49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference <br>16t

7 - 10 April 2008, Schaumburg, IL

AIAA 2008-1918

Optimization of Anisotropic Laminated Composite

Plates Incorporating Non-Conventional Ply
Orientations
Mark W. Bloomfield1, J. Enrique Herencia2,
University of Bristol, Bristol, UK, BS8 1TR
Paul M. Weaver3
University of Bristol, Bristol, UK, BS8 1TR
In laminated composite design, ply orientations are generally restricted to
0, 90, 45 degrees. This is often due to manufacturing limitations. However, studies have shown that
an expanded set of ply orientations may lead to more efficient structures. Using a larger set of ply
orientations, including non-standard fibre orientations, imparts greater elastic tailoring capabilities
which may bring associated gains in efficiency or indeed functionality. In particular it has been
shown that 60 degree plies are optimum for shear in buckling for long anisotropic composite
laminated plates. In this paper a two-level optimization approach, using an expanded design envelope
of 0, 90, 45, 30 60 degree plies, is presented. At the first level, gradient based methods are used
to minimize the mass of a simply supported composite plate under different loading conditions. At
the second level, a particle swarm optimization algorithm is used to determine laminate stacking
sequences. Using an expanded design envelope of ply orientations, it is shown that mass savings of up
to 7.4% can be achieved for an unbalanced symmetric laminate. Specifically, mass savings are made
for geometries where the plate has a large aspect radio and is predominantly shear loaded.
Additionally, it is shown that the particle swarm optimization performs well and achieves feasible
designs which satisfy the design constraints.

Nomenclature
Aij
A
a
b
C
ci
Dij
E11, E22
f
G12
h
M
m
M i

=
=
=
=
=
=
=
=
=
=
=
=
=
=

components of the in-plane stiffness matrix

cross sectional area of the plate
plate length
plate width
compression
weighting factors
components of the out-of-plane stiffness matrix
longitudinal and transverse Young's moduli
fitness function
shear modulus
plate thickness
vector of out-of-plane moments
total number of plies in the lay-up
mass of plate using design envelope i

PhD Student, ACCIS, Department of Aerospace Engineering, Queens Building, University Walk, Student
Member
2
Marie Curie Research Assistant, ACCIS, Department of Aerospace Engineering, Queens Building, 1.69,
University Walk, Student Member
3
Reader, ACCIS, Department of Aerospace Engineering, Queens Building, 2.39, University Walk.
Member
1
American Institute of Aeronautics and Astronautics
Copyright 2008 by the author(s). Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

ms
ni
N
Ni
N icr
Pcr
Pc
p
Qij
RFi j
ri
Ui
s
T
vij (t )

=
=
=
=
=
=
=
=
=
=
=
=
=
=
=

mid-plane symmetric
number of each ply orientation in the lay-up
vector of in-plane loads
load per unit width in the i direction
critical load per unit width
critical buckling load of a column accounting for transverse shear loading
Euler buckling load
size of i
components of the reduced stiffness matrix
reserve factor for strength (i = s) and buckling (i = b) and j = T, C
random weighting factors
material invariants
symmetric
tension
velocity of particle i in dimension j at discrete time step t

xi (t )

position of particle i at discrete time step t

y i (t )
y i (t )
z

local best position of particle i discrete time step t

0
iA, D

=
=
=
=

global best position of particle i discrete time step t

normalized through thickness co-ordinate
vector of mid-plane strains
lamination parameters (i = 1...4)

A, D
iOpt

optimum lamination parameters (i = 1...4)

=
=

vector of plate curvatures

density of material

v12

Poissons ratio

=
=

ith design envelope of ply orientations

swarm inertia coefficient

I. Introduction

he excellent performance of composite materials has been well publicized in recent years. As such, the
aviation industry is rapidly employing composite materials for primary structures such as wings and
fuselages. Their use is clearly seen with the development of many recent commercial and military aircraft
including the Airbus A380, the Boeing 787 Dreamliner and the Airbus A400m. Despite their insertion in
high profile aircraft, there are potential efficiency gains to be made by undertaking stacking sequence (layup) optimization and reducing dependence on use of homogeneous properties, or black metal.
In laminated composite design, ply orientations are generally restricted to 0, 90, 45 degrees. This
is often due to manufacturing limitations. However, studies have shown that an expanded set of ply
orientations may lead to more efficient structures. Using a larger set of ply orientations, including nonstandard fibre orientations, imparts greater elastic tailoring capabilities which may bring associated gains in
efficiency or indeed functionality.1 In particular it has been shown that 60 degree plies are optimum for
shear in buckling for long anisotropic composite laminated plates2. Furthermore, it has been shown that 30
degree plies could be useful for aeroelastic tailoring3. Additionally, the optimization of composite
structures, using lamination parameters as intermediate design variables, has been restricted since the
feasible region of lamination parameters was known only for 0, 90, 45 degree plies. Recently, a method
was derived to determine the feasible region of lamination parameters for any discrete design envelope of
ply orientations4. This development allows the optimization of composite structures using any expanded
design envelope to include, for example, 30, 60 degree plies. Note, if the laminate is orthotropic, using
this particular expanded design envelope of ply orientations is particularly useful since it populates almost
the entire design space of lamination parameters5.
2
American Institute of Aeronautics and Astronautics

The optimization of laminated composites has evolved significantly over the past 25 years. For a
detailed analysis, see Refs. 6-7. Recent focus has been on the optimization of laminated composites using
lamination parameters.6-7 Tsai and Pagano8 & Tsai and Hahn9 characterized the stiffness properties of
composite laminates in terms of material invariants and at most 12 lamination parameters. Lamination
parameters significantly reduce the number of design variables when compared with using ply orientation
and thicknesses. Concerning optimization, using lamination parameters is particularly useful. For example,
Grenestedt and Gudmundson10 proved that the feasible region of lamination parameters is convex. The
convexity of the lamination parameters reduces the complexity of the optimization problem and also
reduces the CPU time. If ply orientation and thickness are used as design variables the resulting feasible
region is often a non-convex and complex response surface. Note, given the convex nature of the
optimization presented herein, a global optimum is guaranteed using gradient based methods.
Recently, a two level optimisation approach has been adopted by Herencia et al.6,7 to optimize
long anisotropic laminated fibre composites with T-shaped stiffeners. At the first level, gradient based
methods were used to determine a vector of optimal lamination parameters. All constraints such as strength,
buckling, practical design rules etc. were considered at this level. At the second level, a genetic algorithm
(GA) was used to determine feasible lay-ups from the optimal vector of lamination parameters. The authors
showed that elastic tailoring could be used to improve the structures performance. However, their study
restricted ply orientations to 0, 90, 45 degrees. Despite the effectiveness of a GA, it has been suggested
that a particle swarm optimization (PSO) algorithm may require less computational effort to achieve the
same high quality solution as a GA..11,12 PSO is a search technique developed in 1995 by Kennedy and
Eberhart13. The development of a PSO algorithm was inspired by the notion of birds flocking to a particular
point. It is useful to note that PSO has already been successfully used in composite structural
optimization.14,15 However, these studies did not use PSO as a discrete optimizer in the manner presented
in this paper.
The aim of this paper is to provide a two level approach, similar to Herencia et al.6,7 in which the
design envelope of ply orientations is expanded to 0, 90, 45, 30, 60 degrees. At the first level,
gradient based methods are used to find the optimum thickness and vector of lamination parameters, subject
to strength, buckling and lamination parameter constraints. The objective function, which is to be
minimized, is the mass of the composite plate. The plate is also considered to be symmetrically laminated
and thus exhibits no flexural-membrane coupling. At the second level, a discrete PSO algorithm is used to
find actual stacking sequences that match the vector of optimum lamination parameters. This work, to the
authors knowledge, represents the first published to use a discrete PSO to match stacking sequences to an
optimal vector of lamination parameters. Consequently, this algorithm represents a new approach to solving
the second level problem. Additionally, a new method is presented to approximate the number of each ply
orientation in the lay-up. The novelty of this work lies in the use of an expanded design envelope of ply
orientations, the use of a new method to approximate the number of each ply orientation and the use of a
discrete PSO to determine laminate stacking sequences.
II. Background
Tsai and Pagano8 & Tsai and Hahn9 characterized the stiffness properties of monolithic laminated
composites as linear functions of material invariants and at most 12 lamination parameters. Since the plate
is symmetric and thus does not exhibit any bending-extension coupling, the constitutive equation for a
symmetric laminated plate using classical laminate theory (e.g. Ref. 9) is given, by,

N A 0 0
=

M 0 D

(1)

where N is a vector of resultant running loads, M is a vector of resultant out-of-plane moments, 0 is the
vector of mid-plane strains and is the vector of plate curvatures. The in-plane and out-of-plane stiffness
matrices are defined in terms of lamination parameters and material invariants,

3
American Institute of Aeronautics and Astronautics

1 1A
A11
2A

A
2A
A22
1 1
A

0
2A
12 = h 0
A66
0
1A
0

A
A
A16
0 3 / 2 4
A
0 3A / 2 4A
26

0
0
1
0
0
0

1 1D
D11
2D

D
2D
D22
1 1
D
3
0
2D
12 = h 0
D66 12 0
0
1D

0 D / 2 D
3
4
D16

D
0 3D / 2 4D
26

0
U 1
0
U 2
0
U3
1
U 4
0
U 5
0

0
0
1
0
0
0

0
U 1
0
U 2
0
U3
1
U 4
0
U 5
0

(2)

(3)

where the lamination parameters are defined as,

1
2
3
=
2

[1A, 2, 3, 4] =
[D1, 2, 3, 4]

1
1

[cos 2 ( z ), cos 4 ( z ), sin 2 ( z ), sin 4 ( z ) ]dz

(4)

[cos 2 ( z ), cos 4 ( z ), sin 2 ( z ), sin 4 ( z ) ]z dz

2

and (z ) is the distribution function of the ply orientations through the normalized thickness co-ordinate

2z'

. For further details regarding ( z ) , see e.g. Ref. 16. Additionally, the material invariants are
h
defined as,

z=

U 1 = [Q11 + 2(Q12 + 2Q66 ) + Q22 ] / 4

U 2 = [Q11 Q22 ] / 2
U 3 = [Q11 2(Q12 + 2Q66 ) + Q22 ] / 4

(5)

U 4 = [Q11 + 2(Q12 2Q66 ) + Q22 ] / 4

U 5 = [Q11 + Q22 2Q12 ] / 4
where Qij are the reduced stiffnesses for unidirectional lamina and defined as,
2
2
2 2
Q11 = E11
/( E11
E 22
12 )
2
2
Q22 = E11 E12 /( E11
E 22 12
)

Q12 = 12 Q22

(6)

Q66 = G12
In Eq. (6), E11 , E 12 , G12 are the longitudinal, transverse, and shear moduli and 12 is the Poissons
ratio for a unidirectional lamina. Note, the eight lamination parameters defined in Eq. (4) are integrals
through the thickness of the sines and cosines of the lay-up orientations. In practice, this integration is
replaced by a through-the-thickness summation at ply level. Moreover, only these eight lamination
parameters are necessary to take into account the stiffness properties of any symmetric composite laminate.
4
American Institute of Aeronautics and Astronautics

III. Optimization Strategy

The optimization of the composite plate is divided into two levels. At the first level, gradient based
techniques are used to obtain the mass of the plate, plate thicknesses and the vector of optimum lamination
parameters. At the second level, a PSO algorithm is used as a discrete optimizer to obtain the actual
stacking sequence for the composite plate. The two level optimization strategy is outlined in Fig. 1,

Loads

Material
Properties

Geometry

1 Gradient Based
Optimization

1st Level Constraints

2 Discrete
Optimization
(PSO)

No
1st Level constraints
check
Yes
Final Design
Figure 1 Two Level Optimization Strategy

IV. Continuous Optimization

The gradient based optimization (first level) can be mathematically formulated as follows,
Minimize

M (x )

Subject to

G j (x) 0

Objective function
j = 1...G

xiu xi xil
where

x = x1, x2 ,K x n

Inequality constraints
Lower and Upper bounds

Design variables
(7)

5
American Institute of Aeronautics and Astronautics

The objective function, which is to be minimized, is the mass of the composite plate. The objective function
is defined as,

M ( x ) = aA

(8)

where a is the length of the plate, is the density of the material and A is the cross sectional area of the
plate. The design variables are the lamination parameters, (1A 2A 3A 4A 1D 2D 3D 4D ) and plate
thickness (t). The length and width of the composite plate are not considered as design variables. The
design constraints consist of the design variable bounds (lower and upper) and inequality constraints. These
constraints are,
1.
2.
3.

Feasible region of lamination parameters for 0, 90, 45, 30 60 and 0, 90, 45 degree plies
Failure strength at laminate level
Critical buckling load determined using closed form solutions.

The constraints on the feasible region of lamination parameters for 0, 90, 45, 30 60 degree
plies were recently derived in Ref 4. The constraints on the feasible region of lamination parameters for
angles restricted to 0, 90, 45 degrees and for a symmetric plate can be found, for example, in Ref. 6. The
optimization is undertaken for both design envelopes. Strains at laminate level are limited by an allowable
strain value. Equation (2) is used to calculate the laminate strains for the composite plate, where,

x0
A11
0y = A12

0
xy
A16

A12
A22
A26

A16
A26
A26

N
x
N y

N xy

(9)

The strength reserve factor is given by the ratio between the allowable and the applied strain from
Eq. (9) Hence,

RFi j =

aij
where i = x, y, xy and j = T, C
i0 j

(10)

The strength constraint applied to the plate is given by,

1
RFi j

1 0 where i = x, y, xy and j = T, C

(11)

Buckling constraints are in the form of closed form solutions (CFS).17-19 The plate is assumed to
be simply supported on all four edges and under normal and shear loads. Non-dimensionalized parameters20
are used to calculate buckling coefficients in order to determine the critical buckling loads. The nondimensionalized parameters are defined in terms of the out-of-plane stiffnesses as follows,

=4

D + 2 D66
D22
, = 12
, =
D11
D11 D22

D16
4 D3 D
11 22

D26

, =
4

3
D11 D22

(12)

In Ref. 17, the critical buckling load of an anisotropic plate which is simply supported on all four
edges and under normal loading was approximated using the following expression,

6
American Institute of Aeronautics and Astronautics

N xxr = K x

2
b2

D11 D12

(13)

where K x is a non-dimensional buckling coefficient given by,

K x = 2(1 + ) 2(3 + + 2 2 )

( + 3 ) 2
(3 + ) 2

4( + 2 3 )

( + 3 ) 3
(3 + ) 3

(14)

It is noted that sufficient accuracy for K x is obtained when , < 0.4 . If , > 0.4 an iterative
procedure is used to determine the value of K x . For further details see Ref. 17. The reserve factor for the
uniaxial compression loading is given by,

RF px =

N xcr
Nx

(15)

In Ref. 18 the shear buckling coefficient was defined in terms of the non-dimensional parameters,

K xy = 3.42 + 2.05 0.13 2 1.79 6.89 + 0.36 (2 + ) 0.25(2 + ) 2

(16)

The critical shear buckling load is calculated as,

cr
N xy
=

2
b

3
K xy 4 D11 D22

(17)

The reserve factor for the shear loading is thus,

RF pxy =

cr
N xy

(18)

N xy

Note, in the case of negative shear, the shear buckling coefficient is calculated assuming that the sign of
each ply angle is reversed. Lastly, the following formula19 is used to address the interaction between normal
and shear buckling,

1
1
1
=
+
RF pb RF px ( RFxy ) (1.9+0.1 )

(19)

The buckling constraint is therefore,

1 RF pb 0

(20)

V. Second Level Optimization

The goal of the second level of the optimization is to determine the laminate stacking sequence.
The stacking sequence is obtained using data from the first level using a PSO as a discrete optimizer. The
block structure of the PSO is shown in Fig. 2,

7
American Institute of Aeronautics and Astronautics

Generate population

Evaluation

Calculate the local and

global best positions

Calculate particle
velocities

New population

Stopping condition

Final design
Figure 2 - PSO Block Structure
The first step of the PSO algorithm is to generate an initial population. A new method is proposed
to generate the initial population. Firstly, the in-plane lamination parameters defined in Eq. (4) are rewritten
as a finite sum for a lay-up of m plies. Therefore,

1
(( z1 z 0 ) cos 21 + ( z 2 z1 ) cos 2 2 K + ( z m z m 1 ) cos 2 m )
2
1
2A = (( z1 z 0 ) cos 41 + ( z 2 z1 ) cos 4 2 K + ( z m z m 1 ) cos 4 m )
2
1
A
3 = (( z1 z 0 ) sin 21 + ( z 2 z1 ) sin 2 2 K + ( z m z m 1 ) sin 2 m )
2
1
A
4 = (( z1 z 0 ) sin 41 + ( z 2 z1 ) sin 4 2 K + ( z m z m 1 ) sin 4 m )
2

1A =

(21)

Assuming that each ply orientation has uniform ply thickness, it follows that z i z i 1 is constant.
Since the lay-up is formed from a finite design envelope , it follows that there will be
orientation, where i = 1...p. Therefore,

8
American Institute of Aeronautics and Astronautics

ni of each ply

cos 21
1A
cos 4
A
1

2 = sin 2
1
A m
3A
sin 41
4
1

cos 2 2 K cos 2 p 1
cos 4 2 K cos 4 p 1
sin 2 2

K sin 2 p 1

sin 4 2

K sin 4 p 1
K

cos 2 p n1

cos 4 p n2
sin 2 p M

sin 4 p n p
1 n p 1

(22)

where m is the total number of plies. Using this background information, a gradient based optimization is
constructed as follows.
p

M ( n) = m

Minimize

Objective Function

i =1

Subject to

cos 21
1A
cos 4
A
1

2 = 1 sin 2
1
A m
3A

sin
4
1

4
1

cos 2 2 K cos 2 p 1
cos 4 2 K cos 4 p 1
sin 2 2

K sin 2 p 1

sin 4 2

K sin 4 p 1

cos 2p n1

cos 4 p n2
sin 2 p M

sin 4 p n p
1 n p 1

n1
n
2
M 0

np
n p 1

Equality Constraints

Inequality Constraints

(23)
A gradient based optimization method is used in MATLAB21 to determine the vector n , where ni
are the design variables. Note, m is obtained at the first level of the optimization and is rounded up to the
nearest ply. This approach is conservative. If rounding down, critical constraints may not be satisfied. Note,
the obtained vector n is continuous although its sum is discrete. The ceiling function is used to give an
upper bound on the number of each ply in the layup, i.e. ni = ceil (ni ) . A population vector (pv) is then
formed from the vector n and the design envelope of ply orientations . For example, if

n = [0 2 0 2 0 1 3 3] and = [0 90 45 45 30 30 60 60] then there are zero 0 degree plies, two 90

degree plies, zero -45 degree plies and so forth. Therefore,
pv = [90 90 45 45 30 60 60 60 60]

(24)

The initial population is formed by randomly selecting vectors of size m / 2 (or ( m + 1) / 2 if midplane symmetric) from pv. The vector pv ensures that laminates are selected with a bias representative of
the vector n . Once the initial population is selected, ply orientations are encoded using Table 1,

9
American Institute of Aeronautics and Astronautics

Table 1- PSO Encoding

90

-45

45

-30

30

-60

60

Code

Next, the PSO algorithm is initiated. A PSO algorithm contains a swarm of particles (lay-ups),
where each particle represents a potential solution to the optimization problem. Particles move through a
multidimensional search space where the position of the particle is adjusted according to its own experience
and also the experience of the other particles in the swarm. Let xij (t ) denote the jth ply (from the outer
surface to mid-surface) in the ith lay-up at the tth iteration. Note t is a positive integer. The position of the
particle (lay-up) is updated by adding a velocity vector, vij (t ) to the current position, i.e.

xij (t + 1) = xij (t ) + vij (t + 1)t

(25)

where

vij (t + 1) =

vij (t ) + c1r1 (t ) ( y ij (t ) xij (t )) + c 2 r2 (t ) ( y j (t ) xij (t ))

t

(26)

where ci , are user-defined constants used to accelerate the swarm and control the 'inertia' of the swarm,
respectively and t = 1 . Functions r1 (t ), r2 (t ) are random numbers generated within the interval [0,
1], yij (t ) is the position (local best) of the jth ply in the ith lay-up in its history over all iterations:

yij (t 1)
y ij (t ) =
xij (t )

if

f ( xij (t )) f ( yij (t 1))

f ( xij (t )) < f ( y i (t 1))

(27)

with f, fitness function (to be minimized),

4

f ( ) =

A
A 2
iOpt i )

+( iOptD iD ) 2

(28)

i =1

and y j (t ) is the global best jth ply in any lay-up,

y j (t ) = min yij (t ) .

(29)

By updating the position vector with the velocity vector over a small time period (iteration ), Eq. (25), each
lay-up is either constrained to stay within its own neighborhood or is 'forced' to a different area of the
search space. Note, the velocity vector vij (t ) drives the optimization and thus the convergence of the
swarm to an optimal solution. The velocity vector vij (t ) also represents the change in ply angle of the jth ply
in the ith lay-up in dimension j = 1...m/2 (or (m+1)/2 if mid-plane symmetric) where m/2 is the dimension of
the search space (i.e. half the number of plies).
Note, ri (t ) introduces a stochastic element into the algorithm, to prevent premature convergence
to local optima whilst for small values of , the swarm conducts a localized search. On the other hand, if
the value of is large, the swarm has enough impetus to explore the wider search space. Additionally, to
10
American Institute of Aeronautics and Astronautics

prevent premature convergence, the algorithm contains a further stochastic element. At each time step, 20%
of the population (none of the members selected are the global best) are randomly selected from the full
design envelope of 0, 90, 45, 30 60 degree plies. For the optimization, is fixed at 0.8, ci = 1.5 11-12
and the population size is selected to be 20. Note, no parameter refinement has taken place. The selected
PSO parameters are taken from the literature and are deemed to give good convergence properties. The
stopping condition of the algorithm is if a good match is found between the target lamination parameters
and the lamination parameters of the best member and if this member has corresponding reserve factors for
strength and buckling greater or equal to 0.98. If the conditions imposed upon the reserve factors are not
met, the PSO is re-looped. In the next section, several numerical examples are given for various plate
geometries and loading conditions.
VI. Numerical Examples
In the following numerical examples, the mass of a simply supported composite plate is optimized
(minimized) subject to the aforementioned design constraints. The plate is also considered to be
symmetrically laminated. Additionally, the plate is optimized using two different design envelopes of ply
orientations, 0, 90, 45 and 0, 90, 45, 30 60 degree plies. Note, transverse shear effects are not
considered. For the numerical examples, the following load cases are considered,
i)
ii)
iii)
iv)
v)

Nx = 0, Nxy = -800
Nx = 0, Nxy = 400
Nx = -100, Nxy = 500
Nx = -300, Nxy = -300
Nx = -500, Nxy = 200

For each load case, three different large aspect ratios are considered. The length of the plate is fixed at
1200mm. However, the plate width is varied between 150 and 300mm, thus changing the aspect ratio of the
plate. It is noted that a) the loads are representative of the minimum and maximum compression and shear
values for a plate of the above dimensions and b) all aspect ratios are sufficiently large to allow use of the
long plate CFS (Eqs. 13-19). The following material properties were used:
E11 = 135000 N/mm2, E22 = 8500 N/mm2, G12 = 4200 N/mm2, =1.610-6 kg/mm3 and ply thickness was
0.25mm. Note, in the following examples ply contiguity constraints were not considered. Results are
produced in Tables 2-10 below. A discussion of results follows Table 10,
Table 2 - Masses for design Envelope of ply orientation 1 and 2 where the aspect ratio is 4

Load Case

M 1

M 2

Mass Saving (%)

M 1 / M 2

i)
ii)
iii)
iv)
v)

3.24
2.57
3.02
3.14
3.41

3.02
2.40
2.91
3.11
3.41

7.3
7.1
3.9
1
0

11
American Institute of Aeronautics and Astronautics

Table 3 - Lay-ups and reserve factors obtained using PSO for 2 and data in Table 2
Load
Case
i)

Lay-up

RF pb

RFscr

[60 5 / 45 / 60 / 90 / 45 / 45 / 45]ms

0.99

0.99

ii)

[60 6 / 90 / 30 / 45]ms

1.06

1.06

iii)

[60 2 / 45 / 60 3 / 45 / 60 / 90 / 45 / 30]ms

1.19

1.11

iv)

[45 / 60 / 45 / 60 2 / 45 2 / 45 / 45 / 30 / 45] s

1.07

1.01

v)

[45 / 45 / 45 / 45 2 / 45 / 60 / 45 / 45 / 45 / 30 / 45 / 30] s

1.04

0.98

Table 4 - Optimum and actual lamination parameters for each case in Table 3
Load
Case

Lamination Parameters
(1A 2A 3A 4A 1D 2D 3D 4D )

i)

Optimum
Actual

ii)

Optimum
Actual

iii)

Optimum
Actual

iv)

Optimum
Actual

v)

Optimum
Actual

( 0.371 0.600 0.429 0.544

( 0.381 0.524 0.447 0.495
( 0.297 0.384 0.562 0.701
( 0.412 0.353 0.568 0.713
( 0.246 0.587 0.105 0.574
( 0.357 0.500 0.168 0.536
( 0.049 0.785 0.026 0.372
( 0.090 0.818 0.067 0.315
(0.049 0.856 0.065 0.007
(0.042 0.875 0.133 0.072

0.479 0.503 0.837 0.850 )

0.473 0.503 0.829 0.778)

0.481 0.483 0.853 0.856)

0.5046 0.470 0.840 0.849 )
0.412 0.587 0.539 0.714)
0.400 0.589 0.464 0.675)
0.194 0.805 0.196 0.338)
0.210 0.784 0.179 0.374)
0.001 0.992 0.102 0.001)
0.021 0.968 0.083 0.036)

Table 5 - Masses for design Envelope of ply orientation 1 and 2 where the aspect ratio is 6

Load Case

M 1

M 2

Mass Saving (%)

M 1 / M 2

i)
ii)
iii)
iv)
v)

1.65
1.31
1.54
1.60
1.73

1.55
1.22
1.48
1.58
1.73

6.5
7.4
4.1
1.2
0

Table 6 - Lay-ups and reserve factors obtained using PSO for 2 and data in Table 5
Load
Case
i)

Lay-up

RF pb

RFscr

[60 4 / 45 2 / 60 / 45 2 ]ms

1.16

1.00

ii)

[60 2 / 45 / 60 / 90] ms

1.06

0.99

iii)

[60 / 45 / 60 3 / 45 / 30 / 60] s

1.15

1.27

iv)

[45 / 45 / 60 2 / 45 / 30 / 45 3 ] ms

1.10

1.04

v)

[45 / 45 / 45 / 45 2 / 30 / 30 / 60 / 30 2 ] ms

1.16

1.04

12
American Institute of Aeronautics and Astronautics

Table 7 - Optimum and actual lamination parameters for each case in Table 6
Load
Case

Lamination Parameters

(1A

i)

Optimum
Actual

ii)

Optimum
Actual

iii)

Optimum
Actual

iv)

Optimum
Actual

v)

Optimum
Actual

2A 3A 4A 1D 2D 3D 4D )

( 0.297 0.690 0.252 0.274

( 0.294 0.706 0.365 0.306
( 0.355 0.505 0.493 0.641
( 0.462 0.462 0.512 0.666
( 0.203 0.615 0.166 0.523
( 0.250 0.625 0.217 0.433
( 0.026 0.881 0.217 0.062
( 0.059 0.824 0.129 0.102
(0.106 0.745 0.099 0.048
(0.131 0.763 0.123 0.046

0.419 0.543 0.786 0.766)

0.436 0.564 0.834 0.720 )

0.482 0.489 0.847 0.855)

0.478 0.522 0.782 0.827 )
0.412 0.587 0.539 0.715)
0.344 0.643 0.381 0.616)
0.206 0.794 0.188 0.357 )
0.127 0.828 0.277 0.220 )
0.001 0.996 0.103 0.002)
0.039 0.947 0.054 0.012)

Table 8 - Masses for design Envelope of ply orientation 1 and 2 where the aspect ratio is 8

Load Case

M 1

M 2

Mass Saving (%)

M 1 / M 2

i)
ii)
iii)
iv)
v)

1.03
0.81
0.95
0.99
1.07

0.99
0.76
0.92
0.98
1.07

4
6.6
3.3
1
0

Table 9 - Lay-ups and reserve factors obtained using PSO for 2 and data in Table 8
Load Case

Lay-up

i)

[60 2 / 45 2 / 60 / 45 2 ]ms

1.08

RFscr
0.98

ii)

[60 4 / 45 2 ]ms

1.14

1.08

iii)

[60 3 / 45 / 30 / 45 / 90]ms

1.09

1.04

iv)

[45 / 45 / 60 / 45 / 30 / 45 / 30]ms

1.08

1.06

v)

[45 / 45 2 / 45 / 30 / 0 / 60 / 45]ms

0.99

1.03

RF pb

13
American Institute of Aeronautics and Astronautics

Table 10 - Optimum and actual lamination parameters for each case in Table 9
Load
Case

Lamination Parameters

(1A

i)

Optimum
Actual

ii)

Optimum
Actual

iii)

Optimum
Actual

iv)

Optimum
Actual

v)

Optimum
Actual

2A 3A 4A 1D 2D 3D 4D )

( 0.116 0.88 0.073 0.119

( 0.214 0.786 0.124 0.124
( 0.368 0.604 0.430 0.544
( 0.364 0.636 0.357 0.630
( 0.284 0.607 0.274 0.464
( 0.231 0.539 0.267 0.533
(0.062 0.792 0.184 0.149
(0.071 0.786 0.124 0.124
(0.151 0.648 0.186 0.074
(0.133 0.600 0.164 0.000

0.261 0.739 0.776 0.432 )

0.346 0.655 0.765 0.504)
0.479 0.503 0.837 0.849 )
0.490 0.501 0.820 0.849 )
0.412 0.588 0.539 0.713)
0.400 0.555 0.605 0.769)

0.207 0.791 0.188 0.362)

0.060 0.882 0.126 0.200)
0.003 0.989 0.104 0.007 )
0.058 0.906 0.091 0.049 )

Analyzing the data from Tables 2, 5 and 8, a trend in the mass savings becomes apparent. When
the plate is under a high shear load, significant mass savings are achieved by using the expanded design
envelope. Notably, for load case ii) in Table 5 a mass saving of 7.4% was achieved. It is further observed
that if the plate is under combined loading, a mass saving of approximately 3.5% is obtained if the loading
is dominated by shear. On the other hand, if the loading is dominated by normal loading, minimal mass
savings are observed. The optimization was also run assuming the laminate was balanced and symmetric.
In these cases it was observed that less mass savings were achieved. For a balanced symmetric laminate,
the maximum mass saving was 3.4% (for loading case i) and with a plate aspect ratio of four). Greater mass
savings could therefore be achieved when the laminate was an unbalanced symmetric laminate. This arises
3
because a balanced laminate ( 3A = 4A = 0 and 3D , 4D ) reduces the amount of in-plane and thus
4
out-of-plane anisotropy, which is gained through using an expanded design envelope of ply orientations.
Concerning the closeness of the match between lamination parameters from the first and second levels, it
can be seen from Tables 4, 7 and 10 that in most cases, a good match was achieved. The PSO algorithm
was able to find stacking sequences (Tables 3, 6 and 9) which closely match the optimum lamination
parameters (obtained at the first level) and which satisfy design constraints. Note the results presented in
Tables 2-10 were achieved using a basic PSO framework with no parameter refinement. Concerning
laminate thickness, it is observed that since no additional increase in the number of plies (after rounding)
occurred at the second level, the percentage saving between continuous and discrete levels is approximately
the same.

Conclusion
A two-level optimization approach with an expanded design envelope of ply orientations has been
presented. At the first level, a gradient based method was used to minimize the mass of an anisotropic
composite plate subject to strength, buckling and lamination parameter constraints. At this level, it was
shown that mass savings of up to 7.4% could be achieved using an expanded design envelope of ply
orientations of 0, 90, 45, 30, 60 compared to the traditional design envelope of 0, 90, 45 degrees.
Additionally, it was observed that the high mass savings occurred when the aspect ratio of the plate was
large and under shear loading. At the second level, a PSO was used to determine the laminate stacking
sequence, where a new method to estimate the number of each ply orientation was used to generate the
swarm population. The PSO performed well and was able to determine stacking sequences which closely
match the optimum lamination parameters (obtained at the first level) and which satisfy design constraints.
Further work includes PSO parameter refinement for improved convergence properties.

14
American Institute of Aeronautics and Astronautics

Acknowledgements
The first author would like to thank Great Western Research and Airbus UK for their ongoing
support. The second author would like to thank the European Commission (EC) for the Marie Curie
Excellence Grant MEXT-CT-2003-002690.

References
1
Weaver, P. M., "Anisotropic Elastic Tailoring in Laminated Composite Plate and Shells in Buckling and
Postbuckling Structures: Experimental, Analytical and Numerical Studies" Editors:Falzon and Aliabadi, 2007
2
Grenestedt, J. L. Layup optimization against buckling of shear panels. Struct. Opt., Vol. 3, pp.115-120,
1991
3
Lemanski, S. L. and Weaver P. M. Flap-Torsion Coupling in Sandwich beams and filled box-sections
Thin-Walled Structures 43(6), pp923-955 ,2005
4
Bloomfield, M.W., Diaconu, C.G., Weaver, P.M., "Pseudo-Tensorial Constraints for Laminated Composite
Structures", Submitted 2008
5
Bloomfield, M.W., Weaver, P.M, " Approximating the General Feasible Region of Lamination Parameters
Using Finite Design Envelopes", In Progress, 2008
6
Herencia, J.E., Weaver, P.M. and Friswell, M.I., "Local Optimization of Long Anisotropic Laminated Fibre
Composite Panels with T Shaped Stiffeners". AIAA Journal Vol.45, No.10, p.2497-2509, 2007.
7
Herencia, J.E., Weaver, P.M. and Friswell, M.I., "Local optimization of anisotropic composite panels with T
shaped stiffeners including transverse shear effects and out-of-plane loading". Struct. Mul. Opt., 2007,
DOI:10.2007/s00158-008-0227-6
8
Tsai, S. W., J. C. Halpin and N. J. Pegano. "Composite Materials Workshop".
Technomic Publishing
Co., Inc., Stamford, Conn. 06902. pp.223-253. 1968
9
Tsai, S. W. and Hahn, H. T. "Introduction to composite materials". Technomic Publishing Co., Inc., 750
Summer St., Stamford, CT. 06902, 1980
10
Grenestedt, J. L., and Gudmundson, P., Lay-up Optimization of Composite Material Structures,
Proceedings of IUTAM Symposium on Optimal Design with Advanced Materials, Elsevier Science, Amsterdam, pp.
311336, 1993
11
Engelbrecht,A.P., Fundamentals of Computational Swarm Intelligence (Hardcover), John Wiley and Sons
Ltd, 2005
12
Suresh, S., Sujit, P.B., Rao, A.K., "Particle swarm optimization approach for multi-objective composite
box-beam design", Composite Structures, Volume 81, 2007
13
Kennedy,J. and Eberhart, R. Particle swarm optimization, in Proc. of the IEEE Int. Conf. on
Neural
Networks, Piscataway, NJ, pp. 19421948, 1995
14
Omkar, S. N., Mudigere, D., Naik, G. N., and Gopalakrishnan, S., "Vector evaluated particle swarm
optimization (VEPSO) for multi-objective design optimization of composite structures". Computers and Strucures. 86,
1-2 (Jan. 2008)
15
Sekishiro, M., Todoroki, A., "Optimizations of stiffened composite panel using fractal branch and bound
method"46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, Apr 18-21
Austin, TX, United States

16

Diaconu, C. G., Sato, M., Sekine, H., "Feasible Region in General Design space of Lamination
Parameters for Laminated Composites", AIAA Journal, v 40, n 3, p 559-565
17

Weaver, P. M., Approximate Analysis for Buckling of Compression Loaded Long Rectangular Plates with
Flexural/Twist Anisotropy, Philosophical Transactions of the Royal Society of London, Series A: Mathematical and
Physical Sciences, Vol. 462, No. 2065, 2006, pp. 5974.
18
Weaver, P. M., On Optimization of Long Anisotropic Flat Plates Subject to Shear Buckling Loads, 45th
AIAA/ASME/ASCE/AHS/ACS Structures, Structural Dynamics and Materials Conference, Palm Springs, CA, AIAA
Paper 2004-2053, 2004.
19
Weaver, P.M., Nemeth, M.P., "Improved Design Formulae for Buckling of Orthotropic Plates under
Combined Loading",49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference,
Chicago, Illinois, 2008
20
Weaver, P.M., Nemeth, M.P., "Bounds on Flexural Properties and Buckling Response for Symmetrically
Laminated Plates", Accepted for Publication by Journal of Engineering Mechanics, 2007
21
MATLAB, Software Package, Ver. 7.2, The MathWorks, Inc., Natick, MA, 19942007.

15
American Institute of Aeronautics and Astronautics