Professional Documents
Culture Documents
(J. H. Argyris D.sc. Energy Theorems and Structural Analysis
(J. H. Argyris D.sc. Energy Theorems and Structural Analysis
Argyris
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
J. H. Argyris 1960.
Originally published by Plenum Press in 1960.
Softcover reprint of the hardcover 1st edition 1960
PREFACE
By J. H. Argyris
T
1. INTRODUCTION GENERAL REFERENCES
(I) Biezeno, C. B., and Gramme!, R. Technische Dynamik, 1st ed., Springer, Berlin, 1939.
HE increasing complexity of aircraft structures and the many exact (2) Engesser, F. Z. Architek u. lng. Verein Hannover, Vol. 35, pp. 733-774, 1899.
or approximate methods available for their analysis demand an (3) Lord Rayleigh. Theory of Sound, 2nd ed., Vols. I and II, Macmillan, London, 1892 and 1896.
integrated view of the whole subject, not only in order to simplify (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.
their applications but also to discover some more general truths and (6) Mueller-Breslau, H. Die neueren Methoden der Festigkeitslehre und der Statik der Baukon-
methods. There are also other reasons demanding a more comprehensive struktilmen, 1st ed., Korner, Leipzig, 1886.
discussion of the basic theory. We mention only the increasing attention 19 ~7. Southwell, R. V. Introduction to the Theory of Elasticity, 2nd ed., Clarendon Press, Oxford,
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 19~8/ Timoshenko, S., and Goodier, J. N. Theory of Elasticity, 2nd ed., MacGraw-Hill, New York,
unified analysis appears more than necessary. (9) Timoshenko, S. History<;{ Strength of Materials, MacGraw-Hill, New York, 1953.
With this present paper we set out to develop a comprehensive system (10) Tretftz, E. Handbuch der Physik, Vol. VI, Springer, Berlin, 1928.
_(II) Westergaard, H. M. 'On the Methods of Complementary Energy,' Proceedings Amer. Soc.
for the determination of stresses and deformations in elastic structures C.v. Engrs., 1941, p. 190.
based on two fundamental energy principles. Although much of the theory (12) Argyris, J. H. Thermal Stress Analysis and Energy Theorems, A.R.C. 16,489, Dec. 1953.
given has naturally been known for many years we believe that some of the (13) Argyris, J. H., and Kelsey, S. Applications to A.R.C. 16,489, A.R.C. 16,513, Jan. 1954.
theorems and the generality of the results are new. The loading systems Additional references are given as footnotes.
considered are of an arbitrary nature and include ab initio the effect of
temperature or other initial strains. Neither do we restrict ourselves to results ar~ identical_ with exist~ng _ideas _they clothed them in a language not
elastic bodies obeying Hooke's law but take account of purely elastic non- too fam1har to engmeers. Th1s d1scuss1on of past authors' work brings us
linear stress-strain laws. This is possibly not of very great importance at to a few points which are preferably stated now. In much of present day
present but may have wider applications in the future. No problems of sta- structur~l analy~is there seems to be an unfortunate tendency to over-
bility will be touched upon in the present series of articles and any other emphasize certam methods of analysing redundant structures and to
considerations of large-deflexion theory are, in general, omitted. Thus the negl~-;t more useful ideas readily available for many years. This refers
purpose is to investigate, within the small-deflexion theory, the stresses particularly to Castigliano's principles which are so often set out as the
and deformations in elastic bodies not necessarily obeying a linear stress- basis of all ·considerations, not only in theory, but also in the actual
strain law and under any load and temperature distribution. Dynamic methods of calculation. This is, in our opinion, unfortunate, even though
effects are initially not considered and hence it is assumed for the present all methods naturally lead to the same results if based on the same
that the loads and temperature are of the quasi-static type. When in- assumptions. For example, if we select forces as redundancies then much
vestigating thermal strain effects we ought strictly to base the analysis on the best means of obtaining the basic equations for their determination is
thermodynamic considerations. These are, however, only slightly touched the long established O;k method of Mueller-Breslau based on the Unit
upon here. L<?ad idea. We do not ne~d, _in fac~, even the concept of strain energy for
As in all theoretical work, we start by discussing the exact implications th1s purpose. All we reqmre 1s the 1dea of work and kinematics as used in
and equations derived from the initial assumptions, but we do not restrict rigid-body_ mec~anics. From such ~deas we can write down immediately
ourselves here to this aspect. On the contrary, we pay close attention to our equatiOns m the unkl!owns Wlt~o.ut b~thering about strain energy.
approximate methods of analysis based on the physical concepts of work Thes~ l!lethods h~ve been m use by CIVIl engmeers for the past sixty years
and strain energy. In particular we attempt to give upper and lower bounds and 1t IS surely t1me that we accepted them in the aeronautical world as
to overall properties of the structure such as its stiffness. No attempt is standard analytical equipment. Actually, the basic principles go much
mad!! to estimate the error of stress and deformations at any particular farther back than Mueller-Breslau and were, in fact, developed independ-
point. ently ~Y Ma:'we_11 4 and Mohr 5 nearly a hundred years ago. The first sys-
This series of papers originally arose12 • 13 from lectures given by the ~ematlc app~1cat1_on of_the ~ik method to stressed skin structures was given
author since 1949-50 at the Imperial College, University of London. m the classical mvest1gat10ns of Ebner.* Regrettably enough this lucid
Naturally, the scope of the present work has grown beyond the narrower wor~ was occasionally referred to in the past as obscure, a lack of compre-
concept of undergraduate teaching, but the basis of the analysis dates back hensiOn, no doubt, at leas_t partly due to the too narrow understanding of
to that time. It is appropriate here to point out that certain of the basic redundant structures ansmg from a concentration on Castigliano's
ideas originate with Engesser2 who unfortunately does not seem to have methods. However, the limitations of Castigliano's formulation of the
followed them up. We refer, of course, to the two complementary con- problem are being at last increasingly recognized in aeronautical circles
cepts of work and complementary work. If we consider an ordinary load due to the demands of calculations for highly redundant systems. Natur-
displacement diagram, then, even if we restrict ourselves to small dis- ally, most of ~he_ alternative methods suggested are really nothing more
placements, this may be curvilinear, if the material follows a non-linear than a transcnpt10n of the Mueller-Breslau and Ebner technique.
stress-strain law. Work is the area between the displacement axis and the We start our investigation in Section 3 with a discussion on work and
curve, while complementary work is that included between the force axis complementary work in the presence of temperature effects and for non-
and the curve. Thus, the two areas complement each other in the rec- linear stress-strain laws. With this basic knowledge we then proceed to the
tangular area (force) x (displacement) which would be the work if the standard principle of virtual displacements or virtual work in Section 4.
ultimate force were acting with its full intensity from the beginning of the This is very similar to the currently used principle in rigid-body mechanics.
displacement. Naturally, in the case of a body following Hooke's law, the Thus, we consider a state of equilibrium, apply virtual displacements to it
two complementary areas are equal, but it is still useful for the purpose of and develop hence the classical principle of virtual work which sub-
analysis to keep them apart. Since writing a previous paper12 on the stitutes, of course, for the equations of equilibrium. Since virtual dis-
subject the author has had the opportunity of consulting the most in- placements are kinematically possible ones this theorem starts from the
teresting latest book 9 of Stephen Timoshenko. There a reference is made assumption of inherent compatibility to find the necessary and sufficient
to the work of Westergaard, 11 who indeed has developed further the basic condition for equilibrium. It is, of course, well known that the theorem
ideas of Engesser, but not on quite such a general basis as here. Since applies also to large displacements but this aspect is ignored here. How-
approximate methods figure prominently in this paper reference ought to ever, temperature effects and an arbitrary law of elasticity are considered
be made to the work of Prager and Synge. They too set out to develop as _lo~g as the latter is '!lonotoni_cally incre.1sing. Having established this
systematically the determination of upper and lower limits to strain prmc1ple we deduce easlly some Important theorems and applications.
energy, restricting themselves, however, to Hooke's law and excluding
• See e.g. H. Ebner and H. Koeller. 'Zur Berechnung des Kraftverlaufes in versteiften Zylin<Jer·
temperature effects. Moreover, it appears that although many of their schalen.' Luftfahrtfnrschung, Vol. 14. No. 12, December 1937.
2
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)
}
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"
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
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
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 Thus,
Fig. a.-Example of an arbitrary virtual displacement • See Biezeno and Gramme! (I), p. 74.
7
B
in terms of the displacements follow also from this analysis. It is im- !ransvers~ direction: -:'nother type of constraint may be achieved by the
portant to note that in all applications it is best to form directly mtroduct10n of a ngt_d support. Also the stress-strain diagram may be
SU1 = f[ailE]a v taken local~y to be II mstead of I (FIG. 9). In all such cases the arguments
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. 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
(B) Castigliano's* theorem Part I generalized for thermal effects agam keepmg constant the prescribed displacements the strain energy is
decreased. The relaxation of constraints may take the form of a stress-
[~~~ 1~= const. =P, · · · · · · · · · · · · · · · · • · • • · · · · • • • • · • • • • • • • · • (49)
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
where P, is the force (moment) applied in the direotion of the deftexion bending stiffness and admit only a membrane state of stress· a current
(~otation) u,_. TJ:lls relation may be obtained immediately if we apply a procedure in wing stressing. '
virtual elastic displacement Su, solely to one external load P,. Note that . Thus we co~cl';lde: the strain energy of an elastic body for a given set of
Eq. (49) applies also for non-linear stress-strain laws and may also be d1sp~acements 1s mcreased (reduced) by the imposition (relaxation) of con-
generalized for large displacements. 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
(C) The Principle of Minimum Strain Energy when U1 is expressed in circular shell under mternal pressure It ts Immaterial whether we take
terms of the displacements and the temperature is not varied. account of or ignore the bending stiffness of the walls.
We arrive immediately at this theorem if we select only such virtual Also since the constraints do no work it follows that the increase
displacements Su which are zero at the applied forces. Then (reduction) of the strain energy can only be produced by the forces P.
SW=O 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
and we conclude from Eq. (44) that mcreased (reduced) by the imposition (relaxation) of constraints that do no
S.U1 =0 and U;=min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (44a) work.
at true position of equilibrium if only such virtual deformations are . If we consi~er now the case of an elastic body under a given set of forces
allowed that no external work is done. mstead of displacements then we conclude immediately from the last
Hence, if we compare all possible compatible states of deformation of a the? rem: the displacements and hence also the strain energy in a body under
body associated with a given set of displacements (not sufficient by them- a g1ven set of forces are reduced (increas!!d) by the imposition (relaxation)
of constraints that do no work.
selves to fix completely the deformed shape of the body) then the true Both the last two theorems illustrate two complementary effects of the
position of equilibrium has the minimum strain energy. This is still true action of constraints on the stiffness of a structure. The last theorem may
if the body is subjected to temperature loading.
The point is of sufficient interest to warrant some elaboration. First it also be expressed as follows: the strain energy of an elastic body under a
may be helpful to point out that when we state that at the position of gil•en set of forces is a maxim!lm when it is subjected to the least number
of constraints that do no work.
equilibrium the total potential energy has a stationary value and that this The immediate application of the above considerations is, of course to
is a minimum we do not compare physically possible adjoining states. the effect of actual constraints on elastic structures as illustrated in 'the
For, in stating that the potential energy has a stationary value, i.e. SU=O,
e~amples m~ntioned. A more important application appears in connexion
we compare the true position u, with a position u+Su assuming in both wtth approxtmate analyses of deformations. Thus, if we reduce the free-
cases that forces and stresses are the same. This can obviously not be true dom of deformation as we do in the Rayleigh-Ritz and related methods we
for the second position since for given forces there is a unique position of always over-estimate the stiffness of tho structure. Hence for a given set
equilibrium. In fact, we mentioned that this arises due to our legitimate of forces (displacements) we under-estimate (over-estimate) the corre-
neglect of the higher order terms in Sa and oP. Also, when we go a step sponding displacements and strain energy (forces and strain energy).
farther and state that the stationary value is a minimum we prove this by _The above theorems on the effects of constraints on strain energy and
considering the influence of terms like !Sa,.,· S~:xx in U, arising from the sttff~ess appear to have been given first by Rayleigh* in 1875 for linearly
variation Sa.,., associated with S~:"'"'' but we still keep the forces constant- elasttc bodtes and no temperature effects. Our arguments indicate how-
although this cannot, in general, be true. ever, that they apply also to elastic bodies with non-linear stress~strain
Having pointed out these aspects of the virtual displacements approach relationship and under thermal loading. The original principles are
we shall, in what follows, discuss the question of the extremum of U1 from occasionally referred to as the static analogues of Bertrand's and Kelvin's
a more physical point of view. Again we prescribe certain displacements theorems in dynamics.
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 (0) The Unit Displacement method. This method will be developed
its natural position of equilibrium from which we can deduce the value in Section 8.
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 .<E) Approximate ~ethods of displacement analysi~, using the Rayleigh-
we must apply certain additional body and surface forces to push the Rttzt proce~ure. In tht~ method we_assume_for_the dtsplacements approxi-
system away from its natural configuration. The work done by these mate functtons or senes of functtons sattsfymg the geometric but not
constraint forces (!'~.;oP· ou, obviously positive), produces by reason of necessarily the static boundary conditions. For example, in a three-
equilibrium in the new position an equal increase in the strain energy dimensional elastic continuum we may express the total displacements
stored. Thus, the strain energy in any neighbouring compatible configura- u, v, w in a finite series as follows:
tion is greater than that for the unconstrained original position and hence
the strain energy there is a minimum. n
u= Uo(X, y, z)+La,u,(x, y, z)
I
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
r~l
I
I
n
the introduction (prior to the imposition of the displacements) of internal · V=' V0 (x,y,z)+Lb,vr(x,y,z) ...................... (50)
or ext~mal constraints that do no work. For example, in a shell or plate r ''I
analysis, we may assume that the middle surface is inextensible and the n
transverse shear strains are zero; thus, in this case we impose infinite W='W (X,y, z)+Lc,w,(x,y, z)
0
values for E and G in the middle surface and an infinite value for G in the r --I
where U0 , V00 w, satisfy the kinematic conditions where these are prescribed
*See A. Castigliano, Theorie de /'equilibre des systemes elastiqu~s, Turin 1879. and u,, v, w, are linearly independent functions which vanish there.
a,, b,, c, are unknown constants to be determined by the Rayleigh-Ritz
_1I procedure. The elastic strains corresponding to (50) are (see Eqs. (I) and (2))
au av aw
I E,,=sx -YJ, Eyy=a-y -YJ, E,,=sz -YJ
(51)
au av av aw aw au
0 E,"=i\Y+ox' E",=oz +s/ 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,
Fig. 9.-Stiffening and relaxation VoL 28. p. 737, 1909; see also J. reme u. angew. Math .• Vol. 135, p. 1. and Gesammelte Werke,
of stress-strain curve Parts, 1911, p. 192.
8
obtained by variation of a" b., c, while the other coefficients and the tem- where the Su's are virtual, i.e. kinematically possible infinitesimal displace-
perature are kept constant are hence virtual elastic displacements. Note ments. Eqs. (57) lead to the principle of virtual work Eq. (44) if the a's
again that the chosen u, v, w functions need n'ot satisfy the given static satisfy not only the internal but also the boundary equilibrium conditions
boundary conditions. Naturally, the accuracy of the analysis is enhanced Eqs. (5).
if the latter are also satisfied. Assume now that the displacements u, v, w are written in the approxi-
To determine the coefficients a., b., c, we use the condition of the mate form of Eqs. (50) where the a, b" c, are unknown coe:ficients. How-
stationary value of the total potential energy in the forms ever, contrary to what we assumed in paragraph (4), series (50) are taken
DU=O or SU;=DW 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-
There are also cases when the theorem of minimum strain-energy, U; =min., brium conditions where prescribed.
is useful, see for example Part H of this series example 4. The applica- Expressing now the stresses in the brackets of (57) in terms of the dis-
tion of DU=O ensures the average satisfaction of the equilibrium con- placements and temperature 0 and selecting as virtual elastic displace-
ditions. ments one of the 3n possibilities (52) we obtain the n equations in the 3n
If we evaluate U in terms of (50) then for a linearly elastic body we unknowns a" b, c,
obtain a quadratic function in a., b., c,. Condition DU=O which in this
case becomes 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 '' '''''' ''' ''' ''
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. It is easy to see that for displacement functions (50) satisfying all boundary
Whilst the Rayleigh-Ritz method can provide a good approximation conditions the Galerkin and Rayleigh-Ritz methods must yield the same
to the deformations, the accuracy of the associated stresses is, in general, equations for a, b" c, and hence also the same deformations. We need
not as good. This is obvious if we remember that the accuracy of any only realize that in this case Eqs. (57) are indeed equivalent to the principle
approximate function is decreased with every differentiation. of virtual work. Hence substitution of u, v, w in Eqs. (57) must give the
In two- or three-dimensional problems it is possible to improve upon same result as substitution into
the original Rayleigh-Ritz procedure by adopting a mixed technique of
(A) and this paragraph. Thus, in a two-dimensional problem it is often SU=O
possible to guess accurately the variation of the displacements parallel to The advantage of Galerkin's method lies in a more direct derivation of
one axis, say the x-axis, while it is much more difficult to make an in- the equations in a., b., c,. However, contrary to what is usually assumed,
telligent assumption about the variation parallel to the other direction. this advantage is small if we calculate SU directly. We note also that
We may then write the displacements u and v in the form Galerkin's method allows only such approximate functions as satisfy all
u=u 1(x)·cp(y) } boundary conditions, while the Rayleigh-Ritz procedure requires only the
V=V (x)·l{l(y) . . . . . . . . . . . . . . (56) satisfaction of the kinematic boundary conditions.
1
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)
--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
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
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
J
I
1
(cl3)
\.-,
X
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 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- p~inciples of virtual displacements and virtual forces. Thus, by substituting
I
TARY VIRTUAL WORK vtrtual.forces (stresses) for virtual elastic displacements (strains), actual
T is natural in reviewing the developments of Sections 3 and 4 to inquire total displacements (strains) for forces (stresses), and invariant state of
if it is possible to enlarge upon the conception of complementary work straining for. invari~nt _state of forces we obtain Eq. (62) from Eq. (44).
and strain energy in a similar way as accomplished for work and strain However, th1s duahty IS only complete for continuous structures which
energy by the introduction of virtual displacements. In fact, if we consider are infinitely redundant. If, on the other hand, we consider a statically
Eqs. (28) and (36) we realize immediately that the functions S W* and determinate structure we find that while it is still possible to describe an
S U; * are independent of the variations Su and DE associated with the force infinite set of virtual displacements O!t associated with a prescribed set of
and stress increments, just as S Wand OU; are independent of the varia- certain of the displacements, only one stress system can exist for given
tions oP and Sa associated with the chosen Su and DE'S. Hence, when external forces; hence no oa can be assigned in the latter case and the
finding oW* and oU;* we may assume that displacements and strains principle of virtual forces has no application. A more fundamental
remain constant. Also, the infinitesimal increments ,Sa, ow, ocp are arbi- limitation of the principle of virtual forces appears if we want to extend
trary as long as they satisfy the equilibrium conditions in the interior and, ?ur th~orems to finite displacements. Here we find that it is, in general,
where such are prescribed, on the surface. Thus, if we fix that the surface tmposs1ble to achieve this for the principle of virtual forces while as
forces are not to be varied over part of the boundary we must have there mentioned in Section 4, no basic difficulty arises in the case of the prin-
ocp=O; however, where kinematic conditions are prescribed on the ciple of virtual displacements. However, for the usual analyses of redundant
boundary the Scp variation cannot be assigned. It is apparent that our systems involving small displacement theory the principle of virtual forces
incremental stress system need not even be an elastically compatible one. with its many particular forms is t:1e most useful one since the standard
It is only restricted by the condition that it must be statically equivalent proce~ure !ntrodu~es forces as unknowns. Naturally, there are many cases,
to the load increments ow and Scp. While these increments are applied it is especially m multi-redundant structures, where it is advantageous to in-
assumed as in Section 4 that the temperature remains constant. troduce displacements as unknowns; here the principle of virtual displace-
Such infinitesimal variations of forces and stresses which are arbitrary ments is the indicated method as shown in Example (b) of Section 5.
as long as they satisfy the prescribed equilibrium conditions we call virtual . We return now to _Eq. (62) and will illustrate its validity on a very
forces and stresses. s1mple example. Constder to that effect the redundant beam of uniform
Before restating theorem (35) for the more general conceptions intro- flexural_ stiffness £/built-in at z =0, simply supported at z ~,I, and subjected
duced here let us consider again its derivation in the light of our new ideas. to a umform load p (see FIG. 14). Under the assumption that the ordinary
Thus, if we multiply the true displacements u, v, w by the internal equili- engineers' theory of bending holds and that the shear deflexions are
brium conditions (29) which the virtual stresses Sa and ow must satisfy negligible the deflexion v is given by
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
14.-Example of an arbitrary virtual force
8U,*~Jy8adV ........................................ (65a)
v
12
but 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
U,* U1* -t- frysdV f [ JEOa]dV f frJ.w/V (66) are not any longer compatible. It is important to realize that the oar
v v 0 v
systems are self-equilibrating since the external loads P remain constant.
since <1 =const. U,* we term the complementary potential energy of total Thus, in a framework we may obtain a system oa,. by cutting a redundant
deformation. Note that it is always simpler to calculate directly oU,* bar and applying a variation oN to the true force N in the bar.
from Eq. (65). Particular care is necessary in evaluating U,* for as Eq. (66) We now investigate the differences in complementary work W* and
shows in the first integral E is taken to vary with a from the initial to the U>~* between the original equilibrium position of the uncut body and the
final state while 7), s in the second integral refer only to the final values. new enforced equilibrium position of the cut body. Comparing Eqs. (38)
Physically speaking we may consider U>~* as the· complementary work and (66) we find
necessary to reach the final true state of deformation from an initial state 1
in which we allowed free thermal expansion and destroyed compatibility. W* = U.t*- f<TJs- fTJOs)dV (70)
Formulae (65) and (66) may be extended immediately to the case of v 0
arbitrary initial straining by substitu1ing In moving from the uncut (compatible) equilibrium state to the cut one we
7Ja -T),.,.a,.,.-t- ...... +7) .•.•,a_,., I .... I 7),...a,... for 7JS 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-
Eq. (62) can now be written more concisely dition for compatible equilibrium of the original body and no first order
o,U*~o.,(U,*tU,*J 0 .................................. (67) increment 8 W* can arise since the loads P remain constant. We are then
left only with second order increments.
where the suffix a indicates that only forces and stresses are varied and For the complementary work this is
U*=U.,*-t-U,* (67a) o 2 WL1-fou,·oac·dS ... ................................. (70a)
c
is defined as the total complementary potential energy of the system. (where the integral is taken over the cut faces) which is the work of the
Eq. (67) states that a state of equilibrium of an elastic body is also one of virtual stresses oa c over the displacements ou c they themselves produce.
elastic compatibility (i.e. the body is at the true position of equilibrium) This is clearly positive. The second order increment of U,* is merely
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 o 2 U,* =Hoa· oE·dV . . . . . . . . . . . . . . . . . . . . . .. . . . . . .. . . . . . . (70b)
v
complementary potential energy. This theorem we call the principle of a
stationary value of total complementary potential energy if the latter is since 7) remains constant; oa and OE are the stresses and strains due to oa ,.
expressed in terms of forces and stresses. Actually the stationary value of Terms (70a) and (70b) are equal and both positive.
U* is a minimum as we may prove without difficulty.* This point is dis- We conclude that the complementary potential energy of total deforma-
cussed in more detail under (C) below. tion U,* and the complementary work W* have for given forces and tem-
Eqs. (62) or (67) may be used to derive the results which follow. perature distribution a minimum at that position of equilibrium of the uncut
(A). The differential equations of the theory of elasticity (for arbitrary
body at which compatibility is satisfied.
loading and temperature distribution), or any particular structural It follows that if U.t* is overestimated by assuming a statically equivalent
problem, in terms of stresses or stress-resultants; the appropriate stress system which does not satisfy all compatibility conditions and we,
kinematic conditions in terms of forces and stresses follow also from this ignoring the latter fact, equate U<~* to W* of the applied loads Palone
analysis. It is important to note that in all applications it is best to form we cannot but overestimate the magnitude of the displacement system
directly under the loads P. Conversely to achieve a given displacement system our
calculations based on a non-compatible stress system must underestimate
oU,*=f[yoa]dV the corresponding load system P. Thus, the latter has its maximum for the
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,*. expressed also as follows :
For given displacements and temperature distribution the complementary
potential energy of total deformation has a maximum when the state of
(B). Castigliano'st Theorem Part II generalized for Thermal Effects and equilibrium satisfies also the compatibility conditions.
non-Unear elasticity The above theorems may be combined to give:
The stiffness of an elastic body in which the equilibrium conditions are
[ (!Ud*]
(JP =u, ..................................... . (68) satisfied is a maximum when the elastic compatibility conditions are all met.
r B=const.
Thus we see that the effect of introducing assumed forms of stress dis-
where u, is the deflexion (rotation) in the direction of the force (moment) tribution for the purpose of approximate solutions is the opposite to that
P,. This relation may be obtained immediately if we apply one virtual of the method of Virtual Displacements and therefore application of both
external force oP, in the direction of the displacement u,. methods to a given problem yields upper and lower bounds to such
aggregate quantities as stiffness. No general conclusion as to bounds can,
(C). The Principle of Stationary (Minimum) value of Complementary of course, be drawn for the details of the stress distribution.
potential energy of total deformation for internally redundant The above theorems which apply also in the presence of initial strains
structures other than those due to temperature do not appear to have been given
before with this degree of generality.
This may be derived from (62) if we do not apply any external virtual
forces; i.e.
(D). The Unit Load Method
8c/>~=oc/>,=oc/>z=owx=ow 11 =~owz=O
Assume that we require the deformation (deflexion or slope) u, at a
while varying the stresses a. given point and direction of an elastic redundant body subjected to given
Then forces and thermal effects. Let the actual total strains in the structure be
8.,(U,*)=0 (69) known and given by*
which is our generalization of the standard principle of Castigliano of Yx~=Eu+7), Yx·y=Exy
Minimum Strain Energy to include temperature effects and non-linear Applying a load (force or moment), oP, in the direction of u, and using
stress-strain laws. Note again that Principle (69) necessarily applies only to Eq. (62) we find
internally redundant structures, since for given external loads only one
stress distribution can exist in statically determinate structures. oP,·u,=JyoadV
v
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- =HYxxOa..,..,+ ..... +Yx110a,.+ .. +yuoa.,]dV (71)
v
patibility conditions are satisfied. Note, as mentioned before, that with the
limitations of the present assumptions, i.e. small displacements and where Oaxx . ... oa~•. ... are the virtual stresses due to oP T• In a linearly
monotonically increasing stress-strain diagram, there is only one position elastic system Sax~ etc., are proportional to oPT and Eq. (69) can be written
possible where both the equilibrium and compatibility conditions are. l·u,= J[y,.p,x-1-YY·P•v+YzPzz+Yx•a,,+y,.a•• +yz,azr.]dV .... · . (71a)
satisfied. We now investigate the nature of the extremum of Ud*, which v
naturally requires the consideration of second order terms as in Section 4. where au . ... a,•.... are the stresses due to a unit load. Since ii,.,.. ...
Consider an elastic body under given loads and temperature distribution ax• . . . . need only satisfy the internal equilibrium conditions and the
in its compatible equilibrium position. We make a series of cuts in the external one for oP,= 1 it is obviously advantageous to determine
body but at the same time apply stresses a c acting across and along the ax., . ... iix·u . ... in the most simple statically determinate basic system.
cuts of the same magnitude as in the uncut body; these are obviously the For a non-linear system Eq. (7la) is still applicable as long as etc. a,,.
stresses required to maintain the compatibility condition of perfect fit at are calculated in a statically determinate basic system. For only in the
• See Biezeno and Grammel', p. 75. • Naturally. this method may .also b.> applied in the case of an initial strain system with shear strains.
t A. Castigliano, Theorie de /'equilibre des systjmes elastiques, Turin 1879.
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
(72)
J[ o2F, o2F, o2F,J
Y"'"oy2 +Y••ox2 -y""oxoy dA =0 . . . . . . . . . . . . . . . . . . . . . . (79a)
A
which satisfy automatically the internal equilibrium conditions (4). 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-
Eliminating the displacements from the strain expressions (I) we obtain knowns b1 to bn. These are only linear if the body follows a linear stress
the compatibility condition for the strains, -- strain law. In the latter case and for constant body forces Eq. (74) shows
o~+o~o~-0 (73)
that the solution must be independent of the Poisson's ratio v if all
oy 2 ox 2 - oxoy- ................................. . boundary conditions are of the static type. Hence we may take v=O and
(79a) becomes
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
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
unknown F, which will, in general, be non-linear. However, in the case A
of bodies obeying Hooke's law we obtain for a=const. the simple linear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (79b)
result 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
• .f. C. Maxwell, Phil. Mag., vol. 27, p. 294, 1864 for all surface conditions are then exactly satisfied by (76). This indicates
t 0. Mohr, Zeit. Ar<-hitek. u. /ng. Ver. Hannover, 1874, p. 509; 1875, p. 17. that it should be possible to express (79a) in a form similar to that given
t The presentation is restricted to singly connected domains. as Galerkin's method under the virtual displacement principle. In fact, if
§ See Timosbenko 8, p. 26. 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.
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)
: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
~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
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 placements, the stress-deformation analysis of engineering structures was,
T
NUMBER OF REDUNDANCIES 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
HE general theorems given in Sections 4 and 6 include, fro~ the for the analysis of redundant systems, when investigating problem (b) in
fundamental point of view, all that is required for the analysis of Section 5 used also the displacement method. The analysis of indetermin-
redundant structures. However, to facilitate practical calculations it ate structures on the basis of redundant forces goes back to Clerk Maxwell'
is helpful to develop more explicit methods and formulae. To find these and Otto Mohr 5 and was ultimately developed by Mueller-Breslau. 6 t This
is the purpose of this Section. technique is, as mentioned in the introduction, more concise and physically
A structure is by common definition redundant if there are not sufficient more illuminating than the Castigliano approach; it derives most naturally
conditions of equilibrium to obtain all internal forces (stresses or stress- from the unit load method.(see Section 60, Eq. (7la)). Mueller-Breslau's
resultants) and reactions; the number of redundancies is the difference technique is generalized here and presented also in matrix form. The effect
between the number of unknown forces (or stresses) and the number of of temperature or other initial strains is included ab initio.
independent equilibrium conditions. Strictly all actual structures are in- Parallel to the rapid development of the force-redundant theory occa-
finitely redundant but for practical purposes it is, in general,. ne~ss~ry sional practical problems were solved by selecting deformations as un-
and justified to simplify and idealize the structure and/or stress d1stnbut10n knowns. Fundamentally this method is equivalent to the virtual displace-
in order to obtain a system with a finite (or even zero) number of redun- ment analysis given in Section 4. Mohrt was probably the first to use such
dancies. Such typical processes of simplification are, for example, the an approach in engineering structures when finding the secondary bending
assumption of pin-joints in frameworks and the assumption of the engin- stresses in frameworks of the type usually assumed to be pin-jointed.
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 • 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,
body. Theorie der Elasticitaet [ester Koerper, Leipzig, 1862; French edition by B. de St. Venant: A. Clebsch,
All our considerations in this Section are restricted to linearly elastic 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.
bodies but Example 2 in Part II shows how the present methods may be
extended to the analysis of non-linear redundant structures. t See also H. Mueller-Breslau, Die graphische Statik der Baukonstruktionen, 1 ed., Koerner,
Leipzig, 1886.
It is curious to note that, while the solution of problems in the theory of
~ 0. Mohr, Zivilingenieur, Vol. 38, p. 577, 1892; see also, Abhandlungen aus dem Gebiete der
elasticity is derived very often from the differential equations in the dis- technischen Mechanik, 2 ed., Berlin, 1914, p. 407.
'· flexibility matrix of g element C; force or moment in the basic system due to X;= 1 acting on an
element which experiences a ~ in the direction of this ~
f flexibility matrix of all elements
N (n), S (s), M (m) normal force, shear force, bending moment
4> flexibility of element of unit length
0, o (rectangular) zero matrix
true and virtual strain due to unit displacement at i
stress due to unit displacement i unit matrix
ku, k 1h direct and cross-stiffnesses 1 .... j ..... h ..... m direction of external forces
K matrix of stiffnesses k;h 1 . . . . . i . . . . . k . . . . . n redundancies
A transformation matrix for displacements a, b, ..... g, ..... .s elements of structure
K., K., generalized and orthogonal stiffness matrices A', A- 1 transpose and inverted (reciprocal) matrix of A
a rectangular transformation matrix for strains { . . . . . } 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
where
R=8Q
L _I } ................ (97)*
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- is the so-called reciprocal or inverse matrix of G. Its determination is
ments and rotations in any linearly elastic body. To each displacement or equivalent to solving Eqs. (95) and therefore involves considerable numer-
rotation r; there corresponds a force or moment R;. Thus, in such a case ical labour if the number of equations is large. In such cases approximate
the matrices are 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
and computing the reciprocal matrix but hope to return in greater detail to this
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=
~,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)
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
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)
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
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)
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
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'
~
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
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)
~~~-
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 Fig. 28.-Singly redundant, pin-jointed framework. Contribution tu 010
from sinking or 'give' of support and excess lengths of bars
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 The 8;, are, of course, the influence or flexibility coefficients of the basic
may sometimes be identified with the process of an actual physical cut of structure for the directions of the redundant forces. We use here the symbol
redundant members (e.g. of bars in a redundant pin-jointed framework). 8 for these flexibilities since it is standard in the literature. To calculate the
However, the simple idea of a cut is not always applicable to continuous 8's we apply the unit load method of Section 6D.
structures typical of aircraft. We discuss this point later but for the sake Thus, using again the abbreviations
of linguistic simplicity continue to use the expression 'cut redundancy'.
Let the stress-system in the basic system be denoted by aE=auExx + ..... . +UzxEzx
} .................. (3)
a7] ~--au7]xx+ . . · . · · +azx7)zx
ao
we find from Eq. (71a),
It must obviously be in equilibrium with the applied loads. We describe
it as a 'statically equivalent stress system', thus drawing attention to the 0;,= fa;(E,+7])dV ............................. . (177)
fact that in its determination only statical conditions enter. We find also v
in the basic structure the stress systems 8;;'~Ja;E;dV, o,k=fa;EkdV= fakE;dV=8ki ................ (178)
a 1 ,a2 , •••••• ,a;, ...... all
v v v
due to where ai, E; (ako Ek) are the stresses and strains corresponding to X;= 1
XI= I, X2= I, ...... , X;·~ I, ...... , X,'= I (Xk=1) and a., Eo are the stresses and strains due to the applied loads.
respectively. The systems a I, a 2, ...... , a;, ...... , a" are obviously self- Eqs. ( 178) reproduce, of course, merely Eqs. (84) for the flexibility co-
equilibrating. Since our structure is by definition linearly elastic the true efficients. The total initial strains 7J imposed upon the basic system may be
stresses a in the uncut original structure can be expressed as separated into thermal and other strains
7J.•·x=a0 +T}xxo• • • • • • · , T}xu =7]xuo• • • • · • • · • • • • • • • · (179)
a a 0 + ~a;X; ........................ (176)
i=-1 where T}xx•• etc. are initial strains due to say lack of fit, 'give' at the supports.
Similar equations may be written down for stress resultants (forces or The effect of the latter upon 8; 0 is best considered separately and expressed
moments). Thus, the problem reduces to the determination of the X's, in terms of the imposed changes of length (rotations) and 'gives'. Consider,
which as already mentioned, need not be simple forces or moments but can for example, the singly redundant framework of FIG. (28) and assume that
be linear combinations of such (generalized forces). 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
The Equations in the unknown X. furthermore that the intermediate support gives or sinks by the amount /:1.
We define the following set of deformations in the basic system. As redundancy we select the force XI in the bar (1, 2) and denote by N_9
O;. Relative movement of ends of cut ith redundancy due to all
and NI the (tension) forces in the bars of the basic system due to the appliea
external causes, i.e. loads, temperature changes, lack of fit, 'give'
at the supports, etc.; i =I to n. loads and X1 = I respectively. The loading case X1 = l, with the correspond-
O;k=Oki relative movement of ends of cut ith (kth) redundancy due to the
ing force applied to support C by the structure is shown in FIG. (28).
self-equilibrating load system Xk = 1 (X;= I); i and k take values We find immediately
1 to n. The 8-coefficients are taken positive if the relative move- NI2f
ments are in the positive direction of the X's. 8u = ~ AE ............................ (180)
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
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
• 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,
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
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 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,
I 4
D=~~ I 4
a~~-~ Ya Y2-
I
II
L _j cosO -sin~ .............. (202)
and clearly a scalar multiple of the D matrix of (c). I 0 -sin 0 -cos()
The final system (d) is recognized as a particular case of the well-known L _j
Three-Moment Equation of Clapeyron. Since the basic system approxi- In practice it is nearly always worthwhile to find the elastic centre and
mates more closely to the actual system than that of (c) it is clearly the mosl eliminate two of the cross-flexibilities but determination of the principal
suitable choice of all. axes, unless possible by inspection, is not usually worth the trouble. A
The differences between the above four systems become even more further point is that the elastic centre concept is still valid if deformations
pronounced when the number of spans is increased. Moreover, the 3w due to normal and shear forces are included whereas the principal axes
coefficients of choices (a) and (b) tend, for a large number of spans, to requirement becomes more complicated.
become linearly dependent. Interestingly enough this solution was first given by Mohr* more than
This discussion shows how important the choice of the redundant seventy years ago, but it appears not to be universally known, for otherwise
forces is for the convenient numerical solution of a problem. An extreme it would not have been necessary to rediscover it so many times. A more
case of simplicity is achieved if in all equations only one unknown appears. recent derivation and application of an orthogonal set of redundancies
The particular 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
Y1 , Y 2 , Y3, - - - - , Y;,---- Yn torsion.
for which this condition is satisfied is called orthogonal. Then all but the Particular forms of the 3;k and 3; 0 coefficients
corresponding direct influence coefficients are zero, i.e.
'ik
=0 if ;,*' k . . . . . . . . . . . . . . . . . . . . . . . . . . (199)
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.
where we introduce the symbol ' for 3 to emphasize the special nature of
this system. Eqs. (182) take now the simple form, Stiff-jointed plane framework. We assume the Engineers' theory of
';;Y;+,;.=O for i=l ton ........................ (200) bending stresses to be true and introduce the special notation:
N., s., M. normal force, shear force, bending moment in basic system
and hence due to applied loads.
N;, S;, M; normal force, shear force, bending moment in basic system
Y;=-~ for i=l ton ............................
~ii
(200a) due to X;= 1 where i= 1 ton.
s coordinate along axis of beam.
This system Y; may always be obtained by a particular linear transforma- 0 temperature at neutral line of cross-section.
tion D.0/h temperature gradient across depth h of beam ; positive if giving
X=B,Y .............................. (201) rise to thermal bending strain of the same sign as that due to
However, the computations involved in determining B, are usually more positive bending moment M.
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- • 0. Mobr, Z. Architek u. lng. Ver. Hannover, Vol. 27, 1881, p. 143. See also the 1enenlization
tageous to find the orthogonal set of redundancies. This is particularly so 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
if physical and not mathematical considerations indicate how to find them. and shear forces are included in the latter analysis.
For example, this is so in arches and singly connected rings where, if 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
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-
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
Example for the b0 and~ matrices.
~10. (35a) shows a five times redundant structure, assumed symmetrical,
matrix formulation of the analysis. Consider a system consisting of s subJected to the loads R1 and Rt· Due to symmetry of loading and struc-
structural elements with a total number n of redundancies which may be ture the system is effectively three times redundant. For the basic system
forces (stresses), moments or any generalized forces. We select a basic we sc:Iect the statically determinate structure of FIG. (35b). The b0 and b1
system by 'cutting' a number r of redundancies where r<n. Thus, the matnces for half the structure including the central vertical member (11)
simple idea of a statically determinate basic system (r=n) is but a particu- are found easily as
lar case of our investigations.
>;-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).
--
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)
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 Longitudinal flanges
are dealing with a multi-web structure and ribs at not too great a distance.
• interrupted at junction
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 N=3
the rib stations if the spacing of the spars and the sheet thickness of the Fig. 39.-Geometry of typical bay for determination of number of
cover are large. Similarly, if the structure has few ribs we may have to redundancies
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 structural or loading discontinuities where the stress distribution is more
longitudinal or transverse reinforcement along the new grid lines. We call difficult to estimate and the Poisson's ratio effect more pronounced. Such
the surface enclosed between two adjoining grid lines in the z- and s- a method is given here at the end of this sub-section but at first w~ d~velop
direction a field, and denote by 'bay' a part of the wing structure which the theory under the assum;Jtion that the effective flang~s areas are known
lies between two cross-sections taken through adjoining grid lines running and that they are constant between two adjoining nodal points.* For the
in the s-direction (see FIG. 37). The assumption of a linear direct stress webs, when considering torque and lift loads, it is alwtys su n:ient to add
distribution along the edges of an orthogonal and flat field yields from I /6 of the web cross sectional area to the longitudin:ll and transv~rse
overall equilibrium conditions a parabolic shear flow distribution along flanges at the intersection of the spars and ribs with the cover.
the edges. Naturally, neither the linear direct stress nor the parabolic We summarize now the main assumptions underlying the idealized
shear flow variation are, in general, exact and violate the internal and structure selected for analysis. Thus, our system consists of an orthogonal
boundary compatibility conditions of the field. This is not serious as long or nearly orthogonal grid of spars and ribs with top and bottom covers.
as we keep the spacing of the grid lines reasonably close. Moreover, we Effective flanges of constant area betw~en adjoining nodal points and carry-
simplify further the problem by neglecting the quadratic and linear terms ing only direct stresses are assumed placed along the grid lines in longi-
in the shear flow and considering it to be constant within each field. We tudinal and transverse directions. For the tim;: being w~ assum;: that the
note that for non-orthogonal grid lines (tapered structure) the uniform flange areas are known. The direct stresses and hence also the flange loads
shear flow offends against the equilibrium conditions even if the direct stress are taken to vary linearly between nodal points. All sheet material for
is constant between adjoining nodal points. The errors introduced by the covers, spars and webs is assigned a purely shear carrying role and a con-
assumption of uniform shear flow are, however, practically insignificant stant thickness within each field. The angles of taper of spars in plan view
for the reometry of structure considered here when the nodal point dis- and elevation are assumed to be small. The shear deformability of covers,
tances are small. ribs and spars is included ab initio in the analysis. For the stresses and loads
If the direct stresses along the grid lines were known we could calculate in the various elem;:nts we adopt the si~n convention illu;trated in FIG. 38.
the fraction of sheet area to be added to the reinforcements to form the Naturally, the idetliZltions and sim;Jiifi:ations introduced are strictly only
equivalent or effective flanges. This applies to the cover, spar-webs and necessary for the calculation of the redundtncies. The btsic or statically
rib-webs and yields an idealized structure in which the fields are only shear equivalent stress system may and should preferably be d~termined in the
carrying and the direct stress carrying ability is concentrated in flanges; (cut) actual structure.
an assumption widely used in aircraft practice. Neglecting the Poisson's
ratio effect and assuming the same material for flanges and sheet material Degree of redundancy of idealized structure
cover, the fraction of sheet cross-sectional area to be added to the flanges We proceed next to the enum;:ration of the redundancies in our idealized
varies between 1/6 and 1/2 if the fields are flat; the former value applies structure. In addition to the sim;>lifications introduc~d previously we
when the field is in pure bending in its own plane and the latter when it is ignore here the bending stiffness of the flanges for displac~m~nts tang~ntial
under uniform stress. Since the stress distribution is unknown we can at to the wing surface. This is, no doubt, su.nciently accurate for th~ present
best only estimate the effective areas of the flanges but may use an iteration exploratory analysis. The wing structure supported at the root and free at
process if the first guess proves inadequate. However, the latter procedure
is really clumsy and lengthy and a direct method, obviating the guessing • Strictly, the latter assumgtion is only necessary when finding the Oexibilities of the Dances for the
of flange areas would evidently be useful, in particular at the root or other 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) Longitudinal flange loads Transverse flange toads
and from (247a) the total number of redundancies is Fig. 41.-Rectangular stiffened panel. Unit self-equilibrating stress system
n=a(3N-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (249) of type X=l
which formula again assumes that all the flanges are held at the root.
Of considerable importance in modern aircraft structural practice are figure. All information as to flange loads and shear flows is given there.
the multispar syst~ms with .few, often only two, end ribs. A typical wing The corresponding equations (182) or (224) for the unknown X; are easily
of the latter type IS shown m FIG. (40). To analyse this structure we sub- seen to be reasonably well conditioned. Naturally, we can further improve
divide it into a number of bays whose length should not exceed say five the conditioning by introducing group loads
times the spar pitch. Effective flanges will by virtue of our idealization X=BX . . . . . . . . . . . . . . . . . . . . . . . . . . . . (252)
process be acting at the junction of these bays although no ribs are pro-
wh~re B is a suitable square matrix. We do not enter at this stage into the
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 chmce of B but hope to discuss these points in Part III. When the panel is
length of the wing and all flanges are held at the root, is given by symmetrical about its middle line it is preferable to combine the %-systems
into symmetrical and antisymmetrical groups.
n=(N-l)+l+a({3-4)=N+a({3-4) . . . . . . . . . . . . . . (250) In a wing structure of the type investigated previously we can describe
The last system to be considered is a flat panel which is of special import- three simple types of self-equilibrating internal systems. They are shown in
ance for diffusion investigations (see FIG. 41). It is assumed built-in at z=O FIGS. 42, 43 and 44 and denoted by
or held with prescribed displacements and free at the other three edges. X=l, Y=1, Z=l
Here the number n of redundancies when there are no unstiffened cut-outs respectively. The first is the generalization of the %-system used in the flat
is simply
II=~ (/3-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (251)
panel and the second and third may be considered as slightly modified
bays four boom load systems taken in the longitudinal and transverse directions
where f3 is defined as in the case of the wing. When M fields are removed respectively. The longitudinal four-boom load systems are applied ex-
without being replaced by stiff-jointed frames the number of redundancies tensively in standard wing analysis.* The three figures are self-explanatory
reduces by M. and give all flange loads and shear flows associated with the unit systems.
The next step in our investigations is the discussion of suitable self- Note, however, that the effect of taper is neglected except that we intro-
equilibrating systems which may be chosen as redundancies. Consider duce the true local dimensions in the evaluation of the self-equilibrating
first the simple case of a rectangular flat panel shown in FIG 41. For the • See J. H. Argyris and P. C. Dunne, 'The General Theory, etc.,' J.R.Ae.S., Vol. LI February,
redundancies we may select n systems of the type X= 1 illustrated in the September, November 1947 •
38
Multi-web wing
Area •·•
1 I + I
q =•-·...:L-..:.1
11 / 2 h, + h 2
q =---1-
Central cross section 1, (h,•h,)
~e
1 ' 1
1
c h a
/
D
Shear field -- / / ' f 2
1 1 ~-Longitudinal flange
//
Multi- web wing
d k b
' ' 1
'
g
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
and hence using the second equation of (261)
·0· .......... ·0 X= -D-1 b1 'fb0 R-D-1 ~,.'H ................... . (262)
0· 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
l·.............. .......
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. s~ I~ (26SJ
We have now all the information to form the matrices L _j
D=b/fb1 and D 0 =b1 'fb0 R where 5. are the true stresses (forces) in the original structure. Applying
and can hence solve the system of equations Eqs. (261), (264) and (265) in (263) we find
b1'fb1{X Y Z}+b1'fb0 R=O . . . . . . . . . . . . . . . . . . . . (259) I I I I I I
for the unknowns X, Y, Z.
Finally, we find the true flange loads and shear flows of the idealized
structure from
I o I Ibh II
5.
bv,
I
= b. R- I1 bl• o-1 b1,.'H . . . . . . . . . . . . (256)
I
L.J L.J L _j
5=b0 R+b1{X Y Z} . . . . . . . . . . . . . . . . . . . . . . (260) Hence
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 or
directions of the loads R may be determined from Eq. (229). H = (blhD- 1b 1,.')- 1 b,.R . . . . . . . . . . . . . . . . . . . . (267)
When the number of equations (259) is too large to be dealt with by our The true stresses in our actual structure are thus
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 5.= [b.- b 1 .D-1 b1 ,.'(bthD- 1 bt~o')- 1 b,.]R . . . . . . . . . . . . (268)
basic system. which solves our problem completely. As mentioned already the method is
If initial strains H are imposed on the structure in addition to the loads ideally suited for finding the alteration of the stresses in a structu!e through
R we have merely to add the column matrix a subsequent introduction of cut-outs, such as access doors which usually
b1 'H 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~
on the left hand side of (259). Thus the analysis includes inter alia the com- lages with window-openings. Naturally, the degree of redundancy IS
plete calculation of wings under thermal loading. increased by the 'filling-in' of the cut-outs but this is of no importance
for the automatic computations envisaged here.
A new approach to the problem of cut-outs
We emphasize that our above analysis is valid in the presence of any A more refined wing stress analysis
kind of cut-out stiffened or unstiffened by closed frames as long as the The above general method of wing stress analysis suffers from the serious
overall geometry and idealization conforms with the initial assumptions. defect mentioned initially that the effective flange areas have first to be
Nevertheless, when we have a structure which is essentially continuous guessed since the stress distributi~n on. which t~ey depend_is. unknown.
with only minor unstiffened cut-outs it may be worthwhile to apply an It is certainly feasible to apply an IteratiOn techmque but this IS not only
artifice which avoids the lack of uniformity in the pattern of the equations necessarily lengthy but als~ rather uninspiring. . . .
inevitably associated with cut-outs. Moreover, it is the idt>.al method of To obviate these difficulties we develop a method which ehmmates the
finding the alteration in the stresses due to the subsequent introduction of determination of the effective flange areas and works directly with effective
cut-outs in our system without having to repeat all the computations ab flange loads. The method has the further virtue that it takes full account
initio. of the Poisson's ratio effect which may be important at the root and at other
The method is as follows. To preserve the pattern of equations disturbed structural and loading discontinuities. The addition of 1/6 of the web
by missing shear panels or flanges we eliminate the cut-outs by introducing area to the flanges is always sufficiently accurate for lift and torque loads
fictitious shear panels or flanges with an arbitrary thickness or area. and is retained here. Hence, our problem is restricted to the wing surface
Naturally, it is usually preferable to select for the latter dimensions those alone. . .
of the surrounding structure. To obtain nevertheless the same flange loads The previously introduced assumptions that the loads are earned m the
and shear flows in our altered structure as in the original system initial idealized structure by a grid system of effective flange loads and fields
strains are imposed on the additional elements of such a magnitude that purely shear-carrying form also the basis of the new method. We assume
their stresses become zero. The effect of the fictitious elements is thus nulli- also that both the direct stress distribution and the effective flange loads
fied whilst the uniform pattern of our equations is retained. vary linearly between consecutive nodal points. However, our analysis
Let the column matrix of the unknown initial strains, in the additional does not presume that the so-called effective flange areas-whiC:h do not
elements only, be enter into our developments-are constant between nodal pomts. ~he
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 the linearly varying direct stresses across a grid line by efftJ7tive flange
matrix f and the matrices b0 and b1 which we write in the partitioned form 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
~~boabo,. ~~ ~~bl•bl,. ~~
in wing structures plate thicknesses a~d pos~ibly flange ar~s may vary
bo= bl = .............. (261) 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
L _j L _j
interrupted the corresponding values should be taken as zl?ro. These
where the sub-matrices with the suffices g and h refer to the forces in the simplifications are not necessary for the purpose of the analys~s but ease
elements of the original structure and the fictitious new elements respec- the problem of notation; moreover they do no! affect s~nously the
tively. accuracy of the computations. Contrary to our previous practice of n~;J:m
Denoting the column matrix {X Y Z} simply by X and taking the bering the flange elements with letters we numb:Jr here only the nodal pomts
initial strains in the original structure as zero the Eqs. (259) in the un- with numerals.
known X become, We detive now the equation connecting the effective flange loads at
II nodal points in the z and s directions of the idealized structure a!ld the
~'fb1 X+ b1'fb0 R+ b1' ~ = 0 I I 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-
L_j 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)
+ L9.3
~~a,a'l + L9.1f> I
i O'ztr.ll
Actual Structure aR3 a.~ 15
and Stresses L _J L _j
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)
L
0 0 lo
fr
_j
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 where the flexibilities of the cover, webs and ribs are as before.
apply a self-equilibrating unit load system X;=l (which may be also a We find for the matrices D and D 0 •
Y- or Z-system) and denote by P 13 and P 19 its longitudinal effective flange D=b1'fb1 and D 0 =b1'fb0 R . . . . . . . . . . . . . . . . . . . . (225b)
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 where b 1 and b 0 are arranged as in Eqs. (280).
Using Eqs. (225b) in (259) we solve the problem completely.
points and directions of X 1 = I due to some given flange strains E. and E •• The refinement introduced by the L matrix need not of course extend
The contribution to S; of the straining of the flanges (3, 9) and (8, 9) is then over the complete wing but may be restricted to the root and other marked
- 9 9 changes of structure and loading. For the rest of the structure it may still
31 = .......... + f P;E,dz+ f Q1E.ds+ be sufficient to estimate the effective flange areas and to use the simple
3 8 form (277a).
and since we assume that both flange loads and strains vary linearly between
nodal points we find, 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.
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
44
TABLE II (continued)
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)
Cf Eqs. (121), (122), (125), (126) C.f Eqs. (153), (154), (159), (160)
R =1 R=1
t ' f 1
r:~--,r~~
I
~·-~~.1--..-~tl----,--~-..--b-..
-~
)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}
"-~~--:;-
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 Problem b
Given a set of displacements r find forces R, stresses 5 and strains v Given a set of forces R find joint displacements r, strains v and stresses 5
From Eqs. (300c) From Eqs. (300d)
(314a) . . . . . . . . . . . . . . . . . . . . . . r=FR R=Kr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (314b)
Hence Hence
(315a) .. .. .. .. .. .. .. .. . .. .. . R=F-1 r r=K-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (315b)
(316a) . . . . . . . . . . . . . . . . . . . . . . 5=bR=bF-tr v=ar=aK-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (316b)
(317a) . . . . . . . . . . . . . . . . . . . . . . v=f5=fbF-1 r 5=kv=kaK-1 R................................ (317b)
Once F is known the question of statical determinacy or indeterminacy Once K is known the question of kinem:ttical d~terminacy or indeter-
is irrelevent in this problem. 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 in which W is known in terms of rand U. We set now the modified prob-
problem (a): Given the set of forces R determine the set of forces X lem (a): Given the set of displacements r determine the set of displace-
necessary to satisfy the compatibility conditions. ments U necessary to satisfy the equilibrium conditions.
Here in the basic system obtained by putting X= 0 the stresses Here in the basic system obtained by putting U = 0 the strains
(303a) . . . . . . . . . . . . . . . . . . . . . . S0 =b0 R v0 =aor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (303b)
are completely known although the system is statically indeterminate. are completely known although the system is kinematically indeterminate.
Similarly we know the stresses Similarly we know the strains
(304a) . ... . . . .. . . ... . . . . . . . . 5 1 =b1 X vl =ax u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (304b)
when R=O when r=O
True stresses in actual structure, True strains in actual structure,
(305a) . . . . . . . . . . . . . . . . . . . . . . 5=50 +51 =b0 R+b1 X v=v0 +v1 =aor+a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (305a)
strains in elements forces on elements
(306a) . . . . . . . . . . . . . . . . . . . . . . V=f5=fb 0 R+fb1 X S=kv=kaor+kaxU . . . . . . . . . . . . . . . . . . . . . . . . . . (306a)
Compatibility condition in actual system at points of application of X Equilibrium condition in actual syst~:n at diJ;:>hc::m::nts U
(324a) . . . . . . . . . . . . . . . . . . . . . . b1 'v=b1 'fb0 R+b1 'fb1 X=O a1 'S=a1 'kaor+i/kaxU=O . . . . . . . . . . . . . . . . . . (324b)
where b1 is a set of stresses statically equivalent to unit X's (and R= 0) where a1 is a set of strains kinematically equivalent to unit U's (and r= 0)
preferably found for Z = 0. In the latter case the rows of b1 are the same preferably found for W = 0. In the latter case the rows of i.1 are the same
as the corresponding rows of b1 of Problem (a). as the corresponding rows of a 1 of Problem (a).
Thus, Thus,
(325a) . . . . . . . . . . . . . . . . . . . . . . X=- (b1 'fb1 )-1 b1 'fb.,R u ~ -(a1 'ka1)-1a 1 'kaor . . . . . . . . . . . . . . . . . . . . . . . . (325b)
True stresses and strains True strains and stresses
(296a) S=bR, v=fbR v=ar , S=kar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
where where
(326a) . . . . . . . . . . . . . . . . . . . . . . b = b 0 - b1 (b1 'fb1 )- 1 b1 'fb0 a=ao-al(al'kax)-lil'kao ...................... (326b)
Displacements r due to R (see Eq. (300a)) Forces R due to r (see Eq. (300b))
(327a) . . . . . . . . . . . . . . . . . . . . . . r =b0 'v = FR R=a0 'S=Kr (327b)
where where
(328a) (328b)
and and
(329a) Ko=ao'kao .. . . .. .. .. . .. . . . . . . . .. . . .. . .. .. . .. . . (329b)
is the flexibility of the basic system. is the stiffness of the basic syst;:m.
The matrix b0 is a set of stresses statically equivalent to unit R's pre- a.l
The matrix is a set of strains kinematically equivalent to unit r's pre-
ferably found for Z = 0 and X= 0. In the latter case 60 is identical with ferably found for W = 0 and U = 0. In the latter case lio is identical with
b0 of problem (a). ao of problem (a).
Cf Eq. (175)
47
TABLE II (continued)
Elimination and rigidification of structural elements Rigidification and elimination of structural elements
Assume a set of initial strains, written as column matrix H, in the struc- Assume a set of initial stresses, written as column matrix J, in the struc-
tural elements to be removed, of such magnitude as to give zero stress in tural elements to be made infinitely rigid, of such magnitude as to give
resultant system. zero strain in resultant system .
Write the b and b 1 matrices of the complete structure in the partitioned Write the a and a 1 matrices of the complete structure in the partitioned
form 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)
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
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
simple when direct and shear deformations are ignored. In fact, for all frame- method in the same grid system.
works the determination of the matrices C and C 0 is trivial once we con- We denote for the purpose of the analysis of the unit panel the dis-
sider all possible degrees of freedom of the joints. See for example, the placements parallel to the s and z axes by u and w respectively and intro-
systems of FIGS. (23), (24) and (48) investigated on pp. 23, and 45, duce also the local coordinate system ~. '· Consider now the state of strain
which show clearly how elementary the matrices a and stiffness k are when and stress arising from a unit displacement
we break up the structure into its simplest constituent components. We va=l ............................... . (341)
need not therefore concern ourselves any more here with frameworks, and
we turn our attention to the membrane type of system characteristic of 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, •
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 0 0 0 EC2 EC2
y-y k= .............. (347)
0 ...... 0 k,, 0 . ,0
0 0 0 0 0 0 ECn EC2
d d
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-
Fla. 50.-Rectanaular stiffened panel. Assembly of stiffness placements being defined at all nodal points formed by grid line inter-
matrix from unit panels 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.
Since the terms in the a matrix are either unity or zero their formation is The analysis of such a structure under vertical loads follows that given
particularly simple. Writing a in the partitioned form under Problems (a) and (b) in TABLE n. Thus, we designate the vertical
displacements as r and take the rotations as the redundant displacements
a={a,.a, ........ a, ........ a.} .................. (348) U.
The strains of the elements are here identified as the displacements of
a, is the sub-matrix of 8 rows and 50 columns relating the displace- the unit panel defined in FIG. 49 and can therefore be written (Eq. 305b).
ments defined for the unit panel g (FIG. 49) to the displacements as defined v=aor+a1U
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
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.*
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
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
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
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*
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
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
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
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
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
0+1200Gc/>~
-125Dxa+466D,. -100 D,s =
.JJx
+ 5400 + IOSOG'I'S y
M dxM aM DxM
~ 10.000 kg.
I -12·17 1·46 -20· 13
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
l:·
reader to the new ideas and show him their power and basic simplicity.
• Actually, we may consider this basic system as .also derived by a single cut from the given system . • The particular case of the four-Bange single-cell tube under a given loading has also been treated
Thus if at the cut Bange we - l y the correspondmg E.T.B . Bange load and allow there any out of by W. J. Goodey in 'Two-spar Wing Stress Analysis', AIRCRAFT ENGINEERING, Vol. XXI, No. :U7,
place' movements, the stress distribution in the other structural elements is obviously that of the E.T.B. 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
and Bredt-Batho theories. equations given here with essentilUiy the same basic idealizations but with tbe effect of taper induded.
t Joe. cit. p. 38. 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
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
" --~.
F"lan;. I oa<H
..,
q,,
OMI
J50 ' -
~
140
.,
-- -·-
1------
--r-
-- - -- -
-,
~-
---~~-=-.:.-=~-~-
:..... ...;
(
'._ _ _ _ _ _ _J
~ r-----i-----~-----4------1------+-
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
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
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
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,
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
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
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 The complete set of flange loads and shear flows due to the forces R is
b. and b1 contain only the loads for the top flange. This applies also for obtained now with the b. and b1 matrices already given and the solution
the flanges of spar 'a' and for the covers. (b43) as
Carrying out the multiplications and assembly indicated in Eq. (b35) 5=bR=[b.-b1(b 1'fb1) - 1 b1'fb.]R .................. (b44)
we obtain the complete matrices b1'fb. and b1'fb1 and hence, on inverting
b1'fb1o the column of redundancies Limitations of space prevent the presentation of the complete matrix,
Y = -(b/fb1) - 1 b1'fb.R ........................ (b43) but as typical results we show above the flange loads in the upper
flange of spar 'a' :
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 bza=(b 0 za -bua(bt'fb1)- 1bt'fb.J ................ (b44a)
be repeated here. The calculation of the reciprocal matrix (b 1'fb1) - 1 was and the shear flows in the web of spar a :
performed using the Jordan technique described on p. 28 and required b,a=(b,owa-btwibt'fb1)-lbl'fb 0 ] • • • • . • • • • • • • (b44b)
approximately 8 hours for completion-by an unpractised and compara-
tively unskilled computor. 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 The numerical result for the Y 8 matrix is printed in full below. It
values calculated there. Such differences as arise are due to the neglect should be emphasized that this matrix gives the complete solution for a
of the terms 8;.1 in the O;k analysis. (See footnote on p. 63). general set of forces R. It only remains to premultiply with bh to obtain
As a further illustration, the 11ange loads and shear flows given in TABLE the flange and shear flows. Thus, taking only a single force R3 =1,000 lb.
III due to a single load R3 =I ,000 lb. are easily obtained from the third we obtain from the third column the additional flange loads and shear
column of the final b matrix of Eq. (b44.) flows given in TABLE III due to the removal of the web w3. Adding these
These values are also depicted graphically in FIG. 69. to the values already given for the structure without the cut-out, we obtain
E.ff'ect of a Cut-Out
the total values for the structure with cut-out. FIG. 69 illustrates the results
graphically.
To show now the application of method of treating cut-outs developed This particular example of the application of the H matrix has been
on p. 41 we determine the effect of removing the web of spar 'a' in the third worked out rather more fully than is strictly necessary in practice in order
bay (i.e. element w3). This, of course, converts this bay into a four-boom to illustrate the way it works. In fact, the final result for the stresses in the
open tube which, if taken into account initially would have required the structure with cut-out is given directly by Eq. (268) on p. 41. However, the
introduction of the special extended Y system over the cut-out discussed amount of work involved is astonishingly small, consisting of a few simple
on p. 62 and destroyed the simplicity and regular pattern of both b 0 and matrix multiplications and a single inversion. In more general cases where
b1 matrices. By our artifice of imposing initial strains in the cut-out element several elements are removed at the same time the additional matrix to be
until the stress there is zero we eliminate all such special considerations. inverted is of order m x m where m is the number of stresses (loads or stress
Moreover, when the cut-out is introduced subsequent to the initial stress flows) which have to be nullified.
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 The Flexibility Matrix
result is obtained with remarkable speed and simplicity, the additional To conclude the present example we calculate now the complete flexi-
matrix to be inverted being of order 1 x 1. bility matrix of the structure for the points and directions of the forces
The shear flow in the element to be eliminated, i.e. b 11 , is given by the R 1 to R 12 •
third row bwa (Eq. (b44b)), (printed in full on p. 67) and the matrix b 111 In Eq. (312a) for the flexibility matrix, F
of the shear flow in this element due to unit redundancies is similarly the
third row of b 1 wa (Eq. (b40)), i.e., F=bo'fb.- bo'fb 1(b 1'fb 1) - 1 b 1'fb. . ................... (b50)
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) fb., b.'fb 1 and (b 1'fb 1) - 1 b1 'fb.
strain H in this element is (Eq. (236) or (262)) and therefore we only have to perform two further matrix multiplications
DY 8 +b 111 'H=O and one addition. The complete matrix obtained is given in full below,
Y n= -D- 1 b111 'H .................... (b46) Eq. (b50a).
and therefore the additional stress (shear flow) in the element due to H is The numerical values given there (as in all other numerical matrices in
this example) have been rounded off to fewer significant figures than were
b11,Y 8 =-b 111 D- 1 bu'H ................ (b47) used in the actual computation.
Adding this to the stress in the element already calculated, b,R,and equat- Although of course the F matrix is symmetrical, this characteristic was
ing the resultant to zero, we find, using the above b1 11 and the D- 1 already not used to shorten the computations and all values were calculated in-
calculated, the required initial strain dependently, the symmetry being then used to give a check on the numeri-
H =(b 11,D- 1 bu')- 1 b 11 R cal work. The same applies, naturally, to the computations of any other
192 2 symmetrical matrix, like D. The whole question of numerical accuracy and
3 · 801 X 10 8 b "R ............ · ........ · (b48) avoidance of computational errors is obviously one of extreme importance
in the practical application of these methods, and it is essential to have com-
Hence from Eq. (b46) the additional Y 8 system required to nullify the 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).
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
69
Su:·flange s1ngte cell tubw
Singly symm•tric~l s iJC·flange
stngle cell lubf'
Two-bay segment
i-sysltm
(equal •net opp05 iU• X~systems
on covvrs (2.)) •nd (4.1 l)
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~
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
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)
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)
Fig. 75.-Extended X system applied at top cover (b70)
one Z system extending over the affected two bays
the one Z and one X system in the affected bay one X system j
} (b67)
and the four Y systems acting at the adjoining rib stations 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
We have a net loss of six standard redundant systems and need hence five symmetry the X system of (b70) is replaced by an X system.
new systems. As such we choose (see FIGS. 74 and 75),
Finally, we consider the case of a cut-out in the cover (FIG. 77). We lose
one stretched Y system (as shown in FIG. (66)) here in the affected bay three of the systems
two standard Y systems applied to the four comer flanges at two Y systems at adjoining ribs }
the adjoining ribs (b68) (b7l)
one Zsystem
two stretched X systems at the top and bottom covers (as the X system of (b66) may be assumed applied to the opposite cover and is
shown in FIG. 75) hence not lost. Our choice of two new redundancies is
Alternatively we could select two stretched Y systems (two adjoining four one stretched Y system in the affected cell
flange tubes) but since these are linearly related to the two above Y systems } (b72)
applied at the comer flanges it is then necessary to drop one of the latter Y 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)
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
1. INTRODUCTION
W
The method of calculation follows closely Section 8 of Part I.
From the symmetry of structure and loading, the shear force in the
E follow up and amplify the general theorems and principles ring on the axis of symmetry is zero. Hence as the basic or statically
developed in Part I by presenting here a number of applications. determinate system we select the structure shown in FIG. Ia. As redun-
It is hoped that the problems considered may prove of more than dancies we choose the bending moment X 1 and normal force X 2 at the
academic interest. Three of them are concerned with stresses due to elastic centre,* which in this case is the centre of the ring.
thermal straining-a current fashion in the aircraft structural field-and Forces and moments in basic system due to X 1 = l and X 2 = l re-
were originally presented in an A.R.C. paper.* spectively :
The first deals with the stresses arising in a circular ring due to a peri-
}
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)
}
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)
(c)
73
a0
~eds= --,ds .......................................... (5)
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)
0·6
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: ('
}
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
('
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
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.
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
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)
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
(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)
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
80
1·0
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
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 - !....... . . . . . . ...........
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 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~)
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
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
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
!!._(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
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