Anda di halaman 1dari 17

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

8, 323-339 (1974)

BUCKLING UNDER COMBINED LOADING OF THIN,


FLAT-WALLED STRUCTURES BY A COMPLEX FINITE
STRIP METHOD
R . J . PLANK AND w. H. WITTRICK?
Department of Civil Engineering, University of Birmingham, England

SUMMARY
A finite strip method is presented for determining the initial buckling stresses of any structure consisting of a
series of thin flat isotropic plates rigidly connected together at their longitudinal edges. Each plate may be
subjected to a combination of longitudinal and transverse compression, longitudinal in-plane bending, and
shear, and it is assumed that the buckling mode, of whatever type, is sinusoidal in the longitudinal direction.
Due to the presence of shear, the perturbation forces and displacements which occur at the edges of component
plates during buckling are out of phase, and this is accounted for by defining their magnitudes in terms of
complex quantities. Stiffness matrices relating the amplitudes of these forces and displacements are derived
using an approximate method based upon assumed displacement functions across the width of the plate. It is
shown how the method can be used to calculate natural frequencies of prismatic structures, and finally an
indication of the accuracy of the method is given together with some illustrative results.

lNTRODUCTI ON
In a recent paper,' Przemieniecki shows how the finite strip method can be used to predict the
initial local buckling stresses of plate assemblies under biaxial compression. His analysis is
applicable to a wide range of structures which consist of a number of thin, flat, rectangular plates
rigidly connected together at their longitudinal edges, and is based upon the usual assumption
that in a local buckling mode all line junctions between component flats remain perfectly straight.
This is undoubtedly a justifiable assumption for this type of mode, but is invalid for longer
wavelength modes, such as torsional or overall modes of stiffened panels, where some component
flats may undergo appreciable in-plane displacements.
In recent years the second author and his colleagues have developed of calculating
the buckling stresses of such structures at any wavelength, using what might be called an exact
finite strip approach. Each component flat is treated as a single element and it is assumed that
all three components of displacement vary sinusoidally along any longitudinal line when buckling
occurs. This enables the partial differential equations governing the in-plane and out-of-plane
deformations of the component flats to be reduced to ordinary differential equations which can
then be solved. Thence stiffness matrices are derived, relating the amplitudes of the sinusoidally
varying forces and displacements on the longitudinal edges of the plate. Each edge has therefore
four degrees of freedom, consisting of a rotation and three translations. These element stiffness
matrices can be assembled into an overall stiffness matrix for the structure, the coefficients of
which are transcendental functions of the load factor and of the wavelength of the mode. The
criterion for buckling is, in general, the vanishing of the determinant of the overall stiffness
matrix, and a theorem has been d e r i ~ e dthat
~ , ~enables the critical load factors to be computed
t Professor.
Received 27 July 1973

0 1974 by John Wiley & Sons, Ltd.

323
324 R. J. PLANK AND W . H. WITTRICK

both automatically and with complete certainty. This theorem is essentially an extension of the
Sturm sequence property of the standard linear eigenvalue problem.
Przemieniecki’s approach‘ utilizes the approximate finite strip formulation of Y. K. Cheung
(e.g. Reference 7) in which the lateral displacement of the strip during buckling is assumed to
vary sinusoidally in the longitudinal direction and according to a cubic polynomial in the direc-
tion of the width. Thus it is essentially a finite element procedure, and leads to an approximate
overall stiffness matrix for the structure. The advantage of this approach over the exact finite
strip method described above is that the coefficients of the overall stiffness matrix are linear
functions of the load factor, and standard eigenvalue routines can therefore be used to extract
the buckling loads. The disadvantage of it is that, in order to achieve sufficient accuracy, it is
almost always necessary to subdivide the component flats into two or more finite strips, so that
the order of the overall stiffness matrix is usually at least twice that arising from the exact finite
strip approach, in which subdivision of the component flats is unnecessary. Moreover the addi-
tional ‘nodes’ introduced by the subdivisions usually increases the optimum bandwidth of
the overall stiffness matrix, and it has been the authors’ experience that the exact approach almost
always involves somewhat less computation time than the approximate one as a result.
Nevertheless, the fact that the approximate finite strip method leads to an eigenvalue problem
of standard type is undoubtedly appealing, and it is the purpose of this paper to extend Przemie-
niecki’s analysis in two ways. Firstly, the assumption that the line junctions between adjoining
plates remain straight during buckling will be abandoned, and each longitudinal edge of the
strip will be assumed to possess four degrees of freedom. It will still be assumed that the mode
is sinusoidal in the longitudinal direction. Secondly, whereas Przemieniecki takes account of
uniform biaxial compression in each plate, we shall admit a more complicated state of stress,
including uniform transverse direct stress, uniform shear and a longitudinal direct stress which
may vary linearly across the width of the strip.
The presence of the shear stress introduces phase differences between the sinusoidally varying
displacements and forces on the edges of the strips, corresponding to the fact that the nodal
lines in the component flats are curved. This implies that the method will give realistic results
for buckling loads when shear is present only if the half-wavelength of the mode is substantially
less than the overall length of the structure. If there is no shear present, however, this restriction
is not necessary provided that the ends of the structure are ‘simply supported’. By this we mean
that, at its ends, each component flat is simply supported as far as out-of-plane displacements
are concerned, constrained against transverse in-plane displacement, but free to displace
longitudinally during buckling. The half-wavelength of the mode is then equal to some integral
fraction of the overall length of the structure, chosen so as to minimize the buckling load.
When shear is present in the component flats the phase differences that occur give rise to
stiffness matrices that are complex and Hermitian rather than the more familiar real symmetrical
ones. This is exactly parallel to the situation that arises from the exact finite strip approach
described previ~usly.~
Finally some possible extensions of the method described in this paper should be mentioned.
Firstly, ifthere is no shear in the plates, the restriction of ‘simply supported’ ends can be removed
by assuming appropriate longitudinal variations of displacements. For example, clamped end
conditions could be dealt with by using clamped-clamped beam vibration modes for the longi-
tudinal variation.’ Secondly, it is conceptually no more difficult to cater for anisotropy in the
component plates, as well as to introduce widthwise curvature, thereby making the method
applicable to stiffened curved panels of laminated construction. Finally the method can also be
used for calculating natural frequencies of vibration as well as buckling loads and it is shown
later how the present analysis can be very simply modified to allow for this.
A FINITE STRIP METHOD 325

DEFINITION O F T H E PERTURBATION STIFFNESS MATRICES


O F AN INDIVIDUAL STRIP
The basic state of stress in an individual strip at the onset of buckling is shown in Figure 1.
It consists of a uniform transverse compressive stress oT,a uniform shearing stress T, and a
longitudinal compressive stress which varies linearly across the width b of the plate from (oL- oB)
at edge 1 to (oL+ a B )at edge 2. Thus oL is the average longitudinal compressive stress and foB is
the stress due to longitudinal in-plane bending.

Figure 1. The basic state of stress in a wall

In addition, if the structure has buckled, the strip will be subjected on its edges to a system of
perturbation forces, distributed sinusoidally in the longitudinal direction with a half-wavelength
2, as shown in Figure 2. The perturbation forces per unit length of edge are defined as

where

and Re { } denotes the real part of the quantity inside the brackets.
All edge forces and displacements t o be
multiplied by exp(im/h)

Figure 2. Sinusoidal system of perturbation edge forces and displacements due to buckling
326 R. J. PLANK AND W. H . WITTRICK

The corresponding perturbation displacements of the edges are similarly defined as


Re { ( + 1 w 1 ~ 1 ~ 1 + 2 ~ 2 ~ei5>
2~2)

The quantities m , , +1,etc are in general complex due to spatial phase differences that arise
from the presence of the shearing stress z.
For infinitesimally small buckling displacements the in-plane and out-of-plane effects are
uncoupled and can be considered separately. Accordingly we define perturbation force and
displacement vectors by the equations

where suffices 0 and I denote out-of-plane and in-plane respectively, and in general all four of the
vectors defined by equations (2) are complex.
Perturbation stiffness matrices so and s, may now be defined by the equations
Po = sodo, PI = s,d, (3)
It will be seen later that, if z # 0 the stiffness matrix so contains complex off-diagonal elements
andis Hermitian in form, whilst ifz = Oit is real and symmetrical. The introduction ofi (=,/( - 1))
in the in-plane force and displacement vectors pI and dl automatically incorporates a 90 degree
phase difference between the u and u displacements, and between the px and p,, forces. As a result
of this the stiffness matrix sI is entirely real and symmetrical. This device is not absolutely neces-
sary but without it the matrix sI contains some elements that are imaginary and is Hermitian in
form.

DERIVATION OF THE OUT-OF-PLANE STIFFNESS MATRIX SO

In deriving the out-of-plane stiffness matrix so we follow standard finite element procedures
based on virtual work. However the mathematical detail is slightly more complicated than usual
because po and do are in general complex.
Let the two edges of the strip undergo virtual displacements corresponding to a variation
Sd, in the vector do. The actual (real) virtual edge displacements are Re {ad, ei5}whilst the edge
forces are Re {po eit}. Thus the virtual work SW', done by the edge forces over a wavelength 21
is given by
sZn
S W,, = @/a) Re {ad; ei'} . Re {po eic}d t
0
(4)

where superscript T denotes the transpose.


To simplify this expression consider the following. Let a and b be any two n x 1 complex vectors
and let G be any real n x n matrix. The elements of a, b and G are assumed to be independent of
5. Let a,, ai and b,. bi be the real and imaginary parts of a and b respectively. Then
so2'Re (aTeit). G . Re {b eir}d5

= (a,? cos 5 - a: sin ()G(b, cos 5 - bi sin 5) d<

= a(afGb, + aTGbi)
= a Re {H'Gb}
A FINITE STRIP METHOD 327

where the bar denotes the complex conjugate. This result will be used extensively in the subsequent
analysis.
Using equation (5), and interpreting G as the unit matrix, we may now write equation (4) as

SW,, = 1Re {G?iXp,} (6)


The lateral displacement w of the plate is assumed to be given by the equation
w = Re { Z d , eir} (7)
in which Z is a 1 x 4 interpolation matrix defined by

where
rl = 2Y/b (9)
It is readily verified that this satisfies the kinematic conditions of deflection and slope at the
longitudinal edges of the strip. It is in fact identical with the interpolation matrix assumed by
Cheung' and Przemieniecki' when due account is taken of the origin of y in Figure 2.
Now define a 3 x 1 vector K of curvatures and a corresponding 3 x 1 vector M of internal
moments in the plate by the equations
K = {a2w/ax2 a2w/ay2 2a2w/axay) (10)

Then

where D = Eh3/12(1- v 2 ) is the flexural rigidity of the plate and

Using equation 7), equation 10) becomes


K = Re {Xdo eir} (14)
where x is a 3 x 4 matrix defined by

($1 [ . z ]
- R2Z

=
4iR Z'
In this equation
R = nb/1

and the primes on Z denote differentiation with respect to q.


328 R. J. PLANK AND W. H. WITTRICK

The internal virtual work Syoin a wavelength 21 of the strip due to the virtual displacement
&lomay now be expressed as

GK’FK d t dy

Re { 6diXTei5}F. Re { x do ei5}d t dy

Performing the integration with respect to by means of equation (5) we obtain


SW,, = A Re (6d:A, do)
where

On using equations (13) and (15) this becomes


+1
A, = (D/2b3) { R4ZTZ+ 8R22”Z’ + 16 Z T Z- 4vR2(ZTZ)”}dq (18)
-1

During the virtual displacement the basic membrane forces acting on the four edges of a
rectangle of width b and length 21 also do work 6 W,, . This is given by the equation

where ox is the local value of the longitudinal stress, given by


0
, =oL + VaB (20)
On transforming x and y into the dimensionless variables 5 and y and taking the variation of
the integrand, we find that

Now from equation (7)


aw/dt = Re { iZd, eir}, aw/dy = Re { Z’d, ei5}
If these expressions are substituted into equation (21) and the integration with respect to t is
carried out by means of equation ( 5 ) the following expression is obtained
6 W,, = 1 Re { ddi(B, + iC,)d,} (22)
where

and
+1
C, = (Rhz/b) (Z“Z - Z’Z’) dy
-1
A FINITE STRIP METHOD 329

We now use the Principle of Virtual Work to obtain the equation


swe,+swm,= sy,
and on substituting equations (6), (17) and (22) and noting that Sd, is arbitrary it is found that

Po = sodo (25)
where
so = A, - B, - iC,

I t is easily seen from equations (18), (23) and (24) that A, = A:, B, = B: and C, = -C:.
Thus A, and B, are symmetrical whilst C, is skew symmetrical and it follows that so is complex
Hermitian. I f z = 0 then C, is null and so is real and symmetrical.
Finally we substitute equations (1 8),(23)and (24)into (26)and rearrange the resulting expression
to give
+
so = ( D / b 3 ){ H, - vQ2H, R2(R2- k,)H, + (2R2- k,)H, - R2k,H, - iRksH,} (27)
where kL, k,, kB and k, are dimensionless stress coefficients defined by the equations
(kL1k,, k,, ks} = (b2h/D){oL,o T 9 oB,z } (28)
and the six 4 x 4 matrices appearing on the right hand side are given by
+1
Z " T Zdq, H, = 2((ZTZ)'}?
1

Expressions for these six matrices, obtained by substituting equation (8) and carrying out the
integrations are listed in Appendix 1.

DERIVATION OF THE IN-PLANE STIFFNESS MATRIX S,

The derivation of the in-plane stiffness matrix s, is based on an exactly parallel argument t o that
of the out-of-plane stiffness matrix. From equations (2) it is seen that the actual (real) edge forces
and displacements are Re { Jp, ei5}and Re { Jd, ei5}where J is a 4 x 4 diagonal matrix defined by

Thus in a virtual displacement Sd, the virtual work done by the edge forces over the wave-
length 21 is

SW,, = (@) Jo2ff Re {dd:J ei5}Re { Jp, eic}d<

and on using equation (5) and noting that JJ = I where I is the unit matrix, this becomes

S We, = I Re { ddTp,} (31)


?
330 R. J. PLANK AND W. H. WITTRICK

The displacements u and D are assumed to vary over the strip according to the equations
u = Re (XJd, eic)
D = Re (YJd, eic}

where X and Y are 1 x 4 linear interpolation matrices defined by the equations

x = [O 31-44 0 #+dl
(33)
y = [31--q) 0 % I + ? ) 01
Now define a strain vector e by the equation
e= {--(-+-)}
au aua0 av
(34)
ax ay ay ax

and let CJ be the corresponding stress vector, so that


t~ = E‘Fe (35)
where E’ = E/( 1 - v 2 ) and F is the matrix defined by equation (1 3).
From equations (32) and (34) we find that
e = Re {TJd, eit;)

where T is a 3 x 4 matrix defined by


iRX
r= (i) 2Y’ (37)
2X’ + iRY

The internal work GM/., in a wavelength 2A of the strip is given by

GeTFed<dq =- Re (Gd:JTT eii)F Re {TJd, ei<)d{ dq

and using equation (5) this becomes


dy, = I Re (&l:A,d,) (38)
where

On substituting equations (13) and (37) it is found that

A, = 1-
(E‘h/2b)
+1

1
J(R2XTX+ s l - v)R2YTY + 2(1 - v)X’~X’+ 4Y’TY
+ ~ ~ v R ( Y-’ ~XTY’)
X + i( 1 - v)R(XfTY- YTX’))J dq (39)
A FINITE STRIP METHOD 33 1

I t is readily seen that AT = A, so that A, is Hermitian. However, despite the fact that the inte-
grand of equation (38) appears to be complex, it is in fact entirely real, the complex terms having
been removed by the matrix multiplications involving J and J . Thus A, is real and symmetrical.
I f we are to be able to use this theory to calculate longer wavelength buckling loads, such as
torsional or overall buckling loads of stiffened panels, it is clearly necessary to take account of
the in-plane destabilizing effect of the basic stress system. However, it is intuitively obvious that
oL is the only stress component that need be considered for this purpose, because there is no
conceivable in-plane instability that can arise from the action of the stresses oB,oT and z. To
calculated the loss of potential energy of the stress o, we make use of the following non-linear
expression for the longitudinal strain E , .
Ex = -+-
au 1
ax 2
{(-)’+(;)’+(g)’}
au
ax
The first term has already been accounted for in deriving 6yIwhilst the last term, involving
w, has been used in equation (19) in the calculation of the loss of potential energy of the basic
stresses due to the out-of-plane displacement. Thus, the appropriate expression for the loss of
potential energy of the basic stresses due to the in-plane displacements is

where the integral extends over the width b and wavelength 21. Substituting equations (32) and
proceeding as before gives
6 W,, = 1Re { Gd:B,d,} (42)
where

B, = (Q2a,h/2b)
J-+ll +
J(XTX YTY)Jdq (43)

Now XTX contains terms only in the second and fourth rows and columns whilst YTY contains
terms only in the first and third rows and columns. Thus the premultiplication by J and post-
multiplication by J in the integrand of equation (43) has no effect and hence B, is real and
symmetrical.
We now use the Principle of Virtual Work to obtain
6We,+6W,, = 64,
and on substituting equations (31), (38) and (42) we find that
PI = SId,
where
S, = A,-B,
Substitution of equations (39) and (43) into (45) gives finally

s, = (E‘h/2b) 1-
+1

1
J (Q’( 1 -&,)XTX +$a’(1 - v - 2c,)YTY

+ 2( 1 - v ) X T X +4Y’TY’+ ~ ~ v Q ( Y-’ ~XTY’)


X + i( 1 - v)Q(XfTY- YTX’)}J dq
332 R. J . PLANK AND W . H. WITTRICK

where

Explicit expressions for the elements of s,, found by substituting equations (33) into (46) and
carrying out the integrations are given in Appendix 11.

MODIFICATION TO ENABLE NATURAL FREQUENCIES OF VIBRATION


TO BE CALCULATED
The stiffness matrices so and sI can be converted very easily into dynamic stiffness matrices to
enable them to be used for the determination of natural frequencies of vibration.
For this purpose we interpret the perturbation forces and displacements shown in Figure 2
as being the maximum values that occur during a vibration, so that they all have to be multiplied
by cos 2 m t , where n is the frequency. Now from equation (40)the loss of potential energy asso-
ciated with the longitudinal stress o, is

io,h s[ {(g)’+ (g)’+ (g)’} dx dy

whilst the kinetic energy is equal to

where pis the density of the plate. This neglects the kinetic energy associated with rotations ofthe
normal to the surface.
Along any longitudinal line, the displacement u, for example, is equal tof(y) cos (nxli) cos 2nnt
where some suitable origin for x has been chosen. Hence the amplitude of (du/dt)’ is equal to
4A2n2 times the amplitude of (duldx)’. Precisely the same result holds for the terms involving
u and w. It follows from this that the inclusion of dynamic terms can be allowed for simply by
+
replacing o, by (or. 4pA’n’).
Hence the dynamic stiffness matrix so is defined by equations (27)<29), and the only modifi-
cation required is that k , is now interpreted as
k, = (b2h/D)(o,+4pizn2)
There is no change in the definitions of k,, k , and k,.
Also the dynamic stiffness matrix sI is obtained from the equations of Appendix 11, with E,
now defined as
EL = ( 1 - v2)(a, + 4pA’n’)/E
It might be argued that, since (duldx) is a strain, the terms involving (du/dx) in equation (41)
can be ignored without causing any significant loss of accuracy. If so, however, it can be seen from
the argument outlined above that this very simple transformation from static to dynamic stiff-
ness matrices would not be valid unless the kinetic energy associated with u was also neglected.
In many cases this would be quite justifiable since it is intuitively equivalent to neglecting
rotatory inertia of the cross-sections of the panel. However the inclusion of the terms involving
u involves so little computation that they might just as well be retained.
A FINITE STRIP METHOD 333

ACCURACY OF THE STIFFNESS MATRICES


The accuracy with which the stiffness matrices derived in the preceding analysis can predict
the buckling loads of single plates will now be investigated. Consider first the case of in-plane
buckling, for which the stiffness matrix sI is relevant. I f the two edges of the plate are entirely
free, one would expect that for large values of A/b, i.e. for small values of R,the in-plane buckling
stress would be approximately equal to that given by the Euler formula, i.e.

oL = n2Eb2/12A2

which corresponds to
EL = (1 - v2)Q2/12

where b is the width of the plate.


The values of the buckling stress given by one, two, three and four element representations
of the plate, using the stiffness matrix s I ,have been investigated analytically and it was found that
in all four cases the ratio r of the buckling stress so derived, to the Euler buckling stress, is given by

r = 1+
V2

N 2 ( 1 - v2)
+ O(R2)
where N is the number of elements into which the width is subdivided. Thus as A -+ 00, i.e. as
R2 -+ 0, the errors involved for a Poisson’s ratio of 0.3 are 9-90, 2.47, 1.10 and 0.62 per cent for
N = 1,2, 3 and 4 respectively.
I t is of interest to observe that, for N = 1, r = 1/( 1 - v2) as Q2 -+ 0. The reason for this is easily
apparent ; since the mode is antisymmetrical, u1 = u 2 , and since the interpolation function Y
(equations (32) and (33)) is a linear one it follows that the assumed u is independent ofy. Thus the
strain au/dy is zero and the effective Young’s modulus relating the longitudinal stress and strain
is E/( 1 - v2), as compared with E in the Euler formula.
Now consider the initial out-of-plane buckling of a single long plate with simply supported or
clamped longitudinal edges, in which case only the stiffness matrix so will be involved. Hence an
indication of the accuracy of so can be obtained by examining the buckling of individual plates
under various loading conditions, and comparing the results thus obtained with known values.
The results ofsuch an analysis for both simply supported and clamped plates in pure compression,
pure shear and pure bending are shown in Tables 1-111. The wavelengths at which the calculations
were performed are those corresponding to the minimum buckling stress in the exact solution.

Table I. Buckling coefficients ( k J X 2 ) for single plates in pure compression

Simply supported plate Clamped plate


No. of strips (A/b = 1.0) (A/b = 0.661)

1 4.2583
2 40086 7.2261
3 4.0017 7.0280
4 40005 6.9908
6 44MO 1 6.9753
8 4@000 6.9724
Exact results’ 4000 6.9709
3 34 R. J. PLANK AND W. H. WITTRICK

Table 11. Buckling coefficients k,/n2 for single plates in pure shear
Simply supported plate Clamped plate
No. of strips (L/b = 1.252) ( l / b = 0.830)
1 7.1666 -
2 5~1110 9.5545
3 5.4407 9.4144
4 5.3703 9.0994
6 5.3432 9.0006
8 5.3385 8.9850
Exact results' 5.3363 8.9778

Table 111. Buckling coefficients k$n2 for single plates in pure bending

Simply supported plate Clamped plate


No. of strips (L/b = 0.667) ( l / b = 0.47)

1 27.4159 -
2 25.4544 49.7601
3 24.2 195 42.5 154
4 23.9654 40.36 13
6 23.8969 39.7240
8 23.8866 39.6144
Comparative results' 23.9 39.6

These results clearly indicate the accuracy which can be achieved by subdividing the plate width
into a relatively small number of strips. Note that a clamped plate cannot be represented by a
single strip since there would then be no degrees of freedom and hence a buckling analysis would
not be possible.
In view of these results, four elements have been used to represent each component plate of
the structures analysed in the remainder of this paper.

SOME ILLUSTRATIVE RESULTS


Based upon the theory outlined herein, a computer program has been developed to determine
the initial buckling stresses of thin walled structures. The results of investigations for various
cross-sections and loading conditions are given below as an illustration of the present method.

Channel section
The initial buckling stresses cBofcertain channel sections, each with constant thickness h and
web depth b,, equal to 50 h, and subjected to bending about the major principal axis, have been
determined. The results for channels with different flange widths are shown in Figure 3 as a series
of curves of a,/E versus Ilb,.
The curves for all channels exhibit the same characteristics, namely two limbs, the first of which
has a minimum value at I / b , between 1 and 2. In this region the buckling mode is predominantly
local. As the wavelength increases, the curves rise to a peak, and the buckling mode becomes
mixed between local and flexural-torsional. Beyond the peak the buckling stress decreases with
increasing 3, and the mode is predominantly flexural-torsional, as predicted by Vlasov theory.
A FINITE STRIP METHOD 335

10

Bending
8 moment

.
-

<
D
6

b* 5

0.2
0.3
2
0.4
as
05
a,
1

0
0.5 1 2 4 8 16 32
A/h, logarithmic scale

Figure 3. us/E versus A/bw for channel sections in pure bending

Interaction curves for shear and compression


The local buckling ofa square tube and an equilateral triangular tube under combined torsion
and compression, and a trough stiffened panel under combined shear and compression have been
investigated. The results are shown in Figures 4 and 5 in the form of interaction curves between
cL/cL0and 7/z0 where cLoand 70 are the values of compressive and shearing stresses respectively
which, when applied alone, would cause buckling of the structure.
The interaction curves are applicable to short wavelength buckling, and were derived in the
following way. The ratio between the compressive stress g L and the shearing stress 7 was fixed,
and a load factory applied to both. Critical values ofy were computed over a range of wavelengths,
and a curve of y versus A was plotted as in Figure 6. The minimum value of the load factor in
the local buckling range was read directly from the curve, and hence the critical values of cLand
t computed, enabling one point on the interaction curve to be plotted. Further points were ob-
tained by repeating the process for various ratios of cL:7,and in this way the interaction curve
was plotted.
The interaction curves for the square and triangular tubes compare almost exactly with those
presented by Wittrick and Curzon" who showed the discontinuity in the curve for the square
tube to be due to a mode change. This is confirmed by the present analysis.
336 R. J. PLANK AND W. H. WITTRICK

-T
SO,
~ ,Square tube

CO

Figure 4. Interaction curves between shear and compression for local buckling of square and triangular tubes

0
' I
=0

Figure 5. Interaction curve between shear and compression for a trough stiffened panel

The interaction curve for a trough stiffened panel also exhibits a discontinuity, which in this
case was found to be due to a switch from one minimum at A/b = 0.6 to another at A/b 21 1.2.
I t is also of interest to compare this curve with the parabola ( ( T J O ~+~( )T / T ~ ) =
~ 1. Although this
parabolic interaction formula gives accurate results for a single plate,8 it is quite clearly a very
conservative approximation for this panel. Note that the distribution of shear stresses in this
panel (Figure 7) is determined by the condition ofzero twist in the closed cell formed by the trough
stiffener and part of the skin.
A FINITE STRIP METHOD 337

0
025 05 1 2 4 8 16 32
Ah logarithmic scale

Figure 6. Typical variation of load factor with wavelength for use in derivation of interaction curves

0.4047 0.4041

0.4047

Figure 7. Shear stress distribution in the walls of a trough stiffened panel

Interaction curves for bending and shear


The buckling of a panel with four unflanged stiffeners under the combined action of shearing
stresses T and a compressive stress which varies linearly from - oB at one edge to oB at the +
other has been investigated, and an interaction curve showing critical combinations of z and oB
is shown in Figure 8(a). The interaction curve for the same panel in combined shear and uniform
compression is shown in Figure 8(b). There are no previously published results with which these
can be compared, but for the case of an unstiffened plate in combined shear and compression,
a parabolic interaction curve oL/oLO +(T/T,,)’ = 1 is generally assumed, and for combined bend-
ing and shear, the interaction curve is often taken as a unit circle (oB/pBO)’(z/zJ2 = 1. Com- +
parison between these formulas for an unstiffened plate, and the interaction curves obtained using
the present method (Figure 8) clearly shows that the former are conservative approximations to
the true behaviour.
338 R. J. PLANK AND W. H. WITTRICK

Figure 8. Interaction curves for local buckling of a panel with four unflanged stiffeners: (a) in combined shear and
in-plane bending; (b) in combined shear and compression

APPENDIX I
Expressions for the six matrices appearing on the right hand side of equation (27) and defined
by equations (29) are given below.

pb2 22b -36' 13bl 4b2 3b -b' -3t


156 -13b 36 3b -36
H, = (did H, = (&
4b' -22b 4b' -3t
LSymm. 156 1 Symm. 36

I"' -86
-84
0
b
b'
b-
0
-8b H, = (&I
i"
-6b
0 -6b
-b2

0
6b'
30
- 6b

Symm.
Symm.
84 0

4b2 6b 2b' -6b1 b 0 0'


12 6b -12 0 0 0
H, = H, =
4b2 -6b 0 -b
Symm.
A FINITE STRIP METHOD 339

APPENDIX I 1
Expressions for the elements s,(m, n) of the real symmetrical in-plane stiffness matrix sr lying on
or above the diagonal are given below. ( mis the row number and n the column number)
Sr(1, 1) = ~,(3,3)= (E‘h/b){1 +(1-~-2E,r)R~/6}
+
~,(2,2)= ~ , ( 4 , 4 )= (E’h/b){ (1 - ~ ) / 2 (1 - &L)R2/3}
sr(1,2) = - sr(3,4) = (E’h/b)(1 - 3v)R/4
Sr(l,3) = (E’h/b)(- 1 +(l-v-2&L)R2/12}
sr(1,4) = -Sr(2, 3) = -(Eh/b)(l+v)R/4
~ , ( 2 , 4 )= ( E h / b ) {- ( 1 - ~ ) / 2 + ( 1 - ~ $ 2 ~ / 6 }
where
= aL/E‘ = (1 - V’)O JE

REFERENCES
1. J. S. Przemieniecki, ‘Finite element structural analysis of local instability’, AIAA J . 11, 33-39 (1973).
2. W. H. Wittrick, ‘A unified approach to the initial buckling of stiffened panels in compression’, Aeronautical Quarterly,
19,265-283 (1968).
3. W. H. Wittrick, ‘General sinusoidal stiffness matrices for buckling and vibration analyses of thin flat-walled struc-
tures’. Int. J . Mech. Sci. 10, 949-966 (1968).
4. F. W. Williams and W. H. Wittrick, ‘Computational procedures for a matrix analysis of the stability and vibration
of thin flat-walled structures in compression’, Int. J. Mech. Sci. 11, 979-998 (1969).
5 . W. H. Wittrick and F. W. Williams, ‘A general algorithm for computing natural frequencies of elastic structures’,
Q.J. Mech. and App. Math. 24,263-284 (1971).
6. W. H. Wittrick and F. W. Williams, ‘An algorithm for computing critical buckling loads of elastic structures’, J.
Struct. Mech., 1,497-518 (1973).
7. M. S. Cheung and Y. K. Cheung, ‘Natural vibrations of thin, flat-walled structures with different boundary condi-
tions’, Univ. of Calgary, Dept. Civ. Engng. Research Report (1971).
8. W. H. Wittrick and P. L. V. Curzon, ‘Stability functions for the local buckling of thin flat-walled structures with the
walls in combined shear and compression’, Aeronautical Quarterly, 19,327-35 1 (1968).
9. F. Bleich, Buckling Strength of Metal Structures, McGraw-Hill, New York, 1952.
10. W. H. Wittrick and P. L. V. Curzon, ‘Local buckling of long polygonal tubes in combined compression and torsion’,
Int. J. Mech. Sci. 10, 849-857 (1968).

Anda mungkin juga menyukai