Professional Documents
Culture Documents
www.elsevier.com/locate/aescte
A new procedure for aerodynamic missile designs using topological
optimization approach of continuum structures
Zhen Luo
a,
, Jingzhou Yang
b
, Liping Chen
c
a
Department of Automation & Computer-Aided Engineering, The Chinese University of Hong Kong, Shatin, NT, Hong Kong
b
Center for Computer-Aided Design, The University of Iowa, 111 Engineering Research Facility, Iowa City, IA 52242, USA
c
School of Mechanical Science & Engineering, Huazhong University of Science & Technology, Wuhan, Hubei, 430074, China
Received 20 June 2005; received in revised form 21 November 2005; accepted 20 December 2005
Available online 6 February 2006
Abstract
This paper presents a multi-objective programming scheme for the conceptual design of aerodynamic missiles structure using topological
optimization approaches, in which both the compliance and eigenfrequency are regarded as static and dynamic optimization objectives, respec-
tively. During the conceptual design of the aerodynamic missile with multiple loadings, both the multilevel sequential programming approach
and the compromising programming method are coupled together to settle the associated difculties when the whole structure of the missile
body is considered as a pre-dened design domain. The compromise programming method is rst applied to describe the statically loaded multi-
stiffness topology optimization, and the dynamic formulation is used to establish the subsequent optimization problem mainly concerned with
free vibration. The main advantage of the proposed scheme is the exibility of dealing with optimal topology designs for the whole structures
of aerodynamic missiles with complicated loading cases. Solid isotropic material with penalization (SIMP) is used as the interpolation scheme
to indicate the dependence of material modulus upon regularized element densities. The sequential convex programming approach is applied to
solve the optimization problem. An engineering application is used to demonstrate the characteristics of the presented methodologies based on
the commercial software package of Hyperworks.OptiStruct at Altair.
2006 Elsevier SAS. All rights reserved.
Keywords: Structural topology optimization; Aerodynamic missiles; Multi-objective; Multilevel sequence programming; Compromise programming;
Mathematical programming
1. Introduction
Presently, topology optimization techniques for continuum
structures [2] are mainly concentrating on single objective op-
timization problems, such as minimizing mean compliance or
maximizing the fundamental eigenvalue. However, in practical
engineering elds, there exist lots of multi-objective topology
optimization problems. A structure subjecting to static and dy-
namic loading cases might be a typical multi-objective topol-
ogy optimization problem, where minimizing compliance and
maximizing fundamental eigenvalue are usually regarded as
the static and dynamic objectives, respectively [1315,21]. Up
*
Corresponding author. Tel.: +852 2609 8041; fax: +852 2603 6002.
E-mail addresses: luozhen_me@yahoo.com.cn (Z. Luo),
jyang@engineering.uiowa.edu (J. Yang).
to now, many methodologies have been developed for solving
multi-objective programming problems [16]. Concerning the
multi-objective optimization problems, it is rarely that all the
single objectives can be optimized simultaneously and a unique
optimal solution generally cannot be expected. It is therefore in-
volved the process of making trade-off decisions to obtain a set
of the compromise solutions. The concept of Pareto solution is
commonly used in characterizing the compromise solution of a
multi-objective optimization problem. The denition of Pareto
set states that the vector is chosen as optimal if no criterion can
be improved without worsening at least one other criterion.
In this work, we are mainly limited our topic upon the
methodologies for multi-objective topology optimization of
aerodynamic missiles. Concerning an aerodynamic missile, the
missile body might be the most important structure and the
loading cases for the whole missile body are very complicated,
1270-9638/$ see front matter 2006 Elsevier SAS. All rights reserved.
doi:10.1016/j.ast.2005.12.006
Z. Luo et al. / Aerospace Science and Technology 10 (2006) 364373 365
Fig. 1. Topology, sizing and shape optimization for A380 droop nose ribs (http://www.altair.com.cn).
in which the static and dynamic loads should be two kinds of
regular loading cases. Currently, most optimization approaches
for aircraft designs are mostly concerning on the size or shape
optimization, and topological optimization methods are rela-
tively rarely adopted. Although some researchers applied topol-
ogy optimization techniques to the design of aircrafts, such as
the designs for aeroelastic wing, A380 leading edges ribs and
aircraft wing box ribs [9,17,18], they are mainly limited on de-
sign of components with the single objective such as static or
dynamic. To our knowledge, the research works concentrating
on topology optimization of aerodynamic missiles regarding
the whole structural body of the missile as the design domain
is very scarce. However, topology optimization for the whole
structure of the missile body during the process of conceptual
design must be benecial to improve the global performance of
the aerodynamic missiles.
Minimizing the strain energy for continuum topological op-
timization is not very capable of dealing with local stress and
buckling. The general way is therefore to fully investigate an in-
tegrated structural topology optimization approach by combin-
ing topology with sizing and shape optimization, where topol-
ogy optimization is used for optimal design concepts, and the
sizing and shape optimization can be used for detailed and fun-
damental designs such as stress, fatigue and bucking (Fig. 1).
However, the purpose of this paper is mainly limited upon the
rst part of this integrated procedure. The goal of this study
is to explore a suitable multi-objective programming procedure
for handing topology optimization problem which is subject to
multiple loading cases. The aim of topology optimization for
aerodynamic missiles is to assist the designer to determine the
most efcient way of an optimal material distribution in the
predened design space with a prescribed amount of material
usage. It is necessary to employ the method of structural topol-
ogy optimization to improve the performances of aerodynamic
missiles by means of the static and dynamic multi-objective ap-
proach.
The rest of this paper is arranged as follows. Section 2 de-
scribes multi-stiffness topological optimization, Section 3 dis-
cusses the topology optimization of free vibrating structures,
Section 4 discusses the multi-objective topological optimiza-
tion with static and dynamic objectives, Section 5 shows the nu-
merical procedure and Section 6 illustrates one typical example
of certain aerodynamic missile and nally gives the conclusion.
2. Topology optimization for statically loaded
aerodynamic missiles
Topological optimization problem of maximizing structural
stiffness is a process of purchasing stiffness distributing forms
in the design space. The maximum stiffness topology designs
subject to multiple loading cases are generally stated as multi-
ple stiffness problems [13]. The different optimal stiffness for
structures with multiple static load cases is called multiple-
stiffness where each independent loading case relates to an
optimal stiffness distributing topology. The multiple loading
cases exerted upon the structure simultaneously will result in
different optimal stiffness in the same design domain. Gen-
erally, the optimal topologies corresponding to different load-
ing cases are usually dissimilar in the same design space and
only an efcient solution can be approached. Considering the
present unsatisfactory situation for multi-stiffness problems [1],
the topology optimization subject to several independent sets of
366 Z. Luo et al. / Aerospace Science and Technology 10 (2006) 364373
loads in this paper is posed as a multi-objective problem. When
we consider more complicated cases, such as aerodynamic mis-
sile with many different static loading cases, to obtain a single
optimum for this multi-stiffness optimization problem needs a
multi-decision making process at the top level.
Among available approaches the weighting method is the
easiest one to be implemented and the OptiStruct package
uses this approach by minimizing a linear combination of all
single criteria subject to non-negative weighting factors. The
set of Pareto optimum solutions can be obtained by varying
the weights. However, for non-convex optimization problems,
this method cannot ensure nding all Pareto optimal solutions.
Therefore, it is difcult to capture the whole Pareto character-
istics when Pareto points are linearly distributed. In addition, it
cannot appropriately approximate the real Pareto optimal curve,
because a uniform variation of the weighting factors often re-
sults in an uneven distribution of the Pareto optimum solutions.
It is not easy to balance objectives only by varying weights
when loading scales are greatly different. The optimal stiffness
distributing paths corresponding to small loading cases might
be smeared out gradually during iteration process, and the con-
tribution of small loading cases to the optimal solution cannot
be embodied properly. This situation will result in an ill-posed
loading condition like the ill-posed stiffness matrix. The ef-
fective optimum solution will be easily controlled by optimal
results corresponding to larger loading cases.
To overcome the advantages of the mentioned approach, the
compromise programming scheme [6,28] was proposed as the
solution strategy. Compromise programming actually identies
the optimal solution that has the shortest distance to the ideal
point where conicting objectives reach their corresponding
minimum values but all the criteria cannot be satised at the
same time. The ideal point is not obtainable practically but can
be used as a base point. Concerning convex or non-convex op-
timization problem, the desired set of whole Pareto optimum
solutions can be obtained by varying the corresponding weight-
ing factors. The feasible point with the shortest distance to the
ideal point is recognized as the optimum solution in the com-
promise programming.
The objective of compromise programming approach is de-
scribed as following:
C(X) =
_
m
k=1
w
q
k
_
C
k
(X) C
min
k
C
max
k
C
min
k
_
q
_1
q
=
_
w
q
1
_
C
1
(X) C
min
1
C
max
1
C
min
1
_
q
+
+w
q
m
_
C
m
(X) C
min
m
C
max
m
C
min
m
_
q
_1
q
, (1)
where C
k
(X) is the objective related to the kth loading case.
C
min
k
and C
max
k
separately indicate the best and the worst ob-
tainable compliance solutions of the kth single objectives. In
other words, C
min
k
and C
max
k
are the minimum value and maxi-
mum value of C
k
(X) in the feasible region. Concerning the kth
stiffness topology optimization, C
min
k
= C
min
k
(X
k
) is the min-
imum compliance and X
k
is the design variable vector for the
minimumcompliance. Similarly, C
max
k
=C
max
k
(X
k
) is the max-
imum compliance which if obtained by using the design vari-
able vector X
k
. The weights satisfy w
k
0. q is a penalty ex-
ponent and q 2, which is used to indirectly control the shares
of single objectives involved in the compromise programming
formulation. According to Goicoechia et al. [8] if q = 1, all
distances from the ideal point are equally weighted and the op-
timization problem simply indicates the traditional weighting
method. The compromise solution can be obtained by solving a
linear programming problem. If q , the maximum diver-
gence between individual discrepancies is minimized, and the
optimization problem will be evolved to a min-max problem.
The compromise solution is obtained by solving this min-max
problem. For other cases rather than q = 1 and q , non-
linear mathematical programming algorithms are commonly
needed to obtain compromise solutions. For example, if q = 2,
each deviation is weighted in proportion to its magnitude, and
the larger value of the objective function in the parentheses will
carry the greater share. In this work q = 2 is suggested to shape
the Eq. (1) in the form of square root.
Then the mathematical formulation of the optimization prob-
lem is dened as
Minimize
X=(x
1
,x
2
,...,x
n
)
T
: C(X) =
_
m
k=1
w
q
k
_
C
k
(X)C
min
k
C
max
k
C
min
k
_
q
_ 1
q
Subject to:
m
k=1
_
n
j=1
V
j
x
k
j
_
V 0,
0 <x
min
x
j
<1,
j = 1, 2, . . . , n; k = 1, 2, . . . , m,
Equilibrium equations,
(2)
where x is the design variables, m is the number of loading
cases, and n is the number of nite elements, The minimum
density x
min
= 0.0001 is generally introduced to avoid singu-
larity of stiff matrix during the approximations of nite element
subroutines, w is the weighting efcient, and
V denotes the
amount of material allowable in the design space. The sensi-
tivity analysis of the single objective function is dened by
The computational model for single objectives can be writ-
ten as:
Minimize
X=(x
1
,x
2
,...,x
n
)
T
: C(X) = {U}
T
[K]{U}
Subject to:
E(x) =E
min
+x
p
j
(E
0
E
min
),
n
j=1
V
j
x
j
V 0, j = 1, 2, . . . , n,
0 <x
min
x
j
<1, j = 1, 2, . . . , n,
Equilibrium equation.
(3)
With regard to the SIMP interpolation scheme, the stiffness
matrix, objective function and the sensitivity with respect to de-
sign variables are given as follows
[K] =
n
j=1
_
E
min
+x
p
j
E
_
[K
j
], (4)
C(x) =
n
j=1
_
E
min
+x
p
j
E
_
{U
j
}
T
[K
j
]{U
j
}, (5)
C
(x) =
n
j=1
px
p1
j
{U
j
}
T
[K
j
]{U
j
}E, (6)
Z. Luo et al. / Aerospace Science and Technology 10 (2006) 364373 367
where E = E
0
E
min
; E
0
and E
min
are material modulus;
for numerical stability we choose E
min
= E
0
/1000; [K] and
{U} represent the global stiffness matrix and displacement, re-
spectively; [K
j
] is the element stiffness matrix; and {U
j
} is the
element displacement vector. Here the exponent p of SIMP is
exactly used to penalize the intermediate densities when p sat-
ises the condition p 2, and the details of SIMP is given in
Section 5.1.
3. Dynamic topological optimization of aerodynamic
missiles
Dynamic topology optimization of continuum structures
plays a very important role in the structural optimization. For
dynamic topology optimization of aerodynamic missiles, free
vibration problems under real working conditions might be the
most typical issues. Therefore, the sympathetic vibration would
be usually occurred when the forced frequencies are concerted
with the inherent frequencies of the structure itself. This unde-
sirable phenomenon will accelerate or cause serious damages
to the real structure. The main purpose of this section is to
explore the means to prevent the potential sympathetic vibra-
tion by increasing the dynamic response ability of the aero-
dynamic missiles by using topology optimization. Maute and
Reich [18], Krog and Tuck [9], Kosaka and Swan [12], Ped-
erson [19] and Luo et al. [14] investigated dynamic topology
optimization problems. However, dynamic topology optimiza-
tion considering the whole body of the aerodynamic missile as
the design domain has not been studied to improve the struc-
tural global performance.
Considering the optimization problem where the objective is
to maximize the kth lowest eigenfrequency of the structure sub-
ject to a given volume constraint, when we maximize the kth
eigenfrequency, the other eigenfrequencies may fall down to
the lower values, and the eigenfrequencies might switch their
orders frequently during the optimization process. Therefore,
the oscillation of the objective function might occur. This oscil-
lation will result in the non-differentiable objective and the non-
smoothness sensitivity of the objective function (eigenvalues).
This may result in a worse divergence of the optimization prob-
lem. In order to overcome this problem the mean-frequency
formulation has been suggested to smooth out the objective
[15].
The mathematical model is described as
Maximize
X={x
1
,x
2
,...,x
n
}
T
: (X) =
_
N
k=1
(w)
k
_
m
k=1
w
k
/
k
_
,
k = 1, 2, . . . , N
,
Subject to:
E =x
p
j
(E
0
E
min
), j = 1, 2, . . . , n,
n
j=1
V
j
x
j
V 0, j = 1, 2, . . . , n,
0 <x
min
x
j
<1, j = 1, 2, . . . , n,
([K]
i
[M]){
i
} = {0},
(7)
where N
x
j
=
_
2
_
m
i=1
w
k
_
w
k
2
i
_
k
x
j
_
. (8)
The equilibrium equations and sensitivities of ith eigenvalue
are dened by
i
x
j
= {
i
}
T
_
[K]
x
j
i
[M]
x
j
_
{
i
}, (9)
where
[K] =
n
j=1
_
E
min
+x
p
j
E
_
[K
j
], (10)
[M] =
n
j=1
_
E
min
+x
1
j
E
_
[M
j
], (11)
where [K] and [M] are the system stiffness and mass matrices
respectively;
i
is the ith eigenvalue of the structure; {
i
} is the
ith orthonormal eigenvector.
Topology optimization problem for eigenvalues will involve
a multi-model eigenvalue (repeated eigenvalues) solution, and
the mean eigenvalue formulation as Eq. (7) cannot prevent re-
peated eigenvalues. Repeated eigenvalues generally are not dif-
ferentiable with respect to the design variables. This will affect
the stability of optimization process. The sensitivity analysis for
repeated eigenvalue problems is complex and it is actually out
of the scope of this paper, and the details can be obtained in the
work of Seyranian et al. [22] and Krog and Olhoff [13].
4. Multi-objective topological optimization of aerodynamic
missiles
Sections 2 and 3 have described multi-stiffness and mean
eigenvalue topological optimizations. In this section we will
discuss the multi-objective topology optimization with static
and dynamic objectives by using tolerance multilevel sequence
approach. The general procedure for the multilevel sequence
approach is rst to divide all objectives into r groups according
to their priority in optimal solution. The rst group has p
1
ob-
jectives, the second group has p
2
objectives, and so on. Finally,
the rth group has p
r
objectives. Also they satisfy the following
equation: p
1
+p
2
+ +p
r
= k. The order of priority levels
p
1
>p
2
> >p
r
are sorted according to their importance in
Pareto solution. p
1
has the highest priority level and the objec-
tive in group one should be satised rst and next the p
2
and so
on.
With multilevel sequence algorithm, if the optimal solu-
tion of the lth level is unique, then the subsequent levels,
(l + 1), (l + 2), (l + 3), . . . , k, are not necessarily to be solved
continuously. The reason is that levels from 1 to l will decide
the Pareto solution and the contributions of latter levels will
be ignored. Thus the optimization process will stop and the
whole solving procedure will be discontinuous. To avoid this
discontinuity, tolerance parameters
i
>0 (i = 1, 2, . . . , k 1)
have introduced into the multilevel sequence algorithm. The
368 Z. Luo et al. / Aerospace Science and Technology 10 (2006) 364373
so-called tolerance multilevel sequence approach will be for-
mulated. By this way, we relax the optimal solution of each
level according to the prescribed tolerance parameters. There-
fore, the searching area for optimal point will be extended with
respect to subsequent sub-optimization problems. Regarding to
the sub-problem on certain level, it is more preferable to limit
the previous optimal solution within a neighboring scope rather
than restrict the previous level strictly to meet its optimal so-
lution. Resultantly, the discontinuity of optimization process
might be prevented, and it is possible to obtain whole Pareto
set.
In this paper the tolerance multilevel sequence algorithm is
employed as a multi-objective programming scheme for solv-
ing multi-objective topology optimization of the aerodynamic
missile. The free vibrating problem is solved in the rst level,
and then the multi-stiffness optimization problem is considered
in the second level where the relaxed optimal result of rst level
is posed as an external constraint.
(1) The formulation of the rst level for mean eigenvalue is
dened as:
Maximize
X={x
1
,x
2
,...,x
n
}
T
:
Subject to:
[]
i
i
, i = 1, 2, . . . , N
n
j=1
V
j
x
j
V 0, j = 1, 2, . . . , n,
0 <x
min
x
j
<1, j = 1, 2, . . . , n,
([K]
i
[M]){
i
} = {0},
(12)
where = 0.95; is an articial variable. This formulation can
ensure that the second eigenvalue is larger than the rst one by
some percent, and the third one is larger than the second one
by some percent, and so on. Pederson [19] employed a simi-
lar formulation to eigenvalue topology optimization. The given
computational formulation can prevent the repeated eigenval-
ues as well as the oscillation of the objectives.
After the optimal solution
Minimize
X=(x
1
,x
2
,...,x
n
)
T
: C(X) =
_
m
k=1
w
q
k
_
C
k
(X)C
min
k
C
max
k
C
min
k
_
q
_ 1
q
,
Subject to:
h
1
k=1
_
n
j=1
V
j
x
k
j
_
V 0, j = 1, 2, . . . , n,
0 <x
min
x
j
<1, j = 1, 2, . . . , n,
, i = 1, 2, . . . , N
,
Equilibrium equations,
(13)
where