Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

2

nd
International Conference on Engineering Optimization
September 6 - 9, 2010, Lisbon, Portugal




1
New developments in material parameterization for optimal fiber orientations
in composite structures

M. Bruyneel
1
, D. Deltour
2
, P. Duysinx
2
, T. Gao
2
, T. Schoonjans
2
, C. Fleury
2


1
SAMTECH s.a., Lige Science Park, Angleur, Belgium, michael.bruyneel@samtech.com
2
Aerospace & Mechanical Department, University of Lige, Lige, Belgium, p.duysinx@ulg.ac.be

Abstract
In this paper, a new parameterization of the mechanical properties is proposed for the optimal selection of ply
orientation in structures made of fibers reinforced composite materials. Recent parameterization schemes from
multi-phase topology optimization (i.e. Discrete Material Optimization DMO) are compared to the novel
approach in the selection of conventional laminates including only 0, -45, 45, 90 and 45 plies. In the new
parameterization, termed SFP (Shape Functions with Penalization), the material stiffness is computed as a
weighted sum of the candidate material properties, and the weights are based on the classical shape functions used
in the finite elements method. When the 4 candidate orientations 0, -45, 45, 90 are considered, each vertex of
the reference quadrangle then represents a candidate ply. In this case, compared to DMO, this method requires
fewer design variables, since the 4 variables representing the presence or the absence of a given candidate ply in
DMO are now replaced in the weights by two design variables, which are the 2 natural coordinates of the reference
quadrangular element sufficient to identify each of the 4 vertices. The method is extended to the case of 3
candidate sub-laminates ([0
2
], [90
2
] and [45]), where a 3-nodes representation based on the shape functions of a
triangular membrane element is used. Numerical applications with in-plane loadings are proposed and solved in
order to demonstrate that the new approach is an interesting alternative to DMO, able to select the optimal
orientations and to combine the material distribution and optimal orientation problems.
Keywords: composites, material selection, topology optimization, DMO, SFP

1. Introduction
The problem of selecting a suitable material has been studied for a long time. One of its applications concerns the
selection of the optimal distribution of fiber orientations in composite structures [1-3] and the identification of the
optimal stacking sequence [4-8]. In most aerospace applications, the candidate materials are restricted to the
conventional angles with plies oriented at 0, 45, -45 and 90. This is by nature a discrete optimization problem.
However, the specific parameterizations discussed here allow to work with a continuous formulation, and reliable
optimization methods developed for problems involving continuous variables can therefore be applied. In this
paper, we concentrate on the selection of the optimal local fiber orientations in membrane structures.

2. Material parameterization for optimal ply selection
2.1 The DMO approach
In their recent work, Lund and co-workers [9-14] have proposed the Discrete Material Optimization (DMO)
approach to determine the optimal fiber orientations and to solve the topology optimization problem of structures
made of fibers reinforced composite materials. This approach is an extension of the multi-phase topology
optimization proposed in [15]. When applied to a composite ply noted l, it consists in writing the linear anisotropic
material stiffness matrix C
l
as a weighted sum over the stiffness of some candidate materials. When conventional
laminates are used, and assuming that materials 1, 2, 3 and 4 correspond to fibers oriented at 0, 45, 90 and -45,
respectively, it comes that (no summation over the index l):


l l l l l l l l
n
i
l
i
l
i
l
w w w w w
l
4 4 3 3 2 2 1 1
1
C C C C C C + + + = =
=
(1)
with 1
4 3 2 1
1
= + + + =
=
l l l l
n
i
l
i
w w w w w
l
(2)
and 1 0
l
i
w ,
l
n i ,..., 1 = (3)

In (1) and (2), n
l
is equal to 4 since 4 candidate orientations are considered. At the optimum, the physical ply l
should be made of one (and only one) of the candidate plies, with the material properties C
1
, C
2
, C
3
or C
4
(Figure


2
1). Equations (2) and (3) needs to be satisfied to obtain physically meaningful results. It results that the weights w
i

must be non negative (3) and must satisfy the partition of unity (2).



Figure 1: Selection of the optimal fiber orientation in a set of 0, 45, -45 and 90 plies

Several schemes have been proposed in [9-14] for the expression of the weighting factors w
i
in (1). According to
[12], the schemes of Equations (4) and (5) are the most reliable ones. These are called DMO4 and DMO5,
respectively, in [12]. Omitting the index l, they are written:

( ) ( ) [ ]

=
=
n
i j j
p
j
p
i
DMO
i
x x w
; 1
4
1 for DMO4 (4)

=
=
n
k
DMO
k
DMO
i DMO
i
w
w
w
1
4
4
5
for DMO5 (5)

In (4) and (5), the parameters x
i
(i = 1,,n) are the design variables, which take their values between 0 and 1. p is
an exponent used to penalize the intermediate values of the design variables. According to the authors [9-14], the
problem with DMO4 is that the condition (2) is only verified for design variables x
i
equal to 0 or 1, which could
lead to convergence problems; the difficulty with DMO5, which is a scaled version of DMO4, is that the effect of
the penalization p is less predominant, and the results can be made of a lot of intermediate values of the design
variables, leading to a mixture of the candidate materials at the solution. Nevertheless, both DMO4 and DMO5
perform well, and interesting solutions have been obtained in [9-14]. The schemes DMO4 and DMO5 are
illustrated in Figure 2.


DMO4: w1 and w2 for p = 3

DMO5: w1 and w2 for p = 3

Figure 2: DMO4 and DMO5 schemes


3
Since composite membranes are studied in this paper, the ply thickness can replace the material stiffness in (1), as
they both have a linear contribution to the in-plane stiffness matrix A, and the selection of the optimal fiber
orientation can be conducted on the candidate ply thickness, with a condition on a non negative thickness (3) and a
total ply thickness to be obtained at the solution (2). The resulting parameterization is used in the applications of
Section 4:

total
l l l l l l l l
n
i
l
i
l
i
l
t t w t w t w t w t w t
l
= + + + = =
=
4 4 3 3 2 2 1 1
1


2.2 The 4-nodes SFP approach
As proposed in [16], the shape functions of the 4-nodes quadrangular element can be used as weights in (1), as
depicted in Figures 3 and 4a. These specific functions satisfy the conditions (2) and (3). The 4 weights of the Shape
Function (SF) parameterization are written in (6) and illustrated in Figure 4a. Two design variables are enough to
identify each of the 4 vertices. These variables, termed R and S, are the coordinates of the reference quadrangle,
classically used for the integration of the stiffness matrix and the force vector in the finite element method [17].
They are sufficient to identify each vertex of the quadrangle, corresponding to each candidate material (Figure 3).
The shape functions in (6) are bilinear in terms of R and S. As a result, SF does not penalize the intermediate values
of the design variables, and a mixture of the 4 candidate plies can be observed at the solution. The Shape Functions
with Penalization (SFP) scheme is then proposed in [16], as written in (7) and illustrated in Figure 4b. The
intermediate values of the weights are now penalized in a scheme which is similar to the SIMP law used in
topology optimization (Figures 4b and 5). The condition (2) is no longer satisfied for the intermediate values of the
design variables appearing during the iterative process. However, the results of the applications considered in this
paper will show that this is not an issue in terms of the number of iterations needed to reach a solution.

( )( ) S R w
SF
i
= 1 1
4
1
for SF
that is ( )( ) S R w = 1 1
4
1
1
( )( ) S R w + = 1 1
4
1
2

( )( ) S R w + + = 1 1
4
1
3
( )( ) S R w + = 1 1
4
1
4
(6)

( )( )
p
SFP
i
S R w
(

= 1 1
4
1
for SFP (7)



Figure 3: SFP for 4 candidate materials


Figure 4: The SF and SFP (with p = 3) schemes for the weights in the material selection (1)


4


One of the advantages of the SFP approach compared to DMO is the lower number of design variables necessary
to parameterize the problem. When 4 candidate materials are considered in the problem, 2 design variables are
sufficient for SFP, while 4 parameters must be used with DMO. These 2 design variables are the 2 natural
coordinates of the reference quadrangular element sufficient to identify each of the 4 vertices corresponding to
each candidate orientation (Figure 3).


Figure 5: Weighting function w
2
in SFP, for different values of p (S = -1)


2.3 The 3-nodes SFP approach
The DMO schemes are general and can be applied to the selection of an arbitrary number of candidate plies. Here,
a SFP formulation is proposed for 3 candidate sub-laminates, i.e. [0
2
], [45] and [90
2
] orientations, therefore
extending the range of application of SFP first described in [16] for 4 candidate plies. In this case, the 3-nodes
triangular element is used (Figure 6), and a double thickness value is associated to the candidate ply with fibers
oriented at 0 and 90, while a unit thickness is assigned for the 45 and -45 plies. The resulting weights w
i
are
given in (8). A penalty can be added in order to force the solution towards a vertex. An additional constraint in (8)
must now be added to the problem, in order to limit the possible values of R and S. The formulation is illustrated in
Figure 6, which can be compared to the SFP scheme with 4 candidate plies of Figure 3.

S R w =1
1

R w =
2

S w =
3

0 1 S R (8)




Figure 6: SFP for 4 candidate materials

2.4 The scaled SFP schemes
As it is the case for DMO5 compared to DMO4, a scaled version of SFP can be proposed. The corresponding
weights are illustrated in Figure 7.



5

Figure 7: The SFP schemes (with p = 3) without and with scaling for the weights in the ply selection (1)

2.5 Discussion on the penalization scheme for SFP
In (7), a SIMP penalization scheme is used to prevent the intermediate values of the design variables at the
solution, and therefore avoid any mixture of candidate materials in the final design. This choice is not unique, and
other penalization schemes can be proposed. If the intermediate values of a variable [0;1] must be penalized,
the following schemes can be proposed:

SIMP [18]
p
f = ) ( (9)
RAMP [19]
) 1 ( 1
) (

+
=
p
f (10)
Halpin-Tsai [20]

+
=
) 1 (
) (
r
r
f (11)
Polynomial penalization

1 1
) ( +

=
p
f (12)

These different penalization schemes are compared in Figure 8, where p = 3, r = 0.269 for Halpin-Tsai and = 16
for the polynomial scheme. Theses values of the parameters will be used in the application of Section 4.4.



Figure 8: Comparison of the penalization schemes

Two strategies can be identified to penalize the intermediate values of the weights in (1). In the first strategy, the
weights are penalized with one of the schemes (9) to (12), as it was done from equations (6) to (7). In the second
strategy, the penalization is applied to the design variables instead of the weights. These two strategies will be
compared in the following applications.

3. Numerical procedure
Our own implementation of MMA [21,22] available in the BOSS Quattro optimization tool box [23] is used,
together with the SAMCEF finite element code for the structural analyses [24]. The objective function of the
following optimization problems consists to maximize the global in-plane structural stiffness, in a linear static
analysis. The sensitivities are computed by finite differences.


6
4. Applications
4.1 SFP for 4 candidate orientations
The structure studied is illustrated in Figure 9. It is divided into 16 regions of independent fibers orientations. In
each region, we are looking for the best ply amongst the set of 0, 45, 90 and -45 plies. The orientations can
therefore vary from region to region at the solution. For the DMO schemes, 64 design variables are needed. For the
SFP parameterizations, 32 design variables are used. The results are provided in Figure 10 and the summary of the
convergence history is given in Table 1. SFP with p larger than or equal to 3 provides the lowest value of the
objective function. For DMO4, a mixture of the candidate plies is observed at the solution.


Figure 9: Non homogeneous membrane test case



Figure 10: Results obtained with DMO5 (left) and SFP with p = 5 (right)


Table 1: Results for the selection amongst 4 candidate plies

DMO4
p = 5
DMO5
p = 5

SFP
p = 1
SFP
p = 2
SFP
P = 3
SFP
p = 4
SFP
p = 5
Mixture of
materials
Larger value
of obj. fct.
Mixture of
materials
Mixture of
materials
Best
solution
Best
solution
Best
solution
20 iterations 5 iterations 12 iterations 18 iterations 4 iterations 4 iterations 4 iterations


4.2 SFP for 4 candidate orientations in a topology optimization problem
The problem of section 4.1 is now solved as a topology optimization problem. Besides the design variables
previously defined for the selection of the optimal orientation, an additional design variable is used in each
region in order to remove material at the solution. The material parameterization is now written as [16]:

( )
|
|

\
|
=
=
l
n
i
l
i
l
i
q
l
l
w
1
C C (13)

A resource constraint is added to the optimization problem, which now aims to maximize the stiffness with respect
to relation (14). The number of regions where material is present is then limited to 11.



7
11
16
1

= l
l
(14)

The solution for SFP is provided in Figure 11. It is reached in 19 iterations. The optimal orientations are the same
as the ones obtained in Figure 10.



Figure 11: Solution of the material selection and topology optimization problems with SFP

4.3 SFP for 3 candidate orientations
The problem of section 4.1 is now solved for 3 candidate sub-laminates with plies oriented at 0, 90 and 45. The
solution is illustrated in Figure 12, and the convergence history is summarized in Table 2.



Figure 12: Results obtained with DMO5 (left) and SFP with p = 5 (right)


Table 2: Results for the selection amongst 3 candidate sub-laminates

DMO4
p = 4
DMO5
p = 5

SFP
p = 1
SFP
p = 2
SFP
P = 3
SFP
p = 4
SFP
p = 5
Mixture of
materials
Larger value
of obj. fct.
Mixture of
materials
Mixture of
materials
Mixture of
materials
Best
solution
Best
solution
23 iterations 5 iterations 6 iterations 14 iterations 8 iterations 17 iterations 13 iterations


4.4 Results for different penalization schemes in SFP
The SFP method is used with the 4 different penalization schemes (9) to (12) and the 2 strategies discussed in
Section 2.5. For the first strategy, the solution of Figure 10b is obtained in 4 iterations, for the 4 penalization
schemes tested. When the second strategy is used (penalization of the design variables instead of the weights),
oscillations appear in the iterative process, and a solution can not be identified. The SIMP penalization on the
weights is then a reasonable choice for the problems solved in this paper.




8
4.5 MBB beam
The structure is illustrated in Figure 13. According to the symmetry, only one half is studied. It is divided in 48
regions of independent fiber orientations. In this problem, 4 candidate orientations are considered, i.e. 0, 45, -45
and 90. The optimal fiber orientations obtained with DMO5 and SFP are illustrated in Figures 14 and 15. Six and
five iterations are necessary for DMO5 and SFP, respectively, to reach the solution. SFP provides a final design
with a smaller value of the objective function. In Figure 16, the directions of the principal stresses are provided. As
demonstrated in [1], in an optimal design, the fiber directions should be aligned with the directions of the principal
stresses. The solution provided by SFP tends to satisfy this criterion.


Figure 13: The MBB beam


Figure 14: Solution with DMO5 (p = 5)


Figure 15: Solution with SFP (p = 2)


Figure 16: Directions of the principal stresses




9
4.6 Membrane with a hole
In the last application, a membrane with a hole is considered (Figure 17). It is submitted to tension, with
1 2
2 = .
According to the symmetries, only one quarter of the structure is studied. The panel is divided in 12 regions, where
the optimal fibers orientations must be selected amongst 0, 45, -45 and 90 plies. The results are reported in
Figure 18. DMO5 (with p = 5) and SFP (with p = 2) provide the same final design, in 6 and 5 iterations,
respectively. The optimal fibers orientations are in good agreement with the principal stress directions.

Figure 17: Membrane with a hole



Figure 18: Results obtained with DMO and SFP, and principal stress directions


5. Conclusions
A new parameterization for the optimization of discrete fiber orientations has been presented. It is based on the
shape functions classically used in finite elements, and is applied to the selection of conventional plies oriented at
0, 45, 90, -45 and 45. Compared to other schemes like DMO (Discrete Material Optimization), the SFP
scheme (Shape Functions with Penalization) requires a smaller amount of design variables. The weights are
penalized in order to avoid the mixture of candidate materials at the solution. The specific shape of the resulting
penalty functions is very similar to the SIMP law for topology optimization, and SFP seems less likely to provide
a mixture of candidate materials at the solution, compared to DMO. The selection of 4 and 3 candidate orientations
is addressed. The proposed method showed interesting results when tested on simple representative problems. SFP
is an interesting alternative to DMO, and is able to select the optimal orientations and to combine the material
distribution and optimal orientation problems.
The SFP method should be further studied and applied to optimization problems for more than 4 candidate
materials. Vibration and buckling optimizations should also be investigated.

Acknowledgements
This work was supported by the Walloon Region of Belgium and SKYWIN (Aerospace Cluster of Wallonia),
through the project VIRTUALCOMP.




10
References
[1] P. Pedersen On optimal orientation of orthotropic materials. Structural Optimization, 1, 101-106, 1989.
[2] P. Pedersen. On thickness and orientational design with orthotropic materials. Structural Optimization, 3,
69-78, 1991.
[3] V. Hammer. Design of composite laminates with optimized stiffness, strength and damage properties. PhD
Thesis, DCAMM Report S72, Technical University of Denmark, 1997.
[4] R.T. Haftka. Stacking sequence optimization for buckling of laminated plates by integer programming. AIAA
Journal 30(3), 814-819, 1992.
[5] R. Le Riche and R.T. Haftka. Optimization of laminate stacking sequence for buckling load maximization by
genetic algorithm. AIAA Journal, 31(5), 951-956, 1993
[6] C. Diaconu and H. Sekine. Layup optimization for buckling of laminated composite shells with restricted
layer angles, AIAA Journal, 42(10), 2153-2163, 2004.
[7] C.C. Kin and Y.J. Lee. Stacking sequence optimization of laminated composite structures using h-genetic
algorithm with local improvement. Composite Structures, 63 (3), 339-345, 2004.
[8] D. Liu, V. Toropov, D. Barton and O. Querin. Two methodologies for stacking sequence optimization of
laminated composite materials. Procceedings of the International Symposium on Computational Structural
Engineering, Shangai, China, June 22-24, 2009.
[9] E. Lund and J. Stegmann. Structural optimization of composite shell structures using a discrete constitutive
parameterization. XXI ICTAM, 15-21 August 2004, Warsaw, Poland.
[10] E. Lund and J. Stegmann. On structural optimization of composite shell structures using a discrete
constitutive parameteriation. Wind Energy, 8, 109-124, 2004.
[11] E. Lund, L. Khlmeier and J. Stegmann. Buckling optimization of laminated hybrid composite shell
structures using discrete material optimization. 6th World Congress on Structural and Multidisciplinary
Optimization, Rio de Janeiro, 30 May 3 June 2005, Brazil.
[12] J. Stegmann. Analysis and optimization of laminated composite shell structures. PhD Thesis, Aalborg
University, Denmark, 2005.
[13] J. Stegmann and E. Lund. Discrete material optimization of general composite shell structure. International
Journal for Numerical Methods in Engineering, 62, 2009-2027, 2005.
[14] E. Lund. Buckling topology optimization of laminated multi-material composite shell structures, Composite
Structures, 91, 158-167, 2009.
[15] O. Sigmund and S. Torquato. Design of materials with extreme thermal expansion using a three-phase
topology optimization method. Journal of the Mechanics and Physics of Solids, 48, pp. 461-498, 2000.
[16] M. Bruyneel. SFP a new parameterization based on shape functions for optimal material selection.
Application to conventional composite plies, Structural and Multidisciplinary Optimization, accepted (under
revision), 2010.
[17] O.C. Zienkiewicz. The finite element method. 3rd edition, McGraw-Hill, New York, 1977.
[18] M.P. Bendsoe. Optimal shape design as a material distribution problem. Structural Optimization, 1, 193-202,
1989.
[19] M. Stolpe and K. Svanberg. An alternative interpolation scheme for minimum compliance topology
optimization. Structural and Multidisciplinary Optimization, 22, 116-124, 2001.
[20] J.C. Halpin and S.W. Tsai. Effects of environmental factors on composite materials. AFML-TR, 67-423,
June 1969.
[21] K. Svanberg. The method of moving asymptotes: a new method for structural optimization. International
Journal for Numerical Methods in Engineering, 24, 359-373, 1987.
[22] M. Bruyneel. A general and effective approach for the optimal design of fiber reinforced composite
structures. Composites Science & Technology, 66, 1303-1314, 2006.
[23] Y. Radovcic and A. Remouchamps. BOSS Quattro: an open system for parametric design. Structural &
Multidisciplinary Optimizarion, 23, 140-152, 2002.
[24] SAMCEF. Systme dAnalyse des Milieux Continus par Elments Finis. www.samtech.com

You might also like