You are on page 1of 9

Composite Structures 62 (2003) 261269

www.elsevier.com/locate/compstruct

Optimization of laminated composites subject


to uncertain buckling loads
S. Adali
a

a,*

, F. Lene b, G. Duvaut b, V. Chiaruttini

School of Mechanical Engineering, University of Natal, Center for Composite Materials and Structures, Durban 4041, South Africa
b
Universit
e P. et M. Curie, LM2S, 8 rue du Capitaine Scott, 75015 Paris, France
c
Ecole Normale Sup
erieure de Cachan, 61 ave du Pdt Wilson, 94235 Cachan, France

Abstract
Optimal design of composite laminates under buckling load uncertainty is presented. The laminates are subjected to biaxial
compressive loads and the buckling load is maximized under worst case in-plane loading which is computed using an anti-optimization approach. The magnitudes of the in-plane loads are not known a priori resulting in load uncertainty subject to the only
constraint that the loads belong to a given uncertainty domain. Results are given for continuous and discrete bre orientations
which constitute the optimization problem coupled to load anti-optimization problem leading to a nested solution method. It is
observed that the stacking sequence of a laminate designed for a deterministic load case only diers considerably from that of a
robust laminate designed taking load uncertainties into account. Consequently the buckling load carried by a deterministic design is
considerably less than the one carried by a robust design when both are subjected to uncertain loads.
2003 Elsevier Ltd. All rights reserved.
Keywords: Anti-optimization; Uncertain loading; Optimal design; Composite laminates; Buckling

1. Introduction
Load uncertainties refer to a situation in which the
loads (either magnitude or location or both) are not
known in a precise manner. Such situations occur quite
often in practice due to a lack of a priori knowledge as
to the exact loading that the structure will be subjected
to under operational conditions. Regardless of these
circumstances, a composite structure has to be designed
optimally in order to realize its potential. In these situations design has to be based on the worst case loading
to avoid structural failure.
Even though uncertainty in loading under working
conditions is a fairly common occurrence, most of the
design studies involving composites are based on deterministic loads. A review of the optimal design of
laminated composites under deterministic buckling
loads is given in [1]. More recent studies of deterministic
buckling optimization of composites include [27] where
various complicating eects including the bending
twisting coupling [2], unsymmetric lamination [3], shear

Corresponding author. Fax: +27-31-260-3203/3217.


E-mail address: adali@nu.ac.za (S. Adali).

0263-8223/$ - see front matter 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2003.09.024

deformation [4], cut-outs [5], shape eciency [6], and


thermal eects [7,8] were taken into account. In these
studies loading conditions were specied and the corresponding optimal congurations were determined
shedding light to the eect of the specic aspect under
consideration. Another avenue of research was directed
towards developing more eective and faster optimization algorithms with particular applications to the
buckling optimization of laminates. The foremost
among these techniques is the genetic algorithms which
were improved and rened by Haftka and his coworkers in several studies [914] with an increasing level
of sophistication.
However, uncertainty in loading, material properties,
manufacturing tolerances and boundary conditions
occur quite frequently and there has been an increasing
amount of research work directed towards formulating
and solving such problems. The review articles by Elishako [15] and Ben-Haim [16] provide in-depth reviews
of the subject with particular reference to buckling and
vibration, respectively. Particular emphasis in the present
study is design optimization under load uncertainty with
application to laminated composites. Earlier examples of
optimization of composites subject to uncertain loads
include [1720] which used the convex modelling of

262

S. Adali et al. / Composite Structures 62 (2003) 261269

uncertainty as developed in the book by Ben-Haim and


Elishako [21] as well as fuzzy modelling [17] in problem
formulation and solution. Reliability techniques were
used in the studies [2226] to deal with uncertainties in
the optimization of composite structures. To design a
structure under unknown loading, material, etc. conditions invariable leads to a mathematical problem which
requires the computation of worst possible case with
respect to the unknown quantities. Elishako et al. [27]
introduced the term anti-optimization to describe this
problem the solution of which basically involves the
minimization (maximization) of the performance index
with respect to uncertain quantities if the design objective
is to maximize (minimize) it and this leads to a minmax
problem. Anti-optimization has been used in the solution
of a number of structural design problems [2831] as well
as in the design of laminated composites [13,3234] involving load uncertainties. Haftka and co-workers
[13,32,34] developed two-species genetic algorithms to
deal with the two levels of optimization in the solution of
anti-optimization problems. Minmax formulation was
applied to the solution of a number of composite optimization problems with uncertain loadings in [3538]
with Refs. [37,38] considering designs under uncertain
buckling loads.
In the present work optimal designs of symmetrically
laminated composites are given to maximize the biaxial
buckling load when the only information available on
loading is that the compressive loads belong to a given
uncertainty domain [39]. Numerical results are given for
two such domains, namely, a triangular and a circular
one for non-hybrid and hybrid laminates. The design
variables are taken as the ply angles which could change
continuously or discretely in which case their values are
taken from a dened set of possible values. Optimization
and anti-optimization problems are solved by treating
the overall problem as a nested optimization problem.

2. Anti-optimization problem and uncertainty domains


A rectangular laminate of length a, width b and
thickness H is subjected to in-plane compressive buck-

ling loads of kNx and kNy in the x and y directions where


k is the load factor. The uncertain loads Nx and Ny lie in
a domain bounded by
Nx 0; Ny 0;

Nxp Nyp 1

where p 1; 2; . . . ; 1. Thus the magnitudes of the loads


Nx and Ny are unknown and can take any value in the
uncertainty domains Up dened by the sets
Up fNx ; Ny j Nx P 0; Ny P 0; Nxp Nyp 6 1g

The shape of the uncertainty domain Up depends on the


exponent p and is given by a triangular, circular and
rectangular region for p 1; 2 and 1, respectively, as
shown in Fig. 1.
The uncertain quantities, Nx , Ny 2 Up are unknown
variables of the anti-optimization problem which consists of determining Nx and Ny such that the buckling
load k of the laminate is minimized for any given conguration involving the design variables.
The stacking sequence (h1 ; h2 ; . . .) of the laminate is
specied as the design variable of the problem with hk
indicating the ply angle of the k th layer. Let
kmn hK ; N khK ; N ; m; n denote the load factor for the
buckling modes (m; n) where hK h1 ; h2 ; . . . ; hK is the
set of design variables, K is the number of layers and
N Nx ; Ny is the set of uncertainty variables. The
buckling load factor is given by
khK ; N min khK ; N ; m; n
m;n

The anti-optimization problem can be stated as


Determine the uncertainty variables N Nx ; Ny 2
Up for a given p such that
khK ; N  min khK ; N
N 2Up

Thus the uncertainty variables are determined so as


to minimize the buckling load which constitutes the
solution of the anti-optimization problem which is
coupled to the optimization problem since hK and N are
interdependent. Thus the solution of the problem (4)
yields the worst case of in-plane compressive loading on
the laminate in terms of load magnitudes for the given
values of the design variables hK .

Fig. 1. The uncertainty domains for p 1; 2 and 1.

S. Adali et al. / Composite Structures 62 (2003) 261269

The buckling load is to be maximized by choosing the


set of ply angles hK h1 ; h2 ; . . . ; hK optimally. Thus
the optimization problem can be stated as
Determine the design variables (h1 ; h2 ; . . . ; hK ) such that
khK ; N  max khK ; N 

hk

Clearly the optimal values of hk depend on the antioptimal values of N Nx ; Ny and vice versa leading to
a nested optimization/anti-optimization problem. The
solution of this problem gives the best stacking sequence
so as to maximize the buckling load under the worst
possible case of biaxial compressive loading. As such it
provides a robust optimal laminate design capable of
operating under any loading which lies within the given
uncertainty domain Up .

3. Basic equations and method of solution


The design problem formulated in Section 2 is solved
for a symmetrically laminated plate with equal layer
thicknesses t with the total thickness given by H Kt.
The boundary conditions are taken as simply supported.
The equation governing the buckling of a rectangular
laminate under biaxial loads is given by
D11 wxxxx 4D16 wxxxy 2D12 2D66 wxxyy 4D26 wxyyy
D22 wyyyy kNx wxx Ny wyy 0

where Dij are the exural stinesses, and w wx; y is


the deection. The results are obtained for a laminate
with a large number of layers for which the bending/
twisting coupling stinesses, i.e., D16 and D26 terms, are
neglected. The results by Nemeth [40] indicate that these
terms are negligible provided the constraints
1=4

d D16 =D311 D22


c

D26 =D11 D322 1=4

6 0:2;
7

6 0:2

are satised.
The buckling load k for the simply supported conditions and with D16 D26 0 is given by
2

hmn HK ; N

p2 D11 m4 2D12 2D66 mnr D22 nr

263

are chosen from a set of discrete values hk 2 hi j i


1; 2; . . . ; I where I is the number of possible bre orientations.
In the nested design optimization problem involving
optimization with respect to design variables and antioptimization with respect to uncertainty variables, the
computational procedure is to anti-optimize at each step
of the optimization to ensure that the nal result corresponds to the worst case loading. The anti-optimization problem for each uncertainty domain Up needs to
be solved dierently. For the case p 1, i.e., the triangular region (Fig. 1), the vertex points of the domain
provide the extremum points due to the linear nature of
the resulting problem. Thus for this case the anti-optimal load cases are given either by Nx 1 kN/m and
Ny 0, or Nx 0 and Ny 1 kN/m both of which need
to be checked for any laminate conguration. For the
case p 2, i.e., the quarter circular region (Fig. 1), the
location of the worst case loading lies on the circle
Nx sin a and Ny cos a with 0 6 a 6 90 and the value
of a giving this location depends on the problem parameters, and in particular, on the aspect ratio. The
exact anti-optimal point of the arc has to be computed
as a separate minimization problem. The anti-optimal
value of a is computed by minimizing the buckling load
with respect to a. The same minimization procedure to
compute the anti-optimal point applies to all cases with
3 6 p < 1. For p 1, the extremum point is given by
Nx Ny 1 kN/m for all laminate congurations, i.e.,
this case becomes a deterministic problem.
The continuous case involving only one design variable is solved by the golden section method. The discrete
case involving several possible design angles is solved by
checking all possible combinations which includes not
only the angle combinations but also the sequence
combinations due the presence of Dij terms, i.e., nonlinear thickness dependence, in the expression (8) for the
buckling load. For a symmetric laminate with K layers,
the number of combinations is I K=2 . This approach
leads to exact solutions, i.e. to global optimum points,
and could be used as a benchmark in studies using approximate techniques.

a2 m2 Nx nr Ny

8
where r a=b is the aspect ratio.
The design problem is solved for two cases of bre
orientations, namely, the continuous case for which the
stacking sequence is taken as (h=  h= h=  h= . . .sym ,
and the discrete case for which the stacking sequence is
(h1 =h2 = . . . =hK=2 sym . In the discrete case, the ply angles

4. Numerical results
Consider a multi-layered laminate made of either
graphiteepoxy layers, or glassepoxy layers, or a combination of both, i.e., a hybrid laminate. The hybrid laminate consists of outer layers of graphite/epoxy and inner
layers of glass/epoxy. The material properties are given as

Graphiteepoxy T300=5280 : E1 181 GPa; E2 10:3 GPa; G12 7:17 GPa; m12 0:28
Glassepoxy Scotch-ply 1002 : E1 38:6 GPa; E2 8:27 GPa; G12 4:14 GPa; m12 0:26

264

S. Adali et al. / Composite Structures 62 (2003) 261269

In the calculations, the total thickness H of the laminate is kept constant. However, the thickness t of a
single layer given by t H =K varies as K changes. The
reason for this is to compare the performance of equal
thickness designs with respect to the design objective
regardless of the number of layers. The set of possible
bre orientations hk 2 hi ji 1; 2; . . . ; I to be used in
the discrete design is taken as (0; 45; 45; 90), i.e.,
I 4. It is noted that the buckling load as given by Eq.
(8) in terms of D11 , etc., does not depend on the positive
and negative values of the ply angle since the terms D16
and D26 are neglected. Thus computationally only the
orientations 0, 45 and 90 need to be taken into account when computing kmn hK ; N . For a 16-layered
symmetric laminate used in this study the number of
angle combinations is 38 6561. However, the angle
)45 will be used to impose the constraints d 6 0:2,
c 6 0:2 as given by Eq. (7). Initially the optimum laminate stacking sequence is determined and if the constraints (7) cannot be satised using )45, then that
conguration is rejected. There could be cases where
more than one conguration gives the same maximum
buckling load. In these situations the optimum conguration is determined by choosing the one which gives

First the results for the case of a continuous optimum


ply angle are given in terms of the aspect ratio. Table 1
shows the optimal bre orientations of a
h=  h= h=  h= . . .sym laminate where the deterministic design refers to an optimal design for a known
loading case, e.g., under the in-plane loads (Nx 1 kN/
m and Ny 0) or (Nx 0 and Ny 1 kN/m) which are
the vertex points of the uncertainty domain U1 . The
robust design refers to optimal designs under the
uncertain loads Nx ; Ny 2 U1 with the buckling load denoted by kmax . It is observed that in the case of the
graphiteepoxy laminate hopt values for the robust design coincide with those of deterministic designs for both
cases of (Nx 1 kN/m and Ny 0) and (Nx 0 and
Ny 1 kN/m) for aspect ratios a=b 1:0 and 1.2 only,
i.e., around the square plate. These values coincide for
a=b 0:8, 1.0 and 1.2 in the case of glassepoxy laminates. For some aspect ratios hopt values of robust and
deterministic designs coincide indicating that the worst
case loading is the same as the corresponding deterministic loading for that aspect ratio. However, for most
aspect ratios the optimal hopt values dier from one or
both deterministic cases and if the design is based on one
of the deterministic loads, the laminate would carry a

Table 2
Buckling load ratios for h=  h= h=  h= . . .sym laminates with
robust and deterministic designs under uncertain loads Nx ; Ny 2 U1

minmaxd; c

The results are given for uncertainty domains U1 and


U2 , i.e., for the triangular and the circular domains. In
the case of a rectangular domain U1 , the anti-optimum
point is always at the same point and as such the antioptimization problem ceases to exist. In the computations H 0:1 m, b 1 m (a br), and the units of Nx
and Ny are in kN/m. In the results given below graphite
epoxy and glassepoxy indicate laminates made out of
these materials and graphiteglass indicates a hybrid
laminate with equal number of graphiteepoxy outer
layers and glassepoxy inner layers, i.e., 50% graphite
and 50% glassepoxy.

a=b

Graphiteepoxy

Glassepoxy

kmax =kx

kmax =ky

kmax =kx

kmax =ky

0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

1.37
1.46
1.22
1.00
1.00
1.00
1.06
1.26
1.40
1.44

2.46
1.78
1.36
1.00
1.00
1.00
1.00
1.00
1.55
1.52

1.00
1.00
1.00
1.00
1.00
1.00
1.05
1.24
1.30
1.35

2.04
1.61
1.26
1.00
1.00
1.00
1.00
1.00
1.00
1.00

Table 1
Comparison of deterministic and robust designs for h=  h= h=  h= . . .sym laminates with Nx ; Ny 2 U1
a=b

0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

Graphiteepoxy

Graphite/glass

Glassepoxy

Deterministic

Robust

Deterministic

Robust

Deterministic

Nx 1

Ny 1

Nx ; Ny 2 U1

Nx 1

Ny 1

Nx ; Ny 2 U1

Nx 1

Ny 1

Robust
Nx ; Ny 2 U1

hopt

hopt

hopt

kmax

hopt

hopt

hopt

kmax

hopt

hopt

hopt

kmax

0.0
0.0
11.6
37.3
45.0
51.1
45.4
40.3
41.7
45.0

45.0
45.8
51.1
40.3
45.0
51.1
58.3
70.3
90.0
90.0

14.1
14.4
16.8
37.3
45.0
51.1
58.3
70.3
74.2
74.7

3423.1
908.4
446.7
334.1
316.1
227.7
184.1
163.9
156.4
151.6

0.0
0.0
11.5
37.3
45.0
51.1
45.4
40.2
41.6
45.0

45.0
45.8
51.6
40.2
45.0
51.1
58.3
70.4
90.0
90.0

12.6
13.8
16.4
37.3
45.0
51.1
58.3
70.4
74.7
75.2

3147.0
824.7
404.6
302.7
286.4
206.3
166.8
148.4
141.7
137.5

0.0
0.0
6.3
36.9
45.0
51.4
45.6
37.0
41.4
45.0

45.0
46.3
51.4
37.0
45.0
51.4
59.2
72.6
90.0
90.0

0.0
0.0
6.3
36.9
45.0
51.4
59.2
72.6
90.0
90.0

822.8
219.7
109.2
82.7
78.2
56.4
45.5
40.2
38.2
37.0

S. Adali et al. / Composite Structures 62 (2003) 261269

265

Table 3
Comparison of deterministic and robust designs for h=  h= h=  h= . . .sym laminates with Nx ; Ny 2 U2
a=b

Graphiteepoxy
Robust design Nx ; Ny 2 U2

0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

Deterministic design under uncertain loads

Graphite/glass

Glassepoxy

Robust design
Nx ; Ny 2 U2

Robust design
Nx ; Ny 2 U2

hopt

kmax

kmax =kx

kmax =ky

hopt

kmax

hopt

kmax

87.3
80.1
70.1
54.9
44.1
35.1
27.9
21.2
19.8
19.3

15.0
15.0
11.7
37.3
45.0
51.1
58.3
70.3
86.4
83.0

3374.3
892.3
421.0
281.6
223.6
187.0
164.0
152.6
151.8
151.3

1.33
1.47
1.21
1.00
1.00
1.00
1.06
1.26
1.39
1.44

2.43
1.77
1.36
1.00
1.00
1.00
1.00
1.00
1.51
1.52

15.0
15.0
16.0
37.3
45.0
51.1
58.3
70.4
74.3
74.8

3046.7
806.5
380.9
255.1
202.5
169.4
148.6
138.3
135.3
133.3

0.0
0.0
5.0
36.9
45.0
51.4
59.2
72.6
90.0
90.0

822.2
216.9
102.7
69.7
55.3
46.3
40.6
37.5
36.5
35.9

lower buckling load under uncertain loads with Nx and


Ny taking any value within the uncertainty domain. This
aspect is investigated further in Table 2 which shows the
ratios kmax =kx and kmax =ky where kx indicates the buckling load of a laminate optimally designed for the load
case Nx 1 kN/m and Ny 0 and subjected to the uncertain loads Nx , Ny 2 U1 . Similarly ky indicates the same
quantity for the load case Nx 0 and Ny 1 kN/m. In

Fig. 2. Optimal ply angle versus the aspect ratio for uncertainty domains U1 and U2 .

the case of graphiteepoxy the buckling ratio becomes


as high as 2.46 for kmax =ky at a=b 0:2 and in general
the robust design provides fairly high buckling loads for
small or large aspect ratios. In the case of glassepoxy
laminate, this ratio is 1.0 for a=b 6 1:2 for laminates
designed under the loading Nx 1 kN/m and Ny 0 and
for a=b P 0:8 for laminates designed under the loading
Nx 0 and Ny 1 kN/m indicating that the material
properties have a distinct inuence on the robust design.
Next the results are given for the uncertainty case
Nx ; Ny 2 U2 in Table 3. The worst case loading lies on the
circular boundary of the domain which is given by
Nx sin a and Ny cos a. The value of a giving the
worst case loading is shown in the second column of
Table 3 which ranges from 87.3 for a=b 0:2 to 19.3
for a=b 2:0. The values of hopt are similar for graphite
epoxy and hybrid laminates, but quite dierent for the
glassepoxy one for a=b 6 0:6 and a=b P 1:8. A comparison of the optimal h values are shown in Fig. 2
which plots the values of hopt against the aspect ratio a=b
for Nx ; Ny 2 U1 and Nx ; Ny 2 U2 . The dierences between
the hopt values of dierent material groups materialize
for small and large aspect ratios with hopt values being
similar for 0:8 6 a=b 6 1:6. For the graphiteepoxy
laminate a comparison of the buckling loads of robust

Table 4
Deterministic and robust designs for graphiteepoxy laminates with the stacking sequence h1 =h2 =h3 =h4 =h5 =h6 =h7 =h8 sym and Nx ; Ny 2 U1
a=b
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

Deterministic design

Robust Design

Nx 1 kN/m, Ny 0

Nx 0, Ny 1 kN/m

Nx ; Ny 2 U1

kmax

0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
)45/45/45/)45/45/)45/)45/45
45/)45/)45/45/)45/45/45/)45
45/)45/45/)45/)45/)45/)45/)45
45/)45/45/)45/)45/)45/)45/)45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
45/)45/)45/45/)45/45/45/)45

45/)45/)45/45/)45/45/45/)45
45/)45/45/)45/)45/)45/)45/)45
)45/45/45/)45/45/)45/)45/)45
)45/45/45/)45/45/)45/)45/45
45/)45/)45/45/)45/45/45/)45
45/)45/45/)45/)45/)45/)45/)45
45/)45/45/)45/)45/)45/)45/)45
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90

0/0/0/45/0/0/)45/45
0/0/0/)45/0/0/90/90
0/0/0/45/)45/0/0/0
)45/45/45/)45/45/)45/)45/45
45/)45/)45/45/)45/45/45/)45
45/)45/45/)45/)45/)45/)45/)45
45/)45/45/)45/)45/)45/)45/)45
90/90/90/45/)45/90/90/90
90/90/90/45/)45/90/90/90
90/90/90/45/90/)45/90/90

3446.7
895.8
431.9
325.5
316.1
223.9
172.5
157.9
151.6
149.0

266

Table 5
Deterministic and robust designs for graphite/glass hybrid laminates with the stacking sequence h1 =h2 =h3 =h4 =h5 =h6 =h7 =h8 sym and Nx ; Ny 2 U1 (underline indicates the graphiteepoxy plies)
a=b

Robust design

Nx 1 kN/m, Ny 0

Nx 0, Ny 1 kN/m

Nx ; Ny 2 U1

kmax

0=0=0=0/0/0/0/0
0=0=0=0/0/0/0/0
0=0=0=0/0/0/0/0
45=45=45=  45=45=45=45=45
45=45=45=  45=45=  45=  45=  45
45=  45=  45=45=  45=  45=  45=  45
45=45=45=  45=  45=  45=  45=  45
45=45=45=  45=45=45=45=45
45=45=45=  45=45=45=45=45
45=  45=  45=45=  45=  45=  45=  45

45=  45=  45=45=  45=  45=  45=  45


45=  45=  45=45=  45=  45=  45=  45
45=  45=  45=45=  45=  45=  45=  45
45=45=45=  45=45=45=45=45
45=45=45=  45=45=  45=  45=  45
45=  45=  45=45=  45=  45=  45=  45
45=45=45=  45=  45=  45=  45=  45
90=90=90=90/90/90/90/90
90=90=90=90=90=90=90=90
90=90=90=90/90/90/90/90

0=0=0=  45=0=0=0=0
0=0=0=  45=0=0=0=45
0=0=0=  45=90=90=90=90
45=45=45=  45=45=45=45=45
45=45=45=  45=45=  45=  45=  45
45=  45=  45=45=  45=  45=  45=  45
45=45=45=  45=  45=  45=  45=  45
90=90=90=45/0/0/0/0
90=90=90=  45=0=45=0=45
90=90=90=  45=90=  45=0=90

3115.8
818.3
385.8
295.0
286.4
202.8
156.3
140.4
137.0
135.3

Table 6
Deterministic and robust designs for glassepoxy laminates with the stacking sequence h1 =h2 =h3 =h4 =h5 =h6 =h7 =h8 sym and Nx ; Ny 2 U1
a=b
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

Deterministic design

Robust design

Nx 1 kN/m, Ny 0

Nx 0, Ny 1 kN/m

Nx ; Ny 2 U1

kmax

0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45

)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90

0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
)45/45/45/)45/45/)45/)45/45
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90

822.8
219.7
109.2
81.1
78.2
55.6
43.3
40.0
38.2
37.0

S. Adali et al. / Composite Structures 62 (2003) 261269

0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

Deterministic design

The problem of stacking sequence design of a symmetrically laminated plate for maximum buckling load
was solved under uncertain biaxial loads which belong
to a given domain. Two such domains, namely, triangular and circular uncertainty domains, have been

822.2
216.9
102.7
68.3
55.3
45.7
38.6
37.3
36.5
35.9
0=0=0=  45=0=45=45=0
0=0=0=45=  45=90=  45=0
0=0=0=-45/45/45/45/45
45=45=45=  45=45=45=45=45
45=45=45=  45=45=45=45=45
45=45=45=  45=45=45=45=45
45=45=45=  45=45=45=45=45
90=90=45=  45=90=90=90=90
90=90=90=45/90/90/90/90
90=90=90=45/90/90/90/90

0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
0/0/0/0/0/0/0/0
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90
90/90/90/90/90/90/90/90
3092.1
796.0
367.7
248.6
202.5
166.6
139.3
133.2
130.9
129.0

Stacking sequence

0/0/0/)45/0/45/0/0
0/0/0/)45/0/45/45/0
0/0/0/45/)45/0/0/0
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
45/)45/)45/45/)45/45/45/)45
90/90/90/45/)45/90/)45/90
90/90/90/45/)45/90/90/90
90/90/90/45/)45/90/90/90

3393.0
878.1
406.4
274.4
223.6
183.9
153.7
146.9
144.8
142.9

Glassepoxy

Stacking sequence
kmax

Graphite/glass

Stacking sequence

kmax
Graphiteepoxy

kmax

267

0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0

5. Conclusions

a=b

and deterministic designs is made by computing the


buckling loads of laminates designed under a deterministic load but subjected to uncertain loads
Nx ; Ny 2 U2 . In columns 5 and 6, kx and ky denote the
buckling loads of laminates designed for the load cases
(Nx 1 kN/m and Ny 0) and (Nx 0 and Ny 1 kN/
m), respectively, but subjected to uncertain loads
Nx ; Ny 2 U2 . The ratios kmax =kx and kmax =ky show that
the laminates designed for a single load case and subjected to load uncertainty will underperform a robust
design substantially for aspect ratios less than 0.6 and
more than 1.6.
The results for discrete designs are given for laminates with 16 layers in Tables 47. Tables 46 show the
optimal stacking sequences of graphiteepoxy, hybrid
and glassepoxy laminates, respectively, designed under
the deterministic and uncertain loads Nx ; Ny 2 U1 . In
Table 5, the graphite layers are indicated with an underline. The robust designs dier considerably from the
deterministic ones, in particular, for small and large
aspect ratios. These dierences are even greater in the
case of the hybrid laminates. The kmax values for the
graphiteepoxy and hybrid laminates are again observed
to be fairly close, i.e., within 10% of each other, indicating the eciency of a hybrid design with the expensive and stier layers (e.g. graphite) on the outside and
cheaper and less sti layers (e.g. glass) on the outside.
Table 7 shows the optimal stacking sequences and
kmax values for all three laminates for the case
Nx ; Ny 2 U2 . The a values indicating the location of the
worst case loading on the circular boundary of the domain U2 are similar to the ones given in Table 3 and are
not shown here. Comparison of Tables 47 shows that
the optimal stacking sequences of the graphiteepoxy
and hybrid laminates dier substantially for the uncertainty cases of U1 and U2 indicating the eect of the
specic uncertainty domain on the optimal design.
However, for the glassepoxy laminate both uncertainty
domains produce the same optimal stacking sequences.
A comparison of kmax values for the graphite and
hybrid laminates in Tables 1,35 and 7 show that these
values are fairly close for with the reduction in the hybrid kmax around 10% for all aspect ratios. It is noted
that the hybrid laminate under consideration uses 50%
less graphite providing a considerable material cost
saving with only 10% reduction in the load carrying
capacity.

Table 7
Deterministic and robust designs with the stacking sequence h1 =h2 =h3 =h4 =h5 =h6 =h7 =h8 sym and Nx ; Ny 2 U2 (underline indicates the graphiteepoxy plies in the hybrid laminate)

S. Adali et al. / Composite Structures 62 (2003) 261269

268

S. Adali et al. / Composite Structures 62 (2003) 261269

investigated quantitatively for graphiteepoxy, glass


epoxy and hybrid laminates. The optimal stacking
sequences were computed for a continuous bre orientation and for discrete bre orientations using only
0, 45 and 90 ply angle combinations. The numerical
solutions were achieved by following a nested optimization strategy whereby for a given stacking sequence,
the least favourable loading (the anti-optimization
problem) was determined and the buckling load was
computed. This load was maximized by a one dimensional optimization routine in the continuous case and
by checking all the combinations in the discrete case.
The comparative numerical were given for aspect
ratios ranging from 0.2 to 2.0. The optimal designs of
laminates under a specied deterministic loading were
also given in order to compare the performance of the
robust designs with the deterministic ones. The two
designs were found to dier considerably for low and
high aspect ratios, but similar around a=b 1:0. The
buckling loads for laminates designed under a deterministic load but subjected to uncertain loads were also
computed. These results indicated that the buckling load
of a robust design could be as high as 2.46 times higher
than that of a deterministic one for low aspect ratios
emphasizing the importance of an optimal design which
takes load uncertainties into account. Discrete designs
(Tables 47) showed that the material properties of the
laminate has a distinct inuence on the optimal values of
the ply angles. A comparison of hybrid and graphite
epoxy buckling loads demonstrated the eciency of a
hybrid design where only 50% of the material is graphite
placed in the outer layers. The decrease in the buckling
loads as compared to a non-hybrid graphiteepoxy
laminate was around 10% for all aspect ratios.
Load uncertainties are normally part of the working
conditions for composites operating under a variety of
loadings which are usually unpredictable. A robust design becomes a necessary part of the optimization procedure since by its very nature an optimized design is the
strongest for the given load conditions and quite weak if
these conditions happen to change.
References
[1] Adali S. Lay-up optimization of laminated plates under buckling
loads. In: Turvey GJ, Marshall IH, editors. Buckling and
postbuckling of composite plates. Chapman and Hall; 1995.
[2] Walker M, Adali S, Verijenko VE. Optimization of symmetric
laminates for maximum buckling load including the eects of
bendingtwisting coupling. Comp Struct 1996;58(2):3139.
[3] Aiello MA, Ombres L. Maximum buckling loads for unsymmetric
thin hybrid laminates under in-plane and shear forces. Compos
Struct 1996;36(12):111.
[4] Adali S, Richter A, Verijenko VE. Optimization of sheardeformable laminated plates under buckling and strength criteria.
Compos Struct 1997;39(34):16778.

[5] Walker M. Optimal design of symmetric laminates with cut-outs


for maximum buckling load. Comp Struct 1999;70(3):33743.
[6] Weaver PM. Design of laminated composite cylindrical shells
under axial compression. Compos Part B: Eng 2000;31(8):66979.
[7] Spallino R, Thierauf G. Thermal buckling optimization of
composite laminates by evolution strategies. Comp Struct
2000;78(5):6917.
[8] Foldager JP, Hansen JS, Olho N. Optimization of the buckling
load for composite structures taking thermal eects into account.
Struct Multidisc Optim 2001;21:1431.
[9] Le Riche R, Haftka RT. Optimization of stacking sequence design
for buckling load maximization by genetic algorithm. AIAA J
1993;31(5):9516.
[10] Gurdal Z, Haftka RT, Nagendra S. Genetic algorithm for the
design of laminated composite panels. SAMPE J 1994;30(3):29
35.
[11] Kogiso N, Watson LT, Gurdal Z, Haftka RT, Nagendra S.
Design of composite laminates by a genetic algorithm with
memory. Mech Compos Mater Struct 1994;1:95117.
[12] Liu B, Haftka RT, Akgun MA, Todoroki A. Permutation genetic
algorithm for stacking sequence design of composite laminates.
Comp Meth Appl Mech Eng 2000;186(24):35772.
[13] Liu B, Haftka RT, Akgun MA. Two-level composite wing
structural optimization using response surfaces. Struct Multidisc
Optim 2000;20:8796.
[14] Soremekun G, Gurdal Z, Haftka RT, Watson LT. Composite
laminate design optimization by genetic algorithm with generalized elitist selection. Comp Struct 2001;79(2):13143.
[15] Elishako I. Uncertain buckling: its past, present and future. Int J
Solids Struct 2000;37(4647):686989.
[16] Ben-Haim Y. Reliability of vibrating structures with uncertain
inputs. Shock Vibr Dig 1998;30(2):10613.
[17] Adali S. Convex and fuzzy modelling of uncertainties in the
optimal design of composite structures. In: Pedersen P, editor.
Optimal design with advanced materials. Amsterdam, The Netherlands: Elsevier; 1993. p. 17389.
[18] Adali S, Richter A, Verijenko VE. Non-probabilistic modelling
and design of sandwich plates subject to uncertain loads and
initial deections. Int J Eng Sci 1995;33:85566.
[19] Adali S, Richter A, Verijenko VE. Minimum weight design of
symmetric angle-ply laminates under multiple uncertain loads.
Struct Multidisc Optim 1995;9:8995.
[20] Adali S, Richter A, Verijenko VE. Minimum weight design of
symmetric angle-ply laminates with incomplete information on
initial imperfections. ASME J Appl Mech 1997;64:906.
[21] Ben-Haim Y, Elishako I. Convex models of uncertainty in
applied mechanics. Amsterdam, The Netherlands: Elsevier; 1990.
[22] Murotsu Y, Miki M, Shao S. Reliability design of bre reinforced
composites. Struct Safety 1994;15:3549.
[23] Boyer C, Beakou A, Lemaire M. Design of a composite structure
to achieve a specied reliability level. Reliab Eng Syst Safety
1997;56(3):27383.
[24] Kogiso N, Shao SM, Murotsu Y. Reliability-based optimum
design of a symmetric laminated plate subject to buckling. Struct
Optimiz 1997;14:1824.
[25] Miki M, Murotsu Y, Tanaka T, Shao S. Reliability-based
optimization of brous laminated composites. Reliab Eng Syst
Safety 1997;56(3):28590.
[26] Kogiso N, Shao SM, Murotsu Y. Reliability-based design of
symmetric laminated plate with initial imperfections. In: Proceedings of AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization, St Louis, USA, 24 September
1998. AIAA Paper 98-4912. p. 152836.
[27] Elishako I, Haftka RT, Fang J. Structural design under bounded
uncertaintyoptimization with anti-optimization. Comp Struct
1994;53(6):14015.

S. Adali et al. / Composite Structures 62 (2003) 261269


[28] Zhiping Q, Elishako I. Antioptimization of structures with large
uncertain-but-non-random parameters via interval analysis.
Comp Meth Appl Mech Eng 1998;152(34):36172.
[29] Lombardi M. Optimization of uncertain structures using nonprobabilistic models. Comp Struct 1998;67:99103.
[30] Ganzerli S, Pantelides CP. Optimum structural design via convex
model superposition. Comp Struct 2000;74(6):63947.
[31] McWilliam S. Anti-optimisation of uncertain structures using
interval analysis. Comp Struct 2001;79(4):42130.
[32] Venter G, Haftka RT. A two-species genetic algorithm for
designing composite laminates subjected to uncertainty. In:
Proceedings of the 37th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, Salt Lake
City, Utah, 1517 April 1996. Paper No. 96-1535, Part 3. p. 1848
57.
[33] Lombardi M, Haftka RT. Anti-optimization technique for
structural design under load uncertainties. Comp Meth Appl
Mech Eng 1998;157(12):1931.
[34] Venter G, Haftka RT. Two-species genetic algorithm for design
under worst case conditions. Evolution Optimiz 2000;2(1):119.
[35] Cherkaev E, Cherkaev A. Design versus loading: minmax
approach. In: Proceedings of the Third ISSMO/UBCAD/UB/

[36]

[37]

[38]

[39]

[40]

269

AIAA World Congress of Structural and Multidisciplinary


Optimization (WCSMO-3), Bualo, NY, 1721 May 1999.
Cherkaev AV, Kucuk I. Optimal structures for various loading.
In: Proceedings of the Third ISSMO/UBCAD/UB/AIAA World
Congress of Structural and Multidisciplinary Optimization
(WCSMO-3), Bualo, NY, 1721 May 1999.
de Faria AR, Hansen JS. Buckling load optimization of composite
plates via min-max formulation. In: Proceedings of the Third
ISSMO/UBCAD/UB/AIAA World Congress of Structural and
Multidisciplinary Optimization (WCSMO-3), Bualo, NY, 1721
May 1999.
de Faria AR, Hansen JS. On buckling optimization under
uncertain loading combinations. Struct Multidisc Optim
2001;21:27282.
Adali S, Lene F, Duvaut G, Chiaruttini V. Optimization of
laminated composites under buckling load uncertainties via antioptimization. In: Proceedings of the 9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimization, Atlanta,
Georgia, USA, 46 September 2002. Available from: AIAA
website.
Nemeth MP. Importance of anisotropy on buckling of compression loaded symmetric composite plates. AIAA J 1986;24:18315.

You might also like