, S. N. Patnaik, and D. A. Hopkins, Element Library for Three-Dimensional Stress
Analysis by the Integrated Force Method (NASA Technical Memorandum 4686, 1996)
6. S. N. Patnaik, L. Berke, and R. H. Gallagher, Integrated Force Method Versus Displacement
Method for Finite Element Analysis (NASA Technical Paper 2937, 1990)
7. C. A. Brebbia and J. Dominguez, Boundary Elements: An Introductory Course, 2nd edn (New
York, NY: McGraw-Hill, 1992)
8. J. T. Katsikadelis, The Boundary Element Method for Scientists and Engineers (Boston, MA:
Academic Press, 2016)
9. M. A. Jaswon and G. T. Symm, Integral Equation Methods in Potential Theory and Elastostatics
(Boston, MA: Academic Press, 1977)
10 Unsymmetrical Beam Bending

The engineering beam-bending theory summarized in Chapter 1 assumed that the beam
cross-sectional area has a plane of symmetry and that bending moments were acting along a
single axis (taken to be the y-axis). In this chapter we want to remove those restrictions and
to examine the multiaxis bending of beams with nonsymmetrical cross-sections. This will
lead to generalizations of the flexure formula (see Eq. (1.20)). In Chapter 1 we also obtained
an expression for the shear stresses induced in symmetrical beams. It is difficult to obtain
similar analytical shear stress forms for beams with unsymmetrical cross-sections except in
the case when the cross-section is thin, a case we will also consider in this chapter.

10.1 Multiple Axis Bending of Nonsymmetrical Beams

Figure 10.1 shows a cantilever beam with a triangular-area cross-section, a typical example
we will use to discuss the extensions needed in elementary beam theory to handle the bending
of such unsymmetrical cross-sectional beams. Also shown in Fig. 10.1 are the positive



directions assumed for the internal bending moments M y , M z , shear forces V y , V z ,
and torsional moment, M x , acting in the beam. These directions as well as the coordinate
system shown will be used consistently throughout this chapter and in Chapters 11, 12 when
we discuss torsion and combined bending and torsion problems. With these assumed
directions, when we look down the negative y-axis at a small element of the beam, we see
the free body diagram of Fig. 10.2a and obtain from equilibrium
X  
dM y
M o ¼ M y þ V z dx  M y þ dx ¼ 0
dx (10.1)
dM y
! ¼ Vz
dx
(Any distributed loads are not shown as they do not affect the moment shear relationships.)
Similarly, looking down the z-axis (Fig. 10.2b) we find
X  
dM z
M o ¼ M z þ V y dx  M z þ dx ¼ 0
dx (10.2)
dM z
! ¼ V y
dx
Equation (10.1) is the same relationship we obtained earlier (see Eq. (1.23)) while Eq. (10.2)
is new because in Chapter 1 the internal bending moment was assumed to act only about the
y-axis.

372
 10.1 Multiple Axis Bending of Nonsymmetrical Beams 373

Figure 10.1 A cantilever beam with an unsymmetrical

z cross-section and the positive directions assumed for the
internal moments and shear forces in the beam.
Mz
y

My

Mx
Vy
Vz x

dVz Figure 10.2 (a) The internal moments and shear forces seen
Vz Vz + dx
dx looking down the negative y-axis. (b) The internal moments and
shear forces seen when looking down the z-axis.
My dM y
O My + dx
dx

The main difference between bending of an unsymmet-

dx rical beam and a symmetrical beam is that, even if the
(a) internal moment acts only about the plus or minus y-axis,
dVy as assumed in Chapter 1, bending displacements will take
Vy Vy + dx
dx place about both the y- and z-axes. Multiaxis bending will
dM z also occur when we have internal moments acting about
Mz Mz + dx
O both the y- and z-axes, as assumed here. Consider, as we
dx
did in Chapter 1, the pure bending case but where now


there are both internal moments M y , M z acting. If we
dx assume, as we did in Chapter 1, that cross-sections that
(b) were originally planar cross-sections normal to the beam
axis in the undeformed beam are still planar and normal to
the neutral axis in the deformed beam, then a cross-section plane AB has the behavior seen
in Figs. 10.3a, b, where the plane experiences small rotations about both the negative y-axis
(Fig. 10.3a) and the z-axis (Fig. 10.3b), and these rotations are just equal to the slopes of the
neutral axis, dw=dx and dv=dx, respectively, where w is the z-displacement of the neutral axis
and v is the y-displacement of the neutral axis. Thus, the beam will experience an
x-displacement, ux , that is given by

dw dv
ux ¼ z y (10.3)
dx dx

This, in turn, will generate an axial normal strain, exx , given by

∂ux d 2w d 2v
exx ¼ ¼ z 2  y 2 (10.4)
∂x dx dx
374 Unsymmetrical Beam Bending

Figure 10.3 The behavior of a cross- z y

section AB (originally normal to the
x-axis in the undeformed beam) during ux ux
bending where it assumed the cross-
B B
dw z dv y
section remains planar and normal to
the neutral axis. (a) The plane AB seen dx dx
in the x–z plane. (b) The same plane AB x x
as seen in the y–z plane.

A A

(a) (b)

where d 2 w=dx2 and d 2 v=dx2 are curvatures of the beam in the x–z and y–z planes, respect-
ively. If this strain produces only the normal flexure stress, σ xx , we have
!
d 2w d 2v
σ xx ¼ Eexx ¼ E z 2  y 2 (10.5)
dx dx

where E is Young’s modulus. Note that, as in the symmetrical cross-section case, there are


also normal strains eyy , ezz generated:
!
d 2w d 2v
eyy ¼ ezz ¼ νexx ¼ ν z 2 þ y (10.6)
dx dx

where ν is Poisson’s ratio.

We can relate the flexure stress to the internal moments via the relations (see Fig. 10.4):
ð
σ xx dA ¼ 0
ð
yσ xx dA ¼ M z (10.7)
ð
zσ xx dA ¼ M y

Placing the normal stress expression of Eq. (10.5) into these relations, we find
ð ð
d 2w d 2v
E zdA  E ydA ¼ 0
dx2 dx2
ð ð
d 2w d 2v
E 2 yzdA þ E 2 y2 dA ¼ M z (10.8)
dx dx
2 ð ð
d w d 2v
E 2 z2 dA  E 2 yzdA ¼ M y
dx dx
 10.1 Multiple Axis Bending of Nonsymmetrical Beams 375

z Figure 10.4 Cross-section of an unsymmetrical beam, showing the normal stress

and internal bending moments acting in the cross-section.
Mz

The first equation in Eq. (10.8) gives

y V xx
ð ð
z My
zdA ¼ ydA ¼ 0 (10.9)
y

i.e., the neutral axis is located at the centroid of the cross-section, as it is

in the symmetrical cross-section case. The second and third equations in
Eq. (10.8) give
x

d 2w d 2v
EI yz þ EI zz 2 ¼ M z
dx2 dx
(10.10)
d 2w d 2v
EI yy 2  EI yz 2 ¼ M y
dx dx

where the second area moments and mixed second area moment of the cross-section are
ð
I yy ¼ z2 dA
ð
I zz ¼ y2 dA (10.11)
ð
I yz ¼ yzdA

We can solve the two equations in Eq. (10.10) for the curvatures, obtaining


d 2v M z I yy þ M y I yz
¼  
dx2 E I yy I zz  I 2yz

 (10.12)
d 2 w  M y I zz þ M z I yz
¼  
dx2 E I yy I zz  I 2yz

If we place these curvatures into Eq. (10.5), we find the flexure stress is given by



M y I zz þ M z I yz z  M z I yy þ M y I yz y
σ xx ¼   (10.13)
I yy I zz  I 2yz

Although we used the case of pure bending to derive Eq. (10.13), as was done in Chapter 1,
we can simply let M y ¼ M y ðxÞ and M z ¼ M z ðxÞ to obtain the equivalent engineering beam
theory for the more practical case when bending is due to various applied loads along
376 Unsymmetrical Beam Bending

the beam. This, in turn, will require the existence of shear stresses as well, which we will
discuss shortly. If either the y- or z-axis is an axis of symmetry then I yz ¼ 0 and we have
M yz M zy
σ xx ¼  (10.14)
I yy I zz
which is just a superposition of the simple flexural stress formula of Chapter 1 for the two
internal moments for combined axis bending, and where the difference in sign simply arises
because of the assumed directions of the internal moments. Even if we only have a bending
moment about the y-axis, as assumed in Chapter 1, for an unsymmetrical cross-section Eq.
(10.13) gives


M y I zz z  M y I yz y
σ xx ¼   (10.15)
I yy I zz  I 2yz

which shows that we have bending occurring about both axes, producing a flexure stress that
varies linearly along both axes in the cross-section if the cross-section is unsymmetrical.
For multiaxis moments and an unsymmetrical cross-section, Eq. (10.13) shows that the
neutral surface in the cross-section (where σ xx ¼ 0) is along the line z ¼ tan λ y (see Fig. 10.5)
where
M z I yy þ M y I yz
tan λ ¼ (10.16)
M y I zz þ M z I yz

which can also be written as (assuming M y ¼ 6 0, which is usually the case)


M z =M y I yy þ I yz
tan λ ¼
 (10.17)
I zz þ M z =M y I yz

Generally, both M z and M y are functions of the x-distance in a beam so that the angle of the
neutral surface will also be a function of x. However, if the ratio M z ðxÞ=M y ðxÞ is a constant,
then Eq. (10.17) shows that the angle of the neutral surface will be a constant in the beam.
The case where this ratio is a constant is called the case of a single plane of loading.
Figure 10.6 shows an example of both a single plane of loading case, where all the applied
loads lie in a common plane, and a case of general loading. In the latter case the angle
neutral surface will vary in the beam.
Equation (10.13) is a general expression for the flexure
stress in an unsymmetrical beam under multiaxis bending.
z
There are, however, other forms of this expression that can
be obtained. For example, one can use Eq. (10.17) in
Eq. (10.13) to eliminate the explicit dependency on M z
neutral surface in Eq. (10.13) and obtain
O
y
Figure 10.5 The line of the neutral surface in the cross-section of
an unsymmetrical beam.
 10.1 Multiple Axis Bending of Nonsymmetrical Beams 377

Figure 10.6 (a) Single plane of loading for a beam.

(b) A general loading.

q M y ðz  tan λ yÞ
T σ xx ¼ (10.18)
I yy  I yz tan λ

P although M z does not appear explicitly in Eq. (10.18) it

(a) T
is still present implicitly through the tan λ terms.
Another choice is to write Eq. (10.13) in terms of a
set of principal axes ðY ; Z Þ as shown in Fig. 10.7 where

Y ¼ z sin θp þ y cos θp
(10.19)
q Z ¼ z cos θp  y sin θp
T
P
p Since principal axes act like axes of symmetry for a
2
general cross-section where I YZ ¼ 0 (see Appendix A),
(b) we obtain the form of Eq. (10.14) where now

MY Z MZY
σ xx ¼  (10.20)
I YY I ZZ

In this form the moments must be resolved along the principal directions and the second area
moments of the cross-section with respect to the principal axes must be calculated, as seen in
Appendix A. A fourth and final choice is to use a set of coordinates ðη; ζ Þ, which are along
and perpendicular to the neutral surface as shown in Fig. 10.8. To obtain the flexure stress in
these coordinates note that we can rewrite Eq. (10.13) as

M y I zz þ M z I yz
σ xx ¼ ðz  tan λ yÞ (10.21)
I yy I zz  I 2yz

and we have

z cos λ  y sin λ ζ
z  tan λy ¼ ¼ (10.22)
z cos λ cos λ

so that we obtain
Z Y M y I zz þ M z I yz ζ
σ xx ¼ (10.23)
I yy I zz  I 2yz cos λ

Tp
y
Figure 10.7 A set of principal axes ðY ; Z Þ for an unsymmetrical cross-
section.
378 Unsymmetrical Beam Bending

z Figure 10.8 A set of ðη; ζ Þ coordinates that lie along and perpendicular
ζ
to the neutral surface.

η
All of these different forms can be used and some texts
emphasize one form over another. We will simply note that
O Eq. (10.20) and Eq. (10.23) require extra calculations to use
y their particular coordinate systems, while Eq. (10.18) can be
used directly if one has already calculated the angle, λ, of the
neutral surface, as is often the case.
In addition to obtaining the flexure stress, we often want to
determine the displacement of the neutral axis. As in the
symmetrical cross-section case this means that we must integrate twice the moment–
curvature relations, which now are given by Eq. (10.12), and apply the boundary condi-
tions. In the case of a single plane of loading this can be made easier since from Eq. (10.12)
we have


d 2 v=dx2  M z I yy þ M y I yz
¼
 ¼  tan λ
d 2 w=dx2 M y I zz þ M z I yz
(10.24)
d 2v d 2w
! 2 ¼  tan λ 2
dx dx

and, since the angle λ is a constant, if we let d 2 w=dx2 ¼ F ðxÞ we find, on integration,

ðð
w¼ F ðxÞdxdx þ C 1 x þ C 2
ðð (10.25)
v
¼ F ðxÞdxdx þ C 3 x þ C 4
tan λ

If the boundary conditions are the same for both v and w we have C 1 ¼ C 3 and C 2 ¼ C 4 ,
and from Eq. (10.25)

v
¼ tan λ (10.26)
w

Geometrically, this condition means that the total displacement of the neutral axis,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δ ¼ v2 þ w2 , is perpendicular to the neutral surface as shown in Fig. 10.9.
 10.1 Multiple Axis Bending of Nonsymmetrical Beams 379

z Figure 10.9 The case where the total displacement of the

–v neutral axis is perpendicular to the neutral surface.
G = total
displacement

w O
y
O

Example 10.1 Multiaxis Bending

As a first example of bending under these generalized conditions, consider the cantilever
beam shown in Fig. 10.10a. This is a symmetrical beam under multiaxial bending.
The load P is applied to a flange at the bottom of the beam and at a location called the
shear center of the beam, which will be defined in a later section. When we discuss shear
stresses, we will see that if the load is applied at other locations the beam will twist as well as
bend. In this case the load is placed so that the beam experiences bending only, as we have
assumed in our present discussions. In calculating the properties of the cross-section,
however, this small bottom flange will be neglected (see Fig. 10.10b). In this example we
will take P ¼ 12 kN, E ¼ 72 GPa. From the properties of the cross-section we can determine
the area moments are I zz ¼ 30:73  106 mm4 , I yy ¼ 39:69  106 mm4 , I yz ¼ 0 and the
distance to the (y, z) centroid axes (Fig. 10.10b) is d ¼ 82 mm (see the MATLAB® function
cross_section discussed in Appendix A). In this example we want to find (1) the maximum
tensile and compressive stresses and their locations in the cross-section and (2) the total
deflection of the beam at the load P.
The maximum bending moments occur at the wall where we have

M y ¼ 3000P sin 60∘ ¼ 31:17  106 N-mm,

M z ¼ 3000P cos 60∘ ¼ 18  106 N-mm,
M z =M y ¼  cot 60∘ ¼ 0:577

Since I yz ¼ 0, the expression for the angle the neutral surface makes with respect to the
y-axis (see Eq. 10.17) is

  
M z =M y I yy 39:69  106
tan λ ¼ ¼ ð0:577Þ ¼ 0:746 (10.27)
I zz 30:73  106
which gives λ ¼ 0:640 rad ¼ 36:7∘ (see Fig. 10.11b). To compute the flexure stress, we
will use Eq. (10.18) with again I yz ¼ 0, giving
380 Unsymmetrical Beam Bending

Figure 10.10 (a) A cantilever beam 140 mm

under multiaxis bending. (b) The
cross-section of the beam (the
cross-sectional properties of the z
small flange at the bottom where
the load is applied are neglected). 200 mm
3m y
d

60° P
20 mm all around
(a) (b)

Figure 10.11 (a) The components of the applied A

load. (b) The orientation of the neutral surface and z z
the points of extreme flexural stress, A and B. y y

36.7o
B
P cos 60˚
P sin 60˚

(a) (b)

My
σ xx ¼ ðz  tan λyÞ
I yy

31:17  106 (10.28)

¼ ðz þ 0:746yÞ
39:69  106
¼ 0:785z  0:585y

where σ xx is measured in N/mm2 or MPa. The extreme flexure stresses in the cross-section at
the wall will occur at points A and B (see Fig. 10.11b) that are the farthest points as possible
from the neutral axis. At point A z ¼ 200 – 82 ¼ 118 mm and y ¼ 70 mm, and at point B we
have z ¼ –82 mm, y ¼ –70 mm, so that the stresses at these points are

ðσ xx ÞA ¼ ð0:785Þð118Þ  ð0:585Þð70Þ
¼ 133:6MPa
ðσ xx ÞB ¼ ð0:785Þð82Þ  ð0:585Þð70Þ
¼ 105:3 MPa
 10.1 Multiple Axis Bending of Nonsymmetrical Beams 381

FL3 Figure 10.12 A cantilever beam under an applied end load, showing the
w=
3EI expression for the vertical deflection, w, at the load.

L
F Alternatively, since the (y, z) axes are principal axes the flexure
stress is given by Eq. (10.14), giving

M yz M zy 31:17  106 18  106

σ xx ¼  ¼ z y
I yy I zz 39:69  10 6
30:73  106 (10.29)
¼ 0:785z  0:585y

which is the same expression as obtained in Eq. (10.28). To compute the deflection of
the beam at the load we note that this is a single plane of loading and the beam is
rigidly fixed at the wall in both the y- and z-directions, so that we only need to compute
the z-deflection, w, and then we can use Eq. (10.26) to find the y-deflection, v. To get
the z-deflection, we could integrate the moment–curvature relationship of Eq. (10.12),
which in this case, since I yz ¼ 0, gives d 2 w=dx2 ¼ M y =EI yy but since this a cantilever
beam with an end load, most elementary strength of materials books give the end
z-deflection explicitly for the geometry seen in Fig. 10.12. Applying that result to our
problem we have

ðP sin 60∘ ÞL3 ð12; 000 sin 60∘ Þð3000Þ3

w¼ ¼

 ¼ 32:72 mm
3EI yy 3 72  103 39:69  106

and, consequently, the y-deflection, v, and the total deflection, δ, are

v ¼  tan λ w ¼ 0:745 w ¼ 24:38 mm

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δ ¼ w2 þ v2 ¼ 40:8 mm

This first example was a case of a single plane of loading, so we now consider a case where
the loading is more arbitrary.

Example 10.2 Multiaxis Bending with Loading in Multiple Planes

Let us use the same beam considered in Example 10.1 but change the loading to a
combination of a uniform distributed load, q ¼ 1000 N/m, and a concentrated end load,
P ¼ 500 N, as seen in Fig. 10.13. In this example we want to determine the angle the neutral
surface makes with respect to the y-axis as a function of the distance, x, from the wall.
382 Unsymmetrical Beam Bending

Figure 10.13 A cantilever beam subjected to z

loading in multiple planes.

q =1000 N/m

1.5 m x

1.5 m P = 500 N

1125 N-m 500 N Figure 10.14 The cantilever beam of Fig. 10.13, showing the
1500 N reactions at the wall.
1500 N-m
Figure 10.14 shows a free body diagram of the entire
beam and the reactions at the wall calculated from force
and moment equilibrium. Also shown in Fig. 10.14 is
0.75 m the resultant force of 1500 N generated by the distrib-
1500 N uted load and the location of this resultant. It is easy to
3.0 m see that the end reactions are in equilibrium with the
applied loads. Figures 10.15a, b show free body dia-
500 N grams of portions of the beam from the wall to a cutting
plane at a distance x taken for (a) 0 < x < 1.5 m, and (b)
1.5 < x < 3.0 m. Taking moments at the cutting plane
about a point O at the centroid of the cross-section, we
find for 0 < x < 1.5 m:
X
M zO ¼ 0

M z þ 1500x  1000x2 =2  1125 ¼ 0

! M z ¼ 500x2  1500x þ 1125 N-m
(10.30)
X
M yO ¼ 0
M y  500x þ 1500 ¼ 0
! M y ¼ 1500 þ 500x N-m

Similarly, for 1.5 < x < 3.0 m:

 10.1 Multiple Axis Bending of Nonsymmetrical Beams 383

0 < x < 1.5 m 1.5 < x < 3 mt

1125 N-m 500 N 1500 N 1125 N-m

500 N 1500 N

1500 N-m
1500 N-m
Mz
My Mz
1000 N/m
O My

0.75 m O
x 1500 N

(a) (b)

Figure 10.15 Free body diagrams: (a) for 0 < x < 1.5 m, (b) for 1.5 < x < 3.0 m.

X
M zO ¼ 0
M z þ 1500x  1500ðx  0:75Þ  1125 ¼ 0
! Mz ¼ 0
(10.31)
X
M yO ¼ 0
M y þ 1500  500x ¼ 0
! M y ¼ 500x  1500 N-m

For the angle that the neutral surface makes, we can again use Eq. (10.27) which gives in this
case
     
I yy M z 39:69  106 Mz
tan λ ¼ ¼ ¼ 1:29 (10.32)
I zz M y 30:73  10 6 M y

For 0 < x < 1.5 m we have

 
500x2  1500x þ 1125
tan λ ¼ 1:29 (10.33)
500x  1500
384 Unsymmetrical Beam Bending

Figure 10.16 The angle of the neutral surface with respect to 0

the y-axis for the beam of Fig. 10.13. –5
–10
–15

angle, degrees
–20
–25
–30
–35
–40
–45
0 0.5 1 1.5 2 2.5 3
x-distance

and for 1.5 < x < 3.0 m

 
0
tan λ ¼ 1:29 ¼0
500x  1500 (10.34)
!λ¼0

The result of Eq. (10.34) is as expected since in that portion of the beam only the bending
moment M y exists, as is the case in elementary bending, as discussed in Chapter 1. If we
solve for the angle λ from Eq.(10.33) and plot the results over the entire beam we see
(Fig. 10.16) that the neutral axis makes an angle with respect to the y-axis of nearly –45
at the wall and gradually becomes less negative, reaching 0 at the end of the distributed
load.

10.2 Shear Stresses in Thin, Open Cross-Section Beams

In engineering beam theory for symmetrical cross-sections, it was shown that when the
flexural stress varies along the x-axis of the beam, shear stresses must also be present. By
using the flexure stress expression and equilibrium of an element of the beam, an expression
was obtained for those shear stresses. For multiaxis bending of unsymmetrical beams, one
cannot obtain a similar analytical result except in the case when the beam has a thin cross-
section, such as the beam shown in Fig. 10.17a. Commonly used beams such as I-beams and
channel sections are composed of thin sections, so analytical results for such thin cross-
sections are of significant practical interest. In this section, in addition to assuming the beam
is thin, we will assume the beam cross-section is also an open cross-section, i.e., there are no
closed loops formed in the cross-section. The beam shown in Fig. 10.17a is such a thin, open
cross-section. To analyze the shear stresses, we cut a small element from the beam, as shown
in Fig. 10.17a. This element is also shown in Fig. 10.17b. Since the beam is assumed to be
 10.2 Shear Stresses in Thin, Open Cross-Section Beams 385

s xx q=τt

z
q + dq
s
ds ds ds xx
Mz dx s xx + dx
dx
y dA
dx
t(s) My

Vy
Vz
x
(a) (b)

Figure 10.17 (a) An unsymmetrical beam with a thin, open cross-section. (b) A small element of the beam
showing the flexure stresses and shear flows acting on it.

Figure 10.18 An element of a thin beam showing the shear flows on a section.
s2 s1

thin, we expect the shear stresses will be nearly uniform across the
q ( s1 ) thickness of a section and parallel to the adjacent free surfaces. Thus,
in thin beam problems, it is common to multiply this uniform stress by
the local thickness and examine instead the shear flow, q, in the beam
Aq
defined as qðsÞ ¼ τ ðsÞtðsÞ, where s is the distance measured along the
q ( s2 ) cross-section from a free end of the section, as shown in Fig. 10.17a.
The changing flexure stress and the shear flows on a small element of
the beam are shown in Fig. 10.17b. If we sum forces in the x-direction for this element
we find
X
Fx ¼ 0
 
dσ xx
ðq þ dqÞdx  qdx þ σ xx þ dx dA  σ xx dA ¼ 0 (10.35)
dx
dσ xx
! dq ¼  dA
dx
If we then integrate Eq. (10.35) between two locations s1 and s2 , as shown in Fig. 10.18, then
we obtain
ð
dσ xx
Δq ¼ qðs2 Þ  qðs1 Þ ¼  dA (10.36)
dx
Aq

where Aq is the area between s1 and s2 , as shown in Fig. 10.18. For an unsymmetrical
section, the normal stress was given by Eq. (10.13) as
386 Unsymmetrical Beam Bending



M y I zz þ M z I yz z  M z I yy þ M y I yz y
σ xx ¼   (10.37)
I yy I zz  I 2yz

so that if we differentiate this stress on x and use the moment–shear relations of Eq. (10.1)
and Eq. (10.2), we obtain



dσ xx V z I zz  V y I yz z  V z I yz  V y I yy y
¼   (10.38)
dx I yy I zz  I 2
yz

Placing Eq. (10.38) into the shear flow expression of Eq. (10.36) gives



V z I yz  V y I yy Qz þ V y I yz  V z I zz Qy
Δq ¼ (10.39)
D
where
D ¼ I yy I zz  I 2yz
ð
Qz ¼ ydA
Aq (10.40)
ð
Qy ¼ zdA
Aq

Note that in obtaining our expression for Δq in Eq. (10.39), we showed the shear flow q
flowing out from the end of the section A under consideration and the shear flow flowing
into A at the beginning of the section (see Fig. 10.18). As shown in Fig. 10.19 for an I-beam,
this same convention is followed regardless of the section taken. For all the sections in
Fig. 10.19, therefore we have



V z I yz  V y I yy Qz þ V y I yz  V z I zz Qy
Δq ¼ qðsÞ  qð0Þ ¼ (10.41)
D

Figure 10.19 An I-beam where various

s q (s) q (s) s
sections are taken, showing the
directions of the shear flows. q ( 0) = 0 q ( 0) = 0
q ( 0) ≠ 0
s
q (s)

s s
q ( 0) = 0 q ( 0) = 0
q (s) q (s)
 10.2 Shear Stresses in Thin, Open Cross-Section Beams 387

In a given problem, the shear flow into, or the shear flow out of, a section can be positive or
negative, so by associating the positive signs with a figure such as Fig. 10.19 it is easy to see
the actual direction in the cross-section
If we have a symmetrical cross-section and if V y ¼ 0, then we are considering the
elementary bending theory case considered in Chapter 1. Furthermore, if we take qðs1 Þ ¼ 0
so we are starting at a free end of the open cross-section then Δq ¼ τ ðs2 Þtðs2 Þ and Eq. (10.39)
becomes simply
V z Qy
τ¼ (10.42)
I yy t

which is of very similar form to Eq. (1.25) for the average vertical shear stress in a
symmetrical beam. A minus sign is present in Eq. (10.42) but not in Eq. (1.25) because of
the difference between the direction of the shear stress in Eq. (1.25) and direction of the shear
stress (shear flow) in Eq. (10.42). These differences are shown explicitly in Fig. 10.20 for a
rectangular section.
Another fact you should be aware of is that the shear flows must be conserved at a
junction between sections such as the junction between the web and flange of an I-beam, for
example. This follows directly from equilibrium considerations in Fig. 10.21, which shows
the shear flows ðq1 , q2 , q3 Þ at a T-junction. These shear flows must be in equilibrium with the
changing flexure stress so that, by summing forces in the x-direction, we obtain from that
figure:
X
Fx ¼ 0
 
dσ xx
ðq1 þ q2 þ q3 Þdx þ σ xx þ dx dA  σ xx dA ¼ 0 (10.43)
dx
dσ xx
! ðq1 þ q2 þ q3 Þ þ dA ¼ 0
dx
Thus, as we shrink dA ! 0 from Eq. (10.43) we find
q1 þ q2 þ q3 ¼ 0 (10.44)
z z
i.e., the net shear flow out of (or into) a junction must be zero.
[Note: this result was shown for a T-junction but the same
result is obtained at a general junction.]
Equation (10.39) can be used to obtain the shear-stress distri-
q =τ t
s xz bution in a thin, unsymmetrical beam under multiaxis loading,
and we will show an example of such explicit calculations
y y

t
Figure 10.20 (a) The average shear stress, σ xz , in a rectangular section
appearing in Eq. (1.25). (b) The shear flow, q, and corresponding
(a) (b) stress, τ, appearing in Eq. (10.42).
388 Unsymmetrical Beam Bending

Figure 10.21 Shear flows at a junction where dx

equilibrium shows that the net shear flow into or out
of the junction must be zero. s xx
q3

q2 ds xx
s xx + dx
dx
q1 dA

Figure 10.22 (a) A beam loaded so z z

as to produce bending only (no
twisting). (b) Only bending is
produced when a vertical load acts
through the z-axis of symmetry y y
(also a centroidal axis). (c)
Bending and twisting occur when
the vertical load is not located
along the symmetry axis. (d) If the
(a) (b) (c)
load acts horizontally through the z z
y-centroidal axis, both bending P
and twisting occur. (e) If the
horizontal load acts at a point
below the beam on the flange
y y
called the shear center, only
bending will occur.

(d) (e)

shortly. However, before we apply that equation, we will discuss in the next section the
important concept of the shear center.

10.3 The Shear Center for Thin, Open Cross-Section Beams

Figure 10.22a shows the thin, open cross-section cantilever beam we previously used to
discuss unsymmetrical bending (see Fig. 10.12). There, we placed the load P on a flange
beneath the cross-section so that we could assume the beam was only in multiaxial
bending. Let us examine that assertion more closely. This beam does have an axis of
symmetry (the z-axis) so that if we placed a vertical load anywhere along that axis (which is
also a centroidal axis), we will produce bending only, as shown in Fig. 10.22b. However, if
the vertical load acted to the right or left of the z-axis, the load will produce twisting
 10.3 The Shear Center for Thin, Open Cross-Section Beams 389

(torsion) as well as bending, as shown in Fig. 10.22c. When we discuss torsion in

Chapter 11 we will see that, in thin sections, there are also shear stresses due to torsion
as well as bending so that shear stresses obtained from our shear flow expression,
Eq. (10.39), which assumed bending deformations only are present, is not a correct
expression for the total shear stresses when both bending and torsion are present. If we
apply a horizontal load, as shown in Fig. 10.22d, through the centroidal y-axis (which is
not an axis of symmetry), we will find that both bending and twisting are produced. There
must be some vertical location, however, where the horizontal load can be applied and not
produce twisting because if the load is applied at a very large positive z-coordinate, it will
obviously produce a large counterclockwise twist, similar to that seen in Fig. 10.22d, while
if the load acts at very large negative z-values, we would expect to see a twist in the
opposite sense. There is some point, therefore, (which in this case turns out to be below the
cross-section and along the z-axis) where bending only is produced (see Fig. 10.22e) about
both the y- and z-directions. This point is called the shear center and it is the point we
placed the load (without specifying its location specifically) when solving this problem
previously. Thus, to use our bending shear-flow expression, Eq. (10.39), we must locate the
shear center and ensure that the lines of action of the shear force components act through
that shear center. There are some special cross-sections where we can locate the shear
center by inspection. The shear center, S, must lie on an axis of symmetry of a cross-
section, as seen in Fig. 10.23a, since the symmetrical shear flows produced by a shear force
acting along that axis will not produce any net twisting moment. We must still calculate,
however, where the shear center lies along that symmetry axis. The shear center must lie at
the corner of an L-section, as shown in Fig. 10.23b, since the shear flows all have directions
that act through that point. For the same reason the shear centers must be located at the
points shown in Figs. 10.23c, d for those cross-sections.

S S
S

(a) (b) (c) (d)

Figure 10.23 Cases where the location of the shear center, S, can be partially or totally specified by
inspection. (a) If the cross-section has an axis of symmetry, the shear center lies along that axis since any
shear force whose line of action is along that symmetry axis will produce symmetrical shear flows, as
shown, that have no tendency to twist the section. (b) The shear center must lie at the corner of an L-
section since all the shear flows must have directions which act through that point. The same reasoning
applies to the cross-sections seen in (c) and (d).
390 Unsymmetrical Beam Bending

s Figure 10.24 (a) A thin semicircular cross-section of radius R

t and thickness t, where the shear center is at a distance, e, from
the center, O. (b) Geometry for calculating the shear flow,
R where z is measured from the centroidal y-axis of symmetry.
S q ds T z
R
O O y
There are basically two ways to calculate the loca-
e
tion of the shear center. The first way is based on the
Vz fact that we can calculate the moments produced by
the shear flows about a reference point in terms of V y
(a) (b)
and V z individually by using the shear-flow expres-
sion in Eq. (10.39), which assumes only bending.
Equating moments generated by those shear flows to the moments of V y and V z when
acting at some unknown distances from that reference point then allows us to solve for
those distances (the V y and V z values cancel out) and hence determine the location of the
shear center. For this method to be viable by hand calculations, the integrals of the shear
flow distributions needed to calculate the moments must be known in analytical forms. As
a simple example, consider the semicircular cross-section shown in Fig. 10.24a. The shear
center must lie somewhere along the axis of symmetry. To locate it with respect to the
center O, note that for a vertical shear force, V z , the shear flow expression, Eq. (10.39)
becomes
ð ðθ
Vz Vz
q ¼ zdA ¼  R cos θ tRdθ
I yy I yy
Aq 0 (10.45)
Vz 2
¼ R t sin θ
I yy

For the section we find I yy ¼ πR3 t=2, which is just one fourth the value of the polar area
moment for a thin full circular section. Calculating the moment generated by the shear flow
about point O and setting it equal to the moment of the shear force in Fig. 10.24a gives
ðπ ðπ
M xO ¼ Rqds ¼ R qdθ 2

0 0
ðπ
2V z 4 4RV z (10.46)
¼ R t sin θdθ ¼ 
πR t
3 π
0
¼ V z e
! e ¼ 4R=π

Thus, the shear center is outside the cross-section as shown in Fig. 10.24a. For a general
unsymmetrical section we can do similar calculations individually for the shear forces V z
and V y and locate the shear center in both the y- and z-directions.
 10.3 The Shear Center for Thin, Open Cross-Section Beams 391

Another way to find the location of the shear center is to recognize that the shear center is,
like the centroid, a purely geometric quantity that depends only on the shape of the cross-
section. Thus, if we can determine the geometric parameters that control the location of the
shear center, we can determine the shear center directly from the geometry. As we will see,
one important geometric parameter is called the sectorial area function. Figure 10.25 shows a
thin, open cross-section where the shear forces act through the shear center. If we locate the
shear center relative to a reference point, B, then the moment generated about B by the shear
forces must equal the moment produced by the shear flow under bending only (no torsion).
Equating moments about the x-axis we find
ðE
V z ey  V z ez ¼ r⊥ qds (10.47)
I

where the integral is over the entire area, A, of the cross-section. Since the shear flow is zero
at the starting point, I, from Eq. (10.39) we obtain

 ðE
 ðE
V z I yz  V y I yy V y I yz  V z I zz
V z ey  V y ez ¼ r⊥ Qz ds þ r⊥ Qy ds (10.48)
D D
I I

But r⊥ ds ¼ 2dAs ¼ dωB is twice the area, dAs , swept out

from point B as we go through a distance ds along the
q ds
cross-section, as shown in Fig. 10.26, a quantity which we
I will call the differential of the sectorial area function, dωB ,
z
as measured from point B (do not confuse dAs with
Vy S dA ¼ tds, which is a differential of the cross-sectional
A y
rA
ez Vz
E Figure 10.25 Geometry of an open, unsymmetrical cross-section
B showing the shear center, S, as measured relative to a reference
ey point, B, and the shear flow, which generates the shear forces.

ds Figure 10.26 Definition of an area measured from the point

B swept out by a length ds along the centerline of the cross-
section.

rA
1
dAs = r^ ds
2

B
392 Unsymmetrical Beam Bending

area, A). If we do an integration by parts on the integrals in Eq. (10.48), we find

ðE ðE ðE
Qz dωB ¼ Qz ωB jEI  ωB dQz ¼  ωB dQz
I I I
(10.49)
ðE ðE ðE
E
Qy dωB ¼ Qy ωB I
 ωB dQy ¼  ωB dQy
I I I

since both Qz and Qy vanish at the starting point I and they are also zero at the end point
E because they are the first moments of the total cross-sectional area as measured from
the centroid. However, from their definitions we also have dQz ¼ ydA, dQy ¼ zdA (see
Eq. (10.40)) so
ðE ð
Qz dωB ¼  yωB dA  I yωB
I A
(10.50)
ðE ð
Qy dωB ¼  zωB dA  I zωB
I A

where I yωB and I zωB are called the sectorial products of areas for the sectorial area function,
ωB . [See Appendix C, which gives a detailed discussion of the sectorial area function and its
properties. You should study that appendix to better understand this function. In torsion
problems we will see the sectorial area function also plays an important role.]
From Eq. (10.47), therefore,



I zz I zωB  I yz I yωB I yz I zωB  I yy I yωB
V z ey  V y ez ¼ V z  Vy (10.51)
D D
which must be true for arbitrary V z and V y , giving the location of the shear center relative to
point B finally as


I zz I zωB  I yz I yωB
ey ¼
I yy I zz  I 2yz

 (10.52)
I yz I zωB  I yy I yωB
ez ¼
I yy I zz  I 2yz

The location of the shear center given by Eq. (10.52) is only a function of the geometry of the
cross-section, as shown in Appendix C. The sectorial area function can be calculated with
respect to any convenient point B. The distances ey and ez are then given with respect to that
point. The distances y and z appearing in the first moments of the sectorial area function are
measured from the centroid of the cross-section to a point on the centerline of the cross-
section. To use Eq. (10.52) directly requires that we calculate a large number of parameters.
 10.3 The Shear Center for Thin, Open Cross-Section Beams 393

However, as pointed out in Appendix C, if we construct a sectorial area function about a

point B where
ð
yωB dA ¼ 0
A
ð (10.53)
zωB dA ¼ 0
A

then ey and ez are both zero; i.e., the point B is at the shear center. Thus, one can use
Eq. (10.53) directly to calculate the location of the shear center without having to calculate
all the area moments appearing in Eq. (10.52). Note that we can start the integration from a
point where the sectorial area function has a nonzero starting value, ω0 . However, we can
always set that constant value equal to zero when using Eq. (10.53) since if ω0 ¼ ω þ ω0 we
have
ð ð ð ð
yω0 dA ¼ yωdA þ ω0 ydA ¼ yωdA
A A A A
ð ð ð ð (10.54)
0
zω dA ¼ zωdA þ ω0 zdA ¼ zωdA
A A A A

where we have used the fact that y and z are measured from the centroid of the total
area A of the cross-section to eliminate the terms involving the constant. Thus, these


sectorial products of area I zω , I yω are not affected by a constant sectorial area value.
This is also discussed in Appendix C. We can also remove the restriction that the ðy; zÞ
coordinates in Eq. (10.53) be measured from the centroid by adding the requirement that
the integral of the sectorial area function be zero, i.e., requiring instead that we satisfy
the three conditions:
ð
ωB dA ¼ 0
A
ð
y0 ωB dA ¼ 0 (10.55)
A
ð
z0 ωB dA ¼ 0
A

where now ðy0 , z0 Þ can be measured from any origin in the cross-section we choose. However,
if we start out the sectorial area function at a location where the sectorial area function has a
constant value, that constant, along with the location of the shear center, are all fixed by
Eq. (10.55). A sectorial area function that satisfies Eq. (10.55) is called a principal sectorial
area function (see Appendix C).
394 Unsymmetrical Beam Bending

Example 10.3 Calculating the Shear Center

As an example of determining the location of the shear center, consider the C-section
considered previously when discussing multiaxis bending. We will assume the thicknesses
ðtb , th Þ of the cross-section are thin. For such thin sections it is customary to measure the
distances of the parts of the cross-section along the centerlines of the section, as shown in
Fig. 10.27a. It is also customary to neglect terms involving higher powers of these thickness.
For example, consider calculating the area moment I zz . We have from the parallel axis
theorem (see Appendix A)
" # "  #
tb b3 ht3h b 2
I zz ¼ þ2 þ 2 hth
12 12 2
(10.56)
3 2
tb b hth b
ffi þ
12 2
where the middle term in the first line of Eq. (10.56) is neglected since it is much smaller than
the other terms. To determine the shear center, we only need to evaluate its location on the
z-axis since it lies on that axis of symmetry as shown in Fig. 10.27b so that ey ¼ b=2. Since
I yz ¼ 0 from Eq. (10.52), we only have to evaluate
I yωB
ez ¼  (10.57)
I zz
The sectorial area function ωB calculated using point B as the pole and point I as the initial
integration point (origin) is shown as the shaded area in Fig. 10.27b. Specifically, we have
ωB ¼ bs in the right-hand side and ωB ¼ 0 otherwise, since there is no projected area
from B for the other two sides. The distance y from the centroid to the right-hand side is
just y ¼ b/2 so
ð ð
s¼h

I yωB ¼ yωB dA ¼ ðb=2ÞðsbÞth ds ¼ th b2 h2 =4 (10.58)

A s¼0

Figure 10.27 (a) Geometry of a C-section. bh

(b) Parameters for calculating the location of the z
shear center. th
h y h h ey
tb C S
dc
ez s

B, I b
b

(a) (b)
 10.3 The Shear Center for Thin, Open Cross-Section Beams 395

z' Figure 10.28 The sectorial area function used in

−ez b / 2 − bh / 2 ez b / 2 + bh / 2
determining the shear center.

_ h B +
ez
s3 I s2
+
_
−ez b / 2 ez b / 2 y¢
s1
b

From this value and Eq. (10.56) and Eq. (10.57) we find, finally,

3th h2
ez ¼  (10.59)
ðtb b þ 6th hÞ

Thus, the shear center is below the bottom of the C-section, as we assumed in the previous
example where we considered multiaxis bending for a beam with this cross-section.
Alternatively, we can use, say, the ðy0 , z0 Þ coordinates and locate the pole B at a distance
along the center axis as seen in Fig. 10.28 and choose the initial point (origin) I as shown.
[Note: we could continue to use the centroidal coordinates ðy; zÞ and obtain the same result.
We have changed the coordinates to emphasize they can be noncentroidal.] The sectorial
area function is then
8
>
> ez s1 b=2  s1  b=2
<
ωB ¼ ez b=2 þ bs2 =2 0  s2  h (10.60)
>
>
:
ez b=2  bs3 =2 0  s3  h

whose graphs are shown as the shaded areas in Fig. 10.28. From Fig. 10.28 we can see that
the conditions
ð
ωB dA ¼ 0
A
ð (10.61)
0
z ωB dA ¼ 0
A

are automatically satisfied by this sectorial area function so we only need to examine the
remaining condition in Eq. (10.55) which gives

ð ð
b=2 ðh
0
y ωB dA ¼ 0 ¼ ðs1 þ b=2Þðez s1 Þtb ds1 þ ðbÞðez b=2 þ bs2 =2Þth ds2 (10.62)
A b=2 0
396 Unsymmetrical Beam Bending

Carrying out the integrals we find

ez tb b3 =12 þ 0 þ ez b2 hth =2 þ b2 h2 th =4 ¼ 0

3th h2 (10.63)
! ez ¼ 
ðtb b þ 6th hÞ

which agrees with our previous result.

We will now return to Eq. (10.41) and consider an example that describes the shear flow in
a thin, open section under bending and examine in detail the calculations needed to (1)
obtain the shear flow explicitly and  (2) locate
 the shear center. Two key quantities in
obtaining the shear flows are the Qy , Qz terms. These quantities are defined as first
moment integrals of a varying cross-sectional area, Aq , as we move in the cross-section
(see Eq. (10.40)). The coordinates ðy; zÞ in these first moment integrals are measured from
 of the entire cross-section, to a small-area element dA ¼ tds in Aq and
 centroid
point C, the
where the Qy , Qz terms can be calculated as
ð
Qy ¼ zdA ¼ zðsÞAq ðsÞ
Aq
ð (10.64)
Qz ¼ ydA ¼ yðsÞAq ðsÞ
Aq

Here, ðyðsÞ, zðsÞÞ are measured from the centroid of the entire cross-section, C, to the


centroid of the area Aq ðsÞ (see Fig. 10.29). In cases
  we can calculate the y; z; Aq
where
terms directly we can, therefore, find the Qy , Qz terms without performing the
underlying integrations.

Figure 10.29 Geometry for z

determining Qy and Qz in the y (s)
shear-flow expression, Eq. s
(10.41).

Aq
z (s)

C y
 10.3 The Shear Center for Thin, Open Cross-Section Beams 397

Example 10.4 Shear Flow and Shear Center in a Thin, Open Section Beam
Consider the thin, open section beam cross-section of Fig. 10.30. It is assumed that a 1000 lb
vertical shear force acts through the shear center at this section, producing bending but no
twisting. We want to determine the bending shear-flow distributions in this section and the
location of the shear center. The thickness of all parts of the cross-section is 0.1 in. and all the
dimensions (in inches) in Fig. 10.30 are measured to the centerlines. In this case V y ¼ 0 and
I yz ¼ 0 so the shear-flow expression of Eq. (10.39) becomes much simpler:
V z Qy
Δq ¼ (10.65)
I yy

If we break the cross-section up into three areas, we can calculate the area moment I yy with
the parallel axis theorem as
1 1
I yy ¼ 2  ð8Þð0:1Þ3 þ ð8Þð0:1Þð5Þ2 þ ð0:1Þð10Þ3
12 12 (10.66)
¼ 48:33 in:4

where the first term which involves the thickness cubed is neglected in the end result. Note
that the centroid C is along the horizontal centerline of the entire cross-section at a distance
yc as shown in Fig. 10.31, which we could also calculate with these areas but which will not
be needed. First, consider the shear flow in a portion of the section AB as shown in
Fig. 10.32a. We have
1000
Δq ¼ qðs1 Þ  0 ¼ Q ¼ 20:69Qy ¼ 20:69zAq
48:33 y
¼ 20:69½ð5Þð0:1Þðs1 Þ (10.67)

! qðs1 Þ ¼ 10:35s1 lb=in:

The shear flow is a linear function in AB and the value at B, q1B ,

is given in Fig. 10.32b. Since the shear flow is negative through-
3 in. 5 in. out AB the direction of the shear flow is as shown. In section
BD similar calculations (see Fig. 10.33) give

z Δq ¼ qðs2 Þ  0 ¼ 20:69zAq
1000 lb
e ¼ 20:69½ð5Þð0:1Þðs2 Þ (10.68)
10 in.
S C y ! qðs2 Þ ¼ 10:35s2 lb=in:

and the value at B, q2B , is shown in Fig. 10.33b along with the
actual shear-flow distribution in BD. Now consider the vertical

Figure 10.30 A beam where the shear force in a section acts through
3 in. 5 in. the shear center producing bending shear flows (shear stresses).
398 Unsymmetrical Beam Bending

3 5 Figure 10.31 Decomposition of the cross-section into three areas

Am ðm ¼ 1; 2, 3Þ.

A B D
A1 z 5 section BF. A portion of this section is shown in Fig. 10.34a. The
A3 yc shear-flow calculations are

y Δq ¼ qðs3 Þ  q3B ¼ 20:69zAq

C
A2 5 ¼ 20:69½ð5  s3 =2Þð0:1Þðs3 Þ (10.69)

E F G ! qðs3 Þ ¼ q3B  10:35s3 þ 1:03523 lb=in:

To determine the starting value q3B we must use conservation of

shear flow at the B junction, as shown in Fig. 10.34b, which gives
3 5 q3B þ 51:75 þ 31:03 ¼ 0
(10.70)
! q3B ¼ 82:78 lb=in:

and the shear flow in BF is explicitly

qðs3 Þ ¼ 82:78  10:35s3 þ 1:035s23 lb=in: (10.71)

At the middle of BF the shear flow is qð5Þ ¼ 109 lb=in: and the entire distribution is
sketched in Fig. 10.34c. For the lower flange Qy has the same magnitude as for the upper
flange but with the opposite sign so that shear-flow distributions in EF and FG are as shown
in Fig. 10.35. Figure 10.36a shows the complete shear-flow distribution in the cross-section.
These shear flows generate the forces shown in Fig. 10.36b where

Figure 10.32 (a) Geometry for determining the s1

shear flow in section AB. (b) The shear- q=0 q ( s1 )
z
flow distribution.

z =5 Aq

y
C
(a)

q1B = −31.03 lb / in.

A B
3

(b)
 10.3 The Shear Center for Thin, Open Cross-Section Beams 399

Figure 10.33 (a) Geometry for determining the shear flow

q ( s2 ) s2 q=0
in section BD. (b) The shear-flow distribution.
z
Aq z =5

C y
(a)

qB2 = −51.75 lb / in.

B 5 D
(b)

qB3 Figure 10.34 (a) Geometry for determining the shear

31.03 51.75 flow in section BF. (b) The shear flows at junction
Aq B. (c) The shear-flow distribution in BF (note the
s3 shear-flow values are negative so the shear flow is
z 5 actually up, as shown, with corresponding positive
z values).
q ( s3 ) qB3

C y
(a) (b)
– 82.78
B

–109

F
– 82.78
(c)

+51.75 Figure 10.35 The shear-flow distributions in

+31.03 the lower flange.

E 3 F F 5 G
400 Unsymmetrical Beam Bending

ð3
R1 ¼ 10:35s1 ds1 ¼ 46:58 lb
0
ð5
R2 ¼ 10:35s2 ds2 ¼ 129:38 lb (10.72)
0
1ð0


R3 ¼ 82:78 þ 10:35s3  1:035s23 ds3 ¼ 1000 lb
0

[Note: we have removed all the minus signs appearing in the shear flows in Eq. (10.72) so
that the forces are positive in the directions shown in Fig. 10.36b.] Since we obtained these
shear-flow distributions under the assumption of bending only, the moment of the 1000 lb
shear force about any point must be the same as the moment generated by the shear flows, or
equivalently the forces seen in Fig. 10.36b. Taking moments about O (Fig. 10.37), we find

Figure 10.36 (a) The shear-flow distribution R1 R2

in the cross-section. (b) The forces carried in
the various parts of the cross-section.

10 10
R3

R1 R2

3 5 3 5
(a) (b)

Figure 10.37 The moment about R1 R2

point O of the 1000 lb force acting
through the shear center must be 1000 lb
equal to the moment about O of the
forces generated by the shear flows. 10
e O R3 10
O

R1 R2

3 5 3 5
 10.3 The Shear Center for Thin, Open Cross-Section Beams 401

s2 Figure 10.38 The parameters used to locate the shear center by use of the sectorial
A B area function.
D

5
s1 X
K M O ¼ 1000e ¼ 10R2  10R1
e I
¼ ð10Þð129:38  46:58Þ (10.73)
5
! e ¼ 0:828 in:
E F G
which gives us the shear center location relative to point O.
s3
Another way to locate the shear center directly from the geometry is to
determine the principal pole location of the principal sectorial area func-
tion. As discussed previously, and as shown in Appendix C, the principal sectorial area
function can be found by evaluating the sectorial area function about a pole where Eq.
(10.55) is satisfied. We will call that pole here pole K (to avoid confusion with the other
labels being used) as shown in Fig. 10.38, and we will take the initial point for the
integration to be point I. Then with the coordinates along the various sections, as shown
in Fig. 10.38, we find the sectorial area function ωK to be:
8
>
< es1 5  s1  5
ωK ¼ 5e þ 5s2 5  s2  3 (10.74)
>
:
5e þ 5s3 3  s3  5

A sketch of this function is shown in Fig. 10.39. It is easy to see that this sectorial area
function satisfies
ð
ωK dA ¼ 0
A
ð (10.75)
yωK dA ¼ 0
A

so that the only remaining condition is

ð ð5 ð3
s2 zωK dA ¼ ðs1 Þðes1 Þtds1 þ ð5Þð5e þ 5s2 Þtds2
+
A 5 5
+ z (10.76)
s1 ð5
þ ð5Þð5e þ 5s3 Þtds3 ¼ 0
y
3

_
_ s3 Figure 10.39 The sectorial area function.
402 Unsymmetrical Beam Bending

where here the distance z is measured from the centroid so that z ¼ s1 in the vertical web and
z ¼ 5, z ¼ 5 for the top and bottom flanges, respectively. Carrying out the integrations and
dividing by the common thickness, t, gives
250
e þ 400e  400 ¼ 0
3
(10.77)
400
! e¼ ¼ 0:828 in:
400 þ 250=3

which agrees with our previous result. Note that with pole K located at the distance e given
by Eq. (10.77), the sectorial area function is a principal sectorial area function since the three
equations of Eq. (10.55) are satisfied, and we rewrite them here:
ð
ωK dA ¼ 0
A
ð
y0 ωK dA ¼ 0 (10.78)
A
ð
z0 ωK dA ¼ 0
A

With our choice of the pole K and the starting point I, the first two equations were
automatically satisfied so that only the third equation was needed. As demonstrated earlier,
it was not necessary to choose the ðy0 , z0 Þ coordinates to be measured from the centroid, even
though that was the choice we made here. However, if in our choice of K and I the first
equation in Eq. (10.78) was not satisfied, then we would be forced to use coordinates as
measured from the centroid in the other two equations in Eq. (10.78). Those two equations
would then locate the principal pole (shear center) even though ωK is not a principal
sectorial area function. The most general way to proceed is to write the starting value of
the sectorial area function ωK in terms of an unknown constant, ω0 , at the starting point I so



that ωK ¼ ωK ω0 , ey , ez where ey , ez are some unknown locations of pole K from an
arbitrary fixed point in the cross-section. Then the equations in Eq. (10.78) are three


equations for the three unknowns ω0 , ey , ez . Problem P10.3 gives you a chance to examine
this general approach for a C-section.

10.4 Thin, Closed Cross-Section Beams

All of the examples considered previously were thin, open cross-sections. One can also
examine the bending of thin, closed cross-sections containing one or more closed “cells.”
 10.4 Thin, Closed Cross-Section Beams 403

Figure 10.40 Bending of a single-cell closed section where the shear forces
B Vy act through the shear center. The shear flow has an unknown value, qB , at a
point B where an imaginary cut is taken through the thickness.
S
qB
The main difference from the thin, open cross-section case is that
the shear flows in closed cross-sections contain one or more
Vz unknown constant values since there are no edges in closed cross-
sections at which the shear flows must be zero. Additional condi-
tions must be considered to evaluate those unknowns. Those conditions will be described in
detail in Chapter 12 when we consider the combined bending and torsion of thin open and
closed cross-sections. Here, we will just summarize the case of a single closed cell where the
shear forces act through the shear center so that the member is in bending only (Fig. 10.40).
We can make an imaginary cut of the cross-section at any point B and call the value of the
shear flow at that location qB . Then the shear flow in the cell is given (Eq. (10.41)) as



V z I yz  V y I yy Qz þ V y I yz  V z I zz Qy
qðsÞ ¼ qB þ
D (10.79)
¼ qB þ q0 ðsÞ

where q0 ðsÞ is the known shear flow for an open section that has a real cut at B. As shown in
Chapter 11, the twist/unit length of the section, ϕ0 , induced by shear flow in a thin section is
given by:
þ
1 qðsÞ
ϕ0 ¼ ds (10.80)
GΩ tðsÞ
C

where t is the thickness of the section, G is the shear modulus, and Ω is twice the cross-
sectional area contained within the center line of the cross-section. When the shear
forces act through the shear center no twisting is induced, so we can find the constant
qB from
þ þ
1 q 1 ðqB þ q0 Þ
ds ¼ ds ¼ 0
GΩ t GΩ t
Þ q0 (10.81)
ds
! qB ¼ Þ t
ds
t
and once this constant is known the shear flow is also known explicitly. The location of the
shear center can then be determined by equating moments of the shear flow to the moments
of the shear forces, as done in Example 10.4. Chapter 12 gives more details when discussing
combined bending and torsion and also considers the case of multiple closed sections, so we
will not consider those extensions here.
404 Unsymmetrical Beam Bending

10.5 PROBLEMS

P10.1 An I-beam (I yy ¼ 937  106 mm4 , I zz ¼ 18:7  106 mm4 ) is subjected to a pure bend-
ing moment, M, at a very small angle (ϕ ¼ 1∘ ) with respect to the y-axis, as shown in
Fig. P10.1.
(a) Determine the orientation of the neutral axis.
(b) Determine the ratio of the maximum tensile stress in the beam for the moment at
this angle to the maximum tensile stress for the symmetrical bending case where
ϕ ¼ 0∘ .
P10.2 Two 10 mm thick steel plates are welded together to form the 120 mm by 80 mm
angle beam shown in Fig. P10.2. The beam, which is simply supported at its ends
in both the y- and z-directions, is subjected to a concentrated load P ¼ 4 kN
acting at point E (which is the shear center) in the y–z plane at an angle of 60
(see Fig. P10.2).
(a) Determine maximum tensile and compressive bending stresses at the section of
the beam where the load is applied.
(b) Determine the total deflection (magnitude and direction) at the section of the
beam where the load is applied.
For the properties of the cross-section, see Example A.1 in Appendix A.
P10.3 As discussed at the end of Section 10.3, the most general way to use the sectorial area
function to find the location of the shear center of a thin open cross-section is to


obtain a sectorial area function ωK ¼ ωK ω0 , ey , ez relative to an arbitrary pole
K where we solve the three equations of Eq. (10.78), rewritten here as:
ð
ωK dA ¼ 0
A
ð
yωK dA ¼ 0 (P10.1)
A
ð
zωK dA ¼ 0
z A


for the values ω0 , ey , ez , which then gives us a principal sectorial area
function and the location of the shear center. Note that the location of
305 mm the origin of the ðy; zÞ coordinates in Eq. (P10.1) is arbitrary so that
y ðy; zÞ here do not have to be centroidal. Consider the C-section of Fig.
I P10.3 where the sectorial area function value at the starting point I is
M

Figure P10.1 An I-beam subjected to a bending moment at a small angle with

181 mm respect to the y-axis.
 10.5 Problems 405

width = 10 mm Figure P10.2 (a) Two 10 mm thick steel

60o plates are welded together to form an angle
P z
beam that is simply supported at its ends in
both the y- and z-directions (details of the
3m 120 mm supports are not shown). The force P acts in
E y a plane parallel to the y–z plane. (b) The
2m cross-section of the beam (dimensions are
E measured to the ends of the cross-section).
80 mm

(a) (b)

z Figure P10.3 A thin C-section of uniform thickness, t. The point, I, is at

the origin of the (y, z) coordinates. All dimensions of the cross-section
b are measured along the center lines.

ey
b K ω0 and the location of the pole K is taken to be at the point


ez ey , ez in the ðy; zÞ coordinates shown. The cross-section has
y a constant thickness, t. Use Eq. (P10.1) and the ðy; zÞ coord-
inates of Fig. P10.3 to obtain the location of the shear center
b and the constant ω0 without explicitly using the symmetry of
I
the geometry. Sketch the sectorial area function for the
cross-section. [Hint: it is advantageous to use the symbolic
integration capabilities of MATLAB®.]
P10.4 One general way to find the shear center is by use of the sectorial area function (see
problem P10.3). Another general way, as discussed in this chapter, is to find the shear
flow due to bending only and then equate the moments of that shear flow to the
moments of the shear forces to locate where those shear forces must act. Consider for
example, a split circular section of radius R and thickness t, as shown in Fig. P10.4.
The shear center must lie along the horizontal axis of symmetry through the center of
the section. Determine:
(a) the approximate value of the area moment, I yy , for this
z
thin section,
t (b) the shear flow distribution due to a vertical shear force, P,
acting through the shear center, in terms of P, θ, and R,
R
T
y
e

Figure P10.4 A thin, split circular cylinder, of uniform thickness, t, where the
P shear force P for the section acts through the shear center.
406 Unsymmetrical Beam Bending

(c) the location, e, of the shear center along the horizontal axis by using the shear
flow distribution,
(d) the angle, θ, at which the shear flow is a maximum and that maximum value.


P10.5 The box beam shown in Fig. P10.5 I yy ¼ 688  106 mm4 supports a vertical shear
force V z ¼ 10 kN acting through the shear center. Determine the location of the shear
center (the distance, e, as shown) and show on a sketch the complete shear flow
distribution. All distances in Fig. P10.5 are measured from the center lines of the
cross-section.
P10.6 For the thin J-section shown in Fig. P10.6a, do the following.


(a) Determine the coordinates cy , cz of the centroid relative to the cross-section
center lines (see Fig. P10.6b) and the area moments and mixed area moment
with respect to the centroid C. Use the thin-
z section approximations where the cross-section
10 mm is decomposed into three rectangle having lengths
equal to the center line lengths shown in Fig.
P10.6b and terms involving the thickness cubed
are neglected when computing the area moments
Vz
10 mm and mixed area moment.
500 mm
C (b) Repeat part (a) where exact expressions for
e the centroid location, area moments and
y
20 mm mixed area moments are obtained with the
10 mm MATLAB® script sections (see Appendix A),

300 mm Figure P10.5 A box beam in bending.

Figure P10.6 (a) A thin J- Z z

section where all distances are
100
measured along the center lines. 100
(b) The center lines of the


section and the location cy , cz zc zc
of the centroid relative to those
cy
center lines. 200 yc 200 yc
C C
t = 6 mm for cz
all sections
y
Y
B 200
200
(a) (b)
 10.5 Problems 407

z Figure P10.7 A thin, unsymmetrical Z-section. All distances are

measured along the center lines.
b

zc
using the ðX ; Y ; Z Þ axes shown in Fig. P10.6a.
Show that your answers agree with those of
2b yc
part (a).
C

(c) Determine the location of the shear center ey , ez
t for all
sections relative to the ðy; zÞ axes shown in Fig. 10.6b. Use
y Eq. (10.52) and the origin of the ðy; zÞ coordinates,
point B, as the pole for calculating the sectorial
2b area function.
P10.7 In the bending of unsymmetrical sections, the quantity
D ¼ I yy I zz  I 2yz appears in the denominator of the expression for the flexure stress
(see Eq. (10.13)). Using the fact that the smallest principal area moment for the cross-
section is always positive (see Appendix A), show that D can never be zero.
P10.8 For the thin unsymmetrical Z-section shown in Fig. P10.7, obtain explicit expressions
for the location of the centroid and the area moments and mixed area moment using
the approximation for thin cross-sections, where terms involving the thickness cubed
are neglected. Determine the accuracy of these thin section results by comparing them
to the more exact values obtained with the MATLAB® script sections for b ¼ 10 in.,
t ¼ 2 in. (see Appendix A). All the dimensions and the (y, z) axes shown in Fig. P10.7
are measured along the center lines.
P10.9 An L-section with b ¼ 3in., h ¼ 5 in., t1 ¼ t2 ¼ 0:25 in: (see Fig. A.2 in
Appendix A) is subjected to a bending moment M ¼ 1000 in.-lb about the
y-axis. Using centroidal axes (y, z) parallel to the axes shown in Fig. A.2:
(a) determine the angle that the neutral axis makes with respect to the y-axis,
(b) find the maximum tensile and compressive flexural stresses and their locations in
the cross-section,
(c) repeat (a) and (b) using the flexure formula written in principal (Y, Z) axes
(see Eq. (10.20)). [Note: these are not the (Y, Z) axes shown in Fig. A.2.]
P10.10 Consider the triangular cross-section shown in Fig. P10.8.
From tables we can find that the area moments and mixed
A
area moment are given by

z bh3 hb3 b2 h2
I yy ¼ , I zz ¼ , I yz ¼
h 36 36 72

C y

B Figure P10.8 A triangular cross-section.

b C
408 Unsymmetrical Beam Bending

z Figure P10.9 A thin, closed semicircular cross-section of constant thickness, t,

where t a.

If a bending moment is applied only along the y-axis, show that

a neutral axis intersects the midpoint of the vertical side AC of
y the triangle.
t
P10.11 Consider the thin, closed semicircular cross-section shown in
t
Fig. P10.9. A shear force V z acts on this section through the
shear center. Determine the location of the shear center and
obtain the complete distribution of the shear flow in the
section in terms of V z and the radius a.
11 Uniform and Nonuniform Torsion

The torsion of a solid or hollow bar with a circular cross-section is one of the important
problems considered in elementary strength of material texts. In this chapter we consider the
torsion of bars having more general cross-sections, where the axial warping deformations
produced require that one develops a much more complex solution procedure. Like most
advanced strength of materials discussions, we will first consider the idealized case of
uniform torsion (also called Saint-Venant torsion) where the bar is completely free to warp.
Solutions of uniform torsion problems are obtained using both a warping function and a
Prandtl stress function approach. The case of nonuniform torsion, where the rate of twist
varies along the length of the bar and/or the warping of the bar is restrained, is also
considered using a warping function approach. Nonuniform torsion for general cross-
sections is not normally treated in advanced strength of materials texts. This omission is
unfortunate as many practical torsion problems likely fall in this category. This chapter will
give a complete description of the governing equations for nonuniform torsion of members
with general cross-sections.
Nonuniform torsion problems for general cross-sections typically require a numerical
solution. However, in the case of thin members, one can obtain more direct solutions for
both open and closed cross-sections. It will be shown that the sectorial area function plays a
major role in the solution of both uniform and nonuniform torsion problems for thin cross-
sections.

11.1 Torsion of Circular Cross-Sections – A Summary

In Chapter 1 we outlined the major elements of the strength of materials solution for the
torsion of circular shafts. To set the stage for dealing with more complex torsion problems in
this section, we want to summarize those results for circular sections so that we can contrast
them with more general cases. For a circular cross-section, we assumed that the displace-
ments were due to a pure rotational twisting, ϕðxÞ, of the cross-section with no displacement
in the axial direction, so that the displacements were given as
ux ¼ 0
uy ¼ zϕðxÞ (11.1)
uz ¼ yϕðxÞ

We also assumed that the rate of twist dϕ=dx ¼ ϕ0 was a constant. In that case the only


strains are exy , exz and the only nonzero stresses are likewise

409
410 Uniform and Nonuniform Torsion

σ xy ¼ Gϕ0 z
(11.2)
σ xz ¼ Gϕ0 y

where G is the shear modulus. These stresses were then related to the internal torque, T, in
the cross-section given by
T ¼ GJϕ0 (11.3)
where
ð ð

2 
J ¼ y þ z dA ¼ r2 dA
2
(11.4)
A A

is the polar area moment of the cross-section. If the rate of twist, ϕ0 , is a constant, Eq. (11.3)
shows that the internal torque is also a constant, a condition we will call uniform torsion.
Figure 11.1a shows a bar, fixed at x ¼ 0 and subjected to a torque, T 0 , at x ¼ L. In this case
the internal torque, T ¼ T 0 is a constant throughout the bar, as is the rate of twist, so this is
a case of uniform torsion. Since ϕ0 ¼ dϕ=dx is a constant ϕðxÞ ¼ ϕ0 x=L, where ϕ0 is the twist
at x ¼ L, and we have from Eq. (11.3) ϕ0 ¼ T 0 L=GJ, which is the familiar torque–twist
relationship used in strength of materials. The shear stresses in Eq. (11.2) are components of
the total shear stress, τ, in the cross-section, which varies radially from the center of cross-
section (see Fig. 11.1b) as
Tr
τ¼ (11.5)
J
so that the maximum stress in the cross-section occurs at the outer radius r ¼ c and is
given by
Tc
τ max ¼ (11.6)
J
For a circular bar, in the case of nonuniform torsion, where the rate of twist, ϕ0 , and the
internal torque, T, are functions of x, the basic assumptions made on the deformations for

Figure 11.1 (a) A bar of circular cross-section z

under uniform torsion. (b) The total shear
z
stress, τ, in the cross-section.
y
T0 τ
I0 r
c y

x
L

(a) (b)
 11.2 Uniform Torsion of Noncircular Cross-Sections – Warping Function 411

the uniform case are not violated so that nonuniform torsion problems of circular cross-
section members involving loading and boundary conditions different from that seen in
Fig. 11.1a can be solved directly without difficulty. As we will see, this is not the case for
noncircular sections. We will begin by considering the uniform torsion case for those more
general sections.

11.2 Uniform Torsion of Noncircular Cross-Sections – Warping function

The strength of materials solution described in the previous section cannot be applied to the
torsion of noncircular sections because the basic assumption of the deformation, Eq. (11.1),
requires that each planar cross-section of the bar remains planar and merely rotates (twists)
about the central axis. If one draws a grid on the surface of a circular rubber rod and then
twists it, as shown in Fig. 11.2a, then one can clearly see the twisting deformation of the
horizontal lines (which are transformed by the twisting from straight lines along the x-axis
into helixes) and that there is no distortion of the cross-section in the horizontal x-direction,
as assumed. However, if one takes a noncircular rubber rod and twists it, then the deform-
ation is as shown in Fig. 11.2b. There is again a twisting deformation of the horizontal lines
but there is also significant distortion of the cross-section in the x-direction, i.e., the cross-
sections both twist and warp. Warping significantly affects the strains and their correspond-
ing stresses in the bar. To include this warping, we will replace Eq. (11.1) by
ux ¼ ϕ0 ψ ðy; zÞ
uy ¼ zϕðxÞ (11.7)
uz ¼ yϕðxÞ

where ψ ðy; zÞ is the warping function, which represents the out-of-plane deformation (that has
for convenience been multiplied by the rate of twist, ϕ0 ), and we will assume that ϕ0 again is a
constant so that we have the bar under uniform torsion. In this case the strains are given by
∂ux ∂uy ∂uz
exx ¼ ¼ 0, eyy ¼ ¼ 0, ezz ¼ ¼0
∂x ∂y ∂z
 
∂ux ∂uy ∂ψ
γxy ¼ þ ¼ ϕ0 z
T0 T0 ∂y ∂x ∂y
  (11.8)
∂ux ∂uz ∂ψ
γxz ¼ þ ¼ ϕ0 þy
∂z ∂x ∂z
(a) ∂uz ∂uy
γyz ¼ þ ¼ ϕðxÞ  ϕðxÞ ¼ 0
∂y ∂z

T0 T0

Figure 11.2 (a) Uniform torsion of a circular bar with no out-

of-plane warping. (b) A noncircular bar showing out-of-
(b) plane warping.
412 Uniform and Nonuniform Torsion

so that the only nonzero stresses again are

 
0 ∂ψ
σ xy ¼ Gϕ z
∂y
  (11.9)
0 ∂ψ
σ xz ¼ Gϕ þy
∂z

These stresses must satisfy the equilibrium equations, which are

∂σ xy ∂σ xz
þ ¼0
∂y ∂z
∂σ xy
¼0 (11.10)
∂x
∂σ xz
¼0
∂x
The second and third equations are satisfied identically while the first equation gives

∂2 ψ ∂2 ψ
þ ¼0 (11.11)
∂y2 ∂z2

so that the warping function must satisfy Laplace’s equation. To solve this equation also
requires that we specify the behavior of ψ on the boundary line that defines the edge of cross-
section. We can obtain this boundary condition by examining the stresses on the boundary.
Figure 11.3a shows the shear-stress components ðσ xn , σ xt Þ at the edge of the cross-section
acting in the n and t directions, where n is a unit outward normal vector to the curved edge


and t is a unit tangent vector to that edge. The corresponding stress components σ xy , σ xz
acting in the y- and z-directions are shown in Fig. 11.3b, together with the angle α that the
normal and tangent vectors make with respect to the y- and z-axes. Relating these compon-
ents, we find
σ xn ¼ σ xy cos α þ σ xz sin α
¼ σ xy ny þ σ xz nz
(11.12)
σ xt ¼ σ xy sin α þ σ xz cos α
¼ σ xy nz þ σ xz ny


in terms of the components ny , nz of the unit normal.
t The shear-stress component σ xn must vanish on the
n boundary because σ xn ¼ σ nx where σ nx is the correspond-
V xt ing shear stress acting on the lateral surface (n ¼ constant)
t n of the bar in the x-direction and there are no such axial
z V xz
V xn
D D
Figure 11.3 (a) The stresses at the edge of the cross-section in the
y V xy directions normal and tangential to that edge. (b) The angles that
the unit normal and tangent vectors make with respect to the
(a) (b) y- and z-axes.
 11.2 Uniform Torsion of Noncircular Cross-Sections – Warping Function 413

surface stresses in our torsion problem. From the boundary condition σ xn ¼ 0 and Eq.
(11.12) and Eq. (11.9), we find
∂ψ ∂ψ
ny þ nz  zny þ ynz ¼ 0 (11.13)
∂y ∂z

or, equivalently,
∂ψ
¼ zny  ynz (11.14)
∂n
where ∂ψ=∂n is the normal derivative of the warping function. We can express this boundary
condition in another equivalent form if we let s be the arc length along the boundary.
Figure 11.4a then shows a small change of this arc length (in the t-direction) and Fig. 11.4b
shows the relationship between ds and changes in the y- and z-direction along the boundary.
It follows that
dz
ny ¼ cos α ¼
ds
(11.15)
dy
nz ¼ sin α ¼ 
ds
so from Eq. (11.15) and Eq. (11.14) we can also write the boundary condition as
∂ψ dz dy
¼z þy
∂n ds ds
(11.16)
1d
2 
¼ y þ z2
2 ds
However, it is usually more convenient to use the boundary condition in the form of Eq.
(11.14). Once the warping function is obtained by solving Laplace’s equation subject to the
boundary condition, we can obtain the relationship between the torque and rate of twist. In
this case we have
ð


T ¼ yσ xz  zσ xy dA
A  
0 ∂ψ (11.17)
σ xy ¼ Gϕ z
 ∂y 
∂ψ
σ xz ¼ Gϕ0 þy
∂z
t
t n n
ds
ds dz D
z D
–dy
Figure 11.4 (a) An element of length ds along the boundary.
y
(b) The geometry relating ds to changes in the y- and z-
(a) (b) coordinates.
414 Uniform and Nonuniform Torsion

so that
ð  

 ∂ψ ∂ψ
T ¼ Gϕ0 y2 þ z2 þ y z dA
∂z ∂y (11.18)
A

¼ Gϕ0 J eff

where J eff is the effective polar area moment for the noncircular cross-section. Equation
(11.18) shows that we can write the torque-rate of twist relation in the same form as for the
circular cross-section. Note that when the warping is absent ψ ¼ 0, and J eff just becomes the
ordinary polar area moment. The form seen in Eq. (11.18) is not the only form possible for
J eff . If we multiply the boundary condition of Eq. (11.13) by ψ and use Gauss’s theorem, we
have
ð     
∂ψ ∂ψ
ψ  z ny þ ψ þ y nz ds
∂y ∂z
C
ð       
∂ ∂ψ ∂ ∂ψ
¼ ψ z þ ψ þy dA
∂y ∂y ∂z ∂z (11.19)
A
ð " 2  2 #
∂ψ ∂ψ ∂ψ ∂ψ
¼ z þ þy dA ¼ 0
∂y ∂y ∂z ∂z
A

where C is the boundary of the cross-sectional area A and we have used the fact that ψ
satisfies Laplace’s equation in obtaining the final form seen in Eq. (11.19). If we add the
integral in Eq. (11.19) to the integral appearing in Eq. (11.18) we can find an alternate
expression for J eff given by
ð "    #
∂ψ 2 ∂ψ 2
J eff ¼ yþ þ z dA (11.20)
∂z ∂y
A

which is a form we will use shortly. It can be shown that the largest total shear stress, τ max ,
occurs at some point on the boundary of the cross-section [1]. Since on the boundary σ xn ¼ 0
the shear stress is tangent to the boundary and we have
τ max ¼ fσ xt gmax ¼ σ xy nz þ σ xz ny max
n
 o (11.21)
¼ Gϕ0 z  ∂ψ ∂y nz þ y þ ∂ψ
∂z ny max

We can define an effective maximum radius, ceff , as

   
∂ψ ∂ψ
ceff ¼ z nz þ y þ ny (11.22)
∂y ∂z max

so from Eq. (11.21) and Eq. (11.18) it follows that the maximum shear stress can also be
expressed in a form entirely similar to that of the circular cross-section, namely
 11.2 Uniform Torsion of Noncircular Cross-Sections – Warping Function 415

Tceff
τ max ¼ (11.23)
J eff

but here the J eff and ceff values are typically obtained from a numerical solution.
We have shown that the shear stresses of Eq. (11.9) produce a twisting moment in the
member. To ensure they generate a purely torsional moment, we also need to show that they
do not produce shear forces. Thus, consider the force generated in the y-direction given by
ð ð 
0 ∂ψ
V y ¼ σ xy dA ¼ Gϕ  z dA (11.24)
∂y
A A

Using the fact that the warping function satisfies Laplace’s equation, we can rewrite Eq.
(11.24) in terms of an equivalent area integral and then use Gauss’s theorem to transform
that area integral into an integral around the boundary of the cross-section:
ð       
0 ∂ ∂ψ ∂ ∂ψ
V y ¼ Gϕ y z þ y þ y dA
∂y ∂y ∂z ∂z
A
ð   (11.25)
0 ∂ψ
¼ Gϕ y  zny þ ynz ds ¼ 0
∂n
C

The boundary integral in Eq. (11.25) is zero because of the boundary condition, Eq. (11.14),
showing that V y ¼ 0. An entirely similar set of steps shows that the shear force V z ¼ 0 also.
This formulation we have been analyzing for the uniform torsion case is also called the
Saint-Venant torsional theory. Generally, we need to numerically solve Laplace’s equation,
Eq. (11.11), subject to the boundary conditions, Eq. (11.13), a task that can be done with
finite elements or with boundary elements. We can, however, obtain an approximate
analytical expression for the torsion of a thin rectangular section, as we will show in the
following example.

Example 11.1 Torsion of a Thin Rectangle

Consider the uniform torsion of the thin rectangular section of length b and thickness t
shown in Fig. 11.5a, where t  b. On the two long faces of the section, the warping function
must satisfy the boundary conditions

∂ψ 
¼z (11.26)
∂y y¼t=2

Since the other two faces of the cross-section at z ¼ b=2 are small, we will assume we can
treat the cross-section as if it were infinitely long and simply ignore the boundary conditions
at those faces. In that case we see that a warping function given by
ψ ¼ yz (11.27)
416 Uniform and Nonuniform Torsion

z Figure 11.5 (a) Torsion of a thin rectangular section. (b) The

shear-stress distributions across the thickness and near an end of the
cross-section.

b satisfies both Laplace’s equation and the boundary condi-

y
T tions of Eq. (11.26). The shear stresses, Eq. (11.9), are then
t
σ xz ¼ 2Gϕ0 y
(11.28)
σ xy ¼ 0

(a) (b) which produces a linear distribution of stress across the

thickness as shown in Fig. 11.5b. This distribution does
represent well the actual stress distribution over most of the
cross-section, although near the ends it is obviously in error since the shear stress must
change direction and “flow” around the ends so as to remain tangent to the boundary at the
edge of the cross-section (see Fig. 11.5b). If we compute the effective polar area moment
from Eq. (11.20), we find
ð
bt3
J eff ¼ 4 y2 dA ¼ 4I zz ¼ (11.29)
3
A

Since the maximum shear stress occurs at y ¼ t=2, from Eq. (11.22) we also find

ceff ¼ 2yjy¼t=2 ¼ t (11.30)

which gives
Tt
τ max ¼ (11.31)
J eff

We should note that if we had calculated J eff with the integral expression given in
Eq. (11.18), which came directly from the equation
ð


T ¼ yσ xz  zσ xy dA (11.32)
A

we would obtain a value for J eff that is smaller by a factor of two from the correct value
given by Eq. (11.29)! This discrepancy occurs because we have ignored the σ xy stress in the
cross-section and those stresses near the ends cannot be neglected when using either
Eq. (11.32) or, equivalently, Eq. (11.18). In contrast, using Eq. (11.20) the end effects are
negligible. Thus, when using approximate expressions in calculating such torsional con-
stants, all forms are not equally good for such approximations (although all forms are
equivalent for exact solutions). We will discuss this point again later.
 11.3 Uniform Torsion of Noncircular Cross-Sections 417

11.3 Uniform Torsion of Noncircular Cross-Sections – Prandtl

Stress function
In the Saint-Venant theory of torsion, the displacements are assumed to be
dϕ
ux ¼ ψ ðy; zÞ
dx
uy ¼ zϕðxÞ (11.33)

uz ¼ yϕðxÞ

and for uniform torsion, the rate of twist dϕ=dx ¼ ϕ0 is a constant. As we have seen, this
assumed deformation produces only shear stresses in the cross-section that we can write in
terms of the displacements as
   
∂ux ∂uz ∂ux 0
σ xz ¼ G þ ¼G þϕy
∂z ∂x ∂z
    (11.34)
∂ux ∂uy ∂ux 0
σ xy ¼ G þ ¼G ϕz
∂y ∂x ∂y

All the equilibrium equations are satisfied if

∂σ xy ∂σ xz
þ ¼0 (11.35)
∂y ∂z

This equation can be satisfied automatically by writing the stresses in terms of a function, Φ,
called the Prandtl stress function, where
∂Φ ∂Φ
σ xy ¼ , σ xz ¼  (11.36)
∂z ∂y

However, from Eq. (11.34) we have

∂ux ∂Φ
G ¼  Gϕ0 y
∂z ∂y
(11.37)
∂ux ∂Φ
G ¼ þ Gϕ0 z
∂y ∂z

which also implies

∂2 ux ∂2 Φ
G ¼  2  Gϕ0
∂z∂y ∂y
(11.38)
∂2 ux ∂2 Φ
G ¼ 2 þ Gϕ0
∂y∂z ∂z

These mixed derivatives of the displacement ux must be equal if we are to be able to integrate
the strains to find this displacement and this compatibility condition requires that the stress
function satisfy
418 Uniform and Nonuniform Torsion

∂2 Φ ∂2 Φ
þ 2 ¼ 2Gϕ0 (11.39a)
∂y2 ∂z

which is also often written as

r2 Φ ¼ 2Gϕ0 (11.39b)

in terms of the Laplacian operator r2 ¼ ∂2 =∂y2 þ ∂2 =∂z2 . Equation (11.39a) is called

Poisson’s equation. On the outer surface of the bar, there is no component of the stress
vector acting in the x-direction so that

T ðxnÞ ¼ σ nx ¼ σ xy ny þ σ xz nz ¼ 0 (11.40)

and the y- and z-components of the stress vector are identically zero. In terms of the Prandtl
stress function Eq. (11.40) becomes

∂Φ ∂Φ
ny  nz ¼ 0 (11.41)
∂z ∂y

By examining a small element along the boundary (Fig. 11.6) we see that Eq. (11.41) also
implies

dΦ ∂Φ dz ∂Φ dy
¼ þ ¼0 (11.42)
ds ∂z ds ∂y ds

so that Φ must be a constant on the boundary. For a simply connected cross-section (a cross-
section with no holes), we can take that constant to be zero so the boundary condition
becomes simply

Φ ¼ 0 on the boundary C (11.43)

To find the internal torque, T, in terms of the stress function we start from the relation
between the torque and the shear stresses and express that relationship in terms of the stress
function, yielding

Figure 11.6 (a) A small element at the boundary. (b) The ds n

z D
geometry relating that element to the unit normal. D
dz ds
y
–dy

(a) (b)
 11.3 Uniform Torsion of Noncircular Cross-Sections 419

ð


T ¼ σ xz y  σ xy z dA
A ð 
∂Φ ∂Φ
¼ yþ z dA
∂y ∂z
ðA  
∂ ∂
¼ ðΦyÞ þ ðΦzÞ  2Φ dA (11.44)
∂y ∂z
ðA h i ð
¼  Φyny þ Φznz dA þ 2 ΦdA
ðC A

¼ 2 ΦdA
A

where we have used Gauss’s theorem and where the integral over the boundary C vanishes
since Φ ¼ 0. Thus, the torque is just twice the integral of the Φðy; zÞ function over the cross-
section.
As mentioned previously, the largest shear stress in the cross-section occurs at some point
on the boundary. Since the total shear stress τ must be tangent to the boundary (Fig. 11.7),
we have
σ xz ¼ τ cos α
(11.45)
σ xy ¼ τ sin α

so that
∂Φ ∂Φ ∂Φ
¼ ny þ nz
∂n ∂y ∂z (11.46)
¼ σ xz cos α þ σ xy sin α
¼ τ

which gives
 
∂Φ
τ max ¼  (11.47)
∂n max on C

As with the warping function, we can make these results appear similar to the familiar
formulas for the torsion of a circular cross-section if we define a normalized stress function

¼ Φ=Gϕ0 where
Φ

¼ 2 in the cross-section
r2 Φ
t n
¼ 0 on the boundary (11.48)
Φ
D τ
z D V xz
τ
−V xy Figure 11.7 (a) The total shear stress on the boundary. (b) The
y relationship of the total shear stress to the stress components in the
(a) (b) (y, z) directions.
420 Uniform and Nonuniform Torsion

If we let T ¼ GJ eff ϕ0 then

ð


J eff ¼ 2 ΦdA (11.49)
A

and
Tceff
τ max ¼ (11.50)
J eff
where
 


∂Φ
ceff ¼  (11.51)
∂n max on C

Prandtl Stress Function Solutions

There are a number of cases where we can obtain explicit results using stress functions. Here
we discuss some of the important examples.

(1) Circular cross-section of radius c

In this case, the normalized Prandtl stress function is




¼  1 y2 þ z2  c2 ¼  1 r2  c2
Φ (11.52)
2 2
The effective polar area moment is just the ordinary polar area moment where
ð ð
r¼c

 πc4
J eff

¼ J ¼ 2 ΦdA ¼ 2π c2  r2 rdr ¼ (11.53)
2
A r¼0

For the maximum shear stress, we have

 
T

∂Φ Tc
τ max ¼  ¼ (11.54)
J eff ∂r r¼c J

note that in this case

 
∂Φ 0 ∂ux 0
σ xz ¼  ¼ Gϕ y ¼ G þϕy
∂y ∂z
  (11.55)
∂Φ ∂ux
σ xy ¼ ¼ Gϕ0 z ¼ G  ϕ0 z
∂z ∂y
which implies that
∂ux ∂ux
¼ ¼0 (11.56)
∂y ∂z
 11.3 Uniform Torsion of Noncircular Cross-Sections 421

so that ux is a constant that we can take equal to zero. Thus, there is no warping, as expected,
in this case.

(2) Elliptical cross-section (with semimajor axes of lengths a and b along the y- and z-axes,
respectively)
Here, we have
 2 

¼
a b  y þ z  1
2 2 2
Φ (11.57)
a2 þ b2 a2 b2

and the effective polar area moment is

πa3 b3
J eff ¼
 (11.58)
a2 þ b2

The maximum shear stress is

2T
ðτ max Þz¼b ¼ for b < a (11.59)
πab2
If we integrate the shear-stress expressions to find the warping displacement, ux , we find


T b2  a2
ux ¼ yz (11.60)
πa3 b3 G

(3) Thin rectangular section

Consider the same case examined with the warping function where the cross-section is of
length b in the z-direction and has a small thickness t in the y-direction (Fig. 11.5a). In this
case we might expect that Φ
¼ f ðyÞ so that

¼ d 2f
r2 Φ ¼ 2
dy2 (11.61)
f ¼ 0 on y ¼ t=2

which gives
2 

¼ t
Φ y2
(11.62)
4

Using this stress function, the effective polar area moment is

ð
t=2

J eff ¼ 2b
¼ 1 bt3
Φdy (11.63)
3
t=2
422 Uniform and Nonuniform Torsion

which agrees with our previous result. The maximum shear stress is
 
T

∂Φ 3T
τ max ¼ ∓ ¼ 2 (11.64)
J eff ∂y y¼t=2 bt

This value of J eff , which was obtained with approximate solutions for both the warping
function and the Prandtl stress function, can be compared with the first term in an infinite
series solution that is good to several percent for a rectangle of arbitrary size given by [2]
  
bt3 192 t πb
J eff ¼ 1  5 tan (11.65)
3 π b 2t

Consider a thin rectangle where t=b ¼ 0:2, for example. In that case Eq. (11.65) gives

bt3
J eff ¼ ð1:017Þ (11.66)
3

so that the approximate value of bt3 =3 is quite good.

If we use the relationship between the torque and the stresses to obtain J eff , we find
ð


T ¼ σ xz y  σ xy z dA
A
ð
t=2
0 (11.67)
¼ Gϕ 2y2 bdy
t=2
3
0 bt
¼ Gϕ
6
so that

bt3
J eff ¼ (11.68)
6

which is an incorrect value that is only one half the value obtained from either the warping
function and Eq. (11.20) or the Prandtl stress function and Eq. (11.49). As mentioned
previously, the reason for this discrepancy is that we neglected in our approximation the
shear stress σ xy that develops near the end of the cross-section. While those shear stresses
may be localized in their importance, they do have a large moment arm (on the order of
length b/2) as opposed to the maximum shear stresses σ xz , whose moment arm is on the order
of the thickness, t/2. To estimate the size of the contribution from the σ xy stress, we will
assume that the stress function near the ends of the cross-section varies linearly from zero to
its maximum value over a parabolic-shaped area, Aend , of length h (Fig. 11.8), i.e.,
2

ffit z
Φ (11.69)
4h
 11.3 Uniform Torsion of Noncircular Cross-Sections 423

y Figure 11.8 (a) A parabolic area of height h over which the shear stress
σ xy is assumed to be significant, and over which Φ
is assumed to vary
2th
Aend = linearly from zero to its maximum value, as shown in (b).
3
t z

and
h t2
(a) σ xy ffi Gϕ0 (11.70)
4h
We can estimate the torque produced by this stress (from both
F max = t 2 / 4 ends) as
2 3
ð 2
 
(b) h 6 b 7 0 t 2th
T ffi 24σ xy dA5 ¼ Gϕ b
2 4h 3
Aend (11.71)
bt3
¼ Gϕ0
6
Thus, we obtain a contribution that is equal to that calculated from only σ xz and the total
torque from both stresses gives the correct result.
To evaluate the warping of the thin rectangle, note that for our approximate solution
 
∂Φ ∂ux 0
σ xy ¼ ¼G ϕz ¼0
∂z ∂y
  (11.72)
∂Φ ∂ux 0 0
σ xz ¼  ¼G þ ϕ y ¼ 2Gϕ y
∂y ∂z

so that
∂ux ∂ux
¼ ϕ0 z, ¼ ϕ0 y (11.73)
∂y ∂z

Integrating both of these equations we find

ux ¼ ϕ0 yz þ f ðzÞ
(11.74)
ux ¼ ϕ0 yz þ gðyÞ

where f and g are both arbitrary functions. Equating these two expressions, we find that we
must have f ¼ g ¼ constant, where we can take this constant value to be zero (since it just
produces a translation in the x-direction), giving the same warping function obtained earlier
in Eq. (11.27).

The Membrane Analogy

One of the reasons the Prandtl stress function has been more widely used to describe torsion
problems than the warping function approach is that there is a correspondence between the
424 Uniform and Nonuniform Torsion

s s Figure 11.9 A thin membrane under a pressure, p. The

w=0 w=0 membrane is stretched with a tension, s. The behavior of the
p
displacement, w, of the membrane is analogous to that of the
w(y, z) normalized Prandtl stress function.

z Figure 11.10 The (exaggerated) shape of the deflection of a

membrane for a thin rectangular cross-section (away from
the ends).

w=0 y
p behavior of the stress function and the deflection of a
w=0 thin membrane under pressure. For example, Fig. 11.9
shows the side view of an inflated membrane under an
t
internal pressure, p, where w is the vertical deflection of
the membrane and s is the constant tension in the mem-
brane. From equilibrium of a membrane element one can show [3] that the membrane
satisfies

∂2 w ∂2 w p
þ 2 ¼
∂y 2 ∂z s (11.75)
w ¼ 0 on the boundary

which is identical to the solution of the torsion problem (for a simply connected cross-
section) in terms of the normalized Prandtl stress function if we make the substitution
p

w¼ Φ (11.76)
s

The slopes of the membrane are proportional to the shear stresses because of Eq. (11.36).
A similar analogy can be used for multiply connected cross-sections [4]. By examining the
deflection and slopes of the membrane at various locations one can obtain a sense of the
distribution of the shear stresses in a given cross-section. Figure 11.10, for instance, shows
the shape of a membrane for thin rectangular section, where the deflected shape of the
membrane follows the same parabolic distribution as the normalized Prandtl function, Eq.
(11.62), over most of the cross-section. Near the ends (not shown) the changing slopes of the
membrane similarly show the development of the σ xy shear stress.

Torsion of Multiply Connected Cross-Sections

We previously showed that the stress function had to be a constant on any boundary of the
cross-section. This is true for all boundaries of a multiply connected cross-section, i.e., the
outer boundary and the holes. When we have a simply connected cross-section, we can
always take the single constant on the outer boundary to be zero since we always can add or
 11.3 Uniform Torsion of Noncircular Cross-Sections 425

) 0 Figure 11.11 A multiply connected cross-section where the Prandtl stress

function takes on different constant values at each hole.
C1 C2
subtract a constant from the Prandtl stress function without
) Km
) affecting the stresses. However, if there is one or more holes in
K1 ) K2
Cm
a cross-section, the constant values on those boundaries cannot
all be taken to be zero. Generally, one takes Φ ¼ 0 on the outer
boundary of the cross-section and Φ ¼ K j ðj ¼ 1; . . . ; mÞ on m
holes (Fig. 11.11), where the constants K j are not zero. To find
these m constants we need to specify m additional conditions on our problem. These
conditions can be obtained by requiring that the ux displacement be single-valued.
A single-valued displacement can be assured if we require
þ
dux ¼ 0 ðj ¼ 1; . . . ; mÞ (11.77)
Cj

where C j ðj ¼ 1; . . . ; mÞ are the closed lines defining the holes. But recall

ux ¼ ux ðy; zÞ ¼ ϕ0 ψ ðy; zÞ, ϕ0 ¼ dϕ=dx ¼ constant

 
∂ux
σ xy ¼ G  zϕ0
∂y (11.78)
 
∂ux
σ xz ¼ G þ yϕ0
∂z

so that placing these relations into

þ þ 
∂ux ∂ux
dux ¼ dy þ dz ¼0 (11.79)
∂y ∂z
Cj Cj

we obtain
þ þ


σ xy dy þ σ xz dz þ Gϕ0 ðzdy  ydzÞ ¼ 0 (11.80)
Cj Cj

where the closed line integrals are taken in a counterclockwise sense. We can write these
integrands in terms of the components of the outward unit normal vector (see Eq. (11.15))
and the Prandtl stress function (Eq. (11.36)) as
þ  þ
∂Φ ∂Φ 0
 nz þ ny ds  Gϕ znz þ yny ds ¼ 0 (11.81)
∂z ∂y
Cj Cj
426 Uniform and Nonuniform Torsion

or equivalently, using Gauss’s theorem

þ  ð 
∂Φ ∂ ∂
 ds ¼ Gϕ0 ðzÞ þ ðyÞ dA
∂n ∂z ∂y
Cj Aj
ð
(11.82)
¼ 2Gϕ0 dA
Aj
0
¼ 2Gϕ Aj

where Aj is the area that is enclosed by the contour C j . Thus, to find the m constants we have
to satisfy the m conditions
þ 
∂Φ
 ds ¼ 2Gϕ0 Aj ðj ¼ 1; . . . ; mÞ (11.83)
∂n
Cj

or, in terms of the total shear stress, τ, on these boundaries,

þ
τds ¼ 2Gϕ0 Aj ðj ¼ 1; . . . ; mÞ (11.84)
Cj

Historically, because of the membrane analogy most texts have emphasized the Prandtl
stress function approach over the formulation based on the warping function. However,
when solving these torsion problems numerically for multiply connected cross-sections the
use of the warping function is easier in that there is only one boundary condition (see
Eq. (11.14)) that is applied to all boundaries and there are no additional conditions such as
Eq. (11.84) that need to be considered.

11.4 Nonuniform Torsion

The case of uniform torsion is a simple condition, much like the case of pure bending for
beams. In Chapter 1, we extended the conditions of pure bending to more practical cases
where the bending moment is not a constant along the axis of the beam. By letting the
bending moment (and stress σ xx ) vary along the beam, however, we found that shear stresses
are induced that also must be considered. In Chapter 10 we saw how to model such shear
stresses explicitly for the bending of unsymmetrical thin cross-section beams. For the case of
uniform torsion, the rate of the angle of twist and the internal torque are constants and the
warping of the cross-section is allowed to occur unimpeded. In real applications, however,
there typically are some constraints on the warping at boundaries. At a rigid boundary, for
example, the warping would be zero. Such restraint of warping will cause the warping
displacement to vary along the length of the member, i.e., we will have ux ¼ ux ðxÞ, and that
variation in turn will produce a varying axial stress, σ xx ðxÞ. We can try to include variable
warping by extending the basic assumptions on the deformation (see Eq. (11.33)) to be
 11.4 Nonuniform Torsion 427

ux ¼ ϕ0 ðxÞψ ðy; zÞ
uy ¼ zϕðxÞ (11.85)
uz ¼ yϕðxÞ

i.e., the in-plane displacements still reflect the same basic twisting deformation and we again
have out-of-plane warping but we will let the rate of twist be a function of the x-coordinate.
Under these conditions we say that the member is under nonuniform torsion since, if the rate
of twist is not a constant, the internal torque will vary also. Note that we can also have
nonuniform torsion when there are applied distributed torsional loads along the length of
the member since those loads will cause the internal torque to vary along the length of the
member and so the rate of twist will vary also. Based on the displacements assumed in Eq.
(11.85) we have the strains
∂ux 00
exx ¼ ¼ ϕ ðxÞψ ðy; zÞ
∂x
 
∂ux ∂uy 0 ∂ψ
γxy ¼ þ ¼ ϕ ðxÞ z
∂y ∂x ∂y
  (11.86)
∂ux ∂uz ∂ψ
γxz ¼ þ ¼ ϕ0 ðxÞ þy
∂z ∂x ∂z
∂uy ∂uz
γyz ¼ þ ¼ ϕðxÞ þ ϕðxÞ ¼ 0
∂z ∂y
00
where ϕ ¼ d 2 ϕ=dx2 . If we assume that σ xx is the only significant normal stress in the member
then for an isotropic elastic material the only nonzero stresses are
00
σ xx ¼ Eϕ ðxÞψ ðy; zÞ
 
∂ψ
σ xy ¼ Gγxy ¼ Gϕ0 ðxÞ z
∂y (11.87)
 
0 ∂ψ
σ xz ¼ Gγxz ¼ Gϕ ðxÞ þy
∂z

Placing these stresses into the local equilibrium equations (see Eq. (3.11)) with zero body
forces gives
∂ h 00 i ∂  
∂ψ

0
Eϕ ðxÞψ ðy; zÞ þ Gϕ ðxÞ z
∂x ∂y ∂y
  
∂ 0 ∂ψ
þ Gϕ ðxÞ þy ¼0
∂z ∂z
  (11.88)
00 ∂ψ
Gϕ ðxÞ z ¼0
∂y
 
00 ∂ψ
Gϕ ðxÞ þy ¼0
∂z
428 Uniform and Nonuniform Torsion

Equation (11.88) shows that, based on our original deformation assumption, we cannot
00
satisfy the second and third equilibrium equations unless either ϕ ¼ 0 (in which case we are
back to the uniform torsion case) or the torsional stresses vanish and we have no torsion.
The first equation gives
00
∂2 ψ ∂2 ψ Eϕ ðxÞ
þ ¼ 0 ψ (11.89)
∂y2 ∂z2 Gϕ ðxÞ
00
which would require that unless ϕ ¼ 0 we must have ψ ¼ ψ ðx; y; zÞ, which contradicts our
original assumption that ψ ¼ ψ ðy; zÞ. Thus, we need to modify our assumptions about the
behavior in this problem. From beam-bending theory, as mentioned previously, we know
that a normal stress that varies along the beam axis requires the development of shear
stresses in the cross-section. In our torsion case a varying normal stress will produce shear
stresses that can cause an additional displacement ux of the cross-section, uSx , called second-
ary warping. Let ψ S ðx; y; zÞ represent this secondary warping, which will vary in the
x-direction as well as the y- and z-directions. This secondary warping will be associated
with secondary stresses σ Sxy ; σ sxz , where

uSx ¼ ψ S ðx; y; zÞ
∂uSx ∂ψ S
σ Sxy ¼ G ¼G
∂y ∂y (11.90)

∂uSx ∂ψ S
σ Sxz ¼ G ¼G
∂z ∂z

This secondary warping and these secondary stresses will be incorporated into the case of
nonuniform torsion in the following manner. Let us again modify the uniform torsion case
where we simply let the rate of twist vary, i.e., we assume

ux ¼ ϕ0 ðxÞψ P ðy; zÞ
uy ¼ zϕðxÞ (11.91)
uz ¼ yϕðxÞ

Here the in-plane displacements represent the twisting of the cross-section and there is
warping present that we will call the primary warping of the cross-section, represented by
ψ P ðy; zÞ. These are the same expressions as used in uniform torsion theory except the rate of
twist is now allowed to vary. This change from the uniform torsion case means that the
primary warping will generate a normal stress that we will call the warping normal stress, σ wxx ,
where
∂ux 00
σ wxx ðx; y; zÞ ¼ E ¼ Eϕ ðxÞψ P ðy; zÞ (11.92)
∂x
 11.4 Nonuniform Torsion 429

in addition to the primary shear stresses, σ Pxy ; σ Pxz , given by

 P 
∂ψ
σ Pxy ¼ Gϕ0 ðxÞ z
∂y
 P  (11.93)
0 ∂ψ
σ xz ¼ Gϕ ðxÞ
P
þy
∂z
Since, as we previously indicated, the warping normal stress produces additional secondary
shear stresses the total shear stresses will be the sum of the primary and secondary shear
stresses, i.e.,
σ xy ¼ σ Pxy þ σ Sxy
(11.94)
σ xz ¼ σ Pxz þ σ Sxz

Assuming the warping normal stress and the total shear stresses are in equilibrium (see the
first equilibrium equation of Eq. (3.11), in the absence of body forces) we have

∂σ wxx ∂σ Pxy ∂σ Pxz ∂σ Sxy ∂σ Sxz

þ þ þ þ ¼0 (11.95)
∂x ∂y ∂z ∂y ∂z

In order to satisfy Eq. (11.95), we will further require that the primary shear stresses are in
equilibrium with themselves while the secondary shear stresses are in equilibrium with the
warping normal stress, leading to the conditions
∂σ Pxy ∂σ Pxz
þ ¼0
∂y ∂z
(11.96)
∂σ wxx ∂σ Sxy ∂σ Sxz
þ þ ¼0
∂x ∂y ∂z
This decomposition makes sense since the primary shear stresses are the same as those in the
uniform torsion case where there is no normal stress, while the secondary shear stresses are
present because of the varying normal stress. It is true that we cannot satisfy the two
remaining equilibrium equations, so this nonuniform torsion theory lacks a completeness
that one should be aware of. This is not unlike engineering beam theory for bending,
however, where in general all the local equilibrium equations are also not satisfied.
If we place the stresses of Eq. (11.90) and Eq. (11.93) into Eq. (11.96), we find the
governing equations for both the primary and secondary warping functions:

∂2 ψ P ∂2 ψ P
r2 ψ P ¼ þ ¼0
∂y2 ∂z2
000
(11.97)
∂2 ψ S ∂2 ψ S Eϕ ðxÞ P
rψ ¼
2 S
þ ¼ ψ
∂y 2 ∂z 2 G
000
where ϕ ¼ d 3 ϕ=dx3 . These governing equations are now consistent with our original
assumptions that ψ P ¼ ψ P ðy; zÞ and ψ S ¼ ψ S ðx; y; zÞ so by the inclusion of the secondary
430 Uniform and Nonuniform Torsion

warping and secondary shear stresses into the nonuniform torsion case, we have removed the
problems associated with Eq. (11.89) described earlier. To solve the equations in Eq. (11.97)
one also needs to specify boundary conditions. Recall from Eq. (11.12)

σ xn ¼ σ xy ny þ σ xz nz
(11.98)
σ xt ¼ σ xy nz þ σ xz ny

so with the decomposition into primary and secondary shear stresses we also have

σ Pxn ¼ σ Pxy ny þ σ Pxz nz

σ Pxt ¼ σ Pxy nz þ σ Pxz ny
(11.99)
σ Sxn ¼ σ Sxy ny þ σ Sxz nz
σ Sxt ¼ σ Sxy nz þ σ Sxz ny

Substituting Eq. (11.90) and Eq. (11.93) into Eq. (11.99) we find
 P 
0 ∂ψ
σ xn ¼ Gϕ ðxÞ
P
 zny þ ynz
∂n
 P 
∂ψ
σ Pxt ¼ Gϕ0 ðxÞ þ yny þ znz
∂t
(11.100)
∂ψ S
σ Sxn ¼ G
∂n
∂ψ S
σ Sxt ¼ G
∂t
At the boundary we have seen previously that σ xn ¼ 0, so requiring that both the primary
and secondary shear stresses satisfy this condition, the boundary conditions for the primary
and secondary warping functions are

∂ψ P ∂ψ S
¼ zny  ynz , ¼0 (11.101)
∂n ∂n

Thus, the case of nonuniform torsion requires that we solve a boundary value problem for
the primary warping given as

∂2 ψ P ∂2 ψ P
þ ¼0 in the cross-section A
∂y2 ∂z2
(11.102)
∂ψ P
¼ zny  ynz on the boundary C
∂n

Once this solution is known, then we must solve a second boundary value problem for the
secondary warping where
 11.4 Nonuniform Torsion 431

000
∂2 ψ S ∂2 ψ S Eϕ ðxÞ P
þ ¼ ψ in the cross-section A
∂y 2 ∂z 2 G
(11.103)
∂ψ S
¼ 0 on the boundary C
∂n

The first boundary value problem is the same problem found earlier in the uniform torsion
case. We see that, because of the presence of ψ P in the governing equation, the secondary
boundary value problem requires the solution of the first boundary value problem.
000
A knowledge of the twist ϕðxÞ is also required because of the ϕ ðxÞ term. Both boundary
value problems must typically be solved numerically. We will say more about their solutions
later. Once these solutions are obtained, then we can find the warping normal stress and the
shear stresses from Eq. (11.90) and Eq. (11.93). The shear stresses must produce the total
twisting moment, M x ðxÞ, which we can also decompose into a primary twisting moment,
M Px ðxÞ, and a secondary twisting moment, M Sx ðxÞ, where
ð
M Px ðxÞ ¼ yσ Pxz  zσ Pxy dA
A
ð 
0

 ∂ψ P ∂ψ P
¼ Gϕ ðxÞ y þz
2 2
þy z dA
∂z ∂y
A
ð (11.104)
M Sx ðxÞ ¼ yσ Sxz  zσ Sxy dA
A
ð 
∂ψ S ∂ψ S
¼G y z dA
∂z ∂y
A

We recognize that the expression for the primary twisting moment is the same as the one
given by the uniform torsion theory where

M Px ðxÞ ¼ GJ eff ϕ0 ðxÞ (11.105)

and
ð 

 ∂ψ P ∂ψ P
J eff ¼ y þz
2 2
þy z dA (11.106)
∂z ∂y
A

is the effective polar area moment of the cross-section (see Eq. (11.18). We can also use the
alternate expression given in Eq. (11.20). For the secondary twisting moment, we note that


by placing the warping functions ψ P , ψ S in the Gauss–Green theorem we have
ð ð

P 2 S 

ψ r ψ  ψ r ψ dA ¼ ψ P ∂ψ S =∂n  ψ S ∂ψ P ∂n ds
S 2 P
(11.107)
A C
432 Uniform and Nonuniform Torsion

But using the boundary value problem relations of Eq. (11.102) and Eq. (11.103) we have
000 ð ð
Eϕ ðxÞ
P 2

 ψ dA ¼  ψ S zny  ynz ds (11.108)
G
A C

and we can use Gauss’s theorem to transform the boundary integral to an area integral,
giving
000 ð ð
Eϕ ðxÞ
P 2

 ψ dA ¼  ψ S zny  ynz ds
G
A
ð C  (11.109)
∂ψ S ∂ψ S
¼ z þy dA
∂y ∂z
A

Comparing this result with the secondary twisting moment expression in Eq. (11.104), we
obtain
000
M Sx ðxÞ ¼ EJ ψ ϕ ðxÞ (11.110)

where the warping constant, J ψ , is given by (see Appendix C where a similar function is
defined in terms of the sectorial area function for thin sections)
ð

2
J ψ ¼ ψ P dA (11.111)
A

Like the primary shear stresses, the secondary shear stresses should not generate any net
shear forces. Thus, we must show that
ð ð S
∂ψ
V y ¼ σ Sxy dA ¼ G dA ¼ 0
∂y
A A
ð ð S (11.112)
∂ψ
V z ¼ σ Sxz dA ¼ G dA ¼ 0
∂z
A A

Consider the V y term. We have

ð S ð     ð  2 S 
∂ψ ∂ ∂ψ S ∂ ∂ψ S ∂ψ ∂2 ψ S
dA ¼ y þ y dA  y þ dA (11.113)
∂y ∂y ∂y ∂z ∂z ∂y2 ∂z2
A A A

The first integral on the right-hand side of Eq. (11.113) can be transformed by Gauss’s
theorem into a line integral on the boundary of the cross-section, while in the second integral
we can use the governing differential equation in Eq. (11.103) to find
ð S ð  S  000 ð
∂ψ ∂ψ ∂ψ S Eϕ ðxÞ
dA ¼ y ny þ nz dC þ yψ P dA (11.114)
∂y ∂y ∂z G
A C A
 11.4 Nonuniform Torsion 433

The first integral in Eq. (11.114) vanishes by the boundary conditions on the secondary
warping function while the second integral vanishes if
ð
yψ P dA ¼ 0 (11.115)
A

so that under these conditions V y ¼ 0. In an entirely similar fashion, we can show that
V z ¼ 0 if
ð
zψ P dA ¼ 0 (11.116)
A

We will see shortly that the primary warping function must satisfy Eq. (11.115) and
Eq. (11.116) so that the secondary shear stresses indeed produce only a secondary twisting
moment. Note that if we integrate the differential equation in Eq. (11.103) over the cross-
section and apply the boundary conditions, we also have
ð 2 S  000 ð
∂ψ ∂2 ψ S Eϕ ðxÞ
þ dA ¼  ψ P dA
∂y2 ∂z2 G
A
ð S  A
(11.117)
∂ψ ∂ψ S
¼ ny þ nz dC ¼ 0
∂y ∂z
C

which gives an additional condition on the primary warping function, namely

ð
ψ P dA ¼ 0 (11.118)
A

The physical meaning of these three conditions on the primary warping function is that in
the case of nonuniform torsion, this primary warping function must not generate any axial
stresses or bending stresses since from Eq. (11.92) we see that Eq. (11.115), Eq. (11.116), and
Eq. (11.118) can also be written in terms of the warping normal stress, σ wxx , i.e.,
ð ð
yψ dA ¼ yσ wxx dA ¼ 0
P

A A
ð ð
zψ P dA ¼ zσ wxx dA ¼ 0 (11.119)
A A
ð ð
ψ P dA ¼ σ wxx dA ¼ 0
A A

We will show that the three conditions of Eq. (11.119) locate the center of twist of the cross-
section, which is the point about which the cross-section rotates, and fix the value of an
arbitrary constant that appears as part of the solution.
434 Uniform and Nonuniform Torsion

z Figure 11.12 The center of twist, S, of a general cross-section as measured

z
with respect to a set of ð

y;
zÞ coordinates with origin O.
yS y
S
zS
In our original assumptions on the deformation, Eq. (11.85), the
O y distances y and z appearing in that equation are measured from the
center of twist, which we will call point S (see Fig. 11.12), but we
did not specify where that point is located. In solving for the
primary warping function, our solution ψ P is dependent on having the origin of the ðy; zÞ
coordinates be at S so we will label that desired function to be ψ PS ðy; zÞ. Now, choose a set of


coordinates ð
y;
zÞ relative to a different arbitrary origin, O, and let
yS ,
zS be the location of
S as measured in those coordinates (Fig. 11.12). From the geometry we have
y ¼
y 
yS
(11.120)
z ¼
z 
zS

Recall that we had

∂2 ψ PS ∂2 ψ PS
þ ¼0
∂y2 ∂z2
(11.121)
∂ψ PS
¼ zny  ynz
∂n

so that by a change of coordinates from ðy; zÞ toð
y;
zÞ we have

∂2 ψ PS ∂2 ψ PS
þ ¼0
∂
y2 ∂
z2
(11.122)
∂ψ PS


¼
z 
zS n
y 
y 
yS n
z
∂n

or, equivalently,

∂2 ψ PS ∂2 ψ PS
þ ¼0
∂
y2 ∂
z2
(11.123)
∂
P 
ψ S þ
y
zS 
z
yS ¼
z n
y 
y n
z
∂n

If we let ψ PO ¼ ψ PS þ y

zS 
z
yS then we see that ψ PO satisfies the same boundary value
problem as ψ PS , namely

∂2 ψ PO ∂2 ψ PO
þ ¼0
∂
y2 ∂
z2
(11.124)
∂ψ PO
¼
z n
y 
y n
z
∂n
 11.4 Nonuniform Torsion 435

This boundary value problem is called a Neumann problem [4], which has a unique solution
(to within an arbitrary constant) so that we can include such a constant,
c, and write
ψ PS ðy; zÞ ¼ ψ PO ð
y;
zÞ 
y
zS þ
z
yS þ
c (11.125)
[We should point out that by using Eq. (11.125), it is easy to show [1] that the primary shear
stresses generated by either ψ PS ðy; zÞ or by ψ PO ð
y;
zÞ are the same so at least for these stresses,
the choice of coordinates is not essential.] Now consider Eq. (11.125) and the conditions of


Eq. (11.119) which becauseðy; zÞ and ð
y;
zÞ just differ by the constants
yS ,
zS can be written
equivalently as
ð

yψ PS dA ¼ 0
ðA

zψ PS dA ¼ 0 (11.126)
ðA
ψ PS dA ¼ 0
A

We have
ð ð ð ð ð

yψ PS dA ¼
yψ PO dA 
z S

y dA þ
y
2 S

y
zdA þ
c
ydA ¼ 0
A A A A A
ð ð ð ð ð

zψ PS dA ¼
zψ PO dA 
zS
y
zdA þ
yS
z2 dA þ
c
zdA ¼ 0
A A A A A
ð ð ð ð
ψ PS dA ¼ ψ PO dA 
zS
ydA þ
yS
zdA þ
cA ¼ 0 (11.127)
A A A A


Equation (11.127) is a set of three equations for the three unknowns y
S ,
zS ;
c . We can solve
these equations directly but they can be simplified by choosing the origin O of the ð
y;
zÞ
coordinates to be at the centroid, C, of the cross-sectional area. In this case we find the set of
equations

yS I
y
z þ
zS I
z
z ¼ I P
y ψ


yS I
y
y þ
zS I
y
z ¼ I
Pzψ

c ¼ I Pψ =A (11.128)
where
ð
I Pψ ¼ ψ PC dA
Að

I P
y ψ ¼
yψ PC dA (11.129)

ðA
I
Pzψ ¼
zψ PC dA
A
436 Uniform and Nonuniform Torsion

The solution of Eq. (11.128) for the coordinates of the center of twist (as measured relative to
the cross-section centroid) is then
I
y
z I P
y ψ  I
z
z I
Pzψ

yS ¼
I
y
y I
z
z  I 2
y
z
I
y
y I P
y ψ  I
y
z I
Pzψ

zS ¼ (11.130)
I
y
y I
z
z  I 2
y
z

You might have noticed the similarity between the coordinates of the center of twist
expression, Eq. (11.130), and the location of the shear center found for thin, open cross-
section beams. As we will show later in this chapter, for thin, open cross-sections the primary
warping function is simply the negative of the principal sectorial area function. As shown in
Appendix C (see Eq. (C.20)), the location ðyP , zP Þ of the principal pole (which was shown to
also be the shear center) as measured in centroidal coordinates ðy; zÞ (where yC ¼ zC ¼ 0) is
given by
I zωC I zz  I yωC I yz
yP ¼
I yy I zz  I 2yz
(11.131)
I zωC I yz  I yωC I yy
zP ¼
I yy I zz  I 2yz

which is the same as Eq. (11.130) with the replacement of the primary warping function
calculated with a coordinate system whose origin is at the centroid by the negative of the
sectorial area function calculated with a pole at the centroid, i.e., ψ PC ! ωC . Thus, for thin


sections ðyP , zP Þ ¼
yS ,
zS . [Note: the P subscript in Eq. (11.131) indicates that ðyP , zP Þ
locates the principal pole in centroidal coordinates while the P superscript in warping
functions such as ψ PC refers to the fact that this is a primary warping function, so one should
be aware of this notational difference.]
As discussed in Chapter 10, texts typically determine the shear center location (for thin
beams) through use of either the sectorial area function or by locating where the shear forces
must be applied in the cross-section for the bending shear-stress expressions to be valid.
However, the center of twist location expression, Eq. (11.130), is generally not derived so the
fact that the two locations are identical, as shown here (at least for thin, open cross-sections),
is omitted. One reason for this omission is that many texts emphasize only the uniform
torsion of general cross-sections and, as described above, it is necessary to consider the case
of nonuniform torsion in detail to identify the center of twist location explicitly.
In the case of nonuniform torsion, it is also necessary to determine the twist of the bar as a
function of the axial coordinate, x. The primary twisting moment, M Px , was given in Eq.
(11.105), and the secondary twisting moment, M Sx , was given by Eq. (11.110). Thus, the total
internal twisting moment, M x , is
M x ðxÞ ¼ M Px ðxÞ þ M Sx ðxÞ
000
¼ GJ eff ϕ0 ðxÞ  EJ ψ ϕ ðxÞ (11.132)
 11.4 Nonuniform Torsion 437

If the bar carries a distributed applied toque/unit length, tx ðxÞ, then from equilibrium we
have

dM x
¼ tx (11.133)
dx

so we have the differential equation for the twist, ϕðxÞ, given by

00
GJ eff ϕ ðxÞ  EJ ψ ϕiv ðxÞ ¼ tx ðxÞ (11.134)

where ϕiv ¼ d 4 ϕ=dx4 . We will rewrite Eq. (11.134) as

d 4ϕ 2d ϕ
2
tx ðxÞ
4
 k 2
¼ (11.135)
dx dx EJ ψ

GJ eff
where k 2 ¼ (11.136)
EJ ψ

To solve Eq. (11.135) we need to specify boundary conditions. The boundary conditions
most frequently occurring are fixed, simple, and free conditions. At a fixed support there is
no twisting or warping allowed ðux ¼ 0Þ so that

fixed support : ϕ ¼ 0, ϕ0 ¼ 0 (11.137)

At a simple support twisting is prevented and the support is free from normal stress so that
00
simple support : ϕ ¼ 0, ϕ ¼0 (11.138)

At a free end there is no normal stress and no shear stresses so that the total internal torque,
000
M x , is zero. These requirements give, since M x ¼ EJ ψ k2 ϕ0  ϕ , the boundary conditions
00 000
free end : ϕ ¼ 0, k 2 ϕ0  ϕ ¼ 0 (11.139)

The general solution of Eq. (11.135) is the sum of a homogeneous solution and a particular
solution, which can be written as [5]

ϕ ¼ C 1 þ C 2 x þ C 3 cosh ðkxÞ þ C 4 sinh ðkxÞ

ðx
1 (11.140)
 ½k ðx  ξ Þ  sinh ðk ðx  ξ ÞÞ tx ðξ Þdξ
kGJ eff
0

where C m ðm ¼ 1; . . . ; 4Þ are constants and one end of the member is assumed to be at

x ¼ 0. From Eq. (11.132) the total internal twisting moment (torque) is
ðx
M x ðxÞ ¼ GJ eff C 2  tx ðξ Þdξ (11.141)
0
438 Uniform and Nonuniform Torsion

Figure 11.13 Torsion of a noncircular bar of length L that is

T0 restrained from twisting and warping at one end and where a torque,
T 0 , is applied at the other end.
x
As a simple example, consider a bar (Fig. 11.13) that is fixed
L
at x ¼ 0 and where an applied torque, T 0 , is applied at the
end x ¼ L. In this case tx ¼ 0, and at x ¼ L we have from
Eq. (11.141)
GJ eff C 2 ¼ T 0 (11.142a)

The other boundary condition at x ¼ L is that the normal stress is zero, so we have
00
ϕ ðLÞ ¼ k2 ½C3 cosh ðkLÞ þ C 4 sinh ðkLÞ ¼ 0 (11.142b)

At the fixed end x ¼ 0 we have

ϕð0Þ ¼ C 1 þ C 3 ¼ 0
(11.142c, d)
ϕ0 ð0Þ ¼ C 2 þ kC 4 ¼ 0

These four boundary conditions determine the constants. In MATLAB® we have the
solution
syms G J T k L
M ¼ [ 0 1 0 0; 1 0 1 0; 0 1 0 k; 0 0 cosh(k*L) sinh(k*L)];
C ¼ inv(M)*[T/(G*J); 0; 0; 0]
C =
-(T*sinh(L*k))/(G*J*k*cosh(L*k))
T/(G*J)
(T*sinh(L*k))/(G*J*k*cosh(L*k))
-T/(G*J*k)

which can be written as

T0
C 3 ¼ C 1 ¼ tanh ðkLÞ
GJ eff k
T0 T 0 (11.143)
C2 ¼ , C4 ¼
GJ eff GJ eff k

and the final solution is therefore

T0
ϕðxÞ ¼ ½kx  sinh ðkxÞ  tanh ðkLÞð1  cosh ðkxÞÞ (11.144)
GJ eff k

and the twist at x ¼ L is

 
T 0L tanh ðkLÞ
ϕðLÞ ¼ 1 (11.145)
GJ eff kL
 11.4 Nonuniform Torsion 439

1 Figure 11.14 Normalized twist versus distance for

different values of restraint of warping at x ¼ 0, where
0.8 kL ¼ ∞ corresponds to the case where there is no
restraint of warping.
kL = 8
kL = ¥
0.6
fGJeff
T0 L 0.4
kL = 2
0.2
kL = 1
0
0 0.2 0.4 0.6 0.8 1
x/L

The first term in Eq. (11.145) is the twist at x ¼ L when the twist is zero at x ¼ 0 but the bar
is free to warp. The second term, which represents the effect of restraint of warping at x ¼ 0,
reduces the twist at x ¼ L; i.e., it makes the bar appear stiffer. As kL ! ∞ the second term
goes to zero. The twist along the entire bar is shown in Fig. 11.14 for different kL values.
To complete our description of nonuniform torsion, we note that in solving for the
secondary warping function, ψ S , the solution of Eq. (11.103) is determined only to within
an arbitrary constant. We see from that equation that this secondary warping is propor-
000
tional to ϕ ðxÞ so that we could write the secondary warping displacement and normal
stress as
000
uSx ¼ ψ S ðx; y; zÞ ¼ ϕ ðxÞ~
ψ ðy; zÞ
∂uSx (11.146)
σ Sxx ¼ E ¼ Eϕiv ðxÞ~
ψ ðy; zÞ
∂x
But this secondary warping must not produce any net axial force so we must have
ð ð
σ xx dA ¼ Eϕ ðxÞ ψ
S iv
~ ðy; zÞdA ¼ 0 (11.147)
A A

which in turn implies that ψ S must satisfy

ð
ψ S dA ¼ 0 (11.148)
A

Equation (11.148) is a constraint on the secondary warping function that fixes the value of
the constant since, if we solve the boundary value problem of Eq. (11.103) and call that
_ _
solution ψ and let a constant C be given as
ð
_ 1 _
C¼ ψ dA (11.149)
A
A
440 Uniform and Nonuniform Torsion

then the secondary warping function

_
ψ S ¼ψ C (11.150)

is a solution of Eq. (11.103), which also satisfies Eq. (11.148).

11.5 General Solution of the Nonuniform Torsion Problem

We now have a complete formulation of the nonuniform torsion problem where we have the
governing equations that determine the primary and secondary warping functions and
the shear stresses, the location of the center of twist, and the angle of twist in the member.
The solution procedure has a number of steps that we want to outline here explicitly. First,
we can locate the centroid of the cross-section and determine the solution for the primary
warping function, ψ PC , using the centroid as the origin of the coordinates. This warping
function satisfies the boundary value problem

∂2 ψ PC ∂2 ψ PC
þ ¼ 0 in the cross-section
∂
y2 ∂
z2
(11.151)
∂ψ PC


z
¼
zn
y  yn on the boundary
∂n

From this solution we can then locate the center of twist explicitly and a constant,
c, as (see
Eqs. (11.128)–(11.130)):

I
y
z I P
y ψ  I
z
z I
Pzψ

y ¼
S
I
y
y I
z
z  I 2
y
z

I
y
y I P
y ψ  I
y
z I
Pzψ

zS ¼
I
y
y I
z
z  I 2
y
z

c ¼ I Pψ =A (11.152)

The primary warping function, which uses the center of twist, S, as the origin of the ðy; zÞ
coordinates, is then given by Eq. (11.125):

ψ PS ðy; zÞ ¼ ψ PC ð
y;
zÞ 
y
zS þ
z
yS þ
c (11.153)

[Note: we can use coordinates with an arbitrary origin O rather than the centroid C in the


above steps to determine
yS ,
zS and
c if we solve the set of more general equations given by
Eq. (11.127).] With this primary warping function known, we can determine the effective
polar area moment and warping constant (see Eq. (11.106) and Eq. (11.111)):
 11.5 General Solution of the Nonuniform Torsion Problem 441

ð 

 ∂ψ PS ∂ψ PS
J eff ¼ y þz
2 2
þy z dA
∂z ∂y
A
ð (11.154)

P 2
Jψ ¼ ψ S dA
A

Having these constants, we can obtain the twist in the member (see Eq. (11.135) and
Eq. (11.136)) as the solution of the equation

d 4ϕ 2d ϕ
2
tx ðxÞ
4
 k 2
¼
dx dx EJ ψ
(11.155)
GJ eff
k2 ¼
EJ ψ

under appropriate specified loads and boundary conditions. This solution also gives us the
primary and secondary internal moments as
M Px ðxÞ ¼ GJ eff ϕ0 ðxÞ
000 (11.156)
M Sx ðxÞ ¼ EJ ψ ϕ ðxÞ

and the corresponding total internal moment M x ðxÞ ¼ M Px ðxÞ þ M Sx ðxÞ. The primary shear
stresses are then also given explicitly as
 P 
∂ψ S
σ Pxy ¼ Gϕ0 ðxÞ z
∂y
 P  (11.157)
∂ψ
σ Pxz ¼ Gϕ0 ðxÞ S
þy
∂z

The secondary warping function, ψ S , can be found from the solution of Eq. (11.103):
_ _ 000
∂2 ψ ∂2 ψ Eϕ ðxÞ P
þ 2 ¼ ψ S in the cross-section A
∂y2 ∂z G (11.158)
_
∂ψ
¼ 0 on the boundary C
∂n
where
ð
_ 1 _
ψ S ¼ψ  ψ dA (11.159)
A
A

which gives the secondary shear stresses as

∂ψ S
σ Sxy ¼ G
∂y
(11.160)
∂ψ S
σ Sxz ¼G
∂z
442 Uniform and Nonuniform Torsion

As part of the nonuniform torsion problem solution, one can also examine the warping
normal stresses, σ wxx , generated, where (Eq. (11.92)):
00
σ wxx ðx; y; zÞ ¼ Eϕ ðxÞψ PS ðy; zÞ (11.161)

Needless to say, a numerical solution is in general needed to implement these steps and solve
the general nonuniform torsion problem. Recently, work by Sapountzakis and coworkers
has given a very efficient solution using the boundary element method [6], which is a nice
illustration of a case where boundary elements is an attractive alternative to the finite
element method. We will not go through the details of such numerical solutions here.

11.6 Torsion of Thin, Open Cross-Sections

If the cross-section is thin, then the torsion problem can be treated more explicitly. Consider
first the case of thin, open cross-sections such as the geometries shown in Fig. 11.15. Except
near ends or junctions, we expect that the primary stress results for a thin rectangle still apply
to these cross-sections, namely that the shear stress varies linearly across the thickness. Thus,
we have (Fig. 11.16)
σ xs ¼ 2Gϕ0 n
(11.162)
σ xn ¼ 0

where n is the distance measured from the centerline in the outward normal direction. This
result can be justified with the membrane analogy, since the shape of the membrane will still
be approximately parabolic in shape across the thickness for most of the cross-section, as
seen in the thin rectangle case (Fig. 11.10). Thus, if we ignore the deviations from this
behavior at ends or junctions, then for the cross-sections seen in Figs. 11.15a, b, c the
effective polar area moments for these sections can be calculated with the thin rectangle
results as

Figure 11.15 Thin, open cross-sections. t t1

b1

t1 C C
t2
b
b1 b2
t2 t3
C C C

b2 b3

(a) (b) (c)

 11.6 Torsion of Thin, Open Cross-Sections 443

n Figure 11.16 Primary shear stress across the cross-section in a thin, open section where s is
directed along the tangent to the center line and n is directed normal to the center line.
s

1
ðaÞ J eff ¼ bt3
3
1 1
ðbÞ J eff ¼ b1 t31 þ b2 t32 (11.163)
3 3
1 1 1
ðcÞ J eff ¼ b1 t31 þ b2 t32 þ b3 t33
3 3 3
When the thickness is slowly varying function along the length, b, of a cross-section, then we
can again use a modified result for the thin rectangle given by
ðb
1
J eff ¼ ½tðsÞ3 ds (11.164)
3
0

which reduces to the thin rectangle result when the thickness is a constant. If we let tmax be
the largest thickness in a cross-section, then the maximum primary shear stress based on
Eq. (11.162) would be
Ttmax
ðσ xn Þmax  τ max ¼ (11.165)
J eff

However, at reentrant corners such as the points C shown in Fig. 11.15, the changing
geometry produces stress concentrations that can produce larger stress than given by Eq.
(11.165). Figure 11.17 shows a typical case where (assuming that t2 > t1 so that
τ ¼ Tt2 =J max is the larger of the two stresses in the sides adjacent to the corner) the stress
distribution near the corner is no longer linear and has a maximum value given by
Tt2
τ max ¼ K (11.166)
J eff

where K is the stress concentration factor. The stress concentration factor is a function of the
radius, r, of the fillet in the corner (Fig. 11.17). Such factors are often determined experi-
mentally or with detailed numerical calculations. Similar concentrations can also appear
locally near notches or keyways in the cross-sections.
Unlike a thin rectangular section where the warping is zero along the centerline of the
rectangle, a general thin, open section can experience significant warping. You can experi-
ence this tendency of thin, open sections to warp by simply twisting a thin rolled-up piece of
paper and see what happens. To obtain an expression for this centerline warping, consider
the tangential displacement along the centerline, us , as seen in Fig. 11.18. Relating this
displacement to the components along the y- and z-axes and using the Saint-Venant
assumptions for those y- and z-displacements (i.e., the displacements due to a rotation about
the center of twist, S, as seen in Eq. (11.7)), we have
444 Uniform and Nonuniform Torsion

t2 Figure 11.17 Stress concentration at a reentrant corner where it is assumed t2 > t1 .

Tt2
W max K us ¼ uy sin α þ uz cos α
J eff
t1 ¼ zϕðxÞ sin α þ yϕðxÞ cos α
C r (11.167)
¼ ϕðxÞren
¼ r⊥ ϕðxÞ

where r ¼ yey þ zez , en ¼ cos αey þ sin αez , and r⊥ ¼ ren is the perpen-
dicular distance from the center of twist, S, to the line of action of us . The primary
engineering shear strain, γPxs , at the centerline is then given by
∂us ∂ux
γPxs ¼ þ
∂x ∂s (11.168)
∂ux
¼ r⊥ ϕ 0 þ
∂s
where s is the distance measured along the centerline (in the s-direction shown in Fig. 11.18)
and the corresponding primary shear stress, σ Pxs , is
∂ux
σ Pxs ¼ GγPxs ¼ Gr⊥ ϕ0 þ G (11.169)
∂s
However, just as in a rectangular section, we expect this centerline stress and strain to be
zero, so that by setting σ Pxs ¼ 0 and integrating Eq. (11.169), we find the primary warping
displacement, ux , along the centerline to be given as
0s 1
ð
ux ¼ ϕ0 @ r⊥ ds þ ω0 A
(11.170)
0
¼ ϕ0 ωðsÞ

where ωðsÞ is a sectorial area function, a function we have previously seen in the bending of
thin sections, and which is examined in Appendix C. The ω0 is a constant of integration,
which depends on the choice of the starting point for the integration (s ¼ 0).This sectorial
area function depends on the choice of the constant of integration (starting point), as well
as the location of the origin, S, of the coordinates (called the pole of the function) which
here is at the center of twist. If we compare Eq. (11.170) with the original Saint-Venant
assumption on the deformation (Eq. (11.33)), we see that the primary warping function,
ψ P , along the centerline is just the negative of the sectorial area function, i.e., ψ P ¼ ω.
Thus, instead of having to solve a differential equation to find the primary warping
function, for thin, open sections we can obtain the centerline value directly from the
geometry by determining the sectorial area function. As in the case of general cross-
sections, to determine explicitly the location of the center of twist, S (which is also the
pole of the sectorial area function in Eq. (11.170)), as well as the constant of integration,
one must consider the secondary warping and stress present for the case of nonuniform
 11.6 Torsion of Thin, Open Cross-Sections 445

us , s xs Figure 11.18 Geometry for a thin section, showing the displacement and stress in the
n
s cross-section along the s-direction (tangential to the centerline). The displacement
en also has a component in the n-direction, as shown.
a un

torsion of the thin member. This secondary warping is generated because

r
z of the primary warping normal stress produced, i.e., in this case
r^ 00
σ wxx ðx; y; zÞ ¼ Eϕ ðxÞωðy; zÞ (11.171)

ez Requiring that this warping stress produces no net axial force or bending
moments, we have
S ey y ð ð
00
σ xx dA ¼ Eϕ ωdA ¼ 0
w

A A
ð ð
00
yσ wxx dA ¼ Eϕ yωdA ¼ 0 (11.172)
A A
ð ð
00
zσ wxx dA ¼ Eϕ zωdA ¼ 0
A A

which are the conditions that require the sectorial area function to be a principal sectorial area
function ω ¼ ωP ðs; ω0 Þ, where the principal pole P is also the shear center S and the constant
ω0 defines a principal origin (see Appendix C). Thus, the primary warping is determined
explicitly by this principal sectorial area function. The changing warping normal stress
produces a secondary shear stress, σ Sxt along the centerline in the thin section. We will assume
this secondary shear stress is uniformly distributed over the thickness, t, of the section so that it
can be expressed in terms of a secondary shear flow, qS ðsÞ, as qS ðsÞ ¼ tðsÞσ Sxs ðsÞ. By examining
equilibrium of a small element of a thin section (Fig. 11.19) we obtain
 w   S 
∂σ xx ∂q
dx tds þ ds dx ¼ 0
∂x ∂s
(11.173)
∂qS ∂σ wxx
! ¼ t
∂s ∂x
Writing the warping normal stress in terms of the principal sectorial area function and
integrating we have
ðs
000
q ðsÞ ¼ Eϕ ðxÞ ωP ðs; ω0 ÞtðsÞds
S
(11.174)
0

where the constant of integration is zero since we have assumed s ¼ 0 is taken to be at the
end of a thin, open section where qS ð0Þ ¼ 0. Recall that for the secondary warping function,
ψ S , we had (Eq. (11.160))
446 Uniform and Nonuniform Torsion

V xxw Figure 11.19 A small element of a thin section showing the relationship
between the secondary shear flow and the warping normal stress.
ds
dx t qS ∂ψ S
wq S σ Sxy ¼ G
q 
S
ds ∂y
ws
∂ψ S
σ Sxz ¼G
wV w ∂z
V xxw  xx
dx
wx Since (see Fig. 11.18)
∂ψ S qS dy
σ Sxy ¼ G ¼ σ Sxs sin α ¼
∂y t ds
(11.175)
∂ψ S q S
dz
σ Sxz ¼G ¼ σ Sxs cos α ¼
∂z t ds
we find
∂ψ S ∂ψ S dy ∂ψ S dz qS
¼ þ ¼ (11.176)
∂s ∂y ds ∂z ds Gt
so, using Eq. (11.174), the secondary warping can be obtained by integration as
2s 3
ð S
1 q ðuÞ
ψ S ðx; sÞ ¼ 4 du þ K 5 (11.177)
G tðuÞ
0

where the constant of integration, K, can be obtained, as before, from the condition (see Eq.
(11.148))
ð
ψ S dA ¼ 0 (11.178)
A

In summary, for thin, open sections we have both the primary shear stresses and the
secondary shear flow given explicitly as
σ Pxs ¼ 2Gϕ0 n
σ Pxn ¼ 0
ðs (11.179)
000
q ¼ Eϕ
S
ωP tds
0

We also have the primary and secondary warping functions given by

ψ P ¼ ωP
2s 3
ð S
1 q (11.180)
ψS ¼ 4 ds þ K 5
G t
0
 11.6 Torsion of Thin, Open Cross-Sections 447

The effective polar area moment, J eff , is the thin section result obtained from the thin
rectangular section, as discussed previously. This is related to the primary torque, M Px , by
M Px ¼ GJ eff ϕ0 . For the secondary torque, M Sx , computed about the center of twist (shear
center) S, we have
ðl ðl
M Sx ¼ r⊥ q ds ¼ qS dωP
S
(11.181)
0 0

where l is the length of the open section and we can take r⊥ ds ¼ dωP in terms of the
principal sectorial area function since r⊥ is measured from the shear center. If we integrate
Eq. (11.181) by parts, we find
ðl
∂qS
M Sx ¼ qS ωP jl0  ωP ds
∂s
0
(11.182)
ðl
∂q S
¼  ωP ds
∂s
0

since qS ð0Þ ¼ qS ðl Þ ¼ 0 at the free ends of an open section. Placing the expression for the
secondary shear flow, Eq. (11.174), into Eq. (11.182) gives
ð
000
M Sx ¼ Eϕ ðxÞ ω2P dA
A (11.183)
000
¼ Eϕ ðxÞJ ω

(where dA ¼ tds is the element area), which is the same expression for the warping constant,
J ψ , obtained previously for a general cross-section but where here the secondary warping
function (see Eq. (11.111)) is replaced by the principal sectorial area function for the thin
open section. We can again relate the primary and secondary torques to a distributed torque/
unit length, tx , acting on the bar, and the twist for the thin open section is thus governed by
the same differential equation (see Eq. (11.134)) as before:
00
GJ eff ϕ ðxÞ  EJ ω ϕiv ðxÞ ¼ tx ðxÞ (11.184)

and appropriate boundary conditions.

Example 11.2 Nonuniform Torsion of an I-Beam

We will use the I-beam cross-section of Fig. 11.20a to give an explicit example of the use of
these thin, open section formulas. Many texts use an I-beam to discuss nonuniform torsion
and restraint of warping since the warping and stress distributions are relatively simple to
derive and understand. We will assume this beam has a length L and is fixed at x ¼ 0, and a
448 Uniform and Nonuniform Torsion

Figure 11.20 (a) z t

f
Cross-section of an
z
I-beam. (b) An
example of restrained
warping and T0
nonuniform torsion
h
y
where the I-beam is tw x
loaded in torsion at
one end and rigidly L
fixed at the other end. b

(a) (b)

has an applied torque, T 0 , acting at its other end at x ¼ L, as shown in Fig. 11.20b. The
effective polar area moment for this section is given by the thin section result as

2bt3f þ ht3w
J eff ¼ (11.185)
3
The principal sectorial area function for this cross-section has its pole, P, located at the
centroid (which is also the shear center) and the initial point of integration, I, can also be
taken at P. The principal sectorial area function is then given by (shown as the shaded areas
in Fig. 11.21a)
h
ωP ¼ s (11.186)
2
which is valid for both flanges when the distance, s, is taken as shown in Fig. 11.21. The
sectorial area function is zero in the web. The largest and smallest values for this function are
bh=4. The warping normal stress induced by the restraint of warping is therefore given by
00 00 h
σ wxx ¼ Eϕ ðxÞωP ¼ Eϕ ðxÞ s (11.187)
2
00
and has the linear distributions in the flanges shown in Fig. 11.21b. [As we will see, ϕ is
positive so the stress distribution is opposite in sign to the primary warping function.] In the
beam cross-section, the primary shear stress distribution is a linear distribution across the
thickness
M Px
σ Pxs ¼ 2Gϕ0 n ¼ 2 n (11.188)
J eff

where M Px is the primary part of the torsional moment and the distance n is normal to the
centerline. The primary shear-stress distribution along the cross-section sides is shown in
Fig. 11.22a. The secondary shear-stress distribution is described by the shear flow, qS , where
in each flange
 11.6 Torsion of Thin, Open Cross-Sections 449

bh/4 wP Figure 11.21 (a) The principal sectoral area function

s for the I-beam. (b) The resulting warping stress
induced when the beam is in nonuniform torsion.

s xxw
–bh/4 s =0
w
xx
wP = 0 P, I
bh/4

s s xxw
–bh/4 wP

(a) (b)

s Figure 11.22 (a) The primary torsional moment and

qS associated primary shear stress distribution. (b) The
secondary torsional moment and associated secondary
shear-flow distribution.

s xsP qS = 0
n
M xS
P
M x q S
s

(a) (b)

2s 3
ð
qS ¼ Eϕ ðxÞ4 ωP tds þ C 5
000

0 (11.189)
 2

000 htf s
¼ Eϕ ðxÞ þC
4

and where we have included a constant of integration, C, because s ¼ 0 is not at a free end.
Since we must have qS ¼ 0 at s ¼ b=2 we have C ¼ htf b2 =16 and
000
!
Eϕ ðxÞhtf 2 b2
q ¼
S
s 
4 4
! (11.190)
M Sx htf b2
¼ s 2
Jω 4 4
450 Uniform and Nonuniform Torsion

which has the parabolic shear flow distribution in the flanges shown in Fig. 11.22b. The
warping constant is

ð ð
b=2
h2 2
Jω ¼ ω2P dA ¼ 2tf s ds
4
A b=2 (11.191)
2 3
tf h b
¼
24

Since the boundary conditions on the I-beam are the same as considered in Section 11.4, we
have from Eq. (11.144)

T0
ϕðxÞ ¼ ½kx  sinh ðkxÞ  tanh ðkLÞð1  cosh ðkxÞÞ
GJ eff k
 
T0  sinh ðkxÞ cosh ðkLÞ þ cosh ðkxÞ sinh ðkLÞ
¼ kx  tanh ðkLÞ þ
GJ eff k cosh ðkLÞ
 
T0 sinh ðkðx  LÞÞ
¼ kx  tanh ðkLÞ 
GJ eff k cosh ðkLÞ
(11.192)

which gives
 
0 T0 cosh ðk ðx  LÞÞ
ϕ ðxÞ ¼ 1
GJ eff cosh ðkLÞ
 
00 T 0 k sinh ðk ðx  LÞÞ
ϕ ðxÞ ¼ (11.193)
GJ eff cosh ðkLÞ
 
000 T 0 k2 cosh ðkðx  LÞÞ
ϕ ðxÞ ¼
GJ eff cosh ðkLÞ

00 000
Note that throughout the beam we have ϕ0 > 0, ϕ > 0, and ϕ < 0. The primary and
secondary torques are then given by

 
0 cosh ðk ðx  LÞÞ
M Px ¼ GJ eff ϕ ¼ T 0 1
cosh ðkLÞ
(11.194)
cosh ðk ðx  LÞÞ
000
M Sx ¼ EJ ω ϕ ¼ T 0
cosh ðkLÞ
 11.6 Torsion of Thin, Open Cross-Sections 451

The primary and secondary warping functions in each flange are given by Eq. (11.180):
hs
ψ P ¼ ωP ¼
2
2 3
ðs S
16 q 7
ψS ¼ 4 ds þ K 1 5
G t (11.195)
b=2
2 3
000 ðs ! 000
" ! #
2
Eϕ 6h b 7 Eϕ h s3 b2 s b3
¼ 4 s 
2
ds þ K 2 5 ¼   þ K2
G 4 4 G 4 3 4 12
b=2

00
where K 1 and K 2 are “constants” (i.e., they not functions of s) and where Eϕ K 2 ¼ K 1 .
Requiring that there be no net axial force from this secondary warping, we have
ð ð
b=2 000
" #
Eϕ tf b4 h
ψ S dA ¼ 2 ψ S tds ¼ 2  þ K 2b ¼ 0 (11.196)
G 48
A b=2

so that K 2 ¼ b3 h=48 and we have in each flange (see Fig. 11.23 for both the primary and
secondary warping distributions):
000
!
3 2
Eϕ h s b s
ψS ¼  (11.197)
4G 3 4

which attains its maximum magnitude at x ¼ 0, s ¼ b=2, where from Eq. (11.193),
000
 S Eϕ ð0Þb3 h T 0 b3 h
ψ  ¼ ¼ (11.198)
max 48G 48GJ ω

y P = −wP yS Figure 11.23 (a) The primary centerline

z z warping function. (b) The secondary
warping function of the I-beam according
to the theory for thin, open sections.

x x
y
y P = −wP y
yS

(a) (b)
452 Uniform and Nonuniform Torsion

while the maximum magnitude of the primary warping function is at s ¼ b=2 where
 P bh
ψ  ¼ (11.199)
max 4
The maximum primary stress in the cross-section (ignoring any stress concentrations) occurs
at x ¼ L and at the outer edge of the cross-section given as
 
 P  P M Px t T 0 t 1
τ   σ  ¼ ¼ 1  (11.200)
max xs max
J eff J eff cosh ðkLÞ

while the maximum secondary shear stress occurs at x ¼ 0 and s ¼ 0 in the flanges and is
 S
 S q  M Sx b2 h T 0 b2 h
τ  ¼   ¼ ¼ (11.201)
max t J ω 16 16J ω
max

We want to compare these explicit results for this I-beam with results obtained by numeric-
ally solving the nonuniform torsion theory of Section 11.4 (and summarized in Section 11.5),
using the boundary element method [6]. The specific parameters are for a cantilever steel
I-beam of length L ¼ 5.0 m, where Young’s modulus is E ¼ 2.1 108 kN/m2, and Poisson’s
ratio is ν ¼ 0.3. The dimensions of the cross-section are h ¼ 14=15 m, b ¼ 2=3 m, tf ¼ tw ¼
1=15 m, and the applied torque at x ¼ L is T 0 ¼ 1=3 kN-m. To compare the thin section
results with the more general theory, we will examine six quantities: (1) the effective polar
area moment of the I-beam, J eff , (2) the warping constant, J ω , (3) the maximum value in the
cross-section of the principal warping function, (4) the maximum value in the cross-section
of the secondary warping function, (5) the maximum value of the primary shear stress in the
beam, and (6) the maximum value of the secondary shear stress in the beam.
For this beam we will first examine the split of the total torque, T 0 , into the primary and
secondary torques, Eq. (11.194). To obtain the distributions of these torques, we need the
value of the parameter
sffiffiffiffiffiffiffiffiffiffiffi
GJ eff
kL ¼ L (11.202)
EJ ω

so we will first find the underlying parameters in Eq. (11.202). From the values given for this
problem we have the results:
E
G¼ ¼ 8:0769 108 kN=m2
2ð1 þ νÞ
2bt3f þ ht3w
J eff ¼ ¼ 2:2387 104 m4
3
(11.203)
tf h2 b3
Jω ¼ ¼ 7:1696 104 m6
24
sffiffiffiffiffiffiffiffiffiffiffi
GJ eff
kL ¼ L ¼ 1:7327
EJ ω
 11.6 Torsion of Thin, Open Cross-Sections 453

1 Figure 11.24 The primary twisting moment (solid line) and

S secondary moment (dashed line) distributions in the cantilever
0.8
M x
I-beam.
T0
M x 0.6
T0
0.4
M xP
0.2
T0

0
0 0.2 0.4 0.6 0.8 1
x/L

With these values we then can obtain and plot the primary and secondary torsional
moments:
 
cosh ðkLðx=L  1ÞÞ
M Px ¼ T 0 1 
cosh ðkLÞ
(11.204)
cosh ðkLðx=L  1ÞÞ
Mx ¼ T0
S
cosh ðkLÞ

Figure 11.24 shows that at the free end, the secondary twisting moment is still about
35 percent of the total torque so that the effects of warping restraint extend over the full
length of the member. Next, we will compute the maximum values of the primary and
secondary warping functions and the primary and secondary shear stresses:
 P bh
ψ  ¼ ¼ 0:1556 m2
max 4
 S T 0 b3 h
ψ  ¼ ¼ 3:3163 108 m
max 48GJ ω
  (11.205)
 P T 0t 1
τ  ¼ 1 ¼ 65:2284 kN=m2
max J eff cosh ðkLÞ
 S T 0 b2 h
τ  ¼ ¼ 12:0536 kN=m2
max 16J ω
In Table 11.1 these results from the theory for thin, open sections are compared to the results
obtained in [6] for the more general theory for arbitrary cross-sections as calculated numer-
ically by the boundary element method (BEM). The agreement in general is good for both
the warping functions and the stresses. Although the thin cross-section results predict the
warping at the centerline and the boundary element method solves for the warping along
the edges, the results from the boundary element method also show that that there is
little variation across the thickness [6]. Note that while the secondary warping function is
the actual warping displacement, uSx , the primary warping displacement, uPx , is given by
uPx ¼ ϕ0 ðxÞψ P ¼ ϕ0 ðxÞωP , which has its maximum value at x ¼ L given by the thin cross-
section theory as
454 Uniform and Nonuniform Torsion

Table 11.1 Comparison of the results of the torsion theory for thin sections with the more
general theory for arbitrary cross-sections as calculated with the boundary element
method (BEM)

Thin cross-section theory BEM

4

J eff m 2.24 E–4 2.47 E–4


J ω m6 7.17 E–4 7.13 E–4
 P
2
ψ  m 1.56 E–1 1.64 E–1
max
 S
 ψ  ð mÞ 3.32 E–8 3.21 E–8
max
 P
τ  ðkPaÞ 65.23 71.29
max
 S
τ  ðkPaÞ 12.05 10.85
max

 P  P
u  0  
x max ¼ ϕ ðLÞ ψ max
 
T0 1  P
¼ 1 ψ  (11.206)
GJ eff cosh ðkLÞ max

¼ 1:8844 106 m
This is small but it is still significantly larger than the secondary warping displacement.
However, the maximum secondary shear stress is not negligibly small in comparison to the
maximum primary shear stress.

11.7 Torsion of Thin, Closed Cross-Sections

When a thin cross-section is closed, the analysis is more complex than the open section case.
Figure 11.25 shows a closed cross-section with two holes. The section is said to be composed of
two “cells.” Since the shear stresses must be parallel to the sides of a given section, the stresses
must act as shown at three positions labeled a–a, b–b, and c–c. We can use the Prandtl stress
function approach to obtain the distribution of these stresses. As discussed previously, the
Prandtl function must be zero on the outer boundary and have constant values K 1 , K 2 at the
inner holes. Since the cross-section is thin, we can approximate the Prandtl function as just a
linear function across the thickness of the sections, as shown in Fig. 11.26a. This also makes
sense from the membrane analogy because, to first order, we expect that the membrane
displacement across the thickness will also be linear. If we let the thickness of the cross-
section at a–a, b–b, and c–c be ta , tb , and tc , respectively, then since the shear stress is just
proportional to the slope of the Prandtl stress function, the stresses at these three locations are
     
∂Φ K1 ∂Φ K1  K2 ∂Φ K2
τa ¼  ¼ , τb ¼  ¼ , τc ¼  ¼ (11.207)
∂n a ta ∂n b tb ∂n c tc
 11.7 Torsion of Thin, Closed Cross-Sections 455

z Figure 11.25 Torsion of a closed cross-section with two “cells,”

showing the shear stresses in the section at three locations and the
a F=0 value of the Prandtl stress function at the outer and inner
ta boundaries of the cells.
tb
a b b
c
F = K1 F = K2
y
tc
c

) K1 ) K1 Figure 11.26 (a) Approximating the

Prandtl stress function (and corresponding
) K2 ) K2
membrane analogy displacement) as linear
functions across the thickness of the
) 0 ) 0 sections. (b) The shear stresses generated in
the different sections.
a a b b c c
ta tb tc

(a)
n
s )
w) K1 0
Wa 
wn ta
s ) K2
) K1 s
n
w) K2
Wc 
wn tc
n
w) K1  K 2
Wb 
wn tb

(b)

(see Eq. (11.46)), as shown in Fig. 11.26b. These shear stresses are uniform across the
thickness in each section but if the thickness changes, these stresses can vary as we go along
the center line of each cell. However, if as in the bending of thin sections, we introduce the
shear flow q ¼ τt, Eq. (11.207) shows that these shear flows depend only on the constants
K 1 , K 2 , so the shear flows in a closed section are constant both across the thickness and
along the centerline. For the two-cell closed cross-section, therefore, we can describe the
shear flows as two constant shear flows qi ði ¼ 1; 2Þ, where qi is the shear flow for the ith cell
(Fig. 11.27). Note that while the shear flow in each cell is a constant, in sections that share
the sides of two cells, such as the vertical section in Fig. 11.27, the total shear flow is a
different constant from the remainder of the cells.
We have also used shear flows to describe the bending of unsymmetrical sections in
Chapter 10. As in the bending case, we can show that these torsional shear flows must be
456 Uniform and Nonuniform Torsion

J Figure 11.27 The constant shear flows ðq1 , q2 Þ in the two cells and the
q1 − q2 = K1 − K 2 total shear flows flowing in the various sections, where at a junction, such
q1 = K1 as J, the sum of the shear flows into or out of the junction must be zero.

q1
q2
conserved at a junction in the sense that the sum of the shear
q2 = K 2 flows flowing into or out of a junction must be zero. This can
easily be seen to be the case, for example, for the shear flows in
Fig. 11.27 at junction J. We will take a positive shear flow to be
flowing in a counterclockwise sense around any given cell, as shown in Fig. 11.27. Such flows
will then generate internal torques which are positive along the positive x-axis of the
member, as we have already assumed in our previous discussions of torsion. To obtain the
torque generated by the shear flows in a closed cross-section, consider the ith cell, as shown
in Fig. 11.28. The torsion moment, M xi , generated about a point P by the shear flow, qi , in
the ith cell is
Þ
M xi ¼ qi r⊥ ds
0 1
ð ð
¼ qi @ r⊥ ds þ r⊥ dsA (11.208)
ABC CDA
¼ qi ðωABC þ ωCDA Þ

In Eq. (11.208) ωABC and ωCDA are the sectorial areas obtained by integrating along the
curved side ABC and the straight side CDA of the cell, respectively. However, the sectorial
area is positive when the area is swept out in a counterclockwise direction and negative when
sept out in a clockwise direction (see Appendix C) so that ωABC is positive and ωCDA is
negative, where ωABC ¼ 2AABC , ωCDA ¼ 2ACDA ; i.e., these sectorial areas are just twice the
areas AABC , ACDA being swept out as shown in Fig. 11.28b. It then follows that
ωABC þ ωCDA ¼ 2AABC  2ACDA
(11.209)
¼ Ωi

where Ωi ¼ 2Ai is just twice the area contained with the centerline of the ith cell, as shown in
Fig. 11.28b. Thus, the torque being carried by the ith cell is just
M xi ¼ qi Ωi (11.210)

The total primary (Saint-Venant) torque for a closed section with N cells, M Px , is therefore
given by
X
N X
N
M Px ¼ M xi ¼ qi Ωi (11.211)
i¼1 i¼1

For a single cell if the primary torque is known then the primary shear flow, qP , in the cell is
simply given as
 11.7 Torsion of Thin, Closed Cross-Sections 457

1 1 Figure 11.28 (a) The geometry of the ith cell

dAs = r^ ds = w ds
2 2 B for calculating the torque produced by the
shear flow. (b) The areas AABC , ACDA swept
qi out from point P and the area, Ai , within the
r^ centerline of the cell.
r

C D A
M xi
P
(a)
B

Ai
AABC C D A
C A
ACDA
Ai = AABC − ACDA

P P
(b)

M Px
qP ¼ (11.212)
Ω
where Ω ¼ 2A, i.e., it is twice the area, A, contained within the centerline of the single cell.
For multiple cells we need to obtain additional relations to solve for the shear flows. These
relations are obtained by requiring that the warping displacement, ux , must be single valued.
Recall, from Eq. (11.169) we found that the primary shear stress on the centerline of a
section can be expressed as
∂ux
σ Pxs ¼ GγPxs ¼ Gr⊥ ϕ0 þ G (11.213)
∂s
For a thin, open section, we set this shear stress to be zero but for a closed section we have
qt
σ Pxs ¼ (11.214)
t
where qt is the shear flow. [Note: qt here is total shear flow anywhere in a cell so that it may be
qi in some parts of the ith cell but combinations of qi and other shear flows qj ðj ¼6 iÞ in other
parts of the ith cell. Thus, we have placed the t subscript on this shear flow to remind you
that it is not just qi in the ith cell.]. Thus, we have
∂ux q
¼ t  ϕ0 r ⊥ (11.215)
∂s Gt
Integrating Eq. (11.215) around each cell and requiring the warping displacement to be
single valued, we find
458 Uniform and Nonuniform Torsion

þ þ þ
∂ux 1 qt
ds ¼ 0 ¼ ds  ϕ0 r⊥ ds (11.216)
∂s G t
ith cell ith cell ith cell

where ϕ0 is the same for each cell and the second integral on the right-hand side of Eq. (11.216)
is just twice the area contained within the ith cell, so we have finally the conditions
þ
1 qt
ds ¼ ϕ0 ði ¼ 1; . . . ; N Þ (11.217)
GΩi t
ith cell

For a single cell Eq. (11.217) and Eq. (11.212) simply give
þ
M Px ds
ϕ0 ¼ (11.218)
GΩ2 t

which shows that for the single-cell closed section we again can write the relationship
between the primary torque and rate of twist as

M Px ¼ GJ eff ϕ0 (11.219)

where the effective polar area moment is

Ω2
J eff ¼ Þ (11.220)
ds
t

For multiple cells we cannot write down a closed form expression for J eff but we can write
Eq. (11.211) in the form of Eq. (11.219) where

X
N
J eff ¼ f i Ωi (11.221)
i¼1

and where f i ¼ qi =Gϕ0 . The normalized total shear flow, ^q t ¼ qt =Gϕ0 , is just a linear function
of the f i so that the equations
þ
^q t
ds ¼ Ωi ði ¼ 1; . . . ; N Þ (11.222)
t
ith cell

are just N linear equations that can be solved for the N unknown f i values and those results
placed into Eq. (11.221) to obtain J eff .
For a general closed section with multiple cells, we can also obtain the primary warping
displacement by returning to Eq. (11.215) and integrating. We find
 11.7 Torsion of Thin, Closed Cross-Sections 459

ðs ðs
0 1 qt
ux ðsÞ ¼ ϕ r⊥ ds þ ds
G t
sP sP
2s 3
ð ðs (11.223)
^
q
¼ ϕ0 4 r⊥ ds  ds5
t
t
sP sP
0 _
¼ ϕ ωP ðs; sP Þ
_
in terms of a function, ω P ðs; sP Þ, given by
ðs ðs
_ ^q t
ω P ðs; sP Þ ¼ r⊥ ds  ds (11.224)
t
sP sP

which acts as a modified sectorial area function for the closed section. For a single closed cell
Eq. (11.224) has the more explicit form
ðs ðs
_ Ω ds
ω P ðs; sP Þ ¼ r⊥ ds  Þ (11.225)
ds t
sP s
t P
Like the ordinary sectorial area function, the modified sectorial area function depends on the
pole P (which is the origin from which the r⊥ distance is measured) and the constant sP
(which is a origin location where this modified function is zero). The modified function is a
function of r⊥ and ^q t (which is obtained from Eq. (11.221)) so, like the ordinary sectorial
area function, it depends only on the geometry. For the nonuniform torsion of open sections
we found that in order to ensure that the warping does not generate any net axial force or
bending moments, the sectorial area function appearing in the primary warping displace-
_
ment had to be a principal sectorial area function. Here, the modified function ω P ðs; sP Þ
must satisfy the same conditions so that it is a principal modified sectorial area function
where the pole P and the origin sP can be found from the three conditions (see Eq. (11.172)):
ð
_
ω dA ¼ 0
A
ð
_
y ω dA ¼ 0 (11.226)
A
ð
_
z ω dA ¼ 0
A

_
where ω is a modified principal sectorial area function calculated with respect to an arbitrary
pole and origin. In Appendix C we derive a wide range of properties of the ordinary sectorial
area function. It is easy to show that the modified sectorial area function defined in
460 Uniform and Nonuniform Torsion

Eq. (11.223) also satisfies the following properties found in Appendix C for the ordinary
sectorial area function.

(1) Relationship for changing the pole and origin of the modified sectorial area function (see
Eq. (C.12)):
_ _ _
ωA ðs; s0 Þ ¼ ωB ðs; s1 Þ  ω B ðs0 , s1 Þ
(11.227)
þ ðyB  yA ÞðzðsÞ  z0 Þ  ðzB  zA ÞðyðsÞ  y0 Þ
_
(2) Relationship between the principal modified sectorial area function, ωP ðs; sP Þ, and a
_
modified function, ωB ðs; s1 Þ, computed with an arbitrary pole B and origin s1 (see Eq.
(C.43)):
_ _ Q_ωB
ω P ðs; sP Þ ¼ ω B ðs; s1 Þ  þ ðzP  zB Þy  ðyP  yB Þz (11.228)
A
where
ð
_
Qω_B ¼ ωB dA (11.229)
A

(3) Location ðyP , zP Þ of the principal pole P (center of twist or shear center) (see Eq. (C.21))
_
as calculated from a modified sectorial area function, ω B ðs; s1 Þ, with an arbitrary pole
B and origin s1 :
I zω_B I zz  I yω_B I yz
ey ¼ yP  yB ¼
I yy I zz  I 2yz
(11.230)
I zω_B I yz  I yω_B I yy
ez ¼ zP  zB ¼
I yy I zz  I 2yz

where, as shown in Appendix C, y and z are centroidal axes. In Eq. (11.230) the distances


ey , ez are the distances of the principal pole P from the pole B in the y- and z-directions.
(4) The warping constant, J ω_P , calculated with the principal modified sectorial area func-
_
tion, as calculated from the same constant using a modified sectorial area function, ω B ,
with arbitrary pole and origin (see Eq. (C.46)):

J ω_  Qω_B 2
J ω_ ¼ B
 ðyP  yB Þ2 I yy  ðzP  zB Þ2 I zz þ 2ðyP  yB ÞðzP  zB ÞI yz (11.231)
P A
Of course, the warping constant can also be calculated directly by integration using its
definition
ð
_
J ω_P ¼ ω 2P dA (11.232)
A
 11.7 Torsion of Thin, Closed Cross-Sections 461

In the case of nonuniform torsion there will again be a warping normal stress generated
where
00 _
σ wxx ðs; xÞ ¼ Eϕ ðxÞω P ðs; sP Þ (11.233)

As in the thin, open section case, equilibrium requires that this normal stress produce a
secondary shear flow, qS ðsÞ, where

∂qS ∂σ w 000 _
¼ t xx ¼ tEϕ ðxÞω P ðs; sP Þ (11.234)
∂s ∂x

As with the case of uniform torsion for a closed cross-section with multiple cells, it is difficult
to obtain explicit results for this secondary shear flow and the secondary warping. Later, we
will outline the steps needed for the multiple cell case. Here, however, we will obtain explicit
results for the case of a single-cell closed section. If we integrate Eq. (11.234) for a single cell,
we obtain
ðs
_ 000
q ðs; xÞ ¼ q0 þ Eϕ ðxÞ ω P dA
S

0
000
¼ q0 þ Eϕ ðxÞS ω_P ðsÞ (11.235)

where q0 is a constant and we have defined S ω_P as

ðs
_
S ω_ ðsÞ ¼ ω P dA (11.236)
P
0

000 Ðs
[Note: in the case of thin, open sections we had qS ðs; xÞ ¼ Eϕ ðxÞ ωP dA, which we could
000
also have written as qS ðs; xÞ ¼ Eϕ ðxÞS ωP ðsÞ.] To obtain the constant
0
q0 we use the fact that
S
the secondary warping displacement, ux , must be single valued so that the following closed
integral around the cell is zero:
þ þ S
∂ux ∂ψ
ds ¼ ds ¼ 0 (11.237)
∂s ∂s

But from Eq. (11.176) this is equivalent to

þ
1 qS
ds ¼ 0 (11.238)
G t

Instead of writing Eq. (11.238) in terms of q0 we will write it instead in terms of the
secondary torsional moment, M Sx . We have
Þ
M Sx ¼ r⊥ qS ds
Þ 000
Þ (11.239)
¼ q0 r⊥ ds þ Eϕ S ω_P r⊥ ds
462 Uniform and Nonuniform Torsion

which we can solve for q0 as

000 þ
M S Eϕ
q0 ¼ x  S ω_P r⊥ ds (11.240)
Ω Ω

which, when placed back into Eq. (11.235) gives

000 þ
M Sx Eϕ 000
q ðs; xÞ ¼
S
 S ω_P r⊥ ds þ Eϕ S ω_P ðsÞ
Ω " Ω #
þ (11.241)
000 M^S 1
¼ Eϕ x  S ω_P r⊥ ds þ S ω_P ðsÞ
Ω Ω

where M ^ S ¼ M S =Eϕ is a normalized moment. Placing Eq. (11.241) into Eq. (11.238) then
000
x x
gives (after canceling common terms)
Þ ds
^ S þ ds
M
þ þ _
S ωP
x
 t S ω_P r⊥ ds þ ds ¼ 0 (11.242)
Ω t Ω t

From Eq. (11.225) we have

Ω ds
dωP ðs; sP Þ ¼ r⊥ ds  Þ
_
(11.243)
ds t
t
so that
þ þ þ
_ Ω S ω_P
S ω_P r⊥ ds ¼ S ω_P dωP þÞ ds (11.244)
ds t
t
Placing Eq. (11.244) into Eq. (11.242) then gives
Þ ds
^ S þ ds
M
þ
_
x
 t S ω_ dω P ¼ 0
Ω t Ω P (11.245)
þ
^ ¼ S _ dω _
! M S
x ω P
P

However, by integrating by parts we have

þ  þ
_ _ sf
_ 2
S ωp dωP ¼ S ω P ω P   ω P dA
_ _
0
¼ J ω P _ (11.246)

where s ¼ sf is at the same point in the cross-section as s ¼ 0 and the first term vanishes since
_
both S _ωp and ωP are single-valued functions. We have also defined the modified warping
constant, J _ωp , as
 11.7 Torsion of Thin, Closed Cross-Sections 463

þ

_ 2
J ω_P ¼ ω P dA (11.247)

(written here as a line integral rather than an area integral since dA ¼ tds). Thus, from
Eq. (11.245) we have the secondary torsional moment given by
000
M Sx ¼ EJ ω_ ϕ ðxÞ (11.248)
P

and the secondary shear flow becomes

" #
000 ^S 1þ
M
q ðs; xÞ ¼ Eϕ ðxÞ
S x
 S ω_P r⊥ ds þ S ω_P ðsÞ
Ω Ω
2 0 1 3
þ
6 J_ 1B Ω S ω_P C 7 (11.249)
¼ Eϕ ðxÞ4 ωP  @J ω_P þ Þ
000
dsA þ S ω_P ðsÞ5
Ω Ω ds t
t
000


¼ Eϕ ðxÞS ω_P ðsÞ

where
þ
S ω_P

_ ðsÞ ¼ S _ ðsÞ  1
S ds (11.250)
ωP ωP Þ ds t
t
The total twisting moment is
M x ¼ M Px þ M Sx
000 (11.251)
¼ GJ eff ϕ0  EJ ω_P ϕ

so that, from equilibrium of moments, we again have the differential equation for the twist
given in the same form as found before for both general sections and thin open sections,
namely
00
GJ eff ϕ ðxÞ  EJ ω_P ϕiv ðxÞ ¼ tx ðxÞ (11.252)

where tx is the distributed torque/unit length along the member.

Finally, the secondary warping displacement for the single closed cell is obtained by
integrating
∂ψ S 000

_ ðsÞ
¼ qS ¼ Eϕ S ωP (11.253)
∂s
giving
ðs
ψ ðsÞ ¼ Eϕ
S 000

_ ds þ K
S (11.254)
ωP
0
464 Uniform and Nonuniform Torsion

where the constant K can again be found from the condition

þ
ψ S dA ¼ 0 (11.255)

For a multicell closed cross-section with N cells, the steps are very similar but we cannot
write them down as explicit expressions. First, we can integrate Eq. (11.235) and define a set
of shear flows flowing around each cell as
ðs
_
000
qSi ðs; xÞ ¼ qi þ Eϕ ðxÞ ωP dA
(11.256)
0
000
¼ qi þ Eϕ ðxÞS ω_P ðsÞ ði ¼ 1; . . . ; N Þ

where qi are constants at the starting integration points in each cell (Fig. 11.29). We can also
define the twisting moments for each cell as
þ
M xi ¼ qSi r⊥ ds ði ¼ 1; . . . ; N Þ (11.257)
ith cell

and, as in the single-cell case, write shear-flow constants qi in terms of the normalized
moments, M ^ xi ¼ M xi =Eϕ000 . The total secondary moment, M S , is then
x

X
N
M Sx ¼ Eϕ
000
^ xi
M (11.258)
i¼1

which we can put in the form used previously

000
M Sx ¼ Eϕ J eff
_
ω
(11.259)

where
X
N
J eff
_ ¼ 
^ xi
M (11.260)
ω
i¼1

To obtain the normalized moments we use the fact that the warping displacement must be
single valued so that now we must have (see Eq. (11.238))
þ S
1 qt
ds ¼ 0 ði ¼ 1; 2; . . . ; N Þ (11.261)
G t
ith cell
s ^ xi values. As in the primary
which are N linear equations for the N M
q1 shear-flow case we have called the total secondary shear flow in each

q2
Figure 11.29 A two-cell cross-section showing the starting shear flow values
s ðq , q Þ at the s ¼ 0 positions in the cells.
1 2
 11.7 Torsion of Thin, Closed Cross-Sections 465

cell qSt since certain walls may contain several qSi ðs; xÞ shear-flow functions (Fig. 11.29).
Solving Eq. (11.261) for the M^ xi values then gives us the individual secondary shear flows,
qSi ðs; xÞ and the effective warping constant, J eff
_ . The secondary warping function can then
ω
be obtained by integration of the total secondary shear flow:
ðs
ψ ðs; xÞ ¼ qSt ds þ K
S
(11.262)
0

where again the constant K can be found from the condition

þ
ψ S dA ¼ 0 (11.263)

Obviously, this is a rather complex process where we have not delineated many of the
explicit details but instead listed the basic steps involved, following our previous approach in
describing the primary shear flows.

Example 11.3 Primary Warping of a Thin Rectangular Closed Cell

Consider a thin rectangular closed cross-section as shown in Fig. 11.30. We will examine the
primary warping of this section, using the expression for the principal modified sectorial area
function given by Eq. (11.225). For a rectangular section with constant thickness, that
equation can be written along each straight side as
ðs
_ ab _
ω P ðsÞ ¼ r⊥ ds  s þ ω P ð0Þ (11.264)
ða þ bÞ
0

Because of the symmetry of the rectangular section, the shear center is located at the center,
O, of the rectangle so we can take the principal pole at O. If we take the starting point of the

s1 Figure 11.30 A thin rectangular cross-section of constant

thickness, t. The lengths (a, b) are measured from the
A
centerlines of the cross-sections and the distances ðs1 , s2 , s3 , s4 Þ
s2 are also measured along the centerlines.

t
b
O
s4

s3

a
466 Uniform and Nonuniform Torsion

integrations to be at point A in Fig. 11.30 and let the value of the modified sectorial area
function be ω0 at that point then in the four parts of the section we have, using Eq. (11.264),

_ bðb  aÞ
ω1 ¼ s1 þ ω0
2ða þ bÞ
_ aða  bÞ abðb  aÞ
ω2 ¼ s2 þ ω0 þ
2ða þ bÞ 2ða þ bÞ
_ bðb  aÞ
ω3 ¼ s3 þ ω0
2ða þ bÞ
_ aða  bÞ abðb  aÞ
ω4 ¼ s4 þ ω0 þ (11.265)
2ða þ bÞ 2ða þ bÞ

To obtain the value of ω0 we must satisfy (see Eq. (11.226))

ð
_
I ¼ ω P dA ¼ 0 (11.266)

We can do the integration in MATLAB®, giving:

syms a b w0 w1 w2 w3 w4 x
w1 ¼ b*(b-a)*x/(2*(a+b)) +w0;
w2 ¼ a*(a -b)*x/(2*(a+b)) + w0 + a*b*(b -a)/(2*(a+b));
w3 ¼ b*(b-a)*x/(2*(a+b)) +w0;
w4 ¼ a*(a-b)*x/(2*(a+b))+ w0 +a*b*(b-a)/(2*(a+b));
I ¼ int(w1,x, 0,a) +int(w2,x,0,b) +int(w3,x,0,a). . .
+int(w4,x,0,b);
simplify(I)
ans ¼ 2*a*w0 + 2*b*w0 + (a*b^2)/2 - (a^2*b)/2

Setting this integral I found in MATLAB® equal to zero, we find

abða  bÞ
ω0 ¼ (11.267)
4ða þ bÞ

We can write the principal modified sectorial area function succinctly in terms of this
constant as
_
ω 1 ¼ 2ω0 s1 =a þ ω0
_
ω 2 ¼ 2ω0 s2 =b  ω0
_
ω 3 ¼ 2ω0 s3 =a þ ω0
_
ω 4 ¼ 2ω0 s4 =b  ω0 (11.268)

which is shown in Fig. 11.31. For a ¼ b the constant ω0 vanishes so that a square section is
free of warping. Having the principal modified sectorial area function, one can calculate the
warping constant as
 11.8 Problems 467

Figure 11.31 The principal modified sectorial area function

−Z0 Z0 Z for a rectangular section.
0

−Z0

Z0

Z0 −Z0
−Z0

ta2 b2 ðb  aÞ2
J ω_P ¼ (11.269)
24ða þ bÞ

Note that it is also easy to evaluate the area moments approximately when the rectangular
section is thin as

tb2 ðb þ 3aÞ
I yy ¼
6
(11.270)
ta2 ða þ 3bÞ
I zz ¼
6

11.8 PROBLEMS

P11.1 For a solid bar whose cross-section is an equilateral triangle of height, h, (see Fig.
P11.1), the normalized Prandtl stress can be obtained exactly as:
   
pffiffiffi pffiffiffi

¼ 1 y  3 z  2h y þ 3 z  2h y þ h
Φ
2h 3 3 3
z
Note that each of the quantities in parentheses in the
2h/3 above expression vanish on an edge of the triangle so that

¼ 0 on the boundary, as required. For this shape one
Φ
2h / 3 can show by integration that
y ðð pffiffiffi 4

3h
J eff ¼ 2 ΦdA ¼
45

h Figure P11.1 A cross-section in the form of an equilateral triangle.

468 Uniform and Nonuniform Torsion

t Figure P11.2 (a) A thin, closed circular cross-section. (b) A thin,

t open circular cross-section.

R r (a) For this cross-section, determine and plot the

shear stress along the y-axis.
(b) Find the maximum shear stress, τ max , in terms of
the torque, T, carried by the section, and h.
(a) (b)
Where is τ max found?
P11.2 Consider a thin, closed circular cross-section of
radius R and thickness t, (see Fig. P11.2a), and a thin open section circular cross-
section of radius r and thickness t, as shown in Fig. P11.2b. In these sections the
effective polar area moment, J eff , determines the stiffness of the cross-section with
respect to torsion. If the effective polar area moments of these sections are equal,
determine the radius, r, of the open section in terms of R and t for the closed section.
What value of r do you obtain in this case if t ¼ 1 mm, R ¼ 10 mm?
P11.3 The thin extruded cross-section shown in Fig. P11.3 is subjected to a uniform torque,
T. Determine the maximum allowable value for T if the maximum permissible shear
stress is 75 MPa. Neglect stress concentrations. All dimensions are in millimeters and
measured with respect to the cross-section centerlines.
P11.4 The thin aluminum (G ¼ 26.7 GPa) C-section torsional member shown in Fig. P11.4
is subjected to a uniform torque T ¼ 1356 N-m.
(a) Determine the angle of twist per unit length.
100 (b) Determine the maximum shear stress and the location
of this maximum stress in the cross-section. Neglect
stress concentrations. All dimensions are in milli-
5 meters and measured with respect to the cross-section-
150 10 centerlines.
15

Figure P11.3 A thin, open cross-section. All dimensions are given in

90 millimeters as measured along the centerlines.

12
200 10
12

Figure P11.4 A thin aluminum C-section. All dimensions are given in

100 millimeters as measured along the centerlines.
 11.8 Problems 469

L Figure P11.5 Three thin cross-

sections in the form of a square, an
equilateral triangle, and a circle. All
d d
L sections have the same uniform
t r t
t thickness, t, and the same length
along their center line perimeters
(i.e., the lengths L, d, r are related).
d

P11.5 Compare the shear stress and twist per unit length of a thin square tube and a thin
tube in the shape of an equilateral triangle with the corresponding shear stress and
twist per unit length, respectively, for a thin circular tube (Fig. P11.5). The three
sections have equal wall thicknesses and equal perimeters. The dimensions are meas-
ured with respect to the cross-section centerlines.
P11.6 Consider a single closed cell that has “fins” as seen in Fig. P11.6a. In the fins we might
expect the stress distributions to be those we have found for thin, open sections while in
the closed cell we might expect the stress to be nearly uniform across the thickness. We
will still keep these assumptions on the behavior of the stresses but will take the
presence of the fins into account in the torque–twist relationship by writing the effective
polar area moment, J eff , as the sum of the closed cell and open cell values, i.e.,

X1ð
bi
Ω2
J eff ¼Þ þ ti ðsÞ3 ds
ds i
3
0
t
where the summation is over all the fins present and we have allowed the fins to have
variable thicknesses (see Eq. (11.164)). Generally, these fin contributions are small.

yields uniform stress

a

a a
a
t
(b)

(a)
(c)

Figure P11.6 (a) A single closed cell section with fins. (b) The shape of a membrane across the thickness of
the closed cell at a–a when using the membrane analogy, showing the difference of the membrane shape
from the assumed straight-line behavior, an assumption which yields a uniform shear stress across the
thickness, t. (c) A plot of a membrane having the shape of the difference seen in (b).
470 Uniform and Nonuniform Torsion

However, there is also a small correction to the closed cell value that we should
include that is of the same order. To see where this correction comes from, consider
the membrane analogy we discussed previously. Figure P11.6b shows the shape of the
membrane across the thickness, t, at some location a–a in the closed cell, as seen in
Fig. P11.6a. We had assumed to first order the membrane has the straight-line shape
seen in Fig. P11.6b. It was this assumption that led to the shear stress being taken as
uniform across the thickness. However, there will be a difference between the actual
membrane shape and the assumed straight-line behavior and that difference is shown
by itself in Fig. P11.6c. The membrane shape of this difference looks much like that
seen in a thin, open section so we will add a similar open section contribution for the
closed cell and write
þ X1ð
bi
Ω2 1
J eff ¼Þ þ 3
tðsÞ ds þ ti ðsÞ3 ds
ds 3 i
3
0
t
where the middle term is a line integral around the closed cell.
Now, let us apply this result to the cross-section seen in Fig. P11.7a where a
torque T ¼ 6000 N-mm is applied to the section and the shear modulus is G ¼ 80
GPa.
(a) Compute the angle of twist/unit length and the maximum shearing stress (neglect-
ing stress concentrations) in the closed section and the fin.
(b) Now, consider the same geometry but where the cross-section is split at A to
produce the open section seen in Fig. P11.7b. Again, determine the twist/unit
length and the maximum shearing stress for this open section. How does the
slitting change these values from those calculated in part (a)?
P11.7 A member with the two-cell cross-section shown in Fig. P11.8a carries a uniform
torque, T. The shear modulus is G ¼ 80 GPa.
(a) If the maximum allowable shear stress is 10 MPa, determine the maximum
allowable torque for this closed cross-section. Ignore any stress concentrations.

Figure P11.7 (a) A box beam with an 2 mm 2 mm

extension. (b) The same cross-section,
which is split at point A. All
dimensions are outside measurements.
A
2 mm 19 mm
1 mm

1 mm

20 mm 15 mm
(a) (b)
 11.8 Problems 471

1 mm thickness all around Figure P11.8 (a) A two-cell closed

cross-section. (b) The same cross-
section when it is slit at point A. All
A dimensions are measured from the
center lines.
20 mm T

16 mm 4 mm
(a) (b)

(b) If the cross-section is split at point A, producing the cross-section shown in Fig.
P11.8b, determine the maximum allowable torque under the same conditions as in
(a). How much do your answers change if you include the small correction to J eff
for the closed part as described in problem P11.6?
P11.8 Two cover plates (shown as dashed lines in Fig. P11.9a) are welded onto the outside
of an I-beam. Both the cover plates and the I-beam have the same uniform thickness,
t. What is the ratio of the torsional stiffness of the I-beam with the plates to the beam
without the plates?
P11.9 Given a thin rectangular section of length b and thickness, t (Fig. P11.9b) what should
the radius, R, be (in terms of b and t) of two thin, closed circular tubes, also of
thickness t, that we weld onto the rectangular section if we wish to double the
torsional stiffness?
P11.10 The two-cell thin-walled box beam shown in Fig. P11.10 is subjected to a pure
torque, T. If the maximum allowable shear stress in the beam is 40 MPa (neglecting
stress concentrations), determine the applied torque, T , and the rate of twist, ϕ0 (G ¼
70 GPa). All wall thicknesses are 0.5 mm and the dimensions shown in Fig. P11.10
are all measured from the centerlines of the cross-section.

b = 10t Figure P11.9 (a) An I-beam with and

2R without cover plates. The thickness of all
t components is t. (b) A thin rectangular
section of length b augmented by two thin
t h = 20t
circular tubes of radius R. The thickness of
all components is t.
b

(a) (b)
472 Uniform and Nonuniform Torsion

Figure P11.10 A two-cell

box beam. 10
T

20

20 10

P11.11 The two-cell box beam shown in Fig. P11.11 carries a pure torque, T. If the
maximum shearing stress in the section is limited to 5 ksi, determine the correspond-
ing allowable torque and twist/unit length. Neglect
any effects of stress
 concen-
trations. Take the box beam to be made of steel G ¼ 12 10 psi and let the
6

thickness t ¼ 0.5 in. and the dimension a ¼ 6 in. All the dimensions are measured
from the centerlines of the beam walls.

Figure P11.11 A two cell box beam where a ¼ 6 in., a a

t ¼ 0.5 in.

t t
t/2
2t T a

t t t

P11.12 Consider a member with the thin, split (open) circular cross-section shown in Fig.
P11.12a. We will assume that this member is subjected to a torque, T 0 , at one end
and is fixed at the other end, as shown (for a general section) in Fig. 11.13, so that it
is a state of nonuniform torsion. Since nonuniform torsion involves quite a few
parameters and functions, we want to describe those functions and parameters in
general terms for this geometry.
(a) In Chapter 10 we showed how to use the shear flow in bending to determine the
shear center. We also have shown in Appendix C how we can use the sectorial
area function to obtain explicit expressions for the shear center location (see Eq.
(C.21)) or instead calculate the location by using the properties of the principal
sectorial area function (see Eq. (C.32)). Here, we can take advantage of the
relationship obtained in Appendix C for how the sectorial area function changes
 11.8 Problems 473

z z Figure P11.12 (a) A thin, open

circular section that is split at
t I point I. (b) A thin, open
t
semicircular section. (c) A thin,
R closed semicircular section.
B T
T I B R
C, A C A y
y
e e
z
(a) (b)

t t
R
T
B y
C A, I
e

(c)

when we change the pole location for the sectorial function (Eq. (C.14)) which
we rewrite here as
ωB ðs; s0 Þ ¼ ωA ðs; s0 Þ þ ðyA  yB ÞðzðsÞ  z0 Þ  ðzA  zB ÞðyðsÞ  y0 Þ (P11.1)
to determine the principal sectorial function for the cross-section of Fig.
P11.12a. Recall that the ðy; zÞ coordinates in this relationship are measured from
the centroid.
First, determine the sectorial area function with the pole located at point
A (see Fig. P11.12a) and with an initial starting point (origin) at point I (this is
a very simple function). Calculate the sectorial area function, in terms of the
distance, e, from Eq. (P11.1) when the pole is located at point B instead. Then
change the origin, adding an unknown constant to the sectorial area function.
Finally, use the criteria of Eq. (C.32) to obtain the principal sectorial area
function, ωP , and the value of e that defines the shear center location (and the
unknown constant). Note that the primary warping function for the cross-
section, ψ P , is then ψ P ¼ ωP .
(b) Determine the sectorial second moment (warping constant), J ωP , the effective
polar area moment, J eff , and the constant, kL (see Eq. (11.136)), where L is the
length of the member. Note that expressions for the twist and derivatives of the
00 000
twist ϕðxÞ, ϕ0 ðxÞ, ϕ ðxÞ, ϕ ðxÞ are then given by Eq. (11.192) and Eq. (11.193).
(c) Determine the secondary warping function, ψ S , and expressions for the primary
and secondary shear stresses in the cross-section as well as the warping normal
stress, σ wxx .
474 Uniform and Nonuniform Torsion

P11.13 Repeat the steps of problem P11.12 for the cross-section shown in Fig. P11.12b.
P11.14 Repeat part (a) of problem P11.12 for the closed cross-section of Fig. P11.12c.
Specifically, determine (1) the principal shear flow, (2) the location of the shear
center, and (3) the principal modified sectorial area function, which gives us the
primary warping function of the cross-section. Equation (P11.1) can also be used for
the modified sectorial area function. Note that in a closed cross-section the warping
is normally much smaller than for open sections and the shear center is usually near
the centroid.

References
1. I. S. Sokolnikoff, Mathematical Theory of Elasticity (New York, NY: McGraw-Hill, 1956)
2. W. B. Bickford, Advanced Mechanics of Materials (Menlo Park, CA: Addison- Wesley, 1998)
3. R. D. Cook and W. C. Young, Advanced Mechanics of Materials, 2nd edn (Upper Saddle River,
NJ: Prentice-Hall, 1999)
4. C. E. Pearson, Theoretical Elasticity (Cambridge, MA: Harvard University Press, 1959)
5. W.D. Pilkey and L. Kitis, Notes on the Linear Analysis of Thin-walled Beams, https://vdocuments
.mx/bmbk.html (1996)
6. E. J. Sapounzakis and V. G. Mokos, “Warping shear stresses in non-uniform torsion by BEM,”
Comp. Mech., 30 (2003), 131–142
12 Combined Deformations

The bending of unsymmetrical beams was considered in Chapter 10 where we showed that,
for thin sections, we could obtain explicit expressions for both the flexure stresses and the
shear flows (shear stresses). In Chapter 11 similar explicit expressions were found for the
shear stresses or shear flows for the torsion of thin, open and closed cross-sections. In this
chapter we will examine cases where a thin member is in a combination of bending and
torsion. We will also consider the case where axial loads acting on a thin section can induce
torsional deformations. A new quantity called a bimoment will be shown to be a resultant of
the axial stresses in that case. Although we will deal only with combined bending and
torsional and combined axial extension/compression and torsion in this chapter, we should
note that parts of our discussion are elements of a theory called Vlasov theory [1], [2], which
examines the more general case where there can be any combinations of bending, torsion,
and axial extension/compression of thin, open sections.

12.1 Bending and Torsion of Thin, Open Cross-Sections

Consider a thin, open section where the shear forces act on a general point in the cross-


section, as defined by the distances d y , d z from a fixed point, O, as shown in Fig. 12.1.
These shear forces will produce both bending and torsion of the section. If the shear forces
acted through the shear center, then bending only would be produced so we can imagine
decomposing the case of Fig. 12.1 into pure bending and pure torsional problems as shown


in Fig. 12.2 where ey , ez are measured from point O to the shear center and the torque T is
given by


T ¼ V z d y  ey  V y ðd z  ez Þ (12.1)

We know from our discussions in Chapter 10 and Chapter 11 how to solve these bending
and torsional problems separately, so by superposition we can obtain the total shear stresses
in the general case. This process is best illustrated with an example so we will use the thin,
open section considered in Example 10.4 in Chapter 10.

Example 12.1 Combined Bending and Torsion of a Thin, Open Cross-Section

In Example 10.4 considered in Chapter 10, there was a vertical 1000 lb shear force acting
through the shear center, producing bending only of the section. In this case we want to

475
476 Combined Deformations

z Figure 12.1 Shear forces acting on an unsymmetrical cross-section.

obtain the maximum shear stress and its location in the cross-section for
Vy the problem where the shear force acts through the center point O on the
dz y web (Fig. 12.3) so that both bending and torsion are produced. For the
O torsional shear stress, we will only consider the primary shear stress for
Vz
dy uniform torsion. The shear center was found to be 0.828 in. to the left of
point O in the web so that the pure bending and torsional problems are as
shown in Fig. 12.3. The effective polar area moment, J eff , and the
maximum primary shear stress, ðτ max ÞT for the pure torsion case (neglect-
ing any stress concentrations) are
1 1
J eff ¼ 2  ð8Þð0:1Þ3 þ ð10Þð0:1Þ3
3 3
¼ 0:008666 in:4
Ttmax ð828:2Þð0:1Þ
ðτ max ÞT ¼ ¼
J eff 0:008666
¼ 9556 psi

Vy Vy
dz ez
O T
= O + O
dy Vz ey Vz

general case bending only torsion only

Figure 12.2 A general loading case decomposed into pure bending and pure torsion problems.

Figure 12.3 The beam of

Example 10.4 in Chapter 10 0.828 in.
where the shear force acts 10 in. =
through the center point O in the O O + O T = 828.2 in.-lb
web, producing the bending and 1000 lb
1000 lb
torsional loadings shown. The
beam has a constant thickness
t ¼ 0.1 in.
3 in. 5 in.
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 477

τ = 9556 −1090
0.828 in. = 8466 psi
T = 828.2 in.-lb
qmax = 109 lb / in.
O
O (τ max ) B = qmax / t O
(τ max )T = 9556 psi
1000 lb = 1090 psi

τ max = 9556 + 1090

= 10646 psi
bending torsion

(a) (b)

Figure 12.4 (a) The maximum bending and torsional shear stresses. (b) The stresses at point O.

This maximum shear stress acts at the outer boundaries of the cross-section and is the same
in all sections since the thickness is a constant 0.1 in. Thus, we need only to find the location
where this maximum torsional stress adds to the maximum bending stress. In the bending
case considered in Chapter 10, we found the maximum shear flow to be 109 lb/in. flowing
upwards at point O so that the maximum bending shear stress (which is uniform across the
cross-section), ðτ max ÞB , is given by ðτ max ÞB ¼ qmax =t ¼ 1090 psi. Both the maximum bending
and torsional shear stresses at O are shown in Fig. 12.4a. Since the primary torsional shear
stress varies linearly across the thickness, the sum of the maximum bending and torsional
stresses are as shown in Fig. 12.4b where the largest total shear stress τ max ¼ 10 646 psi will
occur on the right-hand side of the section at point O.

12.2 Bending and Torsion of Thin, Closed Cross-Sections

12.2.1 Single-Cell Cross-Sections
Consider first the case of a single-cell closed section as shown in Fig. 12.5, where shear


force components V y , V z are located with respect to an arbitrary point P. The torsion
that the shear forces generate will induce a constant shear flow, qT , as shown. However,
in the bending part the shear flow,qB , is known only up to a constant since we have, from
Eq. (10.41),



V z I yz  V y I yy Qz þ V y I yz  V z I zz Qy
qB ðsÞ ¼ qB ð0Þ þ (12.2)
D
and, unlike open sections, we cannot find an end where qB ¼ 0. Thus, the total shear flow will
be determined only to within a constant value (that we must determine) and that constant is
partly due to bending and partly due to torsion. If we sum moments about P we have
478 Combined Deformations

Figure 12.5 Bending and Vy Vy

torsion of a thin, single-cell
closed section decomposed dz ez
= + T
into its bending and r^ P P
P qB ey qT
torsional components. q dy
Vz Vz
bending torsion

X þ
M P ¼ V z d y  V y d z ¼ qr⊥ ds (12.3)
C

and the total shear flow in the section can be written as

qðsÞ ¼ qc þ V y f ðsÞ þ V z gðsÞ (12.4)

where qc is a constant shear flow, part of which comes from bending and part of which
comes from the primary shear stress due to uniform torsion. (We will not include the
secondary torsional shear flows in this discussion.) From Eq. (12.2) we have the remaining
bending shear-flow terms
I yz Qy  I yy Qz
f ðsÞ ¼
I yy I zz  I 2yz
(12.5)
I yz Qz  I zz Qy
g ð sÞ ¼
I yy I zz  I 2yz

Placing Eq. (12.4) and Eq. (12.5) into Eq. (12.3) gives
þ
 
V z d y  V y d z ¼ Ωqc þ V y f ðsÞ þ V z gðsÞ r⊥ ds (12.6)
C

where Ω is twice the area contained within the centerline of the cross-section. Also recall that
we have a relationship between the shear flow and the twist/unit length induced by that shear
flow given as (Eq. (11.217)):
þ
0 1 q
ϕ ¼ ds (12.7)
GΩ t
C

We will examine two types of problems.

(1) The shear forces and their positions are known.

In this case qc can be found directly from Eq. (12.6) since the left-hand side of that
equation is known explicitly as are all the other terms on the right-hand side of that equation
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 479

except for qc . Once qc is known, then the total shear flow q is known completely so we can
use it in Eq. (12.7) to obtain the twist/unit length ϕ0 .

(2) The shear forces are known and assumed to act through the shear center, whose position


ey , ez relative to point P is unknown.

In this case there is no twisting of the section so that this is not a combined deformation
problem and we have ϕ0 ¼ 0. We can also set V y ¼ 0 and solve Eq. (12.7) for an unknown
constant shear flow qc1 . Then Eq. (12.6) gives the location of the shear center d y ¼ ey since
þ
V z ey ¼ qc1 Ω þ V z gðsÞr⊥ ds (12.8)

We can repeat this process by setting ϕ0 ¼ 0 and V z ¼ 0 and solving Eq. (12.7) for a new
constant shear-flow value qc2 . Then Eq. (12.6) gives the location of the shear center d z ¼ ez
since
þ
V y ez ¼ qc2 Ω þ V y f ðsÞr⊥ ds (12.9)

We can then combine these two cases and obtain the total shear flow and corresponding
shear stress when both V y and V z are present.

12.2.2 Multiple-Cell Cross-Sections

Figure 12.6a shows a cross-section with two cells. We know that in the case of torsion only


there are two constant shear flows ðq1 ÞT , ðq2 ÞT in each cell (not shown). In the case of
bending only we can imagine breaking the cells at two locations and writing the bending

q = ( q1 ) B Figure 12.6 (a) A two-cell closed section under

bending and torsion. (b) The two shear-flow
Vy constants appearing in the bending part of the
P dz problem. (c, d) The decomposition of the original
Vz problem into two parts where (c) constant shear
dy flows (due to both bending and torsion) are
q = ( q2 ) B flowing in the cells, and (d) an equivalent open
(a) (b) section where the shear flows are zero at
=

q=0 imaginary cuts in the cross-section.

q1 q2 +

q=0
(c) (d)
480 Combined Deformations


shear stresses in terms of two unknown “starting” shear flows ðq1 ÞB , ðq2 ÞB , as seen in
Fig. 12.6b. Thus, we can combine the two sets of unknown constant shear flows and consider
the original problem of Fig. 12.6a as the sum of a problem where there are constant shear
flows ðq1 , q2 Þ in the cross-section, as shown in Fig. 12.6c ( where qi ¼ ðqi ÞT þ ðqi ÞB ði ¼ 1; 2Þ)
and a problem where there are bending shear flows in an open-like cross-section where the
starting shear flows at the imaginary breaks in the cross-section are both zero, as shown in
Fig. 12.6d. This means that if we sum moments about point P we will find
2 þ 
X 
V z d y  V y d z ¼ Ω1 q1 þ Ω2 q2 þ V y f ðsÞ þ V z gðsÞ r⊥ ds (12.10)
m¼1
Cm

where Ωi ði ¼ 1; 2Þ are twice the areas contained inside the centerlines of the two cells and the
two line integrals are taken around the two closed cells. In addition, we have, as shown in
Chapter 11, the angle of twist/unit length related to the shear flows in each cell as
þ
1 q
ϕ0 ¼ ds
GΩ1 t
C1
þ (12.11)
0 1 q
ϕ ¼ ds
GΩ2 t
C2

As in the single cell problem we can examine two types of problems.

(1) The shear forces and their positions are known.

In this case we can solve the two equations in Eq. (12.11) for ðq1 , q2 Þ in terms of the
unknown ϕ0 and place these values into Eq. (12.10), which is then an equation that can be
solved for ϕ0 . Once ϕ0 is obtained in this manner then the previously obtained solutions of
Eq. (12.11) yield the values of ðq1 , q2 Þ.

(2) The shear forces are known and assumed to act through the shear center, whose position
is unknown.
There is no torsion in this case so this is not a combined deformation problem and we can set
ϕ0 ¼ 0, V y ¼ 0 and solve the two equations in Eq. (12.11) for the unknown ðq1 , q2 Þ values.
Then Eq. (12.10) becomes
X2 þ
V z ey ¼ Ω1 q1 þ Ω2 q2 þ ½V z gðsÞr⊥ ds (12.12)
m¼1
Cm

which can be solved for the shear center location ey . We can repeat this process by setting
ϕ0 ¼ 0 and V z ¼ 0 , solving for new values ðq1 , q2 Þ, and then obtain the shear center location
ez from
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 481

2 þ 
X 
V y ez ¼ Ω1 q1 þ Ω2 q2 þ V y f ðsÞ r⊥ ds (12.13)
m¼1
Cm

In this chapter we have emphasized obtaining the bending and primary torsional shear
stresses in the combined bending and torsion of open and closed sections. There is also
warping present in these combined problems as well as secondary torsional shear stresses for
nonuniform torsion cases that can be obtained with the methods outlined in Chapter 11. We
will now consider in some detail an example of a single-cell closed section to illustrate the
steps outlined previously in an explicit case.

Example 12.2 Combined Bending and Torsion of a Single-Cell Closed

Cross-Section
In this problem we will calculate the shear flows and determine the shear center for the box
beam shown in Fig. 12.7. The beam has a uniform thickness of t ¼ 0.1 in., and all the
dimensions shown in Fig. 12.7 are measured in inches. A 1000 lb vertical force acts along the
centerline of the left-hand vertical section, producing bending and torsion. We have labeled
the four sides of the section as sides (1)–(4). The location of the centroid has been calculated
as Y C ¼ 5:21 in:, Z C ¼ 3:5 in: and the area moments and mixed area moment of the four
sides have been calculated as

I ðyy1Þ ¼ 1:433 I ðzz1Þ ¼ 18:44 I ðyz1Þ ¼ 4:074

I ðyy2Þ ¼ 14:41 I ðzz2Þ ¼ 16:41 I ðyz2Þ ¼ 3:419

(12.14)
I ðyy3Þ ¼ 6:975 I ðzz3Þ ¼ 24:43 I ðyz3Þ ¼ 4:689

I ðyy4Þ ¼ 14:70 I ðzz4Þ ¼ 15:15 I ðyz1Þ ¼ 3:318

Z Figure 12.7 Bending and torsion of a closed single-

cell section of uniform thickness t ¼ 0.1 in. All the
1000 lb dimensions listed are in inches.
z

(3) (2)
9 5.21

C y (1) 4
3.5 (4) Y

12
482 Combined Deformations

13 Figure 12.8 Geometry of the inclined side as seen in

5
centroidal coordinates.
12
ds2 s2
5.21
y all measured in units of in.4. For sides (1), (3), and (4)
z these values follow directly from the parallel axis the-
0.5
orem (see Appendix A) but calculations for the inclined
C
side (2) are a bit more complex. We could use a combin-
6.79 ation of the parallel axis theorem and the change of the
area moment under rotations but it is easier to simply do
the integrations directly, using the geometry of the inclined side relative to the cross-section
centroid as shown in Fig. 12.8. It follows that the ðy; zÞ locations of a small element ds2 are
given by

5
z ¼ 0:5 þ s2
13
(12.15)
12
y ¼ 6:79  s2
13

The area moment, I ðyy2Þ is then just an integral that can easily be calculated:
1ð3

I ðyy2Þ ¼ z2 dA
0

1ð3 (12.16)
¼ ð0:5 þ 5s2 =13Þ2 ð0:1Þds2
0

¼ 14:41 in:4

and the others can be calculated similarly, yielding

1ð3

I ðzz2Þ ¼ y2 dA ¼ 16:41 in:4

0
(12.17)
1ð3

I ðyz2Þ ¼ yzdA ¼ 3:419 in:4

0

From all these values for the individual parts we then have for the entire cross-section
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 483

X
4
I yy ¼ I ðyymÞ ¼ 37:52 in:4
m¼1

X
4
I zz ¼ I ðzzmÞ ¼ 74:43 in:4 (12.18)
m¼1

X
4
I yz ¼ I ðyzmÞ ¼ 15:50 in:4
m¼1

In our problem we have V z ¼ 1000 lb, V y ¼ 0 so the equation for the bending shear flows
(see Eq. (10.41)) is
 
 I zz Qy  I yz Qz V z
Δq ¼
I yy I zz  I 2yz
 
74:43Qy  15:5Qz ð1000Þ (12.19)
¼
ð37:52Þð74:43Þ  ð15:5Þ2
! Δq ¼ 29:16Qy  6:073 Qz

where Δq is the difference between the shear flow at a given location and its constant
value at some starting location. However, we can also use Eq. (12.19) for the total shear
flow since the torsional part simply adds an additional constant value, which we could
combine with the constant in the bending expression, as discussed previously. Thus, we
will now use Eq. (12.19) for the total shear flow in each part of the section. Consider part
(1) as shown in Fig. 12.9. The quantities (Qy, Qz) are calculated for an area As as shown
in Fig. 12.9 as
ð
Qy ¼ zdA ¼ zAs
As
ð (12.20)
z Qz ¼ ydA ¼ yAs
y = 6.79 As

where ðy; zÞ are the distances from the centroid of

the entire cross-section to the centroid of the area
As as shown in Fig, 12.9 for part (1). Thus, we have
y for part (1)
z
3.5 s1
q1
As Figure 12.9 The geometry for part (1) of the cross-section.
484 Combined Deformations

z Figure 12.10 The geometry for part (2) of the cross-section.

13
5 s2 / 2
s2 / 2
12 q1 + 1 Ð
Qðy1Þ ¼ zdA ¼ ð3:5  s1 =2Þð0:1Þðs1 Þ
y z 0.5 ¼ 0:35s1 þ 0:05s21
y (12.21)
Ð
Qðz1Þ ¼ ydA ¼ ð6:79Þð0:1Þðs1 Þ
¼ 0:679s1
6.79 which gives the shear flow in part (1) as
 
qð1Þ ðs1 Þ ¼ q1  29:16 0:35s1 þ 0:05s21  6:073½0:679s1 
(12.22)
¼ q1 þ 6:082s1  1:458s21 lb=in:

where q1 is the starting value in the part, as shown in Fig. 12.9. At the end of part (1) we have
qð1Þ ð4Þ ¼ q1 þ 1 lb=in:, which is also the starting value for the shear flow in part (2) as shown
in Fig. 12.10. In this part we have
y ¼ 6:79  6s2 =13
z ¼ 0:5 þ 5s2 =26
(12.23)
Qðy2Þ ¼ zAs ¼ ð0:5 þ 5s2 =26Þð0:1Þðs2 Þ ¼ 0:05s2 þ 0:1923s22

Qðz2Þ ¼ yAs ¼ ð6:79  6s2 =13Þð0:1Þðs2 Þ ¼ 0:679s2  0:04615s22

so that the shear flow in this part is

   
qð2Þ ðs2 Þ ¼ q1 þ 1  29:16 0:05s2 þ 0:01923s22  6:073 0:679s2  0:04615s22
(12.24)
¼ q1 þ 1  5:58s2  0:2802s22 lb=in:

and at the end of part (2) we have qð2Þ ð13Þ ¼ q1  118:89 lb=in: For part (3), as shown in
Fig. 12.11, the same types of calculations give

Figure 12.11 The geometry for part (3) of the cross- q1 −118.89
section. z

s3
As 5.5

5.21
y
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 485

y ¼ 5:21

z ¼ 5:5  s3 =2

Qðy3Þ ¼ ð5:5  s3 =2Þð0:1Þðs3 Þ ¼ 0:55s3  0:05s23

(12.25)
Qðz3Þ ¼ ð5:21Þð0:1Þðs3 Þ ¼ 0:521s3
 
qð3Þ ðs3 Þ ¼ q1  118:89  29:16 0:55s3  0:05s23  6:073½0:521s3 

¼ q1  118:89  12:87s3 þ 1:458s23 lb=in:

and at the end of part (3) we have qð3Þ ð9Þ ¼ q1  116:62 lb=in: Finally, for part (4), as shown
in Fig. 12.12, we find
y ¼ 5:21 þ s4 =2

z ¼ 3:5

Qðy4Þ ¼ ð3:5Þð0:1Þðs4 Þ ¼ 0:35s4

(12.26)
Qðz4Þ ¼ ð5:21 þ s4 =2Þð0:1Þðs4 Þ ¼ 0:521s4 þ 0:05s24
 
qð4Þ ðs4 Þ ¼ q1  116:62  29:16½0:35s4   6:073 0:521s4 þ 0:05s24

¼ q1  116:62  13:37s4 þ 0:3036s24 lb=in:

We can check the consistency of these many calculations by examining the end value of the
shear flow in part (4), where we find qð4Þ ð12Þ ffi q1 , which was our original starting value.
Since we now have all the shear flows in the parts, we can calculate the total forces generated
in each part approximately as

z Figure 12.12 The geometry for part (4) of the cross-

section.

5.21

As
q1 −116.62 3.5 y

s4
486 Combined Deformations

ð4
ð1Þ


F ¼ q1 þ 6:082s1  1:458s21 ds1 ¼ 4q1 þ 17:55 lb
0
1ð3
ð2Þ


F ¼ q1 þ 1  5:58s2  0:2802s22 ds2 ¼ 13q1  663:7 lb
0
(12.27)
ð9


F ð3Þ ¼ q1  118:89  12:87s3 þ 1:458s23 ds3 ¼ 9q1  1237 lb
0
1ð2


F ð4Þ ¼ q1  116:62 þ 13:37s4  0:3036s24 ds4 ¼ 12q1 þ 611:4 lb
0

These forces are shown in Fig. 12.13 along with the original 1000 lb force they must produce.
These force systems are equivalent so they must generate the same moment about any point
(and the same force). Point A is a convenient point to take moments about so we find
ð12q1  611:4Þð9Þ þ ð4q1 þ 17:55Þð12Þ ¼ ð1000Þð0Þ
(12.28)
! q1 ¼ 33:92 lb=in:

Then the shear flows are explicitly

qð1Þ ðs1 Þ ¼ 33:9 þ 6:08s1  1:46s21 lb=in:

qð2Þ ðs2 Þ ¼ 34:9  5:58s2  0:208s22 lb=in:
(12.29)
qð3Þ ðs3 Þ ¼ 85:0  12:9s3 þ 1:46s23 lb=in:
qð4Þ ðs4 Þ ¼ 82:7 þ 13:4s4  0:304s24 lb=in:

These distributions are shown explicitly in Fig. 12.14.

Figure 12.13 The applied 1000 lb A A

force and the forces generated from
the shear flows. 13q1 − 663.7

1000 lb 9
= 9q1 −1237
4q1 + 17.55
12q1 − 611.4

12
 12.2 Bending and Torsion of Thin, Closed Cross-Sections 487

– 85 Figure 12.14 The shear-flow distributions

in the box beam.
– 85
– 13.2
34.9
(midpoint)
– 113
34.9
(midpoint)
40.2
(midpoint)
– 82.7 33.9
33.9
– 82.7
– 13.5
(midpoint)

Now, to find the location of the shear center for the vertical force we need to set ϕ0 ¼ 0,
which gives
þ
1 q
ϕ0 ¼ ds ¼ 0 (12.30)
2GΩ t

but, since the thickness t is a constant, we have equivalently

þ
qds ¼ 0 (12.31)

which is just the sum of the forces we have already computed, giving
ð4q1 þ 17:55Þ þ ð13q1  663:7Þ þ ð9q1  1237Þ þ ð12q1  611:4Þ ¼ 0
(12.32)
! q1 ¼ 65:65 lb:in:

With this value of q1 then in place of Fig. 12.13 we have the equivalent systems of Fig. 12.15.
Again, equating moments about point A of these two systems we find
ð4q1 þ 17:55Þð12Þ þ ð12q1  611:4Þð9Þ ¼ 1000ey
(12.33)
! ey ¼ 4:95 in:

Having this result, we can then decompose our original combined bending and torsion
problem into a purely bending problem where the shear force acts through the shear center
plus the torsion produced by the moment of the force about the shear center. This decom-
position is shown in Fig. 12.16 where we see that the clockwise torque about the shear center
is just T ¼ 4950 in.-lb. This torque will generate a constant primary shear flow throughout
the cross-section given by (see Eq. (11.212)):
T 4950
q¼ ¼ ¼ 31:7 lb=in: (12.34)
Ω 2½ð4Þð12Þ þ ð5Þð12Þ=2
488 Combined Deformations

A A

13q1 − 663.7

1000 lb 9 9q1 −1237

ey =
4q1 + 17.55
12q1 − 611.4

12

Figure 12.15 The applied 1000 lb force acting through the shear center and the forces generated in the
cross-section by the shear flows.

Figure 12.16 The original 1000 lb 1000 lb

loading which produces T = 4950 in.-lb
bending when the force acts
through the shear center plus = +
torsion of the section 4.95
generated by the moment of
the force acting about the q1 = 33.9 lb / in. ( q1 ) B = 65.5 lb / in. ( q1 )T = −31.7 lb / in.
shear center. The shear flows
at the starting point of the
integrations are also shown.

where the minus sign is present since we have been taking positive torques and shear flows as
acting counterclockwise. This value of the shear flow is also the value of q1 at the starting
point of the integration due to torsion only, so we can see that the original constant value of
q1 for combined bending and torsion is the sum (approximately) of its value ðq1 ÞB due to
bending and a value ðq1 ÞT due to torsion as shown in Fig. 12.16, in agreement with our
previous discussions.

12.3 Twisting Induced by Axial Stresses

In Chapter 11 we saw that, for nonuniform torsion problems, the torsion of the section
induced axial normal stresses. Is it possible to similarly induce torsion by the application of
axial normal stresses? The answer is yes. Consider, for example, the cantilever I-beam shown
in Fig. 12.17a. Because of the restraint of warping, we saw in Chapter 11 that normal stresses
are induced in the flanges (Fig. 12.17b). These normal stresses in turn are statically
 12.3 Twisting Induced by Axial Stresses 489

Figure 12.17 (a) Nonuniform torsion of

a cantilever I-beam. (b) Normal stress
Mx distributions in the flanges. (c) The
normal stresses are equivalent to a pair
of moments called bimoments.
x

(a)
s xx
Mx Mx
s xx = 0

x x

s xx
(b) (c)

equivalent to a pair of moments as shown in Fig. 12.17c called, for obvious reasons,
bimoments. Thus, if torsion can coexist with bimoments (which are statically equivalent to
a set of self-equilibrated axial stress distributions), it follows that we could view this problem
in reverse and consider the axial stress distributions which generate such bimoments as the
applied loads that in turn induce the torsion present. The challenge remains, however, to
describe in more general terms such self-equilibrated stresses, which we will now do.
To model such twisting induced by axial stresses we will consider a general restrained thin,
open section, as shown in Fig. 12.18, subjected to axial stresses that we will assume produce
both an axial displacement of the section, u0 ðxÞ, and nonuniform warping. Then the axial
displacement will be given by
ux ¼ ωp ðy; zÞϕ0 ðxÞ þ uo ðxÞ (12.35)

where ωP is the principal sectorial area function and ϕ0 is the twist/unit length. Recall that
the principal sectorial area function satisfies
ð
ωP dA ¼ 0
A
ð
yωP dA ¼ 0 (12.36)
A
ð
zωP dA ¼ 0
A
490 Combined Deformations

z Figure 12.18 A restrained thin, open section subjected to an axial

stress distribution.
y

If we assume σ xx is the only significant normal stress, then

dux d 2ϕ du0
V xx σ xx ¼ E ¼ Eωp þE (12.37)
dx dx dx
where E is Young’s modulus. If F x is the axial load gener-
x
L ated by the stress distribution, then
ð ð
d 2ϕ du0
F x ¼ σ xx dA ¼ E ωp dA þ EA
dx dx
A A (12.38)
du0
¼ EA
dx
because the integral of the principal sectorial area function over the cross-section is zero (the
first condition in Eq. (12.36)). Thus, Eq. (12.38) is just the ordinary relationship seen
between the axial force and the axial displacement. Also note that the bending moments
vanish since
ð ð ð
d 2ϕ du0
M y ¼ yσ xx dA ¼ E yωp dA þ E ydA ¼ 0
dx dx
A A A
ð ð
ð (12.39)
d ϕ 2
du0
M z ¼ zσ xx dA ¼ E zωp dA þ E zdA ¼ 0
dx dx
A A A

if ðy; zÞ are measured from the centroid of the cross-section. Thus, the axial stress of
Eq. (12.37) does not generate any bending about the centroid, only axial extension/
compression and warping.
The force, F x , is the resultant that describes axial extension/compression of the cross-


section and the moments M y , M z describe bending, so the question now is what type of
resultant describes the warping and associated torsion of the cross-section? We will define
such a resultant, M Γ ðxÞ, as
ð
M Γ ¼ σ xx ωp dA (12.40)
A

which, because of the behavior seen in the I-beam discussed previously, we will call the
bimoment. This definition makes sense because the bimoment is directly related to the
twisting of the section, i.e.,
 12.3 Twisting Induced by Axial Stresses 491

ð ð
d 2ϕ du0
M Γ ðxÞ ¼ E 2 ωp dA þ E
2
ωp dA
dx dx
A A (12.41)
d ϕ
2
¼ EJ ω
dx2
where J ω is the warping constant. If we place Eq. (12.41) and the axial force relation of
Eq. (12.38) into the expression for the axial stress, Eq. (12.37), we find
M Γ ðxÞωp ðy; zÞ F x
σ xx ¼ þ (12.42)
Jω A
If the applied stress is purely uniform across the cross-section, as assumed in axial-load
problems, then no twisting will be induced since
ð ð
M Γ ¼ σ xx ωp dA ¼ σ xx ωp dA ¼ 0 (12.43)
A A

However, other distributions that generate a bimoment as well as an axial force can produce
both twisting and extension/compression. If a given stress distribution does generate a
bimoment, then the equation for the twist becomes (see Eq. (11.135))

d 4ϕ dϕ
 k2 ¼0 (12.44)
dx4 dx
since there is no applied moment, M x , and where, recall (Eq. (11.136)),
GJ eff
k2 ¼ (12.45)
EJ ψ
To solve Eq. (12.44) for the twist ϕ we need three boundary conditions. For the cantilever
beam problem of Fig. 12.18, for example, we have at the wall (x ¼ 0) the conditions
ϕð0Þ ¼ 0, ϕ0 ð0Þ ¼ 0 (12.46)

The other boundary condition comes from the loaded end (x ¼ L), where we assume the
stress distribution (and hence the bimoment) is known, and from Eq. (12.41) we have

d 2 ϕ M Γ ð LÞ
2
¼ (12.47)
dx x¼L EJ ω

Example 12.3 Twisting of an Axially Loaded Member

Consider the cantilever unsymmetrical beam shown in Fig. 12.19, which is loaded by a
concentrated force, P ¼ 100 kN, acting at the centroid. We want to determine the twist,
492 Combined Deformations

Figure 12.19 (a) An z

unsymmetrical beam loaded by 100 mm
an axial load acting through the
centroid of the cross-section. (b)
the geometry of the cross-section z
where the thickness t ¼ 5 mm is
P = 100 kN
constant though out the cross- C y
section. C
x 200 mm
5 mm

L=3m

100 mm
(a) (b)

ϕðxÞ, in the beam and the axial stress distribution at the wall. In this case the centroid C is also
at the shear center. We will take E ¼ 200 GPa ¼ 2  105 N=mm2 , G/E ¼ 0.36. To determine
the principal sectorial area function, we only need to satisfy the first condition on that function
in Eq. (12.36) if we take the pole at the shear center, C. Thus, let us take the initial integration
point at A, where we let ω ¼ ω0 (see Fig. 12.20), and sweep out the sectorial area going from
A to D as shown. Then in the three sections of the cross-section, we have
ω ¼ ω0 þ hs1 =2 ð0  s1  d Þ
ω ¼ ω0 þ hd=2 ð0  s2  hÞ (12.48)
ω ¼ ðω0 þ hd=2Þ  hs3 =2 ð0  s3  d Þ
Ð
Setting A ωdA ¼ 0 gives

ðd ðh ðd
s1 t ðω0 þhs1 =2Þds1 þt ðω0 þhd=2Þds2 þt ðω0 þhd=2hs3 =2Þds3 ¼0
d 0 0 0
A
s2 hd ðhþd Þ ð200Þð100Þ ð200þ100Þ
!ω0 ¼ ¼
h/2 2 ðhþ2d Þ 2 ð200þ200Þ
¼7:5103 mm2
C
(12.49)
h/2

D
d Figure 12.20 The geometry for determining the principal sectorial area
s3
function, where the centroid C is the shear center.
 12.3 Twisting Induced by Axial Stresses 493

2.5×10 3 mm2 Figure 12.21 The principal sectorial area function.

– 7.5×103 mm2

2.5×10 3 mm2

– 7.5×10 3 mm2

which gives the principal sectorial area function shown in Fig. 12.21. It can then be shown
that
1 1
J eff ¼ 2  d t3 þ ht3 ¼ 0:17  105 mm4
3 3
ð
J ω ¼ ω2p dA ¼ 2:08  1010 mm6
(12.50)
A
sffiffiffiffiffiffiffiffiffiffiffiffi
G J eff
k¼ ¼ 5:4  104 mm1
E Jω

We now are in a position to calculate the twist in the beam. General solutions of the
homogeneous equation seen in Eq. (12.44) and its derivatives are
ϕ ¼ A cosh ðkxÞ þ B sinh ðkxÞ þ C

ϕ0 ¼ Ak sinh ðkxÞ þ Bk cosh ðkxÞ (12.51)

00
ϕ ¼ Ak 2 cosh ðkxÞ þ Bk 2 sinh ðkxÞ
00
where A, B, and C are constants. From ϕ0 ð0Þ ¼ 0 we find B ¼ 0. From ϕ ðLÞ ¼
M Γ ðLÞ=EJ ω we then find A ¼ M Γ ðLÞ=Ek 2 J ω cosh ðkLÞ. finally, from ϕð0Þ ¼ 0 we have
C = –A, giving
M Γ ð LÞ
ϕðxÞ ¼ ½1  cosh ðkxÞ (12.52)
Ek J ω cosh ðkLÞ
2

In this case it is easy to calculate M Γ ðLÞ since the force at x ¼ L is a concentrated force at the
origin of the centroidal coordinates so that
494 Combined Deformations

15.7 MPa Figure 12.22 Axial stress distribution at the wall.

61.4 MPa
ð
M Γ ðLÞ ¼ σ xx ðy; zÞωp ðy; zÞdA
A
ð
61.4 MPa ¼ ωp ð0, 0Þ σ xx dA (12.53)
Að0, 0Þ
15.7 MPa
¼ Pωp ð0, 0Þ ¼ 2:5  103 P N-mm

Placing the value of P and the other values into Eq. (12.52) gives approximately


ϕðxÞ ¼ 0:08 1  cosh 5:4  104 x rad (12.54)

so at x ¼ L we have ϕðLÞ ¼ 0:13 rad. At the fixed wall x ¼ 0 we have

d 2ϕ
M Γ ð0Þ ¼  ð0ÞE J ω
dx2 (12.55)
¼ 95  10 N-mm 6 2

and the axial stress is, in terms of the sectorial area function,


P M Γ ð0Þωp 105 95  106 ωp
σ xx ¼ þ ¼ þ
A Jω ð400Þð5Þ 2:08  1010 (12.56)
3
¼ 50 þ 4:57  10 ωp MPa

which has the distribution at the wall as shown in Fig. 12.22. It can be seen that the torsional
warping makes the stress distribution quite different from the constant P/A ¼ 50 MPa value
computed from ordinary axial-load theory even though the load acted through the centroid
of the cross-section.

12.4 PROBLEMS
t = 5 mm
P12.1 A 5mm thick plate is formed into the semicircular
shape of radius R ¼ 250 mm as shown in Fig. P12.1.
R = 250 mm The shear center location for this section was obtained in
S
O
4R
e=
p
Figure P12.1 A 10 kN shear force acting on a thin, open semicircular
10 kN cross-section.
 References 495

z Figure P12.2 A 1 kN shear force acting on a thin, open

t = 5 mm circular cross-section.
R = 100 mm
R
S
O T

e y

1 kN

z Figure P12.3 A 5 kN shear force acting on a closed semicircular section.

R = 100 mm
Chapter 10 (see Eq. (10.46)) to be at a distance e ¼ 4R=π
t = 6 mm
R from point O, as shown in the figure. Determine the
S maximum shear stress and its location in the cross-section
t y if a 10 kN vertical shear force acts as shown in Fig. P12.1.
0.53R
P12.2 Consider the thin, open circular section of Fig. P12.2.
t The thickness is t ¼ 5 mm and the radius of the section
is R ¼ 100 mm. A 1 kN shear force acts through the
5 kN centroid, O, of the cross-section. Determine the maximum
shear stress and its location in the cross-section.
P12.3 The 100 mm radius closed semicircular section of Fig.
P12.3 carries a 5 kN shear force acting on the straight
vertical part of the section. The thickness, t ¼ 6 mm, of the section is constant
throughout the section. The shear center, S, is located as shown in Fig. P12.3.
Determine the maximum shear stress and its location in the cross-section (neglect
stress concentrations). Also, plot the shear stresses separately in the straight vertical
part and the curved part.

References
1. V. Z. Vlasov, Thin-Walled Elastic Beams, 2nd edn (translated from Russian. Washington, DC:
Clearinghouse for Federal and Technical Information, TT61–11400, 1959)
2. D. V. Wallerstein, A Variational Approach to Structural Analysis (New York, NY: John Wiley and
Sons, 2002)
13 Material Failure and Stability

One of the primary reasons for doing stress analysis of structures is to ensure that those
structures remain safe. Structures can fail in different ways, so one needs to examine a
variety of failure modes. In this chapter we will consider (1) a number of the commonly used
static failure theories, (2) fatigue failure under alternating loads, and (3) fracture theory. We
will also briefly discuss how nondestructive inspections can be used in conjunction with
crack growth laws to keep structures safe while in use. Another way that structures can fail is
through a loss of stability. The sudden buckling of columns, also called a bifurcation type of
instability, will be described as well as other types of instabilities such as limit-load instabil-
ities and snap-through buckling instabilities.

13.1 Theories of Static Failure

The basic idea behind theories of static failure is that if nontime-varying stresses are too
large, then a material will fail. To examine static failure, let us consider first the special case
of plane stress (Fig. 13.1a). We can define what we mean by “too large” by plotting
combinations of the two principal stresses and defining a safe region where any of those
combinations do not lead to failure (Fig. 13.1b). If we can determine the boundary of such a
safe region through particular failure theories and experiments, then we have a rational way
to guarantee the safety of the structure. Let us now examine a number of static failure
theories.

13.1.1 Maximum Normal-Stress Theory

Materials are often classified into two types: ductile and brittle. A ductile material under
large loads exhibits significant permanent strains, called “slipping,” when unloaded. Brittle
materials, however, do not have this behavior over the full range of stresses all the way to
ultimate fracture. Many steels, for example, exhibit ductile behavior, whereas materials such
as cast iron are brittle. The


maximum
normal-stress

theory for a brittle material says that
 
fracture occurs if either
σ p1
¼ σ f or
σ p2
¼ σ f , where σ p1 , σ p2 are the principal stresses
and σ f is a fracture stress as determined by pulling a specimen to failure in a uniaxial tension
test (see Fig. 13.2a). A plot of the safe region for this theory is shown in Fig. 13.2b.
One of the limitations of this theory is that it does not distinguish between tensile or
compressive stress behavior and materials do typically show different failure criterion in
tension or compression. There is another theory, called the Coulomb–Mohr theory, that

496
 13.1 Theories of Static Failure 497

V p2 Figure 13.1 (a) A plane state of

V yy
failure stress. (b) The definition of failure
in terms of a region in the plane of
V yx principal stresses.
V xy failure
V xx safe region

V p1

failure
failure

(a) (b)

V p2 Figure 13.2 (a) A uniaxial

tension test where the bar is
Vf loaded to the fracture stress,
σ f . (b) The safe region
Vf (shaded area) for the
−V f Vf maximum normal-
V p1 stress theory.

(a)
−V f

(b)

V p2 Figure 13.3 The safe region for a general state of stress according to the
maximum normal-stress theory.
Vf

does take such tensile/compressive differences into account but we will

Vf
not cover it here. This theory also does not depend on the orientation
V p1 of the principal planes so it is only applicable to isotropic materials.
The generalization of this theory to a 3-D state of stress is easy, where
V p3 the corresponding safe region is the cube shown in Fig. 13.3.

13.1.2 Maximum Shearing-Stress Theory

For ductile materials the maximum shearing-stress theory (also called the Tresca theory) says
that failure by slip (also called yielding) occurs when the maximum shearing stress, τ max ,
exceeds the yield shear stress, τ y , as determined experimentally in a uniaxial tension test, as
shown in Fig. 13.4a. In uniaxial tension, this yield shear stress is τ y ¼ σ y =2, where σ y is the
498 Material Failure and Stability

Figure 13.4 (a) A uniaxial tension s

test where the bar is loaded to the ×
normal yield stress, σ y , and the sy
corresponding shear stress τ y ¼ σ y =2. ty
(b) A stress–strain curve, where the
yield stress, σ y , occurs when, upon
unloading, the material is left with a sy
specified permanent strain, es , at
zero stress. e
es
(a) (b)

Figure 13.5 The safe region (shaded area) for no failure V p2

by slip according to the maximum shearing V p2 V y
stress theory.
V p 2 − V p1 V y

V p1 V y
o
45
V p1
V p1 −V y

V p1 − V p 2 V y

V p2 −V y

normal stress at yielding, and this yield shear stress occurs on planes at 45∘ to the x-axis, as
shown in Fig. 13.4a. The yield stress, σ y , however, is not the stress at which the bar fractures.
Instead it defined as the normal stress that, when the bar is completely unloaded, results in a
specified residual strain, es (also called a permanent slip), as shown in Fig. 13.4b. This
permanent slip is why this type of failure is called failure by slip.
 
In a plane state of stress there are three principal stresses σ p1 , σ p2 ; 0 , and depending on
their values, the maximum shear stress is the largest of the three extreme shear stresses
given by
8




=2 8




=2
>
> σ p1  σ p2 >
> σ p1  σ p2
<

<


τ max ¼ max
σ p1  0
=2 ¼ max
σ p1
=2 (13.1)
>
>

>
>


:
:


σ p2  0
=2 σ p2 =2

The safe region τ max  τ y ¼ σ y =2 is thus the shaded hexagonal area shown in Fig. 13.5. Note
that this theory also does not depend on the orientations of the planes of extreme shear so
that it is only strictly applicable to isotropic materials.
 13.1 Theories of Static Failure 499

13.1.3 Maximum Distortional Strain Energy Theory

Intuitively, we expect a material will fail if we place too much energy into it. Thus, we might
expect that we can use the size of the strain energy density, u0 , as a failure criterion.
However, as discussed in Chapter 8, experiments have shown that a pure hydrostatic
pressure will not cause yielding, so failure due to slip must be independent of the hydrostatic
part of u0 . By removing that hydrostatic part (see Chapter 8), we can define the distortional
stain energy density, ud , as
ð1 þ νÞ h 2  2  2 i
ud ¼ σ p1  σ p2 þ σ p1  σ p3 þ σ p2  σ p3 (13.2)
6E
The maximum distortional strain energy theory (also called the von Mises failure theory)
predicts that failure with respect to slip (yielding) occurs when the distortional strain energy
density is equal to its value, uy , as determined in a uniaxial tension test where there is only
one principal stress and that is equal to the normal stress at yield, σ y , so that
ð 1 þ νÞ h 2 i
uy ¼ 2σ y (13.3)
6E
Thus, the failure criterion ud ¼ uy gives
 2  2  2
σ p1  σ p2 þ σ p1  σ p3 þ σ p2  σ p3 ¼ 2σ 2y (13.4)

which for a state of plane stress reduces to

σ 2p1  σ p1 σ p2 þ σ 2p2 ¼ σ 2y (13.5)

Equation (13.5) represents the equation of a rotated ellipse, as shown in Fig. 13.6. The
corresponding hexagonal shape from the maximum shearing stress theory is also shown in
Fig. 13.6 so we see that there is little difference between the two theories.
In a general 3-D state of stress, we can also interpret the failure criterion in two other
equivalent ways. Recall, in Chapter 2 we obtained the magnitude of the total shear stress on
the octahedral plane as (see Eq. (2.82)):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  2  2  2
jτ s joct ¼ σ p1  σ p2 þ σ p1  σ p3 þ σ p3  σ p2
3
V p2 (13.6)
Vf In the uniaxial load case at the yield stress, σ y , we have this
shear stress on the octahedral plane, τ oct y , given by
pffiffiffi
τ y ¼ 2σ y =3. The failure criterion jτ s joct ¼ τ oct
oct
y for the
−V f 45o
stress on the octahedral plane gives Eq. (13.4) so that it is
Vf V p1

Figure 13.6 The safe region (shaded area) for no failure by slip
−V f
according to the maximum distortional strain energy theory.
500 Material Failure and Stability

equivalent to the distortional strain energy criterion pffiffiffi ud ¼ uy . Alternatively, if we define the
von Mises or effective stress, σ v , as σ v ¼ 3jτ s joct = 2 (see Eq. (2.84)), then at the yield stress
this effective stress is just σ y , so the failure criterion σ v ¼ σ y is again equivalent to ud ¼ uy .
Thus, we can view the distortional strain energy criterion as equivalent to a criterion on
either the octahedral stress or the effective stress.
In formulating this failure theory, we used generalized Hooke’s law so that this theory as
presented is applicable only to isotropic materials, but it can be generalized to anisotropic
materials. The maximum distortional strain energy theory is less conservative than the
maximum shear-stress theory, but generally there is little difference in their predictions of
failure. Most experimental results tend to fall on or between these two theories.
For 3-D states of stress, the maximum distortional strain energy theory is given by
Eq. (13.4) while the maximum shear-stress theory becomes
8


>
σ p1  σ p2
=2
>
<


τ max ¼ τ y ¼ σ y =2 ¼ max
σ p1  σ p3
=2 (13.7)
>
:

σ  σ

=2
>
p2 p3

Both Eq. (13.4) and Eq. (13.7) remain unchanged if we add equal stresses to all principal
stress components, i.e., if we let

σ 0p1 ¼ σ p1 þ σ
σ 0p2 ¼ σ p2 þ σ
σ 0p3 ¼ σ p3 þ σ

This is because both theories are independent of adding or subtracting a hydrostatic

component. What this means is that in the 3-D case the failure (yield) surfaces are cylinders
whose sides are parallel to a line that makes equal axes with all three principal stress
directions, as seen in Fig. 13.7.

Figure 13.7 Failure surface for a 3-D state p3

of stress using the maximum shearing

stress (Tresca) theory or the maximum 1 1 1
n= e p1 + e p2 + e p3
distortional strain energy (von 3 3 3
Mises) theory. e p3 e p 2 p2

e p1

p1

ellipse for von Mises,

hexagon for Tresca
 13.1 Theories of Static Failure 501

Example 13.1 Static Failure Criteria

A solid circular shaft having a 60 mm radius is loaded by both a 10 kN-m bending moment
and a 12 kN-m torque. The isotropic material of the shaft has a measured yield stress of
220 MPa in uniaxial tension. What is the safety factor for failure by slip according to the
maximum shearing-stress theory and the maximum distortional strain energy theory?
The largest flexure normal stress and torsional shear stress occur at the outer radius, c, of
the shaft and are given by
 
Mc 10  106 ð60Þ
σ xx ¼ ¼ ¼ 58:9 MPa
I πð60Þ4 =4
 
Tc 12  106 ð60Þ
τ¼ ¼ ¼ 35:4 MPa
J πð60Þ4 =2

At this outer radius we have a state of plane stress so principal stresses are given by (see
Eq. (2.67))
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
σ xx þ σ yy σ  σ 2 58:9 58:9 2
xx yy 2
σ p1 , σ p2 ¼  þ σ 2xy ¼  þ ð35:4Þ
2 2 2 2
σ p3 ¼ 0

which gives

σ p1 ¼ 75:5 MPa, σ p2 ¼ 16:6 MPa, σ p3 ¼ 0

If we examine the Mohr circles, we see the maximum shear stress is just
σ p1  σ p2
τ max ¼ ¼ 46:05 MPa
2
Thus, according to the maximum shear-stress theory the safety factor, SF, is just

τy σ y =2 110
SF ¼ ¼ ¼ ¼ 2:4
τ max τ max 46:05
In contrast, for the distortional strain energy theory, written in terms of the effective stress:
pffiffiffi
σy 2σ y
SF ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   2  2
σ eff 2
σ p2  σ p1 þ σ p2  σ p3 þ σ p1  σ p3
pffiffiffi
2ð220Þ
¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ¼ 2:6
2 2 2
ð92:1Þ þ ð16:6Þ þ ð75:5Þ
502 Material Failure and Stability

13.2 Fatigue Failure

The failure theories of the previous section are appropriate for structures under static loads.
However, many structures are subjected to significant time-varying loads so we must
develop equivalent failure theories for those cases. In cases such as reciprocating machinery,
for example, the time varying loads are sinusoidal. Under such repetitive loading, small
imperfections in the material or on the surface, for example, develop into microscopic
cracks, which grow over time into larger cracks until eventually a crack of “critical” size is
produced that grows rapidly, causing the material to fracture. This type of process is called
fatigue failure. As you might expect, fatigue failure is in detail a very complex process so that
one cannot develop relatively simple failure criterion as found in the case of static failure.
Thus, in fatigue one relies heavily on the use of experiments. Figure 13.8a shows a typical
experimental setup that is used to generate fatigue failure data. A cantilever beam of circular
cross section is forced to rotate while being loaded by an end force, P through a fixed
surrounding sleeve. In this arrangement, the extreme flexure stresses at the top and bottom
of the beam go through alternating (i.e., completely reversed) tensile and compressive values
as a function of time, as seen in Fig. 13.8b, where σ a is the magnitude of this alternating
stress. The beam is allowed to remain in this setup until the beam ultimately fails at a specific
number of cycles, N 1 . This whole process is then repeated with a new sample at a new value
of P (and, hence, σ a ) resulting in a different number of cycles to failure, N 2 . A collection of
such experiments then allows us to generate an S–N curve, where the alternating stress, σ a
(which is normally labeled as S), is plotted versus the number of cycles to failure, N f , as seen
in Fig. 13.9. Some materials exhibit a stress level, σ e , called the endurance limit, below which
the life of the part is essentially infinite. If such asymptotic behavior is not present, then the
endurance limit is usually defined as the stress at a very large number of cycles such as
N f ¼ 108 . The S–N curve is normally generated with a purely reversed alternating stress, σ a ,
but in practice a component will often also see a static (mean) stress, σ m , as seen in
Fig. 13.10, where the mean stress, alternating stress, and stress range, Δσ, are defined in
terms of the maximum and minimum stresses experienced. Combinations of mean and
alternating stresses are often defined in terms of the stress ratio, R ¼ σ min =σ max , and the
amplitude ratio, A ¼ σ a =σ m . Fig. 13.11b shows a few combinations of ðR; AÞvalues that are
common. If one measures the number of cycles to failure at various combinations of such

Figure 13.8 (a) Generating a stress

sinusoidal state of stress versus time
by rotating a cantilever specimen
with a static end load, P. (b) A plot of Va time
the stress versus time where the stress P
is an alternating (i.e., completely Va
reversed) stress, σ a .

(a) (b)
 13.2 Fatigue Failure 503

800 Figure 13.9 A typical S–N curve for fatigue failure.

600
S Va
(MPa) V e Se
endurance limit
400
103 104 105 106 107 108
N f , cycles to failure

Figure 13.10 A more V

general time-varying
harmonic stress that V max 'V V max  V min stress range
includes a mean (static) Va
stress component as well as 'V V max − V min
'V Va alternating stress
an alternating component. 2 2
V min Vm V max  V min
Vm mean stress
2

time, t

fully reversed R –1, A ¥ Figure 13.11 (a) A loading with mean

stress, σ m , and alternating stress, σ a ,
s max
sa values. (b) Some specific combinations
of ðσ m , σ a Þas defined in terms of the
zero to max R = 0, A = 1 stress ratio, R ¼ ðσ min =σ max Þ, and
s min amplitude ratio, A ¼ σ a =σ m .
sm

zero to min R ¥, A –1
(a) (b)

mean and alternating stresses then a sequence of S–N curves can be generated, as shown in
Fig. 13.12a. At a given constant life (i.e., constant number of cycles to failure) then one can
generate the constant life curves of Fig. 13.12b from the data of Fig. 13.12a. Generating such
constant life curves experimentally in this fashion is very time consuming and costly so
instead one normally uses an empirical constant life curve that has the general behavior seen
in Fig. 13.12b. There are a number of different empirical curves that have been used.
504 Material Failure and Stability

Va Va

–100 S–N curve

0
ultimate static
100 failure stress
200 Vu
Vm 400 MPa

10
6
10
7
Nf –100 0 100 200 400 600 Vm
(a) (b)

Figure 13.12 (a) S–N curves generated for different mean stresses. (b) Constant life curves of alternating
versus mean stress values.

Va Figure 13.13 Empirical constant life curves: (a) Soderberg, (b)

Goodman, (c) Gerber, and (d) Morrow.
V fs
c
The most common of those curves are shown in
Fig. 13.13. The names of their authors (as well as the
d
a b country of origin and the year they were introduced) and
Vy Vu Vf Vm the behavior of these curves are summarized as follows:

σa σm
Soderberg ðUSA; 1930Þ þ ¼1 (13.8a)
σ fs σ y
σa σm
Goodman ðEngland; 1899Þ þ ¼1 (13.8b)
σ fs σ u
 2
σa σm
Gerber ðGermany; 1874Þ þ ¼1 (13.8c)
σ fs σu
σa σm
Morrow ðUSA; 1960sÞ þ ¼1 (13.8d)
σ fs σ f

As can be seen from Fig. 13.13, all these curves are defined by their limiting values at σ m ¼ 0
and σ a ¼ 0. The stress σ fs , which is called the fatigue strength, is common to all these curves.
It is the purely alternating stress at failure for a specified lifetime of N f cycles. The stress σ y is
the static yield stress while the stress σ u is the ultimate static stress at fracture. The stress σ f
is the so-called true stress at fracture. Unlike the ultimate stress, which is given by the ratio of
the failure load at fracture, Pfailure , to the original cross-sectional area, A0 , of the specimen,
i.e., σ u ¼ Pfailure =A0 , the true fracture stress, σ f , is the ratio of the failure load to the actual
cross-sectional area, A, at failure, i.e., σ f ¼ Pfailure =A, so that it accounts for any significant
 13.2 Fatigue Failure 505

S Figure 13.14 A finite life design based on (a)

the Goodman line and (b) the S–N curve for
V fs V fs
the material.

Va constant life of N f cycles

Vm Vu Nf N
(a) (b)

reduction in that area at large strains. Generally, most test data tend to fall between the
Goodman line and the Gerber curve. The Soderberg line is very conservative. If the peak
stresses used in these fatigue calculations are based on stresses at locations where a stress
concentration factor, K t , is present, one should be aware that the use of such stress concen-
tration factors produces overly conservative results so that in cyclic loading the K t value is
often replaced by a modified fatigue stress concentration factor, K f < K t that is based on
experimental data.
If one wants to do finite fatigue life design based on these fatigue concepts, there are
basically three steps.

(1) A given combination of mean and alternating stresses is taken to lie on a constant life
curve such as, for example, the Goodman line (Fig. 13.14a).
(2) The equation of that line is then used to solve for the fatigue strength, σ fs . In the
Goodman case we can have
σa σm
þ ¼1 (13.9)
σ fs σ u

which yields
σa
σ fs ¼ (13.10)
1  σ m =σ u
(3) Using this fatigue strength, the lifetime, N f , for this combination of mean and alternat-
ing stresses is found from the S–N diagram for the given material (Fig. 13.14b).
Even though these three steps are straightforward, note that we must get values for ðσ a , σ m Þ
and it is not at all clear how to do this if a component is in a general 3-D state of stress. There
are a number of possibilities, such as using the largest tensile principal stress behavior or
using the effective (i.e., von Mises) stress to calculate these mean and alternating stress
values, but there is no fundamental reason for making those choices. One can examine a
number of the possible ways to calculate these ðσ a , σ m Þ values and evaluate the sensitivity of
the fatigue life results to those choices, as we will show in the next example. We will not go
into further details here but there is a wide literature on fatigue that can be consulted to help
make judgements in specific cases (see [1]–[3]).
506 Material Failure and Stability

Example 13.2 Fatigue Life Design

A closed cylindrical pressure vessel has a mean radius rm ¼ 60 mm and a wall thickness of
t ¼ 4 mm. The vessel is subjected to an internal pressure that repeatedly cycles between zero
and 25 MPa. How many pressure cycles can be applied before fatigue failure is expected?
For this multiaxial fatigue problem, use a Goodman line for the constant life curve and
compute the mean and alternating stresses using the behavior of the largest tensile principal
stress. Take the yield stress (in tension) of the material of the cylinder to be 320 MPa and the
ultimate stress (in tension) to be 476 MPa. Assume the S–N curve can be written as
S ¼ AN B (13.11)
where S is the alternating stress in MPa, N is the number of cycles to failure, and A ¼ 839,
B ¼ 0.102 for the material of the vessel. How do your answers change if you use the effective
(von Mises) stress to calculate mean and alternating stress values?
The hoop stress in the cylinder is σ h ¼ prm =t and the axial stress is σ x ¼ prm =2t (see
Chapter 7). Both of these are principal stresses, so the hoop stress is the largest, varying
from zero to a maximum value σ h ¼ ð25Þð60Þ=4 ¼ 375 MPa. This is an alternating load with
R ¼ 0, A ¼ 1 (see Fig. 13.11b), so we have σ m ¼ σ a ¼ 187:5 MPa. Placing these values and the
ultimate stress into the Goodman line, we have the fatigue strength
σa 187:5
σ fs ¼ ¼ ¼ 309:4 MPa
1  σ m =σ u 1  187:5=476
If we place this value in the S–N curve, we can solve for the number of cycles as
σ 1=B 309:4 9:804
fs
Nf ¼ ¼ ¼ 1:77  104 cycles
A 839
If we had used the effective stress to calculate the mean and alternating values, then letting
σ p1 ¼ σ x , σ p2 ¼ σ h ¼ 2σ x , σ p3 ¼ 0 we have for the effective stress at the largest pressure
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2  2
1
σ eff ¼ pffiffiffi σ p2  σ p1 þ σ p2  σ p3 þ σ p1  σ p3
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi pffiffiffi
1
¼ pffiffiffi σ 2x þ 4σ 2x þ σ 2x ¼ 3 σ x ¼ 3ð25Þð60Þ=8
2
¼ 324:8 MPa
so that σ m ¼ σ a ¼ 162:4 MPa, which gives a fatigue strength of
σa 162:4
σ fs ¼ ¼ ¼ 246:5 MPa
1  σ m =σ u 1  162:4=476
and from the S–N curve
σ 1=B  9:804
fs 246:5
Nf ¼ ¼ ¼ 1:64  105 cycles
A 839
which is about nine times the number of cycles based on the largest principal stress.
 13.2 Fatigue Failure 507

In practice, a structure typically will see different numbers of cycles at varying stress levels
over its lifetime. To handle this case, consider the following general model where we suppose
that a body can tolerate only a certain amount of total damage, D, before failure. If that
total damage comes from damages, Di ¼ ði ¼ 1; 2; . . . ; N Þ arising from M sources, then we
might expect that failure will occur if
X
M
Di ¼ D (13.12)
i¼1

or, equivalently,
X
M
Di
¼1 (13.13)
i¼1
D

defines failure, where Di =D is the fractional damage from the ith source. We can use this
linear damage concept in a fatigue setting by considering the situation where a component is
subjected to n1 cycles at alternating stress σ 1 , n2 cycles at stress σ 2 , . . . , nM cycles at σ M .
From the S–N curve for this material, we can find the number of cycles to failure, N 1 , at σ 1 ,
N 2 at σ 2 , . . . , N M at σ M . It is reasonable in this case to let the fractional damage at stress
level σ i be Di ¼ ni =N i . This leads to the Palmgren–Miner rule, which says that fatigue failure
occurs when
XM
ni
¼1 (13.14)
i¼1
Ni

Example 13.3 Palmgren–Miner Rule

Consider a part in a fatigue environment where 10 percent of its life is spent at an alternating
stress level, σ 1 , 30 percent is spent at level σ 2 , and 60 percent at a level σ 3 . How many cycles,
n, can the part undergo before failure?
If, from the S–N diagram for this material, the number of cycles to failure at σ i is N i
(i ¼ 1, 2, 3), then from the Palmgren–Miner rule failure occurs when
0:1n 0:3n 0:6n
þ þ ¼1 (13.15)
so solving for n gives N1 N2 N3

1
n¼ (13.16)
0:1 0:3 0:6
þ þ
N1 N2 N3

In evaluating the Palmgren–Miner rule experimentally, in “High–Low” fatigue tests

where testing occurs sequentially at two stress levels ðσ 1 , σ 2 Þ, and where σ 1 > σ 2 , it is found
that failure occurs when
508 Material Failure and Stability

X2
ni
¼c
i¼1
Ni

where c normally is less than one, i.e., the Palmgren–Miner rule is nonconservative. For
“Low–High” tests, on the other hand, c values are generally greater than one. For tests with
random loading histories at several stress levels (rather than the cyclical histories normally
considered), the Palmgren–Miner rule is very good.

13.3 Fracture Mechanics

In fatigue failure the ultimate cause of failure is the growth of cracks until a crack of a
critical size is reached that leads to rapid fracturing of the component. Thus, an analysis of
the behavior of cracks is essential for a more complete understanding of failure. Fracture
mechanics (see [4]–[9]) deals with the stresses that surround cracks and the stress criteria for
failure due to catastrophic crack growth. In Chapter 7 we showed that if the material
behaves elastically, the stress field around cracks is singular near the crack tip. In practice,
stresses cannot be infinite since the material will yield at large stress, creating a region near
the tip where the material behaves in an inelastic manner. However, it has been found that as
long as this region is small, the assumption that the material remains elastic everywhere
allows one to analyze the fracture process with a purely elastic model.
Consider, for example, a thin plate of width 2b that is uniformly loaded at its ends and
contains a centrally located crack of length 2a, as shown in Fig. 13.15a. When the material
behavior in the plate is purely elastic, the stresses near a crack tip (see Fig. 13.15b) can be
written in terms of r and θ as
K I G ij ðθÞ
σ ij ffi pffiffiffiffiffiffiffi (13.17)
2πr

Figure 13.15 (a) A stressed plate containing a V

central crack of length 2a. (b) A magnified view of
the right end of the crack where the origin of the
polar coordinates ðr; θÞ is at the crack tip.
r
T

2a
2b

V
(a) (b)
 13.3 Fracture Mechanics 509

where K I is called the stress intensity factor for the crack in a mode I opening mode (i.e.,
when the loading tends to open the crack as shown schematically in Fig. 13.15b, and G ij ðθÞ is
a non dimensional angular factor that depends on the specific loading and geometry. The
pffiffiffiffiffiffiffiffiffiffiffiffiffi
dimension of a stress intensity factor is stress length. Stress intensity factors are written in
the form
pffiffiffiffiffi
K ¼ f ðg; aÞσ πa (13.18)

where σ is a reference stress value and a is a reference crack length. The function f ðg; aÞ is a
nondimensional configuration factor that depends on the reference crack length a and the
geometry, g, of the component (g can be a complex function of more than one variable but it
is simply designated here by a single symbol). Stress intensity configuration factors have
been tabulated for many different geometries. Figures 13.16 and 13.17 show examples of a
few cases. Note that in contrast to the central crack of Fig. 13.16a, where a is the half-crack
length, for an edge crack (Fig. 13.16b) the length a is taken to be the full length of the crack.
For the rectangular beam in Fig. 13.17b, the crack is on the bottom side of the beam because
it is the tensile stress on this side that will tend to open the crack and cause it to grow. In
contrast, compressive stresses will tend to close a crack and prevent further growth. The
opening mode I stress intensity factor, K I , is the most common stress intensity factor studied
since cracks often tend to grow in a tensile stress field with this type of mode. However, a
crack can also be subjected to a shearing or tearing type of deformation (see Fig. 13.18) and
there are separate stress intensity factors ðK II , K III Þ associated with those cases. Here we will
focus only on type I mode opening stress intensity factors.

s s Figure 13.16 Stress intensity factors for

uniaxial loading of (a) a centrally cracked
plate, and (b) a plate with an edge crack.

2a a
2b b

s s

KI = f ( g, a ) s p a

pa f ( g , a ) = 1.12 − 0.231( a / b ) +10.55 ( a / b )

2
f ( g , a ) = sec
2b
−21.72 ( a / b ) + 30.39 ( a / b )
3 4

~ 1 ( b >> a )
=
(a) (b)
510 Material Failure and Stability

Figure 13.17 Stress intensity V

factors for (a) uniaxial loading t thickness
of a plate with cracks on both
edges, and (b) for a
V 6 M / tb 2 max tensile
rectangular beam under a pure bending stress
bending moment with an edge
crack on the tensile stress side
a a M b M
of the beam.
a
2b

V
KI f g, a V pa

1.12  0.203 a / b
2
f g, a f g, a 1.122  1.40 a / b  7.33 a / b
2 3
1.197 a / b  1.930 a / b
3 4
13.08 a / b  14.0 a / b

(a) (b)

Figure 13.18 The three types of stress

intensity factors.

mode I mode II mode III

KI K II K III
opening shearing tearing
mode mode mode

All of the examples shown so far have been for 2-D through-thickness cracks, but one
can also examine the stress intensity factor for 3-D cases such as a penny-shaped crack in a
block (Fig. 13.19a) or a thumbnail semicircular crack on the edge of a block (Fig. 13.19b).
Stress intensity factors for complex geometries and loadings must typically be obtained
numerically.
The reason that the stress intensity factors are needed is because it has been found that the
rapid growth of a crack to failure occurs when the stress intensity factor exceeds a certain
value, called the fracture toughness, K C . The fracture toughness is a function of the specimen
thickness, temperature, environmental conditions, etc., so that tests to obtain this toughness
 13.3 Fracture Mechanics 511

V V Figure 13.19 Stress intensity factors for (a) a circular

crack under a state of uniaxial tension, and for (b) a
semicircular crack under a state of uniaxial tension.

a
a
2a

V V
2V § 2V ·
KI pa KI 1.12 ¨ pa ¸
p © p ¹
(a) (b)

must be done under conditions that simulate the actual situations found in practice. As the
test specimen thickness increases, the fracture toughness decreases until it reaches a constant
value called the plane strain fracture toughness. For a crack in opening mode I, the plane
strain fracture toughness is usually written as K IC . The size of the crack at the plane strain
toughness is called the critical flaw size, ac , where
pffiffiffiffiffiffiffi
K IC ¼ f ðg; ac Þσ πac (13.19)

Example 13.4 Estimating Critical Flaw Size

If one knows both the yield strength and plane strain fracture toughness of a material, then
you can obtain a rough estimate of the critical flaw size. For example, consider a 4340 steel
(an airframe material) and a A533B steel (a nuclear pressure vessel steel). For the 4340
pffiffiffiffi
steel, the yield stress is σ Y ¼ 240 ksi and the fracture toughness
pffiffiffiffi is K IC ¼ 50 ksi in ., while
for the A533B steel σ Y ¼ 60 ksi and K IC ¼ 140 ksi in . Consider the case where we have a
small half-penny-shaped crack on the stress-free surface. If the crack is stressed in an
opening mode then

2σ pffiffiffiffiffi
K I ¼ 1:12 πa (13.20)
π

Fracture occurs when K I ¼ K IC , a ¼ ac , so under these conditions we can solve Eq. (13.20)
for the critical flaw size. If we also take the reference stress to be the yield stress (which
admittedly is a very large value but here we are just doing a rough comparative estimate),
then we find
512 Material Failure and Stability

 2
K IC
ac ¼ 0:626 (13.21)
σ yield

For the 4340 steel, Eq. (13.21) gives ac ¼ 0:03 in:, while for the A533B steel ac ¼ 3:4 in:
Clearly, the critical flaw size can vary significantly with the choice of material. If one inspects
a material or structure with a nondestructive evaluation (NDE) test such as ultrasound, for
example, in order to ensure that cracks have not grown to near a critical length, the critical
flaw size determines the sensitivity required in the inspection.

Fracture toughness is generally obtained experimentally with standard specimens such as

a beam or uniaxial tension member with an edge crack at the root of a notch. For plate
geometries such as seen in Fig. 13.16 and Fig. 13.17 consistent results can be obtained with
such tabulated values as long as the plate thickness, t, and the crack length, a, satisfy [9]:

K IC 2
t, a  2:5 (13.22)
σ yield

If Eq. (13.22) is not satisfied, then predicting failure on the criterion K I < K IC is conserva-
tive and an elastic–plastic fracture analysis can lead to a more efficient design.
In order to estimate the remaining safe life in a material undergoing a fatigue environ-
ment, it is necessary to know the rate at which cracks are growing in the material. If the
minimum and maximum stresses are ðσ min , σ max Þ then the corresponding stress intensity
factor will change from K min to K max where
pffiffiffiffiffi
K max , min ¼ f ðg; aÞσ max , min πa (13.23)
A commonly used crack growth law is the Paris law (also called the Paris–Erdogan law):
da
¼ CΔK m , ΔK ¼ K max  K min (13.24)
dN
where C, m are material and environmental dependent constants, and a ¼ aðN Þ is the crack
size as a function of N, the number of cycles of loading. On a log–log plot (see Fig. 13.20),
this crack growth law is linear, a behavior that is often found in crack growth studies except
at early stages of growth where there can be a threshold value of ΔK below which there is no
crack growth and the curve bends down, or at late stages of growth near the critical flaw size,
where the crack growth becomes quite rapid and the curve bends up. In terms of stress the
Paris law can be written as
da pffiffiffiffiffi m
¼ C πaf ðg; aÞΔσ Δσ ¼ σ max  σ min (13.25)
dN
which can be rearranged and integrated to yield
ðc
N aðc
da
ΔN ¼ dN ¼ pffiffiffiffiffi m (13.26)
C ½ πaf ðg; aÞΔσ 
N1 a1
 13.3 Fracture Mechanics 513

Figure 13.20 Crack growth on a log–log plot,

showing a linear region where the Paris–Erdogan
rapid growth law holds.
da
log to failure
dN

Paris–Erdogan
law holds
threshold

log ( DK )

received flaw signal Figure 13.21 An ultrasonic NDE inspection setup

for examining components for flaws.
pulser/receiver

transducer
flaw

where a1 is the crack size at N 1 cycles, ac is the critical flaw size at failure after N c cycles,
and ΔN ¼ N c  N 1 is the number of cycles left to failure. Thus, if we know (1) the
configuration factor from fracture mechanics studies, (2) the critical flaw size from stress
analyses and material tests, (3) the range of stresses, Δσ, experienced by the component,
and (4) the measured value of the crack size a1 at N 1 cycles from a nondestructive
evaluation (NDE) inspection of the component, then we can estimate the remaining
life ΔN.
There are various NDE inspection methods in use. Three of the most common inspec-
tion methods are ultrasonics, eddy currents, and X-rays. Other methods include dye
penetrants, magnetic particle, thermography, and more. In the case of an ultrasonic
inspection, a transducer is driven by short voltage pulses from a pulser/receiver (see
Fig. 13.21). The transducer is normally made from a piezoelectric crystal that converts
these voltage pulses into motion of the crystal which is then transmitted into the com-
ponent being inspected as elastic waves. The waves interact with any flaws, such as
cracks, present in the component, which reflect those waves back to the transducer.
The piezoelectric transducer converts the motion of the crystal back into received voltage
pulses, which are sent to the pulser/receiver to be amplified and displayed. In the case
of eddy current inspection (Fig. 13.22), an alternating current and voltage are used to
drive a coil, which is then placed on or near the surface of an electrically conducting
514 Material Failure and Stability

Figure 13.22 An eddy current NDE inspection measure impedance

setup for evaluating surface-breaking flaws.
alternating voltage,
current
coil
~
primary
fields

flaw secondary
fields

eddy currents

component film or image Figure 13.23 An NDE X-ray inspection setup.

capture screen

X-ray tube component. The time-varying fields surrounding the

coil, called the primary fields, induce secondary fields
in the component and corresponding “eddies” of
currents as shown schematically in Fig. 13.22. If a
surface crack is present, it interrupts the flow of these
flaw eddy currents and causes a change of the coupled
coil-component electrical impedance that can be
measured. For X-ray inspections, a stream of elec-
trons is generated in an X-ray tube that then strikes a target, producing X-rays that pass
out through a window in the tube. These X-rays interact with a flawed component,
producing an X-ray image of the flaw on a piece of film or a screen (Fig. 13.23) due to
changes in the absorption of the X-rays caused by the presence of the flaw. Both
ultrasonic and eddy current inspections are indirect measurements since the signals they
produce must be related to the nature and size of the flaws present. X-ray inspections are
more direct since they produce images of the flaws present. There is a radiation hazard
associated with X-ray inspections that is not present in either ultrasonic or eddy current
inspections, so this can limit the use of this method. Eddy current inspections also are
limited since they require the component being inspected to be electrically conducting and
only surface or near-surface flaws can be examined due to the limited depth of penetra-
tion of the eddy currents below the surface. Ultrasonic inspections have few material
constraints and can inspect components for both surface-breaking and interior flaws so
that ultrasonic NDE is one of the most commonly used inspection methods. For more
details on NDE inspections, see [10], [11], and [12].
 13.4 Stability 515

13.4 Stability
The previous sections have analyzed how structures fail either due to excessive static loads or
due to fracture or fatigue. Another type of failure is when a structure loses stability and can
suddenly experience unwanted displacements, and, in some cases, collapse. Most elementary
strength of materials texts discuss this type of loss of stability by considering the simply
supported member shown in Fig. 13.24a that is placed under a uniaxial load, P. As the load
increases, the member compresses without any deflection in the z-direction, but at a certain
critical load, Pcr , the member develops a sudden z-displacement as illustrated by the dashed
line in Fig. 13.24a. The appearance of this sudden lateral deflection is called buckling. As can
be seen from Fig. 13.24a, when buckling occurs the member acts as a deflected beam where
the bending is coupled to the axial load. To examine this type of behavior in general,
consider a small element of a deflected beam as shown in Fig. 13.24b. Although the axial
load is constant, the shear force and bending moment can vary in the beam, as shown, and
we have also allowed for the possibility of a distributed load/unit length, qz ðxÞ. From vertical
force equilibrium we find

dV z
Vz þ dx  V z þ qz dx ¼ 0
dx (13.27)
dV z
! ¼ qz
dx
which is the same relation as found in an ordinary beam problem. From moment equilib-
rium, however, we have, summing moments about point O, as shown,
  
dM y dV z dx
My þ dx  M y  V z þ dx dx  PðdwÞ þ qz dx ¼0
dx dx 2 (13.28)
dM y dw
! ¼ Vz þ P
dx dx
where we have neglected all the higher-order terms in Eq. (13.28) since they vanish as dx
goes to zero. Equation (13.28) show the coupling between the bending moment and the axial
load in the buckled beam. Differentiating Eq. (13.28) and combining it with Eq. (13.27)
we find

z qz dM y Figure 13.24 (a) A simply

My + dx supported member under a
My dx
uniaxial load. (b) A free body
P
P diagram of a deflected member.
O
P x dx dw
w dVz w+ dx
Vz Vz + dx
L dx

(a) (b)
516 Material Failure and Stability

d 2M y d 2w
¼ q z þ P (13.29)
dx2 dx2
From the moment–curvature relation

d 2w
M y ¼ EI (13.30)
dx2
Placing this relation into Eq. (13.29) gives the differential equation

d 4w 2
2d w q
4
þ k 2
¼ z (13.31)
dx dx EI
where k2 ¼ P=EI . If we consider cases where the distributed load is absent, Eq. (13.31) is a
homogeneous fourth order equation. Its general solution is
wðxÞ ¼ A cos ðkxÞ þ B sin ðkxÞ þ Cx þ D (13.32)

where the constants ðA; B; C; DÞ are determined from the boundary conditions.
Let’s consider the simply supported case of Fig. 13.24a where these boundary conditions
are


M y ð0Þ ¼ 0 ! d 2 w=dx2
x¼0 ¼ 0
wð0Þ ¼ 0

(13.33)
M y ðLÞ ¼ 0 ! d 2 w=dx2
x¼L ¼ 0
wðLÞ ¼ 0

The first two of these boundary conditions give A ¼ D ¼ 0 while the second two boundary
conditions give C ¼ 0 and the condition
"rffiffiffiffiffiffi #
P
B sin ðkLÞ ¼ B sin L ¼0 (13.34)
EI

There are two ways in which we can satisfy Eq. (13.34). The first solution is B ¼ 0, which is
certainly a legitimate solution since it says that the beam does not deflect in the z-direction. If
B is not zero, we can divide by it and we can only satisfy the boundary condition if
"rffiffiffiffiffiffi #
P
sin L ¼0 (13.35)
EI

which yields a series of buckling loads given by

nπ2
Pn ¼ EI ðn ¼ 1; 2; . . .Þ (13.36)
L
The smallest (n ¼ 1) of these possible loads occurs first, so it is the critical buckling load, Pcr
(also called the Euler buckling load), given by
 13.4 Stability 517

 π 2
Pcr ¼ EI (13.37)
L
and the associated buckling deflection is
πx
wðxÞ ¼ B sin (13.38)
L
where the amplitude, B, is arbitrary.
From this analysis it is clear that the critical load depends on the boundary conditions so
that if one considered other cases the critical load would be different. We will not consider
such cases here but they can be explored in the problems at the end of this chapter.
Buckling of long, slender members is an example of a bifurcation instability (see
Fig. 13.25a) where, on a load path (a plot of generalized force versus generalized displace-
ment), at the critical load there can be more than one load paths possible. Bifurcation
stability, however, is only one way in which stability can be lost. One can have a limit-load
instability, where the load reaches a maximum value (or, rarely, a minimum value) beyond
which the system is unstable. Some shell structures and structures that are sensitive to small
imperfections of geometry, loading, or boundary conditions can have a limit-load behavior
similar to that seen in Fig. 13.25b. Another way in which stability can be lost is via snap-
through buckling where the system reaches a limit load and then jumps to a new, stable
equilibrium position, as seen in Fig.13.25c. Slender arches and shallow shell structures can
exhibit such snap-through behavior. Analysis of limit loads and snap-through buckling
typically requires a nonlinear stress analysis, because if one assumes a linear behavior (see
the dashed lines in Figs. 13.25b, c), these instabilities are not captured by the analysis.
Similarly, in bifurcation types of instability the behavior of the system at loads beyond the
critical load is nonlinear. Thus, most elementary and advanced strength of materials texts
rarely examine limit-load instabilities, snap-through buckling, or postcritical load behavior
for buckling. However, if one considers simpler deformable structures composed of systems
of rigid bodies and springs, one can easily analyze these types of instabilities, as we will
now discuss.

P P P Figure 13.25 (a) A bifurcation type of

unstable instability. (b) A limit-load type of
unstable
Pcr instability. (c) A snap-through buckling
stable type of instability. Here P represents a
Pcr Pcr generalized force and δ represents a
stable generalized displacement.
stable unstable
stable
G G G
(a) (b) (c)
518 Material Failure and Stability

Figure 13.26 A ball at rest on a surface at either an unstable z

position, A, or a stable position, B.

z = f ( x)

B
x

As illustrated by the buckling bifurcation problem just considered, one can obtain the
critical load from an examination of equilibrium. Energy methods, however, provide an
alternative tool for both obtaining the critical load and for considering nonlinear behavior.
Consider, for example, the simple case where the load paths can be described in terms of a
single generalized load, P (either a force or a moment), and a single generalized displacement
variable, δ (either a displacement or a rotation angle). In these cases the criterion for the
stability of the system is that the second derivative of the total potential energy of the system,
Π, with respect to the deformation variable, must be positive, i.e., d 2 Π=dδ2 > 0, where
ΠðδÞ ¼ V ðδÞ  PuðδÞ, where V ðδÞ is the potential energy of the deformable bodies present
(which is same as the strain energy discussed in Chapter 8) and u is the displacement at the
load P in the direction of P. Also, at equilibrium one has dΠ=dδ ¼ 0, which is an alternate
way to guarantee that equilibrium is satisfied. [Note: P is held fixed when these derivatives
are taken.] These criteria make sense since they are a generalization of the stability criteria
seen, for example, when we consider the stability of a ball on a curved surface, as shown in
Fig. 13.26. In that case the surface z ¼ f ðxÞ is proportional to the total potential energy of
the ball due to gravity and one has equilibrium at points A and B when df =dx ¼ 0, but the
equilibrium is unstable at A and stable at B where d 2 f =dx2 < 0 and d 2 f =dx2 > 0, respect-
ively. The following example will illustrate the use of an energy method to analyze stability.

Example 13.5 Stability Analysis via Equilibrium or Energy Methods

Consider a rigid bar of length L that is allowed to rotate about a smooth pin at O and is
attached to a linear torsional spring at O, which supplies a moment, M 0 , that is proportional
to the angle of rotation, i.e., M 0 ¼ k θ θ, where kθ is the spring constant (see Fig. 13.27a).

(a) From equilibrium of moments about the smooth pin at O, determine all the allowable
values for P. Show that there are multiple load paths that the load can follow and that
there is a bifurcation point at which load paths cross. Determine the force P ¼ Pcr at the
bifurcation point. Show that you obtain the same results by setting dΠ=dθ ¼ 0, where Π
is the total potential energy and θ is the angle of rotation of the bar about the pin at O.
(b) By examining the second derivative of the total potential energy, d 2 Π=dθ2 , determine the
stability of all the possible load paths. What is the physical behavior of the system when
P > Pcr ?
 13.4 Stability 519

P P u L  L cos T Figure 13.27 (a) A rigid bar of length L that is

pinned at O and is attached at O to a linear
P torsional spring whose spring constant is k θ .
A vertical force, P, acts on the bar. (b) Free body
L sin T diagram of the deflected bar.
L
L cos T
torsional
spring rigid T
M0 kT T
kT
O
(a) (b)

P Figure 13.28 The load paths for the bar of Fig. 13.27a, showing a
unstable bifurcation behavior.

stable stable
From a free body diagram of the bar (Fig. 13.27b) and summing
kT
Pcr moments about O we have
L
stable X
M0 ¼ 0
(13.39)
T PL sin θ þ kθ θ ¼ 0

This equation has two solutions. The first is θ ¼ 0 and the load P is
arbitrary. The second solution is P ¼ kθ θ=L sin θ. These solutions are two intersecting lines
so there is a bifurcation behavior as seen in Fig. 13.28. The critical load at the intersection
point is Pcr ¼ k θ =L. The total potential energy of the bar is
1
Π ¼ kθ θ2  PLð1  cos θÞ (13.40)
2
Setting the derivative equal to zero, we find dΠ=dθ ¼ kθ θ  PL sin θ ¼ 0, which gives us the
same result for the critical load as equilibrium. To examine the stability of the under various
paths, we must examine the second derivative of the total potential energy, which is

d 2Π
¼ PL cos θ þ kθ (13.41)
dθ2
Along the θ ¼ 0 path d 2 Π=dθ2 ¼ PL þ k θ , which gives d 2 Π=dθ2 > 0 when P < Pcr and
d 2 Π=dθ2 < 0 when P > Pcr , yielding the stability labels shown in Fig. 13. 28 for the vertical load
path. For the curved path P ¼ kθ θ=L sin θ ¼ Pcr θ= sin θ and d 2 Π=dθ2 ¼ LPcr ð1  θ= tan θÞ,
which is always positive since θ= tan θ < 1. Thus, the curved path is always stable, as shown
in Fig. 13.28. As the load P is increased from a value of zero, the stable vertical load path is
520 Material Failure and Stability

followed and the bar remains vertical until P reaches the critical load. For P > Pcr the vertical
load path is no longer stable so the bar must rotate right or left and follow a stable curved path.
Note that there is no unwanted or catastrophic behavior for P > Pcr so a bifurcation does not
always require some unwanted or catastrophic behavior.

There are additional stability problems at the end of this chapter that will allow you to
explore bifurcation, limit-load, and snap-through stability cases.
All the cases we have shown are static problems. However, time-varying systems can
exhibit dynamic instabilities. These are beyond what we will cover in this book, but you can
see some examples in [13].

13.5 PROBLEMS

P13.1 A circular shaft of diameter d is subjected to a combined loading of a bending

moment, M, and a torque, T. The material of the shaft has a yield stress σ y .
Determine an expression for the failure surfaces for this case in terms of M, T, and
d for the maximum shearing stress theory and the maximum distortional strain energy
theory. If the bending moment is M ¼ 10 kN-m, the torque is T ¼ 20 kN-m, and the
yield stress is σ y ¼ 300 MPa, determine the minimum allowable shaft diameter to
obtain a safety factor of 2 according to these two theories.
P13.2 For a thin plate in a state of plane stress, yielding is observed when σ xx ¼ 90 MPa,
σ yy ¼ 10 MPa, σ xy ¼ 30 MPa. (a) What is the corresponding yield stress in uniaxial
tension predicted by the maximum shear-stress theory and the maximum distortional
strain energy theory? (b) How do your answers change if σ xx ¼ 90 MPa, σ yy ¼ 30 MPa,
σ xy ¼ 40 MPa?
P13.3 The maximum distortional strain energy theory written in terms of the principal stresses
is given by Eq. (13.5) for a state of plane stress. What is the corresponding theory
 
written in terms of the stresses σ xx , σ yy , σ xy ? What is the corresponding theory written
in terms of the stresses, as given in a set of Cartesian coordinates, for a general state of
stress?
P13.4 A 1.0 m diameter thin wall cylindrical vessel carries an internal pressure p ¼ 3 MPa
and is made of a material with a uniaxial yield stress σ y ¼ 400 MPa. Determine the
wall thickness required to prevent yielding with a safety factor of 2.0 according to
the maximum shearing stress theory and the maximum distortional strain energy
theory.
P13.5 A material with a yield stress σ y ¼ 650 MPa, an ultimate strength σ u ¼ 800 MPa, and a
fatigue strength of 350 MPa at 107 cycles is subjected to a mean stress σ m ¼ 300 MPa
together with an alternating stress σ a ¼ 200 MPa. Will the part survive 107 cycles
according to the Goodman curve? Will it survive based on the more conservative
Soderberg curve?
 13.5 Problems 521

P13.6 A thin-wall cylindrical pressure vessel has a radius r ¼ 100 mm and a thickness t ¼ 5 mm.
The pressure in the vessel varies from zero to p. Using the effective (von Mises) stress
to calculate mean and alternating stresses, determine the pressure p so that the vessel
will fail after 108 cycles based on a Goodman curve. The ultimate strength of the
material is σ u ¼ 800 MPa and the fatigue strength at 108 cycles is σ fs ¼ 250 MPa.
P13.7 A bolt is subjected to an alternating axial force with maximum and minimum values
of ðF max , F min Þ, respectively. The ultimate static tensile stress of the material is σ u
and the fatigue strength for an alternating stress at the desired life is σ fs . Based on the
Goodman curve, show that the required cross-sectional area of the bolt, A, is

1 1 σ fs
A¼ F max  ðF max þ F min Þ 1 
σ fs 2 σu

P13.8 A small leaf spring of rectangular cross-section that has a width b ¼ 10 mm wide and
a length L ¼ 150 mm can be modeled as a simply supported beam subjected to a
center concentrated force, P, that varies from zero to 20 N. Using the Goodman
curve, determine the value of the height, h, needed for the spring if the fatigue
strength at the desired life is σ fs ¼ 350 MPa and the ultimate static tensile strength is
σ u ¼ 1200 MPa.
P13.9 Consider a material whose S–N curve is given by
S ¼ 800N 0:12

where S is the alternating stress in MPa and N is the number of cycles to failure.
If this material is placed in an operating environment where it operates for
250 000 cycles at an alternating stress level of 175 MPa and for 250 000 cycles at
150 MPa, will the material fail according to the Palmgren–Miner rule?
P13.10 A 100 mm wide, 10 mm thick plate made with a material having a fracture
pffiffiffiffi
toughness K IC ¼ 80 MPa m is subjected to a tensile load, P (Fig. P13.1a)), which

P Figure P13.1 (a) An axially loaded

cracked plate. (b) A cracked beam under
100 mm combined loading.
case (a) M M
case (b)
h
P a P
2a
a

P
(a) (b)
522 Material Failure and Stability

varies from zero to P ¼ 140 kN. Use the stress intensity configuration factors in
Fig. 13.16.
(a) Determine the size that a through-thickness centrally located fatigue crack must
grow to before failure by fast fracture occurs.
(b) How does your answer to (a) change if the crack is located on the edge of the
plate instead?
pffiffiffiffi
P13.11 A 30 mm wide plate that has a plane strain fracture toughness K IC ¼ 70 MPa m is
subjected to an axial tensile stress of 150 MPa. If an ultrasonic NDE method is
capable of finding a 3 mm long crack, is this testing method able to detect a surface
breaking flaw in the plate before it becomes critical? What is the safety factor for this
inspection? Note, however, that in a fatigue environment the frequency of inspection
is also important to ensure that a crack does not grow to a critical length
between inspections.
P13.12 A rectangular beam of height h ¼ 50 mm and width t ¼ 20 mm carries a combined
loading of P ¼ 10 kN and M ¼ 200 N-m (Fig. P13.1b). If the beam is made of a
pffiffiffiffi
material with a fracture toughness K IC ¼ 40 MPa m, what is the critical crack
length at which the beam fractures due to an edge crack, as shown. Use the stress
intensity configuration factors in Fig. 13.16 and Fig. 13.17b.
P13.13 A rolling container is supported by two parallel beams, each of length L ¼ 40 ft and
having rectangular cross sections of height h ¼ 10 in. and width t ¼ 1.0 in. The
weight of the container and its contents is W ¼ 15 000 lb. A crack of length a exists
at the bottom of one of the beams at its center as shown in Fig. P13.2.
pffiffiffiffiffiffi
(a) If the fracture toughness of the beam material is K IC ¼ 60 ksi in:, determine
the largest crack that could be tolerated at the configuration shown in Fig. P13.2
before unstable crack growth would occur.
(b) If a one-half inch long crack already exists in the beam at the location shown in
Fig. P13.2, how many times could the container be pushed across the beams
before the cracked beam fails? Assume the bending moment varies from zero to
a maximum value given in part (a) and assume the crack growth law is

Figure P13.2 A container traveling over a 10 ft 10 ft

cracked beam.

h
a

20 ft 20 ft
 13.5 Problems 523

da
¼ AðΔK Þm
dN
where A ¼ 1:0  1018 in:=cycle and m ¼ 3. Also, assume the configuration
factor in the stress intensity expression is equal to 1.122, independent of
crack size.
(c) If an NDE inspection is performed every 10 000 crossings, what size of crack
must be reliably detected if one wants to guarantee that the beam will not fail
before the next inspection? Use the same crack growth law and configuration
factor in (b) and the critical flaw size from (a).
P13.14 A rigid bar of length L is attached to a linear spring whose other end rides in a
vertical smooth slot so that the spring always remains horizontal. The spring
constant is k. A vertical load, P, is applied to the bar, as shown in Fig. P13.3a.
(a) From equilibrium of moments about the smooth pin at O, determine all the
allowable values for P. Show that there are multiple load paths that the load can
follow and that there a bifurcation point at which load paths cross. Determine
the force P ¼ Pcr at the bifurcation point. Sketch your results. Show that you
obtain the same results by setting dΠ=dθ ¼ 0, where Π is the total potential
energy and θ is the angle of rotation of the bar about the pin at O.
(b) By examining the second derivative of the total potential energy, d 2 Π=dθ2 ,
determine the stability of all the possible load paths. What is the physical
behavior of the system when P > Pcr ?
P13.15 A rigid bar of length L is attached to a linear spring whose other end rides in a
vertical smooth slot so that the spring always remains horizontal. The spring
constant is k. A vertical load, P, is applied to the bar eccentrically with a small
eccentricity, e, as shown in Fig. P13.3b.
(a) From equilibrium of moments about O, determine the equilibrium values of P as
a function of the angle of rotation, θ, about O. Show that you obtain the same

P Figure P13.3 (a) A vertical force P

k P k e
acts on rigid bar of length L,
which is attached to a linear
spring. (b) The same configuration
but where the load is applied
L eccentrically to the bar, where e is
L
the small eccentricity.
smooth
smooth
rigid rigid
O O

(a) (b)
524 Material Failure and Stability

results by setting dΠ=dθ ¼ 0, where Π is the total potential energy and θ is the
angle of rotation of the bar about the pin at O. Plot the normalized load, P=kL,
versus θ for the normalized eccentricity e=L ¼ 0:1 and for 1:5 < θ < 1:5
radians (because the slope of a portion the load curve is very large, also limit
the plot range to 0 < P=kL < 2).
(b) For P=kL ¼ 0:5, e=L ¼ 0:1 plot the normalized total potential energy, Π=kL2 ,
over the same angles used in part (a) and determine the equilibrium positions
and if they are stable or unstable. Use that information to describe physically
what happens in this problem as the load P increases and justify why this type of
instability is called a limit-load stability. By making the eccentricity very small,
can you see the relationship between this behavior and the bifurcation behavior
of this system when e ¼ 0?
P13.16 Consider a shallow truss, as shown in Fig. P13.4a, which is composed of two
uniaxially loaded members that are pinned at their ends. Because of symmetry, the
pin connecting the two members experiences a displacement, u, as shown in Fig.
P13.4b and the lengths of the members changes from L0 to L. If we assume linear
elastic behavior for the members, then they behave like linear springs, where
F ¼ kðL  L0 Þ. [Note: the minus sign is present because ðL  L0 Þ is the elongation
and we have shown the members in Fig. P13.4c to be compressive.] The spring
constant k ¼ EA=L0 , where A is the cross-sectional area of each member and E is
Young’s modulus.
(a) From equilibrium, obtain an expression for the nondimensional load, P=P0 in
terms of the nondimensional displacement, δ ¼ u=d, and a=d, where P0 ¼ ka2 =d.
Show that you obtain the same result by setting dΠðδÞ=dδ ¼ 0, where Π is the total
potential energy. Plot this nondimensional load versus nondimensional displace-
ment from u=d ¼ 0 to u=d ¼ 2:5 for a shallow truss where d=a ¼ 0:1. As the load
increases from zero, what is the first maximum load, P ¼ Pcr , reached?

P Figure P13.4 (a)

A shallow truss. (b) The
L0 L0 deformation. (c) A free
body diagram of
d P
the forces.

a a F F

(a) T T
u
L0 (c)
L0
T L L
T
(b)
 References 525

z z Figure P13.5 (a) An axially loaded

cantilever beam. (b) An axially loaded
P fixed–fixed beam.
P
x x
L
L
(a) (b)

(b) Plot the total potential energy, ΠðδÞ, for the same values of displacement in part
(a) for P ¼ 0:25P0 . Identify the equilibrium positions and state if they are stable
or unstable. If these stability results are characteristic of the portions of the
PðδÞ=P0 curve on which they occur, describe physically what happens when the
truss is loaded from P ¼ 0 to P > Pcr . This type of loss of stability behavior is
called snap-through buckling. It can occur, for example, if the loading on a
shallow shell roof structure becomes too large. If after snap-through buckling
occurs the load is removed, does the truss return to its original unloaded
position?
P13.17 Consider an axially loaded cantilever beam, as shown in Fig. P 13.5a. Determine the
buckling loads and the corresponding buckled beam shapes. Show that, for the
lowest-order mode, we can use the Euler buckling load expression for a simply
supported column with the length replaced by an effective length for the cantilever.
What is this effective length? If the cantilever has a rectangular cross section of base
b and height h, where b < h, what area moment should be used to determine the
buckling load?
P13.18 Repeat problem P13.17 for the fixed-fixed beam of Fig. P13.5b, solving for the
lowest buckling load, corresponding mode, and effective length.

References
1. E. Zahavi, Fatigue Design – Life Expectancy of Machine Parts (Boca Raton, FL: CRC Press,
1996)
2. A. F. Liu, Structural Life Assessment Methods (ASM International, CRC Press, 1998)
3. B. Farahmoud, Fatigue and Fracture of High-Risk Parts (London, UK: Chapman and Hall, 1997)
4. H. Tada, P. C. Paris, and G. R. Irwin, The Stress Analysis of Cracks Handbook (ASME Press,
2000)
5. D. P. Rook and D. J. Cartwright, Compendium of Stress Intensity Factors, (H.M.S.O, 1976)
6. G. Shih, Ed., Experimental Evaluation of Stress Concentration and Stress Intensity Factors,
(Switzerland: Springer, 1981)
7. J. F. Knott, Fundamentals of Fracture Mechanics, (Oxford, UK: Butterworths, 1973)
8. J. M. Barsom and S. T. Rolfe, Fracture and Fatigue Control in Structures, 3rd edn (ASTM,
1999)
526 Material Failure and Stability

9. T. L Anderson, Fracture Mechanics, 4th edn (Boca Raton, FL: CRC Press, 2017)
10. L. W. Schmerr, Jr., Fundamentals of Ultrasonic Nondestructive Evaluation – A Modeling
Approach, 2nd edn (Switzerland: Springer, 2016)
11. N. Bowler, Eddy-Current Nondestructive Evaluation (Switzerland: Springer, 2019)
12. R. Halmshaw, Industrial Radiology: Theory and Practice (Switzerland: Springer, 2012)
13. L. W. Schmerr, Engineering Dynamics 2.0 – Fundamentals and Numerical Solutions (Switzerland:
Springer, 2019)
APPENDIX A
Cross-Section Properties

Descriptions of bending and torsional deformations involve numerous properties of the

member cross-section. As seen in Chapter 10, the bending cross-sectional properties
included the location ðY c , Z c Þ of the centroid and the second area moments and the


mixed second area moment I yy , I zz , I yz , as calculated with respect to centroidal coord-
inates ðy; zÞ. To describe the shear stresses and the shear center location ðeY , eZ Þ for the
bending of thin sections the principal sectorial area function, ωP (see Appendix C), and


the sectorial products of areas I yω , I zω also appeared, as well as the first area moments
ðQy , Qz Þ, which are closely related to the first area moments ðQyA , QzA Þ of the entire
cross-sectional area, A (see Appendix C). For torsion, in Chapter 11 we saw that the
effective polar area moment, J eff , and the torsional warping constant, J ω , were two
important quantities. The location of the center of twist/shear center, ðeY , eZ Þ, again was
needed as was the principal sectorial area function, ωP , the sectorial products of areas,


I yω , I zω , as well as the first area moments ðQyA , QzA Þ. All of the quantities that are
associated with the sectorial area function are discussed extensively in Appendix C.
Figure A.1 lists some of these important torsional properties for common thin cross-
sectional areas. For bending, we have not listed approximate expressions for thin
sections. Instead, we have written a MATLAB® function cross_section that calculates
the location of the centroid and the area moments and mixed area moments of a cross-
section composed of rectangles lying parallel to a (Y, Z) coordinate system. This allows
one to calculate these properties for all of the cross-sections shown in Fig. A.1 and Fig.
A.1 (continued) whether or not they are thin. The calling sequence for the MATLAB ®
function cross_section is:
[Yct, Zct, Iy, Iy, Iyz] = cross_section( Ly, Lz, Yc, Zc);

where ðYc; ZcÞ are vectors containing the location of the centroids of the rectangular parts in
a ðY ; Z Þ coordinate system, and ðLy; LzÞare the lengths of the rectangular parts along the
ðY ; Z Þ axes. The outputs of the function are the coordinates of the centroid of the total
cross-section, ðYct; ZctÞ, and the area moments and mixed area moment, relative to a set of
centroidal axes, ðy; zÞ. [Note that, for a thin section, all the dimensions are typically given in
terms of distances along the centerlines of the sections, while the function cross_section uses
lengths of the sides of individual rectangular components. Thus, to compare the exact results
from cross_section with thin cross-section values, one must relate the two different descrip-
tions of the geometry. An example of this process is given in Section A.3.] Figure A.2 shows
a number of common cross-sections for which a MATLAB® script (named sections) has

527
528 Appendix A

Figure A.1 Torsional Z

tf
properties of thin b
if t f tw t
sections. S is the 2bt 3f  htw3
J eff
location of the 3 t 3 2b  h
shear center. tw J eff
h t f h 2b 3 3
S JZ
eZ 24
tf h
eY 0, eZ
Y 2

b
Z
tf if t f tw t
b1 b1  b2 t 3f  htw3
J eff
3
t f h 2 b13b23 t 3 b1  b2  h
tw JZ J eff
S 12 b13  b23 3
h
eZ b13 h
eY 0, eZ
tf b  b22
1
2

Y
b2

Z b if t f tw t
tf
2bt 3f  htw3
J eff t 3 2b  h
3 J eff
tw 3
S t f b3 h 2 3bt f  2htw
h JZ tb3 h 2 3b  2h
12 6bt f  htw JZ
eZ tf 12 6b  h
eY 3b 2t f h 3b 2 h
eY , eZ eY , eZ
6bt f  htw 2 6b  h 2
Y
b

been written that uses the function cross_section and allows one to calculate the centroids
and area moments directly from the given dimensions (which again are measured, as shown
in Fig. A.2, from the lengths of the sides, not along the centerlines). A description of the
script and a sample calculation are given in Section A.3.
In the bending of unsymmetrical sections, several relationships for transforming second
area moments and mixed second area moments were discussed in Chapter 10. These
included the parallel axis theorem and the transformation to principal coordinates of the
cross-section. In this appendix we will briefly outline those relations.
 Cross-Section Properties 529

Figure A.1 (cont.) Z

b if t f tw t
Torsional properties
tf
of thin sections. S is 2bt  ht3
f
3
w
the location of the
J eff
3
shear center. tw 3 2
bh t 3 2b  h
h JZ 2 J eff
S 12 2b  h 3
eZ
u ª¬ 2t f b 2  bh  h 2  3twbh º¼ tb h b  2h
3 2
JZ
tf 12 2b  h
h
eY 0, eZ
Y 2
b
Z

if t f tw t

tw bt 3f  htw3 b  h t3
J eff J eff
h 3 3
tf JZ 0
eY eZ 0
S
Y
b

t Z
2t 3 b  h
J eff
3
h tb 2 h 2 h  b
2
S JZ
eZ 24 b  h
h
eY 0, eZ
Y 2
b

A.1 Parallel Axis Theorem

In calculating the second area moments and mixed second area moments for composite
cross-sections, i.e., those that are composed of combinations of simple shapes, it is often
convenient to determine how the values of those area moments change when we use
coordinates relative to different axes that are translated (and not rotated) with respect to
each other. Thus, consider a cross-sectional area A, as shown in Fig. A.3, and two sets of
axes ðy0 , z0 Þ and ðy; zÞ that are parallel to each other and where the ðy0 , z0 Þ axes have their
origin at the centroid, C, of A. If we let ðyc , zc Þ be the coordinate locations of C as measured
in the ðy; zÞ set of axes then from the definition of the area moment, I zz , we have
530 Appendix A

Figure A.2 Various cross-sections. (a) An Z Z

c
L-section. (b) A J-section (which includes
t3
a C-section as a special case).
(c) A T-section. (d) A Z-section.
(e) An I-section. All dimensions are t2 t2
measured along the section sides as h h
shown. Note that in (d) the Z-axis is not t1 t1
along the centerline of the vertical part of
the section as it is in the symmetrical Y
sections of (c) and (e). Y
b b
(a) (b)
Z c Z
t3

t2
h t2 h
t1 t1

Y
Z Y
b c b
(c) (d)

t3
t2 h
t1

Y
b
(e)

zc Figure A.3 Two parallel coordinate systems, where the ðy0 , z0 Þ axes have their
origin at the centroid of the area A.
z
ð ð
yc
I zz ¼ y2 dA ¼ ðyc þ y0 Þ dA
2
A
yc ð 
C (A.1)
y2c þ 2yc y0 þ ðy0 Þ dA
2
¼

zc ¼ ðI z0 z0 Þc þ Ay2c

where the middle integral vanishes since y0 is measured from the

y centroid. In similar fashion one can show
 Cross-Section Properties 531


I yy ¼ I y0 y0 c þ Az2c

 (A.2)
I yz ¼ I y0 z0 c þ Ayc zc

Equations (A.1) and (A.2) represent the parallel axis theorem for the area moments.

A.2 Area Moments in Rotated Coordinates and Principal Axes

The values of the area moments also change when we use a new rotated set of coordinates to
describe their values. Figure A.4 shows a cross-sectional area as measured in a set of ðy; zÞ
axes and a set of rotated ðy0 , z0 Þ axes. One can again use the definitions of these area
moments to relate their values. We will not give the details here but simply show the end
result, namely,

I yy þ I zz I yy  I zz
I y0 y0 ¼ þ cos ð2θÞ  I yz sin ð2θÞ
2 2
I yy þ I zz I yy  I zz
I z0 z0 ¼  cos ð2θÞ þ I yz sin ð2θÞ (A.3)
2 2
I yy  I zz
I y0 z0 ¼ sin ð2θÞ þ I yz cos ð2θÞ
2

These area moments transform like the stresses do for a case of plane stress where we have


only the nonzero stresses σ yy , σ zz , σ yz . From the general stress transformation, Eq. (2.38),
one can show that we find
σ yy þ σ zz σ yy  σ zz
σ y0 y0 ¼ þ cos ð2θÞ þ σ yz sin ð2θÞ
2 2
σ yy þ σ zz σ yy  σ zz
σ z0 z0 ¼  cos ð2θÞ  σ yz sin ð2θÞ (A.4)
2 2
σ yy  σ zz
σ y0 z0 ¼ sin ð2θÞ þ σ yz cos ð2θÞ
2
so the transformations are identical if we use the correspond-




ences σ yy , σ zz , σ yz ! I yy , I zz ; I yz and σ y0 y0 , σ z0 z0 , σ y0 z0 !


z I y0 y0 , I z0 z0 ; I y0 z0 . This correspondence also means that in a
principal set of coordinates, the mixed area moment will be
zc yc zero and the area moments will have their largest and smallest


values I p1 , I p2 . As in the plane-stress case (see Eq. (2.67)), we
have these principal area moments given by

T
O y Figure A.4 An area as measured in a set of ðy; zÞ axes and a rotated set of
A
ðy0 , z0 Þ axes.
532 Appendix A

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
I yy þ I zz I yy  I zz 2
I p1 , I p2 ¼  þ I 2yz (A.5)
2 2

This correspondence between stresses and area moments also means that we can use the
MATLAB® function eig to calculate the principal area moments and their principal direc-
tions as we will see in the following example.

Example A.1 Principal Area Moments and Principal Directions

Consider the thin L-shaped section shown in Fig. A.5a, which has a constant thickness
t ¼ 0.01 in. We have the centroid C located at

Y 1 A1 þ Y 2 A2 ð0Þð0:04Þ þ ð1Þð0:02Þ 1
Yc ¼ ¼ ¼ in:
A1 þ A2 0:04 þ 0:02 3
(A.6)
Z 1 A1 þ Z 2 A2 ð0Þð0:02Þ þ ð2Þð0:04Þ 4
Zc ¼ ¼ ¼ in:
A1 þ A2 0:04 þ 0:02 3

and for the area moments and mixed area moments about the centroid
"  2 # "   #
4 ð0:01Þð4Þ3 4 2
I yy ¼ 0 þ ð0:02Þ  þ þ ð0:04Þ 2  ¼ 0:1067 in:4
3 12 3
"   # "  2 #
ð0:01Þð2Þ3 1 2 1 (A.7)
I zz ¼ þ ð0:02Þ 1  þ 0 þ ð0:04Þ  ¼ 0:0200 in:4
12 3 3
     
1 4 1 4
I yz ¼ 0 þ ð0:02Þ 1   þ 0 þ ð0:04Þ  2 ¼ 0:0267 in:4
3 3 3 3

Figure A.5 (a) A thin L-shaped cross-section. The

Z
dimensions shown are measured along the
centerlines of the sections. (b) The principal z
0.01 in.
coordinate axes (yp, zp) at the centroid C. zp
z yp
4 in. Yc ez
Tp
y y
C C ey
Zc

Y
2 in.

(a) (b)
 Cross-Section Properties 533

If we calculate the values in Eq. (A.7) and place them in a 2  2 matrix with the values
I yy I yz
I yz I zz
we can then use the eig function to calculate the principal values and principal directions. We
have in MATLAB®
Iyy = .02*(4/3)^2 +(.01)*4^3/12 +.04*(2-4/3)^2
Iyy = 0.1067

Izz = .01*8/12 +.02*(1-1/3)^2 +.04*(1/3)^2

Izz = 0.0200
Iyz = .02*(1-1/3)*(-4/3) +.04*(-1/3)*(2-4/3)
Iyz = -0.0267
I = [ Iyy -Iyz; -Iyz Izz]
I = 0.1067 0.0267
0.0267 0.0200
[pdirs, pvals] = eig(I)
pdirs = 0.2723 -0.9622
-0.9622 -0.2723
pvals = 0.0125 0
0 0.1142

We see that the two principal values along the principal (yp, zp) axes are I yp ¼ 0:1142 in:4 and
I zp ¼ 0:0125 in:4 , respectively, since if you examine the unit vectors in the columns of the
pdirs matrix you will see that the corresponding columns of this matrix are in the negative yp
and zp directions. Thus, the components of unit vectors along the positive yp and zp axes are


eyp ¼ 0:9622ey þ 0:2723ez and ezp ¼ 0:2723ey þ 0:9622ez , respectively, where ey , ez are
unit vectors along the ðy; zÞ axes. We can find the angle θp shown in Fig. A.5b via
atand(.2723/.9622)

ans = 15.8014

so that θp ¼ 15:8 , approximately.

A.3 Calculating Centroids and Area Moments in MATLAB®

Consider first the L-shaped cross-section shown in Fig. A.6a. We will decompose the section
into the two rectangles shown in Fig. A.6b. The lengths of these rectangles and the locations
of their centroids in the ðY ; Z Þ coordinates are as shown in Table A.1.
534 Appendix A

Table A.1 The properties of the rectangular parts of the L-shaped section shown in Fig. A.6b

Rectangle Ly Lz Yc Zc

(1) 70 10 45 5
(2) 10 120 5 60

Fig. A.6 (a) An L-shaped cross- Z

Z
section whose centroid is located at
C. (b) The cross-section divided into 10
two rectangles, where the dimensions
thickness = 10 mm
(2)
and centroids of the rectangles are z
given in the (Y, Z) coordinates. Yc
120
120 (5,60)
C y
Zc
(45,5)
10
Y
80 70 (1) Y

(a) (b)

We can place these values into vectors and calculate the cross-sectional area properties in
MATLAB®, giving:
Ly = [70 10];
Lz = [10 120];
Yc = [45 5];
Zc = [5 60];
[yc, zc, Iy, Iz, Iyz] = cross_section(Ly, Lz, Yc, Zc)
yc = 19.7368
zc = 39.7368
Iy = 2.7832e+06
Iz = 1.0032e+06
Iyz = -9.7263e+05

so that the location of the centroid of the entire cross-section is at Y c ¼ 19:74 mm,
Z c ¼ 39:74 mm and the area moments are I yy ¼ 2:78  106 mm4 , I zz ¼ 1:00  106 mm4 .
The mixed area moment is I yz ¼ 9:73  105 mm4 .
We can also use the cross_section function to check on the accuracy of using the thin wall
approximations that are customarily used when dealing with thin sections. Consider, for
example, the thin L-shaped section considered in Fig. A.5a. The dimensions in that figure
were based on distances as measured along the centerline of the cross-section and the
distances to the centroid were also calculated from a set of axes along the centerlines, which
 Cross-Section Properties 535

Zc Z Fig. A.7 (a) Centerline dimensions used

for the L-section of Fig. A.5a as well as
0.01 the distances to the centroid C as
thickness = 0.01 in. measured from those centerlines.
(2)
z (b) The exact dimensions of the cross-
Ycc section when it is treated as two
4
4.005 (0.005, 2.0025) rectangles as well as the centroid
C y
locations of those rectangles as
Z cc measured from the ðY ; Z Þ axes shown.
(1.0075, 0.005)
Yc 0.01
2 1.995 (1) Y
(a) (b)

we will call ðY 0 , Z 0 Þ. The exact dimensions of two rectangles making up the cross-section as
well as the ðY ; Z Þ axes used for locating the centroid are shown in Fig. A.7b. Using those
exact dimensions, we find in MATLAB®:
Ly = [ 1.995 0.01];
Lz = [ 0.01 4.005];
Yc = [ 1.0075 0.005];
Zc = [ 0.005 2.0025];

[Yct, Zct, Iy, Iz, Iyz] = cross_section(Ly, Lz, Yc, Zc)

% centroid locations from the (Y, Z) axes
Yct = 0.3383
Zct = 1.3383
% area moments for centroidal coordinates
Iy = 0.1067
Iz = 0.0200
Iyz = -0.0267
% calculate centroid locations from the centerlines to compare
% with thin section results
Ycp = Yct -.005
Ycp = 0.3333
Zcp = Zct - .005
Zcp = 1.3333

where Ycp ¼ Y 0c , Zcp ¼ Z 0c . Thus, we see that as measured from the centerlines the centroid
coordinates are Y 0c ¼ 0:3333 in:, Z 0c ¼ 1:3333 in:, and I yy ¼ 0:1067 in:4 , I zz ¼ 0:0200 in:4 ,
I yz ¼ 0:0267 in:4 . These are in agreement with the thin cross-section results of the previous
section (see Eq. (A.6) and Eq. (A.7)), showing the validity of the use the thin-section
approximations for this cross-section. Generally, the thin-section approximations are used
536 Appendix A

to make the calculations, when done by hand, more manageable and the expressions for the
area moments less complex. With the availability of functions like cross_section, however,
there is less need to use the thin-section approximations. We have also, as mentioned
previously, developed a MATLAB® script, sections, that makes these calculations even
easier. Five cross-sections are shown in Fig. A.2 that can be described in terms of up to
three lengths ðb; h; cÞ and three thicknesses ðt1 , t2 , t3 Þ. The cross-sections are called L-, J-, T-,
Z-, and I-sections. If one enters a string ‘L’, ‘J’, ‘T’, ‘Z’, or ‘I’ in a variable called type to
identify the type of section and then enters the necessary length and parameters for the
chosen section into the command window, we can simply run the script sections. That script
uses the function cross_section and returns the location of the centroid of the section in the


ðY ; Z Þ coordinates shown in Fig. A.2 as well as the area moments I yy , I zz , I yz with respect
to centroidal axes. For example, consider the L-section of Fig. A.6. In MATLAB® we have
clear
type = ' L';
b = 80;
h = 120;
t1 = 10;
t2 = t1;
sections
yc = 19.7368
zc = 39.7368
Iy = 2.7832e+06
Iz = 1.0032e+06
Iyz = -9.7263e+05

These values agree with the ones calculated previously.

 APPENDIX B
The Beltrami–Michell Compatibility Equations

B.1 Compatibility Equations for Stresses

The compatibility relations discussed in Chapter 4 are relations between the strains in a
body. However, if we use the stress–strain relations and the equilibrium equations, we can
instead write those compatibility relations in terms of the stresses. For a homogeneous,
isotropic elastic solid these compatibility equations are called the Beltrami–Michell rela-
tions. From Eq. (4.50) the 81 compatibility equations are

∂2 eik ∂2 ejl ∂2 ejk ∂2 eil

þ   ¼ 0 ði; j; k; l ¼ 1; 2, 3Þ (B.1)
∂xj ∂xl ∂xi ∂xk ∂xi ∂xl ∂xj ∂xk

If we use the stress–strain relations (see Eq. (5.34)) in these equations, we have the corres-
ponding 81 equations

∂2 σ ik ∂2 σ jl ∂2 σ jk ∂2 σ il
þ  
∂xj ∂xl ∂xi ∂xk ∂xi ∂xl ∂xj ∂xk


ν ∂2 Θ ∂2 Θ ∂2 Θ ∂2 Θ
¼ δik  δjk þ δjl  δil ði; j; k; l ¼ 1; 2, 3Þ (B.2)
1þν ∂xj ∂xl ∂xi ∂xl ∂xi ∂xk ∂xj ∂xk

where δij is the Kronecker delta (see Eq. (2.34)) and

X
3
Θ¼ σ nn (B.3)
n¼1

If we set k ¼ j and sum over j the resulting equations from 1 to 3, we have nine equations
which actually represent six equations because of the symmetry of the stresses:
X3  2 
∂ σ ij ∂2 σ jl ∂2 σ il ∂2 Θ
þ  
j¼1
∂xj ∂xl ∂xj ∂xi ∂xj ∂xj ∂xi ∂xl


ν X 3
∂2 Θ ∂2 Θ ∂2 Θ ∂2 Θ
¼ δij  δjj þ δjl  δil ði; l ¼ 1; 2, 3Þ (B.4)
1 þ ν j¼1 ∂xj ∂xl ∂xi ∂xl ∂xi ∂xj ∂xj ∂xj

These equations are linear combinations of the original 81 equations so they can serve as the
reduced system to represent compatibility. Again, these equations are not all independent, so
that in fact there are only really three independent compatibility relations. If we use the
relations

537
538 Appendix B

X3
∂2 σ il
¼ r2 σ il
j¼1
∂x j ∂x j

X3
∂2 Θ
¼ r2 Θ (B.5)
j¼1
∂x j ∂xj

X
3
δjj ¼ 3
j¼1

Eq. (B.4) can be rewritten as

X3  2 

∂ σ ij ∂2 σ jl ∂2 Θ ν ∂2 Θ
þ   r σ il ¼
2
δil r Θ þ
2
ði; l ¼ 1; 2, 3Þ (B.6)
j¼1
∂xj ∂xl ∂xi ∂xj ∂xi ∂xl 1þν ∂xi ∂xl

We can express Eq. (B.6) in a more compact form by using the equations of equilibrium (see
Eq. (3.10)) written as
X
3
∂σ ij
þfi ¼ 0 ði ¼ 1; 2, 3Þ (B.7)
j¼1
∂xj

where f i are the body force components. Then we have

X3
∂2 σ ij ∂f
¼ i
j¼1
∂x j ∂x l ∂x l
(B.8)
X3
∂2 σ jl X3
∂2 σ lj ∂f
¼  l ði; l ¼ 1; 2, 3Þ
j¼1
∂xi ∂xj j¼1
∂xj ∂xi ∂xi

so that Eq. (B.6) becomes

 
1 ∂2 Θ ν ∂f i ∂f l
r2 σ il þ  δil r2 Θ ¼  þ ði; l ¼ 1; 2, 3Þ (B.9)
1 þ ν ∂xi ∂xl 1 þ ν ∂xl ∂xi

To simplify Eq. (B.9) even more, let i ¼ l in Eq. (B.6) and sum over i which, using the
symmetry of the stresses, gives

X3 X 3
∂2 σ ij ν  2
2  2r2 Θ ¼  4r Θ (B.10)
i¼1 j¼1
∂x i ∂x j 1 þ ν

which gives
X3 X 3
∂2 σ ij 1ν 2
¼ rΘ (B.11)
i¼1 j¼1
∂x i ∂x j 1þν
 The Beltrami–Michell Compatibility Equations 539

But by differentiating the equations of equilibrium again we have

X3 X 3
∂2 σ ij X3
∂f i
¼ (B.12)
i¼1 j¼1
∂xi ∂xj i¼1
∂xi

so using this result in Eq. (B.11) we find

  3
1þν X ∂f i
r2 Θ ¼  (B.13)
1  ν i¼1 ∂xi

Using Eq. (B.13) in Eq. (B.9) gives, finally, the compatibility equations written in terms of
the stresses as
X  
1 ∂2 Θ ν 3
∂f j ∂f i ∂f l
r σ il þ
2
¼ δil  þ ði; l ¼ 1; 2, 3Þ (B.14)
1 þ ν ∂xi ∂xl 1  ν j¼1 ∂xj ∂xl ∂xi

Equations (B.14) are the Beltrami–Michell compatibility equations. For the case of no body
forces they are simply

1 ∂2 Θ
r2 σ il þ ¼ 0 ði; l ¼ 1; 2, 3Þ (B.15)
1 þ ν ∂xi ∂xl
 APPENDIX C
The Sectorial Area Function

In elementary strength of materials problems involving bending of symmetrical sections and

torsion of circular shafts, certain geometrical properties of the cross-section appear. These
are as follows.
Location ðy; zÞ of the centroid of the cross-section and the associated first area moments
ðQyA , QzA Þ:
ð
yA ¼ QzA ¼ ydA
A
ð (C.1)
zA ¼ QyA ¼ zdA
A

[Note: these first area moments are based on the entire cross-sectional area. Similar quan-
tities ðQy , Qz Þ were defined in Chapter 10 to describe the shear flow in bending. Those
quantities are for a portion, Aq , of the cross-section as defined by a distance, s, in the cross-
section. In this appendix ðQyA , QzA Þ will always refer to quantities calculated for the entire
cross-section.]
Second area moments and mixed second area moment:
ð
I zz ¼ y2 dA
A
ð
I yy ¼ z2 dA (C.2)
A
ð
I yz ¼ yzdA
A

Polar area moment:

ð ð
 2 
J¼ y þ z2 dA ¼ r2 dA (C.3)
A A

In the bending and torsion of unsymmetrical thin sections, additional cross-sectional prop-
erties appear that are related to a quantity called the sectorial area function. To define this
function, consider a thin cross-section, as shown in Fig. C.1a. If we let rA be a position vector
from a point A in the cross-section to an arbitrary point, Q, along the centerline of the

540
 The Sectorial Area Function 541

Figure C.1 (a) Geometry en

of a thin section with the es
parameters used to define ds
the sectorial area
Q s
function. (b) The
C
sectorial area function,
ωA , is twice the sectorial rA As
area As ¼ ABC shown. rA B
1 s0
dAs rA ds
2

z Z A s, s0 2 As
A
A

O y
(a) (b)

section and let es be a unit vector tangent to the centerline, then we can define a unit normal
to centerline curve, en , as en ¼ es  ex (shown in Fig C.1a), where ex is a unit vector along the
x-axis. When the point on the centerline moves a small distance, ds, the line from A to
Q sweeps out a small area dAs ¼ rA en ds=2 ¼ r⊥ ds=2, where r⊥ is the perpendicular distance
to a line tangent to the centerline at Q. The differential of the sectorial area function, dωA , is
defined as dωA  2dAs or, equivalently, dωA ¼ rA es ds ¼ r⊥ ds so it is just twice this swept
area. The sectorial area function itself is
ðs ðs
ωA ðs; s0 Þ ¼ rA en ds ¼ r⊥ ds (C.4)
s0 s0

where s0 is a position along the centerline at which the sectorial area is zero called the
sectorial area origin and s is a varying position in the cross-section. There may be more than
one place in the cross-section where the sectorial area is zero so in that case there are
multiple origins. The sectorial area function depends on both the point A (called the pole
of the sectorial area function) as well as the origin, s0 , so when it is important to be explicit
about these points the sectorial area function will be written as ωA ðs; s0 Þ. Otherwise, it will
simply be written as ωðsÞ. From a geometrical standpoint, this sectorial area function
represents the twice the total area ABC swept out along the centerline as we move from s0
to s, as shown in Fig. C.1b. The sectorial area also has a physical meaning since in the
torsion of thin cross-sections the out-of-plane warping of the cross-section (along the
x-direction) is defined by this function. Note that we can write the perpendicular distance
r⊥ as r⊥ ¼ rA en or, equivalently, r⊥ ¼ rA ðes  ex Þ ¼ ðrA  es Þex , which shows that ωA is
positive when rA  es is in the þex direction, corresponding to the area being swept out in the
counterclockwise direction. Similarly ωA is negative when rA  es is in the ex direction,
542 Appendix C

Figure C.2 (a, b, c) Sectorial area functions ZA 0

(shaded plots) for different choices of the pole, A, I a2
A, and initial point (origin), I. All sides of the a 2
a I
C-section are of length a. a

a
a

a
a
a 2 A
2 2
a 2a
2a 2
2a 2

(a) (b)
2a 2
2a 2
a A

a
I
a 2

a 2

(c)

which corresponds to the area being swept out in a clockwise sense. The areas shown in
Fig. C.1a, b are both being swept out in the counterclockwise sense, so they are both
positive. Figure C.2 shows the sectorial area functions for a C-shaped cross-section with
different choices for the pole A and initial point, I, for the integration (which is also an
origin). The sense in which the area is being swept out in these cases is shown by dashed lines
and arrows that determine if the sectorial area is positive or negative. Note that since the
pole A was chosen along one of the legs of the cross-section, in all three cases there is no
additional sectorial area swept out for that leg, so the sectorial area function is a constant in
that leg. When the constant is zero, any point in that leg is an origin. For a given sectorial
area function, ωðsÞ, we can define a number of cross-sectional properties. The first sectorial
moment, Qω , for a thin section of variable thickness, t ¼ tðsÞ is defined as
ð
Qω ¼ ωðsÞdA (C.5)
A

where dA ¼ tðsÞds, and A is the total area of the cross-section. The sectorial products of areas
 
I yω , I zω are given by
 The Sectorial Area Function 543

ð
I yω ¼ yðsÞωðsÞdA
A
ð (C.6)
I zω ¼ zðsÞωðsÞdA
A

where ðy; zÞ are distances along the y- and z-axes as measured from some origin, O to a
general point on the centerline such as point Q (Fig. C.1a). A third quantity, called the
sectorial second moment (or warping constant), J ωP , is defined as
ð
 
J ωP ¼ ω2P s; sp dA (C.7)
A

where ωP ðs; sP Þ is a warping function with a specific pole, P, and origin(s), sP , called the
principal sectorial area function, which will be defined shortly. Comparing Eqs. (C.5)–
(C.7) with Eqs. (C.1)–(C.3) we see that these properties based on the sectorial area
function are analogous to the properties that appear in symmetric cross-section
bending problems.
Since the sectorial area function depends on both the pole used as well as the sectorial area
origin chosen on the centerline, it is instructive to examine how the choice of pole and origin
affect that function. Figure C.3 shows a geometry where there are two different poles (A, B),
and two different sectorial area origins ðs0 , s1 Þ. Let us use those points to define the sectorial
area functions ωA ðs; s0 Þ and ωB ðs; s1 Þ. Then from the geometry we have

ðs
ωA ðs; s0 Þ ¼ rA en ds
s0

ðs
 
¼ rB þ rB=A en ds
s0
(C.8)
ðs1 ðs ðs
s ¼ rB en ds þ rB en ds þ rB=A en ds
s1
s0 s1 s0

rB ðs
rA s0 ¼ ωB ðs0 , s1 Þ þ ωB ðs; s1 Þ þ rB=A en ds
yB , z B s0

z B yA , zA
rB / A
A

Figure C.3 Geometry for examining the effects of using different

O y poles and initial points to define the sectorial area function.
544 Appendix C

Now, let us examine the last integral in Eq. (C.8). We have

en ds ¼ ðes  ex Þds
  
dy dz
¼ ey þ ez  ex ds (C.9)
ds ds
¼ dy ez þ dz ey

and
rB=A ¼ ðyB  yA Þey þ ðzB  zA Þez (C.10)

so that
ðs ðs ðs
rB=A en ds ¼ ðyB  yA Þ dz  ðzB  zA Þ dy
s0 s0 s0
(C.11)
¼ ðyB  yA ÞðzðsÞ  zðs0 ÞÞ  ðzB  zA ÞðyðsÞ  yðs0 ÞÞ
¼ ðyB  yA ÞðzðsÞ  z0 Þ  ðzB  zA ÞðyðsÞ  y0 Þ

where we have let zðs0 Þ ¼ z0 , yðs0 Þ ¼ y0 . Placing Eq. (C.11) into Eq. (C.8) gives
ωA ðs; s0 Þ ¼ ωB ðs; s1 Þ  ωB ðs0 , s1 Þ
(C.12)
þ ðyB  yA ÞðzðsÞ  z0 Þ  ðzB  zA ÞðyðsÞ  y0 Þ

Equation (C.12) gives us a general relationship between sectorial area functions when we
change both the pole and sectorial area origin. Now, consider some special cases. First,
consider the case where we do not change the pole (A ¼ B) but have different origins
ðs0 ¼
6 s1 Þ. Then
ωA ðs; s0 Þ ¼ ωA ðs; s1 Þ  ωA ðs0 , s1 Þ (C.13)

But ωA ðs0 , s1 Þ is just a constant, so changing the origin simply changes the sectorial area
function by a constant amount. Now consider the case where the origins are the same
ðs1 ¼ s0 Þ but the pole locations are different ðA¼6 BÞ. Then we have
ωA ðs; s0 Þ ¼ ωB ðs; s0 Þ þ ðyB  yA ÞðzðsÞ  z0 Þ  ðzB  zA ÞðyðsÞ  y0 Þ (C.14)

Principal Pole
In the beam problems of Chapter 10 it is shown that, in order to produce bending only (no
torsion), the shear forces in the cross-section must be located at a specific point S called the
shear center. For thin sections, the shear center point can be located by a physical approach
where the shear flow expression (which assumes bending only) is used to evaluate where the
shear forces must be located. Alternatively, it is shown that if the pole P is chosen such that
 The Sectorial Area Function 545

 
sectorial products of area I yωP , I zωP are both zero, then P will be the shear center S. In
Chapter 10 that result was obtained with the use of the expression for the shear flow in
bending but here we will see that, if we calculate the location of a specific pole called the
principal pole, which is a function only of the geometry, this principal pole is also the
shear center.
If we take the general relationship of Eq. (C.12) and multiply it by the distance yðsÞ to a
general point s on the centerline of a thin section, as measured from the coordinate system
origin O (Fig. C.3), and integrate both sides over the entire cross-section we find
I yωA ¼ I yωB  ωB ðs0 , s1 ÞQzA þ ðzA  zB ÞðI zz  y0 QzA Þ
  (C.15)
 ðyA  yB Þ I yz  z0 QzA

If, instead, we multiply Eq. (C.12) by z and integrate, we find similarly


I zωA ¼ I zωB  ωB ðs0 , s1 ÞQyA þ ðzA  zB Þ I yz  y0 QyA
 (C.16)
 ðyA  yB Þ I yy  z0 QyA

Now, take the coordinate system origin O (from which the coordinates ðy; zÞ are measured)
to be located at the centroid of the cross-section. Then QyA ¼ QzA ¼ 0 and Eq. (C.15) and
Eq. (C.16) simplify to
I yωA ¼ I yωB þ ðzA  zB ÞI zz  ðyA  yB ÞI yz
(C.17)
I zωA ¼ I zωB þ ðzA  zB ÞI yz  ðyA  yB ÞI yy

Next, we will require pole A to be located at a specific pole, P, called the principal pole,
where
ð
I yωP ¼ yðsÞωP ðsÞdA ¼ 0
ðA (C.18)
I zωP ¼ zðsÞωP ðsÞdA ¼ 0
A

In this case, Eq. (C.17) becomes a set of simultaneous equations

I yωB þ ðzP  zB ÞI zz  ðyP  yB ÞI yz ¼ 0
(C.19)
I zωB þ ðzP  zB ÞI yz  ðyP  yB ÞI yy ¼ 0

which can be solved to obtain the two coordinates of the principal pole ðyP , zP Þ, both
measured from the centroid of the cross-section, as
I zωB I zz  I yωB I yz
yP ¼ yB þ
I yy I zz  I 2yz
(C.20)
I zωB I yz  I yωB I yy
zP ¼ zB þ
I yy I zz  I 2yz
546 Appendix C

which can also be written in terms of the relative distances of the principal pole P from pole
B as

I zωB I zz  I yωB I yz
ey ¼ yP  yB ¼
I yy I zz  I 2yz
(C.21)
I zωB I yz  I yωB I yy
ez ¼ zP  zB ¼
I yy I zz  I 2yz

which is identical to Eq. (10.52) for the location of the shear center S (relative to pole B).
Thus, the principal pole P is indeed the same point as the shear center S. We are free to
choose whatever sectorial area origin we want in calculating the sectorial area terms on the
right-hand side of Eq. (C.21) since if we change the sectorial area origin used in calculating
ωB , this sectorial area only changes by a constant amount, as previously shown in Eq.
 B ¼ ωB þ ω0 where ω0 is a constant, we have
(C.13). Thus, if we let ω
ð
I yωB ¼ yðωB þ ω0 ÞdA ¼ I yωB þ ω0 QzA ¼ I yωB
ðA (C.22)
I zωB ¼ zðωB þ ω0 ÞdA ¼ I zωB þ ω0 QyA ¼ I zωB
A

since QzA ¼ QyA ¼ 0. Because the sectorial products of area remain unchanged when the
sectorial area function origin is changed, the coordinates of the principal pole found from
Eq. (C.20) or Eq. (C.21) are independent of this sectorial area origin. We also placed no
restriction on the pole B used to determine the principal pole, so in practice both the pole
and the sectorial area origin used to calculate ωB can be chosen to make the calculations as
easy as possible. The principal pole location does not depend on the particular choice of the
pole B since if we use another pole A location instead and assume Eq. (C.21) produces a
 
different principal pole location y0P ; z0P from the ðyP , zP Þ values obtained with pole B, then
using the relations of Eq. (C.17) we can easily show that

I zωA I zz  I yωA I yz I zωB I zz  I yωB I yz

y0P  yA ¼ ¼ þ ðyB  yA Þ
I yy I zz  I 2yz I yy I zz  I 2yz
¼ yP  yB þ ðyB  yA Þ
! y0P ¼ yP

and a similar calculation shows z0P ¼ zP also. Thus, the principal pole (shear center) location
is a function of the geometry only and is independent of both the pole and sectorial area
function origin used in its location.
An alternative to using Eq. (C.20) to determine the principal pole is write a sectorial area
function ωB in terms of a set of unknown pole coordinates ðyB , zB Þ relative to the centroid
and then determine those pole coordinates that satisfy Eq. (C.18), i.e., solve
 The Sectorial Area Function 547

ð
yðsÞωB ðsÞdA ¼ 0
A
ð (C.23)
zðsÞωB ðsÞdA ¼ 0
A

since those coordinates will be the coordinates, relative to the centroid, of the principal pole
(shear center). Again, the choice of origin used in obtaining ωB in Eq. (C.23) is arbitrary.
If the y-axis of the cross-section is an axis of symmetry, then the centroid will lie on that
axis of symmetry so that if we place the pole used in calculating ωB also on that symmetry
axis, we will have zB ¼ 0 and one can easily show that I yωB ¼ 0. Since we also have I yz ¼ 0,
Eq. (C.20) shows that zP ¼ 0; i.e., the principal pole also lies on the axis of symmetry.
Similarly, if the z-axis is an axis of symmetry, the principal pole will lie on that axis. If both
the y- and z-axis are axes of symmetry, then the principal pole will be at the centroid.

Example C.1 Shear Center Location

For a simple example of calculating the shear center, consider the thin unsymmetrical
I-beam cross-section shown in Fig. C.4a where the thickness t  a, b, h. This cross-section
does have an axis of symmetry so the centroid and shear center lie on the y-axis and I yz ¼ 0
so that we have, from Eq. (C.21)
I zωB
ey ¼
I yy (C.24)
ez ¼ 0

As in Chapter 10, when calculating parameters such as centroids, area moments, etc., for
thin cross-sections it is common to use the dimensions as measured along the centerline, as
shown in Fig C.4a, to calculate the cross-sectional properties and to neglect terms that
involve the thickness raised to higher powers. Thus, for example we calculate I yy as

ta3 ht3 tb3 t  3 

I yy ¼ þ þ ffi a þ b3 (C.25)
12 12 12 12
where we have neglected the middle term. We will use point B (Fig. C.4b) as both the pole
and the initial point I (origin) to calculate ωB . The sectorial area with this choice is zero
except for the right-hand vertical section where ωB ¼ hs, with the distance s taken along the
section as shown in Fig. C.4b. The sectorial area function is shown as the shaded area in Fig.
C.4b. To calculate I zωB we have

ð
b=2 ð
b=2

I zωB ¼ zðsÞωB ðsÞtds ¼ sðhsÞtds

(C.26)
b=2 b=2

¼ thb3 =12
548 Appendix C

Figure C.4 (a) An unsymmetrical z

I-beam cross-section. (b) Geometrical hb/2
parameters for determining the t ey
centroid, shear center, and sectorial s
a b
area parameters.
B,I S C y

–hb/2
h
(a) (b)

Figure C.5 Parameters for calculating the z h  d y  yB b / 2

principal pole (shear center) for the
unsymmetrical I-beam in Fig. C.4a using  yB  d y a / 2
Eq. (C.23) and the resulting sectorial area yB
s2 s1
functions shown as the shaded areas. C
y
dy B,I
yB  d y a / 2

 h  d y  yB b / 2

which gives

hb3
ey ¼   (C.27)
a3 þ b3

For a ¼ b the I-beam has two axes of symmetry and ey ¼ h=2 so the shear center is at the
centroid, as stated earlier. As an alternative way to calculate the shear center, let us use
Eq. (C.23) directly. Figure C.5 shows the choice of a pole B and initial point I (origin) that is
at an arbitrary distance yB relative to the centroid. We will see we do not have to know
explicitly the centroid location in this example which we have described by the distance, d y ,
as shown. Also shown is the sectorial area function ωB . The first equation in Eq. (C.23) is
satisfied automatically by the symmetry present so consider the second equation. We have
ð ð
b=2 ð
a=2
   
zðsÞωðsÞdA ¼ s1 h  d y  yB s1 tds1 þ s2  yB þ d y s2 tds2 ¼ 0 (C.28)
A b=2 a=2

which, letting yB þ d y ¼ ey gives

 
h  ey tb3 ey ta3
 ¼0 (C.29)
12 12
 The Sectorial Area Function 549

which again gives Eq. (C.27). We see that indeed we did not have to know the distance d y to
the centroid in this example to locate the shear center. Also note that we could have replaced
the distance z, from the centroid by z þ d z in Eq. (C.28) and not changed this final result.
This is because the sectorial area function ωB satisfied
ð
QωB ¼ ωB dA ¼ 0 (C.30)
A

When this is the case, the coordinates y and z appearing in Eq. (C.23) can be measured from
any fixed location we choose in the cross-section since if we let y0 ¼ y þ d y , z0 ¼ z þ d z ,
 
where d y , d z are any two constants, then if Eq. (C.30) is satisfied we have
ð ð ð ð
y0 ωB dA ¼ yωB dA þ d y ωB dA ¼ yωB dA ¼ 0
A A A A
ð ð ð ð (C.31)
z0 ωB dA ¼ zωB dA þ d z ωB dA ¼ zωB dA ¼ 0
A A A A

A sectorial area function, ωP ðs; s0 Þ, which satisfies

ð
ωP dA ¼ 0
A
ð
y0 ωP dA ¼ 0 (C.32)
A
ð
z0 ωP dA ¼ 0
A

for coordinates ðy0 , z0 Þ measured with respect to any fixed point in the cross-section is called a
principal sectorial area function, where s0 is a principal origin of the principal sectorial area
function. The three equations in Eq. (C.32) fix the location of the principal pole P and the
location of the principal origin(s) (there may be more than one origin location in the cross-
section) in the principal sectorial area function.

Principal Origin
Now, consider further the equation for determining the principal origin(s). By definition, if
for a sectorial area function computed with a given pole B there is a sectorial area function
origin, s0 , such that
ð
QωB ¼ ωB ðs; s0 ÞdA ¼ 0 (C.33)
A
550 Appendix C

the point s0 is called a principal origin. To determine the origin(s) consider another sectorial
area function ωB ðs; s1 Þ with a known origin, s1 . Then from Eq. (C.13)
ωB ðs; s0 Þ ¼ ωB ðs; s1 Þ  ωB ðs0 , s1 Þ (C.34)

so that if s0 is a principal origin we must have

ð ð
QωB ¼ ωðs; s0 ÞdA ¼ ωB ðs; s1 ÞdA  ωB ðs0 , s1 ÞA ¼ 0 (C.35)
A A

This is an equation that we can use to determine the principal origin(s) in the following way.
The quantity ωðs0 , s1 Þ is just a constant that we will call ω0 . From Eq. (C.35) we can solve
for ω0 as
ð
1
ω0 ¼ ωB ðs; s1 ÞdA (C.36)
A
A

The right-hand side is known since both the pole B and the center s1 are specified. The
sectorial area function ωB with the principal origin(s) can then be obtained from Eq. (C.34)
as
ωB ðs; s0 Þ ¼ ωB ðs; s1 Þ  ω0 (C.37)

If we evaluate and plot ωB ðs; s0 Þ the zero(s) of this sectorial area function are the location(s)
of the principal origin(s). [Note: generally, we do not have to solve explicitly for the s0 values
at the origins because the ω0 value and ωB ðs; s1 Þ are all that we need to obtain the sectorial
area function with those principal origin(s).] The principal origin(s) do not depend on the
choice of s1 used in Eq. (C.36) since if we had used a sectorial area function ωB ðs; s2 Þ instead
to determine the principal origin we would have
ð
ωB ðs; s2 ÞdA  ωB ðs0 , s2 ÞA ¼ 0 (C.38)
A

But from Eq. (C.13)

ωB ðs; s2 Þ ¼ ωB ðs; s1 Þ  ωB ðs2 , s1 Þ (C.39)

so Eq. (C.38) becomes

ð
ωB ðs; s1 ÞdA  ½ωB ðs0 , s2 Þ þ ωðs2 , s1 ÞA
A
ð (C.40)
¼ ωB ðs; s1 ÞdA  ωB ðs0 , s1 ÞA ¼ 0
A

which leads to the same ω0 value and hence the same principal origin(s). Thus, a principal
origin, like the principal pole, is a cross-sectional property.
 The Sectorial Area Function 551

If we let the pole B in the above discussion be the principal pole P (or, equivalently the
shear center, S) then the sectorial area ωP ðs; s0 Þ ¼ ωP ðs; s1 Þ  ω0 is the principal sectorial
area function with principal pole P and principal origin s0 .

The Sectorial Second Moment (Warping Constant)

The sectorial second moment defined in Eq. (C.7) uses the principal sectorial area function.
We can also define this quantity in terms of a sectorial area function, ωB , that is defined with
respect to an arbitrary pole and origin. To see this let us use Eq. (C.12) to relate the principal
sectorial area function to
 
ωP s; sp ¼ ωB ðs; s1 Þ  ωB ðsP , s1 Þ
(C.41)
þ ðyB  yP ÞðzðsÞ  z0 Þ  ðzB  zP ÞðyðsÞ  y0 Þ

If we integrate Eq. (C.41) and assume that yðsÞ and zðsÞ are measured from the centroid of
the cross-section, then we find

QωP ¼ 0 ¼ QωB  ωB ðsP , s1 ÞA  y0 ðzP  zB ÞA þ z0 ðyP  yB ÞA (C.42)

Placing Eq. (C.42) into Eq. (C.41), we can write that equation as
Qω B
ωP ðs; sP Þ ¼ ωB ðs; s1 Þ  þ ðzP  zB Þy  ðyP  yB Þz (C.43)
A

Using Eq. (C.43), sectorial second moment is

ð
Q2
J ωP ¼ ω2P dA ¼ J ωB þ 2ðzP  zB ÞI yωB  2ðyP  yB ÞI zωB  ωB
A (C.44)
A

þ ðyP  yB Þ2 I yy þ ðzP  zB Þ2 I zz  2ðyP  yB ÞðzP  zB ÞI yz

However, because P is a principal pole we have from Eq. (C.19)

I yωB þ ðzP  zB ÞI zz  ðyP  yB ÞI yz ¼ 0

(C.45)
I zωB þ ðzP  zB ÞI yz  ðyP  yB ÞI yy ¼ 0

which can be used to write Eq. (C.44) as

Q2ωB
J ωP ¼ J ωB   ðyP  yB Þ2 I yy
A (C.46)
þ 2ðyP  yB ÞðzP  zB ÞI yz  ðzP  zB Þ2 I zz

which gives us the final relationship we sought that allows us to compute the sectorial second
moment from a knowledge of the relative location of the principal pole with respect to pole
B, the second area moments of the cross-section, and ωB .
552 Appendix C

Example C.2 Sectorial Second Moment

To illustrate a calculation of the sectorial second moment, consider the unsymmetrical I-
beam of Fig. C.4a. The principal sectorial area function is given in the two flanges (see
Fig. C.5) as
( 
h  ey s1  b=2 s1 b=2
ωP ¼ (C.47)
ey s2  a=2 s2 a=2

where ey is given by Eq. (C.27). Then

ð ð
b=2 ð
a=2
 2
J ωP ¼ ω2P dA ¼t h  ey s21 ds1 þ t e2y s22 ds2
A b=2 a=2 (C.48)
 2
tb h  ey
3
ta3 e2y 2 3 3
th a b
¼ þ ¼  
12 12 12 a3 þ b3

where the final result comes after using the expression for ey and some algebra. Now, let us
use instead Eq. (C.46) to calculate the sectorial second moment. In this case we have QωB ¼ 0
as well as I yz ¼ 0 and also zP ¼ zB ¼ 0 so we find

J ωP ¼ J ωB  ðyS  yB Þ2 I yy
ð
b=2 " #
2 t a3
þ b 3
¼ ðhsÞ tds  e2y
12 (C.49)
b=2

th2 b3 th2 b6 th2 a3 b3

¼    ¼  
12 12 a3 þ b3 12 a3 þ b3

which gives the same value.

Now, consider these sectorial area calculations in a more general unsymmetrical

section example.

Example C.3 Determining Sectorial Properties of an Unsymmetrical Section

All the examples shown so far have been rather simple sections with some symmetry so that
the calculations could be easily illustrated. Now, consider a more general thin unsymmet-
rical section such as the one shown in Fig. C.6a. We will calculate the principal pole (shear
center) for the cross-section and the principal sectorial area function. The thickness, t, is the
same for all sections. Using the sections (1), (2) (3) shown in Fig. C.6b the centroid is
given by
 The Sectorial Area Function 553

zc Figure C.6 (a) The geometry of an

unsymmetrical section, showing the
b s3 (3) location, C, of the centroid. (b) The
parameters needed to define the
z location of the shear center and the
dy (2) 2b 2
principal sectorial area function.
2b C ey
y S
t
dz ez
s2 B, I

yc
2b s1 (1)
(a) (b)

y1 A1 þ y2 A2 þ y3 A3

y ¼
A1 þ A2 þ A3
(C.50)
z1 A1 þ z2 A2 þ z3 A3
z ¼
A1 þ A2 þ A3
where Am is the area of section (m) for m ¼ 1, 2, 3 and where ym is the location of the centroid
of Am in the ðy0 , z0 Þ coordinates. Note that all distances are measured along the centerline of
the cross-section and areas are computed with these distances also, as is customary when
dealing with thin sections. These expressions give
ðbÞð2btÞ þ ð0Þð2btÞ þ ðb=2ÞðbtÞ b
y ¼ ¼
2bt þ 2bt þ bt 2
(C.51)
ð0Þð2btÞ þ ðbÞð2btÞ þ ð2bÞðbtÞ 4b
z ¼ ¼
2bt þ 2bt þ bt 5
where ðy; zÞ here are the coordinates of C in the ðy0 , z0 Þ coordinates. For the second area and
mixed area moments we have from the parallel axis theorem

I yy ¼ I y1 þ A1 d 2y1 þ I y2 þ A2 d 2y2 þ I y3 þ A3 d 2y3

I zz ¼ I z1 þ A1 d 2z1 þ I z2 þ A2 d 2z2 þ I z3 þ A3 d 2z3 (C.52)
I yz ¼ I yz1 þ A1 d y1 d z1 þ I yz2 þ A2 d y2 d z2 þ I yz3 þ A3 d y3 d z3

where I y m ðm ¼ 1; 2, 3Þ are second area moments about a y-axis going through the centroid
of each area Am and I zm ðm ¼ 1; 2, 3Þ are second area moments about a z-axis going
through the centroid of each Am . The mixed area moments I yzm ðm ¼ 1; 2, 3Þ are similarly
measured with to the centroidal axes going through the centroid of each Am . The distances
 
d ym , d zm ðm ¼ 1; 2, 3Þ are distances from the ðy; zÞ centroidal axes, respectively, to the
individual centroids of the areas Am . For our specific geometry we have
554 Appendix C

     
2bt3 4b 2 tð2bÞ3 4b 2 bt3 4b 2
I yy ¼ þ 2bt  þ þ 2bt b  þ þ bt 2b 
12 5 12 5 12 5
52b3 t
¼
15
 2  2
tð2bÞ3 b 2bt3 b tb3
I zz ¼ þ 2bt þ þ 2bt þ þ btð0Þ2
12 2 12 2 12 (C.53)
3
7b t
¼
4
       
b 4b b b 6b
I yz ¼ 0 þ 2bt þ 0 þ 2bt þ 0 þ bt ð0Þ
2 5 2 5 5
¼ b3 t

where we have neglected all the terms involving t3 (which was also done in Chapter 10) since
these terms are much smaller than the remaining terms for a thin section. To calculate the
location of the shear center, we can use point B as both the pole and initial point (origin) to
calculate ωB . These are convenient choices since in the three sections we only have a nonzero
function in one section as shown in Fig. C.6b. Specifically, ωB ¼ 2bs3 in section (3). Then
 
we can calculate the functions I yωB , I zωB as
ð ðb
b4 t
I yωB ¼ yωB dA ¼  ðb=2  s3 Þð2bs3 Þtds3 ¼ 
6
A 0
(C.54)
ð ðb 4
6b t
I zωB ¼ zωB dA ¼ ð2b  4b=5Þð2bs3 Þtds3 ¼ 
5
A 0

which gives, from Eq. (C.21)

     
I zωB I zz  I yωB I yz 6b4 t=5 7b3 t=4  b4 t=6 b3 t 17b
ey ¼ ¼   ¼
I yy I zz  I yz
2
76b t =15
6 2 38
      (C.55)
I zωB I yz  I yωB I yy 6b4 t=5 b3 t  b4 t=6 52b3 t=15 20b
ez ¼ ¼   ¼
I yy I zz  I 2yz 76b t =15
6 2 57

Now, let us compute the sectorial area function with respect to the principal pole P (the same
as the shear center S) and having the same sectorial origin sI ¼ I as before (Fig. C.7). Then
this sectorial area function from the geometry of Fig. C.7 is:
8
< 20bs1 =57 0 s1 2b
>
ωP ðs; sI Þ ¼ 17bs2 =38 0 s2 2b (C.56)
>
:
17b =19  94bs3 =57 0 s3 b
2
 The Sectorial Area Function 555

zc
s3 (3)
(2)

17b
S 38
20b
s2 I
57
yc
s1 (1) Figure C.7 Calculation of the sectorial area about the shear center
with I as the origin.

32b 2
32b 2 (3)
57
+ 57
_
(2) +
62b 2
57
7b 2
_ 19
b 2 +
_
3
(1)
b 2
3 Figure C.8 The principal sectorial area function for the unsymmetrical
section of Fig. C.6a.

The first sectorial moment of this function is

ð 2ðb 2ðb ðb !
20bs1 17bs2 17b2 94bs3
ωP ðs; sI ÞdA ¼ tds1 þ tds2 þ  tds3
57 38 19 57
A 0 0 0
   
40 3 34 17 47 80 102 102 94 3
¼ b t þ b3 t þ b3 t  b3 t ¼ þ þ  bt
57 38 19 57 114 114 114 114
5
¼ b3 t
3
(C.57)

Using Eq. (C.36), let us compute

ð
1
ω0 ¼ ωP ðs; sI ÞdA (C.58)
A
A
556 Appendix C

Then

1 5b3 t
ω0 ¼
5bt 3
(C.59)
b2
¼
3
and the principal sectorial area function is ωP ðs; s0 Þ ¼ ωP ðs; sI Þ  ω0 where
8
>
< 20bs1 =57  b =3 0 s1 2b
2

ωP ðs; s0 Þ ¼ 17bs2 =38  b2 =3 0 s2 2b (C.60)

>
:
32b2 =57  94bs3 =57 0 s3 b

This principal sectorial area function is shown in Fig. C.8 where it can be seen that there are
actually three principal origins (shown as black dots) in the principal sectorial area function.
APPENDIX D
MATLAB® Files

At various points in this book, MATLAB® functions or scripts are used to aid in the
solution process. The listings of the m-files for those functions or scripts are given below.
[Note: because of formatting changes, there may be some minor differences between these
listings and the actual m-files.] The m-files can be obtained from Cambridge University Press
at www.cambridge.org/schmerr or by sending an e-mail with the subject title “Advanced
Mechanics Codes” to the author at [email protected]
stress_invs

function [I1, I2, I3] = stress_invs(s)

% [I1, I2, I3] =stress_invs(s) uses the stress matrix s as its
% input argument and returns the three stress invariants (I1,
% I2, I3)
I1 = trace(s);
I2 = s(1,1)*s(2,2) - s(1,2)^2 + s(1,1)*s(3,3) - s(1,3)^2 ...
+ s(2,2)*s(3,3) - s(2,3)^2;
I3 = det(s);

cantilever_d

%script cantilever_d
% This script plots the 2-d displacement field of a cantilever
% beam from engineering beam theory
% make an 8x4 grid on a beam with normalized length and width
% of 1 and 0.2, respectively.
xn = linspace(0,1,8);
yn = linspace (-0.1, 0.1, 4);
[x,y] = meshgrid(xn, yn);
% choose values for Poisson’s ratio and normalized height hn =
% h/L
nu = 1/3;
hn = 0.1;

557
558 Appendix D

% calculate normalized beam displacements

ux = x.^2.*y/2 + (1+nu)*(hn*y - y.^3/3) + nu*y.^3/6 -y/2;

uy = -x.^3/6 - nu*x.*y.^2/2 + x/2 -1/3;
% make a quiver plot of the displacements
quiver(x, y, ux, uy)
axis equal

stress_strain

function stress = stress_strain(E, nu, strain)

% stress = stress_strain(E, nu, strain) has as inputs Young’s
% modulus, E,(in GPa), Poisson’s ratio, nu, and the tensor
% strain matrix, strain, (in micro strain). The function
% calculates the stress matrix (in MPa) for an isotropic
% elastic solid

E = E*10^3; % put E in MPa

e = strain*10^(-6); % put micro factor in strain
C = E/((1+nu)*(1-2*nu));
G = E/(2*(1+nu));
stress = zeros(3,3);
stress(1,1) = C*((1-nu)*e(1,1) +nu*(e(2,2) +e(3,3)));
stress(2,2) = C*((1-nu)*e(2,2) +nu*(e(1,1) +e(3,3)));
stress(3,3) = C*((1-nu)*e(3,3) +nu*(e(1,1) +e(2,2)));
stress(1,2) = 2*G*e(1,2);
stress(1,3) = 2*G*e(1,3);
stress(2,3) = 2*G*e(2,3);
% make stresses symmetric
stress(2,1) = stress(1,2);
stress(3,1) = stress(1,3);
stress(3,2) = stress(2,3);

strain_stress

function strain = strain_stress(E, nu, stress)

% strain = strain_stress( E, nu ,stress) returns the strain
% matrix strain, measured in microstrain, for a state of
% stress, stress, with the stresses measured in MPa, and where
% E is given in GPa. nu is Poisson’s ratio.
% To get strains in microstrain, multiply E by 10^-3
 MATLAB® Files 559

E = E*10^-3;
G = E/(2*(1+nu));
strain(1,1) = (1/E)*(stress(1,1) -nu*(stress(2,2). . . +stress(3,3)));
strain(2,2) = (1/E)*(stress(2,2) -nu*(stress(1,1). . . +stress(3,3)));
strain(3,3) = (1/E)*(stress(3,3) -nu*(stress(1,1). . . +stress(2,2)));
strain(1,2) = stress(1,2)/(2*G); % note: these are tensor
% strains
strain(1,3) = stress(1,3)/(2*G);
strain(2,3) = stress(2,3)/(2*G);
strain(2,1) = strain(1,2);
strain(3,1) = strain(1,3);
strain(3,2) = strain(2,3);

M_Matrix

function M = M_Matrix(l)
% M = M_Matrix(l) takes as its input argument the direction
% cosine matrix l whose transpose, lt, transforms the x-
% coordinates to the x’-coordinates through the relation {x’}
% = [lt]{x}. The function returns the 6x6 matrix, M, which
% transforms the 6x6 matrix of elastic constants,C, as
% measured in the x-coordinates, to the matrix of elastic
% constants,C’, as measured in the x’-coordinates, through the
% relation [C’] = [M][C][Mt], where Mt is the transpose of M.

M = zeros(6,6);
M(1,1) = l(1,1)^2;
M(1,2) = l(2,1)^2;
M(1,3) = l(3,1)^2;
M(1,4) = 2*l(2,1)*l(3,1);
M(1,5) = 2*l(3,1)*l(1,1);
M(1,6) = 2*l(2,1)*l(1,1);
M(2,1) = l(1,2)^2;
M(2,2) = l(2,2)^2;
M(2,3) = l(3,2)^2;
M(2,4) = 2*l(2,2)*l(3,2);
M(2,5) = 2*l(1,2)*l(3,2);
M(2,6) = 2*l(1,2)*l(2,2);
M(3,1) = l(1,3)^2;
M(3,2) = l(2,3)^2;
M(3,3) = l(3,3)^2;
M(3,4) = 2*l(2,3)*l(3,3);
560 Appendix D

M(3,5) = 2*l(1,3)*l(3,3);
M(3,6) = 2*l(1,3)*l(2,3);
M(4,1) = l(1,2)*l(1,3);
M(4,2) = l(2,2)*l(2,3);
M(4,3) = l(3,2)*l(3,3);
M(4,4) = l(2,2)*l(3,3) + l(3,2)*l(2,3);
M(4,5) = l(1,2)*l(3,3) + l(3,2)*l(1,3);
M(4,6) = l(1,2)*l(2,3) + l(2,2)*l(1,3);
M(5,1) = l(1,1)*l(1,3);
M(5,2) = l(2,1)*l(2,3);
M(5,3) = l(3,1)*l(3,3);
M(5,4) = l(2,1)*l(3,3) + l(3,1)*l(2,3);
M(5,5) = l(1,1)*l(3,3) + l(3,1)*l(1,3);
M(5,6) = l(1,1)*l(2,3) + l(2,1)*l(1,3);
M(6,1) = l(1,1)*l(1,2);
M(6,2) = l(2,1)*l(2,2);
M(6,3) = l(3,1)*l(3,2);
M(6,4) = l(2,1)*l(3,2) + l(3,1)*l(2,2);
M(6,5) = l(1,1)*l(3,2) + l(3,1)*l(1,2);
M(6,6) = l(1,1)*l(2,2) + l(2,1)*l(1,2);

m_to_v

function y = m_to_v(input, type)

% m_to_v converts a 3x3 matrix of strains or stresses to a 6x1
% vector
% first input argument is the 3x3 matrix
% the type argument is either ’strain’ or ’stress’
% if type = ’strain’ then the matrix shear strains are doubled
% to give engineering shear strains
if strcmp(type, ’stress’)
y(1) = input(1,1);
y(2) = input(2,2);
y(3) = input(3,3);
y(4) = input(2,3);
y(5) = input(1,3);
y(6) = input(1,2);
elseif strcmp(type, ’strain’)
y(1) = input(1,1);
y(2) = input(2,2);
y(3) = input(3,3);
 MATLAB® Files 561

y(4) = 2*input(2,3);
y(5) = 2*input(1,3);
y(6) = 2*input(1,2);
else
error(’wrong type’)
end
% put in column vector
y = y’;

v_to_m

function y = v_to_m(input, type)

% v_to_m converts a 6x1 vector of strains or stresses to a 3x3
% matrix
% first input argument is the 6x1 vector
% the type argument is either ’strain’ or ’stress’
% if type = ’strain’ then the vector engineering shear strains
% are halved to generate the tensor shear strains in the 3x3
% matrix
if strcmp(type, ’stress’)
y(1,1) = input(1);
y(2,2) = input(2);
y(3,3) = input(3);
y(2,3) = input(4);
y(1,3) = input(5);
y(1,2) = input(6);
elseif strcmp(type, ’strain’)
y(1,1) = input(1);
y(2,2) = input(2);
y(3,3) = input(3);
y(2,3) = input(4)/2;
y(1,3) = input(5)/2;
y(1,2) = input(6)/2;
else
error(’wrong type’)
end
y(2,1) = y(1,2);
y(3,1) = y(1,3);
y(3,2) = y(2,3);
562 Appendix D

rosette

function [exx, eyy, exy] = rosette(anga, angb, angc, ea, eb,. . . ec)
% [exx, eyy, exy] = rosette(anga, angb, angc, ea, eb, ec)
% takes the three angles (anga, angb, angc) of the three gages
% in a rosette gage (as measured in degrees from the x-
% axis)and the three strains (ea, eb, ec)measured by those
% gages, and returns the tensor strains(exx, eyy, exy)
nax = cos(anga*pi/180);
nay = sin(anga*pi/180);
nbx = cos(angb*pi/180);
nby = sin(angb*pi/180);
ncx = cos(angc*pi/180);
ncy = sin(angc*pi/180);
Coeff = [ nax^2 nay^2 2*nax*nay;...
nbx^2 nby^2 2*nbx*nby; ...
ncx^2 ncy^2 2*ncx*ncy];
vals = [ ea eb ec]’;
strains = Coeff\vals;
exx = strains(1);
eyy = strains(2);
exy = strains(3);

stress_singularity

function stress_singularity()
% stress_singularity determines the singular behavior of the
% stresses near a sharp notch and plots the behavior of the
% parameter that controls the order of the singularity as a
% function of the notch angle.
warning off
a = linspace(0, pi, 100); %angle alpha
out = zeros (1,100);
for k =1:100
val = 0.5 +0.5*a(k)/pi; % choose initial guesses
out(k) = fzero(@singularity, val, [ ],a(k)); % find roots
end
plot((180/pi)*a(1:100), out(1:100))
axis([ 0 180 0 1.0])
xlabel(’\alpha , degrees’)
 MATLAB® Files 563

ylabel(’smallest root, m’)

warning on
function y = singularity(x, a)
y = sin (x*(2*pi-a)) - x*sin(a);
end
end

cross_section

function [Yct, Zct, Iy, Iz, Iyz] = cross_section( Ly, Lz,. . .

Yc, Zc)
% [Yct, Zct, Iy, Iz, Iyz] = cross_section(Ly, Lz, Yc, Zc)
% determines the coordinates of the centroid of a cross-
% section and the area moments relative to that centroid for a
% cross-section composed of rectangular parts. Ly and Lz are
% vectors containing the lengths of the rectangles in the Y-
% and Z-directions, and Yc, Zc are vectors containing the
% locations of the Y- and Z-coordinates of the rectangles. The
% outputs are the Y- and Z-coordinates of the cross-section
% centroid (Yct, Zct) and the area moments (Iy, Iz, Iyz)
% relative a set of centroidal (y,z) axes. Note that the
% rectangles must lie parallel to the (Y,Z) axes.
A = sum(Ly.*Lz); % total area of cross-section
% calculate location of centroid for the total cross-section
Yct = sum((Yc.*Ly.*Lz)/A); % Y-coordinate of centroid in (Y,Z)
Zct = sum((Zc.*Ly.*Lz)/A); % Z-coordinate of centroid in (Y,Z)
% calculate area moments about the (Y,Z) axes
IY = sum(Ly.*(Lz.^3)/12 +Ly.*Lz.*(Zc.^2)); % IYY in (Y,Z)
% coordinates
IZ = sum(Lz.*(Ly.^3)/12 +Ly.*Lz.*(Yc.^2)); % IZZ in (Y,Z)
% coordinates
IYZ = sum(Ly.*Lz.*Yc.*Zc); % IYZ in (Y,Z) coordinates
% use parallel axis theorem to calculate area moments about
% centroidal coordinates (y,z)
Iy = IY - A*(Zct^2); % Iyy for centroidal (y,z) axes
Iz = IZ -A*(Yct^2); % Izz for centroidal (y,z) axes
Iyz =IYZ - A*Yct*Zct; % Iyz for centroidal (y,z) axes
564 Appendix D

sections

% script sections
% This script computes the location of the centroid and area
% moments for a L-section, a T-section, a J-section, a Z-
% section, and a I-section. In the command window one must
% enter a string ’L’, ’T’, ’J’, ’Z’, or ’L’ in a variable
% named type and then enter the lengths b, h, c and the
% thicknesses t1, t2, t3 (or a subset of these values) in the
% command window and then run this script. The script uses the
% function cross_section to return the location of the
% centroid and the area moments
switch type
case{’L’}
Ly(1) = b- t2; Ly(2) = t2;
Lz(1) = t1; Lz(2) = h;
Yc(1) = t2 + (b-t2)/2; Yc(2) =t2/2;
Zc(1) = t1/2; Zc(2) = h/2;
case{’T’}
Ly(1) = b; Ly(2) = t2;
Lz(1) = t1; Lz(2) = h-t1;
Yc(1) = 0; Yc(2) = 0;
Zc(1) = t1/2; Zc(2) = t1 + (h- t1)/2;
case{’J’}
Ly(1) = b -t2; Ly(2) = t2; Ly(3) = c-t2;
Lz(1) = t1; Lz(2) = h; Lz(3) = t3;
Yc(1) = t2 + (b-t2)/2; Yc(2) = t2/2; Yc(3) = t2 . . .
+(c-t2)/2;
Zc(1) = t1/2; Zc(2) = h/2; Zc(3) = h - t3/2;
case{’Z’}
Ly(1) = b - t2; Ly(2) = t2; Ly(3) = c -t2;
Lz(1) = t1; Lz(2) = h; Lz(3) = t3;
Yc(1) = t2 + (b-t2)/2; Yc(2) = t2/2; Yc(3) = -(c-t2)/2;
Zc(1) = t1/2; Zc(2) = h/2; Zc(3) = h -t3/2;
case{’I’}
Ly(1) = b; Ly(2) = t2; Ly(3) = c;
Lz(1) = t1; Lz(2) = h - t1 -t3; Lz(3) = t3;
Yc(1) = 0; Yc(2) = 0; Yc(3) = 0;
Zc(1) = t1/2; Zc(2) = t1 + (h -t1 -t3)/2;Zc(3) = . . .
h -t3/2;
otherwise
disp([’Unknown type’])
end
[yc, zc, Iy, Iz, Iyz] = cross_section(Ly,
 Index

Airy stress function, 149, 153, 190 direction cosine matrix, 35 Gauss’s theorem, 70
Airy stress function, Cartesian distortional strain energy density, 232 Gauss’s theorem -2D, 76
coordinates, 191 dummy load, 261 generalized displacements, 174
Airy stress function, polar coordinates, generalized forces, 174
197 eddy current NDE, 513 generalized Hooke’s law, 119
alternating stress, 502 effective maximum radius, 414 generalized strains, 174
effective polar area moment, 414, 527 Gerber curve, 504
Beltrami–Michell compatibility effective stress, 61, 500 Goodman curve, 504
equations, 149, 539 eigenvalue problem, 44 governing equations, 146
bending moment, 6 eigenvalues, 44
Betti–Rayleigh theorem, see reciprocal eigenvectors, 44 Hertz contact theory, 212
theorem of Betti–Rayleigh elastic bodies in contact, see Hertz Hooke’s law, 5
biaxial state of stress, 51 contact theory
bifurcation instability, 517 elastic constants, 122 interference, 188
biharmonic equation, 158 Engesser’s first theorem, 260 isotropic material, 119, 127
biharmonic equation, Cartesian Engesser’s second theorem, 264
coordinates, 191 engineering beam theory, 9 Kronecker delta, 35
biharmonic equation, polar coordinates, engineering shear strain, 14, 90
197 equilibrium equations, Cartesian Lamé constants, 148
biharmonic operator, 153 coordinates, 71 Laplace’s equation, 412
bimoment, 490 equilibrium equations, cylindrical limit-load instability, 517
boundary conditions, 158 coordinates, 82 local average rotation, 94
boundary element method, 148, 361 equilibrium equations, plane strain,
buckling, 515 154 maximum distortional strain energy
equilibrium equations, plane stress, 84 theory, 499
Castigliano’s first theorem, 256 equilibrium equations, spherical maximum normal stress theory,
Castigliano’s second theorem, 260 coordinates, 83 496
center of twist, 433 equivalent stress, 61 maximum shearing stress theory,
centroid, 527, 540 Euler buckling load, see critical buckling 497
Clapeyron’s theorem, 229 load Maxwell stress function, 149
combined bending and torsion, thin, mean stress, 502
closed sections, 477 fatigue failure, 502 membrane analogy, 423
combined bending and torsion, thin, fatigue strength, 504 Michell solutions, 197
open sections, 475 finite differences method, 148 Miner rule, see Palmgren-Miner rule
compatibility equations, 101, 303 first area moments, 527, 540 mixed second area moment, 8, 527, 540
compatibility matrix, 304, 312 first sectorial moment, 542 modified sectorial area function, 459
complementary strain energy density, flexibility matrix, 299 modified warping constant, 462
233 flexure stress, unsymmetrical bending, Mohr’s circle, 53
compliance matrix, 124 375 moment–curvature relationship, 8
concentrated force on a planar surface, force-based finite element method, axial Morera stress function, 149
209 loads, 295 Morrow curve, 504
concentrated force on a wedge, 208 force-based finite element method, beam
configuration factor, 509 bending, 329 Navier’s equations, 147, 310
constant life curves, 503 force-based finite element method, Navier’s table problem, 166
critical buckling load, 516 general body, 358 Neumann problem, 435
critical flaw size, 511 force-based finite element method, neutral axis, 7
cubic material, 127 specified displacements, 347 nonuniform torsion, 427
curved beam, 202 four pillars, 18 normal strain, 3, 88
fracture mechanics, 508 normal stress, 4, 32
deviatoric stress, 67 fracture toughness, 510 normalized Prandtl stress function, 419
dilatation, 98 fundamental solutions, 362 null space, 314

565
566 Index

octahedral plane, 60 rosette strain gage, see strain gage strain invariants, 95
orthogonal matrix, 37 rosette strain transformation equations, 95
orthotropic material, 124 row-reduced echelon form, 313 strain–displacement relations, 91
strains, cylindrical coordinates, 112
Palmgren–Miner rule, 507 Saint-Venant torsional theory, see stress concentration at a circular hole,
parallel axis theorem, 531 uniform torsion 198
Paris law, 512 Saint-Venant’s principle, 161 stress intensity factor, 509
plane strain, 154 second area moments, 8, 527, 540 stress invariants, 43
plane stress, 51, 73 secondary twisting moment, 436 stress range, 502
planes of extreme shear stress, 56 secondary warping, 428 stress singularities, 214
Poisson’s equation, 418 sectorial area function, 391, 540 stress transformation equations, 34
Poisson’s ratio, 3 sectorial area origin, 541 stress vector, 28
polar area moment, 15, 540 sectorial area pole, 541
Prandtl stress function, 149, 417 sectorial products of areas, 527, 542 tensor shear strain, 90
primary twisting moment, 436 sectorial second moment, see warping theorem of minimum potential energy,
primary warping function, 428 constant 242
principal area moments, 531 shape functions, 281 thick-wall pressure vessel, 183
principal directions, 42 shear stress, 30 total shear stress, 32
principal origin, 549 shear center, 389 traction vector, see stress vector
principal planes, 42 shear flow, 17, 385 transformation of elastic constants, 132
principal pole, 545 shear flow expression, unsymmetrical transversely isotropic material, 126
principal sectorial area function, 527, bending, 386 Tresca theory, see maximum shearing
543 shear force, 9 stress theory
principal strains, 95 shear modulus, 14 true stress, 504
principal stresses, 42 shear strain, 89
principle of complementary virtual shear stress, 9 ultimate stress, 504
work, 250 shrink-fit, 187 ultrasonic NDE, 513
principle of complementary virtual work single plane of loading, 376 uniform torsion, 411
- axial loads, 280 single-valued displacement, 103 uniqueness, 243
principle of complementary virtual work singularity functions, 20 unit load method, 262
- beam bending, 319 S–N curve, 502
principle of least work, 264 snap-through buckling, 517 virtual displacements, 239
principle of minimum complementary Soderberg curve, 504 virtual work, 240
potential energy, 251 Somigliana’s identity, 364 Voigt notation, 124
principle of virtual work, 240 state of strain, 93 von Mises failure theory, 499
principle of virtual work, axial loads, state of stress, 29 von Mises stress, 61, 500
278 stiffness matrix, 148, 283
principle of virtual work, beam bending, stiffness-based finite element method, warping constant, 432, 527
316 148 warping constant, thin sections, 447
proper orthogonal matrix, 37 stiffness-based finite element method, warping function, 411
pseudostiffness matrix, 311, 333 axial loads, 281 warping normal stress, 428
pure bending of a curved beam, 202 stiffness-based finite element method, work, 224
beam bending, 320
Rayleigh–Ritz method, 257 stiffness-based finite element method, X-ray NDE, 514
reciprocal theorem of Betti–Rayleigh, general body, 356
268 strain energy, 226 yield normal stress, 498
reciprocity, 267 strain energy density, 226 yield shear stress, 497
redundants, 263 strain gage rosette, 141 Young’s modulus, 5