(Isaac Elishakoff, Yiwei Li, James H. Starnes JR) (B-Ok - Xyz) PDF

Non-Classical Problems in the Theory of Elastic Stability
When a structure is put under an increasing compressive load, it may become

unstable and buckle. Buckling is a particularly significant concern in designing
shell structures such as aircraft, automobiles, ships, or bridges. This book discusses
stability analysis and buckling problems and offers practical tools for dealing with
uncertainties that inherently exist in real systems. The techniques are based on two
competing yet complementary theories, which are developed in the text. First, the
authors present the probabilistic theory of stability, with particular emphasis on
reliability. Both theoretical and computational issues are discussed. Second, they
present the alternative to probability based on the notion of “anti-optimization,”
a theory that is valid when the necessary information for probabilistic analysis is
absent, that is, when only scant data are available. Design engineers, researchers,
and graduate students in aerospace, mechanical, marine, and civil engineering and
theoretical and applied mechanics who are concerned with issues of structural
reliability and integrity will find this book a particularly useful reference.
Isaac Elishakoff is a professor in the Department of Mechanical Engineering at

Florida Atlantic University.
Yiwei Li is an engineering software developer at Alpine Engineered Products, Inc.
James H. Starnes, Jr., is the head of the Structural Mechanics Branch at NASA
Langley Research Center.
Non-Classical Problems in the
Theory of Elastic Stability
Deterministic, Probabilistic and
Anti-Optimization Approaches
ISAAC ELISHAKOFF
Florida Atlantic University
YIWEI LI
Alpine Engineered Products, Inc.
JAMES H. STARNES, JR.

NASA Langley Research Center
Dedicated to the memory
of Professor Dr. Ir. Warner Tjardus Koiter
– colleague, co-author, mentor, and friend
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 2RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521782104
© Cambridge University Press 2001
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 2001

This digitally printed first paperback version 2005
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data

Elishakoff, Isaac. 1944–
Non-classical problems in the theory of elastic stability / Isaac Elishakoff, Yiwei Li,
James H. Starnes, Jr.
p. cm.
Includes bibliographical references and index.
ISBN 0-521-78210-4
1. Structural stability – Statistical methods. 2. Probabilities. 3. Elastic analysis
(Engineering) 4. Buckling (Mechanics) I. Li, Yiwei. II. Starnes, James H. III. Title.
TA656 .E45 2000
624.1´7 – dc21
00-023667
ISBN-13 978-0-521-78210-4 hardback

ISBN-10 0-521-78210-4 hardback
ISBN-13 978-0-521-02010-7 paperback

ISBN-10 0-521-02010-7 paperback
Contents
Preface: Why Still Another Book on Stability? page ix
1 Mode Localization in Buckling of Structures 1

1.1 Localization in Elastic Plates Due to Misplacement
in the Stiffener Location 1
1.2 Localization in a Multi-Span Periodic Column with a Disorder
in a Single Span 17
1.3 Localization Phenomenon of Buckling Mode
in Stiffened Multi-Span Elastic Plates 30
2 Deterministic Problems of Shells with Variable Thickness 43

2.1 Introductory Remarks 43
2.1.1 Basic Equations for Homogeneous Shells 44
2.1.2 Hybrid Perturbation-Weighted Residuals Method 46
2.1.3 Solution by Finite Difference Method 50
2.1.4 Solution by Godunov-Conte Shooting Method 52
2.1.5 Numerical Results and Discussion 52
2.2 Buckling of an Axially Compressed Imperfect Cylindrical Shell
of Variable Thickness 53
2.2.1 Direct Discussion of Energy Criterion 54
2.2.2 Numerical Technique 57
2.3 Axial Buckling of Composite Cylindrical Shells
with Periodic Thickness Variation 71
2.4 Effect of the Thickness Variation and Initial Imperfection
on Buckling of Composite Cylindrical Shells 89
2.5 Effect of the Dissimilarity in Elastic Moduli on the Buckling 99
2.5.1 Analysis 99
2.5.2 Discussion 105
3 Stochastic Buckling of Structures: Monte Carlo Method 107

v
vi CONTENTS
3.2 Reliability Approach to the Random Imperfection Sensitivity

of Columns 109
3.2.1 Motivation for the Reliability Approach 109
3.2.2 The Linear Problem 110
3.2.3 Deterministic Imperfection Sensitivity:
Mixed Quadratic-Cubic Foundation 112
3.2.4 Single-Mode Solution 114
3.2.5 Multi-Mode Solution 116
3.2.6 Buckling Under Random Imperfections – Monte Carlo Method 117
3.2.7 Numerical Examples 120
3.3 Non-Linear Buckling of a Structure with Random Imperfection and
Random Axial Compression by a Conditional Simulation Technique 127
3.3.1 Deterministic Procedure 127
3.3.2 Formulation of Basic Random Variables 129
3.3.3 Probabilistic Analysis 130
3.3.4 Numerical Example and Discussion 134
3.4 Reliability of Axially Compressed Cylindrical Shells
with Random Axisymmetric Imperfections 137
3.4.1 Preliminary Considerations 137
3.4.2 Probabilistic Properties of Initial Imperfections 139
3.4.3 Simulation of Random Imperfections with Given
Probabilistic Properties 144
3.4.4 Simulation of Random Initial Imperfections
from Measured Data 148
3.4.5 Computation of the Buckling Loads 154
3.4.6 Comparison of the Monte Carlo Method
with the Benchmark Solution 157
with Nonsymmetric Imperfections 161
3.5.1 Probabilistic Properties and Simulation
of the Initial Imperfections for a Finite Shell 161
3.5.2 Multi-Mode Deterministic Analysis for Each Realization
of Random Initial Imperfections 165
3.5.3 Numerical Results and Discussion 167
4 Stochastic Buckling of Structures: Analytical and Numerical

Non–Monte Carlo Techniques 175
4.1 Asymptotic Analysis of Reliability of Structures in Buckling Context 175
4.1.1 Overview of the Work by Ikeda and Murota 177
4.1.2 Finite Column on a Non-Linear Foundation 179
4.1.3 Application of Ikeda-Murota Theory 181
4.1.4 Axially Compressed Cylindrical Shells 183
4.2 Second-Moment Analysis of the Buckling of Isotropic Shells
with Random Imperfections 186
4.3 Use of STAGS to Derive Reliability Functions by Arbocz and Hol 193
CONTENTS vii
4.4 Reliability of Composite Shells by STAGS 196

4.5 Buckling Mode Localization in a Probabilistic Setting 200
4.6 Finite Element Method for Buckling of Structures
with Stochastic Elastic Modulus 210
4.7 Stochastic Finite Element Formulation by Zhang and Ellingwood 214
5 Anti-Optimization in Buckling of Structures 222

5.1 Incorporation of Uncertainties in Elastic Moduli 223
5.1.1 Basic Equations 223
5.1.2 Extremal Buckling Load Analysis 228
5.1.3 Determination of Convex Set from Measured Data 231
5.1.4 Numerical Examples and Discussion 233
5.1.5 Numerical Analysis by Non-Linear Programming 237
5.2 Critical Contrasting of Probabilistic and Convex Analyses 238
5.2.1 Is There a Contradiction Between Two Methodologies? 238
5.2.2 Deterministic Analysis of a Model Structure
for a Specified Initial Imperfection 241
5.2.3 Probabilistic Analysis 247
5.2.4 Non-Stochastic, Anti-Optimization Analysis 251
6 Application of the Godunov-Conte Shooting Method

to Buckling Analysis 257
6.2 Brief Outline of Godunov-Conte Method As Applied
to Eigenvalue Problems 260
6.3 Application to Buckling of Polar Orthotropic Annular Plate 261
7 Application of Computerized Symbolic Algebra

in Buckling Analysis 268
7.2 Brief Review of MATHEMATICA® 271
7.3 Buckling of Polar Orthotropic Circular Plates on Elastic Foundation
by Computerized Symbolic Algebra 274
7.3.1 Results Reported in the Literature 274
7.3.2 Statement of the Problem 276
7.3.3 Buckling of Clamped Polar Orthotropic Plates 277
7.3.4 Buckling of Simply Supported Polar Orthotropic Plates 285
7.3.5 Buckling of Free Polar Orthotropic Plates 288
Bibliography 290
Author Index 327
Subject Index 331
Preface: Why Still Another Book on Stability?
The preface is the most important part of the book. Even reviewers read a preface.
Philip Guedalla
There are at present numerous books available on the theory of stability and its appli-
cations to structures. One author even remarked sarcastically that if they were put in a
single bookcase, it would buckle under their weight. We do not complain that “. . . of
the making of many books there is no end” (Ecclesiastes 12:12), rather we ask a natural
question: Is there a legitimate place for a new book in this field?
The answer to this question is affirmative, if a book has its unique, distinct char-
acteristics. We have chosen to deal with non-classical problems. To the best of our
knowledge, none of the subjects, touched upon in this monograph, have been discussed
exclusively in the existing books on buckling analysis. Thus we feel that this book will
not be just another new book on buckling. Indeed, most existing books may be classified
as belonging to one of the following two categories: textbooks – which often look very
much alike, maybe not without reason, since the subject is the same – and monographs –
which have an encyclopedic nature, trying to comprise an uncomprisable – to cover all
or nearly all pertinent topics. This latter task of listing all the results (even only those
of major importance) appears to be impossible indeed.
The purpose of this book is to present two competing theories, which incorporate
ever-present uncertainty in the stability applications of the real world. These uncertain-
ties are first and foremost due to unavoidable initial imperfections, deviations of the
structure from its intended, nominal, ideal shapes. Other uncertainties are the material
characteristics and/or realizations of the boundary conditions. These topics are almost
never touched upon in the texts or monographs. Here we first present the probabilistic
theory of stability. The bridge is made between a description of random imperfections
as random fields, its description as random vectors, and the Monte Carlo methods.
Special emphasis is devoted to evaluation of reliability, the probability that the struc-
ture will not fail prior to preselected load level; reliability concept is the powerful
tool that needs to be introduced in the practical design of uncertain structures, if the
probabilistic paradigm is adopted. The book presents a unified probabilistic theory of
stability. It elucidates both the theoretical and computational aspects in a single package.
ix
x PREFACE
Special emphasis is placed on asymptotic evaluations and non–Monte Carlo analytical

methods.
Along with the probabilistic theory of stability, we expose the alternative to proba-
bility, based on the notion of “anti-optimization”; this theory is valid when the necessary
information for the probabilistic analysis is absent and only the scant data are avail-
able. Such a theory appears to be of prime importance for estimating the worst case
scenario with limited data present. Thus, these two theories complement each other.
Ideally, if sufficient data are available, one would resort to the probabilistic theory;
however, realistically, in most circumstances its alternative, anti-optimization, should
be preferred.
The main objective of this book is, therefore, to provide a strong impetus to both
theorists and practitioners so that they will become acquainted with the probabilistic and
anti-optimization theories as the most practical tools for dealing with the uncertainties
that are present whether theorists or practitioners want to acknowledge them or not
and whether they know the appropriate analyses or not. Uncertainty analysis is thus
not a luxury but rather a mere necessity to rigorously reflect the working conditions
of the real systems. This book, therefore, fills the gap that exists in the present-day
literature and practice. In fact, engineers and researchers can no longer pretend that
their deterministic approaches represent the truth, that directly, without invoking the
uncertainty aspects, could be utilized by the practicing engineers. Recognition of this
fact is extremely important not only to the practitioners but also for the theorists for they
would try, it is hoped, in their future analyses, to devote additional effort to the rigor of
their models, especially when the overlooked uncertainty may drastically change their
results obtained for ideal, maybe even never existing, situations.
From the engineering point of view, the developments in stability of structures
can be divided into two periods. The first period dates from 1744, when Leonhard
Euler communicated to the world his famous formula for the buckling load of a col-
umn. The second dates from 1945, when Warner Koiter submitted his PhD thesis to
the Delft University of Technology. He uncovered the disastrous effect of initial geo-
metric imperfections on the load-carrying capacity of shell structures. During the two
centuries of the post-Euler, pre-Koiter era, engineers and scientists made outstanding
contributions in shedding light on the buckling of structures. New problems arose with
the technological revolution that took place in the twentieth century. With the mod-
ern need to develop lightweight vehicles, the buckling concept has become even more
eminent. Lorenz, Southwell, and Timoshenko generalized Euler’s classical formula for
columns to the case of thin cylindrical shells. Fortunately, buckling specialists were
interested in experimental validation of their findings. This led to a surprising and
rather disappointing observation: The experimental results did not vindicate the theo-
retical investigations, most of the data were much lower than what the linear analyses
predicted. Koiter’s teachings appeared then to explain that the unavoidable imperfec-
tions, deviations from the nominally ideal shape, play a dominant role in the drastic
reduction of the load-carrying capacity of cylindrical shells and some other structures.
Thus, the notion of imperfection sensitivity came into being. This is how the buckling
topic emerged into healthy adolescence after two hundred years of childhood. The the-
oretical works of Budiansky and Hutchinson at Harvard University, of Thompson at
PREFACE xi
University College, London, and of G. W. Hunt of University of Bath, experimental

studies by Singer at the Technion–Israel Institute of Technology, combined theoretical-
numerical-experimental investigations by Arbocz at the Delft University of Technology,
to name just a few, have been spectacularly instrumental in providing this old yet still
mysterious field with new impetus and ideas.
Practical and theoretical people in the aeronautics and aerospace fields, who must
design real shells and who have invested considerable funds in the research on shell
structures and their stability problems, still have not fully adopted the concept of im-
perfection sensitivity but are using the so-called knockdown factor. This appears to
be the case since basically engineering scientists did not present to them a means of
incorporating theoretical findings into practice. Knockdown factor, which is defined
so that its product with the classical buckling load yields a lower bound to all existing
experimental data, was compiled primarily by Weingarten, Seide, and Peterson. This is
both welcome and disappointing. It is welcome given that many of the shell structures
we employ perform extremely responsible functions; it is better to overdesign than un-
derdesign, if we must choose between these two alternatives and do not want to take
excessive risks. On the other hand, it is disappointing that the results of several decades
of research are mostly ignored and unreflected in engineering practice and that His
Excellency the Knockdown Factor is reigning in the kingdom of the designers.
It has been recognized since 1958, when V. V. Bolotin of the Moscow Power Engi-
neering Institute introduced the notion of randomness into the theory of elastic stability,
that for the buckling theories to become practical, they must be combined with analy-
sis of the uncertain initial imperfections. We are aware that there are no two identical
structures, even if they are produced by the same manufacturing process. The appar-
ent reluctance of the designers to take advantage of theoretical findings stems from
the fact that most theoretical and/or numerical imperfection studies are conditional on
detailed a priori knowledge of the geometric imperfections of the particular structure
that is available. In an ideal case, the imperfections can be measured and incorporated
into the analyses to predict the buckling loads. For example, Horton of Georgia In-
stitute of Technology tested a large-scale shell with a diameter of 60 ft, and Arbocz
and Williams measured the imperfection profile of 10-ft diameter stiffened shells at
the NASA Langley Research Center. More recently, Ariane interstage shells (some of
them 80 ft in diameter) were measured in the Fokker Aircraft Company. This approach,
which is justified for single prototype-like structures, is impractical as a general method
for behavior prediction. Information on the type and magnitude of imperfections of a
particular structure would be too specific and hardly applicable to other realizations of
the same structure, even if they were produced by the same manufacturing process.
With these considerations in view and also bearing in mind the large scatter of the
experimental results, it becomes clear that practical applications of the imperfection-
sensitivity theories are conditional on their being fused with a probabilistic analysis
of the imperfections and buckling loads. This is because engineers do not want to
overdesign or underdesign the structures. Apparently, the knockdown factor approach
penalizes ingeniously designed shells, those with fewer imperfections.
However, appreciation of the probabilistic approach is not sufficient to solve the
problem. Indeed, the early studies on random imperfections fell into two classes. The
xii PREFACE
first class considered overly simplified models of single-degree-of-freedom finite struc-

tures with the initial imperfection amplitude as a random variable and with attendant
straightforward analysis. The second class treated the initial imperfections as random
fields (i.e., random functions of both the axial and circumferential coordinates). To
facilitate the purely (and often restrictive) analytical treatment, researchers adopted
far-reaching assumptions regarding the probabilistic nature of initial imperfection (sta-
tistical homogeneity and ergodicity); they also treated infinitely long structures rather
than ones of finite length. Thus a new thinking was necessary to discard negative char-
acteristics of existing probabilistic treatments, while retaining and reinforcing their
positive characteristics. We take the liberty of quoting Johann Arbocz’s extensive re-
view (1981) on the past, present, and future of shell stability analysis: “. . . it was not until
1979, when Elishakoff published his reliability study on the buckling of a stochastically
imperfect finite column on a nonlinear elastic foundation, that a method has been pro-
posed, which made it possible to introduce the results of the initial imperfection surveys
routinely into the analysis.” Basically, the idea was to utilize the Monte Carlo method
for solution of the stability problem of axially compressed cylindrical shells with ran-
dom initial imperfections. The latter are expanded in terms of the buckling modes of
the perfect structure, and then the Fourier coefficients are treated as random variables.
Next, using a special numerical procedure, the Fourier coefficients of the desired large
sample of random initial imperfection shapes are simulated after a sufficiently large
sample of initial imperfection measurements become available. This is followed by a
deterministic buckling load of each of the simulated shells, with subsequent statistical
analysis.
Meaningful probabilistic analysis is conditional on probabilistic characterization
of the input variables and functions. This can be performed only when appropriate
measurement data are available. Fortunately, Babcock and Arbocz at the California
Institute of Technology, as well as some other investigators realized the need for large-
scale experiments, as a result of which several initial-imperfection data banks have
been developed (those at the Delft University of Technology and the Technion–Israel
Institute of Technology being the most notable). Instead of making unnecessary and
often restrictive assumptions regarding the properties of the initial imperfections, these
data banks are amenable to direct statistical analysis, since they provide indispensable
tools for further probabilistic analysis.
One must recognize that, at first glance, the Monte Carlo method may not appear
to be attractive analytically; it may turn out to be numerically time-consuming espe-
cially for some highly complex structures. The answer to such a possible reservation is
that the Monte Carlo method is the only universal technique for probabilistic analysis
of structures and that its use does not entail some of the heavily simplified solution
procedures that are often required for idealized structural models.
Complex multi-mode non-linear deterministic analysis should be balanced by
maximum resemblance to real-life situations. As Timothy Ferris instructively put it,
“science is said to proceed on two legs, on the theory (or, loosely, of deduction) and the
other of observation and experiment (or induction). Its progress, however, is less often
a commanding stride than kind of halting stagger – more like the path of a wander-
ing minstrel than the straight-ruled trajectory of a military marching band.” We trust
PREFACE xiii
that the direct introduction of the results of experiments into the probabilistic analysis
puts us on the right track, combining both the deductive and the inductive facets of
engineering. This became feasible by employing a numerical tool closely resembling
the experiments themselves, namely by the Monte Carlo method. At the same time,
we do not advocate an abandonment of analytical techniques that, although applicable,
may complement the Monte Carlo method especially where small imperfections are
involved. The Monte Carlo method needs very accurate numerical techniques to evalu-
ate buckling loads of each shell in the ensemble of shells; careful numerical techniques
like STAGS, BOSOR, PANDA, and those based on multi-mode imperfection methods
are consistent with the reliability analysis.
Can one use probabilistic modeling if the data are extremely limited? This is an
intriguing question. The answer is affirmative for those who make a fetish of the proba-
bilistic models, but we feel that it should be negative. Indeed, the central premise of this
book is the need for sound theoretical, experimental, and numerical analyses. Rigorous
deterministic analysis is a cornerstone of every meaningful probabilistic treatment of a
problem at hand. Theoretical analysis elucidates the physical phenomenon but may not
be feasible without some idealizations. Experimental analysis then provides the data
used by the numerical analyst to create statistical “brothers and sisters” of the measured
shells. According to the John Wiley Dictionary of Scientific Terms, reliability is “the
probability that a component part, equipment, or system will satisfactorily perform its
intended function under given circumstances, such as environmental conditions, lim-
itations as the operating time, and frequency and thoroughness of maintenance, for a
specified period of time.” In the structural-stability context, reliability is the proba-
bility that the structure will sustain loads in excess of the specified level. Here either
extremely high reliabilities or equivalently extremely low probabilities of failure are
needed. To perform such a refined analysis, refined data and authentic analytical and
numerical tools are called for. In the absence of experimental data (as may well be
the case with variation of the elastic moduli), probabilistic methods cannot be recom-
mended for design purposes. Fortunately, there exists a new discipline, called convex
modeling of uncertainty (or, in a more general context, set-theoretical modeling of un-
certainty), developed by Ben-Haim and Elishakoff for applied-mechanics problems.
This discipline does not seek to make something out of nothing (i.e., it does not create
the probabilistic model out of extremely limited or absent data), but it does represent an
alternative technique for such special yet often encountered situations. Set-theoretical
uncertainty in effect represents anti-optimization under uncertainty: It operates with the
least favorable scenarios based on the limited experimental data describing uncertain
variables or functions.
As we saw earlier, accurate determination of the reliability or the least favorable
buckling loads calls for a threefold effort in: (a) deepening theoretical insight into
the phenomenon, (b) collecting experimental information, and (c) conducting proper
numerical analysis. Choice of a suitable model of uncertainty is also a prominent
decision to be made, based on the amount and character of the information.
This monograph reflects this philosophy. Neither a textbook nor an encyclopedia,
it covers the deterministic, probabilistic, and set-theoretical approaches for buckling
of structures. Chapters 1, 2 (except Sections 2.1 and 2.2), and 5 are written by three
xiv PREFACE
of us; the first two sections of Chapter 2 are based on our joint papers with Professor
Koiter of the Delft University of Technology on the effect of thickness variation in
perfect or imperfect isotropic shells. The last two sections of Chapter 2 represent a
generalization for the case of composite shells. Chapters 3 (except Section 3.2), 4,
6, and 7 are written by the first author; Section 3.2 is based on our joint paper with
Professor Masanobu Shinozuka of the University of Southern California. Chapters 3
and 4 are based on the work of the first author (Section 3.1), and his joint studies with
Professor Johann Arbocz of the Delft University of Technology (Sections 3.3, 3.4, and
4.2), Professor Koyohiro Ikeda of Tohoku University and Professor Kazuo Murota of the
Kyoto University (Section 4.1) and other investigators. Contents of works of Professor
Wei-Chau Xie of the University of Waterloo and of Dr. Jun Zhang and Professor Bruce
Ellingwood of Johns Hopkins University constitute the central part of the last two
sections of Chapter 4. Section 3.5 is based on our joint investigation with Dr. David
Bushnell of Lockheed Company. Section 5.3 is based on the joint paper by the first and
third authors with Professor Guoqiang Cai of the Florida Atlantic University.
The organization of the present book is as follows: The first two chapters are de-
voted to some new deterministic problems. Chapter 1 discusses the mode localization
in the deterministic setting, an extremely recent topic in the buckling context. In par-
ticular, we consider multi-span columns and plates, with unavoidable misplacements
of the stiffeners location. Generally, a misplacement can be regarded as one type of
imperfection, although it does not lead to as drastic a change in the buckling load as
geometric imperfections do in shell buckling. As a matter of fact, in many experimental
studies (e.g., of Singer or of Tenerelli and Horton) of the shell buckling, only localized
buckling modes were observed. In strong ring-stiffened shells, two or three buckles
emerged when the critical load was reached and were confined to a single span. In weak
ring-stiffened shells, buckles appeared around the whole circumference but extended
only through part of the spans. Chapter 1 advocates that, if each span is treated as an
element of a periodic structure, the localization phenomena can be explained by using
the analytical finite difference calculus. Chapter 2 is devoted to the influence of thick-
ness variation on the buckling of perfect or imperfect, isotropic or composite, circular
cylindrical shells.
Chapters 3 and 4 deal with stochastic buckling of structures with random imper-
fections. Chapter 3 focuses on the Monte Carlo method, whereas Chapter 4 discusses
approximate analytical and numerical techniques, including the asymptotic analysis,
the first-order second-moment method, the mode localization due to random misplace-
ments, and the finite-element method for structures with random material properties.
Convex modeling of uncertainty in buckling problems is the focal point of Chap-
ter 5. Here the realistic situation of data scarcity is considered, and the minimum
buckling loads in an ensemble of plates and shells with uncertain material properties
are derived. Chapter 6 discusses the Godunov-Conte shooting method (representing an
extended version of the paper by Elishakoff and Charmats), whereas Chapter 7 deals
with the application of computerized symbolic algebra – an able and obedient “ser-
vant” of the present-day researchers. It constitutes an extended version of the paper by
Elishakoff and Tang.
We hope that the monograph will prove useful for researchers, engineers, and senior
graduate students specializing in aeronautical, aerospace, mechanical, civil, nuclear,
PREFACE xv
and marine engineering. We hope that our “pluralistic” philosophy will have a definite
impact on the entire engineering profession. As Abraham Maslow remarks, “If the
only tool you have is a hammer, you tend to treat everything as if it were a nail.”
We strongly trust that the modern analyst cannot afford to confine himself/herself to
only one of the trio, namely, (a) deterministic, (b) probabilistic, and (c) set-theoretical
Weltanschauung. Instead, we, as engineers must be pragmatic and flexible in choosing
the most appropriate methods, consistent with available experimental information.
Some pertinent questions remain still to be tackled. Two of the central questions
were posed by Professor Bernard Budiansky in his correspondence with Professor
Johann Arbocz (Arbocz and Singer, 2000): “Are we necessarily doomed to accept
forever the unhappy coexistence of efficient shell design and imperfection-sensitivity?
Or are there some structural design secrets to be discovered that will retain minimum
weight and rid us of the curse of imperfection-sensitivity?”
This book deals with how to live with the above curse most efficiently, combining
deterministic probabilistic and anti-optimization “cures,” in the hope that some “genes”
will be uncovered that will make the multiplicity of studies on the “curse” unnecessary
even if still instructive.
We thank (a) Elsevier Science Publishers for allowing us to reproduce Fig-
ures 3.13–3.29 from the article by I. Elishakoff and J. Arbocz, “Reliability of Axially
Compressed Cylindrical Shells with Random Axisymmetric Imperfections,” Interna-
tional Journal of Solids and Structures, Vol. 18, 563–85, 1982; Figures 4.1–4.4 from
the article by K. Ikeda, K. Murota, and I. Elishakoff, “Reliability of Structures Sub-
ject to Normally Distributed Initial Imperfections,” Computers and Structures, Vol. 59,
463–9, 1995; Figures 5.10–5.14 from the study by I. Elishakoff, G. Q. Cai, and J. H.
Starnes, Jr., “Nonlinear Buckling of a Column with Initial Imperfection via Stochastic
and Non-Stochastic Convex Models,” International Journal of Nonlinear Mechanics,
Vol. 29, 71–82, 1994; Figure 3.34 from the book Buckling of Structures – Theory and
Experiment (Elsevier, 1988), edited by I. Elishakoff, J. Arbocz, C. D. Babcock, Jr.,
and A. Libai; (b) AIAA Press for permission to reproduce Figures 4.5–4.6 from the
article by I. Elishakoff, S. Van Manen, P. G. Vermeulen, and J. Arbocz, “First-Order
Second-Moment Analysis of the Buckling of Shells with Random Initial Imperfec-
tions,” AIAA Journal, Vol. 25, 1113–17, 1987; Figures 4.7–4.11 from the paper by
W.-C. Xie, “Buckling Mode Localization in Randomly Disordered Multispan Con-
tinuous Beam,” AIAA Journal, Vol. 33, 1142–9, 1995; (c) American Society of Civil
Engineering for permission to reproduce Figures 4.12–4.14 from the paper by J. Zhang
and B. Ellingwood, “Effects of Uncertain Material Properties on Structural Stability,”
Journal of Structural Engineering, Vol. 121, 705–16, 1995; and (d) American So-
ciety of Mechanical Engineering for the permission to reproduce Figures 3.30–3.33
and 3.35–3.37 from the paper by I. Elishakoff and J. Arbocz, “Reliability of Axially
Compressed Cylindrical Shells with General Nonsymmetric Imperfections,” Journal
of Applied Mechanics, Vol. 52, 122–8, 1985.
The authors gratefully acknowledge scientific cooperation, over the last two
decades, on various buckling problems with the following colleagues: Professor J.
Arbocz, Ir. J. van Geer, Professor J. Kalker, Professor W. T. Koiter, Ir. A. Scheurkogel,
Professor W. D. Verduyn, and Ir. P. G. Vermeulen of Delft University of Technology;
Professor J. Ari-Gur of Western Michigan University; Professor M. Baruch, Professor
xvi PREFACE
Y. Ben-Haim, Mr. M. Charmatz, Mr. E. Itzhak, Dr. S. Marcus, Professor J. Singer,

and Professor Y. Stavsky of the Technion–Israel Institute of Technology; Professor C.
W. Bert of the University of Oklahoma; Professor V. Birman of the University of
Missouri–Rolla; Dr. D. Bushnell of Lockheed Missiles and Space Company, Inc.;
Professor G. Q. Cai, Dr. L. P. Zhu, and Dr. M. Zingales of Florida Atlantic
University; Professor K. Ikeda of Tohoku University, Japan; Dr. S. van Manen of TNO
(the Netherlands); Professor K. Murota of Kyoto University, Japan; Dr. W. J. Stroud
of the NASA Langley Research Center; Professor M. Shinozuka of the University of
Southern California; and Professor W. C. Xie of the University of Waterloo.
Over the last two decades, helpful discussions on aspects in uncertain buckling
with Professor J. Amazigo of the University of Nigeria; Professor S. T. Ariaratnam of
the University of Waterloo; Professor G. Augusti of the University of Rome; Professor
V. V. Bolotin, Professor B. P. Marakov, and Dr. V. M. Leizerakh of the Moscow Power
Engineering Institute–State University; Professor C. D. Babcock, Jr. of Caltech; Pro-
fessor B. Budiansky of Harvard University; Professor B. Frazer of the University of
New South Wales; Professor S. H. Crandall of M.I.T; Professor N. J. Hoff of Stanford
University; Professor I. Kaplyevatsky and Professor I. Sheinman of the Technion–Israel
Institute of Technology; Professor D. Shilkrut of the University of Negev; Professor
S. Krenk of Lund University; Professor G. Palassoupulos of the Military Academy of
Greece; Dr. M. Stein of the NASA Langley Research Center; Professor R. C. Tennyson
and Professor J. Hansen of the University of Toronto; and others (who may have been
unintentionally overlooked) are most appreciated. Most work reported in this book was
made possible through a grant from NASA Langley Research Center. This help is grate-
fully acknowledged. First author appreciates spending July 2000 at the Delft University
of Technology as the W. T. Koiter Visiting Chair Professor, and especially the warm
and kind hospitality of Professor Rene de Borst and Professor Fred van Keulen; during
this stay the final touches of the manuscript were completed.
The responsibility for any imperfections in this monograph, be they random or oth-
erwise, rests solely upon the authors. We would appreciate hearing your comments via
electronic mail ([email protected]), fax (561-297-2825), or regular correspondence.
CHAPTER ONE
Mode Localization in Buckling of Structures
Think globally, act locally.

Anonymous
The genuine spirit of localism . . .

George Borrow
This chapter investigates the buckling mode localization in the periodic multi-span beams
and plates. We start our discussion with disorder in two- and three-span elastic plates; then
we focus our attention on the multi-span beams and plates with a disorder occurring in
an arbitrary single span. The analytical finite difference calculus is employed to derive the
transcendental equations from which buckling load is calculated. The underlying treatment
is general, and the solution thus obtained is exact within the theory used. Numerical results
show that the buckling mode is highly localized in the vicinity of the disordered span of the
beam or the plate. In the multi-span, elastic plates are considered with transverse stiffeners,
and the discreteness of the stiffeners is accounted for. The torsional rigidity of the stiffener
is found to play an important role in the buckling mode pattern. When the torsional rigidity
is properly adjusted, the stiffener can be used in passive control; that is, it can serve as an
isolator of deformation for the structure at buckling so that the deflection is limited to only
a small area.
1.1 Localization in Elastic Plates Due to Misplacement

in the Stiffener Location
Traditionally, the stability of the stiffened plate has been studied following three differ-
ent settings. One approach consists of replacing the stiffened plate by an “equivalent”
orthotropic plate after the stiffeners are smeared out, in an energetic sense, over the entire
surface of the plate (Brush and Almroth, 1975; McFarland, Smith, and Bernhart, 1972).
This approach appears reasonable for plates with many closely spaced stiffeners but
is invalid for plates with very few stiffeners. The second approach is based on energy
consideration and treats the contributions of the plate and the stiffeners separately;
the Rayleigh-Ritz method has been utilized widely to estimate the buckling load of
1
2 MODE LOCALIZATION IN BUCKLING OF STRUCTURES
the stiffened plate structures (Bulson, 1969; Liew and Wang, 1990). This method may
predict well the global buckling but fails to detect the possible localization of buckling
mode due to the small changes in the location of the stiffeners. The third approach is
the method applicable to equally spaced stiffeners by the analytical finite difference
calculus (Bleich, 1952; Wah and Calcote, 1970). The latter method, though powerful
for studying plates with periodically spaced stiffeners or supports, is inapplicable if
the periodicity is disturbed as is usually the case when misplacement in the location
of the stiffener or support is present through imprecision of construction or assembly.
Despite their usefulness and simplicity, the previously mentioned methods can only
be employed to investigate the global buckling of the structure and appear incapable
of revealing the localization phenomena when the structure is sparsely or irregularly
stiffened and the buckling mode is likely to be localized.
Localization phenomenon was first uncovered by the Nobel laureate in physics
P. W. Anderson (1958). Its occurrence in structures has recently attracted much attention.
Much research has been conducted in recent years to study the localization phenomenon
in vibration of structures (for example, Hodges, 1982; Hodges and Woodhouse, 1983;
Pierre and Dowell, 1987; Pierre and Chat, 1989; Pierre, 1990) and in acoustics (De
Jong, 1994; Maidanik and Dickey, 1996). Cai and Lin (1991) studied the localization
of wave propagation in randomly disordered periodic structures. Apparently, Pierre and
Plaut (1989) were the first investigators to consider the localization in buckling; they
studied the simplest two-span column with a deterministic disorder. A more general
case, the multi-span column, was recently treated by Nayfeh and Hawwa (1994a, 1994b)
using the transfer matrix method. Ariaratnam and Xie (1996) investigated the localiza-
tion in the buckling of a system of rigid bars connected with springs in the stochastic
setting. Xie (1995) studied the buckling mode localization in randomly disordered
multi-span beams by the finite-element method. Tvergaard and Needleman (1983) dis-
cussed the development of localized patterns in the elastic-plastic and thermal buckling
problems. The deterministic buckling localization in cylindrical shells was investigated
by El Naschie (1975a, 1975b, 1977, 1990).
In this section, we investigate the effect of small structural irregularity, due to
the misplacement of stiffeners or interior supports, on both the buckling load and
the buckling mode of the rib-stiffened plate. Since the buckling mode shape is of
main interest, the interaction between the plate and stiffeners should be properly taken
into account. Here, the integration of the general governing differential equation is
performed for the stiffened elastic plate. By considering the rib-stiffened plate as a
physically continuous plate with as many spans as the number of ribs, the stiffeners are
accounted for through the conditions of continuity. Two cases commonly encountered
in practice are considered: one with simple support under the ribs and one without. It is
found that in the presence of small misplacement of stiffeners or interior supports, the
buckling mode shapes experience dramatic changes to become strongly localized. We
will first deal with the localization phenomenon in the buckling of a two-span plate with
a single rib using different parameters for the stiffener. Furthermore, a stiffened three-
span plate will be investigated, and the optimal configuration of stiffener placement,
which yields the highest buckling strength, will be discussed along with the attendant
localization sensitivity to deterministic misplacement.
1.1 LOCALIZATION IN ELASTIC PLATES DUE TO MISPLACEMENT IN THE STIFFENER LOCATION 3
Figure 1.1 A simply supported rectangular plate reinforced with transverse stif-
feners.
We first consider a rib-stiffened rectangular plate subjected in its mid-plane to

uniform compression P in the x direction (Figure 1.1). The differential equation of the
deflection surface of the plate under consideration is
4
∂ w ∂ 4w ∂ 4w ∂ 2w
D + 2 + + P =0 (1.1)
∂x4 ∂ x 2∂ y2 ∂ y4 ∂x2
where w is the transverse displacement, downward positive; D is the flexural rigid-
ity of the plate. The solution of Equation (1.1) can be represented in the following
form:

πy
w(x, y) = X (x) sin (1.2)
b
Substitution of Equation (1.2) into Equation (1.1) results in

d4 X P π 2 d2 X π4
+ − 2 + X = 0, X = Aesx (1.3)
dx4 D b2 d x 2 b4
The corresponding characteristic equation reads

P π2 2 π4
s +
4
−2 2 s + 4 =0 (1.4)
D b b
or

P π2 P P π2
s2 = − − 2 ± −2 2 (1.5)
2D b 2D 2D b
For rib-stiffened, or intermediately supported, plates, we have roots si (i = 1, 2, 3, 4)

as follows:
s1 = iβ1 , s2 = −iβ1 , s3 = iβ2 , s4 = −iβ2 (1.6)
where
1/2
P π2 P P π2
β1 = − 2
+ −2 2 ,
2D b 2D 2D b
1/2
(1.7)

P π2 P P π2
β2 = − 2 − −2 2
2D b 2D 2D b
Even for the unstiffened plate, the buckling load Pcr is always equal to or larger than
4π 2 D/b2 (Timoshenko and Gere, 1961). Therefore, the expression p/2D − 2π 2 /b2 is
non-negative. Thus, β1 and β2 are both real quantities. The solution of Equation (1.1)
thus can be written as
πy
w(x) = [A cos(β1 x) + B sin(β1 x) + C cos(β2 x) + D sin(β2 x)] sin (1.8)
b
where A, B, C, and D are constants of integration, which are to be determined using
the continuity and boundary conditions. For the arbitrary, jth span, the solution can be
written as
w j (x j ) = [A j cos(β1 x j ) + B j sin(β1 x j ) + C j cos(β2 x j )

πy
+ D j sin(β2 x j )] sin , 0 ≤ xj ≤ aj (1.9)
b
where a j is the length of the jth span, and j ranges from 1 to N for an N -span plate. We
consider the plate simply supported along its periphery. The boundary conditions read
w1 |x1 =0 = 0
2
∂ w1 ∂ 2 w1
Mx(1) x1 =0 = −D + ν =0
∂ x12 ∂ y 2 x1 =0
2 (1.10)
∂ wN ∂ 2 w N
Mx(N ) x N =a N = −D + ν =0
∂ x N2 ∂ y 2 x N =a N
w N |x N =a N = 0
where Mx(1) and Mx(N ) are the bending moments in the first and last spans of the contin-
uous plate; ν is the Poisson’s ratio. In view of (1.9), the preceding boundary conditions
become
A1 + C 1 = 0 (1.11)
β12 A1 + β22 C1 = 0 (1.12)
β12 cos(β1 a N )A N + β12 sin(β1 a N )B N + β22 cos(β2 a N )C N + β22 sin(β2 a N )D N = 0
(1.13)
cos(β1 a N )A N + sin(β1 a N )B N + cos(β2 a N )C N + sin(β2 a N )D N = 0 (1.14)
Regarding the continuity conditions between two successive spans, two cases of
practical interest deserve consideration.
Case A: Simple support under the rib. In some applications, the flexural rigidity
of the stiffener is not large enough, and a vertical support is installed under the stiffener
to suppress the transverse displacement. In this case, the continuity conditions between
the two typical neighboring spans numbered j and j + 1 are

w j+1 |x j+1 =0 = 0
w j |x j =a j = 0

∂w j ∂w j+1
=
∂ x j x j =a j ∂ x j+1 x j+1 =0

∂ 3 w j+1
Mx( j+1) x j+1 =0 − Mx( j) x j =a j = (GJ ) j
∂ x j+1 ∂ y 2 x j+1 =0
or
2
∂ 2 w j+1 ∂ 2 w j+1 ∂ wj ∂ 2 w j
−D + ν + D + ν
∂ x 2j+1 ∂ y2 x j+1 =0 ∂ x 2j ∂ y 2 x j =a j

∂ 3 w j+1
= (GJ ) j (1.15)
∂ x ∂ y2
j+1 x j+1 =0
where (GJ ) j denotes the torsional rigidity of the jth rib.

Substituting Equations (1.9) into the preceding conditions of continuity leads to
the following four equations:
A j+1 + C j+1 = 0 (1.16)
cos(β1 a j )A j + sin(β1 a j )B j + cos(β2 a j )C j + sin(β2 a j )D j = 0 (1.17)
−β1 sin(β1 a j )A j + β1 cos(β1 a j )B j − β2 sin(β2 a j )C j
+ β2 cos(β2 a j )D j − β1 B j+1 − β2 D j+1 = 0 (1.18)
−β12 cos(β1 a j )A j − sin(β1 a j )B j − cos(β2 a j )C j − sin(β2 a j )D j
β12 β22 β22
(GJ ) j π 2 (GJ ) j π 2
+ β12 A j+1 + β 1 B j + β2
2
C j + 1 + β2 D j + 1 = 0 (1.19)
D b2 D b2
Case B: No support under the rib. In this case, the bending, in addition to the
torsion, of ribs should be taken into account. The conditions of continuity between two
consecutive spans j and j + 1 read
w j |x j =a j = w j+1 |x j+1 =0

∂w j ∂w j+1
=
∂ x j x j =a j ∂ x j+1 x j+1 =0

∂ 3 w j+1
Mx( j+1) x j+1 =0 − Mx( j) x j =a j = (GJ ) j
∂ x j+1 ∂ y 2 x j+1 =0
or
2
∂ 2w j ∂ 2 w j ∂ w j+1 ∂ 2 w j + 1
+ ν − + ν
∂ x 2j ∂ y 2 x j =a j ∂ x 2j+1 ∂ y2
x j + 1 =0

∂ w j+1
3
= (GJ ) j
∂ x j+1 ∂ y 2 x j+1 =0

( j+1)

( j) ∂ 4 w j+1
Vx − Vx x j =a j = (EI ) j
x j+1 =0 ∂ y 4 x j+1 =0
or
3
∂ 3w j ∂ 2w j ∂ w j+1 ∂ 2 w j+1
+ (2 − ν) − + (2 − ν)
∂ x 3j ∂ x j ∂ y 2 x j =a j ∂ x 3j+1 ∂ x j+1 ∂ y 2 x j+1 =0

∂ 4 w j+1
= (EI ) j
∂ y 4 x j+1 =0
(1.20)
where Vx( j)
and are the shearing forces in the jth and ( j + 1)th spans of the plate;
Vx( j+1)
(EI ) j is the flexural rigidity of the jth rib.
These conditions of continuity can, in turn, be expressed by the following equations
in terms of the constants of integration:
cos(β1 a j )A j + sin(β1 a j )B j + cos(β2 a j )C j + sin(β2 a j )D j − A j+1 − C j+1 = 0
(1.21)
−β1 sin(β1 a j )A j + β1 cos(β1 a j )B j − β2 sin(β2 a j )C j + β2 cos(β2 a j )D j
− β1 B j+1 − β2 D j+1 = 0 (1.22)
−β12 cos(β1 a j )A j − β12 sin(β1 a j )B j − β22 cos(β2 a j )C j − β22 sin(β2 a j )D j
(GJ ) j π 2 (GJ ) j π 2
+ β12 A j+1 + β 1 B j+1 + β 2
2 C j+1 + β2 D j+1 = 0 (1.23)
D b2 D b2
β13 sin(β1 a j )A j − β13 cos(β1 a j )B j + β23 sin(β2 a j )C j − β23 cos(β2 a j )D j

(EI ) j π 4 (EI ) j π 4
− A j+1 + β1 B j+1 −
3
C j+1 + β23 D j+1 = 0 (1.24)
D b D b
By introducing the following non-dimensional quantities
Pb2 aj (GJ ) j (EI ) j
λ= 2 , rj = , τj = , ωj = ( j = 1 ∼ N)
π D b bD bD
1/2 1/2
λ λ λ λ λ λ
β̄ 1 = −1+ −2 , β̄ 2 = −1− −2
2 2 2 2 2 2
(1.25)
the boundary conditions for the simply supported continuous plate, Equations (1.11)–
(1.14), can be written as
A1 + C 1 = 0 (1.26)
β̄ 21 A1 + β̄ 22 C1 =0 (1.27)
β̄ 21 cos(β̄ 1r N π )A N + β̄ 21 sin(β̄ 1r N π )B N + β̄ 22 cos(β̄ 2r N π )C2
+ β̄ 22 sin(β̄ 2r N π )D N = 0 (1.28)
cos(β̄ 1r N π )A N + sin(β̄ 1r N π )B N + cos(β̄ 2r N π )C N + sin(β̄ 2r N π )D N = 0 (1.29)
The conditions of continuity for Case A, Equations (1.16)–(1.19), are transformed
into the following equations:
A j+1 + C j+1 = 0 (1.30)
cos(β̄ 1r j π)A j + sin(β̄ 1r j π )B j + cos(β̄ 2r j π )C j + sin(β̄ 2r j π )D j = 0 (1.31)
−β̄ 1 sin(β̄ 1r j π )A j + β̄ 1 cos(β̄ 1r j π )B j − β̄ 2 sin(β̄ 2r j π )C j + β̄ 2 cos(β̄ 2r j π )D j

− β̄ 1 B j+1 − β̄ 2 D j+1 = 0 (1.32)
− β̄ 21 cos(β̄ 1r j π )A j − β̄ 21 sin(β̄ 1r j π )B j − β̄ 22 cos(β̄ 2r j π )C j − β̄ 22 sin(β̄ 2r j π )D j
+ β̄ 21 A j+1 + τ j π β̄ 1 B j+1 + β̄ 22 C j+1 + τ j π β̄ 2 D j+1 = 0 (1.33)
For Case B, Equations (1.21)–(1.24) are rendered into the following form:
cos(β̄ 1r j π)A j + sin(β̄ 1r j π )B j + cos(β̄ 2r j π )C j + sin(β̄ 2r j π )D j

− A j+1 − C j+1 = 0 (1.34)
− β̄ 1 sin(β̄ 1r j π )A j + β̄ 1 cos(β̄ 1r j π )B j − β̄ 2 sin(β̄ 2r j π )C j + β̄ 2 cos(β̄ 2r j π )D j
− β¯1 B j+1 − β¯2 D j+1 = 0 (1.35)
− β̄ 21 cos(β̄ 1r j π )A j − β̄ 21 sin(β̄ 1r j π )B j − β̄ 22 cos(β̄ 2r j π )C j − β̄ 22 sin(β̄ 2r j π )D j
+ β̄ 21 A j+1 + τ j π β̄ 1 B j+1 + β̄ 22 C j+1 + τ j π β̄ 2 D j+1 = 0 (1.36)
β̄ 31 sin(β̄ 1r j π )A j − β̄ 31 cos(β̄ 1r j π )B j + β̄ 32 sin(β̄ 2r j π )C j − β̄ 32 cos(β¯2r j π )D j
− ω j π A j+1 + β̄ 31 B j+1 − ω j πC j+1 + β̄ 32 D j+1 = 0 (1.37)
For a general N -span continuous plate, we have four equations for boundary con-
ditions in the form of Equations (1.26)–(1.29). Since there are four equations for each
rib or interior support such as Equations (1.30)–(1.33) or Equations (1.34)–(1.37),
4 × (N − 1) equations can be established from the continuity considerations. Alto-
gether, there are 4N algebraic equations for the same number of unknown coefficients
A j , B j , C j , and D j ( j = 1 ∼ N ),
[F(λ)]4×N {∆} N ×1 = 0 (1.38)
where elements of matrix [F(λ)] are composed of such parameters as those denoted
in (1.25) and { } is a column containing A j , B j , C j , and D j . These equations are linear
and homogeneous. A non-trivial solution is obtained by setting the determinant of the
matrix F(λ) equal to zero, which yields a transcendental equation whose smallest root
is the critical buckling load λ. Once we evaluate the buckling load λ, Equation (1.38)
is used to determine, to an arbitrary constant multiple, the coefficients A j , B j , C j , and
D j , which can then be substituted back into Equation (1.9) to obtain the buckling mode
shape of the entire plate. Note that, in the special case of plates with equally spaced
stiffeners, the analytical finite difference calculus discussed by Wah and Calcote (1970)
can be used. In this section, however, since we will concentrate on the two- or three-span
plates with stiffeners that are not necessarily uniformly spaced, the use of the previously
mentioned method is not viable.
To investigate the variation of the buckling mode of the stiffened plate due to a
small structural irregularity, we study the simplest case where there is a single stiffener
that is slightly misplaced from the mid-span (Figure 1.2).
Let us consider a square plate. Intuitively, we know that, to produce the highest
reinforcement on the plate, the single stiffener should be placed equidistantly from
the plate edge parallel to it. We will therefore study the effect of misplacement from
such an idealized situation. Hence, we use the following non-dimensional notations for
Figure 1.2 Notations and positive directions.
specifying the positions of the stiffeners:

1 1 d
r1 = + δ, r2 = − δ, δ= (1.39)
2 2 a
where d denotes the misplacement of the stiffener from the middle, and δ is its non-
dimensional counterpart. Positiveness of d (or δ) indicates that the stiffener is shifted
to the right of its designed position; when d (or δ) is negative, the stiffener is located to
the left of its designed position. F(λ) in Equation (1.38) is now an 8 × 8 matrix with
elements as follows:
Case A:
F1,1 = 1, F1,3 = 1, F2,1 = β̄ 21 , F2,3 = β̄ 22 , F3,5 = β̄ 21 cos (β̄ 1r2 π )
F3,6 = β̄ 1 sin(β¯1r2 π ),
2
F3,7 = β̄ 22 cos(β¯2r2 π ), F3,8 = β̄ 22 sin(β¯2r2 π )
F4,5 = cos(β¯1r2 π ), F4,6 = sin(β1r2 π),
¯ F4,7 = cos(β¯2r2 π )
F4,8 = sin(β¯2r2 π ), F5,5 = 1, F5,7 = 1, F6,1 = cos(β̄ 1r1 π )
F6,2 = sin(β̄ 1r1 π ), F6,3 = cos(β̄ 2r1 π ), F6,4 = sin(β̄ 2r1 π )
F7,1 = −β̄ 1 sin(β̄ 1r1 π ), F7,2 = β̄ 1 cos(β̄ 1r1 π ), F7,3 = β̄ 2 cos(β̄ 2r1 π )
F7,4 = β̄ 2 cos(β̄ 2r1 π ), F7,6 = −β̄ 1 , F7,8 = −β̄ 2
F8,1 = −β̄ 21 cos(β̄ 1r1 π ), F8,2 = −β̄ 1 sin(β̄ 2r1 π ),
2
F8,3 = β̄ 22 cos(β̄ 2r1 π )
F8,4 = −β̄ 22 sin(β̄ 2r1 π ), F8,5 = −β̄ 1 ,
2
F8,6 = τ1 π β̄ 1
F8,7 = β̄ 22 , F8,8 = τ1 π β̄ 1 (1.40)
The remaining elements are zero.
Case B:
F1,1 = 1, F1,3 = 1, F2,1 = β̄ 21 , F2,3 = β̄ 22 , F3,5 = β̄ 21 cos(β̄ 1r2 π )
F3,6 = β̄ 1 sin(β¯1r2 π ),
2
F3,7 = β̄ 22 cos(β¯2r2 π ), F3,8 = β̄ 22 sin(β¯2r2 π )
F4,5 = cos(β¯1r2 π ), F4,6 = sin(β̄ 1r2 π ), F4,7 = cos(β̄ 2r2 π )

F4,8 = sin(β¯2r2 π ), F5,1 = cos(β̄ 1r1 π ), F5,2 = sin(β̄ 1r1 π )
F5,3 = cos(β̄ 2r1 π ), F5,4 = sin(β̄ 12r1 π ), F5,5 = −1, F5,7 = −1
F6,1 = −β̄ 1 cos(β̄ 1r1 π ), F6,2 = β̄ 1 sin(β̄ 1r1 π ), F6,3 = −β̄ 2 cos(β̄ 2r1 π )
F6,4 = β̄ 2 sin(β̄ 2r1 π ), F6,6 = −β̄ 1 , F6,8 = −β̄ 2
F7,1 = −β̄ 21 cos(β̄ 1r1 π ), F7,2 = −β̄ 1 sin(β̄ 1r1 π ),
2
F7,3 = −β̄ 22 cos(β̄ 2r1 π )
F7,4 = −β̄ 22 sin(β̄ 2r1 π ), F7,5 = −β̄ 21 , F7,6 = τ1 π β̄ 1 , F7,7 = β̄ 22
F7,8 = τ1 π β̄ 2 , F8,1 = β̄ 31 sin(β̄ 1r1 π ), F8,2 = −β̄ 31 cos(β̄ 1r1 π )
F8,3 = β̄ 32 sin(β̄ 2r1 π ), F8,4 = −β̄ 32 cos(β̄ 2r1 π ), F8,5 = −ω1 π
F8,6 = β̄ 1 ,
3
F8,7 = −ωπ, F8,8 = β̄ 2 3
(1.41)
The rest of the elements equal zero.
The column { } is now defined as
{ } = {A1 , B1 , C1 , D1 , A2 , B2 , C2 , D2 }T (1.42)
Setting the determinant of [F(λ)] to zero and using the quasi-Newton root-search-
ing method, one can find the smallest root of det[F(λ)] = 0, which corresponds to
the buckling load of the structure. Then, substituting λ back into Equation (1.38) and
taking any seven equations out of the eight equations (1.38), we can solve for { }. The
buckling mode reads

πy
w1 (x1 ) = [A1 cos(β1 x1 ) + B1 sin(β1 x1 ) + C1 cos(β2 x1 ) + D1 sin(β2 x1 )] sin ,
b
a
0 ≤ x1 ≤ − d
2

πy
b
a
0 ≤ x2 ≤ + d
2
(1.43)
Consider now a three-span continuous plate with stiffeners. The plate is all-round
simply supported and subjected to the uni-axial uniform compression in the direc-
tion perpendicular to the stiffeners (Figure 1.3). For this specific problem, a set of 12
algebraic equations can be established in the form of Equation (1.38).
Suppose that the two stiffeners are located at distances ξ1 and ξ2 from the left edge,
respectively. Following the same procedure as discussed earlier, the buckling mode for
the three spans are expressed as

πy
b
0 ≤ x1 ≤ ξ11

πy
b
0 ≤ x2 ≤ ξ2 − ξ1
Figure 1.3 Uni-axially compressed rectangular plate stiffened with a single misplaced rib.

πy
w3 (x) = [A3 cos(β1 x3 ) + B3 sin(β2 x3 ) + C3 cos(β2 x3 ) + D3 sin(β2 x3 )] sin ,
b
0 ≤ x 3 ≤ a − ξ2
(1.44)
We are interested in the variation of the buckling load and the buckling mode
with the small misplacement of the stiffeners. In addition, the optimal position of
the stiffeners, which yields the highest buckling strength, also appears to be of in-
terest. Numerical calculations are performed for both the single rib-stiffened plate
and the stiffened three-span continuous plate. Structures with different parameters for
the torsional and flexural rigidities are also investigated. For the plate with a single
stiffener, attachment of the stiffener to the middle location between the two paral-
lel edges of the plate provides the structure with the most favorable load-carrying
capacity. This conclusion holds true for Case A and, as will be seen later, also for
Case B. It is found that for Case A, where there is a support that prevents the verti-
cal displacement of the plate, the magnitude of the non-dimensional torsional rigid-
ity τ has only a moderate effect on the buckling load when τ is larger than 10
(Figure 1.4). Deviation of the stiffener from the mid-span position reduces the buckling
strength and, more importantly, changes the buckling mode from an overall pattern to
the local pattern in the plate segment with longer span. The more misplaced the stiffener
from the mid-span, the greater the reduction in buckling load is, and the more localized
the buckling mode becomes. That implies that the deflection of the plate on one side of
the stiffener is much greater than that on the other side. For example, for Case A with
a torsional rigidity of τ = 20, the ratio of the maximum deflection in the left segment
to that in the right segment is about 4.5 when δ = 0.01. If a bigger misplacement is
involved, say δ = 0.02, then the ratio of the maximum deflections in the two segments
increases to 7. However, for the stiffener with non-dimensional torsional rigidity τ < 5,
small misplacement does not significantly affect the buckling load; for instance, when
τ = 2, a deviation of magnitude δ = 0.05 produces only 4% reduction in buckling load.
Figure 1.4 Uni-axially compressed three-span stiffened plate.
Figures 1.5 and 1.6 show the buckling mode shape of the plate in Case A for different
values of τ . With a stiffener of τ larger than 30, the shorter segment of the plate is almost
undeflected as buckling mode is localized in the longer segment. For Case B, the flexural
rigidity of the stiffener plays a more important role in the buckling strength than the
torsional rigidity, although the influence of torsional rigidity is still remarkable on the
buckling mode shape. For example, it can be seen from Figure 1.7 that only stiffener
with flexural rigidity ω ≥ 5 has noticeable strengthening effect, and that when ω falls
below two, the position of the stiffener becomes almost irrelevant for the magnitude of
the buckling load. When a plate is reinforced with a rib of moderate flexural stiffness,
Figure 1.5 Loci of buckling loads for a plate stiffened by a single rib with different
values of τ (Case A).
Figure 1.6 Buckling mode localization for a plate stiffened with a single rib of
τ = 20.0 (Case A).
the longer segment of the plate is severely deflected at the onset of buckling while
the short segment experiences only a scant deformation. So the buckling is still fairly
localized, as can be seen from the mode shapes depicted in Figures 1.7, 1.8, and 1.9.
It is interesting to note that the reduction in the overall strength of the plate by mis-
positioning a stronger stiffener can be greater. For instance, when τ = 30 and ω = 20,
a 5% deviation from the mid-point produces 13% decrease in the buckling strength.
Thus, we can see that a unilateral increase in the stiffener’s strength may make the whole
structure highly sensitive to the misplacement (which can be regarded as a special kind
of initial imperfection) in the sense that a small misplacement of the stiffener or interior
Figure 1.7 Buckling mode shape for a plate stiffened by a single rib with misplace-
ment δ = 0.02 (Case A).
Figure 1.8 Loci of buckling loads for a plate stiffened by a single rib with different values
of ω (τ = 10, Case B).
support can lower the buckling load of the plate, and more importantly, localize the
buckling mode shape. Figure 1.10 shows the buckling mode for the plate stiffened by
a single rib with misplacement δ = 0.01 (Case B).
As compared with the single rib-stiffened plate, the three-span plate is even more
sensitive to the misplacement of the stiffener. For example, if one stiffener is precisely
Figure 1.9 Buckling mode localization for a plate stiffened by a single rib of τ = 20.0
and ω = 10.0 (Case B).
Figure 1.10 Buckling mode shape for a plate stiffened by a single rib with misplacement
δ = 0.01 (Case B).
fixed at ξ1 = a/3, and the other stiffener is supposed to be located at ξ2 = 2a/3 but
somehow was misplaced from this position by, say, δ2 = −0.02 (the negative sign rep-
resents the misplacement is in the negative x direction), the buckling load is decreased
by 9.5% from its counterpart without misplacement. Interestingly enough, some pat-
terns of the misplacement are detrimental, while others can be helpful. For instance,
suppose the stiffeners are designed to be located at ξ1 = a/3 and ξ2 = 2a/3, re-
spectively. The combination of the misplacement by the magnitude δ1 = δ2 = 0.02
Figure 1.11 Effect of misplacement on the buckling load of a stiffened, three-

span plate (τ1 = τ2 = 20.0).
Figure 1.12 Buckling mode shape for a stiffened, three-span plate without
stiffener misplacement (τ1 = τ2 = 20.0, δ1 = δ2 = 0.0).
(meaning the misplacements are both in the positive x direction) from ξ1 = a/3 and
ξ2 = 2a/3, respectively, cuts down the buckling load by 9.6%. However, misplace-
ment of δ1 = −δ2 = −0.02 (meaning the left stiffener has been moved slightly to
the left and the right stiffener to the right) boosts the buckling strength by 11%
(Figure 1.11). As for the buckling mode, in the majority of the situations, the buckling
patter is still severely localized (Figure 1.12). This is shown by Figure 1.13 where a
small misplacement of δ2 = −0.01 of the right stiffener triggers the onset of the local
Figure 1.13 Buckling mode shape for a stiffened, three-span plate with slight misplace-
ment (τ1 = τ2 = 20.0, δ1 = 0.0, δ2 = 0.01).
buckling in the third span of the plate at a lower load than its counterpart without the
misplacement. When this happens, the other two spans hardly deflect at all.
For the three-span continuous plate, numerical results show that, as far as the
buckling load is concerned, equally spaced stiffeners such that the three spans of the
plate have the same length are not most beneficial. Numerical analysis shows the op-
timal stiffener layout for stiffeners with torsional rigidity τ = 20 is ξ1 = 0.329a and
ξ2 = 0.671a. The corresponding buckling load is 19.6% above the buckling load with
two identical stiffeners positioned at ξ1 = a/3 and ξ2 = 2a/3. This result can also be
interpreted as follows: for misplacement δ1 = 1/3 − 0.329 ≈ 0.005 and δ2 = 2/3 −
0.671 ≈ −0.004, the buckling load is decreased by 16% ((1. − 1/1.19) × 100%). This
again demonstrates the high sensitivity of optimally designed structures to small im-
perfections. This phenomenon was discussed by Budiansky and Hutchinson (1979) as
well as Zyczkowski and Gajewski (1983) and some other investigators. Moreover, it is
found here that the optimal pattern of the stiffener layout is almost independent of the
specific value of τ , as long as the two stiffeners are identical. For example, even when
the torsional rigidity τ is decreased to 5.0, the most favorable positions of two stiffeners
are hardly changed, as they are now ξ1 = 0.329a and ξ2 = 0.671a. Figure 1.14 depicts
the buckling mode for such a situation, from which we can observe that with the optimal
stiffener layout the buckling mode is a global one (that is, all parts of the place deflect
to a comparable degree, and the potential capability of the structure is fully tapped).
As shown, the buckling mode localization phenomenon resulting from small mis-
placement in the stiffened or continuous plates should not be overlooked, especially
in those applications where the mode shape is a significant concern. Because of the
imprecision in the fabrication, the misplacement is always present and can affect the
buckling characteristics of the structure to a large extent. When the structure is designed
in terms of the optimal stiffener layout, misplacement may reduce the buckling load
Figure 1.14 Buckling mode shape for a stiffened, three-span plate with optimal stiffener
placement (τ1 = ρ2 = 20.0, ξ1 = 0.329a, ξ2 = 0.671a).
1.2 LOCALIZATION IN A MULTI-SPAN PERIODIC COLUMN WITH A DISORDER IN A SINGLE SPAN 17
and may cause the buckling mode to become highly localized in a manner that one
segment of the structure deflects appreciably while the capability of the other parts of
the structure has not been brought into full play.
1.2 Localization in a Multi-Span Periodic Column with a Disorder

in a Single Span
In Section 1.1, we considered two- and three-span plates. We are also interested in
studying the buckling mode localization in multi-span plates. Generally, such an anal-
ysis, though numerically cumbersome, is quite straightforward. In a multi-span plate,
one should write down expressions for the buckling mode for each span, as we did in the
case of the three-span plate [for example, Equation (1.44)]. Satisfaction of the bound-
ary and continuity conditions leads to an eigenvalue problem; the lowest eigenvalue
represents the buckling load, and its corresponding eigenvector depicts the buckling
mode. Extensive numerical effort is usually required to investigate the localization
phenomenon. This is especially true if the number of spans is large. One might be,
however, interested in the situation where numerical analysis is less demanding. This is
where the case of the so-called perfect misplacement comes in. Perfect misplacement
refers to a set of misplacements appearing in an extremely specialized manner. Indeed,
Koiter’s classical work (1963) where he studied the effect of initial imperfections on
the buckling of cylindrical shells was of this kind: he studied the effect of specialized
initial imperfections varying sinusoidally (see p. 107). In this and the following sec-
tion, we address the issue of a multi-span structure, where ideally stiffeners are spaced
periodically but where, due to a production process, the periodicity breaks down in one
single span of the multi-span structure. This case may, at first glance, appear rather
artificial, but we can envision that this idealized situation arises in the following way.
Suppose that the large periodic structure must be assembled; however, from the point
of view of transporting the structure, it must be assembled from several sub-structures.
Each sub-structure is a periodic structure. We can visualize that the exact periodicity
was destroyed in the assembling process. Thus, we may end up with a structure where
misplacement in one single span destroys the overall periodicity of the entire construc-
tion. Because we will use mathematical methods pertinent to ideal periodic structures
(that is, without any misplacement) it appears instructive to start our discussion with
an extremely brief overview of the analysis of periodic structures.
Large-scale periodic structures are often encountered in engineering practice. They
appear in different forms, such as beams on equidistant supports, gridwork structures
with equally spaced and geometrically similar stiffening components, and plates and
shells with uniformly distributed stiffeners. Periodic structures have drawn substan-
tial interest among researchers, and many mathematical methods have been developed,
starting with the pioneering work by Brillouin (1953). The reader may also consult
several studies (for example, Krein, 1933; Miles, 1956; Lin and McDaniel, 1969;
Sen-Gupta, 1970; Mead, 1971; Abramovich and Elishakoff, 1987; Zhu, Elishakoff, and
Lin, 1994; and Elishakoff, Lin, and Zhu, 1994). To analyze perfectly periodic struc-
tures, the method of the analytical finite difference calculus (Bleich, 1952; Wah and
Calcote, 1970) appears to be very instrumental. For the treatment of periodic structures,
this method has decisive advantage over the conventional matrix methods used in the
structural analysis. The method usually leads to a determinative matrix that can be
orders of magnitude smaller than those necessary in either the force or displacement
method or the conventional finite element method. With the size of the matrix reduced,
the numerical accuracy is improved, and the computational effort is cut down dramat-
ically. However, the applicability of this method is confined to only those structures
that are uniform in the spacing and stiffness characteristics of the constituent elements.
When the periodicity is disturbed, this method can no longer be applied. The deviation
from complete periodicity is commonly known as disorder or irregularity. Disorders
may arise from various imprecisions in the fabrication process or from geometric and
material variations in the different parts of the structures. From the standpoint of struc-
tural analysis, if the constituent units of the structure are different from one to another,
we need to analyze each of them separately, with attendant satisfaction of continuity
conditions from one unit to another, as was done in Section 1.1 for the two- and three-
span plates. This procedure usually results in matrices of high order if the structure
is composed of a large number of units. Because inversion and other operations on
such matrices may be involved in the calculations, numerical errors are unavoidable.
Therefore, it is often desirable to reduce the order of the matrices as far as possible. For-
tunately, many large-scale engineering structures are essentially periodic, and disorders
often take place in some localized areas of the structure.
Here, we discuss the general N -span beam with torsional springs at supports, which
is structurally periodic except that one of the spans of the beam contains a disorder. By
combining the finite difference calculus with the conventional displacement method,
we present the exact solution for the buckling of a large-scale, multi-span periodic
beam having disorder in an arbitrary single span of the beam. Section 1.3 will discuss
the buckling of multi-span plates with a single disorder. It is shown that even a single
disorder could be responsible for the highly localized pattern of buckling modes.
The governing differential equation for the typical ith span of axially compressed,
continuous beam with uniform cross section reads:

d 2w xi xi
EI 2 + Pw = Mi−1 R
− 1 − MiL (1.45)
d xi ai ai
or
R
d 2w Mi−1 xi MiL xi
+ k 2
w = − 1 − (1.46)
d xi2 EI ai EI ai
√
where k = P/EI, P = the axial load on the beam, E = the Young’s modulus, and
I = the moment of inertia of the cross-section of the beam; w = the deflection of the
beam and ai = the length of the ith span (Figure 1.14); Mi−1 R
, MiL are the bending
moments at two supports of that span, respectively. The superscript R(L) indicates that
span of the beam is to the right (left) of the support in question.
The general solution to Equation (1.46) is
R
Mi−1 xi M L xi
w = Ci sin(kxi ) + Di cos(kxi ) + −1 − i (1.47)
P ai P ai
where Ci and Di are arbitrary constants that are determined by the use of boundary
conditions. Here we discuss the case of transversely rigid supports. Thus, the boundary
conditions at the supports are
(1) w = 0 at xi = 0; (2) w=0 at xi = ai (1.48)
from which Ci and Di can be evaluated as

R R
Mi−1 cos(kai ) MiL Mi−1
Ci = − + , Di = (1.49)
P sin(kai ) P sin(kai ) P
Substitution into Equation (1.47) results in
R
Mi−1 cos(kai ) xi MiL sin(kxi ) xi
w= − sin(kxi ) + cos(kxi ) + − 1 + − +
P sin(kai ) ai P sin(kai ) ai
(1.50)
from which
R
Mi−1 1 sin(kai ) sin(kxi ) + cos(kai ) cos(kxi )
w = −k
P ai sin(kai )

M L 1 k cos(kxi )
− i − (1.51)
P ai sin(kai )
In the following, we will use the analytical finite difference calculus (see, for
example, Wah and Calcote, 1970). We will use the angles of rotation at supports as
principal variables, since the deflections at all the supports are zero. The angles of
rotation at supports i − 1 and i are obtained from Equation (1.51) by setting xi equal
to 0 and ai , respectively,
R
Mi−1 kai cos(kai ) MiL kai
θi−1 = 2 1− − 2 1− (1.52)
k ai EI sin(kai ) k ai EI sin(kai )
R
Mi−1 kai MiL kai cos(kai )
θi = 2 1− − 2 1− (1.53)
k ai EI sin(kai ) k ai EI sin(kai )
We may also express Mi−1 R
and MiL , the bending moments at supports i − 1 and i,
respectively, in terms of the rotational angles θi and θi+1 as follows:
2EI
R
Mi−1 = [2c1 θi−1 + c2 θi ]
ai
(1.54)
2EI
MiL = − [2c1 θi + c2 θi−1 ]
ai
where
kai [sin(kai ) − kai cos(kai )]
c1 =
4[2 − 2 cos(kai ) − kai sin(kai )]
(1.55)
kai [kai − sin(kai )]
c2 =
2[2 − 2 cos(kai ) − kai sin(kai )]
Note that the preceding derivation could also be performed straightforwardly by

using the stability functions (Horne and Merchant, 1965). Equilibrium at a typical ith
support requires that
MiL − MiR = J θi (1.56)

where J is the torsional modulus of the spring (Figure 1.14), θi is the rotational angle
at support i.
Using Equation (1.54), we obtain
c2 (θi+1 + θi−1 ) + 4(c1 + )θi = 0 (1.57)
where = J a/8EI.
Introducing the shifting operator E (Wah and Calcote, 1970) which is defined as
Eθi = θi+1 , Equation (1.47) can be rewritten as
[c2 (E + E −1 ) + 4(c1 + )]θi = 0 (1.58)
Equation (1.48) is a second-order finite difference equation, whose solution is

obtained by letting.
θi = exp(iφ) (1.59)
A direct substitution of Equation (1.59) into Equation (1.58) yields

2(c1 + )
cosh φ = − (1.60)
c2
Three different cases must be considered:
Case A: −2(c1 + )/c2 ≥ 1

The solution of Equation (1.60) has the following form

−1 2(c1 + )
φ1,2 = ±φ; φ = cosh − (1.61)
c2
The attendant solution for θi reads
θi = Aeφi + Be−φi (1.62)
where A and B are arbitrary constants.
Case B: −2(c1 + )/c2 ≤ −1

The solution to Equation (1.60) takes the form

−1 2(c1 + )
√
φ1,2 = ±(α + jπ ); α = cosh , j= −1 (1.63)
c2
and the solution for θi is
θi = [A cosh(iα) + B sinh(iα)] cos(iπ ) (1.64)

Figure 1.15 A multi-span beam with a single disorder in the (q + 1)th span.
Case C: −1 ≤ −2(c1 + )/c2 ≤ 1

The solution to Equation (1.60) now becomes

−1 2(c1 + )
φ1,2 = ±β; β = cos − (1.65)
c2
and, for θi , we have
θi = A cos(iβ) + B sin(iβ) (1.66)
Suppose that we have an N -span beam (Figure 1.15), which is periodic except
that its (q + 1)th span contains a span length imperfection that makes the span slightly
longer or shorter than the other spans of the structure, that is,
ai = a for i = 1, . . . , q, q + 2, . . . , N ; aq+1 = b (a = b) (1.67)
As a particular case b = a, we recover the perfectly periodic structure.

To facilitate the solution of the problem, we can treat the entire beam as being
composed of three segments. The first q-span periodic beam constitutes segment I;
the (q + 1)th span, namely, the single disordered span, constitutes segment II; and the
last (N − q − 1) spans of periodic beam represent segment III. Assume first that both
segments I and III themselves contain a large number of spans. For segments I and
III per se, the finite difference calculus is applicable due to their structural periodicity.
As to the disordered span, segment II, a separate consideration should be made, and
the conventional displacement method is used here. By following this procedure, we
construct a solution composed of three parts with each part corresponding to a specific
segment of the beam. Continuity conditions between those different segments are uti-
lized in combination with boundary conditions at the ends of the beam to establish an
eigenvalue problem.
For the first q spans of periodic beam, we perform the analytical finite difference
analysis
θr = θrI ; (r = 0, 1, . . . , q) (1.68)
where the superscript denotes the sequence number of the segment in question; θrI takes
one of the three forms represented by Equations (1.62), (1.64), and (1.66), depending
on the physical and geometrical conditions of the segment.
For the disordered span, recalling Equation (1.44), we have
2EI
MqR = 2c̄1 θ0II + c̄2 θ1II

b
(1.69)
2EI
L
Mq+1 =− 2c̄1 θ1II + c̄2 θ0II
b
or, in another form,
b 2c̄1 MqR + c̄2 Mq+1L

θ0II =
2EI 4c̄21 − c̄22
(1.70)
L
b 2c̄1 Mq+1 + c̄2 MqR
θ1II =−
2EI 4c̄21 − c̄22
where
kb[sin(kb) − kb cos(kb)]
c̄1 =
4[2 − 2 cos(kb) − kb sin(kb)]
(1.71)
kb[kb − sin(kb)]
c̄2 =
2[2 − 2 cos(kb) − kb sin(kb)]
Note that c̄1 and c̄2 are obtained from Equation (1.55) by formally replacing ai with b.
The treatment of the last N − q − 1 spans of periodic beam is similar to that of
segment I,
θs = θs−q−1
III
; (s = q + 1, q + 2, . . . , N ) (1.72)
where θs−q−1
III
again adopts one of the three forms denoted by Equations (1.62), (1.64),
and (1.66).
Consider now a beam simply supported at its two ends (other boundary conditions
can be treated in a similar manner). Then the boundary condition at the left end of the
beam can be represented as
−M0R = J θ0I or (2c1 + 4) θ0I + c2 θ1I = 0 (1.73)
while the boundary condition at the right end of the beam reads
M NL = J θ NIII−q−1 or (2c1 + 4) θ NIII−q−1 + c2 θ NIII−q−2 = 0 (1.74)
Conditions of continuity between the periodic spans and the disordered span of the
beam, namely, between segment I and segment II, are
2
MqL − MqR = J θq or (2c1 + 4) θqI + c2 θq−1
I
+ MR = 0
2EI q
(1.75)
b a a
θqI = θ0II or θqI − 2 2 c̄ 1 M R
+ c̄ 2 M L
=0
a 4c̄ 1 − c̄ 22 2EI q 2EI q+1
Analogously, the continuity conditions between the second and the third segments
are
a
L
Mq+1 − Mq+1
R
= J θ0III or (2c1 + 4) θ0III + c2 θ1III − ML = 0
2EI q+1

b a a
θ1II = θ0III or θ0 + 2
III
c̄2 M + 2c̄1
R
M L
=0
a 4c̄ 1 − c̄ 22 2EI q 2EI q+1
(1.76)
Equations (1.75) and (1.76) should be formulated in terms of the three different
cases because, in each case, the resulting expressions are different.
For Case A, the rotation angles in the first and third segments can be expressed as
θrI = A1 eφr + B1 e−φr (r = 0, 1, . . . , q)
φ(s−q−1)
(1.77)
θs−q−1
III
= A2 e + B2 e−φ(s−q−1) (s = q + 1, . . . , N )
Substituting the preceding expressions in boundary conditions (1.73) and (1.74)
and continuity conditions (1.75) and (1.76), we obtain six homogeneous algebraic
equations:
(2c1 + 4 + c2 eφ )A1 + (2c1 + 4 + c2 e−φ )B1 = 0 (1.78)

(2c1 + 4)eφq + c2 eφ(q−1) A1 + (2c1 + 4)e−φq + c2 e−φ(q−1) B1 + M̄ qR = 0

(1.79)
φq −φq 2c̄1 b c̄2 b
e A1 + e B1 − 2 M̄ q − 2
R L
M̄ q+1 =0 (1.80)
4c̄ 1 − c̄ 2 a
2
4c1 − c̄ 2 a
2
(2c1 + 4 + c2 eφ )A2 + (2c1 + 4 + c2 e−φ )B2 − M̄ q+1 L

=0 (1.81)

c̄2 b 2c̄1 b
A2 + B2 + 2 M̄ qR + 2 L
M̄ q+1 =0 (1.82)
4c̄ 1 + c̄ 2 a
2
4c1 + c̄ 2 a
2
(2c1 + 4)eφ(N −q−1) + c2 eφ(N −q−2) A2

+ (2c1 + 4)e−φ(N −q−1) + c2 e−φ(N −q−1) B2 = 0 (1.83)

where
MqR a L
M̄ q+1 a
M̄ qR = ; L
M̄ q+1 = (1.84)
2EI 2EI
For Case B, the solutions for the first and third segments are as follows:
θrI = A1 cosh(αr ) cos(πr ) + B1 sinh(αr ) cos(πr ) (r = 0, 1, . . . , q)
θs−q−1
III
= A2 cosh[α(s − q − 1)] cos[π (s − q − 1)]
+ B2 sinh[α(s − q − 1)] cos[π (s − q − 1)] (s = q + 1, . . . , N )
(1.85)
and performing the substitution similar to that in Case A, we arrive at the following six
homogeneous equations:
[2c1 + 4 − c2 cosh(α)]A1 − c2 sinh(α)B1 = 0 (1.86)
{(2c1 + 4) cosh(αq) cos(πq) + c2 cosh[α(q − 1)] cos[π (q − 1)]}A1

{(2c1 + 4) sinh(αq) cos(πq) + c2 sinh[α(q − 1)] cos[π (q − 1)]}B1 + M̄ qR = 0
(1.87)
cosh(αq) cos(πq)A1 + sinh(αq) cos(πq)B1

2c̄1 b c̄2 b
− 2 M̄ q − 2
R L
M̄ q+1 =0 (1.88)
4c̄ 1 − c̄ 22 a 4c̄ 1 − c̄ 22 a
[2c1 + 4 − c2 cosh(α)]A2 − c2 sinh(α)B2 − Mq+1 L
=0 (1.89)

c̄2 b 2c̄1 b
A2 + 2 M̄ qR + 2 L
M̄ q+1 =0 (1.90)
4c̄ 1 − c̄ 2 a
2
4c̄ 1 − c̄ 2 a
2
{(2c1 + 4) cosh[α(N − q − 1)] cos[π (N − q − 1)]

+ c2 cosh[α(N − q − 2)] cos[π (N − q − 2)]}A2
(1.91)
{(2c1 + 4) sinh[α(N − q − 1)] cos[π (N − q − 1)]
+ c2 sinh[α(N − q − 2)] cos[π (N − q − 2)]}B2 = 0
For Case C, the solutions are in the following form:
θrI = A2 cos(βr ) + B1 sin(βr ) (r = 0, 1, . . . , q)
θs−q−1
III
= A2 cos[β(s − q − 1)] + B2 sin[β(s − q − 1)] (s = q + 1, . . . , N )
(1.92)
and the corresponding equations are
[2c1 + 4 + c2 cos(β)]A1 + c2 sin(β)B1 = 0 (1.93)

{(2c1 + 4) cos(βq) + c2 cos[β(q − 1)]}A1 + {(2c1 + 4) sin(βq)
+ c2 sin[β(q − 1)]}B1 + M̄ qR = 0 (1.94)

2c̄1 b c̄2 b
cos(βq)A1 + sin(βq)B1 − 2 M̄ q − 2
R L
M̄ q+1 =0 (1.95)
4c̄ 1 − c̄ 2 a
2
4c̄ 1 − c̄ 2 a
2
[2c1 + 4 + c2 cos(β)]A2 + c2 sin(β)B2 − M̄ q+1 L

=0 (1.96)

c̄2 b 2c̄1 b
A2 + 2 M̄ qR + 2 L
M̄ q+1 =0 (1.97)
4c̄ 1 − c̄ 2 a
2
4c̄ 1 − c̄ 2 a
2
{(2c1 + 4) cos[β(N − q − 1)] + c2 cos[β(N − q − 2)]}A2

+ {(2c1 + 4) sin[β(N − q − 1)] + c2 sin[β(N − q − 2)]}B2 = 0 (1.98)
Thus, for each case, we have six homogeneous algebraic equations, which can be
expressed in a matrix form as follows:
[F(K )]6×6 {δ}6×1 = 0, K = ka (1.99)
where [F(K )] is the matrix, and {δ}T = {A1 , B1 , M̄ qR , M̄ q+1

L
, A2 , B2 }.
Non-triviality of {δ} requires that the determinant of the coefficient matrix vanish,
det[F(K )] = 0 (1.100)
which constitutes a transcendental equation for the non-dimensional buckling load

parameter K . After the buckling load parameter K is determined, we can use Equa-
tion (1.50) to calculate, span by span, the buckling mode shapes for the entire structure.
It is worth mentioning that, if the disorder occurs in the first or last span of the
beam, the whole beam can be partitioned into two segments: one is the disordered span
and the other is the N − 1 spans of periodic beam. If this happens, only four equations
are needed to characterize the problem so that instead of having a 6 × 6 matrix for
[F(K )], we will have a 4 × 4 determinant matrix.
In the following numerical examples, the non-dimensional spring constant is
fixed at 0.3 because this particular case for perfectly periodic beam was considered by
Wah and Calcote (1970), and a comparison can be made with their results.
As a first example, we discuss a simply supported, 100-span continuous beam. The
disorder arises from a span length imperfection characterized by b/a = 1.1, where a
and b are the lengths of periodic spans and the disordered span, respectively. Because
b/a > 1, the disordered span is longer than other spans of the structure. The disorder
may appear in any span of the beam. Figure 1.16 depicts the results of the buckling load
parameter K for the beams with or without torsional springs. The most critical situation
for the beam without torsional springs occurs when the disorder occurs at either the first
or the last span, for which the buckling load parameter K equals 3.05. (If there is no
disorder, the buckling load parameter is π. Thus, the effect on the buckling load itself is
somewhat detrimental with buckling load reduced by 3%.) If the disorder appears in one
Figure 1.16 Variation of buckling load with the location of the disorder (i , sequence number
of the span where the disorder occurs).
of the spans close to the center of the beam, the buckling load increases. However, for
beams with torsional springs, numerical results display a quite different picture. For this
case, the occurrence of disorder near the boundaries may be more advantageous. The
buckling load is maximum for this configuration, it decreases as the disorder moves
away from the boundaries. Nevertheless, for both cases, the buckling load remains
almost unchanged with the location of disorder, once the disorder is 10 spans away
from boundaries. This mean that the effect of the boundary dies out if the disorder is
sufficiently far from the boundary. The location of the disorder has almost no effect
on the buckling load if the spring modulus = 0.15. If we refer to the buckling load
in the absence of disorder as the classical buckling load, then the classical buckling
load parameter K equals π for the case without torsional springs and has a value of
3.760 for the beam with torsional springs of = 0.03 (Wah and Calcote, 1970). With
the disorder present, numerical results show that the buckling load parameter K is
below the corresponding classical value. For instance, K is less than 3.12 for the beam
without torsional springs and is no more than 3.72 for the beam with torsional springs of
= 0.3. Thus, we can see that such a disorder, namely, the span length imperfection,
may have a degrading effect on the load-carrying capacity of the structure.
The second example is mainly devoted to the discussion of the buckling mode
shapes for disordered periodic beam. Figures 1.17 and 1.18 depict the buckling mode
shapes for an 11-span beam with the disorder appearing at various locations of the
beam. Again, the disorder is introduced by a span length imperfection specified by
b/a = 1.1. A significant phenomenon is that the buckling mode shape exhibits a strong
localization around the disordered span, when the torsional springs are used as sup-
ports. The larger the moduli of the torsional springs, the more localized the mode shape
becomes. Thus, the torsional spring weakens the coupling between different spans of
the structure. This observation is consistent with that found by Pierre and Plaut (1984)
for the two-span beam.
In passing, it is worthwhile to point out that, although the underlying treatment
makes it possible to obtain an exact solution to the buckling problem of disordered
periodic beams with any number of spans by dealing with a determinative matrix of
low order (the matrix is 6 × 6 if only one disorder occurs in the span neither the first nor
the last), some numerical problem can occur when N , the number of spans, is a large
number, say N ≥ 50. This is because, in our calculations, we have to evaluate terms
eφ(N −q−1) and cosh[α(N − q − 1)], which can be so large when q is small, that they may
exceed the upper limit of some digital computers. To avoid this numerical difficulty, we
may divide the corresponding equation by a relevant large term, for example, eφ(N −q−1)
or cosh[α(N − q − 1)], and manipulate the resulting equation by making an asymptotic
approximation
sinh[α(N − q − 1)]
→1 for q N (1.101)
cosh[α(N − q − 1)]
It follows that only Equations (1.101) and (1.91) need to be modified and they adopt,
after the approximations, the following forms, respectively,
[(2c1 + 4) + c2 e−φ ]A2 = 0 (1.102)

Figure 1.17 Buckling mode shapes for a disordered 11-span beam without torsional spring
(i , support sequence number; w, deflection). (a) Disorder occurs in the third span; (b) disorder
occurs in the mid-span (sixth span); (c) disorder occurs in the last span.
and
{(2c1 + 4) cos[π (N − q − 1)] + c2 e−α cos[π (N − q − 2)]}A2

+ {(2c1 + 4) cos[π (N − q − 1)] + c2 e−α cos[π (N − q − 2)]}B2 = 0
(1.103)
Figure 1.19 depicts the buckling modes of a 100-span beam and a 400-span beam with
torsional springs of = 0.3; both beams contain a disorder in the 40th span. From
Figure 1.18 Buckling mode shapes for a disordered 11-span beam with torsional spring of
ν = 10 (i , support sequence number; w, deflection). (a) Disorder occurs in the third span;
(b) disorder occurs in the mid-span (sixth span); (c) disorder occurs in the last span.
Figure 1.19, it is clear that the deflection of the beam at buckling dies out quickly as
the distance from the disordered span increases. The envelope of the buckling mode is
depicted in Figure 1.19. If we take a logarithm of the function, we obtain Figure 1.20,
which displays a nearly straight line. Thus, we establish that the deflection at buckling
decays exponentially. The exponential decay constant (Pierre, 1990) is usually referred
to as the Lyapunov exponent (Arnold, Crauel, and Eckmann, 1991; Ariaratnam and
Xie, 1995) in the literature [note that its counterpart in vibration problems is commonly
known as the logarithmic decrement (Thomson, 1981)] and in our case equals −0.260.
Figure 1.19 Envelope of the buckling mode shape for a 100-span beam.
In the preceding examples, we dealt with the disorder of “detrimental” character;

that is, the disordered span is longer than the periodic spans, and such a disorder leads
to a reduction in the buckling load of the structure. As a third example, we consider
also another possibility with attendant opposite effect; that is, the disordered span
is shorter than the other spans (b/a < 1). In this case, the disorder turns out to be
“beneficial” because it results in slightly higher buckling loads. For example, for a
perfectly periodic 11-span beam with torsional springs of non-dimensional modulus
= 0.3, the buckling load parameter K is, as mentioned earlier, 3.76. For the same
structure but with disorder, K varies in the range from 3.79 to 3.82, depending on the
location of the disorder. Thus, we can see that the “beneficial” effect of such a disorder
on the buckling load is very small – the increase in buckling load is less than 2%.
However, the impact of disorder on the buckling mode shape is noteworthy. Figure 1.21
portrays the buckling mode shapes of 11-span beams containing such a disorder, from
which it can be seen that the change in buckling mode is significant, especially when the
Figure 1.20 Logarithmic plot of the envelope function.

Figure 1.21 Buckling mode shape for a disordered 11-span beam (disordered span
is shorter than the periodic span). (a) Disorder occurs in the third span; (b) disorder
occurs in the mid-span (sixth span).
disorder occurs near the boundaries. Interestingly enough, the localization phenomenon
has not been observed in this “beneficial” case. Finally, it is should be noted that the
present study can be generalized to other more complicated problems, with more than
a single disorder.
1.3 Localization Phenomenon of Buckling Mode in Stiffened Multi-Span

Elastic Plates
In the previous two sections, we discussed the localization phenomena of the buckling
mode in the single-span plate and the multi-span beam due to misplacement either
in the stiffeners or in the internal supports. Now we will extend our investigation to
the multi-span plate (Figure 1.22). The general idea presented in Section 1.2 will be
followed here. Regarding the discussion within an arbitrary span of the plate, the results
obtained in Section 1.1 will be employed.
1.3 BUCKLING MODE IN STIFFENED MULTI-SPAN ELASTIC PLATES 31
Figure 1.22 An N -span continuous plate with a single disorder in the (q + 1)th
span.
Again, the differential equation of the deflection surface of the plate subjected to a
uniform compression P in the x direction (Figure 1.23) is
4
∂ w ∂ 4w ∂ 4w ∂ 2w
Dp + 2 + + P =0 (1.104)
∂x4 ∂ x 2∂ y2 ∂ y4 ∂x2
where w is the transverse displacement, downward positive; D p is the flexural rigidity
Figure 1.23 A piecewise periodic continuous plate.

of the plate; From Section 1.1, we know that, for rectangular plates whose boundaries
are parallel to the x-axis and are simply supported, the solution of Equation (1.104) can
be represented in the following form:

πy
w(x) = W (x) sin
b (1.105)
W (x) = A cos(β1 x) + B sin(β1 x) + C cos(β2 x) + D sin(β2 x)
where A, B, C, and D are constants of integration, which are to be determined by use

of continuity and boundary conditions; and parameters β1 and β2 are denoted by
1/2
π λ λ λ
β1 = −1+ −2
b 2 2 2
(1.106)
1/2
π λ λ λ Pb2
β2 = −1− −2 , λ= 2
b 2 2 2 π Dc
Here b stands for the width of the plate.

Catering to a single, arbitrary ith span, the solution can be written as

πy
wi = Wi sin
b
Wi = Ai cos(β1 xi ) + Bi cos(β1 x j ) + Ci cos(β2 x j ) + Di sin(β2 xi ), (1.107)
0 ≤ xi ≤ ai
where ai is the length of the ith span and i ranges from 1 to N for an N -span plate.
Here we consider a simplified case where there is a roller support under each
interior stiffener. The more complicated situation, namely the plate without the interior
support, is addressed in the study by Elishakoff, Li, and Starnes (1995a). In reality,
we may deviate somewhat from this condition. Instead of the presence of the interior
supports, a more common occurrence in engineering practice is the use of girders of
joists with plate structures, and sometimes the flexure rigidity of these girders can be
so large that the deflection of the girders is negligible. If this happens, the deflection
along the stiffeners can be regarded as zero.
Using boundary conditions for an arbitrary ith span
wi |xi =0 = 0
wi |xi =a j = 0

dwi πy
= θi−1 ; θi−1 = i−1 sin (1.108)
d xi xi =0 b

dwi πy
= θi ; θi = i sin
d xi xi =ai b
coefficients Ai , Bi , Ci , and Di can be determined with the aid of Mathematica (Wolfram,

1991)
1
Ai = {[β2 sin(β1 ai ) − β1 sin(β2 ai )]i
Si
+ [−β2 cos(β2 ai ) sin(β1 ai ) + β1 cos(β1 ai ) sin(β2 ai )]i−1 }
1
Bi = {β2 [−cos(β1 ai ) + cos(β2 ai )]i
Si
+ [β2 cos(β1 ai ) cos(β2 ai ) − β2 + β1 sin(β1 ai ) sin(β2 ai )]i−1 }
1
Ci = {[−β2 sin(β1 ai ) + β1 sin(β2 ai )]i (1.109)
Si
+ [β2 cos(β2 ai ) sin(β1 ai ) − β1 cos(β1 ai ) sin(β2 ai )]i−1 }
1
Di = {β1 [cos(β1 ai ) − cos(β2 ai )]i
Si
+ [−β1 + β1 cos(β1 ai ) cos(β2 ai ) + β2 sin(β1 ai ) sin(β2 ai )]i−1 }

Si = 2β1 β2 [−1 + cos(β1 ai ) cos(β2 ai )] + β12 + β22 sin(β1 ai ) sin(β2 ai )
The expression for the bending moment is

2
∂ wi ∂ 2 wi
Mx = −D p +ν 2 (1.110)
∂ xi2 ∂y
Using Equation (1.107), a moment-slope relationship can be established as follows:
πy Dp
R
Mi−1 = Mx |xi =0 = m i−1
R
sin ; R
m i−1 = [c1 i−1 + c2 i ]
b ai
(1.111)
πy Dp
MiL = Mx |xi =ai = m iL sin ; m iL = [c1 i + c2 i−1 ]
b ai
R
where M j−1 and M jL are the bending moments at the two supports of the span, respec-
tively. The superscript R(L) indicates that span of the plate is to the right (left) of the
support in question (Figure 1.24). The coefficients c1 and c2 are defined as
ai 2
c1 = β1 − β22 [−β2 cos(β2 ai ) sin(β1 ai ) + β1 cos(β1 ai ) sin(β2 ai )]
Si
(1.112)
ai 2
c2 = β1 − β22 [β2 sin(β1 ai ) − β1 sin(β2 ai )]
Si
If a number of spans of the plate have the common length ai = a and are made of
the same material, then the analytical finite difference calculus may be applied in the
discussion of that part of the plate.
Equilibrium at a typical support r reads
∂ 2 θr
MrR − MrL = GJ (1.113)
∂ y2
where GJ is the torsional rigidity of the transverse stiffener.
Figure 1.24 Buckling mode for an 11-span periodic plate.
Substituting Equation (1.111) into Equation (1.113), we have

(2c1 + k)r + c2 (r +1 + r −1 ) = 0 (1.114)
where k = π 2 aGJ/b2 Dc .
Introducing the shifting operator E which is defined as Eθi = θi+1 , Equation (1.114)
can be rewritten as

c2 (E + E −1 ) + (2c1 + k) r = 0 (1.115)
Equation (1.115) is a second-order finite difference equation, whose solution is
obtained by letting
r = eφr (1.116)
Substitution into Equation (1.115) results in
2c1 + k
cosh(φ) = − (1.117)
2c2
Three different cases may arise and deserve separate considerations.
Case 1: −(2c1 + k)/2c2 ≥ 1

The solution of Equation (1.117) has the following form:

2c1 + k
φ1,2 = ±α; α = cosh−1 − (1.118)
2c2
The attendant solution for i reads
i = Aeαi + Be−αi (1.119)
where A and B are arbitrary constants.
Case 2: −(2c1 + k)/2c2 ≤ −1

The solution to Equation (1.117) takes the form

−1 2c1 + k
√
φ1,2 = ±(α + jπ ); α = cosh , j= −1 (1.120)
2c2
and the solution for i is

i = [A cosh(αi) + B sinh(αi)] cos(πi) (1.121)
Case 3: −1 ≤ −(2c1 + k)/2c2 ≤ 1

The solution to Equation (1.117) now becomes

−1 2c1 + k
φ1,2 = ±α; α = cos − (1.122)
2c2
and, for i , we obtain
i = A cos(αi) + B sin(αi) (1.123)
Upon introducing the following notations
f 1 (α, r ) = eαr , f 2 (α, r ) = cosh(αr ) cos(πr ), f 3 (α, r ) = cos(αr )
−αr
g1 (α, r ) = e , g2 (α, r ) = sinh(αr ) cos(πr ), g3 (α, r ) = sin(αr )
(1.124)
the three cases have a unified expression
r,i = A1 f i (α, r ) + B1 gi (α, r ), i = 1, 2, 3 (1.125)
where r,i corresponds to the three different cases when the subscript i varies from
1 to 3.
We will consider two different kinds of continuous plates. The first kind of the multi-
span plate is structurally periodic except for a single disordered span that contains an
imperfection. The second kind is a two-piecewise-periodic plate, which means that its
first qth spans and the rest of the N − q spans are periodic per se, but they do not have
the same periodicity.
Suppose that we have an N -span plate (Figure 1.25), which is periodic except that
its (q + 1)th span contains a span length imperfection that makes that span slightly
longer or shorter than the other spans of the structure, that is,
ai = a for i = 1, . . . , q, q + 2, . . . , N ; aq+1 = a ∗ (a = a ∗ ) (1.126)
As a particular case (a ∗ = a), we recover the original, perfectly periodic structure.
To facilitate the solution of the problem, we can treat the entire continuous plate
as being composed of three segments. The first q-span periodic plate constitutes seg-
ment I; the (q + 1)th span, namely, the disordered span, constitutes segment II; and the
last (N − q − 1) spans of periodic plate represent segment III. Assume first that both
segment I and III themselves contain a large number of spans. For segments I and III,
the finite difference calculus is applicable because of their structural periodicity. With
respect to the disordered span, segment II, a separate consideration should be made. By
following this procedure, we construct a solution composed of three parts with each
part corresponding to a specific segment of the plate. Continuity conditions between
those different segments are utilized in combination with boundary conditions at the
ends of the plate to establish an eigenvalue problem.
Figure 1.25 Buckling mode for a disordered 11-span plate (k = 5).
For the first q spans of periodic plate, we perform the finite difference calculus
analysis
πy
θr = θrI = rI sin ; (r = 0, 1, . . . , q) (1.127)
b
where the superscript denotes the sequence number of the segment in question; rI
takes on one of the three forms represented by Equations (1.119), (1.121), and (1.123),
depending on the physical and geometrical conditions of the segment.
For the disordered span, recalling Equation (1.111), we have
Dc II
MqR = c̄ 1 θ + c̄ 2 θ II
a∗ 0 1
(1.128)
Dc
L
Mq+1 = − ∗ c̄1 θ1II + c̄2 θ0II
a
a ∗ c̄1 MqR + c̄2 Mq+1 L

θ0II =
Dc c̄ 21 − c̄ 22
(1.129)
a ∗ c̄1 Mq+1
L
+ c̄2 MqR
θ1II =−
Dc c̄ 21 − c̄ 22
where c̄1 and c̄2 are obtained from the expressions for c1 and c2 by formally replacing
ai in Equation (1.112) with a ∗ .
The treatment of the last N − q − 1 spans of periodic beam is similar to that of
segment I,
θs = θs−q−1
III
; (s = q + 1, q + 2, . . . , N ) (1.130)
Consider now a plate simply supported at its two ends (other boundary conditions
can be treated in a similar manner). Then the boundary condition at the left end of the
plate can be represented as
∂ 2 θ0I
M0R = GJ1 or (c1 + k1 )0I + C2 1I = 0 (1.131)
∂ y2
while the boundary condition at the right end of the plate reads
∂ 2 θ NIII−q−1
−M NL = GJ1 or (c1 + k1 )III
N −q−1 + c2 N −q−2 = 0
III
(1.132)
∂ y2
where GJ1 is the torsional rigidity of the transverse stiffeners at the two boundaries. In
order to have a purely periodic structure, the stiffeners at the boundaries should have half
the stiffness of those interior stiffeners (that is, GJ1 = GJ/2 or k1 = k/2). This “half-
stiffener” concept has been used widely in the monograph (Wah and Calcote, 1970).
Conditions of continuity between the periodic spans and the disordered span of the
beam, namely, between segment I and segment II, are
∂ 2 θq a R
−MqL + MqR = GJ or (c1 + k)qI + c2 q−1
I
+ m =0
∂2 y Dc q
(1.133)
a∗ a R a L
θqI = θ0II or qI − 2 c̄ 1 m + c̄ 2 m =0
a c̄ 1 − c̄ 22 Dc q Dc q+1
Analogously, the continuity conditions between the second and the third segments
are
∂ 2 θ0III a L
−Mq+1
L
+ Mq+1
R
= GJ or (c1 + k)III
0 + c2 1 −
III
m =0
∂ y2 D q+1

a∗ a R a L
θ1II = θ0III or III
0 + 2 c̄2 m q + c̄1 m q+1 = 0
a c̄ 1 − c̄ 22 D D
(1.134)
Using the unified expression (1.125), the rotation angles in the first and third
segments can be expressed as
rI = A1 f i (α, r ) + B1 gi (α, r ) (r = 0, 1, . . . , q)
III
s−q−1 = A2 f i (α, s − q − 1) + B2 gi (α, s − q − 1) (s = q + 1, . . . , N )
(1.135)
Substituting expressions (1.135) in the boundary conditions (1.131) and (1.132)
and the continuity conditions (1.133) and (1.134), we obtain six homogeneous algebraic
equations,
[(c1 + k1 ) f i (α, 0) + c2 f i (α, 1)]A1 + [(c1 + k1 )gi (α, 0) + c2 gi (α, 1)]B1 = 0
(1.136)
[(c1 + k) f i (α, q) + c2 f i (α, q − 1)]A1
+ [(c1 + k)gi (α, q) + c2 gi (α, q − 1)]B1 + m̄ qR = 0 (1.137)
∗ ∗
c̄1 a c̄2 a
f i (α, q)A1 + gi (α, q)B1 − 2 m̄ qR − 2 L
m̄ q+1 = 0 (1.138)
c̄ 1 − c̄ 22 a c1 − c̄ 22 a
[(c1 + k) f i (α, 0) + c2 f i (α, 1)]A2

+ [(c1 + k)gi (α, 0) + c2 gi (α, 1)]B2 − m̄ q+1
L
=0 (1.139)
∗ ∗
c̄2 a c̄1 a
A2 + B2 + 2 m̄ qR + 2 L
m̄ q+1 =0 (1.140)
c̄ 1 + c̄ 2 a
2
c1 − c̄ 2 a
2
[(c1 + k1 ) f i (α, N − q − 1) + c2 f i (α, N − q − 2)]A2

+ [(c1 + k1 )gi (α, N − q − 1) + c2 gi (α, N − q − 2)]B2 = 0 (1.141)
where
m qR a L
m q+1 a
m̄ qR = ; L
m̄ q+1 = (1.142)
Dc Dc
Thus, we have six homogeneous algebraic equations, which can be expressed in a
matrix form as follows:
[F(λ)]6×6 {δ}6×1 = 0 (1.143)
where [F(λ)] is the coefficient matrix, and {δ}T = {A1 , B1 , m̄ qR , m̄ q+1
L
, A2 , B2 }.
Another kind of the continuous plate is characterized by

a for i = 1, . . . , q
ai = ∗ (a = a ∗ ) (1.144)
a for i = q + 1, . . . , N
For this problem, the first periodic segment, which consists of the first q spans of the
plate, may fall into one of the three different cases; the second periodic segment, which
is comprised of the remaining N − q spans may present itself if another three different
cases. Therefore, there might be nine separate cases altogether, which makes the search
for a solution rather complicated.
Recalling Equation (1.124), a unified solution for this piecewise periodic plate can
be written as

A1 f i (α1 , r ) + B1 gi (α1 , r ), i = 1, 2, 3; r = 0, . . . , q
r =
A2 f j (α2 , r ) + B2 g j (α2 , r ), j = 1, 2, 3; r = q + 1, . . . , N
(1.145)
Using the boundary conditions (simple supports at two ends)
I ∂ 2 θ0I L II ∂ 2 θ NII−q
(1) M0R = GJ1
; (2) − M = GJ1 (1.146)
∂ y2 N
∂ y2
and the conditions of continuity
I II
(1) θqI = θ0II ; (2) MqL = M0R (1.147)
the following homogeneous equations are established
[(k1 + c1 ) f i (α1 , 0) + c2 f i (α1 , 1)]A1 + [(k1 + c1 )gi (α1 , 0) + c2 gi (α1 , 1)]B1 = 0
(1.148)
[(k + c̄1 ) f j (α2 , N − q) + c̄2 f j (α2 , N − q − 1)]A2
+ [(k + c̄1 )g j (ᾱ2 , N − q) + c2 g j (α2 , N − q − 1)]B2 = 0 (1.149)
f i (α1 , q)A1 + gi (α1 , q − 1)B1 − f j (α2 , 0)A2 − g j (α2 , 0)B2 = 0 (1.150)
where the sub-indices i and j take on the value of 1, 2, or 3, depending upon which
particular case the segments fall into.
[c1 f i (α1 , q) + c2 f i (α1 , q − 1)]A1 + [c1 gi (α1 , q − 1) + c2 gi (α1 , q − 1)]B1

a
− ∗
c̄1 f j (α2 , 0) + c̄2 f j (α2 , 1) A2
a

a
+ ∗
c̄1 g j (α2 , 0) + c2 g j (α2 , 1) B2 = 0 (1.151)
a
Again, Equations (1.148) to (1.151) can be written in matrix form
[F(λ)]4×4 {δ}4×1 = 0 (1.152)
where [F(λ)] is the coefficient matrix, and {δ}T = {A1 , B1 , A2 , B2 }.
Both Equations (1.143) and (1.152) are homogeneous algebraic equations. As such,
non-triviality of {δ} requires that the determinant of the coefficient matrix vanish
Det[F(λ)] = 0 (1.153)
which constitutes a transcendental equation from which the non-dimensional buckling
load parameter λ can be solved in terms of other geometric and material properties of
the structure in question. After determining the buckling load parameter λ, we can use
Equation (1.107) to calculate, span by span, the buckling mode shape for the entire
structure, after the type of case has been ascertained.
In this section, we discuss the buckling load and mode shapes of the two different
types of multi-span plates described in Section 1.2.
As the first example, consider the following case:
a a∗
= 1, = 1.1, N = 11, q=5 (1.154)
b a
As is shown in the preceding data, the plate consists of 11 spans, of which the sixth
span contains a length imperfection that makes that span a bit longer than the other
spans. Numerical results show that such an imperfection has a slight degrading effect
on the buckling load. For instance, when k, the parameter characterizing the torsional
rigidity of the stiffener, equals 5, buckling load parameter λ is 5.06. Compared with its
counterpart of the periodic plate, which is λ = 5.26, the reduction rate is only 4%. The
buckling load reduction remains almost unchanged with the torsional rigidity of the
stiffeners. Even when the torsional rigidity doubles, the reduction rate only amounts
to 5%. So, the buckling load decrease induced by the presence of the imperfection is
not significant. However, the buckling modes are appreciably different for the plates
with and without the imperfection (Figures 1.26–1.28). Moreover, as k increases, the
buckling mode of the disordered plate becomes increasingly localized (Figures 1.26
and 1.27). The overall behavior of such plates is very similar to that of the continuous
beams with torsional springs discussed in Section 1.2, despite the difference of structural
dimensionality between beams and plates.
The second example is a 10-span, piecewise periodic plate whose first five spans
have the length of a and the other five spans have the length of a ∗ (assuming a ∗ > a),
Figure 1.26 Buckling mode for a disordered 11-span plate (k = 10).
and the plate is reinforced by transverse stiffeners with torsional rigidity of k = 10.
When the ratio of a ∗ to a equals unity, the plate reduces to a purely periodic plate.
Here we discuss the case where the ratio is different from unity, say a ∗ /a = 1.1. As
far as the buckling load is concerned, the difference between such a structure and the
purely periodic plate is relatively minor. For the preceding structure, the buckling load
parameter λ equals 5.35; for the corresponding, exactly periodic plate, λ is 5.77. So the
difference in the buckling load between the two is only 7%. More significant, however,
is the difference in the buckling mode. For instance, when a ∗ /a = 1.1 and k = 10, the
deflection of the structure at buckling is largely confined to the left end, whereas those
spans of the plate near the other end hardly experience any deformation (Figure 1.28).
From the examples already shown, it is well demonstrated that the torsional stiff-
ness of the stiffeners should not be ignored in the investigation, just for the sake of
Figure 1.27 Buckling mode for a piecewise periodic plate (k = 5).

Figure 1.28 Buckling mode for a piecewise periodic plate (k = 10).
simplification in analysis. As it turns out, the torsional stiffness plays quite an impor-
tant role. It not only boosts the strength of the structure but also localizes the loss of
geometric rigidity of the structure at buckling to a small area so that any damages,
should they occur, are kept to a minimum.
For the plates contemplated here, small structural irregularities do not significantly
alter the load-carrying capacity. However, the presence of such irregularities confines
the buckling pattern associated with large deflection to a limited fraction of the structure.
In this regard, the effect of such irregularities on the buckling loads can be considered
favorable. As Nayfeh and Hawwa (1994b) pointed out, by means of “deliberately”
inducing some irregularities in the system, one may confine the structural buckling to
a limited part of the system only, which can be regarded as a passive control of the
buckling process.
The method used here seems to have a wide application, especially in continuous
plates with a large number of spans. In fact, the larger the number of spans is, the more
advantageous the present method is over the other traditional methods such as the one
based on the integration of the governing differential equation because it leads to a
determinative matrix that can be orders of magnitude smaller than those necessary in
other methods. Besides, the solutions generated by the present method are analytic and
exact and, thus, can be used as benchmarks for other numerical methods.
The method of investigation presented here can be applied to discuss essentially
periodic structures where irregularities occur only in some local areas. It takes into
consideration the discreteness of the stiffeners and, in particular, their torsional rigidity.
As it turns out, the torsional rigidity of the stiffeners are important and should not be
ignored in the discussion of the buckling mode shape. Adjusting the torsional rigidity
of the stiffeners, one can achieve the goal of localizing the deflection of the structure
at buckling to a small area. For further discussion, the reader is referred to the paper by
Elishakoff, Li, and Starnes (1995).
The following consideration appears worthy of addressing: The buckling modes

themselves, although extremely important, constitute only auxiliary information. They
could be utilized for expanding the response quantities in series in terms of these modes.
Yet, studying only buckling modes does not lead to the conclusion about the response
quantities that the designer may be concerned with. Zingales and Elishakoff (2000)
addressed the effect of small misplacements in the same two-span column considered
by Pierre and Plaut (1989), but the column was subjected to transverse loading, in
addition to axial forces. A significant influence of the misplacements was detected on
the response; it included, for specific combinations of parameters and in some locations,
more than a fivefold enhancement of the response, for just 3% of the misplacement in
the middle support.
Buckling mode localization in probabilistic setting is studied in Section 4.5.
CHAPTER TWO
Deterministic Problems of Shells

with Variable Thickness
Shells were first used by the Creator of the earth and its inhabitants. The list of
natural shell-like structures is long, and the strength properties of some of them
are remarkable. Egg shells range in size from those of the smallest insects to the
large ostrich eggs, and cellular structures are the building blocks for both plants
and animals. Bamboo is basically a thin-walled cylindrical structure, as is the
root section of a bird’s feather. The latter structural element develops remarkable
load-carrying abilities.
E. E. Sechler
The degeneration of a bifurcation point into a limit point due to presence of

imperfections . . . suggests that bifurcation is an exception rather than the rule.
In spite of this the literature on elastic stability and buckling of structures deals
preponderantly with bifurcation buckling problems. Human frailty is, perhaps,
the main reason for this state of affairs.
W. T. Koiter
This chapter focuses on the buckling of cylindrical shells with small thickness varia-
tions. Two important cases of thickness variation pattern are considered. Asymptotic
formulas up to the second order of the thickness variation parameter are derived by
combining the perturbation and weighted residual methods. The expressions obtained
in this study reduce to Koiter’s formulas, when only the first-order term of the thickness
variation parameter is retained in the analysis. Results from the asymptotic formula are
compared with the those obtained through the purely numerical techniques of the finite
difference method and the shooting method. We first deal with homogeneous shells;
then we discuss shells made of composite materials in some detail.
2.1 Introductory Remarks

There is a vast literature devoted to buckling of cylindrical shells of constant thickness.
The problem regarding the influence of thickness variation on the buckling load has not
gained attention and remains open even today. To the best of our knowledge, the first
work on the effect of thickness variation on the buckling of shells was undertaken by
43
44 DETERMINISTIC PROBLEMS OF SHELLS WITH VARIABLE THICKNESS
Elishakoff, Li, and Starnes (1992). Both the thickness variation and the initial geometric
imperfections were considered axisymmetrical. Solution was composed of two terms:
The first term was associated with the shell of constant thickness, and the second one
incorporated the effects of the thickness variation. The former coincided with Koiter’s
analytical investigation (Koiter, 1963) for constant thickness shells with axisymmet-
ric imperfection, whereas the latter term was derived numerically using the shooting
method. Koiter (personal communication, 1992) derived an analytical formula for the
buckling load of a perfect, non-uniform cylindrical shell. The attendant derivation, ob-
tained by using the energy method, was included in Koiter et al. (1994b). The latter
study supports the central result of the combined theoretical-numerical investigation
(Elishakoff et al., 1992), namely that the effect of thickness variation becomes re-
markable when the thickness pattern is co-configurational with the initial imperfection.
However, further investigation shows that the most detrimental effect of the axisymmet-
ric thickness variation occurs at twice the wave number of the classical buckling mode.
This chapter examines in detail the buckling of the cylindrical shell with small
thickness variations. Our analysis is based on a system of linearized governing dif-
ferential equations of perfect shells with variable thickness. The asymptotic formulas
in terms of ( is the thickness non-uniformity parameter) are derived by a hybrid
perturbation-weighted residuals methods. In comparison with formulas (Koiter, per-
sonal communication, 1992) that are linear in , these asymptotic formulas also contain
the quadratic term, which results in a higher accuracy. In addition to the analytic inves-
tigation, numerical study was also performed, and the results stemming from different
methods were compared and discussed.
2.1.1 Basic Equations for Homogeneous Shells

The linear equations governing the axially compressed, non-uniform cylindrical shell
(Figure 2.1) are as follows:
2 2
dh ∂ F ∂2 F d 2h ∂ 2 F ∂2 F
h ∇ ∇ F +2
2 2 2
−ν 2 −h 2 −ν 2
dx ∂x2 ∂y dx ∂x2 ∂y
3
dh ∂ F ∂ F
3
dh ∂ F 3
Eh 3 ∂ 2 W
− 2h − ν − 2(1 + ν)h = (2.1)
dx ∂x3 ∂ x∂ y 2 d x ∂ x∂ y 2 R ∂x2
Eh 3 1 ∂2 F 3Eh 2 dh 2
∇ ∇
2 2
W + + ∇ W
12(1 − ν 2 ) R ∂x2 12(1 − ν 2 ) d x
3
3Eh 2 ∂ W ∂ 3 W dh
+ + ν
12(1 − ν 2 ) ∂ x 3 ∂ x∂ y 2 d x
2 2
6Eh dh ∂ W ∂2W
+ +ν 2
12(1 − ν 2 ) d x ∂x2 ∂ y

3Eh 2 ∂2W ∂ 2 W d 2h
+ + ν
12(1 − ν 2 ) ∂ x 2 ∂ y2 d x 2
3Eh 2 dh ∂ 3 W ∂2W
+ + P0 =0 (2.2)
12(1 + ν) d x ∂ x∂ y 2 ∂x2
2.1 INTRODUCTORY REMARKS 45
Figure 2.1 Notations and sign conventions.
where W and F represent the radial displacement (positive outward) and the Airy stress
function, respectively; ν is Poisson’s ratio; E the modulus of elasticity; P0 denotes the
uniform axial load at the ends of the shell; h(x) is the shell thickness, assumed here
varying only axisymmetrically,

2 px
h(x) = h 0 1 − cos (2.3)
R
where h 0 is the nominal thickness of the shell and and p are the non-dimensional
parameters indicating the magnitude and wave of the thickness variation, respectively.
By introducing the following non-dimensional parameters
x y W F h
ξ= , η= , w= , f = , H= (2.4)
L L L D0 h0
where D0 = Eh 30 /12(1 − v 2 ), the governing equations (2.1) and (2.2) can be rewritten
into their non-dimensional form,

d H 2 ∂2 f ∂2 f ∂2 H ∂2 f ∂2 f
H ∇ ∇ f +2
2¯2¯2
−ν 2 − H 2 −ν 2
dξ ∂ξ 2 ∂η ∂ξ ∂ξ 2 ∂η
3
dH ∂ f ∂3 f d H ∂3 f
− 2H −ν − 2(1 + ν)H
dξ ∂ξ 3 ∂ξ ∂η 2 dξ ∂ξ ∂η2
12(1 − ν 2 )L 2 3 ∂ 2 w
= H (2.5)
Rh 0 ∂ξ 2
P0 L 2 ∂ 2 w L2 ∂2 f dH ∂ ¯ 2
H 3 ∇¯ 2 ∇¯ 2 w + + + 3H 2 ∇ w
D0 ∂ξ 2 Rh 0 ∂ξ 2 dξ ∂ξ
3
2dH ∂ w ∂ 3w d H 2 ∂ 2w ∂ 2w
+ 3H +ν + 6H +ν 2
dξ ∂ξ 3 ∂ξ ∂η2 dξ ∂ξ 2 ∂η
2 2
∂ w ∂ w d H
2
dH ∂ w 3
+ 3H 2 +ν 2 + 3(1 − ν 2 )H 2 =0 (2.6)
∂ξ 2 ∂η dξ 2 dξ ∂ξ ∂η2
Furthermore, in view of separation of variables, we seek solution of Equations (2.5)
and (2.6) in the following form:
nL
f (ξ, η) = f¯ (ξ ) cos η
R
(2.7)
nL
w(ξ, η) = w̄(ξ ) cos η
R
where n denotes the number of waves in the circumferential direction during buckling.
Equations (2.5) and (2.6) are thus transformed into ordinary differential equations
2
(4) d H ¯ nL dH 2 d 2 H ¯
H 2 f¯ − 2H f + −2H 2 +2 −H f
dξ R dξ dξ 2
¯ 4
d H nL 2 f 2 nL d H 2 nL 2
+ 2H + H +2 ν
dξ R R dξ R

d2 H n L 2 ¯ 12(1 − ν 2 )L 2 3
− H ν f = H w̄ (2.8)
dξ R Rh 0
2
d H nL P0 L 2 dH 2 d2 H
H 3 w̄ (4) + 6H 2 w̄ + −2H 3 + + 6H + 3H 2 2 w̄
dξ R D0 dξ dξ
2 4 2 2
d H nL nL dH nL
− 6H 2 w̄ + H 3 − 6H ν
dξ R R dξ R
2
d2 H nL L 2 ¯
− 3H 2 2 ν w̄ + f =0 (2.9)
dξ R Rh 0
In this chapter, three different methods are used to obtain the classical buckling load
Pcl . First, we evaluate the buckling load via an analytical technique, and then compare
it with the results of purely numerical calculations.
2.1.2 Hybrid Perturbation-Weighted Residuals Method

In the hybrid perturbation-weighted residuals method (Koiter et al., 1994a), we assume
w̄(ξ ) in the form
pL 3 pL
w̄(ξ ) = A cos ξ + B cos ξ (2.10)
R R
where p is the number of half-waves along the shell length at buckling and A and B
are undetermined constants. This buckling pattern satisfies the boundary conditions of
the simple supports. The first term of the two-term approximation (2.10) is the exact
buckling mode for the shell of constant thickness, and the second term is introduced to
account for the thickness variation.
In order to solve the compatibility equation (2.8) for f¯ , the perturbation procedure
will be employed here. To this end, f¯ is expressed in terms of the thickness variation
parameter as
f¯ (ξ ) = f 0 (ξ ) + f 1 (ξ ) + 2 f 2 (ξ ) + · · · (2.11)
Substituting (2.11) into (2.8) and keeping (2.3) in mind, one has, after collecting
the like terms in ,
(4) (4)
f¯ − 2N 2 f 0 + f 0 − 4c2 z 2 w̄ + f 1 − 2N 2 f 1 + N 4 f 1 − 2 cos 2Pξ f 0
(4)
− 4P sin 2Pξ f 0 + 4N 2 cos 2Pξ f 0 − 4P 2 cos 2Pξ f 0

− 4P N 2 sin 2Pξ f 0 − (2N 4 + 4ν P 2 N 2 ) cos 2Pξ f 0 + 12c2 z 2 cos 2Pξ w̄

(4)
+ 2 f 2 − 2N 2 f 2 + N 4 f 2 + cos2 2Pξ f 0 − 2 cos 2Pξ f 1
(4) (4)
+ 4P cos 2Pξ sin 2Pξ f 0 − 4P sin 2Pξ f 1 + (−2N 2 cos2 2Pξ
+ 8P 2 sin2 2Pξ + 4P 2 cos2 2Pξ ) f 0 − (4P 2 − 4N 2 ) cos 2Pξ f 1
− 4P N 2 cos 2Pξ sin 2Pξ f 0 + 4P N 2 sin 2Pξ f 1
+ (N 4 cos2 2Pξ + 8ν N 2 P 2 sin2 2Pξ + 4ν P 2 N 2 cos2 2Pξ ) f 0

− (2N 4 + 4P 2 N 2 ν) cos 2Pξ f 1 − 12c2 z 2 cos2 2Pξ w̄ + · · · = 0 (2.12)
where
pL nL L
P= , N= , z=√ , c= 3(1 − ν 2 ) (2.13)
R R Rh 0
From Equation (2.12), we obtain
L ( f 0 ) = 4c2 z 2 w̄ (2.14)
+ 4P sin 2Pξ f 0 − 4N 2 cos 2Pξ f 0
(4)
L ( f 1 ) = 2 cos 2Pξ f0
+ 4P 2 cos 2Pξ f 0 − 4P N 2 sin 2Pξ f 0
+ (2N 4 + 4ν P 2 N 2 ) cos 2Pξ f 0 − 12c2 z 2 cos 2Pξ w̄ (2.15)
L ( f 2 ) = − cos2 2Pξ f 0 + 2 cos 2Pξ f 1 − 4P cos 2Pξ sin 2Pξ f 0

(4) (4)
+ 4P sin 2Pξ f 1 − (−2N 2 cos2 2Pξ + 8P 2 sin2 2Pξ

+ 4P 2 cos2 2Pξ ) f 0 + (4P 2 − 4N 2 ) cos 2Pξ f 1
+ 4P N 2 cos 2Pξ sin 2Pξ f 0 − 4P N 2 sin 2Pξ f 1
− (N 4 cos2 2Pξ + 8ν N 2 P 2 sin2 2Pξ + 4ν P 2 N 2 cos2 2Pξ ) f 0
+ (2N 4 + 4P 2 N 2 ν) cos 2Pξ f 1 + 12c2 z 2 cos2 2Pξ w̄ (2.16)
where the operator L (·) is defined as
L ( f ) = f (4) − 2N 2 f + N 4 f (2.17)
Equations (2.14)–(2.16) can be solved analytically with the aid of the computerized
symbolic algebra Mathematica (Wolfram, 1991) for f 0 , f 1 , and f 2 to yield
f 0 = a1 cos Pξ + a2 cos 3Pξ

f 1 = a3 cos Pξ + a4 cos 3Pξ + a5 cos 5Pξ (2.18)
f 2 = a6 cos Pξ + a7 cos 3Pξ + a8 cos 5Pξ + a9 cos 7Pξ
where a1 , a2 , . . . , a9 are coefficients depending on A and B, and are given in the

paper by Koiter et al. (1994a). Consult Chapter 7 for more information on the use of
computerized symbolic algebra in the buckling analysis.
Applying the weighted residuals method, namely, in our case the Boobnov-Galerkin
procedure, to the equilibrium equation (2.9), we arrive at
1/2
d H dH 2 d2 H
H 3 w̄ (4) + 6H 2 w̄ + −2H 3 N 2 + 4αcz 2 + 6H + 3H 2 2 w̄
−1/2 dξ dξ dξ

dH 2 dH 2 2 d2 H
− 6H 2 N w̄ + H 3 N 4 − 6H ν N − 3H 2 2 ν N 2 w̄
dξ dξ dξ

cos Pξ
+ z 2 ( f 0 + f 1 + 2 f 2 + · · ·) dξ = 0 (2.19)
cos 3Pξ
where α is the buckling load reduction factor due to the thickness variation defined as
P0 Eh 20
α= , P0,const = (2.20)
P0,const R 3(1 − ν 2 )
and P0,const is the classical buckling load of the uniform shell with constant thick-
ness h 0 .
Case A: We evaluate the classical buckling load corresponding to the buckling

mode at the top of the Koiter’s semi-circle (Koiter, 1980) (Figure 2.2). In this case, the
buckling mode has the same wave numbers in both the axial and the circumferential
directions, and the buckling wave numbers p and n can be expressed as follows:
p0 R
p=n= , p02 = 2c (2.21)
2 h0
then the thickness variation pattern (2.3) becomes

p0 x
h = h 0 1 − cos (2.22)
R
With this assumption, substituting (2.11) and (2.18) into (2.19) and making some
algebraic manipulations lead, when retaining only the terms up to 2 , to the following
Figure 2.2 Koiter’s semi-circle.
eigenvalue problem:

A
[C(, α)]2×2 =0 (2.23)
B
where [C(, α)] is the coefficient matrix containing the thickness variation parameter
and the buckling load reduction factor α. The elements of matrix [C(, α)] are as
follows:

58 − 4ν + 13ν 2 2
C11 = P 4 4 − 4α − 2ν +
25

336 + 66ν 66 + 300ν + 9ν 2 2
C12 = C21 = P −4
+ (2.24)
25 50

4 1412 − 900α 1571010 − 11988ν + 1377ν 2 2
C22 = P +
25 21125
The requirement of vanishing of the determinant matrix [C(, α)] results in the
following equation, when the terms higher than 2 are neglected:

9248 −8048400 + 169632ν − 400968ν 2 2
144α + −
2
+ 72ν + α
25 21125
5648 2824 1737952ν − 478708ν 2 2
+ − ν − =0 (2.25)
25 24 21125
From Equation (2.25) an asymptotic expression can be obtained for the buckling
load reduction factor due to the thickness variation
1 (832 + 464ν − 23ν 2 ) 2
α = 1 − ν − (2.26)
2 512
which coincides with the formula (Koiter, personal communication, 1992)
1
α = 1 − ν (2.27)
2
if the quadratic term in (2.26) is dropped.
Case B: We now investigate the axisymmetric buckling mode, that is,
R
n = 0, p = p0 , P02 = 2c (2.28)
h0
then the thickness variation pattern (2.3) reads

2 p0 x
h = h0 1 − cos (2.29)
R
For this case, we obtain, by retaining the terms up to 2 , the following asymptotic
expression for the buckling load reduction factor:
25 2
α =1−− (2.30)
32
which again coincides with Koiter’s linear formula (Koiter, personal communica-
tion, 1992)
α =1− (2.31)
if the quadratic term is ignored.
2.1.3 Solution by Finite Difference Method

The finite difference method, which is particularly useful for the buckling problems
of structures of complicated geometry or varying flexural rigidity, is used here. This
method is based on the use of approximate algebraic expressions for the derivatives
of unknown variables that appear in the fundamental governing equations. The follow-
ing expressions of the central difference are used to approximate the corresponding
derivatives:
f i+1 − f i−1 f i+1 − 2 f i + f i−1
fi = , 2 f i =
2d d2
f i+2 − 2 f i+1 − 2 f i−1 − f i−2
3 f i = , (2.32)
2d 3
f i+2 − 4 f i+1 + 6 f i − 4 f i−1 + f i−2
4 f i =
d4
where d is the distance between neighboring nodal points.
Using Equation (2.32), the differential equations (2.8) and (2.9) are approximated
by the finite difference equations,

G 1i G 2i G 1i G 2i G 3i G 4i
− 3 f i−2 + −4 4 + 3 + 2 − f i−1
d4 2d d d d 2d

G 1i G 3i G 1i G 2i G 3i G 4i
+ 6 4 − 2 2 + G 5i f i + −4 4 − 3 + 2 + f i−1
d d d d d 2d

G 1i G 2i G 6i G 6i G 6i
+ + 3 f i+2 + 2 wi−1 − 2 2 wi + 2 wi+1 = 0 (2.33)
d4 2d d d d

G 7i G 7i G 7i G 8i G 9i
f i−1 − 2 2 f i + 2 f i+1 + − 3 wi−2
d2 d d d4 2d

G 8i G 9i G 10i G 11i G 8i G 10i
+ −4 4 + 3 + 2 − wi−1 + 6 4 − 2 2 + G 12i wi
d d d 2d d d

G 8i G 9i G 10i G 11i G 8i G 9i
+ −4 4 − 3 + 2 + wi+1 + + wi+2 = 0
d d d 2d d4 2d 3
(2.34)
Here the derivatives H (ξi ) and H (ξi ) are evaluated analytically. By subdividing the
shell length domain (−L/2, L/2) into M equal segments and applying (2.33) and (2.34)
to each nodal point, points near the ends of the shell are influenced by the boundary
conditions. Here we consider the case of simply supported boundary conditions, namely,
w0 = w0 = f 0 = f 0 = w M = wM = f M = f M = 0 (2.35)
or in view of (2.32),
G 1i = [H (ξi )]2 , G 2i = −2H (ξi )H (ξi )
G 3i = −2N 2 [H (ξi )]2 + 2[H (ξi )]2 − H (ξi )H (ξi )
G 4i = 3N 2 H (ξi )H (ξi ),
G 5i = H 2 N 4 + 2ν N 2 [H (ξi )]2 − ν N 2 H (ξi )H (ξi )
G 6i = −12(1 − ν 2 )z 2 [H (ξi )]3 , G 7i = z 2 (2.36)

G 8i = [H (ξi )] , 3
G 9i = 6[H (ξi )] H (ξi )
2
G 10i = −2N H + 4αcz 2 + 6H (ξi )[H (ξi )]2 + 3[H (ξi )]2 H (ξi )
2 3
G 11i = −6N 2 [H (ξi )]2 H (ξi ),

G 12i = N 4 [H (ξi )]3 − 6ν N 2 H [H (ξi )]2 − 3ν N 2 [H (ξi )]2
w−1 = w1 , f −1 = f 1 , w M+1 = w M−1 , f M+1 = f M−1 (2.37)

Thus, we establish a system of simultaneous algebraic equations,
[C(ξi , α)](2M+2)×(2M+2) {δ}(2M+2)×(2M+2) = 0 (2.38)
where [C(ξi , α)] is the coefficient matrix, whose elements depend on the shell geometry,
nodal point coordinates, and elastic constants as well as the unknown buckling load
reduction factor α; {δ} represents a column vector containing the unknown values of
functions of w and f at the nodal points.
Setting the determinant of [C(ξi , α)] equal to zero gives the approximate value
for the classical buckling load reduction factor α or the classical buckling load that
improves in accuracy with an increase in the number of subdivided segments. In the
implementation of this process, the classical buckling load reduction rate α is sought
through iterations.
2.1.4 Solution by Godunov-Conte Shooting Method

The differential equations (2.8) and (2.9), together with the boundary conditions of

simple supports ( f¯ = f¯ = w̄ = w̄ = 0) at the ends of the shell, can be solved for
the classical buckling load by use of the shooting method. However, as pointed out
by Grigolyuk et al. (1971) in the problem of buckling of cylindrical shells, when the
non-dimensional parameter L(1 − ν 2 )0.25 (Rh)−0.5 exceeds 10, the coefficient matrix of
the algebraic equations, from which the missing initial conditions are solved, becomes
too ill-conditioned, which may lead to the loss of accuracy or even completely incorrect
results. Here the modified version of the shooting method, known as the Godunov-Conte
method (Elishakoff and Charmats, 1977), is employed. It utilizes the Gram-Schmidt
orthogonalization procedure during the integration steps to prevent the ill-conditioning
problem so that more accurate results can be obtained than those furnished by the
ordinary shooting method. For the detailed description of the Godunov-Conte method,
consult Chapter 5.
2.1.5 Numerical Results and Discussion

The results for the classical buckling load reduction α from the preceding three methods
are listed in Tables 2.1 and 2.2 for different values of the thickness variation parameter .
A very good match between the results from different methods is shown up to the
value = 0.05. The increasingly bigger difference is observed between the results of
the first-order approximation given by Equation (2.27) [or Equation (2.31)] and those
Table 2.1. Comparison of buckling loads derived via different methods

for Case A (ν == 0.3)
Asymptotic formula
Koiter’s Second-order
formula, approximation, Shooting Finite
Eq. (2.27) Eq. (2.26) method difference
0.0 1.0 1.0 1.0 1.0
0.01 0.999 0.998 0.999 0.998
0.05 0.993 0.988 0.989 0.988
0.10 0.985 0.966 0.967 0.966
0.15 0.975 0.935 0.939 0.938
2.2 BUCKLING OF AN AXIALLY COMPRESSED IMPERFECT CYLINDRICAL SHELL 53
Table 2.2. Comparison of buckling loads derived via different methods

for Case B (ν == 0.3)
Asymptotic formula
Koiter’s Second-order
formula, approximation, Shooting Finite
Eq. (2.31) Eq. (2.30) method difference
0.0 1.0 1.0 1.0 1.0
0.01 0.990 0.990 0.990 0.990
0.05 0.950 0.948 0.949 0.948
0.10 0.900 0.892 0.895 0.894
0.15 0.850 0.832 0.837 0.836
of numerical solutions as becomes larger. Even though the first-order asymptotic

approximate formula may not be sufficiently accurate as reaches 0.1, the second-
order asymptotic formula (2.26) [or (2.30)] retains a good accuracy even for as big
as 0.15. Thus, owing to its higher accuracy, Equations (2.26) and (2.30) can be used
to obtain a sufficiently good estimate of the buckling load reduction factor due to the
thickness variation.
These results also show that the effect of certain types of thickness variation on
buckling load deserves special attention. Although the thickness variation pattern akin
to the classical buckling mode (Case A) may have some effect on the classical buckling
load (the classical buckling load is decreased by over 6% when = 0.15), the most
detrimental effect of thickness variation occurs when the wave number of axisymmetric
thickness variation is twice that of the classical buckling mode (Case B). In this situation,
even if the amplitude of the thickness variation is as small as 0.1, the thickness variation
reduces the buckling load by 10% from its counterpart on the shell with constant
thickness. When = 0.15, the classical buckling load is decreased by over 15%. Thus, in
the absence of initial geometric imperfection, this particular kind of thickness variation
may constitute the most important factor in the buckling load reduction.
2.2 Buckling of an Axially Compressed Imperfect Cylindrical Shell

of Variable Thickness
In 1963, Koiter investigated the effect of axisymmetric imperfections in the shape
of the axisymmetric buckling mode on the buckling of cylindrical shells. No similar
information seems to be available for the effect of axisymmetric thickness variations
of type h = h 0 [1 − cos( px/R)], where the wave number p was expected (naively)
to be most critical when it coincides with the wave number p0 = [2c R/ h 0 ]1/2 of the
axisymmetric buckling mode of the shell of uniform thickness h 0 . Here we aim at
providing the additional information on the effect of such thickness variations. In this
section, we will follow the work by Koiter et al. (1994b).
The simplest approach to the problem is a direct discussion of the energy criterion of
stability by means of the second variation of the potential energy. Comparable result are
obtained by means of the linear equations of neutral equilibrium in terms of Airy stress
functions F and the normal deflection w. The end conditions are largely ignored in the
energy approach, but the conditions of simple support are fully taken into account in
the solution of the equations of neutral equilibrium. Therefore, the latter results should
be more accurate, even though the energy approach does not claim to be valid beyond
the first order. Results from the asymptotic formulas are compared with those obtained
through the purely numerical technique of the shooting method.
2.2.1 Direct Discussion of Energy Criterion

As discussed in Koiter et al. (1994b), the linear prebuckling state of axial compression
for a cylindrical shell of constant thickness h 0 is characterized by a uniform axial stress
resultant N x = −λEh 20 /c R, where E is the Young’s modulus and R is the radius of the
shell; c = [3(1 − ν 2 )]1/2 (ν is the Poisson ratio), and the critical value of load factor
λ is unity. The associated uniform outward radial deflection is νλh 0 /c. A deflection
from the fundamental prebuckling state is described by the axial, circumferential, and
radial components (u, v, w), the latter positive outward. We employ Flügge’s notation
of primes and dots for derivatives with respect to the axial and circumferential non-
dimensional coordinates x/R and y/R. The second variation of the energy for shallow
buckling modes is now given by (Koiter, 1945, 1980)

Eh 0 d S 1
P2 [u ] = u 2 + (v · + w)2 + 2νu (v · + w) + (1 − ν)(u · + v )2
2(1 − ν )R
2 2 2

h2 λh 0 2
+ 0 2 {(w + w·· )2 + 2(1 − ν)[(w· )2 − w w ·· ]} − (1 − ν 2 ) (w )
12R cR
(2.39)
where the integration is extended over the entire shell area 2π R L. The equations of
neutral equilibrium are obtained from the second variation by putting its first variation
equal to zero. The resulting linear equations in u, v, and w are completely equivalent to
the equations in terms of the Airy stress function and the normal deflection as discussed
by Koiter (1963). To include the effect of mid-surface geometric imperfections of
type w0 = −µh 0 cos( p0 x/R), where p0 = [2c R/ h 0 ]1/2 is the wave number of the
axisymmetric buckling mode, we supplement the second variation with the additional
bilinear term

Eh 20
λP11 [u 0 , u ] = −λ 3 w0 w d S (2.40)
cR
We shall also need the third variation of the potential energy

Eh 0 d S
P3 [u ] = {u + ν(v · + w)}(w )2
2(1 − v 2 )R 3

+ {νu + v · + w}(w · )2 + (1 − ν)(u̇ + v )w w · (2.41)

Cylindrical shells of adequate length L (say, L/R > 1) and with reasonably rigid
support at the ends (say, simply supported or clamped ends) exhibit many simultaneous
buckling modes in the interior at the critical load factor λ = 1. These modes are all
sinusoidal in both directions with wave numbers p and q, connected by the equation
p 2 + q 2 = pp0 , where p0 = [2c R/ h 0 ]1/2 is a large number. In order to obtain a non-
zero result for the third variation, we must retain the axisymmetric mode with wave
number p0 and one or more asymmetric modes such that the sum of positive or negative
wave numbers is zero, both in the axial and circumferential directions. We select the
asymmetric mode p = q = m = 12 p0 at the top of the Koiter’s semi-circle, resulting in
the formulas (Koiter, 1980)
ν 1−ν
u/ h 0 = − b0 sin p0 x/R + Cm sin(mx/R) cos(my/R),
2m 4m
3+ν
v/ h 0 = − Cm cos(mx/R) sin(my/R) (2.42)
4m
w/ h 0 = b0 cos( p0 x/R) + Cm cos(mx/R) cos(my/R)
where Cm and b0 are constants and m is the wave number of the buckling mode in the
circumferential direction of the shell.
Substituting from u, v, and w into the sum of the second and third variations for
the cylindrical shell of constant thickness and length L, the result is Equation (3.15) of
Koiter (1980)

π ELh 30 3c
P2 [u] + P3 [u] = (1 − λ) 8b02 + Cm2 + b0 Cm2 (2.43)
8 R 2
The additional term to allow for axisymmetric imperfections w0 = −µh 0 cos( p0 x/R)
is given by

Eh 20 π E Lh 30
P11 [u 0 , u] = −λ 3 w0 w d S = 16λµb0 (2.44)
cR 8 R
We turn now to the effect of small axisymmetric thickness variations described by
h = h 0 [1 − cos( px/R)] (2.45)
where > 0 in order to achieve a detrimental effect by a “thinning” of the wall thick-
ness in the region around x = 0 where the flexural energy dominates. The single es-
sential change in the second variation of the energy is that the extensional rigidity
now contains a factor 1 − cos( px/R), whereas the flexural rigidity contains a factor
[1 − cos( px/R)]3 ≈ 1 − 3 cos( px/R), if we ignore higher than first-order correc-
tions in . No change occurs in the last term of the second variation or in the additional
term to allow for axisymmetric geometric imperfections of the mid-surface.
We now assume that the buckling modes of the shell with a uniform thickness re-
main a good approximation for the buckling modes of the shell with thickness variations.
We are at least ensured that the critical load factor λ, obtained this way, is by the energy
principle an upper bound for the actual critical load factor. The integrand of the second
variation for shells of uniform thickness contains terms with a factor cos2 (mx/R) or
sin2 (mx/R) as well as those with a factor cos2 ( p0 /R). The additional terms due to the
thickness variation all have an additional factor cos( px/R). For sufficiently long shells,
say L/R > 1, the integrals of type

cos2 (mx/R) cos( px/R) d x, cos2 ( p0 x/R) cos( px/R) d x
are all approximately zero, except for the cases p = 2m = p0 and p = 2 p0 , where the
integrals have the value L/4. It is now a simple matter to evaluate the formulae for
the second variation of shells with thickness variations of wave numbers p = p0 or
p = 2 p0

π ELh 30 1
Case p = p0 : P2 [u] = 8(1 − λ)b0 + 1 − λ − ν Cm2
2
(2.46)
8 R 2
π ELh 30
Case p = 2 p0 : P2 [u] = 8(1 − λ − )b02 + (1 − λ)Cm2 (2.47)

8 R
In the absence of initial geometric imperfection, with the partial derivative of P2 [u]
with respect to Cm equal to zero, we obtain
1
Case p = p0 : λ = 1 − ν (2.48)
2
Case p = 2 p0 : λ=1− (2.49)
Thus, an important result of the present analysis is that axisymmetric thickness

variations of wave numbers p0 or 2 p0 entail a reduction of the critical load factor
below unity by 12 ν in the case p = p0 and by in the case p = 2 p0 . The associated
buckling modes are the non-symmetric mode w = cos(mx/R) cos(my/R) in the case
p = p0 and the axisymmetric mode w(x) = cos( p0 x/R) in the case p = 2 p0 .
It is now also a simple matter to discuss the combined effect of thickness variations
of type, h = h 0 [1 − cos( px/R)], where p = p0 or p = 2 p0 , and the most critical type
of asymmetrical geometric imperfections w0 = −µh 0 cos( p0 x/R). For this purpose,
we need the evaluation of the third variation P3 [u], that is, Equation (2.41) with h 0
replaced by h = h 0 [1 − cos( px/R)]. The result reads:

3π c ELh 30 1
Case p = p0 : P3 [u] = 1 − b0 Cm2 (2.50)
16 R 3

3πc ELh 30 1
Case p = 2 p0 : P3 [u] = 1 − b0 Cm2 (2.51)
16 R 6
To a first approximation, we may ignore the factors (1 − /3) and (1 + /6) in the
cubic terms. Leaving aside the dimensional factor π ELh 30 /8R, the energy expression
to be discussed is:

1 3c
Case p = p0 : 8(1 − λ)b02 + 1 − ν − λ Cm2 + 16λµb0 + b0 Cm2 (2.52)
2 2
3c
Case p = 2 p0 : 8(1 − − λ)b02 + (1 − λ)Cm2 + 16λµb0 + b0 Cm2 (2.53)
2
The equations of equilibrium are obtained by putting the partial derivatives with
respect to b0 and Cm equal to zero. In case p = p0 , we have the equations
3c 2
16(1 − λ)b0 + 16λµ + C =0 (2.54)
2 m

1 3c
1 − ν − λ Cm + b0 Cm = 0 (2.55)
2 2
The solution Cm = 0 of the second equation leads to b0 = −λµ/(1 − λ) from

the first equation, and bifurcation buckling with respect to the asymmetric mode with
amplitude Cm occurs at the value b0 = − 23 c(1 − 12 ν − λ).
The equation for the critical load factor is thus

1 3c
(1 − λ) 1 − ν − λ − λµ = 0 (2.56)
2 2
In case p = 2 p0 , we have the conditions of equilibrium obtained from (2.51)
3c 2
16(1 − − λ)b0 + 16λµ + C =0 (2.57)
2 m
3c
(1 − λ)Cm + b0 C m = 0 (2.58)
2
and the equation for the critical load factor is in this case
3c
(1 − λ)(1 − − λ) − λµ = 0 (2.59)
2
The present analysis by means of the energy criterion already permits some signif-
icant conclusions. Periodic axisymmetric thickness variations may result in a fractional
decrease of the critical load factor under axial compression, up to a fraction equal to
defined by (2.45), achieved for a wave number 2 p0 , twice the wave number of the
axisymmetric buckling mode of the uniform shell. It is a more or less unexpected result
that the fractional decrease of the critical load factor in the case of thickness variations
with the wave number of axisymmetric buckling mode is much smaller, namely ν/2.
We are not surprised, however, that the reduction of the critical load factor of the order
, due to thickness variations is far less detrimental than the similar reduction of order
µ1/2 , due to geometric imperfections of the mid-surface. This is also the justification
a posteriori for our ignoring the factors (1 − /3) and (1 + /6) in Equations (2.50)
and (2.51).
2.2.2 Numerical Technique

Now we will approach the problem from a different angle. Here we propose a combined
analytical-numerical technique and closely follow the procedure by Koiter (1963),
which deals with shells of constant thickness.
The governing equations for the buckling of the cylindrical shell with variable
thickness and axisymmetric initial imperfections, under axial compression read
2
∂2W ∂2W ∂2W d 2 W0 ∂ 2 W 1 ∂2W
h 2 ∇ 2 ∇ 2 F − Eh 3 − − +
∂ x∂ y∂ x 2 ∂ y2 d x 2 ∂ y2 R ∂x2
2 2
dh ∂ F ∂2 F d 2h ∂ 2 F ∂2 F
+2 − ν − h − ν
dx ∂x2 ∂ y2 dx2 ∂x2 ∂ y2

dh ∂ 3 F ∂3 F dh ∂ 3 F
− 2h − ν − 2(1 + ν)h =0 (2.60)
dx ∂x3 ∂ x∂ y 2 d x ∂ x∂ y 2
Eh 3 ∂2 F ∂2W ∂2 F ∂2W ∂2 F ∂2W 1 ∂2 F
∇ ∇
2 2
W − − + 2 +
12(1 − ν 2 ) ∂ y2 ∂ x 2 ∂ x 2 ∂ y2 ∂ x∂ y ∂ x∂ y R ∂x2
d 2 W0 ∂ 2 F 3Eh 2 dh ∂ 2 3Eh 2 dh ∂ 3 W
− + ∇ W+
dx ∂y
2 2 12(1 − ν ) d x ∂ x
2 12(1 + ν) d x ∂ x∂ y 2
2 2
3Eh 2 dh ∂ 3 W ∂3W 6Eh dh ∂ W ∂2W
+ +ν + +ν 2
12(1 − ν 2 ) d x ∂ x 3 ∂ x∂ y 2 12(1 − ν 2 ) d x ∂x2 ∂y

3Eh 2 d 2 h ∂ 2 W ∂2W
+ + ν =0 (2.61)
12(1 − ν 2 ) d x 2 ∂ x 2 ∂ y2
where F is the Airy stress function defined as
∂2 F ∂2 F ∂2 F
Nx = ; Ny = ; Nx y = − (2.62)
∂ y2 ∂x2 ∂ x∂ y
With the non-dimensional notations

x y W F h
ξ= , η= , w= , f = , H= (2.63)
L L h0 D0 h0
where D0 = Eh 30 /[12(1 − ν 2 )] is the flexural rigidity of the shell with nominal thickness
h 0 , the governing equations can be expressed in non-dimensional forms as
∂2 F ∂2W ∂2 F ∂2W ∂2 F ∂2W L2 ∂2 F

H 3 ∇¯ 2 ∇¯ 2 W − − + 2 +
∂η2 ∂ξ 2 ∂ξ 2 ∂η2 ∂ξ ∂η ∂ξ ∂η Rh 0 ∂ξ 2
d 2 W0 ∂ 2 F 2dH ∂ ¯ 2 2dH ∂ W
3
− + 3H ∇ W + 3(1 − ν)H
∂ξ 2 ∂η2 dξ ∂ξ dξ ∂ξ ∂η2
3
2dH ∂ W ∂3W d H 2 ∂2W ∂2W
+ 3H +ν + 6H +ν 2
dξ ∂ξ 3 ∂ξ ∂η2 dξ ∂ξ 2 ∂η

d2 H ∂2W ∂2W
+ 3H 2 2 + ν =0 (2.64)
dξ ∂ξ 2 ∂η2
2 2
2
d H ∂ F ∂ 2
F d 2
H ∂ F ∂ 2
F
H ∇¯ ∇¯ F + 2
2 2 2
−ν 2 − H −ν 2
dξ dξ 2 ∂η dξ 2 ∂ξ 2 ∂η
2
dH ∂ F ∂3 F d H ∂3 F
− 2H − ν − 2(1 + ν)H
dξ ∂ξ 3 ∂ξ ∂η2 dξ ∂ξ ∂η2
2 2
∂ W ∂ 2 W ∂ 2 W d 2 W0 ∂ 2 W L2 ∂2W
− 12(1 − ν 2 )H 3 − − + =0
∂ξ ∂η ∂ξ 2 ∂η2 dξ 2 ∂η2 Rh 0 ∂ξ 2
(2.65)
where the non-dimensional Laplace operator reads
∂2 ∂2
∇¯ 2 = 2 + 2
∂ξ ∂η
We let
W = W I + WII , F = FI + FII (2.66)
where W I , FI represent the non-dimensional prebuckling solutions, and WII and FII
represent non-dimensional small increments at buckling. A direct substitution into
Equations (2.64) and (2.65) and deletion of products of the small increments yields a
set of non-linear governing equations for the prebuckling quantities
∂ 2 FI ∂ 2 W I ∂ 2 FI ∂ 2 W I ∂ 2 FI ∂ 2 W I L 2 ∂ 2 FI
H 3 ∇¯ 2 ∇¯ 2 W I − − + 2 +
∂η2 ∂ξ 2 ∂ξ 2 ∂η2 ∂ξ ∂η ∂ξ ∂η Rh 0 ∂ξ 2
d 2 W0 ∂ 2 FI 2dH ∂ ¯ 2 2 d H ∂ WI
3
− + 3H ∇ W I + 3(1 − ν)H
dξ 2 ∂η2 dξ ∂ξ dξ ∂ξ ∂η2
3
2dH ∂ WI ∂ 3 WI d H 2 ∂ 2 WI ∂ 2 WI
+ 3H +ν + 6H +ν
dξ ∂ξ 3 ∂ξ ∂η2 ∂ξ ∂ξ 2 ∂η2

d 2 H ∂ 2 WI ∂ 2 WI
+ 3H 2 2 + ν =0 (2.67)
dξ ∂ξ 2 ∂η2

d H 2 ∂ 2 FI ∂ 2 FI d 2 H ∂ 2 FI ∂ 2 FI
H ∇ ∇ FI + 2
2¯2¯2
−ν −H −ν
dξ ∂ξ 2 ∂η2 dξ 2 dξ 2 ∂η2

d H ∂ 3 FI ∂ 3 FI d H ∂ 3 FI
− 2H − ν − 2(1 + ν)H
dξ ∂ξ 3 ∂ξ ∂η2 dξ ∂ξ ∂η2
2 2
∂ WI ∂ 2 W I ∂ 2 W I d 2 W0 ∂ 2 W I L 2 ∂ 2 WI
− 12(1−ν )H 2 3
− − + =0
∂ξ ∂η ∂ξ 2 ∂η2 ∂ξ 2 ∂η2 Rh 0 ∂ξ 2
(2.68)
and a set of linearized equations governing the increments at buckling

∂ 2 FI ∂ 2 WII ∂ 2 W I ∂ 2 FII ∂ 2 FI ∂ 2 WII ∂ 2 W I ∂ 2 FII
H 3 ∇¯ 2 ∇¯ 2 WII − − − −
∂η2 ∂ξ 2 ∂ξ 2 ∂η2 ∂ξ 2 ∂η2 ∂η2 ∂ξ 2
∂ FI ∂ WII
2 2
∂ W I ∂ FII
2 2
L ∂ FII
2 2
d W0 ∂ 2 FII
2
+2 +2 + −
∂ξ ∂η ∂ξ ∂η ∂ξ ∂η ∂ξ ∂η Rh 0 ∂ξ 2 dξ 2 ∂η2
dH ∂ ¯ 2 d H ∂ 3 WII
+ 3H 2 ∇ WII + 3(1 − ν)H 2
dξ ∂ξ dξ ∂ξ ∂η2
3
2dH ∂ WII ∂ 3 WII d H 2 ∂ 2 WII ∂ 2 WII
+ 3H +ν + 6H +ν
dξ ∂ξ 3 ∂ξ ∂η2 dξ ∂ξ 2 ∂η2

d 2 H ∂ 2 WII ∂ 2 WII
+ 3H 2 2 +ν =0 (2.69)
dξ ∂ξ 2 ∂η2

d H 2 ∂ 2 FII ∂ 2 FII d 2 H ∂ 2 FII ∂ 2 FII
H ∇ ∇ FII + 2
2 ¯2¯2
−ν −H −ν
dξ ∂ξ 2 ∂η2 dξ 2 ∂ξ 2 ∂η2
3
d H ∂ FII ∂ FII
3
d H ∂ FII
3
− 2H −ν − 2(1 + ν)H
dξ ∂ξ 3 ∂ξ ∂η 2 dξ ∂ξ ∂η2
2
∂ W I ∂ 2 WII ∂ 2 W I ∂ 2 WII ∂ 2 W I ∂ 2 WII
− 12(1 − ν 2 )H 3 2 − −
∂ξ ∂η ∂ξ ∂η ∂ξ 2 ∂η2 ∂η2 ∂ξ 2

d W0 ∂ WII
2 2
L ∂ WII
2 2
− + =0 (2.70)
dξ ∂η
2 2 Rh 0 ∂ξ 2
Let us first perform the prebuckling analysis. When the shell is subjected to a
uniform axial compression and both the thickness variation and the initial geometric
imperfection are axisymmetric, there exists an axisymmetric prebuckling state. This
axisymmetric prebuckling state can be described as
1
FI (ξ, η) = − N̄ 0 η2 + FI∗ (ξ ), W I (ξ, η) = W I∗ (ξ ) (2.71)
2
where
N0 L 2
N̄ 0 = (2.72)
D0
Here N0 is the axial compressive stress resultant.
Substitution of Equation (2.71) into the prebuckling governing equations (2.67)
and (2.68) yields
d 4 W I∗ d 2 W I∗ L 2 d 2 FI∗ d 2 W0 d H d 3 W I∗
H3 4
+ N̄ 0 2
+ 2
+ N̄ 0 2
+ 6H 2
dξ dξ Rh 0 dξ dξ dξ dξ 3
2 2 ∗ 2 2 ∗
d H d WI 2 d H d WI
+ 6H + 3H =0 (2.73)
dξ dξ 2 dξ 2 dξ 2

d 4 FI∗ d H 2 d 2 FI∗ d 2 H d 2 FI∗
H2 + 2 + ν N̄ 0 − H + ν N̄ 0
dξ 4 dξ dξ 2 dξ 2 dξ 2
d H d 3 FI∗ 12(1 − ν 2 )L 2 3 d 2 W I∗
− 2H − H =0 (2.74)
dξ dξ 3 Rh 0 dξ 2
In view of the thickness variation, the prebuckling terms W I∗ and FI∗ are further
assumed to have the expression as follows:
W I∗ = W I,K
∗ ∗
+ W I,T
(2.75)
FI∗ = FI,K
∗ ∗
+ FI,T
The parameter is exactly the same one appearing in the thickness variation pattern
of Equation (2.45). This means that, when thickness is a constant (i.e., = 0), the second
terms in Equation (2.75) vanish. Notice that the terms with subscript ‘K ’ describe the
situation of shells of constant thickness (dealt with by Koiter, 1963) and the terms
with subscript ‘T ’ are the additional terms due to the thickness variation. Substituting
Equation (2.75) into Equations (2.73) and (2.74), and regrouping in terms of , we
obtain the differential equations for the terms with subscripts ‘K ’ and ‘T ’, respectively,
∗ ∗ 2 ∗
d 4 W I,K L 2 d 2 FI,K d W I,K d 2 W0
+ + N̄ 0 + =0 (2.76)
dξ 4 Rh 0 dξ 2 dξ 2 dξ 2
∗ ∗
d 4 FI,K 12(1 − ν 2 )L 2 d 2 W I,K
− =0 (2.77)
dξ 4 Rh 0 dξ 2
and
∗ ∗ 2 ∗
d 4 W I,T 2dH
d 3 W I,T dH 2 2d H
2 d W I,T
H 3
+ 6 H + 6H + 3H + N̄ 0
dξ 4 dξ dξ 3 dξ dξ 2 dξ 2
∗ 4 ∗
L 2 d 2 FI,T pLξ 2 pLξ 3 pLξ
d W I,K
+ 2
− 3 cos − 3 2
cos + 3
cos
Rh 0 dξ R R R dξ 4
∗ 2 ∗
2dH
d 3 W I,K dH 2 2d H
2 d W I,K
+ 6H 3
+ 6H + 3H 2
=0 (2.78)
dξ dξ dξ dξ dξ 2
∗ ∗ ∗
d 4 FI,T d H d 3 FI,T d2 H d H 2 d 2 FI,T
H 2
− 2 H − H − 2
dξ 4 dξ dξ 3 dξ 2 dξ dξ 2
∗
12(1 − ν 2 )L 2 3 d 2 W I,T 12(1 − ν 2 )L 2 pLξ pLξ
− H + 3 cos − 3 2 cos2
Rh 0 dξ 2 Rh 0 R R
2 ∗ 4 ∗
pLξ d W I,K pLξ pLξ d FI,K
+ 3 cos3 2
+ −2 cos + 2 cos2
R dξ R R dξ 4
∗ ∗
d H d 3 FI,K d2 H d H 2 d 2 FI,K
− 2H − H − 2
dξ dξ 3 dξ 2 dξ dξ 2

d2 H dH 2
− ν N̄ 0 H − 2 =0 (2.79)
dξ 2 dξ
Equations (2.76) and (2.77) admit the following solution:
∗ λ p0 Lξ
FI,K = 4µc cos (2.80)
4ρ 4 + 1 − 4λρ 2 R
ν Rh 0 4λρ 2 p0 Lξ
W I,K = N̄ 0 − µ 4 cos (2.81)
12(1 − ν ) L
2 2 4ρ + 1 − 4λρ 2 R
where λ = N̄ 0 Rh 0 /4cL 2 and ρ 2 = h 0 p02 /4c R.

Solutions given by Equations (2.81) and (2.82) are identical to the prebuckling
solutions given by Koiter (1963). Here they are presented in a non-dimensional form.
The terms associated with the thickness variation are obtained by solving numeri-
cally Equations (2.78) and (2.79) with proper boundary conditions. Here the Godunov-
Conte method (Godunov, 1961; Roberts and Shipman, 1972; Conte, 1966; Elishakoff
and Charmats, 1977) is employed. The Godunov-Conte method is described in detail
in Chapter 6. The Godunov-Conte method is a modified shooting method, using the
Gram-Schmidt orthogonalization procedure during the integration steps to prevent the
ill-conditioning of the coefficient matrix of the algebraic equation, from which the miss-
ing initial conditions are solved. It is worth mentioning that the Gram-Schmidt orthog-
onalization procedure is essential for many shells encountered in engineering practice.
According to Conte (1966), when the nondimensional parameter L(1 − ν 2 )0.25 (Rh)−0.5
exceeds 10, the problem of ill-conditioning is bound to occur in the buckling analysis
of cylindrical shells.
The boundary conditions for the terms with subscript T are taken as follows:
∗ ∗
∗
d 2 FI,T ∗
d 2 W I,T
FI,T = = W I,T = =0
dξ 2 dξ 2 (2.82)
1
ξ =±
2
which satisfy the simply supported boundary conditions at the ends of the shell.
Now, let us perform the buckling analysis. As in the prebuckling state treatment,
the terms of FII∗ and WII∗ are assumed to be composed of two parts, respectively,
WII∗ = WII,K
∗ ∗
+ WII,T
(2.83)
FII∗ = FII,K
∗ ∗
+ FII,T
Substitution of Equation (1.83) into Equation (1.70) and again regrouping accord-
ing to the parameter leads to
∗ ∗ ∗
∗
∂ 2 W I,K ∂ 2 WII,K 2
2 d W0
d 2 WII,K
∇¯ 2 ∇¯ 2 FII,K + 12(1 − ν 2 ) + 12(1 − ν )
∂ξ 2 ∂η2 dξ 2 ∂η2
∗
12(1 − ν 2 )L 2 ∂ 2 WII,K
− =0 (2.84)
Rh 0 ∂ξ 2
and
∗ ∗ ∗ ∗
2¯2 ∗ d H 2 ∂ 2 FII,K ∂ 2 FII,T ∂ 2 FII,K ∂ 2 FII,T
H ∇ ∇ FII,T + 2
¯
2
+ 2 2 −ν − ν
dξ dξ 2 ∂ ξ ∂η2 ∂η2
∗ ∗ ∗ ∗
d 2 H ∂ 2 FII,K ∂ 2 FII,T ∂ 2 FII,K ∂ 2 FII,T
−H + −ν − ν
dξ 2 ∂ξ 2 ∂ξ 2 ∂η2 ∂η2
∗ ∗ ∗ ∗
d H ∂ 3 FII,K ∂ 3 FII,T ∂ 3 FII,K ∂ 3 FII,T
− 2H + − ν − ν
dξ dξ 3 ∂ξ 3 ∂ξ ∂η2 ∂ξ ∂η2
∗ ∗
d H ∂ 3 FII,K ∂ 3 FII,T
− 2(1 + ν)H +
dξ ∂ξ ∂η2 ∂ξ ∂η2
2 ∗ 2 ∗ ∗ ∗
∂ W I,K ∂ WII,T ∂ 2 W I,T ∂ 2 WII,K
+ 12(1 − ν 2 ) H 3 +
∂ξ 2 ∂η2 ∂ξ 2 ∂η2
∗ ∗ ∗ ∗
∂ 2 W I,T ∂ 2 WII,T d 2 W0 ∂ 2 WII,T L 2 ∂ 2 WII,T
+ + −
∂ξ 2 ∂η2 dξ 2 ∂η2 Rh 0 ∂ξ 2
∗ ∗
pLξ ∂ 2 W I,K ∂ 2 WII,K
− 12(1 − ν ) H cos
2 2
R ∂ξ 2 ∂η2
∗ ∗
d 2 W0 ∂ 2 WII,K L 2 ∂ 2 WII,K
+ − =0 (2.85)
dξ 2 ∂η2 Rh 0 ∂ξ 2
∗ ∗
Again, as in the prebuckling analysis, the terms WII,K and FII,K follow from the
∗
analysis (Koiter, 1963) for imperfect shells of constant thickness, whereas terms WII,T
∗
and FII,T are associated solely with thickness variation.
Equation (2.84) admits a solution of the following form:
∗ pLξ m Lη
WII,K = C1 cos cos
R R

∗ 8µτ 2 ρ 2 tc2 3 pLξ 8µc2 τ 2 ρ 2 t − 4ρ 2 c pLξ m Lη
FII,K = C1 cos + cos cos
(9ρ 2 + τ 2 )2 R (ρ 2 + τ 2 )2 R R
(2.86)
where C1 is an arbitrary constant, and τ 2 = h 0 n 2 /Rc, t = (1 − λ)−1 .
The solution defined by Equation (2.86) satisfies the boundary conditions for simply
supported edges at the shell ends ξ = ± 12 .
Moreover, an examination of Equation (2.85) reveals that it is subject to separation
of variables of ξ and η. Thus, we assume
∗
pLξ m Lη
WII,T = C1 cos cos
R R
(2.87)
∗
m Lη
FII,T = C1 (ξ ) cos
R
Here (ξ ) is an undetermined function of ξ only.
Substitution of Equation (2.87) into Equation (2.85) leads to

d H n2 L 2 2 dH 2 1 d2 H
H − 2 H
2 (iv)
+ H −2 2 + 2
2
−
dξ R H dξ H dξ 2
4 4
d H n2 L 2 2 n L 2ν d H 2 n 2 L 2 ν d 2 H n2 L 2
+ 2 H + H + −
dξ R 2 R4 H 2 dξ R2 H dξ 2 R 2

dH 2 d2 H 8µτ 2 ρ 2 tc2 −9 p 2 L 2 n2 L 2 3 pLξ
+ 2 −H + ν cos
dξ dξ 2 (9ρ 2 + τ 2 )2 R2 R2 R

8µc2 τ 2 ρ 2 t − 4cρ 2 p2 L 2 n2 L 2 pLξ
+ − + ν cos
(ρ 2 + τ 2 )2 R2 R2 R

d H 8µτ 2 ρ 2 τ c2 27 p 3 L 3 3 pn 2 L 3 3 pLξ
− 2H − ν sin
dξ (9ρ + τ ) 2 2 2 R 3 R 3 R

8µc2 τ 2 ρ 2 t − 4cρ 2 p 3 L 3 pn 2 L 3 pLξ dH
+ − ν sin − 2(1 + ν)H
(ρ + τ )
2 2 2 R 3 R 3 R dξ

24µc2 τ 2 ρ 2 t 3 pLξ 8µc2 τ 2 ρ 2 t − 4cρ 2 pLξ pn 2 L 3
× sin + sin
(9ρ 2 + τ 2 )2 R (ρ 2 + τ 2 )2 R R3

4 p2 n 2 L 4 2 pLξ pLξ
+ 12(1 − ν ) H −µ(t − 1)
2 3
4
cos cos
R R R
∗ n2 L 2 pLξ 2 2
∗ n L pL d 2 W0 n 2 L 2 pLξ
− W I,T 2
cos − W I,T 2
cos ξ − 2 2
cos
R R R R dξ R R

L 2 p2 L 2 pLξ pLξ
+ 2
cos + 12(1 − ν 2 ) H 2 cos
Rh 0 R R R

4 p2 n 2 L 4 2 pLξ pLξ
× −µ(t − 1) cos cos
R4 R R
2 2 2 2 2 2

d W0 n L pLξ L p L pLξ
− cos + cos =0 (2.88)
dξ 2 R 2 R Rh 0 R 2 R
∗
Since the second derivative W I,T is known numerically through the prebuckling
analysis, Equation (2.88), representing a fourth-order ordinary differential equation
with variable coefficients, can be solved for (ξ ) again by the Godunov-Conte method
with the boundary conditions
1
(ξ ) = (ξ ) = 0 at ξ =± (2.89)
2
Taking into account the numerically determined Airy’s stress function (ξ ), sub-
stituting Equations (2.86) and (2.82) into Equation (2.69) and applying the Boobnov-
Galerkin procedure, namely, multiplying each term of the equation by cos( pLξ/R) and
integrating over the shell length, we arrive at

(ρ 2 + τ 2 )2 − 4λρ 2 + 4ρ 4 (ρ 2 + τ 2 )−2 + 16(cµ)2 t 2 ρ 4 τ 4 [(9ρ 2 + τ 2 )−2

+ (ρ 2 + τ 2 )−2 ] − 2cµτ 2 [t − 1 + 8tρ 4 (ρ 2 + τ 2 )−2 ] − (ρ 2 + τ 2 )2

Rh 0 2 2cµτ 2 ρ 2 t
− 4αλρ 2 + 8ρ 2 τ 2 µ(t − 1)I1 + 8 τ I2
L2 (9ρ 2 + τ 2 )2

2µcτ 2 ρ 2 t − ρ 2 2 2 Rh 0 τ 2 2 Rh 0
+ I 3 + I4 + 2 2 I5 + 8µρ 2 τ 2 I1
(ρ + τ )
2 2 2 cL 2 c L
1/2 1/2
h0 h0
+ 6(1 + ) (ρ + τ )p I6 + 6(1 − ν)(1 + )
2 2
ρτ 2 p I6
cR cR
1/2
h0 h0
+ 6(1 + ) (ρ 2 + ντ 2 )ρp I6 + 24(1 + ) (ρ 2 + ντ 2 ) p 2 2 I7
cR cR
Rh 0 2 h0
+ 2cµ(t − 1)τ 2 − 2(1 + ) 2
τ I8 + 12(1 + ) (ρ 2 + ντ 2 ) p 2 2 I9
cL cR

− 2(1 + )(ρ + 2τ ρ + τ )(3 I10 + 3 I11 + I12 ) = 0
4 2 2 4 2 3
(2.90)
where Ii (i = 1 ∼ 12) are integrals listed in the paper by Koiter et al. (1994b).
When the thickness is a constant, all the terms in the curved brackets in Equa-
tion (2.90) vanish; the reduced equation coincides with Equation (5.2) in the study
by Koiter (1963). The smallest solution N̄ 0 of this algebraic equation is the critical
buckling load, or more precisely the upper bound of the critical buckling load, of the
imperfect shell of the variable thickness.
Since the critical buckling load is in need at the very beginning for the solution
∗ ∗ ∗ ∗ ∗
of W I,T , FI,T , and FII,T (WII,T is assumed co-configurational to WII,K in this analysis)
by the shooting method, the critical buckling load is preassigned an initial guess to
start the iterative procedure, which in each step includes a prebuckling analysis and a
buckling analysis as well as a computation of the residue of Equation (2.90). Based
on the knowledge of this residue, the initial guess is modified and used for the next
iteration, and in this way the process is continued. The approximate value of the critical
buckling load is obtained when the residue of Equation (2.90) changes its sign and
is less than a preassumed small quantity such as 10−5 . It should be emphasized that
the initial guess of the critical buckling load is an important consideration for fast
numerical convergence. It is observed that the whole computational process might be
very sensitive to the choice of this initial guess. When is small, the solution furnished
by considerations of shells of constant thickness can be utilized as the initial guess.
Let us investigate a cylindrical shell with different thickness variation and initial
imperfection parameters, The radius R and the length L of this cylinder are fixed at 30 cm
(11.8 in.) and 46.57 cm (18.3 in.), respectively. The nominal thickness h 0 is taken equal
to 0.5 mm (0.0197 in.), Young’s modulus of the material E is 205.8 Gpa (29.8 × 106 psi).
Figure 2.3 Comparison between the numerical results and those of Equation (2.59).
The Poisson ratio ν is equal to 0.3 here. The non-dimensional parameter varies from
zero to 0.2.
For the shell of these dimensions, numerical calculations are performed, and results
are plotted together with those of the asymptotic equation (2.59) (Figure 2.3). It appears
that, for relatively large values of and µ, numerical analysis is needed; however,
for small µ and , Equation (2.59) yields excellent results (Tables 2.3 and 2.4). As is
revealed from Table 2.3, for small values of and µ, results provided by the two methods
are nearly coincident. Comparison becomes less satisfactory for larger values of and
µ. This is understandable if we recall that the asymptotic formula (2.59) represents a
first-order approximation. Table 2.4 demonstrates that, for the perfect shell with the
thickness variation parameter = 0.2, there is a difference between the asymptotic
estimate λ = 0.800 and the numerical result λ = 0.787. This difference for a relatively
large value of is incompatible with our previous study (see Table 2 in Koiter et al.,
1994a), where excellent agreement was documented between the results by the first-
order asymptotic formula and the shooting method for values of only up to 0.05.
Remarkably, however, for the same = 0.2 and the imperfection amplitude µ = 0.01,
the agreement is extremely good: The first-order asymptotic formula yields λ = 0.732,
and the numerical analysis gives λ = 0.730. Such a good agreement appears to be
unexpected. To investigate this transition, additional calculations have been performed
in the range of 0 ≤ µ ≤ 0.01 for fixed at 0.2. The results are listed in Table 2.5. As
can be seen, the agreement between these two methods improves as the imperfection
amplitude µ increases in the range under consideration. Expectedly, a bigger difference
occurs for greater µ; for example, when µ = = 0.2, the first-order asymptotic formula
predicts λ = 0.428 while the numerical method yields λ = 0.415. Still, in view of the
complexity of the problem due to its highly nonlinear nature, the agreement between
these two methods seems to be quite acceptable.
Table 2.3. Comparison of buckling loads from

two different methods for == 0.1 (Case p == 2 p0 )
Imperfection Numerical
amplitude, µ Eq. (2.59) analysis
0.00 0.900 0.894
0.01 0.801 0.800
0.05 0.660 0.649
0.10 0.571 0.549
0.15 0.511 0.482
0.20 0.467 0.432

two different methods for == 0.2 (Case p == 2 p0 )
0.00 0.800 0.787
0.01 0.732 0.730
0.05 0.608 0.607
0.10 0.526 0.519
0.15 0.470 0.459
0.20 0.428 0.415

two different methods for == 0.2 and
0 ≤ µ ≤ 0.01
0.000 0.800 0.787
0.001 0.791 0.778
0.002 0.782 0.772
0.003 0.774 0.765
0.004 0.767 0.759
0.005 0.760 0.754
0.006 0.754 0.749
0.007 0.748 0.744
0.008 0.743 0.739
0.009 0.737 0.734
0.010 0.732 0.730
Figure 2.4 The effect of thickness variation on the buckling load of the perfect shell.
If both the initial geometrical imperfection and the thickness variation are ignored,
the classical buckling load Ncl has a value of 1.04 × 105 N/m (592.5 lb/in.), which is
calculated using the value of the nominal thickness h 0 . If only the thickness variation is
included, then the classical buckling load can be calculated using the methods elucidated
in another study (Koiter et al., 1994a). It is found that the critical thickness variation
(implying the pattern of thickness variation whose wave number is twice that of the
classical buckling mode) has a quite noticeable effect on the classical buckling load.
For example, even if the amplitude of the thickness variation is small, say = 0.1,
the thickness variation produces 10.6% reduction in the classical buckling load. When
= 0.2, the classical buckling load is decreased by 21.3% from its counterpart of the
case without thickness variation. Figure 2.4 shows the change of classical buckling
loads with the critical thickness variation when is between 0. and 0.2.
When the initial imperfection is present, the combination of the initial imperfection
and thickness variation reduces the buckling load even more drastically. Here the initial
imperfection amplitude µ is assumed to range from zero to 0.2. For this imperfect shell,
the reduced buckling load N0 can be readily calculated by use of Equation (5.2) by Koiter
(1963) if the effect of thickness variation is not considered. Taking into account the
critical thickness variation, numerical calculations shows that the influence of thickness
variation is generally not as great as that of the initial imperfection for isotropic shells
(Figure 2.5). In order to assess the effect of the thickness variation on the buckling load
reduction, the so-called thickness variation influence factor β is introduced here, which
is defined as
N0 − N0,T
β= × 100% (2.91)
N0
where N0 and N0,T are buckling loads for shells with and without thickness variation,
Figure 2.5 Buckling load of the imperfect shell with thickness variation.
respectively. In Figure 2.6, the thickness variation influence factor β is plotted versus the
imperfection amplitude µ for = 0.1 and for = 0.2. The curves are seen to be similar
in general form. The thickness effect is most significant in the absence of the initial
imperfection. As Figure 2.5 indicates, for the shell with critical thickness variation
pattern of amplitude, = 0.20 and without the initial imperfection, the buckling load
parameter λ is 0.787, and the thickness variation influence factor β equals 21.3%, which
Figure 2.6 The effect of thickness variation on the buckling load of the imperfect
shell.
Figure 2.7 Imperfection amplitude versus thickness variation influence factor.
indicates a 21.3% reduction from its counterpart of uniform shell. If the shell contains
the initial imperfection of amplitude µ = 0.2 in addition to the thickness variation, the
buckling load parameter λ is reduced to 0.415 and now β is 8%. Thus the inclusion of
thickness variation leads to an 8% decrease in the buckling load from the counterpart
without thickness variation, which is λ = 0.451. For a smaller , the thickness variation
affects the buckling load less appreciably, and the initial imperfection plays the dominant
role in the buckling load reduction. Figure 2.7 depicts the dependence of the buckling
load of the imperfect shells on the thickness variation parameter . With larger values
of , the effect of thickness variation may be more detrimental. When = 0.2 and
µ = 0.02, the thickness variation causes 11% of the further decrease in the critical
buckling load in addition to the reduction from the initial geometrical imperfection,
which is 20%; thus, the total decrease in load-carrying capacity of the shell due to both
geometric and thickness imperfection amounts to 31%. This illustrates that, despite
the fact that the initial geometrical imperfection stands out as the main factor for the
reduction of the critical buckling load and the effect of thickness variation is less
significant in many cases, the thickness variations of certain patterns may cause further
notable decrease in the critical buckling load. Neglect of such a thickness variation,
therefore, is not on the safe side, for design purposes.
The combined effect of thickness variation and initial imperfection on the buckling
load is illustrated in Figure 2.8.
It should be pointed out that all the equations and analytical developments in
Section 2.2 are identical to those in the paper by Koiter (1963), when the thickness
variation parameter is set to be zero. Thus, this part of the monograph can be viewed as
an expansion of Koiter’s work (1963), and it is intended as a contribution to the further
understanding of factors leading to reduction of load-carrying capacity of shells, in
addition to initial imperfections.
2.3 AXIAL BUCKLING OF COMPOSITE CYLINDRICAL SHELLS 71
Figure 2.8 Combined effect of thickness variation and initial imperfection on

the buckling load.
2.3 Axial Buckling of Composite Cylindrical Shells with Periodic

Thickness Variation
Classical buckling of cylindrical shells made of composite materials was studied by
several investigators (Tasi, 1966; Stavsky and Friedland, 1969; Hirano, 1979). The
effect of initial geometric imperfections was pioneered by Khot (1968) and Card (1969).
Further studies were performed by Khot and Venkayya (1970); Tennyson, Chan, and
Muggeridge (1971); Simitses, Shaw, and Sheinman (1985); Arbocz and Hol (1991); and
other investigators. The work on buckling of composite structures done before 1975 was
reviewed by Tennyson (1975). Books specifically devoted to the buckling of composite
shells were written by Rikards and Teters (1975) and Vanin and Semeniuk (1978). The
monographs by Ambartsumian (1974), Vinson and Sierakowski (1986), Vasiliev (1993),
and others also contain chapters on the buckling of composite shell structures.
Due to various factors in the manufacturing process, structures made of composite
materials often exhibit certain types of variations in the wall thickness. In the context
of isotropic shells, the effect of thickness variation on the buckling of axially com-
pressed cylindrical shells has been discussed in the previous two sections. For the case
of composite shells, one may conjecture that thickness variation may affect the buckling
behavior of the structure as well. Despite its importance, so far, buckling of compos-
ite structures with non-uniform wall thickness has not been tackled in the literature.
In the case of composite shells, with the attendant heterogeneity and anisotropy, the
analysis becomes much more complicated. But results from such an analysis appear to
be more important because, unlike the situation with isotropic structures, the designer
may have the opportunity to reduce the deleterious effect by selecting appropriate fiber
orientations.
In this section, we deal with the classical buckling load of the perfect anisotropic
shell with variable thickness. First, starting from the linear shell theory, the governing
partial differential equations are derived. Then a simplification is made, which renders
the equations amenable to the method of separation of variables, and the partial dif-
ferential equations are converted into the ordinary differential equations. Finally, the
finite difference method is applied to obtain the buckling load reduction factor due to
the presence of thickness variation. The numerical results are presented for laminated
shells of three common composite materials such as glass/epoxy, graphite/epoxy, and
boron/epoxy.
We use the linear anisotropic shell theory. The linear strain-displacement rela-
tions are
∂U ∂2W
x = , κx = − 2
∂x ∂x
∂V W ∂2W
y = + , κy = − 2 (2.92)
∂y R ∂y
∂V ∂U ∂2W
γx y = + , κx y = −2
∂x ∂y ∂ x∂ y
where x and y are the axial and circumferential coordinates in the shell middle sur-
face; U and V are the shell displacements along axial and circumferential directions,
respectively; W is the radial displacement, positive outward; x , y , and γx y are strain
components; κx , κ y , and κx y are middle surface curvatures of the shell; and R is the
radius of the cylindrical shell.
Thickness variation of the laminated shell invariably exists due to the imprecision
involved in the fabrication process. Here, we discuss the case of idealized variation of
thickness, namely, the thickness variation is axisymmetric, and in addition, of uniform
nature – each lamina has the same variation pattern:

2 px
h k (x) = h 0,k 1 − cos = h 0,k H (x) (k = 1 ∼ K ) (2.93)
R
where h k and h 0,k are the thickness and the nominal thickness for the kth layer, respec-
tively; and p are the non-dimensional parameters indicating the magnitude and wave
number of the thickness variation, and they are assumed to be coincident for all the
constituent layers; K represents the total number of layers in the laminate. At first sight,
the perfect homology of the thickness variation may appear as a restrictive assumption.
Yet, if the constituent layers are produced by the same manufacturing process and if
they belong to the same fleet of specimens, one can study the case of similar deviations
from uniform thickness. Moreover, such a study may shed some light on the question
of thickness variability and lead to relatively tractable analysis.
With the foregoing assumption, elements of the stiffness matrices [A], [B], and
[D] for the shell with variable thickness are derived as follows:

K
K
Ai j = ( Q̄ i j )k (h k − h k−1 ) = H (x) ( Q̄ i j )k (h 0,k − h 0,k−1 ) = H (x)ai j
k=1 k=1
1 K
1 K

Bi j = ( Q̄ i j )k h 2k − h 2k−1 = [H (x)]2 ( Q̄ i j )k h 20,k − h 20,k−1 = [H (x)]2 bi j
2 k=1 2 k=1
1 K
1 K

Di j = ( Q̄ i j )k kk3 − h 3k−1 = [H (x)]3 ( Q̄ i j )k h 30,k − h 30,k−1 = [H (x)]3 di j
3 k=1 3 k=1
(i, j = 1, 2, 6)
(2.94)
where ai j , bi j , and di j are elements of stiffness matrices for the corresponding uniform
laminate; Q̄ i j s are the transformed reduced stiffnesses of the individual lamina and
have no bearing on the thickness. In the following, we will use the transformed stiffness
matrices [A∗ ], [B ∗ ], and [D ∗ ], which are related to the matrices in (2.94) as follows:
[A∗ ] = [A]−1 , [B ∗ ] = [B][A], [D ∗ ] = [D] − [B][A∗ ][B] (2.95)
thus
1 ∗
Ai∗j = a , Bi∗j = H (x)bi∗j , Di∗j = [H (x)]3 di∗j (2.96)
H (x) i j
where [a ∗ ], [b∗ ], and [d ∗ ] are counterparts, in the corresponding uniform laminate, of
the transformed stiffness matrices [ A∗ ], [B ∗ ], and [D ∗ ], and they are derived from [a],
[b], and [d] as follows:
[a ∗ ] = [a]−1 , [b∗ ] = [b][a], [d ∗ ] = [d] − [b][a ∗ ][b] (2.97)
We will deal with symmetric laminates, for which there is no coupling between
bending and extension. Thus, we have
Bi j = 0, Bi∗j = 0 (i, j = 1, 2, 6) (2.98)
The constitutive relations for the anisotropic laminate are
   ∗  
 x 
  A11 A∗12 A∗16   Nx  
 
y =  A∗12 A∗22 A∗26  N y (2.99)
 
  
 

x y A∗16 A∗26 A∗66 Nx y
   ∗ ∗ ∗
 
 Mx 
  D11 D12 D16 
 kx  
 ∗ ∗ ∗ 
M y =  D12 D22 D26  k y (2.100)

 
 
  
∗ ∗ ∗
Mx y D16 D26 D66 kx y
where N x , N y , and N x y are stress resultants, and Mx , M y , and Mx y are bending and
twisting moments, acting on the mid-surface of a laminate.
The equations of equilibrium for the cylindrical shell read:
∂ Nx ∂ Nx y
+ =0
∂x ∂y
∂ Nx y ∂ Ny
+ =0
∂x ∂y
2
∂ 2 Mx ∂ 2 Mx y ∂ 2 M y ∂2W ∂2W ∂ W 1
+2 + + Nx + 2N x y + Ny − =0
∂x2 ∂ x∂ y ∂ y2 ∂x2 ∂ x∂ y ∂ y2 R
(2.101)
Introducing the Airy stress function F

∂2 F ∂2 F ∂2 F
Nx = , Ny = , Nx y = − (2.102)
∂ y2 ∂x2 ∂ x∂ y
the first two equations of equilibrium (101) are identically satisfied. Substituting (1.99)
into the third equation of equilibrium yields

∗ ∂ W ∗ ∂ W ∗ ∂ W ∗ ∂ W
3 3 3 3
2dH
H ∇d ∗ W + 3H
3 4
2d11 + 2d + 6d + 2d
dx ∂x3 12
∂ x∂ y 2 16
∂ x 2∂ y 26
∂ y3

∗ ∂ W
3
dH 2 2d H
2
∗ ∂ W
2
∗ ∂ W
2
+ 4d66 + 6H + 3H d + d
∂ x∂ y 2 dx dx2 11
∂x2 12
∂ y2

∗ ∂ W ∂2W ∂2W ∂2W 2∂ F
2 2
1
+ 2d66 + Nx + N y + + 2N x y + z =0
∂ x∂ y ∂x2 ∂ y2 R ∂ x∂ y ∂x2
(2.103)
where the differential operator ∇d4∗ is defined as
∗ ∂4 ∗ ∂4 ∗ ∗ ∂4 ∗ ∂4 ∗ ∂
4
∇d4∗ = d11 + 4d + 2(d + 2d ) + 4d + d
∂x4 16
∂ x 3∂ y 12 66
∂ x 2∂ y2 26
∂ x∂ y 3 22
∂ y4
(2.104)
Elimination of U and V from (1.92) leads to the compatibility equation,
∂ 2 x ∂ 2y ∂ 2 x y 1 ∂2W
9 + − − =0 (2.105)
∂ y2 ∂x2 ∂ x∂ y R ∂x2
Substituting (1.98) into Equation (2.105) and using the Airy stress function F, the
equation of compatibility can be written as

d H ∗ ∂3 F ∗ ∂ F
3
∗ ∂ F
3
∗ ∂ F
3
H ∇a ∗ F + H
2 4
a16 3 + a26 − a − 2a
dx ∂y ∂ x 2∂ y 66
∂ x∂ y 2 12
∂ x∂ y 2

∗ ∂ F
3
∗ ∂ F
3
dH 2 d2 H
− 2a22 + 2a + 2 − H
∂x3 26
∂ x 2∂ y dx dx2

∗ ∂ F ∗ ∂ F ∗ ∂ F ∂2W
2 2 2
× a12 + a − a − H 3 z2 2 = 0 (2.106)
∂y 2 22
∂x 2 26
∂ x∂ y ∂x
where
∗ ∂4 ∗ ∂4 ∗ ∗ ∂4 ∗ ∂4 ∗ ∂
4
∇a4∗ = a22 − 2a + (2a + a ) − 2a + a
∂x4 26
∂ x 3∂ y 12 66
∂ x 2∂ y2 16
∂ x∂ y 3 11
∂ y4
(2.107)
Introducing the non-dimensional quantities
x y W F Nx L 2
ξ= , η= , w= , f = ∗
, N̄ x = ∗
,
L L h0 d11 d11
Ny L 2 Nx y L 2 L ai∗j d11
∗
N̄ y = ∗ N̄ x y = , z=√ , āi∗j = ,
d11 d11 Rh 0 h 20
bi j ∗ di∗j
b̄i∗j = , d̄ i j = ∗
(i, j = 1, 2, 6)
h0 d11
(2.108)
where h 0 is the overall thickness of the shell and L is the shell length, (1.103) and
(1.106) can be written in the non-dimensional form:

d H ∗ ∂3 f ∗ ∂ f
3
∗ ∂ f
3
∗ ∂ f
3
H ∇ā ∗ f + H
2¯4
ā16 3 + ā26 − ā − 2 ā
dξ ∂η ∂ξ 2 ∂η 66
∂ξ ∂η2 12
∂ξ ∂η2

∗ ∂ f
3
∗ ∂ f
3
dH 2 d2 H
− 2ā22 + 2 ā + 2 − H
∂ξ 3 26
∂ξ 2 ∂η dξ dξ 2

∗ ∂ f ∗ ∂ f ∗ ∂ f ∂ 2w
2 2 2
× ā12 2 + ā22 2 − ā26 − H 3 z2 2 = 0 (2.109)
∂η ∂ξ ∂ξ ∂η ∂ξ

∗ ∂ w ∗ ∂ w ∗ ∂ w ∗ ∂ w
3 3 3 3
2dH
H ∇d¯ ∗ w + 3H
3¯4
2d̄ 11 3 + 2d̄ 12 + 6 d̄ + 2 d̄
dξ ∂ξ ∂ξ ∂η2 16
∂ξ 2 ∂η 26
∂η3

∗ ∂ w
3
dH 2 2d H
2
∗ ∂ w
2
∗ ∂ w
2
+ 4d̄ 66 + 6H + 3H d̄ + d̄
∂ξ ∂η2 dξ dξ 2 11
∂ξ 2 12
∂η2

∗ ∂ w ∂ 2w ∂ 2w ∂ 2w 2∂ f
2 2
+ 2d̄ 66 + N̄ x 2 + N̄ y + z 2
+ 2 N̄ x y + z =0
∂ξ ∂η ∂ξ ∂η2 ∂ξ ∂η ∂ξ 2
(2.110)
where
∗ ∂ ∂4 ∂4 ∂4 ∗ ∂
4 4
∗ ∗ ∗ ∗
∇¯ ā4∗ = ā22 − 2ā + (2ā + ā ) − 2ā + ā
∂ξ 4 26
∂ξ 3 ∂η 12 66
∂ξ 2 ∂η2 16
∂ξ ∂η3 11
∂η4
(2.111)
∗ ∂ ∂4 ∂4 ∂4 ∗ ∂
4 4
∗ ∗ ∗ ∗
∇¯ d4¯∗ = d̄ 11 4 + 4d̄ 16 3 + 2(d̄ 12 + 2d̄ 66 ) 2 2 + 4d̄ 26 + d̄
∂ξ ∂ξ ∂η ∂ξ ∂η ∂ξ ∂η3 22
∂η4
(2.112)
In a perfect agreement with studies of Hirano (1979) and Vinson and Sierakowski
(1986), the coupling stiffnesses ( A16 , A26 , B16 , B26 , D16 , D26 ) are assumed to be zero.
They are identically zero for the cross-ply laminates. As for symmetric angle-ply lami-
nates, B16 and B26 are zero, and A16 , A26 , D16 , and D26 can be neglected for laminates
with “many” layers. Moreover, we confine our discussion to the buckling of shells under
axial compression (i.e., N x = P0 , N x y = N y = 0).
For the shell of constant thickness, the classical axisymmetric buckling load for
the symmetrically laminated shell reads (Tasi, 1966):

∗
2 d11
Pcl, sym = ∗
(2.113)
R a22
The critical buckling wave number pcl is (Tennyson et al., 1971)
R
pcl2 = ∗ ∗ (2.114)
4 a22 d11
However, in many circumstances, the buckling mode is non-axisymmetric. The

more general expression for the classical buckling load is (Elishakoff, Li, and Starnes,
1993)
2
L C11 C22 C33 + 2C12 C23 C13 − C13
2
C22 − C23
2
C11 − C12
2
C33
Pm,n = (2.115)
mπ C11 C22 − C122
where
2 2 2
mπ n n mπ 2
C11 = A11 + A66 , C22 = A22 + A66
L R R L
4 2 2 4
mπ mπ n n
C33 = D11 + 2(D12 + 2D66 ) + D22
L L R R

A22 B22 n 2 B12 mπ 2
+ 2 +2 +2
R R R R L
(2.116)
mπ n
C12 = C21 = (A12 + A66 )
L R
3
mπ 2 n A22 n n
C23 = C32 = (B12 + 2B66 ) + + B22
L R R R R
3 2
A12 mπ mπ mπ n
C13 = C31 = + B11 + (B12 + 2B66 )
R L L L R
To determine the critical buckling load Pcl for a cylindrical shell with given dimen-
sions and material properties, one determines those integral numbers m and n, which
minimize the value of Pm,n . To encompass the general case, we introduce the following
expression for the classical buckling load of shells of constant thickness:

∗
(0) 2 d11
Pcl = min{Pm,n } = Pcl, sym φ, Pcl, sym = ∗
(2.117)
m,n R a22
where φ is a control parameter accounting for the non-axisymmetric buckling cases.

When the classical buckling mode is axisymmetric, φ takes the value of unity.
In the case of special anisotropy, namely when
A16 = A26 = B16 = B26 = D16 = D26 = 0
the governing equations allow for separation of variables.

We seek solution of Equations (1.109) and (1.110) in the following form:
nL
f (ξ, η) = f¯ (ξ ) cos η
R
(2.118)
nL
w(ξ, η) = w̄(ξ ) cos η
R
where n denotes the number of waves in the circumferential direction during the buck-
ling. Using (1.118), Equations (1.109) and (1.110) are transformed into ordinary dif-
ferential equations,
4 ¯

∗ d f d 2 f¯
H 2 ā22 − N 2
1 (2ā 12 + ā 66 ) + N 4
ā
1 11
¯
f
dξ 4 dξ 2
¯ ¯ 3 ¯

dH 2 ∗ d f 2 ∗ d f ∗ d f
+H N1 ā66 + 2N1 ā12 − 2ā22 3
dξ dξ dξ dξ
2
d2 H d 2 f¯ 3 2 d w̄
2
dH 2 ∗ ¯
+ 2 −H −N 1 ā 12 f + ā 22 − H z =0
dξ dξ 2 dξ 2 dξ 2
(2.119)

∗ d w̄ ∗ d w̄
4 2
∗ ∗
H 3 d̄ 11 4 − 2N12 (d̄ 12 + 2d̄ 66 ) 2 + N14 d̄ 22 w̄
dξ dξ

∗ d w̄ 2 ∗ d w̄ 2 ∗ d w̄
3
2dH
+ 3H 2d̄ 11 3 − 2N1 d̄ 12 − 4N1 d̄ 66
dξ dξ dξ dξ
2
∗ d w̄
2 2
2d H dH 2 ∗
+ 3H + 6H d̄ 11 2 − N1 d̄ 12 w̄
dξ 2 dξ dξ
d 2 w̄ 2 ¯
2d f
+ λ + z =0 (2.120)
dξ 2 dξ 2
() (0)
where λ = Pcl /Pcl and is referred to as the buckling load reduction factor due to the
()
thickness variation; Pcl is the buckling load of shells with variable thickness, and
and N1 are non-dimensional parameters,
2φ L 2 nL
= ∗ ∗ , N1 =
R d11 , a22 R
The governing equations (1.119) and (1.120) constitute a set of fourth-order differ-
ential equations with variable coefficients. Here we employ the finite difference method,
which seems to be particularly useful for the buckling problems of structures of com-
plicated geometry and/or varying flexural rigidity. This method is based on the use of
approximate algebraic expressions for the derivatives of unknown variables that ap-
pear in the fundamental governing equations. The following expressions of the central
difference are used to approximate the corresponding derivatives:

f i+1 − f i−1 f i+1 − 2 f i + f i−1
fi = , 2 f i =
2d d2
f i+2 − 2 f i+1 − 2 f i−1 − f i−2
3 f i = , (2.121)
2d 3
f i+2 − 4 f i+1 + 6 f i − 4 f i−1 + f i−2
4 f i =
d4
where d is the distance between neighboring nodal points.
Using (1.121), the differential equations (1.119) and (1.120) are approximated by
the following finite difference equations:

G 1i G 2i ¯ G 1i G 2i G 3i G 4i ¯
− 3 f i−2 + −4 4 + 3 + 2 − f i−1
d4 2d d d d 2d

G 1i G 3i G 1i G 2i G 3i G 4i ¯
+ 6 4 − 2 2 + G 5i f¯ i + −4 4 − 3 + 2 + f i−1
d d d d d 2d

G 1i G 2i ¯ G 6i G 6i G 6i
+ + 3 f i+2 + 2 w̄i−1 − 2 2 w̄i + 2 w̄i+1 = 0 (2.122)
d4 2d d d d

G 7i ¯ G 7i ¯ G 7i ¯ G 8i G 9i
f − 2 f + f + − w̄i−2
d 2 i−1 d2 i d 2 i+1 d4 2d 3

G 8i G 9i G 10i G 11i G 8i G 10i
+ −4 4 + 3 + 2 − w̄i−1 + 6 4 − 2 2 + G 12i w̄i
d d d 2d d d

G 8i G 9i G 10i G 11i G 8i G 9i
+ −4 4 − 3 + 2 + w̄i+1 + + 3 w̄i+2 = 0
d d d 2d d4 2d
(2.123)
where all the coefficients G ji ( j = 1 ∼ 12) can be analytically evaluated as follows:

∗
G 1i = [H (ξi )]2 ā22 , G 2i = −2H (ξi )H (ξi )ā22 ∗
∗ ∗ ∗

G 3i = −N12 [H (ξi )]2 (2ā12 + ā66 ) + ā22 2[H (ξi )]2 − H (ξi )H (ξi )
∗ ∗
G 4i = (ā66 + 2ā12 )N12 H (ξi )H (ξi ),
∗ ∗

G 5i = [H (ξ )]2 N14 ā11 − N12 ā12 2[H (ξi )]2 − H (ξi )H (ξi )
G 6i = −z 2 [H (ξi )]3 , G 7i = z 2 ,
∗ ∗ (2.124)
G 8i = [H (ξi )]3 d̄ 11 , G 9i = 6[H (ξi )]2 H (ξi )d̄ 11
∗ ∗
G 10i = −2N12 [H (ξ )]3 (d̄ 12 + 2d̄ 66 ) + λ + 6H (ξi )[H (ξi )]2
∗
+ 3[H (ξi )]2 H (ξi ) d̄ 11
∗ ∗
G 11i = −6N12 [H (ξi )]2 H (ξi )(d̄ 12 + 2d̄ 66 )
∗ ∗
G 12i = N14 [H (ξi )]3 d̄ 22 − N12 d̄ 12 6H (ξ )[H (ξi )]2 + 3[H (ξi )]2 H (ξi )
Note that the derivatives H (ξi ) and H (ξi ) are also calculated analytically. By
subdividing the shell length domain (−L/2, L/2) into M equal segments, we apply
Equations (2.122) and (2.123) to each nodal point. Here we consider the case of simply
supported boundary conditions, namely

w̄0 = w̄0 = f¯ 0 = f¯ 0 = w̄ M = w̄M = f¯ M = f¯ M = 0 (2.125)
or in view of (2.121)
w̄−1 = w̄1 , f¯ −1 = f¯ 1 , w̄ M+1 = w̄ M−1 , f¯ M+1 = f¯ M−1 (2.126)
Thus, we establish a system of simultaneous algebraic equations,
[C(ξi , λ)](2M+2)×(2M+2) {δ}(2M+2)×(2M+2) = 0 (2.127)
where [C(ξi , λ)] is the coefficient matrix, whose elements depend on the shell geometry,
nodal point coordinates, elastic constants as well as the bucking load reduction factor
λ; {δ} represents a column vector containing the sought values of functions of w and
f at the nodal points. Equation (2.127), the stability equation in finite-difference form,
is a set of linear, homogeneous, algebraic equations. There exist non-trivial solutions
of this set of equations for some values of load parameters λ. The lowest eigenvalue λ
represents the classical buckling load reduction factor. Its value is determined through
an iteration procedure. Usually, the load parameter λ is increased step by step, and at
each step the determinant of matrix [C(ξ, λ)] is calculated in order to find the interval
where its sign first changes. Once the interval in which the determinant of coefficient
matrix changes sign is located, the secant method is used to expedite the search for
the classical buckling reduction factor λ, the accuracy of which is improved with the
increase in the number of subdivided segments. The numerical results obtained by the
finite difference method will be discussed later.
In many circumstances, the composite shell buckles axisymmetrically. For the cases
where the axisymmetric buckling mode dominates, it is possible to derive an asymptotic
formula relating the buckling load reduction to the thickness variation parameter. This
could be accomplished in the following fashion. We assume w̄(ξ ) in the form
pL 3 pL
w̄(ξ ) = A cos ξ + B cos ξ (2.128)
R R
where p is the number of half-waves along the shell length at buckling; A and B
are undetermined constants. The preceding buckling pattern satisfies the boundary
conditions of the simple supports if p is of the form
R
p = (2k + 1)π (2.129)
L
where k is an integer. The first term of the two-term approximation (1.128) is the exact
buckling mode for the shell of constant thickness, and the second term is introduced to
account for the thickness variation.
To solve the compatibility equation (1.129) for f¯ , the perturbation procedure is
employed. To this end, f¯ is expressed in terms of the thickness variation parameter
as follows:
f¯ (ξ ) = f 0 (ξ ) + f 1 (ξ ) + 2 f 2 (ξ ) + · · · (2.130)
Substituting (1.130) into (1.129), we have after collecting the like terms in ,
∗ ∗ ∗
∗ (4)
)N 2 f 0 + ā11
∗
f 0 − z 2 w̄ + ā22 ∗ ∗
)N 2 f 1
(4)
ā22 f 0 − (2ā12 + ā66 f 1 − (2ā12 + ā66
∗ ∗ ∗
P sin(2Pξ ) f 0
(4)
+ ā11 N 4 f 1 − 2ā22 cos(2Pξ ) f 0 − 4ā22
∗ ∗
+ (4ā12 + 2ā66 )N 2 cos(2Pξ ) f 0 − 4ā22
∗
P 2 cos(2Pξ ) f 0
∗ ∗
+ (4ā12 + 2ā66 )P N 2 sin(2Pξ ) f 0 − (2ā11
∗ ∗
N 4 − 4ā12 P 2 N 2 ) cos (2Pξ ) f 0

∗ (4)
+ 3z 2 cos(2Pξ )w̄ + 2 ā22 f 2 − (2ā12 ∗ ∗
N 2 + ā66 N 2 ) f 2 + ā11
∗
N 4 f2
∗ ∗ ∗
P cos(2Pξ ) sin(2Pξ ) f 0
(4) (4)
+ ā22 cos2 (2Pξ ) f 0 − 2ā22 cos(2Pξ ) f 1 + 4ā22
∗ ∗ ∗
+ (4ā12 N 2 + 2ā66 N 2 − ā22 P 2 ) sin(2Pξ ) f 1 − (2ā12
∗ ∗
N 2 + 2ā66 ∗
N 2 + 4ā22 P 2)
× cos2 (2Pξ ) f 0 + 8ā22
∗
P 2 sin2 (2Pξ ) f 0 + (4ā12
∗ ∗
+ 2ā66 )P N 2 sin(2Pξ ) f 1
∗ ∗
− (4ā12 + 2ā66 )N 2 P cos(2Pξ ) sin(2Pξ ) f 0 + (ā11
∗
N 4 cos2 (2Pξ )
∗ ∗ ∗
+ 8ā12 P 2 N 2 sin2 (2Pξ ) + 4ā12 P 2 N 2 cos2 (2Pξ ) f 0 − (2ā11 N4
∗

+ 4ā12 P 2 N 2 ) cos(2Pξ ) f 1 − 3z 2 cos2 (2Pξ )w̄ + · · · = 0 (2.131)

where
pL nL L
P= , N1 = , z=√ (2.132)
R R Rh 0
Equation (2.131) must hold for any value of the parameter , which means that
the factor for each of the powers of must be zero. Thus, Equation (2.131) may be
separated into the following system of equations:
L ( f 0 ) = z 2 w̄ (2.133)
∗ ∗
P sin(2Pξ ) f 0
(4)
L ( f 1 ) = 2ā22 cos(2Pξ ) f 0 + 4ā22
∗ ∗
− (4ā12 + 2ā66 )N 2 cos(2Pξ ) f 0
∗
+ 4ā22 P 2 cos(2Pξ ) f 0 − (4ā12
∗ ∗
+ 2ā66 )P N 2 sin(2Pξ ) f 0
∗ ∗
+ (2ā11 N 4 − 4ā12 P 2 N 2 ) cos(2Pξ ) f 0 − 3z 2 cos(2Pξ )w̄ (2.134)
∗ ∗ ∗
cos(2Pξ ) sin(2Pξ ) f 0
(4) (4)
L ( f2) = −ā22 2
cos (2Pξ ) f0 + 2ā22 cos(2Pξ ) f1 − 4ā22 P
∗ ∗ ∗
− (4ā12 N 2 + 2ā66 N 2 − ā22 P 2 ) sin(2Pξ ) f 1 + (2ā12
∗ ∗
N 2 + 2ā66 N2
∗
+ 4ā22 P 2 ) cos2 (2Pξ ) f 0 − 8ā22
∗
P 2 sin2 (2Pξ ) f 0 − (4ā12
∗ ∗
+ 2ā66 )P N 2
× sin(2Pξ ) f 1 + (4ā12
∗ ∗
+ 2ā66 )N 2 P cos(2Pξ ) sin(2Pξ ) f 0
∗ ∗ ∗
− (ā11 N 4 cos2 (2Pξ ) + 8ā12 P 2 N 2 sin2 (2Pξ ) + 4ā12 P 2 N 2 cos2 (2Pξ ) f 0
∗ ∗
+ (2ā11 N 4 + 4ā12 P 2 N 2 ) cos(2Pξ ) f 1 + 3z 2 cos2 (2Pξ )w̄ (2.135)
where the operator L (·) is defined as
∗ ∗ ∗
L ( f ) = ā22 f (4) − (2ā12 + ā66 )N 2 f + ā11
∗
N4 f (2.136)
Equations (2.133)–(2.135) are solved analytically with the aid of the computerized
symbolic algebra Mathematica (Wolfram, 1991) for f 0 , f 1 , and f 2 to yield
f 0 = a1 cos(Pξ ) + a2 cos(3Pξ )
f 1 = a3 cos(Pξ ) + a4 cos(3Pξ ) + a5 cos(5Pξ ) (2.137)
f 2 = a6 cos(Pξ ) + a7 cos(3Pξ ) + a8 cos(5Pξ ) + a9 cos(7Pξ )
where a1 , a2 , . . . , a9 are coefficients depending on A and B, and are given in the article
by Koiter et al. (1994b).
Applying the weighted residuals method, namely, in our case the Boobnov-Galerkin
procedure, to the equilibrium equation (2.120), we arrive at

1/2
∗ dw̄ 4
2 ∗ ∗ d w̄
2
4 ∗
H 3
d̄ 11 − 2N (d̄ 12 + 2d̄ 66 ) 2 + N d̄ 22 w̄
−1/2 dξ 4 dξ

∗ d w̄ 2 ∗ d w̄ 2 ∗ d w̄
3
2dH
+ 3H 2d̄ 11 3 − 2N d̄ 12 − 4N d̄ 66
dξ dξ dξ dξ

2d H
2
dH 2 ∗ d w̄
2
2 ∗ d 2 w̄
+ 3H + 6H d̄ 11 − N d̄ 12 w̄ + λ
dξ 2 dξ dξ 2 dξ 2

+ z 2 ( f 0 + f 1 + 2 f 2 + · · ·) ψ j (ξ ) dξ = 0 (2.138)
where ψ j (ξ ) is either cos(Pξ ) or cos(3Pξ ).

We now investigate the axisymmetric buckling mode,
n = 0, p = 2 pcl (2.139)
that is, the modal number of thickness variation is twice as much as that of the classical
buckling mode, which is the case where the thickness variation has the most detrimental
effect on the buckling behavior of the isotropic shells. For the composite shells discussed
in this paper, the aforementioned case is also the most critical one, as is shown by the
numerical results in Figures 2.4, 2.5, and 2.6. Thus, in the subsequent analysis, attention
will be devoted to this case to establish an asymptotic relationship between the buckling
load reduction factor λ and the thickness variation parameter . Substituting (2.128)
and (2.137) into (2.138) and making some algebraic manipulations lead, when retaining
the terms up to 2 , to the following eigenvalue problem:

A
[C(, λ)]2×2 =0 (2.140)
B
where [C(, λ)] is the coefficient matrix containing the thickness variation parameter
and the buckling load reduction factor λ. The elements of matrix [C(, λ)] read

3 2
C11 = P 1 − λ − +
4
4

27 2
C12 = C21 = P −7 +
4
(2.141)
8

243 2
C22 = P 41 − 9λ +
4

4
and the characteristic equation C11 C22 − C12
2
= 0 reads

135 2 85
9λ2 + λ −50 + 9 − + 41 − 41 + 2 = 0 (2.142)
2 2
Thus the following asymptotic expression for the buckling load reduction factor is
obtained,
25 2
λ=1−− (2.143)
32
which is identical to the formula (2.30) in the isotropic case.
In this section, six-layer cylindrical shells with different thickness variation have
been investigated. The radius R and the length L are fixed at 6 in. and 30 in., respec-
tively. The nominal thickness of each lamina is taken to be equal to 0.012 in. The
non-dimensional parameter varies from zero to 0.2. Three kinds of materials are
considered; the material moduli are as follows:
1. Glass/epoxy:
E 11 = 7.5 × 106 psi, E 22 = 3.5 × 106 psi

ν12 = 0.25, G 12 = 1.25 × 106 psi
2. Graphite/epoxy:
E 11 = 20 × 106 psi, E 22 = 1 × 106 psi

ν12 = 0.25, G 12 = 0.6 × 106 psi
3. Boron/epoxy:
E 11 = 40 × 106 psi, E 22 = 4.5 × 106 psi

ν12 = 0.25, G 12 = 1.5 × 106 psi
The laminate configuration is chosen as [θ/–θ/θ ]sym , with θ ranging from 0◦ to 90◦ to
show the interaction between the thickness variation and the fiber orientation.
(0)
Using (2.117), Pcl , the classical buckling load for the shell of constant thickness
h 0 , can be calculated. It is seen from Figure 2.9 that in a certain fiber-angle range (e.g.,
θ between 21◦ and 69◦ for glass/epoxy, θ between 14◦ and 76◦ for graphite/epoxy, and θ
Figure 2.9 Classical buckling load Pcl(0) for shells of three different materials.
between 18◦ and 72◦ for boron/epoxy), the buckling mode is axisymmetric. Numerical
results show that even though the thickness variation generally reduces the load-bearing
capability of the structure, the magnitude of the reduction can vary enormously with
such factors as constituent materials, fiber orientation, and thickness variation pattern.
Figures 2.10, 2.11, and 2.12 illustrate that, when the thickness variation pattern is con-
figurational to the classical axisymmetric buckling mode, the buckling load reduction
Figure 2.10 Buckling load parameter λ versus fiber angle θ for glass/epoxy shells
([θ/– θ/θ ]sym , p = pcl ).
Figure 2.11 Buckling load parameter λ versus fiber angle θ for graphite/epoxy
shells ([θ/– θ/θ]sym , p = pcl ).
is around 5–8% for = 0.1 and around 8–12% for = 0.2, the specific value varying
in terms of the constituent materials and the lamination profile. Interestingly enough,
in the cases where the axisymmetric buckling mode dominates, the buckling load re-
duction rate λ does not vary significantly in terms of the buckling pattern, whether
it is axisymmetric or non-axisymmetric. From Figures 2.13, 2.14, and 2.15, the most
detrimental effect of the thickness variation has been found to occur when the wave
Figure 2.12 Buckling load parameter λ versus fiber angle θ for boron/epoxy shells
([θ/– θ/θ ]sym , p = pcl ).
Figure 2.13 Buckling load parameter λ versus thickness variation parameter p (material:
glass/epoxy; laminate profile: [45◦ /–45◦ /45◦ ]sym , = 0.1).
number of the axisymmetric thickness variation is twice that of the classical buck-
ling mode. Remarkably, analogous phenomena has been analytically predicted in the
isotropic case, where an asymptotic formula for the dependence of classical buckling
reduction factor λ on the thickness variation is given in Equation (1.30) as
25
λ = 1 − − 2
36
graphite/epoxy; laminate profile: [45◦ /–45◦ /45◦ ]sym , = 0.1).
boron/epoxy; laminate profile: [45◦ /–45◦ /45◦ ]sym , = 0.1).
However, compared with the isotropic shells, the effect of thickness variation on the
buckling load is more complicated for the composite shells. From Figures 2.15, 2.16,
2.17, and 2.18, it is seen that the buckling load reduction is more remarkable in the
axisymmetric buckling mode than in the asymmetric mode. For instance, if the material
is glass/epoxy, the shell with the lamination profile [θ/−θ/θ]sym buckles axisymmetri-
cally when the fiber angle θ varies between 21◦ and 69◦ . In these cases of axisymmetric
Figure 2.16 Buckling load parameter λ versus fiber angle θ for glass/epoxy shells
(laminate profile: [45◦ /–45◦ /45◦ ]sym , p = 2 pcl ).
Figure 2.17 Buckling load parameter λ versus fiber angle θ for graphite/epoxy
shells (laminate profile: [45◦ /–45◦ /45◦ ]sym , p = 2 pcl ).
buckling mode, the thickness variation of magnitude = 0.1 brings about 10% buck-
ling load reduction, whereas the buckling load reduction rate could reach 22% when
is 0.2. However, the reduction rate λ remains almost the same in the entire range
of axisymmetric buckling mode, regardless of what the constituent materials or the
laminate fiber angles are, although these factors may affect remarkably the classical
buckling load. Figures 2.17, 2.18, 2.19, and 2.20 illustrate the pattern of change of
Figure 2.18 Buckling load parameter λ versus fiber angle θ for boron/epoxy shells
(laminate profile: [45◦ /–45◦ /45◦ ]sym , p = 2 pcl ).
Figure 2.19 Effect of thickness variation and imperfection on the buckling load (lami-
nate configuration [θ/– θ/θ/– θ/θ ]sym , thickness variation parameter = 0.1).
buckling load reduction factor λ with the fiber orientation when the shell contains the
critical thickness variation.
As has been reported (see Khot, 1970a, 1970b), composite shells are less
imperfection-sensitive than metallic shells. Nevertheless, the present investigation indi-
cates that composite shells are as sensitive to the thickness variation as metallic shells.
Figure 2.20 Effect of thickness variation and imperfection on the buckling load (lami-
nate configuration [θ/– θ/θ/– θ/θ ]sym , thickness variation parameter = 0.2).
2.4 BUCKLING OF COMPOSITE CYLINDRICAL SHELLS 89
This makes the study of the effect of thickness variation more important for composite
shells because they are more vulnerable to fabrication imprecision, which may directly
lead to uneven wall thickness of the structure.
2.4 Effect of the Thickness Variation and Initial Imperfection on Buckling

of Composite Cylindrical Shells
As in Li et al. (1997), this section aims at the combined effect of thickness variation
and intial imperfection on the buckling behavior of the composite shells. We approach
this problem by using Koiter’s (1945, 1966b, 1980) energy criterion of elastic stabil-
ity, as we have done in the isotropic case. Here, we consider the small axisymmetric
thickness variation, and as a first approximation, only the terms up to the first order
of thickness variation parameter are retained. The final product of this discussion is
again an asymptotic formula that relates the thickness variation parameter and initial
imperfection amplitude to the buckling load of the structure. Thus, the present analysis
is actually a direct generalization and extension of our former investigation (Koiter
et al., 1994a, 1994b) to the anisotropic case.
Membrane strain energy of a laminated cylindrical shell of length L is
2π R L
1
Um = (N x x + N y y + N x y γx y ) d xd y (2.144)
2 0 0
Bending strain energy reads

2π R L
1
Ub = (Mx κx + M y κ y + Mx y κx y ) d xd y (2.145)
2 0 0
For the shell under axial uniform end compression N0 , potential energy of the
applied load takes the form
2
1 2π R L
∂w ∂w0
=− N0 + d xd y (2.146)
2 0 0 ∂x ∂x
where w0 is the geometric initial imperfection.

Thus, the total potential energy reads
! = Um + Ub + (2.147)
or, using the constitutive relations (2.99) and (2.100),

2π R L
1
!= A11 x2 + 2A12 x y + 2A16 x γx y + 2A26 y γx y + A22 y2 + A66 γx2y
2 0 0
+ D11 κx2
+ 2D12 κx κ y + 2D16 κx κx y + 2D26 κx κx y + D22 κ y2

∂w ∂w0 2
+ D66 κx y − N0
2
+ d xd y (2.148)
∂x ∂x
Substitution of Equation (2.92) into Equation (2.148) leads to the energy expression
in terms of displacements

1 2π R L
∂u 1 ∂w 2 2 ∂u 1 ∂w 2
!= + A11 + 2A12 +
2 0 0 ∂x 2 ∂x ∂x 2 ∂x
2 2
∂v w 1 ∂w ∂u 1 ∂w ∂u ∂v ∂w ∂w
× + + + 2A16 + + +
∂y R 2 ∂y ∂x 2 ∂x ∂y ∂x ∂x ∂y
2
∂v w 1 ∂w ∂u ∂v ∂w ∂w
+ 2A26 + + + +
∂y R 2 ∂y ∂y ∂x ∂x ∂y
2 2
∂v w 1 ∂w ∂u ∂v ∂w ∂w 2
+ A22 + + + A66 + +
∂x R 2 ∂y ∂y ∂x ∂x ∂y
2 2
∂ w ∂ 2w ∂ 2w ∂ 2w ∂ 2w ∂ 2w ∂ 2w
+ D11 + 2D12 2 + 4D16 2 + 4D26 2
∂x 2 ∂x ∂y 2 ∂ y ∂ x∂ y ∂ y ∂ x∂ y
2 2 2 2 2
∂ w ∂ w ∂w ∂w0
+ D22 + 4D66 − N0 + d xd y (2.149)
∂y 2 ∂ x∂ y ∂x ∂x
For the use of Koiter’s energy criterion of elastic stability, variations of energy are
performed at the fundamental (prebuckling) state.
The second variation of the energy for buckling modes is
2
1 2π R L
∂u ∂u ∂v w ∂u ∂u ∂v
P2 [u] = A11 + 2A12 + + 2A16 +
2 0 0 ∂x ∂x ∂y R ∂x ∂y ∂x

∂v w ∂u ∂v ∂ν w 2 ∂u ∂v 2
+ 2A26 + + + A22 + + A66 +
∂y R ∂y ∂x ∂y R ∂y ∂x
2 2
∂ w ∂ 2w ∂ 2w ∂ 2w ∂ 2w ∂ 2w 2
+ D11 + 2D 12 + 4D 16 + D 22
∂x2 ∂ x 2 ∂ y2 ∂ x 2 ∂ x∂ y ∂ y2
2 2
∂ 2w ∂ 2w ∂ w ∂w 2
+ 4D26 2 + 4D66 − N0 dx dy (2.150)
∂ y ∂ x∂ y ∂ x∂ y ∂x
We will discuss the effect of the most critical type of axisymmetric geometrical
imperfection w0 (x) = −µh 0 cos(2 px/R) (Koiter, 1963; Tennyson et al., 1971), where
h 0 is the nominal thickness of the shell, µ is the non-dimensional parameter describing
the magnitude of the imperfection, and p is the wave number of the axisymmetric
classical buckling mode, which has the following expression (Tennyson et al., 1971),
R
p2 = ∗ ∗ (2.151)
4 a22 d11
We supplement the second variation with the additional bilinear term due to the
geometric initial imperfection

2π R L
∂w dw0
P11 [u 0 , u] = −N0 dx dy (2.152)
0 0 ∂x dx
The third variation of the energy reads
2π R L

1 ∂u ∂w 2 ∂u ∂w 2 ∂w 2 ∂u w
P3 [u] = A11+ A12 + +
2 0 0 ∂x ∂x ∂x ∂y ∂x ∂y R

∂u ∂w ∂w 1 ∂w 2 ∂u ∂v
+ 2A16 + +
∂x ∂x ∂y 2 ∂x ∂y ∂x
2
∂v w ∂w ∂w 1 ∂w ∂u ∂v
+ 2A26 + + +
∂y R ∂x ∂y 2 ∂y ∂y ∂x
2
∂v w ∂w ∂u ∂v ∂w ∂w
+ A22 + + 2A66 + dx dy
∂y R ∂y ∂y ∂x ∂x ∂y
(2.153)
We now assume that the buckling modes of the shell with a uniform thickness
remain a good approximation for the buckling modes of the shell with small thickness
variations. We are, again as in the isotropic case, ensured that the critical load obtained
in this way is by the energy principle an upper bound for the actual critical buckling
load.
According to the study of Tennyson et al. (1971), the following buckling mode
expression can be adopted for the laminated cylindrical shell with the aforementioned
axisymmetric initial imperfection w0 :
2 px px ny
w = b cos + Cn cos cos (2.154)
R R R
where b and Cn are constants, n is the number of waves in the circumferential direction
of the shell. If we recall the shell equilibrium equations in terms of displacements u, v,
and w (Vinson Sierakowski, 1986)
 
    0 
L 13   

L 11 L 12 u  
 
  0
 L 12 L 22 L 23  v = (2.155)

   
 ∂ 2w 

L 13 L 23 L 33 w 
 N0 

∂x 2
where operators L i j are
∂ ∂2 ∂2
L 11 = a11 + 2a 16 + a 66 ,
∂x2 ∂ x∂ y ∂ y2
∂2 ∂2 ∂2
L 12 = a16 + (a 12 + a 66 ) + a 26
∂x2 ∂ x∂ y ∂ y2

1 ∂ ∂ ∂2 ∂2 ∂2
L 13 = a12 + a26 , L 22 = a66 2 + 2a26 + a22 2 ,
R ∂x ∂y ∂x ∂ x∂ y ∂y
(2.156)
1 ∂ ∂
L 23 = a26 + a22
R ∂x ∂y
a22 ∂4 ∂4 ∂4
L 33 = + d 11 + 4d 16 + 2(d12 + 2d 66 )
R2 ∂x4 ∂ x 3∂ y ∂ x 2∂ y2
∂4 ∂4
+ 4d26 + d 22
∂ x∂ y 3 ∂ y4
We can obtain the expressions for u and v as follows:
a12 2 px px ny
u=− b sin + Q n Cn sin cos
2 pa11 R R R
(2.157)
px ny
v = K n Cn cos sin
R R
where

n p 2 a11 a22 + n 2 a66 a22 − a12 p − a66 a12 p 2
2 2
Kn = −
(a11 p 2 + a66 n 2 )(a66 p 2 + a22 n 2 ) − (a12 + a66 )2 p 2 n 2
(2.158)
pa66 ( p 2 a12 − n 2 a22 )
Qn = −
(a11 p 2 + a66 n 2 )(a66 p 2 + a22 n 2 ) − (a12 + a66 )2 p 2 n 2
It should be mentioned that in deriving solution (2.157), we again used an as-
sumption in the studies of Tasi (1966) and Hirano (1979) that the coupling stiffnesses
A16 , A26 , D16 , and D26 can be approximately set to zero.
In the numerical analysis of composite shells with thickness variation, we have
deduced that, in the absence of the geometric imperfection, the thickness variation has
the most degrading effect on the buckling load when the wave number of the thickness
variation is twice that of the classical buckling mode, that is p1 = 2 p. This result is
also observed in the isotropic shell case (Koiter et al., 1994a, 1994b). Now we are
interested in the combined effect of the most critical geometric imperfection and the
most detrimental thickness variation on the reduction of the buckling load.
Substituting Equations (2.154) and (2.157) into the second and third variations, we
obtain, after retaining only the first-order terms in ,
Cn2 π L
P2 [u] = d22 n 4 + 2d12 n 2 p 2 + 4d66 n 2 p 2 + d11 p 4 + a22 (1 + K n n)2 R 2
4R 3
− N0 p 2 R 2 + 2a12 (1 − K n n)Q n p R 2 + a11 Q 2n p 2 R 2

1 2
3
+ a66 (K n p + n Q n ) R +
2 2
b 16d11 p 1 −
4
2R 3 2
2

a12 R2 1 1
− 1 − + a22 R 1 −
2
− 4N0 R p
2 2
(2.159)
a11 2 2
4bh 0 N0 p 2 µπ L
P11 [u 0 , u] = (2.160)
R

bCn π L
2
1 1
P3 [u] = a12 −1 + p + 4 1 + (1 + K n n) p
2 2
8R 2 2 2
2 2

1 a12 n 1 1
+ a22 1 − n + 2
−1 + + a12 1 − p 2
2 a11 2 2

1 1
+ 4a11 p Q n 1 + − 4a66 1 + np(K n p + Q n n)
3
(2.161)
2 2
Thus, the energy expression to be considered is
P2 [u] + P11 [u 0 , u] + P3 [u] (2.162)
The equations for the initial post-buckling behavior are furnished by taking the
partial derivatives of the energy expression with respect to b and Cn to be zero:
2

b 3 a12 R2 1 1
16d11 p 1 − −
4
1 − + a22 R 1 − − 4N0 R p
2 2 2
R3 2 a11 2 2

4N0 h 0 p 2 µ Cn2 1 1
+ + a12 −1 + p + 4 1 + (1 + K n n) p
2 2
R 8R 2 2 2
2 2

1 a12 n 1 1
+ a22 1 − n + 2
−1 + + a12 1 − p 2
2 a11 2 2

1 1
+ 4a11 p 3 Q n 1 + − 4a66 1 + np(K n p + Q n n) = 0 (2.163)
2 2
and
Cn
[d22 n 4 + 2d12 n 2 p 2 + 4d66 n 2 p 2 + d11 p 4 + a22 (1 + K n n)2 R 2
2R 3
− N0 p 2 R 2 + 2a12 (1 + K n n)Q n p R 2 + a11 Q 2n p 2 R 2 + a66 (K n p + n Q n )2 R 2 ]

bCn 1 1 1
+ a 12 −1 + p 2
+ 4 1 + (1 + K n n) p 2
+ a 22 1 − n2
4R 2 2 2 2
2 2

a12 n 1 1 1
+ −1 + + a12 1 − p + 4a11 p Q n 1 +
2 3
a11 2 2 2

1
− 4a66 1 + np(K n p + Q n n) = 0 (2.164)
2
The solution Cn = 0 of Equation (2.164) leads to
4N0 h 0 p 2 µR 2
b=− 2
a12 R2 (2.165)
16d11 p 4 1 − 32 − a11
1 − 12 + a22 R 2 1 − 12 − 4N0 R 2 p 2
from Equation (2.163), and bifurcation buckling with respect to the asymmetric mode
with amplitude Cn occurs at
2
b = − [d22 n 4 + 2d12 n 2 p 2 + 4d66 n 2 p 2 + d11 p 4 + a22 (1 + K n n)2 R 2 − N0 p 2 R 2
R
+ 2a12 (1 + K n n)Q n p R 2 + a11 Q 2n p 2 R 2 + a66 (K n p + n Q n )2 R 2 ]

1 1 1
÷ a12 −1 + p + 4 1 + (1 + K n n) p + a22 1 − n 2
2 2
2 2 2

a2 n2 1 1 1
+ 12 −1 + + a12 1 − p 2 + 4a11 p 3 Q n 1 +
a11 2 2 2

1
− 4a66 1 + np(K n p + Q n n) (2.166)
2
Equating expressions (2.165) and (2.166), we obtain the equation for the critical
buckling load N0
2

3 a12 R2 1 1
16d11 p 4
1− − 1 − + a22 R 1 − − 4N0 R p
2 2 2
2 a11 2 2
× [d22 n 4 + 2d12 n 2 p 2 + 4d66 n 2 p 2 + d11 p 4 + a22 (1 + K n n)2 R 2
− N0 p 2 R 2 + 2a12 (1 + K n n)Q n p R 2 + a11 Q 2n p 2 R 2 + a66 (K n p + n Q n )2 R 2 ]

1 1
− 2N0 h 0 p 2 µR 3 a12 −1 + p 2 + 4 1 + (1 + K n n) p 2
2 2
2 2

1 a12 n 1 1
+ a22 1 − n + 2
−1 + + a12 1 − p 2
2 a11 2 2

1 1
+ 4a11 p Q n 1 + − 4a66 1 + np(K n p + Q n n) = 0
3
(2.167)
2 2
It should be pointed out that, in solving Equation (2.167), integer search should be
performed with respect to the circumferential wave number n to arrive at the lowest
value of N0 .
We define the non-dimensional critical load parameter λ
N0
λ= (2.168)
Ncl
where Ncl is the classical buckling load in the absence of both initial imperfection and
thickness variation and has the following expression (Vinson and Sierakowski, 1986):
Ncl = min{Nm,n }
m,n
2
L C11 C22 C33 + 2C12 C23 C13 − C13
2
C22 − C23
2
C11 − C12
2
C33
Nm,n =
mπ C11 C22 − C122
(2.169)
where
2 2 2
mπ n n mπ 2
C11 = A11 + A66 , C22 = A22 + A66
L R R L
4 2 2 4
mπ mπ n n A22
C33 = D11 + 2(D12 + 2D66 ) + D22 + 2
L L R R R

mπ n A12 mπ
C12 = (A12 + A66 ) , C13 = ,
L R R L
3
A22 n n
C23 = + B22
R R R
(2.170)
Equation (2.167) is very useful in the sense that the axial buckling load can be de-
termined from it for composite cylindrical shells containing small axisymmetric initial
imperfection and thickness variation. For practical purposes, the results thus obtained
should be considered conservative, since the most detrimental case of geometric im-
perfection and thickness variation is investigated. However, since we ignored in our
derivation the higher order terms in , the results from the present study should not be
deemed accurate for shells having large thickness variation.
As a numerical example, we discuss the shells made of carbon/epoxy laminae,
whose elastic moduli are E 1 = 13.75 × 106 psi, E 2 = 1.03 × 106 psi, ν12 = 0.25,
G 12 = 0.42 × 106 psi. The shell is 6 in. in radius and 30 in. in length and is composed
of ten equally thick layers, each being 0.012-in. thick. The laminate configuration is
[θ/–θ/θ/–θ/θ ]sym , with, the fiber angle θ varying from 0◦ to 90◦ .
Solving Equation (2.167) numerically for the critical load N0 with integer search
performed simultaneously with respect to the circumferential buckling wave number
n, and then non-dimensionalizing the result in virtue of (2.168), we obtain the critical
buckling load factor λ for different cases of thickness variation parameter and im-
perfection amplitude µ. The numerical results here obtained are in agreement with the
previous first-order asymptotic formula λ = 1 − , which holds only for the axisym-
metric buckling cases for composite shells without initial imperfection. It is interest-
ing to note that as long as the axisymmetric buckling mode dominates, the buckling
load reduction factor λ remains practically constant, irrespective of the change in the
laminate construction. However, after the shell involves initial imperfection, the buck-
ling mode becomes entirely non-axisymmetric, and the buckling load reduction is
strongly influenced by the stacking sequence of the laminae. Figure 2.21 depicts the
results of the buckling load factor λ for shells of different laminate profiles, such as
Figure 2.21 Buckling load reduction in shells with different laminate configurations (case 1:
isotropic; case 2: [45◦ /–45◦ /45◦ /–45◦ /45◦ ]sym ; case 3: [16◦ /–16◦ /16◦ /–16◦ /16◦ ]sym ).
[45◦ /−45◦ /45◦ /–45◦ /45◦ ]sym and [16◦ /−16◦ /16◦ /−16◦ /16◦ ]sym , together with the results
for corresponding isotropic shells. It can be seen from this figure that composite shells
are sensitive to thickness variation, and especially to initial imperfection, and that such
a sensitivity to the thickness variation is comparable to that of the isotropic shell. Once
again, we can conclude that, despite the fact that the geometric initial imperfection
remains as a dominant factor for the buckling load reduction, the further degradation
in the load-bearing capacity of the structure due to the effect of thickness variations
should not be overlooked.
In order to investigate the accuracy of the preceding asymptotic formula, an effort
has been made to use BOSOR4 (Bushnell, 1976), a well-known commercial software for
buckling analysis, to generate a set of comparable data for the non-dimensional critical
load parameter λ. Since the classical buckling load has been used to non-dimensionalize
the critical buckling load, it is necessary to check the results from Equation (2.169) with
their counterparts from the numerical software so that a common basis can be estab-
lished for the follow-up comparison of results for non-dimensional critical load λ. For
this purpose, software packages BOSOR4 and PANDA2 (Bushnell, 1983) (using both
the linear and non-linear shell theories) were utilized for the classical buckling load and
numerical results are plotted together with those from Equation (2.169) (Figure 2.22).
Figure 2.22 shows that the classical buckling loads from different sources agree quite
well except for the cases where θ lies between 53◦ and 80◦ . When θ is between 53◦
and 80◦ , Equation (2.169) produces a set of data that are very close to the results from
PANDA2 using the shallow shell theory. However, a discrepancy is found between the
results based on the shallow shell theory and those from BOSOR4 and PANDA2 us-
ing Sanders non-linear shell theory (Sanders, 1963). This indicates that, for those cases
Figure 2.22 Comparison of classical buckling loads stemming from different methods.
where the lamination angle θ ranges from 53◦ to 80◦ , the shallow shell theory appears to
be inadequate, and a more refined, non-linear theory may be necessary for the prediction
of buckling loads. Figure 2.23 displays the results of the non-dimensional critical load
parameter λ obtained from the asymptotic formula [Equation (2.167)] and BOSOR4
for a 10-layer composite shell (laminate configuration: [16◦ /−16◦ /16◦ /−16◦ /16◦ ]sym ),
Figure 2.23 Comparison of the non-dimensional critical load λ obtained from the asymptotic
formula and BOSOR4.
which contains both the initial imperfection and the thickness variation. It can be seen
from this figure that the asymptotic formula predicts the knockdown factor quite accu-
rately.
Finally, it is worthwhile to mention that, as a special case, if we let
Eh 0 1−ν
a11 = a22 = , a12 = νa11 , a66 = a11 , a16 = a26 = 0
1 − ν2 2
Eh 3 1−ν
d11 = d22 = , d12 = νd11 , d66 = d11 , d16 = d26 = 0
12(1 − ν 2 ) 2
(2.171)
where E is the Young’s modulus, and ν is the Poisson’s ratio; furthermore, we select
the asymmetric mode at the top of the Koiter’s semi-circle (Koiter, 1980); that is, let

p = n = [ 3(1 − v 2 )R 2 /2h 0 ]1/2 (2.172)
Equation (2.167) reduces to its counterpart in the isotropic shell case,

3 3(1 − ν 2 ) 1
(1 − λ)(1 − − λ) − λµ 1 + = 0 (2.173)
2 6
which is identical to Equation (2.59), if the small term /6 is ignored as compared with
unity (see also Koiter et al., 1994b).
Figure 2.24 shows the comparison of results in the isotropic shell case using Koiter’s
circle and those using integer search with respect to the circumferential wave number
n, and it is seen that the agreement is excellent.
For further details, consult the study by Li et al. (1997).
Figure 2.24 Comparison of results using Koiter’s semi-circle and those using integer search
for isotropic shells with thickness variation = 0.2).
2.5 EFFECT OF THE DISSIMILARITY IN ELASTIC MODULI ON THE BUCKLING 99
2.5 Effect of the Dissimilarity in Elastic Moduli on the Buckling

In this section, it is demonstrated through a simple example of the Roorda-Koiter frame
that the unavoidable dissimilarity in the distribution of elastic moduli may further reduce
the load-carrying capacity in addition to the well-recognized effect of initial geometric
imperfections.
The works by Koiter (1945, 1963) and Budiansky and Hutchinson (1964) estab-
lished that the initial geometric imperfection – deviation from the nominally ideal
shape – plays a major role in the reduction of the load-carrying capacity of structures.
An additional source of such a reduction was identified recently by Koiter et al. (1994b)
as a non-uniform thickness of the structure. To the best of the authors’ knowledge, there
is only a single work, by Ikeda and Murota (1990a, 1990b), that points out that the sym-
metry breaking in the distribution of elastic moduli may cause additional reduction in
the loads the system is able to sustain. In particular, they analyzed the von Mises truss in
such a context. In this section, we perform analytical and numerical investigation on a
frame structure to study the effect of dissimilarity of elastic moduli on the load-carrying
capacity of the structure.
2.5.1 Analysis
In 1965, Roorda (1965) conducted a set of experiments on the two-bar frame, subjected
to an eccentric load. Koiter (1967) performed an initial-imperfection analysis of this
frame, referred to in the literature since then as the Roorda-Koiter frame. Roorda and
Chilver (1970) and Bažant and Cedolin (1989) analyzed this structure from different
perspectives (for additional references, see also the text of Brush and Almroth, 1975).
Here, the vertical and the horizontal segments of the two-bar frame (Figure 2.25) may
Figure 2.25 Roorda-Koiter frame.

have different lengths L 1 and L 2 and different Young’s moduli E 1 and E 2 . Our aim is
to study the effect of small differences in the elastic moduli between the two segments,
namely, the vertical and the horizontal bars of the frame structure. Such a difference
may trigger the dissimilarity in the elastic moduli of the structure.
Following those previous investigators, we first analyze the perfect structure.
The total potential energy of the two-bar frame is
L1
E 1 A dξ 1 dw 2 2 E 1 I d 2 w 2
!= + + dx
0 2 dx 2 dx 2 dx2
L2
E 2 A dη 1 dv 2 2 E 2 I d 2 v 2
+ + + dy − Pv A (2.174)
0 2 dy 2 dy 2 dy 2
The change in potential energy is obtained by letting
ξ → ξ0 + ξ̄ , η → η0 + η̄, v → v0 + ν̄, w → w0 + w̄ (2.175)
where
Px
η0 = v0 = w0 = 0, ξ0 = − (2.176)
EA
represents an equilibrium state on the primary equilibrium path in the neighborhood of
the bifurcation point A. The energy expression may take the following form:
1 2 1 1
! = δ ! + δ3! + δ4! (2.177)
2! 3! 4!
where

1 2 E 1 A dξ 2 P d w̄ 2 E 1 I d 2 w̄ 2
L1
δ != − + dx
2! 0 2 dx 2 dx 2 dx2
L2
E 2 A d η̄ 2 E 2 I d 2 v̄ 2
+ + dy (2.178)
0 2 dy 2 dy 2
L1
1 3 E 1 A d ξ̄ d w̄ 2 E 2 A L 2 d η̄ d v̄ 2
δ != dx + dy (2.179)
3! 0 2 dx dx 2 0 dy dy
L1
1 4 E 1 A d w̄ 4 E 2 A L 2 d v̄ 4
δ != dx + dy (2.180)
4! 0 8 dx 8 0 dy
correspond to second, third, and fourth variations of the potential energy, respectively.
According to Koiter’s initial post-buckling theory (1945), in the initial post-buckling
range, the incremental displacement components are of the form of the classical buck-
ling mode
ξ̄ = β A ξ1 , η̄ = β A η1 , v̄ = β A ν1 , w̄ = β A w1 (2.181)
where ξ1 , η1 , v1 , and w1 are normalized classical buckling load. β A is the rotation angle
at bifurcation point A and could be viewed as the amplitude parameter.
Substituting (2.181) into the second variation leads to

1 2 1 L1
dξ1 2 dw1 2 E 1 I d 2 w1 2
δ != −P
E1 A + 2
d xβ A2
2! 2 0 d x d x 2 d x
2 2
1 L2 dη1 2 d v1
+ E2 A + E2 I dyβ A2 (2.182)
2 0 dy dy 2
Performing variation on expression (2.182), we obtain, according to the Trefftz

criterion,
L1
1 2 dξ1 d(δξ1 ) dw1 d(δw1 )
δ δ ! = E1 A −P
2! 0 dx dx dx dx
2 2
E 1 I d w1 d (δw1 )
+ d xβ A2
2 dx2 dx2
L2 2 2
dη1 2 d(δη) d v1 d (δv1 )
+ E2 A + E2 I dyβ A2 = 0
0 dy dy dy 2 dy 2
(2.183)
Integration by parts and rearrangement gives
L 1 L 2 L 1
dξ1 dη1 d 3 w1 dw1
EA δξ1 + E2 A δη1 − E1 I +P δw1
dx 0 dy 0 dx3 dx 0
L2
d 3 v1 d 2 w1 d(δw1 ) L 1 d 2 v1 d(δv1 ) L 1
− E2 I δv1 + E1 I + E2 I
dy 3 0 dx2 dx 0 dy 2 dy 0
L1 L2
d 2 ξ1 d 2 η1
− E 1 A 2 δξ1 d x − E 2 A 2 δη1 dy
0 dx 0 dy
L1 L2
d 4 w1 d 2 w1 d 4 v1
+ E1 I + P δw 1 d x + E 2 I δv1 dy = 0 (2.184)
0 dx4 dx2 0 dy 4
The satisfaction of Equation (2.184) results in boundary conditions (using ξ |x=L 1 =

−v| y=L 2 ; η| y=L 2 = w|x=L 2 );
d 2 w1 d 2 v1
=0 at x = 0; =0 at y=0 (2.185)
dx2 dy 2
dξ1 d 3 v1 d 2 w1 d 2 v1
E1 A + E2 I = + =0 at x = L 1, y = L2 (2.186)
dx dy 3 dx2 dy 2
dη1 d 3 v1 dw1
E2 A − E1 I 3
−P = 0; at x = L 1, y = L2 (2.187)
dy dx dx
and governing equations:

d 2 ξ1 d 4 w1 d 2 w1
= 0, E1 I + P =0
dx2 dx4 dx2
(2.188)
d 2 η1 d 4 v1
= 0, =0
dy 2 dy 4
Integration of Equations (2.188) gives
x y
β A ξ1 = C1 , β A η1 = C2
E1 A E2 A (2.189)
β A w1 = C3 sin(kx) + C4 x, β A v1 = C5 y + C6 y 3
where k 2 = P/E 1 I , and Ci (i = 1, 2, . . . , 6) are integration constants.

Using boundary conditions at x = L 1 , y = L 2 :
L2
w = η: C3 sin(k L 1 ) + C4 L 1 = C2 (2.190)
E2 A
L1
ξ = −v: C5 L 2 + C6 L 32 = C1 (2.191)
E1 A
dw
= −β A : kC3 cos(k L 1 ) + C4 = −β A (2.192)
dx
dv
= −β A : C5 + 3c6 L 22 = −β A (2.193)
dy
dξ1 d 3 v1
E1 A + E2 I = 0: C1 + 6E 2 I C6 = 0 (2.194)
dx dy 3
d 2 w1 d 2 v1
2
+ = 0: −k 2 C3 sin(k L 1 ) + 6C6 L 2 = 0 (2.195)
dx dy 2
dη1 d 3 v1 dw1
E2 A − E2 I 3
−P = 0: C2 − P[kC3 cos(k L 1 ) + C4 ] = 0
dy dx dx
(2.196)
Equations (2.190)–(2.196) are linear, homogeneous algebraic equations in terms

of β A , C1 , . . . , C6 . The non-triviality condition leads to
6E 2 I L2
6(k L 1 ) cos(k L 1 ) − 2
(k L 1 )2 sin(k L 1 ) − 6 sin(k L 1 ) − 2(k L 1 )2 sin(k L 1 )
E 1 AL 2 L1
2
P 6E 1 I 2 P L2
+ (k L 1 ) sin(k L 1 ) + 2(k L 1 )
2
sin(k L 1 ) = 0
E 1 A E 1 AL 1 L 2 E2a L 1
(2.197)
Because our discussion is limited to the elastic range, terms containing P/E 1 A,
6E 2 I /E 1 AL 22 , P/E 2 A, and 6E 1 I /E 1 AL 1 L 2 are negligibly small and could be omitted
in the first approximation. Thus, we have an approximate characteristic equation as the
following:
k L1
tan(k L 1 ) ≈ (2.198)
1+ 1
3
(k L 1 )2 LL 21
from which, the critical buckling load parameter kcl can be determined. The normalized
classical buckling mode is found to be
3Pcl
ξ1 = x
E 1 A(kcl L 1 )2
3Pcl
η1 = y
E 1 A(kcl L 2 )2
(2.199)
3 sin(kcl x)
w1 = x − L1
(kcl L 1 )2 sin(kcl L 1 )

1 y2
v1 = y 1 − 2 , P = Pcl + (λ − 1)Pcl
2 L2
For the case L 1 = L 2 = L, the buckling load parameter is found to be kcl L = 3.72.
Substituting the preceding expressions back into the energy expressions, we have

1 2 E2
δ ! = 0.478Pcl L + 0.109 Pcl L − 0.587λ β A2 (2.200)
2! E1
If we assume that there is a dissimilarity in the distribution of elastic moduli (i.e.,
generally E 1 = E 2 )
E 2 = E 1 (1 + ) (2.201)
then Equation (2.201) reads
1 2
δ ! = 0.587(1 + 0.186 − λ)Pcl Lβ A2 (2.202)
2!
The third variation is
1 3
δ ! = 0.149Pcl Lβ A3 (2.203)
3!
The approximate expression for the potential energy increment ! is the sum of the
second and third variations of the energy expressions. For equilibrium, d( !)/dβ A =
0. For β A = 0, we obtain
λ = 1 + 0.186 + 0.381β A (2.204)
or, we can rewrite Equation (2.204) as
P 0.381
λ1 =
=1+ βA; Pcl = Pcl (1 + 0.186) (2.205)
Pcl 1 + 0.186
We now proceed with the analysis of the geometrically imperfect structure. For an
eccentric load applied at a distance φ L to the right of point A, the potential energy
expression must be modified by adding a term (Brush and Almorth, 1975)

φ = −P Lφβ A = −λ1 Pcl Lφβ A (2.206)
and the second variation and the third variation take the form
1 2
δ ! = A2 (1 − λ1 )Pcl Lβ A2
2
(2.207)
1 3
δ ! = A3 (1 − λ)Pcl Lβ A3
3!
where
A2 = 0.587/(1 + 0.186)2 , A3 = 0.149/(1 + 0.186)
Then
1 2 1
! = δ ! + δ 3 ! + φ
2! 3!

= (1 − λ1 )A2 β A2 + A3 β A3 − λ1 φβ A Pcl L (2.208)

For the equilibrium, we have d( !)/dβ A = 0:
3A3 1 φ
λ1 = 1 + βA − λ1 (2.209)
2A2 2A2 β A
3A3 2 1
λ1 β A = β A + βA − φλ1 (2.210)
2A2 2A2
Differentiating both sides of Equation (2.210) with respect to β A , we have
dλ1 3A3 1 dλ1
βA + λ1 = 1 + βA − φ (2.211)
dβ A A2 2A2 dβ A
Setting dλ1 /dβ A equal to zero, we obtain the limit-point load factor λ L for the imperfect
structure
3A3
λL = 1 + βA (2.212)
A2
or
(λ L − 1)A2
βA = (2.213)
3A3
Substituting back into Equation (2.210), we have
(3A3 λ L )1/2
λL = 1 − (−φ)1/2 (2.214)
A2
As a first approximation, we can set λ L = 1 at the right-hand side of Equation (2.214),
and we arrive at
(3A3 )1/2 (3A3 )1/2
λL ≈ 1 − (−φ)1/2 = 1 − a(−φ)1/2 , a= (2.215)
A2 A2
Figure 2.26 Buckling load reduction (L 2 /L 1 = 1.0).
where a is a parameter describing the imperfection sensitivity of the structure. For

= 0, a = 1.15; = 0.1, a = 1.17; = −0.1, a = 1.11. Using these data, we can
plot the relationship between the buckling load and the initial imperfection parameter.
This is depicted in Figure 2.26, where the three curves correspond to three different
cases of elastic moduli.
For the case where L 2 /L 1 = 0.5, the buckling load parameter obtained from Equa-
tion (2.198) is (kcl L 1 ) = 3.97. Following the same procedure as before, we again end
up with Equation (42). However, A2 and A3 now have the following expressions:
0.37 0.09
A2 = , A3 = (2.216)
(1 + 0.51)2 1 + 0.51
For = 0, a = 1.40; = 0.1, a = 1.51; = −0.1, a = 1.30. Having these, we
can plot the curves of buckling load reduction as in Figure 2.26 and 2.27.
2.5.2 Discussion
Figures 2.26 and 2.27 demonstrate the usual pattern of the buckling load reduction due
to initial imperfection. However, what is worthwhile to notice here is a further reduction
in buckling load because of the presence of dissimilarity in the elastic moduli and the
geometric configuration. When is positive, the horizontal bar possesses a bigger elastic
modulus than the vertical bar, which results in a stiffer structure. Likewise, when L 2 /L 1
is less than unity (the horizontal bar is shorter), the overall stiffness of the structure
is again increased. On the one hand, a stiffer structure has a larger buckling load.
Figures 2.26 and 2.27 show that the stiffer the structure is, the more sensitive it is to
Figure 2.27 Buckling load reduction (L 2 /L 1 = 0.5).
the initial imperfection. As we can see also from those two figures, when the stiffness
is decreased either through a reduced elastic modulus ( < 0) or from an increased
length (L 2 /L 1 > 1), the structure becomes less sensitive to the initial imperfection.
The change in the overall stiffness of the structure comes, as is shown here, from the
dissimilarity. Thus, one may conclude that the dissimilarity in elastic moduli could
contribute to the change in sensitivity of the structure to the initial imperfection. One
may intentionally introduce such a dissimilarity in elastic moduli so that the structure
has a decreased sensitivity to initial imperfections. In this simple two-bar structure, the
decrease in sensitivity due to dissimilarity in elastic modulus is not remarkable when
the two segments of the frame have the same length, namely L 1 = L 2 . However, for
other geometric configurations, the effect may become more pronounced. For L 1 = 2L 2
(Figure 2.27), say, at φ = 0.25, λ = 0.3 for = 0, and λ = 0.24 for = 0.1, which
indicates a 20% decrease in the limit load; for = −0.1, λ = 0.3, which amounts to a
17% increase in load-carrying capacity. It appears that the subject is worth pursuing in
the direction of shell structures to see how important the effect of non-homogeneity of
elastic moduli is on the imperfection sensitivity.
Recently, Combescure et al. (2000) and Gusic et al. (2000) extended our results
to the buckling of cylindrical shells under external pressure, under non-axisymmetric
thickness imperfections. They proposed an analytical formula showing that the pressure
load reduction is a linear function of thickness imperfection. The reduction factor due
to the coupling of two types of imperfection (thickness imperfection and geometric
imperfection) is represented as a product of each mode of reduction. The comparison
with experimental results was conducted.
CHAPTER THREE
Stochastic Buckling of Structures:

Monte Carlo Method
Most calculations of imperfection-sensitivity have been carried out for simple

shapes of the imperfection distribution, selected for convenience of analysis. It is
generally realized, of course, that actual imperfections are unlikely to follow this
regular pattern.
W. T. Koiter
There is a close connection between the concepts of stability and probability.

Stable states of equilibrium or motion observed in the natural or engineering
systems are the most probable ones; unstable ones are improbable and even unre-
alizable. The more stable a state is, the greater is the probability of its realization.
Hence follows the connection between the concepts of stability and reliability.
V. V. Bolotin
But to us, probability is the very guide of life.

J. Butler
In this chapter, we treat initial geometric imperfections as spacewise random functions

(i.e., random fields). As a result, the buckling load – the maximum load the structure can
sustain – turns out to be a random. We focus our attention on reliability of the shells, namely
the probability that a structure will not fail prior to the specified load. We develop efficient
techniques of simulation of random fields, based on the knowledge of the mean initial
imperfection function and the auto-correlation function. This allows us to conduct, by the
Monte Carlo method, an extensive analysis of the buckling of columns on non-linear elastic
foundations and of circular cylindrical shells. We consider the cases of both axisymmetric
and general, non-symmetric imperfections.

It is now generally recognized that initial geometric imperfections play a dominant
role in reducing the buckling load of certain structures. As is well known, an axially
compressed thin shell is highly imperfection sensitive in this context (see, for exam-
ple, Hutchinson and Koiter, 1970; Tvergaard, 1976; and Budiansky and Hutchinson,
1979).
107
108 STOCHASTIC BUCKLING OF STRUCTURES: MONTE CARLO METHOD
This conclusion is mainly due to the work of a series of investigators, among

them Koiter (1945), Donnell and Wan (1950), and Budiansky and Hutchinson (1966),
who arrived at it through recourse to specialized imperfections. However, despite the
accepted theoretical explanation of the buckling behavior of these structures, the use
of the concept of imperfection sensitivity in engineering practice is still in the ad hoc
stage, and engineers prefer to rely on the “knockdown factor” (NASA, 1968), chosen
so that its product with the classical buckling load yields a lower bound to available
experimental data for the configuration in question. This apparent reluctance to take
advantage of theoretical findings stems from the fact that most imperfection studies
are conditional on detailed advance knowledge of the geometric imperfections of the
particular structure, which is rarely possible. In an ideal case, the imperfections can be
measured experimentally in each shell and incorporated in the theoretical analysis to
predict the buckling loads. This approach, however, while justified for single prototype-
like structures, is impracticable as a general method of behavior prediction. Information
on the type and magnitude of imperfections of a particular structure would be too specific
and thus are not strictly valid for other realizations of the same structure even those
obtained by the same manufacturing process.
With these considerations in view and also bearing in mind the scatter of the
experimental results, it appears obvious that practical applications of the imperfection-
sensitivity theories are conditional on their being combined with a statistical analysis of
the imperfections and critical loads. The notion of randomness of the initial imperfec-
tions was given considerable attention in the literature, and a bibliography can be found
in Amazigo’s paper (1976). For the single-mode solutions, consult Bolotin (1969) (who
pioneered the probabilistic approach to buckling problems in 1958) and Roorda (1980).
Later on, Elishakoff (1978b, 1979a, 1980b) suggested utilizing the Monte Carlo
method for the solution of multi-mode problems involving random initial imperfection
sensitivity. This method represents a logical remedy in view of the difficulties inherent in
purely analytical approaches (based on unnecessary and often very restrictive assump-
tions about the properties of the initial imperfections and/or using heavily simplified
solution procedures).
The first step of the Monte Carlo method consists of simulating the random initial
imperfection profiles via a special numerical procedure (Elishakoff, 1979b); the second
step is simply a numerical solution of the buckling problem for every realization of the
initial imperfection profile; and the third and last step involves a statistical analysis
of the buckling loads (for a detailed description of the Monte Carlo method, consult
Elishakoff, 1983b). The reliability is determined as the fraction of an ensemble of shells
of which the buckling loads exceed the specified load.
As far as we know, three works have been devoted to the analysis of cylindrical shells
with general non-symmetric random imperfections. Makarov (1971), at the Moscow
Power Engineering Institute and State University, carried out systematic analysis of
initial imperfections. He used a series of 30 cylindrical shells made of sheets of electrical
grade pressboard. The initial imperfection was represented as a double Fourier series,
and the coefficients were treated as random variables. The analysis showed that the
assumption of circumferential homogeneity of the initial imperfections was satisfactory
(within the confidence limits used in the analysis), and the assumption of Gaussianity
of their Fourier coefficients did not conflict with the experimental data. Makarov also
3.2 RELIABILITY APPROACH TO THE RANDOM IMPERFECTION SENSITIVITY OF COLUMNS 109
carried out a theoretical analysis of the buckling of stochastically imperfect shells, with
the experimental data obtained earlier serving as the input for the description of the
imperfections. He found a theoretical mean buckling load that exceeds its experimental
counterpart by a factor of 1.35.
Fersht (1974) generalized the method of truncated hierarchy, used earlier by
Amazigo (1969) for axisymmetric random imperfections, to include the non-symmetric
case. It turned out that for non-symmetric imperfections a closed-form expression for
the buckling load is unattainable, and rather cumbersome numerical integrations have
to be performed. Moreover, for the axisymmetric case, Fersht’s numerical results do
not agree with those of Amazigo (1969).
Hansen (1977) generalized his previous deterministic results (Hansan, 1975). The
main conclusion was that the imperfection parameters associated with the non-
axisymmetric modes appear only in three separate summations and that the behav-
ior of the system is governed by the values of these summations rather than by the
individual imperfection amplitudes. It was assumed that the Fourier coefficients of the
initial imperfections are jointly normal random variables with zero mean and that they
are statistically independent and identically distributed. Then the Monte Carlo method
was applied. For each sample problem the buckling load was found via the method
described by Hansen (1975), and then the mean buckling loads were calculated. The
role of the non-axisymmetric imperfections turned out to be very important.
In Sections 3.2 and 3.3, we consider the buckling of beams on non-linear elastic
foundations in order to elucidate the Monte Carlo methods as they are applied to
stochastic buckling of structures in the first section. The other two sections (Sections 3.4
and 3.5) are devoted to discussions on cylindrical shells with random axisymmetric and
non-symmetric imperfections.
3.2 Reliability Approach to the Random Imperfection

Sensitivity of Columns
The reliability approach to the random imperfection sensitivity of columns is based on
Elishakoff (1979a, 1983a, 1985).
3.2.1 Motivation for the Reliability Approach

The following considerations are intended as a contribution to understanding the random
imperfection sensitivity of non-linear elastic structures. Specifically, we are dealing with
a stochastically imperfect column on a non-linear mixed quadratic-cubic foundation.
This section is a generalization of an earlier work (Elishakoff, 1979a), where buckling
of a column on a purely cubic foundation was considered.
Deterministic imperfection-sensitivity studies of structures on a non-linear elastic
foundation were performed by Reissner (1970, 1971) and Keener (1974). The first paper
on random imperfection sensitivity of columns under such conditions was published
by Fraser and Budiansky (1969), with reference to infinitely long columns on a soften-
ing (cubic) elastic foundation. They concluded that every column in the ensemble has
the same buckling load, which depends on the auto-correlation function of the initial
imperfections, assumed to be a random ergodic function of the axial coordinate. Finite
columns on a cubic foundation were considered in several studies cited in our earlier
paper (Elishakoff, 1979a), and finite columns on a quadratic-cubic foundation were
considered by Hansen and Roorda (1973, 1974), who assumed that the initial imper-
fection function has the shape of the buckling mode of the associated linear structure.
Earlier, Fraser (1965) considered a single-term approximation for finite columns on
a cubic foundation. The two approaches – consideration of an infinite structure and
a single-term approximation for a finite one – were bridged (Elishakoff, 1979a), the
main contribution of which was a multi-mode solution of the finite structure. In that
study, the initial and the additional deflection functions were expanded in terms of the
buckling modes of the associated perfect linear structure. Fourier coefficients of the
expansion for the initial imperfection function were simulated numerically. For each
realization of the initial imperfection, the buckling load was found by transformation
of the two-point non-linear boundary-value problem to an initial-value problem via the
Qiria-Davidenko method. When the linear elastic stiffness modulus was taken such that
the associated linear structure had a buckling mode with one half-wave, in the context
of a special class of auto-correlation functions of the initial imperfections, Fraser’s
single-mode solution (1965) was in excellent agreement with the results obtained by
the Monte Carlo method. By contrast, when the modulus was such that the associated
linear structure had a buckling mode with more than one half-wave, and/or the initial
imperfection function was not co-configurational with the buckling mode of the asso-
ciated linear structure, the single-mode solution turned out to be insufficient. When the
linear “spring” constant of the foundation increases as the non-linear spring constant is
kept constant, the variance of the buckling loads of the non-linear structure decreases
throughout the rather wide range under consideration. Still, this decrease does not con-
flict with the conclusion of Fraser and Budiansky (1969) that for the infinite column
the buckling load is a deterministic quantity.
We must digress here and mention that the Monte Carlo method was applied by
Fraser (1965) in the particular case when the buckling mode of the associated linear
structure was represented by one half-wave. He also considered the buckling of struc-
tures involving exponentially correlated random imperfections (private communication
to I. Elishakoff, 1978).
3.2.2 The Linear Problem

We start with the following differential equation:
d 4w d 2 (w + w̄)
EI +P + k1 w = 0 (3.1)
d x4 dx2
where E is the Young’s modulus, I is the section moment of inertia, w̄ is the initial
imperfection function (see Figure 3.1), w(x) is the additional deflection due to the
axial load P, k1 is the linear “spring” constant of the foundation, and x is the axial
co-ordinate. The column is simply supported, so that the boundary conditions are
d 2w
w= =0 at x =0 and x=L (3.2)
dx2
where L is the length of the column.
Figure 3.1 Column on non-linear elastic foundation.
If w̄(x) ≡ 0, we obtain an associated equation for a perfect column on a linear

foundation. We consider, for the buckling problem, solutions of the form
mπ x
w = a sin (3.3)
L
where m is the number of half-waves and a is an arbitrary constant. Substituting (3.3)
into (3.1), it is determined that P must have a value of Pm given by

k1 L 4 π 2 EI
Pm = PE m + 2 2
, P E = (3.4)
m EIπ 4 L2
where PE is the buckling load of the column without elastic foundation. The value of
m = m ∗ , which makes Pm a minimum, determines the number of half-waves during
buckling of the linear structure. Consequently, the classical buckling load of a column
on a linear elastic foundation is

κ1 k1 L 4
Pcl = PE m 2∗ + 4 2 , κ1 = (3.5)
π m∗ EI
Unlike the column without elastic foundation, m ∗ does not necessarily equal unity.
Assuming that the non-dimensional linear foundation coefficient is κ1 , we arrive at a
situation where Pcl in Equation (3.5) is smaller for m ∗ = 1 than for m ∗ . The limiting
value of κ1 , at which transition from m ∗ + 1 occurs, is found from the condition that the
corresponding expression (3.5) should yield the same value for Pcl for m ∗ + 1, namely
κ1 κ1
m 2∗ + 4 2 = (m ∗ + 1)2 + 4
π m∗ π (m ∗ + 1)2
and m ∗ is determined from
π 4 m 2∗ (m ∗ + 1)2 = κ1 (3.6)
As κ1 increases, so does the number of half-waves during buckling, and for m ∗ 1,
Equation (6) reads
κ 4 m 4∗ − κ1 = 0, m ∗ = [m + 1] (3.7)
where [κ] refers to the integer part of κ.
Substituting Equation (3.7) in Equation (3.5), we obtain

√
Pcl = κ1 /π 2 PE (3.8)
Concerning the problem of imperfection sensitivity, we consider the case where

∞
kπ x
w̄(x) = ξ̄k sin (3.9)
k=1
L
√
= I /A being the section radius of gyration; w(x) can then be sought in the form

∞
kπ x
w(x) = ξk sin (3.10)
k=1
L
Substituting Equations (3.9) and (3.10) in Equation (3.1), we obtain as the relation
between ξ and
P
ξk = ξ̄k (3.11)
Pk − P
which is meaningful, provided P is sufficiently small compared to Pcl = min Pk .
3.2.3 Deterministic Imperfection Sensitivity:

Mixed Quadratic-Cubic Foundation
The differential equation for the column’s displacement w(x) is now
d 4w d 2 (w + w̄)
EI + P + k1 w − k2 w 2 − k3 w 3 = 0 (3.12)
dx4 dx2
where the original notation is adhered to and the new symbols k2 and k3 represent
non-linear “spring” constants, assumed positive (Figure 3.1).
Equation (3.12) can be modified by introducing dimensionless independent and
dependent variables in the form
x w w̄ P
η= , u= , ū = , α= ,
L Pcl
(3.13)
k2 L 4
k3 L
2 4
κ1
κ2 = , κ3 = , γ (κ1 ) = m 2∗ π 2 + 2 2
EI E m∗π
whereas Pcl is defined in Equation (3.5), and κ1 is in Equation (3.5). Equations (3.12)
and (3.2) then become
d 4u d 2u d 2 ū
+ αγ (κ 1 ) + κ 1 u − κ 2 u 2
− κ 3 u 3
= αγ (κ1 ) (3.14)
dη4 dη2 dη2
d 2u
u= =0 at η=0 η=1 (3.15)
dη2
The problem is defined as follows: Given the spring coefficients of the foundation
κ1 , κ2 , and κ3 and the function ū(η), find the buckling load α = α ∗ , defined from the
requirement (Fraser and Budiansky, 1969)

dα
= 0, F = F(u, ū) (3.16)
dF
here F(u, ū) is some functional of non-dimensional deflections; in the present case, we
use the end shortening of the column (i.e., the distance the column ends move closer
together under load)
L
∼ 1 dw 2 d w̄ dw
d= + dx (3.17)
0 2 dx dx dx
In terms of the non-dimensional quantities, the functional could be written as
1 2
d 1 du d ū du
F(u, ū) = 2 = + dη (3.18)
0 2 dη dη dη
resorting to Boobnov-Galerkin’s method, we expand u(η) and ū(η) in series in terms
of the modes of stability loss of associated linear structure, given by Equations (3.9)
and (3.10). Substituting them in Equation (3.14), multiplying the resulting equation by
sin(mπη) and integrating, we arrive at the following infinite set of coupled non-linear
algebraic equations for ξm :
2 2
m∗ s2 m∗
αm ξm − α(ξm + ξ̄m ) − s1 Jm − Im = 0, m = 1, 2, . . .
m 8 m
(3.19)
where

∞
∞
∞
∞
∞
Jm = B pqm ξ p ξq , Im = 8 A pqr m ξ p ξq ξr (3.20)
p=1 q=1 p=1 q=1 r =1
1
B pqm = sin( pπ η) sin(qπ η) sin(mπ η)dη
0
= B( p + q, m) + B(m + q, p) + B(m + p, q) − B( p + q, −m) (3.21)

0, m = p+q
B( p + q, m) = 1 1 − (−1) p+q−m
 , m = p + q
4π p+q −m
(mπ)2 + κ1 (mπ )−2 2κ2 2κ3
αm = , s1 = , s2 =
(m ∗ π)2 + κ1 (m ∗ π )−2 κ1 + m 4∗ π 4 κ1 + m 4∗ π 4
(3.22)
and δij is the Kronecker delta.

Closed solution of the infinite set of non-linear equations in Equations (3.19) seems
to be unfeasible; these equations must be truncated and solved numerically. Before doing
so, let us consider the analytic solution obtainable in the single-mode case.
3.2.4 Single-Mode Solution

Retaining only the m ∗ th term in the series in Equations (3.9) and (3.10), Equation (3.19)
reduce to a single equation, namely (asterisks are omitted):

3 m∗ 2 2
α(ξm + ξ̄m ) = ξm αm − s1 ξm − s2 ξm ,
8 m
(3.23)
1 2 2
F = m π ξm (ξm + 2ξ̄m )
4
where
4κ2 [1(−1)m ] 2κ3 2κ3
s1 = , s2 = = (3.24)
3m 2 π 3 γ (κ1 ) (m ∗ π )4 + κ1 (m ∗ π )2 γ (κ1 )
When κ2 = 0 and/or m is even, Equation (3.23) coincides formally with Equation (2.18)
in Fraser’s thesis (1965) for a column on a cubic foundation. It is also worth noting that
according to Equation (3.23), ξm and ξ̄m have the same sign (i.e., on further deformation
of the column, the total ξ̄m + ξm increases by the absolute value of the additional
displacement); this is seen from the fact that the assumption ξ̄m ξm < 0 may imply
α < 0 for 0 < |ξm | < ξ̄m (i.e., a tensile force), which contradicts the terms of the problem.
Moreover, it is seen that αm ≥ am ∗ = 1.
For a perfect column (ξ̄m = 0), Equation (3.23) reduces to

3 m∗ 2 2
ξm α − αm + s1 ξm + s2 ξ = 0 (3.25)
8 m2
It represents two branches, the straight line ξm = 0 and the parabola

3 m∗ 2
α=− s2 ξ m 2 − s1 ξm + αm (3.26)
8 m
which intersects the α-axis at α = αm , the maximum being
s12
αmax = αm + (3.27)
4γ1
Obviously, for a perfect column (ξ̄m = 0), there is no sign restriction on additional
displacements; hereinafter we will confine ourselves to the case κ2 > 0, κ3 > 0.
Differentiating Equation (3.23) with respect to F and setting
dα
= 0, α = α∗ (3.28)
dF
we obtain, after lengthy algebraic manipulations, the relation between the buckling load
α ∗ and the initial imperfection amplitude ξ̄m
2
∗ s12 3 81 m ∗ 2 ∗ s1 2s13 ∗
αm − α + = s2 −(αm − α ) − − α ξ̄m
3γ1 32 m 3γ1 27γ12
(3.29)
3 m∗ 2
γ1 = s2
8 m
When κ2 = 0 and/or m is even, Equation (3.23) reduces to Equation (30) of

Elishakoff (1979a) and formally coincides with Equation (2.18) of Fraser work (1965)

∗ 3 81 m ∗ 2 2 ∗ 2
(αm − α ) = s2 ξ̄m (α ) (3.30)
32 m
Note that Equations (3.30) and (3.31) are particular cases of Equation (26) of
Elishakoff (1980a). The latter deals with static and dynamic buckling of a simple non-
symmetric structure, namely a three-hinge, rigid-rod system, constrained laterally by a
non-linear spring, the force in the latter including both quadratic and cubic terms. In
other words, it is a generalization of the model originally proposed by Budiansky and
Hutchinson (1964). The additional displacement corresponding to α ∗ is given by
!
1
ξm(1,2) = −s1 ± s12 + (αm − α ∗ )3γ1 (3.31)
3γ1
Note that ξm(1,2) depends on ξ̄m via α ∗ :

∗s1 2s13 4 2 m s 2 3/2
ξ̄m(1,2) = (αm − α ) − ± αm − α ∗ + 1 (3.32)
3γ1 α ∗ 27γ1 α ∗ 9α s2 m ∗ 3γ1
For s1 = 0, Equation (3.30) coincides with Equation (41) of Elishakoff (1979a). ξ̄m,1
and ξ̄m,2 are not necessarily meaningful for any α ∗ ; to begin with, α ∗ is bounded by the
following value
s12
α ∗ ∗ = αm +
3γ1
Consider the case κ2 > 0; then for
αm < α ∗ < α ∗ ∗ (3.33)
Equation (3.30), as is readily shown by expansion of its last term in Taylor series in the
vicinity of (αm − α ∗ ), yields ξ̄m,(1,2) > 0, whereas ξm(1,2) < 0 [Equation (3.32)]. This
conflicts with the earlier observation that ξm ξ̄m > 0; hence, only the branch associated
with ξ̄m < 0 has a physical sense.
Figure 3.2 shows a typical α ∗ − ξ̄m curve for m = m ∗ = 1, a = 1, κ1 = π 4 , κ2 =
0.4 κ1 , κ3 = 0.1 κ1 . For κ2 = −0.4κ 4 with other data as before, the α ∗ − ξ̄m graph is
the mirror image of its counterpart for κ2 = 0.4 π 4 with respect to the α ∗ -axis. Dashed
lines represent the meaningless branches of Equation (3.30). Note that the buckling load
can exceed that of a perfect structure; for ξ̄m = 0, the buckling load sets in at the values
s1
ξm = 0 and ξm = (3.34)
2γ1
which in turn are associated with the maxima of α ∗ on the left- and right-hand bran-
ches of the α ∗ − ξ̄m curves, respectively,
s12
αmax = αm and αmax = αm + (3.35)
4γ1
Figure 3.2 Buckling load parameter α as a function of the critical

imperfection amplitude.
Note also that for the case of a cubic foundation s1 = 0, both branches of α ∗ − ξ̄m curves
intersect at αmax = αm .
3.2.5 Multi-Mode Solution

First, we truncate Equations (3.19) and (3.20) to some N (number of terms taken into
consideration). We then apply the Qiria-Davidenko method (1951, 1953) for reducing
the set of non-linear algebraic equations to the initial value problem. We denote
2 2
m∗ s2 m∗
Bm (ξ1 , ξ2 , . . . , ξ N , α) ≡ (αm − α)ξ M − α ξ̄m − s1 Jm −
(N )
Im(N )
m 8 m
m = 1, 2, . . . , N (3.36)
here

N
N
N
N
N
Im(N ) = 8 Apqrm ξ p ξq ξr , Jm(N ) = Bpgm ξ p ξq (3.37)
p=1 q=1 r =1 p=1 q=1
With F as a new independent variables, and α and ξm as its functions, we differentiate

Equation (3.36) with respect to F and obtain
∂ Bm dα N
∂ Bm dξk
+ =0 (3.38)
∂α dF k=1 ∂ξk dF
Equation (3.38), in conjunction with the equation

N
∂ F dξ p
=1 (3.39)
k=1
∂ξ p d F
can be considered as a set of ordinary linear differential equations in α and ξk (k =

1, 2, . . . , N ). Initial conditions are obtainable from the unstressed state of the column,
in the absence of load
α = 0, ξ1 = ξ2 = · · · = ξ N = 0 at F =0 (3.40)
Equations (3.38) and (3.39) take the following final form:

dα N N N N
∂ξ p
−(ξm + ξ̄m ) + (αm − α)δm, p − 2 eimp ξi − 3 Cijmp ξi ξ j =0
dF p=1 i=1 i=1 j=1
∂F
(3.41)

N
p π dξ p
2 2
(ξ p + ξ̄ p ) −1=0 (3.42)
p=1
2 dF
where
2 2
m∗ m∗
eimp = s1 Bimp cijmp = s2 Aijmp (3.43)
m m
The set of Equations (3.41) and (3.42) subject to (3.40) yields the entire α − F curve,
the buckling load being defined as the maximum attainable load on the portion of the
curve originating at zero load. Note that the final equations used to find the α − F
curves, Equations (3.41) and (3.42), differ from the comparable Equations
" N (33) and
(34) of Elishakoff (1979a), in that they contain an additional term −2 i=1 eimp ξi .
3.2.6 Buckling Under Random Imperfections – Monte Carlo Method

Assume now that the initial imperfection function is a Gaussian random function of the
position η with given mean functions
Ū (η) = E[ū(η)] (3.44)
and auto-covariance function
K ū (η1 , η2 ) = E{[u(η1 ) − Ū (η1 )][u(η2 ) − Ū (η2 )]} (3.45)
where η1 and η2 are two (generally distinct) points on the column axis (“observa-
tion” points), and E(· · ·) denotes mathematical expectation. In these circumstances,
the dimensionless buckling load α becomes a random variable. The problem consists
in determining the reliability of the structure, defined as the probability of the α ∗ ex-
ceeding some prescribed level α :
R(α ) = Prob(α ∗ > α ) (3.46)
Purely analytic solution of this problem seems to be unfeasible. With the advent of
high-speed digital computers, statistical simulation, the so-called Monte Carlo method
(Shreider, 1964; Hammersley and Handscomb, 1964; Rubinstein, 1981) can be used
as the logical remedy. Using this method, one usually starts with a simulation of the
stochastic variables involved. Then, the numerical solution of the boundary-value prob-
lem has to be carried out repeatedly for every realization of those simulated stochastic
variables. Finally, statistical analysis is performed on the data generated by the Monte
Carlo simulation by calculating the relative frequencies.
In view of Equation (3.9), the mean imperfection function can be written as

N
Ū (η) = E(ξ̄m ) sin(mπ η) (3.47)
m=1
1
E(ξ̄m ) = 2 Ū (η) sin(mπ η) dη (3.48)
0
Keeping Equations (3.47) and (3.48) in mind, the auto-covariance function is then
written as

N
N
K ū (η1 , η2 ) = σmn sin(mπ η1 ) sin(nπ η2 ) (3.49)
m=1 n=1
where
σmn = E{[ξ̄m − ξ̄m ][ξ̄n − ξ̄n ]} (3.50)
is the covariance of ξ̄m and ξ̄n , given in terms of K ū (η1 , η2 ) as
1 1
σmn = 4 K ū (η1 , η2 ) sin(mπ η1 ) sin(nπ η2 ) dη1 dη2 (3.51)
0 0
A simulation procedure for the random initial imperfections was outlined in
Elishakoff (1979b) and applied to various problems (Elishakoff, 1978b, 1979a). Here,
we give the final result: The random initial imperfections ū(η) with the specified mean
function Ū (η) and auto-correlation function K ū (η1 , η2 ), can be simulated by

N
ū(η) = ξ̄m sin(mπ η) (3.52)
m=1
here

N
ξ̄m = E(ξ̄m ) + cmk dk (3.53)
k=1
d1 , d2 , . . . , d N being independent normal variables with zero mean value and unit vari-
ances, and cmk being elements of the lower triangular matrix C. The matrix [#] =
{σmn } N ×N is positive definite and uniquely decomposable in the form [#] = [C][C T ].
With [#] known, we obtain [C] by Cholesky’s procedure for factoring a positive
definite matrix; the well-known algorithm is not reproduced here (for further details
see Elishakoff, 1983b).
Thus, with different realizations of the independent normal variables d1 , d2 , . . . , d N ,
we obtain corresponding realizations of the dependent random variables ξ̄1 , ξ̄2 , . . . , ξ̄ N
– the Fourier coefficients of the random initial imperfection function. For each such
realization, the initial value problem, Equations (3.40)–(3.43) must be solved in order
to find the realization of α ∗ . The empirical function R ∗ (α ) is then obtainable as
1
R ∗ (α ) = u M (α ) (3.54)
M
where u M (α ) is the number of α ∗ values exceeding α , and M is the ensemble size
(number of trials).
With a single-term approximation, a closed expression is obtainable for R(α ).
In this case, ξ̄m is a normally distributed random variable with mean value ξ̄m as
per Equation (3.48) and variance σmn as per Equation (3.50). It can be seen from
Equation (3.33) (for α < αm ) that
R(α ) = Prob(α ∗ > α ) = Prob(ξ̄m,1 < ξ̄m < ξ̄m,2 ) (3.55)
where ξ̄m,1 and ξ̄m,2 are defined as

s1 2s13 4 2 m s12 3/2
ξ̄m(1,2) = − (αm − α ) − ±
αm − α + (3.56)
3γ1 α 27γ1 α
2
9α s2 m ∗ 3γ1
Finally, we have

ξ̄m,2 − ξ̄m ξ̄m,1 − ξ̄m
R(α ) = erf √ − erf √ (3.57)
σmn σmn
Note that for a column on a cubic foundation and ξm = 0, Equation (3.57) reduces to
Equation (41) of the paper by Elishakoff (1979a):

ξ̄m 4 2
R(α ) = erf √ , ξ̄m = (1 − α )3/2 α (3.58)
σmn 9 s2
and is formally identical to Equation (17) of Fraser (1965).
For α ≥ αm , k2 > 0, we have

ξ̄m,2 − ξ̄m ξ̄m
R(α ) = erf √ + erf √ (3.59)
σmn σmn
Consequently, there is some non-zero probability of α ∗ > αm (i.e., of the buckling load
exceeding that of perfect structure)

ξ̄m,2 − ξ̄m ξ̄m
R(αm ) = erf √ + erf √ (3.60)
σmn σmn
where

2s13 4s13 m 2
ξ̄m,2 =− + (3.61)
27γ12 αm 27αm m ∗ 3s2 γ13
Now the question arises regarding how to choose the imperfection auto-covariance
function K ū (η1 , η2 ) so that it is non-homogeneous, subject to boundary conditions
ū(0) = ū(1) = 0 (3.62)
Following Krenk’s suggestion (private communication, 1979), we visualize that the
finite structure is obtained by cutting an infinitely long one to a given length and fixing
the ends afterward. The initial deviations ν(η) from a middle line before cutting to unit
length and fixing the ends can be assumed to be a normal field with mean function
E[ν(η)] and the autocorrelation function
K ν (η1 , η2 ) = E{[ν(η1 ) − E(ν(η1 ))][ν(η2 ) − E(ν(η2 ))]} (3.63)
The structure is now cut, and the line between the end points is used as the axis.
Assumption of small angles yields
ū(η) = ν(η) − ν(0) − η[ν(1) − ν(0)] (3.64)
and Equation (3.62) is satisfied.
The covariances σmn in Equation (3.51) take the form
4
σmn = σ̃mn + K ν (0)[(−1)m − 1][(−1)n − 1]
mnπ 2
4
+ [K ν (0) − K ν (1)][(−1)m + (−1)n ]
mnπ 2

4 1 1
− [1 + (−1) ] m+n
(K ν , sin(nπη)) + (K ν , sin(mπ η)) (3.65)
π m n
where (K ν , sin(mπn)) denotes the inner product and σ̃mn is determined by a relation
quite analogous to Equation (3.51) with K ū (η1 , η2 ) replaced by K ν (η1 , η2 ).
3.2.7 Numerical Examples

The numerical examples were worked out using the auto-covariance function
A sin B(η1 − η2 )
K ū (η1 , η2 ) = (3.66)
η1 − η2
with the product AB as the variance of the initial imperfections, and with the mean
function chosen as zero. The Fourier coefficients σmn were approximated from Equation
(3.51) by numerical quadrature, and the elements of matrix [C] were obtained by the
Cholesky decomposition procedure. Equations (3.41), (3.42), and (3.43) were integrated
for numerous “trial” values of the random Fourier coefficients ξ̄1 , ξ̄2 , . . . , ξ̄ N , using
Hamming’s predictor-corrector method with varying step size (Elishakoff, 1979a). The
sign of the quantity of dα/dF, the (N + 1) component of the unknown vector ξ ,
ξ T = (ξ1 , ξ2 , . . . , ξ N , α)
was checked throughout the integration process to yield the interval containing the
buckling load, which was in turn identified as the maximal value of α. Continu-
ous nondimensional end-shortening curves are shown in Figure 3.3 for different ξ̄m
Figure 3.3 Typical load end-shortening curves (purely cubic foundation).
values with A = 0.01/2π, B = 1; for m ∗ = 1, the single-mode approximation suf-

fices for the buckling load approximation (Elishakoff, 1979a). The attendant load-
end shortening curves are portrayed in Figure 3.4. With the buckling loads known
for a sufficient number of trials, the empirical reliability function was determined
by Equation (3.55). The number of realizations used was taken as 10,000 so that,
according to the Kolmogorov-Smirnov test of goodness of fit (Massey, 1951), the
critical values of the maximum absolute difference between the theoretical R(α )
and the √empirical R ∗ (α ) reliability function (obtained by the Monte Carlo method)
is 1.36/ 10, 000 = 0.0136 at a level of significance of 0.05. To be able to check
the Monte Carlo method against the exact solution, a single-term Boobnov-Galerkin
Figure 3.4 Load end-shortening curves for different ξ̄ values for quadratic-
cubic foundation (κ1 = π 4 ).
approximation was used. Results from the exact formula Equation (3.58) are given in
Figure 3.5, where Figure 3.5(a) is the solution of Equation (3.57), and Figure 3.5(b) is
the probability density of the initial imperfection amplitude. The shaded area in Fig-
ure 3.5(b) represents the reliability at the non-dimensional load level α = 0.8 with the
following parameters:
m = m ∗ = 1, κ1 = π4 , κ2 = κ3 = 0.1κ1 ,
(3.67)
αmax = 1.012008, σ11 = 0.231477 × 10−3
The calculation results are given in Figure 3.6 based on Equation (3.58) are in excellent
agreement with those of the Monte Carlo method (shown by the stars), the exact solution
practically coinciding with the latter. The maximum difference between R(α ) and
R ∗ (α ) is 0.005, much smaller than the critical value of 0.0136. Thus, the Kolmogorov-
Smirnov test is too conservative, and the actual situation may be much better than
predicted by the test. Note that inclusion of the quadratic terms skews the single-
mode imperfection-sensitivity curve [Figure 3.5(a) by comparison with Figure 2 of
Elishakoff, 1979a], so as to produce more sensitivity for positive values of ξ̄m and less
for negative.
Figure 3.7 shows the influence of the coefficient B on R ∗ (α ), which is seen to
decrease as B increases (for constant A = 0.03). For example, at load level 0.95 the
reliability equals 1.0 for B = 1 (i.e., almost none of the columns buckles), 0.975 for
B = 2 (i.e., about 2.5% of columns buckle), and 0.425 for B = 3 (i.e., about 57.5% of
the columns buckle).
Figure 3.8 portrays the influence of the coefficient A on R ∗ (α ), which decreases
with the increase of A. This is also understandable because a larger A means a larger
variance of the initial imperfections. As is seen in Figure 3.8, the reliability function
furnishes a basis for design of the stochastically imperfect structures. The criterion is
for such a design the reliability should be greater than some required reliability R. The
Figure 3.5 Single-term Boobnov-Galerkin solution: (a) buckling load parameter

as function of critical imperfection amplitude; (b) probability density of critical
imperfection amplitude, with the shaded area equal to the reliability of the
structure at non-dimensional load level α = 0.8.
maximum non-dimensional load αdesign , which satisfies the equation

R(αdesign ) = R (3.68)
is then the design strength for the ensemble of the columns. If, for example, R = 0.9,
then Figure 3.8 provides us with the design load levels 0.888 for A = 0.03, 0.904 for
A = 0.02, and 0.914 for A = 0.01. Figure 3.9 shows a histogram of non-dimensional
buckling loads, while Figure 3.10 portrays the sample variance.
The method shown in this section can be extended to other structures, and is
deemed to provide a sound theoretical basis for engineering design (Elishakoff, 1983a,
1983b, 1999). The first and most difficult step is to compile extensive experimen-
tal information on imperfections, classified according to the manufacturing process,
Figure 3.6 The calculation results based on the exact solution [Equa-
tion (3.57)] coincide with those obtained by the Monte Carlo method.
with a view to determining the statistical measures (the mean imperfection func-
tion and the auto-covariance function). On this basis, available analytic approaches
(Koiter, 1945; Budiansky and Hutchinson, 1972; Arbocz, 1982b) permit prediction
of the buckling loads and, consequently, calculation of the reliability associated with
a given manufacturing process. In this way, the imperfection-sensitivity concept can
be introduced into the design (Elishakoff, 1983a, 1998) in contrast to the existing
knockdown-factor approach (Figure 3.11); as indicated earlier, the knockdown factor
is chosen in such a way that its product with the classical buckling load yields a lower
bound on the available experimental data. For high values of the structural reliability R,
the reliability approach is not as conservative as the knockdown-factor approach and
has a sound theoretical basis; consequently, it permits association of the design loads
with a specified manufacturing process.
Figure 3.7 The reliability function R(α )

versus the non-dimensional actual
load α .
Figure 3.8 Calculation of the non-dimensional design loads associated

with specific required reliability, of the structure (after Elishakoff, Acta
Mechanica, reprinted with permission, Springer Verlag, 1985).
Figure 3.9 Histogram of non-dimensional buckling loads

(after Elishakoff, Journal of Applied Mechanics, 1979a).
Figure 3.10 Sample variance of the non-dimensional buckling load versus κ1 (after
Elishakoff, Journal of Applied Mechanics, 1979a).
Figure 3.11 How to introduce the imperfection-sensitivity concept into design (after Elishakoff,
1983a).
3.3 NON-LINEAR BUCKLING OF A STRUCTURE WITH RANDOM IMPERFECTION 127
3.3 Non-Linear Buckling of a Structure with Random Imperfection

and Random Axial Compression by a Conditional Simulation Technique
In Section 3.2 as well as in almost all other studies, the uncertainty associated with
the external load was not investigated. However, as discussed in Li et al. (1995), in
real situations, the external load is always subject to some amount of variability. Two
exceptions, which treated some simple cases including the uncertainty in loads in
addition to that of the initial imperfection, were offered by Roorda (1980) and Elishakoff
(1983a). In this section, we postulate a probabilistic model for the external load and
assume it to be a random variable. Due to the complexity of the deterministic procedure
involved in the evaluation of the buckling load, we must again resort to a solution by a
simulation technique. Here, an improved simulation technique is introduced to reduce
significantly the computational effort in deriving the reliability of the structure.
3.3.1 Deterministic Procedure

The differential equation for the deflection of the imperfect column on a non-linear
cubic foundation is (Elishakoff, 1979a),
d 4w d 2w d 2 w̄
EI + P + k 1 w − k 3 w 3
= −P (3.69)
dx4 dx2 dx2
For the simply supported column, the boundary conditions read:
d 2w
w= =0 at x =0 and x=L (3.70)
dx2
where w̄(x) is the initial imperfection, w(x) is the additional deflection due to the axial
load P, k1 and k3 are the “spring” constants of the foundation, and EI is the flexural
rigidity of the column.
The classical buckling load of the problem can be obtained by letting w̄(x) = 0
and k3 = 0 and substituting the expression of the classical buckling mode w(x) =
sin(mπ x/L) into Equation (3.69). Thus, one obtains

EI k1 L 4
P(m) = 2 π m + 2 2
(3.71)
L EIπ 2 m 2
We denote by m ∗ the integer m, which corresponds to the smallest value of P in
(3.71), that is,

EI k1 L 4
P(m ∗ ) = 2 π m ∗ +
2 2
(3.72)
L EIπ 2 m 2∗
where m ∗ takes the value of the integer nearest to (k1 /EI)1/4 L/π.
By introducing the non-dimensional quantities,
w w̄ x P
u= , ū = , η= , α=
L Pcl
(3.73)
k1 l 4 k 3 l 4 2 k̄1
k̄1 = , k̄3 = , γ = π m∗ + 2
2
EI EI π m∗
where is the radius of gyration of the cross section, Equations (3.69) and (3.70) can
be transformed into their non-dimensional forms,
d 4u d 2u d 2 ū
+ αγ + k̄ 1 u − k̄ 3 u 3
= −αγ (3.74)
dη4 dη2 dη2
and
d 2u
u= =0 at η=0 and η=1 (3.75)
dη2
We expand u and ū in classical buckling modes,

∞
u= ξm sin(mπ η) (3.76)
m=1

∞
ū = ξ̄m sin(mπ η) (3.77)
m=1
Substituting Equations (3.75) and (3.76) into Equation (3.73) and employing
Boobnov-Galerkin’s method yields the following set of coupled nonlinear algebraic
equations for ξm :
s m 2∗
αm ξm − α(ξm + ξ̄m ) − Im = 0 (3.78)
8 m2
in which
2k̄3 π 2 m 2 + k̄1 /(π 2 m 2 )
s= , α m =
k̄1 + m 4∗ ξ 4 π 2 m 2∗ + k̄1 /(π 2 m 2∗ )
∞ ∞ ∞
Im = ξ p ξq ξr [δ p+q,r +m − δ| p−q|,r +m − δ p+q,|r −m| + δ| p−q|,|r −m| + δ p,q δr,m ]
p=1 q=1 r =1
(3.79)
Here m ∗ is the buckling wave number, and δ p,q is the Kronecker delta.
An approximate solution to Equation (3.67) can be obtained by properly truncating
these equations and retaining the terms that have the most important contribution to
the buckling load. Fraser and Budiansky (1969) as well as Elishakoff, Cai, and Starnes
(1994a, 1994b) have pointed out that the most significant contribution to the solution
comes from the m ∗ th term and its neighboring terms. This is especially true when k3 is
much smaller than k1 , and the system possesses a “weak” nonlinearity. Hence, retaining
a few terms on either side of the m ∗ th term, one can solve Equation (3.67) step by step for
incrementally assumed ξm , using a Newton-Raphson type of iteration procedures. The
buckling load P ∗ is determined as the maximum load the system can carry. Because the
column falls into the category of the imperfection-sensitive structures, the unavoidable
presence of initial imperfection lowers the buckling load, thus reducing the load-bearing
capacity of the structure. This means when initial imperfection is present, the non-
dimensional parameter α(= P ∗ /Pcl ) is below unity.
3.3.2 Formulation of Basic Random Variables

As in the previous section, the initial imperfection expression Equation (3.66) needs to
be truncated,

N
ū(η) = X̄ m sin(mπ η) (3.80)
m=1
where N is the number of retained terms; ξ̄m is a possible value the random variable
X̄ m can assume.
Generally speaking, if the mean initial imperfection Ū (η) = E[ū(η)] is known
through measurements, the mean values µm = E[ X̄ m ] can be calculated as
1
µm = 2 Ū (η) sin(mπ η)dη (3.81)
0
and the auto-covariance function of the imperfections is

K
K
Cū (η1 , η2 ) = E[( X̄ m − µm )( X̄ n − µn )] sin(mπ η) sin(nπ η) (3.82)
m=1 n=1
The elements vmn of the variance-covariance matrix [V ] are determined by
vmn = E[( X̄ m − µm )( X̄ n − µn )]
1 1
=4 Cū (η1 , η2 ) sin(mπ η1 ) sin(nπ η2 ) dη1 dη2 (3.83)
0 0
In this case, the initial imperfection coefficients {ξ̄ } can be simulated in terms of
the following formula (Elishakoff, 1983b):
{ X̄ } = [C]{Z } + {µ} (3.84)
where [C][C]T = [V ], and {Z } is an independent standard normal distribution vector,

{µ}T = {µ1 , µ2 , . . . , µ N }.
When the initial imperfection mean function Ū (η) is not easily available, Elishakoff,
Cai, and Starnes (1994a, 1994b) proposed the use of a truncated normal distribution
for each random variable, that is,
 2
C exp − ξ̄m ,

|ξ̄m | ≤ Am
m
f X̄ m (ξ̄m ) = bm2 (3.85)

0, |ξ̄m | > Am
where f X̄ m (ξ̄m ) is the probability density function of ξ̄m , Am is the maximum value
that can be taken by the random variable X̄ m , bm is a parameter, and the normalization
constant Cm is given by

Am −1
Cm = 2bm erf (3.86)
bm
where erf(·) is the error function, defined as

x 2
1 t
erf(x) = √ exp − dt (3.87)
2π 0 2
The advantage of this probability density function lies in its flexibility. As is easily
verified, when the value of parameter Am increases while bm remains constant, ξ̄m
approaches the (untruncated) normal distribution. As another extreme, when b2 A2 ,
ξ̄m tends to be nearly uniformly distributed.
In real situations, the external loadings are also subject to a degree of uncertainty.
Consequently, the structure may be subjected to overloads. The effect of the actual
load variation in addition to the randomness of the initial imperfection was previously
considered by Roorda (1980) and Elishakoff (1983b) for model structures that admitted
the closed-form analytical solution. When multi-mode analysis is needed, as is generally
the case with realistic structures, such exact analysis is unfeasible, and simulation
techniques should be utilized. In this investigation, we assume that the external load
P satisfies Type I extreme-value distribution, the probabilistic distribution function of
which is as follows (Thoft-Christensen and Baker, 1982):
FP ( p) = exp{−exp[−a( p − c)]}, a>0 (3.88)
where parameters a and c are related to the mean value µ P and the standard deviation
σ P in the following manner:
0.5772 π
µP = c + , σP = √ (3.89)
a 6a
3.3.3 Probabilistic Analysis

Because of the existence of random initial imperfection, the buckling load P ∗ is a
random variable. For each sample of the random initial imperfection {ξ̄ }, the procedure
outlined in Section 3.2.1 can be employed to determine the buckling load P ∗ (ξ̄ ). On the
other hand, the applied load P is generally also a random variable. Thus, the criterion
of the satisfactory performance of the system is expressed as
P ∗ ( X̄ ) > P (3.90)
Hence, the probability of failure is the probability of violation of the inequality in
Equation (20)
P f = Prob[P > P ∗ ( X̄ )] = 1 − Prob[P < P ∗ ( X̄ )] (3.91)
Due to the complexity of the deterministic procedure for the evaluation of the
buckling load, a simulation technique can be successfully applied for the determination
of the probability of failure for the bar resting on the non-linear foundation. However,
in view of the small value of the desired probability of failure for real structures in
engineering practice (P f is normally in the order of 10−4 , 10−5 , or less), the number of
simulations N required by the Monte Carlo method is extremely large: N is proportional
to the value of (1 − P f )/P f 2 , where is the desired accuracy of the simulation result
(Hammersley and Handscomb, 1964). For instance, for a structure with the probability
of failure P f = 2 × 10−4 , over 3 million simulation trials are needed in order to ensure
a 10% accuracy for the estimated result from simulations. Therefore, the use of direct
Monte Carlo simulation is unfeasible in this case, and appropriate modifications are
needed.
There exist several variation reduction techniques that are aimed at improving the
computational efficiency of the Monte Carlo method (Rubinstein, 1981), the conditional
expectation technique (Ayyub and Haldar, 1985) being one of them. Later on, Li and
Gan (1992) made further modification and supplemented the conditional expectation
technique; the simulation of basic variables is combined with the analytical evaluation
of probability (or sometimes numerical integration), and statistical treatment of the
computational results is carried out with the incorporation of the batch means technique.
In this way, both the efficiency and accuracy of the computation are greatly increased
such that the CPU time required for the solution of the problem is substantially reduced.
Here we will apply this technique to the present problem.
As we have assumed that the external load P is a random variable with probability
distribution function FP ( p), Equation (3.91) can be transformed as follows, using the
concept of conditional probability:
{P f | X̄ 1 = ξ̄1 , X̄ 2 = ξ̄2 , . . . , X̄ N = ξ̄ N } = 1 − Prob[P < P ∗ (ξ̄ )] = 1 − FP [P ∗ (ξ̄ )]
(3.92)
where FP (·) is the probability distribution of the external load P, evaluated at the value
of function P ∗ when X̄ 1 = ξ̄1 , X̄ 2 = ξ̄2 , . . . , X̄ N = ξ̄ N . The unconditional probability
of failure becomes, therefore,
+∞ +∞
Pf = ··· {P f | X̄ 1 = ξ̄1 , X̄ 2 = ξ̄2 , . . . , X̄ N = ξ̄ N }
−∞ −∞
× f N̄ (ξ̄1 , ξ̄2 , . . . , ξ̄ N ) d ξ̄1 d ξ̄2 . . . d ξ̄ N

+∞ +∞
= ··· {1 − FP [P ∗ (ξ̄ )]} f X¯ (ξ̄1 , ξ̄2 , . . . , ξ̄ N ) d ξ̄1 d ξ̄2 . . . d ξ̄ N (3.93)
−∞ −∞
It is remarkable that the right-hand side of this equation represents the mathematical
expectation of the expression in the parentheses with respect to the initial imperfection
components X̄ 1 , X̄ 2 , . . . , X̄ K :
P f = 1 − E X̄ {FP [P ∗ ( X̄ )]} (3.94)
An unbiased estimator of the mathematical expectation in (3.94) is
1 N∗
P̄ f = P̃ f (3.95)
N∗ j=1 j
where
P̃ f j = 1 − FP [P ∗ (ξ̄1 , ξ̄2 , . . . , ξ̄ N )] j (3.96)
For each sequence of the simulated random initial imperfection coefficients
{ξ̄1 , ξ̄2 , . . . , ξ̄ N }, use of the deterministic procedure described in Section 3.2.1 results
in a buckling load P ∗ , and substitution of this buckling load P ∗ into Equation (3.96)
yields an estimated value of the probability of failure through an analytical calculation.
If N∗ sequences of random variables {ξ̄1 , ξ̄2 , . . . , ξ̄ N } are generated in terms of their

respective probabilistic distributions by computer, repeating the same procedure de-
scribed previously, one can obtain N∗ approximate values of probability of failure
P̃ f j ( j = 1, 2, . . . , N ). The mean of these N approximate values, as indicated by Equa-
tion (3.95), constitutes another, but a better, estimate of the probability of failure P f .
The sample deviation of the preceding simulations reads
1 N∗
2
s2 = P̃ f j − P̄ f (3.97)
N∗ − 1 j=1
One of the advantages of the present simulation technique is that here only the
initial imperfection needs to be simulated, whereas by the direct Monte Carlo method
one has to simulate not only the initial imperfection but the external load as well. In
addition, the analytical probabilistic evaluation is performed so that the computational
efficiency is further improved.
However, it should be pointed out that P̄ f , defined by Equation (3.95), represents
an approximate value of the probability of failure. So, the following question naturally
arises: How far is this approximate value different from its real value P f , or, in other
words, what is the precision of this approximation? To answer this question, one has to
use the theory of statistics, in particular, the interval estimation.
If P̃ f1 , P̃ f2 , . . . , P̃ f N∗ are random variables satisfying
the same normal distribution
and are independent of each other, then ( P̄ f − P f )/ s 2 /N∗ tends to t(N∗ − 1) as N∗
increases, where t(N − 1) denotes the N − 1 order student distribution. The confidence
interval with 100(1 − α)% level of confidence is
 
s 2 s2 
 P̄ f − tα/2 (N∗ − 1) , P̄ f + tα/2 (N∗ − 1) (3.98)
N∗ N∗
When N∗ > 40, tα/2 (N∗ − 1) can be replaced by z α/2 , the percentile of the standard
normal distribution N (0, 1).
However, it should be pointed out that the confidence interval expressed by Equa-
tion (3.98) is only approximate. This is because, in most situations, we do not have
sufficient evidence to state that random variables P̃ f1 , P̃ f2 , . . . , P̃ f N∗ belong to the same
Gaussian distribution. Besides, P̃ f1 , P̃ f2 , . . . , P̃ f N are the data of simulation tests origi-
nated from the same set of seeds, so, strictly speaking, these data are not independent.
In order to use the classical theory of statistics, Law and Kelton (1982) proposed sev-
eral methods for transforming the simulation results into independent random variables
of the same distribution. Here we adopt the batch means technique. In the actual im-
plementation of this technique, our procedure is sightly different. Suppose the total
number of simulation tests is N∗ . We can carry out the simulation tests in M batches,
each batch performing L ∗ tests (N∗ = M × L ∗ ) and obtaining L ∗ estimated values of
the probability of failure. For the sake of clarity, we will use P to denote the estimated
probability of failure in the following simulation data.
Data from the first batch is denoted by P1 , P2 , . . . , PL ∗ , and data from the second
batch is denoted by PL ∗ +1 , PL ∗ +2 , . . . , P2L ∗ . In a perfect analogy, data from the Mth
batch is indicated as P(M−1)L ∗ , P(M−1)L ∗ +1 , . . . , PM L ∗ . The average of the data from each
batch reads
1 L∗
P̄ j = P( j−1)L ∗ +i ( j = 1, 2, . . . , M) (3.99)
L ∗ j=1
We utilize
1 M
P̌ = P̄ j (3.100)
M j=1
as the final estimator of the probability of failure.

According to the central limit theorem, if L ∗ , the number of tests in each batch,
is sufficiently large, P̄ j ( j = 1, 2, . . . , M) approximately follows the same Gaussian
distribution. In addition, every P̄ j is a mean value of a batch of test data. Therefore,
P̄ j ( j = 1, 2, . . . , M) can be viewed as “almost” independent with each other. In reality,
we use a different seed for a different batch of simulation tests, thus further reducing
the correlation among these data P̄ j ( j = 1, 2, . . . , M). The sample deviation of the
preceding simulation data is
1 M
2
sM = ( P̄ j − P̌)2 (3.101)
M − 1 j=1
Thus, the confidence interval defined by Equation (3.98) can be modified as

 
2 2
 P̌ − tα/2 (M − 1) M , P̌ + tα/2 (M − 1) s M 
s
(3.102)
M M
When using this confidence interval, we propose that the correlatedness of the data
P̄ j ( j = 1, 2, . . . , M) should be assessed. This can be done by utilizing the following
estimator of the correlation coefficient:
" M−1
j=1 ( P̄ j − P̌)( P̄ j+1 − P̌)
ρ̂ = "M (3.103)
j=1 ( P̄ j − P̌)
2
When the value of ρ̂ in expression (3.103) is large, the confidence interval expressed
by Equation (3.101) is not accurate. The method of reducing the value of ρ̂ is to
appropriately add some more tests into each batch and to increase the number of batches.
If we define the ratio of the semi-length of the confidence interval to the estimated
value P̌ of the probability of failure as the relative precision

1 s2
r = tα/2 (M − 1) M (3.104)
P̌ M
then increasing L ∗ , the number of tests in each batch, and M, the number of batches,
makes r attain a required accuracy.
In the course of simulation tests, the number of tests in each batch L ∗ and the
number of batches M should be properly chosen. Usually, L ∗ can not be too small,
otherwise the data P̄ j might not be independent, or it may deviate from the Gaussian
distribution. Similarly, M should not be too small otherwise there might be a remarkable
correlation among data P̄ j ( j = 1, 2, . . . , M). Besides, the choice of M and L ∗ is also

dependent on the accuracy required and the actual value of the probability of failure of
the structure. Generally speaking, the higher the required accuracy, and the smaller the
failure probability itself, the more simulations are needed to perform. It is worth noticing
that the adoption of the batch means technique also saves the computer storage. If the
batch means technique is not employed, the total number of required storage elements
in computer is approximately equal to the total number of simulation tests performed.
By using the batch means technique, the total number of storage elements is reduced
to the number of batches.
3.3.4 Numerical Example and Discussion

As a numerical example, we consider the case where k̄1 = 16π 4 and k̄3 = 0.1 k̄1 such
that m ∗ = 2. Numerical calculations show that here only four terms in the deterministic
procedure are needed to obtain a sufficiently accurate solution of the buckling load P ∗ .
For the initial imperfection simulation, we assume each ξ̄m (m = 1, 2, 3, 4) to have the
same probability distribution density f X¯ (ξ̄m ) [Equation (3.94)] with A1 = A2 = A3 =
A4 = A and b1 = b2 = b3 = b4 = b; A is fixed at 0.5, and b is fixed at 0.1. Further-
more, we assume that the applied load has a Type I extreme-value distribution with
the mean value µP = 0.5Pcl and standard deviation σ P = 0.05Pcl . For each sample of
the simulated initial imperfection, the deterministic procedure is carried out to get the
corresponding buckling load P ∗ and then an analytical probability evaluation is per-
formed with respect to the external load P so that an estimated value of the probability
of failure P̃ f j is obtained. In the actual computation, we carried out the simulations by
batches. The total number of simulations was 2000; here L ∗ (the number of tests in each
batch) and M (the number of batches) were taken to be 50 and 40, respectively. A 99%
confidence interval for the probability of failure is obtained through (3.102) as follows:
[0.237 × 10−4 − 0.108 × 10−5 , 0.237 × 10−4 + 0.108 × 10−5 ]
The relative precision of the approximate value P̌ f = 0.239 × 10−4 is 5%. The
whole process only consumed 4 minutes of CPU time (the numerical calculation was
conducted on a DEC 5000/200 workstation). By contrast, one has to do over 10 million
Monte Carlo simulation tests to reach a similar precision.
In engineering practice, designers usually require a very low probability of failure.
The latter should be determined with high precision. Therefore, the level of confidence
for the estimated probability of failure should be consistent with the value of the
probability of failure itself to result in a safe design. In our case, with 4000 simulations
(L ∗ = 80, M = 50), a 99.999% confidence interval for the probability of failure is
obtained as
[0.237 × 10−4 − 0.209 × 10−5 , 0.237 × 10−4 + 0.209 × 10−5 ]
The computational process requires about 8 minutes of CPU time on a DEC

5000/200 workstation.
In conclusion, it is worth mentioning that the confidence interval for the probability
of failure obtained here can be directly applied to the structural design. For instance,
suppose that the design code requires
P f ≤ [P f ] (3.105)
where [P f ] is an codified value representing the maximum tolerable probability of

failure for the structure. Because the actual value of the probability of failure for the
structure in consideration is generally unknown, use a more conservative expression
than Equation (3.105), such as
!
P̌ + tα/2 s M
2
/M ≤ [P f ]
For further details, consult the study by Li et al. (1995).

In some studies there is a claim that the randomness in the applied load makes for
reduced reliability. To elucidate this question, we consider a simple model structure,
capable of exact solution. Consider a cantilever of Figure 3.12 – a rigid link of length
a, pinned to a rigid foundation and supported by a linear extensional spring of stiffness
x, which is capable of resisting both tension and compression and which retains its
horizontality as the system deflects (Elishakoff, 1983a). The initial imperfection is
Figure 3.12 The propped cantilever.

modeled by the deflection κa from the vertical position; the total displacement is
denoted as ya. Equilibrium dictates

λy = (y − x) 1 − y 2
λ = f /cl , f cl = κa
For the non-dimensional buckling load λs – the maximum load the structure can
sustain – we put dλ/dy = 0, λ = λs . Hence, y = x 1/3 , which yields an exact expression
λs = (1 − x 2/3 )3/2
for the limit load, due to Thompson and Hunt (1973). Let the applied load be a random
variable $, uniformly distributed over the interval (λ1 , λ2 ), that is the probability density
f $ (λ) equals (λ2 − λ1 )−1 in the interval (λ1 , λ2 ) and vanishes elsewhere. The initial
imperfections are random; due to this fact, $s is a random variable too, with probability
distribution function F$s (λs ). With reasonable assumption of independence of $s and
$ due to independence of the applied loads of the manufacturing process, we find
[by analogy with Equation (7.4) of Elishakoff, 1983b]
∞

R= 1 − F$s (z) f $ (z) dz

0
The calculations yield, for initial displacements having a uniform probability density
in the interval (0, ξ ) with α ∗ = (1 − ξ 2/3 )3/2

α∗ 3 3β2 cos β2 sin5 β2 sin2 β2 sin4 β2
R= − + − +
αmax αmax ξ 48 6 24 192

3β1 cos β1 sin β1
5
sin β1 sin4 β1
2
− + − +
48 6 24 192
where
β1 = cos−1 (α ∗ )1/3 , β2 = cos−1 (αmax )1/3
For example, for r = 0.999 and ξ = 0.3, we have αmax = 0.4103, α ∗ = 0.41 and
R = 0.9999995; that is, the reliability of the structure with random applied load exceeds
that of the structure under deterministic load considered. However, if λ1 = 0.4103 and
λ2 = 0.4144, the reliability becomes

1 1 λ2
R= (1 − z 2/3 )3/2 dz = 0.9928439
(λ2 − λ1 ) ξ λ1
that is, less than the required reliability r = 0.999 associated with the deterministic
applied load. The lesson to be learned from this example is that uncertainty in the
applied load is not always accompanied by a decrease in the reliability estimate. For
further details, consult the study by Elishakoff (1983a). Note that some recent studies
do not follow a general methodology outlined in this section for taking into account
combined randomness in initial imperfection and applied load.
3.4 RELIABILITY OF AXIALLY COMPRESSED CYLINDRICAL SHELLS 137
3.4 Reliability of Axially Compressed Cylindrical Shells with Random

Axisymmetric Imperfections
This discussion of the reliability of axially compressed cylindrical shells with random
axisymmetric imperfections is based on Elishakoff and Arbocz (1982).
3.4.1 Preliminary Considerations

It has been established for some time that the theories of imperfection sensitivity of
structures should be combined with the statistical analysis of the initial imperfections.
The first work in this direction for the most controversial structure – the circular cylin-
drical shell under axial compression – was presented by Amazigo (1969). He treated
infinitely long cylindrical shells with spatially homogeneous, ergodic random axisym-
metric imperfections by means of a modified truncated hierarchy method. The conclu-
sion derived was that the buckling stress is a deterministic quantity, depending only on
the spectral density of the random axisymmetric imperfections. Moreover, for small
values of the standard deviations of the initial imperfections, this deterministic buckling
stress depended only on the value of the initial imperfection power spectral density at the
spatial frequency of the classical axisymmetric buckling mode, this dependence being
2/7
9πc2 PBIF
λ = 1 − √ S̄ w0 (1) δ 4/7 , λ= (3.106)
2 2 Pc
where PBIF is the buckling load at which the governing non-linear equations admit an
asymmetric solution infinitesimally adjacent to the prebuckling axisymmetric state, Pc
is the classical buckling load of the perfect structure, λ is the non-dimensional determin-
istic buckling load (and, therefore, also the mean buckling load), c = [3(1 − ν 2 )]1/2 , ν is
Poisson’s ratio, S̄ w0 (ω)
# ∞ is the normalized initial axisymmetric imperfection power spec-
tral density so that −∞ S̄ w0 (ω)dω = 1, ω is the non-dimensional spatial frequency so
that ω = 1 corresponds to the wave number associated with the axisymmetric buckling
mode, and δ is the standard deviation of the initial imperfections. Later on, Amazigo
and Budiansky (1972) modified Equation (3.106) to
2/7
9πc2
λ = 1 − √ S̄ w0 (1) δ 4/7 λ4/7 (3.107)
2 2
In the works of Amazigo (1969) and Amazigo and Budiansky (1972) as well as Tennyson
et al. (1971), the normalized auto-correlation function R̄ w0 (x1 , x2 ) = Rw0 (x1 , x2 )δ −2
was assumed to be of exponential-cosine type
R̄ w0 (x1 , x2 ) = e−β|η| cos(γ η) (3.108)
where η is the difference between the non-dimensional axial coordinates of the points of
observation, and β and γ are some positive constants supposed to depend on the manu-
facturing process. The normalized spectral density associated with this auto-correlation
function is
β ω2 + ω2 + γ 2
S̄ w0 (ω) = (3.109)
π ω4 + 2(β 2 − γ 2 )ω2 + (β 2 + γ 2 )2
so that S̄ w0 (1) entering into Equations (3.106) and (3.107) is

β 1 + β2 + γ 2
S̄ w0 (1) = (3.110)
π 1 + 2(β 2 − γ 2 ) + (β 2 + γ 2 )2
A design criterion based on formula (3.107) was developed by Tennyson et al. (1971).
Two approaches were suggested: to substitute the experimentally determined values of
δ and S̄ w0 (1) into Equation (3.107) or to assume, like Amazigo (1969) and Tennyson
et al. (1971), the exponential-cosine auto-correlation function (3.108) with parameters
β and γ chosen so that the spectral density (3.109) will have a peak at the spatial fre-
quency, which corresponds to the axisymmetric classical buckling mode of the perfect
shell. Experimental results (Arbocz, 1974), however, did not yield good agreement with
the proposed random axisymmetric imperfection model because the measured discrete
power spectral density of initial imperfections did not have a peak at the spatial fre-
quency which corresponds to the axisymmetric classical buckling mode of the perfect
shell. A different approach was used by Roorda and Hansen (1972). They assumed
that the shape of the initial imperfection was specified in the form of the axisymmet-
ric buckling mode of the cylindrical shell and considered its magnitude as a random
variable with given probability distribution. Then the following relationship derived by
Koiter (1945) using his general non-linear theory of elastic stability
2(1 − λ)2 − 3c |µ|λ = 0 (3.111)
between the non-dimensional buckling load λ and the magnitude of non-dimensional

imperfection µ was used as a transfer function to calculate the probability character-
istics of buckling load in terms of the probability characteristics of the imperfection
magnitude. Finally, the reliability function
R(α) = Prob(α < $ ≤ 1) (3.112)
where (α < $ ≤ 1) stands for the random event that the random buckling load $ will
exceed given non-dimensional load α, was calculated.
In a subsequent paper, Roorda (1972a) proposed to consider all kinds of imperfec-
tions in a real shell of given length, radius, thickness, and boundary conditions to be
equivalent to a hypothetical axisymmetric imperfection in a shell of infinite length with
the same radius and thickness. This equivalent imperfection was treated as a random
normal variable with its mean and variance approximated as linear function of the R/t
ratio. The obtained formulas were compared with results of some 360 experiments on
axially compressed cylindrical shells with different length, radius-to-thickness ratios,
boundary conditions, materials, and manufacturing processes, which were reported by
Hart-Smith (1970). Roorda also compared his results with the “lower bound” curve
proposed by Weingarten, Morgan, and Seide (1965) for a large number of test results
from different sources.
Makarov (1970) was apparently the first who performed systematic statistical anal-
ysis of the initial imperfections. He used series of 50 cylindrical shells made of steel
sheets of electrical-grade pressboard. The imperfection function was represented in
Fourier series, and the coefficients were treated as random variables. The analysis
showed that the assumption of homogeneity of the initial imperfection in the circum-
ferential direction was satisfactory and that the normality of their Fourier coefficients
did not conflict with the experimental data. Subsequently, Makarov (1969, 1970) per-
formed a theoretical analysis of the buckling of stochastically imperfect shells with
the experimental data (see Makarov, 1969) serving as an input for the description of
imperfections. The theoretical mean value of the non-dimensional buckling load was
0.31, whereas the experiments yielded 0.23.
General, non-axisymmetric random imperfections were treated by Fersht (1974)
and Hansen (1977). Fersht generalized the approach by Amazigo (1969). It turned out
that, for the non-axisymmetric imperfections, closed formulas of the type (1) or (2) are
not obtainable and rather cumbersome numerical analysis has to be performed to yield
the mean buckling load. Hansen (1977) generalized his previous deterministic results
(Hansen, 1975). The main conclusion of that paper was that the imperfection parame-
ters associated with the non-axisymmetric modes appear in three and only three distinct
summations and that the system behavior is governed by the value of these summations
and not by the individual imperfection parameters. It was assumed that the modal imper-
fection amplitudes are jointly normal random variables with zero mean. Also the strong
assumption was made that these amplitudes are statistically independent and distributed
identically. The Monte Carlo method was then applied: For each sample problem, the
buckling load was determined (see Hansen, 1975), and the mean buckling loads as
well as their confidence levels were calculated. It has been demonstrated that the non-
axisymmetric imperfections play a very important role in the determination of the buck-
ling load statistics. Another important conclusion was that the large dispersion occurs for
small values of R/t and that this dispersion decreases as R/t increases. The same con-
clusion was accounted for by Roorda (1972a, 1972b) by postulating that the mean and
the variance of the imperfection were functions decaying with increasing values of R/t.
In this chapter, contrary to other earlier works, the probabilistic properties (the auto-
correlation functions or spectral densities) are not assumed, instead the mean vector and
the variance-covariance matrix of the Fourier coefficients are calculated from the exper-
imental measurements of the shell profiles. Then the Monte Carlo method is employed.
Thus, at first, a large number of shells is “created.” That is, the Fourier coefficients of
their initial imperfection representations are simulated numerically by a special proce-
dure. Next for each shell a deterministic analysis of buckling load evaluation is carried
out (implying that the usual deterministic approach is a particular case of the proba-
bilistic one). After the buckling loads of an ensemble of shells are available, one then
proceeds by studying their probabilistic behavior. In particular, one determines the relia-
bility function representing the relative number of shells with buckling loads exceeding
the specified load. Finally, the design load for the shells produced by a given manufac-
turing process is obtained as that load for which the reliability function has the desired
value close to unity. This section follows the study of Elishakoff and Arbocz (1982).
3.4.2 Probabilistic Properties of Initial Imperfections

We characterize the random initial imperfections W0 by functions that are the basis
of the study of random processes. The mean function W̃0 (x) of a function W0 is the
expected value of the random variable W0 (x)

∞
W̃0 (x) = E{W0 (x)} = w0 f (w0 ; x) dw0 (3.113)
−∞
where f (w0 ; x) is the first-order probability density function of W0 (x). In general, W̃0 (x)
is a function of the axial coordinate, and E{. . .} denotes the mathematical expectation.
The auto-covariance function Rw0 (x1 , x2 ) of a random function W0 (x) is the covariance
of the random variables W0 (x1 ) and W0 (x2 ):
Rw0 (x1 , x2 ) = E{[W0 (x1 ) − W̃0 (x1 )][W0 (x2 ) − W̃0 (x2 )]}
x x
= [w01 − W̃0 (x1 )][w02 − W̃0 (x)] f (w01 , w02 ; x1 , x2 ) dw01 dw02
−x −x
(3.114)
Here f (w01 , w02 ; x1 , x2 ) is the second-order probability density of the random function
W0 (x). Assume now that the {ϕ1 (x)} represents the complete set of orthogonal functions
in [0, L], where L is the shell length. Then W0 (x) can be expanded in a series in terms
of the ϕi (x)s:

W0 (x) = Ai ϕi (x) (3.115)
i
where Ai is a random variable for every fixed i. The mean function then becomes

W̃0 (x) = E(Ai )ϕi (x), E(Ai ) ≡ Ãi (3.116)
i
Here the Ãi s are the means of the Ai s and are readily found as
L L
1
Ãi = 2 W̃0 (x)ϕi (x) d x, µi =
2
ϕi2 (x) d x (3.117)
µi 0 0
and they form the mean vector { Ã}. The auto-covariance function becomes

Rw0 (x1 , x2 ) = σi j ϕi (xi )ϕ j (x2 ) (3.118)
i i
where
σij = E{[Ai − ã i ][A j − Ã j ]} (3.119)
The σij s are obtained as
L L
1
σij = 2 2 Rw0 (x1 , x2 )ϕi (x1 )ϕ j (x2 ) d x1 d x2 (3.120)
µi µi 0 0
and they form the variance-covariance matrix [#] = [σij ]. Equations (3.113)–(3.120)
imply that the knowledge of the mean and the auto-covariance function, on the one
hand, and the mean vector { Ã} and the variance-covariance matrix [#], on the other
hand, are equivalent.
To be able to treat also the limiting case of the infinite shell, we consider initially
the complex Fourier series, so that, instead of representation (3.115), we use
∞
W0 (x) = Am eimπ x/L (3.121)
m=−∞
For the mean function, we get

∞
W̃0 (x) = E(Am )eimπ x/L
m=−∞
(3.122)
L
1
Ãm = W̃0 (s)e imπs/L
ds
2L −L
The auto-covariance function becomes in terms of the elements σmn of the variance-
covariance matrix

∞ ∞
Rw0 (x1 , x2 ) = σmn eiπ (mx1 −nx2 )/L
m=−∞ n=−∞
L L
(3.123)
1 iπ (−ms1 +ns2 )/L
σmn = Rw0 (s1 , s2 )e ds1 ds2
4L 2 −L −L
Substitution of the σmn into the expression of the auto-covariance function yields

∞ ∞ L L
1 −imπ s1 /L ∈π s2 /L
Rw0 (x1 , x2 ) = Rw (s1 , s2 )e e ds1 ds2
m=−∞ n=−∞
4L 2 −L −L 0
×eimπ x1 /L e−nπ x2 /L (3.124)
Now, defining the spatial frequencies by
mπ nπ
ωm = , πn = (3.125)
L L
and the difference between the successive frequencies by
π π
ωm = , ωn = (3.126)
L L
Equation (3.125) becomes

∞ ∞ L L
1 −iωm s1 iωn s2
Rw0 (x1 , x2 ) = R w (s ,
1 2s )e e ds1 ds2
m=−∞ n=−∞
4π 2 −L −L 0
×eiωm x1 e−ωn x2 ωm ωn (3.127)
If we now define
L L
1
Sw0,L (ωm , ωn ) = Rw0 (s1 , s2 )e−iωm s1 eiωn s2 ds1 ds2 (3.128)
4π 2 −L −L
then Equation (3.127) can be rewritten as follows:

∞ ∞
Rw0 (x1 , x2 ) = Sw0,L (ωm , ωn )eiωm x1 e−ωn x2 ωm ωn (3.129)
m=−∞ n=−∞
Under very general conditions the limit of a sum of the form (3.129) as ωm → 0,
ωn → 0 is the integral
∞ ∞
Rw0 (x1 , x2 ) = Sw0 (ω1 , ω2 )eiω1 x1 e−ω2 x2 dω1 dω2 (3.130)
−∞ −∞
Therefore, since L → ∞ implies ωm → 0, we have

lim Sw0,L (ωm , ωn ) = Sw0 (ω1 , ω2 )
L→∞
where
∞ ∞
1
Sw0 (ω1 , ω2 ) = Rw0 (s1 , s2 )e−iω1 s1 eiω2 s2 ds1 ds2 (3.131)
4π 2 −∞ −∞
so that Sw0 (ω1 , ω2 ) and Rw0 (s1 , s2 ) constitute the double Fourier transform pair,
Sw0 (ω1 , ω2 ) is referred to as the generalized power spectral density of the initial im-
perfections of the infinite shell, and, as was shown earlier, it can be deduced from the
elements of the variance-covariance matrix associated with the finite shell.
Consider now the particular case when the initial imperfections of the finite shell
form a stationary random function, meaning that the mean imperfection is constant and
the auto-covariance depends only on s2 − s1 :
W̃0 (x) = constant, Rw0 (x1 , x1 + ξ ) = Rw0 (ξ ) (3.132)
Substituting into Equation (3.31), Rw0 (s2 − s1 ) instead of Rw2 (s1 , s2 ), making the change
of coordinates
s1 − s2 = ξ, s2 = η
and bearing in mind that
∞
ei(ω2 −ω1 )η dη = 2π δ(ω2 − ω1 )
−∞
yields
Sw0 (ω1 , ω2 ) = Sw0 (ω1 )δ(ω2 − ω1 )
where Sw0 (ω1 ) is the power spectral density of the weakly homogeneous initial imper-
fections (homogeneous in the wide sense). Then with ω1 → ω, one obtains
∞
1
Sw0 (ω) = Rw (ξ ) e−iωξ dξ (3.133)
2π −∞ 0
with Equation (3.130) transforming to
∞
Rw0 (ξ ) = Sw0 (ω) eiωξ dω (3.134)
−∞
Equations (3.133) and (3.134) constitute the Wiener-Khintchine relationship for the
weakly homogeneous random functions. The concept of the power spectral density was
used in the studies by Amazigo (1969) and Amazigo and Budiansky (1972) to represent
the initial imperfections.
Strictly speaking, the initial imperfections of the finite shell cannot be homoge-
neous. Tennyson et al. (1971) used measurements performed on the finite shell in order
to develop the design criterion based on formula (3.107), which was derived from the
infinite shell with weakly homogeneous imperfections. That is, the assumption was
made that the initial imperfections are homogenous in the interval [0, L], that is, condi-
tions (3.132) are valid when x1 , x2 are in this interval. Consider this case in more detail.
Assume, after Tennyson et al. (1971), that the mean imperfection function is identically
zero. Then the initial imperfections (as those of the finite shell) are characterized by the
variance-covariance matrix, Equation (3.119). Assume now in Equation (3.120) that
√
Rw0 (s1 , s2 ) = R√w0 (s2 − s1 ) and introduce a new coordinate system z 1 = (s2 − s1 )/ 2,
z 2 = (s1 + s2 )/ 2 to finally arrive at
1
(−1)n−m
σmn = Rw0 (2αL)(sin 2mπ α − sin 2nπ α) dα, if m = n
π(n − m) 0
(3.135)
and
1
σmn = 2 Rw0 (2αL)(1 − α) cos 2mπ α dα if m = n
0
where
z1 s2 − s 1 γ
α=√ = = (3.136)
2L 2L 2L
We will show that for L → ∞, the diagonal terms σmm of the variance-covariance matrix
reduce to Tennyson’s discrete power spectral density. Indeed, substituting (3.136) into
the expression for σmn results in

1 2L γ
σmm = Rw0 (γ ) 1 − cos mω0 γ dγ
L 0 2L
where the dimensional spatial frequency ω̃0 = π/L was introduced. Then

σmm 1 2L γ
= Rw0 (γ ) 1 − cos m ω̃0γ dγ
ω̃0 π 0 2L
but now, with L → ∞, ω̃ → 0 and
lim m ω̃0 = ω̃
ω̃0 →0
hence,
∞
σmm 1
lim = Rw0 (γ ) cos ω̃γ dγ = Sw0 (ω̃) (3.137)
ω̃0 →0 ω̃0 π 0
which is the desired result. This expression can be related also to the non-dimensional
discrete power spectral density discussed by Arbocz and Williams (1976). Indeed Sw0 (ω̃)
is p(ω̃). In order to non-dimensionalize Sw0 (ω̃), we introduce the dimensionless spatial
frequency ω defined as

m ω̃0 m L 2c
ω= = , i cl =
(2c/Rt) i cl π Rt
where i cl is the number of half-waves in the classical axisymmetric buckling mode for
isotropic
shells. Then the non-dimensional power spectral density S̄ w0 (ω) = Sw0 (ω̃)/
(Rt/2c) becomes
1 σmm
S̄ w0 (ω) = lim = lim σmm i cl (3.138)
(Rt/2c) ω̃→0 ω̃0 ω̃→0
which coincides with formula (A12) in the paper by Arbocz and Williams (1976).
3.4.3 Simulation of Random Imperfections with Given

Probabilistic Properties
Assume that the mean function W̃0 (x) and the auto-covariance function Rw0 (x1 , x2 ) are
given. We then calculate analytically or numerically the vector { Ã} and the variance-
covariance matrix [#]. Having these quantities, we then proceed to the simulation of the
initial imperfections. Our aim is to “create” the desired number M of initial imperfec-
tion profiles having given probabilistic properties. The method of simulation, suitable
for computer realization for a random normal function was developed by Elishakoff
(1979a). Here we are only giving a short outline of the method. We assume that the
mean vector { Ã} is identically zero. This entails no loss of generality because, if vector
{A} is simulated as is shown later, the vector {A + Ã} has the mean { Ã} and the same
variance-covariance matrix [#]. The latter matrix is positive semidefinite and can be
uniquely decomposed in the form
[#] = [C][C]T (3.139)
where T means transpose and [C] is a lower triangular matrix. [C] is found by the
Cholesky’s algorithm (Elishakoff, 1983b). Now, we can form vector {B}, its elements
being normally distributed, statistically independent with zero means and unit variances.
Then the vector of the Fourier coefficient of the initial imperfections is simulated as
follows:
{A} = [C]{B} + { Ã} (3.140)
Having realizations of vector {B}, we obtain the same number of realizations of {A}.
The main feature of this simulation technique is that it is applicable for homogeneous as
well as for non-homogeneous random functions with given mean and auto-covariance
functions. This procedure was applied to different static and dynamic buckling prob-
lems, involving random imperfection sensitivity (see Elishakoff, 1978b, 1979a, 1980b).
Using the auto-covariance function (3.108), we are able to compare the results of
the Monte Carlo method with the expressions (3.106) or (3.107) for the mean buckling
load. To begin with, we assume, after Amazigo (1969), that the initial imperfection
auto-covariance function is of the exponential cosine type,
(x2 − x1 )
Rw0 (x1 , x2 ) = δ 2 e−A(|x2 −x1 |L) cos B (3.141)
L
In order to get Amazigo’s representation in the form of Equation (3.141), the constants
A and B must have the following values:
A = βi cl π, B = γ i cl π (3.142)
where
√
L 2c
i cl = and c = 3(1 − ν 2 ) (3.143)
π RT
The elements of the variance-covariance matrix are calculated via Equation (3.120),
with ϕi = cos(iπ x/L) (i.e., the half-wave cosine representation is used for the initial
imperfections). As is shown (Elishakoff, 1979b), simplification during the calculation
of σij is possible due to the weak homogeneity of W0 . Namely, the σij s are given by
L
1
σij = 2 2 Rw0 (ξ )Mij (ξ )dξ (3.144)
µi µ j 0
where
L
Mij (ξ ) = Pij (ξ ) + Pji (ξ ), µi2 = ϕi2 (x) d x
0
L−ξ/2
1 1
Pij (ξ ) = ϕi η − ξ ϕi η + ξ dη
ξ/2 2 2
(see Equations (18)–(20) of Elishakoff, 1979b). Upon substitution and carrying out of
the integrals, these equations yield the following formulas:
4δ 2 i
σ̄ii = σii i cl =
π ωi + 2ωi (β − γ 2 ) + (β 2 + γ 2 )2
4 2 2
 2
 8δ2 ωi Āi − ω j Ā j ,
2
 (3.145)
if i = j and i + j = even
σij = σij i cl = i cl ωi2 − ω2j


0, if i + j = odd
where

1 P̄i Q̄ i T̄ i R̄ i
i = β Ū i − + + 2V̄i [1 − (−1)i e−βicl π cos γ i cl π ]
2πi cl
R̄ i Q̄ i

β (ωi + γ ) Q̄ i (ωi − γ ) R̄ i
− i cl − + 2γ (−1)i e−βicl π sin γ i cl π
π R̄ i Q̄ I
1 1
A¯i = 2 4 {V̄i [(−1)i e−βicl π cos γ i cl π − 1]
π ωi + 2ωi (β − 152 ) + (β 2 + γ 2 )2
2 2
+ −2βγ (−1)i eβicl π sin γ i cl π }

P̄i = β 2 − (ωi + γ )2 , Q̄ i = β 2 + (ωi − γ )2
R̄ i = β 2 + (ωi + γ )2 , T̄ i = β 2 − (ωi − γ )2
Ū i = ωi2 + β 2 + γ 2 , V̄i = ωi2 + β 2 − γ 2
ωi = i/i cl (3.146)
In order to compare the results of the Monte Carlo method with those obtained by the
application of formulas (3.106) and (3.107), the following shell was used: R = L =
101.6 mm, t = 0.944931 mm and ν = 0.3, resulting in i cl = 6. This shell, with the
initial imperfection auto-covariance function used by Amazigo [see Equation (3.108)
with β = 0.2 and γ = 1.0], will be referred to as the Amazigo shell. Figure 3.13 shows
the normalized variance of the initial imperfections back-calculated with the aid of
the σij s given in Equation (3.109). Twelve terms have been retained in the summation
given by Equation (3.118). The uniformity is disturbed in the vicinity of the edges
due to Gibb’s phenomenon. The auto-covariance function possessed by 100 simulated
Amazigo shells is given in Figure 3.14(a), and the elements of the variance-covariance
Figure 3.13 Recalculated variance of Amazigo’s shell with exponential-

cosine auto-correlation function (after Elishakoff and Arbocz, 1982; Copy-
right
c Elsevier Science Ltd., reprinted with permission).
matrix are shown in Figure 3.14(b). Since i cl = 6, it is seen that, on the diagonal
at i = j = i cl , the elements of the variance-covariance matrix reach a maximum. This
maximum corresponds to the well-known peak in the spectral density [Equation (3.109)]
used by Amazigo in his analysis of the infinite shell.
Notice that, for sufficiently long shells, the number of half-waves associated with
the classical buckling mode i cl 1. Then ωi tends to the continuous spatial frequency
ω, and in the expression for i [see Equation (3.146)] the second and the third terms
become vanishingly small in comparison with the leading term β Ū i . Thus, as L → ∞,
σii in Equation (3.145) reduces to the spectral density S̄ w0 (ω) used by Amazigo for the
infinite shell [see Equation (3.109)]. The extra factor of 4 is due to the fact that the
half-wave cosine representation was employed for ϕi (x) in deriving Equation (3.145)
instead of the complex representation used by Amazigo. Further, as can be seen from
Figure 3.14(b), some off-diagonal terms σij (i = j) are different from zero. However, it
can easily be shown using the expression given in Equation (3.146) that as L → ∞ the
off-diagonal terms approach zero.
When comparing the results of the Monte Carlo method for the mean buckling
load with the prediction based on Equations (3.106) and (3.107), the critical value of
the maximum absolute difference between the unknown √ theoretical and the obtained
simulated distributions of the buckling loads is 1.36/ 100 = 0.136 according to the
Kolmogorov-Smirnov test of goodness of fit (see Massey, 1951) at a level of significance
of 0.05. The variance of the simulated initial imperfections was fixed at δ 2 = 0.005. The
histogram of the buckling loads and the reliability function are shown in Figures 3.15(a)
and 3.15(b), respectively. The non-dimensional buckling loads λ were distributed be-
tween 0.376 and 0.886 so that the design buckling load at the required reliability, 0.98
say, equals 0.37. Obviously, the design buckling load associated with the high level of
required reliability is a more powerful design criterion than the mean buckling load.
Figure 3.14 Statistical properties of the simulated Amazigo shells (after Elishakoff and
Arbocz, 1982; Copyright c Elsevier Science Ltd., reprinted with permission).
The mean buckling load possessed by the “created” shells is 0.608. Formula (3.106)
by Amazigo (1969) predicts 0.468 and formula (3.107) by Amazigo and Budiansky
(1972) yields 0.602. As is seen, the mean buckling load given by Equation (3.107) is
much more reliable than the one predicted by Equation (3.106). This is in agreement
with the conjecture of Amazigo and Budiansky (1972).
It should be remarked here that in the present work no use was made of the ergodicity
assumption adopted by Amazigo and Budiansky (1972), but ensemble averaging was
employed to find the characteristics of the buckling load, which turns out to be a
random variable. Also, for the Monte Carlo simulation, a shell of finite length was used,
not an infinite one as in the papers of Amazigo (1969) and Amazigo and Budiansky
(1972).
Figure 3.15 The reliability function associated with the simulated Amazigo
shells (after Elishakoff and Arbocz, 1982; Copyright
c Elsevier Science Ltd.,
reprinted with permission).
3.4.4 Simulation of Random Initial Imperfections from Measured Data

The techniques for measuring the initial imperfections of shells are well established
(see, for example, Arbocz and Babcock, 1968) and the results are collected in the pa-
pers of Singer, Abramovich, and Yaffe (1978) and Arbocz and Abramovich (1979). The
techniques employed measure the deviation of the shell outer surface relative to an imag-
inary cylindrical reference surface (the “perfect shell” associated with the imperfect
shell under consideration). Consider, for example, cylindrical shells of L = 176.02 mm,
R = 101.6 mm, and t = 0.1160 mm, manufactured by electroplating from pure cooper
and tested in a controlled end-displacement type compression-testing machine. We can
(m)
Table 3.1. First nine Fourier coefficients ai of the initial imperfections expanded
in the series
(m) (m)

8
(m) iπx
w 0 (x) = t W0 (x) = t ai cos
i=0
L
Shell
number, i A-7 A-8 A-9 A-10 A-12 A-13 A-14
0 0.0176 0.0343 0.0226 0.0108 0.0023 0.0018 0.0029
1 0.0669 0.6534 0.0832 −0.0231 0.0158 0.0242 0.0662
2 −0.0164 0.1033 −0.0437 −0.0265 −0.0164 0.0095 0.0041
3 −0.0176 −0.0696 −0.0079 0.0054 0.0274 0.0006 0.0237
4 −0.0403 −0.1997 −0.0519 −0.0232 0.0092 0.0048 0.0013
5 −0.0031 −0.1637 0.0015 −0.0055 −0.0194 −0.0021 −0.0438
6 −0.0313 −0.0787 −0.0347 −0.0187 0.0062 −0.0047 −0.0349
7 −0.0050 −0.0092 −0.0080 −0.0106 0.0115 0.0060 0.0042
8 −0.0326 −0.0821 −0.0370 −0.0158 −0.0116 0.0038 −0.0041
For the group of A-shells (R = 101.6 mm, t = 0.1160 mm, L = 176.02 mm, E = 1.0441×
(0)
105 N/mm2 , ν = 0.3, i cl = 30). Note: Here w0 (x) is positive outward.
visualize that a suitable stock of such shells, referred to as the A-shells (Arbocz and
Abramovich, 1979), is available. Due to the very nature of the manufacturing process,
each realization of the shell will have a different initial shape that cannot be predicted
in advance. The imperfections represent deviations of the initial shape from the perfect
circular cylinder amounting to a fraction of the wall thickness. They can be picked
up and recorded by the special experimental setup developed at Caltech (Arbocz and
Babcock, 1968). The scanning device, moving in both the axial and the circumferential
directions, yields a complete surface map of the shells. Any two shells produced by the
same manufacturing process may have totally different imperfection profiles, as can
be seen from the three-dimensional plot of initial imperfections, shown in Figure 3.16.
Measured imperfection surfaces are represented by different Fourier series. The inte-
grals involved in the determination of the Fourier coefficients are carried out numerically
(m)
using the trapezoidal rule. The axisymmetric Fourier coefficients ai of the different
(m)
A-shells are listed in Table 3.1. Now we are looking at the ai s as realizations of the
random variable Ai in Equation (116). Then the sample mean is estimated as
(e) 1 N
(m)
Ai = ai (3.147)
N m=1
where N is the number of sample shells.

The elements σij of the variance-covariance matrix are estimated as
(e) 1 N
(m) (e)
(m) (m)
σij = ai − Ãi a j − Ã j (3.148)

N − 1 m=1
which is an unbiased estimate (see Table 3.2).

Figure 3.16 Three-dimensional plots of measured initial imperfections (after

Elishakoff and Arbocz, 1982; Copyright
c Elsevier Science Ltd., reprinted with
permission).
The estimates of the mean initial imperfection function and of the auto-covariance
functions become, respectively [see Equations (3.116) and (3.118)]
(e)
(e)
W̃0 (x) = Ãi ϕi (x)
i
(e)
(3.149)
Rw(e)0 (x1 , x2 ) = σij ϕi (xi )ϕ j (x2 )
i j
Table 3.2. Elements of the variance-covariance matrix [Σ(e) ] for the group of A-shells
i/j
0 1 2 3 4 5 6 7
0 0.015 0.230 0.030 −0.035 −0.083 −0.046 −0.029 −0.008
1 0.230 5.530 1.031 −0.658 −1.644 −1.338 −0.556 −0.079
2 0.030 1.031 0.232 −0.113 −0.277 −0.270 −0.092 −0.005
3 −0.035 −0.658 −0.113 0.096 0.217 0.140 0.069 0.016
4 −0.083 −1.644 −0.277 0.217 0.533 0.375 0.180 0.039
5 −0.046 −1.338 −0.270 0.140 0.375 0.353 0.130 0.013
6 −0.029 −0.556 −0.092 0.069 0.180 0.130 0.074 0.016
7 −0.008 −0.079 −0.005 0.016 0.039 0.013 0.016 0.007
Each entry should be multiplied by a factor 10−2 .
(e)
Since [#] = [σij ] is a positive–semi-definite matrix, therefore, according to
Sylvester’s theorem (Chetaev, 1961), all principal minor determinants associated with
matrix [#] are non-negative. The same property must be possessed by the estimate
(e)
[# (e) ]. This property is used in order to “correct” initial values of σij . If, for example,
the r th-order principal minor determinant is non-negative
(e)
σ11 · · · σ1r(e)

..
. ≥0

σ (e) · · · σ (e)
rl rr

σ (e) · · · σ (e) #1,4+1
(e)
11 1r
.
..

(e) ≤0
σr 1 · · · σrr(e) σr,r +1
(e)

(e)
σr +1,1 · · · σr(e)+1,r σr +1,r +1
(e)
but the (r + 1)th-order principal minor determinant is negative

σ11 (e)
· · · σ
(e)
#
(e)
+ x
1r 1,4+1
..
.
≤0
σ (e) ··· σrr
(e)
σr,r +1 + x
(e)
r1
(e)
σ · · · σr +1,r + x σr +1,r +1 + x
(e) (e)
r +1,1 + x
(e)
This “correction” can be used if max j σ j,r +1 + x lies within the confidence interval for
(e) (e)
the element matrix σ j,r +1 , and, moreover, if x max j σ j,r +1 .
(e) (e)
Having the estimates of Ãi and σij , we are proceeding to the simulation of the
initial imperfections, as described in the previous section. Instead of { Ã} and [#] in
Equations (3.139) and (3.140), we use { Ã(e) } and [# (e) ], respectively. As a result we
obtain the desired number M > N initial imperfections profiles, which are statistically
“equivalent” to the initial sample N shells.
Figure 3.17 Elements of the variance-covariance matrix [# e] of the group of A

To check the simulation procedure, the auto-covariance function of the simulated

sample of shells, defined as
(s)
Rw(s)0 (x1 , x2 ) = σi j ϕi (x1 )ϕ j (x2 ) (3.150)
m n
with
(s) 1 M
(s)(m) (s)
(s)(m) (s)
σij = ai + Ãi a j − Ã j (3.151)

N − 1 m=1
must be compared with the auto-covariance function of the initial sample Rw(e)0 . The
variance-covariance matrix for the group of A-shells is given in the paper by Elishakoff
and Arbocz (1982) and its elements are displayed in Figure 3.17. The auto-covariance
function estimated from the measured data is displayed in Figure 3.18. The
Figure 3.18 The estimated auto-covariance function of the axisymmetric part of the
measured initial imperfections of the group of A shells (after Elishakoff and Arbocz,
1982; Copyright c Elsevier Science Ltd., reprinted with permission).
Figure 3.19 Estimated variance of

the measured initial imperfections as
a function of the axial coordinate
for the group of A shells (after El-
ishakoff and Arbocz, 1982; Copyright
c Elsevier Science Ltd., reprinted
with permission).
auto-covariance function of the simulated sample turns out to be indistinguishable

from that of the measured initial sample. The estimated variance of the measured initial
imperfection versus the axial coordinate is shown in Figure 3.19.
A group of B-shells (Arbocz and Abramovich, 1979) was also considered. These
shells were manufactured by putting pieces of thick-walled, seamless, brass tubes onto a
mandrel and then machining them to the desired wall thickness. The average dimensions
of this group of shells are L = 134.37 mm, R = 101.6 mm, t = 0.2007 mm, and
ν = 0.3. The auto-covariance function estimated from the measured data is displayed
in Figure 3.20. Also, in this case, the auto-covariance function of the simulated sample
showed excellent agreement with that of the measured initial sample. The elements of
the variance-covariance matrix are displayed in Figure 3.21. The estimated variance
of the measured initial data versus the axial coordinate is depicted in Figure 3.22.
As shown in Figures 3.19 and 3.22, the variances associated with A- and B-shells
are non-uniform. This implies that their initial imperfections form non-homogeneous
random fields. Note that i cl = 30 for the group of A-shells and i il = 17 for the group
of B-shells.
Figure 3.20 The estimated auto-covariance function of the axisymmetric part of the
measure initial imperfections for the group of B-shells (after Elishakoff and Arbocz,
Figure 3.21 Elements of the variance-covariance matrix [# e] of the group of B-

As can be seen from Figures 3.17 and 3.21, the Fourier coefficients of the axisym-
metric part of the initial imperfections “peak” near ω = 0 and become vanishingly small
near ω = 1. That means that the variance-covariance matrices of the shells investigated
are dominated by the lower order modes and not by the classical axisymmetric buckling
mode. Next we proceed to the second step of the Monte Carlo method: evaluation of
the buckling load for each “created” shell.
3.4.5 Computation of the Buckling Loads

The buckling loads are calculated by using Koiter’s special theory (Koiter, 1963). We
are considering a cylindrical shell with an axisymmetric initial imperfection of the
Figure 3.22 Estimated variance of the measured initial imperfections as a

function of the axial coordinate for the group of B-shells (after Elishakoff and
Arbocz, 1982; Copyright c Elsevier Science Ltd., reprinted with permission).
form
iπ x
w0 (x) = t ξ̄i cos (3.152)
L
where ξ̄i denotes the magnitude of the imperfection as a fractional value of the shell
thickness t and i is an integer denoting the number of half-waves in the axial direction.
Using Koiter’s special non-linear theory, one can derive a relationship between the non-
dimensional axial load level λ (at which the resulting fundamental equilibrium state
bifurcates into an asymmetric deformation pattern) and the imperfection amplitude ξ̄i .
If one assumes the buckling mode
kπ x ly
w(x, y) = tCki sin cos (3.153)
L R
where k and l are integers denoting the number of half-waves and the number of full
waves in the axial and in the circumferential direction, respectively, then the following
non-linear transfer function between λ and ξ̄i is obtained:

c βl2 8αk4
(λci − λ) (λcki − λ) + (λci + λ) 2 λ + λci
2
2 ξ̄i δij
2 αk αk2 + βl2

1 1
+ 8c αk βl
2 2 4
+ 2 λci ξ̄i = 0
2 2
(3.154)
2 2 2 2
9αk + βl
2
αk + βl
where
2
1 2 1 1 αk2 + βk2 αk2
λci = αi + 2 , λcki = + 2
2 αi 2 αk2 αk2 + βt2
(3.155)
Rt π 2 Rt π 2 Rt 1 2
αi2 = i 2 , αk2 = k 2 , βl2 = l 2
2c L 2c L 2c R
and δij is the Kronecker delta with j = 2k.

It can be shown that Equation (3.154) can be reduced to Equation (5.2) of Koiter
(1963). From Equation (3.154) for an imperfection-sensitive structure, ξ̄i must be neg-
ative and i must be an even integer. The use of Equation (3.154) to calculate the critical
buckling load for a given shell then proceeds as follows. Initially with i = 2, k = 1
(since i = 2k) and for different values of l, Equation (3.154) is solved repeatedly, and
the lowest bifurcation buckling load λ is determined. Next the process is repeated with
i = 4, 6, . . . , until all the available imperfection harmonics have been tested. The abso-
lute minimum bifurcation buckling load is then identified as the critical buckling load
for the shell under consideration.
Once the critical buckling loads for the large number of simulated shells (M = 100)
are available, one can then proceed to calculate the histogram of the buckling loads and
to determine the corresponding reliability function. Figures 3.23 through 3.26 display
the histograms of the buckling loads and the calculated reliability functions for the group
of A- and B-shells, respectively. A comparison of Figures 3.24 and 3.26 shows that
Figure 3.23 Histogram of the non-dimensional buckling loads λ for

the group of A-shells (after Elishakoff and Arbocz, 1982; Copyright
c
Elsevier Science Ltd., reprinted with permission).
the reliability of the B-shells is less than that of the A-shells, meaning that machining
thin-walled shells out of thick-walled, seamless, brass tubes is a “rougher” procedure
than electroplating.
Note that the mean buckling load for the simulated sample of A-shells is 0.946,
whereas that for the B-shells is 0.724. These loads are considerably higher than the
experimentally observed mean buckling loads for the corresponding initial samples
(0.643 for the A-shells and 0.592 for the B-shells). The reason is that, as was pointed out
by Arbocz and Babcock (1976), for accurate buckling load predictions, the asymmetric
imperfections must also be taken into account. This work could be viewed as the first
step toward a more general analysis. The investigation of the effect of general, non-
axisymmetric imperfections will be discussed in Section 3.5.
Figure 3.24 Calculated reliability function versus the non-dimensional

buckling load λ for the group of A-shells (after Elishakoff and Arbocz,
Figure 3.25 Histogram of the non-dimensional buckling loads λ for

the group of B-shells (after Elishakoff and Arbocz, 1982; Copyright
3.4.6 Comparison of the Monte Carlo Method

with the Benchmark Solution
At this point, a natural question arises: What is the accuracy of the Monte Carlo method?
In order to provide a check of this method when applied to shell structures, let us consider
a case amenable of an exact solution. If one assumes that the initial imperfections are
of the form
πx
w0 (x) = t ξ̄ cos i cl (3.156)
L
where ξ̄ is a random variable X̄ with given probability density f x̄ (ξ̄ ), then the shape
of the axisymmetric imperfection coincides with the axisymmetric buckling mode
of the perfect shell. Hence, λci = 1, αk2 = 14 (since i = 2k), and Equation (3.144)
Figure 3.26 Reliability function versus the non-dimensional load for

the group of B-shells (after Elishakoff and Arbocz, 1982; Copyright
c
reduces to

8
(1 − λ)2 (λckl − λ) + (1 − λ)2cβl2 λ + 2 ξ̄
1 + 4βl2

1 1
+ 32c βl4 + ξ̄ 2 = 0
2
(3.157)
9 + 4βl 2 2 1 + 4βl2
One can show that Equation (3.157) is equivalent to Equation (5.4) of Koiter (1963).
Equation (3.157) then represents the non-linear transfer function between λ and ξ . If
the amplitude of the imperfection ξ̄ is a random variable X̄ , then the buckling load λ
will also be a random variable $. Thus, Equation (3.157) becomes

8
(1 − λ) (λckl − $) − (1 − $)2cβl $ +
2 2
2 | X̄ |
1 + 4βl2

1 1
+ 32c2 βl4 2 + X̄ 2 = 0 (3.158)
9 + 4βl2 1 + 4βl2
The minus sign has been introduced (as discussed earlier) in order to obtain an
imperfection-sensitive structure. The reliability is then defined as the probability that
the buckling load $ will be greater or equal to some specified value α. It follows from
Equation (3.158) that this is equivalent to the probability that the absolute value of the
imperfection X̄ be less than or equal to some value ξ̄ ∗ , where ξ̄ ∗ is the smallest root of
the transfer function (3.158) for the specified value of $ = α.
It should be noted here that Equation (3.158) contains as a free parameter l, the
number of full waves in the circumferential direction. Thus, finding the smallest root ξ̄ ∗
for a given value of $ = α involves repeated solution of Equation (3.158) for different
values of l.
If we now introduce the distribution function Fx̄ (ξ̄ ), then by definition
ξ̄ ∗
Prob{ X̄ ≤ ξ̄ ∗ } = Fx̄ (ξ̄ ∗ ) = f x̄ (ξ̄ ) d ξ̄ (3.159)
−∞
and the reliability can be written as
R(α) = Prob{$ ≥ α}
ξ̄ ∗
∗
= Prob{| X̄ | ≤ ξ̄ } = f x̄ (ξ̄ )d ξ̄ = Fx̄ (ξ̄ ∗ ) − Fx̄ (−ξ̄ ∗ ) (3.160)
−ξ̄ ∗
Thus, the reliability equals the integral of the initial imperfection amplitude probability
density f x̄ (ξ̄ ) over the interval (−ξ̄ ∗ , ξ̄ ∗ ).
If one now further assumes that the random variable X̄ is normally distributed with
mean m x̄ and variance σx̄2 , then its probability density is given by
1
√ e[−(ξ̄ −m x̄ )/2σx̄ ]
2
f x̄ (ξ̄ ) = (3.161)
σx̄ 2π
and we get from Equation (3.161) the following closed form solution for the reliability:
∗ ∗
1 ξ̄ − m x̄ −(ξ̄ ∗ − m x̄ ) ξ̄ − m x̄
R(α) = erf √ − erf √ = erf √ (3.162)
2 σx̄ 2 σx̄ 2 σx̄ 2
where erf (x) is an error function defined as
x
2
e−t dt
2
erf(x) = √ (3.163)
2 0
The reliability calculation for this benchmark case is illustrated in Figure 3.27,
Figure 3.27 Illustration of the reliability calculation for the benchmark case (after
Elishakoff and Arbocz, 1982; Copyright c Elsevier Science Ltd., reprinted with
permission).
Figure 3.28 Histogram of the non-dimensional buckling loads λ

(after Elishakoff and Arbocz, 1982; Copyright
where the shaded area equals the reliability of the shell at the non-dimensional load
level α.
For the sake of comparison, 1089 imperfect shells were simulated. Their dimensions
were set at L = 141.0 mm, R = 101.6 mm, t = 0.2634 mm, and ν = 0.3 so that i cl =
16.0. The shape of the imperfections coincided with the classical axisymmetric buckling
mode, and their amplitudes were normally distributed random variables with a mean of
m x̄ = 0.1 and a standard deviation of σx̄ = 0.05. The histogram of the buckling loads
is shown in Figure 3.28. The buckling loads were computed following the procedure
outlined in the preceding section with Equation (3.157) in place of Equation (3.165).
The reliability functions are shown in Figure 3.29. The solid line indicates the analytical
solution from Equation (3.157). The results of the Monte Carlo method are shown by
circles. The agreement between the simulated and the analytical results is excellent.
Figure 3.29 Comparison of analytical reliability function with results

of the Monte Carlo method (after Elishakoff and Arbocz, 1982; Copy-
right
This section has demonstrated that the Monte Carlo method can be used success-
fully for the investigation of the stochastic imperfection sensitivity of axially com-
pressed cylindrical shells with axisymmetric initial imperfections. We believe that the
reliability function – the final product of such an analysis – represents a more powerful
design criterion than the ones based on deterministic or mean buckling load formulas.
Computation of the variance-covariance matrices of the Fourier coefficients as en-
semble averages of the experimentally determined values have shown that the measured
initial imperfections of finite shells are non-homogeneous, and hence non-ergodic. Thus
the ergodicity assumption used in many previous investigations is not appropriate for
realistic structures of finite length.
As has been seen, the results of the existing initial imperfection data banks can be
incorporated directly into the Monte Carlo method. Thus, the results presented in this
section further reinforce the need for compiling extensive experimental information on
imperfections classified according to the manufacturing processes.
Using the technique developed in the earlier papers (Elishakoff, 1978b, 1979b,
1980b; Elishakoff and Arbocz, 1982) and in this section (as well as in the following
section where the discussion has been extended to include general non-symmetric
imperfections), one can then calculate via Monte Carlo method the reliability functions
associated with different manufacturing processes. Thus, through this approach, the
imperfection-sensitivity concept may finally be introduced into the design procedure.

with Nonsymmetric Imperfections
This discussion of the reliability of axially compressed cylindrical shells with non-
symmetric imperfections is based on Elishakoff and Arbocz (1985).
3.5.1 Probabilistic Properties and Simulation of the Initial Imperfections

for a Finite Shell
Let us represent the initial imperfection functions as the following series:

N1
N2
N3
Wn (ξ, θ ) = Ai cos(iπ xi) + [Ckl sin(kπ ξ ) cos(lθ ) + Dkl sin(kπ ξ ) sin(lθ )]
i=0 k=l i=l
(3.164)
where
Wn (ξ, θ) x y
Wn (ξ, θ ) = , ξ= , θ= , 0 ≤ x ≤ L, 0 ≤ θ ≤ 2π
t L R
Wn (ξ, θ ) and Wn (ξ, θ) are dimensional and non-dimensional initial imperfections:
t, L, and R are the thickness, the length, and the radius of the shell, respectively;
x is the axial and y is the circumferential coordinate. Notice that in Equation (3.165)
the first summation represents the axisymmetric part of the initial imperfection pro-
file, whereas the second, a double summation, is associated with its non-symmetric
part. The axisymmetric part is expressed in the half-range cosine series, whereas the
non-symmetric part is represented by the half-range sine series so that the series (3.165)
sums up to the measured imperfection profile in the range 0 ≤ x ≤ L, 0 ≤ θ ≤ 2π .
The mathematical expectation of Wn (ξ, θ ) is given by

N1
N2
N3
E[w0 (ξ, θ )] = E(Ai ) cos(iπ ξ ) + [E(Ckl ) sin(kπ ξ ) cos(lθ )
i=0 k=l l=1
+ E(Dkl ) sin(kπ ξ ) sin(lθ)] (3.165)
where E(·) denotes a mathematical expectation.
The auto-covariance function becomes
C W0 (ξ1 , θ1 ; ξ2 , θ2 ) = E{[W0 (ξ1 , θ1 ) − E(W0 (ξ1 , θ1 ))][W0 (ξ2 , θ2 ) − E(W0 (ξ2 , θ2 ))]}
$

N1
=E (Ai − E(Ai )) cos(iπ ξ1 )
i=0

N2
N3
+ (Ckl − E(Ckl )) sin(kπ ξ1 ) cos(lθ1 )
k=1 l=1

N2
N3
+ (Dk l − E(Dk l)) sin(kπ ξ1 ) sin(lθ1 )
k=1 l=1

N1
× (A j − E(A j )) cos( jπ ξ2 )
j=0

N2
N3
+ (Cmn − E(Cmn )) sin(mπ ξ2 ) cos(nθ2 )
m=1 n=1
%

N2
N3
+ (Dmn − E(Dmn )) sin(mπ ξ2 ) sin(nθ2 ) (3.166)
m=1 n=1
For the sake of simplicity, we rewrite Equation (3.164) in an alternative way, replacing
the double summation in Equation (3.164) by a single summation

N1
N
W0 (ξ, θ ) = Ai cos(iπ ξ ) + [Cr sin(kr π ξ ) cos(lr θ ) + Dr sin(kr π ξ ) sin(lr θ )]
i=1 r =1
(3.167)
where the quantities indexed with r are chosen to ensure the equivalence of the two
series given by Equations (3.164) and (3.167) and N = N2 × N3 . The auto-covariance
function can be written as

N1
N1
C W0 (ξ1 , θ1 ; ξ2 , θ2 ) = K Ai A j cos(iπξ1 ) cos( jπ ξ2 )
i=0 j=0

N1
N
+ K Ai Cs cos(iπ ξ1 ) sin(ks π ξ2 ) cos(ls θ2 )
i=0 x=1

N1
N
+ K AiDs cos(iπ ξ1 ) sin(ks π ξ2 ) sin(ls θ2 )
i=0 s=0

N
N1
+ K Cr A j sin(kr π xu 1 ) cos(lr θ1 ) cos( jπ ξ2 )
r =1 j=0

N
N1
+ K Dr A j sin(kr πξ1 ) sin(lr θ1 ) cos( jπ ξ2 )
r =1 j=0

N
N
+ K Cr Cs sin(kr πξ1 ) cos(lr θ1 ) sin(ks π ξ2 ) cos(ls θ2 )
r =1 s=1

N
N
+ K Cr Ds sin(kr π ξ1 ) cos(lr θ1 ) sin(ks π ξ2 ) sin(ls θ2 )
r =1 s=1

N
N
+ K Dr Cs sin(kr π ξ1 ) sin(lr θ1 ) sin(ks π ξ2 ) cos(ls θ2 )
r =1 s=1

N
N
+ K Dr Ds sin(kr π ξ1 ) sin(lr θ1 ) sin(ks π ξ2 ) sin(ls θ2 )
r =1 s=1
(3.168)
where the variance-covariance matrices K Ai A j , . . . , etc. are defined as follows:
K Ai A j = E[((Ai − E(Ai ))(A j − E(A j ))],

K A j Cs = E[(A j − E(Ai ))(Cs − E(Cs ))]
K Ai Ds = E[A1 − E(Ai ))(Ds − E(Ds ))],
(3.169)
K Cr Cs = E(Cr − E(Cr ))(Cs − E(Cs ))]
K Cr Ds = E[(Cr − E(Cr ))(Ds − E(Ds ))],
K Dr Ds = E[(Dr − E(Dr )))(Ds − E(Ds ))]
If the auto-covariance function C W0 (ξ1 , θ1 ; ξ2 , θ2 ) is known, then these quantities can

be obtained as follows:
2π 1 2π 1
1
K Ai A j = C W0 (ξ1 , θ1 ; ξ2 , θ2 )
π2 0 0 0 0
× cos(iπ ξ1 ) cos( jπ ξ2 ) dξ1 dθ1 dξ2 dθ2
2π 1 2π 1
2
K Ai Cmn = 2 C W0 (ξ1 , θ ; ξ2 , θ2 ) cos(iπ ξ1 )
π 0 0 0 0
× sin(mπ ξ2 ) cos(nθ2 ) dξ1 dθ1 dξ2 dθ2
2π 1 2π 1
2
K Ai Dmn = 2 C W0 (ξ1 , θ1 ; ξ2 , θ2 ) cos(iπ ξ1 )
π 0 0 0 0
× sin(mπ ξ2 ) sin(nθ2 ) dξ1 dθ1 dξ2 dθ2
2π 1 2π 1
2
K A j Ckl = 2 C W0 (ξ1 , θ1 ; ξ2 , θ2 ) cos( jπ ξ2 )
π 0 0 0 0
× sin(kπ ξ1 ) cos(lθ1 ) dξ1 dθ1 dξ2 dθ2
2π 1 2π 1
2
K A j Ckl = 2 C W0 (ξ1 , θ1 ; ξ2 , θ2 ) cos( jπ ξ2 )
π 0 0 0 0
× sin(kπ ξ1 ) sin(lθ1 ) dξ1 dθ1 dξ2 dθ2
2π 1 2π 1
4
K Ckl Cmn = 2 C W0 (ξ1 , θ1 ; ξ2 , θ2 ) sin(kπ ξ1 ) cos(lθ1 )
π 0 0 0 0
2π 1 2π 1
4
K Dkl Cmn = 2 C W0 (ξ1 , θ1 ; ξ2 θ2 ) sin(kπ ξ1 ) sin(lθ1 )
π 0 0 0 0
2π 1 2π 1
4
K Dkl Dmn = 2 C W0 (ξ1 , θ1 ; ξ2 , θ2 ) sin(kπ ξ1 ) sin(lθ1 )
π 0 0 0 0
× sin(mπ ξ2 ) sin(nθ2 ) dξ1 dθ1 dξ2 dθ2 (3.170)
Notice that if W0 (ξ, θ) is constant and
K Ai Cs = K Ai Ds = K Cr A j = K Dr A j = K Cr Ds = K Dr Cs = 0 (3.171)
for any combination of indices and, moreover,

K Cr Cs = K Dr Ds
(3.172)
K Cr Cs = K Cr Cs δlr ls
where δlr ls is a Kronecker delta, then the initial imperfection is a weakly homogeneous
function in the circumferential direction. Under the conditions (3.171) and (3.172), the
auto-covariance function takes the form

N1
N1
C W0 (ξ1 , θ1 ; ξ2 , θ2 ) = K Ai A j cos(iπξ1 ) cos( jπ ξ2 )
i=0 j=0

N
N
+ K Cr Cs sin(kr π ξ1 ) sin(ks π ξ2 ) cos[lr (θ2 − θ1 )]
r =1 s=1
(3.173)
(i.e., it depends on θ2 − θ1 rather than on θ1 and θ2 separately). It can be shown that

Equations (3.172) and (3.173) represent not only the sufficient conditions but also
the necessary ones. One could argue that for the closed, nominally circular, seamless
cylindrical shell the probabilistic properties would not be affected by a shift of the origin
of the coordinate axes in the circumferential direction. Interestingly, in the experiments
performed by Makarov (1971), this weak homogeneity was preserved even for the series
of shells with pronounced seams. Due to the frequent use of this property, we will first
show how to simulate the initial imperfections possessing weak homogeneity in the
circumferential direction.
To simulate the large number of initial imperfection profiles needed for the Monte
Carlo method, first the mean values and the variance-covariance matrices of the mea-
sured initial imperfections must be determined. This involves the evaluation of the
following ensemble averages for a sample of experimentally measured initial imper-

fections:
(e) 1 M
(m) (e) 1
M
Ãi = Ai ; C̃ r = C (m)
M m=1 M m=1 r
(e) 1 M
(m) (e)
(m) (e)
K Ai A j = Ai − Ãi A j − Ã j (3.174)
M − 1 m=1
(e) 1 M
(m) (e)
(e)
K Cr C s = Cr − C̃ r Cs(m) − C̃ s
M − 1 m=1
where M is the number of sample shells, and m is the serial number of the shells. The
variance-covariance matrices are positive–semi-definite and can be uniquely decom-
posed in the form
(e)
(e)
K Ai A j = [G][G]T , K Cr Cs = [G ][G ]T (3.175)
where T means transpose and [G], [G ] are lower triangular matrices found by the
Cholesky decomposition algorithm. Now we form the random vectors [B] and [B ],
the elements of which are normally distributed and statistically independent with zero
means and unit variance. Then the vectors of the Fourier coefficients of the initial
imperfections are simulated as follows:
(e) (e)
[A] = [G]{B} + { Ã }, {C} = [G ]{B } + {C̃ } (3.176)
Having the desired large number of realizations of the vectors {B} and {B }, one obtains
the same number of realizations of {A} and {C}. The main feature of this simulation
technique (Elishakoff, 1979b) is that it is applicable for homogeneous as well as non-
homogeneous random functions with given mean and auto-covariance functions.
Equation (3.176) represents the simulated vectors {A} and {C} for the random
imperfections, weakly homogeneous in the circumferential direction. For the imper-
fections that form a general non-homogeneous random field, the refined simulation
procedure (Elishakoff, 1988) has to be utilized. The essence of this refinement is the
replacement of the multiple summations in Equations (3.165) and (3.167) by a single
“string” and the dealing with the resultant mixed series.
3.5.2 Multi-Mode Deterministic Analysis for Each Realization

of Random Initial Imperfections
The buckling load for each created shell is then calculated by the so-called multi-mode
analysis (Arbocz and Babcock, 1974), which allows the incorporation of imperfection
shapes in the form of the double Fourier series given in Equation (3.164). By definition,
the value of the loading parameter λ corresponding to the limit point of the prebuckling
states is the theoretical buckling load. The number of modes of deformation included
in the analysis is limited by practical considerations, like the available core size and
the time required for obtaining the solution. Thus, since the shell buckling load will be
determined by solving the governing equations for a particular set of modes, an attempt
Figure 3.30 Modified 15-mode imperfection model used for

the analysis by the Monte Carlo method (after Elishakoff and
Arbocz, Journal of Applied Mechanics, 1985).
at optimizing the selection of these modes must be made. That is, it is necessary to
locate those modes that dominate the prebuckling and buckling behavior of the shell.
Previous investigations by Arbocz and Babcock (1974, 1976) have shown that to yield a
noticeable decrease from the buckling load of the perfect shell, the initial imperfection
harmonics used must include at least one mode with a significant initial amplitude
and an associated eigenvalue that is close to the buckling load of the perfect shell.
Furthermore, if the modes are so selected that the non-linear coupling conditions are
satisfied, then the resulting buckling load of the shell generally will be lower than the
buckling loads obtained with each mode considered separately.
Based on these considerations and the results of Arbocz and Babcock (1974), we
choose the imperfection model shown in Figure 3.30 for the Monte Carlo simulation. In
this model A2.0 stands for a half-wave cosine axisymmetric Fourier coefficient, with two
half-waves in the axial direction and no waves in the circumferential direction. On the
other hand, C1.10 stands for an asymmetric Fourier coefficient with a single half-wave
in the axial direction and 10 full waves in the circumferential direction.
As pointed out by Arbocz and Babcock (1976), the chosen imperfection model
requires imperfection amplitudes at wave numbers that were not measured. This is due
to the fact that in the early experimental work the mesh-spacing used was not suffi-
ciently close to resolve all the harmonic amplitudes of interest. Therefore, the Donnell-
Imbert imperfection model was fitted over the wave numbers actually measured, and
then the amplitudes of the harmonics of interest were obtained by extrapolation. It
should be stressed here that the averaged (in the axial direction short wavelength)
(m)
Table 3.3. First 15 Fourier coefficients ai of the initial
imperfections expanded as
(m) (m)

14
(m) iπx
w 0 = t W0 (x) = t ai cos
i=0
L
Shell
number, i B-1 B-2 B-3 B-4
0 0.0010 0.0028 0.0111 0.0080
1 −0.0333 −0.1889 −0.6231 0.1096
2 −0.0108 −0.0272 −0.0899 −0.0176
3 −0.0190 −0.0276 −0.0803 0.0407
4 0.0226 −0.0078 −0.0255 −0.0092
5 −0.0025 −0.0097 −0.0230 0.0132
6 0.0018 −0.0049 −0.0223 −0.0059
7 −0.0062 −0.0080 −0.0189 0.0120
8 0.0076 −0.0074 0.0052 −0.0125
9 −0.0050 −0.0023 0.0053 0.0254
10 −0.0013 −0.0041 −0.0133 −0.0265
11 0.0015 −0.0007 −0.0033 0.0200
12 −0.0062 −0.0009 −0.0085 0.0176
13 0.0084 −0.0031 0.0092 0.0103
14 0.0020 −0.0027 −0.0136 −0.0114
For the group of B-shells (R = 101.6 mm, t = 0.2007 mm, L = 134.30 mm,
(m)
E = 1.065 × 105 N/mm2 , ν = 0.3, i cl = 17). Note: Here w0 (x) is positive outward.
modes of the imperfection model shown in Figure 3.30 must be included in the anal-
ysis to satisfy the non-linear coupling conditions. Their initial amplitudes are actually
insignificant.
For the purpose of the Monte Carlo method, the MIUTAM code (see Arbocz
and Babcock, 1976) was incorporated into a new program, which then one by one
automatically starts the calculations for the simulated imperfections and at the end lists
all the buckling loads obtained.
3.5.3 Numerical Results and Discussion

The procedure described in the previous sections was applied to the group of shells,
referred by Arbocz and Abramovich (1979) as B-shells. These shells were originally
cut from thick-walled seamless brass tubing; the pieces were mounted on a mandrel and
the outer surface was machined to the desired dimensions. The geometric and material
properties of the B-shells are summarized in Table 3.3, whereas Table 3.4 lists elements
of the variance-covariance matrix.
As is seen from Figure 3.30, the simulation procedure was applied only to the
eight lower order modes, namely A2,0 , A4,0 , C1,2 , C1,6 , C1,8 , C1,10 , C2,3 , and C2,11 .
The remaining higher order modes, namely A26,0 , C24,3 , C24,11 , C25,2 , and C25,10 were
Table 3.4. Elements of the variance-covariance matrix [Σ(e) ] or the group of B-shells.
i/j
(m)
[Σ ] 0 1 2 3 4 5 6 7 8 9 10
0 0.002 −0.079 −0.013 −0.006 −0.008 −0.002 −0.004 −0.001 −0.0005 0.003 −0.004
1 −0.079 10.059 1.095 1.489 0.398 0.454 0.279 0.364 −0.160 0.145 −0.070
2 −0.013 1.095 0.132 0.145 0.057 0.044 0.036 0.033 −0.013 0.003 0.005
3 −0.006 1.489 0.145 0.247 0.035 0.074 0.033 0.063 −0.034 0.041 0.028
4 −0.008 0.398 0.057 0.035 0.040 0.012 0.018 0.006 0.006 −0.012 0.012
5 −0.002 0.454 0.044 0.074 0.012 0.023 0.010 0.019 0.009 0.012 0.008
6 −0.004 0.279 0.036 0.033 0.018 0.010 0.010 0.007 0.002 −0.002 0.004
7 −0.001 0.364 0.033 0.063 0.006 0.019 0.007 0.016 −0.009 0.012 0.009
8 −0.0005 −0.160 −0.013 −0.034 0.006 −0.009 −0.002 −0.009 0.009 −0.010 0.007
9 0.003 0.145 0.003 0.041 −0.012 0.012 −0.002 0.012 −0.010 0.017 0.014
10 −0.004 −0.070 0.005 −0.028 0.012 −0.008 0.004 −0.009 0.007 −0.014 0.013
Each entry should be multiplied by a factor 10−2 .
Figure 3.31 (a) Elements of cross-correlation matrix K A i A j for simulated group of 500
B-shells, and (b) corresponding part of variance (after Elishakoff and Arbocz, Journal of
Applied Mechanics, 1985).
obtained by extrapolation from the corresponding Donnell-Imbert imperfection model

(Arbocz and Babcock, 1976; Imbert, 1971).
Figure 3.31 shows the elements of the variance-covariance matrices K Ai A j and the
corresponding part of the variance, denoted by σ A2 (ξ )

N1
σ A2 (ξ ) = K Ai A j δi j cos(πiξ ) cos(π jξ ) (3.177)
i=0 j=0
Figures 3.32 and 3.33 show the elements of the variance-covariance matrices K Cr Cs
and K Dr Ds , respectively, and the associated parts of the variance, denoted by σC2 (ξ, θ )
and σ D2 (ξ, θ ):

N
N
σC2 (ξ, θ ) = K Cr Cs sin(kr π ξ ) sin(ks πξ ) cos(lr θ ) cos(ls θ ) (3.178)
r =1 s=1

N
N
σ D2 (ξ, θ ) = K Dr Ds sin(kr π ξ ) sin(ks πξ ) sin(lr θ ) sin(ls θ ) (3.179)
r =1 s=1
Figure 3.34 portrays the probabilistic characteristics of the 500 simulated shells.
Figure 3.35(a) shows the mean imperfection function, whereas Fig. 3.35(b) displays the
Figure 3.32 (a) Elements of cross-correlation matrix K C r C s for simulated group of B-

shells, and (b) corresponding part of variance (after Elishakoff and Arbocz, Journal of
variance. As is seen from Figure 3.35 neither the mean function nor the variance are
constant in the circumferential direction, implying that the random imperfections do not
consitute a circumferentially homogeneous field. This conclusion can also be deduced
from examination of values K Cr Cs and K Dr Ds , which reveals that not only do the corre-
sponding elements of these matrices not coincide but a ratio between them also may well
exceed 10. Thus, the homogeneity assumption adopted in the work of Amazigo (1974)
and other investigators turns out to be unjustifiable even for seamless shells.
To calculate the reliability functions shown in Figure 3.36, the following dimen-
sions corresponding to shell B-1 (Arbocz and Abramovich, 1979) were used: length
of 196.85 mm, radius of 101.60 mm, and thickness of 0.205 mm. In addition, for the
buckling load calculations of shells with axisymmetric imperfections, the one-sided
transfer-function shown in Figure 3.37 was chosen. This was done due to the fact,
pointed out by Babcock and Sechler (1962), that for a finite length shell the buckling
load is sensitive only to those axisymmetric imperfections that point inward at the
mid-plane of the shell (at x = L/2).
In Figure 3.36, curve 1 represents the reliability function for the case of purely
axisymmetric imperfections (with an estimated mean buckling load of 0.935), whereas
curve 2 shows the reliability function for the 15-modes non-symmetric imperfection
mode (with an estimated mean buckling load of 0.739). As can be seen, the reliability
estimate depends strongly on the number of terms taken into account (i.e., it is sensitive
Figure 3.33 (a) Elements of cross-correlation matrix K D r D s for simulated group of 500
B-shells, and (b) corresponding part of variance (after Elishakoff and Arbocz, Journal of
to the adequacy of the underlying deterministic model). As a check on the correctness

of the simulated results the 15-modes imperfection model shown in Figure 3.30 was
used to calculate the collapse loads of the original four B-shells involved, using the ex-
perimentally measured initial imperfections in place of the simulated random variables.
These computations yielded for the shells B-1, B-2, B-3, and B-4, respectively, the fol-
lowing collapse loads 0.751, 0.746, 0.740, and 0.781 with a mean of 0.756. Closeness
to the simulated results is remarkable. It should also be mentioned here that the theo-
retical collapse load of ρs = 0.66, reported in Arbocz and Babcock (1976) for the shell
B-1, was entirely based on the Donnell-Imbert imperfection model. Considering the
results shown in Figure 3.36 further, one sees that the inclusion of some asymmetric
imperfection components has reduced the estimate of the mean buckling load consider-
ably, though it is still higher than the experimental mean buckling load for the group of
B-shells of 0.592 (Arbocz and Abramovich, 1979). However, further refinements in the
non-symmetric random imperfection model may lead to lower simulated buckling loads.
In conclusion, one may summarize the results obtained in this section so far as
follows:
1. It has been demonstrated that the Monte Carlo method can be used successfully to
obtain reliability functions for shells with axisymmetric as well as non-symmetric
imperfections.
2. It has been found that for shells of finite length, non-homogeneous probabilistic
characteristic must be used (thus ergodicity assumption is not applicable).
3. Using the simulation procedure developed, the measured initial imperfections have
been used directly to generate input for the Monte Carlo method.
It is hoped that these preliminary results will encourage many investigators all
over the world to compile extensive experimental information on initial imperfections
classified according to the manufacturing procedures. The existence of these initial
imperfection data banks will make it possible to associate statistical measures with the
Figure 3.34 Three-dimensional plots of initial imperfection profiles of selected sim-

ulated shells (after Elishakoff, 1988; Copyright
with permission).
Figure 3.35 Probabilistic characteristics of simulated group of 500 B-shells; (a) mean
function, (b) variance (after Elishakoff and Arbocz, Journal of Applied Mechanics, 1985).
Figure 3.36 Reliability functions for simulated group of 500 B-

shells (after Elishakoff and Arbocz, Journal of Applied Mechanics,
1985).
Figure 3.37 One-sided transfer function used for axisymmetric im-

perfections only (for details, consult Babcock and Sechler, 1962)
(after Elishakoff and Arbocz, Journal of Applied Mechanics, 1985).
different methods of fabrication. As outlined here, the variance-covariance matrices and

the mean vectors can be used effectively to generate input for the Monte Carlo method,
which in turn yields the reliability functions associated with the different manufacturing
processes. It is felt that by this means the imperfection-sensitivity concept can be
finally introduced routinely into the design procedures since the Monte Carlo method
described in this chapter seems to offer the means of combining the lower bound
design philosophy with the notion of quality class depending on the manufacturing
process. Thus, shells manufactured by a process, which produces inherently a less
damaging initial imperfection distribution, will not be penalized because of the lower
experimental results obtained from shells produced by another process, which may
generate a more damaging characteristic initial imperfection distribution. Compilation
and extensive analysis of the international initial imperfection data banks is actively
pursued at the Delft University of Technology by the group led by Professor J. Arbocz
(Arbocz, 1981, 1982a, 1982b; Arbocz and Abramovich, 1979; Klompé, 1986, 1988,
1989; Klompe and Den Reyer, 1989) and at the Israel Institute of Technology by the
group led by Professor J. Singer (Singer et al., 1978; Yaffe, Singer, and Abramovich,
1981; Abramovich, Singer, and Yaffe, 1981), and by other investigators. These data
banks provide unique material for investigators to further contribute to the explanation
of the perplexing behavior of cylindrical shells and to develop novel design methods of
imperfection-sensitive structures. These data banks can be directly incorporated also
for the finite element analysis of shells with stochastic imperfections and/or material
properties.
CHAPTER FOUR
Stochastic Buckling of Structures: Analytical

and Numerical Non–Monte Carlo Techniques
Analytical procedures should be consistent with the complexity and importance

of the structure. Judgement and experience may be guides as reliable as analytical
procedures based on simplifying assumptions.
E. E. Sechler
It is remarkable that a science which began with the consideration of games of

chance should have become the most important object of human knowledge.
P. Laplace
Probabilistic investigations are essential for any real progress in the design of
imperfection-sensitive structures.
G. Augusti and A. Baratta
In this chapter, we continue our discussion on stochastic buckling of structures. However,

instead of resorting to the Monte Carlo method – the only universal technique to deal
with highly non-linear stochastic boundary value problems – we resort to approximate
analytical techniques. These techniques, although lacking versatility, are still useful (a) to
get additional insights into the problem, (b) to collaborate results obtained via the Monte
Carlo method, and (c) to devise simplified engineering methods for analysis and design of
engineering structures.
4.1 Asymptotic Analysis of Reliability of Structures in Buckling Context

The stochasticity of initial imperfections of structures has been given considerable
attention. As first postulated by Bolotin (1958), the buckling load $c of a structure
can be expressed as a function of a number of stochastic parameters X i (i = 1, . . . , p)
representing the initial imperfections:
$c = φ(X 1 , . . . , X p ) (4.1)
Since X i are random variables, the critical buckling load turns out to be a random
175
176 STOCHASTIC BUCKLING OF STRUCTURES
variable. Hence, uppercase notations are used for both X i and $c . The evaluation
of the probability density of $c through the use of Equation (4.1) entails two major
difficulties: (a) the probability densities of all initial imperfections X i (i = 1, . . . , p)
are difficult to obtain; (b) the function φ usually is a complicated non-linear function,
obtainable only in the form of the sophisticated numerical code.
In order to tackle these difficulties, extensive research has been conducted (see,
for example, the bibliography of Roorda, 1972b, and Amazigo, 1976). The initial im-
perfection data bank has been developed by measuring the initial imperfections of
shells (Arbocz and Abramovich, 1979; and Abramovich et al., 1981). Elishakoff and
Arbocz (1982) investigated the effect of random axisymmetric imperfections on the
buckling of circular cylindrical shells under axial compression. The stochastic proper-
ties of the shells were evaluated based on the measured data, and the reliability function
of the shells was computed by means of the Monte Carlo method. Later on, on the
basis of an assumption that the initial imperfections are represented by normally dis-
tributed random variables, the first-order second-moment method was employed, in
addition to the Monte Carlo method, to greatly reduce computational costs (see, for
example, Elishakoff et al., 1987; Arbocz and Hol, 1991). To sum up, the assumption of
Gaussianity of imperfections has been successfully used to resolve the first difficulty.
In the case of imperfections that do not follow normal distribution, the use should be
made of methods pertinent to non-Gaussian random variables and functions (Grigoriu,
1995).
A series of asymptotic theories on initial imperfections has been developed to tackle
the second difficulty. Roorda and Hansen (1972) extended Koiter’s theory (1945) to a
single mode normally distributed initial imperfection. The critical initial imperfection
vector, which achieves the steepest decline of critical load under the constraint of
a constant norm, was explicitly obtained (Ikeda and Murota, 1990a, 1990b; Murota
and Ikeda, 1991). Based on results of the latter work, with an initial imperfection
vector being assumed to be randomly distributed under the constraint of a constant
norm, explicit forms of the probability density function of critical loads were obtained
(Ikeda and Murota, 1991b; Murota and Ikeda, 1992). As the logical sequel to this, a
theoretical method of deriving reliability of structures with normally distributed initial
imperfections was presented by Ikeda and Murota (1993).
This method, which combines the simplicity of both the normal distribution and
the asymptotic theory, appears to be a promising way to overcome the aforementioned
two difficulties, associated with Bolotin’s postulate in the form of Equation (4.1). The
method by Ikeda and Murota (1993) is applied here to the results of the stochastic
studies on realistic structural models. including the buckling of a stochastically im-
perfect column on a non-linear elastic foundation (Elishakoff, 1979a) and buckling
of the axially compressed cylindrical shells with random axisymmetric imperfections
(Elishakoff and Arbocz, 1982). The initial imperfections on columns were assumed to
be normally distributed. The column problem is re-visited here to ensure the validity
of the present method, whereas the shell problem is studied to assess applicability of
the asymptotic method to realistic structures. This section closely follows the study by
Ikeda, Murota, and Elishakoff (1995).
4.1 ASYMPTOTIC ANALYSIS OF RELIABILITY OF STRUCTURES IN BUCKLING CONTEXT 177
4.1.1 Overview of the Work by Ikeda and Murota

We will start the formulation of the present theory by summarizing the study by Ikeda
and Murota (1993). A critical point may appear in a different type depending on the
specific structural and loading condition; three types of simple critical points, namely,
limit, asymmetric bifurcation, and unstable-symmetric bifurcation points are discussed.
We consider a system of non-linear equilibrium equations
H(λ, u, v) = 0 (4.2)
where λ denotes a loading parameter, u indicates an N -dimensional nodal displacement
(or position) vector, and ν is a p-dimensional imperfection pattern vector. We assume
H to be sufficiently smooth. For a fixed ν, a set of solutions (λ, u) of the preceding
system of equations makes up equilibrium paths. Let (λc , u c ) = (λc (ν), u c (ν)) denote
the first critical point on the main path of engineering interest [(·)c refers to the critical
point], governing the critical load (defined as the buckling load) of the structure with
an imperfection pattern vector ν. In particular, (λ0c , u 0c , ν 0 ) denotes the critical point of
the perfect system (a quantity with superscript 0 refers to the perfect system).
We write
ν = ν 0 + X, λc = λ0c + λ̃c (4.3)
where (>0) denotes the magnitude of the initial imperfection, X = (X 1 , . . . , X p )T is
a vector indicating the pattern of initial imperfections, and λ̃c means the increment of
the critical load.
We are interested in the stochastic behavior of the critical load $c for random
imperfections ν. To be more specific, we consider the case where the imperfection
pattern vector X represents a multi-variate normally distributed vector N (0, W −1 ) with
zero mean and variance-covariance matrix characterized by a positive definite matrix
W −1 . The analyses are asymptotic in the sense that they are valid only when is small.
As has been made clear by Koiter (1945), the asymptotic behavior for the increment
(increase or decrease) λ̃c of the critical load λc of the imperfect system is known to be
expressed as
λ̃c = λc − λ0c ∼ C0 a ρ ρ (4.4)
when is small. Here C0 is positive constant, and ρ varies with the type of points as
follows:

ρ = 1, at limit point
ρ = 12 , at asymmetric bifurcation point (4.5)

ρ = 3,
2
at unstable-symmetric bifurcation point
The variable a, which we call the “effective initial imperfection,” depends on the im-
perfection pattern vector X through

p
a = cT X = ci X i (4.6)
i=1
with some constant vector c = (c1 , . . . , c p )T . The variable a of Equation (4.6), which
is a sum of normally distributed variables ci di (i =" 1, . . ."
, p), is itself normally dis-
p p
tributed N (0, σ̃ 2 ) with zero mean and variance σ̃ 2 = i=1 j=1 ci c j cov (X i , X j ). The
probability density function f a (a) of a is given as

1 a2
f a (a) = √ exp − 2 , −∞ < a < ∞ (4.7)
2π σ̃ 2σ̃
With the use of (4.7), the probability density function of the critical load $c is
evaluated as follows:

 da

 f a (a) = g1 (λc ), −∞ < λc < ∞, at limit point

 dλc




 da at asymmetric
2 f a (a) = 2g1/2 (λc ), −∞ < λc < λ0c ,
f $c (λc ) = dλc bifurcation point



 at unstable-

 da

2 f a (a) = 2g2/3 (λc ), −∞ < λc < λ0c , symmetric

 dλc
bifurcation point
(4.8)
where
& '
λc − λ0 1/ρ−1 −1 λc − λ0c 2/ρ
c
gρ (λc ) = √ exp (4.9)
2π (C0 σ ρ )1/ρ 2 C0 σ ρ
and σ = σ̃ . The mean E[$c ] and the variance Var[$c ] of $c are expressed respectively
as
E[$c ] = λ0c + E[ζ ]C0 σ ρ , Var[$c ] = Var[ζ ](C0 σ ρ )2 , (4.10)
where ζ = λ̃c /(C0 σ ρ ) is now a normalized critical load increment. Its mean E[ζ ] and
variance Var[ζ ] for various kinds of simple critical points are listed in Table 4.1.
Table 4.1. E[ζ] and Var[ζ] for various kinds of simple critical
points [Γ(x) is the gamma function]
Type of critical points E[ζ] Var[ζ]

Limit point 0 12
−23/4 3
Asymmetric bifurcation point √ Γ = −0.822 (0.349)2
2π 4

−25/6 5
Symmetric bifurcation point √ Γ = −0.802 (0.432)2
2π 6
Integration of (4.8) leads to the reliability function:


 λ − λ0c

 1− , −∞ < λ < ∞, at limit point

 C0 σ



 2

 λ − λ0c at asymmetric
1 − 2 − , −∞, λ < λ0c ,
R(λ) = C σ 1/2 bifurcation point


0



 λ − λ0c 3/2 at unstable-



1 − 2 − , −∞, λ < λ0c , symmetric bifur-
 C0 σ 2/3
cation point
(4.11)
Here
ζ 2
1 ζ
(ζ ) = √ exp − dζ (4.12)
−∞ 2π 2
denotes the cumulative probability distribution function of standard normal variable.
It is to be emphasized here that by the present method a mere calculation of the
sample mean E[$c ] and variance Var[$c ] of the critical loads will yield the parameters
λ0c and C0 σ ρ in (4.9), and, in turn, the probability density function in (4.8) and the
reliability function in (4.11). The present method thus is quite simple. As we have seen
in (4.8) and (4.11), the forms of probability density function and reliability function
vary with the type of critical points. The present method, which is based on a firm
theoretical background, can implement such a variation in a systematic way.
4.1.2 Finite Column on a Non-Linear Foundation

In this section we deal again with a model structure of the column on a non-linear
foundation (Ikeda, Morota, and Elishakoff, 1995). This model was extensively studied
in Chapter 3 in the context of the Monte Carlo method. In this section, we will utilize this
model to illustrate the theory of Ikeda and Murota (1993). This section closely follows
the recent article by Ikeda et al. (1995). It is instructive to consider the non-dimensional
form of the equation
d 4u d 2u d 2 ũ
+ λγ (κ 1 ) + κ 1 u − κ 3 u 3
= −λγ (κ1 ) (4.13)
dη4 dη2 dη2
for the deflection of an imperfect column on a non-linear “softening” foundation with
simply supported boundary conditions
d 2u
u= = 0, at η=0 and η=1 (4.14)
dη2
where
w w̃ x P
u= , ũ = , η= , λ=
l Pcl
(4.15)
κ1 k1 l 4 k 3 l 4 2
γ (κ1 ) = n 2∗ π 2 + , κ1 = , κ3 =
(n ∗ π )2 EI EI
w̃(x) is the initial imperfection function; w(x) is the additional deflection due to the
axial load P; λ is the non-dimensional buckling load normalized with respect to the
(classical) buckling load Pcl of a column on a linear elastic foundation; κ1 and κ3 are,
respectively, the linear and non-linear spring coefficients of the foundation; x is the
axial coordinate; l is the length of the column; E is the Young’s modulus; I is the cross-
section moment of inertia; is the cross-section radius of gyration; and n ∗ denotes the
number of the half-waves for the buckling mode of the linear structure.
Here, a procedure to obtain the stochastic properties of the column is introduced
because a summary of Elishakoff (1979a) will be recapitulated. The normalized initial
imperfection ũ(η) is assumed to be a Gaussian random function of the position η with
given mean function Ū (η) and auto-correlation function K ũ (η1 , η2 ), η1 , and η2 being two
(generally distinct) points on the column axis. To employ the Boobnov-Galerkin-type
solution procedure, we resolve the displacement and initial imperfections, respectively,
into finite Fourier series

N
N
u(η) = ξκ sin(κπ η), ũ(η) = ξ̃κ sin(κπ η) (4.16)
κ=1 κ=1
compatible with the boundary condition (4.14), where N denotes the number of terms to
be implemented in the formulation. We take (ξ̃1 , . . . , ξ̃ N ) as the vector of imperfections,
[i.e., ν = (ξ̃1 , . . . , ξ̃ N )T (with p = N ) in the notation of Section 4.1.1].
The mean values X k = E[ξ̃k ](k = 1, . . . , N ) are obtainable as
1
Xk = 2 Ū (η) sin(kπ η) dη (4.17)
0
The auto-correlation function of the imperfections takes the form
K ũ (η1 , η2 ) = E{[u(η1 ) − Ū (η1 )][u(η2 ) − Ū (η2 )]}

N
N
= E[(ξ̃k − X̄ k )(ξ̃r − X̄ r )] sin(kπ η1 ) sin(r π η2 ) (4.18)
k=1 r =1
Hence the elements (W −1 )kr of the variance-covariance matrix W −1 of (ξ̃1 , . . . , ξ̃ N ) are

determined by
1 1
(W −1 )kr = 4 K ũ (η1 , η2 ) sin(kπ η1 ) sin(r π η2) dη1 dη2 (4.19)
0 0
The probabilistic characteristics of imperfections ũ(η) are specified such that
ν 0 = X̄ = (X 1 , . . . , X N )T = 0 (4.20)
−1
K ũ (η1 , η2 ) = A(η1 − η2 ) sin B(η1 − η2 ) (4.21)
with positive constants A and B (due to Boyce, 1961). The substitution of (4.21)
into (4.19) yields the variance-covariance matrix W −1 .
In Elishakoff (1979a), the differential equations (4.13) of the column was dis-
cretized with the use of (4.16) and was numerically solved to arrive at the non-
dimensional buckling load λc , which was governed by the simple unstable-symmetric
Figure 4.1 Comparison of theoretical probability density function fλc (λc ) numerical his-
togram of the critical load λc for the finite column on a non-linear elastic foundation. His-
togram after Elishakoff, Journal of Applied Mechanics, 1979a (κ1 = κ3 = (4π )4 , A = 0.01,
B = 1, m∗ = 4, N = 7) (after Ikeda, Murota, and Elishakoff, 1995; Copyright c Elsevier
Science Ltd., reprinted with permission).
bifurcation point. Figure 4.1 shows the histogram of λc produced by the Monte Carlo
method for an ensemble of 1000 columns with the Gaussian imperfections with the
probabilistic characteristics of (4.17) and (4.18). The non-dimensional buckling load
λ0c for the perfect system satisfies
λ0c = 1 (4.22)
4.1.3 Application of Ikeda-Murota Theory

The sample mean E[$c ] and variance Var[λc ] of buckling load $c are calculated
respectively, as E[$c ] = 0.86 and Var[λc ] = 0.069 based on the histogram of $c in
Figure 4.1. The substitution of these values into (4.10) results in the value of λ0c and
C0 σ ρ listed in Table 4.2 for various kinds of critical points. For the unstable simple-
symmetric bifurcation point, which governs the critical load in this case, the value of
λ0c = 1.03 computed in this manner is close to its theoretical value, which is equal to
the unity by (4.22).
Table 4.2. Parameters for the finite column

computed for various kinds of simple critical
points
Type of critical points λ0c C0 σ ρ Difference

Limit point 0.86 0.069 16.9
Asymmetric bifurcation point 0.99 0.16 22.8
Symmetric bifurcation point 1.03 0.20 6.5
Figure 4.2 Comparison of theoretical and numerical reliability functions R(λ) for the
finite column on a non-linear elastic foundation: (a) theory (unstable-symmetric bifur-
cation point) and (b) Monte Carlo method (Elishakoff, Journal of Applied Mechanics,
1979a) (κ1 = κ3 = (4π)4 , A = 0.01, B = 1, m∗ = 4, N = 7) (after Ikeda, Murota, and
Elishakoff, 1995; Copyright
With the use of the values of λ0c and C0 σ ρ in Table 4.2, we computed the curves of the
probability density function for various kinds of critical points in Figure 4.1, and those
of the reliability function in Figure 4.2(a); Figure 4.2(b) shows the empirical reliability
curve computed by Elishakoff (1979a). The curves for the unstable simple-symmetric
bifurcation point serve as theoretical ones, whereas those for the other points are used
here for comparison. According to χ 2 test of goodness of fit (Kendall and Stuart, 1973),
the critical value of the difference between the observed and the theoretical probability
density function at a level of significance of 0.05 is 11.1 or less. (Strictly speaking, the
true critical value should be slightly smaller than 11.1 because our procedure estimates
the location and scale parameters of the distribution. But 11.1 is employed here to
approximate it.) The value of difference for the unstable simple-symmetric bifurcation
point is equal to 6.5 and, hence, is significantly smaller than the critical value 11.1. The
reliability curve by the present method in Figure 4.2(a) is very close to the empirical
curve in Figure 4.2(b). These may suffice to show the validity of the present method,
which achieves the accuracy, while retaining the desired simplicity.
4.1.4 Axially Compressed Cylindrical Shells

We consider two groups of cylindrical shells, called A-shells and B-shells after Ikeda
et al. (1995). A group of A-shells of L = 176.02 mm, R = 101.6 mm, and t =
0.1160 mm were manufactured by electroplating from pure copper (where L is the
shell length, R is its radius, and t is its wall thickness) (Arbocz and Abramovich, 1979).
A group of B-shells of L = 134.37 mm, R = 101.6 mm, and t = 0.2007 mm were
manufactured by putting pieces of thick-walled, seamless, brass tubes onto a mandrel
and then machining them to the desired wall thickness (Babcock and Sechler, 1962).
Because of the very nature of the manufacturing process, each realization of the shell
will have a different initial shape. The imperfections represent deviations of the initial
shape from the perfect circular cylinder. They were recorded by the special experi-
mental setup developed at Caltech (Arbocz and Babcock, 1968) and later sent to the
data bank of a complete surface map of the shells (Singer et al., 1978; Arbocz and
Abramovich, 1979).
This subsection offers a procedure to arrive at the reliability of the shells (Elishakoff
and Arbocz, 1982). We consider a cylindrical shell with an axisymmetric initial imper-
fection of the form
iπ x
ν(x) = t ξ̃i cos (4.23)
L
where i is an integer denoting the number of half-waves in the (axial) x-direction. We
further assume the following buckling mode:
kπ x ly
w(x, y) = tCk l sin cos (4.24)
L R
where k and l are integers denoting the number of half-waves in the axial and the circum-
ferential directions, respectively. With the use of Koiter’s special theory (Koiter, 1963),
one can derive the following relationship between the non-dimensional axial load λ
(at which the resulting fundamental equilibrium state bifurcates into an asymmetric
pattern) and the imperfection magnitude ξ̃i

2 4
cβ 8α
(λi − λ)2 (λkl − λ) + (λi − λ) l2 λ + λi 2 k 2 ξ̃i δi,2k
2αk αk + βl

1 1
+ 8c αk βl
2 2 4
+ 2 λi ξ̃i = 0
2 2
(4.25)
2 2 2 2
9αk + βl
2
αk + βl
where
2
1 2 1 1 αk2 + βl2 αk2
λi = αi + 2 , λkl = + 2 ,
2 αi 2 αk2 αk2 + βl2
2 2 2 (4.26)
2 Rt π 2 Rt π Rt 1
αi = i
2
, αk = k
2
, βl2 = l 2
2c L 2c L 2c R
Figure 4.3 Comparison of theoretical probability density function fλc (λc ) and numerical
histogram (Elishakoff and Arbocz, International Journal of Solids and Structures, 1982):
(a) A-shells and (b) B-shells (after Ikeda, Murota, and Elishakoff, 1995; Copyright c
c is a constant, and δi,2k is the Kronecker delta, having a value of unity or zero depending
on whether i = 2k.
In Equation (4.25) for an imperfection-sensitive structure, ξ̃i must be negative, and
i must be an even integer. It also implies that the critical load for this shell is governed
by the simple asymmetric bifurcation point. The buckling loads were obtained by
solving (4.25) for different values of k and l(i = 2k) until all the available imperfection
harmonics have been tested. The absolute minimum bifurcation buckling load is then
identified as the critical buckling load for the shell under consideration.
The critical buckling loads for the groups of A- and B-shells, respectively, were
computed for an ensemble of 100 shells, and their histograms and the reliability curves
were computed as shown in Figures 4.3 and 4.4.
Table 4.3 lists the value of λ0c and C0 σ ρ for the A- and B-shells computed for
various kinds of critical points. The values of λ0c = 1.02 and 0.97 for the asymmetric
Table 4.3a. The parameters for A-shells computed for various

kinds of simple critical points
Type of critical points λ0c C0 σ ρ Differencea

Limit point 0.95 0.033 7.3
a Critical value at a significance level of 0.05 is 6.0 or less.
Figure 4.4 Comparison of theoretical and numerical reliability functions

R(λ): (a) A-shells and (b) B-shells. Solid line: theory (asymmetric bifurca-
tion point); dashed line: Monte Carlo method – Elishakoff and Arbocz, 1985
(after Ikeda, Murota, and Elishakoff, 1995; Copyright c Elsevier Science
Ltd., reprinted with permission).
Table 4.3b. The parameters for B-shells computed for

various kinds of simple critical points
Type of critical points λ0c C0 σ ρ Differencea

Limit point 0.73 0.102 5.7
a Critical value at a significance level of 0.05 is 11.1 or less.
bifurcation point, computed in this manner is close to its theoretical value, which equals
the unity.
Figure 4.3 shows the curves of probability density function f $c (λ) of the normalized
buckling load λc computed for various kinds of critical points. Figure 4.4 compares
the theoretical and empirical reliability function R(λ). The curves for the asymmetric
bifurcation point serve as theoretical ones, while those for the other points are used here
for comparison. The theoretical curves can simulate well the empirical histograms and
the reliability curves to show their usefulness. The theoretical curve for the probability
density function passed the χ 2 test at a significance level of 0.05 or less (the true critical
values are slightly smaller than those used in Table 4.3). These demonstrate, at the least
for these shells, the adequacy of the assumption that the initial imperfections are subject
to normal distribution.
In concluding this section, we must mention that the explicit formulas for the
probability density function (4.7) and (4.8) of the critical buckling loads, the validity
of which has been assessed through this study, appears to be of great assistance in
evaluating the reliability of structures. These formulas are expected to be applicable for
various kinds of structures undergoing bifurcation, though caution should be exercised
based on the fact that different formulas must be applied in accordance with the type
of critical points.
4.2 Second-Moment Analysis of the Buckling of Isotropic Shells

with Random Imperfections
In this section, we follow Elishakoff et al. (1987). As was mentioned earlier, a probabili-
stic approach to buckling analysis of structures was suggested in a study of imperfection-
sensitive structures by Bolotin (1958), who postulated that the random buckling load
$c of a structure can be expressed as a function of a finite number of random variables
X i representing the initial imperfection Fourier coefficients, Equation (4.1).
Bolotin applied this method to a cylindrical panel under a uniform compressive
load along its curved edges, with the initial imperfections being represented by a sin-
gle normally distributed amplitude parameter. A single-term Boobnov-Galerkin ap-
proximation yielded Equation (4.1). Makarov (1969) used a simplified expression of
Equation (4.1) to derive the statistical properties of the buckling loads. However, as
Amazigo (1976) mentioned, “it is . . . a non-trivial problem to obtain a relation of type
Eq. (4.1) and to perform the above analysis for p > 2, say. It is this difficulty that limits
4.2 BUCKLING OF ISOTROPIC SHELLS WITH RANDOM IMPERFECTIONS 187
the effectiveness of this method.” Indeed, because there is no formula of Equation (4.1)
in the literature for multi-parametric imperfections, the direct, analytical evaluation of
the buckling loads seems to be a formidable task.
On the other hand, procedures (see Babcock, 1983; Almroth and Brogan, 1978;
Arbocz and Babcock, 1974, 1978) have been developed, dealing with the determination
of the buckling loads numerically. We limit ourselves to mentioning the multi-mode
analysis (MIUTAM) by Arbocz and Babcock (1974), the well-known general-purpose
code STAGS (Almorth and Brogan, 1978; Arbocz and Babcock, 1978), and the Dutch
multi-purpose finite-element package DIANA (Borst et al., 1983). In addition, the
recent results of extensive initial imperfection surveys have been directly incorporated
into the probabilistic analysis of shells with random imperfections (Elishakoff and
Arbocz, 1982, 1985) without resorting to the number of restrictive assumptions used
in the literature on the probabilistic buckling of imperfect structures.
The question that arises in this context follows: Is it possible to develop a simple
but rational method of checking the reliability of the shells, using some statistical mea-
sures of the imperfections involved, to provide an estimate of the structural reliability
without recourse to the Monte Carlo method? Of course, even with the relation of type
of Equation (4.1) available, it would still be an enormous task to use it for the reliability
calculations in view of the cumbersome integration in a multi-dimensional space. Al-
ternatively, the order zero second-moment approach (Rzhanitsin, 1949; Cornell, 1969)
has been known for many years to those engaged in the probabilistic analysis of struc-
tures. This method, requiring only the knowledge of mean values and elements of a
variance-covariance matrix of the basic variables (imperfection Fourier coefficients),
will be adopted in this section.
The cornerstone of the method is the deterministic state equation
Z = Z (X 1 , X 2 , . . . , X p ) (4.27)
where the nature of the so-called performance function Z (. . .) depends on the type of
the structure and the limit state considered. According to the definition, the equation
Z =0 (4.28)
determines the failure boundary; Z < 0 implies failure, and Z > 0 indicates non-failure
(successful performance). The use of the second-moment method then requires lin-
earization of the function Z at the mean point and knowledge of the distribution of
the random vector X. Calculations are relatively simple if X is normally distributed. If
X is not normally distributed, an appropriate normal distribution must be substituted
instead of the actual one.
Because of the randomness of initial imperfection parameters, the buckling load
in Equation (4.1) is a random variable, denoted by $c . In the present case, we are
interested in knowing the reliability of the structure at any given load λ, that is
R(λ) = Prob($c > λ) (4.29)
A function Z can be defined as
Z (λ) = $c − λ = ψ(X 1 , X 2 , . . . , X p ) − λ (4.30)

where λ is the applied deterministic load. At first glance, it appears that, due to the
absence of straightforward deterministic relation connecting $c and the X i , the first-
order second-moment analysis is unfeasible.
Indeed, it is impossible to perform such an analysis analytically. However, it can
be done numerically, as has been performed for a different problem, as reported by
Kazadeniz, van Manen, and Vrouwenvelder (1982). To combine the numerical codes
with the mean value first-order second-moment method, we need to know the lower
order probabilistic characteristics of Z . In the first approximation, the mean value will
be determined as follows:
E(Z ) = E($ p ) − λ
= E[ψ(X 1 ), E(X 2 ), . . . , E(X p )] − λ
∼
= ψ[E(X 1 ), E(X 2 ), . . . , E(X p )] − λ (4.31)
This corresponds to the use of the Laplace approximation of the moments of the non-
linear functions. The value of
ψ[E(X 1 ), E(X 2 ), . . . , E(X p )] (4.32)
should be calculated numerically by any of the codes reported by Arbocz and

Babcock (1974, 1978) and Borst et al. (1983). It corresponds to the deterministic
buckling load of the structure possessing mean imperfection amplitudes.
The variance of Z is given by
p p
∼ ∂ψ ∂ψ
Var(Z ) = Var($s ) = v jk (4.33)
j=1 k=1
∂ξ j ξ j =E(X j ) ∂ξk ξk =E(X k )
where v jk = cov (X j , X k ) is a covariance.

Calculation of the derivatives of ∂ψ/∂ξ j are performed numerically by using the
following numerical differentiation formula:
∂ψ ∼ ψ(ξ1 , ξ2 , . . . , ξ j−1 , ξ j + ξ j , ξ j+1 , . . . , ξ p ) − ψ(ξ1 , ξ2 , . . . , ξ )
= (4.34)
∂ξ j ξ j
at values of ξ j = E( X̄ j ). To find these derivatives it is necessary to carry out p calcu-

lations of the buckling problem.
Having estimated E(Z ) and Var(Z ) we obtain the estimate for the probability of
failure:
P f (λ) = Prob(Z < 0) = (−β) (4.35)
at the load level λ. In Equation (4.35), is the standard normal probability distribution
function and

β = E(Z )/ Var(Z )
is the reliability index. Accordingly, reliability will be estimated as
P(λ) = Prob(Z < 0) = 1 − (−β) = (β) (4.36)

Table 4.4. Geometrical and material data on the group of B-shells

(Arbocz and Babcock, 1974)
Shella R, mm t, mm L, mm LHA , mmb Pexp , mm

B-1 101.60 0.2050 196.85 171.45 11326.0
B-2 101.60 0.1852 144.78 121.92 7178.5c
B-3 101.60 0.1491 140.97 121.92 —
B-4 101.60 0.2634 140.97 121.92 16661.2
a For all shells E = 1.065 × 105 N/mm2 and ν = 0.3.
b L HA = length used for harmonic analysis.
c Shell had a visible stable prebuckle at a spot where the surface had been scratched
by a sharp object.
For the actual calculations, we used the data associated with the so-called B-shells
(Arbocz and Babcock, 1974). The geometrical and material properties of the B-shells
are provided in Table 4.4.
Since the measurements included only 41 points in the axial direction and 49
points in the circumferential direction, it was not feasible to compute those Fourier
coefficients the order of which exceeds the “cut-off ” values of k = 20 and l = 24. For
these coefficients, the Donnell-Imbert imperfection model (Imbert, 1971) was used:
X r,s = X/k r l s (4.37)
where k is the number of axial half-waves and l is the number of circumferential

full waves. Values of measured imperfections are given in Table 4.5. Also shown is the
complete imperfection model used with an indication of which coefficients were actually
Table 4.5. Imperfection model and values of the Fourier coefficients

(Elishakoff and Arbocz, 1985)
Simulated random variables Donnell-Imbert imperfection model

C1.6 C1,10 C25,6 C25,10
C1,2 C25,2
A2,0 + C1,8 + C25,8 + A26,0
C2,3 C24,3
A4,0 + C2,11 + C24,11 + (A26,0 )
B-1 B-2 B-3 B-4
A2.0 −0.010809 −0.027238 −0.089906 −0.017560
A4,0 0.022578 −0.007836 −0.089906 −0.009239
C1,2 0.417400 0.392870 0.741280 0.222900
C1,6 −0.077872 −0.143490 0.017483 0.077668
C1,8 −0.263690 −0.009405 0.112470 0.101510
C1,10 0.036568 0.043628 −0.245610 −0.008853
C2,3 −0.101290 0.034018 −0.064766 −0.001887
C2,11 0.009732 −0.008685 −0.028261 0.013545
Table 4.6. Sample mean vector and sample variance-covariance matrix (Elishakoff et al., 1987)
Mean Elements of variance-covariance matrix

vector (All elements of the variance-covariance matrix are multiplied by 100)
−0.0364 1 0.1319
−0.0050 2 0.0566 0.0402
0.4436 3 −0.7074 −0.1916 4.686
−0.0316 4 −0.0926 −0.0810 −0.0872 0.9670
−0.0148 5 −0.3646 −0.3327 0.6154 0.9956 3.057
−0.0436 6 0.4771 0.1986 −2.478 −0.6529 −1.372 1.868
−0.0335 7 0.0384 −0.0518 −0.5978 −0.0833 0.5647 0.2623 0.3710
−0.0034 8 0.0646 0.0272 −0.3739 0.0205 −0.1497 0.2068 0.0022 0.0369
measured and which ones were extrapolated in accordance with Equation (4.37). The
parameters X, r, and s were determined by the least-square fitting, the distribution of
the measured Fourier coefficients.
The mean vector and the variance-covariance matrix of the Fourier coefficients,
treated as random variables, are given in Table 4.6. In order to apply the first-order
second-moment method, the mean buckling load has to be calculated first. For this
purpose, the multi-mode analysis is used.
Let us give a brief overview of the multi-mode analysis. We can write the Donnell-
type non-linear equations for imperfect stiffened cylindrical shells in the form:
1 1
L H (F) − L Q (W ) = − w,x x − L NL (W, W + 2W̄ ) (4.38)
R 2
1
L Q (F) + L D (W ) = F,x x + L NL (F, W + W̄ ) (4.39)
R
where W is positive inward, and the linear operators are
L D ( ) = Dx x ( ),x x x x + Dx y ( ),x x yy + D yy ( ),yyyy
L H ( ) = Hx x ( ),x x x x + Hx y ( ),x x yy + Hyy ( ),yyyy (4.40)
L Q ( ) = Q x x ( ),x x x x + Q x y ( ),x x yy + Q yy ( ),yyyy
and the non-linear operator is
L NL (S, T ) = S,x x T,yy − 2S,x y T,x y + S,yy T,x x (4.41)
Commas in the subscripts denote repeated partial differentiation with respect to the in-
dependent variables following the comma. The stiffener properties have been “smeared
out” to arrive at effective bending, stretching, and torsional stiffness. The stiffener pa-
rameters Dx x , Hx x , Q x x , Dx y , . . . , etc., are defined as in Babcock (1983). Here W̄ is
the initial radial imperfection, W is the component of displacement normal to the shell
mid-surface and F is the Airy stress function.
Let us represent the (m + 1) approximation to a solution to Equations (4.38) and
(4.39) by
Wm+1 = Wm + δWm , Fm+1 = Fm + δ Fm (4.42)
where Wm , Fm = mth approximation to the solution, and δWm , δ Fm = correction to the

mth approximation. Substitution of Equation (4.42) into Equations (4.38) and (4.39)
and omission of products of the correction quantities yields a set of linear partial
differential equations for determining the correction terms. If one represents the initial
imperfections by

N1
N2
W̄ = t W̄io cos(i x̄) + t W̄kt sin(k x̄) cos(l ȳ) (4.43)
i=1 k,l=1
where x̄ = π x/L and ȳ = y/R, then the linearized governing equations admit sepa-
rable solutions of the form
N1 N2
Wm Wν Wio Wkt
=t + cos(i x̄) + t sin(k x̄) cos(l ȳ)
δWm 0 i=1
δWio k,t=1
δWkt
N1
Fm ERt 2 −λ ȳ 2 /2 ERt 2 Fio
= + cos(i x̄)
Fm c 0 c i=1 δ Fio
N2
ERt 2 Fkt
+ sin(k x̄) cos(l ȳ) (4.44)
c k,l=1
δ Fkt

where Wν = (−ν/c)( H̄x x /1 + µ1 )λ and c = 3(1 − ν 2 ).
The set of linear partial differential equations are satisfied approximately by
Boobnov-Galerkin’s procedure yielding a set of linear algebraic equations in terms
of the unknown correction terms. In matrix notation,

[A]{δ F} + [B]{δW } = − E (1)
(4.45)
[C]{δ F} + [D]{δW } = − E (2)
To obtain the buckling load for a given imperfect cylindrical shell, one begins by
making an initial guess for {W } and {F} at a small initial load level λ. Iteration is then
carried out until the correction vectors are smaller than some preselected value. The
converged solutions then are used as the initial given at the next higher axial load level
λ + λ. The entire process is repeated for increasing values of the axial load parameter
λ. The non-linear analysis then will locate the limit point of the prebuckling states. By
definition the value of the loading parameter λ corresponding to the limit point will be
the theoretical buckling load.
It is shown (Babcock, 1983) that the solution satisfies the circumferential period-
icity. It also contains details of the coefficient matrices A, B, C, and D and the error
vectors E (1) and E (2) .
The result of the calculation of the mean buckling load E($c ) is given in Figure 4.5;
E($c ) = 0.746 (i.e., 74.6% of the classical buckling load). The mean buckling load
calculated via the Monte Carlo method is E($c ) = 0.739; thus, the difference between
mean buckling loads, derived by these two methods is only 0.007 or 0.95%. Next the
sensitivity derivatives were calculated. For the increment of the Fourier coefficients in
Table 4.7. Derivatives of with respect to the Fourier coefficients (Elishakoff et al., 1987)
Xj A2,0 A4,0 C1,2 C1,6 C1,8 C1,10 C2,3 C2,11

∂ϕ/∂ X j 0.09668 0.00340 −0.01854 −0.05687 −0.24686 −0.08183 −0.01214 −0.07173
Equation (4.34), 10% of their original values are used, so that ξ j = 0.10X j . The
calculated derivatives are listed in Table 4.7. In this study the increments of end-
shortening were chosen in such a way that the limit loads were found to an accuracy of
0.0001.
The results of the mathematical expectation and variance of Z are E(Z ) = 0.746 −
λ and Var(Z ) = 0.0175, respectively. The reliability is calculated directly from Equa-
tion (4.36). The reliability functions calculated with the Monte Carlo method and with
the first-order second-moment method are both given in Figure 4.6. These curves ap-
pear to be in excellent agreement, however, in the higher reliability region, which is
important for design load derivation. The deviation is more noticeable here.
From the results of the calculations it may be concluded that:
1. The first-order second-moment method can be successfully used for determining
the reliability function of axially compressed shells.
2. The number of buckling load calculations necessary for first-order second-moment
method is significantly less than with the Monte Carlo method.
3. The mean buckling loads due to both methods are in excellent agreement, but still
higher than the experimental value. This is caused by the simplified determinis-
tic buckling load analysis. Because the present method does not need as many
calculations as the Monte Carlo method, a more advanced and expensive method
Figure 4.5 Response curve with mean imperfections in the λ-δ N

plane; dimensions used for the analysis are the dimensions of shell
B-1 (after Elishakoff et al., Copyright
c 1987 AIAA, reprinted with
permission).
4.3 USE OF STAGS TO DERIVE RELIABILITY FUNCTIONS BY ARBOCZ AND HOL 193
Figure 4.6 The reliability curves for a group of B-shells, calcu-

lated via the Monte Carlo method, and the mean value, first-
order second-moment method (after Elishakoff et al., 1987;
Copyright c AIAA, reprinted with permission).
(in terms of the computer time) can be used for the deterministic analysis of buck-
ling load calculations.
The first-order second moment method was implemented by van den Nieuwendijk
(1997).
4.3 Use of STAGS to Derive Reliability Functions by Arbocz and Hol

In a previous analysis, following Elishakoff et al. (1987), we utilized the multi-mode
analysis for derivation of shell reliability. The highest level of computational complexity
and accuracy in deterministic analysis at present is to employ a two-dimensional non-
linear shell analysis code such as STAGS. STAGS was utilized for reliability analysis
by Arbocz and Hol (1990b). With the use of this type of numerical tool, one can,
in principle, determine the buckling load of a complete shell structure including the
effect of arbitrary initial imperfections represented by a double Fourier series. The
computations by Arbocz and Hol (1990b) were carried out with a STAGS-A code
(Almroth et al., 1973) modified so that, in addition to increments of load and increments
of axial displacement, increments of a ‘path parameter’ (Riks, 1984) can also be used
as a load parameter.
The number of imperfection modes included in the analysis is limited by practi-
cal considerations, like the time required for obtaining the solutions of all buckling
problems needed for calculations of the derivatives ∂ψ/∂ X i . Thus, because the shell
buckling load is determined by solving the governing equations for a particular set of
initial imperfections, an attempt to select a optimal combination of these modes must
be made (i.e., there is a need to locate those imperfection modes which dominate the
prebuckling and the collapse behavior of the shell). Recently, novel group-theoretical
methods have been developed by Wohlever and Healey (1995) to address this question.
Examples of attempts to locate critical imperfection modes, defined as that com-
bination of axisymmetric and asymmetric imperfection modes which would yield the
lowest buckling load have been reported in the literature (Arbocz 1974; Arbocz and
Babcock, 1976, 1980b). These studies have shown that, to yield a decrease from the
Table 4.8. Geometric and material properties and experimental buckling loads of the
AS-shells (Arbocz ad Hol, 1990b)
t A1 e1 I11 × 102 It1 × 102 d1 PEXP

Shells [mm] [mm2 ] [mm] [mm4 ] [mm4 ] [mm] [N]
AS-2 0.1966 0.7987 0.3368 1.5038 4.9448 8.0239 14286.3
AS-3 0.2807 0.7432 0.3614 1.2033 4.0146 8.0289 22357.1
AS-4 0.2593 0.4890 0.2758 0.3474 1.2383 8.0112 17074.9
buckling load of a perfect structure, the initial imperfection harmonics must include at
least one mode with a significant initial amplitude and an associated eigenvalue that
is close to the critical buckling load of the perfect structure. For the calculations, the
data associated with the integrally stringer stiffened aluminum alloy shells tested at
California Institute of Technology; the so-called AS-shells (Arbocz and Abramovich,
1979) will be used. The shell properties are listed in Table 4.8. For the numerical
calculations the properties of shell AS-2 will be used. Relying on the results of ear-
lier investigations of the buckling behavior of the imperfect AS-2 shell (Arbocz and
Babcock, 1978), it was decided to employ the following initial imperfection model for
the collapse load calculations:
2π x πx
w(x, θ ) = h X 1 cos − sin (X 2 cos 2θ + X 3 cos 9θ + X 4 cos 10θ
L L
+ X 5 cos 11θ + X 6 cos 19θ + X 7 cos 21θ )
(4.46)
where θ = y/R, y = circumferential coordinate, R = radius. Note that the shape of

this imperfection model is symmetric in the axial direction about the center of the shell;
hence, only half of the shell length needs to be modeled. However, the imperfection
model includes modes with both even and odd number of circumferential waves. This
implies that, to be able to use the symmetry conditions, half the shell perimeter must
be modeled. Based on the results of the numerical convergence studies (Arbocz and
Babcock, 1978), this leads to the use of the discrete model consisting of 21 × 131 mesh
points (21 mesh points in the axial direction, and 131 mesh points in the circumferential
direction). The foregoing imperfection model requires eight collapse load calculations
to be able to evaluate derivatives ∂ψ/∂ X i .
To apply the first-order, second moment method, the mean buckling load must
be calculated first. Using the foregoing imperfection model with the mean values
of the corresponding equivalent imperfection amplitudes listed in Table 4.9, the cal-
culation result is E(λc ) = 0.87538, whereby the mean buckling load is normalized
by −223.079 N/cm, the buckling load of the perfect AS-2 shell computed using
non-linear prebuckling condition. The derivatives ∂ψ/∂ X i are listed in Table 4.10.
Increments of the path parameter were chosen in such a way that the limit loads
were determined with accuracy of 0.01%. The variance-covariance matrix of ini-
tial imperfection parameters are listed in Table 4.11. Mathematical expectation and
4.3 USE OF STAGS TO DERIVE RELIABILITY FUNCTIONS BY ARBOCZ AND HOL 195
Table 4.9. Values of the equivalent Fourier coefficients

and the reduced sample mean vector (Arbocz and Hol,
1990b)
Xj AS-2 AS-3 AS-4 E(Xi )

1 0.00455 0.01378 −0.01126 0.00236
2 0.33691 0.08298 0.54217 0.32069
3 0.08843 0.02445 0.00297 0.03862
4 0.05524 0.03148 0.00414 0.03028
5 0.05494 0.01912 0.00502 0.02636
6 0.01106 0.00689 0.00424 0.00740
7 0.00879 0.00475 0.00095 0.00483
Table 4.10. Derivatives of ψ with

respect to the equivalent Fourier
coefficients (seven-mode
imperfection model) (Arbocz and
Hol, 1990b)
∂ψ/∂ X j
Xj SS-3 C-4
1 −0.6354 −0.5986
2 0.1498 0.1582
3 0.6924 0.3678
4 0.9138 0.6672
5 0.6233 1.0844
6 0.2449 0.2922
7 0.1811 0.4202
Table 4.11. The reduced sample variance-covariance matrix (all terms are
multiplied by 100) (Arbocz and Hol, 1990b)
 
0.01604
 0.28485 5.29110 Symmetric 
 
−0.02165 −0.18590 0.19763 
 
−0.02122 0.28341 0.10789 0.0537 
 
−0.01354 −0.12712 0.11436 0.06313 0.06625 
 
−0.00226 −0.02590 0.01511 0.00866 0.00879 0.00118 
−0.00303 −0.0384 0.01683 0.01001 0.00983 0.00134 0.00154
variance of the performance function Z are E(Z ) = 0.87538 − λ, Var(Z ) = 0.00486

Note that, at the reliability level of 0.98, one obtains a theoretical estimate of the de-
sign load λdesign = 0.73, which also represents a theoretical estimate of the knockdown
factor.
These results are associated with so-called SS-3 boundary conditions stipulating
that at the boundaries
N x = v = w = Mx = 0 (4.47)
Another set of boundary conditions, designated in the literature as C-4 boundary

conditions were also studied. In this case, the boundary conditions read
u = v = w = ∂w/∂ x = 0 (4.48)
The derivatives ∂ψ/∂ X i are listed in Table 4.10. In this case, the calculated mean
buckling load equals E($c ) = 0.96298, a value normalized by −315.323 N/cm, the
buckling load of the perfect AS-2 shell computed using non-linear prebuckling and
the same C-4 boundary conditions. The computation of the mathematical expectation
and the variance of Z yields: E(Z ) = 0.96298 − λ, Var(Z ) = 0.00400, respectively.
Notice that in this case for a reliability of 0.98 one obtains a knockdown factor of
λdesign = 0.84, where now λdesign is normalized by −315.323 N/cm, the buckling load
of the perfect AS-2 shell using the C-4 boundary conditions.
Comparing the buckling loads predicted for a reliability of 0.98 of Nss−3 =
−162.848 N/cm and Nc−4 = −264.871 N/cm based on the seven-modes imperfection
model of Equation (4.46) with the experimental buckling load Nexp = −223.793 N/cm,
one notices that the calculated results seem to support the suggestion made by
Arbocz (1982b) that the experimental boundary conditions of the test setup used
to buckle the AS-shells at the California Institute of Technology imposed a special
sort of elastic boundary conditions. For all shells E = 6.895 × 104 N/mm2 , ν = 0.3,
R = 101.60 mm, L = 139.70 mm, NR × NC = 21 × 49, 80 stringers.
4.4 Reliability of Composite Shells by STAGS

The methodology developed by Elishakoff et al. (1987) has been applied to cylindrical
shells made of composite materials by Arbocz and Hol (1989). They have investi-
gated the glass-epoxy layered (30◦ , 0◦ , −30◦ ) composite shell, previously studied by
Booton (1976) in the deterministic setting. Values of the Fourier coefficients and their
mean values are listed in Table 4.12. Reliability of composite shells was derived using
STAGS computer code. The following initial imperfection model was utilized for the
collapse load calculations:

πx 10 10
w̄(x, θ ) = h sin C1,i cos iθ + Di j sin jθ (4.49)
L i=5 j=5
This imperfection model requires 13 calculations of the collapse load to evaluate the
derivatives ∂ψ/∂ X j . Note that the required number of the collapse load calculations
4.4 RELIABILITY OF COMPOSITE SHELLS BY STAGS 197
Table 4.12. Values of the Fourier coefficients and the sample mean vector
(Arbocz and Hol, 1989)
Xj B-1 B-2 B-3 B-4 Mean vector

C1,5 1 −0.081 0.115 0.024 0.166 0.0560
C1,6 2 −0.078 −0.143 0.017 0.078 −0.0315
C1,7 3 0.005 0.258 −0.062 −0.014 0.0538
C1,8 4 −0.264 −0.009 0.112 0.102 −0.0148
C1,9 5 −0.116 0.025 −0.051 −0.002 −0.0350
C1,10 6 0.037 0.044 −0.246 −0.009 −0.0435
D1,5 7 0.115 0.177 0.100 0.025 0.1043
D1,6 8 −0.390 0.058 0.128 −0.248 −0.1130
D1,7 9 −0.185 −0.086 0.213 0.060 0.0005
D1,8 10 −0.029 0.087 0.185 −0.050 0.0483
D1,9 11 0.063 0.018 0.102 0.064 0.0618
D1,10 12 −0.013 0.039 −0.051 −0.051 −0.0140
can be reduced to 7 if one replaces the foregoing imperfection model with
πx
10
w̄(x, θ ) = h sin ξ1,i cos(iθ − ϕ1,i ) (4.50)
L i=5
where ξkl = (Ckl + Dkl )1/2 and ϕkl = tan−1 (Dkl /Ckl ). The coefficients ξkl are listed in
Table 4.13. The mean vector and the variance-covariance matrix of coefficients ξkl ,
treated as random variables, are given in Tables 4.13 and 4.14, respectively. For cal-
culation of the mean buckling load, initially the reduced six-mode imperfection in
Equation (4.50) is utilized, with ϕ1,i set to zero. The result of calculation is E($c ) =
0.5942, whereby the mean buckling load is normalized by −730.939 N/cm, the buck-
ling load of the perfect “Booton-shell” computed using linear prebuckling analysis.
The derivatives ∂ψ/∂ξi are listed in Table 4.15. The results of the calculation of the
mathematical expectation and variance of the performance function Z are E(Z ) =
0.5942 − λ, Var(Z ) = 0.00926. The specified reliability of 0.98 corresponds to the
non-dimensional design load, or theoretical knockdown factor of 0.39, a value that ap-
pears to be too low for the published experimental results for composite shells. Analysis
of Arbocz and Hol (1989) shows that the imperfection model in Equation (4.50) results
Table 4.13. Values of the equivalent Fourier coefficients and the reduced sample
mean vector (Arbocz and Hol, 1989)
Xj B-1 B-2 B-3 B-4 Mean vector

ξ1,5 1 0.1407 0.2111 0.1028 0.1679 0.1556
ξ1,6 2 0.3977 0.1543 0.1291 0.2600 0.2353
ξ1,7 3 0.1851 0.2720 0.2218 0.0616 0.1851
ξ1,8 4 0.2656 0.0875 0.2163 0.1136 0.1707
ξ1,9 5 0.1320 0.0308 0.1140 0.0640 0.0852
ξ1,10 6 0.0392 0.0588 0.2479 0.0518 0.0994
in a strongly localized collapse pattern. This, however, disagrees with the observed
global collapse phenomenon, reported in experimental investigations.
Arbocz and Hol (1989) have repeated the calculations of the reliability function
R(λ) using the 12-mode imperfection model in Equation (4.49). Elements of variance-
covariance matrix read:
v11 = 1.1791 v21 = 0.3428 v22 = 0.9646
v31 = 0.6020 v32 = −1.0158 v33 = 1.9691
v41 = 1.4424 v42 = 0.9960 v43 = −0.1996
v44 = 3.0618 v51 = 0.6406 v52 = 0.0117
v53 = 0.5528 v54 = 0.7609 v55 = 0.3939
v61 = 0.1470 v62 = −0.6514 v63 = 1.2005
v64 = −1.3733 v65 = 0.1082 v66 = 1.8778
v71 = −0.1921 v72 = −0.5832 v73 = 0.5992
v74 = −0.4017 v75 = 0.0210 v76 = 0.1786
v77 = 0.3902 v81 = 0.8492 v82 = −0.3093
v83 = 0.8634 v84 = 2.8270 v85 = 0.7949
v86 = −2.0265 v87 = 0.6379 v88 = 0.6759
v91 = 0.6685 v92 = 1.1697 v93 = −1.1862
(4.51)
v94 = 2.6540 v95 = 0.2879 v96 = −2.1160
v97 = −0.4635 v98 = 2.6591 v99 = 3.0196
v10,1 = −0.0771 v10,2 = −0.1618 v10,3 = −0.0081
v10,4 = 0.8447 v10,5 = 0.0920 v10,6 = −1.1303
v10,7 = 0.3065 v10,8 = 2.4748 v10,9 = 1.1397
v10,10 = 1.1941 v11,1 = −0.1264 v11,2 = 0.2340
v11,3 = −0.4582 v11,4 = 0.1600 v11,5 = −0.1096
v11,6 = −0.3934 v11,7 = −0.1173 v11,8 = 0.0523
v11,9 = 0.4080 v11,10 = 0.1164 v11,11 = 0.1180
v12,1 = −0.0179 v12,2 = −0.3611 v12,3 = 0.4738
v12,4 = −0.2140 v12,5 = 0.0667 v12,6 = 0.2295
v12,7 = 0.2290 v12,8 = 0.3228 v12,9 = −0.3528
v12,10 = 0.1096 v12,11 = −0.1028 v12,12 = 0.1489
The calculations of the mean buckling load yields, in this case, E($c ) = 0.8854,
whereby again −730.939 N/cm is used for non-linearization purposes. The results
of the computations of the derivatives ∂ψ/∂ξi , are listed in Table 4.16. The math-
ematical expectation and the variance of Z are, in this case, E(Z ) = 0.8854 − λ,
Var(Z ) = 0.00014, respectively. In this case for the targeted reliability of 0.98, the
use of the knockdown factor λdesign = 0.79 is implied; it is a value that is considerably
higher than the value obtained for the reduced six-mode imperfection model. Utilizing
the initial imperfection pattern given in Equation (4.49), or, alternatively, inclusion of
4.4 RELIABILITY OF COMPOSITE SHELLS BY STAGS 199
Table 4.14. The reduced sample variance-covariance matrix (all terms are
multiplied by 100) (Arbocz and Hol, 1989)
 
0.2078
−0.0338 1.4941 
 
 0.0455 −0.4663 0.8048 
 
−0.3047 0.5302 0.0497 0.7090 
 
−0.1833 0.2807 −0.0351 0.3830 0.2143 
−0.3258 −0.7811 0.0260 0.2385 0.1561 0.9868
the phase ϕ1,i in the imperfection model [Equation (4.50)] results in a global deforma-
tion pattern and in a less localized collapse mode. The results agree better with existing
experimental evidence. Comparison of the results obtained through the use of 6- or
12-mode imperfection models clearly illustrates that one should exercise extreme cau-
tion when choosing the deterministic model for predicting the buckling loads for per-
forming the stochastic analysis.
An analogous conclusion was also arrived at in another study by Arbocz and Hol
(1989). On the other hand, the success of the deterministic buckling load analysis
very heavily depends on the appropriate choice of the non-linear model, which in turn
requires considerable knowledge by the analyst of the expected physical behavior of
imperfect shell structures.
We digress, in words of Arbocz and Hol (1989) that “. . . only a shell design spe-
cialist who is aware of the latest theoretical developments and who is familiar with
the theories upon which the non-linear structural analysis codes he uses are based,
can achieve the accurate modeling of the collapse behavior of complex structures that
guarantees a successful application” of stochastic theories of buckling as developed
by Elishakoff (1978b, 1979b, 1980b), Elishakoff and Arbocz (1982, 1985), Elishakoff
et al. (1987), and other investigators. According to Arbocz and Hol (1989),
. . . one can not repeat this warning often enough. The danger of incorrect predictions
lies in the use of sophisticated computational tools by persons of inadequate theoretical
background. For a successful implementation of the proposed improved shell design
procedure the companies involved in the production of shell structures must be prepared
to do the initial investment in carrying out complete imperfection curves on a small
sample of shells that are representative of their production line. With the modern
measuring and data acquisition systems complete surface maps of very large shells can
be carried out, at a negligibly small fraction of their production cost.
Table 4.15. Derivatives of ψ with respect to the equivalent Fourier coefficients

(6-mode imperfection model) (Arbocz and Hol, 1989)
Serial number, j Xj ∂ψ/∂ X j Serial number, j Xj ∂ψ/∂ X j

1 ξ1,5 −0.1558 4 ξ1,8 −0.5342
2 ξ1,6 −0.3492 5 ξ1,9 −0.4527
3 ξl,7 −0.5001 6 ξ1,10 −0.2997
Table 4.16. Derivatives of ψ with respect to the Fourier coefficients (12-mode

imperfections model) (Arbocz and Hol, 1990b)
Serial number, j Xj ∂ψ/∂ X j Serial number, j Xj ∂ψ/∂ X j

1 C1,5 0.2719 7 D1,5 0.0251
2 C1,6 0.0689 8 D1,6 0.4428
3 C1,7 −0.5224 9 D1,7 0.2000
4 C1,8 −0.3292 10 D1,8 −0.4526
5 C1,9 0.3150 11 D1,9 −0.03892
6 C1,10 0.3854 12 D1,10 0.1856
4.5 Buckling Mode Localization in a Probabilistic Setting

In this and the following section, we will combine the probabilistic treatment with the
finite element method, although in different contexts (Xie, 1995). Here we will con-
sider mode localization in a randomly disordered multi-span continuous beam with
deterministic elastic modulus. In Section 4.6, in contrast, we will deal with the situa-
tion where the geometric disorder is not present, but the elastic modulus constitutes a
random field. In both cases, the discretization through the finite element method will
be conducted. One must stress, however, that the disorder (or irregularity) that can
occur in geometry of configuration and material properties of the structure is gener-
ally of a uncertain structure. In Chapter 1, we dealt with the disorder in the geometry
of configuration, in the deterministic setting. Here uncertain disorder is modeled as
having a random nature. Buckling mode localization in probabilistic setting was ap-
parently first treated by Ariaratnam and Xie (1996). This section follows the study by
Xie (1995).
An N -span continuous beam under compressive load P as shown in Figure 4.7(a)
is considered. The length of the ith span is L i , and the flexural stiffness is ki = E Ii /L i .
The nominal span length and flexural stiffness are L and k, respectively. The torsional
spring connecting span i − 1 and span i has spring stiffness Ti , which reflects the
Figure 4.7 (a) Multi-span continuous beam under compressive axial load, (b) sta-
bility functions (s) and (c) (after Xie, 1995; Copyright
c 1995 AIAA, reprinted with
permission).
4.5 BUCKLING MODE LOCALIZATION IN A PROBABILISTIC SETTING 201
coupling between these two spans. When Ti is small, span i − 1 and span i are strongly
coupled; when Ti is large, the two spans are weakly coupled. As an extreme case, when
Ti approaches infinity, each span is individually clamped: there is no coupling between
the adjacent spans in this case.
Consider a beam of length L and flexural stiffness k = EI/L as shown in Figure
4.7(b). It is known (Horne and Merchant, 1965) that in order to have rotation θ at
support A and no rotation at support B, a moment M A = skθ at support A and a moment
M B = sckθ at support B are required, where s and c are the stability functions defined as
(1 − 2α cot 2α)α 2α − sin 2α π√
s= , c= , α= ρ (4.52)
tan α − α sin 2α − 2α cos 2α 2
For the N -span continuous beam, the rotation at support i is denoted by θi . Employing
the preceding result, the equilibrium condition at support i requires
si−1 ci−1 ki−1 θi−1 + (si−1 ki−1 + si ki + Ti )θi + si ci ki θi+1 = 0 (4.53)
where
(1 − 2αi cot 2αi )αi 2αi − sin 2αi
si = , ci = ,
tan αi − αi sin 2αi − 2αi cos 2αi
π√ P P PE,i
αi = ρi , ρi = = , ρ̂i = (4.54)
2 PE,i ρ̂i PE
PE,i = EIi (π/L i )2
From Equation (4.52), θi+1 may be expressed as
si−1 k̂i−1 + si k̂i + ti si−1 ci−1 k̂i−1

θi+1 = − θi − θi−1 (i = 2, 3, . . . , N ) (4.55)
si ci k̂i si ci k̂i
where k̄i = ki /k, ti = Ti /k, or, in the matrix form,

θi+1
xi = Ti xi−1 , xi =
θi
 
si−1 k̂i−1 + si k̂i + ti si−1 ci−1 k̂i−1 (4.56)
− θi − 
Ti =  si ci k̂i si ci k̂i 
1 0
in which T i is the transfer matrix. The state vector x n = {θn+1 , θn }T can be related to the
initial state vector x 1 = {θ2 , θ1 }T by a product of transfer matrices x n = T n T n−1 · · ·T 2 x 1 ,
where the superscript T denotes transpose.
It is assumed that the disorders of spans are random, are statistically independent of
disorders in other spans, and have a common probability distribution. Then the transfer
matrices T i will also be independent and identically distributed. The rate of growth of
the state vector x n or θn+1 is governed by the behavior of the product of random matrices
T n T n−1 · · ·T 2 . The asymptotic properties of such a product have been studied by many
researchers. In this paper, Furstenberg’s theorem (Furstenberg, 1963) on the limiting
behavior of products of random matrices will be utilized. Furstenberg’s theorem may

be stated as follows.
Furstenberg’s Theorem: Let T1 , T2 , . . . , Tn be non-singular, independent, iden-

tically distributed 2 × 2 matrices, where T i = T i () is a function of the random vector
with probability density p(). If at least two of the random transfer matrices do not
have common eigenvectors, and if
. n
1
lim ln |det T i | = 0 (4.57)
n→∞ n
i=1
then there exists a constant δ > 0 such that, for each x 0 = 0,

1
lim ln T n T n−1 · · · T 1 x 0 = δ (4.58)
n→∞ n
with probability 1, and
1
lim ln T n T n−1 · · · T 1 = δ (4.59)
n→∞ n
with probability unity, where x is a suitable norm of the vector x and T denotes a
suitable norm of the matrix T.
It can easily be seen that the transfer matrices defined in Equation (4.56) satisfy
the condition (4.57). Applying Furstenberg’s theorem to Equation (4.56), one obtains,
with probability unity,
1
lim ln x n = lim ln T n T n−1 · · · T 2 x 1 = δ > 0 (4.60)
n→∞ n − 1 n−1
implying that, for large n,
x n = eδ(n−1) x 1 (4.61)
The positive number δ characterizes the average exponential rate of growth of the norm
of the state vector xn = {θn+1 , θn }T ; $ is called the Lyapunov exponent. It can be shown
that Equation (4.61) implies for large n
|θn+1 | = eδ(n−1) |θ1 | (4.62)
Therefore, the Lyapunov exponent λ characterizes the exponential rate of growth of the
angles of rotations. The positivity of the Lyapunov exponent δ for randomly disordered
structure results in the localization in the buckling modes because the non-zero angles
of rotation growing exponentially from each end of the large multi-span beam must
match at the maximum; in other words, the buckling mode is localized with amplitudes
decaying exponentially at the average rate δ on either side of some region. The Lyapunov
exponent δ is, therefore, the localization factor.
On the other hand, letting Bn = T n T n−1 · · ·T 2 , one obtains x n = Bn x 1 . The
Euclidean norm of x n is x n 2 = x 1T BnT Bn x 1 . Let σmax2
= σ12 ≥ σ22 = σmin
2
be the
eigenvalues of BnT Bn and e1 and e2 the corresponding orthonormal eigenvectors. By

Rayleigh’s principle,
e1T BnT Bn e1 x 1T BnT Bn x1 e2T BnT Bn e2
σmax
2
= ≥ ≥ = σmin
2
(4.63)
e1 2 x 1 2 e2 2
or
σmax x 1 ≥ x n ≥ σmin x 1 (4.64)
The upper and lower bounds are reached when x 1 takes the values e1 and e2 , respec-
tively. Hence, taking the natural logarithm, dividing by n − 1, and taking the limit as
n → ∞ results in
1 1 1
lim ln σmax ≥ lim ln x n ≥ lim ln σmin (4.65)
n→∞ n − 1 n→∞ n − 1 n→∞ n − 1
The larger Lyapunov exponent is then given by

1
δ1 = δmax = lim ln σmax (4.66)
n→∞ n − 1
From the well-known multiplicative ergodic theorem of Oceledec (1968), x n

grows exponentially at the average rate δmax for any x 1 except when x 1 = e2 , x n
grows at the rate
1
δ2 = δmin = lim ln σmin (4.67)
n→∞ n−1
It may be noted that δ2 = −δ1 because the determinant of matrix BnT Bn is unity.
If a monocoupled structure is perfectly periodic, T 2 = T 3 = · · · = T n = T. Let
$1 , $2 (|$1 | ≥ |$2 |) be the eigenvalues of T and E 1 , E 2 be the corresponding
orthonormal eigenvectors. It can be shown that σmax 2
= |$1 |2n−2 , σmin
2
= |$2 |2n−2 are
the eigenvalues of BnT Bn . Hence, from Equations (4.66) and (4.67), for any x 1
not parallel to E 2 , x n grows exponentially at the rate δ1 = ln |$1 |, whereas when
x1 is parallel to E 2 , x n grows exponentially at the rate δ2 = ln |$2 |.
If the multi-span continuous beam shown in Figure 4.7 is perfectly periodic (i.e.,
L 1 = L 2 = · · · = L N = L, k1 = k2 = · · · = k N = k, t1 = t2 = · · · = t N +1 = t), the
transfer matrix given by Equation (4.56) is then

2γ 1 1 t
T= , γ =− 1+ (4.68)
1 0 c 2S
If the absolute value of the trace of the matrix T satisfies the inequality, |tr(T)| < 2
(i.e., |γ | < 1), the eigenvalues of T are complex and the Lyapunov exponents are
δ1,2 = 0. For the values of axial compressive load P corresponding to |γ | < 1, buckling
can take place. These ranges are known as the pass bands.
If |tr(T)| < 2 (i.e., |γ | < 1), the eigenvalues of T are real, and x n grows expo-
nentially at the rate
$
sgn(γ ) ln |γ + γ 2 − 1|, if θ2 /θ1 = γ − sgn(γ ) γ 2 − 1
δ= (4.69)
sgn(γ ) ln |γ + γ 2 − 1|, if θ2 /θ1 = γ − sgn(γ ) γ 2 − 1
For the values of the axial load P corresponding to |γ | > 1, the larger Lyapunov
exponent is positive; therefore, buckling cannot take place. These regions are known
as the stop bands.
When the structures are randomly disordered, the transfer matrices T2 , T3 , . . . , Tn
are random matrices. The Lyapunov exponents or the localization factors γ can be
determined using the following algorithm. An arbitrary non-zero unit vector x̂ 1 is
chosen first. The state vector xi is determined iteratively. At the ith iteration,
xi = T i x̂i−1 (4.70)
xi is then normalized to give xi = xi x̂i . It is easy to show that
.
n
T n T n−1 · · · T 2 x̂ 1 = xi (4.71)
i=2
From Equation (4.59), the Lyapunov exponent or the localization factor is obtained as
1 n
δ = lim ln xi (4.72)
n→∞ n − 1
i=2
For perfectly periodic multi-span continuous beams, the Lyapunov exponents or

localization factors δ are calculated by Equation (4.69) and are plotted in Figure 4.8
for different values of the non-dimensional torsional spring stiffness t. The pass bands
are determined by |tr(T)| ≤ 2. The values of ρ corresponding to the upper ends of the
pass bands are given by 1/s = 0 and c = −1 for odd-numbered pass bands or c = 1 for
Figure 4.8 Localization factors for the periodic structure, σ L = 0.0 (after
Xie, 1995; Copyright c 1995 AIAA, reprinted with permission).
even-numbered pass bands. The values of ρ corresponding to the lower ends of the pass
bands are given by the roots of |[1 + t/(2s)]/c| = 1, in which 1/s = 0. It is seen that
the larger the value of the non-dimensional torsional spring stiffness t or the weaker
the coupling between the adjacent spans is, the smaller the width of the pass bands is.
As an extreme case, when t approaches infinity, the width of the pass bands becomes
zero; the pass bands are pass points. The pass bands corresponding to different values
of t have the same upper-end value P or ρ because the corresponding buckling modes
have zero slope at all the supports and the torsional springs do not affect the buckling
of the multi-span continuous beam.
As an example of disordered periodic structures, the lengths of the spans are as-
sumed to be uniformly distributed random numbers with a mean value L and a standard
deviation ρ L , whereas other parameters are constants. Equation (4.72) is employed to
determine the Lyapunov exponents or the localization factors numerically. In perform-
ing the simulation, for the ith iteration, a standard uniformly distributed random number
Ri is generated, the length of the ith span is calculated as L i = L(1 + σ L Ri ), and the
entries of the transfer matrix are calculated by Equation (4.72). Iteration is carried out
for a large number of transfer matrices (e.g., n = 105 ). Numerical results are plotted
in Figure 4.9 for σ L = 0.01 and in Figure 4.10 for σ L = 0.1 and different values of ti .
It can be seen that when ti = 0 (i.e., when the adjacent spans are strongly coupled),
the localization factors are very small, and localization in the buckling modes is weak.
However, if ti is large or if the adjacent spans are weakly coupled, the localization
factors are large, especially when ρτ is close to the ends of pass bands, which means
Figure 4.9 Localization factors for the disordered structure, σ L = 0.01

(after Xie, 1995; Copyright
c 1995 AIAA, reprinted with permission).
Figure 4.10 Localization factors for the disordered structure, σ L = 0.1

(after Xie, 1995; Copyright
c 1995 AIAA, reprinted with permission).
that localization in the corresponding buckling modes is strong. At the middle of the
pass band, the localization factors are relatively small, and localization in the buckling
modes is relatively weak. From Figures 4.9 and 4.10, it is also seen that the larger the
disorder in the periodicity of the structure, the larger the degrees of localization in the
buckling modes.
When the cross sections in each span of the N -span continuous beam are not
uniform, the preceding exact formulation is not applicable; an approximate formulation
has to be employed. Let us establish the equations of equilibrium of the multi-span beam
by a finite element method.
Each span of the multi-span beam is divided into M finite elements. A two-node
beam element has four degrees of freedom [i.e., two translational displacements ν1e and
ν3e and two rotational displacements ν2e and ν4e ; Figure 4.11(a)]. Hence, each span has
2M − 1 degrees of freedom as shown in Figure 4.11(b), with the global degrees of
freedom marked at each node. Each element of the beam is assumed to have a uniform
cross section; the length and the flexural rigidity of the jth element in the ith span are
L ij and EIij , respectively.
The stiffness matrix of a two-node element is given by (Xie, l995)
 
6 3L ij −6 3L ij
3L 2L 2ij −3L ij L 2ij 
2EIij  ij 
K ije =   (4.73)
L 3ij  −6 −3L ij 6 −3L ij 
3L ij L 2ij −3L ij 2L 2ij
Figure 4.11 (a) Two-node beam elements, (b) global coordinates of a multi-
span beam, (c) M -element span (after Xie, 1995; Copyright
c 1995 AIAA,
whereas the geometric stiffness matrix under axial compressive load P is

 
36 3L ij −36 3L ij
P   3L ij 4L 2ij −3L ij −L 2ij  
K ijG,e =   (4.74)
30L ij −36 −3L ij 36 −3L ij 
3L ij −L 2ij −3L ij 4L 2ij
where the superscript e denotes element.
For the span i as shown in Figure 4.11(c), let the displacement vector be νis =
{νi1 , νi2
T T
, . . . , νiTM , νi,M+1
T
}T , where νi1
T
= {νi1 }, νiTj + {νi,2 j−2 νi,2 j−1 }, j = 2, 3, . . . , M,
and νi,M+1 = {νi+1,1 }. By assembling the element stiffness matrix (4.73) and the element
T
geometric stiffness matrix (4.74), one obtains the stiffness matrix K̃ is and the geometric
stiffness matrix K̃ iG,s for span i, in which the superscript s stands for span. The equations
of equilibrium are then given by
2
K̃ i − K̃ iG,s νis = 0 (4.75)
Employing the following notation
u i1 = νi1 , u i,2 j−2 = νi,2 j−2 /l, u i,2 j−1 = νi,2 j−1
(4.76)
j = 2, 3, . . . , M, u i+1,1 = νi+1,1
and dividing equations 1, 3, . . . , 2M − 1, and 2M of Equation (4.75) by l 2 PE , and

equations 2, 4, . . . , 2M − 2 of Equation (4.75) by PE results in
s
K I − ν K iG,s u is = 0 (4.77)
where u is = {u i1 T
, u i2
T
, . . . , u im
T
, u i,M+1
T
}T , u i1
T
= {u i1 }, u iTj = {u i,2 j−2 , u i,2 j−1 }, j = 2, 3,
. . . , M, u i,M+1
T
= {u i+1,1 }, K is , and K iG,s are the non-dimensional stiffness and geomet-
ric stiffness matrices for span i, respectively, given by
 T 
K i1s s
ki1 ... 0
 s T .. 
ki1 K i2s ki2s
. 
 
 .. .. .. 
K is =  . . .  (4.78)
. s T 
. 
. s
ki,M−1 K isM ki M 
0 ... kisM K i,M+1
 T 
K i1G,s s
ki1 ... 0
 G,s T .. 
ki1 K i2G,s ki2G,s
. 
 
 .. .. .. 
K iG,s = . . .  (4.79)
. G,s T 
. G,s G,s 
. ki,M−1 K iM kiM 
0 ... G,s
kiM G,s
K i,M+1
For the elements of K is and K iG,s one should consult the study of Xie (1995). Equa-
tion (4.77) may also be written as
Ais u is = 0 (4.80)
where Ais = K is − ν K iG,s .

For the span considered, u iB = {u i1 , u i+1,1 }T are boundary degrees of freedom,
and u iI = {u i2 , u i3 , . . . , Ui,2M−1 }T are interior degrees of freedom. Using the method
of condensation, the interior degrees of freedom may be expressed in terms of the
boundary degrees of freedom. Equation (4.80) can be rearranged as
$ %
AiBB AiBI u iB
=0 (4.81)
AiIB AiII u iI
or
AiBB u iB + AiBI u iI = 0 (4.82)

AiIB u iB + AiII u iI =0 (4.83)
From Equation (4.83) u iI may be expressed in terms of u iB , for non-singular AiII , u iI =

−(AiII )−1 AiIB u iB . Substituting u iI into Equation (4.82) yields
AiB u iB = 0 (4.84)
where AiB is a 2 × 2 symmetric matrix given by

−1 αi1 βi
AiB = AiBB − AiBI AiII AiIB = (4.85)
βi αi2
To compare with the exact results obtained for uniform beams in the following
numerical examples the M elements in each span are taken to be identical; for example,
κκi1 = κi2 = · · · = κi M = κi , li1 = li2 = · · · = li M = li , and ri1 = ri2 = · · · = ri M =
ri . The elements of matrix AiB are
αi1 = αi2 + ti /M, αi2 = [2(κi − 2νi ) + Nα /D]li2

(4.86)
βi = Nβ li2 /D
where νi = νri , and for a two-element span (i.e., M = 2),

Nα = 2 3νi3 − κi νi2 + 4κi2 νi − κi3

Nβ = κi 13νi2 − 4κi νi + κi2 (4.87)
D = 2(6νi − κi )(2νi − κi )
whereas for a four-element span (i.e., M = 4),

Nα = 3 11475νi7 − 24330κi νi6 + 32890κi2 νi5 − 27248κi3 νi4

+ 11123κi4 νi3 − 2066κi5 νi2 + 160κi6 νi − 4κi

Nβ = − 10125νi7 − 33300κi νi6 + 37590κi2 νi5 − 21818κi3 νi4 (4.88)

+ 6301κi νi − 960κi νi + 64κi νi − 2κi
4 3 5 2 6

D = 8(6νi − κi ) 2025νi5 − 3645κi νi4 + 2195κi2 νi3 − 516κi3 νi2 + 42 κi4 νi − κi5
For the N -span beam as shown in Figure 4.7, assembling Equations (4.85) gives
the equations of equilibrium
  θ 
α 1 β1 ... 0 
 1 

..   

 
 θ 

β1 α2 β2 .  
2 

 . . . 
 .. .. ..  .
.  .  = 0
. (4.89)
.  

. β N −1 α N β N  θN 

 

0 ... β N α N +1 θ N +1 
where α1 = α11 , αi = αi−1,2 + αi1 , i = 2, 3, . . . , N , α N +1 = α N 2 + t N +1 /M, and

θi = u i1 , i = 1, 2, . . . , N + 1.
Equation (4.89) may be written in the form of finite difference equations. From
Equation (4.90), one writes
βi−1 θi−1 + αi θi + βi θi+1 = 0 (4.90)

$ %
θi+1
xi = T i xi−1 , xi =
θi
 α βi−1  (4.91)
i
− −
T i =  βi Bi 
1 0
Equations (4.53) and (4.91) are of identical form; the only differences are the entries of
the transfer matrices Ti . Therefore, the method employed earlier may be used to deter-
mine the localization factors for the buckling analysis. Numerical results are evaluated
and critically discussed by Xie (1995). It turns out that if only two finite elements are
taken for each span, the buckling loads in the first-pass bands are reasonably accurate.
When more finite elements are taken from each span, the buckling loads in the higher
pass bands turn out to be more accurate. Xie (1995) shows that when four elements are
taken for each span, the buckling loads in the first two pass bands are quite accurate;
the numerical results obtained from a finite element formulation turn out to agree quite
well with those obtained from an exact formulation. To sum up, the method of finite
elements is suitable for studying the localization phenomenon in buckling modes of
multi-span beams having non-uniform cross sections. Xie (1995) also utilized Green’s
function formulation, to determine the localization factor for one-dimensional disor-
dered systems, whose governing equations form a tridiagonal system. The drawback of
this method is that, when the number of finite elements is increased, the size of the ma-
trices involved becomes very large and, hence, may create a computational problem. If
a multi-span beam is randomly disordered, the localization factor turns out to be always
positive, with attendant localization in the buckling modes. When the adjacent spans
are strongly coupled, the localization factors are small, and localization in the buckling
modes is weak. If the adjacent spans are weakly coupled, the localization factors are
large, especially when ρ is close to the ends of the pass bands, with attendant strong lo-
calization in corresponding buckling modes. Generalization of the study by Xie (1995)
to rib-stiffened plates with randomly misplaced stiffeners, by the Kantorovich method,
was performed by Xie and Elishakoff (2000).
4.6 Finite Element Method for Buckling of Structures with Stochastic

Elastic Modulus
In previous sections, we concentrated on the initial geometric imperfections, thickness
variations, or misplacements in the stiffeners. A natural question arises: how to treat
the variation of the elastic moduli? As we saw in Section 2.5, the dissimilarity in elastic
moduli may cause the additional imperfection sensitivity. In general, elastic modulus is
not a constant but rather a function of axial and circumferential parameters x and y. Also,
it can be considered to be an uncertain function. If there is a sufficient data available
4.6 BUCKLING OF STRUCTURES WITH STOCHASTIC ELASTIC MODULUS 211
on elastic moduli, they can be treated as random fields of the coordinates. At present,
a large body of literature treats the elastic modulus as a random field, in the context
of the finite element method for stochastic problems (FEMSP; i.e. structures with
random elastic moduli). At present there exist three monographs, namely by Nakagiri
and Hisada (1985), Ghanem and Spanos (1991), and Kleiber and Hien (1993). Much of
the analyses on the FEMSP is concerned with the second-moment analysis of response
(i.e., computing the mean and the variance of the displacement, strain, or stress). Such
analyses involve employment of the perturbation method (Cambou, 1975), Neumann
expansion (Yamazaki and Shinozuka, 1988), Karhunen-Loeve expansion by Ghanem
and Spanos (1991), and Monte Carlo simulation by Shinozuka and his associates. For
further references, consult with the aforementioned monographs and the review papers
by Vanmarke et al. (1986), Benaroya and Rehak (1988), Brenner (1991), Ghanem and
Spanos (1997), Matthies and Bucher (1999), Der Kiureghian and Zhang (1999), and
Elishakoff and Ren (1999).
Here we will concentrate on the orthogonal series expansion of random fields.
This method is chosen because, in our analysis of random initial imperfections, we
have extensively used the orthogonal series expansion in terms of the eigenfunctions
of the appropriate linear operators (Elishakoff, 1979b, 1980b, 1983b). We will con-
centrate in this section primarily on the Karhunen-Loeve expansion representation,
extensively used in the FEMSP context by Ghanem and Spanos (1991). The Karhunen-
Loeve expansion was derived apparently independently by a number of investigators
(Karhunen, 1947; Loeve, 1948; Kac and Siegert, 1947).
The Karhunen-Loeve expansion involves a set of orthogonal deterministic func-
tions, the eigenfunctions of the auto-covariance function of the random field, and uncor-
related random variables. The Karhunen-Loeve expansion can be implemented only if
the eigenvalues and eigenfunctions of the auto-covariance function are known. The de-
termination of these eigenvalues and eigenfunctions involves the solution of an integral
equation. For illustration purposes, let us study a one-dimensional problem. Consider a
continuous random function E(x), with mean M[E(x)] and the auto-covariance func-
tion C E (x1 , x2 ); the operator of mathematical expectation is denoted as M[·], in contrast
to the previous use for this purpose of notation E[·]; the new notation is used so that
the elastic modulus is not confused with the notation of the mathematical expectation.
We expand the function E(x) in a series

∞
E(x) = M[E(x)] + ci νi h i (x) (4.92)
i=0
where ci = constant coefficients; νi = random variables with zero means; and h i (x) =
any complete set of orthogonal deterministic functions. The auto-covariance function
is expressed as
C E (x1 , x2 ) = M{E(x1 ) − M[E(x1 )]}{E(x2 ) − M[E(x2 )]}

∞
∞
= cic j Mνi ν j h i (x1 )h j (x2 ) (4.93)
i=1 j=1
Multiplying both sides of Equation (4.93) by h m (x1 ), integrating over domain of

variation of x, and using the orthogonality property of h i (x)

0, for i = m
h i (x1 )h m (x1 )d x = (4.94)
1, for i = m
yields

∞
C E (x1 , x2 )h m (x1 )d x1 = cm c j Mνm ν j h j (x2 ) (4.95)
j=1
Multiplying both sides of Equation (4.95) by h n (x2 ) and integrating over domain
with respect to x2 results in

C E (x1 , x2 )h m (x1 )h n (x2 )d x1 d x2 = cm cn Mνm νn (4.96)

Hereinafter we will follow Zhang and Ellingwood (1994). Equation (4.92) becomes a
Karhunen-Loeve expansion when the eigenvalue and eigenfunction of C E (x1 , x2 ) are
chosen as the coefficient ci2 and function h i (x), respectively. Let ci2 and h i (x) be the
ith eigenvalue and eigenfunction of the auto-covariance function C E (x1 , x2 ). Then, by
definition,

C E (x1 , x2 )h i (x2 )d x2 = ci2 h i (x1 ) (4.97)

Rearranging Equation (4.96) gives

h n (x2 )d x2 C E (x1 , x2 )h m (x1 ) d x1 = cm cn Mνm νn (4.98)

Using Equation (4.97) and the orthogonality of h i (x) yields

cm cn Mνm νn = h n (x2 )cm2 h m (x2 ) d x2

(4.99)
cm cn Mνm νn = cm2 δmn
where δmn = Kronecker delta. Thus the random variables νm in the Karhunen-Loeve
expansion are uncorrelated:
Mνm νn = δmn (4.100)
Expansion (4.92) in the applied mechanics context was used apparently for the first
time by Elishakoff (1979b), who utilized trigonometric functions
iπ x
h i (x) = sin (4.101)
l
for beams simply supported at both ends. The following functions, representing the
modes of free vibrations of a beam in vacuo, were used for beams clamped at both ends,
h i (x) = sin λi ξ − sinh λi ξ + Ai (cos λi ξ − cosh λi ξ )
(4.102)
Ai = (cos λi − cosh λi )(sin λi + sinh λi )−1
where eigenvalues are the roots of the transcendental equation
1 − (cos λi )(cosh λi ) = 0 (4.103)
4.6 BUCKLING OF STRUCTURES WITH STOCHASTIC ELASTIC MODULUS 213
and have the approximate values
λ1 = 4.730027, λ2 = 7.853185, λ3 = 10.995588,

λ4 = 14.137146, λ5 = 17.278740, λ6 = 20.4203327 (4.104)

1
λi ≈ i + π, i ≥4
2
Other representations of the random field as a sum of continuous deterministic func-
tions with random coefficients include method of shape functions and method of optimal
linear estimation. Zhang and Ellingwood (1994) used Legendre polynomials.
To elucidate the relationship between other orthogonal expansions and the
Karhunen-Loeve expansion, the correlated random variables vi in Equation (4.92) first
are transformed to uncorrelated random variables. We define a random vector V as
follows:
V T = (V1 , V2 , . . . , Vi , . . .)
= (c1 ν1 , c2 ν2 , . . . , ci νi , . . .) (4.105)
The mean vector and the variance-covariance matrix of V read

E(Vi ) = 0

(4.106)
C(Vi , V j ) = C E (x1 , x2 )h i (x1 )h j (x2 ) d x1 d x2

We want to transform the correlated random vector V into a vector of standard

uncorrelated random variables U . We first determine the eigenvectors of the variance-
covariance matrix C = [C(Vi , V j )]:
CA = A$ (4.107)
where $ diagonal matrix of eigenvalues λi ; the matrix A has columns consisting of the
corresponding eigenvectors. The correlated random vector V then can be transformed
into the vector U by
V = (A T )−1 $1/2 U (4.108)
Substituting Equation (4.108) into Equation (4.92) results in

∞
1/2

∞
(i)
E(x) = M[E(x)] + λi Ui a j h j (x) (4.109)
i=1 j=1
where {a j , j = 1, 2, . . .} is the ith column of (A T )−1 . Because U = vector of standard

(i)
independent random variables, Equation (4.109) represents a Karhunen-Loeve expan-

sion. The following part of the ith term in the expansion, after truncating the series by
N terms
N
(i)
E i (x) = a j h j (x) (4.110)
j=1
represents the ith eigenfunction of C E (x1 , x2 ).

On the other hand, the nth eigenvalue and eigenfunction that are employed in
the Karhunen-Loeve expansion also can be obtained by solving the following integral
equation (Ghanem and Spanos, 1991).

λn f n (x) = C E (x1 , x2 ) f n (x2 ) d x2 (4.111)

The numerical solution for each eigenfunction can be symbolically written as

N
(i)
f i (x) = d j h j (x) (4.112)
j=1
(i)
where d j = constant coefficients. Requiring the truncation error to be orthogonal to
each term in the expansion base results in the following eigenvalue problem (Ghanem
and Spanos, 1991)
CD = DΛ (4.113)
(i)
where $ diagonal matrix, {d j , j = 1, 2, . . . , N } is the ith column of D and

Cij = C E (x1 , x2 )h i (x1 )h j (x2 )d x1 d x2 (4.114)

The matrix of eigenvectors for the positive symmetric C must be related by
[A T ]−1 = A = D (4.115)
Thus, Equations (4.110) and (4.112) are identical to each other. This result implies
that any orthogonal expansion of a random field E(x) can be related to the Karhunen-
Loeve expansion of that random field. Expanding a random function on an orthogonal
base h n (x) is equivalent to expanding the random function using the Karhunen-Loeve
expansion, in which the eigenfunctions are determined numerically by expanding them
on the same orthogonal base.
4.7 Stochastic Finite Element Formulation by Zhang and Ellingwood

Let us turn now to the finite element formulation following Zhang and Ellingwood
(1995b). Consider a simply supported beam-column subjected to axial compression
P. The material property of interest is the flexural rigidity F = E(x)I (x), which is
modeled as a random field with mean m F and covariance function C F (x1 , x2 ). The
potential energy for each element
. 2 e 2 e 2
e
1 d w 1 dw
= F 2
d x − P dx (4.116)
2 le dx 2 le dx
in which w e = element displacement field, l e = element length. The element displace-
ments are interpolated from the element nodal degrees of freedom (Re )
w e = (N e )(Re ) (4.117)
4.7 STOCHASTIC FINITE ELEMENT FORMULATION BY ZHANG AND ELLINGWOOD 215
where N e = the shape function matrix. From Equation (4.117), ∂we /∂ x and ∂ 2 we /∂ x 2
can be obtained, respectively,
∂we d
= (N e )(Re ) = (C e )(Re )
∂x dx
(4.118)
∂ 2 we d2
= (N )(R ) = (B )(R )
e e e e
∂x2 dx2
The random rigidity F can be expanded using the truncated orthogonal series expansion:

M
F(x) = m F + Vm h m (x) (4.119)
m=1
where V = zero-mean random variables, which have the following covariance matrix:

E(Vm Vn ) = C F (x1 , x2 )h m (x1 )h n (x2 )d x1 d x2 , m, n = 0, 1, 2, . . .

(4.120)
Substituting Equations (4.118)–(4.119) into Equation (4.114), the potential energy can
then be written as
& '
. e
1 e T M
e 1
= (R ) K F + e
Vm K m F (Re ) − P(Re )T K eg (Re ) (4.121)
P
2 m=1
2

e
( K̄ e ) F = (Be )T m F (Be ) d x; K m F = (Be )T h m (x)(Be ) d x (4.122)
le le

e
K g = (C e )T (C e ) d x (4.123)
le
Applying the principle of stationary potential energy yields

M

( K̄ e ) F + Vm K me F − P K ge (R e ) = 0 (4.124)
m=1
Equation (4.124) constitutes the stiffness matrix equation for a stochastic beam element
e
with axial force P. The first-order and geometric element stiffness matrices ( K̄ ) F and
(K eg ) in Equation (4.124) are exactly the same as in the deterministic case; the stochastic
beam-column stiffness has the one additional term involving V .
The global stiffness matrix can be assembled using standard finite-element analysis:

Ne
K = (K e ) (4.125)
e=1
where Ne = total number of elements; the matrices in this summation involve the
expanded element stiffness matrices expressed in global coordinates. Finally Equa-
tion (4.124), expressed in the global coordinate system, reads:

M
( K̄ ) F + Vm (K m ) F − P(K g )(R) = 0 (4.126)
m=1
where R = global nodal displacement vector.

At bifurcation, the determinant of the stiffness matrix must vanish. Thus, the insta-
bility analysis involves finding the smallest eigenvalue of Equation (4.126). Because the
stiffness matrix involves the random variables Vm (m = 1, 2, . . . , M), the eigenvalues
and eigenvectors F also turn out to be random. Equation (4.126) can be evaluated by
perturbation analysis or by Monte Carlo simulation to obtain the probabilistic charac-
teristics of the buckling load P.
Let us now consider the perturbation method. The perturbation technique has been
implemented in stochastic finite-element analysis by several researchers (e.g., Nakagiri
and Hisada, 1985; Liu, Belytschko, and Mani, 1984). In the following, the second-
order perturbation method is used to identify the bifurcation loads involving random
parameters.
Equations (4.126) can be converted to the following general form:
[K − λ(K g )](R) = 0 (4.127)

L
K = K̄ + αi (K i ) (4.128)
i=1
λ= P (4.129)
where L = the total number of random variables; K̄ = the stiffness matrix based on
mean values of the random fields; K g = the geometric stiffness matrix; α = a set of
zero-mean correlated random variables; and K i = the set of corresponding stiffness
matrices. Here, K̄ , K i , and K g are all deterministic.
The stiffness matrix K involves the random variables α(i = 1, 2, . . . , L). Using the
mean-centered second-order perturbation method, the stiffness matrix K , eigenvalues
λ, the eigenvectors R can be expanded in a Maclaurin series with respect to the random
variables αi :

L
1 L L
II
K = K0 + K iI αi + K ij αi α j + · · · (4.130)
i=1
2 i=1 j=1
L
I 1 L L
II
λ = λ0 + λi αi + λ αi α j + · · · (4.131)
i=1
2 i=1 j=1 ij

L
1 L L
II
R = R0 + RiI αi + R αi α j + · · · (4.132)
i=1
2 i=1 j=1 ij
where
K 0 = K̄ (4.133)

∂ K
Ki =
I
= Ki (4.134)
∂αi α=0

∂ 2 K
K ij =
II
=0 (4.135)
∂αi ∂α j α=0
and similarly for R. Substituting Equations (4.130)–(4.132) into Equation (4.127), the
random eigenvalues λ and eigenvectors R are found recursively as
Zero order:
[(K 0 ) − λ0 (K g )](R0 ) = 0 (4.136)
First order:

(R0 )T (K g )(R0 )λiI = (R0 )T K iI (R0 ) (4.137)

[λ0 (K g ) − (K 0 )] RiI = K iI − λiI (K g ) (R0 )) (4.138)

Second order:

(R0 )T (K g )(R0 )λIIij = (R0 )T K iI − λiI (K g ) R Ij + (R0 )T K Ij − λ Ij (K g ) RiI
(4.139)

[λ0 (K g ) − (K 0 )] RIIij

= −λij (K g )(R0 ) + K iI − λiI (K g ) R Ij + K jI − λli (K g ) RiI (4.140)
The mean and the variance of the buckling load can be obtained from Equa-
tion (4.131) using either the first-order or second-order approximation. The first-order
or second-order approximation is obtained by considering the first two terms on the
right-hand side of Equation (4.131), which results in
E(λ) ≈ λ0 (4.141)

I
I
Var(λ) ≈ λiI λ Ij E(αi α j ) (4.142)
i=1 j=1
Similarly, the second-order approximation is obtained as

L
L
E(λ) ≈ λ0 + λIIij E(αi α j ) (4.143)
i=1 j=1

L
L
1 L L L L
Var(λ) ≈ λiI λ Ij (αi α j ) + [E(αi αl )E(α j αk )
i=1 j=1
4 i=1 j=1 k=1 l=1
+ E(αi αk )E(α j αl )]λIIij λIIkl
(4.144)
In the second term in Equation (4.144), the fourth moments have been expressed in
terms of second moments using a relation that is only exact when the random vector
(αi ) is Gaussian.
To evaluate the accuracy and any limitations in the preceding formulations and to
investigate the effects of uncertain material properties on elastic structural stability, con-
sider buckling of a simply supported column. Monte Carlo simulation is used to validate
the results obtained from the perturbation analysis by Zhang and Ellingwood (1995b).
The random fields in the following examples are assumed to be weakly homo-
geneous and are described in the second-moments sense by the exponential
auto-correlation function:

−|x1 − x2 |
C(x1 , x2 ) = σ exp
2
(4.145)
DEI
where the parameter DEI is commonly known as the correlation length and σ 2 is the
variance of the random field. For DEI tending to infinity, the random field reduces to a
random variable; for DEI tending to zero, the random field becomes an ideal white noise,
under some conditions. Without loss of generality, the random field domain is defined
as (−1, 1). The Legendre polynomials are chosen as the orthogonal functions h(x) to
represent the basis of the random field (Zhang and Ellingwood, 1994). Figure 4.12
illustrates the exponentially decaying auto-correlation function and its approximation
recovered from evaluating the Legendre polynomial series expansion of the random
field. The number of terms M in the expansion is chosen such that the integral error
measure defined as

1 1 M M
M = C(x1 , x2 ) − E(Vi , V j )h i (x1 )h j (x2 )d x1 d x2 (4.146)
A σg 2
i=0 j=0
is less than 0.005. In Equation (4.146), A is the area of the domain of the covariance
function, and σg2 is the variance of the random field. The correlation lengths 24.0
and 0.15 (see Figure 4.12) are used as the lower- and upper-limit correlation lengths
in numerical calculations. Intensities at any two points in the random field can be
Figure 4.12 Normalized exponential auto-correlation function: exact ex-

pression versus results obtained via Legendre expansion (after Zhang
and Ellingwood, 1995; Copyright c 1995 ASCE, reprinted with permis-
sion).
Figure 4.13 Convergence of the mean value of the buckling load of a

stochastic simply supported column with number of elements (after
Zhang and Ellingwood, 1995; Copyright
c 1995 ASCE, reprinted with
permission).
considered as almost perfectly correlated when DEI = 24, in which case the random
field can be approximated simply as a random variable. Conversely, intensities within
the random field are weakly correlated, and spatial fluctuations are relatively large when
DEI = 0.15. For a constant M , the number of terms required to represent the random
field increases as γ decreases (Zhang and Ellingwood, 1994).
As an example, a simply supported column wilh stochastic flexural rigidity, sub-
jected to the axial load P, is investigated. The flexural rigidity EI(x) is assumed to be a
weakly homogenous Gaussian random field with mean m F and coefficient of variation
ρEI . Figure 4.13 shows the convergence of the mean of the buckling load Pcr when
ρ F = 0.3 as the number of finite elements is increased. The results are presented in
terms of the non-dimensional parameter Pcr L 2 /m F , so that this parameter equals π 2
when the column is deterministic. The convergence of Pcr occurs more slowly as the cor-
relation length of the random field decreases (i.e., a finer finite-element mesh is needed
to achieve convergent results when intensities of the random field at two points a given
distance apart become uncorrelated). With the continuous representation of the random
field, there is no need to discretize the field by a mesh of random variables, and the
orthogonal series expansion of the random field is incorporated in the finite-element for-
mulation [see Equation (4.124)]. It is apparent from Figure 4.13 that the finite-element
mesh necessary to achieve satisfactory results is affected by the random field, partic-
ularly for random fields with short correlation lengths. This occurs because the shape
functions used in the finite-element analysis are not exact; as γ decreases, either more
finite elements or higher order shape functions are necessary to achieve the same level
Figure 4.14 Influence of the correlation length and coefficient of variation of

elastic stiffness on probabilistic characteristics of the buckling load of a sim-
ply supported column: (a) mean value of buckling load and (b) coefficient of
variation of buckling load (after Zhang and Ellingwood, 1995; Copyright c
1995 ASCE, reprinted with permission).
of accuracy. The results from the second-order perturbation analysis appear to agree
well with results of Monte Carlo simulation, but the accuracy of the second-order per-
turbation analysis decreases as the correlation length decreases. Based on these results,
the column was modeled with 10 finite elements by Zhang and Ellingwood (1995b).
Figure 4.14 illustrates the effects of the correlation length DEI and the coefficient
of variation ρEI on the buckling load statistics. As DEI decreases, the mean value and
the coefficient of variation of Pcr decrease, indicating that higher spatial fluctuations in
the rigidity of the column tend to reduce both the mean value and the variability of Pcr .
Conversely as the correlation length becomes large, the mean of Pcr L 2 /m F approaches
π 2 , which is what one would expect if EI were treated as a random variable. As ρEI
increases (i.e., the variability in the intensity of the random field increases), the mean
value of Pcr decreases, and the coefficient of variation of Pcr increases. The second-
order perturbation results again agree well with the Monte Carlo simulation results, but
the accuracy of the perturbation analysis decreases as either DEI or ρEI increases. Note
that the Kahrunen-Loeve expansion method was utilized recently by Schenk, Schuëller
and Arbocz (2000a, 2000b) to deal with buckling analysis of cylindrical shells with
random imperfections.
We conclude this chapter on stochastic buckling by observing that according to
some researchers probabilistic considerations may prove to be useful not only for design
of structures with imperfections, but also for attacking, as Budiansky and Hutchinson
(1979) designate it, “. . . the unresolved, nagging problem of the quest for a basic,
general stability theorem.” In their elegant overview of modern problems of buckling,
these authors mention:
But there is no doubt that at least an esthetic problem remains, and that a new, congenial
definition of stability is desirable. Perhaps some statistical concepts may be fruitful.
When the second-variation method fails, it appear that the minimum in ϕ is destroyed
only by presence of obscure secret passages in function space into which no self-
respecting structure would venture except by wildly improbable accident. Accordingly,
an appropriately defined probability of failure should, under these circumstances, be
absurdly low. But we do not have any helpful suggestions concerning this definition,
which, in order to be useful in the assessment of the practical stability of structure,
should permit easy evaluation of the desired probability.
For the discussion of the fundamental dilemma in the theory of elastic stabil-
ity, readers are referred to articles by Koiter (1965, 1975, 1976), Budiansky (1974),
Como and Grimaldi (1975, 1995), Budiansky and Hutchinson (1979), and Potier-
Ferry (1981b). Interrelation between the concepts of stability and stochasticity is dis-
cussed by Bolotin (1967). Potier-Ferry (1987) stresses that
. . . difficulty no longer exists within Kelvin-Voigt viscoelasticity . . . or within theory
of beams with moderate rotations (Ball, 1974). Since a century, the energy criterion has
been extensively applied. A discrepancy between theory and experiment has never been
explained by removing this criterion. Hence, those mathematical difficulties should not
be used to question the validity of this stability test. On the contrary, mathematics use to
comply to physical evidence and some norms or some constitutive assumptions must be
rejected of the corresponding notion of stability is not equivalent to the energy criterion.
It remains to be seen if the notion of probability will prove useful for the very
foundation of buckling of structures; yet we illustrated that the probabilistic concepts
are useful for a more modest objective, namely, to provide the explanation both of the
knockdown factors and for their numerical determination, based on the data derived
from the initial imperfection data banks.
For additional aspects of stochastic imperfection sensitivity of structures, consult
papers by Arbocz and Hol (1995), Stam (1996), Stam and Arbocz (1997), as well as
review articles by Chryssanthopoulos (1997) and Elishakoff (1998).
CHAPTER FIVE
Anti-Optimization in Buckling of Structures
So far as the laws of mathematics refer to reality, they are not certain. And so far
as they are certain, they do not refer to reality.
A. Einstein
To a person who is studying algebra, it is often more useful to solve the same
problem with three or four different methods, than to solve three or four different
problems. By solving problems by different methods, one can by the comparison
clarify which of them is shorter and more effective.
W. W. Soyer
The subject of probability is over two hundred years old and for the whole period
of its existence there has been dispute about its meaning.
D. V. Lindley
A thousand probabilities do not make one truth.

English proverb
Probability does not exist.

B. de Finetti
I see and approve better things, but follow worse.

Publius Ovidius Naso
In a traditional probabilistic analysis, the statistical parameters of uncertain quantities –

initial geometric imperfections or elastic moduli – are presumed to be known, which must
be inferred from on-site measurements. Because the available data of such measurements
are often limited to permit the probabilistic analysis, a new discipline, called convex mod-
eling of uncertainty, is applied to obtain estimates of the upper and lower bounds of the
buckling loads. From a structural safety point of view, the least favorable lower buckling
load should be used in design. Critical comparison of the probabilistic and convex analyses
is performed.
222
5.1 INCORPORATION OF UNCERTAINTIES IN ELASTIC MODULI 223
5.1 Incorporation of Uncertainties in Elastic Moduli

Previous studies on plates and shells made of composite have been based on the assump-
tion that the properties of the materials are characterized by predetermined elastic mod-
uli, and no uncertainties of these moduli have been taken into account (Tennyson, 1975;
Vinson and Sierakowski, 1986; Whitney, 1987). However, the composite materials
are always subject to a certain amount of scatter in their measured elastic moduli
(Tewary, 1978). Such uncertainties in elastic moduli of composite materials are due
to the many factors that influence the actual properties of composites. For example,
among other things, misalignment of fibers or imperfect bonding between the fibers
and the matrix may contribute to the scattered values of the measured elastic moduli.
To a large extent, the properties of composite materials are dependent on the fabrica-
tion process. But even the composite materials manufactured by the same process may
demonstrate significant differences in their elastic properties. Therefore one should be
aware of the potential variations in load-carrying capacity of composite structure that
can arise due to the uncertainty in elastic moduli. A more realistic analysis of composite
structures should be performed with the variations of the elastic moduli being taken
into consideration at the same time.
The effect of scatter in material properties on buckling of structures is usually
treated within the realm of the new tool in stochastic analysis of structures, namely the
finite element method for stochastic structures (Ramu and Ganesan, 1993; Zhang and
Ellingwood, 1994, 1995a, 1995b, 1995c, 1995d) as discussed in Chapter 4. However,
as Shinozuka (1987) mentions,
it is recognized that it is rather difficult to estimate experimentally the autocorrelation

function, or in the case of weak homogeneity, the spectral density function of the
stochastic variation of material properties. In view of this, the upper bound results are
particularly important, since the bounds derived . . . do not require knowledge of the
autocorrelation function.
A somewhat analogous idea was expressed by Ariaratnam (private communica-

tion, 1992). In the stochastic context, Shinozuka and Deodatis (1988) and Deodatis and
Shinozuka (1989) derived upper and lower bounds on the probabilistic characteristics
of the response in terms of probabilistic characteristics of the material variability.
The present investigation represents, to the best knowledge of the authors, the
first study in the literature, which develops analytical tools to incorporate the uncer-
tainties in elastic moduli into analysis. Here, we deal with the buckling problems of
laminated plates and shells by use of a convex modeling of uncertainty (Ben-Haim
and Elishakoff, 1990; Elishakoff et al., 1994; Elishakoff, 1995). Both the upper and
lower bounds of the buckling load are derived so that the designer may have a better
understanding of the actual load-carrying capacities possessed by these structures.
5.1.1 Basic Equations

We start our discussion from the buckling of axially compressed composite shells.
Here we use the Donnell-Mushtari-Vlasov shell theory (Timoshenko and Gere, 1961).
224 ANTI-OPTIMIZATION IN BUCKLING OF STRUCTURES
Strain-displacement relations are
∂u ∂ 2w
x = , κx = − 2
∂x ∂x
∂v w ∂ 2w
y = + , κy = − 2 (5.1)
∂y R ∂y
∂v ∂u ∂ 2w
γx y = + , κx y = −2
∂x ∂y ∂ x∂ y
where x and y are the axial and circumferential coordinates in the shell middle surface;
u and v are the shell displacement along axial and circumferential directions; w is the
radial displacement, positive outward; x , y , and γx y are strain components; κx , κ y , and
κx y are middle surface curvatures of the shell; and R is the radius of the cylindrical
shell. The constitutive relations for the composite laminate read
    
 Nx  A11 A12 A16 B11 B12 B16  x 

 
 
 


 Nθ     
B26  y  

  
 A12 A22 A26 B12 B22 
 

N 
    
 A16 A26 A66 B16 B26 B66  x y 
= 
xy
 (5.2)


 Mx  
  B11 B12 B26 D11 D12 D16 
κx  
M 
   





 y   B12 B22 B26 D12 D22 
D26  κ 


 
  
 y

Mx y B16 B26 B66 D16 D26 D66 κx y
where N x , N y , and N x y are stress resultants; Mx , M y , and Mx y are bending and twisting
moments, acting on a laminate; the laminate stiffnesses Ai j , Bi j , and Di j are defined as
h/2
(k)
(Ai j , Bi j , Di j ) = Q̄ i j (1, z, z 2 ) dz
−h/2
where h is the total thickness of the laminate, and z is the coordinate in the direction of
the laminate thickness; Q̄ i j are the transformed reduced stiffnesses and can be expressed
in terms of the lamina orientation and four independent engineering material constants
in principal material directions [i.e., E 1 , E 2 , ν12 , and G 12 (Jones, 1975)].
The equations governing the buckling of the cylindrical shell under axial compres-
sion read
∂ Nx ∂ Nx y
+ =0
∂x ∂y
∂ Nx y ∂ Ny
+ =0 (5.3)
∂x ∂y
∂ 2 Mx ∂ 2 Mx y ∂ 2 My 1 ∂ 2w
+ 2 + − N y − N x =0
∂x2 ∂ x∂ y ∂ y2 R ∂x2
where N x is axial loading applied at the ends of the shell.

Using Equations (5.1) and (5.2), Equation (5.3) can be written as

L̄ ȳ = f¯
In Equation (5.4),
 
L 11 L 12 L 13
 
L̄ =  L 12 L 22 L 23 
L 13 L 23 L 33
 

u 
ȳ = v (5.4)

  
w
 

 0  
 
f = 0
 2 
 Nx ∂ w 
 
∂x2
where operators L i j are
∂ ∂2 ∂2
L 11 = A11 + 2A 16 + A 66
∂x2 ∂ x∂ y ∂ y2
∂2 ∂2 ∂2
L 12 = A16 + (A 12 + A 66 ) + A 26
∂x2 ∂ x∂ y ∂ y2

1 ∂ ∂ ∂2 ∂3
L 13 = A12 + A26 − B11 2 − 3B16 2
R ∂x ∂y ∂x ∂x ∂y
∂3 ∂3
− (B12 + B66 ) − B26
∂ x∂ y 2 ∂ y3
∂2 ∂2 ∂2
L 22 = A66 + 2A 26 + A 22 (5.5)
∂x2 ∂ x∂ y ∂ y2

1 ∂ ∂ ∂3
L 23 = A26 + A22 − B22 3
R ∂x ∂y ∂y
∂3 ∂3 ∂3
− 3B26 − (B12 + 2B 66 ) − B16
∂ x∂ y 2 ∂ x 2∂ y ∂x3

2 ∂2 ∂2 ∂2 A22 ∂4
L 33 =− B12 2 + 2B26 + B22 2 + 2 + D11 4
R ∂x ∂ x∂ y ∂y R ∂x
∂4 ∂4 ∂4 ∂4
+ 4D16 + 2(D 12 + 2D 66 ) + 4D 26 + D 22
∂ x 3∂ y ∂ x 2∂ y2 ∂ x∂ y 3 ∂ y4
We consider the axially compressed cylindrical shell with simply supported bound-
ary conditions. In this case, the boundary conditions are represented by
w = Mx = v = N x = 0 at x = 0, L (5.6)
The preceding boundary conditions are satisfied by the following displacement func-
tions:
 mπ x ny 
  
U cos cos 



mn
L R
 u
   ∞  
∞
mπ x ny 
v = Vmn sin sin (5.7)

   m=1 n=0 
 L R 
w 
 


 Wmn sin mπ x cos ny 
L R
Similar to Hirano (1979), here the coupling stiffnesses (A16 , A26 , B16 , B26 , D16 , D26 )
are assumed to be zero. They are actually zero for symmetric cross-ply laminates. As
for symmetric angle-ply laminates, B16 and B26 are zero, and A16 , A26 , D16 , and D26
can be neglected for laminates with many layers.
Substitution of Equations (5.1), (5.2), and (5.7) into Equation (5.3) leads to a set
of homogeneous linear algebraic equations, and the existence of non-trivial solutions
requires that the determinant of the coefficient matrix vanishes,
 
C11 C12 C13
C 
 21 C22 C23 
det  =0 (5.8)
 mπ 2 
C31 C32 C33 − Ncl
L
where
2
mπ 2 n
C11 = A11 + A66
L R
2
n mπ 2
C22 = A22 + A66
R L
4 4
mπ mπ 2 n 2 n
C33 = D11 + 2(D12 + 2D66 ) + D22
L L R R

A22 B22 n 2 B12 mπ 2
+ 2 +2 +2 (5.9)
R R R R L

mπ n
C12 = C21 = (A12 + A66 )
L R
3
mπ 2 n A22 n n
C23 = C32 = (B12 + 2B66 ) + + B22
L R R R R
3 2
A12 mπ mπ mπ n
C13 = C31 = + B11 + (B12 + 2B66 )
R L L L R
Thus, we arrive at the classical buckling load,

L 2 C11 C22 C33 + 2C12 C23 C13 − C13 2
C22 − C23
2
C11 − C12
2
C33
Ncl = (5.10)
mπ C11 C22 − C12
2
To determine the critical buckling load, Ncl , for a cylindrical shell with given
dimensions and material properties, one determines those integer values of m and n
which make Ncl a minimum.
Let us now consider the case of the symmetrically laminated plate subjected to
uni-axial loading. This case could be easily handled by putting R → ∞ and Bi j = 0
in Equation (5.4). The governing equation becomes
∂ 4w ∂ 4w ∂ 4w ∂ 4w ∂ 4w
D11 + D16 3 + 2(D12 + D66 ) 2 2 + 4D26 + D22 4
∂x 4 ∂x ∂y ∂x ∂y ∂ x∂ y 3 ∂y
∂ 2w
= Nx (5.11)
∂x2
In the simplest case, when the behavior of the composite plate is of a special
orthotropy, the coupling terms D16 and D26 vanish, and the governing equation (5.11)
reduces to
∂ 4w ∂ 4w ∂ 4w ∂ 2w
D11 + 2(D 12 + D 66 ) + D 22 = N x (5.12)
∂x4 ∂ x 2∂ y2 ∂ y4 ∂x2
The following displacement function satisfies the simply supported boundary con-
ditions:
∞ ∞
mπ x nπ y
w(x, y) = Amn sin sin (5.13)
m=1 n=1
a b
where a and b are the length and width of the plate, and m and n are integers, denoting
the buckling wave numbers along x and y directions, respectively.
Substituting Equation (5.13) into Equation (5.12) yields
4 2 2 4
π 2a2 m m n n
Ncl = − 2 D11 + 2(D12 + 2D66 ) + D22 (5.14)
m a a b b
It is clear that the integer n should be equal to unity in order to result in the lowest
value for the buckling load. However, the value of the integer m depends on the ratio
a/b of side lengths as well as the material properties of the plate considered. In practice,
m can be determined by a search for a smallest value of the buckling load Ncl .
For the general anisotropic plate, the twisting coupling stiffnesses D16 and D26
may not be neglected. In these cases, the boundary conditions for the simply supported
plates become
∂ 2w ∂ 2w ∂ 2w
w = Mx = −D11 − 2D 16 − D 12 =0 at x = 0, a (5.15)
∂x2 ∂ x∂ y ∂ y2
∂ 2w ∂ 2w ∂ 2w
w = M y = −D12 − 2D 26 − D 22 =0 at y = 0, b (5.16)
∂x2 ∂ x∂ y ∂ y2
Due to the presence of stiffness moduli D16 and D26 in the governing equation (5.11)
and the boundary conditions, the closed form solution is unattainable. Whitney (1987)
employs a method of weighted residuals to obtain an approximate solution of the
problem. Using the displacement function as described by Equation (5.13) and taking
the boundary conditions (5.11) into account, the equation of weighted residues becomes
[the misprints of Whitney (1987) relative to Equations (5.17) and (5.18) have been
corrected here]
b a
∂ 4w ∂ 4w ∂ 4w ∂ 4w
D11 4 + 4D16 3 + 2(D12 + 2D66 ) 2 2 + 4D26
0 0 ∂x ∂x ∂y ∂x ∂y ∂ x∂ y 3

∂ 4w ∂ 2w mπ x nπ y
+ D22 4 + Ncl 2 sin sin d xd y
∂y ∂x a b
a 2 2
∂ w ∂ w nπ mπ x
− 2D26 (−1) n
− sin dx
0 ∂ x∂ y y=b ∂ x∂ y y=0 b a
b 2 2
∂ w ∂ w mπ nπ y
− 2D16 (−1) m
− sin dy
0 ∂ x∂ y x=a ∂ x∂ y x=0 a b
$
m = 1, 2, . . . , M
=0 (5.17)
n = 1, 2, . . . , N
Further substitution of Equation (5.13) into Equation (5.17) yields the following
set of homogeneous linear algebraic equations:

m 2a2
π D11 m + 2(D12 + 2D66 )m n R + D22 n R − Ncl 2 Amn
4 4 2 2 2 4 4
π

M
N
− 32mn Rπ 2 Mi j [(m 2 + i 2 )D16 + (n 2 + j 2 )D26 R 2 ]Ai j = 0 (5.18)
i=1 j=1
(m = 1, 2, . . . , M; n = 1, 2, . . . , N )
where

ij m ± i odd
Mi j =
(m − i )(n − j )
2 2 2 2
n ± j odd

i = m, m ± i even
=0 (5.19)
j = n, n ± j even
a
R=
b
Again, the non-trivial solution of Equation (5. 18) depends on the condition that the
determinant of the coefficient matrix must vanish, from which the uni-axial buckling
load Ncl can be determined.
5.1.2 Extremal Buckling Load Analysis

The classical buckling load of the structure depends on the four basic elastic moduli,
mentioned in Section 5.1.1. For the sake of generality, in the following analysis, a
generic formula for the classical buckling load is adopted instead of relying on more
concrete expressions, such as Equations (5.10) and (5.14).
Suppose that the classical buckling load Ncl takes the form
Ncl = F(E 1 , E 2 , ν12 , G 12 ) (5.20)
or, more simply,
Ncl = F(E i ), (i = 1, 2, 3, 4) (5.21)
where E 3 = ν12 and E 4 = G 12 . The function F in Equation (5.21) also depends on the
form of structure (plate or shell), boundary conditions as well as geometric properties.
Let E io be the nominal values of the elastic moduli, which might be visualized as
the average values of those elastic moduli data from measurements. Then, the elastic
moduli of values slightly different from those nominal values could be denoted as
E io + δi , δi being small quantities. The classical buckling load corresponding to these
elastic moduli can be written, retaining only the first order in δi , as follows:

o o 4
∂ F E io
F E i + δi = F E i + δi (5.22)
i=1
∂ Ei
We introduce the following notations:
o o o
∂ F E o
i ∂ F E i ∂ F E i ∂ F Ei
fT = , , ,
∂ E1 ∂ E2 ∂ E3 ∂ E4 (5.23)
δ T = (δ1 , δ2 , δ3 , δ4 )
where T denotes transpose operation. Then Equation (5.22) can be rewritten as

F E io + δi = F E io + f T δ (5.24)
The deviation δ from the nominal elastic moduli is assumed to vary in the following
ellipsoidal set:
$ %
4
δi2
Z (α, e) = δ: ≤α 2
(5.25)
e2
i=1 i
where the size parameter α and the semi-axes ei (i = 1, 2, 3, 4) are based on the exper-
imental data available and will be discussed later.
The problem is formulated as follows: given an ellipsoid of the elastic moduli
[Equation (5.25)], find the extremal buckling load,

Next = extremum F E io + f T δ (5.26)
δ∈Z (α,e)
In Equation (5.26), Next is the lowest or highest buckling load of the composite
structure with the elastic moduli varying within the range of the ellipsoidal set Z .
Because Equation (5.26) calls for finding the extremum of the linear functional f T δ
on the convex set Z (α, e), the extremal values take place on the set of extreme points
(Ben-Haim and Elishakoff, 1990), or the boundary, of Z , that is on the ellipsoidal shell
defined as follows:
$ %
4
δi2
C(α, e) = δ: 2
= α2 (5.27)
i=1
ei
Thus, the extremum buckling load becomes

Next = extremum F E io + f T δ (5.28)
δ∈C(α,e)
The analysis of the extremal buckling load is quite similar mathematically to

the other problems considered by Ben-Haim and Elishakoff (1990) and Elishakoff
et al. (1994). Let us define as an 4 × 4 diagonal matrix whose ith diagonal element
is 1/ei2 . Then the constraint given by Equation (5.27) becomes
δ T δ − α 2 = 0 (5.29)
We use the method of Lagrange multipliers, and define the Lagrangean as
H (δ) = f T δ + λ(δ T δ − α2 ) (5.30)
where λ is the Lagrange multiplier. For the extremum, the derivative of the Hamiltonian
must vanish,
∂H
= 0 = f T + 2λδ (5.31)
∂δ
or
1 −1 T
δ=− f (5.32)
2λ
In view of Equation (5.29), we have
1 T −1
λ2 = f f (5.33)
4α 2
and
α
δ = ± −1 f (5.34)
T −1
f f
It follows from Equation (5.34) that the maximum and minimum buckling loads
have the following expression:
!
Nmax
= F E io ± α f T −1 f
Nmin
/
0 4 & '2
o 0 ∂ F E io
= F Ei ± α 1 ei (5.35)
i=1
∂ Ei
From Equation (5.35), the upper and lower bounds of the critical buckling load are
calculated. Equation (5.35) shows explicitly that the uncertainties in elastic moduli have
a direct effect on the value of the buckling load. It is remarkable that the semi-axes of
the uncertainty ellipsoid, as well as the sensitivity derivatives are directly incorporated
in Equation (5.35) to yield the least and most favorable buckling loads due to the
uncertainty in elastic moduli.
5.1.3 Determination of Convex Set from Measured Data

Before any prediction can be made on the critical buckling load of the composite plate
or shell, the values of α and ei (i = 1, 2, 3, 4) should be known in advance. In fact, these
values are dependent on the manufacturing process by which the composite structure has
been fabricated. It is understandable that the more advanced the manufacturing process
and the better the workmanship, the smaller these parameters will be in value. Besides,
the evaluation of these parameters is also linked with the amount of information one
has about the properties of the composite considered; in other words, with increased
information on the experimental data, more precise calibration of the parameters can
be performed.
In the paper by Goggin (1973), some measurements of the elastic moduli for
the composite materials were described; they clearly show a scatter of the measured
data for the elastic moduli. In particular, for the transverse and longitudinal Poisson’s
ratio, the experimental values have a large scatter. In this study, the parameter α is set
equal to unity. As long as α is fixed, the other parameters could be readily determined
by the evaluation of the existent experimental data. Normally, if a sufficient amount
of experimental data is available, the average value of these data could be used as
the nominal value E i0 for the corresponding elastic modulus, whereas parameters ei s
could be chosen as the proper deviations from the average values of the corresponding
measured data. Even if one has only limited knowledge of the uncertain elastic moduli,
it is still possible to judge what value the maximum buckling load of the composite
structure will be, provided the variation ranges of these elastic moduli are known.
Suppose that we deduce from the experimental data that the elastic moduli are
varying in the following ranges, respectively:
E iL ≤ E i ≤ E iU (i = 1, 2, 3, 4) (5.36)
where E iU and E iL correspond to upper and lower bounds of the elastic moduli E i ,
respectively.
We introduce the following notations:
E iU + E iL
E io =
2
(5.37)
E iU − E iL
i =
2
Then Equation (5.36) can be rewritten as

0
E i = E i + δi
(5.38)
|δi | ≤ i
The second of Equation (5.38) describes a “box.” To use the theory developed in
the last section, we need to enclose this “box” by an ellipsoid,

4
δ2 i
≤1 (5.39)
i=1
ei2
The question arises as to how to determine the semi-axes of this ellipsoid. Naturally,
such an ellipsoid should have a minimum volume. The volume of the preceding ellipsoid
is given by
V = Ce1 e2 e3 e4 (5.40)
where C is a constant.
Because the corner points of the “box” [Equation (5.38)] should be on the surface
of the ellipsoid, we have
21 22 23 24
+ + + =1 (5.41)
e12 e22 e32 e42
We are interested in minimizing the volume V of the ellipsoid, defined by Equa-
tion (5.40), subject to the constraint of Equation (5.41). To do this, we use the Lagrange
multiplier technique. The Lagrangean L reads,
& '
21 22 23 24
L = Ce1 e2 e3 e4 + λ 2 + 2 + 2 + 2 − 1 (5.42)
e1 e2 e3 e4
Requirements
∂L
=0 (i = 1, 2, 3, 4) (5.43)
∂ei
lead to equations
2λ 21
Ce2 e3 e4 − =0 (5.44)
e13
2λ 22
Ce1 e3 e4 − =0 (5.45)
e23
2λ 23
Ce1 e2 e4 − =0 (5.46)
e33
2λ 24
Ce1 e2 e3 − =0 (5.47)
e43
We multiply Equation (5.44) by e1 , Equation (5.45) by e2 , Equation (5.46) by e3 ,
Equation (5.47) by e4 , and sum up all four equations to yield
& '
21 22 23 24
4V − 2λ 2 + 2 + 2 + 2 = 0 (5.48)
e1 e2 e3 e4
Bearing in mind Equation (5.41), Equation (5.48) becomes
λ = 2V (5.49)
Substituting Equation (5.49) into Equation (5.44) leads to
V 2
− 4 31 V = 0 (5.50)
e1 e1
which implies that
e1 = 2 1 (5.51)
Likewise, we can determine the other semi-axes of the ellipsoid as

e2 = 2 2 , e3 = 2 3 , e4 = 2 4 (5.52)
Thus, once the size of the “box” containing the experimental data is known, the semi-
axes of the ellipsoid can be determined, and the analysis of convex modeling can be
carried out.
5.1.4 Numerical Examples and Discussion

We will now proceed to investigate several cases of the buckling problem of composite
plates and shells with a view of gaining some insight into the effect of uncertainties in
the material properties on the load-carrying capacity of these structures.
In the following analysis, the experimental data for the elastic moduli are based on
the available data (Goggin, 1973). The material of the lamina is composed of carbon
fibers and resin matrix, with a volume fraction of fiber being 40%. From the figures in
the preceding study, the following data were deduced:
E 1U = 100 Gpa (14.5 × 106 psi), E 1L = 90 Gpa (13.0 × 106 psi)
E 2U = 7.3 Gpa (1.06 × 106 psi), E 2L = 6.8 Gpa (0.99 × 106 psi)
(5.53)
E 3U = 0.28, E 3L = 0.22
E 4U = 3.2 Gpa (0.46 × 106 psi), E 4L = 2.6 Gpa (0.38 × 106 psi)
From these data, the nominal elastic moduli E i0 and the semi-axes ei can be evalu-
ated, by using Equations (5.32), (5.46), and (5.47), as the following:
E 1o = 13.75 × 106 psi, e1 = 1.5 × 106 psi
E 2o = 1.03 × 106 psi, e2 = 0.07 × 106 psi
(5.54)
E 3o = 0.25, e3 = 0.06
E 4o = 0.42 × 106 psi, e4 = 0.08 × 106 psi
Let us concentrate first on a specially orthotropic laminated rectangular plate sub-
ject to the simply supported boundary conditions. We investigate cross-ply laminates,
in which fibers of adjacent plies are oriented at 90◦ to each other and parallel to the
plate edges. The critical buckling load for the uni-axially compressed plate is given by
Equation (5.14). Note here that the basic elastic moduli E i are implicitly contained in
the flexural stiffness Di j .
Here, the width b is fixed at 10 in., while the length a varies, and the thickness of
each lamina is 0.012 in. The following three cases are studied:
Case 1: A 5-ply plate, with laminate configuration (0◦ , 90◦ , 0◦ , 90◦ , 0◦ ).
Case 2: An 11-ply plate, with odd-numbered layers oriented at 0◦ and even-

numbered layers oriented at 90◦ .
Case 3: A set of l0-ply square plates (a = b = 10 in.), with laminate configuration

as [θ/−θ/θ/−θ/θ]s , θ ranging from 0◦ to 90◦ .
Figure 5.1 Uncertainty in buckling load for a 5-ply laminated plate (Case 1).
The analysis described in previous sections were carried out for these three cases
and numerical results are shown in Figures 5.1, 5.2, and 5.3. In the first two cases, the
analyses were relatively straightforward. However, the calculations in Case 3 were more
involved; here the results in Figure 5.3 were calculated with M = N = 12 in Equa-
tion (5.18), and the differentiations as required by Equation (5.35) were performed
numerically. From these figures, one can see that for different dimensions of the plate,
Figure 5.2 Uncertainty in buckling load for an 11-ply laminated plate (Case 2).
Figure 5.3 Uncertainty in buckling load for a set of 10-ply laminated plates (Case 3).
the effect of uncertainty in elastic moduli on the buckling load is different. The per-
centagewise variability is defined as
vu − vl
ν= × 100% (5.55)
2vn
where vu , vl , and vn are the upper bound, the lower bound, and the nominal values,
respectively. As suggested by Equation (5.53), the percentagewise variabilities are:
5.3% in E 1 , 3.5% in E 2 , 12% in E 3 , and 10.3% in E 4 . These variabilities in elastic
moduli lead to up to 11% (for Case 1) and 9% (for Case 2) of variation in buckling load
of the plate. However, when one of the dimensions of the plate increases, such an effect
tends to stabilize; in the cases considered, the uncertainty in buckling load is about
8.5% of its nominal value. It is interesting to note that the uncertainty of buckling load
induced by uncertainty in elastic moduli becomes more “stabilized” with the increase of
layers. Besides, the variability of the buckling load is also dependent on the lamination
configuration of the plate; Figure 5.3 indicates that the scatter in the buckling load Ncl
is more noticeable when θ is in the vicinity of 45◦ .
Now we consider the buckling of the symmetric angle-ply cylindrical shell subjected
to axial compression. Equation (5. 10) is used in conjunction with Equation (5.35). Here,
an integer search is performed for the determination of the axial buckling wave number
m. The shells investigated have a 6.0-in. radius and are composed of 0.012-in.-thick
layers. The following two cases are considered:
Case 1: The 3-layer laminated shell, with ply angle being [θ, −θ, θ ], θ ranging
from 0◦ to 90◦ .
Case 2: The 5-layer laminated shell, with ply angle being [θ, −θ, θ, −θ, θ ], θ
ranging from 0◦ to 90◦ .
Figure 5.4 Uncertainty in buckling load for a set of 3-ply laminated shells (Case 1).
Figures 5.4, 5.5, 5.6, and 5.7 portray the variability of buckling load due to the
uncertainty in elastic moduli for Cases 1 and 2. Again, the effect of uncertainty in elastic
moduli on the buckling load varies with the laminate configuration and the number of
layers that make up the laminated shell. The maximum variabilities in buckling load of
the shell constitute 8% for Case 1 and 9% for Case 2 (Figures 5.5 and 5.7).
Figure 5.5 Effect of uncertainty in elastic moduli on the buckling load of a set of 3-ply lami-
nated shells (Case 1).
Figure 5.6 Uncertainty in buckling load of a set of 5-ply laminated shells (Case 2).
5.1.5 Numerical Analysis by Non-Linear Programming

The preceding problem can also be treated as a non-linear programming problem with
bounds that are mathematically stated as follows:
Find min F(E 1 , E 2 , E 3 , E 4 ) or max F(E 1 , E 2 , E 3 , E 4 )
(5.56)
Subject to E iL ≤ Ei ≤ E iU , for i = 1, 2, 3, 4
Figure 5.7 Effect of uncertainty in elastic moduli on the buckling load of a set of 5-ply
laminated shells (Case 2).
For this problem, it is not advisable to apply the gradient methods or the Davidon-
Fletcher-Powell method (Himmelblau, 1972), which are used most often in non-linear
optimization, because the directional derivative of the objective function F cannot
be easily calculated analytically. So, a direct search should be implemented. Here we
choose the complex method (Beveridge and Schechter, 1970), which is based exclusively
on function comparison; no derivatives are used. In the search for a minimum F, the
complex method starts with 2n (n = 4 in our case) points E (1) , E (2) , . . . , E (2n) , where
(i) (i) (i) (i)
E (i) = {E 1 , E 2 , E 3 , E 4 }. At each search cycle, a new point is generated by a certain
rule in terms of the previous 2n points, and the worst point E ( j) , which has the largest
value of F among these 2n points, is rejected and replaced by the new point. Whenever
the new point generated is beyond the bound, it will be set to the bound. Progress will
continue with repeated rejection and regeneration until some criteria are met. For a
complete description of this method, consult several monographs (Himmelblau, 1972;
Beveridge and Schechter, 1970). Nowadays, performing non-linear programming could
also be realized through use of such computational tools as gradient projection, feasible
direction, and penalty-function methods (Kirsch, 1981). We must digress that, since the
expression of the buckling load is available explicitly, one can choose to optimize a two-
term or three-term Taylor expression of the buckling load about the nominal values of
elastic moduli subject to non-linear quadratic constraint function (Li et al., 1996). The
constraint function, given in Equation (5.17), represents the equation of the ellipsoid of
minimum volume that encloses the rectangular parallelepiped representing the original
inequality constraints. Van den Nieuwendijk (1997) confirmed the present approach
numerically and extended it.
It appears to be remarkable that the present treatment exhibits a basic philosophi-
cal difference from the classical optimization studies. In classical optimum design of
structures, one looks to maximize the buckling loads; here we look for the least favor-
able scenarios (i.e., we determine, for design purposes, minimum buckling loads). This
procedure has been dubbed by Elishakoff (1991) as “anti-optimization.” A hybrid study
that uses both optimization and anti-optimization in the structural buckling context was
conducted by Adali et al. (1994). Combined optimization and anti-optimization is not
unlike the folk wisdom, that advises “Make the best out of the worst.”
5.2 Critical Contrasting of Probabilistic and Convex Analyses
5.2.1 Is There a Contradiction Between Two Methodologies?

In Chapters 3 and 4 we extensively used probabilistic analysis to predict the behavior
of the structures due to uncertain initial imperfections modeled as random fields. In
addition, in Section 4.6 we used a stochastic version of the finite element method to
deal with randomly varying elastic moduli as a function of axial coordinate. On the
other hand, in Section 5.1, an alternative, set-theoretical analysis of elastic moduli was
presented; in particular, we dealt with the often-encountered situation when the data
were unavailable to justify the traditional probabilistic analysis.
A natural question arises: Is there any contradiction between these two methodolo-
gies? We feel that the answer to this question is negative. If sufficient data are available,
5.2 CRITICAL CONTRASTING OF PROBABILISTIC AND CONVEX ANALYSES 239
one can utilize probabilistic method; however, if the data are not available, we refrain
from recommending, as it is often done, to “invent data out of nowhere,” and to perform
yet another academic study on the effect of random material properties on the buckling
of a structure. In these circumstances, one should utilize non-probabilistic methods
of dealing with uncertainty, such as convex modeling. For the detailed survey of the
developments in this new alternative to probabilistic modeling, consult the essays of
Ben-Haim (1994) and Elishakoff (1995).
The probabilistic analyst in actuality claims, “Give me the joint probability densities
of random variables involved, and I will calculate the reliability of the structure through
sophisticated numerical methods.” This reminds us of the well-known statement by
Archimedes: “Give me a firm spot on which to stand, and I will move the earth.”
It is clear that this analogy is not perfect. However, the resemblance of these two
statements is clear. In this connection, the following quotation of Blekhman, Mishkis,
and Panovko (1983) appears to be instructive:
Significantly, the weakness of numerous works on stochastic models sometimes ruling

out any application lies in the choice of statistically hypotheses, especially of assump-
tions regarding the probabilistic features of the given quantities and functions. These
features are often regarded as fully known (like an assumption of a normal distribution
with known parameters), or as capable of determination. In real situations, it mostly
turns out that the needed information is lacking.
Moreover, as was shown by Ben-Haim and Elishakoff (1990) in the series of prob-
lems, Elishakoff and Hasofer (1992), Neal, Mathews, and Vangel (1992) and Elishakoff
(1995), even small errors in probabilistic data may lead to large errors in estimating
probabilities of failure. It should be borne in mind that (Wentzel, 1988)
it is infrequent that the theory is viewed to be a sort of magic wand yielding information
from nothing, i.e., from total ignorance. Those who think so are under a misapprehen-
sion since probability theory is used but to transform data on observed phenomena to
infer the behavior of those which cannot be observed.
In some circumstances, as Wentzel (1988) notes, utilization of probabilistic method

is questionable from the very start.
A little comment is here in order on the difference between uncertainty and randomness.
As will be recalled, probability theory refers the term “random event” to the events
which recur and, more important, have a property of statistical stability. The latter
implies that similar trials with random outcome tend to assume a stable distribution
upon manifold repetition. The frequencies of the events tend then to the respective
probabilities, and the sample means to the mathematical expectations. . . .
There is however, nonstochastic uncertainty. . . . The factors . . . are as ever un-
known beforehand, but additionally there is no point in speaking of, or trying to evaluate
their distributions or other probability characteristics.
As Freudenthal (1956) digresses, “Ignorance of the cause of variation does not

make such variation random.” One of the present writers (Elishakoff, 1995) notes that
“Uncertainty and randomness are not reciprocal.”
Under these circumstances it makes sense not to abandon the alternatives to proba-
bilistic methods, namely the theory of fuzzy sets (Zadeh, 1965) and anti-optimization,
also known in the literature as the unknown-but-bounded uncertainty approach, guar-
anteed performance approach, or convex modeling of uncertainty, whose simplest
form – interval analysis – was known for decades. First hints of the latter method
appeared a long time ago (Bulgakov, 1940, 1946). The idea that bounding techniques,
rather than the probabilistic methods, may be preferable in some circumstances has
reappeared in works by Drenick (1968, 1970), Shinozuka (1970), Schweppe (1968,
1973), and Chernousko (1981), although in different engineering contexts. However,
it was not until the monographs of Schweppe (1973), dealing with control theory,
and Ben-Haim (1985), dealing with applications in nuclear engineering, that the non-
probabilistic methods started to develop intensively. The convex modeling in the applied
mechanics context was developed in monographs of Ben-Haim and Elishakoff (1990),
Chernousko (1994), and Elishakoff et al. (1994). For extensive discussion on the con-
vex models of uncertainty, consult with the review article of Ben-Haim (1994) and the
essay of Elishakoff (1995).
Several additional quotations appear to be in order. Freudenthal (1961), one of the
main architects of the modern probabilistic theory of structures, pinpoints the difficulties
associated with probabilistic methods:
when dealing with probabilities a clear distinction should be made between conditions
arising in design of inexpensive mass products on which the probability figures are
derived by statistical interpretation of actual observations or measurements (since a
sufficiently large number of observations are actually obtainable), and conditions arising
in design of structures of complex systems. In the latter, probability figures are used
simply as a scale or measure of reliability that permits the comparison of alternative
designs. The figures can never be checked by observations or measurements since they
are obtained by extrapolations so far beyond any possible range of observation that such
extrapolation can no longer be based on statistical arguments but could only be justified
by relevant physical reasoning. Under these conditions the absolute probability figures
have no real significance.
Bolotin’s (1961) reasoning resonates well with the latter one:
small probabilities of failure, if they are correctly found, still retain their importance
as some objective characteristics of possibility of random events taking place. They
become sensible when comparing them with each other, allowing contrast of the risk
of failure of different structures or of the same structures in different working condi-
tions. . . .
In the context of the probabilistic modeling, once the probability of failure of an en-
semble of structures is evaluated, it should be compared with acceptable probability of
failure. Grandori (1991) addresses a problem of assigning the acceptable probability
of failure:
the probabilistic approach to structural safety is today a well established paradigm. All
overwhelming part of research effort, in fact, has been and still is devoted to estimating
failure probabilities. By contrast, only sporadic research deals with the problem of
choosing an acceptable risk of failure. . . . It is true that the adaptation of a probabilistic

approach is in any case progress. . . . However, the concept of structural safety will not
leave the “realm of metaphysics” unless we devise a method for justifying the choice
of risk acceptability levels.
In this context, a natural question arises: Is there a possibility to directly contrast the
probabilistic and the non-probabilistic methodologies? This question was posed to us
by Crandall (personal communication, 1990). The reply to this question is affirmative. It
was given in the recent study by Elishakoff, Cai, and Starnes (1994b). In order to conduct
the probabilistic analysis, the authors of the preceding study first have “pretended”
that the experimental information was sufficient to justify the traditional probabilistic
analysis. Namely, the Fourier coefficients of the initial imperfections were treated as
having a truncated normal distribution. Then the assumption of sufficiency of available
information was abandoned, and the preceding Fourier coefficients were assumed to be
uncertain but non-random: They were assumed to belong to a multi-dimensional box.
The results of this comparison follow.
5.2.2 Deterministic Analysis of a Model Structure for a Specified

Initial Imperfection
Consider buckling of an initial imperfection-sensitive structure-column on a non-linear
elastic foundation. The system is illustrated in Figure 5.8, and the governing equation is
d 4w d 2w d 2 w̄
EI + P + K 1 w − K 3 w 3
= −P (5.57)
dx4 dx2 dx2
Figure 5.8 Comparison of results for the axial buckling load from the anti-optimization method
and numerical non-linear programming.
subject to the boundary conditions

d 2w
w= = 0, at x =0 and x =l (5.58)
dx2
where w̄ is the initial deflection, w is the additional deflection due to the axial load
P, and EI is the bending rigidity; both K 1 and K 3 are positive constants, representing,
respectively, the linear and nonlinear spring constants of the foundation.
If the initial deflection w̄ is given as a deterministic function of x, then the Boobnov-
Galerkin method can be applied to solve Equation (5.57) approximately. By expressing
both the initial deflection w̄ and additional deflection w as Fourier series, Equation (5.57)
leads to a set of algebraic equations for the unknown Fourier coefficients of the additional
deflection. If the foundation is linear, every term in the Fourier series represents a
“separate” mode, different modes are uncoupled, and the Fourier coefficients (i.e., the
magnitudes of the modes in the additional deflection) can be evaluated from the set of
equations. However, in the case of a non-linear foundation, different modes are coupled,
and the set of algebraic equations is an infinite hierarchy. To obtain an approximate
solution for the additional deflection, this infinite set has to be truncated, and only
a finite number of the “most important” modes should be taken into consideration.
Based on the Galerkin method, two different numerical schemes (Fraser, 1965; Fraser
and Budiansky, 1969; Elishakoff, 1979a) were developed to calculate the buckling
load for the non-linear column with a given initial deflection. Fraser (1965) carried
out Monte Carlo simulations – a general numerical method to deal with stochastic
problems – to calculate the reliability of the column. In Fraser’s simulation procedure, a
single-mode Boobnov-Galerkin approximation was used to evaluate the buckling load
in each sample. To significantly improve the accuracy, a multi-mode Boobnov-Galerkin
approximation was used by Elishakoff (1979a), in conjunction with a development of
a special procedure to simulate the multi-mode random initial imperfections.
In the present section, which closely follows the study by Elishakoff, Cai, and
Starnes (1994a), the buckling problem of the column, represented by Equation (5.57)
and boundary conditions (5.58) is investigated. The concept of “modal buckling load”
is defined for the column on a linear foundation. For the case of a non-linear founda-
tion, a criterion based on the concept of modal buckling load is proposed to determine
which modes are the most significant and should be retained in calculating the buckling
load by the Boobnov-Galerkin method. For the reliability study, a random field model –
Fourier series with truncated normally distributed coefficients – is suggested to describe
a random initial deflection. Monte Carlo simulations are performed to obtain the prob-
ability density of the buckling load and the reliability of the column. In the frequently
encountered case where the sufficient knowledge about the initial imperfection is ab-
sent for substantiation of the stochastic analysis, an alternative non-stochastic approach
(Ben-Haim and Elishakoff, 1990; Elishakoff and Ben-Haim, 1990; Elishakoff, 1991)
is applied to model the initial geometric imperfections and to obtain the minimum
buckling load. The objective is to critically contrast the results from the stochastic and
non-stochastic approaches.
First, consider a perfect column on a linear elastic foundation, namely, K 3 = 0 and
w̄ = 0 in Equation (5.57). Substituting the non-trivial solution w = sin (mπ x/1) into
Equation (5.57), we obtain

EI 2 2 K 1l 4
P(m) = 2 π m + (5.59)
l EIπ 2 m 2
When the axial load reaches the minimum value of P(m), m = 1, 2, . . . , buckling
occurs. Namely, the buckling load P∗ is determined by
P∗ = P(m ∗ ) = min[P(m), m = 1, 2, . . .] (5.60)
where m ∗ is such an integer that P(m ∗ ) is the minimum of P(m). From (5.59) and (5.60),

[d], if P([d]) < P([d + 1])
m∗ = (5.61)
[d + 1], if P([d]) > P([d + 1])
where d = 1/π(K 1 /EI)1/4 , and [d] denotes the integer part of d. The buckling load is
then

EI 2 2 K 1l 4
P∗ = Pcl = 2 π m ∗ + (5.62)
l EIπ 2 m 2∗
Equations (5.56) and (5.62) indicate that the buckling occurs in the m ∗ th mode once the
external load reaches the buckling load Pcl , which depends on the system parameters.
Now, define the following non-dimensional quantities
w w̄ x P
u= , ū = , η= , α=
Pcl
(5.63)
K 1l 4
K3 l
2 4
k1
k1 = , k3 = , γ = π 2m∗ + 2 2
EI EI π m∗
where is the radius of gyration of the cross section. Equation (5.57) and boundary
conditions (5.58) can be transformed to their non-dimensional forms, respectively,
d 4u d 2u d 2 ū
+ αγ + k 1 u − k 3 u 3
= −αγ (5.64)
dη4 dη2 dη2
and
d 2u
u= = 0, at η=0 and η=1 (5.65)
dη2
In what follows, we only discuss the problem in non-dimensional form. The qualification
phrase “non-dimensional” will be omitted for convenience; for example, u and ū are
called the initial and additional deflections, α is the external axial load, and so on.
The initial deflection can be expanded as

∞
ū(η) = ξ̄m sin(mπ η) (5.66)
m=1
We seek the solution of Equation (5.64) also in the form of Fourier series

∞
u(η) = ξm sin(mπ η) (5.67)
m=1
Substitution of Equations (5.66) and (5.67) into (5.65) results in (Fraser, 1965; Fraser
and Budiansky, 1969)
s m 2∗
αm ξm − α(ξm + ξ̄m ) − Im = 0 (5.68)
8 m2
where m ∗ has been given in Equation (5.61),
2k3 π 2 m 2 + k1 /(π 2 m 2 )
s= , αm = (5.69)
k1 + ξ 4 m 4∗ π 2 m 2∗ + k1 /(π 2 m 2∗ )
and

∞
∞
∞
Im = ξ p ξq ξr
p=1 q=1 r =1

× δ p+q,r +m − δ| p−q|,r +m − δ p+q,|r −m| + δ| p−q|,|r −m| + δ p,q δr,m (5.70)
in which δ p,q is the Kronecker delta.

If the elastic foundation is linear, s = 0, and Equation (5.68) is reduced to
α
ξm = ξ̄m (5.71)
αm − α
which indicates that the mth mode in the initial deflection can only induce the mth
mode in the additional deflection. The buckling load is usually defined as one that
causes the additional deflection to become unbounded (Ziegler, 1969). If the initial
deflection contains the m ∗ th mode, namely ξ̄m ∗ = 0, it can be seen from Equation (5.71)
that the buckling occurs always in the m ∗ mode and the buckling load αcl = αm ∗ =
min [αm , m = 1, 2, . . .] = 1, which is the same as that of the column without initial
imperfection. However, if the initial deflection contains only the kth (k = m ∗ ) mode,
the additional deflection becomes unbounded not at unity load, but at αk . Thus, we call
αm the mth modal buckling load. It is seen from Equation (5.69) that
$
>1, if m = m ∗
αm (5.72)
=1, if m = m ∗
The m ∗ th mode is called the classical critical mode. In general, an initial deflection
contains all the modes, including the m ∗ th mode. For a non-linear column with initial
deflection and under an axial force, an explicit expression for the additional deflection
is not obtainable, and a numerical algorithm has to be used to calculate the additional
deflection. Depending on the non-linearity, the system may exhibit initial imperfection
sensitivity. Fraser and Budiansky (1969) defined the buckling load α ∗ as

dα
=0 (5.73)
d F α=α∗
where

1
1 du 2 du d ū
F = F(u, ū) = + dη (5.74)
0 2 dη dη dη
representing the end-shortening of the column. According to this definition, the buckling
load α∗ represents a maximum load the structure can sustain (i.e., the buckling load is
defined as a limit load). If the initial and additional deflections are of the forms (5.66) and
(5.67), respectively, the end-shortening can be obtained by substituting Equations (5.66)
and (5.67) into (5.74) as follows:
π2
∞
F= m 2 ξm (ξm + 2ξ̄m ) (5.75)
4 m=1
Equation (5.73) cannot be solved either analytically or numerically since Equations

(5.66)–(5.68) and (5.75) have an infinite number of terms. Approximate solutions
were obtained by truncating these equations and retaining the “most important modes”
(Fraser and Budiansky, 1969; Elishakoff, 1979a). It is obvious that for “weak” non-
linearity (i.e., k3 k1 ), the most significant contribution to the buckling load is ex-
pected to come from the m ∗ th mode, which is the critical mode for the corresponding
linear case. In this case, an explicit relation between the buckling load and the initial
deflection has been obtained (Fraser, 1965). However, if the non-linearity is not small
and/or a high accuracy for the approximate solution is required, more modes must
be taken into consideration. Thus, a criterion for determining which mode is more
important is desirable.
Let us first consider a specific case where k1 = 16π 4 . It can be easily determined that
m ∗ = 2 and that α1 = 2.125, α2 = 1, α3 = 1.347, α4 = 2.125, and α5 = 3.205. This
specific example will be investigated in numerical calculations throughout this study.
Figures 5.9(a)–(c) show the buckling load for the column with k3 = 0.1k1 computed
from Equations (5.70)–(5.72) and (5.79) by retaining the first four modes. In all these
figures, the magnitude of the second mode of the imperfection is taken as a variable
for the horizontal coordinate. The buckling load curves in Figure 5.9(a) for different
values of ξ1 and in Figure 5.9(c) for different values of ξ4 remain very close, whereas
those in Figure 5.9(b) for different values of ξ3 are, although not very close, still not
significantly separated. The closeness of these curves indicates that neither ξ1 , ξ3 , nor
ξ4 has a profound influence on the buckling load. However, all these curves have sharp
slopes, especially near the origin (ξ2 = 0), which indicates a drastic change of the
buckling load with an increasing ξ2 . Therefore, it can be concluded that the second
mode plays a dominant role, the third mode also has significant contributions to the
buckling load, but less important than the second mode, and the first and fourth modes
are even less important. It is noticed that these results “correlate” well with the fact
that 1 = α2 < α3 < α1 = α4 , namely, the closer the modal buckling load is to unity,
the more significantly the corresponding mode contributes to the accurate evaluation
of the buckling load. For the simplest approximation, only the critical mode needs
to be considered. However, if more accurate results are required, the magnitude of
the modal bucking load can be considered as a criterion to decide whether or not the
corresponding mode should be included in calculating the buckling load by the Galerkin
method.
To further confirm this, we return to the preceding specific case of k1 = 16π 2 , but
with the non-linearity k3 as a variable. The approximate buckling loads are calculated for
Figure 5.9 Buckling load α versus ξ̄2 : (a) different ξ̄1 ; (b) different ξ̄3 ;
(c) different ξ̄4 (after Elishakoff, Cai, and Starnes, 1994; Copyright c
two different initial deflections – (a) ξ1 = ξ2 = ξ3 = ξ4 = ξ5 = 0.1, (b) ξ1 = ξ3 = ξ4 =

ξ5 = 0.3, ξ2 = 0.1 – and are depicted in Figures 5.10(a) and (b). Different combinations
of modes are selected to carry out numerical computations. The second mode is included
in all cases because it is the critical mode. It can be observed from Figures 5.10(a) and (b)
that:
1. If the initial deflection is small and the non-linearity is weak, one mode approxi-
mation may give acceptable results.
2. Within two mode approximation, these two modes should be taken as the second
mode and the third mode. This can be clearly seen from Figure 5.10(b) and coincides
with the fact that α1 and α2 are less than the other modal buckling loads.
3. If the first, second, and third modes are included in the computations, the accuracy
of the results is indeed improved, but not significantly because the fourth mode is
almost as important as the first mode since α1 = α4 . Thus, neglect of the fourth
mode may lead to a considerable error.
4. Four-mode approximation gives very accurate results, and the contribution of the
fifth mode is negligible because α5 = 3.125 is “far away” form unity.
The preceding analysis confirms that the modal buckling load reflects the importance
of the corresponding mode to the buckling load and can serve as a criterion for selecting
the modes involved in calculating the buckling load. In order to improve accuracy for
the approximate results, the mode with a lower modal buckling load should be taken
into account successively.
It is noted from Equation (5.71) that if the critical mode is a high mode (i.e., m ∗
is large), the αm value for m near m ∗ is close to unity; thus, the mth mode is almost
as important as the critical mode. In this case, a large number of modes are needed to
obtain a highly accurate result.
5.2.3 Probabilistic Analysis

The probabilistic analysis in this section is based on Elishakoff et al. (1994b). In some
cases, we can obtain statistical properties of a initial deflection by measurements or
experience. In these cases, the initial deflection should be treated as a random field.
Every sample in this random field can still be described by a Fourier series, namely,
Equation (5.66). However, different samples have different coefficients in their respec-
tive Fourier series. Thus, a random initial deflection can be described by the Fourier
expansion (5.66) with random coefficients ξ̄m = (m = 1, 2, . . .). To proceed with the
reliability analysis, a knowledge of the joint probability distribution for random vari-
ables ξ̄m is necessary. The normal distribution is a popular choice due to its simplicity in
analysis, but it may not be appropriate for practical cases because the initial deflection
can be visualized as limited in a certain range. (Indeed, quality control will discard
“very imperfect” columns.) To improve this model, we propose a truncated normal
distribution for each random variable ξ̄m as follows:
 2
c exp − ξ̄m ,

|ξ̄m | ≤ Am
m
p(ξ̄m ) = bm2 (5.76)

0, |ξ̄m | > Am
Figure 5.10 Buckling load α computed by using different combination of modes versus non-
linearity k3 /k1 : (a) ξ̄1 = ξ̄2 = ξ̄3 = ξ̄4 = ξ̄5 = 0.1; (b) ξ̄1 = ξ̄3 = ξ̄4 = ξ̄5 = 0.3, ξ̄2 = 0.1 (after
Elishakoff, Cai, and Starnes, 1994; Copyright c Elsevier Science Ltd., reprinted with permis-
sion).
where p(ξ̄m ) is the probability density of ξ̄m , each Am is a maximum possible value
for the random variable ξm , bm are parameters, and the normalization constants cm are
derived from
1
cm =
(5.77)
2bm erf Abmm
in which the error function erf (·) is defined as

x
e−t dt
2
erf(x) = (5.78)
0
Also, we assume that the random coefficients ξ̄m are independent for different m. The
probability density functions of the random coefficients are shown in Figure 5.11 with
ξ̄m , Am , and bm replaced by ξ̄ , A, and b for simplicity. With a given A, the probability
density depends exclusively on b. A large b corresponds to a large deviation of ξ̄ .
When b2 A2 , ξ̄ is nearly uniformly distributed, as shown by the case of b = 1 in
Figure 5.11.
The auto-correlation function and the mean square value of the initial deflection
are, respectively,

∞

Rū (η1 , η2 ) = E ξ̄m2 sin(mπ η1 ) sin(mπ η2 ) (5.79)

m=1
Figure 5.11 Probability density function for a truncated normally distributed random vari-
able (after Elishakoff, Cai, and Starnes, 1994; Copyright
with permission).
and

∞

E(ū 2 (η)] = E ξ̄m2 sin2 (mπ η) (5.80)

m=1
Equations (5.79) and (5.80) show that the random field of the initial deflection is
inhomogeneous. From the probability density function (5.77) and the auto-correlation
function (5.79), it can be seen that the model is quite general and feasible for describing
a real initial deflection.
Another advantage of the random field model is its simplicity to be simulated. The
realization of ξ̄m , denoted by (#̄m )k , k = 1, 2, . . . , can be generated by

−1 Am
(#̄m )k = bm erf (2δk − 1)erf (5.81)
bm
where δk , k = 1, 2, . . . , are independent random numbers uniformly distributed in [0,1].
In fact, the probability distribution function for #̄m is
F#̄m (ξ̄m ) = Prob [#̄m ≤ ξ̄m ]
= Prob{bm erf −1 [(2δk − 1)erf(Am /bm )] ≤ ξ̄m }

1 erf(ξ̄m /bm )
= Prob δk ≤ +
2 2 erf(Am /bm )
1 erf(ξ̄m /bm )
= + (5.82)
2 2 erf(Am /bm )
where the last equality is due to the uniform distribution of δk in [0,1]. One differentiation
of F#̄m with respect to ξ̄m leads to the probability density of #̄m
 2 2
 exp −ξ̄m /bm ,

|ξ̄m | ≤ Am
p#̄m (ξ̄m ) = 2bm erf(Am /bm ) (5.83)


0, |ξ̄m | > Am
which is the same as Equation (5.76).
With given parameter Am and bm in the probability density functions p(ξ̄m ) for the
initial deflection, Monte Carlo simulations can be carried out to obtain the probability
density for the buckling load. For every sample, the Galerkin method can be applied
by retaining finite modes in both initial and additional deflections according to the
criterion proposed in the last section. Figure 5.12 shows the computed probability
densities of the buckling load for the example column with k1 = 16π 4 and k2 = 0.1k1 .
Four modes were retained in computations, and every mode was assumed to have the
same probability distribution, namely, A1 = A2 = A3 = A4 = A and b1 = b2 = b3 =
b4 = b. Three different cases of b = 0.1, 0.25, and 1 were considered, and 105 samples
were calculated for each case. Figure 5.12 shows that with the same bound Am = A for
the initial deflection’s Fourier coefficients, the distributions of the buckling load can be
significantly different, depending on the distributions of the initial deflections. For a
larger deviation of the initial deflection (the case b = 1), the probability that a smaller
buckling load occurs increases.
Figure 5.12 Probability density function of the buckling load (after Elishakoff, Cai, and
Starnes, 1994; Copyright c Elsevier Science Ltd., reprinted with permission).
The reliability function for the column with a random initial imperfection, subject
to a prespecified axial load α, is defined as (Bolotin, 1958)
R(α) = Prob[α∗ ≥ α] (5.84)
where α∗ is the random buckling load. Figure 5.13 depicts the reliability functions
for the preceding example column with A = 0.5 and four different values of b. As
expected, the reliability for a given load α significantly depends on the parameter b,
which indicates the deviation of the initial deflection from the nominal value. If we
design a column based on the stochastic approach, the value of the load corresponding
to R equal to a codified required reliability r is the maximum value for the admissible
axial load. The latter load should be used as a design load for an ensemble of columns
on non-linear elastic foundation with stochastic imperfection.
5.2.4 Non-Stochastic, Anti-Optimization Analysis

The non-stochastic, anti-optimization analysis in this section is based on Elishakoff
et al. (1994b). In some cases, it is even difficult to estimate the probability distribution of
the initial imperfection. In these circumstances, the stochastic approach is not applicable
and a non-probabilistic model for the initial imperfection must be adopted. If we still
expand the initial deflection as a Fourier series (5.66), then a simple non-probabilistic
model for the initial deflection is its Fourier expansion coefficients vary in a hyper-
cuboid set:
Z (A): |ξ̄m | ≤ Am (Am ≥ 0, m = 1, 2, . . .) (5.85)
Figure 5.13 Reliability function versus actual axial load (after Elishakoff, Cai, and Starnes,
where A = {A1 , A2 , . . .} is a constant vector. Our objective now is to find the minimum,
least favorable limit load for all possible initial deflection ξ̄ = {ξ̄1 , ξ̄2 , . . .} belonging to
the set Z (A). If we design a column based on the non-stochastic approach, the minimum
buckling load is the maximum value for the admissible axial load. Thus, we arrive at
the alternative way of determining the admissible axial load, which can be applied to
an ensemble of columns with bounded Fourier coefficients.
As discussed before, the buckling load is a function of ξ̄ and can be expressed
formally as
α∗ = ψ(ξ̄) (5.86)
where the function form ψ in (5.86) is implicit and non-linear; in fact, it represents a
procedure of numerical evaluation of the buckling load as described previously. Then,
the problem to find the minimum buckling load becomes an extreme value problem
min (α∗ ) = min ψ(ξ̄) (5.87)

ξ̄ ∈Z (A) ξ̄ ∈Z (A)
According to the method of Lagrange multipliers, we construct an auxiliary function

N

G(ξ̄, A, x) = ψ ξ̄) + λm ξ̄ 2m − A2m + xm2 (5.88)
m=1
where N is the number of the most significant modes involved in computations, x =

(x1 , x2 , . . . , x N ) is an auxiliary vector variable, and λm (m = 1, 2, . . . , N ) are Lagrange
multipliers. By letting ∂G/∂ ξ̄m = 0 and ∂G/∂ xm = 0, we obtain

∂ψ
+ 2λm ξ̄m = 0, m = 1, 2, 3, . . . , N (5.89)
∂ξm
λm xm = 0, m = 1, 2, 3, . . . , N (5.90)
which, combined with
ξ̄m2 − A2m + xm2 = 0, m = 1, 2, . . . , N (5.91)
constitute a set of non-linear algebraic equations for unknown ξ̄m , xm , and λm . If deriva-
tives ∂ψ/∂ ξ̄m exist in the entire set Z (A), the extreme values of the function ψ(ξ̄) are
reached at some of the solution points of the set of equations. For example, if N = 4,
the solutions for the set of Equations (5.89)–(5.91) are
(i) ξ̄1 = ±A1 , ξ̄2 = ±A2 , ξ̄3 = ±A3 , ξ̄4 = ±A4 ;
(ii) ξ̄i = ±Ai , ξ̄ j = ±A j , ξ̄k = ±Ak , ∂ψ/∂ ξ̄l = 0;
(iii) ξ̄i = ±Ai , ξ̄ j = ±A j , ∂ψ/∂ ξ̄k = 0, ∂ψ/∂ ξ̄l = 0;
(iv) ξ̄i = ±Ai , ∂ψ/∂ ξ̄ j = 0, ∂ψ/∂ ξ̄k = 0, ∂ψ/∂ ξ̄l = 0;
(v) ∂ψ/∂ ξ̄i = 0, ∂ψ/∂ ξ̄ j = 0, ∂ψ/∂ ξ̄k = 0, ∂ψ/∂ ξ̄l = 0;
(5.92)
where i, j, k, l = 1, 2, 3, 4 and i = j = k = l. However, ∂ψ/∂ ξ̄m may not exist at some
points; for example, ∂ψ/∂ ξ̄2 does not exist at ξ̄2 = 0 in the example column, as known
from the single-mode approximation (Elishakoff, 1979) and the numerical results shown
in Figures 5.9(a)–(c). Then, besides the solution points of Equations (5.89)–(5.91), the
points at which any of the derivatives ∂ψ/∂ ξ̄m does not exist have to be included as
candidate points to seek the extreme values of the function ψ(ξ̄).
Let us consider the example column: Because of the symmetry of the problem,
function ψ is an even function of ξ̄m . Thus, at ξ̄m = 0, either ∂ψ/∂ ξ̄m does not exist
or ∂ψ/∂ ξ̄m = 0. Therefore, if four modes are retained in computations, the following
points:
(i) ξ̄1 = ±A1 , ξ̄2 = ±A2 , ξ̄3 = ±A3 , ξ̄4 = ±A4 ;
(ii) ξ̄i = ±Ai , ξ̄ j = ±A j , ξ̄k = ±Ak , ξ̄l = 0;
(iii) ξ̄i = ±Ai , ξ̄ j = ±A j , ξ̄k = 0, ξ̄l = 0; (5.93)
(iv) ξ̄i = ±Ai , ξ̄ j = 0, ξ̄k = 0, ξ̄l = 0;
(v) ξ̄i = 0, ξ̄ j = 0, ξ̄k = 0, ξ̄l = 0;
contain all the solution points for the set of Equations (5.89)–(5.91), as well as the points
at which any of the derivatives ∂ψ/∂ ξ̄m do not exist. Therefore, the points in (5.93) are
the candidate points at which the ψ function may reach extreme value. Among these,
the point (v), (i.e., the origin) corresponds to the case where the initial imperfection is
absent. At this point, the function ψ obtains its maximum value for an imperfection sen-
sitive structure. The minimum buckling load can be found by comparing the values of
ψ at those points listed in (i) to (iv). Note that points in (i)–(iv) correspond, respectively,
to vertices, middle points of side, center of planes, and centers of three-dimensional
projected cuboids for the four-dimensional cuboid.
Figures 5.14(a) and (b) show the minimum buckling load for the example column
with the initial deflection bound A j = constant = A as a variable. Also, the admissible
loads corresponding to different values of reliability are calculated from the stochastic
approach and depicted in the same figures. For the case of b = 1 in Figure 5.14(b),
namely, large deviation of the initial deflection, the minimum buckling load and the
admissible load corresponding to different required reliability levels r = 0.9, 0.99,
and 0.999 do not exhibit much difference regardless of the magnitude of the boundary
A. Therefore, design can be made based on the non-stochastic approach because it is
much simpler than the stochastic one. The same situation can also be found for the
case of b = 0.1 and small A (A < 0.2 in the present case), shown in Figure 5.14(a).
However, if the deviation of the initial deflection is small and the boundary for the initial
deflection is large [b = 0.1 and A > 0.2 in the present case, shown in Figure 5.14(a)],
the admissible value for the axial load obtained from the stochastic approach may be
well above the minimum buckling load. This implies that the use of the non-probabilistic
method when the sufficient probabilistic information is available may not be advisable;
the latter technique may lead to a conservative design.
One of the non-stochastic models of uncertainty, namely, the Fourier coefficients
varying within a hyper-cuboid set, is adopted for the initial deflection in this study, and
the minimum buckling load is computed for this model. The comparison of the results
with those obtained from the stochastic approach indicates that the design based on the
non-stochastic approach is acceptable for a initial deflection with a large deviation but
may be conservative for that with a small deviation.
It is remarkable that in some circumstances both approaches, although being of car-
dinally different nature, may yield close values for the design axial loads. If probabilistic
information is unavailable, one should not propose a probabilistic model, based on an
arbitrary assumption on the distribution of the Fourier coefficients. Rather, in such cir-
cumstances one should use the non-stochastic approach to uncertainty. Only when the
full probabilistic information is available and the initial deflection’s Fourier coefficients
have a relatively small deviation, will use of the non-stochastic approach be unadvis-
able; at this time, purely stochastic analysis should be conducted. However, even when
probabilistic information is available to substantiate the probabilistic analysis, if the
distribution of the initial imperfections is “close” to uniform, one may prefer a simpler
non-stochastic, convex analysis, because it yields admissible axial loads comparable
to the results of the stochastic approach. Naturally, when the probabilistic information
is unavailable, the probabilistic methods cannot be utilized. Use of the probabilistic
modeling in such circumstances would be equivalent to treating it as a “magic wand,”
producing information, as Wentzel (1988) notes, from a void. In the case when scarce
information is available on initial imperfections, application of non-stochastic methods
appears to be a natural alternative to the probabilistic methods.
Thus we conclude that there are classes of problems where solely probabilistic or
solely non-probabilistic convex modeling can be conducted to describe the uncertain
quantities involved. It is also remarkable that there is a region of parameter varia-
tions, where either probabilistic and convex modeling can be applied. In this case the
non-probabilistic convex modeling appears to be superior to that of the probabilistic
approach because the former is numerically and conceptually easier than the latter.
Figure 5.14 Comparison of admissible axial loads computed from stochastic and non-
stochastic, anti-optimization approaches: (a) b = 0.1 and (b) b = 1 (after Elishakoff, Cai,
and Starnes, 1994; Copyright c Elsevier Science Ltd., reprinted with permission).
It is instructive to conclude by quoting Bolotin (1969) from the first edition

(Russian) of his book devoted to applications of statistical methods in mechanics:
There is a wide range of problems of structural mechanics in which the use of statistical
methods is the most adequate means of investigation – yet there are problems where
statistical methods may only play the role of auxiliary methods of analysis. Here the
statistical and deterministic methods could successfully coexist, complementing each
other . . . the overestimation of the role of statistical methods can only be harmful.
To sum up, Chapters 3 and 4 illustrate the range of cases where the use of proba-
bilistic methods is quite successful. In Chapter 5, we propose an alternative technique –
convex modeling of uncertainty – to deal with problems where the probabilistic method
cannot be justified. Remarkably in special circumstances, as it was demonstrated by
Elishakoff and Colombi (1993), the combination of these two approaches may be
needed.
Most importantly, the anti-optimization procedure (i.e., looking for the least buck-
ling load under uncertainty) can be combined with the optimization ideas. Namely, Adali
et al. (1994) studied the effect of uncertainty on optimal design of the laminate. The lam-
inate is optimized with respect to the ply angle θ with the geometric parameters fixed.
Let Ncl designate the buckling load for deterministic characteristics; let N ∗ = min Ncl
denote the minimum critical buckling load of the structure subject to uncertainty; the
minimization is conducted in the region of variation of uncertain moduli. The optimal
design problem is stated as follows: Determine Nopt = max N ∗ = max min Ncl and
the optimal ply angle θopt so as to maximize N ∗ ; the maximization is conducted in the
region of variation of the geometric parameters, characterizing area, volume, weight,
and so on. The sequential or nested max-min optimization procedure gives the values
of the optimal ply angle θopt and corresponding buckling load Nmax under the worst
case of material properties (see also Adali, Richter, and Verijenko, 1997).
In their recent article, Arbocz and Singer (2000) quote from the correspondence
with Professor Bernard Budiansky who posed the following question: “One other related
thought, that is only vaguely in my mind, is this: Is it possible that a more prominent role
should be given to “worst-case” imperfections?” It appears that the anti-optimization
procedure, combined with optimization around the “worst” behavior, provides at least
a partial answer to the above inquiry.
CHAPTER SIX
Application of the Godunov-Conte Shooting

Method to Buckling Analysis
The purpose of computing is insight, not numbers. . . . Usually the first question
to ask is “What are we going to do with the answers?”
R. W. Hamming
There is safety in numbers.

Euripides
I have no satisfaction in formulas unless I feel their numerical magnitude.

Lord Kelvin
This chapter does not give a general overview of numerical methods utilized in the buckling
of linear and non-linear structure; rather, it addresses a specific numerical method, namely
the Godunov-Conte method, which is often used in the buckling analysis. The Godunov-
Conte, which avoids the loss of accuracy resulting from the numerical treatment often
associated with stability and vibration analysis of elastic bodies, consists of parallel inte-
gration of the set of k homogeneous equations under the Kronecker-delta initial conditions,
which are orthogonal (k being the number of “missing” conditions); after each step, subject
to Conte’s test, the set of solutions is reorthogonalized by the Gram-Schmidt procedure
and integration is continued. The procedure prevents flattening of the base solutions, which
otherwise become numerically dependent.

Consider the eigenvalue problem defined by the set of n = 2k linear ordinary differential
equations
ẏ = A(t, λ)y (6.1)
where y(t, λ) is a (n × 1)-vector with components y1 (t, λ), y2 (t, λ), . . . , yn (t, λ); A(t, λ)
is an n × n matrix whose i, j-element is ai j (t, λ), λ being the sought eigenvalue; the
initial conditions are
yi (t0 ) = 0 i = 1, 2, . . . , k (6.2)
257
258 APPLICATION OF THE GODUNOV-CONTE SHOOTING METHOD TO BUCKLING ANALYSIS
The terminal conditions
yk+m (t f ) = 0 (m = 1, 2, . . . , k) (6.3)
This problem can be solved by the following simple procedure, known in literature as
the Goodman-Lance method (1956), or the method of complementary functions. Let
u ( j) (t, λ) ( j = 1, 2, . . . , n; t0 ≤ t ≤ t f ) be the set of n linearly independent solutions
under the initial conditions
( j)
u i (tδ , λ) = δi j (i, j = 1, 2, . . . , n) (6.4)
where δi j is Kronecker’s delta. The general solution of (6.1) is then found in a linear
combination

n
( j)
yi (t, λ) = b j u i (t, λ) (6.5)
j=1
where the b j constants are to be determined by the boundary conditions. By virtue of

(5.4)
yi (t0 , λ) = bi = 0, i = 1, 2, . . . , k (6.6)
yk+i (t0 , λ) = bk+i , i = 1, 2, . . . , k (6.7)
In other words, the constants bk+i , i = 1, 2, . . . , k, are equivalent to “missing” initial

conditions. Using (5.3)

k
(i)
yk+i (t f , λ) = bk+α u k+i (t f , λ) = 0 (i = 1, 2, . . . , k) (6.8)
α=1
The homogeneous system (6.8) has a solution only if the determinant of the coefficient
matrix [ f (λ)] vanishes
det[ f (λ)] = 0 (6.9)
where
 (k+1) (k+2) (n)

u (k+1) (t f , λ) u k+1 (t f , λ) · · · u k+1 (t f , λ)
 
 (k+1) (k+2) (n) 
u (t f , λ) u k+2 (t f , λ) · · · u k+2 (t f , λ)
[ f (λ)] =  k+2  (6.10)
 .. 
 . 
u (k+1)
n (t f , λ) u (k+2)
n (t f , λ) ··· u (n)
n (t f , λ)
The procedure thus consists of determining a sequence of values λ until (6.9) is sat-
isfied to a specified accuracy. Any root-finder technique may be used to arrive at the
successive approximations of the eigenvalue λ∗ , for which the eigenfunctions are read-
ily obtainable. Let the rank of [ f (λ∗ )] be k − 1. Then the system for determining the
bs is (Grigolyuk et al., 1971)

k
bk+α u k+i (t f , λ∗ ) = −bk+r u k+r (t f , λ∗ ),
(i) (i)
i, α = r, bk+r = b0 (6.11)
α=1
where b0 is any constant, and the eigenfunctions are found by substituting bk+1 , bk+2 , . . . ,
kk+r −1 , bk+r , bk+r +1 , . . . , bn in (6.8); since the system is homogeneous, they are deter-
mined only up to a multiplicative constant, b0 . This procedure, although mathematically
exact, may often lead to very poor or even completely incorrect results when applied
in numerical form. In some cases, the vectors u (k+1) , u (k+2) , . . . , u (n) , which form the
columns of [ f (λ)], become numerically dependent as t increases, irrespective of the
integration procedure used. For example, Grigolyuk et al. (1971) found that, in sta-
bility problems of cylindrical shells, this situation occurs when the non-dimensional
parameter Z satisfies the inequality
Z = L(1 − ν 2 )0.25 (RH)−0.5 > 10 (6.12)
where L , R and H are, respectively, the length, radius, and thickness of the shell, and
ν is Poisson’s ratio of the shell material. The matrix [ f (λ)] turns out to be poorly
conditioned, the eigenvalues ν ∗ are inaccurate, and the expansion (6.8) leads to poor
numerical results.
A method avoiding loss of accuracy was proposed by Godunov (1961), whereby the
matrix [ f (t, λ)] of base solutions remains orthogonal throughout. This is achieved by
integrating the set of k homogeneous equations in parallel under the Kronecker-delta
initial conditions (which are orthogonal), with the set of solutions reorthogonalized
after each integration step by the Gram-Schmidt procedure. The method was success-
fully applied by Grigolyuk and Lipovtsev (1975) and Valishvili (1975) in stability
problems of axisymmetric shells, and by Novichkov and Indenbaum (1972) in non-
homogeneous two-point boundary value problems (specifically, axisymmetric defor-
mation of a circular cylinder made of low-modulus material and enclosed in a plastic
shell subjected to internal pressure and a thermal effect). The method was subsequently
modified and adapted for more practical use by Conte (1966), with the angle between
the relevant pairs of vectors u (k+1) (t, λ), u (k+2) (t, λ), . . . , u (n) (t, λ), for the points ti at
which the solution is computed – serving as criterion of the need for reorthogonaliza-
tion. If the least angle is less than a specified tolerance β, the solution vectors are to be
orthogonalized; otherwise, we proceed with the next step. Conte’s test reads:
(i,q−1)
u (t), u ( j,q−1) (t)
−1
min cos 2 22 2 < β, i = j (6.13)
i, j 2u (i,q−1) (t)22 2u ( j,q−1) (t)22
2 (i,q−1) 22 (i,q−1)
2u (t)2 = u (t), u (i,q−1) (t) (6.14)
2 ( j,q−1) 22 ( j,q−1)
2u (t)2 = u (t), u ( j,q−1) (t) (6.15)
the parentheses denoting an inner product. Here u (m,q) (t) is a (k × 1) vector solution
of Equations (6.1)–(6.4) – m denoting the index counter for the number of the set of
linear conditions m = 1, 2, . . . , k, and q that of the point at which the vectors were
last orthogonalized. For q = 0, u (m,q) (t0 ) ≡ u (m) (t0 ) coincides with Kronecker-delta
initial conditions in Equation (6.4); u (m,q) (t) hold for the interval t (q) ≤ t ≤ t (q+1) . In
principle, the tolerance β can lie anywhere between 0◦ and 90◦ ; at the lower bound
reorthogonalization is unnecessary altogether, whereas at the upper one it is in order
for every integration step. Usually, some numerical experience is needed for finding a
suitable value of β for a given problem.
6.2 Brief Outline of Godunov-Conte Method As Applied

to Eigenvalue Problems
The method consists in solving the equations
u̇ = A(t, λ)u (6.16)
k times, under k sets of initial conditions
(m)
u i (t0 ) = δi,m+k (m = 1, 2, . . . , k) (6.17)
(m)
where u i (t, λ) is the solution for the ith component of u with λ given for the mth
integration of the homogeneous equations. The general solution, which reads

k
( j)
yi (t, λ) = b j u i (t, λ), i = 1, 2, . . . , n (6.18)
j=1
automatically satisfies (6.2). We subdivide the interval (t0 , t f ) into γ sub-intervals ti ,

i = 0, 1, . . . , γ − 1, and use any standard integration method to obtain the base solu-
tions u (1) , u (2) , . . . , u (k) at the mesh points, orthonormalizing them wherever they fall
below Conte’s criterion. Next, we define U (q) (t, λ) – an (n × k) matrix whose elements
are the solutions u (m,q) (t, λ)(m = 1, 2, . . . , k), which were last orthonormalized at t (q)
and whose columns are the vectors u (1,q) (t, λ), u (2,q) (t, λ), . . . , u (k,q) (t, λ), q again the
index counter for the number of orthonormalizations, 0 ≤ q ≤ Q:
 (1,q) (2,q) (k,q) 
u 1 (t, λ) u 1 (t, λ) · · · u 1 (t, λ)
 (1,q) 
u (t, λ) u (2,q) (t, λ) · · · u (k,q) (t, λ)
U (t, λ) = 
(q)
..
2 2 2 
 (6.19)
. 
u (1,q)
n (t, λ) u n (t, λ) · · · u (k,q)
n (t, λ)
(m,0) (m)
and u i (0, λ) = u i (0, λ) = δi,m+k . At t (q) (which need not be evenly spaced), we
construct for the set of k linearly independent vectors u (m,q−1) (t (q) , λ), m = 1, 2, . . . , k
the set of orthonormal vectors u (m,q) (t (q) , λ), which reads in matrix form

u (q) t (q) , λ = u (q−1) t (q) P (q) (6.20)
where P (q) is non-singular and upper triangular with elements Pij
 1

 , i= j

 w

 jj
Pij = 0, i> j (6.21)

 (i) ( j)

 z ,u

− , i< j
wii wjj
and
1/2
w11 = u (1) , u (1) , z (1) = u (1) /w11

η(2) = u (20) − u (2) , z (1) z (1)
6.3 APPLICATION TO BUCKLING OF POLAR ORTHOTROPIC ANNULAR PLATE 261
1/2
w22 = η(2) , η(2) , z (2) = η(2) /w22 (6.22)

j−1

η( j) = u ( j) = u ( j) , z (s) z (s)
s=1
( j)1/2

wjj = η , η ( j)
, z ( j) = η( j) /wjj , j = 1, 2, . . . , k
The solution at t (q) = t f is
y(t f , λ) = U (Q) (t f , λ)b(Q) (6.23)
(Q) (Q)
where b(Q) is a (k × 1) vector of constants with components b1 , b2 , . . . , bn(Q) . The
eigenvalue λ∗ derives from the singularity condition of the matrix U (Q) (t f , λ).
6.3 Application to Buckling of Polar Orthotropic Annular Plate

Several studies of buckling of homogeneous circular plates under radial compres-
sion are reported in the literature. The axisymmetric buckling problem was solved by
Woinowsky-Krieger (1958) with the aid of Bessel functions. Later, Pandalai and Patel
(1965) used power-series expansion in the same problem. The axisymmetric buckling
problem was solved by Mossakowsky (1960) and Mossakowsky and Borsuk (1960)
using Frobenius’s method with the aid of hypergeometric functions. Swarmidas and
Kunukkasseril (1973) examined the axisymmetric and first asymmetric buckling of a
circular orthotropic plate, formulating the displacement of the buckled plate with the
aid of Bessel and Lommel functions. Kaplyevatsky (1975) presented a solution for
specific values of plate orthotropy measure by a generalization of Frobenius’s method.
Approximate values of the axisymmetric buckling loads, at different edge conditions, in
a plate under uniform compression were obtained by Vijayakumar and Joga Rao (1971)
using the Rayleigh-Ritz method, and by Uthgennant and Brand (1970) using a finite-
difference technique and the Vianello-Stodola iterative procedure. The approach of
Vijayakumar and Joga Rao (1971) was extended by Ramaiah and Vijayakumar (1975)
for an annular plate under uniform internal pressure (in which case it always buckles
axisymmetrically).
The axisymmetric and general asymmetric buckling problem of annular orthotropic
plates, under arbitrarily specified internal and external pressures and with an arbitrary
orthotropy measure, is still unsolved analytically. In what follows, we report results
obtained by the Godunov-Conte method as described earlier.
The governing buckling equation in terms of the transverse deflection w and in-
plane radial and circumferential forces N2 and Nθ , respectively, reads:
4
∂ w 2 ∂ 3w 1 ∂ 2w 1 ∂w 2 ∂ 2w 1 ∂ 4w
Dr + + D θ − + + +
∂r 4 r ∂r 3 r 2 ∂r 2 r 3 ∂r r 4 ∂θ 2 r 4 ∂θ 4

1 24 w 1 ∂ 3w 1 ∂ 2w
+ 2Drθ 2 2 2 − 3 +
r 2r ∂θ r ∂r ∂θ 3 r 4 ∂θ 2

∂ 2w 1 ∂w 1 ∂ 2w
= Nr 2 + N θ + 2 2 (6.24)
∂r r ∂r r ∂θ
where
Er h 3 Eθ h3 G rθ h 3
Dr = , Dθ = , Dr = Drθ νθ + (6.25)
12(1 − νr νθ ) 12(1 − νr νθ ) 12
The plane stress-strain relations for the orthotropic plate material are
     
σ,r E r E rθ 0 r
     
σ,θ  =  E θr E θ 0  θ  (6.26)
τrθ 0 0 G rθ γrθ
where Er , Erθ (=E θr ), E θ , and G rθ are elastic stiffness moduli. In the isotropic case, Er =
E θ = E/(1 − ν 2 ), Erθ = ν E/(1 − ν 2 ), G rθ = G = E/2(1 + ν), Dr = Dθ = Eh 3 /
12(1 − ν 2 ) where E is Young’s modulus, ν is Poisson’s ratio, and D is the flexural
rigidity.
Using the equilibrium equation for in-plane forces in the axisymmetric radial in-
plane prebuckling state
d
(rNr ) − Nθ = 0 (6.27)
dr
The following differential equation is obtained:
4
∂ w 2 ∂ 3w 1 ∂ 2w 1 ∂w 2 2 1 24 w
Dr + + Dθ − 2 2 + 3 + 4 ∂ w∂θ + 4 4
2
∂r 4 r ∂r 3 r ∂r r ∂r r r ∂θ

1 2w 4
1 ∂ w3
1 ∂ w
2
1 ∂ ∂w
+ 2Drθ 2 2 2 − 3 + − rNr
r ∂r ∂θ r ∂r ∂θ 3 r 4 ∂θ 2 r ∂r ∂r
∂ 1 ∂ 2w
− (rNr ) 2 2 = 0 (6.28)
∂r r ∂θ
where Nr and Nθ are given by the plane-stress elasticity solution
ψ dψ
Nr = , Nθ = (6.29)
r dθ
−k
ψ = Ar + Br
k
(6.30)
ψ being the force resultant function, with

A = −N0ri−k−1 + Ni r0−k−1 R −1 (6.31)

B = N0rik−1 − Ni r0k−1 R −1 (6.32)
2k
−k−1 k−1 ri
R = ri r0 1− . (6.33)
r0
k 2 = Er /E θ = Dr /Dθ is an orthotropy measure, N0 = p0h , Ni = pih , pi and p0 denote

uniform in-plane pressures at the inner and outer edges, respectively.
Now, assuming a deflection function such as
w(r, θ ) = W (r ) cos(nθ) (6.34)

where n is the number of nodal diameters, Equation (6.27) reduces to an ordinary

differential equation with variable coefficients:
4
d W 2 d3W 1 d2W 1 dW 2n 2 n4
Dr + 3 + Dθ − 2 + 3 − 4 W + 4W
dr 4 r dr 3 r dr 2 r dr r r
2 2 3 2
2
n d W n dW n 1 dW n d
− 2Drθ 2 2
+ 3 + 2W − rNr + 2 (rNr )W = 0
r dr r dr r r dr r dr
(6.35)
For further convenience and generality, Equation (5.34) can be rewritten in non-
dimensional form, namely
ρ 4 W + 2ρ 3 W − µ + (ρ 2 W − ρW ) + κ W
+ λ[αρ 1+k (ρ 2 W + kρW − n 2 kW ) + Bρ 1−k (ρ 2 W − kρW + n 2 kW )] = 0
(6.36)
where
r ri Pi Dr θ P0 b2 h
ρ= , η+ , ξ= , k12 = , λ= ,
r0 r0 p0 Dr Dr

µ = k 2 + 2k12 n 2 , κ = n 2 k 2 n 2 − 2k 2 − 2k1 2
(6.37)
1 − ξη 1+k
η 1+k
(ξ − η k−1
)
α= , β=
1 − η2k 1 − η2k
ri , r0 , and h are inside and outside radii and thickness, respectively; k1 is a supplementary
orthotropy measure, ρ = 1 and ρ = η represent the outer and inner edges, respectively;
λ is the non-dimensional critical buckling load for a specified load ratio ξ .
Equation (5.35) is supplemented by two conditions at each boundary point, which
read in general form
Fγ (W ) = G γ (W ) = 0, γ = r1 , r0 (6.38)
Fγ and G γ being linear functionals.
Numerical results by the Godunov-Conte method were obtained, the lowest critical
loads were calculated with the radius ratio η as parameter. As in the study of isotropic
plates by Yamaki (1958), it turned out that the assumption of symmetrical loading often
leads to a stability overestimate.
The plates under consideration were clamped at the inner and outer edges, with
functionals Fγ and G γ , as follows:

∂(· · ·)
Fri (· · ·) = (· · ·)|r =r1 , G ri (· · ·) = (6.39)
∂r r =r1

∂(· · ·)
Fr0 (· · ·) = (· · ·)|r =r0 , G r0 (· · ·) = (6.40)
∂r r =r0
Curve 1 in Figure 6.1 presents the buckling behavior of the plate under uniform compres-
sion (ξ = 1) for different values of k, and curve 2 refers to axisymmetric buckling loads
(n = 0) found by the Rayleigh-Ritz method. The nodal diameters n, which correspond
Figure 6.1 Buckling load parameter λ = p0 b2 h/D r as a function of orthotropy measure k;

both edges clamped and p0 = pi . Curve 1, asymmetric buckling loads (Vijayakumar and Joga
Rao, 1971); curve 2, asymmetric buckling loads.
Table 6.1. Least values of λ and corresponding values of n and N (plate under
uniform compression)
η 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

√
k = 0.5 λmin 6.013 6.997 8.127 9.574 11.540 14.478 19.340 29.047
(β = 5◦ ) n 2 2 3 5 6 8 12 19
N 6 0 1 1 6 2 20
√
k=1 λmin 6.710 7.502 8.637 10.099 12.155 15.200 20.282 30.437
(β = 14◦ ) n 1 2 2 3 4 5 7 11
N 6 5 7 7 2 25 45
√
k = 1.5 λmin 7.622 8.251 9.231 10.650 12.617 15.655 20.733 30.995
(β = 5◦ ) n 1 2 2 2 3 4 5 8
N 3 2 1 0 0 1 2
Table 6.2. Least values of λ for the plate under internal pressure, β = 5◦
n = 0.1 n = 0.2 n = 0.3 n = 0.4 n = 0.5 n = 0.6 n = 0.7 n = 0.8

k = 0.5 5.1737 14.109 30.039 58.869 113.63 227.034 499.13 1354.6
k=1 7.6725 17.261 33.705 62.968 118.10 231.85 504.26 1360.0
7.678a 17.26a 33.705a 62.972a 118.10a 231.86a 504.32a 1360.2a
k=2 20.016 30.992 48.994 79,668 136.18 251.21 524.83 1381.7
a Results from Ramaiah and Vijayakumar (1974).
Table 6.3. Least values of λ and corresponding values of n for the plate under external
pressure, β = 5◦
η 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

k = 0.5 λmin 35.625 45.655 56.007 68.266 84.309 107.99 146.75 223.84
n 2 3 5 6 9 12 18 30
k=1 λmin 44.515 55.684 70.241 89.101 113.80 150.01 208.64 324.36
λmin 44a 55a 69.5a 88.5a 114.5a
n 2 2 3 4 6 8 12 20
k=2 λmin 75.310 82.847 99.975 124.50 161.47 219.98 314.57 505.18
n 1 2 2 3 4 6 8 14
a Results from Ramaiah and Vijayakumar (1975).
Table 6.4. Least values of λ and corresponding values of n and N (isotropic plate under
internal tension, β = 5◦ )
η 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

k = 0.5 λmin 7.7591 14.808 23.628 34.850 50.586 73.767 112.39 189.78
n 4 5 6 8 11 14 20 30
N 21 22 16 17 19 31 28 41
k=2 λmin 66.37 75.413 98.866 128.21 166.66 225.86 322.16 512.70
n 3 3 4 5 6 8 11 16
N 18 19 17 18 20 8 7 7
Table 6.5. Least values of λ and corresponding values of n and N for isotropic plate under internal tension, β = 5◦
η 1/8 1/7 1/6 1/5 1/4.5 1/2 1/3.5 1/3 0.4 0.5 0.6 0.7 0.8
λmin 19.433 22.371 24.935 27.980 30.643 34.688 39.704 46.826 58.779 82.775 118.09 176.33 291.36
λmin 19.5a 22.0a 25.7a 29.0a 32.0a 36.6a 42.5a 51.5a
n 3 3 4 4 4 4 5 5 6 8 8 14 22
N 18 20 21 17 14 15 15 14 19 20 20 22 25
a Results from Lipovtsev (1970).
to the least values of the non-dimensional buckling load λ, are given in parentheses.
Table 6.1 lists the least values of λ for different radius ratios η; N denotes the number
of reorthogonalizations required. The tolerance β was chosen as 14◦ . [Note that the
results for the isotropic plates are practically coincident with those of Yamaki (1958).]
Table 6.2 lists the least values of λ for a plate under uniform internal pressure (ξ →
∞), λ being defined in this case as
λ = pi ri2 h/Dr (6.41)
for different values of η and k (here, n = 0). Those for the isotropic case (k = 1) are
compared with those of Ramaiah and Vijayakumar (1975). The percentage is of the
order of 0.1%.
Table 6.3 present the results for a plate under uniform external pressure; those
for the isotropic case are compared with those of Ramaiah and Vijayakumar (1974).
Table 6.4 lists the non-dimensional buckling loads
λ = pi ri2 h/Dr (6.42)
for a plate under internal tension, as a function of k and η; Table 6.5 presents a com-
parison with Lipovtsev’s (1970) results for the isotropic case, obtained by the double-
sweep method (see Gelfand and Lokutsievski, 1964; Biderman, 1967; Grigolyuk and
Lipovtsev, 1975; Godunov, 1971).
To sum up, in this chapter, application of the Godunov-Conte method to eigenvalue
problems was demonstrated. Loss of accuracy was avoided by orthonormalizing the
base solutions at a point where the least angle between the relevant pair of base vectors
at each computation point is less than a specified tolerance. The method was applied to
the buckling of a polar orthotropic plate. Numerous examples were evaluated, showing
the effect of the orthotropy measure and radius ratio.
CHAPTER SEVEN
Application of Computerized Symbolic Algebra

in Buckling Analysis
For the public at large and even for most scientists, numerical calculation and
scientific calculation have become synonymous. . . . However, numerical calcu-
lation does not rule out algebraic calculation. . . . And, the power of computers
does not solve everything.
J. H. Daventport, Y. Siret, and E. Tournier
There are . . . problems in mechanics for which standard numerical procedures are
available but which could be solved more elegantly and accurately using analytical
methods if the formula work could be overcome.
J. Jensen and F. Niordson
Computer algebra systems can solve many problems more quickly than a human
being. In our experience it is not unusual for a computer system to solve a problem
which has been taxing a capable mathematician for several months, in a few
minutes. One wonders how many tractable problems remain unsolved or have
been forgotten about simply because they are making excessive demands on a
researcher’s time and sanity!
C. Wooff and D. Hodgkinson
In this chapter, we deal with the analytic, symbolic computation for buckling analysis.
Symbolic algebra can significantly reduce the tedium of analytic computation and simulta-
neously increases its reliability. It has great impact on scientific computation, as more and
more analytically minded researchers are using computers, which have been traditionally
associated with “number crunching.” Analytic work can now be extended as far as possi-
ble, and the numerical side of the analysis can be “delayed.” Calculations with arbitrary
arithmetic precision along with the automatic generation of computer codes have opened
up more possibilities of computer use. In this chapter, we use a classical problem – buckling
of non-isotropic plate – to illustrate the application of symbolic algebra, a neo-classical
analytic-numerical tool.

With the advent of high-speed computers, the scope of the structural analysis in all of its
manifestations, including buckling, experienced a dramatic impact. As a result of this
268
quiet “computer revolution,” the analysis of even the most complex structures became
tractable in many circumstances. Parallel with this increase in the capacity for solving
large numbers of problems was a shift in emphasis so that classic analytic techniques
became somewhat neglected, and engineers and researchers started to favor purely
numerical analysis. A new discipline computational mechanics rightfully experienced
huge developments. “Thus with the exception of relatively simple closed-form solu-
tions, there is an increasing tendency to assume indiscriminately numerical methods
are to be preferred even when long-established analytical techniques (whether exact
or approximate) are readily available” (Pavlovic and Sapountzakis, 1986). Analytically
minded researchers often associate this trend with “number crunching syndrome.” Ac-
cording to Beltzer (1990a), researchers were even discussing the dilemma, “join the
computers or fight them.” Some researchers associated results of the numerical analyses
as those that were “untouched by human mind.” Human intellect, however, is getting
around this dilemma by endowing computers with the capacity of performing analytic
operations. Noor and Andersen (1979) stress that “. . . symbolic computation can pro-
vide an important link between analysis on the one hand, and numerical calculations
on the other.” In Beltzer’s words (1990a),
surprisingly enough, the same digital computers now help to restore a delicate equi-
librium between analytic and numerical methods, which promises to be particularly
fruitful. Yet, the arising field of the so-called “computer algebra” or “symbolic compu-
tation” may be useful in overcoming the difficulties due to limited speed the numerical
calculations can be performed with. As noted in Davenport et al. (1988), even CRAY
still needs much more than 48 hours to fairly forecast the weather 48 hours ahead
[present numbers are unavailable to us].
According to Kleiber (1990), the coupled numeric-symbolic computation serves the

purpose of “humanizing” structural mechanics computations. Although this may appear
to be an over-dramatization of the state of affairs, it at least indicates the possibility
that we can exercise more control over computation when computers are equipped with
symbolic algebra software.
From the start of the electronic calculation, the main use of computers has been
to conduct numerical calculations. Computers, or calculating machines, as they were
initially known, deal effectively with numerical calculations in modern engineering
and sciences. Yet, the very idea that computers would perform algebraic calculation
apparently dates back to nearly a century and a half ago, first suggested by Lady
Lovelace. According to Wooff and Hodgkinson (1987), Lady Lovelace was a patron of
Charles Babbage, who is usually credited with the development of the first computer
in the world. Her idea was very timely. Need for huge algebraic computations always
existed. Famous large calculations of the nineteenth century include a large proportion
of algebraic manipulation. The most widely known is Le Verrier’s calculation of the
orbit of Neptune, which started from the disturbances of the orbit of Uranus and which
led to the discovery of Neptune. As Davenport, Siret, and Tournier (1988) mention, the
most impressive calculation with pencil and paper was also in the field of astronomy:
Delaunay took 10 years to calculate the orbit of the moon, and another 10 years to
check the results of his calculations. The result was not numerical because it consisted
270 APPLICATION OF COMPUTERIZED SYMBOLIC ALGEBRA IN BUCKLING ANALYSIS
primarily of a formula, which by itself occupied all the 128 pages of Chapter 4 of his
book.
The first software to realize the idea of Lady Lovelace was developed in the 1950s;
two Masters theses were written almost simultaneously – one at Temple University
by Kahrimanian (1953) and the other one at M.I.T. by Nolan (1953) – and both were
written on the computer implemention of analytic differentiation. About 10 years later,
the first general-purpose systems appeared for algebraic calculations, namely, ALPAK
(a forerunner of ALTRAN) and FORMAC. Since then, the development of systems and
algorithms for processing formulas on computers has been extremely rapid. Presently,
there are a number of general as well as more specialized languages on a broad variety
of computers. Here are a few of the most renowned: DERIVE, MACSYMA, MAPLE,
MATHEMATICA, muMATH, REDUCE, and SCRATCHPAD. Even handheld calcu-
lators with symbolic algebra capability are presently available (Patton, 1987). Some of
these software are discussed in the books by Rayna (1987), Davenport et al. (1988),
and Wolfram (1996). As Davenport et al. (1988) mention, the name of this discipline
has long hesitated between “symbolic and algebraic calculation,” or “symbolic and
algebraic manipulations,” and finally settled down as “computer algebra” in English
and “Calcul Formel,” abbreviated as CALSYF, in French. At the same time, societies
were formed to bring together the research workers and the users of this new discipline:
SIGSAM (Special Interest Group in Symbolic and Algebraic Manipulations) is the
worldwide group of ACM (Association for Computing Machinery). SIGSAM organizes
congresses SYMSAC and EUROSAM and publishes the bulletin entitled SIGSAM. The
European group is called SAME (Symbolic and Algebraic Manipulation in Europe)
and organizes congresses call EUROCAM. For French research workers, there is a
body of the CNRS (Centre national de la recherche scientifique), the GRECO (Groupe
de REcherches COordonées) of computer algebra. Since 1985, a specialized Journal
of Symbolic Computations has been published by Academic Press. Presently, com-
puter algebra in essence became an able and obedient “butler” – a reliable assistant to
researchers and engineers.
In applied mechanics, computer algebra was apparently pioneered by Jensen and
Niordson (1977), Pedersen (1977), Noor and Andersen (1979), Korncoff and Fenves
(1978), Hussain and Noble (1983), and others. The state of the art of applications of
computer algebra in applied mechanics is reflected in books by Rand (1984), Rand and
Ambuster (1987), Klimov and Rudenko (1989), Noor, Elishakoff, and Hulbert (1990),
Beltzer (l990a,b), and others. Extensive literature on applications of symbolic algebra
in vibrations and buckling exists today. Crespo da Silva and Hodges (1986) used it
in rotorcraft dynamics; Nagabhushnan, Gaonkar, and Reddy (1981) developed special
analytic-numerical software for helicopter dynamics; Elishakoff, Hettema, and Wilson
(1989) applied it in deterministic vibration problems; and Rehak et al. (1987) and El-
ishakoff and Hettema (1988) developed special symbolic manipulation techniques for
random vibration analysis of structures; Klimov (1990) developed widely utilized sym-
bolic algebra for solving non-linear dynamics problems; optimal design of laminated
shells was performed by Verijenko et al. (1994); Yoakimidis (1994) used symbolic
algebra for inverse design problems.
7.2 BRIEF REVIEW OF MATHEMATICA® 271
Applications to buckling problems can be found in the papers by Rizzi and Tatone
(1985a, 1985b), Elishakoff and Tang (1988), and Elishakoff and Pletner (1990). This
chapter represents an extended version of the study by Elishakoff and Tang (1988).
Before illustrating the application of the symbolic algebra to a particular buckling
problem, we will briefly review the main points of MATHEMATICA, one of the popular
systems in this field.
7.2 Brief Review of MATHEMATICA®

Among the numerous systems for symbolic computation, MATHEMATICA® appears
to be a newcomer. It was first announced in 1988, and Version 2 was released in 1991.
MATHEMATICA is now available for a wide range of computers, including IBM-
compatible PCs, Macintosh computers, UNIX workstations, and some mainframes. It
is designed to handle analytical calculations as well as those numerical calculations
that are usually carried out in FORTRAN.t Remarkably, in numerical computations,
MATHEMATICA can handle numbers of any precision, which is usually impossi-
ble in traditional numerical analysis due to the hardware limitation in digit length.
Besides, most special functions of mathematical physics, for example, Bessel func-
tions and hypergeometrical functions, as well as linear algebra operations are built into
MATHEMATICA. In addition, MATHEMATICA is a complete graphics programming
language and has an extensive graphics capability for visualizing results of calculations.
In fact, MATHEMATICA’s popularity over several other symbolic languages rests in
its graphics capability.
We now present some simple examples of MATHEMATICA. MATHEMATICA
can be used in the interactive or batch mode. When MATHEMATICA is started, which
is typically done by typing the command math at an operating system, the prompt
In[1]: = pops up, signifying that it is ready for the user’s input. The user can type the
input and then press the RETURN key. MATHEMATICA will process the input and
print out the result, starting with Out[1]: = . Before we start our first example, here are
a few important points pertinent to the use of MATHEMATICA:
■ Arguments of functions are given in square brackets such as in Sqrt[5], Sin[x].
■ Names of built-in functions have their first letters capitalized; for example, Inte-
grate[f(x), x] and DSolve [eqns, y[x], x].
■ Multiplication can be represented by a space, though ∗ could also be used. For
example, a b means a times b.
■ Powers are denoted by ∧ ; for example, y∧ 5 denotes y 5 .
Differentiation
∂f
D[f, x] Partial derivative,
∂x
∂ ∂ ∂
D[f, x1 , x2 , . . . , xn ] Multiple derivative, ··· f
∂ x1 ∂ x2 ∂ xn
∂n f
D[f, {x, n}] Repeated derivative,
∂xn
Example 1:
In[1]: = D[Sin[4x],{x, 2}]

Out[1]: = −(16 Sin[4x])
Integration
#
Integrate[f, x] Indefinite integral, f d x
#b
Integrate[f, {x, a, b}] Definite integral, a f d x
#b #d
integrate[f, {x, a, b}, {y, c, d}] Multiple integral, a d x c f dy
Example 2:
In[2]: = Integrate[Sin[x]∧ 3 + y∧ 2, {x, 0, 1}, {y, 0, x}]

3 − 27 Cos[1] + 3 Cos[3] + 27 Sin[1] − Sin[3]
Out[2]: =
36
Algebraic Equations
Solve[lhs == rhs, x] Solve an equation for x (lhs and rhs stand for
the left-hand side and the right-hand side,
respectively)
Solve[{lhs1 == rhs1 , lhs2 = rhs2 , . . .}, {x, y, . . .}]
Solve a set of simultaneous equations for
x, y, . . .
Eliminate[{lhs1 == rhs1 , lhs2 = rhs2 , . . .}, {x, . . .}]
Eliminate x, . . . in a set of simultaneous
equations
Example 3:
In[3]: = Solve [ax + b == c, x]

b−c
Out[3]: = x−>−
a
Example 4:
In[4]: = Solve [x∧ 2 + y∧ 2 == 1, x + y == 1}, {x, y}]

Out[4]: = {{x − > 0, y − > 1}, {x − > 1, y − > 0}}
Differential Equations
DSolve[eqn, y[x], x] Solve a differential equation for y(x) with x as

the independent variable
7.2 BRIEF REVIEW OF MATHEMATICA® 273
Example 5:
In[5 ]: = DSolve[{y [x] == ay[x], y[0] == 1}, y[x], x]

Out[5 ]: = {{y[x] − > Eax }}
Operations with Polynomials
Expand[expr] Multiply out expression expr

Expand[expr, Trig − > True] Expand out trigonometric functions such as
writing sin(x)2 in terms of sin(2x) and so on
Factor[expr] Reduce expr to a product of factors
Collect[expr, x] Collect terms in expr with like powers of x
Coefficient[expr, form] Collect the coefficient of form in expr
Simplify[expr] Try to obtain a simplest form of expr
Example 6:
In[6 ]: = Expand[(x∧ 2 + y + 1)∧ 2]
Out[6 ]: = 1 + 2x2 + x4 + 2y + 2x2 y + y2
In[7 ]: = Collect[%, y] (% means the last result generated)
Out[7 ]: = 1 + 2x + x + (2 + 2x2 )y + y2
2 4
In[8 ]: = Coefficient[%, x∧ 2y]

Out[8 ]: = 2
In[9 ]: = Simplify[% %] (% % means the next-to-last result generated;
in general, % % · · · % (k times) means the
kth previous result)
Out[9 ]: = (1 + x2 + y)2
Substitution
expr/.x − > value Replace x by value in the expression expr
expr/.{x − > xval, y − > yval} Perform several replacements in an
expression
Example 7:
In[10 ]: = x∧ 2 + y/. x − > z + y

Out[10 ]: = y + (y + z)∧ 2
Arbitrary-Precision Calculations
expr//N or N[expr] Approximate numerical value of expr

N[expr, n] Numerical value of expr with n-digit precision
Example 8:
In[11 ]: = N[Sqrt[7], 30]
Out[11 ]: = 2.64575131106459059050161575364
These are just a few of MATHEMATICA’s capabilities. The interested reader may
consult the comprehensive texts on MATHEMATICA by Wolfram (1996), Abell and
Braselton (1992), Bahder (1995), and Höft and Höft (1998).
7.3 Buckling of Polar Orthotropic Circular Plates on Elastic Foundation

by Computerized Symbolic Algebra
7.3.1 Results Reported in the Literature

In this section, we discuss the application of the symbolic algebra to buckling problems.
It appears instructive to demonstrate its usefulness on specific examples. Since we used
circular plates to demonstrate the Godunov-Conte method in Section 6.1, we will also
utilize circular plates to elucidate the power of symbolic algebra. This will be done in
conjunction with description and use of Rayleigh’s method presently referred to as the
method of optimized parameters.
In a number of interesting papers, Schmidt (1981, 1982, 1983) revived the quite
forgotten version of Rayleigh’s method, suggested by Rayleigh. He gave several appli-
cations of Rayleigh’s method to the buckling of isotropic circular plates clamped or
simply supported at their circumference. The essence of the method is an introduction
of the undetermined power coefficient as one of the parameters describing the trial
function. The desired eigenvalue then becomes a function of this undetermined power.
Because the Rayleigh method provides the upper bound of the eigenvalues (buckling
loads or natural frequencies), the presence of the additional “degree of freedom” – the
non-integer power coefficient – allows one to minimize the eigenvalue with respect to
this coefficient. Thus, the “best” value of the eigenvalue is found within the class of
trial functions created by the undetermined power coefficient. The detailed descrip-
tion of Rayleigh’s method with the new applications of this wonderful idea is given by
Bert (1987).
Since we will focus our attention on buckling of circular plates, we will describe
only the appropriate results reported in the literature. For circular isotropic plates, the
Rayleigh quotient reads
# R 2 2 2 2ν dw d 2 w
D 0 ddrw2 + r1 dw + r dr dr 2 r dr
Nr = dr
# R dw 2 (7.1)
0 dr
r dr
where Nr is the radial compressive load per unit length, D is the flexural rigidity, R is
the radius, w is the displacement, ν is Poisson’s ratio, and r is the radial coordinate. In
non-dimensional form, (6.41) becomes
#1
Nr R 2 [(w,ρρ )2 + (w,ρ /ρ)2 + 2ν(w,ρ /ρ)(w,ρρ )]ρ dρ
N̄ r = = 0 #1 (7.2)
D
0
(w,ρ )2 ρ dρ
7.3 BUCKLING OF POLAR ORTHOTROPIC CIRCULAR PLATES ON ELASTIC FOUNDATION 275
where ρ = r/R and w,ρ = dw/dρ. For the clamped plates, instead of using the trial
functions in either the form ρ − ρ 2 or ρ − ρ 3 utilized in the literature (and yielding
the approximate buckling loads N̄ r = 15 or N̄ r = 16, respectively), Schmidt (1982)
suggested a trial function
w,ρ = ρ − ρ n (7.3)
where n is an undetermined power coefficient.

Using MATHEMATICA, we write the following program segment for calculating
the buckling load:
wp = ro − ro∧ n [Equation (7.3) with wp − > w,ρ , ro − > ρ]
fden = wp∧ 2 ro [fden represents the integrand in the denominator of
Equation (7.2)]
fnum = (D[wp, ro]∧ 2 + (wp/ro)∧ 2 + 2 nu wp/ro D[wp, ro]) ro
[fnum represents the integrand in the numerator of Equation (7.2);
nu − > ν]
nr = Integrate[fnum, {ro, 0, 1}]/Integrate[fden, {ro, 0, 1}] (nr − > N̄ r )
The resulting expression for N̄ r is

2(n + 1)(n + 3)
N̄ r = (7.4)
n
Minimizing N̄ r with respect to n, or requiring N̄ r,n = 0, yields n 2 = 3, with buckling
load
√
N̄ r = 4(2 + 3) 14.928 (7.5)
which is 1.66% higher than the exact value 14.682.

For the clamped plate on the elastic foundation, Bert (1987) applied the following
buckling shape:
w = (1 − ρ n )2 (7.6)
which satisfies the boundary conditions w(1) = 0, w (1) = 0 and the regularity condi-
tions
w,ρ (0) = 0, w,ρρ (0) < ∞ (7.7)
provided that n > 1. For the plate without elastic foundations, the buckling load equals

1 8 4
N̄ r,o = 6n 3 − + (7.8)
n − 1 3n − 2 3n − 1
The minimal value is attained at n = 1.722 and is N̄ r,o = 15.161 or 3.25% higher than
the exact value. For the plate on elastic foundation with modulus q,

q R4 3 8 6 8 1
N̄ = N̄ o + 1− + − + (7.9)
D 2n n + 2 n + 1 2n + 2 2n + 1
and the minimal value occurs at n = 1.63 with N̄ = 29.2 for qR4 /D = 100. This turned
out to be 5.07% higher than the exact value found by Kline and Hancock (1965).
In this section, we consider the buckling of clamped and simply supported polar
orthotropic plates on elastic foundations.
7.3.2 Statement of the Problem

Several studies of buckling of polar orthotropic circular plates under radial compres-
sion are reported in the literature. The axisymmetric buckling problem was solved
by Woinowsky-Krieger (1958) with the aid of Bessel functions. Later, Pandalai and
Patel (1965) used a power-series expansion for the same problem. The axisymmetric
buckling problem was treated by Mossakowsky (1960) and Mossakowsky and Borsuk
(1960) by the Frobenius method with the aid of hypergeometric functions. We are fo-
cusing here on using the Rayleigh quotient, with an objective of obtaining the accurate
results within the variable parameter approach. For the polar orthotropic circular plate,
the Rayleigh quotient is obtained from the requirement
d
(V + T ) = 0 (7.10)
dC
where the trial function is represented for the axisymmetric buckling as
W (r, θ ) = Cw(r ), (7.11)
and
2
R
d2W 2 d w 1 dW 2 1 dW 2
V =π Dr + 2D1 + Dθ + qW r dr
2
0 dr 2 dr 2 r dr r dr
(7.12)
R 2
dW
T = πh σr r dr (7.13)
0 dr
where Dr , D1 = νθ Dr , and Dθ are the flexural rigidities of the plate, νθ is Poisson’s
ratio, σr is the normal stress that equals (Woinowsky-Krieger, 1958)
σr = −Nr (r/R)k−1 (7.14)
where Nr is the radial compression, k is the orthotropy coefficient

k = E θ /Er = Dθ /Dr (7.15)
where Er and E θ are Young’s moduli in radial and circumferential directions, respec-
tively, and q is the elastic foundation modulus. Using the non-dimensional coordinate
ρ = r/R, we get for the buckling load the Rayleigh quotient
#1
Nr R 2 [(w,ρρ )2 + 2νθ (w,ρρ )(w,ρ /ρ) + k 2 (w,ρ /ρ)2 + Qw 2 ]ρ dρ
N̄ = = 0 #1
Dr ρ k (w,ρ )2 dρ
0
(7.16)
where
Q = qR4 /Dr (7.17)
The problem involves evaluating this quotient so as to yield accurate estimates of

buckling loads of plates with or without elastic foundations, under different boundary
conditions.
7.3.3 Buckling of Clamped Polar Orthotropic Plates

We first employ the variable power method. We use the trial function
w,ρ = (1 − ρ n+1 )2 (7.18)
[with n + 1 instead of n in (7.6) for numerical convenience]. For a buckling load, at

fixed non-dimensional stiffness of the elastic foundations, Q = 100, we get
N̄ = I /J (7.19)
I = 288n 13 + n 12 (312k + 4200) + n 11 (252k 2 + 4360k + 27,236)
+ n 10 (168k 3 + 3468k 2 + 26, 854k + 125,654)
+ n 9 (54k 4 + 2254k 3 + 20,977k 2 + 119,818k + 125,654)
+ n 8 (6k 5 + 693k 4 + 13,205k 3 + 81,839k 2 + 368,273k + 886,080)
+ n 7 (73k 5 + 3840k 4 + 45,481k 3 + 212,123k 2 + 736,814k + 1,296,693)
+ n 6 (378k 5 + 12,087k 4 + 100,585k 3 + 361,745k 2 + 959,417k + 1,289,712)
+ n 5 (1091k 5 + 23,862k 4 + 147,447k 3 + 406,559k 2 + 819,022k + 873,523)
+ n 4 (1924k 5 + 30,687k 4 + 144,685k 3 + 301,061k 2 + 454,435k + 396,520)
+ n 3 (2127k 5 + 25,740k 4 + 94,039k 3 + 144,745k 2 + 157,666k + 115,371)
+ n 2 (1442k 5 + 13,593k 4 + 38,947k 3 + 43,347k 2 + 31,029k + 19,434)
+ n(549k 5 + 4104k 4 + 9339k 3 + 7344k 2 + 2640k + 1440)
+ 90k 5 + 540k 4 + 990k 3 + 540k 2 (7.20)
J = 4n(36n + 468n + 2639n + 8509n + 17,377n + 23,473n
10 9 8 7 6 5
+ 21,211n 4 + 12,631n 3 + 4727n 2 + 999n + 90) (7.21)
The requirement of N̄ ,n = 0 yields a twenty-third-order polynomial equation (k = 1)

for n:
864n 23 + 23,616n 22 + 302,640n 21 + 2,385,756n 20 + 13,003,362n 19

+ 52,546,379n 18 + 165,046,136n 17 + 416,094,782n 16 + 858,351,464n 15
+ 1,459,534,754n 14 + 2,038,352,060n 13 + 2,304,580,458n 12
+ 2,048,146,644n 11 + 1,342,033,680n 10 + 528,482,612n 9 − 40,547,266n 8
− 248,719,048n 7 − 214,576,018n 6 − 113,350,828n 5 − 41,363,230n 4
− 10,511,286n 3 − 1,780,011n 2 − 179,820n − 8100 = 0 (7.22)
For the plate without an elastic foundation, the expressions for I and J are much
simpler:
I = k 5 + k 4 (9n + 6) + k 3 (28n 2 + 35n + 11) + k 2 (42n 3 + 67n 2 + 35n + 6)
+ k(52n 4 + 94n 3 + 53n 2 + 11n) + 48n 5 + 116n 4 + 104n 3 + 41n 2 + 6n
(7.23)
J = 4n(6n + 5n + 1)
2
(7.24)
The equation for N̄ ,n = 0 for k = 1 yields an equation
24n 7 + 72n 6 + 78n 5 + 23n 4 − 28n 3 − 30n 2 − 10n − 1 = 0 (7.25)
Simpler equations are obtained if one uses (7.3) as the trial function:
w,ρ = ρ − ρ n (7.26)
for the plate without elastic foundation. The buckling load becomes
I = n 3 (6 + 2k) + n 2 (2k 3 + 9k 2 + 14k + 15) + n(3k 4 + 15k 3 + 21k 3 + 11k + 6)
+ (k 5 + 6k 4 + 11k 3 + 6k 2 ) (7.27)
J = 4n(n + 1) (7.28)
The minimization requirement is
n 4 (2k + 6) + n 3 (4k + 12) + n 2 (−3k 4 − 13k 3 − 12k 2 + 3k + 9)
+ n(−2k 5 − 12k 4 − 22k 3 − 12k 2 ) − k 5 − 6k 4 − 11k 3 − 6k 2 = 0 (7.29)
For k = 1, we have the following roots:
√
n 1 = −1.00078, n 2 = −√ 3,
(7.30)
n 3 = −0.999218, n4 = 3
Only the last root is meaningful, with the buckling load 14.928 reported in Section 7.3.2.
For k = 2, we get
n 1 = 0.695678, n 2 = −1.88,
(7.31)
n 3 = −4, n 4 = 4.57915
The last root has a significance and yields
N̄ = 43.7843 (7.32)
Finally, for k = 3,
n 1 = −0.6259, n 2 = −2.63,
(7.33)
n 3 = −6.78, n 4 = 8.04277
with
N̄ = 92.0826 (7.34)
associates with n = n 4 .
The undetermined power variant of Rayleigh’s method in this problem appears to

be relatively complex. For the trial function (7.18), it leads to the necessity to solve a
polynomial of the twenty-third degree, whereas the trial function in (7.26) leads to one
of the fourth degree. Under these circumstances, it appears that this undetermined power
version may not always be preferable to the predetermined power variant of it. Indeed,
substituting n = 2 in (7.26), or using w jρ = ρ − ρ 2 , leads to N̄ = 15, which is also
2.166% higher than the exact value; the minimization procedure led to N = 14.928. The
actual “gain” may appear illusory. This is especially true if one recalls that Rayleigh’s
method was designed, as well as applied, to get quick estimates.
These considerations naturally lead us to question whether it is possible to ap-
ply another version of the Rayleigh method to eliminate these disadvantages while
retaining all the advantages of the original non-integer-power method. Here we use
another version of the variable parameter method. It is equivalent to using the two-term
Rayleigh-Ritz method (Schmidt, 1982).
To illustrate this undetermined multiplier method, consider first the plate without
elastic foundation. Instead of the conventionally used trial functions, ρ − ρ 2 or ρ − ρ 3
for the plate’s shape w,ρ , Schmidt (1982) used a variable power ρ − ρ n . Here we use a
variable multiplier to form the following trail function:
w,ρ = (ρ − ρ 2 ) + n(ρ − ρ 3 ) (7.35)
The expressions for I and J for the buckling load become
I = n 2 (10k 7 + 250k 6 + 2480k 5 + 12,500k 4 + 34,890k 3 + 60,450k 2

+ 82,620k + 75,600) + n(14k 7 + 350k 6 + 3464k 5 + 17,300k 4
+ 46,886k 3 + 75,230k 2 + 93,636k + 85,680) + (5k 7 + 125k 6
+ 1235k 5 + 6125k 4 + 16,220k 3 + 24,350k 2 + 27,540k) (7.36)
J = 120[n 2 (4k 2 + 40k + 96) + n(4k 2 + 46k + 126) + (k 2 + 13k + 42)] (7.37)
As we observe, the buckling load is given as a ratio of two quadratic expressions

in terms of the undetermined power. Therefore, the minimization requirement N̄ ,n = 0
is also a quadratic equation:
n 2 (−8k 9 − 250k 8 − 3260k 7 − 22,890k 6 − 92,524k 5 − 207,710k 4

− 188,740k 3 + 204,690k 2 + 735,732k + 650,160)
+ n(−10k 9 − 320k 8 − 4270k 7 − 30,660k 6 − 126,890k 5 − 295,180k 4
− 307,430k 3 + 148,560k 2 + 801,000k + 756,000)
+ (−3k 9 − 99k 8 − 1359k 7 − 10,014k 6 − 42,483k 5 − 101,961k 4
− 116,571k 3 + 13,434k 2 + 208,656k + 211,680) = 0 (7.38)
with roots
−5k 4 − 35k 3 − 35k 2 − 5k + 150 ± A
n 1,2 = (7.39)
2(4k 4 + 25k 3 + 25k 2 − 5k − 129)
where
A = [k 8 + 8k 7 + 33k 6 + 58k 5 + 111k 4 + 32k 3 + 7k 2 − 18k + 828]1/2 (7.40)
Upon substituting n 1,2 into Equations (7.36) and (7.37), we obtain the buckling
loads for specific values of the orthotropy coefficients. For k = 1, we obtain for n = n 1
7 √
N̄ = (29 − 265) 14.841 (7.41)
6
which is 1.08% higher than the exact value. For k = 1.5,
143 √
N̄ = (459 − 5 2085) 27.118 (7.42)
1024
For k = 2, N̄ = 42. For k = 2.5, we get
221
N̄ = (3409 − 6,602,773) 60.388 (7.43)
3072
The buckling load equals N̄ = 84 for k = 3. The values of the buckling loads re-
ported here virtually coincide with the values read from the paper by Woinowsky-
Krieger (1958, Figure 3), who obtained an exact solution in terms of Bessel functions.
Moreover, the comparison with the results obtained through the use of the undetermined
power method reveals that the variable parameter method yields both more straightfor-
ward (without recourse to high-degree polynomial equations) and more reliable results
(the furnished upper bounds are lower).
Consider now the polar orthotropic plate on elastic foundation. Rayleigh’s quotient,
given in (7.16), contains also a term w2 so that we approximate the displacement rather
than the slope:
w = (1 − 3ρ 2 + 2ρ 3 ) + n(1 − 4ρ 3 + 3ρ 4 ) (7.44)
Equation (7.16) can again be put as a ratio I /J with
I = n 2 (1008k 7 + 25,200k 6 + 56k 5 Q + 253,008k 5 + 1400k 4 Q + 1,335,600k 4

+ 13,720k 3 Q + 4,257,792k 3 + 65,800k 2 Q + 9,646,560k 2 + 154,224k Q
+ 16,656,192k + 141,120Q + 15,240,960) + n(2016k 7 + 50,400k 6
+ 89k 5 Q + 500,976k 5 + 2225k 4 Q + 2,545,200k 4 + 21,805k 3 Q
+ 7,280,784k 3 + 104,575k 2 Q + 13,371,120k 2 + 245,106k Q
+ 19,432,224k + 224,280Q + 17,781,120) + (1260k 7 + 31,500k 6
+ 36k 5 Q + 311,220k 5 + 900k 4 Q + 1,543,500k 4 + 8820k 3 Q
+ 4,087,440k 3 + 42,300k 2 Q + 6,136,200k 2 + 99,144k Q
+ 6,940,080k + 90,720Q + 6,350,400) (7.45)
J = 30,240[n (4k + 28k + 48) + n(4k + 40k + 84) + (k + 13k + 42)]
2 2 2 2
(7.46)
Again, to find the minimal value of the buckling load, we should solve a quadratic
equation
n 2 (−4032k 9 − 116,928k 8 − 132k 7 Q − 1,407,168k 7 − 3552k 6 Q
− 9,047,808k 6 − 38,208k 5 Q − 32,727,744k 5 − 206,040k 4 Q
− 58,427,712k 4 − 553,788k 3 Q + 8,543,808k 3 − 519,048k 2 Q
+ 280,482,048k 2 + 554,688k Q + 578,140,416k + 210,325,248k
+ 1,088,640Q + 426,746,880) + n(−8064k 9 − 245,952k 8 − 176k 7 Q
− 3,128,832k 7 − 4960k 6 Q − 21,434,112k 6 − 55,872k 5 Q
− 84,518,784k 5 − 312,800k 4 Q − 183,976,128k 4 − 836,944k 3 Q
− 149,764,608k 3 − 519,360k 2 Q + 245,331,072k 2 + 2,025,792k Q
+ 773,515,008k + 3,144,960Q + 670,602,240) + (−3024k 9
− 99,792k 8 − 55k 7 Q − 1,369,872k 7 − 1658k 6 Q − 10,094,112k 6
− 19,836k 5 Q − 42,822,864k 5 − 116,110k 4 Q − 102,776,688k 4
− 309,065k 3 Q − 117,503,568k 3 − 79,032k 2 Q + 13,541,472k 2
+ 1,253,196k Q + 210,325,248 + 1,799,280Q + 213,373,440) = 0
(7.47)
with roots
n 1,2 = [−2016k 4 − 11,088k 3 − 44k 2 Q − 11,088k 2 − 140k 2 Q + 4032k
+ 312Q + 66,528 ± B][6(336k 4 + 1344k 3 + 11k 2 Q + 1344k 2
+ 21k Q − 3696k − 36Q − 14,112)]−1 (7.48)
where
B = [14(53Q 2 − 22,752Q + 4,808,160)]1/2 (7.49)
Roots n 1,2 being substituted into N̄ = I /J yield the minimal buckling loads for
any combination of the orthotropy coefficient k and the non-dimensional stiffness Q to
the elastic foundation. Consider a number of specific cases. For an isotropic plate, we
get
N̄ = [C(1525Q 2 − 139,104Q − 18,797,184)
+ (41,552Q 3 ± 122,602,016Q 2 + 1,522,063,872Q + 474,969,277,440]
× {144[68CQ − 15,372C + 1855Q 2 ± 796,320Q + 168,285,600]}−1
(7.50)
where
C = [14(53Q 2 − 22,752Q + 4,808,160)]1/2 (7.51)
For a plate without elastic foundation, Q = 0 and
√
14(1855 + 37 265)
N̄ = √ 14.841 (7.52)
1325 + 61 265
which coincides with Equation (7.41), which was derived by using the trial function
w,ρ = (ρ − ρ 2 ) + m(ρ − ρ 3 ). The coincidence of the buckling loads is understandable
due to the proportionality of the constituents of the trial function
(1 − 3ρ 2 + 2ρ 3 ),ρ ∼ ρ − ρ 2 (7.53)
(1 − 4ρ + 3ρ ),ρ ∼ ρ − ρ
3 4 3
(7.54)
For an isotropic plate with elastic foundation, without non-dimensional stiffness Q =

100,
1
N̄ = (4654 − 2,680,090) 27,934 (7.55)
108
which is only 0.5% higher than the value reported by Kline and Hancock (1965).
Interestingly, the present approach yields a lower value than an approximate method
utilized by Dinnik (1955). For the isotropic clamped plate, he suggested using the
formula
PR2 2Q
N̄ = = α12 + 2 (7.56)
D α1
√
where α1 = 3.8317 is the first root of the equation J1 ( P/DR) = 0, with J1 (x) being
the Bessel function of the first order. For Q = 0 Dinnik’s approximation coincides with
the exact result, but for Q = 100 Dinnik’s formula yields N̄ = 28.3041.
For orthotropy coefficient k = 2, we arrive at
N̄ = [D(881Q 2 − 710,640Q − 152,409,600) + 7728Q 3 ± 10,039,680Q 2

+ 4,343,673,600Q − 1,152,216,576,000]
× [360(9DQ − 10,080D + 92Q 2 ± 136,060Q + 76,204,800)]−1 (7.57)
where
D = [3(23Q 2 − 34,020Q + 19,051,200)]1/2 (7.58)
For the specific value Q = 100, we obtain

1
N̄ = (2553 − 4 119,094) 65.144 (7.59)
18
When k = 3, we get
N = [E(43Q 2 − 93,120Q − 27,820,800) + 56Q 3 ± 11,760Q 2

+ 254,016,000Q + 79,543,872,000][24(2EQ − 14,880E + 7Q 2
± 3,360Q + 32,659,200)]−1 (7.60)
where
E = (Q 2 − 480Q + 4,665,600)1/2 (7.61)

For Q = 100, the buckling load is

1
N̄ = (2942 − 21 11,569) 113,876 (7.62)
6
It also appears instructive to study how the choice of the trial function influences
the results of the buckling loads. Bert (1987) used the trial function w = (1 − ρ n )2 ,
within the undetermined power version of the Rayleigh method, for the isotropic plate
on elastic foundation. We will utilize trial functions of this class with predetermined
powers, but with an undetermined multiplier,
w = (1 − ρ 2 )2 + n(1 − ρ 3 )2 (7.63)
for the orthotropic plates on elastic foundation. For I and J , we get, respectively,
(for Q = 100)
I = n 2 (8019k 10 + 425,007K 9 + 9,789,741k 8 + 129,854,583k 7
+ 1,115,908,713k 6 + 6,698,662,173k 5 + 29,784,804,759k 4
+ 100,676,192,517k 3 + 247,223,400,750k 2 + 382,186,910,520k
+ 269,168,961,600) + n(15,246k 10 + 808,038k 9 + 18,526,805k 8
+ 242,337,205k 7 + 2,018,314,298k 6 + 11,422,984,804k 5
+ 46,426,816,205k 4 + 141,823,272,745k 3 + 322,949,067,606k 2
+ 483,513,979,608k + 340,532,216,640) + (9240k 10 + 489,720k 9
+ 11,187,330k 8 + 144,696,090k 7 + 1,173,812,640k 6
+ 6,295,078,020k 5 + 23,257,835,370k 4 + 62,234,428,410k 3
+ 125,371,558,620k 2 + 176,747,099,760k + 124,480,540,800) (7.64)
J = n 2 (81k 5 + 2349k 4 + 26,325k 3 + 142,155k 2 + 368,874k + 367,416)
+ n(72k 5 + 2412k 4 + 30,996k 3 + 190,548k 2 + 557,892k + 617,760)
+ (16k 5 + 608k 4 + 9008k 3 + 6486k 2 + 226,176k + 304,128) (7.65)
The minimization requirement yields (for Q = 100)
n 2 (−657,558k 15 − 51,321,600k 14 − 1,821,568,257k 13 − 38,923,767,378k 12
− 558,486,724,554k 11 − 5,676,827,194,026k 10 − 41,947,587,724,428k 9
− 226,518,671,150,406k 8 − 875,831,248,231,164k 7
− 2,239,968,412,170,126k 6 − 2,605,843,279,603,827k 5
+ 5,359,615,554,667,824k 4 + 32,351,700,422,843,676k 3
+ 71,812,607,632,193,952k 2 + 83,002,742,955,266,112k
+ 41,164,832,809,013,760) + n(−1,240,272k 15 − 99,392,832k 14
− 3,624,987,492k 13 − 79,652,738,664k 12 − 1,176,254,942,760k 11
− 12,323,476,885,464k 10 − 94,159,522,997,280k 9
− 529,897,561,143,624k 8 − 2,179,562,582,640,600k 7
− 6,313,972,266,390,024k 6 − 11,393,884,883,247,228k 5
− 5,596,409,574,852,912k 4 + 32,469,254,939,758,032k 3
+ 100,263,966,172,139,520k 2 + 132,511,904,737,977,600k
+ 72,251,351,149,824,000) + (−421,344k 15 − 35,348,544k 14
− 1,348,043,488k 13 − 30,932,567,512k 12 − 476,369,683,504k 11
− 5,198,129,429,856k 10 − 41,333,185,703,040k 9
− 242,248,300,830,096k 8 − 1,043,053,957,383,680k 7
− 3,222,489,305,035,616k 6 − 6,661,690,903,240,736k 5
− 7,048,828,900,788,312k 4 + 4,678,986,305,136,528k 3
+ 29,890,548,464,078,016k 2 + 45,436,368,005,259,264k
+ 26,666,283,097,681,920) (7.66)
with roots
n 1,2 = [−34,452k 7 − 934,956k 6 − 9,661,221k 5 − 47,659,833k 4 − 112,317,597k 3
− 59,691,411k 2 + 416,475,270k + 1,005,582,600 + 2F]
× [9(8,118k 7 + 203,346k 6 + 1,937,087k 5 + 8,824,507k 4
+ 18,042,461k 3 − 9,824,949k 2 151,882,146k − 254,633,544)]−1 (7.67)
where
F = (331,822,656k 14 + 16,583,919,264k 13 + 367,934,944,608k 12
+ 4,787,539,400,556k 11 + 40,663,247,789,641k 10
+ 237,527,161,809,208k 9 + 979,989,247,011,847k 8
+ 2,877,371,177,782,438k 7 + 5,808,079,295,573,071k 6
+ 6,446,393,327,811,628k 5 − 2,663,017,967,877,591k 4
− 19,081,572,469,289,766k 3 − 3,269,554,711,073,064k 2
+ 88,717,428,179,238,672k + 160,660,980,513,392,832) (7.68)
We consider specific cases to get more insight. For the isotropic case without elastic
foundation, we get
1
N̄ = (9,961 − 37,187,185) 14.6877 (7.69)
263
which coincides with the exact solution within four significant digits. This is probably
the approximate result that is closest to the exact solution, using the trial functions,
reported here. For an isotropic plate on elastic foundation with Q = 100, we have
1
N̄ = (2,465,703 − 7 21,325,774,110) 27.7196 (7.70)
52,074
For the plate with orthotropy coefficient k = 2 and without elastic foundation, we arrive
at
693
N̄ = (60,331 − 4 73,778,091) 43.2902 (7.71)
415,787
Table 7.1. Values of buckling loads for clamped plates without elastic foundation
k=1 k=2 k=3

w,ρ = ρ − ρ n 14.928a 43.7843 92.0826
w,ρ = (ρ − ρ 2 ) + n(ρ − ρ 3 ) 14.841 42 84
w = (1 − 3ρ 2 − 2ρ 3 ) + n(1 − 4ρ 3 + 3ρ 4 ) 14.841 42 84
w = (1 − ρ 2 )2 + n(1 − ρ 3 )2 14.6877 43.2902 82.8093
a Result reported by Schmidt (1982).
which is higher than the value N̄ = 42 obtained with Equation (7.35) as the trial func-
tion. For Q = 100, the result is N̄ = 67.5251. For k = 3, Q = 0,
3
N̄ = (54,523 − 9 18,760,889) 82,8093 (7.72)
563
which is lower than the value N̄ = 84 obtained with (7.35). For k = 3, Q = 100,
1
N̄ = (2,136,503 − 9 23,807,753,219) 110.7120 (7.73)
6756
The lesson to be learned by “trying” various trial functions in the same problem is
that the degree of accuracy depends on the orthotrophy coefficient k: Whereas for one
particular value of k the specified trial function may be “better,” the same trial function
may perform “worse” for another value of k. Generally, however, the undetermined
multiplier (or, in other words, the Rayleigh-Ritz) method is much easier to implement
than its undetermined power counterpart (Table 7.1).
7.3.4 Buckling of Simply Supported Polar Orthotropic Plates

For a simply supported isotropic plate, Schmidt used the trial function
n+ν
w,ρ = ρ − ρn (7.74)
1+ν
satisfying both the boundary conditions at the circumference and the regularity con-
ditions at the center; ν is the Poisson’s ratio. The minimum occurs at n = 2.5 for
which N̄ = 4.20, agreeing precisely with the exact value reported by Timoshenko and
Gere (1961) to three significant figures.
We use the following simple function:
ψ = 1 − ρ 3 + n(ρ 2 − ρ 3 ) (7.75)
which satisfies the geometric boundary condition at ρ = 1 – namely w(1) = 0 – as well
as the regularity conditions at the center.
I = n 2 (k 5 + 12k 4 + 4k 3 νθ + 59k 3 + 48k 2 νθ + 204k 2 + 188kνθ + 564k
+ 240νθ + 720) + n(2k 5 + 24k 4 + 24k 3 νθ + 134k 3 + 288k 2 νθ + 600k 2
+ 1,128kνθ + 1880k + 1440νθ + 2400) + (9k 5 + 108k 4 + 36k 3 νθ + 459k 3
+ 432k 2 νθ + 972k 2 + 1692kνθ + 1692k + 2160νθ + 2160) (7.76)
J = 4[n 2 (k 2 + 3k + 8) + n(6k 2 + 30k + 36) + (9k 2 + 63k + 108)] (7.77)

For νθ = 0.3, the minimization requirement leads to the quadratic
n 2 (5k 7 + 90k 6 + 660k 5 + 2808k 4 + 8793k 3 + 20,310k 2 + 24,814k + 7320)
+ n(90k 6 + 1350k 5 + 9198k 4 + 41,454k 3 + 128,952k 2
+ 227,196k + 157,680) + (−45k 7 − 720k 6 − 4230k 5 − 8658k 4
+ 19,125k 3 + 146,106k 2 + 318,222k + 255,960) = 0 (7.78)
with roots
n 1,2 = [3(−15k 3 − 45k 2 − 288k − 438 ± G)]
× [5k 4 + 30k 3 + 65k 2 + 318k + 122]−1 (7.79)
where
G = (25k 8 + 250k 7 + 1225k 6 + 3000k 5 + 8400k 4 + 11,520k 3 + 21,620k 2
+ 74,960k + 134,016)1/2 (7.80)
For the case of the isotropic plate, we arrive at, for n = n 2 ,
√
101 − 7081
N̄ = 4.213 (7.81)
4
For the orthotropy coefficient k = 2, for n = n 2 ,

3 √
N̄ = (98 − 5579) 13.984 (7.82)
5
Whereas for k = 3, we get for n = n 2 ,
7
N̄ = (609 − 237,801) 28.135 (7.83)
30
For the plate on elastic foundation, to the numerator J should be added the following
increment:
J = Q[5k 3 n 2 + 54k 3 n + 189k 3 + 60k 2 n 2 + 648k 2 n + 2268k 2 + 235kn 2
+ 2538kn + 8883k + 300n 2 + 3240n + 11,340] (7.84)
Accordingly, the minimization condition changes too. For Q = 100, the minimal values
of N̄ occur at
n 1,2 = [3(−105k 3 − 115k 2 − 1666k − 1716 ± H )]
× [35k 4 + 210k 3 + 355k 2 + 2176k − 196]−1 (7.85)
where
H = (1225k 8 + 12,250k 7 + 62,125k 6 + 189,00k 5 + 554,600k 4 + 521,480k 3
+ 1,064,380k 2 − 429,760k + 3,506,784)1/2 (7.86)
Let us list some results. For k = 1 we have, for n = n 1 ,

1
N̄ = (2,971 − 1,370,521) 21.432 (7.87)
84
which is very close to the extrapolated value 20 read from the curve in the paper by
Kline and Hancock (1965). Our result is also in good agreement with the approximate
formula by Dinnik (1955) for isotropic plates,
Q α 2 − 2 + 2ν 2
N̄ = α 2 + (7.88)
α2 α2 − 1 + ν 2
where α = 2.047 and which is the first root of the transcendental equation
3 3 3
P P P
R J0 R − (1 − ν)J1 R =0 (7.89)
D D D
for the buckling of the plate without elastic foundation, ν being the Poisson ratio.
For Q = 100, ν = 0.3, Dinnik’s formula yields
N̄ = 20.7799 (7.90)
so that our result is 3.13% higher than Dinnik’s approximation. Since Dinnik’s value is
also an upper bound, this means that the percentagewise error of (7.87) in comparison
to the exact solution is at least 3.13%. We are unaware of results in the literature for the
polar orthotropic plate.
Let us list some results; at k = 2,
3 √
N̄ = (861 − 13 1309) 33.485 (7.91)
35
for k = 3, for n = n 1 ,
1 √
N̄ = (2469 − 49 1281) 47.682 (7.92)
15
The values of buckling loads are listed in Table 7.2.
Note that Makushin (1962) suggested an analogous method for the buckling of
isotropic circular plates. He utilized the beam deflection functions stemming from two
different loads, T1 and T2 , which resulted in deflections
w1 = T1 ϕ1 , w2 = T2 ϕ2 (7.93)
Table 7.2. Values of buckling loads for clamped plates with elastic foundation
k=1 k=2 k=3

w = (1 − ρ )n 2
29.2[5] — —
w = (1 − 3ρ 2 + 2ρ 3 ) + n(1 − 4ρ 3 + 3ρ 4 ) 27.934 65.144 113.879
w = (1 − ρ 2 )2 + n(1 − ρ 3 )2 27.7196 67.5251 110.7120
The Makushin method combined these two functions to construct the trial function
T2
w = ϕ1 + ϕ2 = ϕ1 + nϕ2 (7.94)
T1
where n is an undetermined multiplier. He then solved the problem of the buckling of

a clamped circular plate with trial functions
w = (2 + n) − (6 + 2n)ρ 2 + 4ρ 3 + nρ 4 (7.95)
or
w = (10 + 3n) − 5(4 + n)ρ 2 + 10ρ 3 + 2nρ 5 (7.96)
For the isotropic plate, which is simply supported, he used
w = (8 + 5n) − (12 + 6n)ρ 2 + 4ρ 3 + nρ 4 (7.97)
or
w = (25 + 9n) − 3(10 + n)ρ 2 + 5ρ 4 + nρ 5 (7.98)
Equations (7.93)–(7.98) are based on the following observation. For the axisym-
metric buckling modes, the shape w,ρ vanishes. Accordingly, Makushin (1962) used
the functions that satisfy the condition w,ρ = 0 at ρ = 0 and the appropriate boundary
condition at the circumference. For clamped plates, functions in (7.93) correspond to
the displacements due to uniformly distribute (T1 ) and to concentrate force at the cen-
ter (T2 ), respectively; note that the slope w,ρ coincides with the assumption we used
in (7.35), from other considerations. Therefore, our numerical result for the isotropic
plate coincides with that of Makushin (Table 7.2).
7.3.5 Buckling of Free Polar Orthotropic Plates

We employ the following trial function:
w = ρ 2 + nρ 3 (7.99)
The results for I and J read
I = n 2 (9k 5 + 108k 4 + 36k 3 νθ + 459k 3 + 432k 2 νθ + 972k 2 + 1692kνθ

+ 1692k + 2160νθ + 2160) + n(16k 5 + 192k 4 + 48k 3 νθ + 784k 3
+ 576k 2 νθ + 1344k 2 + 2256kνθ + 1504k + 2880νθ + 1920)
+ (8k 5 + 96k 4 + 16k 3 νθ + 384k 3 + 192k 2 νθ + 576k 2 + 752kνθ
+ 376k + 960νθ + 480) (7.100)
I = 4[n 2 (9k 2 + 63k + 108) + n(12k 2 + 96k + 180) + (4k 2 + 36k + 80)] (7.101)
The minimization requirement is also a quadratic equation

n 2 (−9k 7 − 144k 6 − 846k 5 + 108k 4 νθ − 1764k 4 + 1620k 3 νθ + 3339k 3
+ 8964k 2 νθ + 26,532k 2 + 21,708kνθ + 57,132k + 19,440νθ + 45,360)
+ n(−18k 7 − 306k 6 − 1962k 5 + 144k 4 νθ − 5346k 4 + 2304k 3 νθ − 1332k 3
+ 13,680k 2 νθ + 28,548k 2 + 35,712kνθ + 71,136k + 34,560νθ + 60,480)
+ (−8k 7 − 144k 6 − 984k 5 + 48k 4 νθ − 3024k 4 + 816k 3 νθ − 2952k 3
+ 5136k 2 νθ + 5952k 3 + 14,160kνθ + 18,920k + 14,400νθ + 16,800) = 0
(7.102)
with roots
n 1,2 = [3k 4 − 15k 3 − 6k 2 + 24kνθ + 66k + 96νθ + 168 ± L]
× [3(k 4 + 4k 3 − k 2 − 12kνθ − 40k − 36νθ − 84)]−1 (7.103)
where
L = (k 8 + 10k 7 + 45k 6 + 120k 5 + 336k 4 − 144k 3 νθ + 504k 3 − 144k 2 νθ
+ 908k 2 + 288kνθ + 2912k + 576νθ2 + 2016νθ + 4707)1/2 (7.104)
Substitution of specific values of νθ = 0.3 and k = 1, 2, 3 leads to the conclusion
that the buckling loads of the free plate coincide with those of the simply supported
plates. Indeed, in the absence of an elastic foundation, the trial function itself does not
enter in the analysis but rather in the slope.
The first derivative of the trial function used for the simply supported cases can be
represented as

(m + 1) 2
[1 − ρ 3 + m(ρ 2 − ρ 3 )],ρ = m 2ρ − 3ρ (7.105)
m
Then, with
−(m + 1)/m ≡ M (7.106)
the simply supported plate’s slope coincides with that of the free plate.
As we have seen, within trial functions’ choice, the free and simply supported plates
share the same buckling load when they are unsupported by the elastic foundation. Note
that the computerized symbolic algebra proved extremely useful to obtain closed-form
solutions in buckling of structures for columns (Elishakoff and Rollot, 1999) and plates
(Elishakoff, 2000).
Bibliography
He who is able to write a book and does not write it is as one who has lost a child.
Nahman of Bratslav
Abell, M. L., and Braselton, J. P., The Mathematica Handbook, AP Professional, Boston, 1992.
Aboudi, J., Cederbaum, G., and Elishakoff, I., Dynamic stability analysis of viscoelastic plates by
Lyapunov exponents, Journal of Sound and Vibration, Vol. 139, 459–68, 1990.
Abramovich, H., and Elishakoff, I., Application of the Krein’s method for determination of natural
frequencies of periodically supported beam based on simplified Bresse-Timoshenko equations,
Acta Mechanica, Vol. 66, 39–59, 1987.
Abramovich, H., Singer, J., and Yaffe, R., Imperfection characteristics of stiffened shells-Group 1,
TAE Report 406, Technion, Israel Institute of Technology, Department of Aeronautical Engineering,
Haifa, 1981.
Adali, S., Elishakoff, I., Richter, A., and Verijenko, V. E., Optimal design of symmetric angle-ply
laminates for maximum buckling load with scatter in material properties, Paper AIAA-94-4365-CP,
A Collection of Technical Papers, 5th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary
Analysis and Optimization (J. Sobieski, ed.), Part 2, pp. 1041–5, 1994.
Adali, S., Richter, A., and Verijenko, V. E., Minimum weight design of symmetric angle-ply laminates
under multiple uncertain loads, Structural Optimization, Vol. 9, 89–95, 1995.
Adali, S., Richter, A., and Verijenko, V. E., Minimum weight design of symmetric angle-ply laminates
with incomplete information on initial imperfections, Journal of Applied Mechanics, Vol. 64, 90–7,
1997.
Almroth, B. O., Influence of edge conditions on the stability of axially compressed cylindrical shells,
AIAA Journal, Vol. 4, 134–40, 1966.
Almroth, B. O., and Brogan, F. A., The STAGS computer code, NASA CR-2950, 1978.
Almroth, B. O., and Rankin, C. R., Imperfection sensitivity of cylindrical shells, Advances in Engi-
neering Mechanics, 1071–4, 1983.
Almorth, B. O., and Starnes, J. H., Jr., The computer in shell stability analysis, Journal of the Engi-
neering Mechanics Division, Vol. 101, 873–88, 1974.
Almroth, B. O., Brogan, F. A., Miller E., Zelle, F., and Peterson, M. T., Collapse analysis or shells
of general shape, User’s Manual for the STAGS-A Computer Code, Air Force Flight Dynamics
Laboratory, Wright-Patterson AFB, AFFDL-TR-71-8, 1973.
Amazigo, J. C., Buckling under axial compression of long cylindrical shells with random axisym-
metric imperfections, Quarterly of Applied Mathematics, Vol. 26, 537–66, 1969.
Amazigo, J. C., Buckling of stochastically imperfect columns on non-linear elastic foundations,
Quarterly of Applied Mathematics, Vol. 28, 403–9, 1971.
291
292 BIBLIOGRAPHY
Amazigo, J. C., Asymptotic analysis of the buckling of externally pressurized cylinders with random
imperfections, Quarterly of Applied Mathematics, Vol. 31, 429–42, 1972.
Amazigo, J. C., Dynamic buckling of structures with random imperfections, Stochastic Problems in
Mechanics (H. H. E. Leipholz, ed.), University of Waterloo Press, Waterloo, pp. 234–54, 1974.
Amazigo, J. C., Buckling of stochastically imperfect structures, Buckling of Structures (B. Budiansky,
ed.), Springer Verlag, Berlin, pp. 172–82, 1976.
Amazigo, J. C., and Budiansky, B., Asymptotic formulas for the buckling stresses of axially com-
pressed cylinders with localized or random axisymmetric imperfections, Journal of Applied Me-
chanics, Vol. 39, 179–84, 1972.
Amazigo, J. C., Budiansky, B., and Carrier, G. F., Asymptotic analysis of the buckling of imperfect
columns on nonlinear elastic foundation, International Journal of Solids and Structures, Vol. 6,
1341–56, 1970.
Ambartsumyan, S. A., Theory of Anistropic Shells, Technomic Publishers, Lancaster, 1970.
Ambartsumyan, S. A., General Theory of Anisotropic Shells, “Nauka” Publishing House, Moscow,
1974 (in Russian).
Ambartsumyan, S. A., Some problems of the theory of shells from composite materials, Uspekhi
Mekhaniki, Vol. 6, Nos. 3–4, 1983 (in Russian).
Ambartsumyan, S. A., Theory of Anisotropic Plates: Strength, Stability and Vibration, Hemisphere
Publishing, New York, 1991.
Anderson, P. W., Absence of diffusion in certain random lattices, Physical Review, Vol. 109, 1492–
1505, 1958.
Aoki, T., and Fukumoto, Y., On scatter in buckling strength of steel columns – Effect of residual stress
distributions, Proceedings of JSCE, No. 201, 31–41, 1972 (in Japanese).
Arbocz, J., The effect of initial imperfections in shell stability, Thin Shell Structures: Theory, Ex-
periment and Design (Y. C. Fung and E. E. Sechler, eds.), Prentice-Hall, Englewood Cliffs, NJ,
pp. 205–45, 1974.
Arbocz, J., Present and future of shell stability analysis, Zeitschrift für Flugwissenschaften und Wel-
traumforschung, Vol. 5, 335–48, 1981.
Arbocz, J., The imperfection data bank: A means to obtain realistic buckling loads, in Buckling of
Shells (E. Ramm, ed.), Springer Verlag, Berlin, pp. 535–67, 1982a.
Arbocz, J., Shell stability analysis: theory and practice, Collapse: Buckling of Structures in Theory
and Practice (J. M. T. Thompson and G. W. Hunt, eds.), Cambridge University Press, Cambridge,
pp. 43–74, 1982b.
Arbocz, J., About the state of the art of shell design, Memorandum M-629, Faculty of Aerospace
Engineering, Delft University of Technology, Delft, the Netherlands, 1990.
Arbocz J., Towards an improved design procedure for buckling critical structures, Buckling of Shell
Structures on Land, in the Sea and in the Air (J. F. Jullien, ed.), Elsevier Applied Science, London,
pp. 270–6, 1991.
Arbocz, J., Future directions and challenges in shell stability, 38th AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference and Exhibit, AIAA paper 97-1077,
pp. 1949–62, 1997.
Arbocz, J., and Abramovich, H., The initial imperfection data bank at the Delft University of Technol-
ogy, Part 1, Report LR-290, Delft University of Technology, Department of Aerospace Engineering,
Delft, the Netherlands, 1979.
Arbocz, J., and Babcock, Jr., C. D., Experimental investigation of the effect of general imperfections
on the buckling of cylindrical shells, NASA CR-1163, 1968.
Arbocz, J., and Babcock, Jr., C. D., A multi-mode analysis for calculating buckling loads of im-
perfect cylindrical shells, GALCIT Report SM-74-4, California Institute of Technology, Pasadena,
1974.
Arbocz, J., and Babcock, Jr., C. D., Prediction of buckling loads based on experimentally measured
initial imperfections, Buckling of Structures (B. Budiansky, ed.), Springer Verlag, Berlin, 1976.
Arbocz, J., and Babcock, Jr., C. D., Utilization of STAGS to determine knockdown factors from
measured initial imperfections, Report LR-275, Delft University of Technology, Department of
Aerospace Engineering, Delft, the Netherlands, 1978.
BIBLIOGRAPHY 293
Arbocz, J., and Babcock, Jr., C. D., Computerized stability analysis using measured initial imper-
fections, 12th Congress of International Council of Aeronautical Science, Munich, pp. 688–701,
1980a.
Arbocz, J., and Babcock, Jr., C. D., The buckling analysis of imperfection sensitive structures, NASA
CR-3310, 1980b.
Arbocz, J., and Hol, J. M. A. M., ANILISA – Computational module for Koiter’s imperfection theory,
Report LR-582, Delft University of Technology, Faculty of Aerospace Engineering, Delft, the
Netherlands, 1989.
Arbocz, J., and Hol, J. M. A. M., Koiter’s stability theory in a computer aided engineering environment,
International Journal of Solids and Structures, Vol. 26, 945–73, 1990a.
Arbocz, J., and Hol, J. M. A. M., Recent developments in shell stability analysis, Report LR-633, Delft
University of Technology, Faculty of Aerospace Engineering, Delft, the Netherlands, 1990b.
Arbocz, J., and Hol, J. M. A. M., Collapse of axially compressed cylindrical shells with random
imperfections, AIAA Journal, Vol. 29, 2247–56, 1991 (also reprinted in Thin-Walled Structures,
Vol. 23, 131–58, 1995).
Arbocz, J., and Singer, J., Professor Bernard Budiansky’s contributions to buckling and postbuck-
ling of shell structures, AIAA Paper 2000-1322, 41st AIAA/ASME/ASCE/AHS/ASC Structures,
Structural Dynamics and Materials Conference and Exhibit, 3–6 April 2000, Atlanta, Ga.
Arbocz, J., and Williams, J. G., Imperfection surveys on a 10 ft. diameter shell structure, AIAA Journal,
Vol. 15, 949–56, 1976.
Arbocz, J., Poiter-Ferry, M., Singer, J., and Tvergaard, V., Buckling and Postbuckling, Springer Verlag,
Berlin, 1987.
Arbocz, J., and Singer, J., Professor Bernard Budiansky’s contributions to buckling and postbuckling of
shell structures, (preprint), 41st AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics,
and Materials Conference and Exhibit, Atlanta, GA, 2000.
Arbocz, J., Starnes, J., and Nemeth, M., A comparison of probabilistic and lower bound methods for
predicting the response of buckling-sensitive structures, Paper No. AIAA-2000-1382, AIAA Press,
Washington, D. C., 2000.
Ari-Gur, J., and Elishakoff, I., Influence of the shear deformation on the buckling of a structure with
overhang, Journal of Sound and Vibration, Vol. 139, 165–9, 1990.
Ariaratnam, S. T., and Xie, W. C., Wave localization in randomly disordered nearly periodic long
continuous beams, Journal of Sound and Vibration, Vol. 181, 7–22, 1995.
Ariaratnam, S. T., and Xie, W. C., Buckling node localization in randomly disordered continuous
beams using a simplified model, Chaos, Solitons and Fractals, Vol. 7, 1127–44, 1996.
Arnold, L., Crauel, H., and Eckmann, J. P. (eds.), Lyapunov Exponents, Springer Verlag, Berlin, 1991.
Aston, P. J., Analysis and computation of symmetry-breaking bifurcation and scaling laws using group
theoretic methods, SIAM Journal of Mathematical Analysis, Vol. 22, 181–212, 1991.
Augusti, G., Probabilistic treatments of column buckling problems, Stochastic Problems in Mechanics,
University of Waterloo Press, Waterloo, pp. 255–74, 1974.
Augusti, G., and Baratta, A., Teoria probabilistica della resistenza delle aste compresse, Construzioni
Metalliche, No. 1, 1–15, 1971 (in Italian).
Augusti, G., and Baratta, A., Limit analysis of structures with stochastic strength variations, Journal
of Structural Mechanics, Vol. 1, 43–62, 1972a.
Augusti, G., and Baratta, A., Probabilistic theory of slender columns, Proceedings of 13th Polish
Solid Mechanics Conference, Warsaw, 1972b.
Augusti, G., and Baratta, A., Reliability of slender columns: Comparison of different approximations,
Buckling of Structures (B. Budiansky, ed.), Springer Verlag, Berlin, pp. 183–98, 1976.
Augusti, G., Baratta, A., and Casciati, F., Probabilistic Methods in Structural Engineering, Chapman
and Hall, London, 1984.
Ayyub, B. M., and Chia, C. Y., Generalized conditional expectation for structural reliability assess-
ment, Structural Safety, Vol. 11, 131–46, 1992.
Ayyub, B. M., and Haldar, A., Improved simulation techniques as structural reliability models,
Proceedings, International Conference on Structural Safety and Relibility, Kobe, Japan, pp.
I. 17–I. 26, 1985.
294 BIBLIOGRAPHY
Ayyub, B. M., and Lai, K. L., Selective sampling in simulation-based reliability assessment, Interna-
tional Journal of Pressure Vessels and Piping, Vol. 46, 229–49, 1991.
Ayyub, B. M., and McCuen, R. H., Simulation-based reliability methods, Probabilistic Structural
Mechanics Handbook (C. R. Sundararajan, ed.), Chapman and Hall, New York, pp. 53–69, 1995.
Babcock, Jr., C. D., Shell stability, Journal of Applied Mechanics, Vol. 50, 935–40, 1983.
Babcock, Jr., C. D., and Sechler, E. E., The effect of initial imperfections on the buckling stress of
cylindrical shells, NASA TN D-1510, pp. 135–42, 1962.
Badala, A. Marinetti, A., and Obiveto, G., The collapse load of pin-jointed networks of randomly
imperfect rods, Engineering Structures, Vol. 2, 9–14, 1980.
Bahder, Th. B., Mathematica for Scientists and Engineers, Addison-Wesley, Reading, MA, 1995.
Baker, E. H., Cappelli, A. P., Kovalevsky, L., Rish, F. L., and Verette, R. M., Shell analysis manual,
NASA CR-512, 1968.
Ball, J. M., Extensible beam stability theory, Journal of Differential Equations, Vol. 14, 399–418, 1974.
Bažant, Z. P., and Cedolin, L., Stability of Structures, Oxford University Press, New York, 1991.
Bažant, Z., Structural stability, International Journal of Solids and Structures, Vol. 37, 55–67, 2000.
Bellini, P. X., and Chulya, A., An improved automatic incremental algorithm for the efficient solution
of non-linear finite element equations, Computers and Structures, Vol. 26, 99–110, 1987.
Beltzer, A. I., Engineering analysis via symbolic computation - A breakthrough, Applied Mechanics
Reviews, Vol. 43, 119–23, 1990a.
Beltzer, A. I., Variational and Finite Element Methods – A Symbolic Computation Approach, Springer
Verlag, Berlin, 1990b.
Benaroya, H., and Rehak, M., Finite element methods in probabilistic structural analysis: A selective
review, Applied Mechanics Reviews, Vol. 41, 201–13, 1988.
Benedetti, A., On the reliability analysis of elastic-plastic struts with initial imperfections, IASS
Symposium – 91, Vol. III, pp. 123–30, 1991.
Benedetti, A., On the probabilistic stability analysis of tubular struts with initial imperfections, Struc-
tural Safety and Reliability (G. I. Schuëller, M. Shinozuka, and J. T. P. Yao, eds.), Balkema Publishers,
Rotterdam, Vol. 2, pp. 745–52, 1994.
Ben-Haim, Y., The Assay of Spatially Random Material, Kluwer, Dortrecht, 1985.
Ben-Haim, Y., Convex models for uncertainty in radial pulse buckling of shells, Journal of Applied
Mechanics, Vol. 60, 683–8, 1993.
Ben-Haim, Y., Convex models of uncertainty: Applications and implications, Erkenntnis, Vol. 41,
139–56, 1994.
Ben-Haim, Y., On convex models of uncertainty for small initial imperfections, Zeitschrift für ange-
wandte Mathematik und Mechanik, Vol. 75, 901–8, 1995.
Ben-Haim, Y., Robust Reliability in the Mechanical Sciences, Springer-Verlag, Berlin, 1996.
Ben-Haim, Y., and Elishakoff, I., Non-probabilistic models of uncertainty in the nonlinear buckling
of shells with general imperfections: Theoretical estimates of the knockdown factor, Journal of
Applied Mechanics, Vol. 56, 403–10, 1989a.
Ben-Haim, Y., and Elishakoff, I., Dynamics and failure of structures based on the unknown-but-
bounded imperfection model, Recent Advances in Impact of Engineering Structures (D. Hui and
N. Jones, eds.), AMD – Vol. 105, AD – Vol. 17, pp. 89–95, 1989b.
Ben-Haim, Y., and Elishakoff, I., Convex Models of Uncertainty in Applied Mechanics, Elsevier
Science Publishers, Amsterdam, 1990.
Bernard, M. C., and Bogdanoff, J. L., Buckling of columns with random initial displacements, Journal
of Engineering Mechanics, Vol. 97, 755–71, 1971.
Bernard, P., and Fogli, M., Utilization des méthodes de Monte-Carlo en sécurité structurale, Technical
Report, Université de Clermont II, Laboratoire de Génie Civil, 1984 (in French).
Bert, C. W., Analysis of shells, Composite Materials, Vol. 7, Structural Design and Analysis, Part 1
(C. C. Chamis, ed.), Academic Press, New York, pp. 205–58, 1975.
Bert, C. W., Applications of the Rayleigh technique to problems of bars, beams, columns, membranes,
and plates, Journal of Sound and Vibration, Vol. 119, 317–26, 1987.
Bert, C. W., Techniques for estimating buckling loads, Handbook of Civil Engineering (P. N.
Cheremisinoff, ed.), Technomic Publishers, Lancaster, 1987.
BIBLIOGRAPHY 295
Beveridge, G., and Schechter, R., Optimization: Theory and Practice, McGraw Hill, New York,
pp. 453–6, 1970.
Biderman, V. L., Application of the double-sweep method to problems in structural mechanics,
Mekhanika Tverdogo Tela, Vol. 5, 62–6, 1967 (in Russian).
Bielewicz, E., Gorski, J., Schmidt, R., and Walukiewicz, H., Random fields in the limit analysis of
elastic-plastic shell structures, Computers and Structures, Vol. 51, 267–75, 1994.
Birkemoe, P. C., Stability: Directions in experimental research, Engineering Structures, Vol. 18, 807–
11, 1996.
Bjaerum, D., Finite element formulation and solutions algorithms for buckling and collapse analysis
of thin shells, PhD Thesis, Division of Structural Engineering, University of Trondheim, 1992.
Bjorhovde, R., Deterministic and probabilistic approaches to the strength of steel columns, PhD
Dissertation, Department of Civil Engineering, Lehigh University, 1972.
Bljuger, F., Probabilistic analysis of reinforced concrete columns and walls in buckling, ACI Structural
Journal, 124–31, 1987.
Bleich, F., Buckling Strength of Metal Structures, McGraw-Hill, New York, 1952.
Blekhman, I. I., Myshkis, A. D., and Panovko, Ya. G., Mechanics and Applied Mathematics: Logics
and Specifics of Applications of Mathematics, “Nauka” Publishers, Moscow, pp. 205–9, 1983
(in Russian).
Bogdanovich, A. E., Nonlinear Problems of Dynamics of Cylindrical Composite Shells, “Zinatne”
Publishers, Riga, 1987 (in Russian).
Bogdanovich, A. E., and Yushanov, S. P., Analysis of buckling of cylindrical shells with a random
field of initial imprefections under axial dynamic compression, Mechanics of Composite Materials,
Vol. 17, 552–60, 1981.
Bogdanovich, A. E., and Yushanov, S. P., About reliability analysis of anisotropic shells via the
probability of rare out-crossing of vector random field through the limit surface, Mechanics of
Composite Materials, No. 1, 80–9, 1983a (in Russian).
Bogdanovich, A. E., and Yushanov, S. P., Computing the reliability of anisotropic shells from the
probability of infrequent excursions of a random vector field beyond the limiting surface, Mechanics
of Composite Materials, Vol. 19, 67–75, 1983b (in Russian).
Bolotin, V. V., Statistical methods in the nonlinear theory of elastic shells, Izvestija Academii
Nauk SSSR, Otdelenie Tekhnicheskikh Nauk, No. 3, 1958 (English translations, NASA TTF-85,
1–16, 1962).
Bolotin, V. V., Statistical Methods in Structural Mechanics, Grosstroiizdat, Moscow, 1961 (in Russian).
Bolotin, V. V., Non-Conservative Problems of Theory of Elastic Stability, Pergaman Press, London,
1964.
Bolotin, V. V., Application of methods of the theory of probability in the theory of plates and shells,
Theory of Shells and Plates (S. M. Durgaryan, ed.), Israel Program for Scientific Translation,
Jerusalem, 1966a.
Bolotin, V. V., Theory of layered medium with random initial imperfections, Mechanics of Polymers,
Vol. 1, 11–19, 1966b (in Russian).
Bolotin, V. V., Statistical aspects in the theory of structural stability, Dynamic Stability of Structures
(G. Herrmann, ed.), Pergamon Press, New York, pp. 67–81, 1967.
Bolotin, V. V., Statistical Methods in Structural Mechanics, Holden-Day, San Francisco, 1969.
Bolotin, V. V., Stochastic edge-effects in subcritical deformations of elastic shells, Mechanics of
Solids, Vol. 5, 82–6, 1970.
Bolotin, V. V., Reliability theory and stochastic stability, Stability (H. H. E. Leipholz, ed.), University
of Waterloo Press, Waterloo, pp. 385–442, 1972.
Bolotin, V. V., and Makarov, B. P., On correlation of displacements in thin elastic shells with random
initial imperfections, Reliability Problems in Structural Mechanics (V. V. Bolotin and A. Cyras,
eds.), “RINTIP” Publishers, Vilnius, pp. 163–8, 1968a (in Russian).
Bolotin, V. V., and Makarov, B. P., Correlation theory of pre-critical deformations in thin elastic shells,
Prikladnaya Matematika i Mekhanika, Vol. 32, 428–34, 1968b (in Russian).
Bolotin, V. V., and Novichkov, Yu. N., Mechanics of Multilayered Structures, “Mashinostroenie”
Publishers, Moscow, 1981 (in Russian).
296 BIBLIOGRAPHY
Bonello, M., Reliability assessment and design of stiffened compression flanges, PhD Thesis, De-
partment of Civil Engineering, Imperial College, University of London, 1992.
Bonello, M. A., Chrystanthopoulos, M. K., and Dowling, P. J., Probabilistic strength modeling of
unstiffened plates under axial compression, Proceedings of the 10th Offshore Mechanics and Arctic
Engineering Symposium, ASME, Norway, 1991.
Booton, M., Buckling of imperfect anisotropic cylinders under combined loading, Report No. 203,
Institute for Aerospace Studies, University of Toronto, 1976.
Borst, R. de, Kusters, G., Nauta, P., and Witte, F. de, DIANA, a complex but flexible finite element
system, Finite Element Systems (C. A. Brebbia, ed.), 3rd ed., Springer Verlag, Berlin, 1983.
Boyce, W. E., Buckling of a column with random initial displacements, Journal of Aerospace Sciences,
Vol. 28, 308–20, 1961.
Brabré, R., Stabilität gleichmässig gedrückter Rechteckplatten mit Längs- oder Quersteifen, Ingenieur
Archiv, Vol. 8, 117–50, 1937 (in German).
Brasil, R. M., and Hawwa, M. A., The localization of buckling modes in nearly periodic trusses,
Computers and Structures, Vol. 56, 927–32, 1995.
Brenner, C. E., Stochastic finite element method, Report No. 35-91, Institute of Engineering Mechan-
ics, University of Innsbruck, Austria, 1991.
Brillouin, L., Wave Propagation in Periodic Structures, Dover, New York, 1953.
Britvec, S. J., Stability and Optimization of Flexible Space Structures, Birkhauser, Boston, 1995.
Brown, W. S., ALTRAN Users Manual, Bell Laboratories, Murray Hill, NJ, 1973.
Brush, D. D., and Almroth, B. O., Buckling of Bars, Plates and Shells, McGraw-Hill, New York,
1975.
Budiansky, B., Theory of buckling and postbuckling behavior of elastic structures, Advances in
Applied Mechanics (C. Yih, ed.), Academic Press, New York, pp. 1–65, 1974.
Budiansky, B., and Hutchinson, J. W., Dynamic buckling of imperfection-sensitivity structures, Pro-
ceedings of the Eleventh International Congress of Applied Mechanics (H. Goertler, ed.), Munich,
pp. 636–51, 1964.
Budiansky, B., and Hutchinson, J. W., A survey of some buckling problems, AIAA Journal, Vol. 4,
1505–10, 1966.
B̀udiansky, B., and Hutchinson, J. W., Buckling of circular cylindrical shells under axial compression,
Contributions to the Theory of Aircraft Structures, The Van Der Neut Anniversary Volume, Delft
University Press, Delft, pp. 239–59, 1972.
Budiansky, B., and Hutchinson, J. W., Buckling: Progress and challenge, Trends in Solid Mechanics
(J. F. Besseling and A. M. A. Van der Heijden, eds.), Proceedings of the Symposium Dedicated to
the 65th Birthday of W. T. Koiter, Sijthoff and Noordhoff, Delft, pp. 93–116, 1979.
Budiansky, B., and Sanders, Y. L., Jr., On the “best” first order linear theory, Progress in Applied
Mechanics, Macmillan, New York, pp. 129–40, 1963.
Bulgakov, B. V., Fehleranhaeufung bei Kreiselapparaten, Ingenieur-Archiv, Vol. 11, 461–9, 1940
(in German).
Bulgakov, B. V., On the accumulation of disturbances in linear systems with constant coefficients,
Doklady Akademii Nauk SSR, Vol. LI, No. 5, 339–42, 1946 (in Russian).
Bulson, P. S., The Stability of Flat Plates, American Elsevier, New York, 1969.
Bushnell, D., Stress, stability, and vibration of complex branched shells of revolution: Analysis and
user’s manual for BOSOR 4, NASA CR-2116, 1972.
Bushnell, D., BOSOR-54 – Program for buckling of elastic-plastic complex shells of revolution
including large deflections and creep, Computers and Structures, Vol. 6, 221–39, 1976.
Bushnell, D., Buckling of shells – Pitfall for designers, AIAA Journal, Vol. 11, 1183–226, 1981.
Bushnell, D., PANDA – Interactive program for minimum weight design of stiffened cylindrical panels
and shells, Computers and Structures, Vol. 16, 167–85, 1983.
Bushnell, D., Computerized Buckling Analysis of Shells, Martinus Nijhoff Publishers, Dortrecht,
1985.
Cai, G. Q., and Lin, Y. K., Localization of wave propagation in disordered periodic structures, AIAA
Journal, Vol. 29, 450–6, 1991.
BIBLIOGRAPHY 297
Calladine, C. R., Theory of Shell Structures, Cambridge University Press, London, 1983.
Cambou, B., Application of first-order uncertainty analysis in the finite element method in linear
elasticity, Proceedings of the Second International Conference on Applied Statistics and Probability
in Soil and Structural Engineering, Aachen, pp. 67–87, 1975.
Capsoni, A., and Poggi, C., The role of symbolic algebra in the initial post-buckling and imper-
fection sensitivity analysis of axially compressed composite cylindrical shells, Technical Report,
Politecnico di Milano, Dip. di Ingegneria Strutturale, Milan, Italy, 1990.
Card, M. F., The sensitivity of buckling of axially compressed fiber-reinforced cylindrical shells to
small geometric imperfections, PhD Dissertation, Department of Engineering Science and Me-
chanics, Virginia Polytechnic Institute, 1969.
Carstensen, C., and Wagner, W., Detecting symmetry-breaking bifurcation points of symmetric struc-
tures, International Journal for Numerical Methods in Engineering, Vol. 36, 3019–39, 1993.
Čausević, M., Prikaz uzroka opadanju vrijednosti kriticnoy opterecenja za neke staticke sisteme,
Tehnika, Vol. 36, No. 7–8, 1087–91, 1981 (in Serbian).
Cederbaum, G., and Arbocz, J., Reliability of axially compressed cylindrical shells, Delft University
of Technology, Report LR-767, Faculty of Aerospace Engineering, Delft, the Netherlands,
1994.
Cederbaum, G., and Arbocz, J., Reliability of shells via Koiter formulation, Thin-Walled Structures,
Vol. 24, 173–87, 1996a.
Cederbaum, G., and Arbocz, J., On the reliability of imperfection-sensitive long isotropic cylindrical
shells, Structural Safety, Vol. 18, 1–9, 1996b.
Cederbaum, G., and Arbocz, J., Reliability of imperfection-sensitive composite shells via the Koiter
Cohen criterion, Reliability Engineering and System Safety, Vol. 56, 257–63, 1997.
Cederbaum, G., Aboudi, J., and Elishakoff, I., Dynamic stability of viscoelastic laminated plates via
the Lyapunov exponents, International Journal of Solids and Structures, Vol. 28, 317–27, 1991.
Champneys, A. R., van der Heijden, G. H. M., and Thompson, J. M. T., Spatially complex localization
after one-twist-per-wave equilibria in twisted circular rods with initial curvature, Philosophical
Transactions of the Royal Society of London, Series A – Mathematical, Physical and Engineering
Science, Vol. 355, 2151–74, 1997.
Cheriachukin, V. V., Application of the Monte Carlo method to some statistical problems of stability
and reliability, Reliability Problems in Structural Mechanics (V. V. Bolotin and A. Cyras, eds.),
Vilnius, Rintip, pp. 79–85, 1968 (in Russian).
Chernousko, F. L., Optimal guaranteed estimates of indeterminacies with the aid of ellipsoids,
Tekhnicheskaya Kibernetika, pp. 1–9, 1981 (in Russian).
Chernousko, F. L., State Estimation for Dynamic Systems, CRC Press, Boca Raton, 1994.
Chetaev, N. G., The Stability of Motion, Pergamon Press, New York, 1961.
Cho, H.-N., Lee, S.-J., Choi, Y.-M., and Shin, J.-C., Buckling reliability of laminated cylindrical shells,
Structural Safety and Reliability (G. I. Schuëller, M. Shinozuka, and Y. T. P. Yao, eds.), Balkema
Publishers, Rotterdam, Vol. 2, pp. 753–60, 1994.
Choi, S.-T., and Liao, Y. L., Nonlinear forced vibration of rectangular plates with random geometric
imperfections, Structural Safety and Reliability, (G. I. Schuëller, M. Shinozuka, and J. T. P. Yao,
eds.), Balkema, Rotterdam, pp. 101–8, 1994.
Chryssanthopoulos, M. K., Probabilistic buckling analysis of plates and shells, Thin-Walled Structures,
Vol. 30, 135–57, 1997.
Chryssanthopoulos, M. K., and Poggi, C., Stochastic imperfection modelling in shell buckling studies,
Thin-Walled Structures, Vol. 23, 179–200, 1995a.
Chryssanthopoulos, M. K., and Poggi, C., Probabilistic imperfection sensitivity analysis of axially
compressed composite cylinders, Engineering Structures, Vol. 17, 398–406, 1995b.
Chryssanthopoulos, M. K., Baker, M. J., and Dowling, P. J., Statistical analysis of imperfections in
stiffened cylinders, Journal of Structural Engineering, Vol. 117, 1979–97, 1991a.
Chryssanthopoulos, M. K., Baker, M. J., and Dowling, P. J., Imperfection modelling for buck-
ling analysis of stiffened cylinders, Journal of Structural Engineering, Vol. 117, 1998–2017,
1991b.
298 BIBLIOGRAPHY
Chryssanthopoulos, M. K., Giavotto, V., and Poggi, C., Statistical imperfection models for buckling
analysis of composite shells, Buckling of Shell Structures on Land, in the Sea and in the Air (J. F.
Jullien, ed.), Elsevier Applied Science, London, pp. 43–52, 1991.
Chryssanthopoulos, M. K., Giavotto, V., and Poggi, C., Characterization of manufacturing effects for
buckling-sensitive composite cylinders, Composites Manufacturing, Vol. 6, 93–101, 1995.
Chung, B. T., Random-parameter analysis of stability of inelastic structures, PhD Thesis, State
University of New York, 1962.
Chung, B. T., and Lee, G. T., Buckling strength of column based on random parameters, Journal of
the Structure Division, Vol. 97, 1927–45, 1971.
Ciavarella, M., Modelli probabilistici di imperfezioni geometriche per lo studio dell instabilita di
gusci cilindrici in materiale composito, Diploma Thesis, Department of Aerospace Engineering,
Politecnico di Milano, 1991 (in Italian).
Como, M., and Grimaldi, A., Stability, buckling and post-buckling of elastic structures, Meccanica,
Vol. 4, 254–68, 1975.
Combescure, A., Gusik, G., and Jullien, J. F., Influence of thickness imperfections on the buckling of
cylinders under external pressure. Fourth Int’l Coll. on Computation of Shell & Spatial Structures,
Chania-Crete, Greece, pp. 62–63, June 4–7, 2000.
Como, M., and Grimaldi, A., Theory of Stability of Continuous Elastic Structures, CRC Press, Boca
Raton, 1995.
Conte, S. D., The numerical solution of linear boundary value problems, SIAM Review, Vol. 8, 309–21,
1966.
Cornell, C. A., Bounds on the reliability of structural systems, Journal of Structural Division, Vol. 93,
171–200, 1967.
Cornell, C. A., Probability based structural code, ACI Journal, 66, 974–85, 1969.
Crandall, S. H., Engineering Analysis, A Review of Numerical Analysis Procedures, McGraw-Hill,
New York, pp. 297–8, 1956.
Crandall, S. H., Personal communication to I. E., 1990.
Crespo da Silva, M. R. M., and Hodges, D. H., Dynamic modeling of rotating beams undergoing
flexure in two directions and torsion, with application to helicopter rotor blades, Computers and
Mathematics, 1986.
Crisfield, M. A., A fast incremental/iterative procedure that handles “snap-through,” Computers and
Structures, Vol. 13, 387–99, 1979.
Crisfield, M. A., An arc-length method including line searches and accelerations, International Journal
of Numerical Methods in Engineering, Vol. 19, 1269–89, 1983.
Damil, N., and Potier-Ferry, M., A new method to compute perturbed bifurcations: Application to
the buckling of imperfect elastic structures, International Journal of Engineering Science, Vol. 28,
943–57, 1990.
Das, P. K., Frieze, P. A., and Faulkner, D., Structural reliability modelling of stiffened components of
floating structures, Structural Safety, Vol. 12, 1984.
Day, W. B., Buckling of a column with nonlinear restraints and random initial displacement, Journal
of Applied Mechanics, Vol. 47, 204–5, 1980.
Davenport, J., Siret, N., and Tournier, E., Computer Algebra, Academic Press, London, 1988.
Davidenko, D. F., On one new method of numerical solution of the systems of non-linear equations,
Doklady Akademii Nauk SSSR, Vol. 88, pp. 601–2, 1953 (in Russian).
Davister, M., and Britvec, S., Evaluation of the most unstable dynamic buckling mode of pin jointed
space lattices, Third International Conference on Space Structures, pp. 468–73, 1984.
Davletkhanova, A. D., Stability of multi-part column and discrete supports with random charateristics,
Stroitelnayia Mekhanika i Raschet Sooruzhenii, No. 6, 48–53, 1980.
De Jong, R. G., Localization and transmission loss in ribbed plates, Proceedings of the NOISE –
CON ’94, Progress in Noise Control for Industry (J. M. Cuschieri, S. A. L. Glegg, and D. M.
Yeager, eds.), Fort Lauderdale, pp. 603–8, 1994.
Dellnitz, M., and Werner, B., Computational methods for bifurcation problems with symmetries, with
special attention to steady state and Hopf bifurcation points, Journal of Computer Applications in
Mathematics, Vol. 26, 97–123, 1989.
BIBLIOGRAPHY 299
Deodatis, G., and Shinozuka, M., Bounds on response variability of stochastic systems, Journal of
Engineering Mechanics, Vol. 115, 2543–63, 1989.
Der Kiureghian, A., and Zhang, J., Space-variant finite element reliability analysis, Computer Methods
in Applied Mechanics and Engineering, Vol. 168, 173–83, 1999.
Dimentberg, M. F., Some problems of stability of shells subjected to random excitations, Problems of
Stability in Structural Mechanics, “Stroiizdat” Publishers, Moscow, pp. 217–22, 1965 (in Russian).
Diniz, S. M. C., Reliability evaluation of high strength concrete columns, PhD Thesis, Department
of Civil Engineering, University of Colorado at Boulder, 1994.
Diniz, S. M. C., and Frangopol, D. M., Reliability basis for high-strength concrete columns, Journal
of Structural Engineering, Vol. 123, 1375–81, 1997.
Diniz, S. M. C., and Frangopol, D. M., Reliability of high-strength concrete columns, Materials for
the New Millennium (K. P. Chong, ed.), ASCE Press, New York, Vol. 1, pp. 213–22, 1996.
Dinnik, A. N., Selected Papers, Vol. 2, Academy of Ukrainian SSR Publishing House, Kiev, 1955
(in Russian).
Doedel, E., AVTO: Software for continuation and bifurcation problems in ordinary differential equa-
tions, California Institute of Technology, 1986.
Donnell, L. H., Beams, Plates and Shells, McGraw-Hill, New York, 1976.
Donnell, L. H., and Wan, C. C., Effect of imperfections on the buckling of thin cylinders and columns
under axial compression, Journal of Applied Mechanics, Vol. 17, 73–83, 1950.
Drenick, R. F., Functional analysis of effects of earthquakes, 2nd Joint United States-Japan Seminar
on Applied Stochastics, Washington, DC, Sept. 19–24, 1968.
Drenick, R. F., Model-free design of aseismic structures, Journal of Engineering Mechanics Division,
Vol. 96, 483–93, 1970.
Dugundji, J., Nonlinear problem of aeroelasticity, Computational Nonlinear Mechanics in Aerospace
Engineering (S. N. Atluri, ed.), AIAA Press, Washington, D.C., pp. 127–55, 1992.
Duprat, F., Pinglot, M., and Lorrain, M., Structural reliability analysis of reinforced concrete columns:
A comparison of methods to assess the probability of buckling failure, Materiaux et Constructions,
Materials and Structures, Vol. 29, 485–93, 1998 (in French).
Edlund, B. L. O., Computer simulation for structures, Publication S73-6, Chalmers University of
Technology, Göteborg, 1973.
Edlund, B. L. O., and Leopoldson, U. L. C., Monte Carlo simulation of the strength of steel structure:
Part 1; Method and basic data, Publication S71-3, Chalmers University of Technology, Department
of Structural Engineering, 1971a.
Edlund, B. L. O., and Leopoldson, U. L. C., Monte Carlo simulation of the strength of steel structure:
Part 2; Rolled I- and H-beam, statistical variation of the load carrying capacity, Publication S71-5,
Chalmers University of Technology, Department of Structural Engineering, 1971b.
Edlund, B. L. O., and Leopoldson, U. L. C., Simulation of strength of axially loaded cylindrical shells,
CTH Institute für Konstruktonsteknik, Stål-och Träbyggnad, Inst. Strifter S 73:7, Göteborg, 1974.
Edlund, B. L. O., and Leopoldson, U. L. C., Computer simulation of the scatter in steel member
strength, Computers and Structures, Vol. 5, 209–24, 1975.
Eggwertz, S., and Palmberg, B., Structural safety of axially loaded cylindrical shells, Technical Report
FFA-TN-1985-50, Aeronautical Research Institute of Sweden, Bromma, Sweden, 1985.
Elishakoff, I., Axial impact buckling of a column with random initial imperfection, Journal of Applied
Mechanics, Vol. 45, 361–5, 1978a.
Elishakoff, I., Impact buckling of thin bar via Monte Carlo method, Journal of Applied Mechanics,
Vol. 45, 586–90, 1978b.
Elishakoff, I., Buckling of a stochastically imperfect finite column on a nonlinear elastic foundation –
A reliability study, Journal of Applied Mechanics, Vol. 46, 411–16, 1979a.
Elishakoff, I., Simulation of space-random fields for solution of stochastic boundary-value problems,
Journal of Acoustical Society of America, Vol. 65, 399–403, 1979b.
Elishakoff, I., Remarks on the static and dynamic imperfection-sensitive of nonsymmetric structures,
Journal Applied Mechanics, Vol. 47, 111–15, 1980a.
Elishakoff, I., Hoff ’s problem in a probabilistic setting, Journal of Applied Mechanics, Vol. 47, 403–8,
1980b.
300 BIBLIOGRAPHY
Elishakoff, I., How to introduce initial-imperfection sensitivity concept into design, Collapse: The
Buckling of Structures of Theory and Practice (J. M. T. Thompson, and G. W. Hunt, eds.), pp. 345–
57, 1983a.
Elishakoff, I., How to introduce initial-imperfection sensitivity concept into design 2, M. Stein
Volume, NASA CP, 1998, 206280.
Elishakoff, I., Probabilistic Methods in the Theory of Structures, John Wiley, New York, pp. 433–68,
1983b (2nd ed., Dover, Mineola, 1999).
Elishakoff, I., Reliability approach to the initial imperfection sensitivity, Acta Mechanica, Vol. 55,
151–70, 1985.
Elishakoff, I., A remark on the adjustable parameter version of Rayleigh’s method, Journal of Sound
and Vibration, Vol. 118, 163–5, 1987.
Elishakoff, I., Simulation of an initial imperfection data bank, Part I: Isotropic shells with general
imperfections, TAE Report 500, Technion–Israel Institute of Technology, Department of Aero-
nautical Engineering, Haifa, 1982. Also in Buckling of Structures – Theory and Experiment
(I. Elishakoff, J. Arbocz, C. D. Babcock, Jr., and A. Libai, eds.), Elsevier Science Publishers,
Armsterdam, pp. 195–210, 1988.
Elishakoff, I., An idea of the uncertainty triangle, Shock and Vibration Digest, Vol. 22, 1, 1990.
Elishakoff, I., Convex versus probabilistic modelling of uncertainty in structural dynamics, Structural
Dynamics – Recent Advances (M. Petyt, H. F. Wolfe, and C. Mei, eds.), Elsevier, London, pp. 3–21,
1991.
Elishakoff, I., Essay on uncertainties in elastic and viscoelastic structures: From A. M. Freudenthal’s
criticisms to modern convex modeling, Computers and Structures, Vol. 56, 871–96, 1995.
Elishakoff, I., How to introduce the imperfection sensitivity concept into design 2, 237–267 NASA/CP-
1998-206280, 1998.
Elishakoff, I., Unusual Closed-form Solutions for Eigenvalues of Inhomogeneous Structures, 2000 (in
press).
Elishakoff, I., and Arbocz, J., Reliability of axially compressed cylindrical shells with random
axisymmetric imperfections, International Journal of Solids and Structures, Vol. 18, 563–85,
1982.
Elishakoff, I., and Arbocz, J., Reliability of axially compressed cylindrical shells with general non-
symmetric imperfections, Journal of Applied Mechanics, Vol. 52, 122–8, 1985.
Elishakoff, I., and Ben-Haim, Y., Dynamics of a thin cylindrical shell under impact with limited
deterministic information on initial imperfections, Journal of Structural Safety, Vol. 8, 103–12,
1990.
Elishakoff, I., and Charmats, M., Godunov-Conte method for solution of engineering problems and
its applications, Journal of Applied Mechanics, Vol. 44, 776–9, 1977.
Elishakoff, I., and Colombi, P., Combination of probabilistic and convex models of uncertainty when
scarce knowledge is present on acoustic excitation parameters, Computer Methods in Applied
Mechanics and Engineering, Vol. 104, 187–209, 1993.
Elishakoff, I., and Hasofer, A. M., On the accuracy of Hasofer-Lind reliability index, Structural Safety
and Reliability (I. Konishi, A. H-S. Ang, and M. Shinozuka, eds.), Proceedings of ICOSSAR’85,
the 4th International Conference on Structural Safety and Reliability, Kobe, Japan, Vol. 1, pp.
1. 229–1.239, 1985.
Elishakoff, I. and Hasofer, A. M., Effect of human error on reliability of structures, Proceedings, 33rd
AIAA/ASME/ASCE/AMS/ASC Structures, Structural Dynamics and Materials Conference, AIAA
Press, Washington, DC, pp. 3222–37, 1992.
Elishakoff, I., and Hettema, Ch. D., Nonstationary random vibration with REDUCE, Probabilistic
Methods in Civil Engineering (P. D. Spanos, ed.), ASCE, New York, pp. 523–32, 1988.
Elishakoff, I., Hettema, Ch. D., and Wilson, E. L., Direct superposition of Wilson trial functions by
computerized symbolic algebra, Acta Mechanica, Vol. 74, 69–79, 1989.
Elishakoff, I., and Pletner, B., Computerized symbolic algebraic evaluation of buckling loads by
Snitko’s method, Symbolic Computations and Their Impact on Mechanics (A. K. Noor, I. Elishakoff,
and G. Hubbert, eds.), ASME Press, New York, pp. 175–88, 1990.
BIBLIOGRAPHY 301
Elishakoff, I., and Ren, I. J., The bird’s eye view on finite element method for stochastic structures,
Computer Methods in Applied Mechanics and Engineering, Vol. 168, 51–61, 1999.
Elishakoff, I., and Rollot, New Closed-form solutions for buckling of variable stiffness column by
Mathematica, Journal of Sound and Vibration, Vol. 224, 172–182, 1999.
Elishakoff, I., and Tang, J., Buckling of polar orthotropic circular plates on elastic foundation by com-
puterized symbolic algebra, Computer Methods in Applied Mechanics and Engineering, Vol. 68,
229–47, 1988.
Elishakoff, I., Arbocz, J., Babcock, Jr., C. D., and Libai, A. (eds.), Buckling of Structures – Theory
and Experiment, Elsevier Science Publishers, Amsterdam, 1988.
Elishakoff, I., Cai, G. Q., and Starnes, J. H., Jr., Probabilistic and convex models of uncertainty in
buckling of structures, Structural Safety and Reliability, (G. I. Schuëller, M. Shinozuka, and J. T. P.
Yao, eds.), Balkema Publishers, Rotterdam, Vol. 2, pp. 761–6, 1994a.
Elishakoff, I., Cai, G. Q., and Starnes, J. H., Jr., Nonlinear buckling of a column with initial imperfection
via stochastic and non-stochastic, convex models, International Journal of Nonlinear Mechanics,
Vol. 29, 71–82, 1994b.
Elishakoff, I., Gana Shvili, Y., and Givoli, D., Treatment of uncertain imperfections as a convex opti-
mization problems, Proceedings of the Sixth International Conference on Applications of Stochas-
tics and Probability in Civil Engineering (L. Esteva and S. E. Ruiz, eds.), Mexico, Vol. 1, pp. 150–7,
1991.
Elishakoff, I., Li, Y. W., and Starnes, Jr., J. H., The combined effect of the thickness variation and
axisymmetric initial imperfection on the buckling of the isotropic cylindrical shell under axial com-
pression, Preliminary Report, Florida Atlantic University, Department of Mechanical Engineering,
Boca Raton, May 1992.
Elishakoff, I., Li, Y. W., and Starnes, Jr., J. H., A deterministic method to predict the effect of unknown-
but-bounded uncertain elastic moduli on the buckling of composite structures, Computer Methods
in Applied Mechanics and Engineering, Vol. 111, pp. 157–67, 1993.
Elishakoff, I., Li, Y. W., and Starnes, Jr., J. H., Buckling mode localization in elastic plates due to
misplacement in the stiffener location, Chaos, Solitons and Fractals, Vol. 5, 1517–31, 1995.
Elishakoff I., Li, Y. W., and Starnes, J. H., Jr., Imperfection sensitivity due to the elastic moduli in the
Roorda-Koiter frame, Chaos, Solitons and Fractals, Vol. 7, 1179–86, 1996.
Elishakoff, I., Lin, Y. K., and Zhu, L. P., Probabilistic and Convex Modeling of Acoustically Excited
Structures, Elsevier Science Publishers, Amsterdam, 1994.
Elishakoff, I., Van Manen, S., Vermeulen, P. G., and Arbocz, J., First-order second-moment analysis
of the buckling of shells with random initial imperfections, AIAA Journal, Vol. 25, 1113–17,
1987.
Ellingwood, B., Statistical analysis of RC beam-column interaction, Journal of Structural Division,
Vol. 103, 1377–88, 1977.
El Naschie, M. S., Exact asymptotic solution for the initial post-buckling of a strut on a linear elastic
foundation, Zeitschrift für angewandte Mathematik und Mekhanik, Vol. 54, 677–83, 1974.
El Naschie, M. S., Local postbuckling of compressed cylindrical shells, Proceedings of the Institution
of Civil Engineers, London, Part 2, Vol. 59, 523–5, 1975a.
El Naschie, M. S., Localized diamond shaped buckling patterns of axially compressed cylindrical
shells, AIAA Journal, Vol. 13, 837–8, 1975b.
El Naschie, M. S., Nonlinear isometric bifurcation and shell buckling, Zeitschrift für angewandte
Mathematik und Mechanik, Vol. 57, 293–6, 1977.
El Naschie, M. S., Stress, Stability and Chaos in Structural Engineering: an Energy Approach,
McGraw-Hill, London, 1990.
Englestad, S. P., and Reddy, Y. N., A probabilistic postbuckling analysis of composite shells, Com-
putational Stochastic Mechanics (P. D. Spanos and C. A. Brebbia, eds.), Elsevier Applied Science,
London, pp. 839–50, 1991.
Ernst, L. J., A finite element approach to shell problems, Research Report WTHD-114, Delft University
of Technology, Department of Mechanical Engineering, Delft, the Netherlands, 1979.
Farshad, M., Stability of Structures, Elsevier Science Publishers, Amsterdam, 1994.
302 BIBLIOGRAPHY
Fedorov, E. I., About probabilistic analysis of beam-column, Problems of Reliability in Structural

Design (S. A. Timoshenko, ed.), Sverdlovsk, pp. 255–61, 1972 (in Russian).
Fersht, R. S., Buckling of cylindrical shells with random imperfections, PhD Thesis, California
Institute of Technology, 1968a.
Fersht, R. S., Almost sure stability of long cylindrical shells with random imperfections, NASA CR-
1161, 1968b.
Fersht, R. S., Buckling of cylindrical shells with random imperfections, Thin Shell Structures: Theory,
Experiment and Design (Y. C. Fung and E. E. Sechler, eds.), Prentice Hall, Englewood Cliffs, NJ,
pp. 325–41, 1974.
Floris, C., and Mazzucchelli, A., Reliability assessment of RC column under stochastic stress, Journal
Folz, B., and Foschi, R. O., Stochastic stiffness properties and buckling of laminated wood columns,
Structural Safety and Reliability (G. I. Schuëller, M. Shinozuka, and J. T. P. Yao, eds.), Balkema,
Rotterdam, pp. 577–89, 1994.
Frangopol, D. M., Ide, Y., Spacone, E., and Iwaki, I., A new look at reliability of reinforced concrete
columns, Structural Safety, Vol. 18, 123–50, 1996.
Fraser, W. B., Buckling of a structure with random imperfections, PhD Thesis, Division of Engineering
and Applied Physics, Harvard University, 1965.
Fraser, W. B., and Budiansky, B., The buckling of a column with random initial deflections, Journal
Applied Mechanics, Vol. 36, 232–40, 1969.
Freudenthal, A. M., Safety and probability of structural failure, Transactions of the ASME, Vol. 121,
1337–75, 1956.
Freudenthal, A. M., Fatigue sensitivity and reliability of mechanical systems, especially aircraft struc-
tures, WADD Technical Report 61–53, Wright-Patterson AFB, Dayton, OH, 1961.
Frieze, P. A., Das, P. K., and Faulkner, D., Partial safety factors for stringer stiffened cylinders under
extreme compressive loads, Proceedings, Second International Symposium on Practical Design in
Shipbuilding, Tokyo, pp. 475–82, 1983.
Fukumoto, Y., and Itoh, Y., Statistical study of experiments on welded beams, Journal of Structural
Division, Vol. 107, 89–102, 1981.
Fukumoto, Y., and Itoh, Y., Basic compressive strength of steel plates from test data, Transactions of
Japan Society of Civil Engineers, Vol. 344/I-1, 129–39, 1984.
Fujimoto, M., and Iwata, M., A study on buckling strength of steel columns by the application of prob-
abilistic method – Consideration of random initial imperfection, Translation of the Architectural
Institute of Japan, No. 218, 17–25, 1974 (in Japanese).
Fukimoto, Y., Itoh, Y., and Kubo, M., Strength variation of laterally unsupported beams, Journal of
Structural Division, Vol. 106, 165–81, 1979.
Fujimoto, M., Iwata, M., and Nakatani, F., A study on buckling strength of steel columns by the
application of probabilistic method – Weak axis buckling of H-section members with random
residual stress, random yield stress and initial deflection, Transactions of the Architectual Institute
of Japan, No. 229, 53–61, 1975 (in Japan).
Fukimoto, Y., Kajita, N., and Aoki, T., Evaluation of column curves based on probabilistic concept,
Preliminary Report on Second International Colloquium on Stability, Tokyo, 1976.
Furstenberg, H., Noncommuting random products, Transactions of the American Mathematical So-
ciety, Vol. 108, 377–428, 1963.
Galambos, T. V., Reliability of axially loaded columns, Engineering Structures, Vol. 5, 73–8, 1983.
Galambos, T. V., Guide to Stability Design Criteria for Metal Structures, John Wiley and Sons, New
York, 1988.
Galoussis, E., and Vasilas, V., On the investigation of the stochastic non-linear snap-through phe-
nomenon in structures, Zeitschrift für angewandte Mathematik und Mechanik, Vol. 61, 40–1,
1981.
Gáspár, Z. S., Probability of the instability of imperfect structures, Zeitschrift für angewandte Math-
ematik und Mechanik, Vol. 63, T48–49, 1983.
Gatermann, K., and Hohmann, A., Symbolic exploitation of symmetry in numerical pathfollowing,
Impact of Computational Science and Engineering, Vol. 3, 330–65, 1991.
BIBLIOGRAPHY 303
Gauthier, R. W., Maximum statistical strength of steel columns, MS Thesis, Department of Civil
Enginering, MIT, 1971.
Geier, B., and Rohwer, K., On the analysis of the buckling behavior of laminated composite plates
and shells, International Journal of Numerical Methods in Engineering, Vol. 27, 403–27, 1989.
Geier, B., Klein, H., and Zimmerman, R., Buckling tests with axially compressed unstiffened cylin-
drical shells made from CFRP, Buckling of Shells Structures on Land, in the Sea and in the Air (J. F.
Jullien, ed.), Elsevier Applied Science, London, pp. 498–507, 1991.
Gelfand, M., and Lokutsievski, O. V., The double sweep method for solution of difference equa-
tions, Theory of Difference Schemes (S. K. Godunov and V. S. Ryabenkii, eds.), North Holland,
Amsterdam, pp. 239–62, 1964.
Ghanem, R. G., and Spanos, P. D., Stochastic Finite Elements: A Spectral Approach, Springer, New
York, 1991.
Ghanem, R. G., and Spanos P. D., A spectral formulation of stochastic finite elements, Probabilistic
Methods for Structural Design (C. Guedes Soares, ed.), Kluwer, Dordrecht, pp. 289–312, 1997.
Giavotto, V., Poggi, C., and Chrissanthopoulus, M., Buckling of imperfect composite cylindrical shells
under compression and torsion, Annual Forum of the American Helicopter Society, Atlanta, Paper
43, 1991.
Godunov, S. K., On the numerical solution of boundary value problems for systems of linear ordinary
differential equations, Uspekhi Matematicheskikh Nauk, Vol. 16, 171–4, 1961 (in Russian).
Godunov, S. K., Equations in Mathematical Physics, “Nauka” Publishing House, Moscow, pp. 394–
403, 1971 (in Russian).
Godunov, S. K., and Ryabenkii, V. S., Theory of Difference Schemes, North Holland, Amsterdam,
1964.
Goggin, P. R., The elastic constants of carbon-fiber composites, Journal of Materials Science, Vol. 8,
233–44, 1973.
Goncherenko, V. M., Towards determination of probability of stability loss of the shell, Izvestiya
Akademii Nauk SSSR, Otdelenie Tekhnicheskykh Nauk, Mekhanika i Mashinostoenie, No. 1, 159–
60, 1962a (in Russian).
Goncharenko, V. M., Investigation of probability of snapping of a long cylindrical panel under random
pressure, Prikladnaya Matematika i Mekhanika, Vol. 26, 740–4, 1926b (in Russian).
Goncharenko, V. M., On one aspect of the statistical method in the theory of stability of shells, Theory
of Plates and Shells, Academy of Ukraine Press, Kiev, pp. 368–71, 1962c (in Russian).
Goncharenko, V. M., Application of Markov processes in the statistical theory of stability of shells,
Ukrainskii Matematicheskii Zhurnal, Vol. 14, 198–202, 1962d (in Russian).
Goncharenko, V. M., Snap-through of a panel in presence of random excitation, Theory of Shells and
Plates, Academy of Armenia Press, Erevan, pp. 383–90, 1964 (in Russian).
Goncharenko, V. M., About dynamic problems of statistical theory of stability of elastic systems,
Problems of Stability in Structural Mechanics, “Stroiizdat” Publishers, Moscow, pp. 210–16, 1965
(in Russian).
Goodman, T. R., and Lance, G. N., The numerical solution of two-point boundary value problems,
Mathematical Tables and Other Aids of Computation, Vol. 10, 82–6, 1956.
Goodman, R. A., and Movatt, G. A., Allowance for imperfections in ship structural design, Journal
of Strain Analysis, Vol. 12, 153–62, 1977.
Grandori, G., Paradigms and falsification in earthquake engineering, Meccanica, Vol., 26, 17–21,
1991.
Greenberg, J. B., and Stavsky, Y., Buckling and vibration of orthotropic composite cylindrical shells,
Acta Mechanica, Vol. 36, 15–29, 1980.
Greenberg, J. B., and Stavsky, Y., Stability and vibration of compressed, aelotropic, composite cylin-
drical shells, Journal of Applied Mechanics, Vol. 49, 843–8, 1982.
Greenberg, J. B., and Stavsky, Y., Buckling of composite orthotropic cylindrical shells under non-
uniform axial loads, Composite Structures, Vol. 30, 399–406, 1995.
Greene, W. H., Effects of random member length errors on the accuracy and internal loads of truss
antennas, Proceedings of AIAA/ASME/ASCE/AHS 24th Structures, Structural Dynamics and Ma-
terials Conference, Lake Tahoe, pp. 697–704, 1983.
304 BIBLIOGRAPHY
Grigolyuk, E. I., and Kabanov, V. V., Stability of Shells, “Nauka” Publishers, Moscow, 1978a
(in Russian).
Grigolyuk, E. I., and Kabanov, V. V., Stability of Circular Cylindrical Shells, “VINITI” Publishers,
Moscow, 1978b (in Russian).
Grigolyuk, E. I., and Lipovtsev, Yu. V., On the solution of a class of eigenvalue problems in thin
shells of revolution, Problems of Mechanics of Deformed Solids, Volume Dedicated to 60th An-
niversary of V. V. Novozhilov, “Sudostroenie” Publishing House, Leningrad, pp. 129–41, 1975
(in Russian).
Grigolyuk, E. I., Maltsev, V. P., Myachenkov, V. I., and Frolov, A. N., On a solution method for sta-
bility and vibration problems in shells of revolution, Mekhanika Tverdogo Tela, Vol. 1, 9–19, 1971,
(in Russian).
Grigorenko, Yu. M., Isotropic and Anisotropic Rotational Shells of Varying Stiffness, “Naukova
Dumka” Publisher, Kiev, 1970 (in Russian).
Grigoriu, M., Applied Non-Gaussian Processes, PTR Prentice Hall, Englewood Cliffs, NJ, 1995.
Grove, T., and Didriksen, T., Buckling experiments on large axial stiffened and one ring stiffened
cylindrical shells, Report 76-432, Det norske Veritas, 1976.
Groves, S. E., and Highsmith, A. L. (eds.), Compression Response of Composite Structures, ASTM,
New York, 1994.
Guedes Soares, C., Uncertainty modeling in plate buckling, Structural Safety, Vol. 5, 17–34, 1988.
Guedes Soares, C., Probabilistic modelling of the strength of flat compression members, Probabilistic
Methods for Structural Design (C. Guedes Soares, ed.), Kluwer Academic Publishers, Dortrecht,
pp. 113–40, 1997.
Guedes Soares, C., and Silva, A. G., Reliability of unstiffened plate elements under in-plane com-
bined loading, Proceedings 10th Offshore Mechanics and Arctic Engineering Symposium, ASME,
Norway, 1991.
Guliaev, V. I., Bazhenov, V. A., and Gozuliak, E. I., Stability of Nonlinear Mechanical Systems, Lvov
University Press, Lvov, 1982 (in Russian).
Guliaev, V. I., Bazhenov, V. A., and Lizunov, P. P., Non-classical Theory of Shells and Its Applications
to Solution of Engineering Problems, Lvov University Press, Lvov, 1978 (in Russian).
Gusic, G., Combescure, A. and Jullien, J. F., Influence of circumferential thickness variations on
buckling of cylindrical shells under external pressure, Computers and Structures, 2000 (in press).
Guz, A. N., Stability of Three Dimensional Deformable Bodies, “Naukova Dumka” Publishers, Kiev,
1971 (in Russian).
Guz, A. N., and Babich, I. Yu., Three Dimensional Theory of Stability of Beams, Plates and Shells,
“Vischa Shkola” Publishers, Kiev, 1980 (in Russian).
Haines, W., Hierarchy methods for random vibrations of elastic strings and beams, Journal of Engi-
neering Mathematics, Vol. 1, 293–305, 1967.
Hammersley, J. M., and Handscomb, D. G., Monte Carlo Method, Methuen, London, 1964.
Handbook of Structural Stability, Column Research Committee of Japan, Corona, Tokyo, 1971.
Hangai, Y., and Kawamata, S., Analysis of geometrically nonlinear and stability problems by static
perturbation method, Report of the Institute of Industrial Science, The University of Tokyo, Vol. 22,
No. 5 (Serial No. 143), 1973.
Hansen, J. S., Buckling of imperfection-sensitive structures: A probabilistic approch, PhD Disserta-
tion, Department of Civil Engineering, University of Waterloo, 1973.
Hansen, J. S., Influence of general imperfections in axially loaded cylindrical shells, International
Journal of Solids Structures, Vol. 11, 1223–33, 1975.
Hansen, J. S., General random imperfection in the buckling of axially loaded cylindrical shells, AIAA
Journal, Vol. 15, 1250–6, 1977.
Hansen, J. S., and Roorda, J., Reliability of imperfection sensitive structures, Stochastic Problems in
Mechanics, (S. T. Ariaratnam and H. H. E. Leipholz, eds.), University of Waterloo Press, Waterloo,
pp. 229–42, 1973.
Hansen, J. S., and Roorda, J., On a probabilistic stability theory for imperfection sensitive structures,
International Journal of Solids and Structures, Vol. 10, 341–59, 1974.
BIBLIOGRAPHY 305
Harris, L., Suer, H. S., Skene, W. T., and Benjamin, R. J., The stability of thin-walled unstiffened
circular cylinders under axial compression including the effects of internal pressure, Journal of the
Aeronautical Sciences, 587–96, 1957.
Hart, D. K., Rutherford, S. E., and Wickham, A. H. S., Structural reliability analysis of stiffened plates,
Transactions of the Royal Institution of Naval Architects, Vol. 128, 293–310, 1986.
Hart-Smith, L. J., Buckling of thin cylindrical shells under uniform axial compression, International
Journal of Mechanical Sciences, Vol. 12, 299–313, 1970.
Hasofer, A. M., and Lind, N. C., Exact and invariant second-moment code format, Journal of the
Engineering Mechanics Division, Vol. 100, 111–21, 1974.
Hawranek, R., and Rackwitz, P., Reliability calibration for steel columns, Bulletin d’Information No.
112, Comité Européen du Beton, pp. 125–57, 1976.
Healey, T. J., A group theoretic approach to computational bifurcation problems with symmetry,
Computational Methods in Applied Mechanics and Engineering, Vol. 67, 257–95, 1988.
Herrmann, G. (ed.), Dynamic Stability of Structures, Pergamon Press, New York, 1967.
Hilton, H. H., Yi, S., and Danyluk, M. J., Stochastic delamination buckling of elastic and viscoelastic
columns, Computational Stochastic Mechanics (P. D. Spanos, ed.), Balkema, Rotterdam, pp. 687–
96, 1995.
Himmelblau, D. M., Applied Nonlinear Programming, McGraw-Hill, New York, pp. 177–8, 1972.
Hirano, Y., Buckling of angle-ply laminated circular cylindrical shells, Journal of Applied Mechanics,
Vol. 46, 233–4, 1979.
Hirano, Y., Optimization of laminated composite shells for axial buckling, Transactions of the Japan
Society for Aeronautical and Space Science, Vol. 26, 154–62, 1983.
Hodges, C. H., Confinement of vibration by structural irregularity, Journal of Sound and Vibration,
Vol. 82, 441–4, 1982.
Hodges, C. H., and Woodhouse, J., Vibration isolation from irregularity in a nearly periodic struc-
ture: Theory and measurements, Journal of the Acoustical Society of America, Vol. 74, 894–905,
1983.
Hoff, N. J., The accuracy of Donnell’s equations, Journal of Applied Mechanics, Vol. 23, 329–34,
1955.
Hoff, N. J., The perplexing behavior of thin circular cylindrical shells in axial compression, Israel
Journal of Technology, Vol. 4, 1–28, 1966.
Höft, H. F. W., and Höft, M., Computing with Mathematica, Academic Press, San Diego, 1998.
Hoh, Y., Usami, T., and Fukumoto, Y., Experimental and numeral analysis and database on structural
stability, Engineering Structures, Vol. 18, 812–20, 1996.
Horne, M. R., and Merchant, W., Stability of Frames, Pregamon Press, Oxford, 1965.
Horton, W. H., and Durham, S. C., Imperfection, a main contributor to scatter in experimental values
of buckling load, International Journal of Solids and Structures, Vol. 1, 59–72, 1965.
Hoshiya, M., and Shah, H. C., Free vibration of a stochastic beam column, Journal of Engineering
Mechanics, Vol. 97, 1239–55, 1971.
Hoshiya, M., and Spence, S. T., Reliability of a single flexible column with three spring supports.
Proceeding of JSCE, No. 183, 1970.
Hour, K. Y., Damage characterization of a random-oriented fiber reinforced composite, Mechanics
of Composite Materials–Nonlinear Effects (M. W. Hyer, ed.), AMD-Vol. 159, ASME Press, New
York, pp. 303–18, 1993.
Hu, H. T., Buckling analysis of fiber composite laminate plates with material nonlinearity, Finite
Element Analysis and Design, Vol. 19, 169–80, 1995.
Hunt, G. W., Elastic stability in structural mechanics and applied mechanics, Collapse: The Buckling of
Structures in Theory and Practice (J. M. T. Thompson and G. W. Hunt, eds.), Cambridge University
Press, London, pp. 123–47, 1983.
Hunt, G. W., Hidden asymmetries of elastic and plastic bifurcation, Applied Mechanics Reviews,
Vol. 39, 1165–86, 1986.
Hunt, G. W., and Wadee, M. K., Comparative Lagrangian formulations for localized buckling, Pro-
ceedings of the Royal Society of London, Vol. A 434, 485–502, 1991.
306 BIBLIOGRAPHY
Hunt, G. W., and Wadee, M. A., Localization and mode interaction in sandwich structures, Proceed-
ings of the Royal Society of London, Series A–Mathematical, Physical and Engineering Sciences,
Vol. 454, 1197–216, 1998.
Hunt, G. W., Bolt, H. M., and Thompson, J. M. T., Structural localization phenomena and the
dynamical phase-space analogy, Proceedings of the Royal Society of London, Vol. A 425, 245–67,
1986.
Hussain, M. A., and Noble, B., Applications of MACSYMA to calculations in applied mechanics,
Report No. 83CRD054, General Electric, Schenectady, New York, March 1983.
Hutchinson, J. W., Plastic buckling, Advances in Applied Mechanics (C. S. Yih, ed.), Academic Press,
New York, pp. 67–143, 1974.
Hutchinson, J. W., and Amazigo, J. C., Imperfection-sensitivity of eccentrically stiffened cylindrical
shells, AIAA Journal, 392–401, 1967.
Hutchinson, J. W., and Frauenthal, J. C., Elastic postbuckling behavior of stiffened and barreled
cylindrical shells, Journal of Applied Mechanics, Vol. 37, 784–90, 1969.
Hutchinson, J. W., and Koiter, W. T., Postbuckling theory. Applied Mechanics Reviews, Vol. 23, 1353–
66, 1970.
Ibrahim, A., Buckling mode localization in rib-stiffened plates with misplaced stiffness, Master of
Applied Science Thesis, Department of Civil Engineering, University of Waterloo, 1997.
Ikeda, K., and Murota, K., Computation of critical initial imperfections of truss structures, Journal
of Engineering Mechanics, Vol. 116, 2101–17, 1990a.
Ikeda, K., and Murota, K., Critical initial imperfection of structures, International Journal of Solids
and Structures, Vol. 26, 865–86, 1990b.
Ikeda, K., and Murota, K., Bifurcation analysis of symmetric structures using block diagonalization,
Computational Methods in Applied Mechanics and Engineering, Vol. 86, 215–43, 1991a.
Ikeda, K., and Murota, K., Random initial imperfections of structures, International Journal of Solids
Structures, Vol. 28, 1003–21, 1991b.
Ikeda, K., and Murota, K., Statistics of normally distributed initial imperfections, International
Journal of Solids Structures, Vol. 30, 2445–67, 1993.
Ikeda, K., Chida, T., and Yanagisawa, E., Imperfection sensitive strength variation of soil specimens,
Journal of Mechanics of Physics of Solids, Vol. 45, 293–315, 1997.
Ikeda K., Murota, K., and Elishakoff. I., Reliability of structures subject to normally distributed initial
imperfections, Computers and Structures, Vol. 59, 463–9, 1995.
Ikeda, K., Murota, K., and Fuji, H., Bifurcation hierarchy of symmetric structure, International Journal
of Solids and Structures, Vol. 27, 1551–73, 1991.
Ilin, V. A., Ovcharov, N. P., and Sukalho, A. A., Statistical diagnosis of critical forces during the
investigation of stability of imperfect cylindrical shells, Izvestiya Akademii Nauk SSSR, Mekhanika
Tverdogo Tela, No. 5, 191–4, 1975 (in Russian).
Imbert K., The effect of imperfections on the buckling of cylindrical shells, Aeronautical Engineer
Thesis, California Institute of Technology, Pasadena, 1971.
Itoh, Y., and Fukimoto, Y., Stochastic evaluation of compressive strength of unstiffened plate
components, 4th International Colloquium on Stability of Plate and Shell Structures, Belgium,
1987.
Itoh, Y., Kubo, M., and Fukimoto, Y., Statistical experiment of strength of laterally unsupported rolled
H-beams, Japan Society of Civil Engineers, Annual Conference, pp. 579–80, 1978.
Itoh, Y., Usami, T., and Fukumoto, Y., Experimental and numerical analysis database on structural
stability, Engineering Structures, Vol. 18, 812–20, 1996.
Ivanov, L. D., and Rousev, S. H., Statistical estimation of reduction coefficient of ship’s hull plates
with initial deflections, The Naval Architect, No. 4, 158–60, 1979.
Iwatsubo, T., Tomita, K., and Kawai, R., Evaluation of unstable probability of rotors having stiffness
errors of asymmetry, Bulletin of the JSME, Vol. 25, No. 206, 1299–1305, 1982.
Jaquet, J., Essais de flambement et exploitation statistique, Construction Métallique, No. 3, 1970
(in French).
BIBLIOGRAPHY 307
Jasquot, R. G., Nonstationary random column buckling problem, Journal of the Engineering Mechan-
ics Division, Vol. 98, 1173–82, 1972.
Jensen, J., and Niordson, F., Symbolic and algebraic manipulation languages and their application in
mechanics, Structural Mechanics Software Series, Vol. l, 541–76, 1977.
Jeong, G. D., Critical buckling load analysis of uncertain column, Probabilistic Mechanics and Struc-
tural and Geotechnical Reliability, 6th Specialty Conference, pp. 563–6, 1992.
Jin, W-L., and Luz, E., Improving importance sampling method in structural reliability, Nuclear
Engineering Design, Vol. 147, 393–401, 1994.
Johns, K. C., Some statistical aspects of coupled buckling structures, Buckling of Structures
(B. Budiansky, ed.), Springer, Berlin, pp. 199–207, 1974.
Johns, K. C., Strength statistics for thin shell structures, Proceedings of the Fifth Canadian Congress
of Applied Mechanics, Fredericton, pp. 169–70, 1975.
Jones, R. M., Mechanics of Composite Materials, Hemisphere Publishing Company, Fredericton,
pp. 45–54, 1975.
Julien, J. F. (ed.), Buckling of Shell Structures on the Land, in the Sea, and in the Air, Elsevier Applied
Science, London, 1991.
Kabanov, V. V., Investigation of stability of shells by methods of finite difference, Izvestiya Akademii
Nauk SSSR, Mekhanika Tverdogo Tela, No. 1, 24–9, 1971 (in Russian).
Kabanov, V. V., Stability of Non-homogeneous Cylindrical Shells, “Mashinostroenie” Publishers,
Moscow, 1982 (in Russian).
Kac, M., and Siegert, A. J. F., An explicit representation of stationary Gaussian process, Ann. Mat.
Stat, Vol. 18, 438–42, 1947.
Kahrimanian, H. G., Analytic differentiation by a digital computer, MS Thesis, Temple University,
Philadelphia, PA, 1953.
Kaplyevatsky, Y., Generalization of Frobenius’ method for stability analysis of orthotropic annular
plates, Acta Mechanica, Vol. 22, 295–307, 1975.
Karadeniz, H., Van Manen, S., and Vrouwenvelder, A., Probabilistic reliability analysis for the fatigue
limit state of gravity and jacket type structures, Proceedings of the Third International Conference –
BOSS, McGraw-Hill, London, 1982.
Karhunen, K., Ueber lineare methoden in der wahrscheinlichkeitsrechnung, Amer. Acad. Sci., Fenni-
cade, Ser. A, I, Vol. 37, 3–79, 1947 (in German); English translation: RAND Corp., Santa Monica,
(A, Report T-131, Aug. 1960).
Von Karman, T., and Tsien, H. S., The buckling of thin cylindrical shells under axial compression,
Journal of Aerospace Sciences, Vol. 8, 303–12, 1941.
Kasagi, A., Interactive buckling of ring stiffened composite cylindrical shells, D. Sc. Thesis, Wash-
ington University, 1994.
Kasagi, A., and Sridharan, S., Postbuckling analysis of layered composite using p-version finite strips,
International Journal of Numerical Methods in Engineering, Vol. 33, 2091–107, 1992.
Kasagi, A., and Sridharan, S., Buckling and postbuckling analysis of thick composite cylindrical
shells under hydrostatic pressure using axisymmetric solid elements, Composites Engineering,
Vol. 3, 467–87, 1993.
Kasagi, A., and Sridharan, S., Imperfection sensitivity of layered composite cylinders, Journal of
Keener, J. P., Buckling imperfection sensitivity of columns and spherical caps, Quarterly Journal of
Applied Mathematics, Vol. 32, 173–99, 1974.
Keller, J. B., Stochastic equations and wave propagation in random media, Proceedings of Sym-
posia on Applied Mathematics, XVI, Providence, American Mathematical Society, RI, pp. 145–70,
1964.
Keller, J. B., and Antman, S., Bifurcation Theory and Non-Linear Eigenvalue Problems, W. A.
Benjamin, New York, 1969.
Kendall, K., and Stuart, A., The Advanced Theory of Statistics, Vol. 2, Charles Griffin & Co. Ltd.,
London, 1973.
308 BIBLIOGRAPHY
Kennedy, D., and Williams, F. W., More efficient use of determinants to solve transcendental structural
eigenvalue problems reliably, Computers and Structures, Vol. 41, 973–9, 1991.
Khot, N. S., On the influence of initial geometric imperfections on the buckling and postbuckling
behavior of fiber-reinforced cylindrical shells under uniform axial compression, AFFDL-TR-68-
136, 1968.
Khot, N. S., Buckling and postbuckling behavior of composite cylindrical shells under axial com-
pression, AIAA Journal, Vol. 8, 229–35, 1970a.
Khot, N. S., Postbuckling behavior of geometrically imperfect composite cylindrical shells under axial
compression, AIAA Journal, Vol. 8, 579–81, 1970b.
Khot, N. S., and Venkayya, V. B., Effect of fiber orientation on initial postbuckling behavior and
imperfection sensitivity of composite cylindrical shells, AFFDL-TR-79-125, 1970.
Kirsch, U., Optimum Structural Design, McGraw-Hill, New York, 1981.
Kisliakov, S., An investigation of buckling danger of bars, columns and other structural ele-
ments as a result of random overloadings or small deviations from the standard of materials,
Annuaire des Ecoles Techniques Supérieures, Mechanique Appliqué, Vol. VI, Livre, 1, 43–58, 1971
(in Bulgarian).
Kissel, G. I., Localization in disordered periodic structures, Proceedings of 28th AIAA/ASME/
ASCE/AHS Structures, Structural Dynamics and Materials Conference, AIAA Paper 87–0819,
Monterey, 1987.
Kissel, G. I., Localization on disordered periodic structures, PhD Thesis, Department of Aeronautics
and Astronautics, MIT, 1988.
Kissel, G. I., Localization factor for multichannel disorder system, Physical Reviews, Vol. 44, 1008–
14, 1991.
Kleiber, M., Coupled numeric-symbolic computations in structural mechanics, Discretization Meth-
ods in Structural Mechanics (G. Kuhin and H. Mang, eds.), Springer, Berlin, 317–26, 1989.
Kleiber, M., and Hien, T. D., Stochastic Finite Element Method, Wiley, New York, 1993.
Klimanov, V. I., and Timashev, S. A., Nonlinear Problems of Stiffened Shells, Academy of Sciences,
Ural Scientific Center, Sverdlovsk, 1985 (in Russian).
Klimov, D. M., Computer algebra in mechanics, General and Applied Mechanics (A. Yu. Ishlinsky et
al., eds.), Hemisphere, New York, pp. 163–78, 1990.
Klimov, D. M., and Rudenko, V. M., Computer Algebra in Applied Mechanics, “Nauka” Publishing
House, Moscow, 1989 (in Russian).
Kline, L. V., and Hancock, J. O., Buckling of circular plate on elastic foundation, Journal of Engi-
neering and Industry, Vol. 87B, 323–4, 1965.
Klompé, A. W. H., Initial imperfection survey on a cylindrical shell (C. S. E.) at the Fokker B. V.,
Report LR-495, Delft University of Technology, Department of Aerospace Engineering, Delft, the
Netherlands, 1986.
Klompé, A. W. H., UNIVIMP–A universal instrument for the survey of initial imperfections of
thin-walled shells, Report LR-570, Delft University of Technology, Department of Aerospace En-
gineering, Delft, the Netherlands, 1988.
Klompé, A. W. H., The initial imperfection data bank at the Delft University of Technology, Part IV,
Netherlands, 1989.
Klompé, A. W. H., and Den Reyer, P. C., The initial imperfection data bank at the Delft University
of Technology, Part III, Report LR-568, Delft University of Technology, Department of Aerospace
Engineering, Delft, the Netherlands, 1989.
Kmiecik, M., Jastrzeboski, T., and Kuzniar, J., Statics of ship plating distortions, Marine Structures,
Vol. 8, No. 2, 119–32, 1995.
Kogiso, N., Reliability analysis and reliability-based optimization of composite laminated plate for
buckling, PhD Thesis, Department of Aerospace Engineering, Osaka Prefecture University, Sakai,
Osaka, Japan, 1997.
Kogiso, N., Shao, S., and Murotsu, Y., Reliability-based optimum design of symmetric laminated
plate subject to buckling, Structural Optimization, Vol. 14, 184–92, 1997.
BIBLIOGRAPHY 309
Kogiso, N., Shao, S., and Murotsu, Y., Effect of correlation on reliability-based design of composite
plate for buckling, AIAA Journal, Vol. 36, 1706–13, 1998.
Koiter, W. T., On the stability of elastic equilibrium, PhD Dissertation, Delft University, 1945 (English
translations: a, NASA TT-F-10, 833, 1967; b, AFFDL-TR-70-25, 1970).
Koiter, W. T., The effect of axisymmetric imperfections on the buckling of cylindrical shells under
axial compression, Proceedings of Royal Netherlands Academy of Sciences, Amsterdam, Series 13,
Vol. 66, 265–79, 1963.
Koiter, W. T., The energy criterion of stability for continuous elastic bodies, I, II, Proceedings,
Koninklijke Nederlandse Akademie van Wetenschappen, Series B, Vol. 68, 178–202, 1965.
Koiter, W. T., Purpose and achievements of research in elastic stability, Proceedings of 4th Technical
Conference, Society of Engineering Science, North Carolina State University, Raleigh, NC, pp. 197–
218, 1966a.
Koiter, W. T., General equations of elastic stability for thin shells. Proceedings of the Symposium on
the Theory of Shells, Houston, pp. 187–227, 1966b.
Koiter, W. T., Postbuckling analysis of single two-bar frames, Recent Progress in Applied Mechanics,
(B. Broberg et al., eds.), Almquist and Wiksell, Göteborg, 1967.
Koiter, W. T., A basic open problem in the theory of elastic stability, Proceedings, IUTAM/IMU
Symposium on Application of Methods of Functional Analysis to Problems of Mechanics, Marseille,
Part XXIX, 1975.
Koiter, W. T., Current trends in the theory of buckling, Buckling of Structures (B. Budiansky, ed.),
Springer Verlag, Berlin, pp. 1–16, 1976.
Koiter, W. T., Buckling of cylindrical shells under axial compression and external pressure, Thin
Shell Theory. New Trends and Applications (W. Olszak, ed.), CISM Courses and Lectures No. 240,
Springer Verlag, Vienna, pp. 77–87, 1980.
Koiter, W. T., Elastic Stability, 28th Ludwig Prandtl Memorial Lecture, Zeitschrift für Flugwis-
senschaften und Weltraumforschung, Vol. 9, 205–10, 1985.
Koiter, W. T., Elishakoff, I., Li, Y. W., and Starnes, J. H., Buckling of an axially compressed cylindrical
shell of variable thickness, International Journal of Solids and Structures, Vol. 31, pp. 797–805,
1994a.
Koiter, W. T., Elishakoff, I., Li, Y. W., and Starnes, J. H., Buckling of an axially compressed imperfect
cylindrical perfect shell of variable thickness, Proceedings of 35th AIAA ASME/ASCE/AHS/ASC
Structures, Structural Dynamics and Materials Conference, Hilton Head, SC, AIAA Press,
Washington, DC, pp. 277–289, 1994b.
Kondubhatla, S. R. V., Stability of elastic-plastic columns with random geometric imperfections,
SECTAM, Southeastern Theoretical and Applied Mechanics Conference, Deerfield Beach, FL,
1998.
Korncoff, A. R., and Fenves, S. J., Symbolic generation of finite element stiffness matrices, Computers
and Structures, Vol. 10, 119–24, 1978.
Kotoguchi, H., Leonard, J. W., and Shiomi, H., Statistical evaluation of steel beam-column resistance,
Engineering Structures, Vol. 3, 573–88, 1985.
Kowal, Z., and Raducki, K., Interaction of random critical load of spatial truss joint loaded at the joint
and on the bars, Third International Conference on Space Structures (H. Nooshin, ed.), Elsevier
Applied Science Publishers, Amsterdam, pp. 606–12, 1984.
Köylüoǧlu, H. U., Nielsen, S. R. K., and Cakmak, A. A., Uncertain buckling load and reliability of
columns with uncertain properties, Department of Building Technology and Structural Engineering,
Paper No. 141, Aalborg University, Denmark, 1995.
Krätzig, W. B., and Oňate, E. (eds.), Static and Dynamic Stability of Shells, Springer Verlag, Berlin,
1990.
Krätzig, W. B., Basar, Y., and Wittek, U., Nonlinear behaviour and elastic stability of shells: Theoretical
concepts, numerical computations and results, Buckling of Shells: Proceedings of a State-of-the-Art
Colloquium (E. Ramm, ed.), Springer Verlag, Berlin, pp. 19–57, 1982.
Krein, M. G., Vibration theory of multi-span beams, Vestnik Inzhenerov i Tekhnikov, Vol. 4, 142–5,
1933 (in Russian).
310 BIBLIOGRAPHY
Kweon, J. H., and Hong, C. S., Improved arc-length method for postbuckling analysis of composite
cylindrical panels, Computers and Structures, Vol. 53, 541–9, 1994.
La Tegola, A., Il carico di collasso delle strutture con imperfezioni random, Giornale del Genio Ciorle,
1976 (in Italian).
La Tegola, A., Simulation methods for the analysis of pin-jointed structures with randomly imperfect
members, Second International Symposium on Innovative Numerical Analysis in Applied Engi-
neering Science, Ecole Polytechnique de Montreal, 1980.
Lagace, P. A., Jensen, D. W., and Finch, D. C., Buckling of unsymmetric composite laminates, Com-
posite Structures, Vol. 5, 101–23, 1986.
Langley, R. C., Forced response of 1D periodic structures: Vibration localization by damping, Journal
of Sound and Vibration, Vol. 178, 411–28, 1994.
Law, A., and Kelton, W., Simulation Modeling and Analysis, McGraw-Hill, New York 1982.
Lebedev, N. N., Special Functions and Their Applications, Dover Publications, New York, 1972.
Leipholz, H. H. E., Stability of Elastic Systems, Sijthoff and Noordhoff, Alphen san den Rijn, 1980.
Leissa, A. W., Buckling of laminated composite plates and shell panels, AFWAL-TR-85 3069, AF
Wright Aeronautical Laboratory, Dayton, OH, 1985.
Leizerakh, V. M., Statistical analysis of random imperfections in cylindrical shells with the aid of the
computer, Proceedings of the Moscow Power Engineering Institute, Subsection of Dynamics and
Strength of Machines (V. V. Bolotin, ed.), MEI Press, Moscow, pp. 45–56, 1969 (in Russian).
Leizerakh, V. M., and Makarov, B. P., Statistical correlation analysis of deformation of compressed
shell with initial imperfections, Theory of Plates and Shells, Proceedings of the 8th All-Union
Conference, Rostov-on-Don, pp. 320–324, 1971 (in Russian).
Lenz, J. C., Reliability based design rules for column buckling, PhD Thesis, Department of Civil
Engineering, Washington University, 1972.
Lenz, J., Ravindra, M. K., and Galambos, T. V., Reliability based rules for column buckling, Computers
and Structures, Vol. 3, 573–88, 1973.
Li, Y. W., Prediction of natural frequency variability due to uncertainty in material properties, Pro-
ceedings of the NOISE-CON’94, Progress in Noise Control for Industry (J. M. Cuschieri, S. A. L.
Glegg, and D. M. Yeager, eds.), Fort Lauderdale, pp. 917–22, 1994.
Li, Y. W., and Gan, H. L., Evaluation of the probability of failure for engineering structures by a
modified simulation method, Journal of Shanghai Institute of Railway Technology, Vol. 13, 29–39,
1992 (in Chinese).
Li, Y. W., Elishakoff, I., and Starnes, J. H., Jr., Axial buckling of composite cylindrical shells with
periodic thickness variation, Buckling and Postbuckling of Composite Structures (A. K. Noor, ed.),
ASME Press, pp. 95–114, 1994 (see also Computers and Structures, Vol. 56, pp. 65–74, 1995).
Li, Y. W., Elishakoff, I., and Starnes, J. H., Jr., Buckling mode localization in a multi-span periodic
structure with a disorder in a single span, Chaos, Solitons and Fractals, Vol. 5, 955–69, 1995a.
Li, Y. W., Elishakoff, I., and Starnes, J. H., Jr., Effect of thickness variation and initial imperfection
on buckling of composite cylindrical shells, Computers and Structures, Vol. 56, 65–74, 1995b.
Li, Y. W., Elishakoff, I., Starnes, J. H., Jr., and Bushnell, D., Effect of the thickness variation and initial
imperfection on buckling of composite cylindrical shells: Asymptotic analysis and numerical results
by BOSOR 4 and PANDA2, International Journal of Solids and Structures, Vol. 34, 3755–67, 1997.
Li, Y. W., Elishakoff, I., Starnes, J. H., Jr., and Shinozuka, M., Nonlinear buckling of a structure
with random imperfection and random axial compression by a conditional simulation technique,
Li, Y. W., Elishakoff, I., Starnes, J. H., Jr., and Shinozuka, M., Prediction of natural frequency and
buckling load variability by convex modelling, Fields Institute Communications, Vol. 9, 139–54,
1996.
Libai, A., and Durban, A., A method of approximate stability analysis and its application to circu-
lar cylindrical shells under circumferentially varying edge loads, Journal of Applied Mechanics,
Vol. 40, 971–6, 1973.
Librescu, L., and Lin, W., Classical versus non-classical postbuckling behaviour of laminated com-
posite panels under complex loading conditions, Report, Department of Engineering Science and
Mechanics, Virginia Polytechnic Institute and State University, 1993.
BIBLIOGRAPHY 311
Liew, K. M., and Wang, C. M., Elastic buckling of rectangular plates with curved internal supports,
Journal of Structural Engineering, Vol. 118, 1480–93, 1990.
Lim, K. B., and Junkins, J. L., Probability of stability: New measures of stability robustness for linear
dynamical systems, The Journal of the Astronautical Sciences, Vol. 35, 383–97, 1987.
Lin, S. C., and Kam, T. Y., Buckling analysis of imperfect frames, using a stochastic finite element
method, Computers and Structures, Vol. 42, 895–901, 1992.
Lin, S. C., Kam, T. Y., and Chu, K. H., Evaluation of buckling and first-ply failure, probabilities of
composite laminates, International Journal of Solids and Structures, Vol. 35, 1395–410, 1998.
Lin, Y. K., and McDaniel, T. J., Dynamics of beam type periodic structures, Journal of Engineering
for Industry, Vol. 93, 1133–41, 1969.
Lindberg, H. E., Random imperfection for dynamic pulse buckling, Journal of Engineering Mechan-
ics, Vol. 114, 1144–65, 1990.
Lindberg, H. E., Dynamic response and buckling failure measures for structures with bounded and
random imperfections, Journal of Applied Mechanics, Vol. 58, 1092–5, 1991.
Lindberg, H. E., An evaluation for convex modeling for multimode dynamic buckling, Journal of
Applied Mechanics, Vol. 59, 923–36, 1992a.
Lindberg, H. E., Convex models for uncertain imperfection control in multimode dynamic buckling,
Journal of Applied Mechanics, Vol. 59, 937–45, 1992b.
Lipovskii, D. E., Altukher, G. M., Koz, V. M., Nazarov, V. A., Todchuk, V. A., and Shun, V. M.,
Statistical estimation of the effect of random disturbances on the stability of stiffened shells based on
the experimented investigations, Analysis of Spacewise Structures, Vol. 17, “Stroiizdat” Publishers,
Moscow, pp. 32–44, 1977 (in Russian).
Lipovtsev, Y. V., Features of application of the double-sweep method to stability problems in shells
and plates, Mekhanika Tverdogo Tela, Vol. 3, 43–9, 1970 (in Russian).
Litle, W. A., Reliability of shell buckling predictions based on experimental analysis of plastic models,
ScD Thesis, Department of Civil Engineering, MIT, 1963.
Liu, J. K., and Zhao, L. C., Mode localization and eigenvalue loci veering phenomena in disordered
aircraft structures, Proceedings of 12th International Modal Analysis Conference, Honolulu, Vol. 1,
pp. 374–8, 1994.
Liu, W. K., Belytschko T., and Mani, A., Random field finite elements, International Journal of
Numerical Methods in Engineering, Vol. 23, 1831–45, 1984.
Lockhart, D., Dynamic buckling of damped imperfect columns on a nonlinear foundation, Quarterly
of Applied Mathematics, Vol. 36, 49–55, 1978.
Lockhart, D., and Amazigo, J. C., Dynamic buckling of externally pressurized imperfect cylindrical
shells, Journal of Applied Mechanics, Vol. 42, 316–20, 1975.
Loeve, M., Functions aleatoires du second ordre, Supplement to P. Levy, Processus Stochastic et
Mouvement Brownian, Gauthier Villars, Paris, 1948 (in French).
Lombardi, M., and Haftka, R. T., Anti-optimization technique for structural design under load uncer-
tainties, Computer Methods in Applied Mechanics and Engineering, Vol. 157, 19–31, 1998.
Loo, T. T., An extension of Donnell’s equations for circular cylindrical shell, Journal of Aeronautical
Science, Vol. 24, 390–1, 1957.
Lorenz, R., Achsensymmetrische Verzerrung in dünnwandigen Hohlzylindern, Zeitschrift der Vereines
Deutscher Ingenieure, Vol. 52, 1706–13, 1908 (in German).
Lorenz, R., Die nichtachsensymmetrische Knickung dünnwanger Hohlzylinder, Physikalische
Zeitschrift, Vol. 13, 241–60, 1911 (in German).
Lottati, L., and Elishakoff, I., Influence of shear deformation and rotary inertia on flutter of a can-
tilevered beam-exact and symbolic computerized solutions, Refined Dynamical Theories in Beams,
Plates and Shells and Their Applications (I. Elishakoff and H. Irretier, eds.), Springer Verlag,
Berlin, pp. 261–73, 1987.
Lukoshevichus, R. S., Rikards, R. B., and Teters, G. A., Probabilistic analysis of stability and min-
imization of mass of the cylindrical shell made of composite materials with random geometric
imperfections, Mekhanika Polimerov, No. 1, 80–9, 1977 (in Russian).
Lust, S. D., Friedmann, P. P., and Bendiksen, O. O., Free and forced response of multi-span beams and
multi-bay trusses with localized modes, Journal of Sound and Vibration, Vol. 180, 313–32, 1995.
312 BIBLIOGRAPHY
Maidanik, G., and Dickey, J., Localization and delocalization in one-dimensional dynamic systems,
Proceedings of the IEEE Ultrasonic Symposium, Vol. 1, pp. 507–14, 1996.
Maimon, G., and Libai, A., Dynamics and failure of cylindrical shells subjected to axial impact, AIAA
Journal, Vol. 15, 1624–30, 1977.
Mak, C. K. K., Probabilistic buckling behavior of arches with random imperfections, Recent Advances
in Engineering Science, Vol. 7, 443–50, 1970.
Mak, C. K. K., and Kelsey, S., Statistical aspects in the analysis of structures with random imperfec-
tions, Statics and Probability in Civil Engineering (P. Lumb, ed.), Hong Kong University Press,
Hong Kong, pp. 569–84, 1971.
Mak, C. K. K., and Kelsey, S., Probabilistic buckling behavior of structure with random lack-of-fit and
geometric imperfections, Proceedings of the 2nd International Conference on Structural Mechanics
in Reactor Technology, Berlin, Vol. V – Part M, Paper M7/10, 1973.
Makar, P. F., Application of an implicit linear statistical analysis of the estimation of the resistance of
a reinforced concrete beam-column, Miscellaneous Paper N-78-7, Defense Nuclear Agency, 1978.
Makarov, B. P., Application of a statistical method for analysis of experimental data on shell stability,
Izvestiya Akademii Nauk, SSSR, Otdelenie Tekhnicheskikh Nauk, Mekhanika i Mashinostroenie,
No. 1, 157–8, 1962a (in Russian).
Makarov, B. P., Application of statistical method for analysis of nonlinear problems of shell stability,
Theory of Plates and Shells, Proceedings of the All-Union Conference, Kiev, pp. 363–7, 1962b
(in Russian).
Makarov, B. P., Analysis of nonlinear shell stability problems with the aid of a statistical method,
Inzhenernyi Zhurnal, Vol. 3, 100–106, 1963 (in Russian).
Makarov, B. P., Statistical analysis of stability of imperfect cylindrical shells, Proceedings of 7th
All-Union Conference on the Theory of Plates and Shells, Dnepropetrovsk, pp. 387–91, 1969
(in Russian).
Makarov, B. P., Statistical analysis of nonideal cylindrical shells, Izvestija Akademii Nauk SSSR,
Mekhanika Tverdogo Tela, Vol. 5, 97–104, 1970 (in Russian).
Makarov, B. P., Statistical analysis of deformation of imperfect cylindrical shells, Strength Analysis,
Vol. 15, “Mashinostroenie” Publishing House, Moscow, 240–56, 1971 (in Russian).
Makarov, B. P., Postcritical deformations of non-ideal spherical shells, Problems of Reliability in
Structural Design, Sverdlovsk, pp. 126–131, 1972 (in Russian).
Makarov, B. P., and Chechenev, N. A., About snap-through of thin elastic panels under random
impulsive loads, Raschety na Prochnost, Vol. 11, “Mashinostroenie” Publishing House, Moscow,
pp. 378–84, 1965 (in Russian).
Makarov, B. P., and Leizerakh, V. M., Towards stability of cylindrical shells with random initial
imperfections, Proceedings of the Moscow Power Engineering Institute (V. V. Bolotin, ed.), Moscow,
pp. 35–45, 1969 (in Russian).
Makarov, B. P., and Leizerakh, V. M., On stability of nonideal cylindrical shell in axial compression,
Proceedings of the Moscow Power Engineering Institute, Dynamics and Strength of Machines
(V. V. Bolotin, ed.), Vol. 74, 24–9, 1970 (in Russian).
Makarov, B. P., Leizerakh, V. M., and Sudakova, N. I., Statistical investigation of initial imperfections
of cylindrical shells, Problems of Reliability in Structural Mechanics (V. V. Bolotin and A. Cyras,
eds.), “RINTIP” Publishing House, Vilnius, 1968 (in Russian).
Makushin, V. M., Effective application of an energy method of investigation of elastic stability of
columns and plates, Raschety na Prochnost’ (Strength Calculations) 8, “Mashinostroenie” Pub-
lishing House, Moscow, pp. 225–52, 1962.
Manevich, A. I., Stability and Optimum Design of Stiffened Shells, “Visha Shkola” Publishers, Kiev,
1975 (in Russian).
Manevich, L. I., and Prokopalo, E. F., On statistical properties of load carrying capacity of smooth
cylindrical shells, Problems of Reliability in Structural Mechanics (V. V. Bolotin and A. Cyras,
eds.), “RINTIP” Publishing House, Vilnius, 1968.
Manevich, L. I., Milzin, A. M., Mossakvskii, V. I., Probkpalo, E. F., Smelii, G. N., and Sotnikov, D.
I., Experimental investigation of stability of unstiffened cylindrical shells of various scales in axial
compression, Izvestiya Akademii Nauk SSSR, Mckhanika Tverdoygo Tela, No. 5, 1968 (in Russian).
BIBLIOGRAPHY 313
Massey, F. J., Jr., The Kolmogorov-Smirnov test for goodness of fit, Journal of American Statistical
Assessment, Vol. 46, 68–78, 1951.
Matthies, H. G., and Bucher, Ch., Finite elements for stochastic media problems, Computer Methods
in Applied Mechanics and Engineering, Vol. 168, 3–17, 1999.
McDaniel, T. J., The deflection of columns with random centroidal loci, M. Sci. Thesis, Department
of Aeronautical and Astronautical Engineering, University of Illinois, Urbana, 1962.
McFarland, D., Smith, B., and Bernhart, W., Analysis of Plates, Spartan Books, New York, 1972.
Mead D. J., Vibration response and wave propagation in periodic structures, Journal of Engineering
for Industry, Vol. 41, 467–86, 1971.
Mead, D. J., Free wave propagation in periodically supported infinite beams, Journal of Sound and
Vibration, Vol. 11, 181–97, 1973.
Mester, S. S., and Benaroya, H., Periodic and near-periodic structures, Shock and Vibration, Vol. 2,
69–95, 1995.
Miles, J. W., Vibration of beams on many supports, Journal of Engineering Mechanics, Vol. 82, 1–9,
1956.
Miller, R. K., and Hedgepeth, J. M., The buckling of lattice columns with stochastic imperfections,
Mirza, S. A., Probability based strength criterion for reinforced concrete slender columns, ACI Struc-
tures Journal, Vol. 84, 459–66, 1987.
Mirza, S. A., and MacGregor, J. G., Slenderness and strength reliability of reinforced concrete columns,
ACI Structures Journal, Vol. 86, 428–38, 1989.
Mirza, S. A., and Skrabek, B. W., Reliability of short composite beam-column strength interaction,
Journal of Structural Engineering, Vol. 117, 2320–1, 1991.
Mirza, S. A., Lee, P. M., and Morgan, D. L., ACI stability resistance factor for RC columns, Journal
Mlakar, P. F., Reliability based safety factors for concrete structures sliding stability, Proceedings of
the 26th Conference on the Design of Experiments in Army Research Development and Testing,
ARO Report 81-82, 1981.
Moghtaderi-Zadeh, M., and Madsen, H., Probabilistic analysis of buckling and collapse of plates
in marine structures, Proceedings 6th Offshore Mechanics and Arctic Engineering Symposium,
ASME, Houston, 1986.
Morandi, A. C., Computer aided reliability based design of externally pressurized vessels, PhD Thesis,
Department of Naval Architecture and Ocean Engineering, University of Glasgow, 1994.
Morandi, A. C., Das, P. K., and Faulkner, D., Finite element analysis and reliability based design of
externally pressurized ring stiffened cylinders, The Royal Institution of Naval Architects, pp. 1–18,
W4, 1996.
Mork, K. J., Bjornsen, T., Venas, A., and Thorkildsen, F., Reliability-based calibration study for
upheaval buckling of pipelines, Journal of Offshore Mechanics and Arctic Engineering, Vol. 119,
203–8, 1997.
Morley, L. S., An improvement of Donnell’s approximation for thin walled circular cylinders, Quarterly
Journal of Mechanics and Applied Mathematics, Vol. 12, 89–99, 1959.
Morris, N. F., Shell stability: The long road from theory to practice, Engineering Structures, Vol. 18,
801–6, 1996.
Mossakowsky, J., Buckling of circular plates with cylindrical orthotropy, Archiwum Mechaniki
Stosowanej, Vol. 12, 583–96, 1960.
Mossakowsky, J., and Borsuk, K., Buckling and vibrations of circular plates with cylindrical or-
thotropy, Applied Mechanics (A. Rolla and W. T. Koiter, eds.), Elsevier, Amsterdam, pp. 266–9,
1960.
Movsisian, L. A., On stability of cylindrical shell with random initial stresses, Reliability Problems in
Structural Mechanics (V. V. Bolotin and A. A. Cyras, eds.), “RINTIP” Publishing House, Vilnius,
pp. 187–91, 1968 (in Russian).
Muggeridge, D. B., The effect of initial axisymmetric shape imperfections on the buckling behavior of
circular cylindrical shells under axial compression, UTIAS Report No. 148, Institute for Aerospace
Studies, University of Toronto, 1969.
314 BIBLIOGRAPHY
Mura, T., and Koya, T., Variational Methods in Mechanics, Oxford University Press, New York, 1992.
Murota, K., and Ikeda, K., Critical imperfection of symmetric structures, SIAM Journal of Applied
Mathematics, Vol. 51, 1222–54, 1991.
Murota, K., and Ikeda, K., On random imperfections for structures of regular-polygonal symmetry,
SIAM Journal of Applied Mathematics, Vol. 52, 1780–803, 1992.
Murzewski, J. W., Random instability of elastic-plastic frames, Structural Safety and Reliability
(A. H.-S. Ang, M. Shinozuka, and G. I. Schuëller, eds.), ASCE Press, New York, pp. 1011–14,
1990.
Murzewski, J. W., Overall instability of steel frames with random imperfections, International Col-
loquium, East-European Session, Stability of Steel Structures (M. Ivanji and B. Veröci, eds.),
Bucharest, pp. II/221–8, 1990.
Mutoh, I., Kato, S., and Chiba, Y., Alternative lower bounds analysis of elastic thin shells of revolution,
Engineering Computations, Vol. 13, 41–75, 1996.
Muzean, J. P., Fogli, M., and Lemaire, M., Buckling of structures in probabilistic context, Third Inter-
national Conference on Space Structures (H. Nooshin, ed.), Elsevier Applied Science Publishers,
London, pp. 486–91, 1984.
Nagabhushnan, J., Gaonkar, G. K., and Reddy, T. S. R., Automatic generation of equations for rotor-
body systems with dynamic inflow for a priori ordering schemes, Paper No. 7, 7th European
Rotorcraft Forum, Garmish-Partenkirchen, Federal Republic of Germany, 1981.
Nakagiri, S., and Hisada, T., An Introduction to Stochastic Finite Element Method: Analysis of Un-
certain Structures, Bai Fukan, Tokyo, 1985 (in Japanese).
Nakagiri, S., Takabatake, M., and Tani, S., Uncertain eigenvalue analysis of composite laminated
plates by the stochastic finite element method, Journal of Engineering for Industry, Vol. 109,
9–12, 1987.
NASA, Buckling of thin-walled circular cylinders, NASA SP 8007, Aug. 1968.
Nash, W. A., Instability of thin shells, Applied Mechanics Surveys (H. N. Abramson, H. Liebowitz,
J. M. Crowley, and S. Juhasz, eds.), Spartan Books, New York, pp. 339–56, 1966.
Nayfeh, A. H., and Hawwa, M., Buckling mode localization in multi-span columns, Proceedings
of 35th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference,
April 18–20, Hilton Head, pp. 277–89, 1994a.
Nayfeh, A. H., and Hawwa, M. A., The use of mode localization in the passive control of structural
buckling, AIAA Journal, Vol. 32, 2131–3, 1994b.
Neal, D. M., Mathews, W. T., and Vangel, M. G., Uncertainties in obtaining high reliability from
stress-strength models, Proceedings, Ninth DoD/NASA/FAA Conference on Fibrous Composites In
Structural Design, Lake Tahoe, pp. 503–21, 1992.
Nemeth, M. P., Importance of anisotropy on buckling of compression-loaded symmetric composite
plates, AIAA Journal, Vol. 24, 1831–5, 1986.
Nemirovskii, B. Ya., Towards probabilistic analysis of a flexible shallow shell, Stroitelnaya Mekhaniki
i Raschet Sooruzhenii (Structural Mechanics and Analysis of Buildings), No. 3, 17–21, 1968
(in Russian).
Nemirovskii, B. Ya., Probabilistic design of flexible shallow shells under combined loading, Stroi-
tel’naya Mekhanika i Raschet Sooruzhenii, No. 3, 38–42, 1969 (in Russian).
Newton, D. A., and Ayaru, B., Statistical methods in the design of structural timber columns, The
Structural Engineer, Vol. 50, 191–5, 1972.
Niedenfuhr, F. W., Scatter of observed buckling loads of pressurized shells, AIAA Journal, Vol. 1,
1923–5, 1963.
Nieuwendijk van den, H. L., Preliminary study of a new design criterion for shells under axial com-
pression, Memorandum M-805, Faculty of Aerospace Engineering, Delft University of Technology,
July 1997.
Nikolaidis, E., Hughes, O., Ayyub, B. M., and White, G. J., Reliability analysis of stiffened panels,
Structures Congress XII (N. C. Baker and B. J. Goodno, eds.), Vol. 2, ASCE Press, New York,
pp. 1466–71, 1994.
Niordson, F. I., Shell Theory, North Holland, Amsterdam, 1985.
BIBLIOGRAPHY 315
Nolan, J., Analytic differentiation on a digital computer, SM Thesis, MIT Cambridge, MA, 1953.
Noor, A. K., Survey of computer programs for solution of non-linear structural and solid mechanics
problems, Computers and Structures, Vol. 13, 425–65, 1981.
Noor, A. K., Recent advances and applications of reduction methods, Applied Mechanics Reviews,
Vol. 47, 125–46, 1994.
Noor, A. K., and Andersen, C. M., Computerized symbolic manipulation in structural mechanics-
progress and potential, Computers and Structures, Vol. 10, 95–118, 1979.
Noor, A. K., and Peters, J. M., Recent advances in reduction methods for instability analysis of
structures, Computers and Structures, Vol. 16, 67–80, 1983.
Noor, A. K., Andersen, C. M., and Peters, J. M., Reduced basis techniques for collapse analysis of
shells, AIAA Journal, Vol. 19, 393–7, 1981.
Noor, A. K., Elishakoff, I., and Hulbert, G. (eds.), Symbolic Computations and Their Impact on
Mechanics, ASME Press, New York, 1990.
Nordgren, R. P., On One-Dimensional Random Fields with Fixed End Values, Probabilistic Engi-
neering Mechanics, Vol. 14, 301–10, 1999.
Nordgren, R. P., and Hussain, K. S., Probabilistic stability analysis of shallow arches, Engineering
Mechanics, Proceedings of the 11th Conference (Y. K. Lin and T. C. Su, eds.), ASCE Press,
New York, pp. 124–7, 1996.
Novichkov, Yu. N., and Indenbaum, V. M., Deformation of a circular cylinder made of low-modulus
material and enclosed in a shell, Proceedings of the Moscow Energy Institute, Dynamics and
Strength of Machines, Vol. 101, pp. 48–55, 1972 (in Russian).
Oceledec, Y. I., A multiplicative ergodic theorem. Lyapunov characteristic number for dynamical
systems, Transactions of the Moscow Mathematical Society, Vol. 19, 197–231, 1968.
Oh, D. M., and Librescu, L., Free vibration and reliability of composite cantilevers featuring uncertain
properties, Reliability Engineering and System Safety, Vol. 56, 265–72, 1997.
Ohsaki, M., and Nakanura, T., Optimum design with imperfection sensitivity coefficients for limit
point loads, Structural Optimization, Vol. 8, 131–7, 1997.
Pai, S. S., and Chamis, C. C., Probabilistic progressive buckling of trusses, Journal of Spacecraft and
Rockets, Vol. 31, 466–74, 1994.
Palassopoulos, G. V., A probabilistic approach to the buckling of thin cylindrical shells with gen-
eral random imperfections: Solution of the corresponding deterministic problem, Theory of Shells
(W. T. Koiter and G. K. Mikhailov, eds.), North Holland, Amsterdam, pp. 417–43,
1980.
Palassoupoulos, G. V., On the effect of stochastic imperfections on the buckling strength of certain
structures, Computational Stochastic Mechanics, (P. D. Spanos and C. A. Brebbia, eds.), Elsevier
Applied Science, London, pp. 211–22, 1991a.
Palassopoulos, G. V., Reliability-based design of imperfection sensitive structures, Journal of Engi-
neering Mechanics, Vol. 117, 1220–40, 1991b.
Palassopoulos, G. V., Response variability of structures subjected to bifurcation buckling, Journal of
Palassopoulos, G. V., A new approach to the buckling of imperfection-sensitive structures, Journal of
Palassopoulos, G. V., An improved approach to the reliability-based design of structures subjected to
bifurcation buckling, Structural Safety and Reliability (G. I. Schuëller, M. Shinozuka, and J. T. P.
Yao, eds.), Balkema Publishers, Rotterdam, Vol. 2, pp. 767–76, 1994.
Palassopoulos, G. V., Efficient finite element analysis of imperfection-sensitive frames and arches
with stochastic imperfections, Computational Stochastic Mechanics (P. D. Spanos, ed.), Balkema
Publishers, Rotterdam, pp. 461–8, 1995.
Palassopoulos, G. V., Buckling analysis and design of imperfection sensitive structures, Uncertainty
Modeling in Finite Element, Fatigue and Stability of Systems, (A. Haldar, A. Guran, and B. M.
Ayyub, eds.), World Scientific, Singapore, pp. 311–56, 1997.
Palassopoulos, G. V., and Shinozuka, M., On the elastic stability of thin shells, Journal of Structural
Mechanics, Vol. 1, 439–49, 1973.
316 BIBLIOGRAPHY
Pandalai, K. A. V., and Patel, S. A., Buckling of orthotropic circular plates, Journal of Royal Aero-
nautical Society, Vol. 69, 279–80, 1965.
Panovko, Ya. G., and Gubanova, I. I., Stability and Oscillations of Elastic Systems, Paradoxes, Fallacies
and New Concepts, Consultants Bureau, New York, 1964.
Pantelides, C. P., Buckling of elastic columns using convex model of uncertain springs, Journal of
Parkus, H., Die Beulwahrscheinlichkeit einer Platte in einem transversalen magnetfeld, Sitzungs-
berichten der Österreichischen Akademie der Wissenschaften, Mathem.-naturw, Klasse, Abteilung
II, 180, Bcl. 4. bis. 7, Heft, 185–91, 1971 (in German).
Patton, C. M., Symbolic computation for handheld calculators, Hewlett-Packard Journal, Aug., 21–25,
1987.
Pavelle, R. (ed.), Applications of Computer Algebra, Kluwer, Boston, 1985.
Pavlovic, M. N., and Sapountzakis, E. J., Computers and structures: Non-numerical applications,
Pedersen, P., On computer-aided analytic element analysis and the similarities of tetrahedron elements,
International Journal of Numerical Methods in Engineering, Vol. 11, 611–22, 1977.
Peek, R., Worst shapes of imperfections for space trusses with multiple global and local buckling
nodes, International Journal of Solids and Structures, Vol. 30, 2243–60, 1993.
Peek, R., and Triantafylidis, N., Worst shapes of imperfections for space trusses with many simul-
taneously buckling members, International Journal of Solids and Structures, Vol. 29, 2385–403,
1992.
Perov, V. A., Roev, V. A., Riashina, N. A., and Sharygin, I. A., About post-critical deformations of thin
shells with random initial imperfections, Proceedings of the Moscow Power Engineering Institute
(M. M. Orakhelashvili, ed.), pp. 86–95, 1970 (in Russian).
Perry, S. H., Statistical variation of buckling strength, PhD Thesis, Department of Civil and Mechanical
Engineering, University College, London, 1966.
Pflüger, A., Stabilitätsprobleme der Elastostatik, Springer Verlag, Berlin, 1967 (in German).
Pflüger, W., Die Stabilität de Kreiszylinderschale, Ingenieur-Archiev, Vol. 3, 463–506, 1932
(in German).
Pi, H. N., Ariaratnam, S. T., and Lennox, W. C., First-passage time for the snap-through of a shell-type
structure, Journal of Sound and Vibration, Vol. 14, 375–84, 1971.
Pierre, C., Weak and strong vibration localization in disordered structures: A statistical investigation,
Journal of Sound and Vibration, Vol. 139, 111–32, 1990.
Pierre, C., and Chat, P. D., Strong mode localization in nearly periodic disordered structures, AIAA
Journal, Vol. 27, 227–41, 1989.
Pierre, C., and Dowell, E. H., Localization of vibrations by structural irregularity, Journal of Sound
and Vibration, Vol. 114, 549–64, 1987.
Pierre, C., and Plaut, R. H., Curve veering and mode localization in a buckling problem, Journal of
Applied Mathematics and Physics (ZAMP), Vol. 40, 758–61, 1989.
Pignataro, M., Rizzi, N., and Luongo, A., Stability, Bifurcation and Postbuckling Behavior of Elastic
Structures, Elsevier Science Publishers, Amsterdam, 1991.
Potier-Ferry, M., Multiple Bifurcation, Symmetry and Secondary Bifurcation, Pitman Press, London,
1981a.
Potier-Ferry, M., Critères de l’énergie en élasticité et viscoélasticité, Le Flambement des Structures
(R. L’Hermite, ed.), Editions Bâtiment Travaux Publics, Paris, pp. 23–37, 1981b (in French).
Potier-Ferry, M., Foundations of elastic postbuckling theory, Buckling and Post-Buckling (J. Arbocz,
M. Potier-Ferry, J. Singer, and V. Tvergaard, eds.), Springer Verlag, Berlin, pp. 1–82, 1987.
Power, T. L., and Kyriakides, S., Localization and propagation of instabilities on long shallow panels
under external pressure, Journal of Applied Mechanics, Vol. 61, 755–63, 1994.
Pshenichnov, G. I., Statistical analysis of net-cylindrical shells, Inzhenernyi Sbornik, Vol. 27, 171–8,
1960 (in Russian).
Pu, Y., Das, P. K., and Faulkner, D., Ultimate compression strength and probabilistic analysis of
stiffness plates, Proceedings of the 15th Offshore Mechanics and Arctic Engineering Conference,
Vol. II, ASME Press, New York, pp. 151–7, 1996.
BIBLIOGRAPHY 317
Qiria, V. S., Motion of the bodies in resisting media, Proceedings of the Tbilisi State University,
Vol. 44, 1–20, 1951 (in Russian).
Qiu, Z. P., and Elishakoff, I., Antioptimization of structures with large uncertain-but-nonrandom pa-
rameters via interval analysis, Computer Methods in Applied Mechanics and Engineering, Vol. 152,
361–72, 1998.
Ramaiah, G. K., and Vijayakumar, K., Buckling of polar orthotropic annular plates under uniform
internal pressure, AIAA Journal, Vol. 12, 1045–50, 1974.
Ramaiah, G. K., and Vijayakumar, K., Elastic stability of annular plates under uniform compressive
forces along the outer edge, AIAA Journal, Vol. 13, 832–5, 1975.
Ramm, E. (ed.), Buckling of Shells, Proceedings of the State-of-the-Art Colloquium, Springer Verlag,
Berlin, 1982a.
Ramm, E., The Riks/Wempner approach – an extension of the displacement control method in nonlin-
ear analysis, Recent Advances in Nonlinear Computational Mechanics (E. Hinton, D. R. J. Owen,
and C. Taylor, eds.), Pineridge Press, Swansea, pp. 63–86, 1982b.
Ramu, S. A., and Ganesan, R., Stability analysis of a stochastic column subjected to stochastically
distributed loadings using the finite element method, Finite Elements in Analysis and Design,
Vol. 11, 105–15, 1992.
Ramu, S. A., and Ganesan, R., Parametric stability of stochastic columns, International Journal of
Solids and Structures, Vol. 30, 1339–54, 1993.
Ramu, S. A., and Ganesan, R., Response and stability of a stochastic beam-column using stochastic
FEM, Computers and Structures, Vol. 54, 207–22, 1995.
Rand, R. H., Computer Algebra in Applied Mathematics: An Introduction to MACSYMA, Pitman,
London, 1984.
Rand, R. H., and Armbruster, D., Perturbation Methods, Bifurcation Theory and Computer Algebra,
Springer Verlag, Berlin, 1987.
Rankin, C. C., Riks, E., Starnes, Jr., J. H., and Waters, Jr. A., An experimental and numerical verification
of the postbuckling behavior of a composite cylinder in compression, NASA Contract Report, to
appear.
Rankin, C. S., Stehlin, P., and Brogan, F. A., Enhancements to the STAGS computer code, NASA
CR-4000, 1986.
Rantis, T. D., and Taingjitham, S., Probability based stability analysis of a laminated composite plate
under confined in-plane load, CCMS-94-17, Virginia Polytechnic Institute and State University,
Center for Composite Materials and Structures, Blacksburg, VA, 1993.
Rayleigh, B., The Theory of Sound, Dover Publications, New York, 2nd ed., 1945 (1st ed., Macmillan,
New York, 1894).
Rayna, G., REDUCE: Software for Algebraic Computation, Springer, Berlin, 1987.
Rehak, M., Dillaggio, F., Benaroya, H., and Elishakoff, I., Random vibrations with MACSYMA,
Computer Methods in Applied Mechanics and Engineering, Vol. 61, 61–70, 1987.
Reissner, E., On postbuckling behavior and imperfection sensitivity of thin plates on a non-linear
foundation, Studies on Applied Mathematics, Vol. XLIX, 45–7, 1970.
Reissner, E., A note on imperfection sensitivity of thin plates on a non-linear elastic foundation,
Instability of Continuous Systems (H. H. E. Leipholz, ed.), Springer Verlag, Berlin, pp. 15–18,
1971.
Reznikov, B. S., Probability of failure of thermoelastic reinforced shells, Strength of Materials, Vol. 14,
1364–8, 1982 (in Russian).
Richardson, J. M., The application of truncated hierarchy techniques in the solution of stochastic
linear differential equations, Proceedings of Symposium in Applied Mechanics, pp. 290–302, 1964.
Rikards, R. B., and Teters, G. A., Stability of Shells Made of Composite Materials, “Zinatne” Pub-
lishers, Riga, 1974 (in Russian).
Riks, E., An incremental approach to the solution of snapping and buckling problems, International
Journal of Solids and Structures, Vol. 15, 529–51, 1979a.
Riks, E., Development of an improved analysis capability of the STAGS computer code with respect to
the collapse behavior of shell structures, Memorandum SC-79-054, National Aerospace Laboratory
NRL, the Netherlands, 1979b.
318 BIBLIOGRAPHY
Riks, E., Some computational aspects of the stability analysis of nonlinear structures, Computer
Methods in Applied Mechanics and Engineering, Vol. 47, 219–59, 1984.
Riks, E., Progress in collapse analysis, Journal of Pressure Vessels Technology, Vol. 109, 33–41, 1987.
Riks, E., Brogan, F. A., and Rankin, C. C., Aspects of the stability analysis of shells, Static and
Dynamic Stability of Shells (W. B. Krätzig and E. Oňate, eds.), Springer Verlag, Berlin, 1988.
Rizzi, N., and Tatone, A., Symbolic manipulation in buckling and postbuckling analysis, Computers
and Structures, Vol. 21, 691–700, 1985a.
Rizzi, N., and Tatone, A., Using symbolic computation in buckling analysis, Journal of Symbolic
Computation, Vol. 1, 317–21, 1985b.
Roberts, S. M., and Shipman, J. S., Two-Point Boundary Values Problems: Shooting Methods, Amer-
ican Elsevier, New York, pp. 50–86, 1972.
Rokach, A. J., A statistical study of the steel columns, M.Sc. Thesis, Report R0-60, Department of
Civil Engineering, MIT, 1970.
Romstad, K. M., Hutchinson, J. R., and Runge, K. H., Design parameter variation and structural
response, International Journal of Numerical Methods in Engineering, Vol. 5, 337–49, 1973.
Roorda, J., Stability of structures with small imperfections, Journal of Engineering Mechanics Divi-
sion, Vol. 91, 87–106, 1965.
Roorda, J., Some statistical aspects of the buckling of imperfection-sensitive structures, Journal of
Mechanics, Physics and Solids, Vol. 17, 111–23, 1969.
Roorda, J., An experience in equilibrium and stability, Technical No. 3, University of Waterloo, Solid
Mechanics Division, Waterloo, 1971.
Roorda, J., Buckling of shells; an old idea with a new twist, Journal of Engineering Mechanics
Division, Vol. 98, 531–8, 1972a.
Roorda, J., Concepts in elastic structural stability, Mechanics Today (S. Nemat-Nasser, ed.), Pergamon
Press, New York, Vol. 1, pp. 322–72, 1972b.
Roorda, Y., The random nature of column failure, Journal of Structural Mechanics, Vol. 3, 239–57,
1975.
Roorda, J., Buckling of Elastic Structures, University of Waterloo Press, Waterloo, 1980.
Roorda, J., and Chilver, A. M., Frame buckling: An illustrative of the perturbation technique, Inter-
national Journal of Non-Linear, Vol. 5, 235–46, 1970.
Roorda, J., and Hansen, J. S., Random buckling behavior in axially loaded cylindrical shells with
axisymmetric imperfections, Journal of Spacecraft, Vol. 9, 88–91, 1972.
Rubinstein, R., Simulation and the Monte Carlo Method, John Wiley, New York, 1981.
Ruiz, S. E., and Aguilar, J. C., Reliability of short and slender reinforce concrete columns, Journal of
Structural Engineering, Vol. 120, 1850–65, 1994.
Rurlykina, V. N., and Minaev, V. A., About a statistical problem of stability of elastic beam, Analysis
of Strength, Stability and Vibrations of Elements of Engineering Structures, Vononezh, pp. 114–17,
1981 (in Russian).
Rzadkowski, J. W., Random ultimate bearing capacity of elastic-plastic space trusses, Third Inter-
national Conference on Space structures (H. Nooshin, ed.), Elsevier Applied Science Publishers,
London, pp. 561–6, 1984.
Rzhanitsin, A. R., Structural Design with Consideration of Plastic Properties of Materials, “Stroivoen-
morizdat,” Moscow, 1949 (in Russian).
Rzhanitsin, A. R., Statistical Method of Determination of Allowable Stresses for Beam-Columns,
Gosudarstvennoe Izdatelstvo Stroitelnoi Literatury, Moscow, 1951 (in Russian).
Rzhanitsin, A. R., Statistical stability of a compressed column, Reliability Problems in Structural
Mechanics (V. V. Bolotin and A. Cyras, eds.), “RINTIP” Publishing House, Vilnius, pp. 192–8,
1968 (in Russian).
Rzhanitsin, A. R., and Visir, P. L., Stability of equilibrium and motion under conditions of random
initial and boundary conditions, Loadings and Reliability of Building Constructions Proceedings
of ZNIISK, Vol. 21, “Stroiizdat” Publishers, Moscow, 1973 (in Russian).
Sabag, M., Stavsky, Y., and Greenberg, J. B., Buckling of edge-damaged, cylindrical composite shells,
Journal of Applied Mechanics, Vol. 56, 121–6, 1989.
BIBLIOGRAPHY 319
Sadovsky, Z., Stochastic resistance of square plate under uniaxial compression, Engineering Struc-
tures, Vol. 19, 827–33, 1997.
Salim, S., Iyengar, N. G. R., and Yadav, D., Buckling of laminated plates with random material
characteristics, Applied Composite Materials, Vol. 5, 1–9, 1998.
Salim, S., Yadav, D., and Iyengar, N. G. R., Analysis of composite plates with random material
characteristics, Mechanics Research Communications, Vol. 20, 405–14, 1993.
Sanders, Jr., J. L., Nonlinear theories for thin shells, Quarterly of Applied Mechanics, Vol. 21, 21–36,
1963.
Sankar, T. S., and Ariaratnam, S. T., Snap buckling of shell-type structures under stochastic loading,
Schenk, C. A., Schuëller, G. I. and Arbocz, J., On the analysis of cylindrical shells with random
imperfections, Fourth Int’l Coll. on Computation of Shell & Spatial Structures, Chania-Crete,
Greece, June 4–7, 2000.
Schenk, C. A., Schuëller, G. I. and Arbocz, J., Buckling analysis of cylindrical shells with random
imperfections, International Conference on Monte Carlo Simulation, Monte Carlo, June 18–21,
2000.
Scheurkogel, A., and Elishakoff, I., On ergodicity assumption in an applied mechanics problem,
Journal of Applied Mechanics, Vol. 52, 133–6, 1985.
Scheurkogel, A., Elishakoff, I., and Kalker, J., On the error that can be induced by an ergodicity
assumption, Journal of Applied Mechanics, Vol. 48, 654–6, 1981.
Schmidt, R., A variant of the Rayleigh-Ritz method, Industrial Mathematics, Vol. 31, 37–46, 1981.
Schmidt, R., Estimation of buckling loads and other eigenvalues via a modification of the Rayleigh-
Ritz method, Journal of Applied Mechanics, Vol. 49, 639–40, 1982.
Schmidt, R., Modification of Timoshenko’s technique for estimating buckling loads, Industrial Math-
ematics, Vol. 33, 169–73, 1983.
Schmidt, H., and Krysik, R., Towards recommendations for shell stability design by means of numer-
ically determined buckling loads, Buckling of Shell Structures on Land, in the Sea and in the Air
(J. F. Jullien, ed.), Elsevier Applied Science, London, pp. 508–19, 1991.
Schweppe, F. C., Recursive state estimation: unknown but bounded errors and system inputs, IEEE
Transactions on Automatic Control, Vol. AC-13, 22–8, 1968.
Schweppe, F. C., Uncertain Dynamic Systems, Prentice Hall, Englewood Cliffs, NJ, 1973.
Sebec, R. W. L., Imperfection surveys and data reduction of Ariane interstage I/II and II/III, Ir. Thesis,
Department of Aerospace Engineering, Delft University of Technology, 1981.
Semeniuk, N. P., About initial post-buckling behavior of composite cylindrical shells in axial com-
pression, Prikladnaya Mekhanika, Vol. 88, No. 5, 34–60, 1988 (in Russian).
Semeniuk, N. P., Formulas for analysis of stability, initial post-buckling behavior and imperfection
sensitivity of layered composite cylindrical shells, Prikladnaya Mekhanika, Vol. 27, No. 7, 31–6,
1991a (in Russian).
Semeniuk, N. P., Towards analysis of stability, initial post-buckling behavior and imperfection-
sensitivity of layered composite cylindrical shells, Prikladnaya Mekhanika, Vol. 27, No. 6, 74–80,
1991b (in Russian).
Semeniuk, N. P., and Zhukova, N. B., Towards solution of a problem on initial post-buckling behavior
of stringer-stiffened cylindrical shells, Prikladnaya Mekhanika, Vol. 27, 81–8, 1991 (in Russian).
Semeniuk, N. P., and Zhukova, N. B., About stability and initial post-buckling behavior of stiffened
cylindrical composite shells, Prikladnaya Mekhanika, Vol. 28, 48–53, 1992 (in Russian).
Semeniuk, N. P., and Zhukova, N. B., Towards the problem of interaction of modes of stability loss for
imperfect composite cylindrical shells, Prikladnaya Mekhanika, Vol. 30, 69–75, 1994 (in Russian).
Sen-Gupta, G., Natural flexural wave and the normal modes of periodically supported beams and
plates, Journal of Sound and Vibration, Vol. 13, 89–101, 1970.
Seward, L. R., REDUCE User’s Guide, for IBM 360 and Derivative Computers, Version 3.2, Rand
Publication CP83, 1985.
Shah, H. C., Regression analysis of reinforced concrete columns, Proceedings of the ASCE, Vol. 90,
1964.
320 BIBLIOGRAPHY
Sheinman, I., Cylindrical buckling load of laminated columns, Journal of Engineering Mechanics,
Vol. 115, 659–61, 1989.
Sheinman, I., and Simitses, G. J., Buckling of geometrically imperfect stiffened cylinders under axial
compression, AIAA Journal, Vol. 15, 374–82, 1977.
Shilkrut, D. I., The influence of the paths of multi-parametrical conservative loading on the behaviour
of a geometrically nonlinear deformable body, Buckling of Structures: Theory and Experiment
(I. Elishakoff, J. Arbocz, C. D. Babcock, Jr., and A. Libai, eds.), Elsevier Science Publishers,
Amsterdam, 1988.
Shilkrut, D. I., and Vyrlan, P. M., Stability of Nonlinear Shells, “Shtiinza” Publishing House, Kishinev,
1977 (in Russian).
Shinozuka, M., Maximum structural response of seismic excitations, Journal of Engineering Me-
chanics Division, Vol. 96, 729–38, 1970.
Shinozuka, M., Structural response variability, Journal of Engineering Mechanics, Vol. 113, 825–42,
1987.
Shinozuka, M., and Deodatis, G., Response variability of stochastic finite element systems, Journal
Shinozuka, M., and Deodatis, G., Simulation of stochastic processes by spectral representation, Ap-
plied Mechanics Reviews, Vol. 44, 191–204, 1991.
Shinozuka, M., and Henry, L., Random vibration of a beam column, Journal of the Engineering
Mechanics Division, Vol. 91, 123–43, 1965.
Shinozuka, M., and Yamazaki, F., Stochastic finite element analysis: An introduction, Stochastic
Structural Dynamics (S. T. Ariaratnam, G. I. Schuëller, and I. Elishakoff, eds.), Elsevier, London,
pp. 241–91, 1988.
Shiraki, W., and Takaoka, N., Reliability analysis of compression members with rectangular cross
section, Theoretical and Applied Mechanics, Vol. 27, 247–58, 1979.
Shiraki, W., and Takaoka, N., Reliability analysis of compression members with nonstationary random
initial deflection, Proceedings of JSCE, No. 297, 37–46, 1980 (in Japanese).
Shreider, Ya. A., Method of Statistical Testing: Monte Carlo Method, Elsevier, Amsterdam, 1964.
Shulga, S. A., Sudol, D. E., and Nishino, F., Influence of the mode of initial geometrical imperfec-
tions of the load-carrying capacity of cylindrical shells made of composite materials, Thin-Walled
Structures, Vol. 14, 89–103, 1992.
Simitses, G. J., Buckling and postbuckling of imperfect cylindrical shells, Applied Mechanics Reviews,
Vol. 39, 1517–24, 1986.
Simitses, G. J., and Anastasiadis, J. S., Shear deformable theories for cylindrical laminates equilibrium
and buckling with applications, AIAA Journal, Vol. 30, 826–34, 1992.
Simitses, G. J., Shaw, D., and Sheinman, I., Stability of imperfect laminated cylinders: A comparison
between theory and experiments, AIAA Journal, Vol. 23, 1086–92, 1985.
Singer, J., The influence of stiffener geometry and spacing on the buckling of axially compressed
cylindrical and conical shells, Theory of Thin Shells (F. I. Niordson, ed.), Springer Verlag, Berlin,
pp. 234–63, 1969.
Singer, J., Buckling experiments on shells-a review of recent developments, Solid Mechanics Archives,
Vol. 7, 213–313, 1982.
Singer, J., and Rosen, A., The influence of boundary conditions of the buckling of stiffened cylindrical
shells, Buckling of Structures (B. Budiansky, ed.), Springer Verlag, Berlin, pp. 227–50, 1976.
Singer, J., Abramovich, H., and Yaffe, R., Initial imperfection measurements of integrally stringer-
stiffened cylindrical shells, TAE Report 330, Technion–Israel Institute of Technology, Department
of Aeronautical Engineering, Haifa, Israel, 1978.
Singer, J., Abramrovich, H., and Yaffe, R., Initial imperfection measurements of stiffened shells and
buckling predictions, Israel Journal of Technology, Vol. 17, 324–38, 1979.
Singer J., Arbocz, J., and Weller, T., Buckling Experiments: Experimental Methods in Buckling of
Thin-Walled Structures, Wiley, New York, 1997.
Singh, M. P., Analysis and reliability-based design of columns with random parameters, PhD Thesis,
Department of Civil Engineering, University of Illinois at Urbana, 1972.
BIBLIOGRAPHY 321
Singh, M. P., and Ang, A. H.-S., Columns with random imperfections and associated bracing require-
ments, Journal of Structural Mechanics, Vol. 4, 161–80, 1976.
Slaughter, W. S., and Fleck, N. A., Microbuckling of fiber composites with random initial fiber
waviness, Journal of Mechanics of Physics and Solids, Vol. 42, 1743–66, 1994.
Slezinger, I. N., and Barskaya, S. Ya., On influence of random form imperfections on stability of
shallow double-curvative panel, Dynamics and Strength of Machines, Kharkov, pp. 96–104, 1977
(in Russian).
Sobel, L. H., Effect of boundary conditions on the stability of cylinders subjected to lateral and axial
pressure, AIAA Journal, Vol. 2, 1437–40, 1964.
Southwell, R. V., On the collapse of tubes by external pressure, Philosophical Magazine, Part I, Vol. 25,
687–8, 1913; Part II, Vol. 26, 502–11, 1913; Part III, Vol. 29, 66–7, 1915.
Spencer, H. H., Are measurements of geometric imperfections of plates and shells useful?, Experi-
mental Mechanics, Vol. 18, 107–11, 1978.
Stam, A. R., Stability of imperfect cylindrical shells with random properties, Paper AIAA-96-1462-
CP, 37th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference,
Salt Lake City, 1307–15, 1996.
Stam, A., Experimental Verification of Probabilistic Design Analysis, Paper No. AIAA-2000-1511,
AIAA Press, Washington, D. C., 2000.
Stam, A. R., and Arbocz, J., Stability analysis of anisotropic cylindrical shells with random layer
properties, Reliability in Nonlinear Structural Mechanics, (O. D. Ditlevsen and J. C. Mitteau, eds.),
Euromech-372, University of Blaise Pascal, Clermon-Ferrand, pp. 77–84, 1997.
Starnes, Jr., J. H., and Haftka, R. T., Preliminary design of composite wings for buckling, strength
and displacement constraints, Journal of Aircraft, Vol. 16, 564–70, 1979.
Starnes, Jr., J. H., Kright, Jr., N. F., and Rouse, M., Postbuckling behavior of selected flat stiffened
graphite-epoxy panels loaded in compression, AIAA Journal, Vol. 23, 1236–46, 1985.
Stavsky, Y., and Friedland, S., Stability of heterogeneous orthotropic cylindrical shells in axial com-
pression, Israel Journal of Technology, Vol. 7, 111–19, 1969.
Stavsky, Y., and Hoff, N. J., Mechanics of composite structures, Composite Engineering Laminates
(A. G. H. Dietz, ed.), MIT Press, Cambridge, pp. 5–59, 1969.
Stein, M., Some recent advances in the investigation of shell buckling, AIAA Journal, Vol. 6, 2239–45,
1968.
Strating, J., Probabilistic theory and buckling curves, Report 6-70-8, CEACM 8-1, Stewin-
Laboratorium, Delft University of Technology, Delft, the Netherlands, 1970.
Strating, J., and Vas, H., Computer simulation of the R.C.C.S. buckling curve using a Monte-Carlo
method, Heron, Vol. 19, 1–25, 1973.
Streletskii, N. S., Osnovy Statisticheskogo Ucheta Koeffizienta Zapasa Prochnosti Sooruzhenii (Ba-
sics of Statistical Analysis of Safety Factor of Structures), “Stroiizdat” Publishers, Moscow, 1947
(in Russian).
Stroud, W. J., and Anderson, M. S., PASCO – Structural panel analysis and sizing code, capability
and analytical foundations, NASA TM-80181, 1981.
Stroud, W. J., Davis, Jr., D. D., Maring, L. D., Krishnamurthy, T., and Elishakoff, I., Reliability of
stiffened structural panels: Two examples, Reliability Technology (T. A. Cruse, ed.), ASME Press,
New York, pp. 196–216, 1992.
Stroud, W. J., Krishnamurthy, T., Sykes, N. P., and Elishakoff, I., Effect of bow-type initial imperfection
on the reliability of minimum-weight stiffened structural panels, NASA TN-3263, 1993.
Swamidas, A. S. J., and Kunukkasseril, V. X., Buckling of orthotropic circular plates, AIAA Journal,
Vol. 11, 1633–36, 1973.
Takaoka, N., and Shiraki, W., Probabilistic design method of elastic compression members,
Reports of the Faculty of Engineering, Tottori University, Vol. 10, No. 1, 140–52, 1979
(in Japanese).
Takaoka, N., Shiraki, W., and Fujiwara, T., Reliability analysis of H-shaped compression members
with random initial imperfections, Reports of the Faculty of Engineering, Tottori University, Vol. 11,
No. 1, 134–48, 1980 (in Japanese).
322 BIBLIOGRAPHY
Talaslidis, D., Anwendung statistischer Methoden auf das Stabilitatsproblem “ungenauigkeitsemfind-

licher” Schalen, Konstruktver Ingenieurbau-Berichte (E. H. Zerna, ed.), Vulkan Verlag, Essen,
pp. 151–63, 1976 (in German).
Tasi, J., Effect of heterogeneity on the stability of cylindrical shells under axial compression, AIAA
Journal, Vol. 4, 1058–62, 1966.
Tatsa, E., Tene, Y., and Baruch, M., Remarks on the application of Koiter’s general post-buckling
theory, Acta Mechanics, Vol. 22, 153–62, 1975.
Temple, S. G., and Bickley, B. G., Rayleigh’s Principle and Its Applications to Engineering, Oxford
University Press, New York, 1933 (reprinted by Dover, New York, 1956).
Tenerelli, D. J., and Horton, W. M., Theoretical and experimental study of the local buckling of ring-
stiffened cylinders subject to axial compression, Israel Journal of Technology, Vol. 7, 181–94, 1969.
Tennyson, R. C., Buckling of laminated composite cylinders: a review, Composites, Vol. 6, 17–24,
1975.
Tennyson, R. C., Interaction of cylindrical shell buckling experiments with theory, Theory of Shells,
North-Holland, Amsterdam, pp. 65–116, 1980.
Tennyson, R. C., Buckling of composite cylinders under axial compression, Developments in Engi-
neering Mechanics (A. P. S. Salvadurai, ed.), Elsevier Science Publishers, Amsterdam, pp. 229–58,
1987.
Tennyson, R. C., and Muggeridge, D. B., Buckling of axisymmetric imperfect cylindrical shells under
axial compression, AIAA Journal, Vol. 7, 2127–31, 1969.
Tennyson, R. C., and Muggeridge, D. B., Buckling of laminated anisotropic imperfect circular cylin-
ders under axial compression, AIAA 10th Aerospace Sciences Meeting, AIAA Paper No. 72-139,
1972.
Tennyson, R. C., Chan, M. K., and Muggeridge, D. B., The effect of axisymmetric shape imperfections
on the buckling of laminated anisotropic circular cylinders, Canadian Aeronautics and Space,
Transactions of the Institute, Vol. 4, 131–8, 1971.
Tennyson, R. C., Muggeridge, D. B., and Caswell, R. D., Buckling of circular cylindrical shells having
axisymmetric imperfections distributions, AIAA Journal, Vol. 9, 924–30, 1971.
Tewary, V. K., Mechanics of Fiber Composites, John Wiley and Sons, New York, pp. 108–21, 1978.
Thangjitham, S., and Rantis, T. D., Probability-based buckling instability analysis of a laminated
composite plate, Proceedings of 33th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynam-
ics and Materials Conference, AIAA Press, New York, pp. 454–62, 1992.
Thoft-Christensen, P., and Baker, M. J., Structural Reliability Theory and Its Applications, Springer,
Berlin, 1982.
Thompson, J. M. T., Toward a general statistical theory of imperfection-sensitivity in elastic post-
buckling, Journal of Mechanics, Physics and Solids, Vol. 15, 413–17, 1967.
Thompson, J. M. T., and Hunt, G. W., A General Theory of Elastic Stability, Wiley, London, 1973.
Thomson, W. T., Theory of Vibration with Applications, 2nd ed., Prentice Hall, Englewood Cliffs, NJ,
pp. 30–4, 1981.
Tichy, M., and Norlicek, M., Safety and eccentrically loaded reinforced concrete columns, Journal of
Structural Division, Vol. 88, 1–10, 1962.
Timashev, S. A., Statistical investigation of stability of thin convex orthotropic shells, Reliability
Problems in Structural Mechanics (V. V. Bolotin and A. Cyras, eds.), “RINTIP” Publishing House,
Vilnius, pp. 157–64, 1968 (in Russian).
Timashev, S. A., Stochastic stability and reliability of stiffened shells, ScD Dissertation, Leningrad
Civil Engineering Institute, 1975 (in Russian).
Timashev, S. A., and Kantor, S. L., Stochastic stability and reliability of stiffened shells, International
Conference on Light-Weight Spacewise Structures in Concentional and Seismic Regions, Alma-Ata,
pp. 140–5, 1977 (in Russian).
Timoshenko, S. P., Einige Stabilitätsprobleme der Elastizitätstheorie, Zeitschrift für angewandte Math-
ematik und Physik, Vol. 58, 337–85, 1910 (in German).
Timoshenko, S. P., and Gere, J. M., Theory of Elastic Stability, 2nd ed., McGraw-Hill, New York,
pp. 94–8, 1961.
BIBLIOGRAPHY 323
Trendafilova, I., and Ivanova, J., Loss of stability of thin, elastic, strongly convex shells of revolution
with initial imperfections, subjected to uniform pressure. A probabilistic approach, Thin-Walled
Structures, Vol. 23, 201–14, 1995.
Triantafylidis, N., and Peek, R., On stability and the worst imperfection shape in solids with nearby
simultaneous eigenmodes, International Journal of Solids and Structures, Vol. 29, 2281–99, 1992.
Troshenkov, M. K., Application of variational methods in nonlinear stochastic problems of dynamics
and statics of shells, PhD Dissertation, Novosibirsk Electrotechnic Institute, 1976 (in Russian).
Tsai, C. T., and Palazotto, A. N., A modified Riks approach to composite shell snapping using a
high-order shear deformation theory, Computers and Structures, Vol. 35, 221–6, 1990.
Tvergaard, V., Buckling Behavior of Plate and Shell Structure, Theoretical and Applied Mechanics
(W. T. Koiter, ed.), North-Holland, Amsterdam, pp. 233–47, 1976.
Tvergaard, V., and Needleman, A., On the localization of buckling pattern, Journal of Applied Me-
chanics, Vol. 50, 935–40, 1980.
Tvergaard, V., and Needleman, A., On the development of localized buckling patterns, Collapse:
The Buckling of Structures in Theory and Practice, (J. M. T. Thompson and G. W. Hunt, eds.),
Cambridge University Press, Cambridge, pp. 299–332, 1983.
Uthgenannt, E. B., and Brand, R. S., Buckling of orthotropic annular plates, AIAA Journal, Vol. 8,
2102–4, 1970.
Vainberg, M. M., and Trenogin, V. A., Theory of Branching of Solution of Nonlinear Equations,
Noordhoff International Publishers, Leyden, 1974.
Valishvili, N. V., Stability of spherical shells under finite displacements, Mechanics of Deformable
Solids and Structures, volume dedicated to 60th anniversary of Ju. N. Rabotnov, “Mashinostroenie”
Publishing House, Moscow, pp. 92–8, 1975 (in Russian).
Van Eekelen, A. J., Computation module for the buckling and postbuckling behavior of a cylindri-
cal shell with a two-mode imperfection, Memorandum LR 773, Delft University of Technology,
Department of Aerospace Engineering, Delft, the Netherlands, 1994.
Van Slooten, R. A., and Soong, T. T., Buckling of a long, axial compressed, thin cylindrical shell with
random initial imperfection, Journal of Applied Mechanics, Vol. 38, 1066–71, 1972.
Vanin, G. A., and Semeniuk, N. P., Stability of Shells Made of Composite Materials with Imperfections,
“Naukova Dumka” Publishers, Kiev, 1978 (in Russian).
Vanin, G. A., Semeniuk, N. P., and Emelianov, R. F., Stability of Shells of Composite Materials,
“Naukova Dumka” Publishers, Kiev, 1978 (in Russian).
Vanmarcke, E., Shinozuka, M., Nakagiri, S., Schuëller, G., and Grigoriu, M., Random fields and
stochastic finite-elements, Structural Safety, Vol. 3, 143–66, 1986.
Vásárhelyi, A., Collapse load analysis of panel structures by stochastic programming, Periodica
Politechnica, Vol. 25, No. 314, 243–50, 1981.
Vasiliev, V. V., Mechanics of Composite Structures, Taylor and Francis, Washington, DC, 1993.
Vassalos, D., Turan, O., and Pawlowski, M., Dynamic stability assessment of damaged passenger/ro-ro
ships and proposal of rational survival criteria, Marine Technology, Vol. 34, 241–66, 1997.
Verduijn, W. D., and Elishakoff, I., A testing machine for statistical analysis of small imperfect shells,
Netherlands, 1982.
Verijenko V. E., Adali, S., Summers, E. B., and Reiss, T., Optimal design of laminated shells with shear
and normal deformation using symbolic Computation, AIAA-94-4428-CP, AIAA Press, pp. 1066–
71, 1994.
Videc, B. P., and Sanders, J. L., Application of Khas’ minskii limit theorem to the buckling problem
of a column with random initial deflections, Quarterly of Applied Mathematics, Vol. 33, 422–8,
1976.
Vijayakumar, K., and Joga Rao, C. V., Buckling of polar orthotropic annular plates, Journal of Engi-
neering Mechanics Division, Vol. EM3, 701–10, 1971.
Vinson, J. R., and Sierakowski, R. L., The Behavior of Structures Composed of Composite Materials,
Martinus Nijhoff Publishers, Dortrecht, 1986.
Volkov, S. D., Statistical Strength Theory, Gordon and Breach, New York, 1962.
324 BIBLIOGRAPHY
Volkov, S. D., and Stavrov, V. P., Statistical Mechanics of Composite Materials, Belorussian State
University, Minsk, 1978 (in Russian).
Volkov, S. I., Branching of probability functionals’ extremals and stability of nonlinear elastic column
with random imperfections, Technical Report, Novosikirsk Electrotechnic Institute, Novosikirsk,
1978 (in Russian).
Volkov, S. I., An example of stochastic bifurcation in the theory of bending of non-ideal plates,
Prikladnaya Matematika i Mekhanika, Vol. 45, 876–83, 1981 (in Russian).
Volmir, A. S., Biegsame Platten und Schalen, Verlag für Bauwsen, Berlin, 1962 (in German).
Volmir, A. S., Stability of Elastic Systems, Wright-Patterson Air Force Base, Dayton, OH, FTD-MT
64-335, 1964, and NASA AD-628508, 1965.
Volmir, A. S., and Kildibekov, I. G., Probabilistic characteristics of behavior of cylindrical shells
subjected to acoustic loading, Prikladnaya Mekhanika, Vol. 1, 1–9, 1965 (in Russian).
Volmir, A. S., and Kildibekov, I. G., Probabilistic characteristics of behavior of shallow shells in an
acoustic field, Reliability Problems in Structural Mechanics (V. V. Bolotin and A. A. Cyras, eds.),
“RINTIP” Publishing House, Vilnius, pp. 169–74, 1968a (in Russian).
Volmir, A. S., and Kildibekov, I. G., Nonlinear vibrations and buckling of a shallow shell in acoustic
field, Reliability Problems in Structural Mechanics (V. V. Bolotin and A. Cyras, eds.), pp. 169–74,
1968b (in Russian).
Volosovich, O. V., and Timashev, S. A., Initial imperfections and buckling modes of rectangular convex
shells, Theory of Shells and Plates, Proceedings of All-Union Conference, “Nauka” Publishing
House, Rostov-on-Don, pp. 254–8, 1973 (in Russian).
Von Slooten, R. A., and Soong, T. T., Buckling of a long, axially compressed, thin cylindrical shell
with random initial imperfections, Journal of Applied Mechanics, 1066–71, 1972.
Vorovitch, I. I., Statistical method in the theory of stability of shells, Prikladnaya Matematika i
Mekhanika, Vol. 23, 885–92, 1959 (in Russian).
Vorovich, I. I., Some problems of applications of statistical methods in the theory of stability of plates
and shells, Theory of Plates and Shells, Academy of Armenia Press, Erevan, pp. 69–94, 1964
(in Russian).
Vorovich, I. I., Paths of developments of problem of stability in the theory of shells, Actual Problems
of Science, Rostov University Press, Rostov-na-Donu, pp. 111–26, 1967 (in Russian).
Vorovich, I. I., and Minakova, N. I., Stability problem and numerical methods in the theory of spherical
shells, Mechanics of Solid Deformable Bodies, Vol. 7, “VINITI” Publishers, pp. 1–86, 19732 (in
Russian).
Vorovich, I. I., and Zipalova, V. F., Towards solution of nonlinear boundary-value problems of elasticity
by transfer to the Cauchy problems, Prikladnaya Matematika i Mekhanika, Vol. 29, 894–901, 1965
(in Russian).
Wah, T., and Calcote, L. R., Structural Analysis by Finite Difference Calculus, Van Nostrand Reinhold,
New York, 1970.
Wang, D., Zhang, S., and Yang, G., Improved Monte Carlo method and its application to structural
reliability, Acta Mechanica Solida Sinica, Vol. 7, 256–64, 1994.
Wang, I-Y., Monte Carlo analysis of nonlinear vibration of rectangular plates with random geometric
imperfections, International Journal of Solids and Structures, Vol. 26, 99–109, 1990.
Wang, L. R. L., and Rodriguez-Agrait, L., Statistical tests on structural models for the buckling
behavior of spherical shells, Statistics and Probability in Civil Engineering (P. Lumb, ed.), Hong
Kong University Press, Hong Kong, pp. 489–512, 1971.
Waszczyszyn, Z., Cichon, C., and Radwanska, M., Stability of Structures by Finite Element Methods,
Elsevier Science Publishers, Amsterdam, 1994.
Weingarten, V. I., Morgan, E. J., and Seide, P., Elastic stability of thin-walled cylindrical and conical
shells under axial compression, AIAA Journal, Vol. 3, 500–5, 1965.
Weingarten, V. I., Seide, P., and Peterson, J. P., Buckling of thin-walled structures, NASA SP-8007,
1968.
Wentzel, E. S., Operations Research, “Mir” Publishers, Moscow, 1988.
Whitney, J. M., Structural Analysis of Laminated Anisotropic Plates, Technomic Publishing Company,
Lancaster, PA, pp. 151–6, 1987.
BIBLIOGRAPHY 325
Williams, F. W., Development history of efficient algorithms for finding roots with certainty in dynamic
stiffness methods, Proceedings of Second ESA Workshop on Modal Representation of Flexible
Structures by Continuum Methods, ESA WPP-034 European Space Agency, Noordwijk, pp. 75–
94, 1992.
Wirsching, P. H., and Yao, J. T. P., Random behavior of columns, Journal of the Engineers Mechanics
Division, Vol. 97, 605–18, 1971.
Wittrick, W. H., and Williams, F. W., An algorithm for computing critical buckling loads of elastic
structures, Journal of Structural Mechanics, Vol. 1, 497–518, 1973.
Wohlever, J. C., and Healey, T. J., A group theoretic approach to the golbal bifurcation analysis of an
axially compressed cylindrical shell, Computer Methods in Applied Mechanics and Engineering,
Vol. 122, 315–50, 1995.
Woinowsky-Krieger, S., Ueber die Beulsicherheit von Kreisplatten mit kreiszylindrischer Aelotropie,
Ingenieur Archiev, Vol. 26, 129–31, 1958.
Wolfram, S., The MathematicaR Book, 3rd ed., Cambridge University Press, 1996.
Wooff, G., and Hodgkinson, D., muMATH: A Microcomputer Algebra System, Academic Press,
London, 1987.
Wriggers, P., and Simo, J. C., A general procedure for the direct computation of turning and bifurcation
points, International Journal of Numerical Methods in Engineering, Vol. 30, 155–67, 1990.
Wu, C. H., Buckling of anisotropic circular cylindrical shells, PhD Thesis, Case Western Reserve
University, 1971.
Wu, Y. T., and Burnside, D. M., Computational methods for probability of instability calculations,
Proceedings, AIAA/ASME/ASCE/AHS/ASC 31st SDM Conference, Structures, Structural Dynamics
and Materials, Long Beach, 1990.
Wu, Y. T., and Burnside, O. H., Computational methods for probability of instability calculations,
Proceedings of the 31st Structures, Structural Dynamics and Materials Conference, AIAA Press,
Washington, DC, pp. 1081–91, 1990.
Wunderlich, W., Rensch, H. J., and Obrecht, H., Analysis of elastic-plastic buckling and imperfection-
sensitivity of shells of revolution, Buckling of Shells: Proceedings of a State-of-the-Art Colloquium
(E. Ramm, ed.), Springer Verlag, Berlin, pp. 137–74, 1982.
Xie, W. C., Buckling mode localization in randomly disordered multispan continuous beam, AIAA
Journal, Vol. 33, 1142–9, 1995.
Xie, W. C., Buckling mode localization in rib-stiffened plates with randomly misplaced stiffness,
Xie, W. C., Buckling mode localization in nonhomogeneous beams on elastic foundation, Chaos,
Solitons and Fractals, Vol. 8, 411–31, 1997.
Xie, W. C., Buckling mode localization in rib-stiffened plates with randomly misplaced stiffener,
Xie, W. C., and Elishakoff, I., Buckling mode localization in rib-stiffened plates with mis-
placed stiffeners-Kantorovich approach, Chaos, Solitons and Fractals, Vol. 11 (10), 1559–1574,
2000.
Yadav, D., and Verma, N., Buckling of composite circular cylindrical shells with random material
properties, Composite Structures, Vol. 37, 385–91, 1997.
Yaffe, R., Singer, J., and Abramovich, H., Further initial imperfection measurements of integrally
stringer-stiffened cylindrical shells-series 2, TAE Report 404, Department of Aeronautical Engi-
neering, Technion – Israel Institute of Technology, 1981.
Yamaki, N., Buckling of a thin annular plate under uniform compression, Journal Applied Mechanics,
Vol. 25, 267–72, 1958.
Yamaki, N., Elastic Stability of Circular Cylindrical Shells, North-Holland, New York, 1984.
Yamaki, N., and Kodama, S., Postbuckling behavior of circular cylindrical shells under compression,
International Journal of Non-Linear Mechanics, Vol. 11, 99–111, 1976.
Yeigh B. W., Imperfections and instabilities, PhD Dissertation, Department of Civil Engineering and
Operations Research, Princeton University, 1995.
Yamazaki, F., and Shinozuka, M., Neumann expansion for stochastic finite element analysis, Journal
326 BIBLIOGRAPHY
Yeigh B. W., Stochastic modeling of imperfections in beams, Probabilistic Mechanics and Struc-
tural Reliability (D. M. Frangopol and M. D. Grigoriu, eds.), ASCE Press, New York, pp. 676–9,
1996.
Yeigh, B. W., and Shinozuka, M., Uncertainty modeling in structural stability, Uncertainty Modeling
in Finite Element, Fatigue and Stability of Systems (A. Haldar, A. Guran, and B. M. Ayyub, eds.),
World Scientific, Singapore, pp. 215–60, 1997.
Yoakimidis, N. I., Symbolic computations for the solution of inverse/design problems with UAPLE,
Yushanov, S. P., Probabilistic model of layerwise collapse of a composite and analysis of reliability of
layered cylindrical shells, Mechanics of Composite Materials, No. 4, 642–53, 1985 (in Russian).
Yushanov, S. P., and Bogdanovich, A. E., Method of analysis of reliability of layered cylindrical shells
with allowance for scatter in the strength characteristics of the composite, Mechanics of Composite
Materials, Vol. 22, 725–30, 1987 (in Russian).
Yusupov, A. K., Stability of compressed column during random rotation of the support, Stroitelnaya
Mekhanika i Raschet Sooruzhenii, No. 1, 38–41, 1976 (in Russian).
Zadeh, L. A., Fuzzy sets, Information and Control, Vol. 8, 338–53, 1965.
Zhang, J., and Ellingwood, B., Orthogonal series expansions of random fields in reliability analysis,
Journal of Engineering Mechanics, Vol. 120, 2660–77, 1994.
Zhang, J., and Ellingwood, B., Orthogonal series representations of random fields in stochastic FEA,
Computational Stochastic Mechanics (P. D. Spanos, ed.), Balkema Publishers, Rotterdam, pp. 57–
65, 1995a.
Zhang, J., and Ellingwood, B., Effects of uncertain material properties on structural stability, Journal
of Structural Engineering, Vol. 121, 705–16, 1995b.
Zhang, J., and Ellingwood, B., SFE-Based structural stability analysis, Application of Statistics and
Probability (M. Lemaire, J. L. Favre, and A. Mebarki, eds.), Balkema Publishers, Rotterdam,
pp. 1041–6, 1995c.
Zhang, J., and Ellingwood, B., Error measure for reliability studies using reduced variable set, Journal
of Engineering Mechanics, Vol. 121, 935–7, 1995d.
Zhu, L. P., Elishakoff, I., and Lin, Y. K., Free and forced vibrations of periodic multi-span beams,
Shock and Vibration, Vol. 1, 217–32, 1994.
Zhu, L. P., Elishakoff I., and Starnes J. H., Jr., Derivation of multi-dimensional ellipsoidal convex
model for experimental data, Mathematical Computing and Modelling, Vol. 24, 103–14, 1996.
Zhukova, N. B., and Semeniuk, N. P., Timoshenko-type theory in cubic approximation with application
to the initial post-buckling behavior of composite shells, Prikladnaya Mekhanika, Vol. 28, No. 10,
41–6, 1992 (in Russian).
Ziegler, H., Principles of Structural Stability, Blaisdell, Boston, 1968; Birkhauser, Basel, 1977.
Zimmermann, R., Optimierung axial gedrückter CFK-Zylinderschalen, VDI Report Nr. 207, VDI-
Verlag, Düsseldorf, 1992 (in German).
Zingales, M., and Elishakoff, I., Localization of the bending response in presence of axial load,
International Journal of Solids and Structures, (to appear).
Zipalova, V. F., and Nenastieva, V. M., On application of Galerkin method in temperature-force
problems in nonlinear stability of shells, Izvestiya Akademii Nauk SSSR, Mekhanika Tverdogo
Tela, Vol. II, No. 1, 100–3, 1971 (in Russian).
Zucarro, G., Elishakoff, I., and Baratta, A., Antioptimization of earthquake excitation and response,
Mathematical Problems in Engineering, Vol. 4, 1–19, 1998.
Zyczkowski, M., and Gajewski, A., Optimal structural design under stability constraints, in Collapse:
The Buckling of Structures in Theory and Practice, (J. M. T. Thompson and G. W. Hunt, eds.),
Cambridge University Press, Cambridge, pp. 299–332, 1983.
Author Index
Not too many intellectuals have the courage to admit they pick a book and look
first for their name in the index.
I. Howe
Abell, M. L., 274, 291 Belytschko, T., 216, 311

Abramovitch, H., 17, 148, 149, 150, 153, 167, 170, Benaroya, H., 211, 270, 294, 317
174, 179, 183, 194, 291, 292, 320 Ben-Haim, Y., xv, xvii, 223, 229, 230, 239, 240,
Adali, S., 238, 256, 270, 291, 323 242, 294
Almroth, B. O., 1, 99, 187, 193 Bernhart, W., 1, 313
Amazigo, J., xvii, 108, 109, 137, 138, 142, 145, 170, Bert, C. W., xvii, 274, 283, 294
176, 186, 291, 292 Beveridge, G., 238, 294
Ambuster, D., 270, 317 Biderman, V. L., 267, 295
Andersen, C. M., 269, 270, 314 Bleich, F., 2, 17, 295
Anderson, P. W., 2, 292 Blekhman, I. I., 239, 295
Arbocz, J., xiii, xiv, xvii, 71, 124, 137, 138, 139, 143, Bolotin, V. V., xiii, xviii, 108, 186, 240, 293
144, 146, 147, 148, 149, 150, 152, 153, 154, 156, Booton, M., 196, 197, 296
157, 159, 160, 161, 165, 166, 167, 169, 170, 171, Borst, de R., 187, 296
173, 174, 179, 183, 187, 188, 189, 190, 192, 193, Borsuk, K., 261, 276, 313
195, 196, 197, 198, 199, 200, 201, 221, 256, 292, Boyce, W. E., 180, 296
293 Brand, R. S., 261, 322
Ariaratnam, S. T., xvii, 2, 28, 200, 223, 293 Braselton, J. P., 274, 291
Ari-Gur, J., xvii Brenner, C. E., 211, 296
Arnold, L., 28, 293 Brillouin, L., 17, 296
Ayyub, B. M., 131, 293 Brogan, F. A., 187, 193, 291
Brush, D. D., 1, 99, 296
Babcock, C. D., Jr., xiv, xviii, 148, 149, 150, 156, 165, Bucher, Ch., 211, 312
166, 167, 169, 170, 183, 187, 188, 189, 190, 191, Budiansky, B., xii, xviii, 16, 99, 107, 108, 109, 110,
193, 194 113, 124, 128, 137, 142, 147, 221, 242, 244, 295
Bahder, Th. B., 274, 294 Bulgakov, B. V., 240, 296
Baker, M. J., 130, 322 Bulson, P. S., 2, 296
Ball, J. M., 221, 294 Bushnell, D., xvi, xvii, 96, 296
Baratta, A., 175, 293
Baruch, M., xvii Cai, G. Q., xvi, xvii, 2, 128, 129, 241, 242, 245, 246,
Bažant, Z. P., 99, 294 247, 248, 249, 251, 252, 255, 296
Beltzer, A. I., 269, 270, 294 Calcote, L. R., 2, 17, 19, 26, 37, 324
Only those authors who are referenced in the text are listed. The reader may also consult the extensive bibliography
of about 900 references. We tried to cover the subject as completely as we could, but some inadvertent omissions
may have occurred.
327
328 AUTHOR INDEX
Cambou, B., 211, 296 Geer van, J., xvii

Caswell, R. D., 90, 91, 322 Gelfand, M., 267, 302
Cedolin, L., 99, 294 Gere, J. M., 4, 223, 322
Chan, M. K., 71, 137, 138, 322 Ghanem, R., 211, 214, 302
Charmatz, M., xvi, xvii, 52, 62 Godunov, S. K., xvi, 52, 62, 257, 259, 261, 263,
Chat, P. D., 2, 326 267, 274
Chernousko, F. L., 240, 297 Goggin, P. R., 231, 233, 303
Chetaev, N. G., 15, 297 Grandori, G., 240, 303
Chilver, A. M., 99, 318 Grigoliuk, E. I., 52, 258, 259, 267, 303
Chryssanthopoulos, M. K., 221, 297 Grigoriu, M., 176, 304, 323
Colombi, P., 256, 300 Grimaldi, A., 221, 298
Combescure, A., 106 Gusic, G., 106
Como, M., 221, 298
Conte, D., xvi, 52, 62, 257, 259, 261, 263, 274 Haldar, A., 131, 293
Cornell, C. A., 187, 298 Hammersley, J. M., 118, 130, 304
Crandall, S. H., xviii, 241, 279, 298 Hamming, R. W., 257
Cravel, H., 28, 293 Hancock, J. O., 282, 287, 308
Handscomb, D. G., 118, 130, 304
Daventport, J. H., 268, 269, 270, 298 Hansen, J., xviii, 109, 110, 138, 139, 176, 221, 304
Davidenko, D. F., 110, 116, 298 Hart-Smith, L. J., 138, 304
De Yong, R. G., 2, 298 Hasofer, A. M., 239, 300
Deodatis, G., 223, 298 Hawwa, M. A., 2, 41, 314
Der Kiureghian, A., 211, 298 Healey, T. J., 193, 324
Dickey, J., 2, 311 Hettema, Ch. D., 70, 300
Dimaggio, F., 270, 317 Hien, T. D., 211, 308
Dinnik, A. N., 282, 287, 299 Himmelblau, D. M., 285, 305
Donnell, L. H., 108, 166, 171, 189, 299 Hirano, Y., 71, 92, 226, 305
Dowell, E. H., 2, 316 Hisada, T., 211, 314, 323
Drenick, R. F., 240, 299 Hodges, C. H., 2, 270, 298, 305
Hodgkinson, D., 268, 269, 324
Eckmann, J. P., 28, 294 Hoff, N. J., xviii
Einstein, A., 222 Höft, H. F. W., 274, 305
Elishakoff, I., xiv–xvii, 17, 32, 41, 42, 44, 46, 48, Höft, M., 274, 305
52–54, 62, 65, 68, 76, 81, 89, 92, 98, 99, 108–110, Holl, J. M. A. M., 71, 176, 193, 194, 195, 196, 197,
115, 117–121, 123–130, 135–137, 139, 144–148, 198, 199, 200, 201, 221, 293
152–157, 159–161, 165, 166, 170–174, 176, Horne, M. R., 20, 201, 305
179–182, 184–187, 190, 192, 193, 196, 199, 210, Horton, W. H., xiii
211, 212, 223, 229, 230, 238–242, 245–249, Hulbert, G., 270, 315
251–256, 270, 271, 291, 294, 301, 315, 326 Hunt, G. W., xvii, 136, 322
El-Naschie, M. S., 2, 301 Hussain, M. A., 270, 305
Ellingwood, B., xvi, xvii, 212–214, 217–219, 220, Hutchinson, J., xii, 16, 99, 107, 108, 124
223, 326
Euripides, 275 Ikeda, M., xvi, xvii, 99, 176, 177, 179, 181, 182,
Euler, L., xii 184, 185
Imbert, K., 166, 169, 189, 306
Fenves, S. Y., 270, 309 Indenbaum, V. M., 259, 315
Fersht, R. S., 109, 139, 301 Itzhak, E., xvii
Finetti de, B., 222
Frazer, B., xviii, 109, 110, 113, 128, 242, 244, 245 Jensen, J., 268, 270, 306
Freudenthal, A. M., 239, 240, 302 Joga Rao, C. V., 261, 264, 323
Friedland, S., 71, 321 Jones, R. M., 224, 306
Frolov, A. N., 52, 258, 259, 303 Jullien, J. F., 106
Furstenberg, H., 201, 202, 302
Kac, U., 211, 307
Gajewski, A., 16, 326 Kahrimanian, H. G., 270, 307
Gan, H. L., 131, 310 Kaplyevatsky, I., xviii
Ganesan, R., 223, 317 Karadeniz, H., 188, 307
Gaonkar, G., 270, 314 Karhunen, K., 211, 213, 214, 221, 307
AUTHOR INDEX 329
Keener, J. P., 109, 307 Nayfeh, A. H., 2, 41, 314

Kelton, W., 132, 309 Neal, D. M., 239, 314
Kendall, K., 182, 307 Needleman, A., 2, 322
Khot, N. S., 71, 88, 307 Niordson, F., 268, 270, 306
Kirsch, U., 238, 307 Noble, B., 270, 305
Kleiber, M., 211, 269, 308 Nolan, J., 270, 314
Klimov, D. M., 270, 308 Noor, A. K., 270, 289, 314, 315
Kline, L. V., 282, 287, 308 Novichkov, Y. N., 259, 315
Klompé, A. W. H., 174, 308
Koiter, W. T., xv, xvii, 16, 43, 44, 46, 48, 50, 52, 53, Oceledec, Y. I., 203, 315
54, 55, 57, 61, 62, 66, 68, 70, 81, 89, 90, 92, 98, 99,
107, 108, 124, 138, 154, 155, 176, 177, 183, 221, Palassoupulos, G., xviii, 315
308, 309 Pandalai, K. A. V., 261, 276, 315
Korncoff, A. R., 270, 309 Panovko, Ya. G., 239, 295
Krein, M. G., 17, 309 Patel, S. A., 261, 315
Krenk, S., xviii, 120 Patton, C. M., 270, 315
Kusters, G., 187, 296 Pavlovic, M. N., 269, 316
Pedersen, P., 270, 316
Laplace, P., 175 Peterson, M. T., 193, 291
Law, A., 132, 309 Pierre, C., 2, 26, 28, 42, 316
Leizerakh, V. M., xviii, 310, 312 Plaut, R. H., 2, 26, 42, 316
Li, Y. W., 32, 41, 44, 46, 48, 53, 54, 65, 68, 76, 81, 89, Pletner, B., 271, 300
92, 98, 99, 127, 131, 135, 238, 301, 310 Potier-Ferry, M., 221, 316
Libai, A., xvii
Liew, K. M., 2, 320 Qiria, V. S., 110, 116, 316
Lin, Y. K., 2, 17, 230, 240, 296, 301, 310, 326
Lipovtsev, Yu. V., 259, 266, 267, 303, 311 Ramaiah, G. K., 261, 265, 267, 316
Liu, W. K., 216, 311 Ramu, S. A., 223, 317
Loeve, M., 211, 213, 214, 221, 311 Rand, R. H., 270, 317
Lokutsiewski, O. V., 267, 302 Rayna, G., 270, 317
Reddy, T. S. R., 270, 314
Maidanik, G., 2, 311 Rehak, M., 211, 270, 294, 317
Makarov, B. P., xviii, 108, 138, 139, 186 Reiss, T., 270, 323
Makushin, V. M., 287, 288, 312 Reissner, E., 109, 317
Manen, S., van, xvii, 186, 188, 190, 192, 193, 196, 307 Ren, Y. J., 211, 301
Mani, A., 216, 311 Reyer, P. C. den, 174, 308
Markus, S., xvii Richter, A., 238, 256, 291
Maslov, A., xvi Rikards, R. B., 71, 317
Massey, F. J., Jr., 121, 146, 312 Rizzi, N., 271, 317
Mathews, W. T., 239, 314 Roberts, S. M., 62, 318
Matties, H. G., 211, 312 Roorda, J., 99, 108, 110, 127, 130, 138, 139, 176,
McFarland, D., 1, 313 304, 318
Mead, D. J., 17, 313 Rubinstein, R., 118, 131, 318
Merchant, W., 20, 201, 305 Rudenko, V. M., 270, 308
Miles, J. W., 17, 313 Rzhanitsin, A. R., 187, 318
Miller, E., 193, 291
Morgan, E. J., 138, 324 Sanders, J. L., Jr., 96, 318
Mossakowsky, J., 261, 276, 313 Sapountzakis, E. J., 269, 316
Muggeridge, D. B., 71, 90, 91, 137, 138, 322 Schechter, R., 238, 294
Murota, K., xvi, xvii, 99, 176, 177, 179, 181, 182, Schenk, C. A., 221, 319
184, 185 Schmidt, R., 274, 279, 285, 318
Myachenkov, V. I., 52, 258, 259 Schuëller, G. I., 221, 319
Myshkis, A. D., 239, 295 Schweppe, F. C., 240, 319
Sechler, E. E., 43, 170, 175
Nagabhushnan, J., 270, 314 Seide, P., 138, 324
Nakagiri, S., 211, 314, 323 Semeniuk, N. P., 71, 323
Naso, P. O., 222 Sen-Gupta, G., 17, 319
Nauta, P., 187, 296 Shaw, D., 71, 320
330 AUTHOR INDEX
Sheinman, I., xviii, 71, 320 Vanmarke, E., 211, 323

Shilkrut, D., xviii Vasiliev, V. V., 71, 323
Shinozuka, M., xvii, 127, 129, 135, 211, 223, 238, Venkayya, V. B., 71, 307
240, 298, 323, 325 Verijenko, V. E., 238, 256, 270, 291, 323
Shipman, J. C., 62, 318 Vermeulen, P. G., xvii, 186, 190, 192, 193,
Shreider, Ya. A., 118, 320 196, 301
Siegert, A. J. F., 211, 307 Vijayakumar, K., 261, 264, 265, 267, 316, 323
Sierakowski, R. L., 71, 91, 223, 323 Vinson, J. R., 71, 91, 223, 323
Simitses, G. J., 71, 241, 242, 245, 246, 320 Vrouwenvelder, A., 188, 307
Singer, J., xvi, xvii, 148, 174, 176, 183, 231, 256, 320
Siret, N., 268, 269, 270, 298 Wah, T., 2, 17, 19, 26, 37, 324
Smith, B., 1, 323 Wan, C. C., 108, 299
Soyer, W. W., 222 Wang, C. M., 2, 310
Spanos, P., 211, 214, 302 Weingarten, V. I., 138, 324
Starnes, J. H., Jr., xvii, 32, 41, 44, 46, 48, 53, 54, 65, Wentzel, E. S., 239, 254, 324
68, 76, 81, 89, 92, 98, 99, 127, 128, 129, 135, 223, Whitney, J. M., 223, 228, 324
238, 240, 247, 248, 249, 251, 252, 256, 298, 300, Williams, J. G., xiii, 143, 293
301, 310, 321 Wilson, E. L., 270, 300
Stuart, A., 182, 307 Witte, F. de, 187, 296
Summers, E. B., 270, 323 Wohlever, J. C., 193, 324
Swamidas, A. S. J., 261, 321 Woinowsky-Krieger, S., 261, 276, 280, 324
Wolfram, S., 33, 48, 81, 270, 274, 324
Tang, J., xvi, 271, 300 Woodhouse, J., 305
Tasi, J., 71, 75, 92, 321 Woof, C., 268, 269, 324
Tatone, A., 271, 317
Tennyson, R. C., xviii, 71, 90, 91, 137, 138, Xie, W.-C., xvi, xvii, 2, 28, 200, 204, 205, 206, 207,
223, 322 208, 210, 293
Teters, G. A., 71, 317
Tewary, V. K., 223, 322 Yaffe, R., 148, 176, 183, 320
Thoft-Christensen, P., 130, 322 Yamaki, N., 263, 267, 325
Thompson, J. M. T., xii, 136, 322 Yamazaki, F., 211, 325
Thomson, W. T., 28, 322 Yoakimidis, N. I., 270, 325
Timoshenko, S. P., 4, 223, 322
Tournier, E., 268, 269, 270, 298 Zadeh, L., 240, 326
Tvergaard, V., 2, 107, 322 Zelle, F., 193, 291
Zhang, J., xvi, xvii, 211, 212, 213, 214, 217, 218, 219,
Uthgennant, E. B., 261, 322 220, 223, 298, 326
Zhu, L. P., xvii, 17, 230, 240, 301, 326
Valishvili, N. V., 259, 323 Ziegler, H., 244, 326
Vangel, M. G., 239, 314 Zingales, M., xvii, 42, 326
Vanin, G. A., 71, 323 Zyczkowski, M., 16, 326
Subject Index
algebra bending strain energy, 89

computerized, xvi, 33, 48, 81 Bessel function, 261
Airy stress function, 45, 54, 58, 64, 74 bifurcation, 43
analysis bilinear term, 54
asymptotic, xvi, 175, 177 Boobnov-Galerkin method, 48, 64, 81, 113, 121,
interval, 240 123, 180, 186, 191, 242
multi-mode, 165 BOSOR4 code, xv, 96, 97
probabilistic, xii, 130 boundary conditions, 4, 19, 227
uncertainty, xii, 222, 230, 234, 239 box, multi-dimensional, 241
angle buckling, 94
lamination, 97 dynamic, 144
optimal, 256 load, xii, xvi, 14, 16, 25, 40, 103, 175
rotation, 18, 100 local, 16
approach localization, 2, 200
deterministic, xii modal, 244
probabilistic, xii, 130 nonlinear, xvi
approximation strength, 10
first-order, 66 buckling load
auto-correlation function, 109, 139, 180, 218, 249 extremal, 228, 230
exponential-cosine, 146 greatest, 229, 234, 237
average ensemble, 165 lower bound, 222
axial compression, 54 lowest, 229, 234, 237, 252
uniform, 60 mean, 144, 146, 156, 171, 191, 194, 197
axial load, uniform, 45 upper bound, 222
axial stress resultant, 54 buckling load reduction, 39, 77, 79, 105
critical, 85, 177
beam design, 146
clamped, 212 buckling mode, 1, 11, 13, 36
-column, 214 axisymmetric, 79, 81, 94, 95, 137, 138, 143,
continuous, 200 157, 160
disordered, 26, 27 classical, 146
multi-span, 200, 206
periodic, 26, 36 case
11-span, 28 benchmark, 159
100-span, 27 worst, xii
400-span, 27 central-limit theorem, 133
behavior Cholesky procedure, 118, 120, 144, 165
bifurcation, 184 classical buckling load, 71, 75, 82, 94, 111, 180, 191,
initial postbuckling stochastic, 107, 187 226, 227
bending moment, 4, 18, 223 classical buckling mode, 1, 11, 13, 36, 53, 90
331
332 SUBJECT INDEX
code second-order probability, 140

ALTRAN, 270 spectral, 139, 146
BOSOR4, xv, 96, 97 deviation, standard, 160
CALSYF, 270 DIANA code, 187
DERIVE, 270 difference, central, 50, 77
DIANA, 187 Dinnik’s formula, 282, 287
FORMAC, 270 disorder
MACSYMA, 270, 271, 272, 273, 274, 275 beneficial, 29
MAPLE, 270 detrimental, 29
MATHEMATICA, 270 displacement, 31, 91
MIUTAM, 167, 187 incremental, 100
mu MATH, 270 radial, 45
PANDA, xv, 96, 97 total, 136
REDUCE, 270 dissimilarity in elastic moduli, 99, 106
STAGS, xv, 193, 196 distribution
coefficient of variation, 219, 220 cumulative, 179
collapse load, 194 extreme-value, 130, 134
compatibility equation, 47, 74, 79 normal, xv, 118, 132, 176, 177, 187, 188, 238
composite materials uniform, 136
boron/epoxy, 72, 82, 83, 84, 86, 87 truncated-normal, 129, 247, 249
carbon/epoxy, 95 Donnell-Imbert model, 166, 171
glass/epoxy, 72, 82, 83, 85, 86, 196 dynamics
graphite/epoxy, 72, 82, 83, 84, 85, 87 helicopter, 270
computational mechanics, 269 nonlinear, 270
computerized symbolic algebra, 33, 48, 81, 268 rotorcraft, 270
conditions
boundary, 6, 21, 32, 37, 101, 196 eigenfunctions, 211, 214
continuity, 4, 21, 32, 35, 37 eigenvalues, 213, 214
geometric, 21 eigenvalue problems, 267
constitutive relations, 89 eigenvectors, orthogonal, 203, 213
Conte’s test, 257, 259 elastic foundation
convex analysis, 222, 238 linear, 242, 276
model, xvi nonlinear, 176, 179, 241
modelling, 222, 223, 240, 254 elastic moduli, ellipsoid, 229
correlation length, 218, 230 elastic properties, 126
coupling electroplating, 148
nonlinear, 166 ellipsoid, semi-axes of, 230, 231, 233
stiffness, 75, 92 end
covariance, 118, 120 clamped, 54
critical buckling load simply supported, 54
maximum, 230 end-shortening, 113, 120, 121, 122, 192, 245
minimum, 230 energy
critical imperfection model, 193 potential, 89, 103, 214, 215, 216
cumulative distribution function, 179 second variation, 54, 56, 92, 101, 104
cylindrical shell strain, 89
imperfect, 191 third variation, 55, 91, 92, 104
non-uniform, 44 total potential, 89, 100
energy criterion, 53, 54, 57, 89, 90, 221
data, limited, xii equations
experimental, 108, 109 Donnell, 190
data bank, xiv, 176 stiffness matrix, 215
measured, 148, 174, 231 transcendental, 212
deflection, 18, 54 equilibrium
additional, 180 equation, 57, 74
density primary, 100
first-order probability, 140, 158 stable, 107
joint probability, 239 state, 100
power spectral, 142, 143 ergodicity, 137, 147, 171
probability, 157, 175, 176, 178, 180 Euler’s formula, xii
SUBJECT INDEX 333
expectation, mathematical, 140, 162, 194, 196, Hamiltonian, 230

198, 211 Hamming’s predictor-corrector method, 120
expression, asymptotic, 49, 82, 90, 93 histogram, 123, 125, 146, 148, 157, 160
homogeneity, weak, 145, 164, 165, 217
fabrication process, 72 hypergeometric function, 276
factor, knockdown, xiii, 98, 108, 124, 196, 197
failure, 187, 188 ill-conditioning, 62
field imperfection
perfectly correlated, 219 amplitude, 66, 114, 116, 122, 183, 187, 194
random, xi, 107, 165, 211, 214 asymmetric, 156, 171
finite difference calculus, xvi, 2, 35, 36 axisymmetric, 44, 53, 90, 154, 161, 176, 183
finite difference equation, 3, 4, 20, 210 critical, 176, 193
second-order, 34 data bank, xiv
finite difference method, 2, 7, 17, 19, 21, 72, 77 effective, 177
finite element method, 174, 187, 200, 210, 214, function, 118, 150
216, 219 Gaussian, 180
flexural stiffness, 200 geometric, xiii, xvi, 54, 89, 109
Fokker Aircraft Company, xiii initial, xi, xii, xvi, 12, 17, 44, 91, 94, 98, 105,
force resultant function, 262 139, 175
foundation length, 21, 25
cubic, 109 mean, 142, 150, 169, 170, 171
linear, 111 model, 169, 171, 189, 199
non-linear, 111 non-symmetric, xvii, 139, 161, 170, 171
quadratic-cubic, 110, 112 random, xi, xvii, 108, 109, 139, 170
Fourier sensitivity, xii, 107, 108, 112, 124, 137, 174,
coefficients, xiv, 108, 139, 187, 191, 197 175, 210
equivalent, 194, 197, 199 independence, statistical, 201
series, 108, 243 inhomogeneity, 165
transform, 142 integer part, 111
frame interval analysis, 240
Roorda-Koiter, 99 iterative procedure, 65
two-bar, 99, 100
frequency, spatial, 141, 143 Kahrunen-Loeve expansion, 211, 212, 213, 214, 221
function knockdown factor, xiii, 98, 108, 124, 196, 197, 212
auto-correlation, 109, 139, 180, 218, 249 Koiter’s semi-circle, 48, 49, 55, 98
auto-covariance, 117, 118, 120, 129, 137, 140, 141, Kolmogorov-Smirnov test, 121, 146
144, 145, 150, 152, 153, 162, 163, 211 Kronecker delta, 113, 164, 184, 244, 257
distribution, 158
ergodic, 109 Lagrange multiplier technique, 230, 231, 252
error, 130, 159, 249 lamina, 72, 73, 82
Gaussian random, 117, 180 laminate, 72, 73, 88, 224, 235
homogeneous, 164 angle-ply, 75, 226, 233
hypergeometric, 276 cross-ply, 75, 226, 233
mean, 117, 129, 139, 141, 180 Laplace operator, 59
orthogonal, 140 Legendre polynomials, 213, 218
performance, 187, 194, 196 level of significance, 146
probability density, 130 limit point, 43, 165, 177, 178, 181, 182, 185, 186, 191
random, 109 load
stability, 20, 201 axial, 18
transfer, 158 buckling, xii, 1
variance, 139 carrying capacity, 10, 41
collapse, 171
Gaussianity, 108, 176, 181 compressive, 200
non-Gaussianity, 176 design, 196, 198
girder, 32 increment, 178
Godunov-Conte method, xvi, 52, 62, 64, 257, 260, random applied, 136
261, 263, 267, 274 localization, xvi, xvii, 1, 202
Gram-Schmidt procedure, 257 factor, 202, 204, 205, 206, 210
guaranteed performance approach, 240 phenomenon, 210
334 SUBJECT INDEX
logarithmic decrement, 28 orthogonalization procedure, 62

Lommel function, 261 orthotropy coefficient, 276
Lyapunov exponent, 28, 202, 203, 204, 205
PANDA code, xv, 96, 97
Mathematica® , 33, 48, 81 paradigm, probabilistic, xi, 240
matrix pass-band, 203, 204, 205
auto-covariance, 129, 140 periodicity, 2, 17, 18, 21, 191
cross-correlation, 169 perturbation procedure, 47
diagonal, 213 plate
geometric, 207, 208 circular, 261, 274
lower-triangular, 165 composite, 222, 227
shape-function, 215 continuous, 6, 7, 31, 35, 41
stiffness, 207, 208 equivalent orthotropic, 1
transfer, 201, 204 homogeneous, 261
upper-triangular, 260 multi-span, 30
variance-covariance, 140, 142, 143, 144, 146, 147, N -span, 31, 35
149, 151, 152, 153, 154, 163, 167, 168, 174, 180, orthotropic, 262
187, 188, 194, 195, 197 polar-orthotropic, 277
mean value, 178, 187 rectangular, 3, 10, 32
sample, 181 rib-stiffened, 210
membrane strain energy, 89 simply supported, 6, 37
method stiffened, 1, 2
Boobnov-Galerkin, 48, 64, 81, 113, 121, 123, 180, three-span, 2, 7, 9, 16
191, 242 two-span, 2, 7
complementary functions, 258 point
complex, 238 asymmetric bifurcation, 177, 178, 181, 182, 184,
Donnell-Imbert, 166, 169, 171, 189 185, 186
feasible directions, 238 bifurcation, 100, 177, 178, 180, 181, 182, 184,
Frobenius, 274 185, 186
Godunov-Conte, xvi, 52, 62, 64, 257, 260, 261, critical, 182, 186
263, 267, 274 design, 193
gradient projection, 238 limit, 165, 177, 178, 181, 182, 182, 184, 185,
group-theoretical, 193 186, 191
Hamming’s predictor-corrector, 120 unstable bifurcation, 177, 178, 180, 181, 182, 184,
Kantorovich, 210 185, 186
Qiria-Davidenko, 110, 116 Poisson’s ratio, 4, 45, 54, 66, 98, 231
Rayleigh, 274 prebuckling state, 54, 59, 62, 165
Rayleigh-Ritz, 279, 285 analysis, 60, 62, 65
Riks’ path parameter, 193 fundamental, 54, 90
second-moment, 187, 192, 193, 194 pressure, external, 106
truncated hierarchy, 109 probability of failure, 130, 132, 134, 135, 188, 240
variable parameter, 279 problem
misplacement, xv, 2, 8, 12, 15 boundary value, 116
perfect, 17 initial value, 116
MIUTAM code, 167, 187 process, manufacturing, xiii, 136, 137, 149, 174, 183
mode localization, xvi, 1
modulus of elasticity, 45, 82, 99 radius, 194
moment of inertia, 110 gyration, 194
Monte-Carlo method, xi, xii, xiv, xv, xvi, 107, 108, random field, 214, 218
109, 110, 117, 118, 121, 124, 126, 131, 139, 144, random function
145, 147, 154, 157, 160, 161, 164, 166, 171, 174, auto-correlation, 109, 139, 180, 218, 249
175, 176, 181, 187, 191, 192, 213, 217, 220, 242, auto-covariance, 117, 118, 120, 129, 137, 140, 141,
250 144, 145, 150, 152, 153, 162, 163, 211
Gaussian, 117, 180
NASA Langley Research Center, xiii homogeneous, 144, 164
non-linear programming, 237 inhomogeneous, 144
number of half-waves, 183, 188 random variable(s)
axial, 183, 188 basic, 187
circumferential, 188, 194 correlated, 213
SUBJECT INDEX 335
dependent, 119 shell, cylindrical, xii, 2, 43, 52, 65, 71, 148, 154
independent, 119, 218 Ariane interstage, xiii
normal, 119, 158, 160, 177, 178, 187 simulated, xiv, 123, 145, 169, 172
uncorrelated, 213 shifting operator, 34
randomness, xiii, 136, 239 shooting method
Rayleigh quotient, 276, 280 simulation, 118, 144, 148, 171, 172, 189
Rayleigh-Ritz method, 285 conditional, 127, 131
Rayleigh’s method, 274, 283 sample, 156
relation, moment-slope, 33 solution
reliability, xi, xiv, xv, xvii, 109, 118, 122, 124, 136, multi-mode, 110, 130
138, 139, 146, 158, 161, 170, 175, 182, 183, 187, purely analytic, 118, 161, 169
193, 196 single-mode, 108, 110, 114, 119, 121
reliability index, 188 span, disordered, 22, 25, 35
representation, half-wave cosine, 144 spring, 2
rigidity constant, 25, 180
flexural, 3, 4, 10, 31, 58, 77, 214, 219 torsional, 25, 26, 205
random, 216, 262 stability, theory, xi
torsional, 1, 5, 10, 16, 40, 41 elastic, xii
Roorda-Koiter frame, 99 structures, xii
STAGS code, xv, 193, 196
sample mean, 181 stiffener, 1
scanning device, 149 half-stiffener, 37
second moment method, 186, 192, 193, interior, 32
194 optimal position of, 10
sensitivity derivatives, 230 stiffness matrix, 72, 73, 215, 216
separation of variables, 46, 63, 72, 76 global, 215, 216
series transformed, 73
double Fourier, 165 stress, resultant, 73
Fourier, 140 structure
half-range cosine, 166 gridwork, 17
half-range sine, 162 mono-coupled, 203
orthogonal, 211 multi-span, 17
set perfect, xiv, 115, 119, 157, 194
convex, 231 periodic, 2, 17, 35, 205
fuzzy, 240 randomly disordered, 2
hyper-cuboid, 256 Sylvester’s theorem, 151
shape, ideal, xi, xii, 99 symbolic algebra, 33, 48, 81, 268
shearing force, 6
shell technique, batch means, 132
Amazigo, 146 test of significance
anisotropic, 71 χ 2 test, 182, 186
axially compressed, 183 theorem
Booton, 197 central-limit, 133
Caltech A-shells, 149, 150, 151, 152, 153, 183, 197 Furstenberg, 201, 202
Caltech B-shells, 153, 154, 157, 167, 168, 170, 171, Sylvester, 151
173, 183, 197 theory
Caltech AS-shells, 195, 196 asymptotic, 176
composite, 71, 79, 88, 89, 92, 97, 196, 197, Donnell-Mushtari-Vlasov, 223
222 Ikeda-Murota, 181
ellipsoidal, 229 initial postbuckling, 100
finite, 142 Koiter’s special, 183
imperfect, 70, 148 probabilistic, xi, xii
infinite, 140, 152, 160 Sanders’ shell, 96
isotropic, 71, 82, 96 thickness
laminated, 72, 91, 235 amplitude, 53
metallic, 88 constant, 43, 44, 61, 65
simulated, 145, 169, 172 nominal, 58, 65, 72, 82
six-layer, 82 variable, 43, 53, 58
symmetrically laminated, 75 variation of, xv, 43, 44, 47, 61, 65, 68, 79, 82, 98
336 SUBJECT INDEX
Trefftz criterion, 101 vector

truncation error, 214 correlated, 213
twisting moment, 224 mean, 140, 197
normal, 129, 144, 216
unbiased estimate, 149 random, xi, 144, 158, 175, 176, 178, 198
uncertainty, xi, 222, 230, 234, 239 uncorrelated, 213
in elastic moduli, 230 viscoelasticity, 221
ellipsoid, 230
external load, 127 wave number, 55
set-theoretical modeling of, xv axial, 48
unknown-but-bounded, 240 circumferential, 48
critical, 76
variable
basic, 187 Young’s modulus, 16, 65, 100, 110, 180, 189,
dependent, 119 200, 262
independent, 119 lower bound, 231
normal, 119, 158, 160, 177, 178, 187 nominal, 233
random, 138, 140, 176, 211 semi-axes, 233
variance, 110, 120, 122, 158, 176, 178, 198 stochastic, 210, 223
sample, 126, 169 upper bound

(Isaac Elishakoff, Yiwei Li, James H. Starnes JR) (B-Ok - Xyz) PDF

Uploaded by

Copyright:

Available Formats

(Isaac Elishakoff, Yiwei Li, James H. Starnes JR) (B-Ok - Xyz) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Isaac Elishakoff, Yiwei Li, James H. Starnes JR) (B-Ok - Xyz) PDF

Uploaded by

Copyright:

Available Formats

Non-Classical Problems in the Theory of Elastic Stability

When a structure is put under an increasing compressive load, it may become

Isaac Elishakoff is a professor in the Department of Mechanical Engineering at

Yiwei Li is an engineering software developer at Alpine Engineered Products, Inc.

JAMES H. STARNES, JR.

Cambridge University Press

© Cambridge University Press 2001

This publication is in copyright. Subject to statutory exception

First published 2001

Library of Congress Cataloguing in Publication data

ISBN-13 978-0-521-78210-4 hardback

ISBN-13 978-0-521-02010-7 paperback

Preface: Why Still Another Book on Stability? page ix

1 Mode Localization in Buckling of Structures 1

2 Deterministic Problems of Shells with Variable Thickness 43

3 Stochastic Buckling of Structures: Monte Carlo Method 107

3.2 Reliability Approach to the Random Imperfection Sensitivity

4 Stochastic Buckling of Structures: Analytical and Numerical

4.4 Reliability of Composite Shells by STAGS 196

5 Anti-Optimization in Buckling of Structures 222

6 Application of the Godunov-Conte Shooting Method

7 Application of Computerized Symbolic Algebra

Special emphasis is placed on asymptotic evaluations and non–Monte Carlo analytical

University College, London, and of G. W. Hunt of University of Bath, experimental

first class considered overly simplified models of single-degree-of-freedom finite struc-

Y. Ben-Haim, Mr. M. Charmatz, Mr. E. Itzhak, Dr. S. Marcus, Professor J. Singer,

Mode Localization in Buckling of Structures

Think globally, act locally.

The genuine spirit of localism . . .

1.1 Localization in Elastic Plates Due to Misplacement

We ﬁrst consider a rib-stiffened rectangular plate subjected in its mid-plane to

For rib-stiffened, or intermediately supported, plates, we have roots si (i = 1, 2, 3, 4)

the two typical neighboring spans numbered j and j + 1 are

where (GJ ) j denotes the torsional rigidity of the jth rib.

−β̄ 1 sin(β̄ 1r j π )A j + β̄ 1 cos(β̄ 1r j π )B j − β̄ 2 sin(β̄ 2r j π )C j + β̄ 2 cos(β̄ 2r j π )D j

cos(β̄ 1r j π)A j + sin(β̄ 1r j π )B j + cos(β̄ 2r j π )C j + sin(β̄ 2r j π )D j

Figure 1.2 Notations and positive directions.

specifying the positions of the stiffeners:

F4,5 = cos(β¯1r2 π ), F4,6 = sin(β̄ 1r2 π ), F4,7 = cos(β̄ 2r2 π )

Figure 1.4 Uni-axially compressed three-span stiffened plate.

Figure 1.11 Effect of misplacement on the buckling load of a stiffened, three-

1.2 Localization in a Multi-Span Periodic Column with a Disorder

(1) w = 0 at xi = 0; (2) w=0 at xi = ai (1.48)

from which Ci and Di can be evaluated as

Note that the preceding derivation could also be performed straightforwardly by

MiL − MiR = J θi (1.56)

[c2 (E + E −1 ) + 4(c1 + )]θi = 0 (1.58)

Equation (1.48) is a second-order ﬁnite difference equation, whose solution is

A direct substitution of Equation (1.59) into Equation (1.58) yields

Case A: −2(c1 + )/c2 ≥ 1

θi = Aeφi + Be−φi (1.62)

where A and B are arbitrary constants.

Case B: −2(c1 + )/c2 ≤ −1

θi = [A cosh(iα) + B sinh(iα)] cos(iπ ) (1.64)

Case C: −1 ≤ −2(c1 + )/c2 ≤ 1

and, for θi , we have

θi = A cos(iβ) + B sin(iβ) (1.66)

ai = a for i = 1, . . . , q, q + 2, . . . , N ; aq+1 = b (a = b) (1.67)

As a particular case b = a, we recover the perfectly periodic structure.

[c2 (E + E −1 ) + 4(c1 + )]θi = 0 (1.58)

Case A: −2(c1 + )/c2 ≥ 1

Case B: −2(c1 + )/c2 ≤ −1

Case C: −1 ≤ −2(c1 + )/c2 ≤ 1

ai = a for i = 1, . . . , q, q + 2, . . . , N ; aq+1 = b (a = b) (1.67)

−M0R = J θ0I or (2c1 + 4) θ0I + c2 θ1I = 0 (1.73)

M NL = J θ NIII−q−1 or (2c1 + 4) θ NIII−q−1 + c2 θ NIII−q−2 = 0 (1.74)

(2c1 + 4)eφq + c2 eφ(q−1) A1 + (2c1 + 4)e−φq + c2 e−φ(q−1) B1 + M̄ qR = 0

(2c1 + 4 + c2 eφ )A2 + (2c1 + 4 + c2 e−φ )B2 − M̄ q+1 L

(2c1 + 4)eφ(N −q−1) + c2 eφ(N −q−2) A2

+ (2c1 + 4)e−φ(N −q−1) + c2 e−φ(N −q−1) B2 = 0 (1.83)

{(2c1 + 4) cosh(αq) cos(πq) + c2 cosh[α(q − 1)] cos[π (q − 1)]}A1

{(2c1 + 4) cosh[α(N − q − 1)] cos[π (N − q − 1)]

[2c1 + 4 + c2 cos(β)]A1 + c2 sin(β)B1 = 0 (1.93)

[2c1 + 4 + c2 cos(β)]A2 + c2 sin(β)B2 − M̄ q+1 L

{(2c1 + 4) cos[β(N − q − 1)] + c2 cos[β(N − q − 2)]}A2

[(2c1 + 4) + c2 e−φ ]A2 = 0 (1.102)

{(2c1 + 4) cos[π (N − q − 1)] + c2 e−α cos[π (N − q − 2)]}A2

and the solution for i is

of numerical solutions as becomes larger. Even though the ﬁrst-order asymptotic