Int. J. Mech. Sci. Vol. 30, No. 10, pp. 719 732, 1988 002(P7403/88 $3.00+.

00
Printed in Great Britain. © 1988PergamonPress plc

THE ELASTIC WRINKLING OF RECTANGULAR SHEETS

R. H. SEGEDIN,I. F. COLLINS and C~ M. SEGVOIN

Department of Theoretical and Applied Mechanics, University of Auckland, Private Bag, Auckland,
New Zealand

(Received 5 August 1987; and in revised form 23 April 1988)

Abstract--Two models for predicting the elastic behaviour of rectangular sheets pulled in tension are
investigated. In one, a parabolic loading distribution is assumed; in the other, the normal
displacement is taken to be uniform on the loaded edges. Collocation techniques are used to find the
wrinkling modes and the associated eigenvalues for the tensile loads. In the second problem the in-
plane stress distribution is found using finite difference procedures. The stress singularities in the
corners are allowed for. Buckling under compression is also studied for both models.

NOTATION

a
length of free edge y = 0
b
length of loaded edge (width) x = 0
c~ a/b (aspect ratio)
u, v, w displacements in x, y, z directions respectively
t thickness of plate
x/a
17 y/b
axx, ax r, a ,
in-plane stresses
D=Et3/[12(1-v2)] flexural rigidity
m mode number along length of plate
n mode number across width of plate
o o load stress factor
A=oob2t/D dimensionless load factor
C,,, Fourier coefficients of the displacement surface
v Poisson's ratio
v/( l + v)
E Young's modulus
y [12ao(1-v2)/E] 1/2
2 singularity index
K singularity constant
a,, average value of axx along x = 0
Y uniaxial yield stress

INTRODUCTION

One of the major factors affecting the forming limits of sheet metal is the appearance of
wrinkles or buckles. An understanding of the factors governing the formation and growth of
these wrinkles is of prime importance in many press forming operations, notably in the
manufacture of automobile panels. Frequently these wrinkles appear during elastic recovery
or springback. In order to understand the mechanics of this process, Yoshida and his co-
workers developed a test [1] in which a square specimen is stretched diagonally to produce
one or more wrinkles. Yoshida et al. [2] and Aoki et al. 1-3] developed an approximate
theoretical model in which the onset of wrinkling is attributed to elastic buckling arising
from the compressive lateral stresses induced in the nonuniform stress fields. The growth of
these wrinkles is then governed by the plastic behaviour of the material through parameters
such as its r-value (anisotropy) and n-value (work-hardening). In the present paper we are
concerned only with a rectangular sheet pulled in tension with the lateral edges free from in-
plane stresses (Fig. 1). Yoshida et al. [2] assumed the sheet to be subject to a parabolic tensile
traction distribution on the loaded edges. The associated stress distribution has lateral
compressive stresses in the central region of the plate, so that under sufficiently large loads
719
720 R . H . SEGEDIN et al.

w,T__> u
( a
• ItI
"0 . . . . . . . ->x
FIG. 1. Layout of the plate, showing its dimensions and direction of displacements.

the plate can be expected to wrinkle. The analysis of this phenomenon given in [2] is further
approximated, however, since the influence of the in-plane shear stresses was ignored. An
exact analysis of the onset of buckling under this stress field is presented in this paper,
together with a parallel study of the much more realistic model in which the two loaded
edges are clamped and pulled apart uniformly. The stress distributions corresponding to the
two boundary loading conditions are found to be very different and in consequence the
predictions of the onset of wrinkling are also markedly different. It is to be emphasised that
the analysis presented here is classical in the sense that it is based on the linearized buckling
equations. No attempt is made to look at the post-buckling or growth behaviour which
would require the use of either the nonlinear von Kfirm~in equations if the deformation
remained elastic or of an elastic-plastic analysis.
Although the original motivation for this investigation stemmed from the need for an
analysis of wrinkling of sheets under tension, the resulting algorithm predicts the com-
pressive instabilities as well. It is of interest to compare these compressive buckling modes
with those induced by the application of uniform boundary traction distributions as given
by classical elastic plate buckling theory. (N.B. We are using the term wrinkling here in the
sense of the onset of wave-like instabilities in situations where the dominant applied stresses
are tensile.)

PLATE BUCKLING CALCULATIONS

The notation used to describe the plate geometry and deformation is shown in Fig. 1. The
origin is located in the left front corner, while the x and y axes are measured along the length
and width of the plate respectively. The aspect ratio ~ is taken to be a/b, the ratio of the
length ~f the free edge to that of the loaded edge. The lateral displacement of the plate w(x, y)
satisfies the standard plate buckling equation I-4]:
- // 02W ~2 W ~2 W "~
DVZw=t~rrX~x2 + 2rrxyo~y+arr~y2 ). (1)

This equation can be recast in a more useful dimensionless form if all the lengths are
nondimensionalized by the plate width b, w by the thickness t and all stresses by some
suitable load stress factor Oo:

V4w=A[ffxx t~2w l~2w 02wl, (2)

7
where

A=
aob2t
D =
(~)12(l_vZ)(b/t)2 (3)

is the dimensionless load factor.

For the purpose of calculating the critical buckling loads, all four sides of the plate are
taken to be simply supported, so that solution of equation (1) or (2) can be sought in the form
of a double Fourier sine series:

w(x, y) = ~ C,,,,sin (mrc~) sin (nru/), (4)

m,'n= 1

where ~ = x/a and t/= y/b are dimensionless lengths in the x and y directions respectively.
 The elastic wrinkling of rectangular sheets 721

This means that the b o u n d a r y conditions pertaining to simple support are automatically
satisfied.
Substituting equation (4) into (2) yields the system of equations

[Amn(x, y)-AB..n(x, y)] C m . = 0 , (5)

m,n=l
where
(m2 n2 ~2
Am~(x,y) = n4 k-dy + ~-i] sin (mnO sin (nnr/)
\ /
and
m 2
Bm~(x,y ) = - axx(x, y) n 2 ~ - sin (mn~)sin (nnr/)

mn
+ 2a,r(x, y) n 2 ~ - cos (mn~)cos (nnq)
n 2
- ayr(x, y)/r 2 ~- sin (mn¢) sin (nnq). (6)

In order to solve for the coefficients Cm., the double series is truncated at m = M and n = N
and the left hand side of(5) evaluated at each o f p x q interior points lying on the regular grid

(xp, yq)=( pa qb )
M~l' N+I " (7)

This gives a set of MN h o m o g e n e o u s linear equations for these MN coefficients. These were
solved for the eigenvalue A (the load factor) using the N A G library routine F02BJF based on
the Q Z algorithm [5].
The buckling is either symmetric or antisymmetric in both the x and y directions, so that
just one quarter of the plate needs to be considered. Also, this means that each of the m and n
values are either all odd or all even. T o investigate all possible buckling modes, all four
c o m b i n a t i o n s of m, n o d d and even are needed (e.g. m odd, n even etc.). It was found in
practice that values of M, N of a r o u n d 17 or 18 were usually sufficient to give the basic
eigenvalues correct to 3 significant figures, although for the highly wrinkled modes, values
up to 35 were required.

PARABOLIC TRACTION D I S T R I B U T I O N

Stress distribution
The in-plane stress distribution used in Ref. [2] is, in the notation of the present paper:
(axx/ao) = 4t/(1 - t / ) - fl~2(1 - ~)2~2[1 - 3(2r/- 1) 2]
(ayr/ao) = -/311 - 3 ( 2 ~ - 1)2] ~/2(1 - r / ) 2 (8)

(axr/ao) = - 4fl~(1 -- ~)~(2¢ -- 1)(1 -- r/)r/(2r/- 1),

where
49~ 2
fl = 7 ~ 4 + 4 ~ 2 + 7"

O n the loaded edges x = 0 and a, the n o r m a l stress distribution is symmetric and parabolic:
axx = 4aot/(1 -- r/), (9)
with a peak value of ao, while axy = 0. This stress field is self-equilibrated, and has been used
by T i m o s h e n k o and G o o d i e r [6] as an a p p r o x i m a t i o n to an elastic stress field. There is a
sizeable area in the centre of the plate, between

x = a(3 ___x//3)/6, (10)

i.e. of width 0.577a, in which a , is compressive and which will hence tend to induce
wrinkling.
MS 30 : l O - B
722 R.H. SEGEDINet al.

Numerical results
Values of the dimensionless loading parameter A have been computed for various aspect
ratios c~= a/b in the range 1/2 to 4. Outside this range, the A values are much higher.
Although our main concern is with studying the wrinkling loads under tension (A > 0), the
program also computed the corresponding compressive buckling loads. For each critical A,
the coefficients C,,. are calculated--these values define the shape of the associated mode of
buckling (or wrinkling). While for simple buckling under uniform compressive stress fields,
only one Fourier coefficient C,,, is nonzero, for the nonuniform stress fields considered
herein this is not so, each buckling mode now containing many Fourier components.
Inspection of the numerical results reveals that for m odd, the lowest tensile eigenvalues
have modes which are dominated by components for which m = 1, i.e. for which there is just
one wave along the length of the plate. There are sometimes, however, minor components for
m = 3, but their amplitudes are never more than 20% of the amplitude of the dominant
component. For m = 2, the lowest eigenvalues are about ten times as high as those with m = 1
and so will not be observed in practice.
Table 1 shows the lowest tensile eigenvalue A for selected aspect ratios ~, for both even
and odd numbers of lateral wrinkles (i.e. for n odd or even) and Fig. 2 illustrates this
graphically. The table also shows the dominant modes with the corresponding wave
numbers n. It can be seen that the value of the critical load decreases rapidly as the aspect
ratio increases from 0.5 to a minimum around 1.5 and then increases again (see also Fig. 2).

TABLE 1. PARABOLIC LOADING DISTRIBUTION. TABLE OF LOAD FACTORS AND DOMINANT WAVE NUMBERS OF
WRINKLING MODES

n odd n even
A/10 3 nI n2 n3 n4 A/10 3 n I n2 n3 n4

0.50 52.20 11 13 (99%)15 (60%) 9 (51%) 52.20 12 10 (76%)10 (74%)16 (37%)

0.55 33.41 11 9 (96%)13 (57%) 7 (38%) 33.41 10 12 (76%) 8 (66%)14 (35%)
0.60 22.75 9 7 (74%)11 (64%)13 (25%) 22.75 8 10 (90%)12 (44%) 6 (42%)
0.75 9.63 7 5 (96%) 9 (44%) 9.64 6 8 (64%) 4 (44%)
0.85 6.56 5 7 (59%) 3 (33%) 6.57 6 4 (91%) 8 (39%)
1.00 4.54 5 3 (80%) 7 (35%) 4.50 4 6 (54%)
1.25 3.42 3 5 (58%) 3.46 4 2 (51%) 6 (33%)
1.50 3.25 3 5 (64%) 3.38 4 2 (82%) 6 (26%)
1.75 3.47 3 5 (33%) 3.65 2 4 (93%)
2.00 3.91 3 5 (30%) 1 (25%) 4.12 2 4 (79%)
3.00 6.90 3 1 (33%) 5 (26%) 7.23 2 4 (61%)
4.00 11.38 3 1 (34%) 11.87 2 4 (56%)

25000
A
20000

15000'

J
10000'

5000

0 i ! i

1 2 3
(~=a/b

FIG. 2. Variation of wrinkling loads with aspect ratio for plate under tension subjected to parabolic
traction distribution, symmetric wrinkling (n odd); . . . . antisymmetric wrinkling (n even).
 The elastic wrinkling of rectangular sheets 723

2.0.

I.O-

0.0"

4.5"

.1.13 -

-1.5"

.2.0"

,t o •
Pu['[Q0d . ~ ~ . o.~o.~
FIG. 3. Antisymmetric wrinkling mode under tension for plate with aspect ratio a/b=0.55 under
parabolic traction stress loading.

The dominant lateral wave number n also decreases as ct increases and settles down to 3
(odd) or 2 (even) for long plates with ct > 1.5. The table also gives the ratio, in the form of a
percentage, of the amplitudes of the second and, if significant (i.e. /> 25%), third and fourth
modes to that of the first. A comparison of the lowest even and odd modes shows that the
latter gives a slightly lower result for ~ t> 1.25, but that there is no significant difference for
smaller values of ~. A typical example of a wrinkling mode is shown in Fig. 3, where ~t= 0.55.
There are ten waves across the plate, but they are of course of different heights because of the
influence of other sub-dominant components.
Buckling limit diagram
For a given plate material and thickness it is possible to show in an (a, b) diagram the
range of plate sizes and loads for which wrinkling can be expected to occur. This wrinkling
region can be determined for an arbitrary plate from the 'universal curve' shown in Fig. 4(a),
in which ya/t is plotted against yb/t, where
y2 = 12tro( 1 _ v2)/E = trot3/D. (11)
This curve can be obtained by replotting the values of A in Fig. 2, where
A =~2(b2/t2) is plotted against ~ = a/b.
A particular example is given in Fig. 4(b), together with the result of the approximate
computation given in [23 for t = 1 mm, in which the effect of shear stress was ignored.
Inclusion of the shear stress is seen to increase the size of the wrinkling zone.
Buckling under compression
The negative eigenvalues give the buckling loads under a parabolic compressive boundary
traction distribution. The results are similar to those obtained under uniform compression.
The lowest dominant buckling modes all have n = 1, so that only one wave spans the width of
724 R.H. SEOEDIN et al.

300

7(a/t)
200

100
~ wrinkling
zone

I I

0 100 200 300

7(b/t)
FIG. 4(a). Universal nondimensionalized buckling limit diagram for the parabolic traction
distribution.

4000

3000 i/

b 2000 / wrinkling
[i /
zone

1000

I I I

0 1000 2000 3000 4000

a

FIG. 4(b). Particular example of buckling limit diagram. The inclusion of shear stress terms increases
the size of the buckling zone for the parabolic distribution, analysis neglecting axy terms as in
I-2]; exact analysis of the present paper.

the plate. Plots of the eigenvalues for buckling patterns dominated by modes with m = l, 2, 3
against aspect ratio (dashed lines) are shown in Fig. 5. Also plotted on the graph are the first
three curves for uniform compression (dotted lines). The buckling induced by the parabolic
stress distribution occurs at lower A values--approximately 20% lower. The turning point
does not occur at exactly a - - 1, 2, 3 as with the uniform loading case. Note that to compare
the two cases the same total load is applied to the end, so a 0 is one and a half times the
uniform stress value. The lowest compressive buckling load is A ~ 5 l, which occurs at c~just
less than unity. The corresponding minimum tensile eigenvalue is about one hundred times
as large.

UNIFORM DISPLACEMENT

The second stress distribution model considered here is obtained by stretching the plate
with constant normal displacement, inhibiting any lateral movement at the gripped edge; i.e.
u = const and v = 0. This better models the experiments carried out by Yoshida et al. [l, 2].
Firstly, the displacements u, v are calculated by finite difference methods at each point on the
plate using Navier's equations in the field and zero stress conditions on the free edges. The
 The elastic wrinkling of rectangular sheets 725

120

A
100

80
li~" ~.:.. \\ ':".. : I" I
i....." .""
\'.... ".. ../'"
\ °og
60 - . . . . . • .7. ...........

m=l m=2 m=3

40 i i !
0 1 2 3 4
(x=a/b

FIG. 5. Buckling under c o m p r e s s i o n ~ o m p a r i s o n of buckling loads for three stress distributions.

. . . . . . constant compressive stress; - - - parabolic traction stress; - - uniform displacement
including singularities.

corresponding stresses are then calculated and s u b s t i t u t e d i n t o the buckling equation (2).
The Navier equations for plane stress are
~2 u ~2 u ~2/)
2(I - 9 ) ~ + (I - 2 9 ) ~ + ~ = O, (12)

- ~2U 1 - ~2/) ~2 u
(1-2v)~x2+2 ( -v)~+~y=0, (13)

where ~ = v/(1 + v) and so has a range of (0,-~) (see Love I-7]).

As shown in Fig. 6, on the free edges y = 0 and y = b, both trxr and tryy vanish,
whilst on x=0, u=-U, v=0,
and on x=a, u= U , v=0. (14)
Again, because of symmetry, one need only consider a quarter of the plate, but the
b o u n d a r y conditicms along the lines of symmetry need to be invoked: on the line x = a/2 we
k n o w that u = 0, and, as v is an even function in the x direction here, it follows that dv/dx = O.
Similarly on y = b/2, du/ay = 0 and v = 0. The stress-free b o u n d a r y conditions must also be

¢r=O
G xy= 0
YY

aulay = 0
u=-U v=O u=U
v=O V=0
u=0
Y 3v/ax = o

x O'=0
(l ~ = 0
W

FIG. 6. Boundary conditions for uniform displacement distribution.

726 R.H. SEGEDIN et al.

written in terms of displacements, so that the conditions axr = 0 and ayr = 0 imply that

~YY
+ t 3 x Ov
=0~u and 2~+2(1-~)~=0 (15)

on y = 0 and y = b.
Finite difference approximation
Finite difference techniques were implemented to solve the Navier equations (12), (13). The
standard central difference approximations for first and second derivatives were used, except
on the boundary where the one sided formula
8F
c3~ "~ ( -- 3Fi'j + 4Fi + 1,j -- Fi + 2,j)/2h (16)

was used, where F~,j is the value o f F at the (i, j)th position on the grid. These approximations
introduced errors of order h 2.
The final form of the equations, suitably nondimensionalized, are summarized in Fig. 7.
The equations in Fig. 7 are solved iteratively for u, v at each mesh point. Initially v was set to
zero everywhere and u varied linearly from - 1 to 0 along the plate from x = 0 to x = a/2. The
speed of convergence was increased by the use of successive over-relaxation, with the
relaxation factors being determined by trial and error. The number of points needed in each
direction in the grid was generally between 20 and 60. The program was run until either a
maximum of about 500 iterations was reached, or the sum of errors between the ith and
(i+ 1)th iteration was less than 10-3. The load factor A for the parabolic distribution was
earlier written as in equation (3). For the present problem the displacements are propor-
tional to U, so (U/a) is a representative longitudinal strain. The buckling equation (1) is now
nondimensionalized by dividing w by t, x and y by b and the stresses by E, so that the
eigenvalue load factor parameter A is now modified to:
A ' = (U/a) 12(1 - v2)(b/t) 2. (3)'
(Note that since b was set equal to 1 in computations, A' is obtained from the eigenvalue in
the dimensionless equation (2) by dividing by the aspect ratio cc)

4u 1
U='~ i,j_l--'~ U i,j_2

v=O

U - - 1 - ' o (U "1- ) -t- 1-L~ (U +U )

3-4"o i+l,j U i-l,j 3-4~ i.j+l i,j-1
1
"t" (Vi.l,j+ 1 -- Vi_l,j+ 1 + Vi_l,j_ 1 -- Vi+1,i_1 ) u=O
8(3--4;)
U=--I
v=O V= V -- l V
i-l,j 3 i-2.j
V = 31 -- 42 ;; ( Vi.l,)_l" Vi_l,i ) + 31--4; ; (vl'j*l + vi'j-1 )
1
+ -8 ( 3- - 4 ~ ) (U i*l,j+l--Ui-l,j+l "1"Ui-l,j-1 --Ui÷l,j-1 )

~. u=O

U = ~•3 ( 4 U i,j+l - U ,j+2 + v i+l,j - v i-1,~ ) V-- 1 ( 4 V i . i + l --Vi,j+2 )

1 (4v -v ) + + ~ (4u~_,.j-u~_~,j)
v=3 ~,m+l ~,j+2 3(1-,7i (uj+,,j-u~_,,j) 3( 1-',~ )

FIG. 7. Final form of the finite difference equations for finding the displacement distribution in the
sheet, when subjected to uniform normal displacement.
 The elastic wrinkling of rectangular sheets 727

Form of the singularity

The displacements are finite in each corner of the plate, but the stresses become infinite, so
that extra singular terms have to be incorporated in the finite difference calculation to take
account of this. These extra terms will only be significant near the corner. The form of the
singularity is well known and has been given by Williams 1,8,9] and England [10], for
example. The resulting equations for u, v are as follows:
1- ( 3 - 2 - 4~) sin (1 - 2)0 ( - 5 + 2 + 4~) cos (1 - 2)0
2 ( 1 _ 2 ~ , ( : ) = K / + ( 1 - - 2 ) s i n ( 2 + 1)0 + (1 - 2) cos (2 + 1)0

] ( - 2-- 3 + 4~) cos (1 -- 2)0 ( - 2 - 1 + 4~) sin (1 - 2)0

L + ( 2 - 1)cos(1 +2)0 + (1 - 2) sin (2 + 1)0
(17)
The boundary condition u = v = 0 at 0 = re/2 gives a transcendental equation for the index 2
which, for v =0.33 (i.e. ~ = 0.248), takes the value 0.254; q~ is a constant involving 2 and ~, so
that in this case ~b=2.301 1,8-10,1. The remaining constant K is found from the field
equations when the singular terms are incorporated into the finite difference scheme using
the method outlined by Emery and Segedin r 11-]. This procedure is summarized in Appendix
A.

Numerical results
In order to compare the critical eigenvalues from the parabolic traction distribution with
those corresponding to the uniform displacement model, the modified eigenvalue A' is
converted to the equivalent value A by the relation

1.5 ~-r~ A'

The actual normal traction distribution on the loaded boundary corresponding to the
uniform displacement model is shown in Fig. 8. It is of course very different from the simple
parabolic distribution assumed in [2-1, and is such that axx becomes infinite at the two ends.
(A special quadrature formula used to work out the average stress (aav) enabled the effect of
the singularities to be taken into account.)
The variation of the compressive load parameter A with aspect ratio is shown in Fig. 5
(solid lines). These curves correspond to buckling modes with a single wave in the lateral
direction (n = 1). The three separate curves correspond to different numbers of longitudinal
waves (m = 1, 2 or 3). The values for n = 2 show a similar general trend, but are consistently

2.6
(~
XX

2.4

2.2

2.0

0.0 0.1 0.2 0.3 0.4 0.5

b

FIG. 8. Nondimensionalizednormal traction distribution ox~along x =0 for uniformdisplacement

model with a square plate (a=b = 1).
728 R.H. SEGEDIN et al.

higher and hence will not occur in practice. The values shown in Fig. 5 include the effect of
the corner singularities, and were computed using the procedure described in the Appendix.
If the singularity is ignored in the finite difference scheme, the calculated values of A' under
compressive loadings are not changed significantly except for the small deviation in the range
0.5 < ~ < 1. This is not found when the singularity is not included in the calculations, and
corresponds to the aspect ratios where there are sizeable bands of lateral tensile stress (see
discussion at the end of this section)• Typically, values calculated by ignoring the singularity
are between 0% and 5% higher than those obtained by the more exact procedure.
Also shown in Fig. 5 is a comparison of the compressive buckling loads for the parabolic
distribution, constant compressive stress and the uniform displacement, including the effect
of the singularities. Despite their different forms, the two nonuniform traction distributions
are both seen to make the'plate buckle at a lower critical load than when the applied stress is
uniform. The parabolic distribution curves rise more steeply than the other two as the load
increases.
Inclusion of the singularities has a profound effect with the tensile wrinkling modes. If the
singularities are ignored, no tension modes are found for any aspect ratio. But when the
singularities are allowed for, tensile wrinkling modes are found for a narrow range of aspect
ratios between ~ = 1/2 and ~ = 1. These are summarized in Table 2. The wrinkling loads
under tension obtained under uniform edge displacement with inclusion of the singular
terms are shown in Fig. 9. The solid line corresponds to the parabolic loading distribution

.TABLE 2. U N I F O R M DISPLACEMENT LOADING DISTRIBUTION. TABLE OF LOAD FACTORS [CONVERTED AS IN EQUATION

(18)] AND DOMINANT WAVE NUMBERS OF WRINKLING MODES

n odd n even
A/103 n1 n2 n3 A/103 nI n2 n3

0.571 45.12 6 8 (80%) 4 (64%)

0.600 33.60 6 4 (74%) 8 (65%)
0.667 24.87 5 3 (61%) 7 (56%) 21.36 4 6 (95%) 8 (40%)
0.750 15.20 5 3 (89%) 7 (32%) 14.37 4 6 (60%) 2 (40%)
0.800 12.33 3 5 (91%) 1 (30%) 11.90 4 2 (46%) 6 (26%)
0.900 8.40 3 5 (63%) 1 (33%) 8.51 4 2 (60%) 6 (26%)
0.938 13.61 3 5 (96%) 1 (39%) 13.21 4 6 (38%) 2 (50%)
0.952 63.52 7 5 (98%) 9,3 (37%) 60.09 6 8 (88%) 10 (60%)

70000
A
60000
J
50000 J
40000

30000

20000
\ J
10000 ~'.-.~'-" .~.,.,L,~,....,..¢. j ~"

0 I l I I

0•5 0.6 0.7 0.8 0.9 .0

(x=a/b

FIG. 9. Wrinkling loads under tension for the small range of aspect ratios where wrinkling occurs
under the uniform displacement model. The influence of the corner singularities is included. • .....
uniform displacement (n odd); - - - uniform displacement (n even); - - parabolic traction
distribution.
 The elastic wrinkling of rectangular sheets 729

(both even and odd modes are effectively the same for this range of aspect ratio). When
ct < 0.8, the odd modes for the uniform displacement distribution (dotted lines) start to have
significantly higher loads than the even modes (dashed lines), and when ~ < 0.65 there are no
positive real eigenvalues for n o d d - - t h e y are replaced by complex conjugate roots with
positive real parts. There is an extremely sharp increase in load between at = 0.9 and 0.95 for
both even and odd modes, and when ct > 0.95 no positive eigenvalues could be found at all,
and no complex eigenvalues with positive real parts (see also Table 2). An example of
symmetric wrinkling for a plate of aspect ratio ct=0.8 is shown in Fig. 10. There are three
main wrinkles across the plate and two small ones at the edges.
As with the parabolic traction distribution, there is a direct correspondence between
bands of compressive values and tension buckling modes. In the range of aspect ratios
0.6 < ~ < 0.9, there is a large band almost half of the plate wide, between x ~ a(1 _ 1/2)/2, in
which o-rr is compressive. (It should be noted that if the loading is compressive, this band is
tensile.) The band has narrowed in width to the centre third when 0t=0.5, where the only
eigenvalues with a positive real part are complex. When ~t= 1/3 there are only a few small
localized regions of compressive stress near the centre of the span, but which do not spread
across the width of the plate. When ct > 0.952 there are no compressive lateral stresses at all
and no eigenvalues, real or complex, with positive real parts. As was stated earlier, no tensile
instabilities are found when the singularities are not taken into account and there are no
regions of compressive lateral stress.

CONCLUSION

This investigation is concerned with evaluating the loads at which elastic wrinkles will
occur when a rectangular sheet is stretched. Two types of loading distribution have been
considered: (a) parabolic normal traction distributions; (b) uniform normal edge displace-
ments. Whilst we are mainly concerned with tensile instabilities, the corresponding
compressive buckling loads are also discussed. The principal conclusions of this study are as
follows.
(1) The inclusion of the shear stresses in Yoshida et al.'s [2] original parabolic traction
distribution model does not alter the form of the elastic tensile wrinkling zone as shown in
Fig. 4(b), but does increase the size of this zone, so that for a given load, wrinkling occurs for
a wider range of aspect ratios than predicted in [2-1.

0.5 -

0.0"

-0.5"

-t-0"

_!.5"

~,h,~ .Io Fr'-

FIG. 10. Symmetricwrinkling mode under tension for plate with aspect ratio ct=a/b=0.8 for the
uniform displacement model including the influence of the corner singularities.
730 R.H. SEGEDINet al.

(2) The load required for a plate to wrinkle under the parabolic tensile traction
distribution is about one hundred times greater than the corresponding load for buckling
under compression. The variation of the tensile loads with aspect ratio does not follow the
same 'intersecting' loops pattern as is observed under compression (see Fig. 2). As the aspect
ratio changes, the lowest wrinkling load corresponds to different buckling mode forms (see
Table 1).
(3) The minimum compressive buckling loads for the parabolic traction distribution are
about 20% lower than with constant compression for the same total load. Otherwise the
variations with the aspect ratio follow the same general form, although the parabolic
distribution curves rise more steeply.
(4) The boundary conditions encountered in the experimental tests are modelled more
accurately by assuming that the displacement on the loaded edges is uniform and normal to
the edge. The resulting displacement and stress distributions in the plate are found by using
finite difference techniques. These require the incorporation of terms to take account of the
stress singularities at the corners of the plate.
Wrinkling under tension for this uniform displacement model, including the effects of
singularities, occurs only in a narrow band of aspect r a t i o s ~ . 5 < ~ < 1, as shown in Fig. 10.
The wrinkling load increases very rapidly as the aspect ratio approaches 1. These loads are
always higher than those predicted using the parabolic traction distribution. The percentage
difference is least at ~ ~ 0.65, where the load is about 50% higher. There is good correlation
between the occurrence of bands of compressive lateral stress and tensile wrinkling. When
0 . 5 < ~ < 1, the distribution of compressive 0"rr stresses is similar to that of the parabolic
loading distribution.
In reality, the sheet will develop small plastic zones at the gripped corners. However, as is
found in linear fracture mechanics, one may anticipate that outside the plastic zone the
elastic stresses will not differ significantly from those predicted by elastic singularity theory.
Nevertheless the detailed wrinkling behaviour of the plate is clearly highly sensitive to the
boundary conditions on the loaded edges on the plate, although--since both the parabolic
traction and uniform displacement conditions predict tensile wrinkling loads of the order of
100-200 times larger than those obtained in compression--it suggests that perhaps the order
of magnitude of this ratio is not so dependent on the detailed loading conditions.
(5) The level'of applied stress at which wrinkling occurs is given, using equation (3), by the
formula

E 12(1-v 2)
Hence the sheet will wrinkle elastically before it deforms plastically provided that

\ r/E / "
With v~0.33, Y / E ~ I O - 3 .for a typical sheet steel and A~104, it follows that the
width/thickness ratio has to be larger than 103 (approximately). This suggests the rough rule
of thumb that a steel sheet t rnm thick must be at least t m wide before it can wrinkle
elastically.
(6) The variation of the compressive buckling loads for the uniform displacement model
with aspect ratio follows the corresponding constant compression curves more closely than
for the parabolic distribution, but are again slightly lower (see Fig. 5). The presence of large
regions where the lateral stress art is compressive for aspect ratios in the range 0.5 < ~ < 1
appears to cause a slight distortion to the curves.
(7) In practice, in both industrial sheet forming processes and the laboratory tests, the
onset of wrinkling will frequently occur in the plastic range. Matsui et al. I12-] have used the
nonlinear finite element package "MARC" to study the elastic-plastic deformation of a
square steel sheet pulled diagonally as in the Yoshida buckling test. In their analysis,
however, the. applied loads are simulated by a uniform tensile traction distribution, in
contrast to the uniform displacement boundary conditions used here.
 The elastic wrinkling of rectangular sheets 731

(8) The method for including the singularity proposed in Ref. [11] and summarized in the
Appendix appears to be an effective way of incorporating stress singularities without
excessive mesh refinement. The importance of including these singularities has been
demonstrated in this problem, since if they are neglected, no tensile wrinkling loads are
predicted at all.

Acknowledgements--The authors are grateful to Dr D. Bhattacharyya and Prof. R. Sowerby for helpful comments
on this investigation.

REFERENCES
1. K. YOSHIDA, H. KOMORIDA,M. USUDA,H. HAYASHI,H. OIKAWA,T. OHWUEand T. AMAIKE,Analysis and
control of buckling behavior due to elastic recovery in press forming. 13th Biennial Congress, IDDRG,
Melbourne, Australia, pp. 73-84 (1984).
2. K. YOSH1DA,S. HAYASHI,K. MIYAUCHI,Y. YAMATO,K. ABE, i . USUDA,R. ISHIDAand Y. OIKE, The effects of
mechanical properties of sheet metals on the growth and removing of buckles due to non uniform stretching.
Scient. Papers Inst. Phys. Chem. Res. 68, 85-93 (1974).
3. 1. AOKI, T. MATOBAand M. ATAKA,Effect of mechanical properties on the growth of wrinkles in sheet metal
forming. 12th Biennial Congress, IDDRG, S. Margherita, Ligure, Italy. pp. 221-229 (1982).
4. D. D. BRUSHand B. O. ALMROTH,Buckling of Bars, Plates and Shells, p. 94. McGraw-Hill, New York (1975).
5. C. B. MULLERand G. W. STEWART,An algorithm for generalised matrix eigenproblems. J. Numer. Anal. 10,
241-256 (1953).
6. S. TIMOSHENKOand J. N. GOODIER, Theory of Elasticity, p. 167. McGraw-Hill, New York (1951).
7. A. E. H. LOVE, A Treatise on the Mathematical Theory of Elasticity, p. 204. Dover, New York (1944).
8. M. L. WILLIAMS,Stress singularities resulting from various boundary conditions in angular corners of plates
under extension. J. appl. Mech. 19, 526-528 (1952).
9. M. L. WILLIAMS,Surface singularities resulting from various boundary conditions in angular comers of plates
under bending. 1st U.S. National Cong. of Appl. Mech., Illinois Institute of Technology, Chicago (1951).
10. A. H. ENGLAND, On stress singularities in linear elasticity. Int. J. Engng Sci. 9, 571-585 (1971).
11. A. F. EMERYand C. M. SEGEDIN,Singularity programming--A numerical technique for determining the effect
of singularities in finite difference solutions illustrated by application to plane elastic problems. Int. J. Numer.
Meth. Engng 6, 367-380 (1973).
12. M. MATSUI,N. IWATAand N. MORI, Initiation and growth of buckling in the biaxial diagonal tensile test on
steel sheet. J. Mech. Working Tech. 14, 283-294 (1987).

APPENDIX A

Notation

U vector of (u, /))T

Su, Sv singularity terms for u, v respectively
S =(Su, SO T
L finite difference linear operator of Navier equations
regular part of u
K singularity constant ('strength')
IIO, a 1 constituents of u: K =0, 1 respectively
L* modified one sided linear operator
Singularity programming
The poor accuracy of finite difference formulae near a singularity can be improved, without having to refine the
mesh excessively, by writing
u=~+KS, (AI)
where u =(u, v)r is the displacement vector, fi is the regular part of u, S are the singular terms referred to in equation
(18) and K is the 'strength' of the singularity, which is as yet undetermined. Let L be the linear operator arising from
the finite difference form of the Navier equations (12, 13): then,with the above decomposition of u into regular and
singular constituents, in equation (A1), it follows that

Lu = L~ + K L S . (A2)
Clearly u and S = (Su, SO T both satisfy the Navier equations and hence so does ~; therefore fi, being regular, satisfies
Lfi ~ 0. Thus, when account is taken of the singular nature of the displacements, the finite difference equations in the
interior of the plate are no longer homogeneous, but become

Lu = K L S . (A3)
The boundary conditions must also be modified to allow for the singular terms. For example, one of the boundary
732 R.H. SEGEDIN et al.

conditions on y=O is
Ou Ov
2~ ~xx+ 2 ( 1 - 9 ) ~ y = 0 . (A4)

The left hand side of equation (A4) in finite difference form is

9[ui+ l a - u l - 1j] +(1 - f:)[--3Vl.j+4Vi,j+ 1-- Vi,j+2].

With the use of equation (AI), this becomes

9[Ui+ l,j--Ui- l,j] -J-(1-- 9)[-- 3~i,j"['4~i,j+ l --~i,j+ 2]

+ K{v[.Sui+ 1 . j - S u l - 1.j] + [ 1 - 9 ] [ - 3 S v l . j + 4 S v i . j + 1 -Svi.j+2]}. (A5)
As fi is regular, the finite differences approximate to the derivatives, so that equation (A5) becomes

oa o~ - 9)
2 ~ x + 2 ¢ l - - 9 ) ~ y + K { }= 29 +20

--K[2q~+2(1 dSv-]
- 9 ) ~ - y J + K { }. (A6)

The first two bracketed termsare equal to zero as both u and S Satisfy the boundary conditions. The new finite
difference equation for the boundary condition then becomes
1 ,7
vij = ~ [4v, j +1 - vij +2] + ~ Lui +l j - - ui-1 j ]

-K {~[Sui+ L y - Su,_ Lj] +E4Sv,.j+ l - Sv,.i+ 2] } + KSvij. (A7)

The boundary conditions on the lines of symmetry have an extra term to calculate--for example, the condition
Ov/Ox = 0 on the line x = a/2 is
J_
2 (3v,j - 4v i_ 1j + vi- 2j ) -_ 1 ~ (3v,j--
- -
4v~ _ 1.1+ f~- 2j ) + ½K { 3SvLj -- 4Sv, _ ,.j + Sv~ _ 2.j }
dE Ov OSv
=h--+K{ }=h---Kh--+K{ }. (A8)
c~x dx 8x
As before, Ov/Ox=O, but this does not hold for OSv/Ox, so this must be calculated analytically and used in the
equation.
The following strategy is used to solve the modified finite difference algebraic equations.
(1) Put K = 0 and obtain u ° satisfying Lu = 0 and the non-homogeneous boundary conditions as shown in Fig. 7.
u ° is thus the standard finite difference solution found when no account is taken of the singularities.
(2) Put K = 1 and obtain another set of displacements u a satisfying
Lul = LS, (A9)
u ~ now being subject to homogeneous boundary conditions modified as indicated above.
(3) It is clear that u = u ° + K u a also satisfies equation (A3).
(4) K cannot yet be determined without using further information. The field equations have already been used in
the interior region, but on the boundary only the boundary conditions are invoked. The required new information
is obtained by insistence on the field equations at the boundary. The finite difference form of these equations will
involve one sided finite differences so that equation (A3) now becomes
L*u= KL*S

thereby yielding L*u ° + K L * u I = KL*S. (A 10)

Equation (A 10), being a vector equation, gives two values for K at each point of the boundary, but only points near
the singularity give well conditioned equations for K. If the K values are not consistent, the grid may not be fine
enough, or higher order singularities may be interfering. Obtaining consistent values of K is taken as confirmation
of the accuracy of the solution process. In these computations, a high degree of consistency of the K values was
achieved. When u has been found, the stresses are determined by use of a modified calculation to take account of the
singularity. Again, the analytical forms of the partial derivatives of Su and Sv are required.