Anda di halaman 1dari 16

Research paper

DOI 10.1007/s00158-003-0362-z
Struct Multidisc Optim 27, 2742 (2004)
Conceptual design of aeroelastic structures
by topology optimization
K. Maute and M. Allen
Abstract A topology optimization methodology is pre-
sented for the conceptual design of aeroelastic structures
accounting for the uidstructure interaction. The ge-
ometrical layout of the internal structure, such as the
layout of stieners in a wing, is optimized by material
topology optimization. The topology of the wet surface,
that is, the uidstructure interface, is not varied. The
key components of the proposed methodology are a Se-
quential Augmented Lagrangian method for solving the
resulting large-scale parameter optimization problem,
a staggered procedure for computing the steady-state so-
lution of the underlying nonlinear aeroelastic analysis
problem, and an analytical adjoint method for evalu-
ating the coupled aeroelastic sensitivities. The uid
structure interaction problem is modeled by a three-eld
formulation that couples the structural displacements,
the ow eld, and the motion of the uid mesh. The
structural response is simulated by a three-dimensional
nite element method, and the aerodynamic loads are
predicted by a three-dimensional nite volume discretiza-
tion of a nonlinear Euler ow. The proposed method-
ology is illustrated by the conceptual design of wing
structures. The optimization results show the signi-
cant inuence of the design dependency of the loads
on the optimal layout of exible structures when com-
pared with results that assume a constant aerodynamic
load.
Key words topology optimization, aeroelasticity, uid
structure interaction, Sequential Augmented Lagrangian
method, coupled adjoint sensitivity analysis
Received: 13 October 2002
Revised manuscript received: 27 June 2003
Published online: 30 January 2004
Springer-Verlag 2004
K. Maute
u
and M. Allen
Center for Aerospace Structures, Campus Box 429, University
of Colorado at Boulder, Boulder, CO 80309, USA
e-mail: kurt.maute@colorado.edu,
maute@pegasus.colorado.edu
1
Introduction
The performance of numerous structural systems is dom-
inated by coupling eects resulting from the interaction
between the structural deformations and one or multiple
other physical elds, such as ow, electrostatic, or elec-
tromagnetic elds. For example, the aerodynamic pres-
sure acting on aircraft may strongly depend on the uid
structure interaction. Electrostatic-mechanical coupling
is often used to actuate microsystems. For the design of
such systems, coupling eects need to be taken into ac-
count to avoid failure due to instability phenomena, such
as aeroelastic divergence and utter, and to optimize for
performance.
This study will focus on the optimization of aeroe-
lastic structures, as it is relevant to the design of aero-
nautical systems and components, such as wings and ro-
tor blades. Traditionally, the aeroelastic design of such
systems has been split up into two consecutive tasks:
(1) optimizing the aerodynamic shape assuming a rigid
structure, and (2) nding the internal structural lay-
out and material composition such that the deformed
shape due to the aerodynamic loads of the conguration
found in task (1) is identical to the optimum aerodynamic
shape.
The advantage of this so-called jig-shape approach
is that aerodynamic and structural analysis and design
are decoupled. Aerodynamic shape optimization and the
design of the internal structure can be performed by
methods tailored to each subproblem. However, static
and dynamic instability eects, such as divergence and
utter, cannot be directly considered as the jig shape
is unrelated to the calculations of these instability ef-
fects, which are pure coupling phenomena. Another ma-
jor drawback of the jig-shape approach is that, in general,
there is no exact solution for the second step; one can-
not guarantee that there exists a structural layout whose
deformed shape is identical to the aerodynamically opti-
mum shape. Even if such a structural layout exists, the
mass required for this design may lead to dierent lift
requirements. This in turn may change the constraints
28
for the aerodynamic optimization step (1) and may lead
to a dierent aerodynamic shape. Consequently, the jig-
shape approach leads to an iterative design procedure and
may produce non-optimal results.
Therefore, aeroelastic optimization procedures have
been developed in which structural parameters, such as
the thickness of stieners and the orientation of bers of
composite materials, are used to directly optimize the de-
formed aerodynamic shape. These approaches allow for
the simultaneous consideration of structural design crite-
ria, such as stresses and displacements, and aerodynamic
criteria, such as lift and drag. However, these design pro-
cedures require the prediction of the coupled aeroelastic
response of the system.
Initially, linear ow theories and simplied structural
models were applied for predicting the aerodynamic loads
in the aeroelastic design procedure, for example, by Haft-
ka (1986), Bowman et al. (1989), Friedmann (1991), and
Barthelemy et al. (1994). Recently, computational aeroe-
lastic optimization procedures have been presented that
allow for detailed linear and non-linear structural nite
element models and non-linear ow theories, such as Eu-
ler and NavierStokes ow discretized by nite volume
or nite element methods. These high-delity optimiza-
tion procedures have been applied to the shape optimiza-
tion of three-dimensional exible wings with and without
varying the thickness of internal ribs and stieners, for
example, by Hou and Satyanarayana (2000), Gumbert
et al. (2001), and Maute et al. (2001). These optimiza-
tion problems require a well-dened initial design and are
characterized by a moderate number of optimization vari-
ables (< 50).
In this paper a new approach, illustrated in Fig. 1,
is introduced for the conceptual design of the internal
layout of aeroelastic structures. At the beginning of the
optimization procedure only the wet shape of the unde-
formed structure is given and the so-called design domain
for the internal structure is dened. Areas in which no
structure should be generated can be additionally de-
ned. The goal of the optimization process is to determine
the structural layout that yields a deformed shape that is,
for example, aerodynamically optimal and satises given
structural design constraints. While no restrictions are
imposed on the internal structural layout, for practical
purposes, the topology of the wet shape is not subject to
optimization; that is, no holes can be inserted into the
skin.
Fig. 1 Aeroelastic design by topology optimization
The proposed approach combines material topology
optimization and recent developments in computational
aeroelastic optimization. It allows for the exploration of
conceptually novel designs and leads to new algorithmic
and computational challenges. The novel aspects of this
approach can be summarized as follows:
Topology optimization is extended to uidstructure
interaction problems.
The proposed approach allows for the optimization
of aeroelastic structures without requiring a close-to-
optimal design.
Topology optimization has mostly focused on displace-
ment-related optimization criteria for which robust
and ecient optimization algorithms have been de-
veloped. Considering aerodynamic optimization cri-
teria, such as lift and drag, requires non-standard
optimization algorithms, which have not been applied
to topology optimization so far.
In contrast to aeroelastic optimization problems ad-
dressed so far, which are characterized by a small
number of optimization variables (< 50), material
topology optimization leads typically to thousands
of optimization variables. Due to the size of the
topology optimization problem, established aeroelas-
tic optimization procedures are not ecient or even
applicable.
To address these challenges a Sequential Augmented
Lagrange (SAL) method is introduced, which allows the
solution of large aeroelastic topology optimization prob-
lems, and an adjoint method is applied for computing the
coupled aeroelastic sensitivities.
Preliminary results on aeroelastic topology optimiza-
tion were rst presented by Maute et al. (2002) and
Maute (2002). In this paper a comprehensive descrip-
tion of the theoretical background and the computational
methods for aeroelastic topology optimization is given.
The proposed approach is veried by numerical examples,
which may serve as benchmarks for further studies.
The rest of this paper is organized as follows: in Sect. 2
the formulation of material topology optimization, on
which the proposed approach is based, is presented, and
recent developments in topology optimization for coupled
multi-physics problems are summarized. In Sect. 3 the
Sequential Augmented Lagrangian method, which drives
the aeroelastic optimization process, is outlined. In order
to keep this paper as self-contained as possible, the three-
eld formulation of the underlying aeroelastic problem
and the method for computing the aeroelastic steady-
state response are summarized in Sect. 4. The adjoint
method for computing the coupled aeroelastic sensitivi-
ties is presented in Sect. 5. The potential of the proposed
method is illustrated by the layout optimization of three-
dimensional exible wings with respect to weight, sti-
ness, stress, and aerodynamic lift and drag in Sect. 6.
Finally, in Sect. 7, the key features of the proposed ap-
proach are summarized, and the future developments are
discussed.
29
2
Material topology optimization
The proposed aeroelastic optimization method is based
on the concept of material topology optimization, den-
ing the body of the structure by its material distribution
in a given design domain
d
(see Fig. 2):
(x) =
_
_
_
1 x
m
0 x
d
/
m
. (1)
The geometry of the structure is dened by the set
m
of material points (x) in the design domain
d
. In order
to allow for arbitrary structural layouts, the indicator
function (x) is a function of L

(
d
). The topology op-
timization problem is mapped into a discrete material
distribution problem to nd the indicator function

(x)
that minimizes an objective (or objectives) subject to
constraints.
In general, this problem cannot be solved analyti-
cally, but a nite element approach is applied for dis-
cretizing the material distribution (see Fig. 2b). How-
ever, this formulation leads to an ill-posed integer op-
timization problem that can be regularized by intro-
ducing porous material models. Parameters dening the
properties of the porous material serve as optimization
variables. From a mathematical point of view, porous
material models with extreme properties are necessary
to obtain a well-posed optimization problem. However,
these material models lead to optimized distributions
with a large amount of porous material, which can
hardly be converted into structural layouts that can be
manufactured.
Therefore, the so-called SIMP (Solid Isotropic Mate-
rial with Penalization) approach has become popular in
the engineering optimization community and is applied in
this study. Originally, the SIMP approach was introduced
by Bendse (1989) and Rozvany et al. (1992) for maximiz-
ing the structural stiness while prescribing the mass in
the design domain. It relates the Youngs modulus E to
the density as follows:
E() =
_

0
_

E
0
, 3 , (2)
where the subscript 0 denotes the properties of the bulk
material. The nonlinearity of the SIMP model implicitly
penalizes intermediate densities and leads approximately
to a 01 material distribution in the design domain.
Typically, the density of each nite element in the design
domain is treated as an optimization variable.
Since the results of the SIMP approach depend on
the orientation and renement of the mesh, the formula-
tion of the optimization problem needs to be augmented
by additional regularization methods. Examples of such
methods are the perimeter method by Ambrosio and But-
tazzo (1993) and Haber et al. (1996), the slope control
approach by Petersson and Sigmund (1998) and Zhou
Fig. 2 Material topology optimization
et al. (2001), or the ltering method by Sigmund (1994)
and Sigmund and Petersson (1998). The latter method
restricts the discontinuity of the optimized material dis-
tribution by smoothing the gradients of the objective
and/or constraints, which translates into a restriction
of the minimum size of optimal structural members.
In addition, the ltering technique eliminates checker-
board modes (Sigmund 1994). In this study a modi-
ed ltering method is employed that compensates for
the errors due to gradient smoothing and therefore im-
proves the convergence in SAL iterations. This ltering
method is embedded into a continuation method, which
serves the purpose of successively eliminating fuzziness in
the material distribution introduced by the ltering ap-
proach. This procedure was introduced by Maute et al.
(2002).
For a comprehensive survey of structural topology
optimization, the reader is referred to Bendse (1999),
Olho and Eschenauer (1999), Rozvany et al. (1995),
Rozvany (1997), Rozvany and Olho (2001), and Es-
chenauer and Olho (2001). Of relevance to the prob-
lem of optimizing the layout of exible aeroelastic struc-
tures are the studies on topology optimization with de-
sign dependent loads by Hammer and Olho (2000) and
Chen and Kikuchi (2001). In these approaches, however,
the loads depend on the initial structural geometry but
not on the structural response. Eschenauer et al. (1998)
have applied the so-called bubble-method to the con-
ceptual layout of airfoils but also neglected the depen-
dency of aerodynamic loads on the structural design and
deformations.
So far, topology optimization techniques have mainly
been applied to problems in structural mechanics. Only
recently, topology optimization has been extended to
coupled systems, such as thermo-mechanical, electro-
thermo-mechanical, and piezoelectric problems by Ro-
driques and Fernandes (1995), Silva et al. (1997), Sig-
mund (1998, 2001a,b), and Sigmund et al. (1998). Pre-
liminary results on topology optimization applied to
fully coupled uidstructure-interaction problems were
recently presented by Maute et al. (2002) and Maute
(2002).
30
3
Sequential Lagrangian method
The relaxed material formulation of the aeroelastic top-
ology optimization problem leads to smooth, nonlinear
constrained optimization problems that can be eciently
solved by gradient-based optimization algorithms. In ma-
terial topology optimization the resolution of the opti-
mum structural layout is higher the ner the design do-
main is discretized, which typically leads to a large num-
ber of optimization variables. In particular, for three-
dimensional topology optimization problems, the number
of variables may easily exceed 10
5
. Because of the size
of the optimization problem, optimality criteria methods
tailored to the specic optimization problem have tradi-
tionally been applied (Rozvany 1989; Bendse 1989). As
robust and ecient as these methods are for the classical
stinessmass topology optimization problem and linear
elastic structural response, they are often not applicable
to more general problems with dierent optimization cri-
teria and/or nonlinear structural response.
For complex topology optimization problems, such as
problems with stress constraints, eigenfrequencies, the
design of micro-structured materials, and problems con-
sidering coupled electro-thermal-mechanical response,
Sequential Linear Programming (SLP) methods have
been advocated by Tenek and Hagiwara (1993) and Sig-
mund et al. (1998), and Sequential Convex Program-
ming (SCP) methods by Ma et al. (1995) and Duys-
inx and Bendse (1997). SCP methods construct a se-
ries of convex subproblems to approximate the ori-
ginal optimization problem by explicit, separable func-
tions. The explicit subproblems can then be eciently
solved, for example, by a dual method. In particular,
for topology optimization of coupled problems, Sigmund
(2001a,b) proposed the Method of Moving Asymptotes
(MMA), which belongs to the class of SCP methods,
but he reported on very large numbers of iterations
( 1000) in order to obtain a well-converged mate-
rial distribution. In addition, Maute (2002) experienced
that the MMA method is not well suited and may
not even be applicable to highly nonlinear optimization
problems, which are characterized by a strong second-
order coupling of the optimization variables in the op-
timization criteria. In this case the optimization prob-
lem cannot be suciently approximated by separable
functions.
Therefore, a limited-memory quasi-Newton method,
embedded into a Sequential Augmented Lagrangian (SAL)
approach, is proposed for solving the aeroelastic topology
optimization problems. The SAL approach, often also
referred to as augmented Lagrangian, constructs approxi-
mated bound-constrained subproblems that are solved by
a limited-memory quasi-Newton method, which accounts
for second-order coupling of the optimization variables.
In the SAL approach all constraints other than simple
bounds are included in an augmented Lagrangian func-
tion L
a
, which is minimized with respect to the optimiza-
tion variables s, subject to the simple lower and upper
bounds s
L
and s
U
, respectively:
min
s
L
a
= z +
T
h+
t
(gv) +
1
2
r
_
h
T
h+(gv)
T
(gv)
_
(3)
subject to
s
L
s s
U
, v ={v R
ng
|v 000} , (4)
R
n
h
, ={ R
ng
| 000} , (5)
where z is the objective of interest, h the vector of n
h
equality constraints, and g the vector of n
g
inequality
constraints. The associated vectors of Lagrange multipli-
ers are denoted by and , and r is a penalty factor. The
inequality constraints g 000 are transformed into equal-
ity constraints via non-negative slack variables v, which
can then be explicitly eliminated. For a given set of La-
grange multipliers and the penalty factor {, , r}, the
simply bounded optimization problem is minimized with
respect to the optimization variables s. In an outer loop
the Lagrange multipliers {, } and the penalty factor r
are updated such that for r the original optimiza-
tion problem is solved.
In this paper the globally convergent procedure of
Conn et al. (1991) is employed for updating the Lagrange
multipliers, and the local minimization problems are
solved by the quasi-Newton limited-memory algorithm of
Byrd et al. (1995). For robustness purposes, the step size
is restricted in the local minimization problems.
4
Aeroelastic analysis
The proposed topology optimization method allows for
the design of the internal layout of aeroelastic struc-
tures such that structural and aerodynamic criteria can
be optimized simultaneously. In order to obtain reliable
optimization results, the aerodynamic response of the
wing should be predicted as accurately as possible. In
this study a high-delity aeroelastic simulation method
is employed for predicting the aeroelastic steady-state re-
sponse within the proposed optimization methodology.
The steady-state response of the aeroelastic system
is modeled by the three-eld formulation of Farhat et al.
(1995), which allows for potentially large structural defor-
mations. The discretized form of the aeroelastic problem
can be written as follows:
G =
_

_
S(s, u, x, w)
D(s, x)
F(s, w, x)
_

_
= 000 , (6)
where S, D, and F denote the discrete equations of equi-
librium governing respectively the structure, the motion
31
of the computational uid grid, and the uid system; u is
the vector of generalized structural displacements, x de-
notes the displacement vector of the uid grid points, and
w is the uid state vector.
The structural, ow, and uid grid motion equations
are coupled through the following interface boundary
conditions:
n =p
a
(w, x) on
FS
, (7)
x =u on
FS
, (8)
where is the structural stress tensor, and n is the
normal vector on the structureuid interface
FS
. The
aerodynamic loads p
a
acting on the wet surface
FS
de-
pend on the uid state w and the geometry of the uid
mesh x. The resulting deformations u on
FS
are Dirich-
let boundary conditions for the uid mesh motion equa-
tions. The positions of the grid points in the uid domain
are updated by solving a ctitious structural problem in
which the uid mesh is treated as an elastic medium. The
ow solution is computed on the deformed mesh.
In this study the structural response is modeled by
a linear elastic nite element formulation:
S : Kuf
a
(x, w) f = 0 , (9)
where Kis the structural stiness matrix, f
a
is the vector
of aerodynamic forces, and f is the vector of other forces
acting on the structure. The Euler ow is approximated
by a second-order nite volume scheme based on the Ar-
bitrary Lagrangian Eulerian (ALE) form of the Roe ux
(Roe 1981):
F : F
2
(x, w) = 0 , (10)
where the subscript 2 emphasizes that the vector of
ALE convective uxes F is accurate to second-order in
space. The motion of the uid grid points is computed by
the improved torsional spring analogy method of Farhat
et al. (1998a):
D :
_
_
K

F
K

K
t

_
_
. .
K
_
_
x

F
x

_
_
=
_
_
0

f
_
_
,
(11)
with x

=Pu on
FS
, (12)
where K is a ctitious stiness matrix associated with
the computational uid grid, the subscript
F
designates
the uid grid points lying inside the computational uid
domain
F
, the subscript
FS
denotes those lying on
the uidstructure interface
FS
, and

f is the ctitious
reaction force needed to enforce the Dirichlet boundary
conditions (12). The matrix P projects the structural
displacements u onto
FS
and accounts for potentially
non-matching uid and structure meshes while conserv-
ing energy and momentum (Farhat et al. 1998b).
The aeroelastic governing equations specic to the
models employed in this study can be written as
_

_
S
_
_
D

F
D

_
_
F
_

_
=
_

_
Kuf
a
f
_
_
K

F
x

F
+K

P u
x

Pu
_
_
F
2
_

_
.
(13)
If n
u
denotes the number of structural degrees of freedom
(DOFs), then K is an n
u
n
u
symmetric positive de-
nite matrix. The number of uid mesh motion DOFs is
n
v
=3n
fg
, where n
fg
is the number of uid grid points,
and K is an n
v
n
v
symmetric positive-denite matrix.
For an Euler ow each grid point has ve uid DOFs,
leading to an overall number of DOFs n
w
= 5n
fg
. F
2
is an n
w
-long vector.
4.1
Staggered solution procedure
The aeroelastic governing equations (13) form a system
of nonlinear equations that is stymied by a large number
of unknowns for practical applications. Each of the three
components of (13) has dierent mathematical and nu-
merical properties, well-established and distinct optimal
solution algorithms, and dierent data structures. There-
fore, staggered solution procedures also called parti-
tioned and segregated procedures have prevailed for
solving steady and unsteady aeroelastic problems (Str-
ganac and Mook 1990; Pramono and Weeratunga 1994;
Piperno et al. 1995; Lesoinne and Farhat 1998).
In this study an algorithm is applied for solving the
steady-state equations (13) that merges a quasi-Newton
approach for building linear subproblems and a block
GaussSeidel scheme for solving these subproblems. Lin-
earizing the system of nonlinear equations (13) around
a given conguration {u, x, w}
n
yields the following sys-
tem of linear equations with the superscript n denoting
the n-th iteration:
_

_
S
u
S
x
S
w
D
u
D
x
000
000
F
x
F
w
_

_
n
. .
A
n
_

_
u
_

_
x

F
x

_
w
_

_
n
=
_

_
S
_

_
D

F
D

_
F
_

_
n
,
(14)
where the linearized operator A
n
is more specically
32
Table 1 Staggered algorithm for computing the aeroelastic steady-state response
1. Initialize u
(0)
, update the uid mesh x
(0)
by solving (11), and compute the steady-state ow w
(0)
around the structure with the displacement eld u
(0)
. Set w
(0)
= 0.
2. Given the previously computed aerodynamic forces f
a
(n)
s
, solve
K u
(n+1)
=f
a
(n)
+f (16) ,
and for the sake of numerical stability, update u
(n+1)
using the following under-relaxation scheme:
u
(n+1)
=(1) u
(n)
+ u
(n+1)
with < 1.0 (17) .
3. Transfer the structural displacements to the uid grid and update the position of this dynamic grid,
set x
(n+1)

=Pu
(n+1)
, and solve K
F F
x
(n+1)
F
= K
F
Pu
(n+1)
(18) .
4. Evaluate the new uid state vector w
(n+1)
by computing
w
(n+1)
= w
(n)
+w
(n+1)
, where w
(n+1)
is the solution of H
(n)
w
(n+1)
= F
(n)
2
(19) .
5. Compute the aerodynamic pressure force f
a
(n+1)
f
in the uid mesh and project it onto the structure
f
a
(n+1)
s
=P
t
f
a
(n+1)
f

w
(n+1)
, x
(n+1)

(20) .
6. Check convergence by inspecting the following criteria:

p, u
(n+1)
, x
(n+1)
, w
(n+1)

2

AA

p, u
(0)
, x
(0)
, w
(0)

2
(21) ,

p, x
(n)
, w
(n)

AA

p, x
(0)
, w
(0)

2
(22) ,
where
AA
is a specied tolerance, and the superscript AA stands for Aeroelastic Analysis. If convergence
is not reached, extract from w
(n+1)
the uid pressure on , compute f
a
(n+1)
, and go back to step 2.
A
n
=
_

_
K
_

f
a
x

f
a
x

_

f
a
w
_
_
K

P
P
_
_
_
_
K

F
000
000 I
_
_
_
000
000
_
000
_
F
2
x

F
F
2
x

_
H
_

_
.
(15)
In each quasi-Newton-like iteration an approximate
Jacobian is built by dropping the o-diagonal terms in
A
n
. The resulting linear system is approximately solved
by only one block GaussSeidel step. The algorithm is
summarized in Table 1 and described in detail by Maute
et al. (2001, 2003).
In this paper the structural subproblem (9) is solved
by a sparse direct method, the uid mesh motion sub-
problem (18) by the Jacobi-preconditioned conjugate
gradient algorithm (PCG), and the linearized ow prob-
lem (19) by the Generalized Minimum Residual (GM-
RES) algorithm preconditioned by the Restricted Addi-
tive Schwarz (RAS) method of Cai et al. (1998).
5
Adjoint aeroelastic sensitivity analysis
Of crucial importance for the eciency and eventually
for the applicability of topology optimization to realis-
tic problems is the computation of the gradients of the
optimization criteria with respect to the optimization
variables, that is, the sensitivity analysis. For the classi-
cal stinessmass problem in topology optimization, the
gradients of the strain energy with respect to material
parameters are self-adjoint and therefore can be com-
puted without noticeable additional eort (Haug et al.
1986). In general, computing the gradients of the objec-
tive and constraints leads to linear problems of the size of
the underlying analysis problem. Since in topology opti-
mization the number of optimization variables is typically
several orders of magnitudes larger than the number of
constraints, the adjoint approach for computing the sen-
sitivities is the method of choice.
A general framework for the sensitivity analysis of
coupled systems has been introduced by Sobieszczanski-
Sobieski (1990), and sensitivity analysis methods have
been studied for coupled problems in the case of thermo-
elasticity by Meric (1985, 1986) and Tortorelli et al.
(1991), thermo-elastoplasticity by Michaleris et al. (1995),
and magnetohydrodynamics by Meric (1990), among
others. In particular, relevant to this study are recent
approaches for the sensitivity analysis of nonlinear aeroe-
lastic problems by Ghattas and Li (1998), Mller and
Lund (2000), Hou and Satyanarayana (2000), Maute
et al. (2001), and Lund et al. (2002). While these methods
follow the direct approach, and therefore are of limited
use for aeroelastic topology optimization, Maute et al.
(2000, 2003) and Martins and Alonso (2002) have re-
33
Table 2 Staggered algorithm for computing the aeroelastic adjoint solution
1. Compute the steady-state aeroelastic solution {u, x, w}
eq
and set a
(0)
x
= 0.
2. Given the previously computed adjoint mesh motion state a
(n)
x
, solve
K a
(n+1)
u
=
q
u
P
t

K
t
F
a
(n)
xF
a
(n)
x

(28) ,
where q/u is computed analytically or by automatic dierentiation, and for the sake of
numerical stability, update a
(n+1)
u
using the following under-relaxation scheme:
a
(n+1)
u
= (1)a
(n)
u
+ a
(n+1)
u
with <1.0 (29) .
3. Compute the adjoint uid state a
(n+1)
w
by solving
H
t
2
a
(n+1)
w
=
q
w
+
f
at
w
a
(n+1)
u
(30) ,
where q/w is computed analytically or by automatic dierentiation.
4. Compute the adjoint mesh motion state a
(n+1)
x
F
in the computational uid domain
F
by solving
K
F F
a
(n+1)
x
F
=
q
x
F
+
f
at
x
F
a
(n+1)
u

F
t
2
x
F
a
(n+1)
w
in
F
(31) ,
and compute a
(n+1)
x
on the interface as follows:
a
(n+1)
x
=
q
x

+
f
at
x

a
(n+1)
u

F
t
2
x

a
(n+1)
w
on (32) ,
where q/x is computed analytically or by automatic dierentiation.
5. Check convergence by inspecting the relative norm of the residual of the adjoint equation
Ra
(n+1)
/ Ra
(0)

SA
(33) ,
where
SA
is a specied tolerance, and the superscript SA stands for Sensitivity Analysis.
If convergence is not reached, go back to Step 2.
cently presented an adjoint formulation of the aeroelastic
sensitivities.
Since the adjoint approach presented herein is a cru-
cial ingredient of the proposed topology optimization
method, the essential formulations and computational is-
sues of the adjoint approach based on the three-eld for-
mulation are summarized below. The derivative of the
optimization criterion q
j
, that is, the objective or a con-
straint, with respect to the optimization variable s
i
can
be computed as follows:
dq
j
ds
i
=
q
j
s
i
+
q
j
u
T
du
ds
i
+
q
j
x
T
dx
ds
i
+
q
j
w
T
dw
ds
i
,
(23)
where the derivatives du/ds
i
, dx/ds
i
, and dw/ds
i
of the
aeroelastic response are governed by a system of linear
equations, which is obtained by dierentiating the gov-
erning equations (6). This yields the following coupled
system of linear equations:
A
eq
_

_
du
ds
i
dx
ds
i
dw
ds
i
_

_
=
_

_
S
s
i
D
s
i
F
s
i
_

_
eq
,
(24)
where the superscript eq emphasizes that these quanti-
ties are evaluated at the aeroelastic equilibrium. For the
model employed in this study the transpose of the aeroe-
lastic Jacobian A
eq
takes on the following form:
A
eq
t
=
_

_
K
_
P
t
K
t

P
t
_
000
_

f
at
x

f
at
x

_
_

_
K

F
000
000 I
_

_
_

_
F
t
2
x

F
F
t
2
x

f
at
w
[ 000 000 ] H
t
2
_

_
.
(25)
Pursuant to the adjoint approach, the derivatives of the
optimization criterion dq
j
/ds
i
are evaluated in the follow-
ing order:
dq
j
ds
i
=
q
j
s
i

_
_
_
_
_
_
_
_
_
A
eq
t
_

_
q
j
u
q
j
x
q
j
w
_

_
eq_
_
_
_
_
_
_
_
_
t
. .
a
t
_

_
S
s
i
D
s
i
F
s
i
_

_
eq
,
(26)
34
where a is the solution of the adjoint system, which needs
to be solved only once for each optimization criterion q
j
.
The associated system of linear equations can be written
as follows, with R
a
being the residual of the aeroelastic
adjoint problem:
R
a
=A
eq
t
_

_
a
u
_
a
x

F
a
x

_
a
w
_

_
q
u
_

_
q
x

F
q
x

_
q
w
_

_
eq
=000 .
(27)
For elliptic problems with a symmetric Jacobian, such
as elastic and heat conduction problems, computing the
adjoint response bears the same computational com-
plexity as solving the linearized governing equations
because A
t
=A. In contrast to these problems, the aeroe-
lastic Jacobian A
eq
is non-symmetric and, in particu-
lar, the evaluation of the sub-matrices (F/x)
T
and
(F/w)
T
deserves special attention. The reader is re-
ferred to Maute et al. (2003) for a thorough discussion of
the computational complexity of the adjoint aeroelastic
operator A
eq
t
.
For problems with a moderate number of degrees of
freedom (< 10
5
), which are, for example, typical in struc-
tural optimization, it is common practice to factorize the
transpose of the Jacobian once and store the factorized
matrix. So, computing the adjoint response requires only
a forward-backward substitution. However, the modeling
of realistic aeroelastic problems leads to a large number
of degrees of freedom, in particular for approximating the
oweld. Therefore, the adjoint solution needs to be com-
puted by iterative methods, making the computational
eort comparable to computing the aeroelastic steady-
state response.
In this study the adjoint system is solved by another
staggered procedure, which is summarized in Table 2 and
described in detail by Maute et al. (2003). For accuracy
and eciency reasons, all partial derivatives with respect
Fig. 3 Structural model and design domain of 3-D wing
to the optimization variables are computed analytically
as opposed to nite dierencing. The numerical eciency
of the algorithm can be increased in the case of top-
ology optimization by exploiting the fact that the partial
derivatives D/s
i
and F/s
i
of the uid mesh motion
equations and the ow equations, respectively, vanish in
(26). For solving the linear subproblems (28), (30), and
(31) the same algorithms are used as in the staggered
procedure for computing the aeroelastic response (see
Sect. 4.1).
6
Numerical examples
The potential of the proposed method is veried with the
stiener layout optimization of three-dimensional wings.
In the rst example, the structure is modeled by a three-
layer plate model, and the design domain is formed by
the upper and lower layers. The second example begins
with a general rib and spar internal structure, and top-
ology optimization is used to further rene the layout of
the internal structure.
For both problems, the computational methodology
described in the previous sections is employed. All com-
putations are performed on a 9-node PC LINUX cluster.
Each node is equipped with a 1.7-GHz P4 processor and
1 GB of RDRAM memory. The nodes are connected by
100-MBit fast Ethernet.
6.1
Layered plate model
The purpose of the rst example problem is twofold. The
rst objective is to demonstrate the importance of ac-
counting for aeroelastic loading versus constant loading.
The second purpose is to illustrate the ability of the pro-
posed methodology to develop a conceptual layout of
stieners, which ccan then be interpreted as spars and
ribs.
For the aeroelastic analysis of practical problems the
structure is often represented by a simplied beam or
35
plate model. Following this common practice in the rst
example, the structure of a three-dimensional backswept
tapered wing is modeled by a plate, which is built up of
three layers and clamped at its root (Fig. 3). The plate
is discretized by 12960 multi-layer three-node ANS plate
elements of Militello and Felippa (1991). In order to re-
duce the mesh-dependency of the three-node elements,
the nite element mesh consists of two overlaying triangu-
lations with cross diagonals. The upper and lower layers
dene the design domain, and the thin middle layer is
not subject to optimization. The material distribution
in the design layers represents the sought-after layout of
stieners.
In order to reveal aeroelastic eects, the thickness of
the middle layer is made articially small compared with
the layers forming the design domain; the thickness ratio
is t
m
/t
t
=0.024, with a total thickness t
t
=0.1025 m. The
material distribution in the upper and lower layers is de-
scribed by 3240 optimization variables, employing the
SIMP model (2) with a exponent of 3.5. In each layer,
the density of a group of four corresponding elements is
treated as one optimization variable leading to an iden-
tical density distribution in the upper and lower layers.
The material properties of the bulk material are: Youngs
modulus E
0
= 1.4810
11
N/m
2
, Poisson ratio = 0.3,
and density
0
= 2.710
3
kg/m
3
.
The wet surface of this wing is formed by extruding
a NACA0012 airfoil along a nite span. The computa-
tional uid domain is discretized by 19384 grid points.
The free stream conditions correspond to a ight at an al-
titude of 10000 m and a Mach number of Ma = 0.8. The
angle of attack is 5

. The uid mesh on the wet surface of


the wing and on the symmetry plane, as well as the pres-
sure contours of the rigid conguration, is shown in Fig. 4.
The results of two optimization problems are com-
pared to study the inuence of aeroelastic eects on
the optimum layout of the stieners. In the rst case,
Fig. 4 Right: uid mesh on wet surface of wing and symmetry wall. Left: pressure contours for undeformed conguration
the mass in the design domain is minimized subject to
lift, drag, and displacement constraints while accounting
for the uidstructure interaction. This case is labeled
FSI. In the second case, the mass is minimized for
a constant aerodynamic pressure, neglecting any uid
structure interaction. This case is labeled CPR. The
aerodynamic pressure in the CPR case is computed as-
suming a rigid structure. The aerodynamic constraints
cannot be taken into account. In addition to the displace-
ment constraints the strain energy is constrained in order
to characterize the overall structural response. The for-
mulations of the two optimization problems are compared
in Table 3.
In the aeroelastic case (FSI) the aerodynamic lift
L is required to be at least 90% of the lift L
rigid
=
23.65 kN of the rigid conguration. The aerodynamic
drag D may not exceed the drag D
rigid
= 1.69 kN of the
rigid conguration by more than 2.5%. The maximum
feasible vertical displacement v
max
is 10% of the span
of the wing, that is, v
max
= 0.3 m. In the simplied de-
sign problem (case CPR), the same displacement con-
straints are applied as in the aeroelastic case. In order
to make both optimization problems comparable, the up-
per bound on the strain energy constraint is set to the
strain energy of the optimal aeroelastic design, that is,

i
max
=
i
opt,FSI
. In both cases, the optimization process
starts from a uniform density distribution in the design
layers of /
0
= 0.50.
The aeroelastic performance of the initial and optimal
congurations for the aeroelastic case (FSI) are compared
in Table 4. The optimal density distribution is shown in
Fig. 5a. In the optimization process, the mass is decreased
to 13.99% with resect to the mass of the initial congu-
ration. At the optimum, the lift and drag constraints, as
well as the displacement constraint at point D, are active.
The reader may note that these constraints are slightly vi-
olated, which is due to the external penalty term of the
36
Table 3 Optimization problems with and without accounting for uidstructure interaction
Case FSI: aeroelastic response Case CPR: constant aerodynamic pressure
min mass min mass
subject to subject to
lift L0.9 L
rigid
0 lift not applicable
drag D1.025 D
rigid
0 drag not applicable
displacement v
C
vmax 0 displacement v
C
vmax 0
displacement v
D
vmax 0 displacement v
D
vmax 0
strain energy
i

i
max
0
Table 4 Case FSI: comparison of mass, lift, drag, and dis-
placements of initial and optimum congurations in aeroelas-
tically deformed state
conguration initial optimized
mass [kg] 817.755 114.441
lift [kN] 22.599 21.090
drag [kN] 1.590 1.751
displacement in C [m] 0.085 0.209
displacement in D [m] 0.082 0.303
Table 5 Case CPR: Comparison of mass, displacements, and
strain energy of initial and optimum congurations for con-
stant aerodynamic pressure
conguration initial optimized
mass [kg] 817.755 90.539
displacement in C [m] 0.047 0.218
displacement in D [m] 0.056 0.300
strain energy [Nm] 194.709 741.354
SAL algorithm (3). The strain energy of the optimal con-
guration is
i
max
= 740.66 Nm =
i
opt,FSI
.
For the constant pressure case (CPR), the initial and
optimum congurations are compared in Table 5. The op-
Fig. 5 Optimized material distribution left: case FSI, right: case CPR
Table 6 Comparison of aeroelastic responses of the opti-
mized congurations for the FSI and CPR cases with respect
to aerodynamic lift and drag
case FSI CPR
lift [kN] 21.090 21.709
drag [kN] 1.751 4.034
timal density distribution is shown in Fig. 5b. In this case,
the mass could be decreased to 11.07%. The displace-
ment constraint at point D and the energy constraint are
active.
The layout and the mass of the stieners are consid-
erably dierent for the FSI and CPR cases. To illustrate
the importance of accounting for the uidstructure in-
teraction for the design of exible wings, the aeroelastic
response of the CPR optimum design is analyzed. The
aeroelastic responses of both designs are compared in
Table 6 with respect to aerodynamic lift and drag. The
deformed aerodynamic shapes and the associated pres-
sure contour plots of both designs are shown in Figs. 6 and
7. While the lift is approximately the same, the drag of
the CPR optimum is more than twice as large as the drag
of the FSI optimum. The reader may note the buckling-
pattern-like deformation at the leading edge of the CPR
optimum. In addition, the shock on the wing of the CPR
37
Fig. 6 Aeroelastic response of FSI optimum deformed conguration: mesh and pressure contours
Fig. 7 Aeroelastic response of CPR optimum deformed conguration: mesh and pressure contours
optimum is stronger than the shock of the FSI optimum.
Both features lead to a larger drag in the CPR case.
Comparing the aeroelastically optimized result with
the result assuming a constant aerodynamic pressure de-
monstrates that uidstructure interaction phenomena
need to be accounted for when determining the internal
structural layout of exible wings by topology optimiza-
tion. The optimization result in Fig. 5 can be interpreted
as a more realistic result in a post-processing step. For
example, the density distribution can guide the engineer
in selecting the number and location of ribs and spars as
shown in Fig. 8.
The authors would also like to point out that the jig-
shape approach, outlined in the introduction, is not af-
icted with the shortcomings of the CPR approach as the
aeroelastic eects are accounted for in an iterative fash-
ion (see Sect. 1). In the above example, the CPRapproach
serves an illustrative purpose only and should not be con-
fused with the jig-shape approach.
6.2
Three-dimensional rib-spar model
In the second example, the proposed topology optimiza-
tion method is applied to the layout optimization of in-
ternal stieners in a 3-D build-up wing. The wing model
is shown in Fig. 9. In contrast to the rst example, the
second example starts with an initial structural layout
of ribs and spars, which is then further optimized. The
goals of the topology optimization procedure are to deter-
mine which ribs and spars are needed and to optimize the
geometry of the remaining structure.
To illustrate the nature of this design problem, an
aeroelastic analysis was performed on the initial design
with no ribs and on the initial design with no spars.
The resulting displacements of the designs are shown in
Fig. 10. The reader may note from Fig. 10a that the main
purpose of the ribs in the structure is to keep the skin in
place. With no ribs in place the skin of the wing undergoes
38
Fig. 8 Practical conceptual design for wing supports
large deformations, and severe aerodynamic penalties re-
sult. The main purpose of the spars is to add stiness to
resist the aerodynamic loads, the importance of which is
illustrated in Fig. 10b. With no spars in place the wing
structure becomes too exible to support the aerody-
namic loads suciently, causing unrealistic deformations
and increases in the stress on the skin. The ability of the
proposed methodology to account for both of these fac-
tors is illustrated by the following optimization problem.
The design domain is the initial layout of the struc-
tural supports, two spars and twenty evenly spaced ribs.
The design domain is covered by a layer of shell elements
that represents the skin of the wing and is not subject to
optimization.
The shape of the wing is formed by extruding a NACA
2412 airfoil along a nite span. The design domain is
discretized by 25856 multi-layer three-node ANS plate
elements (Fig. 9b). Again, in order to reduce the mesh-
dependency of the three-node elements, the nite elem-
ent mesh consists of two overlaying triangulations with
cross diagonals (see example in Sect. 6.1). The skin of
the wing is represented by 15424 three-node shell elem-
ents (Fig. 9c). The wing is translationally clamped at
the internal supports of the root of the wing. The ma-
terial distribution in the design domain is described by
6464 optimization variables, which represent the density
of a group of four corresponding elements. The material
properties in each group are variable, pursuant to the
SIMP approach (2), and the exponent is set to 3.0.
The material properties of the bulk material in the de-
sign domain, and that of the skin, are: Youngs modulus
E
0
= 7.410
10
N/m
2
, Poisson ratio = 0.3, and dens-
ity
0
=2.710
3
kg/m
3
. The spar member elements have
a thickness of 0.02 m, and the ribs and skin each have
a thickness of t
rib
= t
skin
= 0.0025 m. The uid domain
is discretized by 28740 grid points (Fig. 9d). The free-
stream ow conditions correspond to a ight at Mach 0.8
and an angle of attack of 5.0

at an altitude of 10000 m.
The objective of the optimization problem is to min-
imize the mass of the wing subject to constraints on the
Fig. 9 Ground plan and dimensions, design domain of struc-
ture, wing surface, and uid mesh on wet surface
39
Fig. 10 Aeroelastic divergence: large local deformations
Fig. 11 Optimum material distribution: elements with relative density /
0
0.50 (light) and /
0
0.75 (dark)
Table 7 Optimization problem for 3-D aeroelastic layout
optimization
objective min mass
subject to
lift
L+g(mm
initial
)
L
initial
1.0 0
lift/drag
(L/D)
(L/D)
initial
1.0 0
skin stress
max

initial
max
+1.0 0
aerodynamic lift, the liftdrag ratio, and the maximum
von Mises stress in the skin. The lift may only decrease as
the weight of the wing decreases, the liftdrag ratio needs
to be larger/equal to the ratio of the initial congura-
40
Table 8 Comparison of aeroelastic performance of initial
and optimized design
conguration initial optimized
mass (design domain) 1556.100 790.400
lift [kN] 32.307 30.652
drag [kN] 3.223 2.911
liftdrag 10.023 10.528
tion, and the maximum von Mises stress may not exceed
the maximum stress of the initial design. The maximum
stress in the skin is approximated by the Kreisselmeier
Steinhauser function. The design domain of the initial
conguration is lled with the bulk material with a rela-
tive density of /
0
= 1.0. The optimization problem is
summarized in Table 7, with
g
being the gravitational
acceleration.
In the initial conguration all constraints are active.
Over the course of the optimization process the mass is re-
duced by 49.21%. The mass reduction allows for a smaller
lift in the optimum, which in turn reduces the induced
drag. As the overall aerodynamic loading decreases, the
mass of the spars can be reduced while satisfying the
stress constraint. In the optimized conguration, only
the constraints on the masslift ratio and the maximum
stress in the skin are active. The liftdrag ratio is slightly
increased in the optimized design. The aeroelastic per-
formance of the initial and the optimized congurations
are compared in Table 8.
The optimized density distribution in the design do-
main is depicted in Fig. 11, showing only elements with
a density ratio larger than /
0
= 0.5 and /
0
= 0.75.
The comparison between the two density plots indicates
that there remains some amount of material with low
porosity in the nal conguration. This material is noted
Fig. 12 Deformed shape with pressure contour plot of opti-
mized conguration
in light grey in Fig. 11. The presence of porous material in
the optimized conguration may indicate that foam-type
materials, even with low stinessmass ratios, are advan-
tageous for exible aeroelastic structures. As the material
accumulation along the upper surface indicates, the foam
is mainly used to preserve the aerodynamically compati-
ble shape. The deformed shape and the pressure contour
plot of the optimized design are shown in Fig. 12.
7
Conclusions
A computational methodology has been presented for
the internal structural layout of exible structures un-
dergoing coupled uidstructure phenomena. The ap-
proach combines material topology optimization, high-
delity three-dimensional aeroelastic analysis, coupled
adjoint sensitivity analysis, and a Sequential Augmented
Lagrange optimization method.
The structural geometry is described by the distri-
bution of a porous material in the design domain. The
SIMP approach is used to relate the material properties
to the optimization variables. Accounting for the coupled
aeroelastic response in the topology optimization process
leads to a computationally challenging problem, which
is characterized by a large number of optimization and
state variables. The optimization problem is solved by
a SAL method. The aeroelastic problem is modeled by
a three-eld formulation, and the aeroelastic response is
analyzed by a staggered scheme. A crucial component of
this approach is the adjoint method for evaluating the
sensitivities of the objective and constraints, as the top-
ology optimization problem often results in thousands of
optimization variables.
The proposed methodology has been applied to the
layout optimization of aeroelastic wings represented by
an equivalent plate structure and a detailed build-up -
nite element model. The importance of accounting for the
uidstructure interaction is illustrated by comparing the
optimized layout for a design and deformation-dependent
aerodynamic pressure and the result assuming a constant
aerodynamic load. Neglecting aeroelastic eects in the
topology optimization procedure may lead to structures
with aerodynamically unfavorable large deformations. In
future studies, aeroelastic divergence phenomena need to
be studied as they may generate local minima and cause
convergence problems for very exible structures. Also,
nonlinear structural models need to be included to ac-
count for buckling phenomena of the skin and stieners,
for example.
Acknowledgements The rst author acknowledges the sup-
port of the National Science Foundation under Grant No.
DMI-0300539 and the Air Force Oce of Scientic Research
under Grant No. F49620-02-1-0037. The second author ac-
knowledges the support of the Department of Defense through
41
a National Defense Science and Engineering Graduate Fellow-
ship. The authors also thank ICEM CFD Engineering Inc. for
providing their mesh generator software ICEM CFD.
References
Ambrosio, L.; Buttazzo, G. 1993: An optimal design problem
with perimeter penalization. Calc. Var. 1, 5569
Barthelemy, J.-F.; Wrenn, G.A.; Dovi, A.R.; Hall, L.E. 1994:
Supersonic transport wing minimum design integrating aero-
dynamics and structures. AIAA J. Aircraft 31, 330338
Bendse, M.P. 1989: Optimal shape design as a material dis-
tribution problem. Struct. Optim. 1, 193202
Bendse, M.P. 1999: Variable-topology optimization: Status
and challenges. Technical report, Department of Mathemat-
ics, Tech. Univ. of Denmark, Lyngby, Denmark
Bowman, K.B.; Grandhi, K.V.; Eastep, F.E. 1989: Structural
optimization of lifting surfaces with divergence and control
reversal constraints. Struct. Optim. 1, 153161
Byrd, R.H.; Lu, P.; Nocedal, J.; Zhu, C. 1995: A limited mem-
ory algorithm for bound constrained optimization. SIAM J.
Sci. Comput. 16, 11901208
Cai, X.-C.; Farhat, C.; Sarkis, M. 1998: A minimum over-
lap restricted additive Schwarz preconditioner and appli-
cations in 3D ow simulations. In: Cai, X.-C.; Mandel, J.;
Farhat, C. (eds.) The Tenth International Conference on Do-
main Decomposition Methods for Partial Dierential Equa-
tions
Chen, B.-C.; Kikuchi, N. 2001: Topology optimization with
design-dependent loads. Finite Elem. Anal. Des. 37, 5770
Conn, A.R.; Gould, N.I.M.; Toint, P.L. 1991: A globally con-
vergent augmented Lagrangian algorithm for optimization
with general constraints and simple bounds. SIAM J. Numer.
Anal. 28, 545572
Duysinx, P.; Bendse, M.P. 1997: Topology optimization of
continuum structures with stress constraints. In: Gutkow-
ski, W.; Mr oz, Z. (eds.) Proceedings of the 2nd World Congress
of Structural and Multidisciplinary Optimization, pp. 527
532, Institute of Fundamental Technological Research, War-
saw, Poland
Eschenauer, H.A.; Becker, W.; Schumacher, A. 1998: Multi-
disciplinary structural optimization in aircraft design (in Ger-
man). Technical report, Final report of BMBF DYNAFLEX
Eschenauer, H.A.; Olho, N. 2001: Topology optimization of
continuum structures: a review. Appl. Mech. Rev. 54(4),
331389
Farhat, C.; Degand, C.; Koobus, B.; Lesoinne, M. 1998a: Tor-
sional springs for two-dimensional dynamic unstructured uid
meshes. Comput. Methods Appl. Mech. Eng. 163, 231245
Farhat, C.; Lesoinne, M.; LeTallec, P. 1998b: Load and motion
transfer algorithms for uid/structure interaction problems
with non-matching discrete interfaces: momentum and energy
conservation, optimal discretization and application to aeroe-
lasticity. Comput. Methods Appl. Mech. Eng. 157, 95114
Farhat, C.; Lesoinne, M.; Maman, N. 1995: Mixed explicit/
implicit time integration of coupled aeroelastic problems:
threeeld formulation, geometric conservation and dis-
tributed solution. Int. J. Numer. Methods Fluids 21,
807835
Friedmann, P.P. 1991: Helicopter vibration reduction using
structural optimization with aeroelastic/multidisciplinary
constraints a survey. AIAA J. Aircraft 28, 821
Ghattas, O.; Li, X. 1998: Domain decomposition methods for
sensitivity analysis of a nonlinear aeroelastic problem. Int. J.
Comput. Fluid Dyn. 11, 113130
Gumbert, C.R.; Hou, G.J.-W.; Newman, P.A. 2001: Simultan-
eous aerodynamic analysis and design optimization (SAADO)
for a 3 d exible wing. In: AIAA 2001-1107, 39th Aerospace
Sciences Meeting & Exhibit (held in Reno)
Haber, R.B.; Jog, C.S.; Bendse, M.P. 1996: A new approach
to variable-topology shape design using a constraint on
perimeter. Struct. Optim. 11, 112
Haftka, R.T. 1986: Structural optimization with aeroelastic
constraints a survey of U.S. applications. Int. J. Veh. Des. 7,
381392
Hammer, V.B.; Olho, N. 2000: Topology optimization of
continuum structures subjected to pressure loading. Struct.
Multidisc. Optim. 19, 8592
Haug, E.J.; Choi, K.K.; Komkov, V. 1986: Design sensitivity
analysis of structural systems. Orlando: Academic Press
Hou, G.J.-W.; Satyanarayana, A. 2000: Analytical sensitivity
analysis of a statical aeroelastic wing. In: AIAA 20004824,
8th AIAA/USAF/NASA/ISSMO Symposium on Multidisci-
plinary Analysis and Optimization (held in Long Beach)
Lesoinne, M.; Farhat, C. 1998: A higher-order subiteration
free staggered algorithm for nonlinear transient aeroelastic
problems. AIAA J. 36, 17541756
Lund, E.; Mller, H.; Jakobsen, L.A. 2002: Shape optimiza-
tion of uidstructure interaction problems with large dis-
placements and two-equation turbulence models. In:
Mang, H.A.; Rammerstorfer, F.G.; Eberhardsteiner, J. (eds.)
Proceedings of the Fifth World Congress on Computational
Mechanics (WCCM V) (held in Vienna). Vienna: Vienna Uni-
versity of Technology, ISBN 3-9501554-0-6,
http://wccm.tuwien.ac.at
Ma, Z.-D.; Kikuchi, N.; Cheng, H.-C.; Hagiwara, I. 1995:
Topological optimization technique for free vibration prob-
lems. ASME J. Appl. Mech. 62, 200207
Martins, J.R.R.A.; Alonso, J.J. 2002: High-delity aero-
structural design optimization of a supersonic business jet.
In: AIAA 2002-1483, 43rd AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference
(held in Denver)
Maute, K. 2002: Topology optimization for non-linear aeroe-
lastic uidstructure interaction problems. In: Mang, H.A.;
Rammerstorfer, F.G.; Eberhardsteiner, J. (eds.) Proceedings
of the Fifth World Congress on Computational Mechanics
(WCCM V) (held in Vienna). Vienna: Vienna University of
Technology, ISBN 3-9501554-0-6, http://wccm.tuwien.ac.at
42
Maute, K.; Nikbay, M.; Farhat, C. 2000: Analytically based
sensitivity analysis and optimization of nonlinear aeroelas-
tic systems. In: AIAA 20004825, 8th AIAA/USAF/NASA/
ISSMO Symposium on Multidisciplinary Analysis and Opti-
mization (held in Long Beach)
Maute, K.; Nikbay, M.; Farhat, C. 2001: Coupled analytical
sensitivity analysis and optimization of three-dimensional
nonlinear aeroelastic systems. AIAA J. 39(11), 20512061
Maute, K.; Nikbay, M.; Farhat, C. 2002: Conceptual lay-
out of aeroelastic wing structures by topology optimization.
In: AIAA 2002-1480, 43rd AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference
(held in Denver)
Maute, K.; Nikbay, M.; Farhat, C. 2003: Sensitivity analysis
and design optimization of three-dimensional nonlinear aeroe-
lastic systems by the adjoint method. Int. J. Numer. Methods
Eng. 56, 911933
Meric, R.A. 1985: Coupled optimization in steady-state ther-
moelasticity. J. Therm. Stresses 8, 333347
Meric, R.A. 1986: Material and load optimization of thermoe-
lastic solids. Part I: sensitivity analysis. Part II: numerical
results. J. Therm. Stresses 9, 359372, 373388
Meric, R.A. 1990: Optimal cross-sectional shape for MHD
channel ows. Int. J. Numer. Methods Eng. 30, 919929
Michaleris, P.; Tortorelli, D.A.; Vidal, C.A. 1995: Analysis
and optimization of weakly coupled thermoelastoplastic sys-
tems with application to weldment design. Int. J. Numer.
Methods Eng. 38, 12591285
Militello, C.; Felippa, C. 1991: The rst ANDES elements:
9-dof plate bending triangles. Comput. Methods Appl. Mech.
Eng. 91, 217246
Mller, H.; Lund, E. 2000: Shape sensitivity analysis of
strongly coupled uidstructure interaction problems. In:
AIAA 20004823, 8th AIAA/USAF/NASA/ISSMO Sympo-
sium on Multidisciplinary Analysis and Optimization (held in
Long Beach)
Olho, N.; Eschenauer, H. 1999: On optimum topology design
in mechanics. In: ECCM99, European Conference on Compu-
tational Mechanics (held in Munich)
Petersson, J.; Sigmund, O. 1998: Slope constrained topology
optimization. Int. J. Numer. Methods Eng. 41, 14171434
Piperno, S.; Farhat, C.; Larrouturou, B. 1995: Partitioned
procedures for the transient solution of coupled aeroelas-
tic problems Part I: model problem, theory, and two-
dimensional application. Comput. Methods Appl. Mech. Eng.
124, 79112
Pramono, E.; Weeratunga, S.K. 1994: Aeroelastic compu-
tations for wings through direct coupling on distributed-
memory MIMD parallel computers. In: AIAA-94-0095, 32nd
Aerospace Sciences Meeting and Exhibit (held in Reno)
Rodriques, H.; Fernandes, P. 1995: A material based model for
topology optimization of thermoelastic structures. Int. J. Nu-
mer. Methods Eng. 38, 19511965
Roe, P.L. 1981: Approximate Riemann solvers, parameter
vectors and dierence schemes. J. Comput. Phys. 43, 357372
Rozvany, G.I.N. 1989: Structural design via optimality crite-
ria. Dordrecht: Kluwer Academic Publishers
Rozvany, G.I.N. (ed.) 1997: Topology optimization in struc-
tural mechanics, Vol. 374 of CISM Course and Lectures. Vi-
enna: Springer
Rozvany, G.I.N.; Bendse, M.P.; Kirsch, U. 1995: Layout opti-
mization of structures. Appl. Mech. Rev. 48, 41119
Rozvany, G.I.N.; Olho, N. (eds.) 2001: Topology optimization
of structures and composite continua. (NATO ARW held in
Budapest). Dordrecht: Kluwer Academic Publishers
Rozvany, G.I.N.; Zhou, M.; Birker, T. 1992: Generalized shape
optimization with homogenization. Struct. Optim. 4, 250252
Sigmund, O. 1994: Design of material structures using top-
ology optimization. PhD thesis, Danish Center for Ap-
plied Mathematics and Mechanics, Technical University of
Denmark
Sigmund, O. 1998: Topology optimization in multiphysics
problems. In: AIAA 98-4905, Proceedings of the 7th AIAA/
USAF/NASA/ISSMO Symposium on Multidisciplinary An-
alysis and Optimization (held in St. Louis), pp. 14921500
Sigmund, O. 2001a: Design of multiphysics actuators using
topology optimization Part I: one-material structures. Com-
put. Methods Appl. Mech. Eng. 190, 65776604
Sigmund, O. 2001b: Design of multiphysics actuators using
topology optimization Part II: two-material structures.
Comput. Methods Appl. Mech. Eng. 190, 66056627
Sigmund, O.; Petersson, J. 1998: Numerical instabilities in
topology optimization: a survey on procedures dealing with
checkerboards, mesh-dependencies and local minima. Struct.
Optim. 16, 6875
Sigmund, O.; Torquato, S.; Aksay, I.A. 1998: On the design of
1-3 piezo-composites using topology optimization. J. Mater.
Res. 13(4), 10381048
Silva, E.C.N.; Fonseca, J.S.O.; Kikuchi, N. 1997: Optimal de-
sign of piezoelectric microstructures. Comput. Mech. 19,
397410
Sobieszczanski-Sobieski, J. 1990: Sensitivity of complex, inter-
nally coupled systems. AIAA J. 28, 153160
Strganac, T.W.; Mook, D.T. 1990: Numerical model of un-
steady subsonic aeroelastic behavior. AIAA J. 28, 903909
Tenek, L.H.; Hagiwara, I. 1993: Static and vibrational shape
and topology optimization using homogenization and math-
ematical programming. Comput. Methods Appl. Mech. Eng.
109, 143154
Tortorelli, D.A.; Subramani, G.; Lu, S.C.Y.; Haber, R.B. 1991:
Sensitivity analysis for coupled thermoelastic systems. Int. J.
Solids Struct. 27, 14771497
Zhou, M.; Shyy, Y.K.; Thomas, H.L. 2001: Checkerboard and
minimum member size control in topology optimization.
Struct. Multidisc. Optim. 21, 152158

Anda mungkin juga menyukai