369

DESIGN OF THE ELASTIC PROPERTIES OF LAMINATES

WITH A MINIMUM NUMBER OF PLIES
M. Montemurro,
1,2
A. Vincenti,
1
and P. Vannucci
3*
Keywords: anisotropy, mechanical properties, laminate mechanics, numerical analysis
The design of laminates with a minimum number of layers for obtaining the given elastic properties is addressed in
the paper. The problem is treated and solved in the general case no simplifying hypotheses are made about the type of their
stacking sequence. The problem is stated as a nonlinear programming problem, where a unique objective function includes all
the requirements to be satisfed by the solutions. The optimum solutions are found within the framework of the polar-genetic
approach, since the objective function is written in terms of polar parameters of the laminates, while a nonclassical genetic
algorithm is used as the optimization scheme. The optimization variables include the number of layers, layer orientations and
layer thicknesses. In order to include the number of plies among the design variables, certain modifcations of the genetic
algorithm have been done, and new genetic operators have been developed. Some examples and numerical results concerning
the design of laminates with a minimum number of layers for obtaining some prescribed elastic symmetries, bending-extension
uncoupling, and quasi-homogeneity are shown in the paper.
1. Introduction
The design of elastic properties is very important in many applications, e.g., for aircraft and space structures. Unlike
classical materials, composite laminates can be designed to obtain certain properties, such as bending-extension uncoupling,
in-plane and/or bending orthotropy, isotropy, and so on. This can be done mainly by a correct design of the stacking sequence
of laminate plies. The problem of tailoring a composite plate to realize a given elastic or hygrothermoelastic behavior has at-
tracted the attention of several researchers. A rather complete review of the state of the art of the problem, at least what concerns
the design with respect to stiffness, can be found in [1, 2]. The design of laminates considered as an optimization problem
is rather cumbersome and presents diffculties in its solution due to the high nonlinearity and nonconvexity of the objective
Mechanics of Composite Materials, Vol. 48, No. 4, September, 2012 (Russian Original Vol. 48, No. 4, July-August, 2012)
1
Institut dAlembert UMR7190 CNRS Universit Pierre et Marie Curie Paris 6, Case 162, 4, Place Jussieu, 75252 Paris
Cedex 05, France
2
Centre de Recherche Public Henri Tudor, av. 29, J.F. Kennedy, L-1855 Luxembourg-Kirchberg, Luxembourg
3
Universit de Versailles et St Quentin, 45 Avenue des Etats-Unis, 78035 Versailles, France
*
Corresponding author; e-mail: paolo.vannucci@uvsq.fr
0191-5665/12/4804-0369 2012 Springer Science+Business Media, Inc.
Russian translation published in Mekhanika Kompozitnykh Materialov, Vol. 48, No. 4, pp. 539-570 , July-August,
2012. Original article submitted July 8, 2011; revision submitted April 13, 2012.
370
function, which is caused by the fact that laminate properties depend on a combination of powers of the circular function of
layer orientations, the latter being natural design variables. As a consequence, designers generally limit the search for solu-
tions to a restricted class of laminates, usually to symmetric stacking sequences to ensure the bending-extension uncoupling,
or balanced sequences to have an in-plane orthotropy and so on.
The problem of designing laminates as an optimization problem received a general formulation, especially concern-
ing the design of elastic symmetries, with the works of Vannucci, Vincenti, and Verchery [3-5]. They have shown that it is
possible to construct, by using the so-called polar method, a unique objective function which is able to take into account
several design criteria, e.g., elastic properties, such as uncoupling, orthotropy, and many others, and given thermal responses
in extension and/or in bending. The general problem is therefore reduced to the classical nonlinear programming problem,
and its solutions are the minima of a nonlinear, nonconvex function in the design space of layer orientations. In these studies,
many optimum solutions were found for several different problems. In all the investigations made by these authors, the number
of plies was always fxed a priori, and the design process was focused only on the geometry of stacking sequence, i.e., the
only design variables were layer orientation angles.
As a natural continuation of [3-6], the topic of this paper is a new formulation for the laminate design with a minimum
number of layers which satisfes assigned elastic symmetries. For this purpose, the number and orientations of plies, as well
as the thickness of each layer, are taken into account as design variables. More precisely, this paper tries to give an answer to
a question which is usually left apart by designers, but which is classical and fundamental in any mathematical problem, i.e.,
the question about the existence of a solution. In the case of laminate design, this question should be: what is the minimum
number of layers that guarantees the existence of at least one solution to a given problem of tailoring the elastic properties of
a laminate?
To the knowledge of the authors, only in one case the minimum number of layers to obtain some prescribed properties
is known exactly thanks to a theoretical result. This is the case of in-plane isotropy, solved by Werren and Norris [7]: at least
three unidirectional plies are needed to obtain a laminate that will be isotropic in tension, although with be membrane-bending
coupling. But as soon as one adds another or a different requirement, for instance, uncoupling or bending isotropy, the result
is unknown. Finding the minimum number of layers for which a given optimum laminate design problem can be solved is
actually a very diffcult task. In fact, the number varies with the type of elastic requirements to be satisfed: the results are
strictly problem-dependent, and unfortunately in all the cases, the optimum solutions are unknown and there is no analytical
model which could describe their evolution with the number of layers. Therefore, a numerical investigation seems to be an
appropriate approach to the investigation.
It is worth noting that the optimum design of a laminate in terms of the number and properties (orientation, materials,
and thickness) of its layers is a combinatorial optimization problem, which is harder to solve for small numbers of layers. In
fact, the smaller the number of plies, the smaller becomes the design space, and the number of available solutions decreases.
However, solutions with a minimum number of plies are important when the problem of minimum weight of laminates is
addressed.
The way proposed in this paper is to formulate the problem in the form of search for the minimum of a positive
semidefnite form, including the number of layers n among the variables. The function takes into account the number n in the
form of a penalization term in order to strongly drive the search for an optimum solution towards laminates with the lowest
number of layers.
The numerical technique adopted here is a genetic algorithm (GA), BIANCA, which stands for Biologically Inspired
Analysis of Composite Assemblages, created by the authors and already used in the previous research [3, 8]. This code has
been specially conceived for the optimum design of laminates; here, a modifed version of this code, able to include the number
of layers into design variables, is used. In this way, the code BIANCA becomes a GA able to cross not only individuals, but
also species, see Sect. 5. In order to obtain an effective formulation of the problem, the polar formalism was used. It is based
upon an algebraic formulation making use of tensor invariants for representing planar tensors, see [9, 10], which has proven
to be rather effective in solving several design problems concerning laminates.
371
The paper is organized as follows. In the frst section, the general equations of the Classical Laminated Plate Theory
(CLPT) are recalled, and then the design problem is stated within the framework of the polar method and formulated as an
optimization problem. Further, an account of the numerical procedure, in particular, a concise description of the new genetic
algorithm for deriving solutions, is presented. Finally, several numerical examples are given in order to show the effectiveness
of the approach proposed, and then some general considerations ends the paper.
2. General Equations
The general equations of the Classical Laminated Plate Theory (CLPT) (see for instance [11]) describing the behavior
of a composite laminate, in the case of zero hygrothermal loads, are

N
M
A B
B D

`
)
=

`
)

, (1)
where
N =

N
N
N
x
y
s
, M =

M
M
M
x
y
s
, =

x
y
s
, =

x
y
s
.
In Eq. (1), the classical Voigt notation is assumed, and all the quantities included are expressed in the global reference
frame of the laminate, i.e., R = {O; x, y, z} (see Fig.1). N and M are the tensors of in-plane forces and bending moments, is
the tensor of in-plane strains for the midplane, and is the curvature tensor. The tensors A and D describe the in- and out-of-
plane behavior of the plate, but B is the coupling tensor. For example, if B is not equal to zero, the in-plane forces applied to
the laminate cause both in-plane strains and curvatures of the midplane and vice versa. The tensors A, B, and D depend on
the mechanical and geometrical properties of layers. For a multilayer laminate with n plies, the expression of A, B, and D are
A Q = ( ) ( )
=
k k
k p
p
k k
z z
1
,
B Q = ( )
( )
=

1
2
2
1
2
k k
k p
p
k k
z z , (2)
D Q = ( )
( )
=

1
3
3
1
3
k k
k p
p
k k
z z .
x
1
z = x
3k
y
x
2k
x
1k
d
k
Fig. 1. Laminate reference systems.
372
Here, Q
k
(
k
) is the reduced stiffness tensor of a kth ply oriented at an angle
k
with respect to the global reference
frame of the laminate, see Fig. 1; n is the number of plies and is defned as
n
p
p
=
+

2
2 1
if even
if odd
,
;

z
k
and z
k
1 are z coordinates of the top and bottom surfaces of the kth layer. Figure 2 illustrates the defnition of z
k
used in
this work. The homogenized stiffness tensors A*, B*, and D* of the laminate and its homogeneity tensor C are defned as
A A
*
=
1
h
tot
, B B
*
=
2
2
h
tot
, D D
*
=
12
3
h
tot
, C A D
* *
= ,
where h
tot
is the total thickness of the laminate. The inverse of Eq. (1) is

a b
b d
N
M
T

`
)
=

(
(

`
)
,
where
a A BD B
1
=
( )

1
, b aBD
1
=

, d D BA B
1
=
( )

1
. (3)
In Eqs. (3), a, b, and d are the in-plane, coupling, and bending compliance tensors.
3. Polar Representation Method
The polar representation method is a mathematical technique introduced in 1979 by Verchery [9]. By using this method,
it is possible to represent a plane tensor of any order by its invariants. This technique has already been used to treat many
optimum design problems for laminates. A complete overview of the polar method can be found in [12]. There are different
ways to represent a tensor, and the Cartesian representation is the most-used one. The main drawback of the Cartesian method
is the fact that the components are frame-dependent.
The idea of Verchery is to replace these components with other parameters that are frame-independent, i.e., tensor
invariants. There are several ways to choose these invariants. The quantities introduced by Verchery are directly linked to the
elastic symmetries and to the decomposition of strain energy [12].
In the case of a plane fourth-rank elasticity tensor L, such as the layer stiffness tensor Q or the layer compliance
tensor S = Q
1
, the polar method ensures that the Cartesian components can be expressed in terms of four moduli T
0
, T
1
, R
0
,
and R
1
and two polar angles
0
and
1
. The components of the tensor L in the material frame {O; x
1
, x
2
, x
3
} are expressed as
x
-p
a
z
h/2
h/2
z
k
z
k-1
p
-k
k
-1
1
0
b
x
z
h/2
h/2
z
k
z
k-1
-p
p
-k
k
-1
1
Fig. 2. Sketch of laminate layers and interfaces.
373

L T T R R
L R R
L
1111 0 1 0 0 1 1
1112 0 0 1 1
2 4 4 2
4 2 2
= + + +
= +
cos cos ,
sin sin ,


11122 0 1 0 0
1212 0 0 0
1222 0 0
2 4
4
4 2
= +
=
= +
T T R
L T R
L R
cos ,
cos ,
sin

RR
L T T R R
1 1
2222 0 1 0 0 1 1
2
2 4 4 2
sin ,
cos cos .

= + +
(4)
The norm of the tensor L as a function of polar parameters is
L = + + + T T R R
0
2
1
2
0
2
1
2
2 4 . (5)
The inverse relations of Eqs. (4) are

8 2 4
8 2
8
0 1111 1122 1212 2222
1 1111 1122 2222
0
4
T L L L L
T L L L
R
= + +
= + +
,
,
e
ii
i
L iL L L iL L
R L

0
1
1111 1112 1122 1212 1222 2222
1
4
1
4 2 4 4
8
= + +
=
,
e
1111 1112 1222 2222
2 2 + + iL iL L .

The polar moduli T
0
, T
1
, R
0
, and R
1
and the angular difference
0

1
are invariants of the tensor L. In axes rotated
through an angle , the components of the tensor L are expressed as

L T T R R
L R
xxxx
xxxy
= + + +
=
0 1 0 0 1 1
0 0
2 4 4 2
4
cos ( ) cos ( ),
sin ( )

++
= +
=
2 2
2 4
4
1 1
0 1 0 0
0 0
R
L T T R
L T R
xxyy
xyxy
sin ( ),
cos ( ),
cos

(( ),
sin ( ) sin ( ),
c

0
0 0 1 1
0 1 0
4 2 2
2

= +
= + +

L R R
L T T R
xyyy
yyyy
oos ( ) cos ( ). 4 4 2
0 1 1
R
(6)
These relations show that the rotation changes the polar angles
0
and
1
into
0
and
1
and that T
0
and
T
1
describe the isotropic part of the tensor, while R
0
and R
1
the anisotropic one. These are two reasons that make the polar
method very effective when it is employed to represent the layer stiffness and compliance tensors within the framework of
the CLPT theory, especially in design problems. In particular, the conditions for elastic symmetries are expressed in a simple
way when using polar invariants. A summary of these conditions is given in Table 1. We recall here that, in the plane case, the
square symmetry corresponds to the cubic syngony, i.e., it is characterized by the /2-periodicity of the elastic moduli, while
the R
0
-orthotropy is a special case of plane orthotropy, see [13] for more details.
TABLE 1. Conditions for Elastic Symmetries in Terms of Polar Invariants
Elastic symmetry Polar condition
Orthotropy

0 1
4
= K

R
0
- orthotropy
R
0
0 =
Square symmetry
R
1
0 =
Isotropy
R R
0 1
0 = =
374
In this paper, T
0
, T
1
, R
0
, R
1
,
0
, and
1
will denote the polar components of the reduced stiffness tensor of the basic
layer, Q. It is also possible to deduce the polar components of the laminate stiffness tensors A, B, and D as functions of polar
components of the lamina stiffness tensor Q according to Eq. (2) of CLPT:

T T T
m
T z z
T T T
m
T
A B D
k
k p
p
k
m
k
m
A B D
k
k p
p
0 0 0 0 1
1 1 1 1
1
1
, , ,
, ,
=
( )
=
=

zz z
R R R
m
R
k
m
k
m
A i B i D i
k
i
A B D
k k

( )
=

+
1
0
4
0
4
0
4
0
4
0 0 0 0
1
,
, ,
(
e e e e
))
,
, ,
k p
p
k
m
k
m
A i B i D i
k
i
z z
R R R
m
R
A B D
=


( )
=
1
1
4
1
4
1
4
1
2
1 1 1
1
e e e e
(( )
.

1
1
k k
k p
p
k
m
k
m
z z
+
=


( )

(7)
where T
A
0
, T
A
1
, R
A
0
, R
A
1
, F
0
A
, and F
1
A
stand for the polar components of the tensor A; T
B
0
, T
B
1
, R
B
0
, R
B
1
, F
0
B
, and F
1
B

stand for the polar components of the tensor B; T
D
0
, T
D
1
, R
D
0
, R
D
1
, F
0
D
and F
1
D
stand for the polar components of the ten-
sor D. In Eq. (7), m = 1, 2, 3 for the extension, coupling, and bending stiffness tensors, respectively.
From Eq. (7), it can be noticed that the symmetries of the laminate in terms of extension, coupling, or stiffness behavior
depend on its stacking sequence, layer properties, layer orientation and thickness, and of course on the number of plies. When
concerned with laminate design, a designer has to satisfy several conditions at the same time, including not only the com-
mon objectives, such as the buckling load or strength, but also the general properties of the elastic response of the laminate,
such as uncoupling, extension orthotropy, bending orthotropy, and so on. In fact, it is not easy to take into account all these
aspects, and normally designers use some shortcuts (rules of thumb) to automatically get some properties, e.g., uncoupling or
the extension orthotropy. Vannucci and Vincenti showed in the previous studies (see [4-6]) that, within the framework of the
polar method, it is possible to formulate all problems of the optimum design of laminates, including the requirements on the
elastic symmetries; therefore, a general approach to the design of laminates is possible. The reader is referred to the works
cited previously for a deeper insight into the matter. The next section presents an important modifcation to this approach that
also includes the number of layers among the design variables.
4. Formulation of the Optimization Problem
In formulating the design of the elastic properties of a laminate as an optimization problem, the key point is the con-
struction of the objective function. For a laminate with n plies, the design variables can be the number of layers n, the vector
of layer orientation , and the vector of layer thickness h. In order to formulate the laminate design problem in a general way,
the objective function f = f (n, , h) should include all design requirements, in particular, the elastic symmetries of the laminate.
Vannucci and Vincenti [3-6] showed that, within the framework of the polar method, the laminate design for given
elastic properties can be reduced to the search for the minima of a positive semidefnite function of polar elastic parameters of
the laminate. In those works, the number of layers, layer thickness, and material properties were fxed, while layer orientations
were assumed as the only optimization variables. The optimization problem was defned as
min

f ( )

.
Since the objective function f() is positive semidefnite, as mentioned above, its minima are also zeros of the func-
tion. For more details on the defnition of this objective function for different combinations of elastic symmetries, see [3-6].
375
As specifed previously, the objective of this paper is the design of a laminate with assigned symmetries at a minimum
number of layers. In such a case, the number of plies, and eventually the thickness of each layer, must belong to the set of
design variables, i. e., a modifcation of the objective function is necessary. The new, unconstrained optimization problem is

min , , ,
, , , , .
, , n
s
F n
F n f n n

h
h
h h
( )

( ) = ( )
(8)
It can be noticed that the new objective function F(n, , h) is defned in such a way that it is a positive semidefnite
function whose zeros are solutions of our problem. The infuence of the number of layers n is introduced as a penalization
term, where s is a power whose value can be chosen from a certain range. The large number of numerical tests that the authors
have been conducted show that the best results are obtained when s is in the interval [1, 4].
Equation (8) formalizes the classical nonlinear programming problem without constraints, for which several numeri-
cal solving techniques can be used. Since f (n, , h) is a nonconvex function with several nonglobal minima, a suitable and
robust solving algorithm must be chosen.
In conclusion of this section, it can be noticed that the approach proposed is general, i.e., no simplifying assumptions
are introduced, such as, for example, restrictions to symmetric, balanced, or angle-ply laminates.
5. Search for Solutions by the Genetic Algorithm BIANCA
Some general considerations have determined the choice of a GA as the numerical technique for solving the optimization
problem expressed by Eq. (8). The key point is the nature of the objective function, which is highly nonlinear and nonconvex.
Previous works, see for instance [14,15], showed that, due to the nonlinearity of the problem, in many cases, there are several
sets containing an infnite number of solutions that describe geometrical sets as continuous hypersurfaces in the design space.
Another very important aspect is the nonconvexity of the problem. In fact, in problems concerning the optimum design of
laminates, as a rule, there are many local and global minima, and the GA is able to look for a global minimum exploring the
entire research space in an effective way.
Another important point is that, in the case of laminates, problems often depend on continuous, discrete, or abstract
variables. For example, layer orientations can be considered as continuous or discrete variables the angles can take all
values in the range [90, 90] or they can be restricted to a limited and discrete set of fxed orientations, such as 0/45/90.
The same considerations can also be applied to the layer thickness. Finally, an abstract variable can represent a group of dif-
ferent variables, such as in the case of the constitutive material of layers when the material is chosen from a database: once a
particular material is associated with a ply, the entire set of properties of the layer are determined. Therefore, a suitable numeri-
cal technique should offer the same effectiveness for different types of design variables; GAs can deal with all these aspects
because they are basically the zeroth-order numerical strategies, which require only computations of the objective function.
The GA is a metaheuristic method for solving optimization problems, which is based on two biologic metaphors the
natural evolution, according to the Darwinian principle of survival of the fttest, and the genetic coding of the phenotype. A
wide and detailed presentation of genetic algorithms can be found in the books written by Goldberg [16] and Michalewicz [17].
Vannucci and Vincenti have developed a GA, BIANCA [8, 18], specially intended for solving many optimization
problems for laminates. BIANCA has several original features, mainly concerning the representation of information, i.e.,
coding of the genotype of an individual. This program is applicable to all classical genetic operators, such as adaptation and
selection, crossover, mutation and elitism operators. In addition, specifc strategies were developed within BIANCA in order
to take into account constrained optimization problems. We give here only some elements of the code BIANCA, more details
can be found in [8, 18].
Unlike most of the GAs dedicated to laminate problems, where generally a unique string of real variables composes a
sort of surrogate of the genetic code of the laminate, in BIANCA, the information concerning the laminate has an articulated
structure: a laminate is considered as an individual phenotype whose genome genotype is stocked in a binary coded matrix.
376
Each row of the matrix is a chromosome coding all data concerning a layer. In BIANCA, some chromosome corresponds to
each layer. In this way, the number of layers determines, in the biological metaphor, a species (in the sense that laminates with
a different number of layers belong to different biological species). Each chromosome is divided into genes; a gene codes a
particular quantity concerning the chromosome, i.e. the layer, such as, e.g., the orientation angle, material, thickness, etc. The
genes corresponding to various chromosomes form columns of the genome matrix. It is possible to code real and discrete
variables in the same genome at the same time. As programmed in BIANCA, the classical genetic crossover and mutation
operators mimic Nature very closely they act on each gene separately and independently.
5.1. The modifed genetic algorithm BIANCA
In order to include the design with respect to the number of layers in the objective function, we introduced some
modifcations in the code. As said previously, in BIANCA, the number of layers n determines the number of chromosomes,
i.e., the biological species of layers. So, if n has to be included among the variables and has to evolve during generations, a
crossover of the species has to be performed: the new version of BIANCA is a GA based upon the evolution of species and
individuals at the same time. The modifcations of the code BIANCA are substantially new genetic operators. We were partially
inspired by the previous work of Park et al. [19]. In our case, their strategy was modifed in order to adapt it to a scheme of
genetic coding and genetic operations much more complicate and rich than those proposed in [19].
In particular, the classical reproduction phase, see Fig. 3, was changed by introducing new genetic operators called
chromosome shift operator, chromosome reorder, chromosome number mutation, and chromosome addition-deletion.
A brief description of these new operators and their use in the reproduction phase is given in the next paragraphs.
5.1.1. Structure of the individual in BIANCA. The frst modifcation concerns the structure of the generic individual
genotype. As briefy mentioned previously, in BIANCA, the genotype of each individual is represented by the binary array
shown in Fig. 4. In this fgure, the quantity g
ij
k
represents a jth gene of an ith chromosome of a kth individual. The letter e
stands for an empty location, i.e., there is no gene in this location, while n is the number of chromosomes of the kth individual.
In this paper, the number of layers and their orientation angles and thickness are assumed as optimization variables.
The new structure of the individual-laminate is shown in Fig. 5. In this case, the number of layers in the kth laminate is n, but
the orientation and thickness of the ith ply are
i
and h
i
, respectively.
No
Yes
INITIAL
POPULATION
Random extraction
of individuals N
ADAPTATION
Fitness evaluation
of the individuals
SELECTION
Making parents
couples
CROSSOVER
Mixing of parents
genotype
REPRODUCTION
BEST
INDIVIDUAL
NEW
GENERATION
STOP
CRITERION
MUTATION
Stochastic alteration
of individuals
genotype
Final populations
average adaptation
Fig. 3. Standard GA scheme.
377
The classical reproduction operators, such as crossover and mutation, act on the genotype of parents couples. However,
to take into account the variable number of layers, i.e., the number of chromosomes, a new approach is required in executing
the reproduction phase. This new procedure is detailed below.
As mentioned previously, this new version of BIANCA is able to evolve species and individuals at the same time:
the best value of n is the outcome of biological selection, and the most adapted species issues automatically as a natural
result of the Darwinian selection.
5.1.2. The modifed crossover phase and the role of chromosome shift and reorder operators. To explain the way
whereby the reproduction phase takes place one can consider the following case. There are two parents, P1 and P2, with three
and fve layers, respectively, see Fig. 6. In this example, for the sake of synthesis, the maximum number of plies is assumed
equal to six. The minimum number can be chosen arbitrary between 1 and 6. Before realizing the crossover between these two
individuals, it can be noticed that there are different ways to pass the information restrained in parents genotype to the next
generation, i.e., to their children. At the successive generation, two new individuals will be produced from this couple, one
with three and the other with fve layers. To improve the effciency of the genetic algorithm in terms of exploration and exploi-
tation of the information on the design space, the concept of shift factor is introduced. The shift factor is sorted randomly,
with a given probability p
shift
, in the range [0, n
P1
n
P2
], where n
P1
n
P2
is the absolute value of difference between the
numbers of parents layers. With the shift factor, various combination of crossover are possible. In the example mentioned
before, the minimum shift factor is 0 and the maximum is 2. For example, if the shift factor is 1, all the genes of P1 with a
smaller number of chromosomes are shifted up to down by a quantity equal to 1, as shown in Fig. 7.
( ) g
k
11
( ) g
k
12
( ) g
m
k
1
n
( ) g
k
21
( ) g
k
22
( ) g
m
k
2


( ) g
n
k
1
( ) g
n
k
2

( ) g
nm
k
e e e e
Fig. 4. Structure of an individual genotype in BIANCA.
( )
1
k
( ) h
k
1
n
( )
2
k
( ) h
k
2


( )
n
k
( ) h
n
k
e e
Fig. 5. Individual-laminate in BIANCA.
378
After the shift operation, the real crossover phase takes place. The crossover operator acts on every single gene. The
position of crossover is chosen randomly for each gene of both individuals. This operator naturally involves all the chromo-
somes of the parent with a smaller number of plies, i.e., in the case shown in Fig. 8, all genes of P1, while only the homologous
genes of P2 undergo the action of crossover operator. At this point, two new individuals can be created, C1 and C2, that have
three and fve chromosomes-layers respectively, see Fig. 9. It can be noticed that the 1st and 5th chromosome of P2 have not
e e 3 ( )
1
2 P
( ) h
P
1
2
5
( )
1
1 P
( ) h
P
1
1
( )
2
2 P
( ) h
P
2
2
( )
2
1 P
( ) h
P
2
1
( )
3
2 P
( ) h
P
3
2
( )
3
1 P
( ) h
P
3
1
( )
4
2 P
( ) h
P
4
2
e e ( )
5
2 P
( ) h
P
5
2
e e e e
Fig. 6. Structure of parents couple.
e e 3 ( )
1
2 P
( ) h
P
1
2
5
( )
1
1 P
( ) h
P
1
1
( )
2
2 P
( ) h
P
2
2
( )
2
1 P
( ) h
P
2
1
( )
3
2 P
( ) h
P
3
2
( )
3
1 P
( ) h
P
3
1
( )
4
2 P
( ) h
P
4
2
e e ( )
5
2 P
( ) h
P
5
2
e e e e
Fig. 7. Effect of the shift operator on parents couple.
e e 3 ( )
1
2 P
( ) h
P
1
2
5
( )
1
1 P
( ) h
P
1
1
( )
2
2 P
( ) h
P
2
2
( )
2
1 P
( ) h
P
2
1
( )
3
2 P
( ) h
P
3
2
( )
3
1 P
( ) h
P
3
1
( )
4
2 P
( ) h
P
4
2
e e ( )
5
2 P
( ) h
P
5
2
e e e e
Fig. 8. Position of crossover for every gene.
379
undergone the crossover phase, so, according to the notation of Figs. 8 and 9, it is possible to write the equalities
1
2
1
2
( ) = ( )
P C
,

5
2
5
2
( ) = ( )
P C
, h h
P C
1
2
1
2
( ) = ( ) , and h h
P C
5
2
5
2
( ) = ( ) . At this point, before the mutation phase, a readjustment of chromo-
somes position is required. The chromosome reorder operator achieves this phase by a translation of all chromosome down
to up in the structure of the individual with the smaller number of layers, see Fig. 10.
5.1.3. The modifed mutation phase and the role of chromosomes number mutation and addition-deletion operators.
Mutation is a random process that is applied to the genotype to better explore the feasibility domain. Mutation is articulated
in two phases: it acts frst on the number of chromosomes-layers and then on the genes value.
In the frst phase, the number of layers is changed arbitrarily by one at time for each individual with a given probabil-
ity p
mut
chrom
( ) ; then the chromosome addition-deletion operator acts on the genotype of both individuals by adding or delet-
ing a chromosome-layer. The location of layer addition-deletion is also selected randomly. Naturally, if the number of layers
is equal to the maximum number of chromosomes, only deletion occurs. Similarly, if the number of plies is equal to the
minimum number, only addition is applied. In the case shown in Fig. 11, the number of layers of C1 is increased by one, and
a new chromosome
a
C
a
C
h ( ) ( )
{ }
1 1
, is randomly added in position 3, while the number of layers of C2 is decreased by one
and the chromosome deletion is randomly done in position 2.
e e 3 ( )
1
2 C
( ) h
C
1
2
5
( )
1
1 C
( ) h
C
1
1
( )
2
2 C
( ) h
C
2
2
( )
2
1 C
( ) h
C
2
1
( )
3
2 C
( ) h
C
3
2
( )
3
1 C
( ) h
C
3
1
( )
4
2 C
( ) h
C
4
2
e e ( )
5
2 C
( ) h
C
5
2
e e e e
Fig. 9. Structure of new individuals after a crossover.
( )
1
1 C
( ) h
C
1
1
3 ( )
1
2 C
( ) h
C
1
2
5
( )
2
1 C
( ) h
C
2
1
( )
2
2 C
( ) h
C
2
2
( )
3
1 C
( ) h
C
3
1
( )
3
2 C
( ) h
C
3
2
e e ( )
4
2 C
( ) h
C
4
2
e e ( )
5
2 C
( ) h
C
5
2
e e e e
Fig. 10. Effect of the chromosome reorder operator on the new individuals.
380
In the second phase, the mutation of gene values, i.e., orientations and thickness, takes place with a probability p
mut
after the rearrangement of chromosome positions. In the example of Fig. 12, the mutation is undergone by the genes
2
1
( )
C

and h
a
C
( )
1
of the individual C1 and by the gene
1
2
( )
C
of the individual C2.
6. Sample Problems and Results
In order to demonstrate the effectiveness of the polar formulation and of the new genetic algorithm BIANCA in obtain-
ing composite laminates with a variable number of plies and with certain elastic properties, several calculations were carried
out, and a number of solutions that satisfy different combinations of design objectives was found. Among all possible design
cases, the following ones are discussed in this paper:
1) uncoupling, total orthotropy with K = 0 and coincidence of axes, i.e., the in-plane and bending orthotropy with the
same axes;
2) uncoupling, total orthotropy with K=1 and coincidence of axes, i.e., the in-plane and bending orthotropy with the
same axes;
3) uncoupling and total isotropy , i.e., the in-plane and bending isotropy;
4) uncoupling and homogeneity, i.e., an identical behavior of the homogenized in-plane and bending stiffness tensors.
The corresponding polar conditions for the elastic symmetries are summarized in Table 1. Here, uncoupling is intended
to be the bending-extension uncoupling determined by the fact that the tensor B is zero.
( )
1
1 C
( ) h
C
1
1
4
( )
1
2 C
( ) h
C
1
2
4
( )
2
1 C
( ) h
C
2
1
e e
( )
a
C1
( ) h
a
C1
( )
3
2 C
( ) h
C
3
2
( )
3
1 C
( ) h
C
3
1
( )
4
2 C
( ) h
C
4
2
e e
( )
5
2 C
( ) h
C
5
2
e e e e
Fig. 11. Effect of the chromosome mutation and addition-deletion operators on the new individuals.
( )
1
1 C
( ) h
C
1
1
4 ( )
1
2
m
C
( ) h
C
1
2
4
( )
2
1
m
C
( ) h
C
2
1
( )
3
2 C
( ) h
C
3
2
( )
a
C1
( ) h
am
C1
( )
4
2 C
( ) h
C
4
2
( )
3
1 C
( ) h
C
3
1
( )
5
2 C
( ) h
C
5
2
e e e e
e e e e
Fig. 12. Effect of mutation operators on the new individuals.
381
6.1. Sample problems
Before formulating the objective function F (n, , h) for each case, it is worth specifying its general properties. In all
the problems considered in the present section (cases 1 to 4), this function is a dimensionless, homogenized, convex function
of polar parameters of the tensors A, B and D, Eq. (7), while it is a highly nonlinear, nonconvex function of design variables,
i.e., of the number of layers, orientations, and thickness. In all the cases, the power s in Eq. (8) is assumed equal to 2.
The objective function for each problem is defned in such a way that the solutions, i.e., the minima, are also zeros
of the function. Since each case corresponds to a given combination of elastic symmetries, the global objective function is
constructed as the sum of partial objective functions, and the values of each partial objective function are normalized between
zero and unity.
As a conclusive remark, before expressing the objective function for each particular case, we should note that, when-
ever a requirement on polar angles occurs in the expression of the objective function, these quantities are expressed in radians.
Cases n. 1 and n. 2. In order to obtain the elastic uncoupling, i.e., B = 0, the norm of the coupling tensor B must be
zero. To obtain orthotropy, the difference between the polar angles
0
and
1
must be a multiple of /4 (see the corresponding
polar condition in Table 1), both for the membrane and bending stiffness tensors. The last required elastic property considered
here is the coincidence of orthotropy axes, i.e., the angle
1
must be the same for A and D. The expression of the global objec-
tive function including all these conditions is
F n
K
A A A D
, ,
*
* * *
h
B
Q
( ) =
|
\

|
.
|
|
|
+

|
\

|
.
|
|
|
|
+
2
0 1
2
0
4
4

** * *
* *

|
\

|
.
|
|
|
|
+

|
\

|
.
|
|
|
|

1
2
1 1
2
4
4 4
D D
A D K

(
(
(
(
n
2
. (9)
In Eq. (9), B
*
is the homogenized coupling tensor, and Q is one of layer stiffness tensors. Their norms are calculated
according to Eq. (5). All other polar parameters are referred to the norms of their respective homogenized tensors, i.e., A
*
and
D
*
. The constants K
A
*
and K
D
*
can assume the values 0 or 1, depending upon the different kinds of orthotropy (case 1:
K K
A D
* *
= = 0 ; case 2: K K
A D
* *
= =1 ).
Case n. 3. In this case, along with the elastic uncoupling, the requirement of total isotropy was formalized. The partial
objective function for uncoupling is expressed as in cases 1 and 2, see Eq. (9). In order to ensure the isotropy of in-plane and
bending stiffnesses, the anisotropic part of the tensors A
*
and D
*
must be zero. Therefore, the global objective function has
the form
F n
R R
R R
R R
A A
Q Q
D D
, ,
* * * * *
h
B
Q
( ) =
|
\

|
.
|
|
|
+
+
+
+
+
2
0
2
1
2
0
2
1
2
0
2
1
4
4
4
22
0
2
1
2
2
4 R R
n
Q Q
+

(
(
(
. (10)
In Eq. (10), R
A
0
*
, R
A
1
*
and R
D
0
*
, R
D
1
*
are referred to the homogenized laminate in-plane and bending stiffness ten-
sors, respectively; R
Q
0
and R
Q
1
are referred to the layer stiffness tensor, and they are used for the sake of normalization.
Case n. 4. In this case, the requirements are uncoupling and homogeneity, i.e., the laminate has the same behavior in
extension and bending. To realize the objective of homogeneity, the polar parameters T
0
, T
1
, R
0
, R
1
,
0
, and
1
must assume
the same values for both the tensors A
*
and D
*
, and therefore the homogeneity tensor is equal to zero, C = 0. The objective
function is
F n n , , .
*
h
B
Q
C
Q
( ) =
|
\

|
.
|
|
|
+
|
\

|
.
|
|

(
(
(
2
2
2

382
6.2. Numerical results
Since the elastic behavior of a laminate depends on the elastic properties of its elementary plies, the results must
refer to a given material; in this paper, a highly anisotropic unidirectional carbon/epoxy ply (T300/5208) [20] was chosen. Its
properties were as follows: E
1
= 181 GPa, E
2
= 10.3 GPa, G
12
= 7.17 GPa, n
12
= 0.28, T
0
= 26.8804311 GPa, T
1
= 24.7448933
GPa, R
0
= 19.71043 GPa, R
1
= 21.433122 GPa, and F
0
= F
1
= 0 deg.
For each of the cases from 1 to 4 mentioned, two kinds of simulations were performed. In the frst one, the thickness
of the elementary ply was assumed equal to 0.125 mm, and thus the design variables were only the number and orientations
of layers. In the second case, the thickness also was a variable of optimization.
We also considered laminates with layers of variable thickness in order to assess the infuence of such a variable on the
minimum number of layers needed to obtain some specifed elastic properties. In practice, this corresponds to increasing the
number of design variables for the same kind of problem and should result in a better quality of results relative to those in the
cases of a fxed thickness and perhaps in a lower fnal minimum number of layers. Actually, it was so, which can be observed
in the results shown below. Of course, in doing this, we do not consider the practical (e.g., manufacturing) aspects of such a
choice, because we were concerned merely with the theoretical solution of the mathematical problem of fnding a laminate
having the minimum number of layers to satisfy some elastic requirements; the practical aspects are not discussed in this paper,
at least for the cases of laminates with layers of different thickness. We considered only layers with the same elastic proper-
ties, but of different thickness, which implies the assumption that the volume fraction and arrangement of fbers are constant.
For each simulation, the number of plies n varied in the range [4, 16], while the orientation of each layer
k
(k = 1,..,n)
could assume any integer value in the interval [90, 90] discretized by a step of 1. In simulations where the layer thickness
was also a design variable, the thickness h
k
(k = 1,..,n) varied in a continuous way in the interval [0.1, 0.2] mm. The genetic
parameters common for all the cases were as follows: N
ind
= 500, N
gen
= 500, p
cross
= 0.85, p
mut
= 1/N
ind
, p
shift
= 0.5, and
( )
( ) ( )
.
max min
p
n n
N
mut chrom
chrom chrom
ind
=

A remark regarding the mutation probability p
mut
and the mutation probability of the
TABLE 2. The Best Stacking Sequences for Design Problems 1 to 4
Objective Stacking sequence (angles in degrees)
No. of
plies
Residual
Fixed layer thickness
Case n. 1 [9/0/18/4/14/5] 6
3 5873 10
6
.

Case n. 2 [16/65/67/8/10/59] 6
1.7547 10
2

Case n. 3 [0/50/61/42/87/90/49/10/12/36/26/47/83] 13
1 6117 10
2
.

Case n. 4 [64/71/74/65/66/63/70] 7
1.7547 10
2

Variable layer thickness
Case n. 1
[9/6/4/11/9/6]
[0.118/0.126/0.140/0.126/0.103/0.152 mm]
6
3 2976 10
7
.

Case n. 2
[24/73/18/67]
[0.100/0.200/0.200/0.100 mm]
4 7 3315 10
3
.

Case n. 3
[19/74/45/20/75/17/53/83/2/51]
[0.113/0.168/0.200/0.137/0.149/0.190/0.199/0.113/0.106/0.116 mm]
10 2 0476 10
3
.

Case n. 4
[15/2/12/10/21/-6]
[0.156/0.188/0.158/0.151/0.111/0.182 mm]
6 2 9945 10
5
.

383
number of chromosomes, (p
mut
)
chrom
: their values were chosen according to De Jongs study [21]. In this study, it was sug-
gested that the mutation probability be inversely proportional to the size of population. Hence, for all simulations, the mutation
probability was chosen according to this rule.
Table 2 shows the examples of stacking sequences for laminates satisfying the design criteria for cases n. 1 to n. 4 at
fxed and variable layer thicknesses. The residual in the last column is the value of the global objective function F (n, , h) for
the solution indicated aside in the adjacent column (recall that exact solutions correspond to zeros of the objective function). As
usual in a numerical technique, the real solution is found to within a small numerical tolerance; this tolerance is the residual.
A discussion on the importance of the numerical residual in this type of problems can be found in [3, 18].
Tables 3-6 show the values of polar parameters for all the stacking sequences found in both the cases of constant and
variable thicknesses of plies. As seen, all the laminates are extension-bending uncoupled, although the stacking sequences are
not symmetric. Actually, some of these sequences are antisymmetric (the condition that guarantees the bending orthotropy, but
not always the bending-extension uncoupling, see [15]). For instance, the sequence of case n. 1 with plies of constant thick-
ness can be reduced to the sequence [2/7/11/11/7/2], which is antisymmetric, simply by a rotation through 7. Actually,
such an angle (see Table 3) corresponds to the polar angle
1
, and in this case, since K = 0, also to
0
. In fact, for a given
elasticity tensor L, in the case of orthotropy with K = 0, the direction determined by the angle
1
corresponds to the strong
axis of orthotropy, i.e., to the direction of the highest value of the component L
xxxx
, as easily seen from Eq. (6) (see also [10]).
A similar result is valid also for case n. 2 if the plies have a constant thickness, as well as for cases n.1 and n.4 if the plies have
a variable thickness. Nevertheless, the condition of antisymmetrical stacks is a suffcient condition for obtaining the bending
orthotropy, which is valid only for laminates with identical layers; normally, this condition cannot be applied to laminates hav-
ing plies of variable thickness. In our calculations, we preferred not to fx the orthotropy direction, because the properties that
we were looking for were intrinsic, i.e., frame-independent. The use of the polar formalism allows one not only to fx a frame,
for instance, by imposing a given value of the polar angle
1
, but also to do completely abstract from the frame whenever
intrinsic properties are sought for independently of any frame. Of course, a postprocessing operation of frame rotation, like
that described above, can always been done if one wishes to have the fnal result in a particular frame.
TABLE 3. Polar Parameters for the Laminate in Case n.1
Elastic properties Tensor A
*
Tensor B
*
Tensor D
*
At a constant ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 17.0334619 0.0195909 18.7724089
R
1
, MPa 20.6832749 0.00283 21.1735205

0
, deg 7.00 - 7.00

1
, deg 7.00 - 7.00

0

1
, deg 0.00 - 0.00
At a variable ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 19.4379543 0.0025019 19.5838459
R
1
, MPa 21.3588283 0.0009081 21.3986655

0
, deg 7.32 - 7.32

1
, deg 7.32 - 7.32

0

1
, deg 0.00 - 0.00
384
Figures 13-20 show the directional plots of some elastic properties of laminates for cases n. 1-n. 4; for the sake of clarity
and brevity, not all elastic properties have been illustrated, but those presented here are suffcient to show that the properties
prescribed have really been obtained.
TABLE 4.Polar Parameters for the Laminate in Case n.2
Elastic properties Tensor A
*
Tensor B
*
Tensor D
*
At a constant ply thickness
T
0
, MPa 26.804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 4.8733071 1.4079761 1.1229394
R
1
, MPa 13.0025339 0.1135579 14.6267514

0
, deg 7.50 - 7.50

1
, deg 37.50 - 37.50

0

1
, deg 45.00 - 45.00
At a variable ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 4.0359329 1.3458389 1.0298979
R
1
, MPa 13.4207539 0.1588238 14.6732203

0
, deg 0.50 - 0.50

1
, deg 45.50 - 45.50

0

1
, deg 45.00 - 45.00
TABLE 5. Polar Parameters for the Laminate in Case n.3
Elastic properties Tensor A
*
Tensor B
*
Tensor D
*
At a constant ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 0.1010911 0.5360837 0.1209914
R
1
, MPa 0.0093146 0.0907468 0.0636677

0
, deg - - -

1
, deg - - -

0

1
, deg - - -
At a variable ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 0.0454948 0.2537272 0.0304769
R
1
, MPa 0.0293042 0.0465314 0.0068212

0
, deg - - -

1
, deg - - -

0

1
, deg - - -
385
A detailed discussion of the results is presented below only for the laminates in cases n. 2 and n. 3, but similar con-
siderations can also be repeated verbatim in the other cases.
One can consider frst the requirement expressed by case n.2. From Table 4, it is seen that the laminate solution, both
at constant and variable thicknesses of plies, respects the design criteria (such verifcations are particularly simple in the polar
formalism, which is frame-independent, see Table 1):
1. in-plane orthotropy with K
A
*
=1:
a) plies of identical thickness,
0 1
7 50 37 50 45 00
A A
* *
. . . = ( ) = ;
b) plies of nonidentical thickness,
0 1
0 50 45 50 45 00
A A
* *
. . . = ( ) = ;
2. bending orthotropy with K
D
*
=1 :
a) plies of identical thickness,
0 1
7 50 37 50 45 00
D D
* *
. . . = ( ) = ;
b) plies of nonidentical thickness,
0 1
0 50 45 50 45 00
D D
* *
. . . = ( ) = ;
3. elastic uncoupling expressed by the polar condition B
*
= 0 . The norm of tensor the B
*
is negligible compared
with that of the tensor A
*
or D
*
:
a) plies of identical thickness,
B
A
*
*
. = 0 0270 ;
b) plies of nonidentical thickness,
B
A
*
*
. = 0 0260 ;
4. coincidence of orthotropy axes, polar condition
1 1
A D
* *
= :
a) plies of identical thickness,
1 1
37 50
A D
* *
. = = ;
b) plies of nonidentical thickness,
1 1
45 50
A D
* *
. = = .
TABLE 6. Polar Parameters for the Laminate in Case n.4
Elastic properties Tensor A
*
Tensor B
*
Tensor D
*
At a constant ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
, MPa 24.7438933 0 24.7438933
R
0
, MPa 18.9297208 0.0064786 18.9216397
R
1
, MPa 21.2193737 0.0008899 21.2174539

0
, deg 22.15 - 22.14

1
, deg 67.85 - 67.85

0

1
, deg 90.00 - -89.99
At a variable ply thickness
T
0
, MPa 26.8804311 0 26.8804311
T
1
MPa 24.7438933 0 24.7438933
R
0
, MPa 18.0907546 0.0444380 18.0881731
R
1
, MPa 20.9823915 0.0192386 20.9836104

0
, deg 10.05 - 10.05

1
, deg 10.08 - 10.08

0

1
, deg 0.03 - 0.03
386
For case n. 2, it is important to note that, concerning laminates with identical layer thickness, the laminate which sat-
isfes all requirements at a minimum number of plies involvs 6 layers, see Table 2. When the ply thickness becomes a design
variable, the solution found has many improvements, as shown in Table 3. First, the minimum number of layers decreases from
6 to 4, and this solution has a lower value of the residual, i.e., the laminate satisfes the requirements in a more satisfactory
way. Figures 15 and 16 show the polar diagrams of elastic properties of the laminate in this case.
From Table 5, it is seen that the laminate solution of problem n.3, both at constant and variable thicknesses of plies,
respects the design criteria:
1. in-plane isotropy, expressed by the polar condition that the anisotropic part of the tensor A
*
must be zero, i.e.,
R R
A A
0
2
1
2
4 0
* *
+ = . The ratio between the anisotropic and isotropic parts of the tensor is very close to zero, i.e., the anisotropic
components are negligible compared with the isotropic ones:
a) plies of identical thickness,
R R
T T
A A
A A
0
2
1
2
0
2
1
2
4
2
0 0023
* *
* *
.
+
+
= ;
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
180
150
210
120
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
E
D
11
, MPa
G
D
12
, MPa
G
A
12
, MPa
E
A
11
, MPa
Fig. 13. Polar variations for laminate n.1 at an identical ply thickness. (a) stiffness components, (b)
membrane Youngs modulus E
11
and shear modulus G
12
and (c) bending Youngs modulus E
11
and
shear modulus G
12
.
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
Fig. 14. The same for laminate n. 1 at a nonidentical ply thickness.
387
b) plies of nonidentical thickness
R R
T T
A A
A A
0
2
1
2
0
2
1
2
4
2
0 0017
* *
* *
.
+
+
= ;
2. bending isotropy, expressed by the polar condition that the anisotropic part of the tensor D
*
must be zero, i.e.,
R R
D D
0
2
1
2
4 0
* *
+ = . The ratio between the anisotropic and isotropic parts of the tensor is very close to zero, i.e., the anisotropic
components are negligible compared to the isotropic ones:
a) plies of identical thickness,
R R
T T
D D
D D
0
2
1
2
0
2
1
2
4
2
0 0040
* *
* *
.
+
+
= ;
b) plies of nonidentical thickness,
R R
T T
D D
D D
0
2
1
2
0
2
1
2
4
2
0 0007
* *
* *
.
+
+
= ;
150
210
120
180
240
90
60
30
0
270
300
330
1 0,000 0
50,000
150,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
1 0,000 0
50,000
150,000
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
80,000
60,000
100,000
40,000
20,000
Fig. 15. Polar diagrams for laminate n.2 at an identical ply thickness.
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
Fig. 16. The same for laminate n. 2 at a nonidentical ply thickness.
388
3. the elastic uncoupling is expressed by the polar condition B
*
= 0 . The norm of the tensor B
*
is negligible com-
pared to that of the tensor A
*
or D
*
:
a) plies of identical thickness,
B
A
*
*
. = 0 0130 ;
b) plies of nonidentical thickness,
B
A
*
*
. = 0 0060 .
In this case, concerning the laminate with identical layer thickness, the composite that satisfes all requirements at a
minimum number of plies can be made from 13 layers, see Table 2. When the ply thickness becomes a design variable, the
best solution is really improved, as shown in Table 2: the minimum number of layers to obtain a solution decreases from 13
to 10, and the value of the residual is smaller, i.e., the laminate satisfes the requirements more accurately. Figures 17 and 18
show the polar diagrams of elastic properties of the laminate in this case.
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
Fig. 17. Polar diagrams for laminate n.3 at an identical ply thickness.
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
60,000
40,000
20,000
80,000
Fig. 18. The same for laminate n.3 at a nonidentical ply thickness.
389
7. Conclusions
The problem of determining the minimum number of layers ensuring some given elastic properties of a laminate has
been addressed. The approach proposed in the paper to solve this problem is to reduce it to the classical unconstrained optimiza-
tion problem of search for the minima of a semidefnite positive function. This method is based on the so-called polar-genetic
approach and is totally general, i.e., no simplifying assumptions are required. The problem formulation is based on the polar
representation of plane tensors, while the numerical strategy for the search for solutions is a genetic algorithm, BIANCA,
which has been modifed to include the number of layers among the variables. The numerical results presented in this paper,
which are completely new and nonclassical, show the effectiveness of the method proposed.
Aknowledgements. The frst author was supported by the National Research Fund (FNR) in Luxembourg through
Aides la Formation Recherche Grant (PHD-09-139).
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
E
D
11
, MPa
G
D
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
Fig. 19. Polar diagrams for laminate n.4 at an identical ply thickness.
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
E
D
11
, MPa
G
D
12
, MPa
E
A
11
, MPa
G
A
12
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
A
*
11
, MPa
D
*
11
, MPa
B
*
11
, MPa
150
210
120
180
240
90
60
30
0
270
300
330
150,000
100,000
50,000
200,000
Fig. 20. The same for laminate n.4 at a nonidentical ply thickness.
390
REFERENCES
1. H. Ghiasi, D. Pasini, and L. Lessard, Optimum stacking sequence design of composite materials Part I: Constant
stiffness design, Composite Structures, 90, 1-11 (2009).
2. H. Ghiasi, K. Fayazbakhsh, D. Pasini, and L. Lessard, Optimum stacking sequence design of composite materials
Part II: Variable stiffness design, Composite Structures, 93, 1-13 (2010).
3. A. Vincenti, P. Vannucci, and G. Verchery, Design of composite laminates as an optimisation problem: a new genetic
algorithm approach based on polar tensor invariants, Proc. of WCSMO 5 (ffth world congress on structural and
multidisciplinary optimisation), Lido di Jesolo, Venezia, Italy, (2003).
4. P. Vannucci, Designing the elastic properties of laminates as an optimisation problem: a unifed approach based on
polar tensor invariants, Structural and Multidisciplinary Optimisation, 31, No. 5, 378-387 (2006).
5. P. Vannucci and A. Vincenti, The design of laminates with given thermal/hygral expansion coeffcients: a general
approach based upon the polar-genetic method, Composite Structures, 79, 454-466 (2007).
6. A. Vincenti, P. Vannucci, and M. R. Ahmadian, Optimization of laminated composites by using genetic algorithm and
the polar description of plane anisotropy, Mechanics of Advanced Materials and Structures, accepted for publication
(2011).
7. F. Werren and C. B. Norris, Mechanical properties of a laminate designed to be isotropic, US Forest Products Labo-
ratory, report 1841 (1953).
8. A. Vincenti, M. R. Ahmadian, and P. Vannucci, BIANCA: a genetic algorithm to solve hard combinatorial optimisa-
tion problems in engineering, Journal of Global Optimisation, 48, 399-421 (2010).
9. G. Verchery, Les invariants des tenseurs dordre 4 du type de llasticit, Proc. Of colloque Euromech 115, Villard-
de-Lans, France, (1979).
10. P. Vannucci, Plane anisotropy by the polar method, Meccanica, 40, 437-454 (2005).
11. R. M. Jones, Mechanics of Composite Materials, Taylor and Francis, USA (1975).
12. P. Vannucci, Thse dHabilitation Diriger les Recherches, Universit de Bourgogne, France, (2002).
13. P. Vannucci, A special planar orthotropic material, Journal of Elasticity, 67, 81-96 (2002).
14. A. Vincenti, G. Verchery, and P. Vannucci, Anisotropy and symmetries for elastic properties of laminates reinforced
by balanced fabrics, Composites Part A, 32, 1525-1532 (2001).
15. E. Valot and P. Vannucci, Some exact solutions for fully orthotropic laminates, Composite Structures, 69, 157-166
(2005).
16. D. E. Goldberg, Genetic algorithms, Addison and Wesley, New York (1994).
17. Z. Michalewicz, Genetic algorithms + data structures = evolutionary programming, Springer, (1994).
18. A. Vincenti, Conception et optimisation de composites par mthode polaire et algorithmes gntiques, PhD Thesis,
Universit de Bourgogne, France, (2002).
19. C. H. Park, W. I. Lee, W. S. Han, and A. Vautrin, Multiconstraint optimisation of composite structures manufactured
by resin transfer moulding process, J. Composite Materials, 39, No. 4, 347-374 (2005).
20. S. W. Tsai and T. Hahn, Introduction to Composite Materials, Technomic, (1980).
21. K. A. De Jong, An Analysis of the Behaviour of a Class of Genetic Adaptive Systems, PhD Thesis, University of
Michigan, USA (1976).