Bok:978 94 017 0243 0 PDF

Analysis of Geometrically Nonlinear Structures
Analysis of Geometrically
Nonlinear Structures
Second Edition
by
Robert Levy
Technion - Israel Institute of Technology,
Haifa, Israel
and
William R. Spillers
New Jersey Institute of Technology,
Newark, New Jersey, U.S.A.
Springer-Science+Business Media, B.V.

A c.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-6438-7 ISBN 978-94-017-0243-0 (eBook)

DOI 10.1007/978-94-017-0243-0
Printed on acid-free paper
All Rights Reserved

© 2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003.
Softcover reprint of the hardcover 1st edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher, with the
exception of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
nr.Jt111;)1;) n"i'\t1 yj~m
1" ,.., 11\u1n'"
)Inc! tlie £ana liac! rest from war

Josliua 14 15
To Lydia, Itai and Shelly

R.L.
To the late Jewell Garrelts of

Columbia University
W.R.S.
VII
BOOK CONTENTS
Preface XIII
Using the CD xv
Chapter 1 Overview 1
1.1 Introduction 1
1.2 Newton's Method 2
1.3 Restrictions of Small Strain 4
1.4 Stress Stiffening 5
1.5 Buckling 6
1.5.1 Snap Through 7
1.5.2 Thermal Buckling 8
1.5.3 Euler's Column 8
1.5.4 Moment Distribution 9
1.5.5 Eigenvalue Approach 12
1.5.6 An Exact Approach 15
1.6 Prestress 17
1.6.1 A Fundamental Theorem 18
1.7 Problems 21
Chapter 2 Linear Structural Analysis 23

2.1 Introduction 23
2.2 The Truss Problem 25
2.3 Computer Programs 29
2.3.1 Programs PI-TR3D.FOR and P2-TR2D.FOR 30
2.4 Examples 30
2.4.1 Example 2.1. A 3-Bar Truss 30
2.4.2 Example 2.2. A 24-Bar Space Truss Dome 32
2.4.3 Example 2.3. A 72-Bar Double Layered Grid 35
2.5 Problems 39
Chapter 3 "Exact" Analysis of Trusses 41

3.1 Introduction 41
3.2 Linearization of the Joint Equilibrium Equations 42
3.3 The Geometric Stiffness Matrix 43
3.4 Overall Buckling 47
VIII

3.5.1 Programs P3-TR3DNL.FOR and P4-TR2DNL.FOR 49
3.5.1.1 Nonlinear Analysis and Newton's Method 51
3.5.2 Programs P5-BUCK3D.FOR and P6-BUCK2D.FOR 51
3.6 Examples 52
3.6.1 Example 3.1. Biot's 2-Bar Prestressed Truss 52
3.6.2 Example 3.2. A Prestressed Cablenet 54
3.6.3 Example 3.3. Plane Truss Buckling 57
3.6.4 Example 3.4. Buckling of a Symmetric 2-Bar Truss 61
3.6.5 Example 3.5. Buckling of a 4-Bar Shallow Space Truss 63
3.6.6 Example 3.6. Buckling of the 24-Bar Dome 66
3.7 Problems 68
Chapter 4 Nonlinear Analysis of Plane Frames 71

4.1 Linear Analysis 71
4.2 Computer Program P7-FR2D.FOR 75
4.2.1 Example 4.1 A Simple Plane Frame 76
4.3 The Geometric Stiffness Matrix 78
4.4 Computer Program P8-FR2DNL.FOR 81
4.4.1 Examples 82
4.4.1.1 Example 4.2. Buckling of a Portal Frame 82
4.4.1.2 Example 4.3. Buckling of a Plane Frame 85
4.4.1.3 Example 4.4. Large Rotations of a Circular Cantilever
Beam 87
4.5 Problems 90
Chapter 5 Nonlinear Analysis of Space Frames 93

5.1 Introduction 93
5.2 Linear Analysis 93
5.3 Computer Program P9-FR3D.FOR 98
5.3.1 Example 5.1. A 2-Storey Simple Space Frame 99
5.3.2 Example 5.2. A More Complex Space Frame 102
5.4 Nonlinear Effects 105
5.5 The Geometric Stiffeness Matrix 107
5.6 Computer Programs PI0-FR3DNL2.FOR, PII-FR3DNL3.FOR and
P12-FR3DNLSR 113
5.6.1 Example 5.3. Lateral Torsional Buckling 114
5.7 Problems 119
Chapter 6 Nonlinear Analysis of Membranes 121

6.1 Introduction 121
6.2 The Geometric Stiffness Matrix of the Plane Stress Triangular Finite
Element 121
IX
6.3 Three Dimensional Members 127

6.4 A Direct Alternative Derivation of the Geometric Stiffness Matrix of
Three-Dimensional Membranes 131
6.5.1 Program P13-FEMPS.FOR 132
6.5.2 Program PI4-MEMBR.FOR 132
6.5.3 Program PI5-MEMNL.FOR 133
6.6 Examples 133
6.6.1 Example 6.1. A Deep Beam 133
6.6.2 Example 6.2. A Spherical Cap 136
6.6.3 Example 6.3. A Flat Stretched Membrane 139
6.7 Problems 148
Chapter 7 Cablenets and Fabric Structures 151

7.2 Basic Methods of Shape Finding 152
7.2.1 Deformed Shape 152
7.2.2 Force Density Method 153
7.2.3 Grid Method 154
7.2.4 Smoothing 155
7.3 The Grid Method 155
7.3.1 Example 7.1. A Piece of a Cross Arched Skylight 156
7.3.2 Example 7.2. A Squared Base Skylight 158
7.4 Smoothing 163
7.5 A More Complex Example 166
7.6 Membrane Finite Element Model 177
7.7 Patterning 180
7.8 Computer Programs For Cable Nets And Fabric Structures 182
7.8.1 LAYOUT.FOR (The grid method) 182
7.8.2 LPLOT1.FOR 183
7.8.3 LAYOUTPLOT.FOR 183
7.8.4 PATTERN. FOR 184
7.9 Problems 186
Chapter 8 Three-Dimensional Beam-Columns 187

8.2 The Equation of Three-Dimensional Beam-Columns 188
8.3 The Member Stiffness Matrix 191
8.4 Numerical Solution 194
8.5 Special Cases 196
8.5.1 The Elastic Beam 196
8.5.2 Two-Dimensional Beam-Column 196
x
8.5.3 Lateral Buckling 197
8.5.4 A More Complex Case 197
8.5.5 The Effect ofInitial Torsion 199
8.6 Problems 203
Chapter 9 Nonlinear Analysis of Shells 205

9.2 The Geometric Stiffness Matrix of Triangular Eelement Shells 206
9.2.1 In-Plane Contribution of the Triangular Membrane
Element 207
9.2.2 In-Plane Contribution of the Triangular Plate Bending
Element 207
9.2.3 Out-Of-Plane Contribution to the Shell Geometric Stiffness
Matrix 210
9.3 Element Pure Deformational Rotations and Translations 214
9.3.1 Stress Retrieval in the Membrane Finite Element 214
9.3.2 Stress Retrieval in the Plate Finite Element 215
9.4 Computer Program PI6-SHELLNL.FOR 218
9.5 Examples 219
9.5.1 Example 9.1. Bending of a Cantilever Plate 220
9.5.2 Example 9.2. Simply Supported Plate 223
9.5.3 Example 9.3. Analysis of a Shallow Cylindrical Shell 226
9.5.4 Example 9.4. Leicester's Shallow Spherical Shell 233
9.5.5 Example 9.5. Mescall's Shallow Spherical Shell 234
9.5.6 Example 9.6. Open Hemispherical Shell 235
9.5.7 Example 9.7. Lateral Buckling of an L-Frame 236
9.6 Some Remarks 237
9.7 Problems 237
References 239
Appendix 1 Member Stiffness When Beam-Column Effects are

Included 243
Appendix 2 Determinants 247
Appendix 3 The Rotation Matrix 249
Appendix 4 Perturbation Methods Applied to Plane Beams 257
Appendix 5 Introduction to Computer Programs 259

A5.1 Introduction 259
XI
AS.2 Space Trusses 259

AS.3 Plane Frames 260
AS.4 Listing for TR3D.FOR 261
AS.S Listing for FR2D.FOR 263
Appendix 6 Graphics on a PC 267

A6.1 Introduction 267
A6.2 Plotting in 2-D 268
A6.3 Drawing Lines in 2-D 269
Index 271
XIII
PREFACE
The availability of computers has, in real terms, moved forward the practice
of structural engineering. Where it was once enough to have any analysis
given a complex configuration, the profession today is much more
demanding. How engineers should be more demanding is the subject of this
book.
In terms of the theory of structures, the importance of geometric
nonlinearities is explained by the theorem which states that
"In the presence ofprestress, geometric nonlinearities are of the same order
of magnitude as linear elastic effects in structures. "
This theorem implies that in most cases (in all cases of incremental
analysis) geometric nonlinearities should be considered. And it is well
known that problems of buckling, cable nets, fabric structures, ... REQUIRE
the inclusion of geometric nonlinearities.
What is offered in the book which follows is a unified approach (for
both discrete and continuous systems) to geometric nonlinearities which
incidentally does not require a discussion of large strain. What makes this all
work is perturbation theory. Let the equations of equilibrium for a system be
written as
where P represents the applied loads, F represents the member forces or

stresses, and N represents the operator which describes system equilibrium.
(This equation can also be thought of as the matrix equation of node
equilibrium for a discrete system.) Under a load perturbation dP this system
responds as
dN TF + NT dF = dP
It is the first term in the above equation which describes so-called geometric
nonlinearity and it is the second term which returns linear theory. For a
discrete system it turns out to be a relatively simple matter to convert this
equation into the usual
XIV PREFACE
Here K E and KG are the geometric and elastic stiffness matrices and 0
represents the system displacements. With this approach, nonlinear structural
analysis becomes simply an application of Newton's method.
In terms of organization, this book begins with an introduction
which cites many practical phenomena involving geometric nonlinearities.
We then procede through a hierarchy of discrete and continuous systems ...
from trusses to frames and from beams to a membrane and shell finite
elements. In all cases computer programs are a central part of the discussion.
Finally in a slightly different spirit, a chapter on fabric structures is included.
Fabric structures of course require nonlinear analysis but this chapter goes
outside the other material of this book when talking about shape finding and
patterning. And we leave out much of the beautiful work which comprises
much of what is now called nonlinear elasticity primarily on the grounds that
it depends on issues of large strain.
Finally, we hope that the material which follows will have appeal to
the practicing engineer. In fact, this work appears in its present form largely
because of a comment made by Horst Berger (Horst Berger Partners, New
York City) in the 1980's who suggested that since fabric structures were
nonlinear anyway, their analysis should include the nonlinear effects (such
as buckling) in the supporting frame. He might well amend this today to
include more general classes of structures. That is, there are many reasons
today to include nonlinear effects in structures rather than simply settling for
linear analysis. And the technology to do so is available and inexpensive.
Any modem discussion of the incremental analysis of structures
takes much from the now classic work ofBiot (1965). We would be happy to
be regarded to be in his company.
Robert Levy
William R. Spillers
Haifa and Newark, August, 1994
Preface to the Second Edition
With the exception of new material on three-dimensional beams

(Chapter 8) and shells (Chapter 9) this text retains most of the material
of the first edition.
Robert Levy
William R. Spillers
Haifa and Newark, July, 2003
xv
USING THE CD
This book is sold with a CD which contains FORTRAN source code and
data files to be used with these programs. It is presumed that the reader has
access to a personal computer (PC) and a FORTRAN compiler. The authors
have attempted to make computer programs an integral part of this text and
see the reader with a PC at his or her side. This approach probably derives
from the old saw which tells us that you don't really understand something
until you code it for a computer. Having said this, it seems to the authors that
the issue of how best to combine computer code and text remains largely
unsettled.
The following comments attempt to deal with some preliminary
items that the reader should understand at the outset:
pes Versus Workstations. The computer programs associated with

this text all run on a PC running DOS. This is an expression of the authors'
view that PCs not workstations will be computers of choice in both the
workplace and the university. But the reader may note that with the
exception of computer graphics features, these programs move easily to a
Sun workstation. (We run them on the SUN under X-windows which is used
to supply the graphics routines.)
Computer Graphics. Computer graphics is fun but also a serious

component analysis programs, particularly with regard to checking data and
visualizing three-dimensional problems. The reader who wishes to take
advantage of computer graphics features should go directly to Appendix 6.
Hardware. The programs were run on a Pentium 4, 1.9GHz with

Windows® XP.
Practical Applications. This book is written more for use as a

learning tool than for production. That is particularly true of the computer
programs which do not use sparse matrix methods and are thus limited in
terms of the size of problems which can be treated. But 1) adding sparse
matrix methods to a computer algorithm is by now a fairly routine matter and
2) the code used in this text has served as the basis of many commercial
applications.
XVI
Contact With the Authors. We would be pleased to receive E-mail

from the readers. We can be reached at
cvrlevy@tx.technion.ac.il
spillers@adminl.njit.edu
CHAPTER 1
OVERVIEW
1.1 Introduction
There are a number of practical phenomena in structures which simply can

not be described using a linear formulation. Among these are buckling, the
behavior of cable nets and fabric structures, the formation of plastic hinges,
the nonlinear material behavior of concrete, structural collapse, ... In all these
cases (and many others) what has come to be known as the elementary
theory of structures does not work or at least needs to be augmented. The
question is of course how.
Since the elementary theory of structures is the low end - the most
simple attempt to describe the response of structures - what is to be added to
it is an open-ended question. In the case of the work which follows what is
added to elementary theory is in some sense the "next step". To make this
point it is convenient to introduce the terms "geometric" and "material"
nonlinearities. Over the years effects due to constitutive equations (stress-
strain relations) which are nonlinear, such as those describing the yielding of
mild steel and the behavior of reinforced concrete, have been referred to as
material nonlinearities. These effects are NOT considered in the work which
follows.
truss bar
""
r0
truss bar
"" rJinn
J/ih"
IIff
L
lilliE
L
III.
Figure 1.1 Geometrically unstable structure.
This book addresses the low end of what is then left of structural
theory when material nonlinearities are not considered. Elementary structural
theory is of course restricted to issues of small strain. With the exception of
the truss problem, this book is also restricted to problems of small strain.
(The truss is an anomaly in which it is a simple matter to work with large
strain.) What is left is the effect of deformed geometry.
2 ANALYSIS OF GEOMETRICALLY NONLINEAR STRUCTURES
Elementary theory is a strange mix in which behavior is described in

the "undeformed configuration". That is, in elementary theory the effect of
deformations - even though they are computed - are neglected when writing
the equations of equilibrium and motion. Biot (1965) has used the structure
of Figure 1.1 to characterize these "geometric" difficulties which are
inherent in elementary theory. This structure is said to be (geometrically)
unstable, but it can function quite well in carrying load when sufficiently
prestressed. In fact, it is not even necessary for this structure to undergo
large deformations to work well, but its behavior is not included in the linear
theory of structures.
The central idea of this book is quite simple: Real structures are in
equilibrium in their deformed configuration, not their undeformed
configuration as implied by elementary theory. Using the deformed
configuration implies nonlinear analysis in the work below which will
typically involve applying Newton's method to compute the effect of a load
perturbation upon some given initial state. Buckling occurs when the system
matrix becomes singular. This is not only easy to say. It is easy to do for
many common structural systems.
This chapter attempts an overview by discussing familiar if eclectic
problems of structures with the idea of setting the stage for a consistent
approach to be developed in the later chapters.
1.2 Newton's Method
Central to the solution process used in this book is Newton's method. It is

utilized whenever need arises to solve nonlinear equations. It has the
advantage of rapid convergence and solutions are obtained with relative
ease. Newton's method is usually described for a scalar system (Figure 1.2).
Here an attempt is being made to find the value x * for which f (x*) = 0 .
The typical step of Newton's method involves starting at some given point
Xo and linearizing the system to find a new point Xl. Happily, Newton's
method works for an n x n system of n equations in n unknowns, f(x) = 0 .
Formally,
r{X) = 0 ~ r{xo)+ vr{xo)· dx ~ 0 (Ll)

or
dx = -vr{xo tl r(xo) (1.2)
with
dx = Xl - Xo (1.3)
OVERVIEW 3
linear
approximation
f(x)
Figure 1.2 Newton's method.
Equation (1.1) only involves keeping the linear term in a Taylor series
expansion. Vf is of course the gradient matrix of the n-vector f. Newton's
method linearizes the nonlinear system f = 0 at a point and requires the
solution of a system of linear equations for the next point. But solving linear
equations is something structural engineers are very good at. It might also be
noted that Newton's method requires the gradient matrix which will be
computed explicitly in the work which follows. In structures, the gradient
matrix of the equations of equilibrium is the tangential stiffness matrix
containing within both the usual elastic stiffness matrix, K E and the
geometric stiffness matrix, KG which will be discussed below.
One of the remarkable facets of Newton's method is that it possesses
the property of "quadratic convergence". The remainder of this section
discusses quadratic convergence for the simple case of a single variable x
and a single equationf(x). For the n-dimensional case, the reader can refer to
any of the standard texts on numerical analysis. The Taylor series expansion
offix) about Xo with remainder, is
where'; is a point in the interval (x, xo) . For a single variable and a scalar
function Eq. 1.1 reduces to
(1.5)
For x ~ x * so that f(x*) = 0 the substitution ofEq. 1.4 into Eq. 1.5 yields
x - X* = (X * -X )2 I"(r;) (1.6)
1 0 2f'(xo)
For ::;; 1 it follows that
(1.7)
which implies quadratic convergence in the neighborhood of x*. That is

Ix
when you are sufficiently close to a root so that * - Xo << I, the next step I
is much closer.
1.3 Restrictions of Small Strain
With the exception again of the truss, this book deals with the special case of
large rotations (deformations) but small strain. There are several reasons for
doing so but the principal one is simplicity. Using the restrictions of small
strain it is possible, in terms of mechanics, to keep much of classical theory.
Another reason for dealing with this special case has to do with materials.
The common structural components, with the notable exception of mild
steel, cannot tolerate large strains without fracture. This is true of cables,
fabrics commonly used in permanent structures, wood, aluminum, ...
When dealing with nonlinear problems, it is particularly important to
be careful with regard to the semantics used. A case in point is the
implications of the restrictions of "small strain" used in linear theory (see
Figure 1.3). Given a bar and a displacement, linear theory takes the member
length change to be the projection of the displacement in the initial direction
of the bar. The actual length change IS, on the other hand, obtained by
A'
displaced position L *" ." .,.

..
~ .... displacement
...... (OA -oe)
9
c initial pOSition (linear theory)
Figure 1.3 A truss bar. Approximation oflinear theory.

OVERVIEW 5
drawing an arc of a circle from the displaced end of the bar, in this case, to
the horizontal. The error is indicated in the figure. For the case of small
rotations this error can be shown to be small.
The point is that linear theory, if misused, can lead to the idea of
"fictitious" or "spurious" strain - strain when there really isn't any. In the
work below, steps will be taken to prevent the appearance of spurious strain.
This will typically be done by removing the effect of rigid bodymotion prior
to computing strain. (This step is fundamental to the concept of large
rotations but small strain.)
1.4 Stress Stiffening
Figure 1.1 shows the classic example of stress stiffening, the phenomenon in
which prestressing a structure increases its stiffness. This simple structure
can not work within linear theory because equilibrium is impossible in the
undeformed configuration as indicated in Figure 1.4. Clearly, since the
undeformed structure implies horizontal bar forces there is no mechanism to
carry an applied vertical load. The kinematic version of the impossibility of
equilibrium is the fact that a small vertical displacement of the loaded point
produces no length change in the bars (within elementary theory).
Another look at this structure comes from the theory of strings
(string vibrations from partial differential equations). Under the assumptions
of string theory, the structure may displace without changing the string
tension. If that is the case (Figure 1.5), deformed equilibrium may be used to
compute a joint stiffness. Here, vertical equilibrium implies that
P = 2T sin e where T is the prestress (string tension). But
sin e == tan e == e == d / L so that
p=(2:}d (1.8)
and the "joint stiffness" is, thus, equal to 2TIL with each member
contributing an equal amount of TIL to the stiffness.
F ...
r
o
Figure 1.4 Undeformed joint equilibrium.
!P
,j};;"'" .........dl?p
'" ","""
~~~~~~
O~
deformed equilibrium
IP
geometric
T
rdT~
T e
+
~
~
T
effect joint free body diagram
Figure 1.5 Joint equilibrium.
The "geometric effect" is the change dT in the internal force in

response to the change in geometry of the structure.
The "geometric stiffness" TIL of the members is their ability to carry
load normal to their centerline. Clearly this stiffness is a function of the
tension T and increases with increasing T The geometric stiffness matrix
discussed below will include this effect of stress stiffening.
1.5 Buckling
Figure 1.6 shows the classic buckling example used by Timoshenko (1936).
In this case a rigid bar is in equilibrium under the applied force P. Within
linear theory, the only solution available to this problem is the one indicated
in the schematic in which the spring is undeformed. In order to study
buckling, it is common to apply a "perturbation", in this case a displacement
d, to joint A and look for a position of deformed equilibrium. For deformed
equilibrium the spring force is kd; for a small angle e the vertical component
of the force P becomes pe. For deformed equilibrium it follows that
pe=kd (1.9)
Here k is of course the spring constant. If e = d / L, Eq. 1.9 becomes

Pd =kd => PeR =kL (1.10)
L
OVERVIEW 7
rigid bar
P
\.
c~
schematic deformed equilibrium
Figure 1.6 Buckling example.
It is said that the buckling load is equal to kL implying that for P::;; PCR the
system will return to its unperturbed configuration when disturbed and that
P ~ PCR implies catastrophic failure. Again, this solution is quite intuitive
but offers little help in terms of a general approach for problems of buckling.
1.5.1 Snap Through
Figure 1.7 shows the simple case of a symmetric two bar truss and its load
deflection curve for large displacements. This curve can be produced in the
following manner. First of all, it is necessary to use a nonlinear strain-
displacement relationship such as requiring the force to be proportional to
the "real" bar length change rather than the approximate length change of
elementary theory. Then the load P can be computed directly from the
displacement d:
• Assume any d
• Compute the real length change M
• Compute the bar forces F=k &
r P
B JI.
B
schematic pod plot
Figure 1.7 Snap-through buckling.

8 ANALYSIS OF GEOMETRICALLY NONUNEAR STRUCTURES
• Compute P using the deformed equilibrium (P = 2F sin g )

Here again k is the linear spring constant, g is the angle of inclination of the
bar in its deformed position, and F is the bar force.
There are several points of interest in Figure 1.7. First of all, point D
is typically what would be referred to as buckling; point A corresponds to
the truss deformed to the point at which it becomes flat; and point B
corresponds to the structure hanging down below the horizontal without
load. After point B the structure continues to stiffen as it is loaded.
In order to generate this behavior in a laboratory, it would be
necessary to control the displacement and measure the load. Otherwise, as
the load P is increased from the value zero, the structure wants to "snap
through" when point D is reached and jump to point C since following the
curve downward to point A requires unloading.
1.5.2. Thermal Buckling
The solution of the problem of the preceding section implies the solution of
the thermal buckling problem of Figure 1.8. (This problem looks a little like
pavement buckling.) Here the bars are heated. If they are heated sufficiently
the structure will buckle.
truss bar weight truss bar
~-~\-.. . .-d-T----tt-;/.-(~)-u-n+I---dT-.\-2im7
.
Figure 1.8 Thermal buckling.
The analysis of this buckling problem can be carried out in the

following manner. For any temperature change, dT, a "free bar length" can
be computed. (This is the length each bar would have if it were free to
expand.) For this bar length the analysis of Section 1.5.1 can be carried out.
If the attached weight is greater than the buckling load computed, thermal
buckling will not occur; otherwise it is unstable for the given dT. That is, the
structure will not buckle as long as the attached weight is sufficient to hold it
down.
1.5.3. Euler's column
There is of course a commonalty between discrete and continuous systems.

This can be seen in the classical problem of the buckling of a pin-ended
OVERVIEW 9
initial shape
'>. p
---;&---;
p p
~,'-7--- -+ '';It ;~M

~y deformed shape If: ..
v q(x)
tim - -----+,.,M
~ fOOO++
M \.
I. dx .'V
Figure 1.9 Euler's column.
column (Figure 1.9). In this simple case, deformed equilibrium implies that
M = Py. That is, the bending moment M is equal to the axial load P times
the lateral displacement y. If the constitutive equation M = -Ely" is
invoked (E=Young's modulus and 1 is the moment of inertia of the beam), it
follows that,
Py = -Ely" or y"+k 2 y = 0 (1.11)

with k 2 = P / El . Clearly,
y = A sin kx + B cos kx (1.12)
Using the boundary condition y = O@ x = 0 it follows that B = 0; if

y = 0@x = L it follows that sin kL = O. This implies that
kL = nJf (n = 1,2, ... ) . By picking the smallest buckling load it is argued that
kL = Jf or eL2 = Jf2 and PeR = Jf2 El / L2 .
Note that the approach used here is again one of perturbation. The
initially straight column is given a small lateral deformation and conditions
for deformed equilibrium used to identify the buckling load.
1.5.4. Moment Distribution
Gere (1963) gives an example of the use of the method of "moment

distribution" to determine the buckling load of a frame. His discussion
includes the simple frame of Figure 1.10. Using Gere's analysis this frame
buckles when the rotational stiffness of the joint becomes zero. Since this
frame has no side sway and if member length change is neglected, this frame
becomes a single degree of freedom system. Since the horizontal member
r
10 ANALYSIS OF GEOMETRlCALLyNONLINEAR STRUCTURES
L
L=192in
1=20 in'
E=30,000 ksi
L
,.
Figure 1.10 Frame buckling.
has no axial load, its stiffness is simply 3EI/ L. This is also the stiffness of
the vertical frame member when axial load effects are neglected. In fact the
linear elastic solution of this problem simply has a compressive load P in the
vertical member.
Gere introduces buckling by considering "beam-column" effects.
(See Appendix 1.) Within beam-column theory, the stiffness of a member
(the end moment required to produce unit rotation) decreases as the
compressive load in a member is increased and increases as the tension in a
member increases. This is another example of stress-softening/stiffening.
Consider a specific case of the frame in Figure 1.10. As the load increases
from zero in this problem, at some point the stiffness of the vertical member
goes to zero, (kL~3.125), (see Figure 1.11). The structure is at this point
not yet unstable because the horizontal member can still provide stiffness to
the joint. If the load P is increased further the stiffness of the vertical
member becomes increasingly negative. At some point the stiffness of the
vertical member decreases to - 3EI / L and thus the total rotational stiffness
of the joint becomes zero. The condition for zero stiffness for this case can
be written according to Appendix 1, as
l+~=O (1.13)
a
This is an implicit trigonometric equation whose solution was obtained using

a fixed point iterative scheme to yield a critical value, PeR' of 226 kips.
Returning to Figure 1.11 the stiffness of - 3EI / L occurs at kL=3. 725 which
translates to:
OVERVIEW 11
p. _ E/(kL)2 30,000 x 20 x (3.725) 2 "" 226 ki s

CR - L2 (192)2 p
It may be noted here that this example illustrates the importance of

"member stiffness". That is, without beam-column effects it is impossible to
discuss buckling in this case. The example is interesting from another point
of VIew. When dealing with nonlinear problems it is often difficult
8
Ten
For
6
L
ElI.l
4
-2
-4
-6
-8
Stiffness factors for beams with axial loads. K - stiffness; k = /P

VEi
Figure 1.11 (after Gere (1963).)
to keep track of the level of approximation involved. But in this case it is

only member length change which is neglected. It remains to determine how
sensitive the buckling load is to member length change.
1.5.5 Eigenvalue Approach
This book argues that buckling is most appropriately described by

incrementing the load on a structure to failure - that is until its stiffness
matrix becomes singular. This section describes a simplified, alternative
approach to buckling which is sometimes found in the literature. (Note that
this approach depends on unit vectors in the undeformed configuration and
thus does not properly represent deformed equilibrium.)
Consider the structure of Figure 1.12. It is a simple two-bar truss
under the action of an applied load P for which a deformed equilibrium
position exists in addition to the undeformed equilibrium described above.
Linear theory, of course, does not give a buckling load.
If the "string effects" described in Section 1.4 are included it is
possible to compute a buckling load as follows. First of all note that linear
theory gives a load/deformation relationship, KEo = P,
(1.14)
Here K E is the elastic stiffness matrix, D is the node displacement matrix

(ox = d), and P is the joint load matrix. The "string effects" can be included
by noting that the horizontal displacement of the loaded joint produces a
horizontal force component PdlL (Figure 1.12) and that there is no effect of
this type from the horizontal bar since it is unloaded. This type of
phenomenon is commonly described using the so-called "geometric stiffness
matrix" KG.
Writing this string effect on the joint forces as KGD,
(1.15)
and combining both effects will then give equilibrium as (K E + KG)o = P .

In general, given P, it is possible to solve this system. The singular case
r
OVERVIEW 13
2
A,E,L cons!.
deformed
equilibrium
Figure 1.12 Eigenvalue buckling.
(buckling) corresponds to the problem of finding P so that the determinant of

KE + KG is zero. In this trivial case the determinant is zero only when
PeR =-AE (1.16)
At buckling then a horizontal displacement, d, of the loaded joint produces

an elastic force AEdiL in bar 1 which is balanced by the string effect force
from bar 2 of PdlL.
This example, of course, still leaves open the question of a general
approach. But it is simple enough so its shortcomings are obvious. The
problem here is that the effect of the deflections is not introduced in any
consistent manner. (Both the elastic and the geometric stiffness matrices
make use of the undeformed rather than the deformed configuration.) It will
be seen later that a better approach is not difficult
Figure 1.13 is an attempt to introduce a more general perspective to
geometric effects and the contribution of a generic member to the overall
geometric stiffness matrix. A more rigorous derivation based on a
perturbation of the equilibrium equations will be discussed later. It will be
seen to produce identical results.
Member i connects nodes A and C with a positive incidence toward
node A. If the bar force F; remains constant in the deformed and
undeformed configurations (e < < 1), the string effect or geometric effect is a
force Fie in the direction of vector d that can be written as
d
dF = Fjejdf (1.17)
~dF,"""
F geometric
, effect
c
undeformed
config uration
Figure 1.13 The geometric effect.
Now f) = M
Li
and d = 0 - (0· n)n where n is a unit vector in the direction of
the undeforrned configuration of member i. dF (Eq. 1.17) now becomes
dF = Fi [0 -(0· n)n] (1.18)

Li
In matrix form Equation 1.18 becomes
F
dF=-1 (I-nn )0
T
(1.19)
Li
In the general case of a member for which both ends are displaced, so that
0= 0 A - oe and dFA = -dFe the contribution of member i to the geometric
stiffness matrix associated with the appropriate degrees of freedom
bAr ~ r = x,y,z and oCr ~ r = x,y,z is simply
colA colC
bAx
- (I - nin;)1~: row A (1.20)

(I - nin;) b~x row C
bey
bez
OVERVIEW 15
1.5.6 An Exact Approach
This section introduces the notion of nonlinear buckling to describe overall

stability when large rotations and equilibrium in the deformed state are
considered. Moreover exact member length changes are used here to define
member deformations. The eigenvalue formulation of the last section is
restricted to small rotations and equilibrium in the initial state and is referred
to as linear buckling. Both of these formulations naturally fall within
nonlinear structural analysis since they describe buckling which is a
nonlinear phenomenon. Differentiation between the two levels of complexity
within buckling leads to the linear and nonlinear terminology. The
discussion in section 1.5.1 relates, of course, to nonlinear buckling.
Consider the fundamental non symmetric two-bar truss of Figure
1.14. When treated "exactly" it no longer is a simple statically determinate
structure but rather a complex system governed by implicit equations of
equilibrium and conditions for nonlinear stability. This section will only
discuss the symmetric case in which BI = B2 = B and v=O. For a discussion
of the general case see Levy and Brill (1989).
Assuming that member stiffnesses, k, are constant and Hooke's law
applies, equilibrium in the deformed state may be written in terms of the
displacement, U, as
HI
P-2k(Lo -L)-=O (1.21 )
L
where HI =H - U and Lo and L are the initial and deformed lengths of the
members respectively.
'" II ••
••
Figure 1.14 The non-symmetric two-bar truss.
16 ANAL YSrS OF GEOMETRICALLY NONLINEAR STRUCTURES
As the loading is increased proportionally the derivative dP / du will

vanish at a certain critical point along the equilibrium path and provide the
following nonlinear buckling condition:
dP =0
du
( Hr)2 + L-Lo =0 (1.22)
L Lo
The critical height, HI' is first obtained from Eq. 1.22. In terms of e,
the original angle of inclination ( e = arctan( H / B) ), it can be written as
(1.23)
Then HI is substituted in Eq. 1.21 to yield the critical buckling load
nonlinear PeR = 2kLo (1- cos 2 / 3 8)3/2 (1.24)
When linear buckling, as described by the eigenvalue approach of

section 1.5.5, is used to determine the buckling load of this truss, two modes
are evident. Modes that interchange roles at 8 = 45°. The critical loads for
these buckling modes are
(1.25)
Linear and nonlinear (exact) buckling can now be compared (Figure1.15)

and the result is striking. Up to an angle of e = 62° the linear eigenvalue
representation of stability overestimates the buckling load broadly. Defining
the "linear buckling load factor", A, as the ratio of the critical buckling load
obtained via the eigenvalue representation to the critical nonlinear buckling
load, a curve of A versus e can be generated. A peak of A = 7.55 at 45° is
observed which implies that the eigenvalue representation is largely
ineffective. This is an important result. It suggests that in most cases linear
buckling analysis is highly inaccurate. The error of linear buckling here is
reminiscent of shell theory where elementary buckling loads do not agree
with experiment. In that case it is common to cite "imperfections" as the
culprit. Here the difference between linear buckling and nonlinear buckling
cannot be explained in such terms.
OVERVIEW 17
8,---,---,----.---,----,---,---,----.---,
, • • I J
- - :- - :- - :- - :- .55
I • • •
7 ····-·-1-·····-1-··· · ··~·· ··· ··: + .. ..................... ..... .

I
~ ~
: : : : I :
5 6 ······[·······i······ ~ -·-----~---1 -T-·· .. ----.--- -------!-.... -
I : :
------r-------r-------r---··--:---r- r-
~ 5 : : : :
-g w
.--- ~-
••••••• ••••••• ••• --
I .
.s.4 _.- ._._._ .. _.. _-- _._----,._._--
.!2 • • • •
-------~------- ~-- -----~-----.-~---
::;;<
g "
~ 3 ·----··~--·-···~··-···-~··-···-~·--l---~- ---.- ------- ------- t------
,§'" . .. I .
2 · - -·-- ~·· --··· i·---·-·i···--·-i·--l·--!---· -- ------.. -.---.! ..... .
I ~ !
10 20 lJ 40 50 60 70 00 90
angle of inclination (degrees)
Figure 1.15 The linear buckling load factor.
1.6 Prestress
This section addresses the relative position of geometric stiffness effects in

structures. It will be argued that "prestress" is the driving force and that quite
generally geometrically nonlinear effects are of the same order of magnitude
as classical linear effects and should be included in most analyses. This
speaks to the question raised by Figure 1.1 concerning why a simple
"unstable" truss can perform well under conditions of small deformation.
Specifically, it will be shown using a uniform Taylor series expansion that:
Without prestress-
-Keeping terms of first order in displacement components leads to
the classical linear theory of structures.
-Keeping second order terms leads to the "geometric stiffness"
effect but also leads to a length change term due to rotation which is of the
same order.
With prestress-
-A zeroth order term occurs which corresponds to the presence of
the prestressing force.
- Keeping first order terms leads to the classical elastic response and
also to the "geometric stiffness" effect.
eLength change due to rotation appears as a higher order term.

The point is that when there is prestress, nonlinear effects which are
incorporated in the geometric stiffness matrix can be of the same order of
magnitude as the effects ordinarily included in the linear theory of structures
and must therefore be included in any analysis in order to develop a
consistent analysis procedure. This type of conclusion appears in other areas
of mechanics (Baumhauer and Tiersten, 1973).
1.6.1 A Fundamental Theorem
In Figure 1.16 a truss bar is shown in both the undeformed and deformed
position. In general the length is
and the force components are
(1.27)
(1.28)
where k is the bar stiffness. IfEq. 1.26 is now written as,
L=LO~ (1.29)
where
(1.30)
a Taylor series expansion can be used to eliminate the radical in the

expression for length and write Eqs. 1.27 and 1.28 in powers of t5x and t5y •
Now
(1.31 )
OVERVIEW 19
_Fx
displaced
position ______ ............
...
...
L
...... ......
... ... ... ...
e
La lix
inilial
position
Figure 1.16 A truss bar.
smce
1 1 3 2 - ... for x «I
--=I--x+-x II (1.32)
.JI+x 2 8
Equation 1.27 thus becomes
(1.33)
or, after substitution ofEq. 1.30
0; 1+ ... higher order terms

Fx = k(Ox + 2Lo (1.34)
Similarly Eq. 1.28 becomes
(1.35)
or, after substitution ofEq. 1.30
o
Fy = ko x ----2'... + ... higher order terms (1.36)
Lo
20 ANALYSIS OF GEOMETRICMLY NONLINEAR STRUCTIJRES
From Eqs. 1.34 and 1.36 it is clear that keeping terms uniformly in
the first order of the joint displacements gives rise to the classical x -
component of the first order or linear theory of structures,
F(l)
x
= k<5x
(1.37)
Going to quadratic terms gives rise to two new components,
(1.38)
If kt5x is identified with the classical bar force and 5 y / Lo with the angle e
in Fig. 1.16 (assuming small angle changes), the F~2) component which
becomes FY)e can be identified with the "string effect" (Eq. 1.8). But at the
same time Eq. l.38 indicates that there is a "length change due to rotation"
which is of the same order in the displacement components and must also be
included in order to be consistent. Specifically, to second order terms,
1
sece::::: 1 + _e 2 (1.39)
2
and
(1040)
which can be viewed as a second order correction to length change. Note that
e e
Eqs. 1.39 and lAO are identical since sec = L / Lo and = 5 y / Lo .
These results leave open the question of the relationship of linear
buckling theory to this Taylor series analysis since in the linear buckling of
trusses it is common to neglect terms corresponding to length change due to
rotation. That point will now be clarified by considering the effect of
prestress Fo in a member. In this case, it is only necessary to modifY Eqs.
OVERVIEW 21
1.27 and 1.28 to read
Fx = [Fo + (L - Lo)k] (Lo : Ox) (1.41)
(1.42)
Then Eqs. 1.34 and 1.36 become
Fx = Fo + kox + . .. higher order terms (1.43)
Oy
Fy = Fo - + ... higher order terms (1.44)
Lo
In Eqs. 1.43 and 1.44 there is now a zeroth order term due to the prestress.
Keeping first order terms uniformly leads to the classical linear term and the
geometric term with the length change due to rotation appearing as a higher
order term. The right-hand-side term of Eq. 1.44 is the now familiar
"geometric effect" that appeared in this chapter as the "string effect" (section
1.4) and later as FjB (Eq. 1.17), a more general term, that was used to
construct the contribution of member i to the geometric stiffness matrix (Eq.
1.20). A fundamental theorem as depicted from Eqs. 1.43 and 1.44 may be
stated as
In the presence o/prestress, geometric nonlinearities are o/the same order

0/ magnitude as linear elastic effects in structures.
1.7 Problems
1. Redo the proof of geometric convergence for Newton's method given in

section 1.2 for the case in which f is an n-vector.
2. Show geometrically the conditions under which the "error" indicated in

Figure 1.3 is small. (Hint: B« 1)
3. Take a specific 2-bar truss and recreate the P-d relationship of Figure 1.7
numerically.
4. Redo Gere's example of Figure 1.10 for the case in which the frame has
"fixed" supports rather than the "pinned" supports shown in the figure.
5. Show that Eqs. 1.18 and 1.19 are identical.
6. Take a specific 2-bar truss and verify the "linear buckling load factor" for
the case in which B= 45°.
CHAPTER 2
LINEAR STRUCTURAL ANALYSIS
2.1 Introduction
This chapter is restricted to a review of the displacement or node method of

linear structural analysis. Many good references on this topic are available to
the reader. (See for example Spillers, 1972.)
U sing the node method, the equilibrium equations of each structural
joint are written in terms of the joint displacements giving a square, linear
system of equations to be solved by the computer. These equations are
commonly written in matrix form as
(2.1)
where
elastic stiffness matrix (n x n)

joint displacement matrix (n xl)
joint load matrix (n x 1)
Physically, an element of K E, (K E)!i' is the force at "i" due to a unit

displacement at ")". That can be seen clearly by selecting 3 as indicated in
Eq. 2.2 below and forming the product KEel,
(K E )!) (K E )!}
0
(KE )!i (KE )!i row I (2.2)
rowj
0
The elastic stiffness matrix is also symmetric. This may be argued from
many points of view, for example, in terms of the conservation of energy but
that issue will not be pursued here.
A
m
member i element i
j m
A C
[[Xl [xl X1
[[X] [X]] row A
(KE)i = [X] [X] rowC (KE)i = [X] [x] [[x] 1
[X] [x] [x] m
Figure 2.1 Contributions to the stiffness matrix.
In practical terms the matrix K E is usually constructed by summing

the contribution of each bar or finite element to that matrix. Given the above
physical interpretation of the elements (KE)ij' it follows that a member with
two ends contributes four (matrix) terms to the stiffness matrix and that an
element with three nodes contributes nine. (See Figure 2.1). The size of
these matrix terms depends on the degrees of freedom of each node. The size
of the matrix term for plane trusses, for example, is 2 X 2 whereas for plane
frames it is 3 x 3 etc. This is because trusses have two degrees of freedom at
each node and plane frames have three degrees of freedom at each node.
The equations of the node method can be written in many forms. The
form used here is motivated by both network methods and the equations of a
linear elastic continuum:
the node method elasticity
node equilibrium NTF=P (J ij,i +fj =0

constitutive eqs. F=KA (Jij = 2f.1Gij + AbijGkk
member node disp. eqs L1=No Gij = (ui,j +u j ,J/2
Here
LINEAR STRUCTURAL ANALYSIS 25
F member force matrix stress tensor O"ij
A member displacement matrix strain tensor cij
K primitive stiffness matrix Lame constants A.,fJ-

0 node displacement matrix displacement vector ui
P node load matrix body force vector Ii
and 6ij is Kronecker's delta.
It follows directly from these equations that NT F=P =>
NTKA = P=> NTKNo = P=> KE = NTKN and that once has KEo = P
been solved for 0, F and A follow as A = No and F = KA . Furthermore,
when K E is written as K E = NT KN, its symmetry only requires the
symmetry of the matrix K.
2.2 The Truss Problem
In order to be specific and anticipating some of the material to follow, the

matrix formulation of the truss problem will now be described. Other types
of structures will be discussed later as the need arises.
f;"' -
\',
\
_
'
---
\ displaced
position
z initial
position
Typical Joint i
(RI)x (PI)x (01 )x

(RI)y (PI)y (01 )y
(RI)z (PI)z (0 1 )z
(R 2 )x (P2 )x (02)x
(R 2 )y (P 2 )y (02)y
(R 2 )z (P2 )z (02)z
Figure 2.2 Joint (node) quantities.

Figure 2.3 Typical bar i.
For the purposes here, a truss is a collection of pin-connected

members in which loads are only applied to nodes. In a typical truss,
associated with each node i is a position vector R;, a load vector P;, and a
displacement vector 0;. In matrix formulations these are collected in column
matrices as indicated in Figure 2.2. (At this point, the use of matrices is only
a matter of bookkeeping.)
Next, a typical truss member can be described in terms of its member
force F;, its member displacement (length change) it;, and a unit vector n;
which describes the slope of the bar. (See Figure 2.3.)
Figure 2.4 can now be used to discuss the equations of node
equilibrium. There are two cases to be considered here: a bar is either
positively or negatively incident upon any node. For the case of Figure 2.4
node equilibrium is simply
(2.3)
It can be shown that node equilibrium in matrix form can be written as
(2.4)
where the matrix N is defined as
if node j is the + end of member i

if node j is the - end of member i (2.5)
otherwise
Here n; is the 3 x 1 column matrix of unit vector components.

Figure 2.4 Equilibrium of node i.
Before the matrix formulation can be completed two more relations

are needed. First, there is the approximate (linear) member/node
displacement relationship which was discussed in Chapter 1
(2.6)
It can be verified by direct calculation that the matrix equation
A=No (2.7)
reproduces Eq. 2.6 for each bar in the system.

Finally there is the constitutive equation Fj = k j L1 j where
kj = AjE / L j ; A i = cross sectional area of member i; L i = length of
member i and E = Young's modulus. The constitutive relations can be
written in matrix form as
F=KA (2.8)
where K is diagonal with elements Ku = k j = AjE / L j •

Given the matrices of the node method it is a simple matter to form
the system matrix K E = NT KN. Rather than doing so directly, the
contribution of member i to the system matrix, (K E) j' is usually formed
allowing the system matrix to be assembled as the sum of the contributions

of all of the members. For the truss bar of Fig. 2.3 this contribution turns out
to be
colA colC
row A
(2.6)
rowC
with
(nj)x(n;)y
2
(n; )y (n;)y(n;)z (2.7)
2
(nj)z(nj)y (n; )z
and
This contribution can be computed in many ways. For example,

Figure 2.5 describes how this can be done by applying the interpretation, that
the elements of the stiffness matrix are forces due to unit displacements. In
this figure n; indicates the original direction of the undeformed bar and also
(from linear theory considerations) the direction of the deformed bar. Let,
(0 A)x = 1 with all other displacement components held at zero. The member
length change is then:
the member force:
and the end-of-member forces are:

r---~,--,r---------------~,~,------~,,~--+X
________ ~,L----------------------~----------~
---------..!~
z
A·E
11 = I.\n;),. -'-;
L;
A·E A-E {_ '\ A-E
I1x = (n;),.(n;)"-'-;l1y = (n;),.(n;\-'-;l1z = (n;),.'I.'i~ -'-
~ ~ ~
Figure 2.5 Truss element stiffness.
This step has produced the elements of (KE)j which lie in column
(3*A-2) of the global stiffness matrix. The terms in the other columns follow
in a similar fashion.
The computer programs that come with the CD which is included with
this book show how these ideas can be implemented. Here it will simply be
noted that all programs for linear structural analysis have three steps:
Step 1. Form the stiffness matrix K E by summing the contributions

to it of the various structural elements.
Step 2. Solve the system KEo = P for 0
Step 3. Compute the forces F and the member displacements A in

the structural elements.
2.3 Computer Programs
This section presents the program TR3D.FOR that performs linear analysis
on three dimensional trusses. A two dimensional version, TR2D.FOR, is also
discussed. A listing ofTR3D.FOR is provided in Appendix 5. This program
illustrates the concepts of linear structural analysis and serves as a basis for
the nonlinear analysis computer programs to follow.
It should be noted that sparse matrix methods are not used in this
book and that the solver is typically a Gaussian elimination program which
does NOT pivot. All supports are required to be completely fixed with
support joints numbered last. Fictitious fully fixed members with infinite
areas of cross section may be added to partially fixed joints (roller supports)
to meet the above requirement. All loads are joint loads.
2.3.1 Programs P1-TR3D.FOR and P2-TR2D.FOR
It is appropriate to start by noting that TR2D.FOR is directly obtained from

TR3D.FOR by changing 3-D vectors to 2-D vectors wherever they are
associated with joint entities. With regard to the coding:
1. MAXC is the dimension of array C.
2. Young's modulus E is set to 30,000,000 psi. (This is the only
place units appear in the program. For example if E has units of psi then the
load matrix P should have units of pounds and the coordinate matrix R
should have units of inches.) Conversion of the program to other system of
units is thus trivial.
3. Read as input and echoed as output are the number of bars NB, the
number of nodes NN, the number of supports NS, the joint coordinate array
R ( I) and the joint load array P ( I ) •
4. The system matrix KE is constructed as matrix C (I, J). Its
elements are first set to zero and then the contributions of all members (Eq.
2.7) are added. This step uses subroutine UNITV to generate the unit vector
lli of member i and subroutine INSERT to actually assemble c. Associated
with each member the quantities NP (I) , NM (I), and S (I) are read and
echoed. These are the "+ end", the "- end" and the area of bar i respectively.
2.4 Examples
2.4.1 Example 2.1. A 3-Bar Truss
A statically indeterminate three bar plane truss is fully fixed at nodes 2, 3

and 4 and loaded at node 1 with a horizontal load of -10,000 lbs and a
vertical load of -5,000 lbs. The connectivity of the members is shown in
Figure 2.6 which was generated using PLOT3D.FOR with the data file
TR2D21.DAT. Member lengths are 10, 1O...J2 and 20 inches. All member
areas are .657 square inches and E=30,OOO,OOO psi.
Listed below is the data file which is read by the linear analysis
program TR2D.FOR. The first line provides the number of bars, the number
of nodes and the number of supports. The next four lines supply information
on each of the four nodes. Each line gives the x-coordinate, the y-coordinate,
2 1 1
3 3
Figure 2.6 The three bar truss.
the horizontal load and the vertical load respectively. Finally each of the last
three lines supplies connectivity details (node number of the + end and the
node number of the - end) and the member area. Young's modulus is
specified within the program.
Data file TR2D21.DAT
3 4 3
10. 20. -10000.0 -5000.0
20.
10.
10.0
1 2 .657
1 3 .657
1 4 .657
The output file below first echoes the data and then lists the nodal
displacements, member elongations, member forces and member stresses.
This is also the order of their computation. The linear system KEo = P is
solved for the nodal displacements, 0, then the member deformations, ~, are
calculated from No = ~, the member forces, F, are obtained from F = ~
and the member stresses, (J, from (ji = Fj / Ai' where Ai is the area of cross-
section of member i.
Output file TR2D21.0UT

3 NO. MEMBERS
4 NO. NOOES
3 NO. SUPPORTS
COOROINATES LOAOS
X y PX PY
1 .100000000+02 .200000000+02 -.100000000+05 -.500000000+04
2 .000000000+00 .200000000+02 .000000000+00 .000000000+00
3 .000000000+00 .100000000+02 .000000000+00 .000000000+00
4 .100000000+02 .000000000+00 .000000000+00 .000000000+00
MEMBER + ENO - ENO AREA

1 1 2 .65700000E+00
2 1 3 .65700000E+00
3 1 4 . 65700000E+00
OISPLACEMENTS
X Y
1 -.333259380-02 -.159162100-02
2 .000000000+00 .000000000+00
3 .000000000+00 .000000000+00
4 .000000000+00 .000000000+00
MEMBER OL FORCE STRESS

1 -.333259380-02 -.656854250+04 -.999778150+04
2 -.348194570-02 -.485281370+04 -.738632220+04
3 -.159162100-02 -.156854250+04 -.238743150+04
Stop - Program terminated.
2.4.2 Example 2.2. A 24-Bar Space Truss Dome
A twenty four bar space truss which is fixed at its edges and loaded
vertically downwards at its central node with a load of 220.46 Ibs. is shown
in Figure 2.7. EA is constant for all the members with E=30,000,000 psi and
A=0.0155 square inches.
Here again PLOT3D.FOR was activated to generate the drawings of
Fig. 2.7 with the analysis data in TR3D22.DAT. That data was read by the
three dimensional truss analysis program, TR3D.FOR and results saved in
TR3D22.0UT. This example falls in the category of single layered
reticulated shallow shells that are apt to experience loss of stability (snap
through) at rather unexpectedly low loads. The example will be further
studied in the next chapter when the geometrically nonlinear analysis tools
will have been introduced.
24 13 6
.000000 .000000 3.234646 -220.46
4.921303 8.523848 2.447244
-4.921246 8.523880 2.447244
13
12
11
Figure 2.7 Plan and schematic views of the twenty four bar truss dome.
-9.842520 .000023 2.447244

-4.921286 -8.523857 2.447244
4.921228 -8.523890 2.447244
9.842520 -.000043 2.447244
.000025 19.685040 .000000
-17.047720 9.842553 .000000
-17.047770 -9.842475 .000000
-.000065-19.685040 .000000
17.047700 -9.842588 .000000
17.047790 9.842440 .000000
1 3 .1550E-Ol
2 3 .1550E-Ol
3 8 .1550E-Ol
2 8 .1550E-Ol
1 4 .1550E-Ol
3 4 .1550E-Ol
4 9 .1550E-Ol
3 9 .1550E-Ol
1 5 .1550E-Ol
4 5 .1550E-Ol
5 10 .1550E-Ol
4 10 .1550E-Ol
1 6 .1550E-Ol
5 6 .1550E-Ol
6 11 .1550E-Ol
5 11 .1550E-Ol
1 7 .1550E-Ol
6 7 .1550E-Ol
7 12 .1550E-Ol
6 12 .1550E-Ol
1 2 .1550E-Ol
7 2 .1550E-Ol
2 13 .1550E-Ol
7 13 .1550E-Ol
Output file TR3D22.0UT
24 NO. MEMBERS
13 NO. NODES
6 NO. SUPPORTS
COORDINATES
x y Z
1 .OOOOOOOOD+OO .OOOOOOOOD+OO .32346000D+Ol
2 .49212500D+Ol .85239000D+Ol .24472000D+Ol
3 -.49212500D+Ol .85239000D+Ol .24472000D+Ol
4 -.98425000D+Ol .OOOOOOOOD+OO .24472000D+Ol
5 -.49212500D+Ol -.85239000D+Ol .24472000D+Ol
6 .49212500D+Ol -.85239000D+Ol .24472000D+Ol
7 .98425000D+Ol .OOOOOOOOD+OO .24472000D+Ol
8 .OOOOOOOOD+OO .19685000D+02 .OOOOOOOOD+OO
9 -.17047200D+02 .98425000D+Ol .OOOOOOOOD+OO
10 -.17047200D+02 -.98425000D+Ol .OOOOOOOOD+OO
11 .OOOOOOOOD+OO -.19685000D+02 .OOOOOOOOD+OO
12 .17047200D+02 -.98425000D+Ol .OOOOOOOOD+OO
13 .17047200D+02 .98425000D+Ol .OOOOOOOOD+OO
LOADS
PX PY PZ
1 .OOOOOOOOD+OO .OOOOOOOOD+OO -.22046000D+03
2 .OOOOOOOOD+OO .OOOOOOOOD+OO .OOOOOOOOD+OO
MEMBER + END - END AREA

1 1 3 .15500000E-Ol
2 2 3 .15500000E-Ol
3 3 8 .15500000E-Ol
4 2 8 .15500000E-Ol
5 1 4 .15500000E-Ol
6 3 4 .15500000E-Ol
7 4 9 .15500000E-Ol
8 3 9 .15500000E-Ol
9 1 5 .15500000E-Ol
10 4 5 .15500000E-Ol
11 5 10 .15500000E-Ol
12 4 10 .15500000E-Ol
13 1 6 .15500000E-Ol
14 5 6 .15500000E-Ol
15 6 11 .15500000E-Ol
16 5 11 .15500000E-Ol
17 1 7 .15500000E-Ol
18 6 7 .15500000E-Ol
19 7 12 .15500000E-Ol
20 6 12 .15500000E-Ol
21 1 2 .15500000E-Ol
22 7 2 .15500000E-Ol
23 2 13 .15500000E-Ol
24 7 13 .15500000E-Ol
DISPLACEMENTS
X Y Z
1 -.48536751D-17 -.11946443D-17 -.20641184D+00
2 .37159444D-02 .64361834D-02 .91781931D-02
3 -.37159444D-02 .64361834D-02 .91781931D-02
4 -.74321153D-02 -.29720790D-19 .91780850D-02
5 -.37159444D-02 -.64361834D-02 .91781931D-02
6 .37159444D-02 -.64361834D-02 .91781931D-02
7 .74321153D-02 -.96950537D-18 .91780850D-02
MEMBER DL FORCE STRESS

1 -.97840054D-02 -.46076256D+03 -.29726617D+05
2 .74318888D-02 .35111285D+03 .22652442D+05
3 -.24987545D-02 -.93394741D+02 -.60254673D+04
4 -.24987545D-02 -.93394741D+02 -.60254673D+04
5 -.97838205D-02 -.46075567D+03 -.29726172D+05
6 .74319838D-02 .35111595D+03 .22652642D+05
7 -.24986957D-02 -.93394458D+02 -.60254490D+04
8 -.24987116D-02 -.93396601D+02 -.60255873D+04
9 -.97840054D-02 -.46076256D+03 -.29726617D+05
10 .74319838D-02 .35111595D+03 .22652642D+05
11 -.24987116D-02 -.93396601D+02 -.60255873D+04
12 -.24986957D-02 -.93394458D+02 -.60254490D+04
13 -.97840054D-02 -.46076256D+03 -.29726617D+05
14 .74318888D-02 .35111285D+03 .22652442D+05
15 -.24987545D-02 -.93394741D+02 -.60254673D+04
16 -.24987545D-02 -.93394741D+02 -.60254673D+04
17 -.97838205D-02 -.46075567D+03 -.29726172D+05
18 .74319838D-02 .35111595D+03 .22652642D+05
19 -.24986957D-02 -.93394458D+02 -.60254490D+04
20 -.24987116D-02 -.93396601D+02 -.60255873D+04
21 -.97840054D-02 -.46076256D+03 -.29726617D+05
22 .74319838D-02 .35111595D+03 .22652642D+05
23 -.24987116D-02 -.93396601D+02 -.60255873D+04
24 -.24986957D-02 -.93394458D+02 -.60254490D+04
2.4.3 Example 2.3. A 72-Bar Double Layered Grid
A seventy two bar double layered square grid (see Figure 2.8) having
members of equal length (54 inches) which is 118 symmetric is solved in this
example. (See Levy and Hanaor, 1986 for a solution with optimal prestress.)
The grid is loaded vertically upwards at its central node with a load
of 13,534 lbs. All circumferential nodes are restrained in the vertical
direction only. The central node is restrained in both horizontal x and y
directions and one of the comer nodes is restrained perpendicular to the
main diagonal in a plan projection to prevent rigid body rotation.
Figure 2.9 shows 118 of the structure. It is plotted using
PLOT3D.FOR with the appropriate analysis data file TR3D23.DAT. The
plot shows 20 members instead of 12 and this is because an additional 8
"fictitious" members having areas of cross-section of 1,000,000 square
inches were added to accommodate non fully fixed joints. The supports were
established as follows: two fictitious bars are introduced at the
circumferential nodes 4 and 5 in the vertical direction, two bars
accommodate the original restraints of the central node, three fictitious bars
perpendicular to the main diagonal in the plan projection restrain movement
in that direction (because of symmetry) at nodes 2, 3, and 4, and finally one
Figure 2.8 Plan and schematic views of the 72-bar double layered grid.
fictitious bar was added to node 6 in the x-direction (symmetry). Two fixed
supports are added to the "cut" bars 3 and 10. Note that the areas of cross-
section are halved for members 1,2,4 and 5 and the resulting forces should
be doubled in these members when considering the full grid. The load is 1/8
that of the full load i.e. 1691.75 Ibs.The output file TR3D23.0UT lists the
analysis results in the now familiar format.

20 16 10
O. O. 38.1834 1691.75
27 . 27.
54. 54. 38.1834
111 6 14
1 13
101 1 12
13
Figure 2.9 Node map of 118 of the double layered grid.

8I. 8I.
27. 8I.
54. 38.1834
27.
8I.
-10. 38.1834
-10. 38.1834
-10. 54. 38.1834
54.
8I. 27. 38.1834
108. 54.
8I. 8I. -10.
27. 8I. -10.
1 6 . 11065e+00
1 2 .11065e+00
7 2 .2213
2 3 .11065e+00
3 4 .11065e+00
5 4 .2213
6 3 .2213
2 5 .2213
6 5 .2213
8 5 .2213
3 5 .2213
2 6 .2213
1 9 10.e5
1 10 10.e5
6 11 10.e5
2 12 10.e5
3 13 10.e5
4 14 10.e5
4 15 10.e5
5 16 10.e5
OUtput file TRJD23.0UT

20 NO. MEMBERS
16 NO. NOOES
10 NO. SUPPORTS
COOROINATES
X Y Z
1 .000000000+00 .000000000+00 .381834000+02
2 .270000000+02 .270000000+02 .000000000+00
3 .540000000+02 .540000000+02 .381834000+02
4 .810000000+02 .810000000+02 .000000000+00
5 .270000000+02 .810000000+02 .000000000+00
6 .000000000+00 .540000000+02 .381834000+02
7 .000000000+00 .270000000+02 .000000000+00
8 .000000000+00 .810000000+02 .000000000+00
9 .000000000+00 -.100000000+02 .381834000+02
10 -.100000000+02 .000000000+00 .381834000+02
11 -.100000000+02 .540000000+02 .381834000+02
12 .540000000+02 .000000000+00 .000000000+00
13 .810000000+02 .270000000+02 .381834000+02
14 .108000000+03 .540000000+02 .000000000+00
15 .810000000+02 .810000000+02 -.100000000+02
16 .270000000+02 .810000000+02 -.100000000+02
LOAOS
PX PY PZ
1 .000000000+00 .000000000+00 .169175000+04
2 .000000000+00 .000000000+00 .000000000+00
3 .000000000+00 .000000000+00 .000000000+00


1 1 6 .11065000E+00
2 1 2 .11065000E+00
3 7 2 .22130000E+00
4 2 3 . 11065000E+00
5 3 4 .11065000E+00
6 5 4 .22130000E+00
7 6 3 .22130000E+00
8 2 5 .22130000E+00
9 6 5 .22130000E+00
10 8 5 .22130000E+00
11 3 5 .22130000E+00
12 2 6 .22130000E+00
13 1 9 .10000000E+07
14 1 10 .10000000E+07
15 6 11 .10000000E+07
16 2 12 .10000000E+07
17 3 13 .10000000E+07
18 4 14 .10000000E+07
19 4 15 .10000000E+07
20 5 16 .10000000E+07
DISPLACEMENTS
X Y Z
1 .39875312D-09 .87725703D-09 . 16718938D+00
2 -.18000648D-01 -.18000657D-01 .86690986D-Ol
3 .97300728D-02 .97300753D-02 .25456898D-01
4 -.43785307D-02 -.43785310D-02 .56391543D-I0
5 -.24325188D-02 -.26757731D-01 .50752512D-09
6 .39875279D-09 .23352202D-01 .53665901D-Ol

1 .23352201D-01 . 14355117D+04 .12973445D+05
2 .38919949D-01 .23925073D+04 .21622297D+05
3 -.18000648D-01 -.44261595D+04 -.20000720D+05
4 - .15567972D-Ol -.95700244D+03 -.86489147D+04
5 .38919864D-02 .23925020D+03 .21622250D+04
6 -.19460118D-02 -.23925135D+03 -.10811177D+04
7 .97300724D-02 .11962584D+04 .54055958D+04
8 -.87570745D-02 -.10766337D+04 -.48650414D+04
9 .11675988D-01 .14355048D+04 .648669130+04
10 -.24325188D-02 -.59812937D+03 -.27027987D+04
11 .58379927D-02 .71775224D+03 .32433449D+04
12 - .11675988D-01 -.14355048D+04 -.64866913D+04
13 .87725703D-09 .26317711D+04 .26317711D-02
14 .39875312D-09 . 11962594D+04 .11962594D-02
15 .39875279D-09 .119625840+04 .11962584D-02
16 -.624447440-08 -.49061224D+04 -.490612240-02
17 .17226127D-08 .13534123D+04 .13534123D-02
18 -.21532622D-09 -.16917625D+03 -.169176250-03
19 .56391543D-10 .169174630+03 .16917463D-03
20 .507525120-09 .152257540+04 .152257540-02
2.5 Problems
1. Show that Eq. 2.4 (NT F = P ) reproduces the node equilibrium in Eq. 2.3
for each node in the system.
2. Show that Eq. 2.7 (A = No) reproduces Eq. 2.6 for each bar in the system.
3. Modify TR2D.FOR to accommodate prestress through "lack of fit". (Hint:

use Fi = Ki (L1i - D i ) instead of Fi = KiL1i for the constitutive relation. Here
Di is the "lack of fit" of member i in units of length. )
4. Comment on the use of the above modified program to accommodate

thermal changes in truss members. (Hint: To what extent are "lack of fit"
problems equivalent to thermal stress problems?)
5. Member 2 of Example 2.1 was manufactured 0.001667 inches longer than

necessary. Use the modified program of problem 1 to analyze this truss. (For
a discussion on use of "lack of fit" in the optimal design of trusses see Levy,
1989)
6. If the allowable stress is 10,000 psi, what percentage savings in material

does prestress introduce if all members retain equal cross sectional areas?
CHAPTER 3
"EXACT" ANALYSIS OF TRUSSES
3.1 Introduction
A structure must satisfy the conditions of equilibrium (deformed equilibrium

in the nonlinear case), the pieces of the structure must fit together in their
loaded state and a constitutive equation must be satisfied. In this chapter all
these conditions are met for the truss. The only restriction here is that the
truss bars or elements be linear springs whose forces are proportional to their
true length change.
This exact analysis of trusses is a nonlinear problem and therefore
some iterative analysis procedure can be expected. (Actually, Newton's
method is used). A typical step of this analysis can be described as follows.
Given a fixed joint load matrix P and a starting configuration which is not in
equilibrium with this joint load matrix (If it were there would be no analysis
to perform.), the following sequence of actions must be taken:
• Compute the unbalanced load P'. Since the member forces F are not in
equilibrium with the given load P, the unbalanced load can be computed as
P'=P-NTF.
• Solve for the incremental displacements. Under the unbalanced load P' the
structure will displace. This computation involves solving the system
(KE + KG)o = P' for the node displacement matrix 0 . (The geometric
stiffness matrix KG will be discussed below).
• Update the coordinates. The coordinate matrix R is updated to describe the
deformed shape as R-+R + O.
• Compute new member forces. The new coordinates of the structure imply
new member lengths which in turn imply new member forces. The "exact"
length change will be used here rather than the approximate expression
A=No.
• Repeat this sequence of calculations. Computation stops when the
unbalanced load P' -+ O. It will be argued below that these calculations
comprise the application of Newton's method to the equations of node
equilibrium.
3.2 Linearization of the Joint Equilibrium Equations
The above analysis is based on the node equilibrium equations which will be
written here as NT F = P. While this is the same as the form used for
equilibrium in linear structural analysis, here it is assumed that NTF = P.
describes deformed equilibrium. Under a perturbation this equation takes the
form,
(3.1)
It is assumed that all variables can be described in terms of the joint

coordinate matrix R and its perturbation, the node displacement matrix 0 . In
Eq. 3.1 the term NT dF describes (under the chain rule of partial
differentiation) the change in member forces with the matrix N fixed which
is simply linear elastic analysis. That is,
(3.2)
It is argued here that the variation of the member forces, dF, with the
equilibrium "operator", N, fixed simply returns linear elastic theory. It is
then the term dN TF of Eq. 3.1 which gives rise to the geometric stiffness
matrix KG ' i.e.
(3.3)
How the matrix KG can be computed as a gradient will now be explained.

This argument is based on the fact that a small variation df of a function f
can be represented as
df-Vf ·dx (3.4)
Vf is of course the gradient matrix of the function f. A vector, f, of n

functions with m variables has an nxm gradient matrix as indicated in Eq. 3.5
below.
"EXACT" ANALYSIS OF TRUSSES 43
oil/oXI Ofl/oX2
X2 Of2 1oX I Of2 1oX2

f= x= Vf= (3.5)
... O.f
:J n
lax m
3.3 The Geometric Stiffness Matrix
In order to complete the details of the derivation of the geometric stiffness

matrix as a gradient, it is convenient to work with (N T F) i' the contribution
of bar i to the equilibrium equations. More explicitly, this term is shown
below for the typical bar of Fig. 2.3 as
(nJx
(ni )y -rowA
(ni)z
(NTF)i = Fi (3.6)
- (ni)x
-(nJy -rowe
-(nJz
It is the gradient of Eq. 3.6 which gives the contribution of bar ito
the geometric stiffness matrix. Referring back to the original discussion of
Newton's method, for the case of the equilibrium equations discussed here,
f ~ (NTF)i = NiTFi
X ~ R (3.7)
dx ~ ()
and
Now since NT is a function of the coordinates xA, YA' zA' and

xc' Yc , Zc , chain rule of differentiation yields,
oNT dx oNT d oNT dz

( - - A + - - !YA + - - A
OXA ~A OZA
(3.9)
With dxA = (OAt, dYA = (OA)y' dzA = (OA)z' dxc = (Oct,

dyc = (oc)y, and dzc = (oc)z, i.e. the changes in coordinates are
interpreted as nodal displacements, the contribution of member i to the
geometric stiffness matrix becomes
l
colA colC
(KG)i = T
V(Ni F)i =
V(NT)AA V(NT)
AC
1Fi row A
(3.10)
V(NT)cA V(NT)cC rowC
where
o(n;)x o(ni )x o(ni )x

oXA ~A oZA
o(ni )y o(ni)y o(nJ y
v(NTj = oXA oZA
~A
o(ni )z B(ni )z B(ni )z
OxA ~A oZA
8Cn i )x 8Cn i )x 8Cn i )x

&C eyc ac
8Cn i )y 8CnJy 8CnJy
&C eyc ac
8CnJz 8CnJz 8Cn i )z
&c eyc ac
8Cn i )x 8Cn i )x 8Cn i )x

8xc ayc 8z c
T)CC 8CnJy 8CnJy 8CnJ y
V(N j = - ------''--
8xc ayc 8z c
8Cn i )z 8CnJz 8Cn i )z
8xc ayc 8z c
and
8Cn i )x 8CnJx 8Cn i )x

8XA ayA 8z A
T)CA 8CnJy 8Cn i )y 8CnJy
V (N j = ------'-
8xA ayA 8z A
8Cn i )z 8Cn i )z 8Cn i )z
8xA ayA 8z A
The derivatives indicated in Eq. 3.10 can now be computed

completing the description of the geometric stiffness matrix. Symbolically,
colA colC
C3 .11)
rowC
with
1- (nJ; -(nit(ni)y -(ni)x(nit

(KG)~ = ;1
1
-(ni)y(nJx 1-(n i ) / - (ni)y(nJz (3.12)
-(nit(nit -(nit(ni)y 1-(n i ) /
or
(3.13)
where
(3.14)
Equation 3.12 follows directly from differentiation as will be shown below.

It is appropriate to mention though that the geometric matrix of Eq 3.11 is
identical to that of Eq 1.19 which was derived via a geometric observation
concerning string effects. It is also identical to that of Eq. 1.44 which was
derived via the fundamental theorem of prestress.
Consider the x-component of the unit vector n i when written in
terms of nodal coordinates,
(3.15)
Differentiation yields,
(3.16)
8(n i )x =_! XA -xc 2( _ )= __1 (n.) (n.) (3.17)

8 2 3 Y A Yc L 1 X 1 Y
YA Li i
(3.18)
The derivatives of (n J y and (n J z with respect to oxA, 8y A and OZ A

follow in a similar fashion given the form of Eg. 3.15. Derivatives with
respect to the coordinates at end C are the negative of derivatives with
respect to the coordinates at end A.
3.4 Overall Buckling
The approach to buckling used in this book is one in which load is

incremented and the determinant of the system matrix, IKE + KG I, is
monitored. When this determinant goes to zero the structure is said to be
unstable. (See Appendix 2.)
The relationship between the physical phenomenon of buckling
(collapse) and the formal concept of the determinant becoming zero is non-
trivial except perhaps in the scalar case shown in Fig. 3.1. In this case it
seems clear that when the load displacement curve becomes nearly flat, a
"small" increment of load will produce a "large" increment of displacement.
But even in this simple case it is not common to follow the phenomenon to
collapse which is what bucking implies. (Perhaps the structure will "snap-
through" and recover its stability.)
There are several ways to argue the relationship of the determinant
of the system matrix and buckling. It is most simple to argue that the
determinant is the generalization of the slope of the curve in the scalar case
above to the case of n dimensions. A more formal approach is through the
idea of a "condition number" (Isaacson and Keller, 1966). Roughly a system
Ax=f can be considered in its perturbed state as (A+dA)(x+dx)=f+df. In a
well-posed problem, a small perturbation of the system parameters produces
a small perturbation of the state of the system. The condition number p,
J.1 = I/A -ll/·IIAII

can be used to formalize the idea of an ill-conditioned system. It can be
shown that when the condition number is not too large, bounded df and dA
imply bounded dx.
There are approaches to buckling other than incrementing the load
on a structure and monitoring its behavior (the value of the determinant) but
these other approaches are argued here to be less accurate. One such
approach involves reducing buckling to an eigenvalue problem. This
approach was briefly discussed in Chapter 1 and termed "linear buckling".
For the case of the truss linear (eigenvalue) buckling is particularly simple to
discuss.
load
_L _________ :.::_,___
buckling load ;;. _
displacement
Figure 3.1 Load deflection curve showing buckling.
For trusses as has just been indicated, the geometric stiffness matrix
is linear in the member forces. For example, if KG(FI ) is the geometric
stiffness matrix corresponding to some load PI' then PK o (FI ) is the
geometric stiffness matrix which corresponds to the load pPI . It is argued
that, at buckling, for a given load P and corresponding buckling load,
PeR = f3P there are many deformed equilibrium positions. If that is true,
(3.19)
and
(3.20)
which implies that there exists a () such that
(3.21)
or
(3.22)
which implies a generalized eigenvalue problem. That is, the given load P
implies KG and buckling becomes an eigenvalue problem for the unknowns
f3 and (). Clearly, P itself can be computed from the determinantal equation
(3.23)
This type of analysis depends heavily upon the initial configuration which is
used to construct both K E and KG' Since incrementing the loads to find
the buckling load does not depend upon the initial configuration (this relates
also to the question of "imperfection sensitivity") in this manner, it is argued
that doing so is superior to the eigenvalue approach. There are other
arguments against the eigenvalue approach which have been presented in
section 1.5.6 where it was shown that the two approaches to buckling can
lead to quite different results.
This section presents four computer programs which are direct extensions of
the linear analysis programs PI-TR3D.FOR and P2-TR2D.FOR that are
discussed in Chapter 2. The first two programs perform nonlinear analysis
and nonlinear buckling for two and three dimensional trusses. The other two
programs determine linear buckling loads using the eigenvalue approach.
3.5.1 Programs P3-TR3DNL.FOR and P4-TR2DNL.FOR
These two powerful nonlinear analysis programs are iterative but converging
rapidly. They signal buckling as well and indeed achieve critical loads if fine
tuned to desired accuracy.
A brief comment on "exact" buckling loads. The procedures
described below involve incrementing the load and monitoring the
determinant of the system matrix. When this determinant becomes negative
the buckling load has been exceeded. Clearly, smaller steps will give a closer
bound on the buckling load.
There are two reasons for not attempting extremely small load steps.
From an engineering point of view it is typically unnecessary to have an
"exact" buckling load; from a computational point of view, as you get closer
and closer to the buckling load the system becomes more and more singular
and convergence becomes more and more difficult.
Following are some of the steps that have to be taken to go from
linear analysis to nonlinear analysis:
• Assembling the geometric stiffness matrix, KG'
• Iterating. In order to solve a nonlinear system it is typically necessary to

use some type of iterative method. (Here Newton's method is used.) In the
programs this iteration is described by the parameter ITER. These are
iterations perfonned with fixed load. Another parameter LDSTP describes

the load incrementing. This is first of all required for buckling problems
where the buckling load is not known and some load range which includes it
is to be specified. Alternatively as a practical matter, nonlinear problems are
sometimes sensitive to the manner in which load is applied. It is not
uncommon for an analysis to diverge if the load is applied in one step but to
converge when it is applied in increments. (This relates to the convergence
of Newton's method which is most commonly shown to be local. That is,
Newton's method can be shown to converge when you start near the answer.)
• Saving data. Since more than one analysis is to be performed, some data
which would ordinarily be discarded must be kept. This is true for the joint
loads PSAVE and the member forces FSAVE.
• Computing the unbalanced load. It is the unbalanced load rather than the
total load which drives incremental analysis. The unbalanced load is
computed here during the construction of the system matrix where the
appropriate data is available.
• Computing Error. The error is measured here by computing the "length" of

the unbalanced load P' . That is the error is defined to be ~ L (Pi' )2 .
• True length. When member forces are updated their true length change is
used rather than the approximate length change of linear structures.
• Buckling. The linear equation solver is set up so that the detenninant can
be monitored by simply looking at its diagonal terms during the process of
Gaussian elimination. A zero or negative term along the diagonal generates a
message in the output.
• Computer output. Full output for linear analysis is collected from each step
of Newton's method and each load increment. This is generally too much
output but the selection of output has been left to the reader who may wish
to "scan" the output file. It is for example sometimes useful to scan the
output file for the tenn ERROR to examine convergence of the algorithm. (It
would have been better to plot this error and display it at as the program
runs.) It can also be useful to scan the output file for the tenn NEG when
concerned with buckling.
3.5.1.1 Nonlinear Analysis and Newton's Method
The iterative process for the nonlinear analysis of trusses is basically of the
form
(3.24)
Letting f(x(n») = P - NTF(n) and dx = c')(n+l) Newton's method of solving a

nonlinear system implies
which becomes
(3.25)
when applied to Eq. 3.24. For a direct correspondence between Eqs. 3.24 and
3.25 it remains to show that the gradient of the unbalanced load is the
tangential stiffness matrix. This gradient is now written as
But VP is zero since P is constant for that particular load step and it has
been shown in the previous sections that the gradient of the equilibrium
equations i.e. V(NTF(n)) reduces to the tangential stiffness matrix,
(KE + KG(F(n»)) making the recursive procedure of Eq. 3.24 identical to
Newton's method with its quadratic convergence properties.
3.5.2 Programs P5-BUCK3D.FOR and P6-BUCK2D.FOR
These two programs demonstrate the eigenvalue approach to truss buckling.

For eigenvalue truss buckling it is necessary to solve the "generalized "
eigenvalue problem
(3.26)
These programs have been assembled informally for specific

eigenvalue buckling problems. This has been done by taking parts from P3
and P4 which are used to generate KE and KG and then calling a "canned"
eigenvalue solver NROOT taken from the old IBM Scientific Subroutine
Package. Since NROOT requires that KE and KG be dimensioned exactly,
some re-coding must be done each time a problem of a different size is run.
(See also problem 7 at the end of the chapter.)
As can be seem from the listing of the code, this program starts with
an elastic analysis program which is first used (ITER=l) to generate the
member forces. As a second step (ITER=2) the stiffness matrices are
generated and the eigenvalue routine called. It should be noted that NROOT,
because of its structure generates -11 r.3 as eigenvalues. This implies that the
critical buckling load is obtained by multiplying the given load by a factor of
(-1 )/( smallest eigenvalue).
3.6 Examples
In this section six problems have been selected to give the reader an
overview of geometrically nonlinear analysis for trusses.
3.6.1 Example 3.1. Biot's 2-Bar Prestressed Truss
Nonlinear analysis is now performed on Biot's 2-bar prestressed truss of

Figure 1.5 which is "linearly unstable". (See Figure 3.2 generated by
PLOT3D.FOR.) It is, of course, the geometric stiffness matrix which makes
the analysis possible.
The analysis is performed using TR2DNL.FOR on the data file
TR2DNL31.DAT. Both the data file and a partial listing of the output are
given below. The first line in the data file includes the number of load steps
(lOin this case) and the number of iterations in each load step (5 in this
case) in addition to the usual number of bars, number of nodes and number
of supports. Node information is supplied in the same manner as for linear
analysis whereas member information includes the prestress (in this case
1,000 lbs) for each member. Thus the center node of this truss is loaded
downwards with 70 lbs and each member carries a prestress of 1,000 lbs for
equal areas of cross section of 0.0127 square inches. The given program was
edited and Young's modulus was compiled to be E=10,000,000 psi.
As for the output, the input data is first echoed and then for each
iteration of each load step nodal displacements, member deformations,
2 1 1 2 3
Figure 3.2 The linearly unstable 2-bar truss.
member forces and member stresses are provided in the same manner as in
the linear analysis. Moreover, here the new deformed equilibrium position
(UPDATED COORDINATES) and the resulting member forces at that
deformed position (UPDATED FORCES) are also given. In this particular
output set the last iteration (ITERATION 5)of the last load step
(LOADSTEP 10) is listed. The equilibrium position was attained with an
ERROR of .62413028D-ll which is basically zero error implying that the
structure is in complete equilibrium with no unbalanced forces.
Data file TR2DNL31.DAT
2 3 2 10 5
200. 00. 0.0 -70.
00. o.
400. O.
1 2 .0127 1000.0
1 3 .0127 1000.0
Output file TR2DNL31.0UT

2 NO. MEMBERS
3 NO. NODES
2 NO. SUPPORTS
10 NO. LOAD STEPS
5 NO. ITERATIONS
COORDINATES LOADS
x y PX PY
1 .20000000D+03 .00000000D+00 .00000000D+00 -.70000000D+02
2 .00000000D+00 .00000000D+00 .00000000D+00 .00000000D+00
3 .40000000D+03 .00000000D+00 .00000000D+00 .00000000D+00
MEMBER + END - END AREA PRESTRESS

1 1 2 .12700000E-01 .10000000E+04
2 1 3 .12700000E-01 .10000000E+04
****ITERATION NUMBER 5
LOAD STEP 10
LOADSTEP 10; ITERATION 5; ERROR .62413028D-11
DISPLACEMENTS
x Y
1 -.42816502D-16 .51889630D-12
MEMBER DL FORCE STRESS UPDATED FORCES

1 -.17044482D-13 -.10817434D-10 -.85176650D-09 .10682100D+04
2 -.16958895D-13 - .10763116D-10 -.84748945D-09 .10682100D+04
UPDATED COORDINATES
x y
1 .20000000D+03 -.65565399D+Ol
3.6.2 Example 3.2. A Prestressed Cablenet
In this case nonlinear analysis is performed on a prestressed cable-net.

A 4x4 square grid with 6 ft. spans for a total of 24 ft is transversely
loaded downwards with 10,000 lbs. Each cable has an area of cross section
of 0.25 square inches and carries a prestressing force of 20,000 lbs. Figure
3.3 displays the net. It was plotted using PLOT3D.FOR and the analysis data
file TR3DNL32.DAT.
The net is an extension of Example 3.1 to three dimensions and
resembles a trampoline in its loading. The structure is linearly unstable and
cannot be solved without the aid of the geometric stiffness matrix. The
nonlinear analysis program TR3DNL.FOR is invoked with E = 30,000,000
psi to yield a converged solution at iteration 6 of load step 20. (These
parameters are read in from the first line of the data file.) Listed below are
the first few lines of output that echo the input as well as the results of the
last iteration.
24 NO. MEMBERS
25 NO. NODES
16 NO. SUPPORTS
20 NO. LOAD STEPS
6 NO. ITERATIONS
COORDINATES
X Y Z
1 .60000000D+02 .18000000D+03 .OOOOOOOOD+OO
2 .12000000D+03 .18000000D+03 .OOOOOOOOD+OO
3 .18000000D+03 .18000000D+03 .OOOOOOOOD+OO
4 .60000000D+02 .12000000D+03 .OOOOOOOOD+OO
5 .12000000D+03 .12000000D+03 .OOOOOOOOD+OO
6 .18000000D+03 .12000000D+03 .OOOOOOOOD+OO
7 .60000000D+02 .60000000D+02 .OOOOOOOOD+OO
8 .12000000D+03 .60000000D+02 .OOOOOOOOD+OO
9 .18000000D+03 .60000000D+02 .OOOOOOOOD+OO
11 .60000000D+02 .24000000D+03 .OOOOOOOOD+OO
12 .12000000D+03 .24000000D+03 .OOOOOOOOD+OO
13 .18000000D+03 .24000000D+03 .OOOOOOOOD+OO
14 .24000000D+03 .24000000D+03 .OOOOOOOOD+OO
19 .60000000D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
10 11 12 13 14
13 17 21
15 1 1 2 2 3 3 4 23
14 18 22
16 5 4 6 5 7 6 B 24
15 19 23
17 9 7 10 8 11 9 12 25
16 20 24
18 19 20 21 22
Figure 3.3 The transversely loaded flat cable-net.
21 .18000000D+03 .000000000+00 .OOOOOOOOD+OO

22 .240000000+03 .000000000+00 .OOOOOOOOD+OO
23 .240000000+03 .180000000+03 .000000000+00
24 .240000000+03 .120000000+03 .000000000+00
25 .24000000D+03 .600000000+02 .000000000+00
LOADS
PX PY PZ
23 .000000000+00 .000000000+00 .000000000+00

24 .000000000+00 .000000000+00 .000000000+00
25 .000000000+00 .000000000+00 .000000000+00
MEMBER + ENO - ENO AREA PRESTRESS

1 15 1 .25000000E+00 .20000000E+05
2 1 2 .25000000E+00 .20000000E+05
3 2 3 .25000000E+00 .20000000E+05
4 3 23 .25000000E+00 .20000000E+05
5 16 4 .25000000E+00 .20000000E+05
6 4 5 .25000000E+00 .20000000E+05
7 5 6 .25000000E+00 .20000000E+05
8 6 24 .25000000E+00 .20000000E+05
9 17 7 .25000000E+00 .20000000E+05
10 7 8 .25000000E+00 .20000000E+05
11 8 9 .25000000E+00 .20000000E+05
12 9 25 .25000000E+00 .20000000E+05
13 11 1 .25000000E+00 .20000000E+05
14 1 4 .25000000E+00 .20000000E+05
15 4 7 .25000000E+00 .20000000E+05
16 7 19 .25000000E+00 .20000000E+05
17 12 2 .25000000E+00 .20000000E+05
18 2 5 .25000000E+00 .20000000E+05
19 5 8 .25000000E+00 .20000000E+05
20 8 20 .25000000E+00 .20000000E+05
21 13 3 .25000000E+00 .20000000E+05
22 3 6 .25000000E+00 .20000000E+05
23 6 9 .25000000E+00 .20000000E+05
24 9 21 .25000000E+00 .20000000E+05
LOAO STEP 20
ERROR = .647003480-02
OISPLACEMENTS
X y Z
1 .395412890-08 -.395412840-08 -.270709980-09
2 .246475820-15 -.116174000-07 -.447321380-09
3 -.395412840-08 -.395412850-08 -.270709750-09
4 .116173990-07 .229245380-15 -.447321680-09
5 -.216719710-15 -.227423960-15 .784636530-09
6 -.116174000-07 .229233670-15 -.447321420-09
7 .395412890-08 .395412890-08 -.270710470-09
8 .246490880-15 .116173990-07 -.447321830-09
9 -.395412840-08 .395412890-08 -.270710050-09
MEMBER OL FORCE STRESS UPOATEO FORCES

1 .395921120-08 .494776170-03 .197910470-02 .218980100+05
2 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
3 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
4 .395921070-08 .494776110-03 .197910440-02 .218980100+05
5 .116257760-07 .145053320-02 .580213300-02 .338899300+05
6 - .116765490-07 -.145684970-02 -.582739890-02 .339850200+05
7 - . 116765490-07 -.145684970-02 -.582739890-02 .339850200+05
8 .116257760-07 .145053330-02 .580213320-02 .338899300+05
9 .395921120-08 .494776170-03 .197910470-02 .218980100+05
10 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
11 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
12 .395921070-08 .494776110-03 .197910440-02 .218980100+05
13 .395921070-08 .494776110-03 .197910440-02 .218980100+05
14 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
15 -.394268300-08 -.492711880-03 -.197084750-02 .218795400+05
16 .39592112D-08 .49477617D-03 .19791047D-02 .21898010D+05
17 .11625776D-07 .14505333D-02 .58021332D-02 .33889930D+05
18 -.11676549D-07 - .145 68497D-02 -.58273989D-02 .33985020D+05
19 -.11676549D-07 -.14568497D-02 -.58273989D-02 .33985020D+05
20 .11625776D-07 .14505332D-02 .58021330D-02 .33889930D+05
21 .39592107D-08 .49477611D-03 .19791044D-02 .21898010D+05
22 -.39426830D-08 -.49271188D-03 -.19708475D-02 .21879540D+05
23 -.39426830D-08 -.49271188D-03 -.19708475D-02 .21879540D+05
24 .39592112D-08 .49477617D-03 .19791047D-02 .21898010D+05
UPDATED COORDINATES
X y z
1 .60000097D+02 .17999990D+03 - .13456794D+Ol
2 .12000000D+03 . 17994911D+03 -.26924913D+Ol
3 .17999990D+03 .17999990D+03 - .13456794D+Ol
4 .60050886D+02 .12000000D+03 -.26924913D+Ol
5 .12000000D+03 .12000000D+03 -.71144376D+Ol
6 .17994911D+03 .12000000D+03 -.26924913D+Ol
7 .60000097D+02 .60000097D+02 -.13456794D+Ol
8 .12000000D+03 .60050886D+02 -.26924913D+Ol
9 .17999990D+03 .60000097D+02 -.13456794D+Ol
3.6.3 Example 3.3. Plane Truss Buckling
A seven bar plane truss is solved for buckling (Figure 3.4 ). This problem is
taken form Timoshenko (1951) who solves it for
Here A and L refer to the horizontal bars and LI and Al refer to the inclined
bars. Two solutions are presented to this problem. The first is obtained by
iterative nonlinear analysis using TR2DNL.FOR whereas the second is
obtained using BUCK2D33.FOR to solve a generalized eigenvalue problem
("linear buckling"). For Al =2 square inches, A=200 square inches,
E=30,000,000 psi, Ll =120 inches and 8=45° Timoshenko's solution yields
PeR =21xlo6 Ibs.
A maximum load of 30,000,000 lbs is chosen as input for the
nonlinear buckling analysis. The output file TR2DNL33.0UT indicates a
negative term in the diagonal at load step 10. The last load for which the
analysis converged is then load step 9. This implies that the buckling load is
bounded by the loads at load steps 9 and 10. Thus:
10
Load step 9: 30,000,000 x - = 20,000,000 Ib
15
1 2 2
1 6 3
6 4 3 5 4
p p
P~P L L
.t • •.
Figure 3.4 Schematic and node map of the 7-bar truss.
Timoshenko:=> PCR =21,000,000 lbs
11
Load step 10: 30,000,000 x - = 22,000,000 lbs
15
If a more accurate value for the buckling load is needed a different

scheme can be used to generate load. But it should be noted again that as you
get closer to the buckling load, computation becomes more and more
difficult. In this sense a very accurate determination of the buckling load can
be counter productive.
The same load of 30,000,000 lbs is selected as input for the linear
buckling analysis. The data from BUCK2D33.DAT is first echoed in the
output file BUCK2D33.0UT, then the linear analysis results are printed and
finally the column of -1113 is listed. The last entry in this column provides the
smallest eigenvalue which is to be multiplied by the selected initial load, i.e.
PCR =30,000,000x(-1I-.14292360E+01)=21xlo 6 lbs.
It is interesting to note that both the linear and nonlinear buckling
loads are the same for this truss. This is rather an exceptional case. In the
examples that follow (single layered reticulated type) the linear buckling
load is much higher than the nonlinear buckling load.
8 NO. MEMBERS
6 NO. NODES
2 NO. SUPPORTS
15 NO. LOAD STEPS
5 NO. ITERATIONS
COORDINATES LOADS
X y PX PY
1 .84850000D+02 .84850000D+02 .30000000D+08 .OOOOOOOOD+OO
2 .25456000D+03 .84850000D+02 -.30000000D+08 .OOOOOOOOD+OO
3 .16970000D+03 .OOOOOOOOD+OO .OOOOOOOOD+OO .OOOOOOOOD+OO
4 .33941000D+03 .OOOOOOOOD+OO -.30000000D+08 .OOOOOOOOD+OO
5 .33941000D+03 -.10000000D+03 .OOOOOOOOD+OO .OOOOOOOOD+OO
6 .OOOOOOOOD+OO .OOOOOOOOD+OO .OOOOOOOOD+OO .OOOOOOOOD+OO

1 1 6 .20000000E+01 .OOOOOOOOE+OO
2 2 1 .20000000E+03 .OOOOOOOOE+OO
3 2 4 .20000000E+01 .OOOOOOOOE+OO
4 3 6 .20000000E+03 .OOOOOOOOE+OO
5 3 4 .20000000E+03 .OOOOOOOOE+OO
6 1 3 .20000000E+01 .OOOOOOOOE+OO
7 2 3 .20000000E+01 .OOOOOOOOE+OO
8 4 5 .20000000E+08 .OOOOOOOOE+OO
LOAD STEP 9
LOADSTEP 9; ITERATION 5; ERROR .82974001D+00
DISPLACEMENTS
X Y
1 .14560879D-07 -.75826807D-07
2 .25488995D-07 -.69831642D-07
3 .21600028D-07 -.12735170D-06
4 .37381528D-07 .42232683D-09

1 -.43532891D-07 -.21767026D-Ol -.10883513D-Ol .38932170D+03
2 .10927904D-07 .38764006D+00 .19382003D-02 -.20000150D+08
3 -.41461073D-07 -.20731433D-Ol -.10365717D-Ol -.60647490D+03
4 .21602921D-07 .76635474D+00 .38317737D-02 -.20000280D+08
5 .15779581D-07 .55974110D+00 .27987055D-02 -.19999580D+08
6 .41516244D-07 .20758601D-01 .10379300D-Ol .60644710D+03
7 .43545338D-07 .21772249D-01 .10886125D-Ol -.38933380D+03
8 -.10561904D-14 -.63371423D-02 -.31685712D-09 -.73047570D+03
UPDATED COORDINATES
X Y
1 .84569203D+02 .85130968D+02
2 .25371433D+03 .85124994D+02
3 .16913515D+03 -.38420104D-02
4 .33828030D+03 -.63813696D-02
LOAD STEP 10
LOAOSTEP 10; ITERATION 2; ERROR .381434570+09
**NEG TERM ON THE OIAGONAL AT ROW 6
Output file BUCK2D33.0UT
8 NO. MEMBERS
6 NO. NOOES
2 NO. SUPPORTS
COOROINATES LOAOS
X y PX PY
1 .848500000+02 .848500000+02 .300000000+08 .OOOOOOOOD+OO
2 .254560000+03 .848500000+02 -.300000000+08 .OOOOOOOOD+OO
3 .169700000+03 .000000000+00 .000000000+00 .OOOOOOOOD+OO
4 .339410000+03 .000000000+00 -.300000000+08 .000000000+00
5 .339410000+03 -.100000000+03 .000000000+00 .OOOOOOOOD+OO
6 .000000000+00 .000000000+00 .000000000+00 .OOOOOOOOD+OO
1 1 6 .20000000E+01
2 2 1 .20000000E+03
3 2 4 .20000000E+01
4 3 6 .20000000E+03
5 3 4 .20000000E+03
6 1 3 .20000000E+01
7 2 3 .20000000E+01
8 4 5 .20000000E+08
OISPLACEMENTS
X Y
1 -.424225000+00 .42422500D+00
2 -.127277500+01 .42427500D+00
3 -.848500000+00 -.50001473D-04
4 -.169705000+01 -.159269950-21
5 .000000000+00 .000000000+00
6 .000000000+00 .000000000+00

1 .278691460-14 .139350350-08 .696751750-09
2 -.848550000+00 -.300000000+08 -.150000000+06
3 -.283139400-14 -.141574390-08 -.707871970-09
4 -.848500000+00 -.300000000+08 -.150000000+06
5 -.848550000+00 -.300000000+08 -.150000000+06
6 -.272774420-14 -.136391740-08 -.681958680-09
7 .274698880-14 .137345900-08 .686729510-09
8 -.159269950-21 -.955619720-09 -.47780986D-16
-liP
-.68074820E-12
-.17469420E-11
-.24195970E-11
-.20062000E-08
-.75163970E-08
-.14731450E-07
-.70959640E+00
-.14292360E+01
3.6.4 Example 3.4. Buckling of a Symmetric 2-Bar Truss
The symmetric 2-bar truss of Figure 3.5 with E=10,000,000 psi and
A=0.0127 square inches was already discussed in Chapter 1 and is included
here in order to demonstrate the distinct features of the nonlinear analysis
program with respect to the manner in which length change is represented.
Nonlinear buckling analysis for an initial load of 30,000 lbs reveals
that buckling occurs somewhere between load step 43 and load step 44.
Printed below in TR2DNL34.0UT is the last converged iteration number 6
of load step 43. This implies that the buckling load is bound between 44/50x
30,000=26,400 lbs and 45/50x30,000=27,000 lbs.
Linear buckling analysis is performed using BUCK2D34.FOR on
BUCK2D34.DAT and the results are listed below in BUCK2D34.0UT. The
critical buckling load is PeR =30,000x(-1I-.l6703310E+00)=179,605 lbs.
The details of the computational model used can have some
surprising effects. Equation 1.22 models nonlinear behavior by considering
equilibrium in the deformed configuration and a strain model which defines
strain as the total length change divided by the initial length. It is followed
by Eq. 1.24 which provides the critical buckling load. For the 2-bar truss of
this example the critical load is calculated as, PeR =23,800 lbs using Eq.
1.24. Now if strain is modeled as the total length change divided by the final
length, the following relationships replace Eqs. 1.23 and 1.24 respectively:
dP = 0 ~ (HI)2 = Lo - L (3.27)
du L 2Lo-L
and
(3.28)
Equation 3.27 yields u =106.355 (making HI =93.645) which is

substituted into Eq. 3.28 to yield PeR =30,240 lbs. The result of
PeR =26,400 lbs which is obtained using iterative incremental analysis is
approximately the mean of the above two models. It comes from the fact that
this program is coded with strain modeled as the length change divided by
the actual length but this is done for each increment of load rather than the
total load. Thus,
pn+I
I
= Fn + LlFn+I
I I
where LlFn+I
I
= LAE
n+I
(Ln+I _ Ln )
I I
I
1 2
2 3
Figure 3.5 The symmetric 2-bar truss.
compares to Fj = AE(LOi - Li)/ Li in Eq. 3.28. The question of whether the

sum of the increments is always bounded by the two linear models that
consider total length changes remains open. Irrespective of the model chosen
the crucial buckling load factor ,,1,=179,605126,400=6.80 remains.
2 NO. MEMBERS
3 NO. NODES
2 NO. SUPPORTS
50 NO. LOAO STEPS
6 NO. ITERATIONS
COOROINATES LOAOS
X y PX PY
1 .200000000+03 .200000000+03 .000000000+00 -.300000000+05
2 .000000000+00 .000000000+00 .000000000+00 .OOOOOOOOD+OO
3 .40000000D+03 .OOOOOOOOD+OO .000000000+00 .000000000+00

1 1 2 .12700000E-01 .OOOOOOOOE+OO
2 1 3 .12700000E-01 .OOOOOOOOE+OO
LOAD STEP 43
LOADSTEP 43; ITERATION 6; ERROR .637136720-03
OISPLACEMENTS
x Y
1 -.77705446D-15 .11648657D-04

1 .547632030-05 .306920930-02 .241670030+00 -.280776600+05
2 .547632030-05 .306920930-02 .241670030+00 -.280776600+05
UPDATED COORDINATES
x y
1 .20000000D+03 .10653176D+03

2 NO. MEMBERS
3 NO. NODES
2 NO. SUPPORTS
COORDINATES LOADS
X y PX PY
1 .20000000D+03 .20000000D+03 .OOOOOOOOD+OO -.30000000D+05

1 1 2 .12700000E-01
2 1 3 .12700000E-01
DISPLACEMENTS
X Y
1 -.20753441D-14 -.66813240D+02
2 .OOOOOOOOD+OO .OOOOOOOOD+OO
3 .OOOOOOOOD+OO .OOOOOOOOD+OO

1 -.47244095D+02 -.21213203D+05 -.16703310D+07
2 -.47244095D+02 -.21213203D+05 -.16703310D+07
-1/~
-.74505810E-08
-.16703310E+00
3.6.5 Example 3.5. Buckling of a 4-Bar Shallow Space Truss
The nonlinear and linear buckling loads of the 4-bar space truss (E
10,000,000 psi) of Figure 3.6 are determined using TR3DNL.FOR and
BUCK3D35.FOR on TR3DNL35.DAT and BUCK3D35.DAT respectively.
Areas of cross section and loads may be read from the output files
TR3DNL35.0UT and BUCK3D35.0UT which are listed below.
This shallow truss is rather "soft" in terms of overall buckling. Once
again loss of stability is encountered at loads much lower than anticipated by
the eigenvalue formulation. In fact, it is established that the nonlinear
buckling load occurs at a load bounded by 17118x180=170 lbs and 18/18x
180= 180 lbs in the downward direction. It will be taken as 170 lbs since load
step 16 is the last converged load step. For the linear buckling load it is
calculated from the output as -1I-.19274340E+00xI80=934 lbs. The
buckling load factor is A=9341170=5.5 which is very high and warrants
added caution in design.
TR3DNL35.0UT
4 NO. MEMBERS
5 NO. NODES
4 NO. SUPPORTS
18 NO. LOAD STEPS
10 NO. ITERATIONS
COORDINATES
x y Z
1 .14400000D+03 .12000000D+03 .30000000D+Ol
5 .20400000D+03 .19200000D+03 .OOOOOOOOD+OO
LOADS
PX PY PZ

1 1 2 .19421000E+Ol .OOOOOOOOE+OO
LOAD STEP 16
ERROR = .23648477D-03
3 2
Figure 3.6 The 4-bar space truss.

OISPLACEMENTS
X y Z
1 -.104185590-08 -.752078080-09 -.578743190-07

1 -.109130580-08 -.157955030-03 -.81332076D-04 -.269221000+04
2 -.216657920-10 -.600674740-05 -.231082070-05 -.360353700+04
3 - .133271170-08 -.128476350-03 -.827758260-04 -.215142000+04
4 -.19150728D-08 -.152229910-03 - .102l67720-03 -.206535900+04
UPOATEO COOROINATES
X y Z
1 .14398836D+03 .119993350+03 .205115930+01
LOAO STEP 17
ERROR = .119372310+02
4 NO. MEMBERS
5 NO. NOOES
4 NO. SUPPORTS
COOROINATES
x y Z
1 .144000000+03 .120000000+03 .300000000+01
2 .204000000+03 .000000000+00 .000000000+00
3 .000000000+00 .000000000+00 .000000000+00
4 .000000000+00 .192000000+03 .000000000+00
5 .204000000+03 .192000000+03 .000000000+00
LOAOS
PX PY PZ
2 .000000000+00 .000000000+00 .000000000+00
3 .000000000+00 .000000000+00 .000000000+00
4 .000000000+00 .000000000+00 .000000000+00
5 .000000000+00 .000000000+00 .000000000+00

1 1 2 .19421000E+01
2 1 5 .25994000E+01
3 1 4 .15521000E+01
4 1 3 .14900000E+01
OISPLACEMENTS
X Y Z
1 -.842680280-02 -.481652640-02 -.578368260+00

1 -.134687920-01 -.194919570+04 -.100365360+04
2 -.941342990-02 -.260947100+04 -.100387440+04
3 -.l61575970-01 -.155741220+04 -.100342260+04
4 - .188112640-01 -.149510770+04 -.100342800+04
-1//3
-.55872000E-04
-.18070120E-03
-.19274340E+00
3.6.6 Example 3.6. Buckling of the 24-Bar Dome
The 24-bar dome was encountered in Chapter 2 where it was linearly

analyzed for a vertically downward load of 220.46 lbs. As is observed from
the output below it actually buckles non linearly at a load 14115x150=140
lbs. Linear analysis cannot anticipate buckling where the phenomenon is
simply unaccounted for. The linear buckling load is calculated from the
output below as -l/-0.21208881E+00x150=707 lbs. Note that the listing
from NROOT changes sign in the calculation of the 14-th eigenvalue. This
does not imply negative eigenvalues which of course are not possible!
Rather, NROOT loses accuracy in single precision arithmetic in the very
high frequency range. Finally the linear buckling load factor becomes A
=7071140=5.1, a very high factor indeed.
24 NO. MEMBERS
13 NO. NODES
6 NO. SUPPORTS
15 NO. LOAD STEPS
6 NO. ITERATIONS
COORDINATES
X Y Z
1 .OOOOOOOOD+OO .OOOOOOOOD+OO .32346000D+Ol
2 .49212500D+Ol .85239000D+Ol .24472000D+Ol
3 -.49212500D+Ol .85239000D+Ol .24472000D+Ol
4 -.98425000D+Ol .OOOOOOOOD+OO .24472000D+Ol
5 -.49212500D+Ol -.85239000D+Ol .24472000D+Ol
6 .49212500D+Ol -.85239000D+Ol .24472000D+Ol
7 .98425000D+Ol .OOOOOOOOD+OO .24472000D+Ol
8 .OOOOOOOOD+OO . 19685000D+02 .OOOOOOOOD+OO
9 -.17047200D+02 .98425000D+Ol .OOOOOOOOD+OO
10 -.17047200D+02 -.98425000D+Ol .OOOOOOOOD+OO
12 .17047200D+02 -.98425000D+Ol .OOOOOOOOD+OO
13 .17047200D+02 .98425000D+Ol .OOOOOOOOD+OO
LOADS
PX PY PZ
11 .000000000+00 .000000000+00 .000000000+00
12 .000000000+00 .000000000+00 .000000000+00
13 .000000000+00 .000000000+00 .000000000+00

1 1 3 .15500000E-Ol .OOOOOOOOE+OO
LOAO STEP 13
ERROR = .405126820-04
OISPLACEMENTS
X Y Z
1 -.750342620-18 -.184837980-18 .366100390-08
2 -.151017190-09 -.268435170-09 -.928473110-09
3 .151017180-09 -.268435170-09 -.928473100-09
4 .288973760-09 .224570350-18 -.827290470-09
5 .151017180-09 .268435170-09 -.928473100-09
6 -.151017180-09 .268435170-09 -.928473100-09
7 -.288973760-09 -.539079120-18 -.827290480-09

1 -.554293560-10 -.261274540-05 -.168564230-03 -.424807800+03
2 -.302034370-09 -.142584220-04 -.919898220-03 .355932600+03
3 -.279358580-11 -.104427750-06 -.673727410-05 -.589620800+02
4 -.279358590-11 -.104427750-06 -.673727440-05 -.589620800+02
5 -.420080620-10 -.198012050-05 -.127749710-03 -.424802300+03
6 -.301450010-09 -.142307790-04 -.918114800-03 .355933700+03
7 .350236730-11 .130925580-06 .844681160-05 -.589612000+02
8 -.823250550-11 -.307752820-06 -.198550210-04 -.589617100+02
9 -.55429356D-I0 -.261274550-05 -.168564230-03 -.424807800+03
10 -.301450010-09 -.142307790-04 -.918114800-03 .355933700+03
11 -.823250590-11 -.307752840-06 -.198550220-04 -.589617100+02
12 .350236770-11 .130925590-06 .844681240-05 -.589612000+02
13 -.554293550-10 -.261274540-05 -.168564220-03 -.424807800+03
14 -.302034370-09 - .l42584220-04 -.919898220-03 .355932600+03
15 -.279358620-11 -.104427760-06 -.673727510-05 -.589620800+02
16 -.279358570-11 -.104427750-06 -.673727400-05 -.589620800+02
17 -.420080610-10 -.198012050-05 -.127749710-03 -.424802300+03
18 -.30145001D-09 -.14230779D-04 -.91811480D-03 .35593370D+03

19 .35023672D-ll .13092557D-06 .84468112D-05 -.58961200D+02
20 -.82325051D-ll -.30775281D-06 -.19855020D-04 -.58961710D+02
21 -.55429356D-10 -.26127455D-05 -.16856423D-03 -.42480780D+03
22 -.30145001D-09 -.14230779D-04 -.91811480D-03 .35593370D+03
23 -.82325059D-ll -.30775284D-06 -.19855022D-04 -.58961710D+02
24 .35023680D-ll . 13092560D-06 .84468133D-05 -.58961200D+02
UPDATED COORDINATES
X y Z
1 -.63623184D-17 .18422895D-17 .30031710D+01
2 .49250183D+01 .85304268D+01 .24613183D+01
3 -.49250183D+01 .85304268D+01 .24613183D+01
4 -.98500367D+01 .30818523D-17 .24613180D+01
5 -.49250183D+01 -.85304268D+01 .24613183D+01
6 .49250183D+01 -.85304268D+01 .24613183D+01
7 .98500367D+01 -.18711817D-17 .24613180D+01
-liP
0.51235702E-01
0.26026258E-01
0.26026156E-01
0.90162409E-03
0.56519511E-03
0.56519359E-03
o . 17111958E-03
o . 17111933E-03
-0.54412139E-04
-0.14497516E-03
-0.14497591E-03
-0.18869033E-03
-0.24152090E-03
-0.26560493E-03
-0.26560537E-03
-0.11734122E-02
-0.11734152E-02
-0.43631340E-02
-0.13986786E-01
-0.13986799E-01
-0.21208881E+00
stop - Program terminated.
3.7 Problems
1. Find the linear buckling load of the 3-bar truss of Example 2.1 using
BUCK2D.FOR given that only one load acts vertically downwards, the areas
of cross section of members 1 and 2 are 0.1 square inches and that member 3
is rigid. (For computational purposes an area of 1,000,000 square inches has
been used.)
2. Calculate the buckling load of problem 1 using principles of mechanics.

3. Using TR2DNL.FOR find the nonlinear buckling load of the 2-bar truss of
Figure 1.14 given that E=lO,OOO,OOO psi, H=10 inches, B} =200 inches,
B 2 =50 inches, and Al = A2 = 0.0127 in 2 . (Ans: P=60 Ibs).
4. Write a few lines of code that will verify your results in problem 3.
(Increment the displacement and obtain the load via equilibrium in the
deformed configuration. Buckling occurs when the slope of the P-o curve
changes sign.)
5. a) Derive Eq. 3.20.

L2 _L2
b) Use E = 2 0 and re derive Eqs. 3.20 and 3.21.
2Lo
c) Use these expressions to obtain a buckling load for Example 3.4.
6. Figure 3.7 shows a 12-member cable net which consists of a two-

directional network of cables attached to a rigid frame. The cross sectional
area of the cable is 0.0003038 square inches, L=96.85 inches andf=22.24
inches. The horizontal component of initial prestress is 26.455 lbs and
E=30,000,000 psi. If one of the free nodes is loaded downwards with 18.298
lbs find the forces in the members and the displacements of the nodes using
TR3DNL.FOR. (For a discussion on the type of analysis required for cable
networks (linear vs. nonlinear) the reader is referred to Levy et ai., 1991.)
-~
~ L ..
ti---(:P
Figure 3.7 The 12-member cable network.
7. Modify programs P5 and P6 for more general usage when computing

eigenvalue buckling loads. (Hint: If NROOT is placed within a subroutine,
arrays with variable DIMENSION statements can be used avoiding re-coding
when problems of different size are to be run.)
CHAPTER 4
NONLINEAR ANALYSIS OF PLANE FRAMES
4.1 Linear Analysis
The discussion of plane frames adds little to the earlier discussion of trusses.
Plane frames are, of course, more complex and include trusses as a special
case as the moment of inertia of frame members is set to zero.
node i
,,
',deformed
undeformed
Figure 4.1 Node i.
Figure 4.1 shows the typical variables associated with the joint of a
plane frame. In terms of kinematics each joint is described by a two-
dimensional displacement vector OJ and a scalar rotation OJ' The conjugate
variables are Pi and M i , the joint vector and the scalar moment. These
node quantities give rise to the joint load matrix P and the joint displacement
matrix 0
(Pit (° 1 )x
(pJ)y (oJ) y
P= (p2t
MJ
0= (02t
°1
(4.1)
(p2 )y (02)y
M2 O2
local coordinate
system
global coordinate
system
Figure 4.2 Typical member.
Clearly, the size of the system matrix is three times the number of movable
joints.
Figure 4.2 describes a typical member. At each end there is a shear
Vi' a moment m; and thrust t i . Obviously, mt "* mi . The member force
matrix, Fi , which contained only bar forces in the case of the truss, now
contains three terms per member. It can be shown that the number of these
terms is invariant but that the specific choice of terms is somewhat arbitrary.
Here the three force components that are chosen per member are the two end
moments and the thrust of the "+ end". The other three quantities i.e. the two
end shear forces and the thrust in the "- end" are related to the first three
quantities through the equations of equilibrium. The member force matrix,
thus, becomes
11
mt
ml
F= 12 (4.2)
m~
m2
Let the member forces at the positive and negative ends be represented by
f: and f i- as
NONLINEAR ANALYSIS OF PLANE FRAMES 73
(4.3)
Member equilibrium relates ft and fj- to F j as
(4.4)
where
° °
_ I.-I
_L-1
I
I
°1
~ ; Nj =
r- ~
1 I.-I
I
I.-I
I
(4.5)
Since the equilibrium equations at each node must be written in the global
coordinate system, these end of member forces must be transformed using
the usual rotation matrix R j which is described in Appendix 3 as
~l (4.6)
It is the transpose of the rotation matrix that is directly applied to the

member forces because member end forces are considered "local". The
transformation here goes from a "local" to a "global" system whereas the
rotation matrix, R j , (Appendix 3) takes vectors from the "global" to a
"local" system. This inverse rotation is performed using the transpose of R j
T T T T
to yield R j N7 F j and R j Nj F j as the end of member forces.
Consistent with the notation of the node method that was introduced
in Chapter 2 it is now possible to define the generalized incidence matrix N
(in NTF=P)as

otherwise
Proceeding to the constitutive relations, it is obvious that the

member displacements b.. i entering the equations must be the member
elongation and the two member end rotations. The node method relates
member displacements to nodal displacements through No = A. With /).
described as
LiLl
a(
al
A= LiL2 (4.8)
a;
a2
and Nand 8 already defined, No = A reduces for a single member to
(4.9)
If RiO A and Rio C are regarded as the nodal displacements of member i in

the local coordinate system indicated by the subscript Xi, then the above
expression, when expanded, becomes
(4.10)
with
LiLi = member length change of member i

at = rotation of the "+ end" minus the rigid body rotation of the member
at = rotation of the "- end" minus the rigid body rotation of the member
Finally the member "primitive" or "local" stiffness matrix is
presented in its usual form as
EAi
0 0
Li
4Eli 2Eli
Ki = 0 (4.11 )
Lj Li
2El j 4Eli
0
Li Li
where Ai,E,L j are the member area, Young's modulus, and the member
length respectively.
In terms of global assemblage for use in the node method
(N T KNo = P), the contribution of each member to the global stiffness
matrix becomes
... }
(4.12)
or
colA colC
RiTN7TKiNiRi] row A
(4.13)
RiTNiTKiNiRi rowC
4.2 Computer Program P7-FR2D.FOR
This program performs linear analysis of plane frames. The coding follows
the development of the node method for plane trusses. Appendix 5 contains a
full listing of this program.
The program begins with some initialization of variables and some
input. The system matrix K E is then generated as C by summing the
contribution of each member to it. Joint displacements are computed and

finally member forces are determined.
The subroutines in the program are RNK, INSERT and PROD.
Subroutine RNK defines the matrices R, N+ , N- and K and then forms the
products N+R and N-R as NRP and NRM. Subroutine INSERT is used to
place elements in the system matrix C. Subroutine PRO D forms the product
of three matrices D = AT BC .
The variables in the program are the rotation matrix R=R, the
member stiffness matrix SK = K, SNP = N+ , SNM = N-, the number of
nodes NN, the number of bars NB, the joint load matrix P, the system matrix
C, the member area A, the member length AL, the member moment of inertia
SI, the member orientation e, as TH, and the member end nodes NP and MI.
Here again as in the truss programs all supports are regarded to be
fully fixed and only nodal loads are considered.
4.2.1 Example 4.1. A Simple Plane Frame
This example is presented to familiarize the reader with some of the details
of FR2D.FOR in terms of both input and output. Figure 4.3 shows a plane
frame which is loaded with a horizontal force P = 10,000 lbs, a downward
vertical force of 5,000 lbs and a moment M = 12,000 in-Ibs. The span is 24
feet (L=12 ft.) and H=8 ft. One support is fully fixed whereas the other is a
roller. The roller is handled by modeling it as a fictitious fixed bar of zero
moment of inertia and an infinite area of cross section.
Program FR2D.FOR is invoked with the input file FR2D41.DAT
and the results are listed in the output file FR2D41.0UT. The moment
diagram of Figure 4.4 is drawn from the output columns "~,, and "M-". The
moments are drawn on the tension side of the beam.
to.SP
2
~H I=O;A =00
2
L L
14 .... ••
Figure 4.3 A three member plane frame.
The first line of the input file supplies NB, NN and NS. The next
four lines provide nodal load information and the last five lines provide
member information in terms of cross sectional area, length, moment of
inertia, angle of orientation and "+ end" node number and "- end" node
number respectively. Note that the last line in the member information which
relates to the fictitious bar of length of one inch, zero moment of inertia and
infinite area (actually 10,000,000 square inches was used).
Input file FR2D41.DAT
5 6 2 (input format: 315)

10000. (input format: 3010.2)
-5000. (input format: 3DIO.2)
-12000. (input format: 3DIO.2)
(input format: 3D 10.2)
100. 144. 800. 90. 1 5 (input format: 4EIO.2,215)
100. 144. 800. O. 2 1 (input format: 4E10.2,215)
100. 144. 800. O. 3 2 (input format: 4EIO.2,215)
100. 96. 800. 90. 3 4 (input format: 4EIO.2,215)
10000000. l. 0.0000000 90. 4 6 (input format: 4E10.2,215)
Output file FR2D41.0UT
5 NO. MEMBERS
6 NO. NOOES
2 NO. SUPPORTS
JOINT LOAOS
PX PY M
1 .100000000+05 .000000000+00 .000000000+00
2 .000000000+00 -.500000000+04 .000000000+00
3 .000000000+00 .000000000+00 -.120000000+05
4 .000000000+00 .000000000+00 .000000000+00
MEMBER PROPERTIES
AREA LENGTH I ANGLE+ENO-ENO
1 .10000000E+03 .14400000E+03 .80000000E+03 .90000000E+02 1 5
2 .10000000E+03 .14400000E+03 .80000000E+03 .OOOOOOOOE+OO 2 1
3 .10000000E+03 .14400000E+03 .80000000E+03 .OOOOOOOOE+OO 3 2
4 .10000000E+03 . 96000000E+02 .80000000E+03 .90000000E+02 3 4
5 .10000000E+08 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 4 6
3.2638 x 10 5
11.136 x 10 5
Figure 4.4 Moment diagram.

JOINT DISPLACEMENTS
DX DY TH
1 .28316217D+OO -.66131430D-04 -.24431519D-02
2 .28316217D+OO -.17759724D+OO .17483097D-03
3 .28316217D+OO -.12198879D-03 .17426643D-02
4 .45045795D+OO -.53409112D-06 .17426643D-02
MEMBER DISPLACEMENTS-FORCES
DL AL+ AL- T M+ M-
1 -.65796D-04-.47675D-03 .19664D-02-.13251D+04 .32638D+06 . 11136D+07
2 .21635D-09 .14077D-02 -.12103D-02 .43571D-02 .51719D+06 -.32638D+06
3 .21635D-09 .51020D-03 -.10576D-02 .43571D-02 -.12000D+05 -.51719D+06
4 -.12165D-03-.16079D-17 .86736D-18-.36749D+04 - .11350D-08 .61321D-10
5 -.12672D-l0 .45220D+OO .45046D+OO-.36749D+04 .OOOOOD+OO .OOOOOD+OO
The geometric stiffness matrix for plane frames will be derived in an

identical manner to that used for trusses and similar arguments will be made
to those of Section 3.2. First the contribution of member i to the equilibrium
equations is formulated and then the gradient of this contribution is taken in
order to generate the geometric stiffness matrix. Following the node method
for plane frames member i will contribute the following terms to the node
equilibrium equations
T +T
Ri N j F j joint A
(NTF)i = (4.14)
T T
R j Nj Fj joint C
which, after multiplication of the matrices yields
tj cos (fJj - Vi sin (fJj Ail

tj sin (fJi + Vj cos (fJj Ai2 jointA
m+ m+
I I
(NTF)j = = (4.15)
- Ij cos (fJj + Vj sin (fJj -Ail
- tj sin (fJj - Vj cos (fJj -Ai2 jointC
mj mj
The geometric stiffness matrix can be computed directly as
colA colC
row A
(4.16)
rowC
where the sub matrices are produced in the same manner as Eq. 3.10 to
become
BAil BAil BAil BAil

0 0
axA ayA BXe aye
v(NiF) = BAi2 BAi2
0 ;v(NiFje = BAi2 BAi2
0
ax A ayA BXe aye
0 0 0 0 0 0
and
BAil BAil BAil BAil

0 0
ax A ayA BXe aye
( TfA =
VNiF BAi2 BAi2
0 ;v{NiF)ee = BAi2 BAi2
0
BXA ayA BXe aye
0 0 0 0 0 0
The derivatives are now computed to produce the terms of the geometric
stiffness matrix. Symbolically
colA colC
row A
(4.17)
rowC
Figure 4.5 Member projections.
where
(4.18)
and
(4.19)
(4.20)
(4.21 )
These terms can be derived with the help of the following (Fig 4.5):
x x
cos() = - = =n
L ~x2 + y2 x
o ()
-cos = -1 (1- n 2)
ox L x
o
-cos()=-xx +y
(2 2 )-3/2 y = -nxny
--
By L
o cos() =__
___ 1- 2n2 ::...x
ox L L2
a
-sinB=--Y-
I-n
2
Oy L
8 sine
8y L
4.4 Computer Program P8-FR2DNL.FOR
This program implements the material described above and is quite similar
to the truss problem with one exception. The truss problem uses linear
springs for members and does not include the possibility of member
buckling. (This could of course have been built in but it would be necessary
in this case to introduce the member moment of inertia. There is also the
problem for cable nets of members having no stiffness in compression.) As
was pointed out in Chapter 1, beams (beam-columns) have stress stiffening
and softening in the presence of axial load. For that reason the member
stiffness matrix for plane frames is based upon beam-column theory. The
careful reader will see that complex exponentials are used in computing the
member stiffness which allow a single form of solution to include both
compression and tension.
New subroutines encountered in this program include CORLO
which computes the unbalanced load, GS TIFF which computes the
geometric stiffness matrix, RESET which computes the new member
deformations and BMCOL which is used to construct the member stiffness
matrix including the effect of axial load. Subroutine RNK has been expanded
82 ANALYSIS OF GEOMETRIC ALL Y NONLINEAR STRUCTURES
to accommodate and change the member stiffness matrix by calling

BMCOL.
New variables include AI which is used for the triple product that
comes out of subroutine PROD3. AJ is the transpose of AI and the matrix
AK is the member geometric stiffness matrix. AI is later destroyed by
inserting member deformations in its first row and member forces in its
second row.
4.4.1 Examples
Three examples are presented in this section. The first two involve frame
buckling whereas the third involves an analysis for large deflections (small
strain) on a circular cantilever beam.
4.4.1.1 Example 4.2. Buckling of a Portal Frame
This is a simple buckling example taken from Timoshenko (1936). Again, no

attempt is made to obtain an "exact" value of the buckling load. (See earlier
comments.) Timoshenko gives the buckling load in this (E = 29xl 0 6 psi
I = 800 in 4 L = 144 in ) case to be
( \
Ij; l:1
L
.'" III
@
fictitious fictitious
members members
Figure 4.6 Plane frame buckling.

Figure 4.6 shows Timoshenko's buckling problem which is described

in detail by the input file FR2DNL42.DAT which in tum is read by
FR2DNL.FOR. Only partial results from the lengthy output file are listed
below as the file FR2DNL42.0UT. The full load step is in this case
specified to be 2,200,000 lbs. Note that fictitious bars with 1=0 and 1.0 inch
length have been added to model the pinned supports. Also large areas have
been used in some members to simulate Timoshenko's neglect of member
length change. By scanning the output file it is found that a negative term
first appears on the diagonal of the system matrix during the 9th load step.
The buckling load lies, therefore, between the 8th and 9th load steps. The 8th
load step is of ~ x 2,200,000 = 1,980,000 lbs. and the 9th load step is of
10
.!.Q x 2,200,000 = 2,200,000
lbs. This implies that the buckling load occurs
10
between 1,980,000 lbs and 2,200,000 lbs which agrees with Timoshenko's
result.
Input file FR2DNL42.DAT
7 8 4 10 4
O. -2200000.
-2200000.
O.
O.
100. 144. 800. 90. 1 3
100. 144. 800. 90. 2 4
1. 144. 800. O. 2 1
100. 1. O. 90. 3 6
1. 1. O. O. 3 5
1. 1. O. O. 4 7
100. 1. O. 90. 4 8
O.
O.
O.
O.
O.
O.
O.
Output file FR2DNL42.0UT
7 NO. MEMBERS
8 NO. NODES
4 NO. SUPPORTS
10 NO. LOAD STEPS
4 NO. ITERATIONS/STEP
JOINT LOADS
PX PY M
MEMBER PROPERTIES
AREA LENGTH I ANGLE +END -END
1 .10000000E+03 .14400000E+03 .80000000E+03 .90000000E+02 1 3

2 .10000000E+03 .14400000E+03 .80000000E+03 .90000000E+02 2 4
3 .10000000E+01 .14400000E+03 .80000000E+03 .OOOOOOOOE+OO 2 1
4 .10000000E+03 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 3 6
5 .10000000E+01 .10000000E+01 .OOOOOOOOE+OO .OOOOOOOOE+OO 3 5
7 .10000000E+03 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 4 8
CURRENT MEMBER FORCES

1 .OOOOOOOOE+OO .OOOOOOOOE+OO .OOOOOOOOE+OO
LOAD STEP 8
JOINT DISPLACEMENTS
DX DY TH
1 -.64286104D-04 -.85467514D-07 .13680851D-06
2 -.64287406D-04 .96456597D-07 .13680924D-06
3 -.19722560D-10 -.60868362D-09 .61633452D-06
4 .14104490D-10 .64596191D-09 .61635144D-06
DL AL+ AL- T M+ M-
1-.87626D-07 -.30962D-06 .16990D-06 -.17647D+01 -.13103D+03 -.19380D-06
2 .93044D-07 -.30963D-06 .16991D-06 .18738D+01 -.13103D+03 .29267D-06
3-.13017D-08 .13555D-06 .13555D-06 -.26216D-03 .13103D+03 .13103D+03
4-.60868D-09 .61631D-06 -.19722D-IO -.17652D+Ol .OOOOOD+OO .OOOOOD+OO
5-.19307D-10 .61694D-06 .60870D-09 -.55990D-03 .OOOOOD+OO .OOOOOD+OO
6 .13664D-10 .61571D-06 -.64597D-09 .39624D-03 .OOOOOD+OO .OOOOOD+OO
7 .64596D-09 .61637D-06 .14104D-10 .18733D+01 .OOOOOD+OO .OOOOOD+OO
MEMBER PROPERTIES
1 .10000000E+03 .14400000E+03 .80000000E+03 .89997610E+02 1 3
2 .10000000E+03 .14400000E+03 .80000000E+03 .89997610E+02 2 4
3 .10000000E+01 .14400000E+03 .80000000E+03 -.67403680E-05 2 1
4 .10000000E+03 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 3 6
5 .10000000E+01 .10000000E+01 .OOOOOOOOE+OO -.39115870E-01 3 5
6 .10000000E+01 .10000000E+01 .OOOOOOOOE+OO -.39122570E-01 4 7
7 .10000000E+03 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 4 8

1 -.19798300E+07 .12208010E+05 -.25477250E-06
2 -.19801700E+07 .12208020E+05 .27050080E-06
3 .71968580E-02 -.12208020E+05 -.12208010E+05
4 -.19798300E+07 .OOOOOOOOE+OO .OOOOOOOOE+OO
5 .23545330E+01 .OOOOOOOOE+OO .OOOOOOOOE+OO
*** ERROR .18657140E+00
LOAO STEP 9
4.4.1.2 Example 4.3. Buckling of a Plane Frame
This example which is described in Figure 4.7 is taken from the test
problems that appear in the ANSYS (1987) manual. Ten load steps with 4
iterations per load step are used to solve the problem using FR2DNL.FOR
and the input file FR2DNL43.DAT with a starting load of 30,000 lbs. Some
of the output is listed in FR2DNL43.0UT. It includes the echo of the data
file and the results of the last iteration of the last converged load step (load
step 8, iteration 4). The buckling load is bounded between 9/10 x 30,000 =
27,000 lbs and 10/10 x 30,000 = 30,000 lbs. This result agrees with the
ANSYS result of27,789Ibs. (ANSYS, 1987, example no. 12.)
Output file FR2DNU3.0UT

6 NO. MEMBERS
7 NO. NOOES
4 NO. SUPPORTS
10 NO. LOAO STEPS
JOINT LOAOS
PX py M
1 .000000000+00 .000000000+00 .000000000+00
2 .000000000+00 .000000000+00 .000000000+00
3 -.300000000+05 .000000000+00 .000000000+00
MEMBER PROPERTIES
AREA LENGTH I ANGLE +ENO -ENO
1 .38730000E+01 .14400000E+03 . 12500000E+01 .OOOOOOOOE+OO 2 1
2 .33540000E+01 .96000000E+02 .93750000E+00 .90000000E+02 2 6
144 120
E, A = const.
96 I
7111111111
fictitious
member
fictitious
member
Figure 4.7 Plane frame of example 4.3.

3 .42430000E+01 .12000000E+03 .15000000E+01 .OOOOOOOOE+OO 3 2

5 .10000000E+05 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 1 5
6 .10000000E+05 .10000000E+03 .OOOOOOOOE+OO .90000000E+02 3 7

LOAD STEP 8
JOINT DISPLACEMENTS
DX DY TH
1 -.23518923D-14 -.10649286D-16 -.15353358D-09
2 -.87463440D-09 -.10712609D-ll .75974923D-10
3 -.10371566D-08 -.60452218D-12 -.67537903D-10
DL AL+ AL- T M+ M-
1-.87462D-09 .75982D-10 -.15353D-09 -.70571D-03 -.87157D-04 .66174D-22
2-.76739D-12 .66864D-10 -.91108D-ll -.80432D-06 .73017D-04 .28502D-04
3-.16252D-09-.67542D-10 .75971D-10 -.17239D-03 -.16272D-19 .16048D-04
4-.23315D-14-.15353D-09 .10649D-16 -.69944D-03 .OOOOOD+OO .OOOOOD+OO
5 .00000D+00-.15354D-09 -.23519D-14 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
6 .00000D+00-.77909D-10 - .10372D-10 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
MEMBER PROPERTIES
1 .38730000E+01 .14400000E+03 .12500000E+01 -.29413460E-05 2 1
2 .33540000E+01 .96000000E+02 .93750000E+00 .90019980E+02 2 6
3 .42430000E+01 .12000000E+03 .15000000E+01 -.47256770E-05 3 2
4 .10000000E+05 .10000000E+01 .OOOOOOOOE+OO -.19312930E-10 1 4
5 .10000000E+05 .10000000E+01 .OOOOOOOOE+OO .90000000E+02 1 5
6 .10000000E+05 .10000000E+03 .OOOOOOOOE+OO .90033780E+02 3 7

1 -.26993110E+05 .19375060E+02 .49922000E-19
2 -.16752790E+01 -.23694280E+03 -.42474940E+03
3 -.27000000E+05 -.22042710E-19 .21756780E+03
*** ERROR .56081750E-03
LOAD STEP 9
**NEG TERM ON THE DIAGONAL AT ROW

4.4.1.3 Example 4.4. Large Rotations of a Circular Cantilever Beam
This example is intriguing because an initially circular cantilever beam is

taken for an excursion of large rotations going through a straightened
geometry and finally curving in the opposite direction. The beam is an
aluminum (E = 10,000,000 psi) 1.0 inch square bar with a 45° circular bend
and a radius of 100 in. It is subjected to incremental loading of up to 600 lbs
in ten load steps. It is analyzed here, of course, as a plane beam with no
thought given to lateral buckling effects which are discussed in Chapter 5.
The plane frame nonlinear analysis program uses the input file
FR2DNL44.DAT with results of the input echoed at 300 lbs and 600 lbs
loads listed below in the file FR2DNL44.0UT. For the 300 lbs load
convergence is achieved at load step 4 and iteration 5 whereas at 600 lbs
convergence is achieved at load step 10 and iteration 5. Figure 4.8 shows the
beam unloaded, at its 300 lbs load position, and at the 600 lbs load position.
The coordinates for these positions are calculated using its output data to be
(64.40,44.54) and (49.76,59.3) respectively.

8 NO. MEMBERS
9 NO. NOOES
1 NO. SUPPORTS
10 NO. LOAD STEPS
JOINT LOADS
PX PY M
1 .000000000+00 .000000000+00 .000000000+00

2 .000000000+00 .000000000+00 .000000000+00
3 .OOOOOOOOD+OO .000000000+00 .000000000+00
4 .000000000+00 .000000000+00 .000000000+00
5 .000000000+00 .OOOOOOOOD+OO .000000000+00
6 .000000000+00 .000000000+00 .000000000+00
7 .000000000+00 .OOOOOOOOD+OO .000000000+00
8 .000000000+00 .60000000D+03 .OOOOOOOOD+OO
MEMBER PROPERTIES
1 .10000000E+Ol .98140000E+Ol .83300000E-Ol .19690000E+02 1 9

2 .10000000E+Ol .98140000E+Ol .83300000E-Ol .14060000E+02 2 1
3 .10000000E+Ol .98140000E+Ol .83300000E-Ol .84400000E+Ol 3 2
4 .10000000E+Ol .98140000E+Ol .83300000E-Ol .28100000E+Ol 4 3
5 .10000000E+Ol .98140000E+Ol .83300000E-Ol -.28100000E+Ol 5 4
6 .10000000E+Ol .98140000E+Ol .83300000E-Ol -.84400000E+Ol 6 5
7 .10000000E+Ol .98140000E+Ol .83300000E-Ol -.14100000E+02 7 6
8 .10000000E+Ol .98140000E+Ol .83300000E-Ol -.19700000E+02 8 7

59.29
44.54
Figure 4.8 The Circular Cantilever Beam

LOAO STEP 4
JOINT OISPLACEMENTS
OX OY TH
1 -.284412400-06 .585327420-06 .128889020-06
2 -.114860640-05 .199090440-05 .187727230-06
3 -.201533630-05 .319992020-05 .792095490-07
4 -.165715710-05 .274024120-05 -.233869310-06
5 .114706880-05 -.721137090-06 -.689484870-06
6 .678643000-05 -.782773210-05 -.114895700-05
7 .144664920-04 -.183158210-04 -.147018090-05
8 .225665320-04 -.310939120-04 - .158014110-05
MEMBER OISPLACEMENTS-FORCES
OL AL+ AL- T M+ M-
1-.144740-10 .625790-07 -.663100-07 -.147480-04 .100030-01 - .119040-01

2-.266570-09 .196010-07 -.392370-07 -.271630-03 -.620860-07 -.100030-01
3-.810910-09-.723690-07 .361480-07 -.826280-03 -.184530-01 .620780-07
4-.156120-08-.174490-06 .138590-06 -.159080-02 -.357660-01 .174760-01
5-.243840-08-.235570-06 .220050-06 -.248460-02 -.426950-01 .347900-01
6-.328510-08-.224540-06 .234940-06 -.334730-02 -.364200-01 .417190-01
7-.396540-08-.145610-06 .175620-06 -.404050-02 -.196680-01 .349550-01
8-.436990-08-.385560-07 .714040-07 -.445280-02 -.980660-03 .177150-01
MEMBER PROPERTIES
1 .10000000E+01 .98140000E+01 .83300000E-01 .25914010E+02 1 9
2 .10000000E+01 .98140000E+01 .83300000E-01 .31575310E+02 2 1
3 .10000000E+01 .98140000E+01 .83300000E-01 .35605110E+02 3 2
4 .10000000E+01 .98140000E+01 .83300000E-01 .38079040E+02 4 3
5 .10000000E+01 .98140000E+01 .83300000E-01 .39043940E+02 5 4
6 .10000000E+01 .98140000E+01 .83300000E-01 .38454120E+02 6 5
7 .10000000E+01 .98140000E+01 .83300000E-01 .36231480E+02 7 6
8 .10000000E+01 .98140000E+01 .83300000E-01 . 32387170E+02 8 7

1 . 13110640E+03 .16673050E+05 -.19321220E+05
2 .15708560E+03 .14164740E+05 -.16673050E+05
3 .17465850E+03 . 11770960E+05 -.14164740E+05
4 .18502430E+03 .94533960E+04 -.11770960E+05
5 .18897470E+03 .71667450E+04 -.94533960E+04
6 .18656620E+03 .48611220E+04 -.71667450E+04
7 .17731460E+03 .24862240E+04 -.48611220E+04
8 .16069120E+03 .34999940E-09 -.24862240E+04
*** ERROR = .15305700E-03
LOAO STEP 10
JOINT OISPLACEMENTS
ox OY TH
1 -.752690630-08 .134621350-07 .440110950-08
2 -.422101380-07 .543115440-07 .348560490-08
3 -.375271550-07 .501537580-07 -.388893690-08
4 .110750020-07 .149458840-07 -.818526630-08
5 .822735360-07 -.306757560-07 -.755555220-08
6 .135020840-06 -.630180060-07 -.635422190-08
7 .196942030-06 -.102545730-06 -.840397210-08
8 .267602070-06 -.153486580-06 -.911119530-08
MEMBER OISPLACEMENTS-FORCES
OL AL+ AL- T M+ M-
1-.720850-11 .282950-08 -.157160-08 -.734510-05 .695110-03 -.541120-04
2-.116420-10-.197470-08 -.105920-08 - .118630-04 -.851110-03 -.695110-03
3-.939340-11-.325080-08 .412370-08 -.957140-05 -.406190-03 .851110-03
4 .141290-10-.207010-08 .222630-08 .143970-04 -.326560-03 .406190-03
5-.129250-10 .106080-08 .431100-09 -.131700-04 .433970-03 .326560-03
6 .347990-11-.496320-10 -.125100-08 .354580-05 -.229040-03 -.433970-03
7 .134950-10-.918540-09 .113120-08 .137510-04 -.120620-03 .229040-03
8-.127580-10-.235300-09 .471920-09 -.130000-04 -.750570-18 .120620-03
MEMBER PROPERTIES
1 .10000000E+01 .98140000E+01 .83300000E-01 .29183540E+02 1 9
2 .10000000E+01 .98140000E+01 .83300000E-01 .40320540E+02 2 1
3 .10000000E+01 .98140000E+01 .83300000E-01 . 48485710E+02 3 2
4 .10000000E+01 .98140000E+01 .83300000E-01 .54066620E+02 4 3
5 .10000000E+01 .98140000E+01 .83300000E-01 .57358410E+02 5 4
6 .10000000E+01 .98140000E+01 .83300000E-01 .58482090E+02 6 5
7 .10000000E+01 .98140000E+01 .83300000E-01 .57437180E+02 7 6
8 .10000000E+01 .98140000E+01 .83300000E-01 .54219520E+02 8 7

1 .29256510E+03 .24714460E+05 -.29855400E+05
2 .38823760E+03 .20224930E+05 -.24714460E+05
3 .44927400E+03 .16322060E+05 -.20224930E+05
4 .48581960E+03 .12866480E+05 -.16322060E+05
5 .50523640E+03 .96903770E+04 -.12866480E+05
6 .51148580E+03 .66121240E+04 -.96903770E+04
7 .50568080E+03 .34428410E+04 -.66121240E+04
8 .48675760E+03 .58803160E-14 -.34428410E+04
*** ERROR = .35856650E-03
stop - Program terminated.
4.5 Problems
l.a. Show that the member stiffness matrix for a member with a hinge at the
"+ end" is
(Af); 0 o
(K E ); = o o
o
b. What is the stiffness matrix of a member with a hinge at its "-end".

c. Show that the two matrices can be used to perform an incremental plastic
collapse analysis for a structure in which the formulation of plastic hinges
can lead to collapse.
2. It is common to assume that the beams in a structural frame are rigid in

order to simplify buckling calculations. Construct a real frame example and
perform parametric studies to determine the extent to which this is true.
3. The AISC manual provides charts which can be used to determine the
effective length of frame members appropriately. Construct examples and
run parametric studies which will verify the accuracy of these charts.
4. Discuss the "complex" form of the solution to the beam column equations.
Verify the code which is included in subroutine BMCOL of program
FR2DNL.FOR. (Hint: Refer to section 1.5.3 and Appendix 1 .)
5. The geometric stiffness matrix for the plane frame is symmetric (as is the
case of the truss). Find the potential from which the geometric stiffness
matrix can be computed as a derivative. (Hint: For the truss let the potential
fjJ = L F; L i ' When the F;' s are held constant V¢ = NT F and
members
KG =VV¢.)
6. Compare the examples of this chapter with the results of some commercial
program such as ANSYS. Discuss the effect of subdividing a member into
many finite elements. How many element should be used? What is the effect
of using a single finite element to model each member?
7. If a beam such as the one used in Example 4.4 is made with sufficient
curvature it is possible to have such a beam buckle before reverse curvature
develops. Construct such a case.
CHAPTER 5
NONLINEAR ANALYSIS OF SPACE FRAMES
5.1 Introduction
This chapter completes the sequence which began with trusses some pages
back. As you move up in terms of complexity from trusses, space frames add
new issues which revolve about the rotation matrix (see Appendix 3). Both
trusses and plane frames can be described geometrically in terms of node
coordinates. That is, certainly for the case of straight members, the geometry
of trusses and plane frames can be described by giving the coordinates of the
ends of the members. That is not the case for space frames where members
can still be rotated about their axes after their ends have been located. From
another side, joint rotations in plane frames can still be treated as scalars
while joint rotations in space frames require the concept of a rotation matrix.
Figure 5.1 Typical joint i.
5.2 Linear Analysis
A typical joint i of a linear elastic space frame (Fig. 5.1) has four physical
vectors associated with it,
displacement vector
(small) rotation vector
applied load vector
applied moment vector
These vectors are used to form the joint load matrix P and the joint
displacement matrix 3 as
PI i)I
P2 i)2
P= , i)= (5.1)
PNN i)NN
with
(Pit (i);}X
(Pi)y (i)i )y
(Pi)z (i)i t
Pi = (Mit i)i= (ait (5.2)
(Mi)y (a i )y
(M;}z (a i )z
Here NN is the number of moveable nodes in the structure.

The members of the structure are described by the member force
matrix F and the member displacement matrix A
F= (5.3)
FNB ANB
with
(Ft ti (i) Ati - (i)c ti

{MtL (aAti -(acti
(Mt )/ (a A)/ + [(i) Ati - (i)c )zi ];Li
(5.4)
(Mtti (a A)zi - [(i) A)yi - (i)c)/ YLi
(Mi )/ (a c )/ + [(i) A ti - (i)dzi ];Li
(Miti (a cti - [(i) A)/ - (i)c)/ YLi
NONLINEAR ANALYSIS OF SPACE FRAMES 95
Here NB is the number of bars in the structure; Ft and Mt indicate the

member forces and moments at the "+" end of the member; Fi- and Mi the
member forces and moments at the"-" end of the member and the subscripts
Xi , / ,zi indicate vector components in the local coordinate system of
member i. The generalized incidence matrix N has elements

otherwise
where
1 o 000
o o 100
o o o 0
Nt = 0
-II Li o 0
(5.6)
o o 000
000
and
-1 o o o 0 0
o o o -1 0 0
o o -II Li o 0 0
Nj= (5.7)
o 11 Li o o 0 0
o o -II Li o 1 0
o II Li o o 0 1
and the matrix ~i is formed from the three-dimensional rotation matrix R i

which describes the transformation from global to local coordinates for
member i as
96 ANALYSIS OF GEOMETRICALLY NONLINEAR SmucTUREs
Equations 5.4-5.7 were obtained using identical steps to those of Chapter 4

for plane frames. If the member forces at the positive and negative ends are
represented by It and Ij- then
Member equilibrium relates It and Ij- to F j as
where N7 and Nt have already been presented in Eqs 5.6 and 5.7. In order
to develop the equations of equilibrium ( NT F = P ) which must be
assembled in a global coordinate system, it is necessary to transform the
local member end forces using the transpose of the full rotation matrix R j
that is described in Appendix 3. Since 6 quantities (two vectors) are being
transformed and not 3 as in the case of plane frames it is essential to stack
two rotation matrices diagonally. This transformation yields lJ{j T N7 T F j and
TNt T F j • Thus for example the x-global component of a moment vector in

lJ{j
the local coordinate system becomes
Member displacements are related to nodal displacements in the

local coordinate system of the member i as
and the primitive stiffness matrix is given by

EAi
0 0 0 0 0
Li
GJi
0 0 0 0 0
Li
i i
4E1IY 2E1IY
0 0 0 0
Li L;
(K)i = 4E1t
i
2ErI
i
0 0 0 0
Li Li
i i
2E1IY 4E1IY
0 0 0 0
L; Li
i i
2ErI 4E1Iz
0 0 0 0
Li Li
where
length of member i
area of member i
torsional stiffness of member i
bending stiffness about the local y-axis of member i
bending stiffness about the local z-axis of member i

Young's modulus
modulus of rigidity
Global assemblage is identical to that presented for plane frames CEq 4.11)
to yield:
colA colC
(KE)i
[ ~i T Ni +' KiNi ~i
+ ~i Ni
T +T
KiNj~i
1 row A
T T + TN-TK N- rowC
~i Nj KiNi~; ~i ; i i ~i
5.3 Computer Program P9-FR3D.FOR
This program performs linear analysis of space frames. It is identical in form

to FR2D.FOR and in fact is its direct extension into 3 dimensions. The only
complication is the rotation matrix which will be discussed in detail below.
The orientation of members is described using three angles as input
(instead of one for plane frames) thus paving the way to define the rotation
matrix as a sequence of three rotations about the global coordinate axes that
take the global coordinate system into the local system. Appendix 3 provides
the three individual rotation matrices about the different axes and discusses
compound rotations. One generic angle, e, has been used in the appendix for
illustrative purposes and ease of reference to the same 2-dimensional Figure
A3.2. In general each member will posses three different angles of
orientation. Figure 5.2 shows a member in space with the angles of
orientation e (anticlockwise), ¢, (clockwise) and r (clockwise) that the Z-
axis, Y-axis and X-axis undergo respectively to rotate the global coordinate
system (XYZ) into the local coordinate system (xyz). Anticlockwise
directions are positive.
It is now a matter of direct matrix multiplication to obtain the
rotation matrix for member i from the individual rotations about the X, Y,
and Z-axes. This rotation matrix performs the following transformation
or
The product of the three rotation matrices forms R i as
o
cosy Si~ 1rco~ ~ ~
Y
- sin ~1r cos {}
o - sin{}
-siny cosy sin~ 0 cos~ 0
to result in
z,z' Y ,Y
~----L-----~------------------_X
Figure 5.2 Rotations in 3D.
R i = [ - sin
COS rp cos e COS rp sin e
ecos y + sin y sin rp cos e cos y cos e + sin ysin rp sin e

- sin rp
sin ycos rp
J
sin y sin e + cos ysin rp cos e - sin y cos e + cos y sin rp sin e cos ycos rp
Finally it is worth noting that the terms of each row are the projections of the
local base vectors (n x' ny, n z) of member i with respect to the global
coordinate system.
5.3.1 Example 5.1 A 2-Storey Simple Space Frame
A two storey frame is loaded horizontally by two point loads as shown in

Figure S.3 L = 100 inches and P = 1.0 lb. Also shown are the global
coordinate system and the joint numbering. Young's modulus is taken as
E=29,000,000 psi and the shear modulus is taken as G=12,000,000 psi.
This problem is solved using FR3D.FOR and the data file
FR3DS1.DAT. The results are listed in the output file FR3DS1.0UT. The
reader is particularly alerted to the manner of supplying the angles of
member orientation. The X; Y,Z global coordinates are assigned numbers
1,2,3 respectively and if member i rotates by, say, -90 0 , about the Y-axis only
(clockwise) then in the I th line of input for member information use
NT (1, I) =2 and TH (1, I) =-90; NT (2, I) =1 and TH (2, I) =0; and
NT (3, I) =1 and TH (3, I) =0. Any order of inserting a pair of axis and
angle is actually valid as long as the zero value for that member does not
appear first since NT(J,I) is used for specifying the angle of rotation for the
J-th rotation of member I and TH(J,I) is the value of the J-th rotation about
the axis NT(J,I) of member I (See input file below).
j\
Y L 9
P 13
11
®
@ 16
L ®
10
®
L
~
14 )1
Figure 5.3 Simple space frame linear analysis.
Input file FR3D51.DAT

16 12 4
1 O.OEO O.OEO 1.0EO O.OEO O.OEO O.OEO
1 O.OEO O.OEO 1.0EO O.OEO O.OEO O.OEO
1 O.OEO O.OEO O.OEO O.OEO O.OEO O.OEO
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 2 1
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 3 4
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 6 5
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 7 8
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 4 1
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 3 2
1.0EO 1.0E2 2.0EO 1.0EO 1.0+0 8 5
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 7 6
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 1 5
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 5 9
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 2 6
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 6 10
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 4 8
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 8 12
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 3 7
1.0EO 1.0E2 2.0EO 1.0EO 1.0EO 7 11
1 0.0000 1 0.0000 1 0.0000
1 0.00 1 .00 1 0.00
1 0.00 1 0.00 1 0.00
1 0.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
3 -90.00 1 0.00 1 0.00
16 NO. MEMBERS
12 NO. NODES
4 NO. SUPPORTS
APPLIED LOADS-MOMENTS
PX PY PZ MX MY MZ
1 0.00 0.00 1. 00 0.00 0.00 0.00
2 0.00 0.00 1. 00 0.00 0.00 0.00
3 0.00 0.00 0.00 0.00 0.00 0.00
4 0.00 0.00 0.00 0.00 0.00 0.00
5 0.00 0.00 0.00 0.00 0.00 0.00
6 0.00 0.00 0.00 0.00 0.00 0.00
7 0.00 0.00 0.00 0.00 0.00 0.00
8 0.00 0.00 0.00 0.00 0.00 0.00
MEMBER PROPERTIES
AREA LENGTH IX IY IZ +END -END
1 1.0000 100.0000 2.0000 1.0000 1.0000 2 1
2 1.0000 100.0000 2.0000 1.0000 1.0000 3 4
3 1. 0000 100.0000 2.0000 1.0000 1.0000 6 5
4 1. 0000 100.0000 2.0000 1.0000 1. 0000 7 8
5 1.0000 100.0000 2.0000 1.0000 1.0000 4 1
6 1.0000 100.0000 2.0000 1.0000 1. 0000 3 2
7 1.0000 100.0000 2.0000 1.0000 1.0000 8 5
8 1.0000 100.0000 2.0000 1.0000 1. 0000 7 6
9 1.0000 100.0000 2.0000 1.0000 1. 0000 1 5
10 1.0000 100.0000 2.0000 1.0000 1.0000 5 9
11 1. 0000 100.0000 2.0000 1.0000 1. 0000 2 6
12 1.0000 100.0000 2.0000 1.0000 1.0000 6 10
13 1.0000 100.0000 2.0000 1.0000 1. 0000 4 8
14 1. 0000 100.0000 2.0000 1.0000 1. 0000 8 12
15 1.0000 100.0000 2.0000 1.0000 1.0000 3 7
16 1.0000 100.0000 2.0000 1.0000 1.0000 7 11
MEMBER PROPERTIES
AXIS/ANG AXIS/ANG AXIS/ANG
1 1 0.0000 1 0.0000 1 0.0000
2 1 0.0000 1 0.0000 1 0.0000
3 1 0.0000 1 0.0000 1 0.0000
4 1 0.0000 1 0.0000 1 0.0000
5 2 -90.0000 1 0.0000 1 0.0000
6 2 -90.0000 1 0.0000 1 0.0000
7 2 -90.0000 1 0.0000 1 0.0000
8 2 -90.0000 1 0.0000 1 0.0000
9 3 -90.0000 1 0.0000 1 0.0000
10 3 -90.0000 1 0.0000 1 0.0000
11 3 -90.0000 1 0.0000 1 0.0000
12 3 -90.0000 1 0.0000 1 0.0000
13 3 -90.0000 1 0.0000 1 0.0000
14 3 -90.0000 1 0.0000 1 0.0000
15 3 -90.0000 1 0.0000 1 0.0000
16 3 -90.0000 1 0.0000 1 0.0000
JOINT DISPLACEMENTS
x Y Z TH X TH Y TH Z
1-.66223D-08 -.65806D-05 .60241D-02 -.15806D-04 .38511D-10 -.26220D-10
2-.66223D-08 -.65806D-05 .60241D-02 -.15806D-04 .38511D-10 -.26220D-10
3 .30631D-ll .65806D-05 .60224D-02 -.15792D-04 .38516D-10 -.12565D-10
4 .30 632D-ll .65806D-05 .60224D-02 -.15792D-04 .38516D-10 -.12565D-10
5-.27658D-08 -.47005D-05 .26166D-02 -.23601D-04 .62792D-11 -.35262D-10
6-.27658D-08 -.47005D-05 .26166D-02 -.23601D-04 .62792D-ll -.35262D-10
7 .42579D-09 .47005D-05 .26166D-02 -.23593D-04 .62814D-ll - .11539D-10
8 .42579D-09 .47005D-05 .26166D-02 -.23593D-04 .62814D-11 - .11539D-10
MEMBER FORCES
THRUST TORQUE MY(+) MZ(+) MY(-) MZ(-)
1-.93113E-12 .23039E-05 .10460E-04 -.45321E-04 .10460E-04 -.45321E-04
2-.38750E-12 .23039E-05 .10468E-04 -.21647E-04 .10468E-04 -.21647E-04
3-.15192E-11 .26915E-05-.65876E-05 -.61143E-04 -.65876E-05 -.61143E-04
4-.56272E-12 .26915E-05-.65837E-05 -.19934E-04 -.65837E-05 -.19934E-04
5-.49989E+00 .32813E-05-.48302E-04 .27257E+02 -.48305E-04 .27266E+02
6-.49989E+00 .32813E-05-.48302E-04 .27257E+02 -.48305E-04 .27266E+02
7-.17124E-03 .56956E-05-.44606E-04 .40893E+02 -.44607E-04 .40898E+02
8-.17124E-03 .56956E-05-.44605E-04 .40893E+02 -.44607E-04 .40898E+02
9 .54523E+00-.55178E-05 .27266E+02 .16276E-04 .22745E+02 .11031E-04
10 .13631E+01-.82227E-05 .18153E+02 .73185E-05 .31841E+02 .27770E-04
11 .54523E+00-.55178E-05 .27266E+02 .16276E-04 .22745E+02 .11031E-04
12 .13631E+01-.82227E-05 .18153E+02 .73185E-05 .31841E+02 .27770E-04
13-.54523E+00-.55164E-05 .27257E+02 -.13951E-04 .22732E+02 -.13356E-04
14-.13631E+01-.82211E-05 .18161E+02 -.20891E-04 .31845E+02 -.14198E-04
15-.54523E+00-.55164E-05 .27257E+02 - .13951E-04 .22732E+02 -.13356E-04
16-.13631E+01-.82211E-05 .18161E+02 -.20891E-04 .31845E+02 - .14198E-04
5.3.1 Example 5.2 A More Complex Space Frame
A 16 member space frame having four members rigidly held at the corners of
a square layout of S04 inches is loaded at nodes 1-6 (Fig. S.4). Three slanted
members reduce the base of the second floor to a rectangle of 324x396
inches. Young's modulus is taken as E=29,000,000 psi, the shear modulus is
taken as G=12,OOO,OOO psi and P=10,OOO lbs. The problem is solved using
FR3D.FOR and the data file FR3DS2.DAT which is listed below together
with the results FR3DS2.0UT.
Input file FR3DS2.DAT
16 12 4
O.OEO 1.0E4 -1.0E4 O.OEO O.OEO O.OEO
O.OEO 1.0E4 -1.5E4 O.OEO O.OEO a.OEO
O.OEO O.OEO -0.75E4 O.OEO O.OEO O.OEO
O.OEO O.OEO -0.5E4 O.OEO O.OEO O.OEO
O.OEO 1.25E4 O.OEO O.OEO O.OEO O.OEO
O.OEO 1. 25E4 O.OEO O.OEO O.OEO O.OEO
O.OEO O.OEO O.OEO O.OEO O.OEO O.OEO
O.OEO O.OEO O.OEO O.OEO O.OEO O.OEO
1.5EO 324.0EO 1.5EO 1.0EO 0.5EO 2 1
1.5EO 324.0EO 1.5EO 1.0EO 0.5EO 3 4
1.5EO 324.0EO 1.5EO 1.0EO 0.5EO 6 5
1.5EO 324.0EO 1.5EO 1.0EO O.5EO 7 8
P/2 3/4P
Figure 5.4 A more complex space frame linear analysis.
1. SEO 396.0EO 1.SEO 1.0EO O.SEO 4 1

1.SEO 396.0EO 1.SEO 1.0EO O.SEO 3 2
1.SEO 396.0EO 1.SEO 1.0EO O.S+O B S
1.SEO 396.0EO 1.SEO 1.0EO O.SEO 7 6
1.5EO 1BO.OEO 1.5EO 1.0EO 0.5EO 1 5
1.5E0375.B510EO 1.5EO 1.0EO O.SEO 5 9
1.5EO 1BO.OEO 1.5EO 1.0EO 0.5EO 2 6
1.5E0416.7301EO 1.5EO 1.0EO 0.5EO 6 10
1.5EO 1BO.OEO 1.5EO 1.0EO 0.5EO 4 8
1.5EO 360.0EO 1.5EO 1.0EO O.SEO 8 12
1.5EO 1BO.OEO 1.5EO 1.0EO 0.5EO 3 7
1.SE0402.4922EO 1.SED 1.DED D.SEO 7 11
1 0.00 1 0.00 1 0.00
1 0.00 1 0.00 1 0.00
1 0.00 1 0.00 1 0.00
1 0.00 1 0.00 1 0.00
3 90.00 1 0.00 1 0.00
3 90.00 1 0.00 1 0.00
3 90.00 1 0.00 1 0.00
3 90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
3 90.00 2 -73.300B 1 0.00
2 -90.00 1 0.00 1 0.00
3 149.0362 2 -59.7537 1 0.00
2 -90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
2 -90.00 1 0.00 1 0.00
3 180.00 2 -63.4349 1 0.00
16 NO. MEMBERS
12 NO. NODES
4 NO. SUPPORTS
PX PY PZ MX MY MZ
1 0.00 10000.00 -10000.00 0.00 0.00 0.00
2 0.00 10000.00 -15000.00 0.00 0.00 0.00
3 0.00 0.00 -7500.00 0.00 0.00 0.00
4 0.00 0.00 -5000.00 0.00 0.00 0.00
5 0.00 12500.00 0.00 0.00 0.00 0.00
6 0.00 12500.00 0.00 0.00 0.00 0.00
7 0.00 0.00 0.00 0.00 0.00 0.00
8 0.00 0.00 0.00 0.00 0.00 0.00
MEMBER PROPERTIES
AREA LENGTH IX IY IZ +END -END
1 1.5000 324.0000 1.5000 1.0000 0.5000 2 1
2 1.5000 324.0000 1. 5000 1.0000 0.5000 3 4
3 1. 5000 324.0000 1. 5000 1.0000 0.5000 6 5
4 1. 5000 324.0000 1.S000 1.0000 O.SOOO 7 8
5 1.5000 396.0000 1.5000 1.0000 0.5000 4 1
6 1.5000 396.0000 1.5000 1.0000 0.5000 3 2
7 1.5000 396.0000 1. 5000 1.0000 O.SOOO 8 5
8 1.5000 396.0000 1.5000 1.0000 0.5000 7 6
9 1.S000 180.0000 1.5000 1.0000 0.5000 1 5
10 1.5000 375.8510 1.5000 1.0000 0.5000 5 9
11 1.5000 180.0000 1.5000 1.0000 0.5000 2 6
12 1.5000 416.7301 1.5000 1.0000 0.5000 6 10
13 1.5000 180.0000 1.5000 1.0000 0.5000 4 8
14 1.5000 360.0000 1.5000 1.0000 0.5000 8 12
15 1.5000 180.0000 1. 5000 1.0000 0.5000 3 7
16 1.5000 402.4922 1. 5000 1.0000 0.5000 7 11
MEMBER PROPERTIES
AXIS/ANG AXIS/ANG AXIS/ANG
1 1 0.0000 1 0.0000 1 0.0000
2 1 0.0000 1 0.0000 1 0.0000
3 1 0.0000 1 0.0000 1 0.0000
4 1 0.0000 1 0.0000 1 0.0000
5 3 90.0000 1 0.0000 1 0.0000
6 3 90.0000 1 0.0000 1 0.0000
7 3 90.0000 1 0.0000 1 0.0000
8 3 90.0000 1 0.0000 1 0.0000
9 2 -90.0000 1 0.0000 1 0.0000
10 3 90.0000 2 -73.3008 1 0.0000
11 2 -90.0000 1 0.0000 1 0.0000
12 3 149.0362 2 -59.7537 1 0.0000
13 2 -90.0000 1 0.0000 1 0.0000
14 2 -90.0000 1 0.0000 1 0.0000
15 2 -90.0000 1 0.0000 1 0.0000
16 3 180.0000 2 -63.4349 1 0.0000
JOINT DISPLACEMENTS
X Y Z TH X TH Y TH Z
1 .31121D+03 .32938D+04 -.86769D+03 .77391D+00 .15139D-02 .20835D+Ol
2 .31120D+03 .39788D+04 -.93953D+03 .4S530D+00 .35516D+00 .17371D+01
3-.45338D+03 .39788D+04 -.24543D+03 .17178D+00 .53939D+00 .16669D+Ol
4-.45339D+03 .32938D+04 -.23672D+00 .14223D+00 .52633D+00 .21566D+01
5 .30038D+03 .28925D+04 -.86766D+03 -.39089D+01 .42206D+00 .20801D+01
6 .30040D+03 .36323D+04 -.93948D+03 -.27790D+01 -.S3028D+00 .S5792D+00
7-.49041D+03 .36323D+04 -.24539D+03 -.20742D+01 -.11810D+00 .37S85D+00
8-.49039D+03 .2892SD+04 -.19700D+00 -.19504D+01 -.13323D+00 .22746D+01
THRUST TORQUE MY(+) MZ(+) MY(-) MZ (-)

1-.18385E+04-.17700E+OS .83465E+04 -.70244E+OS -.54960E+OS -.39243E+05
2 .30677E+03 .16418E+04-.11907E+06 -.76283E+05 -.12141E+06 -.32451E+OS
local coordinate
system
global coordinate
system
~
X
Figure 5.5 Typical Member i
3 .25319E+04 .62773E+05-.23334E+06 -.32712E+06 -.62864E+05 -.19087E+06

4-.33754E+04-.68816E+04-.47254E+06 -.34226E+06 -.47525E+06 -.17231E+06
5-.67818E+04 .23855E+05 .80750E+06 .44256E+05 .71498E+06 .38902E+05
6-.55644E+04 .83744E+04 .65316E+06 -.52823E+05 .61163E+06 -.47679E+05
7-.77236E+03-.25240E+05 .21064E+07 .46774E+05 .23932E+07 .32525E+05
8-.51463E+04 .18736E+05 .17848E+07 -.34283E+06 .18880E+07 -.32950E+06
9-.62992E+04 .34248E+03 .78815E+05 -.69728E+06 .21432E+06 .57182E+05
10 .88253E+04 .10122E+06-.25132E+07 .21485E+06 -.31164E+07 .19992E+06
11-.11662E+05 .11792E+06 .27161E+02 -.62933E+06 -.28528E+06 -.10825E+06
12 .24232E+04 .66729E+05-.13442E+07 .14232E+07 -.16065E+07 .15305E+07
13-.95869E+04-.11803E+05-.97552E+05 .80914E+06 -.11497E+06 -.47200E+06
14-.23875E+05 .11373E+06 .61546E+06 -.16275E+07 .63692E+06 -.17846E+07
15-.99517E+04 .12911E+06 .11070E+06 -.65152E+06 - .10116E+06 -.28966E+06
16-.16825E+05 .56519E+05-.55497E+06 .17076E+07 -.57199E+06 .18291E+07
5.4 Nonlinear Effects
While the approach to geometric nonlinearities is consistent throughout the

text, it is not the only approach possible. It is of some interest to note that
there is an on-going discussion of three-dimensional frames in the literature.
This discussion includes the work of Chan and Zhou (1994), Kuo, Yang and
Chou (1993), Izzudin and Elnashai (1993), Krysl (1993).
Trusses and plane frames have moved forward with few surprises
when developing the geometric stiffness matrix. That is not the case with
three-dimensional frames where the use of the three-dimensional rotation
matrix (see Appendix 3) leads to a geometric stiffness matrix which is
unsymmetric. There is another piece to the description of the space frame.
Earlier in the case of the plane frame it was argued that the member stiffness
matrix should come from beam-column theory. For the space frame there is
an analogous three-dimensional beam column which should be used for the
member stiffness. That is discussed below in Chapter 8.
c A
Figure 5.6 Unit vector notation.
If the discussion of space frames is to include finite rotations (large

displacements but small strains) then it is not possible to use a rotation
vector to describe either joint rotations or member rotations. What must be
used in both of these cases is a rotation matrix. Euler's Theorem (see
Appendix 3) shows the equivalence between a rotation matrix and a finite
rotation about a single axis. This allows the small rotation vector of linear
analysis to be described as a perturbation of this rotation about a single axis.
The assumption of small rotational strain comes in when member
forces are to be computed given a deformed structure. Their computation is
essential if you want to perform incremental analysis since these internal
forces (the equivalent of prestress in trusses) must be saved within the
members for the next increment of load to be applied. It is argued that small
strain implies that the rotation of one end of a member with respect to the
other end can be described using a small rotation vector. This is done in the
following manner. Figure 5.5 represents a typical case of deformed
equilibrium. Associated with joints A and C are the rotation matrices R A
and Rc which describe the finite rotation of these joints (similar to the
rotation matrix, R, of Eq. A3.3 in Appendix 3); the rigid body motion of
member i is described by two rotation matrices R71d and R~ew which
correspond to the undeformed and deformed rigid body positions.
In order to apply linear structural theory to compute member forces,
it is necessary to have the rotation of joints A and C relative to the rigid body
motion of member i. Let the unit vector OJ (Figure 5.6) represent the
undeformed member axis (i.e. 0 j describes the direction of the initial line
from joint C to joint A). In general, the local coordinate system of member i
defines the rigid body motion of this member. Note that the rigid body
motion of a member is not fully determined by the unit vector 0 j since
rotation about the member axis is still possible given OJ. In order to
complete the description of the rigid body motion it will be assumed that
each member is fixed to its negative joint with regard to rigid body rotation
(twist) about the member axis.
Let the unit vector n; describe the deformed member axis. (n; is
determined from the deformed coordinates of joints A and C.) Furthermore,
let the unit vector n~ = R~ n i describe the deformed member axis at joint
C. The end of member rotation of member i at joint C is then the vector roc
(see Appendix 3, Figure A3.4) with the properties
roC-nixnC*
(5.8)
lroc 1= cos- 1 (n; . n~)
Equation 5.8 implies that the direction of roc is perpendicular to the plane
containing the vectors n; and n~ and has a magnitude in radians equal to
the angle between the two vectors. It remains to compute the end of member
rotation vector ro A for joint A. Let ro AC be the relative rotation of joint A
with respect to joint C. It follows that
(5.9)
However, ro AC is simply the vector form (Eqs A3.4, A3.5 of Appendix 3) of

1
the matrix product R Rc' The matrix product itself is the compound
rotation describing the relative rotation of joint A with respect to joint C.
Adding the two rotations in Eq. 5.9 is valid since we are actually dealing
with pure rotations that are small. It follows that the kinematic description of
a structure is now complete.
Again, the starting point for constructing the geometric stiffness matrix is the
contribution of bar i to the equilibrium equations
(5.10)
T Trow C
1{i Nt Fi
The gradient of Eq. 5.10 is then the contribution of member i to the

geometric stiffness matrix. This contribution (Eq. 5.11) has the usual form of
four sub matrices that have to be properly inserted into the global geometric
stiffness matrix according to the member's incidence. Here the (KG )iIJ
terms are 6 x 6 matrices.
colA colC
(KG )i11 (KG )i12 row A

(5.11)
(KG)i =
(KG )i21 (KG )i22 rowC
There are two points to be made with regard to the member

equilibrium equation. First, the end-of-member forces constitute two vectors,
a force and a moment at each end of a member. Second, because of the
choice which has been made in defining the member force matrix, the shears
are written in terms of the moments. This will lead to two kinds of terms
when a gradient is used to form the geometric stiffness matrix. In one type of
terms (,stretch' terms) the moments are kept fixed and the gradients of the
L ~l terms are computed; in the other type the vectors are held constant and
the gradient is computed as a small rotation ('rotation' terms). This can be
seen by looking directly at the terms in Eq 5.10. Using the product rule for
differentiation, it follows that a small perturbation of either term in this
equation has the form
in which the member force matrix Fi

is held fixed.
The stretch terms are most easy to obtain. These are derived from
the first term of the right hand side of Eq. 5.12 as the gradient,
V(il{i T N~ T Fi) .Note that the matrices Nt only contain constants
fixed 8i ,tpi ,ri
and the term Li 1 which arises when member shears are obtained from
moments by dividing by the member length. Clearly, Li depends only on the
coordinates of the end of the member. The gradient of Li 1 then follows
directly as
aD! aD!
_1_== _ _1_
aXe aXA
aD! aD!
- -1= = - -1- (5.13)
aye ayA
OL-:! OL-:!
- -1= = - -1-
aZ e aZ A
Let each (KG )i1J term in Eq. 5.11 be written as the sum of a stretch term
and a rotation term according to Eq. 5.12
(KG )iIJ == (KG )~J + (KG )~J
The stretch term can be written directly as
kGll kG 12 kG 13 -kGll -kG12 -kG13

kG 21 kG 22 kG 23 -kG2! -kG22 -kG23
kG3! kG 23 kG33 -kG3! -kG32 -kG33

(KG )7!! == (KG )722 == (5.14)
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
where
In this equation fj + is the 6 x 1 column matrix that originates from the nodal
equilibrium vector CJ{j T N7 T F j with the thrust term set equal to zero. The
term 17, j = 1---+6 is then the j th element of that column matrix. The result
is a nonsymmetric contribution. Clearly
(5.16)
Before proceeding with the rotation term it is instructive to

demonstrate the derivation of one of the terms of Eq. 5.15, say, the term
kG 13 . This term is evaluated from the partial derivative with respect to
XA of the 3rd term ofEq. 5.10. In expanded form it becomes
and appears in Eq. 5.15.

The rotation term in the geometric stiffness which comes from
(d ~i T )N ~ T Fi , the second term on the right hand side of Eq. 5.12, can be
computed using the relationship from rigid body mechanics (Goldstein,
1957) which states that the change of a vector G under a small rigid body
rotation that is described by the small rotation vector 0) can be written as
dG = 0> x G = -Gxo> (5.18)
This is a much more simple procedure to use than computing the required
derivatives by the direct use of the gradient.
In matrix form the cross product ofEq 5.18 becomes:
dG = Go>, (5.19)
or
In this discussion G plays the role of a force vector and a moment vector in
the 'plus' and 'minus' ends of the member respectively, i.e. Gforce' Gjorce'
G ~oment and G ~oment .

Several steps are required to find the 'rotation' tenns of the geometric
stiffness matrix. First of all it is necessary to write the rotation vector (0 in
terms of the joint displacements. This is typically done by working first in
the local coordinate system of member i and then transfonning these results
into the global coordinate system. Finally, the cross product of Eq S.18 is
written as a matrix product.
Let 0); be the 3 x 1 column matrix representing the small rotations in
the local coordinate system. The components of 0); are simply
(S.20)
(O)i)~ = [(OA)/ -(Oc)/]/ Li

Note that the first of Eq S.20 comes from the arbitrary assumption of fixing
the rigid body rotation of the negative end C to the rotation of joint C. Put
into matrix fonn Eq S.20 becomes
0); =A~o~ +A~o~

where
~l A~ ~[~ ~l
1 0
A~ ~[~
0 0 0 0 0 0
0 _rl
I 0 0 0 + rl
i 0 0 (S.21)
+ rl
i 0 0 0 r:l
I 0 0 0
These matrices are transfonned into the global coordinate system as
(S.22)
The small rotations vector (OJ' can now be transfonned into global
coordinates as
(S.23)
In this equation the matrices of Eq 5.21 are "stacked" so that they can
operate simultaneously on two vectors.
The "rotation" terms of the geometric stiffness matrix can now be
completed. These rotation terms that stem from (d~;r)N~ T F j are in essence
the change in nodal force and moment equilibrium equations of member i
(Eq. 5.10) due to the small rotations O>j. Using Eq. 5.19 to describe the
changes, the rotation terms emerge from
~.T N·+T F· }
~dIP.
"'\.1
+T
T)N -:-
I
l i T
F. = d {
I I
~;rNi F j
(5.24)
as
(KG)~l =G+ AA
(KG)~2 =G+ Ac
(5.25)
(KG)~l =G- AA
(KG )~2 = G - Ac
where
+
Gmoment
l; G- =[G force
and
l
-gZl 0
gi ; G~oment = - g6_
o -gs
Here gt is the 6 x 1 column matrix obtained by computing the product
~i T Nt T F i • The term g t is then the i-th element of that column matrix.
Moreover gj is the 6 x 1 column matrix obtained by computing the product

T
~i N i
_T
F i · The term g i is then the i-th element of that column matrix.
The result is a nonsymmetric contribution.
Displayed below is a 12x12 layout of the geometric stiffness matrix
with a mapping of the stretch terms, S, and the rotation terms, R. Columns
10-12 are due to (m i ); =(ecti that takes the value of "1" inside the matrix
A~ (Eq. 5.21). When A~ is transformed to global coordinates it fills out
these columns.
1 2 3 4 5 6 7 8 9 10 11 12
1 RS RS RS S S S RS RS RS RS RS RS
4 R R R 0 0 0 R R R R R R
5 R R R 0 0 0 R R R R R R
6 R R R 0 0 0 R R R R R R
10 R R R 0 0 0 R R R R R R
11 R R R 0 0 0 R R R R R R
12 R R R 0 0 0 R R R R R R
5.6 Computer Programs PIO-FR3DNL2.FOR, Pll-FR3DNL3.FOR

and P12-FR3DNLSR.FOR
FR3DNL2.FOR is a three dimensional version of the nonlinear plane frame

program incorporating the nonlinear effects and the geometric stiffness
matrix as described in the previous sections. It also uses the two dimensional
beam-column equations in its member stiffness matrix. This is a matter of
expediency. It is proper to use the three-dimensional beam-column of
Chapter 8 below when computing the member stiffness matrix for three-
dimensional frames and that is done here in program Pll-FR3DNL3.FOR.
Unfortunately, the program which uses the finite difference solution of
Chapter 8 is computationally intensive and runs too slowly for practical
applications on a PC. The program P12-FR3DNLSR uses a power series
solution and is much faster. The compromise of program P10-
FR3DNL2.FOR is to use two-dimensional beam column effects in the
member stiffness together with short enough member lengths to mitigate the
absence of three-dimensional beam column effects such as lateral buckling.
5.6.1 Example 5.3 Lateral Torsional Buckling
This example is a good test for computer programs for space frames because
it fails in torsional buckling with the out-of-plane bending around the weak
axis. The particular parameters used in this case come from an experiment
(Spillers et ai" 1993) for which the buckling load was measured to be about
25 lbs. (See Figure 5.7.) With regard to input data, this program requires
coordinates to be specified and therefore only one angle of orientation is
necessary. That angle is y. If the other angles are also specified they will be
automatically replaced with calculated angles for reasons of internal
consistency. The input file for this example is FR3DNL53.DAT.
The output file FR3DNL53.0UT is partially listed below. First an
echo of the data is returned. Then with each iteration after the displacements
are calculated both nodal and member quantities are updated. Updated nodal
coordinates are listed under NEW COORDINATES. For members there is the
INCREMENTAL ELASTIC RESPONSE that provides member
deformations in the first line and member forces in the second line for the
current iteration and load step. UPDATED NONLINEAR RESPONSE
provides the accumulated deformations and 'forces'. The accumulated
member forces are assembled and listed again under CURRENT MEMBER
FORCES.
The output file is scanned for the word NEG which appears when
the system becomes singular. In this case the system is stable at load step 7
and unstable at load step 8 which implies that the buckling load lies between
811 Ox28=22.4 lbs and 911 Ox28=25.2 lbs which agrees with the observed
buckling load.
Interestingly, this problem exhibits two different bifurcation loads,
depending on the direction of loading. For the upward loading, buckling of
the frame was reached at 62% of that for the downward loading.
Figure 5.7 L-frame in lateral torsional buckling.
Input file FR3DNL53.DAT
10 11 1 10 5
28.
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
. 12 2.1 .000576 .000144 .01 2 1 o.
.12 2.1 .000576 .000144 .01 3 2
.12 2.1 .000576 .000144 .01 4 3
.12 2.1 .000576 .000144 .01 5 4
.12 2.1 .000576 .000144 .01 6 5
. 12 2.1 .000576 .000144 .01 7 6 -90 .
. 12 2.1 .000576 .000144 .01 8 7 -90 .
. 12 2.1 .000576 .000144 .01 9 8 -90 .
. 12 2.1 .000576 .000144 .01 10 9 -90 .
.12 2.1 .000576 .000144 .01 11 10 -90.
o. 10.64
2.1 10.64
4.2 10.64
6.2 10.64
8.4 10.64
10.5 10.64
10.5 8.512
10.5 6.384
10.5 4.256
10.5 2.128
10.5
10 NO. MEMBERS
11 NO. NODES
1 NO. SUPPORTS
10 NO. LOAD STEPS
5 NO. ITERATIONS PER STEP
PX PY PZ MX MY MZ
1 .OOOOOD+OO .28000D+02 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
2 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
NODE COORDINATES
X Y Z THX THY THZ
1 .OOOOOE+OO .10640E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
2 .21000E+01 .10640E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
11 .10500E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
MEMBER PROPERTIES
AREA LENGTH IX IY IZ +END -END THETA PHI GAMMA
1 0.12 2.10 0.000576 0.000144 0.010000 2 1 0.00 0.00 0.00
2 0.12 2.10 0.000576 0.000144 0.010000 3 2 0.00 0.00 0.00
3 0.12 2.00 0.000576 0.000144 0.010000 4 3 0.00 0.00 0.00
4 0.12 2.20 0.000576 0.000144 0.010000 5 4 0.00 0.00 0.00
5 0.12 2.10 0.000576 0.000144 0.010000 6 5 0.00 0.00 0.00
6 0.12 2.13 0.000576 0.000144 0.010000 7 6 -90.00 0.00 0.00
7 0.12 2.13 0.000576 0.000144 0.010000 8 7 -90.00 0.00 0.00
8 0.12 2.13 0.000576 0.000144 0.010000 9 8 -90.00 0.00 0.00
9 0.12 2.13 0.000576 0.000144 0.010000 10 9 -90.00 0.00 0.00
10 0.12 2.13 0.000576 0.000144 0.010000 11 10 -90.00 0.00 0.00
******** NEW LOAD STEP ***********

THRUST TORQUE MY{+) MZ{+) MY{-) MZ{-)
1 .OOOOOE+OO OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
*** ERROR = .56000000E+Ol

LOAO STEP 1
ITERATION NUMBER 1
APPLIEO LOAOS-MOMENTS
PX PY PZ MX MY MZ
1 .000000+00 .560000+01 .000000+00 .000000+00 .000000+00 .000000+00
2 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
3 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
4 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
5 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
6 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
7 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
8 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
9 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
10 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
JOINT OISPLACEMENTS
1 .33284E-Ol .87350E-Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.93433E-02
6 .33284E-Ol .49652E-04 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.62563E-02
9 .53254E-02 .19861E-04 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.25025E-02
10 .13313E-02 .99306E-05 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO - .12513E-02
NEW JOINT COORDINATES

1 .33284E-Ol .10727E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.93433E-02
2 .21333E+Ol .10708E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.92198E-02
4 .62333E+01 .10672E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.82670E-02
8 .10512E+02 .63840E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.37538E-02
INCREMENTAL ELASTIC RESPONSE

1 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .82320E-04 .OOOOOE+OO -.41160E-04
1 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO . 11760E+02 .OOOOOE+OO -.15102E-11
UPOATEO NONLINEAR RESPONSE
1 .90855E-04 .OOOOOE+OO .OOOOOE+OO .82052E-04 .OOOOOE+OO -.41428E-04
1 .51917E+02 .OOOOOE+OO .OOOOOE+OO . 11683E+02 .OOOOOE+OO -.76655E-Ol

2 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .20580E-03 .OOOOOE+OO -.16464E-03
2 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .23520E+02 .OOOOOE+OO - .117 60E+02
UPOATEO NONLINEAR RESPONSE
2 .49197E+02 .OOOOOE+OO .OOOOOE+OO .23449E+02 .OOOOOE+OO - .11831E+02

10 .56000E+01 .OOOOOE+OO .OOOOOE+OO .58800E+02 .OOOOOE+OO -.58800E+02
UPDATED NONLINEAR RESPONSE

THRUST TORQUE MY(+) MZ(+) MY(-) MZ(-)
1 .51917E+02 .OOOOOE+OO .OOOOOE+OO . 11683E+02 .OOOOOE+OO -.76655E-01
2 .49197E+02 .OOOOOE+OO .OOOOOE+OO .23449E+02 .OOOOOE+OO - . 11831E+02
8 . 11471E+02 .OOOOOE+OO .OOOOOE+OO .58781E+02 .OOOOOE+OO -.58819E+02
*** ERROR = . 64062560E+02
MEMBER PROPERTIES
1 0.12 2.10 0.000576 0.000144 0.010000 2 1 -.53 0.00 0.00
2 0.12 2.10 0.000576 0.000144 0.010000 3 2 -.52 0.00 0.00
3 0.12 2.00 0.000576 0.000144 0.010000 4 3 -.49 0.00 0.00
4 0.12 2.20 0.000576 0.000144 0.010000 5 4 -.45 0.00 0.00
5 0.12 2.10 0.000576 0.000144 0.010000 6 5 -.39 0.00 0.00
6 0.12 2.13 0.000576 0.000144 0.010000 7 6 -90.32 0.00 0.00
7 0.12 2.13 0.000576 0.000144 0.010000 8 7 -90.25 0.00 0.00
8 0.12 2.13 0.000576 0.000144 0.010000 9 8 -90.18 0.00 0.00
9 0.12 2.13 0.000576 0.000144 0.010000 10 9 -90.11 0.00 0.00
10 0.12 2.13 0.000576 0.000144 0.010000 11 10 -90.04 0.00 0.00
LOAD STEP 7
ITERATION NUMBER 5
JOINT DISPLACEMENTS
1 -.13128E-08 -.75124E-08 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .92972E-09
3 - .11744E-08 -.36984E-08 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .83103E-09
4 - .1l238E-08 -.22078E-08 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .63826E-09
6 -.10583E-08 .16148E-IO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .30153E-09
7 -.58958E-09 .61334E-ll .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .17268E-09
8 -.25959E-09 .30372E-12 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .13391E-09
9 -.44756E-10 -.22361E-ll .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .59862E-10
10 .57997E-ll -.19451E-ll .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .80584E-12
NEW JOINT COORDINATES

6 .10632E+02 .10639E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.24802E-Ol
8 .10547E+02 .63839E+01 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO - .14841E-Ol

THRUST TORQUE MY(+) MZ(+) MY(-) MZ (-)
1 .82812E+00 .OOOOOE+OO .OOOOOE+OO .47008E+02 .OOOOOE+OO .19273E-07
*** ERROR = .15142980E-04
MEMBER PROPERTIES
1 0.12 2.10 0.000576 0.000144 0.010000 2 1 -2.12 0.00 0.00
2 0.12 2.10 0.000576 0.000144 0.010000 3 2 -2.06 0.00 0.00
3 0.12 2.00 0.000576 0.000144 0.010000 4 3 -1. 95 0.00 0.00
4 0.12 2.20 0.000576 0.000144 0.010000 5 4 -1.78 0.00 0.00
5 0.12 2.10 0.000576 0.000144 0.010000 6 5 -1. 55 0.00 0.00
6 0.12 2.13 0.000576 0.000144 0.010000 7 6 -91.28 0.00 0.00
7 0.12 2.13 0.000576 0.000144 0.010000 8 7 -90.99 0.00 0.00
8 0.12 2.13 0.000576 0.000144 0.010000 9 8 -90.71 0.00 0.00
9 0.12 2.13 0.000576 0.000144 0.010000 10 9 -90.42 0.00 0.00
10 0.12 2.13 0.000576 0.000144 0.010000 11 10 -90.14 0.00 0.00
LOAD STEP 8
ITERATION NUMBER 3
NEG TERM ON THE DIAGONAL AT ROW 58 $$$$
5.7 Problems
1. Prove that the relative rotation of joint A with respect to joint C is a

1
compound rotation given by the matrix product R Rc .
2. Redo Example 4.4 as a three dimensional beam.
3. Introduce a small lateral load in the y-direction in Example 5.2 and plot
the coordinates of the tip as P is incremented.
CHAPTER 6
NONLINEAR ANALYSIS OF MEMBRANES
6.1 Introduction
This chapter is an addition to the sequence of discussions of discrete

systems. It is included here for several reasons. In terms of applications, if
you are going to discuss fabric structures you need access to a membrane
finite element. The plane stress/strain element which follows is probably the
most simple finite element which can be discussed. Since the functional
dependence of the element upon the node coordinates is so simple, it is an
easy matter to apply perturbation methods to it.
Subsequently the book discusses three-dimensional beam-columns
as continuous systems rather than discrete systems. This membrane finite
element is in these terms something of a hybrid. Membranes are, of course,
continuous in the sense that they are properly described by their field
equations. The most direct way to come up with a nonlinear membrane finite
element would be to first of all derive the appropriate field equations and
then formulate some discrete solution of them such as the finite element
solution. In the material below an alternative approach is taken. That is, the
discretization of the finite element method for the linear membrane problem
is accepted as a starting point and then perturbation methods applied to it.
Certainly the procedure is simple.
The work below proceeds as follows. First, Zienkiewicz's plane
stress constant strain finite element is introduced and its corresponding
geometrically nonlinear finite element is derived. This is a two-dimensional
result. It is then a simple matter to construct the three-dimensional finite
element using the rotation matrix. An application is included with the
computer program which is discussed subsequently.
6.2 The Geometric Stiffness Matrix of the Plane Stress Triangular

Finite Element
For the purpose of this book, the triangular plane stress (constant stress)
finite element (Fig. 6.1) described by Zienkiewicz (1977) is used. This
section derives its element's nodal force vector in the 'global' two-
dimensional coordinate system of the element's plane. This global
coordinate system makes sense in problems of two-dimensional elasticity

only. It reduces to a member's local coordinate system for the three-
dimensional membrane shells considered herein.
Let the nodal displacement vector be defined as
Ui
Vi
uj
ae = (6.1)
Vj
urn
vrn
and the displacement field be approximated using linear shape functions

Gr (x, y), r = i, j, m in both the x and y directions such that
The shape function coefficients are determined using the condition that for a
given degree of freedom its shape function will attain a value of unity and
the shape functions that are linked to the other degrees of freedom will attain
a value of zero. The shape functions may, thus, be written as
(r = i,j,m) (6.3)
and each is required to satisfy the following three conditions
X,
I
Y~
I
L -_ _ _ _ ~I-----------------------+x
Figure 6.1 Plane stress triangular finite element.

NONLINEAR ANALYSIS OF MEMBRANES 123
(6.4)
for a given r. When Eq. 6.3 is expanded for the case of r=i and s=i, j, m it
becomes
Ai + Bixi +CiYi =1 [1
Ai + BiXj +CiYj =o=> 1 Xj (6.5)
Ai + Bix m + CiYm =0 1 xm
and yields the following shape function coefficients:
A. = xjYm -xmYj . B. = Yj - Y m . C = Xm -Xj (6.6)

I 2A 'I 2A' I 2A
where A is the area of the triangle ijm so that
1 x·
[
2A =det 1 X~ (6.7)
1 xm
Proceeding in the same manner and calculating the rest of the shape
function coefficients by introducing the remaining two conditions into Eq.
6.3, i.e.
and changing variables so that ay = 2AAy; by = 2ABy; cy = 2ACy ;

(r = i, j, m) the full set of shape function coefficients becomes:
a i = xjYm - xmYj; bi = Yj - Y m = Yjm; c i = xm - Xj = x mi ;

a j =xmYi -xiYm; b j =Ym -Yi =Ymi; c j =xi -xm =x im ; (6.8)
am =xiYj-xjYi; bm =Yi-Yj =Yij; c m =xj-xi =x ji
The strains are described using the following linear relationship

o
-o {u(x,y)} = LU(x,y) (6.9)
cy V(X,y)
o
Substitution for u(x,y) from Eq. 6.2 yields constant strains since
or
(6.11 )
where
iG r
0
c~
a
Br =LIG r = 0
iGr =_1 0
0: 2A
[b' ] r=i,j,m (6.l2)
iG r iG r Cr br
0: a
Now since Br is constant it makes the strains E(X, y) = B(x, y) = constant.

Stresses are obtained, using the usual linear constitutive rule, as
o
j
O"X(X'Y))
a(x,y)= O"y(x,y) =~ v
I-v
1 v
I-v
o E(X,y)= DE(X,y) (6.13)
T xy (x, y) 0 0
2
where E is Young's modulus and v is Poisson's ratio and, O"x(X,y) ,

0" y(x, y) and T xy (x, y) are the usual stresses of plane elasticity. Substitution
ofEq. 6.11 into Eq. 6.13 yields constant stresses of
(6.14)
The linear elastic stiffness matrix which takes the general form of
(6.15)
reduces to a constant matrix, for the case of the constant stress triangular
element,
KE = fBTDBtdxdy=BTDBtA (6.16)
v
where t is the thickness of the element. Equation 6.16 may be expanded to

yield a 6x6 two-dimensional stiffness matrix
(KE )u (KE )ij (KE )im

(K E)= (KEt (KEtJ (KEtm (6.17)
(KE)mi (KE )mj (KE)mm

with
I-v
\K E rs 4 A 2 r
( e) =_t_._E_ brb s +-2- crcs
1- v
I-v vb r bs +--b c
(6.18)
2 r s
for r,s i, j, m.
=
The element's nodal force vector in the local coordinate system is
defined as
(6.19)
and when Eq. (6.16) is substituted it becomes Fe = tABTDBa e or
(6.20)
More explicitly the two-dimensional nodal force vector (Eq. 6.20) in the
element's plane coordinate system may be written as
biO'x +Cil"xy (Fi )x

rowi
ciO' y + bil" xy (F;)y
b jO' x + C jl" xy t (Fjt

Fe =~ row j (6.21)
2 CjO'y +bjl"xy 2 (Fj )y
bmO'x +cml"xy (Fmt

rowm
cmO'y +bml"xy (Fm)y
As was done in the previous chapters the geometric stiffness matrix

is derived as the gradient of the nodal force vector, Fe (Eq. 6.21). The
gradient of Fe has a particularly simple form. It comprises the geometric
stiffness matrix having 9 sub matrices (instead of 4 in two noded elements
l
like trusses and beams) arranged in an anti symmetric manner as
OA
KG = -A 0 (6.22)
A -A
where
A=-t [-l" xy (6.23)

2 -O'y
The gradient has been taken with respect to the node coordinates Xi' Y i ;
Xj'Yj; xm,Ym; while holding down the stresses O'x,O'y,l"xy fixed.
Equations 6.18 and 6.22 give K~ and KG explicitly for plane elasticity
problems. An additional step is required to deal with three dimensional
membranes.
6.3. The Geometric Stiffness Matrix of Three-dimensional

Membranes
A plane stress problem defines a two-dimensional (x, y) displacement field

while a three-dimensional membrane in general defines a three-dimensional
(x,y,z) displacement field. For linear elastic problems, a membrane has no
out-of-plane stiffness so that the 3-D element stiffness can be constructed by
simply "rearranging" terms from Eq. 6.18 and Eq. 6.23 in a 3-D array i.e.
expanding the 2x2 matrices to 3x3 by inserting a zero third row and a zero
third column. That is not true in the nonlinear case. Just as the string (truss
bar) has a "geometric" stiffness component normal to its axes, the nonlinear
membrane has an out-of-plane "geometric" stiffness. That effect will be
described below using the relationship from mechanics (that was introduced
in Chapter 5) which gives the change dF of a vector force F which is
subjected to a small rigid body rotation vector co as
dF = co x F = -Fxco (6.24)
This approach is possible since within small strain theory an out-of-

plane displacement produces no strain in a membrane. Certainly this is a
simple procedure to use and will in general produce rather accurate results in
incremental nonlinear analysis where equilibrium is satisfied at each
increment of load.
Equation 6.24 expands for local quantities, into
(6.25)
First of all, it should be clear that small, in-plane (z component)

rotations are included within the geometric stiffness matrix shown in Eq.
6.22. It is the effect of out-of-plane rotations (x, y, components only) which
must be added to the existing formulation to complete the 3-dimensional
model. Given the x, y, z displacement components of the nodes of a typical
finite element (Figure 6.2) the out-of-plane rotation components OJx " OJ y ' in
the local coordinate system, (x', y', z') are
(6.26)
J-x
z·
z
Figure 6.2 A triangular finite element in its local coordinate system.
(6.27)
These rotations are obtained from Figure 6.3 as
t5 ji
and - ¢ = OJ • = - - (6.28)
Y e
where
(6.29)
e
It .1
Figure 6.3 Out-of-plane rotations.

Now since the rotation components are linear in displacements (Eqs. 6.26,
6.27), the rotation vector 0> can be written as
o>=A·o e (6.30
or
(OJ)x'
(OJ )y.
(OJ )z.
rl
e-c c
0 0 --- 0 0 0 0 lIa (Oft.
ae ae
lOy' - 0 0 lie 0 0 -lie 0 0 0 (0 f )y. (6.31)
lOz' 0 0 0 0 0 0 0 0 0 (oJz.
(omt·
(om )y.
(omt·
It only remains to construct a matrix representation of the incremental forces
Eq. 6.24, produced by these rotations acting upon the element nodal forces
ofEq. 6.21. Noting that out-of plane components of nodal forces are zero i.e.
(Frt. =0 for r = i,j,m, the incremental forces (Eq. 6.24) for each node
may be written as
(6.32)
(6.33)
(6.34)
and the incremental force vector for member e becomes
(6.35)
where F* is a 9x3 matrix obtained by stacking the 3x3 matrices ofEqs. 6.32-
6.34. Substitution ofEq. 6.31 into Eq. 6.35 yields
(6.36)
The required out-of-plane contribution, K~, to the geometric

stiffness matrix, in the local coordinate system is then simply
K~ =-F*A* (6.37)
or more explicitly
(6.38)
where
~.].
ar
, K eG*jr =[~ ~ ~ l·
0 0 f3r
' e
KG *mr =[~ ~ ~ 1
0 0 Ar
(6.39)
where
(6.40)
f3r =-~(Fr)y. +..!..(Frt.; r =i,j,m (6.41)

ae e
1
Ar =-(F,)y'; r =i,j,m (6.42)
a
6.4 A Direct Alternative Derivation of the Geometric Stiffness

Matrix of Three-Dimensional Membranes
As was previously noted the geometric stiffness matrix is the gradient of the
element nodal force vector in global coordinates, i.e.
(6.43)
The matrix ~e in Eq. 6.43 is formed from the three-dimensional rotation

matrix R e which describes the transformation of member e from global to
local coordinates as follows:
(6.44)
Here, the rotation matrix R e is assembled from
(i't] ={i'r}
(rt (6.45)
(k't k'
where i', j', k' are the unit vectors in the x', y', z' directions of the local
coordinate system which is described in Figure 6.2.
No attempt is made here to perform the gradient of Eq. 6.43 but for
the sake of completeness of the discussion of the previous section of in-plane
and out-of-plane effects, the gradient is split in two to yield
(6.45)
The first term on the right-hand side of Eq. 6.45 denotes the 'in-
plane' effect of the geometric stiffness matrix, which requires only the
gradients of member length, i.e., b" Cr. The second term on the right-hand
side of Eq. 6.45 relates to the 'out-of-plane' effect of the matrix, which
stems from the gradient of the rotation matrix. It is the calculation of this
term that poses some hardship and the previous section was devoted to
circumventing this difficulty by deriving a geometric stiffness matrix that
assumes small rigid body rotations.
6.5.1 Program P13-FEMPS.FOR
This program is one of the most simple finite element analysis programs. It
is the constant stress finite element program carefully described in
Zienkiewicz's book and is in fact quite similar to the other programs
described in this book. In terms of loads and coordinates this program is
identical to the earlier programs; it also includes the same solver used in
these earlier programs. The coefficients b (here B) and C (here CZ) are those
ofEq.6.S.
6.5.2 Program P14-MEMBRN.FOR
This program is a step along the way toward a computer program for fabric
structures or membranes. It is a linear analysis program which does not take
into account changes in geometry. The elements themselves only have in-
plane stiffness so that you end up with a structure which can only transmit
axial forces and in-plane shear which is of course the membrane shell.
It is well known that to construct a membrane shell finite element
program it is simply necessary to start with a plane stress finite element and
rotate the element until it has the proper geometric orientation in space. That
is what is done here and also the reason that much of the code which follows
is taken from program P 13.
This program has the structure of the other analysis programs in the
book. It uses the following subroutines:
-ABC. This subroutine computes the coefficients band c and the surface
areaA.
-COMPKT. This subroutine collects terms from a global array and places
them in a local array.
-PCOORD sets up so-called "plane coordinates" and determines the rotation

matrix from the element coordinates.
-PLSTR sets the member stiffness for a plane stress/strain element.
-FORCES computes the member stresses/strains given the member

displacements.
-TRANS rotates a vector from global to local coordinates.
-RASM rotates the member stiffness to global coordinates.
6.5.3 Program PI5-MEMNL.FOR
This program is simply program P14 with geometrical nonlinearities added.

Truss bars are also included as an added capability since they would be
needed for a typical fabric structure application. New subroutines in this
program include:
-MGS T I F generates the geometric stiffness matrix.
-UPDATE calculates the new member forces once rigid body rotations have
been removed.
This program uses the same input as program P14.
6.6 Examples
6.6.1 Example 6.1. A Deep Beam
This example is a "deep beam", that is a beam in which the depth is not
small in comparison to its span. In this case, the problem data is generated by
a simple computer program FEMPSDAT.FOR. This program generates a
rectangular finite element grid which is subsequently divided into triangles.
The variables in this program include: NX = number of grid lines in the x
direction; NY = number of grid lines in the y direction; T = beam thickness;

H = beam height and AL = beam length.
It this case the program places a unit load at the end of the beam.
The easiest way to run these programs is to compile both the data generation
program, FEMPSDAT.FOR, and the finite element analysis program,
FEMPS.FOR, and then use the DOS pipeline command to concatenate their
execution FEMPSDATIFEMPS>FEMPS.OUT. The beam is 30 inches long
and 28 inches deep and 1 inch thick. Young's modulus is 29,000,000 psi and
Poison's ratio is .32. Figure 6.4 shows the structural analysis model.
The program reads joint coordinates R, loads P, member incidence nodes
NP, and thickness S. It computes joint displacements P, strains AZ and
stresses, AI. The output file is listed below. Assessment of the results as
representative of the phenomenon is left as an exercise to the reader.
Output file FEMPS.OUT
COORDINATES LOADS
X Y PX PY
1 .30000000D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO -.10000000D+Ol
2 .30000000D+02 .93333330D+Ol .OOOOOOOOD+OO .OOOOOOOOD+OO
3 .30000000D+02 .18666670D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
13 .75000000D+Ol .OOOOOOOOD+OO .OOOOOOOOD+OO .OOOOOOOOD+OO
14 .75000000D+Ol .93333330D+Ol .OOOOOOOOD+OO .OOOOOOOOD+OO
15 .75000000D+Ol .18666670D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
16 .75000000D+Ol .28000000D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
18 .OOOOOOOOD+OO .93333330D+Ol .OOOOOOOOD+OO .OOOOOOOOD+OO
19 .OOOOOOOOD+OO .18666670D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
20 .OOOOOOOOD+OO .28000000D+02 .OOOOOOOOD+OO .OOOOOOOOD+OO
ELEMENT ELEMENT NODES THICKNESS

1 5 1 2 .10000000E+Ol
2 5 2 6 .10000000E+Ol
3 6 2 3 .10000000E+Ol
4 6 3 7 .10000000E+Ol
5 7 3 4 .10000000E+Ol
6 7 4 8 .10000000E+Ol
7 9 5 6 .10000000E+Ol
8 9 6 10 .10000000E+Ol
9 10 6 7 .10000000E+Ol
10 10 7 11 .10000000E+Ol
11 11 7 8 .10000000E+Ol
12 11 8 12 .10000000E+Ol
13 13 9 10 .10000000E+Ol
14 13 10 14 .10000000E+Ol
15 14 10 11 .10000000E+Ol
16 14 11 15 .10000000E+Ol
17 15 11 12 .10000000E+01
18 15 12 16 .10000000E+Ol
20 6 4
24 16 12 6
22 16 10 4
20 14 8 2
19
17
Figure 6.4 The plane stress cantilever beam.
19 17 13 14 . 10000000E+Ol
20 17 14 18 .10000000E+Ol
21 18 14 15 .10000000E+Ol
22 18 15 19 .10000000E+Ol
23 19 15 16 .10000000E+Ol
24 19 16 20 .10000000E+Ol
DISPLACEMENTS
x Y
1 -.12561651D-06 -.29653664D-06
2 -.24175047D-07 -.23195584D-06
3 .25915595D-07 -.20460964D-06
4 .79456403D-07 -.19862350D-06
5 -.96539144D-07 -.16937614D-06
6 -.20086171D-07 -.15520392D-06
7 .24816674D-07 -.14673469D-06
8 .75709248D-07 -.14552769D-06
9 -.72706518D-07 -.95010275D-07
10 -.15657549D-07 -.86800959D-07
11 .19421555D-07 -.84794679D-07
12 .61803372D-07 -.87771887D-07
13 -.41962270D-07 -.40817802D-07
14 -.80423984D-08 -.32318339D-07
15 .10651986D-07 -.32806949D-07
16 .35986177D-07 -.37874193D-07
STRAINS STRESSES
EX EY GA SX SY TAU
1-.38770D-08 .69194D-08-.60860D-08 -.53722D-Ol .18347D+OO -.66854D-Ol
2-.54518D-09 .15185D-08-.20422D-08 -.19152D-02 .43422D-Ol -.22433D-Ol
136 ANALYSIS OF GEOMElRICALLY NONLINEAR STRUCTURES
3-.54518D-09 .29299D-08-.48667D-08 .12678D-01 .89025D-01 -.53460D-01

4 .14652D-09 .90742D-09-.29056D-08 .14115D-01 .308320-01 -.31918D-01
5 .14652D-09 .64137D-09-.19801D-08 . 11365D-01 .22237D-01 -.21752D-01
6 .49962D-09 .12932D-09-.16267D-08 .17479D-01 .93436D-02 -.17869D-01
7 .31777D-08 .15185D-08-.17241D-08 -.86967D-01 .16206D-01 -.18939D-01
8 .59048D-09 .87957D-09-.30080D-08 -.99840D-02 .22313D-01 -.33042D-01
9 .59048D-09 .907420-09-.43094D-08 -.969600-02 .23212D-01 -.47338D-01
10 .71935D-09 .214960-09-.45002D-08 .25463D-01 .14382D-01 -.49434D-01
11 .71935D-09 .12932D-09-.28059D-08 .24578D-01 .116150-01 -.308220-01
12 .18541D-08-.31899D-09-.31599D-08 .566060-01 .886320-02 -.34711D-01
13-.40992D-08 .87957D-09-.11133D-08 -.123350+00 - .139630-01 -.12229D-01
14 .10154D-08 .91066D-09-.36301D-08 -.23389D-01 .189240-01 -.398760-01
15-.10154D-08 .21496D-09-.35059D-08 -.305820-01 -.355250-02 -.385120-01
16 .11693D-08-.523510-10-.49287D-08 .37236D-01 .103970-01 -.541410-01
17 .11693D-08-.318990-09-.23908D-08 .34479D-01 .178280-02 -.262620-01
18 .34423D-08-.54292D-09-.39386D-08 .10560D+00 .180480-01 -.43265D-01
19 .55950D-08 .91066D-09-.18081D-08 -.17135D+00 -.28423D-01 -.19862D-01
20 .10723D-08 .000000+00-.43091D-08 -.34645D-01 -.11086D-01 -.47335D-01
21-.10723D-08-.523510-10-.23061D-08 -.35186D-01 -.127780-01 -.253330-01
22 .14203D-08 .000000+00-.43743D-08 .458860-01 .14684D-01 -.480510-01
23 .14203D-08-.54292D-09-.16599D-08 .40273D-01 - .28572D-02 -.182340-01
24 .47982D-08 .00000D+00-.50499D-08 .15502D+00 .49607D-01 -.554720-01
6.6.2 Example 6.2. A Spherical Cap
This example analyses a spherical cap for membrane stresses. Figure 6.5
shows a plan view of the analysis model. Again, the data is generated using a
computer program (MEMBRNDAT.FOR) with the following variables: RAD
= shell radius; THICK = shell thickness; PHI = shell central angle; NSEG =
number of sectors taken around the shell and NRING = number of levels into
which the shell is subdivided. Again, both the data program and the analysis
program must first be compiled. At run time they can be concatenated using
the DOS pipeline command MEMBRNDATIMEMBRN>MEMBRN.OUT.
The cap has a radius of 4.76 inches, a central angle of 10.9 degrees
and a thickness of 0.01576 inches. Young's modulus and Poisson's ratio are
10,000,000 psi and 0.3 respectively. The output for this example is listed
below.
Output file MEMBRN.OUT
NUMBER OF BARS 0
NUMBER OF NODES 25
NUMBER OF SUPPORTS 8
NUMBER OF CONSTRAINTS 0
NUMBER OF ITERATIONS 1
NUMBER OF FINITE ELEMENTS 40
COOROINATES LOADS
X Y Z PX PY PZ
1 .OOOOOD+OO .000000+00 .47600D+01 .OOOOOD+OO .OOOOOD+OO -.10000D+01
2 .301650+00 .000000+00 .47504D+01 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
3 .21330D+00 .21330D+00 .475040+01 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO
4 .38236D-06 .30165D+00 .475040+01 .OOOOOD+OO .000000+00 .OOOOOD+OO
5-.21330D+00 .21330D+00 .47504D+01 .000000+00 .OOOOOD+OO .OOOOOD+OO
19
24
Figure 6.5 The spherical membrane shell .
6-.301650+00 .764730-06 .475040+01 . 000000+00 .000000+00 .000000+00

7-.213300+00-.213300+00 .475040+01 .000000+00 .000000+00 .000000+00
8-.114710-05-.301650+00 .475040+01 .000000+00 .000000+00 .000000+00
9 .213300+00-.213300+00 .475040+01 .000000+00 .000000+00 .000000+00
10 .556250+00 .230410+00 .472180+01 .000000+00 .000000+00 .000000+00
11 .230410+00 .556250+00 .472180+01 .000000+00 .000000+00 .000000+00
12-.230410+00 .556250+00 .472180+01 .000000+00 .000000+00 .000000+00
13-.556250+00 .230410+00 .472180+01 .000000+00 .000000+00 .000000+00
14-.556250+00-.230400+00 .472180+01 .000000+00 .000000+00 .000000+00
15-.230410+00-.556250+00 .472180+01 .000000+00 .000000+00 .000000+00
16 .230400+00-.556250+00 .472180+01 .000000+00 .000000+00 .000000+00
17 .556250+00-.230410+00 .472180+01 .000000+00 .000000+00 .000000+00
18 .900090+00 .000000+00 .467410+01 .000000+00 .000000+00 .000000+00
19 .636460+00 .636460+00 .467410+01 .000000+00 .000000+00 .000000+00
20 .114100-05 .900090+00 .467410+01 .000000+00 .000000+00 .000000+00
21-.636460+00 .636460+00 .467410+01 .000000+00 .000000+00 .000000+00
22-.900090+00 .228190-05 .467410+01 .000000+00 .000000+00 .000000+00
23-.636460+00-.636460+00 .467410+01 .000000+00 .000000+00 .000000+00
24-.342290-05-.900090+00 .467410+01 .000000+00 .000000+00 .000000+00
25 .636460+00-.636470+00 .467410+01 .000000+00 .000000+00 .000000+00
FINITE ELEMENT OATA

NO. NOOES THICKNESS
I J M
1 1 2 3 .1576E-Ol
2 1 3 4 .1576E-Ol
3 1 4 5 .1576E-Ol
4 1 5 6 .1576E-Ol
5 1 6 7 .1576E-Ol
6 1 7 8 .1576E-Ol
7 1 8 9 .1576E-Ol
DISPLACEMENTS
x Y z
1 -.28711713D-10 .17348089D-10 -.23013107D-02
2 .26981282D-04 -.58449789D-11 .12403418D-03
3 .19078607D-04 .19078568D-04 .12403315D-03
4 .28488593D-10 .26981224D-04 .12403335D-03
5 -.19078584D-04 .19078653D-04 .12403344D-03
6 -.26981239D-04 .74573549D-10 .12403344D-03
7 -.19078664D-04 -.19078536D-04 .12403328D-03
8 -.85858794D-10 -.26981189D-04 .12403299D-03
9 .19078543D-04 -.19078664D-04 .12403294D-03
10 .82616315D-05 .34220653D-05 .34118350D-04
11 .34220896D-05 .82616316D-05 .34118449D-04
12 -.34220756D-05 .82616545D-05 .34118522D-04
13 -.82616462D-05 .34221082D-05 .34118544D-04
14 -.82616561D-05 -.34220602D-05 .34118502D-04
15 -.34221060D-05 -.82616211D-05 .34118430D-04
16 .34220644D-05 -.82616785D-05 .34118600D-04
17 .82616260D-05 -.34221213D-05 .34118451D-04
MEMBER ANALYSIS
STRAINS STRESSES
EX EY GA SX SY TAU
1-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42271E+02 -.81153E+03
2-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42271E+02 -.81153E+03
3-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42271E+02 -.81153E+03
4-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42271E+02 -.81153E+03
5-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42271E+02 -.81153E+03
6-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42270E+02 -.81153E+03
7-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42270E+02 -.81153E+03
NONLINEAR ANALYSIS.OF MEMBRANES 139
8-.16540E-03 .45773E-04-.21100E-03 -.16667E+04 -.42277E+02 -.81153E+03
9 .14852E-04-.33342E-04-.30056E-10 .53294E+02 -.31744E+03 -.11560E-03
10-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
11 .14852E-04-.33342E-04-.30224E-11 .53294E+02 -.31744E+03 - . 11625E-04
12-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
13 .14852E-04-.33342E-04 .18448E-11 .53294E+02 -.31744E+03 .70953E-05
14-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
15 .14852E-04-.33342E-04-.56942E-11 .53294E+02 -.31744E+03 -.21901E-04
16-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
17 .14852E-04-.33342E-04 .13423E-11 .53294E+02 -.31744E+03 .51628E-05
18-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
19 .14852E-04-.33342E-04-.28137E-11 .53294E+02 -.31744E+03 -.10822E-04
20-.11794E-04 .76648E-04 .71993E-04 .12309E+03 .80341E+03 .27689E+03
21 .14852E-04-.33342E-04 .16695E-09 .53293E+02 -.31744E+03 .64210E-03
22-.11794E-04 .76648E-04 .71992E-04 .12309E+03 .80341E+03 .27689E+03
23 .14852E-04-.33342E-04 .22670E-09 .53294E+02 -.31743E+03 .87192E-03
24-.11794E-04 .76648E-04 .71992E-04 .12308E+03 .80341E+03 .27689E+03
25 .OOOOOE+OO-.77662E-05-.44590E-10 -.25603E+02 -.85343E+02 -.17150E-03
26-.24579E-05 .72260E-05 .22979E-04 -.31876E+01 .71304E+02 .88383E+02
27 .OOOOOE+OO-.77662E-05-.24496E-11 -.25603E+02 -.85343E+02 -.94214E-05
28-.24580E-05 .72261E-05 .22979E-04 -.31882E+01 .71305E+02 .88382E+02
29 .OOOOOE+OO-.77662E-05 .61971E-11 -.25603E+02 -.85343E+02 .23835E-04
30-.24580E-05 .72262E-05 .22979E-04 -.31883E+01 .71305E+02 .88382E+02
31 .OOOOOE+OO-.77662E-05-.33468E-11 -.25603E+02 -.85343E+02 - .12872E-04
32-.24580E-05 .72262E-05 .22979E-04 -.31883E+01 .71305E+02 .88382E+02
33 .OOOOOE+OO-.77662E-05-.84355E-11 -.25603E+02 -.85343E+02 -.32444E-04
34-.24580E-05 .72261E-05 .22979E-04 -.31883E+01 .71305E+02 .88382E+02
35 .OOOOOE+OO-.77662E-05 .12291E-10 -.25603E+02 -.85343E+02 .47275E-04
36-.24580E-05 .72262E-05 .22979E-04 -.31882E+01 .71305E+02 .88382E+02
37 .OOOOOE+OO-.77662E-05 .19779E-10 -.25603E+02 -.85343E+02 .76073E-04
38-.24580E-05 .72262E-05 .22979E-04 -.31883E+01 .71305E+02 .88382E+02
39 .OOOOOE+OO-.77661E-05-.88205E-10 -.25603E+02 -.85342E+02 -.33925E-03
40-.24580E-05 .72262E-05 .22979E-04 -.31884E+01 .71305E+02 .88382E+02
6.6.3 Example 6.3. A Flat Stretched Membrane
The transversely loaded flat cable-net of Example 3.2 is transformed to a flat

stretched membrane ending a sequence of examples that started with Biot's
prestressed 2-bar truss. The symmetric analysis model of Figure 3.3 is turned
to a symmetric finite element analysis model. (See Figure 6.6.) The thickness
of the membrane is taken as the cross-sectional area of a single cable divided
by one span i.e. 0.25 = 0.004167 in. Similarly a uniform stress of
5x 12
20,000
- - - -7- - - , pSI" IS app I'Ie d aII around the membrane. Young's
0.25 = 80000
5x12 5x12
modulus and Poisson's ratio are taken as 30,000,000 psi and 0.3 respectively.
Input file MEMNL.DAT is invoked by MEMNL.FOR and the results are
partially listed in MEMNL.OUT. Actually only the echo of the input file and
the computations of the last iteration of the last load step are listed. It is
important to mention here that LINEAR INCREMENTAL ANALYSIS
lists the strains and stresses that are computed from the usual plane nodal
displacements/plane strain relations of the plane stress triangular element i.e.
nodal displacements are transformed to local coordinates and the z-

coordinate discarded. INCREMENTAL NONLINEAR ANALYSIS,
however, lists strains and stresses which are computed after rigid body
rotations have been removed. (See problem 6.2.) UPDATED PRESTRESS
simply adds the stresses from the current increment to those of the previous
increment.
It might be of interest to observe that the central deflection here 6.62
inches whereas the central deflection of the cable net is 7.11 inches. But then
again this is to be expected because of the shear in the membrane that tends
to stiffen it.
Output file MEMNL.OUT

NUMBER OF BARS o
NUMBER OF NODES 25
NUMBER OF CONSTRAINTS o
NUMBER OF LOAD STEPS 5
COORDINATES LOADS
x Y Z PX PY PZ
1-.60000E+02 .60000E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
Figure 6.6 The transversely loaded stretched membrane.

4-.60000E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
5 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO -.10000E+05
7-.60000E+02-.60000E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
8 .00000E+00-.60000E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
FINITE ELEMENT DATA

NO. NODES THICKNESS PRESTRESS
I J M SIGX SIGY TAU
1 15 1 10 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
2 11 10 1 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
3 1 2 11 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
4 12 11 2 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
5 13 12 2 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
6 2 3 13 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
7 14 13 3 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
8 3 23 14 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
9 16 4 15 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
10 1 15 4 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
11 4 5 1 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
12 2 1 5 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
13 3 2 5 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
14 5 6 3 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
15 23 3 6 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
16 6 24 23 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
17 4 16 17 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
18 17 7 4 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
19 5 4 7 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
20 7 8 5 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
21 8 9 5 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
22 6 5 9 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
23 9 25 6 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
24 24 6 25 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
25 7 17 18 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
26 18 19 7 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
27 19 20 8 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
28 8 7 19 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
29 20 21 8 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
30 9 8 21 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
31 21 22 9 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
32 25 9 22 .41600E-02 .80000E+05 .80000E+05 .OOOOOE+OO
LOAD STEP 5
COORDINATES LOADS
X Y Z PX PY PZ
1-.59985E+02 .59985E+02-.14309E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
2-.25607E-15 .59983E+02-.26046E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
3 .59985E+02 .59985E+02-.14309E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
4-.59983E+02-.82747E-15-.26045E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
5 .39495E-16 .36615E-16-.66418E+Ol .OOOOOE+OO .OOOOOE+OO -.10000E+05
6 .59983E+02 .82668E-15-.26045E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
7-.59985E+02-.59985E+02-.14309E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
8 .27357E-15-.59983E+02-.26046E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
9 .59985E+02-.59985E+02-.14309E+Ol .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
ERROR = .12389175D-02
DISPLACEMENTS
X Y Z
1 .52016588D-09 - .10951429D-08 -.46234717D-09
2 .11365291D-15 -.21571404D-08 -.16859462D-09
3 -.52016577D-09 -.10951434D-08 -.46234508D-09
4 -.29205406D-09 .72583523D-15 .28383451D-09
5 -.12545721D-16 -.14985675D-16 .45354019D-09
6 .29205406D-09 -.72999134D-15 .28383420D-09
7 .52016577D-09 .10951434D-08 -.46234496D-09
8 - .12116067D-15 .21571404D-08 -.16859486D-09
9 -.52016589D-09 .10951429D-08 -.46234720D-09
MEMBER ANALYSIS
MEMBRANE FINITE ELEMENTS
ELEMENT 1
NODES 15 1 10
LINEAR INCREMENTAL ANALYSIS
EX EY GA SX SY TAU
1 .88504E-ll-.44298E-14-.18238E-I0 .29173E-03 .87385E-04 -.21044E-03
INCREMENTAL NONLINEAR ANALYSIS
EX EY GA SX SY TAU
1 .88505E-ll-.43817E-14-.18239E-I0 .29173E-03 .87388E-04 -.21044E-03
UPDATED PRESTRESS
SX SY TAU
.9737767E+05 .8521242E+05 -.2801543E+04
ELEMENT 2
NODES 11 10 1
EX EY GA SX SY TAU
2 .OOOOOE+OO .18421E-I0-.86649E-ll .18219E-03 .60729E-03 -.99979E-04
EX EY GA SX SY TAU
2 .OOOOOE+OO .18421E-I0-.86649E-ll .18219E-03 .60729E-03 -.99979E-04
UPDATED PRESTRESS
SX SY TAU
.8521326E+05 .9737754E+05 -.2802885E+04
ELEMENT 3
NODES 1 2 11
EX EY GA SX SY TAU
3-.87633E-ll .18403E-I0-.26405E-I0 -.10689E-03 .52004E-03 -.30467E-03
EX EY GA SX SY TAU
3-.87633E-ll .18403E-I0-.26405E-I0 -.10689E-03 .52004E-03 -.30467E-03
UPDATED PRESTRESS
SX SY TAU
.8351018E+05 .9685909E+05 -.8657109E+04
ELEMENT 4
NODES 12 11 2
EX EY GA SX SY TAU
4 .OOOOOE+OO .35996E-I0-.18919E-17 .35601E-03 .11867E-02 -.21830E-I0
EX EY GA SX SY TAU
UPDATED PRESTRESS
SX SY TAU
.9211395E+05 .1203798E+06 .2734976E-I0
ELEMENT 5
NODES 13 12 2
EX EY GA SX SY TAU
EX EY GA SX SY TAU
5 .OOOOOE+OO .35996E-I0 .OOOOOE+OO .35601E-03 .11867E-02 .OOOOOE+OO
UPDATED PRESTRESS
SX SY TAU
.9211395E+05 .1203798E+06 .OOOOOOOE+OO
ELEMENT 6
NODES 2 3 13
EX EY GA SX SY TAU
6-.87633E-ll .18403E-I0 .26405E-I0 -.10689E-03 .52004E-03 .30467E-03
EX EY GA SX SY TAU
6-.87633E-ll .18403E-I0 .26405E-I0 -.10689E-03 .52004E-03 .30467E-03
UPDATED PRESTRESS
SX SY TAU
.8351018E+05 .9685909E+05 .8657109E+04
ELEMENT 7
NODES 14 13 3
EX EY GA SX SY TAU
7 .OOOOOE+OO .18421E-I0 .86649E-ll .18219E-03 .60729E-03 .99979E-04
EX EY GA SX SY TAU
7 .OOOOOE+OO .18421E-I0 .86649E-ll .18219E-03 .60729E-03 .99979E-04
UPDATED PRESTRESS
SX SY TAU
. 8521326E+05 .9737754E+05 .2802885E+04
ELEMENT 8
NODES 3 23 14
EX EY GA SX SY TAU
8 .88504E-ll-.44298E-14 .18238E-I0 .29173E-03 .87385E-04 .21044E-03
EX EY GA SX SY TAU
8 .88505E-ll-.43817E-14 .18238E-I0 .29173E-03 .87388E-04 .21044E-03
UPDATED PRESTRESS
SX SY TAU
.9737767E+05 .8521242E+05 .2801543E+04
ELEMENT 9
NODES 16 4 15
EX EY GA SX SY TAU
9-.50619E-ll .16643E-33 .12082E-16 -.16688E-03 -.50063E-04 .13941E-09
EX EY GA SX SY TAU
9-.50619E-ll .OOOOOE+OO .12082E-16 -.16687E-03 -.50062E-04 .13941E-09
UPDATED PRESTRESS
SX SY TAU
.1203582E+06 .9210745E+05 -.1955792E-I0
ELEMENT 10
NODES 1 15 4
EX EY GA SX SY TAU
10 .88504E-l1-.18497E-I0-.45423E-l1 .10883E-03 -.52226E-03 -.52411E-04
EX EY GA SX SY TAU
10 .88505E-ll-.18497E-I0-.45423E-ll .10884E-03 -.52226E-03 -.52411E-04
UPDATED PRESTRESS
SX SY TAU
.9686449E+05 .8350182E+05 -.8644788E+04
ELEMENT 11
NODES 4 5 1
EX EY GA SX SY TAU
11 .46574E-ll-.18474E-I0 .14410E-I0 -.29171E-04 -.56298E-03 .16627E-03
EX EY GA SX SY TAU
11 .46575E-l1-.18474E-I0 .14410E-I0 -.29168E-04 -.56298E-03 .16627E-03
UPDATED PRESTRESS
SX SY TAU
.1446910E+06 .9783068E+05 -.1561558E+05
ELEMENT 12
NODES 2 1 5
EX EY GA SX SY TAU
12-.87633E-l1-.36519E-I0-.17152E-I0 -.65007E-03 -.12906E-02 -.19791E-03
EX EY GA SX SY TAU
12-.87633E-ll-.36519E-I0-.17152E-I0 -.65008E-03 -.12906E-02 -.19791E-03
UPDATED PRESTRESS
SX SY TAU
.9784664E+05 .1446473E+06 -.1565376E+05
ELEMENT 13
NODES 3 2 5
EX EY GA SX SY TAU
13-.87633E-l1-.36519E-I0 .17152E-I0 -.65007E-03 -.12906E-02 .19791E-03

EX EY GA SX SY TAU
13-.87633E-ll-.36519E-I0 .17152E-I0 -.65008E-03 -.12906E-02 .19791E-03
UPDATED PRESTRESS
SX SY TAU
.9784664E+05 .1446473E+06 .1565376E+05
ELEMENT 14
NODES 5 6 3
EX EY GA SX SY TAU
14 .46574E-ll-.18474E-I0-.14410E-I0 -.29171E-04 -.56298E-03 -.16627E-03
EX EY GA SX SY TAU
14 .46575E-ll-.18474E-I0-.14410E-I0 -.29168E-04 -.56298E-03 -.16627E-03
UPDATED PRESTRESS
SX SY TAU
.1446910E+06 .9783068E+05 .1561558E+05
ELEMENT 15
NODES 23 3 6
EX EY GA SX SY TAU
15 .88504E-ll-.18497E-I0 .45423E-ll .10883E-03 -.52226E-03 .52411E-04
EX EY GA SX SY TAU
15 .88505E-ll-.18497E-I0 .45423E-l1 .10884E-03 -.52226E-03 .52411E-04
UPDATED PRESTRESS
SX SY TAU
.9686449E+05 .8350182E+05 .8644788E+04
ELEMENT 16
NODES 6 24 23
EX EY GA SX SY TAU
16-.50619E-l1 .14390E-32 .12152E-16 -.16688E-03 -.50063E-04 .14021E-09
EX EY GA SX SY TAU
16-.50619E-l1 .OOOOOE+OO-.59973E-27 -.16687E-03 -.50062E-04 -.69199E-20
UPDATED PRESTRESS
SX SY TAU
.1203582E+06 .9210745E+05 - .1366110E-08
ELEMENT 17
NODES 4 16 17
EX EY GA SX SY TAU
17-.50619E-ll .14309E-32 .12082E-16 -.16688E-03 -.50063E-04 .13941E-09
EX EY GA SX SY TAU
17-.50619E-l1 .OOOOOE+OO-.59973E-27 -.16687E-03 -.50062E-04 -.69199E-20
UPDATED PRESTRESS
SX SY TAU
.1203582E+06 .9210745E+05 - . 1366110E-08
ELEMENT 18
NODES 17 7 4
EX EY GA SX SY TAU
18 .88504E-ll-.18497E-I0 .45423E-ll .10883E-03 -.52226E-03 .52411E-04
EX EY GA SX SY TAU
18 .88505E-ll-.18497E-I0 .45423E-ll .10884E-03 -.52226E-03 .52411E-04
UPDATED PRESTRESS
SX SY TAU
.9686449E+05 .8350182E+05 .8644788E+04
ELEMENT 19
NODES 5 4 7
EX EY GA SX SY TAU
19 .46574E-11-.18474E-10-.14410E-10 -.29171E-04 -.56298E-03 -.16627E-03
EX EY GA SX SY TAU
19 .46575E-11-.18474E-10-.14410E-10 -.29168E-04 -.56298E-03 -.16627E-03
UPDATED PRESTRESS
SX SY TAU
.1446910E+06 .9783068E+05 .1561558E+05
ELEMENT 20
NODES 7 8 5
EX EY GA SX SY TAU
20-.87633E-11-.36519E-10 .17152E-10 -.65007E-03 -.12906E-02 .19791E-03
EX EY GA SX SY TAU
20-.87633E-11-.36519E-10 .17152E-10 -.65008E-03 -.12906E-02 .19791E-03
UPDATED PRESTRESS
SX SY TAU
.9784664E+05 .1446473E+06 .1565376E+05
ELEMENT 21
NODES 8 9 5
EX EY GA SX SY TAU
21-.87633E-11-.36519E-10-.17152E-10 -.65007E-03 -.12906E-02 -.19791E-03
EX EY GA SX SY TAU
21-.87633E-11-.36519E-10-.17152E-10 -.65008E-03 -.12906E-02 -.19791E-03
UPDATED PRESTRESS
SX SY TAU
.9784664E+05 .1446473E+06 -.1565376E+05
ELEMENT 22
NODES 6 5 9
EX EY GA SX SY TAU
22 .46574E-11-.18474E-10 .14410E-10 -.29171E-04 -.56298E-03 .16627E-03
EX EY GA SX SY TAU
22 .46575E-11-.18474E-10 .14410E-10 -.29168E-04 -.56298E-03 .16627E-03
UPDATED PRESTRESS
SX SY TAU
.1446910E+06 .9783068E+05 -.1561558E+05
ELEMENT 23
NODES 9 25 6
EX EY GA SX SY TAU
23 .88504E-11-.18497E-10-.45423E-11 .10883E-03 -.52226E-03 -.52411E-04
EX EY GA SX SY TAU
23 .88505E-11-.18497E-10-.45423E-11 .10884E-03 -.52226E-03 -.52411E-04
UPDATED PRESTRESS
SX SY TAU
.9686449E+05 .8350182E+05 -.8644788E+04
ELEMENT 24
NODES 24 6 25
EX EY GA SX SY TAU
24-.50619E-11 .16722E-33 .12152E-16 -.16688E-03 -.50063E-04 .14021E-09
EX EY GA SX SY TAU
24-.50619E-11 .OOOOOE+OO .12152E-16 -.16687E-03 -.50062E-04 .14021E-09
UPDATED PRESTRESS
SX SY TAU
.1203582E+06 .9210745E+05 -.1860152E-10
ELEMENT 25
NODES 7 17 18
EX EY GA SX SY TAU
25 .88504E-11-.44298E-14 .18238E-10 .29173E-03 .87385E-04 .21044E-03
EX EY GA SX SY TAU
25 .88505E-11-.43817E-14 .18238E-10 .29173E-03 .87388E-04 .21044E-03
UPDATED PRESTRESS
SX SY TAU
.9737767E+05 .8521242E+05 .2801543E+04
ELEMENT 26
NODES 18 19 7
EX EY GA SX SY TAU
26 .OOOOOE+OO .18421E-10 .86649E-11 .18219E-03 .60729E-03 .99979E-04
EX EY GA SX SY TAU
26 .OOOOOE+OO .18421E-10 .86649E-11 .18219E-03 .60729E-03 .99979E-04
UPDATED PRESTRESS
SX SY TAU
.8521326E+05 .9737754E+05 .2802885E+04
ELEMENT 27
NODES 19 20 8
EX EY GA SX SY TAU
27 .OOOOOE+OO .35996E-10-.20169E-17 .35601E-03 .11867E-02 -.23272E-10
EX EY GA SX SY TAU
27 .OOOOOE+OO .35996E-10 .OOOOOE+OO .35601E-03 .11867E-02 .OOOOOE+OO
UPDATED PRESTRESS
SX SY TAU
.9211395E+05 .1203798E+06 .OOOOOOOE+OO
ELEMENT 28
NODES 8 7 19
EX EY GA SX SY TAU
28-.87633E-11 .18403E-10 .26405E-10 -.10689E-03 .52004E-03 .30467E-03
EX EY GA SX SY TAU
28-.87633E-11 .18403E-IO .26405E-10 -.10689E-03 .52004E-03 .30467E-03
UPDATED PRESTRESS
SX SY TAU
.8351018E+05 . 9685909E+05 .8657109E+04
ELEMENT 29
NODES 20 21 8
EX EY GA SX SY TAU
29 .OOOOOE+OO .35996E-10-.20169E-17 .35601E-03 .11867E-02 -.23272E-10
EX EY GA SX SY TAU
29 .OOOOOE+OO .35996E-10-.20169E-17 .35601E-03 .11867E-02 -.23272E-10
UPDATED PRESTRESS
SX SY TAU
.9211395E+05 .1203798E+06 .2927084E-10
ELEMENT 30
NODES 9 8 21
EX EY GA SX SY TAU
30-.87633E-11 .18403E-10-.26405E-10 -.10689E-03 .52004E-03 -.30467E-03
EX EY GA SX SY TAU
30-.87633E-11 .18403E-10-.26405E-10 -.10689E-03 .52004E-03 -.30467E-03
UPDATED PRESTRESS
SX SY TAU
.8351018E+05 .9685909E+05 -.8657109E+04
ELEMENT 31
NODES 21 22 9
EX EY GA SX SY TAU
31 .OOOOOE+OO .18421E-10-.86649E-11 .18219E-03 .60729E-03 -.99979E-04
EX EY GA SX SY TAU
31 .OOOOOE+OO .18421E-10-.86649E-11 .18219E-03 .60729E-03 -.99979E-04
UPDATED PRESTRESS
SX SY TAU
.8521326E+05 .9737754E+05 -.2802885E+04
ELEMENT 32
NODES 25 9 22
EX EY GA SX SY TAU
32 .88504E-11-.44298E-14-.18238E-10 .29173E-03 .87385E-04 -.21044E-03
EX EY GA SX SY TAU
32 .88505E-11-.43817E-14-.18239E-10 .29173E-03 .87388E-04 -.21044E-03
UPDATED PRESTRESS
SX SY TAU
. 9737767E+05 .8521242E+05 -.2801543E+04
6.7 Problems
1. For the plane stress cantilever Example 6.1 study the effect of changing
the grid size (element density) in an attempt to obtain reasonable deep beam
stresses. (Timoshenko and Goodier, 1951, pp. 361 ff)
2. Discuss the removal of rigid body rotations for the cases of a truss bar and
a plane stress triangular element. (Hint: for the truss bar see figures 1.13 and
1.16)
3. An aluminum (E = lOx 10 6 psi, v=0.3) prestressed sheet is shown in

Figure 6.7. The sheet is L=4 inches square, 0.1 inches thick and there is a
uniform initial prestress of 1000 psi. Using the method of load incrementing
find the buckling load.(Ans: between 430 lbs and 440 lbs.)
L
•• ••
finite
elements~+--t7"'- wrinkles
~--~~--~~--~~--~~
all boundary nodes are fixed
Figure 6.7 Finite element grid for the prestressed sheet.
4. For fabric structures applications, rather than buckling, fabric "wrinkling"

would be the appropriate phenomenon. Fabric wrinkling can be modeled by
monitoring the principal stresses as the structure is loaded. When the initial
prestress which is of course tension is taken to zero in terms of either
principal stress during the load application, the isotropic finite element
stiffness can be changed to an anisotropic one with only stiffness in the
direction of the principal tensile stress. (This is the "tension field" concept
used in plate girders.) Modify MEMNL.FOR to include principal stresses
and monitoring of the loss of prestress and find for the prestressed sheet of
Figure 6.7, at what load do wrinkles occur and their angle of inclination.
(Ans: P=240 Ibs at an angle of 27° to the horizontal. Up to a load of 600 lbs
wrinkles do not form in the other triangular facets.)
5. Model the 24-bar truss of Example 3.6 using nonlinear membrane plane
stress elements and find the buckling load. Comment on the expected and
attained closeness of the membrane results vs. truss results.
CHAPTER 7
CABLE NETS AND FABRIC STRUCTURES
7.1 Introduction
This chapter is something of a digression from the other chapters of this text
which are concerned strictly with the analysis of specific types of structures.
Here the concern is with how cable nets and fabric structures are designed.
This in tum raises the question of how shape is to be determined which in
tum returns to the question of the application of some methods of nonlinear
analysis. But no new analysis topics are introduced in this chapter.
In the material which follows there will not always be a clear
distinction made between cable nets and fabric structures. Obviously, they
are physically quite similar but the matter goes deeper than this. The most
simple model of a piece of fabric is one with fibers running in two directions
(a cable net). Arguments against this model usually revolve around the fact
that a cable net can not transmit shear. The counter argument is of course
that fabric is not good at transmitting shear anyway and that the cable net
model of fabric is physically intuitive and simple. The simple cable net also
does not model to warp/woof load interchange found in a piece of fabric
when originally straight fibers become deformed under biaxial load ( Stubbs
and Fluss, 1979).
The design of cable nets and fabric structures can be described in
terms of three events: shape finding, analysis, and patterning. In the process
of shape finding, the designer specifies a set of parameters and then
computes other parameters finally resolving the details of the shape of the
structure. Under analysis, loads are applied to a structure whose shape is
known and the response to these loads computed. Patterning is concerned
with how a curved surface is to be formed from rolls of fabric.
This chapter is first concerned with the process of finding the details
of the shape of a structure and later with how a structure is to be patterned. It
is argued that methods of analysis have already been discussed in earlier
chapters. Finally, only occasional references will be provided below as part
of the text. The more ambitious reader may wish to consult the excellent
book of Leonard (1988) on tension structures.
7.2 Basic Methods of Shape Finding
The process of shape finding can be thought of roughly in terms of stretching

fabric over a frame of arbitrary shape. (For example, in the skylight problem
of Figure 7.1 , the geometry is fixed along both crossed arches and the base.)
Clearly the fabric must follow the frame at the boundaries and certain
tensions can be specified on these boundaries. But just as clearly, the
locations of the fabric points within the frame must be determined from the
equations of equilibrium and in some cases the material parameters. Finding
the locations of these internal points is the process of shape finding.
In the early days (Frei Otto, 1973) and in the absence of the
computer, physical models were commonly used in the design of fabric
structures and cable nets. It is now conventional wisdom that small scale
models are not sufficiently accurate either for the prediction of forces or the
patterning of the fabric. In the following some computer based approaches
which have been used to find shape are discussed.
Figure 7.1 Skylight example: crossed circular arches on structural frame.
7.2.1 Deformed Shape
If you need to find a shape which is in equilibrium, this can be done by

applying loads to, for example, a stretched elastic sheet and then using the
deformed sheet or a scaled version of it as the shape. There is nothing wrong
with doing so but care is required since loads applied to a sheet may
introduce stress concentrations which may not be desired in the structure
under design. The basic reference to this method is Argyris, et al. (1964).
Any computer program for nonlinear structural analysis can be used
to achieve shape in this manner but it does not appear common to do so.
Pertinent to this is the fact that the fabrics now commonly used in permanent
structures can not tolerate large strains without tearing.
CABLE NETS AND FABRIC STRUCTURES 153
7.2.2 Force Density Method
It has been noted by Schek (1974) that if the ratio of the bar force to its
length is held constant in a cable net, the associated geometry can be found
by solving a system of linear equations. (A similar statement can be made for
a finite element fabric model.) This approach is frequently used in the design
of fabric structures.
The force density method is based on the fact that the force on the
end of a truss bar can be represented by the product of the bar force and a
unit vector in the direction of the bar as shown in Figure 7.2. Here n i is the
unit vector of member i, Fi is the bar force of member i, and Li is the
length of member i. The components of the force vector can be written as
(Fit = 2I
(XA -xc)
(Fj) =Fi(yA-yC ) (7.1)

Y Li
(Fi)z = ;i (ZA -Zc)

I
Clearly, if the "force density", F/Li the ratio of the bar force to its
bar length is some known constant, the force at the end of the bar becomes a
linear function of the coordinates X A' YA , ZA andXc' Yc , Zc of the ends of
the bar. What this means is that when the equilibrium equations are written,
they are linear in the node coordinates. Thus problems can be formulated in
which the node coordinates (shape) are easily found by solving a system of
linear equations.
typical bar
Figure 7.2 The force density method for a bar element.

z·
y'
IE
e
Figure 7.3 Force density method for a membrane element.
One of the shortcomings, of course, with this method is that it is

usually preferable to specify the member force Fi rather than the force
density F/L i . In fact, Li is not really known until the node coordinates
have been computed by some shape finding algorithm.
As Figure 7.3 indicates, similar arguments can be made for a simple
membrane finite element. Roughly, given a simple finite element such as
Zienkiewicz's plane stress element (Zienkiewicz, 1977) and an existing state
of stress, nodal forces can be computed. (They depend upon both the state of
stress and the geometry of the finite element.) These nodal forces can be
expressed as linear combinations of the unit vectors n and m (Figure 7.3)
which describe the slopes of two sides of the finite element. As was done for
the bar, these unit vectors can be factored into a term linear in the node
coordinates and whatever remains. Obviously, when the coefficients of these
linear terms are specified as in the case of the bar, the equilibrium equations
become linear in the coordinates of the elements.
7.2.3. Grid Method
Siev and Eidelman (1964) showed that if equilibrium is satisfied over a grid
in the horizontal plane (see Figure 7.4), vertical equilibrium can be used to
compute the elevation (shape) at the grid points. This is probably the most
simple way to find shape and will be discussed at some length below.
Figure 7.4 The grid method.
7.2.4 Smoothing
Given a shape, a computer program for nonlinear analysis can be used

(Haber and Abel, 1982) to smooth out the forces or stresses sometimes
resulting in a state of uniform stress. How this can be done is described
below along with variations which allow shape to be found directly using
computer programs for nonlinear analysis.
In the material which follows, a path will be described from the grid
method, through smoothing, and on to a procedure which allows shape to be
determined using computer programs for nonlinear analysis which are of
course central to this text.
7.3 The Grid Method
The grid method is probably the most simple way to find shape. It requires,
first of all, a grid in the horizontal plane but any numerical method for
finding shape requires some kind of a grid of points. Member forces must be
assumed or determined so that the cable net or finite element system is in
equilibrium in the horizontal plane but this too can be a trivial step. That is
clearly the case in the simple example of Figure 7.6 below where it is
assumed that the horizontal cable force components are all 1,000 lbs.
(Constant force components over a rectangular grid are clearly in
equilibrium in the horizontal plane.) If only the rise of the circular arches has
been specified, it is also required to compute the elevations of the points of
attachment along these arches as is done below. The final step is to write the
equations of equilibrium in the vertical direction and then solve a system of

linear equations for the unknown elevations at these points. Following are
two examples of shape finding using the grid method. The first, which has
the form of part of a skylight, is solved by hand whereas the second skylight
example is solved using the computer program LAYOUT.FOR.
7.3.1 Example 7.1 A Piece of a Cross Arched Skylight
This example is a piece (type B in Figure 7.5) of a cross arched skylight of

dimensions 20ft x 40ft. Since the base of this skylight is not square, it would
be necessary to find the shape of another piece (type A in Figure 7.5) to
complete its design.
The piece of skylight is taken as a 9-member cable net laid over a
right angled triangular structural base having two edges of 10ft and 20 ft as
shown in Figure 7.6. It is supported at nodes 4,5,6 by the third edge which is
a circular arch having a rise of 5 ft. With horizontal force components of
1,000 lbs in each of the cables it is required to find the vertical coordinates
of nodes 1,2,3 so that equilibrium is maintained.
The elevations of nodes 5 and 6 are first established from the arch
geometry. The radius of the arch is obtained from:
e = 2 x arctan( 22.3607
5 ) = 25.2087°
R= 22.3607 = 52.500 ft.

sin 25.2087°
Now using the equation for a circle, x 2 + y2 = R2, the z-coordinates of

nodes 5 and 6 are obtained as
20 ft 20 ft
••
plan
Figure 7.5 Rectangular skylight.

I. 22.3607 ft .,. 22.3607 ft .,.
7.453
xJi
arch geometry
5ft
.(
20 ft ...
cable net
Figure 7.6 A simple cable net.
Z6 = ~ R2 -14.907 2 - Rcose = 2.8391 ft
The next step is to write vertical joint equilibrium for each free joint by
adding the contribution of each member. (Remember that the grid method
starts with the satisfaction of equilibrium in the horizontal plane.) The
simple relation of Figure 7.4 is used for that purpose, i.e.
(7.2)
Here H=I,OOOlbs and LH has only two values which are ~Y = 3.333 ft and
A¥ = 6.667 ft. Considering vertical equilibrium:
node 1: H(5 - zl + 2 x 4.468 - zl _ zl - z2 ) =0

L1X L1Y L1X
node 2: H(_ZI0--_Z"'-2 + 2 x z3 - z2 -Z2)

- -0
_
L1X L1Y L1X
node 3: H(2.839 - z3 + 4.468 - z3 _ z3 - z2 _~) = 0

L1Y L1X L1Y L1X
Observing that A¥ / ~Y = 2 and canceling H the above equations are

rewritten and solved as
-6z 1 +z2 +22.872=0

zl - 6z 2 + 4z3 = 0
2z2 - 6z 3 + 10.146 = 0
Substituting z2 from the first equation and z3 from the third equation into
the second equation yields
-27 zl + 113.5 = 0 =::} zl = 4.203 ft and z2 = 2.350 ft; z3 = 2.474 ft. This
=::}
completes the example since an initial shape in equilibrium has been found.
7.3.2 Example 7.2 A Squared Base Skylight
This example, which is described by Figure 7.7, is in fact the skylight of

Figure 7.1. This time the computer program LAYOUT.FOR will be used to
determine shape. Figure 7.8 shows a node map of 1/8 of this structure which
is all that it is required for design due to symmetry.
Following are the calculations which are necessary to determine the
elevations of the fixed points on the arch (nodes 5, 9, 12, 14 and 15) having
a specified rise of 30 inches. Here the base of the skylight is 172 inches
square and the cables are equally spaced in the horizontal plane. The
elevations Z5 and Z15 are given as 30 inches and 0 inches respectively. The
angle Band the radius of the arch are obtained as
B == tan -1 30 == 13.856°=> B == 27.71250

2 121.622
RsinB == 121.622=> R == 261.533 in
Ar-----------~
.E
'"'"
I. 86 in 1.1
86 in
M
C
plan
1
121.622 in 121.622 in
Ie lillIE '1
Zg
arch geometry
Figure 7.7 Skylight dimensions for example 7.2.

Using the equation of a circle x 2 + y2 = R2 the elevations for nodes 9, 12

and 14 become
Z9 = ~ R2 - 30.406 2 - Rcose = 28.2265 in
Z12 = ~ R2 - 60.811 2 - Rcose = 22.832 in
Z14 =~R2 -90.2165 2 -Rcose=13.578in
DOME which is listed below is the input file required by the

program LAYOUT.FOR. First the number of bars and nodes is specified.
Then a set of lines describes the bar ends and the horizontal component of
the force in each cable here taken to be 1000 lbs; the second part of the file
describes the node coordinates (which are taken to be zero when they are not
known), some symmetry conditions to be discussed later, and the member
connectivity. This connectivity is simply a list of the nodes which are
connected by cables to the node under consideration. Once the shape has
been determined the final cable forces can be computed as indicated in
Figure 7.9. Obviously, the final cable forces can not be equal since the cable
slopes are not equal. The output file FORT8 lists for each node the
coordinates and the FORCE BALANCE (from which reactions can be
computed) and for each member its prestress. Fictitious unbalanced
components indicated at points of symmetry are simply a product of the
manner by which LAYOUT.FOR deals with symmetry. See Section 7.8 for
more details regarding the running of the computer program LAYOUT.FOR.
Input file DOME

24 15
1000. 1 2
1000. 2 3
1000. 3 4
1000. 4 5
1000. 6 7
1000. 7 8
1000. 8 9
1000. 10 11
1000. 11 12
1000. 13 14
1000. 15 14
1000. 14 12
1000. 2 7
1000. 7 11
1000. 11 14
1000. 3 8
1000. B 12
1000. 4 9
4 0
4 19
3 9
3 16 17
2 6 9
2 13 14 11 15 4
5 8 10
21 22 10 23 13 24 15
Figure 7.8 Node map of 1/8 of the skylight structure.
1001. 1008.
1097. 1097. 1125 o.
1000. 1002 1004.
1104. 1113. 1139. 1183. o.
1000. 1000 1000. 1QOO.
Figure 7.9 Computed prestress on 1/8 skylight structure.
1000. 12 9
1000. 9 5
1000. 1 6
1000. 6 10
1000. 10 13
1000. 13 15
O. O. O. 1 2 6 6
O. 21. 5 O. 1 7 3 7
O. 43. O. 2 8 4 8
O. 64.5 3 9 5 9
O. 86. 30. 1 4 4
21.5 O. 1 1 7 10
21.5 21. 5 O. 2 6 8 11
21.5 43. O. 3 7 9 12
21. 5 64.5 28.227 1 4 8 4
43. O. O. 1 6 11 13
43. 21.5 O. 7 10 14 12
43. 43. 22.832 1 4 8 11
64.5 O. O. 1 10 14 15
64.5 21. 5 13.577 1 4 13
86. O. O. 1 4
Output file FORT8
COORDINATES FORCE BALANCE

X Y Z x y Z
1 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .20000E+04 .10000E+04 .46736E+03
2 .OOOOOE+OO .21500E+02 .10048E+02 .20000E+04 -.23820E-04 -.11462E-03
3 .OOOOOE+OO .43000E+02 . 19211E+02 .20000E+04 - . 11285E-03 -.80640E-04
4 .OOOOOE+OO .64500E+02 .26416E+02 .20000E+04 .46334E-04 -.16394E-03
5 .OOOOOE+OO .86000E+02 .30000E+02 .OOOOOE+OO -.10000E+04 -.16668E+03
6 .21500E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO .10000E+04 .48795E+03
7 .21500E+02 .21500E+02 .10491E+02-.14171E-05 .44862E-04 -.65020E-04
8 .21500E+02 .43000E+02 .20190E+02-.12679E-04 -.42546E-04 -.24605E-03
9 .21500E+02 . 64500E+02 .28227E+02-.10000E+04 -.10000E+04 -.45802E+03
11 .43000E+02 .21500E+02 .11725E+02-.25455E-04 .15764E-04 .32629E-04
12 .43000E+02 .43000E+02 .22832E+02-.10000E+04 -.10000E+04 -.63948E+03
14 .64500E+02 .21500E+02 . 13577E+02 .OOOOOE+OO -.10000E+04 -.63149E+03
MEMBER OUTPUT
+END -END FORCE
1 1 2 .11038236E+04
2 2 3 .10870308E+04
3 3 4 .10546581E+04
4 4 5 .10137963E+04
5 6 7 .11126964E+04
6 7 8 .10970520E+04
7 8 9 .10675797E+04
8 10 11 .11390364E+04
9 11 12 .11255585E+04
10 13 14 .11826992E+04
11 15 14 .OOOOOOOOE+OO
13 2 7 .10002119E+04
14 7 11 .10016460E+04
15 11 14 .10037032E+04
16 3 8 .10010362E+04
17 8 12 .10075203E+04
18 4 9 .10035400E+04
21 1 6 .10000000E+04
22 6 10 .10000000E+04
23 10 13 .10000000E+04
24 13 15 .10000000E+04
The shape of the 118 model is shown in Figure 7.10. For practical
reasons it is desirable to look at the entire structure rather than simply 118 of
it. That is done here by running two graphics routines: LPLOTl.FOR takes
the geometry of 118 of the structure and reflects it about the diagonal to get a
view of 114 of the structure as shown in Figure 7.11; the routine
LAYOUTPLOT.FOR can then be used to produce a picture of the entire
structure of Figure 7.1. (See also Section 7.8 below.)
CABLE NETS AND FABRlC STRUCTURES 163
Figure 7.10 Initial equilibrium shape of lI8 skylight structure.
Figure 7.11 Initial equilibrium shape of lI4 skylight example.
7.4 Smoothing
Having obtained a shape using in this case the grid method, it is natural to
hope that a better shape perhaps close to the given shape might be obtained.
For example, there are many reasons such as efficiency in the use of material
to look for a shape in which the bar forces or the membrane stresses are
constant.
When such a shape exists, it may be possible to generate it using a
nonlinear structural analysis computer program. This is the idea of
"smoothing" which has been described by Haber and Abel (1982). There are
two steps to the smoothing procedure:
I) The output of the shape finding program is used as input to the
nonlinear analysis program except that the desired cable forces, here taken to
be 1,000 lbs rather than the computed cable forces are used. Clearly this
initial configuration is not in equilibrium.
2) A fictitious, small value of E, Young's modulus, is used during
the subsequent analysis. This essentially disables the elastic stiffness matrix
164 ANALYSIS OF GEOMETRICALLY NONLINEAR STRUCTIJRES
and prevents the cable forces from changing. When the nonlinear analysis
converges, a shape has been obtained for which the cable forces are constant.
(See Figure 7.12) In this case the cables have moved to form geodesics in the
surface of the structure.
A final note with regard to this example. It is, in this case, possible
to start from the beginning using the analysis program avoiding the grid
method entirely. Doing so allows a shape to be achieved directly using an
analysis program while specifying the cable forces. The three dimensional
nonlinear analysis program TR3DNL.FOR was slightly modified to find an
equilibrium shape given a state of prestress. This modified version,
TR3DSM.FOR basically paralyzes the elastic stiffness matrix by inserting
zero for Young's modulus leaving the task of shape finding to the geometric
stiffness matrix.
The input file TRDOMESM.DAT is used with TR3DSM.FOR to
resolve Example 7.2 above for the case of constant cable forces (1000 lbs).
Two output files, FORT60 and FORT600, are saved but only FORT600,
listed below, provides information on the number of bars, the number of
nodes, the number of supports, the number of load steps, and the number of
iterations. Member forces, connectivity, and area follow in subsequent lines.
The second part of the output gives the final coordinates of the nodes. The
input file is the same as DOME except for the information on symmetry
which is not supplied here. Instead, 'constraints' are specified for all degrees
°
of freedom: the number 1 indicates a constrained degree of freedom whereas
the number indicates a free degree of freedom.
Input file TRDOMESM.DAT

24 15
1000. 1 2
1000. 2 3
1000. 3 4
1000. 4 5
1000. 6 7
1000. 7 8
1000. 8 9
1000. 10 11
1000. 11 12
1000. 13 14
1000. 15 14
1000. 14 12
1000. 2 7
1000. 7 11
1000. 11 14
1000. 3 8
1000. 8 12
1000. 4 9
1000. 12 9
1000. 9 5
1000. 1 6
1000. 6 10
1000. 10 13
1000. 000
1000. 1000.
1000 1000.
1000. 000.
10
1000. 1000. 1000. 1000. 000.
1000. 1000 1000 1000.
Figure 7.12 Skylight structure with constant cable forces.

1000. 13 15
o. O. O. 111
o. 21. 5 O. 1
o. 43. O. 1
o. 64.5 1
O. .86. 30. 111
21.5 O. 111
21.5 21.5 O.
21. 5 43. O.
21.5 64.5 28.227 111
43. O. O. 111
43. 21.5 O.
43. 43. 22.832 111
64.5 O. O. 111
64.5 21.5 13.578 111
86. O. O. 111
Output file FORT600
24 15 0 1 10
1000.000 1 2 .006
1000.000 2 3 .006
1000.000 3 4 .006
1000.000 4 5 .006
1000.000 6 7 .006
1000.000 7 8 .006
1000.000 8 9 .006
1000.000 10 11 .006
1000.000 11 12 .006
1000.000 13 14 .006
1000.000 15 14 .006
1000.000 14 12 .006
1000.000 2 7 .006
1000.000 7 11 .006
1000.000 11 14 .006
1000.000 3 8 .006
1000.000 8 12 .006
166 ANALYSIS OF GEOMETRlCALL Y NONLINEAR STRUCTURES
1000.000 4 9 .006
1000.000 12 9 .006
1000.000 9 5 .006
1000.000 1 6 .006
1000.000 6 10 .006
1000.000 10 13 .006
1000.000 13 15 .006
.000 .000 .000 111 O. o. o.
.000 23.335 10.558 100 O. o. o.
.000 44.720 19.276 100 O. o. o.
. 000 65.121 25.835 100 O. o. o.
.000 86.000 30.000 III O. o. O.
21. 500 .000 . 000 111 O. O. O.
21. 405 23.015 11.302 000 O. O. O.
21.363 44.164 20.805 000 O. O. O.
21. 500 64.500 28.227 111 O. O. O.
43.000 .000 .000 111 O. O. O.
42.977 22.377 12.723 000 O. O. O.
43.000 43.000 22.832 111 O. O. O.
64.500 . 000 . 000 111 O. O. O.
64.500 21.500 13.578 111 O. O. O.
86.000 .000 .000 111 O. O. O.
7.5 A More Complex Example
Figures 7.13 (a)-(b) show a cable net which looks something like a
hyperbolic paraboloid; it is supported at four points and bounded by four
edge cables on four sides. This structure is more complex than the example
above because of its four edge cables. A node map has been created and
displayed in Fig. 7.14. (Note that cable node 15, which defines the "sag" of
the edge cable has been forced to lie on a grid point.)
In order to get started with the grid method in this case it is first
necessary to go off and solve the (horizontal) cable problem by hand. Figure
7.14 is a free body diagram of half of one edge cable. Symmetry requires
that the force components at the exposed end be equal. Moment equilibrium
then provides R (see below) after which individual cable forces and reactions
may be calculated using force equilibrium starting with cable 32 and moving
upwards up to cable 29. Following are the required calculations:
L moments about top ~ 1000x2+ 1000x4+ I 000x6+ 1000x2+Rx4-Rx6=0 ~

14000=2R ~R=7000.
force in cable 32 is ~60002 + 7000 2 = 9219lbs
force in cable 31 is ~ 5000 2 + 7000 2 = 8602 lbs

Figure 7.13 (a) Plan of the 'hyperbolic paraboloid' cable net.
Figure 7.13 (b) Schematic of the hyperbolic paraboloid cable net.

The grid method again gives a solution using HYPAR.DAT as the
input file for LAYOUT.FOR following the previous example. One of the
two output files FORT8 is also listed. In the final design (Fig. 7.15) the
interior cable forces are of course not constant. Smoothing can be used to
determine a shape with constant internal cable forces but in this case the real
stiffness of the edge cable must be used if the plan shape of this structure is
to be preserved.
H = 4000 tv = 8000
+-
1000
3@2ft
1000
.............. +---~
2@2ft
•• III'
Figure 7.14 Edge cable for the hyperbolic paraboloid cable net.
Input file HYPAR.DAT

36 23
1000. 1 7
1000. 07 13
1000. 13 17
1000. 17 20
1000. 20 23
1000. 2 8
1000. 814
1000. 14 18
1000. 18 22
1000. 3 9
1000. 9 15
1000. 410
1000. 1016
1000. 512
1000. 2 1
1000. 8 7
1000. 14 13
1000. 18 17
1000. 21 20
1000. 3 2
1000. 9 8
1000. 15 14
1000. 19 18
1000. 4 3
1000. 10 9
1000. 5 4
1000. 11 10
1000. 6 5
8944. 21 23
9433. 22 21
8602. 19 22
9219. 15 19
9219. 16 15
2019 1
0
4
17 18
3 8 2
13 17 14 22
2 7 11
7 16 8 21 9 25
6 10 12
15 2 20 3 24 4 6
Figure 7.15 Node map for the hyperbolic paraboloid.
1233.
1138.
1053.
1006.
Figure 7.16 Cable forces in the hyperbolic paraboloid.
8602. 11 16
9433. 12 11
Figure 7.17 Equilibrium shape of 114 of the hyperbolic paraboloid.
8944. 6 12
O. O. O. 3 2 7 2 7
2. 0.0 0.0 1 8 3 8 1
4. 0.0 0.0 1 9 4 9 2
6. 0.0 0.0 1 10 5 10 3
8. 0.0 0.0 1 12 6 12 4
10. 0.0 10. 1 1
O. 2. 0.0 2 1 8 13 8
2. 2. 0.0 2 9 14 7
4. 2. 0.0 3 10 15 8
6. 2. 0.0 4 11 16 9
6.40 2. 0.0 10 12 16
8. 1. O. 5 6 11
O. 4. O. 2 7 14 17 14
2. 4. O. 8 15 18 13
4. 4. O. 9 14 16 19
6. 2.285 O. 10 11 15
O. 6. O. 2 13 18 20 18
2. 6. O. 14 19 22 17
2.285 6. O. 18 15 22
O. 8. O. 2 17 21 23 21
1. 8. O. 20 22 23
2. 6.40 O. 18 19 21
O. 10. O. 1 2
Output file FORT8 for the hyperbolic paraboloid
COORDINATES FORCE BALANCE

X Y Z x y Z
1 .OOOOOE+OO .OOOOOE+OO .50000E+01 .20000E+04 .20000E+04 .23817E-02
2 .20000E+01 .OOOOOE+OO .52205E+01 .26542E-04 .20000E+04 .11969E-02
3 .40000E+01 .OOOOOE+OO .58820E+01 .90543E-05 .20000E+04 .69918E-03
4 .60000E+01 .OOOOOE+OO .69699E+01 .13404E-04 .20000E+04 .19375E-02
5 .80000E+01 .OOOOOE+OO .84115E+01 .77365E-05 .20000E+04 .13206E-04
6 .10000E+02 .OOOOOE+OO .10000E+02 .OOOOOE+OO .OOOOOE+OO .OOOOOE+OO
7 .OOOOOE+OO .20000E+01 .47795E+01 .20000E+04 -.25352E-04 .14257E-02
8 .20000E+01 .20000E+Ol .50000E+01 .26542E-04 -.40700E-04 . 16566E-02
9 .40000E+Ol .20000E+01 .56688E+01 .24420E-04 -.37987E-04 .73296E-03
10 .60000E+Ol .20000E+01 .67931E+01 .17458E-04 .17434E-04 .38531E-02
11 .64000E+01 .20000E+01 .70671E+01 -.64792E+01 -.79566E+Ol .90093E-02
12 .80000E+01 .10000E+Ol .83747E+01 .58832E+00 -.39761E+00 -.19493E-03
13 .OOOOOE+OO .40000E+01 .41180E+01 .20000E+04 -.53616E-04 .21760E-02
14 .20000E+01 .40000E+Ol .43312E+01 -.15173E-04 .13650E-04 .84525E-03
15 .40000E+01 .40000E+Ol .50000E+Ol -.27346E+01 -.27344E+01 .63961E-02
16 .60000E+01 .22850E+Ol .67329E+01 .72955E+01 .95622E+01 .17164E-01
17 .OOOOOE+OO .60000E+Ol .30301E+Ol .20000E+04 .66451E-04 .99830E-03
18 .20000E+01 .60000E+01 .32069E+01 .23749E-04 .37939E-04 .46322E-02
19 .22850E+01 .60000E+01 .32671E+01 .95622E+01 .72954E+01 .18472E-01
20 .OOOOOE+OO .80000E+01 .15885E+01 .20000E+04 -.37339E-05 .12033E-02
21 .10000E+01 .80000E+01 .16252E+01 -.39796E+00 5.8840E+00 .41335E-02
22 .20000E+01 .64000E+01 .29329E+01 -.79562E+01 -.64792E+01 -.88023E-03
MEMBER OUTPUT
+END -END FORCE
1 1 7 0.10060590E+04
2 7 13 0.10532775E+04
3 13 17 0.11383743E+04
4 17 20 0.12326866E+04
5 20 23 0.12770435E+04
6 2 8 0.10060590E+04
7 8 14 0.10544263E+04
8 14 18 o . 11471840E+04
9 18 22 0.12120830E+04
10 3 9 0.10056666E+04
11 9 15 0.10544264E+04
12 4 10 0.10039003E+04
13 10 16 0.10220560E+04
14 5 12 0.10006747E+04
15 2 1 0.10060591E+04
16 8 7 0.10060591E+04
17 14 13 0.10056667E+04
18 18 17 0.10039001E+04
19 21 20 0.10006747E+04
20 3 2 0.10532782E+04
21 9 8 0.10544270E+04
22 15 14 0.10544270E+04
23 19 18 0.10220565E+04
24 4 3 o.11383762E+04
25 10 9 0.11471854E+04
26 5 4 0.12326910E+04
27 11 10 0.12120878E+04
28 6 5 0.12770487E+04
29 21 23 o. 11056922E+05
30 22 21 0.11476976E+05
31 19 22 0.10404198E+05
32 15 19 0.11034478E+05
33 16 15 o .l1034482E+05
34 11 16 0.10404225E+05
35 12 11 o .11477012E+05
36 6 12 o .11056961E+05
In this case of smoothing it is not possible to go directly from scratch

to the final shape in one step using the analysis program. (The analysis does
not converge.) But a two step procedure does work:
1) Fix the plane coordinates, specify the desired cable forces, set
E=O in all cables and iterate using the analysis program. The analysis
program in this case acts like a modified form of the grid method except that
the cable forces rather than their horizontal projections are specified.
HYPARSM1.DAT is the input file used and TR3DSM.FOR compiled with a
very small ec=l.
2) Release the artificial constraints of the above step and restore the
real stiffness of the edge cable and iterate using the analysis program. When
this step has been completed a proper shape has been found in which the
interior cables have constant forces.(See Figure 7.18.) The steps require that
FORT600 be saved as HYPARSM2.DAT which is appropriately edited to
become the input file for another run with a recompiled TR3DSM.FOR with
ec=25. D06. Editing frees nodes 8, 9,10,11,12,14,15,18,19,21 and 22
from the horizontal constraints.
Listed below are the output file FORTEC1600 (FORT600 for the
case of ec=l, step 1 above), the input file HYPARSM2.DAT (edited file
FORT600, step 2 above) and output file FORTEC25600 (FORT600 for the
case of ec=2 5. DO 6, step 2 above).
Input file HYPARSMl.DAT
36 23
1000. 1 7
1000. 07 13
1000. 13 17
1000. 17 20
1000. 20 23
1000. 2 8
1000. 8 14
1000. 14 18
1000. 18 22
1000. 3 9
1000. 9 15
1000. 4 10
1000. 10 16
1000. 5 12
1000. 2 1
1000. 8 7
1000. 14 13
1000. 18 17
1000. 21 20
1000. 3 2
1000. 9 8
1000. 15 14
1000. 19 18
1000. 4 3
1000. 10 9
1000. 5 4
1000. 11 10
1000. 6 5
8944. 21 23
9433. 22 21
8602. 19 22
9219. 15 19
9219. 16 15
8602. 11 16
9433. 12 11
8944. 6 12
o. o. o. 11
2. 0.0 0.0 11
4. 0.0 0.0 11
6. 0.0 0.0 11
8. 0.0 0.0 11
10. 0.0 10. 111
o. 2. 0.0 11
2. 2. 0.0 11
4. 2. 0.0 11
6. 2. 0.0 11
Figure 7.18 Hypar with constant cable forces.
6.40 2. 0.0 11
8. 1. O. 11
O. 4. O. 11
2. 4. O. 11
4. 4. O. 11
6. 2.285 O. 11
O. 6. O. 11
2. 6. O. 11
2.285 6. O. 11
O. 8. O. 11
1. 8. O. 11
2. 6.40 O. 11
O. 10. O. 111
Output file FORTEC1600

36 23 0 1 10
1000.000 1 7 .006
1000.000 7 13 .006
1000.000 13 17 .006
1000.000 17 20 .006
1000.000 20 23 .006
1000.000 2 8 .006
1000.000 8 14 .006
1000.000 14 18 .006
1000.000 18 22 .006
1000.000 3 9 .006
1000.000 9 15 .006
1000.000 4 10 .006
1000.000 10 16 .006
1000.000 5 12 .006
1000.000 2 1 .006
1000.000 8 7 .006
1000.000 14 13 .006
1000.000 18 17 .006
1000.000 21 20 .006
1000.000 3 2 .006
1000.000 9 8 .006
1000.000 15 14 .006
1000.000 19 18 .006
1000.000 4 3 .006
1000.000 10 9 .006
1000.000 5 4 .006
1000.000 11 10 .006
1000.000 6 5 .006
8944.275 21 23 1. 000
9433.247 22 21 1. 000
8602.237 19 22 1. 000
9219.220 15 19 1.000
9219.220 16 15 1.000
8602.236 11 16 1. 000
9433.246 12 11 1.000
8943.412 6 12 1. 000
.000 . 000 5.000110 o. o. o.
2.000 . 000 5.313110 o. o. o.
4.000 . 000 5.964110 o. o. o.
6.000 . 000 6.995110 o. o. o.
8.000 . 000 8.396110 o. o. o.
10.000 . 000 10.000111 O. o. o.
.000 2.000 4.687110 o. o. o.
2.000 2.000 5.000110 O. o. o.
4.000 2.000 5.664110 O. o. o.
6.000 2.000 6.762110 o. o. o.
6.400 2.000 7.036110 o. o. o.
8.000 1.000 8.344110 o. o. o.
.000 4.000 4.035110 o. o. O.
2.000 4.000 4.336110 O. o. O.
4.000 4.000 5.000110 O. o. o.
6.000 2.285 6.705110 O. o. o.
.000 6.000 3.005110 O. o. o.
2.000 6.000 3.237110 o. o. o.
2.285 6.000 3.295110 O. o. o.
.000 8.000 1.604110 O. o. o.
1.000 8.000 1. 656110 o. o. o.
2.000 6.400 2.964110 o. o. O.
.000 10.000 .000111 o. o. o.
Input file HYP ARSM2.DAT
36 23 0 1 10
1000.000 1 7 .006
1000.000 7 13 .006
1000.000 13 17 .006
1000.000 17 20 .006
1000.000 20 23 .006
1000.000 2 8 .006
1000.000 8 14 .006
1000.000 14 18 .006
1000.000 18 22 .006
1000.000 3 9 .006
1000.000 9 15 .006
1000.000 4 10 .006
1000.000 10 16 .006
1000.000 5 12 .006
1000.000 2 1 .006
1000.000 8 7 .006
1000.000 14 13 .006
1000.000 18 17 .006
1000.000 21 20 .006
1000.000 3 2 .006
1000.000 9 8 .006
1000.000 15 14 .006
1000.000 19 18 .006
1000.000 4 3 .006
1000.000 10 9 .006
1000.000 5 4 .006
1000.000 11 10 .006
1000.000 6 5 .006
8944.275 21 23 1.000
9433.247 22 21 1.000
8602.237 19 22 1.000
9219.220 15 19 1. 000
9219.220 16 15 1. 000
8602.236 11 16 1.000
9433.246 12 11 1. 000
8943.412 6 12 1.000
.000 .000 5.000110 O. O. O.
2.000 .000 5.313110 O. O. O.
4. 000 .000 5.964110 O. O. O.
6.000 .000 6.995110 O. o. O.
8.000 . 000 8.396110 O. o. O .
10.000 .000 10.000111 O. O. O.
. 000 2.000 4.687110 O. O. O .
2.000 2.000 5.000000 O. O. O.
4.000 2.000 5.664000 O. O. O.
6.000 2.000 6.762000 O. O. O.
6.400 2.000 7.036000 O. O. O.
8.000 1.000 8.344000 O. O. O.
.000 4.000 4.035110 O. O. O.
2.000 4.000 4.336000 O. O. O.
4.000 4.000 5.000000 O. O. O.
6.000 2.285 6.705000 O. O. O.
. 000 6.000 3.005110 O. O. O.
2.000 6.000 3.237000 O. O. O.
2.285 6.000 3.295000 O. O. O.
.000 8.000 1.604110 O. O. O.
1.000 8.000 1.656000 O. o. O.
2.000 6.400 2.964000 o. o. o.
. 000 10.000 . 000111 o. O. O .
Output file FORTEC25600

36 23 0 1 10
1000.000 1 7 .006
1000.000 7 13 .006
1000.000 13 17 .006
1000.000 17 20 .006
1000.000 20 23 .006
1000.000 2 8 .006
1000.000 8 14 .006
1000.000 14 18 .006
1000.000 18 22 .006
1000.000 3 9 .006
1000.000 9 15 .006
1000.000 4 10 .006
1000.000 10 16 .006
1000.000 5 12 .006
1000.000 2 1 .006
1000.000 8 7 .006
1000.000 14 13 .006
1000.000 18 17 .006
1000.000 21 20 .006
1000.000 3 2 .006
1000.000 9 8 .006
1000.000 15 14 .006
1000.000 19 18 .006
1000.000 4 3 .006
1000.000 10 9 .006
1000.000 5 4 .006
1000.000 11 10 .006
1000.000 6 5 .006
10588.110 21 23 1.000
11025.270 22 21 1.000
10161.760 19 22 1. 000
10747.010 15 19 1.000
10747.010 16 15 1.000
10161. 760 11 16 1. 000
11025.270 12 11 1.000
10588.110 6 12 1.000
.000 .000 5.000110 O. O• O.
2.000 . 000 5.309110 O. O. O.
4.000 . 000 5.966110 O• O. O.
6.000 . 000 7.008110 O. O. O.
8.000 . 000 8.412110 O. O. O.
10.000 . 000 10.000111 O. O. O.
.000 2.000 4.691110 O. O. O.
1. 916 1. 916 5.000000 O. O. O.
3.926 1. 907 5.675000 O. O. O.
5.973 1.982 6.784000 O. O. O.
6.384 2.005 7.058000 O. O. O.
7.984 1.002 8.364000 O. O. O.
.000 4.000 4.034110 O. O. O.
1.907 3.926 4.325000 O. O. O.
3.993 3.993 5.000000 O. O. O.
5.984 2.288 6.725000 O. O. O.
.000 6.000 2.992110 O. O. O.
1.982 5.973 3.216000 O. o. O.
2.288 5.984 3.275000 O. o. o.
.000 8.000 1.588110 O. O. O.
1. 002 7.984 1.636000 O. o. O.
2.005 6.384 2.942000 O. O. O.
. 000 10.000 .000111 O. O . O.
One final comment on the case of edge cables. The above analysis
does not seem to be sensitive to the edge cable forces specified initially. This
allows the designer to start with a rough approximation of the edge cable
forces (which of course correct themselves in the second step of the
analysis) thus avoiding some tedious cable calculations.
The two steps required above in the use of the analysis program can
be argued in terms of Newton's method where it is sometimes possible to
improve convergence by taking smaller steps. In the example here, rather
than trying to start from scratch and get shape in a single step, two steps are
used.
CABLE NETS AND FABRIC STRUCTIJRES 177
7.6 Membrane Finite Element Model
If you prefer to use a membrane finite element rather than a cable element to
represent fabric, that can be done as suggested above for cable nets. (See
Figures 7.19-22 where the above two examples are redone using a finite
element model.) These examples add little to the above discussion but it
should be noted that the analysis seems to converge more slowly in these
cases. Similarly, cases in which the geometry is "radial" rather than "flat" as
it is here can be handled using appropriate variations of the above
discussions.
Listed below are the data files MEMDOMSM.DAT and
MEMHYPSM.DAT which are used with the membrane finite element
program MEMSM.FOR to analyze the skylight dome and the hyperbolic
paraboloid respectively. In terms of detail, FORT600 of step I for the
DOME example is edited and saved as MEMDOMSMl.DAT which is used
as input for step 2 whose FORT60 and FORT600 output files are saved as
MEMDOMSM60 and MEMDOMSM600 respectively. Similarly
MEMHYPSMl.DAT, MEMHYPSM60 and MEMHYPSM600 are obtained
for the hyperbolic paraboloid example.
Input file MEMDOMSM.DAT
15 10 16
o. o. o. III
o. 21.5 o. 11
o. 43. o. 11
o. 64.5 11
o. 86. 30. 111
21.5 o. 111
21. 5 21.5 o. 11
21. 5 43. o. 11
21. 5 64.5 28.227 111
43. o. o. 111
43. 21.5 o. 11
43. 43. 22.832 111
64.5 o. o. III
64.5 21.5 13.578 III
86. o. o. III
1 6 2 1. 46.51 46.51
7 2 6 1. 46.51 46.51
2 7 3 1. 46.51 46.51
8 3 7 1. 46.51 46.51
3 8 4 1. 46.51 46.51
9 4 8 1. 46.51 46.51
4 9 5 1. 46.51 46.51
6 10 7 1. 46.51 46.51
11 7 10 1. 46.51 46.51
7 11 8 1. 46.51 46.51
12 8 11 1. 46.51 46.51
8 12 9 1. 46.51 46.51
10 13 11 1. 46.51 46.51
14 11 13 1. 46.51 46.51
11 14 12 1. 46.51 46.51
13 15 14 1. 46.51 46.51
15
Figure 7.19 Finite element map of 1/8 of the skylight example.
Figure 7.20 Finite element map of 1/4 of the hyperbolic paraboloid.

Input file MEMHYPSM.DAT
8 23 10 26
o. o. o. 11
2. 0.0 0.0 11
4. 0.0 0.0 11
6. 0.0 0.0 11
8. 0.0 0.0 11
10. 0.0 10. 111
o. 2. 0.0 11
2. 2. 0.0 11
4. 2. 0.0 11
6. 2. 0.0 11
6.40 2. 0.0 11
8. l. o. 11
o. 4. o. 11
2. 4. o. 11
4. 4. o. 11
6. 2.285 o. 11
o. 6. o. 11
2. 6. o. 11
2.285 6. o. 11
o. 8. o. 11
1. 8. o. 11
2. 6.40 o. 11
o. 10. o. 111
8944. 21 23 l.
9433. 22 21 l.
8602. 19 22 1.
9219. 15 19 l.
9219. 16 15 l.
8602. 11 16 l.
9433. 12 11 l.
8944. 6 12 l.
20 21 23 l. 500. 500.
21 20 22 l. 500. 500.
20 17 22 l. 500. 500.
17 18 22 l. 500. 500.
18 19 22 l. 500. 500.
18 17 14 l. 500. 500.
13 14 17 l. 500. 500.
18 14 19 l. 500. 500.
14 15 19 l. 500. 500.
14 13 7 l. 500. 500.
7 8 14 l. 500. 500.
15 14 9 l. 500. 500.
8 9 14 l. 500. 500.
15 9 16 l. 500. 500.
9 10 16 l. 500. 500.
10 11 16 l. 500. 500.
7 1 8 l. 500. 500.
1 2 8 1. 500. 500.
8 2 9 l. 500. 500.
2 3 9 1. 500. 500.
9 3 10 l. 500. 500.
3 4 10 l. 500. 500.
11 10 4 l. 500. 500.
4 5 11 l. 500. 500.
5 12 11 l. 500. 500.
5 6 12 l. 500. 500.
180 ANALYSIS OF GEOMETRICALLY NONLINEAR STRUCnJRES
7.7 Patterning
The final step in the design of fabric structures is patterning. Fabric

structures are usually made by joining strips of fabric which have been cut
from rolls of cloth. How that is to be done is the subject of this section.
The object of the design process is to produce a structure which will assume
a specified shape when it is prestressed. This implies that cables, for
example, must be fabricated "too short" so that they will fit together when
prestressed. Similarly, there is a step in patterning called "compensation" in
which the stretch of the material under prestress is introduced into the
patterning process. That step can be easily appended to the end of the
patterning process and will not be discussed here.
Figure 7.21 Patterning strips for the 1/8 skylight example.

The basic problem of patterning is the construction of a curved

surface out of flat pieces of material. This can be done in the following
manner. First, the surface must be "triangulated". This is the process by
which the curved surface of the structure is approximated by a collection of
flat triangular facets. It is usually done during the analysis phase of fabric
structure design in order to deal with, for example, environmental loads. The
next step is to select "strips" of triangles over the structure. These strips must
Figure 7.22 Patterning strips for 1I4 hyperbolic paraboloid.

cover the structure and must be SERIALLY connected. The point is that
strips of serially connected triangles can be deformed into a flat sheet
without stretching.
A patterning for the two examples discussed above is described in
Figures 7.21-22. The patterning process will be described in more detail
under the discussions of computer programs which follow.
7.8 Computer Programs For Cable Nets and Fabric Structures
7.8.1 LAYOUT.FOR (The grid method)
This program uses an iterative scheme (Gauss-Seidel) to compute the z

coordinates over a grid of a cable net (finds shape) using the equations of
equilibrium in the z direction. It might be noted that it would have been more
efficient to solve these equations directly using for example Gaussian
elimination since they are linear. At least, the iterative scheme used is simple
and requires little computer storage.
With regard to the details of LAYOUT.FOR the example of Figure
7.1 is useful. In this case it is required to find the shape ( compute the z
coordinates) of a skylight which is formed using crossed circular arches with
a rise of30".
The data required by layout is included in the file DOME. In order
to get started it is necessary to prepare a node map Figure 7.8 and to
compute the z coordinates along the arch as described above in Example 7.2.
Since this structure is 118 symmetric it is possible to work with the small
piece of the structure shown.
The data file DOME first lists the number of bars and the number of
nodes. It then lists the member description. (In this case the horizontal
component of the member force and the two ends of each member.) Finally,
node data is specified over the grid. At each node the z coordinates are set to
be zero unless they are know to be fixed otherwise. Next, a "I" is used to
indicate that the node is "fixed" in the z direction; then "4" is used to
indicate terms along the diagonal (This is required later by the graphics
routines.); node connectivity is then specified by listing adjacent nodes. (The
repeated listing of nodes is used to cover the case of symmetry.)
With regard to the details of coding, following statement number
678, there is a primary DO loop set at 200 iterations. For each iteration a new
set of z coordinates is computed keeping all but the coordinate which
corresponds to diagonal term fixed. Figure 7.4 explains how the vertical
force components in each bar are proportional to the z coordinates of the end
of the bar. For each bar, the coefficients of these force terms are collected in
SUM and SUMl which are subsequently used to compute new z coordinates.
The final DO loop computes the residual error associated with the final
coordinates. To run LAYOUT.FOR the DOS command is
LAYOUT DOME FORT8 FORTl50
Here FORT8 is an output file and FORTl50 is set up to pass information to

subsequent programs. (DOME is, of course, the input file described above.)
When the analysis program is used to find shape the DOS command is:
TR3DSM DOME FORT60 FORT600
7.8.2 LPLOT1.FOR
(This graphics routine constructs a 114 symmetric piece of a structure from a

118 symmetric piece.) As described for the example DOME above, it can be
useful to construct the entire geometry of a structure even when less is
required for reason of symmetry. This program handles the first step in so
doing by reflecting the 118 solution about the 45 degree line. That is the
function of the primary DO loop of this short program.
It can be run, immediately following LAYOUT.FOR, using the DOS
commands
COPY FORTl50 FORT50

DEL FORTl50
LPLOTl FORT50 FORTl50
LPLOTl at one point waits for keyboard input. The coordinate shift of the y-
axis is then specified (Figure 7.7). In this case it is 86".
7.8.3 LAYOUTPLOT.FOR
(This graphics routine creates a full symmetric model of a structure given 114
of it.)The final graphics step takes the" DOME" from 114 symmetry to full
symmetry. Generally, this program projects each node and bar in the 114
model into three new pieces to fill out the model. There are of course many
cases in which, for reasons of symmetry, all three reflections are not
required. It can be run, in this case immediately following LPLOT1.FOR,
using the DOS commands
COPY FORTI 50 FORT50

DEL FORTI 50
LAYOUTPLOT FORT50 FORTI50
7.8.4 PATTERN.FOR
(This program produces "patterns" from which fabric is to be cut. Note that
patterning is done here using grid points from the shape finding algorithm. It
could as well have been done using points interpolated over the shape.) This
program creates the actual fabric dimensions so that material strips can be
cut from rolls of fabric. It begins by reading output from the layout program
(FORTI 50) from which it gets the three-dimensional shape of the structure.
It then reads strip definitions. In the case of DOME these are shown in the
file DOMEPAT.DAT which is listed below. It then produces coordinates
and graphic output which describe how the fabric should be cut. It can be run
with the following DOS commands after running LAYOUT.FOR:
COPY FORTI 50 FORT50

PATTERN FORT50 FORT60 FORTI 0 DOMEPAT.DAT
Here FORT60 is a scratch file. The cutting data appears in FORTIO.
5 40
24 37
4 11::-_ _'" 9 ....- - - , j
3...,._ _.... 8 23 11::---'" 36
22 19
2 iE----I1JlI 7 1&:---....
21 ' - - _ - - I . 18
1 ' - - _ - - I.. 6
strip 1 from "DOME" example "fan" strip defined by sequence
defined as 1 6 2 7 3 8 4 9 5 21 182219233637 -24 40
Figure 7.23 Numerical descriptions of strips.

Input file DOMEPAT.DAT
4
9
1 6 2 7 3
8 4 9 5
7
6 10 7 11 8
12 9
5
10 13 11 14 12
3
13 15 14
Given a physical "strip" across a structural surface, this strip may be

described by listing a sequence of nodes as indicated in Fig. 7.21. This
sequence is understood by the computer as follows: Points 1 and 6 define a
starting line; each additional node defines a triangle which is serially
connected to the preceding one by using the line formed by the new node
and the second node in the sequence back from it. The "fan strip" of Fig.
7.23 indicates a configuration in which this description does not work. In
this case triangle 37-24-23 requires going back in the strip sequence 3 rather
than 2 nodes. That fact is indicated by assigning a negative sign to node 24
in the strip sequence.
A three-dimensional description of a strip is now in place. The
projection of this representation onto a plane can be done in the following
manner. The orientation of the first edge (edge 1-6 from DOME above) of
the strip on the plane is arbitrary. In a typical step, a new node (eg. node 2)
describes a new triangle in space one side of which is the line from the last
node to the new one. The real three-dimensional angle between the last two
lines defined on the strip in then computed using the scalar product of unit
vectors which describe the slopes of these lines. This angle allows the next
point to be located on the plane projection of the strip. When all the points
have been projected onto the plane, the strip is rotated so that it will lie
within a piece of fabric of minimum width. Finally, suitable coordinates for
the strip which are convenient for cutting are printed out.
With regard to programming details, PATTERN. FOR begins with
some now familiar input and then produces a plot showing the length of lines
in the surface to be patterned. This plot can be useful when checking
patterns. Its primary do loop, (DO 11 1=1, NSTR), produces the
patterning information for each strip. It begins by laying out arbitrarily the
first line in the strip as described above. In the typical step, which begins
with the statement 1200 I P= lABS (L 1ST ( J + 1 ) , it constructs the
unit vector which describes the slope of one side of the new triangle. It then
computes the real three-dimensional angle between itself and the last
triangle. This allows a new point to be projected onto the plane of the fabric.
After all points of the strip have been projected onto the fabric
plane, the entire strip is rotated so that it can be contained within a piece of
fabric of minimum width.
With regard to subroutines called by PATTERN.FOR:
• ROTATE which is used to "rotate" a strip into a piece of fabric of
minimum width is constructed from the plotting routines which themselves
determine the maximum dimensions of a rotated object.
• CONVRT creates a single array R given three arrays X, Y, and Z.
• S PLOT and PLOT are the usual plotting subroutines.
7.9 Problems
1. Set up the iterative scheme to determine the shape of a radially symmetric

cable net. (Hint: The discussions of this chapter have been limited to
relatively "flat" configurations; in the case of a radially symmetric
configuration another approach may be used. Let Pi be the radial coordinate
at level i. Write the radial equilibrium equation at level i, J;(p) = o. Solve
these equations iteratively.
2. The square-based tents at the "Great Adventure Amusement Park" have a

more complex radial geometry in which the cross sections of the structure
are not circular. How would you find shape in cases like this.
CHAPTER 8
THREE-DIMENSIONAL BEAM-COLUMNS
8.1 Introduction
This chapter is concerned with the application of the perturbation method to

the continuous problem of a three-dimensional beam. The reason for doing
so is quite simple. It is now common to use the so-called beam-column
equations to generate the element stiffness for plane frames. The question is
then what should be done in the case of three-dimehsional frames. This
chapter suggests that if "beam-columns" are used in plane frames then
"three-dimensional beam-columns" should be used in space frames. Just
what is to be included in three-dimensional beam-columns is the subject of
this chapter. The use of two-dimensional beam-columns has already been
discussed above in Chapter 1 and Appendix 1. The derivation of the well-
known beam-column equations using the methods of this section is discussed
in Appendix 4.
Structural engineers have been well-served by the so-called "beam-
columns" over the years. Roughly speaking, the beam-column equations
include the effect of member displacement upon bending moment to the end
that the bending stiffness becomes a function of the axial load (Gere, 1963).
The beam-column equations, of course become singular at the flexural
buckling load.
Typically, the term beam-column implies a two-dimensional beam in
which there is bending about a single axis. This chapter is concerned with
straight, three-dimensional prismatic beams in which there is bending about
two axes and torsion. As will be seen, three-dimensional beams have the
potential of many complex beam-column like interactions. A finite
difference solution and a power series solution are proposed below which
includes all possible interactions of initial stress resultants and can be used
to generate the member stiffness in a three-dimensional frame.
Of the various interactions which can occur in three-dimensional
beam problems, the best known is probably the phenomenon of torsional
(lateral) buckling (Timoshenko, 1936). In the most simple case, the presence
of bending moment about the strong axis of a beam weakens its effective
stiffness about its weak axis (stress softening) leading eventually to lateral
buckling. The presence of axial compression, of course, only makes matters
worse (see below). Biot (1965) in his classic book discusses another case of
interaction in which axial tension increases the torsional stiffness in a

prismatic bar. More generally, it would not be a simple matter to
characterize all possible types of interactions which can occur in three-
dimensional beam-columns. Not knowing what to expect places an
increasing importance on the problem formulation used. In fact, the
interactions of stress resultants in three-dimensional beam-columns are not
well understood toady. An indication of the possibilities is given in
(MacBain et al. 1999) using the using 3D graphical displays.
The fact is that we are comfortable when considering the softening
effect of axial compression upon member stiffness in two-dimensional beam-
columns and at least willing to think about the softening effect of bending
moment in lateral buckling problems in three-dimensional beams. It will be
seen below that the entire initial state of stress in a member can interact with
the member stiffness matrix in a rather complex manner. Since any
incremental analysis has initial states of stress associated with it which are
not simply compression, there is every reason to use the appropriate three-
dimensional "beam-column" described below.
This chapter describes a general approach which subsumes all the
effects cited above. Numerical procedures for computing the member
stiffness for three-dimensional beams are presented in which four coupled
differential equations are solved numerically. In order to find the 6x6
member stiffness matrix for a three-dimensional beam, this system of
equations must be solved 6 times. Several applications are discussed.
8.2 The Equations of Three-Dimensional Beam-Columns
The three-dimensional beam-column equations are now available in the

literature (Spillers, 1990) and are presented in Table 1 below. This section
reviews their derivation. For an alternative derivation see Chen et al. (1993).
First of all, a perturbation method (Green et al., 1968) is used to
describe the beam response. That is, a given straight beam, in equilibrium
under given forces, is subjected to a small load perturbation. It will be seen
that this method has the advantage of describing what are typically thought
of as nonlinear effects without recourse to theories of large strain. It also
produces the tangent stiffness now common in nonlinear analysis and
describes buckling as the response to load perturbation becomes singular.
The starting point is the equilibrium of a three-dimensional beam segment
(Reissner, 1962) described by the force and moment equilibrium as
P'+p =0
(8.1)
M'+m+txP =0
THREE-DIMENSIONAL BEAM-COLUMNS 189
Here P and M are the usual force and moment stress-resultants with p and m
the applied forces and moments and the prime symbol refers to the
differentiation with respect to arc length. The vector t is of course the unit
tangent vector which will sometimes be referred to as i in the applications
cited below. In component form
p = Pxi + Pyj + Pzk

(8.2)
M=Mxi+Myj+Mzk
and the base vectors i, j, k selected so that Px and M x represent thrust and
torque, respectively. Going from initial to the perturbed configuration,
changes are indicated as
pO ~pl =po +EP

MO ~Ml =Mo +EM (8.3)
i O~il =i o +d;jO ~jl =jO +EJ;k o ~kl =ko +Ek
where E is the anticipated small parameter and the bar is used to distinguish
the perturbation term. The perturbed terms of Eq. 8.2 are now inserted into
the equilibrium equations of the perturbed configuration given in Eq. 8.1. In
component form these equations are
M1'i1 + M1i1'+Ml'Jol + M1Y J.1'+M1'k 1 + M1kl'

x x Y z z
(8.5)
+m1i 1 +m1J.1 +m1k1 +t1 xp1 =0
x Y z
In order to complete the analysis, it is simply a matter of inserting

the definition of the perturbations (Eq. 8.3) into Eqs. 8.4 and 8.5 and
collecting terms in the coordinate directions. Before doing so it is convenient
to introduce displacements in the following manner. Let ro represent the
rotation vector associated with any beam element. If in the initial
configuration the beam is straight and lies along the x-axis, for the case of
small rotations ro can be written (from Fig. 5.2) as
(8.6)
where Bx is the torsional rotation, b z is the z-component of the beam

displacement vector, by is the y-component of the beam displacement can,
of course, be used to determine the changes of the base vectors (from Eq.
5.18) as
i O ~il =io +d=io +roxio

jO ~ jl = t + e} = jO + ro x jO (8.7)
°
k ~kI = k °+ ek = k °+ ro x k °
Clearly
ro x..°= s:
uy,xJ
·0 s: kO
+ uz,x
ro x jO = -8y,x iO + BxkO (8.8)
rox k o = - Us:.o B'o

zx • - xJ
Table 1 First-order equations.

Force Equilibrium
°
i component: ~'+Px -P;by,xx -PzObz,xx =0
jO component: Py'+ Py + Pxoby,xx - PzOBx,x =0
°
k component: ~'+Pz +P~bz,xx +P;Bx,x =0
Moment Equilibrium
°
i component: Mx'+m x -M~by,XX -M~bz,xx =0
jO component: -,- ° 0-
My +my + Mxby,xx -MzBx,x -Pz =0
°
k component: -,-
M ° + My°Bx,x +Py
z +m z + Mxbz,xx - =0
Table 2 Zero-order equations.
Force Equilibrium
°
i component: pO'+po
x x =0
jO component: pO'+po
y y
=0
°
k component: pO'+pO
z z =0
Moment Equilibrium
°
i component: MO'+mo
x x =0
j °component: MO'+m
y
o _po =0
y z
k °component: MO'+m
z
o o
z +py =0
The first-order equilibrium equations in the jO, t, kO directions are

collected in Table 1. They are obtained by writing the equilibrium equations
in the perturbed configuration and keeping terms which are linear in the
small parameter E. For convenience this parameter can then be set to one in
which case the terms indicated by bars (Eq. 8.3) then represent the full
perturbation. The zero-order equations are assumed to satisfy equilibrium in
the iO, jO, kO directions and in the initial configuration. They are a direct
outcome ofEq. 8.1 in component form and presented in Table 2.
It should be noted that the assumptions of small displacement theory
have been invoked above to allow arc length differentiation to be replaced by
differentiation with respect to the space variable x.
8.3 The Member Stiffness Matrix
In the most simple case of elastic analysis, a straight beam segment with no
cross-sectional warping can be described by four uncoupled events: axial
deformation, torsional deformation, and bending deformation about 2 axes.
If the two-dimensional beam-column equations are used to introduce
nonlinear effects, there is a coupling between axial forces and bending. If the
equations of Table 1 are used, they introduce a coupling between all the
stress resultants present in the three-dimensional frame element. The
resultant member stiffness matrix is a tangent stiffness matrix due to the use
of the perturbation method.
First of all, from considerations of equilibrium it can be argued that
the member stiffness matrix is a 6x6 matrix. Second, using Newton's method
for nonlinear structural analysis, each step (iteration) of the nonlinear
analysis becomes simply a linear analysis which uses the local tangent
stiffness. Therefore the terms in the member stiffness matrix are "forces" due
to unit "displacements". Third, if the member forces are chosen properly, the
terms in the member stiffness matrix may be computed by introducing
sequentially 6 discontinuities into the boundary conditions of the system of
equations in Table 1:
• A unit axial discontinuity (\(L) = 1.0, is introduced first (i=l).

• A unit torsional discontinuity Bx(L) = 1.0, is introduced next (i=2).
• Four flexural discontinuities are finally introduced, (i=3, 4, 5, 6):
bz,x(L) =-1.0; by,x(L) =1.0;

bz,x(O) = -1.0; by,x(O) = 1.0;
Note that the four flexural discontinuities are those which are used in
moment distribution: a unit rotation about an axis of flexure is applied at one
end of a beam while the other end is held fixed. Two beam ends and 2 axes
of flexure then imply four flexural discontinuities.
The next step will combine some of the equations in Table 1 to
produce a system of four displacements Ox,Oy,Oz, and ex as unknowns.
Quite generally and following linear beam theory, the second equation can
be combined with the sixth equation eliminating the shear term Py ; similarly
the third equation can be combined with the fifth equation in the table to
eliminate the shear term Pz . Furthermore at this point all the member loads
can be eliminated as not of interest,
(8.9)
This implies that the initial axial thrust and torque, Pxo, P~, Pzo and M2,
must be constant and that the initial bending moments, M~ and M~ , can at
most be linear functions of x. The six equations of Table 1 then reduce to the
following four equations:
-, 0 0
Px -Py Oy,xx - Pz oz,xx =0
--, ° 0-O
Mx -Myoy,xx -Mzoz,xx-
(8.10)
My"-2M~'ex,x +M2 o y,xxx -M~ex,xx +Pxooz,xx =0
Mz"+2M~'ex,x +M2 o z,xxx +M~ex,xx -P~Oy,xx =0
To complete the formulation, four constitutive equations are appended,
Here the k's are the usual spring constants from considerations of strength of
materials. Some general comments on this system of equations can now be
made:
• The last three equations are coupled and must be solved simultaneously;
then the first equation can be integrated to complete the solution .
• The last two equations are fourth order in the beam displacements (like the
linear elastic beam equations); the other two equations are second order .
• The equations themselves are linear in x since the initial moments M~ ,
M~ may be linear in x. (Timoshenko (1936) remarks that equations of this

type may be solved using Bessel functions.)
The rest of this section is concerned with some of the details
required to construct the three-dimensional beam-column stiffness matrix
which incidentally, is nonsymmetrical.
The 6x6 stiffness matrix is first obtained from the displacements that
are computed for each discontinuity, i, from the following relations and then
transformed into the perturbed coordinate system:
~i(L) = (k x8 x,x(L»)i
MXi(L) = (krBx,x(L»)i
Myi(L) = (- ky8 z,xx(L»)i
(8.12)
Mzi(L) = (k z 8y,xx(L»)i
M yi (0) = (- ky8 z,xx (O»)i
Mzi(O) = (k z 8y,xx(0)\
The perturbed, stiffness matrix having elements, K ij , becomes:
(8.l3)
The transformed components ~l (L),

M ~ (0), in the perturbed coordinate system are computed from the
components of Eq. 12 by considering the change in the unit vectors due to a
small rotation vector roo Thus
(8.14)
(8.15)
The 3-D beam-column stiffness matrix may now be finalized as:
~1(L) ~2(L) Px3(L)+P~(L) ~4(L)-P~(L) ~S(L) Px6 (L)
M x1 (L) M x2 (L) Mx3(L)+M~(L) MX4(L)-M~(L) MXS(L) M x6 (L)
My1(L) MY2(L)-M~(L) M y3 (L) MY4(L)+M~(L) MyS(L) My6 (L)

Mz1(L) Mz2(L)+M~(L) Mz3(L)-M~(L) M z4 (L) MzS(L) M z6 (L)
Myl(O) M y2 (O) M y3 (O) M y4 (O) MyS(O) My6(O)+M~(O)

M z1 (O) M z2 (O) M z4 (O) MZS(O)-M~(O) M z6 (O)
8.4 Numerical Solutions
As a practical approach, each beam is divided into a number of segments and

the solution of Eqs. 8.10 represented by a finite number of points using the
method of finite differences. In the case of the solutions described below,
twenty spaces have been taken along each beam. Using central differences
and introducing fictitious points to handle the boundary conditions, seventy
five simultaneous equations were solved on a PC for every case which
involved the last three of Eqs. 8.10; once these equations have been solved
together, the first of Eqs. 8.10 can be solved. The computed solutions are in
agreement with the charts of Gere (1963) when those solutions were
appropriate. Other numerical checks are described below. A FORTRAN
subroutine, BMCOL3.FOR, which produces a 6x6 member stiffness matrix
is included on the CD supplied with the text. A similar subroutine for three-
dimensionanal beam elements of varying stiffness is described in Levy and
Gal, (2002 A ).
Dividing members into short segments can be costly when thinking
in terms of doing large, practical, three-dimensional frames. An alternative
approach (Spillers and Rashidi, 1997) is to construct the member stiffness
matrix using a power series approach. Eisenberger has used this approach for
the solution of a rich variety of buckling and vibration problems over the last
fifteen years. His elegant method is referred to, in the literature, as the exact
element method (see for example Eisenberger, 1990). In it the four
1HREE-DIMENSIONAL BEAM-COLUMNS 195
displacements are expanded in power series
(S.16)
(S.17)
(S.lS)
(S.19)
When these power series are inserted into the differential equations,
recurrence relations occur between the coefficients. Setting the expression,
which multiplies typically the term x m , gives
(S.20)
{m+2)[k r a m+2 +M~f3m+2 +M~Ym+2]

(S.21)
+ m[- Pzof3m+1 + P~ Ym+l]= 0
-ky{m + 4Xm + 3Xm + 2)Ym+4 + M2{m + 3Xm + 2)f3m+3

(S.22)
+ 2P~am+l + M~{m + 2)am+2 + P~mam+l + PxO(m + 2)Ym+2 =0
kz{m + 4Xm + 3Xm + 2)f3m+4 + M2(m + 3Xm + 2)Ym+3

(S.23)
+ 2Pzoa m+l -M~(m + 2)am+2 + Pzomam+l - PxO(m + 2)f3m+2 =0
Finally, a system of coupled, linear, algebraic equations in these coefficients

results when the boundary conditions are applied.
By inspection of Eqs. S.20 - S.23, an elimination scheme for the
coefficients can be constructed: the second of these equations can be used to
eliminate a in terms of 13 and r, the fourth equation can be used to eliminate
13; and the third equation can be used to eliminate y. Then in a secondary
scheme, the first equation can be used to eliminate A in terms of the
coefficients already computed.

A FORTRAN subroutine BMCOLSR.FOR has been written that
uses this series solution to compute the member stiffness matrix for 3D
frames for the case of geometrically nonlinear structures. This routine
replaces BMCOL3.FOR in the three-dimensional nonlinear frame analysis
program that is described in Chapter 5 and FR3DNLSR.FOR is the resulting
new program.
8.5 Special Cases
8.5.1 The Elastic Beam
When the initial forces are all zero, Eqs. 8.10 degenerate to the solution for
an elastic beam in which axial compression, torsion, and y and z axis
bending are all uncoupled:
kxox,xx =0
kTBx,xx = 0
(8.24)
- kyoz,xxxx =0
kzoy,xxxx= 0
8.5.2 Two-Dimensional Beam-Columns
When only an initial axialload,p~, is present, Eqs. 8.10 again uncouple as

described above. In this case, the y and z axis bending is controlled by the
familiar two-dimensional beam-column equations:
kxox,xx= 0
kTBx,xx= 0
(8.25)
- kyoz,xxxx + Pxo oz,xx= 0
kzoy,xxxx - P~ Oy,xx = 0
This is the case which is most commonly used today in the nonlinear
analysis of three-dimensional frames.
8.5.3. Lateral Buckling
If a constant initial moment M~ is added to the problem, the member twist

ex becomes coupled with tSz while tSx and tSy remain uncoupled (Eq. 8.18
below). The third of these equations is particularly interesting and can be
thought of as a beam-column equation in which the constant Pxo / k y is
2
replaced by (Pxo -M~ / kT)/ k y . With this correction Gere's beam-column
charts can also be used in this case.
(8.18)
8.5.4 A More Complex Case.
If an initial constant moment M~ is now added, ex, tSy , and tSz ,

become fully coupled and it is no longer possible to invoke charts for well-
known solutions. In this case the equations take on the form
kxtSx xx =0
kTex,xx - M~tSy,xx - M~ tSz,xx =0
(8.19)
kytSz,xxxx +M~ex,xx -Px°tSz,xx=O
kztSy,xxxx +M~ex,xx -Px°tSy,xx =0
The solution of Eqs. 8.19, the case of a beam subject to an initial

state of axial load Pxo and constant biaxial bending M~ and M~ is now
discussed. The first of Eqs. 8.19 is trivial. The remaining equations can be
written in operator form as
"'0
co
.9 2
c
o
.~ ~
~ :::::- 1.5
a.~
§~
-
U~
"'0 0':::
.~
iii
E
o 0.5
z
O+-----,,----. .----~----~._--~r_----~~
o 0.5 1.5 2 2.5 3
Normalized Bending Moment
My = M~LI ~k/T
Figure 8.1 Buckling curves.
o
The determinantal equation for this system is
which is basically quadratic in D2 and thus should be regarded to be

accessible. The solutions of this system generally have the form.
Ox = Al + A 2x + A3X2 + A4X3 + Asx4 + A6 Xs

+ A7 sin kI x + Ag cos kI X + A9 sin k 2x + AlO cos k 2x
{)z = BI + B 2x + B3 X2 + B4 X3 + Bsx4 + B6 XS
+ B7 sinkIx + Bg coskIx + B9 sink2x + B IO cosk2x
u~y = C1 +C2 X+C 3 X2 +C4 X3 +CSX 4 +C6 X5

+ C 7 sink1x + C g cosk1x + C 9 sink2x + C lO cosk2X
where kl and k2 are the roots of the quadratic described above. Coupling of
these coefficients requires that the polynomial terms of order quadratic or
higher must vanish and also provides eight relationships between the
coefficients of the trigonometric terms. The ten boundary conditions then
complete the problem statement.
For a particular case of a 3D cantilever under biaxial bending and
axial compression, buckling loads are obtained using the finite difference
method and presented in Figure 8.1. For M~ = M~ = 0 the result is in full
agreement with Timoshenko's (1936) analytical solution. More cases,
including added initial torsion are discussed in Levy and Gal (2002 B ).
8.5.5 The Effect of Initial Torsion I
The discussion of special cases has thus far been motivated by

practical applications and available solutions and therefore has been
dominated by the effect of axial load. An alternative would have been to
start with the case of initial torsion M ~ . In this case the equations are
kxox,xx =0
=0
kT()x ,xx
(8.20)
- kyoz,xxxx + M2oy,xxx = 0
kzoy,xxxx +M2 o z,xxx=0
Here only Oy and 0 z are coupled and it is possible to solve the

system by hand. For
1Professor Maciej Bieniek of Columbia University suggests that this case is of interest to and
known to those working on the mechanics of drill shafts.
Here D represents the derivative dldx. If the determinant of this system is set
to zero it follows that
and that 0 z and Oy have the form:
Oz = Al + A2x + A3X2 + A4X3 + Asx4 + A6XS

+ A7 sinh+ As cosh
Oy = BI + B2x + B3X2 + B4X3 + Bsx4 + B6XS
+ B7 sinh + Bs cosh
with k 2 = M22 /(k ykz ). Coupling of these coefficients implies that

A4 = As = A6 = B4 = Bs = B6 = 0 and that
- ky A7k + BsM2 =0
kyAsk + B7M2 =0
Given the eight boundary conditions (flexure about two axes) of this
system, the A's and B's can be computed explicitly by hand.
An interesting two-dimensional 'extension' to this case would be the
effect of twist on the classical 2D beam-column. This case was solved in
Levy and Gal (2002 B) using the finite difference method and the results are
presented in Figure 8.2 below.
There is a subset of Eqs. 8.20, in which torsion can produce flexural
buckling which is accessible to hand calculation (Figure 8.3). In the case of
Euler buckling a lateral displacement produces a bending moment
Mz = P~ Oy; in the torsional buckling problem, lateral displacements
generate bending moments from components of the applied torsion M2. For
this problem, Oy and Oz couple as
The determinantal equation in this case is

1HREE-DIMENSIONAL BEAM-COLUMNS 201
4.5 ~--------------------------------------------.
T=O (classical 20 beam-column)
4 1 - -_ __
~ 3. 5 -r:T::-=_3-:--_ _~
T=4
~
U5
T=5
g> 2.5
'i5
<::
&l 2
T=6
al
.~
1.5
iii
o
E
Z 0.5
o +----+----+---~----~--~--~~--~--~--~
o 0.5 1.5 2 2.5 3 3.5 4 4.5
Normalized Compression Load, P*
k = kyUk z = kzUk y
P* = ~p~L2/ky
Figure 8.2 Effect of initial twist on classical beam-column interactions.
which implies solutions of the form
Oz = Al + A2x+ A3 sinkx + A4 coskx

Oy =BI +B2x+B3sinkx+B4coskx
with k 2 = M22 I(kykz). In this case coupling of the coefficients implies that
A2 = B2 = 0 and that
-kyA3k + B4M2 =0
kyA4k + B3M2 =0
The problem statement is completed with the 4 boundary conditions

oy = 0 z = 0 @ x = 0 and x = 1. Setting the determinant of the matrix of the
coefficients equal to zero in this case implies the characteristic equation
coskl = 1.0
The smallest nontrivial root is kl = 27r or the critical buckling

torque is
Euler Buckling
M z - PxoOy = 0 (equilibrium); M z = kzoy,xx (constitutive equation)

pO
Oy,xx = / Oy
z
r.te r.te
- - f " J l ly
y
r.te -
0
~ My = -Mxlly,x
-- zf"JB~ M;z - 0
= -Mxo z x
o '
x -y plane r.te x- z plane Mx
Torsional Buckling
My + M20 y,x = 0 (deformed equilibrium, x-y plane)

M z + M20z x = 0 (deformed equilibrium, x-z plane)
M z = kzoy,xx , My = -kyoz,xx (constitutive equations)
Figure 8.3 Simple buckling cases.

This chapter has explored several special cases of three-dimensional

beam-column equations and suggests that the numerical procedures such as
method of finite differences and the power series approach can be used when
a solution of the general case is required. While not elegant, the method of
finite differences has the added advantage of being unconcerned with the
signs of the coefficients in the equations and thus includes both stress
hardening and softening in a single algorithm.
One of the driving forces of this chapter is the need to construct a
member stiffness matrix for the nonlinear analysis of three-dimensional
frames. At the moment, many of the available computer programs and even
theoretical discussions such as See and McConnell (1986) use two-
dimensional beam-column theory to construct such a member stiffness
matrix. Clearly such an approach leaves out the effect of lateral buckling. As
pointed out here it also leaves out other possible three-dimensional couplings
produced by initial loads.
8.6 Problems
1. The finite difference solution of the three-dimensional beam-columns is

rather slow to execute. Reconstruct the alternative solution using the power
series method.
2. Investigate the effect of member weight upon the buckling of a drill shaft.
3. Timoshenko claims that Bessel functions can be used to solve the lateral
buckling problem when unequal end moments are used. Compare the use of
Bessel functions with the use of the series method in this case.
4. Run the program BMCOL3.FOR using a single member cantilever and

verify Timoshenko's lateral buckling results.
CHAPTER 9
NONLINEAR ANALYSIS OF SHELLS
9.1 Introduction
This chapter extends the material of Chapter 6, Nonlinear Analysis

of Membranes, to include the effects of bending and the material of Chapter
5, Nonlinear Analysis of Space Frames, to remove rigid body rotations (Levy
and Gal 2002 c). Chapter 6 discussed a "membrane shell", using a finite
element developed for fabric structures applications where the stress levels
are almost constant. The approach to shell buckling of this chapter is quite
general, building upon this work in Chapter 6.
The history of finite element applications to shell buckling is
extensive going back to the work of Clough and Johnson (1968). The natural
mode contribution of Argyris et al. (1977) was a major addition to shell
theory. It was recently modified to include elastoplastic effects (Argyris et
al. 2000). Horrigmoe and Bergan (1978) used the co-rotational method for
nonlinear analysis and Bathe and Ho (1980), Hsiao (1987), Mohan and
Kapania (1997), Peng and Crisfield (1992) improved element performance
along those lines. The 3-D elasticity "degenerate" element of Ahmad et al.
(1970) was followed by, among others, Bathe and Balourchi (1980), Hughes
and Lui (1981), Dvorkin and Bathe (1984), and Buechter and Ramm (1992).
In an excellent review, Ibrahimbegovic (1997) addresses the various
approaches and the complex issues involved.
Here the derivation of the geometric stiffness matrix is somewhat
different but consistent with the approach used throughout this text. The
linear equilibrium equations for a flat triangular shell element in its local
coordinates system are first perturbed to yield the in-plane geometric
stiffness matrix. Then out-of-plane considerations that involve the effect of
rigid body rotations on member forces yield an out-of-plane geometric
stiffness matrix. The shell element that was chosen for that purpose
combines the constant stress triangle (CST) flat triangular membrane
element (Zienkeiwicz (1977)) and of the discrete Kirchhoff theory (DKT)
flat triangular plate element (Batoz et al. (1980)).
Finally a computer program, featuring incremental analysis and
Newton's method, geometric effects, pure deformation isolation, internal
stress retrieval and updating of nodal forces and coordinates is presented and
used to solve a number of problems that have appeared in the literature.
These problems include experiments, analytical results and numerical

problems that were solved by the finite element method. They compare well
with the results in the current literature. Some details in the derivation have
been omitted. The interested reader can find them in Gal (2002).
9.2. The Geometric Stiffness Matrix of Triangular Element Shells
The local geometric stiffness matrix of the shell element is split into three
distinct matrices:
e ]Shell
[K G TOTAL
= [K G
e ]mem
IP
[K e ]Plate [ e ]Shell
+ G IP + KG OP (9.1)
where the first, second and third terms on the R.H.S. ofEq. 9.1 represent the
in-plane geometric stiffness matrix of the membrane, the in-plane geometric
stiffness matrix of the plate and the out-of-plane geometric stiffness matrix
of the shell element respectively. The total, 'tangential' stiffness matrix for
use in nonlinear analysis will include, in addition, the linear elastic stiffness
matrices of a plane stress triangular element (membrane) and that of a
triangular plate element.
The geometrically nonlinear triangular shell element has eighteen
local degrees of freedom (DOF's): 3 displacements and 3 rotations at each
node. The membrane element contributes to nine displacement DOF's only.
The basic three noded constant stress triangular flat element has only six
local displacement DOF's that are shown in Figure 9.1. The out-of-plane
contribution (the normal stiffness) of the membrane element to the basic
local shell element is a displacement DOF in the direction normal to the
plane of the element. This nonzero contribution was encountered earlier in
prestressed trusses as the ability to carry load perpendicular to the bars.
Vm
-x
Figure 9.1 Triangular membrane element.

NONLINEAR ANALYSIS OF SHELLS 207
Wm
i
Wi
Figure 9.2 Triangular plate element.
The flexural element contributes to eighteen local DOF's. The basic

three noded plate triangular flat element has only nine DOF's that are shown
in Figure 9.2. The in-plane contribution to the basic local element adds two
displacement DOF's in the plane of the element. The out-of-plane
contribution adds a rotation DOF in the direction normal to the plane of the
element.
9.2.1 In-Plane Contribution of the Triangular Membrane Element
The elastic stiffness matrix in the local coordinate system of the membrane
element has the general form
(9.2)
and is explicitly presented in Eqs. 6.17 and 6.18.

The local in-plane contribution of the membrane element to the
geometric stiffness matrix is the gradient of the nodal force vector (Eq. 6.21)
]mem = V'F e
lrKeGIP mem (9.3)
and is explicitly given in Eqs.6.22 and 6.23.
9.2.2 In-Plane Contribution of the Triangular Plate Bending Element
The general form ofthe elastic stiffness matrix ofthe DKT triangular
plate element may be written as:
[ e ]Plate
KE == 2Af11-17f Bplate
T (t3-D )Bplated4d17 (9.4)
o 0 12
where 4 and 17 are the usual area coordinates and Bplate, which appears
explicitly in Batoz et al. (1980) as a function of 4 and 17 is slightly more
complex than the Bmem of the membrane. Here again the local in-plane
contribution of the plate element to the geometric stiffness matrix will be
derived as the gradient for fixed M, of the element nodal force vector which
is given as:
. (9.5)
where
(9.6)
The in-plane geometric stiffness matrix may be written symbolically

as:
(A )Plate (A 1m )plate
IP
IJ IP
[K eG ]Plate
IP
== VF e == (A .. \plate
plate jl rIP
(A jj )~ate (A )Plate
Jm IP
(9.9)
(Aml.)plate
IP
(A.mJ )Plate
IP
(A mm )Plate
IP
where
oFrz oFrz
0
Ox s Oys
(A rs )plate = oMrx oMrx
IP 0 (9.10)
oXs Oys
oMry oMry
0
oXs Oys
The expressions for the individual terms of the geometric stiffness matrix
were obtained in closed form using symbolic algebra.
The above follows directly from the discussion, in Chapter 3 on the
derivation of the geometric stiffness matrix for trusses. When applied to the
plate, chain rule differentiation with respect to the coordinates yields
(9.11)
where dx r =u r ; dYr =vr ; r=l,j,m.

The first expression on the R.H.S. of Eq. 9.l1 returns the elastic
stiffness matrix whereas the second expression becomes the geometric
stiffness matrix. It is left to clarify what is meant by M fixed in Eq. 9.11. The
moment vector M, ofEq. 9.6 contains three components that are functions of
~ and '1 and defined as:
Mxx(~''l)= (M~ -M~~ +(M: -M~h+M~ (9.12)
(9.13)
Mxy(~,'l)= (M~ -M~~ + (M; -M~h +M~ (9.14

where
{M~ M~y M~yy =M(q=O,1J=O) (9.15)
{M~ M~y M~yy =M(q=l,1J=O) (9.16)
{M; M'; M;y =M(q=O,1J=l) (9.17)
When the components ofM (Eqs. 9.12-9.14) are inserted into Eq. 9.11 it is
the values of these components at the nodes (Eqs. 9.15-9.17) that are held
fixed.
9.2.3 Out-of-Plane Contribntion to the Shell Geometric Stiffness Matrix
The out-of-plane stiffness matrix for the membrane was discussed in section
6.3. This section will derive the out-of-plane geometric stiffness matrix for
the plate along the same lines and subsequently present a combined out-of-
plane contribution. The derivation starts with the change in a vector G, due
to a small rotation that is given by Goldstein (1950) as
dG =o>xG (9.18)
where 0> is the rigid body rotation vector due to changes in the geometry. In
terms of joint displacements with respect to the local coordinate system, two
components Band -¢, of the rigid body rotation are obtained from Figure 6.2.
The third component is chosen for the plate arbitrarily, as the local z-rotation
of node i. Recall that for the membrane this component is included in the in-
plane contribution.
(9.19)
OJ z ' = BiZ
where (3 r t, is the displacement in the local (primed) z-direction of node r.

At each node the forces, moments and rotations may be written as
for r = i, j, m and the changes in the force and moment vectors are given in
more detail as
(9.20)
(9.21)
l
The out-of-plane stiffness contribution of the plate is now obtained as:
(dF'l,.., = [KG ~;" a' =-F' x., = n[A, Aj Am {:jc9.22l

where
F; 0 Frz 0
M; -Frz 0 0
Fj 0 0 0
Fe = ; Fr = -Mry
(9.23)
Mj 0 Mrz
Fm -Mrz 0 Mrx
Mm Mry -Mrx 0
Now ro can be described in terms of the displacement vector as:
(9.24)
where
e-c
0 0 --- 0 0 0
ae
(Ai)= 0 0 lie 0 0 0 (9.25)
0 0 0 0 0 1
c
0 0 0 0 0
ae
(Aj)= 0 0 -l/e 0 0 0 (9.26)
0 0 0 0 0 0
(Am)~l~
~1
0 l/a 0 0
0 0 0 0 (9.27)
0 0 0 0
and
(Oit, (oit, (omt,

(Oi )y' (0 it, (Om)y'
(Oi )z' (oJz' (Om)z'
{d i }= ; ~i}= ; {d m}=
(9 i t, (9 i t,
(9 m t, (9.28)
(9 i )y' (9 i )y' (9 m)y'

(9 i t, (9 i t, (9 mt,
Carrying out Eq. 9.22 for the plate and adding Eq. 6.38 (adjusted to 18
DOF's) for the membrane, results in the following out-of-plane shell
geometric stiffness matrix with respect to the local coordinate system:
(Ai)! (A i)2 (A i )3
[K~ t;ll = (A j)! (A j)2 (A j)3 (9.29)
(Am)! (Am)2 (Am)3

where
o 0 ar 0 0 0
o br 0 000
( ) 0 0 ar 000
Ar 1 = 0 o C
r o 0 -Mry
(9.30)
o 0 dr o 0 M rx
o 0 er 0 0 0
o -0 fr 0 0 0
o -0 gr 000
() 0 0 f3r 000
Ar 2 = 0 -0 hr (9.31)
000
o - 0 ir 000
o 0 ir 000
o 0 0 000
o 0 kr 000
( ) 0 0 Itr 000
Ar =0 0 0 (9.32)
3
000
o 0 lr 000
o 0 mr 0 0 0
=-(F) e-c _ (Frt,. f3 =-(F) ~ (Frt,. ~ = (Fr)y. = Frz.

a r r y' 'r r y' + , Ar , ar ,
ae e ae e a e
e-c M e-c e-c M
br =Frz - - ; c r =-E...;dr =Mrz --;e r =-Mry ---~;
ae e ae ae e
fr =- Frz ; g r = Frz ~; hr =- Mrz ; ir = M rz ~; i r = - M ry ~ + M rx ;

e ae ae e ae e
F M M
k r =-~·l, r =-~·m' r =~., .
a a a
214 ANALYSIS OF GEOMETRICALLY NONLlNEAR STRUCTURES
9.3 Element Pure Deformational Rotations and Translations
For a given set of nodal displacements it is necessary to isolate the pure

elastic deformations from the rigid body ones so as to proceed with the
determination of the internal stresses and the evaluation of the element nodal
forces. The internal stresses are needed to construct the geometric stiffness
matrix and the nodal forces are used to calculate the unbalanced load as the
analysis progresses. It is the local shell element pure translations and
rotations that are of interest and will be discussed below. The translations
are related to the membrane element whereas the rotations to the plate
element. Internal stresses follow from the linear constitutive relations: I)
0' = DE for the membrane, where 0' comprises the stress components (J x ,
(J y and 't, and E comprises the strain components Ii x' Ii y and r xy

t3
respectively and 2) M = -DBplateqe
for the plate, where qe is the pure
12
local deformation vector that includes one lateral displacement and two in-
plane rotations at each node. Since lateral displacements are trivially zero in
the local coordinates system only rotations are discussed.
9.3.1 Stress Retrieval in the Membrane Finite Element
Figure 9.3 describes the membrane element exact geometric size in its initial
and current configurations. The membrane strains are obtained directly as:
AC'-AC
Ii =--- (9.33)
x AC
B'D'-BD
liy =--B-D-- (9.34)
(AC'-AC)AD
= LDBD' _ LDBD" = AD'-AD AC (9.35)
Yxy BD BD
The second term of Eq. 9.33 is argued as follows. Each point along AC will
be stretched by a fraction of the total stretch CC'. That fraction is determined
by the relative position of the point along AC. Point D stretches to DD" and
therefore the length DD" will be equal to CC' multiplied by the fraction
AD/AC.
initial
configuration current
configuration
~----------~~--~----~~
A o on 0' C C'
Figure 9.3 Membrane pure strains of the shell element.
9.3.2 Stress Retrieval in the Plate Finite Element
Figure 9.4 describes two positions of a given element at two consecutive

iterations k and k+ 1 with normal unit vectors nand n ' respectively.
Moreover the unit vectors n;, nj, nrn , which are normal to tangent planes at the
nodes of the elastically deformed element define the pure nodal rotations
relative to n'. Note that the rigid body motion of the element is not fully
determined by the unit vector n ' since rotation about the element plane is still
possible given n'.
At any given iteration, analysis will provide for each node, three
rotation components Bx , By, Bz (in addition to three displacements).
These rotations are given with respect to the global coordinate system.
Euler's theorem implies that any finite rotation can be described as a simple
rotation of magnitude a about some fixed axis described by a unit vector
n*. Let
O=n*a (9.36)
describe this rotation. Given the rotation components: 9 = Bxi + By j + Bzk

and a = ~BI + B1: + Bi and the unit vector n* is computed from
n* = (Bxi + By j + Bzk)/ a . Here i, j, k, belong to the global coordinates X,
Y, Z. Following Noble (1969), a rotation matrix R that transforms the
components of a vector from global to local coordinates is related to Euler's
rigid body motion through
deformed
configuration
...... ::..
·k+1
I .............................................
initial
configuration
·k
J ~ k
X local
Figure 9.4 Pure flexural rotations of the shell element.
cosa + cn x
.2 • • -n z• sin a
cnxny • z• + ny• sina
cnxn
• • + n z• sin a
R T = cnxny
.2
cosa + cny • y• - nx* sina (9.37)
cnzn
• z* -ny• sina cnzn
cnxn * y* + nx* sina cosa + cn z
.2
where c = 1 - cos a. Actually any rigid body motion may be described

equivalently using either a rotation matrix R or the above representation
which involves n* and a. Let the small deformational rotation vector Oli
(Figure 9.4) describe the rotation between the vector n' and nj with the
properties
or equivalently
. -1(1 ,) nixn'
0) i =Slll ~Ini xnil '1ln i xn'l (9.39)
where 0) i is the pure rotation at node i in global coordinates. The pure

rotations at the other elemental nodes,} and m, may be computed as:
(9.40)
Here O)ri describes the relative rotation of joint r with respect to joint i and
is the vector form (Euler's rigid body motion) of the compound rotations
R/Ri . Let this compound rotations matrix be called Rij so that Rij= R/Ri .
Then the unit vector describing the rigid body rotation (Eq. 9.34) is given
(Noble, 1969) as:
0) .
* - jl (9.41)
n ji -110) ji I
where
0) . = (R: 23 -
JI IJ,
R:IJ, 32 )i + (RT31
IJ,
- R:IJ, 13 )J' + (RT21
IJ,
- RIJ,TI2 )k (9.42)
and the magnitude of rotation as:
a·· =cos
-I (RJ,II + RJ.22 + RJ,33 - 1J (9.43)
fl 2
Finally the relative rotation vector becomes
0) ji = n *ji a ji (9.44)
These pure rotations vector has components in the global coordinate system.
In order to complete the computation of the vector qe it is necessary to
transform this vector to the present local coordinate system.
218 ANALYSIS OF GEOMElRICALLY NONLINEAR STRUCTURES
9.4 Computer Program P16-SHELLNL.FOR
This program performs nonlinear analysis of thin shells in one of two

distinct ways - force control (constant force increments) or generalized
displacement control (the cylindrical arc-length method). The total stiffness
matrix C (N , N) is assembled at every iteration from the individual
contributions which are explicit. Subroutine RASM constructs this stiffness
matrix, C (N, N) , (size ISx IS) in the global coordinates system given the
five individual stiffness matrices in local coordinates. This five stiffness
matrices are: the linear elastic stiffness matrix of the membrane element
PLSTR; the linear elastic stiffness matrix of the shell, element
KEPLATE1 +KEPLATE2; the in-plane geometric stiffness matrix of the
membrane element MGSTIF; the in-plane geometric stiffness matrix of the
shell element KGPLATE and KGOUTPLANE, the out of plane geometric
stiffness matrix for both the membrane and shell elements.
The unbalanced load vector is stored in in the array P (N) and its
norm stored in ERROR. Convergence at any iteration n is achieved when
ERROR n +1<10- 5 *ERROR n . The system of equations is solved using
SOLV. Membrane stresses are calculated in UPDATE. For element number
L the membrane stresses (J' x' (J' y' and r xy are stored in the 1st, 2nd and 3rd
columns of PSIG respectively. The internal moments for the plate element
are calculated in UPOATE1 and stored in PSIG as follows: The moments
M xx' M y.y and M xy of node i are stored in the 4th, 5th and 6th columns
respectively and those of nodes j and k are stored in the 7th, Sth, 9th , 10th, 11th
and 12th columns respectively. UPOATE1 uses the pure deformation vector
in local coordinates (0 P ( 9) ) that is obtained in RE SET for I S T RE S S = 0
according to the method of section 9.3.2 and linear stress evaluation (section
9.3). The option ISTRESS=l works with full nodal deformations rather
than pure deformations. These (array OS) are transformed to element
coordinates using TRANS and stored in array OSl. PSOP then extracts
oP for linear stress evaluation. One must be aware that this option is load
step dependent for convergence and in many cases, having converged, might
result in a wrong answer.
As for the input the program uses formatted read statements. General
data is read form an input file using:
REAO(15,150)
NN,NS,NCONS,NIT,NFE,NSTEP,E,ANU,IARC, IO, ISTRESS
150 FORMAT(6I5,2010.2/,3I5)
where NN, NS, NCONS, NIT, NFE and NSTEP are the number of
nodes, the number of fixed supports, the number of additional constraints,
the number of iterations, the number of finite elements and the number of
load steps respectively. E is Young's modulus, ANU is Poisson's ratio,
IARC is the index that will determine force control analysis (IARC=O) or
displacement controlled analysis (IARC= 1), IO is the number of iterations
to be used for the arc-length method and ISTRESS will determine which of
two methods are to be used for determining internal moments.
Data pertaining to nodes is read using:
READ ( 15, 156) (R ( 3 * K- 2) , R ( 3 * K-
1),R(3*K) ,PSAVE(6*K-5),PSAVE(6*K-4)
1 , PSAVE(6*K-3) ,PSAVE(6*K-2),PSAVE(6*K-
1) ,PSAVE(6*K),K=1,NN)
156 FORMAT (9010.2)
where the arrays Rand P SAVE store nodal coordinates and external loads
respectively.
Data pertaining to the finite elements is read using:
IF(NFE.GT.O)
REAO(15,1158) ((IJM(L,I) ,I=1,3) ,TH(L), (PSIG(L,I)
1 ,I=1,12) ,L=l,NFE)
1158 FORMAT(5X,3I5,13010.2)
Here I JM is the element connectivity matrix. For element number L the i
node number is stored in the 1st column of I JM and the j and m node
numbers are stored in the 2nd and 3rd columns respectively. TH is the element
thickness and initial prestress is read into PSI G.
Finally it is left to the reader to study the main program and its
subroutines before producing any meaningful results from the output that is
generated. Care has to be taken with the data parameters and generated
output when invoking the arc-length method that is coded in the subroutine
OISPLACEMENTCONTROL.FOR.
9.S Examples
This section describes the analysis of seven examples of diverse nature and
shows the general usefulness of the shell element in identifying snap through
buckling, reverse snap, bifurcation and large rotation elastica like problems
of rigid nature. All the examples were run using the same SHELLNL.FOR.
Each example was run as a project bearing its name. Input and output data
files are identified by the example they represent.
9.5.1 Example 9.1. Bending of a Cantilever Plate
An initially flat cantilever plate (Figure 9.5) is subjected to a bending

moment at its tip. The analytical solution to this classical elastica problem
corresponds to a beam "rolled-up" into a circular arc of radius r and is given
by lIr = MlEI where M is the applied end moment. A complete circle is
achieved at M = 2nEIIL where L is the length of the cantilever. This
example was run using twenty elements with the following physical and
geometrical properties: Young's modulus E=1.2 xl06 kNlm2, Poisson's ratio
of 0.0, L=IO.O m, b=1.0 m and h=O.1m.
SHELLNL91.EXE activates the datafile SHELLNL91.DAT to
produce an outputfile SHELLNL91.0UT. Actually the program runs with
a.dat and produces a.out. The files have been renamed accordingly.
Data file SHELLNL91.DAT
22 2 0 10 20 100 120000.00 0.0

0 0 1
10.000000 .000000 .000000 .00 .00 .00 0.3d1
10.000000 1.000000 .000000 .00 .00 .00 0.3d1
9.000000 .000000 .000000 .00 .00 .00
9.000000 1.000000 .000000 .00 .00 .00
8.000000 .000000 .000000 .00 .00 .00
8.000000 1.000000 .000000 .00 .00 .00
7.000000 .000000 .000000 .00 .00 .00
7.000000 1.000000 .000000 .00 .00 .00
6.000000 .000000 .000000 .00 .00 .00
6.000000 1.000000 .000000 .00 .00 .00
5.000000 .000000 .000000 .00 .00 .00
5.000000 1.000000 .000000 .00 .00 .00
4.000000 .000000 .000000 .00 .00 .00
4.000000 1.000000 .000000 .00 .00 .00
3.000000 .000000 .000000 .00 .00 .00
3.000000 1.000000 .000000 .00 .00 .00
2.000000 .000000 .000000 .00 .00 .00
2.000000 1.000000 .000000 .00 .00 .00
1.000000 .000000 .000000 .00 .00 .00
1.000000 1.000000 .000000 .00 .00 .00
0.000000 .000000 .000000 .00 .00 .00
0.000000 1.000000 .000000 .00 .00 .00
1 2 4 0.1
1 3 4 0.1
3 4 6 0.1
3 5 6 0.1
5 6 8 0.1
5 7 8 0.1
7 8 10 0.1
7 9 10 0.1
9 10 12 0.1
9 11 12 0.1
11 12 14 0.1
11 13 14 0.1
13 14 16 0.1
13 15 16 0.1
15 16 18 0.1
15 17 18 0.1
__"'=~:::::------------ M=O
• Analytical Solution
• Current Study
•
•
M=2PI*EIIL
•
•
•
• • M=O.5PI*EIIL
M=PI*EIIL
Figure 9.5 Bending of a cantilever plate.
17 18 20 0.1
17 19 20 0.1
19 20 22 0.1
19 21 22 0.1
Output file SHELLNL91.0UT
NUMBER OF NODES 22
E= 120000.000000000
NU= O.OOOOOOOOOOOOOOOE+OOO
IARC= 0 Id= 0 ISTRESS= 1
COORDINATES LOADS
X Y Z PX PY PZ MX MY MZ
1 10.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0300 0.0000
2 10.000 1. 0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0300 0.0000
22 0.0000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
FINITE ELEMENT DATA

NO. NODES TH PRESTRESS
I J M SIGX SIGY TAU MXXI MYYI MXYI MXXJ MYYJ MXYJ MXXM MYYM MXYM
1 1 2 4 0.1 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2 1 3 4 0.1 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
20192122 0.1 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
LOAD STEP 1
COORDINATES LOADS
1 10.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0300 0.0000
2 10.000 1.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0300 0.0000
22 0.0000 1. 0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
ERROR = 0.42426406D-01
DISPLACEMENTS
X Y Z THX THY THZ
1 .OOOOOD+OO .00000D+00-.29999D+00 -.27749D-14 .59999D-01 .OOOOOD+OO
2 .OOOOOD+OO .00000D+00-.29999D+00 -.26976D-14 .59999D-01 .000000+00
20 O.OOOOOD+OO .00000D+00-.29999D-02-.36180D-15 .59999D-02 .OOOOOD+OO
LOAD STEP 1
COORDINATES
X Y Z
1 0.10000D+02 0.00000D+00-0.30000D+00
2 0.10000D+02 0.10000D+01-0.30000D+00
22 O.OOOOOD+OO 0.10000D+01 O.OOOOOD+OO
UNBALANCED LOADS
PX PY PZ MX MY MZ
1-.97233D+01-.66613D-12 .55423D+00 -.37484D-15 .19360D-14 .10301D-15
2-.97233D+01-.33199D-12 .55423D+00 .84504D-15 .20123D-15 .33434D-16
ERROR = 0.14705609D+02
LOAD STEP 100
COORDINATES
X Y Z
1 -0.48448D+00-0.95310D-12-0.64903D-01
2 -0.48448D+00 0.10000D+01-0.64903D-01
22 0.000000+00 0.100000+01 0.000000+00
UNBALANCEO LOAOS
PX PY PZ MX MY MZ
1 .764280-12-.221600-12-.173640-12 .170390-11 -.137670-13 -.252410-11

2-.150340-11 .141420-13-.153890-12 .169340-11 .444090-15 -.253120-11
22 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00 .000000+00
ERROR = 0.106582000-10
9.5.2 Example 9.2. Simply Supported Plate
A square plate is subjected to a uniform pressure q as shown in Figure 9.6.

One quarter of the plate was analyzed using eighteen elements. The response
is compared in Figure 9.7 to the analytical solution reported by Levy (1942).
This example was run using Young's modulus E=107 psi, Poisson's ratio
v = J(ij and h=0.12in and a=24in. Following is a truncated listing of
SHELLNL92.DAT (full geometry) and SHELLNL92.0UT that have been
invoked by SHELLNL92.EXE.
Data file SHELLNL92.DAT
49 0 72 10 72 2010000000.0.31622777
o 6 1
.0000 .0000 .0000
.0000 4.0000 .0000
a a
1: * *I * *I ***1
Figure 9.6 Simply supported uniformly loaded plate.
2 ,----------------------------------------------------,
LEVY(1942)
:2 --~.~-- SHELLNL
! 1.6
"E
Q)
E
~ 1.2
CO
a.
III
'6
"0 0.8
"/
.~ V
(ij
oE 0.4
Z
o ~--_r--~--~---,----r_--r_--~--,_--,_--_r--_r--_4
o 50 100 150 200 250 300
Normalized load (qa 4 /Eh4)
Figure 9.7 Displacement at Point 0 in Simply Supported Plate .
.0000 8.0000 . 0000

.0000 12.0000 .0000
.0000 16.0000 .0000
.0000 20.0000 .0000
.0000 24.0000 .0000
4.0000 .0000 .0000
4.0000 4.0000 .0000 -30.0
4.0000 8.0000 .0000 -30.0
4.0000 12.0000 .0000 -30.0
4.0000 16.0000 .0000 -30.0
20.0000 16.0000 .0000 -30.0

20.0000 20.0000 .0000 -30.0
20.0000 24.0000 .0000
24.0000 .0000 .0000
24.0000 4.0000 .0000
24.0000 8.0000 .0000
24.0000 12.0000 .0000
24.0000 16.0000 .0000
24.0000 20.0000 .0000
24.0000 24.0000 .0000
1 1
1 2
1 3
2 1
2 2
2 3
48 1
48 2
48 3
49 1
49 2
49 3
1 2 8 .1200
2 3 9 .1200
3 4 10 .1200
46 47 40 .1200
47 48 41 .1200
48 49 42 .1200
Data file SHELLNL92.0UT
NUMBER OF NODES 49
E= 10000000.0000000
NU= 0.316227770000000
COORDINATES LOADS
1 .00000 .00000 .00000 .00000 .00000 .00000 .00000 .03000 .00000
2 .00000 4.0000 .00000 .00000 .00000 .00000 .00000 .03000 .00000
49 24.000 24.000 .00000 .00000 .00000 .00000 .00000 .00000 .00000
CONSTRAINTS DATA
NO. NODE NO DOF NO
111
212
313
72 49 3
FINITE ELEMENT DATA

NO. NODES TH PRESTRESS
1 1 2 8 .12 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2 2 3 9 .12 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
72484942 .12 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
LOAD STEP 1
COORDINATES LOADS
1 .00000 .00000 .00000 .00000 .00000 .00000 .00000 .03000 .00000
2 .00000 4.0000 .00000 .00000 .00000 .00000 .00000 .03000 .00000
49 24.000 24.000 .00000 .00000 .00000 .00000 .00000 .00000 .00000
ERROR = 0.75000001D+01
DISPLACEMENTS
X Y Z THX THY THZ
1 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO -.15193D-04 .15193D-04 .OOOOOD+OO
2 .OOOOOD+OO .OOOOOD+OO .OOOOOD+OO .94171D-04 .53708D-02 .000000+00
49 .000000+00 .000000+00-.299990-02 .151930-04 -.151930-04 .OOOOOD+OO

LOAD STEP 1
COORDINATES
x Y Z
1 O.OOOOOD+OO O.OOOOOD+OO O.OOOOOD+OO
49 0.24000D+02 0.24000D+02 O.OOOOOD+OO
UNBALANCED LOADS
PX PY PZ MX MY MZ
1 .OOOOOD+OO .OOOOOD+OO .27713D+01 -.97145D-15 -.13878D-16 .OOOOOD+OO
2 .82918D+02 .35790D+02-.23393D+01 -.16772D-03 -.16952D-05 .23976D-03
49 .OOOOOD+OO .OOOOOD+OO .27713D+01 .26645D-14 .14988D-14 .OOOOOD+OO
ERROR = 0.62699235D+03
LOAD STEP 20
COORDINATES
x Y Z
1 O.OOOOOD+OO O.OOOOOD+OO O.OOOOOD+OO

25 0.12000D+02 0.12000D+02-0.22360D+00
49 0.24000D+02 0.24000D+02 O.OOOOOD+OO
UNBALANCED LOADS
PX PY PZ MX MY MZ

2 .47620D+03 .15675D+03-.22866D+02 .19789D-07 .39397D-07 .30931D-06
25 .11033D-08 .40631D-09 .76922D-10 .18616D-10 -.45734D-13 .21851D-09
ERROR = 0.80893953D-03
9.5.3 Example 9.3. Analysis of a Shallow Cylindrical Shell
A cylindrical shell of dimensions L=254mm and h=12.7mm (Figure 9.8) that

is pinned along its longitudinal edges and free at its curved edges is loaded at
its center point. Analysis is performed for two cases of thickness: t=12.7mm
and t=6.35mm. Young's modulus was taken as E=3.l0275 kN/mm2 and
Poisson's ratio as 0.3. Using symmetry one quarter of the shell was analyzed
with a 10xlO grid. Figure 9.9 shows a comparison to the results reported by
Hsiao (1987) with the phenomenon of snap-back clearly seen for the case of
t=6.35mm. Following is a truncated listing of SHELLNL93.DAT (for the
case t=12.7mm) that have been invoked by SHELLNL93.EXE. Only
highlights of the enormous output file SHELLNL93.0UT are presented. It
starts with the usual echo of the data and then goes through the iterative
process. Following loadstep 1 note that at "* * * * ITERATION NUMBER
3" the error was "ERROR = 0.211054810-03". Error
convergence criteria are met at iteration 4. Therefore, iterations cease and
the next load step takes over. Two more parameters that relate to the arc-
length method are printed. These are the length squared "OL2=
2 9 . 3281303301849" and the load factor "FACT=
9.314575418727726E-002". From FACT and the updated
coordinates at this stage one point on each curve of Fig. 9.8 (t=12.7) has
been generated. (In this case by looking at nodes 1 and 111.) As the analysis
proceeds a "NEGATIVE TERM ON THE DIAGONAL AT 724"
is encountered. This is expected since critical points are bypassed by the arc-
length method (loadsteps 5 to 7).
Data file SHELLNL93 .DAT
121 a 97 10 200 10 3.10275 0.3

1 6 1
.0000 .0000 12.6895
.0000 25.3996 12.5625
p
2L
h~
I(
*
---....I...--L-=;;;;;;::::±IIh
L )I( )1
Figure 9.8 Longitudinally simply supported shallow cylindrical shell.

.0000 228.2915 2.4094

.0000 253.5769 .0000
25.4000 .0000 12.6895
25.4000 253.5769 .0000

50.8000 .0000 12.6895
50.8000 253.5769 .0000

76.2000 .0000 12.6895
76.2000 253.5769 .0000
3.5
t=12.7mm
- - - - - Hsiao
3
... SHELLNL
I
I
2.5 I
I
Z
~ 2
'-'
"C
IU
.9 1.5
0.5
0
0 5 10 15 20 25 30
Displacement (mm)
t=6.35mm
0.8 - - - -- Hsiao
... SHELLNL
0.6
0.4
Z
~
"C 0.2
IU
0
...J
0
-0.2
-0.4
-0.6
0 5 10 15 20 25 30
Displacement (mm)
Figure 9.9 Displacements at points a and b of the shallow cylindrical shell.

101.6000 .0000 12.6895
101.6000 253.5769 .0000

127.0000 .0000 12.6895
127.0000 253.5769 .0000

152.4000 .0000 12.6895
152.4000 253.5769 .0000

177.8000 .0000 12.6895
177.8000 253.5769 .0000

203.2000 .0000 12.6895
203.2000 253.5769 .0000

228.6000 .0000 12.6895
228.6000 253.5769 .0000

254.0000 .0000 12.6895 -1.2
254.0000 253.5769 .0000

1 2
1 4
1 6
11 1
11 2
11 3
22 1
22 2
22 3
121 1
121 2
121 3
121 5
121 6
120 1
120 5
120 6
112 1
112 5
112 6
111 1
111 2
111 4
111 5
111 6
100 2
100 4
100 6
12 2
12 4
12 6
1 2 12 12.7000
2 3 13 12.7000
10 11 21 12.7000
12 13 23 12.7000
21 22 32 12.7000
23 24 34 12.7000
32 33 43 12.7000
34 35 45 12.7000
43 44 54 12.7000
45 46 56 12.7000
54 55 65 12.7000
56 57 67 12.7000
65 66 76 12.7000
67 68 78 12.7000
76 77 87 12.7000
78 79 89 12.7000
87 88 98 12.7000
89 90 100 12.7000
98 99 109 12.7000
100 101 111 12.7000
109 110 120 12.7000

12 13 2 12.7000
21 22 11 12.7000
23 24 13 12.7000
32 33 22 12.7000
34 35 24 12.7000
43 44 33 12.7000
45 46 35 12.7000
54 55 44 12.7000
56 57 46 12.7000
65 66 55 12.7000
67 68 57 12.7000
76 77 66 12.7000
78 79 68 12.7000
87 88 77 12.7000
89 90 79 12.7000
98 99 88 12.7000
100 101 90 12.7000
109 110 99 12.7000

111 112 101 12.7000
120 121 110 12.7000
Data file SHELLNL93.0UT

NUMBER OF NODES 121
E= 3.10275000000000
NU= 0.300000000000000
COORDINATES LOADS
1 .00000 .00000 12.690 .00000 .00000 .00000 .00000 .00000 .00000
2 .00000 25.400 12.562 .00000 .00000 .00000 .00000 .00000 .00000
111 254.00 .00000 12.690 .00000 .00000 .00000 .00000 .00000 .00000
121 254.00 253.58 .00000 .00000 .00000 .00000 .00000 .00000 .00000
CONSTRAINTS DATA
NO. AT NODES DOF NO
1 1 2
2 1 4
3 1 6
66 111 1
67 111 2
68 111 4
69 111 5
70 111 6
95 12 2
96 12 4
97 12 6
FINITE ELEMENT DATA

NO NODES TH PRESTRESS
1 1 212 12.7 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2 2 313 12.7 0.00 0.00 .00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
LOAD STEP 1
ERROR = 0.12000000D+00
DL2= 29.3281303301849 FACT= 0.100000000000000

LOAD STEP 1
ERROR = 0.24398085D-01
DL2= 29.3281303301849 FACT= 9.999999999999987E-002
LOAD STEP 1
ERROR = 0.21105481D-03
DL2= 29.3281303301849 FACT= 9.314575418727726E-002
LOAD STEP 1
ERROR = 0.21182225D-04
LOAD STEP 2
ERROR 0.14683566D+00
DL2= 58.6562606603699 FACT= 0.215506995139774
LOAD STEP 2
ERROR = 0.93162090D-03
DL2= 58.6562606603699 FACT= 0.203092997001754
LOAD STEP 2
ERROR 0.75035584D-04
DL2= 234.625042641479 FACT= 0.421975519876304
LOAD STEP 4
ERROR = 0.14925216D-03
LOAD STEP 5
ERROR = 0.11689717D+00
DL2= 469.250085282959 FACT= 0.519347694081416
LOAD STEP 5
ERROR = 0.20282994D+00
NEGATIVE TERM ON THE DIAGONAL AT 724
DL2= 469.250085282959 FACT= 0.519338641344191
LOAD STEP 5
ERROR = 0.25778760D-01
DL2= 469.250085282959 FACT= 0.463308867908064
ERROR = 0.60527602D-03
DL2= 469.250085282959 FACT= 0.463216649218310
LOAD STEP 5
ERROR = 0.81589471D-04
LOAD STEP 6
ERROR = 0.13534852D-01
DL2= 703.875127924438 FACT= 0.474493978037810
LOAD STEP 6
ERROR = 0.40802921D-04
DL2= 703.875127924438 FACT= 0.392607019349170
LOAD STEP 10
ERROR = 0.17222246D-03
DL2= 1140.27770723759 FACT= 0.414133102394352
LOAD STEP 10
ERROR = 0.14731258D-03
DL2= 1140.27770723759 FACT= 0.414103471180779
LOAD STEP 10
ERROR = 0.12847919D-03
DL2= 1140.27770723759 FACT= 0.414076522020857
9.5.4 Example 9.4. Leicester's Shallow Spherical Shell
A spherical cap having a square base whose side is 61.8034in originates

from a sphere of radius R=100in. It is subjected to a concentrated load at the
crown with all edges hinged and immovable. This example was run using a
lOx 10 grid on a quarter of the cap with the following physical and
geometrical constants: Young's modulus E=10 3 ksi, Poisson's ratio v = 0.3
and a thickness of 3.9154in. Figure 9.10 show how well results compare to
those reported by Leicester (1968) who used Fourier transforms to solve his
nonlinear differential equations.
0.45
0.4
- - - -- Leicester
M • SHELLNL
.c 0.35
W I
~ 0.3
I
e:- 0.25 9'

r
&
I
'0
~ 0.2
ru 0.15
~
E
0 0.1
Z
0.05
0
0 0.5 1 1.5 2 2.5 3 3.5
Normalized displacement (w/h)
Figure 9.10 Displacement at crown of spherical shell.

9.5.5 Example 9.5. Mescall's Shallow Spherical Shell
Mescall (1965) reports analytic results for large displacement analysis of a

spherical shallow shell subjected to a concentrated load at the crown. Two
cases of support conditions are considered: rigid supports and vertically
restrained supports. Figure 9.11 shows how well results compare for the
rigidly supported shell that strain hardens and the vertically restrained shell
that snaps. The geometry of the shallow shell is formed as the intersection of
a cone whose tip lies at the center of a shell of radius 4.76in and having a tip
angle of 21.8°. The thickness of the shell is taken as t=O.OI576in, Young's
modulus E=107psi and Poisson's ratio v = 0.3.
0.8
- - - -- Mescall
M
0.7 ... SHELLNL
.c
W 0.6
t!
a:
N
0.5
~
"0
til 0.4
.2
"0
Q)
.!::! 0.3
- - - --
"iii
E
0
0.2
Z
0.1
0
0 2 3 4 5 6
Figure 9.11 Displacements of Me scali's shell.
3
- - - -- Experiment
M 2.5 ... SHELLNL
.c
W
t!
a:
N
2
~
"0
til 1.5
.2
"0
Q)
.!::!
"iii
E
0
Z 0.5
0
0 0.5 1 1.5 2 2.5 3 3.5
Figure 9.12 Displacements of Penning's experiment.

Penning (1966) perfonned an experiment on a shallow spherical

shell with a circular base of 4.0in that is cut from a sphere of radius 19.8in.
The thickness of the shell is taken as t=0.01576in, Young's modulus
E=107psi and Poisson's ratio v = 0.3. Results are displayed in Figure 9.12.
9.5.6 Example 9.6. Open Hemispherical Shell
A hemispherical shell (Figure 9.13) with an 18° opening at the top is loaded
by two orthogonal pairs of diametrically opposite equal forces. One pair is
directed inwards whereas the other pair is directed outwards. Using
symmetry, only one quarter of the shell was loaded and modeled using 16
latitudes by sixteen longitudes. In this case E==6.825*10 6 milliN/mm2;
t=0.04mm; Poisson's ratio of OJ. Results are compared in Figure 9.14 with
those of Buechter and Ramm (1992) and Simo et al. (1990).
F F
.([)
F
10( 2R )1
Figure 9.13 Open hemispherical shell.

450
- - - - - Buechler and Ramm
400
Simo el al
350 • SHELLNL
C-
O 300
]
250
~
g 200
outward
/.
"'C
ro 150
0
...J
100
50
0
0 2 3 4 5 6 7 8 9
Displacement (mm)
Figure 9.14 Open hemispherical shell.
I( )1
.-
p
or
Figure 9.15 Lateral buckling of an L-shape frame.
9.5.7 Example 9.7. Lateral Buckling of an L-Frame
An L-shaped frame that is clamped on one end and subjected to an in-plane

load at the other end (Figure 9.15) experiences lateral buckling.
Interestingly, this problem exhibits two different bifurcation loads,
depending on the load direction. Buckling was reached at between 1. IN-
1.2N for the outward loading and between 0.7N-0.8N for the inward.
Excellent agreement was found with the results of Simo et al. (1990) for
both bifurcation loads and with that of Battini and Pacoste (2001) for the
outward direction. Here L1=L2=240 mm; h=30.0 mm; t=0.6 mm; E=7l240
N/mm 2 ; Poisson's ratio of 0.3.

An experiment for the outward loading case was carried out by
Spillers et al. (1993). They reported a buckling load of 25 lbs. Here the
buckling load was attained at between 24-25 lbs for L]=10.64 in; L2=10.5 in;
h=1.0 in; 1=0.12 in; E=1.08 x 107 psi; Poisson's ratio of 0.3.
9.6 Some Remarks
A somewhat more complex approach for the derivation of the

geometric stiffness matrix for triangular shell elements has been
presented in this chapter. It separates the in-plane and the out-of-plane
geometric effects when taking the gradient of the nodal forces that
defines the geometric stiffness matrix. This approach is complete in
the sense that all contributions to the response of the same order of
magnitude are included.
Whereas this approach depends on an a priori chosen linear elastic
finite element it is independent of large strain formulations that are needed
otherwise. The nonlinear element is as good as the linear element it
originates from. This approach has its rigor and its success is ensured in
incremental analysis since equilibrium is satisfied in the deformed geometry.
Moreover the new technique is presented for the isolation of the
pure deformations that allows stress retrieval via linear constitutive relations
proved successful with respect to convergence and accuracy.
The presented shell element is overall efficient, reliable, accurate
and stable not failing in any of the diverse examples on which it was tested.
9.7. Problems
1. ModifY the computer program to solve an orthotropic thin shell.
2. ModifY the computer program so as to inflate a rubber balloon.
3. ModifY the computer program to inflate a rubber balloon that IS

strengthened by a cablenet.
REFERENCES
Ahmad, S., Irons, B. M., and Zienkiewicz, O. c., "Analysis of Thick and Thin Shell Structures
by Curved Finite Elements", Int. J. Numer. Methods Eng., Vol. 2, pp. 419-451, 1970.
Aitken, A.C., Determinants and Matrices, Oliver and Boyd, NY City, 1958.
ANSYS, Engineering Analysis Systems, Swanson Analysis Systems Inc., Box 65, Houston,
PA,1987.
Argyris, J. H., Angelopoulos, T., and Bichat, B., "A General Method for the Shape Finding of
Lightweight Tension Structures", Computer Methods in Applied Mechanics and Engineering,
3, 135-149, 1974.
Argyris, 1. H., Dunne, P. C., Malejannakis, G., and Schelkle, E., "A Simple Triangular Facet
Shell Element With Applications to Linear and Non-Linear Equilibrium and Elastic Stability
Problems", Compo Meth. Appl. Mech. & Engng., Vol. 11, 1977, pp. 97-131. and Vol. 10,
1977, pp.371-403.
Argyris, 1. H., Papadrakakis, M., and Karapitta, L., "Elastoplastic Analysis of Shells With The
Triangular Element TRIC", The Fourth International Colloquium on Computational of Shell
& Spatial Structures, Chania-Crete, Greece, 2000.
Bathe, K. 1., and Ho, L. W., "A Simple and Effective Element for Analysis of General Shell
Structures", Computers & Structures, Vol. 13, pp. 673-681,1980.
Bathe, K. J., and Bolourchi, S., "A Geometric and Material Nonlinear Plate and Shell
Element", Computers & Structures, Vol. 11, pp. 23-48, 1980.
Batoz, J. L., Bathe, K. J., and Ho, L. W., "A Study of Three Noded Triangular Plate Bending
Elements", Int. J. Num. Meth. Engng., Vol. 15, pp. 1771-1812, 1980.
Battini, J.-M., and Pacoste, C., "Co-rotational Beam Elements With Warping Effects in
Instability Problems", Comput. Methods Appl. Mech. Engrg., Vol. 191, pp. 1755-1789,2001.
Baumbauer, 1. c., and Tiersten, H. F., "Nonlinear Electrostatic Equations for Small Fields
Superimposed on a Bias", Journal of the Acoustic Society of America, 54, 4, 1017-1034,
1973.
Biot, M. A., Mechanics ofIncremental Deformation, John Wiley & Sons, New York, 1965.
Buechter, N., and Ramm, E., "Shell Theory Versus Degeneration - A Comparison in Large
Rotation Finite Element Analysis", Int. J. for Num. Meth Eng., Vol. 34, pp. 39-59, 1992.
Chan, S. L and Zhou, Z. H, "Pointwise Equilibrating Polynomial for Nonlinear Analysis of

Frames", Journal of Structural Engineering, ASCE, 120, 6 1703-1717, 1994.
Chen, W. F., and Lui, E. M., Stability Design of Steel Frames, CRC Press, Boca Raton, FL,
1993.
Clough, R., W., and Johnson, C., J., "A Finite Element Approximation for the Analysis of
Thin Shells", Int. J. Num. Meth. Eng., Vol. 4, pp. 43-60, 1968.
Crisfield, M. A, "A Fast Incremental/Iterative Solution Procedure That Handles "snap-

through"", Comput. Struct., Vol. 13, pp. 55-62,1981.
Dvorkin, E., N., and Bathe, K, J., "A Continuum Mechanics Based Four-Node Shell Element
for General Non-Linear Analysis" Eng. Comp., Vol. 1, 1984, pp. 77-88.
Eisenberger, M., ""Exact static and dynamic stiffness matrices for generalvariable cross
section member", AIAA Journal., Vol. 28, pp. 1105-1109, 1990.
Gal, E., "Triangular Shell Element for Geometrically Nonlinear Structural Analysis", Ph.D.
Thesis, Technion, (2002).
Gere, James M., Moment Distribution, D. Van Nostrand Co., Princeton NJ, 1963.
Goldstein, Herbert, Classical Mechanics, Addison-Wesley Publishing Company, Reading,

MA,1950.
Green, AE., Knops, RJ.,and Laws, N., "Large Deformations, Superimposed Small
Deformations, and the Stability of Elastic Rods", Inter. Journal Solids and Structures, 4, 555-
577, 1968.
Haber, R.B., and Abel, J.F., "Initial Equilibrium Solution Method for Cable Reinforced
Membranes. Part 1 - Formulation", Computer Methods in Applied Mechanics and
Engineering, 30, 263-284, 1982.
Horrigmoe, G., and Bergan, P. G., "Nonlinear Analysis of free-form Shells by Flat Finite
Elements", Compo Meth. Appl. Mech. & Engng., Vol. 16, pp. 11-35, 1978.
Hsiao, KM., "Nonlinear Analysis of General Shell Structures by Flat triangular Shell
Element", Computers & Structures, Vol. 13, pp. 665-675, 1987.
Hughes, TJ.R., and Lui, W.K, "Nonlinear Finite Element Analysis of Shells: Part 1 Three-
Dimensional Shells", Compo Meth. Appl. Mech. & Engng., Vol. 26, pp. 331-362,1981.
Isaacson, Eugene, and Keller, Herbert Bishop, Analysis of Numerical Methods, John Wiley
Co., New York City, 1966.
Ibrahimbegovic, A, "Stress Resultant Geometrically Exact Shell Theory for Finite Rotations
and its Finite Element Implementation", Appl. Mech. Rev., Vol. 50, no. 4, 1997, pp. 199-226.
Izzuddin, B. A, and Elnashai, A S. "Eulerian Formulation for Large-Displacement Analysis

of Space Frames", Journal o/Engineering Mechanics, ASCE, 119,3,549-569,1993.
Krysl, Petr, "Computer Stiffness Matrices for Buckling Analysis of Frames", Journal 0/
Engineering Mechanics, ASCE, 119,2,225-237, 1993.
REFERENCES 241
Kuo, Shyh-Rong, Yang, Yeong-Bin, and Chou, Jung-Hua, "Nonlinear Analysis of Space
Frames with Finite Rotations", Journal o/Structural Engineering, ASCE, 119, 1, 1-15, 1993.
Leicester, R. H., "Finite Deformations of Shallow Shells", Proc. ASCE, Vol. 94 (EM6, pp.
1409-1423), 1968.
Leonard, John William, Tension Structures, McGraw Hill, N Y City, 1988.
Levy, R., "Linear Programming in Prestressed Trusses", Computers and Structures, Vol. 33,
No.2, pp. 421-425, 1989.
Levy, R., and Brill, D.R., "Optimization for Exact Stability of 2-Bar Trusses", Computer
Utilization in Structural Engineering, pp. 520-529, (edited by James K. Nelson, Jr.),
American Society of Civil Engineers, New York, 1989.
Levy, R., and Gal, E., "Three-Dimensional Non-Prismatic Beam-Columns", International

Journal of Structural Stability and Dynamics, Vol. 2, No.3, pp. 395-408, 2002 A•
Levy, R., and Gal, E., "Buckling and Stress Softening of Beam-Columns under complex
Three-Dimensional Loading", International Journal of Structural Stability and Dynamics, Vol.
2, No.4, pp. 487-498, 2002B .
Levy, R. and Gal, E., "Three Noded Shell Finite Element for Large Rotations", Proceedings of
WCCM V, Fifth World Congress on Computational Mechanics, Vienna, July 7-12, 2002c,
paper no. 80327.
Levy, R., and Hanaor, A., "Optimal Design of Prestressed Trusses", Shells, Membranes and
Space Frames, Proceedings of the lASS Symposium on Membrane Structures and Space
Frames, Osaka, September 15th-19th, 1986, Vol. 3, pp. 207-214, (edited by K. Heki), Elsevier
Science Publishers B.V., Amsterdam, 1986.
Levy, R., Vilnay, 0., and Acheampong, K.B., "Exact Geometry Considerations in Buckling
Analysis of Trusses", Computers and Structures, Vol. 4-1, No.6, pp. 1241-1248, 1991.
Levy, S., "Bending of Rectangular Plates With Large Deflections", Tech. Notes, NACA, No.
846,1942.
MacBain, K., Saadeghvaziri, A., and Spillers, W. R., "Examples of Stress Hardening and
Softening in Three-Dimensional Beams", Journal 0/ Structural Engineering, ASCE, 129, 9,
1073-1075, 1999.
Mescall, J. F., "Large Deflections of Spherical Shells Under Concentrated Loads", ASME
Applied Mechanics Division, pp. 936-938, 1965.
Mohan, P., and Kapania, R. K., "Geometrically Nonlinear Analysis of Composite Plates and
Shells Using a Flat Triangular Shell Element", 38th AlAAIASMEIASCEIAHSIASC Structures.,
Struct. Dyn. and Mat. Coni, Kissimmee, FL, pp. 2347-2361 (AlAA Paper 97-1233), 1997.
Noble, B., Applied Linear Algebra, Prentice-Hall, New Jersey, 1969.
Otto, Frei (editor), Tensile Structures, MIT Press, Cambridge, MA 1973.

Peng, X., and Crisfield, M. A., "A Consistent Co-rotational Formulation for Shell Using The
Constant Stress/ Constant Moment Triangle" Int. J Num. Meth. Engng., Vol. 35, pp. 1829-
1847, 1992.
Penning, F. A., "Experimental Buckling Modes of Clamped Shallow Shells Under

Concentrated Load", ASME Applied Mechanics Division, pp. 247-304, 1966.
Reissner, Eric, "Variational Considerations for Elastic Beams and Shells", J Engr. Mech.
Division, Proc. ASCE, 88, EMl, Feb. 1962, pp. 23-57.
Schek, H.J., "The force density method for form finding and computation of general
networks", Computer Methods in Applied Mechanics and Engineering, 3, 1974, pp 115-134.
See, Thomas, and McConnell, Richard E., "Large Displacement Elastic Buckling of Space
Structures", Journal o/Structural Engineering, Vol. 112, no. 5, May 1986, pp. 1052-1069.
Siev, A., and Eidelman, 1.,"Stress Analysis of Prestressed Suspended Roofs", Proc. ASCE, 90,
ST4, pp. 103-121, August 1964.
Simo, J. C., Fox, D. D., and Rifai, M. S., "On A Stress Resultant Geometrically Exact Shell
Model. Part III: Computational Aspects of The Nonlinear Theory", Comput. Methods Appl.
Mech. Engrg., Vol. 79, pp. 21-70,1990.
Spillers, W. R., Automated Structural Analysis: An Introduction, Pergamon Press, New York,
1972. (This out of print text has been re-released in soft cover by Franklin Book Company,
Elkins Park, P A, 1992)
Spillers, W. R., "Geometric Stiffness Matrix for Space Frames", Computers and Structures,
36, 1, 1990, pp 29-37.
Spillers, W. R., "A Note on Some Qualitative Results for Beams and Rods", Quarterly
Applied Mathematics, XLVIII, 3 September 1990, pp 575-580.
Spillers, W. R., Saadeghvaziri, A., and Luke A., "An Example of Three Dimensional Frame
Buckling", Computers and Structures, Vol. 47, No.3, pp 483-486, 1993.
Spillers, W. R., Schlogel, Mark, and Pilla, Dominick, "A Simple Membrane Finite Element",
Computers and Structures, 45, 1, 181-183, 1993.
Spillers, W. R., and Rashidi, S., "Member Stiffness for Three-Dimensional Beam-Columns",
Journal o/Structural Engineering, ASCE, 123,7,971-972, 1997.
Stubbs, N., and Fluss, H., "Modelling Fabric Reinforced Composites", Proc. 7th Canadian
Congo Aool. Mech., Sherbrooke, Quebec, pp. 113-114, 1979.
Systeml360 Scientific Subroutine Package, IBM Corp., White Plains,New York, 1970.
Timoshenko, S., Theory of Elastic Stabilitv, McGraw Hill Book Co.,New York, 1936.
Timoshenko, S., and Goodier, 1. N., Theory of Elasticity, McGraw Hill, NY 1951.
Zienkiewicz, O. c., The Finite Element Method, 3 rd edition, McGraw-Hill, NY, 1977.
APPENDIX 1
MEMBER STIFFNESS WHEN BEAM-COLUMN
EFFECTS ARE INCLUDED
--+x
Figure A1.1 Simply supported beam.
This appendix derives the (moment distribution stiffness) of a member when

beam-column effects are included for two cases: the pinned-fixed and the
fixed-fixed beams. Consider the statically determinate beam of Figure Al.I
which is subjected to a compressive load P and an end moment m. In this
case the internal bending moment at any point x can be written as
mx
M=-+Py (Al.I)
L
If the elastic constitutive equation, M = -Ely", is used it follows that
mx
- + Py = -Ely" (A.l.2)
L
In this case y is simply
y = A sin kx + B cos kx _ mx
PL
(Al.3)
e=~
EI
Using the boundary conditions:
y=O @ x=O => B=O

y=O @ x=L => A=m/(P sin kL)
244 GEOMETRICALLY NONLINEAR STRUCTURES
The slope at x=L is then
y' x=L = ; ( k cot kL - ~) (AlA)
and, therefore
BB =- ~ ( k cot kL - ~) (Al.5)
or
mL
BB =--a (Al.6)
3EI
where
a=~(_l - cotkL) (Al.7)

kL kL
The member stiffness which is the moment per unit rotation and is
denoted for this pinned case K rs, becomes
Krs = 3EI.~
L a
(Al.8)
This stiffness was used in Chapter 1 and is described in Figure 1.11 by the
curve marked "Compression. Far End Simply Supported".
The member stiffness for the fixed-fixed case will be derived via
superposition by supplying a moment at A to eliminate the rotation BA and
then adding the effect of this moment on the rotation at B. Figure Al.2
illustrates the process. BA is derived from {y't=o as
B - ml fJ (Al.9)
A - 6EI
where
(Al.10)
MEMBER STIFFNESS WHEN BEAM-COLUMN EFFECTS ARE INCLUDED 245
A B ~
~-----~
e _ mL A mL
eB=-o.
A - 6EII-' 3EI
e '= mL A e '= mLJ32

A 6EII-' B 12Elo.
mJ3( A~--_~ B P
~'.E,~-
mJ3( A B ~, p
2c;~<' _____ ~r y -
zero slope eF = mL (40.2 - 13 2 )

B 4EI 30.
I. L .,
Figure A1.2 Rotation via superposition.
The moment required to produce a zero slope at A is mf31(2a) with

an additional rotation 0B 1 , at B of
o '= _(mf3)~f3 (A 1. I I)
B 2a 6El
The total rotation for the fixed case is thus
OF = _ mL/3 2 + mLa = mL (4a 2 - /3 2 J (Al.12)

B 12Ela 3El 4El 3a
and the member stiffness becomes,

(A 1. 13)
This stiffness is described in Figure 1.11 by the curve marked "Compression.

Far End Fixed".
APPENDIX 2
DETERMINANTS
A structure has buckled when its system matrix has become singular or
IK KG I
E + = O. When that happens a zero will appear on the diagonal of the
system matrix during the process of Gaussian elimination. Since it is
unlikely to hit upon the exact buckling load when the applied load is
incremented in steps, it is most common to go past the buckling load which
will generally result in the determinant becoming negative. The computer
programs discussed in these notes monitor the diagonal terms of the system
matrix during the solution process looking for terms which are not positive.
The logic of doing so is the subject of this appendix.
Cramer's rule has the inverse of a matrix A as the transpose of the
matrix of the cofactors divided by the determinant. Similarly, the
determinant of the matrix A can be described in terms of its first row
IA I
elements alj and their cofactors 1j as
(See Aitkin, 1958.) This allows the determinant of an n x n matrix to be

defined inductively in terms of the determinant of a smaller matrix and
eventually only having to deal with a scalar.
There are two rules of determinants which are important here:
1. A determinant is unaltered in value when any row or column is added to a

constantmultiple of any other row or column.
2. The determinant of a triangular matrix is the product of its diagonal

elements.
It is clear from the definition of a determinant that if you divide any

row of a matrix by a constant, you change its determinant by this constant.
The point is that rule I implies that Gaussian elimination can be performed
in such a manner that the value of the determinant of the system matrix does
not change. Rule 2 implies that when the value of the determinant goes to
zero, a zero must appear along the diagonal at the end of the elimination
process.
APPENDIX 3
THE ROTATION MATRIX
Geometric modeling, including the modeling of structures, typically involves
the concept of global and local coordinate systems (Figure A3.1). The global
coordinate system is of course the coordinate system in which the equations
of equilibrium are written and is fixed in space; the local coordinate system
is fixed in the member and describes the orientation of the member. Since
these two coordinate systems differ only by a rotation (if they are both
orthogonal) the rotation matrix becomes an important tool of structural
analysis.
Z
global
coordinate
~
system
X
x
,/'
...--x
z
Figure A3.1 Global and local coordinate systems.
It is most easy to deal with the two-dimensional rotation matrix

(Figure A3.2). From simple geometric arguments it follows that
Ax' = Ax cose + Ay sine
Ay' = -Ax sin e + Ay cose
or in matrix form
A'=RA
where
A'= { AX'} ; R= [cose sine] ; A= {Ax}

Ay' - sine cose Ay
By direct multiplication it follows that R T R = RR T = I .

250 GANALYSIS OF GEOMElRICALL Y NONLINEAR STRUCTURES
Figure A3.2 Two-dimensional rotation
There is a simple transition from two-dimensional rotation matrices

to three-dimensional rotation matrices. First of all, two-dimensional
rotations can be thought of as (three-dimensional) rotations about the z-axis.
Then
sine 01
cose 0 (A3.I)
o 1
By permutations of the labels of the coordinate axes it follows that rotations

about the x and y axes can be described as
o
sine
cose
One way to describe three-dimensional rotations is through the use of

compound rotations. A compound rotation is simply a sequence of rotations.
This description is particularly useful when the rotations in the sequence are
rotations about coordinate axes (which are essentially two-dimensional
rotations). In this case the use of compound rotations has the effect of
reducing three-dimensional rotations to two-dimensional rotations. When
rotations are performed in sequence, the composite rotation matrix is the
product of the individual rotation matrices used. For example, let a rotation
THE ROTATION MATRIX 251
be described physically as a sequence in which rotation 1 => R J ,

rotation 2 => R 2 , rotation 3 => R3 then
and
which implies that
This gives a way to construct the rotation matrix using rotations about
coordinate axes: That is, describe the physical rotation as a sequence of
rotations about coordinate axes and then form the rotation matrix as the
product of the rotation matrices described above.
Another approach to the rotation matrix comes from the fact that
"The rows of the rotation matrix are the base vectors

(coordinate unit vectors) of the local coordinate system /I
This result requires the fact that
"The transpose o/the rotation matrix is its inverse, i.e. R T = R- 1 /I
To show the latter is an easy matter. By definition length does not change
under rotation so neither does the square of the length. Ifx'=Rx then
or
Since this is true for arbitrary x it follows that

252 GANALYSIS OF GEOMETRICALLY NONLINEAR STRUCTURES
or
It is now possible to show that the rows ofR are the base vectors of the local
coordinate system. Clearly
x'=Rx
and
x=RTx'
Let
X.=(:l=i'
Since x = R T x' it follows that
I{i'}
Repeating this procedure for the other two coordinate axes it follows that
(i')y
(r)y (i't
(rt = r (A3.2)
(k')y (k')z k'
Finally some topics which will be returned to when three-dimensional

frames are discussed. First of all, it is clear that finite rotations do not behave
as vectors since the order in which rotations are applied makes a difference,
(see Figure A3.3). Clearly to describe the finite rotation of an object it is
necessary to use a rotation matrix rather than a rotation vector. What then is
the rotation vector used in linear elastic analysis? It will be shown now that
the rotation matrix is equivalent to a "vector-like" quantity. It is then the
Taylor series representation of this quantity which gives rise to the small
rotation vector of linear mechanics.
First, note that the rotation matrix which is used to describe the
relationship between the local and the global coordinate systems can be used
in kinematics to describe rigid body rotation. To do so, the local coordinate
system can be regarded as fixed in a rotating rigid body. Let xoJd and xnew
refer to the coordinates of a point before and after rotation, (see Figure
A3.4). It follows that Euler's theorem of rigid body motion implies that any
finite rotation (or sequence of rotations) can be described as a simple
rotation of magnitude a about some fixed axis described by a unit vector
Following Noble, (Noble, 1969, p. 421),

This implies that the rotation matrix R may be written in terms of a
and the components of n as
T
xnew =R xold (A3.3)
Xnew = xold - {1- cos a }{Xold - {X old . n *)n *} + sin a{n * XXold}
x
Lex
z z
y
y z
90 deg. about -90 deg. about
zaxis yaxis
z x
~y
z
y x
-90 deg. about 90 deg. about
yaxis z axis
x
y
Figure A3.3 Finite rotations A+B:;t:B+A.

254 GANALYSIS OF GEOMETRICALLY NONLINEAR STRUCTIJRES
*2
cnxn z* + ny* sin a
* *
cosa + cn x Cnxny - n z* sin a *
*2
RT = cnxny
* *
+ n z* sina cosa +cn y * *
cnyn *
z - nx sin a
y + nx• sina
* * * * * .2
cnzn x -ny sina cnzn cosa + cn z
where
c = l-cosa
Furthermore as Noble indicates, given R it IS a simple matter to
reconstruct n * and a since
(A3.4)
and
n *x''n*y''n*z -R T RT T T
- 23 - 32''R 31 - RT
13''R 12 - RT
21
It follows that any rigid body motion may be described equivalently using
either a rotation matrix R or the above representation which involves n* and
a.
Figure A3.4 Rotation about an axis.

Let
O=n*a (A3.5)
be called a rotation. If O(x) possess a Taylor series expansion, then

dO = VO . dx. It is this dO which is referred to as small rotation vector of
linear structural analysis.
APPENDIX 4
PERTURBATION METHODS APPLIED TO PLANE
BEAMS
This appendix examines the special case of pertrubation methods applied to

plane beams. The more general case of three-dimensional beams is
addressed as a separate chapter. The results obtained here are simply the
beam-column equations commonly found in discussions of strength of
materials. This appendix is included in an attempt to show how the
perturbation method works for a simple case of continuous systems.
The starting point is the vector form of the equations of equilibrium
for a three-dimensional beam
P'+p=O
(A4.l)
M'+m+txP=O
Here P, M = force and moment stress resultant vectors respectively; p, m =

force and moment applied loads respectively. For plane beams these
equations take on the form
Px' + px =0; Py '+p y =0·, (A4.2)
For an initially straight beam, t =iO . This reduces Eq. A4.2 to
pO'+pO
x
=0.
x'
pO'+p
y
°=0·
y'
MO'+mo
z z
+poy=0·
' (A4.3)
When written in their perturbed form, where iI = i ° + ro x i ° and

t
jl = + ro x t
and ro = y'kO, Eq. A4.1 appear as
~[(p~ +E~Jio + Y'jo)+(P~ +~)(jo - Y'i)] (A4.4)

dx
+ (p~ + sPx)(io + y' jO) + (p~ + EPy)(j0 - y'i) = 0
where the perturbed quantities such as the thrust appear as the zero or initial
solution plus the perturbation written as E times a term with a bar over it i.e.
pO ~ pI = pO + EP
MO ~MI =MO +EM (A4.6)
i O ~il =i O +ei;jO ~jl =jO +EJ;k O ~kl =kO +Ek
Setting p~ = P~ = m~ = Px = Py = m z = 0 as of no interest and c=l, and

collecting first order terms it follows that
(~-p~ y') =0
(~ + Pxoy') = 0 (A4.7)
Mz'+Py =0
The second and third of these equations can be combined (with the
perturbation parameter set to 1) to give the well-known beam column
equations
M z "-pOy"=O
x (A4.8)
The first equation which describes the perturbed thrust is uncoupled from the
other two.
APPENDIX 5
INTRODUCTION TO THE COMPUTER PROGRAMS
AS.l Introduction
It is the contention of the authors that the best use of this book will be made
by those who become actively involved with the computer programs
included in it. Actively involved can mean many things. At the best level, the
reader will use the work here as a springboard into other things of his or her
interest, perhaps even finding something that will help with an ongoing
project. On a more casual level, the reader my simply come to understand
nonlinear effects more clearly.
The way we learn is a complex matter. Can you learn something by
simply reading it once? How many hours must be spent to comprehend a
page of difficult material? What IS difficult material? How much of a
concept is it necessary to understand in order to be able to apply the
concept? If canned computer programs exist for structural analysis, how
much do you need to know in order to be able to run them? What should a
structural engineer know? Without answering any of these questions, it is
our contention that a structural engineer can learn much from simple
programming even if he or she has no intention of programming for real
applications. This is particularly true in the case of nonlinear analysis which
is inherently more difficult than the linear analysis with which we are now
so familiar.
If the reader is to understand the computer programs of this text, it is
first necessary to understand the computer programs of linear structural
analysis. This appendix discusses two of the most simple computer programs
of linear structural analysis, the space truss and the plane frame. These
programs are listed in full here whereas the other computer programs
discussed must be printed by the reader from the disk supplied with the text.
AS.2 Space Trusses
The space truss program listed here has three parts like most linear analysis
computer programs which use the displacement or node method: the system
matrix is set up, it is solved for displacements given loads, and finally
internal forces and stresses are computed.
In more detail, following input, the system matrix C ( I , J) is
zeroed and then formed by adding the contribution of each member to it. In
the case of bars with two ends, each member contributes at most four matrix
terms to the system matrix. In the case of the truss, each of these terms has
the form of
and all that remains is to determine exactly where these terms go in the
matrix C. That is done by the subroutine INSERT. In the program shown
here, these terms require unit vector components which are provided by the
subroutine UNITV.
In the center of the listing is a section of computer code which
solves the system matrix given the joint loads. This code is simply Gaussian
elimination: In the first half, rows of C are combined to eliminate terms
below the diagonal; in the second half a triangular system of equations (back
substitution) is solved. Like most linear equation solvers this code is entered
with a description of the system matrix and the right hand side of the
equations; at time of exit, the answers (the node displacements) are found in
the array which originally contained the right hand side (the joint loads).
Note also that this code does not include pivoting. There are two reasons for
this: First, a zero along the diagonal of the system matrix would imply a
singular (unstable) joint in the structure which can not occur; second, the
equations of the node method are typically well-conditioned.
Once the node displacements have been computed, the member
forces can be computed, bar by bar as
A·E
F. =_'_LJ·
I L. I
I
(It is left to the reader to determine the details, for example, of where each of
the above terms is to be placed in the various arrays.)
A5.3 Plane Frames
Once one computer program for linear structural analysis has been written,
others follow directly. The plane frame is a case in point. Like the three-
dimensional truss it has three degrees of freedom per node, it requires four
matrix terms in the system matrix per member, it uses the same equation
solver, but is a little more complex to deal with than the truss.
The difficulties of the plane frame arise from the fact that its
members are beams rather than rods. This fact also drives the introduction of
the concept of a local coordinate system and the use of the rotation matrix. In
terms of programming, the impact of these differences is simply that the
contribution of each member to the system matrix is left as a matrix product
INTRODUCTION TO COMPUTER PROGRAMS 261
T +T +
Ri Nj KiNjR i
rather than being written out explicitly. This, incidentally, makes it an easy
matter to modify the stiffness matrix which now is set up for straight,
uniform elements. The subroutine RNK generates the matrices just
mentioned and then PROD3 performs the required matrix multiplications.
Otherwise, this program uses much from the space truss program.
AS.4 Listing of TR3D.FOR

C SPACE TRUSS
DIMENSION NP(100),NM(100),S(100)
DOUBLE PRECISION R(100),P(100),C(100,100),UVEC(3)
1,C1,D1,F1,F2,FAC
MAXC=100
C
C INITIALIZE PARAMETERS/ARRAYS
C
E = 30.0D06
100 READ(5,150)NB,NN,NS
150 FORMAT (3(I4,3X))
WRITE(6,1)NB,NN,NS
1 FORMAT (I5, , NO. MEMBERS'/I5,' NO. NODES'/I5,
l' NO.SUPPORTS'//)
READ(5,156) (R(3*K-2),R(3*K-1),R(3*K),P(3*K-2),
1P(3*K-1),P(3*K),K=1,NN)
WRITE (6, 157) (K,R(3*K-2) ,R(3*K-1) ,R(3*K) ,P(3*K-2),
1P(3*K-1),P(3*K),K=1,NN)
157 FORMAT (lH1,25X,11HCOORDINATES,40X,5HLOADS//
114X,lHX,19X,lHY,19X,lHZ,18X,2HPX,18X,2HPY,
118X,2HPZ//(I4,6D20.8))
NNN = NN - NS
N=3*NNN
C
C SET UP SYSTEM MATRIX
C
DO 30 I = 1,N
DO 30 J = 1,N
30 C(I,J) = O.
WRITE(6,159)
DO 999 L=l,NB
READ(5,151)NP(L),NM(L),S(L)
WRITE(6,160)L,NP(L),NM(L),S(L)
151 FORMAT (2I5,8X,E10.6)
160 FORMAT (3I10,E20.8)
K = 3*NP (L)
M = 3*NM (L)
CALL UNITV(K,M,Cl,UVEC,R)
CALL INSERT(C,K,M,UVEC,MAXC,N,E,S(L),Cl)
999 CONTINUE
C
C SOLVE FOR JOINT DISPLACEMENTS
C
M= N - 1
DO 17 I 1,M
L = I + 1
DO 17 J = L,N
IF(C(J,I)) 19,17,19
19 DO 18 K = L,N
18 C(J,K) = C(J,K) - C(I,K)*C(J,I)/C(I,I)
P(J) = P(J) - P(I)*C(J,I)/C(I,I)
17 CONTINUE
P(N) = P(N)/C(N,N)
DO 20 I I,M
K = N - I
L = K + 1
DO 21 J = L,N
21 P(K) = P(K) - P(J)*C(K,J)
P(K) = P(K)/C(K,K)
20 CONTINUE
WRITE(6,161) (I,P(3*I-2) ,P(3*I-1) ,P(3*I) ,I=l,NNN)
161 FORMAT (lH1,13HDISPLACEMENTS/20X,lHX,19X,lHY,19X,lHZ
1//(IlO,3D20.8) )
WRITE(6,162)
162 FORMAT(lH1,3X,6HMEMBER,9X,2HDL,17X,5HFORCE,
1 14X,6HSTRESS//)
C
C COMPUTE MEMBER FORCES AND DISPLACEMENTS
C
DO 998 I=l,NB
K = 3*NP(I)
M = 3*NM (I)
CALL UNITV(K,M,C1,UVEC,R)
K1=K
D1=0.
FAC=l.
DO 997 J=1,2
IF(K1.GT.N) GO TO 996
D1=D1+FAC*(P(Kl-2)*UVEC(1)+P(K1-1)*UVEC(2)+P(K1)*UVEC(3))
996 FAC=-l.
K1=M
997 CONTINUE
F1=D1*E*S(I)/C1
F2=F1/S(I)
WRITE(6,1000) I,D1,F1,F2
998 CONTINUE
GO TO 100
156 FORMAT (8X,6F11.6)
1000 FORMAT (I10,3D20.8)
159 FORMAT (lH1,3X,6HMEMBER,5X,5H+ END,5X,5H-
END, 6X, 4HAREA/ /)
END
C
SUBROUTINE UNITV(K,M,C1,UVEC,R)
DOUBLE PRECISION R(1),C1,UVEC(3)
C1=0.
DO 1 1=1,3
UVEC(I)=R(K+I-3)-R(M+I-3)
1 C1=C1+UVEC(I) **2
C1=DSQRT (C1)
DO 2 1=1,3
2 UVEC(I)=UVEC(I)/C1
RETURN
END
C
SUBROUTINE INSERT{C,K,M,UVEC,MAXC,N,E,S,C1)
DOUBLE PRECISION C{MAXC,MAXC),UVEC{3),C1
K1=K
DO 1 1=1,2
IF{K1.GT.N) GO TO 1
M1=K
DO 2 J=1,2
IF{M1.GT.N) GO TO 2
FAC=l.
IF{I.NE.J) FAC=-l.
DO 3 L=1,3
I1=Kl-3+L
DO 3 L1=1,3
J1=Ml-3+L1
3 C{I1,J1)=C{I1,J1)+UVEC{L)*UVEC{L1)*S*E*FAC/C1
2 M1=M
1 K1=M
RETURN
END
AS.S Listing of FR2D.FOR

C PLANE FRAMES
DIMENSION A(100),AL{100),SI{100),TH{100),NP{100),MI{100)
DOUBLE PRECISION
P(100) ,C{100,100) ,R{3,3) ,SK{3,3) ,SNP{3,3)
1,SNM{3,3),AI{3,3),AJ{3,3),NRP{3,3),NRM{3,3),ANG
MAXC=100
C
C INITIALIZE PARAMETERS/ARRAYS
C
DO 6 1=1,3
DO 6 J=1,3
R{I,J)=O.
SK{I,J)=O.
SNP{I,J)=O.
6 SNM{I,J)=O.
R{3,3)=l.
SNP (I, 1) =l.
SNP{2,3)=1.
SNM ( 1 , 1) =-l.
SNM{3,3)=l.
PI=3.14159 /180.
E=29000000.
5 READ{5,2,END=999) NB,NN,NS
2 FORMAT (3I5)
WRITE{6,1)NB,NN,NS
1 FORMAT {I5, , NO. MEMBERS '/15, , NO. NODES'/I5,' NO.
SUPPORTS'//)
NNS=NN-NS
N=3*NNS
READ{5,3) ( P{3*I-2),P{3*I-1),P{3*I),I=1,NNS)
3 FORMAT (3D10.2)
WRITE (6, 903) (I, P (3*I-2) , P (3*I-1) , P (3*I) ,1=1, NNS )
903 FORMAT {lH1, 11HJOINT LOADS /13X,2HPX,18X,2HPY,19X,lHM//
1 (I4,3D20.8))
WRITE (6,901)
901 FORMAT(18H1MEMBER PROPERTIES
/11X,4HAREA, 15X, 6HLENGTH, 16X,lHI,
1 17X,5HANGLE,14X,5H+ END,5X,5H- END)
DO 904 I=l,N
DO 904 J=l,N
904 C(I,J)=O.
DO 926 K=l,NB
READ(5,31) A(K),AL(K),SI(K),TH(K),NP(K),MI(K)
31 FORMAT(4E10.2,2I5)
WRITE(6,900) K, A(K),AL(K),SI(K),TH(K),NP(K),MI(K)
900 FORMAT(I4,4E20.8,2I10)
ANG=TH(K)*PI
CALL RNK(R,SK,SNP,SNM,NRP,NRM,AL,SI,A,E,ANG,K)
IF(NP(K) .GT.NNS) GO TO 23
CALL PROD3(NRP,SK,NRP,AI)
CALL INSERT(C,MAXC, NP(K),NP(K),AI)
IF(MI(K) .GT.NNS) GO TO 926
CALL PROD3(NRP,SK,NRM,AI)
CALL INSERT(C,MAXC, NP(K),MI(K),AI)
DO 14 1=1,3
DO 14 J=1,3
14 AJ(I,J)=AI(J,I)
CALL INSERT(C,MAXC, MI(K),NP(K),AJ)
23 CALL PROD3(NRM,SK,NRM,AI)
CALL INSERT(C,MAXC, MI(K),MI(K),AI)
926 CONTINUE
C
C SOLVE FOR DISPLACEMENTS
C
927 M=N-1
DO 91 I=l,M
L=I+1
DO 91 J=L,N
IF (C(J,I)) 93,91,93
93 DO 92 K=L,N
92 C(J,K)=C(J,K)-C(I,K)*C(J,I)/C(I,I)
P(J)=P(J)-P(I) *C(J,I)/C(I,I)
91 CONTINUE
P (N)=P(N)/C(N,N)
DO 94 I=l,M
K=N-I
L=K+1
DO 95 J=L,N
95 P(K)=P(K)-P (J)*C(K,J)
94 P (K)=P(K)/C(K,K)
WRITE (6,231) (I, P (3*1-2) , P (3*1-1), P (3*1) ,1=1, NNS
231 FORMAT(20H1JOINT DISPLACEMENTS
/13X, 2HDX, 16X, 2HDY, 16X,2HTH/
1 (I3,3X,3(D15.8,3X)))
C
C COMPUTE MEMBER FORCES AND DISPLACEMENTS
C
WRITE (6,230)
230 FORMAT(28H1MEMBER DISPLACEMENTS-
FORCES/13X, 2HDL, 15X,3HAL+, 16X,
1 3HAL-,16X, 1HT, 16X,2HM+, 16X,2HM-)
DO 207 I=l,NB
ANG=TH(I)*PI
CALL RNK(R,SK,SNP,SNM,NRP,NRM,AL,SI,A,E,ANG,I)
DO 222 J=1,3
222 AI(2,J)=0.
IF(NP(I) .GT.NNS) GO TO 213
DO 208 J=1,3
DO 208 K=1,3
L4=3*NP(I)-3+K
208 AI(2,J)=AI(2,J)+NRP(J,K)*P(L4)
213 IF(MI(I) .GT.NNS) GO TO 215
DO 308 J=1,3
DO 308 K=1,3
L4=3*MI(I)-3+K
308 AI(2,J)=AI(2,J)+NRM(J,K)*P(L4)
215 DO 226 J=1,3
AI(l,J)=O.
DO 226 K=1,3
226 AI(l,J)= AI(1,J)+SK(J,K)*AI(2,K)
207 WRITE(6,217) I,AI(2,1),AI(2,2),AI(2,3), AI(l,l),
6 AI(1,2), AI(1,3)
217 FORMAT (I3,3X,6(D15.8,3X))
GO TO 5
999 STOP
END
C
SUBROUTINE RNK(R,SK,SNP,SNM,NRP,NRM,AL,SI,A,E,ANG,K)
DIMENSION AL(l),SI(l),A(l)
DOUBLE PRECISION
R(3,3) ,SK(3,3) ,SNP(3,3) ,SNM(3,3) ,NRP(3,3),
1 NRM ( 3 , 3) , ANG
R(l,l)=DCOS(ANG)
R(2,2)=R(1,1)
R(1,2)=DSIN(ANG)
R(2,1)=-R(1,2)
SK(l,l)=E*A(K)/AL(K)
SK(2,2)=E*4.*SI(K)/AL(K)
SK(3,3)=SK(2,2)
SK(2,3)=SK(2,2)*.5
SK(3,2)=SK(2,3)
SNP(2,2)=-1./AL(K)
SNP(3,2)=SNP(2,2)
SNM(2,2)=1./AL(K)
SNM(3,2)=SNM(2,2)
DO 1 1=1,3
DO 1 J=1,3
NRP(I,J)=O.
NRM(I,J)=O.
DO 1 L=1,3
NRP(I,J)=NRP(I,J)+SNP(I,L)*R(L,J)
1 NRM(I,J)=NRM(I,J)+SNM(I,L)*R(L,J)
RETURN
END
C
SUBROUTINE INSERT(C,MAXC,I1,J1,A)
DOUBLE PRECISION C(MAXC,MAXC),A(3,3)
DO 1 1=1,3
I2=3*I1-3+I
DO 1 J=1,3
J2=3*Jl-3+J
1 C(I2,J2)=C(I2,J2)+A(I,J)
RETURN
END
C
SUBROUTINE PROD3(A,B,C,D)
DOUBLE PRECISION A(3,3),B(3,3),C(3,3),D(3,3)
DO 1 1=1,3
DO 1 J=1,3
D(I,J)=O.
DO 1 K=1,3
DO 1 L=1,3
1 D(I,J)=D(I,J)+A(K,I)*B(K,L)*C(L,J)
RETURN
END
APPENDIX 6
GRAPHICS ON A PC
A6.1 Introduction
Computer graphics can be fun to use when computing. It can also be very
useful when you are attempting to check input data and it can be
indispensable when dealing with three-dimensional visualization of the type
required for fabric structures and cable nets. Unfortunately, graphics code is
the least portable code of all.
In order to get started with graphics, it is instructive to compile and
run a simple program on the disk called PLOTDEMO.FOR. That is done on
our system using the (Microsoft FORTRAN) commands
FL32 PLOTDEMO.FOR
PLOTDEMO
It should be noted that the use of graphics requires loading a graphics library
in addition to the usual FORTRAN libraries. With the compiler used above,
these libraries are loaded automatically. Other versions of FORTRAN
require a specific reference to this library. For example, on some compilers
the command is
FL PLOTDEMO.FOR /LINK GRAPlllCS.LIB

PLOTDEMO
If the graphics libraries have not been loaded properly, the system will be
quite vocal about so informing you.
PLOTDEMO.FOR generates a three-dimensional figure which can
be rotated using keyboard commands. It generates this figure with and
without node numbers.
If this program will not compile, there can be several reasons. The
most likely one has to do with the use of fonts when writing letters and
numbers. Our program uses the command
FONTPATH= '\F32\LIB\COURB.FON'
This statement implies that we are running our programs in our root
directory and that the fonts reside in a file called 'COURB.FON' which can
be reached through the commands 'CD F32' and 'CD LIB'. If PLOTDEMO
does not run on your system you may have to change this statement.
Otherwise you may have to redo the command
DUMMY=SETVIDEOMODE($VRES16COLOR)
The program PLOTDEMO.FOR has two pieces. Its main program simply
generates the data for the three-dimensional structure which is displayed.
(Once the structure has been displayed, it is a straightforward matter to read
the code which generated it.) The main program subsequently calls plotting
routines.
There are two plotting routines. The main program calls a subroutine
SPLOT which in turn calls a subroutine PLOT. SPLOT is a more or less
portable subroutine which rotates a three-dimensional object. PLOT is a very
system dependent subroutine which produces the actual drawing on the
CRT.
A6.2 Plotting in 2-D
Any plot on a flat surface is of course a two-dimensional drawing. Three-

dimensional drawings are produced here by first rotating a three-dimensional
object to obtain a desired view of it and then plotting the x-y plane in the
rotated coordinate system. Only stick figures which can be described by
points and lines are considered here.
Figure A.I gives some of the basic information about creating two-
dimensional plots on a PC. First of all the image to be drawn has its own
coordinate system which must be transformed so that this image will appear
on the screen when drawn using the coordinate system of the screen. In the
case of the subroutine PLOT, the transformation is simply a shift and a
scaling.
Otherwise there is system overhead to be dealt with. That is, when
you have a line to draw or a piece of text to write, in addition to the draw and
write commands, many other commands must be given to "set up the
system". These are called "overhead" here since they add nothing to the
plotting process and only enable the system; they are more or less
recognizable do to their otherwise lack of function. The subroutine PLOT
scales a two-dimensional picture and draws it on the CRT line by line. The
subroutine SPLOT performs the required three-dimensional rotations (see
Appendix 3) using the idea of a compound rotation.
With regard to detail, in order to plot on the face of the CRT the
main program calls a subroutine SPLOT as
GRAPHICS ON A PC 269
CALL SPLOT(NP,MI,NN,NB,R,FOR,IWRITE)
with
NP,MI - arrays which describe member node numbers

NN - number of nodes
NB - number of bars or members
R - coordinate array
FOR - array of member forces
IWRITE - describes text features
IWRITE=O no text
IWRITE=l shows member forces
IWRITE=2 shows node map
Subroutine SPLOT generates the new x-y coordinates of all points

after the coordinate system has been rotated to establish a point of view. It
begins by setting the rotation matrix ROT to be the identity matrix which
subsequently produces an initial top view of the structure. It then sets the
rotation increment DTH to be 10 degrees. It finally forms the rotation matrix
as the product of rotations about coordinate axes and computes new plane
coordinates RXY which are subsequently plotted by the subroutine PLOT.
Hard copy can be obtained using the PRINTSCREEN key.
It may be noted that the code for SPLOT comes primarily from the
three-dimensional frame program which also uses a rotation matrix.
PLOTDEMO has utility beyond the truss which is displayed.
PLOT3D is a program which is created from it by replacing the main
program with read routines. In this form PLOT3D can be used to display
data sets. For example, the command
PLOT3D TR2DNL.DAT
will produce a picture of the structure described by the data set

TR2DNL.DAT.
A6.3 Drawing Lines in 2-D
Coordinate Systems
• Real coordinates - describe the object to be drawn
• Screen coordinates(draw) - are used to draw the picture on the screen of
the CRT
y
object screen(draw) screen(text)
Figure A 7.1 Three coordinate systems of a picture"
• Screen coordinates(text) - are used to write text on the screen when fonts
are not used
Real coordinates must be transformed so that the picture will appear on the
screen:
Xscreen = (Xreal - Xshift) / scale

Yscreen = (y real - Yshift) / scale
Pick "scale" to make the object fit screen
scale = max((x max - X min) / 640; (Ymax - Ymin)/480)
Pick Xshift' Yshift to center object in screen
x sh"ft =
Xmax + Xmin 640 I
- - - x sea e
I 2 2
Y " = Ymax + Ymin _ 480 x scale
shift 2 2
INDEX
Ahmad, S., 203 error, 43

ANSYS,76 Euler buckling, 7
Argyris, J. H., 130, 203 Fluss, H., 129
Balourchi, S., 203 force density method, 130
Bathe, K. J., 203, 204 frames
Batoz, J. L., 204 Plane frames (Chapter 4),63
Battini, J.-M., 235 three dimensional (Chapter 5),
beam-column, 8, 167 85ff
three-dimensional (Chapter 8), fundamental theorem, 16
162 geometric instability, 1
Bergan, P. G., 203 geometric stiffness matrix
Beuchter, N., 204 trusses, 37
Bieniek, M., 169 . plane frames, 70
Biot, M. A., 1,45, 162 space frames, 94
buckling (Section 1.5), 5ff membranes, 127
exact (nonlinear), 13 shells, 204
linear buckling load factor, 15ff Gal, E., 194, 197, 199,203
overall, 40ff Gere, James M., 8, 10, 162
snap-through, 6 Goldstein, Herbert, 97
thermal,7 gradient matrix, 37
torsional, 99, 167, 171 Green, A. E., 163
Chan, 91 grid method, 132
Chen, W. F., 163 Ho, L. W., 203
Chou, Jung-Hua, 91 Horrigmoe, G., 203
Clough, R. W., 203 Hsiao, K M., 203
condition number, 41 hyperbolic paraboloid example,
convergence 143
quadratic (Newton's method), 3 Ibrahimbegovi6, A., 239
deformed shape Irons, B. M., 203
method of, 130 Isaacson, E., 41
Crisfield, M. A., 203 Izzuddin, B. A., 91
dome example, 29 Johnson, C. 1., 203
Dvorkin, E. N., 204 Kapania, R. K, 203,
Eidelman, J., 132 Keller, H. B., 41
eigenvalue buckling, 10ff, 41 ff Kirkhoff
Eisenberger, 172 DKT plate F.E., 204, 206
Elnashai, A. S., 91 Knops, R. J., 163
Krysl, Petr, 91 Shells, (chapter 9), 203

Kuo,Shyh-Rong,91 skylight example, 135
lack of fit, 35 smoothing, 13 2
Laws, N., 163 Siev, A., 132
Leonard, John William, 129 Simo, J. C., 233
Levy, R., 14,31,35,62, 194, 197, Spillers, W. R., 20 99,
199,203 163,188,194
linear buckling load factor, 15ff strain
linear structural analysis (Chapter small,3ff
2),20 spurious, 4
Levy, S., 221 stress stiffening, 4
Lui, E. M., 163 string effects, 11
MacBain, K., 188 Stubbs, N., 129
material nonlinearities, 1 symmetry, 82
Mescall, J. F., 232 thermal buckling, 7
McConnell, Richard E., 172 thermal stresses, 35
Mohan, P., 203 Timoshenko, S., 5,49, 73, 162
moment distribution, 8 trusses
Newton's method, 2 exact analysis (Chapter 3), 36ff
nonlinear analysis, 43 linear analysis (Chapter 2), 20ff
quadratic convergence, 3 geometric stiffness matrix, 12
NROOT,62 Yang, Yeong-Bin, 91
Otto, Frei, 129 Zienkiewicz, O. c., 131,204
Pacoste, C., 235
patterning, 156
Peng, X., 203
Penning, F. A., 232
perturbation, 36
potential, 82
prestress
fundamental theorem, 15ff
"lack of fit" ,35
Ramm, E., 204
Rashidi, S., 194
rotation
matrix, 65, 88
relative, 93
pure, 203, 212
Saadeghvaziri, A., 188
Schek, H. J., 130
See, Thomas, 172

Bok:978 94 017 0243 0 PDF

Uploaded by

Copyright:

Available Formats

Bok:978 94 017 0243 0 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

Bok:978 94 017 0243 0 PDF

Uploaded by

Copyright:

Available Formats

Analysis of Geometrically Nonlinear Structures

Springer-Science+Business Media, B.V.

ISBN 978-90-481-6438-7 ISBN 978-94-017-0243-0 (eBook)

Printed on acid-free paper

All Rights Reserved

)Inc! tlie £ana liac! rest from war

To Lydia, Itai and Shelly

To the late Jewell Garrelts of

Chapter 2 Linear Structural Analysis 23

Chapter 3 "Exact" Analysis of Trusses 41

3.5 Computer Programs 49

Chapter 4 Nonlinear Analysis of Plane Frames 71

Chapter 5 Nonlinear Analysis of Space Frames 93

Chapter 6 Nonlinear Analysis of Membranes 121

6.3 Three Dimensional Members 127

Chapter 7 Cablenets and Fabric Structures 151

Chapter 8 Three-Dimensional Beam-Columns 187

Chapter 9 Nonlinear Analysis of Shells 205

Appendix 1 Member Stiffness When Beam-Column Effects are

Appendix 2 Determinants 247

Appendix 3 The Rotation Matrix 249

Appendix 4 Perturbation Methods Applied to Plane Beams 257

Appendix 5 Introduction to Computer Programs 259

AS.2 Space Trusses 259

Appendix 6 Graphics on a PC 267

where P represents the applied loads, F represents the member forces or

Preface to the Second Edition

With the exception of new material on three-dimensional beams

pes Versus Workstations. The computer programs associated with

Computer Graphics. Computer graphics is fun but also a serious

Hardware. The programs were run on a Pentium 4, 1.9GHz with

Practical Applications. This book is written more for use as a

Contact With the Authors. We would be pleased to receive E-mail

There are a number of practical phenomena in structures which simply can

Figure 1.1 Geometrically unstable structure.

Elementary theory is a strange mix in which behavior is described in

1.2 Newton's Method

Central to the solution process used in this book is Newton's method. It is

r{X) = 0 ~ r{xo)+ vr{xo)· dx ~ 0 (Ll)

Figure 1.2 Newton's method.

For ::;; 1 it follows that

which implies quadratic convergence in the neighborhood of x*. That is

1.3 Restrictions of Small Strain

displaced position L *" ." .,.

c initial pOSition (linear theory)

Figure 1.3 A truss bar. Approximation oflinear theory.

1.4 Stress Stiffening

effect joint free body diagram

Figure 1.5 Joint equilibrium.

The "geometric effect" is the change dT in the internal force in

Here k is of course the spring constant. If e = d / L, Eq. 1.9 becomes

Figure 1.6 Buckling example.

1.5.1 Snap Through

Figure 1.7 Snap-through buckling.

• Compute P using the deformed equilibrium (P = 2F sin g )

1.5.2. Thermal Buckling

truss bar weight truss bar

The analysis of this buckling problem can be carried out in the

1.5.3. Euler's column