You are on page 1of 10

ARTICLE IN PRESS

Thin-Walled Structures 44 (2006) 507516


www.elsevier.com/locate/tws

Stability analysis of simply-supported rectangular plates under


non-uniform uniaxial compression using rigorous and
approximate plane stress solutions
P. Jana, K. Bhaskar
Department of Aerospace Engineering, Indian Institute of Technology, Madras, Chennai 600 036, India
Received 13 July 2005; accepted 26 April 2006
Available online 22 June 2006

Abstract
The non-classical problem of buckling of a simply-supported rectangular plate due to various types of non-uniform compressive edge
loads is analysed here. For each case, the elasticity solution for the internal in-plane stress eld is obtained rigorously using a
superposition of Airys stress functions and also approximately using extended Kantorovich method. Subsequently, the convergent
buckling loads are obtained using Galerkins method. Results are presented to highlight the dependence of the total buckling load and
the corresponding buckled shape on the edge load distribution, as well as to illustrate the applicability of the approximate plane stress
solutions.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Rectangular plates; Buckling; Non-uniform loading; Exact plane stress solutions; Extended Kantorovich method

1. Introduction
Compared to the attention devoted to the classical
problem of buckling of a rectangular plate due to normal
and shear loads uniformly distributed along the edges and
the corresponding literature on exact and approximate
solution techniques, studies on non-uniformly loaded
plates have been very meagre. However, plate problems
are often idealisations of portions of a much larger overall
stiffened or built-up structurean aircraft wing or a ship
or a multi-storeyed building, for instance, and hence the
loads that cause buckling are those exerted by the adjoining
free-body on the plate; thus, uniform loading is an
exception rather than the rule because the elastic forces
between the free bodies depend on their relative stiffnesses.
Since such elastic force distributions are expected to vary
from one structure to another and cannot be captured
without a formal complicated analysis of the complete
structure, it is necessary to analyse plates subjected to
Corresponding author. Tel.: +11 91 44 22574010;
fax: +11 91 44 2350509.
E-mail address: kbhas@ae.iitm.ac.in (K. Bhaskar).

0263-8231/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2006.04.009

various types of simple, assumed edge load distributions so


as to understand their qualitative and quantitative inuence on the buckling behaviour.
The simplest case of non-uniform loading is that of a
linear antisymmetric variation along a pair of opposite
edges to cause in-plane bending; in this case, as for uniform
edge loading, the solution for the plane stress problem is
trivial in that the internal stress eld coincides with the
applied load distribution. While approximate solutions for
the corresponding buckling problem have been discussed
even in classical textbooks [1,2], an exact series solution has
been worked out for this case only recently [3].
The other cases which have been studied by various
investigators are those of a pair of concentrated collinear
forces on opposite edges, and symmetrically applied
uniform partial edge loads. While early solutions were
based on either simplied approximate plane stress elds
[4,5], or by directly invoking the simple energy method
[6,7], the use of a more complicated functional that
obviates the need for plane stress analysis has also been
explored [810].
Studies on plates under non-linear compressive loads
have been very few. As pointed out by Bert and

ARTICLE IN PRESS
508

P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

Fig. 1. Various load cases considered.

Devarakonda [11] recently, the early solutions for plates


under uniaxial compressive loading with sinusoidal [12]
and parabolic variations [13] were based on the totally
unrealistic assumption that only the corresponding normal
stress would be present in the interior as well and that even
this normal stress would be same at all sections parallel to
the loaded ends. This deciency was overcome in Ref. [11]
by the use of a more realistic, yet approximate, in-plane
stress eld, and it was shown that all the three in-plane
stress components need to be accounted for in order to
obtain a correct value of the critical load.
In the light of the above discussion, the objective of the
present work is to study the buckling of rectangular plates
due to a set of simple edge load distributions causing
uniaxial compression (Fig. 1), without compromising on

analytical rigour and accuracy, so that a set of standard


results can be generated for future use in judging the
applicability of approximate solutions. Towards this end,
the plane stress analysis is carried out using a superposition
of three Airys stress functions taken in appropriate
Fourier series form such that all the eld equations and
boundary conditions are satised, and later the corresponding buckling loads are obtained using Galerkins
method with a complete set of admissible functions which
ensure convergence.
Further, since closed-form expressions for the in-plane
stresses are preferable to innite series counterparts for use
in the subsequent buckling analysis, such closed-form
expressions are obtained using the extended Kantorovich
method based on the principle of minimum complementary

ARTICLE IN PRESS
P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

energy, and their utility in yielding reasonably accurate


buckling loads is examined. Results are also generated
using a commercial nite element package and compared
with those of the analytical solutions.
2. Rigorous plane stress solution
The methodology adopted here is based on superposition of three Airys stress functions taken as Fourier
series, as discussed by Little [14], and illustrated by
Gorman and Singhal [15] with reference to vibration of a
tensioned membrane.
2.1. Building block 1

1
X
r1;3;5...

sr cos

2.2. Building block 2


The second building block is due to shear stresses applied
at y b=2 and represented by the Fourier sine series
npx
X
,
(6)
An sin
txy2 
a
n1;3;5...
where An are undetermined coefcients.
By a similar procedure as for building block 1, and by
applying the zero normal stress and imposed shear stress
boundary conditions at y b=2, one can obtain the stress
function f2 for this case as
npxh
npy
npyi
X
f2
cos
C 1n cosh
C 4n y sinh
,
a
a
a
n1;3;5
(7)

For the general case of a rectangular plate (a/2pxpa/2,


b/2pypb/2) subjected to uniaxial load symmetric about
x-axis, an appropriate series representation is taken as
sx

509

rpy
.
b

(1)

Substituting the Airy stress function f1 given by


rpy
X
f x cos
f1
b
r1;3;5

where C1n and C4n are given in Appendix A. This solution


gives a residual stress (t2) at x a=2, where
nph
npy

X np
np
sin
t2 
C 1n C 4n sinh
a
2
a
a
n1;3;5
npyi
np
C 4n y cosh
.
8
a
a
2.3. Building block 3

(2)

into the governing biharmonic equation r f 0, one can


obtain the general solution for the function f(x) as
rpx
rpx
X
f x
C 1r cosh
C 2r sinh
b
b
r1;3;5
rpx
rpx
C 3r x cosh
C 4r x sinh
.
3
b
b
in which C1r through C4r are constants which are to be
obtained from the boundary conditions.
It has to be noted that this stress function solution
automatically gives a zero normal stress at y b=2.
Enforcement of the zero shear stress boundary condition as
well as the normal stress distribution as dened in Eq. (1)
at x a=2, yields
rpyh
rpx
rpxi
X
f1
cos
C 1r cosh
C 4r x sinh
,
b
b
b
r1;3;5
(4)
where C1r and C4r are given in Appendix A.
However, the above in-plane stress solution gives a
residual shear stress (t1) at the y b=2 edges as given by
rph
rpx

X rp
rp
sin
t1 
C 1r C 4r sinh
b
2
b
b
r1;3;5

i
rp
rpx
C 4r x cosh
.
5
b
b
To eliminate these residual stresses two additional stress
function solutions (building blocks) are superposed on the
rst solution.

The third building block corresponds to imposed shear


stresses at x a=2 as given by
mpy
X
txy3 
Bm sin
(9)
b
m1;3;5...
where Bm are undetermined coefcients.
The stress function for this building block has the same
form as that of the rst; the constants however, will take on
different values when the condition of zero normal stress
and applied shear stress are enforced along x a=2. The
nal solution is given by
mpyh
mpx
X
f3
cos
C 1m cosh
b
b
m1;3;5
mpxi
C 4m x sinh
,
10
b
where C1m and C4m are given in Appendix A. This solution
gives a residual stress (t3) at y b=2 as given by
mpx

X mp mph
mp
sin
C 4m sinh
t3 
C 1m
b
2
b
b
r1;3;5


mp
mpx
x cosh
C 4m
.
11
b
b
2.4. Superposition
Now by superposing the three building blocks and
applying zero shear boundary conditions, one gets
t1 t3 txy2 0 at y b=2,

(12)

t2 txy3 0 at x a=2

(13)

ARTICLE IN PRESS
510

P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

to yield An and Bm. It should be noted that this step


involves expansion of t1, t2 and t3 into a Fourier sine
series, and thus Eqs. (12) and (13) yield as many algebraic
equations as are required for the determination of the
unknown coefcients An and Bm.
2.5. Results
Results are obtained for all the nine load cases as shown
in Fig. 1. The central load of Case 1 and the end loads of
Case 9 are taken to be localised over a small patch and not
as point loads because the former is more realistic in
practice. The value of b/200 chosen for the patch size is on
the basis of a convergence studystudies with a smaller
patch size yield more or less the same nal critical load. It
should also be noted here that the load cases 69 have to be
looked upon as the superposition of a uniform load minus
centrally applied loads of cases 14, so that the series
representation of Eq. (1) is applicable. For each load case,
a sufcient number of terms are considered in all the
Fourier series (see Eqs. (1), (4), (7) and (10)) to ensure fourdigit convergence of the nal in-plane stress results.
The superposition of three stress functions is required so
as to enforce the zero shear stress conditions at the edges of
the plate. The convergence of the nal superposed value
of this stress towards zero with increase in the number of
terms considered in f2 and f3 (i.e. nmax and mmax, which
are taken to be equal for the sake of convenience) is
illustrated for one load case (Case 3) in Fig. 2. This kind of
study is carried out for each load case so as to ensure that
the zero shear boundary condition is always numerically
satised, so that the nal numerical results for the in-plane
stress eld can be taken as rigorously correct.
By virtue of St. Venants principle, the applied edge load
is expected to diffuse into a uniform state of stress in the
interior of the plate; however this diffusion is complete
only for a long plate. The interior sx distribution for a
square plate for different edge loads is shown in Fig. 3; for
each case, the total load is kept at the same value P. This
gure shows that the interior stresses depend very much on
the actual edge load distribution and cannot be assumed to
be uniform or nearly so. The effect of such an internal
stress distribution on the buckling load cannot be taken to
be small, unless the plate is very long.
Another characteristic feature of non-uniform uniaxial
loading is the presence of non-zero sy and txy, in addition
to sx. Such a biaxial stress eld is as shown in Fig. 4 for the
case of sinusoidal loading (Case 4). Quite clearly, the
maximum values of sy and txy are not negligible as
compared to that of sx, and this is true for all the other
load cases also except that of uniform loading.
3. Use of extended Kantorovich method (EKM)
As against the innite series solutions for the in-plane
stress eld obtained above, closed-form counterparts with
just a few terms would prove more convenient for

Fig. 2. Shear stress distribution for Case 3 along: (a) y b/2, (b) x a/2.

subsequent buckling analysis. Towards this end, one can


employ Kantorovich method [16] in which an n-term trial
function for the two-dimensional elasticity problem can be
taken as
f f0 x; y

n
X

X p xY p y,

(14)

p1

where f0 x; y has specied value of applied traction on the


boundary and Xp(x) and Yp(y) are such that they
correspond to zero traction conditions.
Here, Xps are specied a priori and Yps are to be
determined by solving a set of n ordinary differential
equations derived from a variational principle. In this case,
the rst variation of the simplied complementary strain
energy U* expressed in terms of Airy stress function has to
be zero, where U* is given by [17]
Z
1
U
r2 f2 dA; where r2 f fxx fyy .
(15)
2E A
The resulting solution would obviously agree more
closely with the exact solution with respect to the
variation in the y-direction than with respect to that in
the x-direction. To get over this difculty, Kerr [18]
proposed an extended Kantorovich method, in which,
after a Kantorovich solution is obtained as described

ARTICLE IN PRESS
P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

511

Fig. 3. Normal stress diffusion for different edge loads.

above, the nal Yps are used to re-determine Xps through


another Kantorovich solution process; such iterations can
be repeated until the result converges to a desired degree.
Successful applications of the extended Kantorovich
method have been illustrated earlier with reference to
static exure [19] and stability [20] of a uniformly
compressed clamped plate using a single-term approach,
and with reference to buckling of uniformly compressed
plates with different boundary conditions using a multiterm approach [21].
In the present study, one and two-term (i.e. n 1 or 2 in
Eq. (14)) extended Kantorovich solutions are examined for
their accuracy, with the iterations repeatedly carried out
within each solution till the numerical results converge.
3.1. Methodology
With reference to any of the nine load cases considered
earlier (Fig. 1), it is advantageous to carry out the analysis
for the rst quadrant in view of the symmetry of the
problem. Further, the load cases divide themselves
naturally into two categories, one comprising of the
smoothly varying loads (Cases 37) and the other,
discretely applied loads (Cases 1,2,8,9). For the former,
f0 is a continuous function, while for the latter, it becomes
a function with discontinuities at the points where the
patch load terminates; thus, for the latter cases, one needs
to solve the ordinary differential equations separately in
different sub-regions of the plate, as made clear below. For
the simplest case of uniform load, there is no need for any
Kantorovich iterations, because f0 alone represents the

complete solution corresponding to a uniform internal


state of stress.
The derivation of the ordinary differential equations and
the corresponding boundary conditions is detailed below in
non-dimensional form, with the Airy stress function
given by
!
X
X p xY p Z ,
(16)
f s0 b2 f s0 b2 f0
p1;2;3...

where x x/b and Z y/b, and s0 is a reference stress


value (see Fig. 1),
The minimum complementary energy criterion reduces
to
dI 0,
with
Z
I

l=2
0

(17)
Z

1=2

r2 f 2 dx dZ,

(18)

where l a/b.
Assuming a two-term solution, the EulerLagrange
equations of this functional, for the cases of smoothly
varying loads, turn out to be:
(a) When Xps are assumed:
K x fd M x fd;ZZ 2Ax Y 1 2Dx Y 001 2F x Y 0000
1
Bx Y 2 E x G x Y 002 I x Y 0000
2 0,
Lx fd N x fd;ZZ Bx Y 1 E x Gx Y 001 I x Y 0000
1
00

2C x Y 2 2H x Y 2 2J x Y 0000
2 0,

19

ARTICLE IN PRESS
512

P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

and
Y 01 Y 02 ; 0
000
Dx Y 01 E x Y 02 2F x Y 000
1 I x Y 2 M x fd;Z 0,
000
GY 01 HY 01 IY 000
1 2JY 2 Nfd;Z 0 at Z 0,

20b

where fd fZZ, and the constants Ax, etc. are integrals


dependent on X1 and X2.
(b) When Yps are assumed:
00
K y 2Ay X 1 2Dy X 001 2F y X 0000
1 By X 2 E y G y X 2

I y X 0000
2 0,
Ly By X 1 E y G y X 001 I y X 0000
1 2C y X 2
2H y X 002 2J y X 0000
2 0,

21

with
X 1 X 2 0 and X 01 X 02 0 at x 1=2

(22a)

and
X 01 X 02 0,
000
Dy X 01 E y X 02 2F y X 000
1 I y X 2 0,
000
Gy X 01 H y X 02 I y X 000
1 2J y X 2 0 at x 0,

22b

where Ay, etc. are integrals dependent on Y1 and Y2.


For the cases of discrete loading, the quarter of the
plate is further subdivided into two sub-regions as shown in
Fig. 5 for the case of a central patch, and the approximation for f* is taken separately in the two sub-regions as
f f0 X 1 xY 11 Z X 2 xY 21 Z in the loaded subregion
X 1 xY 12 Z X 2 xY 22 Z in the unloaded subregion:
23
When Xps are assumed, the variational procedure yields a
set of four differential equations (similar to Eq. (19));
further, in addition to the boundary conditions at Z 0
and 1/2 as in Eq. (20a), one gets continuity conditions at
the interface between the two regions. The derivation of
these equations is straightforward [22] and hence they are
not presented in detail here.
As regards the other portion of the Kantorovich
iteration wherein Yps are assumed, the discrete approximation as in Eq. (23) does not alter the character of the
Euler-Lagrange equations (Eqs. (21)(22)), except that the

Fig. 4. The stress eld for sinusoidal loading.

with
Y 1 Y 2 0 and Y 01 Y 02 0 at Z 1=2

(20a)

Fig. 5. Sub-regions for EKM analysis (Case 2).

ARTICLE IN PRESS
P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

513

Table 1
Comparison of exact and approximate plane stress results
Location Stress

Case 3: Triangular load

Case 2: Central patch load

1-term EKM (4 iterations) 2-term EKM (1 iteration) Superposition method 2-term EKM (4 iterations) Superposition method
sx/s0
sy/s0
(a/4,b/4) sx/s0
sy/s0
txy/s0
(a/2,0)
sx/s0
sy/s0
(0,0)

0.6601
0.1422
0.5143
0.0319
0.0871
1.0000
0.4441

0.6939
0.1366
0.5208
0.0269
0.0867
1.0000
0.5172

0.6900
0.1368
0.5214
0.0261
0.0869
1.0000
0.5679

integrals Ay, etc. occurring therein would now be sums of


integrals over the two sub-regions with the appro
priate Yps.
3.2. Results
The f0s assumed for the different cases are as given in
Appendix B. For every case, a two-term extended
Kantorovich solution is obtained by starting with
X 1 x2  1=42 ; X 2 x2 x2  1=42 .

(24)

The iterations are carried out till the numerical results


converge. The corresponding one-term solutions (i.e.
X2 Y2 0) are also obtained. All the symbolic and
numerical manipulations are carried out using the computer algebra package MATHEMATICA. The plane stress
results based on these solutions are compared with those of
the earlier exact superposition approach in Table 1 for
triangular load (Case 3) and central patch load (Case 2).
From this table, one can see that the accuracy of the
converged one-term EKM results is not satisfactory, which
indicates that, in general, a single variables-separable term
of the type X1Y1, besides f0, is not adequate to capture the
correct stress eld. The two-term results are signicantly
improved as compared to the one-term results and require
but a few iterations to converge; however, they are of
acceptable accuracy only for the triangular load and not
for the patch load. Similar observations are also true of the
results of the other cases which are not presented here in
detail; from such results, one can conclude that the discrete
load cases (Cases 1,2,8,9) require three or more terms in the
EKM approximation. However, such higher term solutions
would turn out to be quite complicated because of a larger
number of coupled differential equations, and are hence
not attempted here.
4. Stability analysis and results
The governing differential equation for thin plate
buckling is
r4 w

h
q2 w
q2 w
q2 w
sx 2 2txy
sy 2 0.
D
qx
qxqy
qy

(25)

0.1747
0.0534
0.0679
0.0073
0.0329
1.0000
0.9844

0.1908
0.0620
0.0741
0.0093
0.0385
1.0000
0.9396

Owing to the complexity of this equation when each of


the in-plane stresses is a function of x and y, an exact
analytical solution is not possible. Therefore, the buckling
solution is obtained by using the Galerkin method for the
case of simply supported rectangular plates with the
admissible functions
X
X
W mn cosmpx=a cosnpy=b, (26)
wx; y
m1;3;5::: n1;3;5:::

Or
wx; y

W mn sin 2mpx=a cos npy=b;

m1;2;3::: n1;3;5:::

(27)
which are symmetric and antisymmetric, respectively, with
respect to the y-axis. Their choice is based on the expected
buckled shape which depends on the aspect ratio of the
plate; however, as will be shown later, the effect of aspect
ratio depends on the actual distribution of the edge load
and hence trials with both Eq. (26) and (27) are often
required to identify the lowest buckling load. It should be
pointed out that the shape of the buckled conguration in
the direction parallel to the loaded edge is not of a simple
harmonic half-wave except for uniform compressive load,
and hence trial functions with different harmonic wave
numbers in y are required. However, the lowest buckling
load is expected to correspond to a symmetric buckled
shape with no nodal lines parallel to the x-axis and hence
only odd n-terms occur in the above equations. It is also
appropriate to point out that these trial functions form a
complete set and thus convergence to the exact solution
with increase in number of terms is guaranteed.
The convergent buckling loads for different edge load
cases are presented in Table 2; the results correspond to the
total integrated value of the edge load at the onset of
buckling. Focussing attention on the square plate rst, one
can see that the actual distribution of the uniaxial load
does have a signicant inuence on its critical value; this
value is minimum for the case of a central localised load
and maximum for the case of end localised loads and
ranges from 0.65 to 1.64 times that for uniform load. In
fact, the load cases of Fig. 1 are numbered in an order such
that the compressive action of the load moves gradually

ARTICLE IN PRESS
P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

514

Table 2
Total buckling loads (values for K in Pcr Kp2D/b)
Load cases

1.
2.
3.
4.
5.
6.
7.
8.
9.

Central localised load (c b/200)


Central patch load (c b/10)
Triangular load
Sinusoidal load
Uniform load
Reverse triangular load
Reverse sinusoidal load
End patch loads (c/2 b/20)
End localised loads (c/2 b/400)

a/b 1

a/b 2

Super-position method

2-term EKM

NISA

Super-position method

2.604
2.621
3.339
3.450
4.000
4.934
5.438
6.499
6.541

2.232
2.612
3.343
3.450
4.000
4.936
5.437
5.566
6.910

2.584
2.607
3.324
3.442
3.986
4.826
5.241
6.427
6.409

2.947
2.962
3.576
3.656
4.000
4.350
4.456
4.520
4.580

Fig. 6. The buckled shape along: (a) y 0, (b) x 0.

from the centre of the edge to the corners, and Table 2


shows that the corresponding total buckling load increases
in the same order. The increase is more dramatic as one
moves down from load case 5 to 9 than from 1 to 5.
In all these cases, the lowest buckling load corresponds
to a shape with no nodal lines parallel to x or y. The mode
shapes for load cases 1 and 9 are compared with those for
uniform load (Case 5) in Fig. 6; it is clear that the variation
along the two central lines is not signicantly different
from the sinusoidal shape.
Regarding the use of EKM, one nds that the buckling
loads based on the two-term closed-form plane stress
solutions are quite accurate for all the smoothly varying

load cases (37) though the in-plane stresses themselves


display slightly greater error with respect to the exact
counterparts (see Table 2). This suggests that point-wise
agreement of the plane stress results is not essential to
obtain accurate buckling loads because they contribute to
the buckling equation only in an integrated form. However, for the discrete load cases (1,2,8,9), serious discrepancies occur between the exact solution and the two-term
EKM based solution both for plane stress results and the
buckling loads; very clearly, these discrete load cases
correspond to stress-elds with steep gradients and care
must be taken to carry out the plane stress analysis
reasonably rigorously for such cases.
Table 2 also includes buckling results for the square plate
as generated by using the commercial nite element
software NISA; these results are based on the use of the
eight-noded quadrilateral element with a mesh density
sufcient for convergence. The only idea behind the
inclusion of these results in this paper is to give an estimate
of the errors involved in a straightforward use of a finite
element package. It is clear from Table 2 that the agreement
with the exact solution is satisfactory, but the results do not
coincide in spite of careful convergence studies. Further, it
is necessary to note that the present nite element
predictions for the buckling load are not overestimates as
commonly expected, presumably due to the use of an
incompatible plate bending element or due to errors in
replacing a continuous load distribution by approximately
equivalent nodal forces.
For a longer plate (a/b 2), because of greater diffusion
towards a uniform state of stress in the interior of the plate,
the difference between the various load cases is expected to
be smaller than for the square plate; Table 2 indicates that
this is indeed so, with the total buckling load now ranging
from 0.74 to 1.15 times that for uniform load. However,
greater differences occur here for changes within the group
of 15, than within the group of 59. For all these cases, the
lowest buckling load corresponds to one nodal line parallel
to the y-axis (as is well known for uniform load) except
for load case 9, which corresponds to a shape with no
nodal lines.

ARTICLE IN PRESS
P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

515

estimates for the buckling load when the edge load is


distributed in a smooth fashion, but a larger number of
terms is required for patch and localised loads.
Appendix A
The coefcients appearing in the building blocks of the
superposition approach are
C 1r

sr b2 arp cosh arp=2b 2b sinh arp=2b


,
r2 p2 ap b sinh arp=b

C 4r 
Fig. 7. Buckling coefcient K Pcr b=p2 D versus aspect ratio.

The variation in the number of nodal lines parallel to the


y-axis with respect to a/b is shown in Fig. 7. It is well
known that a plate with integral value of a/b buckles into
square panels when subjected to uniform compression with
the buckling load invariant irrespective of the actual value
of a/b. However, this is not true for other load distributions. From Fig. 7, it is clear that as a/b increases the
buckling load approaches that for uniform compression as
expected. The qualitative nature of the buckling load
variation with a/b for any particular edge load distribution
is the same as that for uniform load, but there is a shift
towards left or right of the transition points which
correspond to two equally probable buckled congurations
for the same buckling load. As can be easily imagined, a
plate with a predominantly central load buckles into
multiple half-wave-like congurations for smaller a/b
ratios than one with a predominantly off-centre load. It
is quite interesting to see that this shift can be very
signicantin fact, for the plate with end loads, even for
a/b 2, the lowest buckling load corresponds to one halfwave-like conguration with no nodal lines.
5. Conclusions
Based on rigorous plane elasticity solutions for a set of
uniaxial edge load distributions, the corresponding total
buckling loads have been determined here using Galerkins
approach with a set of trial functions that ensure
convergence. The results show that the actual distribution
of the edge load has a signicant inuence on the critical
value of the total load. It is expected that the convergent
results tabulated herein will serve as a standard for judging
the accuracy of various approximate methods in future.
It has also been shown that the dependence of the
buckling load on the aspect ratio of the plate is
qualitatively similar for all the edge loads considered here
though there is a shift of the transition points. With respect
to the use of extended Kantorovich method for approximate plane stress analysis, it has been shown that a
solution with two variables-separable terms can yield good

2sr b2 sinh arp=2b


,
rparp b sinharp=b

a2 bAn sinh npb=2a


,
bn2 p2 anp sinh bnp=a
2a2 An cosh npb=2a
,
2 2
bn p anp sinh bnp=a

C 1n 
C 4n

C 1m 
C 4m

ab2 Bm sinh mpa=2b


,
am2 p2 bmp sinh amp=b

2b2 Bm cosh mpa=2b


.
am2 p2 bmp sinh amp=b

Appendix B
The rst term of the extended Kantorovich solution is
given by
f cospZ
for sinusoidal load;
p2 
 02
Z
cospZ
for reverse sinusoidal load;
2  p2
2

3
Z
Z
for triangular load;
2  3
Z3
3
Z2
2

for reverse triangular load;


(within the loaded sub-region) for patch loads.

References
[1] Timoshenko SP, Gere JM. Theory of elastic stability. New York:
McGraw-Hill; 1961.
[2] Bulson PS. The stability of at plates. London: Chatto and Windus;
1970.
[3] Kang JH, Leissa AW. Exact solutions for the buckling of rectangular
plates having linearly varying in-plane loading on two opposite
simply supported edges. Int J Solids Struct 2005;42:422038.
[4] Leggett DMA. The effect of two isolated forces. Proc Cambridge
Philos Soc 1937;33:3259.
[5] Yamaki N. Buckling of a rectangular plate under locally distributed
forces applied on the two opposite edges, Report of the Institute of
High Speed Mechanics, vol. 34, Tohoku University, Japan, 1954;
p. 5571.
[6] Timoshenko SP. Stabilitat einer rechteckigen platte, die durch
einzelne krafte gedrckt wird. Z Math Phys 1910;58:35760.
[7] Hopkins HG. Elastic stability of innite strips. Proc Cambridge
Philos Soc 1949;45:58794.
[8] Alfutov NA, Balabukh LI. Energy criterion of the stability of elastic
bodies which does not require the determination of the initial stressstrain state. Prikl Mat Mekh 1968;32:7037.

ARTICLE IN PRESS
516

P. Jana, K. Bhaskar / Thin-Walled Structures 44 (2006) 507516

[9] Khan MZ, Walker AC. Buckling of plates subjected to localized edge
loading. Struct Engi 1972;50(6):22532.
[10] Spencer HH, Surjanhata H. Plate buckling under partial edge
loading, In: Developments in mechanics, Proceedings of the 19th
mid-western mechanics conference, vol., 13, 1985; p. 834.
[11] Bert CW, Devarakonda KK. Buckling of rectangular plates subjected
to nonlinearly distributed in-plane loading. Int J Solids Struct
2003;40:4097106.
[12] Van der Neut A. Buckling caused by thermal stresses. In: High
temperature effects in aircrafts structure. AGARD report, vol. 28,
1958, p. 21547.
[13] Benoy MB. An energy solution for the buckling of rectangular plates
under non-uniform in-plane loading. Aeronaut J 1969;73:9747.
[14] Little RW. Elasticity. New Jersey: Prentice-Hall; 1973.
[15] Gorman DJ, Singhal RK. A superposition RayleighRitz method for
free vibration analysis of non-uniform tensioned members. J Sound
Vib 1993;162:489501.

[16] Kantorovich LV, Krylov VI. Approximate methods of higher


analysis. New York: Interscience Publishers; 1966.
[17] Richards TH. Energy methods in stress analysis with an introduction
to nite element techniques. New York: Halsted Press; 1977.
[18] Kerr AD. An extension of the Kantorovich method. Quart J Appl
Math 1968;6:21929.
[19] Kerr AD, Alexander H. An application of the extended Kantorovich
method to the stress analysis a clamped rectangular plate. Acta Mech
1968;6:18096.
[20] Kerr AD. An extended Kantorovich method for the solution of
eigenvalue problems. Int J Solids Struct 1969;5:55972.
[21] Yuan S, Jin Y. Computation of elastic buckling loads of rectangular
plates using the extended Kantorovich method. Comput Struct
1998;66(6):8617.
[22] Jana P. Analytical solutions for buckling of rectangular plates due to
non-linearly distributed uniaxial and biaxial loads. MS thesis, IIT,
Madras, 2005.

You might also like