Agard Ag 72

S
AGARDograph 72
EFT.
ID
1 0 APR 1964
Matrix Methods
of Structural
Analysis
Editor*.
F. de VEUBEKE
University of Liege
Pergamon Press
ACARDograph 69
A Correlation Study
of Methods of Matrix
Structural Analysis
Report to the 14th Meeting. Structures and
Materials Panel. Advisor) Group far Aero-
nautical Research and Development.
NATO AGARDograph 72
R. H . G a l l a g h e r , Bell Aeroivslems Company, Buffalo.

New York
This book presents the result of a survey

Matrix Methods of
and review of matrix structural analysis
as practised in the airframe industry.
Structural Analysis
Concepts of the force and displacement
approaches are described, and areas of
agreement between published methods Edited by F. de Veubeke, u-ivmity ofLUt*
that are within the framework of these
approaches are delineated. Techniques
for determining the force-displacement This is a collection of papers presented to
properties of discrete elements employed the Structures and Materials Panel of
in the analytical idealizations of structures AGARD.
are reviewed. The use and limitations of Besides the progress reported in the linear
these techniques are exemplified in the theory, the most conspicuous development
determination of properties for the quadri- is the penetration of matrix structural
lateral plate in direct stress. analysis in the field of large deflection
The report also includes descriptions and theory.
results of extensive analyses of multiweb The papers report progress and the current
low aspect ratio wings. The results of state of the art in the application of
past evaluative studies are taken as a matrix algebra and computer language
starting point in the selection of com- to the structural analysis of complicated
parative idealization possibilities for the structures. Recent advances covered
present analyses. Idealizations described in include the use of subroutines in the com-
past reports to the AGARD Structures puter programming, choice of redun-
and Materials Panel arc also selected. dancies, methods of obtaining upper and
The evaluation ofthe idealizations is speci- lower bounds to structural influence
fically based upon comparisons of the coefficients, treatment of non-linear prob-
results, in the form of displacement in- lems such as rigidity in the presence of
fluence coefficients, with each other and thermal stresses, the determination of
with the results of wing model tests. More buckling loads, and plastic and creep
general conclusions are then drawn as to deformations.
convergence and error sources in matrix
structural analysis.
The study concludes with a detailed
discussion of future research topics in
matrix structural and aeroelastic analysis.
70s. net
£5 net
S4-
" • " - 3 DZPT
r D
10 APR 1964
AGARDograph 72
M A T R I X METHODS OF
S T R U C T U R A L ANALYSIS
MATRIX M E T H O D S
OF
S T R U C T U R A L ANALYS 5IS
Edited by
P R O F E S S O R B. FRAEIJJ 5 de V E U B E K E
Univci'siu'-s cle Liege et dc Louvain
Belgium
Published for and on behalf of

ADVISORY GROUP FOR
A E R O N A U T I C A L R E S E A R C H AND D E V E L O P M E N T
N O R T H ATLANTIC; TREATY O R G A N I Z A T I O N
by
PERGAMON PRESS
Oxford • London • New York • Paris
1964
P E R G A M O N PRESS LTD.
Headington Hill Hall, Oxford
4 & 5 Fitzroy Square, London, W. 1
PERGAMON PRESS INC.
122 East 55.7i Street, New York 22, N.Y.
GAUTHIER-VTLLARS ED.
55 Quai des Grands-Augustins, Paris 6
PERGAMON PRESS G.m.b.H.
Kaiserstrasse 75, Frankfurt am Main
Distributed in the Western Hemisphere by

T H E MACMILLAN COMPANY • NEW YORK
pursuant to a special arrangement with
PERGAMON PRESS L I M I T E D
Copyright
©
1964
ADVISORY CROUP FOR
AERONAUTICAL RF.SKARC.lt AND DEVELOPMENT
N1PU III ATLANTIC TREATY ORGANIZATION
Library of Congress Card No. 63-21973
Set in Monotype liaskerville 11 on 12 /'(. and printed in Great Britain by

T H E PITMAN PRESS, BATH
CONTENTS
Matrix Methods of Structural Analysis—A Precis of Recent

Developments . . . . . . . . 1
J . H. ARC-YRIS, S. KEI.SEY, H. KAMEL (Institut fur Statik und
Dynamik der Flugkonstruktioncn) and Aeronautical Struc-
tures Laboratory, Imperial College—University of London
Upper and Lower Bounds in Matrix Structural Analysis . .165
B. M. FRAEIJS DE VEUBEKE, Universities of Liege and Louvain
Further Development and Applications of the Stiffness Method . 203
M. J . TURNER, H. C. MARTIN, R. C. W E I K E L
Some Problems in the Discrete Element Representation of Aircraft

Structures . . . . . . . . . 267
I. C. T A I O , R. I. K E R R
Digital Analysis of Nonlinear Structures by the Force Method . 317

P. H. DENKE
Author Index . . . . . . . . . 343

MATRIX METHODS OF STRUCTURAL
ANALYSIS
A PRECIS OF RECENT DEVELOPMENTS
J. H. ARGYRisf, S. K E L S E Y J , and H. KAMEI.§

Insiitm fiir Slatik und Dynainik der Flugkonslruktioncn
IVi hull al University, Stuttgart
and
Aeronautical Structure! Laboratory, Imperial College, University of London
CONTENTS
Pati
PREFACE . . . . . . . 3
A. FORCE METHOD . . . . . 1(1

a. General Theorv of the Force Method for Compl ex
Structure . . . . . 1(1
b. Primary Redundancies in the Fuselage . 11
c. Major Cut-Outs in the Regularized Structure 21
d. Experimental Work on a Fuselage p-)
e. Postscript to the Force Method; Flexibilities; Therma
Stresses . . . . . . 37
I. DISPLACEMENT METHOD . . . . Id
a. Idealization of Membrane Cover; Stiffnesses Id
I). Idealization of Spars; Stiffnesses . 55
c. Assembly of the K-matrix dl
a. The matrix assembly of K 62
ji. On the Regularization of the Matrix a 64
y. The "Long-Hand" Formation of K 68
d. Some Further Comments on the K Matrix 68
e. Experimental Work on a Wing Structure 71
f Professor and Head.

* Lecturer and Deputy Head.
§ Lecturer and Section Ix-ader.
j. H . A R G Y R I S , S. KELSEY and H . KAMEL
PaV
C. M A T R I X FORGE M E T H O D IN N O N - L I N E A R PROBLEMS . . 8 1
a. Introduction to an Elasto-Plastic Analysis . . .81
b. The Idealized Stress-Strain Relations . . . .81
c. The Generalized Initial Strains for Typical Elements . 84
a. Uniform flange under constant end load . . 84
p. Shear panel . . . . . . . 86
y. Uniform flange under linearly varying end load . 86
d. Determination of Stresses and Deformations . . . 8 8
e. Main Flow Diagram and Example . . . . 9 2
f. A Precis on Large-Displacement Analysis . . . 9 8
D. M A T R I X DISPLACEMENT M E T H O D IN N O N - L I N E A R PROBLEMS 105
a. Introduction and General Principles . . . .105
b. Determination of Geometrical Stiffnesses for Typical
Structural Elements . . . . . .112
c. Determination of Invariant Vector C*: Stresses; Natural
Stiffness 116
d. Buckling 120
APPENDIX I
O n the Stillness of Triangular Elements . . . . 120
Introduction . . . . . . . .120
a. Basic Geometry . . . . . . .122
b. Stiffness Matrix . . . . . . .126
APPENDIX II
Stiffness Matrix for an Arbitrary but PI me Quadri lateral
Panel 128
Introduction . . . . . 128
a. Basic Geometry and Kinematics 129
b. Strains and Stiffness Matrix 131
APPENDIX III
Stiffness Matrix of Spar Elements 135
Introduction . . . . 135
A. The Tapered Spar Element 136
a. Translational Bending . 137
b. Rotational Bending 138
c. Shearing 139
d. Stiffness Matrix 140
2
MATRIX METHODS OF STRUCTURAL ANALYSIS
Page
e. Strains in Spars . . . . . . .142

f. Comments . . . . . . . . 143
B. The "Exact" Stiffnesses of a Uniform Web . .144
a. Mode b t = 1 . . . . . . .145
b. Mode r2 = 1 146
c. Mode s2 = 1 . . . . . . . 147
d. Stiffness Matrix . . . . . .147
e. Comments . . . . . . . .147
APPENDIX IV
Stiffness Matrix of a Parallelogram Element . . . 148
APPENDIX V
On the Buckling and Large Displacement Theory of Elastic
Systems 152
I'Rl.l ACI.
THIS paper summarizes some of the progress achieved in developing

further the first matrix theory of structures, to be given in dual form
both with forces and displacements as unknowns. 1 Certain parts of
the current report were actually presented at the September 1959
meeting of the Structures Panel of AGARD in Aachen and included
some of the advances in the matrix force and displacement methods,
and their experimental verification effected in the period 1956-59. 2
Unfortunately the proceedings of the Conference were not issued and
no time was found to publish the main conclusions in another form.
Bearing this in mind we have interwoven some of the salient points of
our previous account in this paper which covers hence the most
important but to a considerable extent unpublished contributions of
the years 1956 to 1962. As far as the recent work from 1959 to 1962
is concerned, this was largely originated at the Institut fur Statik und
Dynamik der Flugkonstruktionen of the Technical University in
Stuttgart.
The range of subject matter covered here is now summarized below
under separate headings.
A. Matrix Force Method

Considerable development of the matrix force method has taken
place since 1955 which, although mainly centred on fuselage analysis,
is also pertinent to wings and indeed to any complex structure. Since
we have published in the last few years a fair number of papers on
fuselage theory, which are to be issued in an extended book form before
3
j. H. A R G Y R I S , s. KELSEY and H . KAMEL
long, it will suffice here if we restrict our account to the highlights of

the analysis and to its yet unpublished experimental confirmation.
But before discussing them it may be appropriate if we emphasize
once more a few aspects that are all too easily overlooked when writing
in general terms about the matrix force method. Thus, to express the
stress matrix in the (idealized) structure in the form (notation of Refs.
1 and 3),
b^bo-MVD,,, (P.l)
docs not require any particular insight into the theory of structures.
The real test of the practicability of a theory lies in the proposed
procedures leading, in the computer itself, to the matrices b 0 —the
static equivalent stress system—and h x —the self-equilibrating stress
systems. At the same time as we have pointed out repeatedly the
underlying redundant systems must not be understood as single-cut
forces but as generalized groups of forces applied over part or even
the complete cross-section. Also important is the formation of the
flexibility matrix f of the unassembled elements which enters into the
formation o f t h e matrices D n and Di 0 . A further significant aspect of
the theory emerges through the possibility—if not practical necessity
—of splitting the redundancies into distinct groups, denoted by us as
primary, secondary and tertiary ones, which allow the computations
for a complex structure to be divided in a number of steps corresponding
to distinct subdivisions of the structure and involving only one group
of the redundancies at a time. There is little doubt that a number of
publications have either ignored these issues or dealt with them
perfunctorily. Now once the idealization of the structure is carried out
in a systematic and consistent manner (another point to which all too
often not sufficient attention is paid) it is imperative to automatize
completely the procedure and to obtain the b 0 , b , and f matrices in
the computer starting from the simplest initial data, like coordinates,
cross-sectional areas, etc., input as column or row matrices. How
this is to be achieved for fuselages is shown in great detail in Ref. 3,
and the techniques developed there have all been applied with con-
siderable success both in our Institutes as also in industry. One aspect
of these techniques which we cannot help mentioning is their elegance
and simplicity. To anyone truly interested in structures—both from
the fundamental and practical points of view—it is always a revelation
how the elegant and concise approach nearly always proves the most
suitable one for practical applications, being probably most in harmony
with nature itself. In this connexion we draw the reader's attention to
one of the new methods which involves the determination of b„
and b j submatrices appertaining to a cross-section by a single inversion
of a simple matrix. Also very valuable in practice has proved the
partial orthogonalization technique of Rcf. 3, making use of standard
4
MATRIX METHODS OF S T R U C T U R A L ANALYSIS
subprogrammes for the determination of the eigenvectors of a sym-

metrical matrix and leading to any desired degree of excellence in the
conditioning of D n . However, all these techniques could not be
applied to their full advantage without the help of two further allied
devices which were initiated in 19551. These are the rcgularization
and cut-out techniques. In any complex structure, and this is particu-
larly true of aeronautical applications, no complete continuity of the
system can be achieved in practice, any number of minor or major
discontinuities or cut-outs interrupting the regular pattern of the
structure. Now, although the direct analysis of such systems on the
digital computer is in principle possible, it clearly cannot make use of
standard general-purpose programmes. As a result any automatization
of the exceedingly cumbersome and inelegant computations would
require an ab initio major effort in individual programming. All these
difficulties are circumvented by the beforementioned regularization
technique which transforms the given system into a regular one by
filling-in all missing members. Clearly, the resulting so-called regu-
larized (and idealized) system may easily be computed by a standard
procedure. Moreover, the complete analysis of this structure suffices,
in principle, to yield also the stresses and deformations of the dis-
continuous idealized structure. This is proved in Ref. 1 by the applica-
tion of the so-called cut-out or elimination technique which removes
effectively the supernumerary members. There is little doubt that this
procedure achieved a major breakthrough towards the rational analysis
of highly complex systems.
Part A of our present report summarizes these theories and demon-
strates, inter alia, on the evidence of comprehensive tests on a fuselage
with non-circular cross-section and cut-outs carried out at the Aero-
nautical Structures Laboratory of Imperial College, what remarkable
agreement can be achieved between theory and experiment.
It is fairly evident that the automatization techniques described
above are, in principle, also applicable to wings. In fact, this is
demonstrated in Refs. 3 and 4 on a particular example and these
procedures are at present evolved even further in conjunction with the
Boeing Transport Division to cover wings of any planform. In this
connexion it is necessary to drop the idealized concept of lumped
flanges and shear-carrying walls and to allow for full direct stress
carrying and Poisson's ratio effects in the sheet. It is hence imperative
to establish flexibility matrices for elements of arbitrary quadrilateral
shape, whose formulation is not simple. Wc only reproduce here the
flexibility matrix for parallelogram elements. Naturally, to account for
these refinements we have to introduce additional redundant systems,
and the point is soon reached where one of the significant advantages
of the matrix force method over the displacement method, arising
through a smaller number of unknowns, is lost. Under these conditions
5
the application of the matrix displacement method to wings and tail

units proves more advantageous, especially since arbitrary structural
configurations are much more amenable to a displacement than a
force analysis.
B. Matrix Displacement Method

As mentioned before, the original matrix theory of structures was
evolved in Ref. 1 in dual form both with forces and displacements as
unknowns. However, for all the fundamental similarity of the corre-
sponding formulae there exist some significant differences in the
detailed application of the methods. First of all, in practice duality is
not so immediately evident since forces and not displacements are, in
general, prescribed in both cases. It follows that when working with
displacements instead of forces as unknowns, we find it as a rule
convenient to introduce all nodal displacements as primary unknowns.
Although these unknowns are mostly selected as single nodal displace-
ments, more complex generalized displacement functions may also
prove fruitful and appear, in fact, as such—even if in a very elementary
form—in the spar analyses. However, two more significant practical
differences from the force method may be noted.
T h e first arises in the initial idealization process of the structure. It
is advantageous, as we have emphasized in many instances, to select
an idealized system which is most natural and accessible to the tech-
nique used. Thus, in the force method we may prescribe the nature
of the stress state, the simplest ensuing model being that of Wagner
involving effective or lumped flanges and pure shear-carrying walls.
But while this is a most appropriate presentation of many membrane
structures in the force approach, it is inconvenient and incongruous
in the displacement method and its enforcement would only prove a
lack of structural flair of the investigator. A much more sensible idea
in this case is to specify the variation of the displacements within the
elements and between adjacent nodal points, say by prescribing linear
or second order functions in the coordinates, and to deduce hence the
simplified strain and stress patterns. The elastic properties of the
elements which are, of course, measured by the stiffnesses corresponding
to the nodal displacements, may be determined concisely by the unit
displacement method developed in Ref. 1, which is dual to the unit
load method also presented there. In fact, the expressions for the
stiffnesses of membrane elements prove as a rule much more simple
than the corresponding ones for the flexibilities. This is strikingly-
illustrated in the case of the stiffness matrix for arbitrary but plane
quadrilaterals which is derived in Appendix II to this report; the
corresponding analysis for the flexibilities is much more complex and
less convenient from the computational point of view. At the same time
the most judicious choice for the elements in the cover of an arbitrary
6
wing or any irregular membrane structure analysed by the displace-

ment method is found to be the triangular panel which can claim two
distinct advantages. First, it yields trivially simple stiffness formulae
and secondly, it allows the most straightforward representation of a
tapered, swept and curved cover, since the three vertices of a triangle
must always lie in a plane. To achieve, on the other hand, an idealiza-
tion of such a cover with quadrilateral shapes requires, in general
non-planar elements which introduce considerable complications in the
analysis. It is nevertheless of some interest that the rectangular
elements of Rcf. 1 are—where applicable—for the same array of nodal
points, slightly more accurate in their estimate of stresses and deforma-
tions than the triangular ones.
But in contrast to the inherent simplicity of the analysis of cover
elements by the displacement method there arises some difficulty—
surprising at first sight—in specifying a sufficiently accurate kinematic
representation of the spar elements. The reader will remember that
we prescribed in Ref. 1 essentially the same deformation modes for the
spars as for the cover. They involve a pure shearing displacement
and a rotation of a spar station not entailing any vertical displacements.
Now it is fairly evident that in the limit as the spar elements become
shorter, our idealization leads necessarily to the exact result but—and
this is decisive in the context—the convergence ofthe procedure proves
very slow. The deflexions of the spars may hence be underestimated
seriously. This deficiency is clearly due to the neglect of the bending
of the individual elements. A different approach lo this problem was
adopted within the Turner Group by Melosh and Merritt who again
prescribed two modes which involve, however, both bending and
shearing. Essentially they obtained the spar stiffnesses from the concept
of flexibilities. Not unexpectedly this second technique yields for few
spar stations a more reliable estimate of the deflexions—but not
necessarily of the stresses in the cover. At the same time we note that
with increasing and more realistic number of spar stations the deflexions
by the M.-M. approach hardly seem to vary, thus conveying the
impression of an excellent convergence which is nevertheless fictitious.
In fact, onc observes that the deformations are once more under-
estimated and become, indeed, less accurate than those of Ref. 1. T o
cut short our comments, we may say that the difficulty in either of the
two methods arises at least impact from the imposition of two modes
only which do not suffice to describe with sufficient accuracy the
kinematic behaviour of the spars. For this reason wc have developed
a new approach to this problem, which specifics at present three modes,
two being pure bending deformations and onc involving pure shearing.
The investigation is reproduced in full in Appendix I I I to this report.
Good agreement with experimental deflexion results has been obtained.
This is not surprising since a single element would suffice, by virtue
7
j. H. ARGYRIS, s. K E I . S E Y and H. KAMEL
ofthe nature ofthe assumed modes, to yield the correct deflected shape
of a uniform cantilever. However, although a significant advance has
been achieved in the displacement method, as may be seen from the
comparison of the theoretical and experimental results given in Part B
of this report, some fundamental questions remain unanswered and
more modes than three may be required to obtain a still better answer
to the spar problem. At the same time we must be aware that the
more refined kinematic analysis of a spar induces some lack of
compatibility at the intcrnodal junction of covers and spars.
Having discussed our first point—that of the choice and specification
of kinematic modes in the displacement method—we turn to the second
point where a substantial difference from the force method is in practice
apparent. Thus, in computing the stresses from Eq. (P.l) we have to
set up the matrices D u and D 1 0 which are given by
D u = bjfbj and D 1 0 - bjfb,, (P.2)
and involve an elaborate pattern of matrix multiplications. No such
complexities arise in the displacement method. Here we have only to
set up the stiffness matrix K of the assembled structure and then to
compute the unknown displacements r (and hence stresses) from the
relation
R = K r or r - K *R (P.3)
where R are the prescribed loads in the directions ofthe displacements.
T o deduce K from the matrix k of the stiffnesses of the unassembled
elements, we may use—as first demonstrated in Ref. 1—the concise
congruent transformation
K a'ka (P.4)
where a is a matrix relating the displacements defining the behaviour
of individual elements to the general displacement vector laid down
for the complete structure. Now, by specifying the k matrix of the
elements immediately in the directions of r the matrix a is seen to
contain merely a few unit elements and a large number of zero elements.
T h e formation of K is hence reduced to the physically obvious opera-
tion of adding the elements of k into appropriate addresses of K
specified as to row and column position by the unit elements of a.
This individual addition of the stiffness elements has, in fact, been
proposed by other authors, who did not establish the matrix transform-
ation of Eq. (P.4). However, as described in Part B of this paper, the
matrix operation (P.4) is still useful if it entails solely the specification
or instruction within single words of the computer store of the row and
column positions of the unit elements. This proves to be particularly-
advantageous in conjunction with the idea of kinematic regularization
introduced here. T h e formation of K is achieved in this way very
speedily and elegantly by general-purpose programmes. As a matter
8
of interest the setting-up of the K matrix of dimensions 500 X 500

takes by this procedure less than 11 min on a computer like the
Remington Rand 1107. We are currently developing these ideas
further in conjunction with the Boeing Transport Division.
Part B of this report, and Appendices I to IV discuss in greater
detail some of the above issues and other related questions of the
displacement method. Also described is an experiment on a wing-type
structure which confirms satisfactorily our theories.
C. Matrix Force Method in Non-Linear Problems

The non-linearities dealt with under this heading are of two kinds:
those directly associated with the stress-strain properties of the struc-
tural material and those which arise when the deflexions are large
enough to cause significant changes in the structural geometry.
The non-linear clastic and plastic behaviour of elements is allowed
for within the framework of the standard matrix force method by
writing the stress-strain curve as
v = H + fS (P.5)
where H and f a r e functions o f S and, in the case of plastic straining,

depend also on the previous stress history. An iterative method of
solution is clearly indicated (at least in the present state of the art)
and the problem essentially resolves into the correct determination of
the initial strains H and the tangent flexibility f a t each stage of loading
or unloading. In the elasto-plastic analysis presented in Part C, the
stress-strain curve is simplified to a series of linear portions, which
representation is particularly suited to computer operations. The
theory is illustrated with reference lo the idealized flange/shear panel
model for which a unidimensional yield criterion suffices. The case of
partial yielding of a flange is dealt with, as is also the reversal of strain
in a plastically deformed element. Such an analysis can, of course,
only be developed in close conjunction with the associated computer
programme and, in fact, the flow diagram shown is as much a guide
to the former as to the latter.
Although the matrix force method is generally regarded as a small
displacement theory, it can be applied, with the help of some artifices,
to problems where the finite nature ofthe displacements can no longer
be neglected. These artifices allow for the geometrical changes by
introducing fictitious external forces to represent the modified equi-
librium conditions and also the useful and ubiquitous initial strains
which effectively correct the compatibility conditions for the deformed
structure. However, the analysis ofthe structure continues to be based
on its undeformed geometry. Since the fictitious external forces and
initial strains are non-linear functions ofthe displacements, the method
9
j. H. A R G Y R I S s. K E L S E Y and H. KAMEI.
of solution must be iterative and admits of various degrees of approxi-

mation. The first approximation yields essentially the classical (Euler)
type of buckling analysis. A simple example which exhibits a "snap-
through" type of instability requires a higher order approximation.
D. Non-Linear Matrix Displacement Method

In the final part ofthe report we give an extension and generalization
of the displacement method to deal with geometrical non-linearities.
In principle, a geometrical stiffness K, ; (like that of a stretched string)
is introduced, as well as the elastic stiffness K f ; , to take account of the
effect of displacements on the equilibrium conditions. The nature of
the displacement method permits the direct construction ofthe elastic
and geometrical stiffness matrices as functions of the displacements,
which are then obtained by an iteration procedure.
An iterative procedure, obtaining the sequences of approximations
r , . . . r„ for the displacements may be represented as follows:
K, - K , ( 0 ) - r i
K, = K E (r.) + K e ( r I ) - r i
K;1 - KA.(r8) + K„(r8) -> r 3
In the first iteration which gives the usual small displacement result
there is no geometric effect since no internal forces are known.
The theory is illustrated for structural elements consisting of spar
flanges and flat triangular panels.
The methods put forward in this communication are being currently
applied in a number of Research and Industrial Organizations in the
United Kingdom, United Slates, German Federal Republic and other
countries. A translation of our previous main publications on this
subject, covering the period 1954-59, has, moreover, appeared in
Russian.
A. FORCE METHOD
A. a. General Theory of the Force Method for a Complex Structure

For many types of structure, and particularly aircraft, a systematic
breakdown is possible into a series of major parts, each of which is
itself statically indeterminate and is attached to its neighbours in a
statically indeterminate manner. T h e fuselage, wing and tail unit are
the obvious constituent parts of a conventional aircraft, into which it
may be divided, although more subtle or comprehensive breakdowns
may be appropriate in unusual cases. Such a structure is, in the
present context, described as complex and the parts which together
10
form the complete structure or system are called substructures or

major components.
The analysis of a complex structure may, of course, be developed
at the most elementary level simply by ignoring its natural divisions
and regarding it as a very large system, which we reduce to a primitive
statically determinate basic system, and introducing a set of redun-
dancies which are all handled in the same way. This approach,
however, takes no account of the fact that the computational analysis
must inevitably proceed in a piecemeal manner, owing to the integral
part it plays (or should play) in the design process. The situation in
which we are confronted with an established and already designed
structure, requiring stress analysis, is extremely rare. Instead we are
concerned with onc major component at a time, for much of which
the design iteration can proceed independently. It is therefore natural
that the structural analysis be built around the separate substructures,
but within the framework of their synthesis into the complete structure.
An approach of this kind has been developed in Refs. 3 and 4 and
requires the idea of different classes of redundancy, which enter into
the analysis at different times and in different ways, as well as the
extension of the general theory of the matrix force method to take
account of them. Thus, we distinguish the redundancies internal to
the substructures, and which are obviously required in their analysis,
from those which are involved in the connexions between neighbouring
substructures and are only called into play in the final synthesis. These
latter are called tertiary or interaction redundancies: they are most
obviously exemplified at a wing/fuselage attachment. Though the
tertiary redundancies are generally of a local character, between two
adjoining substructures, cases can arise where a group of substructures
forms a closed ring. The appropriate tertiary redundancies then affect
all the substructures of the ring system.
In addition we make a further division of the internal substructure
redundancies into primary and secondary. The secondary redundan-
cies are most commonly (though not necessarily) the internal ones of
local sub-assemblies, which can be separated out into distinct groups,
e.g. the rings or diaphragms in a fuselage. The sub-assemblies may
therefore be brought into the main or primary analysis of the sub-
structure as "closed" redundant systems in much the same way as the
substructures themselves enter into the final synthesis. In both cases,
i.e. the primary and the interaction or tertiary analysis, it is clear that
the redundancies arc associated with basic systems which are statically
indeterminate and it is, of course, largely to allow for this systematically
that the general theory requires enlargement. The necessary develop-
ment is presented in detail in Refs. 3 and 4 both for the specific problem
of the aircraft and for a more arbitrary complex system.
The essence of the three-stage analysis may be represented by the
11
j. H . A R G Y R I S , S. K E L S E Y and H. KAMEL
"staircase" diagram of Table A.1, which shows the various stress

matrices calculated in the course of the computations. The notation
is that of Chapter V I I I in Refs. 3 and 4.
Table A.1
R W Y X
Complete system B X + Y + W closed

Substructure
B„ b,.- X + Y closed
B„ b;, b, X closed
Sub-assembly Bun b
30 bio b2 Basic system
The symbols R, W, Y, X denote external loads, tertiary, primary

and secondary redundancies respectively and the columns which they
head contain the stress matrices due to unit values of each of them in
turn. Each row represents a stage in the analysis and in the right-hand
column the state of compatibility at that stage is indicated. Thus,
X + Y closed means that the compatibility conditions associated with
and defined by the secondary and primary redundancies are all
satisfied.
To pass from a given level in Table A.1 to the one above, we require
always the same type of operation to "close" the relevant group of
redundancies which may be represented here as a premultiplication
by a standard square matrix for the class of redundancy involved. For
example
b 1 = [i - b 2 D 2 .;b 2 f]b 1 0 (A.i)
and
B„ = [I - b 1 D 1 ) 1 b|f]B ( ) (A.2)
where
D, b 2 fb 2 D it b;fb,
and f is the flexibility matrix of the unassembled structural elements.

The construction of similar equations for the other stress matrices in
Table A.1, is mnemonic and obvious. It is also obvious that the
operation to close secondary redundancies is confined to the redundant
sub-assemblies, and the closing of the primary redundancies to the
substructures. Only in the final stage is the complete complex system
involved.
The stress matrices of the bottom row in Table A.1 are clearly those
appropriate to the primitive basic system, with all redundancies cut.
A "one-shot" analysis of the whole complex structure would pass
directly from this to the final stress matrix B. Apart from our earlier
argument, the scale of such a single, self-contained computation would
12
militate against its adoption even for a medium-sized aircraft and

with a large modern computer available. The step-wise progression
to B illustrated in Table A.1 may, of course, be regarded as deriving
from a systematic partitioning of the matrices involved in the single
computation. The partitioning follows logically the physical divisions
of the structure and automatically exploits the large zero areas in the
complete matrices. Further to this, however, some useful simplifica-
tions follow from the partial satisfaction of the compatibility conditions
at the various stages, which arc not so obvious if the process is viewed
merely as a computational partitioning. Thus, we find that the
stresses S,, in the complete system due to initial (thermal) strains H
may be derived from the equation**4
S„ |b„D i r !,.b! ( . • b A V b j | b 2 D 2 2 'b 2 JH (A.3)
Although the above arguments are relevant to the secondary as well
as to the tertiary redundancies, a further consideration is important
and was, in fact, the original reason for their separation into a special
class. As already mentioned, .secondary redundancies usually arise in
local, redundant sub-assemblies which can conveniently be treated in
isolation. In the open or cut condition these may display grossly
magnified flexibility which, if the secondary and primary redundancies
are taken together, inflates their contributions to the coefficients of the
compatibility equations associated with the primary redundancies, and
may have such a swamping effect (hat serious ill-conditioning and loss
of accuracy occur. In a typical fuselage, for example, the terms in D , ,
arising from the rings are greater by a factor of up to 103 for open rings
than for closed rings. In addition these increased terms arc themselves
less well-conditioned since their inflation stems basically from the
(usually) small number of the secondary redundancies. Thus, if bj
is the compatible stress matrix in the sub-assembly due to unit primary
redundancies and hence satisfies the condition
b 2 f b l - 0, (A.4)
we can reverse the normal procedure and represent any merely
statically equivalent stress matrix as
b 1() = ba. + b 2 X (A.5)
where X contains columns of secondary redundancies and defines the
way the system is open or cut. Taking into account Eq. (A.4) the
contribution of this open system to the matrix D n is clearly
bi„fb l 0 biTb. + X'b 2 fb 2 Xl
= b{fb, + X'D 2 2 X
The first term on the right-hand side is the contribution of the closed
sub-assembly and the second term the inflated part due to "opening"
13
J, H. A R O Y R I S , s. KELSEY and H. KAMEL
it. The inflation term derives from a congruent transformation on

D 2 2 and hence cannot have a higher rank than D 2 2 . This will generally
be much less than the order of D u and hence if the inflated contribu-
tion dominates D n , as with the fuselage rings quoted above, this latter
becomes nearly singular. I n such cases, the separate treatment of the
secondary redundancies is clearly vital.
A.b. Primary Redundancies in the Fuselage

In general terms, the selection of redundancies in a large structure
or sub-structure such as a fuselage or wing should aim to satisfy a number
of conditions. Firstly, they must yield a reasonably well-conditioned
matrix D n in order to ensure accuracy and confidence in the final
results. Secondly, the associated bj matrices must be capable of simple
and efficient generation in the computer, if possible by means of a
standard routine and using the simplest of input data. Thirdly, we
remember that the essential function of the redundancies is to bring
about changes, within the equilibrium conditions, of the overall stress
distribution. For them to be able to do this efficiently and unambigu-
ously, it is important for the individual columns of b j to be as different
as possible. This requirement is of course instrumental in satisfying
the first condition; but, further to this consideration for the regularized
structure, it is necessary in order to ensure the accuracy of the cut-out
analysis, in which severe local changes in the stress distribution may-
be brought about. Finally, and this is important in avoiding ill-
conditioning, it is desirable that the chosen systems yield well balanced
contributions from the different types of structural elements.
In our earliest work on the subject, 1 emphasis was laid on the use
of standardized self-equilibrating stress systems of a localized character.
Such systems, since they are elastically coupled only to their neighbours,
are invariably well conditioned and also suited to the cut-out analysis.
Their simplicity, however, is more suited to manual formation and in,
for example, a fuselage with non-conical taper, they are not so well
adapted to automatic generation on the computer. In Chap. IV, of
Refs. 3 and 4, therefore, a technique was developed for the automatic
setting up of the b t matrices in a fuselage for the effective flanges and
shear panels associated with the primary redundancies centred on a
given cross-section. This technique produces at the same time the
appropriate b 0 matrices of statically equivalent stresses.
Thus, for the flange loads at frame i and the shear flows (field forces)
in bay t, i' + 1, we calculate for a singly-connected cross-section with /
flanges, the t x t matrices (Fig. A.1),
[b0i b„],. == {a,'.. A|}-- (A.7)

I K KAi.nt = { - < , + ! . - K } - 1 (A.8)
14
where b 0 „ b 0 o provide the statically equivalent systems and b 1 ( , b j ,

arc the matrices of flange loads and field forces, respectively, associated
with unit redundancies Y,. The matrices a „ a, are the geometry
matrices which specify the cross-section and local taper. Their function
Fig. A . 1 . Single-cell cross-seclion. Fig. A.2. Two-cell cross-section.
in Eqs. (A.7), (A.8), is to express the equilibrium conditions on the

cross-section. The redundant systems themselves are defined by the
two related transformation matrices
A,; A;« (A.9)
where o is the (square) difference matrix for a closed circuit. The
matrices b u , b,„ are, of course, related through the flange-panel
equilibrium conditions
otb i m . i + i bin- (A.10)
This last relation is used as a computational check on the indepen-

dently calculated flange load and field force systems.
In principle, the choice of A„ A„ is arbitrary, except that they must
yield non-singular matrices for inversion in Eqs. (A.7) and (A.8), and
a number of simply constructed forms has been tried. For example,
linear independent rows, each with the four element pattern 1, — 1 , 1,
— 1 or the three element pattern — 1 , 2 , —1 differently positioned in
each row, can be combined lo form an A,. However, much the best
results so far have been obtained by putting
A, = SI, A.ll)
where SI, is the t X (/ — 3) trigonometric matrix whose columns are

the orthogonal, self-equilibrating end-load systems in a circular cross-
section with cyclic symmetry. Not the least of its advantages is that it
can very simply be formed in the computer. Moreover, for a cross-
section which is not greatly non-circular, the self-equilibrating systems
15
j. H. ARGYRIS, S. K E L S E Y and H. KAMEL
obtained by the above process (including the bending moments in the

rings) are extremely well-conditioned (see e.g. under Section A.d).
As far as the flange load and shear flow distributions are concerned,
this holds also for more markedly non-circular sections, like the outer
70
* 3 * ' " B
CPT: 6 7 a 9 » 11 '2 '3 "

Wing C r o s s - s e c t i o n
005 o -oos oi oos o -oos -01 01 OOS 0 DOS -»1

I 1 1 1 1
—z-3*-
17 17 17
15 r-ai N ^Z~~—af 15
* *Si ~--5s^
13 J
/ ' If-**
13
^êif^
13
" 5a - :r -^r-~-~
Ai '"
11
^^*2---^
n
,-'
>
*N*V:>
9 4
"V
7 ".^1 ^ 7 * jm^m*.
XP £>=*• _-- —-^
~~~~^L
S -*^ 1 5 r——•*
£ S ^ •?a---~ j
--::<^
1 3 3
^
IL 1
Irom fi,
1 1
by o r l h o g o n a l i z a l i o n method
Fig. A.3. Self-equilihrnling flange load systems in wing cross-seclion.
cover of the double-cell cross-section shown in Fig. A.4, and the flat,
nearly rectangular section characteristic of a wing. As shown in Fig.
A.3, for the flange loads in a typical wing cross-section, the peripheral
distributions retain the characteristic wave patterns of the Si, systems
in the circular cross-section and the non-diagonal terms due to flanges
and shear panels in the sub-matrix of D u associated with the set of
redundancies at the rib station are small. The maximum off-diagonal
terms in the normalized D n sub-matrices (i.e. with all leading diagonal
terms reduced to unity) are, in fact, 0-236 and 0-362 for flanges and
shear panels respectively; most of the others are much smaller.
In such strongly non-circular sections, however, the position is not
so satisfactory as regards the bending moment distributions in the
16
>
100 in —
-
y.
-
-
"» 'VJ \ r "

-
7th Antisymmetric system (a) 7th Antisymmetric system (b) —

m cm on in.
-
C'T.10 -OflOS-l o f "
>
-
>
y.
-04J-"0 Slh.Antisymmetric system (a 9th Antisymmetric system (b) -ooos.
Fig. A.4. B.M. diagrams ( b l V M + 1 ) in doubly coiuuc'.ed ring determined (a) from SI, and (b) from orlhngoimlizalion process.
rings or frames which result from the b ^ , b,„ distributions. The left-
hand diagrams (a) of Fig. A.4 show three of the antisymmetric bending
moment distributions in a doubly-connected ring in the double-cell
fuselage due to the £2,-bascd stress systems in the outer cover. It will
be seen that, apart from a change of sign, which is in the context
immaterial, all three distributions are remarkably similar in shape
and in magnitude. T h e increase of waviness and decrease of magnitude
with higher order redundancies, characteristic ofthe parent systems in
a circular section, is entirely absent. The three systems shown have
been selected as the worst examples from the total of nine antisym-
metric distributions for the cross-section. Though the others display
more variety of form, the improvement is slight and it is not surprising
that the resulting D n is not well conditioned, the non-diagonal terms
in the normalized D u being mostly greater than 0-95. Moreover, the
contributions to D n from the rings arc excessively large for the higher
order systems and tend to mask those from the flanges and shear panels.
It is of interest that the symmetrical modes for this cross-section are
very much better and also that meaningful practical results are still
obtainable with these systems, at least for the continuous fuselage.
However, the improvement of the redundancies, to obtain stress
systems of more variegated form and a better conditioned D n matrix
is, fortunately, a simple matter. As detailed in Chap. V I I of Refs. 3
and 4, we utilize the standard latent root programme of the computer
to establish the {t — 3) X {t — 3) matrix US, whose columns are the
latent or eigenvectors of the (/ — 3) X (f — 3) sub-matrix of D u
associated with the set of redundancies Y, at a given cross-section. A
revised transformation matrix A „ where
A, = n , m (A. 12)
is then used in Eq. (A.7), etc., to obtain a new set of self-equilibrating
stress systems. For the portion of structure yielding the original D n
sub-matrix, the new systems are now, of course, orthogonal. However,
much of the value of this method consists in applying the revised
transformation matrix also for neighbouring regions where the cross-
section, though possibly slightly different owing to non-conical taper,
has the same general shape. In these circumstances, and with possibly
different elastic properties, we obtain not an orthogonal set, but a very
well conditioned one.
In Fig. A.4, the new fourth, seventh and ninth antisymmetric
bending moment distributions for the doubly-connected ring and
obtained by the above method are shown. For these, only the con-
tribution of the rings themselves to D u was used in the latent root
programme. This simplification is often possible, owing to the much
smaller sensitivity of the flange load and field force distributions. As
shown by the broken lines in Fig. A.3, the end-load systems are not
18
very much affected by an orthogonalization process which uses only

the D u contributions from the shear panels and a rib formed by a
single peripheral ring.
The physical picture of the improvement brought about in the
bending moment distributions is in Fig. A.4, immediately clear in the
enormous reduction of magnitude and in the characteristic waviness
which is retrieved. The various practical aspects of this method are
dealt with at length in Refs. 3 and 4, We discuss here briefly, however,
the extension of the general method to double-cell fuselages which, in
contrast to that shown in Fig. A.3, have direct stress-carrying flanges
in the floor.
For such a case, Refs. 3 and 4 use Eqs. (A.7) and (A.8) to set up the
b ^ , b 1(( matrices for the outer cover only. In the floor, which was
there assumed transversely flat, simple, local sclf-cquilibrating systems
(essentially the X-systcms of Ref. 1) were suggested. Though these
latter are invariably well conditioned, they are not applicable to a
floor or other internal wall which is curved in section. To overcome
this limitation and give at the same time a unified method, applicable
to the whole cross-section, we discuss how the use of Eqs. (A.7) and
(A.8) may be generalized for a double-cell cross-section. Since it is
still somewhat experimental in its details we limit ourselves to an
outline of the method.
We suppose that the cross-section has t flanges in the outer cover
and a further s flanges in an interior wall joining flanges m and p
(Fig. A.2). We consider first the flange load systems b , , , arising at a
frame station due to the primary redundancies there. It is clear that
b , , , is now a {t + s) X (t + s — 3) matrix and hence needs to be
enlarged by the addition of s rows and columns, i.e.
h u = {b l a h l b } { (A.13)
where suffices a and b refer to the outer cover and the floor respectively.
The extension of a,, to the (/ + s) X 3 form
»i. = {a-„ •»«,}. (A. 14)
is obvious and needs no comment.

Again the choice of the transformation matrix A, [now {t + s) x
{t A- s — 3)] to define the redundancies is, in principle, arbitrary,
subject only to the condition that it yields non-singular matrices for
inversion in Eqs. (A.7) and (A.8). However, the success ofthe trigono-
metric matrix St, in this role for the single cell suggests that it be
retained and a further trigonometric matrix introduced for the floor.
Thus, we may define an s x s matrix SI, where
ft, = [c« C j . . . c r . . . c. +1 ] (A. 15)

19
j. H. ARGYRIS, S. K E L S E Y and H, KAMEL
and
rtr rsn A.1.,]
cos cos
s+ ll
We mention here that though SI, has its origin in the orthogonal,
self-equilibrating systems in a circular cylinder with identical, equally
spaced flanges, Sl s is not the equivalent matrix deriving from a panel
with s flanges. It is formed purely to generate wave-like distributions
in the flanges of the floor.
We may thus put
SI, On
(A.17)
0 SI
and proceed to establish b , , according to Eq. (A.7).
The derivation of the shear flow matrix h l q is not quite so simple
since it now contains t 4- s 4- I rows and must hence be obtained from
an inversion of that order. We remember also that an extra redundancy
arises in the bay between two frames owing to the two-cell cross-section.
Wc first generalize the / x t matrix a into the {( -4- s) X (, + j + I) a e
for the doubly-connected cross-section
/ s 4- I
a. = (A. 18)
I) ».« i
in which a s + 1 is the {s f- I) x s difference matrix for an open chain.

(Sec Eq. (IV.97) of Refs. 3 and 4.)
This matrix is seen to perform the function of the original a in the
single-cell cross-section in, for example, the equilibrium condition
between flange loads and shear flows. Similarly to a „ the matrix a„,
which defines the overall equilibrium condition, requires extension to
include the s 4- I panels of the floor.
If we now follow the analysis of Refs. 3 and 4, we obtain the equation
defining the shear flows of the self-equilibrating systems as
•^l*«"*l9i,i+l — J(+s-3 (A.19)
where
AJa,, is of dimensions (/ 4- s — 3) X (t* 4- s 4- I)
and
b l 9 i < + 1 is of dimensions {t 4- s 4- I) X (i* 4- J — 3)
20
So far our equations allow only for t 4- s — 3 redundancies. To

introduce the extra redundancy in the bay requires that AJa, be
expanded to {t + s — 2) rows and b1(7I l + 1 to (/ 4 s — 2) columns.
The choice of additional rows is arbitrary but if we write
A|ot,.
(A.20)
t eS +1J
where o, and e s + 1 are zero and unit columns with t and s 4- 1 rows
respectively, this choice imposes the condition that the sum of the
shear flows (field forces) in the floor is zero, except in the case of the
additional redundancy. Thus, for the related statically equivalent
system for a torque in the bay, it enforces a generalized (for taper)
Bredt-Batho system in the outer cover only.
The final equation for b„ / I - 1 + 1 is therefore
fcu...+i == {- <.+!.- A3 ' IO I<+_J (A.21)

where the circumflex over h l Q i i + 1 indicates that it contains the extra
column for the Z-rcdundancy in the bay.
A.c. Major Cut-Outs in the Regularized Structure

A vital part in the production of standard routines and programmes
for the analysis of complicated structures is played by the idea of the
regularized structure, i.e. the structure which is obtained by filling in
all the cut-outs and which is topologically, though not geometrically,
regular. All calculations are based on this regular system and, inter
alia, the problem of devising special redundant systems to allow for
the cut-outs is thereby avoided. At the same time, the usefulness of
this procedure, and indeed its existence, depend on a comprehensive
and systematic technique for determining subsequently the effect of
the cut-outs in the regularized structure.
The method originated in Rcf. 1 uses the physical idea of imposing
initial strains on the elements of the regularized structure which are
to be eliminated, until the load or stress in them is zero. These elements
arc thereby rendered structurally ineffective and the resulting stress
distribution in the remaining structure is clearly that for the original
structure with cut-outs. Equivalent mathematical derivations of the
necessary analysis are of course feasible but experience has demon-
strated the value ofthe physical picture provided by the idea of initial
strains, particularly in unravelling some of the more subtle aspects
which can present themselves. In this present sub-section we discuss
some of these aspects in relation to the elimination of a large area of
structure which itself can exist as an independent sub-system. The
related question of modifications, which arise through changes in
21
j, H . A R G Y R I S , S. K E L S E Y and H. KAMEI,
element flexibility, is, in comparison, straightforward and is not

mentioned. (See, however, Refs. 3 and 4.)
The essential result of the cut-out analysis, as far as the stress distri-
bution is concerned, is contained in the equation:
S c = S - [h w D l v \ v h'. v l t + b 1 D 11 1 bL i + h ^ h M ^ (A.22)
where suffix c refers to the structure with cut-outs and suffix h refers to
the stresses which are to be independently nullified by the cut-out
process.
The matrix H)h<. is given by the equation
»* = \>W/PWWKH + b l h D u ' b ' 1 A 4 b 2A D 22 -b 2)l ]-i (A.23)
Eqs. (A.22) and (A.23) are for a complex structure whose substructures
have both primary and secondary redundancies and therefore include
more limited cases. We emphasize that all data in the above two
equations are for the regularized structure. These equations also
assume the cut-outs to be made after the assembly of the substructures
into the complete system. However, this does not affect the present
discussion. We see that the cut-out is defined by the rows of h l t etc.,
which together make up b,„, etc., in Eqs. (A.22) and (A.23) and our
present discussion is confined to the selection of the rows, i.e. the
selection of the particular stresses whose liquidation will produce the
effect of the cut-out.
It is immediately clear that in a large cut-out, it is not necessary to
nullify independently all the prescribable stresses, i.e. those given by-
rows of S, within the cut-out. This follows from the equilibrium
conditions among the stresses in the cut-out, which ensure that some
of the stresses vanish automatically when others are made zero. Alter-
natively we may note that only in statically indeterminate elements
can the stresses be affected by the presence of initial strains. If therefore
the removal of certain elements leaves behind a group which is statically
determinate, the stresses in this group cannot be further affected by
initial strains, so that any attempt to nullify them directly must prove
nugatory. Hence in approaching this problem it is better to think in
terms of making the regularized structure ineffective within the cut-out
boundary rather than doing away with it altogether. From this angle
it is also clear that in large cut-outs we may leave behind a redundant
group or sub-system which remains connected to, but in such a way
that it cannot work with, the surrounding structure. We illustrate
these points now and the systematic techniques for making a proper
selection of cuts to which they lead, with reference to the plane frame-
work shown in Fig. A.5. A more detailed discussion together with
further examples is given in Chap. V I I I of Refs. 3 and 4. For simplicity
of expression wc describe the operation of nullifying directly a particular
stress as "making a cut" or simply "cutting".
22
la) Ragutan/ad Structure (c) Minimal t u t ayatam - b o u n d a r y (a) M a . i m a l c u l ayatam - m l a r n a l

r a d u n d a n c i a a ralaaaad r a d u n d a n c i a a o l a u b - a y a l a r n alao ralaaaad
lb) C u l out ai r a g u l a n z a d alruclura (d) Inadmiaaibla cul t y a l a m d u a to (I) Oiamambanncj o l aub - ayalam to b a
COncurranca o l ra».dual link* a l i m m a t a d - m a x i m a l c u l ayalam
Fig. A.5. Elimination techniques in large cut-out.
The illustration (a) shows a highly redundant, plane, pin-jointed

framework which must be supposed to extend beyond the bounds of
the figure and has a regular, X-braced panel construction. The cut-out
to be produced is shown beneath, (b), and, as drawn, extends over
7 bays horizontally and 5 bays vertically. It involves the effective
elimination of 128 bars from the regularized structure.
Minimal cut system. The first technique we consider, illustrated in
(c), is to cut bars around the boundary until there remains a self-
contained sub-system connected to the outside structure by statically
determinate links only. This residual system is obviously ineffective
as far as the main structure is concerned, since deformation of the cut-
out boundary causes merely rigid body displacement of the 3 x 5
bay framework left behind and cannot induce any loads in it. That
the framework left behind is redundant is clearly immaterial; it has
ceased to be an effective part of the main structure.
We must, of course, be careful to ensure that the bars left connected
to the rest of the structure are not all parallel or concurrent. T h e
arrangement shown in (d) is inadmissible, having a so-called critical
form which, while yet possessing a kinematic freedom (of rotation
about the point of intersection of the bars) has a redundant connexion
to the boundary. Mathematically this would give linearly dependent
rows in b 1 A and hence a singular matrix to be inverted. Though an
exact singularity in an actual numerical case is unlikely, arrangements
approximating to (d) can be expected to be sensitive and lead to an
ill-conditioned matrix for inversion.
23
j. H . A R G Y R I S , S. KELSEY and H. KAMEL
Returning to a correct arrangement (c) we note that the effective

elimination of 128 elements has been achieved with only 57 actual
cuts. Since we have left behind the maximum number of internal
redundancies, it is evident that this is the minimum number of cuts
which can reproduce the required effective cut-out. In general, the
minimal cut system is obtained by cutting all redundant boundary-
conditions and the number of cuts required is hm where
h m = P t , = m b ~ eb (A.24)
and p b is the degree of redundancy along the boundary; mb is the
total number of stresses nullified at the boundary; eb is the number
of independent equilibrium conditions relevant to the interior sub-
system as a whole.
Maximal cut system. It is, of course, possible to make further cuts
in the configuration (c), provided they are confined to redundant
elements of the residual sub-system. As shown, this has a degree of
redundancy of 23 and hence a further 23 cuts may be made, leading
to a residual system like that of (e). This has the maximum possible
number of cuts, being reduced to a statically determinate system which
cannot further be affected by initial strains. We need hardly add that
these additional cuts arc quite superfluous. However, the system serves
to establish the maximum number of cuts /(,,.
hM = h m + P i (A.25)
where p i is the redundancy of the internal sub-system.
h M may also be fixed from the total number m T of stresses to be
nullified, both directly and indirectly, and the number of e T indepen-
dent equilibrium conditions between these stresses as
hy = m T — e T (A.26)
For our present problem, e T is equal to twice the number of nodes
internal to the cut-out. Hence
hM - 128 - 2 x 24 - 80
Dismembering technique. In this approach we set aside formal con-
siderations of redundancies and equilibrium conditions. Instead we
consider how we may set about the physical task of systematically and
progressively dismantling the regularized structure within the cut-out
boundary until only a statically determinate skeleton remains. Wc
Start by removing all the diagonal (shear) bracing members within the
panels, which is clearly permissible. There remains a system of roughly-
horizontal and vertical bars which still, of course, constrain the cut-out
boundaries and constitute a multi-redundant system. To eliminate
this, however, it is only necessary to cut one bar in each ofthe horizontal
and vertical chains to give the final arrangement of (e). The number
24
of cuts made is seen to be 7 x 5 x 2 + 4 + 6 = 80 which is the

maximum number permitted.
The simplicity of this approach is most attractive, with its systematic
treatment of the groups of elements according to their structural
functions. It proves particularly successful in some of the cut-outs in
stressed skin structures where further subtleties may be associated with
the minimal technique. On the other hand, it yields inevitably the
maximum number of cuts and hence a larger matrix for inversion.
A.d. Experimental Work on a Fuselage

A series of experimental investigations has been carried out in the
Aeronautical Structures Laboratory at Imperial College, London, on
a cylindrical fuselage model to give comparisons between the experi-
mentally measured stress distributions and those calculated by the
matrix force method. The model (Fig. A.6) was designed specifically
for this purpose and the tests so far carried out include the effects of
two cut-outs as well as the complete, uncut structure. For the structure
without cut-outs, the aim was to make the comparison as wide as
possible by obtaining an experimental picture of the overall stress
distribution in a structure of representative construction on a scale
comparable to that yielded by the analysis. The stresses measured
were the longitudinal direct stresses and the shear stresses in the skin
as well as the direct stresses in the rings or frames. Measurements were
made at 427 points, which compares with a total of 594 non-zero
values obtained from the calculations. For the effect of cut-outs,
however, a more detailed experimental picture was obtained for the
regions adjacent lo the cut-outs.
The geometry and construction of the fuselage specimen are illus-
trated in Fig. A.7. Its outer skin is of sandwich construction with faces
0 0 1 2 in. thick and 0*15 in. thick aluminium honeycomb core. In
addition, longitudinal reinforcing strips or booms are bonded between
the sandwich faces at an approximate circumferential pitch of 8 in.
These increase the total cross-sectional area of material that carries
direct stress to nearly twice the shear-strcss-carrying area and hence
represent the main effect of the stringers in the usual stringer/sheet
construction. In addition they provide natural boundaries for a series
of cut-outs without the necessity for special reinforcement of the thin
sandwich faces. The non-circular cross-section, which for construc-
tional reasons has a tri-symmctric form, is maintained by six light and
three heavy rings. The light rings have a uniform top hat section with
honeycomb filling, while the heavy rings, (1, 3 and 9), which are
designed to take the transverse concentrated loads, are of tapered,
double-channel cross-section. Two vertical posts with ball seats are
glued and bolted between the channel webs at identical positions on
each heavy frame to provide the support and loading points.
25
j. H . A R G Y R I S , S. K E L S E Y a n d H. KAMEI.
Fig. A.6a. View of fuselage tesl.
Fig. A.6b. View of fuselage test.

(Note that cut-outs shown refer to tests not discussed in the paper.)
26
Vr -•••• i.:---^
-
•
Support ••
framaa
1
\
i-Honayeorrpb Cora ^ * /
0 1175 in. True,. ' B o o m 1 1 i n a 0 1175 In
Fig. A.7. Construction and geometry of tested fuselage.
All joints in the structure arc made with cither Redux, or Araldite
epoxy resin adhesive, bolts and rivets being used only as jury-attach-
ments while the adhesive was cured. Thus, dubious corrections to the
results for rivet and bolt slip are avoided. The use of sandwich con-
struction, besides permitting an all-bonded structure without requiring
too many complicated jigs and tools, gives a smooth structure in which
the interpretation of strain-gauge results is unambiguous and also
allows the development of a high elastic stress level without buckling,
which enhances the accuracy of the stress measurements. At the same
time, the thin sheet which can be utilized in a sandwich means that
the applied loads are kept reasonably small.
For testing, the fuselage model was supported on ball scats on two
portal frames at rings 1 and 3. At ring 1 the ball seats were located
inside the ring and the support reactions transmitted via adjustable
C-links to the portal. The adjustment on the links was provided for
levelling the fuselage and also allowed the (statically indeterminate)
support reactions to be equalized in symmetric loading cases.
Vertical loads were applied to the two loading points on frame 9 by
means of a powered screw jack, the load being measured by a hydraulic
capsule and pressure gauge. The loading cases investigated arc
illustrated in the figures which show the results. Basically they con-
sisted of a pair of symmetric loads at frame 9 with symmetrical support
reactions and two unsymmetrical load systems obtained by loading on
onc load point only with one or other of the C-links at frame 1 removed.
A further loading consisted of a bending moment applied as concen-
trated loads to four of the booms (reinforcing strips) at the end cross-
section (see Figs. A.6, A . l l ) .
27
All measurements of stress were made with electrical resistance wire

strain gauges, the gauge patterns and distribution being chosen so that
the results would be directly comparable with theoretical values.
Thus, the direct stresses in the cover were measured at the centres of
the reinforcing strips, which coincide with the effective flanges of the
idealized structure used in the analysis. For the shear stresses, rosettes
of four gauges (two inside and two outside) were positioned centrally
in the panels and wired into a complete bridge circuit to give an output
directly proportional to shear strain. In the rings, single gauges were
used to pick up the circumferential direct strain on the inside of the
rings and also on the attachment angle to the skin.
The interpretation of stress from the measured strain, as well as the
calculation of flexibilities in the analysis, were in all cases based on
experimentally determined clastic properties of the material, longi-
tudinal and transverse strains being measured on simple tensile
specimens made from surplus material. From these results E, v and
hence G were deduced. For the cover material, sandwich specimens
were used, in preference to ones made only of the face material, thus
allowing for the slight influence of the core.
For the idealized structure o f t h e analysis, 18 effective flanges were
positioned to coincide with the reinforcing strips in the sandwich skin.
The degree of primary redundancy is hence 15 at each interior ring
station, giving a total 105 for the complete structure, with 9 x 3 = 27
secondary (ring) redundancies. From the point of view of modern
computations, this is a very modest problem indeed.
T h e cross-sectional areas of the effective flanges include the skin
area between the centre lines ofthe adjacent panels, the skin area being
calculated for the polygonized cross-section. Similarly, in the compu-
tation of the ring properties / and A an effective width of skin was
included. The width used, about 3 in. on each side of the ring, was
largely an intuitive guess. However, it is hoped that the experimental
results for ring stresses and deflexions will give some guidance for
future use on this awkward question. The area C, effective in shear,
was taken simply as the web area of the ring cross-section.
Two elements in each of the heavy rings possess considerable taper.
In assessing the flexibility of these elements, a parabolic variation of
1 j E I was assumed, based on the values at the ends and the middle of
the elements. Otherwise, all elements have a uniform cross-section.
Although values of /, A, C were calculated by hand and regarded as
input data, all subsequent calculations of flexibility were carried out
on the computer. The same basic data was also used in the detailed
calculation of stresses from bending moments, end loads, etc., as the
final stage ofthe computations.
In the analysis the self-equilibrating stress systems produced by
Eqs. (A.7 and A.8) with SI, as the generating matrix showed themselves
28
extremely well conditioned, as may be judged by the following extract

of D n for the symmetrical redundancies at a frame station. The
numbers over the columns refer to the redundancies and a common
scalar factor has been dropped.
D 11s —
2 3 4 5 6 7 8
n i s i -2*203 27*34 66-59 ()•()()() 7-985 19*64 0*044
53-86 -0-232 0-060 4-302 - 0 - 0 6 2 -0*030 1*320
15-55 1-777 - 0 - 0 2 3 0-623 0-678 0*034
14-44 - 0 - 0 3 2 0-553 1-318 0-029
9-722 0-037 0*054 0*227
8*741 0-240 0-030
8*692 0-009
4-154
The most severe coupling is between 1 and 4 yielding a normalized

off-diagonal term of 0-504; none ofthe others rise above 0-2, and are
mostly much less, which is extraordinarily good.
As will also be seen, the diagonal terms show the desirable falling
off in magnitude with increasing order of redundancy. Contributions
from flanges, shear panels and rings arc also well balanced, as may be
judged from the following comparison. Taking the (almost constant)
flange load contribution as unity, the (monotonically) decreasing
ranges of shear panel and ring terms are approximately,
shear panels 4-1-0*48

rings 208 -0*72
The antisymmetric systems are equally well conditioned and demon-
strate the effectiveness of SI, for defining the redundancies. The analysis
was otherwise straightforward and requires no further comment.
The results of the computations, together with the experimentally
determined stresses are illustrated in Figs. A.8 to A.11, for four loading
cases. They are shown as peripheral distributions around the fuselage,
starting at the top, for cross-sections midway between adjacent frames.
The theoretical values of shear stress are the constant values in the
inter-frame bay while the direct stresses, because of the assumed
linear variation of flange load within the elements, are the means of
those calculated for the two neighbouring frame stations. The depar-
ture from the elementary theories (E.T.B. and Bredt-Batho) is note-
worthy. This is particularly brought out by a comparison of the
direct stresses in Fig. A.8, with those of Figs. A.9 and A. 10, and also
by the smallness of the direct stresses at the top of the fuselage.
Agreement between experimental and calculated values is in general
very good, especially for the direct stresses and in the way the shapes
29
^9,
4 -4000 0 4000 lb/ ,
. i 1 'in
-300 0 300 kg/ 2

/cm
>
-
-
Direct Stress
Calculated
-
-
-
-
Q.
-4000 0 4000 l b / 2
. . i /in >
-
-
0 J0
-300 ° k9/-m2
/CITl
Shear Stress
Fig. A.8. Stress distribution in cover. Loading case I.
4 -4000 0 4000 lb/jn2
- 30O- ~~0 300 kg /( . m 2

>
-.
-
-
-.
-
Z
Direct Stress
a
s.
o
T
Calculated
UB
H
~
~
O
H
C
-
>
-
4000 0 4000 lb/. 2
i • 'in
>
•300 0 300 k9 /cm J
Shear Stress
Fig. A.9. Stress distribution in cover. Loading case III.
-
*
OS w
r
C/J
w
-
H
PI
H
=
D
-
-
O
-7
>
r
>
y.
-300 0 300 ^ / c m 2
Shiear Stress
Fig. A.11. Stress distribution in cover. Loading case V.
J . H . A R G Y R I S , S. KELSEY a n d H . KAMEL
of the calculated distributions are faithfully copied experimentally in

the unsymmetrical loading cases. Numerically the agreement between
individual values is generally better than about 5 per cent which is
not much in excess of the likely experimental tolerance.
With the shear stresses, there is a consistent tendency for the experi-
mental stresses to fall slightly below those calculated. This is, of
course, a familiar feature of experimental distributions where it is
regarded as unusual to be able to account for more than about 85 pet-
cent of the applied shear. In the present case it is likely that a contri-
butory factor is the bending stiffness of the reinforcing strips or booms
in the plane of the cover.
O n e other discrepancy is in the diffusion of the bending moment
applied as a four point loading. Experimentally the diffusion is more
rapid than is calculated. The reason here lies in the construction of
the heavy frame (9) which is a double frame (see Fig. A.7) and gives
shear reinforcement to the cover over a length of about 3 in. It is
expected that further calculations, with frame 9 separated into two
and enclosing a short bay reinforced in shear, will improve the agree-
ment in this respect.
Subsequent to the measurement of the overall stress distribution in
the complete fuselage, a progressive scries of cut-outs was made. In
the remaining figures of this section some results, both calculated and
experimental, are given for the first two cut-out cases which have been
investigated. T h e first of these, A, was the simplest cut-out which could
be made within the grid framework of flanges over a single bay length
(Fig. A. 12). Although no reinforcement was applied, this simple
cut-out was, from the point of view of the idealized structure, a com-
bined cut-out and modification because of the loss of effective cross-
section to the bordering flange and frame elements. The order of the
cut-out matrix h l h was therefore 11. It should, of course, be remembered
that the modification would have been unnecessary if, as would usually
be the case, the modified flexibilities had been used in the initial
calculations. These, however, would not have been correct for the
continuous fuselage.
Figures A. 12 and A.13 show details of the direct and shear stress
distributions in the neighbourhood ofthe cut-out for two loading cases.
Possibly of most interest are the (average) direct stresses in the flanges
which agree satisfactorily with measured values. The changes in shear
stress due to the cut-out are less pronounced, but the theoretical values
appear to be upheld by experiment. In the panel immediately below
the cut-out, a more comprehensive system of shear gauges was applied
in order to detect variation from the constant shear assumed in the
idealized structure. From the evidence of these two figures, the concept
of the constant shear field appears to be surprisingly realistic.
T o obtain cut-out B, the already existing opening was extended
34
Shear Stress Direct Stress
With Cutout 101 HI With Cutout -*-^>

Without Cutout Withouf Cutout —
Experimental . . . . Experimental
-
-
-
z
-
3000 2000
-.
-
r-0 ,b
/n2 0 -.
o
-.
-3000 -2000
LlfPP -
-
>
-
150
>
r- 200 z
k 0-
-0 Vc,
-200 _--i——. i - :::::

-150
Fig. A.12. Stresses in cover around cut-out A. .Loading case I.

With Cutout IHI .1 With Cutout ~-s-~

Without Cutout Without Cutout —
Experimental • • • Ejtperimental . • •
>
20O0lb O
3000 2000-. T.
OS
-
- -
imiTiimnîmiJ3
l-o %' 0- H
-3000 -2000
__ 150
200 >
K
w
k -
o «/cmJ oH
-200 -150
Hyil'llllll-miimliini
Fig. A. 13. Stresses in cover around cut-out A. leading case III.

upwards over two further panels (see Figs. A.7, A.14 and A.15). The
idealized structure thereby suffered altogether the loss of three shear
fields and two flange elements as well as experiencing modifications to
the bordering frames and flanges. b 1A was in this case of order 23,
though again, had the modifications been anticipated, this would have
been reduced to 5, for the cuts alone.
As with cut-out A, the calculated direct and shear stresses are
supported by the experimental results, shown in Figs. A.14 and A.15.
The constancy of the shear stress over the specially treated panel is
possibly not quite so good, but it is noteworthy that the theoretical
shears do tend to be conservative.
For cut-out B, some results arc also shown for the direct stresses in
the rings which border the cut-out. These are calculated, for the
modified cross-section, from the bending moments and normal forces
at the nodal points. (Since the normal force is assumed constant along
each ring clement in the analysis, a mean value was taken at the
junctions.) The agreement between experiment and theory is remark-
ably good on the inside of the rings, though less satisfactory on the
outside. This latter effect is, no doubt, due to coupling with the longi-
tudinal direct stresses in the cover. However, the magnitude of the
bending stresses on the outside is much less and the most important
ring stresses—which can be at least as large as the cover direct stresses
—are predicted with very high accuracy.
The experimental points also reveal some bending of the rings in
the local plane of the cover, which requires further investigation. It
is, of course, an effect not allowed for in the current idealization and
must imply, inter alia, some bending of the flanges in this plane.
The selection of results presented here already gives a most com-
forting support to the structural representation adopted in the idealized
system, especially in regard to the cut-out analyses. Further investiga-
tions will, however, provide more information on the ring stresses and
also on local bending effects near cut-outs. The cut-out B is to be
extended over three bays (an intermediate stage is shown in Fig. A.6),
thus involving the loss of some ring segments as well as cover area.
T o complete the picture, measurements of overall deflexions and the
deformation of the rings arc planned.
A.e. Postscript to the Force Method; Flexibilities; Thermal Stresses

In our previous review of the force method we referred only cursorily
to the flexibility matrix f of the unassembled elements. An exhaustive
discussion on the formation o f f may be found in Refs. 1, 3, 4 and 8.
However, all but Ref. 8 did essentially assume that the Poisson's ratio
effect could be ignored and that we can represent the structure by an
assembly of effective or lumped flanges and shear-carrying walls. It
is true, of course, that Ref. 1 did produce the so-called L-matrix
37
With Cutout With Cutout ~rs-
Without Cutout Without Cutout
Experimental Experimental
>
-
7-
W
OS
00
~
J.
<
tt
—
r
>
R
x
-
Fig. A. 14. Stresses in cover around cut-out B. Loading case I.

i •
2 •§ _
if) AAA 9 S
-Sol
03 t
• C
•"*••••
I S.I
£
S K
=* * LU
~l 1 1^
s O
E
^
o
O r*»
_l I
o
_L_
a
I-
]f 1 *a
bj
I • « _ P
_. I
in
l/)
ii
E.
•a—
taO
3
O
3
SI
(0 o •e E
r. |=» i| |iS «
39
J. H. ARGYRIS, S. KELSEY a n d H. KAMEL
"f
'I -4000 0 4000 I t / ^ J
/
/ « -300 0 WO ""Vcrai*
ll
. ..
CafcuUM £
E«p»rim»fH«l . . . .
Fig. A. 16. Direct stresses in rings al cut-out B. Loading case I.
-4000 0 4000 ">/,„•
1
1 I .
-300 0 300 "9ji„J
Calculated
Experimental
Fig. A. 17. Direct stresses in rings al cut-out B. Loading case III.
Ill
technique which does not involve any guessing of flange areas, but
assumes a linear variation of direct stress in the sheet between nodal
points taking full account of Poisson's ratio effect. Although the
procedure as given in Ref. 1 is restricted to rectangular grids it could
no doubt be generalized to oblique fields and slight taper. The
technique has proved very valuable in investigating problems w h i n
cross-strains are of major importance, as in the immediate neighbour-
hood of concentrated loads diffusing into panels where a good agree-
ment between this theory and experiment has been achieved in a
set its of tests carried out at the Aeronautical Structures Laboratory of
Imperial College. Moreover, R. E. Miller J r . obtained satisfactory
confirmation of the procedure for small delta wing models tested at
the Boeing Transport Division. However, the technique, as it stands,
is not really a practical tool, making considerable demands on the
computer which could probably only be met efficiently by large
machines like the Fcrranti ATLAS and the Remington Rand 1107.
We believe now that if the relevant problem is to be solved by the
force method this is best realized by a different approach (Ref. 8).
Thus, it is clear that wc may allow for all direct stress-carrying effects
in the sheet by introducing additional systems of redundancies. Their
number is evidently larger the more closely our assumed stress pattern
approximates to the actual stress distribution. We have investigated
a number of schemes which take account of sweep and slight taper;
more general configurations arc under review. At present we use the
simplified scheme assuming a constant stress distribution between
adjacent nodal points. It follows hence that the antisymmctrical
direct stress distribution in the sheet must be carried by the flanges.
As a consequence the area has to be increased by a sixth of the cross-
sectional area of the immediately adjoining two panels. Thus, by this
procedure the Poisson's ratio enters into the uniform (symmetrical)
component of the direct stress (between the nodal points) but not in
the linear (antisymmetrical) bending component. There is some
evidence that the method yields a reliable estimate of the stresses
although it is admittedly not as accurate as the L-matrix technique.
We now turn our attention to the question of additional redundant
systems arising from the more refined stress system in the sheet
admitted here. Assuming for simplicity of presentation a rectangular
grid, a typical generalized redundancy is illustrated in Fig. A. 18, its
main lines of action being along the ^-direction. Corresponding
systems operating in the ^-direction can immediately be set up. It
will be noted that the constant crosswise stress distribution applied to
the sheet at onc end must fall linearly to zero at the other end. By
virtue of equilibrium on a sheet element this yields a shear flow
constant in the ^-direction but varying linearly along the j-direction.
This linear distribution of shear flow must be superimposed on the
41
J. II. ARGYRIS, S. KE1.SEY a n d H. KAMEL
-'«•) 1111 ,-!|

I IP"'
f
.1 •1
-1/2
imfmttTtTnv-
:
-1/2 rlM Mi, -1/2
Normal force in flanges Normal force in flanges
-1/8 -1/8
.1/2
-1/2
Field force in panel Normal Force in panel
Fig. A. 18. Ceneralized redundancy for direct stress carrying panels.
» x
g = x/l
Fig. A.1 !>. Flexibility for flange under three generalized forces.
VI
Ni, = 1 Ni, =1
Fig. A.20. Flexibility for oblique panel under five generalized forces.
(To each loading case, N x and N v , there also corresponds a field force distribution.)
constant one associated with the standard systems of redundancies

discussed in (A.a) and (A.b). We also note that the loads in the flanges
are no longer linear between nodal points, there existing also a parabolic
component due to the linear variation of the shear flow. For reasons
of space we cannot develop further the theme of the new system of
redundancies but it may be useful if we reproduce the flexibility
matrices corresponding to the expanded stress system. At the same
time we allow for parallelogram fields for which the stresses in the
sheet are referred to the oblique axes as shown in Fig. A.20.
We first consider the flange, the loads on which may be described
as consisting of three generalized forces, P u P 2 , P 3 made up of two
linear and one parabolic variations of the axial load: see Fig. A. 19.
The flexibility matrix f associated with JPJ, P 2 , P 3 may now be obtained
from the unit load method of Ref. 1. In its more general formulation
for a three-dimensional element it may be written as
f = JJJb'<t>b dx dy dz (A.27)
where b is the assumed (6 X n) stress matrix due to the n unit general-
ized forces prescribed on the surface of the element and <j> is the
flexibility of the unit cube
1 -v 0
1
4> -v 1 0 (A.28)
I:
0 0 2(1 + v)_
For a flange under an axial load the flexibility per unit length is
•, J_ (A.29)
'EA
43
J . H . A R O Y R I S , S. KELSEY a n d H, KAMEL
The internal stress matrix due to the three unit generalized forces of
Fig. A. 19 is
b , = [S (1 - I) 4 | ( 1 - £)] (A.30)
Substitution of Eqs. (A.29) and (A.30) into Eq. (A.27) yields for a
uniform flange the (3 x 3) flexibility matrix f„
,.
/
(A.31)
EA s
1 \ I ,-.
The 2 x 2 submatrix bounded with dotted lines is the standard

flexibility matrix corresponding to the linear variation of a flange
load. The generalized displacements due to P l t P 2 , P 3 may now be
derived from
v = K v2 v3} - f,P = f.{P, P , P3} (A.32)
Next we determine the flexibility matrix of the parallelogram panel
of Fig. A.20. Since five generalized forces are here prescribed the
flexibility matrix is of order 5 x 5 and may be expressed as
t P - j$th' P 4, P h P dx dy (A.33)
where b , , is the (3 x 5) matrix of the oblique stresses due to unit
generalized forces. Introducing the non-dimensional coordinates
x y
-. and « =— (A.34)
/ a
and noting Fig. A.20 we have
/(I -- f) IS 0 0 0"
1
hP = 0 0 d{\ - r i ) dii 0 (A.35)
tdl
- d { \ - ;/) d{l - r,) - / ( i - S) l{\ - S) d
In Eq. (A.33) <(>/. is the 3 x 3 flexibility of a unit parallelogram clement
and may be put in the form,
1 ;. 2c '
I
•4P -A i 2c (A.36)
Es
2r 2c 2(1 +f*)i
where
A = v sin 2 0 — cos-
os 0,
0,
2 2
( 20
/* == v sin 0 4- :os (A.37)
and
s = sin 0, c = cos 0
Eq. (A.36) reduces to the standaid expression for the flexibility of a unit
square element when 0 = 90°.
44
St
oj UOISIAIQ uodsuEJ^L Suioog j o *.if JJ|[ipv "3 M*^lBH ° l 3 n P MnjiHUJi
Jiqnqnj pajsaq B ui sassajjs jBoijajoaqj puB pnuouiuadxa j o uon
-BJ3JJOD B si n5irtcfv poqiaui aruoj .mo ui p.uopisuoa aq oj uiajt JSB| aqj^
0 11 (I o
9 9
(rf+0?I (rf+i) I 3
9 9
(rf + I)
(38S*V) J 3)
?_ •1 .; 9
3 i 1
9 'I 'i
3 3 i 1
•(rf+ 1)2 3 3 I)
!-: 9 !) 9
i I 3
9 £ 9 9
(qarv) 1 I 3 3
a
9 '» 9
3
(-/+ I
1
(rf+l)2-
9 'I
(i
3 3
( ' / + I)
II
t_ i
(I 0
v Y
I 1
(«8£"V) r X = ft
t I
(i 0
r
I i
(i II
If
•'•'•'ii
(8£-V)
K> '•(>• 7.7
i
= rf
j
SB 3-.U/A 3M q j i q . w XI.HBIU
Aitpqixay aqj uiBjqo 3M (gg*y) "bg ui (gg*y) P " B (cg-y) -sbg Suisq
SISA'IV.NJV 'iv*tin.i:)n*H.i.s .-o s a o n x a w XIHXVK

I- A R G Y R I S , S. KELSEY a n d H . KAMEL
whom the authors are indebted for the present information. Figure
A.21 shows the rectangular box beam and gives results for two tem-
perature distributions but the experimental evidence is restricted to
the lower range. As far as the theoretical investigations are concerned
kg/cm' lb/In
1100
'at croaa-iaclloa,
—— f l m l . dtrlarenc.
— — Arghor line gr to
»-— Arghor coatee grid
a experimental
Temperature
diatribulion 'C
Fig. A . 2 1 . Heated box beam by Arghor and comparison with finite difference
and experiment as in WADC TR 54-384.
these had first been carried at Wright Field using the finite difference
approximation for which it was considered necessary to introduce a
very fine grid. The Boeing Transport Division and in particular
R. E. Miller J r . applied subsequently the matrix force method of
Ref. 1 and operated with two grids of which the finer one is still very
much coarser than the finite difference one ( A R G H O R is a Boeing
code standing for Argyris-horizontal-partitioning). It is gratifying lo
observe the close agreement of theory and experiment. In general, the
experimental results lie closest to the finite difference computations
which differ, however, only insignificantly from the results of the fine
grid for the force method. It is, moreover, remarkable, that even the
extremely coarse grid of the force method yields stresses sufficiently
accurate from the practical point of view.
B. DISPLACEMENT METHOD
B.a. Idealization of Membrane Cover; Stiffnesses

We referred to the matrix displacement method in general terms in
the Preface and emphasized the basic simplicity of the method when
analysing membrane covers of arbitrary shape. We mentioned in
46
particular how straightforward the procedure becomes if the membrane

is partitioned into triangular elements. Appendix I presents a full
account ofthe theory leading to the strain-stress and stiffness expressions
of triangular elements as used by us in the analysis of wings. An
alternative more elegant approach is given under D.c below. We note
here only that the triangularization necessarily induces a constant
stress and strain pattern within each panel. However, a few practical
considerations ma\ be helpful to the application of this technique.
6 • i, 2234 x 10 ' i n lib
Id)
F = 10 5 it 10' lb/In! v i n . I . OOS in
Fig, B.l. Diffusion of loads into rectangular panels; comparison of triangular

and rectangular elements.
Thus, it is fairly evident that the triangles should not be of any extreme
shape like thin wedges, the most appropriate configuration being, in
general, not far removed from an equilateral triangle. Another aspect
may best be described with the help of a particular example. Figure
B.l illustrates three different solutions to the diffusion problem of
symmetrical loads into a rectangular panel. In each case the same
number and position of nodal points is used, the three approaches
differing only in the direction of the "diagonal" sides. It is clear that
the best solution must provide the largest displacement in the direction
of the load since the kinematic technique overestimates the stiffness
47
(see Ref. 1). It follows that the most advantageous triangularization

is that of (a). At the same time it is evident that the decisive factor
determining the accuracy of the deflexion for the given number of
nodal points is the corner configuration, as is seen from the arrangement
(c) which approximates closely to that of (a). A still more accurate
answer is obtained by using the rectangular elements of Rcf. 1. No
difficulty arises in allowing for elements of varying thickness /—it being
only necessary to substitute the total mass or volume S)l0 of the corre-
sponding element in place of Lit in Eq. (1.35). An additional and
practically important advantage of the triangularization is that it
allows where necessary a local refinement of the grid system whilst this
is much more difficult, if not impossible, with other types of elements.
This is shown in Fig. B.2, which indicates how a closer network may
Fig. B.2. Panel with varying triangular elements.
Fig. B.3. Strains in triangle.
be applied to the diffusion problem in regions where steep stress

gradients occur.
Altogether, it appears from our experience that the triangular
element if used judiciously yields a very accurate stress distribution in
the cover, possibly only exceeded—for the same number of nodal points
—by that of rectangular or parallelogram elements, which arc however
of limited applicability. Since the stresses given in Appendix I are
referred to the local system of axes O'x'y'z' it is necessary when requiring
the strains and stresses in any other prescribed direction, to apply tin-
standard transformation formulae—an operation which is easily carried
out in the computer. Of course, such evaluations must be performed
with some engineering sense. For example, in Fig. B.3, it may be of
48
interest to check experimental direct and shear strain measurements

associated with the indicated two directions f and tj. Then, it proves
usually most appropriate to take the mean value of the corresponding
strains e f{ , e m , fitl of each pair of triangles I and II and to assign these
values to the point of intersection A where the £ line intersects the
common side of the two triangles. The same rule may be applied to
the stresses. We may then assume a linear and continuous distribution
ofthe strains and stresses along the £ line. In fact, this procedure has
been applied in the comparison of theoretical and experimental cover
strains of a wing-type structure shown in Figs. B.l9 and B.20. In
modern wing systems with their small aspect ratio and pronounced
sweep, it is desirable to determine the principal values of the stresses
either for each triangle or pair of triangles as in Fig. B.3. Summarizing,
we may state that the method of triangularization proves so successful
and simple to handle that there is hardly anything of significance to
add. However, the pure membrane state of stress in the cover—which
continues at this moment to be a fundamental assumption in most
aeronautical applications—cannot be upheld any longer for many-
high-speed configurations with their relatively thick walls and reduced
profile thickness. Here it is imperative to take account of bending
and the simple triangularization of Appendix I has to be extended to
include shell effects.
Apart from the triangular panels, wc have also applied a number of
elements of different geometry like arbitrary but plane quadrilaterals.
Their analysis, although much more elaborate than for triangles, can
yet be formulated in a concise matrix form as shown in some detail
in Appendix II. Interestingly enough it appears from our investiga-
tions on wing structures that the triangular panels lead in most cases
to more accurate stress and deformation estimates than the quadrilateral
panels based on the same nodal points. This may be due to the fact
that the compatibility conditions between quadrilateral panels are, in
general, violated when using the kinematic pattern laid down in
Appendix II*. On the other hand, for the special case ofthe parallelo-
gram clement it is possible to generalize in a consistent manner the
corresponding investigation for rectangular panels of Ref. 1, which
does not violate the compatibility conditions along the boundaries.
The detailed derivation of the stiffnesses for the parallelogram is
reproduced in Appendix IV. We have introduced the parallelogram
element in a number of applications covering diffusion of loads into
oblique panels and swept-back wings not tapered in plan view. The
experimental evidence in hand indicates good agreement with theory.
A particular diffusion example is illustrated in Fig. B.7, and shows
convincing evidence of the reliability of the parallelogram and trian-
gular idealizations. We note en passant that for all but the triangular
* See, however, the remark at the end of this Appendix.
49
clement, the assumed kinematic pattern does not yield a constant

stress and strain pattern in a panel. In our present discussion on
stiffnesses we did not refer to flanges or stringers reinforcing the cover.
Since these have been investigated in Ref. 1 there is no point in
reproducing the expressions here.
We may round off this sub-section by reproducing the unit displace-
ment formulation for the stiffnesses which has been used by us in all
investigations to date. To obtain a concise presentation we apply the
following notation. Assume that the kinematic freedoms allowed to
the panel as part of the complete structure are n. Thus, for a triangle
(plane quadrilateral) n is equal to six (eight) since there are three
(four) nodal points, and two in-plane displacements are significant at
each nodal point of a membrane element. On the other hand if we
want to include in our evaluation of the stiffness matrix also the zero
coefficients corresponding to out-of-plane displacements n is increased
to nine (twelve) respectively. Let now e be the 3 x n strain matrix
€ = [ C 1 £ 2 . . . £ , . . .£„] (B.l)
corresponding to the n unit displacements at the vertices. In Eq. (1)
c, signifies the column matrix
«. = {««. e»v «••}. ( B *2)*
which determines the strains in the membrane due to the .th unit
displacement. Wc also introduce the 3 X n stress matrix o associated
with £. Then the n X n stiffness matrix k is simply given by
k = [*,,] = J..'«•'«-• dx dy = f J-Vx£ dx dy (B.3)
where
"1 v 0
B.l
1 - v)/2j
Fig. B.4. Alphagon panel.
is the stiffness of unit square element and / the thickness of the sheet.
Formula (3) has been applied in all analyses given in Appendices I
to IV.
• x , y signifies i n - p l a n e cartesian c o o r d i n a t e s .
50
We may generalize the principal reasoning underlying our investiga-

tions for triangular and quadrilateral panels of Appendices I, II and
I V to a polygon with a vertices by proceeding as follows (Fig. B.4).
In setting up the kinematic displacement patterns within such an
alpha-gon, wc note that to each of the n — 2a in-plane unit displace-
ments, which may be imposed at the a vertices, there correspond in
general n constants Ax to A,„ which may be considered as generalized
displacements. This assumes that we express the displacement functions
as polynomials which should be usually the most convenient specifica-
tion of the kinematics. The order of the polynomial depends on the
number of vertices and on the type of deformation variation selected
for our purpose. For example, for a triangle and quadrilateral, linear
and second order functions have to be chosen, on the other hand, for
a pentagon we may have either a second or third order function
depending on how many second order terms (i.e. x 2 ,y 2 , xy) are affixed
with a different coefficient A. This depends again on the variation
ofthe displacements along the polygon sides which we wish to prescribe,
the best distribution being a linear one. Anyhow, we may evidently
deduce from the assumed kinematic functions that for an arbitrary
deformation mode j associated with n = 2a given vertex displacements
«i . . . Ma, J-J . . . vx the following relation holds
V
Pi = ("r .) == {«! «2 * * * "a "l P« • • • Vm) = 4>,A, (B.S)
where <J>,, is a 2a x 2a matrix depending only on the vertex coordinates
x,,,y r and A^ is the (2a X 1) column matrix
A, = { A 1 A i . . . A„} (B.6)
Hence
A, = <fr- J Pi (B.7)
which determines uniquely the deformations corresponding to the
chosen mode j . Now if we consider the 2a unit displacement modes
at the vertices, we immediately derive from Eq. (B.7) that the 2a X 2a
set of coefficients
A |A, A , . . . A , . . .A„] (B.S)
is given by
A = -K1 (B.9)
In practice the inversion will usually be reduced to onc of an a X a
matrix for the reasons discussed in Appendix II. Having established
the matrix A, wc obtain from the kinematic functions and the standard
strain-displacement relations the 3 X 2a strain matrix £ due to 2a
unit modes in the form,
£-•>(*, j)A (B.10)
Here
£ |£, £ . . . . £ , . . .£„] (B.11)
51
where £, is given by Eq. (B.2). Note that the elements o f t h e 3 X 2a

matrix >|» are functions of x a n d j .
Substituting finally Eq. (B.10) into Eq. (B.3) we find the 2a X 2a
stiffness matrix k corresponding to the 2a in-plane vertex displacements
in the x and y directions as
k = J"J7A'«J>'X4»A dx dy = A'[j\(7ij>'xv|> dx dy]A (B.l 2)
The above procedure is effectively that used in Appendix II to derive

the stiffness matrix of the plane quadrilateral.
For programming reasons it may be convenient to include in Eq.
(B.l2) also the zero stiffnesses associated with the a out-of-plane w
displacements. This is immediately achieved in the computer by-
substituting the matrix
A -[A O] (B.13)
in place of A in Eq. (B.12). T h e matrix k is then of dimensions 3a X 3a.

The matrix k established in Eqs. (12) and (13) strictly refers to a
local coordinate system whose z-axis stands perpendicular to the plane
of the element or elements it serves. For such a system we use in
Appendix I the notation O'x'y'z' the corresponding stiffness being k ' .
Adopting for the moment this new notation, the problem arises of
deriving the stiffness matrix k associated with a fixed coordinate
system Oxyz which is chosen to describe the complete structure. T o
this purpose we require the 3 x 3 matrix <£ set up in Appendix I
which reproduces the direction cosines of the axes O'x'y'z' with respect
to those of Oxyz. We next introduce the 3a X 3a super diagonal
matrix
«, = [d C ttl (B.l 4)
which is a generalization of Eq. (1.16). T h e matrix k of dimensions

3a X 3a may now be obtained by an argument similar to that leading
to Eq. (1.40). Thus, we find
k = e,,k -a,, 15.15
C„ = T - « , T (B.l 6)
Here T is the 3a x 3a orthonormal matrix
EJJ r, 21 E 3 1
*-12 ''22 *"32
T — (B.l 7)
-Eia E 2ot E 3oL
52
where E l 7 are 3 X a submatriccs in which the only non-zero element

is a " o n c " placed in the .th row a n d j t h column.
It is often desirable to rearrange, prior to the assembly of the stiffness
matrix K of the complete structure, the stiffness matrices k of two
triangles into a single stiffness matrix k of the quadrilateral formed by
the two triangles. This is from the programming point of view par-
ticularly helpful if we intend to carry out an alternative analysis of" the
structure using the stiffness matrix k of a quadrilateral, as derived in
Appendix II, in place of the k's corresponding to the two triangles.
Considering the configuration shown in Fig. B.S, we note without
difficulty that
k - Mlk.M, | M,k,,M2 (B.18)
where k,, k,, are the 9 x 9 stiffness matrices of the two triangles and
M , and M 2 are the 9 X 12 transformation matrices (see Fig. B.S, for
the sequence of nodal points adopted).
Fig. B.S. Formation of quadrilateral from two triangles.
R g . B.6. Spar element.

53
IQOOO l b 10,000 Ib
10.000 lb
101 6 c m .
Parallelogram panel
a
~r- 'r'r~-/--r-y-r-r~7—
/ / / / / / / / /
/ / / / / / / / /
/
/ /
/ / / f / / s/ // // j/ / / X
-'
.
/ / /
S ^ ^ f S - A /
/ / / J -/ j i /- / / /,/_., ~
y
X
Parallelogram Grid •<
-
—
T
Experimental a a a a >
W
-
Crosswise distribution o l longitudinal strain
Triangular Grid
Fig. B.7. Diffusion in parallelogram panel.

+ 1 0 0 0
0 + 1 0 0 o o
0 0 0+1
+1 0 0 0
M, o 0+1 0 0 o
0 0 0+1
+1 0 0 0
o o 0+1 0 0
0 0 0+1
(B.19)
0 0+1 0
0 0 0+1 o o
0+1 0 (l
0 0 + 1 0
M2 o 0 0 0+1 o
0 + 1 0 0
0 0 + 1 0
o o 0 0 0+1
0 4-1 0 0
(B.20)
B.b. Idealization of Spars; Stiffnesses
The reader may remember that in our original matrix theory of
structures (Ref. 1) the stiffnesses for a uniform spar element were
established by applying the same modes of deformation as for a rect-
angular element in the cover. For a singly symmetrical wing this leads
to the specification of two kinematic modes at each end of an element.
This presumes, of course, that we neglect the compressibility of the
spar in the transverse direction. The two modes are shown diagram-
matically in Fig. B.S—where they arc actually applied to a tapered
element—and may be considered to represent a pure shear strain (for
a uniform element) and a rotation of one end which is not allowed to
cause any transverse displacements along the element. Now this
representation is, in principle, permissible since with decreasing length
of an clement the shear mode becomes the only determining factor.
However, as pointed out in the Preface, the convergence of the pro-
cedure for the deflexions is, in general, deficient, yielding in many
55
J . I I . A R G Y R I S , S. KELSEY a n d H . KAMEI.
Spar Element
Two Modes Three Modes

(Rel t)
Mode r. Mode r.
Mode b,
Mode s. Mode s.
Fig. B.8. Modes of deformation of spar elements.
cases a significant underestimate of the transverse displacements.

This is forcibly brought to our attention in Fig. B.10, which indicates
that the convergence deteriorates with decreasing ratio of flange area
to web area. It is evident that this shortcoming of our original spar
theory is due to the fact that our modes do not include the bending
deformation ofthe individual spar element—which is after all the most
significant factor for the deflexion, unless the spar element is very
short. O n the other hand, our original kinematic assumption can
claim the advantage that it does not violate anywhere the compatibility
between cover and spar. Moreover, the stresses in the cover are
estimated with good accuracy even when the deflexions are seriously
underestimated.
Before considering the alternative methods developed by us since
the appearance of Ref. 1, it may be of some relevance to refer to a
different solution of the spar stiffness problem given by Melosh and
Merritt, 3 to which brief reference was made in the Preface. These
authors again consider two modes only which may be said to arise
from a constant bending moment and a uniform shear. It is interesting
56
cm. in.
0-03 B - 0 2 in 1
2»7tJ 2 M70 • W in
0-06
Flange a r * a B-0 2 a a
R a n g * ar*a B-2 0 • •
R a n g * ar»a B-100 • • •
002-
0 04
1 - 200 In. - 508 cm.
h,- 33 In - 83fl cm.
h,» 3 in. - 762 cm.
oot
002
B - 2 0 in 2
04781 0*»77 t>*»17 a t o ' in
2
B - 10 0 in
01048 01I21 01158 • 10 J in.
ni
i -A e
fig. B.(.). Tip displacements "/ tapered spar.
and possibly not unexpected that if the Melosh-Mcrritt spar stiffnesses

arc used on a wing analysis, wc immediately notice an apparently
excellent convergence of the deflexions with increasing number of
stations. However, we know that the displacements are not estimated
correctly being, in fact, underestimated significantly- for a cantilever
structure. This has been confirmed also by independent computations
of the Boeing Transport Division. In effect, when the partitioning of
the spar is carried beyond a certain number of elements the deflexions
given by our original theory of Ref. 1 prove more accurate. In a way,
as pointed out in the Preface, the Melosh-Merritt technique derives
essentially from an idealization which is more akin to the force than
the displacement method. Gallagher has reviewed carefully the
Melosh-Merritt procedure in Ref. 7, and subsequent publications, and
it is superfluous for us to elaborate his comments except to point out
57
J . H . A R G Y R 1 S , S. KELSEY a n d H . KAMEL
that the stresses given by Ref. 6 do not prove more reliable than those
of Ref. 1.
Inevitably at this stage the question poses itself why the idealization
of the spar could not be achieved with a triangularization which is,
after all, so appropriate for the cover. A little thought shows, however,
that this approach cannot be rational since the antisymmctrical direct
stress distribution in the spar cannot be reproduced with reliable
accuracy by an assembly of constant stress triangular elements unless
we use a large number of elements. Figure B. 11 illustrates this aspect
and shows that eight triangles prove, in fact, less reliable than a single
rectangular element as analysed in Ref. 1.
As far as our own theoretical developments arc concerned, they are
based on a number of different assumptions, most of which make use
of prescribed kinematic displacements. Effectively, the solution to our
problem demands, as mentioned before, the inclusion of the bending
deformation of the individual element. This necessitates a greater
number of kinematic modes than hitherto allowed and we have
operated with three and four kinematic patterns. However, as far as
computational experience on large structures is concerned, this is
restricted at present to three modes which have given us excellent
agreement in the deflexions and cover strains in a number of applica-
tions. We included ab initio the effect of taper, and Fig. B.S illustrates
the three assumed modes to which there corresponds a 6 X 6 stiffness
matrix for each spar element. Two of the modes are effectively those
of the classical E.T.B. which neglects shear strains, whilst the last one
is again a pure shearing action. Of the two bending modes, the first
involves a translation of one end cross-section with respect to the other,
not allowing any rotation at either end. The second mode, on the
other hand, consists of the rotation at one end which is held at zero
deflexion. In fact, these two bending modes would suffice if shearing
deformation of the spar could be neglected, and are, indeed, the
standard displacement functions which one would apply for a deforma-
tion analysis of a rigid jointed frame. For the mathematical specifica-
tion of the two first modes, we use at present the E.T.B. deformations
of a uniform element. It follows that for a constant cross-section
cantilever, the deflexion derived from the present analysis is indepen-
dent of the number of the elements, being equal to that of E.T.B. plus
shearing effects. This is shown in Figs. B. 10 and B.ll where the tip
displacement plotted against the number of nodal stations is seen to
be independent of the chosen partitioning. However, this is not true
for a tapered cantilever, since the two bending modes arc, as mentioned
previously, strictly only true for a uniform element. At the same time,
the dependence on the number of stations is small, as indicated by the
example of a tapered cantilever in Fig. B.9. The detailed analysis of
the associated stiffnesses is given in Appendix I I I , where the complete
58
B-0
0 03
R-1
0 06
Three Modes
Spot Welded Flange
Three Modes
Integral Flange
Cn:
tid i
1-400 in - 1016 cm,
Two Modes. h - 2 5 * 4 i n - 64 6 c m ,
t - O 05 i n - 0 127 cm,
0 02
T04 B - 0 2 in'
0 01 / A
• ) .
/ s
\ / f
P\ * Af'
\ ' ' A
\• \ - ' '
fAf J
\ -^/V B - 2 0 In.
—\-f'' ' Z ~ - zr=r=
a^- ••» B - 10 0 i n
o • o.
1 4 8 12
fig. B. 10. Comparison of spar idealizations.
6 x 6 stiffness matrix is derived. T h e reader will note that two

transverse displacements arise at each station, one being due to mode
" 2 " , and the other to mode " 3 " . Hence when setting up the
R Kr (B.21)
relation for the structure, we observe that the transverse load R at
each nodal station where a spar is attached, has to be entered twice.
As referred to above, the present approach has yielded excellent
agreement on a number of wing specimens, and this is demonstrated
on a particular example in Section B.d. Nevertheless, a number of
open questions remain to be clarified. Thus, the cover-spar compati-
bility conditions are, in general, only satisfied at the nodal points, due
to the different kinematic idealization adopted in the respective sub-
systems. A number of investigations arc in hand to finalize the solution
59
J. H. ARGYRIS, S. KELSEY and H. KAMEL
m.
B-0
0 03
Two Modes
Three Modes
I n t e g r a l Flange
006 I - 400 in • 1006 cm.

h-25 44 i n . - 6 4 6 c m
1-0 05 in.-0-127 e n . . . » • '
002 Triangular
Elements
y
0 04 s
/
/
/ /
0 01
/ / .'
/ ' . S A-
0 02 / / , * A-
/ / / / B-2 0m*
.*«C .
A / ^ " B T 10-0 i n '

oi—i^ s "~""~'~
1 4 8 12
Fig. B.l 1. Comparison of spar idealizations.
to the spar problem—one of them being based on four instead of three

modes. I t should also be noted that any flange in the cover where a
spar web is provided is integrated with the spar and not the cover.
As we shall sec below, this is a much more satisfactory idealization,
especially when the number of spar stations is relatively small.
Before attempting the tapered spar analysis given in Appendix
III.A, we considered an alternative scheme which may be of some
interest. Thus, we derived for a uniform spar web the "exact" stiff-
nesses associated with the three modes shown in Fig. I I I . 2 . In contrast
to the simplified analysis, this takes account of the parabolic shear
stress distribution present in modes " 1 " and " 2 " . However, a finite
shear flow is allowed along the upper a n d lower edges of the web,
whose magnitude is determined by the condition that the transverse
shear force vanishes for zero Poisson's ratio. Naturally, this theory
60
involves warping of the web (see Fig. III.2) and an incompatibility

would occur at spar stations where a transverse load is applied. O n
the other hand, no difficulty arises for a single tip load. Again the
result is inevitably independent of the number of stations. Moreover,
as may be verified from Fig. B.10, the deflexions differ only insignifi-
cantly from those of the elementary theory of Appendix III.A. We
have applied the corresponding stiffnesses which are derived in
Appendix IV.B to a number of problems, and even took account of
taper by applying factors to the stiffnesses necessary to uphold equili-
brium. The real difficulty associated with this approach lies in the fact
that the flange cannot be integrated with the web, being effectively
only "spot welded" at the nodal stations. This leads to an underestimate
of the stiffness of the cantilever as may be seen from Fig. B.10, which
is all the more marked the larger the flange area is. The technique
is of no practical value except for demonstrating that the integration
of flange and web is absolutely necessary.
The 6 x 6 stiffness matrix k of the spar clement is referred to a
local coordinate system Ozx (see Appendix I I I ) , the Oz axis of which
will usually be parallel to the corresponding axis of the fixed Oxyz
system. At the same time the Ox axis of the spar will, in general, be
directed at an angle to the fixed Ox axis. Denoting for the moment
the lengthwise direction of the spar with s we require the stiffness
matrix kJ(, of the spar with respect to the fixed axes Oxyz. Wc prove
without difficulty that k„„ which is of dimensions 8 X 8, is given by
(sec Fig. B.6)
ksp = N'kN i-B.22)
where
cos {s,x) 0 cos (s,y) 0 0 0 0 0
I) cos (s,x) 0 cos {s,y) 0 0 0 0
N (B.23)
O. I,
B.C. Assembly of the K-Matrix

Having established the individual stiffnesses k„ of the s component
elements of the structure, wc write the stiffness matrix k of the un-
assembled elements as the diagonal matrix
k = [k1k1. . kf. . k.J (B.24)
T o obtain hence the stiffness matrix K of the assembled idealized
structure we may use, following the theory expounded in Ref. 1, the
congruent matrix transformation
K - a'ka (B.25)
61
J . I I . A R G Y R I S , S. KELSEY a n d H . KAMEL
where the matrix a expresses the relations between the kinematic

freedoms (nodal displacements) assigned to the individual elements
and the complete structure respectively. That the matrix a will always
be very sparsely populated is evident. Now if the specifications and
directions of the nodal displacements arc taken to be the same for the
elements and the structure, the non-zero elements are clearly reduced
to " 1 " . In fact, in this approach all geometrical data are included in
the k a matrices and in consequence the matrix a is seen to contain
merely instructions specifying in which addresses of the K matrix
individual entries of the sub-matrices k have to be entered and added.
O u r discussion on the matrices k„ in the two preceding sections—as
also in Appendix I—where we referred the stiffness of the elements
ultimately to the same fixed coordinate system 0,xyz is in line with
the present argument. In what follows we assume this to be, through-
out, the case. Of course, the technique is then in principle identical
to that advanced by subsequent authors (e.g. Ref. 5) who simply
prescribe the addition of individual stiffness coefficients to assemble
the complete K. At the same time this literal interpretation of the
algorithm expressed by the matrix a proves in our experience less
convenient for speedy programming and ease in checking. This is
particularly true if we introduce a number of devices "regularizing"
the a matrix which are discussed further below in sub-section (y).
Here we first describe in general terms the matrix assembly of Eq. (B.25),
and then proceed to the alternative technique of direct addition of
individual stiffnesses to form K.
(a) The matrix assembly oj K. We start by partitioning the a matrix
in s sub-matrices corresponding to the s stiffnesses k„. Thus,
(B.26)
Introducing Eq. (B.26) into Eq. (B.25) we obtain (see also Fig. B.12),
K (B.27)
= 1 KK*u
which is given in Rcf. 1 and used also in our current work. Now, in
view of the extreme sparsencss of the population of the matrix a, it
would be incongruous and waste valuable space in the fast store, as
also computing time, if we were to introduce the complete a„ matrices
into the computer. To circumvent this inflation of work a number of
62
techniques may be used. Thus, we may ignore in the specification of

the matrix a all zeros and merely indicate the position of the ( + 1)
elements within single word entries. In particular, we may assign to
the first half-word //I the row number i and to the second half H2,
the column number j of a particular unit element. This is shown
diagrammatically below for i = 58 a n d j = 107 within a 36-bit word.
II] Ill
111.1. 11.1.11 (B.28)
row n u m b e r i column n u m b e r j
We mention here parenthetically a possible extension to this tech-

nique. It is, for example, sometimes desirable to allow not only for
an instruction as to a ( + 1 ) but also to a ( — 1). This gives us the
freedom to affix the positive sign of the nodal displacements for the
complete structure and the component elements alternatively to the
same or opposite directions. This may easily be achieved by reserving
the first bit for the specification of the sign. Furthermore, we may
assign another bit to indicate whether the element is a " o n e " or a
"two". These two bits may prove advantageous if we want to use the
same basic programme for symmetrical and antisymmetrical loadings.
But let us consider again our present matrix a in its more prosaic
form containing only the addresses of (-f-1) elements. In our experience
it has proved invaluable both in the speed of writing the necessary
programmes and also in the ease of checking them to apply a so-called
kinematic regularization to the structure. Since we discuss this
technique in more detail below, it suffices here if we just mention that
it involves inter alia the inclusion of all degrees of freedom (actual or
otherwise) of the nodal points when setting up initially the a matrix.
The actual matrix a for the given structure is then obtained by extrac-
tion of the columns allied to degrees of freedom which have to be
suppressed.
At this point we refer also to another possible method of storing the
elements of a which we have not tried out but which may be advan-
tageous under certain conditions. This consists of storing all entries of
a (be they zero or otherwise) in series of consecutive bits. Although
the complete a is hence placed in the computer, each clement requires
merely a binary bit in place of a whole word. The procedure appears
advantageous for computers with variable length words in which each
bit is separately addressable. It should be economical where the
density of the unit elements exceeds say, one in 36 (assuming this to
be the standard word length).
We are now in a position to describe briefly the principles of the
assembly programme for K using Eq. (B.27). This is best discussed in
63
J . H . A R G Y R I S , S. KELSEY a n d H. KAMEL
conjunction with the simplified flow diagram of Fig. B. 12, showing

the formation of ajkâ,,, a typical component matrix of K. The first
main step involves the operation k„a„. For a specific instruction {i,j)
of a unit element this entails the transfer of the ith column of k„ to
the j t h column of the space provided for k 9 a„. Since more than one
unit element will, in general, appear in the jth. column of \%a our
process is seen to lead to the addition of two or more columns of k,.
Anyhow the operations merely involve transfers and additions. Having
obtained k ^ we premultiply with the transpose of a„. There follows,
for the same unit element in question, a transfer of the ith row of
k„a 0 to the j t h row of the store allocation cleared for K. Again in the
course of these instructions, rows of kâ,, will be added in accordance
with the unit elements present in the 7th column of a„. The square
matrix a ^ k ^ has, of course, the dimensions of K, say n x n. The
procedure is carried out in turn for g = 1 to J and yields the complete
stiffness matrix K since all component matrices are directed to the
same reserved store space. O u r programme also provides for elemental
stiffnesses of varying dimensions. For example, k„ for the cover is of
dimensions 12 X 12 but only 8 x 8 for the spar elements. The 8 x 8
matrices for the spar presume three modes of deformation and would
be enlarged to 10 x 10 if four modes were to be included. As an
illustration of the speed of this programme we mention that the setting-
up of a K matrix of 500 x 500 on the Remington-Rand 1107 required
less than 1.1 min. In this connexion it may be of interest that the
formation of the associated k and a matrices was achieved in less than
a minute.
(/?) On the Regular!zation of the Matrix a. Let us first recall that the
rows and columns of K correspond to the degrees of freedom usually-
defined as nodal displacements u t , vt, WA measured parallel to the axes
of the fixed system Oxyz. More than one type of displacement may be
assigned to a direction; for example, wb and w, in a spar corresponding
to the modes b and s (see Appendix I I I ) . In order to keep the pro-
gramme as simple as possible and make possible the development of
general purpose programmes, all prescribable directions at the nodal
points are allowed in the formation of a irrespective ofthe corresponding
degree which has ultimately to be suppressed or is ignorable. The
displacements to be suppressed are associated with actual supports of
the structure which do not allow one or more movements u, v or w.
Thus, even the rigid body freedoms of the structure are assumed
released when setting-up the a matrix, the structure being considered
to be in a free-free state in space. Now, while the kinematic freedoms
which have to be suppressed are self-evident, the concept of the
ignorablc kinematic freedoms is somewhat more subtle. In fact, the
ignorable freedoms are in their most elementary form, kinematic
directions at nodal points where no forces can be applied since the
64
General view of K assembly
k Formation of k a a
a *o
(diagonal) ( s p a r s e l y populated )
i.
All data connected
with geometry are " k„

a3
"® ,
included in k
mmw.
memory
J —
-
-
=
Hence all e l e m e n t s location
o
-
i n a a r e either o n e •/-,
Formation of K0 = a ^ ( k g a 0 ) and finally K c

or z e r o . x
- *s-l a s -i -J-.
Ul ~
X
K = a'ka = l_*a kt7i

0.1
u X
O
-
-
>
•
y
Only non-zero elements of a 9 are stored. >
r
When a l l c o m p o n e n t <
Computer memory matrices K) to Ks
location.
are c o m p u t e d matrix
L
WTWYhJ K in store
memory location
Fig. B. 12. Diagram for assembly ofK..

J. II. A R G Y R I S , S. KELSEY a n d H . KAMEL
structure displays no corresponding stiffness. Thus, it is not the dis-

placement r but the force R which is necessarily zero in such a direction.
For example, this is true for the direction normal to a membrane cover
at a nodal point where no spar or rib is attached (see Fig. B.l3).
However, in practice the ignorable degree will not present itself as
simply since the stiffnesses are measured along the fixed directions
Oxyz, the Oz axis of which is tilted away from the normal to the cover.
All stiffnesses are hence finite but one degree is again ignorable since
the stiffnesses are linearly dependent, the matrix being of degeneracy-
one. Alternatively, we may say that a kinematic mechanism links the
Fig. B. 13. Ignorable degree of freedom.
directions Oz, Ox, Oy. Now, if we eliminate an ignorable degree of

freedom, we do not impose any strains on the structure since a zero
displacement in this direction may be considered achieved by the
superposition of a rigid body motion with respect to the adjoining nodal
points. Hence the displacements along the prescribable true degrees
of freedom are not affected by the elimination of an ignorable displace-
ment. The argument is illustrated in Fig. B. 14, on the simple example
of a two-dimensional rigid jointed frame. Interestingly enough, any
of the degrees of" freedom along Ox, Oy, Oz could strictly be ignored
but it is evident from the conditioning point of view that the axis
nearest to the normal, i.e. Oz, should be chosen for this purpose (see
Section B.d).
It is clear that the a matrix thus set up will display a high degree of
regularity even if the structure is complex. Moreover, we may intensify
the rcgularization effect by introducing, if necessary, any number of
additional elements. This may go as far as prescribing additional spars
and ribs if this is considered helpful in establishing the programme for
a. Now, in order to obtain the actual a matrix we have to extract both
the suppressed and ignorable degrees of freedom. This is achieved
by a simple and standard subroutine which collapses the matrix
columnwise, striking out the supernumerary columns and modifying
66
Z',
Ignorable degree of freedom of A
l"
Good choice of ignorable degree of freedom
Bad choice of Ignorable degree ol freedom
Fig. B. 14. Kinematic freedom in a frame.
the addresses of the unit elements in the following columns. Let us

consider the specific example of a regularized matrix a of dimensions
2500 x 500. The columns 2,6,235,239,451,455,496,500, arc taken to
correspond to actual supports and have consequently to be suppressed*.
This reduces the size of a lo 2500 x 492. The corresponding stiffness
matrix K is of order 492 X 492 and is not degenerate as the 500 X 500
would be. The standard subroutine requires merely instruction lo the
computer quoting the number of columns and their position. Thus,
in the present case
(8 — 2,6,235,239,451,455,496,500)
the change of the addresses follows automatically.
We have found the above procedure exceedingly simple to handle
allowing the speedy programming for large scale structures with many
* Note that this duos not imply that there arc t-iglil supporting actions. For example, if
the restraint refers only to the z direction and both n , and iv, modes arc allowed in the spars,
the number of supports in the present example would be only four.
67
hundreds of unknowns. T h e ease of checking flow diagrams and

programmes achieved by the high degree of regularization has been
particularly striking.
T h e reader will notice how the principle of regularization enters
differently in the force and displacement methods. Thus, in the force
method we seek structural regularization to ensure a standard forma-
tion of the basic matrices b 0 and b , . In the displacement method on
the other hand, we pursue a kinematic regularization (which may or
may not involve additional elements) in order to achieve a repetitive
formation of the basic instruction matrix a. Also the elimination of
the supernumerary structural elements is initiated in the force method
after the completion of the analysis of the regularized system whilst
in the displacement method the suppression or ignoration of degrees
of freedom is carried out on the matrix a itself prior to the assembly
ofK.
[y) The "Long-Hand" Formation of K. We briefly discuss below the
outline of a programme for the formation of K by direct instruction to
the individual entries ofthe k a matrices. The basic steps ofthe routine
are as follows:
(1) The unknown displacements or degrees of freedom of the struc-
ture are specified (1,2 . . . «). This may also involve an
instruction as to how the matrix is to be partitioned.
(2) The dimensions n X n of the K matrix having been fixed, a
corresponding place is cleared in the large backing store of the
computer. Naturally, depending on the sizes of computer and
matrix, storage space may have to be provided elsewhere.
(3) The k matrix is computed by special programme and stored,
the corresponding addresses being in the meantime suitably-
registered.
(4) The computer is given a set of instructions which may read as
shown in Table B.l (p. 69).
The computer will read these instructions one by one and proceed
to add the elements of the k matrices into the proper address of the K
matrix.
As it stands the method is much more cumbersome than that of the
assembly with the help of the a matrix. On the other hand, the
statements themselves could be "calculated" by a proper algorithm.
B.d. Some Further Comments on the K Matrix

We discussed under B.c. (a) and B.C. (/?) the main principles under-
lying the programme leading to the formation of the stiffness matrix
K of the complete structure and turn now our attention to two im-
portant aspects concerning its nature and morphology. We start by
considering the typical sub-assemblies of an idealized membrane
68
Table B.l
Addresses of Size Displacements associated with the element in question.

stiffness k,, ofK set out in the order they arise in k„
Drum —10(11 6x6 375, 376. 412, 413, 7,8

Drum—1037 6x6 376, 377,413,414, 18. 19
Drum—1073 6x6 377, 378, 414, 415. 28, 29
Drum—4601 6x6 1, 15. 24. 37, 9, 25

Tape 1 — 101613 8x8 2, 15. 37, 49,414,415, 712, 730
Tape 10—4740 12 x 12 110. 1. 12, 153, 24, 26, 751, 1024, 9, 111, 14, 512
structure shown in Figs. B. 15 and B. 16, which may be taken to form

part of, say, a wing cover. The first is flat and as we noted in sub-
section B.c (a), displays no stiffness normal to the plane. In setting-
up the matrix a we have hence to extract one column in order to
eliminate the ignorablc kinematic freedom. Wc pointed out that it is
Fig. B. 15. Flat stiffened membrane.
Fig. B.16. Idealized curved stiffened membrane.

69
J . H . A R G Y R I S , S. K E 1 S E Y a n d H. KAMEL
best to choose for this purpose the direction which is nearest to the
normal—in the present case evidently Oz. When we view, however,
the membrane assembly of Fig. B.l5 we observe that it offers resistance
to displacements in the x,y,z directions and that as a consequence all
degrees of freedom must be fully allowed in the matrices a and K.
But the case does easily arise in wing structures that the curvature of
the cover of which Fig. B.l6 may be part is small. Then the stiffness
normal to the cover, and as drawn in Fig. B. 16 probably also in the
Oz direction, may be much smaller than the stiffnesses associated with
the other two directions nearly tangential to the wing cover. This
may lead to a local bad conditioning of the K matrix. That this must
occur follows immediately from the limiting case when the cover
becomes plane and the matrix K singular—an extreme case of ill
conditioning. Two possible methods of circumventing this source of
trouble are suggested.
(1) We may consider, for example, the offending displacements as

secondary "redundancies" (unknown displacements) and "elimi-
nate"* them first in the similar way as we close the rings in a
fuselage before solving for the primary redundancies in the
cover. This involves a subroutine which rearranges the matrix
a by transferring the trouble-causing columns to the right-hand
end and renumbering the other column entries. The complete
matrix K will now appear as
Kn K12
K = (B.29)
K21 K.,.,
where the suffices " 1 " and " 2 " refer to the primary and secondary
degrees of freedom respectively. Eliminating the secondary-
displacements by the procedure set out in Ref. (1) we obtain,
K n = K u - K 1 2 K W *K 2 1 (B.30)
where K n is now a fully populated matrix.

(2) Another technique is that of the local orthogonalization similar
to that applied in the fuselage analysis. Wc may, for example,
determine for one nodal point or even a group of adjoining
nodal points, a set of directions for which the stiffnesses are
orthogonal, thus not giving rise to any coupling terms. Selecting
then these directions as alternative kinematic freedoms, we
effectively diagonalizc the corresponding sub-matrix of K and
isolate the trouble. This may be achieved by a local transforma-
tion matrix applied to the relevant part ofthe original K matrix
* Note that the elimination is to be understood here as an initial solution for the secondary
displacements which do not appear subsequently as unknowns.
70
or an equivalent re-evaluation of the k matrices along the new-

local directions. After solving the transformed system of equa-
tions, we refer again the displacements and forces to the original
x,y, z directions.
The first method should always prove satisfactory in modern wing

structures. At the same time it leads to a fully populated stiffness
matrix K n which does not allow us to use a programme for a band
matrix inversion which is valuable for large matrices of order greater
than, say, 300 x 300. Also if we have to analyse a membrane tubular
structure the upper and lower faces of which are at a steep angle, the
first procedure would not lead to a substantial improvement of the
local conditioning for the upper cover if the x-y plane were more or
less tangential lo the lower cover.
The second technique can easily be automatized. The programme
determines at each joint the orthogonal directions and may even
suppress a column in the matrix a if the corresponding kinematic
freedom is ignorable which may be ascertained by a zero eigenvalue.
Certainly whilst this method orthogonalizes the diagonal sub-matrices
it cannot achieve this for the off-diagonal ones. In general, this should
not be a source of ill-conditioning at least as long as there is no violent
change in the form ofthe structure (see also the similar argumentation
in Refs. 3 and 4 illustrated with examples on the orthogonalization
technique in fuselages).
The last item to be taken up in this section concerns the sequential
Gs
I** ,GJ ,GI ,GI
J- i i
Fig. B. 17. ExampU of grouping of nodal points.
Fig. B. 18. Example of grouping of nodal points in a membrane roof.

71
order of the displacements in the matrix r . For convenience of pro-

gramming and assembly, we order the kinematic freedoms into groups
which need not have the same number of nodal points (see Figs B. 17
and B.l8). Let us consider the effect of three different arrangements
of r on the form of the matrix K.
(1) If in each group G, the displacements u, v and w* are collected
in sequential column matrices u„ vt and w{ then the stiffness K
will appear as the three-band matrix of Table B.2, each sub-
matrix of which consists of 9 sub-submatrices. In its turn each
of these sub-submatrices is formed by a three-band matrix with
scalar elements. This method was actually applied to the
example discussed in Section B.e.
T —•—-•-<
G
Pi
G,
Table B.2
J
I
* 2 im _ t
G,J
^ _ ^ _ - in i ••.- i w r n i w — ——•
m],i
k- I * ' *— •'' *
Table B.3
(2) The second alternative arrangement differs from the first only
in that the u, v, w displacements within each group G, are col-
lected pointwise, starting with u n , v n , w a and followed by
* For brevity we refer here only to w; clearly W may consist of independent wb and w,
kinematic freedoms.
72
11,2, *-i2> w nr c t c * Then each sub-matrix of Table B.2 will be

itself a three-band matrix containing fully populated sub-sub-
matrices as indicated in Table B.3, for a typical sub-matrix.
In this case the complete matrix K may be represented as follows:
Table B.4
Now the order of these sub-submatriccs will be for a membrane

structure (3 X 3) or (4 x 4), assuming that not more than four degrees
of freedom are prescribed at each nodal point. Evidently, if in the
programme we can allow for all zero sub-submatrices shown in Table
B.4, the computing time will be enormously reduced.
(3) The last order provides for three main sequences consisting of
all u's, followed by the v's and finally the o-'s. In each of these
groups the displacements are arranged according to the num-
bering of the nodal points. The complete K matrix will now
appear as in Table B.S, where each sub-submatrix consists as in
the first arrangement of three-band matrices of scalars.
From the programming point of view the most economical way of
forming K is that described under (2) since all non-zero sub-submatrices
are fully populated. On the other hand as far as operations on the
matrices, e.g. inversion, are concerned, arrangement (1) is more
advantageous. The third grouping is not to be recommended. As
mentioned before if we want to eliminate certain w displacements, they
are placed at the end of the r vector but this rearrangement is best
73
J . H . A R G Y R I S , S. KEI.SEY a n d H . KAMEI.
G2
—
1
' G,
0,
fG,
Gj
___ ,,, .
Table B.S
achieved by a completely automatic programme operating on the a

matrix.
The matrix displacement method expounded in these sections has
been used to analyse complex structures with many hundreds of
unknowns, both in our Institutes and in industry. By considering more
extensive systems as assemblies of sub-structures which are at first
computed independently, we may determine the deformations and
stresses of the complete system by a final investigation involving the
actual kinematic links between the sub-systems which may be denoted
as tertiary displacements in analogy to the tertiary redundancies in
the force method.
B.e. Experimental Work on a Wing Structure

In order to provide experimental data for comparison with the
results of displacement method analysis, tests were carried out on the
wing-like structure illustrated in Figs. B.19 and B.20. It consists of a
non-conically tapered five-spar section, with ribs at the tip and root
only, overhung from a triangular root portion. In crossing the root
rib, the spars arc bent through a considerable angle to run parallel to
one of the sides of the root triangle. Both upper and lower covers are
flat, for manufacturing simplicity, but are so inclined as to yield con-
siderable taper over the structure as a whole. As can be seen from the
cross-section in Fig. B.20, the horizontal plane is a plane of symmetry.
The perverse form of the structure and its somewhat capricious
geometry were deliberately chosen to provide a severe test for the
method of analysis. Nevertheless, it is, in its main features, character-
istic of a swept wing and provides a natural example for the exploitation
of the displacement method.
74
Fig. B. 19. View of tested wing structure.
Monaycomb Cora 0 15 in
0 37 cm
Croat Section XX
Support Frame*,
Fig. B.20. Geometry and construction of wing structure.
In the construction ofthe model aluminium alloy, honeycomb sand-

wiches were used throughout for the rib and spar webs, as well as for
the covers. All joints were made by Redux bonding or with Araldile
epoxy-rcsin adhesive, bolts and rivets being used only for location and
temporary attachments while the adhesive was cured.
For the tests, the "wing" was mounted horizontally on three ball
supports at the vertices of the root triangle, the ball supports being
designed to provide only the requisite three kinematic constraints in
75
the horizontal plane. Loads were applied, normal to the mid-plane,

over local areas of the surface through a loading pad with a rubber
covered surface, by a loading lever which spanned the structure in an
approximately chordwise direction, one end of the lever being sup-
ported by a vertical link while the other was attached to a powered
screw-type straining unit (sec Fig. B.20). Direct measurement of the
load was by a hydraulic capsule with a ball seating on the loading pad.
T o obtain a uniformly distributed load over the overhung portion ofthe
"wing", a tailored pressure bag was inflated between its upper surface
and a flat, externally supported platen.
To obtain experimentally the deflexions normal to the middle
surface, dial gauges were mounted on a stiff datum frame with a three-
point attachment to the support points of the wing itself. Direct strains
in the spar flanges and in the cover were measured with wire resistance
strain gauges.
The theoretical analysis made use of triangular panel elements for
the cover, the stiffnesses of which are given in Appendix I. For the
spars, the kinematic deformation modes described under B.b, and in
Appendix I I I were used. Two analyses have been performed, using
the coarse and fine grid patterns shown in Fig. B.2 la and b, the fine
Fig. B.21 a. Coarse grid for wing.
Fig. B.2 lb. Fine grid for wing.
grid being obtained by introducing intermediate nodal points between

those of the coarse grid. The axis system is shown in Fig. B.22.
76
Fig. B.22. Axis system for wing.
In the coarse grid analysis, all nodal points lie on spars and hence
the only rcgularizalion required of the complete a matrix was the
suppression of the six vertical displacements (3 b and 3 s modes) at
the support points leaving a total of 1 74 displacements. The prelimi-
nary assembly of the triangular panels into quadrilaterals is described
in B.a Eq. (B.l 8), while the formation of a (800 rows) which, it should
be noted, includes the cover triangles for b and s displacements and
the final stiffness matrix K, are in B.c and B.d.
Exclusive of the ignorablc displacements at 56 grid points which do
not lie on a spar and the six vertical displacements at the support
points, the order of the stiffness matrix for the fine grid is 494. Its
inversion, on a Remington Rand 1107, occupied 55 min as compared
with 5 min for the small grid K.
Some results obtained from the analysis with the rough grid are
shown in Figs. B.23 to B.30, together with experimental points, for
1500 Ib Experimental
680 kg Calculated
Fig. B.23. Deflexions of wing Load case A.
77
J . H . A R G Y R I S , S. KELSEY a n d H. KAMEI.
three cases of loading. The shapes of the deflexion surfaces are seen to
agree very well with experiment, though as would be expected, the
rather coarse kinematic idealization ofthe cover tends to underestimate
the deflexions.
The theoretical distributions of flange direct stress are based on
values calculated at the middle of the spar elements, as in Appendix
III. It is mentioned there that there is good agreement between those
calculated on the spar idealization [Eq. (III.31)] and those derived
Experimental
Calculated
F"ig. B.24. Deflexions of wing—Load case C.
5 lb. Experimental
Calculated
Fig. B.25. Deflexion of wing—Load case D.

78
Experimental • • •
Calculated
Fig. B.26. Direct stresses in spar flanges—Load caê A.
Fig. B.27. Direct stresses in spar flanges—Load case C.
Fig. B.28. Direct stresses in cover—lj>ad case C.

79
J, H. ARGYRIS, S. KELSEY a n d H. KAMEL
0 35 kg/ ,
Fig. B.29. Direct stresses in spar flanges—Load case D .
5
Xrl
035k
«/araX
Fig. B.30. Direct stresses in cover—Load case D .
from the cover triangles [Eq. (III.35)] at these points. This is in spile
of the lack of compatibility between the deformation modes of these
elements. It is noticeable that agreement with experiment is better on
the front two spars, where it is very good, than on the rear spar. This
is, however, not surprising, since the latter suffers much more from the
roughness of the grid system, being divided into only four elements as
against eight elements on the front spar.
80
C. MATRIX FORCE M E T H O D IN NON-LINEAR PROBLEMS
C.a. Introduction to an Elasto-Plastic Analysis

In the following paragraphs we consider a few examples of the
application of a generalized matrix force method to non-linear prob-
lems. We start with a presentation of an analysis for elasto-plastic
systems. It was pointed out in our original publications (see, for
example, Rcf. 1) that the solution of this problem effectively only
requires the determination of the "generalized" initial strains H
corresponding to the plastic deformation of the component members.
In fact, the by now standard formula for the stress matrix S for a linear
system under known loads R and unknown initial strains H may be
written as
S - bR b.D.i'blH (C.l)
which we may put in the form

S = S L - (3H (C.2)
where
p = baD./b; (C.S)
In Eq. (C.2) S,_ may be assumed to represent the generalized stresses

in the linear state due to any cause, be they external loads R or known
initial (e.g. thermal) strains H„. Hence, Eq. (C.2) may also be used in
the elasto-plastic regime as long as the strains H are correctly assessed.
The elasto-plastic theory developed below actually allows also for
unloading of members initially subjected to plastic straining. For
simplicity of presentation we refer in our discussion to a sequence of
positive load increments starting from a purely elastic straining of the
system. The method is, however, equally applicable to any form of
loading cycle. Moreover, it may easily be extended to account for a
collapse or buckling of individual members. In extremis such a behaviour
amounts to an elimination or cutting-out ofthe corresponding elements
and leads to the verification (or otherwise) ofthe fail-safe characteristics
of a structure.
Before proceeding to the mathematical derivation of H and S we
introduce a number of useful physical and analytical concepts.
C.b. The Idealized Stress-Strain Relations

Figure C.l shows a stress-strain curve typical of materials used in
aeronautical structures. It is advantageous for computational purposes
to approximate the diagram by an analytical expression. Inevitably
it proves thereby necessary to reach a compromise between the degree
of accuracy (often anyhow illusory due to the scatter of experimental
results) and the mathematical convenience of the proposed formula.
81
Well known empirical relations arc inter alia those of Ramberg-Osgood*

(power law) and o f t h e R.Ae.S. data sheets (exponential law). The
incorporation of either formula in our elasto-plastic analysis of redun-
dant systems is, in principle, possible. However, the considerable
effort in the associated process of iteration, which may be based, for
example, on the Newton-Raphson technique, and the (apparent) high
degree of the achieved accuracy cannot, in general, be justified in view
ofthe uncertainties in the initial assumptions. These refer, in particular,
to the idealization of the system and the onset of plastic deformations
in a two- or three-dimensional stress state. In such cases it would be
Fig. C. 1. Typical stress-strain diagram.
necessary to operate with an appropriate yield criterion and to intro-

duce, for example, the Prandtl-Reuss equations. The presentation of
the corresponding theory would exceed the scope of this paper although,
in principle, not unduly elaborate. For the present purpose of demon-
strating the main effects of plasticity it is hence reasonable to introduce
an idealized stress-strain diagram which while as convenient as possible
docs not mask or modify to a significant extent the physical phenomena
involved. To this effect we use a funicular stress-strain law consisting
of a number of straight lines as illustrated in Fig. C.2. For the present
expose we restrict ourselves to two straight lines the intersection of
which represents the nominal yield stress au. We also assume that but
for the sign the diagram is identical in compression and tension; the
specification of different yielding conditions for positive and negative
stresses is actually straightforward. The gradients of the two lines are
E and E, respectively, where E, is the so-called tangent modulus.
Unloading at a point A always takes place parallel to the first or
Hookean segment as shown in Fig. C.2. In fact, any straining (either
positive or negative) of the material starting from a point C within the
range AB will be purely elastic unless the loading increment carries it
* Sec, for example, Ref. 13.

82
beyond the point A, whereupon positive straining again takes place

along the second segment. Our above discussion describes the uniaxial
stress-strain behaviour but the argument is the same for shear.
cola, t i/E
c o m , = WE,
E, 8 e
F"ig. C . 2 . Idealized stress strain diagram.
The mathematical presentation of the stress-strain relationship is

now
a
e = for a ^ o*„ (C.4)
a f
E (E El ^ ~ ^' °r incrcasin
S a
\ > \a*\
It is convenient from the mathematical and programming point of

view to introduce the following non-dimensional transformation of
Eqs. (C.4). Thus, writing
EE E
and y = - (C.S)
where av is here the positive yield stress and e„ the corresponding

strain we may express Eqs. (C.4) in the form (see Fig. C.3)
for 1 ^Tt^ 1
and
i
y = T -f- IT(T T 1)> 'or T . (C.6)
Here -n, denoted by us as the plasticity factor, stands for
(C.7)
= ( ! - )
Formula (C.6) applies not only for a state of direct stress but also
one of shear with the proviso that G/G, is substituted in Eq. (C.l) in
place of E/E, and av is taken as the yield stress for pure shear.
83
' m .cot p j - c o i p ,
cotp, < 1
cotP, a E/E,
V Y
Fig. C.3. Non-dimensional stress-strain diagram.
The initial or plastic strain arising in the second of Eqs. (C.6) is
rj = rr{r - 1) = =S (C.8)
A computationally very convenient unification of the two formulae

(C.6) may be achieved by introducing the function
HI 1 -IT -11] (C.9)
which takes in turn the values, zero for — \ •= T •= \, (T — 1) for

T ^ 1, and (T -f- 1) for T ^ 1. It follows that the general expression
of the chosen stress-strain law is simply,
y = r + TTTA (CIO)
where
•? = -"A (Call)
represents the permanent straining.
C.c. The Generalized Initial Strains for Typical Elements

For brevity of exposition we derive here only the generalized initial
strains, associated with the permanent straining •/, for uniform flanges
under constant end load and linearly varying flange loads and uniform
walls under constant shear flow.
(a) Uniform flange under constant end load. For a flange of length /,
simple physical argument yields
II le y TTT A — A c Tj (C.I2)
where
Xc = TTEyl ( C I 3)
84
(11
i-.r.. z/l
(a) Linearly varying

tlang* load
and stress
(b)Compi*i*l, p i * .
elk - .-
(c) Partially plastic

W Hang*
q = 1/1
i-5= • i
(d) Actual »h*ar flow
and Hang* load
du* to und load
-^rrJTTTmTTTllTTM
q»"E,i
1*J'" iZ
(•) 'Ers*t-r' component
l/l-U.,1
shear Hows
1
•" • I .
p^rrrflTiTM
(f)TrMti'component
Hang* loads
(t-r,)
l--a-rrrn?TfrTTTTTT-r-r-r-r^.
Fig. C.4. Plastic strains in flanges with linearly varying load.
85
J . H . A R G Y R I S , S. KELSEY a n d H . KAMEI.
(/?) Shear panel. For a panel of dimensions I x d carrying a shear

force (field force) S along the edge / we obtain (e.g. by the unit load
theorem)
H = e
' " j j Tt n l dx dy = e"dr> = A,TA (C.14)
where
TTE u d (C.15)
Note that for shear E„ = ayjG,

(y) Uniform flange under linearly varying end load. For the third
type of member we first consider the most elementary case, when the
complete flange (1,2) of length / experiences permanent straining (sec
Fig. C.4). If the stresses at the two ends are TX and T 2 respectively,
wc note that for the plastic deformation in question r A 1 ^ 0, T A 2 / 0.
T h e corresponding ( 2 x 1 ) generalized initial strain matrix H may
immediately be written down from Eqs. (24), (25) in Ref. 12 in the
form
'Hi 1 "2 1"

H
H%. 1 2_ f'i.
(C.16)
TTE y l 2 1 •Al
X,T A
1 2
where
2
X, and ( C I 7)
.1 'A-.JJ
However, Eq. (C.16) is not valid when the plastic straining extends only
over part of the flange. Figure C 4 ( c ) illustrates the case when
T
A2 ^ 0> Du>t T A 1 = 0 and the initial straining affects the length £2/.
Here
'A2
•:* = (C.18)
Similarly, if T A 1 -f 0 whilst T A2 vanishes, permanent straining occurs in

the range C.l from the end (1) where
T
A1
£. (C.18a)
For programming purposes a marked simplification is achieved, if

we can develop a single general formula for the matrix H , applicable
in all cases of partial or complete permanent elongation. This does
not appear, at first sight, easy to accomplish but the following device
86
in the application of the unit load method will be found to lead to the
required formulation.
Thus, we initially assume that the shear flows equilibrating a unit
load, say applied at (2), extend only over the plastic range of the
flange. The associated generalized initial strains, denoted by H, are
evidently given by a formula similar to Eq. (C.16) but with £2/ in place
of / and r A 1 = 0. Now, since the equilibrating shear flows must act
along the whole length / of the flange it is necessary to superimpose a
self-equilibrating shear flow system [illustrated in Fig. C.4(e)] which
couples with the plastic strains and leads to another set of generalized
strains H M . Using an additional logical operation we may unite the
total initial strain matrix H as
H = H, HM
= [X, + X „ ] T A X,T A (C.19)
where
(C.20)
6 J 2J
and
nE v {l - l ) l r I , -£,1
>M, (C20a)
6 [Ha Ul
and
= Ci + u (C.21)
As may easily be verified Eqs. (C. 19) to (C.21) give the correct generalized
H strains under all conditions of plastic deformations. It should be
noted that the "lengths" ). depend on the plastic stresses r A1 or T A 2
only when the plastic straining is partial.
Assume now that the members, under complete or partial permanent
strain, are specified by / stress stations. We write the (/ x 1) column
matrix H , of the corresponding generalized strains in the form
H, = A,TAI (C.22)
where
A, = fX| (C.23)
is the diagonal (/ X /) matrix of the X's, the (diagonal) elements of

which are 2 x 2 matrices or scalar values if they correspond to flanges
or sheet elements respectively. Clearly the ( I x 1) column matrix
T S , must include the zero elements associated with any elastic stations
of flanges in a partial plastic state.
O u r above derivation of the generalized strains is restricted to
uniform idealized elements but the theory can easily be extended to
cover non-uniform elements and other members with a more complex
stress specification.
87
J, H. ARGYRIS, S. K E L S E Y a n d H . KAMEL
C d . Determination qf Stresses and Deformations

We now proceed to the formal solution of the elasto-plastic problem
and assume first for simplicity of argument that members are either in
the fully clastic or fully plastic regime. Moreover, we specify that for
any positive load increment no unloading takes place in any element.
In the subsequent elaboration of the theory these restrictions will be
lifted.
We start by defining the diagonal matrix P„ of the yield loads at
all prescribed stress stations. In elements with a linearly varying force
(e.g. flanges) the yield loads corresponding to the two end stations (1)
and (2) must be listed separately. We have, introducing the 'area' A,
P„ - \Aa,\ (C.24)
Premultiplying Eq. (C.2) with P v ' we obtain
T = TA - P p - p H (C.25)
Introducing the initial strain matrix H
H - {O, H,} (C.26)

and using Eq. (C.22) we find
x = x L - P„ \ % A f i u (C.27)
where (3,, is the submatrix of p containing the / columns corresponding
to z A l . T h e partitioned (3 matrix appears thus as
Pf/f/ P'//
P = [P„ P.,1 (C.28)
.Pi. PIIJ
in which the second form will presently be found useful.

Let us now consider the / stresses T, specified in the members which
are at least under partial plastic straining. For these stations Eq. (C.27)
becomes
T, == x L l - PaTrtiiAfti (C.29)
where (3,, is defined in Eq. (C.28). We next introduce the concise

notation
B„ - P „ , l p „ A , (C.30)
which is a non-symmetrical matrix and rewrite Eq. (C.29) as

T, = x L l - B „ T A , (C29a)
This equation is best solved for T A , by the following device. We

subtract from both sides of the equation
HI-T.-e.l-l'.-e.l] (C.9a)
88
The left-hand side reads then TAi and a simple argument shows that
the first term on the right-hand side may be expressed as T / A I where
*L&I = *LI ~ t [ | - » i i - e,| - |T/., - e,|] (C9b)
It follows that Eq. (C29a) becomes
* A I = TLA, - B „ T A , (C.29b)
(II
TA/ = n„T y . A , (C.31)

Here
n „ = [I + B „ | • (C32)
It is interesting that the matrix [I + B„] of order (/ x /) can never
be singular in practice. At the same time it is possible to reduce its
order by excluding the zero elements of T A ( associated with the elastic
stresses in partially plastic members. This may be achieved by sup-
pressing the corresponding rows ofb 1 ( and columns of A, but the method
is not so attractive from the programming point of view.
The neat Eq. (C.3I) which expresses the unknown T A I in terms of the
known linear increments T A A „ represents in conjunction with Eq. (C.27)
the complete solution of our problem. We may unify the presentation
by substitution of Eq. (C.31) into Eq. (C.27) to obtain
T = xL - B.,!!,.-^, (C.33)
where
B., = P v l P..A, (C.34)
We observe that i f / i s known as to number and position, Eq. (C.33)
yields a direct determination of the elasto-plaslic stress distribution
subject only to the two restrictions initially mentioned in our argument.
It is interesting that Eq. (C.33) can also be derived by the modification
technique developed in Refs. (11) and (12).
We now proceed to lift the restrictions on our theory and start by
allowing also a partial yielding of members as may occur in flanges.
The problem may be solved by a number of different approaches, all
of which are necessarily based on an iteration technique. For example,
we may assume that the state of all members subject lo plastic straining,
but for a single one under partial plastic, straining, remains fixed at
the values computed in the preceding iteration cycle. We may then
set up for this one member a cubic equation in £ the solution of which
determines the extent of its plastic state. The initial condition of the
complete system is correspondingly modified and it is hence necessary
to repeat its analysis by direct solution or iteration. A new cubic
equation may next be established and solved, the process of iteration
being self-evident. T h e influence of other members under partial
plastic straining may be similarly accounted for. T o anyone versed in
programming techniques it will be apparent that this method, whilst
89
I. II. A R G Y R I S , S. KELSEY a n d H . KAMEL
possibly useful for small problems, is not really suitable for large scale
structures. Here it proves more appropriate to use the automatic
iteration programme detailed below in conjunction with a basic flow
diagram. Essentially this technique amounts to computing the £ values
from the T A column matrix of the preceding iteration cycle. The con-
vergence of the procedure is satisfactory and does not require any
manual checks.
More subtle is the question of members which after complete or
partial plastic straining are subjected to unloading. For such cases it
is necessary to extend the above theory. In what follows we denote
the stations where unloading enters directly or indirectly by u. Since
any unloading must take place along a line AB as in Fig. C.2 the station
in question remains under a constant permanent strain »rfA„, where
fAl, represents the plastic stress of the preceding loading increment
which was still giving rise to a permanent straining at this station. In
the course of loading the structure a number of members initially
subject to plastic deformations may experience at different loading
increments a partial or complete unloading. At the same time as the
plastic state spreads in the structure some elements may pass from an
unloading to a new loading state. However, as long as the corre-
sponding TAU remains below the original fAu the member or station
must still be taken to experience the fixed straining 7rfAl(. Once T A „
exceeds fA„ the station has returned to the normal plastic stress-strain
behaviour and must be reclassified under the /-group. For the under-
standing of the theory and the programme we have next to discuss a
number of particular cases. Thus, if in a flange one of the stations is
subject to unloading whilst the other is still purely elastic—having
never reached the plastic condition—the stresses determining the
generalized initial strains are then at the two end stations, the (associated)
fA„ and T A „ = fA„ = 0 respectively. The important point to note is
that the vanishing TAU of the elastic station must be entered in the T A „
vector. As mentioned in the corresponding case of the partially plastic
member under an increasing stress, these zero elements in the stress
vector could strictly be omitted but this is not to be recommended for
programming purposes. In some instances it may occur that a com-
pletely plastic flange experiences at the two stations a loading and an
unloading respectively. Here it is mandatory to identify the two
stations by a T A , in the /-group and a fA„ in the M-group.
Let us now develop the generalized elasto-plastic theory which
includes also unloading of plastic members. The total number of
stations entering in the loading and unloading states is given by
p= l+ u (C.35)
and the corresponding plastic stress vector is
Ta = {rA, fAt(} (C.36)
90
We express next the generalized initial strain matrix as

H - fO„ H,} = {O. H, HJ (C26a)
where
H, |H, Hj A„T„
I A, A„1{TA, TA„| (C37)
The matrices A,, A„ are no longer diagonal when the two (or more)
stations of one member are registered in the / and u group respectively.
However, as referred to briefly in the subsequent paragraph our
automatic programme uses the theory in a slightly different form,
which retains the diagonal set-up of A p .
Following an evident generalization of Eq. (C.27) the stresses r in the
structure may now be written as
T i - P , 1 p. J ,A,T, = T i , - B i pfn
T,, - B . , T A , - B . „ T A „ (C.38)
11 ere
B.„ |B ( B.J = [P/p.jA, P^P..A.] (C.39)
and (3 ,, (3 „ are the submatriccs of
P IP.* P.J - [P., P.. P.J (C40)

Selecting now the / stresses x, we obtain
T
t = ' i j B„TA, B,„T A „ (C.41)
where B„ is given by Eq. (C.30) and B,„ expresses

B,„ ^ P „ , ' p „ A , (C30a)
Proceeding as in the case of Eq. (C.29) we find
*A. = n..l>/.A, - B , „ T A J (C.42)
where the specification of T / A , and n,, is as in Eq. (C.26) and (C.32).
Substituting Eq. (C.42) into Eq. (C.38) we have the final formula for
the stresses
T = Tr B
'n„ n„B,„ (C.43)
o I
which completes the formal solution of the problem.
Before outlining the programme we deduce also the deformations r
of the structure under the given loading. Let us denote with h a set
of stresses statically equivalent to unit loads applied at the points and
directions where we wish to determine the deflexions. Applying the
unit load theorem in the form applicable both to linear and non-linear
svstcms wc have
r = b ' | f S + H] - o'fS | fc;,H„ (C.44)
91
f is the flexibility matrix of the elements, S the "stresses" due to the

external loading and H„ the generalized strains of Eq. (26a) arising
from the plastic deformations: t p is the submatrix o f t corresponding
to H„. We easily confirm from our previous analysis that
H
n,, n„B (1 ;
(C.45)
o i, .
substitution of which into Eq. (C.44) yields the desired deformations.
Note that the first term in Eq. (C.44) may also be written as
t ' f S = fc'H, = fc'A/c (C.46)
where A., is effectively given by expressions of the type of Eqs. ( C 1 3 ) ,
(C.l5) and (C.l7) and extends over the complete structure.
If the deformations are measured at the points and directions ofthe
loads R giving rise to x or S, we may put Eq. (C.44) in the form
r - FR + b;H„ (C44a)
Here F is the elastic flexibility matrix of the system. The effect of
known imposed initial, e.g. thermal, strains H„ may easily be accounted
for by introducing in the bracket of Eq. (C.44) the additional column
H0.
C.e. Main Flow Diagram and Example
For the theory developed in the preceding section we have set up an
automatic programme suitable for large-scale systems. We cannot
describe it in this brief summary in detail and present only in Fig. C.S
the main flow diagram. It will be noted that the arrangement of the
computations is slightly different from that specified under C d . This
is simply due to considerations of reducing the central processing time.
The diagram should be self-explanatory but for notational details.
The operation starts with the input of the elastic stresses T K due to a
given set of loads or thermal strains whose relative magnitude is fixed.
The coefficient p E determines then the absolute magnitude of the
loading and corresponding stresses for which yielding is just setting in.
We may next start a sequence of rising multiples p of the loads for
which plastic straining will increasingly spread in the system. The
increments />A are thereby fixed in advance but may also be altered at
will during the course ofthe operations. To each loading case specified
by a p there corresponds the main loop shown with a thick line. The
problem of determining the associated stresses is solved by the iteration
process of the inner loop which follows the dotted line. The stress T_
is there the result of the preceding iteration cycle. It is evident that
X ; / represents the incremental redundancies arising from the plastic
(generalized) strains H,,. Having completed the iteration for a load
factor p we proceed to the next one (p + pf) which is solved once
more within the inner loop.
92
Compai* "t ,X
r-
and tubtlitu.* X ,
' ,o, x ,
II
P ap.P 4
• •
1
pa II.u)
t , = PX£
p . .1
1i K--
X -
b
X
;:1
*'
x =x c T
' " ta.
t.t, i •
I. J = b D 'b '
'\Q H II IP
i
A
P • TA. A
.J!!,<t-i1
i
B,P = p ; p t „ A p
TT, • [I * B J
TT, -TTn B„
H, = A ,
0 L
D.H = b,; H .
X„ i - O j D„
x =x, •P,'b,xK
I • "' \ ,n Pnnt out /"

load ^ S T O P )
---CT -x. = o ?)-»- x -*-Q Limit
Fig. 0 . 5 . Alain flow diagram for elasto-plastic analysis of systems.
The following programming points, not illustrated in the flow

diagram, on the formation of the matrices b,„ and T,, are of interest.
These matrices are, in fact, tabulated not in the formal sequence shown
in the corresponding block but in the same order as they appear in the
matrices b , or T. The same applies to the arrangement of the matrix
H„. To obtain hence the tabulation as shown in the flow diagram we
use sparsely populated location matrices, analogous to the matrix a,
93
J. 11. A R G Y R I S , S. KELSEY a n d H. KAMEL
the only non-linear elements of which are ( + 1 ) which are handled

as in Section B. Thus if asterisks denote for the moment the arrange-
ment shown in the figure and used in the theory, we have
bu K. (C.47)
N„
where N, and N„ arc the aforementioned location matrices and are of

dimensions (/ X p) and [u -\- p) respectively. In fact, we do not
determine bjf^ as such, but find directly
N/bI,DII1b|, N| = B„|
[
N^î'bJX = BJ ^°'
The matrix A p remains thereby under all conditions diagonal having
always the two stations of a flange at adjacent locations and displaying
the same sequence as x.
On the other hand T* is found from the first of Eqs. (C.47). We then
compute
°»B«"]T« n*^{r.* n*} (C.49)

o •V, J
and hence
Nfof + Niitf = ij (C.50)
The generalized strains H„ are then given by
H , = A„r) (C.51)
An example of the elasto-plastic analysis proposed here is illustrated

in Figs. C.6, C.7 and C.S. Figure C.6 shows in particular the con-
struction of a two-cell cantilever tube subjected to tip and half-span
loads; all geometrical and material data are listed in a table. Figures
C.7 and C.S reproduce the direct stresses in the flanges and the shear
stresses in the three webs. The coefficient p = 1 signifies the extreme
loading for which Hookean elasticity still holds. The higher loads
under which plasticity sets in and gradually spreads arc thus repre-
sented by p > 1, as also in the flow diagram of Fig. C.S. T h e diagrams
are self-explanatory and give a clear indication of the power of the
method. The reader should note that partial plasticity of the flanges
holds even for the highest loading case considered here for which
p = 1*20. This proves that it would have been erroneous to base the
elasto-plastic analysis on a mean flange load in each direct stress-
carrying member. The convergence of the iteration procedure is
satisfactory and we intend to discuss our programming technique in
greater detail in a subsequent publication.
94
STATION 1 2 3 4 5 6 7
front 0.85 0.85 0.80 0 70 0.60 0.50

e E
middle 1.50 1.10 0 90 0 90 0 90 0 90
°. *
-r m rear 0 95 0.95 090 0.80 0.70 0.60
front 06 0.6 06 06 06 0.6 0.35

rear 0.6 06 06 06 06 0.6 0.35
front 0.048 0.048 0.048 0 036 0036 0.036
middle 0.08 006 0.064 0.064 0 036 0.036

a —
i rear 0.048 0048 0.048 0.036 0 036 0036
front 0.048 0.048 0.048 0.048 0.048 0 048

a c
> —
o —
u rear 0.048 0.048 0.048 0.048 0 . 0 4 6 , 0.048
front 0.08 0.036 0.036 0036 0.036 0 036 0036

n 1
or *- rear 0 08 0.036 0 036 0.036 0.036 0 036 0.036
E = 10 x 10s lb/in* G = 3 .85 x 106 l b / i n '

E/E, = G/G, =30 (D = 29
0y = 48 000 lb/in* o y , _ = 36 000 l b / i n 2
Fig. C.G. Example of elasto-plastic analysis of structures.
95
p • 1 DO
p . 1 »
p • 1 15
p • 1 10
P • 1.00
. plaalic mambara >

O

w
r
p • 1.20
Fig. C.7. Direct stresses inflangesfor elasto-plastic regime p = 1 signifies highest loading for elastic slate.
•9/D/S njsvp jofSuwooj issijSiu sufiuSis j = d mtfyj }i)sv]if-oist>ji w f s q m ut sissnjs Jivsifg -g-;3 - 3 I J
-
<
y.
<
aJ
<
tt
S 01
-
-
03
t-.
r
- OC
0
CA
a ( 0l «9E~=''o
z ,vJ|/qi ,01 » 07 •
x
r-
M
J. H. ARGYRIS, S. K E L S E Y a n d H. KAMEI,
Cf. A Precis on Large-Displacement Analysis

The matrix force method may be generalized by a number of
schemes to account for the effect of large displacements and the associ-
ated phenomena of buckling and post-buckling. We cannot give here
even the briefest of apercus of the various digital devices we can use and
will restrict ourselves to a single technique to which we referred already
in our Aachen contribution 2 . This method, although successful on a
number of examples, is not necessarily the best under all conditions
and other approaches may be more appropriate in the presence of
truly large displacements and such intricate instability manifestations
as oil canning.
T o understand the underlying idea, consider the simple statically
determinate pin-jointed system under a central load R shown in Fig.
C.9, for which the load S and the central deflexion r are to be deter-
mined. The deformability ofthe structure—an important factor when
Fig. C.9. Simple example for large displacement analysis.
Fig, C.10. Effect of large displacements.
w n . s (i-^.i
r\ s sin •
n
1, = " •,
»„' •
Fig. C.l 1. Equilibrium conditions; simulation of large displacements with "Ersatz"
loads R,. S L and S,, indicate the loads in members of undeformed and
deformed structure respectively.
the elevation ratio hja is small—is to be allowed for. For brevity of

presentation we initially ignore buckling of the individual members
which may lead to snapping or oil canning of the structure.
The standard analysis which neglects the deformability of the
system when writing the equilibrium conditions and assumes it as
infinitely small when deriving the deformations themselves, yields in
the present case (see Fig. C 1 1 )
S = b R , v = fS, r --- F R (C.52)
98
where, retaining the matrix notation, we have
b = = £= F ( Q 5 3 )
~2lii^ ~2? EA' = 2t,*EA
Consider next the influence of the deformations, the quantities de-
pending on them being specified where necessary by a suffix " d " .
Figure C 1 1 gives
sin .£„ ,,„ - [h - r ) l l „ (C.54)
The length /,, of the deformed bars derives from,
ld= l\\ - , / V ( 2 - p }]. (C.55)
where
P = r/h (C.56)
is the non-dimensional measure of displacement adopted here. Equa-
tion (C.54) may be put in the form
(C 54a)
*="[i !M^W -
Following the first of Eqs. (C.52) the actual force in the members of the
deformed system may be expressed as
S b,R ' R (C.57)

*-rld
The coefficient b,, is clearly a non-linear function of r. Now our

aim is to retain as far as possible the formulation ofthe small deflexion
theory and to use, in particular, the basic static and flexibility data b
and F of the linear analysis also in our present more refined investiga-
tion. As shown below this may be achieved by the introduction of
imaginary or Ersatz (additional) loads R ; and initial strains H,. In
fact, for the statically determinate example of Fig. C.9, R, may imme-
diately be written down in terms of the actual external load R .
However, for our subsequent generalization to redundant systems it
proves necessary to express R, in terms of the generalized stresses S.
Referring to Fig. C l l or directly from Eq. (C.57) wc obtain
S = b ( R + R,) (C.58)
where
R,= -2{ v -t},)8=-?-l ''"R (C.59)
Writing* Vn
wc have a,'= - 2 ( 7 ? - ^ ) (C60)
R, = alS (C59a)
• We already use for a f which is here a scalar, the matrix notation required for our
general theory.
99
Equations (C59a) and (C.60) show that the imaginary or Ersatz loads
are directly proportional to the stresses S but non-linear function of the
displacements. In fact, application of Eq. (C54a) yields in the present
case
1
2 (C.61)
"I'-[T V*p{2 - P } ] >\
Expansion of the root gives in turn
a*^-2w{«* + ^-2(3-,,)) (C61a)
^ —2prj(*? (C61b)
where
a = cos <ft = ajl
In the approximate expression (C.6lb) we only retain the linear term
in p ofthe expansion. Wc observe that the second term in Eq. (C.61 a)
carries a factor tf and is hence negligible for the configurations with
small i/ (say 0*1) which interest us here.
Fig. C. 12. Derivation of deflexions: simulation of large displacements with

" E r s a t z " initial strains H,.
We now turn our attention to the condition of compatibility of the

system. Following Fig. C.12 we may express the finite deflexion r in
the form
r = - i[v4-H(] (C.62)
V
Here H , is evidently an imaginary or Ersatz initial strain, the intro-
duction of which permits us to use for the r the formula (C.62), which
is effectively the same as in the small deflexion theory.
/
fS b R (C.63)
FA. I
100
We also derive from Fig. C.l2,

H:= -ft,-/(I -ptf)] (C.64)
We observe that while aj contains a linear term in r, the lowest order

power of r in H , is quadratic. This is a characteristic property of our
type of problems and will be used later on. We note en passant that
Eq. (C.62) is but a particular case of the small deflexion matrix equation
r = b'[v + H | (C.65)
derived in Ref. 1 for systems under loads and initial strains H . Sub-
stituting into Eq. (C.65) the elongation given in Eq. (C.63) we may
write the result in the form
r = F[R + R J + b'H, (C.66)
Having established the two equations in the unknowns S and r, we
demonstrate their solution in what follows. Using Eqs. (C59a) and
(C.61) in Eq. (C.58) wc obtain
S = |I baj] " b R !B,bR b„R (C.67)
The exact and approximate expressions for b,, are
bd = S b ( }
' "2rT I-P
= - ^ [- " P {«2 + P .f (3 - p))]" 1 (C68a)
[1 pa2]_1 (C 68b)
- ~ h, ~ -
where Eqs. (C.68), ( C 6 8 a ) , (C68b) represent the exact and the
approximate expressions, the last of which contains only the linear
term in r. Introducing the strains
es = A and BB = _ (C.69)
Eq. (C.67) may be written as

e s = b d E l{ (C.70)
Considering next Eq. (C.62) we reduce it, using Eqs. (C.63), (C.69)
and (C.67), to the formula
_if
101
We do not reproduce here the exact equation for p which can easily
be written down in conjunction with Eq. (C.55). For the purpose ofthe
present exercise Eqs. (C.70) and (C.71) are best solved by determining
the loading strain En and member strain E S for selected values of (rjh).
This is shown for the specific value of r\ — 0.1 in Fig. C . l 3 . It is evident
Fig. C. 13. Numerical example of large displacement analysis.
that the structure is, in general, not in stable equilibrium over the
whole range of compressive e.s strains. In fact, if the buckling strain
Eb of the member is smaller than the maximum compressive strain e g ,
the system will snap into a new and stable equilibrium position in
which the members experience tension. This is indicated in Fig. C.13
by the thick vertical lines. An important point for our subsequent
developments is that within the wide range of p values from zero to
2.5, the differences between the exact and approximate formulae of
Eq. ( C 6 8 b ) are negligibly small and cannot be reproduced in Fig. C.13.
On the other hand, if px2 < 1 we may approximate IB, of Eq. (C.68)
further to the linear expression
»,.= ! + p«2 (C68c)

However, in this case the accuracy of the results becomes acceptable
only for the lower values of pec2, say ^ 0-25. Anyhow, the above
developments show that for many purposes it will suffice to assume the
102
Ersatz loads as proportional to r. In the present case this amounts to

putting [see Eq. (C61b)]
R * 2/, V /-S - poem (C.72)
Following the above discussion of the simple example, it is fairly
evident that our conceptual model may also be applied to arbitrary
redundant structures. Naturally, il proves necessary there to allow
for the generation of internal stresses S due to the Ersatz initial strains
H,. If wc assume, moreover, that the system experiences in addition
to the external loads R also a given set of initial strains H due to say
thermal causes or lack of fit, Eqs. (C.58) and (C.66) take their most
general form,
S = b ( R + R,.) - p ( H + H,) (C.73)
r = F ( R + R,) + b ' ( H + H.) (C.74)
where as in our elementary example b , (5, F arc obtained by the
standard matrix analysis of linear systems: p is incidentally given by
Eq. (C.3). The non-linearity enters as stated previously through the
dependence of R, and H , on the displacements. Al the same time the
linear relationship of R, and S given in Eq. (C.59) is in all cases true.
It is not possible to consider in this expose the matrix formulation of
R, and H, for typical configurations.
The two non-linear matrix Eqs. (C.73) and (C.74) can, in general,
only be solved by iteration. We cannot attempt here to discuss the many
possible iteration techniques. For the purpose of illustration, it suffices
if we outline briefly a method we have found successful. It also leads
naturally to the matrix equation determining the classical or small
deflexion instability phenomena.
We start by substituting formula (C59a)—which must be understood
now as a matrix equation—into Eq. (C.73). Solving then for S wc
obtain formally
S = [I ba;.]-* [bR - p ( H + H,)] (C.75)
and hence
R, aj[l - ba;.]-* | b R - (3(H + H,)] (C.76)
Introducing Eq. (C.76) into (C.74) yields a highly non-linear matrix
equation in the unknown vector r which we may solve, for example,
by the Newton-Raphson technique. For this purpose it is convenient
to expand the inverted matrix and to approximate it by a finite series
[I ba;] * = I + (ba|) -\ (ba,') 2 + . . . (C.76a)
Any approximation in Eq. (C76a) should logically entail a correspond-
ing simplification in the expression for a,'. As an extreme measure of
approximation, let us assume that
[I - b a j ] - 1 w I (C76b)
103
This effectively means that the imaginary loads are deduced from the
internal stresses S /y given by the linear theory*, i.e.
R, =•= a J 8 z (C.77)
As a consequence it is logical to reduce a' to its simplest form whereby
it is merely proportional to r . The degree of approximation is then
equivalent to that of Eq. (C68c) and should be sufficient as long as the
deformations arc not too large.
Considering next only the effect of the given loads R and initial
strains H wc may reformulate the expression
aJbR a|(JH to read
- k,„r - k . „ r (C.78)
where k l f l and k, 7 / arc Ersatz stiffnesses generated by the linear
stresses due to the external loads and initial strains respectively, when
the nodal points are subjected to unit displacements. For convenience
we also introduce the expression
- a;p = b,„ (C.79)
which defines the Ersatz loads arising from unit initial strains.
Taking into account now Eqs. (C.77), (C.78) and (C.79), Eq. (C.74)
is transformed into
[I + F(k, 7 , + k,.„)]r = F R + b ' H + (Fb.„ + b ' ) H , (C.80)
where the non-linear dependence on r is retained in the third term
on the right-hand side. Equation (C.80) is relatively simple to solve by
iteration. We have applied Eq. (C.80) to the large torsional and bend-
ing deformations of double wedge wings for which analytical results had
been derived in Ref. 14.
When the deflexions are truly small, H „ being of the second order
in r, can be neglected. Equation (C.70) can then be solved directly for
r . It differs from the usual expression for the deflexion by allowing for
the modification of the equilibrium equations due to the presence of
the deformations. It may hence be used for setting up the instability
criteria of the standard buckling phenomena associated with small
displacements. For such loading cases Eq. (C.80) is reduced to
|I +- F(k.. w + k..„)]r = 0 (C.81)
Introducing X and 0 as measures ofthe intensity of loading and heating
we may write
hiR = A,w, k , „ = 6k.,, (C82)
[ 1 + (AFfc,7H G F k , „ ) I i — 0 (C81a)
* T h e influence of H , m u s t , of course, IK- i n c l u d e d .
104
which yields via the standard techniques for eigenvalue problems the
critical combinations of ?. and 0 as well as the modes of deformations r
occurring at buckling.
1). MATRIX DISPLACEMENT METHOD IN NON-LINEAR

PROBLEMS
D.a. Introduction and General Principles

We present in this last Section an extension of our matrix displace-
ment theory, which takes account of some of the non-linear effects due
to large displacements. As in the original theory of Ref. 1 the influence
of non-uniform heating or similar stress sources arising from give of
supports or lack of fit is included ab initio. In what follows we classify
all these phenomena under the (generalized) title "heating applica-
tions". O u r reader will no doubt remember that wc introduced in the
dual matrix analysis "initial strains" in the force method and "initial
stresses" in the displacement method respectively to determine the
consequences of these heating applications. Just as initial strains H are
assumed to be generated in the force technique on the free unassembled
elements, so the initial stresses J arc generated in the displacement
method on the elements with all nodal points blocked or fixed. Refer-
ring to Refs. 1 and 12 the following formulae apply for the generalized
stresses and loads:
S kar | J (D.l)
and
R = K r | a'J (D.2)
where a is the same matrix as is used for the assembly of the K in B.C.
Naturally this presumes that the initial stresses J are measured in the
same fixed coordinate system as R and r . These devices of initial
strains and stresses have proved most fruitful both from the theoretical
and practical points of view. Proofs of general theorems and verifica-
tions of special techniques, which can be derived by the more traditional
argumentations only through tortuous arguments, follow most naturally
with the help of these simple, but at the same time subtle, concepts.
Returning to our present task, we extend here the ideas on large
displacement analysis wc first proposed in Ref. 2 and in a subsequent
note to analyse the influence of change of geometry occurring in large
displacements. Although the method is in certain aspects similar to
that put forward in Ref. 10 it will be found to be more general and
at the same time more concise and symmetrical in its formulation. Its
matrix language should be ideal for programming purposes.
As given here the theory presumes elastic behaviour ofthe material.
It could be generalized, however, as in Cf, to allow for clasto-plastic
deformations or more complex rheological properties of the material.
1(15
The analysis uses a step by step procedure, each increment of which

is linearized. It is assumed that at any onc instant the incremental
strains may be taken to be so small that the standard formulae for the
strains and stresses in membrane elements subject to load and thermal
effects are valid. Just as in Section C f our large displacement approach
includes the effect of the change of geometry on the equilibrium
conditions and yields in the limit, for infinitesimal displacements, the
initial buckling criteria.
T h e basic idea is very simple. The external loads and thermal
actions are represented by a series of conjoint or separate small incre-
ments leading from the initial, unstressed state to the final one where
the full complement of external loads and thermal actions is present.
for each loading and/or thermal increment we introduce, using the
beforementioncd linearization, the effect of the change of geometry
on the vectors of the nodal loads which have built up in the course of
the preceding incremental steps. In some cases, it may be appropriate
or may suffice if the thermal stresses are initially obtained through a
linearized (or non-linearized, if necessary) analysis and then assumed
as invariant throughout the sequence of external load applications. It
is evident that a divergence of the supposedly small-displacement
increments will occur when a secondary instability phenomenon like
snapping is approached, as may occur in thin wings under heating
actions 14 .
Consider an intermediate equilibrium condition of a system after a
number, say m of incremental steps of external loads and/or generalized
heating actions, the corresponding external load and thermal gradients
being R,„ and 0,„. Let us denote this state by '0'. Figure D.l shows a
particular element of the structure in equilibrium under the action of
the nodal loads or generalized stresses P 0 which have been set up in the
course ofthe previous loading history. [We find it here more convenient
to use the notation P 0 instead ofthe S of Eq. (D.l).] If we now apply
an increment R A and/or 0 A of external loads and/or non-uniform
thermal gradients the structure moves to a new equilibrium state which
we denote by (<-/). In the course of this displacement the element may-
experience a rigid body rotation and/or deformation in its own plane.
If for the moment we forget the elastic nodal loads, we observe that
the forces P 0 will, in general, be subject to a change in their components
both with respect to the basic coordinate system Oxyz of the structure
and the local system O'x'y'z' which is taken to remain fixed as the
element moves from (0) to (</). It is at the same time helpful to intro-
duce a local moving system 0|r,-£ which is initially coincident with
O'x'y'z' in state (0). It is evident that the components of P 0 with
respect to 0|^^ may only be modified if the element experiences
deformation. However, whatever the nature of the displacements may-
be, it is clear that certain measures of the vector P 0 have to remain
106
Fig. D . l . Incremental linearization of large displacement problem.

Geometrical and elastic incremental Iniuls.
«.™-") . JoûTj.y^z,)
Fig. D.2. Flange in fixed coordinate system. Derivation of geometrical

stiffness matrix.
107
Fig. D.3. Flange in local coordinate system. Derivation of geometrical
stiffness matrix.
Fig. O.4. Triangular element in fixed and local coordinate system.

Derivation of geometrical stiffness using three fictitious flanges
(1,2), {2,3), C U ) .
invariant. These are clearly selected resultants or the specified internal

stresses o. For example, for the flange of Fig. D.2 the constant axial
load of magnitude P 0 will not vary as it moves from the position (1,2)
to another one (I,II). At the same time the components of P 0 with
respect to Oxyz or O'x'y'z' will change unless the flange does not
experience a change of direction as it displaces from (0) to (</). Inter-
estingly enough a pure axial elongation /A of the flange will also not
give rise to incremental components of P 0 . If we consider a more
elaborate clement, like that in Fig. D . l , the conditions of invariancc
are evidently not as simple unless we select for this purpose the internal
stresses, which is not very convenient. We note en passant that the
components of P 0 cannot experience any change if the element is subject
108
to a pure translation, which may or may not be accompanied by a

similar contraction or enlargement.
It follows from the above reasoning that in order to satisfy the
equilibrium conditions, with respect to Oxyz or O'x'y'z', on the
displaced and possibly deformed clement, we must introduce an
incremental component loading P A r ; which does not affect the invariant
measures of the vector P 0 . To determine PA,r. we may use cither
kinematic or static considerations which are, of course, dual in their
formulation. Nevertheless the statical argument appears in general
the simpler, especially when followed by the aforementioned lineariza-
tion in the incremental displacements r A . Starting with the state (0),
we may express the equilibrium conditions with respect to the fixed
local system O'x'y'z' or the basic axis system Oxyz of the vector P 0
by a matrix equation of the type
»*P„ O (D.3)
where 33* is a matrix in general of order (6 X 3n), where n is the
number of nodal points. We may now expand 93* to a square matrix
of order (3B X 3B) by introducing the necessary number, (3B — 6), of
conditions of invariance defined for the load vector P 0 . Equation
(D.3) becomes then
-p. - [°:] D.l
where C* is a vector with (3B — 6) elements which are the invariant

measures. 93 has, of course, to be a non-singular matrix, which is
anyhow true if we choose independent equilibrium relations and the
appropriate invariance conditions. We may solve Eq. (D.4) for P„
to obtain
P 0 = 93 *C (D.4a)
which illustrates the significance of the vector C. Equation (D.4a)
may be expressed in the form
P 0 = 2C,M. (D.4b)
i
where the C, are the elements of the vector C and 31, column matrices
of order (3B X 1). As a matter of fact Eq. (D.4b) can often be written
down by inspection.
Considering now the displaced state {d), Eq. (D.4) will be trans-
formed into
[ » + » J | P o H PAI.-I = C (D.5)
where 93A, an increment of the matrix 93, is a function of the displace-
ment vector p A and may to the first approximation be linearized in
p A . Carrying out the multiplication in Eq. (D.5), neglecting the higher
111')
J. H. ARGYRIS, S. K E L S E Y a n d H. KAMEL
order matrix in the displacements, i.e. 9JAPA,,- and allowing for tin-
original equilibrium condition (D.4), we find the simple relation
93PA„. + 93AP0 = O (D.6)
which yields
PA„ - -93-*93 A P 0 = 93 '93A95 'C (D.6a)
It is interesting that Eq. (D.6a) may alternatively be derived by-
considering the differential of expression (D.4a). Thus
d(P 0 ) = //(93--C) = -93-^(93)93-'C (D.6b)
Equation (D.6a) may be established either in the Oxyz or the O'x'y'z'
system. For the further developments it is necessary to transform Eq.
(D.6a) and to express it not as a function of P 0 , but of the vector of
nodal displacements p A or p' A . More simply, wc may obtain the
increment P A , ; directly from Eq. (D.4b) in the form
PA« = ZC,** (D.6c)

I
It proves more straightforward to express each column as
«.A O,PA (D-6d)

where o, is easily seen to be a symmetrical square matrix of order
(3B x 3 B ) . It follows that Eq. (D.6c) is transformed into
PA/; = [IC.a,.]p A (D.7)
/
It is apparent that the expression in the square brackets of Eq. (D.7)
may effectively be considered as a kind of stiffness matrix. It is a
function ofthe geometry ofthe element in the state (0) and the invariant
measure C* of the load system P 0 . We denote it in what follows as
the instantaneous geometrical stiffness matrix k,,. of the element.
Equation (D.7) may hence be written
PAG = k,.-PA (D.7a)

w lien-
kf;-=2C,a,. (D.7b)
i
Let us now extend our argument and include the effect of elasticity
o f t h e element. As indicated in Fig. D.l the deformed and displaced
element will be in the state [d) in equilibrium under the action of the
complete nodal loads
P / = Po + P A / ; + P A , (D.8)
where P A/ , ; is the matrix of incremental loads due to the clastic deform-
ations of the element. Wc now assume that the clastic behaviour of
the element may be described with sufficient accuracy in the incre-
110
mental stage (0) to {d) by a constant stiffness k^,. which is evaluated on

the basis of the geometric and material data at (0). Actually, the
geometry of a single element will hardly vary over a number of steps
and it will often be sufficient to use the same dimensions in the local
fixed coordinate system O'x'y'z' over a wide range, allowing only,
where necessary, for the change of the clastic moduli with temperature.
It is apparent that the assumption of a constant stiffness within a step
ignores the second order strain effects as the element moves from id
to (//). This should be permissible since these terms cannot be but
negligible when the increment p A is sufficiently small. The total
incremental nodal loads on the element may now be written as
PA - PAG- + P A * - [k« + k«]p A = kp A (D.9)

which is formally the same relation as in the standard linear matrix
displacement theory. The difference is solely due to the incremental
loads P A ,j or the associated fictitious stiffness K f ; introduced through
the change of geometry.
It follows that the analysis of the incremental step may be carried out
in principle in the same way as in the small displacement theory.
Thus, the stiffness matrix of the complete structure is obtained from
K = K „ + K„ ; = a ' k a (D.10)
Since the equilibrium ofthe structure is satisfied in the state (0) under
the action of the external load vector R,„ and the nodal load vector
P 0 we need only consider the equilibrium condition under the incre-
mental loads, for which we have
RA Kr A , (D.l 1)
which reduces the solution of the incremental problem to that of the
linear regime in Section B.
Equation (D. 11) presumes that no incremental thermal loading
arises between (0) and (//). Should, on the other hand, an incremental
initial strain Y)A be present, a slight modification of our solution becomes
necessary. This may be done most simply and consistently by applica-
tion of Eq. (D.2). Thus, introducing the initial stress vector J A
corresponding to y)A, Eq. (D.2) reads in the present case
RA K r A + a'J A (D.l la)

Having completed the analysis of the mth step, we proceed to the
[m -f- l)th. For this purpose we set up the new invariant vector C*,
which is derived by including in the vector C* of the mth step the
corresponding elastic and thermal contributions of the mth step. It is
then possible to determine the new K,,. and to set up and solve the new
Eq. (D.l l a ) .
Ill
D.b. Determination qf Geometrical Stiffnesses for Typical Structural Elements

We have derived the geometrical stiffnesses for a number of elements
including flanges under constant load, triangular elements under
membrane and/or bending action. We restrict here our account to
flanges and triangular elements under membrane stresses. As a matter
of fact we need only obtain the geometrical stiffness for the flanges
which suffices to establish the corresponding stiffness for the triangular
elements.
Consider the flange, under an axial load P 0 and length / shown in
Fig. D.2. Its position with respect to the basic coordinate system
Oxyz of the complete structure is fixed by the column matrices 4-i a n d
"4»2 (see Appendix I ) . Its direction cosines may be expressed as the
row matrix [sec Eq. (1.7)]
c = [1 m n] [«W - 4»a - } [*.- A. Zn] (D-l2)
we find it convenient to write initially the vectors P 0 and pA in the

sequence
l"lx Pin "lz "ix "iv "izl (D.13)
PN [U1A "1A Wr II.. «-«A} (D.H)
It is evident that the invariant measure of the force is in the present

case the axial load P 0 itself. On the other hand the components \? lT
to P 2 j of the load at the points 1 and 2 may change as the flange
moves from the original position (1,2) to the incremental one (I,II).
The equilibrium equations for the six components at (1) and (2) are
trivially simple and may be immediately written down in matrix
form. At the same time one of the moment equilibrium relations is
linearly dependent on the other two and cannot be entered in our
equations. Instead we introduce the invariance condition in the form
P*/l = Po (D.15)
Dropping the equilibrium condition for M z , the matrix 93, here of

dimensions (6 x 6), may be expressed as
I.
1/1 0 0
« (D.16)
o. n 0 -1
0 —n m
to which there corresponds the column matrix

C = {0 0 0 P0 0 0} (D.17)
112
Either by solution of Eq. (D.4) or directly by inspection wc obtain
P„ - P«c,! (D.18)
where
t,. = [ - c e] (D.19)
which is but a particular case of Eq. (D.4b). We immediately derive
from Eq. (D.18) the incremental vector P A in the form of Eq. (D.6c)
P A == POÂ (D.20)
A simple argument yields the following concise matrix expression
for the increments ofthe direction cosines:
I'A -"A nA}
= '.' [*s - «'c]{«ii »« «-2i} (D.21)
Equation (D.21) confirms that for pure translational movements
'21 •'21
(D.22)
1 111 n
which may include an elongation /A of the flange, no change in the
direction cosines takes place and the vector P A vanishes. Following
Eq. (D.21) we may write the column matrix c,'A as,
*t< * -—{ - « i <A, opA (D.23)

\\ here
ejfj (D.24)
/
It follows from Eqs. (D.7b) and (D.20) that the geometrical stiffness
matrix of the flange is given by the concise expression,
I,
k, ; = P0a (D.25)
/
Naturally we may obtain the explicit expressions for the coefficients of
the (6 X 6) stiffness matrix k,, by carrying out the term by term
multiplication of t',,tr. However, this is inadvisable for computational
purposes on the digital machine.
It is interesting to consider at this stage the corresponding clastic
stiffness matrix kA; of the flange. Following Rcf. 1 and Eq. (IV.12) of
Appendix (IV) this may be written in the form
-Ma* (D.26)
/ '
113
J . H . A R G Y R I S , S. KEI.SEY a n d H. KAMEL
where B is the area of the flange and EH the Young's modulus corre-
sponding to the current temperature 6. Comparison of Eqs. (D.25)
and (D.26) yields the interesting relation
I,
k (D.27)
°~ / BE,}"
which may also be confirmed by direct physical argument. We can

combine k, ; and h.K into the following single formula for k,
i:, -VI BE,.

Vc (D.28)
/ -I3 ij + I
For practical operations wc require, in general, the matrices k and not

k, for which the corresponding load and displacement vectors are
"0 ==
\"li "ix "\v "iy "\z "izi (D.29)
PA (a 1 \ <i\ <•> ,
w.. (D.30)
Using the notation introduced in Appendix I [Eqs. (1.20), (1.21)], we

have
PA = Tp A TP, (D.31)
where
En E 21 E3,
(D.32)
_E12 E22 E32_
The (3 x 2) location sub-matrices E l 7 are similar to those defined on

page 53 and include merely "ones" for non-zero elements. The
stiffness matrices k,,. and k may now be obtained from the congruent
transformations
k,,. T'k„T]
and (D.33)
k = T'kT
Again it is not recommended to set up the individual elements of the

rearranged stiffness matrices k,,. and k.
As a useful exercise for the reader we next consider the case when
the flange is first referred to the local fixed coordinate system O'x'y'z',
as illustrated in Fig. D.3. Wc obtain cither through an independent
derivation or directly from Eq. (D.25)
K Po ' r- l-uh -lal r

.—•En
E„ ~EU
E„
(D.34)
114
To find now k,,. for the fixed coordinate system Oxyz we have merely
to apply a congruent transformation
k... - b,.k,';b, (D.35)
where
b o:,
(D.36)
_o;! b.
in which b is the (3 X 3) matrix of the direction cosines of the axes
O'x'y'z' with respect to Oxyz. Thus, the first row of b would read
[1 m n|
It is easily confirmed that operation (D.35) leads to the same result
as Eq. (D.25).
We next derive the geometrical stiffness k, ; of an arbitrary triangle
(1,2,3) whose geometry is fixed by the coordinate vectors t]> ofthe three
vertices with respect to the basic coordinate system Oxyz. It may at
first appear rather surprising that this problem may be solved quite
simply by the specification of three fictitious flanges (1,2), (2,3) and
(3,1) along the sides of the triangle. This may be seen immediately if
we transform the nodal loads P 0 that have accumulated in the preceding
m incremental steps into forces P 02 i, POM* a n c * Pom acting at the vertices
1,2,3 along the directions ofthe sides ofthe triangle, as shown in Fig.
D.4. It follows from the equilibrium conditions which must be satisfied
by the original components of P 0 that (see Fig. D.4)
=
Po.i.i+i Po.i+i.i (D.37)
Clearlv the "flange loads" P 02 i, P 032 , P 0 1 3 form the invariant vector
C* o r '
P|)7- = (P021 Po:(2 P|>i;il (D.38)
of the ( 9 x 1 ) load vector
Po = {Pix P|» P u P2J- P21/ Pa* PBJ- Pan P32} (D.39)
to which there corresponds the displacement vector
PA = ("IA vi± «-IA "2A ^2A »>%& «&4 "UA « W (D.39a)
Let us now consider a typical fictitious flange l t , i + y Its geometrical
stiffness k,,., I t l is immediately given by
"o.i+l.i I. 1.1
V/i.I + l tii.ifi *H.i + i (D.40)
/,
H.M L L - »I.
3 I,
where, in accordance with Eq. (D. 19)
trt.i+i = [ - « «].,.+. (D.41)
and
t{.i+i = j - "l»«+i.i = 7 - - [ « m . i yi+lA 2m,,] (D.42)
115
Having established the three stiffnesses k,, 21 , k,,..,2, k,, 13 the stiffness
k r , for the displacement vector (D.39a) can be assembled following
the procedure of Eq. (B.l8) by congruent transformations applying
position matrices M , , , , and a subsequent summation. Thus
k 0 - M 2 1 k,. 2 1 M 2 1 + M£ 2 k,,, 2 M 3 2 -! M' 13 k„ 13 M 13 (D.43)
where the (6 X 9) matrices M , + 1 , are given by
or
MM =
fe o. M«
•o,
o3 o.
I. o3i
I3-
o
M13 =
Eo 3
3
la
o,_
The stiffness matrix k,,. for the displacement vector
(D.44)
V J
PA = {«1A «2A «3A *PA i\ '3A «"1A "-2A *"3A} (D.45)
follows now from,

k, Tk,.T (D.46)
where T is defined in Eq. (1.21).
There only remains to determine the invariant vector P o r in terms
ofPg of Eq. (D.29). For this purpose we introduce the (3 x 9) matrix
«12 o
ti. O o (D.47)
o «31J
where C llM1 are the direction cosines of Eq. (D.42). Then following
the argument leading to Eq. (D.58) on p. 118 we may prove that (see
also Eqs. (D.56), (D.59)),
P„, I *tt 7 DP 0 = I'djDTP,, (D.48)
D.c. Determination oj'Invariant Vector C*; Stresses; Natural Stiffness

T o complete the analysis we still require the current C*—the
invariant vector of P 0 —as a function of the previous loading history
ofthe elements ofthe structure. It is evident that C*, being effectively
measured in the moving coordinate system 0£//£, is obtained by direct
summation of the incremental C* arising from the preceding m linear-
ized steps of the analysis and involving the corresponding elastic and
thermal contributions.
We start by considering the simpler case of a flange under a constant
load P 0 which forms in this case the only element of C*. To determine
the increment P A for a typical incremental step it is effectively only
116
necessary to obtain the elastic strain along the axis of the flange. Now
it is simple enough to see that the total axial strain t A is given by
eA = J c,.TpA (D.49)
where (.. and T arc defined in Eqs. (D.l9) and (D.32).

If we operate with p A instead of p A , the location matrix T has to be
omitted in Eq. (D.49). Denoting the incremental thermal gradient of
the flange by 0A, the corresponding thermal strain is
•/A = 0.0A (D.50)
It follows that the pure elastic strain (fA — »;A) is
J «,TPA - »VA
Since the stiffness of the flange corresponding to the "clastic strain"

is merely K, E„B the incremental load P A is
E
PA= f c,Tp A - EmB.ls (D.51)
Thus, the total axial load P 0 after m steps is
E
Po = S'PA - P 0 K + Po« = S f U>A - S£iAA (D.52)
where ^ is the standard summation or integration sign applied in the

finite difference technique.
We now proceed to the more complex case of a triangular membrane
element. It is relatively easy to see that the pure elastic deformations,
measured along the axes of the fixed system Oxyz, may be expressed as
P*A = PA - »?AT+ (D.53)
where «4> is given by
+ = W»i *. +3} (D.54)
and T is the same as in Eq. (D.46). If pA is used in place of p A , T has
to be suppressed in Eq. (D.53). Any rigid body movements may be
added lo Eq. (D.53) and it may prove more convenient to operate
with the vector |o \\».,, i|i al I instead of {<]», 4-2 4*3}. The incre-
mental vector PA—not the invariant measure P r A — i s hence
PA - krfjw (D.55)
where the (9 X 9) stiffness matrix kA. is obtained in Appendix I as k.
The invariant ( 3 x 1 ) incremental vector P r A may now be determined
from
D'C'7PrA ^ PA- TPA (D.56)
117
J. H. ARGYRIS, S. K E L S E Y a n d H . KAMEL
where
o3 - I 3 (D.57)
I. o 3
is a "difference" matrix and C r is defined in Eq. (D.47). Prcmulti-
plying by tX r D we find,
P r A = I -e7DTPA (D.58)
where
2 -c 12 c!, a c12t311 2 cos/* cos x
-c 2 3 cj 2 2 -c 2 3 c 3 i = cos ji 2 cosy (D.59)
-f3ifi2 ~t3i«23 2 J [cos a cosy 2
and a, ft, y are the angles of the triangle at the vertices 1,2,3 respec-
tively. Equation (D.58) yields in conjunction with Eqs. (D.53) and
(D.55) the incremental invariant load P 7 , A as a function of p A and the
thermal strains r/A. T h e vector p A itself is obtained by the linearized
analysis of the step under consideration, in accordance with D.b. It
follows that the invariant P r required for the (m -f- l ) t h step may be
put in the form
P r = Sk'[PA - »/AT<H = Sk'p A - Sk'. /A Td- (D.60)
where the (3 X 9) matrix k ' is given by
k' = I ' C y D T k , , ; (D.61)
using the relations developed in Appendix I for the local fixed co-
ordinate system O'x'y'z'. T h e stresses in the triangle may be obtained
simply from (see Eq. (1.29)),
a .Sxlopl - / /A c] (D.62)
where
c = {1 1 0} (D.62a)
A more elegant derivation of the increment P r A may be obtained
by the specification of the (3 X 3) stiffness matrix k 7 of the triangular
clement corresponding to triangular forces P T A and the associated
displacement vector p r A . Wc first note that the total relative displace-
ment vector (in the directions of the forces i? T fj is determined from
the pA vector by the transformation
P7A = G7-DpA G 7 .DTp A (D.63)
where (t 7 . and D are given in Eqs. (D.47) and (D.57).
The pure elastic displacement vector p r A / ,; is now
PTSK — PTA — '/A 1 = C / D T p A — J /A 1 (D.64)
where
1 l'l2 '23 '3: (D.65)
118
To set tip k 7 . in a manner natural to the geometry of the triangle, we

introduce a new specification ol the stresses, different from thai used
in Appendix I where it is bound up with the local coordinate system
O'x'y'z'. Thus, we represent the actual stress distribution in the
triangle as the superimposition of three stress components aVic, a i3r ,
a 3lr * parallel to the three sides (1,2), (2,3) and (3,1) respectively.
We write the vector
a a
°7- = {<*tU i3c 3lc} (D.66)
Note that these stresses are not the total stresses in the selected directions
since no two components are, in general, orthogonal to each other.
It is easy to write down, by matrix transformation, the actual stresses
o-12, a i 3 , a 31 or for that matter any other system of direct and shear
stresses, but we omit here the straightforward equations. Wc may now
relate the stress vector directly to the vector P r * by the matrix equation 2
<J/1„ (D.67)
where 12 and t are the area and thickness of the triangle (see Appendix
I) and 1,, is the diagonal matrix
h = T / 1 2 Imm / . . J (D.68)
Consider next the strains
'.,i+i.r
"i.i + l.c and (D.69)
E -..•i-i.i E
parallel and perpendicular respectively to the side i,i + 1, which

correspond to the definition of the stress in Eq. (D.66). Using the
standard strain transformation formulae, we find the following relation
between a r and the elongations p r of the sides of the triangle
- l,..tf T tj T = p T (D.70)
where
sf T = (1 + v ) s / - ve 3 e 3 (D.71)
in which
• 1 cos2 ft cos2 a
cos2 ft 1 cos2 y (D.71a)
2 2
cos a cos y 1
and e 3 is the unit column matrix
e , = {l 1 1} (D.71b)
Connecting equations (D.70) and (D.67) we obtain the force-displace
ment expression
k 7 .p r (D.72)
where
kr Eiltl,, '.V.,,1!,, ' (D.73)
is the required expression for the stiffness matrix of the triangle.
* W<* limit, for the current derivation o f k r . the suffix A.
119
It follows that the invariant vector P 7 . which is operative for the

current displacement increment p A is now obtained from
P 7 . = SkrpTtmM = ..S^PTi - S ' / A - M (D.74)
or P r = ,<jk r tt r DTp A - ^'-/P-M (D.74a)
which is much simpler than Eq. (D.60).
Naturally, the matrix k,.. itself, which enters in each incremental
investigation is best determined from k 7 . and not from the local k^, as
in Appendix I. It may be shown that
k„. = T D ' «!,. k 7 t t 7 D T (D.75)
The reader should note that the order of k r is merely (3 X 3).
This deflation of the stiffness matrix arises through the ignoration of
the three in-plane rigid body freedoms which cannot affect the stiffness
of an element. For a flange the corresponding stiffness is evidently the
( 1 x 1 ) expression EBjl. We propose the connotation natural or
invariant for such stiffnesses. Their introduction has proved a most
significant advance both from the fundamental and programming
point of views, see also Appendix V.
D.d. Buckling
Any initial buckling problem which may under certain conditions
arise in a system under the given static loading is easily analysed by
the theory of the preceding paragraph. In fact we only need consider
the first incremental step from a given equilibrium condition into an
adjoining one which may set in for a zero incremental load vector R A .
Thus, using equations (D.10) and (D.l 1) in the form
[K, ; + K , ; ] r = o (D.76)
the determination ofthe buckling loads and the corresponding buckling
modes is reduced to a typical eigenvalue problem. Introducing an
intensity factor X to denote the unknown magnitude of the load system
under which instability may develop we can rewrite Eq. (D.76)
[K K + AK„]r = o (D.77)
which clearly shows the standard programme to be used in this case.
Evidently we may also define an intensity factor 0 of the temperature
distribution, if such is applied, and split the matrix K „ into two
component matrices proportional to H and 0 respectively. For a more
detailed discussion see Appendix V.
APPENDIX I
On the Stiffness of Triangular Elements
INTRODUCTION
In this Appendix we establish the stiffness of triangular elements
as used by us since 1956. Following on the idealization o f t h e cover
120
we draw a grid system which partitions each of these surfaces into a

number of triangular elements. To fix the geometry of our structure
we use in addition to a common coordinate system Oxyz also for each
plane cover a "local" system O'x'y'z', so chosen that the axis O'z'
stands perpendicular to the corresponding surface. The origin 0 ' and
the direction of the pair of axes O'x', O'y' arc strictly arbitrarily (see
Fig. 1.1). At the same time it proves convenient to apply a standard
Fig. 1,1, Triangular elements in wing cover.
Fig. 1.2. Triangular element in curved cover.
rule for their selection when the cover displays such pronounced
curvature that it appears in its idealized state as a polyhedron, each
facet of which contains either one or at most very few triangles. In
this case it is advantageous to take for each plane the axis O'x' coalescent
with one ofthe sides ofthe triangle (or one ofthe triangles) say (1,2)
and to fix the origin 0 ' at say the vertex " 1 " (see Fig. 1.2). Anyhow,
by whatever procedure the local coordinate system is chosen, we first
determine the stiffness of the triangles in the direction of its axes.
121
J, II. A R G Y R I S , S. K E L S E Y , a n d H . KAMEL
Applying next a standard congruent transformation, we derive the

stiffness matrix k of all constituent triangles with respect to the basic
system Oxyz. T h e present procedure is actually being superseded in
our programmes bv the natural stiffness k T derived on p. 119 (see
Eq. (D.73)).
(a) Basic Geometry

Consider the triangle (1,2,3) of Fig. 1.3 whose position relative to
the basic coordinate system Oxyz is fixed by the ( 3 x 1 ) column
matrix
A.,
+1 = •I*, y% (1-1]
Z-l
L---2J.-P.
of the coordinates of the three vertices. Assuming for the moment that
the origin 0 ' and directions of the axes of the local coordinate system
O'x'y'z' appertaining to the flat surface containing the triangle in
question is known, we may write the 3 x 3 matrix of the relevant
direction cosines in the form,
a «i 11
1
t£ b = K bt b. (1.2)
c Ji < • •
' :i
or in detail
"cos {x',x) cos {x',y) cos {x',z)'
d = cos (y',x) cos (y',y) cos (y',z) (1.3)
_cos (Z',A-) cos [z',y] cos {z',z)_
Note that the third row specifics the direction cosines of the O'z' axis,
which is normal to the plane of the triangle. To find the column
matrix d/ of the coordinates in the local system
* ' = {*' / A (1.4)
we apply the standard transformation rule
d/ = CN> - - K l (1*5)
where
+» = {*o- Jo- -V) (1-6)
defines the origin 0 ' with respect to 0 . In what follows we also make
use of the notation
•K = *+\ - *\>t
= {(*.•--*,) ( j i - y , ) (*.-*,)} (1-7)
Let us now turn to the determination of the matrix C when the
system O'x'y'z' is fixed with reference to a particular triangle by the
122
rule mentioned in the Introduction. Then the row matrix a of Eq.

(1.2) may be immediately expressed as
T Mt (1.8)
'12
where
/12 = V'x\x -r-^li + ^21 (1.9)
From the programming point of view the specification of a merely
requires the standard instruction of a (mathematical) normalization
of the row <|»{2. Next we fix the direction cosines of O'z'. To this
effect wc introduce the column matrices
X = {X1 Xr, x3}

Y = {Ji J>i Ja\ (1.10)
z = {z t z 2 z3)
which may be obtained in the computer as the three respective columns
of
Wc also define with the symbols

x
+> y+> z +
the column matrices whose first element corresponds to vertex " 2 "
instead of " 1 " ; for example
y+ = {fi y* y.) (1.10a)

These matrices arc again obtained by simple instruction in the com-
puter. With this notation it is easy to prove that the row matrix c
may be found by mathematical normalization of the matrix
[(«'y+ - y' x +) (y'« + z'y + ) (z'x + - x'z + )] (i.ii)

To prove our assertion we observe that the projection of the triangle
1,2,3 on to the planes Oxy, Oyz, Ozx yields triangles with the respective
areas Llxy, il,,,, il, x determined from (see Fig. 1.3)
*] y. 1 7i 1
2flU - *i 72 2i2„ Ji *A%
A
;l y» yz -:>
-1 *i
2"„ = -P X. (1.12)
Zl *a
123
J . H. A R G Y R I S , S. KELSEY a n d H. KAMEL
il 2 . W
x.u
Fig. 1,3. Analysis of triangular element.
The row matrix c is now simply
QVI Ii„] (1.13)
where ii is the area o f t h e triangle 1-2-3 itself. Using Eqs. (1.12) wc

immediately confirm the correctness of the statement (1.11). If our
programme library also includes the evaluation of determinants, direct
application of Eqs. (1.12) may be preferable.
There only remains to find the direction cosines of O'y'. This is
most simply achieved by drawing the line (3,4) perpendicular to (1,2)
and obtaining the coordinates of point " 4 " . Simple geometry in
matrix form vields
+M = |I o'oW.3 (1.14)
124
The mathematical normalization of i|»34 yields finally the row matrix

b which concludes the determination of the transformation matrix tt.
Application of the specific transformation matrix tt obtained above
to Eq. (1.5) yields
* ' = tt[v|> - t ^ ] (I.5a)
and, in particular, for the three vertices ofthe triangle (1,2,3),
«|»J -= o, u>2 ttvi>2, <K = GvJ>3 (I.5b)
For the derivation of the stiffnesses under (b) it is convenient to

introduce
'' = w *i] (1.15)
y' = iy'i yi yS
which are best obtained as the first two columns of the matrix
We also require in the subsequent analysis the extended 9 x 9 matrix

tt, of the direction cosines which is given by
ti o o
-i 0 ti (i (1.16)
0 0 li
The usefulness of Eq. (1.16) is immediately apparent if we consider two

alternative sequential arrangements of the displacement components
u, v, w at the vertices of a triangle. Thus, we may use either in the
Oxyz or O'x'y'z' system the following specification of the displaced
state of (1,2,3) which involves in all cases a (9 X 1) column matrix
P = {«! "2 "i Pi •' ':i •'I wt "•:!

(1.17)
P' = {U\ III, U3 !'', V'm V3 ';• w'i '•' '• '':',
and
P - l«i i'i VDt ":, ':. W3
(1.18)
p ' = {u[ v[ u., V. •a'.. «8 »a ' ' • : :
Application of Eq. (1.16) immedialcly yields
p' = tt,p (1.19)
The elements of the vectors Eq. (1.17) may be rearranged into those of
Eq. (1.18) by the straightforward matrix operation
Tp and p' TP' (1.20)

125
J . H. A R G Y R I S , S. KELSEY a n d H . KAMEL
where T is the 9 x 9 location matrix
1 1) 0 (1 1) (1 (1 1) (1
0 0 0 1 1) 0 0 (I (1
(1 0 0 0 0 0 1 (1 0
0 1 0 0 0 0 0 0 0
(i 1) 0 0 1 0 (1 (1 0
(i 0 0 0 0 0 0 1 1)
0 0 1 0 0 0 0 0 0
0 (1 0 II 1) 1 (1 ll 0
0 0 0 II (1 0 0 (1 1
Ell E21 *"3l

E
12 E
2 E:12 (1.21)
E13 E2
In the condensed form of Eq. (1.21) E 0 signifies a 3 x 3 submatrix in

which the only non-zero clement is a " o n e " placed in the ith row and
j t h column. Some properties of the T matrix are pertinent. Thus, we
note that it is symmetrical and orthonormal,
T* = T and T'T
and hence
T* = T = T _ 1 LLP':
If we now substitute Eqs. (1.20) into Eq. (1.19) and premultiply with
T ( = T we find
T'Tp - T'ttfTp
or
P' = tt,-P (1.23)
where
tt„ = T ' t t T = Ttt.T (1.24)
(b) Stiffness Matrix

We start by deriving the stiffnesses corresponding to the local co-
ordinate system O'x'y'z'. It is thereby only necessary to investigate
specifically the u',v' displacements since any stiffness associated with a
displacement w' is necessarily zero in view of our assumption of a
membrane state of stress in the cover. T o determine now the stiffnesses
arising from the displacements u',v' we stipulate that the variation of
displacements between nodal points is linear, the corresponding strains
(and also stresses) being hence constant within each triangular element.
126
It is convenient to specify any strained state due to a unit displacement

mode i at a vertex by the ( 3 x 1 ) column matrix
e E
«l = K v v-v- x'v] (L25)
The expressions for these strains are actually very simple as may be
seen in Refs. 3 and 4. However, from the practical point of view it is
preferable to obtain the «' strains directly in the computer by a single
inversion. This procedure is described here briefly.
Let us first define the (3 X 9) matrix of all column matrices t't—
including those zero ones due to w' displacements—arrayed in the
sequence corresponding to p' of Eq. (1.17)
ft = [ < ] (1.26)
It may be shown that the matrix a may be expressed as
p' 0 0"
0 Y' o (1.27)
Y P' oj
where the ( 1 x 3 ) row matrices (J' and y' are determined in the
computer by the straightforward inversion,
{a' p' Y'1 == [•« *' y']-1 (1-28)
The column matrices x ' , y ' a r e given in Eq. (1.15). We may now obtain
the ( 3 x 1 ) strain matrix c' due to any displacement vector p' from
e' = ap' (1.29)
The associated stress matrix
o' = {<r. °xy} (1.30)
follows from
o' = xe' = x a p ' (1.31)
in which x is the stiffness of a unit membrane element
I V (I
I (I
x = 1 (1.32)
-
0 0
To find now the (9 X 9) stiffness matrix k' associated with p' we apply
most conveniently the unit displacement principle established in Ref. 1.
This states that any stiffness coefficient kjt\ is determined from the
relation
k'lh = JjVejx€„ clx'dy'
= J K x e , dx'dy' = k'hi (1.33)
127
J . H . A R G Y R I S , S. KELSEY a n d H. KAMEI.
where t, the thickness ofthe sheet, may vary arbitrarily with x' and_y'.
We may next generalize Eq. (1.33) to refer to all displacements p' in
the sequence of Eq. (1.17) and obtain in conjunction with Eq. (1.31)
k ' - [k'ih]
= f ffo'xo dx' dy' (1.34)
Assuming uniform thickness o f t h e element Eq. (1.34) reduces to our
well-known expression of Refs. 3 and 4,
k ' - i2/o'xo (1.35)
in which il, the area of the triangle, is given by
2i2=[e3 x' y'] (1.36)
The last three columns and rows of Eqs. (1.34) and (1.35) are, of course,
zero. However, in practice we do not make use of this fact and find
k ' directly in the computer, applying for a uniform element the matrix
operations of Eq. (1.35). This has proved the most convenient
procedure of establishing the stiffness matrix of a triangular clement.
If for a moment we assume that the nodal loads R ' on an clement
arc also measured in the O'x'y'z' system we have
R' = k'p' (1.37)
T o find now the stiffness matrix k referred to the loads R and the
displacement vector p of Eq. (1.17) for which the matrix condition
R = kp (I.37a)
holds, we note the dual relations which arise from Eq. (1.23),
R = C,;R' and p' - tt„p (1.38)
Introducing Eq. (1.23) into Eq. (1.37) and premultiplying with tt,' we
find using the first of Eqs. (1.38)
tt(1R' = R = tt/,k'tt„P (1.39)
Hence from Eq. (I.37a)
k - tt,',k'tt„ (1.40)
APPENDIX II
Stiffness Matrix for an Arbitrary but Plane Quadrilateral Panel
INTRODUCTION
Following on our analysis of a triangular panel we present a corre-

sponding investigation for an arbitrary but plane quadrilateral panel.
As in our previous work on the displacement method we assume that
128
the displacements u and v within a panel are uniquely defined by the

eight displacement components at the vertices or nodal points. Inter-
estingly enough this necessitates, in contrast to the procedure for
triangular and rectangular elements, the adoption of displacement
functions with a certain non-linearity. But in any case these functions
represent the simplest possible choice in order to satisfy the kinematic
boundary conditions. Although more complicated displacement forms
are being investigated, there are a number of reasons why their applica-
tion may not prove very fruitful. The following analysis is restricted
to the derivation of the in-plane stiffnesses. The extension to an
arbitrary coordinate system is in principle similar to that of Appendix I
and is omitted here. For convenience wc adopt in what follows a
slightly different notation.
(a) Basic Geometry and Kinematics

As in Appendix I we start by defining a number of basic matrices.
First, the geometry of the quadrilateral is fixed with reference to an
in-plane coordinate system Oxy by the 4 x 1 column matrices of the
vertex coordinates (see Fig. I I . 1)
Fig. 11.1. Quadrilateral panel element.
IP Xa Xn Pl
(II.1)
y , = {y. y-i y 3 y*}
Following on what we mentioned before we propose to express the
state of deformation solely in terms ofthe ( 8 x 1 ) column matrix
[th p; (II.2)
of the displacements measured in the x a n d y directions [cf. Eq. (1.17)].
We deduce from Eq. (II.2) that eight constants must, in general, be
determined to describe the deformation pattern in the panel for any
set of nodal displacements. With this in mind we introduce the
129
J- ARGYR1S, S. KELSEY a n d H . KAMEL
following two specific states of deformation associated with unit values

of p in the x andj-directions respectively.
Thus, we fix for any u, in the x-direction
u( = a,. + fttx + y,y + btxy
(II.3a)
v, = 0
and for any pf in thej-direction
U, = 0
(II.3b)
Vj = Xj + ftjX + y j -f- bjxy
Note that the suffices i and j in Eqs. (II.3a) and (II.3b) take the values
1, 2, 3, 4 respectively. It follows from our definition of the u and 0
functions that the coefficients a, ft, y, d due to a unit displacement in
the x orjv-dircction at one and the same vertex are identical.
It will be appreciated that our specification of the displacements u
and v involving a term in xy does not allow for full compatibility of the
deformations along the common side of two adjoining quadrilaterals,
but for the limiting case of rectangular elements with sides parallel to
the x and^> axes. In fact, in the last case our expressions for the dis-
placements are identical with those given in "Energy Theorems" 1 .
We also note that in contrast to the triangular panel linear functions
do not suffice to satisfy the boundary conditions.
The 4 x 4 set of coefficients arising in Eqs. (II.3a) and (II.3b) may
now be determined very simply and in a concise matrix form. To this
purpose we define the current ( 1 x 4 ) row matrix
4* = [1 x y xy] (II.4)
Next we assemble the a, ft, y, d coefficients as the 4 x 4 matrix,
A jot p Y 8} (11.5)
or
Pi Pi Ps Pi
A (II.5a)
Yi Yi Ya Yi
d, d2 d3 ii4
where we use the numerals 1) to (4) to den ote quantities associated
with unit displacements of the vertices 1 to 4 in either the x or y -
directions. Application of Eq. (H.5) to Eqs. (II.3a) and (II.3b) yields
u = t\,A (II.6a)
and
u„ = o
(II.6b)
v„ = u>A
130
where the ( 1 x 4 ) matrices u „ v x , etc., stand for,
« . = ["l "2 «3 ««]

(II.7a)
v , = [0 0 0 0]
and
u„ [0 0 0 OJ
(H.7b)
v„ = [«', '-.I
The suffices I to 4 specify displacements arising from unit u or v at
the corresponding nodal point.
To determine the matrix A in one operation it is useful to set up the
4 x 4 matrix of the vertex coordinates
*Pr = le 4 x,, y,. x,,. y„] (II.8)
or
1 *i y i *iyi
1 x2 y 2 x2y2
T, (H.8a)
1
*3 ya x3y3
1 xt y t *4j4.
The notation x,, . y,, has been chosen to denote an element by element
multiplication of the two column matrices x,, and y,.. In a less sym-
metrical but more standard symbolism it can be represented by the
operation x^y r or yfx,,.
The boundary conditions for the u and v displacements in Eqs.
(H.6a) and (II.6b) respectively may now be written as
«F, A = I. (II.9)
and hence [cf. Eq. (1.28)]
A = V-1 (II.9a)
Note that the inversion of T,, is always feasible since the matrix *P„ can
never become singular for a quadrilateral.
The total displacements for any displacement vector p may now be
expressed as
"==*[A °JP (11.10)
v = iHO A]p
(b) Strains and Stiffness Matrix
Considering next the strains due to the prescribed unit displacements,
we observe that for the states u1 to u4 we have
e«N = Pt + -5ty, e„t = 0 (II. 11a)
and
d X
£ « , = Yi + i
131
Similarly for the states St to vA

e
°, vvi = Yi + -V
and (II.lib)
W. = Pi + b>y
The reader will note that the strains within a quadrilateral are not any
longer constant as for a triangular panel. At the same time the
expressions for the strains in Eqs. (11.11) are in principle similar to those
for the rectangular panel. Thus, the direct strains EXX arising from unit
displacements in the x-direction are seen to vary linearly not with x,
but with j , whilst the associated shear strains depend only on x. The
opposite is seen to be true for the strains £„„ and EXU corresponding to
unit displacements in the j-direction.
As in "Energy Theorems" and in Appendix I, we represent the state
of strain connected with a unit displacement p t = 1 by the 3 X 1
matrix
*/ X^XX *VV XV} I (11.12)
T h e corresponding stress matrix
<*i = { a xx a
vv a
xv\i (11.13)
is deduced, following Appendix I, from
o, = xe, (11.14)
in which the stiffness x is once more given by

"1 v 0
V 1 0
x = (11.15)
1
0 0
We now assemble the two 3 x 8 matrices [cf. Eq. (1.27)]
P o- [8 0
0 = 0 Y b 0 8 (11.16)
Y P_ 8 8
Using Eqs. (11.16) we may write the (3 X 8) matrix
«« = [ £ u ' -is IV •1', ll (11.17)
reproducing the strains in the quadrilateral due to the eight unit

displacements at the vertices in the concise form
e, = a + Xb (H.18)
132
where x is the 3 x 3 matrix

' y 0 0"
X 0 x 0 (11.19)
x y 0
and is the only matrix in Eq. (11.18) containing the independent
variables x a n d y . Note that the matrix product x& stands for
[>8 0
Xb 0 vS (11.20)
.«8 J8J
[[see also Eqs. (11.11)].
The ( 3 x 1 ) strain and stress matrices e and o corresponding to an
arbitrary set of vertex displacements p may be obtained from the
self-evident formulae
e = €.p and o = xe (H-21)
Eqs. (11.18), (II. 14) and (11.15) yield now all the necessary informa-
tion for the derivation of the stiffness matrix k corresponding to p. In
fact, direct application of the unit displacement theorem yields for the
8 < 8 stiffness matrix k
k JJe.'xe,., dxdy (11.22)
in which I is again the sheet thickness. Substituting Eq.s. (11.18) and
(11.15) and integrating over x,y we obtain the following simple matrix
formulation of the stiffness matrix
k o'k.,0 I [o'k.b -f- b'k|o] + b'k.b (11.23)
where
" E' vE' 0"
ko = M 0 vE' E ' 0 (11.24)
0 0 G
E'M X vE'M„ 0
k.= vE'M x E'M U 0 (11.25)
GM„ CM. 0
and
'E'M XX + GMUU {vE' + G)M xy
{vE' + G)M xy E'M y y + GMx, (11.26)
0 0
The scalar values M 0 , M „ M v , M xx , M yy , M xy represent the standard
integrals
M n = $$t dxdy, M x = JJ/jv dxdy, M u == jijitx dxdy
and (IL27)
M , . - Hty 2 dxdy. Mm, = Sitx- dxdy, M xy = jjtxy dxdy
133
J. H . A R G Y R I S , S. KELSEY a n d H. KAMEL
which reproduce the total "mass" as well as first and second order
"mass" moments of the quadrilateral. Programmes are available for
their evaluation as functions of x v ,y v and a constant thickness t.
Naturally, provision ought also to be made for their evaluation in the
presence of a thickness t varying linearly with x stndy. For numerical
convenience it is advantageous to operate with a local system of axes
which may be taken to originate at one of the vertices, say 1'. Then,
«i = (° *i —*i «t ~ x i x i - x i) , TT 0Q s
rn * (11.28)
yK = {0 y 2 - j i y 3 - j i y t - y - t ]
The column vectors Eq. (11.28) are easily established automatically in
the computer.
I n our experience the formulation of the stiffness matrix k as set out
in Eq. (11.23) is the most suitable from the programming point of view.
Nevertheless, it may be of some interest to reproduce the matrix
expressions arising when the transformation operations of the second
and third terms are explicitly evaluated in terms of the matrices |3, y
and 8. We find
VE'M&h + GMrfl v E ' M & h + GM x y'81
1 [
\_vE'M£'h + GM y y'S E ' M y y ' 8 + G M ^ ' b ] ' '
and
b k
*b ==
L (.'£' + G)M x y VS {E'M yy + GM XX )S'S] [Ll
"i^>
The drawback of the displacement specification applied in our above
analysis of the quadrilateral is the aforementioned lack of kinematic
compatibility arising, in general, along the common side of two
adjoining quadrilaterals. A possibly more serious objection to our
method is the fact that the stiffnesses are not invariant with the direc-
tions of the axes. On the other hand the advantage of the approach
lies in the relative simplicity and compactness of the formulae. It is,
of course, feasible to prescribe a kinematic pattern which does not
violate anywhere the compatibility conditions. In fact, the technique
in question must involve a suitable generalization ofthe procedure laid
down for the rectangle and parallelogram (sec Appendix I V ) . T o this
effect we introduce a non-dimensional parametric presentation {x,y>) of
a point within the quadrilateral, which obeys the following rule.
Thus, if we draw straight lines through the point {x,y) which intersect
opposite sides in the same ratio, say, % and xp, there is a unique relation-
ship between {x,y) and {x,y)', see Fig. I L L The four sides o f t h e
quadrilateral (1,2), (2,3), (3,4), (4,1) are now denoted by y> = 0,
% = 1 ( y> = 1 f y_ = 0, respectively. If we express now the cartesian
displacements u, v as,
") = « + Px + Yy + ozv (n-30)

134
the displacements along the edges of the panel are seen to vary linearly
and satisfy hence the compatibility at the junctions of quadrilaterals.
Moreover, for any of the unit displacements the coefficients a, ft, y, 6
can only take the values ( + 1), ( — 1) or zero. The expressions for %
and tf in terms of x and y are, however, not simple and the cartesian
strain matrix e will be found to be involved. Inevitably, the stiffness
matrix k cannot be given in as neat a form as in Eq. (11.23).
APPENDIX III
Stiffness Matrix of Spar Elements

INTRODUCTION
Wc follow up our general discussion of the assumptions which must

underline a rational displacement analysis of spars by deriving the new
6 x 6 stiffness matrices of a spar element we have used in our latest
wing investigations. The results are presented below under two head-
ings (A) and (B). In the first section we obtain the stiffness matrix for
a tapered spar element with constant area flanges using the E.T.B.
modes of deformation strictly appropriate for a uniform spar. An
example on a spar with pronounced taper, reproduced in Fig. B.9,
illustrates the very satisfactory convergence of" the method. At the
same time, to verify this analysis, which ignores the Poisson's ratio
effect, we develop under (B) the exact expressions for the stiffnesses of
a uniform web without flanges. Not unexpectedly we confirm for this
limiting case the very good accuracy of our simplified investigation
under (A). [We note incidentally that the theory of elasticity solution
of (B) is only exact as long as no incompatibility of warping arises at
the joints of elements.]
Although our new specification of spar modes and stiffnesses has
yielded results in good agreement with experimental evidence and
represents a significant advance on our past and other parallel work,
there still remain a number of open questions requiring clarification in
future.
Figure I I I . 1 illustrates the coordinate systems x,z and x,z specifically
introduced for the analysis of the spar element. The local x-axis
applied here is, of course, in general different from the x-axis of the
xyz system used in the complete wing analysis. We now introduce the
following symbols
C = z/d
X = x\d. x = *M
U = zfd, ti = zjd
(IIII)
Zm = (Z,. + Zm)/2, Cn, = Z j d
Z
A =
Z2 Zi, U = zjd
135
J . H. A R G Y R I S , S. K t L S E Y a n d H . KAMEL
—mX- —i-*X —
mode b.
mode r.
S2-1
mode s 2
F"ig. I I I . 1. Kinematic deformation modes for tapered spar element.
We express furthermore the current coordinate z' of the upper flange

in the concise form,
z' = z m + Z A X
or (III.2)
r = *« + CAZ
where
£' - z'/rf (III.la)
(A) The Tapered Spar Element
We first define the displacement modes which arc to describe
completely the deformations of a spar element. Since it is reasonable
to assume that the web is incompressible in the z-direction, three modes
generated at each end of an element yield the complete deformation
pattern. They are shown in Fig. I I I . 1 and may be denoted as:
(a) A translational bending associated with zero rotation of the
displaced end (mode b)
136
(b) A rotational bending associated with zero deflexion at the

rotated end (mode r)
(c) A pure shearing (mode s).
The assumed kinematics for modes (a) and {b) are selected in such a
way that they give rise to zero shear strain (classical E.T.B.).
We quote below the expressions for the displacements and associated
strains and stresses. In each case we give for completeness and for
convenience of the reader the formulae for the three modes generated
at either end ofthe element (sec also Fig. I I I . l ) . At the same time we
do not reproduce the a zz stress (which does not vanish), since with e„
throughout zero a2Z does not enter into the stiffness expressions.
Furthermore, for consistency of the stresses at the joints of flanges and
webs and also in order to obtain the standard E.T.B. results for uniform
spars, we use E instead of E ' = Ej{\ — v2) in the formulae for axx.
To describe completely the strain pattern in the spar we must also
have the explicit expressions for the strains in the flanges. Considering
the upper flange / shown in Fig. I I I . 1 inclined at an angle 0 to the
x-axis we deduce from the standard strain transformation formula
e,f = EXX cos2 0 -f E zz sin2 0 + f „ sin 6 cos 6
in conjunction with r z l = 0, that
Fff = cos2 0{EXX + EZX tan (1)
= co8«0(e„ | k « J (III.3)
For the small angles of taper usually encountered in practice we may
put cos 0 == 1, However, no computational difficulty arises in retaining
the cos2 0 factor.
(a) Translational Bending

The generating nodal displacements in the z-dircction, denoted here
with b, are preferably designated with wb when forming the displace-
ment vector r ofthe complete structure.
(a) Mode h2 —• 1. Displacements
w = p { 3 - 2x)
(IIL4)
Strains
.--w-a.
du 12 „
=
Yx = - d *
(IIL5)
= Yz + Yx = °
= — cos 2 0/X'
137
Stresses
12E
°xx = ~f xi
(III.6)
12E
Off = — r cos2 Oyf
Note again that the assumed displacements are those of the classical
E.T.B. associated with zero shear strains; also the displacements are
strictly appropriate for a uniform member.
{ft) Mode / > ! = ! . Displacements
«• = (! - * ) • ( ! +2jF)
(III.7)
« = 6 x i l - z)£
Strains
' XJ- / /*=
0 (III.8)
12
eff = - • - j cos 2 fi^'
Stresses
12£ ,
oM = 0 (III.9)
12£ . . ,,
(T/r = ^ COS" 0 / 4
(b) Rotational Bending

The generating nodal displacements in the local x-direction, denoted
here with r, could also be designated with u s p (see p. 61). T h e rela-
tion between u 8p and the standard displacements u,v measured in
the basic system is given by [see also Eq. (B.23)],
« s p = [cos (s,x) cos {s,y)] (III.A)
(a) Mode r2 = 1. Displacements
w
= yX 2 i l - X)
t2
« - T X&X - 2)C (111.10)
138
Strains
•V- = 4-(6z+ l)C

£ra = 0 (111.11)
2
cos 0
c
// </£,
(6/ + \)i'
Stresses
<l,m
az, 0 (111.12)
2
£ cos 0
"" " dim (6* + or
{ft) Mode r, = 1. Displacements
"• = - r (1 - *)2*
« = 7 ( 1 -*)(! -Sf){ (IIL13)

ti
Strains
' (1 4-6 z )C
/ •
rj-
= 0 (111.14)
cos
°n LC \f
Stresses
*« = ~ -5- (i + 6*K
«ti
trM = 0 (III.15)
£ cos2 0
J
/r (1 + 6Z)C
dU
(c) Shearing
The generating nodal displacements, denoted here with s, are
preferably designated w, when forming the displacement vector r of
the complete structure.
(a) Mode s2 — 1. Displacements
w = x, « - •- (IIL16)
139
Strains
= E„ = 0
= \/d, E„ = {CJd) COS2 0 (111.17)
Stresses
<*xx = o „ = 0
a „ = G/d, a „ = { E i j d ) cos2 0 (111.18)
{ft) Modes. = 1. Displacements
w = 1 — x,
u = 0 (111.19)
Strains
E
xx = £« = 0
e„ = -\\d, E „ = - { C J d ) cos2 0 (111.20)
Stresses
a
xx = <*zz = °
azx= -G/d, a,f= {ECJd) cos* 6 (111.21)
(d) Stiffness Matrix
We may now evaluate any stiffness coefficient k tj associated with
two kinematic modes i and ; from the unit displacement method which
in the present case may be written in the partitioned form
kii = {<7 ri ) di dy, + 2 d j + *BaffE„ A

h (IH.22)
^^ Sc ' +C
where B is the flange area. Straightforward evaluation of the integrals
leads to the 6 x 6 stiffness matrix ofthe spar. To write it in as concise
form as possible it is advantageous to define a number of basic matrices.
First we introduce three purely numerical matrices of dimension
6 x 6 which are independent of the specific geometry of the spar
elements. Thus,
1 2 -6 ii 0 0"
2 1 -6 i. 0 0
(. (i 12 -12 (i 0
'.'I (111.23)
6 6 -12 12 0 0
II 0 i) 0 3 3
1) 0 I) 1) -3 3.
2 0 2 -2 1 r
(I 2 -2 2 -1 i
2 -2 0 i) 0 ii
(III.23a)
2 2 0 0 (l 0
i 1 n 0 0 i)
1 1 (i 0 0 0
1 K)
ind
11) 1 1 -27 27 -15 15
11 16 27 27 -15 15
27 -27 54 -54 30 30
a (III.23b)
27 27 51 54 30 30
15 -15 30 -30 30 -30
15 15 30 30 -30 in
Then, we identify three 6 x 6 diagonal matrices which in contrast to

Eqs. (111.23) depend on the geometry ofthe trapezoidal web. In par-
ticular.
1 i G\_ (. i
I
di =
Ci w lie...
1 i
d, -
r, w i)
i 1
(111.24)
Xi u
Using Eqs. (111,23) and (III.24) the 6 x 6 stiffness matrix
r. ft, ft.
r.
P
x
ih
(111.25)
ht
•1
or
k = k„t + k,. (HI.25a)
where k„t represents the contribution of the web and k w that of the
flanges, is obtained from the expressions
[gt == iftc&fu,
Ft
(111.26)
+ V5CmCld.Mi
Ft
+ 2 0 & W .
141
or
k s t = }Ett rt m d 1 md 1
t-2
+ Et [Cl + H) ^d 2 »d 2 (III.26a)
Et,
+ ~ £„.£d2ttd
15
and
2EB
B= -c«md,«d2
d
2/<7?
+ 5 C m C^./Bd, (111.27)
The expression for the contribution k / ; of the flanges assumes that

cos 0 * 1, but any value of cos 0 may be allowed by introducing in
Eq. (27) the common factor
CTO3fl =
[ i + a ] * * ' 1 " •** (IIL28)
which differs, in general, only to an insignificant degree from unity.

Furthermore, a perusal of Eqs. (111.26) and (111.27) indicates that we
could in many cases neglect safely the component matrices which carry
the factor TA or i \ . O n the other hand, having a general programme
for k based on Eqs. (111.26) and (111.27), its computation on a large
digital machine is so straightforward and fast that it proves unnecessary
and inconvenient to consider the limiting forms for small JA<
(e) Strains in Spars

Having obtained the nodal displacements of the structure by inver-
sion of the stiffness matrix K or otherwise, we have then to compute
inter alia the main strain and stress values in the spars. We reproduce
here only the formula for the flange strain eff at the midpoint of the
element and the average shear strain in the element. It is easy to
derive also the flange strains on cither side of the nodal points but the
expressions are more complex than at the midpoint and should not
be required in practice. 2
(a) Shear strain in web. The average shear strain in the web is
immediately seen to be given by
w w
e zx = " ~ * (111.29)
Hence the shear stress
a „ = GEZX (III.29a)
142
{ft) Flange strain at midpoint. O u r general flange strain expression is

e„ = cos2 0{EXX + UE Z X ) (111.30)
Having obtained EZX in Eq. (111.29) we merely have to compute sxx.
Noting that at the midpoint the translational bending has a point of
inflexion and that EXX is zero for the third or shearing mode, we find
that only the rotational displacements, denoted here as in Eq. (III.A)
with Msp, enter into the required expression for s /f —which shows the
great advantage in seeking the strain at midpoint. Now, application
of Eqs. ( I I I . l l ) and (III.14) for x = 0 and I = Cm yields the simple
formula,
_ £m "sp2 £ m M-SP1
z2 d Zj d
_ Cm l U S P i _ Wspij / T I T rj,v
or {
" rf\C, ' cJ '
Having found rxx the determination of the corresponding E U is straight-
forward.
In our practical computations we have also used a different approach
in the definition of c lf which is not so attractive as the above one but
may serve as a check. Thus, defining the displacement vectors or
column matrices
* - { * * *} ( n L 3 2 )
r , = [«, v, a.,}
at the end points (1) and (2) ofthe element whereby
w = w„ + w, (III.32a)
we obtain the change of length ofthe line (1,2) from the relation
M, = H,[r 2 - r , ] (111.33)
Here C r is the 1 x 3 row matrix of the direction cosines of the flange
element (1,2). Thus,
H, = [cos (/,x) cos {fy) cos (/,z)] (111.34)
The "average" strain i f t of the spar element is now
i„ = % ! (111.35)
h
where I, is the length of the flange element. It is interesting that in
computed wing examples relatively small differences were, in general,
only noted between the strain measures Eff and l l f .
(f) Comments
For uniform spars our assumed modes are effectively those of the
Engineers' Theory of Bending which incorporates in the present case
143
J . 11. A R G Y R I S , S. KBL8EY a n d H . KAMEL
a shear strain constant over any cross-section and within each element.
It is hence evident that the deflexion of a cantilever under a tip load
deduced by our stiffness method must be independent of the number
of elements in which the cantilever is partitioned. In fact, the deflexion
must be identical to those obtained by a straightforward application
o f t h e E.T.B. in question. This is confirmed for a particular cantilever
in Fig. B.10 where the continuous line parallel to the horizontal axis
indicates the invariance of the results with the number of elements.
O n the other hand, our conclusion does not hold for tapered spars or
for untapered ones when the flange area is varying. Figure B.9
illustrates this for a tapered spar and for three different but constant
flange areas. As to be expected, we observe a slight rise in the com-
puted deflexion with the number n of elements. However, the difference
in the displacement corresponding to four or eight elements proves less
than 1 per cent and the convergence of the procedure appears very
satisfactory. We have also substantiated the correctness ofthe estimated
deflexion through direct analytical solution ofthe tapered spar problem.
(B) The "Exact" Stiffnesses of a Uniform Web

In the course of our investigations on a more rational displacement
analysis of spars, we considered a number of schemes alternative to that
described under (A). O n e of these may be of interest since it yielded
the exact stiffnesses of a rectangular web without flanges. The resulting
6 x 6 matrix—which includes, of course, the Poisson's ratio effect—
is surprisingly simple. Not unexpectedly, the application of these
stiffnesses to a rectangular cantilever under a tip load yields a deflexion
which differs only to an insignificant degree from that given by the
elementary theory of (A). At the same time the value of the present
analysis is very much limited by the exclusion of taper and any attached
flanges. To take account of the flanges requires—as mentioned in the
main text—their being spot-welded at the nodal points which may lead
to a serious underestimate of the spar stiffness. Thus, the consistent
investigation under (A) is for all practical purposes to be preferred,
the value of the exact analysis lying mainly in its confirmation of the
simplified theory for the limiting case of a uniform web.
We reproduce below the derivation of the "exact" spar or web
stiffnesses in a very condensed presentation. Since this theory-of-
elasticity solution is associated with a parabolic shear stress distribution,
it inevitably involves warping of the cross-section. No constraint of
warping—which does not vary lengthwise within an element—is
allowed for, the supported end of the element being held at its upper
and lower nodal points as shown in Fig. I I I . 2 . T h e unrestrained
warping requires, of course, that the shear force is constant along the
spar {i.e. a tip load). In order to simplify the analysis of the wing it is
advantageous to assume equal vertical displacements at the upper and
144
lower surfaces. To achieve this here for the spar elements wc have to
admit a finite shear stress on the upper and lower edges of each
clement; in this respect the stress distribution differs from the usual
one for a cantilever. At the same time this assumption essentially
contributes to the simplicity of the final expressions for the stiffnesses.
z z
!
t . h
— a . * — -1 ••/
1
— - d -
mode b.
mode r 2
mode Sj
Fig, 111.2. "Exact" deformation modes for rectangular spar web element.
We start by presenting the three modes which may be generated at

one end of an element, here taken to be end. 2
(a) Mode b t Displacements
U-
4(|(3//-2x)+^(//-2x)(z2-^))
rf»\6 (111.36)
- -•(*-*+^(,-S))
145
[. H. ARGYRIS, S. KELSEY a n d H . KAMEL
Strains and Stresses
Cxx = - J XC e z z = - - — vxC
(111.37)
1|(1 +v)(p_/l.)
and
VIE
<*xx — d XC azz = 0
(111.38)
?(P-^
where
i'^V
and
»7 = -j (III.39a)
Note again that the shear stress at the upper and lower edges is not
zero, having in fact the value,
E 2+ v
2
*"=-2^ TT^ <IH.38a)
(b) Mode r2 — 1. Displacements
- = ^{|^-2^)+V2(*i-2.)(z2-i2))
(III 40)
-sH +^'-v)) -
U=
-M\* { d -* ) + — V - 4 ) ) 2 +
//A
Strains and stresses
12 2£ L2 _2»i
C — e V
JK» h h' "~ h h
(111.41)
2
j(l + v)({ -^)
and
12£ 2££
<*«. = y Zt + - y , «r„ = 0
(111.42)
2 2
tr» = - ^ (C - /- )
146
(c) Mode s 2 = I
This mode is identical to that given in Eqs. (A. 16) to (A. 18) under
Section (b) of (A).
(d) Stiffness Matrix

Application of the unit displacement method in the form,
C+l r' J
d
*« = l i ) \*^mfimt + *m*mt) dC dX (111.43)
yields the stiffness matrix reproduced below.

r, r, *, i, j l s,
•a9l»{. + | V ) Elr,(i + /),,•) - E l f , 1 ' ' + /Jii») Etri'd + /Jr/) -vOtf yGl,,*-
/•.(,/(•+ / V ) - » r j « ( l +/Jrj") Etifd + fitf) -rGlif rllnf
E l r f ( l + fir,1) - f t r j ' d + /V) rGttf -tGttf
-t,t
£•«(•( 1 + (Jrj") -tGltf rGltf
•jmunetric Cirj — Ci»j
i.a, .
(111.44)
where
fi =
m + v)(4 + 8v + 9"2) (III,44a)
(e) Comments
Comparing our "exact" stiffness matrix with the approximate one
of Eq. (III.26), as reduced when putting £A = 0, we note that all ele-
ments {b,s) and {r,s) are zero in the approximate stiffness matrix. O n
the other hand these terms vanish also in the exact matrix if we put
v = 0. However, even for v = 0 the two 6 x 6 matrices are still not
identical, the difference being due to the presence of the coefficient ft
in Eq. (44) which is due to the inclusion of the shear dcformability in
the b and r modes of the exact theory. Nevertheless, the difference in
the corresponding stiffness coefficients is really very small (but for the
unrealistic case of a very short element). T h e effect offtand the cross-
terms {b,s) and {r,s) on the deflexion is actually insignificant, as is
indicated for a particular example in Fig. B.10 where it is impossible
to differentiate graphically between the results of (A) and (B).
Before concluding this Appendix we draw attention to some further
characteristic features of the strains and stresses as obtained under (A)
and (B). That EXX and a xx are in both analyses identical is self-evident.
On the other hand, EZZ is zero for modes b and r under (A) but finite
under (B) whilst the opposite is true for a zz . We have already drawn
attention to the nature of the shear stress a zx in the exact analysis.
147
APPENDIX IV
The Stiffness Matrix of a Parallelogram Element
Another type of component we considered in our displacement analysis

of structures is that of a parallelogram sheet element as it may occur,
for example, in the cover of a wing whose spars are parallel in plan
view. To establish in this case a standard programme in the computer
depending essentially only on a single geometric parameter, we used
the following kinematic assumption. Thus, in extension of the work
on rectangular elements in which the variation of displacements
between adjacent nodal points was taken to be linear, we stipulate for
parallelogram elements that the same applies for the displacements
along the oblique net of grid lines. It will be noticed that the assump-
tion is not identical with the first one adopted in Appendix II for an
arbitrary quadrilateral. Hence application of the latter method to
parallelograms would yield stiffnesses different from those obtained
here. O n the other hand the more refined approach referred to at the
end of Appendix I I essentially uses the same kinematic idealization as
the current investigation.
It follows from our assumption that the deformed state of an element
is uniquely determined by the nodal displacements p which we write
as the column matrix
P = {"i f«] (iv.i;

Introducing now the oblique and cartesian coordinate systems y',x'
a n d y , x respectively as shown in Fig. IV.I we stipulate that for a unit
y
"
>
d'
3
I
.*'
w • K,
.i
a^2
K.
U V
*• X
1
d .
Fie;. I V . I . Parallelogram panel element.
148
nodal displacement in the x-direction, say a, = 1, the associated u and

V displacements are given by
.. $(.-') :f -'-/-») (IV.2)

». = o
Also for a unit displacement in thej-direction, say v t = 1, we have,
Similar expressions apply for the other unit displacements. The

complete displacement pattern is reproduced in fable IV. 1.
Table IV. I Specified functions «,, v{for Unit

Cartesian Nodal Displacements
Nodal Displacements u, (or v.) Due to Unit u,

Point [or v . ) ; note v, = 0 {or n, = 0)
1
i'-DM)
2
•H) x y
3
dl
4
(' Vi
Considering next the strains, we obtain for the typical displacement
Um= 1,
1
*m = J ( ii ~ y + 2* tan 0)
y
r = 0 (IV.4)
n
X
1
J--/ = —
///
149
and for z/4 = 1
(IV 5)
^--IO-S) -
•m ^ l > + id - 2x) tan 0]
For each unit displacement «, or p, we form the strain vector
*t - {«•• e „ £xy},- (IV.6)
As in the other Appendices, a typical stiffness coefficient k u may be

obtained from the formula,
k it = Mtefrtj dx dy = fjt&mA dx dy (IV.7)
where the integral extends over the area of the parallelogram and x
the stiffness of a unit square is given in Eq. (1.32).
Substitution of expressions of the type (IV.4) and (IV.5) into Eq.
(IV.7) yields after integration the 8 x 8 stiffness matrix k which is
written most conveniently in the form
^xx *^xy
(IV.8)
tM-vx *I/I/J
where
E'lt Gdt
Kx= d [« + ? » + ? 2 e ] + ,-«
(IV.9)
E'dt Git
2
Kv= J tt+ ^ [ « + ? » + 9 ll]
and
K* = K v = { G + vE')- 6 [*-cp<l]
Here <p is the parameter of "obliquity"
cp = - t a n 0 (IV.10)
and 91, 5B, (£, $ are the numerical matrices,

150
2 -2 1 1
-2 2 1 -1
1 1 2 -2
1 -1 -2 2
-1 0 1 0
l) 1 0 -1
» = i
1 (i -1 it
(i 1 0 1
(IV.ll)
2 1 1 2
1 2 -2 1
(i
1 2 2 1
-2 1 1 2
1 1 1 -1
-1 1 1 I
-1 1 1 I
1 1 1 -1
For a rectangular element the parameter 17 becomes zero and Eqs.
(IV.8) and (IV.9) reduce to the stiffness matrices given in Ref. 1.
Since wc did not quote in any other Appendix the stiffnesses of
stringers or flanges in the cover as given in Ref. 1, it may be appropriate
if for completeness it is done here. Assuming that the flanges run along
thej-gridlines and retaining the representation of Eq. (IV.8), we obtain
for the component matrices
k —k mM
= Kx = 0
and XX xu
r B. (i 0 -B,
/-; 0 fi. B2 n
(IV.12)
/ 0 B. B2 i)
L-*i 0 0 B.
APPENDIX V
On the Buckling and Large Displacement Theory of Elastic Systems
INTRODUCTION
We amplify in this last Appendix some aspects of the non-linear dis-

placement theory developed in Section D. In particular, we present
151
J . H . A R G Y R I S , S. KELSEY a n d H . KAMEI.
the basic information for the application of our method to elastic

systems consisting of beam elements under bending and end load. For
simplicity of presentation all loads are assumed to act at the nodal
points of the idealized system but the analysis may easily be extended
to account for intermediate forces by simulating their effect with
fictitious initial stresses J , as demonstrated in Ref. (4) for the dual
case of the force method with the device of initial strains H . The
specification ofthe elastic stiffness of an clement is based as on pp. 109,
119 on the natural elastic modes of deformation which exclude the
superfluous rigid body motions.
At this state it may be helpful for the understanding of the reader
if we note that the two alternative formulations of the initial buckling
problem developed in C.f and D.d are fully equivalent. In fact, the
Ersatz stiffnesses k, H , k ( / / introduced in the force method are seen to
be merely a transcription of the geometrical stiffness K „ put forward
in the displacement method. Furthermore, we confirm by premulti-
plication of Eq. (D.77) with K K l = F the equivalence of this formula
with Eq. (C.81). We have used both approaches with considerable
success not only in hinged and stiffjointed frameworks but also in plates
of a wide variation of shapes. The last application necessarily requires
the specification of the bending stiffness of a plate-like clement. The
reader may consult Refs. (15), (16) for a more detailed discussion of
these topics. We limit our current account—but for an explanatory
interpolation on triangular elements—to a consideration of two-
dimensional frameworks. Unavoidably the following paragraphs are
restricted to an outline of the techniques.
(a) Stiffjointed Frameworks; Main Assumptions

The application of Eq. (C.81) or (D.77) and (C.80) or (D.l la) to
investigations on buckling and large displacements of hinged frame-
works is straightforward. However, as far as stiffjointed systems are
concerned we require a slight amplification of the theory given in
D.b and D.c.
Figure V.l shows a typical stiffjointed assembly of bars or beams for
which we want to determine a possible onset of buckling and/or its
large displacement behaviour. T o analyse such a system—which may
also contain curved bars—we first idealize it into an assembly of straight
segments or elements, the length of which is chosen to be so small
that the magnitude of the gradient tp' remains limited and, in par-
ticular, its square negligible in comparison to unity. Then extreme
variations ofthe curvature of an element are excluded and we may use
the standard formula for the bending curvature of a beam (actually
the assumption is only required when investigating large displace-
ments). We also neglect here transverse shear deformations although
our theory may easily be extended to include their effect by the
152
adoption of the procedures given in B.b and Appendix I I I . We

assume, furthermore, that the chosen length of a segment is so small
that the contribution to the bending moment arising from the end
load {—Nd) may be ignored (see Fig. V.2). This is equivalent to
assuming that the instability load of the independent segment is much
CV x.u
Fig. V . l . Arbitrary stiff jointed framework.
S.-CLA-'.- ^ ^
• y
HP
6 —*-»„
Fig. V.2. Large deformation of element (1,2) in stiff jointed framework.
higher than the actual end load. Amplifying this statement our
technique may be said to presume that an element at the state (0) of
every incremental step of the large displacement analysis is straight
(see D.a). Hence, the idealization o f t h e element effectively involves
its representation by two torsion springs at the end station, deforming
under the corresponding bending moments, and a direct spring,
deforming under the end load (see Fig. V.2a). It is, of course, possible
to allow for a variation of the spring stiffnesses through the loading
history ofthe system (e.g. due to the effect of temperature on the elastic
moduli).
153
It follows from our above assumptions that the rule of linear super-
position in the moving co-ordinate system 0£//£ (Fig. V.2)—introduced
in the large displacement theory under D.c for the formation of the
invariant vector C* of the force increments in the directions of the
natural stiffnesses—is also applicable here. To simplify the argument
—without in any way restricting the applicability of the theory—we
assume that all bending displacements are taking place in a principal
plane of the beam elements.
(b) The Natural Elastic Stiffness of an Element: A Postscript to the Triangle

T h e natural stiffness k, of a flange under a constant end load given
in D.b
kf = B f (V.l)
has to be extended here to include the effect of bending moments.
Since shear dcformability is neglected our natural force and associated
displacement vectors become of order (3 X 1). The simplest and most
elegant formulation is obtained by the specification of the applied
forces and corresponding modes shown in Figs. V.3 and V.4 for a
typical element (1,2),
?«.,-{# M, Ma}r,2 (V.2)
Pm,. — {« V, vJi.2 (V.3)
where the suffices s and a refer to symmetrical and antisymmetrical

bending mode distributions respectively. T h e relation connecting
the two vectors is
Til.i km.2p/n,2 (V.4)
where k / n 2 is the (3 X 3) natural stiffness matrix of the element
(1,2). For a uniform element, whose direct and bending stiffnesses per
unit length are EB and Ef a simple argument yields for k w the diagonal
matrix,
k«.. = r-" 5 1 3/Ji2 (V,5)

*12
For a system with elastic properties varying axially it is, in general,

permissible to substitute an idealized assembly of uniform elements, as
long as the chosen length / is sufficiently small. Of course, it is possible
to establish the natural stiffness for non-uniform elements. To illustrate
the point let us consider the case when the pertinent flexibilities cj>s
and tftf,t per unit length vary linearly between the two end stations (1)
and (2).
154
N„ - • i 2 N,. . 1
Normal Force N = 1
M, - 1 M, H
Symmetrical Bending Moment M. =1
M, s i M2 =1
. • :
Antisymmetncal Bending Moment M. =1
Fig. V . 3 . Unit natural forces for a beam element (1,2).
u, •= 1/2 , 2 «, ' " 2

• • • * • -
Unit Elongation u 2] z 1
• , ' 1/2 •,=-1/2
Symmetrical End Rotation iji. = 1
• , =1/2
"*, =1/2
Antisymmetrical End Rotation t|i, =1
Fig. V.4. Unit natural deformations for a beam element (1,2).
155
Wc form the expressions

1 J_
<f>N = *.>
— = ™ ~ ^Vm + X'<f>.\d
"" EB
(V.6)
, 1 1 ,
VjU = = IF? = YMm + X 9.,Jd
where
<f*.vm — {<f>M + 4>M)I^, .\d = <A.\2 — -AM
(V.7)
'I'M,,, = {<f>Ml + 9.Mi)l^> -A.Ud = <rV./2 — -r^Wl
#' is the non-dimensional co-ordinate,
X = [x' ~ l f)ll„ (V.8)

Suffices (I) and (2) indicate values at the respective end stations.
Using the unit load method of Ref. (1) in the generalized form
given in Ref. (4) the bending flexibility matrix fM may be written as
(suffix (1,2) is omitted),
f
» " C t I -2y'] [ *«" + *'fl'"l[1 ~2x'] dx'
—^Md
6 (V.9)
L — <ft Md '^•9 Mm
The (3 x 3) stiffness matrix may next be written in the partitioned

form
k B = r * . v k.ul (V.10)
where
kN = m N m (v.ii)
and the stiffness k w is obtained by inversion of Eq. (V.9) which yields
6
\^9Mm 4>Md
k „ = l{^4>Mm (V.l 2)
— 4>lld) \-<f>Md
i/</) SJJ>M„ ty.,1
Having derived the natural stiffness matrix k w of a beam element we
set up the stiffness matrix W\K' in the coordinate system Ox'y'z' referred
to the complete displacement and force vectors. Following the nota-
tion of Appendix I and Fig. V.2 we have
V' = {N[2 Si, Mr N 21 521 M2}

., r r I r r • (*•••*>'
P = {«i t>i Vi "2 v2 tp2j
where
N> 2 = - N 2 l and S[ 2 = - S 2 l (V.l 4)
156
The relations between the natural and complete vectors may be put
in the standard dual form,
p w = Sip' and P' = «'P/; (V.l 5)
Here 81 is easily seen to be the (3 X 6) matrix
1 (1 0 1 0 0
'i\ II 0 1 () 0 -1 (V.16)
(1 2/1 1 (i -2// 1
The stiffness matrix k^, is now given by,
fci = M'k„«i (V.17)
which is valid for any type of beam element. For any specified matrix
k ^ we can obtain from Eq. (V.17) the explicit form of the (6 X 6)
matrix W'K. It is, however, more convenient from the computational
point of view to carry out the congruent transformation of Eq. (V. 17)
on the digital machine itself; (see below).
Since it is necessary to set up ultimately the stiffness matrix K,. or
K.h. for the displacements, u, v, tp measured in the basic co-ordinate
system Oxyz (Fig. V.2), we must determine the matrix fcf; for each
element. This is achieved by the simple transformation
(^W'k^lG,,, (V.l 8)
where ii l l e is the (6 X 6) matrix
(V.19)
I let-l-
(V.20)
•••-Ba
and
t£
t cos {x',x)
cos {y',x)
(see also Eq. (1.3) for the three-dimensional case).
cos {x',y)
cos {y',y).
(V.21)
In general, it is preferable to apply for the force and displacement

vectors the sequential order,
P = [Nu N2l Su S 2l M1 M2}
(V.22)
P = ("l "2 "l l-i Vl V-2 )
We have,
P - Tp (V.23)
where T is given by Eq. (D.32).
The corresponding stiffness matrix kA; is now
k, - T'k f c T (V.24)
157
and the matrix K ^ of the assembled structure may be obtained from

Eq. (B.25) which reads here,
KK = a ' k , a (V.25)
Before concluding the discussion on the elastic stiffness of the beam it
is instructive to check Eq. (V.17) for a uniform beam element. In
this case it is preferable to base the stiffness not on p of Eq. (12) but
on the sequential order
p' = W «J »i »i w'i Vt) (V.26)
(Note that relation (23) applies also to the local system O'x'y'z'.)
Introducing then the appropriate congruent transformation we obtain
B -B
O
B
E 12///2 -12///2 6/// 6///
K (V.27)
I symmetric 12///2 -6/// -6///
4/ 2/
4/
which checks with the expression given in Ref. (1). Alternatively k^.
follows from Eq. (V.24) which becomes here
k; = T k ; T (V.27a)
In many cases it is possible—and, in fact, often computationally
advisable—to assume an infinite direct stiffness for the elements. As
a result there arise certain rigid links which significantly reduce the
number of kinematic freedoms. Lack of space precludes the discussion
of this intrinsically simple topic.
Note on the natural stiffness of the triangle.

The derivation of the natural stiffness k T of a triangle established
in D.c is possibly too concise and a few explanatory remarks may be
helpful. To this purpose we indicate in Fig. V.5a the specification of
the stiffnesses in the directions of the sides of the triangle. In Fig.
V.5b wc illustrate a typical unit elastic mode of deformation. The
last diagram (c) should assist in the understanding of equilibrium
Eq. (D.67) and shows the triangle (1,2,3) under the action of a typical
component stress n v i c . If we consider the corresponding statically
equivalent forces al the nodal points we observe that the resultant
reaction vanishes at the corner (3). On the other hand we obtain for
nodal points (1) and (2) the loads
p« =
.p aga * (V>28)
lu
158
(a)
Definition of Natural
Stillnesses lor a Triangle
(c)
Equilibrium Condition
lor a Triangle
Fig. V.5. Derivation of natural stiffness for a triangle.
Application in turn of stresses ir2.,r and a 3 U yields then the matrix

Eq. (D.67).
(c) Large Displacement Analysis; Buckling

Following the general theory developed under D.a the large dis-
placement analysis of stiffjoinled systems may be based on the linear
incremental formula (see Eq. (D.l la))
R A = \K M + K,,]r A + a'J A
= K r A + a'J A (V.29)
To apply Eq. (V.29) we only require the instantaneous geometrical

stiffness K, ; . In view of our assumptions on the effective straightness
of a beam clement at the initial state (0) of each incremental step the
geometrical stiffness of an element is solely dependent on the end load
N 0 and shear force S0 and may hence be derived by the formulae ofthe
type of Eq. (D.25) developed in D.b for a flange under a constant end
load. It goes without saying that JV0 is measured along the line con-
necting the C.G.'s ofthe two end stations ofthe segment (see Fig. V.2).
Note that in the present case N 0 and S0 are the only entries ofthe ( 3 x 1 )
invariant vector C* required for the formation of the geometrical
159
J . H . ARGYR1S, S. KELSEY a n d H . KAMEL
stiffness. (If we refine, however, the theory by allowing for the effect
of truly large curvatures this statement must be qualified.) To obtain
N 0 or S 0 from the incremental N± or 5 A of the preceding loading steps
we merely have to introduce a suitable adaptation of the summation
formula in the moving co-ordinate system 0£rjt, given in Eq. (D.52)
for a flange. This completes our all too brief apercju for the practical
application of Eq. (V.29) to stiffjointed frameworks.
The limiting case of initial instability is straightforward, since we
have all the data to set up the governing equation (D.42)
|KA; + /K„|r = O Vpn
where A is a suitable measure of the loading intensity.
When seeking the low est buckling load it is convenient to prcmultiply
Eq. (V.30) with K K ' = F and to use the relation
FK,.,.. . « , (V.31
Application of the vector iteration procedure yields the lowest value

of the buckling intensity factor ), and the corresponding cigenmode r .
For a broader outline of the technique and possible alternative methods
the reader may consult Refs. (15), (16).
(d) Two Examples

We first illustrate application of Eq. (V.31) on the simple strut
problem of Fig. V.6 for which the exact solution is given by Timoshenko
X
0.1 I
EI.
(a)
0.1 I
0.41
11
" ' " • • '
,
EL 0.05 I
(b)
0.05 1
T T,
l-'ig. V.6. Buckling of stmt with discontinuous bending stiffness.
160
in Theory of Elastic Stability, 2nd edition, p. 115. The basic matrices k A

and k,,. may be obtained either directly or from Eqs. (V.27) and
(D.25) respectively. Denoting here the length of the clement with d
and neglecting (as in Timoshenko) the direct compressibility of the
strut, we find
'12/d* -12///2 6/d 6/d' •'I
2
El 12/// -6/d -6/d V-i
(V.32)
d symmetric 4 2 Vi
1 Vt
and
' 1 -1 V,
O.,
Pit -1 1 V.
(V.33)
d V'i
O,
o. _ Vt
Application of Eq. (V.31) lo the two alternative idealizations shown

in the diagrams (a) and (b) of Fig. V.6 yields an excellent convergence
with increasing number of subdivisions. Thus, writing the buckling
load in the form
P
r K
Eh (V.34)
Ocr — /2
we obtain for the coefficient a the values given in Table V.l
Table VJ. Buckling of a strut; Effect of idealization
Number of Coefficient
elements a
10 4-28
20 4-23
4-22
(Timoshenko)
As a final example we consider the buckling and large displacement

behaviour of the pin-jointed framework shown in Fig. V.7. As far as
instability under the pair of end loads R is concerned the analysis is
straightforward using formulae (31), (D.25), (D.26). Fig. V.8 repro-
duces the first buckling load Rb and the corresponding eigenmode. In
Fig. V.9 we indicate an alternative loading case which consists of
161
J. . HARGYRIS, S. KELSEY and H. KAMEL
• *, n n n A
/ Araat ol
M^>t\ y\\ S \ /f\ / x^ Mambara (in*)
A, I
*. y<C*> j><C y\. V*C ^C 'f. jXT i.
A, A, *, 1/4
A, 1/)
— 4—1—4—. *. l/l
f a 10' l b / I n '
. 6x4
Fig. V.7. Geometry of framework.
<?, a «5 960 lb
Fig. V . 8 . Buckling of framework.
859 60 Ib •59 60 1b
Xri 7H
Fig. V . 9 . Za-jf-j»c* displacement analysis of framework.
constant transverse forces RJ100 at the upper nodal points and a

varying end load P applied to the upper member. As P increases there
arises a typical large deflexion problem for which we applied the
incremental technique of Eq. (V.29). Its programming proved
extremely simple and the resulting deflexion r versus load P is logarith-
mically plotted in Fig. V. 10. Note the pronounced non-linearity of
the r-jP relationship and, in particular, the steep rise of r as P approaches
the value of 2Rb of the buckling case in Fig. V.8.
Pft
0 50.10' 100.10' ISOiK)' 2«, JOOalO'
Fig. V.10. Tip deflexion r for varying end load P ; r„ deflexion for P = 0.
162
Acknowledgments
The authors wish to express their appreciation to the considerable
number of collaborators who assisted in the preparation of the paper.
In particular, Mr. Patton, M . S c , is to be mentioned for his inspired
help in programming the matrix assembly technique given in B.C.
Herr Pretsch contributed to the spar data reproduced in Figs. B.9,
10, 11, and Hcrr Dcbatin solved the buckling and large displacement
examples in Appendix V. Frau Rics and Miss Pook are to be com-
mended for sharing in the arduous task of typing the manuscript.
Finally, Miss Niedermeier is responsible for most of the drawings.
REFERENCES
1. AROYRIS, J . H . Energy theorems a n d structural analysis. Hart I, General theory.
Aircraft Engineering, 26, O c t . p . 137, Nov. p . 383, 1954; 2 7 , F e b . p . 42, M a r c h
p . 80, April p . 125, M a y p . 145, 1955. Sec also Energy Theorems and Structural
Analysis. Butterworth's Scientific Publications, London, 1960.
2. ARGYRIS, J . H . Recent developments of matrix theory of structures. Paper
presented at the 10th meeting o f t h e Structures and Materials Panel; Structures
G r o u p , A G A R D , Aachen, Sept. 1959.
3. ARGYRIS, J . H . and KEI.SEY, S. T h e analysis of fuselages of arbitrary cross-section
and taper. Aircraft Engineering, 3 1 , M a r c h p . 62, April p. 101, M a y p . 133,
J u n e p . 169, J u l y p . 192, August p . 224, Sept. p. 272, 1959; 3 3 , F e b . p . 34,
M a r c h p . 71, J u n e p. 164, J u l y p . 193, August p . 227, 1961.
4. ARGYRIS, J . H . and KELSEY, S. Modern Euselage Analysis and the Elastic .Aircraft.
Bulterworth's Scientific Publications, London, 1962/63.
5. T U R N E R , M . J . , CLOLIGH, R. J . , M A R T I N , H . C. a n d TOPP, L.J. Stiffness and
deflexion analysis of complex structures. Journal of the Aeronautical Sciences, 2 3 ,
Sept. 1956.
6. MF.LOSH, R. J . and M E R R I T T , H . G. Evaluation of spar matrices for stiffness
analysis. Journal oj the Aeronautical Sciences, 26, Sept. 1959.
7. G A L L A G H E R , R. H . and RATTINGER, I. T h e experimental and theoretical
determination ofthe elastic characteristics of modern airframes. Paper presented
at the 12th meeting of the Structures and Materials Panel, A G A R D , Athens,
Greece, Sept. I960.
8. ARGYRIS, J . H . Fifth Research Report to the Boeing Go. Transport Division.
Institut fur Statik und Dynamik d e r I-lugkonslmklionen Stuttgart, M a y 1962.
9. DENKE, P. H . A matrix method of structural analysis. Proceedings o f t h e 2nd
U.S. National Congress of Applied Mechanics, A S M E , New York, 1954.
10. T U R N E R , M . J . T h e direct stiffness method of structural analysis. Paper pre-
sented at the 10th meeting of the Structures and Materials Panel, A G A R D ,
Aachen, Sept. 1959.
11. ARGYRIS, J . H . and KELSEY, S. T h e matrix force method of structural analysis
and some new applications. A . R . C . ; R. & M . No. 3034, Feb. 1956.
163
12. ARGYRIS, J. H . Die Malrizentheorie der Statik. Ingenieur Archiv, 2 5 , 174,

1957.
13. D E N K E , P. H . T h e matrix solution of certain nonlinear problems in structural
analysis. Journal of the Aeronautical Sciences, 2 3 , No. 3, M a r c h 1956.
14. KOCHANSKI, S. L. and ARGYRIS, J . H . Some effects of kinetic heating on the
stiffness of thin wings, I and I I . Analysis of flexural and torsional slilfnessess for
large deformations of thin solid wings. Aircraft Engineering, October 1957,
February, M a r c h , April 1958.
15. ARGYRIS, J . H. Recent Advances in Matrix Methods of Structural Analysis; Progress
in Aeronautical Sciences, 6. Pergamon Press, 1963 (in Press).
16. ARGYRIS, J . H . lectures on Matrix Theory of Structures. Oslo University Press,
1963 (in Press).
164
UPPER AND LOWER BOUNDS IN MATRIX
STRUCTURAL ANALYSIS
B. M. FRAEIJS DE VEUBEKE
Universities of Liege ami Louvain, Belgium
CONTENTS
Par,
1. INTRODUCTION . . . . . . . . . 166
2. DUAL TREATMENT OF THE TRIANGULAR PANEL . . 167

2.1. Uniform Field . 167
2.2. Compatible Net of Triangular Fields . 168
2.3. Equilibrium Net of Triangular Fields . . 170
3. T H E STIFFNESS M E T H O D OF RESOLUTION . . . .171
4. T H E QUADRILATERAL PANEL SUBDIVIDED BY DIAGONALS . 175
5. T H E FORCE M E T H O D OF RESOLUTION . . . .177
APPENDIX A
Principles for the Determination of Upper and Lower Bounds
to Influence Coefficients . . . . . .180
1. Linearized Theory . . . . . . .180
2. Scalar Product of Two Fields 182
3. Potential Energy and Complementary Potential Energy of
the External Loads . . . . . . .182
4. Principle of Minimum Total Energy . . . .184
5. Principle of Minimum Complementary Energy . .185
6. Clapeyron's Exterior Theorem . . .185
7. Upper and Lower Bounds to Direct Influence Coefficients . 185
8. Upper and Lower Bounds to Cross Influence Coefficients . 186
165
Pag,
APPENDIX B
Examples and Numerical Results . . . . . 188
1. Structural Model 188
2. Methods of Analysis . . . . . . . 189
3. Reduction of Self-Straining Conditions from Symmetry Con-
siderations . . . . . . . .192
1. I N T R O D U C T I O N
The numerical calculation of static influence coefficients for a complex

structure is necessarily based on simplifying assumptions. The results
of such calculations are of value if the accuracy of the approximation
is known. Within some measure of probability a quantitative estimate
of the accuracy can be gained through previous experience with a
similar structure. A theoretically more satisfactory situation arises if
the calculation procedures are such that both upper and lower bounds
are obtained for the coefficients. The maximum possible errors are
then known quantitatively and no further investigation is needed
should they lie inside the margin acceptable to the designer. Should
this margin be exceeded, further refinements are required in the
structural idealization.
The analytical procedures leading to upper and lower bounds are
well established for the small deflexions case. It is not mandatory,
though preferable for obvious reasons of economy, that the two pro-
cedures be applicable to the same idealization of the given structure.
For the sake of completeness a short account of these procedures is
given in the Appendix A of the present paper. Their chief character-
istic is that they should be " p u r e " . By " p u r e " is meant that no
assumptions are allowed that would violate both the compatibility
conditions and the equilibrium equations.
The approach yielding lower bounds must be purely compatible
and, to this purpose, built on a continuous single-valued field of small
displacements. The approach yielding upper bounds must satisfy
everywhere the equilibrium conditions.
O n e must here distinguish carefully between
(1) the structural decomposition into elements: plates, beams . . .

assumed to behave in accordance with simplified displacement
or stress fields,
(2) the assumptions involved in the procedures for connecting the
elements together, especially with regard to continuity of the
displacements or stresses,
(3) the procedures adopted for solving the redundancies.
166
U P P E R AND LOWER BOUNDS IN MATRIX S T R U C T U R A L ANALYSIS
The terminology "displacement method" and "force method" used

in the literature too often refers only to the last aspect of the overall
problem and is not entirely relevant to the distinction made here
between a "purely compatible approach" and "a purely equilibrium
approach".
A displacement method of solution, or stiffness method, can be used
to solve the connexion problem between elements and still violate some
continuity of displacement between panels. A force method of solution
can be applied and still violate local equilibrium conditions.
The important consideration to apply, if one wishes to be certain
that the coefficients are upper or lower bounds, is to ensure that the
geometry of the elements, their assumed displacement or stress field
and the geometry of their connexions produce either a purely com-
patible overall field or a pure overall equilibrium field. Once this is
ascertained it will be seen that, in either case, both the force method
of resolution or the stiffness method are generally applicable, though
perhaps with different degrees of efficiency.
2. DUAL T R E A T M E N T OF THE T R I A N G U L A R PANEL
The subdivision of panels in triangular elements is an attractive pro-
position for complicated geometries (taper, sweepback, skew ribs . . .).
The triangular panel is also the natural shape for the purely compatible
or purely equilibrium connexions between uniform stress-strain fields.
2.1. Uniform Field

A linear two-dimensional displacement field
u = a + xem + X h ' m — °>) ( 2 l ]

v = b + x { \ y „ -f- w) + y E y
depends on six parameters: two translations a,b, a rotation co account-
ing for the rigid body displacements, and three uniform strain com-
ponents {Ex, Yxy, e,).
The corresponding uniform stress field is given by
(2.2)
with a stress-strain matrix M, which for isotropic panels is

/I v 0
167
B. M. FRAEIJS D E V E U B E K E
The reciprocal relations are

1 0 V
Iff,, I 1 0 1, (2.4)
\ y xy (1 0 2(1 + v)f (2.5)
Retaining the assumption that there are no body forces, the uniform
stress field satisfies the internal equilibrium equations [see Eqs. (A.6)].
So, with regard to its own interior domain, the uniform stress-strain
field is both purely compatible and a purely equilibrium field.
The duality between compatibility and equilibrium arises solely
when such uniform fields are edged and pieced together.
2.2. Compatible Net of Triangular Fields

Along any straight boundary line both the « and v components vary
linearly and are entirely determined by the values taken in two
different points. Consequently, if two linear displacement fields of
type (2.1) arc made to coincide in two points, they are coincident along
the whole straight line joining these points. This property leads to the
construction of a compatible net of triangular fields by enforcing the
coincidence of the displacements at each common vertex. A natural
requirement for such a construction is then that the parameters of each
field be expressed in terms of the six displacement components at the
vertices. Let the vertices be numbered {p,q,r) in a counter-clockwise
sense. Then
1
*, y v
A = 1
x, y , •y rp x Qr - xTvy„ (2.6)
x
i r yr
is equal to twice the area of the triangle, where
x pq Xp XQ etc.
An elementary calculation yields for the strain components
Wxyl
where N is the rectangular (3 x 6) matrix
/ Ar yr, y*« o o o \
N = 0 0 0 -Xm, -xrv -x„ (2.8)
X X X
\ ar rp JKI y<ir /rj Jiml
and u the column matrix of displacements, whose transpose or row

matrix will be written
«' = Op U, U. vp va vr) (2.9)
168
U P P E R AND L O W E R BOUNDS IN MATRIX S T R U C T U R A L ANALYSIS
The strain energy of the panel of uniform thickness t is conveniently

obtained from Clapeyron's theorem (see Eq. (A.5) for the energy
density)
A
rr • ,
Y
2 2^' *"
Substitution of Eqs. (2.2) and (2.7) gives
U = J - u'N'MNu = \u'Ku (2.10)

4/1
with K — {t/2A)N'MN a symmetrical (6 X 6) matrix (2.11)
The elements of the u matrix are to be considered as generalized

coordinates for the field. The corresponding generalized forces (or the
so-called corner forces) are obtained from Castigliano's formulae. If
/ d e n o t e s their column matrix, the variational procedure equivalent to
Castigliano's theorem gives
SU = l-du'Ku + \u'Kf>u = du'Ku == bu'f
Hence
f=Ku (2.12)
and A' is the so-called "stiffness matrix". When formula (2.11) is

expanded the identity of A' with the forms given by Turner or Argyris
can easily be checked. The elements of t h e / m a t r i x will be denoted
as follows
/ ' = (//„ //„ H t V, VQ Vr)
The relations between stresses and displacements follows from Eqs.
(2.2) and (2.7)
(a* .
= \MNu (2.13)
.1
VT.
and from this can be derived the equation
/ t
f=Ku = 2 N'MNu = - iV'j ay | ( 2 --4)
yielding an interpretation of the generalized forces. For example, it is

found that
/
T
Hp — 9 O-JW + xv X rv)
169
B . M. F R A E I J S DE V E U B E K E
It can then be stated 5 that each generalized force at a vertex is equal

to half the resultant of stresses acting on the opposite edge or to the
resultant of stresses acting on half of the adjacent edges (Fig. 1).
HP
Fig. 1
Obviously a compatible net of triangular fields is not an equilibrium

field. The equilibrium equations between adjacent fields are of a
global character; they state that the vectorial sum of all generalized
loads meeting in a common vertex is equal to the external load applied
there. They do not prevent the stresses to vary discontinuously across
a common edge.
2.3. Equilibrium Net of Triangular Fields

In this case it is the continuous transmission of stresses that is
required and this can only be achieved by relaxing compatibility
requirements. The equilibrium requirements are met by taking as
generalized displacement coordinates the displacements at the middle
of each edge of the triangle. The mid-points define the vertices of a
"skeleton triangle" (hatched in Fig. 2), whose area is four times smaller
than that ofthe real panel:
\ A = y r p x v r - xrvyQT (2.15)
U P P E R AND L O W E R BOUNDS IN MATRIX S T R U C T U R A L ANALYSIS
Otherwise the theory of par. 2.2 applies without modifications and,

keeping the definition (2.8) of the A' matrix, one obtains the stiffness
matrix
K*=^N'MN (2.16)
- - Nu and / 2/A" j a, |
From this last result follows that each generalized load applied to a
mid-point is the resultant of the stresses acting along the corresponding
side (Fig. 2).
Whereas in the compatible net a vertex or node is common to several
fields, a node of an equilibrium net is only common to two panels.
The global equilibrium condition is, barring external loading, equiva-
lent to stating that the generalized forces at the node are equal and
opposite (Fig. 2). In view of the fact that the stresses are constant
along each edge this implies that the stresses are transmitted continu-
ously as was required. Any external load, represented by a generalized
resultant applied at a node, will really have to consist of a uniformly
distributed load along the edge.
If local equilibrium is satisfied in an equilibrium net, compatibility
is now generally violated for two reasons:
(1) the strain along a common edge can have different values in
each panel;
(2) the orientation of the edge can be different.
This possibility of rotation of one panel with respect to the adjacent
one leads to peculiar difficulties to be discussed later.
3. THE STIFFNESS M E T H O D O F R E S O L U T I O N
The essential steps of the stiffness method of resolution are briefly
recalled:
(1) Addition of the stiffnesses ofthe individual elements in a stiffness
matrix for the complete structure. This step expresses the single-value
of the displacement vector at a node where the elements are assembled
and the additivity of the corresponding internal generalized loads,
balancing the external load;
(2) Elimination of externally unloaded nodes, where deflexions are
not to be determined. Suppose Eq. (2.12) has been extended by step 1
to represent the relations between the external loads m a t r i x / a n d the
displacement matrix u for the complete structure. If the ith element
o f / i s zero, we have
I AP>«, - 0
i
171
B. M. F R A E I J S DE V E U B E K E
This equation is used to express u, as a linear combination ofthe other

displacements. The j t h column of the original stiffness matrix is
thereby modified by the addition of the ith column multiplied by the
factor —AP/A,. T h e ,'th column itself becomes a column of zeros
and is deleted together with the ith row from which Eq. (2.18) was
derived.
Steps 1 and 2 can be taken progressively as an intertwined process
of growth and reduction until the complete stiffness matrix A ( / ) (free-
structure matrix) is obtained for the loaded nodes.
(3) Elimination of all rigid body modes to produce a final non-
singular stiffness matrix A U ) (supported structure matrix). This step
consists in prescribing certain displacements to be zero in order that
the structure be at least isostatically, or even hyperstatically, supported.
Rows corresponding to the suppressed displacements are deleted in the
equation / = K, f) u. They are used afterwards to know the reaction
loads due to the remaining independent loads.
(4) Inversion of the non-singular matrix A,,, to produce the matrix
of influence coefficients.
No difficulties are encountered in implementing those steps in the
case of compatible triangular fields associated to beam elements or,
more generally, in the case of fields where the displacements are defined
at the corners. The treatment of equilibrium fields is by no means
restricted to the use of the dual "force" method and there are some
obvious advantages to be gained from attempting to solve them also
by the stiffness procedure. T h e same basic programme can be used in
the computer and the sometimes delicate choice of convenient self-
strainings is avoided. To achieve this, however, some care must be
exercised with respect to the addition of beam elements and the
elimination of rigid-body modes:
(a) Addition of a beam element (Fig. 3). The total load g„ applied by
the fields (and possibly by an external source) to the isolated beam
9p
Fig. 3
element is in the form of a uniform shear flow g v /c v . There are also

end loads g m and g„ with conventional positive sense identical to the
sense chosen for the end displacements vm and vn. T h e internal tension
172
UPPER AND LOWER BOUNDS IN MATRIX STRUCTURAL ANALYSIS
T in a cross-section positioned by the non-dimensional coordinate

—-J < £ < £ is to be calculated by equilibrium considerations
T = - g m - fa - tg, (3.1)
The (complementary) strain-energy of the beam, assumed here for
simplicity to have a uniform cross-sectional area, is
8
2ES Urn h k,) + 12
TKA (3.2)
2ESJ V
A simple application of Castigliano's formulae yields
dU c
vm ~ v„ = = ^ , (fc + 1*,) (3.3)
as relative displacement of the end-sections, and
P
» ~ v " = i)a =
2ES ( gm ' t_ * g J (3-4)
It should be noted that •„ is not the displacement of the mid-span

cross-section; it is a generalized displacement corresponding to the
uniform shear flow. It is equal to the ordinary average of all cross-
section displacements as calculated by integration of the strain. A
best fit in compatibility is reached when this vv is identified with that
belonging to the field edges which are loading the beam. In that
respect it would have been more appropriate, when discussing the
equilibrium nets of triangular fields, to define the displacements as
averages along each edge. However, since the field is uniform the
average is identical with the displacement ofthe middle ofthe edge.
While the addition of a beam segment to a compatible net introduces
no additional nodes, in the present case the end-sections of the beam
are new points where displacements must be introduced and where
concentrated external loads can be applied. To obtain the component-
stiffness matrix due to the beam, Eqs. (3.3) and (3.4) are solved for g m
and g v ; the result substituted in Eq. (3.2) yields the strain-energy
2ES
U = {v2m + vl + 3v% + vmv„ - 3vpvm - S t v O (3-5)
The other Castigliano's formulae give the load-displacement relations

expressed by means of the required stiffness matrix in the form
(3.6)
These equations arc also directly obtainable from Eqs. (3.3) and (3.4)
together with (3.1) for £ = \ and T = g n . It should be noted that at
the point where beam end sections meet together equilibrium will
automatically be satisfied by the stiffness addition process.
(b) Elimination of rigid body modes. T h e possibility of relative panel
rotation already mentioned is generally checked by the presence of
connected beam segments. T h e case can, however, arise of structural
deformation modes due to panel rotations without strain energy. They
will be called rigid deformation modes and are best determined by
inspection of the kinematics of the skeleton triangles.
Two cases must be distinguished. In the first, passing from the free
to the supported structure, the nature of the support prevents the rigid
deformation modes as well as the rigid body displacements and no
special treatment is required.
In the second case some rigid deformation modes remain even after
the structure is supported. They are characterized by non-zero dis-
placement vectors satisfying the matrix equations
K U) w (r) = 0 r = 1.2. . ./
The theory of linear equations systems gives as necessary and sufficient
conditions for s o l v i n g / = K U) u that the loading /" verifies the virtual
work equations
w'lr)f=0 (3.7)
Suppose now that the rigid deformation modes are just suppressed by
preventing the first t displacements (this will generally require a re-
numbering of the displacements and is only advocated for simplicity
of exposition). T h e first / components o f f and u are now isolated in
submatrices / ( 1 ) and u n ) , their complements denoted by f(2) and u (a)
and the matrix A,,, subdivided in four component blocks so that
.hi) — -MiMii» + A12«(.2)

==
fii) ^-21U(1) + ^*22U(2)
By making u n ) = 0 wc have, since K22 is by assumption non-singular
"(i) = 0 "(2) = K n f w (3-8)

= :
/<!) ^ l i ^ i i /(2) (3-9)
The last equation furnishes the reactions in the rigid links suppressing
the w(1) displacements as linear functions of the independent loads of
the / ( 2 ) matrix.
A unit load on the structure is now considered as a group formed by
a unit vector of thcy ( 2 ) matrix associated with the corresponding set
of/ ( 1 ) loads so that the reactions in the links disappear and the « (1)
displacements can be freed. Under this circumstance the general
174
expression for the displacements contains arbitrary contributions from

the rigid deformation modes
i
«(1) = I < W r l ( l )
"(2) — K i i f s i ) + Z, a r W lr)ii)
1
The analysis shows that the situation is unsatisfactory except for those
influence coefficients that would happen to be independent of the
arbitrary coefficients a r .
4. THE QUADRILATERAL PANEL SUB-DIVIDED BY DIAGONALS

A remarkable case where the relative rotations between panels docs not
affect the loading possibilities and constitutes a convenient new
building block is the quadrilateral panel sub-divided into four triangular
fields by the internal diagonals (Fig. 4). The four skeleton triangles
Fig. 4
articulated in points /, ./, A and L, form a kincmatically dcformable

chain. Yet the 2 x 4 load components applied on the outside vertices
E, F, G and H arc only restricted by three conditions expressing their
statical equivalence to zero (rigid-body equilibrium).
One proof consists in obtaining 2 X 4 - 3 = 5 independent states
of loading which can be transmitted through the kincmatical chain.
Four of them are obvious and consist in the equal and opposite force
pairs numbered from 1 to 4. The transmission is here based on the
fact that the sides of the skeleton triangles are aligned two by two to
form a skeleton parallelogram.
A fifth independent state of loading has been analysed for clarity
on Fig. 5. The pair of forces numbered 5 can be transmitted through
the sides /./ and KL, keeping each skeleton triangle in equilibrium,
provided the extensions of the lines /./, KL and HG intersect in a
175
B. M. F R A E I J S DE VEUBEKF.
Fig. 5
common point 0 . Suppose this to be true; then by Ceva's theorem

applied to the triangle FGH cul by the transversal KLO:
0 0 LF K H
1
(ill ' LG ' K F
Again from Ceva's theorem applied to the triangle GHE cut by the
transversal I J O :
GO J E I H
GH'JG'IE^
Whence a necessary and sufficient condition that 0 be a common
intersection is that we should have
LF KH J E III
LG ' KF JG ' IF
This, however, is true since L F = fH, LG = IE, K H = J E and
K F = JG.
Another way to look at this property of the quadrilateral is to prove
that the kinematical dcformability of the chain can take place with
each skeleton rotating about the outside vertex. The geometrical
proof is similar. There follows that no virtual work of the loads
applied to the outside vertices is ever involved and no other restrictions
placed on them than static equilibrium with respect to rigid body-
motion. For the purpose of building up the structure the skeleton
parallelogram of the quadrilateral can be considered as rigid.
Obviously a sixth state of loading is provided by a pair of opposite
forces along EF. This is, however, easily seen to be a linear combination
of the preceding states.
176
5. THE FORCE M E T H O D OF R E S O L U T I O N
A drawback of the stiffness method, as applied to equilibrium fields, is
a relatively large increase in the number of nodes requiring a corres-
pondingly large number of eliminations. Experience should tell
whether the accuracy of the computations is thereby seriously affected.
The opposite situation occurs when one wishes to apply the force
method of resolution simultaneously to compatible fields and equi-
librium fields. We shall here assimilate the force method to the comple-
mentary energy method and, since the prescribed displacements are
usually zero, it will further reduce to the minimum principle of the
strain energy expressed in terms of stresses.
The stress field is built up by adding a convenient particular equi-
librium field with a unit load, ignoring the compatibility restrictions,
and a complete independent set of self-strainings. The most convenient
self-strainings are those confined to a minimum number of elements.
In that respect the compatible triangular nets are very simply dealt
with. A complete set of independent self-strainings is obtained by
considering the interactions between pairs of triangles (Fig. 6) or
between a triangle and a beam segment (Fig. 7). The effect of such
lug. (i
Fig. 7
a self-straining on a triangular field is to produce a state of simple

traction in a direction parallel to the interacted edge. This is easily
verified by Eq. (2.14) with the relation (Fig. 8)
\thpc = F p
Turning to the cartesian axes of reference, we find
2Fm'~"
=
"' Wm
., \ e , f
2FJy„rV
th. : ( • : : ) '
2Fp
d>ii
( X
Cn
QT fqr\
Cm )
Adding the effect of the self-strainings F v and F r along the other edges
and substituting into
tA
U 2T
= ^ 2 E («S + °l + 'V - -M-V, - •&))
T h e following standard form of the strain energy is obtained
v _ - /(/•>, 4 - / ^ , + f i g , ) ' IFpF, FmF. /yFAj

Et\ A \CpC0 t-/ r csp/i
T h e use of this standard form in adding together all the energy contri-
butions implies that the particular equilibrium field be locally resolved
in a {Fp,FQ,Fr) system. In that case each force of this system is the sum
of a component from the particular equilibrium field and one hyper-
static unknown.
The beam energies are o f t h e type {cpN2)/{2ES) where the traction
load N is the sum of a component fixed by the particular equilibrium
state and unknowns from the interaction with adjacent fields.
The disadvantage of a large number of self-strainings is partially
offset by their simplicity and the possibility of eliminating them
gradually before the structure is completed. Indeed a self-straining
intensity X can be eliminated by a Mcnabrea type of equation
3U/8X = 0 as soon as the elements concerned by this self-straining
have been added to the structure.
The equilibrium net of triangular fields contains fewer self-strainings
but of a more complex character. One example is shown on Fig. 9.
As already observed 6 the setting up of the simplest independent self-
strainings by the computer itself represents a major step in the auto-
mation of the computations.
The state of stress in a triangle can this time be regarded as a super-
position of three F states (Fig. 10) similar to those of Fig. 10 used in
178
the compatible case. By virtue of Eq. (2.17) we have here th p a = F p .

Hence, with the exception of a numerical factor the strain energy of
the superposed fields is the same as before.
1 j{FpCp + F,c, + F r c r )
U =
\Et V <V« Vr C r Cp/l
To make use of this standard form, both the particular equilibrium

and the effects of the various self-strainings to which a triangle would
happen to be subjected to must be resolved in an F system. Note
incidentally that the partial derivatives of this expression would yield
the elongations of the skeleton triangle sides under the F forces. The
Fig. 9
strain energy of a beam segment as a function of loads was already

dealt with. The loads appear again as linear combinations of a term
due to the particular equilibrium condition considered and the self-
straining terms.
REFERENCES
1. T R E F F T Z , E. Ein gegenstuck zum Ritzschen v e r f a h m i . Proceedings of the 2nd
Int. Cong, for Appl. Mech., Zurich (1926).
2. WF.BF.R, C. Eingrenzung von verschiebungen und zerrungen mit hilfe der mini-
malsatze, Z . A . M . M . , 22, 130 (1942).
3. PRAOER, W. and SYNCE, J . L. Approximations in elasticity based on the concept
of function space, Quart. Appl. Math.. 5 (1947).
4. FRAEIJS DE VEUBEKE, B. Sur certaines inegalites fondamentales et leur generalisa-
tion dans la theorie des bornes superieures et inferieures en elasticite. Rev. Univer-
sale des Mines, X V I I , N o . 5 (1961).
5. T U R N E R , M . J . T h e direct stiffness method of structural analysis. Structures
a n d Materials Panel, A G A R D , Aachen, 1959.
6. DENKE, P. H . A general digital c o m p u t e r analysis of statically indeterminate
structures. Structures and Materials Panel, A G A R D , Aachen, 1959.
7. G A L L A G H E R . R. H . A correlation study of methods of matrix structural analysis.
O r a l progress report to the Structures and Materials Panel of A G A R D , Paris, 1961.
8. AROYRIS, J . H . and KEI.SEY, S. Note on the theory of aircraft structural analysis,
Zeitschrift fiir Flugwissenschaften, 7 heft 3 (1959).
APPENDIX A
Principles for the Determination of Upper and Lower Bounds to

fnfluence Coefficients*
1. L I N E A R I Z E D THEORY
The principles apply only to the fully linearized theory of elasticity.
This means that
(a) Rotations and strains are small enough to justify the linear
relationships
du dv du dv .. ,,
e A
«• = al v = -rT Yxv = T + a" -'
ex oy cy ex
between strain components {Ex,Ey,yxy) and displacement vector {u,v).
No distinction need be made between Lagrangean and Eulerian
coordinates: the cartesian coordinates {x,y) are either those of the
initial reference state or those of the deformed state.
(b) The strain energy density W {Ex,yxv,ev) relating stresses and
strains
d W
' " ' * --' ,A. n .
x = T = a (A 2)
° BEX » 8yr '=BE-y -
* For simplicity the two-dimensional case is considered. The extension to three dimensions
is trivial.
180
is taken to be a quadratic, positive definite, homogeneous form. The

stress-strain relations arc thus ofthe homogeneous linear type (general-
ized Hooke's law).
(c) The complementary energy density O {crx,Txy,ay), whose general
definition is through the Legendre contact transformation
0 = dx£x + TmrnYmm + Ct y E , - W (A.3)
is also a quadratic homogeneous form (in the stress components). This

is an obvious consequence ofthe assumption of a generalized Hooke's
law.
Differentiating Eq. (A.3)
/ dW\ J ( dW\ .
i i>W\ J
+ \av - . , . } «*•« + e x d(J
x + Yxv d T xy + Fv da
v
and using Eqs. (A.2) there follows
do t)«D a«D
Y {AA)
ta. » - ir... JK ta,
Equations (A.4) are dual to Eqs. (A.2); they state again the generalized
Hooke's law, resolved this time with respect to strains. Eulcr's theorem
on homogeneous quadratic forms is applicable both to W and to <l>.
In each case, by virtue of Eqs. (A.2) or (A.4), the same result is
obtained:
W = <t = \{a x E x + r xyYxy + ayEy) (A.5)
This is conveniently referred to as the local form of Clapeyron's

theorem: the energy densities are both numerically equal to half the
sum ofthe products of stresses by corresponding strains.
(d) The internal equilibrium equations
%+*£ + X--J ^ + - ^ + 7-0 (A.6)

ox dy dx oy
and the surface equilibrium equations
la x + mrxy == p lrxy + mcry = q (A.7)

are linear. Here {X,Y) and {p,q) denote cartesian components of body
forces and surface forces, while {l,m) arc the direction cosines of the
outward normal at the surface.
(e) As a consequence of the linear character of all equations the
principle of superposition applies.
181
B. M. F R A E I J S DE VEUBEKE
2. S C A L A R P R O D U C T O F T W O FIELDS
By "field" is implied an arbitrary field of stresses and corresponding
strains, related by the generalized Hooke's law. This field need not
satisfy cither the compatibility conditions Eqs. (A. 1) or the equilibrium
Fqs. (A.6) and (A.7). It should, however, be integrable in the sense
that the total strain energy:
!//.>••>• + TxvYxy + -V,) dS
should exist.
If two fields are distinguished by the subscripts A and B, consider
the following mixed expression
i A , B ) = SS{a*.Aem.B + TXV.AYXV.U + °y.Aev.ll) dS
(A-8)
The integrand can be transformed as follows:
,<l> Pl> <-!>
* — a x.A n~ V T xv.A fl r a
y,A .,
dCx.it dTxyl. da y B
The last member is clearly the bilinear form attached to <1) and is
equivalent to
,.|. ,.|. ( il>
a
' — x . n .. h T XV.B ^ I" C y . l l
dox.A dr xy , 6ayA
Hence we obtain the commutativity property
{A,B) = {B,A) (A.9)
which is recognized as a statement of the Betti-Rayleigh reciprocity
principle. On the other hand if a and fi denote scalar multipliers we
obviously have:
{*A,fiB) = *fi{A,B)
and also
{A,B + € ) = {A,C) + {B,C)
Expression (A.8) has the properties of a scalar product, it is conveni-
ently referred to as the scalar product of the two fields.
By virtue of Clapeyron's theorem the norm of a field
{A,A) > 0
is equal to twice the strain energy; it vanishes if and only if the field
itself vanishes.
3. I'on.vilAI. ENERGY AND C O M P L E M E N T A R Y PO I KN 1 I Al.

ENERGY OF T H E EXTERNAL LOADS
The following notations will prove to be useful by their concision:
Cwill denote a particular compatible field. This implies that the strains
can be derived from continuous single-valued displacement functions
182
{u,v) according to Eqs. (A.1) and that those displacements take

values
u = u v= v one, (A. 10)
prescribed on parts of the boundary.

/ / will denote a general homogeneous compatible field. Here the
values prescribed to the displacements on c2 are zero. That this
field is "general" means either that it is the most general one, or, if
approximate solutions are sought, that it contains unknown functions
or unknown parameters to be determined by the application of the
energy theorems.
According to these definitions and to the superposition principle the

general compatible field of a problem can be denoted by
C + H.
E will denote a particular equilibrium field. This implies that the

stresses satisfy the equilibrium Eqs. (A.6) and (A.7) on the comple-
mentary part c, of the boundary.
A will denote a general self-straining field. It is an equilibrium field
with the prescribed loads set equal to zero. It also contains unknown
functions or parameters.
According to these definitions the general equilibrium field of a problem

can be denoted bv
E f A
S will denote the exact field of the problem, or the solution. It is both
a compatible and an equilibrium field.
In a scalar product, if one field is compatible and the other an

equilibrium field, the value of the product is expressible in terms of
the displacements of the compatible field and the external loads with
which the other field is in equilibrium. Take the case of the product
{C,E). Since C is a compatible field, the product can be written as
If we now integrate by parts
{C,E) = {{lax,K + mr x y E )u c + ' h „ i S + may_K)vc] ds

Jc.-\ r,
183
B . M. F R A E I J S DE VEUBEKE
However, the stresses ofthe equilibrium field verify Eqs. (A.6) and also
Eqs. (A.7) on the part c, of the boundary, while the displacements of
C verify Eq. (A. 10) on c2. Hence
(CaaS) = ~ P t ~ Q, (A.ll)
where
X
Pe = ~ \ \ ^ + '**) dx dy - J [ {pu. + qve) ds
is the potential energy of the prescribed loads associated with the

displacements ofthe field C, and
Q e = —\ {*{ta,.s + W T . , I E ) + e{hmm.E + ma y h : )} ds
is the complementary potential energy of the reactions of field E

associated with the prescribed displacements. From this result, which
has a general character, onc finds immediately the following similar
properties:
{H,E) = - P h (A.12)
{C,A) = - Q Q (A.13)
{H,A) = 0 (A.14)
This last result is specially useful and can be stated as the following
theorem: " T h e scalar product of a homogeneous compatible field and
a self-straining field vanishes."
In most problems the prescribed boundary displacements are zero
and the particular displacement field associated with C does not appear.
This simplification will be used henceforward. Should it arise, the
problem involving non-zero prescribed displacements can be treated
by the same techniques as presented below.
4. P R I N C I P L E OF M I N I M U M T O T A L ENERGY
Under the assumption that C = 0, the minimum total energy principle
can be stated as follows:
ti = \{H,H) - {H,E)
takes its minimum value for H = S.
Proof: Since the solution S is also a compatible field, the difference
between S and H is a compatible and homogeneous field. This is
written
l l = S + dll
Substitution of this into the expression of//, expansion o f t h e product
and rearrangement of the terms produces
/i = 1(5,5) -- {S,E) + \{6H,t>H) + { S - E,dH)
184
Now S — E is the difference between two particular equilibrium fields

and is consequently a self-straining field. Its product with a compatible
homogeneous field vanishes, so the last term vanishes. The two first
terms are the value taken by pi when H = S. T h e third term is positive
unless o i l vanishes. This completes the proof.
5. P R I N C I P L E OF M I N I M U M C O M P L E M E N T A R Y ENERGY
Since the prescribed boundary displacements are assumed to be zero
there is no complementary potential energy. The principle states that
the true solution makes the complementary strain energy (the strain
energy expressed in terms of stresses) a minimum. In terms of a scalar
product
t\ — \ { E -f- A,E -f A) is minimum for E f A = S
If we subtract from E + A the particular equilibrium field S we
obtain a self-straining field; this can be written
E + A = S f dA
Substitution of this into the scalar product and rearrangement produces
A = \{S,S) + \ ' 6 A M ) + {S,dA)
Again the last term vanishes as the product of a self-straining field and
a homogeneous compatible field. T h e first term is the value taken by
X for E f A = S. The second term is positive unless dA vanishes.
This completes the proof.
6. CLAPEYRON'S E X T E R I O R T H E O R E M
Since S is a homogeneous compatible field and S — E a self-straining
field wc have
{S,S - E) = 0
This can also be written in the form
\{S,S) == \{S,E) = ~ \ P (A.15)
the last equality following from property (A. 12) so that P denotes the
potential energy of the prescribed loads under the displacements of
the true solution. This is also the usual theorem of Clapcyron according
to which the strain energy is equal to half the virtual work done by
the loads under the true displacements.
7. UPPER AND LOWER BOUNDS TO D I R E C T INFLUENCE

COEFFICIENTS
From the principle of minimum total energy and Clapcyron's theorem
(A.15) follows
\{H,H) - {H,E) > i{S,S) - {S,E) = -±{S,S) = \ P
185
From the principle of minimum complementary energy follows

\ { E + A,E + A ) > i{S,S) = -\P
Hence by changing signs and the sense of the last inequality
\{H,H) - {H,E) > \ P > - \ { E + A, E + A) (A.16)
Under restrictive assumptions used for displacements in some approxi-
mate but compatible approach, the left-hand side of the inequality-
will be the value estimated for half the potential energy; denote it by
\ P . Denote similarly by i P the right-hand side which will be the value
estimated under restrictive assumptions for the stresses in some equi-
librium approximation to the same problem. The free parameters or
functions contained in the approximations are naturally determined
by the minimum principles themselves in order that the bracket for P
P > P > P
be as tight as possible.
If the load consists of a single force F with corresponding displace-
ment w, the influence coefficient c is defined by
w = cF
and the potential energy can be written
P = —Fw = — cF 2 (exact solution)
In a compatible approximation we would find
P = — cF 2 (compatible approximation)
In an equilibrium approximation
P = —cF 2 (equilibrium approximation)
Insertion of these expressions in the bracket for the potential energy
produces after cancellation of the common factor F 2 and a change of
sign altering the sense of the inequalities
c > c > c (A.17)
Thus an approximate compatible approach produces a lower bound
and an approximate equilibrium approach an upper bound to a direct
influence coefficient.
8. UPPER AND LOWER BOUNDS TO CROSS I N F L U E N C E

COEFFICIENTS
Let
«-i = fuî + c 12 F 2
=
••21 *-12
w2 = c-jaF, + c 22 F 2
186
x -
i 9- > 9 o 5 ~s rs -
|fr a -
-.
K
—
z__
~ is. - •
0
fi V-
-1 &s n
c
9
fi T3 ; 2 c 3" -
-
, is. *•* =* S" 5 .••3 1? ^-,
3" O CJ!
c- c Si M o
— 85 ff." x A-
— K-
<r* *»
•*.
y
: S £a - -
i
n
'-
-
< *- to — — rc Sc_^ A*
-
IC ic
a —- - — a
~. rr. crq <•> 9
— ii
=;i .
—
ti
< *° - f rm. — t
n
-z J' IA — — d zr ~
. — IV IA c - ^ r VJ x
CL JW.I — 5' :
9) -. x z fi 2?
V
X
a •a >^ - r- *"" — -.
- 9 IV — Q, 6s r» X
IV £•
VJ
3
—
_-. IT
-J
< 9
+ -.
c
——
• - — =
TD fi Z
< ¥— 1 - — B 0 » , ^ 3 3
/.
 CL * . ""
-.
ff. -X < >aj
I! 3 :•
5 —. - Z
c - as
— p-
l
"**i
• - - .
I — "™ S S
<" f " —.
o— to - IC SS
— ic
-. IC — — li
-
— -. -- o : S R_
i-.
-.
-JI
c -. -
=- 1-
ti
— x li fa 9 = 9 , -ii 5 ' 5
n- — - Z — 9 ,.
+s . + ps -
I -^ —
s ~ .
0 x
> c — n -r
3a
N 9 „a. f». PB
J-. •: - 5
y
- ~gS9 " -
*- li DO ti
re- -. z 9 P
<"
ti
* ' ro -
K le.
rf M
9
j-.
rt r> -
-. 1 1 9 5 - irs -. n
-. n Z
N< l«
ti
=-
'- • ^
P
ft
1'^. + -
5
+ o
t fi'°
n 9
-:
-
< ti i- - to 3. 6 R O -
- <' N .
iS ~. ° cr, -j CL
- ft. —
*: IO 9 3 a n vi
~. o rx - z
- z B
9 N- 9 0 = !" -
— tc -: 3 <W 3 ><
S 9 >
- =
-. — —' n
-. -. S' -2 -.
from which follows

C
^12 > 12 V(-5U-5U)(«5M-5M) (A*20)
T h e second inequality can be treated by the same technique and
produces the new bounds
c
12 ^ £12 T" * ( C l l £ll)(c22 £22) (A.21)
C
I2 ^ f
Cl2 ~" * ( I l f
£ll)( 22 £22) (A.22)
m e
If, as usually the case, it is found numerically that c12 > C12 h 08 *
bracket is made of inequalities (A.20) and (A.21).
APPENDIX B
Examples and Numerical Results
B. M. FRAEIJS DE VEUBEKE and Guv SANDER
1. S T R U C T U R A L MODEL
The ultimate step is the analysis of a complete and complicated
structure. Very instructive results are, however, obtained concerning
the validity of the dual approach by the consideration of a rather
simple structural element.
A rectangular element, or panel, of length 2a and height 2b, edged
by beams of identical cross-section S, was analysed by various methods
and the influence coefficients compared. The beams are assumed to
be devoid of flexural rigidity. The two structural parameters of the
structure arc then
r = a/b
R = S/{bt)
where t is the constant thickness of the panel. Exploratory values of
1 and 5 for r, and 0-4-10 2-0 and 4-0 for R were adopted.
The reference to Model 1 or Model 2 relates to the type of support
of the structure and is explained in Fig. 1 1.
T h e loading cases investigated arc shown in Fig. 12 and referred to
as Cases I, I I , I I I or IV.
2b
model 1 model 2
Fig. 11
188
> > • t>

model 1
!> -> &- •a '-&
easel caseH caseTH case TY
model 2.
Fig. 12
2. M E T H O D S OF ANALYSIS
The methods of analysis that were compared arc the following:
(1) Tension 4. This denomination refers to a subdivision of the
panel in four triangles by the diagonals and construction of a pure
equilibrium field as explained in Sections 2.3 and 3a of the paper.
It is found that the state of stress in the panel is either zero or a state
of pure shear.
(2) Tension 16. The panel is first subdivided into four equal
rectangular fields of dimensions a and b. Each rectangle is then sub-
divided in four triangles by the diagonals. Again a pure equilibrium
field is constructed.
In contrast to the "Tension 4 " case there arc redundancies in the
form of self-strainings. Figure 13 shows the self-strainings involved in
"•-_ Ii _ - •
f
X
<s>
V--P-PX
< & >
X
t
(//ffî
<sl> Y
<T y >
X type self-straining '— m-~
c
Y type self-straining
— Za —
t
Zb
mm. Za _^
Z type selfstraining
Fig. 13
189
B. M. F R A E I J S DE VEITBEKE a n d GUY SANDER
the Model 1 type of support. In the case of Model 2 there are additional
self-strainings depicted in Fig. 14.
"Tension 4 " and "Tension 16" arc the only methods used to provide
upper bounds to the influence coefficients.
(3) Turner 16. T h e subdivision of the panel in triangles is the same
as for "Tension 16". However, the field of displacements is purely
V type self-straining
w—
2W*-
2W-*
"""" —a.
W-— 1
W type self-straining
Fig. 14
compatible. It is identical to the Argyris-Turner triangularization

method and the calculations were essentially made according to the
Turner stiffness method. 5
(4) Turner 4. It is the triangularization and stiffness method
applied to the subdivision used in Tension 4.
(5) Argyris 4. T h e panel is first subdivided into the four equal
rectangular fields. In each field the displacement assumption is used
u = u0 + px + qy + r xy
v = v0 -\- Ix -\- my -\- n xy
The eight parameters can be expressed in terms of the corner displace-
ments. As in the case of triangular fields continuity of the displace-
ments for the whole panel is obtained by simply requiring identical
displacements at the common corners. Corner loads and stiffness
190
Table 1
I--,
a a a t>
r = = 1 r= - = 5
6 b
t
-J 0-4 10 2-0 4-0 0-4 1-0 20 40
>
Methods Case I
i i'>
Tension 4 0-2106 01196 00893 00742 9-840 4113 2-204 1-249
Tension 16 0-2038 01185 00890 0-0741 9-182 3-989 2171 1*241
Turner 16 0-1287 00991 00825 00719 2-139 1-699 1-303 0-939
Turner 4 01059 00894 0-0767 00690 0-810 0-765 0-561 0-606
Argyris 1 0-1099 0-0895 0-0773 00692
Argyris 4 01238 00948 00807 006945
Beam, theory 00986 00813 00719 006603 5-239 3068 1-896 11630
Fict. spars 01374 0-1045 00850 00730 1-985 1-556 1159 0-788
>
Case II *P
la*
Tension 4 0-1136 00454 00227 001136 0-5681 0-227 0113 00568
Tension 16 00930 00415 00217 00111 0-365 0178 0-0999 00529
Turner 16 00545 0-0315 00185 00102 0143 0103 00705 00434
Turner 4 00393 0-0256 00162 00094 0101 00749 00534 0-0349
Argyris 1 00413 0-0263 00165 0-00953
Argyris 4 00468 0-0299 00177 000982
Fict. spars 00588 00340 00194 0-01048 0109 0085 0-0619 00401
191
II. M. E R A E I J S DE V E U B E K E a n d GUY SANDER
Table 2
A— P
Methods Case III
— P
Tension 4 01136 00454 0-0227 00113 0-5681 0-227 0113 00568
Tension 16 0-0976 0-0425 0-0220 0-01117 0-4201 0-1973 01054 005466
Turner 16 0-0570 00325 00189 00103 0-117 00895 00642 0-0410
Turner 4 0 0468 00291 00176 00099 00409 00370 0-0317 00248
Argyris I 0-0508 00304 0-0182 0-0101
Argyris 4 0-0632 00321 00186 0-0102
Fict. spars 00588 0-0340 00194 00104 0109 00852 00619 00401
IS -A-P
Case IV L->-
— P
Tension 4 0113 00454 0-0227 00113 0-568 0-227 0113 00568
Tension 16 00885 0 0406 0-0214 00110 0-310 0-162 00935 00511
Turner 16 0-0520 00305 00182 00100 0169 0116 00768 00458
Turner 4 00318 0-0222 00148 00089 0161 0112 00750 00450
Argyris 1 00318 0-0222 0-0148 0-00895
Argyris 4 00399 00267 00166 0-00953
Fict. spars 00588 0-0340 0-0194 0-0104 0109 00852 00619 00401
192
ip.l
IOH-00 61900 S800 6010 8t0l00 teio-o Ot£00 8850-0 SJBds *1DLJ
£96000 5/1100 0630-0 9i.W)-0 f SuA3jy*
3S6000 UIO-O 1930-0 01 WH) j suXSjy
BtSOO 0£G00 OfLOO 6600 H-600-0 OL 10-0 GG300 06£00 fjoujn j
3£K)0 00£00 IOI-0 0+1*0 8 6 6 0 0 0 8ZI0-0 9630-0 96tTj-0 91 J s u j n x
63eo-o 6 6 6 0 0 8iI00 S9£0 01100 I I 600 moo U800 91 UOISUDJ.
89500 £110 LZZO 89C0 £1100 £3300 rGt-0-0 9£II-0 JJ, UOISU3J_
</ —
i / / >™D
S6L-0 3SI-I W I S96-I G690-0 SCiO-O I £ 6 0 0 [ 6/III-0 sjcds -JOIJ
0£9ll 968-1 890£ 6£3S £09900 6U0-0 £1800 98600 A.io.-)i)i -tuB-jg
8." 9 0 0 6G/100 £3800 6W50-0 if n i X S i y
5990-0 £S£00 i0800 3*60-0 J sijXSjy
1090 £89-0 gei-o 0080 W900 CSiO-0 S0800 03600 ^ jatun j
££6*0 00E-I OOL-1 I£I-3 Z890-0 L9L0-0 £6800 3II-0 91 J s u j n x
991* I £06-1 680£ c+£-S I0/100 £0800 £G600 8* 1 0 Q[ u o i s u a x
9*2-1 961 Z 860+ £08-6 i-OiO-O 8 1 8 0 0 i-OIO ZL 1 0 J(, UOISU3J_
d \7
/ >SV
D spoi/i>fy
\
s= *
11
0t 0-2 0-1 H) 0+ OS 0-1 H)
5 =i ? = i
\ 1 i - V
s mvj
S I S A I V M V iVHfiion**txs xi*a.i.vi\ NI s a x n o a *aa.v\oi CINV «5i<i<in
B. M. FRAE1JS DE V E U B E K E a n d GUY SANDER
Table 4
•>- P
Methods Case Hi
' i ~* P
Tension 4 01136 00454 0-0227 00113 0-568 0-227 0113 00568
Tension 16 0-0971 0-0418 00216 00110 0-420 0-197 0105 0-0544
Turner 16 00569 00324 00189 00103 0-117 00896 00641 0-0410
Turner 4 00469 00259 00150 0-0099 00409 00370 00317 00248
Argyris 1 00508 00304 00182 0-0101
Argyris 4 0-0631 0-0321 00186 00102
Fict. spars 00588 00340 00194 001048 0109 0-085 0-0619 00401
-*• P
Case IV i --P
Tension 4 01136 0-0454 00227 00113 0-568 0-227 0113 00568
Tension 16 00783 00390 00206 00110 0-310 0162 00935 00511
Turner 16 0-0424 00268 00167 000963 0-163 0114 00759 00454
Turner 4 00312 0-0219 00146 0-0089 0158 0111 00744 00448
Argyris 1 0-0312 00219 00147 000893
Argyris 4 00390 00260 00165 0-00950
Fict. spars 00588 00340 00194 001048 0109 0085 00619 00401
194
561
•spunoq .IOMOI* pui; j o d d n jsoq
oqj A[I*EJOUOS OJB qoiqM 'spoqjoui g\ uoisuoj^ puE J o u j n L *>q. uooMjoq
pojuooj OJE sopiAOjd ,i sjuopyjooo oouonyui oq L "II--M SSJOM u o p
-dumssE oqi oopoBjd u j "sppy JBoqs j n o j j o uuojsAs E OJ p o o n p o j uoqj
OJB ojE|d oqj Aq pojjuusuEjj SOSSOJJS oqj^ *poA.iosuoo OJB BIJJOUI j o
JUOUIOUI JEJOJ o q , puE EOJE [EUOUOOS-SSOJO JBJOJ oqj q i o q .rnds p u j u o o
B SE spjiqj-owj p u s JBds oSpo UB SB pOJBJJUOOUOO SI o u q d oqj JO EOJB
OSJOASUEJJ E }o qjxis-ouo u o q \ \ *sji*ds JEJJUOO puB o3po s n o i q p q Aq
ojBjd oqj j o AjioBdBO SUIAJJBO-SSOJJS joojip oqj SuioB|doj ui SJSISUOO J J
• p u n o q JOJVVOJ J O j o d d n UB opiAOjd Ajumjjoo qjiM jou soop jBqj poju'S
-IJSOAUI u o p d u m s s B p u q A q A[UO oqj si siq ^ -tCxodui xvds s n o t p p i j {f)
•ojEjnooB AJOA uoqj OJE qjoq
Apuonbosuoo p u E poqjoui 91 uotsuo j oqj o j o s o p AJOA oq o j uoos si J J mx
j o sonpÂ OSJEJ JOJ p u n o q JOMOJ SUIJSOJOJUI UE sopiAOjd poqjoiu oqj^
'Z I J P°IV ° - s*; I I ° P ° W 0 1 I P M s , ; soqddB p o q j o m
oqj 'ouEjd un;uio.i oj poumssE OJB X OJ JEUJJOU suopoos oqj osnEOog
*||EUis OJB g puB x uoq/v\ j o o d 81 jmso.1 oqj uoijdiunssi; siqj j o osnBOog
•uoijoojip it"oqj ui Ajii;iqisuoixoui j o uoijdomssB oqj oj o n p u o p n q u j u o o
qo.w posEOJOui UB qjLv. uopoos-ssojo oqj j o BHJOUI j o juouioui oqj si
.1.1,)i( u
o m , Tii <i
n uta) " p
|E0issEp si juoioiyooo oouoiqpii
oqj j o j p o u m j q o orqBA o q j "sjuouiooiqdsip JOJ SUOUBIJBA J O ojchouiid
oqj UJOJJ p o u i u u o j o p OJB {X)A puB {x)g s u o p o u n j u.wmi-jim o q j
ix)A = a
{x)giC=- n
odAj oqj J O p|oy E p u E d oqj JOJ

b'uiumssE ui sjsisuoo J J 7 OSEO SuipBoj oj oijqEoijddE 'A.iooqj S u i p u o q j o
odAj q j n o u j o g Oi*qpBduioo .([.und E OJ SJOJOJ siqj^ •iCxo.tu) uxvig (9)
•ouo JSE[ oqj
oj j o s o p AI-JEJOUOS 'spoqjom g\ puB ^ j o u j n x oqj uoojwjoq sjpisoj OABS
JI 'posn joAouoqAA *spunoq JOMO| ppiA j s m u '91 puE f JOUJII j qji.u
joqjoSoj 'poqjoui siq ^ -soiipo u o m u i o o oqj SSOJOE popjoooj OJB SOSSOJJS
oqj ui sopmupuoDsip jo[|Ems j n q Ajsnonupuoo pojEjoiA OJE s u o p i p u o o
l u n u q q m b o oqj^ " p p y ium.iqi[inbo UE JOU SI u o p d u m s s E juouiooEjdsip
oqj qji.w pojBpossE p p y SSOJJS oip jEqj si SOISUEIJJ OJUI uoisiAipqns E OJ
joodsoj qjiM oouojojjqo JOIEIU oq j ^ ^ s p p y jEpiSuEi.ij oj poqddi; poqjoiu
ssoujjps oqj JOJ posn SB sojnpooojd OUJES oqj Aq pouiBjqo OJB SOOUJEUI
SISAIVNV i v a n i D n m s XIMXVIV NI s o N i i o a U J M O - I O N V ua.1.1.1

B. M. F R A E I J S DE V E U B E K E a n d GUY SANDER
It is, however, recognized that in this idealization both the Poisson's

ratio effect and sweep-back effects are lost.
It would therefore be interesting to pursue the investigations on a
model incorporating sweep-back and perhaps taper.
Note: The values recorded in the tables and compared in Figs. 17, 18, 19 and 20 arc
displacements corresponding to specific values:
load P = 1000 kilos
E = 2-2 10- kilo/mm 2
b= 1000 mm
1= 2 mm
To obtain the non-dimensional quantities
Et
dp -/(',«)
where d is Ihe displacement cm responding to the load, the table values should Ix- multiplied
by the factor 44. In the figures, heavy continuous lines are for lower bounds, dotted lines
lijr upper bounds. The thin continuous line is for the fictitious spar theory.
3. R E D U C T I O N O F S E L F - S T R A I N I N G C O N D I T I O N S FROM
SYMMETRY CONSIDERATIONS
Sometimes advantage can be gained from the symmetries in reducing
the number of self-strainings to consider. Take for example Model 1
with loading case I I . Figure 15 shows that this case can be considered
as the superposition of loading cases IV and I I I .
•• —a.
IV + in
Fig. 15
p*— ] —»»P*Za
"^Cz7
1 X CDy S
2 ®x®
I'
Zd
\ Zb
1 t 1
Zb Zb
1_ 1
1
Fig. 16
196
I ITER Wl) LOWER HOUNDS IN MATRIX STRUCTURAL ANALYSIS
(a) Case IV. It is symmetrical with respect to both the x and y

axes. Hence we need only consider a self-straining of type Z when
dealing with the Tension 16 method. It is readily found by investiga-
tion of the equilibrium conditions of the triangles (Fig. 16) that in
triangles 1 and 2 the state of stress is one of pure shear:
am = 0 ay = 0 Tmy - Z
while in triangles 3 and 4 it is the state
Om= ~ Z ~ b Om= - Z - T„ = C
The corresponding strain energies (see p. 166) are
2 -fi{2(l+.«.)=|(l-M2-
1 ab I
2 ^(•^•^H-Se+5-*)"
Q25.u
Displacements
Model 1 r=i easel
1 tension L
i ® .. 16
a?o rr\ 3! Turner 16
L Turner /i
(6) Beam theory
(7) Fict. spars
OJS
mo
006
Fig. 17
197
11. M. FRAEIJS DE VEUBEKE a n d GUY SANDER
Adding and multiplying by 4 (the other three rectangular fields have

symmetrical states of stress) the panel energy is
".-f(»+S+3»
In half of a vertical beam the tension, computed from equilibrium
considerations, is
T = Zt{b - y )
and the corresponding strain energy for both complete verticals

2 t2b3 2
U — Z
U
" ~ 3 ES
In half of a horizontal beam the tension is the superposition of the
self-straining effect a n d a particular equilibrium state under the
external horizontal load P applied at the e n d :
T = P + Zt{a + x)
0.125 , u
,1
Displacements
li
i i
Model 1 r:1 easel
0.10 (\ tension 4
(Tj tension 16
( 3 ) Turner 16
f • Turner 4
( 5 : A r g y n s t,
0.075 •1 Fict. spars
0.05
0.026
The corresponding strain energy for both complete horizontals
Ub = * s {aP2 + a 2 PZt P \aH 2 Z 2 }
The total energy of the structure is thus found to be
z i
v ; f i , , ( 2 + r ^ ' ? ) + 23RT ]: \ i 2 zE s 'R4 + rE' 2Rt'
From Menabrea's theorem t U / d Z = 0 follows
bt\3R{2 + r2 + r 2 ) + 2r2\
and finally from Castigliano's formula
,_1W IfiW 4rp

2dP E\ R Rt
025
Displacements
Model 2 R1 case I
020 kD 1 ' tension 4

2 tension 16
3) Turner 16
4 Turner 4
6 Beam theory
CIS 7) Ret. spars
0.10
Q05
Fig. 19
199
B. M. FRAEIJS DE VEUBEKE and GUY SANDER
where d is the deflexion under each P load of Model 1. After substitu-

tion of the Z value there comes
3r2
2
EtR 3R{2 4- r + r~2) + 2r \
2
Note that in the tension 4 approach no self-strainings exist and we

would have Z = 0 with a corresponding deflexion
Et" R
This is the first term of the previous formula which represents the
elongation of a horizontal beam under the end load when the plate is
under no stress and does not participate to the general stiffness. The
second term represents the plate contribution in reducing the deflexion.
(b) Case III. It is symmetrical with respect to the y axis, anti-
symmetrical with respect to the x axis. It can be dealt with by con-
i:c
r, i
Displacements.
90
Model 2 r=5 easel
80 1 tension 4
2 tension 16
3 Turner 16
70
4 Turner 4
6 Beam theory
60 (7) Fict. spars
50
40
30
2.0
10
30 R = -gt 40
Fig. 20
200
sidering only the Y type of self-straining. By similar calculations there

is found
t R 2 T l 2 1 + 2 V
U ^ 4 - ( )
+ + 3r+
3R 7* ~T
d
= EiU2
tR[ f r4 + R + 3Rr* + 2(1 + 2v)r2R
Again the first term represents the beam elongation in the absence of
the plate, as would be found in the tension 4 approach. The second
term is the plate effect given by the tension 16 approach.
(c) Case II. Returning to case I I , it is obvious from the principle
of superposition that the deflexion under each load is the average
between those of case I I I and case IV.
201
FURTHER DEVELOPMENT AND
APPLICATIONS OF THE STIFFNESS
METHOD
M. J. TURNER-)-, H. C. MARTINJ, and R. C. WEIKEL§
CONTENTS
Pa,,
ABSTRACT . . . . . 204
1. INTRODUCTION 204
2. STRESS CALCULATIONS . 213

Theory . . . . 213
Examples—Stresses 219
Sheet under parabolic loading 219
Ring-stiffened circular cylinder 221
Sweptback wing model 227
3. THERMOELASTIC ANALYSIS AND ELASTIC STABILITY 241

Structural Elements 242
Stability Analysis . 250
Wing Analysis 253
Plate Buckling Analysis 259
Stability of a Tapered Column 260
4. APPLICATIONS TO COMPLEX STRUCTURES 263
REFERENCES. . . . . . 266
j Chief, Structural Research and Development Section, Aero Space Division, The
Boeing Company.
• Consultant and Professor, Department of Aeronautical Engineering, University of
Washington.
§ Senior Croup Engineer, Stress Methods Group, Aero Space Division, The Boeing
Company.
II 203
II, j . T U R N E R , H . C. MARTIN and R. C. W E I K E L
Further development of the direct stiffness method of structural analysis is

reported in this second AGARD paper.
The paper is divided into four parts. These are {I) Introduction, containing
general information associated with computer programming of the method; (2)
Stress Calculations; {3) Thermoelastic .Analysis and Elastic Stability;
{4) Applications to Complex Structures.
Of particular interest in Part 1 are the discussion on choice and use of axes
systems, transformation matrices, procedure for handling mixed boundary value
problems, and details associated with imposing constraints on certain classes of
problems.
Part 2 presents in detail the calculation of stresses. It is shown that stress
determination should be based on averaging results obtained from the so-called
stress matrix, or on a procedure employing nodal forces. Equivalence of the two
procedures under certain conditions is demonstrated on theoretical grounds. Stress
calculation along an edge or boundary is also included. Three examples are
• given to illustrate the theory.
In Part 3 attention is given to problems possessing geometric nonlinearity.
A new derivation is given for the triangular element undergoing in-plane deforma-
tion in the presence of initial stresses. This derivation leads to simpler results
than those previously published. Examples illustrating the theory include several
stability {eigenvalue) calculations, plus the analysis of a healed sweptback wing.
Finally Part 4 illustrates the application of the direct stiffness method to the
structural and vibration analysis of an actual helicopter structure. The structural
idealization used in the analysis is shown and calculated results are compared
with experimental data.
I. I N T R O D U C T I O N
IN a paper, 1 presented at the 1959 meeting o f t h e A G A R D Structures

and Materials Panel in Aachen, the essential features of a system for
numerical analysis of structure, termed the direct-stiffness method,
were described. T h e characteristic feature of this particular version of
the displacement method is the assembly procedure, whereby the
stiffness matrix for a composite structure is generated by direct addition
of matrices associated with the elements of the structure. Applications
discussed in the earlier paper were limited to analyses of deflexions and
vibration modes of a few simple structures. In the present paper we
shall describe subsequent improvements and extensions of the direct
stiffness method, including the development of stress analysis capa-
bilities. Results of applications will be presented from deflexion
analysis of a complex helicopter fuselage, stress analysis of a model
sweptback wing and thermoelastic analysis of a nonuniformly heated
sweptback wing.
Before entering into a discussion of analytical developments and
applications it may be of interest to mention some of the general
features which we believe should be included in developing a general-
purpose program based on the stiffness method. For those who may
204
DEVELOPMENT AND APPLICATIONS OF T H E STIFFNESS METHOD
not be familiar with the earlier account, it is noted that the structure
is decomposed for analysis by a network of intersecting cuts into an
assemblage of one- and two-dimensional elements. Points of inter-
section are termed nodes; structural deformations are defined by the
nodal displacements relative to a basic reference system. By intro-
ducing a sufficiently small mesh size (and correspondingly large number
of elements) the local deformations within an element may be approxi-
mated by simple functions containing a small number of independent
parameters; these parameters are expressed by simple kinematic
considerations in terms of the displacements of neighboring nodes.
For an individual element the generalized nodal force increments
{AA''} required to maintain a set of nodal displacement increments
{Au} are given by a matrix equation
{AZ'} = K'{Au} (1.1)
in which K' denotes the stiffness matrix of the individual element.
Resultant nodal force increments acting on the complete structure are
{L\AX} = 2{AZ'} = K{Au} (1.2)
wherein, K, the stiffness matrix of the complete structure is given by
the summation
K = AZK* (1.3)
which provides the basis for the matrix assembly procedure noted
earlier.
For the solution of linear problems involving small deflexions of a
structure at constant uniform temperature which is initially stress-free
in the absence of external loads, the matrices AP arc defined in terms
of initial geometry and elastic properties of the materials comprising
the elements; they remain unchanged throughout the analysis.
Problems involving nonuniform heating of redundant structures and/or
large deflexions are solved in a scries of linearized steps. Stiffness
matrices are revised at the beginning of each step to account for
changes in internal loads, temperatures, and geometric configuration.
Stiffness matrices are generally derived in local reference systems
associated with the elements (as prescribed by a set of subroutines) and
then transformed to the basic reference system. It is essential that the
basic program be able to accommodate arbitrary additions to the
collection of subroutines as new elements are encountered. Associated
with these are a set of subroutines for generation of stress matrices iS*
relating matrices of stress components {a'] in the local reference system
to nodal displacements
M = S'{u} (1.4)
The matrix {u\ denotes a matrix of resultant displacements relative to
a local reference system which is attached to the element. This concept
205
M. J. T U R N E R , H . C MARTIN a n d R. C. WEIKEL
will be discussed further in Section 3. Provision should also be made

for introduction of numerical stiffness matrices generated outside the
program. This permits the utilization and evaluation of new element
representations which have not yet been programmed. It also provides
a convenient mechanism for introducing local structural modifications
into the analysis.
Subroutines for the treatment of standard components, consisting of
assemblages of simple elements are also very useful. For example,
computation of stiffness and stress matrices for a complete stringer
joining many nodes may be programmed for automatic computation
utilizing the formulae for the simple two-node segment. Descriptive
data for such a component would consist of a list of nodes, stringer
areas at the node points and material identification.
Another useful component is the four-node beam segment consisting
of a flat quadrilateral web plus chord members, as indicated in Fig.
1.1a. Descriptive data include node identification, web thickness, chord
3
4 L
Fig. 1.1a. Four-node beam segment.
areas, and material identification. The web is treated as an assemblage

of four triangular plate elements in plane stress. Equilibrium con-
ditions at the interior node point (where there are no applied loads)
are employed to eliminate the displacements at that node. Generally
the ratio of length to average depth of such a segment should not
exceed 1*5, since the strain components are treated as constants
throughout each of the triangular plate elements comprising the web
and the use of elements which are excessively elongated leads to poor
approximations.
The subroutine for the four-node beam segment may be used in
turn as basic unit for the generation of matrices representing a complete
spar, as suggested in Fig. 1.1b. Provision should be made for checking
length to depth ratios ofthe segments and introducing additional nodes,
such as 1' and 2', as required to keep these ratios within acceptable
limits. Displacements at these supplementary nodes will generally be
eliminated (since they do not appear on adjacent components).
A further application of the four-node beam segment is suggested in
Fig. 1.1c wherein a fuselage frame is approximated as a ring with
polygonal boundaries. Certain of the displacements at interior nodes
206
i V 5 6 7
1
1
s
1i 2' tO 9
14 '3
Fig. 1.1b. Spar composed offour-node segments.
Fig. 1.1c. Fuselage frame composed of four-node segments.
identified by primes, where there are no applied loads or connexions

to neighboring components, may be specified for elimination in
deriving the final stiffness matrix for the component which will be used
in the structural analysis.
Perhaps it is inevitable that users of a general-purpose program for
structural analysis will be tempted to overload the computer by trying
to perform a complete analysis of a large, complex structure in a
single continuous operation. However, it is certainly evident (and
experience has unfortunately confirmed the obvious) that there is a
practical limit to the complexity of structures that can be analyzed on
a given computer by means of a single, closed-form program; this
limit is dictated by storage capacity and the maximum reliable running
time that can be obtained with the available equipment. In view of
this limitation it is advisable to perform deflexion and stress analyses
of complex structures in separate operations. In addition, it is fre-
quently advisable to divide the structure into major components for
separate analysis; this will be illustrated later in discussing the analysis
of a helicopter fuselage which was analyzed in three parts (see Fig. 4.2,
p. 264). In analyzing a major component the basic input data will be
used initially to generate both stiffness and stress matrices for the
elements; the stress matrices arc stored on separate tapes for later use
in the stress analysis. The stiffness matrix for the component will then
207
M. J . TURNER, H. C. MARTIN a n d R. C. WEIKEL
be generated by merging the stiffness matrices of the elements. At this

stage the component is treated as a free body. Its nodal displacements
may be classified in three distinct categories: (a) those associated with
points of support or application of external loads; (b) those occurring
at points of attachment to adjacent components; (c) displacements
occurring at "internal" node points where no external loads are
applied—these freedoms arc required to achieve satisfactory repre-
sentation of strain distributions. The next step is to employ equilibrium
conditions at the internal nodes to express the displacements in
Category (c) as linear functions of the remaining displacements.
Matrices occurring in the equations representing these functional rela-
tions are stored for later use in the stress analysis, since all nodal
displacements are required to define strains (and stresses) in the
elements. T h e stiffness matrix is then reduced to lower order by
eliminating Category (c) displacements and stored for use in the final
displacement analysis of the complete structure. In this final analysis
the reduced stiffness matrices are merged and support conditions are
imposed. The resultant stiffness matrix is inverted and used to calculate
displacements due to external loads. These displacements are then
used as input data for stress analyses of the major components.
In order to provide a capability for analyzing structures of general
form it is essential that the program be able to handle assemblages of
elements having arbitrary relative orientations. As noted earlier
stiffness and stress matrices are generated in a local fixed reference
system having some convenient relation to the element in its initial
condition. For example, as indicated in Fig. 1.2, it is convenient in
deriving the stiffness matrix for a short, initially straight, untwisted
Fig. 2.2. Local reference axes associated with a beam-column element.

208
beam-column element to employ a local reference system in which the

x-axis passes through the two terminal nodes and the j-axis is oriented
so that a principal axis of the cross-section at one end of the member
lies in the jf-J-plane. Input data for the member will identify the end
nodes, 1 and 2, for which numerical coordinates referred to a basic
.v^-z-reference arc available. In addition direction cosines of the
specified principal axis arc given. Coordinates of the node points are
employed in calculating direction cosines of the jf-axis; orthogonality
relations for a right-handed system are then employed to calculate the
remaining direction cosines for y- and z-axes. For the general case,
including axial elongation, bending in two planes and torsion, the
member is treated as a system with twelve degrees of freedom com-
prising three translations and three rotations at each end. Referred
to the local reference system the displacement matrix and the corres-
ponding force matrix are, respectively
{Au} = {AM,, AVX, A W U Aa^ A/J\, Afx, Au2, Aii2, Aw2,

Aa 2 , A&, Ay2} (1.5)
{AX} ==- {AA„ A ? . , A2„ A,\f u , AM lB , AA/,,, Mm, A? 2 , AZ2,
A M V , A.\l.19, AM,.}
The stiffness matrix in this system is denoted by R and the force-

displacement relations are
{AX} = R{Au} (1.6)
In idealizing the structure it is assumed that beam-column members

are divided into short enough segments so that linear approximations
may be employed in representing the clastic behaviour of a single
segment. This implies that throughout the analysis loads on such a
segment must remain small relative to critical buckling loads for the
segment.
In transforming Eq. (1.6) to the basic reference system it is convenient
to employ the third-order matrix
v
p- ,"- i
A = h IH v
i ;i.7)
' • !> ' ' •
whose rows are composed of the direction cosines of x-, y-, and z-axes
(in that order) relative to the x^-z-reference system. Then it is easily
shown that displacement and force matrices transform as follows
{Au}=H{Au}
{AX} = H T { A X } [
'
209
M. J . TURNER, H . c. MARTIN and H. C . WEIKEL
where
"A 03 03 o , '
03 A 03 03
H =
03 03 A o3
_03 03 o3 s.m
0, = third-order null matrix
and by substituting the expressions for {Au] and {AA'j from Eqs. (1.8)
into Eq. (1.6) it follows that
1A.VI = K{Au} (1.9)
where
K = ffTKff (1.10)
As a further option it is sometimes advantageous for the analyst to be
able to introduce fixed reference axes at each node. To illustrate the
procedure we consider in Fig. 1.3, a beam-column member in one
Fig. 1.3. Nodal axes.
quadrant of a structure which is assumed to have axial symmetry

about the z-axis. In this example the local axes at a node are oriented
so that the z 0 ) -axis is parallel to the z-axis and the x'-'-axis intersects
the axis of symmetry. In general the analyst would be required to
indicate at the beginning of a problem whether nodal axes or a common
reference system will be employed for forces and displacements. If the
former option is selected then it will be necessary to list direction
cosines as initial input data (or specify a procedure for computing
them) along with the nodal coordinates.
210
For the member 12 in Fig. 1.3 displacements and force matrices

referred to nodal axes are
{A«<"} = {Aa'-», Ar«>, AM><-», Act-1-, A/?'1', Ay'1*, A«-2',
2
A * , ' ' , A. W {2) > ^ ( 2 ^ A.£(2I ; ^(2)}
(1.11)
{AA™} = {AA-«», . . ., AMf>}
Furthermore the column matrices defined in Eq. (1.11) are related to
{A«}, {AA'} by the equations (analogous to Eqs. (1.8))
{Au<"} = 3f{Au}
(1.12)
{AX} = ,3f r{AA,<*>}
in which
A<" o3 o. o3
o3 A (1) o, o3 (1.13)
0a o3 A'2> 03
o3 o, 03 A' 2 '
and
) _ (1.14)
A* i"j° H°.
Since
W - 1 = JtfT (1.15)
Equations (1.12) may be solved for {A«}, {AA'"'} to give
{Au} = Ji" T {Au^}
(1.16)
jAA'-*'} = j r { A X }
and Eqs. (1.16) may be used to transform Eq. (1.9) into
{AX<"} = j r { A « ' " [ (1.17)
with
J f = -WK.W T = WH T RB-/f T
(1.18)
Introduction of nodal axes in the analysis of a structure with axial
symmetry makes it possible to retain that symmetry in the analytical
formulation. If the applied loading also exhibits axial symmetry, the
analysis may be formulated for a sector ofthe structure with appropriate
boundary conditions.
A special procedure is often required for solving mixed boundary-
value problems. Forces corresponding to a selected set of displacements
are assigned and the remaining displacements are required to assume
prescribed values. It is then required to determine numerical values
for the displacements corresponding to the prescribed forces and vice
211
M. J . T U R N E R , H . C. MARTIN a n d R. C. W E I K E L
versa. Elements of the displacement and force matrices are rearranged

as necessary and partitioned
{Au} = {Aa«"; Au<"»}
{AAX} = {AZ"*>; AA™} (1.19)
, ) <M
The submatrices jA.V " j, JAu j are given and it is required to deter-
mine {AA'""}, {At/-"'}. For this purpose the stiffness matrix K is
rearranged and partitioned in conformity with Eq. (1.19)
•âa A-ab 1
A (1.20)
-^tra KbbJ
Then the force-displacement equations may be arranged in the form

of two matrix equations
{AA-""} = ^{A«<->} + Kab{Au<»}
{AX*>} - K ba {Au^} + Kbb{Au*>}
from which there follow
{A.""} = K J { A X ^ } - K,jK ab {Au^}
{AA™} = K^KîAX™} + {Kbb - K ^ K ^ K J W } .
Finally wc shall conclude this section by noting that a procedure for
imposing linear constraints in structural analysis is frequently useful.
The constraint equations may be expressed in the form of a single
matrix equation
M{Au} = 0 (1.23)
It is assumed that the matrix M is given; it is necessarily of rectangular
form. The number, m, of constraints (assumed independent) is equal
to the number of rows in M. The constraint conditions are to be
employed in eliminating a set of m displacements specified by the
analyst. As before we rearrange and partition the displacement
matrix as indicated symbolically in the first of Eqs. (1.19), the elements
of {Aub} having been selected for elimination. M is partitioned con-
formably and Eq. (1.23) is written in the form
Ma{AM°} + Mh{Aub} = 0 (1.24)
There follows from Eq. (1.24)
{Au6} = -Af 6 - 1 Af.{Ai«"} (1.25)
and the force-displacement relations for the reduced system take the
form
{AA"-} = ( * „ - K M l M a ){Au°} (1.26)
This type of procedure may be employed to introduce inextensional
conditions governing flexural deformations of curved beams and
212
D E V E L O P M E N T AND A P P L I C A T I O N S OF T H E STIFFNESS METHOD
frames. Another application arises in wing analyses wherein bending

stiffnesses of cover plates are neglected. If in the skin segmentation
process nodes are introduced at interior points not supported by ribs
or spars the resulting stiffness matrix will be singular unless constraints
are employed to eliminate components of displacement perpendicular
to the cover plate.
2. STRESS CALCULATIONS
This section of the paper will be divided into two parts. The first will
present a brief account of theory pertinent to stress determination,
while the second will be devoted to application of the theory to several
examples.
Theory
The discussion will be restricted to the problem of finding stresses in
planar sheets. These sheets may be individual pieces, or they may be
components of a larger structure. In the aggregate they may form a
non-coplanar surface. Such a case will be considered later (Example
2).
Whatever the geometry of the overall sheet may be, it will always
be replaced by an assemblage of triangular elements. These elements
will be connected at their vertices (nodes). A typical element is shown
in Fig. 2.1.
Within any single triangular element the strains (and therefore
stresses) arc taken as constants. The constant stresses are replaced by
stress resultants shown as nodal forces X i t Y( {i = 1,2,3) on Fig. 2.1.
x.,u.
Fig. 2.1.
Furthermore, these nodal forces may be related to corresponding nodal

displacements through the stiffness matrix K r , for the triangle. This
stiffness matrix was first given in Ref. 2.
213
M. J . T U R N E R , II. C. MARTIN a n d R. C. W E I K E L
Now the theory necessary in deriving K' is also sufficient for deter-
mining the stresses within the triangle once a„ Vt have been found.
Finding these nodal displacements is the primary goal of the stiffness
method; consequently they may be considered as known and hence
available for determining stresses.
It therefore becomes a simple matter to relate stresses, a, to displace-
ments, u. The relevant matrix equation may be written as
ial = [•*]{«} (2.a)

where
f/f.
.... = = stresses in triangle
[u] nodal displacements for triangle
"3 J
[S] = so-called stress matrix and depends on nodal co-

ordinates, Young's modulus, and Poisson's ratio. Reference 3 contains
a detailed discussion of these points.
At this point a practical matter of considerable importance arises.
Since stresses and strains are constants within any given triangle, dis-
continuities in these terms will occur across common edges of adjacent
triangles. The confusion which results in plotting and interpreting
stress data is augmented by the fact that these discontinuities can be
seemingly erratic as well as large. As a result the direct-stiffness
method has proved disappointing, or at best severely limited, to most
investigators as a procedure for determining stresses.
The purpose here is, therefore, to present new information which
will help resolve these apparent difficulties.
At the outset it is important to notice that two procedures are
possible for finding stresses. One is based on the previous discussion
and Eq. (2.a). T h e other is based on using the element stiffness matrix,
K', and known nodal displacements to determine the stress resultants
X : , Y,. The equation is,
{X} = [*']{«} (2ab)
T h e stress resultants may then be used to determine stresses.

214
In pursuing this further it is instructive to consider a sheet, decom-

posed into triangles in such a way that the triangles form a set of
rectangles as illustrated by Fig. 2.2. The separate triangles I, . . . I V
have nodes 1, . . . 5 as shown. We suppose nodal displacements to
be known.
A very interesting result will be discovered if this problem is studied
in detail. First, suppose stresses in the triangles are calculated using
Eq. (2.a). In order to simplify the calculations assume nodes 2, 3,
L.
Fig. 2.2.
and 4 have zero displacements (the general case follows by super-

position). Calculating these stresses and then taking the averages gives,
ff =
- 4(l r2)ab{-bUl-VaV>)
E
a»i) (2.1)
°'» - 4(1 - v»)ab ( "*"•
E
1 A
T
-v.v - 4(1 - v > i ( A,flWl - --b>.)
where
v = Poisson's ratio
Ax - (1 - »0/2
These average values for stresses as given by the four component
triangles will be assumed to apply at node 5, the centroid of the
circumscribing rectangle.
215
M. J. TURNER, H. C. MARTIN and R. C. WEIKEL
Next we determine nodal forces at node 5 by applying Eq. (2.b) for

triangles I, I I , and IV. This gives,
vm _
5
4(1
Et
- v2)ab
[--,^ + (-2^ + ^ , 4
V3* 2
yd) =
4(1 - i f ah [(" 2vab +
a2 + X J ) "* ~ a * V ]
E t
vdi) _
5
4(1 i f ah
(2.2)
2
*" - 4(T^ ) a b l M ^ + H
AjaA*
«,v, = 4(r^>- [--* + ( 2vaA + At.2 + b2) * ]
vdv) = *&
5
4(1 - ••••)«/,.
where A2 = (1 + v)/2.
These nodal forces may now be used to calculate stresses. Two
possibilities exist and these are illustrated by Fig. 2.3. We have,
AI1- = AT + x<5"\ n u = n" + n ll)
xf> = x f + AP, IVI , F<52> = yd> + y a V )
(2)
(2)
IV
-*X5
^
Fig. 2.3.
Now we define stresses associated with these nodal forces as follows:

a ^ = normal stress on diagonal 1-3
-X^ cos 0 + r , " sin 0
.•Va! • b2
(11
6 = shear stress along diagonal 1 -3
-AT sin 0 F<,->cosO
=
t \ a- ' b*
216
where positive shearing stress acts from 3 to 1 and 2 to 4, and

a12* = normal stress on diagonal 2-4
A T cos Q + Fi 2 ' sin 6
tVo* + b*
i2)
T = shear stress along diagonal 2-4
- X f > sin 9 + Y f cos 6
tVa* + b 2
where t = thickness of sheet.
cos 0 = b/Va 2 + b2, sin 6 = a/Va 2 + b2
The stresses given above can now be used to plot a Mohr's circle,
which will give the complete stress state existing at node 5. If this is
done and if ax, ay, and TXV are then determined with the aid of Mohr's
circle, it will be discovered that the results are precisely those obtained
by averaging stresses for the four triangles as given in Eqs. (2.1).
Therefore for rectangular elements composed of triangles as in Fig.
2.2, the stress averaging process is seen to give identical results to the
procedure employing nodal forces. As stated earlier a superposition
of states similar to that discussed above will provide the proof for the
case of general nodal displacements.
Stresses based on the averaging process will be found to give results
which arc: (a) not subject to the large discontinuities existing for the
individual triangles, and (b) subject to increasing accuracy as the
nodal network is refined. To a large extent this process removes
the objections normally raised with regard to stress calculations by the
direct-stiffness method.
A similar procedure can be used when the geometry dictates the use
of quadrilateral elements of more general form. However, the equiva-
lence between stress averaging and the use of nodal forces to determine
stresses has not been investigated for this general case. It is suspected
that the equivalence docs not exist in this case.
Nodal forces are at times to be preferred to stress averaging. This
is particularly true when stresses are being found on a boundary. In
view of this some additional remarks will be presented on the use of
nodal forces in finding stresses.
Consider the problem of finding stresses at node 1 in Fig. 2.4. The
nodal forces at 1, acting over areas \{a f- b)t and \{c + d)t will then
yield the desired stresses. First consider a vertical cut along 2 - 1 - 3 .
From the four triangles to the left having node 1 in common we obtain,
X u = A-} + XY + X\ 11 + A T
Ylx = Y \ + F} 1 + F { " + Y\ v
217
M. J . TURNER, H. C. MARTIN and R. C. WEIKEL
,2
Fig. 2.4.
Equal but opposite forces could have been obtained from the forces to
the right of the vertical section. It must be recognized, however, that
stress discontinuities will occur in wing analysis at lines of intersection
of wing skins and spar webs. Next consider a horizontal cut along
4 - 1 - 5 . Then the four triangles below this line may be used to obtain,
X - x\n + AT + xj + x r
Yly = Y\u + Y\v + Y J + F}'1
Again equal but opposite forces could have been obtained from the
four triangles lying above the horizontal section.
Using the above forces, stresses at 1 are then given by
X.
v
lx
°lx ~ ${c + d)t
F,In
aly = (2.3)
i(« + b)t
A Iv •la- \r
'liy I +
" * &_\{a + b)t \{c + d)t] {a + b)t {c + d)t
If a nonrectangular nodal pattern is necessary a modification of the
above procedure may be used. This situation is illustrated by Fig. 2.5.
Stresses are desired at node 1.
Segments 2-1 and 1-3 can be considered as chords of a curved
section tangent to angle bisector at node 1.
Separate the triangles to the left of 2-1-3 as shown in part (b) of
Fig. 2.5. The forces normal and tangential to a line bisecting the angle
between lines 1-2 and 1-3 are then F, and F, . These are given by
the expressions,
FK (aVj • A] 1 i A]") cos 0 ( (F{ + F} 1 + FJ") sin 6
PM
F.] = {X\ + A} 1 + AJ") sin 6 + (FJ + Y\ l + Y\ u ) cos 0
218
-V-
111 I lv *
(A) (B)
Fig. 2.5.
The corresponding stress components arc then,
F
a = >"
'" K'I-I + 'i-s)-* (2.5)
'' Wl-M + ll-*)*

where lx_2 — length of line 1-2, etc.
In a similar manner stresses can be found on another section through
node 1. When this has been done Mohr's circle can again be con-
structed. The complete state of stress at 1 then becomes known.
The stress averaging procedure is obviously better suited to pro-
gramming for machine computation than the nodal force method.
However, the nodal force approach is useful for hand computation of
stresses based on nodal displacements in an irregular network and on
boundary or line of symmetry.
The information briefly sketched up to this point will enable the
analyst to devise those special improvisations which are necessary in
coping with special situations.
Several examples will now be discussed.
Examples—Stresses
Three examples will be given. The first is for a simple sheet subjected
to parabolic edge loading. The second and third problems represent
composite structures of moderate complexity. In each case compara-
tive data are available for checking results given by the stiffness method.
Sheet under parabolic loading. The problem is shown in Fig. 2.6 and
has been previously discussed by Timoshenko and Goodicr. 4
219
M. J . T U R N E R , H . C. MARTIN a n d R. C. WEIKEL
Due to symmetry only one quadrant need be considered in using the

stiffness method. The applicable boundary conditions are as follows:
u = 0, T« = 0 on x = 0
v = 0, Txv - 0 on j = 0
..-si f" ,,2 I* Tj,, = 0 on x = a (2-c)
a„ = 0, = 0 on y = a
T h e nodal network used in the analysis is shown in Fig. 2.6. Thirty-six
triangles making up a total of nine rectangles are used to represent the
quadrant of the plate.
••-X
•^-^va^-q
T
Fig. 2.6.
Distributed loading between nodes is replaced by nodal forces (along

edge 4, . . ., 16). This is done by taking the total distributed load
between two adjacent nodes and replacing it by corresponding nodal
forces having the same centroid and total magnitude as the distributed
loading.
A total of 25 nodes are used to represent the quadrant. Therefore a
total of 50 components of nodal displacement exist. Equations (2.c)
will reduce these by 8; hence, the total stiffness matrix for the quadrant
will be of order 42 X 42. This is obtained by properly superimposing
the stiffness matrices for the individual triangles.
Stresses have been computed along each axis by using the nodal
force concept developed previously in this section. Actual values are
based on E = 30 X 10 s psi and v = 0.30. Timoshenko and Goodier
present an approximate solution based on the method of least work.
Comparative results are shown in Figs. 2.7a, b. Deflexion data were,
of course, also obtained but they will not be included here.
220
The agreement evidenced in Figs. 2.7a, b is regarded as satisfactory.

Timoshenko and Goodier used three terms in a series for the stress
function to obtain their solution. As a result both answers are approxi-
mate. An exact solution is unknown.
Ring-stiffened circular cylinder. Next we examine the problem of a
ring-stiffened circular cylinder. This has been investigated experi-
mentally by Kuhn, Dubcrg, and Griffith.5 Theoretical analyses have
been reported by Hoff, Jensen, and Denke. 6 ' 7 i 8
Fig. 2.7a. ox/S versusy on x = 0
STIFFNESS METHOD
LEAST WORK
*/(a/3)
Fig. 2.7b. ayjS versus x ony = 0.

221
The cylinder and loading arc shown in Fig. 2.8, while the nodal
network chosen for the stiffness method solution is shown in Fig. 2.9.
As shown by Figs. 2.8 and 2.9 the circular cylinder is replaced for
purposes of analysis by a polygonal cylinder with twelve flat sides. The
stiffening rings are replaced by regular twelve-sided polygonal frames.
Straight-frame members have the same axial stiffness and moment of
BAY 1 2 3 4
A A
_J
1000 Ib. « 15 • - 15
15
1000 Ib. RING
p-i.oo-H
.032
.78
_i_
S»et. A-A Through Ring
Fig. 2.8. Cylindrical shell with ring stiffeners.
Fig. 2.9. Idealization of ring-stiffened cylinder.

222
inertia in bending as the circular rings. Torsional and out-of-plane

bending deflexions of the rings are neglected. Ccnterlines of the poly-
gonal frame members have vertices which lie on the locus of centroids
of the original frame. The flat rectangular sheets used to approximate
the cylinder arc broken up into triangles as shown in Fig. 2.9.
Due to symmetry only half the structure need be analyzed. The
nodes at the clamped end have zero displacements and hence may be
eliminated from the total set of force-displacement equations; that is,
from the overall stiffness matrix. However, special consideration must
be given to the points lying at the centroids of the rectangles. Dis-
placements normal to such panels are not considered in forming the
stiffness matrix from the inscribed triangles. As a result transformation
to the reference axes will give rise to difficulties which will not be
Table 2 . 1
Nodal Displacements {in.) Ring

Bending
Node
Moment
lOu \0v lOai (in.-lbs)
1 -0.018 -0.601 0 372

2 + 0.009 -0.364 0.064 280
3 0.068 + 0.062 0.490 -3
4 0.101 0.206 1.030 -447
5 0.045 0.223 0.970 805
6 -0.107 0.946 0.251 :,'l:,
7 0.220 1.903 0 + 2230
8 0.017 -0.397 0 281
9 + 0.009 0.220 0.047 198
1(1 0.066 0.081 0.349 -57
II 0.094 0.165 0.665 -381
12 0.037 0.204 0.519 -413
13 -0.101 0.627 0.096 163
14 -0.192 0.988 0 698
15 -0.014 -0.203 0 176
16 0.009 -0.097 0.029 106
17 0.056 0.070 0.196 -75
IK 0.073 0.104 0.330 -218
19 0.019 0.134 0.214 -124
20 0.082 0.319 0.030 + 122
21 -0.135 0.441 0 201
22 0.008 -0.057 0 68
23 0.006 -0.018 0.010 32
24 0.034 0.037 0.067 -41
25 0.040 0.042 0.097 -68
26 0.006 0.053 0.050 -13
27 0.046 0.102 0.003 + 36
28 -0.071 0.130 0 39
223
discussed here. This difficulty is most easily handled by eliminating

nodal displacements at the centroids of the rectangles prior to trans-
formation of the stiffness matrix for the rectangle to reference axes.
If this is done the problem will involve 28 active nodes. These arc
shown in Fig. 2.9. At each node four displacement components must
be considered. Two of these are in the plane of the sheet and involve
the sheet stiffness plus axial stiffness of the frame segment. The other
two represent translation and rotation in bending ofthe frame segment.
As a result a total of 112 degrees of freedom exist for this problem.
Symmetry provides some reduction in degrees of freedom. For
example, nodes lying in the x-y plane of symmetry will remain in that
plane after loading has occurred. Likewise, these nodes will not
experience rotation in bending. Hence the degrees of freedom will be
reduced from 112 to 96.
Numerical data are based on E = 10.6 X 106 psi and v == 0.33.
Nodal displacements and ring bending moments are given in Table
2.1. A positive bending moment induces compressive stresses in the
outer-ring fibres.
Ring displacements are shown in Fig. 2.10, while ring-bending
moments are given in Figs. 2.11 and 2.1 la. Angle 0 and ring identifi-
cation are shown on Fig. 2.8. Agreement between measured and
calculated moments, as shown on Fig. 2.11, is seen to be quite good.
Calculated stresses existing in each triangular element are shown in
Fig. 2.12. These data are listed in detail to exhibit some o f t h e com-
ments made earlier in this section. Although at first glance the stress
1/10 IN
DISPLACEMENT
SCALE
Fig. 2.10. In-plane ring deflexions.

224
3000
2500
2000-
M,
RING 1 - MEASURED _©_
in-lbs ]5oo.
COMPUTED -4-
1000
500 4
120 150 180
-500 * DEG.
Fig. 2.11. Ring bending moments.
RING 2 - MEASURED
COMPUTED
RING 3 - MEASURED
COMPUTED
Fig. 2.1 la. Ring bending moments.
discontinuities between adjacent triangles seem quite erratic, closer

observation will show that a consistent trend does occur in groups of
corresponding triangles. As discussed previously, average values of
stresses for the triangles within a given rectangle will be assumed to
act at the centroid of the rectangle.
Shear stresses are then plotted in Fig. 2.13. These are considered
constant within each rectangle.
225
M. J . T U R N E R , H . 0 . MARTIN a n d R. C. W E I K E L
9 15 22
(J2 229 415 525
40^- s ^ / - " ^ 5 9 iie\ ^ - 6 0 ^ - " U 8 1 3 > - ""159 69
-49 15 -155 137 -254 284
-19 • ^ r t ^
-55
- i
-60
-78
^r -510
-<58>-
-25
-690
-551
- 7 6 ^ \ - 4 6 / ^ < 7 6 - 5 8 > - \ - l > - -"-385 -93>K "<950 •1550v \ 1 5 > - •^560

• 4 -97 195 -222 506 •^JP" -529 740 -450
^<15i> "^69-4^ ^550*- <242V
-46 -14 88 160
-205 -640 -1740 -2660
-M^ \ 1 3 3 ^ - ^ 5 ^ 3 - U 9 0 ^ \ 4 2 J ^ -<fl90 -21t>9v \ 5 9 ° " -2690 •286IK VaSP"-2870
227 38 658 182 850 550 750 480
<2450*- ^5OTT>V
152 418 590 620
-560 -1610 -2590 -2660
- 6 2 6 ^ \ 6 8 C i > . < - 6 2 1 -1500-- \ 9 9 J - - <1500 - I 7 5 V ^ 9 5 > ••1750 •1780^
722 656 920 1073 500 1570 130
\55°" -1790
" < 7 1 > - 1590
678 994 950 860
-655 -llSo -670 -90
-199^ •v^fia^ -<166 - ^ 9 0 2 / -" 51 870- ^ 6 4 0 / ' ^ 8 7 0 157fr- \ 6 m - "1570
1102 1728 111 1694 -400 1690 -510 1740
^269^ <i220^- •»^4lb^ <525b^
1412 905 640 610
4?6 1750 2890 3770
U9C> V 4 0 5 O , •"1210 2910^- ^ 1 2 0 / ^ 2 9 0 0 3 7 8 >
560 '-1900'"-* 1520 -670 918 -450
\J5P-" "2770 4 5 5 0 "
760 -540
4560
"466V -"556>* 750
1040 126 160 190
14 21 26
Order of L i s t i n g : ( ax lbs/in2
1 7- X 0 l b s / i r 2
T*
d
4 0
,
+ 0> I1
+ x -a»
Fig. 2.12. Computed stresses in triangular shell elements.
Measured shear stresses are also shown in Figs. 2.13 and 2.13a.
These were obtained as reported in Ref. 5, at two stations located 2 in.
inboard of rings 1 and 3 respectively. Normal stresses were measured
at the station adjacent to ring 3. Also measured stresses were obtained
at a number of values for 6 at the two stations.
Normal stresses, a x , were likewise plotted for nodal positions corres-
ponding to centroids of rectangles. These plots, not shown here, were
then used to obtain stresses at the rings. These data are given in Figs.
2.14 and 2.14a.
226
1500-
COMPUTED -
MEASURED
x e
Fig. 2.13. Shear stress between ring I and ring 2.
1000
lb,/in
'x«
-250
« DEG
Fig. 2.13a. Shear stress between ring 3 and ring 4.
Although, in this example, a rather coarse representation for the

skin was employed, the use of stress averaging is seen to give results
which portray the trends and values of experimental data in a con-
sistent and reasonably satisfactory manner.
Sweptback wing model. To further illustrate the application of the
direct-stiffness method in calculating stresses, a sweptback model wing
will be treated as a final example. The wing which will be studied is
that investigated experimentally and theoretically (force method) by
Eggwertz and Noton. 9 - 10 These investigators obtained deflexion and
strain data on a model structure having a moderate degree of com-
plexity.
Figures 2.15 and 2.16 show the model. Material used in the structure
is 24S-T. Dimensions have been converted to inches. The wing has
227
Y in Y in
I0" 3 (i x lb./in 2
Fig. 2.14. Longitudinal stresses in shell.
-3 2
10 a v lbs/in.
RING 4 BUILT IN END

Fig. 2.14a. Longitudinal stresses in shell.
11.81
Fig. 2.15. Planform and dimensions of model.

228
30 degrees sweepback and is untapered throughout. Five identical

spars and three identical ribs are bonded to top and bottom cover
skins which run continuously from wing tip to wing tip.
Support is provided by a solid, rectangular centre member. Loading
is shown in Fig. 2.17. Support and loading arc such that the semi-span
can be considered as a cantilever supported 1.535 in. outboard of the
vertical plane of symmetry (see Fig. 2.15).
Cover skins arc assumed to carry plane stress only; that is, their
bending stiffness is neglected. Representation of the top cover skin by
39 rectangles and 4 triangles is shown in Fig. 2.18. As in the previous
examples, each rectangle is further decomposed into 4 triangular
1 __
-2.953- O.IIE
(TYP.) -0.059 2.48
d 1-
11.81-
TYPICAL CHORDWISE SECTION THROUGH MODEL
—
.197 r
.197
.157- -.157 .157- .157
DETAILS OF ATTACHMENT OF SPAR AND RIB WEBS TO COVER SKIN OF MODEL

F i g . 2.16. Structural details o f model.
1/2 Ib
1/2 Ib.
LOADING CONDITION N O . 1 LOADING CONDITION N O . 2
Fig. 2.17. Loading conditions.

229
>*> -,1
% — J U, -<
,.65 ,-o* -87 {
•? •ir • * •
x« A"
9 yr ?\i B
7T7 >7'
Eh
y» XT» >*
IV Ys
,*7» ;<n
15
V
r.sv. r
4» -67
\«0
.'**
T
ii
x«
5 *
| « - S.MJ — a )
5
i
. BraaJedown and lloda Jtuabarlnc Sjrataat
230
DEVELOPMENT AND A P P L I C A T I O N S OF T H E STIFFNESS METHOD
elements. Since a horizontal plane of symmetry exists, only the upper

cover skin need be considered.
Spar and rib webs are treated as pure-shcar-carrying members.
Chord members are assumed to carry only axial loads. Therefore
conventional beam bending stiffness matrices are not used in this
example. Spar webs are assumed to carry uniform shear stress between
ribs and between the inboard rib and the root section. Likewise, rib
webs are assumed to carry pure shear between adjacent spars.
Of interest here are cover plate stresses in the root region. Con-
sequently attention will be restricted to the area inboard of the section
joining nodes 31 and 35, Fig. 2.18.
Nodal displacements, from which cover skin stresses may be deter-
mined, arc given in Table 2.2. Vertical deflexions along the front and
rear spars are given in Fig. 2.19 which also includes experimental data.
Y — INCHES FROM TIP
Fig. 2.19. Deflexions—load condition No. 2.
Stresses existing within each triangular element of the root region

are shown in Fig. 2.20. This figure also shows the coordinate system
which will be used in the discussion to follow.
Nodal forces for the same inboard region are listed in Fig. 2.21. As
can be seen, the listing in this figure is complete in that every node is
included.
Table 2.2 also lists stresses at the centroids of the rectangles. These
have been obtained by stress averaging. Since no particular difficulty
arises in obtaining these values, they will not be discussed further.
Nodes lying on the spars require special treatment. T o illustrate the
231
2£S
» ECC/o- OSFO~ L&

6FII- I88'0- 96
8W0- 5is - o- 56
FSi'I- 922M- W
EES'I- 688"0- E6
?a. oioi- 6I5-Q- 26
EFS2~ 9FF4- 16
3 LWZ~ £214- (Hi
9£2-2- 68
I- 999 I -
66J1*0-
22F'0~ 88
s. 69S£- SSF4- LS
619E- 2214- 98
1I9E- osro- 68
1• 0I9-£- IIE'O- F8
0 0 0 65
0 0 0 85
0 0 0 /IS
0 0 0 9C
0 0 0 ss
0920 90S'0- £8£'0- w
0 0 0 £5
8i.ro— 9104- 9£8'0- 25
F8£Q- 6/190- £8F0- IS
0 0 0 OS
EEEI- 8t*Sl- 922'!- 6F
689'I IFE I - t-06'0- 8F
I60M- 0F6'0- EISO- /IF
0 0 0 9i
2E2-E- 860'2- 1154- SF
£26'£- eiZO'z- F8II - Ft
I2F£~ 108" I — 8S8'0- EF
F002- 2S24~ 08F0- 2F
0 0 0 IF
ZWL- I66-2- 289'I OF
££6'8 690-E- LZM'\- 6£
i*6"8- EFO'E- t-IO i - 8E
L6L'L~ 0F8'2- 0990- L£
S69S~ 9992- OEOTJ- 9E
699E1- 090 • F - 5854- S£
69091~ ^Fn•- G224- f-£
-
6£6'9I- IS2 F- 0S8'0- E£
069'9I EEEF- EFFO" 2E
016SI~ I6FF- 8000 IE
17? (7 n >P°N
{trOI X •«.)
x uoijtpuoj pvcrj—s}usw93V}dst(j ppofsj 'Z'Z W J .
13-MI3M *D "& p u c m i H V i v *o *H ' n a n u n x *f w

x t*- /
\ 1 -J X \ 1 H- / B
\ O ~4 ro y \ vfi • * - * > / x
\ vî \ > * 7 / <
1 ON \
Q SO -O
\|«5/ / ] - 4 1 t \0 \
X ro M O •
V - i CD m
/ * - * vO H
• • > . / • • • \
x
-
CD CD O / \ f - J CD —J VJI VJ** / \ VDCD N 0
tM ro o /
/
\ V J < CD v^« 0 * 0 - 4 /
o\V /( -JtV
\vO-J f
1 G9 \
-
i.
/ ro f f \ /•JH ro\ V» OviO \ x
I M • T H W \
/ SP Q \ / r o CD V M \
novo \ AT
I- / O vji >j. \
\ 1 CN /
\ • ^ X\ -** /
\ Q \ o t?v X
>
1 Wl X
\SI-sX \ glJ-x
\—J -J V j - /
. / vj. 1 C N \ yi ON 1
\\ \
ON \
j C,D H-»
< y Nro
vo/ /
y i OV
y
D
••> •03 •f \ \ / X*f C • D •CN
*-t \ Q \ & Y . ^j-, \ j \ _ j H H ^ \ / \ j * t \ i . ^ O C N G P \ / O —J - f
I---J CO / \ 1 - ' VjH V>a
V^-^CP X >vO Ov> f f X f O H
v-* O f / X. ro vo i r -o o co X \ *> ^ •*••
A Vj. \ A \T- \
/ I vj» \ 1 VJl \ f
X ^M \>. CN \ X H - V . ON \
/
/ vji O
\ j i v>.
ON
Q>
\
\
ON N) \B \ r.
/ VaO VJl VJl X
vo ro -~~\ \ >
ro 1 •*» Q S V i vff /
-.
\ \Si /
\ vî X\ \J* s
M S" fl "<* * \ t— iv'w- / 0
\ o\* V X \ $ & % / \ | 2 g / /
1 *" V jf f \ -J-J f / \ f - J M/ WS
cr
/ H ' J J v . n H \ D V\r J
*" x ro A
ro f \ > / *" f \ X *>0
*" ro *" \ / \r i
f O VO V ^ \ o> \ / 0 \ 0 f VJ« CD f \ . X v > V-Mv o ro \ X vw O O c
• «—v-p o V CD 6 o t!°^ X •" ^ ^ o -^v* X O H - > vjn
CNO H / \ a\0«w.y CO / X ro f r o
iw v> X \ j n o t5v
a*. f O —J N i / \ CD C N \ > . VO O M / \ - ^ J 0 5 ON
-
3 / -rX / *" X / *" \ / f \ f l\ -.
/
X-— CN f 1
fO O v *
\
\
/
X Q vji -Jt \
« £ v j l VJl \ / - r v j VJd \
X ro H O \ r o O v « \
H ON-J \
-
/ V J I H iy \ / VJF! O ON \ / -J o * \ H !-• vO \ x
\ * *" / \ X\
\ O \Ji fO X \ fO f M X \ f MH X
*" \
f
CNO O
y\ X
/
\
f 1
V H »-• X
*" X \
\ X
\ 0 i \ 0 ^ / \ - r p o y \ ON—J H - /
\ \ \ >oi vo
—J P• \ f vJ M / X
\ v j i \ j \ ro / \ r o CD v j * / \0N-*J \ J \ / \J*S /
\>* \ / V^* \ v o o rvX \ X
D \ K N \ / Q 1 \ j \ ro » • • \ /• • • **\ X *•
v^ \ / W. 1 X
J i CC \ > i Y U CC\aT \ 3 H - Q \ / r O t - ' l - ' f CC H \ / f CD H iJi CTv O \ / vji ON O*Ui ON Q \ X VJ. ON H
O CD O N / \ rovr O O ro VaD V ro O vo -Jv-n*-" X C N f r o - o o v c N X *-JV£>O*
\> tywo / \ o ro >*** O H O / \ vo - J r o a . * - H / \ a j , < h f H ^ M / \ O ON c
NJ H ON X ro f o \
/ \
A *" X \J*
X M CO H- \
.X ^^ X X VJ* l \/ ^^ \ ro i >. r:
X
/ —3 ro co \
/ f f H \ / vp i v ) o \ X f »-• H \ rovpvM \
/ CD CO f \ / O O CO \ / fCBH \ f V O C J H X »-• vO - J X -
/ \m* O - J \
/ -J H^/l X r o vO O >v M\rf«VJl \ X
z
r
11
J. TURNER, H . C. MARTIN a n d R. C. WEIKEL
Iod«i POI-CM Acting an Trltiagalu* FUMnntc

S5
Load Condition l o . 2 Ordar-of Liatlngi
( \ '"••
/0OO5
1 Tn ""•
/.0378
• < -.0212
/.
0 5 6 e \ -.0575
/ .5215 V^. 2 501
/JOO05 -.0573*V
53 / - . 2 7 1 » -.2501 \
V . .0535 .0375 S 54
V.291. .5.57/
/T0007 .055.V -0906 /C0575
52 2 5
/ .3644 .3559 X " * " / " ^ • "
57/.056I -.0112V/.0110
>C-.0212 ,0552^V-.o»95
/ . 0 5 6 a \ -.0375 -.0.22 / . 0 9 0 7 V —0683
/ .7076 \ - . ) J , } 2 - 3 8 9 1 / ^ ,629» V—2**1
/o007 -0575\ yC»t, -.oksaP^
90 / - . 5 6 4 4 -.5*52 V /-.3409 -.2685 V 5?
V .°545 • IN > V \ 2.0516 930
.OW*
.5506/
J
/Toon .0547V —0625VM6» .4545 / ^ .051>\ -.1001 / 0 4 8 4
/ / o 2 8 0
.50}? .1.1*4 \ - . 8 7 . 2 7 . / - . 5 7 1 1 .3250 V ^ - 6 . 3 9 S .5135
58 X<1097 .ooeK ./ooej .ooiv /Tool 5
/X.0O69 .0447_X—0274
• 06O6VC-0551
S - , oi2oV .oioa -.0.28 > / o 6 2 6 X -.0195 -.0532 / f l O O T V —°»70
_ / 1.0145 \ ^ . 5 i o 6 - . 5 2 9 1 / .855. \ - 3 » 5 < - . 3 8 5 6 ^ / .6585 V^,2 5 > 5
/oil .Oloe\ /042J -.0197V /-.0552 —0»70V
46 / - . 5 0 I 7
\-.u«j
-5106
.-i?j
\ /—4458
y V, -".'O
-..096
.-><~
\/
V
—JJ^l
.03.7
-.2902
.0.70
,,
X
/
1 X-4129 .4433 x ' X . .3U5 .,jSt/
/0017 X * " .5889/^
~ 6 8 V ) \ .0075 / T o i 7 4 .0148V -.0526 -/038I .05118V -.01)17 /C0.70
/ .B70S Nl.1821 / .1,715 ..530 V ~ , 5 « i -^ ,j 9 1 2 .5491 X . . 6 7 0 ? / .5717
.065V / < 0 6 5 7 .0336X/.0338 .011K / . 0 1 1 7
.OOo^X. .0167 .0465T>V-'0*76
.0564
/ 0 5>X—0245
7 / - 0 /9. 70 5. 6\ 5 -OWl - * "
0 1
-O^S -.0484 28V -.00.5 -.0464 / \ o e i e \ —0553
*. . /
/ M \
1.5970 V - ' * ' - . 6 , 3 9 / - ; ^ X ^ * * ..4»94 / ^ .8442 V^.)M7 - . 3 9 6 0 / .6709 \ — 2742
/.0017 .096K /0595 .046l\ /0485 —0044*X / V . 0464 -.0354\^
/ -.6702 -.7267 \ /-.JT'*
u V . -.iysa .00111 > V -.0001
-.WO.
x'.^l
\ S -Ml,
V -.0559
-4068 V J /
.0367 / V,
-.3591
.0049
-3118 V
.OJ34 / 45
X.88H , 7 1 9 4 / X-5646 .5172/ \AJ18 .4|42 - ^ \ ^ .3412 .3736/
— UOoS. .1913 P?03l4-.0672V -05M /<0li05 ..0262V. -.0028 ./o464 .0134X -.0385 /.0574
1.0653 \ - 1 . 6 a » S / .5655 .6172 Xf-OOlS / , l * a .1,626 V - , 8 6 6 0 ^ ^ .4015 .5662 \ - . 7 1 4 8 / / ,5526
.09()*V/a904 .095&v/a245 .0650V/—0825 .0152X-'—0277
-.157!VX'l078 -.0355/\.0567 ,OlO6/V-,0Oee .055i/\-0597
.052? / x .1590 ..0276 / 0 2 6 6 \ •°»J» -.0388 /.0201V ,0360 -.0287 / ^ . 0 5 0 e \ -.0096
-•5«59/i.07«F, X.4979 - . 4 7 5 5 , / .86*2 X - 5 9 2 7 —4018_/ .7169 \ - . 3 1 2 9
/.0027 .096&V /J444 .0710V. /»84 .02BVV /^-.0527 —OIBIX
/ -7753 -6571 V / — 5 2 8 0 -.5505 \ /-.4406 .1,236 V/ -.5809 -5380 X 40
56 V. — 0153 -.027. >•C -urna :moi—"•? V -0554 .OWI sV -.0229 .0139 /
X.6434 .6073/ \.543i W l / X.4412 .4156 , / V . .4032 .5956/
-.019V. 1 .0427 J
,/oo54 -.072!K .0765 /01B3 .,0.28V .0527 / o 2 1 7 --0221X .0090 / C 0 1 3 8
.6763 \ - - 5 0 7 ^ ' ,5562 .5916 Xj.0197 . / .4558 ..668 \ — 8 5 4 8 X ,5968 .4041 \ — 7 9 8 8 1 / .5955
.031V / o 5 9 6 .074X/To9«4 .0556v/To«5l .0357V/-.0360
-.0070 >C .0*12 -.0599VS. .°525 -.0271 >-v ,018. — 0 0 7 8 / X 'OOTl
-.0115 j r V . ,05.2 **°°" 5,1 / 0 5 W \ - « J l -.0107 / T r a u V •<*3* -.0152 / - . O O 8 6 X -0222
- . 6 7 1 5 / . .Jo, i « . \ - - " 9 * • /l.0274 V~48B1 . . . 3 9 6 / .8656 V-,1,152 - . 5 9 6 4 / .7994 \ _ . 4 0 2 4
/ o 4 4 3 ' - ' 5 .058K /0096 .0636 V . /oil7 .0528 V^ >^-.015l .0217X
/ —6021 -.611,5 \/-4908 -.5)66 V / -..229 -.1,407 V/ -.5961 -.0405 X —aa
Fig. 2.21.
procedure Fig. 2.22 takes u p in detail the calculation of stresses at

node 42.
By considering a section parallel to the *-axis and passing through
node 42, nodal forces X i 2 , Y i2 may be used to determine ay and TXV.
Similarly, nodal forces at 42 may be applied to a section parallel to
the j - a x i s . This enables a x and TXII to be calculated. As seen from
Fig. 2.22 (bottom sketch) a unique result is obtained for a x , while r m
is dependent on choice of section for determining nodal forces.
Next we consider a section parallel to the x-axis. This permits a y
and r xy to be determined. This time the result for ay is not unique,
while Txy turns out to be unique.
This varied set of stress data is due to the structural idealization
234
4o 47
T - .114 ,, .... 2
xy " " .118 . 2.95 - - - 3 3 l b * / , n
41 43 2.724
»y .118 « 2.95
• 7.33 \ W m
Y . , - 2.447 lbs. T = - .33 l b s / i n 2

x
y
X., - - . 1 1 4 lb.
" .ni?'«^.9j - 7 - 1 0 lb
-/i"2
46i^ n47
Cx - .65 I b i / i n
T
. „ " - •*> l b l / i n 2
Fig. 2.22. Determination of stresses from nodal forces.
employed in applying the direct-stiffness method. Usually spar webs

are continuously attached to cover skins. This results in cover-skin
stresses which arc continuous in normal stresses but discontinuous in
shear stresses at the spar webs. In this instance, however, the spar caps
are discrete members and therefore introduce j-components of force
at nodes such as 42. As a result a y can be expected to be discontinuous
at a section parallel to the *-axis at node 42.
A study of Fig. 2.22 will enable the reader to verify these trends. In
the present stress calculations r xy will be chosen as the value obtained
from sections parallel to the x-axis; ay will be chosen as the average
value given by these same sections. A section parallel to the j-axis will
give ax. At node 42 we then have,
a x = 0.65 psi, fr„ — 7.46 psi, r xy — —0.33 psi
A modification of the above procedure is necessary if we are to find

the stresses at a node lying on the edge of the cover skin. Node 41,
Fig. 2.18, represents such a node. Stress calculations in this case are
carried out with the aid of Figs. 2.23, 2.23a, and 2.23b. This will be
discussed in some detail.
First by using Fig. 2.23, plus the data in Fig. 2.21, it is found that
VCD 1.946 1b.
X% = - 0 . 3 2 4 lb, 1
41
These arc assumed to be stress resultants acting over one-half the length
of the rectangle edges meeting at node 41. Furthermore, a x is assumed
to be zero at the left-hand edge (boundary condition). Consequently
Hi 235
Fig. 2.23. Calculation of r n and a y at node 41 from nodal forces.
Fig. 2.23a. Calculation offerees at node 41 due to stresses on oblique root section.
46
Fig. 2.23b. Verification of a y and xxy at node 41.

236
the stresses equivalent to the above nodal forces are ay and TXV and these
are given by,
X™ = 1(2.953) (0.11 8 ) T „ , or T „ = - 1 . 8 6 lb/in. 2
F<4V - \(2.953)(0.118)a y -- J ( 2 . 4 9 5 ) ( 0 . 1 1 8 ) T W or ay = 9.60 lb/in. 2
Next we consider the nodal forces Xf,\ FJ- given by all the triangles
having node 41 in common. These are also obtained from Fig. 2.21
and are,
*,*•> = - 0 . 3 2 2 lb, y<2> = 1.276 lb.
Figure 2.23a shows these forces plus stresses <r0, T 0 on the inclined face
representing the fixed root of the wing. Finally, Fig. 2.23b shows nodal
forces as given by the upper triangle only at node 41. These are,
Xfl = 0.002 lb, y4V = - 0 . 6 7 0 lb.
With the aid of these figures the stresses given above {ax = 0, ay = 9.60,
TXV = —1.86) may be verified. Since measured values of a0, T 0 are not
available for comparison the theoretical values have not been deter-
mined. It will be found that no discontinuity in ay exists at a section
Table 2.3. Summary of Computed Stresses
From Nodal F trees Stress Averages

Node
{psi) {psi)
a* a. T,n "r o-. T

xy
36 0 8.69 0.20 0.36 8.93 0.45

37 0.79 7.07 -0.53 0.61 6.68 -0.38
38 0.62 5.43 -0.19 0.60 5.41 -0.17
39 0.36 4.62 -0.01 0.34 4.60 +0.04
40 0 4.14 0.16 0.20 4.28 0.12
41 0 9.60 -1.86 0.98 9.55 -0.87
42 0.91 7.45 -0.33 0.91 7.43 -0.41
43 0.63 5.55 0.014 0.59 5.56 -0.003
44 0-30 4.41 0.29 0.29 4.42 0.25
45 0 3.77 0.41 0.15 3.80 0.38
Ii, 0.23 7.04 -0.15 0.52 7.13 0.15
47 0.28 5.63 0.18 0.29 5.60 0.17
48 0.18 4.28 0.42 0.15 4.27 0.41
49 0 3.53 0.54 0.08 3.55 0.52
50 -0.35 5.09 0.19 -0.067 5.18 0.24
51 -0.10 1.10 0.46 0.079 4.10 0.44
52 0 3.44 0.56 -0.047 3.43 0.53
53 -0.54 3.69 0.30 - 0.26 3.71 0.37
54 I) 3.27 0.43 -0.22 3.25 0.41
237
J. TURNER, H . C. MARTIN a n d R. C. WEIKEL
through node 4 1 . T h e reason is that at this node the spar cap is

directly reacted by the fixed root condition at node 41.
Table 2.3 presents stresses at nodes lying on spars and at the fixed
end. These are listed as given both by stress averaging and from nodal
forces. The results obtained are seen to be in good agreement with
Table 2.4. Measured Stresses—Load Condition 2
Stress Components {psi)

•V V
{in.) {in.)
1A 0 4.72 0.57 13.2 0.23

2A 1.48 5.57 0.70 9.75 -0.93
3A 2.95 6.42 -1.55 7.39 -0.67
4A 4.42 7.26 0.14 6.74 -0.52
5A 5.90 8.12 -0.48 5.36 -0.22
6A 7.38 8.95 -0.97 4.74 -0.71
7A 8.85 9.80 0.32 3.95 0.13
8A 10.33 10.64 -0.21 3.50 0.40
9A 11.81 11.50 0.46 3.24 0.48
IB 0 3.15 -0.48 9.83 0.62
2B 1.48 4.00 0.90 8.76 -0.50
5B 5.90 6.53 -0.11 5.59 -0.06
6B 7.38 7.39 0.15 5.03 0.90
7B 8.85 8.23 -0.36 4.11 0.28
8B 10.33 9.08 -0.06 3.88 0.46
9B 11.81 9.92 -0.97 3.59 0.44
3C 2.95 4.72 -0.79 7.15 -0.53
4C1 4.42 4.72 0.20 6.36 -0.18
5C 5.90 4.72 0.08 5.55 -0.15
6C 7.38 4.72 -0.26 4.84 0.16
7(; 8.85 4.72 0.39 4.78 0.16
8C 10.33 4.72 -0.03 4.21 0.28
9C 11.81 4.72 0.27 3.80 0.33
ID 0 1.58 -0.24 7.90 0.74
2D 1.48 1.58 0.30 7.26 -0.12
3D 2.95 1.58 -0.40 6.36 -0.32
1.58 -0.12 5.90 -0.19
4D 4.42
5D 5.90 1.58 0.24
5.43 -0.23
7.38 1.58 0
5.08 -0.14
6D
1.58 0.38 5.00 -0.15
7D 8.85
1.58 -0.12 4.70
8D 10.33 0.05
1.58 4.79
•M) 11.81 0.08 0.11
5.52
IE 0 0 -0.64 0.77
6.19
2E 1.48 0 -0.13 0.85
5.86
3E 2.95 0 -0.39 0.59
5.52
5E 5.90 0 -0.46 0.55
5.41
7E 8.85 0 -0.15 0.37
5.59
9E 11.81 0 0.28 0.59
238
DEVF.I.OPMEN-I AND V I' P 1.1 C A T K) N S i l l Mil STIFFNESS METHOD
each other. The greatest discrepancies occur at the lateral edges of

the wing for the two smaller stresses {ax and T XU ). A finer nodal break-
down would have given values for a x much closer to zero than those
obtained in the present calculation.
Measured strains were obtained by Eggwertz and Noton in the root
region by means of strain gauges. Locations of these are shown on Fig.
2.24. Note the manner by which strain gauges may be located on this
figure. Measured data, converted to psi, arc then listed in Table 2.4.
This data applies to the outer-skin surface on the tension side of the
wing. The origin of the coordinate system, applicable to Table 2.4,
is at node 31 and positive directions are as shown on Fig. 2.20.
Figure 2.25 shows a comparison of calculated and measured ay
SPAJ!
SPAR
SPAR
E D C
Fig. 2.24. Location of strain gauge rosettes.
LBS/IN'
il II
5T r 6 7 8 9 10 11 12 13 14 15
Fig. 2.25. Comparison of measured and calculated values of oy on spar centerlines.

239
M. J . TURNER, H. C. MARTIN a n d R. C. WEIKEL
stresses along the five spars. Some explanation is necessary to the

proper interpretation of this figure.
First measured stresses have been reduced by the ratio 1.24/1.30 =
0.954. This accounts for the fact that the measured stresses exist on the
outer-wing surface while calculated stresses apply to the mid-plane of
the cover skin. Also, on Fig. 2.25 the curves have been drawn through
the experimentally determined points. Both measured and calculated
0 1 2 3 4 5 6 7 8 9 10 11 12
y INCHES
I'ig. 2.26. Comparison of measured and calculated values of a t between spars.
.1 I 1 1
COMPUTED
.7
V ALUtS
-fl
.6
•P \ '*•
y oat"
u
.4
.3
91 • AVERAGE OF MEASURED
.2 IS STRESSES BETWEEN LINE 7
AND L N E ' )
.1
Z
>
3
° a?!-
2 ) i ; 6
1
' i 1> 10 11 1I
K
-.i
1 'f
*- ,< -.2
-.3
ZT
1 COMPUTED AV ERAGE BETWEEN
-/-VALUES - Lll -JE 1 A N D LINE 3
-.4
s
-.5
\
-.6 \
-.7
i8 •
Fig. 2.27. Comparison of calculated mid measured shear stresses.

240
points arc indicated by the same symbol; however, calculated points

are identified by nodal numbers corresponding to Fig. 2.18, while
measured data is referenced to locations specified by Fig. 2.24.
Figure 2.26 presents data similar to that given on Fig. 2.25, except
locations at which stresses are given now lie between the spars.
Finally Fig. 2.27 gives a comparison of calculated and measured
shear stresses at locations which can be determined from previous
figures.
The agreement between calculated and experimental ay values is
seen to be good at all points except node 41. Poorer agreement is
observed for ax and r xy stress components. This is perhaps not too
surprising. These stresses are quite small and furthermore, the strain
measurements were made at points directly over the spars where
gradients in the shear stresses arc large. Eggwcrtz has noted 10 that the
experimentally determined shear stresses are likely to be in error by
much larger amounts than the normal stresses. Nevertheless, the
trends in calculated and measured shear stresses are similar.
3, T H E R M O E L A S T I C ANALYSIS AND ELASTIC STABILITY
At the Aachen meeting a proposed step-by-step procedure was described
for analyzing the behaviour of elastic structures under conditions in
which geometric non-linearities must be taken into account. In the
present section we shall discuss an application of that procedure in the
analysis of deflexions and free-vibration characteristics of a heated wing
structure. Two additional applications in elastic stability analysis of
a plate and a tapered column will also be discussed.
Typically the structure is held in an initial state of stress by a system
of nodal constraints; initial stresses may be generated by forcibly-
joining a set of mismatched components or by nonuniform heating
while the structure is restrained against deformation. We then seek to
determine the equilibrium configuration of the structure and corres-
ponding internal loads as the constraints are relaxed; in addition the
relaxation of constraints may be followed by application of external
loads and further temperature changes or both. All of these processes
are assumed to take place at a slow enough rate so that the structure
passes through a sequence of static thermoclastic equilibrium states.
In general it is assumed that the loading (or constraint relaxation)
and heating process is subdivided into a scries of small steps to permit
linearization. The stillness matrix for a structural element in an initial
state of stress is typically assembled as a sum of two parts
A' = Ro + K1 (3.1)
This corresponds to the segregation of nodal force increments, resulting
from nodal displacements {Au}, into a set R°{Au} which are generated
by incremental strains and the remainder A 1 ! AM}, which arc of
241
M. J . T U R N E R , H . C. MARTIN a n d R. C. WEIKEI.
g e o m e t r i c origin, resulting from a n g u l a r d i s p l a c e m e n t of t h e initial

n o d a l forces.
Structural Elements
In analyzing the heated wing structure only two structural elements
h a v e b e e n e m p l o y e d — t h e axially l o a d e d stringer a n d t h e t r i a n g u l a r
p l a t e in p l a n e stress. Stiffness m a t r i c e s for t h e stringer a r e [Ref. 1 1 ,
p p . 101, E q . (10)]
X2
Xpt SYMMETRIC
AEr, Xv flV
A" (3.2)
-X2 All -Xv
—Xpi -V —ptv I'-
Xv —ptv — u2 Xr ll r
A-
(1 - X2)
-X h (1 - f ) SYMMETRIC
2
Xv -(IV (1 - v )
(1 A2) Xpi Xv (1 - X 2 )
2
Xu -(1 - n ) ptv -Xpi
?.v ptv -(1 - v 2

) -Xv —plV [1 - V2)
(3.3)
T h e m a t r i x of d i s p l a c e m e n t s referred to fixed Ar-^-z-axes is
{At/} = {An,., Av l t Aw x , Au 2 , Av 2 , Aw 2 } (3.4)

P a r a m e t e r s a p p e a r i n g in E q s . (3.2, 3.3) a r e defined as follows:
X, pi, v = d i r e c t i o n cosines of m e m b e r d i r e c t e d from 2 { x 2 , y 2 , z 2 ) to
P 0 = initial load in m e m b e r (tension positive) d u e to p r i o r

straining a n d heating = l,AETAe — 1,AETxAT
L = original l e n g t h of m e m b e r in stress-free state a t reference
temperature
E T = m o d u l u s of elasticity a t existing t e m p e r a t u r e
A = cross-sectional a r e a of m e m b e r
I n d e a l i n g w i t h t h e elastic t r i a n g l e it is c o n v e n i e n t t o e m p l o y a
system of jf-^-z-axes w h i c h m o v e w i t h t h e e l e m e n t , as i n d i c a t e d in
Figs. 3.1a a n d 3 . 1 b . T h e origin of this system is located a t n o d e 1
a n d t h e x-axis passes t h r o u g h 2 ; t h e o r i e n t a t i o n of t h e f-axis is t h e n
242
3 (53, y3, 0)
, 0, 0)
s
s
s
y
Fig. 3.1a. Position of local axes before incremental displacement.
3{; 3 +Au 3 , ?3 + Av 3 , 0)
/ s /.
s.i
Fig. 3.\h.' r Position of local axes after incremental displacement.
determined by requiring 3 to remain in the Jc-J'-plane. In the local

moving reference system the triangular element has three degrees of
freedom. The displacement matrix and the corresponding matrix of
applied loads are, respectively
{Ati} = {A«2) AM3, M.}
(3.5)
{AX} - {AX 2 , AX 3 , A ?,\
243
II, J . T U R N E R , H . C. MARTIN a n d R. C. W E I K E L
In addition to the Jc-.y-.T-axes it is also helpful to introduce a local

fixed reference system, consisting of the A-'-^'-z'-axes, which coincide
with the barred axes in their initial positions. The displacement matrix
referred to the primed system contains nine elements. It is of form
{Au'} = {Au{, Av[, Aw'j, Au2, Av'2, Aw2, Au!,, Av3, Aw3} (3.6)
By restricting displacement components to sufficiently small values the

orientation of the displaced Jf-v-z-trihedron may be attained by a
sequence of three small rotations Aw^, Aw,,., Aco2- (order immaterial)
about x'-,y'- and z'-directions, respectively, with
1
(l -*AAw[-lf^\Aw 2 + lAw^
Pa \ x2) U2v3/ v3
Aw , = — Aw[ Aw!, (3.7)
Aftij. = - - — Az-j + — Av 2
If unit vectors associated by the barred and primed axes are denoted
by f,, ~tg,~t, and T,., "?„-, T^ then for the displaced state of the barred
axes shown in Fig. 3.1b wc have
{ t , + Af,, fe + Af e , t t + At 4 } = ii {t z ., tm., t z ] (3.8)

wherein
1 An -Aw,.
Q Acoz. 1 Aw,- (3.9)
Aw,,. —Aw., 1
and it is assumed that cos (Aw r ) = 1, sin (Aw,,..) = Acox., etc. Rela-
tions between the displacement matrices {Au}, {Au'} are readily derived
from the equations
{Au2 - Au[) fx. + {Av2 - Av[) t „ . + {Aw2 - Aw!.)t,. =

{X. + Aii 2 )(t, + A t , ) - Sml, (3.10)
(Aa 3 - A n f l t , + (At-3 - &>{)t, + (A«v3 - Aw'JH,. =
{X. + Aii 3 )(t, + A t , ) — xft x . + {y3 + Av 3 ){t 9 + A t s ) - yft^
together with Eqs. (3.7) and (3.8). The result obtained after deleting
second-order terms in the displacements may be summarized in the
form of a matrix equation
{A«} = G{Au'} (3.11)
244
I
CM U C en — — m - I -
o o o
1" Z -2 -A
CO CO CO
z o o —
5
o cc i .ti B
o o -
•S.8.
O O
—
£_
O S
U *ZJ
1 £
a
o —
o
o
o
ni
—
o
ni
o
o
w •c .9 - H - -:
z a - a 4- + -z
- H —c u
§ I
Ml Ml
• -
o — £ i
n
•*. '*3
ea
•a -a u
< **: 13 .a 1 ^ S
— - o N
.5 a a«
1 " .
a
I
» i
z •_ — —
- a c CI cou IM
< rt _ «J
1
Crt •—a
— ^£ o a
N - + + a»J U
- 1/5 -^3 u
rt xt'3
- c
Ni C. rt "C
- G o •5.J. o
•a ** -a "• J3 V a
o o
O "5
u
«J
++ + i 1
—i
1
- - be rt H 60 J C - "' _ x "S
y.
<
a . c- *«-
—
a tU
o
>- B S ni ni a.
h i'<
• -
u _- - •
c d
I + ' K I IH
M
<H\ in u K,
—— S3 . 5
z
n t 9a *a , . J * C4 c
&
a -. ' " r c + Us
•J Ss
'Ai . 0
.- uci £
> iA
__ a
s« — u
V
1
I s g—- >
—
- X. Cj ^ 1 - -
is
B " 31 0 X
Z ts > *- + a 5.
n
£ o
r—
—
M. J . T U R N E R , H. C. MARTIN a n d R. C. W E I K E L
and
AF =
0 — Aa> 2 . m\0>. 0 0 11 0 0 0 1
Aw,. 0 -Aw,. II 0 II 0 0 0
-Aw„. A-'V 0 II 0 II 0 0 0
0 0 0 0 -A..,, A
"V 0 (i 0
0 (1 0 Aw,. 0 - An, 0 i) 0
0 0 0 - A w , . Aw^ 0 0 (i 0
0 (1 I) 0 0 11 0 — Aw,."V A
0 0 0 0 0 (1 Aii) 0 — Awx.
0 0 0 1) 0 0 -Aw„- Aw^ 0
(3.18)
From Eqs. (3.14) to (3.17) there follows, after linearization,
{AX'} = [AFG T + AG T ]{X 0 } + G r {AX} (3.19)
A significant simplification is now achieved by utilizing a constant

stiffness matrix to represent the behaviour of the elastic triangle
referred to the moving reference system. Since stresses are within the
linear clastic range and the element is small enough to justify the
constant strain approximation, it is evidently consistent with these
assumptions to employ the superposition principle in representing the
behaviour of the elastic element relative to the barred system.
In the paper presented at the Aachen meeting and in the subsequent
paper on thermoelastic wing analysis 11 the stiffness matrix, including
effects of initial stresses, was derived in a stationary x^y-z-reference
system with the triangle initially lying in the x^y-plane. If the formulae
presented in the latter report (pp. 104, 107-110) were adapted to the
present circumstances we should have
{A^} = X{Au} (3.20)

with
R= X°+ Rl (3.21)
In this last equation K l represents the stiffness associated with initial

stresses as the triangle is strained with respect to the local coordinate
system. Since strains are assumed to be small throughout the entire
history of deformation, superposition may be applied in formulating
force-deflexion relationships relative to the local moving axes. Fibre
rotations relative to this reference system are of the same order as the
strains and therefore the nodal force increments due to initial stresses
246
D E V E L O P M E N T AND A P P L I C A T I O N S OF IT IK STIFFNESS METHOD
(nodal forces), which arc proportional to the fibre rotations, may be

neglected. We therefore have
0
and
{AX} = R°{Au] (3.22)
The last term on the right side of Eq. (3.19) then becomes,
GT{AX} = G r R°{Au}
(3.23)
= G r R°G{Au'} = K°{Au'}
Here K° is the elastic stiffness matrix for the triangle which was first
given in Rcf. 2, pp. 814, Eq. 22.
Returning again to Eq. (3.19) we now examine the first term on the
right side. First we find that,
AFGr+ AGT=
X
I) * 3 Aw, i3 A
Am,.
X, x2
P. 1
— Aw,. - A w , . + f. Au. - . Av3 _l Au2 | - At
%2 -^2 x2 x2
X
i-i A
Aw,. An,.. ' - Aw, Ai'j,,.
X. X.
h
i) A,o, Aw, (3.24)
X,
P» 1
Aw,. -j A«, . A«j
'J ^ + f **» x2 x2
-Aw,. Aw, - -* Aw,.
Xi
0 I) — Aw,
0 Aw, 0
(I -Aw„. Aw,
Second, on writing [AFG r + AG T ]{X 0 }, where {X0} == {X?2, 2 J , f%)

wc find the results can be collected as (with terms retained as indicated
in the equations which follow),
[AFG r + A6' T ]{* 0 } == a^{Att'} (3.25)

where
* - - XpCl + XlKl + Y°3Kl (3.26)
247
and
11
1
(1
v..
1
I) (1 SYMMETRIC
.v..
II (1 (1 0
A! 1 1 (3.27)
I) II I)
1 1
(l 0 0 II
Xi
n (l 0 0 0 0 0
i) (i 0 (1 II 0 1) (1
ll () 1) I) II II (1 II
Kl =
' 0 -
Pi 2x 2 - Pi
2
Y x\
x2
Zx 2 — x s
0 0 SYMMETRIC
0 i) (I
Pa x
ia ^3 xa
0
x\ ~ x% X2 X\
x3
0 (1 1) 0
x2 x\
0 (1 0 (1 0 0 0
1 1
0 0 I) II II 11
x2 •VJ
1 1
0 0 0 0 II I) (1
•Vg A,
(3.28)
218
Kl =
(1
*23 Pa
v2 v-
x
0 II
h
Z
SYMMETRIC
X
A
2 .\'l 3
*3
0 II (1
x
i3 Pt Pa
0
X2 v- yi v2
A
2
X3X23 x-.i
(I 0 1) 0
x x
iP'a 'iPa
1 1
(1 1) 0 II 0
X
i
II (1 II (1 0 (1 (1 (1
x
i3
x
a 1
0 (1
X
(1 II (1 (1
x
'P:. iPa Pa.
(3.29)
There remains the problem of determining initial forces XI, XI, and
f3 in Eq. (3.24). To accomplish this we first recognize that strain
increments are related to nodal displacements in the barred system by
the matrix equation
{Ai} - |A.,AÂ X f i ; [ R\Au} (3.30)
with
P.,
:-.t 0 0 "
1
R = 0 I) Xm (3.31)
XiPa (I
—x v.,
and stress increments are given by
{Ad} = {Aa, Aa s AT*} = *({Ac} - aAT{l 1 0}) (3.32)
wherein
I v 0 '
v 1 0 (3.33)
0 0 Xy_
The nodal forces generated by a temperature increase A T with fixed
nodes arc equal to the forces required to produce nodal displacements
{AtlST} - - * A T { x 2 , Xm. y3} (3.34)
249
Hence the matrix of nodal forces due to an increment in temperature is

{ A X ^ } = -cxATX»{x 2 , x 3 , y3} (3.35)
and the forces XI, XI, Y3 to be employed in Eq. (3.26) to calculate

K l arc given by
{X0} = AHX0{AU) -- ^ A 7 - ^ 2 x 3 y3} (3.36)
in which the summations extend over all preceding steps in the analysis.
From Eqs. (3.31) and (3.32) the matrix of final stresses is
a
{a} = ^ R { A u } -- ^ f r A r {1 1 0} (3.37)
1 — v
Transformation of K° and K 1 to the basic reference system is accom-
plished by operations ofthe type represented by Eqs. (1.10) and (1.18).
Also it may be noted that the K 1 matrix presented in earlier papers can
be obtained by expressing the nodal forces X\, X%, Y3 appearing in
Eq. (3.26) in terms of initial stresses, transforming R l by premultipli-
cation and postmultiplication by G T and G, and forming the sum
K 1 -f G T X l G. However, for the reasons presented earlier, it is advan-
tageous to omit the second term. Also, in solving a non-linear problem
by the stiffness method it docs not appear worthwhile to compute
stresses at intermediate steps in the analysis. Hence the form given for
the K l matrix in Eq. (3.26) in terms of initial nodal forces is preferred.
Stability Analysis
For certain types of loading of idealized structures having axial or
planar symmetry determination of the critical-load intensity at the
onset of instability and the corresponding mode of buckling can be
readily formulated as an eigenvalue problem. For example, the unique
equilibrium configuration of an initially straight column under axial
loading or of a flat plate subjected to in-plane loads of subcritical
magnitude may be determined by linear analysis. If the stiffness
matrix is formulated with lateral or out-of-plane displacements in-
cluded, then incremental displacements from equilibrium are related
to incremental forces by an equation
[K° + P ] { A a } = {AX} (3.38)
Furthermore, if the initial distribution of axial or in-plane loads is

fixed and the magnitude is determined by an intensity factor, F, i.e.
{*"} = F{E*}
and {H°} = column matrix of constants

then K l = FK 2 (3.39)
250
and both K°, K2 are constant matrices. At the critical load level the
homogeneous equation
[K° + FK2]{Au} = 0 (3.40)
has a non-trivial solution. For low order systems the critical load may
be determined as the smallest root of the dcterminantal equation
\K° + FK2\ = 0 (3.41)
For higher order systems it is convenient to transform Eq. (3.40) into
|A»1 l K 2 {Au} -=-p{Au] (3.42)
The smallest eigenvalue, F u and corresponding eigenvector may then

be determined by iterative postmultiplication, starting with an arbitrary-
column vector, as commonly practiced in vibration mode analysis.
In analyzing the stability limits of more general structures or of
ideal, symmetrical structures subjected to asymmetrical loading it is
usually necessary to resort to the step-by-stcp procedure to determine
equilibrium configurations at successively increasing load levels.
In analyzing the clastic stability of a beam-column member it is
sufficient to consider only the axial load in computing the A'*-matrix,
provided that the member has been subdivided into sufficiently short
segments. For simplicity we shall consider displacements of the beam-
column element in only one plane. As in the analysis of the triangular
plate element in plane stress we shall employ a local moving reference
system (see Fig. 3.2). Since we are considering only clastic deflexions
r
/ iff*}' rrrrj
INITIAL STATE
k VH DEFORMED STATE
Fig. 3.2. Local moving reference system for beam-column element.
17
251
M. J . T U R N E R , H . C. MARTIN and R. C. W E I K E L
the curvature ofthe member is bounded. By limiting the length ofthe

element, the slope relative to the chord (i.e. to the Jc-axis) may be
limited to small values, such that the square of the slope is negligible
compared to unity. It may be assumed further that the axial load is
small relative to the critical Euler load of the element. Under these
circumstances the bending moment due to axial load may be neglected
and the superposition principle may be applied in computing deflexions
relative to the barred system. Displacement and force matrices are
{Au} = {AM,, Ay,, Ay2}
(3.43)
{AX} = {AX,, AM U , AM,,}
and the corresponding stiffness matrix is
'AE
il i)
4£7 2EI
(3.44)
/. I.
2EI \El
i)
L I.
If wc treat the member as a free body referred to a fixed .v'^y'-rcference

system displacement and force matrices referred to that system are
{Au'} = {AMl', AV[, Ayl, AM.;, A ^ , Ay£
(3.45)
{AX'} = {A*;, AY'U A M . , , AX!„ AY!„ AM 2 ,}
A simple derivation gives for the corresponding sixth-order stiffness
matrix
K = K° + K1 (3.46)
with
AE
L
\2EI
i) SYMMETRIC
6EI 1/-.7
0
U L
A" (3.47)
AE
(i
~L
\2Ef 6EI 12£/
(i
(I L3 ~U L>
6EI 2E1 UP: i \El
(l If L /:- I.
252
'11
0 —-) SYMMETRIC
L
0 0 0
A1 (3.48)
0 0 0 0
0 ^ 0 0
•v,
/.
0 0 0 0 0 0
A simple geometric transformation of the usual form may be applied

to Eqs. (3.47) and (3.48) to obtain stiffness matrix referred to arbitrary
fixed axes.
Finally it may be noted that a stiffness matrix for the triangular
plate in bending may be derived by employing third-order polynomials
in two variables to represent normal displacements. This approach is
regarded as an improvement over that described in Ref. 1, pp. 22-27.
The same local moving reference system may be employed here as in
the development at the beginning of this section, with simple supports
at the corners. In this system the plate has six out-of-plane freedoms,
represented by the angular inclinations in the x- and _p-directions at
the nodes, and three in-plane freedoms. It is assumed that the element
is small enough to render the superposition principle valid in the
barred system, as in the case of the column. As a result the A 1 matrix
referred to local fixed axes involves only the initial in-plane nodal
forces X 2 , X 3 , Y3. Lack of space precludes a detailed discussion of
formulae.
Wing Analysis
The basic phenomena associated with the thermoelastic analysis
described in this section are best exhibited by vehicles having relatively
thin lifting surfaces subjected to thermal gradients and applied loads.
A model typical of such a lifting surface was designed and analyzed
by the Structural Analysis Unit of the Boeing Acro-Space Division to
determine the effect of extreme thermal gradients and loading con-
ditions on natural frequencies and modes of free vibration about the
equilibrium configuration. This work was performed under contract
No. AF 33(616)-7898 to the United States Air Force and is presented
in Ref. 12.
The model which has been analyzed is a cropped delta wing shown
in Figs. 3.2a and 3.2b. It was designed to be fabricated from Inconel
X. Three supports arc provided by bipod legs containing flexure
pivots to furnish translational constraint in two directions at each
support. The complete system provides statically determinate supports
253
M. J . TURNER, H. G. M A R T I N and R. C. WEIKEL
I r
1 — '
SUP"
C=pr;
1
• ' TYPICAL BAY DErAIL^/,1-—<--x

LlNiVDa- B.B
-\ y-
SUPT
(OK4 3~ FWD
V ^
-C7-^-y-'-L/ V ^ T *
l F 7 -• LlNB'D—
M U 4- MODEL PLAN VIEW
(SPARV^-T> "" " V

v^ 1
v^--\^-v/^CORRUGATED
SPAR WEB I
""C7- i .109 - " ™ -
1
FWO
TYPICAL BAY DETAIL f
UP j=
l-FWD —
Fig. 3.2a. Construction details—/Am wing.
PLANE OF MODEL <t
PLANE OF STRONGBACK
TYPICAL BIPOD SUPPORT
Fig. 3.2b. Thin wing support.
254
DEVELOPMENT AND A P P L I C A T I O N S OF T H E STIFFNESS METHOD
without imposing any constraints on elastic deformations. T h e skin-

gauge and the skin-panel sizes were selected so that the skin would not
buckle under the maximum stress condition. T h e webs of the spars
and ribs are corrugated to minimize thermal stresses. T h e spar caps
are spot welded to the webs and riveted to the skin. T h e wing has a
span of 130 in., root chord of 120 in., tip chord of 40 in. and a thickness
of 2.5 in.
The analysis consisted of (1) predicting the equilibrium configuration
for each of the load and temperature profiles, (2) computing the matrix
of the deflexion influence coefficients, and (3) calculating vibration
modes and natural frequencies.
The wing was idealized as an assemblage of triangular plate elements
as shown in Fig. 3.3. Nodes were located on each rib at points of
LEADING EDGE
SINGLE BAY
OF
FWD IDEALIZED STRUCTURE
SUPPORT POINT
INBD B > 1/2 I (f-SPAR THICKNESS)

MODEL PLAN VIEW TO EACH SPAR BETWEEN
EE> SUPPORT POINT NOOAl POINTS.
F i g . 3.3. Thin wing idealization.
intersection with alternate spars; separate node points were located on

upper and lower surfaces. T h e effective areas of intermediate spar
webs were distributed to adjacent spar webs. T h e temperature profiles
are given in Figs. 3.4 and 3.5.
It was found that the effect of thermal stresses was negligible and
that the change in equilibrium configuration accounted for the major
changes in natural frequencies. While the changes in curvature
increased the frequencies the reduction in Young's modulus due to
heating produced partial compensation of this effect.
The thermal deformations as computed by the linear theory and the
non-linear theory are compared in Fig. 3.6. T h e fourth symmetric
mode of vibration computed for the cold, unloaded structure and the
255
M. J. T U R N E R , H . C. MARTIN a n d R. C. W E I K E L
1200 TRAILING EDGE ^

y - LEADING EDGE
1100
i > HOT SIDE
900
> COLO SIDE
121 117 111 105 97 89 79 69 57 45 33 21 9

121 118 112 106 98 90 80 70 58 46 34 22 10
NODE NUMBERS
Fig. 3.4. Temperature profile.
0 3 6 9 12 15
TIME (MINUTES)
1 2 3 4 5 6 7 8 9 10 II
COMPUTATIONAL STEPS
Fig. 3.5. Mean temperature of surfaces vs. time.
256
hot structure are compared in Fig. 3.7. The variation ofthe frequency
ratio (ratio of frequency at time t to the initial frequency with time) is
shown in Fig. 3.8 for constant material properties. Methods for
accounting for variable material properties were not available for this
analysis; however, if it is assumed that the frequency is proportional
to the square root of the ratio of the mean value of E to the initial
value, the variation of (ojco0, with time is as shown in Fig. 3.9. (In the
analysis the initial value of E was determined for a temperature of
700°F.)
-.AO
Fig. 3.6. Hot structure deflexions.
+ 1.00
^•-..O
FREQUENCY 126.64 CPS FREQUENCY 124.67 CPS
• SUPPORT POINT S SUPPORT POINT
COLD UNLOADED STRUCTURE HOT STRUCTURE

Fig. 3.7. Fourth symmetric mode.
257
FREQUENCY RATIO ( u / w )
6 9
TIME (MINUTES)
Fig. 3.8. Frequency ratio vs. time: constant E
FREQUENCY RATIO ( u A ^ r ) I .01
1.00
6 9
TIME (MINUTES)
Fig. 3.9. Frequency ratio vs. time corrected for \ { E / F l l w .
258
In analyzing thcrmoelastic behaviour of complex structures it is

recommended that rough, preliminary computations be made before
undertaking a refined study including non-linear effects. Such a
preliminary study might consist of the following steps: (1) compute
thermal stresses without deformation; (2) use linearized analysis to
estimate total deflexion in one step; (3) compute modes and fre-
quencies for approximate equilibrium configuration and approximate
thermal stresses.
Plate Buckling Analysis

Finally to illustrate the applicability of the stiffness method to a
buckling problem, the deflexion behaviour of a square plate loaded at
its edges was studied. T h e square plate had the dimensions and
material shown in Fig. 3.10. A concentrated load of 100 lb was applied
TOTAL DEFLECTION
P (UNIFORMLY DIST LOAD) LBS/IN.
NORMAL TO PLATE AT POINT A
mi m 100-LB. LOAD NORMAL
TO PLATE AT A
POINTS COMPUTED
.3 -- z
FROM DEFLECTIONS
• /p do" 5 )
OBTAINED FROM D I G -
IN. ITAL PROGRAM
LBS/IN. UNIFORMLY DISTRIBUTED
.2 • - LOAD O N EDGE
.02
-5
- .224 X 10
Ai
» At .02
.1 . . P .224 X 10
- 8928
PCR DEFINED BY R.V.SOUTHWELL PCR (THEORETICAL)' 8894
PRO. ROY. S O C , L O N D O N , SERIES A , E - 10 7

V O L . 135, p . 601, 1932.
0 .01 02 .03
DEFLECTION AT POINT A (INS)
Fig. 3.10. Plate buckling.
normally at the centre and maintained constant. Then an edge load

was applied in a series of equal increments. The lateral deflexion at
the centre of the plate was computed for each increment of load in a
piecewise linear manner, as were the deflexions computed for the
heated wing of the preceding example. T h e results were plotted as
6/P vs. 6 in Fig. 3.10 where Ad = deflexion and P = edge load. This
plot is sufficiently linear to apply the Southwell technique of computing
the critical buckling edge load from the slope of the plot, i.e.
AS
P
A cr -
Act/P
259
Stability of a Tapered Column

Application of the stiffness method to linearized column analysis
will be demonstrated for the case of the conical segment shown in Fig.
3.11. From the figure the following relations may be seen to apply:
rmln (*), 'mi

-(;)'
For this example we choose,
•min
0.20
We idealize the actual column by replacing it with n segments of equal

length; each segment has a constant cross-sectional area given by the
true value existing at the midpoint station of the segment.
p
'7TT7T7T7T
Fig. 3.11. Tapered column.
For purpose of illustration consider the simple case for which n = 2.

Let nodes be established as follows: node 1 at x = 0, node 2 at
x = L/2, node 3 at x = L. Also let A u I, be the area and moment of
inertia for segment 1-2. Since node 3 is clamped, the displacements at
that node are not considered in forming the total stiffness matrix.
Equations (3.47) and (3.48) may be used to find A0 and A 1 for each
segment. These may then be combined in the usual manner to give
260
the total stiffness matrix for the two segment idealization. In doing
this initial loading A', in Eq. (3.48) will be replaced by the initial
compressive load P. Carrying this out and properly arranging terms
leads to the following force-deflexion relation:
-1 -1
•-r
rp
•-I k* --1
•5-I
*H
-I-
t Cl
f.
CN ~ CN ~
+
1*5 • ^ CM c-i
en
• ~
5.
( •*, o - 3 c
-7
C*l
~7 "-, o o 3
-M c-i
1 1
II
1
M
iAA> —
' .
-1 -1 "1 <
1 _ 1
261
A;/,, Avit Ay, denote incremental displacements from the initial

equilibrium configurate due to incremental loads AX if AY„ AM,.
For convenience this last equation may be expressed in the form
AX 0 " 'Au
AY Av
0
AM .Ay.
This separates into the two equations,
(AA'} = [A aa ]{A«}, [ ^ ] = [Kbb] [^]
in which P occurs in [Kbb\ only. In other words the eigenvalue problem

associated with bending stability is not coupled with the problem of
direct axial force. This corresponds to the inextensional condition
commonly introduced in deriving the linearized differential equation
for column stability analysis.
When P = critical loading, arbitrary displacements [Av, Ay] may
be sustained in the absence of external loads [AY, A M ] . This result
is obtained if
A,„ n.
Alternatively [A,,,,] may be separated into a matrix [Kbb] which is

independent of P, and a second matrix [A',,] which depends only on P.
This enables us to write in accordance with the derivation on pp.
250-251
and
VAY1 = [TO-
LA../J I n
{[*Z\ f P[K2b])
ca
'Av Av
[A*,I 'i*;,',, Ay. LAy.
This last equation is now in standard form for matrix iteration. It has
been applied to the problem at hand for three cases: n = 2, 4, and 8.
If we express the theoretically correct result as,
mEIr,
PcrM =
L2
where m = 1.505 we can observe the accuracy of the stiffness-method

solution by referring to Table 3.1.
262
Table 3.1
n
Number
m % Error
of
Segments
1 1.491 0.93
4 1.502 0.20
8 1.5035 0.01
4. APPLICATIONS TO COMPLEX S T R U C T U R E S
Typical of the comprehensive problems that have been solved by the
stiffness method is that of determining static deflexions of the entire
fuselage of a helicopter.
Fig. 4.1. Helicopter fuselage structure—Vertol.
The actual structure ofthe fuselage is shown in Fig. 4 . 1 ; essentially

it consists of a thin cylindrical shell stiffened fore and aft by simple
extruded stringers and circumfcrentially by press-formed channels.
The cross-section ofthe fuselage is a rectangle with "rounded" corners.
An extruded longeron is placed at each of the four corners. The for-
ward end is closed by the pilot's cabin and the aft end is closed by a
large cargo ramp door, and window openings are located along both
sides of the fuselage. Streamlined stubs near the aft end ofthe fuselage
transmit landing gear loads to the fuselage. Rotors are supported by
263
forward and aft pylon structures on top of the fuselage. Due to rotor
torque reaction the stringers are heavier on one side than the other,
making it necessary to analyze the entire structure because of lack of
symmetry. Two engines and the transmission gear box are mounted
in the rear pylon.
Static deflexions were computed by the stiffness method for three
load conditions: (1) a side load applied at the forward rotor h u b ;
(2) a side load applied at the aft rotor h u b ; and (3) a forward load
applied at the aft rotor hub, and simultaneously an aft load applied at
the forward rotor hub. The axles of the main landing gear were fixed
to give longitudinal, vertical, and lateral constraint; the castored
forward wheel was constrained in the vertical direction only.
The formulation o f t h e idealized structure was based on (1) maxi-
mum number of structural elements and nodes that could be handled
simultaneously by available automatic digital programs, (2) numbcr
of structural elements and nodes necessary to give a sufficiently accurate
estimate of the structural behaviour, and (3) number of mode shapes
desired. T h e two-dimensional triangle and the stringer are the basic
structural elements employed in the analysis. Many more nodes were
needed to describe the elastic behaviour than were needed to form the
mass matrix.
The fuselage was partitioned into three sections, the forward pylon
section, the cabin section, and the aft pylon section as shown in Fig.
4.2. Lumped masses are indicated by the balls. Nodes are located at
Fig. 4.2. Structural idealization of helicopter fuselage.

264
160- STA 497
140-
120-
100
WATER LINE (INCHES)
-.30 -.20 -.10 0

6 (INCHES)
Fig. 4.3. Comparison of analytical and lest deflexions—case I I I 2500-lb
longitudinal loading.
STA 101 ._^

74 l . . . ^
80 _r-Q
A**
60
n\.....jT
73 &
Alj
TEST / j THEORY a
20
TEST
0
WATER LINE
(INCHES) 0
[J
n -453-61
-20
-40
it 6
O-fTFG
j LONGITUDIN AL DEFLECTI DNS
-60 i AT FWD ROTC;R STATION

i
i
S (INCHES) .10 0 .10 .20
VERTOL FUSELAGE
Fig. 4.4. Comparison of analytical and test deflexions—case I I I 2500-lb

longitudinal loading.
265
M. J . T U R N E R , H . C . MARTIN a n d R. C. W E I K E L
the masses and at the black points in the figure. Twelve nodes were
located on each circumferential and three or four masses were located
on each circumferential of the cabin. It will be noted from Fig. 4.2
that groups of stringers have been lumped together in defining the
theoretical model for analysis. Quadrilateral skin panels were sub-
divided into two triangles. Nodal displacements not explicitly desired
in the deflexion analysis were deleted by matrix techniques described
earlier.
T h e gross structure used in the deflexion analysis contained 179
degrees of freedom. For the dynamic analysis only the displacements
associated with the lumped masses were retained.
T h e deflexions due to static loads were determined by imposing
suitable boundary conditions and inverting the stiffness matrix. T h e
predicted deflexions for nodes located on station planes 101 and 497
are compared with the experimental deflexions for loading condition
(3) in Figs. 4.3 and 4.4.
REFERENCES
1. T U R N E R , M . J . T h e direct stiffness method of structural analysis. AGARD
Meeting, Aachen, G e r m a n y , paper presented 17 September, 1959
2. T U R N E R , M . J . , C L O U G H , R . W . , M A R T I N , H . C . a n d T O P P , L . J . Stiffness a n d
deflexion analysis of complex structures. J . Aeron. Sci., 2 3 , September 1956, p p .
805-823.
3. M A R T I N , H . G. Plane elasticity problems by the direct stiffness method. The
Trend in Engineering, Engr. E x p . Sta., University of Washington, Seattle, J a n u a r y
1961.
4. TIMOSHENKO, S. a n d GOODIER, J . N . Theory of Elasticity, M c G r a w - H i l l , 2nd Ed.,
1951, p p . 167.
5. K U H N , P., U U B E R O , J . a n d G R I F F I T H , G . T h e effect of concentrated loads on
flexible rings in circular shells, N A C A , A R R N o . L 5 H 2 3 , 1945.
6. H O F F , N . J . Stresses in a reinforced monocoque cylinder under concentrated
transverse loads. J . Appl. Mech., 66, p p . A-235-A-239, Decermber 1944.
7. JENSEN, W . R . O n simplified fuselage—structure stress distributions. J . Aero/
Space Sci., 2 5 , No. 10, p p . 6 5 6 - 7 .
8. DENKE, P . H . A general digital computer analysis of statically indeterminate
structures. A G A R D Meeting, Aachen, G e r m a n y , paper presented 17 Sep-
tember, 1959.
9. E G G W E R T Z , S. a n d N O T O N , B. Stress a n d deflexion measurements on a multicell
cantilever box b e a m with 30° sweep. Aeronautical Research Institute of Sweden,
Report 53, Stockholm, 1954.
10. E G O W E R T Z , SIGGE. Calculation of stresses in a swept multicell cantilever box
beam with ribs perpendicular to t h e spars a n d comparison with lest results.
T h e Aeronautical Research Institute of Sweden, Report 54, Stockholm 1954.
11. T U R N E R , M . J . , D I L L , E. H . , M A R T I N , H . C . a n d M E L O S H , R . J . L a r g e deflexions
of structures subjected to heating a n d external loads. J . Aero/Space Sci.. 2 7 ,
Feb. 1960, p p . 97-107.
12. WEIKF.L, R . C , J O N E S , R . E., S E I L E R , J . A., M A R T I N , H . C. a n d G R E E N E , B. E.
Nonlinear a n d thermal effects on elastic vibrations. U . S . Air Force Report,
A S D - T D R - 6 2 - 1 5 6 , F e b . 1962.
13. TIMOSHENKO, S. a n d G E R E , J . M . Theory of Elastic Stability, 2nd Ed., M c G r a w -
Hill, 1961.
266
SOME PROBLEMS IN THE DISCRETE
ELEMENT REPRESENTATION OF
AIRCRAFT STRUCTURES
I. C. TAIC and R. I. KERR

English Electric Aviation Limited, VV'arton Aerodrome, near Preston, Lancashire
77ir? paper reviews a number of problems encountered when idealizing complex

aircraft structures into discrete elements fin the purpose i f matrix stress analysis.
The first part is concerned with the problems arising in the analysis of wing-type
structures by the matrix displacement method. A number of general topics are
also covered in this section. The second part deals with a different .set of problems
which have been encountered in analysing fuselage structures by the matrix force
method.
In both pails qf the paper particular problems experienced in actual analyses
are discussed and some results qf controlled studies of idealization variations are
presented.
C O N I I.N r s
INTRODUCTION 268
P A R T I. APPLICATION OF DISPLACEMENT ANALYSIS TO M U L T I -

CELL Box STRUCTURES . . . . . . 269
STRUCTURAL IDEALIZATION 269

1.1. Choice of structural grid 269
1.2. Idealization into line and sheet elements 272
1.3. Taper, curvature and twist . 273
1.1. Idealization of element behaviour 277
1.5. Miscellaneous idealizations . 284
2. SOME ADDITIONAL PROBLEMS ENCOUNTERED IN PRACTICE . 285

2.1. Structure size and computer limitations . . . 285
2.2. Ill-conditioning resulting from structural idealization . 287
2.3. Data accuracy 289
267
I. C TAIG a n d R. I . KERR
Page
3. T H E INFLUENCE OF SOME PARTICULAR IDEALIZATIONS ON A

SAMPLE BOX ANALYSIS . . . . . . .291
3.1. Choice of structure and scope of investigation . .291
3.2. Results and discussion . . . . . . 294
3.3. Conclusions . . . . . . . . 304
P A R T II. APPLICATION OF FORCE ANALYSIS TO FUSELAGE STRUC-

TURES . . . . . . . . 304
1. STRUCTURAL IDEALIZATION 304

1.1. Choice of structural grid 304
1.2. Idealization into line and sheet elements 306
1.3. Idealization of clement behaviour 306
2. SOME PRACTICAL PROBLEMS 306

2.1. Structure size and computer limitations 306
2.2. Ill-conditioning arising from structural idealization 308
2.3. The effects of boundary conditions 310
3. COMPARISON OF FORCE AND DISPLACEMENT ASSUMPTIONS 312
INTRODUCTION
During the past 15 years the approach to aircraft structural analysis
has undergone a major change as a result of the introduction of elec-
tronic computation. A large number of papers have been written
during this period suggesting methods of analysis applicable to various
types of aircraft structures. Many of the early papers were logical
extensions of the engineering analysis of continuous beams and tubes
which had been generally applied to aircraft design beforehand. More
recently, however, a new technique has been developed in which
structures are considered as an assembly of discrete elements, each
possessing simplified properties of continuity and elasticity, coupled
together in systematic manner by the formal use of matrix algebra.
Fairly extensive literature now exists dealing with the basic principles
and the algebraic formulation of these methods but very little informa-
tion has been published concerned with the practical problems en-
countered in applying matrix analysis to actual structures.
This paper presents a brief discussion of some such problems which
have been encountered during ten years of practical application of
matrix stress analysis. The discussion covers such topics as geometrical
idealizations employed in the analysis and some of their consequences;
the representation of the elastic behaviour of structural members and
268
DISCRETE REPRESENTATION O F A I R C R A F T STRUCTURES
difficulties arising during computation which limit, or are influenced

by, the structural idealizations employed.
Matrix methods of stress analysis fall into two broad categories:
those employing displacements as the basic unknowns and those
formulated in terms of forces. Displacement techniques have been
found by the authors to be particularly well suited to the analysis of
low aspect ratio multiccll box structures due to the case of handling
irregularly shaped structural elements and their efficiency in solution
of structures with a large number of "parallel" interconnections.
Force methods, on the other hand, have found their major application
in fuselage analyses where the "series" type of structure leads to an
economical problem formulation in terms of redundant forces compared
with displacement degrees of freedom. Some considerable difference
has been encountered between the problems arising in idealizing a
wing-type structure for displacement analysis and a fuselage structure
for force analysis. Hence the paper has been divided into two parts
dealing with each application separately. There are, of course, many
common problems and these are generally discussed under the heading
of matrix displacement analysis, simply because the authors can call
upon a wider range of experience in dealing with this type of analysis
and it is in this context that most of these problems first arose.
It is appreciated that most of the questions arising in idealizing a
structure do not lead to rigorous solutions and little attempt is made
in the ensuing discussion to prove, by formal means, the validity of the
principles employed. In many cases a simple statement of current
policy is given with no more than qualitative arguments to justify it.
It is generally true, however, that many of the more controversial
solutions to idealization problems have been reached as a result of
careful assessment of the results of numerous practical analyses, and
after discarding alternative ideas on the basis of this assessment.
PART I. APPLICATION OF DISPLACEMENT ANALYSIS

T O MULTICELL BOX STRUCTURES
1. S T R U C T U R A L IDEALIZATION
1.1. Choice of Structural Grid
The first problem of idealization of an actual structure into discrete
elements is the choice ofthe structural grid. It is considered axiomatic
that every attempt should be made to choose grid lines as nearly as
possible to correspond with actual structure. Departure from this
principle may be necessary due to various difficult design features and
to limitations in the geometry of structural elements which can be
analysed by a particular method.
Within these limitations a certain amount of choice exists as to the
269
I . C. TAIG a n d R. I. KERR
size of mesh to be employed and the breakdown of actual panels into

suitable panels for analysis.
1.1.1. Size of mesh. The size of mesh chosen depends on the nature
of the assumptions to be made concerning element behaviour and on
the degree of accuracy required from the analysis. Using linearized
displacement assumptions it is considered suitable to conduct pre-
liminary analyses of low aspect ratio structures by sub-dividing the
larger structure-wise dimension into six bays of approximately equal
size and the smaller dimension into bays of similar length. Evidently
this rule can be modified to suit the actual internal structure if a better
representation of local members can be obtained using, for instance,
five or seven bays.
For comprehensive analysis the structure should be divided into at
least ten bays in the direction of the major structural dimension (or a
number equal to the number of" primary internal members il greater
than ten) and similar-sized bays in the minor direction. Methods of
analysis based on a reasonably representative model of element
behaviour generally converge towards a more accurate solution as the
mesh size is decreased. It should be noted, however, that maintenance
of uniform sheet element proportions is important. Rattinger and
Gallagher 1 have shown that reduction of element size in one direction
only can have an adverse effect on accuracy when this involves the
introduction of panels of high aspect ratio. This point will be enlarged
upon later.
Sub-division of actual skin panels by grid lines between spars and
ribs is generally a worthwhile objective particularly in regions of rapid
load diffusion.
Limitations on mesh size may be introduced by particular computer
programmes as will be mentioned later.
1.1.2. Choice of suitable panel shapes. The choice of panel shapes is
largely dictated by the method of analysis adopted. Most current
analytical methods represent spar and rib elements as trapezoidal or
quadrilateral flexural elements or sheet/flange combinations. Skin
panels are represented in earlier analyses as rectangular or trapezoidal
plane sheets and in current analyses either by triangular sheet elements
(e.g. Turner 2 ' 3 , or by combinations of non-planar quadrilateral and
triangular elements 4 - 5 ).
The latter idealizations arc theoretically capable of representing any
structure adequately, but sometimes considerations of mesh and size
and panel proportions necessitate slight repositioning of structural
members to fit the idealized model. Figure 1 shows how preliminary
and detailed grids have been selected to analyse a delta fin structure.
In most analyses conducted to date element boundaries are assumed
straight and hence curved contours are represented by a scries of
straight lines. As a result of such a simplification transverse loading
270
DISCRETE R E P R E S E N T A T I O N OF A I R C R A F T STRICTURES
*-*3
ItCONOIIIIY STBUCTUFlt'
- I ! ' . LINES COINClOCNT <XITH STRUCTURE
• aDoiTioiai. (ait) LIHCS
lNTt««». WCIS
tlilNltli
J" . JOINTS
fb) REFINED GRID.
Pig. 1. Representation of a delta fin.
of internal webs due to skin and boom curvature can only be repre-
sented by a scries of point loads at grid nodes. This is a limitation
which can be corrected to a reasonable degree of accuracy after
performing an analysis.
In earlier published analyses no attempt was made to simulate in
detail the taper or curvature of box structures and considerable
differences in box depth could exist between idealized and actual
structures.
271
t . C. TAIG a n d R. I. KERR
1.2. Idealization into Line and Sheet Elements

Structural idealizations currently employed normally represent
reinforced sheet structures by a combination of line elements (booms)
and thin sheet elements (panels). Actual structures consist of sheets of
finite thickness and profiles slightly different from the idealized sections
and stringer/flange elements of finite size whose material is distributed
over an appreciable proportion of the box depth.
In earlier analyses 6 - 7 the direct load-carrying capacity of skin,
stringers and flanges was lumped together into the boom elements, a
matter calling for considerable judgment when dealing with irregular
structures or widely-supported skins. The more sophisticated analyses
only require the lumping of stringers and flanges into the booms and
the process is less intuitive when a suitable grid has been chosen. The
main requirement is that the flexural and torsional properties of the
actual local section should be adequately represented in the structural
model and a simple example ofthe idealization process is now described.
Consider the portion of box structure illustrated in Fig. 2 which is
to be represented by the linearized line/sheet elements also shown. It
Fig. 2. Idealization of typical box section.
will be assumed that half of the area of stringer / will be associated

with booms A,. x and A, and half the area of n with booms A t and A , + v
The effective area Af is then determined so that the moment of
inertia of A t about the local neutral axis is equal to the moment of
inertia of the combined flanges.
A t 7 * = */. + I m + J / .
If the stringer moments of inertia about their centroids are small we
can write
1. = A,Z 2 , etc.,
272
DISCRETE R E P R E S E N T A T I O N OF A I R C R A F T STRUCTURES
where Z, is the distance of the centroid from the neutral axis and
A,Zf = \ A , Z 2 + AmZ2n + UnZl
or in general
AZ 2
A
'-*4;
the summation being taken over all associated areas A.
Ideally the effective skin thickness tt in the bay between stringers,
i -- 1 and i should be chosen to yield flexural and torsional stiffness
equivalence. The two relations, however, are not in general compatible
and in regions of pronounced curvature the simple device has been
adopted of satisfying torsional stiffness equivalence and modifying
adjacent flange areas to adjust flexural stiffness. For relatively thin
skins the relations can be written as follows:
Flexural equivalence tZ 2 ds = I fZ'f ds
l f { Z 2 + Z A Z ^ + Z t _\)
3
Torsional equivalence about 0
[' tfds = [ii_tip\ds = lttip\

When taper and curvature are small both conditions arc approximately
satisfied by
/,/,(Z'+1Z^l)2= \tZ 2 ds
4
22
£*i
This is the commonly used form of expression for correcting skin

thickness.
It should be noted that the effect of stringer centroid depth correc-
tions is to introduce an effective area A,- which is generally less than
the actual relevant stringer area A. If appreciable in-plane loads are
present it may be necessary to restore the additional area {A — A,) in
the form of boom area at the neutral axis.
1.3. Taper, Curvature and Twist

In many of the early proposals for discrete element analysis, and
indeed in some ofthe models currently employed for electronic analogue
computation, no attempt is made to simulate in detail the effects of
variation in box depth. Flcxural members are represented merely as
possessing moment of inertia and perhaps shear flexibility but not as
273
t. c T A I G and R. KERR
tapered beams carrying moment in non-parallel flanges. The effect of

carrying bending moment in tapering flanges is to react a large
proportion of the applied shear force by means of the vertical compo-
nent of flange load: the familiar slope-shear effect. In Fig. 3a a
typical conventional spar is shown with the distributed spanwise load
having a centre of pressure at 42 per cent span. If the wing tapers in
(b) D I S C R E T E ELEMENT REPRESENTATION
Fig. 3. Effect of taper.
depth from d„ al the root to dx = \dQ at the tip then it is readily seen
thai the vertical component of the root flange loads P 0 is given by:
I',, 0-42 O-fi/t . - :'.A-
Thus for these typical proportions ', ofthe shear in a straight tapered
spar is carried by the flange bending forces.
Figure 3b illustrates how a Garvey* type shear panel and inclined
flanges can satisfactorily represent element behaviour in a tapered spar.
Using the notation illustrated, the panel end loads and shear forces
are related to the resultant outboard shear force by the equations:
{a + b)
P. - I • R Is'
acf
R
l»l,
{a + J ) a \ . _ a\
bdb hi •
274
Alternative spar representations will be discussed later when considering

stress and strain distributions in discrete elements.
The effect of curvature of boom elements is to introduce distributed
normal forces in the webs of flexural members. Similarly, curvature
of skins transmitting direct stresses introduces normal forces equivalent
to a pressure distribution over the skin which must be transmitted to
adjacent webs.
In a linearized grid representing a curved structure it has already
been observed that the normal forces due to curvature are replaced
by equivalent point loads at the structure nodes. This limitation is
exactly equivalent to the lumping of applied aerodynamic loads into
localized forces and hence imposes no further restriction on the analysis.
A simple method has been proposed for making allowance for distri-
buted loads along clement boundaries (4) in linearly elastic structures.
The applied loading is represented by two components: the equivalent
point loads as illustrated in Fig. 4 and self-equilibrating force groups
• • r, . • . n *
Fig. 4. Representation of curvature in a linearized grid.
representing the actual distributed loads reacted by forces equal and

opposite to the equivalent point loads.
The elements are solved independently under the action of the self-
equilibrating force groups using the appropriate boundary conditions
(zero boundary displacements for displacement analyses and zero con-
straint forces for force analyses). The boundary forces arc computed
in a displacement analysis and these are used as initial internal
forces/displacements in the conventional analysis of the complete struc-
ture. Finally the local and overall solutions arc combined to yield the
complete solution.
In the case of curvature loading, which is not known before com-
mencing the analysis, an approximation to this procedure can be used
in which the effects of the initial internal forces/displacements on the
overall analysis are ignored and the local self-equilibrating force
systems are computed after solving the main problem.
Most wing-type structures are designed as conical or near-conical
275
I. G. T A I G a n d R. I. KERR
surfaces with straight generators, but it is rarely possible to orientate

the internal structure so that all the main members run along genera-
tors. Consequently, if the structure is sub-divided into quadrilateral
cells by the internal members it is usual to find that a linearized grid
passing through the natural corner points gives rise to surface elements
which are not planar. Except in the vicinity of the leading edge the
amount of twist involved is generally small and it is usual to derive the
stiffness or flexibility characteristics of the sheet elements from those
of a similar flat sheet. Some important consequences of panel twist
must, however, be noted and accounted for in a comprehensive stress
analysis.
Firstly, the twist gives rise to a change in the slope-shear character-
istics of the skin between adjacent spars and the effect of twist on the
direct stress components should therefore be represented to the same
order of accuracy as other geometrical factors. Secondly, a set of
tangential forces around the boundaries of a twisted panel cannot be
in equilibrium and if shear-type forces are present in a panel it is
necessary to introduce forces normal to the local surface in addition to
tangential forces. Allowance for this factor is particularly important
in displacement-type analyses where, as will be explained later, equi-
librium of element forces must be established to a very high degree of
accuracy.
T o account for both these features of twisted panels, the following
procedure has been devised for use in matrix-displacement analyses.
The panel stiffness matrix for a non-planar quadrilateral as shown
in Fig. 5a is obtained by first deriving the stiffness of a plane panel
obtained by projection on to the median plane as shown. The element
forces thus obtained arc then resolved into self-equilibrating axial force
pairs along the projected panel boundaries and shear-type forces
tangential to these boundaries. The axial force pairs are then assumed
to act along the true panel boundaries thereby simulating the influence
of twist on direct stresses. The shear-type forces are, however, assumed
to act as point loads at the nodes of the actual idealized element
parallel to the median plane (Fig. 5c). It is readily demonstrated that
such a system is in equilibrium and intuitively it appears to be a plaus-
ible representation of the behaviour of a twisted panel. Resolving the
corner forces back into the reference axes a transformation matrix can
be derived relating the forces in the twisted element to those in its
plane projection. This transformation is then applied symmetrically
to the plane panel stiffness matrix to yield the twisted panel stiffness.
An alternative approach, employed by Turner, Clough Martin, and
Topp 2 , is to eliminate the effect of twist by sub-dividing each quadri-
lateral clement into triangles, each of which will, of course, be planar
in a linearized grid system. This procedure overcomes this particular
problem but needs to be employed with discretion.
276
DISCRETE REPRESENTATION OF A I R C R A F T STRUCTURES
«•) TWISTED PANEL AND MEDIAN PLANE PROJECTION.
- =»<.ar.. *: IESJ.TAST f t C E H-HMtL

m , •-. >L»St JO UlOltM P I . I . E
b> LOSS OF E Q U I L I B R I U M UNDER SHEAR -TYPE FORCES.
C.) REPRESENTATION OF BOUNDARY FORCES.
F i g . "). Non-planar quadrilateral sheet.
1.4. Idealization qf Element Behaviour

Perhaps the most complex facet of the structural idealization
problem is the representation of the behaviour of discrete structural
elements in terms of a finite number of force or displacement para-
meters. Whilst early attempts were made to obtain analytical solutions
to complex structural problems employing continuous variables,
results were only achieved by making very restrictive physical assump-
tions such as the presence of rigid ribs, and Gallagher and Rattingcr 1
have demonstrated conclusively that these early methods could yield
277
I. c T A I O and R. I. KEKR
very unsatisfactory results when applied to low aspect ratio multi-cell

structures.
To achieve satisfactory results for complex structures it has been
found necessary to employ discrete displacement or force variables,
whether these take the form of Fourier Series components or simplified
element functions. We are concerned here with the latter representa-
tion and in particular with the assumptions made in displacement
analyses of multi-cell box structures.
The simplest idealization of element displacements is the assumption
of linear distribution along clement boundaries. The displacements
along a boundary in a given direction are usually represented by two
triangular distributions as shown in Fig. 6 defined by the magnitudes
Fig. 6. Definition of linearized displacements.
of the displacements at each end. For line elements this definition is

sufficient to determine the displacements, and hence strains, at all
points within the member and the stiffness properties follow auto-
matically.
For two-dimensional elements it is necessary to define an internal
force or displacement pattern in terms of the linearized boundary
deformations and this can be carried out either by solving the internal
equations of elasticity, equilibrium, and compatibility with linear dis-
placement boundary conditions or by making further simplifying
assumptions.
In the case of triangular elements an exact solution of very simple
form exists. The plane clement t of Fig. 7 has boundary deformations
Fig. 7. Triangular element with linear displacements.
U, (referred to arbitrary cartesian axes) consisting of triangular

displacement functions as described above.
1], E . {U X Vi U3 V3}
278
DISCRETE REPRESENTATION O F AIRCRAFT STRUCTURES
A typical displacement of this set, Vlt is illustrated. If we now assume

that internal displacements
U = {U{x,y) V{x,y)}
arc also linear with respect to the cartesian coordinates then
1 (ji ~.y)*3i - (*i - x

)f.a-i II .
x
aiy-n -J 3 2-v 2 i L 0 {y2 -y)x.m - {.u - x)y„ . \\
u.
where x a . {x3 — x.,), etc. v..
I....
LV,.
The three components of plane strain are then given by
em p. -V]
Sm' dy
Cjfy •= ti Ut
< ,. '.
YTV
A •.
y-ai I) P'l.l y-n

1
where e. = 0 _ ,
- '32 0 (1 A
21
x-aiy-n ~Jr,w&n
—Xm. ^32 -731
Thus the linearized displacement pattern gives rise to the simple

constant strain system defined by the above matrix which satisfies the
conditions of internal compatibility and is readily shown to be inde-
pendent of the choice of axes.
The corresponding stress system is obtained by premultiplying by a
matrix of elastic constants thus:
a
xx 1 V 0
E V 1 0
=
(1 - . * ) 1 - V
* xv 0 0
2
which evidently represents a constant stress system. If the panel is of

uniform thickness the internal stresses are evidently in equilibrium and
the internal state ofthe element is fully compatible with the linearized
boundary deformations. A good case might therefore be made for
sub-division of thin sheets into triangular elements when a linearized
displacement analysis is employed. This case is valid provided the
279
I. c T A I G and R. I. KERR
elements can be made sufficiently small but some serious limitations

can arise near structural boundaries and in regions of discontinuity.
The type of problem arising is illustrated in Fig. 8 where a number
of adjacent triangular panels are shown in a boundary region loaded
primarily in a direction parallel with the boundary. If forces exist
which give appreciable direct load along the boundary then the
diagram shows how such forces can load two panels a and c without
introducing a corresponding load into the intermediate element b.
Fig. 8. Triangular elements near structure boundary.
The result could be a fictitious discontinuity in the stresses in adjacent

panels which it would be very difficult to interpret in terms of actual
stresses. This type of behaviour is found in other groups of multiple
connected panels but can frequently be overcome by suitable choice
of elements. With triangular elements of finite size, however, some
effect of this nature is inevitable.
A more natural sub-division of multi-cell structures is obtained when
the sheet material is divided into quadrilateral elements whose boun-
daries can be made to correspond with actual boundaries or major
structural members. The general problem of a quadrilateral sheet
under specified boundary deformations has no exact solution known to
the authors even for the simplest boundary conditions. It is not,
however, considered necessary to search for a solution of ideal accuracy
since the boundary conditions themselves are severely restrictive at
element scale. Consequently a number of relatively arbitrary assump-
tions have been made regarding internal element behaviour and it is
frequently very difficult to establish the basic physical idealizations
which have been adopted by particular investigators.
The simplest idealization of the quadrilateral sheet element is the
shear-carrying panel. This is basically a force-idealization and as
applied to shear webs by Melosh and Merritt 9 for example, a simple
Garvey type of shear force distribution is assumed (Ref. Fig. 3b). The
280
Stiffness of such an clement for use in a displacement-type analysis is

in fact computed by inverting a flexibility matrix derived from strain
energy formulae.
A second approach adopted by the authors is to assume a constant
shear strain in the element defined by the change in angle 0 between
the panel median lines ac and bd (Fig. 9). The stiffness ofthe sheet is
Fig. 9. Shear strain in quadrilateral sheet.
obtained by setting up the matrix relating average shear strain to node

displacements, and using this to transform the sheet stiffness GAt
expressed in terms of constant shear strain over the panel volume At.
A simpler formulation results from this approach but the two methods
yield similar solutions in practical problems where the amount of taper
is relatively small. The same is not true of an earlier idealization
adopted by Turner et al. 2 in which the shear web was combined with
flanges and analysed by beam flexural theory with allowance for shear.
Appreciable errors were reported by Mclosh and Merritt for analyses
conducted using the flcxural element idealization, which were greatly
reduced in comparison with test results when the shear-web representa-
tion was employed.
The direct idealization of continuous sheet surfaces into line elements
and shear panels is not considered adequate when irregular structures
are to be analysed. Some authors such as Argyris*0 and Grzedzielski"
make use of this simple physical concept but obtain the effective
flexibilities of the idealized elements as a result of extensive subsidiary
computation. In an analysis based on displacement assumptions such
devices are unnecessary and a simple two-dimensional treatment of
rectangular panels is given by Argyris*0. The assumption of linear
displacement distribution within the clement is employed and this
leads to linear strain distributions from which corresponding stresses
are determined. The Virtual Work principle yields the clement
stiffness in terms of corner displacements.
The stiffness matrix of a rectangular sheet element i (Fig. 10) of
constant thickness / is the sum of two terms k u and k t , representing
281
Fig. 10. Rectangular panel geometry.
direct stress and shear stress contributions respectively. Both matrices

can be written in terms of three simple sub-matrices as follows:
*-=K a°
r cn 'Ci.- Et
AP,
yC[ 2 CML 1
(the partitioning corresponds with u- and -'-displacements)
where
'1 1 1 -2 " 2 - 2 I 1
1 2 -2 -1 -2 2 1 1
Qi = c . !
1 _2 2 1 - >2
60 1 1 2 2
2 1 1 2 I - 1 - 2 2
1 -1 1 1
/i = panel aspect ratio
1 -1 I 1
c 12 h
I 1 1 1
1 1 1 1_
First attempts at determining stiffness matrices for trapezoidal and
quadrilateral sheet elements assumed either that the panel was sub-
divided into triangles and/or rectangles or that internal displacements
were linear with respect to chosen cartesian coordinates. Both approxi-
mations yielded anomalous results and both gave quite dissimilar
solutions according to the actual details of the internal breakdown or
choice of axes. The solution eventually adopted 4 assumed linear dis-
plaeement variation with respect to a unique system of panel co-
ordinates defined by straight lines f = constant and »y = constant
making equal intercepts on opposite sides of the panel as in Fig. 11.
The panel boundaries are defined by £ = 0, tj = 1, 1 = 1 , tj = 0.
Internal strains are obtained by differentiation of the displacement
function with respect to cartesian coordinates, and stiffnesses by appli-
cation of the Virtual Work principle. A stiffness matrix is obtained
in the same form as for the rectangular panel but the sub-matrices
282
DISCRETE REPRESENTATION OF A I R C R A F T STRUCTURES
C n , C l2 , C22 are very much more complex functions ofthe panel dimen-
sions. The general formulation obviously reduces to the previously
quoted form when the quadrilateral becomes a rectangle.
It will be observed that the constituent sub-matrices ofthe rectangular
panel stiffness matrix are linear functions of the panel aspect ratio.
For aspect ratios of approximately unity the constituent sub-matrices
are of similar orders of magnitude, the shear terms being smaller than
,'sT
Fig. 11. General quadrilateral coordinate system.
direct stress terms in the approximate ratio G\E. If fi !> 1 or /J <^ 1,

however, C'n : C22 varies as fi2 and hence a magnified difference in
order of magnitude between the sub-matrices results. Now il can be
readily appreciated that a linear displacement representation can yield
some exaggerated shear strains and in panels of aspect ratio far from
unity certain shear terms can become very large in relation to direct
stress terms. In such cases a poor representation of structural behaviour
can be obtained, and this is probably the cause of the reduction in
accuracy noted by Gallagher and Rattinger 1 when mesh size is reduced
in the direction ofthe smaller element dimension.
This effect reaches the most serious proportions when the general
element stillness is used to represent the webs of slender flexural
members. In this application it is found that shear forces made a
considerable contribution to the bending resistance of the panel as
can be seen by imposing equal and opposite rotations on the two ends
of a rectangular element thus
U t = jl 1 1 1 : 0 0 0 0}
which gives shear forces Q kitUt
Q = {\ i i l o o o 0}^
which are shown in Fig. 12.
As ji decreases these forces can become very large and the element
representation quite inadequate. Far more realistic solutions are in
283
I. c T A I G and R. I. KERR
Ua.-I
"7 * -3/J
/
Gb Gl
5/i ' -V u 4 --i-<- 33
Fig. 12. Shear forces due to linearized flexure.
fact obtainable by using the shear-carrying elements previously

described, or by arbitrarily modifying the shear stiffness coefficients to
eliminate this bending resistance.
1.5. Miscellaneous Idealizations

We have considered so far the representation of continuous stringer
and sheet material by line and sheet elements. Practical structural
members are in fact discontinuous and may be reduced in stiffness due
to the presence of lightening holes, rib castcllations, riveted and bolted
joints, etc. Some of the problems of accounting for these factors will
now be briefly discussed.
Lightening holes and castellations in webs are generally allowed for
by reducing the effective modulus of the web, and since web stiffness
is usually a parameter of secondary importance very simple formulae
are regarded as adequate for this purpose.
For a square panel of side b, the modulus reduction quoted by
Kuhn 1 2 for a lightening hole of diameter D is given by
= 11 — — I in the clastic range

0
0
= 1 — I - - I for loading up to proof strength.
An clement representing a portion of web with several lightening

holes can be split up into approximately square regions around each
hole of side bx, b. . . . and a solid region of length a. Then the elastic
stiffness can be approximately represented by
G a + (A - A) + ip. - D m ) + .
<l b
where b = a -\- b, -f- />., . . . = total length.

Similarly, approximate formulae have been derived for castellated
webs.
284
When stringers or reinforcing members terminate abruptly in a

highly stressed region of structure they cannot be regarded as fully
effective in contributing to structural stiffness. It is usually possible to
perform an approximate diffusion calculation to determine the rate of
transfer of load to a stringer over the length of the first one or two bays.
Such a calculation can be used to determine the (average stringer
strain) : (adjacent skin strain) ratio and this factor can be used to
reduce the effective stiffness.
Major structural joints can sometimes contribute appreciably to the
local flexibility of a structure. Two methods have been used in making
allowance for continuous joints between sheet elements. In the case
of detachable panels located by flexible fasteners, the latter arc fre-
quently allowed for by estimating their distributed flexibility, adding
this to the direct and shear flexibility of the panel and computing new
stiffness properties equivalent to this combined flexibility.
A better idealization is obtained by introducing joints as discrete
elements having stiffness properties equivalent to the group of fasteners
associated with each node. To introduce such elements it is standard
practice to replace single node points by coincident node pairs, one
member of each pair being assumed to lie on the portion of structure
on either side of the joint. Joint stiffness is introduced as a special
matrix interconnecting the coincident nodes. This method has the
obvious disadvantage of doubling, locally, the number of nodes and
thereby increasing the size of compulation.
Finally, brief mention should be made of buckling and non-linear
load deflexion relations. 'I liese phenomena (In not necessarily affect
the basic structural idealization as methods of analysis can be adopted
which are based on the initial linear, unbuckled conditions. Discussion
ofthe methods employed in non-linear numerical analysis is, however,
outside the scope of this paper.
2. SOME ADDITIONAL PROBLEMS E N C O U N T E R E D IN

PRACTICE
In performing a large number of analyses of wing-type structures by
matrix displacement techniques, a number of practical problems have
arisen which have not been adequately covered in the general dis-
cussion. Some of these are described in the follow ing section.
2.1. Structure Size and Computer Limitations

In the analysis of complex, low aspect ratio wings considerable
restriction has been imposed on the choice of analysis grid by size and
speed limitations of the computer available. The stiffness matrices of
typical structures are of magnitude well outside the computer storage
capacity and extensive use has been made of matrix partitioning. This
285
I. G. T A I G a n d R. I. KERR
is in any case desirable from the point of view of computational effici-

ency since it provides a simple means of mathematically isolating those
regions ofthe structure which are not directly connected in the physical
sense. The problem solution is undertaken in terms of non-zero sub-
matrices of the stiffness matrix at an enormous saving in computer
time. For efficiency in operation size limits have been imposed on the
number of degrees of freedom associated with individual structural
partitions and on the extent to which partitions can overlap through
common structural elements.
These considerations make it uneconomic to include, in a single
analysis, overall stress distribution calculation and local stresses in
regions of rapid diffusion. They have also necessitated, in certain
cases, the sub-division of a structure into a number of separate regions
analysed independently and subsequently coupled by a comprehensive
elastic interaction calculation.
Onc ofthe largest analyses undertaken to date in a single unit is the
swept-back wing for which the grid is illustrated in Fig. 13. The wing
*s
y \ ^ K
^.•""^ / /1 -A *' *• \Â -,
\ \ ^x. .-•'
•1 \ •' u ' c ,'•"'•? ; y , aiutON
' \. ' **'
—-
--
Li / 7 ^r
fFUM.ai aiauCTuaai. t o n
-.«>« urns i i ' i s s i . i w INIUKAL mac m i l

- ABOlTlOKAl. CUD U " 1 S, «l»rtIORM OUIUNr,
Fig. E{. Plan view of a swept-back wing structural analysis grid.
was treated as asymmetric about the chordplane and the total number
of degrees of freedom involved was 690. l.vrn with this complexit)
certain regions such as that within the fuselage required more detailed
treatment. The standard procedure adopted in this situation is to
isolate the local region of structure for particular consideration and
break it down into a finer grid. Boundary conditions for this region
are derived from the force distributions (and sometimes deflexions)
obtained from the overall analysis. The local region can then be
analysed in detail without further reference to the overall analysis.
This procedure is at best approximate and its success depends on a
wise choice of boundary conditions. It is advisable to choose the
286
boundary at which compatibility with the overall structure analysis is

established as far as possible from the region of critical concern.
A more sophisticated treatment has been employed in the trailing
edge region of a wing where it was required to introduce the diffusion
of large discrete loads from flap hinges. A portion of the basic analysis
grid is shown in Fig. 14a together with the unrepresentative type of
HAILING t D O t
ra*p LOAD
W miMJUY QUID WITH LINCAILlSCD OtFLCC TI ON
r- UNCAFt I S CD i BOUNDARY J
1
I 1 1 , i
I I I
rH-F-.--[-H-i-i-i-
-•{—J-i-a..}-
1 1 1 1 1
" "• - r - v - T "

T - • - - - . J. ! (
1 •r
W JJECONDAtY GBIP WITH FUPFUSLNTATI V t BOUNDARY DtlLtCTIQN
Fig. 14. Fine-mesh analysis in a diffusion zone.
edge deformation which the linearized analysis would entail. To

represent the structure more adequately the aft bay ofthe analysis grid
was sub-divided into the much finer network shown in Fig. 14b and
a preliminary analysis was conducted for unit displacements imposed
at each of the primary structure nodes in turn. Deformations were
constrained to be linear between basic nodes along the forward boun-
dary hij and were unconstrained between basic nodes elsewhere. The
stiffness of the sub-divided structure in terms of basic node deformations
was computed and utilized in the primary analysis. Back-substitution
of the displacements along hij from the primary analysis yielded the
necessary boundary conditions for the secondary calculation.
2.2. til-conditioning Resulting from Structural fdealization

The matrix-displacement type of analysis formulated in terms of
absolute displacements tends to be ill-conditioned since the equations
of equilibrium ofthe structure nodes are derived from small differences
287
I. C. TAIG a n d R. I. KERR
of the displacement variables. Provided that sufficient care is taken in

formulating the stiffness matrices this ill-conditioning does not usually
pose problems when present-day computers arc used for the solution.
Several instances have, however, occurred where solutions have not
been obtainable by standard programmes due to the stiffness matrix
becoming too nearly singular. These have generally been traced back
to basic structural idealization and it is therefore relevant to discuss
the physical aspects of the problems.
Two primary causes of severe ill-conditioning have so far been
discovered.
2.2.1. Range qf magnitude of stiffness coefficients. Most o f t h e matrix
programmes written for the Deuce computer employ block-floating
arithmetic in which sub-matrices are stored and manipulated with a
fixed binary point. If the elements of an individual sub-matrix vary
widely in magnitude, then, since the number of available binary places
is limited, the smaller numbers must lose a number of significant
figures. Structural idealizations which yield widely differing stiffnesses
in adjacent regions must suffer from this type of rounding-off error.
It should be noted here that small numbers in a stiffness matrix are
not necessarily less significant in terms of their influence or the solution
than adjacent large numbers.
This trouble frequently arises when near-planar sheet elements occur
in a box analysis with node points unsupported by internal webs. The
stiffness of the sheet in a direction normal to the local surface is then
smaller than the in-plane stiffness by a factor of the order of tan 2 a
where a is the change in slope of adjacent structural members. Since
a may typically be <0-01 radians this could mean a range >10* in
the stiffnesses at a given point in two directions. This may be sufficient
to cause programme failure or an unacceptable loss in solution accuracy.
A simple expedient is currently adopted to overcome this difficulty. In
all cases where near planar panel groups have unsupported nodes the
number of degrees of freedom is reduced by relating the local normal
displacements to the displacements of adjacent supported nodes. Tin-
error thereby introduced is negligibly small as a direct result of the
low stiffness values which are the source of the trouble.
Another fairly frequent cause of this trouble is the occurrence of
slender llextiral elements and substantial sheet or boom elements in
close proximity. A particular instance arose when analysing an all-
moving tailplane mounted on a flexible support spigot. The mounting
was represented by adding its flcxural stiffness to an otherwise singular
structural stiffness matrix. The mounting stiffness was several orders
of magnitude smaller than the remainder of the stiffness matrix and
the problem could not be solved by standard methods. It was nccessary
to assume a rigid mounting and introduce the support flexibility
subsequent to the solution of the main problem.
288
68S
JEIIJOB 3qj m o j j s.iii.udo.id puisXqd jU3J3jpp ApqSqs j o ajnjanjjs
3[qEuosE3j A|[i33iSj*.qd E sju3S3jd3J *a*i 'ju3jsisuo3-jps Apj3i*dui03 si EjEp
ojEjnDDKUt a q i j s q j pjpiAo.id .in.u AjjEJanisS SI s i q x 'sujsjjEd UOIJIIIU
- . i q p p put? SSOJJS ip;.UAO J33JJE Ajsnouas j o u pjnoqs s j u j u i o p [Ejnjsnjjs
S m j u 3 S 3 j d 3 j u i sapuanDDEUi JEDOJ JEqj S J 3 3 U I S U 3 SSSJJS JSSUOUIE psjd333B
XppiM si ji 3DUIS ASJOAOJJUOD 3|qE.i3pisuoo oj osu UOAIS SEq juiod
s i q j , 'EjEp 3isEq OJEJUODEUI j o ssn s q j UIOJJ pajjnsoj 3AEq suopnjos
ui3|qo.id ui SJOJJO s n o u o s 3J3qAA s3urmE.tSo.id .i3jnduio3 DpEuiojnE Suisn
S 3 j n j 3 n j J S X 3 i * d u i 0 3 J 0 SlSAi*EUE 3 q j Ul p3JjnDDO 3AEq S33UEJSUI pUOAOg
iCwm33y V}V(j - g ' g
'S3UIEJJ 3UEjd JOpUO|S

SUISAJEUE U3q.w UOIJEUI.IOJSUE.IJ mop33Jj j o 33J§3p 33.iqj o j anoj B
,).)npo.ijut o j put; J3UUEUI siq j ui oAy oj 3.mj3n.ijs adAj-SuiM E u t sspou j o
JIEd q3E3 JE UJOp33JJ JO S33J§3p X1S 3 q j UIJOJSUBJJ OJ 33U3E-ld p.IEpUEJS
3JOJ3J3qj SI JJ *S3iqEUEA JUOpUodjp-A'JBOU 3qj JO 3UO S m j E u u i n p
Aqo.ioqj puE 3UE|dp.ioq3 s q j o j JEUIJOU u o p a a j i p E UI sjpjd s p o u j o
sju3m33Ejdsip 3 q j SuijBnb.) sajqEUBA j o UOIJELUJOJSUEJJ E S u p n p o . i j u i
Aq 'qdE.tSE.iEd snotA3.id oqj j o JEqj oj ,i3uuEm JEJIUIIS E u i 3UIODJ3AO
st uj3jqo.id 3 q j j ' X I J J E U I SS.IUJJIJS s q j j o AjuE[nSuis-jE3u o j s n p so.mpEj
uoijtqos S n i o n p a i d jnoipi.w j u o p u o d o p u i j\pjo[duioD SB psjE3Jj s q
jou p j n o s sspou j o sjiEd q a n s j o s u o i x . q p p 3qj jBqj S3SXJBUB jBnjDE j o
j . i q u m u B UI p u n o j u.")jq si;q J J *3UBi;dpjoq3 3qj oj (BIUJOU u o p o j . n p 3qj
tn p3jdno3 X p s o p 3q SAEJW|B \\X.W f puE . sjuiod juaoEfpB j o s u o p c s j p p
-siujiiUJii/ilsi// j/ion jiujuCpo fo Suiji/no-) •(•• -iii.[
N0liYH>J0J3a irjldxj. /q*
iKl«l3»i.9ia 3Mi\rll» I I I I O I I D I N
aids aiiNUOJIaM ("0)
s q j g J ' S i J UI uoij33s jBds [B3idXj E Xq po]B,ijsn|jT SutM uiqj B SB

q a n s sojnjonjjs .i3pu3i;s JOJ s n q j , -sajdpuijd (BoisAqd .qdmis m o j j sojqi;
-1JBA JU3m33B|dsip UIEJJ33 U33MJ3q diqSUOpBpj JE3UIJ-JB3U E JOipjud
oj sjqissod si JI msjqo.id j o sossEp UIEJJSO U J 'uioiqo.id |Ejnj3njjs 3qj
sjEjnuijqj oj pssn s3|qBUEA j u s p u o d . i p u i Ajposoddns 3qj j o sjU3oqj303
3qj u33A\joq d i q s u o p B p j JE3uq-JE3U E j o 33U3JSIX3 s q j si S u i u o p i p u o 3
-JJI XUJEUI j o 3snB3 s q j ^ •suoijvuuo f p pudjvCpv fo Suijrfnoj "2"g"2
KaTaaxomaxs X-IVISOHIV d o N O i x v j . N a s a a a a a axaaosiq

I. c . T A K ; and R. I. KERR
structure. The important type of data inaccuracy is that which leads

to inconsistency in the mathematical representation of the structure.
Three sources of such errors arc worthy of mention:
2.3.1. Special stiffness matrices. It is frequently desirable to introduce
into a stress analysis the influence of non-standard structural elements
for which stiffness matrices are computed manually. Often the non-
standard elements arc not amenable to accurate calculation and still-
nesses have been quoted to, say, slide-rule accuracy based on very
simple calculations. A particular instance can be quoted where a fm
was attached to a flexible beam whose stiffness was represented by
flexural stiffness coefficients in one direction at two adjacent points.
I IK stillness matrix was computed to slide-rule accuracy and con-
sidered to be an adequate representation of the support. However, the
coefficients ofthe matrix did not satisfy equilibrium conditions to more
than three significant figures and this inconsistency was magnified to
such an extent by the problem sensitivity that a 10 per cent error in fin
root bending moment resulted. It is therefore recommended that in
displacement analyses any manually computed stiffness matrices should
satisfy equilibrium conditions to as many significant figures as can be
accepted by the standard computer programmes to be used.
2.3.2. Boundary transformation matrices. A similar source of error to
the above has arisen when special transformations of variables have
been introduced. This is frequently necessary at structural boundaries
where it is required to relate displacements at idealized nodes to dis-
placements at adjacent datum points such as fuselage attachments, or
where two incompatible structural grids meet. Transformation
matrices are normally computed manually and it has again been found
necessary to ensure absolute consistency in using geometrical dimen-
sions, however approximate the dimensions themselves may be.
2.3.3. Biased round-off errors in element stiffnesses. An instance occurred
recently where a serious loss of solution accuracy resulted from the use
of a new element stiffness programme which was itself accurate in all
respects except that a biased rounding-off process was inadvertently
used. This resulted in a small equilibrium error in the stiffness matrix
which was magnified by the analysis and occurred in the same sense
in some 200 elements distributed all over the structure. The result
was an error which appeared to be of the order of 50 per cent in the
displacements of a large portion of the structure although the original
error was in the seventh or eighth significant decimal digit. The same
problem formulated using a programme employing unbiased rounding
off resulted in a perfectly satisfactory solution. This was, however, an
exceptional case since the problem had already been rendered sensitive
by the inclusion ol a wide range of element stillnesses.
It should be noted that the large error magnifications quoted above
are only associated with stiffness matrix inconsistencies. It is readily
290
shown that an error of the scale-factor type (such as a wrongly estimated

effective boom area) leads lo a local error in solution which is approxi-
mately proportional to the element strain. Errors which result in loss
of equilibrium, however, can be shown to be magnified in the ratio
of absolute displacement : element internal deformation, and this
magnification factor can easily reach values ofthe order of 10*.
3. THE INFLUENCE OF SOME PARTICULAR IDEALIZATIONS

ON A SAMPLE BOX ANALYSIS
When preparing this paper it was decided to undertake a small pro-
gramme of work on a sample structure to provide an illustration of
some of the factors discussed qualitatively in the previous sections.
Ideally a parallel programme of experimental work would have been
desirable but this has been ruled out on timescale and cost grounds.
In the absence of experimental evidence, the basis of comparison for
the studies undertaken is a fine-mesh analysis representing the same
structure and making use ofthe best techniques suggested by experience
and a review ofthe literature.
3.1. Choice qf Structure and Scope of fnvestigalion

To illustrate a number of useful points it was decided to perform a
series of analyses on a structure which was reasonably representative
of present-day practice, demonstrated the influence of low aspect ratio
and non-rectangular geometry and yet was sufficiently simple to be
analysed economically. The basic structure chosen is illustrated by the
grid shown in Fig. 16 and the fine-mesh grid used as the basis of
comparison is illustrated in Fig. 17.
In previously published comparisons of various methods of structural
analysis such as Rcf. 1 it is difficult to establish which of several
differences in idealization contributes most to the large observed
differences in .HI uracy. In this investigation the same basic n chnique
and the identical set of computer programmes are used throughout and
variations in the idealization or problem data are made independently
to eliminate this source of confusion.
The investigation covers the influence of a range of idealizations on
the structure stresses and deformations under a unit tip load, the
flexibilities, and on the vibration characteristics as defined by the lower
natural frequencies and normal modes.
The basic structural idealization is described as follows:
The cover sheets are treated as triangular and non-planar trapezoidal
elements as appropriate. Symmetry about the chordplane is assumed.
Internal shear material is represented by constant thickness webs
concentrated at the grid lines. The web elements are treated as anti-
symmetrically loaded two-dimensional trapezoidal sheets with a correc-
tion made for the shear contribution to flexural stiffness.
291
OKI
ro s
to Q.
ic
r.
-
-
-
Fig. 16. Basic delta box structure grid. Skin thicknesses are shown thus 0.064 (in.). Boom areas {in.*)
are shown on grid lines. Web thickness {in.) are shown alongside.
-
x-
-
-
-
-
-
•d
73
-
x
y
-
>
—
1-0 y.
• z
x
73
-
x-
>
x
-.
•s.
Fig. 1 7. Delia box structure. Fine mesh grid. Modified boom areas shown on grid lines.
-
x
All other sizes as for basic structure.
-
O
H
-
-
-
Stringer and flange material is represented by concentrated booms

at the grid lines.
T h e loading case used for the following comparisons consists of a
single load of 10 lb applied at the tip node 1 (node 66 in the fine-mesh
grid).
The coordinates of the grid and the lumped masses associated with
each node are given in Table 2. Sizes of the structural members are
indicated in the grid diagrams Figs. 16 and 17.
The variants on the basic analysis are defined by Table 1 below.
Table I. Sample Analysis Programme
Run Idealization Changes Information Computed
1 Basic structure Deflexions, stresses

Flexibilities
Normal modes
2 Skin elements replaced by shear panels Deflexions, stresses
and increased area booms Normal modes
3 Displacement transformation introduced Deflexions, stresses
to make ribs rigid Normal modes
4 All quadrilateral skin panels replaced by Deflexions, stresses
pairs of triangles Normal modes
5 Constant-depth structural grid used, all Deflexions, stresses
thicknesses and areas scaled to suit
6 Internal web elements with shear contribu- Deflexions, stresses
tion lo (lexural stiffness Normal modes
7 Local errors introduced in stillnesses Deflexions, local stresses
7.1. Large local scaling error
7.2. Small consistency errors
a Mesh size halved in both directions, re- Deflexions, stresses
allocate boom areas. Redistribute in- Normal modes for (a) coarse.
ertias over finer grid (b) fine mesh mass distribution
3.2. Results and Discussion

Figure 18 shows the deflexions at outboard sections which result
from applying unit tip load. All the coarse mesh analyses under-
estimate the tip deflexion compared with the fine-mesh case with the
exception of that for the untapered box. The latter result is regarded
as fortuitous since the increased shear deformations in this analysis
appear to be compensating for the local effects of mesh coarseness.
At stations other than the tip the basic case and the triangular panel
analysis give results which are indistinguishable from each other and
barely distinguishable from the fine-mesh results. Most analyses give
the same form of deformation, the principal exceptions being the rigid
rib case and the boom/shear panel cover idealization.
294
S62
m o o q OSEJOAE oqj j n q SISXJBUB q s o i u - o u y oqj Xq d n u.woqs si s.ioSuujs
ojtnpoiujojui oqj ui joojjo S E J JEoqs ojqEOopou y 'SOSBO j o q j o oqj qjiw
pojBdtuoo uoijnqijjsq} ui UOUEIJBA ojqEJopisuoo A\oqs SOSXJBUE p u s d
.iBoqs/mooq p u E q u p i S u oqj j n q ojnjonjjs >pEq-jdoMS oqj j o JBOJ
oqj SpjEMOJ SSOJJSJO UOIJEJJUOOUOD opsijojoBJBqo oqj Moqs SOSXJBUE oqj
[JE jBqj uoos si J J *XEq oqj uiqjiM j u o m o p qoBO u i SOSSOJJS OMJ oqj j o
UEom oqj OJB SISXJBUE q s o u i - o u y oqj J O J p o j o n b SOSSOJJS o q x 'u.woqs si
AEq JOOJ oqj u i SOSSOJJS m o o q J O j o j d osiMpjoqo E 51 - S i j u j 'SUOIJBZI
-jBopt [EoujomooS oqj (JOJ pojoojjoo Xjdims oq UED x o q pojodEjun oqj
j o OSEO oqj ui J O ) Xq pojoojjBun O.IE osoqj ooms SOSSOJJS qo<w JBoqs puB
m o o q oqj o j OOUOJOJOJ Xq p o j E d u i o o XJJUOIUOAUOO jsoui OJE SOSSOJJ$
•suopEDqddB DIJSBJOOJOB ui juBoymSis XJOA oq pjnoo qoiqM oouoj)
-toui ui oouojojjip ojqEioo.iddE SMoqs SISXJBUE jouBd jBoqs/tuooq o q j ^
•suopdmnssB q i j p i S u oqj JJUB sqoM pozyjEOuij oqj ui joojjo S u i p u o q
-jEoqs oqj oj onp osoqj XJJBJOUOS OJB s o p u E d o j o s i p JSOSJEJ oqj^
Z£ oo-z 0 OS I LZ
09 oe-z 0 531 9Z
li 8t--E 0 001 SZ
91 99-t; 0
uOS w
H 9G-E 0 £Z
99 HE 0 SZ zz
8£ Ot-Z 0 0 rz
IS 99-1 oz 09 02
06 K-Z 06 SE1 Gl
SO I OOE 02 Oil 81
1)11 ZO-E 0Z <i8 il
101 ZL-Z 02 09 91
Ee OOZ oz SE SI
ZE J-E-I Of- Oil fl
ot* 6*1 £1
u
88
n-z Of Z\
n-z Of
06 I
II
T-8 tz-z S6
Et 091 Of Oi 01
S9 mi 1 09 081 6
99 9M 09 SSI 8
69 08-1 09 0E1 i
9E 0c.I 09 S01 9
ot: 99-0 08 061 S
9<J- ZZ-I 08 S91 f
OE 08-0 08 OH £
si J-E-0 001 ooz Z
in OH) 001 Sil I
ssupy z A X 3
P°N
t>]t)Q VI/JJUJ p u s tCqmtOttf) •£ fjqvj^
sa-anxoawixs XJVHOHIV J O NoixviNasaadaa axanosio

I . C . TAIG a n d R. I. K E R R
stress is predicted with remarkable accuracy by the basic analysis and

those involving triangular skin panels and constant depth idealizations.
Spar shear stresses arc plotted in Fig. 20 for the same bay and these
show much greater discrepancies between the various assumptions.
Once again the rigid rib and boom/shear panel idealizations show by
far the greatest discrepancies. The shears for the untapered box differ
from the majority of curves by an amount approximately proportional
to the local direct stress, which suggests that a slope-shear correction
could be introduced after the analysis to give good agreement with the
1 1 J
1 y— .
BASIC CASE
CSSH/kHCAU PANEL ANALYSIS/

1 1 //
ill-ID mi ANALYSIS / /
/
' 1 / /
FINE MC
1 / /
SH ANALYSIS / /
SJ
/ XtiSiSf
. -^-"""^
Fig. 18. Effect of idealization on deflexion under lip load.

296
other predictions. Results for triangular and quadrilateral skin panel

assumptions are again barely distinguishable but a noticeable deviation
from the fine-mcsh analysis occurs locally at the rear spar. The dis-
crepancy due to the shear-bending stiffness of the webs is again
approximately proportional to the bending stress as might be antici-
pated.
Figure 21 illustrates the effects of two types of error introduced in
runs 7.1 and 7.2. In run 7.1 the stiffness of the spar web element
coupling nodes 19 and 26 was halved. In run 7.2.1 the stiffness ofthe
same clement was modified by the introduction of a numerical error
-MOM/SHIAR M « L ANALYSIS
(UGIO tit ANALYSU
sfA» smssJ
Hut MtSH
STtlNOtt STBISSJ
Fig. 19. Chordwise variation of direct stress in root bay.

297
in a single term in the stiffness matrix of magnitude approximately

0*2 per cent ofthe correct term. In run 7.2.2 a similar error was intro-
duced into the stiffness of the web element coupling nodes I and 4.
The effect of the large scaling error on deflexions is seen to be very
small and that of the root bay consistency error is negligible. A larger
effect is seen to result from the tip bay consistency error indicating a
high magnification of errors of this type. The influence of the errors
on stresses in an inboard and outboard bay shows that the inboard
consistency error has negligible effect. Both the scaling error and the
B AS I C CA S I
BOOM / S r i ' . AH H i l l ANALYSIS.
t I G I D RIBS
- RIANi ULAd SKIN PANELS
UN T A P ' a t 0 BOX
Fig. 20. Chordwise variation of spar shear stress in root bar.

298
outboard consistency error have large effects on local shear stresses

but only the latter has any ajjpreciablc influence on direct stresses.
These results confirm the statements made in section 2.3 with regard
to the importance of numerical consistency as opposed to the accuracy
of representation of local stiffness.
The foregoing results have indicated no appreciable difference
between triangular and quadrilateral panel representation of the box
skins. Figures 22 and 23 have been prepared to highlight the differ-
ences in skin stresses obtained by the two idealizations and compare
Fig. 21. Effect of errors on de/lexions mid local spar stresses.

299
them with the fine-mesh analysis. The average stress in the triangular
sub-panels agrees very closely with the stress in the quadrilateral panels
in the basic analysis. The discontinuity in adjacent panel stresses
across an arbitrary boundary, discussed in Section 1.4 and illustrated
in Fig. 8 also appears in this analysis in a quite pronounced form. In
(«,. BASIC CASE.
STtims IN L I S / I N ' -
lot I0L8S.TIP LOAD APPLIED
(b) TRIANGULAR SKIN PANELS.
Fig. 22. Comparison of skin principal stresses.
regions removed from the point of load application the difference in

stresses between adjacent triangular sub-panels is considerably greater
than the stress variation over the four corresponding trapezoidal sub-
panels of the fine-mesh analysis. A chordwise plot of panel stresses
based on the trapezoidal representation would agree with the fine-mesh
analysis better than the triangular representation.
300
n
-
x
X
STRESSES IN L B S / I N * FOR IOLBS. TIP LOAD APPL -
-
x
[JO
W
X
-:
X> c
-
-
* i -
>
*' I ' a / i V i i f V«o AP ZIPS' /' -

-
Fig. 23. 5iWfl principal stresses from fine-mesh analysis. a
o
-
-
-
-
Both of the coarse-mesh analyses seriously underestimate the peak

stresses in the immediate neighbourhood of the applied load. This is
an artificial situation since point loads of any magnitude will never be
applied to the tip of an actual wing and the local effects of distributed
loadings arc not likely to be nearly so marked. It appears that the
coarsc-mesh trapezoidal idealization can be used with confidence more
than two bays away from the point of application of the load provided
that some allowance is made for local stress peaks in diffusion areas
such as the rear panel at the root. The stress obtained from the fine-
mesh analysis in this region could have been closely approximated by
extrapolation ofthe coarsc-mesh stresses using simple graphical means.
Table 3. Comparison of Natural Frequencies
Natural Frequency {cjs)

Idealization
Mode 1 Mode 2 Mode 3
Basic case 10-93 26-70 38-4

Boom/shear panel 11-24 26-6 38-3
Rigid ribs 11-66 30-9 57-4
Triangular skin panels 10-83 25-9 38-9
U n t a p c r e d box
Webs with shear-bending stiffness 11-84 30-8 42-5
Fine-mesh structure basic inertias 10-84 260 36-8
Fine-mesh structure fine-mesh inertias 10-91 26-75 38-2
The influence of structural idealizations on vibration characteristics

is indicated by Table 3 which records the first three natural frequencies
of the structure computed by the various analyses. The results are all
in reasonably close agreement with the exception of the rigid rib
idealization and, to a lesser degree, the analysis using shear web
stiffnesses including the shear-bending terms.
If wc assume that the fine-mesh analysis with fine mass distribution
represents the best approximation to the correct solution an interesting
observation results:
T h e basic analysis yields a noticeably better approximation than
either the triangular skin panel analysis or the fine-mesh structure with
basic mass distribution. This result suggests that there may generally
be adverse effects on accuracy if a structure grid is used which is finer
than the inertia grid.
Most ofthe analyses yield similar normal mode shapes but in Fig. 24
it is seen that the rigid rib analysis is quite unable to represent the
more complex deformation patterns, which accounts for the poor
estimate of frequencies.
302
DISCRETE R E P R E S E N T A T I O N OK A I R C R A F T STRUCTURES
M BASIC CASE 38 4. C.P-5.
(b) RIGID RIB CASE 57 a C.P.S.
(C) FINE MESH ANALYSIS 36 8 C P . S .
Fig. 24. Comparison of contours for '.In! vibration mode.
303
i. c . T A I G a n d R. I. KERR
3.3. Conclusions
The above brief discussion leads to a number of general conclusions.
3.3.1. For structures without major discontinuities in geometry or
loading the coarse mesh of the basic analysis appears to give results of
adequate accuracy for preliminary structural assessment. This agrees
with the recommendations of section 1.1.1.
3.3.2. Geometrical idealizations appear, generally, to have less
influence on stress distribution than those affecting clement behaviour.
Even in the case of shears in the untapered box a "slope-shear" correc-
tion could be introduced to yield quite good results. The least satis-
factory of the idealizations investigated appears to be the assumption
of rigid ribs. It is also confirmed that the representation of shallow-
web stilfnesses is a matter of considerable importance.
3.3.3. Vibration analyses are most accurately performed when the
grid for structural analysis corresponds with that for lumping of masses.
3.3.4. There is little to choose in accuracy between triangular or
quadrilateral panel representation of skins. The latter, however, have
an advantage in ease of interpretation and appear slightly better for
the calculation of natural frequencies.
3.3.5. It is confirmed that accuracy of specification of individual
element stiffness properties is of secondary importance compared with
numerical consistency in computations based on the assumed properties.
PART II. APPLICATION O F FORCE ANALYSIS T O

FUSELAOE STRUCTURES
1. S T R U C T U R A L IDEALIZATION
1.1. Choice qf Structural Grid

The chief application of the matrix force method by the authors
has been in the solution ofthe stress fields in aircraft fuselages.
Since it is obvious that the structural grid should follow- the actual
members, one set of lines is fixed along the frames and the other along
the stringers. In many cases it is not necessary or possible to split the
panels along every stringer, and so a certain amount of choice is
available as to the actual choice of the panel sizes.
1.1.1. Size qf mesh. Because of the complexity of the structure of
modern high-speed aircraft, the size of the mesh employed is often
governed by the capacity of the computer available.
Thus, sub-division of the skin panels between frames is considered
to be important only in the case of certain specialized local loads (e.g.
pressure) and so such sub-division is considered unwarranted in large
over-all analyses.
Sub-division ofthe panels normal to the frames is largely an intuitive
304
procedure, but it has been shown by the analysis of part of a fuselage

on a coarse grid and then on a fine one, that if the load paths, cut-outs,
etc., are all reasonably well represented, then the resultant stress
distribution obtained is largely independent of the details of the
idealization.
1.1.2. Choice of suitable element shapes. Because o f t h e large depth of
the structure of a fuselage the flexibility is hardly dependent on the
variation of flexibilities within the individual elements. Thus, the
simplest possible elements representation may be used. The authors
have thus chosen three basic elements which are as follow:
(a) Booms. These are assumed to be constant area members and

each is capable of carrying two unit loads. Each load consists
of a unit end load applied at one end, reacted by a shear dis-
tributed uniformly along the length of the element.
(b) Shear Panel. These are panels which are capable of carrying
shear only. Since in the analyses so far carried out by the
authors, the fuselage frames have been assumed plane and
mutually parallel, these panels arc, in general, twisted trapezia.
The unit loads along the non-parallel edges are assumed to be
uniform shears, whilst along each parallel edge the loads are
such that the resultant in any direction is proportional to the
projection of the opposite edge in that same direction. It will
be seen that this definition of shear gives rise in the limiting
case, to the normal concept of shear in a plane rectangular
panel. This unit loading is also considered acceptable in the
other limit of a triangular panel. Whatever the shape of the
panel the flexibility is taken somewhat arbitrarily as A/Gl where
A is the panel area (taken as the mean of the lengths of the
parallel sides multiplied by the distance between them), G is the
shear modulus, and / the panel thickness. In regions of rapid
curvature, the authors have used panels which arc formed from
a series of the simple panels given above, so that the actual skin
contours may be followed more closely. The flexibility of the
panel is then taken as the sum of the flexibilities of its con-
stituents.
(c) Frames. Each of these is treated as a composite element, whose
flexibility in terms of horizontal and vertical loads at its nodes
is assumed to be found by a previous separate calculation. Al-
though this approach has been found to be acceptable in the
past, it will present difficulties when, with the advent of a faster,
larger computer, iterative analyses arc carried out to cope with
design problems and the effects of non-linearity. Thus, the
existing programmes are at present being modified to include
the computation of the frame flexibility in the overall analysis.
305
I. c . T A I G and R. I. KERR
1.2. Idealization into Line and Sheet Elements

The idealization of the structure of a fuselage is easier than that of a
shallow box, such as a wing, in that the bend and shear stillness of
each element is small compared with the overall fuselage. Thus, it is
sufficient to lump the end-load carrying area of the skin and stringers
into concentrated booms, and to connect these by panels which carry-
shear only. O n the few occasions when it is necessary to represent the
bending stiffness of longerons, such members can be split up into two
booms and a shear panel without much trouble.
1.3. Idealization of Element Behaviour

T h e boundary conditions for a force element are assumed to be
linearly varying loads along the edges, and are thus the exact counter-
part of those normally used for displacement structure. This leads to
non-compatible deflexions except at the nodes, in place of the hypo-
thetical unbalanced restraining forces assumed in the displacement
method, and explains why the force method gives a more flexible
answer than docs the displacement method.
Because of the simplicity of the elements used in the force method
error magnification of the type experienced in the displacement
method (due to an incompatible clement flexibility) cannot occur.
2. SOME P R A C T I C A L PROBLEMS
2.1. Structure Size and Computer Limitations

In analyses so far carried out by the authors, fuselages with approxi-
mately 30 frames, 3000 nodes, 5000 redundancies have been success-
fully solved. With the use of interaction technique, structures much
greater than this have been analysed.
Thus, analyses of this size stretch the capacity of DEUCK to ils
limits, and so, as in the displacement method, extensive use is made
of partitioned matrices and of the fact that many of the partitions are
null.
However, in spite of the large number (30) of panels in a bay with
which the computer can cope, limitations on the idealization, especially
round cut-outs, have had lo be imposed in the past.
A further result of the large size of analyses mentioned previously is
that it becomes beyond the scope of manual calculation lo set up b{]
and h, matrices, and thus automatic methods to set up these two
matrices have been developed.
The method used to set up bx is basically similar to that described
by Argyris. 10 However, since the number of redundancies is dependent
almost entirely on the number of shear panels within a bay, the
redundancies are set up first in terms of these panels, and not, as by
306
Argyris, in terms of the boom loads. A further difference as compared

with Argyris is that the whole structure is considered from the start,
rather than a basic cylindrical shell.
Thus, the shear flows in the panels must satisfy certain equilibrium
conditions (zero vertical load in the symmetric case, and zero lateral
load and torque in the antisymmetric onc), and these may be
represented as:
ah, = 0
where h, are the terms in h, corresponding to shears in the panels.

Similarly if h0 arc the shears due to that part of bu representing the
overall loading, then:
ab 0 = I
and thus,
a[b 0 i £,] - [/ : 0]
or
a /
[h0 \ b.]
A L-\i .V„
a
where A is any matrix which makes square, and non-singular,
.1
and defines the form of the redundant force system and X 1 and X.2 arc
unknown.
Thus,
I i 0
ha:hl
X X \Xm m
and then, if A\ and A'2 are chosen as Xx = 0XX2 = I
a
/)„:/),
A
and thus
a
hn:bx T
A
where T is a transformation depending solely on the geometry of the

idealized structure.
No attempt is made to choose a matrix A\ SO that the resultant D
matrix is well conditioned. In fact the simplest possible A matrix is
307
used and in the symmetric and antisymmetric cases .4. and .4 , have
been extensively used.
1 -1 o. . . 0 1 -2 1. . . 0
(1 1 -1 . . . 0 1) 1 -2. . . 0
(1 0 i . . . 0 1) 0 1. . . 1
AA =
0 0 0... -1 0 0 0 1
In the past, D has been set up using the redundancies formed above,
and then, using an A matrix based on the latent vectors of the corres-
ponding diagonal partition of D, a second set of redundancies, well
conditioned within the two-bay system, has been formed.
However, with the advent of larger, faster computers which will be
used to solve finer mesh problems without perhaps sufficient increase
in the number of digits, the problem of conditioning is being further
investigated by the authors, and the possibility of curing ill-conditioning
due to the cross-coupling of adjacent two-bay sets is also being con-
sidered. This cannot be covered completely without the use of
redundancies covering more than two-bays, and hence losing the
banded form of D, but some improvement should be possible.
The part of b0 formed above covers only the overall loading (shear,
bending moment and torque) and not the actual application of local
loads. O n onc analysis carried out by the authors there were several
hundred of these local loads, and thus several programmes have now-
been incorporated into the overall analysis to deal with such problems
as fuel loading, etc. A further difficulty experienced in the past was the
writing of R matrices. Thus the concept of h„ and R as two separate
entities has now been abandoned by the authors to some extent. The
effects of the six unit accelerations on the aircraft inertias are now
supplied to the computer together with point loads equivalent to the
aerodynamic loads.
Then, the computer automatically evaluates the resulting applied
loads and then the aircraft acceleration. Next the unit inertias are
factored and added to the aerodynamic effects, and lastly the overall
effects (bending moments, etc.) are automatically added in giving the
final b0R complete.
2.2. Ill-conditioning Arising from Structural Idealization

It has been found in the past that one of the major difficulties of
ill-conditioning is recognizing whether or not it has occurred. Thus,
it has been the authors' experience that the normal DD ' check is
virtually useless. (Note. D = b[fb l and the nomenclature used is that
308
given by Argyris.1") For instance on one analysis, the DD • check

gave a matrix with off-diagonal terms greater than those of the
diagonal.
However, after extensive modification to the redundancies to improve
this conditioning, a new D matrix was formed which gave DD ' with
a reduction of a factor of 10 on the oil-diagonal terms compared with
the first D.
o LOAD C»«RY:KIJ MATERIA-.
FAIRING.
Fig. 25. End section of a fuselage.
However, the stress matrices based on the two sets of redundancies

differed by only some 3 per cent. Thus in later analyses the only
check on the accuracy of the results has been the displacement com-
patibility check obtained by formation of b',fs (which should, of course,
be a null matrix). If the terms of this matrix are uniformly small then
309
the results have been accepted. Sometimes, evidence of slight ill-

conditioning round extensive cut-outs has been found, but this only
shows up as a loss of two or three binary places in b[fs. With the form
of the redundancies used, the authors have found no evidence of ill-
conditioning between adjacent two-bay sets.
In one of the very early analyses carried out, due to design changes
in the actual structure between the idealization and the actual compu-
tation of the stress matrix, some panels were inserted with a very small
ion ooo
80000
s.
a***
\
60 000
/ \
r
/
/ \ \\
/ \
10,000 f
. A S i
\
• ^*AAy
\
\
10,000
V \
v.
^ s
^Va*N
TRUE VALUE
*ENCINEER5 (END m o R T " C.AStO ON L0AOS)
FUSELAGE CROSS-SECTION CONSTRAINED

TO R E M A I N PLANE.
Fig. 26. Effects of boundary conditions on the end load in boom " A " ,
thickness in an attempt to represent cut-outs, and this did give rise to

conditioning trouble, but since programmes were written to improve
the conditioning of the redundancies, variations of panel flexibility of
100: 1 have been treated without trouble.
2.3. 'The Effects qf Boundary Conditions

Some difficulties have been experienced over the choice of suitable
boundary conditions especially when only a section of a fuselage has
been considered. Thus, as an example, the half fuselage section shown
in Fig. 25 was the end section of one analysis. The end loads were
calculated from engineers' theory (e.g. L = A,yx -f- C. due to a
bending moment) and the resulting loads in two ofthe booms are shown
in Figs. 20 and 27.
310
The results showed a much greater proportion of shear than would

be expected being carried in the side wall (between A and B) and also
because of the rapid changes in boom loads over the end bay, the
boom load distribution at the end was suspect. Thus, the end con-
ditions were modified to give a more representative shear distribution
and also the end section was constrained to remain plane. This effect
is also shown in Figs. 26 and 27.
•*._
^ - • '
/ N, XtV
/s
-10.000
•
S
t
1
-10.000
1
I
/
/
/
/
/
/
-.«0,000 /
1 '
, /
1 /
1
/
, /
1 /
1 /
II
HUE VALUE r .
"ENGINEERS B I N D T H t O d v ' [ » A S E S 0M L O A D S )
• l O S I . L A G E. C A 0 S S S E C T I O N CONSTRAINED
TO R E M A I N PLANE.
Fig. 27. Effects of boundary conditions on the end load in boom " B " .
At a later date, the analysis of the fuselage section forward of this

station was carried out and on completion of the interaction, the results
shown as the correct value on Figs. 26 and 27 were obtained.
This shows that with a little ingenuity and experiment a reasonable
set of end conditions can be specified, but any remaining errors lake
rather a long time to die out.
311
I. C . T A I G and R. I. KERR
3. C O M P A R I S O N OF FORCE AND DISPLACEMENT

ASSUMPTIONS
To investigate the effects of various assumptions made concerning the
idealization of the structure, a simple section of fuselage has been
analysed.
A typical frame of this fuselage was analysed by the displacement
method (using both coarse and fine grids) and by the force method
(using the fine grid and beam elements).
O DENOTES NODE UtED IN COAASE 6 RIO
Fig. 28. Idealization of a sample frame.
The two grids are shown in Fig. 28 and the differences in the
flexibilities obtained are demonstrated in Table 4 which gives the
deflexions due to a unit vertical load applied at node 1 and reacted
at node 9.
In view of differences in these results two loading cases were applied
to the frames. The first case consisted of a uniform floor loading (i.e.
312
Table 4. Vertical Deflexions from Sample Fuselage Frame
Displacement Displacement
Node Force Method Method Method
{Fine Mesh) {Coarse Mesh)
1 7-552 7-623 6-587

2 5-081 5-167 4-568
3 1-861 1-870 1-649
4 1-862 1-822 1-654
5 1-902 1-857 1-692
6 1-796 1 -800 1-629
7 1-776 1-738 1-817
8 0-824 0-824 0-800
9 0 0 0
11) 1-539 1-619 1-378
11 1-439 1-572 1-308
FRAME NEUTRAL AXIS
LOADING CASE I.
LOADING CASE I
Fig. 29. Bending moments round sample frame from force method analysis.
313
I. C. TAIG a n d R. I . KERR
between nodes 6 and 11), reacted by a shear distribution based on

engineers' bend theory. T h e second case was a uniform internal
pressure. The results obtained from the force and displacement (fine-
mesh) methods are shown in Figs. 29 and 30 respectively. It will be
seen that the differences between these curves are well within normal
stressing tolerances.
FRAME NEUTRA. AXIS
LOADING CASE I.
LOADING CASE I.
Fig. 30. Bending moments round sample frame from displacement method analysis.
Finally overall loading cases were applied to the fuselage, and the
difference in the final stress matrices caused by the use of the various
forms ofthe stillness matrix were less than 0*5 per cent.
314
SI£
(-)jodoj
jjuEApt* V'OV\) '209 " " i s Moda-g [punorj q.).n;.>s.)>j- p;.)inn;no.i.i\- -sj|ot[
S u u u i q S q )iioi|ii.v\ pin; qn.w sq.i.w ir.iqs )(i ss.mjjiis pin; ipiin.i.ns ,iq ] -,j *NII.I}J -7\
'L6Z >!'I lnx'-'H ]K.)tinBUo.i.>y *BpBUE;*ijo p j u n o ; ) qjJB.isoy |BUOIJE\J
•s,).in).)n,us SHIM qi.i-iqnui pins .lucls-iqnui |o .-i.iojq [ " i \ " I \ ' ' n w i a i z a a z H Q - \ \
SS6I A B K - q j J 'FS6I M Q - A O M
•Xiiujiuili'tni f o i m y -sisXpntE pMin.in.ijs put* sui.uo.iqi AS.mu-j -j-f -f -siMAoay - Q \
"6S6I '-iqiu-iKl-JS '(> ' " \ ' '9i5 'sttttmps jvjiinimimy nyi fo pnunop -sis-ipinE
ss.limiis .IOJ SOOI.1JBIU .n»cls JO UOIlEn|BAg -Q ">J '.EXIlf afaj\ puB • [ " y •itscnai'Aj •(*
' I l l i I x i ' ] \ 'rHiiiJA-uiSirj ifoAMiy -ptiEd .lEjqs [Ej,nE]i.ipBnb j q x [ ' s '•* i AMVQ -g
'9561 VA 'tZ£ '°N '82 ,A\my*wOua
ifonuiy -SSUIM OIIBJ is.idsB M O | JO siu.qqo.id . i i i s B p o j j y -y a T) 'NNVKOHJ •/
'£561 -M'-f 'L " ° \ '02 'mutijy ji/Jiliwurmy jqi pi
/iiiuiio/- -s8ii|A\ B i p p .IOJ siuoijqjjoj o;>uon[(ui piiE SISA[EUE |Ejnjjnj)^; i«; -AAaq -y
7.S6I •i.>qui.>,\o1V ' I I o x
'fcZ 'WaWW/y tvnjmruooy JI/I fit jmunof -,>.nuBAjnj j o nopiuapisaoa E piiB .I,X1EI
j o e p a g g — j i rn;,! :SIS.<|BUB jBjnianJis DUIBUI JO poqjjui .qduiis y JJ 'NiaT^ 'g
' L l ' O ' S MOd»->j p-sjiuirj uoijBiAy DU)33|g
qsqiiii'i -|IIII|I,IIII |u.)iu.).)E|(Isq3 M.III-III . >i 11 \i\ STSArBUB |f in 1 >n i is ' ) | 'OIVJ, '*
• ( l i t l | i.ii(ni.ml.>s | l | i , - , | S[BI.I.IIBJ.\ pin; s.xinimqg '(1>1\ ' ) \ " ' I'-'m M
-.),ul .i,)di;(| -SIS.<|BIIB pMnpiuis j o poqi.itu ss.m(|iis i.j.j.np o q j | ' j \ - 'waxM.i ] •-;
•9-.fi I i-)qin-)ld->S '6 • ° \
'CZ ' SJ tuDug \vninmiony JI/I fo ttnu.nop 's.i.ini in.iis jrajduioj jo SISAJEUE uoixayap
pin; ssiiqiiis '{' " I ' d d o x pin; '••) \ \ •NI.I.HVJAJ' '• \ \ -JJ ' H o n o i g *'f - j \ 'HaNHnj, 'Z
'S6~85 uodo-î [BOiuqjJX ' O ' Q ' V ' M ' • w i j a n f l S — I I -'"-d qo-'iuoo puE
AiqiqEis UO s j j j j p .)|1SB|JO.I.H; .iMios.i.idnv,' H >| '» r i n v n v ) pin; -J"*M:I;)NI.L,I v>* [
s:4 0 N ; i > r . i , K i > i
ssanimiH.i.s I . I V H O M I V J O MOI.I.V.I,\-.-ISMM.I;-IJI xaaaosia

D I G I T A L ANALYSIS OF NON-LINEAR
S T R U C T U R E S BY THE FORCE M E T H O D
PAUL H. DENKE
Douglas Aircraft Company, I n c . Long Beach, California
CONTENTS
1'iAgr
INTRODUCTION . . . . . . . . 318
IDEALIZATION . . . . . . . . 319
Lumped Parameter Idealizations. . 319
Non-Rectangular Gridworks . . . . . 321
Example of Non-Rectangular Gridwork . 323
Machine Selection of Redundants . 325
LARGE DEFLEXIONS . . . . . . . 326

Application ofthe Large Deflexion Kquations . 328
SMALL DEFLEXION STABILITY . . . . . 330
PLASTICITY . . . . . . . . . 332
C R E E P BUCKLING . . . . . . . . 335
NATURAL MODES AND FREQUENCIES . 335
CONCLUSIONS AND RECOMMENDATIONS . 336
APPENDIX A. T H E LARGE DEFLEXION EQUATIONS . . 338

B. U S E OF THE LARGE DEFLEXION EQUATIONS . 339
C. BUCKLING OF A PIN-ENDED COLUMN . . 340
1). PLASTIC ANALYSIS . . . . . . 342
Current linear and non-linear problems in Ihe field of digital structural analysis
are considered, and Ihe extension of Ihe Douglas Force Method to provide solutions
In Ihe non-linear problems is presented.
The idealization of structures utilizing rectangular and non-rectangular grid-
works is discus set/. Basic equations for the geometric large deflexion problem are
317
1-. II. DENKE
derived and some applications air presented. The small deflexion stability problem
is considered and the solution to a problem involving the buckling of a stiffened
plate is presented. A method of solution lo the structural plasticity problem, based
on Ihe I'l audi! Reuss plasticity equations, is described. The creep problem is
briefly considered, and some results relative lo the calculations of natural modes
and frequencies for loiv-aspecl-ralio wings are presented.
INTRODUCTION
A paper describing a digital method of Structural analysis, developed

at the Douglas Aircraft Company, Inc., was presented at the September
1959 meeting of the A G A R D Structures and Materials Panel 1 . An
outline of this approach, based upon the work of Langefors 2 , and
Wchlc and Lansing 3 , was presented initially at the Second U.S.
Congress ol Applied Mechanics in 19544.
The Douglas Force Method was developed to provide a means for
the analysis of complex aircraft structures, which previously were
beyond the reach of any rigorous approach; to take advantage ofthe
newly available capabilities of the large digital computer; and to
prepare for the task of designing the hypervelocity aircraft and orbital
vehicles that were to come. The events of the past three years have
justified the difficulty and expense of developing this and similar
methods, as well as the effort that was necessary to secure acceptance
of the new approach. During this period the computer program has
been completely automated so that human intervention is unnecessary
except in the event of an input or machine error; and applications in
production and preliminary design have become routine.
A number of features believed to be novel were introduced during
the development ofthe method. These features include the following:
(1) A complete formulation of the force approach in vector and
matric form was produced. This formulation included the equi-
librium equations, the continuity equations, and the transforma-
tions which facilitate passing from one set of equations to the
other. These equations were translated into a computer pro-
gram 1 - 4 . Curiously, the need for programming the equilibrium
equations was questioned at the time, although the solution of
these equations by hand is laborious.
(2) The forces which act upon an element were classified into
clement forces and element reactions. This distinction clarified
the formulation of the problem 4 .
(3) The notions "generalized force" and "generalized force com-
ponent" were introduced. This nomenclature, and a corres-
ponding generalized notation, eliminated the need for separate!)
considering forces, moments, translations, and rotations'.
318
DICilTAL ANALYSIS OF N O N - L I N E A R STRUCTURES
(4) A machine method for the automatic selection of redundants

was devised 1 .
Although considerable progress has been made in the digital analysis
of structures, much remains to be done.
In the field of linear structures, the automatic selection of optimized
redundants has eliminated the difficulty and loss of computing accuracy
that sometimes attended the selection of redundants for the force
method. The only remaining source of appreciable error is in the
structural idealization. More study should be devoted to this area,
especially for odd-shaped panels.
The computation of thermal stresses in linear structures is a routine
problem if the thermal gradients are known.
In the field of non-linear problems, further development is necessary.
This area includes the computation of stresses and deflexions in the
presence of plasticity, creep, large deflexions, and large deflexion
stability.
The following paragraphs describe progress that has been made in
the fields of idealization, stability, plasticity, and large deflexions. The
automatic selection of redundants, and the computation of natural
modes and frequencies also are discussed.
IDEALIZATION
This discussion considers discrete idealizations. It is understood that
a discrete idealization consists of a finite number of elements, joined at
a finite number of points. Idealizations of this kind seem to fall into
two categories: lumped parameter and finite element.
As an example of an especially useful kind of lumped parameter
idealization, a stiffened plate, subjected to edge loads in the plane of
the plate, is idealized as a gridwork of axially loaded bars and panels
subjected to shear. Panels are joined to bars at the midpoints of panel
edges. The ability ofthe sheet to carry tensile or compressive stress is
lumped into the bars, while the shear-carrying ability remains in the
panels.
In finite element idealizations, the ability of the sheet to carry
tensile or compressive stress remains in the finite sheet elements.
Lumped Parameter Idealizations
Lithcr of the idealizations can be used in connexion with the force
method. However, the lumped parameter idealization is very con-
venient and satisfactory, especially when the gridwork is rectangular.
This idealization is quite suitable, even in the case of an unslilfened
plate.
A rectangular plate containing a centrally located crack is shown in
the lower right-hand corner of Fig. 1. The plate is subjected to longi-
tudinal tensile stress. In the analysis, only one-fourth of (he plate needs
319
P . II. DENKE
L-°""-J
CRACK
Fig. 1. Idealized structure for crack analysis.
to be considered because of symmetry. The figure also shows the

lumped parameter idealization of one-fourth of the plate. It should
be noted that the lumping is finer in the region ofthe crack tip, where
the stress gradients arc steepest, than elsewhere.
The cracked plate was analyzed by the redundant force method.
Figure 2 shows the computed distribution of longitudinal tensile stress,
and the stress concentration at the crack tip. These results are taken
from a study on the notch strength of sheet materials, performed by
the Douglas Company under contract to the Aeronautical Systems
Division, USAF 5 . Figure 2a shows a comparison of the computed
longitudinal stresses along a transverse section through the crack with
values computed from the theoretical result of Wcstergaard 6 . The only
appreciable error is at the lip ofthe crack, where the theoretical result
320
DIGITAL ANALYSIS OF NON-LINEAR STRUCTURES
is infinite and the computed result is not; however, this discrepancy

has no practical significance.
The accuracy of the solution and the amount of detail shown in
Figs. 2 and 2a demonstrate that the force method is an effective
approach to elasticity problems. This approach, involving the use of
redundants, in a sense may be new and unique, since apparently no
Fig. 2. Distribution of longitudinal tensile stress.
direct analogue to it is known in the field of mathematical analysis;

whereas the difference equation approach has its analogue in the
differential equations of elasticity. The redundant force approach
appears to result in a simpler procedure for satisfying boundary con-
ditions and in better conditioned simultaneous equations than is
provided by the difference equation approach.
Non-Rectangular Cridworks
When the gridwork is not rectangular, the lumped parameter concept
remains valid and gives satisfactory results, but the procedures for
establishing the flexibility matrix are more complicated. Figure 3
shows a parallelogram panel with notation for the geometric parameters,
321
-
1
Z
z I
X
-
Ti
-
— •; o Z ". P 2 a 3 r". O b J I?
—i
o SN 3 IP- 3- PS- i-2. a-*" 13 . tq n n 3" < =
: z s." K* C o
n =r
n 2. = & ^ E T C " ^ 3 3 i c 5
r
"5_ "" H- —•
1
n
1
3"
— n P ? ! B STT3 S *••> •5 *-° 3 " ry n
r* n Si- 3
x 3, - I a .
I =.
- ; - •
FM n 1 H 3
0 ' i: W o rs rs -a* X _ — O O "C l\3 s.
n ~ n rji 5 3 C/3
12 COW 3
o. 3 •< [5 1*3 C>
o • 2 3- Sa Ot O
P 3 FT o — I—1 _ [-J i " . rj 3_ -^- n
n
1
3'
x
a, w H
r =• B 5 2 P " O -j CT 1 "S-l ft
5- -r-3 xr —-y s: o 8" 2 rs
1/1
ST -
O n 8 3 B a 3 n 2. a z•ji M
-
Sfi
X
a.
tf. O* M n *~>
S> JL S 8L S 3 O-o 3 5 9--S ^ E-

r^ ry C — •
s
- i - c p *o
3 M P c
3
P
z
(1
x -r
~ 3
3
jq
Ifl
3'
J
z\. -
3
— n 5 — P w — - c V. — —
X
5- 5 - X sr y
o 2 n is - - » 3 *
•r = ——
. o 3
r-r ° r^
-. P
M
X
3
z
r. - —C - ~p n3 13
-5 * — XX, £_
S. -
3 ,_, K"
r)
C
do
33
i ia
3 . 1MI 3* a T3 3
s r> s>r°-
n c - T3 ^ = = ^ 3__, j IC
z A---.' 3WJ (OJ . rs
0
sr•
5T S -• i X ;
z X rs O
3 5 ^ -. 3 o z ft »"
c, L —^ g-? 3-
rs Q.CK5
3 ' 3 ~z 5 = i 5 " - s 3 33 "'
J.
-J-
03 B
a. 5^ c -- rssr ^ —-
crq n —.
—
-x
jq r ~ 0 P
f-cn AX-XET. 2 -r^s rt -"-. H
n
- 3 "n* £L
3
- -
3 y
X - a. W ' o o 3 •< 3 3 jq z
x n
—n
< SJ CT)
5''
n
CD P
O X — N
—^ ^ 3- O •
3
-
3 —
- 323
Pis
—. ~ 3
Z-
s
z
o
y
c
Qa 3 5
r. W (7
8 ^ eg p P
S
3
-.
. p
a
c
3
W *
-J -
3* '
w
3
3
^
K .
3
K3 3 y.
x
3 P- §•5 5 O ^~~ . — — O H- 3
3 rt -—
Er — — n Z~. 3 3 fl P a.
- n - zl £^ V - CL ! ? r*
B sr
P
-
f. n /
C
-i
i
D- °
3
3 Of> C: x
â. ly
X
—t
— 5
" Z I T W T
o ° 5' -
s- i—
—
p
= 5-1" - 3 -
— B n Cl < r ~ -; o w &- Z~Z_ N" - c 5' ^ " - 3 3: 3" W
-: - o n y
n -^ p -. 5J -' 3 it 3 o
2). 1
K>
Crt
-« : 2. 3 and o 3
s. o 3 OA. — 3 P
C l " % r3^- E/J
1—
5' - - . Q, 3 3J 3 3"
Gr A- 3 3 -. " T ^ < CT5 J i '
p
P —c P o* - et S. K >
— r. rtr"S
A
Z C« ^ " 3
S" PN
3 • 3v „ 3 3'
X
p
a
-
I | —
rs ^ a
— V
IAA ^
""ciq
& 3
S 3 rsO 3l rs
^ 1 3 " rs
z 3 rs C/q 3 -
rs
H
5'
3
3 ^ ^ ^ P
•y. CL
3-
ra o '
3 S
rt
\z.
p>; s. c •S* s* a. s.
P ff. — - -
o pi 3 n —• 3
0- - n rt O . D- ' 3 Ei -Z - -X - — 3 £ C X p
— S O
•n 3 3' — fcj U ) rs p ^ I sT 3 r
—. x ' « ° " 3 -s
rs
3 - ST
rs rs
3
rs
3 3
E.S-
.'. H. OKNKE
t
..173*
_L_j_
079 *——
. . 34"
-
Fig. 4a. Wing structure idealized.
324
.-Eai,ecTiou 1
OE.LeCTlON 1
3
0
y
>
—
>
X.
-
T.
-. P 3
_ D- P
p < c
-1 S3 O
39 3 3
P . I I . DENKE
ANALYSIS (CASES 1 AND 3

ARE COINCIDENT!
30
X-INCHES
l.) DEFLECTIONS. CASES 1 AND 3
REAR SPAR
ANALYSIS (CASE ll
A STARBOARD \
• PORT J
FRONT SPAR
(d! FRONT S REAR SPAR CAP STRESS CASE
Fig. (i. Deflexions and stress, cases I and 3.
systems of equations. On the other hand, the need for selecting

redundants is a disadvantage because skill is required on the part of
the analyst. The development of a machine method for selecting
optimized redundants has removed the disadvantage and retained the
advantage. The operation of a program to select redundants, called
the "Structure Cutter" was described in Ref. 1.
LARGE DEFLEXIONS
A useful approach to the large deflexion problem is to introduce
corrective terms into the small deflexion equations to eliminate the
326
OlGlTAI. ANALYSIS OF NON-I.INKAR STRUCTURES
errors resulting from the small deflexion assumptions. This approach

permits the use of previously developed small deflexion methods. The
principle is illustrated by the simplified pin-ended column, consisting
of two rigid bars joined by a spring, shown schematically in Fig. 7.
When the load on the undeflected column (a) in Fig. 7 passes the
critical value, the column buckles as in (b). T h e forces and moments
which act within the column are shown in (c). The bending moment
(b) Id I
Fig. 7. Large deflexion of a column.
results from the deflexion of the column and is not predicted by the
theory of small deflexions. According to small deflexion theory, the
deflexions are supposed to have a negligible effect on the geometry of
the column, so that the two bars remain colincar, as shown in (d).
Although the bars in this figure have been rotated through the angle 0
to the colinear position, the true forces acting on the bars are shown.
In (e) the forces on the bars have been resolved into horizontal and
vertical components, and the bars have been shortened to give the
same column length as in (b). Also shown are the forces acting on the
joint between the bars.
The small deflexion theory predicts only the forces labeled " P " and
takes no account of the shortening of the bars. Two corrections
therefore are required as follows: (1) a fictitious load, equal in magni-
tude to 2 P sin 0 must be applied to the joint, (2) a fictitious deforma-
tion, equal in magnitude to — a{\ — cos 0) must be applied to each
bar. These fictitious loads and deformations will cause the small
deflexion theory to give the correct results. Note that the fictitious
327
('. II. IIKNKI,
loads and delbrmations are non-linear functions of the deflexions, while

the fictitious loads arc linear functions of the element forces.
This approach, which appears to be general, has been expressed
in matric form. 9 The element forces and deflexions of a structure
accounting for large deflexions are given by
F = F M + <ftn) + F f e + O (3)
A = A M + qtn) + Afe + O (4)
where F is a matrix of clement forces, A is a matrix of deflexions, F$ is
an influence matrix of element forces resulting from unit values of the
external loads, F e is an influence matrix of element forces resulting
from unit values of element deformations, \ is an influence matrix
of deflexions resulting from unit values of the external loads, A,, is an
influence matrix of deflexions resulting from unit values of element
deformations, <> / is a matrix of external loads, 0„ is a matrix of fictitious
external loads, e is a matrix of element deformations, and en is a matrix
of fictitious element deformations.
T h e influence matrices F^, F,., \ , and A,, can be computed by linear
force or displacement methods, <f> and e are known, and </>, and e„ are
functions of the deflexions. Because of the dependence of <£„ and e„
on A, Eqs. (3) and (4), in general, are non-linear and must be solved
by repetitive means.
T o reduce the magnitude of the matrices to be manipulated in the
repetitive solution, the clement forces, which appear explicitly in (3)
and implicitly in (4) through <f>„, can be eliminated from the equations.
This elimination is accomplished in Appendix A with the following
result:
[/ - A,(<^ A -(- t n J ] N - ( A ^ + A,,,) + (A, + -V>„>„ (5)
where <>
/ is a matrix of fictitious loads resultine from unit values of
the deflexions, produced by the linear element forces resulting from
the known external loads; and </>„ . is a matrix of fictitious loads
T
"e\
resulting from unit values of the deflexions, produced by the linear
element forces resulting from the known element deformations. Equa-
tion (5) involves a number of approximations which are explained in
Appendix A. The deflexions A appear explicitly on the left of Eq. (5),
and implicitly on the right, in </>„,, and en. The equation therefore is,
in general, non-linear.
An elaboration of Eq. (5), suitable for direct matrix programming,
has been developed by Warren 1 0 .
Application of the Large Deflexion Equations
As a simple example of the above equations, consider the structure
shown in Fig. 8, which shows a highly idealized representation o f t h e
328
Ole.lTAI, ANALYSIS OK N O N - I . I N E A R ST R l ! C T f R ES
low aspect ratio wing. In the idealization the torsional stillness of ihe
wing is represented by the lube EF, considered to be inextcnsional.
The members representing the leading and trailing edges, AB and CD,
carry tension only. The bar IC, is considered rigid. The leading and
trailing edges are assumed to be heated to temperatures relative to
(.0 LOW-ASPECT RATIO
III) IDEALIZATION
Fig. fi. Simplified idealization of a low-aspect ratio wing.
E F such that the unconstrained elongations of AB and CD are equal

to e,.
The matrices appearing in Eq. (5) were evaluated as in Appendix B
and substituted into the equation to yield the following expression
involving the angle of twist:
L Le, G.I D \ U* =
.,
\ a- EA a* ' EAa*
This result also can be obtained by elementary analysis.
Figure 9 is a graph of angle of twist as a function of torque for the
heated {e, =f 0) and unhealed [e, = 0) cases. For the cold case the
slope ofthe curve shows that the torsional stillness increases with twist
angle, which is the expected non-linear effect. For the hot case the
slope is initially negative, indicating that the zero deflexion position
is unstable. After the deflexion becomes sufficiently large, the stiffness
329
I>. H. DENKE
ARGE DEFLECTION
HEATr L
*— SMALL
A L L OEFLE
DEFLECTION
\ UNHEATE
Fig. \). Torsional response of the idealized wing of Fig. H.
becomes positive and thereafter increases. The shape of the curve

indicates the existence of a "snap-through"' phenomenon for the
structure.
This example demonstrates the ability of Eq. (5) to account for non-
linear effects.
SMALL D E F L E X I O N STABILITY
For small deflexions the matrix e„ is zero. In this case Eq. (5), when
solved for the deflexions, gives
A-[/-A,.# -f-^Jl -(A^-f- \e) .7)
Equation (7) involves an eigenvalue problem. When the determinant

of the matrix which is inverted is zero, the deflexions tend toward
330
DIGITAL ANALYSIS OE NON-LINEAR STRUCTURES
infinity. Thus the small deflexion buckling conditions and modes are
defined by the equation
[ / - A , ( ^ + <*nJ]A = 0 (8)
T h e condition
(/ - A J ) A = 0 (9)
governs the load buckling of the structure, while (he condition
( / - V J A = 0 (10)
governs the thermal buckling, since the "known clement deformations"
mentioned previously in the definition of </>„ can be clement defor-
mations resulting from thermal gradients. Appendix C contains an
example of the use of Eq. (9) in computing the buckling of a pin-ended
column.
Figure 10 shows an idealized skin panel, reinforced longitudinally
by corrugations and transversely at intervals by stiffencrs. The
ENDS OF STIFFENERS
AND PANEL EDGES
ARE SIMPLY SUPPORTED
F'ig. 11). Idealized stiffened panel.

331
P . H . DENKE
PANEL IS SYMMETRICAL
ABOUT LONGITUDINAL AND
TRANSVERSE CENTERLINES
TRANSVERSE STIFFENER
Fig. 11. Buckling of a stiffened panel.
buckling of the panel under longitudinal compression was computed

according to Eq. (9). Figure II shows the computed buckling mode.
This example shows the generality of the digital approach to small
deflexion stability problems.
PLASTICITY
An approach to plasticity problems has been developed on the basis
of the force method. When the structure contains plate elements, or
when plates or solid bodies arc analyzed, the approach incorporates
the Prandtl Reuss equations of plasticity. The resulting non-linear
equations are solved by the Newton-Raphson method.
The force approach is well adapted to the solution of plasticity
problems because the non-linear equations, which must be solved
iterativcly, are equal in number only to the number of redundants.
For statically determinate structures no iterations are necessary, since
the stress distribution is governed solely by the equations of statics.
A complete description of the approach is beyond the scope of this
paper, however the details are presented in Ref. 11, which is a report
on compression fatigue prepared under the sponsorship of the Aero-
nautical Systems Division, AFSC, USAF. Appendix D contains a
general outline of the method. Some of the basic principles arc pre-
sented in the following discussion.
Plastic and clastic problems are basically different, because in plastic
problems no relationship exists between stress and strain except under
special circumstances. Specifying the state of stress in problems
involving plasticity does not specify the state of strain, which also
depends upon the previous loading history of the material. Relation-
ships can be specified only between increments of stress and increments
332
DIGITAL ANALYSIS OF N O N - L I N E A R STRUCTURES
of strain, and the relationships for increasing stress (positive increments),

are generally different from the relationships for decreasing stress
(negative increments). The Prandtl Reuss equations, therefore, relate
increments of stress to increments of strain. These equations, for the
case of plane stress, are as follows:12
3<T'A<T1
Ae? =
2aIV
'hcr'Aa
-.1-
AI >: ( • • :
25H'
3Â5
AyL = 5H'
where Af f., Af* and Ay'fy arc increments ofthe plastic strain components,
IT is the equivalent stress, A<7 is the increment of equivalent stress, / / '
is the slope of a curve of equivalent stress vs. equivalent strain, and
a'x = {2a„ - <T.)/3 a', - [2a. - a x )/3 (12)
The a m ay and r „ are the stress components, while the equivalent stress
is defined by
a = V a \ + a\ — a.a2
where a, and <r2 are the principal stresses. Except for a scalar constant,
a is the quantity known as the octahedral stress. As a reasonable
approximation H ' can be taken as the slope of a stress-strain curve of
the material. Equations (11) apply when the equivalent stress is
positive and increasing or negative and decreasing. Otherwise,
Ac* = Af I —- Aylv = 0. The Prandtl-Rcuss equations have been
integrated into the method.
A second important ingredient is the Newton-Raphson iterative
procedure for solving the non-linear simultaneous equations. The
question of convergence is of decisive importance in the solution of
non-linear problems, particularly when the amount of computation per
iteration is large, as in the present case. A method that converges too
slowly, or does not converge, has no applications. For this reason the
Newton-Raphson method was chosen, since it shows strong uniform
convergence for plasticity problems in cases where the simple iterative
approach is divergent.
Figure 12 shows one ofthe results produced by the plastic analysis 5 .
The figure shows the residual stress distribution in the cracked plate
of Fig. 1, resulting from the application and removal of longitudinal
tensile stress sufficient to cause yielding of the material. The analysis
is based on a realistic stress-strain curve for the material. Figure 13
shows the convergence of the stress in a fiber near the tip of the crack
as a function of iteration number. Note the uniformity of convergence.
333
P. H. DENKE
RESIDUAL STRESS RESULTING FROM

15.000 PSI GROSS APPLIED STRESS
-v
V
COMPRESSION
I INCH - 10,000 PSI
Fig. 12, Residual stress distribution in the cracked fdate.
"CRACMNO = _S KM
• • M S
PQ
ITERATION
Fig. 13, Convergence of the iterative process.
334
D I G I T A L ANALYSIS OF N O N - L I N E A R STRUCTURES
CREEP BUCKLING
An investigation of creep buckling of panels has been undertaken
under USAF contract 1 ". This project includes the testing of coupon
creep specimens, the testing of built-up stiffened sheet panels and
sandwich panels, and digital analysis of the results. The problem
involves both creep and large deflexions.
Since creep is a time dependent process, the natural approach is to
make stepwise calculations, utilizing small increments of time. If
necessary, an iterative cycle can be introduced within the time incre-
ment to minimize the error resulting from the finite length of the
increment 13 . In the creep buckling program a stepwise approach has
been followed, with iterative cycles within each increment to solve the
large deflexion problem that results from the incipient buckling of the
panel. The process involves the repeated use of Eq. (5). At the time
of writing, this work was still in progress.
The two greatest difficulties associated with the computation of
creep effects in structures are the performance of coupon creep tests,
especially in compression, to the degree of accuracy required by the
analysis; and the establishment of a valid law for cumulative creep
under conditions of variable stress and temperature.
NATURAL MODES A M ) F R E Q U E N C I E S
Figure 14 shows a low aspect ratio wing, 11 which was idealized and
analyzed by the force method. A small deflexion influence matrix was
Fig. 14. Idealization of a low aspect ratio wing.

335
P. H. DENKE
Computed and used as a basis for determination of the natural modes

and frequencies. Figure 15 shows a comparison o f t h e results, based
on linear theory, with test data.
If deflexions become sufficiently large, and especially if thermal
gradients are present, the relationship between load and displacement
TEST
ANALYSIS (I' .3)
5TH MODE
FREQUENCY. CPS
MODE NO. 1 2 3 4 5
TEST 4S.3 as.a 122 B Ifi4.2 179.7
ANALYSIS 46. 1 86.7 123.0 152.7 133.'
DEVIATION
% OF TEST S -2
• -7 2
Fig. 15. Symmetric, free-free, vibration modes and natural frequencies,
can become non-linear, as shown in Fig. 9. In this event, Eq. (5) can
serve as a basis of calculation to account for the non-linear effects on
structural vibrations.
C O N C L U S I O N S AND R E C O M M E N D A T I O N S
Ihe digital force method is a useful tool for the elastic analysis of plates
and other continuous bodies. The method has been extended to cover
the analysis of stability problems and problems involving large deflex-
ions. Solutions have been obtained for plastic plate problems on the
basis of the Prandtl-Reuss equations and realistic stress-strain curves
for the material. These methods should be developed further to reduce
the required machine time and to simplify the applications.
336
D I G I T A L ANALYSIS OF NON-I.I N E A K. STRUCTURES
ACKNOWLEDGEMENT
This work was accomplished under the direction of M. Stone, Chief,
Aerostructural Mechanics Section, Aircraft Division, Douglas Aircraft
Company, Inc. The examples on small deflexion buckling, plasticity,
and vibrational modes and frequencies were worked out under the
direction of G. R. Eide, J . O'Kelley, and R. C. Gloria. The work was
carried out under the sponsorship ofthe Aeronautical Systems Division,
AFSC, USAF, and the Douglas Aircraft Company independent
research and development program.
REFERENCES
1. DENKE, P. H . A general digital computer analysis of statically indeterminate
structures. Douglas Aircraft C o m p a n y Engineering Paper No. 834, September
1959.
2. LANGEFORS, B. Analysis of elastic structures by matrix transformation with
special regard to semimonocoque structures. J . Aeron. Sci., 19, No. 7, p p . 4 5 1 - 8 ,
J u l y 1952.
3. W E H L E , L. B. and LANSING, W. A method for reducing the analysis of complex
r e d u n d a n t structures to a routine procedure. J . Aeron. Sci., 19, No. 10, p p .
677-84, October 1952.
4. D E N K E , P. H . A matric method of structural analysis. Proceedings of the Second
U.S. Congress of .ipplied Mechanics. 14-18 J u n e 1954.
5. CHRISTENSEN, R . H . a n d DENKE, P. H . Crack strength a n d crack propagation
characteristics of high strength metals. Aeronautical Systems Division Technical
R e p o r t 61-207, May 1961.
6. WESTERGAARD, H . M . Bearing pressures a n d cracks. J . Appl. .Mech., p p . A 4 9 -
A 5 3 , J u n e 1939.
7. W A R R E N , D . S., CASTLE. R . A. a n d G L O R I A , R . G. A n evaluation of the state-
of-the-art ofthermo-mechanical analysis of structures. Wright Air Development
Division Technical Report 61-152, M a y 1961.
8. v. G R O L , H . J . , HAKKELING, B. a n d SCHUBRMAN, J . A. Results of stress a n d
deflexion measurements performed on a swcpl-back box beam a n d their
comparison with theoretical results. National Aeronautical Research Institute,
Amsterdam, N L L - T N S.536. O r d e r N u m b e r 101.354, 1959.
9. DOUGLAS AIRCRAFT COMPANY. I N C . Nonlinear a n d thermal effects on elastic
vibrations, theoretical a n d experimental. Douglas Report N u m b e r SM-30426.
Wright Air Development Division Purchase Request N u m b e r 89407, October
1960.
HI. W A R R E N , D. S. Buckling of structural panels in creep environments. A study in
progress per Contract N u m b e r A F 33(616)-8359.
11. DOUGLAS AIRCRAFT COMPANY. I N C . Effect of compressive loads on structural
fatigue at elevated temperature. Aeronautical Systems Division Contract
N u m b e r A F 33(616)-8103, April 1962.
12. H I L L , R . The Mathematical Theory of Plasticity, Oxford at t h e Clarendon Press,
1950.
13. A L E S C H , C . W., R I P A R B E I . M . ('... S T E I R E R , W. H . , a n d R O B E , W. Prediction of
creep effects in aircraft structures. Wright Air Development Division Technical
Report 60-41 l ^ P a r t I, August 1961.
14. L E V Y , SAML-EL. Structural analysis a n d influence coefficients for delta wings.
J. .\eron. Sin., July 1953.
337
1'. II. DENKE
APPENDIX A
The Large Deflexion Equations
THE fictitious loads are linear functions ofthe element forces, therefore
<A„ = BF (A.1)
where B is a matrix of fictitious loads resulting from unit values ofthe
element forces, and from the deflexions. Eliminate <f>M and F from Eqs.
(3) and (4).
.'. F = {I - FJi)-i[F,cP + Ffe + ef] (A.2)
Therefore from (A.1),
<f,n = B[I - F , B ) ^ [ F J + Ffe + O ] (A.3)
eliminate <£„ from (4) and (A.3).
.'.
A = AJ + A,B{I - F,B)-i{FJ + Fe{e + en)] + Afe + e„)
(A.4)
Expand the matrix (/ — F,B)~ l as a power series,
.'. A = A ^ + A,B[I + F,B + [ F J ) * + • • •]
X [FJ, + Fe{e + O ] + \ { e + O (A.5)
Rearrange the terms in Eq. (A.5), making the substitutions^,^ = BIf
and <pnc = Blf
.'. A = A,(7 + </,„,+ 4,1, + . . .)</>
+ [A, + A,{I + </>„* + < & + ' • - M J O + O (A.6)
In this equation <pn4i is a matrix of fictitious loads resulting from the
linear element forces resulting from unit external loads, and </>ni. is a
matrix of fictitious loads resulting from the linear element forces
resulting from unit element deformations. As an approximation, delete
all powers o[tf,ni higher than the first, and all terms involving products
of the form 4";ulf/,ne.
.'. A = AM + <f>J<f> + (A„ + V J 0 + O (A.7)
The matrices <f>M and <f>tl.e involve the deflexions A. As an approxima-
tion assume that <f>nJ> and c/>lire arc linear in A, and write <f>nj> = <f>n A
and „,.e ^",\-^' Make these substitutions in Eq. (A.7), and collect
terms involving A.
.-. [/ • AM„, A + * « J ] A = ( V f A,.e) + (A. + AJ n r )e n (A.8)
The matrices <>
/ and </>„ are defined in the body of the paper.
338
D I G I T A L ANALYSIS OF N O N - L I N E A R STRUCTURES
APPENDIX B
Use of the Large Deflexion Equations
REFER to Fig. 8b. Consider a single external load applied to the

structure: the torque <j>. Consider a single deflexion: the twist A.
Consider three element forces and the corresponding clement deforma-
tions: tension in AB, tension in CD, and torque in EF. Linear small
deflexion analysis gives the following results:
A. = A (B.9)
A. = [0 0 1| (B.10)
From elementary calculation, the fictitious torque resulting from the

linear element forces resulting from unit element deformations arc
4>ne = EA
/ p E A If
O] (B.11)
The fictitious element deformations resulting from the twist A are
(HP 2)
The fictitious torque resulting from a unit value of the deflexion A,

produced by the linear clement forces resulting from the torque <f> is
K* = ° (B.l 3)
The fictitious torque resulting from a unit value of the deflexion A,
produced by the linear element forces resulting from the simultaneous
element deformations e, in AB and CD is
K± = *EAfy> (B-14)
The matrix of element deformations is
e=\el\ [B.15)
P. H. DENKE
Substituting the above matrices into Eq. (5) and rearranging the terms
gives
Lc, C J ff\ A L34>
A--
K EAa*
k
) -m<- (B.16)
APPENDIX C
Buckling of a Pin-Ended Column
FIGURE C.l shows a pin-ended column. Because of symmetry, the

problem of computing the buckling load for this column can be
T
l_ 2-a
1 7777
T K \
II
v •< \
7777
_L Ar
-i
Fig. C. I. Buckling analysis of a pin-ended column.

340
DIGITAL ANALYSIS OF N O N - L I N E A R STRUCTURES
replaced by the problem represented in Fig. C . l b . T h e equation

governing the buckling is given by Eq. (9) as:
(/-.VwJA-0 (C.l)
The matrix A^ is a deflexion influence matrix: a matrix of deflexions

resulting from unit loads. Figure C.lc shows the points where unit
loads arc applied and deflexions are measured. The influence matrix
can be determined either by elementary or matrix methods to be
"1 5"
aa 3 48
(C.2)
\EI 5 1
48 24
The matrix <f>„ is a matrix of fictitious loads resulting from unit
deflexions. T h e first column of this matrix is formed by giving the
first deflexion a unit value and observing what fictitious loads are
produced at points 1 and 2, as shown in Fig. C.Id. Similarly the
second column of-/** is obtained from Fig. C . l c .
2
ill
"*A \_.
(C.3)
a 4
Substituting A, and </>"./.A
„ into Eq. (C.l) and rearranging gives:
11 I
24 4
A = 0 (C.4)
8A +
24
where A = EI/{Pf,). T h e buckling condition is
11
24
ll (C.S)
1 . 1
8 ; " + 24
from which X = (5 ± A / 1 8 ) / 2 4 . Therefore P = 2.61 EI/a 2 or 31.7EI/a 2

The lowest eigenvalue substituting a = L/2, is
EI
P = 10.44 (C6)
/-
The error is 10.44/rr2 — 1 = 6.5 per cent.
341
P. II. DENKE
APPENDIX D
Plastic Analysis
IN the force method, the continuity condition is satisfied by the

equations
AZ = 0 (D.l)
where A Z is a column matrix of relative deflexions at the cuts intro-
duced to produce a statically determinate structure. Here the view is
taken that the redundanls arc forces at actual physical cuts, for ease
of visualization, although this interpretation is not the most general.
The more limited interpretation docs not impair the generality of the
resulting equations.
T h e deflexions at the cuts are given by
A Z = d ^ A X + dÂob + PxAe" (D.2)
where, according to the notation of Ref. 1, i)xx is a square matrix of
deflexions at the cuts resulting from unit values of the redundants,
AA' is a column matrix of redundants, o^ is a rectangular matrix of
deflexions at the cuts resulting from unit values of the external loads,
At/, is a column matrix of external loads, fx is a rectangular matrix of
clement forces resulting from unit values of the redundants, and Ac1'
is a column matrix of plastic element deformations.
The unknowns in Eq. (D.l) are the redundants, A'. According to
the Newton Raphson iterative method for simultaneous non-linear
equations, the j + 1st approximation to the redundants for the Jth
load increment is given in terms of the j t h approximation by
'' AX t i J . „ + AX (iJ) - A ^ A Z , ^ (D.3)
where the first subscript indicates load increment number, and the
second subscript indicates approximation number. Also
Alt,n = AZ(M)0 (D.4)
and
8 = | djdXfi | (D.5)
Note that d is a row operator matrix such that the element in the ath
row and flth column of AZc is c'A7.fi AX..
The matrix A can be shown to be given by
where
'•'•',,,„ ' f M o f . (D*7)
1
and D ' is a square matrix such that the clement in the ath row and
/ith column is given by
D'vS - 8e$ldFa. (D.8)
p
The matrix D is a plastic flexibility matrix analogous to the elastic
flexibility matrix D.
342
Et£
'£££ 'Vi H 'n>ivv:i>riis,i\\ •Sl£ •H'*-'"N

'992 '£02 "D >I "ri*i-i.V\ Y92 "1 >I " m m
r •081 'Wl 'I S 'A3sia->[
YEE 'sit: a i '* "".w
'081 "0 '>n'i-'M I rH'mv)]
YEE '82E s (i 'N-IMHV.-A
'992 ".1 » 'HWOf
'992 'E02 '061 '081 '691 I TV 'namuix '992'122 "a "M 'NasNHl'
•081 •;•! • / i . i . i i M . i .
'992'WI 'f""l'<wox -
992 '122 fN'-n-'H
'992 '612 '191 "091 S '"NSJMIIS.UVI |
YE£ H ' i i ' H
£!£ V H O 'NNVKOIIJ, 'Lot 3 'OMITHMVH
SI£ 7 9 2 "0 1 'oivx
'081 TfnoMAS 'S.€ '182 Vi "I V *i*siaizaaz>io

Ltt "H 'M • " r a u j 'Lot f H ' i o » 0 "A
'992 - V T •«•!•'is 992 " 122 O 'Hxiaamo
'Lot V f'NVwuaiiHOS '992 "3 a ' • " • • o
881 O 'MaaNVs '992'612 N f '».-"(ioo«)
i££ O y ' v m o i o
i££ M 'aa°>I 992 Vi f '•"•••o
i££ O 'niHauvdi-H •'JIE'082'K2 T'S'*«"0
Sl£ '£82 'LLZ '1-91 I 'aaoNijj.v*a "50E
'£82 7 i 2 '081 'Wl 7 S H H ' a a H o v n v o
081 M •MU'VM.I
881 "081 '591 a 'a*aa.i;- ; \ aa stiavaj
992 '6£2 'LZZ H 'NOXOM
-
992'Ii-2'6£2 7 2 2 S 'MWMOog
'9t* 'St Mi- 3 "H a f ' I J T I I M
SI£
'182 '082 'Wl 'iS '99. ' l O « 'J-ii«« >I\i 992 '122 T ' o w w o a
S1E '182 992 H ' 3 *T"a
'082 '992 'Wl 7 5 '95 ' t T N "MOTMM 'Ltt ' i l E "992 '122 '081 'Wl H d 'a»«a
'992 '£02 'Wl O H 'NILSVJM
'91Z '99Z'*9\ f t i •"•' " " : )
Y££ H >I 'N3SN3XSIcIH--)
•i£t: 'SIS "S '***n
YEE '81£ A\ 'OMBUVl i££ V •& ' z t u - v o
7EB'8IE tl 'SMOIIIIW-I
•?1E'60E Pot:
l '<I0E '061 '081 '691 'Wl 'I H 'f'snuiowv
'SIE'W2'992'122 .1 *<> '»
'Wl "I 'S -I*<SNVII:)(IN( iEE 'M D 'HOfHTV
xacmi >ioHiav
ACARDograph 62
Advances in Materials Research
in the NATO Nations
Editorial Committee: H. Brooks, U.S.A.: N. H. Mason, U.K.: N. E. Promisel.
U.S.A.: G. H. Cooper, AGARD
In 1959, Dr. Fred Seitz, then Science Adviser to the Secretary-General of

NATO, became convinced thai the demands of modern technology for new and
improved materials were so pressing that a NATO level study of the whole
field was warranted. He visualized that such a study should embrace a review
of basic research in national organizations, should clarify the relationship
between basic and applied research activities, and should explore the best
methods for the administration of such programmes. AGARD, through its
Structures and Materials Panel, arranged the Symposium on Materials
Sciences and this book records the papers presented by experts
of international repute. They cover basic research on materials, description
of administration of research programmes, and study ofthe complexity of rela-
tionship between basic and applied research as they exist in the NATO nations.
The Symposium consisted of three parts: the first concerned fundamental re-
search; the second dealt with the role of basic research in development; and
Ihe third contained the organization of materials research. The book is a valuable
aid to research workers and those responsible for ihe planning and administra-
tion of effective material research programmes.
£5 net
International Series of Monographs in Aeronautics and Astronautics, Symposia

Division, Volume 12
Fatigue of Aircraft Structures
E d i t e d h v W . B a r r o i s , Service Technique tie I'Ai'rnnnuliiiue. I'uris
K. I.. R i p l e y , Royal Aircraft Establishment. Fnrnbnroui/h
This book contains the full text of all papers presented at the Symposium on
Fatigue of Aircraft Structures, held in Paris. It was organized by the "Direction
Technique et Industrielle de I'Aeronauliqtie", and the French Centre of the
International Committee on Aeronautical Fatigue (ICAF). with the support
and assistance of the other National Centres of ICAF and the North Atlantic
Treaty Organization (AGARD).
The first three papers deal with new basic phenomena, and are followed bv
three papers devoted to the statistical aspects of fatigue problems. . ^

Agard Ag 72

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

Agard Ag 72

Uploaded by

Copyright:

Available Formats

S

R. H . G a l l a g h e r , Bell Aeroivslems Company, Buffalo.

This book presents the result of a survey

Published for and on behalf of

Distributed in the Western Hemisphere by

Library of Congress Card No. 63-21973

Set in Monotype liaskerville 11 on 12 /'(. and printed in Great Britain by

Matrix Methods of Structural Analysis—A Precis of Recent

Some Problems in the Discrete Element Representation of Aircraft

Digital Analysis of Nonlinear Structures by the Force Method . 317

Author Index . . . . . . . . . 343

A PRECIS OF RECENT DEVELOPMENTS

J. H. ARGYRisf, S. K E L S E Y J , and H. KAMEI.§

A. FORCE METHOD . . . . . 1(1

f Professor and Head.

e. Strains in Spars . . . . . . .142

THIS paper summarizes some of the progress achieved in developing

A. Matrix Force Method

long, it will suffice here if we restrict our account to the highlights of

subprogrammes for the determination of the eigenvectors of a sym-

the application of the matrix displacement method to wings and tail

B. Matrix Displacement Method

wing or any irregular membrane structure analysed by the displace-

of interest the setting-up of the K matrix of dimensions 500 X 500

C. Matrix Force Method in Non-Linear Problems

where H and f a r e functions o f S and, in the case of plastic straining,

of solution must be iterative and admits of various degrees of approxi-

D. Non-Linear Matrix Displacement Method

A. a. General Theory of the Force Method for a Complex Structure

form the complete structure or system are called substructures or

"staircase" diagram of Table A.1, which shows the various stress

Complete system B X + Y + W closed

The symbols R, W, Y, X denote external loads, tertiary, primary

and f is the flexibility matrix of the unassembled structural elements.

militate against its adoption even for a medium-sized aircraft and

it. The inflation term derives from a congruent transformation on

A.b. Primary Redundancies in the Fuselage

[b0i b„],. == {a,'.. A|}-- (A.7)

where b 0 „ b 0 o provide the statically equivalent systems and b 1 ( , b j ,

Fig. A . 1 . Single-cell cross-seclion. Fig. A.2. Two-cell cross-section.

in Eqs. (A.7), (A.8), is to express the equilibrium conditions on the

This last relation is used as a computational check on the indepen-

where SI, is the t X (/ — 3) trigonometric matrix whose columns are

obtained by the above process (including the bending moments in the

CPT: 6 7 a 9 » 11 '2 '3 "

005 o -oos oi oos o -oos -01 01 OOS 0 DOS -»1

Fig. A.3. Self-equilihrnling flange load systems in wing cross-seclion.

"» 'VJ \ r "

7th Antisymmetric system (a) 7th Antisymmetric system (b) —

-04J-"0 Slh.Antisymmetric system (a 9th Antisymmetric system (b) -ooos.

very much affected by an orthogonalization process which uses only

»i. = {a-„ •»«,}. (A. 14)

is obvious and needs no comment.

ft, = [c« C j . . . c r . . . c. +1 ] (A. 15)

in which a s + 1 is the {s f- I) x s difference matrix for an open chain.

So far our equations allow only for t 4- s — 3 redundancies. To

fcu...+i == {- <.+!.- A3 ' IO I<+_J (A.21)

A.c. Major Cut-Outs in the Regularized Structure

element flexibility, is, in comparison, straightforward and is not

la) Ragutan/ad Structure (c) Minimal t u t ayatam - b o u n d a r y (a) M a . i m a l c u l ayatam - m l a r n a l

Fig. A.5. Elimination techniques in large cut-out.

The illustration (a) shows a highly redundant, plane, pin-jointed

Returning to a correct arrangement (c) we note that the effective

SISA'IV.NJV 'ivtin.i:)nH.i.s .-o s a o n x a w XIHXVK