J.H.

Argyris

Energy Theorems and

Structural Analysis
EN ER GY THEOREMS AN D
STR UC TU RA L ANALYSIS
A Generalised Discourse with Applications on
Energy Principles of Structural Analysis
Including the Effects of Temperature and
Non-Linear Stress-Strain Relations.

by
J. H. ARGYR IS, D.Sc. (Eng)
Professor of Aeronautical Structures,
University of London,
Imperial College of Science and TechnoiOKY

Co-auth or of Part II
S. KELSEY, B.Sc. (Eng)
Lecturer in Aeronautical Structures,
Imperial College of Science and Technology

Springer Science+ Business Media, LLC

 First published by
Butterworth & Co. (Publishers) Ltd.

Originally published in a series of articles in

AIRCRAFT ENGINEERING
Oct., Nov., 1954; Feb., March, April, May, 1955.

ISBN 978-1-4899-5852-5 ISBN 978-1-4899-5850-1 (eBook)

DOI 10.1007/978-1-4899-5850-1

J. H. Argyris 1960.
Originally published by Plenum Press in 1960.
Softcover reprint of the hardcover 1st edition 1960
 PREFACE
T HE present work was originally published as a series of articles in
Aircraft Engineering between October 1954 and May 1955. The
purpose of these papers was two-fold. Firstly to generalize and extend
but at the same time abo to unify the fundamental energy principles of
analysis of elastic structures. Although much of the corresponding
theory has been available for a number of years, to the best of the author's
knowledge it has not been given before in such generality. As an example,
whilst keeping within the small deflection theory the argumeats have been
developed ab initio to include non-linear elasticity and arbitrary initial
strains e.g. thermal strains. The first assumption introduces naturally
the twin concepts of work and complementary work first put forward
by Engesser. The author has attempted in this connexion to refer to all
relevant and historically important papers. Since the appearance of the
present articles, a few papers have been published which touch upon the
same subject but suffer, unfortunately, from a rather incomplete list of
references.
Secondly, the writer developed in considerable detail practical methods
of analysis of complex structures-in particular for aeronautical engineer-
ing applications. The most important contributions are the matrix methods
of analysis. Since they are only cursorily referred to in the Introduction,
it may be appropriate here to describe their use and origin in greater
detail. The matrix formulation besides providing an elegant and concise
expression of the theory of such structures, is ideally suited for modern
automatic computation because of the systematic ordering of numerical
operation which the matrix calculus affords. The necessary programming
for the digital computer is simplified since it can be preprograp1med to
carry out matrix operations with only simple orders as to location and
size of the matrix concerned and the operation to be performed. The
specific programming for a particular problem may therefore be written
comparatively quickly and easily and, moreover, follows closely the
algebraic analysis.
As developed here, the matrix methods of analysis follow from particular
forms of the two fundamental energy principles applicable to structures
made up as an assembly of discrete elements. The one principle leads to
an analysis in terms if displacements as unknowns (displacement method),
while the second leads to an analysis in terms of forces (force method).
Besides revealing more clearly the duality of the two methods, this
derivation shows also the close connexion between the aproximate
methods (like the Rayleigh-Ritz method) for continuous systems and the
matrix methods for finite assemblies. This is particularly valuable in
providing suitable techniques for establishing the basic properties-
stiffness and flexibility-of the individual elements of a complex structure
where these elements have to be assigned simplified stress or strain
patterns.
But in stressing the advantages of a unified approach to these diverse
problems, a word of caution is necessary against carrying over into the
modern methods too many ideas associated with practical calculations
by the established or classical methods. The ability to tackle successfully
problems in which the number of unknowns is measured in hundreds
carries with it the necessity of rethinking one's practical approach if
maximum advantage is to be gained from modern computational tech-
niques. In the force method of analysis the choice of basic system and
of the redundant forces must be governed primarily by the requirements
of simplicity and standardization, in orJer to reduce the m:mual
preparation of data to a minimum, and reduce the proba!Jility of errors.
At the time of publication of the original articles it was intended to
reprint them as a single volume and to follow up the Parts I and II,
contained here, with further parts dealing specifically with the practical
application of the matrix methods. Unfortunately it was not possible, for
a number of reasons, to complete this plan and the articles have for some
time teen unavailable. Sir.ce there appears to be a persistent interest in
them the present reprint has been produced to meet the deficiency.
Grateful thanks and acknowledgment are due to the Editor of Aircraft
Engineering for permission to reprint the articles in this form. The method
of reproduction has not permitted complete rearrangemc:1t of the text
into book form, so that the divisions into monthly instalments are still
marked by blank spaces. However, errors in the text have been corrected
as far as possible, and the pages have been renumbered consecutively
to make for easier reference. Grateful tha:1.ks are due to Miss J. A. Bergg
for her care and skill in effecting these changes. The author would also
like to thank here those correspondents who have written to point out
textual errors and misprints.
A list of references to further work is also appended. These are all
concerned with the matrix methods of analysis whose basic theory is
developed here. In particular, Ref. 6 is an expanded and developed
form of part of the work which was initially planned for the original
series.
FURTHER REFERENCES TO RECE1'11T WORK
1. J. H. Argyris and S. Kelsey. "Structural Analysis by the Matrix
Force Method, with applications to Aircraft Wings". Wissenschaft-
/iche Gese/lschaft fur Luftfahrt, Jahrbuch 1956, p. 78.
2. J. H. Argyris and S. Kelsey. "The Matrix Force Method of Structural
Analysis and some new applications". Brit. Aeron. Research Council,
R. & M. 3034, February, 1956.
3. J. H. Argyris. "Die Matrizen-Theorie der Statik". lngenieur Archiv,
Vol. 25, No.3, p. 174, 1957.
4. J. H. Argyris. "On the Analysis of Complex Elastic Structures".
Applied Mechanics Reviews, Vol. I I, No.7, 1958.
5. J. H. Argyris and S. Kelsey. "Note on the Theory of Aircraft
Structures". Zeitschrift fur Flugwissenschaften, Vol. 7, No. 3, 1959.
6. J. H. Argyris and S. Kelsey. "The Analysis of Fuselages of Arbitrary
Cross-Section and Taper". Aircraft Engineering, Vol. XXXI, No.
361, p. 62; No. 362, p. 101; No. 363, p. 133; No. 364, p. 169; No.
365, p. 192; No. 366, p. 244; No. 367, p. 272; 1959. (To be published
in book form by Butterworths Scientific Publications.)
 Part I. General Theory
By J. H. Argyris
T
1. INTRODUCTION
HE increasing complexity of aircraft structures and the many exact
or approximate methods available for their analysis demand an
integrated view of the whole subject, not only in order to simplify
their applications but also to discover some more general truths and
methods. There are also other reasons demanding a more comprehensive
discussion of the basic theory. We mention only the increasing attention
paid to temperature stresses and the realization of the importance of non-
linear effects. When viewed from all these aspects the idea of presenting a
unified analysis appears more than necessary.
With this present paper we set out to develop a comprehensive system
for the determination of stresses and deformations in elastic structures
based on two fundamental energy principles. Although much of the theory
given has naturally been known for many years we believe that some of the
theorems and the generality of the results are new. The loading systems
considered are of an arbitrary nature and include ab initio the effect of
temperature or other initial strains. Neither do we restrict ourselves to
elastic bodies obeying Hooke's law but take account of purely elastic non-
linear stress-strain laws. This is possibly not of very great importance at
present but may have wider applications in the future. No problems of sta-
bility will be touched upon in the present series of articles and any other
considerations of large-deflexion theory are, in general, omitted. Thus the
purpose is to investigate, within the small-deflexion theory, the stresses
and deformations in elastic bodies not necessarily obeying a linear stress-
strain law and under any load and temperature distribution. Dynamic
effects are initially not considered and hence it is assumed for the present
that the loads and temperature are of the quasi-static type. When in-
vestigating thermal strain effects we ought strictly to base the analysis on
thermodynamic considerations. These are, however, only slightly touched
upon here.
As in all theoretical work, we start by discussing the exact implications
and equations derived from the initial assumptions, but we do not restrict
ourselves here to this aspect. On the contrary, we pay close attention to
approximate methods of analysis based on the physical concepts of work
and strain energy. In particular we attempt to give upper and lower bounds
to overall properties of the structure such as its stiffness. No attempt is
mad!! to estimate the error of stress and deformations at any particular
point.
This series of papers originally arose12 • 13 from lectures given by the
author since 1949-50 at the Imperial College, University of London.
Naturally, the scope of the present work has grown beyond the narrower
concept of undergraduate teaching, but the basis of the analysis dates back
to that time. It is appropriate here to point out that certain of the basic
ideas originate with Engesser2 who unfortunately does not seem to have
followed them up. We refer, of course, to the two complementary con-
cepts of work and complementary work. If we consider an ordinary load
displacement diagram, then, even if we restrict ourselves to small dis-
placements, this may be curvilinear, if the material follows a non-linear
stress-strain law. Work is the area between the displacement axis and the
curve, while complementary work is that included between the force axis
and the curve. Thus, the two areas complement each other in the rec-
tangular area (force) x (displacement) which would be the work if the
ultimate force were acting with its full intensity from the beginning of the
displacement. Naturally, in the case of a body following Hooke's law, the
two complementary areas are equal, but it is still useful for the purpose of
analysis to keep them apart. Since writing a previous paper12 on the
subject the author has had the opportunity of consulting the most in-
teresting latest book 9 of Stephen Timoshenko. There a reference is made
to the work of Westergaard, 11 who indeed has developed further the basic
ideas of Engesser, but not on quite such a general basis as here. Since
approximate methods figure prominently in this paper reference ought to
be made to the work of Prager and Synge. They too set out to develop
systematically the determination of upper and lower limits to strain
energy, restricting themselves, however, to Hooke's law and excluding
temperature effects. Moreover, it appears that although many of their
2
GENERAL REFERENCES
(I) Biezeno, C. B., and Gramme!, R. Technische Dynamik, 1st ed., Springer, Berlin, 1939.
(2) Engesser, F. Z. Architek u. lng. Verein Hannover, Vol. 35, pp. 733-774, 1899.
(3) Lord Rayleigh. Theory of Sound, 2nd ed., Vols. I and II, Macmillan, London, 1892 and 1896.
(4) Maxwell, J. C. Phil. Mag., Vol. 27, p. 294, 1864.
(5) Mohr, 0. Z. Arch. u. lng. Verein Hannover, 1874, p. 509, and 1875, p. 17.
(6) Mueller-Breslau, H. Die neueren Methoden der Festigkeitslehre und der Statik der Baukon-
struktilmen, 1st ed., Korner, Leipzig, 1886.
19 ~7. Southwell, R. V. Introduction to the Theory of Elasticity, 2nd ed., Clarendon Press, Oxford,
19~8/ Timoshenko, S., and Goodier, J. N. Theory of Elasticity, 2nd ed., MacGraw-Hill, New York,
(9) Timoshenko, S. History<;{ Strength of Materials, MacGraw-Hill, New York, 1953.
(10) Tretftz, E. Handbuch der Physik, Vol. VI, Springer, Berlin, 1928.
_(II) Westergaard, H. M. 'On the Methods of Complementary Energy,' Proceedings Amer. Soc.
C.v. Engrs., 1941, p. 190.
(12) Argyris, J. H. Thermal Stress Analysis and Energy Theorems, A.R.C. 16,489, Dec. 1953.
(13) Argyris, J. H., and Kelsey, S. Applications to A.R.C. 16,489, A.R.C. 16,513, Jan. 1954.
Additional references are given as footnotes.
results ar~ identical_ with exist~ng _ideas _they clothed them in a language not
too fam1har to engmeers. Th1s d1scuss1on of past authors' work brings us
to a few points which are preferably stated now. In much of present day
structur~l analy~is there seems to be an unfortunate tendency to over-
emphasize certam methods of analysing redundant structures and to
negl~-;t more useful ideas readily available for many years. This refers
particularly to Castigliano's principles which are so often set out as the
basis of all ·considerations, not only in theory, but also in the actual
methods of calculation. This is, in our opinion, unfortunate, even though
all methods naturally lead to the same results if based on the same
assumptions. For example, if we select forces as redundancies then much
the best means of obtaining the basic equations for their determination is
the long established O;k method of Mueller-Breslau based on the Unit
L<?ad idea. We do not ne~d, _in fac~, even the concept of strain energy for
th1s purpose. All we reqmre 1s the 1dea of work and kinematics as used in
rigid-body_ mec~anics. From such ~deas we can write down immediately
our equatiOns m the unkl!owns Wlt~o.ut b~thering about strain energy.
Thes~ l!lethods h~ve been m use by CIVIl engmeers for the past sixty years
and 1t IS surely t1me that we accepted them in the aeronautical world as
standard analytical equipment. Actually, the basic principles go much
farther back than Mueller-Breslau and were, in fact, developed independ-
ently ~Y Ma:'we_11 4 and Mohr 5 nearly a hundred years ago. The first sys-
~ematlc app~1cat1_on of_the ~ik method to stressed skin structures was given
m the classical mvest1gat10ns of Ebner.* Regrettably enough this lucid
wor~ was occasionally referred to in the past as obscure, a lack of compre-
hensiOn, no doubt, at leas_t partly due to the too narrow understanding of
redundant structures ansmg from a concentration on Castigliano's
methods. However, the limitations of Castigliano's formulation of the
problem are being at last increasingly recognized in aeronautical circles
due to the demands of calculations for highly redundant systems. Natur-
ally, most of ~he_ alternative methods suggested are really nothing more
than a transcnpt10n of the Mueller-Breslau and Ebner technique.
We start our investigation in Section 3 with a discussion on work and
complementary work in the presence of temperature effects and for non-
linear stress-strain laws. With this basic knowledge we then proceed to the
standard principle of virtual displacements or virtual work in Section 4.
This is very similar to the currently used principle in rigid-body mechanics.
Thus, we consider a state of equilibrium, apply virtual displacements to it
and develop hence the classical principle of virtual work which sub-
stitutes, of course, for the equations of equilibrium. Since virtual dis-
placements are kinematically possible ones this theorem starts from the
assumption of inherent compatibility to find the necessary and sufficient
condition for equilibrium. It is, of course, well known that the theorem
applies also to large displacements but this aspect is ignored here. How-
ever, temperature effects and an arbitrary law of elasticity are considered
as _lo~g as the latter is '!lonotoni_cally incre.1sing. Having established this
prmc1ple we deduce easlly some Important theorems and applications.
• See e.g. H. Ebner and H. Koeller. 'Zur Berechnung des Kraftverlaufes in versteiften Zylin<Jer·
schalen.' Luftfahrtfnrschung, Vol. 14. No. 12, December 1937.
 Firstly the principle of virtual displacements may always be used to
derive, for any particular structural problem, the governing differential
equations and the appropriate static boundary conditions in terms of the ~/_;_1____,..../

)::_wx
displacements. This method, however, is not recommended in general as a
substitute for the derivation from consideration of equilibrium and elastic
compatibility.
Next the principle of virtual work is used to derive Castigliano's theorem Wz
Part I, generalized for thermal effects. As is well known, this principle ""'.,.. / L - - - - - - -
applies not only for non-linear stress-strain laws but also for large dis-
placements. Our line of argument leads us then naturally to the principle
of minimum strain energy for a fixed set of displacements and a given
temperature distribution. This theorem applies also for non-linear stress-
strain laws and is of great interest for approximate calculations in terms Stresses Body Forces
of assumed forms of displacements. It shows us that, while the strain
energy is for a given set of displacements a minimum when the compatible Fig. !.·-Stresses and body forces
state is also one of ·equilibrium, it is on the other hand a maximum for a
given set of forces under the same conditions. These theorems were first
developed for linearly elastic bodies by Lord Rayleigh more than seventy-
five years ago. They are shown to apply also in the presence of thermal
strain and for non-linear elasticity. In the remainder of the chapter we in-
vestigate in more detail approximate methods of analysis using the
Rayleigh-Ritz procedure and it is in such applications that the principle
of virtual displacements shows its greatest power. The particular form of
the Rayleigh-Ritz procedure known as the Galerkin method is also dis-
cussed. It is of importance when the assumed deformations satisfy all
boundary conditions. The methods indicated apply again in the presence
of thermal strains and non-linear stress-strain laws. The next, Section 5,
gives simple illustrations to the method of virtual displacements.
The second fundamental principle is developed in Section 6. We call it
the principle of virtual forces or complementary virtual work. Here we
consider a state of equilibrium, apply a statically consistent and infinitely z
small virtual force and stress system and find, by using the idea of com-
plementary work, the second principle. This is a necessary and sufficient Stresses Surface Forces
condition that the position of equilibrium is also one of elastic com-
patibility. Again this theorem may be used to derive the differential Fig. 2.-Stresses and surface forces
¢quations of any particular problem, this time in terms of stresses or stress
resultants. However, our comments on the parallel method in the case of a""' a •• , a.. direct stresses }As shown in

l
the virtual displacements are equally applicable here. It should never be axu=a.z, a •• =a •• , a.,=a,, shear stresses FIGS. I and 2
used as a substitute for more physical and geometric reasoning. vu vv vw
Next, we derive what is essentially a generalization of Castigliano's Yxx=;)X' Y••=S)i' Yzz=vz total direct strains
Part II theorem. Contrary to what is generally believed this theorem does
apply for non-linear stress-strain laws as long as we replace strain energy
by complementary strain energy, which is defined in the same way as y.,.=~ +~;, y•• =~ +~;. y.,=~: +~ total shear strains (
(I)

complementary work. It is extended to include temperature effects. We

proceed then with the generalization of Castigliano's principle of minimum
g=y.,.,+y•• +y.. J
strain energy (or least work) for non-linear stress-strain laws and thermal Y)xx• YJ••• Y)zz initial direct strains (e.g. thermal strains)
strains. Some interesting developments derive from this and are given in Y),., Y)., Y)zx initial shear strains
the form of maximum and minimum theorems complementary to those etc. elastic direct strains

}
Exx=y,.,~Y)xx
developed under the virtual displacement method. They do not seem to
have been given previously in this form and provide a useful background Exu=y,.~Y),. etc. elastic shear strains (2)
to approximate methods. They show us that any assumed statically equi- e =Exx+Euu+Ezz
valent stress distribution must always under-estimate the stiffness. This is dV=dxdydz element of volume
most valuable for practical purposes and is exactly opposite to the effect
of assumed displacement distributions which always overestimate the dS element of surface
stiffness. The two in conjunction give us hence lower and upper bounds a linear coefficient of thermal expansion (may vary
to overall characteristics of the structure such as its stiffness. In this with 0)
section we discuss also the Unit Load Method which, as mentioned 0 rise of temperature
previously, provides the basis for one of the more convenient methods for
the calculation of displacements and of redundant forces. It is shown to E Young's modulus }
be applicable to structures with non-linear stress-strain laws. Section 7 G shear modulus May vary with 0
presents some simple illustrations of the principle of virtual forces. v Poisson's ratio
In the last section we develop a slightly more generalized version of the
D;k method of Mueller-Breslau. These equations lend themselves readily ay=axxYxx+a •• y •• +azzYzz+axu/'xu+a•• y •• +azx/'zx ................ (3)
to presentation in matrix form. Next we obtain the corresponding equations The corresponding explicit expressions for UYJ and aE
when displacements and not forces are introduced as the unknowns. are obtained by substituting the strains Y)xx etc. and
Exx etc. respectively for Yxx etc.
A Note on the Mathematics W work of external forces
The mathematics used in this paper is, in general, elementary and U,= ~ W+const. potential (energy) of external forces
should be familiar to any university graduate. We have avoided the more
formal application of the .calculus of variations which can be singularly strain energy (or potential energy of elastic de-
unattractive to those more physically inspired. Chapter 3 and parts of formation)
Chapters 4 and 6 may prove, at first, rather difficult for a student. How- W*, U, *, Ui * complementary work, complement:uy potential of ex-
ever, it is always possible to gain an understanding of the basic ide:1s by ternal .forces and complemenhry potential energy of
substituting simple examples (e.g. frameworks) for the necessarily more elastic deformation
general proofs given here. Ud* complementuy pot-:::-~ti:ll energy of totli deformation
The later parts of this series of papers will present a number of applic:l-
tions of the basic methods developed here. From a consideration of equilibrium on an ele:nent dV=dxdydz, illus-
trated for the x-direction in FIG. 3
2. BASIC EQUATIONS AND NOTATION OCI,,+oa., . Oazx w =O
0Y -f' 0Z + X

I
body forces (e.g. gravity forces) per unif1 p II 1 OX
volume ara ~ to a
Cartesian
cp.,
c/>z, c/>z s.urface forces per unit surface co-ordinat-:: system (4)
I, m, n direction cosines of external normal to ] Ox, Oy, Oz
surface
(see FIGS. I and 2)
u, v, w displacements
3
 where
L1·1· o"'I/J X'I/J ()''f
~~+~-,,+~and/, m, n are the direction cosines on the surface.
'I' oX" o)'" oZ"

This theorem can be proved by integration by parts.

Examples of application:
or/; or/; orf;
take cp=am sx=ou, ~y~Sl =0
Then
· J J
Jo;~, ou·dV -- -- aa· 8yu·dV + lau· ou·dS
v v s

where 8y,,~c8~=~8u is the increment of the strain y,, corresponding

to ou.
. . or/; or/; 81/J
Fig. ).-Internal equilibrium conditions Stmtlarly take cp=a,., ox=ov, oy=ou, oz =0
Then
From a consideration of equilibrium on the surface (see FIG. 2) J[();~"ov+();,•ou]dV=- Ja,.oy£.dV+ Jax.[ mou+lov]ds
fa,x+mavx+ nazx=cfx v y v s
where
" --o[()u o v] _(lou ()ov
(5)
oy,.- ()y +ox -- oy +ox
nazz+ laxz+ma.z=cfz
Note that although Green's theorem is helpful for the mathematical under-
Over part of the surface the boundary conditions may be expressed in standing of the present theory it is not really necessary for the physical
terms of stresses or forces (static boundary _condi~ions) ~nd over the r~­ understanding; a reader unfamiliar with these aspects of the integral
mainder in terms of displacements or strams (kmemattc or geometnc calculus may omit the relevant parts.
boundary conditions). Naturally, the boundary conditions may be of
both types over the sat:I?-e part of the surf~ce: Consider, for example, the
tube shown in FIG. 4. It ts assumed fully bmlt mat the roo_t (z=O) and _free 3. WORK AND COMPLEMENTARY WORK-STRAIN ENERGY
at the tip (z=/). Ribs rigid in their own plane but freely flextble to defle_x~ons AND COMPLEMENTARY STRAIN ENERGY
out of their plane are assumed at z =0 and z =I. The boundary condttlons The analysis of the present paper is restricted to small strains which can
are: at z=O, u=v=w=O, i.e. pure kinematic conditions; at z=l, a•• =O, be expressed by the linear formulae given in the notation. Such displace-
ments and corresponding strains are obviously additive (algebraically).
( ov)
oy z~! =0 for the vertical walls and (~u) =0 for the horizontal Thus, if u 1 , y 1 , and u 2, y 2 are displacements and strains in a deformed
uX zd
state 1 and 2 respectively, then u 1 + u 2, y 1 +y 2 represent also a compatible
walls, i.e. both static and kinematic conditions. state of deformation of the body. Our assumption does not impose,
however, a linear stress-strain relationship; hence if P 1 , a 1 and P 2 , a 2
To denote infinitesimal elements of geometric properties of the structure are the forces and stresses corresponding to the above two states of
(e.g. co-ordinates, area, volume) we use the standard symbol. d. deformation of the body, the forces and stresses corresponding to the
To denote infinitesimal increments of forces, stresses, dtsplacements, deformed state u 1 +u 2 are not P1 +P2 and a 1 +a 2 except in the case of a
strains and work we use the symbol o. . . . linearly elastic body. In all cases, however, the stress-strain law is assumed
Thus, dV=infinitesimal element of volume=dxdydz, oP mfimtestmal to increase monotonically as shown in FIG. 5. In conclusion we can state
increment of force P. that the law of superposition is assumed to hold for strains and displace-
The symbols ments but not necessarily for the stresses.
J( ..... .)dV, J(...... )dS In general, we assume also that the displacements are so small that the
v s equilibrium conditions can be written down for the undeformed body. It
denote integrations over a volume and surface respectively. follows then that the question of stability or instability of equilibrium does
not enter in the analysis of this paper and there is a unique solution to
The formal mathematical proof of some of the basic theorems in this every problem.
paper is shortened by using Green's theor~m. * ~t and if! be two con-f Consider a three-dimensional deformable body (not necessarily elastic)
tinuous functions and let also the first partial denvattves of(/> and the first in equilibrium subjected to a self-equilibrating system of body forces Wx
and second partial derivatives of rf; be also continuous. Green's theorem etc., surface forces cfx etc. and a temperature 0. These forces and tem-
states: perature may vary with time but the variations are assumed so slow that
the dynamic effects are negligible. Let, in a time interval ot, the forces
J[oo/
ox
or/;+ oo/ or/;+ oo/ oi/J]dv
ox oy oy oz oz increase by Owx, Ocfx etc. and the temperature by 80. The displacements
increase at the same time by ou etc. There arises hence an increment of
v
work (see FIG. 6)
oW= J[wxou +w.ov+w.ow)dV
v
+sHo/xou+cp.ov+cp.ow]dS
• See Courant, Differential and Integral Calculus, translated by J. E. McShane, Blackie and Son
Ltd., London and Glasgow, 1949, Vol. II. + terms of higher order.

Y,V

Rigid rib E
Fig. 4.-Kinematic and static boun- Fig. 5.-Stress-strain diagram Fig. 6.-Work and complementary work; strain energy and comple-
dary conditions mentary strain energy
4
 t
and that it is subjected both to external loads and thermal effects. In view
Oyy of our initial assumption about the smallness of the strains we can write
lhdy
_EY________________ , 0Yxx=O€xx+01]
(12)

-
I
I
Oxx I~
0Yzz=O€zz+01]
I where OExx .... , 811 are the increments of the true elastic strains and the
thermal strains
e
'Y/xx=1]yy=1]zz=1] = faoe (13)
0
a is the linear coefficient of thermal expansion which may vary with e.
In view of the local isotropy of the body, thermal expansion does not give
rise to any angular displacements.
Hence
Oyxy=O€xy, 0Yuz=O€yz. 0Yzx=O€zx • · · · · · · · · · .. · · • •. • • • • ... • • (14)
Substituting Eqs. (13) and (14) into (ll) we obtain
Fig. 7.-Work of direct and shear stresses
oW=oU;+s811 . . . .. . . . . . .. . . . .. . . .. . .. . .. .. . .. .. . .. .. . . . (15)
and
The terms of higher order involve expressions -!-Swx · Su, -!-Sc/J, · Su etc,
and may be neglected to the first order of magnitude considered here. Thus. W=D;+js811 (16)
0
I
J ...................... (
8W=f£wx8u+w.8v+w.8w]dV where
+~[c/Jx8u+c/J.8v+c/Jz8w]dS 7)
D;=fao€ (17)
s 0
Note that Eq. (7) does not presume any specific force-displacement law, and
be it elastic or non-elastic.
It is simple to derive an alternative expression by considering the aOE=axxDExx+ ........ +axuDExu+ ... +azxDEzx=OU; . . . . . . (l7a)
additive effect of the work done by the stress resultants on each volume Note that the second term in (16) and thus also the work depend on the
element dV. A perusal of FIG. (7) shows that the deformation Su, Sv, ow sequence of application of loads and temperature.
gives rise to an increment of work for an element dV We assume now that the elastic properties of the material depend only
(axxDYxx+a •• Sy •• +a •• Sy •• +axuDYxu+a•• Sy •• +a.xSy.,)dV=aoy·dV
on the instantaneous temperature e
but not on the previous history of
deformation, e.g. if adiabatic or isothermal; the error involved in this
again neglecting terms of higher order. assumption is indeed negligible. fhus, the expression U; is only a single
The incremental (infinitesimal) strains Sy,x etc. are those due to the dis- valued function of the instantaneous state of elastic straining. In a closed
placements Su etc. Thus, cycle of deformation 0; is zero and SD;=dU; is a total differential. The
function U; is commonly called the strain energy per unit volume; other
8Yxx =o~!!
~X
etc ., 8Yx• =o{~ ~}
~Y +~X
={~Su ~Sv}
~Y +~X · · · · · ·'' '· (8) names are strain energy function or density of strain energy.
It follows from Eq. (17a) that
It follows that the increment of work 8 W may also be expressed as "- "
oUi=axxDEu+ · · · •· " ~Di"
+azxD€zx=~O€xx+ ••••· ~Di"
+~O€zx
·sw=J(aoy)dV .. . .. . .. .. .. .. . . .. .. .. .. .. .. .. .. .. .. .. .. .. (9) UE'xx UE'zx

v Hence
The formal equivalence of Eqs. (7) and (9) may be proved without difficulty ~Di ~Di
by using Green's theorem. To this effect multiply each of the internal Uxx=()Exx • ••• , ., Uxy=(>E'xy (18)
equilibrium conditions (4) by Su·dV, Sv·dV, Sw·dV respectively, and We conclude that the stress-strain law is uniquely determined by the strain-
integrate over the body. We obtain, energy function and vice versa. Note that the law of elasticity is arbitrary
in Eqs. (12) to (18).
J{[~ax"+ ~a.x+ ~a."'+w"']ou+[~a." +~a•• + ~ax• +w• ]ov
~x ~Y ~z ~Y ~z ~x
Integrating Eq. (16) over the body we find,
v e
+ [~;;· + ~;;• +~:;· +w. JSw} dV =0
or
fWdV= fUidV+ f [ fso'Y/]dV
v v v 0

Applying now Green's theorem to the terms involving the stresses, as (~

shown in the previous section, we find, using the surface equilibrium con- W=U;+J [ fs01]]dV (19)
v 0
ditions, Eqs. (5),
where
f[aoy]dV= f£wx8u+w.8v+w.8w]dV + f£c/Jx8u+c/J.8v+c/J.8w]dS=8 W
v v s U;= fU;dV= f [ faOE]dV (19a)
where the Sy's satisfy Eqs. (8) and are hence compatible strains. Note that v v 0
where c/J, is unknown but the displacement (say u) fixed, the corresponding For e =0 Eq. (19) reduces to the well-known equality
Su is zero and hence the relevant terms in the last relation vanish. In- W=U; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)
tegrating Eq. (9) from the initial unstressed state 0 to a final state I we
find the total work U; is the strain energy or the internal elastic potential (or potential energy
I
of elastic deformation) of the body.
W= f { f[w,8u+w.8v+w.8w]dV+ f£c/Jx8u+c/J.8v+c/Jz8w]dS}= fWdV If we now assume that the law of elasticity is a linear one then
0 v s v
.................................. (10) (21)
where and similar expressions for Euu and Ezz· Also
- I
W= f {a,,Sy,,+a •• Sy•• +a•• Sy •• +axuDYxu+a•• Sy•• +azxOYzx} . . (I I)
a,.=Gy,. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22)
0 Expressing (21) in terms of stresses,
Note that, in general, the work W done to reach a state I starting from
a state 0 depends on the path chosen due to say plasticity, viscous effects,
etc. In such cases 8 W is not the total differential* of the right-hand side
axx=2G[Exx+ 1 ~ 2 ve]=2G[rxx+ 1 v2vg]- 1:1v .......... (23)
of Eq. (10). and two similar expressions for a .•.• and a zz· The moduli E and G and the
In what follows we assume that the body is fully elastic and isotropict ratio v may vary with e.
Substitution of the stresses given by Eqs. (22) and (23) into Eq. (15)
yields
• We say that 6W~dWis a total differential of Wif fcdW~ 0, where the integration is taken around
a closed curve; if this applies W is obviously independent of the path of deformation taken between
states 0 and I. 8w=8D;+f~8iv........................................ (24)
t The isotropy need only be assumed at each point; the properties of the body may vary from
point to point. and if temperature ami loads are increased together from zero
5
 - - EeT} - I
W=U;+2(l-2v)........................................ (25) W*=fyoa (34)
0
where Note that as in the case of work W the complementary work depends, in
- { I _-vVe 2 -2[E,,Eyy+EwEzz+E"zzE",,- -;r(E"xy
1 · 2 2 2 }
general, on the chosen path of stressing between 0 and /.
U;=G 1 2 +~: •• +E"zx )] (26) We assume now the body is fully elastic and isotropic as described
previously and find from Eq. (34)
is the strain energy as a function of the elastic strains.
The following relations hold only for linear elasticity, 8W*=o0;*+TJos........................................ (35)
and
oO; oO;
OUxx =E",,=y,,-TJ .. .... , oa,. =£,•...... (27) - - I
W*=U;*+ST}os · · · · · · · · · · · · · · · · · · · · . · · . . . . . . . . . . . . . . . . . (35a)
0
For two-dimensional stress distributions substitute the factor 1/(1-v)
for l/(l-2v) in Eqs. (24), (25) and (26). The corresponding strain energy where
function 0; is,
O;*=f~:ou (36)
0;=G[(I~v)(~:,,+~:,.) 2 -2~:,,~:•• +t~:,. 2 ] .•••••...••.•••••• (26a)
and
0

Parallel to the conceptions of work and strain energy two further ideas ~:ou=~:,,Sa,,+ ....... +~:,.8u,.+ ..... =o0;*
essentially due to Engesser 2 , are of particular importance to our investiga-
tions. Consider to that effect a one-dimensional force-displacement and a Note that the second term in (35a) depends on the sequence of application
stress-elastic strain diagram (FIG. 6). The vertically shaded areas are of loads and temperature.
obviously those of work and strain energy respectively. lt is natural to Under the same assumptions as for the internal potential energy, U;*
inquire if the horizontally shaded areas complementing the previous areas depends now only on the instantaneous state of stress. In a closed cycle of
in the rectangular areas Pu and a~: respectively are of any importance (see deformation U;* is zero and 8U;* is a total differential of the right-hand
FIG. 6). In fact, as is shown farther on, the introduction of these new con- side of Eq. (36). Hence
ceptions is proved a particularly happy one when generalizing some
theorems, currently assumed to be valid only for linear elasticity, to bodies oO;• oO;*
£,,= oa,, ........... , ~:,.= oa,• ........... (37)
with non-linear stress-strain relations. Although the complementary areas
are equal for linear elasticity it is still useful in such cases to differentiate (see also Eqs. (18)). Note that the law of elasticity is arbitrary in Eqs. (37).
between them. If 0 1* is given in terms of the stresses Eqs. (37) show that this determines
It is interesting to note that in thermodynamics two similar comple- uniquely the strain-stress laws and vice versa. It is natural hence to call
mentary functions are used : the free energy function H of von Helmholtz 0; * the complementary strain energy function.
and the function G of Gibbs.* In what follows we call the horizontally
shaded areas complementary work and complementary strain energy and Integrating over the body we find
denote them by w• and U;* respectively. I
We generalize next our new conceptions by considering the three- W*=U;*+f [ fTJos]dV (38)
dimensional case. Let the actual displacements in a body subjected to body v 0

forces w, surface forces cp and temperature 0 be u, v, w. The increment of where

i.he complementary work as these displacements increase to u + 8u,
v+8v, w+ow due to load increments 81/J, .... , 8w, . .... etc. is given by U;*=fO;*dV=f [f~oa]dV ............................ (38a)
v v 0
S W* = f[uSw, + v8w. + wSw.]dV + f[uoc/J, + voc/J. + wSI/J.JdS where U;* is called the complementary strain energy or complementary
v s elastic potential energy. When 0 =0 then
+terms of higher order W*=U;* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (39)
or In the case of linear elasticity W* = Wand U; • = U;, but it is useful to
8 w• = f[u8w, + v8w. + wSw.]dV + f[u81/J, + voc/J. + wScp.]dS (28) differentiate still between them. For the linear stress-strain laws ofEqs. (21)
v s and (22), Eq. (35a) becomes, if temperature and loads are increased to-
since terms of higher order like tor/J,· Su can be neglected. It is simple, gether from zero,
as in the case of work, to derive an alternative expression to Eq. (28) in
terms of stresses and total strains. To find it, note that the increments w*=U;*+t .......................................... (386)
Sa,, etc. of the stresses must be in equilibrium with the corresponding
increments of the body forces 8w, . . . and surface forces 81/J, ..... and
There are thus six relations of the type of Eqs. (4) and (5). We write here
only the two for equilibrium in the x-direction -. _ _!_[L
U; - 4G l+v- 2<a,,a•• +a••a •• +a ..a,, _ a,. 2_ a.. 2_ a., 2)] (40)
o(8u,,) o(Sa.,) o(Sa.,) "'w _ 0 which for two-dimensional stress distributions reduces to
ox + oy + oz +o .,- ........................ (29)
and
- * =4G
U; 1 [(u,,+a
I+v•• )2 - 2(a,,u•• -a,. 2)] ....... ······· ..... . (40a)
loa,,+moa.,+nou.,=oc/J, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (30) The above considerations on strain energy and complementary strain
Multiplying now each of the first set of equations by the displacements energy may be used to derive all Castigliano theorems as generalized for
u, v, w respectively, summing and integrating over the body we obtain by non-linear stress-strain relations. We postpone, however, these investiga-
applying Green's Theorem similarly to when we derived Eq. (9), and tions to subsequent chapters.
using Eq. (30) When the body obeys linear stress-strain relations the principle of super-
position applies also to forces and corresponding stresses. Some important
f[u8w, +vow.+ wSw.]dV + f[uoc/J, + voc/J. + w8cp.]dS theorems derive from this property.
v s
Thus, if the forces on a body are increased from zero to their final value
= f[y,,Sau +y•• Sa•• +y•• Sa •• +y,.Sa,. +y•• Sa•• +y.,oa.,]dV then
v W= W*=tf[w,u+w.v+w.w]dV+tf(c/J,u+c/J.v+c/J.w]dS I

J
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (31) v s
where y,, etc. are the total strains. Note that where the forces, for example, (41)
U=U*=tf(a~:)dV
1/J,, have fixed values or/J,=O. Thus ultimately v
Sw• = fyoadV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (32) where all symbols in the brackets refer to final values. Eqs. (41) are known
v as Clapeyron's theorem. Another very useful theorem is due to Betti. •
where Assume that the body is subjected to two force systems P 1 and P 2 • Let
yoa=y,,Sa,x+ ......... +y, 11 8a,,,+ . . . . . . . . . . . . . . . . . . . (32a) the deflexions due to system P 1 alone be u 1 and due to P 2 alone u 2 • By
applying first system P 1 and subsequently P 2 a~d _then reversing the pro-
Integrating Eq. (32) between the initial unstressed state 0 and a final cess we prove easily-noting that the final state IS m each case the same--
state I we find the total complementary work that
w• = vf W*dV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (33) W 12 = W21 • • • • • • • • • • . • • • • • • • • . • . • • • • • • • • • • • • • • • • • • • • • • • • (42)
where
where
W12 =~P 1 u 2 and W2 =~P 2 u 1 • • • • • • • • • • • • • • • • • • • • • • • • • • • • (42a)

• The sign of the latter function is taken as that fixed by the Committee of the International Union
ol Physics. • See Num'o Cimento (2), Vols. 7, 8, 1872.

6
are the work done by the system of forces P 1(P2) over the displacements
u2(u1) respectively. Relation (42) is known as the generalized reciprocal
theorem of Betti.
A special form of Eq. (42) is Maxwell's reciprocal theorem. Thus, if
systems l and 2 consist each of one force (or moment) only then
l. u12 =1. u21 •••. . •••. •••. . ••. •. . •. . . . . . . . . . . . . . . . •. •. •
where u12(u 21 ) is the displacement or rotation in the direction of force or
moment P 1(P2) due to force or moment P 2(P1 ).
4. THE PRINCIPLE OF VIRTUAL DISPLACEMENTS OR VIRTUAL
WORK
We assumed in the previous section when discussing work and strain
energy that the displacements Su, Sv, Sw arise from an actual variation
of the applied forces and/or temperature distribution. However, this is an
unnecessary restriction. We need only remember that to the first order of
magnitude considered Sw and SU; are independent of the SP's and
corresponding Sa's since we ignore terms of the order SP· Su. Also for
the purpose in hand no variation in the temperature is called for. Hence,
when finding Sw and SU; we can assume that forces, stresses and tem-
perature remain constant while the displacements are varied to u+Su,
v+Sv, w+Sw. It is only necessary that the Su, Sv, Sw's are compatible
infinitesimal displacements (see Section 3, p. 4); thus they must be piece-
wise continuous in the interior and satisfy the kinematic boundary con-
ditions. For example, if the u, v, ware prescribed on part of the boundary
then the selected variations Su, Sv, Sw must be zero there too. Similar
arguments apply if the derivatives of any of the displacements are fixed.
However, where the forces and stresses are prescribed the variation of the
8u's is necessarily free. Note that there are cases when it is useful to relax
even the kinematic boundary conditions when selecting the Su's.
Such geometrically possible infinitesimal displacements are used ex-
tensively in rigid body mechanics and are called virtual displacements.
Noting that the temperature and hence thermal strains remain constant
we can restate now Eqs. (15), (19) and (19a) more generally as follows:
An elastic body is in equilibrium under a given system of loads and
temperature distribution if for any virtual displacements Su, Sv, Sw from
a compatible state of deformation considered
8W=SU; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
which is the standard principle of virtual displacements or virtual work.
As stated here it is valid for an elastic body subjected both to loads and
temperature effects.
Note that Eq. (44) is also valid for large displacements but the formula-
tion of U; is in such cases more complicated since the strain expressions
are not any longer linear in the displacements and the equilibrium con-
ditions have to be considered on the deformed element.
The point made above that the virtual displacements are arbitrary as
long as they are infinitesimal and satisfy the internal compatibility con-
ditions and kinematic boundary conditions is worth emphasizing. To fix
ideas consider the statically determinate framework shown in FIG. ts)
subjected to a transverse force P. We apply to the system a virtual dis-
placement Su in the form shown in the figure which allows only an elastic
deformation of the upper flange 1, 2. This virtual displacement satisfies
the kinematic boundary conditions
8u7"0 at A and B
but bears obviously no relation to the actual displacements of the frame-
work due to the force P.
We find easily:
. Pab
force m member l, 2: N 12 = -7112
virtual displacement of force P: Su 1 =Su·afb 1
Suh Suh lh
virtual elongation of member I, 2: !i/12 = - b - b =-SuiT
1 2 1 2
-,--~~~~~~~--~--~---
h bars. In the more general cases of self-straining not only may the "YJ~x•
t "YJ•• and 7]zz be different, but there may also arise initial shear strains 7].,.,
Fig. a.-Example of an arbitrary virtual displacement
B
Thus,
SW=PSu~ and SU;=( _f'7: 2)x( -t~~)=PSu~
Hence
Sw=SU;
(43)
as indicated by the principle of virtual displacements or virtual work.
In order to realize best some of the implications of the principle of
virtual displacements let us consider again the derivation of Eq. (44)
which applies when S7J =0. In accordance with the analysis of p. 5, if
we multiply the internal equilibrium Eqs. (4) with the virtual displace-
ments Su, Sv, Sw, sum the three expressions, integrate over the body,
apply Green's theorem Eq. (6), and note the boundary conditions (5) we
obtain Eq. (44). Again if we start from Eq. (44) we can apply Green's
theorem in the opposite direction and are led to
J{[ ()a,, ()a• ., ()a •., Jc;,
i)x +~+~+w., ou+[ .....
c;,
.]ov+[..... . ]ow dV
c;, }
v
= J{ [la,.,+ma•.,+na•.,-4>.,JSu+[..... .]8v+[..... .]Sw}ds
s
For arbitrary virtual displacements Su, Sv, Sw this relation can only be
true if each of the brackets vanishes separately; thus, starting from the
principle of virtual work we have re-established the conditions of equili-
brium. An exception occurs, of course, where a displacement (say u) is
fixed on the surface and Su is automatically zero there.
We conclude that the principle of virtual displacements Eq. (44) is a
necessary and sufficient condition for the existence of equilibrium of an
elastic body. Or otherwise we can state that by using virtual, i.e. kinema-
tically possible compatible displacements, we substitute Eq. (44) for the
internal and external equilibrium conditions. Note that the idea of strain
energy is not necessary to the establishment of the principle of virtual
work.
Since the forces are assumed to be applied on the undistorted system
and to remain constant during the virtual displacements we can regard
S Was the variation of a potential - U,. Thus
Sw= -Su, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (45)
(44) and
U,= -.f[w.,u+w.v+w.w]dV-H4>,u+4>.v+4>.wJdS (46)
v s
Thus
()U,
w.,=-7iil
which is the usual definition of a potential of forces. Note that Su. is a
total differential of the elastic displacement increments Su. U, is denoted
as potential of external forces.
We can write now Eq. (44) in the concise form
S,U=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (47)
where the suffix E indicates that only elastic strains and displacements are
varied.
U=U;+U.+const. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (47a)
is the total potential energy of the system. Eq. (47) states that a position
of elastic compatibility of an elastic body is also one of equilibrium
(i.e. the body is at the true position of equilibrium) if any virtual variation
of the displacements and strains whilst forces, stresses and temperature
remain constant does not give rise to any (first order) variation of the total
potential energy. The particular form (47) is known as the principle of a
stationary value of total potential energy, if the latter is expressed in
terms of displacements. Note that U; itself may be calculated from
formula (l9a).
Actually the stationary value of U is in our case always a minimum
and this confirms our previous assertion that with the assumptions of our
analysis all systems are stable. The mathematical proof that U is a minimum
at the true position of equilibrium is straightforward •; the point is dis-
cussed in greater detail under (C) below.
We have assumed until now that the initial strains 7J are due to a tem-
perature variation. However, this is an unnecessary restriction and there
may be strains arising from any source of self-straining. For example, in a
framework they may be due to manufacturing errors in the lengths of the
7Jvz and 7]zxo In such problems substitute
a87] =a.,.,S7]xx+a•• 87]vu+a ..S7]zz +a.,.S7].,. +a.,S"Y)vz +a,.,S7]zx . . . . (48)
for sS7J in Eqs. (15) and (19) and in the other related expressions.
Equations (44) or (47) may be used to derive the results which follow.
(A) The differential equations ofthe theory of elasticity for arbitrary loading
and temperature distribution or any particular structural problem in
terms of the displacements; the appropriate static boundary conditions
• See Biezeno and Gramme! (I), p. 74.
7
in terms of the displacements follow also from this analysis. It is im-
portant to note that in all applications it is best to form directly
SU1 = f[ailE]a v
v of the prevtous paragraph show that the strain energy of the constrained
and not to evaluate first U1 and then to take its increment S.
(B) Castigliano's* theorem Part I generalized for thermal effects
[~~~ 1~= const. =P, · · · · · · · · · · · · · · · · • · • • · · · · • • • • · • • • • • • • · • (49)
where P, is the force (moment) applied in the direotion of the deftexion
(~otation) u,_. TJ:lls relation may be obtained immediately if we apply a
virtual elastic displacement Su, solely to one external load P,. Note that
Eq. (49) applies also for non-linear stress-strain laws and may also be
generalized for large displacements.
(C) The Principle of Minimum Strain Energy when U1 is expressed in
terms of the displacements and the temperature is not varied.
We arrive immediately at this theorem if we select only such virtual
displacements Su which are zero at the applied forces. Then
SW=O
and we conclude from Eq. (44) that
S.U1 =0 and U;=min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (44a)
at true position of equilibrium if only such virtual deformations are
allowed that no external work is done.
Hence, if we compare all possible compatible states of deformation of a
body associated with a given set of displacements (not sufficient by them-
selves to fix completely the deformed shape of the body) then the true
position of equilibrium has the minimum strain energy. This is still true
if the body is subjected to temperature loading.
The point is of sufficient interest to warrant some elaboration. First it
may be helpful to point out that when we state that at the position of
equilibrium the total potential energy has a stationary value and that this
is a minimum we do not compare physically possible adjoining states.
For, in stating that the potential energy has a stationary value, i.e. SU=O,
we compare the true position u, with a position u+Su assuming in both
cases that forces and stresses are the same. This can obviously not be true
for the second position since for given forces there is a unique position of
equilibrium. In fact, we mentioned that this arises due to our legitimate
neglect of the higher order terms in Sa and oP. Also, when we go a step
farther and state that the stationary value is a minimum we prove this by
considering the influence of terms like !Sa,.,· S~:xx in U, arising from the
variation Sa.,., associated with S~:"'"'' but we still keep the forces constant-
although this cannot, in general, be true.
Having pointed out these aspects of the virtual displacements approach
we shall, in what follows, discuss the question of the extremum of U1 from
a more physical point of view. Again we prescribe certain displacements
on the body and do not allow any forces P other than those arising due
to and in the direction of the given displacements. The structure takes up
its natural position of equilibrium from which we can deduce the value
of the forces P. If we want to force the body to assume a position u+ou,
v+ov, w+ow while keeping the set of prescribed displacements constant
we must apply certain additional body and surface forces to push the
system away from its natural configuration. The work done by these
constraint forces (!'~.;oP· ou, obviously positive), produces by reason of
equilibrium in the new position an equal increase in the strain energy
stored. Thus, the strain energy in any neighbouring compatible configura-
tion is greater than that for the unconstrained original position and hence
the strain energy there is a minimum.
An alternative way of producing a state of equilibrium different from the
natural one in an elastic body under a prescribed set of displacements is
the introduction (prior to the imposition of the displacements) of internal ·
or ext~mal constraints that do no work. For example, in a shell or plate
analysis, we may assume that the middle surface is inextensible and the
transverse shear strains are zero; thus, in this case we impose infinite
values for E and G in the middle surface and an infinite value for G in the
*See A. Castigliano, Theorie de /'equilibre des systemes elastiqu~s, Turin 1879.
_1I
I E,,=sx -YJ, Eyy=a-y -YJ, E,,=sz -YJ
0 E,"=i\Y+ox' E",=oz +s/
Fig. 9.-Stiffening and relaxation
of stress-strain curve
8
!ransvers~ direction: -:'nother type of constraint may be achieved by the
mtroduct10n of a ngt_d support. Also the stress-strain diagram may be
taken local~y to be II mstead of I (FIG. 9). In all such cases the arguments
body for ~he given prescribed set. of displacements is greater than for the
unc~:mstran:'ed body. Conversely, tf we relax any existing constraints whilst
agam keepmg constant the prescribed displacements the strain energy is
decreased. The relaxation of constraints may take the form of a stress-
strain line III instead of I. Alternatively we may introduce a hinge in the
structure. Another example is the case of a shell where we ignore the
bending stiffness and admit only a membrane state of stress· a current
procedure in wing stressing. '
. Thus we co~cl';lde: the strain energy of an elastic body for a given set of
d1sp~acements 1s mcreased (reduced) by the imposition (relaxation) of con-
stramts that do no wor~. ~ n el_(cephon occurs if the effect of imposition or
r~moval of a constr~mt ts mi. For ex_a~pl~, in an infinitely long, thin
circular shell under mternal pressure It ts Immaterial whether we take
account of or ignore the bending stiffness of the walls.
Also since the constraints do no work it follows that the increase
(reduction) of the strain energy can only be produced by the forces P.
Thus:. the fo~·ce system P which is set up at the an_d in the direction of the
f?rescnbed displacements and hence also the st(ffness of the structure is
mcreased (reduced) by the imposition (relaxation) of constraints that do no
work.
. If we consi~er now the case of an elastic body under a given set of forces
mstead of displacements then we conclude immediately from the last
the? rem: the displacements and hence also the strain energy in a body under
a g1ven set of forces are reduced (increas!!d) by the imposition (relaxation)
of constraints that do no work.
Both the last two theorems illustrate two complementary effects of the
action of constraints on the stiffness of a structure. The last theorem may
also be expressed as follows: the strain energy of an elastic body under a
gil•en set of forces is a maxim!lm when it is subjected to the least number
of constraints that do no work.
The immediate application of the above considerations is, of course to
the effect of actual constraints on elastic structures as illustrated in 'the
e~amples m~ntioned. A more important application appears in connexion
wtth approxtmate analyses of deformations. Thus, if we reduce the free-
dom of deformation as we do in the Rayleigh-Ritz and related methods we
always over-estimate the stiffness of tho structure. Hence for a given set
of forces (displacements) we under-estimate (over-estimate) the corre-
sponding displacements and strain energy (forces and strain energy).
_The above theorems on the effects of constraints on strain energy and
sttff~ess appear to have been given first by Rayleigh* in 1875 for linearly
elasttc bodtes and no temperature effects. Our arguments indicate how-
ever, that they apply also to elastic bodies with non-linear stress~strain
relationship and under thermal loading. The original principles are
occasionally referred to as the static analogues of Bertrand's and Kelvin's
theorems in dynamics.
(0) The Unit Displacement method. This method will be developed
in Section 8.
.<E) Approximate ~ethods of displacement analysi~, using the Rayleigh-
Rttzt proce~ure. In tht~ method we_assume_for_the dtsplacements approxi-
mate functtons or senes of functtons sattsfymg the geometric but not
necessarily the static boundary conditions. For example, in a three-
dimensional elastic continuum we may express the total displacements
u, v, w in a finite series as follows:
n
u= Uo(X, y, z)+La,u,(x, y, z)
I
r~l
I
I
n
V=' V0 (x,y,z)+Lb,vr(x,y,z) ...................... (50)
r ''I
n
W='W (X,y, z)+Lc,w,(x,y, z)
0
r --I
where U0 , V00 w, satisfy the kinematic conditions where these are prescribed
and u,, v, w, are linearly independent functions which vanish there.
a,, b,, c, are unknown constants to be determined by the Rayleigh-Ritz
procedure. The elastic strains corresponding to (50) are (see Eqs. (I) and (2))
au av aw
(51)
au av av aw aw au
Eu=ax+az
The chosen series satisfy the displacement boundary conditions and the
infinitesimal deformations
OU=Oa,· u, OV=Ob,· v,, OW=,Oc,· w, . . . . . . . . . . . . . . . . . . . . . . . . (52)
*See Rei.. 3., VoL II. Jl. 94, and also 'General Theorems Relating to Equilibrium and Initial and
Steady Motion , Phd. Mag., March 1875. They have been discussed more recently by D. Williams in
Phrl. Mag. Ser. 7. Vol. 26, 1938. p. 617.
t W. Ritz, "Theorie der Transvers~lschwingungen einer quadratischen Platte.' Ann. d. Physik,
VoL 28. p. 737, 1909; see also J. reme u. angew. Math .• Vol. 135, p. 1. and Gesammelte Werke,
Parts, 1911, p. 192.
obtained by variation of a" b., c, while the other coefficients and the tem-
perature are kept constant are hence virtual elastic displacements. Note
again that the chosen u, v, w functions need n'ot satisfy the given static
boundary conditions. Naturally, the accuracy of the analysis is enhanced
if the latter are also satisfied.
To determine the coefficients a., b., c, we use the condition of the
stationary value of the total potential energy in the forms
DU=O or SU;=DW
There are also cases when the theorem of minimum strain-energy, U; =min.,
is useful, see for example Part H of this series example 4. The applica-
tion of DU=O ensures the average satisfaction of the equilibrium con-
ditions.
If we evaluate U in terms of (50) then for a linearly elastic body we
obtain a quadratic function in a., b., c,. Condition DU=O which in this
case becomes
where the Su's are virtual, i.e. kinematically possible infinitesimal displace-
ments. Eqs. (57) lead to the principle of virtual work Eq. (44) if the a's
satisfy not only the internal but also the boundary equilibrium conditions
Eqs. (5).
Assume now that the displacements u, v, w are written in the approxi-
mate form of Eqs. (50) where the a, b" c, are unknown coe:ficients. How-
ever, contrary to what we assumed in paragraph (4), series (50) are taken
here to satisfy not only the kinematic conditions wb.ere prescribed on the
surface but also by substitution into the stress-strain relations the equili-
brium conditions where prescribed.
Expressing now the stresses in the brackets of (57) in terms of the dis-
placements and temperature 0 and selecting as virtual elastic displace-
ments one of the 3n possibilities (52) we obtain the n equations in the 3n
unknowns a" b, c,
J[ oa~"'+oa.""+oau
ox oy oz
J
+w u dV=O

l
r
ou ou ou v
.t:

J[ oa""+Oaz"+oaM+w JvdV=O
oal = ...... =oa, = ...... =oa, =O
(5S)
ou
Ob 1 = · · · · · ·
ou
=ob, =
ou
=obu =O f ............... . (53)
v 0\'
. OZ OX II r '' '''''' ''' ''' ''

J[o;;z +o;;z+ o;;-- +wz]w,dV=O

· · · · · ·

ou ou ou v
ocl = ...... =oc, = ...... =oc, =O
_j
This is the Galerkin *t procedure usually only given for linear elasticity. To
leads to a set of 3n linear equations in the 3n unknowns a,,b" c,. It is, fix ideas take the case of linear elasticity and write axx etc. in terms of
however, superfluous to evaluate first. U and then to differentiate with u, v, w. We find easily three equations, the first of which is
respect to the unknown coefficients. We can obtain directly the final
equations by forming the 3n expressions
J[~u+l~ 2 v ~!+~_,~ 2i~i~) ~~Ju,dV=O .............. (59)
S.,( U; + U,) =0, 8b,( Ui + U,) =0, De,( U; + U,) =0 . . . . . . . . . . . . . . (54) v
where the suffices ar, br, cr indicate that the virtual displacements are where
chosen respectively as in Eqs. (52).
Using the first of Eqs. (52) in the first of Eqs. (54) we obtain the more o 2u o2u o2u ou ov ow
explicit form
~u=ox2+oy2+oz2 and g=ox +s:Y+oz . . . . . . . . . . . . . . . . . . . . (60)
{faE,dV ~~Pu}Da,=O The bracket in Eq. (59) is, of course, the equilibrium equation in the
v x-direction expressed in terms of displacements.
or since Sa, is arbitrary The procedure in any particular structural problem is to form the
equilibrium conditions in the stresses or stress resultants and express each
faE,dV~~Pu=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (55) in terms of the displacements. Next we multiply each by the corresponding
v
u, (which may be a deflexion, slope, twist) and then integrate over the
where E, are the elastic strains due to u, a the stress due to the elastic body. Thus, for a beam subjected to a distributed loading p in the y-direc-
strains given by Eqs. (51), P the applied forces and u the displacements .tion, the equilibrium equation in terms of the deflexion v is, assuming
in their directions due to the set (u,). There are in all 3n equations in engineers' theory of bending to hold
a., b,, c, which are non-linear if the stress-strain relations are non-linear.
By a judicious choice of the u, v, w functions it is possible to obtain
very good approximations to the deformations of the body. The number ::2 (EI;;~)~p=O .................. .................... (61)
of necessary functions for a good estimate depends on the problem and on and the Galerkin form of the virtual work equation is
the choice of these component functions. The proof of the convergence I
to the exact solution with increasing n is a difficult question which cannot
be considered here, see Trefftz 10, p. 130. We note only, referring to the
previous paragraph, that the Rayleigh-Ritz method always over-estimates
J{;; 2( El;;~) ~p }8v·dz=O .......................... (61a)
0

the stiffness.
Whilst the Rayleigh-Ritz method can provide a good approximation
to the deformations, the accuracy of the associated stresses is, in general,
not as good. This is obvious if we remember that the accuracy of any
approximate function is decreased with every differentiation.
In two- or three-dimensional problems it is possible to improve upon
the original Rayleigh-Ritz procedure by adopting a mixed technique of
(A) and this paragraph. Thus, in a two-dimensional problem it is often
possible to guess accurately the variation of the displacements parallel to
one axis, say the x-axis, while it is much more difficult to make an in-
telligent assumption about the variation parallel to the other direction.
We may then write the displacements u and v in the form
u=u 1(x)·cp(y) }
V=V (x)·l{l(y) . . . . . . . . . . . . . . (56)
1
It is easy to see that for displacement functions (50) satisfying all boundary
conditions the Galerkin and Rayleigh-Ritz methods must yield the same
equations for a, b" c, and hence also the same deformations. We need
only realize that in this case Eqs. (57) are indeed equivalent to the principle
of virtual work. Hence substitution of u, v, w in Eqs. (57) must give the
same result as substitution into
SU=O
The advantage of Galerkin's method lies in a more direct derivation of
the equations in a., b., c,. However, contrary to what is usually assumed,
this advantage is small if we calculate SU directly. We note also that
Galerkin's method allows only such approximate functions as satisfy all
boundary conditions, while the Rayleigh-Ritz procedure requires only the
satisfaction of the kinematic boundary conditions.

where u1 , v1 are assumed crosswise distributions of the u, v displacements * Timoshenko 8 mentions on p. 159 that equations of this type appear already in W. Ritz's work.
and f, if! are unknown non-dimensional functions of y. Substituting See references on p. 353 and also Gesammelte Werke, p. 228.
Eqs. (56) into (55) one obtains after some transformations the differential t B. G. <,;alerkin, 'Series Solutions of Some Problems of Elaftic Equilibrium of Rods and Plates,'
equations in f and if! together with the necessary boundary conditions. Wjestnik lngenerow Petrograd (1915), No. 19, p. 879; see also W. J. Duncan, 'Application of the
Galerkin Method to the Torsion and Flexure of Cylinders and Prisms,' Phil. Mag., Ser. 7, Vol. 25,
Such an analysis can yield a very accurate result. It will be illustrated on 1938. p. 636.
a number of examples of some complexity in Part I I to this report.
Note that the Rayleigh-Ritz procedure as presented here is also valid for
non-linear stress-strain relations.
(F) Galerkin's method of approximate stress analysis. Consider the
internal equilibrium Eqs, (4) which we write here in the by now familiar
form z
J[Oaxx+ oayx+ oau+w ]8u ·dV=O
v
ox oy oz

J[Oayy+Oazx+oa_,"+w
v
oy oz ox y
]1)v ·dV=O
x

(57)
I~
v
fjg, 10.-Cantilever beam

9
 Consider the cantilever under transverse load shown in FIG. to. To Hence the complete virtual work Eq. (a2) becomes
obtain an approximate expression for the deflexions v by the Rayleigh- I I I
Ritz procedure we need only select a function or sequence of functions 2J {Efvlv_p}Sv·dz+2Elv"(Sv)' I 2Eiv'"(8v) I +PSV~O .... (all)
giving 0 0 0

dv But
v=az=O at z=O (Sv)z' =0, ( Sv) 0 =0 and (8v), = 8 V
However, when applying the Galerkin method Eq. (6la) must satisfy also Since otherwise 8v is arbitrary we conclude that to satisfy Eq. (all)
the static conditions at z=l, i.e. we must have
Shear force=Bending moment=O (Elv"L~o=O, 2(Elv"')zd-P=O ............................ (al2)

dav) (d2v) and v must satisfy the differential equation

or ( dz 3 z~l = dz 2 z~l =O
5. ILLUSTRATIONS OF THE METHOD OF VIRTUAL DISPLACE-
MENTS
In this section we present a number of applications of the principle of
virtual work. These are not meant to give the shortest possible solutions
to the problems considered but merely to illustrate the way in which the
method can be applied in some simple cases. More complicated problems
are investigated in Part II
(a) Continuous Beam with Non-Linear Spring Support
The uniform beam of bending stiffness EI shown in FIG. II carries a
uniformly distributed load p and is simply supported at the ends. At the
centre an additional support consists of a spring with the load displace-
ment law
P=kv[ I+ l-(~/ v.)J ............................. .....
Thus k is the initial spring stiffness and v. is the displacement at which
the spring becomes solid.
With the assumption of the Engineers' Theory of Bending, establish by
means of the Principle of Virtual Displacements the differential equation
for the deflexion v of the beam and the boundary conditions. Find also
the deflexion Vat the spring.
p Yr
z The total strain is made up of the elastic strain Er together with the
EJviv_p=O .........................•.................. (al3)
Eqs. (al2) and (al3) together with the kinematic conditions
(v)z~o=O and (v')zd=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4)
give all the necessary information for the determination of v.
Integrating Eq. (al3) and using the boundary conditions (al2) and (al4)
we find finally for V the quadratic equation
(I +!{i)(V/ V.) 2 -[l +!{!(!+a)+ V1 /Vo]( V/V.)+ V1 /Vo=O . .. . .. (al5)
where
!{i=kl 3 /6EI, V1 =Sp/4 /24
(b) Plane Redundant Framework
A plane framework consisting of a single joint connected by a number
of hinged bars to a rigid foundation is loaded by forces X and Y along the
(a I)
axes Ox, Oy respectively (FIG. 12). In addition, the bars are heated to
arbitrary temperatures and have also initial strains due to errors in
manufacture.
Find by application of the Principle of Virtual Displacements the forces
in the bars.
Let u, v be the displacements of the loaded hinge, measured from the
position for which all bars have the correct length and are at zero tem-
perature. Then the total direct strain in the rth bar for these displacements
is
u cos Br+v sin Br
fr • • • • • • • • • • · · · · · · · · · · . . . . . . . . . . . . . . . . (bl)

--4---El/J
thermal strain 7Jr=a0 and the initial strain 'Y]oro where the initial strain is
p 'Y]or=t1fr/lr · ... ·......................................... (b2)
t11r being the additional length of bar (in excess of the correct length) due
·I· to manufacturing errors or other causes. Hence the elastic strain due to
u, vis
~ Er=yr-'Y)r-'Y]or
Fig. 11.-Virtual displacements: Example (a) Simply supported beam
with non-linear spring u cos BrtV sin Br -('Y]r+'Y)orl ......................... . (b3)

The first part of the problem is, of course, trivial and the result known and the direct stress in the bar is
to any undergraduate, but we want to show here how the Virtual Displace-
ments method can be applied in such a case.
Nr_ -EEr-
Ar -Ur- -E[ucosBr+vsinBr
fr - ('Y)r+'Y)or >] ........... . (b4)
We consider virtual displacements consisting of small arbitrary addi-
tional. deflexions (Sv) of the beam from its equilibrium position. The If we now impose on the joint the virtual displacements Su, Sv there
Principle of Virtual Displacements then expresses the equilibrium con- arises an increment of strain energy S U;, and an increment of work S W
of the applied forces, where

r
dition in the form
SU1 -SW=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a2)
The virtual displacements must satisfy certain kinematic boundary con-
ditions, namely :
Sv=O for z=O and z=2/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a3)
and since both structure and loading are symmetrical we need only con-
sider symmetrical virtual displacements. Hence
d(Sv)/dz=(Sv)' =0 for z=l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a4)
Strains and stresses in beam due to bending:
E=Eu= -yv", a=a .. = -Eyv" . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a5) .. x
and therefore the virtual strain due to Sv is
SE=-yS(v")=-ji(Sv)" ... . .. . . .. . . . ... . . . . . . . . . . . . .. . .. . (a6)
The increment of strain energy in the beam due to bending is thus
I I I
2J fa·SE·dA·dz=2Ef { fy 2·dA}v"(8v)"dz=2Elfv"(8v)"dz (a7)
oA o A o
which becomes, on integrating twice by parts,
I I I
2£/{v"(Sv)' I -v"'(Sv) I +Jviv.Sv·dz} (a8)
0 0 0

The increment of strain energy in the spring is

PSV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a9)
and the increment of work done by the distributed load is Area Ar, length lr
I
SW=2fp·Sv·dz (alO) Fig. 12.-Virtual displacements: Example (b) Redundant
0 system of bars
10
 n and the work done by the distributed torque and the end torques
8U;="2:.Ararlr0yr I
r~l
8 W= Jm.88dz+ T 1(88) 1 - To(88)o ................... , ..... . (c7)
=fEArlr[u cos 8r~v sin 8, ~(7],+7]or)] [Su cos 8,t8v sin 8,] 0

r=l r Applying the Principle of Virtual Work

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (b5) I I
8U;~8W=O= J[£r8"(88)" +GJ8'(88)' -m.<88)]dz- T88 (c8)
8W=X8u+ Y8v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (b6) n
1
0
By the Principle of Virtual Displacements Integrating by parts the first term twice and the second once, we finally
8U1 ~8W=O obtain
and therefore • I I
n n n j [Er8iv ~GJ8" ~m.]88dz+[GJ8' ~Er8"' ~T]88 I +£r8"(88)' I =0
au[ u I~:r cos 8,+v I~:T sin 8,cos8,~ LEA,(7],+7]or)cos8,-x]+
0 0 0
2 (c9)
r~l r~l r~l
For the integral to vanish, since 88 is arbitrary, 8 must satisfy the dif-
n n n

sv[u 2:~:r sin 8,cos 8,+v L:Etr sin 2 8r~ IEAi7]r+7]or)sin8,- y J=0
ferential equation
Er8iv -GJ8" ~m.=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (ciO)
r~l r~l r~l

(b7) which is recognized as the usual Wagner equation differentiated with

respect to z.
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

If Eq. (b7) is to be satisfied, since Su, Sv are arbitrary, the two expres- If the twist at both ends of the tube is specified then 88 must there be
sions in brackets must separately be equal to zero and hence we have two taken zero and the first bracketed term in (c9) vanishes also. If in addition
equations in the displacements u, v. Solving for u, v the stresses and forces the warping (and hence 8') is specified (e.g. built-in end) then (88)' is also
in the bars can be calculated from Eq. (b4). zero at z=O and z=l and the remaining term also vanishes.
The two equations are, of course, the equilibrium conditions in the If, however, the end z=l is free to twist and warp, (88) 1 and (88)/ are
x andy directions and could be derived directly by statics without recourse arbitrary and we have as further conditions from (c9)
to the Principle of Virtual Displacements.

}
Note that there are always only two unknowns in this approach, regard- GJ(8')t~Er(8)'"t~ Tt=O
less of the number n of bars. Hence it is preferable to operate with the dis- and (ell)
placements u, vas unknowns than with forces in the bars when n>4. £r<8")t=o
(c) The Open Tube Under Torque which are the necessary static boundary conditions at the free end. The
first is of course the condition for equality of external and internal torque
A uniform, open tube of length I is subjected to a distributed torque and the second the condition for zero direct stress.
m. per unit length and end torques To and Tt. For the series solution, we represent the twisted shape by the Fourier
With the assumption that shear strains due to restrained warping are series
zero (Wagner assumption) establish the differential equation for the angle
of twist 8 and the static boundary conditions. Give also a series solution X

for the case when the ends z=O, z=l are prevented from twisting but are 8= '\:"" . (nTTZ)
L...Ja, Sin -1~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (cl2)
free to warp. n~l

and take for virtual displacements the increments of twist produced by a

small variation 8a.., of the coefficient a,...
Using Eq. (c8) we find

'\:"" nmTT 2 J mTTZ

n=l
X
nTTZ
8U;~8W=8a.., [ GJ L...Ja,~ cos - 1- cos ~1~dz
o
I

J
I
1

(cl3)
\.-,
X

(11mTT 2) 2 mTTZ . nTTZ J.

I
mTTZ
+Er f:;.a" f2 o s1n - 1-sm ~1~dz~ om. sm - 1-dz
J .
I
J
which gives on integrating
I

Jm. sin mTTZ

-1 dz
. . . . . . . . . . . . . . . . . . . . . . . . . . (c·l4)

Fig. 13.-Virtual displacements: Example (c) The open tube

If m. is constant
The warping displacement w of the cross-section is given by*
w=w(d8/dz)=w8' (cl)
4[2
I
I
where wis the warping per unit rate of twist. The direct strain
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

form odd

J
Ezz and . . . . . . . . . . . . (cl5)
stress a_. due to non-constant rate of twist are and
a,..=O form even
E.,=~ =w8", a_.=EE_.=Ew8" (c2)
which gives for the twist distribution
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

The torque due to the St Venant torsional shear stresses is 00

TJ=GJ8' . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c3) (nTTz/1)

8 -~ 41 2 m.~ [ sin(n7T)2£r] ........................ (cl6)
We obtain now virtual displacements by giving to the twist 8 an incre-
ment 88. The corresponding virtual strain 8E is given by GJTT3 n~l 113 I+ T GJ
odd
8E.,=w8(8")=w(88)" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c4)
In this case, since the assumed form of solution satisfies also the static
and the increment of strain energy due to the direct (torsion-bending) boundary conditions 8" =0 for z =0 and I we can alternatively use the
stresses is Galerkin form of the Virtual Work equation, which is in this case
I S I S I
J [ Ja ..SE ..tds]dz=J£8"(88)" [ fw 2 tds]dz=JEr8"(88)"dz .... (c5) I
J[Er8iv -GJ8" -m.]88dz=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . (cl7)
0 0 0 0 0

where r is the torsion-bending constant.*

0

The increment of strain energy du~ to the St Venant shear stresses is and is given immediately from Eq. (c9).
I I If we approximate our solution for 8 by retaining only the first term in
fGJ8'8(8')dz=JGJ8'(88)'dz .............................. (c6) (cl6) we underestimate the average angle of twist. This ties up with our
n ,. statements on p. 8; for, by putting 8=ao sin (TTz/1) we apply constraints
on the tube and hence overestimate the stiffness.
• See Argyris, 'The Open Tube', AIRCRAFT ENGINEERING, Vol. XXVI, No. 302, April
1954, p. 101 et seq.

11
6. THE PRINCIPLE OF VIRTUAL FORCES OR COMPLEMEN-
I
TARY VIRTUAL WORK
T is natural in reviewing the developments of Sections 3 and 4 to inquire
if it is possible to enlarge upon the conception of complementary work
and strain energy in a similar way as accomplished for work and strain
energy by the introduction of virtual displacements. In fact, if we consider
Eqs. (28) and (36) we realize immediately that the functions S W* and
S U; * are independent of the variations Su and DE associated with the force
and stress increments, just as S Wand OU; are independent of the varia-
tions oP and Sa associated with the chosen Su and DE'S. Hence, when
finding oW* and oU;* we may assume that displacements and strains
remain constant. Also, the infinitesimal increments ,Sa, ow, ocp are arbi-
trary as long as they satisfy the equilibrium conditions in the interior and,
where such are prescribed, on the surface. Thus, if we fix that the surface
forces are not to be varied over part of the boundary we must have there
ocp=O; however, where kinematic conditions are prescribed on the
boundary the Scp variation cannot be assigned. It is apparent that our
incremental stress system need not even be an elastically compatible one.
It is only restricted by the condition that it must be statically equivalent
to the load increments ow and Scp. While these increments are applied it is
assumed as in Section 4 that the temperature remains constant.
Such infinitesimal variations of forces and stresses which are arbitrary
as long as they satisfy the prescribed equilibrium conditions we call virtual
forces and stresses.
Before restating theorem (35) for the more general conceptions intro-
duced here let us consider again its derivation in the light of our new ideas.
Thus, if we multiply the true displacements u, v, w by the internal equili-
brium conditions (29) which the virtual stresses Sa and ow must satisfy
sum, integrate over the body, apply Green's Theorem and note th~
boundary conditions (30) we obtain Eq. (35) where y,,. etc. are the total
strains associated with the displacements u, v, w. Next let us apply Green's
Theorem in the opposite direction by starting from the right-hand side
of Eq. (35). We find that this function can only be equal to the left-hand
side if the terms Yxx etc. are indeed the expressions for the strains (3) and
( 12) and satisfy the kinematic boundary conditions.
Thus, we conclude that an elastic body is in an elastically compatible
state under a given system of forces and temperature distribution if for any
virtual increments of forces and stresses from a position of equilibrium
oW*=oU;*+ f'YJOsdV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (62)
where v Also since the Engineers' theory of bending applies (y 11 ,=0)
oU;*=fEoa·dVand 7JOS=7J(0.7n+oa""+oa,z) . . . . . . . . . . . . . . (63)
v
See also Eqs. (35) and (38a).
Eq. (62) is, in fact, a necessary and sufficient condition for elastic com-
patibility of the equilibrium.
Theorem (62) we call the principle of virtual forces or virtual comple-
mentary work for elastic bodies subjected to loads and temperature dis-
tribution. Note that (62) applies for non-linear stress-strain laws.
The above discussion indicates that there is a close parallel between the
14.-Example of an arbitrary virtual force
12
p~inciples of virtual displacements and virtual forces. Thus, by substituting
vtrtual.forces (stresses) for virtual elastic displacements (strains), actual
total displacements (strains) for forces (stresses), and invariant state of
straining for. invari~nt _state of forces we obtain Eq. (62) from Eq. (44).
However, th1s duahty IS only complete for continuous structures which
are infinitely redundant. If, on the other hand, we consider a statically
determinate structure we find that while it is still possible to describe an
infinite set of virtual displacements O!t associated with a prescribed set of
certain of the displacements, only one stress system can exist for given
external forces; hence no oa can be assigned in the latter case and the
principle of virtual forces has no application. A more fundamental
limitation of the principle of virtual forces appears if we want to extend
?ur th~orems to finite displacements. Here we find that it is, in general,
tmposs1ble to achieve this for the principle of virtual forces while as
mentioned in Section 4, no basic difficulty arises in the case of the prin-
ciple of virtual displacements. However, for the usual analyses of redundant
systems involving small displacement theory the principle of virtual forces
with its many particular forms is t:1e most useful one since the standard
proce~ure !ntrodu~es forces as unknowns. Naturally, there are many cases,
especially m multi-redundant structures, where it is advantageous to in-
troduce displacements as unknowns; here the principle of virtual displace-
ments is the indicated method as shown in Example (b) of Section 5.
. We return now to _Eq. (62) and will illustrate its validity on a very
s1mple example. Constder to that effect the redundant beam of uniform
flexural_ stiffness £/built-in at z =0, simply supported at z ~,I, and subjected
to a umform load p (see FIG. 14). Under the assumption that the ordinary
engineers' theory of bending holds and that the shear deflexions are
negligible the deflexion v is given by
v=fs~/1) T3 -s(l) +2(7) 2]
As a virtual force we select oP at the centre of the beam as shown in
FIG. 14(a). However, since we require only a statically equivalent stress
system to equilibrate the applied virtual load we may eliminate the one
redundancy and select a statically determinate beam. The two alternative
choices leading to a simply supported beam and a cantilever are seen in
FIGS. 14(b) and 14(c). Denoting the true deflexion at the centre v1 we have
pf4
oW*=oP· v1 =3P· 192 £}
I I
oU;*= J[j E:,Da,,dA }z=- J~:~oMdz
o A n
where the integral in the square bracket refers to the integration over the
cross-section. For the case shown in FIG. 14(c).
oM= -oPG-z) for o~~z~t/2
oM=O for f/2<z<l
Also since,
~;~=:~~ [I -sf +4GYJ
l/2
DU;*=oP:~I J(£-z)[ I -s] +4(TY}z,=oP 1 :~~/ =oW* q.e.d.
0
If we apply now a force -- fooP at the free end our virtual system (c) is trans-
formed into (a). No additional oW* arises since v=O at z=l. The additional
bending moment 3M produced by - ~1 3P is
f6 ( I-l)oP.I
and this is easily found not to create an additional 3 U; *. By relaxing the
moment restraint at A we may finally prove without difficulty that
OU;*=oW* applies also for the virtual system (b).
We return now to Eq. (62) and note that since the displacements are
assumed constant when the virtual forces are applied we may regard 3 W*
as the variation of a potential - U, * where
U, * = - f[uw,+vw 11 +ww,]dV- f[ucp,+vcp,,+wcp,]dS . . . . . . . . (64)
v s
Thu~, SW* = -- oU, * and U, * may be termed the complementary
potent1al of the external forces. Note, however that W* is not - U *
since !n obtaining W* from S W* we must, naturally, perform the i~­
tegratton for displacements varying with load. In fact, for a linear system
and no temperature effects W* = - U, * /2; compare also Eqs. (64) and (46).
Also, since the thermal strains are kept constant we may write the right-
hand side of (62) as
OU;*+friosdV=o(U;*+frr~dV) ~ oU,* (65)
v v
where
8U,*~Jy8adV ........................................ (65a)
v
but
U,* U1* -t- frysdV f [ JEOa]dV f frJ.w/V (66)
v v 0 v
since <1 =const. U,* we term the complementary potential energy of total
deformation. Note that it is always simpler to calculate directly oU,*
from Eq. (65). Particular care is necessary in evaluating U,* for as Eq. (66)
shows in the first integral E is taken to vary with a from the initial to the
final state while 7), s in the second integral refer only to the final values.
Physically speaking we may consider U>~* as the· complementary work
necessary to reach the final true state of deformation from an initial state
in which we allowed free thermal expansion and destroyed compatibility.
Formulae (65) and (66) may be extended immediately to the case of
arbitrary initial straining by substitu1ing
7Ja -T),.,.a,.,.-t- ...... +7) .•.•,a_,., I .... I 7),...a,... for 7JS
Eq. (62) can now be written more concisely
o,U*~o.,(U,*tU,*J 0 ..................................
where the suffix a indicates that only forces and stresses are varied and
U*=U.,*-t-U,* (67a)
is defined as the total complementary potential energy of the system.
Eq. (67) states that a state of equilibrium of an elastic body is also one of
elastic compatibility (i.e. the body is at the true position of equilibrium)
if any virtual variation of the stresses and forces, while displacements
remain constant, does not give rise to any (first order) variation of the total
complementary potential energy. This theorem we call the principle of a
stationary value of total complementary potential energy if the latter is
expressed in terms of forces and stresses. Actually the stationary value of
U* is a minimum as we may prove without difficulty.* This point is dis-
cussed in more detail under (C) below.
Eqs. (62) or (67) may be used to derive the results which follow.
(A). The differential equations of the theory of elasticity (for arbitrary
loading and temperature distribution), or any particular structural
problem, in terms of stresses or stress-resultants; the appropriate
kinematic conditions in terms of forces and stresses follow also from this
analysis. It is important to note that in all applications it is best to form
directly
oU,*=f[yoa]dV
v unique equilibrium position which is also truly compatible. This may be
and not to evaluate first U.,* and then to take its increment oU,*.
(B). Castigliano'st Theorem Part II generalized for Thermal Effects and
non-Unear elasticity
[ (!Ud*]
(JP =u, ..................................... .
r B=const.
where u, is the deflexion (rotation) in the direction of the force (moment)
P,. This relation may be obtained immediately if we apply one virtual
external force oP, in the direction of the displacement u,.
(C). The Principle of Stationary (Minimum) value of Complementary
potential energy of total deformation for internally redundant
structures
This may be derived from (62) if we do not apply any external virtual
forces; i.e.
8c/>~=oc/>,=oc/>z=owx=ow 11 =~owz=O
while varying the stresses a.
Then
8.,(U,*)=0 (69)
which is our generalization of the standard principle of Castigliano of
Minimum Strain Energy to include temperature effects and non-linear
stress-strain laws. Note again that Principle (69) necessarily applies only to
internally redundant structures, since for given external loads only one
stress distribution can exist in statically determinate structures.
Eq. (69) itself only indicates that Ud* has a stationary value in that
particular state of equilibrium in which all the elastic and kinematic com-
patibility conditions are satisfied. Note, as mentioned before, that with the
limitations of the present assumptions, i.e. small displacements and
monotonically increasing stress-strain diagram, there is only one position
possible where both the equilibrium and compatibility conditions are.
satisfied. We now investigate the nature of the extremum of Ud*, which
naturally requires the consideration of second order terms as in Section 4.
Consider an elastic body under given loads and temperature distribution
in its compatible equilibrium position. We make a series of cuts in the
body but at the same time apply stresses a c acting across and along the
cuts of the same magnitude as in the uncut body; these are obviously the
stresses required to maintain the compatibility condition of perfect fit at
• See Biezeno and Grammel', p. 75.
t A. Castigliano, Theorie de /'equilibre des systjmes elastiques, Turin 1879.
the cuts. If we impose the vii-tual stresses oa, it is apparent that since
these produce corresponding deformations ou,. on the cut faces the latter
are not any longer compatible. It is important to realize that the oar
systems are self-equilibrating since the external loads P remain constant.
Thus, in a framework we may obtain a system oa,. by cutting a redundant
bar and applying a variation oN to the true force N in the bar.
We now investigate the differences in complementary work W* and
U>~* between the original equilibrium position of the uncut body and the
new enforced equilibrium position of the cut body. Comparing Eqs. (38)
and (66) we find
1
W* = U.t*- f<TJs- fTJOs)dV (70)
v 0
In moving from the uncut (compatible) equilibrium state to the cut one we
note first that the integral does not vary since 7) is constant in this step.
Also the first order increment, oUd*, of Ut~* is zero since this is the con-
dition for compatible equilibrium of the original body and no first order
(67) increment 8 W* can arise since the loads P remain constant. We are then
left only with second order increments.
For the complementary work this is
o 2 WL1-fou,·oac·dS ... ................................. (70a)
c
(where the integral is taken over the cut faces) which is the work of the
virtual stresses oa c over the displacements ou c they themselves produce.
This is clearly positive. The second order increment of U,* is merely
o 2 U,* =Hoa· oE·dV . . . . . . . . . . . . . . . . . . . . . .. . . . . . .. . . . . . . (70b)
v
since 7) remains constant; oa and OE are the stresses and strains due to oa ,.
Terms (70a) and (70b) are equal and both positive.
We conclude that the complementary potential energy of total deforma-
tion U,* and the complementary work W* have for given forces and tem-
perature distribution a minimum at that position of equilibrium of the uncut
body at which compatibility is satisfied.
It follows that if U.t* is overestimated by assuming a statically equivalent
stress system which does not satisfy all compatibility conditions and we,
ignoring the latter fact, equate U<~* to W* of the applied loads Palone
we cannot but overestimate the magnitude of the displacement system
under the loads P. Conversely to achieve a given displacement system our
calculations based on a non-compatible stress system must underestimate
the corresponding load system P. Thus, the latter has its maximum for the
expressed also as follows :
For given displacements and temperature distribution the complementary
potential energy of total deformation has a maximum when the state of
equilibrium satisfies also the compatibility conditions.
The above theorems may be combined to give:
The stiffness of an elastic body in which the equilibrium conditions are
(68) satisfied is a maximum when the elastic compatibility conditions are all met.
Thus we see that the effect of introducing assumed forms of stress dis-
tribution for the purpose of approximate solutions is the opposite to that
of the method of Virtual Displacements and therefore application of both
methods to a given problem yields upper and lower bounds to such
aggregate quantities as stiffness. No general conclusion as to bounds can,
of course, be drawn for the details of the stress distribution.
The above theorems which apply also in the presence of initial strains
other than those due to temperature do not appear to have been given
before with this degree of generality.
(D). The Unit Load Method
Assume that we require the deformation (deflexion or slope) u, at a
given point and direction of an elastic redundant body subjected to given
forces and thermal effects. Let the actual total strains in the structure be
known and given by*
Yx~=Eu+7), Yx·y=Exy
Applying a load (force or moment), oP, in the direction of u, and using
Eq. (62) we find
oP,·u,=JyoadV
v
=HYxxOa..,..,+ ..... +Yx110a,.+ .. +yuoa.,]dV (71)
v
where Oaxx . ... oa~•. ... are the virtual stresses due to oP T• In a linearly
elastic system Sax~ etc., are proportional to oPT and Eq. (69) can be written
l·u,= J[y,.p,x-1-YY·P•v+YzPzz+Yx•a,,+y,.a•• +yz,azr.]dV .... · . (71a)
v
where au . ... a,•.... are the stresses due to a unit load. Since ii,.,.. ...
ax• . . . . need only satisfy the internal equilibrium conditions and the
external one for oP,= 1 it is obviously advantageous to determine
ax., . ... iix·u . ... in the most simple statically determinate basic system.
For a non-linear system Eq. (7la) is still applicable as long as etc. a,,.
are calculated in a statically determinate basic system. For only in the
• Naturally. this method may .also b.> applied in the case of an initial strain system with shear strains.
13
 Normal

A
: B
'
C
.
(l,m)

v
A

/\1\1\-- I
-,~r,-
I \1
\1
-,..,\--
'1
',
,..,..~
\1\
- -;',
'' \ y,vL B
~~~--a--~2~--~~--~--~- ~- ~
A Unit load 8 ~ x.u
15.-Unit load method for displacement of redundant framework 16.-Two-dimensional stress case, static and kinematic boundary
conditions

latter case w1ll

(74)
Sa.,.,j'8P,=oa.,.,/oP,
be the stresses corresponding to a unit load. F must, of course, also satisfy the given boundary conditions. In fact,
The Unit Load Method is the most suitable tool in the calculation of where surface forces are prescribed we have (see Eqs. (5) and FIG. 16)
structures with a finite number of redundancies expressed as stresses or o (oF oF) I
stress resultants.
This will be shown in some detail in Section 8.
oy 1oy -mox =f~
J ............................
I
(75)
Example of the application of Eq. (71a).
Consider the plane framework with a redundant support as shown in An approximate method may proceed as follows. Assume that the stress
FIG. 15(a). We seek the deflexion vat joint 2 for the loading case shown. function F is expressed in the form of a finite series
Let the actual elongation of the members due to loads P 1 •••• P 5, tem- n
perature and manufacturing errors be denoted by D./. Next we apply a F=Fo+'J:,b,F, (76)
unit load at 2 in the direction of v and find the forces iv in the bars. Since r= I
we need only consider a statically determinate case we ignore the support where F0 , F1 , • • • • • F, are known functions of which F0 satisfies the static
at C and are left with the very simple problem of finding N in the left-hand boundary conditions (75) where these are prescribed and the functions
span only. Application of Eq. (71a) yields the simple formula F, vanish there. b1 to b. are constants to be determined by the virtual
I·v='J:,ND.I ............................................ (71b) forces principle. The system (76) satisfies by definition all given equili-
brium conditions and the increment
where the summation extends only over the ccmtinuously drawn bars. The
formula given is due essentially to Maxwell* and Mohrt who applied it to
OF=Ob,·F, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (77)
statically determinate frameworks. Actually Mohr derived this type of may be regarded as a virtual stress system corresponding to zero incre-
equation by using the principle of virtual displacements with the actual ments of external loads where the latter are fixed.
elongations taken as virtual ones and the unit load system as the actual Since SW*=O (either forces are given or displacements are zero) the
one. Although such a procedure is in the present case of small displace- principle of minimum complementary potential enengy of total deforma-
ments permissible, it should nevertheless be avoided since Eq. (71b) tion is applicable here and takes the form
follows more naturally from the principle of virtual forces. SUa*= J[y.,.,Sa.,.,+y.vSa•• +y.,.Saxv]dA =0 . . . . . . . . . . . . . . . . . . (78)
A
(E). Approximate method of stress analysis We use here J( ... .)dA to denote the integration ff( ....•. )dxdyover"
A
Consider an elastic bodyt subjected to external loads (body and surface
forces) and a temperature distribution 0. The boundary conditions are the area of the two dimensional continuum.
assumed to be both of the static and kinematic kind; however, where the Substituting Sa.,., etc. in terms of (77) we find
latter are prescribed they are taken to be of the rigid kind, e.g. rigidly
built-in or sliding in a rigid groove (see FIG. 16). To limit the present
analysis we restrict our investigation to a state of plane stress. The solution A
J[ 02F, 02F, 02F,]
Yu 0y2 +Yw 0x2 -yxvoxoy Sb4A=O .................. (79)
of such problems is often expedited by the use of the Airy stress function F.
Then, the stresses are given by§ and since Sb. is arbitrary.
X

a.,.,=~~- Jwxdx, a.,v=- 0~~· avv=~~- JwvdY............

y

(72)
J[ o2F, o2F, o2F,J
Y"'"oy2 +Y••ox2 -y""oxoy dA =0 . . . . . . . . . . . . . . . . . . . . . . (79a)
A

which satisfy automatically the internal equilibrium conditions (4).
Eliminating the displacements from the strain expressions (I) we obtain
the compatibility condition for the strains, --
o~+o~o~-0
oy 2 ox 2 - oxoy- ................................. .
where Yxx etc. are the total strains E.,.,+'l') etc. For a given stress-strain law
we cart express y.,., in terms of the stresses (72) and the temperature 0.
Hence by substituting into (73) we obtain the differential equation in the
unknown F, which will, in general, be non-linear. However, in the case
of bodies obeying Hooke's law we obtain for a=const. the simple linear
result
• .f. C. Maxwell, Phil. Mag., vol. 27, p. 294, 1864
t 0. Mohr, Zeit. Ar<-hitek. u. /ng. Ver. Hannover, 1874, p. 509; 1875, p. 17.
t The presentation is restricted to singly connected domains.
§ See Timosbenko 8, p. 26.
If the total strains are expressed in terms of the stresses (72) and tem-
perature distribution 0 we obtain from (79a) n equations for the un-
knowns b1 to bn. These are only linear if the body follows a linear stress
strain law. In the latter case and for constant body forces Eq. (74) shows
(73)
that the solution must be independent of the Poisson's ratio v if all
boundary conditions are of the static type. Hence we may take v=O and
(79a) becomes
J{[o2F o2F, o2F o2F, o2F o2F,] [o2F, o2F•]}
ox 2 oy 2 + oy ox 2 + 2oxoy oxoy + E7J ox 2 + oy 2
2 dA =O
A
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (79b)
from which we may obtain without difficulty the n linear equations for
b1 to b,..
The case when all prescribed boundary conditions are static is interesting
for all surface conditions are then exactly satisfied by (76). This indicates
that it should be possible to express (79a) in a form similar to that given
as Galerkin's method under the virtual displacement principle. In fact, if
we integrate (79a) twice by parts, or better, if we apply Green's Theorem

14
 .(Area 8
p
T..,_ . . (SOc)

b lO~q c1z where Yt and Yu are the extreme (boundary) values of y corresponding to

'
Areaw;, ..1!.:_ the same x (see FIG. 16). Thus, the coefficients of the homogeneous part
.__, 1-.P. of (SOc) are only constants in the case of a rectangular field.

_t__ ~-ckness t IOtq il+~·l:1

This method can yield very accurate results and is actually the one
adopted in Part II.

7. ILLUSTRATIONS OF THE PRINCIPLE OF VIRTUAL FORCES

p ~I ~z- In this section we present a number of applications of the principle of

virtual forces to quite simple problems. Again, it is not necessarily sug-
gested that the method is the most suitable one for the problems con-
sidered. It is only intended to show how it can be applied in these simple
17.-Virtual Forces: Example (a) Diffusion problem cases. In subsequent parts of the paper some rather more complex pro-
blems will be dealt with. In all the examples of this section linear
elasticity is assumed.
(a) Diffusion Problem
The panel shown in FIG. 17 is subjected to loads P applied at the free
ends of the edge members. Assuming that the sheet carries only shear
stress which is constant across the width b of each half (usual diffusion
+J[1oy•• +moy= -!(lOYx• +moy.,•)JF:ds
r assumption) obtain by application of the principle of virtual forces the
ox oy oy ox differential equation for the load P. in the central stringer. Find also the
c
displacement w of the free end of the stringer.
+.f[( my.,,-2y.,.
I )oFr m )oFrJ From the equilibrium of an element of the stringer, we find for the
oy +( ly,,-pxu ox ds shear flow (q ~ a .,f) in the sheet
c
Ic<- ... .)ds
denotes the line integral along the boundary of the two- q=-t~·'~-!P/ . .. .. .. .. ..... .. . .. .. .. ..... ...... .. .. (al)
dimensional continuum.
But, on the boundary Fr =0 and also and from the equilibrium of the free end of the panel we find for the load
Pn in the edge members
~~ =0 and ; ; =0 Pn=•P- ~" ............................................ (a2)
(see Eqs. (75)).
Hence Eq. (79a) reduces to the slightly simpler form, For the virtual forces we consider a variation 8P. in the stringer load.

J[o2y,., o2y•• o2y.,.J

oy2 + ox2 -oxoy FrdA =0 for r= I to n ............ (SO)
The applied forces P are maintained constant and hence to satisfy the
equilibrium conditions on the free end (P,=O) we must take 8P. to be
zero there
A
which ·shows clearly how the method of virtual forces satisfies in the (oP,),d=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
i.e. (a3)
average the compatibility condition (73). When the body is linearly elastic
Eq. (SO) may be written as (a=const.) Otherwise the variation 8P, is arbitrary.
The virtual shear flow in the sheet is thus

J[(oox4+oy4+2ox2oy2
F o4F o4F )
4 (o20 o20)
+Ea ox2 +oy2 and the virtual load in the edge members
(a4)

A
ow., ow•)] FrdA =0 for r= I ton
+(1-v) ( ox+ oy . . . . . . (SOa) 8Pn=- 8;• ........................................... . (a5)
Since the applied forces are not varied, the virtual forces principle (Eq. (62)
the expression in the bracket being of course Eq. (74). Note again the in- for 0 =0) reduces to
dependence of the solution from v when body forces are constant.
The above application of the principle of virtual forces is a generaliza- 8U;*=0
tion of a method developed by Timoshenko 8 , p. 167. A thermal stress The virtual complementary energy due to 8P. is
example of the above analysis is given in Part H. I
Naturally, the method can be extended to three-dimensional cases.
W. Ritz* proposed as early as I 90S a similar procedure for the solution DU;*= J[%,4oP,+2;;oPB+2t1b8q]dz (a6)
of St. Venant's torsion problem; this method is illustrated on an example 0
of considerable complexity in Part II. q, 8q in terms of P., 8P. and integrating the last
A slightly more refined approach than that shown above may be Substituting for PB, 8PB,
adopted when it is possible to estimate accurately the variation f of F term by parts we find
I I
parallel to one co-ordinate say y while the distribution parallel to the other
co-ordinate is more difficult to guess. Then, we may set oU;*= J[i(1+ 2~)- 2 ~ 1 d;~·-~:B]oP.·dz+ch[~·oP.]=o
F=f(y)·cp(x) . . . .. . . .. . .. . .. . . . . .. . . . . . . .. . . . . . .. . .. . .. . . (Sl) 0 0
(a7)
where cp(x) is an unknown function of x. It is, of course, possible to formu- and therefore, since 8P. is arbitrary, we must have
late the analysis in any other suitable co-ordinate system. Substituting (S1),
with oF=f8cp in place of F., into (79a) or (SO) (or related expressions),
we obtain after some simple integrations the differential equation in cp; d;~·-~:(1+2~)P.+~~P=O ....................... . (aS)
when there are also kinematic boundary conditions the corresponding or
boundary expressions for cp follow also from (79a).
Consider, for example, the case of linear elasticity, zero body forces and d2P. 2 - 2
pure static boundary conditions. Eq. (SOa) takes here the simple form, dz2 -fL P.- -fL Pso · · · · · · · · ·. . . . . . . . . . . . . . . . . . . . . . . . . . . (a9)

J[Ll 2F+EaLl0]JocpdA=O ................................ (SOb) where

A 2 2Gt(l 1) 2A . . . . . . . . . . . . . . . . . . . . . . . . (aiO)
Using (SI) in (SOb), integrating with respect to y and noting that (SOb) fL = Eb A+2B • P.a= A+2J!'
must be true for any virtual variation ocp, we obtain the differential which is the required differential equation in P•.
equation in cp Since 8P. is zero fop z=l, the remaining term in Eq. (a7) vanishes for
the upper limit z=l. At the lower limit z=O, however, oP, is arbitrary
• W. Ritz, J. reine angew. Math., vol. 135, 1908, and Ann. d. Physik, series 4, vol. 28, p. 737,
and hence
1909.
15
 - a••

:1
2
""q0 faa.,'~
7 ' ';
~

- -1
A

~ E00 dbcos~
qo1
~ b ~~~,,,-' ;r~
Am (a)
1 -----::1
............. ...../t
- ~
bb (\..,+ 1

l ~JOt o.x
'I Qry I

3 0,.. --- q..y I

4
,-t/ /·'
...-..,- -- ..L
- -I - , y O.r +ayy ,~ aa. ,'
~ It""'"'~
' - -- l~:1
/

~I ~~~'~"" ~
I I
$~ 1s (b)

--
\ I
I
----s -- v
J
l-
b.Ytt:
-1.
18.-VIrtual Forces: Example (b) Unit load method for twist of
multicell tube X
19.-Virtual Forces: Example (c) Stresses and strains for oblique
coordinates

easily from statics

(a II)
GXJ' :_:_-:Gaa sin 8

which with the equilibrium conditions on the free end,

P,=O for z=l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al2) (cl)
gives the necessary boundary conditions. u •• =(ubb+CTaa cos 2 0+2uab cos 0)/sin ()
The differential equation (a9) is, of course, alternatively derived by a The rectilinear strains are
direct consideration of the deformations. In such case the kinematic
boundary condition (all) appears as the condition of zero shear strain I I 2(1 +v)
E.1·.r= E(a.~.,. --va 1,_11 ), €,,,,~-= E(a 1111 ·-van.), E,~..!J =-E-- CTxv (c2)
at the built-in end.
·Solution of (a9) with the boundary conditions (all) and (al2) gives and hence the virtual complementary energy per unit thickness of the
element dxdy is
P,=P•• [.l co~~~(~/z)J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al3) DU/• =dxdy[c rxDCTu+EvvDCTuu +ExyDCTry]
dxdy
To determine the displacement w of the free end of the stringer we apply = E [unDCTu +u•• DCTyy -v(uu8u.,+uuuDCTu)+2(l +v)CTxyDCT~•]
there a unit force and since we need only consider a statically determinate
system we assume the stringer alone loaded by the unit force. We find . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c3)
then (see Eq. (71)) Substituting for CTxx• u ••• u,. from (cl) we can now express 3U;* in terms
I_ of the oblique stresses and virtual stresses. Thus for the virtual stress
I·w= Ju ..E..Adz 8ubb we find for the virtual complementary energy
0

and since
* dadb ,\
DU; = E sin y[ubb- CTaa + 2uab cos 0]8ubb .. .•.. . . .. .. . .. .. . (c4)
a..A=unit load= I, Ezz=P./EA where
I A=v sin 2 0 -cos 2 0 (c5)
J
w= J,4dz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4) From FIG. (19) the virtual complementary work of 8ubb is seen to be
0 8W*=da8ub!,.Ebbdb .. .. .. .. .. . • .. .. .. .. .. .. .. .. .. .. .. .. .. (c6)
which is, of cour3e, merely the extension of the stringer under the varying and therefore from the Virtual forces principle
end load P,.
8W*=SU;* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c7)
we find, using Eqs. (c4) and (c6) in (c7)
(b) Rate of Twist of Multi-cell Tube
l
In the uniform thin-walled tube whose cross-section is shown in Ebb= £sin o[ubb-Au,..+2uab cos{)] 0 0 0 0 0 0 0 0 0 0 (c8)
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

FIG. 18(a), q. is the known shear flow distribution in the walls of the tube
due to a given loading. Using the unit load method, find the rate of Applying a virtual stress 8uaa in the same way we obtain for the strain
twist dO /dz of the tube. Eaa the corresponding expression
We consider unit length of the tube and apply a unit torque T= I. I
Since we only need consider a statically determinate system for the unit Eaac= Esin o[CTaa-ACTbb+2uab cos{)] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 .. 0 0 (c9)
load stresses we select the single cell (I, 2, 3, 4) shown by full lines in
FIG. 18(b). The unit torque gives then merely a constant shear flow Consider now the virtual shear stress 8uab. From Eq. (c3) we find for the
virtual complementary strain energy due to 8uab
I
ql =2Am . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (bl) da·db .
8U;*= £sin o{2[(l +v) sm 2 0+2 cos 2 0]uab+2(u•• +ubb) cos 0}8uab
around the single cell and the rate of twist is given immediately as . . . . . . . . . . . . . . . . . . . . . . . . (clO)
~~ = J~tq 1ds= 2 ~J 'fitds ................................ (b2) Calculating the complementary work of the virtual shear stress uab
we find (see FIG. 19)
c c
where the integral is obviously taken only around the single cell (1, 2, 3, 4). 8W*=DCTabkaa cos 0+Ebb cos 0+Eab sin {)]da·db (ell) 0 0 0 0 0 0 0 0 0 0 0 0

Eq. (b2) is, of course, a well-known result in the theory of closed thin- Thus the virtual shear stress 8uab does work not only due to the shear
walled tubes. strain Eab, but also due to the strains Eaa and Ebb.
Substituting from Eqs. (c8) and (c9) for the strains E00 , Ebb and equating
(c) Plan. Stress-Strain Relations for Oblique Co-ordinates
3U;* and 8W* of Eqs. (ciO) and (ell) respectively we finally obtain for
the shear strain:
In a uniform isotropic plate, the stresses u •• , ubb, uab=CTba are referred to
the oblique co-ordinates Oa, Ob (FIG. 19). Using the principle of virtual 2(1 +v)[ cos () J
Eab=-E-- uab+ 2 sin2{)(uaa+ubb) ...................... (cl2)
forces and assuming the stress-strain relations for rectilinear stresses, find
expressions for the strains Eaa, Ebb and Eab in terms of the stresses. Note that with the strains defined as above the increment of com-
The oblique strains Eaa, Ebb, Eab are defined as the elongations in the direc- plementary energy is
tions Oa and Ob and the decrease in the angle () respectively of the unit
parallelogram (see FIG. 19). 8U;*={E00 DCTaa+Eb1JDubb+[(Eaa+Eb1J) cos 0+Eab sin 0]8uab}dadb . . (cl3)
For the stresses CTxx• u ••• CTxv equivalent to the oblique stresses, we find as compared with the simple result for rectilinear axes in Eq. (c3).
16
8. METHODS OF ANALYSIS OF STRUCTURES WITH A FINITE
T
NUMBER OF REDUNDANCIES
HE general theorems given in Sections 4 and 6 include, fro~ the
fundamental point of view, all that is required for the analysis of
redundant structures. However, to facilitate practical calculations it
is helpful to develop more explicit methods and formulae. To find these
is the purpose of this Section.
A structure is by common definition redundant if there are not sufficient
conditions of equilibrium to obtain all internal forces (stresses or stress-
resultants) and reactions; the number of redundancies is the difference
between the number of unknown forces (or stresses) and the number of
independent equilibrium conditions. Strictly all actual structures are in-
finitely redundant but for practical purposes it is, in general,. ne~ss~ry
and justified to simplify and idealize the structure and/or stress d1stnbut10n
in order to obtain a system with a finite (or even zero) number of redun-
dancies. Such typical processes of simplification are, for example, the
assumption of pin-joints in frameworks and the assumption of the engin-
eers' theory of bending in the analysis of beams. Note, moreover, that the
Rayleigh-Ritz procedure discussed in Section 6F amounts also, in fac_t,
to the substitution of a finitely redundant structure for the actual elastic
body.
All our considerations in this Section are restricted to linearly elastic
bodies but Example 2 in Part II shows how the present methods may be
extended to the analysis of non-linear redundant structures.
It is curious to note that, while the solution of problems in the theory of
elasticity is derived very often from the differential equations in the dis-
Additional Notation
R, Q, p single, generalized, orthogonal force (moment)
RQP corresponding column matrices
r, q, p single, generalized, orthogonal displacement (rotation)
r q p corresponding column matrices
(true) stress and virtual (statically equivalent) stress due to
unit load at i
- external loads and initial strains
corresponding column matrices
si, si
E; strain due to unit load at i
e; corresponding column matrix
/;;. /;h direct and cross-flexibility
F matrix of flexibilities /;A
B transformation matrix for forces
F., fv generalized and orthogonal flexibility matrices
b rectangular transformation matrix for internal forces (stresses)
5 column matrix of internal forces (stresses)
v column matrix of strains
'· flexibility matrix of g element
f flexibility matrix of all elements
4> flexibility of element of unit length
true and virtual strain due to unit displacement at i
stress due to unit displacement i
ku, k 1h direct and cross-stiffnesses
K matrix of stiffnesses k;h
A transformation matrix for displacements
K., K., generalized and orthogonal stiffness matrices
a rectangular transformation matrix for strains
placements, the stress-deformation analysis of engineering structures was,
until a few years ago, generally based on the concept of force-redundancies.
Interestingly enough, Navier, *who was the first to evolve a general method
for the analysis of redundant systems, when investigating problem (b) in
Section 5 used also the displacement method. The analysis of indetermin-
ate structures on the basis of redundant forces goes back to Clerk Maxwell'
and Otto Mohr 5 and was ultimately developed by Mueller-Breslau. 6 t This
technique is, as mentioned in the introduction, more concise and physically
more illuminating than the Castigliano approach; it derives most naturally
from the unit load method.(see Section 60, Eq. (7la)). Mueller-Breslau's
technique is generalized here and presented also in matrix form. The effect
of temperature or other initial strains is included ab initio.
Parallel to the rapid development of the force-redundant theory occa-
sional practical problems were solved by selecting deformations as un-
knowns. Fundamentally this method is equivalent to the virtual displace-
ment analysis given in Section 4. Mohrt was probably the first to use such
an approach in engineering structures when finding the secondary bending
stresses in frameworks of the type usually assumed to be pin-jointed.
• C. L. Naviec. Resume des lecons sur /'application de Ia mecanique a l'etablissement des construc-
tions et des machines. Paris 1826, 3 ed. par B. de St.-Venant 2 vols. Paris 1864. See also: A. Clebsch,
Theorie der Elasticitaet [ester Koerper, Leipzig, 1862; French edition by B. de St. Venant: A. Clebsch,
Theorie de /'elasticite des corps so/ides. avec de notes etendues de B. de St. Venant etA. Flamant, Paris,
1883. W. Thomson and D. G. Tait, Treatise on natural philosophy; I ed. Oxford, 1867.
t See also H. Mueller-Breslau, Die graphische Statik der Baukonstruktionen, 1 ed., Koerner,
Leipzig, 1886.
~ 0. Mohr, Zivilingenieur, Vol. 38, p. 577, 1892; see also, Abhandlungen aus dem Gebiete der
technischen Mechanik, 2 ed., Berlin, 1914, p. 407.
k. stiffness matrix of g element
k stiffness matrix of all elements
K stiffness of element of unit length
ao stress system of basic structure
a. self-equilibrating stress systems
X;, Y;, Z; redundant force (moment)
S;0 , { ;0 relative displacement at cut i-redundancy in basic system due to
'ik
siko influence (flexibility) coefficients of basic system for the direc-
tions of redundant forces
D matrix of S;k
D. column matrix of S; 0
T triangular matrix
M elimination matrix
~ prescribed relative displacement (linear or angular) either in-
side the structure (e.g. lack of fit) or at the supports ('give'
of foundations)
C; force or moment in the basic system due to X;= 1 acting on an
element which experiences a ~ in the direction of this ~
N (n), S (s), M (m) normal force, shear force, bending moment
0, o (rectangular) zero matrix
unit matrix
1 .... j ..... h ..... m direction of external forces
1 . . . . . i . . . . . k . . . . . n redundancies
a, b, ..... g, ..... .s elements of structure
A', A- 1 transpose and inverted (reciprocal) matrix of A
{ . . . . . } column matrix
17
Following Mohr's <:~nalysis his ideas were applied to stiff-jointed frame-
works, the first systematic work being that of the Danish engineer Axel
Bendixen.* However, the great potentialities of the method were only dis-
covered by Ostenfeld,t a compatriot of Bendixen. He was the first to point
out the duality of the force and displacement ap;:-roach. In fact, his equa-
tions for the unknown displacements in a structure complement Mueller-
Breslau's equations for the redundant forces. It is regrettable that Timo-
shenko in his fascinating History 9 does not mention Ostenfeld's classical
book. We give here a considerable generalization of Ostenfeld's ideas to
include any structures under any load and temperature distribution.
The 'slope-deflexion' equations of Bendixen form the basis of the method
of successive approximation due to Calisevt and developed by Hardy-
Cross!! as the well-known moment distribution method. The technique
used is essentially a particular example of the relaxation method of
Southwell§ which has been successfully applied to a wide range of prob-
lems. In its application to elasticity and structural problems this latter
method is particularly representative of the modern tendency in making
practical the numerical solution of highly redundant systems and has been
used in conjunction with both forces or stresses and displacements as
unknowns. Further discussion of this method is beyond the scope of the
present work which is not concerned with iteration methods but the reader
is referred to the original literature on the subject.
In this Section we make use, where appropriate, of the matrix notation.
Although the complete analysis could be developed ab initio in this form
it is thought preferable to give first most of the basic principles in the more
familiar 'long-hand' notation. Only the most elementary properties of the
matrix algebra like matrix partition, multiplication, transposition and
inversion are necessary for the understanding of our theory. The reader
may consult the classical work of Frazer, Duncan and Collar~] for these
and more advanced matrix operations. Another modern and readable
account is given in the recent book of Zurmuehl. * * The most comprehen-
sive work to date on the formulation of aircraft structural analysis in
matrix notation, anyhow on this side oftheAtlantic, isthatofB. Langefors. tt
D. Williamst t presented r~cently an interesting account of some aspects of
matrix operations in static and dynamic elastic problems.
Before proceeding to a discussion of the general methods for the analysis
of redundant structures we introduce some concepts helpful to the under-
standing of the following theories and their subsequent matrix formulation.

A. Flexibilities
Consider a cantilever beam with a plane of symmetry yz consisting of
three connected segments a, b and c with bending stiffnesses for deftexions
in the yz-plane (£1) 0 , (Elh and (£!),respectively (see FIG. 20). Let the cor-
responding shear stiffnessesfll! be (GA')., (GA'h and (GA'),. Transverse
forces R 1 , R 2 and R3 are applied in the yz plane at the joints B, C and D.
Since the system is assumed to be linear the principle of superposition
holds and we can express the deflexions r 1 , r 2 and r 3 in terms of the loads
as follows:
~"1 =fit R1 +fi2 R2+/13 R3 } Fig. 20.-Fiexibilities and stiffnesses of a cantilever
r2=f21 R1 +/22 R2+/2a R3 . ·.................... (82)
r3=/31 R1 +/32 R2+/33 R3
where (see also Eq. (3))
where/;;, jjh are, of course, the well-known influence coefficients.§§ In fact,
jjh is the displacement in the j-direction due to a unit force Rh =I in the O;E;=axxiE.rxi+ . . . . . . . . . . . . +az.x;Ezxi I
h-direction. We call also /;; and /;h the direct- and cross-flexibilities re- ~.......... (84a)
spectively and deduce immediately from Maxwell's rrciprocality theorem O';Eh=a,,;Exxh+ · · ·: · · · · • · · · +a,,;Ezxh j
(Eq. (43), Section 3) that E;, a 1 (E1., ah) are the strains and stresses corresponding to a unit load at
I ·jjh ~I ·];.; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (83) and in the direction of j (h). Under load we understand either force or
To find the flexibilitiesfin any linearly elastic body we may use the unit moment. Similarly the f!exibilities may represent either displacements or
load method developed in Section 6D. Thus, from Eq. (71a), rotations. Naturally, formulae (84) yield also linear (angular) flexibilities

1r...........
due to moments (forces) respectively. Note that while E;, EA must be the
true strains due to unit loads at j and h respectively, a;, ah need only be
(84) virtual, i.e. statically equivalent, stresses due to the same loads. This is of
great importance in redundant structures. Thus, denoting by a 1, ii,. any
statically equivalent stress system due to unit loads at j and h respectively
in a redundant structure we cah write Eqs. (84) also in the form,

I ..........
• Axel Bendixen. Die l\4cllwdt• tier Alplu~Gieichungen :.ur Berechnung t•on Ruhmenkonstruktionen,
Springer, Berlin, 1914. If;;= fa;E;dV
v
I
t A. Ostenleld, Die Defurmationsmethod,, Springer, Berlin, 1926.
~ Calisev, K. A .• Te<hniski List No. 1-2, 1922, Nos. 17-21. 1923. See also Timo3henko and
(84b)
Young, Theory of Structures, McGraw·Hill, New York, 1945. l·jjh= fii;EhdV= fiiAE;dV=I fhi
II Hardy Cross, 'Analysis of continuous frames by distributing fixed end moments'. Paper N<•·
v v
1793, Vol. 96, Trans. A.S.C., 1932, pp, 1-10.
It is, of course, possible to substitute in the above formulae true stresses
§ R. V. Southwell, Relaxation methods in engineering science. Oxford Univ. Press, 1940. R. V.
Southwell, Relaxation methods in theoretical physics, Oxford Univ. Press, 1946. and virtual strains for true strains and virtual stresses but for reasons of
'T R. A. Frazer, W. J. Duncan, A. R. Collar, Elementary Matrices, Cambridge Univ. Press, logical consistency this is best avoided.*
Cambridge, 1938.
Assuming in the case of the beam shown in FIG. 20 that the Engineers'
•• R. Zurmuehl, Matrizen, Springer, Berlin, 1950.
theory of bending stresses is true we find, noting that the system is statically
tt B. Langefors, 'Analysis of Elastic Structures by Matrix Transfonnation with special regard to determinate and hence a=a,
Monocoque Structures'. Journ. of Aero. Sci., Vol. 11, No.7, 1952. Structural Analysis of Swept-
Back Wings by Matrix Transformation, Saa~. T.N. 3. August, 1951.
U D. Williams, 'Recent Developments in the Structural Approach to Aero2lastic Prohlerrs'.
J.R.Ae.S. Vol. 58, No. 522, June, 1954.
.. .............. (85)
ill\ These shear stiftnesses in bending are commonly based on the a~sumption of the Engineers' theory
of bending shear stresses. See Argyris and Dunne, Structural Ana/.vsis (Handbook of Aeronautics,
Vol. 1), Pitman 1952, for a derivation of the area A'.
§§The influence coefficients were discovered indep2ndently by E. Winkler, Mitt. Architek. u. lng.
Ver. Boehmen 1868, p. 6 and 0. Mohr,Zeit. Architek. u.lng. Ver. Hannover. 1868, p. 19. • See Section 6D, p. 14.

18
where M;, S; (M,., S,.) are the moments and shear forces corresponding to a3 a b
R; = 1 (R~> = 1). Eqs. (85) yield easily the following set of influence co- hiB=3(El).J23s=(GA')a +(GA'h .................. (92a)
efficients,
Such a splitting of the flexibility matrix is extremely useful in numerical
(a+b+c) 3 -(b+c) 3 (b+c) 3-c3 c3 calculations, particularly when obtaining first approximations in which we
faa= 3(E/)a + 3(£/)b +3(E/)c neglect certain flexibilities. Thus, in a first approximate wing analysis we
a b c may neglect the rib deformability; at a later stage we can ascertain its
+(GA')a +(GA'h +(GA')c influence by adding the corresponding flexibility matrix to the original

1 [ (a+b)2(2a+2b+3c)
Ja2 =j"23 = 6(El)a b2
-b 2 (2b+ 3c) + 6(Elh(2b+3c) J flexibility (see Example (b) of Section 9). The method is, of course, quite
general as Eqs. (84) and (84b) show, for it is always possible to write
I ·fih = f [(a xx;Euh +a._.;Eyyh +a zz;E zzh) +(axy;Exyh +a·yz;Eyzh +
a b V a,x;Ezxh)]dV ............ (93)
+(GA')a +(GA'h
where the first expression in round brackets gives the contribution of the
direct strains to /;h whilst the second expression gives the contribution of
f 31 =.1;. 3 = 6 (~~)J 2a+3(b+c)] +(G~')a .................... (86) the shear strains. Eq. (93) shows also that the splitting may be carried a
step farther by considering separately, for instance, the effect of the
To obtain f 22 and .1;.1 from the expression for faa. omit in th~ latte~ the strains Eyy or of the three shear strains Exy, Ey., E zx· Similarly, in fuselage
terms c and b c respectively. Also to find.l;. 2=f21 omit the terms mvolvmg c ring analysis where we usually neglect the deformations due to shear and
in the last of'Eqs. (86). Naturally, we can also derive the influence coeffi- normal forces we can check their influence by adding the corresponding
cients (86) by direct integration of the differential equation for the de- flexibility F8 and FN to the matrix FB for pure bending deformations.
flected beam when shear deformations are included. A systematic method These matrices are,
for deriving the flexibility coefficients for compound engineering structures
is given later. .
Influence or flexibility coefficients are of great importance in the static
FB=[J ~;whds} FN=[J 'i~hds} fs=[J ~~~ds] ........... · (9~)
and dynamic analysis of linearly elastic engineering structures. In this where M;, N;, S; (Mh, Nh, Sh) are the bending moment, normal and shear
connexion it is most appropriate to make use of the matrix notation not forces due to a unit load atj (h).
only for conciseness of presentation but also for the ~ystemati~ progr~m­ It is often convenient not to operate in single loads (or moments) but in
ming of the considerable computational work usually mvolved m practical groups of loads (or moments), which are known as generalized forces. To
problems. The matrix algebra is, in fact, ideally suited for the automatic fix ideas, consider that in the example of FIG. 20 we select as applied
digital computers now available. generalized forces the three loads QI, Q2 and Q3 given by,
The matrix form of Eqs. (82) is,
r=FR .................................. (87)
where r and R are the column matrices of the displacements and forces
t ................
I
(95)

I I
II II or in matrix form
r= I~: ={r1 '2 'a}, R= I~: ={R1 R 2 R 3 } .... (88) Q=GR (96)
where G is the square matrix, in general not symmetrical, of the coeffi-
L__i L _I cients G;h· We call G a load transformation matrix and assume that it is
and non-singular, i.e. that the determinant I G I of the coefficients is different
from zero. We may solve Eq. (96) for R by premultiplying with G- 1 and

Uh2 .h~
ll
f= /21 /22 /2a
fa1 fa2 faa
................................ (89)
obtain

L _I
is the so-called flexibility matrix; note that F is a symmetrical square
matrix. The relation (87) is, of course, valid for any number m of displace-
ments and rotations in any linearly elastic body. To each displacement or
rotation r; there corresponds a force or moment R;. Thus, in such a case
the matrices are
and
where
R=8Q
} ................ (97)*
is the so-called reciprocal or inverse matrix of G. Its determination is
equivalent to solving Eqs. (95) and therefore involves considerable numer-
ical labour if the number of equations is large. In such cases approximate
methods may have to be used. However, with the advent of the automatic
digital computers this difficulty is no longer insuperable. We give later in
this Section a systematic procedure suitable for punch-card machines for
computing the reciprocal matrix but hope to return in greater detail to this

~r-1~ !i.2_ ~ ·_ ·_·_·_I;~ ~ ~ ~ :-!i.-:l

and similar questions in Part III. Next we have to determine the general-
ized displacements q corresponding to the generalized forces Q. By defini-
tion q are obtained from the equality of the two expressions for work in the
f= ~;~ !:_2_ ._. _·:: !:_;_. _· _·::: ~-m-1 .................. (91) two sets of variables R, r and Q, q. Thus, in matrix notation
/m1/m2 • • • • /m; • • • · • fmm W=tr'R=tq'Q .................................. (98)
_j where r' and q' are the transposed matrices of the column matrices r and
where the/;~> can be calculated always from Eqs. (84). The flexibilities in q and are hence the row matrices
(91) need not necessarily refer tom different points. For example, we may r' = [r1 '2 ra], q' = [q1 q2 qa] · · · · · · · · · · · · · · · · · · · · · · · · · · (99)
choose three directions x, y, z at a particular point of a three dimensional Using the first of Eqs. (97) in (98) we obtain
body and define six flexibilities
q'=r'8
J.,.,,/yy,Jzz or
J.,.=J• .,,f•• =f•• ,J•.,=J.,. q=8'r (100)
corresponding to the three forces R.,=1, R.=1, R,=1 at the same point. where 8' is the transpose of the matrix 8, i.e.
Similarly for a beam in which we assume that the engineers' theory of
bending is true we may require the slope and the deflexion at a cross- I I
section under transverse force and moment applied there. Three flexi-
bilities are required for this information; note, however, that if shear
deformations are included we must specify that the bending moment is
Bn B21 Bal I
I
8'= I~~: ~: ~:I ........................ (101)

applied as engineers' theory direct stresses at the particular cross-section.

L _I
A characteristic property of the influence coefficients is that any/;~> in a Substituting (87) into (100) we find,
given elastic body depends only on the points and directions j and h but q=8'FR=8'F8Q=F.Q ........................ (102)
not on any other directions selected for the calc].Jlation of a flexibility Eq. (102) shows that
matrix (91).
A perusal of the flexibilities of Eqs. (86) shows that it is possible to split F.=8'F8 ........................................ (103)
the complete flexibility matrix F into two additive and distinctly different is the flexibility matrix corresponding to the generalized forces and dis-
matrices. Thus, placements Q, qt.
f=FB+Fs ...................................... (92) We illustrate now the application of generalized forces on a simple
where FB and F8 are the flexibilities corresponding to pure bending and
• In practice the generalized forces are more naturally defined directly by the matrix B ofEq. (97).
shear deformations respectively. The first contains only terms involving
Eland the second only terms involving GA'. For example, t Formula (103) is also given by W. J. Duncan, Mechankal Admittances and their Applications to
Oscillation Problems, A.R.C., R. & M. 2000, 1947.

19
 r---a--l---a--1----a-----j and therefore
A B I I I
I - 0·6a+ 1·5a -0·4a I I a/2 I
I r= 2·4a+ I·Oa+0·4a
--3 · 6a ·I ·Oa--0·4a
_j
I
L
5: I _j

r, =-sa in agreement with the previously given values of r. Each of the three
columns of the intermediate expressions represents obviously the r-

~~-st
. components of the corresponding q-coordinate. Fro. (21) illustrates in
detail the three q-modes .
Naturally, Eqs. (97), (100) and (103) are valid for any linearly elastic
body and any number m of forces (moments) Rand displacements (rota-

-0-61 -Nil

----~-3-&Q
l Q1 mode
(Q,=-6a)
tions) r. The load transformation matrix takes then the form
I
8 11 . . . . . . B1; . . . . . . B1 m
I

8=G- 1 = Bil ... ... B;; .. .... B;.,. (108)

_.............--,-·511--L I
0 0 0 0 0 0 0 0

Bm1 ...... Bm; . . . . . . Bmm

L _j

and is not, in general, symmetrical. However, the transformation (103),

called a congruent transformation, ensures that the flexibility matrix F • is
still symmetrical.
Attention is drawn to the dual relationship (97) and (100). Thus, if we
transform a load system R by the transformation
R=8Q
Fig. 21.-Generalized force and displacements the corresponding displacements r are transformed as
q=8'r .............................. (109)
example. Fro. (21) shows a uniform cantilever of bending Stiffness £/and It is often required to find the set of forces and displacements P, p for which
negligible shear deformability. The flexibility F for forces R1 , R2 , R3 at the all cross-flexibilities fvih (when jc#h) are zero. These are, of course, the
equidistant points B, C, Dis easily found from formulae (86) to be, elastic eigenmodes corresponding to the set of displacements r 1 to rm. The
load-displacements law is then given by
1 1 p=FpP
F=a 125 165 288 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (104) where F v is the diagonal matrix,
8 28 54
L _j I I
where fvu 0 ...... 0 ...... 0
a3
I0 fv22· . . . . . 0 ...... 0
a= 6 El ........................................ (l04a) I
We seek now the components of forces and the generalized displacements
corresponding to the generalized force,
I
Q2=1
for tbe assumed load transformation matrix, fv= I0 0 ..... .fvii· ..... 0 ............ (110)

8=
~,01 I
0· 5 -1
1
11
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (105) I
1 -0·5 0·5
L _j I0 0 ...... 0 ..... . fvmm
Using Eq. (97), L _j
R=8Q=8 {0 l 0} Thus,
we deduce immediately, P1=fvuP1, ...... ,p;=/v;;P;, ...... ,pm=fvmmPm .... (Ill)
R={R1 R 2 R3}={1 0·5 -0·5} It is always possible to find the unique load transformation matrix 8v
Hence, which transforms our system R, r into the orthogonal system P, p. We do
r=FR={a/2 -a -5a} not enter here into its detailed derivation since the reader can consult a
The values ~f Rand rare shown in FIG. (21). number of textbooks on this subject.*
Next we find the generalized flexibility F • by straightforward matrix Our above considerations and in particular Eqs. (84) and (84b) show that
multiplication from Eq. (103), the ftexibilities are particularly simple to derive for a statically determinate
structure, e.g. the beam of FIG. 20. For a redundant structure we must first
find the forces or stresses in the redundant members before we can obtain
1
F 0 =8'F8=a 1126 -6 -1
-6 2·5 10~ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (106)
the true strains €; for the unit loads. The necessary analysis is developed
later but it is helpful to give here a formal matrix derivation of the flexi-
10 -1 1·5 bility F of an engineering structure, the stress distribution of which is
L _j known. To this effect we use again the unit load method given by Eq.
(7la). We denote by eh, sh two column matrices for the true strains and
We calculate now the generalized displacements from Eq. (102) as, stresses respectively, corresponding to a unit load Rh = 1 at the point and
q=F.Q=F.{O 1 0} direction h. Thus,
or
q={-6a 2·5a -a}
} 0 0 0 0 0 0 0 0 0 0 0 0 (112)
Finally, we analyse the three generalized displacements q in their r-com-
ponents. From Eq. (100),
r=G'q =(8- 1)'q =(8- 1)' { -6a 2 · 5a -a} where the elements of these matrices may, of course, vary with x, y, z.
It is always possible to write
For the inverted and transposed matrix of 8 we obtain from Eq. (105),
eh=fvsh ........................................ (113)
I I
(8- 1)'=G'= I 0·1 0·6 0·41
0·4 0·4 -0·4
0·6 -0·4 0·4
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (107)
where fv is the flexibility matrix of a unit cube at the point x, y, z. Thus for
an isotropic body
L _j • See Frazer;Duncan, Collar, Toe. cit. Zurmuehl, foe. dt.

20
 I -l The matrix b has submatrices with m columns. f is a partitioned diagonal
I 1/E -v/E -v/E 0 0 0 I matrix whose elements are the flexibility matrices f •·
I -viE 1/E -v/E 0 0 0 I We denote now by b a matrix whose m columns are loading systems on
fv= I -v/E --v/E 1/E 0 0 0 I ........ ( 114) the s members statically equivalent to the external loads R1 =I, R2 = I,
0 0 0 1/G 0 0 I ..... , R,. =I respectively. If we choose these systems to be also elastically
0 0 0 0 1/G 0 I compatible then, of course,
I 0 0 0 0 0 1/G
b=b .............................. .. (124)
L _j
Let also 5; be a column matrix for any statically equivalent stress system Applying now the unit load method and using Eq. (122) and the transpose
corresponding to a unit load R; = 1 at the point and direction j. Thus, matrix -b' we find by an argument similar to that leading to Eq. (117)
i;={CT~..~i1 11 u}Gzzjij;,·yiUvzi1z.-:i1 that the deflexions rat the points of application and in the directions of the
• • ••• · · . . . . • . • . . . · ... · (115)
loads R are given by
For a statically determinate system only one 5; system can be given-the
true one, 5;. We derive now the flexibility coefficients};h from the unit load r=b'fbR ............................ (125)
method as, Therefore the flexibility matrix F for the prescribed m directions in the
1·/;h= Js;'ehdV= fs/fv5hdV= J sh'f~·5;dV= I ·J,; complete structure is
v v v f=b'fb ............................ ( 126)
(116)
I f;; = f 5/fv5;dV The matrix operations in (126) are again congruent and thus F is indeed
v symmetrical. Eqs. (123) show that Eq. (126) can also be written in the form
Hence the total flexibility matrix F for m points and directions is,
f=~.b'.f.b. . ..................... (126a)
F=[/;h]=Js'fv5dV ............................. . (117)
v Eq. (126) is the general expression for obtaining the flexibility of a
where 5, 5 are the partitioned row matrices complete structure from the flexibilities of the constituent elements. The
configuration of the elements is said to be in series since the assembly con-
5 [51 ...... 5; ...... 5m] 1
5=[5 1 . . . . • • 5; ...... sm]
j ··········
(118)
dition is expressed by the matrix b which derives from conditions of
equilibrium. Thus, Eq. (126) may be regarded as the most genera/formula-
tion of the flexibility matrix of a structure consisting of elastic elements in
series.
We shall apply now formula ( 117~ t~ an engineering st~ucture consisting It is also clear why Eq. (103) for the flexibility matrix of generalized
of any numbers of simple elements JOtned together at the1r ends <?r ?o.und- forces has the same form as Eq. ( 126). In the first case we derive general-
aries. These elements may be plates, flanges, beams, rods, pm-Jomted ized forces from single forces and in the second internal forces from ex-
trusses, etc., and take in such a structure the ~lace of the volume element ternal forces but in both cases this entails a linear transformation matrix
dV in a continuum. Let the structure be subjected to the force (and/or B or b. Note also that F is in the first instance the flexibility matrix of the
moment-) system complete structure for the single forces and f in the second instance the
R={R1 . . . . • • R; . . . . . . Rm} ................ (119) flexibility matrix of the individual elements. It is seen, however, that
whereas B is always a square matrix b is, in general, rectangular.
where R. itself need not necessarily be a single force or moment but may be Before illustrating the application of Eq. (126) we draw attention to an
a gener;lized force. Due to these Io~ds the typical g member is subj~cted interesting dual relationship (see alsop. 20). Thus, Eqs. (121) and (125)
at its ends or boundaries to a Ioadmg expressed as a column matnx s. prove that if the internal loads S are derived from the external load system
whose elements are direct and shear stresses, or stress resultants, e.g. R with the relationship
torque, bending moment, shear force, normal force, etc. Now s. is ob-
viously linear in the R's. Thus, S =bR .............................. <121)
S,c b,R ............................. ........... (ll9a) the deflexions r at the points of application of the R-loads are found from
the internal relative displacements (strains) v from the relationship
where b is a rectangular matrix with m columns, and corresponds, of
course, t~ the stress matrix 5 at a point (x, y, z) of a three-dimensional r b'v b'v ...................... (125a)
continuum. If our structure is redundant b, cannot be determined by Naturally, Eq. ( 125a) merely restates the unit load theorem. We stress
statics alone. However, for the present we assume that b, is known. It is again the fact that 6 need only be the matrix of statically equivalent stress
obvious that the relative displacements (shear angle, elongation, deflexion, systems.
slope, twist, etc.) v • at the ends or boundaries of the g element can be
written as a column matrix Illustration of Eq. ( 126).
v.=f.S.=f 9 b.R ............................. ... (120) We observe first that Eq. (126) includes as a particular case Eq. (92)
for the splitting of a flexibility matrix. This may be seen as follows.
where f, is the flexibility matrix of the g elem~nt and has as many rows as Splitting the flexibility matrix is equivalent to considering the combined
v . Each element being assumed to have a simple geometry it is usually effect of two or more geometrically identical structures (elements) to each
c;sy to write down-often merely by inspection-the matrix f •. Since of which is assigned only part of the complete flexibility of the structure
there are, in general, alternative but equivalent ways o_f expressing ~he (e.g. flexibility in bending or shear, or normal force). Thus, the constituent
loading s. on the element, there are als? correspondmg!y alternat1ve elements are in this case geometrically identical and hence the load trans-
expre~sions for the flexibility f •. This aspect 1s elaborated on m an exa~ple ference matrices b, etc. are merely unit matrices I. If then, flexibilities of
at the end of our main argument. Note that f 9 represents all deformations each of the elements are written as f., f 0 etc., we find
that are necessary to ensure the compatibility of the g element within the
complete structure.
The internal stress and deformation matrix of the aggregate structure f" 0 ..... 0 ..... 0
may now be expressed as 0 fb ••••• 0 . . . . . 0
5={5.5 0 • • • • • • S, ...... S,)c-bR 021)

V={VuVo ...... v, ...... VJ=fbR .............. (122)

where Sand v are single column partitioned matrices and
F cc [I I .... I ..... I] 0 0 ..... f, ..... 0
f" 0 ..... 0 ..... 0
0 f,, ..... 0 ..... 0

0 0 ..... 0 ..... f,
f 0 0 ..... f, ..... 0 . . . . . . . . (123)
=f,+f,+ .... +f"+ .... +f, ........ (127)
q.e.d. The order of the unit matrices and the f matrices ism, the number of
assigned directions in the structure.
Consider next a beam built-up by two uniform component beams a and b
b., 0 0 ..... 0 . . . . . f,
as shown in FIG. 22. We seek the flexibility matrix for the transverse forces
-I R1 , Ra and moments R2 , R4 under the assumption that the E.T.B. holds
*The usual presentation of the strain and strC''\S matrix as a 3 / 3 squar~.. 111 ~ ... !.""~ Oi n:•nsor is r.''~.. _,. and shear deflexions are negligible. We analyse first each beam separately
suitable for our considerations her('. Se~· also ~e'-·twil uu. There too attention js drawn to the facl as a cantilever built-in at the L.H.S. and subjected to transverse force and
that i i need only be determined in the most suitable statically determinate svstem.
hPnding moment at the tip, the signs of which are taken to be those of the

21
 where
lOtS(+) 'D~M<+> c/>a,h=(L)., b . . . • . . . •• • . . . . . . . . . . . . . . (132)
tR1 tR 3
gives the flexibility per unit length of the cantilevers. The negative sign in
\Ela the cross-flexibilities arises from the sign convention. The total flexibility
R4 SJ
0
R2Jf"""""\ \Elb follows as
A Element a 8 Element b
cl I I
I f. o I
r,
f= o f" I I
.......................... (131a)
L _j
Applying now Eq. ( 126) the flexibility F of the complete structure is
f=b'fb=b' .f.b.+b' bfbbb ..••............•..•..•. (133)
la lb
~ In the present case ii = b since the system is statically determinate. The
deformations of the structure may finally be obtained from
r=FR ............................. ... (87)
The expressions for the deflexions r1 and r3 derived from Eq. (133) agree
with those of Eq. (86) when the shear deformations are neglected.
An alternative approach to the problem is to express the loading on the
component beam by the end moments (see FIG. 22b). The internal loading
matrix is now,

.................... (134)

where
I
1
-1. -I _,I I
0 -1 -- 1
I o o -- 1 I .................... (135)
I o o I I
L _j

G MBb
~J
-1£-/..................;...;....;,_A
Ele men! b
M~
The internal flexibilities fa and fb derive in this case solely from the end
bending moments and to find them we have to consider only the end slopes
of the simply supported beams shown in FIG. 22b; thus, taking end slopes
positive in the direction of positive moments we find
I I

fla,b= l 1cp bI .. ·............................. (136)

i/6El 11-~. 1-~. I
I 6'~' 3'~' I
L _j
Fig. 22.-Break-down of cantilever for calculation of flexibility matrix Hence,
(137)
where,
corresponding + S (shear force in the beam) and + M (bending moment in
the beam). The tip deflexion and slope are fixed as positive if they are in the
direction of the positive applied shear and end moment respectively. With fl =
lfla
l
0 lI .......................... (136a)
this sign convention, o flb I
L _j
Scb=Ra
} ............ (128) It is easily seen that the flexibility matrix F of Eq. (137) is identical with
MBa= -R2 -R4 -R3lb Mcb= -R4 that of (133). We observe that the cantilever position of the constituent
elements is derived merely by a rigid body rotation from the simply sup-
The loading matrices s. and Sh of the two elements are accordingly ported beams shown in FIG. 22b. No additional total virtual work is asso-
s.={SBaMBal=b.{Rl R2 Ra R4}=b.R I
(129) J............ ciated with such a linear transformation and the flexibility F is, hence the
same.
The form (136) of the flexibility of the element is important for it applies
to a linear variation of any deforming stress, force or moment as long as
where we substitute forcp the appropriate unit flexibility. Thus, for a normal force

j j
I varying linearly in a beam with constant direct stiffness EA, we can use
0 Eq. (136) with cp= 1/ EA .
b.= I~ -1 -lb
.. .. .. .. .. .. (130)
B. Stiffnesses
L _j _j
We return now to Eqs. (87) and solve them for R 1 to Rm to find
and hence equations of the type
II R1 =k 11 r 1 +k12r 2 + ...... +kvr;+ ...... +k 1 ,.rm I
I b. I R2 =k21 r 1 +k 22 r2+ ...... +k 2;r;+ ...... +k2mrm I
b= · · · · · · · · . · ................ (130a)
I bb I R;=k;lrl +k;2r2+ ~-~ ~+k;;r;+~. ~-~k;mr,..- ~ · · · · · · · · 0 38 )
L .J
The flexibility matrices f., f b follow immediately from FIG. (22),
-=----------
1 1 2
-----1
Rm -km r +k,. 2r + . ..... +k,.;r;+ . ..... +k,..,..r,.. .J

The coefficients k;i and k;h are known as direct- and cross-stiffnesses in
I I the directions of the selected m displacements. In fact, it follows directly

f,, b =
I f3
3cfo -2cfo I
12
J2 I
1 . . . . . . . . . • . . • . . . . . . . . . <13t >
from Eqs. (138) that the general stiffness k ih is the force (or moment)
applied in the direction j if we displace the body by rh =I whilst keeping
the remaining (m -I) r's at zero. Using matrix notation the solution (138)
I -2cfo tcp I •. b
of Eqs. (87) may be written as
L _j R=F- 1 r=Kr ............................. ....... (139)
22
where R and r are the column m:1trices defined by Eqs. (90) and K is the
m x m stiffness matrix
I I
k 11 ...... k 1; ...... k 1.,. I
K= kn ...... k;; ...... k 1, .................. (140)

1 kml• · · · km;• • · · . . kmm

L _j
The stiffness matrix may be determined either directly or from the identity
K=F- 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . (140a)
by inversion of the flexibility matrix F. Eq. (140a) shows that K is sym-
metrical, i.e.
k ih cc k hi • . . . • . . • . . . • . . . . . . . . . . . . . . . . . • ( 141 )
This may be seen also as follows:
Let a1, E1 and a\ E" be the stresses and strains corresponding to unit dis-
placements r1 =I and rh =I respectively while all other r displacements are
kept at zero. Applying now the principle of virtual work or displacements
to the true state j (h) and virtual state h(j) we obtain
l·k;h=fci'E 1 dV=fa 1 EhdV~cl·kh; ........................ (142)
v v
Where ci'E 1 etc. stands for an expression analogous to Eq. (84a). This
application of the pr nciple of virtual displacements, by an obvious
analogy with Eq. 84, is called the 'unit displacement method'.
We remarked on page 19 that the direct- or cross-flexibilities depend
for a given structure only on the points and directions to which they refer.
This is not so for the stiffnesses which by definition depend on the complete
set of points and directions selected to describe the stiffness of the body.
Thus, if we choose an additional direction m + l to augment our descrip-
tion of the elastic behaviour of the structure all the original k;h will in
general change whilst the jjh remain unaffected.
Consider again now the example of FIG. 20. A study of Eqs. (86) and
(140a) shows that the stiffnesses k 1" corresponding to unit deflexions at
B, C and D are considerably more complicated than the expressions for
the flexibilities. However, this is not always the case. Naturally, we can Fig. 23.-Stiffnesses of a cantilever
calculate the stiffnesses directly. For example, the k 13 may be obtained by
analysing a continuous beam built-in at A and simply supported at B, C is the stiffness per unit length of a bar.
and D at which last support there is a fixed 'give' of unity. We may solve The stiffnesses at a point associated with the unit displacements at the
this thrice redundant problem either with the three-moment equation or same point are, in fact, already derived by the method of vittual displace-
by the slope-deflexion method. ments in Example 5b where they appear as the coefficients to the displace-
Assume now that not only transverse forces but also moments are ments u, v. Naturally, the problem of deriving the single stiffness k22 at
applied at the junctions of the component beams and at the tip (FIG. 23). the point 2 from the set ( 145) still remains. A general method for solving
To simplify the argument we ignore, moreover, the effect of shear deform-
ability. The modes and stiffnesses corresponding to unit displacements in
the directions, I, 2, 3, 4 can now be determined very easily. For example,
for the modes r 1 =I and r 4 = l shown in FIG. 23 we find respectively,
R,. r,,
ku =12( ~{) + 12(~{) k31 12( ~:) R,
r,,l
n 0 • = 1,
...... (143)
k21 -6( ~:)a +6( ~:)
= k41- 6( ~:)I>
b ' h

k44=4(1!{). k24=2(~1 )b ,d,---~..::...-~-~-~_1

................ (144)

The important point about this example is that it shows how easy the
determination of the stiffnesses can be once we consider all possible modes
of deformation a_(joints connecting simple component elements of a structure.
Another example will help to clarify the argument further. Consider the
symmetrical framework of FIG. 24 and assume that we seek the flexibility
or stiffness at the central point 2 for vertical displacements. In the first
case we must solve a thrice redundant problem and in the second a four
times redundant problem with a central unit "give'. If, on the other hand,
we select the complete set of stiffnesses corresponding to vertical and hori-
zontal displacements at all movable joints then the calculations are most
simple. In fact, for the typical cases shown in FIG. 24 we find by inspection,
Kh Kd(h) 2 Kh
k2.2=·7i+2d d ' k4.2= -71
Kd ha Kd(h) 2
k7,2= -ku.2={j {j2 , ks.2=k12.2=' -d (J
k,~-·-
k1.2 =k3,2 =ks.2 =ks.2 =k9,2 =kw.2 =0 ~1<,1
.. (145)

, k s.1- --~
-k 9,1- a

I Unit stiffness of all honzontal bars 'Ka

Un!l sttffness of all vertrcal bars ')(."
k2,1 =k3,1 =k4,1 =ks,I =k10.1 =0 _j
Untl &ttlfness of all dtagonal bars 'Kd
where
K=EA (145a) Fig. 24.-Stiffnesses of pin-jointed framework
c
23
this and related problem" ts given further below. It is characteristi..: that f--·--·- a - - - + - - a a- _ _j
the direct calculation of the llcxibilities corresponding to ( 145) is as com- A B c D
plicated as that for the single flexibility at 2.
I
I
In both examples we see that the stiffnesses are determined most straight-
forwardly, in fact, practically by inspection, once we find the set of unit I
deformations for which it is simple to calculate the strains and hence I
stresses and forces. The advantage of first deriving the stiffnesses may be
particularly marked in highly redundant structures but it requires, as the
example of FIG. 24 shows, the consideration of many degrees of freedom ';=04 t;= -0 4

which may also have its disadvantages. On the other hand flexibilities are
always easier to calculate if the stresses corresponding to unit forces can be
found without difficulty as in statically determinate structures.
We gave in Eq. (142) a general formula for the determination of the
stiffnesses. Let us consider it again in more detail. Observing first that,
while ui and if must be the true stresses corresponding tor;= I and rh = I
respectively, the strains Eh and €; need only be virtual strains (h and (i fii~-----------R-•.._fo_•_••_>l_a_ ____,R11-o-006l/a O, mod"
(i.e., compatible but not necessarily statically consistent strains*), corres- , (Q, =- o-oo62/Q)

ponding to rh =I and r; =I respectively; this may contribute to a con-

siderable simplification of the calculations. However, the actual practical
use of Eqs. ( 142) rewritten here in the form
k;1 =Jai£.1dV
v .......... (146)
k 1h =.fif£ 1dV = fai€hdV -=k 1.;
v v
is somewhat limited. This follows from the previous discussion which
shows that stiffnesses are best found either by considering all possible
degrees of freedom at the joints, in which case the determination of the
k's is usually performed by inspection, or by inverting the flexibility
t
Q3 mod•

mJtrix F. Moreover, even if we calculate the stiffnesses k for a restricted Ill,= o•os/al
o-2Dl/a
total number of degrees of freedom at the joints (e.g. example of FIG. 20) RJ

Eqs. (146) are really superfluous. Thus, in the example of FIG. 20 we have
to find the true stresses a fvr a four times redundant structure, the analysis Fig. 25.-Generalized displacement and forces
of which includes the derivation of the forces k;h and the use of Eq. ( 146)
is hence unnecessary. Nevertheless, Eq. 146 is of considerable value when
the elements into which the structure is broken down are characterized, We illustrate now in FIG. (25) the application of formulae (147)-(149)
not by simple loading systems (e.g. beam elements or bars) for which on the simple example of FIG. (21) and for the same load transformation
the k's are determinable by inspection, but by simple (assumed) dis- matrix B of Eq. ( 105), but seek here the components of forces correspond·
placement patterns. An example of this application is given in D of this ing to a generalized displacement
section. Also given later is the matrix formulation of Eq. 142, which is q2 ~,I
most useful in practical cases. The displacement transformation matrix A of Eq. (147a) is given by Eq.
We now find a generalization of the concept of stiffness corresponding ( 107)
to the generalized flexibility given on page 19. Thus, following the argu- I I
ment there we introduce the generalized displacements q and forces Q I 0·1 0·6 0·41
defined by A=G'=(B- 1 )'= I 0·4 0·4 -0·41 ............ (107a)
r·· Aq Q=A'R .............................. (147) I 0·6 -0·4 0·41
where [ __ .J
A=G' =(B- 1 )' . . . . . . . . . . . . . . . . . . . . . . (147a) Hence
see also Eqs. (96), (97) and ( 100). Substituting the expressions for rand R r~Aq,=A{O 1 0}={0·6 0·4 -0·4]
in Eq. (139) we find immediately
The generalized stiffness matrix K,, is best obtained by inversion of F,, in
Q=K,,q .............................. ( 148) Eq. ( 106) and is
where
K0 =A'KA ............................ (149)
i-- -l
is the generalized stiffness corresponding to the m generalized displace- 1I 0·0169
--0·00615 -o · 117 1

ments q 1• Eq. (149) may naturally also be derived by inversion from Eq. K =-~-0·00615 0·548 0·406 1 .......... (l06a)
(103), i.e.
'' a -0·117 0·406 0·218 I
K.=(F.)- 1 . . . . . . . . . . . . . . . . . . . . . . . . . . (149a) L _j

The particular linear transformation B (or the corresponding matrix A) where a is given by Eq. (104a). Hence,
which reduces the cross-flexibilities[;h to zero nullifies also the correspond- Q=K.q=K.(O I 0}
ing cross-stiffness k;h· In fact, we obtain from Eq. (1o::l) or
P=KvP .............................. (150)
Q=l{ -0·00615 0·548 0·406}
where Kv is the diagonal matrix
I I We analyse finally the generalized forces Q in their R-components. From
lkv 11 0 ...... 0 ...... 0 Eq. (97) or ( 147)
\0 kv22······0 ...... 0
I
I
Io
R=BQ~-. I
1
r-
I Ill
0·5 --1
I
1 0·548 I
I
-0·00615 I
a II -0·5 0·5 I 0·406 I
L_ .J L .J
Ii ............ (151) and carrying out the multiplication,
K"=
0 0 ..... k,,;; ...... 0
I I I I
I I ·954/a I
0 +O· 548 +0·4061
R=,-1 -0·006 +0·274 -0·406 I +O_
-0·138/a i
a -0·006 -0·274 +0·203 I I -0·077/a 1
L j L .J
II o o ...... 0 · · · • · · kvmm
Each of the three columns of the intermediate expression represents ob-
viously the R components of Q1 , Q2 , Q3 respectively. We can check now
L .J the previously given result for r from
and r=FR
I I I
kvu =f..,-' kp; 3 --r , .. • .. • , kvmm =~,-- where F is given in Eq. (104).
1)11 } pn pmm Next we derive a general formula for the stiffness K of a structure con-
.................... ( 152) sisting of a finite number of simple elements. The expression given is the
• See Section 4 • matrix formulation of Eqs. (146) and corresponds to the flexibility matrix
24
F of Eq. (126). Since the analysis follows closely the arguments on page 21 TABLE I
we need present here only the outlines of the proof. Duality of Force and Displacement Methods
Consider again an assembly of s structural elements joim•d together (it is always possible to substitute a, b for i, b respectively)
at their ends or boundaries. m displacements r are selected to describe the
stiffness K of the complete structure. Let k. be the stiffness matrix of the g Method of Forces Method of Displacem~tnl s
element due to the characteristic strains of the element arising from the
displacements v. at the boundaries. Naturally, there are usually several
different but equivalent possible ways of expressing the straining of the Force I R Displacement r
element. Let a. be the matrix, in general rectangular, which transforms the I K
displacements r into the true strains v. of the element. i.e. Flexibility F FK=I=KF Stiffness

Force +
Then
(153) +
Displacement r R
s.=k.v.=k.a.r ........................ (154)
is the matrix for the forces (moments, etc.) applied on the element due to Generalized Force Q q
'Generalized Displacement
the displacements r. The internal force (or stress) matrix S of the aggregate
R:BQ r=Aq

1
structure is now given by
S=kar .............................. (155) I
where Generalized F~exibility ~ A'B=l=B'A Generalized 1 Stiffness
S={SaSb ...... 5 0 • • • • • • S,} (!56) ~= BFB ~ Kq=l= Kq~ Kq=~ KA
and t t

l
a={a.ab ...... a •...... a,} .................. (157)
Generalized Displacement Generalized Force
k is the symmetrical diagonal partitioned matrix,
q: B'r = ~Q Q :.A'R = Kqq
I
...... 0 ...... o
0 ...... 0 Generalized Series ASSE'mbly Generalized Parallel Assembly

I Stress on elements s Stra~n of elements v

S = bR v = ar
k= ................ (158)
Strain of elements v Stress on elements s
0 0 k •...... 0 r= 6'v R=aS
Flexibility of elements f Stiffness of E'lements k
(for stresses S) (for strains V>
Flexibility of complete structure Stillness of complete structure
F= 6'f b K = a'ka
0 ...... 0 ...... k.
_j
Addition of Flexibilities Addition Of Still nesses
Applying now the principle of virtual work, taking the internal forces S
.
(Spe-i:ial uriu a•umbly) ( Sp•ci.al parallel aaembly)

and external forces R as the true state and selecting as virtual state the

r= r .riJ?-
internal strains corresponding to unit displacements ' I = I, r 2 = 1, ...... ,
rm = 1 respectively we find .
b

~ -r --~
k• ~
R=a'kar ............................ (!59) 1
where a' is the transpose of a. Thus, the stiffness matrix K of the compound F. + F. = F K.
structure is
K=a'ka .............................. (160)
Eq. (160) may also be written as Before illustrating applications of Eq. (160) we draw attention to the
by now all too apparent complete parallel between the flexibility and
K=~a'.k•a• ........................ (160a) stiffness approach in the analysis of structures. We may express this con-
g
cisely by the tabular arrangement under the two headings: 'Methods of
Since the virtual strains need only satisfy the compatibility but not neces- Forces' and 'Method of Displacements'.
sarily the equilibrium conditions we may select for the virtual states a The analogy between the two methods is developed considerably in
simpler matrix a which satisfies only the former. Eq. (160) becomes then what follows and is shown in greater detail in TABLE 11.
K=a'ka ............................ (l60b) Illustrations to Eq. (160).
However, the application of a possibly simplified matrix a is really not Consiqer the beams I and II of FIG. (26) joined by inextensional bars
required in practice. As mentioned on page 23 the stiffness matrix K is which connect the set of points B, C, D and B', C', D' respectively. Let KI
best calculated for all degrees of freedom at the joints, yielding very simple and K 1 I be the stiffness matrices of the upper and lower beam respectively
matrices. defined for vertical displacements 'I• 'a and r 5. From the definition of the
The configuration of the elements of the compound structure is said to stiffness it follows immediately that the stiffness K for the displacements
be in parallel in Eq. (160) since the assembly condition is expressed by the 'I• 'a and r 5 in the compound structure is given by
matrix a which derives from conditions of compatibility. Thus Eq. (160) K=K 1 +Ku .........................• (161)
may be regarded as the most general formulation of the stiffness matrix
for a structure with constituent elements in parallel. It is immediately This simple result may also be derived from the general Eq. (160). For in
apparent why Eq. (l49) which expresses the stiffness matrix for generalized this special case the joint displacements r etc. of the complete structure
displacements must have the same form as Eq. (160). In the first case we and the straining displacements vI and vI I of the component beams are
derive generalized displacements from single displacements and in the the same. Thus,
second, internal strains from external displacements. In both applications VI=a1 r, Vu=aur .............................. (162)
this entails a linear transformation matrix which, however, is a square where
matrix in the former case. Also K is the stiffness of the complete structure
for the single displacements while k is the stiffness matrix of the individual and
members.
II 00 01 01~
Eqs. (153) and (159) show that there is a most illuminating parallel 1 0 0
development to Eqs. (121) and (125a). Thus, if the internal relative dis- a 1 =au=l= ........................ (163)
placements (strains) v derive from the external displacements r with the 1

relationship L _j
v=ar .............................. (l53a) We conclude,
Then the external forces R derive from the internal forces (stresses) S with
the relationship
R=i'S=a'S .......................... (l59a)
Eq. (l59a) restates, of course, the principle of Virtual Work.
25
 ~ ~~ tt; !'5 I
'i.~ r:.~ rs~
I
B
I
c
I
0
I 1/T/~Ti = 1/l~r +/
t r, J---X-~~~ Jt--~

~
j'3 Irs I

TI B' c· D'

f'~ f'3 f'S

"""'
~
~ 'i" '6~
?di B c 0
~ Fig. 27.-Addition of stiffnesses for arbitrary structures

~
further by comparison of Eqs. (92) and (161). The first shows the case of
c• o· additive partial flexibilities for series assembly and the second, the case of
~
B'
additive partial stiffnesses for parallel assembly. A very simple application
of Eq. (160) is given by the pin-jointed framework shown in FIG. 12 of
Fig. 26.-Parallel combination of cantilever. Addition of stiffnesses example 5b. *Thus, the stiffness k, of the bar corresponding to unit elonga-
tion ~I,= I is,
Eq. (161) applies to any compound structure in which the stiffnesses are -k _ (EA), -~
defined for at least all common degrees of freedom associated with the k, -- r -- I, -I, · · · · · · · · · · · • · . . . . . . . . . . . . . . . . . (167)
joining of structures I and II. Thus, in the example of FIG. (26) the common
degrees of freedom are t~e vertical displacements r 1 , r 3 and r 5 • Formula where K,=(EA), is the stiffness per unit length of the rth component bar.
(161) is, however, still tru~ if we define the stiffnesses of the upper beam The transformation matrix a, for displacements in the x- and y-directions
and the complete structure for both the vertical displacements r 1 , r 3 and
r 5 and the slopes r 2 , r 4 , r6 • Then K1 can be calculated by the methods given ~. [
a,= cos 8, sine, J .................... (168)
previously. Ku is still only definable for vertical displacement, the corre-
sponding entries associated with r 2 , r4 and r6 being zero. Naturally, we can Hence
define the stiffness matrix K and say KI for points not connected to IJ.
Again the corresponding terms of Ku are zero. FIG. (27) shows the joining . ......... { 168a)
of two arbitrary structures to give K= KI + KII· Note that at a joint point and
like (2) we must define the stiffnesses for two displacements, say the x and -1
y-directions. ..... 0
Formula (161) may, of course, also be applied in obtaining the stiffness ..... 0
matrix of the compound cantilever consisting of elements a and b, FIG. I
(23). Again, we must define the stiffness for all common deflexions and I
slopes at the joints, assuming the E.T .B. to be true and the shear deflexions I
zero. The total stiffness K is then
K=K.+Kb .......................... (t61b) I
where the elements of the split matrices may be found from Eqs. (143) 0 .... . K,/1, ... .. 0
and (144) for the displacements r 1 and r4 and similar equations for r 2 and I a
r 3 • Thus,
I I
127: El
0 -6[2 0
0 0 ..... 0
.
.... • Kn/1, I
I
0 0 0 0 L _j
K.= ................ (164)
El 4£/ 0 I
-6[2 0 I I I
I
ol
I "\'Kr e . eI
L.JT; cos , sm , I I kn kxy I
0 0 0
L _ja ........ (169)
I
12gj -12£/ 6£[2/
I I:~ cos e, sine,
L
L:? sin 2 e, I I k
_j L
Y> kyy
_j
I
[3 [3

_ 12£ / El El El in agreement with Eq. (67).

13 12 [3 -6[2 -6[2
............ (164a) We mentioned previously that the simplest method of calculating stiff-
El -6£/ 4~! 2£/ nesses is to define them for as many degrees of freedom as are necessary
6[2 [2 I I to obtain simple deformation patterns of the elements of the structure.
Having calculated such a stiffness matrix it will become necessary to
6£1 6EI 2£1 41!! I 'condense' it-i.e. to refer it to the smaller number of displacements in
I I
L [2 - [2 I
_jb
which we may be interested. This changes, of course, all the stiffness
coefficients, but the necessary analysis is easily arranged in matrix form
for automatic computation. Let the original stiffness matrix be of the order
where the columns and rows refer to displacements r 1 , r 3, r 2, r4 respectively.
Formula (16lb) may be generalized for any numbers of component m x m and denoted by K We want to find the matrix K referred to p-direc-
0 •

beams, and for any structure in which the joint displacements r expre3s tions only, where p<m. We have,
also the straining displacements v of the elements (i.e. a= I). In such cases R=K r 0
the stiffness matrix K can be written where
K=Ka+Kb+ ...... +K.+ ...... +K•............ (165) R={R 1 .....• Rv Rv+t ...... Rm}={RI Ru}
Note that the only non-zero coefficients in K. are the stiffnesses k corre- } .... (171)
sponding to the displacements at the ends or boundaries of the g-element.
The flexibility matrix F corresponding to (165) is in which we write first the p-directions required for the condensed matrix
K. Ko may be expressed as a partitioned matrix as follows,
F= K- 1 =(K. +Kb + ...... +K.)- 1 =(F .-1 +Fb- 1 + ...... + F,- 1)- 1 (166)
The parallel between the displacement and force method is underlined • See p. 10.

26
 X,
........................ ( 172)

where K 1 and K 11 are square matrices of order p and 111-p respectively.

Eq. ( 170) can now be transformed to
R1 =K 1 r 1 +K' 111 r 11
R 11 = K 111 r 1 + K 11 r 11
1
J......................
(173)

Also, by definition, in the structure with stiffness defined in p directions

only
R/·' Kr/ I
r
(173a)
R11 =0
Putting R 11 =0 in Eq. (173) and eliminating r 11 we find
Rlc~(K/ ·· K' ///K//-IK 1 u )r 1 ( 174)
and hence comparing with Eq. ( 173a)
K- K 1 -K' 111 K 11 -IK 111 . . • . . • • . • • • . . . . . . . • . • . . . (175)
Eq. (175) gives naturally the solution to the particular problem of FIG. (24)
discussed on page 23. Another example illustrating the application of
Eq. (175) is discussed under D of this Section.
The above method, is, of course, the basis of the solution of partly "give" at support C C, = Force appli•d to support • b ( 1 + ll)
homogeneous equations. A parallel relationship exists also in the 'force- by structure due to X1• I. a, b
method' investigation of structures. Thus, in this case, we have to find the
flexibility F of a redundant structure in which we know the flexibility F, of
the basic structure. The analysis is given under C below.

C. Th~ Calculation of Redundant Structures by the Force-Method

-J !!• AI /'\

We develop now a generalization of the Mueller-Breslau 6 * technique

~~~-
for the calculation of linearly elastic redundant structures. Following X,:l /
our investigations under (A) we could easily formulate immediately the
complete analysis in matrix notation. However, since the basic ideas do
not appear to be generally known we think it preferable to develop them
first in the more standard form.
Consider a structure subject to arbitrary external loads R, temperature Excus length of typical bar
strains a0 and any other initial strains 7J· We assume that the system has 11 over correct length I
N, • Tension 1n sam• bar du• to X,• I
internal or external redundancies
xi, X 2 , ••..•. , xi, ...... x,
which may be stresses, forces, moments or linear combinations of such
(generalized forces). By including the supporting body-assumed rigid-
in our structure we can denote all redundancies as internal. The stress
Contribution to 610 due to give A at
suppolt C and excess lengths 1!.1 of bars ....
C,A • l: N,AI

distribution in the body remains statically indeterminate until an elastic
analysis yields then unknown X's. If, irrespective of compatibility we assume
the X; to be zero, we obtain the 'basic' (principal or null) system which is
statically determinate. This procedure of obtaining the basic structure
may sometimes be identified with the process of an actual physical cut of
redundant members (e.g. of bars in a redundant pin-jointed framework).
However, the simple idea of a cut is not always applicable to continuous
structures typical of aircraft. We discuss this point later but for the sake
of linguistic simplicity continue to use the expression 'cut redundancy'.
Let the stress-system in the basic system be denoted by
ao
It must obviously be in equilibrium with the applied loads. We describe
it as a 'statically equivalent stress system', thus drawing attention to the
fact that in its determination only statical conditions enter. We find also
in the basic structure the stress systems
a 1 ,a2 , •••••• ,a;, ...... all
due to
XI= I, X2= I, ...... , X;·~ I, ...... , X,'= I
respectively. The systems a I, a 2, ...... , a;, ...... , a" are obviously self-
equilibrating. Since our structure is by definition linearly elastic the true
stresses a in the uncut original structure can be expressed as
a a 0 + ~a;X; ........................ (176)
i=-1
Similar equations may be written down for stress resultants (forces or
moments). Thus, the problem reduces to the determination of the X's,
which as already mentioned, need not be simple forces or moments but can
be linear combinations of such (generalized forces).
The Equations in the unknown X.
We define the following set of deformations in the basic system.
O;. Relative movement of ends of cut ith redundancy due to all
external causes, i.e. loads, temperature changes, lack of fit, 'give'
at the supports, etc.; i =I to n.
O;k=Oki relative movement of ends of cut ith (kth) redundancy due to the
self-equilibrating load system Xk = 1 (X;= I); i and k take values
1 to n. The 8-coefficients are taken positive if the relative move-
ments are in the positive direction of the X's.
Fig. 28.-Singly redundant, pin-jointed framework. Contribution tu 010
from sinking or 'give' of support and excess lengths of bars
The 8;, are, of course, the influence or flexibility coefficients of the basic
structure for the directions of the redundant forces. We use here the symbol
8 for these flexibilities since it is standard in the literature. To calculate the
8's we apply the unit load method of Section 6D.
Thus, using again the abbreviations
aE=auExx + ..... . +UzxEzx
} .................. (3)
a7] ~--au7]xx+ . . · . · · +azx7)zx
we find from Eq. (71a),
0;,= fa;(E,+7])dV ............................. . (177)
v
8;;'~Ja;E;dV, o,k=fa;EkdV= fakE;dV=8ki ................ (178)
v v v
where ai, E; (ako Ek) are the stresses and strains corresponding to X;= 1
(Xk=1) and a., Eo are the stresses and strains due to the applied loads.
Eqs. ( 178) reproduce, of course, merely Eqs. (84) for the flexibility co-
efficients. The total initial strains 7J imposed upon the basic system may be
separated into thermal and other strains
7J.•·x=a0 +T}xxo• • • • • • · , T}xu =7]xuo• • • • · • • · • • • • • • • · (179)
where T}xx•• etc. are initial strains due to say lack of fit, 'give' at the supports.
The effect of the latter upon 8; 0 is best considered separately and expressed
in terms of the imposed changes of length (rotations) and 'gives'. Consider,
for example, the singly redundant framework of FIG. (28) and assume that
the manufactured length of the bars exceeds the correct length I by 1:11. Let
also each bar be subjected to a different thermal straining a0. We assume
furthermore that the intermediate support gives or sinks by the amount /:1.
As redundancy we select the force XI in the bar (1, 2) and denote by N_9
and NI the (tension) forces in the bars of the basic system due to the appliea
loads and X1 = I respectively. The loading case X1 = l, with the correspond-
ing force applied to support C by the structure is shown in FIG. (28).
We find immediately
NI2f
8u = ~ AE ............................ (180)

• See also footnote, p, 17.

27
Applying now the unit load method to the state XI= I and the true total Consider a system of n independent equations
strains (e.+'l) and displacements in the basic system we derive (see FIG. aX=Z .............................. (186)
(28) and Eq. (7la)),
0 NIN• h (
l·810 =~NI(e.+'7)/=~ AE l+~NI(a0!+~l)+a;: I +11 ~ .... (181)
. a) where a is a square matrix and Z is a column matrix. Thus,
I I
Naturally, we can alternatively deduce by kinematical reasoning the con- au · · · · · ·at; · · · · · · al" I
tribution to 810 of the initial elongations a0 and ~/ and the 'give' ~­ a=
------------ li
ai1 ••••• • aii ..... . ain .............. (187)
However, the unit load method yields the results much more conveniently
and systematically. -- --- -------
an 1 ••••.. ani . . . . . . ann
More general formulae for the 8 -coefficients are given further below. L
1
_j
The condition of consistent deformations at the cut ends of the n re-
dundancies in the actual structure or application of the unit load method To eliminate in the first column of a all its elements but au and to reduce
yields the following n equations in the n unknown X. • the latter to l we premultiply a with the matrix
811X1 +812X2 + ...... +81;X;+ ...... +81 n X.+810 =0
rl- 1/au o...................... ~I
821 X1 +8 22 X2 + ..... +8 21 X;+ ...... +8 2 .. X,.+8 20 =0
I -a,.!au I 0 ................... 0 I

8,.1X1 +8,.2X2 + ...... +8,.1X 1 + ...... +8,.,.X,.+8,.0 =0 .. (182)

The solution of these equations determines the X and hence also the total
stress distribution after substitution into Eqs. (176)
To solve Eqs. (182) by elimination • is particularly simple when they are I
of the three-moment or five moment type; see for example the tube analysis
in Section 9(b). In general, however, all unknowns may appear in each I
equation and we need a systematic and mnemonic method for the deter-
mination of the X. The most convenient method for this purpose is the o........ o I .... (188)
shortened elimination process of Gausst (known also as Gaussian al-
gorithm). This method is so well known that we need not discuss its
formulation in the present 'long-hand' notation but may use immediately
the matrix notation. Accordingly, we write the system of Eqs. (182) in
the form
DX+D0 =0 ........................ (182a)
where D is the symmetrical square matrix of the 81k-cocfficients, D. the
column matrix of the 81.-coefficients, X the column matrix of the unknown
X 1 and 0 a zero column matrix. The Gaussian elimination process reduces
the system (l82a) to
-a.1 (a 11 0 .................. 0 l
TX=T0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (183) L _j
where T is a triangular matrix, i.e. a matrix whose elements above (or which square matrix has, but for the first column, unit diagonal and zero
alternatively below) the principal diagonal are zero and T 0 is again a

l
cross-elements. We find,
column matrix. For example,
I I l ht2 • • •• • .bli
• • • • • .bln
t 11 0 ...................... 0 0 b22 ...... b2i •..•• . bz ..
ft2 122 0 .................... 0 b=M 1a= [ - - · - - - - - - - - - - -
0 b12 ••••• • b;; ..... . b;,. · · · · · ·.... (l87a)
- - --- -- - - --- --
0 bn2••••••bni••••••bnn
L _j
where
b1 k=a1 k Ia 11 an dbtk=a 1k -a;tatk
-- .............. (189)
au
Next we eliminate the elements of the second column of b, except for b"
which we reduce to l, by premultiplying with <2

T= 1;1 t; 2 ••••••••• • t;; 0 ...... 0 .. (184)

I
1 0 .................. 0
0 0 .................. 0

111 ; l 112 •••••••••••••••••••• Inn I

L _j

The solution of Eqs. (183) and (184) is straightforward by substitution,

starting from the first of Eqs. (183) or in matrix language it is simple to
find the inverse T- 1 ofT and write 0 ...... 0 .... (186a)
X=T- 1T0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . (185)
For the automatic computation techniques now available and, in particu-
lar, for the punched card machines it is usually preferable to modify
slightly the Gaussian elimination process and obtain directly the inverse
of D and hence also the column of X. This method is known as the Jordan
technique and we restrict here our discussion to this process.

• Natur.ally we may .also use iteration techniques but such methods are n3t discussed in this series
of papen.
t See: Simultaneous linear Equation• and the determination of the eigenvalues, National Bureau
of Standards, Applied !t'fath. Series 29 (1953), and 0. He~k: 'Ueber ~en Zeitaufw'!J'd fiUr das
Bereclmea von Detemunanten und fuer das Anftoesen von lmearen Gletchungen.' DISsert. Techri. 0 ~b,.2(b22 0 ................ 0 1
Hnchsch. Damutadt, 1946. Zurmuehl, loc. cit. L _j

28
We obtain
I I I~ = 1/El
1 0 Cta •.•..•••.•. . Cln I
0 1 c23 •..•...••.• . C2n I
----- I
C= 0 0 C;a ............ ci, I (187b)
- --- -
••• 0 ••••••

L
0 0 c,. 3 ........... . c.,.
_j
I (a)
where
b;2b2k 9
f X, fx,
c2 ~~:=b2 ~~:/b22 and c;k=bik-~ for i*2 .............. (18 a)
(b)
The procedure will by now be clear. Thus, at the (g-1) 1h = f1h elimina-
tion step we obtain a matrix g of the form
I I
1\
fx. t X,

g= G .............. (187c) ;\ I~" (c)

Y, Y,
0
L _j -!'/- -.,_x%: (d)
6./<C"-----;d ~-----zLs -----..2':.
where I is the unit matrix with f columns and 0 is a zero matrix with f
columns and (n-/) rows. G is a rectangular nx(n-/) matrix. For the Fig. 29.-Continuous beam. Good and bad choices of basic system and
next step, i.e. to obtain the h matrix, we premultiply with M. which has a redundancies
gth column
is necessary to avoid serious accumulation of round-off errors in the most
{ -gl.fg•• -g2ofg••. 000000.1/g••. 0000000-g,..fg•• } (188b) important digits. For if aii <,a;k the limited number of digits of the machine
and otherwise unit diagonal elements and zero cross-elements. The result- is not sufficient to ensure reliable computations. Thus, requirement(l93)
ing h-matrix is of the form is seen to be essential for well conditioned equations.
Our Eqs. ( 186) are also ill-conditioned if two or more columns are
I nearly linearly dependent, e.g.
1 0. 0000.0
0 l. 0000.0 {a1 ;a2 ; . • . . . . a;; ..... . a,;}~C{a1 A:02 k· •... . akk· ..... a,.d .. (194)
The one diagonal element g •• will inevitably become very small and grave
errors will again arise. Naturally, in actual structural problems two
columns could never be exactly linearly dependent for otherwise this
would indicate that we overestimated by one our number of redundancies.
Nevertheless, a bad choice of redundant forces or moments may give an
approximate linear dependence which would yield a result grossly in error.
h=M.g= ............ (187d) If it is found that our initial choice of redundancies leads to an ill-
conditioned set of equations then we can always obtain more suitable
equations by introducing as unknowns appropriate linear combinations
Y of the initial unknown X. Such a transformation may be represented
always by
0 0 .......... . hAll. ..... . hAn X=BY .............................. (195)
where B is a non-singular square matrix n x n. If X are initially single
0 0 ........... h,.A ..... . h,.n forces or moments then Y represents groups of forces or generalized forces.
L __l Such groups of forces were first introduced by Mueller-Breslau• guided
where by pure physical reasoning and this is still the best method of finding them.
Kk•g•"' The transformation (195) may be introduced directly into Eqs. (186a)
h•m=K.m Ig•• and h~~:m=g~~:m--g-- for k*g .......... (189b) yielding
•• DBY+D.=O
and so on until the last premultiplication with or
I (196)
:-n1 n/n,.,.
where D is determined by matrix multiplication from
D=DB 00000000.. 000000000000000000 (196a)
Physically, Eqs. (196) express the compatibility conditions at the cuts
of the original unknowns X in terms of the new unknowns Y. If the trans-
1/n,.,. formation matrix B is unsymmetrical then the resulting matrix will also be
•• 0 ••••

0
L __l unsymmetrical. Although, in general, the simple substitution of (195) into
(186) can lead to a slight improvement of ill-conditioned equations the
(where I has (n-1) rows) gives effect is usually small. The next obvious step is to express the compatibijity
M,.M,._1 . • . . . . M; ...... M1a=l .................. (190) condition (186) in terms of the generalized displacements at the cuts
Eq. (190) then yields the inverse matrix corresponding to the generalized forces Y of ( 195). Following our dis-
cussion on generalized displacements and ftexibilities on p. 19 the general-
a-1 =M,.M,._1 . . . . . . M; ...... M1 . . . . . . . . . . . . . . . . (191)* ized compatibility equations are derived by premultiplying Eqs. (196) by
From which we find our unknown B' as
X=a- 1 Z ............................ (192) B'DBY + B'D. ~. 0
or
In practice it is usually preferable not to determine a- 1 explicitly but to D.Y+D •• =O (197)
perform first the multiplication M1 Z and continue then premultiplying to where
find X directly. D.=B'DB and o •• =B'D•........................ (l97a)
It is apparent that if in the above procedure any hM becomes zero the It is evident that the column matrix o •• and the symmetrical matrix D.
elimination process cannot b: continu~d. An interchange of rows is in- would be obtained directly by selecting ab initio the generalized forces Y
dicated but this is obviously inconvenient for automatic computation. as redundancies and deriving the corresponding S;. and S;k from the
Clearly, if standard formulae given.
a;;>>a;~~: In many cases it is best, when the equations are ill conditioned, to select
no hM can vanish. Moreover, the condition a different basic system and corresponding redundancies Y. Naturally,
(193) the latter are again statically related by a transformation matrix B with
any previous choice X of redundancies.
• As a matter of interest we point out that tba actual operations on the dieital computer to obtain
a-• do not follow exactly the typical matrix multiplication rules as implied by formula (191). • Mueller·Breslau, lo<'. dt. p. 17.

29
 We illustrate the above considerations on ill-conditioned equations on a
very simple structural example. FIG. (29) shows a uniform, continuous
beam of three equal spans simply supported at A, B, C, D-a twice re-
dundant system. We discuss four alternative choices for the two redun-
dancies. ·
(a) X1 and X 2 are taken as the reactions at the supports A and B. The
D matrix for this system is
1: I I I
3u 312 24 9
D=
9 4
L _j (a) (b)

A remarkably bad choice since 312 > >322 Non-orthogonal redundancies Orthogonal redundancies
(b) X1, X 2 taken as the reactions at the intermediate supports Band C.
Then,

7l
System Y1 = 1: "'1 =1• N1 = 0, s1 = 0

_aaq, ~~ 8 System r, =1 : M, • Y, N, = coso&, s2 = - sino&

D- 18 7 8
System r, = 1: M, • Y, N, =- sin-3, S, = - cos-3
L _j
Still a bad choice since all 3's are of the same order of magnitude.
(c) Y1 and Y2 are generalized redundancies formed from system (b) by t,. t, = [ :Z' ds}
r:r·
the transformation matrix = 0 for elastic centre C.

B= Ill -0·5
-0·~
1 I .................. (198)
t13 = t" =

t,= ,.r , = f EI
ds

Y,Y, ds +
f 1 1 )
sin-3 cos-3 (&A' - EA ds
L _j
The D. matrix may be obtained either directly or from Eq .. (197a) and is
I I = 0 for "principal axes" Cy2 , Cy3 •

4 1
Fig. 30.-0rthogonal and non-orthogonal redundancies-elastic centre of
singly connected ring
1 4 we restrict ourselves to bending deformations and assume the E.T.B. to
L _j
hold, an orthogonal set of redundancies is obtained by referring them to
The improvement over (b) and (a) is immediately apparent. the principal axes (Cy2 Cy3 )of the ring neutral axis weighted with the ring
(d) X1 and X 2 are the bending moments at supports B and C. This flexibility c/J =I/ El (see FIG. 30). The origin C of this system which is, of
choice of unknowns is statically identical to Y1 and Y2 of (c), but the basic course, the centroid of the elastically weighted ring is known as the elastic

tl
system is different. The D matrix is
I 4
D=~~ I 4
L _j
and clearly a scalar multiple of the D matrix of (c).
The final system (d) is recognized as a particular case of the well-known
Three-Moment Equation of Clapeyron. Since the basic system approxi-
mates more closely to the actual system than that of (c) it is clearly the mosl
suitable choice of all.
The differences between the above four systems become even more
pronounced when the number of spans is increased. Moreover, the 3w
coefficients of choices (a) and (b) tend, for a large number of spans, to
become linearly dependent.
This discussion shows how important the choice of the redundant
forces is for the convenient numerical solution of a problem. An extreme
case of simplicity is achieved if in all equations only one unknown appears.
The particular set of redundancies
Y1 , Y 2 , Y3, - - - - , Y;,---- Yn
for which this condition is satisfied is called orthogonal. Then all but the
corresponding direct influence coefficients are zero, i.e.
'ik
=0 if ;,*' k . . . . . . . . . . . . . . . . . . . . . . . . . . (199)
where we introduce the symbol ' for 3 to emphasize the special nature of
this system. Eqs. (182) take now the simple form,
';;Y;+,;.=O for i=l ton ........................ (200)
and hence
Y;=-~ for i=l ton ............................
~ii
(200a)
This system Y; may always be obtained by a particular linear transforma-
tion
X=B,Y .............................. (201)
However, the computations involved in determining B, are usually more
laborious than the direct solution of Eqs. ( 182) if these are well conditioned.
Nevertheless there are structures in which it is simple and hence advan-
tageous to find the orthogonal set of redundancies. This is particularly so
if physical and not mathematical considerations indicate how to find them.
For example, this is so in arches and singly connected rings where, if
centre. The transformation matrix B relating the orthogonal set of re-
dundancies Y to a set X consisting of bending moment X1 , normal force
X 2 and shear force X 3 at some point is in the notation of FIG. 30,
a~~-~ Ya Y2-
I
II
cosO -sin~ .............. (202)
I 0 -sin 0 -cos()
L _j
In practice it is nearly always worthwhile to find the elastic centre and
eliminate two of the cross-flexibilities but determination of the principal
axes, unless possible by inspection, is not usually worth the trouble. A
further point is that the elastic centre concept is still valid if deformations
due to normal and shear forces are included whereas the principal axes
requirement becomes more complicated.
Interestingly enough this solution was first given by Mohr* more than
seventy years ago, but it appears not to be universally known, for otherwise
it would not have been necessary to rediscover it so many times. A more
recent derivation and application of an orthogonal set of redundancies
(in general infinite) is the system of self-equilibrating eigenloads developed
by Argyris and Dunne t for their general theory of tubes in bending and
torsion.
Particular forms of the 3;k and 3; 0 coefficients
We return now to Eqs. (177) and (178) for the 3-coefficients and give,
following our expressions (180) and (181) for a pin-jointed framework,
some further explicit formulae for more complex structures.
Stiff-jointed plane framework. We assume the Engineers' theory of
bending stresses to be true and introduce the special notation:
N., s., M. normal force, shear force, bending moment in basic system
due to applied loads.
N;, S;, M; normal force, shear force, bending moment in basic system
due to X;= 1 where i= 1 ton.
s coordinate along axis of beam.
0 temperature at neutral line of cross-section.
D.0/h temperature gradient across depth h of beam ; positive if giving
rise to thermal bending strain of the same sign as that due to
positive bending moment M.
• 0. Mobr, Z. Architek u. lng. Ver. Hannover, Vol. 27, 1881, p. 143. See also the 1enenlization
and tabular presentation of this method in: J. H. Aqyris and P. C. Dunne, Structural Analysis
(Vol. I of Handbook of Aeronautics), Pitman, 1952, Table 17.1. Both the deformations due to normal
and shear forces are included in the latter analysis.
t See J. H. Aqyris, P. C. Dunne, 'The ~ral'lbeory', ete., J.R.Ae.S., Vol. LI (1947), Feb.,
Sept., Nov. and Vol. Lm (1949), May, June.

30
 r·~·l
tx,
-
r ·'
II
X1 ~~X 1 o..- -~ ::-=t::A.
I
I

A.
I LI
\~!= :
o., -o ..

JJ::h=_
-1
·· -1C::::====T=:r
.
N •..!..
tx, • 1
N
a
=-

;-~::.::.:~~-·-~~A.
h

c,IRJ
a '

:~ . :fc. •
1I N, • -h
X2 •1
I~
'1
1
1 r - -,
Correct shape

U
a~
I I
of panel

J
L - _I
I
n~c · h
1
I
A1
AL :_ ----- ~ tJ Manufactured
shape of panel

a life,.= -(a+ b)
c.. • =-
h

A.~ 1- X1 X, X3 Redundancies

Prescribed shearing deformation due to

LOADING X, • I LOADING X2 • I incorrect manufacture

C11 Shear flow, due to X1 =I, acting on the element

which experiences the A , in the direction of this A
N, Normal force in horizontal member due to X, =I

A. A1 A0 A. Prescribed deformation ("give") c, = T1 . c, • - TI . c, = o

c,. c,. c,. c,. Force (moment), due to X, sl, acting on the element Contributions to 0 10 due to A
which experiences a A, In the direction of this A

Contributions to 0., and 0 20 due to prescribed displacements A

0,0 = fe,Adf, =-f-A, o,. = 0
0., • IC1A • ~ A.+ o.A1 + fA• + o.A. • *A• + fAn Fig. 32.--3; 0 due to initial shear strain arisin1 from incorrect manufacture
020 • Ic,A • ~ A.- (a+ blA1 + fA.+ t.A. c ni• cti direct and tangential forces per unit length g in the basic
system due to X;= I acting on the element which experiences
Fl1. 31.--3~ 0 due to initial strains In ri1id jointed frame and manufacturin1 a a in the direction of this ~.
errors and 'live' at supports t thickness.
We deduce from Eq. (I 78),
prescribed relative displacements (linear or angular) either
inside the structure (e.g. lack of fit) or at the supports ('give'
of foundations).
S;k =i:J J[ a,,;O"xxk +a•• ;O"vvk -v(a,,;O"vvk +a••;O"xzk)+ 2(1 +v)a,.;a,.k Jtdxdy
force or moment in the basic system due to X;= I acting on an .................. (205)
element which experiences a ~ in the direction of this ~. Also from Eq. (177),
EA, GA', EI direct, shear, bending stiffness of beam.
We deduce immediately from Eq. (I 78), see also Eq. (94),
~
a;k=
JN;Nk
EA ds+ JS;Sk
GA'ds+ JM;Mk
-ads .................... (203)
............ (206)
Also from Eq. (177) or directly from the unit load method,
S;.= JN;N. J
a~0 +l::C;~
EA + N;a0ds+ M;-h- J ................. . (204)
The immediate application of the above formulae is to major aircraft com-
ponents like wings and fuselages. Their matrix formulation is discussed
further below.
For pin-jointed frameworks we omit the terms involving M and S. On the FIG. (32) illustrates how on a thrice redundant beam with shear carrying
other hand in stiff-jointed frameworks we can, in general, omit the terms web the contribution of a prescribed displacement ~ to S;. is calculated.
involving Sand often also the terms inN. FIG. (31) illustrates on a twice ~ is in this case an initial shearing displacement of a panel due to error of
redundant system how the contribution of the prescribed displacements manufacture.
a to S;. is calculated. X1 and X2 are the chosen redundancies and ~"' ~b. It was assumed in all our above considerations that the basic or cut
a. and ~ are linear or angular imposed displacements arising from errors system is statically determinate. However, nothing in the theory so far
given restricts us to such a choice. We can select in a structure with a total
in the manufacture of the frame.
number of redundancies n a statically indeterminate basic system with
Two-dimensional stress distribution (n-r)(r<n)redundancies by'cutting'only rredundantmembers. Equations
We restrict ourselves here to a presentation in cartesian co-ordinates of the type (182) can then be written down for the cut r unknowns, the
x, y but the formulae are not restricted to stress distributions in flat plates. corresponding S;k·coefficients being still defined as in our previous analysis
They are applicable to stress-states in any curved membrane which, by in the basic system. In fact, to calculate the S;k we may apply again
definition, does not allow for any variation of the stress over the thickness. Eqs. (I 78). Similarly, for S;. we may use Eq. (177) if we substitute e'l for 7J
Hence, x, y are, in general, orthogonal curvilinear co-ordinates; for ex- where e'l is the true strain in the basic system due to the prescribed initial
ample, in a cylindrical membrane y may be measured along the generator strains. This modification is necessary since the basic system is now re-
and x along the periphery of the cross-section. dundant and the imposed initial strains 7J are not free to develop. However,
The following special notation applies. both formulae for S;k and S;. may be simplified considerably if we remem-
~ ... at prescribed direct (n) and tangential (t) relative displacements ber that in the unit load method from which they derive (see Eq. (7la)) only
either inside the membrane (lack of fit of parts, slip of rivets) the strains must be the true ones for the system considered-in the present
or at the boundaries ('give' at supports). ease the redundant basic system. The stresses corresponding to the unit
g distance along part which experiences~ .. and/or~~ load may be any suitable statically equivalent stress system and may hence

31
be found in the simplest statically determinate system. Thus, if we introduce
the notation
a;=Statically equivalent stress system in redundant basic system due
to X;= I
we may write,
3; 0 =J(1;(t: 0 +t:,1)dV .................. (177a)
v
3,,=fa,t:;dv. 3/A= fa,t:kdV=Jah-t:,dV=3k, ................ 078a)
v v v
The introduction of a,· instead of a, in Eqs. (177) and (178) may shorten
the analysis greatly. Naturally, Eqs. ( 178a) are again identical with formu-
lae (93).
The above method presumes that the strains e and e,1 in the redundant
basic system are known. Such information may be available either from
previous calculations or the literature. Alternatively, we may have to
analyse first the basic system for the external loads (and/or imposed
strains) and the r X,= I by the method given previously. From a
~
' x,-
X,
f x;
~J
mathematical point of view the selection of a redundant basic system
means that we solve the problem of 11 equations with n unknowns in two
steps involving respectively the solution of r equations with r unknowns
~x~~x:~
and (11-r) equations with (11 -r) unknowns. This method is particularly '~
useful if we have available information on the stress distribution of the
redundant basic system or if the number 11 is very high. (a) (b)
Consider, for example, FIG. (33) showing a fuselage ring with transverse
beam under uniform load p. The loading is equilibrated by tangential Statically determinate R•dundant basic
shear flows q applied by the fuselage to the ring. The structure is six times basic sys1Pm syste-m
redundant and as redundancies we select the two groups X 1 , X2 , Xa and
X 4 , X.,. X6 at the intersection of the axis of symmetry with the upper part Fig. 33.-Statically indeterminate basic system. Doubly connected ring
of the ring and the transverse beam. Due to symmetry of loading and
structure
X:1 =X6 =0 It is useful to write down the 3,k and 3," for a two-dimensional stress
distribution when the basic system.is redundant. Thus, with the definition
and hence the problem reduces to finding the remaining four redundancies. of a, given previously we find,
We may solve the system by direct application of Eqs. (182), which in the
present case take the form, 3,. ~-i: JJ[a~,.,a_,.,k +a.",;a•••k -- v(a_,.,.,a••k +a••;a.•. •. k) +2( I +vla.ruta.r•k] tdxdy
S21X, +322X2+324 X4 _L32r, Xr, +32o=O
S41 X1 +3 42 X2+3 44 X 4 ~'-3 4 ,,X,, +3 40 =0
1
311 X1 +3 12 X2 _L3u X 4 ..:..3~;; Xr, +3 10 ~~o
r ................ (207)
............ (205a)
and a similar formula with a and a interchanged. Also,
3,,1 X1 +3r,2 X2 -+-3,, 4 Xt...:. 3,;r,X,-, +3r, 0=-0
where the 3,k and 3,, are calculated with formulae (203) and (204) in which
3,, =iJ I[ a_,._,.,a_,_,.,_L(],,,,;a.,,,, --~·(a,. ,;CJyyo·H1•• ;a_,.,..)+2( I +v)a.ryiCJ.ryo]tdxdy
the integrals extend over ring and transverse beam. Note that if the un- .................. (206a)
knowns X 1, X2, Xa are referred to the elastic centre of the ring the co- where a_,.,., etc., may be here not only the stresses due to external loads but
~fficient 312 vanishes. Having solved Eqs. (207) we find N, S, M in the also due to the imposed initial strains YJ· (Remember that the latter cannot
actual structure from, develop freely in the statically indeterminate basic system and give rise
N ·- N0 +N1 X1 _t_N2 X2 +N4 X 4 +N;;Xr, to some strain-stress state e,,, a,,)
S=· S0 +S1 X1 +S2X2+S4 X4 +S;X; } ................ (208) The use of a redundant basic system arises continuously in wing theory.
M=·M 0 +M1 X 1 +M2 X2 +M4 X 4 +M5 X 5 Thus, following Ebner and Koller* and Argyris and Dun net it is customary
where M 0 , M 1 are defined on page 30. Alternatively to this standard in wing analysis to express the actual stresses in the form,
method we may solve the problem by cutting only the beam at the axis of a=a.+a. . ........................... (213)
symmetry. Then the structure is only twice redundant (X4 and X5 ) with a where the stress system a -the choice of which is at our discretion-
0
basic system that is itself twice redundant. We denote by satisfies both the external and internal equilibrium condition and is
m,H nfa So therefore the statically equivalent stress system, a, are the self-equilibrating
m 1, 11 1, s1 (i =·4 or 5) stress systems (in general, infinite in number), necessary to ensure external
the normal force, shear force and bending moment in the new basic system and internal compatibility. In our present terminology a. is the basic
due top and X 1 =I respectively and assume that they are known. stress system and a. the redundant stress systems which for practical
The stress distribution in the actual structure is then found from purposes are approximated to a finite number. In fact,
M=m 0 +m 4 X 4 +m;X5 I a,=a1 X 1 +a2 X2 + . . . . +a,Xn=:Ea;X;
n
.......... (214)
N=n 0 +n 4 X 4 +n 5 X 5 ~ ••.•••.••••.••.. (209) i=l
S=s0 +s 4X4+s 5 X,; j
The equation of compatibility in the unknowns X 4 and X 5 are now of the It is advantageous in the selection of the basic system a. to try and satisfy
form the two, at times conflicting, requirements of simplicity and not too great
difference from the exact system a. For it is obvious that small a,-systems
'uX4+,4aX5+,4o=--O } ................ (210) are highly desirable from both the theoretical and practical point of view.
';4X4 +';;;X5 +'•o=O Now for wings with not too small an aspect ratio an excellent choice for
where we write ' instead of 3 to stress that these coefficients are different the direct stresses of a. is given by the Engineers' theory of bending for
from the correspondrng 3's in Eqs. (207). To find the ''s we apply Eqs. beams since it combines simplicity with reasonable accuracy. If the wing
( 177a) and ( 178a) in the new basic system and remember that the virtual forms a single cell tube we deduce the shear stresses from the boom load
stresses due to X1 =· l may be selected in a statically determinate system. gradients, the undetermined constant of integration being found from the
Thus, omitting the contributions of the normal and shear forces for overall torque equilibrium; thus, in this case the basic system a. is statically
convenience of printing, we find determinate. If on the other hand the wing is an N-cell tube we see that
I
m 1mk JM;mk Jm;Mk r
,,k= -----erds= El ds= ~ds=~:,kt .......... C2ll)
whereas there is still only one torque equilibrium condition there are N
undetermined constants of integration. To calculate them we must intro-

' "= Jm~7"ds= JM~7°ds

duce conditions of deformation and those are the equality of rate of twist
of all cells. Hence, our basic system is redundant, the degree of redun-
.......................... (212) dancy being N -l. The solution to this problem is reproduced in Example
(a) of Section 9. General considerations on the calculation of the redun-
If, in addition, any initial strains YJ are imposed on our structure giving dant self-equilibrating stress systems a. are given later in this Section in
rise to moments m'l in the basic system the corresponding contribution matrix form (see also Example (b) of Section 9).
to ''" becomes The example of the tube is useful also to illustrate another point. We
r
<..to= Jm;m'l
El ds= JM;m'l
~ds ....................... . (212a) stated on p. 27 that the basic structure is obtained by cutting redundant
members but mentioned that in continuous structures the idea of a physical
The effects of the normal and shear forces may be included without diffi-
culty, the terms following immediately from Eqs. (203) and (204) and the • Loc. cit. p. I.
arguments leading to the contribution of the moments. t Lo..-. cit.
p. 30.
cut is not always applicable. Thus, in the case of the tube in the last
paragraph, when obtaining the basic stresses we do not actually cut any
redundant member but rather select the engineers' theory of bending direct
stresses and the associated shear flows as statically equivalent to the
applied bending moment. In general, all members are found to be load
carrying. The stresses a. only exceptionally satisfy the elastic compati-
bility conditions-for example, due to warping varying parallel to the
axis of the tube and to rib deformability. We may give some physical reality
to the basic structure in which a. is true by releasing in the actual structure
the warping restraints at every cross-section and by assuming the ribs as
rigid; the former idealization does no doubt require a complicated mech-
anism for its realization. The idea of selecting a. as any suitable statically
equivalent stress system without reference to actual cuts may, of course,
also be applied to frameworks.
The use of a redundant basic structure is important also from a further
point of view. Consider the wing of FIG. (34) the main portion I of which is
swept and attached to some root structure II. It is assumed that the ribs
of I are taken perpendicular to a longitudinal axis approximately parallel
to the spars. The necessity may arise of investigating alternative angles of
sweep obtained merely by changing the root structure II. Thus, in FIG. (34)
we show two alternative arrangements. In such instances it is obviously
advantageous to have as much as possible of the stress analysis in common
in the two alternative calculations. To this purpose we release at the junc-
tion ofl and II all redundant forces or groups of forces X 1 to X,. appearing
there. The tube I is then connected to tube ll by some statically determin-
ate arrangement and this new structure is taken as the basic system. The
scheme of the analysis is as follows. Analyse first Tube I for all external
forces and also for X 1 = I to X n = 1 respectively. This investigation in-
volves, of course, the solution of a highly redundant system. Irrespective,
however, of the form tube H takes the analysis of tube I remains unaffected.
Next we analyse tube II for the external forces (which include the statically
determinate reactions P from tube I), and also for X 1 =I to X n = 1 re-
spectively. Again this may involve the solution of a redundant problem.
Finally we can write down equations of the type ( 182) for the unknowns
X1 = 1 to X n = 1 and note that the S-coefficients are in each case
8=81 +8u . ........................... (215)
Hence if we change structure II only S11 but not S1 is altered-an obvious
advantage. The solution of Eqs. (182) yields ultimately the stress distribu- Fig. 34.-Aiternative root structures for swept-back wing. Redundant
tion in the actual wing. basic systems

33
H AVING discussed in the standard longhand notation the main
ideas and methods for the calculation of redundant structures on the
basis of forces as unknowns we now turn our attention to the
matrix formulation of the analysis. Consider a system consisting of s
structural elements with a total number n of redundancies which may be
forces (stresses), moments or any generalized forces. We select a basic
system by 'cutting' a number r of redundancies where r<n. Thus, the
simple idea of a statically determinate basic system (r=n) is but a particu-
lar case of our investigations.
Example for the b0 and~ matrices.
~10. (35a) shows a five times redundant structure, assumed symmetrical,
subJected to the loads R1 and Rt· Due to symmetry of loading and struc-
ture the system is effectively three times redundant. For the basic system
we sc:Iect the statically determinate structure of FIG. (35b). The b0 and b1
matnces for half the structure including the central vertical member (11)
are found easily as

>;-I l
The structure is assumed subjected to a system of m loads (generalized 0 -a/h a/h afh 0 0
forces) -d/h 0 0 0
R={R1 R2 •••••.. . R; ....... . .R,.} ................ (90) 0 -a/2h a/2h afh 0 0 -d/2h-d/2h 0 1/2 1
We denote by X the column matrix of the r cut (unknown) redundancies, L

l
.J
I 2 3 4 5 6 7 8 9 IO 11
X={X1X2 •...•.• • X, ....... . X.} .............. (216)
(218)
The column matrix S of the stresses and forces in the actual (uncut)
structure can always be written in the simple form and
5=b0 R+~X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (217)
where b0 and b1 are rectangular matrices with m and r columns respec-
tively and the sante number of rows as S. In fact, the elements of b1 are or
~-afd 0 -afd
.,;= 0 -a/d 0
0 0 0 -hfd-hfd 0
correspond to the stresses a 1 given previously (see Eq. (176)). If the basic -a/d 0 0 0 -h/d-2h/d
system is statically determinate the two matrices b0 and b1 are found 0 0 -1 -1
merely by statical reasoning. 0 0 0 0 0 0 0
L .J
For a redundant basic system we may obtain the necessary data either I 2 3 4 5 6 7 8 9 IO 11
by analysing it first for the loads R and the r forces X,= I or in many cases (219)
by using existing standard information.
~Jb.re the numbers under the columns refer to the numbered bars of FIG.

Additional Notation
X,Y, Z column matrices of X,, Y1, Z 1 respectively.
H column matrix of initial strains (displacements).
c rectangular matrix of forces (moments) C1 (seep. 31).

B,C areas of actual longitudinal and transverse flanges.

B., C. areas of effective longitudinal and transverse flanges.
1 length of longitudinal flanges between nodal points.
d length of transverse flanges between nodal points.
h height of web.
0 area enclosed by cell.
~. ~.. areas of cover panel and web panel respectively.
L1,, 11,. 2 x 2 matrices.
L=[L1,], 1=[11,]
P, Q effective longitudinal and transverse flange load respectively.
k., k.,, k1 Partial stiffnesses due to shear strains in sheet, direct strains
in sheet and direct strains in flanges respectively.
U, W column matrix of kinematically indeterminate joint displace-
ments U, W respectively.
8o column matrix of strain of elements due to unit r's and
U=O.
~. ii1 column matrix of actual and kinematically equivalent
(virtual) strain of elements due to unit U's when r=O re- X,
spectively.
C=i,.'k~ =~'k~
Ce=~'kaor=~'kaor
Flp. 35 (cr and b).-5tatlcally Indeterminate pin-jointed framework.
J column matrix of initial stresses. Illustration of b 0 and b 1 matrices
34
 Following our previous analysis of structures when the basic system
is redundant we introduce now the matrix
~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (220)
to denote any suitable statically equivalent stress (force) matrix corres-
ponding to
X={l 1. ....... 1. ....... I}

--
Thus, b1 corresponds to the stress system cr; introduced before. When the
basic system is statically determinate only one 61 can be found,
61 = b1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (220a) Complete structure Basic system
We define also by twice redundant )
60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (221)
any suitable matrix statically equivalent to
R={l I. ....... 1. ....... 1}
b0 may, in fact, be determined in a different statically determinate system
from b1 .
Next we derive the matrix equation for compatibility of deformation in
the actual structure. Denoting the relative displacements at the r cuts of
the basic system due to loads Rand the r redundancies X; by Vr the com-
patibility condition is
Vr=O . . . . . . . . . . . . . . . • . . . . . . . . . . . . . . . . . . (222)
where v, is a column matrix with r elements. To express Eq (222) in terms Ring element
of Rand X we note from Eq. (122) that the relative deformations v (these
may be elongations of bars or flanges, shearing angles of plates), at the
ends or boundaries of the s elements are,
v=f5=fb0 R+fb1 X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (223) R/2~ (12
f, the flexibility matrix of the s elements, is the partitioned diagonal matrix
of Eq. (123). We find now vr directly from the argument leading to ~J[~
Eq. (125) as x, X.
Cantilever beam elements
and hence Fig. 36.-Doubly connected ring. Analysis with redundant basic system
(224)
These are the required equations in the r unknown X;, and are, in fact,
equivalent to formulae (182). The symmetrical square matrix Simple example of Eq. (224)
D=b1'fb1 Consider the symmetrical fuselage ring with transverse beam and central
(to use the notation of Eq. (182)) is the flexibility matrix for the directions load R shown in FIG. 36. As in page 32 we select as a basic system the
of the r unknown X; in the basic system. Also in the notation ofEq.(l82a) structure with the beam cut at the centre. For the components s of the
basic system we take the two statically determinate cantilever beams and
D0 =b1'fb0 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . (225a) the closed ring. It is assumed that we know the stress distribution and
Eqs. (224) are the most general formulation in matrix algebra of the equa- hence the flexibilities due to the pairs ofloads applied to the ring (FIG. 36).
tions for the r unknown X; in a structure with a redundant basic system. The basic system is thrice redundant but due to symmetry X 3 =0.
Solving for X we find The load transformation matrices b0 and ~ are
X=- (b1 'fb1 )-1b1'fb0 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . (226) I I
Substituting (226) in (217) we determine 5 solely as a function of the R's.
Thus, _,boB
bu- I •• 0 ••••••••••••••••••••••••• (231)
bOR
(227) L _j
Comparing (227) with Eq. (121) we can write where
S=bR I I
where
(227a) boa=
0
1/2 bOR=
r,~~l •• 0 ••••••••••• (231a)
Naturally, it is always possible to substitute b1 ' for 61 ' in Eqs. (224) to
(227a). However, the introduction of the statically determinate matrix ~ 0 1/2
when the basic system is redundant simplifies the calculations, often L _j L _j
considerably. and
We can apply now Eq. (227a) to derive the flexibility

-l~l
F
of the actual structure for the m points and directions of the applied loads. bl- 0 •• 0 0 •• 0 •• 0 ••• 0. 0 ••••• (232)
Eq. (126) gives blR
L _j
f=b'fb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (126) where
Forb we may use
I I I

~~
b = b0 or even simpler b = b0 I1 01
1
We obtain
F = b0'f[b0 - b1 (61 'fb1 )-1 b1 'fb0 ]
btB= lo Ol b1R= 0 • 0 ••• 0. 0 0 ••••• (232a)
I
or IO I I 0 QlI
L _j L _j

where Note that in the present case b1 = b1 since the loadings on the two elements
of the structure are statically determinate.
fo=60'fb0 =b0'fb0 . . . . . . . . . . . . . . . . . . . . . . . . (230) The flexibilities of the elements for the forces and moments may be
is the flexibility of the basic system for the loads R. written as

35
 ~u '~lI
I [2 fu
EI -2EI O
[2 [3 fB= fm /22 /23
fB= -2EI 3EI 0 fal /82 faa
I
0 0 EA L ..J
L ........ (233)
and hence

f= ~B
L
0
ol
fR
..J
...................... (234)

It is now possible to write down the equation for X,

~'fb1 X+ ~'fb 0 R = 0
Returning to the general argument we investigate now in a system with
a total number n of redundancies the effects ofinitial strains e.g. those due
to temperature rise, excess length of bars due to manufacturing errors, Cross s~ction A-A
'give' of foundation at supports. Assume that the column matrix of the
relative displacements at the ends or boundaries of the s elements due to
71 is
H ························
(235)
We assume first that the basic system is statically determinate: then the
elements of Hare merely the integrated effect of the imposed 71· For exam-
ple, in a pin-jointed framework subjected to temperature rise the elements
of Hare
H ={(a0/) 0 • • • • • • • • • • • • • • • • (a0i),} (235a)
Fig. 37.-Aircraft wing structure. Geometry and definition
If the bars are of an excess length fl./ due to inaccurate manufacture, then
\.hese form directly the elements of H. Independently of the ,nature of H,
however, the corresponding relative displacements at the cut redun- and since x is usually small in comparison with X approximate methods
dancies are simply may often be used in the evaluation of the right hand side. This technique
b1'H
is immediately applicable to the correction of a completed stress analysis
and the equation for the n unknowns X is when subsequent small changes in the flexibilities of some elements are
b1'fb1 X+b1'H=0 . . . . . . . . . . . . . . . . . . . . . . . . . . (236) introduced by modifications to their cross-sectional dimensions.
When the number of equations is too large for performing the matrix
Note that in the present case b1 =b1 since the basic system is taken to be operation on a digital computer then we may apply the following method
statically determinate. When the deformations arise from p 'gives' fl. at which is basically identical with the idea of a redundant basic system.
the foundations it is advantageous to express H as Assume that in a structure with a number n of redundancies we select first
H ={!l.1 •.•••..••.•..••• fl.,} . . . . . . . . . . . . . . . . (235b) a basic system with t=n-r redundancies and that we write down the r
Then for ~' we must substitute the matrix c' where equations in X; in the standard form (224),
DX+Do=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . (224)

_·_·: ~: ~ ·_·_ :~~-~·-·: ~: ~ ·_ ~~"l

:~2
r----------------------
If the number r is still too large for handling by the digital computer we
solve the problem in two steps. Eqs. (224) are first put in the partitioned
C= C,l ........ C;, ........ C;p ....... 0 •••••• (237) form
c,.l ........ c,., ........ c,.,
L ..J (240)
the n rows of which are the forces due to X;= 1, applied by the structure
to the foundations in the directions of the gives fl.. Eq. (236) becomes now,
~'f~X+c'H=O . . . . . . . . . . . . . . . . . . . . . . (236a) where the number of rows in the matrices with suffices I and II is p and
r-p respectively and Db Du are square matrices. Eq. (240) gives
If we select a redundant basic system we cannot derive the elements of H
immediately from the prescribed initial strain since the latter are not free D/XJ+Du/Xu+Do/=0 I
to develop in a redundant structure. In this case unless we have the neces- ~ . . . . . . . . . . . . . . . . (240a)
sary fnformation from previous calculations we must first analyse the DuiXI+DuXu+Dou=O j
basic system by the method of the previous paragraph. Having found the We split next the column matrix Xu into matrices
column matrix H the r equations in the r unknowns take the form Xu=x+y ......................... . (241)
b1 'fb1 X+b1'H=0 . . . . . . . . . . . . . . . . . . . . . . (236b) where x satisfies the equation
where we may write ~ for ~ since the basic system is now redundant. Dux+Dou=O or X= -Du-1 D0 u.............. (242)
The systematic solution of (224) and related equations was discussed on Hence
page 28 but there are a few further points arising in practical calculations Du 1 X 1 +Duy=0 or y=-Du-1 Du1 X 1 . . . . . . . . (242a)
which are best investigated here. Thus, we mentioned on page 20 that it Substituting for x and y into the first of Eqs. (240a) we find
is often possible and justified to neglect certain part flexibilities of the
elements; for example, in a ring analysis we can usually ignore the direct D1 X 1 +D u/[ -Du-1 D0u-D11- 1 Du1 X 1 ]+D01 = 0
and shear flexibility. This applies not only to the evaluation of the external or
flexibility F but also to the determination of the internal redundancies X. X1=- (DJ-Du/Du-1 DuJ)-1(D0 I-Dm'Du-1 Dou) ..... (243)
We write now the D and D0 matrices in the split form from which we deduce y and hence X. Eqs. (243) are identical with the
D=Da+Dband D0 =Doa+Doo . . . . . . . . . . . . . . . . . . . . . . (238) elastic compatibility equation for a basic system with n- p redundancies.
The matrix form (224) of the equation of compatibility is particularly
where the suffices a and b refer to the two flexibilities into which we sepa- suitable to illustrate the transformation (seep. 29),
rate the total flexibility of each element. An approximate solution X a to
the unknown column X is then obtained by ignoring the flexibility b. X=BY .............................. (195)
Then, when the equations are ill-conditioned. Thus, by substitution of(l95) into
DaXa+Doa;;=O or X 0 =-Da-1 Doa . . . . . . . . . . . . . . . . . . (239) (224) we find
Occasionally we may require subsequently the correction x to Xa to find ~'fb1 BY +b1 'fb0 R=0
the true column X, and premultiplying by 8'
X=Xa+x . . . . . . . . . . . . . . . . . . . . . . . . (239a)
B'b1 'f~BY +Bif»1 'fb0 R=0
Substituting (238) and (239a) into Eq. (224) we derive easily . or
X=-D-1 (D 00 +DbXa) b 2'fb2Y +~'fboR=O . . . . . . . . . . . • . . . . . . . . . . . . . . . . (244)

36
where (see also Eqs. (196) and (197)),
b2 =b1 B and b2 =~B . . . . . . . . . . . . . . . . . . . . . . . . . . (245)
are merely the matrices for the true and statically equivalent stress systems
in the basic system due to
Y={l 1. ............... 1}
The form (244) of the equations of compatibility may, of course, be written
down directly when starting ab initio with the group unknowns Y.
Application to a typical aircraft structure
We present now a detailed investigation of a type of system characteristic
of aircraft wings. Consider to this purpose the structure shown in FIG. 37
which consists essentially of an orthogonal or nearly orthogonal grid of Fig. 38.-5lgn convention for flange loads and shear flows
spars and ribs covered with sheet material. Longitudinal and transverse
flanges may be placed at the intersections of spar and rib webs with the
covers. In addition the covers may be stiffened with further longitudinal Unstiffened

I
and/or transverse flanges. The cross-section is assumed arbitrary and the cut- out
spars may taper differently in plan view and elevation but the angle of
taper 20 is taken to be so small that cos 20R:> I and sin 20R:>20. The analysis
is not restricted to structures with continuous rib and spar-webs, covers
and flanges and includes hence any kind of minor or major cut-out.
The geometry considered excludes swept-back wings with ribs parallel
to the line of flight. On the other hand swept back wings with ribs perpen-
dicular to the main wing axis can be analysed by the present method as
long as we are given the necessary information for the triangular root-
section. Delta wings may also be investigated by our theory as long as the
grid of ribs and spars conforms to the geometry stipulated here. Naturally,
many of the restrictions imposed limit the applicability of the- method. bay
Indeed we intend our analysis only as an exploratory and tentative first
attack on the more general problem. We hope to return to this and similar
points in later publications. 12~CD~~~~--L
The problem of finding the stress distribution in the shell type of
structure considered is strictly infinitely redundant. Hence 'it is necessary
to introduce for practical calculations considerable simplifications. First
we adopt the standard assumption in wing stress analysis of a membrane Unstiffened
state of stress, i.e. we exclude any bending of covers and flanges normal to
the surface of the wing. For the very thin wings now coming into promin- cut-out
ence this idealization is open.to grave doubts and will no doubt have to be
reconsidered in future. An essential characteristic of our theory is the
assumption that the longitudinal and direct stresses vary linearly between
the nodal points of a three-dimensional grid of lines traced on the wing
0
Longitudinal flang~s
continuous at junction } ~ = 12

cover. This system of lines should, in general, be at least as fine as the grid
of spars and ribs whose intersection with the covers forms the best mini-
mum set of grid lines. The latter grid will often be sufficiently close if we
are dealing with a multi-web structure and ribs at not too great a distance.
However, many instances occur where it is necessary to select additional
nodal points between which the direct stress is taken to vary linearly.
For example, we may choose points intermediate between spar webs on
the rib stations if the spacing of the spars and the sheet thickness of the
cover are large. Similarly, if the structure has few ribs we may have to
introduce new transverse stations in order to reduce the spacing of the
grid in the longitudinal direction. In either case there need not be an actual
longitudinal or transverse reinforcement along the new grid lines. We call
the surface enclosed between two adjoining grid lines in the z- and s-
direction a field, and denote by 'bay' a part of the wing structure which
lies between two cross-sections taken through adjoining grid lines running
in the s-direction (see FIG. 37). The assumption of a linear direct stress
distribution along the edges of an orthogonal and flat field yields from
overall equilibrium conditions a parabolic shear flow distribution along
the edges. Naturally, neither the linear direct stress nor the parabolic
shear flow variation are, in general, exact and violate the internal and
boundary compatibility conditions of the field. This is not serious as long
as we keep the spacing of the grid lines reasonably close. Moreover, we
simplify further the problem by neglecting the quadratic and linear terms
in the shear flow and considering it to be constant within each field. We
note that for non-orthogonal grid lines (tapered structure) the uniform
shear flow offends against the equilibrium conditions even if the direct stress
is constant between adjoining nodal points. The errors introduced by the
assumption of uniform shear flow are, however, practically insignificant
for the reometry of structure considered here when the nodal point dis-
tances are small.
If the direct stresses along the grid lines were known we could calculate
the fraction of sheet area to be added to the reinforcements to form the
equivalent or effective flanges. This applies to the cover, spar-webs and
rib-webs and yields an idealized structure in which the fields are only shear
carrying and the direct stress carrying ability is concentrated in flanges;
an assumption widely used in aircraft practice. Neglecting the Poisson's
ratio effect and assuming the same material for flanges and sheet material
cover, the fraction of sheet cross-sectional area to be added to the flanges
varies between 1/6 and 1/2 if the fields are flat; the former value applies
when the field is in pure bending in its own plane and the latter when it is
under uniform stress. Since the stress distribution is unknown we can at
best only estimate the effective areas of the flanges but may use an iteration
process if the first guess proves inadequate. However, the latter procedure
is really clumsy and lengthy and a direct method, obviating the guessing
of flange areas would evidently be useful, in particular at the root or other
Longitudinal flanges
• interrupted at junction
N=3
Fig. 39.-Geometry of typical bay for determination of number of
redundancies
structural or loading discontinuities where the stress distribution is more
difficult to estimate and the Poisson's ratio effect more pronounced. Such
a method is given here at the end of this sub-section but at first w~ d~velop
the theory under the assum;Jtion that the effective flang~s areas are known
and that they are constant between two adjoining nodal points.* For the
webs, when considering torque and lift loads, it is alwtys su n:ient to add
I /6 of the web cross sectional area to the longitudin:ll and transv~rse
flanges at the intersection of the spars and ribs with the cover.
We summarize now the main assumptions underlying the idealized
structure selected for analysis. Thus, our system consists of an orthogonal
or nearly orthogonal grid of spars and ribs with top and bottom covers.
Effective flanges of constant area betw~en adjoining nodal points and carry-
ing only direct stresses are assumed placed along the grid lines in longi-
tudinal and transverse directions. For the tim;: being w~ assum;: that the
flange areas are known. The direct stresses and hence also the flange loads
are taken to vary linearly between nodal points. All sheet material for
covers, spars and webs is assigned a purely shear carrying role and a con-
stant thickness within each field. The angles of taper of spars in plan view
and elevation are assumed to be small. The shear deformability of covers,
ribs and spars is included ab initio in the analysis. For the stresses and loads
in the various elem;:nts we adopt the si~n convention illu;trated in FIG. 38.
Naturally, the idetliZltions and sim;Jiifi:ations introduced are strictly only
necessary for the calculation of the redundtncies. The btsic or statically
equivalent stress system may and should preferably be d~termined in the
(cut) actual structure.
Degree of redundancy of idealized structure
We proceed next to the enum;:ration of the redundancies in our idealized
structure. In addition to the sim;>lifications introduc~d previously we
ignore here the bending stiffness of the flanges for displac~m~nts tang~ntial
to the wing surface. This is, no doubt, su.nciently accurate for th~ present
exploratory analysis. The wing structure supported at the root and free at
• Strictly, the latter assumgtion is only necessary when finding the Oexibilities of the Dances for the
calculation of the redundancies.

37
the tip is assumed stiffened by ribs at least at the root and the tip. These Root. with rib
ribs need not necessarily consist of a web with flanges but may take the
form of a &tiff-jointed frame or ring. However, independently of the design
of the ribs we may always substitute an equivalent shear web with flanges.
The wing structure is subdivided into a number of bays of which we show
a typical intermediate one in FIG. (39). The cross-section at the junction
nearer to the tip may be stiffened by a rib carried across some or all cells;
FIG. (39) indicates also those longitudinal flanges which are continuous
across the same junction. It should be noted that if there is a change of
transverse slope of the cover at a longitudinal flange the latter must be
connected to a spar web.
We use the following notation :
fl=number of longitudinal effective flanges which are continuous across
the junction, i.e. are not interrupted there.
N =number of closed cells stiffened by ribs at the tip end of the bay.
Then the number of redundancies arising from the geometry of the bay is
{1-3+N-l
Tip
Hence, in a tubular structure of the type shown in FIG. 37, free at the tip
and either fully built-in at the root or with prescribed displacements there Fig. 40.-Multi-web wing without intermediate ribs
at all longitudinal flanges, the total number of redundancies is
n=~ [@-3)+(N-l)] . . . . . . . . . . . . . . . . . . . . . . . . (247)
bays

>
s
If certain of the flanges are not held at the root section the number of
redundancies reduces accordingly. For example, if the root-section is at
the aircraft centre line and the wing is subjected to anti-symmetrical z
loading the number of unknowns reduces by p,- 3, {3, being the number
of longitudinal flanges at the root. The number in the square brackets in
(247) can, of course, vary from bay to bay since effective flanges may be
interrupted at such stations. Also the number N of stiffened cells may be panel
Rectangular stiffened
made different in each bay by the addition or removal of spar webs.
However, when f3 and N are the same in all bays and all the flanges are
held at the root formula (247) becomes simply
n=a[f3+N-4] . . . . . . . . . . . . . . . . . . . . . . . . (247a)
where a=number of bays. If the sheet cover is missing between two ad-
joining longitudinal flanges in a bay and the cut-out is not provided with a
stiff-jointed closed frame to restore partially the lost shear stiffness of the
sheet then the corresponding cell is open in this bay and by definition
is not included in N. Similarly, if there is no rib or equivalent frame in a
cell at the section considered this cell is excluded from the numbering
for N. Note that spar webs need not be continuous throughout the length
of the wing and may be discontinued at any junction. Formula (247) still
remains valid.
If the cross-section is singly symmetrical the n redundancies of Eq. (24 7)
split into two groups :

}
Self.equillbratmg stress system X: 1 (Flat panel)

n1 = I;[(~-1 )+(N-t)J
bays
(248)
na= I;(~-2)
bays
of which n 1 applies for the lift and torque loads and n2 for the drag loads.
If all cells are closed, with the same number N in all bays and effective
flanges are only placed at the corners of the cells, then
f3=2(N+1)
and from (247a) the total number of redundancies is
n=a(3N-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (249)
which formula again assumes that all the flanges are held at the root.
Of considerable importance in modern aircraft structural practice are
the multispar syst~ms with .few, often only two, end ribs. A typical wing
of the latter type IS shown m FIG. (40). To analyse this structure we sub-
divide it into a number of bays whose length should not exceed say five
times the spar pitch. Effective flanges will by virtue of our idealization
process be acting at the junction of these bays although no ribs are pro-
vided th~re. For such a system the number n of redundancies when there
are no cut-outs in the sheet, when all spars are continuous for the full
length of the wing and all flanges are held at the root, is given by
n=(N-l)+l+a({3-4)=N+a({3-4) . . . . . . . . . . . . . . (250)
The last system to be considered is a flat panel which is of special import-
ance for diffusion investigations (see FIG. 41). It is assumed built-in at z=O
or held with prescribed displacements and free at the other three edges.
Here the number n of redundancies when there are no unstiffened cut-outs
is simply
II=~ (/3-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (251)
bays
where f3 is defined as in the case of the wing. When M fields are removed
without being replaced by stiff-jointed frames the number of redundancies
reduces by M.
The next step in our investigations is the discussion of suitable self-
equilibrating systems which may be chosen as redundancies. Consider
first the simple case of a rectangular flat panel shown in FIG 41. For the
redundancies we may select n systems of the type X= 1 illustrated in the
Longitudinal flange loads Transverse flange toads
Fig. 41.-Rectangular stiffened panel. Unit self-equilibrating stress system
of type X=l
figure. All information as to flange loads and shear flows is given there.
The corresponding equations (182) or (224) for the unknown X; are easily
seen to be reasonably well conditioned. Naturally, we can further improve
the conditioning by introducing group loads
X=BX . . . . . . . . . . . . . . . . . . . . . . . . . . . . (252)
wh~re B is a suitable square matrix. We do not enter at this stage into the
chmce of B but hope to discuss these points in Part III. When the panel is
symmetrical about its middle line it is preferable to combine the %-systems
into symmetrical and antisymmetrical groups.
In a wing structure of the type investigated previously we can describe
three simple types of self-equilibrating internal systems. They are shown in
FIGS. 42, 43 and 44 and denoted by
X=l, Y=1, Z=l
respectively. The first is the generalization of the %-system used in the flat
panel and the second and third may be considered as slightly modified
four boom load systems taken in the longitudinal and transverse directions
respectively. The longitudinal four-boom load systems are applied ex-
tensively in standard wing analysis.* The three figures are self-explanatory
and give all flange loads and shear flows associated with the unit systems.
Note, however, that the effect of taper is neglected except that we intro-
duce the true local dimensions in the evaluation of the self-equilibrating
• See J. H. Argyris and P. C. Dunne, 'The General Theory, etc.,' J.R.Ae.S., Vol. LI February,
September, November 1947 •

38
 Multi-web wing

Multi- web wing

Area •·•
1 I + I
q =•-·...:L-..:.1
11 / 2 h, + h 2

q =---1-
Central cross section 1, (h,•h,)

Self.equilibrating str~ss system X= 1

p.~!;_, Self-equilibral1ng stress system Y= 1

dld1 h
-I
___.1iJ{"+,-,.,~"""'

Longitudinal flange loads

Fig. 42.-Multi·web wing. Unit self-equilibrating stress system of type

X= I
systems. This ignoring of the influence of taper gives rise only to small
errors as long as the lengths I of the bays are reasonably small and the
angles of taper restricted to the order of magnitude mentioned initially.
The expressions for the shear flows in the ribs for the Y- and Z-systems are
also approximate, being derived for an equivalent rectangular rib. Again
the error introduced by this assumption is for practical purposes insigni-
cant. The conditions of equilibrium for the X- and Z-systems yield a load
in an 'effective' vertical flange at the intersection of the ribs and webs.
This flange load may always be neglected.
We enumerate now the number of X, Y, Z systems, independent within
their own group, which can possibly be applied. We find easily, with the
notation of Eq. (247) that we can use for each bay
({J -4) X-systems, N Y-systems and (N -I) Z-systems
Thus there are more systems than we require, the difference from the
number n of redundancies being obviously linearly dependent systems.
Evidently the
(N-l) Z-systems
are independent of the X- and Y-systems and hence must all be chosen as
redundancies. To complete the number n of unknowns we may use
({J -4) X- and one Y-system
However, it is preferable to apply more longitudinal four-boom ( Y)
systems since they are better conditioned, and to reduce accordingly the
number of X-systems. Thus, if we introduce all
N Y-systems
we have to adopt
({J- 3)- N X-systems
The last number reduces to N -I when the effective longitudinal flanges
are only placed at the corners of the N-cells.
It is often advisable to improve the conditioning of the D-matrix by
the introduction of group loads X and Y where
X=B1 X Y=B2Y . . . . . . . . . . . . . . . . . . . . . . . . . . (252a)
Longitudinal flange loads
Fig. 43.-Multi-web winJ. Unit self-equilibrating stress system of type
Y- I. (Longitudinal four-boom tube)
When the cross-section is singly symmetrical the number of redun-
dancies for lift and torque loads reduces to n1 given by the first of Eqs.
(248). The N - l Z-systems must still be included in the analysis for such
loading cases. If, in addition, we use all N Y-systems the necessary number
of X-systems becomes
f32-l-N
and is zero when the effective flanges are arranged merely at the corners of
the N-cells.
For the multispar wing of FIG. (40) with ribs only at the root and tip
the n redundancies of Eq. (250) may be selected as
( N- l) Z-systems at the root
one Y-system at the root
and
({3 -4) X-systems at each junction of bays and at the root.
The Y-system may involve a considerable length of the tube and if the
latter is tapered a more accurate estimate of the longitudinal variation of
the flange loads may become necessary.
Having selected a suitable system X, Y, Z of redundancies we can write
down the b1 matrix with the information given in FICiS. 42, 43, 44. To obtain
the b0 matrix we may use any suitable statically equivalent stress system
in the actual or idealized structure, but preferably the former. It was
mentioned on page 32 that it is advantageous to select a basic stress system
which, while being simple, approximates as closely as possible to the true
stress system and reference was made to the method of example (a) of
Section 9. Nevertheless, if the work in finding such a b0 matrix proves
excessive it may be preferable-since the choice of b0 does not affect
the conditioning of the D matrix-to sacrifice the closeness to the true
stress system and to select a b0 as simple as possible. Thus, we can calculate
a b0 matrix for a basic system in which the spars act independently; a
choice differing, in general, widely from the final b matrix.
39
 Tr,.nsverse II ange ·-·· . ,

~e
1 ' 1
1

c h a
/

D
Shear field -- / / ' f 2

1 1 ~-Longitudinal flange
//
Multi- web wing
d k b

' ' 1
'
g

Fig. 45.-5imple panel to Illustrate b 0 and b 1 matrices

Consider the simple grid shown in FIG. 45 and denote by P0 (P1), Q0 (Q1),
q0 (q,) the longitudinal flange loads, transverse loads and shear flows
corresponding to some R = 1 (X;= 1). The corresponding columns in the
b0 and b1 matrices are
( P oa1 P.oa2PObi P Ob2pOcl P oc2POdlpOd2QO<lQOe2QOflQOf2QOulQOu~OAqok}}
and
{Pial Pia2Pibl Pib2pi c!Pi c2Pidlpid2QieiQie2Qif1Qif2Qi o1Qiu2qihqik}
q=•t/m.,• n.,> . . . . . . . . . . . . . . . . . . . . . . . . (254)
respectively.
To find the D and 0 0 matrices it only remains to give the flexibility
matrix f of the elements. We write it in the partitioned form associated
with the b0 and b1 matrices of Eqs. (253),
I I
I f1 0 0 0 o

0 ft 0 0 0

f= 0 0 0 (255)
q:-1/2Q,
q:+t/(Q,•Q,) 0

L
lo0 0 0 fr
_j
Self-equilibrating stress system z= 1 where the suffices have the same meaning as in Eqs. (253). The matrices
f 1 and f 1 are themselves partitioned diagonal matrices, the sub-matrices
being the flexibility matrices of the longitudinal and transverse flange
elements respeetively. Since the flange loads vary linearly and the effective
flange area of each element is assumed constant within each element the
flexibility of the flange elements is that given on p. 22. Thus, for the
grid of FIG. 45 the f 1 and f 1 are,
I I
fa o o o I I
f. 0 0
0 fb 0 0
fz= 0 (256)
0 0 fc 0

lo
0 0 f,
0 0 fd L _j
Transverse flange loads L _j
typical sub-matrices being
Fig. 44~-Multi-web wing. Unit self-equilibrating stress system of type ,-
Z= I. (Transverse four-boom tube) I I I I
lb lb d. d.
3EBeb 6EB.b 3EC•• 6EC••
We write now the b0 and b1 matrices in the partitioned form: fb= f.= (257)
lb lb d. d.
6EB.b 3EB.b 6EC•• 3EC••
L _j L _j
(253) The flexibility matrices f., fw, fr are diagonal matrices with elements
4>/Gt, 4> .. /Gfw and O.jGtr respectively.
r 1 1 r
where the suffices /, t, s, w and r denote matrices for the longitudinal Io~· ·········· ·O ·0·
0·
.......... ·0
flange loads, transverse flange loads, shear flows in the fields of the cover,
shear flows in the webs and shear flows in the ribs respectively.
Since the flange loads are assumed to vary linearly between nodal points
we need at least two entries in the b matrices to describe the loads in each I 0· .. ·0 ___2£_0·.
4>
·0
flange. As such we use the loads at the ends. (nodal-points) of each flange Gtw
, fw=
element and denote them by the suffices 1 and 2 where 1 is the end first
met when we proceed along the +z or +s direction. These two associated
loads are entered in the assigned column of the appropriate sub-matrix of
b0 or b1 in two consecutive rows, the first of which always corresponds to
the end 1. Since the shear flow is assumed constant in each field only one
entry appears for a field. In the b1 matrix we arrange the columns in three ·0 ·0
groups, the first referring to the X, the second to the Y- and the third to 0· .......... ·0· o............ o· 1
the Z-systems. It is of the utmost importance to organize ab initio a rigid
and consistent system for the setting up of the b0 and b1 matrices. L .:J L _j

40

·0· .......... ·0
0·
·0
0· .......... ·0·
L _j
(258)
where<}), <})w, Q are the areas of the shear fields in the wing surface, webs
and ribs respectively and t, tw, t, are the corresponding thicknesses.
We have now all the information to form the matrices
D=b/fb1 and D 0 =b1 'fb0 R
and can hence solve the system of equations
b1'fb1{X Y Z}+b1'fb0 R=O . . . . . . . . . . . . . . . . . . . . (259)
for the unknowns X, Y, Z.
Finally, we find the true flange loads and shear flows of the idealized
structure from
5=b0 R+b1{X Y Z} . . . . . . . . . . . . . . . . . . . . . . (260)
It will then be necessary to translate the results (260) into stresses of the
actual structure. Finally the flexibility F of the structure at the points and
directions of the loads R may be determined from Eq. (229).
When the number of equations (259) is too large to be dealt with by our
digital computer we may proceed in two (or more) steps by the method
given on page 36. Essentially this introduces into our analysis a redundant
basic system.
If initial strains H are imposed on the structure in addition to the loads
R we have merely to add the column matrix
b1 'H
on the left hand side of (259). Thus the analysis includes inter alia the com-
plete calculation of wings under thermal loading.
A new approach to the problem of cut-outs
We emphasize that our above analysis is valid in the presence of any
kind of cut-out stiffened or unstiffened by closed frames as long as the
overall geometry and idealization conforms with the initial assumptions.
Nevertheless, when we have a structure which is essentially continuous
with only minor unstiffened cut-outs it may be worthwhile to apply an
artifice which avoids the lack of uniformity in the pattern of the equations
inevitably associated with cut-outs. Moreover, it is the idt>.al method of
finding the alteration in the stresses due to the subsequent introduction of
cut-outs in our system without having to repeat all the computations ab
initio.
The method is as follows. To preserve the pattern of equations disturbed
by missing shear panels or flanges we eliminate the cut-outs by introducing
fictitious shear panels or flanges with an arbitrary thickness or area.
Naturally, it is usually preferable to select for the latter dimensions those
of the surrounding structure. To obtain nevertheless the same flange loads
and shear flows in our altered structure as in the original system initial
strains are imposed on the additional elements of such a magnitude that
their stresses become zero. The effect of the fictitious elements is thus nulli-
fied whilst the uniform pattern of our equations is retained.
Let the column matrix of the unknown initial strains, in the additional
elements only, be
H shear flow is again taken to be constant within each field. When replacmg
In the new structure (i.e. without the cut-outs) we determine the flexibility
matrix f and the matrices b0 and b1 which we write in the partitioned form
~~boabo,. ~~ ~~bl•bl,. ~~
bo= bl = .............. (261)
L _j L _j
where the sub-matrices with the suffices g and h refer to the forces in the
elements of the original structure and the fictitious new elements respec-
tively.
Denoting the column matrix {X Y Z} simply by X and taking the
initial strains in the original structure as zero the Eqs. (259) in the un-
known X become,
II
~'fb1 X+ b1'fb0 R+ b1' ~ = 0 I I
L_j
and hence using the second equation of (261)
X= -D-1 b1 'fb0 R-D-1 ~,.'H ................... . (262)
where
D=~'f~
The stress matrix 5 follows as,
5=[b 0 -b1 D- 1 b1 'fb0 ]R-~D-1 ~,.'H . . . . . . . . . . . . . . . . (263)
The expression in the square bracket is the matrix b which we write in the
partitioned form
·~ 1:: l·..................... ~64)
L _j
To find now the column matrix H we put the stresses in the additional
elements to zero. Thus, the matrix 5 must be
l·.............. .......
s~ I~ (26SJ
L _j
where 5. are the true stresses (forces) in the original structure. Applying
Eqs. (261), (264) and (265) in (263) we find
I I I I I I
I o I Ibh II
5.
bv,
I
= b. R- I1 bl• o-1 b1,.'H . . . . . . . . . . . . (256)
I
L.J L.J L _j
Hence
or
H = (blhD- 1b 1,.')- 1 b,.R . . . . . . . . . . . . . . . . . . . . (267)
The true stresses in our actual structure are thus
5.= [b.- b 1 .D-1 b1 ,.'(bthD- 1 bt~o')- 1 b,.]R . . . . . . . . . . . . (268)
which solves our problem completely. As mentioned already the method is
ideally suited for finding the alteration of the stresses in a structu!e through
a subsequent introduction of cut-outs, such as access doors which usually
seem to materialize at a late stage of design. Another particularly meful
application of the new approach may be found ih the analysis of fu~­
lages with window-openings. Naturally, the degree of redundancy IS
increased by the 'filling-in' of the cut-outs but this is of no importance
for the automatic computations envisaged here.
A more refined wing stress analysis
The above general method of wing stress analysis suffers from the serious
defect mentioned initially that the effective flange areas have first to be
guessed since the stress distributi~n on. which t~ey depend_is. unknown.
It is certainly feasible to apply an IteratiOn techmque but this IS not only
necessarily lengthy but als~ rather uninspiring. . . .
To obviate these difficulties we develop a method which ehmmates the
determination of the effective flange areas and works directly with effective
flange loads. The method has the further virtue that it takes full account
of the Poisson's ratio effect which may be important at the root and at other
structural and loading discontinuities. The addition of 1/6 of the web
area to the flanges is always sufficiently accurate for lift and torque loads
and is retained here. Hence, our problem is restricted to the wing surface
alone. . .
The previously introduced assumptions that the loads are earned m the
idealized structure by a grid system of effective flange loads and fields
purely shear-carrying form also the basis of the new method. We assume
also that both the direct stress distribution and the effective flange loads
vary linearly between consecutive nodal points. However, our analysis
does not presume that the so-called effective flange areas-whiC:h do not
enter into our developments-are constant between nodal pomts. ~he
the linearly varying direct stresses across a grid line by efftJ7tive flange
loads at the nodal points we introduce the additional ass~p.hOJ?. that. the
actual flange areas and thicknesses do not vary across this grid hne. Smce
in wing structures plate thicknesses a~d pos~ibly flange ar~s may vary
just there it is suggested to take fo~ this particular c~lc~lauon the mean
values of areas and thicknesses on either Side of the gnd hne; on the other
hand when there is a cut-out on one side of the grid line or the flange is
interrupted the corresponding values should be taken as zl?ro. These
simplifications are not necessary for the purpose of the analys~s but ease
the problem of notation; moreover they do no! affect s~nously the
accuracy of the computations. Contrary to our previous practice of n~;J:m­
bering the flange elements with letters we numb:Jr here only the nodal pomts
with numerals.
We detive now the equation connecting the effective flange loads at
nodal points in the z and s directions of the idealized structure a!ld the
direct stress distribution in the plate material. It is more convement to
fix a particular point and for this purpose we select the point 9 in the grid-
system shown in FIG. 46.
41
 The conditions of equilibrium for the actual and idealized systems yield
in conjunction with Eq. (269)
P 9 ==a,9[B 9 +2(A 8 ,9 +A 9,10)] +a , 8 A8 ,9+a,,10A9.lo I
+a.,92v(A 8 ,9 + A9 , 10) +a, 8 vA 8 , 9 +a.,10 vA 9,10 J (270)

Qg=O'.g[Cg+2(A3,& +As.ts)] +a,aA3,9 +a•tsAs.ts I

+a,g2v(Aa.s +A9,I5) +a,avAa.9 +a,IsVAs.ts J ....... . (271)

These equations are expressed more concisely in the form,

I i I~ i I I
I a,s I I I O'zto I
Nodal Points
=L9.91
a,s
I
I
+ Lg,s I a,g I + Ls.to
I a.,s I a,to
I
L _j I ._I L _j
(272)

+ L9.3
~~a,a'l + L9.1f> I
i O'ztr.ll
Actual Structure aR3 a.~ 15
and Stresses L _J L _j

where the matrices L are as follows:

I
Ls.s= I
8g+2(As.s+As.w>
2v( Aa.9 + A9,15)

l ...
Dire-ct stress ·
...~ Direct stress L
in f-lange'- , _,.-..;v-
a..,- EEu ~ in sheet I u:, I
a:.= E:(E 11+VE:z.)= f<a,.• VOzz)
Ls.3= I vAa,9
0 A
0
(273a)
3.9
L _j

The matrix L9 ,10 (L9 ,15 ) is obtained from L9 , 8 (L9 , 3) by substituting 10(15)
o, for 8(3). Equations corresponding to (272) may be written down for any
other nodal point. We see immediately that
(274)
and that,
(274a)
when r and s are not adjoining nodal points of the grid. We deduce also
that Eqs. (273a) are the general formulae of the L-matrices for adjoining
Idealized Structure and Equivalent Flange Loads nodal points in the z- and s- directions respectively.
Consider now the column matrices for the flange loads and stresses at
all p nodal points

Purely shear
carrying
5={P1 Q1 P 2Q 2 . . • . . . . . P 9 Q 9 •.•••••• P,Q,}
}
.. (275)

From the set of equations of the type (272) we find,

5=Ls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (276)
where L is the symmetrical partitioned matrix

Lt2· · · · · · · · Lt9· ·······Ltv

Fig. 46.-Equlvalent flange loads in idealized structure

L92 . . . . . . . . Lgg ....... . L9 , (277)
The following notation is used: all values apply to nodal point 9.
P9 , Q9 effective flange loads in z and s directions
89 (actual) area of longitudinal flange+ ! (spar-web area) LP2 ........ Lvs ........ L,,,
_j
C9 (actual) area of transverse flange+ ! (rib-web area)
E., E Young's modulus of sheet and flanges respectively From (274a) the submatrices with suffices referring to non-adjoining nodal
a, 9, a.<B longitudinal and transverse flange stresses points are zero. Solving Eq. (276) for s we find
E,9, E.09 longitudinal and transverse flange strains s=L-15............. ... . . . . . . . . . . . . . . . . . . . . . . . . . (278)
a,9 ', a.<B· longitudinal and transverse stresses in sheet and hence also
s I
E.:= 1~·v2 effective elastic modulus of sheet e=e=.EL- 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (279)
where e is the column matrix of the flange strains at the nodal points, i.e.
_1_ E.' _1_ E.' .
A3,9= 0 E 13.913.9• A8 ,9= 0 ed8 ,9t 8 ,9 effective ~ange areas due to plate in e={EztEst· ......... E,gE.g .......... E,,E8 ,} • • • • • • • • • • • • (279a)
pure bendmg
Thus, once we have determined the effective flange loads the flange
Note that the material of flanges and sheet is assumed to be different. stresses and strains follow from Eqs. (279) and the direct stresses in the
We have, sheet from Eqs. (269). No guessing of effective flange areas is involved in
and a,9'=E,'(E,9+vE.,)= ~(a.,+va.,) } .. . .. .. .. .. . (269)
this procedure but we have on the other hand to invert the matrix L with
2p rows and columns. It is apparent that if we knew the effective flange
areas B., c. at the nodal points we could immediately write down the in-
verted matrix as a diagonal matrix whose elements are the unit flange
a,9'=E,'(E.,+vE,9)= ~'(u.,+va,9) flexibilities at the nodal points. In fact, then
42
 I
I
EB.1
0 ···························0 l ~. ~ 1: ~.a=
I
ll9.a/6
I 0
:l (283a)
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ·0 L L _j
EC.1
The matrix 19 ,10(19 ,15) is obtained from ~.s~.a> by substituting 10(15) for
0· ....... ·0 _I_ 0· ............ ·0 8(3). It is simple now to write down in Eq. (282) the terms for any other
EBet~ (277a) pairs of strains (E.E,). We deduce immediately that,
0· ............ ·0 E~et~ O· ....... ·0 1.,=1.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (284)
and that all lr., matrices are zero when they do not refer to adjoining points.
Moreover, matrices (283a) are typical for adjoining nodal points in the z
0· ...................... ·0 _I_ 0 and s directions respectively.
EB.'P
Introducing the matrix
0······ .................... ···0 _I_
I EC.'P I I I
L _j
111~1~-~ ·_· ~-~ ·_· ~~·~·-· ~-~ ·_· ~-~ •:.'P
Consider now the b 0 and b 1 matrices of the basic system. We emphasize
that the b 0 system can in the present method only be derived from pure
I= I ~.1· 0 0 0 0 0 0 0 0 .I.J.9· 0 0 0 0 0 0 0 0 0 l;,,p (285)
static considerations since the effective flange areas are unknown. Thus,
b 0 cannot be derived for a multicell wing from an Engineers' theory of
IPl 0 0 0 0 0 0 0 0 0 0 l'P9 0 0 0 0 0 0 0 0 0 0 ···p
L _j
bending cum Bredt-Batho analysis unless we assign arbitrary areas to the
flanges. To find, in fact, the b 0 system in this example it is probably best to we can express Eq. (282) concisely as
select a set of longitudinal flange loads in equilibrium with the applied S;=[P;1 Q;1 . . . . . • . . P;9 Q;R· ....... P;pQ;p]le . . . . . . . . (286)
loads and derive a statically consistent set of shear flows and transverse
flange loads in equilibrium with the applied shear forces and torque. Con- Observing that the matrix
trary to the system adopted in the previous method we write here the [P;IQ;1....... . P;9 Q; 9 ••••••.• P;pQ;,]
statically determinate b 0 and b 1 matrices in the partitioned form
is the ith row of the b11 matrix and noting Eqs. (281) we find that the
I I contributions of the flange strains to the D and D0 matrices are
I b11 I L _, L-'
D1 =b11'1 E b11 D0,=b11'1E b 01 R.................. (287)
b1=
II bl. I (280) We conclude that the flexibility of the flanges at the nodal points is given
I blW by
I bl, (288)
L _j
Note the structural similarity between the I and L matrices. It is particu-
where the submatrices bot, b 11 give the flange loads at the nodal points. The larly pronounced when v=O. The total flexibility of the elements of the
rows in bot, ht1 are arranged in p pairs corresponding to the p nodal points; structure is now
the first and second row in each pair refers to longitudinal and transverse I I
flange loads respectively. f, 0 0 0
Returning now to Eq. (279) we apply it in the basic system for the
external loads R and the redundancies {X Y Z} respectively. We "find 0 f. 0 0
with the notation of Eq. (280) the flange strains f= • 0 0 ••••••• 0 ••••• 0 •• 0 •• (289)
0 0 f., 0

We seek next the contribution of the flange strains e 1 and e 0 to the matrices
D and D0 of the relative displacements S;k and 310• For this purpose we
apply a self-equilibrating unit load system X;=l (which may be also a
Y- or Z-system) and denote by P 13 and P 19 its longitudinal effective flange
loads at the points 3 and 9 and by Q18 and Q19 its transverse effective
flange loads at the points 8 and 9. LetS~ be the relative displacement at the
points and directions of X 1 = I due to some given flange strains E. and E ••
The contribution to S; of the straining of the flanges (3, 9) and (8, 9) is then
- 9
31 = .......... + f P;E,dz+ f Q1E.ds+
3 8
and since we assume that both flange loads and strains vary linearly between
nodal points we find,
L
0 0 lo
fr
_j
where the flexibilities of the cover, webs and ribs are as before.
We find for the matrices D and D 0 •
D=b1'fb1 and D 0 =b1'fb0 R . . . . . . . . . . . . . . . . . . . . (225b)
where b 1 and b 0 are arranged as in Eqs. (280).
Using Eqs. (225b) in (259) we solve the problem completely.
The refinement introduced by the L matrix need not of course extend
over the complete wing but may be restricted to the root and other marked
9 changes of structure and loading. For the rest of the structure it may still
be sufficient to estimate the effective flange areas and to use the simple
form (277a).
D. The Analysis of Structures by the Displacenrent Method
The analogy between the developments for the flexibilities and stiffnesses

-
1
11/3 l/6llI 1
1
d/3 d/6 1
1 1
I sl
given under A and Band summarized in TABLE 1 shows clearly that parallel
to the analysis of structures with forces as unknowns there must be a cor-
responding theory :with d~formations as unkno~s. ~s ~entioned in the
S;= .... +[P;aPt9]., I Eia I
j+[Q;sQ;9]· EEs
89 + .... introduction to this section Ostenfeld • when mvestigatmg frameworks
Ez9 I was the first to draw attention to such an analogy. In fact, his equations
1/61/3 L :.J d/6 d/3 L _j are the exact counterpart of the classic S;k equations given by Mueller-
L _ja.9 L _j s.9 Breslau for forces as unknowns. In more recent times Southwellt and his
(282) pupils have used his relaxation technique to solve the elasticity equations
The complete expression of S1 is arranged by pairing the terms involving in the finite difference form with displacements as unknowns for a great
number of problems. Hofft has applied the latter method to diffusion and
the longitudinal and transverse strains at the same nodal point. Thus, related problems in aircraft structures and has solved also the correspond-
showing only the typical terms involving {EzsE89}, ing equations directly. Lately Williams II has outlined an analysis of wing-
3;= .. · · +{[P;aQ;a]J9,a+[P;sQtsl~.s+[P;9Q;9]~,9+ structures of the standard or solid type by introducing the deflexions at a
finite grid of points as unknowns ; his technique, which is intended for use
~[P,loQ;lO]~.lo+[P;lsQtlsl~.1s{::J + (282a) in combination with the automatic digital computer, neglects however the

r
where the I matnces are, shear deflexions, which may have an important influence.

• toe. cit. p. 43.

3•15/3 t toe. cit. p. 43.
~.9= (283) ~ N. J. Hoft and Paul A. Libby, 'Rec:ommendations for numeric&! solution of reinforced-panel and
0 fuselage-ring problem. N.A.C.A. Rep. 934 (1949).
L llloe. cit. p. 43.

43
 Naturallly, a theory using displacements as unknowns would only be of
value if it could show some concrete advantages. It is clear that such an
advantage may possibly arise when the stiffnesses are simpler to calculate
than the flexibilities, which is, as we have seen previously, very often the

' s---T
case. In particular in the egg-box structure, characteristic of aircraft wings,
the stiffnesses k;~ are much easier to find than the influence or flexibility
coefficients S;h· Another obvious advantage arises when the number of un-
knowns is smaller for the displacement analysis. This may occur in frame-
works. especially the stiff-jointed type with few degrees of freedom at the
joints. The equations in the displacements for stiff-jointed frameworks are h
almost invariably well conditioned; a further point in their favour, not
only for iteration techniques but also for the direct solution. On the other
hand in continuous structures, like wings and fuselages, this is not the
case. Here, in fact, the equations in the displacements are nearly always
ill-conditioned and it then becomes necessary to introduce generalized
" % "
J
or group displacements as unknowns in order to improve the conditioning.
This is a pronounced drawback of the displacement method when applied
to aircraft structures. Furthermore, in such continuous systems the dis-
placement method will usually involve a considerably greater number of
unknowns than the force analysis in order to achieve a comparable degree
of accuracy. It is apparent then that the choice between the two parallel
techniques must be made on an ad hoc basis after careful consideration
of the possible advantages and disadvantages of each method for a parti-
cular problem. It would, however, appear that at least with the present
types of construction the force method is to be preferred for aircraft
structures.
Before proceeding to the general development of the displacement Frozen system
analysis we introduce first a simple example to familiarize ourselves with
the ideas. Consider the framework shown in FIG. 47, symmetrical both in Displacement Analysos - 2 Unknowns r; and r,
structure and loading. The number of unknown forces or moments when
the engineers' theory of bending is assumed to hold is evidently six. On the Fig. 47.-Displacement analysis of stiff-jointed frame
other hand, if we neglect the deformations due to shear and end load, two
deformations alone, the rotations r 1 and r 2 at the stiff joints. suffice to
specify completely the deformation of the system. The analysis may pro- We obtain from Eqs. (292)
ceed as follows (moments and rotations are taken as positive if in the anti-
clockwise sense). We freeze first the joints, i.e. put
r 1=r2 =0 (292a)
Then, due to the loading on the upper member, moments M 0 are applied
at the joints and are. with the notation of FIG. 47
where
P1l1 2 P1l1 2
M12o=-u M21o=+12
(290) (293a)
M - _p2/2:
230- 12
Having the rotations r 1 and r 2 and using the stiffnesses of the individual
elements contained in (293) we easily derive the actual moments in the
structure. Thus, introducing again the usual sign convention giving posi-
(291) tive bending moment when upper fibres are in compression, we find for
Consider next the system with free joints and no transverse loading sub- the bending moment M 12 at the junction (I) of element (I, 2)
jected to the loading by the joint-moments
P1l1 2 P2l2 2 Pt/1 2
M12= _P:r- 4~11r1-Z~l1,.2 . • •• • • •• . • . • •• • •• • •• (294)
R1=-M1=+u • R2=-M2=u - 12 .......... (291a) This method forms also the basis of the Hardy-Cross or the more general
The superposition of this and the previous case yields the true solution of Southwell relaxation technique in stiff-jointed frameworks.
the given system under the transverse loading. To analyse the second We develop next the general theory of the displacement method. We
problem we apply Eqs. (138) which take the form introduce immediately the matrix notation and assume that the structure
consists of a finite numbers of elements whose stiffnesses k, due to relative
~ ...................... (292)
displacements at the ends or boundaries of each element, are known. In
order to show most clearly and concisely the striking analogy between the
k21r1 +k2'J!2=R2 J force and displacement methods we present them side by side in the follow-
The stiffnesses k;A are easily found as (see also Eqs. (144)) ing TABLE 11; most of the information with respect to forces has been given

}
k _ 4£/A .. 4Eit k _ZEit -k previously under C. The complete duality between the two theories is,
11 - h + /1 ' 12- /1 - 21
of course, a direct consequence of the twin principles of virtual displace-
(293) ments and virtual forces from which they derive most naturally. We
believe that the analysis has not been given previously in this generality.
k 4EIA 4Eit . 4Eit It includes ab initio any effects of initial strains like temperature, lack of
22=JI+t;+T fit and •give' at the foundations. The great advantage obtained in deeper
which assumes that the horizontal beams and supportilll struts have the insight and new theorems and applications by developing the theory on
constant bending stiffness £/1 and £/A respectively. the most general lines is too apparent to need stressing.

TABLE II
A COMPARATIVE PRESENTATION OF STRUCTURAL ANALYSIS BY THE FORCE AND DISPLACEMENT METHODS

METHOD OF FORCES METHOD OF DISPLACEMENTS

External forces R Joint displacements r

Flexibility F Stiffness K
Displacements r=FR Forces R=Kr
FK=I KF=I
See also TABLE I See alSO TABLE I

44
 TABLE II (continued)

Unit Load Method Unit Displacement Method

Given the true strains £ in a structure the kinematically related displace- Given the true stresses a in a structure the equilibrating force Rat a given
ment r at a given point and direction can be calculated from point and direction can be calculated from
(295a) ...................... l·r=fa£dV I. R= JiadV ................................ (295b)
v v
a
where is a virtual or otherwise statically equivalent stress system due to where € is a virtual strain system due to unit displacement in given direc-
unit load in given direction. Statically equivalent stresses ignore compati- tion. In what follows we denote virtual strains as kinematically equivalent
bility conditions. (See also Section 6 and FIG. 15). strains. Kinem"ltically equivalent strains ignore equilibrium conditions.
(See also Section 4 and FIG. 8.)
Cf Eq. (84b) Cf Eq. (146)

StaticaUy Determinate System Kinematically Determinate System

Internal forces (stresses) on elements determined from Internal relative displacements (strains) of elements determined from
(296a) . . . . . . . . . . . . . . . . . . . . . . S=bR v =ar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
where matrix b is obtained by static reasoning alone. where matrix a is obtained by kinematic reasoning alone by displacing
one joint at a time whilst keeping all others fixed.
Flexibility of individual (unassembled) elements Stiffness of individual (unassembled) elements k
Internal strains, 1nternal stresses,
(297a). . . . . . . . . . . . . . . . . . . . . . . . v =fS =fbR S=kv=kar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (297b)
External displacements, External forces,
(298a) ..................... . r=b'v=b'fbR R=a'S=a'kar (298b)
Flexibility, Stiffness,
(299a) ..................... . F0 =b'fb Ko=a'ka (299b)

Cf Eqs. (121), (122), (125), (126) C.f Eqs. (153), (154), (159), (160)

Statically Indeterminate System Kinematically Indeterminate System

In the relation In the relation

(296a) S=bR v = ar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
b cannot be determined by statics alone. a cannot be determined by kinematics alone.
Flexibility of structure needs to be considered, entering as compatibility Stiffness of structure needs to be considered, entering as equilibrium
conditions. conditions.
On the oth.!r hand if the internal strains v are known the kinematically On the other hand if internal stresses 5 are known the equilibrating
related external displacements may be derived from external forces may be derived from
(300a) r=b'v R =ca'S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (300b)
where b is merely a statical~•· equivalent (virtual) matrix due to unit R's. where a is merely a kinematically equivalent (virtual) matrix due to unit
r's.
Hence flexibility Hence stiffness
(301a) K=a'ka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . (30lb)
Eq. (300a) is a special form of the Unit Load method (Principle of Eq. (300b) is a special form of the Unit Displacement method (Principle
Virtual Forces). of Virtual Displacements).
FIG. 48a illustrates the matrices b and ii on a particularly simple FIG. (48b) illustrates the matrices a and i on the same example as in
example of a singly redundant system. FIG. 48a.

R =1 R=1
t ' f 1
r:~--,r~~
I

~·-~~.1--..-~tl----,--~-..--b-..
-~

Trul! displacl!ment systl!m

True system Statically •quivatent systl!m
(a]=~ { 3 3 1 -2 1 3 -3 3 3 -3 3 }

)5l5li,'>
lbl = 113 3 1 -2 1 3 -3 3 3 -3 3 } [b)= 113 3 3 0 3 3 -3 3 3 -3 3}

Fig. 48(a).-True and statically equivalent stress systems in singly redund-

ant, pin-jointed framework
.{- ----- ,'\- -.---- ., ;. -T
' \
' '

"-~~--:;-

Kinematically l!quival•nt system-1 Kinl!matically equivalent system-2

lal={ooooooo'j-'t oaf oo = {13 o o o o o o o o o o 1

Fig. 48(b).-True and kinematically equivalent displacement systems In
pin-jointed framework
C/. Eqs. (125a), (126) Cf Eqs. (l59a), (1"60b)

45
 TABLE ll (continued)

Problem a Problem a
Given a set of forces R, determine a set of statically indeterminate forces Given a set of joint displacement r, determine the set of kinematically
X necessary to satisfy the compatibility conditions. Find also the displace- indeterminate joint displacements U necessary to satisfy the equilibrium
ments r in the directions of R. conditions. Find also the forces R in the directions of r.
Complete force matrix Complete displacement matrix
(302a) . . . . . . . . . . . . . . . . . . . . . . {R X} {r U} . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (302b)
By putting X= 0 we obtain the so-called basic system which is statically By putting U = 0 we obtain the so-called basic system which is kine-
determinate within limits of idealization. matically determinate within limits of idealization.
Stresses in basic system Strains in basic system
(303a) . . . . . . . . . . . . . . . . . . . . . . So=b0 R v0 =a.or... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (303b)
Stresses due to X (with R= 0) Strains due to U (with r=O)
(304a) . . . . . . . . . . . . . . . . . . . . . . 5 1 =b1 X v 1 =a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (304b)
where b 0 and ~ are obtained from statics alone. where ao and a 1 are obtained by kinematics alone.
True stresses in actual structure True strains in actual structure
(305a) . . . . . . . . . . . . . . . . . . . . . . 5=5o+51 =b0 R+b1 X v=v0 +v1 =a.or+a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (305b)
Strains of elements Forces on elements
(306a) ...................... v=f5=fb0 R+f~X S=kv=ka.or+ka1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (306b)
Compatibility condition in actual system at points of application of Equilibrium condition in actual system at non-prescribed displace-
forces X ments U
(307a) . . . . . . . . . . . . . . . . . . . . . . b1 'v=b1 'fb0 R+b1 'fb1 X=O ~'5=a 1 'ka.or+a 1 'ka1 U =0 . . . . . . . . . . . . . . . . . . . . (307b)
or or
(308a) CU+C.=O (308b)
where where
(309a) (309b)
Hence Hence
(310a) U = -C-1C0 = -(a1 'ka1)-1at'ka.or (310b)
True stresses True strains
(296a) S=bR v=ar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
where where
(3lla) (311 b)
. True strains True stresses
(297a) . . . . . . . . . . . . . . . . . . . . . . v=fS=fbR 5=kv=kar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (297b)
Displacements r due to R Forces R due to r
(300c) ...................... r=b'v=b'fbR=FR R=a'5=a'kar=Kr . . . . . . . . . . . . . . . . . . . . . . . . . . . . (300d)
where where
(312a) .................. F= F0 -b0'f~(b1 'fb1 )-1 b1 'fb0 K=K 0 -a.o'k~(a 1 'ka 1 )- 1 a 1 'ka.o . . . . . . . . . . . . . . . . . . (312b)
and and
F0 =bo'fb0 is the flexibility of the basic system since we may choose Ko=a.o'kao is the stiffness of the basic system since we may choose
(313a) . . . . . . . . . . . . . . . . . . . . . . b=b0 a=ao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3f3b)
Cf. Eqs. (217), (223), (222a), (224), (225), (226), (227a), (228), (229), (230)

Problem b
Given a set of displacements r find forces R, stresses 5 and strains v
From Eqs. (300c)
(314a) . . . . . . . . . . . . . . . . . . . . . . r=FR
Hence
(315a) .. .. .. .. .. .. .. .. . .. .. . R=F-1 r
(316a) . . . . . . . . . . . . . . . . . . . . . . 5=bR=bF-tr
(317a) . . . . . . . . . . . . . . . . . . . . . . v=f5=fbF-1 r
Once F is known the question of statical determinacy or indeterminacy
is irrelevent in this problem.
Problem b
Given a set of forces R find joint displacements r, strains v and stresses 5
From Eqs. (300d)
R=Kr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (314b)
Hence
r=K-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (315b)
v=ar=aK-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (316b)
5=kv=kaK-1 R................................ (317b)
Once K is known the question of kinem:ttical d~terminacy or indeter-
minacy is irrelevent in this problem.

Problem c Problem c
Given a set of initial strains H imposed on· free unassembled elements Given a set of initial stresses J imposed on elements with frozen joints
due to temperature, lack of fit, 'give' at foundations, find stresses 5 and (i.e. all joint displacements zero) due to temperature, lack of fit, 'give' at
total strains v when forces R=O. foundations, find strains v and stresses 5 when displacements r=O.
Total strains of elements Total stresses on elements
(318a) . . . . . . . . . . . . . . . . . . . . . . v=fb1 X+H 5=k~ U +J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (318b)
Note that the column matrix U must here include all unknown joint
Compatibility condition in actual system at points of application of displacements.
forces X Equilibrium condition in actual system in the directions of U
(319a) . . . . . . . . . . . . . . . . . . . . . . b1 'v=bt'f~X+b1 'H=O ~'5=at'k~U+a1 'J=0 ...................... (319b)
Hence, Hence,
(320a) ...................... X=-(~'fb1 )- 1 b1 'H (320b)
and and
(32la) ...................... 5= -~(b1 'fb1 )-1 b1 'H v=-a 1 (~'k~)- 1 ~'J ............................ (32lb)
(322a) ...................... V=-f~(bt'f~)-lbt'H+H 5=-k~(~'k~)- 1 ~'J+J . . . . . . . . . . . . . . . . . . . . . . (322b)
Note, Note,
H=-fJ J=-kH
Cj. Eq. (236)
46
 TABLE ll (continued)

Problem d Problem d
Assume that we write the total set of forces (including the statically Assume that we write the total set of joint displacements in the parti-
indeterminate forces) in the partitioned form tioned form

(323a)
l~l
L _j
~~l
L _j
(323b)

in which Z is known in terms of R and X. We set now the modified
problem (a): Given the set of forces R determine the set of forces X
necessary to satisfy the compatibility conditions.
Here in the basic system obtained by putting X= 0 the stresses
(303a) . . . . . . . . . . . . . . . . . . . . . . S0 =b0 R
are completely known although the system is statically indeterminate.
Similarly we know the stresses
(304a) . ... . . . .. . . ... . . . . . . . . 5 1 =b1 X
when R=O
True stresses in actual structure,
(305a) . . . . . . . . . . . . . . . . . . . . . . 5=50 +51 =b0 R+b1 X
strains in elements
(306a) . . . . . . . . . . . . . . . . . . . . . . V=f5=fb 0 R+fb1 X
Compatibility condition in actual system at points of application of X
(324a) . . . . . . . . . . . . . . . . . . . . . . b1 'v=b1 'fb0 R+b1 'fb1 X=O
where b1 is a set of stresses statically equivalent to unit X's (and R= 0)
preferably found for Z = 0. In the latter case the rows of b1 are the same
as the corresponding rows of b1 of Problem (a).
Thus,
(325a) . . . . . . . . . . . . . . . . . . . . . . X=- (b1 'fb1 )-1 b1 'fb.,R
True stresses and strains
(296a) S=bR, v=fbR
where
(326a) . . . . . . . . . . . . . . . . . . . . . . b = b 0 - b1 (b1 'fb1 )- 1 b1 'fb0
Displacements r due to R (see Eq. (300a))
(327a) . . . . . . . . . . . . . . . . . . . . . . r =b0 'v = FR
where
(328a)
and
(329a)
is the flexibility of the basic system.
The matrix b0 is a set of stresses statically equivalent to unit R's pre-
ferably found for Z = 0 and X= 0. In the latter case 60 is identical with
b0 of problem (a).
in which W is known in terms of rand U. We set now the modified prob-
lem (a): Given the set of displacements r determine the set of displace-
ments U necessary to satisfy the equilibrium conditions.
Here in the basic system obtained by putting U = 0 the strains
v0 =aor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (303b)
are completely known although the system is kinematically indeterminate.
Similarly we know the strains
vl =ax u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (304b)
when r=O
True strains in actual structure,
v=v0 +v1 =aor+a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (305a)
forces on elements
S=kv=kaor+kaxU . . . . . . . . . . . . . . . . . . . . . . . . . . (306a)
Equilibrium condition in actual syst~:n at diJ;:>hc::m::nts U
a1 'S=a1 'kaor+i/kaxU=O . . . . . . . . . . . . . . . . . . (324b)
where a1 is a set of strains kinematically equivalent to unit U's (and r= 0)
preferably found for W = 0. In the latter case the rows of i.1 are the same
as the corresponding rows of a 1 of Problem (a).
Thus,
u ~ -(a1 'ka1)-1a 1 'kaor . . . . . . . . . . . . . . . . . . . . . . . . (325b)
True strains and stresses
v=ar , S=kar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
where
a=ao-al(al'kax)-lil'kao ...................... (326b)
Forces R due to r (see Eq. (300b))
R=a0 'S=Kr (327b)
where
(328b)
and
Ko=ao'kao .. . . .. .. .. . .. . . . . . . . .. . . .. . .. .. . .. . . (329b)
is the stiffness of the basic syst;:m.
a.l
The matrix is a set of strains kinematically equivalent to unit r's pre-
ferably found for W = 0 and U = 0. In the latter case lio is identical with
ao of problem (a).

Cf Eqs. (226), (227a), (229), (230)

Condensation offlexibility matrix Condensation of stf(f'ness matrix

The calculation of the flexibility matrix F given under problem (a) can The calculation of the stiffness matrix K given under problem (a) can be
be developed concisely as a condensation of the complete flexibility matrix developed concisely as a condensation of the complete stiffness matrix for
for the forces Rand X. the ...~isplacements r and U.
This matrix may be written as This stiffness may be written as
I I I
Fn/ I K1 Ku/ I
I I
Fn I Ku, Ku I
_j L _j
where 1(11) is for forces R (X) only and was denoted by F0 (0) in problem where I(ll) is for displacements r (U) only and was denoted by K0 (C) in
(a). Evidently FlllR=D0 • problem (a). Evidently Kmr=C 0 •
The flexibility matrix F of the actual structure under the forces R is then The stiffness matrix K of the actual structure for the displacements r is
then
(33la) ........................ f=F 1 -fu/Fu-1 fu 1 K=K 1 -Ku/Ku-1 Km . . . . . . . . . . . . . . . . . . . . . . . . (33lb)
Naturally this condensation may be performed in two or more stages Naturally this condensation may be performed in two or more stages
and is then equivalent to the method of problem (d). and is then equivalent to the method of problem (d).

Cf Eq. (175)

47
 TABLE II (continued)

Elimination and rigidification of structural elements
Assume a set of initial strains, written as column matrix H, in the struc-
tural elements to be removed, of such magnitude as to give zero stress in
resultant system.
Write the b and b 1 matrices of the complete structure in the partitioned
form
Rigidification and elimination of structural elements
Assume a set of initial stresses, written as column matrix J, in the struc-
tural elements to be made infinitely rigid, of such magnitude as to give
zero strain in resultant system .
Write the a and a 1 matrices of the complete structure in the partitioned
form

(332a) ••••• 0 b=
••
I
I
b. I
I b,.
L _j
I

I
.
I
bl- bu -I I

bl,. I
L _j
I a= I
I
I a, I

I a,. I
L _j
I

I . at=
I
I at• i
I
al,.
L _j
I

I
• ••••••••••• 0 •••••(332b)

where the suffix h refers to those elements that are to be removed. where the suffix h refers to those elements that are to be made infinitely
rigid.
We find We find
(333a) ...................... H = (b11,D- 1 b 1 ,.')- 1 b,.R J =(a1 ,.C- 1a 111 ')- 1 a,.r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (333b)
and hence forces in the new structure and hence strains m the new structure
(334a) .......... S.={b. -b11 D- 1 b 111 '(blhD- 1b 111 ')- 1 b 11 }R v. ={a. -a1 .C-1&t,.'(a1,.C- 1a 1,.')-• a,.}r . . . . . . . . . . (334b)
In this process the number of statically indeterminate forces X has been In the process of making elements infiniteiy rigid (stiff) we introduce
reduced to a degree depending on the number of elements removed. kinematic relations between displacements and hence reduce the number
of unknown displacements U accordingly.
In the inverse process of making infinitely rigid certain of the structural In the inverse process of eliminating certain of the structural elements
elements we have merely to put f,. =o for the affected elements. The we have merely to put k,. = o for the affected elements. The number of
number of statically indeterminate forces remains the same. kinematically indeterminate displacements remains the same.
Cf. Eqs. (264), (267), (268)

Generalized Forces Generalized Di~placements

Generalized forces given by Generalized displacements given by
II II I I II
IR I r I I Ao 0 I
Ir I
lx
(335a) I I I I (335b)
I I I I
o\
IL_j
••••• 0. 0 ••• 0 ••••••••

I u I I 0 AlI
L_j L_j L _j
The equation for the unknown X is The equation for the unknown U is
(336a) .. .. .. .. .. .. .. .. .. .. .. DX+D0 =0 CU+Co~O ................................ (336b)
where where
(337a) (337b)
Then Then
(338a) S=bR v =ar ........................................ (338b)
where where
(339a) (339b)
and the flexibility of the actual structure for the forces R is and the stiffness of the actual structure for the displacements r is
(340a) f=f0 -B0 'b0 'fb1 B1 (B1 'DB1 )-1 B1 'b1 'fboB0 K=Ko-Ao'ao'ka 1A 1(A1 'CA1)-1 A 1 'a1 'kaoAo (340b)
where where,
(340c) (340d)
is the flexibility of the basic system under the forces R is the stiffness of the basic structure for the displacements r

48
T HE application of the general theory with displacements as unknowns
to frameworks-both of the pin-jointed and stiff-jointed type-is
straightforward. For the stiff-jointed system the method is particularly
simple when direct and shear deformations are ignored. In fact, for all frame-
works the determination of the matrices C and C 0 is trivial once we con-
sider all possible degrees of freedom of the joints. See for example, the
systems of FIGS. (23), (24) and (48) investigated on pp. 23, and 45,
which show clearly how elementary the matrices a and stiffness k are when
we break up the structure into its simplest constituent components. We
need not therefore concern ourselves any more here with frameworks, and
we turn our attention to the membrane type of system characteristic of
panels reasonably small. Nevertheless, it is inevitable that the stress dis-
tribution derived from an approximate deformation analysis should, in
general, be less accurate than the one obtained from the approximate force
method in the same grid system.
We denote for the purpose of the analysis of the unit panel the dis-
placements parallel to the s and z axes by u and w respectively and intro-
duce also the local coordinate system ~. '· Consider now the state of strain
and stress arising from a unit displacement
va=l ............................... . (341)
Following our assumption the internal displacements are given by

w,,= j( 1-~)
aircraft applications. Essentially, a major aircraft structure like a wing
consists of an assembly of plates (fields) stiffened by flanges along their (34Ia)
edges. The field may be a curved and/or tapered surface but we ignore here
both these effects and consider only rectangular flat elements of constant where the suffix 3 indicates that these displacements are due to v3 =l.
thickness. For convenience the element formed by the plate (sheet) and its The strains and stresses in the sheet are*:

'
fouredge members is denoted by the term unit panel. It is assumed that
Uz.a= -GTtJ

l
the flange areas are constant along each edge.

(342)

Uzz3~~( 1-~)
where £'=£/(l-v 2 )
Strain .: 1 and load P 1 in flange 8 1
I I
E1.a~T , P 1 ,3 =B17 ................ (342a)
all other flange strains and loads are zero.
Similar formulae are obtained for the strains and stresses due to any
other v;=• I. To derive the stiffnesses we apply the unit displacement
method Eq. (295b), which takes here the form, •

l.k;n'~ J:J~ U;End~d'

where the integral extends over sheet and flange. For example, for the
k.J
' -k":l7
stiffnesses associated with va =I we obtain,
_ E'dt £81 Gtl
k.la ___ 3/- I +6d
I

I
E'dt Gtl
k2.1~c --6{ + 0 - 6d
(343)
.bL'-=---
..bz.~_ ____, E'dt £81 Gtl
: v,= kaa=+ 31 +1 +3d
lk., L,,
1

E'dt Gtl
v, = 1
k4.1-= +7J + 0 -3d
J
and

}
Fig. 49.-5tiffnesses of unit panel vE't Gt
k _ vE't Gt
;;a-- 4 +4 k63=+4+4
(343a)
We determine first the stiffnesses k;n of the unit panel shown in FIG. 49 k _ _ vE't _ Gt vE't Gt
for unit displacements in the z- and s-directions at the four corners or nodal 73 - 4 4 ' k63=+4-4
points of the idealized system. The stiffness of our element is hence an
8 x 8 matrix. As in the case of the force method it is necessary for the It is simple now to write down the stiffnesses corresponding to any other
practical evaluation of the k;A to introduce simplifying assumptions which unit displacement. For convenience we express the total stiffness matrix
are, naturally, concerned here with the state of deformations. Thus, we in the form
assume that the displacements vary linearly between the nodal points. k=k.+kd+k, . . .. .. .. .. . . .. .. .. .. .. . . (344)
Although thi!> idealization offends against the equilibrium conditions its * Contrary to our usual notation subscripts are used here to denote stresses and strains due to •it
effect upon the stiffness is not pronounced as long as we keep the unit displacements.

49
where the suffices s, d, .f indicate the partial stiffnesses for shear strains sion in the direction of the height of th:' w~b. The corresponding formulae
and direct strains in sheet, and direct strains in flanges, We find may be obtained by putting 1':, ~ 1·,, and 1· 7 ~ I'H which apply when C1 and
c2 are infinite.
The stiffnesses k., and k<~ contract to 6 >< 6 matrices and are
Gtl Gtl Gtl Gtl Gt Gt Gt Gt
"""3d-3d 6d - 6d ~ 4 -~--4
Gtl Gtl Gtl Gtl Gt Gt
Gtl Gtl Gtl Gtl Gt Gt Gt Gt 3d 3d -6d -6d 2-2
-"""3d """3d- 6d 6d -. -;r- .~ 4 4-
Gtl Gtl Gtl Gtf Gt Gt
Gtl Gtl Gtl Gtl Gt Gt Gt Gt -TiT- -~" 6rl 6rl -T -_:r-
6d . -- 6d 3d . . 3d -4- 4 - 4 ·4-
Gtf Gtf Gtl Gtl Gt Gt
Gt. Gt 6d 6d 3d -3d 2 2 (346)
44 (345) Gtf Gtl Gtl Gtl Gt Gt
Gt Gt Gtd Gtd Gtd Gtd - 6d 6d .Jd 3d---r -..,-
4-4 ~ 61_1l_M_
Gt Gtd Gtd
Gt Gt Gt Gt Gtd Gtd Gtd Gtd 2 -,-. -,-
4--4- 4 - 4 61 31-61-31
Gt Gt Gt Gt Gtd Gtd
-T -..,- -i-- 2 7 -,-
Gt Gt Gt Gt Gtd Gtd Gtd Gtd
.4 4 - 4 - 4-31--61 31 61
Gt Gt Gt Gtd
---4- -4--4 v- E'dt
v
E'dt
6/
E'tlt
v-
E'dt
61 0 0
L
E'dt E'dt E'dt E'dt
I 6r 31 y- 31 0 0
E'dt E'dt E'dt E'dt vE't vE't vE't vE't
v &-v-61 4·-4 4 - 4 E'dt E'clt E 'tit
E'dt E'dt E'dt E'dt vE't vE't v£ 't v£ 't
v 61 7J I) 0 (346a)
& v-v-31 4-4 4-4 E'dt E'dt E'dt E'dt
61 3/ 0 0
E'dt E'dt E'dt E'dt vE't vE't vE't vE't 61 11
-y-(;1 v 61--4 4 - 4 4 0 0 0 0 0 0
E'dt E'dt E'dt E'dt vE't vE't vE't vE't 0 0 0 0 0 0
-61-31 61 y·-·4 4 . 4 4
(345a) No contribution of the flanges is called for when evaluating the stiff-
vE't vE't vE't vE't E'lt E'lt E'lt E'lt
4 4 - 4 - 4 v-Td 6d -6d nesses of the webs since k is best included in the top and bottom panels.
Further simplification of the stiffness matrices for the webs is possible
vE't vE't £'/; E'lt E'lt E'lt when the top and bottom panels of our wing structure are identical. Then
for vertical loads alone the horizontal displacements in the two covers are
4 4-3d 3d - 6d- 6d antisymmetrical and the stiffness matrices (346) and (346a) may be con-
tracted to 4 > 4 matrices.
E'lt E'lt We illustrate now the application of the unit panel stiffnesses to the
3d -3d diffusion problem shown in Fl<i. (50). The plate is reinforced longitudinally
and laterally by stiffeners of area 8 and C respectively, and edge members
vE't vE't vE't vE't E'lt E'lt E'lt E'lt of area B 1. Displacements in the .1· and z directions are defined at all nodes
-4-4 4 4 - 6d 6d -3d 3d of the grid formed by lateral and longitudinal stiffening. Naturally the
L _I grid does not have to be restricted to this definition and we can always
choose a finer one if the stiffeners are widely spaced so that the assumption
of linear variation between adjacent nodal points can represent adequately
0 ---,-
EB1
0 0 0 0 0
the displacement pattern. Using the stiffness matrix of the unit panel
already derived, the setting up of the complete stiffness matrix follows
quite simply. It is oniy necessary to identify quickly and easily the dis-
placements defined for the unit panels separately with those defined for
0 0 0 0 0 0 the assembled panel. The complete stiffness matrix is obtained as (Eq.

--,-
EB1
0 -,-
EB1
0 0 0 0 0
299b).
K a'ka
where k is the stiffness matrix of the unassembled unit panels and may be
0 --,-
EB2
0 0 0 0 0
written in the diagonal partitioned form

0 0 0 EC1 EC1 (345b) k, 0 .............. 0

y-y 0 0
0 k,, 0 .......... 0
EC1 EC1
0 0 0 0 ---d d 0 0

0 0 0 0 0 0 EC2 EC2
y-y k= .............. (347)
0 ...... 0 k,, 0 . ,0
0 0 0 0 0 0 ECn EC2
d d

In assembling the panels of FIG. 50 to form a wing structure the stiff-

nesses (345) and (345a) may be simplified considerably when applied to 0 .............. 0 k,
rib and spar webs. Thus, for these cases we can always neglect the expan-
50
k, is the (8 x 8) stiffness matrix for the unit panel g as derived previously, 4E'dt 2EB 4Git
see Eq. (344). • To preserve the symmetry of formulation, the stiffnesses kts.ts=-y+-/ + 3d
of the reinforcement elements are included with the panels. Thus each of
the areas B or C is split in two and B /2 or C /2 associated with the panel on
each side. For the boundary member, of course, the whole area must be
included with the panel it bounds.
E'dt 2Git
(350)
k2s.1s =ks.ts ="3/- 3d

Fla. 50.-Rectanaular stiffened panel. Assembly of stiffness
matrix from unit panels
Since the terms in the a matrix are either unity or zero their formation is
particularly simple. Writing a in the partitioned form
a={a,.a, ........ a, ........ a.} .................. (348)
a, is the sub-matrix of 8 rows and 50 columns relating the displace-
ments defined for the unit panel g (FIG. 49) to the displacements as defined
for the complete system of FIG. (50). Superimposing the unit panel on
panel g of the complete assembly we find that the directions I, 2, 3, 4, 5, 6,
7, 8 of the unit panel coincide with 15, 25, 17, 27, 16, 26, 18, 28 respectively
and the sub-matrix a, is thus
All the remaining k's associated with 15 are here zero due to symmetry.
If R is the column matrix (50 rows) of forces applied at the nodes then
the displacements r are given by
r=K-1 R
Naturally, loads may not be applied at all nodes (joints) in which case
it may be desirable partially to solve the problem by eliminating the dis-
placements where forces are not applied and to use the condensed matrix.
(See TABLE II.)
Finally we apply the unit panel to the assembly and analysis of the egg
box type of structure illustrated in FIG. 51 where upper and lower plates are
connected together by longitudinal and transverse webs. Any stiffeners
on the plates are assumed for the present example to be along the lines of
web-plate intersections. The structure is taken to be symmetrical about the
horizontal middle surface and we consider the application of vertical loads
only. With these assumptions it is only necessary to specify three displace-
ments at each web intersection : the vertical displacement and the two
rotations of the web intersection line (FIG. 51). In many cases the webs may
be too widely spaced for the assumed linear variation of displacements
between them to give satisfactory accuracy. It then becomes necessary to
introduce further grid lines intermediate between the actual webs, the dis-
placements being defined at all nodal points formed by grid line inter-
sections. Where such nodal points do not lie on a web then obviously we
define there only the two rotations, since vertical displacement does not
affect the cover plates. Naturally, further lateral and longitudinal rein-
forcement of the plates can lie along the extra grid lines.
The analysis of such a structure under vertical loads follows that given
under Problems (a) and (b) in TABLE n. Thus, we designate the vertical
displacements as r and take the rotations as the redundant displacements
U.
The strains of the elements are here identified as the displacements of
the unit panel defined in FIG. 49 and can therefore be written (Eq. 305b).
v=aor+a1U

1 .... 15 16 17 18 ...... 25 26 27 28 ...... 50

I I
0 ..... 1 0 0 0 ....... 0 0 0 0 ....... 0 I
0 ..... 0 0 0 0 ....... 1 0 0 o....... o I
I
0 ..... 0 0 0 ....... 0 0 0 0 ....... 0 I
0 ..... 0 0 0 0 ....... 0 o....... o I
··=
0
(349)
0 ..... 0 0 0 ....... 0 0 0 o....... o 1
1 o..... 0 o o o....... 0 0 0 ....... 0

I 0 ..... 0 0 0 I ....... 0 0 0 0 ....... 0

Fig. 51.-EII·box type of structure. Analysis by displace-
ment method

1 o..... o o o o....... o 0 0 1 ....... o I

L _j The equation for the unknowns U, which is here the condition of equili-
brium of the moments corresponding to the rotations U at the joints is
(TABLE n, Eqs. (310b)),
All remaining columns in a, are zero.
a 1'ka1U +~'kagr = 0
As typical terms in the complete stiffness matrix the stiffnesses associated or
with direction I 5 are, for a uniform panel with t, d, I, B and C, the same in
each bay: U =- (a1 'k~)- 1 a1 'kagr
The stiffness for displacements r only is then
• Sutllx s refen here to the number of unit panels and should not be ~onlused with suffix s for stitr-
neu due to lhear stresses in (344). K = &o'k&o---&o'ka1(a1'ka1)- 1a 1 k&o (312b)
51
from which the displacements r under loads R are found as I I
l2k. 0 ........................ 0
r=K-1R
The total strains of the elements due to R are then
v=ar={a 0 -~(~'k~)- 1 ~'kao}K- 1 R (316b)
0 0 ................ 0
1
from which the stresses in the unit panels are calculate<.i. ~covers
Due to the simple geometry of the structure the matrices a 1 are again ao.
quite straightforward. Thus writing ao in the partitioned form
ao={aoo~~o~, ........ aog· ....... ao,} . . .•. .•. .. . .••. . (351) II
it is apparent that for the cover plates the ao's are all zero since vertical I j
displacements r can cause no strain in the plates• (with U = 0). I0 (355)

For the web/, the 8of matrix is easily seen to be

k=
I
.......... 0 k, 0 ... o
l I

1 ..... 5 6 ..... 9
I\
I I
0 ..... 0 0 ..... 0 webs
0 ..... 0 0 ..... 0
0 ..... 0 0 ..... 0
0 ..... 0 0 ..... 0
aw= 0 ..... 1 0 ..... 0 (352)
0 ..... 1 0 ..... 0 0 .•..................•... o k.
0 ..... 0 1 ..... 0 L ,J
0 ..... 0 1 ..... 0
L _j The factor 2 is introduced for the cover plates to take advantage of the
symmetry of the structure by including the unit panels of the lower cover
with their opposite numbers in the top cover. The k., kb etc., are of course
Likewise the a 1 matrix for unit panel c of the top cover is the stiffnesses of the unit panels discussed earlier.
The method of formulation of the matrices ~'k~, etc., given above is
..... 9 10 11 12 .... . 15 16 17 18 probably the most convenient for use with the automatic digital computer
I I since the various terms in the constituent matrices are reduced to their
.... . h/2 0 0 0 0 0 0 0 I simplest and most standard forms. However, it is instructive to consider
directly the components of a1 'k~, etc., and gain some physical insight
0 0 h/2 0 0 0 0 0 into their formation.
We call
0

0
0

0
0

0
0 .... . h/2

0 0
0
0
0

h/2
0
0
ao'kao=Ko
a 1 'k~=C
~'kao=C
}
........... . (356)
~.= (353) Thus the complete stiffness matrix for the displacement column { r U} is

I .............................. .
0 h/2 0 0 0 0 0 0
I _I
0 0 0 h/2 ..... 0 0 0 0
I~ ~I (357)
0 0 0 0 0 h/2 0 0
L ,J
0 0 0 0 0 0 0 h/2 ..... Ko is clearly the set of vertical forces R which arise due to unit r displace-

l
L _j ments when U is zero. Evidently only the webs are involved and we find
easily as a typical example the vertical forces at the joints due to r 5 = 1
and for the web plate f

t......
k _ 2Ghtw 2Ght,
10 11 12 o.s.s- 1 + d

I
h/2 0 0 (358)
..... -h/2 0 0
Ghlw
0 0 h/2 ..... ko.s.s =ko.2.s =- -~-
Similarly C is the set of moments arising at the joints due to unit dis-
at,= 0 0 -h/2 ..... ••••••••••••••••••• 0 (354) placements (rotations) U. By using the stiffnesses of the unit panel (or by
carrying out the matrix multiplication a1 'k~) we find for the moments
0 0 0 due to U9 =1

l
0 0 0 2E'h 2dt EBh 2 E'h 3 tw 2Gh 2lt Ghltw
ce,e=-3-1- +-~-+~+3d+ -3-
0 0 0
E'h 2dt EBh 2 E'h 3 tw Gh 2lt Ghltw
0 0 0
Ca,e=Cts.9= ---:rr- ---zr-'U/ + 6d +-6-
L _j E'h2dt Gh 2lt
cu,9=c7,9=~---g (359)

f
All other columns are zero.
2
---.zr--
E'h dt Gh lt 2

j
The matrices for the web plates may of course be reduced to six or even Ct7,g=Cta,9=cs.9=c1.9= I2d
four rows by using the assumption of zero vertical direct strain and the
antisymmetric character of the U displacements (see also Eqs. (346) and vE'h 2t Gh 2t
-cts.g=cu,g·= -c2,9=cs,9= -g-+-g-
(346a)). However, we retain here the full eight displacements of the unit
panels to show the simple formation of the a matrices with a completely Finally C is the set of moments arising at the joints due to the vertical

l
standard unit panel. displacements. Again only the webs are involved, and we obtain easily
The stiffness matrix k of the unassembled unit panels is written again in

J ...................... (
the diagonal partitioned form: _ _ Ght,
c12.s= -cs.s=r 360)
• This, of course, is no longer true if the webs are tapered in depth, when an appreciable part of _ _ Ghtw
the shear force carried here by the web is equilibrated by the vertical component of the direct stresses Cts.s= -ca.s=z
in tbe covers.

52
 In some problems it may be possible to neglect the shear deformations
of the webs. Quite obviously this introduces kinematic relations between
the rotations and the vertical displacements. However, the above break-
down of the structure into the simple unit panels is then not the most
suitable. A suggested method for the setting up of the K matrix for this
case, based on the theory of bending of plates, has been given by Williams.*

9. ILLUSTRATIONS TO THE ANALYSIS OF REDUNDANT

STRUCTURES BY THE FORCE METHOD
In this section we present two very simple applications of the force
method developed in Section SC. The first example shows how to deter-
mine the statically equivalent stress system in an N cell tube typical of a
wing structure.t With the direct stresses distributed according to E.T.B.
we find using the sik method the corresponding shear flow distribution
for the multi-cell cross-section with the assumption that the ribs are rigid in
their plane. Naturally, the axial constraint stresses and the effect of rib de-
formation remain to be investigated but the statically equivalent stress
system derived here is particularly useful, being, in general, a reasonable
first approximation when the structural design has to be based merely on
a statically equivalent stress system. This, of necessity, has been the
approach in most cases up to the present owing to the difficulty of
computing highly redundant systems. (a) Loading
Although the problem may, with some justice, be described as trivial in
relation to the powerful analytical techniques of Section 8 it is astonishing
to see what an unfortunate treatment it often receives~ven today.
The second example analyses-first by the S;k method-the axial con-
straint stresses in a four flange tube with shear carrying walls and deform-
able ribs under arbitrary loading at the rib stations. The solution of the
same problem is also obtained by the matrix method and the effect of a
cut-out is investigated by the H matrix device of p. 41. The 'exact' flexi-
bility of the structure is derived and compared with that given by E.T.B.
and Bredt-Batho. A thermal loading is also investigated by the matrix
~· Is,Dxb
method.
These two problems are only meant as preliminary illustrations of the (b) Basic system for calculation of Dxb
force method. Thermal applications of the S;k method are investigated in
Part II. More complicated structures, particularly suitable for showing
the power of the matrix formulation of the theory, will be analysed in
a later publication.
s
(a) Shear Flow Distribution in a Multi-cell Tube Due to E.T.B. Direct
Stresses
Consider the uniform cylindrical and multi-cell tube of the type shown
~
( Dxl 1 DxM-11 P.M
~
~
SM..-
1~ P..M•1 1~ £?. H~ l
.V-1

in FIG. 52 subjected to shear forces s., Sx through and a torque Tn about

the point 0. Find the corresponding distribution of shear flow if the direct
stresses are given by the engineers' theory of bending; thus axial constraint (c) DxM Distribution
stresses due to restrained warping are ignored. Instead of referring the
loads to the arbitrary point 0 we may alternatively give the point D Fig. 52.-Shear flow distribution in multi-cell tube. Sign conventions and
through which the resultant of all transverse forces is acting, i.e. T D ~ 0. equilibrium conditions
The direct stress a due to bending moments M"" M. about the axes Gx, (lay
Gy through the section centroid G and parallel to Ox, Oy, is By(!;:, ~-q.~ --q.,,+qY+=O .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. (a4)
- y -x Eqs. (a3) and (a4) yield, except for a constant of integration in each of the
a=Mxr+M.1 .................... .................... .. (al)
X y N cells, a shear flow distribution q whose resultants in they and x directions
where are s. and S" respectively. Since there remains one further equilibrium
condition,
M __ Mx-MuU:r 11 /fy) M _M11 -M,,(I.,yjl;r)
(a2)
x- l-[(/;ry} 2 jf,.J.]' . - l-·[(/;ry)2/fxf.J To= fqpods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a5)
for torque about 0, the degree of redundancy is (N- I); (the integration
are the effective bending moments for the chosen axes which are, in general, in (a5) extends over all walls and the normal Po is taken positive (negative)
not the principal axes of the cross-section. Physically M,., M. are the if movement along s leads to an anticlockwise (clockwise) rotation about
combinations of M:r, M. which give rise to pure bending strains about 0). It follows that the shear flow distribution in a single-cell tube under
Gx, Gy respectively. We could alternatively restrict ourselves to principal prescribed transverse loading is statically determinate once we stipulate
axes of the cross-section but in practice, unless these are obvious, it is that the direct stresses are distributed as per E.T.B.
preferable to use Eqs. (al), (a2). They are not only more convenient from For the analysis of the general case of an N cell tube we find it more
the computational point of view but permit also the retention of parallel convenient to use a slightly different approach. Thus, for the moment,
axes Ox, Oy at all cross-sections of a wing regardless of the change of we prescribe instead of the torque equilibrium condition, the rate of twist
directions of the principal axes.
The condition of equilibrium in the z-direction of an element dsdz of a c/>=8'=c~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a6)
wall gives,
in all cells. The prescribed c/> may be considered as an initial 'give' ex-
oq oa
os+t•oz=O . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a3) perienced by the ribs maintaining the shape of the cross-section and is
subsequently determined from the torque equilibrium equation. Our
where modified problem has now N redundancies. The basic system is obtained
q=ta. 8 =shear flow in the wall. by cutting the wall in each of the N cells and the unknowns
(,=effective direct stress carrying thickness of the wall (i.e. in- X[, Xu, ......... , X.~·
cluding an allowance for the stringers). are then the shear flows at the cuts. They are determined from the com-
patibility equations,
Similarly, we find from the equilibrium of an element dz of a typical N
flange g placed at a web-cover intersection (see FIG. 52), ~DMrXr+DMo=-cO (a7)
r=J
*lac. cit. p. 17.
t See J. H. Argyris and P. C. Dunne, 'The General Theory, etc.', Part V. J.R. Aer. Soc., Vol. Ll, which express the conditions of zer-o relative warping oat the cuts.*
September 1947, p. 770, and J. H. Argyris and P. C. Dunne, Structural Analysis, Part II of Handbook
of Aeronautics, Pitman 1952. .*We denot.e here the warping by the unconventional symbol 6 to apply directly Eqs.(282)intheir
ongmal notation.

53
 The shear flow distribution qb in the open tube forming the basic system
is obtained from Eqs. (a3) and (a4). Integrating (a3) with respect to s
and using Eq. (a2) we find
- Dxb -Dub
qb=S.T+SxT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a8)
X V
where
- s.~~Sx<I,..,I.> _ Sx-Sifxu/1,)
(a9)
S. 1-[(1.,.)2/f.,/y]' Sx 1-[(1,..) 2/fx/y)
and
s ..
Dxb=- JYt.,ds, D.b= -- fxt,ds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (aiO)
0 0
To determine completely the Db distributions from 'Eqs. (a 10) we require
also the equilibrium conditions of the type (a4) at each joint of spar web

-
and cover. Using (at) and (a8) in (a4) we have
(Dx+- Dxd- Dx- +By)bo =0 I
?- .. . . . .. .. .. .. . .. .. .. .. (all) unit shear flow

(Dv+- D.d- D._+ Bx)b• =0 j

The positive directions of q and s are indicated on
Choice of Basic System
The reduction of the multi-cell tube to an open section can, of course,
be achieved in a variety of ways. For example, we may cut the upper or
lower cover in each cell or we may cut N of the vertical walls (see FIG. 52).
However, consideration of the form of the 8's shows that the compatibility
equations (a7) are very much simpler for the former choice. We confirm
this immediately by applying the unit load method for the calculation of
the 8's which measure the relative warping at the cuts. Thus, if we apply
unit shear flows at each of the cuts of the open section in FIG. 52 we
FIG. 52. Warping due to twist - 6.,., =- 2 Q""'
Fig. 54.-Warping in Mth cell of basic system due to rate of twist cp
f( ... . )dsM
M
denotes _integration over the M cell. It must be emphasized that the sign
conve~tiOn used for qb and s in the basic (open) system is in opposition
to sM m t~e left-hand wall of the M cell and hence in evaluating the above
term the stgn of qb must be reversed over this wall.
Similarly, we obtain for the relative warping 8Motp due to cp (see FIG. 54).
8.\1otp =-2D.Mcp ............................. ............. (al8)
0

produce merely constant shear flow around each of the individual cells. The standard derivation of &i_s. (al7) and (al8) is by kinematics. Thus
Eq. (a 17) is obtained by integration of the shear strain expression corre:
The Redundant Shear Flows sponding to zero rate of twist. Also, we find (al8) from the condition of
zero shear strain along the middle line of the wall.*
s.r' The unit load method also yields directly the coefficients of the unknown
X's. Thus, the relative warping 8.11 M due to unit shear flow in the M cell is
~<t:::: I9 - _ Jds.
0.11.11 ~~ 11 ={3M
0
D $
• Gb..-.--" Gt G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al9)
where
J
f3 .·11 •c• dstM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a20)
Transverse loading s, through 0 Transverse loading S, through 0 M
~=0 S,.=O All cross-terms but 8.11. M- 1 and 8.u. :u n vanish since only unit shear flows
Fig. 53.-Effective shear forces for bending about non-principal axes
in adjoining cells act over a common wall. We find
8.u• .\1-1 = O.11-b .11 = - f3M-1•
If we consider now the total transverse loading s. and s. as acting
.\1
-G--
through D and split it into the two component loads and S.r (see FIG. 53) s. 0.H.M+1 =0 .11+1 ..I t-- _{3.\1,GM+l
. . . . . . . . . . . . . . . . . . . . . . (a21)
we can express the statically indeterminate shear flows in the form
xM=qM=s.D;"'111 +S.,D/.111 ••••••••••••••••••••••••••••• ••• <at2> where

where the DxM and D. 111 are unknown. For the basic system of FIG. 52 the !3.\1-1• M = J~' f3 :lldl H ~ Jo/ .. .... .. ..... ... .. ... .. .. ... (a22)
shear flows in the M cell of the actual system can then be written as: M-I,M M,M+I

external walls q=qb+qM I are the integrals extending over the common walls (spar webs) of cells

webbetweenM-landMcells q=qb+q 111 _ 1 -q 111 J~ .... (al3)

M -I, M and M, M +I respectively. The minus sign arises since the shear
flows due to unit redundancies have opposite signs in the common walls.

web between M and M+l cells q=qb+qM-qM+I Determination of Redundant Shear Flows
We may always put the total shear flow in the form We denote the unknown rates of twist associated with and S'x by s.
-.D., -D. c/>. and c/>x re~pectively. For the loading due to
the M'th compatibility s.
q=S. lx +S., I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4) Eq. (a7), whtch expresses the condition of zero relative warping at the
M'th cut is obviously

s.{ tdsM-2n. G'I'•'S_-

and Eqs. (al2), (al3), (al4) yield for the cross-sectional function D:x: in
theM cell 7., JD.rh .~. 1" f3 11 M- 1 , MD,., 111_ 1
external walls Dx=Dxb+D.rM I M

web between M-1 and M cells D.,=Dxb+D.,,M_ 1 -D.rM ~ (al5) +{3.11DrM -~{3.11, .II +I Dr, .11+1} =0

web between M and M+l cells D.,=D.,b+DxM-Dx.M+I j We obtain hence the set of N equations in the N unknowns D, 111
Similar equations may be written down for D •. f3£Dxl-{3h nDxn = - Dtdsl J
l
+2D.l Gcp.i
y
The 8 Coefficients for the Basic System o/FIG. 52 .....................................................................................................................................................
The 8Mo coefficients consist of two parts 8Mob and SMotp corresponding to
the shear flow qb in the basic system and the initial 'give' cp (see also
Eq. (177)). We have:
0 Mo =0 Mob +8 Motp
Application of the unit load method yields immediately for 8Mob -
NDr, .V-1 +f3NDxN JDtdsN+2D.NGcp,~
J~;dsM
-{3,\'-1• =-
OMob= ............................. ............. (al7) N •
M . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a23)
where s 111 is the circumferential anticlockwise co-ordinate in the M cell • _See e.g. J. H. Argyris, 'The Open Tube', AIRCRAFT ENGINEERING, Vol. XXVI, No. 302,
and the integral Apr1l 1954, p, 101.

54
These formulae are usually derived more simply from the condition of
equal rate of twist*

.P= ~~= 2iLaHdsM .................................... (a24) Similarly . . . . . . . . . . . . (a32)

M N I N
in all N cells in the actual system. In this approach we specify initially 2'La 111n 111 • G.P.of = - JDubP nd5 - 2~d"_,,n .11
I .r I
zero relative warping at all cuts and express the compatibility condition by
the equality of .p in all cells. On the other hand in our present method we where the integrals extend over all walls.
specify initially the same rate of twist .p in all cells and express the com- Having derived the complete set of redundancies we can calculate using
patibility by the condition of zero relative warping at the cuts. formulae (a14) and (al5) the shear flow distribution in the actual system
The solution of Eqs. (a23) and the corresponding ones for Du.lt is under any given transverse loads.
straightforward and may be put in the form If we have found the shear centre £, and the torque TE it is more
convenient to proceed as follows. Thus, using the equilibrium condition
I (FIG. 55)
I Tg=fqpgds ............................................ (a33)
· · · · · · · · · · . · ·. . . . . . . . . . . . . . . . . (a25)

J
and remembering that the engineers' theory of bending shear flows have
no torque contribution about £." we obtain with Eq. (a29)
N
dx 111 , d. 111 are the values of the redundancies corresponding to zero rate of Tf;=G.f»2'Lawfl.w ........................................ (a34)
twist. They yield hence the shear flow distribution qE---commonly known I
as the engineers' theory of bending shear flows-due to transverse shear or
forces acting through the shear centre E, · q E may be written Tg
G.f» = --.-,N---'-'-- .......................................... (a34a)
-d, -d. 2'LaMnM
qE=Svy+Sq (a26) I
"' v
where the cross-sectional functions d, (d.) are obtained from Eqs. (a 15) Hence the redundant q BM due to torque Tf; are given by:
with dxM (d.M) in place of DxM (D.M). The co-ordinates Xg, YE of E., can Tf:
be determined from a consideration of the two loading cases and Sr s. q ll Jf =c a .II ____,N___,:__
2'La.un.,r
(a35)
through £, shown in FIG. 55. We find I
I The shear flow is, of course, constant in each wall between two consecutive
I joints. Thus, we have in the M'th cell
N
I . . . . (a27)
external walls qll=qBM I
-IxvYE+IvYE=- fdvPods=- fDvbPods-2LduMnM

where the integrals extend over all walls and the normal Po is taken positive
I j
web between M-1 and M cells qn=qno.~r- 1 -qllM t .. .. (a36)
(negative) if movement along the positive s direction produces anticlock- web between M and M +I cells qu =q RM -q B• 111+ 1 J
wise (clockwise) rotation about 0.
The distribution q 8 is known as the Bredt-Batho shear flows in a multi-
cell tube. The special case of Eqs. (a23) corresponding to pure torque is
now written more conveniently in the form
~~qB ,-_8,, uqB u =znl a.p

-~ .v-1• sqn, s-1 +~ 1\·qn .v =2n.vGr/> . . . . . . . . . . (a37)

Transverse loading S, through Transverse loading S, through which are most simply derived from the rate of twist formula Eq. (a24).
·shear centre Es S, = 0 Expressing the solution as in Eq. (a29) we find the unknown G.f» from
shear centre Es S,=O Eq. (a34a).
Fig. 55.-Effective shear forces applied through the shear centre Es Note that the total shear flow q of Eq. (al4) may be written as
q=qE+qn .............................................. (al4a)
The shear centre allows us to define the loading alternatively by the where the engineers' theory shear flows qg are given by Eq. (a26).
shear forces through and the torque With the chosen basic system Eqs. (a23) and (a37) are particularly well
TE=-S.(xE-Xn)+Sx(yE-Yn)=To-SuxE+S,.yg .......... (a28) conditioned since the diagonal coefficients are predominant and at the most
only three unknowns are involved in a given quation. Direct solution by
about£,. elimination is quite easy and may be performed by slide rule even for a
To obtain the total redundancies Dx 111 and D. 111 we have still to determine high degree of redundancy; application of the relaxation technique is
the rates of twist .P. and .p,. Observing that the distribution coefficients superfluous. An additional virtue of the basic system of FIG. 52 is that the
a 111 are the same in the two equations (a25)-a direct result of the rigid Dxb distribution is quite close to the final dx distribution and the values of
diaphragm assumption which allows us to displace the transverse forces the redundancies are small.
anywhere along their lines of action-we can combine the indeterminate In finding on the other hand the d. 111 redundancies there is a conflict in
shear flows due to s.
and S"' into a single set the choice of the basic system. For when we cut the external wall in each
cell the Dub distribution is vastly different from the final one although the
Su A.. Ix 'S,. A.. I. A.. A.. A..
...... (a29) equations are as well conditioned as for the D, 111 since the matrix D is
qB.11= jaMG'~'"s +/aMG'I'"'s =aMG('f'x+'I',J=a.,1G'I'
;t y y X the same in both cases. Naturally, the criterion of good conditioning of
where the equations is always the most important one. The alternative basic
system in which the webs are cut gives a D.b distribution close to the
dO actual d. but the equations are not so easy to solve since each equation
cf>={{z=.f»,+.P• .......................................... (a30) involves all the unknowns. To satisfy both the above requirements it is
is the total rate of twist due to the given loading. necessary to make composite cuts, i.e. to cut both upper and lower
external walls and instead of the condition D.b =0 at a cut to take say
To derive the rate(s) of twist we may apply the equilibrium condition equal and opposite values of D 11 b at the two cuts of each cell. In any case
about D (see FIG. 53) the d. distribution is, in general, of small importance and a more approxi-
fqpnds=O .............................................. (a31) mate solution is acceptable.
where the sign of Pn is defined as for Po· Applying Eq. (a31) to the s. Generalizations of the Above Analysis
loading we obtain
The method given above for the determination of the statically equivalent
fDxbPnds+2'LdxMnM+2'LaMnM · G.f».t=O stress system in uniform cylindrical tubes may be generalized and applied
or • to tubes with conical or non-conical taper and with sheet thicknesses and
boom areas varying lengthwise. The angle of taper 28 is taken, however,
*See foe. cit. p. 129, 'The General Theory'. See also derivation of Eq. (al4) by the Unit Load
to be so small that cos· 28R:>! and sin 28R:>28.
method in Section 7.

55
 from Eqs. (a27). The engineers' theory of bending shear flows for trans-
verse forces through the flexural axis are now
Q.d., Q., d.
qE=p T,.+p T. .. . .. . .. . .. . . .. .. . .. . . .. . . .. . .. . .. . . .. (a45)
where d.,, d. are obtained for the root dimensions from Eqs. (a23) for
c/>.=c/>.,=0.
If the shear centre E, has been found we may calculate at any cross-
section the torque T E of all applied transverse forces about the flexural
axis. It is then preferable to calculate the Bredt-Batho shear flows qB of
the total statically equivalent shear flow
q=qE+qB .............................................. (a4la)
by a slightly modified version of the method on p. 55. Thus, Eqs. (a37)
for the redundant q 8 M become
(3IQBI-f3J.IIQB/l=2QI pi/JGcp
T, is the torque about axis V-A

-(3 N-1• NQB, N-1 +f3 NQBN =20.Npi/JGcp (a46)

The solution of which may be written
QBM=a.wGpi/Jcf> . .. . . .. . . .. . .. . . .. . . . . . . . . . . . . . . .. .. . . . .. . (a47)
Fig. 56.-Geometry and loading for conical tube Next we deduce from the equilibrium condition about the flexural axis
TE
Gpi/Jcf> N (a48)
1. Conical and Cylindrical Tubes with Similar Distribution of Material at 2p2~aMQM
All Cross-Sections I
Consider the conical tube shown in FIG. 56. At the root the sheet thick- Hence the shear flows QBM are
nesses for direct and shear stresses are t. and t respectively and the TE
(effective) boom areas B. At any intermediate section the corresponding QBM=aM N (a49)
thicknesses and areas are given by 2p 2 ~aMQM
I I
Finally, we derive the shear flow q 8 in all walls from Eqs. (a36).
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a38) If the torque is given initially about an arbitrary axis VA we can cal-

J
culate TE with the formula
TE=TA-pQ.(xE-XA)+pQx(YE-YA) ...................... (a50)
where where xA, YA are the co-ordinates of the point A at the root.
p=r/r. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a39)
2. Conical and Cylindrical Tubes with Arbitrary Variation of Boom Areas
rand r. being the distances from the apex V of the current and root cross- and Wall Thicknesses
sections respectively; for cylindrical tubes p = 1. 1/J and 1/J. are non-
dimensional functions of z or p. Here we investigate conical or cylindrical tubes with an arbitrary length-
In what follows under the present heading (l) all cross-sectional dimen- wise variation of skin thicknesses and boom areas. The engineers' theory
sions, areas and functions refer to the geometry of the root-section. shear flows are not any longer proportional to Q. and Q., and Eq. (a42)
As in the previous analysis we assume that the direct stresses are given does not apply. The concept of a flexural axis, either straight or curved,
by the engineers' theory of bending and write them in the form is also not any longer strictly true.
Under this heading all cross-sectional dimensions, areas and functions
M~ y Mv x are based on the current cross-section.
(J= p2!fJ, fx + p21/J, lv (a40)
where Mx, Mv are given by Eqs. (a2). Note that y, lx are based on the root
geometry. Since the stresses (a40) act along the generators they give rise
to shear resultants at the apex V in the y and x directions which are easily
found to be

r
M~
and r
M. .
respectively.
Hence the shear forces resisted solely by shear flows q are
M~ My
Q.=S.+-;:-. Q.,=S.,+,- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a41)
If now the transverse forces are applied through a line VD where D
is a point at the root the shear flow q at any cross-section may be expressed Fig. 57.-Equilibrium condition in element of conical tube
as (see also Eqs. (at4))
For the subsequent analysis we require the modified forms of the internal
q=~· ~"'+~"' ~: ·· · ·· · · · · ·· · ·· · · ·. · ................... (a42)
equilibrium conditions (a3) and (a4). Thus, we deduce immediately from
the geometry of FIG. 57 the equilibrium condition on a conical element on
where Q., Q., are determined from Eqs. (a9) with Q., Qx in place of s., S.,. the surface
Thus, - ~(rf) +~q =0

············. ....... (a43)

~r ~w
Q~= ?v_[(~~j~iL~:~· Q.,= ?:_[(~~j!j/~~:~ or
The cross-sectional distribution functions Dx. D. are obtained at the root (a51)
cross-section by the method of the previous analysis but with
where
p 2 1/Jcf>.~. (p 2 fcf>.,~) in place of c/>.t. ( cp.,;:) .... (a44) Hence
f =at, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a52)
~q I ~(r() .
in Eqs. (a23), (a24), (a25), (a29), (a32). Note that p21/Jcf>J,JQ., etc., are ~s=r ~=Df .................................. (aSia)
constants for the type of loading considered as the transformed Eqs. (a32)
show. where
~f f ~f f
The flexural axis is in the present case a straight line VE, through the Df="iir+-,.=-sz+-,. .............................. (a53)
apex and the shear centres at all cross-sections. We define this axis by the
shear centre E. at the root. The co-ordinates XE, YE are found as before In practical calculations we find '?!f/(Jz numerically by finite differences
56
The equilibrium condition Eq. (a4) for a flange B. at the intersection of
cover and web becomes here
d~. -q._ -q.d+qo+=O .................................. (a54)
where
P. =B.a. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a55)
To find the statically equivalent stresses for an arbitrary system of
transverse loads defined by the shear forces (and associated bending
moments)
S.(M.,) and Sz(M.)
and the torque Top and Bottom Nose Rear
Webs
Covers cell cell
TA 2. 3 3 ••
1. 2 •• 5
about an axis VA we proceed as follows. First we equilibrate the applied 9.10 1.1 7. I 6 7
0
1. 10 2.' 3. 8 • 0 7 5,6

bending moments by the E.T.B. direct stresses Length \<ml 30·0 30-0 30·1 30·2 17 20 20 16 10 65·7 14·6

-Y - x-
a=M.,T.,+Mvf. ........................................ (al) Th1l~iss 25 B B 2·5 1-6 1-6 1'1 1·1 ,.. 1·0 1·0

The co-ordinates x, ji from the centre of gravity and the corresponding •

moments of inertia may vary arbitrarily from section to section. Boom
1
10
2
9
3
I 7
5
&
Cell 1 n lll Il y Yl.
The statically equivalent shear flows are calculated as before in two parts Area
l<m'l
I 12 12 12 10 Area
(em~)
350 551 100 540 310 210

q=qE+qB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a56)
Fig. 58.-Geometrical data for six-cell tube
where qE are in equilibrium with Q., Q., with the imposed condition of
zero twist throughout the tube and additional shear flow necessary to The above analysis of a statically equivalent stress system in tapered or
balance the applied torque.* The solution for an N cell tube follows cylindrical multicell tub;:s could b;: used to write down the b0 matrix.
closely the method given initially and the selection of a basic system Using the information develop!d in S:!ction SC for the b 1 and f matrices
derives from similar considerations. The calculations must, of course, be we can calculate next the axial constraint stresses in the structures con-
repeated for every section to be analysed. sidered here. However, when an 'exact' analy3is is to b;: performed by
the matrix method in what is essentially a single continuous operation
Calculation of qE: the choice of the b 0 matrix should b;: guided entirely by the requirement
The shear flow qb in the basic system is obtained at all sections we want of simplicity. Hence, in such cases th:! sy3tem of indep;:ndent spars is
to analyse by integration from (a5la) greatly to be preferred to that of the multicell tube in the present examp.le:
see also Example 9b.
s
qb=JDfEds ............................. ............... (a57)
0
Numerical Example
where !E is calculated from Eqs. (al) and (a52). The necessary constants To illustrate the application of the foregoing analysis we determine
of integration to compute the shear flow in the webs, etc., are derived now the shear flow distribution in the single symmetrical six-cell section of
from equations of the type (a54). TheN unknown shear flows qEM at the a uniform cylinder illustrated in FIG. 58. It was found convenient to use the
cuts of each section are determined by the same method leading to metric system in this example. The direct stress carrying ability of the walls
Eq. (a23). We find has been replaced by effective flange areas at the web-cover intersections
(augmenting the actual flange areas there) so that the walls carry only
f3IqEI-/3[,JiqEJI=- I~bdS[ shear flow which is constant between flanges (t.=O). This procedure
I enormously simplifies the work and is quite accurate enough for practical
computations although naturally additional effective flanges may be
introduced at intermediate stations (see Section SC, p. 37). It should be
-f3M-l•MqE,M-1 +f3MqEM-/3M, M+lqE, M+l = - J~bdsM noted that since the distribution of direct stress is known (E.T.B.) the
M appropriate effective flange areas can be calculated explicitly without any
need for guesswork. Strictly, different areas have to be calculated for
- f3N-l> NqE, N-1 +/3 Nq EN =-
. . . . . . (a58)
N
J~bdsN bending about Gx and Gji but since in practice the horizontal shear force
S., is of much less importance it is usually sufficient to apply throughout
Note that all dimensions are based on the current cross-section. Having those calculated for bending about Gx.
solved these equations we calculate the shear flow qE with Eqs. (a13). For a singly symmetrical section under a shear force normal to the axis
of symmetry we are obviously only concerned with the D., (or d.,) dis-
Calculation of qB: tribution. Due to the assumption of t,=O the (constant) values of D.,b
in the upper and lower walls are zero and the values in the webs can be
The torque T Q of the shear flows q E about the VA axis is written down by inspection using the equilibrium equation (all) for the
N flanges. The Dxb distribution is shown diagrammatically in FIG. 59a. In
TQ=fqEPAds=JqbpAds+1'2:.Q.MqE M ........................ (a59) FIG. 59b we illustrate also the Dxb distribution for the alternative basic
I
system mentioned previously in which the multi-cell tube is reduced to an
where p A is the normal from the point A to the tangent at the wall; note open section by cutting N vertical webs. No further comment should be
the usual sign convention. Hence the torque T 11 to be resisted by the shear necessary to point the moral of this illustration.
flows qB is
TB=TA-TQ ............................. ............. (a60) Table I of Example 9a
The shear flows QB at any section may now be determined from Eqs. (a37) Values of JD~bdsMft and coefficients {3
and (a29) with M

G'I'
-1.= N Tn .......................................... (a61) Cell M I II III IV v VI
1'2:.aMQIII 350 555 600 540 390 210
I
----

3. Tubes with Non-Conical Taper 763 471 490 466 40-l 908
For tubes with non-conical taper we can find the statically equivalent
stress system
!E and qE+qB
by the method under (2). However, when finding the torque T B to be
•
+68
106 125
+120
125
+120
100
t96
--"!.'I-+50
carried by the qB shear flows we must make allowance for the torque Tp
carried by the direct flow IE and the boom loads. Thus
TB=TA-TQ-TP ...... , ............................. .. (a62)
D_,._b_f3M_._~_r+_l_____ 72~~500~\_1500~-- 9l0J _ 3120---

• qE may be regarded as quasi-engineers shear flow; see also J. H. Argyris and P. C. Dunne,
Joe. cit., p • .53 (Handbook). M
JD J'b
d.~.'i
t \ 7200 1 780J 1 o 1 -5~0:) 1 -6.1-80 \ -3120
57
 With D,b having (constant) values in the inter-cell webs only the six
compatibility equations are formed easily and systematically by means
of the arrangement in TABLE 1. All values in the table are obtained
directly from the dimensional data of FIG. 58 and the D,b distribution of
FIG. 59a.
The equations for the unknown D.,M are therefore (see Eqs. (a23)) (a) GOOD CHOICE
763D., 1 -106Dxo = -7200+ 700Gc/>fy

-106D., 1 +471Dxo-125Dxa = -7800+ lllOGc/>~y

0+1200Gc/>~
-125Dxa+466D,. -100 D,s =
.JJx
+ 5400 + IOSOG'I'S y

-lOODx<+ 404Dxs-62·5D.,.=+6480+ 780Gfl

s.
-62·5Dxs+ 908Dxo=+3120+ 420Gc/>&_
Sv
· · · · · · .. · ·........................ (a63)
Also the equilibrium Eq. (a3l) for torque about Dis
700Dx1 + 1110Dxo+ 1200Dxa+ l080DX4 + 780Dxs+420Dxa+94600=0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a64)
Writing the solution of Eqs. (a63) in the form
(b)
D.,M=d.,M+aMGc/>~v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a25)
BAD CHOICE
Fig. 59.-Dxb distribution in six-cell tube for alternative choices of basic
Grpl.,/S. is determined from (a64) and the results are as in TABLE n. system
Table II of Example 9a
Values of dxM and DxM (em. 3)

M dxM aM DxM
~ 10.000 kg.
I -12·17 1·46 -20· 13

II -19·62 3·90 -40·92

III - 1·15 4·52 -25·85
IV +15·79 4·20 - 7·16
I
v +20·71 3·08 + 3·89 f-z2·3cm-; q in kg/em.
VI + 4·88 0·68 + 1·16 Fig. 60.-Final shear flo;v distribution in six-cell tube

From Eq. (a64) G s.=-5·46 l"'

Shear Centre
Since the present cross-section is singly symmetrical, the shear centre
Adding these DxM to the Dxb we obtain the total Dx distribution due to lies on the axis of symmetry and the first of Eqs. (a27) (with 1.,.=0) gives
Sv through D. Since Dxb is zero in the outer walls we have simply (see also immediately the XE co-ordinate. Alternatively we note that the shear
Eqs. (al5)) flows dx have a resultant s.=I., through the shear centre and we can
for outer walls D.,=DxM modify slightly Eq. (a64) for torque about D to give
for web between M and M+1 cells D,=Dxb+D.,M-Dx•M+l 700d.,1 + 111 Od.,2 + 1200d.,3 + 1080dX4 + 780dx5
The total actual shear flows under the 10000 kg. load at Dare +420dx6 +94600=/.,(xE-XD)
Dx 10000
Therefore
q=Su I.,= 7992 Dx=1·25D., . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a64)
in kg./cm., the actual values being shown diagrammatically in FIG. 60.

58
(b) The Four-Boom Tube with Deformable Ribs and some more General
Structures
In this example we determine, using the force method of analysis, the
'exact' stress distribution in the idealized four-boom tube shown in
FIG. 61. The investigation is carried out first by the Swmethod and
illustrated in a numerical example. Subsequently, we analyse the same
example by the general matrix method of Section 8C. We show also how
the matrix formulation may be used with advantage in the more interesting
case of a six-boom tube with or without intermediate spar web. It is hoped
that these simple applications of the matrix method will condition the
reader to the new ideas and show him their power and basic simplicity.
Consider the cylindrical tube of FIG. 61 with a singly symmetrical trape-
zoidal cross-section the flange cross-sectional areas Band the wall thick-
nesses t of which may vary arbitrarily length-wise. Loads are applied only
at the rib positions
1, 2, ........ , i, ........ n
in the form of shear forces
Rat• Rbl, ...... .. , R.;, Rbt• ........ , Ran• Rbn
and moments
Mato Mbto ........ , Ma;. Mb; • .. . ... . . , Man. Mbn
at the front (a) and rear (b) spars. Following the general discussion on the
idealization of aircraft structures given in Section 8 C (pp. 'J7and 40) we
assume that the walls carry only shear stresses (the direct stress carrying
ability is allowed for by suitably increasing the flange areas). The shear
flow is hence constant in any field of each bay since changes in the z-direc-
tion can only be brought about at the ribs. It follows then that the end
loads in the flanges vary linearly between ribs and that a knowledge of the
flange loads at the rib positions suffices to determine them everywhere.
Having found the flange loads the corresponding shear flows are deter-
mined easily from the flange load gradients and the condition of equilibrium
with the applied shear force and torque.
In the tube shown in FIG. 61 there are n ribs including those at the free
and built-in ends. At each rib position there are four flange loads and only
three equilibrium equations are available for their determination. Hence,
nothg that the flange loads at the tip are zero, the degree of redundancy is
(n -I). This trivial result is confirmed by the general Eq. (247a) on p. 38
by substituting N= I, f1=4, a=n.
In selecting the basic system many choices are open to us. We may, for
example, make a single cut in one of the flanges at each rib statiori to reduce
the structure to a statically determinate three-flange basic system. Here,
however we calculate the statically equivalent stress system by the E.T.B.
and the Bredt-Batho theory of torsion, a more general example of which
was investigated in Section 9a. In this choice, instead of making a single
cut we have, in fact, cut all the flanges to allow the relative warpi?g con-
sistent with the statically equivalent stresses while at the same tune the
direct stresses are transmitted across the cuts ; • see also the discussion in
Section 8C, pp. 32 and 33. The redundancies then consist of self-equili-
brating flange load systems at the (n -1) rib stations. A suitable and sym-
metrical measure of such a system is the boom load function P introduced
by Argyris and Dunne.t We prefer to use instead a slight variant of P~ the
Y-system introduced on p. 39. The (n -I) redundant Y are determmed
from the compatibility conditions of warping at the (n -I) rib stations.
As a further alternative procedure we could, of course, choose as basic
system the very simple structure consisting only of the two spars acting in-
dependently, i.e. we cut the top and bottom covers of the tube. However,
• Actually, we may consider this basic system as .also derived by a single cut from the given system .
Thus if at the cut Bange we - l y the correspondmg E.T.B . Bange load and allow there any out of
place' movements, the stress distribution in the other structural elements is obviously that of the E.T.B.
and Bredt-Batho theories.
t Joe. cit. p. 38.
l:·
z
Fig. 61.-Four-flange tube. Geometry and loading.
except for very flexible covers and ribs our previous choice is much closer
to the final correct solutions. The latter method is hence to be preferred
when the design work has to be checked solely by a good statically equiva-
lent stress system as is, in general, the case when no automatic digital
computer is available. Subsequently the exact distribution, if found at all,
is obtained only after completion of the design work-usually by a more
clumsy version of the O;k method.
On the other hand when we use the matrix formulation in conjunction
with a digital computer and derive the complete stress distribution as a
single process it is preferable to select the simplest possible b 0 matrix (see
also the discussion on p. 39). Then the basic system formed by the inde-
pendent spars is obviously indicated. An important advantage of this
system is the absence of rib stresses if external loads are only applied in the
plane of the spars. In multispar construction the simplicity of the basic
system consisting of independent spars in comparison with the multi-cell
system of Example (9a) is even more striking. If it is necessary to taKe into
account external loads applied at intermediate points in the ribs and/or
if spars are interrupted we may select as basic system for the matrix type of
analysis the grid formed by spars and ribs (without covers}-still a very
simple structure in which to find the b 0 matrix.
(I) Analysis by the O;k Method*
The statically equivalent stress system
Following our previous discussion we select here the E.T.B. (or quasi
E.T.B.) and Bredt-Batho stresses as statically equivalent stress system.
The flange loads are calculated at the rib stations with the effective areas
B; there. Thus, the m'th flange load at the i'th rib is,
Pom;=M.,l'mi Bmi ........ • • • ... • •;. • • • • • • • • • • • • (bl)
lcr.i
where M.,; is the bending moment at the i rib due to the applied loads Ra
• The particular case of the four-Bange single-cell tube under a given loading has also been treated
by W. J. Goodey in 'Two-spar Wing Stress Analysis', AIRCRAFT ENGINEERING, Vol. XXI, No. :U7,
p. 287, September 1949; No. 288, p, 313, October 1949; No. 289, p. 358, November 1949. 'J'here.
the author uses the Castigliano technique to formulate equations analogous to the three- and f!Ye-joiat
equations given here with essentilUiy the same basic idealizations but with tbe effect of taper induded.
A comparison of the theoretical results with experimental strain measurement• is aloo JliYeo.
59
 l
n;~

"'

,.,
'
o•2

Fig. 63.-Fiange loads due toY; = I and Y;+ 1 =I (See Fig. 4l)
,.,

i,,
Fig. 62.-Rib shear flow In four-flange tube

and Rb. The corresponding shear flows in the walls, observing that they are
statically determinate in a single cell tube, may be expressed in ·the i bay as
TAi
q .; = q..4;+2n .. . ... . . . .. ............. . . .. . .. ... (b2)
where q...4, is calculated from the flange load gradients
(d~;"') i ' ba.. p omi-tom, i-1 . . . . . . . . . . . . . . . . . . . . . . (bJ)
with the condition of zero torque about the chosen reference point A and
T.t is the torque about this axis (see p. 56). There appears little virtue in
single-cell uniform tubes or tubes with similar longitudinal variation of
stress carrying material to select A at the shear centre £ ., i.e. to write q. as
q . = qg+qB •. • . . . ............... . .. ... ... . . ,(b4)
(see alsop. 56). Similarly it is of no advantage in tubes with arbitrary
longitudinal variation oft and B to split q. as in (b4), in a quasi-engineers'
theory shear flow qE for zero rate of twist and a Bredt-Batho shear flow ,.,I
qB for a torque TB to be found by Eq. (a60). Naturally, in multi-cell tubes w,., .~h!h
it is on the other hand preferable to use the presentation (b4) for the n. • •
calculation of q •. Note that since we assume that flange loads vary linearly I,,,
between rib stations, systems (bl) and (b2) only represent E.T.B. \
w~,::::- h.•/\
at any cross-section between rib station if the flange areas are constant
or vary similarly within each bay. 4•2
To complete the statically equivalent stress system we have to calculate Fig. 64.-Shear flows due to Y; = l and Y; +t = I (See Fig. 4l)
the rib shear flows. Consider to this effect FIG. 62 where rib i is shown Arrows indicate actual direction of shear flows for positive Y systems
isolated under the action of the equilibrating forces; in this it is assumed
that the moments M a;, M bi are applied as flange loads The coefficients O;o and Otk
± Mad h., ± Mb;/hb We write all 8 coefficients in the split form
For a trapezoidal rib the shear flow is not, of course, constant but re- 0= 8, + 8. + 8, .... . .. .. .... . .... .. ..... . ..... . (b8)
solving forces horizontally for the upper or lower half of the rib we find the
average shear flow where the suffices f, s, r refer to the flanges (shear), walls and ribs respec-
tively.
Ma; M bt
qor; = q o,i-l -q 0 ;- 7l; - 71; ......... ... . . ...... . . (b5)
The coefficients 0; 0
where q 0 ,;_ 1 and q 0 ; are the shear flows in the covers of the i - I and i bays For tubes with similar longitudinal variation 1/J, of flange areas Bin each
respectively. bay O;o1 is zero since the direct stresses associated with flange loads (bl)
are at any cross-section those of E.T.B. and hence generate only planar
The self-equilibrating stress systems Y; displacements of the flanges. On the other hand for arbitrary variation of
We show in FIGS. 63 and 64 the flange loads and shear flows due to flange areas system (bl) does not represent, in general, E.T.B. and there
Y, = I and note that they extend only over the bays (i - I) and i; these arises then a term 0; 01 which may be calculated from
figures reproduce in a more convenient form for the present purpose the
., [ JP;,.P J.. .... .. ...... .. . .....
i+ l
information given in FIG. 43. The system Y is connected to the boom load ~
function P of Argyris and Dunne by O;ot = "'-- om
~dz ( b9)

U.+ ~J
m= l m
i-1
Y= P . . . . . .. . . . . ... .... ...... . . .. (b6) where P;,., P ... are the (linearly) varying flange loads due to Y, = I and
external loads respectively. The integration in (b9) is restricted to bays
The Y-systems are determined from (n - 1) equations of the type (282a) (i - 1) and i since the influence of Y, is restricted to this region. In general,
which take here the form the contribution of (b9) to 8;. is negligible and could only become signifi-
n- 1 cant in a bad design.
~ o,kYk + o;. = O .. ... . . ... . . . . . . ... . . . .. . ... (b7) The general formula for 8,., follows immediately from the unit load
k=l
method as,
s••.=iJ Jq1'ds dz
where 8,. is the relative warping at the rib i due to the chosen statically
equivalent stress system. Having the Y redundancies we find the 'exact' .... . . . . .. . ... . ... .. ... . (b!O)
stress distribution by superposition of the statically equivalent stress
system and that due to the Y;'s. where q; are the shear flows due to Y, = I and the subscript s denotes inte-
60
gration over the walls. Again we need only consider bays (i- I) and (i) The idealization of constant flange areas in each bay has been discussed
and we obtain immediately with the notation of FIG. 64 on p. 37.
I 4 Proceeding now to the contribution of the Sib we note that its general
Slo• = G ~ W,.,,.+l{ [q om,m+1,8m,m+1]i- [q om,m +1,8m,m+di-1} (btl) expression is
m-1
where s
ik.•= IIqiqkd
Gt s d z .......................... (b20)
,\'m,m+l
,8
m,m+1=t--
m,m+l
···························· (b12) and derive easily from FIG. 64 that
and the subscripts i and (i -I) indicate that the expressions in the square
brackets have to be evaluated in the i and (i -I) bays respectively.
st.1 -·1··"
-s -
t-t.tR--
~ 'Yi-1
G /;_ 1
I 'Y,
Note S;.;,,.,=8q1.i .• =c-c; 7; .... (b20a)
Using Eq. (b2) in each of the expressions in the square brackets we find
4 4 TA 4
m ~ 1Wm,m+1q om,m+1,8m,m+1 =;,~,}qAw,8]m,m+1 + 20 m~ .£w,8]m,m+1· ..... (b 13)
where
4
This formula may be simplified considerably if we refer the torque to a 'Y= ~ (Wm.mn) 2,8m.mt-1 .................... (b21)
point S whose co-ordinates x 8 , y 8 are m I
I
J
4 For the last term sikr of sik we find from FIG. 64 that it is zero but for
~ £qAw,8]m,m+1 j
k-i 2, i --1, i, i+ 1, i+2
Xs-XA=-20 m~l 4 We obtain,
S .... [ ,8] ............ (b14)
Y m~l W m,m+1 n (w 41 ) 2 d I 1
St.i-2,r=S;-2.ir=(-Gt) ,-,-=2(h a +h)
r r i-1 i-2 i--1
(-Gt) ,-,-
r r i-1 i-2 i-1
Ys-YA=O b

where s. is the shear force. Eq. (b13) reduces then to

4 ~0 .
m~.£wq.,8]m,m+1= 2 0G' ....................... (bl3a)

where r.~ is the torque about S and

4
0= ~ [w,8]m,m+1 (b15)
m=l
The coefficient S1., of Eq. (btl) can now be written concisely as (b22)
1 Finally we find the total S1k coefficient from
S; •• = 2oG [Ts181- r.~. t-10t-1l ............... . (blla)
S;k =Sikf+Stks+otkr
S 1 is, of course, the zero warping centre* for the bay i of the tube. The
expressions given above for its co-ordinates allow for the effect of different This concludes the derivation of all formulae for the S-coefficients in the
taper of the flanges (within the present assumptions). For the uniform Eqs. (b7) for the unknown Yi; Solution of the latter system of equations
bay Eqs. (b14) reduce to the well known result* yields the total stress distribution in the tube. We illustrate our analysis on
4
a numerical example further below.
~ [Dr.,Bw]..,,m+l
Xs - XA
r.m~"l
2.1.~~ .............. (b14a) I::jfect of Cut-outs
cc - 4 -'-------
~ [,8w]m,m+1
m=l
where D,. of Eq. (al4) is obtained by statics, seep. 53.
The last contribution to S1• is that of the ribs. FIG. 64 shows that
stor =o[(q.,ql•)
G,t, 1-1
+(q.,q;,) +(q.,q;,)
G,t, ; G,t, 1+1
J
Rib k removed

I f"\ ha+hb
w 41 =-ha+hb and .1.~=-2-d
1
G,1 and t, 1 are the shear modulus and thickness of the i'th rib. h.
The total value of 010 is now given by
s~.=S;.,+St •• +S;., ........................ (b17) '1
h.
The coefficients Slk k-2

FIGS. 63 and 64 indicate that sikf and sib can only arise for
k=i-1, i, i+l
The contribution S1kt of the direct strains in the flanges may be expressed
as
slkt=IY;Yk ~ ~8m2 dz= JY;Yk<l>dz ....................
m=l
(bl8)
where wm Y1(z) and w,. Yk(z) are the linearly varying flange loads due to
Y1 = I and Yk =I respectively and Yk. 1 system when rib k is removed
4 w,.2
<I>= ~ EB · ............................. (bl9) Fig. 65.-System Yk-t =I when rib k is removed
m=l m
If we assume that the flange areas and hence <I> are constant in each bay If a rib k is removed bays k --I and k merge into a single bay and the
we find immediately, calculation proceeds by the previous analysis observing that the number of

ll ....
unknowns is, of course, also reduced by one. FIG. 65 shows schematically
the flange loads due to Yk-" the influence of which extends now over the
three bays k- 2, k --I and k; there is no system Yk.
(bl8a) If in the (k ~I) bay a wall is removed it is again possible to use our
analysis with few alterations. Thus,

J (a) the statically equivalent stress system Pmo• q. in the open tube bay k is
calculated with the engineers' theory and torsion-bending theory. •
• See AJ'IIYris and Dunne loc. cit. p. 84 and also Argyris and Dunne, Stress Distribution in Conical
Tubes. puhli•hPd by R.Ae.S. as Vol. Ill of Stressed Skin Data Sh-.ts. • See J. H. Ar~yris. 'The Open Tube', AIRCRAFT ENGINHRING, Vol, XXVI. No. 302. p. ll2.

61
 " --~.

All loads rn lb.

10 __::..:.-,... ••

Pa·-lf '•xact '-defarmablt ribs

(IIHO')
5 '•11ac:t' · rigid ribs
E T.B

F"lan;. I oa<H
..,

q,,
OMI
J50 ' -

~
140

.,
-- -·-

1------
--r-

-- - -- -
-,
~-
---~~-=-.:.-=~-~-
:..... ...;
(

'._ _ _ _ _ _ _J
~ r-----i-----~-----4------1------+-

... Rib stahon

Fig. 66.-Fiange loads and shear flows In extended Yk system when web Fig. 67.-Numerlcal example of 81~~; method. Flange loads and shear
or cover Is removed in (k - I) bay flows In spar 'a'
(b) the unknown system Y~:-1 and Y~~; merge into a single system which we
denote by Yt and illustrate in FIG. 66. In this system the flange loads All other 8;;. and 81, , are given by formulae (b20a).
are constant in bay k due to the kinematic freedom of our open tube. Formulae (b21) for 8;;, etc. take for i= k the form,
Observe that the system of (n - 2) unknown Y;'s consists now of d 1 1
Y1, Y2, ... . .... , Y1:-2, Yk, Y "+b ... . .... , Y n-1 8k-s,kr=2(ha+h&) (Grtr)i:-2 l~c-alk-2
(c) there is no term 81:-1,o·
In general, the coefficients 81• 1 are again either zero or negligible. Should
8
k- 2 , kr -
_ d _1 [-1-- c--~ +-~ )+
2(ha + h&) /k- 2 (Grtr)k-2 /1:-2 /k-3
1 _1
(Grtr)i:-1 /1:-2
J
they be important they may be calculated with Eq. (b9) noting that for
8k-1,kr = O
8~~;.1 the integral extends from ribs k - 2 to k + 1 and that in the open bay
P. includes the torsion-bending flange loads. 8 _ d { [ - 1 - + -~-Jc-1 )2+
The terms 8,.,
may be obtained from formula (b11) or (blla) but for kkr - 2(ha + hb) (G,trh- s (Grtr)i:-1 /k-2
8~:-1 .•, which is zero and 8~~;., which is calculated from (see also FIG. 66)
1 4
8~:o,=-G ~ Wm,m+1{[qom,m;-1,8m,m+1Jk-[q..,,m+1,8m,m+lh - z} · • · · • • (bllb)
+ [(Gr~r)k + (G,t~)"+J (~) '}
m= l
Similarly, the 8 1. . are found from Eq. (b16) except for 8kor which is, 8"+1•kr =
d I[ 1 I I ( I 1 )]
2(ha + hb) /;, (G,t,)k /;,+(G,t,h+1 "J;,+/"+1 · · · · · · · · (b22a)
8 _ tj_{_1 [(qor)
~~;.,. - 2 1~:- 2 G.t. 1:-1
-(qor)
Grtr t-2
] -J. [(q.,.)
It G,t, "+t
- (qor)]}
G,t, 1c
The remaining 8, coefficients are derived as before from (b21). It will be
seen later under the matrix analysis in (3) how much more convenient the
use of the H matrix device is for the treatment of cut-outs. This concludes
(d) there are no terms 8~:-1,i· the long-hand analysis by the 8;~~; method. In what follows we illustrate it
Under the assumption that flange areas are constant within each bay by a numerical application.
we obtain
(bl8) 2. Numerical Example of the S;k·Method
All other 8;;1 and 8;,1 are unaffected except that O;,k-1,1 =0 Consider the singly symmetrical cylindrical tube with six bays shown in
FIG. 67 under the action of rib loads Ra1 and Rbi· All.bays have the same
Formulae (b20a) for 811, etc. become in the case of i= k, length
8~:-s,to= -J 'r i:-2
k- 2
ll 1= 20 in.
The overall cross-sectional dimensions are given in FIG. 67 from which
8t-1,h = O ( .... (b20b) we derive, using FIGS. 63 and 64, thew-functions as,
~ _G
_!['Yk-2+'Yk]
Otk•- 1~:-a /~
J W1=+1/6, W2 = - ·1 /6, W3 = + 1/10, w,=-1/10 in.- 1
w 12 =+5/48,w 23 = - l/16,w 84 =- 3/80,w 41 =-1/16 in.- 1 .. (b24)
62
The ratio GjE, of shear modulus of walls to Young's modulus of flanges, Similarly we find for the contributions 8;or ofEq. (b16) using the values
is taken as of (qor/G,t,); in TABLE II.
G/£=0·385
in all bays. I I I I
TABLE I gives the flange areas Bm and moment of inertia lx at the rib +I -1 0 0 0 0 0 I I (q or/G,t,)! I
stations, the wall thicknesses t,.,m+1 in each bay, and the thicknesses t, and
shear moduli Gr of the ribs. For the calculation of the influence coefficients -1 +2 -1 0 0 0 0 I I (q or/G,/,)21
etc., we assume constant flange areas in each bay and compute these as the
mean values of the flange areas at the rib stations. 0 -1 +2 -1 0 0 0 I I (q or/G,t,)al
Having calculated the ratios flm.m+l of the walls and the functions <I> and
'Y of Eqs. (bl9) and (bll) we determine easily the coefficients 8,kt and 0 0 -1 +2 -1 0 0 (q or/Grlr)J I
8,ks of Eqs. (bl8a) and (b20a). These are also given .in TABLE I.
TABLE II presents the statically equivalent stress system from which we 0 0 0 -1 +2 -1 0
derive the term 8,., of Eq. (bll). •
With the information given in TABLES I and II we can obtain the re- 0 0 0 0 -1 +2 --1 (q or/G,I,)s
dundancies Y1 to Y6 for the case when the ribs are taken as infinitely rigid. _j
The relevant equations are in matrix notation (q ;,..(G,fr)1 I
_j

14. 108 - 3 . 560 0 0 0 I

0 1151 I
-3·560 29·103 -3·116 0 0 0 - 174
0 -3·116 31·635 -2·380 0 0 -2821
=10-6
0 0 --2·380 35·394 -2·642 0 2247
0 0 0 -2·642 39·583 -1·9381 582
0 0 0 0 -1·938 44·742 1 I Ys 479 .............................. (b27)
_j L _j

I I We may now compute the total 8;k and 8;. coefficients to find the redun-
-2776·4 1
I dances Y; for deformable ribs. The equations are
541·8 I
2818·7
I 16·800 -8·33~ 2·083 0 0
............ (b24) I -8 · 334 40·126 -II ·448 2·083 0
- 364·2
272·2 2·083 -11·448 44· 133 -10·712 2·083 0

141·8 0 2·083 -10·712 47·892 -10·974 2·083

L 0 0 2·083 -10·974 52·081 -10·270
the solution of which is in lb. in.
0 0 0 2·083 -10·270 57·240 I Ys I
Y 1 = -19573,Y2 =422, Y3 =8915, Y 4 = -485, Y5 = -737, Y 6 = -349 .. (b25) L _j L _j
Next we consider the effect of rib deformability. Eqs. (b22) for 8;kr are easily
calculated directly but it may be convenient also to use their matrix formu- (continued on next page)
lation for their evaluation (see also Eq. (258) on p. 41 ). We have

I I
-1 0 0 0 0 0 I -1 0 0 0 0
-1 2 -1 0 0 0 0
(G,t,h 0 0
I -I 2 -1 0 0 0
O (G,t,h O
0 -1 2 --1 0 0 0 0 -1 2 -1 0 0
0 (G,t,)a 0
0 0 -1 2 -1 0 0 0 0 -1 2 -1 0
0 0 0 -1 2 -1 0 0 (G,t,)4 0 0 0 0 -1 2 -1
0 0 0 0 -1 2 -1 0 (G,t,); 0 0 0 0 0 -1 2
L
I
_j
0 (G,t,)s 0 I o 0 0 0 0 -1
L _j
0
O (Grtr)1 I
L _j

2·691 - 4·774 2·083 0 0 0

4·774 11·023 8·332 2·08~ 0 0
2·083 8·332 12·498 8·332 2·083 0
=10-" i .......................................... (b26)
0 2·083 8·332 12·498 8·332 2·083
0 0 2·083 8·332 12·498 8·332

0 0 0 2·083 - 8·332 12·498

L
• The small term 6iof arising due to the dissimilar taper of front and rear spar flanges has heen
ignored here (See p. ~0.)

63
 TABLE I of Example 9b
Cross-sectional dimensions and 3;kf, 3;ks coefficients
Rib 2 3 4 5 6 7
Bay 2 3 4 5 6
B 1 =B2 (in. 2) 0·65 0·65 0·65 0·45 0·35
0·55 0·25 I
B 3 =B, (in. 2) 1·20 I 0·80 0·40 0·40 0·40
0·40 0·40
1., (in.')
tr (in.)
71· 7
0·08
51·7
0·036
31·7
0·036
28 ·1
0·036
26·3
29·9
0·036
0·036
I 24·5
0·036
G, (lb./in. 2) 3·85 I 2·50 2·50 I 2·50 2·50
2·50 2·50
Mean B 1 = Bz (in. 2) I 0·65 0·65 0·60 0·50 0·40 0·30
Mean B3 =B• (in. 2) I 1·00 0·60 0·40 I 0·40 0·40 0·40
t 12 (in.) 0·048 0·048 0·048 I0·036 0·036 0·036
t 34 (in.)
t 23 = t n (in.)
I 0·080
I 0·048
0·080
0·048
0·064
0·048
0·064
0·048
0·036
0·048
0·036
0·048
~12 I 125 125 125 167 167 167
~34 I 125 I 125 156 156 278 278 I
~2a=f3n
<D (in.-•) Eq. (bl9)
I 501
0·1054
501
0·1186
501
0·1425
501
0·1610
501
0·1887
501
0·2350
I
'I" (in.- 2) Eq. (b21) I 5·446 I
5·446 5·490 I 5·945 I 6·118 6·118 I
3i,i-1.! X 10"} 3 516 3 960 4 753 5 371 6 297
3iifxl0 8 Eq.(bl8a) 7·031 14·952 17·427 20·247 23·333 28·271
3i,i+I,f X 10 8 3·516 3·960 4·753 5·371 6·297
3i,i-t,s X 10 8 } -7·076 7·076 7·133 8·013 8·235
3;;. x 10 8 8 Eq.(b20a) 7·076 14· 151 14·208 15·146 16·248 16·469
3;,;+1•• X 10 -7·076 -7·076 -7·133 -8·013 -8·235

TABLE II of Example 9b
Statically equivalent stress system and 3; 0 ., coefficients
k.ib 2 3 4 5 6 7
Bay I 2 I 3 4 I 5 I 6 I
Ra (lb.) 1000 750 150 600 200 450
Rb (lb.) 200 750 1000 400 700 100
s. (lb.) 6300 I
5100 I 3600 I 2450 1450 550
Ta (lb. in.) about spar 'a' 75600 I 70800 I 52800 I 28800 I 19200 2400
M, (lb. in.) X 10 3 38lJ I 263 I 161 89 40 II 0
10570 9920 9870 4910 1920 440 0
I
·--Pol ='Po2 (lb.)
Poa- --Po• (lb.) 32510 I 20350 I 10160 5950 2850 I I 840 0
q 012 (lb./in.)
q o34 (lb. /in.)
1 -2t6·9
I -424·8
134·9
-377·2
I 243·4
-216·3 -133·2
I
171· 9 87·0
-87·6
I 41·8
-21·9
qo2a=qon (lb./in.) --184·0 -137·2 5·5 -21·8 -12·8 I I -19·9
q 0 , (lb./in.) Eq.(b5) 184·0 I -50·8 I 138·7 I 27·3 I 9·0 I 7·1 I 19·9
'L.w{3qo I 10689 I 8603 I - 2249 I 847 I 201 I 747 I
3;., x 10 6 Eq. (btl) 2776 -542 I 2819 364 272 142
(q ;,_;G,tr) X WS 597 I -564 I -1541 I 303 I -100 79- I -221

I I leading to the three or five-joint equations, provides a practical and power-

I -3473·0 I ful method for the determination of the primary stresses in such a struc-
I I ture. However, in more complicated cases, with more than one redundancy
arising in each bay, the more systematic technique of the matrix formula-
1 652·2!
i tion and solution is the only natural tool. Even for the four-boom tube
4511·3 the analysis under a general loading and the calculation of the flexibilities
. . . . . . . . . . . . (b28) requires the use of matrices for its practical realization .
-1712·4 Therefore, for comparison with the long-hand solution and to illustrate
the way in which the method is applied in practice we analyse the same tube
77·0 I as previously under an arbitrary set of forces applied at the rib stations
and calculate also the flexibility matrix for the points at spar-rib inter-
I 429·2 sections. Finally, we apply the H-matrix device of p. 39 to find the stress
L _j distribution when one web field is removed.
which yield (in lb. in.) Although the matrix method is naturally associated with the use of
automatic digital computers and large numbers of redundancies it also
y 1 = -21898, Y2 =297, Y3 = 11066, Y. = -1243, Y 5 = -721, Ys= -834 (b29) proves extremely useful in smaller problems with more elementary methods
Having the Y; we can find the total st~ss dis_tribl!tion in the given struc- of computation. Thus, the calculations described here were all performed
ture. Thus, for the flange loads at the nb statiOn 1, on a conventional desk type calculating machine and it may be of interest
that the complete calculation-up to and including the flexibility matrix-
Pm;=Pomi+ Y;Wm .......... · · · · · · • • • • • • • • • • (b30) occupied a total time of about 40 man hours.
and for the shear flows in bay i The applied loads R; are numbered consecutively from I to 12 as shown
in FIG. 68 which gives also the numbering of the elements. As mentioned
.......... (b30a) in the introduction to (9b) we consider as basic system the two spars acting
independently as beams and it is obvious that the top and bottom covers
and the ribs are unloaded in such a system. The rib booms are loaded
Typical results for the flange load P and shear flow q are plotted in FIG. 67.
neither in the basic system nor due to the redundancies and need not be
considered at all.
3. Application of the Matrix Force Method to the Same Example Due to our choice of basic system it is apparent that in the matrix a
The long-hand method of formulating t~e equations for the re~un­ considerable number of zeros will arise and to obviate carrying these we
dancies in the analysis of the four-boom smgle-cell tube JUSt descnbed, advance the partitioning of bo in Eq. (253) a stage further and write,

64
 1 ........ 6 7 ....... . 12
,----A-----, ,-_A__---,
r
bola 0
r I
b.z 0 bolb

b •• 0
b=
0 - -- - - ..... (b31)
bow bon· a 0
I Edernal loads

~ bOT I
I - - - - - - - - - ,_ - - -
0 bnwb
L _j
I 0
L Basic system
for b0 matrix
where the numbers above the columns refer to the loads R;. Suffices a and
bare used to denote front and rear spars respectively. In addition, we take
advantage wherever possible of the symmetry of the structure by writing
terms for the top surface only. Thus b. 1• is the matrix of longitudinal
flange loads in the top flange of spar 'a' due to loads R 1 • ••••• R 6 and has
only 12 rows, i.e. one for each end of each element of flange (see discussion
on p. 40). Numbering of elements is from root to tip in each case (see
FIG. 68).
The six redundant force systems of the Y-type are applied as in the
previous analysis and the matrix b 1 is partitioned in a similar way to b •.

Numbering of elements

.......................... (b32)
blu:a

Y,

Y,

The corresponding form for the flexibility matrix f of the unassemblcd

elements is

flu 0 0 0 0 0

0 f,b 0 0 0 0

0 0 ( 0 0 0 Y- systems
for b: matrix
f= . . . . . . . . . . . . (b33)

,,
(se-e F1gs. 63 and 64)
0 0 0 fwa 0 0

0 0 0 0 fwb 0
Fig. 68.-Four-flange tube. Key diagram for matrix method of analysis
0 0 0 0 0 f,
I 2 3 4 5 6
where f,., f," are obtained from Eqs. (257) and f., to f, from Eqs. (258). r
2 3 4 5 6
Carrying out the various matrix multiplications we can put the products
b 1'fb. and b 1'fb 1 of Eq. (259) in the forms, 0 2 3 4 5 JJ
bl'fb.= [[bl'fb.],. [bl'fb.J,b] _j_ [[bl'fb.],.." [bl'fb.],b] ...... (b34)
0 2 3 4 5
I
bl'fbl = [bl'fbr], a+ [b1'fb1J1 b + [bl'fblJ., + 0 0 2 3 4
}2
[bl'fbdwa+[bl'fbl]wb+[bl'fbl]r ...................... (b35)
In the same way, the product F.=b/fb, which is the flexibility matrix of
th~ basic system under the external loads (see Eq. (230) is
r 1 20
0
0
0
0 0
2
2
3 4
3
}3
+ I [bo'fb.],." I bola·= -6 0 0 0 2 3
........ (b37)

}4
0

I o [bo'fb,J,.." .... (b36)

I 0 0 0 0 2
L _j
I
The above partitioning of the various matrices and the computation of I 0 0 0 0 2 I
the tlexibilities as the sums of the component f!exibilities is not of course
strictly necessary. However, it usually happens that the number of struc- I 0 0 0 0 0 J5
tural elements is considerably greater than either the number of external
forces or the number of redundancies and hence controls the order of 0 0 0 0 0
matrices to be handled. For example, in the present problem the full b 1
matrix has 49 rows whereas in the scheme used above the highest order
I
I 0 0 0 0 0 0
}6
matrix is 12 x 12 determined by F •. Although the reduction is necessary L _j
here for manual computation it is obvious that such a method is required
even with automatic computation for actual aircraft structures where the The numbers in italics over the columns of the matrix refer to the exter-
degree of redundancy may be of the order of 200 or so. nal loads R1 • . • . . . R 6 and the numbers opposite the pairs of rows are the
With the extremely simple basic system chosen the b. sub-matrices can number of the elements of the flange 'a' in each pair of rows; the first one
now be written down by inspection. As a typical case we find for the top is for the end of the element nearer the root and the second row for the
flange of spar a the matrix b. 1• of tensile flange loads due to R end away from the root. (See FIG. 68 and p. 40)
65
Fig. 69 (right).-Four-flange tube. Flange loads and web shear flows due to
single load R3 = 1.000 lb. Application of H matrix to cut-out problem.

For the web of spar 'a' the matrix b •• ,., of shear flows due to the loads R
in the basic system is
2 3 4 5 6
r-
1
''
''
0 2 '' Cut-out web (w3)
''
1

0 0 ''

bowa=-~·1 I
0 0 0 4
(b38)
P,
(lb.)
',,',,,

''
0 0 0 0 5 - 3ooo . ''
I
I '
\
\
no cut-ou'
I 0 0 0 0 0 6
\
\
\
L \
\ with cut-out
\

Corresponding matrices for spar 'b' are identical with those of Eqs. I

(b37) and (b38) except that they are multiplied by the scalar factors 20/10
and -1 flO instead of- 20/6 and --I /6 respectively.
Using FIG. 68 which shows the part of the structure affected by each self- 0
equilibrating stress system in combination with the information of FIGs. 63
and 64 on the Y-type system the b 1 sub-matrices can now be formed. We
find for the top flange of spar 'a' 10oo

12'3456
-1
0
0
0
0
0
0
0
0
}1
q_,
0 0 0 0 1 q_, Upper flange loads

(2
(lb/.n)
I (lb./In)

0 0 'i4o
0
I. '12o
-so .

0 0 0 I - 4Q

0 0 0 J3 '1o 0 .
-so
••• 0 •••••• (b39) -so
0 0 0 I
r4
- 4Q

0 0 0 - 2o

0
0
0
0 0
0
15
j
2o
4o
so
0 0 0 I
0 0 0 0 J6
_j

Numbers over the columns refer here to the Y systems.

7 8 9 10 11 12
For the web of spar 'a';
I 2 3 4 5 6 I
2 ·00 I ·001 0 0 I
-1 0 0 0 0 1 I
I ·00 2 ·001 0 0 I
0 -1 0 0 0 2 ----~----~----]

0 0 13·331·66l 0 01
0 0·. l -I 0 0 3
(b40) 0 0 lt·66 3·331 0 0 I
---· -1---·---1----~----1
•••••• 0 •••••

0 0 0 -I 0 4
I 0 0 15·00 2·50 1 0 0 1
0 0 0 0 -I 5 I I I 1
1 0 0 12·50 5·001 0 0I
I o o o o 0 6 2
f1b=3x ws"
,----~----1---- ----1
L I o o 15·00 2·501 o o 1
I I I
The corresponding matrices for web 'b' and the top cover differ from I 0 0 12·50 5·001 0 0 1
Eq. (b40) only in the scalar factor which, instead of 10/6 x 320, is 1----~----~----~----
1 o o 15·00 2·50 1 o o
I 0 0 12·50 5·001 0 0
and } ............ (b41) 1----~----~----
1 0 0 15·00 2·50
I
The sub-matrix b 1r for the shear flows in the ribs has been given already 0 0 12·50 5·00
in Eq. (b25) and need not be repeated here. L J
. ............. (b42)
The flexibility matrix f of the unassembled elements of the structure is
straightforward, using Eqs. (256), (257), (258) and (b33). As an illustration, where the numbers in italics over the columns refer to the flange elements.
we obtain for the upper flange of spar ,b', using the data of TABLE I: Note that the only non-zero elements appear in the diagonal submatrices.
66
 TABLE III of Example 9b
Flange loads and web shear flows, due to R 3 =1000 lb., in tube with or without cut-out
I Rib station I 1 I 2 I 3 I 4 5 6 7

I
Upper flange load Tube without cut-out -2726 -1658 -971 323 -30 -17
in spar 'a'
(lb.)
I Effect of cut-out
Tube with cut-out
1 78
-2648
1 599
-1059
I 1762
791
I -1532
-120)
-396
-426
-15
-32
0
0
0

I
Upper flange load Tube without cut-out -4364 -3005 -1418 -194
I
18 10 0
in spar 'b' I Effect of cut-out 1 -47 1 -359 1 -1057 919 238 9 0
(lb.) Tube with cut-out -4411 -3364 -2475 725 256 19 0
I Bay 2 I 3 I 4 I 5 6

I
Shear flow
in web 'a'
(lb./in.)
I
Tube without cut-out
Effect of cut-out
Tube with cut-out
-95·9
-16·3
-112·1
I
1
-84·0
-36·4
--120·4
1
-102·9
102·9
0
I
11·0
-35·5
-24·5 1
-1?:;
-12·3
-0·5
-0·5
-1·0
Tube without cut-out -25·5 -23·0 I
I
Shearflow -29·8
I
-4·0 0·2
in web 'b' I Effect of cut-out 5·9 I 13 ·1 -37·0
I 12·8
I ~:~ 0·2
(lb. fin.) Tube with cut-out -19·6 -16·7 -60·0 8·8 4·4 0·4
I
-1·704 -3·413 -5·123 -6·832 -8·541 -10·251 0·389 0·789 1·189 1·589 1·989 2·389
-0·006 -3·419 -6·838 -10·294 -13·676 -17·094 0·012 0·978 2·044 3 ·111 4·178 5 ·244
0 -0·006 -3·514 -7·075 -10·636 -14·198 0 0·100 1·586 3·253 4·919 6·586
0
0
0
0
-0·053 -4·233 -8·307 -12·381
0 0·104 -4·815 -9·815 0
0 0 0·181 2·000
0 0 0·181
4·000
2·118
6·000
4·118
I
1 o o o o -0·081 -6·173 o o o 0 0·063 2·000
L _j

I
6·6384 1·3916 0·0347 -0·0558 -0·0132 -0·0003
1·3916 2·9840 0·7156 0·0249 -0·0244 -0·0053
0·0347 0·7156 2·5828 0·5494 0·0088 -0·0184
-0·0558 0·0249 0·5494 2·3167 0·4661 -0·0007
-0·0132 -0·0244 0·0088 0·4661 2·0885 0·3578
-0·0003 -0·0053 -0·0184 -0·0007 0·3578 1· 8112
L _j

I 2 3 4 5 6 7 8 9 IO 11 I2
I I
11·320 27·415 43·613 59·689 75·572 91·509 -2·595 -6·598 -10·780 -14·873 -18·934 -22·993
2·389 14·956 30·050 45·396 60·395 75·470 -0·575 -4·087 -8·894 -13·867 -18·802 -23·72l
0·063 2·581 14·176 28·202 42·159 56·034 -0·022 -0·986 -5·700 -11·782 -17·980 -24·142 R
-0·095 -0·102 1·938 13·520 27·196 40·097 0·021 -0·036 -1·275 -6·494 -12·950 -19·434 °
-0·023 -0·128 -0·179 1·477 13·604 28·050 0·005 0·033 -0·033 -1·242 -6·227 -12·012
-0·001 -0·019 -0·103 -0·227 1·593 14·328 0·000 0·007 0·041 0·013 -0·755 -4·942
L _j ..•. (b43a)
I 2 3 4 5 6 7 8 9 IO 11 I2
I I
-1·447 -2·098 -2·726 -3·385 -4·071 -4·749 -0·433 -1·100 -1·797 -2·479 -3·156 -3·832 I
0·398 -0·841 -1·658 -2·434 -3·268 -4·088 -0·096 -0·681 -1·482 -2·311 -3·134 -3·954 2
0·011 0·430 -0·971 -1·966 -2·974 -3·994 -0·004 -0·164 -0·950 -1·964 -2·997 -4·024 3
-0·016 -0·017 0·323 -1·080 -2·134 -3·182 0·004 -0·006 -0·213 -1·082 -2·158 -3·239 4
-0·004 -0·021 -0·030 0·246 -1·066 -1·992 0·001 0·006 -0·006 -0·207 -1·038 -2·002 5
-0·000 -0·003 -0·017 -0·038 0·266 -0·945 0·000 0·001 0·007 0·002 -0·126 -0·824 6
L _j 0. 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (b44a)
I 2 3 4 5 6 7 8 9 IO 11 I2

r
I
-120·15 -101·78 -95·87 -92·22 -87·62 -83·13 -10·52 -13·08 -9·82 -5·24 -0·69 3·81 wi
12·11 -102·21 -83·99 -77·11 -71·69 -65·44 -2·88 -16·15 -16·63 -10·86 -4·28 2·18 w2
b,a=lo- 3 x 0·83 13·97 -102·93 -90·20 -88·74 -87·88 -0·22 -4·95 -23·05 -27·55 -26·20 -2~·52 w3
-0·38 0·14 11·03 -103·94 -95·87 -89·35 0·08 -0·36 -6·47 -27·35 -35·02 -38·66 w4
-0·12 -0·57 -0·40 8·87 -104·11 -99·70 0·03 0·14 -0·38 -6·54 -28·49 -36·82 w5
-0·00 -0·10 -0·53 -1·18 8·30 -95·20 0·00 0·04 0·21 0·07 -3·93 -25·74 w6
L _j 00 00 00 00 (b44b)
Note
The numbers above columns refer to the loads R1 to R12 •
In b 1a the numbers opposite rows refer to values at the rib stations. In the b., and b11 matrices two values were used at each rib station to denote
the (equal) load in the adjoining flange element sections. Here in baz these two values are merged.
In b,a the numbers opposite rows refer to values in the web fields.

The additional factor 2 is introduced to take care of the lower flange since
b. and b1 contain only the loads for the top flange. This applies also for
the flanges of spar 'a' and for the covers.
Carrying out the multiplications and assembly indicated in Eq. (b35)
we obtain the complete matrices b1'fb. and b1'fb1 and hence, on inverting
b1'fb1o the column of redundancies
Y = -(b/fb1) - 1 b1'fb.R ........................ (b43)
These results are printed in full above except for b1'fb1 which is
identical with the matrix D of the S;k coefficients in Eq. (b28) and need not
be repeated here. The calculation of the reciprocal matrix (b 1'fb1) - 1 was
performed using the Jordan technique described on p. 28 and required
approximately 8 hours for completion-by an unpractised and compara-
tively unskilled computor.
The complete set of flange loads and shear flows due to the forces R is
obtained now with the b. and b1 matrices already given and the solution
(b43) as
5=bR=[b.-b1(b 1'fb1) - 1 b1'fb.]R .................. (b44)
Limitations of space prevent the presentation of the complete matrix,
but as typical results we show above the flange loads in the upper
flange of spar 'a' :
bza=(b 0 za -bua(bt'fb1)- 1bt'fb.J ................ (b44a)
and the shear flows in the web of spar a :
b,a=(b,owa-btwibt'fb1)-lbl'fb 0 ] • • • • . • • • • • • • (b44b)
Calculation of the flange loads and shear flows for the particular set of
67
forces R 1 used in the previous analysis gives results in agreement with the
values calculated there. Such differences as arise are due to the neglect
of the terms 8;.1 in the O;k analysis. (See footnote on p. 63).
As a further illustration, the 11ange loads and shear flows given in TABLE
III due to a single load R3 =I ,000 lb. are easily obtained from the third
column of the final b matrix of Eq. (b44.)
These values are also depicted graphically in FIG. 69.
E.ff'ect of a Cut-Out
To show now the application of method of treating cut-outs developed
on p. 41 we determine the effect of removing the web of spar 'a' in the third
bay (i.e. element w3). This, of course, converts this bay into a four-boom
open tube which, if taken into account initially would have required the
introduction of the special extended Y system over the cut-out discussed
on p. 62 and destroyed the simplicity and regular pattern of both b 0 and
b1 matrices. By our artifice of imposing initial strains in the cut-out element
until the stress there is zero we eliminate all such special considerations.
Moreover, when the cut-out is introduced subsequent to the initial stress
analysis, our method does not require any additional special stress systems
but works solely with the existing information. Since the web carried only
shear stress, the matrix of initial strains has here only one row and the
result is obtained with remarkable speed and simplicity, the additional
matrix to be inverted being of order 1 x 1.
The shear flow in the element to be eliminated, i.e. b 11 , is given by the
third row bwa (Eq. (b44b)), (printed in full on p. 67) and the matrix b 111
of the shear flow in this element due to unit redundancies is similarly the
third row of b 1 wa (Eq. (b40)), i.e.,
1 we have already available the products
b 11,= 192 [0 0 1 -1 0 0] .................. (b45)
Now the equation for the set of Y H redundancies due to initial (shear)
strain H in this element is (Eq. (236) or (262))
DY 8 +b 111 'H=O
Y n= -D- 1 b111 'H .................... (b46)
and therefore the additional stress (shear flow) in the element due to H is
b11,Y 8 =-b 111 D- 1 bu'H ................ (b47)
Adding this to the stress in the element already calculated, b,R,and equat-
ing the resultant to zero, we find, using the above b1 11 and the D- 1 already
calculated, the required initial strain
H =(b 11,D- 1 bu')- 1 b 11 R
192 2
3 · 801 X 10 8 b "R ............ · ........ · (b48)
Hence from Eq. (b46) the additional Y 8 system required to nullify the
The numerical result for the Y 8 matrix is printed in full below. It
should be emphasized that this matrix gives the complete solution for a
general set of forces R. It only remains to premultiply with bh to obtain
the flange and shear flows. Thus, taking only a single force R3 =1,000 lb.
we obtain from the third column the additional flange loads and shear
flows given in TABLE III due to the removal of the web w3. Adding these
to the values already given for the structure without the cut-out, we obtain
the total values for the structure with cut-out. FIG. 69 illustrates the results
graphically.
This particular example of the application of the H matrix has been
worked out rather more fully than is strictly necessary in practice in order
to illustrate the way it works. In fact, the final result for the stresses in the
structure with cut-out is given directly by Eq. (268) on p. 41. However, the
amount of work involved is astonishingly small, consisting of a few simple
matrix multiplications and a single inversion. In more general cases where
several elements are removed at the same time the additional matrix to be
inverted is of order m x m where m is the number of stresses (loads or stress
flows) which have to be nullified.
The Flexibility Matrix
To conclude the present example we calculate now the complete flexi-
bility matrix of the structure for the points and directions of the forces
R 1 to R 12 •
In Eq. (312a) for the flexibility matrix, F
F=bo'fb.- bo'fb 1(b 1'fb 1) - 1 b 1'fb. . ................... (b50)
fb., b.'fb 1 and (b 1'fb 1) - 1 b1 'fb.
and therefore we only have to perform two further matrix multiplications
and one addition. The complete matrix obtained is given in full below,
Eq. (b50a).
The numerical values given there (as in all other numerical matrices in
this example) have been rounded off to fewer significant figures than were
used in the actual computation.
Although of course the F matrix is symmetrical, this characteristic was
not used to shorten the computations and all values were calculated in-
dependently, the symmetry being then used to give a check on the numeri-
cal work. The same applies, naturally, to the computations of any other
symmetrical matrix, like D. The whole question of numerical accuracy and
avoidance of computational errors is obviously one of extreme importance
in the practical application of these methods, and it is essential to have com-
prehensive methods of testing or checking calculated results. We hope to

l
stress in the cut-out element is discuss these aspects more fully in Part III of the series with special refer-
ence to the use of automatic digital computers.
I 0·0238 For comparison with the present analysis, the flexibility matrix was also
computed on the assumption that direct stresses in the flanges are given
0·1817 by E.T.B. i.e.-the statically equivalent system as used in the Otk long hand
analysis. Thus
0·5350 F.=b.'fb. . ......................... (b51)
b"R ................ (b49)
-0·4650 where b. is the corresponding statically equivalent matrix. Particular
attention is drawn to the fact that Eq. (b51) includes all shear strains asso-
-0·1203 ciated with b,. The usual artificial separation of deflexions into those due
to bending shear and twist does not even arise in these calculations.
-0·0047 In the following table we compare some of the/; 11 calculated by Eqs.
L _j (b50) and (b51).

Increments of redundancies Y 8 due to cut-out,

1 2 3 4 5 6 7 8 9 10 1/ 12
I I
-0·004 -0·064 0·471 0·412 0·406 0·402 0·001 0·023 0·105 0·125 0·120 0·112
-0·029 -0·488 3·591 3·147 3·096 3·066 0·008 0·173 0·804 1·961 0·914 0·856
-0·085 -1·435 10·573 9·266 9·115 9·027 0·023 0·509 2·367 2·830 2·691 2·519
0·074 1·247 -9·189 -8·053 -7·922 -7·846 -0·020 -0·442 -2·058 -2·459 -2·339 -2·189 .R
0·019 0·323 -2·378 -2·084 -2·050 -2·030 -0·005 -0·114 -0·532 -0·636 -0·605 -0·567
0·001 0·013 -0·092 -0·081 -0·079 -0·079 0·000 -0·004 -0·021 -0·025 -0·023 -0·022
L _j
Exact flexibility of structure in in./lb., .................. (b49a)
1 2 3 4 5 6 7 8 9 10 11 12
I I
21·54 27·62 35·67 41·42 48·35 55·32 4·43 11·27 18·42 25·43 32·38 39·32
27·62 73·68 103·4 132·9 164·0 19~·8 10·82 36·50 67·23 93·25 129·0 159·7
35·67 103·4 192·-l 265·9 342·3 419·1 17·3() 65·22 136·1 212·8 289·4 365·9
41·42 132·9 265·9 423·3 576·0 722·2 23·71 94·30 210·9 355·1 504·3 653·1
48·36 164·0 342·3 576·0 843·6 1090·3 30·05 122·9 285·1 501·9 747·7 997·3
F=lo- 8 x 55·32 194·8 419·1 722·2 1090·3 1510·6 36·42 151·6 359·3 649·3 995·9 1373·0
4·43 10·82 17·30 23·71 30·05 36·42 10·81 17·21 23·53 29·88 36·25 42·61
11·27 36·50 65·22 94·30 122·9 151·6 17·21 49·90 78·73 107·4 136·1 164·9
18·42 67·23 136·1 210·9 285·1 359·3 23·53 78·73 157·7 232·1 306·2 380·4
25·43 98·25 212·8 355·1 501·9 649·3 29·88 107·4 232·1 383·8 530·6 677·5
32·38 129·0 289·4 50~·3 747·7 9)6·9 36·25 136·1 306·2 530·6 788·1 1036·9
39·32 159·7 365·9 653·1 997·3 1373·0 42·61 164·9 380·4 677·5 103.6·9 1424·7 1
L _j
.................. (b50a)
68
 TABLE IV of Example 9b
Comparison of •exact' and approximate ftexibilities /;h (in.flb.) e·c
120
!t,l !t.4 /2,2 /2.4 /a,a !t2,12
Exact x 10 6 21·54 41·42 73·68 132·9 192·4 1424·7 so
Approx. x 10 6 10·50 29·32 49·61 106·1 154·3 1415·1 40
Error -51% -27% -33% -20% -20% -0·7°1-,o
0
P• (lb X 100)
As would be expected, the errors are much greater for points near to the q•• (lb/in)
root. Towards the tip they become negligible but it should be remembered
that the structure analysed has quite a high aspect ratio. The example so b
shows all too clearly that for modern low aspect ratio wings it will be
necessary not only to take account of the shear strains associated with
E.T.B. but to base the calculations of flexibility on the exact stress dis-
tribution. Moreover, as shown by our simple example the computations Temperature distribution
leading to F appear in the matrix method as a natural final step of the stress on upper flange of spar 'a'
analysis. P. (lbxlOO)
Q., (lb/in)
Thermal Stresses
zo
We consider now the stresses which arise in the four-flange tube due to
non-uniform temperature. For this purpose we assume that the upper flange
0
of spar 'a' has the temperature distribution shown in FIG. 70. The varia-
tion is taken to be linear between nodal points and hence is defined '2o
by the values at the nodal points. Since also the flange loads of the
Y systems vary linearly between these points the argument on p. 43 leading
-40
to the second of Eqs. (287) may be used here too. Thus in place of the
matrix e (Eq. 281) of the direct strains at the nodal points due to the loads
R in the basic system we introduce here the thermal strains (a8) at the ·so
same points. As transverse flanges do not enter into the present considera-
tions the matrix I simplifies considerably and observing that' I is the same ·eo
in each bay, becomes for the affected flange
I

I I I

I 2 0 0 0 0 0 I

'
I
I
upper flange load ............
4 0 0 0 0
I web shear flow
0 4 0 0 0
Fig. 70.-Four-flange tube. Flange loads and shear flows due to temp-
I erature rise of upper flange 'a'
1=6' 0 0 4 0 0 0 0 0 0 • 0 0 0 0 0 (b58)
0 0 0 4 0 and taking a=23 x 10- 6 per degree centigrade we find
I I I
0 0 0 0 4 I 35
0 0 0 0 0 2 I 66
L _j
57
Hence, the matrix D.e is from Eq. (287).
I I
a81 37
a82 42 00 00 •• 00 00 00 •• 00 (b59a)
a8a
L _j

D,e=(b 11 .)'l.l I
ae~ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (b59)
Using the inverted matrix D- 1 of Eq. (b43) we obtain the column Y a
of redundancies due to the given temperature distribution as (in lb. in.).
Ya={ -41340 -36860 -28740 -18210 -13980 -11220} .. (b6l)
a8 5
from which premultiplying by the appropriate b 1 matrix, we derive the
I flange loads and shear flows. These are given graphically in FIG. 70.
aG&
IL a07
_j
4. Examples of More Complex Structures
Having investigated at some length the relatively simple example of a
01> 0 2, •••••• 0 7 are the temperatures at the nodal points (see FIG. 70). four-tlange tube both by the sik and matrix methods we turn our attention
In b 1az the two equal values in the elements on either side of the nodal to some more complex and representative aircraft structures where the
points are merged into one so that btaz gives here single flange loads at the matrix method alone is indicated. In what follows we discuss mainly the
nodal points. selection of the appropriate self-equilibrating stress systems for the
Thus setting-up of the b1 matrix; (see also the more general analysis on pp. 38
and 39).
I I We start with a two-spar six-flange tube, the two additional flanges being
0 0 0 0 0 0 I introduced to represent more accurately the direct stress carrying ability
of the covers. FIG. 71 illustrates a structure of this type fully supported
0 0 0 0 0 0 at the root with six bays and no cut-outs. The total number of redun-
dancies is easily seen to be n = 6 x 3 = 18 which may be confirmed also by
0 0 0 0 0 0 Eq. (247a) on p. 38 ; thus,
(buaY=~· n=a(,B+N-4)=6(6+1-4)=18 (b62)
0 0 0 0 0 0 00 00.00 00 00.00

The calculation of the b. matrix is straightforward and the very simple

0 0 0 0 0 0 choice of two independent spars made in (3) is here too the most ad-
vantageous.
0 0 0 0 0 0 00 00 00 00 00 (b60) For a cross-section with no axis of symmetry we may form the b1 matrix
L _j for the following redundant systems:

69
 Su:·flange s1ngte cell tubw
Singly symm•tric~l s iJC·flange
stngle cell lubf'

Two-bay segment

r -~y5tem apph•d to corner flanges

bet F;gs 63 and 64.or 43)

i-sysltm
(equal •net opp05 iU• X~systems
on covvrs (2.)) •nd (4.1 l)

X- §yt.lem for cover {4,1)

Fig. n.--Singly symmetrical six-flange single-cell tube. Assemby of
(sn Fig 41) X and Y systems
Fig. 71.--Slx-ftange, single-cell tube. Assembly of X and Y systems

6 Y systems applied to the corner flanges,

6 X systems in the top cover and 6 X systems in the
bottom cover
s
I
(b63)

as indicated on FIG. 71; see also FIGS. 41, 43 (or (63) and (64)) for a full
two-cell tube
description including shear flows of these systems. S••-flang~

If the cross-section of the tube in FIG. 71 has an axis of symmetry the

number of redundancies for lift loads and torque is
n1 =6 X2 = 12
As redundant systems we choose in this case
6 Y systems applied to the corner flanges
} (b64)
6 X systems
An X system consists of an antisymmetrical combination of two equal and
opposite X systems in the top and bottom covers as illustrated on FIG. 72.
These coupled X systems were not mentioned on p. 38.
Next we discuss the six-flange tube with intermediate web shown in
· FIG. 73. There are again six bays and for the moment we assume that there
are no cut-outs. When the root is fully supported the degree of redundancy
is from Eq. (247a)
n=a(,B+N-4) = 6(6+2-4) = 24 ........... . .... (b65)
The b. matrix may be obtained simply from the basic system consisting X-system
(see Fig. 42)
of three independent spars. As redundant systems we select,
6 x 2 = 12 Y systems applied to the adjoining four flange tubes }
6 Z systems (b66) Two - bay segment

6 X systems in the top (or bottom) cover

The structural elements affected by these systems are shown schematically
in FIG. 73. Fxos. 42, 43 (or (63) and (64)) and (44) describe fully the X, Y, Z Y- systems.
systems. Naturally, half of the Y systems may alternatively be based on the ( ue Figs. 63 and 64. or 43)
tube formed by the four corner flanges.
Assume now that in one bay the intermediate web is missing and that no
stiff-jointed closed frame is provided to restore partially the lost shear
stiffness of the sheet. The number of redundancies is obviously reduced by
one. The b. matrix may be determined for the basic system consisting of Z-syst~ms

the spars and the two ribs shown in FIG. 74. Of the redundancies given in (see Fog. " )
system (b66) we lose Fig. 73.-Six.ftange, two-cell tube. Assembly of X, Y and Z systems
70
 iJ.

Basic system
Missing rib

X-system
(see Fig. 42)

Missing web

Z-system over two bays

(see Fig. 44)
Extended Y-system
(see Fig. 66)

Y-systems over three bays

(see Fig. 65 )

Fig. 76.-six-flange, two-cell tube. Modified Y and Z systems when rib of

one cell Is removed

systems. On the other hand it may be best to stick to the choice (b68) for
the bay with the missing web and to work also in the other bays with Y
systems applied at the four comer flanges.
If the cross-section has an axis of symmetry and we consider only torque
loads we lose through the missing web in one bay five of the redundant
systems (b66), i.e.
Y-system applied
to comer flanges
Extended X-systems one Z system in the affected bay
(see Fig. 75) } (b67a)
Fig. 74.-slx..ftange two-cell tube. Modified X and Y systems when Internal four Y systems at the adjoining rib stations
web removed The required four new systems may be taken as,
one stretched Y system
i-2
two standard Y systems applied to the four comer flanges
at the adjoining ribs (b68a)
one stretched X system (equal and opposite stretched X
systems at the top and bottom covers)
Again it is better to adopt the Y systems applied at the four corner flanges
at all rib stations.
Another internal cut-out may occur through a missing rib, for example
in the right hand side cell (see FIG. 76). Here we lose in the affected two
bays five of the redundant systems (b66), i.e.
the two adjoining Z systems I
+ three Y systems applied at the affected and adjoining two
ribs in the same cell J
~
(b69)

Fig. 75.-Extended X system applied at top cover
the one Z and one X system in the affected bay
} (b67)
and the four Y systems acting at the adjoining rib stations
We have a net loss of six standard redundant systems and need hence five
new systems. As such we choose (see FIGS. 74 and 75),
one stretched Y system (as shown in FIG. (66))
two standard Y systems applied to the four comer flanges at
the adjoining ribs (b68)
two stretched X systems at the top and bottom covers (as
shown in FIG. 75)
Alternatively we could select two stretched Y systems (two adjoining four
flange tubes) but since these are linearly related to the two above Y systems
applied at the comer flanges it is then necessary to drop one of the latter Y
The required four new systems may be chosen as (see FIG. 76)
two Y systems extending over three bays and applied at the}
adjoining ribs in the right hand side cell (see FIG. 66)
(b70)
one Z system extending over the affected two bays
one X system j
Note that the latter X system is in addition to and in the opposite cover
to the one chosen in system (b66). If the cross-section has an axis of
symmetry the X system of (b70) is replaced by an X system.
Finally, we consider the case of a cut-out in the cover (FIG. 77). We lose
here in the affected bay three of the systems
two Y systems at adjoining ribs }
(b7l)
one Zsystem
the X system of (b66) may be assumed applied to the opposite cover and is
hence not lost. Our choice of two new redundancies is
one stretched Y system in the affected cell
} (b72)
one stretched X system at the affected cover
71
 ('

Cut-out tn cover

Cut-out in cover
Extended Y-system
(see Fig . 66)

' --
Extended X-system
{see F1g 75)

Fig. 77.-Six-flange, two-cell tube. Modified Y and Z systems when one

cover is removed

I
;;.
The above discussion of the effect of missing shear fields and the choice
of appropriate redundancies may be applied with little alteration to the
more general multi-cell structures considered in 8c. However, we require
in addition for tubes wi•h three or more cells the stretched Z systems illus-
trated in FIG. 78 when an intermediate web or cover is missing.
The above discussion shows that every type of cut-out requires special Extl'nded !-system
consideration and inevitably disturbs the pattern of the b" and b 1 matrices. Fig. 78.-Extended Z system in multi-cell tube when one cover is removed
We emphasize also that particular care is necessary in the selection of the
necessary number of redundancies at cut-out regions to avoid the danger
of linear dependency. It is evident from the previous example of a cut-out
that all these complications are avoided by the use of the new technique
on p. 41. With the help of this powerful method we may carry out the stress
analysis in the continuous structure (always a simpler and smoother pro-
cess) and derive hence (by the indirect use of initial strains in the elements
to be removed) the stress distribution in the actual structure with cut-outs.
Naturally we introduce in this approach additional redundancies but this
should be, in general, of little or no importance when a digital computer
is used.

In concluding Part I we point out again that the matrix force method
developed on p. 37 et seq. of Section 8C and illustrated here on some
very simple problems is directly applicable to delta wings and in fact to
any wing configuration as long as the grid of spars and ribs is orthog-
onal or nearly so. Emphasis in the examples has been on the force
method of analysis since it is, in general, more suited to aircraft applica-
tions. Although the a., a 1 matrices of the displacement method (Section
80) may be slightly simpler to form than the b0 , b1 matrices of the force
method 1 it is on the other hand easier to write down the f than the k
matrix of the unassembled elements. Possibly the greatest advantages of
the force method in wing and fuselage problems lie in the better con-
ditioning of the Eqs., the much greater accuracy in the derived stresses
and the smaller number of unknowns.

72
 Part II. Applications to Thermal Stress Problems and St. Venant Torsion

By J. H. Argyris and S. Kelsey

1. INTRODUCTION

W
E follow up and amplify the general theorems and principles
developed in Part I by presenting here a number of applications.
It is hoped that the problems considered may prove of more than
academic interest. Three of them are concerned with stresses due to
thermal straining-a current fashion in the aircraft structural field-and
were originally presented in an A.R.C. paper.*
The first deals with the stresses arising in a circular ring due to a peri-
The method of calculation follows closely Section 8 of Part I.
From the symmetry of structure and loading, the shear force in the
ring on the axis of symmetry is zero. Hence as the basic or statically
determinate system we select the structure shown in FIG. Ia. As redun-
dancies we choose the bending moment X 1 and normal force X 2 at the
elastic centre,* which in this case is the centre of the ring.
Forces and moments in basic system due to X 1 = l and X 2 = l re-
spectively :

}
pherally varying temperature rise on the inside of the ring, which is System X 1 =I;
linearly elastic. Solution in terms of the bending moment and normal (2)
force in the ring is obtained easily by a straightforward application of the
D;k-method described in Section 8 of Part I.
The second example is concerned with the same problem but the System X 2 = I; I
material of the ring is taken to have a non-linear elastic stress-strain
relation. The analysis is based on the unit load formulation of the virtual
forces principle (Section 6 of Part I). It is of some interest that this not
M 2 =rcos 1/J, N 2 = -cos 1/J
Hence total moments and normal forces:
J (3)

too trivial problem is amenable to a complete solution without excessive

}
algebraic and numerical work. M=M 1 X 1 +M2 X 2
The use of the general principle of virtual forces for thermal stress (4)
problems is illustrated in the third example. An approximate solution is
obtained for the stresses in a rectangular plate due to a temperature Straining of basic system due to temperature 0 (see FIG. lc). Relative
distribution symmetrical about the longitudinal axis of the plate and rotation c/J tds of two cross-sections at distance ds apart due to temperature
uniform along the length. gradient 0 I h across the section
In the fourth example, the twin principles of virtual displacements and
virtual forces are both used to obtain approximate solutions to the
St. Venant torsion problem for thin section bars. General differential • See J. H. Argyris and P. C. Dunne, Structural Analysis (Part II of Handbook of Aeronautics,
Vol. I) Pitman, 1!152, Table 17.1.
equations for the approximate warping and stress functions are derived
and applied to rectangular and double-wedge cross-sections. Use of the
two methods gives upper and lower bounds to the torsion constant and
an estimate of the maximum shear stress. The closeness of these bounds
and comparison with some exact solutions indicate that the results achieve
a high degree of accuracy.

2. FIRST EXAMPLE h
Thermal Stresses in a Circular Ring
A uniform circular ring of symmetrical cross-section and mean radius r
is subjected on the inside circumference to a temperature distribution
0o 71' 71' l
0=y(l+cosmi/J)for -m<I/J<iii ~ ............ (I)
0 =0 over the remainder of the circumference J (a)
the temperature varying linearly across the depth h of the section and
falling to zero on the outside circumference (FIG. Ib).
The bending moment and normal force distribution in the ring are to
be found with the assumption that:
(a) the material follows Hooke's law;
(b) the bending deformations of the ring may be calculated by the
engineers' theory of bending for straight beams; and
(c) the deformations arising from the normal force are to be considered Fig. I.-First Example: thermal
but the shearing deformations are to be neglected. stresses in circular ring
(a) basic system
Additional Notation (b) temperaturedistributlon
A : area of cross-section of ring (c) thermal straining of ring
I: moment of inertia of cross-section of ring
M: bending moment ( +ve if producing tension outside)
N: normal force ( +ve tension)

• See Ref. (13) of Part I.

(c)
73
 a0
~eds= --,ds .......................................... (5)

The negative sign arises since positive bending strain ~ corresponds to

increased length on the outside of the ring. Increase of length 7Jds at
centroid (at mid-height of section)
a0
7JdS='J: (6)

The two redundancies X 1 and X 2 are calculated from the compatibility

equations (182) given in Part I which take here the form
SuX1+3 10 =0 I
j ······················ (7)

and hence
_ slO _ 32o
X1- -3u' X2- -322 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
From the definition of the 3-coefficients or equations (203) and (204) of
Part I
J
c
J
s,,= ~2ds+ ~~ds ................................... . (9)
c
M,. =
a§.EI
h

S,.,= JM,~etfs+ JN,7Jds .................................. (10) Fig. 2.-Bending moment and normal force distribution in circular ring
c due to thermal strain, m= I
for p = 1 and 2. The sign f( • .• . )ds denotes integration around the circum-
c
ference of the ring. Using (2), (3), (5) and (6) in (9) and (10) we find:

J
dlfl 2TTr I
Su =r El= El I

J
(II)

c c
and
+n/m
a0.rJ TTra0 0
3 10 =- 2h (I +cos mlfl )dlfl=- ---,;zjl
-n/m
1
a0.r 2
3 20 =- 2h 1 +z,.
h c )J +n/m
(I +cosmlfl) cos lfldlfl ~ (12) e= e.(1+cos31jl)/2
-n/m Mro = a§.EI

= --h-
a0.r 2 ( ·
1
h) m2
+lr m2-1 .
SID m
7T
I
h
j
Thus
Fig. ].-Bending moment and normal force distribution in circular ring
310 El
x1 = -sll =a0o2mk .................................... (13) due to thermal strain, m=3

and a0El a0 0 El
Mt= -~~El=-h- =211 (l +cos mlfl)

X2= -~::=a0o!~ m';~ 1 sin f[z } !~ .................... (14)

and at !fl=O
a0 0 El
where Mto=-h- ············································ (19)
h I Hence
f3=2r and K= r2A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (15) MM ~ ~
M 1 = z-0 + cos m'f') for -iii<. 'f' <.
7T
m ... · ·· · ·· · ·· · ·· · ·· · ·
7T
(20)
Substituting (13) and (14) into (4) we obtain the final bending moment M
and normal force N in the actual structure as Comparison of (20) with (l6a) shows that in the case m = 1 the thermal
straining is completely prevented if we ignore the deformations due to
M- a0oEI[ 1 2 ~ sin (TT/m) 1 +{3 normal force.
- 2mh + m2-1 TT/m I+K- cos 'f'·'·] ............ (16) Using Eq. (19) in (16) and (17) we find finally for the bending moment
a0oEI +{3 and normal force:
m 11+K
m2 . 7T
N=- TTrh m2-1 SID cos lfl (17)
M _ _I_
Mto--2m
[t +2m~sin
-l
(TT/m) 1 +{3
TT/m
2 I +K
cos ·'·]
'f' ·············· (2 1)
which for m=1 reduce to
a0oEI[ 1 +fJ
M=-u~ 1 +1 +K cos lfl
J (l6a)
Nr ~ _ ~ sin (TT/m) 1 +{3
Mto ~ mL-} 7T
.t.
1 +K cos 'f' ...................... (22)

N- _ a0oEI 1 +fJ Numerical Applications

- 2rh I +K cos lfl · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (l7a) The distribution of bending moment and normal force around the
It is often possible to neglect the deformations due to normal force periphery for a ring with r/h=l2 (i.e. {3=1/24) and K/{12=1 is shown in
which appear only in 3 22 . To obtain the corresponding formulae we put FIG. 2 for the case m=1 and in FIG. 3 for m=3. Values of M/M 10 and
in Eqs. (16) to (l7a) -Nr/Mto are plotted radially with the centre line of the ring as base.
1+{3 3. SECOND EXAMPLE
l+K=J .... • ............ • .. • • • • ... • •. • ........ •........ (18)
Thermal Stresses in a Circular Ring with Non-Linear Elasticity
On the other hand no difficulty arises if the effect of shear deformations Consider again the ring of the first example subjected to the sam:!
on 3 22 has to be considered. • This term is, however, usually negligible. temperature distribution. The cross-section of the ring is the idealized
Next we calculate the bending moment set up in the ring if the thermal I-section shown in FIG. 4 and the material is assumed to obey a non-linear
straining is completely prevented. Thus : stress-strain relationship of the form

• See Aqyris and Dunne, loc. cit. E ·=i+~(a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23)

74
 1·0

--o-- - - . - - - - - - - - - --0--~:>---iJ--- 0·8

0·6

Fig. 4.-Second Example: idealized 1-section

where cp (a) is an odd function of a (same stress-strain diagram in com-

pression and tension). Moreover, cp =dcp/da =0 for a =0 which ensures
the usual linear relation for very small stresses (see FIG. 5).
The bending moment and normal force distributions in the ring are to 0·2
be found with the assumptions that :
(a) Only the flanges of the section carry direct stresses, which are taken
at any cross-section to be constant in each flange.
(b) The direct strains in the flanges derive both from bending moment
0·5 1·5 2·0
and normal force.
(c) Shear deformations of the web are negligible.

Fig. 5.-Non-linear stress-strain curve

Additional Notation
A 1, A,: area of inner and outer flange respectively where
A : total direct stress carrying area o 10 =Ja0P; 1ds, o 20 =Ja0P, 2ds . . . . . . . . . . . . . . . . . . . . . . . . . . . . (31)
c
P1, P,: loads in inner and outer flange respectively are the relative displacements of the point of application of X 1 and X 2
The structure is twice redundant as in the first example and as basic due to free thermal expansion of the inner flange.
system we take once more that of FIG. Ia. Also as redundancies we select Using Eqs. (25), (26) and (28) for the flange loads and strains in (30)
again the bending moment X 1 and the normal force X 2 at the centre. we find
To find the unknowns X 1 and X 2 it is convenient to work in terms of
flange loads and strain~.
o 11 X 1 +o 12 X 2 + f<Pncp; + P, 1cp,)ds +8 10 -- o
.. (32)
Flange loads due to X,,= I : o2lxl +o22x 2 + f<P, 2 c/J, + P, 2cp,)ds-t o o
211 -
.-
P,,,= N,,A M,,.· p ,., = N,,A,+M,,
---;r-1 -h A h ........................ ..
(24)

Note that M,. is taken about the centroid G of the cross-section. Using the (33)
expression for N,, and M,, of Eqs. (2) and (3) in (24) we obtain:
for system X 1 =I:
I I
o
are the usual displacement coefficients arising from the linear part, a/E,
Pq=-h; P, 1 =+r, .................................... (25)
of the strain, E, and are the same as in the first example. In fact, 812 =8 21 =0
due to the selection of the elastic centre as origin of X 1 and X 2 and
and for system X 2 = I : Eqs. (32) simplify to
o 11 X 1 +J(Pqcp,+P, 1cp,)dl·+o 111 - 0
P, 2 - (1' +!,) cos 1{1, P,2- - (1' -;;) cos 1{1 (26) .-
(34)
S22 X 2 + J<P, 2c/J, + P, 2c/J,Id1· +o20 --o
Hence the total flange loads P, and P, can be expressed as: ('

These reduce to Eqs. (7) of the first example when cp -- 0.

(27) S - coefficients:
P,- P, 1 X 1 +P, 2X 2=a,A,
Substituting these expressions for the stresses in Eq. (23) the elastic
strains• in the flanges are found to be ~ ............ (35)
I
l - I + A.
-' -") --.
_ PnX 1 +P, 2X 2 ..J..(PqX1 +P; 2X 2) 820 = -
r2(
aH°h m2 sin -7T
E;- EA; +'f' A; A r m2 - I m )
(28) which are essentially the same as in (12), but for a small difference in 8 20
E,-
__ PnX 1 +P, 2X 2 ..1..(P, 1 X 1 +P, 2X 2)
EA, +'f' A, r arising from the geometry of the cross-section chosen.
Furthermore
27Tr
The total strains are 7r1· 3
8 11 = £/' o22 = £7(1 +K) ................................ (36)
y 1 - E; -!-a(-1
y,- E, where
Noting that in the actual system no relative movements take place at
the cut ends and applying the unit load method as given by Eq. (71a) in
K-A~·2and/-A,~,It2 .................................. (37)
Part I we obtain the following two equations Eqs. (36) are identical with (II) in the first example.

}
f{E;Pq +E,P, 1 +a0Pq)ds=0 Non-linear terms of Eq. (34):
,. Using Eqs. (24) and (27) for P,, P, and Eqs. (2) and (3) for M,, N,, the
(29)
f(E;P 12 + E,P, 2+a0P; 2)ds=O arguments of the cp-function in (34) are conveniently written in the form
,-
or c/J,= c/J{AI;h[ -X 1 - X 2 r cos 1{1 (1 +~1 ~)]}
l
(38)
f(E 1Pil + E,Pn)ds+8 10 =0
('

(30) cp,=c/J{)h[ X1 +X2r cos 1{1 (I -1' ~)]}

+ E,P, 2 )ds +8 20 =0
,-
J(E 1P 12
J Substitution of (38) into (34) shows that for a given cp-function we obtain
after integration with respect to 1{1 two non-linear equations in the un-
known X 1 and X 2 . The solution of these equations determines the com-
• Dinrtional suffices in the strains arE' omitted siner direet bendinte strains only in thl'
auly!i!iis.
OCl'ur
plete stress distribution in the ring.
15
 e =0.,(1+ cos!JI)/2

0·1 P=O
p=1
P=2
2{) 3·0 a0.,EI
M,o • - h -

Fig. 6.-Second Example: thermal stresses in non-linearly elastic ring. Fig. 7.-Bending moment and normal force distributions in non-linearly
Values of redundancies x 1 and x2 elastic ring due to thermal strains

We select now the specific a-E relationship

E=i[, +C~r-lJ .................................. (39)
2x 2+p 2x 2[6x 1 2 +3x 2 2]-1 ~ 0 } (46)
or
aE (a)"
cp(a)=-3 -
a.
(40)
The linearly elastic case is represented by p =0. Then
x 1 =x 2 = 1/2
where a. is an arbitrary reference stress (a. may be defined as the stress at which values may also be obtained from Eqs. (13), (14) and (19).
which the total strain is twice the linear strain component) and n is an
odd integer. By varying a. and nit is possible to obtain a versus E diagrams
with a wide variety of shapes (see FIG. 5). Numerical Results
We assume also that Eqs. (46), which have only one pair of real roots, have been solved
for values of p between 0 and 3 and FIG. 6 shows values of x 1 and x 2
A 1 =A,=A/2 plotted against p. The roots of the simultaneous cubic equations were
then, K=I/Ar 2=(h/2r)2=f32 obtained by an iteration method, successive values of x 1 and x 2 being
The cp-functions of Eqs. (38) take now the form calculated, using the. previous approximations for the terms in the brackets.

lr ............
For p =0 · 5, successive values of x 1 and x 2 , calculated in this way are:
cp;=-p"[x 1 +x 2(1+f3> cos ,Pr~ X1 0·5 0·433 0·452 0·449 0·449
(41) x2 0·5 0·390 0·416 0·410 0·412
cp.= pn[x 1 +x2(1-{J) COS ,Plni j The initial approximation in this case was the linearly elastic solution;
for further values of p, extrapolation of previous results gives satisfactory
where starting points. Since successive approximations are alternately greater
Jr1 Jr2r and less than the final values of x 1 and x 2 it is quite easy to improve the
X!= Mto and X2= Mto .................................... (42) natural convergence of the process by a little judicious anticipation.
in which Mt •• the moment at ,P=O for fully restrained thermal expansion, Direct solution ofEqs. (46) was also obtained. Eliminating x 2 an equation
is given by (19), and cubic in p 2x 1 2 and x 1 is obtained, which can be solved for either para-
meter and the corresponding values of p. x 1 and x 2 calculated.
Ea0. h The bending moment and normal force distribution around the ring is
p= 2a. 'fJ=z, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (43)
plotted in FIG. 7 for p=O (linear elasticity), p=l·O, and p=2·0, in the
non-dimensional form M/Mto and Nr/M 10 • As would be expected, the

l
Substituting from Eqs. (25), (26) and (41) in the integrals of (34) we find
effect of the non-linear stress-strain curve (in which da/dE decreases with a)
!. P; 1 cp;ds=p"~ 1-f [x +x (1 +fJ) cos ,P ]nd,P
1 2
is generally to reduce the maximum bending moment produced by a given
temperature rise and also to smooth out the variation around the ring.

JP, cp,ds=p"Jz ~J [x +x (1-{J) cos ,P ]"d¢'

1
c
1 2 I 4. THIRD EXAMPLE
(44) Thermal Stresses in a Rectangular Plate
Consider a thin rectangular plate of length 2a and width b with a
c temperature distribution
0=0.g(y/b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (47)
symmetrical about the longitudinal axis of the plate and constant along
c c
the length (see FIG. 8). The stresses in the plate are to be found with the
assumptions that they are constant across the thickness (plane stress) and
which can be integrated without difficulty since n is an integer. that the material is isotropic and follows Hooke's law.
Taking n=3 Eqs. (44) reduce to the simultaneous cubic equations

l
1 Additional Notation
2x 1 +p 2x 1[2x 12 +3(1 +f3 2)x 22 ]-;n =0
2a, b: length and width respectively of plate
2(1 +f3 2)x 2+p 2x 2 [3(2+f3)x 1 2+3(1 +6f3 2 +{3 4)x 2 2] - (45) x, y: cartesian co-ordinates with respect to the
centre of the plate
-(l +fJ) 2m 2 sin (1rjm) =O x'=a-x
m 2 -1 1r g=x/a, 7J=y/b, 'o;=x'/b: non-dimensional co-ordinates
which form= l and {3=0 (2r»h) take the particularly simple form a0 =a0.g(7J): thermal strain
76
 We apply the principle of virtual forces in the form of Eq. (80b) of
~..F(y/b) ~,.
(Vyy}y.o Part I, i.e.
0
JJ[.6. 2F+Ea.6.0]SFdxdy=0 .. .. . . .. . . . . . .. . . . .. . . .. . .. . .. (58)
where the virtual force system SF is here due to the increment Stfo of the
function tfo
8F=Ea0b 218t/J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (59)
Substituting for F and 0 from Eq. (51) and (47) respectively, and
noting that
d 2g d 4/
dTJ2=-dTJ4 ···················· ···················· ···· (60)
b/2 we obtain from Eq. (58)

_l J J[ d~d
+I +1'2
d 4 t/J a2 d 2t/J d 21 a4 d'l]
:------'L--- - ' - - - - ' ' - - - ------"'---- ---'---'-------' +27)2 de dTJz+ b'<t/J-l)dTJ' I St/Jd~dTJ=O (61)

t------a--~ I -·1/2
which may be written, on integrating with respect to TJ

Fig. 8.-Third Example: thermal stresses in rectangular plate; transverse

distribution of temperature and stresses
J[
+I
d4tfo a2 d2t/J a'
A d~c2b2B de+IJ4C(t/J-I) Stfod~=O
J .. .. . .. .. .. . .. (62)
-I

g, /., 1: functions of TJ where the coefficients A, B, C are given by

A, B, C: constants
J+ I;2 ·rl·l/2do/ r+ I''Jy
A= f2dTJ, B= - I ([2dTJ, C= I dTJ 4dTJ .. .. .. .. .. .. .. .. (63)
tfo : function of ~ or ' -1/2 -1/2 TJ -1/2
{J 1, 2 =JL±iw: roots of auxiliary equation
Note that A, B and C are all positive.
Ji=J-Lbfa, iii =wbfa Eq. (62) can only be satisfied for an arbitrary Scp if
The thickness of the plate is taken as unity. d4t/J a2 d2tfo a 4
The problem can be solved exactly but we shall derive here an approxi- Ad~ 4 -21J2 B d~ 2 +l)4C(t/J-1)=0 .. .. . . .. . .. . .. . .. . . . . . .. (64)
mate solution using the principle of virtual forces in the form discussed in
Part I, Section 6E. which is the required differential equation in tfo.
Consider first a plate of infinite length for which the stress distribution The particular solution of (64) is
is well known.* Thus, for the symmetrical temperature distribution con- t/J= l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (65)
sidered here, the self-equilibrating stress system which is, of course, the solution for an infinite plate and the complementary
I
J I ......................
+1/2 function can be written in the form
axx=Ea0{J g(TJ)dTJ -g(TJ) t/J =K 1 cosh {1 1 ~ + K 2 cosh {1 2 ~ + K 3 sinh {1 1 ~ + K 4 sinh {1 2 ~ . • • • (66)
-1/2 (48)
where
a .•• =a.,.=O
flt·z-b
·-i![
8±(82-AC)vz]vz
A (67)
is set up. To obtain the stress distribution in a plate of finite length, we can
apply a stress-system -a.,., at the two ends x= ±a and superimpose the are the roots of the auxiliary equation
resulting (diffusion) stress-system on (48). We shall, however, adopt here
a slightly different approach in order to illustrate the application of our a2 a4
A{J4-21)2Bfl2+pC=0 . .. . .. . . .. . . . . .. . . .. . . . . .. . . . . . .. . (68)
principle of virtual forces in the presence of thermal strains.
Guided by the solution (48) for an infinite plate, we assume that it is A B C are functions solely of the temperature distribution and are in-
possible to express the longitudinal stress in the form d~pe~dent of a/b. Thus, the length/width ratio of t~e plate _ente~s only as
a.,.,=Ea0J.(TJ)tP(~) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (49) a common factor in the arguments of the hyperbolic functions m (66).
where The tfo functions must obviously be symmetrical with respect to the
+1/2
centre of the plate (~ =0) and therefore
K 3 =K4 =0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (69)
f.(TJ)= Jg(TJ)dTJ-K(TJ) .. .. .. . .. .. .. . .. . .. .. .. . .. .. . . .. .. .. (50)
-1/2
At the ends of the plate (~ = ± l) the stresses axr. a.,. are zero and
therefore
and tfo is a function only of x. We know that Eq. (49) cannot, in general,
represent the true distribution since the cross-wise distribution will also tfo= ~~=0 for~=± I .................................... (70)
vary with x. However, for a smooth function g(y) it can be expected that
(49) will yield a sufficiently accurate answer. The boundary conditions (70) determine the constants K 1, K 2 and the
The distribution (49) can be represented by the stress function t/J-function is found to be
F=Ea0,)J2J(TJ)t/J(~) .. .. .. . .. . .. .. .. .. . .. .. .. .. .. .. . .. . . .. (51) J. t _ _{1 1 sinh {1 1 cosh {1 2 ~ -{1 2 sinh fJz cosh fJ1~ ( 71 )
which gives the stresses 'l'(s)-l {J 1 sinh{J 1 cosh{J 2 -{J 2 sinh{J 2 cosh{J1 ··········
o2F do/
au= 0 yz=Ea0.dTJ 2t/J(~)=Ea0.J.t/J(~) . . . . . . . . . . . . . . . . . . . . . .
In general
(52)
AC>B2
o 2F bdl dt/J and the roots {1 1, 2 are therefore complex. They can be written in the form
a.,.=-oxoy=-Ea0•ad1Jd~ . . .. . .. . . . . . . . . . . . . . .. . . . . . . (53)
{J 1 , 2 =JL±iw . . . . . . . . . . . . . . . . . . . . . . . • . . . . . . . . . . . . . . . . . . . . (72)
o2F b2 d2t/J where
a •• = 0xz=Ea0.-(i21 de ................................ (54)
I
The a.,.,, a•• distributions are symmetrical and the a~• distribution anti-
symmetrical with respect to the x-axis. From Eq. (52) and the conditions
JL=bii=bnc v/~+~)r 12 ~ . .. . .. .. .. . .. . (73)
(a••)•=±b; 2=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (55) a_ a[l(~ /C
w=bw=b 2 VA-A
112 8)] j
(a.,.)._ 0 =0 (antisymmetrical distribution) . . . . . . . . . . . . . . . . . . (56) Substituting (72) into (71) we obtain
the function I is :
JL[cosh JLO +t) sin w(l-~)+cosh JL(l-~) sin w(l-r~)]
'1 'I
tP = I - JL sin 2w +w sinh 2JL
I(TJ)=J J!.(TJ)dTJdTJ .................... ................ (57) w[sinh JLO +0 cosw(l-~)+sinhJL(l-~) cosw(l +~)]
-1/2 0
JL sin 2w +w sinh 2JL
••••••••••••••••••• 0 0 •••••• 0 ••• 0 ••
(74)
• See Timoshenko and Goodier, Th~ory of Elasticity, 2nd Edition, McGraw-Hill, New York
1951, p. 399. • which, at the centre of the plate (~=0) reduces to

77
 1 _ 211- cosh IL sin w +w sinh IL cos w
.JJ ) _
(75) 1·2 ...---.-----. ----
'f'\ 0 - IL sin 2w+w sinh 211-
For the discussion of Eqs. (74) and (75) it is best to consider a particular
form of temperature distribution. Take, for example, the function
1 + cos 2mrT} . l() lt----+--
K(T))= 2 where n IS odd . . . . . . . . . . . . . . . . . . . . . . . . (76)
which is shown in FIG. 8 for n= i.
Then
" __ cos 2n7TTJ (77)
Jo- 2
y
I = 1 +~~~:1TTJ ....................................... . (78)
and the constants A, B, C are, from Eq. (63)
3 1 1
A=128n«1T«' B=32n21T2' C=g ........................... . (79)

also
w
iJ.=1Tn[t(v3+ 1)]11 2= 1· 351Tn, =1Tn[t(v'3-1)]11 2=0· 6981Tn .. (80)
A perusal of Eqs. (74) and (75) on the basis of the arguments (80)
shows that even for the lowest value (n=1) the function !fo, which is zero
at g= 1, increases rapidly with decreasing g towards the value 4> =I for
an infinite plate. In fact, not only does it reach it but, due to the influence
of the trigonometric terms, it exceeds, very slightly (,..._,1·002) the infinite
plate value if a> 1 · 2h for n = 1.
The above discussion indicates that, for plates with alb> 1, the end
effects for g= + 1 and g= - l are practically independent of each other.
Thus, for such plates the stress distribution between g= 1 and g=0 may
be taken to be the same as in a semi-infinite plate. -0·2
Introducing the co-ordinate
x'=a-x
from the free end and putting - 0·4 r---+r---t----t---+---+---+-----l
x' a-x
{=b="=o-e>b
a I
I Method of Virtual Forces
we find that for a-+oo, Eq. (74) reduces to -06-
r/>(~)= 1 ;isinw{:wcosw{e-il; ....................... . (81) I Assuming £ 11 =0
which is the appropriate solution for the semi-infinite plate. Note that,
for the particular temperature distribution of Eq. (76), ji and w are pro-
portional ton and hence lj>({) is solely a function of n{. The corresponding
complete stress system is : Fig. 9.-Third Example: therm!'l stresses in rectan1ular plate; lon1itudinal
w{+w cos w{ -ii:]
var1ation of stresses
~_tf21 -I.- cos2n7TTJ[1-ii sin I
Ea0.- d7J 2 'f'- 2 w e
df d'l'
·J. • 2n 7TTJ w -z . -r -~<·
-2 +11-
I~
sm
Un
Ea0. = dTJ d{=- ~ ----w-
sm we, e • la~ter method with t~ose derived from our principle of virtual forces. To
this purpose we consider for the temperature function (76) the plate under
u1J11 tP!fo +(l+cos2nry) C.02+;i2[_ -r _ . _,.1-;;· Jl an end load stress-system
Ea0. =1 d{2= 81r2n2 -w w cos w.,-IL sm we, e '
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (82) -u = + Ea 0 cos 2n7TTJ
.. 0 2 ...............................• (84)
ji and ware defined in general by Eq. (73) and for the particular tem-
perature distribution of (76) by Eq. (80). The stress system (82) is illus- app~ied at x~ ±a. Now, in .a plat~ ~ith Ev.• =O, the s~ress system (84) is
trated in FIGS. 8 and 9 which show the transverse and lengthwise varia- an etgenload, I.e. the cross-wise vanat1on of Its stresses IS invariant with x. •
tions respectively of the stresses for the case when n = 1. The eigen-value ,\ corresponding to (84) is simply
u .. is the longitudinal direct stress u:u in an infinite plate
,\=~1rn/b
u.. = -Ea0. cos f'7TTJ .................................. (83) and the diffusion coefficient
Maximum stresses: JLD=M(G/E) 112 =21Tnba(G)112
E
For any T} the transverse direct stress (u••) reaches its greatest value at
the end of the plate ({=0) and is given by The lengthwise variation of the u.,.,-stresses is obviously in this case
un 2 - .~ /c cosh ILDg
Ea0. =f(fi. +w 2)=f'V A cosh IL»
The maximum value (at TJ =0) is Thus, the stresses in a plate subjected to temperature function (76) are
( ) -2 -2 when Evv-O• easily found to be '
max.=~ t~ =1·155
l
Uww
27T n
(u ..) •...,
.!!E-.=_ cos 2n7TTJ(t- cosh IL»g)
For the maximum shear stress (u.,.) Ea0. 2 cosh ILD
YJ=l/4, {=0·218n
and thus
~= _
Ea00 E
112 (Q)
sin 2n7TTJ sinh ILDg
2 COShJLD J ... . ....... (8~
(uo:w)max. =0· 302
(u.. ).=<> ..5L = _ (Q) I+ cos 2n7TTJ cosh IL»g
We mentioned on page 77 that it is possible to find the stress distribu- Ea0. E 2 cosh ILD
tion in a finite plate by superposing on (48) the diffusion stress-system or for a long plate
arising from -u.,., applied at the two ends x= ±a. In aircraft structural
theory such diffusion analyses are often based on the simplifying assump-
tion Ew=O. It is interesting now to compare the results obtained by the p. ~~;' Argyris, Dunne, 'The General Theory, etc.', Part V, J,R.Ae.S .. Vol. Ll, November 1947,

78
 __!!iu_ _
Ea0o-
-(Q)
E
112sin 2mr7J .-- •
2 e 11 Ir (86)

a.. GE l+cos2mr7J
rae.=-- 2 e
_-.
1'D·

where
fLD ~~ 27Tn(~) 112
For G/£=0·385 we have: Fig. 10.-Fourth example; thin solid section in torsion
liD = I . 2047TII For the general background of the St. Venant theory of torsion of solid
sections, the reader is referred to: Timoshenko and Goodier8 , Theory of
which is in reasonable agreement with the value of IL in (80). However. Elasticity, 2nd edition, ch. II, McGraw-Hill, 1951.
the corresponding stresses are very different as may be seen from FIG. (9)
for n = 1. It appears that the assumption €•• =0 overestimates considerably (a) Principle of Virtual Displacements
the rate of diffusion of the end load system -au in the range 0· 7> ~>0. To obtain the linear au variation we assume that the warping displace-
5. FOURTH EXAMPLE ment w of the cross-section can be written in the form
II' _c, ~.()' =yj(X). B' • ' • •• • •. • • . • ' • •. •. •• • • ' • • • •• • •• • • •• • (92)
St. Venant Torsion of a Thin Section Bar
In treating the torsion of uniform bars of thin section, approximate where j(x) is a function of x only and is to be determined.*
Then in terms of Eq. (92), the shear strains and stresses are:
expressions for the maximum shear stress and the stiffness may be
obtained by assuming the distribution of shear stress across the thick- ,,cGB'[~W-)']=,GB'y[
ness to be identical with that for a very wide strip of constant thickness
a u =G€ g ~X ~
df.-l] (93)
equal to that at the point considered in the cross-section. Thus for a
section as shown in FIG. 10 we obtain the linear variation of a.x acro!.s the a,.=-GE,.=GB'[~ +x] =GB'[f+x] .................... (94)
thickness
dB We consider virtual displacements defined by an increment 8/ of the
a •.,=-2yGdz (90) function f. For such virtual displacements B' =dB/dz remains constant
and therefore no virtual work is done by the applied torque T. Thus, 8W
with the maximum value is zero and Eq. (44) of Part I reduces to the principle of minimum strain
dB energy Eq.(44a) of Section 4C which takes here the form:
a,.,max.=tGdz 8U;=~Jf[au8E,r+a, 11 8E, 11 ]dxdy -~0 . . . . . . . . . . . . . . . . . . . . . . . . (95)
and the torsion constant where the virtual shear strains due to the increment of are from (93)
and (94)
J=Hy0 3dx .......................................... (91)
8€,,,~B'y8(:fx) O'_,.t!!.Jt> -~ . . . . . . . . . . . . . . . . . . . . (96)
To obtain a more accurate solution we assume still that azr varies
linearly across the thickness, but determine the variation over the width b€,11- 8'8.f J
by application of the energy theorems enunciated in Part I. A solution is Substituting from (96), (93) and (94) into (95), integrating with respect
obtained firstly by application of the Principle of Virtual Displacements to y between ±Yo, we find
with an assumed transverse warping distribution to correspond to the
linear au distribution and secondly by application of the Principle of
Virtual Forces with the assumed stress variation. The two methods provide 1{'~\X -I Y~f/+yo(f+x)b{ }dx=O
x,
(97)
upper and lower limits respectively for the St. Venant torsion constant J
(see Sections 4 and 6 of Part 1). These two analyses are fully worked out and integrating the first term by parts
for singly symmetrical double-wedge sections and the results illustrated
in a number of figures. The closeness of the upper and lower limits prove
the methods to be very accurate. This is underlined by the coincidence of
the approximate and exact solutions in three cases. (98)
Considerable effort has been devoted in the past in developing many
approximate methods for the solution of the St. Venant torsion problem: Since 8/ is an arbitrary virtual increment off
some of them are discussed at the end of this example. Particularly relevant d [y}(df
-- -- --I )] v <!-+ x) "~0
to our present analysis is the approximate solution for the isosceles dx --3 dx ·"
(99)
triangle developed by Duncan, Ellis and Scruton* which is, in fact, identical which is the required differential equation for/, and also
with our solution of this case based on the Virtual Forces Principle. t
Additional Notation
Y·/(:fx -- I)= 0 at the limits x 1 and x 2 •••• . ••••• •••••••• (100)

a, al> b, c, !=dimensions of cross-section (see FIGS. 10 and 15)

giving the necessary boundary contitions at the ends of the cross-section.
These latter arise from the virtual work of the stresses on the edges of the
a, a 1 =a/b, a 1 /b respectively strip and are, in fact, the equilibrium conditions there in terms of the
Ox, Oy=axes in plane of cross-section assumed transverse stress distribution.
(Ox along axis of symmetry of cross-section)
Torsion Constant J
Oz=longitudinal axis of bar To determine J by the Principle of Virtual Displacements, we consider
Yo= y-eo-ordinate of section boundary a virtual increment 8B' of the rate of twist of the bar. Then from Eq. (44)
g, g1, ?]=non-dimensional co-ordinates (see nu. 15 and of Part I
Eqs. (142) and (143)) 8U,: -8W=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (101)
B' =dB /dz =rate of twist of bar about z-axis where
T=applied torque about z-axis 8W=T8B' . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (102)
J= T/GB' =~St. Venant torsion constant and
8U;=Jf[a,x8€u+a, 118€:u]dxdy . . . . . . . . . . . . . . . . . . . . . . . . . . (103)
w=wB'=warping displacement of cross-section
In calculating 8U,. due to the increment 80' we note that part of the
«!>=stress function virtual strains arises from the corresponding increment of the warping
/, /1> cf>, 1> 1 =functions of x displacements which are proportional to B' (cf. Eqs. (92), (93) and (94)).
n=a,,max./GtB'=maxi mum stress parameter But the increment of strain energy due to any virtual warping displace-
ments is zero (Eq. 95). Hence for the virtual strains giving 8U; in (101)
we can omit the part due to warping and take simply
• W. J. Duncan, D. L. Ellis and C. Scruton, 'The Flexural Centre and the Centre of Twist of an
Elastic Cylinder', Phil. Mag., Series 7, Vol. 16, 1953. p. 231.
t After completion and submission for publication of the present analysis the authors' attention was ::::::::::
drawn to: E. H. Mansfield, The torsional rigidity o.fsolid cylinders of double-wedge section, A.R.C. } .. .. .. .. .. .. .. .. .. . .. . .. .. . .. . (104)
Report, 1954, to be published in the R. & M. series. In tbis paper the same upper and lower limits
are obtained for the torsion constant of a double-wedge section by a basically identical method.
The stress distribution is, however, not discussed. • See Part I, Section 4E.

79
 Using Eqs. (93), (94), (104), (103) and ( 102) in ( 101 ), and integrating which gives, on integrating by parts the first two terms,
with respect toy, we obtain for the torque T due to the shear stresses x,

i2~os~~Scf>+y/Jx·c/>ocf> I
J[ ~_Y]\1;
x,
oU;*=G(0')2;{
T=2GO' -I )+xyo<f+x)Jdx (105) x,
x,
Substituting from the differential Eq. (99) for Yo(f+x), integrating by - J{fx(l~os ~~) +Yoa[Ji(Yo1xo) -1 ]c/> }ocf>dx}
parts and noting the boundary conditions (100), we finally obtain ................................ (121)
Since ocf> is zero at the ends of the section the first term vanishes and
(106) there remains only
x,
and therefore the torsion constant is
oU;* - -G(0') 2 ·H {fx(l~o5t)+y.{fx(Yo1x0)--l]c/> }ocf>dx (122)
_ _I_ -~Jx,
J~G0'-3
df)a(
Yo l - dx dx (107)
From Eq. (119) for S T, the virtual complementary work is

x, ow• =OT·O' =G(0') 2· Hyo 3ocf>dx (123)

(b) Principle of Virtual Forces Substituting now Eqs. (122) and (123) in Eq. (116) we conclude that in
To obtain a shear stress system satisfying the internal equilibrium order to satisfy (116) for any arbitrary ocf>, the function c/> must satisfy the
conditions, we express the shear stresses in terms of a stress function <f) as differential equation
()<f} ()<f}
a,.,= ()y, au= -()x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (108) dx T de/>)
d (2yo 5 d ( Yodx
dx +Yo dx a[
dyo) -I 'I'+ Yo 3 =0 ].1. ........... . (124)

The equilibrium conditions on the boundary will also be satisfied if* and must be zero at the ends of the cross-section.
<f)=O at all points on the boundary. (Aoq)
APPLICATIONS TO PARTICULAR SECTIONS
For equilibrium of the shear stresses with the applied torque
We now apply the equations developed in the previous section to obtain
T=ff[a.wx-a,,y]dxdy approximate solutions for the shear stress distribution and torsional
which gives on integration by parts with respect to x and y for the first stiffness of two particular types of cross-section: first the simple rectangular
and second terms respectively and noting the boundary condition (109) section and second a double wedge section formed by two isosceles triangles
T=2JJ<f}dxdy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (110) having a common base. Solutions are obtained by both methods and the
results compared with one another and also with relevant exact solutions.
We assume now that the stress function <f) can be written in the form
2
<f}=G0'(y0 -y )c/>(x) .. . ... . . .. . .. . . . .. . ... . .. . . . .. . .
2 . . (Ill) 1. Rectangular Section
where c/>(x) is a function of x only.t (a) Solution by the Method of Virtual Displacements with Assum?d Trans-
This satisfies (109) at the upper and lower boundaries and gives our verse Warping Variation
assumed linear transverse distribution of azx· Substituting from Eq. (Ill)
in (110) we find for the torque Consider a rectangular section of width s and thickness t. Taking the
origin of axes at the centre of the section, we find for the differential
T=G0'Hyo 3 c/>dx ...................................... (112) Eq. (99) in this simple case
d2[
whence d~2 -p,2f== ph ........................................ (125)

J=Jo,=Hyo 3 c/>dx ............................. ....... (113) where

JL 2= 12/t2 • • • • . • • • . . • • . . • . • • • • • • . • • • • • . • . • . • • • . • • • . • . • (126)
The stresses and strains corresponding to (Ill) are from Eq. (108)
The boundary conditions (100) are here
a,,=~C:=-2yc/>·G0'=GEzx .......................... (114) d{
dx =I for x= ±s/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (127)
()<f}-
a•• ---()x de/> dyO,J.] G0'-
- - [ (Yo 2-y2)dx+2Yodx'l' -GEzu . . .. . . .. (115) and the appropriate solution is therefore
2 sinh JLX
In the general equation of the Virtual Forces principle (Eq. (62) of f=ji:coshJLs/2-x ················ ··················· (128)
Part I, for 0 =0).
sui• -ow•=o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (116> Stress Distribution:
in which Substituting for /in Eqs. (93) and (94)
oU;* = ff(E.,oa._,+E •• oa•• ]dxdy . . . . . . . . . . . . . . . . . . . . . . . . . . (117) , [ cosh JLX
az,=GO · ly cosh JLS/2 -I
J II
and the virtual stresses oa•.,, oa•• are obtained by giving to the function c/> ?- (129)
a virtual increment ocf>. Such virtual stresses automatically satisfy the
internal equilibrium conditions. We note moreover that since <f) =0 is
prescribed all along the boundary, we must take Scf> zero at the ends of
a
•• -
~GO'.~ sinh JLX
JL cosh JLS /2 J
The distribution of shear stress according to Eq. ( 129) is illustrated
the section.
graphically in FIG. II.
From Eqs. (114) and (115), we find for the virtual stresses due to ocf>
oa•..,= -lySe/> . GO' I
J ......... . (118)

and from (112), the virtual torque due to ocf> is

sr=Go'Hyo3ocf>dx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (119)

Substituting for the strains from (114) and (115) and for the virtual
stresses from (ll8), integrating with respect toy between the limits ±Yo,
we obtain from Eq. (117) the virtual complementary energy

sui• =G(0'>2H [2~os;t fx<ocf>>+ y.'!fxo fx<y.acf>Sc/>>+Y.ac/>ocf> Jc~x

Assumed C1z•
distribution
(120)
tis • 0·20
• See Timoshenko and Goodier•, and also Part I, Eqs. ( 5). Fig. !I.-Rectangular s~tion; shear stress distribution according to
t See Part I, Section 6E. Virtual Displacements solution

80
 1·0

--- Virtual Forces

Virtual

m· _jJ
.........
.........
---Virtual Forces

~s--C' o
-Virtual Displacements
Exact Solutions

0·4
0·2 0·4 0·6 0·8 1·0
tis
Fig. 12.-Maximum shear stress for uniform torsion of rectangular section

The maximum stress is reached at the mid-point of the longer sides and is
amax. 1
Gre'=n=l-cosh Oa · · .. · · · · · · .. · · .. · · · · · .. · · · .. · · · · · · <130)
where
8a=p.s/2=v3s/t . .. . . . .. . . . . . .. . . . .. . . .. . . . .. . .. . . .. . . (131)
Eq. (130) is plotted as the broken line in FIG. 12 with t/s as abscissa. 0·2 0·4 0·6 0·8 t·O
We note that in the extreme case of a square section (s/t=l) n=0·657, t/s
which is 2 · 7 per cent less than the exact value 0 · 675. * An obvious error Fig. 13.-Torsion constant J for uniform rectangular section
in the detailed stress distribution is the constant a .. with y leading to
unbalanced stresses on the boundary.

Torsional Stiflness:
The general expression (I 07) for the torsion constant gives, with the
/-function of Eq. (128)
3 +s/2
J=t_J[l- cosh p.x ]dx
3 cosh p.s/2
-s/2

= s~ll-ta~~ OaJ (132)

The broken curve in FIG. 13 shows the variation of 3J/st3 with t/s
according to (132). In accordance with the general Principle of Virtual Assumed C1zx
Displacements this approximate solution must always overestimate the 0 1 .JL distribution
stiffness. L.........l.... Gt~
tis • 0·20
(b) Solution by the Principle of Virtual Forces with Assumed Form of Stress
Fig. 14.-f\ectangular section; shear stress distribution according to
Function Virtual Forces solution
The differential Eq. (124) becomes in this case
where
~~->..2cf>= -,\2 ...................................... (133)
where
8b=As/2=V~s/t (139)
The full line in FIG. 12 shows the variation of the maximum stress para-
A2 =10/t2 (134) meter n with tfs.
and the boundary conditions are
cp=O for x=±s/2 .. .. . . . .. . . . .. . . . . . . . . . .. .. .. . .. . .. . (135) Torsional Stiffness:
From (133) and (135) we find easily Eq. (113), in conjunction with the cp function (136), gives for the torsion
constant
cosh Ax
tP = 1 cosh As /2 · · · · · · · · · .. · · · · · .. · · · · · · .. · · .. · .. · · .. · ( 136) J=s~l 1 _tan8~eb] .................................. 040

Stress Distribution: which is shown as the full line in FIG. 13. We find remarkably close agree-
ment with values given by the exact solution.* Thus for a square section,

I;
Substitution for cp from Eq. (136) in Eq. (114) gives for the stresses which can hardly be considered as thin, Eq. (141) gives the value 0·419st3 /3
which is only 0 · 7 per cent below the exact value. In accordance with the
a =-2y[l-cos\.\x]·G8' general Principle of Virtual Forces our present approximation must under-
""' cosh 1\S /2
(137) estimate the stiffness.
t2 ) , sinh Ax (}' ............ · ...... ·
( 2
a •• = 4 -y 1\ cosh >..s/2 · G 2. Double Wedge Section
FIG. 14 illustrates these distributions for a section with s/t=5. We note Since the centre of twist for unrestrained torsion is arbitrary, we assume
that although the a,., distribution is generally similar to that of FIG. 11, for convenience that it is at the centre of the common base of the two
the a •• distribution now satisfies all the boundary conditions. isosceles triangles which form the section (See FIG. 15). We introduce also
For the maximum shear stress (at the mid-point of the longer sides) the non-dimensional co-ordinates
we find g=(x+a)/a, g1 =(a 1 -x)/a 1 . .. . .. . .. . . . . .. .. . . . . .. . .. .. (142)
amax. 1 measured inwards from the ends of the section and
Gt8' =n=l-cosh 8b · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · (138) 7J=y/b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (143)

• Timosbenko and Goodier', foe. cit. • Timosbeoko and Goodier', /oc. cit.

81
 r_·

1--------- c
Fig. IS.-Double wedge section 0 I _Q_
I Gt .:j
Then for the left-hand part of the section
....I..,JIU...,!. L . J •
I - !....... . . . . . . ...........

x=a(' -1), y.=,b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (144)

and for the right-hand part
tic • 0·20 0/C • 0·50
x=a1(l-,1), y.=,1b ·............................... (l44a) Fig. 16.-5hear stress distribution in symmetrical double wedge section
t/C=0·2; approximate solution by Virtual Displacement method
(a) Solution for Assumed Warping Functions:
For this type of cross-section we have separate expressions for the
function fin the two triangular parts of the section. Writing f and / 1 for
the function in the left ( -a<x<O) and right-hand (0<x<a 1 ) parts re-
spectively, Eq. (98) becomes here
0 0

au;= Iy.3(~ -I )at I - J{~[y.3 (1x -I)

-a -a
J-y.(J+x) }otdx
+a. a1

+ IY. 3 (~-t)otii-J{~[y. 3 (~-I)]-y.(f1 +x)}ot1 dx=O

0 0
................................ (145)
0 1 ..sL

l
To satisfy Eq. (145), the functions/and/1 must, as before, satisfy the
differential Eq. (99) and the conditions at the ends of the section I I .. I ... I I Gt~

(Oyz)y.o
[Yo3 (~-t)l=-a =[Y.a(~1 -t)l=+a, =0 .......... (146)
In addition, however, we see that we must have -- -·~---------~--~~-
[y.3 (~ -I )atJz..o -[y.3 (1x1 -I )ot1].,.. 0 =0 (147) tic • 0·10 a/C • 0·10
Since we assume a priori that the displacements are compatible, it Fig. 17.-5hear stress distribution in single wedge section t/c=O·I;
follows that at the junction of the two regions approximate solution by Virtual Displacement method
f,._o=/ltz-o · ·.... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (148)
and therefore

l
(ofl-o=(o/1)-.o ···································· (149)
which gives in Eq. (I 47) (155)
[y.s(~-1 )l_o -[y.s(~I-1 )l=o =0
d' d'
(150)
![df] +_!_[df1] =0
Eqs. (148) and (150) are the boundary conditions at the junction of the a o=l a 1 1 o,= 1
two parts of t~e section, giving the compatibility and equilibrium require- The first of these gives immediately
ments respecttvely. B=B 1=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (156)
With x and Yo from (144), the differential Eq. (99) becomes, for the and substitution from (153) and (153a) in the last two of (155) leads to
left-hand part of the section
do/ df
' 2iif2+3'a,-3a 2/=3a'(l +a 2)-3a 2a .................. (151)
A _ 2a({J1 2 +fJ 1-2)+a 1(fJ2+fJ 1-2)
- ({J 1+2)[a({J1-I)+a 1({J-I)]
A -2a1(fJ2+fJ-2)+a(fJ12+fJ-2)
I
I
............. .
(157)
where
a=a/b (152) 1- ({J+2) [a(fJ 1-I)+a 1({J-I)]
The complete solution of ( 151) is For a doubly symmetrical section (a=a 1), Eqs. (157) reduce simply to
a [
f=as_ 1 A'
/1-1
+B'
-/1-1
_,(a 2+l)+(aLl) ........... .
J (153)
A=A 1=2 ................ ·~·..........................
and when a 1 =0, i.e. the section consists of a single isosceles triangle
(158)

where A=2(l +fJ)/3 . . . . . . . . . . . . . . .. . .. . . . . . . . . . . . . . . . . . . . . . (159)

fP=l +3a 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (154)
Similarly we find for the right-hand part of the section (substituting Stress Distribution:
from (144a)) the differential equation From Eqs. (94) the stresses corresponding to the general solution (I 53)
are
i:. 2do/1 i:. d/1
!>1 d' 12+Js 1~~ -Ja1o/1 = -3a 1, 10 +a 1 2)-i-3a 12a . . . . . . . . (ISla) a
O"zv=-2a2-J' l-2g
[ A
bG8' ..................... . (160)fJ-2]
and the solution
a1 [ /1,-1 -/1,-1 J a
2a2-)7J
2[
3A ~:.fJ-2] 8 ,
l-2(1+fJ)S bG ................ .. (161)
/1=a 12_1 -A1'1 -B1'1 +' 1(a 12+1)-(a 1 2-l) . . (I53a) O"zz=-

where The maximum stresses obviously occur on the boundary of the section
(7]=,). Resolving there into tangential and normal components we find
a 1 =a 1 /b, {J 12= I +3a 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (154a)
Boundary Conditions:
In terms of the non-dimensional co-ordinates, the boundary conditions (162)
(146), (148), (150) may be written:
82
and TABLE I
Comparison of Values of Maximum Stress Parameter n with Exact Soluti,.,.ns
aazy-U.rz
O'zn = (l +a2)1/2 Infinitely Right Angled
Thin Square Equilateral Isosceles
a 2A
=(a2-l)(l+a2)112 P+t ~
(fl-2)
fl-1 ,
bGB #0 .. . .. . ... ... .. (163)
Section
t/c= l/(a 1 +a,)-->0 a=a 1 =1
Triangle
a=Vl, a,=O
Triangle
a=l.a 1 =0

Expressions analogous to (160)-(163) are obtained for the right-hand Exact Solution 1·00 0·-4n• v'3/-4=0·-433• 0·32lt
part of the section and need not be quoted here. The distribution of shear Virtual Displacements 1·00 0·-4-43 1=0·375
0·-461
stress is illustrated in FIGS. 16 and 17 for some particular cases. As in the Solution (a) Exact (-3·5%) (+l·3%) (+ll·5%)
previous example, our assumed form of warping displacements, yielding
a constant u ·~withy, leads to unbalanced shear stresses u zn on the boundary. Virtual Forces 1·00 0·-485 v'3/-4=0·-433 t. ==0·3125
The maximum u,. stress is given by Solution (b) Exact (+2·0%) Exact ( -60~)

u,.max. a(a 2 +1) 1 12 [2(a 2 +1) Jl/(fl-2)

n= GtO' =J(fi+2)(fi-1) Afi(fi-1) (164) • Timoshenko and Goodier'. Ch. II, pp. l66 and ln.
t Galerkin, B. G. 8ui/.I'Acad:miedesSciencesdeRussie, p. Ill, 1919; Kolossol, G. Compt. rend., Vol. 178,
and occurs at p. 1057, 1924.
fl-2 2(a 2 + I)
~ = AfiCfi -1) (165) TABLE II

In an unsymmetrical section (a#a 1) the maximum stress is always on Comparison of Values of 1 l IT With Exact Solutions
et'

the longer side and hence in general we use in Eq. (164) the greater value Infinitely Richt Ancled
of a. In FIG. 18 the broken lines show values of the maximum stress Thin Square Equilateral Isosceles
parameter n calculated for the cases a=a 1 and a 1 =0 and plotted against Section Trian&le Triancle
I fa. TABLE I gives some particular results for sections where exact solution t/c-->0 a=a 1 =-l a~ v'l, a,==O ao:.~ I, a 1 =0

is available.
Exact Solution 1·00 0·-422• 0·300· 0·1567t

Torsional St!ffness: Virtual Displacements 1·00 .,\=0·-437 0·338 ... ==0·1B7S

Solution (a) Exact (+3·7%) (+ll·6%) (+19·7%)
Eq. (107) takes here the form
Virtual Forces 1·00 ~~=0·406 0·300 ,;:,~0·1563

J~a( t-1 ~)d~+a 1 J ~~a( I+i

I I Solution (b) Exact (-3·7%) Exact (-0·16%)

1 ~ )d~ 1
1ta=a 1 ....... . (166)
• Timoshenko and Goodier'. Ch. II, pp. l66 and ln.
0 0
t Galerkin, B. G. 8uii.I'Academiedes Sciences de Russie, p. Ill, 1919; Kolossol, G. Compt. rend., Vol. 178,
which with the values off and / 1 from (153) and (153a) reduces to p. 1057, 1914.

3J aa 2 [ 6A a 1a 12 [
2ba=aLJ I-~ +a1LJ I
J 6A1
(f1t+0(f1t+2)J
,
(167) Similarly, we find for the isosceles triangle
12J fi2-t
where A, A 1 are given by Eq. (157). (170)
eta =(/3+2) 2
For a very thin strip, i.e. when a, a 1 become infinitely large, the limiting
form of (167) is Eqs. (169) and (170) are plotted in FIG. 19 against the ratio 1/a and in
TABLE u particular results are given for comparison with the corresponding
(168) exact solutions and with the results of the Virtual Forces method.
where t=2b and c=a+a 1 (see FIG. 15).
Eq. (168) is the result given by the rough approximation of Eq. (91); (b) Virtual Forces Solution With Assumed Form of Stress Function
it is exact for the infinitely thin section. As in 2(a) we have separate expressions for cp and cp 1 in the two triangular
For a doubly symmetrical section (a=a 1 ) the values of A and A 1 of parts of the section. Following the argument used there and noting that
Eqs. (158) lead to the simple result at the junction of the two parts the two functions must be equal
12J <11 +5)([1-1)
eta = <fi +2)2 (169) cf>x=o=c/>~o:r=O · · • • · · • · · · • ·• · ·•..•.•••..•.•.•.•••••...• (171)

0·8

0·6

c/12
0·4

Virtual Forces
0·2 Virtual Displacements 0·2 Virtual Forces
0 Exact Solutions Virtual Displacements
0 Exact Solutions

0·4 0·6 08 1·0 0·8 1·0

b/a
Fig. 18.-Maximum shear stress for uniform torsion of double wedge Fig. 19.-Torsion constant J for double wedge section: comparison of
section results by Virtual Displacements and Virtual Forces methods

83
 0 ..JL
Gt.f/

0 1 _g_
1.. 1.._,......_'""'-'-_._.'-''.._,wl Gt:5
f/C : 0·10 a/C = 1·0
f/C • 0·20 ale • 0·50
Fig. 21.-Shear stress distribution in single wedge section tjc--0· 1:
Fig. 20.-Shear stress distribution in symmetrical double wedge section approximate s&lution by Virtual Forces method
t/c=0·2; approximate solution by Virtual Forces method

and hence the virtual increments are also equal

[Sc/Jl.--o=[Sc/J1lx-o .............................. ...... (172)
we deduce from (121) and (116) the additional condition
[ 2yo
5
5 dc/J+
dx Yo '+' x-o
4-1..]
5 dx Yo '+' 1 x-o
=[2yo5 dc/>1+ 4-1.. J
which may be written more concisely as

!!._(Yo5/2J..)J
[ dx dx '+' 1)] x-o =0
'+' x-o -[!!_(Yo5/2J.. ............... ·.·. (173) • I

~~==------~-~--
r-
Eqs. (171) and (173) provide the necessary additional boundary conditions
for this section. Eq. (171) is the equilibrium condition (equality of au)
and (173) is a compatibility condition. From Eq. (115) we see that it is
r""'==..:J::;;..,.., _I

not possible to satisfy the compatibility condition (equality of *'••) for all Oyz'
values of y because of the discontinuity in dy 0 /dx. Eq. (173) expresses the
requirement for the virtual complementary work at the junction to be t/c : 0·05 ate • 025
zero and is thus an averaged compatibility condition. Fig. n.-Shear stress distribution in unsymmetrical double wedge section
The differential Eq. (124) becomes here, on substituting for x and Yn t/c=0·05, afc=0·25; approximate solution by Virtual Forces method
from (144), for the left-hand part of the section*

~ de+ 5~d~+2~( 1 _
c2t!:.P. ct!J! a2)-1..'+'- ~ 2 ....................... .
__ 2a (174) Stress Distribution:
From Eqs. ( 108) or ( 114) we find, for the function c/J of ( 175), the stresses
the complete solution of which may be written
b-2] bG8'
a2 [
c/>=a:2_ 1 1-cg
6-2
-ng -6-2] ....................... . (175)
a2 [
u,.= -2a2 _ 1l] 1-C~
l~ (181)
Similarly we find for the right-hand part of the section
Uzy= -2a2a_ 1g{ 1- s;l-l~ 2 -( 1-Dl] 2] }bG8'
J
I
a 12 [ 6,--2 .. 1),-2]
cP1=a 12_J 1-c1~1 -D1g1 (175a) and at the boundary (lJ =g)
where Uz.,=2-
a(a2+J)ll2 [
a2-=r-~ 1-C~
/j-JJ bG8, I
~ (182)
S2=(3+5a 2)/2, S 12=(3+5a12)/2 (176) Uzn=O J
The maximum value of u .. is given by
Boundar.y Conditions: Uz 8 max. a(a 2 + 1)1! 2 [ I ]1/(1)-2)
Since <I> must be zero at the ends of the section g=g 1 =0 (equilibrium n= Gt8' = 52(S+2)(S-I) C(S-1) (183)
on boundary)
Similar expressions are ebtained for the right-hand side of the cross-
D=D 1 =0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (177) section and need not be given here. Again we find that, in general, the
Eq. (173) becomes, on substituting for x and Yo from (144) and (144a) maximum stress occurs on the longer sides of the section. In FIGS. 20, 21
and 22, some typical distributions given by Eqs. (181) and (182) are shown
graphically. Note the discontinuity in Uz• at the junction of the two
![.!!.
a d~
(g512c/J)] +_! [_!!__<g 5/2-1.. )]
a1 dg1 1 '+'1 =0 ·············· (173a) regions. This appears because the variation with y is fixed by the assumed
<-1 <,~1 form (Ill) of the stress function which cannot represent adequately the
For the functions cP and c/> 1 of (175) and (175a) the boundary con- effect of the sudden change of slope of the boundary. It does not arise
ditions (171) and (173a) yield in the case of the single wedge (a 1 =0) for which we obtain more accurate
_ a 2(S 1 +l)+aa 1 (S 1 +2)+a 1 2 I results. In fact, for an equilateral triangle (a 1 = v'J, S =3) our assumed
form for the stress function gives the exact solution. Thus, the stress

J ··········
C- 5a(S 1 +2)[a(2S 1 +1)+a 1(2S+1)] I function* for this case is
(178)
(184)
C 1 = 5al(S+2)[a(2S 1 + 1)+a 1(2S+ I)]
When a =a 1 Eqs. (178) give simply which is easily shown to satisfy the compatibility equation
5 0 2<1>
ox 2 + 0oy2<1>2 +2GB ' =0 .................................. (185)
C=Cl=2S+I ....................................... . (179)
and when a 1 =0 Values of the maximum stress parameter n are plotted in FIG. 18 against
I fa for comparison with the results of the previous method and TABLE 1
C=l ............................................... . (180) gives some particular results for cases where the exact solution has been
given.
• This equation, _in slightly different form, was used lor the isosceles triangle by Duncan Ellis
and Scuton, /oc. Cit. ' • See Timoshenko and Goodier!, Joe. cit.
84
 0-9

0·8

J
cto/12
0·6

Fig. 23.-Torsion constant j for general double wedge section

Torsional Stiffness: Accuracy of Results

Using Eq. (115) we find for the torsion constant In an approximate solution it is always desirable to be able to assess
the order of accuracy achieved. In the two methods applied in this example,
3J I I the method of Virtual Displacements gives an overestimate whereas the
8ba=alg 3 cfodg+at bgt 3cPtdgl .............. .. ............ (186) method of Virtual Forces gives always an underestimate for the stiffness.
Thus, by using both methods we have succeeded in establishing upper
which with Eqs. (175) and ( 175a) reduces to and lower bounds for the stiffness.

12J a2 [ 4C
8ba=aar::-t 1- 3+2 +atUT!
J
a 12 [ 4C 1
1-sl +2
J ......... . (187)
Some idea of the relative accuracy of the two methods can be obtained
from TABLES 1 and n by comparison with results of exact solutions. Note,
however, that these are for sections well beyond the intended range of
application of the solutions developed here and the basic approximations
where C and C 1 are given by Eqs. ( 178). are not really suitable for such sections. Nevertheless the agreement
Eq. (187), which is very similar in form to (167), yields the same limit appears to be quite good.
as a and a 1 become infinitely large. In general the Virtual Forces method appears to give the most accurate
solution, in particular for the triangular section where the general formula
When a=a1 we find (189) for J gives the exact value at both ends of the range lfa~o and
12J 5 a 2(2S+9) a= v3. In FIG. 23 we show finally the torsion constant J calculated by the
(188) Virtual Forces method for the general double wedge section. Note that
ct3 =2 (2S+I)(S+2) 2 FIG. 23 shows minimum values of stiffness. For t/c<O· 5, however, the
and when a 1 =0 error is less than about 0 · 5 per cent.
Although the general theorem as to upper and lower bounds does not
12J 5 a2 apply to the maximum stress, since it depends much more on the details
(189)
ct 3 =i ~ of the initial approximation, it appears here (see TABLE I and FIG. 18) that
Eqs. (188) and (189) are plotted in FIG. 19 as the full lines and on TABLE 11 the exact solution does lie between the results given by the two methods.
some particular results are given.

85