Professional Documents
Culture Documents
Argyris
EN ER GY THEOREMS AN D
STR UC TU RA L ANALYSIS
A Generalised Discourse with Applications on
Energy Principles of Structural Analysis
Including the Effects of Temperature and
Non-Linear Stress-Strain Relations.
by
J. H. ARGYR IS, D.Sc. (Eng)
Professor of Aeronautical Structures,
University of London,
Imperial College of Science and TechnoiOKY
Co-auth or of Part II
First published by
Butterworth & Co. (Publishers) Ltd.
ISBN 978-1-4899-5852-5
ISBN 978-1-4899-5850-1 (eBook)
DOI 10.1007/978-1-4899-5850-1
J. H. Argyris 1960.
Originally published by Plenum Press in 1960.
Softcover reprint of the hardcover 1st edition 1960
PREFACE
1.
2.
3.
4.
5.
6.
1. INTRODUCTION
GENERAL REFERENCES
(I) Biezeno, C. B., and Gramme!, R. Technische Dynamik, 1st ed., Springer, Berlin, 1939.
(2) Engesser, F. Z. Architek u. lng. Verein Hannover, Vol. 35, pp. 733-774, 1899.
(3) Lord Rayleigh. Theory of Sound, 2nd ed., Vols. I and II, Macmillan, London, 1892 and 1896.
(4) Maxwell, J. C. Phil. Mag., Vol. 27, p. 294, 1864.
(5) Mohr, 0. Z. Arch. u. lng. Verein Hannover, 1874, p. 509, and 1875, p. 17.
(6) Mueller-Breslau, H. Die neueren Methoden der Festigkeitslehre und der Statik der Baukonstruktilmen, 1st ed., Korner, Leipzig, 1886.
19 ~7. Southwell, R. V. Introduction to the Theory of Elasticity, 2nd ed., Clarendon Press, Oxford,
19~8/
Timoshenko, S., and Goodier, J. N. Theory of Elasticity, 2nd ed., MacGraw-Hill, New York,
results ar~ identical_ with exist~ng _ideas _they clothed them in a language not
too fam1har to engmeers. Th1s d1scuss1on of past authors' work brings us
to a few points which are preferably stated now. In much of present day
structur~l analy~is there seems to be an unfortunate tendency to overemphasize certam methods of analysing redundant structures and to
negl~-;t more useful ideas readily available for many years. This refers
particularly to Castigliano's principles which are so often set out as the
basis of all considerations, not only in theory, but also in the actual
methods of calculation. This is, in our opinion, unfortunate, even though
all methods naturally lead to the same results if based on the same
assumptions. For example, if we select forces as redundancies then much
the best means of obtaining the basic equations for their determination is
the long established O;k method of Mueller-Breslau based on the Unit
L<?ad idea. We do not ne~d, _in fac~, even the concept of strain energy for
th1s purpose. All we reqmre 1s the 1dea of work and kinematics as used in
rigid-body_ mec~anics. From such ~deas we can write down immediately
our equatiOns m the unkl!owns Wlt~o.ut b~thering about strain energy.
Thes~ l!lethods h~ve been m use by CIVIl engmeers for the past sixty years
and 1t IS surely t1me that we accepted them in the aeronautical world as
standard analytical equipment. Actually, the basic principles go much
farther back than Mueller-Breslau and were, in fact, developed independently ~Y Ma:'we_11 4 and Mohr 5 nearly a hundred years ago. The first sys~ematlc app~1cat1_on of_the ~ik method to stressed skin structures was given
m the classical mvest1gat10ns of Ebner.* Regrettably enough this lucid
wor~ was occasionally referred to in the past as obscure, a lack of comprehensiOn, no doubt, at leas_t partly due to the too narrow understanding of
redundant structures ansmg from a concentration on Castigliano's
methods. However, the limitations of Castigliano's formulation of the
problem are being at last increasingly recognized in aeronautical circles
due to the demands of calculations for highly redundant systems. Naturally, most of ~he_ alternative methods suggested are really nothing more
than a transcnpt10n of the Mueller-Breslau and Ebner technique.
We start our investigation in Section 3 with a discussion on work and
complementary work in the presence of temperature effects and for nonlinear stress-strain laws. With this basic knowledge we then proceed to the
standard principle of virtual displacements or virtual work in Section 4.
This is very similar to the currently used principle in rigid-body mechanics.
Thus, we consider a state of equilibrium, apply virtual displacements to it
and develop hence the classical principle of virtual work which substitutes, of course, for the equations of equilibrium. Since virtual displacements are kinematically possible ones this theorem starts from the
assumption of inherent compatibility to find the necessary and sufficient
condition for equilibrium. It is, of course, well known that the theorem
applies also to large displacements but this aspect is ignored here. However, temperature effects and an arbitrary law of elasticity are considered
as _lo~g as the latter is '!lonotoni_cally incre.1sing. Having established this
prmc1ple we deduce easlly some Important theorems and applications.
See e.g. H. Ebner and H. Koeller. 'Zur Berechnung des Kraftverlaufes in versteiften Zylin<Jer
schalen.' Luftfahrtfnrschung, Vol. 14. No. 12, December 1937.
cp.,
~/_;_1____,..../
)::_wx
Wz
""'.,.. / L - - - - - - -
Body Forces
Stresses
z
Surface Forces
Stresses
a""' a , a..
direct stresses
axu=a.z, a =a , a.,=a,, shear stresses
vu
Yxx=;)X'
vv
Y=S)i'
y.,.=~ +~;,
g=y.,.,+y +y..
Y)xx YJ Y)zz
Y),., Y)., Y)zx
Exx=y,.,~Y)xx
Exu=y,.~Y),.
etc.
etc.
e =Exx+Euu+Ezz
dV=dxdydz
dS
a
0
E
G
vw
Yzz=vz
}As shown in
FIGS. I and 2
=~ +~;. y.,=~: +~
(I)
(2)
element of volume
element of surface
linear coefficient of thermal expansion (may vary
with 0)
rise of temperature
Young's modulus }
shear modulus
May vary with 0
Poisson's ratio
v
ay=axxYxx+a y +azzYzz+axu/'xu+a y +azx/'zx ................ (3)
The corresponding explicit expressions for UYJ and aE
are obtained by substituting the strains Y)xx etc. and
Exx etc. respectively for Yxx etc.
W
work of external forces
U,= ~ W+const. potential (energy) of external forces
strain energy (or potential energy of elastic deformation)
W*, U, *, Ui *
complementary work, complement:uy potential of external .forces and complemenhry potential energy of
elastic deformation
Ud*
complementuy pot-:::-~ti:ll energy of totli deformation
From a consideration of equilibrium on an ele:nent dV=dxdydz, illustrated for the x-direction in FIG. 3
OCI,,+oa., . Oazx w =O
OX
0Y -f' 0Z + X
(4)
where
or/;
or/; orf;
sx=ou, ~y~Sl =0
where
to ou.
. .
FIG.
nazz+ laxz+ma.z=cfz
Over part of the surface the boundary conditions may be expressed in
terms of stresses or forces (static boundary _condi~ions) ~nd over the r~
mainder in terms of displacements or strams (kmemattc or geometnc
boundary conditions). Naturally, the boundary conditions may be of
both types over the sat:I?-e part of the surf~ce: Consider, for example, the
tube shown in FIG. 4. It ts assumed fully bmlt mat the roo_t (z=O) and _free
at the tip (z=/). Ribs rigid in their own plane but freely flextble to defle_x~ons
out of their plane are assumed at z =0 and z =I. The boundary condttlons
are: at z=O, u=v=w=O, i.e. pure kinematic conditions; at z=l, a =O,
(~u)
=0
uX zd
v
s
denote integrations over a volume and surface respectively.
The formal mathematical proof of some of the basic theorems in this
paper is shortened by using Green's theor~m. * ~t and if! be two continuous functions and let also the first partial denvattves of(/> and the first
and second partial derivatives of rf; be also continuous. Green's theorem
states:
J[oo/
ox
y,, corresponding
or/;
or/;
81/J
cp=a,., ox=ov, oy=ou, oz =0
(5)
2)
fa,x+mavx+ nazx=cfx
( ov)
oy z~! =0
8y,,~c8~=~8u
Stmtlarly take
Then
where
Note that although Green's theorem is helpful for the mathematical understanding of the present theory it is not really necessary for the physical
understanding; a reader unfamiliar with these aspects of the integral
calculus may omit the relevant parts.
3. WORK AND COMPLEMENTARY WORK-STRAIN ENERGY
AND COMPLEMENTARY STRAIN ENERGY
The analysis of the present paper is restricted to small strains which can
be expressed by the linear formulae given in the notation. Such displacements and corresponding strains are obviously additive (algebraically).
Thus, if u 1 , y 1 , and u 2, y 2 are displacements and strains in a deformed
state 1 and 2 respectively, then u 1 + u 2, y 1 +y 2 represent also a compatible
state of deformation of the body. Our assumption does not impose,
however, a linear stress-strain relationship; hence if P 1 , a 1 and P 2 , a 2
are the forces and stresses corresponding to the above two states of
deformation of the body, the forces and stresses corresponding to the
deformed state u 1 +u 2 are not P1 +P2 and a 1 +a 2 except in the case of a
linearly elastic body. In all cases, however, the stress-strain law is assumed
to increase monotonically as shown in FIG. 5. In conclusion we can state
that the law of superposition is assumed to hold for strains and displacements but not necessarily for the stresses.
In general, we assume also that the displacements are so small that the
equilibrium conditions can be written down for the undeformed body. It
follows then that the question of stability or instability of equilibrium does
not enter in the analysis of this paper and there is a unique solution to
every problem.
Consider a three-dimensional deformable body (not necessarily elastic)
in equilibrium subjected to a self-equilibrating system of body forces Wx
etc., surface forces cfx etc. and a temperature 0. These forces and temperature may vary with time but the variations are assumed so slow that
the dynamic effects are negligible. Let, in a time interval ot, the forces
increase by Owx, Ocfx etc. and the temperature by 80. The displacements
increase at the same time by ou etc. There arises hence an increment of
work (see FIG. 6)
See Courant, Differential and Integral Calculus, translated by J. E. McShane, Blackie and Son
Ltd., London and Glasgow, 1949, Vol. II.
+sHo/xou+cp.ov+cp.ow]dS
+ terms of higher order.
Y,V
Rigid rib
Fig. 4.-Kinematic and static boundary conditions
E
Fig. 5.-Stress-strain diagram
Fig. 6.-Work and complementary work; strain energy and complementary strain energy
and that it is subjected both to external loads and thermal effects. In view
of our initial assumption about the smallness of the strains we can write
0Yxx=Oxx+01]
Oyy
lhdy
_EY________________ ,
(12)
Oxx
0Yzz=Ozz+01]
where OExx .... , 811 are the increments of the true elastic strains and the
thermal strains
e
(13)
'Y/xx=1]yy=1]zz=1] = faoe
I~
I
e.
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.
J ...................... (
8W=fwx8u+w.8v+w.8w]dV
+~[c/Jx8u+c/J.8v+c/Jz8w]dS
7)
Note that Eq. (7) does not presume any specific force-displacement law,
be it elastic or non-elastic.
It is simple to derive an alternative expression by considering the
additive effect of the work done by the stress resultants on each volume
element dV. A perusal of FIG. (7) shows that the deformation Su, Sv, ow
gives rise to an increment of work for an element dV
(axxDYxx+a Sy +a Sy +axuDYxu+a Sy +a.xSy.,)dV=aoydV
again neglecting terms of higher order.
The incremental (infinitesimal) strains Sy,x etc. are those due to the displacements Su etc. Thus,
8Yxx =o~!!
etc ., 8Yx =o{~ ~}
~X
~Y +~X
={~Su
~Sv}
~Y +~X
(8)
'' '
~Y
~z
~Y
~z
JSw} dV
~x
where the Sy's satisfy Eqs. (8) and are hence compatible strains. Note that
where c/J, is unknown but the displacement (say u) fixed, the corresponding
Su is zero and hence the relevant terms in the last relation vanish. Integrating Eq. (9) from the initial unstressed state 0 to a final state I we
find the total work
I
s
v
..................................
where
D;=fao
(17)
and
aOE=axxDExx+ ........ +axuDExu+ ... +azxDEzx=OU; . . . . . . (l7a)
Note that the second term in (16) and thus also the work depend on the
sequence of application of loads and temperature.
We assume now that the elastic properties of the material depend only
on the instantaneous temperature
but not on the previous history of
deformation, e.g. if adiabatic or isothermal; the error involved in this
assumption is indeed negligible. fhus, the expression U; is only a single
valued function of the instantaneous state of elastic straining. In a closed
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
names are strain energy function or density of strain energy.
It follows from Eq. (17a) that
""
oUi=axxDEu+
Hence
~Di
"
~Di"
+azxDzx=~Oxx+
UE'xx
~Di"
+~Ozx
UE'zx
~Di
(18)
Uxx=()Exx , ., Uxy=(>E'xy
We conclude that the stress-strain law is uniquely determined by the strainenergy function and vice versa. Note that the law of elasticity is arbitrary
in Eqs. (12) to (18).
Integrating Eq. (16) over the body we find,
e
fWdV= fUidV+ f [ fso'Y/]dV
v
or
(~
W=U;+J [ fs01]]dV
(19)
v 0
where
(10)
where
-
(16)
=0
W=D;+js811
(I I)
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
of Eq. (10).
In what follows we assume that the body is fully elastic and isotropict
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.
t The isotropy need only be assumed at each point; the properties of the body may vary from
point to point.
(22)
(23)
axx=2G[Exx+ 1 ~ 2 ve]=2G[rxx+ 1
and two similar expressions for a .. and a zz The moduli E and G and the
ratio v may vary with e.
Substitution of the stresses given by Eqs. (22) and (23) into Eq. (15)
yields
8w=8D;+f~8iv........................................
(24)
EeT}
W=U;+2(l-2v)........................................
(25)
where
{ I _-vVe 2 -2[E,,Eyy+EwEzz+E"zzE",,- -;r(E"xy
1
2
2
2 }
U;=G
+~: +E"zx )]
1 2
(26)
oO;
OUxx =E",,=y,,-TJ
oO;
.. .... , oa,. =,......
(27)
.....
(26a)
Parallel to the conceptions of work and strain energy two further ideas
essentially due to Engesser 2 , are of particular importance to our investigations. Consider to that effect a one-dimensional force-displacement and a
stress-elastic strain diagram (FIG. 6). The vertically shaded areas are
obviously those of work and strain energy respectively. lt is natural to
inquire if the horizontally shaded areas complementing the previous areas
in the rectangular areas Pu and a~: respectively are of any importance (see
FIG. 6). In fact, as is shown farther on, the introduction of these new conceptions is proved a particularly happy one when generalizing some
theorems, currently assumed to be valid only for linear elasticity, to bodies
with non-linear stress-strain relations. Although the complementary areas
are equal for linear elasticity it is still useful in such cases to differentiate
between them.
It is interesting to note that in thermodynamics two similar complementary functions are used : the free energy function H of von Helmholtz
and the function G of Gibbs.* In what follows we call the horizontally
shaded areas complementary work and complementary strain energy and
denote them by w and U;* respectively.
We generalize next our new conceptions by considering the threedimensional case. Let the actual displacements in a body subjected to body
forces w, surface forces cp and temperature 0 be u, v, w. The increment of
i.he complementary work as these displacements increase to u + 8u,
v+8v, w+ow due to load increments 81/J, .... , 8w, . .... etc. is given by
S W* = f[uSw, + v8w. + wSw.]dV + f[uoc/J, + voc/J. + wSI/J.JdS
v
s
+terms of higher order
or
(28)
........................
(29)
and
loa,,+moa.,+nou.,=oc/J, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (30)
Multiplying now each of the first set of equations by the displacements
u, v, w respectively, summing and integrating over the body we obtain by
applying Green's Theorem similarly to when we derived Eq. (9), and
using Eq. (30)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (31)
where y,, etc. are the total strains. Note that where the forces, for example,
1/J,, have fixed values or/J,=O. Thus ultimately
Sw = fyoadV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
where
(32)
W*=fyoa
(34)
W*=U;*+ST}os
0
. . . . . . . . . . . . . . . . . . (35a)
where
O;*=f~:ou
(36)
and
~:ou=~:,,Sa,,+ ....... +~:,.8u,.+ ..... =o0;*
Note that the second term in (35a) depends on the sequence of application
of loads and temperature.
Under the same assumptions as for the internal potential energy, U;*
depends now only on the instantaneous state of stress. In a closed cycle of
deformation U;* is zero and 8U;* is a total differential of the right-hand
side of Eq. (36). Hence
oO;
oO;*
(37)
,,= oa,, ........... , ~:,.= oa, ...........
(see also Eqs. (18)). Note that the law of elasticity is arbitrary in Eqs. (37).
If 0 1* is given in terms of the stresses Eqs. (37) show that this determines
uniquely the strain-stress laws and vice versa. It is natural hence to call
0; * the complementary strain energy function.
Integrating over the body we find
I
(38)
W*=U;*+f [ fTJos]dV
v
where
U;*=fO;*dV=f [f~oa]dV
............................ (38a)
v
v 0
where U;* is called the complementary strain energy or complementary
elastic potential energy. When 0 =0 then
W*=U;* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (39)
In the case of linear elasticity W* = Wand U; = U;, but it is useful to
differentiate still between them. For the linear stress-strain laws ofEqs. (21)
and (22), Eq. (35a) becomes, if temperature and loads are increased together from zero,
w*=U;*+t
.......................................... (386)
and
-. _ _!_[L
- * =4G
1 [(u,,+a
U;
I+v )2 - 2(a,,u -a,.
2)]
(40)
where all symbols in the brackets refer to final values. Eqs. (41) are known
as Clapeyron's theorem. Another very useful theorem is due to Betti.
Assume that the body is subjected to two force systems P 1 and P 2 Let
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 process we prove easily-noting that the final state IS m each case the same-that
W 12 = W21 . . . (42)
where
W12 =~P 1 u 2 and W2 =~P 2 u 1
(42a)
See Num'o Cimento (2), Vols. 7, 8, 1872.
are the work done by the system of forces P 1(P2) over the displacements
u2(u1) respectively. Relation (42) is known as the generalized reciprocal
theorem of Betti.
A special form of Eq. (42) is Maxwell's reciprocal theorem. Thus, if
systems l and 2 consist each of one force (or moment) only then
l. u12 =1. u21
. . . . . . . . . . . . . . . . . . . . . . . . . .
(43)
8W=SU; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(44)
8u7"0 at A and B
but bears obviously no relation to the actual displacements of the framework due to the force P.
We find easily:
force m member l, 2: N 12 =
Pab
-7112
Suh Suh
lh
1 2
-,--~~~~~~~--~--~--h
Thus,
Hence
Sw=SU;
as indicated by the principle of virtual displacements or virtual work.
In order to realize best some of the implications of the principle of
virtual displacements let us consider again the derivation of Eq. (44)
which applies when S7J =0. In accordance with the analysis of p. 5, if
we multiply the internal equilibrium Eqs. (4) with the virtual displacements Su, Sv, Sw, sum the three expressions, integrate over the body,
apply Green's theorem Eq. (6), and note the boundary conditions (5) we
obtain Eq. (44). Again if we start from Eq. (44) we can apply Green's
theorem in the opposite direction and are led to
J{[
v
c;,
c;, }
.]ov+[..... . ]ow dV
For arbitrary virtual displacements Su, Sv, Sw this relation can only be
true if each of the brackets vanishes separately; thus, starting from the
principle of virtual work we have re-established the conditions of equilibrium. An exception occurs, of course, where a displacement (say u) is
fixed on the surface and Su is automatically zero there.
We conclude that the principle of virtual displacements Eq. (44) is a
necessary and sufficient condition for the existence of equilibrium of an
elastic body. Or otherwise we can state that by using virtual, i.e. kinematically possible compatible displacements, we substitute Eq. (44) for the
internal and external equilibrium conditions. Note that the idea of strain
energy is not necessary to the establishment of the principle of virtual
work.
Since the forces are assumed to be applied on the undistorted system
and to remain constant during the virtual displacements we can regard
S Was the variation of a potential - U,. Thus
Sw= -Su, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(45)
and
U,= -.f[w.,u+w.v+w.w]dV-H4>,u+4>.v+4>.wJdS
v
(46)
Thus
()U,
w.,=-7iil
which is the usual definition of a potential of forces. Note that Su. is a
total differential of the elastic displacement increments Su. U, is denoted
as potential of external forces.
We can write now Eq. (44) in the concise form
S,U=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (47)
where the suffix E indicates that only elastic strains and displacements are
varied.
U=U;+U.+const. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (47a)
is the total potential energy of the system. Eq. (47) states that a position
of elastic compatibility of an elastic body is also one of equilibrium
(i.e. the body is at the true position of equilibrium) if any virtual variation
of the displacements and strains whilst forces, stresses and temperature
remain constant does not give rise to any (first order) variation of the total
potential energy. The particular form (47) is known as the principle of a
stationary value of total potential energy, if the latter is expressed in
terms of displacements. Note that U; itself may be calculated from
formula (l9a).
Actually the stationary value of U is in our case always a minimum
and this confirms our previous assertion that with the assumptions of our
analysis all systems are stable. The mathematical proof that U is a minimum
at the true position of equilibrium is straightforward ; the point is discussed in greater detail under (C) below.
We have assumed until now that the initial strains 7J are due to a temperature variation. However, this is an unnecessary restriction and there
may be strains arising from any source of self-straining. For example, in a
framework they may be due to manufacturing errors in the lengths of the
bars. In the more general cases of self-straining not only may the "YJ~x
"YJ and 7]zz be different, but there may also arise initial shear strains 7].,.,
7Jvz and 7]zxo In such problems substitute
(48)
for sS7J in Eqs. (15) and (19) and in the other related expressions.
Equations (44) or (47) may be used to derive the results which follow.
(A) The differential equations ofthe theory of elasticity for arbitrary loading
and temperature distribution or any particular structural problem in
terms of the displacements; the appropriate static boundary conditions
Fig. a.-Example of an arbitrary virtual displacement
B
in terms of the displacements follow also from this analysis. It is important to note that in all applications it is best to form directly
SU1 = f[ailE]a v
v
and not to evaluate first U1 and then to take its increment S.
(B) Castigliano's* theorem Part I generalized for thermal effects
[~~~ 1~=
(49)
const. =P,
where P, is the force (moment) applied in the direotion of the deftexion
(~otation) u,_. TJ:lls relation may be obtained immediately if we apply a
virtual elastic displacement Su, solely to one external load P,. Note that
Eq. (49) applies also for non-linear stress-strain laws and may also be
generalized for large displacements.
(C) The Principle of Minimum Strain Energy when U1 is expressed in
terms of the displacements and the temperature is not varied.
We arrive immediately at this theorem if we select only such virtual
displacements Su which are zero at the applied forces. Then
SW=O
and we conclude from Eq. (44) that
S.U1 =0 and U;=min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (44a)
at true position of equilibrium if only such virtual deformations are
allowed that no external work is done.
Hence, if we compare all possible compatible states of deformation of a
body associated with a given set of displacements (not sufficient by themselves to fix completely the deformed shape of the body) then the true
position of equilibrium has the minimum strain energy. This is still true
if the body is subjected to temperature loading.
The point is of sufficient interest to warrant some elaboration. First it
may be helpful to point out that when we state that at the position of
equilibrium the total potential energy has a stationary value and that this
is a minimum we do not compare physically possible adjoining states.
For, in stating that the potential energy has a stationary value, i.e. SU=O,
we compare the true position u, with a position u+Su assuming in both
cases that forces and stresses are the same. This can obviously not be true
for the second position since for given forces there is a unique position of
equilibrium. In fact, we mentioned that this arises due to our legitimate
neglect of the higher order terms in Sa and oP. Also, when we go a step
farther and state that the stationary value is a minimum we prove this by
considering the influence of terms like !Sa,., S~:xx in U, arising from the
variation Sa.,., associated with S~:"'"'' but we still keep the forces constantalthough this cannot, in general, be true.
Having pointed out these aspects of the virtual displacements approach
we shall, in what follows, discuss the question of the extremum of U1 from
a more physical point of view. Again we prescribe certain displacements
on the body and do not allow any forces P other than those arising due
to and in the direction of the given displacements. The structure takes up
its natural position of equilibrium from which we can deduce the value
of the forces P. If we want to force the body to assume a position u+ou,
v+ov, w+ow while keeping the set of prescribed displacements constant
we must apply certain additional body and surface forces to push the
system away from its natural configuration. The work done by these
constraint forces (!'~.;oP ou, obviously positive), produces by reason of
equilibrium in the new position an equal increase in the strain energy
stored. Thus, the strain energy in any neighbouring compatible configuration is greater than that for the unconstrained original position and hence
the strain energy there is a minimum.
An alternative way of producing a state of equilibrium different from the
natural one in an elastic body under a prescribed set of displacements is
the introduction (prior to the imposition of the displacements) of internal
or ext~mal constraints that do no work. For example, in a shell or plate
analysis, we may assume that the middle surface is inextensible and the
transverse shear strains are zero; thus, in this case we impose infinite
values for E and G in the middle surface and an infinite value for G in the
*See A. Castigliano, Theorie de /'equilibre des systemes
elastiqu~s,
Turin 1879.
_1I
I
0
u= Uo(X, y, z)+La,u,(x, y, z)
r~l
V=' V0 (x,y,z)+Lb,vr(x,y,z)
r ''I
r --I
......................
(50)
where U0 , V00 w, satisfy the kinematic conditions where these are prescribed
and u,, v, w, are linearly independent functions which vanish there.
a,, b,, c, are unknown constants to be determined by the Rayleigh-Ritz
procedure. The elastic strains corresponding to (50) are (see Eqs. (I) and (2))
aw
av
au
E,,=sx -YJ, Eyy=a-y -YJ, E,,=sz -YJ
(51)
aw au
av aw
au av
Eu=ax+az
E,"=i\Y+ox' E",=oz +s/
The chosen series satisfy the displacement boundary conditions and the
infinitesimal deformations
OU=Oa, u, OV=Ob, v,, OW=,Oc, w, . . . . . . . . . . . . . . . . . . . . . . . . (52)
*See Rei.. 3., VoL II. Jl. 94, and also 'General Theorems Relating to Equilibrium and Initial and
Steady Motion , Phd. Mag., March 1875. They have been discussed more recently by D. Williams in
Phrl. Mag. Ser. 7. Vol. 26, 1938. p. 617.
t W. Ritz, "Theorie der Transvers~lschwingungen einer quadratischen Platte.' Ann. d. Physik,
VoL 28. p. 737, 1909; see also J. reme u. angew. Math . Vol. 135, p. 1. and Gesammelte Werke,
Parts, 1911, p. 192.
obtained by variation of a" b., c, while the other coefficients and the temperature are kept constant are hence virtual elastic displacements. Note
again that the chosen u, v, w functions need n'ot satisfy the given static
boundary conditions. Naturally, the accuracy of the analysis is enhanced
if the latter are also satisfied.
To determine the coefficients a., b., c, we use the condition of the
stationary value of the total potential energy in the forms
DU=O or SU;=DW
There are also cases when the theorem of minimum strain-energy, U; =min.,
is useful, see for example Part H of this series example 4. The application of DU=O ensures the average satisfaction of the equilibrium conditions.
If we evaluate U in terms of (50) then for a linearly elastic body we
obtain a quadratic function in a., b., c,. Condition DU=O which in this
case becomes
ou
ou
ou
oal = ...... =oa, = ...... =oa, =O
ou
ou
ou
Ob 1 =
=ob, =
=obu =O
ou
ocl =
......
ou
=oc, =
......
ou
=oc, =O
l
f ............... .
J
J[ oa""+Oaz"+oaM+w JvdV=O
+w u dV=O
J[ oa~"'+oa.""+oau
ox oy
oz
r
.t:
(53)
_j
leads to a set of 3n linear equations in the 3n unknowns a,,b" c,. It is,
however, superfluous to evaluate first. U and then to differentiate with
respect to the unknown coefficients. We can obtain directly the final
equations by forming the 3n expressions
S.,( U; + U,) =0, 8b,( Ui + U,) =0, De,( U; + U,) =0 . . . . . . . . . . . . . . (54)
where the suffices ar, br, cr indicate that the virtual displacements are
chosen respectively as in Eqs. (52).
Using the first of Eqs. (52) in the first of Eqs. (54) we obtain the more
explicit form
{faE,dV ~~Pu}Da,=O
v
where the Su's are virtual, i.e. kinematically possible infinitesimal displacements. Eqs. (57) lead to the principle of virtual work Eq. (44) if the a's
satisfy not only the internal but also the boundary equilibrium conditions
Eqs. (5).
Assume now that the displacements u, v, w are written in the approximate form of Eqs. (50) where the a, b" c, are unknown coe:ficients. However, contrary to what we assumed in paragraph (4), series (50) are taken
here to satisfy not only the kinematic conditions wb.ere prescribed on the
surface but also by substitution into the stress-strain relations the equilibrium conditions where prescribed.
Expressing now the stresses in the brackets of (57) in terms of the displacements and temperature 0 and selecting as virtual elastic displacements one of the 3n possibilities (52) we obtain the n equations in the 3n
unknowns a" b, c,
0\'
.
OZ
OX
(5S)
This is the Galerkin *t procedure usually only given for linear elasticity. To
fix ideas take the case of linear elasticity and write axx etc. in terms of
u, v, w. We find easily three equations, the first of which is
..............
(59)
ou ov ow
g=ox +s:Y+oz . . . . . . . . . . . . . . . . . . . .
(60)
where
o 2u o2u o2u
~u=ox2+oy2+oz2 and
The bracket in Eq. (59) is, of course, the equilibrium equation in the
x-direction expressed in terms of displacements.
The procedure in any particular structural problem is to form the
equilibrium conditions in the stresses or stress resultants and express each
in terms of the displacements. Next we multiply each by the corresponding
u, (which may be a deflexion, slope, twist) and then integrate over the
body. Thus, for a beam subjected to a distributed loading p in the y-direc.tion, the equilibrium equation in terms of the deflexion v is, assuming
engineers' theory of bending to hold
::2 (EI;;~)~p=O
.................. ....................
(61)
J{;;
2(
El;;~) ~p }8vdz=O
..........................
(61a)
It is easy to see that for displacement functions (50) satisfying all boundary
conditions the Galerkin and Rayleigh-Ritz methods must yield the same
equations for a, b" c, and hence also the same deformations. We need
only realize that in this case Eqs. (57) are indeed equivalent to the principle
of virtual work. Hence substitution of u, v, w in Eqs. (57) must give the
same result as substitution into
SU=O
The advantage of Galerkin's method lies in a more direct derivation of
the equations in a., b., c,. However, contrary to what is usually assumed,
this advantage is small if we calculate SU directly. We note also that
Galerkin's method allows only such approximate functions as satisfy all
boundary conditions, while the Rayleigh-Ritz procedure requires only the
satisfaction of the kinematic boundary conditions.
* Timoshenko 8 mentions on p. 159 that equations of this type appear already in W. Ritz's work.
See references on p. 353 and also Gesammelte Werke, p. 228.
t B. G. <,;alerkin, 'Series Solutions of Some Problems of Elaftic Equilibrium of Rods and Plates,'
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,
1938. p. 636.
I~
II
(57)
v
fjg, 10.-Cantilever beam
(a I)
--4---El/J
I
The first part of the problem is, of course, trivial and the result known
to any undergraduate, but we want to show here how the Virtual Displacements method can be applied in such a case.
We consider virtual displacements consisting of small arbitrary additional. deflexions (Sv) of the beam from its equilibrium position. The
Principle of Virtual Displacements then expresses the equilibrium condition in the form
SU1 -SW=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a2)
The virtual displacements must satisfy certain kinematic boundary conditions, namely :
Sv=O for z=O and z=2/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a3)
and since both structure and loading are symmetrical we need only consider 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)
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
oA
2J {Efvlv_p}Svdz+2Elv"(Sv)' I
0
But
(Sv)z' =0, ( Sv) 0 =0 and (8v), = 8 V
Since otherwise 8v is arbitrary we conclude that to satisfy Eq. (all)
we must have
(Elv"L~o=O, 2(Elv"')zd-P=O ............................ (al2)
and v must satisfy the differential equation
EJviv_p=O ........................................... (al3)
Eqs. (al2) and (al3) together with the kinematic conditions
(v)z~o=O and (v')zd=O
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4)
give all the necessary information for the determination of v.
Integrating Eq. (al3) and using the boundary conditions (al2) and (al4)
we find finally for V the quadratic equation
(I +!{i)(V/ V.) 2 -[l +!{!(!+a)+ V1 /Vo]( V/V.)+ V1 /Vo=O . .. . .. (al5)
where
!{i=kl 3 /6EI, V1 =Sp/4 /24
(b) Plane Redundant Framework
A plane framework consisting of a single joint connected by a number
of hinged bars to a rigid foundation is loaded by forces X and Y along the
axes Ox, Oy respectively (FIG. 12). In addition, the bars are heated to
arbitrary temperatures and have also initial strains due to errors in
manufacture.
Find by application of the Principle of Virtual Displacements the forces
in the bars.
Let u, v be the displacements of the loaded hinge, measured from the
position for which all bars have the correct length and are at zero temperature. Then the total direct strain in the rth bar for these displacements
is
u cos Br+v sin Br
Yr
fr
. . . . . . . . . . . . . . . . (bl)
The total strain is made up of the elastic strain Er together with the
thermal strain 7Jr=a0 and the initial strain 'Y]oro where the initial strain is
'Y]or=t1fr/lr ... ......................................... (b2)
t11r being the additional length of bar (in excess of the correct length) due
to manufacturing errors or other causes. Hence the elastic strain due to
u, vis
Er=yr-'Y)r-'Y]or
2J faSEdAdz=2Ef { fy 2dA}v"(8v)"dz=2Elfv"(8v)"dz
......................... .
>]
........... .
I -v"'(Sv) I +Jviv.Svdz}
.. x
(a7)
SW=2fpSvdz
0
10
(a8)
(a9)
(b4)
(b3)
and the work done by the distributed torque and the end torques
8U;="2:.Ararlr0yr
r~l
=fEArlr[u cos
r=l
r
0
(b5)
(b6)
8W=X8u+ Y8v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
By the Principle of Virtual Displacements
and therefore
(c8)
1
0
Integrating by parts the first term twice and the second once, we finally
obtain
8U1 ~8W=O
n
(c9)
r~l
r~l
r~l
sv[u 2:~:r sin 8,cos 8,+v L:Etr sin 2 8r~ IEAi7]r+7]or)sin8,- y J=0
r~l
r~l
(c7)
r~l
For the integral to vanish, since 88 is arbitrary, 8 must satisfy the differential equation
Er8iv -GJ8" ~m.=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (ciO)
(b7)
If Eq. (b7) is to be satisfied, since Su, Sv are arbitrary, the two expressions in brackets must separately be equal to zero and hence we have two
equations in the displacements u, v. Solving for u, v the stresses and forces
in the bars can be calculated from Eq. (b4).
The two equations are, of course, the equilibrium conditions in the
x andy directions and could be derived directly by statics without recourse
to the Principle of Virtual Displacements.
Note that there are always only two unknowns in this approach, regardless of the number n of bars. Hence it is preferable to operate with the displacements u, vas unknowns than with forces in the bars when n>4.
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
and the second the condition for zero direct stress.
For the series solution, we represent the twisted shape by the Fourier
series
Tt=O
GJ(8')t~Er(8)'"t~
and
r<8")t=o
(ell)
. (nTTZ)
8= '\:""
L...Ja, Sin
-1~
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (cl2)
n~l
n=l
J.
J .
mTTZ
nTTZ
'\:"" nmTT 2 J
8U;~8W=8a.., [ GJ L...Ja,~ cos - 1- cos ~1~dz
(cl3)
(11mTT 2) 2
mTTZ . nTTZ
mTTZ
+Er f:;.a" f2 o s1n - 1-sm ~1~dz~ om. sm - 1-dz
\.-,
If m. is constant
0
Ezz
(cl)
and
(c2)
J [ Ja ..SE ..tds]dz=J8"(88)"
0
[ fw 2 tds]dz=JEr8"(88)"dz
0
....
(c5)
,.
fGJ8'8(8')dz=JGJ8'(88)'dz
..............................
AIRCRAFT ENGINEERING,
(c6)
4[2
form odd
and
form even
a,..=O
which gives for the twist distribution
I
I
. . . . . . . . . . . . (cl5)
00
(nTTz/1)
8 -~ 41 2 m.~ [ sin(n7T)2r]
GJTT3 n~l
113
I+
GJ
........................ (cl6)
odd
In this case, since the assumed form of solution satisfies also the static
boundary conditions 8" =0 for z =0 and I we can alternatively use the
Galerkin form of the Virtual Work equation, which is in this case
I
. . . . . . . . . . . . . . . . . . . . . . . . . . . . (cl7)
11
FIG.
pf4
oU;*=
where the integral in the square bracket refers to the integration over the
cross-section. For the case shown in FIG. 14(c).
o~~z~t/2
for f/2<z<l
If we apply now a force -- fooP at the free end our virtual system (c) is transformed 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 righthand side of (62) as
OU;*+friosdV=o(U;*+frr~dV)
12
~ oU,*
(65)
where
8U,*~Jy8adV
........................................ (65a)
but
(66)
U1* -t- frysdV f [ JEOa]dV f frJ.w/V
v
v 0
v
since <1 =const. U,* we term the complementary potential energy of total
deformation. Note that it is always simpler to calculate directly oU,*
from Eq. (65). Particular care is necessary in evaluating U,* for as Eq. (66)
U,*
shows in the first integral E is taken to vary with a from the initial to the
final state while 7), s in the second integral refer only to the final values.
Physically speaking we may consider U>~* as the complementary work
necessary to reach the final true state of deformation from an initial state
in which we allowed free thermal expansion and destroyed compatibility.
Formulae (65) and (66) may be extended immediately to the case of
arbitrary initial straining by substitu1ing
for
7JS
0 ..................................
(67)
where the suffix a indicates that only forces and stresses are varied and
U*=U.,*-t-U,*
(67a)
is defined as the total complementary potential energy of the system.
Eq. (67) states that a state of equilibrium of an elastic body is also one of
elastic compatibility (i.e. the body is at the true position of equilibrium)
if any virtual variation of the stresses and forces, while displacements
remain constant, does not give rise to any (first order) variation of the total
complementary potential energy. This theorem we call the principle of a
stationary value of total complementary potential energy if the latter is
expressed in terms of forces and stresses. Actually the stationary value of
U* is a minimum as we may prove without difficulty.* This point is discussed in more detail under (C) below.
Eqs. (62) or (67) may be used to derive the results which follow.
(A). The differential equations of the theory of elasticity (for arbitrary
loading and temperature distribution), or any particular structural
problem, in terms of stresses or stress-resultants; the appropriate
kinematic conditions in terms of forces and stresses follow also from this
analysis. It is important to note that in all applications it is best to form
directly
oU,*=f[yoa]dV
v
and not to evaluate first U.,* and then to take its increment oU,*.
(B). Castigliano'st Theorem Part II generalized for Thermal Effects and
non-Unear elasticity
[ (!Ud*]
(JP
r
B=const.
=u, ..................................... .
(68)
the cuts. If we impose the vii-tual stresses oa, it is apparent that since
these produce corresponding deformations ou,. on the cut faces the latter
are not any longer compatible. It is important to realize that the oar
systems are self-equilibrating since the external loads P remain constant.
Thus, in a framework we may obtain a system oa,. by cutting a redundant
bar and applying a variation oN to the true force N in the bar.
We now investigate the differences in complementary work W* and
U>~* between the original equilibrium position of the uncut body and the
new enforced equilibrium position of the cut body. Comparing Eqs. (38)
and (66) we find
1
(70)
v
0
In moving from the uncut (compatible) equilibrium state to the cut one we
note first that the integral does not vary since 7) is constant in this step.
Also the first order increment, oUd*, of Ut~* is zero since this is the condition for compatible equilibrium of the original body and no first order
increment 8 W* can arise since the loads P remain constant. We are then
left only with second order increments.
For the complementary work this is
o 2 WL1-fou,oacdS ... ................................. (70a)
W*
(where the integral is taken over the cut faces) which is the work of the
virtual stresses oa c over the displacements ou c they themselves produce.
This is clearly positive. The second order increment of U,* is merely
o 2 U,* =Hoa oEdV . . . . . . . . . . . . . . . . . . . . . .. . . . . . .. . . . . . . (70b)
v
since 7) remains constant; oa and OE are the stresses and strains due to oa ,.
Terms (70a) and (70b) are equal and both positive.
We conclude that the complementary potential energy of total deformation U,* and the complementary work W* have for given forces and temperature distribution a minimum at that position of equilibrium of the uncut
body at which compatibility is satisfied.
It follows that if U.t* is overestimated by assuming a statically equivalent
stress system which does not satisfy all compatibility conditions and we,
ignoring the latter fact, equate U<~* to W* of the applied loads Palone
we cannot but overestimate the magnitude of the displacement system
under the loads P. Conversely to achieve a given displacement system our
calculations based on a non-compatible stress system must underestimate
the corresponding load system P. Thus, the latter has its maximum for the
unique equilibrium position which is also truly compatible. This may be
expressed also as follows :
For given displacements and temperature distribution the complementary
potential energy of total deformation has a maximum when the state of
equilibrium satisfies also the compatibility conditions.
The above theorems may be combined to give:
The stiffness of an elastic body in which the equilibrium conditions are
satisfied is a maximum when the elastic compatibility conditions are all met.
Thus we see that the effect of introducing assumed forms of stress distribution for the purpose of approximate solutions is the opposite to that
of the method of Virtual Displacements and therefore application of both
methods to a given problem yields upper and lower bounds to such
aggregate quantities as stiffness. No general conclusion as to bounds can,
of course, be drawn for the details of the stress distribution.
The above theorems which apply also in the presence of initial strains
other than those due to temperature do not appear to have been given
before with this degree of generality.
The Unit Load Method
Assume that we require the deformation (deflexion or slope) u, at a
given point and direction of an elastic redundant body subjected to given
forces and thermal effects. Let the actual total strains in the structure be
known and given by*
(D).
Yx~=Eu+7),
Yxy=Exy
(71)
where Oaxx . ... oa~. ... are the virtual stresses due to oP T In a linearly
elastic system Sax~ etc., are proportional to oPT and Eq. (69) can be written
lu,= J[y,.p,x-1-YYPv+YzPzz+Yxa,,+y,.a +yz,azr.]dV .... . (71a)
where
au . ... a,.... are the stresses due to a unit load. Since ii,.,.. ...
a,,.
13
'
/\1\1\--
A
-,~r,I
I
-,..,\--
\1
'1
\1
',
,..,..~
~~~--a--~2~--~~--~--~A
Unit load
Normal
(l,m)
y,vL
- -;',
\1\
''
~-
x.u
(74)
F must, of course, also satisfy the given boundary conditions. In fact,
where surface forces are prescribed we have (see Eqs. (5) and FIG. 16)
I
I
o (oF
oF)
oy 1oy -mox =f~
J ............................
(76)
F=Fo+'J:,b,F,
r= I
We use here
(72)
o~+o~o~-0
oy 2 ox 2 - oxoy................................. .
(73)
where Yxx etc. are the total strains E.,.,+'l') etc. For a given stress-strain law
we cart express y.,., in terms of the stresses (72) and the temperature 0.
Hence by substituting into (73) we obtain the differential equation in the
unknown F, which will, in general, be non-linear. However, in the case
of bodies obeying Hooke's law we obtain for a=const. the simple linear
result
14
(75)
J( ... .)dA
(78)
J[
(79)
J[
. . . . . . . . . . . . . . . . . . . . . . (79a)
If the total strains are expressed in terms of the stresses (72) and temperature distribution 0 we obtain from (79a) n equations for the unknowns b1 to bn. These are only linear if the body follows a linear stress
strain law. In the latter case and for constant body forces Eq. (74) shows
that the solution must be independent of the Poisson's ratio v if all
boundary conditions are of the static type. Hence we may take v=O and
(79a) becomes
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (79b)
from which we may obtain without difficulty the n linear equations for
b1 to b,..
The case when all prescribed boundary conditions are static is interesting
for all surface conditions are then exactly satisfied by (76). This indicates
that it should be possible to express (79a) in a form similar to that given
as Galerkin's method under the virtual displacement principle. In fact, if
we integrate (79a) twice by parts, or better, if we apply Green's Theorem
.(Area 8
T..,_
b
. . (SOc)
lO~q
Areaw;,
_t__
c1z
.__,
..1!.:_
il+~l:1
~-ckness t IOtq
'
1-.P.
~I ~z-
-!(lOYx +moy.,)JF:ds
+J[1oy +moy=
r
ox
oy
oy
ox
c
m )oFrJ
I )oFr
ox ds
oy +( ly,,-pxu
+.f[( my.,,-2y.,.
denotes the line integral along the boundary of the twodimensional continuum.
But, on the boundary Fr =0 and also
~~ =0 and ; ; =0
=0 for
r= I
to n
............
(SO)
which shows clearly how the method of virtual forces satisfies in the
average the compatibility condition (73). When the body is linearly elastic
Eq. (SO) may be written as (a=const.)
(o20 o20)
F o4F o4F )
J[(oox4+oy4+2ox2oy2
+Ea ox2 +oy2
ow., ow)] FrdA =0 for r= I ton
+(1-v) ( ox+ oy
............................................
(a2)
For the virtual forces we consider a variation 8P. in the stringer load.
The applied forces P are maintained constant and hence to satisfy the
equilibrium conditions on the free end (P,=O) we must take 8P. to be
zero there
(oP,),d=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Otherwise the variation 8P, is arbitrary.
i.e.
(a3)
(a4)
. . . . . . (SOa)
the expression in the bracket being of course Eq. (74). Note again the independence of the solution from v when body forces are constant.
The above application of the principle of virtual forces is a generalization of a method developed by Timoshenko 8 , p. 167. A thermal stress
example of the above analysis is given in Part H.
Naturally, the method can be extended to three-dimensional cases.
W. Ritz* proposed as early as I 90S a similar procedure for the solution
of St. Venant's torsion problem; this method is illustrated on an example
of considerable complexity in Part II.
A slightly more refined approach than that shown above may be
adopted when it is possible to estimate accurately the variation f of F
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
F=f(y)cp(x) . . . .. . . .. . .. . .. . . . . .. . . . . . . .. . . . . . .. . .. . .. . . (Sl)
where cp(x) is an unknown function of x. It is, of course, possible to formu-
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;
when there are also kinematic boundary conditions the corresponding
boundary expressions for cp follow also from (79a).
Consider, for example, the case of linear elasticity, zero body forces and
pure static boundary conditions. Eq. (SOa) takes here the simple form,
(al)
and from the equilibrium of the free end of the panel we find for the load
Pn in the edge members
Pn=P- ~"
o2y o2y.,.J
J[o2y,.,
oy2 + ox2 -oxoy FrdA
(SOb)
Using (SI) in (SOb), integrating with respect to y and noting that (SOb)
must be true for any virtual variation ocp, we obtain the differential
equation in cp
W. Ritz, J. reine angew. Math., vol. 135, 1908, and Ann. d. Physik, series 4, vol. 28, p. 737,
1909.
8; ........................................... .
8Pn=-
(a5)
Since the applied forces are not varied, the virtual forces principle (Eq. (62)
for 0 =0) reduces to
8U;*=0
The virtual complementary energy due to
8P. is
DU;*= J[%,4oP,+2;;oPB+2t1b8q]dz
0
(a6)
d;~-~:(1+2~)P.+~~P=O
....................... .
(aS)
. . . . . . . . . . . . . . . . . . . . . . . . . . .
(a9)
or
2
d2P. 2 dz2 -fL P.- -fL Pso
where
2
fL =
2A
1)
Eb A+2B P.a= A+2J!'
2Gt(l
. . . . . . . . . . . . . . . . . . . . . . . . (aiO)
15
:1
""q0
- -1
...../t
.............
...-..,- --
Am
2
~
1 -----::1
..L
I
\
-1.
(a)
I
l-
--
~
y
(b)
faa.,'~
7
' ';
A
b ~~~,,,-'
bb
- -I - ,
1s
$~
qo1
' - -- l~:1
----s --
E00 dbcos~
;r~
(\..,+
l ~JOt o.x
'I Qry
--- q..y
0,..
O.r
+ayy
1
I
,-t/
/'
aa. ,'
,~
~ It""'"'~
~I ~~~'~"" ~
b.Ytt:
X
P,=P
[.l
co~~~(~/z)J
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al3)
Iw= Ju ..E..Adz
0
and since
a..A=unit load= I, Ezz=P./EA
I
w= J,4dz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4)
0
which is, of cour3e, merely the extension of the stringer under the varying
end load P,.
(b) Rate of Twist of Multi-cell Tube
I
ql =2Am . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(bl)
around the single cell and the rate of twist is given immediately as
................................
(b2)
where the integral is obviously taken only around the single cell (1, 2, 3, 4).
Eq. (b2) is, of course, a well-known result in the theory of closed thinwalled tubes.
(c) Plan. Stress-Strain Relations for Oblique Co-ordinates
16
(c2)
and hence the virtual complementary energy per unit thickness of the
element dxdy is
DU/ =dxdy[c rxDCTu+EvvDCTuu +ExyDCTry]
dxdy
= E
[unDCTu +u DCTyy -v(uu8u.,+uuuDCTu)+2(l +v)CTxyDCT~]
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (c3)
Substituting for CTxx u u,. from (cl) we can now express 3U;* in terms
of the oblique stresses and virtual stresses. Thus for the virtual stress
8ubb we find for the virtual complementary energy
* dadb
,\
DU; = E sin y[ubb- CTaa + 2uab cos 0]8ubb
.. ... . . .. .. . .. .. .
(c4)
where
A=v sin 2 0 -cos 2 0
(c5)
From FIG. (19) the virtual complementary work of 8ubb is seen to be
8W*=da8ub!,.Ebbdb .. .. .. .. .. . .. .. .. .. .. .. .. .. .. .. .. .. ..
(c6)
and therefore from the Virtual forces principle
8W*=SU;* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(c7)
we find, using Eqs. (c4) and (c6) in (c7)
l
Ebb= sin o[ubb-Au,..+2uab cos{)]
(c8)
0
Applying a virtual stress 8uaa in the same way we obtain for the strain
Eaa the corresponding expression
I
..
(c9)
Consider now the virtual shear stress 8uab. From Eq. (c3) we find for the
virtual complementary strain energy due to 8uab
dadb
.
8U;*= sin o{2[(l +v) sm 2 0+2 cos 2 0]uab+2(u +ubb) cos 0}8uab
. . . . . . . . . . . . . . . . . . . . . . . . (clO)
Calculating the complementary work of the virtual shear stress uab
we find (see FIG. 19)
8W*=DCTabkaa cos 0+Ebb cos 0+Eab sin {)]dadb
(ell)
Thus the virtual shear stress 8uab does work not only due to the shear
strain Eab, but also due to the strains Eaa and Ebb.
Substituting from Eqs. (c8) and (c9) for the strains E00 , Ebb and equating
3U;* and 8W* of Eqs. (ciO) and (ell) respectively we finally obtain for
the shear strain:
2(1 +v)[
cos ()
J
Eab=-E-- uab+ 2 sin2{)(uaa+ubb)
...................... (cl2)
0
Note that with the strains defined as above the increment of complementary energy is
8U;*={E00 DCTaa+Eb1JDubb+[(Eaa+Eb1J) cos 0+Eab sin 0]8uab}dadb . . (cl3)
as compared with the simple result for rectilinear axes in Eq. (c3).
'
t See also H. Mueller-Breslau, Die graphische Statik der Baukonstruktionen, 1 ed., Koerner,
Leipzig, 1886.
~ 0. Mohr, Zivilingenieur, Vol. 38, p. 577, 1892; see also, Abhandlungen aus dem Gebiete der
technischen Mechanik, 2 ed., Berlin, 1914, p. 407.
'ik
17
Following Mohr's <:~nalysis his ideas were applied to stiff-jointed frameworks, the first systematic work being that of the Danish engineer Axel
Bendixen.* However, the great potentialities of the method were only discovered 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 equations for the unknown displacements in a structure complement MuellerBreslau's equations for the redundant forces. It is regrettable that Timoshenko 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 HardyCross!! 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 problems. 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 comprehensive 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 understanding 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 corresponding 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:
. ....................
(82)
(83)
To find the flexibilitiesfin any linearly elastic body we may use the unit
load method developed in Section 6D. Thus, from Eq. (71a),
1r...........
(84)
Axel Bendixen. Die l\4cllwdt tier Alplu~Gieichungen :.ur Berechnung ton Ruhmenkonstruktionen,
Springer, Berlin, 1914.
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
Young, Theory of Structures, McGrawHill, New York, 1945.
II Hardy Cross, 'Analysis of continuous frames by distributing fixed end moments'. Paper N<
1793, Vol. 96, Trans. A.S.C., 1932, pp, 1-10.
R. V. Southwell, Relaxation methods in engineering science. Oxford Univ. Press, 1940. R. V.
Southwell, Relaxation methods in theoretical physics, Oxford Univ. Press, 1946.
'T R. A. Frazer, W. J. Duncan, A. R. Collar, Elementary Matrices, Cambridge Univ. Press,
Cambridge, 1938.
R. Zurmuehl, Matrizen, Springer, Berlin, 1950.
tt B. Langefors, 'Analysis of Elastic Structures by Matrix Transfonnation with special regard to
Monocoque Structures'. Journ. of Aero. Sci., Vol. 11, No.7, 1952. Structural Analysis of SweptBack 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.
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.
18
O';Eh=a,,;Exxh+ : +a,,;Ezxh
~..........
(84a)
E;, a 1 (E1., ah) are the strains and stresses corresponding to a unit load at
and in the direction of j (h). Under load we understand either force or
moment. Similarly the f!exibilities may represent either displacements or
rotations. Naturally, formulae (84) yield also linear (angular) flexibilities
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
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 ..........
I
If;;= fa;E;dV
(84b)
where M;, S; (M,., S,.) are the moments and shear forces corresponding to
R; = 1 (R~> = 1). Eqs. (85) yield easily the following set of influence coefficients,
(a+b+c) 3 -(b+c) 3 (b+c) 3-c3
c3
faa=
3(E/)a
+
3(/)b +3(E/)c
a
b
c
+(GA')a +(GA'h +(GA')c
1 [ (a+b)2(2a+2b+3c)
b2
Ja2 =j"23 = 6(El)a
-b 2 (2b+ 3c) + 6(Elh(2b+3c)
a
b
+(GA')a +(GA'h
f 31 =.1;.
.................... (86)
To obtain f 22 and .1;.1 from the expression for faa. omit in th~ latte~ the
terms c and b c respectively. Also to find.l;. 2=f21 omit the terms mvolvmg c
in the last of'Eqs. (86). Naturally, we can also derive the influence coefficients (86) by direct integration of the differential equation for the deflected beam when shear deformations are included. A systematic method
for deriving the flexibility coefficients for compound engineering structures
is given later.
.
Influence or flexibility coefficients are of great importance in the static
and dynamic analysis of linearly elastic engineering structures. In this
connexion it is most appropriate to make use of the matrix notation not
only for conciseness of presentation but also for the ~ystemati~ progr~m
ming of the considerable computational work usually mvolved m practical
problems. The matrix algebra is, in fact, ideally suited for the automatic
digital computers now available.
The matrix form of Eqs. (82) is,
r=FR .................................. (87)
where r and R are the column matrices of the displacements and forces
II
r=
I~:
={r1
'2
'a},
I~:
R=
L__i
and
Uh2 .h~
ll
II
={R1 R 2 R 3 }
....
(88)
_I
................................ (89)
_I
.................. (91)
_j
where the/;~> can be calculated always from Eqs. (84). The flexibilities in
(91) need not necessarily refer tom different points. For example, we may
choose three directions x, y, z at a particular point of a three dimensional
body and define six flexibilities
J.,.,,/yy,Jzz
J.,.=J .,,f =f ,J.,=J.,.
corresponding to the three forces R.,=1, R.=1, R,=1 at the same point.
Similarly for a beam in which we assume that the engineers' theory of
bending is true we may require the slope and the deflexion at a crosssection under transverse force and moment applied there. Three flexibilities are required for this information; note, however, that if shear
deformations are included we must specify that the bending moment is
applied as engineers' theory direct stresses at the particular cross-section.
A characteristic property of the influence coefficients is that any/;~> in a
given elastic body depends only on the points and directions j and h but
not on any other directions selected for the calc].Jlation of a flexibility
matrix (91).
A perusal of the flexibilities of Eqs. (86) shows that it is possible to split
the complete flexibility matrix F into two additive and distinctly different
matrices. Thus,
f=FB+Fs ...................................... (92)
where FB and F8 are the flexibilities corresponding to pure bending and
shear deformations respectively. The first contains only terms involving
Eland the second only terms involving GA'. For example,
a3
a
b
hiB=3(El).J23s=(GA')a +(GA'h
.................. (92a)
........... (9~)
where M;, N;, S; (Mh, Nh, Sh) are the bending moment, normal and shear
forces due to a unit load atj (h).
It is often convenient not to operate in single loads (or moments) but in
groups of loads (or moments), which are known as generalized forces. To
fix ideas, consider that in the example of FIG. 20 we select as applied
generalized forces the three loads QI, Q2 and Q3 given by,
t ................
(95)
or in matrix form
Q=GR
(96)
where G is the square matrix, in general not symmetrical, of the coefficients G;h We call G a load transformation matrix and assume that it is
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
obtain
R=8Q
where
} ................ (97)*
is the so-called reciprocal or inverse matrix of G. Its determination is
equivalent to solving Eqs. (95) and therefore involves considerable numerical labour if the number of equations is large. In such cases approximate
methods may have to be used. However, with the advent of the automatic
digital computers this difficulty is no longer insuperable. We give later in
this Section a systematic procedure suitable for punch-card machines for
computing the reciprocal matrix but hope to return in greater detail to this
and similar questions in Part III. Next we have to determine the generalized displacements q corresponding to the generalized forces Q. By definition q are obtained from the equality of the two expressions for work in the
two sets of variables R, r and Q, q. Thus, in matrix notation
W=tr'R=tq'Q .................................. (98)
where r' and q' are the transposed matrices of the column matrices r and
q and are hence the row matrices
r' = [r1 '2 ra], q' = [q1 q2 qa] (99)
Using the first of Eqs. (97) in (98) we obtain
q'=r'8
or
q=8'r
(100)
where 8' is the transpose of the matrix 8, i.e.
I
Bn B21 Bal I
8'= I~~: ~: ~:I
........................
(101)
_I
19
and therefore
r---a--l---a--1----a-----j
B
r=
I
- 06a+ 15a -04a
24a+ IOa+04a
--3 6a I Oa--04a
_j
~~-st
-Nil
Q1 mode
(Q,=-6a)
----~-3-&Q
_.............--,-511--L
where
(104)
8 28 54
_j
a3
Q2=1
fv=
(105)
_j
10~
1
-6 -1
F 0 =8'F8=a 1126
-6 25
10 -1 15
L
_j
(106)
(8- 1)'=G'=
I
01
06 041
04 04 -04
06 -04 04
L
20
_j
5: I
_j
(107)
(108)
_j
I0
fvu 0
...... 0 ...... 0
fv22 . . . . . 0 ...... 0
I
I0
I
I0
~,01
a= 6 El ........................................ (l04a)
8=
a/2
8 11 . . . . . . B1; . . . . . . B1 m
8=G- 1 =
F=a
r, =-sa
-0-61
............ (110)
_j
Thus,
P1=fvuP1, ...... ,p;=/v;;P;, ...... ,pm=fvmmPm
.... (Ill)
(112)
where fv is the flexibility matrix of a unit cube at the point x, y, z. Thus for
an isotropic body
See Frazer;Duncan, Collar, Toe. cit. Zurmuehl, foe. dt.
I
fv=
I 1/E
I -viE
I -v/E
-v/E -v/E 0
1/E -v/E 0
--v/E
1/E 0
0
0
1/G
0
0
0
0
0
0
0
0
0
0
0
0
0
1/G
0
-l
0
0
0
0
0
1/G
I
I
I
I
........ ( 114)
_j
L
Let also 5; be a column matrix for any statically equivalent stress system
corresponding to a unit load R; = 1 at the point and direction j. Thus,
i;={CT~..~i1 11 u}Gzzjij;,yiUvzi1z.-:i1
. . . . . . . . ... (115)
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
method as,
1/;h= Js;'ehdV= fs/fv5hdV= J sh'f~5;dV= I J,;
v
v
v
(116)
I f;; = f 5/fv5;dV
v
Hence the total flexibility matrix F for m points and directions is,
F=[/;h]=Js'fv5dV ............................. . (117)
v
where 5, 5 are the partitioned row matrices
5 [51 ...... 5; ...... 5m]
1
(118)
5=[5 1 . . . . 5; ...... sm]
b.,
0 ..... 0
f,, ..... 0
0 .....
0 ..... 0
F cc [I I .... I ..... I]
f"
0 ..... 0 ..... 0
fb 0 . . . . . 0
0 .....
0 ..... 0 .....
f, ..... 0
..... 0
..... 0
f, .....
. . . . . . . . (123)
. . . . . f,
-I
*The usual presentation of the strain and strC''\S matrix as a 3 / 3 squar~.. 111 ~ ... !.""~ Oi n:nsor is r.''~.. _,.
suitable for our considerations her('. Se~ also ~e'-twil uu.
There too attention js drawn to the facl
that i i need only be determined in the most suitable statically determinate svstem.
f,
21
where
lOtS(+)
\Ela
'D~M<+>
tR1
R2Jf"""""\
\Elb
Element a
tR
Element
c/>a,h=(L).,
R4 SJ
cl
(132)
lb
. . . . . . . . . . . . . . . . . . . .
gives the flexibility per unit length of the cantilevers. The negative sign in
the cross-flexibilities arises from the sign convention. The total flexibility
follows as
I
I
I f. o I
.......................... (131a)
f= o f"
r,
la
_j
.................... (134)
where
I
1
_,I I
-1. -I
0 -1
-- 1
I o o
~J
G
-1-/..................;...;....;,_A
Ele men!
fla,b=
i/6El
Mcb= -R4
} ............ (128)
J............
where
b.=
I~
L
0
-1 -lb
_j
and hence
b=
II
I b. I
.. .. .. .. .. .. (130)
_j
. ................ (130a)
I bb I
f3
22
(22),
12
. . . . . . . . . . . . . . . . . . . .
tcp I .
_j
.. .............................
<13t >
(136)
3'~'
_j
(137)
where,
lfla
fl =
flb
lI
.......................... (136a)
_j
L
It is easily seen that the flexibility matrix F of Eq. (137) is identical with
3cfo -2cfo I
I -2cfo
L
J2
FIG.
Hence,
bI
11-~. 1-~. I
-----1
-=---------Rm -km r +k,. 2r + . ..... +k,.;r;+ . ..... +k,..,..r,..
L .J
l 1cp
I 6'~'
Scb=Ra
I .................... (135)
I o o
I I
_j
L
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
M~
MBb
-- 1
0 38 )
.J
The coefficients k;i and k;h are known as direct- and cross-stiffnesses in
the directions of the selected m displacements. In fact, it follows directly
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)
of Eqs. (87) may be written as
R=F- 1 r=Kr ............................. ....... (139)
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=
.................. (140)
km; . . kmm
L
_j
The stiffness matrix may be determined either directly or from the identity
K=F- 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . (140a)
by inversion of the flexibility matrix F. Eq. (140a) shows that K is symmetrical, i.e.
k ih cc k hi . . . . . . . . . . . . . . . . . . . . . . . . . ( 141 )
This may be seen also as follows:
Let a1, E1 and a\ E" be the stresses and strains corresponding to unit displacements r1 =I and rh =I respectively while all other r displacements are
kept at zero. Applying now the principle of virtual work or displacements
to the true state j (h) and virtual state h(j) we obtain
lk;h=fci'E 1 dV=fa 1 EhdV~clkh;
........................ (142)
Where ci'E 1 etc. stands for an expression analogous to Eq. (84a). This
application of the pr nciple of virtual displacements, by an obvious
analogy with Eq. 84, is called the 'unit displacement method'.
We remarked on page 19 that the direct- or cross-flexibilities depend
for a given structure only on the points and directions to which they refer.
This is not so for the stiffnesses which by definition depend on the complete
set of points and directions selected to describe the stiffness of the body.
Thus, if we choose an additional direction m + l to augment our description of the elastic behaviour of the structure all the original k;h will in
general change whilst the jjh remain unaffected.
Consider again now the example of FIG. 20. A study of Eqs. (86) and
(140a) shows that the stiffnesses k 1" corresponding to unit deflexions at
B, C and D are considerably more complicated than the expressions for
the flexibilities. However, this is not always the case. Naturally, we can
calculate the stiffnesses directly. For example, the k 13 may be obtained by
analysing a continuous beam built-in at A and simply supported at B, C
and D at which last support there is a fixed 'give' of unity. We may solve
this thrice redundant problem either with the three-moment equation or
by the slope-deflexion method.
Assume now that not only transverse forces but also moments are
applied at the junctions of the component beams and at the tip (FIG. 23).
To simplify the argument we ignore, moreover, the effect of shear deformability. The modes and stiffnesses corresponding to unit displacements in
the directions, I, 2, 3, 4 can now be determined very easily. For example,
for the modes r 1 =I and r 4 = l shown in FIG. 23 we find respectively,
ku =
R,.
r,,
R,
r,,l
1,
...... (143)
'
................ (144)
,d,---~..::...-~-~-~_1
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 horizontal 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
k7,2= -ku.2={j {j2
ks.2=k12.2='
Kd(h)
-d
(J
k,~-~1<,1
k s.1-k 9,1--~
a
I
k2,1 =k3,1 =k4,1 =ks,I =k10.1 =0
where
K=EA
_j
(145a)
Unit
stiffness of
all
honzontal
Un!l
sttffness of
all
vertrcal
Untl
&ttlfness of
all
dtagonal
bars
bars
bars
'Ka
')(."
'Kd
23
and
I
kvu =f..,-'
1)11
See Section 4
24
';=04
t;=
-0 4
(Q, =- o-oo62/Q)
Ii
RJ
...... 0
q2
kp; 3 --r
} pn
~,I
i--
1I
00169
--000615
K =-~-000615
0548
'' a -0117
0406
L
where a is given by Eq. (104a). Hence,
Q=K.q=K.(O I 0}
or
-l
-o 117
, .. .. ,
kvmm =~,--
pmm
.................... ( 152)
0406
0218
.......... (l06a)
_j
r-
Io
R=BQ~-. I
II
L_
Ill
I
I
-000615 I
1 0548 I
I 0406 I
.J
.J L
05 --1
-05 05
.J
o-2Dl/a
kvmm
Q3 mod
Ill,= oos/al
............ (151)
I I
II o
_ _j
a-
K"=
f----- a - - - + - - a
A
-0077/a
954/a I
I +O_
-0138/a i
.J
Each of the three columns of the intermediate expression represents obviously the R components of Q1 , Q2 , Q3 respectively. We can check now
the previously given result for r from
r=FR
where F is given in Eq. (104).
Next we derive a general formula for the stiffness K of a structure consisting of a finite number of simple elements. The expression given is the
matrix formulation of Eqs. (146) and corresponds to the flexibility matrix
F of Eq. (126). Since the analysis follows closely the arguments on page 21
we need present here only the outlines of the proof.
Consider again an assembly of s structural elements joimd together
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
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
element. Let a. be the matrix, in general rectangular, which transforms the
displacements r into the true strains v. of the element. i.e.
(153)
Then
s.=k.v.=k.a.r ........................ (154)
is the matrix for the forces (moments, etc.) applied on the element due to
the displacements r. The internal force (or stress) matrix S of the aggregate
structure is now given by
S=kar .............................. (155)
where
S={SaSb ...... 5 0 S,}
(!56)
and
a={a.ab ...... a ...... a,}
.................. (157)
k is the symmetrical diagonal partitioned matrix,
...... 0
0
......
TABLE I
Method
Force
Displacement
S = bR
k=
0
k ......
................ (158)
of
r= 6'v
S)
Flexibility of complete
...... 0
......
k.
_j
Applying now the principle of virtual work, taking the internal forces S
and external forces R as the true state and selecting as virtual state the
internal strains corresponding to unit displacements ' I = I, r 2 = 1, ...... ,
rm = 1 respectively we find
R=a'kar ............................ (!59)
where a' is the transpose of a. Thus, the stiffness matrix K of the compound
structure is
K=a'ka .............................. (160)
Eq. (160) may also be written as
K=~a'.ka
........................ (160a)
g
Since the virtual strains need only satisfy the compatibility but not necessarily the equilibrium conditions we may select for the virtual states a
simpler matrix a which satisfies only the former. Eq. (160) becomes then
K=a'ka ............................ (l60b)
However, the application of a possibly simplified matrix a is really not
required in practice. As mentioned on page 23 the stiffness matrix K is
best calculated for all degrees of freedom at the joints, yielding very simple
matrices.
The configuration of the elements of the compound structure is said to
be in parallel in Eq. (160) since the assembly condition is expressed by the
matrix a which derives from conditions of compatibility. Thus Eq. (160)
may be regarded as the most general formulation of the stiffness matrix
for a structure with constituent elements in parallel. It is immediately
apparent why Eq. (l49) which expresses the stiffness matrix for generalized
displacements must have the same form as Eq. (160). In the first case we
derive generalized displacements from single displacements and in the
second, internal strains from external displacements. In both applications
this entails a linear transformation matrix which, however, is a square
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
members.
Eqs. (153) and (159) show that there is a most illuminating parallel
development to Eqs. (121) and (125a). Thus, if the internal relative displacements (strains) v derive from the external displacements r with the
relationship
v=ar .............................. (l53a)
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.
Addition
(Spe-i:ial
r= r
.
F.
Stra~n
of
Stress
on
Assembly
elements
elements
Stillness of
s
k
complete
structure
K = a'ka
Addition
( Spci.al
aumbly)
F. =
V>
(for strains
Flexibilities
of
uriu
Parallel
Stiffness of E'lements
structure
6'f b
F=
Q :.A'R = Kqq
R=aS
Flexibility of elements
(for stresses
Force
v = ar
elements
KA
Generalized
ASSE'mbly
Stress on elements
Strain
Kq=~
Generalized
q: B'r = ~Q
Series
Generalized 1 Stiffness
~ Kq=l= Kq~
r=Aq
1
l
F~exibility ~ A'B=l=B'A
BFB
Generalized
'Generalized Displacement
Generalized
Force +
Force
~=
I
Stiffness
FK=I=KF
R:BQ
Generalized
Displacem~tnl s
Displacement
+
Displacement
Generalized
of
Method
Flexibility
...... 0
Forces
-r
--~
K.
Of
Still nesses
parallel
aaembly)
.riJ?-
We conclude,
II 00 01 01~
1 0 0
........................ (163)
_j
25
~~
r:.~
c
'i.~
B
TI
rs~
0
j'3
Irs
B'
D'
f'~
'i"
?di
f'3
"""'
c
f'S
Eq. (161) applies to any compound structure in which the stiffnesses are
defined for at least all common degrees of freedom associated with the
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
(161) is, however, still tru~ if we define the stiffnesses of the upper beam
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
previously. Ku is still only definable for vertical displacement, the corresponding entries associated with r 2 , r4 and r6 being zero. Naturally, we can
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
of two arbitrary structures to give K= KI + KII Note that at a joint point
like (2) we must define the stiffnesses for two displacements, say the x and
y-directions.
Formula (161) may, of course, also be applied in obtaining the stiffness
matrix of the compound cantilever consisting of elements a and b, FIG.
(23). Again, we must define the stiffness for all common deflexions and
slopes at the joints, assuming the E.T .B. to be true and the shear deflexions
zero. The total stiffness K is then
K=K.+Kb .......................... (t61b)
where the elements of the split matrices may be found from Eqs. (143)
and (144) for the displacements r 1 and r4 and similar equations for r 2 and
r 3 Thus,
I
I
El
0 -6[2 0
further by comparison of Eqs. (92) and (161). The first shows the case of
additive partial flexibilities for series assembly and the second, the case of
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
example 5b. *Thus, the stiffness k, of the bar corresponding to unit elongation ~I,= I is,
-k _ (EA), -~
k, -- r -- I, -I,
El
-6[2 0
~.
. ......... { 168a)
[3
[3
61
[2
4~!
I
2/
I
6EI
21
I
41!! I
I I
[2
..... 0
L:?
sin
I
I
I
I
.... Kn/1, I
_j
I kn
e, I I k
_j
Y>
kxy
kyy
........ (169)
_j
_jb
26
-6/
[2
El
El
El
12 [3 -6[2 -6[2
_ 12 /
13
El
6[2
0
L
6[2/
-1
..... 0
..... 0
"\'Kr
e . eI
L.JT; cos , sm , I
.................... (168)
and
_ja
12gj -12/
Hence
................ (164)
4/ 0 I
I
I
ol
0
. . . . . . . . . . . . . . . . . (167)
where K,=(EA), is the stiffness per unit length of the rth component bar.
The transformation matrix a, for displacements in the x- and y-directions
127:
0
'6~
K.=
+/
J---X-~~~ Jt--~
~
~
B'
1/T/~Ti = 1/l~r
t r,
~
~
!'5
tt;
We mentioned previously that the simplest method of calculating stiffnesses is to define them for as many degrees of freedom as are necessary
to obtain simple deformation patterns of the elements of the structure.
Having calculated such a stiffness matrix it will become necessary to
'condense' it-i.e. to refer it to the smaller number of displacements in
which we may be interested. This changes, of course, all the stiffness
coefficients, but the necessary analysis is easily arranged in matrix form
for automatic computation. Let the original stiffness matrix be of the order
m x m and denoted by K We want to find the matrix K referred to p-directions only, where p<m. We have,
R=K r
where
0
R={R 1
.....
} .... (171)
in which we write first the p-directions required for the condensed matrix
K. Ko may be expressed as a partitioned matrix as follows,
See p. 10.
X,
........................ ( 172)
J......................
I
R/' Kr/
R11 =0
(173a)
C.
"give" at support
-J
Th~
ao
due to
a 0 + ~a;X;
i=-1
........................ (176)
C, =
!! AI
b ( 1 + ll)
a,
/'\
~~~X,:l
N,
X, I
....
C,A l: N,AI
010
The 8;, are, of course, the influence or flexibility coefficients of the basic
structure for the directions of the redundant forces. We use here the symbol
8 for these flexibilities since it is standard in the literature. To calculate the
8's we apply the unit load method of Section 6D.
Thus, using again the abbreviations
+ ..... .
aE=auExx
+UzxEzx
a7] ~--au7]xx+ . . . +azx7)zx
............................. . (177)
} .................. (3)
................ (178)
where ai, E; (ako Ek) are the stresses and strains corresponding to X;= 1
(Xk=1) and a., Eo are the stresses and strains due to the applied loads.
Eqs. ( 178) reproduce, of course, merely Eqs. (84) for the flexibility coefficients. The total initial strains 7J imposed upon the basic system may be
separated into thermal and other strains
7J.x=a0 +T}xxo , T}xu =7]xuo (179)
where T}xx etc. are initial strains due to say lack of fit, 'give' at the supports.
The effect of the latter upon 8; 0 is best considered separately and expressed
in terms of the imposed changes of length (rotations) and 'gives'. Consider,
for example, the singly redundant framework of FIG. (28) and assume that
the manufactured length of the bars exceeds the correct length I by 1:11. Let
also each bar be subjected to a different thermal straining a0. We assume
furthermore that the intermediate support gives or sinks by the amount /:1.
As redundancy we select the force XI in the bar (1, 2) and denote by N_9
and NI the (tension) forces in the bars of the basic system due to the appliea
loads and X1 = I respectively. The loading case X1 = l, with the corresponding force applied to support C by the structure is shown in FIG. (28).
We find immediately
NI2f
8u = ~ AE ............................ (180)
27
Applying now the unit load method to the state XI= I and the true total
strains (e.+'l) and displacements in the basic system we derive (see FIG.
(28) and Eq. (7la)),
0 NIN
.
h (
l810 =~NI(e.+'7)/=~ AE l+~NI(a0!+~l)+a;: I +11 ~ .... (181)
a)
Naturally, we can alternatively deduce by kinematical reasoning the contribution to 810 of the initial elongations a0 and ~/ and the 'give' ~
However, the unit load method yields the results much more conveniently
and systematically.
More general formulae for the 8 -coefficients are given further below.
The condition of consistent deformations at the cut ends of the n redundancies in the actual structure or application of the unit load method
yields the following n equations in the n unknown X.
811X1 +812X2 + ...... +81;X;+ ...... +81 n X.+810 =0
821 X1 +8 22 X2 + ..... +8 21 X;+ ...... +8 2 .. X,.+8 20 =0
t 11
ft2
0 ...................... 0
122 0 .................... 0
a=
ai1
1;1 t; 2 t;;
an 1 .. ani . . . . . . ann
rl-
111 ; l 112
1/au
I -a,.!au
I 0 ................... 0 I
I
o........ o I ....
-a.1 (a 11 0 .................. 0
(188)
l
_j
which square matrix has, but for the first column, unit diagonal and zero
cross-elements. We find,
l
0
ht2 .bli
.bln
b22 ...... b2i .. . bz ..
bn2bnibnn
.... (l87a)
- - --- -- - - --- --
_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
I
0 .................. 0
0 .................. 0
_j
28
_j
o...................... ~I
1
0
Inn
.............. (187)
b=M 1a= [
.. (184)
li
To eliminate in the first column of a all its elements but au and to reduce
the latter to l we premultiply a with the matrix
0 ...... 0
..... . ain
-- --- -------
T=
aii
0 ...... 0
~b,.2(b22
0 ................ 0 1
_j
.... (186a)
We obtain
1
0
0
L
where
and
c23
...... . C2n
----0 0
C;a
- --- c,.
C=
c2 ~~:=b2 ~~:/b22
0
1
............ ci,
3 ........... . c.,.
I
I
I
I
I~ = 1/El
(187b)
_j
9
for i*2 .............. (18 a)
b;2b2k
c;k=bik-~
X,
The procedure will by now be clear. Thus, at the (g-1) 1h = f1h elimination step we obtain a matrix g of the form
I
I
1\
.............. (187c)
;\
g=
1
0
0. 0000.0
l. 0000.0
............ (187d)
h=M.g=
where
__l
Kkg"'
hm=K.m Ig and h~~:m=g~~:m--g--
fx.
-!'/-
_j
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
next step, i.e. to obtain the h matrix, we premultiply with M. which has a
gth column
(188b)
{ -gl.fg -g2ofg. 000000.1/g. 0000000-g,..fg }
resultThe
cross-elements.
zero
and
elements
and otherwise unit diagonal
ing h-matrix is of the form
fx,
Y,
1/n,.,.
__l
(a)
(b)
X,
I~"
(c)
Y,
-.,_x%:
(d)
Fig. 29.-Continuous beam. Good and bad choices of basic system and
redundancies
29
D=
1:
I
3u 312
I
9
24
9 4
L
_j
A remarkably bad choice since 312 > >322
(b) X1, X 2 taken as the reactions at the intermediate supports Band C.
Then,
_aaq, ~~ 8
D- 18
7l
7 8
_j
Still a bad choice since all 3's are of the same order of magnitude.
(c) Y1 and Y2 are generalized redundancies formed from system (b) by
the transformation matrix
B=
Ill
-0~
-05
(b)
(a)
System
Y1 = 1:
"'1
System
r,
=1 :
M, Y,
System
r, = 1:
M, Y,
t,. t, = [
t13 = t" =
.................. (198)
t,=
_j
Orthogonal redundancies
Non-orthogonal redundancies
,.r , =
:Z'
r:r
ds}
s1
N, = coso&,
s2
= - sino&
N, =- sin-3,
S,
= - cos-3
N1 =
=1
0,
ds
f EI
Y,Y, ds
1
1 )
sin-3 cos-3 (&A' - EA ds
D=~~
I
L
tl
_j
'ik
a~~-~
~ii
30
Cy3
1 4
L
_j
The improvement over (b) and (a) is immediately apparent.
(d) X1 and X 2 are the bending moments at supports B and C. This
choice of unknowns is statically identical to Y1 and Y2 of (c), but the basic
system is different. The D matrix is
I 4
I 0
Ya
Y2-
cosO -sin~
-sin 0 -cos()
II
.............. (202)
_j
L
In practice it is nearly always worthwhile to find the elastic centre and
eliminate two of the cross-flexibilities but determination of the principal
axes, unless possible by inspection, is not usually worth the trouble. A
further point is that the elastic centre concept is still valid if deformations
due to normal and shear forces are included whereas the principal axes
requirement becomes more complicated.
Interestingly enough this solution was first given by Mohr* more than
seventy years ago, but it appears not to be universally known, for otherwise
it would not have been necessary to rediscover it so many times. A more
recent derivation and application of an orthogonal set of redundancies
(in general infinite) is the system of self-equilibrating eigenloads developed
by Argyris and Dunne t for their general theory of tubes in bending and
torsion.
Particular forms of the 3;k and 3; 0 coefficients
We return now to Eqs. (177) and (178) for the 3-coefficients and give,
following our expressions (180) and (181) for a pin-jointed framework,
some further explicit formulae for more complex structures.
Stiff-jointed plane framework. We assume the Engineers' theory of
bending stresses to be true and introduce the special notation:
N., s., M. normal force, shear force, bending moment in basic system
due to applied loads.
N;, S;, M; normal force, shear force, bending moment in basic system
due to X;= 1 where i= 1 ton.
s
coordinate along axis of beam.
temperature at neutral line of cross-section.
0
D.0/h temperature gradient across depth h of beam ; positive if giving
rise to thermal bending strain of the same sign as that due to
positive bending moment M.
0. Mobr, Z. Architek u. lng. Ver. Hannover, Vol. 27, 1881, p. 143. See also the 1enenlization
and tabular presentation of this method in: J. H. Aqyris and P. C. Dunne, Structural Analysis
(Vol. I of Handbook of Aeronautics), Pitman, 1952, Table 17.1. Both the deformations due to normal
and shear forces are included in the latter analysis.
t See J. H. Aqyris, P. C. Dunne, 'The ~ral'lbeory', ete., J.R.Ae.S., Vol. LI (1947), Feb.,
Sept., Nov. and Vol. Lm (1949), May, June.
tx,
r~l
r '
X1 ~~X 1
o..-
-~ ::-=t::A.
I
I
II
A.
-1
N
tx,
..!..
a
N, -h
:~ .
1
'
a~
-o ..
JJ::h=_
-1C::::====T=:r
.
1
N
a
=-
n~c h
A.~
1-
c,IRJ
:fc.
J
'1
I~
1
X2 1
X1
X,
A0
A.
Manufactured
shape of panel
X3
Redundancies
Contributions to 0 10 due to A
0,0 = fe,Adf,
=-f-A,
o,. = 0
Fig. 32.--3; 0 due to initial shear strain arisin1 from incorrect manufacture
c ni cti
Fl1. 31.--3~ 0 due to initial strains In ri1id jointed frame and manufacturin1
errors and 'live' at supports
JN;Nk
EA ds+ JS;Sk
GA'ds+ JM;Mk
-ads
.................... (203)
Also from Eq. (177) or directly from the unit load method,
a~0 +l::C;~
S;.= JN;N.
EA + N;a0ds+ M;-h-
................. . (204)
... at
tJ
Correct shape
of panel
c, = T1 . c, - TI . c, = o
C11
A. A1
L - _I
life,.= -(a+ b)
LOADING X2 I
LOADING X, I
r - -,
AL :_ ----- ~
A1
a
c.. =h
\~!= :
o.,
;-~::.::.:~~--~~A.
1I
N,
LI
S;k
=i:J J[
Jtdxdy
.................. (205)
............ (206)
The immediate application of the above formulae is to major aircraft components like wings and fuselages. Their matrix formulation is discussed
further below.
FIG. (32) illustrates how on a thrice redundant beam with shear carrying
web the contribution of a prescribed displacement ~ to S;. is calculated.
~ is in this case an initial shearing displacement of a panel due to error of
manufacture.
It was assumed in all our above considerations that the basic or cut
system is statically determinate. However, nothing in the theory so far
given restricts us to such a choice. We can select in a structure with a total
number of redundancies n a statically indeterminate basic system with
(n-r)(r<n)redundancies by'cutting'only rredundantmembers. Equations
of the type (182) can then be written down for the cut r unknowns, the
corresponding S;kcoefficients being still defined as in our previous analysis
in the basic system. In fact, to calculate the S;k we may apply again
Eqs. (I 78). Similarly, for S;. we may use Eq. (177) if we substitute e'l for 7J
where e'l is the true strain in the basic system due to the prescribed initial
strains. This modification is necessary since the basic system is now redundant and the imposed initial strains 7J are not free to develop. However,
both formulae for S;k and S;. may be simplified considerably if we remember that in the unit load method from which they derive (see Eq. (7la)) only
the strains must be the true ones for the system considered-in the present
ease the redundant basic system. The stresses corresponding to the unit
load may be any suitable statically equivalent stress system and may hence
31
m,H
nfa So
m 1, 11 1, s1 (i =4 or 5)
the normal force, shear force and bending moment in the new basic system
due top and X 1 =I respectively and assume that they are known.
The stress distribution in the actual structure is then found from
M=m 0 +m 4 X 4 +m;X5
I
N=n 0 +n 4 X 4 +n 5 X 5
~ ..... (209)
S=s0 +s 4X4+s 5 X,;
j
The equation of compatibility in the unknowns X 4 and X 5 are now of the
form
'uX4+,4aX5+,4o=--O
} ................ (210)
';4X4 +';;;X5 +'o=O
where we write ' instead of 3 to stress that these coefficients are different
from the correspondrng 3's in Eqs. (207). To find the ''s we apply Eqs.
( 177a) and ( 178a) in the new basic system and remember that the virtual
stresses due to X1 = l may be selected in a statically determinate system.
Thus, omitting the contributions of the normal and shear forces for
convenience of printing, we find
m 1mk
JM;mk
Jm;Mk
r
,,k= -----erds= El ds= ~ds=~:,kt .......... C2ll)
.......................... (212)
If, in addition, any initial strains YJ are imposed on our structure giving
rise to moments m'l in the basic system the corresponding contribution
to ''" becomes
r
Jm;m'l
<..to=
El ds= JM;m'l
~ds ....................... . (212a)
The effects of the normal and shear forces may be included without difficulty, the terms following immediately from Eqs. (203) and (204) and the
arguments leading to the contribution of the moments.
x,-
'
X,
f x;
~x~~x:~
'~
(b)
(a)
Statically
basic
determinate
Rdundant
basic
syste-m
sys1Pm
3,. ~-i:
............ (205a)
and a similar formula with a and a interchanged. Also,
=iJ I[
3,,
.......... (214)
t Lo..-. cit.
cut is not always applicable. Thus, in the case of the tube in the last
paragraph, when obtaining the basic stresses we do not actually cut any
redundant member but rather select the engineers' theory of bending direct
stresses and the associated shear flows as statically equivalent to the
applied bending moment. In general, all members are found to be load
carrying. The stresses a. only exceptionally satisfy the elastic compatibility conditions-for example, due to warping varying parallel to the
axis of the tube and to rib deformability. We may give some physical reality
to the basic structure in which a. is true by releasing in the actual structure
the warping restraints at every cross-section and by assuming the ribs as
rigid; the former idealization does no doubt require a complicated mechanism for its realization. The idea of selecting a. as any suitable statically
equivalent stress system without reference to actual cuts may, of course,
also be applied to frameworks.
The use of a redundant basic structure is important also from a further
point of view. Consider the wing of FIG. (34) the main portion I of which is
swept and attached to some root structure II. It is assumed that the ribs
of I are taken perpendicular to a longitudinal axis approximately parallel
to the spars. The necessity may arise of investigating alternative angles of
sweep obtained merely by changing the root structure II. Thus, in FIG. (34)
we show two alternative arrangements. In such instances it is obviously
advantageous to have as much as possible of the stress analysis in common
in the two alternative calculations. To this purpose we release at the junction ofl and II all redundant forces or groups of forces X 1 to X,. appearing
there. The tube I is then connected to tube ll by some statically determinate arrangement and this new structure is taken as the basic system. The
scheme of the analysis is as follows. Analyse first Tube I for all external
forces and also for X 1 = I to X n = 1 respectively. This investigation involves, of course, the solution of a highly redundant system. Irrespective,
however, of the form tube H takes the analysis of tube I remains unaffected.
Next we analyse tube II for the external forces (which include the statically
determinate reactions P from tube I), and also for X 1 =I to X n = 1 respectively. Again this may involve the solution of a redundant problem.
Finally we can write down equations of the type ( 182) for the unknowns
X1 = 1 to X n = 1 and note that the S-coefficients are in each case
8=81 +8u . ........................... (215)
Hence if we change structure II only S11 but not S1 is altered-an obvious
advantage. The solution of Eqs. (182) yields ultimately the stress distribution in the actual wing.
33
>;-I
L
0 -a/h
a/h
afh 0 0
5 6
(218)
-d/h
-d/2h-d/2h 0
7
0
1/2
IO
11
and
0 -afd
~-afd
.,;= 0 -a/d 0
L
(219)
-a/d 0
l
.J
-hfd-hfd 0
0 -h/d-2h/d
-1
-1
0 0
5 6
IO
11
.J
~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.
column matrix of initial strains (displacements).
H
rectangular matrix of forces (moments) C1 (seep. 31).
c
B,C
B., C.
1
d
h
0
~. ~..
L1,, 11,.
L=[L1,], 1=[11,]
P, Q
effective longitudinal and transverse flange load respectively.
k., k.,, k1
U, W
8o
~.
ii1
X,
C=i,.'k~ =~'k~
Ce=~'kaor=~'kaor
34
--
Basic system
twice redundant )
structure
Complete
R/2~ (12
~J[~
x,
Cantilever
and hence
(224)
These are the required equations in the r unknown X;, and are, in fact,
equivalent to formulae (182). The symmetrical square matrix
D=b1'fb1
(to use the notation of Eq. (182)) is the flexibility matrix for the directions
of the r unknown X; in the basic system. Also in the notation ofEq.(l82a)
D0 =b1'fb0 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . (225a)
Eqs. (224) are the most general formulation in matrix algebra of the equations for the r unknown X; in a structure with a redundant basic system.
Solving for X we find
X=- (b1 'fb1 )-1b1'fb0 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . (226)
Substituting (226) in (217) we determine 5 solely as a function of the R's.
Thus,
(227)
X.
elements
beam
Consider the symmetrical fuselage ring with transverse beam and central
load R shown in FIG. 36. As in page 32 we select as a basic system the
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
the closed ring. It is assumed that we know the stress distribution and
hence the flexibilities due to the pairs ofloads applied to the ring (FIG. 36).
The basic system is thrice redundant but due to symmetry X 3 =0.
The load transformation matrices b0 and ~ are
I
_,boB
bubOR
L _j
(231)
where
(227a)
Naturally, it is always possible to substitute b1 ' for 61 ' in Eqs. (224) to
(227a). However, the introduction of the statically determinate matrix ~
when the basic system is redundant simplifies the calculations, often
considerably.
We can apply now Eq. (227a) to derive the flexibility
1/2
boa=
bOR=
where
. . . . . . . . . . . .
(230)
r,~~l
(231a)
1/2
L _j
_j
and
-l~l
of the actual structure for the m points and directions of the applied loads.
Eq. (126) gives
f=b'fb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (126)
Forb we may use
b = b0 or even simpler b = b0
We obtain
F = b0'f[b0 - b1 (61 'fb1 )-1 b1 'fb0 ]
or
fo=60'fb0 =b0'fb0 . . . . . . . . . . . .
is the flexibility of the basic system for the loads R.
element
Ring
bl-
0.
(232)
blR
_j
where
I
I1
btB=
lo
Ol
IO
I
01
_j
b1R=
1
0
0
~~
0.
(232a)
QlI
_j
Note that in the present case b1 = b1 since the loadings on the two elements
of the structure are statically determinate.
The flexibilities of the elements for the forces and moments may be
written as
35
~u '~lI
[2
I
EI -2EI O
[2
fB=
-2EI
0
and hence
fu
[3
3EI 0
I
0 EA
f=
fal /82
faa
~B
0
ol
..J
........
(233)
......................
(234)
fR
..J
C=
r----------------------
.......
(237)
Cross
s~ction
A-A
..J
L
the n rows of which are the forces due to X;= 1, applied by the structure
to the foundations in the directions of the gives fl.. Eq. (236) becomes now,
. . . . . . . . . . . . . . . . . . . . . . (236a)
~'f~X+c'H=O
If we select a redundant basic system we cannot derive the elements of H
immediately from the prescribed initial strain since the latter are not free
to develop in a redundant structure. In this case unless we have the necessary fnformation from previous calculations we must first analyse the
basic system by the method of the previous paragraph. Having found the
column matrix H the r equations in the r unknowns take the form
b1 'fb1 X+b1'H=0 . . . . . . . . . . . . . . . . . . . . . . (236b)
where we may write ~ for ~ since the basic system is now redundant.
The systematic solution of (224) and related equations was discussed on
page 28 but there are a few further points arising in practical calculations
which are best investigated here. Thus, we mentioned on page 20 that it
is often possible and justified to neglect certain part flexibilities of the
elements; for example, in a ring analysis we can usually ignore the direct
and shear flexibility. This applies not only to the evaluation of the external
flexibility F but also to the determination of the internal redundancies X.
We write now the D and D0 matrices in the split form
D=Da+Dband D0 =Doa+Doo . . . . . . . . . . . . . . . . . . . . . . (238)
where the suffices a and b refer to the two flexibilities into which we separate the total flexibility of each element. An approximate solution X a to
the unknown column X is then obtained by ignoring the flexibility b.
Then,
DaXa+Doa;;=O or X 0 =-Da-1 Doa . . . . . . . . . . . . . . . . . . (239)
Occasionally we may require subsequently the correction x to Xa to find
the true column X,
X=Xa+x . . . . . . . . . . . . . . . . . . . . . . . . (239a)
Substituting (238) and (239a) into Eq. (224) we derive easily .
X=-D-1 (D 00 +DbXa)
36
(240)
where the number of rows in the matrices with suffices I and II is p and
~ . . . . . . . . . . . . . . . . (240a)
DuiXI+DuXu+Dou=O j
We split next the column matrix Xu into matrices
Xu=x+y ......................... . (241)
where x satisfies the equation
Dux+Dou=O or X= -Du-1 D0 u.............. (242)
Hence
(242a)
. . . . . . . .
Du 1 X 1 +Duy=0 or y=-Du-1 Du1 X 1
Substituting for x and y into the first of Eqs. (240a) we find
D1 X 1 +D u/[ -Du-1 D0u-D11- 1 Du1 X 1 ]+D01 = 0
or
X1=- (DJ-Du/Du-1 DuJ)-1(D0 I-Dm'Du-1 Dou) ..... (243)
from which we deduce y and hence X. Eqs. (243) are identical with the
elastic compatibility equation for a basic system with n- p redundancies.
The matrix form (224) of the equation of compatibility is particularly
suitable to illustrate the transformation (seep. 29),
X=BY .............................. (195)
when the equations are ill-conditioned. Thus, by substitution of(l95) into
(224) we find
~'fb1 BY +b1 'fb0 R=0
and premultiplying by 8'
B'b1 'f~BY +Bif1 'fb0 R=0
or
b 2'fb2Y +~'fboR=O . . . . . . . . . . . . . . . . . . . . . . . . . . . (244)
Unstiffened
cut- out
bay
12~CD~~~~--L
Unstiffened
cut-out
Longitudinal flang~s
continuous at junction
Longitudinal flanges
interrupted at junction
~ = 12
N=3
Fig. 39.-Geometry of typical bay for determination of number of
redundancies
37
the tip is assumed stiffened by ribs at least at the root and the tip. These
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
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
at all longitudinal flanges, the total number of redundancies is
. . . . . . . . . . . . . . . . . . . . . . . . (247)
n=~ [@-3)+(N-l)]
Root. with
rib
Tip
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
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
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
. . . . . . . . . . . . . . . . . . . . . . . . (247a)
n=a[f3+N-4]
where a=number of bays. If the sheet cover is missing between two adjoining 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 :
n1 = I;[(~-1 )+(N-t)J
bays
na= I;(~-2)
bays
stiffened
stress
panel
system
X: 1
(Flat panel)
(248)
where f3 is defined as in the case of the wing. When M fields are removed
without being replaced by stiff-jointed frames the number of redundancies
reduces by M.
The next step in our investigations is the discussion of suitable selfequilibrating systems which may be chosen as redundancies. Consider
first the simple case of a rectangular flat panel shown in FIG 41. For the
redundancies we may select n systems of the type X= 1 illustrated in the
38
Rectangular
Self.equillbratmg
of which n 1 applies for the lift and torque loads and n2 for the drag loads.
If all cells are closed, with the same number N in all bays and effective
flanges are only placed at the corners of the cells, then
f3=2(N+1)
and from (247a) the total number of redundancies is
n=a(3N-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (249)
which formula again assumes that all the flanges are held at the root.
Of considerable importance in modern aircraft structural practice are
the multispar syst~ms with .few, often only two, end ribs. A typical wing
of the latter type IS shown m FIG. (40). To analyse this structure we subdivide it into a number of bays whose length should not exceed say five
times the spar pitch. Effective flanges will by virtue of our idealization
process be acting at the junction of these bays although no ribs are provided th~re. For such a system the number n of redundancies when there
are no cut-outs in the sheet, when all spars are continuous for the full
length of the wing and all flanges are held at the root, is given by
n=(N-l)+l+a({3-4)=N+a({3-4) . . . . . . . . . . . . . . (250)
The last system to be considered is a flat panel which is of special importance for diffusion investigations (see FIG. 41). It is assumed built-in at z=O
or held with prescribed displacements and free at the other three edges.
Here the number n of redundancies when there are no unstiffened cut-outs
is simply
II=~ (/3-2) . . . . . . . . . . . . . . . . . . . . . . . . . . (251)
bays
Longitudinal
flange
loads
Transverse
flange
toads
figure. All information as to flange loads and shear flows is given there.
The corresponding equations (182) or (224) for the unknown X; are easily
seen to be reasonably well conditioned. Naturally, we can further improve
the conditioning by introducing group loads
X=BX . . . . . . . . . . . . . . . . . . . . . . . . . . . . (252)
wh~re B is a suitable square matrix. We do not enter at this stage into the
chmce of B but hope to discuss these points in Part III. When the panel is
symmetrical about its middle line it is preferable to combine the %-systems
into symmetrical and antisymmetrical groups.
In a wing structure of the type investigated previously we can describe
three simple types of self-equilibrating internal systems. They are shown in
FIGS. 42, 43 and 44 and denoted by
X=l, Y=1, Z=l
respectively. The first is the generalization of the %-system used in the flat
panel and the second and third may be considered as slightly modified
four boom load systems taken in the longitudinal and transverse directions
respectively. The longitudinal four-boom load systems are applied extensively in standard wing analysis.* The three figures are self-explanatory
and give all flange loads and shear flows associated with the unit systems.
Note, however, that the effect of taper is neglected except that we introduce the true local dimensions in the evaluation of the self-equilibrating
See J. H. Argyris and P. C. Dunne, 'The General Theory, etc.,' J.R.Ae.S., Vol. LI February,
September, November 1947
Multi-web
wing
Multi- web
wing
Area
1 I + I
q =-...:L-..:.1
11 / 2 h, + h 2
q =---11, (h,h,)
str~ss
system
X= 1
p.~!;_,
dld1 h
Self-equilibral1ng
stress
system
Y= 1
-I
___.1iJ{"+,-,.,~"""'
Longitudinal
flange
loads
systems. This ignoring of the influence of taper gives rise only to small
errors as long as the lengths I of the bays are reasonably small and the
angles of taper restricted to the order of magnitude mentioned initially.
The expressions for the shear flows in the ribs for the Y- and Z-systems are
also approximate, being derived for an equivalent rectangular rib. Again
the error introduced by this assumption is for practical purposes insignicant. The conditions of equilibrium for the X- and Z-systems yield a load
in an 'effective' vertical flange at the intersection of the ribs and webs.
This flange load may always be neglected.
We enumerate now the number of X, Y, Z systems, independent within
their own group, which can possibly be applied. We find easily, with the
notation of Eq. (247) that we can use for each bay
({J -4) X-systems, N Y-systems and (N -I) Z-systems
Thus there are more systems than we require, the difference from the
number n of redundancies being obviously linearly dependent systems.
Evidently the
Z-systems
(N-l)
are independent of the X- and Y-systems and hence must all be chosen as
redundancies. To complete the number n of unknowns we may use
({J -4) X- and one Y-system
However, it is preferable to apply more longitudinal four-boom ( Y)
systems since they are better conditioned, and to reduce accordingly the
number of X-systems. Thus, if we introduce all
N
Y-systems
we have to adopt
({J- 3)- N X-systems
The last number reduces to N -I when the effective longitudinal flanges
are only placed at the corners of the N-cells.
It is often advisable to improve the conditioning of the D-matrix by
the introduction of group loads X and Y where
X=B1 X
Y=B2Y . . . . . . . . . . . . . . . . . . . . . . . . . . (252a)
Longitudinal
flange
loads
When the cross-section is singly symmetrical the number of redundancies for lift and torque loads reduces to n1 given by the first of Eqs.
(248). The N - l Z-systems must still be included in the analysis for such
loading cases. If, in addition, we use all N Y-systems the necessary number
of X-systems becomes
f32-l-N
and is zero when the effective flanges are arranged merely at the corners of
the N-cells.
For the multispar wing of FIG. (40) with ribs only at the root and tip
the n redundancies of Eq. (250) may be selected as
( N- l) Z-systems at the root
one
Y-system at the root
and
({3 -4) X-systems at each junction of bays and at the root.
The Y-system may involve a considerable length of the tube and if the
latter is tapered a more accurate estimate of the longitudinal variation of
the flange loads may become necessary.
Having selected a suitable system X, Y, Z of redundancies we can write
down the b1 matrix with the information given in FICiS. 42, 43, 44. To obtain
the b0 matrix we may use any suitable statically equivalent stress system
in the actual or idealized structure, but preferably the former. It was
mentioned on page 32 that it is advantageous to select a basic stress system
which, while being simple, approximates as closely as possible to the true
stress system and reference was made to the method of example (a) of
Section 9. Nevertheless, if the work in finding such a b0 matrix proves
excessive it may be preferable-since the choice of b0 does not affect
the conditioning of the D matrix-to sacrifice the closeness to the true
stress system and to select a b0 as simple as possible. Thus, we can calculate
a b0 matrix for a basic system in which the spars act independently; a
choice differing, in general, widely from the final b matrix.
39
Tr,.nsverse
II ange - . ,
~e
'
1
1
h
/
Shear
field --
D
f
'
Multi- web
wing
d
~-Longitudinal
'
'
flange
//
'
FIG.
q0 (q,) the longitudinal flange loads, transverse loads and shear flows
and
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
o
0
0
I f1 0
ft
q:+t/(Q,Q,)
Self-equilibrating
Transverse
Fig.
stress
flange
system
z=
(255)
lo
fr
0
0
0
_j
L
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
I
I
fa o o o
0
f. 0
0
0
fb 0
0
(256)
fz=
0
0
0
fc
f,
0
0
_j
L
0
0
fd
_j
L
typical sub-matrices being
,I
I
I
I
lo
loads
44~-Multi-web
lb
3EBeb
lb
6EB.b
lb
6EB.b
lb
3EB.b
d.
d.
3EC
6EC
d.
d.
f.=
fb=
(253)
40
f=
q:-1/2Q,
(257)
3EC
6EC
_j
_j
L
L
The flexibility matrices f., fw, fr are diagonal matrices with elements
4>/Gt, 4> .. /Gfw and O.jGtr respectively.
1
1
0
..........
0
O
o~
0
r
I
, fw=
0
0 .......... 0
L
.:J
4>
I 0 .. 0 ___2_0.
0
Gtw
o............ o
L
1
_j
0 .......... 0
(262)
D=~'f~
0
0 .......... 0
L
_j
(258)
where<}), <})w, Q are the areas of the shear fields in the wing surface, webs
and ribs respectively and t, tw, t, are the corresponding thicknesses.
We have now all the information to form the matrices
D=b/fb1 and D 0 =b1 'fb0 R
and can hence solve the system of equations
b1'fb1{X Y Z}+b1'fb0 R=O . . . . . . . . . . . . . . . . . . . . (259)
for the unknowns X, Y, Z.
Finally, we find the true flange loads and shear flows of the idealized
structure from
5=b0 R+b1{X Y Z} . . . . . . . . . . . . . . . . . . . . . . (260)
It will then be necessary to translate the results (260) into stresses of the
actual structure. Finally the flexibility F of the structure at the points and
directions of the loads R may be determined from Eq. (229).
When the number of equations (259) is too large to be dealt with by our
digital computer we may proceed in two (or more) steps by the method
given on page 36. Essentially this introduces into our analysis a redundant
basic system.
If initial strains H are imposed on the structure in addition to the loads
R we have merely to add the column matrix
b1 'H
on the left hand side of (259). Thus the analysis includes inter alia the complete calculation of wings under thermal loading.
A new approach to the problem of cut-outs
We emphasize that our above analysis is valid in the presence of any
kind of cut-out stiffened or unstiffened by closed frames as long as the
overall geometry and idealization conforms with the initial assumptions.
Nevertheless, when we have a structure which is essentially continuous
with only minor unstiffened cut-outs it may be worthwhile to apply an
artifice which avoids the lack of uniformity in the pattern of the equations
inevitably associated with cut-outs. Moreover, it is the idt>.al method of
finding the alteration in the stresses due to the subsequent introduction of
cut-outs in our system without having to repeat all the computations ab
initio.
The method is as follows. To preserve the pattern of equations disturbed
by missing shear panels or flanges we eliminate the cut-outs by introducing
fictitious shear panels or flanges with an arbitrary thickness or area.
Naturally, it is usually preferable to select for the latter dimensions those
of the surrounding structure. To obtain nevertheless the same flange loads
and shear flows in our altered structure as in the original system initial
strains are imposed on the additional elements of such a magnitude that
their stresses become zero. The effect of the fictitious elements is thus nullified whilst the uniform pattern of our equations is retained.
Let the column matrix of the unknown initial strains, in the additional
elements only, be
H
In the new structure (i.e. without the cut-outs) we determine the flexibility
matrix f and the matrices b0 and b1 which we write in the partitioned form
bo=
~~boabo,. ~~
bl =
~~blbl,. ~~
..............
(261)
L
_j
L
_j
where the sub-matrices with the suffices g and h refer to the forces in the
elements of the original structure and the fictitious new elements respectively.
Denoting the column matrix {X Y Z} simply by X and taking the
initial strains in the original structure as zero the Eqs. (259) in the unknown X become,
II
~'fb1 X+ b1'fb0 R+ b1' ~ = 0
I I
L_j
_j
To find now the column matrix H we put the stresses in the additional
elements to zero. Thus, the matrix 5 must be
s~ I~
L
l.............. .......
(26SJ
_j
where 5. are the true stresses (forces) in the original structure. Applying
Eqs. (261), (264) and (265) in (263) we find
I I
I I
I
I
5.
= b. R- I1 bl o-1 b1,.'H . . . . . . . . . . . . (256)
I o I Ibh II
Hence
L.J
L.J
bv,
_j
or
41
The conditions of equilibrium for the actual and idealized systems yield
in conjunction with Eq. (269)
J
I
J ....... .
(270)
(271)
I a,s I
Points
Nodal
a,s
+ Lg,s
L _j
+ L9.3
Structure
Actual
and
=L9.91
Stresses
~~a,a'l
aR3
L _J
I~
_j
._I
+ L9.1f>
I O'zto
I
I + Ls.to
a,to
I a.,s I
I a,g
I
(272)
i O'ztr.ll
a.~ 15
_j
Dire-ct stress
in
...~
f-lange'- , _,.-..;vEEu ~
a..,-
a:.=
Direct stress
in
sheet
E:(E 11+VE:z.)=
u:,
f<a,. VOzz)
Ls.3=
0
I vAa,9
L
l ...
3.9
_j
The matrix L9 ,10 (L9 ,15 ) is obtained from L9 , 8 (L9 , 3) by substituting 10(15)
for 8(3). Equations corresponding to (272) may be written down for any
other nodal point. We see immediately that
o,
(274)
and that,
Idealized
Structure
and
Equivalent
Flange
(273a)
(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
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
5={P1 Q1 P 2Q 2 . . . . . . . P 9 Q 9 . P,Q,}
Loads
..
(275)
(276)
Purely shear
carrying
.
_1_ E.'
_1_ E.'
A3,9= 0 E 13.913.9 A8 ,9= 0 ed8 ,9t 8 ,9 effective ~ange areas due to plate in
pure bendmg
Note that the material of flanges and sheet is assumed to be different.
We have,
~(a.,+va.,)
a,9'=E,'(E.,+vE,9)= ~'(u.,+va,9)
42
.. . .. .. .. .. .
_j
and a,9'=E,'(E,9+vE.,)=
(277)
(269)
where e is the column matrix of the flange strains at the nodal points, i.e.
(279a)
e={EztEst ......... E,gE.g .......... E,,E8 ,}
Thus, once we have determined the effective flange loads the flange
stresses and strains follow from Eqs. (279) and the direct stresses in the
sheet from Eqs. (269). No guessing of effective flange areas is involved in
this procedure but we have on the other hand to invert the matrix L with
2p rows and columns. It is apparent that if we knew the effective flange
areas B., c. at the nodal points we could immediately write down the inverted matrix as a diagonal matrix whose elements are the unit flange
flexibilities at the nodal points. In fact, then
EB.1
0
EC.1
0 ....... 0 _I_
0 ............ 0
EBet~
E~et~
0 ............ 0
(277a)
O ....... 0
0 ...................... 0 _I_
0
EB.'P
0 .................... 0 _I_
EC.'P
~. ~ 1:
I b11 I
b1=
II bl. I
(280)
I blW
I bl,
L
We seek next the contribution of the flange strains e 1 and e 0 to the matrices
D and D0 of the relative displacements S;k and 310 For this purpose we
apply a self-equilibrating unit load system X;=l (which may be also a
Y- or Z-system) and denote by P 13 and P 19 its longitudinal effective flange
loads at the points 3 and 9 and by Q18 and Q19 its transverse effective
flange loads at the points 8 and 9. LetS~ be the relative displacement at the
points and directions of X 1 = I due to some given flange strains E. and E
The contribution to S; of the straining of the flanges (3, 9) and (8, 9) is then
9
11/3
l/6llI
I
d/3
Eia I
Ez9
1/61/3
L :.J
L
_ja.9
d/6
1 1
I sl
EEs
89
j+[Q;sQ;9]
d/6 d/3
L _j
L
_j s.9
+ ....
(282)
3;= .. +{[P;aQ;a]J9,a+[P;sQtsl~.s+[P;9Q;9]~,9+
~[P,loQ;lO]~.lo+[P;lsQtlsl~.1s{::J +
where the I matnces are,
~.9=
315/3
(283a)
111~1~-~
I=
I ~.1
L
IPl
.I.J.9
l'P9
(285)
l;,,p
_j
_j
where the submatrices bot, b 11 give the flange loads at the nodal points. The
rows in bot, ht1 are arranged in p pairs corresponding to the p nodal points;
the first and second row in each pair refers to longitudinal and transverse
flange loads respectively.
Returning now to Eq. (279) we apply it in the basic system for the
external loads R and the redundancies {X Y Z} respectively. We "find
with the notation of Eq. (280) the flange strains
~.a=
_j
L
L
The matrix 19 ,10(19 ,15) is obtained from ~.s~.a> by substituting 10(15) for
8(3). It is simple now to write down in Eq. (282) the terms for any other
pairs of strains (E.E,). We deduce immediately that,
1.,=1.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (284)
and that all lr., matrices are zero when they do not refer to adjoining points.
Moreover, matrices (283a) are typical for adjoining nodal points in the z
and s directions respectively.
Introducing the matrix
_j
:l
I
ll9.a/6
0
(282a)
Note the structural similarity between the I and L matrices. It is particularly pronounced when v=O. The total flexibility of the elements of the
structure is now
I
I
0
0
f, 0
f=
f.
f.,
lo
(289)
0
0
fr
_j
L
where the flexibilities of the cover, webs and ribs are as before.
We find for the matrices D and D 0
D=b1'fb1 and D 0 =b1'fb0 R
. . . . . . . . . . . . . . . . . . . . (225b)
where b 1 and b 0 are arranged as in Eqs. (280).
Using Eqs. (225b) in (259) we solve the problem completely.
The refinement introduced by the L matrix need not of course extend
over the complete wing but may be restricted to the root and other marked
changes of structure and loading. For the rest of the structure it may still
be sufficient to estimate the effective flange areas and to use the simple
form (277a).
(283)
~ N. J. Hoft and Paul A. Libby, 'Rec:ommendations for numeric&! solution of reinforced-panel and
fuselage-ring problem. N.A.C.A. Rep. 934 (1949).
llloe. cit. p. 43.
43
' s---T
h
"
% "
system
Frozen
Displacement
Analysos -
Unknowns
r; and r,
(290)
230-
12
(291)
Consider next the system with free joints and no transverse loading subjected to the loading by the joint-moments
P2l2 2 Pt/1 2
P1l1 2
R1=-M1=+u R2=-M2=u - 12 .......... (291a)
The superposition of this and the previous case yields the true solution of
the given system under the transverse loading. To analyse the second
problem we apply Eqs. (138) which take the form
k21r1 +k2'J!2=R2
~
J
......................
The stiffnesses k;A are easily found as (see also Eqs. (144))
k _ZEit -k
k _ 4/A .. 4Eit
21
12- /1 '
h + /1
11 k
(293a)
- _p2/2:
22=JI+t;+T
(292)
(293)
which assumes that the horizontal beams and supportilll struts have the
constant bending stiffness /1 and /A respectively.
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 positive bending moment when upper fibres are in compression, we find for
the bending moment M 12 at the junction (I) of element (I, 2)
M12=
_P:r- 4~11r1-Z~l1,.2
. . .
This method forms also the basis of the Hardy-Cross or the more general
Southwell relaxation technique in stiff-jointed frameworks.
We develop next the general theory of the displacement method. We
introduce immediately the matrix notation and assume that the structure
consists of a finite numbers of elements whose stiffnesses k, due to relative
displacements at the ends or boundaries of each element, are known. In
order to show most clearly and concisely the striking analogy between the
force and displacement methods we present them side by side in the following TABLE 11; most of the information with respect to forces has been given
previously under C. The complete duality between the two theories is,
of course, a direct consequence of the twin principles of virtual displacements and virtual forces from which they derive most naturally. We
believe that the analysis has not been given previously in this generality.
It includes ab initio any effects of initial strains like temperature, lack of
fit and give' at the foundations. The great advantage obtained in deeper
insight and new theorems and applications by developing the theory on
the most general lines is too apparent to need stressing.
TABLE II
A COMPARATIVE PRESENTATION OF STRUCTURAL ANALYSIS BY THE FORCE AND DISPLACEMENT METHODS
METHOD OF DISPLACEMENTS
METHOD OF FORCES
External forces
Flexibility
Displacements
See also
44
TABLE I
R
F
r=FR
FK=I
(294)
Joint displacements
Stiffness
Forces
See alSO TABLE
R=Kr
KF=I
TABLE II (continued)
Unit Load Method
Given the true strains in a structure the kinematically related displacement r at a given point and direction can be calculated from
Given the true stresses a in a structure the equilibrating force Rat a given
point and direction can be calculated from
(295a)
......................
I. R= JiadV
lr=fadV
v
Cf Eq. (84b)
................................
(295b)
where is a virtual strain system due to unit displacement in given direction. In what follows we denote virtual strains as kinematically equivalent
strains. Kinem"ltically equivalent strains ignore equilibrium conditions.
(See also Section 4 and FIG. 8.)
Cf Eq. (146)
In the relation
S=bR
(296a)
b cannot be determined by statics alone.
Flexibility of structure needs to be considered, entering as compatibility
conditions.
On the oth.!r hand if the internal strains v are known the kinematically
related external displacements may be derived from
In the relation
v = ar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
a cannot be determined by kinematics alone.
Stiffness of structure needs to be considered, entering as equilibrium
conditions.
On the other hand if internal stresses 5 are known the equilibrating
external forces may be derived from
r=b'v
(300a)
where b is merely a statical~ equivalent (virtual) matrix due to unit R's.
R =ca'S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (300b)
where a is merely a kinematically equivalent (virtual) matrix due to unit
r's.
Hence stiffness
Hence flexibility
(301a)
Eq. (300a) is a special form of the Unit Load method (Principle of
Virtual Forces).
FIG. 48a illustrates the matrices b and ii on a particularly simple
example of a singly redundant system.
R =1
'
K=a'ka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . (30lb)
Eq. (300b) is a special form of the Unit Displacement method (Principle
of Virtual Displacements).
FIG. (48b) illustrates the matrices a and i on the same example as in
48a.
FIG.
R=1
r:~--,r~~
-~~.1--..-~tl----,--~-..--b-..
-~
True
system
lbl = 113 3 1 -2 1 3 -3 3 3 -3 3 }
Statically
Trul!
quivatent systl!m
[b)= 113 3 3 0 3 3 -3 3 3 -3 3}
Fig. 48(a).-True and statically equivalent stress systems in singly redundant, pin-jointed framework
)5l5li,'>
(a]=~ { 3
displacl!ment
systl!m
3 1 -2 1 3 -3 3 3 -3 3 }
'
'
"-~~--:;-
Kinematically
l!quivalnt
system-1
lal={ooooooo'j-'t oaf
oo = {13 o o o o o o o o o o 1
45
TABLE ll (continued)
Problem a
Given a set of forces R, determine a set of statically indeterminate forces
X necessary to satisfy the compatibility conditions. Find also the displacements r in the directions of R.
Complete force matrix
(302a) . . . . . . . . . . . . . . . . . . . . . . {R X}
By putting X= 0 we obtain the so-called basic system which is statically
determinate within limits of idealization.
Stresses in basic system
(303a) . . . . . . . . . . . . . . . . . . . . . . So=b0 R
Stresses due to X (with R= 0)
(304a) . . . . . . . . . . . . . . . . . . . . . . 5 1 =b1 X
where b 0 and ~ are obtained from statics alone.
True stresses in actual structure
(305a) . . . . . . . . . . . . . . . . . . . . . . 5=5o+51 =b0 R+b1 X
Strains of elements
(306a)
......................
v=f5=fb0 R+f~X
where
(309a)
Hence
(310a)
True stresses
(296a)
where
(3lla)
. True strains
(297a) . . . . . . . . . . . . . . . . . . . . . .
Displacements r due to R
(300c)
where
(312a)
and
......................
..................
S=bR
v=fS=fbR
r=b'v=b'fbR=FR
Problem a
Given a set of joint displacement r, determine the set of kinematically
indeterminate joint displacements U necessary to satisfy the equilibrium
conditions. Find also the forces R in the directions of r.
Complete displacement matrix
{r U} . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (302b)
By putting U = 0 we obtain the so-called basic system which is kinematically determinate within limits of idealization.
Strains in basic system
v0 =a.or... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (303b)
Strains due to U (with r=O)
v 1 =a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (304b)
where ao and a 1 are obtained by kinematics alone.
True strains in actual structure
v=v0 +v1 =a.or+a1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (305b)
Forces on elements
S=kv=ka.or+ka1 U . . . . . . . . . . . . . . . . . . . . . . . . . . (306b)
Equilibrium condition in actual system at non-prescribed displacements U
~'5=a 1 'ka.or+a 1 'ka1 U =0 . . . . . . . . . . . . . . . . . . . . (307b)
or
(308b)
CU+C.=O
where
(309b)
Hence
U = -C-1C0 = -(a1 'ka1)-1at'ka.or
(310b)
True strains
v=ar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (296b)
where
(311 b)
True stresses
5=kv=kar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (297b)
Forces R due to r
R=a'5=a'kar=Kr . . . . . . . . . . . . . . . . . . . . . . . . . . . . (300d)
where
K=K 0 -a.o'k~(a 1 'ka 1 )- 1 a 1 'ka.o . . . . . . . . . . . . . . . . . . (312b)
and
Ko=a.o'kao is the stiffness of the basic system since we may choose
a=ao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3f3b)
Problem b
Given a set of displacements r find forces R, stresses 5 and strains v
From Eqs. (300c)
(314a) . . . . . . . . . . . . . . . . . . . . . . r=FR
Hence
(315a) .. .. .. .. .. .. .. .. . .. .. . R=F-1 r
(316a) . . . . . . . . . . . . . . . . . . . . . . 5=bR=bF-tr
(317a) . . . . . . . . . . . . . . . . . . . . . . v=f5=fbF-1 r
Once F is known the question of statical determinacy or indeterminacy
is irrelevent in this problem.
Problem b
Given a set of forces R find joint displacements r, strains v and stresses 5
From Eqs. (300d)
R=Kr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (314b)
Hence
r=K-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (315b)
v=ar=aK-1 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (316b)
5=kv=kaK-1 R................................ (317b)
Once K is known the question of kinem:ttical d~terminacy or indeterminacy is irrelevent in this problem.
Problem c
Given a set of initial strains H imposed on free unassembled elements
due to temperature, lack of fit, 'give' at foundations, find stresses 5 and
total strains v when forces R=O.
Total strains of elements
(318a) . . . . . . . . . . . . . . . . . . . . . . v=fb1 X+H
Problem c
Given a set of initial stresses J imposed on elements with frozen joints
(i.e. all joint displacements zero) due to temperature, lack of fit, 'give' at
foundations, find strains v and stresses 5 when displacements r=O.
Total stresses on elements
5=k~ U +J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (318b)
Note that the column matrix U must here include all unknown joint
displacements.
Equilibrium condition in actual system in the directions of U
~'5=at'k~U+a1 'J=0
...................... (319b)
Hence,
and
(32la)
(322a)
Note,
......................
......................
......................
46
X=-(~'fb1 )- 1 b1 'H
(320b)
and
Note,
J=-kH
(32lb)
(322b)
TABLE ll (continued)
Problem d
Assume that we write the total set of forces (including the statically
indeterminate forces) in the partitioned form
l~l
(323a)
_j
Problem d
Assume that we write the total set of joint displacements in the partitioned form
~~l
L
(323b)
_j
in which W is known in terms of rand U. We set now the modified problem (a): Given the set of displacements r determine the set of displacements U necessary to satisfy the equilibrium conditions.
Here in the basic system obtained by putting U = 0 the strains
v0 =aor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (303b)
are completely known although the system is kinematically indeterminate.
a.l
I
Fn/
Fn
I
I
Ku/
Ku,
Ku
_j
where 1(11) is for forces R (X) only and was denoted by F0 (0) in problem
(a). Evidently FlllR=D0
The flexibility matrix F of the actual structure under the forces R is then
K1
I
I
_j
where I(ll) is for displacements r (U) only and was denoted by K0 (C) in
problem (a). Evidently Kmr=C 0
The stiffness matrix K of the actual structure for the displacements r is
then
K=K 1 -Ku/Ku-1 Km . . . . . . . . . . . . . . . . . . . . . . . . (33lb)
Naturally this condensation may be performed in two or more stages
and is then equivalent to the method of problem (d).
Cf Eq. (175)
47
TABLE II (continued)
Elimination and rigidification of structural elements
Assume a set of initial strains, written as column matrix H, in the structural elements to be removed, of such magnitude as to give zero stress in
resultant system.
Write the b and b 1 matrices of the complete structure in the partitioned
form
I
I
I
I
b. I
b=
(332a)
bl- bu
I b,.
bl,. I
L _j
L _j
where the suffix h refers to those elements that are to be removed.
-I
We find
(333a)
......................
..........
Generalized Di~placements
Generalized displacements given by
Generalized Forces
Generalized forces given by
II
IR
(335a)
II
I r I
I
I
I
I
I u I
L_j
lx
.. .. .. .. .. .. .. .. .. .. ..
II
Ir I
IL_j
o\
0.
DX+D0 =0
CU+Co~O
................................ (336b)
where
where
(337b)
(337a)
Then
Then
(338a)
S=bR
where
where
v =ar
........................................
(339a)
R is
(339b)
(340b)
(340d)
(340c)
r is
(338b)
where,
where
48
(335b)
_j
L
The equation for the unknown U is
L_j
I
I
I Ao 0 I
I
I
I
I
I 0 AlI
panels reasonably small. Nevertheless, it is inevitable that the stress distribution derived from an approximate deformation analysis should, in
general, be less accurate than the one obtained from the approximate force
method in the same grid system.
We denote for the purpose of the analysis of the unit panel the displacements parallel to the s and z axes by u and w respectively and introduce also the local coordinate system ~. ' Consider now the state of strain
and stress arising from a unit displacement
va=l ............................... . (341)
Following our assumption the internal displacements are given by
w,,= j( 1-~)
(34Ia)
where the suffix 3 indicates that these displacements are due to v3 =l.
The strains and stresses in the sheet are*:
Uz.a=
'
-GTtJ
Uzz3~~( 1-~)
where '=/(l-v 2 )
Strain .: 1 and load P 1 in flange 8 1
I
E1.a~T
I
P 1 ,3 =B17
................
(342)
(342a)
where the integral extends over sheet and flange. For example, for the
stiffnesses associated with va =I we obtain,
_ E'dt 81 Gtl
k.la ___
3/- I +6d
k":l7
E'dt
k2.1~c --6{
.bL'-=---
lk.,
: v,=
..bz.~_
____,
k4.1-=
Gtl
+ 0 - 6d
E'dt 81
Gtl
E'dt
Gtl
+7J +
0 -3d
and
vE't Gt
vE't Gt
4 +4
k63=+4+4
_ _ vE't _ Gt
vE't Gt
k _
;;a--
We determine first the stiffnesses k;n of the unit panel shown in FIG. 49
for unit displacements in the z- and s-directions at the four corners or nodal
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
practical evaluation of the k;A to introduce simplifying assumptions which
are, naturally, concerned here with the state of deformations. Thus, we
assume that the displacements vary linearly between the nodal points.
Although thi!> idealization offends against the equilibrium conditions its
effect upon the stiffness is not pronounced as long as we keep the unit
I
(343)
kaa=+ 31 +1 +3d
L,,
v, = 1
73 -
4 ' k63=+4-4
(343a)
* Contrary to our usual notation subscripts are used here to denote stresses and strains due to it
displacements.
49
where the suffices s, d, .f indicate the partial stiffnesses for shear strains
and direct strains in sheet, and direct strains in flanges, We find
Gtl
Gtl
"""3d-3d
Gtl
-"""3d
Gtl
Gtl
6d - 6d
Gtl
Gtl
"""3d- 6d
Gtl
Gtl
6d . -- 6d
Gt
~
Gtl
Gt
6d -. -;r-
Gtl
Gtl
3d . . 3d
Gt
-4-
Gt
Gt
Gt
-~--4
Gtl
3d
Gtl
Gt
3d
-6d -6d
Gtl
Gtl
Gt
-TiT-
Gtl
Gt
Gt.
Gt
6d
Gtf
- 6d
Gt
Gt
Gt
.~ 4
4-
4 - 4 444
Gt
Gt
4-4
Gtd
~
sion in the direction of the height of th:' w~b. The corresponding formulae
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
Gtd
Gtd
Gtd
Gtf
(345)
61_1l_M_
Gt
Gt
Gt
Gtd
Gtd
Gtd
Gtd
Gt
Gt
Gt
Gt
Gtd
Gtd
Gtd
Gtd
Gt
Gt
E'dt
&
E'dt
-4--4
Gtd
v-
E'dt
I
vE't
E'dt
E'dt
E'dt
E'dt
E'dt
v-v-31
E'dt
E'dt
vE't
vE't
vE't
E'dt
vE't
vE't
v 't
Gtl
Gtl
vE't
vE't
vE't
Gt
Gt
vE't
vE't
vE't
vE't
vE't
vE't
E'lt
E'lt
E'lt
6d
E'lt
-6d
vE't
vE't '/;
4-3d
E'lt E'lt
3d - 6d-
E'lt
(345a)
6d
vE't
vE't
0
0
---,EB1
vE't
--,-
EB2
E'lt
E'lt
-3d 3d
_I
EC1
EC1
EC1
EC1
-,EB1
y-y
---d
E'lt
6d
E'lt
4 - 6d
EC2
EC2
ECn
EC2
Gt
Gtd
Gtd
(346)
Gt
E'dt
E'tlt
v-
61
E'dt
E'dt
E'dt
E'dt
6/
E'dt
31
E 'tit
E'clt
7J
E'dt
E'dt
61 3/
Gtd
-,-
I)
(346a)
No contribution of the flanges is called for when evaluating the stiffnesses 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
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
antisymmetrical and the stiffness matrices (346) and (346a) may be contracted to 4 > 4 matrices.
We illustrate now the application of the unit panel stiffnesses to the
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
of area B 1. Displacements in the .1 and z directions are defined at all nodes
of the grid formed by lateral and longitudinal stiffening. Naturally the
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
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 displacements defined for the unit panels separately with those defined for
the assembled panel. The complete stiffness matrix is obtained as (Eq.
299b).
K
a'ka
where k is the stiffness matrix of the unassembled unit panels and may be
written in the diagonal partitioned form
(345b)
y-y
Gt
Gt
Gtd
E'dt
vE't
vE't
Gt
-_:r-
-,-. -,-
-i--
E'dt
vE't
-4-4 4
Gt
-..,-
E'dt
y-4 4 . 4 4
4 - 4 - 4 v-Td
Gt
Gt
E'dt
E'dt
Gt
2-2
-T
Gtl
6r 31 y-
E'lt E'lt
3d -3d
k,
k,,
.............. 0
0
k=
.......... 0
k,,
..............
...... 0
.............. 0
. ,0
In assembling the panels of FIG. 50 to form a wing structure the stiffnesses (345) and (345a) may be simplified considerably when applied to
rib and spar webs. Thus, for these cases we can always neglect the expan-
50
Gtl
61 11
E'dt
-3d
Gt
4-4 4-4
E'dt
EB1
Gtl
3d
v 't
E'dt
--,-
Gtf
61
61--4 4 - 4 4
6d
-61-31 61
vE't
6rl
-T
&-v-61 4-4 4 - 4
-y-(;1
6rl
4 - 4 - 4-31--61 31 61
E'dt
Gtf
-~"
Gt
4--4- 4 - 4 61 31-61-31
Gt
---4-
Gtl
Gt
.4
Gtl
k,
(347)
of the reinforcement elements are included with the panels. Thus each of
the areas B or C is split in two and B /2 or C /2 associated with the panel on
each side. For the boundary member, of course, the whole area must be
included with the panel it bounds.
(350)
E'dt 2Git
k2s.1s =ks.ts ="3/- 3d
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 displacements where forces are not applied and to use the condensed matrix.
(See TABLE II.)
Since the terms in the a matrix are either unity or zero their formation is
particularly simple. Writing a in the partitioned form
.................. (348)
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 displacements 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 displacements being defined at all nodal points formed by grid line intersections. 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 reinforcement of the plates can lie along the extra grid lines.
The analysis of such a structure under vertical loads follows that given
under Problems (a) and (b) in TABLE n. Thus, we designate the vertical
displacements as r and take the rotations as the redundant displacements
U.
ments defined for the unit panel g (FIG. 49) to the displacements as defined
for the complete system of FIG. (50). Superimposing the unit panel on
panel g of the complete assembly we find that the directions I, 2, 3, 4, 5, 6,
7, 8 of the unit panel coincide with 15, 25, 17, 27, 16, 26, 18, 28 respectively
and the sub-matrix a, is thus
1 .... 15
16
17
18 ...... 25
26
27
28 ...... 50
0 ..... 1
0 ....... 0
0 ....... 0
0 ..... 0
0 ....... 1
0 ..... 0
0 ....... 0
o....... o I
I
0 ..... 0
0 ....... 0
0 ....... 0
0 ..... 0
1
o..... 0 o o o....... 0
0 ..... 0
1 o..... o
L
0 ....... 0
o....... o I
0
o....... o 1
0 ....... 0
I ....... 0
0 ....... 0
o o o....... o
1 ....... o I
v=aor+a1U
(349)
_j
The equation for the unknowns U, which is here the condition of equilibrium of the moments corresponding to the rotations U at the joints is
(TABLE n, Eqs. (310b)),
or
a 1'ka1U
+~'kagr =
~onlused
K = &o'k&o---&o'ka1(a1'ka1)- 1a 1 k&o
(312b)
51
l2k.
r=K-1R
0 ........................ 0
0
................ 0
(316b)
~covers
ao.
ao={aoo~~o~,
it is apparent that for the cover plates the ao's are all zero since vertical
displacements r can cause no strain in the plates (with U = 0).
k=
aw=
0 ..... 0
0 ..... 0
0 ..... 0
0 ..... 0
0 ..... 1
0 ..... 1
0 ..... 0
0 ..... 0
L
~.=
h/2
0 .... . h/2
h/2
h/2
h/2 ..... 0
0 h/2
0 h/2 .....
16
17 18
12
h/2
..... -h/2
h/2 .....
-h/2 .....
,J
ao'kao=Ko
........... . (356)
~'kao=C
Thus the complete stiffness matrix for the displacement column { r U} is
a 1 'k~=C
(353)
I .............................. .
_I
I~ ~I
(357)
,J
Ko is clearly the set of vertical forces R which arise due to unit r displace-
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
_j
t......
_ 2Ghtw 2Ght,
o.s.s- 1 + d
(358)
Ghlw
ko.s.s =ko.2.s =- -~
(354)
Similarly C is the set of moments arising at the joints due to unit displacements (rotations) U. By using the stiffnesses of the unit panel (or by
carrying out the matrix multiplication a1 'k~) we find for the moments
due to U9 =1
2E'h 2dt EBh 2 E'h 3 tw 2Gh 2lt Ghltw
ce,e=-3-1- +-~-+~+3d+ -3Ca,e=Cts.9=
_j
52
k.
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
the stiffnesses of the unit panels discussed earlier.
The method of formulation of the matrices ~'k~, etc., given above is
probably the most convenient for use with the automatic digital computer
since the various terms in the constituent matrices are reduced to their
simplest and most standard forms. However, it is instructive to consider
directly the components of a1 'k~, etc., and gain some physical insight
into their formation.
We call
11
...................... o
_j
12 .... . 15
(355)
(352)
at,=
0 ...
webs
11
10
k,
.......... 0
I\
I
0 ..... 0
0 ..... 0
0 ..... 0
0 ..... 0
0 ..... 0
0 ..... 0
1 ..... 0
1 ..... 0
10
.... . h/2
6 ..... 9
I
I0
I
II
(351)
. . .. .. .. . .. .
E'h 2dt
EBh 2 E'h 3 tw
---:rr- ---zr-'U/ +
Gh 2lt Ghltw
6d +-6-
E'h2dt Gh 2lt
cu,9=c7,9=~---g
Ct7,g=Cta,9=cs.9=c1.9=
E'h dt Gh lt
---.zr-I2d
2
vE'h 2t Gh 2t
-cts.g=cu,g= -c2,9=cs,9= -g-+-g-
(359)
Finally C is the set of moments arising at the joints due to the vertical
displacements. Again only the webs are involved, and we obtain easily
_
_
Ght,
c12.s= -cs.s=r
_
_
Ghtw
Cts.s= -ca.s=z
J ...................... (
360)
- y
-x
M __ Mx-MuU:r
11 /fy)
x- l-[(/;ry} 2 jf,.J.]'
_M11 -M,,(I.,yjl;r)
. - l-[(/;ry)2/fxf.J
(al)
(a2)
are the effective bending moments for the chosen axes which are, in general,
not the principal axes of the cross-section. Physically M,., M. are the
combinations of M:r, M. which give rise to pure bending strains about
Gx, Gy respectively. We could alternatively restrict ourselves to principal
axes of the cross-section but in practice, unless these are obvious, it is
preferable to use Eqs. (al), (a2). They are not only more convenient from
the computational point of view but permit also the retention of parallel
axes Ox, Oy at all cross-sections of a wing regardless of the change of
directions of the principal axes.
The condition of equilibrium in the z-direction of an element dsdz of a
wall gives,
oa
oq
os+toz=O
. . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(a3)
where
q=ta. 8 =shear flow in the wall.
(,=effective direct stress carrying thickness of the wall (i.e. including an allowance for the stringers).
Similarly, we find from the equilibrium of an element dz of a typical
flange g placed at a web-cover intersection (see FIG. 52),
(a)
Loading
~ I
s,Dxb
(b)
(c)
SM..-
1~
P..M1
1~ ?. H~ l
.V-1
DxM Distribution
By(!;:, ~-q.~
--q.,,+qY+=O
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
Eqs. (a3) and (a4) yield, except for a constant of integration in each of the
are s. and S" respectively. Since there remains one further equilibrium
condition,
To= fqpods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a5)
for torque about 0, the degree of redundancy is (N- I); (the integration
in (a5) extends over all walls and the normal Po is taken positive (negative)
if movement along s leads to an anticlockwise (clockwise) rotation about
0). It follows that the shear flow distribution in a single-cell tube under
prescribed transverse loading is statically determinate once we stipulate
that the direct stresses are distributed as per E.T.B.
For the analysis of the general case of an N cell tube we find it more
convenient to use a slightly different approach. Thus, for the moment,
we prescribe instead of the torque equilibrium condition, the rate of twist
c/>=8'=c~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(a6)
in all cells. The prescribed c/> may be considered as an initial 'give' experienced by the ribs maintaining the shape of the cross-section and is
subsequently determined from the torque equilibrium equation. Our
modified problem has now N redundancies. The basic system is obtained
by cutting the wall in each of the N cells and the unknowns
X[, Xu, ......... , X.~
are then the shear flows at the cuts. They are determined from the compatibility equations,
N
~DMrXr+DMo=-cO
r=J
(a4)
(a7)
.*We denot.e here the warping by the unconventional symbol 6 to apply directly Eqs.(282)intheir
ongmal notation.
53
The shear flow distribution qb in the open tube forming the basic system
is obtained from Eqs. (a3) and (a4). Integrating (a3) with respect to s
and using Eq. (a2) we find
- Dxb -Dub
qb=S.T+SxT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
where
S.
and
s.~~Sx<I,..,I.>
Sx-Sifxu/1,)
1-[(1,..) 2/fx/y)
1-[(1.,.)2/f.,/y]' Sx
s
..
(a8)
(a9)
?-
.. . . . .. .. .. .. . .. .. .. ..
j
The positive directions of q and s are indicated on
(Dv+- D.d- D._+ Bx)b =0
FIG.
52.
s.r'
~<t::::
$
0
D
Transverse loading
s, through
~=0
I9 Gb..-.--"
<at2>
....
(al3)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al4)
and Eqs. (al2), (al3), (al4) yield for the cross-sectional function D:x: in
theM cell
I
Dx=Dxb+D.rM
external walls
web between M-1 and M cells
D.,=Dxb+D.,,M_ 1 -D.rM
All cross-terms but 8.11. M- 1 and 8.u. :u n vanish since only unit shear flows
in adjoining cells act over a common wall. We find
.\1
8.u .\1-1 = O.11-b .11 = - f3M-1
-G--
. . . . . . . . . . . . . . . . . . . . . . (a21)
J~'
f3 :lldl H
~ Jo/
(a22)
M,M+I
M-I,M
are the integrals extending over the common walls (spar webs) of cells
M -I, M and M, M +I respectively. The minus sign arises since the shear
flows due to unit redundancies have opposite signs in the common walls.
Determination of Redundant Shear Flows
where the DxM and D. 111 are unknown. For the basic system of FIG. 52 the
shear flows in the M cell of the actual system can then be written as:
I
q=qb+qM
external walls
J~
s.
denotes _integration over the M cell. It must be emphasized that the sign
webbetweenM-landMcells
f( ... . )dsM
11 ={3M
_ Jds.
0.11.11 ~~
G . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (al9)
Gt
where
xM=qM=s.D;"'111 +S.,D/.111
(all)
s.
and S'x by
We denote the unknown rates of twist associated with
the M'th compatibility
Eq. (a7), whtch expresses the condition of zero relative warping at the
M'th cut is obviously
.~. 1" f3
s.{
tdsM-2n. G'I''S_7., JD.rh
11
M- 1 ,
s.
MD,., 111_ 1
(al5)
+2D.l Gcp.i
y
.....................................................................................................................................................
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 OMob=
J~;dsM
............................. .............
(al7)
-{3,\'-1
NDr,
54
=-
JDtdsN+2D.NGcp,~
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a23)
.V-1 +f3NDxN
AIRCRAFT ENGINEERING,
These formulae are usually derived more simply from the condition of
equal rate of twist*
.P=
~~= 2iLaHdsM
....................................
(a24)
Similarly
I
I
. . . . . . . . . . . . . . . . . . (a25)
twist. They yield hence the shear flow distribution qE---commonly known
as the engineers' theory of bending shear flows-due to transverse shear
forces acting through the shear centre E, q E may be written
-d.
"'
(a26)
qE=Svy+Sq
where the cross-sectional functions d, (d.) are obtained from Eqs. (a 15)
with dxM (d.M) in place of DxM (D.M). The co-ordinates Xg, YE of E., can
and Sr
be determined from a consideration of the two loading cases
through , shown in FIG. 55. We find
s.
I
j
. . . . (a27)
where the integrals extend over all walls and the normal Po is taken positive
(negative) if movement along the positive s direction produces anticlockwise (clockwise) rotation about 0.
I
G.P.of =
.r
JDubP nd5 -
Transverse loading
S, through
S,=O
shear centre Es
shear centre Es
S, through
S, = 0
s.
Su
A..
Ix
'S,.
A..
I.
A..
A..
A..
...... (a29)
where
dO
cf>={{z=.f,+.P ..........................................
(a30)
s.
fDxbPnds+2'LdxMnM+2'LaMnM G.f.t=O
or
*See foe. cit. p. 129, 'The General Theory'. See also derivation of Eq. (al4) by the Unit Load
method in Section 7.
2~d"_,,n .11
I
or
G.f =
Tg
--.-,N---'-'--
.......................................... (a34a)
2'LaMnM
I
(a35)
q ll Jf =c a .II ____,N___,:__
2'La.un.,r
I
The shear flow is, of course, constant in each wall between two consecutive
joints. Thus, we have in the M'th cell
external walls
qll=qBM
I
web between M-1 and M cells
qn=qno.~r- 1 -qllM
t .. ..
(a36)
Transverse loading
dx 111 , d. 111 are the values of the redundancies corresponding to zero rate of
-d,
. . . . . . . . . . . . (a32)
55
from Eqs. (a27). The engineers' theory of bending shear flows for transverse 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 crosssection 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
=20.Npi/JGcp
(a46)
QBM=a.wGpi/Jcf> . .. . . .. . . .. . .. . . .. . . . . . . . . . . . . . . .. .. . . . .. . (a47)
Next we deduce from the equilibrium condition about the flexural axis
M~
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a38)
Mv
+ p21/J, lv
fx
TE
(a48)
2p2~aMQM
I
QBM=aM
TE
N
(a49)
2p 2 ~aMQM
I
where
p=r/r. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a39)
rand r. being the distances from the apex V of the current and root crosssections respectively; for cylindrical tubes p = 1. 1/J and 1/J. are nondimensional functions of z or p.
In what follows under the present heading (l) all cross-sectional dimensions, areas and functions refer to the geometry of the root-section.
As in the previous analysis we assume that the direct stresses are given
by the engineers' theory of bending and write them in the form
(J= p2!fJ,
Gpi/Jcf>
(a40)
Finally, we derive the shear flow q 8 in all walls from Eqs. (a36).
If the torque is given initially about an arbitrary axis VA we can calculate TE with the formula
TE=TA-pQ.(xE-XA)+pQx(YE-YA) ...................... (a50)
where xA, YA are the co-ordinates of the point A at the root.
2. Conical and Cylindrical Tubes with Arbitrary Variation of Boom Areas
and Wall Thicknesses
Here we investigate conical or cylindrical tubes with an arbitrary lengthwise variation of skin thicknesses and boom areas. The engineers' theory
shear flows are not any longer proportional to Q. and Q., and Eq. (a42)
does not apply. The concept of a flexural axis, either straight or curved,
is also not any longer strictly true.
Under this heading all cross-sectional dimensions, areas and functions
are based on the current cross-section.
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
M~
and
M.
respectively.
My
(a42)
Q., are determined from Eqs. (a9) with Q., Qx in place of s., S.,.
. ....... (a43)
p 2 1/Jcf>.~.
(p 2 fcf>.,~)
in place of c/>.t.
cp.,;:) ....
(a44)
in Eqs. (a23), (a24), (a25), (a29), (a32). Note that p21/Jcf>J,JQ., etc., are
constants for the type of loading considered as the transformed Eqs. (a32)
show.
The flexural axis is in the present case a straight line VE, through the
apex and the shear centres at all cross-sections. We define this axis by the
shear centre E. at the root. The co-ordinates XE, YE are found as before
56
For the subsequent analysis we require the modified forms of the internal
equilibrium conditions (a3) and (a4). Thus, we deduce immediately from
the geometry of FIG. 57 the equilibrium condition on a conical element on
the surface
- ~(rf) +~q =0
or
~r
~w
(a51)
where
Hence
where
f =at, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a52)
~q
I ~(r()
.
~s=r ~=Df
.................................. (aSia)
~f f
~f f
Df="iir+-,.=-sz+-,.
.............................. (a53)
..................................
(a54)
where
P. =B.a. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a55)
To find the statically equivalent stresses for an arbitrary system of
transverse loads defined by the shear forces (and associated bending
moments)
S.(M.,) and Sz(M.)
and the torque
TA
........................................
(al)
............................. ...............
qb=JDfEds
(a57)
where !E is calculated from Eqs. (al) and (a52). The necessary constants
of integration to compute the shear flow in the webs, etc., are derived
from equations of the type (a54). TheN unknown shear flows qEM at the
cuts of each section are determined by the same method leading to
Eq. (a23). We find
I~bdS[
f3IqEI-/3[,JiqEJI=-
- f3N-l> NqE,
N-1
+/3 Nq EN
=-
J~bdsN
. . . . . . (a58)
Note that all dimensions are based on the current cross-section. Having
solved these equations we calculate the shear flow qE with Eqs. (a13).
N
Calculation of qB:
The torque T Q of the shear flows q E about the VA axis is
N
TQ=fqEPAds=JqbpAds+1'2:.Q.MqE M ........................
I
(a59)
where p A is the normal from the point A to the tangent at the wall; note
the usual sign convention. Hence the torque T 11 to be resisted by the shear
flows qB is
............................. .............
TB=TA-TQ
QB
Tn
Length
\<ml
Th1l~iss
5
6 7
1. 10
2.'
3. 8
300
30-0
301
302
17
20
20
16
25
25
1-6
1-6
1'1
11
,..
&
Cell
lll
Il
Yl.
12
10
Area
350
551
100
540
310
210
Area
12
12
5,6
(em~)
10
657
146
10
10
Table I of Example 9a
Values of JD~bdsMft and coefficients {3
M
Cell M
(a61)
1'2:.aMQIII
I
qE may be regarded as quasi-engineers shear flow; see also J. H. Argyris and P. C. Dunne,
Joe. cit., p .53 (Handbook).
3
7. I
2
9
(a60)
2. 3
1.1
1
10
1. 2
9.10
Boom
l<m'l
Nose Rear
cell cell
Webs
II
III
IV
350
555
600
540
390
471
763
490
466
JD
J'b
d.~.'i
t
106
125
125
100
+68
+120
+120
t96
\ 7200 1 780J 1
210
----
908
40-l
VI
--"!.'I
-+50
_ 3120---
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))
763D., 1 -106Dxo
(a)
GOOD CHOICE
(b)
BAD
= -7200+ 700Gc/>fy
= -7800+ lllOGc/>~y
-106D., 1 +471Dxo-125Dxa
0+1200Gc/>~
.JJx
+ 5400 + IOSOG'I'S
y
s.
Sv
(a63)
.. ........................
D.,M=d.,M+aMGc/>~v
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a25)
TABLE
n.
CHOICE
Table II of Example 9a
Values of dxM and DxM (em. 3)
M
dxM
aM
DxM
-1217
146
-20 13
II
-1962
390
-4092
III
115
452
-2585
IV
+1579
420
+2071
308
+ 389
VI
+ 488
068
+ 116
10.000 kg.
716
f-z23cm-;
q in kg/em.
Fig. 60.-Final shear flo;v distribution in six-cell tube
l"'
58
FIG.
60.
Shear Centre
Since the present cross-section is singly symmetrical, the shear centre
lies on the axis of symmetry and the first of Eqs. (a27) (with 1.,.=0) gives
immediately the XE co-ordinate. Alternatively we note that the shear
flows dx have a resultant s.=I., through the shear centre and we can
modify slightly Eq. (a64) for torque about D to give
700d.,1 + 111 Od.,2 + 1200d.,3 + 1080dX4 + 780dx5
+420dx6 +94600=/.,(xE-XD)
Therefore
(b) The Four-Boom Tube with Deformable Ribs and some more General
Structures
In this example we determine, using the force method of analysis, the
'exact' stress distribution in the idealized four-boom tube shown in
FIG. 61. The investigation is carried out first by the Swmethod and
illustrated in a numerical example. Subsequently, we analyse the same
example by the general matrix method of Section 8C. We show also how
the matrix formulation may be used with advantage in the more interesting
case of a six-boom tube with or without intermediate spar web. It is hoped
that these simple applications of the matrix method will condition the
reader to the new ideas and show him their power and basic simplicity.
Consider the cylindrical tube of FIG. 61 with a singly symmetrical trapezoidal cross-section the flange cross-sectional areas Band the wall thicknesses t of which may vary arbitrarily length-wise. Loads are applied only
at the rib positions
1, 2, ........ , i, ........ n
in the form of shear forces
l:
z
and moments
Mato Mbto ........ , Ma;. Mb; .. . ... . . , Man. Mbn
at the front (a) and rear (b) spars. Following the general discussion on the
idealization of aircraft structures given in Section 8 C (pp. 'J7and 40) we
assume that the walls carry only shear stresses (the direct stress carrying
ability is allowed for by suitably increasing the flange areas). The shear
flow is hence constant in any field of each bay since changes in the z-direction can only be brought about at the ribs. It follows then that the end
loads in the flanges vary linearly between ribs and that a knowledge of the
flange loads at the rib positions suffices to determine them everywhere.
Having found the flange loads the corresponding shear flows are determined easily from the flange load gradients and the condition of equilibrium
with the applied shear force and torque.
In the tube shown in FIG. 61 there are n ribs including those at the free
and built-in ends. At each rib position there are four flange loads and only
three equilibrium equations are available for their determination. Hence,
nothg that the flange loads at the tip are zero, the degree of redundancy is
(n -I). This trivial result is confirmed by the general Eq. (247a) on p. 38
by substituting N= I, f1=4, a=n.
In selecting the basic system many choices are open to us. We may, for
example, make a single cut in one of the flanges at each rib statiori to reduce
the structure to a statically determinate three-flange basic system. Here,
however we calculate the statically equivalent stress system by the E.T.B.
and the Bredt-Batho theory of torsion, a more general example of which
was investigated in Section 9a. In this choice, instead of making a single
cut we have, in fact, cut all the flanges to allow the relative warpi?g consistent with the statically equivalent stresses while at the same tune the
direct stresses are transmitted across the cuts ; see also the discussion in
Section 8C, pp. 32 and 33. The redundancies then consist of self-equilibrating flange load systems at the (n -1) rib stations. A suitable and symmetrical measure of such a system is the boom load function P introduced
by Argyris and Dunne.t We prefer to use instead a slight variant of P~ the
Y-system introduced on p. 39. The (n -I) redundant Y are determmed
from the compatibility conditions of warping at the (n -I) rib stations.
As a further alternative procedure we could, of course, choose as basic
system the very simple structure consisting only of the two spars acting independently, i.e. we cut the top and bottom covers of the tube. However,
Actually, we may consider this basic system as .also derived by a single cut from the given system .
Thus if at the cut Bange we - l y the correspondmg E.T.B . Bange load and allow there any out of
place' movements, the stress distribution in the other structural elements is obviously that of the E.T.B.
and Bredt-Batho theories.
t Joe. cit. p. 38.
except for very flexible covers and ribs our previous choice is much closer
to the final correct solutions. The latter method is hence to be preferred
when the design work has to be checked solely by a good statically equivalent stress system as is, in general, the case when no automatic digital
computer is available. Subsequently the exact distribution, if found at all,
is obtained only after completion of the design work-usually by a more
clumsy version of the O;k method.
On the other hand when we use the matrix formulation in conjunction
with a digital computer and derive the complete stress distribution as a
single process it is preferable to select the simplest possible b 0 matrix (see
also the discussion on p. 39). Then the basic system formed by the independent spars is obviously indicated. An important advantage of this
system is the absence of rib stresses if external loads are only applied in the
plane of the spars. In multispar construction the simplicity of the basic
system consisting of independent spars in comparison with the multi-cell
system of Example (9a) is even more striking. If it is necessary to taKe into
account external loads applied at intermediate points in the ribs and/or
if spars are interrupted we may select as basic system for the matrix type of
analysis the grid formed by spars and ribs (without covers}-still a very
simple structure in which to find the b 0 matrix.
(I) Analysis by the O;k Method*
The statically equivalent stress system
Following our previous discussion we select here the E.T.B. (or quasi
E.T.B.) and Bredt-Batho stresses as statically equivalent stress system.
The flange loads are calculated at the rib stations with the effective areas
B; there. Thus, the m'th flange load at the i'th rib is,
Pom;=M.,l'mi Bmi
lcr.i
where M.,; is the bending moment at the i rib due to the applied loads Ra
The particular case of the four-Bange single-cell tube under a given loading has also been treated
by W. J. Goodey in 'Two-spar Wing Stress Analysis', AIRCRAFT ENGINEERING, Vol. XXI, No. :U7,
p. 287, September 1949; No. 288, p, 313, October 1949; No. 289, p. 358, November 1949. 'J'here.
the author uses the Castigliano technique to formulate equations analogous to the three- and f!Ye-joiat
equations given here with essentilUiy the same basic idealizations but with tbe effect of taper induded.
A comparison of the theoretical results with experimental strain measurement is aloo JliYeo.
59
n;~
"'
,.,
'
o2
Fig. 63.-Fiange loads due toY; = I and Y;+ 1 =I (See Fig. 4l)
,.,
i,,
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
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
qB for a torque TB to be found by Eq. (a60). Naturally, in multi-cell tubes
it is on the other hand preferable to use the presentation (b4) for the
calculation of q . Note that since we assume that flange loads vary linearly
between rib stations, systems (bl) and (b2) only represent E.T.B.
at any cross-section between rib station if the flange areas are constant
or vary similarly within each bay.
To complete the statically equivalent stress system we have to calculate
the rib shear flows. Consider to this effect FIG. 62 where rib i is shown
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
Mad h., Mb;/hb
For a trapezoidal rib the shear flow is not, of course, constant but resolving forces horizontally for the upper or lower half of the rib we find the
average shear flow
qor; = q o,i-l -q 0 ;-
Ma;
M bt
where q 0 ,;_ 1 and q 0 ; are the shear flows in the covers of the i - I and i bays
respectively.
The self-equilibrating stress systems Y;
We show in FIGS. 63 and 64 the flange loads and shear flows due to
Y, = I and note that they extend only over the bays (i - I) and i; these
figures reproduce in a more convenient form for the present purpose the
information given in FIG. 43. The system Y is connected to the boom load
function P of Argyris and Dunne by
U.+ ~J
Y= P
k=l
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'
stress distribution by superposition of the statically equivalent stress
system and that due to the Y;'s.
60
,.,I
w,., .~h!h
n.
I,,,
\
w~,::::- h./\
42
.,
O;ot =
~
"'--
m= l
i-1
om
~dz
m
( b9)
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
(i - 1) and i since the influence of Y, is restricted to this region. In general,
the contribution of (b9) to 8;. is negligible and could only become significant in a bad design.
The general formula for 8,., follows immediately from the unit load
method as,
s.=iJ Jq1'ds dz
where q; are the shear flows due to Y, = I and the subscript s denotes inte-
gration over the walls. Again we need only consider bays (i- I) and (i)
and we obtain immediately with the notation of FIG. 64
Slo =
where
(btl)
,\'m,m+l
,8
m,m+1=t- (b12)
m,m+l
The idealization of constant flange areas in each bay has been discussed
on p. 37.
Proceeding now to the contribution of the Sib we note that its general
expression is
ik.= IIqiqkd
Gt s d z .......................... (b20)
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.
t.1 -1"
TA
+ 20
4
m~ .w,8]m,m+1
..... (b 13)
where
~ qAw,8]m,m+1 j
Xs-XA=-20 m~l 4
S .... [ ,8]
Y m~l W
m,m+1
'Y=
For the last term
............ (b14)
.... (b20a)
n (w 41 ) 2
d
I
1
St.i-2,r=S;-2.ir=(-Gt)
,-,-=2(h a +h)
(-Gt)
,-,r r i-1 i-2 i--1
b
r r i-1 i-2 i-1
~0
~ 'Yi-1
G /;_ 1
sikr
We obtain,
Ys-YA=O
where
t-t.tR--
I 'Y,
S;.;,,.,=8q1.i . =c-c; 7;
Note
Using Eq. (b2) in each of the expressions in the square brackets we find
4
4
m ~ 1Wm,m+1q om,m+1,8m,m+1 =;,~,}qAw,8]m,m+1
-s
where
r.~
0=
(b15)
~ [w,8]m,m+1
m=l
~ [Dr.,Bw]..,,m+l
Xs - XA
cc
r.m~"l
2.1.~~
4 -'-------
.............. (b14a)
(b22)
Finally we find the total S1k coefficient from
S;k =Sikf+Stks+otkr
This concludes the derivation of all formulae for the S-coefficients in the
Eqs. (b7) for the unknown Yi; Solution of the latter system of equations
yields the total stress distribution in the tube. We illustrate our analysis on
a numerical example further below.
I::jfect of Cut-outs
~ [,8w]m,m+1
m=l
stor =o[(q.,ql)
G,t,
1-1
+(q.,q;,) +(q.,q;,)
G,t, ;
G,t, 1+1
Rib k removed
I
f"\
ha+hb
w 41 =-ha+hb and .1.~=-2-d
G,1 and t, 1 are the shear modulus and thickness of the i'th rib.
h.
'1
h.
slkt=IY;Yk
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
4 w,.2
<I>= ~ EB
............................. (bl9)
m=l
k-2
Yk. 1
Fig. 65.-System
Yk-t =I
If we assume that the flange areas and hence <I> are constant in each bay
we find immediately,
ll ....
(bl8a)
J
See AJ'IIYris and Dunne loc. cit. p. 84 and also Argyris and Dunne, Stress Distribution in Conical
Tubes. puhlihPd by R.Ae.S. as Vol. Ill of Stressed Skin Data Sh-.ts.
If a rib k is removed bays k --I and k merge into a single bay and the
calculation proceeds by the previous analysis observing that the number of
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.
If in the (k ~I) bay a wall is removed it is again possible to use our
analysis with few alterations. Thus,
(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 J. H.
Ar~yris.
AIRCRAFT ENGINHRING,
61
"
--~.
10
__::..:.-,...
Pa-lf
(IIHO')
E T.B
..,
F"lan;. I oa<H
J50 ' -
~
q,,
1------
140
OMI
-,
--r-
-- --
.,
-- - -- -
---~~-=-.:.-=~-~:.....
~-
...;
(
'._ _ _ _ _ _ _J
~ r-----i-----~-----4------1------+-
...
Rib stahon
Fig. 66.-Fiange loads and shear flows In extended Yk system when web
or cover Is removed in (k - I) bay
(b) the unknown system Y~:-1 and Y~~; merge into a single system which we
denote by Yt and illustrate in FIG. 66. In this system the flange loads
are constant in bay k due to the kinematic freedom of our open tube.
Observe that the system of (n - 2) unknown Y;'s consists now of
Y1, Y2, ... . .... , Y1:-2, Yk, Y "+b ... . .... , Y n-1
8 _
Similarly, the
8"+1kr =
8,.,
m= l
1. .
8 _ tj_{_1 [(qor)
~~;.,. - 2
1~:- 2
G.t. 1:-1
-(qor)
Grtr
t-2
] -J. [(q.,.)
- (qor)]}
It G,t, "+t
G,t, 1c
(bl8)
All other 8;;1 and 8;,1 are unaffected except that O;,k-1,1 =0
Formulae (b20a) for 811, etc. become in the case of i= k,
8~:-s,to= -J
'r
i:-2
k- 2
8t-1,h = O
~ _G
_!['Yk-2+'Yk]
1~:-a
/~
Otk-
62
ll
.... (b20b)
d
_1 [-1-- c--~ +-~ )+
1
_1
2(ha + h&) /k- 2 (Grtr)k-2 /1:-2 /k-3
(Grtr)i:-1 /1:-2
k- 2 , kr -
8k-1,kr = O
d
{ [ - 1 - + -~-Jc-1 )2+
(G,trh- s (Grtr)i:-1 /k-2
+
d
I[
1
I
I
( I
1 )]
2(ha + hb) /;, (G,t,)k /;,+(G,t,h+1 "J;,+/"+1
(b22a)
-3560
29103
-3116
-3116
31635
-2380
--2380
39583
-19381
-1938
44742
I
0
I I (q or/G,t,)!
-1 +2 -1
0 -1 +2 -1
0 -1 +2 -1
0 -1 +2 -1
_j
-833~
16800
I -8 334
............ (b24)
(q or/G,I,)s
_j
2 -1
2 --1
0 -1
2 -1
0 -1
0
0
2 -1
2 -1
2083
44 133 -10712
(G,t,h
2083
52081 -10270
2083 -10270
57240
_j
Ys
_j
O (G,t,h
I -I
0
O
-1
2 -1
2 -1
2 -1
0 -1
(G,t,)a
(G,t,)4
0
0 -1
0 -1
0 -1
I o
(G,t,);
(G,t,)s
_j
O (Grtr)1
L
2691
- 4774
2083
4774
11023
8332
208~
2083
8332
12498
8332
2083
2083
8332
12498
8332
2083
2083
8332
12498
8332
2083
8332
12498
2083
47892 -10974
0 -1
I
I
2083
2083 -10974
Next we consider the effect of rib deformability. Eqs. (b22) for 8;kr are easily
calculated directly but it may be convenient also to use their matrix formulation for their evaluation (see also Eq. (258) on p. 41 ). We have
.............................. (b27)
2083 -10712
1418
=10-" i
0 -1 +2 --1
2083 -11448
(q or/Grlr)J
2247
_j
2722
I I (q or/G,t,)al
We may now compute the total 8;k and 8;. coefficients to find the redundances Y; for deformable ribs. The equations are
- 3642
I (q or/G,/,)21
174
479
0 -1
-2821
Ys
-1
582
-1
_j
28187
(q ;,..(G,fr)1
=10-6
-2642
5418
+I -1
-2642
I
-27764
I
1151 I
35394
Similarly we find for the contributions 8;or ofEq. (b16) using the values
of (qor/G,t,); in TABLE II.
I
I
2 -1
2
0 -1
_j
_j
.......................................... (b26)
The small term 6iof arising due to the dissimilar taper of front and rear spar flanges has heen
ignored here (See p. ~0.)
63
TABLE I of Example 9b
Cross-sectional dimensions and 3;kf, 3;ks coefficients
Rib
(in. 2)
B 3 =B, (in. 2)
1., (in.')
tr (in.)
G, (lb./in. 2)
Mean B 1 = Bz (in. 2)
Mean B3 =B (in. 2)
t 12 (in.)
t 34 (in.)
t 23 = t n (in.)
~12
~34
~2a=f3n
<D (in.-) Eq. (bl9)
'I" (in.- 2) Eq. (b21)
B 1 =B2
3i,i-1.! X 10"}
Eq.(bl8a)
3iifxl0 8
3i,i+I,f X
10 8
3i,i-t,s X 10 8 }
3;;. x 10 8 8 Eq.(b20a)
3;,;+1 X 10
Bay
035
045
025
040
040
040
263
28 1
245
0036
0036
0036
250
250
I 250
040
065
050
065
030
060
I
040
040
040 I 040
060
I 100
0036
0048
0036
0048
0036
0048
0036
0080
0064
0080
0036
0064
0048
0048
0048
0048
0048
I 0048
167
167
167
125
125
I 125
278
278
156
156
I
I 125 I 125
501
501
501
501
501
I 501
02350
01425
01887
01186
01610
01054
6118
5490 I 5945 I 6118
5446
I 5446
6 297
5 371
3 960
3 516
4 753
28271
23333
20247
17427
14952
7031
6297
5371
4753
3960
3516
-7076
7133
8235
7076
8013
16248
16469
14208
14 151
15146
7076
-8235
-7076
-8013
-7076
-7133
065
120
71 7
008
385
055
040
299
0036
250
065
040
317
0036
250
065
080
517
0036
250
TABLE II of Example 9b
Statically equivalent stress system and 3; 0 ., coefficients
k.ib
Bay
Ra (lb.)
Rb (lb.)
s. (lb.)
Ta (lb. in.) about spar 'a'
M, (lb. in.) X 10 3
--Pol ='Po2 (lb.)
Poa- --Po (lb.)
q 012 (lb./in.)
q o34 (lb. /in.)
qo2a=qon (lb./in.)
q 0 , (lb./in.) Eq.(b5)
'L.w{3qo
I
I
I -34730
I
1
6300
75600
770
4292
4
150
1000
600
400
1450
5100 I 3600 I 2450
I 70800 I 52800 I 28800 I 19200
89
161
40
263 I
I
i
. . . . . . . . . . . . (b28)
_j
FIG.
67.
200
700
38lJ I
9870
1920
9920
4910
10570
5950
2850
32510 I 20350 I 10160
870
1 -2t69
171 9
2434
1349
-1332
-2163
-876
-3772
-4248
-218
55
-128
-1372
--1840
273
1840 I -508 I 1387 I
I 90 I
I 10689 I 8603 I - 2249 I 847 I 201
272
-542 I
364
2819
2776
303 I -100
597 I -564 I -1541 I
7
6
450
100
64
750
750
1000
200
-17124
3
2
6522!
45113
550
2400
II
440
840
0
0
0
418
-219
-199
747
71
142
79-
199
I
-221
leading to the three or five-joint equations, provides a practical and powerful method for the determination of the primary stresses in such a structure. However, in more complicated cases, with more than one redundancy
arising in each bay, the more systematic technique of the matrix formulation and solution is the only natural tool. Even for the four-boom tube
the analysis under a general loading and the calculation of the flexibilities
requires the use of matrices for its practical realization .
Therefore, for comparison with the long-hand solution and to illustrate
the way in which the method is applied in practice we analyse the same tube
as previously under an arbitrary set of forces applied at the rib stations
and calculate also the flexibility matrix for the points at spar-rib intersections. Finally, we apply the H-matrix device of p. 39 to find the stress
distribution when one web field is removed.
Although the matrix method is naturally associated with the use of
automatic digital computers and large numbers of redundancies it also
proves extremely useful in smaller problems with more elementary methods
of computation. Thus, the calculations described here were all performed
on a conventional desk type calculating machine and it may be of interest
that the complete calculation-up to and including the flexibility matrixoccupied a total time of about 40 man hours.
The applied loads R; are numbered consecutively from I to 12 as shown
in FIG. 68 which gives also the numbering of the elements. As mentioned
in the introduction to (9b) we consider as basic system the two spars acting
independently as beams and it is obvious that the top and bottom covers
and the ribs are unloaded in such a system. The rib booms are loaded
neither in the basic system nor due to the redundancies and need not be
considered at all.
Due to our choice of basic system it is apparent that in the matrix a
considerable number of zeros will arise and to obviate carrying these we
advance the partitioning of bo in Eq. (253) a stage further and write,
r
r
b.z
bola
bolb
b=
0
bow
~ bOT
L
7 ....... . 12
,-_A__---,
1 ........ 6
,----A-----,
- -- - -
..... (b31)
bon a
Edernal loads
I - - - - - - - - - ,_ - - 0
I
0
_j
bnwb
Basic system
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 .
for
b0
matrix
Numbering of elements
.......................... (b32)
blu:a
Y,
Y,
f=
flu
f,b
fwa
fwb 0
. . . . . . . . . . . . (b33)
Y- systems
for b: matrix
(se-e F1gs. 63 and 64)
where f,., f," are obtained from Eqs. (257) and f., to f, from Eqs. (258).
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,
bl'fb.= [[bl'fb.],.
[bl'fb.J,b]
_j_
[[bl'fb.],.."
...... (b34)
[bl'fb.],b]
I [bo'fb.],."
I
L
,,
f,
1
0
[bo'fb,J,.."
bola=
.... (b36)
_j
20
-6
I
I
I
I
I
I
JJ
}2
}4
}6
}3
........ (b37)
J5
_j
The numbers in italics over the columns of the matrix refer to the external loads R1 . . . . R 6 and the numbers opposite the pairs of rows are the
number of the elements of the flange 'a' in each pair of rows; the first one
is for the end of the element nearer the root and the second row for the
end away from the root. (See FIG. 68 and p. 40)
65
Fig. 69 (right).-Four-flange tube. Flange loads and web shear flows due to
single load R3 = 1.000 lb. Application of H matrix to cut-out problem.
For the web of spar 'a' the matrix b ,., of shear flows due to the loads R
in the basic system is
2
r-
6
1
0
1
bowa=-~1
I
L
''
(b38)
-1
}1
0
0
(2
(b39)
with cut-out
q_,
-so .
- 4Q
-so
- 4Q
2o
4o
so
_j
-1
-1
0. l -I
-I
-I
no cut-ou'
\
J6
o o o o
- 2o
15
j
2 00 I 001
0 0
I 00 2 001
0 0
0 0
0
12
1333166l
0 01
0 0 lt66 3331 0 0 I
--- -1------1----~----1
I 0 0 1500 250 1 0 0 1
(b40)
1 0 0
2
f1b=3x ws"
1250 5001
0 0
I o o
I
----1
1500 2501 o o 1
I 0 0
1250 5001
,----~----1----
The corresponding matrices for web 'b' and the top cover differ from
Eq. (b40) only in the scalar factor which, instead of 10/6 x 320, is
0 0
1----~----~----~----
1
and
11
10
----~----~----]
-so
0
r4
(lb./In)
'1o 0 .
10oo
'i4o
J3
'
'12o
''
',,',,,
''
q_,
''
P,
(lb/.n)
I.
12'3456
''
- 3ooo .
Corresponding matrices for spar 'b' are identical with those of Eqs.
(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 selfequilibrating 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'
''
(lb.)
''
o o
I 0 0
} ............ (b41)
1500 250 1 o o
1250 5001 0 0
1----~----~----
1 0 0 1500 250
I
The sub-matrix b 1r for the shear flows in the ribs has been given already
in Eq. (b25) and need not be repeated here.
The flexibility matrix f of the unassembled elements of the structure is
straightforward, using Eqs. (256), (257), (258) and (b33). As an illustration,
we obtain for the upper flange of spar ,b', using the data of TABLE I:
66
0 0
1250 500
J
. ............. (b42)
where the numbers in italics over the columns refer to the flange elements.
Note that the only non-zero elements appear in the diagonal submatrices.
I
I
Shear flow
in web 'a'
(lb./in.)
Shearflow
in web 'b'
(lb. fin.)
Rib station
Tube without cut-out
Effect of cut-out
Tube with cut-out
Tube without cut-out
I Effect of cut-out
Tube with cut-out
Bay
Tube without cut-out
Effect of cut-out
Tube with cut-out
Tube without cut-out
I Effect of cut-out
Tube with cut-out
66384
13916 00347 -00558
13916 29840 07156 00249
25828
05494
00347 07156
-00558 00249
05494 23167
-00132 -00244 00088
04661
-00003 -00053 -00184 -00007
I
b,a=lo- 3 x
r
L
11320
2389
0063
-0095
-0023
-0001
3
-971
I 1762
791
-1418
1 -1057
-2475
I
1
2
-840
-364
--1204
-298
13 1
-167
I
1
5
-30
-396
-426
18
238
256
4
323
-1532
-120)
-194
919
725
3
-1029
1029
0
-230
-370
-600
I
I
-2098
-0841
0430
-0017
-0021
-0003
I
-00132
-00244
00088
04661
20885
03578
5
-00003
-00053
-00184
-00007
03578
1 8112
75572
60395
42159
27196
13604
1593
110
-355
-245
-40
128
88
6
-17
-15
-32
10
9
19
6
-05
-05
-10
02
02
04
-1?:;
-123
~:~
44
1989
4178
4919
4000
2118
0063
2389
5 244
6586
6000
4118
2000
0
0
0
0
0
0
_j
_j
IO
I2
11
I
91509 -2595 -6598 -10780 -14873 -18934 -22993
75470 -0575 -4087 -8894 -13867 -18802 -2372l
56034 -0022 -0986 -5700 -11782 -17980 -24142
40097 0021 -0036 -1275 -6494 -12950 -19434
28050 0005 0033 -0033 -1242 -6227 -12012
14328 0000 0007
0041
0013 -0755 -4942
... (b43a)
_j
-959
-163
-1121
-255
59
-196
-1447
0398
0011
-0016
-0004
-0000
-1658
599
-1059
-3005
-359
-3364
-1704 -3413 -5123 -6832 -8541 -10251 0389 0789 1189 1589
-0006 -3419 -6838 -10294 -13676 -17094 0012 0978 2044 3 111
0
-0006 -3514 -7075 -10636 -14198 0 0100 1586 3253
0
0
-0053 -4233 -8307 -12381
0
0 0181 2000
0
0
0
0104 -4815 -9815 0
0
0
0181
o
o
o
o
-0081 -6173
o
o
o
0
1
-2726
78
1
-2648
-4364
-47
1
-4411
-2726
-1658
-0971
0323
-0030
-0017
2
-3385
-2434
-1966
-1080
0246
-0038
3
-4071
-3268
-2974
-2134
-1066
0266
4
IO
I2
11
I
-4749 -0433 -1100 -1797 -2479 -3156 -3832
-4088 -0096 -0681 -1482 -2311 -3134 -3954
-3994 -0004 -0164 -0950 -1964 -2997 -4024
-3182 0004 -0006 -0213 -1082 -2158 -3239
-1992 0001
0006 -0006 -0207 -1038 -2002
-0945 0000 0001
0007 0002 -0126 -0824
_j
IO
11
I
2
3
4
5
6
0.
(b44a)
I2
I
-8313 -1052 -1308 -982 -524 -069
381
-6544 -288 -1615 -1663 -1086 -428
218
-8788 -022 -495 -2305 -2755 -2620 -2~52
-8935
008 -036 -647 -2735 -3502 -3866
-9970
003
014 -038 -654 -2849 -3682
-9520
000
004
021
007 -393 -2574
_j
wi
w2
w3
w4
w5
w6
00
00
00
00
(b44b)
Note
The numbers above columns refer to the loads R1 to R12
In b 1a the numbers opposite rows refer to values at the rib stations. In the b., and b11 matrices two values were used at each rib station to denote
the (equal) load in the adjoining flange element sections. Here in baz these two values are merged.
In b,a the numbers opposite rows refer to values in the web fields.
The additional factor 2 is introduced to take care of the lower flange since
b. and b1 contain only the loads for the top flange. This applies also for
the flanges of spar 'a' and for the covers.
Carrying out the multiplications and assembly indicated in Eq. (b35)
we obtain the complete matrices b1'fb. and b1'fb1 and hence, on inverting
b1'fb1o the column of redundancies
Y = -(b/fb1) - 1 b1'fb.R ........................ (b43)
These results are printed in full above except for b1'fb1 which is
identical with the matrix D of the S;k coefficients in Eq. (b28) and need not
be repeated here. The calculation of the reciprocal matrix (b 1'fb1) - 1 was
performed using the Jordan technique described on p. 28 and required
approximately 8 hours for completion-by an unpractised and comparatively unskilled computor.
The complete set of flange loads and shear flows due to the forces R is
obtained now with the b. and b1 matrices already given and the solution
(b43) as
5=bR=[b.-b1(b 1'fb1) - 1 b1'fb.]R .................. (b44)
Limitations of space prevent the presentation of the complete matrix,
but as typical results we show above the flange loads in the upper
flange of spar 'a' :
bza=(b 0 za -bua(bt'fb1)- 1bt'fb.J ................ (b44a)
and the shear flows in the web of spar a :
.
b,a=(b,owa-btwibt'fb1)-lbl'fb 0 ]
(b44b)
Calculation of the flange loads and shear flows for the particular set of
67
forces R 1 used in the previous analysis gives results in agreement with the
values calculated there. Such differences as arise are due to the neglect
of the terms 8;.1 in the O;k analysis. (See footnote on p. 63).
As a further illustration, the 11ange loads and shear flows given in TABLE
III due to a single load R3 =I ,000 lb. are easily obtained from the third
column of the final b matrix of Eq. (b44.)
These values are also depicted graphically in FIG. 69.
E.ff'ect of a Cut-Out
To show now the application of method of treating cut-outs developed
on p. 41 we determine the effect of removing the web of spar 'a' in the third
bay (i.e. element w3). This, of course, converts this bay into a four-boom
open tube which, if taken into account initially would have required the
introduction of the special extended Y system over the cut-out discussed
on p. 62 and destroyed the simplicity and regular pattern of both b 0 and
b1 matrices. By our artifice of imposing initial strains in the cut-out element
until the stress there is zero we eliminate all such special considerations.
Moreover, when the cut-out is introduced subsequent to the initial stress
analysis, our method does not require any additional special stress systems
but works solely with the existing information. Since the web carried only
shear stress, the matrix of initial strains has here only one row and the
result is obtained with remarkable speed and simplicity, the additional
matrix to be inverted being of order 1 x 1.
The shear flow in the element to be eliminated, i.e. b 11 , is given by the
third row bwa (Eq. (b44b)), (printed in full on p. 67) and the matrix b 111
of the shear flow in this element due to unit redundancies is similarly the
third row of b 1 wa (Eq. (b40)), i.e.,
1
b 11,= 192 [0 0 1 -1 0 0] .................. (b45)
To conclude the present example we calculate now the complete flexibility matrix of the structure for the points and directions of the forces
R 1 to R 12
In Eq. (312a) for the flexibility matrix, F
F=bo'fb.- bo'fb 1(b 1'fb 1) - 1 b 1'fb.
fb., b.'fb 1 and (b 1'fb 1) - 1 b1 'fb.
Now the equation for the set of Y H redundancies due to initial (shear)
strain H in this element is (Eq. (236) or (262))
DY 8 +b 111 'H=O
Y n= -D- 1 b111 'H
.................... (b46)
and therefore the additional stress (shear flow) in the element due to H is
b11,Y 8 =-b 111 D- 1 bu'H ................ (b47)
Adding this to the stress in the element already calculated, b,R,and equating the resultant to zero, we find, using the above b1 11 and the D- 1 already
calculated, the required initial strain
H =(b 11,D- 1 bu')- 1 b 11 R
192 2
3 801 X 10 8 b "R ............ ........ (b48)
Hence from Eq. (b46) the additional Y 8 system required to nullify the
stress in the cut-out element is
00238
01817
05350
-04650
-01203
-00047
_j
2154
2762
3567
4142
4836
5532
443
1127
1842
2543
3238
3932
68
10
1/
12
I
-0004 -0064 0471
0412 0406 0402 0001
0023 0105 0125 0120 0112
-0029 -0488
3591
3147 3096 3066 0008 0173 0804
1961
0914 0856
-0085 -1435 10573 9266 9115 9027 0023 0509 2367 2830 2691
2519
0074
1247 -9189 -8053 -7922 -7846 -0020 -0442 -2058 -2459 -2339 -2189 .R
0019 0323 -2378 -2084 -2050 -2030 -0005 -0114 -0532 -0636 -0605 -0567
0001
0013 -0092 -0081 -0079 -0079 0000 -0004 -0021 -0025 -0023 -0022
L
Exact flexibility of structure in in./lb.,
F=lo- 8 x
. ................... (b50)
2762
7368
1034
1329
1640
1948
1082
3650
6723
9825
1290
1597
3567
1034
192-l
2659
3423
4191
1730
6522
1361
2128
2894
3659
_j
4142
1329
2659
4233
5760
7222
2371
9430
2109
3551
50~3
6531
.................. (b49a)
4835
1640
3423
5760
8436
10903
3005
1229
2851
5019
7477
9973
5532 443
1082
4191 173()
7222 2371
10903 3005
15106 3642
3642 1081
1516 1721
3593 2353
6493 2988
9)69 3625
13730 4261
19~8
1127
1842
3650 6723
6522 1361
9430 2109
1229 2851
1516 3593
1721
2353
4990 7873
7873 1577
1074 2321
3062
1361
1649 3804
10
2543
9325
2128
3551
5019
6493
2988
1074
2321
3838
5306
6775
11
3238
1290
2894
5043
7477
9959
3625
1361
3062
5306
7881
103.69
12
I
3932
1597
3659
6531
9973
13730
4261
1649
3804
6775
10369
14247
_j
.................. (b50a)
TABLE IV of Example 9b
Comparison of exact' and approximate ftexibilities /;h (in.flb.)
Exact x 10 6
Approx. x 10 6
Error
/2,2
7368
4961
-33%
!t.4
4142
2932
-27%
!t,l
2154
1050
-51%
ec
As would be expected, the errors are much greater for points near to the
root. Towards the tip they become negligible but it should be remembered
that the structure analysed has quite a high aspect ratio. The example
shows all too clearly that for modern low aspect ratio wings it will be
necessary not only to take account of the shear strains associated with
E.T.B. but to base the calculations of flexibility on the exact stress distribution. Moreover, as shown by our simple example the computations
leading to F appear in the matrix method as a natural final step of the stress
analysis.
Thermal Stresses
We consider now the stresses which arise in the four-flange tube due to
non-uniform temperature. For this purpose we assume that the upper flange
of spar 'a' has the temperature distribution shown in FIG. 70. The variation is taken to be linear between nodal points and hence is defined
by the values at the nodal points. Since also the flange loads of the
Y systems vary linearly between these points the argument on p. 43 leading
to the second of Eqs. (287) may be used here too. Thus in place of the
matrix e (Eq. 281) of the direct strains at the nodal points due to the loads
R in the basic system we introduce here the thermal strains (a8) at the
same points. As transverse flanges do not enter into the present considerations the matrix I simplifies considerably and observing that' I is the same
in each bay, becomes for the affected flange
I
4
0
I
1=6'
0
4
4
4
40
0
P (lb X 100)
q (lb/in)
so
P.
Temperature distribution
on upper flange of spar 'a'
(lbxlOO)
Q., (lb/in)
zo
0
'2o
-40
so
eo
I
I
I
0
(b58)
Fig. 70.-Four-flange tube. Flange loads and shear flows due to temperature rise of upper flange 'a'
a81
D,e=(b 11
37
a8a
.)'l.l
ae~
(b59)
a8 5
aG&
IL
a07
0
0
(buaY=~
42
00
00
00
00
00
00
(b59a)
_j
_j
01> 0 2, 0 7 are the temperatures at the nodal points (see FIG. 70).
In b 1az the two equal values in the elements on either side of the nodal
points are merged into one so that btaz gives here single flange loads at the
nodal points.
Thus
I
57
a82
35
66
_j
............
'
so
120
/a,a
!t2,12
14247
1924
1543
14151
-20% -071,o
/2.4
1329
1061
-20%
00
_j
00
00
00
00
(b60)
00.00
00
00.00
69
Two-bay segment
i-sysltm
(equal net opp05 iU X~systems
(b63)
as indicated on FIG. 71; see also FIGS. 41, 43 (or (63) and (64)) for a full
description including shear flows of these systems.
If the cross-section of the tube in FIG. 71 has an axis of symmetry the
number of redundancies for lift loads and torque is
S-flang~
two-cell tube
n1 =6 X2 = 12
As redundant systems we choose in this case
6 Y systems applied to the corner flanges
} (b64)
6 X systems
An X system consists of an antisymmetrical combination of two equal and
opposite X systems in the top and bottom covers as illustrated on FIG. 72.
These coupled X systems were not mentioned on p. 38.
Next we discuss the six-flange tube with intermediate web shown in
FIG. 73. There are again six bays and for the moment we assume that there
are no cut-outs. When the root is fully supported the degree of redundancy
is from Eq. (247a)
n=a(,B+N-4) = 6(6+2-4) = 24 ........... . .... (b65)
The b. matrix may be obtained simply from the basic system consisting
of three independent spars. As redundant systems we select,
X-system
(see Fig. 42)
(b66)
70
Y- systems.
Z-syst~ms
iJ.
Basic system
Missing rib
X-system
(see Fig. 42)
Missing web
Extended Y-system
(see Fig. 66)
Y-system applied
to comer flanges
Extended X-systems
(see Fig. 75)
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.
one Z system in the affected bay
} (b67a)
four Y systems at the adjoining rib stations
The required four new systems may be taken as,
one stretched Y system
two standard Y systems applied to the four comer flanges
at the adjoining ribs
} (b67)
and the four Y systems acting at the adjoining rib stations
We have a net loss of six standard redundant systems and need hence five
new systems. As such we choose (see FIGS. 74 and 75),
one stretched Y system (as shown in FIG. (66))
two standard Y systems applied to the four comer flanges at
the adjoining ribs
(b68a)
(b68)
one X system
j
Note that the latter X system is in addition to and in the opposite cover
to the one chosen in system (b66). If the cross-section has an axis of
symmetry the X system of (b70) is replaced by an X system.
Finally, we consider the case of a cut-out in the cover (FIG. 77). We lose
here in the affected bay three of the systems
}
two Y systems at adjoining ribs
(b7l)
one Zsystem
the X system of (b66) may be assumed applied to the opposite cover and is
hence not lost. Our choice of two new redundancies is
one stretched Y system in the affected cell
one stretched X system at the affected cover
(b72)
71
('
Cut-out tn cover
Cut-out in cover
Extended
Y-system
'
Extended
--
X-system
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 wih three or more cells the stretched Z systems illustrated in FIG. 78 when an intermediate web or cover is missing.
The above discussion shows that every type of cut-out requires special
consideration and inevitably disturbs the pattern of the b" and b 1 matrices.
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 process) 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 orthogonal or nearly so. Emphasis in the examples has been on the force
method of analysis since it is, in general, more suited to aircraft applications. 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 conditioning of the Eqs., the much greater accuracy in the derived stresses
and the smaller number of unknowns.
72
;;.
Extl'nded !-system
Part II. Applications to Thermal Stress Problems and St. Venant Torsion
By J. H. Argyris and S. Kelsey
1. INTRODUCTION
System X 2 = I;
M 2 =rcos 1/J, N 2 = -cos 1/J
0o
71'
71'
-m<I/J<iii
~ ............
(3)
(4)
2. FIRST EXAMPLE
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
0=y(l+cosmi/J)for
(2)
(I)
(a)
Additional Notation
A : area of cross-section of ring
(c)
73
a0
~eds= --,ds
..........................................
(5)
(7)
and hence
_ slO _ 32o
X1- -3u' X2- -322 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(8)
J
c
S,.,=
M,. =
a.EI
h
JM,~etfs+ JN,7Jds
..................................
(10)
for p = 1 and 2. The sign f( . . )ds denotes integration around the circum-
ference of the ring. Using (2), (3), (5) and (6) in (9) and (10) we find:
dlfl 2TTr
I
Su =r El= El
I
J
c
and
(II)
+n/m
a0.rJ
TTra0 0
3 10 =- 2h (I +cos mlfl )dlfl=- ---,;zjl
-n/m
c )J
+n/m
a0.r 2
h
3 20 =- 2h 1 +z,.
a0.r 2 (
= --h-
-n/m
h) m2
+lr m2-1
El
x1 = -sll =a0o2mk
and
X2=
SID
m
7T
Thus
310
1
j
Mro = a.EI
h
....................................
(13)
(14)
where
h
I
f3=2r and K= r2A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(15)
Substituting (13) and (14) into (4) we obtain the final bending moment M
and normal force N in the actual structure as
M- a0oEI[ 1 2 ~ sin (TT/m) 1 +{3
- 2mh
+ m2-1
TT/m
I+K- cos 'f'']
a0oEI
N=- TTrh
m2-1
m2
SID
+{3
m 11+K
cos lfl
7T
............
(16)
(17)
(l6a)
N- _ a0oEI 1 +fJ
2rh I +K cos lfl (l7a)
It is often possible to neglect the deformations due to normal force
which appear only in 3 22 . To obtain the corresponding formulae we put
in Eqs. (16) to (l7a)
1+{3
l+K=J .... ............ .. ... . ........ ........ (18)
74
e= e.(1+cos31jl)/2
(12)
a0El a0 0 El
Mt= -~~El=-h- =211 (l +cos mlfl)
and at !fl=O
a0 0 El
Mto=-h-
Hence
MM
7T
m ...
7T
(19)
(20)
Comparison of (20) with (l6a) shows that in the case m = 1 the thermal
straining is completely prevented if we ignore the deformations due to
normal force.
Using Eq. (19) in (16) and (17) we find finally for the bending moment
and normal force:
M _ _I_
Mto--2m
[t +2m~sin
(TT/m) 1 +{3
-l
TT/m
I +K
2
cos ']
'f'
(2 1)
......................
(22)
Numerical Applications
The distribution of bending moment and normal force
periphery for a ring with r/h=l2 (i.e. {3=1/24) and K/{12=1
FIG. 2 for the case m=1 and in FIG. 3 for m=3. Values of
-Nr/Mto are plotted radially with the centre line of the ring
around the
is shown in
M/M 10 and
as base.
3. SECOND EXAMPLE
Thermal Stresses in a Circular Ring with Non-Linear Elasticity
Consider again the ring of the first example subjected to the sam:!
temperature distribution. The cross-section of the ring is the idealized
I-section shown in FIG. 4 and the material is assumed to obey a non-linear
stress-strain relationship of the form
E
=i+~(a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(23)
10
--o-- - - . - - - - - - - - -
08
--0--~:>---iJ---
06
Fig. 4.-Second Example: idealized 1-section
where cp (a) is an odd function of a (same stress-strain diagram in compression and tension). Moreover, cp =dcp/da =0 for a =0 which ensures
the usual linear relation for very small stresses (see FIG. 5).
The bending moment and normal force distributions in the ring are to
be found with the assumptions that :
(a) Only the flanges of the section carry direct stresses, which are taken
at any cross-section to be constant in each flange.
(b) The direct strains in the flanges derive both from bending moment
and normal force.
(c) Shear deformations of the web are negligible.
Additional Notation
A 1, A,: area of inner and outer flange respectively
A : total direct stress carrying area
P1, P,: loads in inner and outer flange respectively
The structure is twice redundant as in the first example and as basic
system we take once more that of FIG. Ia. Also as redundancies we select
again the bending moment X 1 and the normal force X 2 at the centre.
To find the unknowns X 1 and X 2 it is convenient to work in terms of
flange loads and strain~.
Flange loads due to X,,= I :
M,,. p ,., = N,,A,+M,,
P,,,= N,,A
---;r-1 -h
A
h ........................ ..
....................................
(25)
(26)
EA;
+'f'
A;
E,-
EA,
+'f'
A,
(28)
(31)
..
.-
(32)
o o
211 -
(33)
are the usual displacement coefficients arising from the linear part, a/E,
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
Eqs. (32) simplify to
o 11 X 1 +J(Pqcp,+P, 1cp,)dl+o 111 - 0
.-
(34)
r2(
m2 sin -7T
aHh
- I + A.
-' -") --.
A r m2 - I
m
............
(35)
)
which are essentially the same as in (12), but for a small difference in 8 20
arising from the geometry of the cross-section chosen.
Furthermore
27Tr
7r1 3
7(1 +K)
................................
(36)
..................................
(37)
where
,.
,-
,-
8 11 = /' o22 =
J(E 1P 12
20
where
o 10 =Ja0P; 1ds, o 20 =Ja0P, 2ds . . . . . . . . . . . . . . . . . . . . . . . . . . . .
820 =
y,- E,
or
15
('
(27)
05
(24)
Note that M,. is taken about the centroid G of the cross-section. Using the
expression for N,, and M,, of Eqs. (2) and (3) in (24) we obtain:
for system X 1 =I:
Pq=-h; P, 1 =+r,
02
+ E,P, 2 )ds +8 20 =0
(29)
c/J,= c/J{AI;h[ -X 1 -
(30)
Dinrtional suffices in the strains arE' omitted siner direet bendinte strains only
auly!i!iis.
K-A~2and/-A,~,It2
OCl'ur
in thl'
X 2 r cos 1{1 (1
+~1 ~)]}
(I -1' ~)]}
(38)
Substitution of (38) into (34) shows that for a given cp-function we obtain
after integration with respect to 1{1 two non-linear equations in the unknown X 1 and X 2 . The solution of these equations determines the complete stress distribution in the ring.
15
e =0.,(1+ cos!JI)/2
01
P=O
p=1
P=2
30
2{)
a0.,EI
M,o - h Fig. 7.-Bending moment and normal force distributions in non-linearly
elastic ring due to thermal strains
..................................
E=i[, +C~r-lJ
or
aE (a)"
cp(a)=-3
a.
(40)
lr ............
(41)
....................................
(42)
pn[x 1 +x2(1-{J)
Jr1
X!= Mto
and
Jr2r
X2= Mto
COS
,Pr~
,Plni
Substituting from Eqs. (25), (26) and (41) in the integrals of (34) we find
I
(44)
1r
which form= l and {3=0 (2rh) take the particularly simple form
76
(46)
Numerical Results
Eqs. (46), which have only one pair of real roots, have been solved
for values of p between 0 and 3 and FIG. 6 shows values of x 1 and x 2
plotted against p. The roots of the simultaneous cubic equations were
obtained by an iteration method, successive values of x 1 and x 2 being
calculated, using the. previous approximations for the terms in the brackets.
For p =0 5, successive values of x 1 and x 2 , calculated in this way are:
X1
05
0433
0452
0449
0449
x2
05
0390
0416
0410
0412
The initial approximation in this case was the linearly elastic solution;
for further values of p, extrapolation of previous results gives satisfactory
starting points. Since successive approximations are alternately greater
and less than the final values of x 1 and x 2 it is quite easy to improve the
natural convergence of the process by a little judicious anticipation.
Direct solution ofEqs. (46) was also obtained. Eliminating x 2 an equation
cubic in p 2x 1 2 and x 1 is obtained, which can be solved for either parameter and the corresponding values of p. x 1 and x 2 calculated.
The bending moment and normal force distribution around the ring is
plotted in FIG. 7 for p=O (linear elasticity), p=lO, and p=20, in the
non-dimensional form M/Mto and Nr/M 10 As would be expected, the
effect of the non-linear stress-strain curve (in which da/dE decreases with a)
is generally to reduce the maximum bending moment produced by a given
temperature rise and also to smooth out the variation around the ring.
4. THIRD EXAMPLE
Thermal Stresses in a Rectangular Plate
(39)
Additional Notation
(45)
~..F(y/b)
~,.
(Vyy}y.o
JJ[.6. 2F+Ea.6.0]SFdxdy=0
b/2
:------'L--- - ' - - - - ' ' - - - ------"'---- ---'---'-------'
t------a--~
_l
-1/2
I ......................
(48)
a . =a.,.=O
is set up. To obtain the stress distribution in a plate of finite length, we can
apply a stress-system -a.,., at the two ends x= a and superimpose the
resulting (diffusion) stress-system on (48). We shall, however, adopt here
a slightly different approach in order to illustrate the application of our
principle of virtual forces in the presence of thermal strains.
Guided by the solution (48) for an infinite plate, we assume that it is
possible to express the longitudinal stress in the form
a.,.,=Ea0J.(TJ)tP(~) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(49)
where
+1/2
f.(TJ)= Jg(TJ)dTJ-K(TJ)
.. .. .. . .. .. .. . .. . .. .. .. . .. .. . . .. .. ..
(50)
-1/2
and tfo is a function only of x. We know that Eq. (49) cannot, in general,
represent the true distribution since the cross-wise distribution will also
vary with x. However, for a smooth function g(y) it can be expected that
(49) will yield a sufficiently accurate answer.
The distribution (49) can be represented by the stress function
F=Ea0,)J2J(TJ)t/J(~) .. .. .. . .. . .. .. .. .. . .. .. .. .. .. .. . .. . . ..
(51)
which gives the stresses
o2F
do/
au= 0 yz=Ea0.dTJ 2t/J(~)=Ea0.J.t/J(~) . . . . . . . . . . . . . . . . . . . . . .
(52)
o 2F
bdl dt/J
a.,.=-oxoy=-Ea0ad1Jd~
o2F
b2 d2t/J
a = 0xz=Ea0.-(i21 de
. . .. . .. . . . . . . . . . . . . . .. . . . . . .
(53)
................................
(54)
The a.,.,, a distributions are symmetrical and the a~ distribution antisymmetrical with respect to the x-axis. From Eq. (52) and the conditions
(a)=b; 2=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (55)
(a.,.)._ 0 =0 (antisymmetrical distribution) . . . . . . . . . . . . . . . . . .
(56)
the function I is :
'1
I(TJ)=J
'I
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)
we obtain from Eq. (58)
J J[ d~d
+I +1'2
d 4 t/J
a2 d 2t/J d 21
a4
d'l]
(61)
I -1/2
J[
+I
g, /., 1: functions of TJ
A, B, C: constants
tfo : function of ~ or '
{J 1, 2 =JLiw: roots of auxiliary equation
Ji=J-Lbfa, iii =wbfa
The thickness of the plate is taken as unity.
The problem can be solved exactly but we shall derive here an approximate solution using the principle of virtual forces in the form discussed in
Part I, Section 6E.
Consider first a plate of infinite length for which the stress distribution
is well known.* Thus, for the symmetrical temperature distribution considered here, the self-equilibrating stress system
+1/2
(58)
.. .. . . .. . . . . . .. . . . .. . . .. . .. . ..
(57)
-1/2 0
d4tfo
a2 d2t/J a'
A d~c2b2B de+IJ4C(t/J-I) Stfod~=O
.. .. . .. .. .. . ..
(62)
(63)
-I
J+ I;2
A= f2dTJ, B= -
rll/2do/
r+
I''Jy
-i![
(68)
A B C are functions solely of the temperature distribution and are ind~pe~dent of a/b. Thus, the length/width ratio of t~e plate _ente~s only as
a common factor in the arguments of the hyperbolic functions m (66).
The tfo functions must obviously be symmetrical with respect to the
centre of the plate (~ =0) and therefore
K 3 =K4 =0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(69)
At the ends of the plate (~ = l) the stresses axr. a.,. are zero and
therefore
tfo= ~~=0
for~= I
....................................
(70)
In general
AC>B2
and the roots {1 1, 2 are therefore complex. They can be written in the form
{J 1 , 2 =JLiw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(72)
where
JL=bii=bnc
v/~+~)r 12
8)]
. .. . .. .. .. . .. .
112
a_ a[l(~ /C
w=bw=b 2 VA-A
j
Substituting (72) into (71) we obtain
JL[cosh JLO +t) sin w(l-~)+cosh JL(l-~) sin w(l-r~)]
tP = I JL sin 2w +w sinh 2JL
w[sinh JLO +0 cosw(l-~)+sinhJL(l-~) cosw(l +~)]
JL sin 2w +w sinh 2JL
See Timoshenko and Goodier, Th~ory of Elasticity, 2nd Edition, McGraw-Hill, New York
1951, p. 399.
(64)
(73)
(74)
77
1 +~~~:1TTJ
l()
lt----+--
(77)
....................................... .
(78)
(79)
12 ...---.-----. ----
also
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 (,..._,1002) 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.
Introducing the co-ordinate
x'=a-x
- 04 r---+r---t----t---+---+---+-----l
x' a-x
-02
{=b="=o-e>b
-06(81)
which is the appropriate solution for the semi-infinite plate. Note that,
for the particular temperature distribution of Eq. (76), ji and w are proportional ton and hence lj>({) is solely a function of n{. The corresponding
complete stress system is :
~_tf21 -I.-
cos2n7TTJ[1-ii sin
2
J.
2n 7TTJ w
-2 +11-z . -r -~<
df d'l'
sm
Ea0. = dTJ d{=- ~
sm we, e
----w-
Un
u1J11
tP!fo +(l+cos2nry) C.02+;i2[_
-r _ . _,.1-;;
Ea0. =1 d{2=
81r2n2
-w w cos w.,-IL
sm we, e '
..................................
Jl
(83)
Maximum stresses:
For any T} the transverse direct stress (u) reaches its greatest value at
the end of the plate ({=0) and is given by
un
.~ /c
2 Ea0. =f(fi. +w 2)=f'V A
The maximum value (at TJ =0) is
(
Uww
-2
-2
max.=~ t~ =1155
27T n
For the maximum shear stress (u.,.)
YJ=l/4, {=0218n
and thus
(u ..) ...,
(uo:w)max. =0 302
(u.. ).=<>
We mentioned on page 77 that it is possible to find the stress distribution in a finite plate by superposing on (48) the diffusion stress-system
arising from -u.,., applied at the two ends x= a. In aircraft structural
theory such diffusion analyses are often based on the simplifying assumption Ew=O. It is interesting now to compare the results obtained by the
78
Assuming
11 =0
I~
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (82)
ji and ware defined in general by Eq. (73) and for the particular temperature distribution of (76) by Eq. (80). The stress system (82) is illustrated in FIGS. 8 and 9 which show the transverse and lengthwise variations respectively of the stresses for the case when n = 1.
u .. is the longitudinal direct stress u:u in an infinite plate
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
an end load stress-system
-u = + Ea 0 cos 2n7TTJ
0
..
...............................
(84)
app~ied at x~ a. Now, in .a plat~ ~ith Ev. =O, the s~ress system (84) is
an etgenload, I.e. the cross-wise vanat1on of Its stresses IS invariant with x.
112 =21Tnba(G)112
E
Ea0.
..5L =
Ea0.
(Q)
(Q) I+ cos 2n7TTJ
E
or for a long plate
cosh ILg
cosh ILD
p. ~~;' Argyris, Dunne, 'The General Theory, etc.', Part V, J,R.Ae.S .. Vol.
__!!iu_ _
Ea0o-
-(Q)
E
112sin
2mr7J .--
2
e 11 Ir
a.. GE l+cos2mr7J
_-.
rae.=-2
e
(86)
1'D
where
5. FOURTH EXAMPLE
St. Venant Torsion of a Thin Section Bar
In treating the torsion of uniform bars of thin section, approximate
expressions for the maximum shear stress and the stiffness may be
obtained by assuming the distribution of shear stress across the thickness to be identical with that for a very wide strip of constant thickness
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
thickness
dB
a .,=-2yGdz
(90)
with the maximum value
dB
a,.,max.=tGdz
and the torsion constant
J=Hy0 3dx
..........................................
(91)
Additional Notation
a, al> b, c, !=dimensions of cross-section (see FIGS. 10 and 15)
a, a 1 =a/b, a 1 /b respectively
Ox, Oy=axes in plane of cross-section
(Ox along axis of symmetry of cross-section)
Oz=longitudinal axis of bar
Yo= y-eo-ordinate of section boundary
g, g1, ?]=non-dimensional co-ordinates (see nu. 15 and
Eqs. (142) and (143))
B' =dB /dz =rate of twist of bar about z-axis
T=applied torque about z-axis
J= T/GB' =~St. Venant torsion constant
w=wB'=warping displacement of cross-section
!>=stress function
/, /1> cf>, 1> 1 =functions of x
n=a,,max./GtB'=maxi mum stress parameter
W. J. Duncan, D. L. Ellis and C. Scruton, 'The Flexural Centre and the Centre of Twist of an
Elastic Cylinder', Phil. Mag., Series 7, Vol. 16, 1953. p. 231.
t After completion and submission for publication of the present analysis the authors' attention was
drawn to: E. H. Mansfield, The torsional rigidity o.fsolid cylinders of double-wedge section, A.R.C.
Report, 1954, to be published in the R. & M. series. In tbis paper the same upper and lower limits
are obtained for the torsion constant of a double-wedge section by a basically identical method.
The stress distribution is, however, not discussed.
For the general background of the St. Venant theory of torsion of solid
sections, the reader is referred to: Timoshenko and Goodier8 , Theory of
Elasticity, 2nd edition, ch. II, McGraw-Hill, 1951.
(a) Principle of Virtual Displacements
To obtain the linear au variation we assume that the warping displacement w of the cross-section can be written in the form
II' _c, ~.()' =yj(X). B'
(92)
' . . ' . . '
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:
a u =G
,,cGB'[~W-)']=,GB'y[
df.-l]
~X
~
(93)
....................
(94)
8,,,~B'y8(:fx) O'_,.t!!.Jt>
-~ . . . . . . . . . . . . . . . . . . . .
J
b,11- 8'8.f
(96)
Substituting from (96), (93) and (94) into (95), integrating with respect
to y between Yo, we find
(97)
x,
(98)
Since 8/ is an arbitrary virtual increment off
d [y}(df
--- --I )] v <!-+ x) "~0
dx -3 dx
"
which is the required differential equation for/, and also
Y/(:fx --
and x 2
(99)
(100)
Torsion Constant J
To determine J by the Principle of Virtual Displacements, we consider
a virtual increment 8B' of the rate of twist of the bar. Then from Eq. (44)
of Part I
8U,: -8W=O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(101)
where
8W=T8B' . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(102)
and
8U;=Jf[a,x8u+a, 118:u]dxdy . . . . . . . . . . . . . . . . . . . . . . . . . .
(103)
In calculating 8U,. due to the increment 80' we note that part of the
virtual strains arises from the corresponding increment of the warping
displacements which are proportional to B' (cf. Eqs. (92), (93) and (94)).
But the increment of strain energy due to any virtual warping displacements is zero (Eq. 95). Hence for the virtual strains giving 8U; in (101)
we can omit the part due to warping and take simply
::::::::::
.. .. .. .. .. .. .. .. .. . .. . .. .. . .. .
(104)
79
Using Eqs. (93), (94), (104), (103) and ( 102) in ( 101 ), and integrating
with respect toy, we obtain for the torque T due to the shear stresses
J[ ~_Y]\1;
x,
T=2GO'
-I )+xyo<f+x)Jdx
(105)
oU;*=G(0')2;{
- J{fx(l~os
x,
a(
df)
_ _I_ -~Jx,
Yo l - dx dx
J~G0'-3
(107)
x,
(121)
................................
Since ocf> is zero at the ends of the section the first term vanishes and
there remains only
oU;* - -G(0') 2
i2~os~~Scf>+y/Jxc/>ocf> I
H {fx(l~o5t)+y.{fx(Yo1x0)--l]c/> }ocf>dx
x,
(122)
Substituting now Eqs. (122) and (123) in Eq. (116) we conclude that in
order to satisfy (116) for any arbitrary ocf>, the function c/> must satisfy the
differential equation
T=ff[a.wx-a,,y]dxdy
which gives on integration by parts with respect to x and y for the first
and second terms respectively and noting the boundary condition (109)
(110)
T=2JJ<f}dxdy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
form
the
We assume now that the stress function <f) can be written in
2
2
(Ill)
.
.
<f}=G0'(y0 -y )c/>(x) .. . ... . . .. . .. . . . .. . ... . .. . . . .. . .
where c/>(x) is a function of x only.t
This satisfies (109) at the upper and lower boundaries and gives our
assumed linear transverse distribution of azx Substituting from Eq. (Ill)
in (110) we find for the torque
T=G0'Hyo 3 c/>dx
......................................
(112)
............................. .......
(113)
The stresses and strains corresponding to (Ill) are from Eq. (108)
a,,=~C:=-2yc/>G0'=GEzx
..........................
dyO,J.] G0'de/>
()<f}-GEzu
- - [ (Yo 2-y2)dx+2Yodx'l'
a ---()x
. . .. . . ..
(114)
(115)
].1.
........... .
(124)
d~2 -p,2f== ph
whence
J=Jo,=Hyo 3 c/>dx
a[
5
dyo) -I 'I'+ Yo 3 =0
d ( Yodx
d (2yo
dx +Yo dx
T de/>)
dx
........................................
where
JL 2= 12/t2 . . . . . . .
The boundary conditions (100) are here
(125)
. . . . . .
(126)
dx =I for x= s/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(127)
(128)
d{
Stress Distribution:
Substituting for /in Eqs. (93) and (94)
,
[ cosh JLX
II
(129)
?~GO'.~ sinh JLX
JL cosh JLS /2
The distribution of shear stress according to Eq. ( 129) is illustrated
graphically in FIG. II.
J ......... .
(118)
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
Assumed C1z
distribution
(120)
See Timoshenko and Goodier, and also Part I, Eqs. ( 5).
t See Part I, Section 6E.
80
tis 020
Fig. !I.-Rectangular s~tion; shear stress distribution according to
Virtual Displacements solution
10
---
Virtual Forces
Virtual
.........
.........
_jJ
---Virtual Forces
~s--C'
04
02
04
-Virtual Displacements
o
tis
06
08
Exact
Solutions
10
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.
We note that in the extreme case of a square section (s/t=l) n=0657,
which is 2 7 per cent less than the exact value 0 675. * An obvious error
in the detailed stress distribution is the constant a .. with y leading to
unbalanced stresses on the boundary.
02
04
t/s
06
08
tO
Torsional Stiflness:
The general expression (I 07) for the torsion constant gives, with the
/-function of Eq. (128)
3 +s/2
s~ll-ta~~ OaJ
(132)
The broken curve in FIG. 13 shows the variation of 3J/st3 with t/s
according to (132). In accordance with the general Principle of Virtual
Displacements this approximate solution must always overestimate the
stiffness.
(b) Solution by the Principle of Virtual Forces with Assumed Form of Stress
Function
The differential Eq. (124) becomes in this case
~~->..2cf>= -,\2
......................................
(133)
where
A2 =10/t2
(134)
(135)
( 136)
Assumed C1zx
distribution
1 .JL
L.........l.... Gt~
tis 020
where
8b=As/2=V~s/t
(139)
The full line in FIG. 12 shows the variation of the maximum stress parameter n with tfs.
Torsional Stiffness:
Eq. (113), in conjunction with the cp function (136), gives for the torsion
constant
J=s~l 1 _tan8~eb]
..................................
040
(137)
which is shown as the full line in FIG. 13. We find remarkably close agreement with values given by the exact solution.* Thus for a square section,
which can hardly be considered as thin, Eq. (141) gives the value 0419st3 /3
which is only 0 7 per cent below the exact value. In accordance with the
general Principle of Virtual Forces our present approximation must underestimate the stiffness.
Stress Distribution:
Substitution for cp from Eq. (136) in Eq. (114) gives for the stresses
=-2y[l-cos\.\x]G8'
cosh 1\S /2
)
,
sinh Ax
(}'
t2
2
(
a = 4 -y 1\ cosh >..s/2 G
""'
I;
............ ......
81
r_
1---------
I _Q_
I Gt .:j
....I..,JIU...,!. L . J
I - !....... . . . . . . ...........
(144)
(l44a)
tic 020
0/C
050
+a.
-a
-a
J-y.(J+x) }otdx
a1
................................
(145)
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
=0
..........
(146)
(150)
Eqs. (148) and (150) are the boundary conditions at the junction of the
two parts of t~e section, giving the compatibility and equilibrium requirements respecttvely.
With x and Yo from (144), the differential Eq. (99) becomes, for the
left-hand part of the section
do/
df
' 2iif2+3'a,-3a 2/=3a'(l +a 2)-3a 2a
..................
(151)
where
a=a/b
The complete solution of ( 151) is
a [
/1-1
-/1-1
f=as_ 1 A'
+B'
_,(a 2+l)+(aLl) ........... .
(152)
(153)
where
fP=l +3a 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(154)
Similarly we find for the right-hand part of the section (substituting
from (144a)) the differential equation
i:. 2do/1
i:. d/1
!>1 d' 12+Js 1~~ -Ja1o/1 = -3a 1, 10 +a 1 2)-i-3a 12a . . . . . . . . (ISla)
and the solution
a1 [
/1,-1
-/1,-1
/1=a 12_1 -A1'1
-B1'1
+' 1(a 12+1)-(a 1 2-l) . .
where
a 1 =a 1 /b, {J 12= I +3a 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(I53a)
(154a)
Boundary Conditions:
In terms of the non-dimensional co-ordinates, the boundary conditions
(146), (148), (150) may be written:
82
--
.. I ...
1 ..sL
I
Gt~
(Oyz)y.o
-~---------~--~~tic 010
(147)
a/C 010
d'
d'
(155)
![df]
+_!_[df1]
=0
a
1 o,= 1
o=l a 1
The first of these gives immediately
B=B 1=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(156)
and substitution from (153) and (153a) in the last two of (155) leads to
A _ 2a({J1 2 +fJ 1-2)+a 1(fJ2+fJ 1-2)
I
({J 1+2)[a({J1-I)+a 1({J-I)]
(157)
A -2a1(fJ2+fJ-2)+a(fJ12+fJ-2)
............. .
1({J+2) [a(fJ 1-I)+a 1({J-I)]
For a doubly symmetrical section (a=a 1), Eqs. (157) reduce simply to
A=A 1=2 ................ ~..........................
(158)
and when a 1 =0, i.e. the section consists of a single isosceles triangle
A=2(l +fJ)/3 . . . . . . . . . . . . . . .. . .. . . . . . . . . . . . . . . . . . . . . .
(159)
Stress Distribution:
From Eqs. (94) the stresses corresponding to the general solution (I 53)
are
a
[
A
(160)
O"zv=-2a2-J' l-2g
bG8'
..................... .
fJ-2]
O"zz=-
2[
a
3A ~:.fJ-2] 8 ,
2a2-)7J
l-2(1+fJ)S
bG
................ ..
(161)
(162)
TABLE I
Comparison of Values of Maximum Stress Parameter n with Exact Soluti,.,.ns
and
O'zn =
aazy-U.rz
(l +a2)1/2
(fl-2)
a 2A
fl-1
,
=(a2-l)(l+a2)112 P+t ~
bGB #0
.. . .. . ... ... ..
1
1 ~ )d~ 1
....... .
(166)
6A1
(f1t+0(f1t+2)J
(167)
Exact Solution
100
0-4n
Virtual Displacements
Solution (a)
100
Exact
0-461
(-35%)
0-4-43
(+l3%)
Virtual Forces
Solution (b)
100
Exact
0-485
(+20%)
v'3/-4=0-433
Exact
Square
Right Angled
Isosceles
Triangle
Equilateral
Triangle
a=Vl, a,=O
a=l.a 1 =0
032lt
v'3/-4=0-433
1=0375
(+ll5%)
t. ==03125
( -60~)
TABLE II
et'
Comparison of Values of 1
Infinitely
Thin
Section
a=a 1 =-l
Equilateral
Trian&le
a~ v'l, a,==O
Richt Ancled
Isosceles
Triancle
ao:.~ I, a 1 =0
100
0-422
0300
01567t
Virtual Displacements
Solution (a)
100
Exact
.,\=0-437
(+37%)
0338
(+ll6%)
... ==01B7S
(+197%)
Virtual Forces
Solution (b)
100
Exact
~~=0406
0300
Exact
,;:,~01563
t/c-->0
Exact Solution
Square
(-37%)
(-016%)
which with the values off and / 1 from (153) and (153a) reduces to
3J
aa 2 [
6A
a 1a 12 [
2ba=aLJ I-~ +a1LJ I
a=a 1 =1
(163)
Infinitely
Thin
Section
t/c= l/(a 1 +a,)-->0
(169)
(170)
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
exact solutions and with the results of the Virtual Forces method.
(b) Virtual Forces Solution With Assumed Form of Stress Function
...........
(171)
08
06
c/12
04
Virtual Forces
02
02
Virtual Displacements
0
Virtual
Exact Solutions
0
04
b/a
06
08
10
Forces
Virtual Displacements
Exact Solutions
08
10
83
..JL
0
1 _g_
Gt:5
Gt.f/
1.. 1.._,......_'""'-'-_._.'-''.._,wl
f/C : 010
ale 050
f/C 020
a/C = 10
.............................. ......
(172)
dc/J+
=[2yo5 dc/>1+
4-1..]
4-1..
dx Yo '+' x-o
5 dx Yo '+' 1 x-o
which may be written more concisely as
!!._(Yo5/2J..)J
[ dx
'+' x-o -[!!_(Yo5/2J..
dx
'+' 1)] x-o =0
............... ..
(173)
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
not possible to satisfy the compatibility condition (equality of *') for all
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
zero and is thus an averaged compatibility condition.
The differential Eq. (124) becomes here, on substituting for x and Yn
from (144), for the left-hand part of the section*
ct!J!
c2t!:.P.
de+ 5~d~+2~( 1 _
__ 2a
~ 2 ....................... .
a2)-1..'+'-
(174)
(175)
6,--2
.. 1),-2]
t/c : 005
(176)
Boundar.y Conditions:
Since <I> must be zero at the ends of the section
on boundary)
<-1
g=g 1 =0 (equilibrium
=0
(173a)
<,~1
For the functions cP and c/> 1 of (175) and (175a) the boundary conditions (171) and (173a) yield
_
a 2(S 1 +l)+aa 1 (S 1 +2)+a 1 2
C- 5a(S 1 +2)[a(2S 1 +1)+a 1(2S+1)]
I
I
C=Cl=2S+I ....................................... .
and when
r-
_I
Stress Distribution:
From Eqs. ( 108) or ( 114) we find, for the function c/J of ( 175), the stresses
b-2] bG8'
a2 [
u,.= -2a2 _ 1l] 1-C~
-2a2a_ 1g{ 1-
s;l-l~ 2 -( 1-Dl] 2]
}bG8'
/j-JJ bG8,
Uzn=O
(181)
I
~
J
(182)
I
J
D=D 1 =0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(177)
Eq. (173) becomes, on substituting for x and Yo from (144) and (144a)
![.!!.
(g512c/J)]
a d~
ate 025
Uz.,=2-
(178)
a(a 2 + 1)1! 2
]1/(1)-2)
a 1 =0
............................................... .
Uz 8 max.
(183)
n= Gt8' = 52(S+2)(S-I) C(S-1)
Similar expressions are ebtained for the right-hand side of the crosssection and need not be given here. Again we find that, in general, the
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
regions. This appears because the variation with y is fixed by the assumed
form (Ill) of the stress function which cannot represent adequately the
effect of the sudden change of slope of the boundary. It does not arise
in the case of the single wedge (a 1 =0) for which we obtain more accurate
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
function* for this case is
(184)
(180)
This equation, _in slightly different form, was used lor the isosceles triangle by Duncan Ellis
and Scuton, /oc. Cit.
'
84
Oyz'
a(a2+J)ll2 [
where
C=l
~~==------~-~--
r""'==..:J::;;..,..,
Uzy=
(175a)
-D1g1
0 2<1>
..................................
(185)
0-9
08
cto/12
06
Torsional Stiffness:
.............. .. ............
a 12
4C
a2 [
12J
8ba=aar::-t 1- 3+2 +atUT!
4C 1
1-sl +2
......... .
(186)
(187)
(188)
(189)
FIG.
Accuracy of Results
In an approximate solution it is always desirable to be able to assess
the order of accuracy achieved. In the two methods applied in this example,
the method of Virtual Displacements gives an overestimate whereas the
method of Virtual Forces gives always an underestimate for the stiffness.
Thus, by using both methods we have succeeded in establishing upper
and lower bounds for the stiffness.
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
are not really suitable for such sections. Nevertheless the agreement
appears to be quite good.
In general the Virtual Forces method appears to give the most accurate
solution, in particular for the triangular section where the general formula
(189) for J gives the exact value at both ends of the range lfa~o and
a= v3. In FIG. 23 we show finally the torsion constant J calculated by the
Virtual Forces method for the general double wedge section. Note that
FIG. 23 shows minimum values of stiffness. For t/c<O 5, however, the
error is less than about 0 5 per cent.
Although the general theorem as to upper and lower bounds does not
apply to the maximum stress, since it depends much more on the details
of the initial approximation, it appears here (see TABLE I and FIG. 18) that
the exact solution does lie between the results given by the two methods.
85