Beruflich Dokumente
Kultur Dokumente
Finite element analysis of linear plates buckling under in-plane patch loading
Ghania Ikhenazen a, , Messaoud Saidani b , Abdelkrim Chelghoum a
a
article
info
Article history:
Received 14 August 2009
Accepted 4 March 2010
Keywords:
Linear buckling
Thin plates
Patch compression
Finite element method
abstract
The elastic buckling load is physically important in design because it is actually the critical step in the
changing plate configuration that will eventually lead to complete failure. The present work investigates
the problem of linear buckling of simply supported thin plates subjected to patch compression. In order
to satisfy the boundary conditions in a rigorous way, the authors chose the finite element method using
the exact stress distribution throughout the plate.
In the present paper, the stability problem treated using the total energy is briefly outlined. The plate
modelling is made by means of an eight noded rectangular element and a reduction of variable strategy
is applied to estimate the number of degrees of freedom leading to little or no loss in seeking solution
accuracy.
The buckling coefficient is determined for different load cases applied to a range of plate with various
edge ratios. The achieved results are summarised through different graphs representing variation of the
buckling coefficient against the plate ratio for each load case treated. A comparison with previous works
is made. Finally, it is shown that the resolution of the plate buckling problem using true stress distribution
with the finite element method leads to a good agreement with results previously obtained by means of
analytical methods using an exact stress distribution.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
The use of thin panels in many technical fields such as
aerospace, mechanical and civil engineering is nowadays quite
common. Since modern design process requires the evaluation of
appropriate safety levels, many studies have been carried out in
the last decades in order to describe the buckling due to uniform
compression [14], tension [5,6] and shear [1] for such structures.
On the other hand, a limited number of studies have been carried
out to evaluate the influence of patch loading on the critical
buckling load in the compressed plates although designers are
always confronted with this issue. Such a problem is encountered
in airframe where the action of the airloading on an aircraft wing
develops an axial loading that gives a non-uniform compression
that can lead to loss of stability. Also, the aerodynamic heating of
panels in supersonic aircraft can be approximated by non-uniform
thermal stresses as the temperature distribution is not uniform
throughout the volume of the restrained plate. In civil engineering
structures, engineers are often confronted with designs involving
partial edge loading, such as the buckling of the web plate of a
Corresponding author.
E-mail address: ghaniai@yahoo.fr (G. Ikhenazen).
0143-974X/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2010.03.006
crane girder under the action of heavy wheel loads applied to the
flanges.
In addition, even when this issue has been tackled, the
problem of the influence of the use of exact stress distribution
has not received sufficient attention and still remains open.
Indeed, authors such as [1,79] used simplified stress patterns in
investigating this problem, this simplification led to considerable
error. Later on, few authors published their works in which they
applied true stress distribution and consequently obtained reliable
results. Pavlovic and Baker [10] used an analytical method to
investigate thin plate buckling, Rockey [11] used finite element
method to investigate the buckling stiffened plate, and Stephen
and Steven [12] worked on the error estimation for plate buckling
element.
It is worth to point out that since constructional elements are
frequently subjected to in-plane patch loading and often prone
to buckling, it is important that further design data should be
provided to deal with this important stability problem. If such an
issue has so far received relatively little attention from researchers,
the reason for this is undoubtedly due to the additional theoretical
difficulties involved in obtaining rigorous solutions to the buckling
of plate when subjected to non-uniform compression. Undeniably,
the solution of this stability problem is mathematically difficult
to obtain as the stress distribution throughout the plate varies
0
0
0
0
1/2(1 )
0
0
0
1
0
0
0
0
1113
0
0
0
h2 /12
h2 /12
0
0
0
0
h2 /12
h2 /12
0
0
0
0
0
0
(2)
h2 (1 )/24
(3)
[ ] =
x
0
0
.
0
(4)
Utotal = Ue + We
Utotal = 1/2 {ue }T [Se ] Se0
Utotal
= 0.
{ue }
(8)
[S] S 0
[D] =
Eh
1 2
(7)
{ue } = 0
(9)
for one element, and for the whole plate it can be written as:
{ue } .
{ } = [D]{}
(6)
hence
[Se ] Se0
(5)
(1)
{u} = 0.
(10)
{u} = 0.
(11)
1114
has been treated analytically and has led to the well-known expression [1]:
K = Nxcr
b2
for a/b 1
2D
(12)
Eh3
12(1 2 )
u2
un [S] u1
u2
un
T
(15)
u2
un
00
u1
u2
un
T
(16)
The condition that minimizes the strain energy with respect to the
slave variable ur is
Sr1 u1 + + Srr ur + + Srn un = 0.
(17)
(18)
00
00
(13)
Sij00 = Sij00 Sir00 Sjr /Srr Sjr00 (Sir /Srr ) + Srr00 (Sir /Srr ) Sjr /Srr . (19)
00
[Sm ] f Sm
{ um } = 0
(14)
00
where [Sm ] and [Sm
] are respectively the reduced elastic and
geometric stiffness matrices.
An important question that arises is how to choose which
degrees of freedom are to be reduced out and which are to be
kept. In the buckling problem, the master degrees of freedom
should be those which are important in describing the membrane
strain energy. Therefore, the masters should be concentrated in the
regions of high in-plane stresses and low flexural stiffness, these
areas that are expected to buckle easily should contain masters.
On this basis, an automatic procedure has been implemented in the
program. This has been achieved by monitoring the ratio between
Sii and Sii00 , the leading diagonal stiffness and geometric stiffness
terms of degree of freedom i. When the ratio Sii /Sii00 is large then
either the in-plane stress at degree of freedom i is small or its
stiffness is large and hence it is well connected into the structure;
it is unwise to keep degree i as a master. On the other hand,
if Sii00 is large and/or Sii is small then the degree of freedom i is
likely to give rise to appreciable instability effects. The selection
technique involves keeping as masters (the degrees of freedom
that are to be kept after reduction) the degrees of freedom for
which the ratio Sii /Sii00 is small. After each elimination, the search
is applied to the reduced problem obtained by eliminating the
previous slave, not the original full system, and is repeated until
a specified number of masters remains. In practical application,
the matrix operations implied by the above equations need a great
computer core storage, thus the frontal solution is used.
Gi = Sir / Srr
(20)
Hi =
1
Gi
Sir Srr00
2
Srr
00
Srr
(21)
(22)
(23)
Three analyses are performed using different uniformly discretized meshes. In all cases the discretization is based on the
rectangular eight noded finite element. The division of the three
meshes is constructed in such a way that too long elements are
avoided.
Table 1 reports for each discretization, the total number of
elements used, the total number of degrees of freedom, the number
of degrees of freedom actually considered after reduction (or
masters), the ratio of number of masters/total number of degrees
of freedom, the minimal critical buckling coefficient K value and
the relative error reached.
As it is expected, it can be noticed from the figures that the
K value is slightly over estimated [14] and as the uniform grid is
more refined, the results reached are improved, such a tendency
characterises the finite element method.
It can be seen also that the increase in the number of master
degrees of freedom from 60 to 100 in both 50 and 100 elements
models has little effect on the minimal critical load coefficient
K even if the results obtained with 100 elements are slightly
improved.
It appears from the results that it would be better in using 200
master degrees of freedom in a refined grid than in using the same
number of masters in a coarser one. Indeed, the 200 unknowns
second idealisation led to a result of 4.0317, whereas in the third
one the result is definitely better with a value of 4.0023.
On the other hand, a ratio of about 17% of the total degrees of
freedom as master seems to be the more appropriate to conduct
with the present buckling analysis as in the three tests this
ratio represents the stating level from which the results become
interesting. Consequently, if good results have to be achieved, this
value should be used as a minimal ratio associated with, of course,
a refined grid.
It can be concluded that the buckling analysis requires a
significant number of nodal points to yield answers of good
accuracy, and elimination of unknowns appears to be of potential
use to solve this type of problem.
3. Results and discussion
3.1. Introduction
As stated previously, the aim of this investigation is to examine
the linear buckling behaviour of a range of mild steel plate, simply
supported and subjected to in-plane patch loading as depicted in
Fig. 1.
Knowing the importance of the plate thickness impact on the
buckling load [2123], the b/h ratio is fixed to 100 (thin plate).
The plate ratio a/b varies from 0.5 to 10, the modulus of elasticity
E = 209 109 N/m2 and Poissons ratio is taken equal to
0.3. The analyses are performed using uniformly refined meshes.
In all cases the discretization is based on the rectangular eight
noded finite element and the division of the meshes is constructed
in such a way that too long elements are avoided. The out of
plane displacements uz at master degrees of freedom for the first
three buckling loads are considered. The buckling coefficient K
corresponding to the lowest in-plane load which gives the plate
buckling is recorded.
3.2. Interpretation and comparison of results for load cases 0 <
l/b 1
The obtained numerical results give important quantitative and
qualitative information about the buckling behaviour of thin plates
under in-plane compressive patch loading.
Analyses are performed for six different loading cases corresponding to various relative load breath l/b = 0.1, 0.2, 0.4, 0.6,
1115
0.8 and 1 where l is the length (or breath) on which the loading is
applied and b is the plate width.
The buckling coefficient K is displayed against the plate ratio
a/b (a is the plate length) for the different load cases in Figs. 4 and
5. It can be noticed that the well known garland K vs. a/b curve [1],
corresponding to the uniform compression of simply supported
plate which can be obtained entirely by translating the particular
case where the buckling is in one half wave, has been obtained.
Moreover, it is immediately apparent that this particular curve
shape no longer applies when the compression is not uniform.
In the present work it can be seen from the drawn graphs, that
for short plates, i.e. for plates with a/b < 1, a very small increase in
the aspect ratio leads to a large decrease in the buckling coefficient
K . This effect is much less apparent when a/b 1. It can be observed also as a/b increases, the buckling coefficient K approaches
the value: l/4b which correlates with Saint Venants principle which
states that for large a/bs, the edge effect can be neglected and most
plates can be considered to be under a uniform state of stress. These
limiting values are shown on the right side of Figs. 3 and 4.
Table 2 illustrates the K values obtained, in the one hand, by
Yamaki [9] who used a simplified stress distribution, and in the
other hand, those obtained numerically in the present work and
analytically by Pavlovic and Baker [10], both investigations using
an exact stress distribution. First of all, it can be seen from this table
that in the case of l/b = 1, even though the authors have used
different methods of calculation the results are in accordance with
each others. Conversely, the figures are different when the ratio l/b
is different from 1. Indeed, it can be noticed that if Yamakis results
are reasonably accurate for uniformly compressed plate (l/b = 1),
they underestimate K for l/b 6= 1. Furthermore, in the case of
1116
Table 1
Minimal critical buckling stress coefficient K for a uniform compressed plate with a/b = 5 ratio.
Number of elements
Number of degrees of
freedom
Number of masters
Number of masters/number of
degrees of freedom in %
Relative error in %
25 2
349
60
100
17.19
28.65
4.0806
4.0802
2.02
2.01
20 5
640
60
100
200
9.38
15.63
31.25
4.0506
4.0506
4.0317
1.27
1.27
0.79
20 10
1260
60
80
100
130
200
4.76
6.35
7.94
10.32
15.87
4.4483
4.2252
4.0853
4.0051
4.0023
11.21
5.63
2.13
0.13
0.05
Table 2
K values for the problem depicted in Fig. 1. Comparison of results stemming from the use of true (PavlovicBaker and the present work) and approximate (Yamaki) stress
distributions.
a/b
0.5
1.0
1.4
2.0
l/b
1.0
0.1
Yamaki
PavlovicBaker
Present work
Yamaki
PavlovicBaker
Present work
6.25
4.00
4.47
4.00
6.25
4.00
4.47
4.00
6.27
4.00
4.49
4.00
27.09
19.34
21.12
19.34
28.68
21.43
26.78
25.78
29.59
22.12
27.30
26.06
Table 3
K values for the problem depicted in Fig. 2. A comparison of results stemming from
the use of true (PavlovicBaker and the present work) and approximate (Leggett
and Timoshenko) stress distributions.
a/b
1.00
0.66
0.50
0.40
0.33
Timoshenko
Leggett
PavlovicBaker
Present work
1.91
1.42
1.31
1.28
1.28
2.46
1.51
1.47
1.46
2.13
1.47
1.42
1.45
1.47
2.54
1.75
1.49
1.45
1.47
using the finite element in the present work and the analytical
method [10] and their corresponding counterpart obtained by
means of the approximate stress pattern employed by Leggett [7]
and Timoshenko [24]. It can be seen from the table that except
for the square plate case, the results obtained from the finite
element method, which use the true stress distribution are closely
approaching PavlovicBakers results. From Table 3, it appears also
that, for the range of square to narrow plate which are compressed
symmetrically by a pair of point loads, the computed error in
the buckling coefficient, in accordance to the other exact stress
distribution is not too large. Indeed, the discrepancy varies from
0% up to 16% in respect to PavlovicBaker results.
1117