You are on page 1of 9

Aeroelastic design optimization of thin-walled subsonic wings

against divergence
Liviu Librescu
a,1
, Karam Y. Maalawi
b,
a
Department of Engineering Science and Mechanics, Virginia Tech, VA 24061, USA
b
Department of Mechanical Engineering, National Research Center, Dokki, P.O. 12622, Cairo, Egypt
a r t i c l e i n f o
Article history:
Received 6 December 2007
Received in revised form
20 May 2008
Accepted 20 May 2008
Available online 27 June 2008
Keywords:
Optimization
Thin-walled wings
Aeroelasticity
Torsional divergence
a b s t r a c t
The present paper deals with aeroelastic design optimization of a slender, thin-walled wing-type
structure against divergence. The main goal is to avoid torsional instability, which might occur at critical
ow conditions, by maximizing the divergence speed without the penalty of increasing the total
structural mass. Divergence provides a useful measure of the general stiffness level of the wing
structure. The model formulation considers a large aspect ratio unswept wing of rectangular planform,
while the ow conditions are restricted to those of subsonic incompressible ones. Both continuous and
piecewise models are analyzed, where exact analytical solutions are obtained within the context of
linear elasticity and aerodynamic strip theories. The nal optimization problem is formulated as a
nonlinear mathematical programming problem solved by implementing the interior penalty function
technique, which interacts to eigenvalue calculation routines. Results show that optimum patterns with
decreasing wall thickness from the inboard portion toward the outboard one produce signicant
improvement in the overall torsional stiffness level. It is also shown that global optimality can be
achieved from the proposed mathematical model, provided that the wing is constructed from piecewise
uniform portions having not-equally spaced lengths and different torsional rigidities.
& 2008 Elsevier Ltd. All rights reserved.
1. Introduction
A ight vehicle is a complex structural system with numerous
variables and constraints. The number of design variables and
alternate constructions is too large to be xed by the governing
equations and constraints. Airframe designers usually resort to
past experience and similar existing designs to x the values of
undetermined variables. This can result in non-optimal aero-
dynamic and structural design.
Optimization techniques are used today to improve aircraft
designs and determine the values of unknown variables. Ref. [1]
presented a brief for the potential applications of optimization in
the eld of aeronautical engineering, where several design aspects
can be found. Concerning aerodynamic optimization, Harvey and
Johnson [2] developed two candidate wings with reduced
transonic drag. Rohl et al. [3] presented a combined aerodynamic
and structural optimization of a high-speed civil transport wing
including strength, buckling, and aeroelastic constraints. Attempts
in other directions of aircraft wing optimization include Edwins
optimization of the Grumman X-29, forward swept wing [4] with
the objective of minimizing structural weight under constraints
imposed on strength and divergence velocity. Grossman et al. [5]
investigated the interaction of aerodynamic and structural
optimization for a sailplane wing via a sequential design
procedure. The number of design variables selected ranged
between 25 and 35, and the chosen criteria for the design were
maximization of aerodynamic performance and minimization of
structural weight under aeroelastic constraints. Rao [6] calculated
the optimum design of aircraft wing subjected to taxiing loads.
The optimization problem was formulated and solved as a
constrained nonlinear programming problem based on nite
element modeling.
The importance of aeroelastic stability in aerospace structural
design has provided strong motivation for development of
numerous research works. One of the most important design
aspects is the determination of the aeroelastic response and the
critical ight speeds at which static and/or dynamic instability
might occur. The basic underlying concepts of the theory of
aeroelasticity can be found in Refs. [79]. Divergence instability of
swept composite wings was considered by Librescu and Thangji-
tham [10], who presented an exact method for determining the
critical divergence speed. Their approach was based on Laplace
transform technique, and the analysis included the effect of both
free and constrained warping models for wing torsion. Another
study by Librescu and Song [11] analyzed divergence instability of
ARTICLE IN PRESS
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/tws
Thin-Walled Structures
0263-8231/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2008.05.007

Corresponding author. Tel.: +20224189599; fax: +2023370931.


E-mail address: maalawi@netscape.net (K.Y. Maalawi).
1
Deceased.
Thin-Walled Structures 47 (2009) 8997
swept-forward wings constructed of composite materials. A thin-
walled beam model was implemented, which incorporates
anisotropy of the material, transverse shear deformation as well
as warping effects. On the other hand, Karpouzian and Librescu
[12] developed a plate-beam model suitable for aeroelastic
analysis of composite wings. Results revealed the importance of
incorporating transverse and warping restraint effects for accurate
prediction of the static aeroelastic response of swept-forward
wings. Aeroelastic optimization of aircraft wings was reviewed by
Done [13], who compared the different aerodynamic and
structural modeling with further comments on the future
progress in wing aeroelasticity. Schuster [14] considered the
optimization of wing torsional stiffness distribution required to
generate a specic aeroelastic twist variation along the wingspan.
Butler et al. [15] calculated the minimal mass design of high
aspect ratio composite wing under constraints imposed on utter
and divergence speeds. The wing was modeled as a series of box
beams whilst aeroelastic loads were based on strip theory. Design
variables included engine position, spar locations, as well as ply
thickness variation. A multi-objective optimization was applied by
Layton [16] to nd the optimal static and dynamic aeroelastic
design of a swept-back wing with the design variables selected to
be the sweepback angle, aspect ratio, and chord distribution.
The present work develops a novel mathematical model for
optimizing the design of a slender wing-type structure against
static torsional instability. The objective function is measured by
maximizing the divergence speed without the penalty of increas-
ing structural weight. The study considers unswept, rectangular,
thin-walled subsonic wings with high aspect ratio. Based on an
exact differential equation formulation, analytical solutions are
given for congurations having continuous or stepped wall
thickness distribution using Bessels functions of the rst and
second kinds. The selected design variables encompass the wall
thickness and length of each panel composing the structure.
Closed-form mathematical expressions are given in appropriate
dimensionless forms for calculating the divergence speed. The
resulting optimization problem has been formulated as a non-
linear programming problem solved by the interior penalty
function method. It has been shown that for a prescribed total
structural mass the global maxima of the divergence speed is
attainable, no matter the number of portions composing the
lifting surface is. The objective function, even though implicit, can
be imagined as an explicit function in the selected optimization
variables. The given mathematical approach has shown the
possibility of dealing with an inverse eigenvalue problem, that is
the divergence speed can be prescribed and the resulting
characteristic equation is solved for any of the chosen optimiza-
tion variables. In all, within the main assumptions of the proposed
model formulation, the overall torsional stability of a typical wing
structure is improved signicantly as compared to conventional
designs.
2. Theoretical analysis
Divergence is an aeroelastic phenomenon dened as static
torsional instability of a lifting surface [79]. When the ow speed
exceeds a certain value called divergence speed, V
div
, the
aerodynamic moment overpowers the restoring elastic moment
of the structure. This creates an unstable condition and the
structure twists to destruction. This problem is one of the crucial
design considerations in aerospace industry. High divergence
speed can be regarded as a major aspect in designing an efcient
lifting surface. It is a sign of high structural torsional stiffness,
which can have many desirable effects on the overall structural
design of the lifting surface. It helps in avoiding the occurrence of
large displacements, distortions, and excessive vibrations. It may
also reduce fretting among structural parts, which is a major
cause of fatigue failure.
The present study considers an unswept, slender lifting surface
having thin-walled airfoil cross-section, as shown in Fig. 1. The
ow is taken to be steady and incompressible, and, in this context,
the aerodynamic strip theory is adopted [8]. Rectangular wing
planform with sufciently large aspect ratio and with no major
cutouts through the structure is considered so that the classical
engineering theory of torsion can be applicable and the state of
deformation described in terms of one space coordinate. The
elastic axis is assumed to be well dened and, hence, the bending
and torsion degrees of freedom can be decoupled. For low aspect
ratio wings made of composites, the cross-section warping ought
to be included to account for three-dimensional deformation
effects [1012]. This will be considered by the authors in an
extension of the present study.
In the present wing model, the governing differential equation
of torsion is given by [79]
d
dy
GJ
da
e
dy
_ _
Ty 0 (1a)
where GJ is the torsional stiffness of the wing, a
e
the elastic angle
of attack (i.e. elastic twist), and T(y) the applied torque distribu-
tion about the elastic axis given by (refer to the nomenclature and
Fig. 1b):
Ty qc
2
eaa
e
a
o
qc
2
C
m:a:c
N
z
mgd (1b)
where a
o
is the rigid angle of attack ( setting angle at wing
root+rigid twist) and N
z
the ight load factor. Substituting from
ARTICLE IN PRESS
Nomenclature
a airfoil lift-curve slope
b semi-span of the lifting surface
b
o
semi-span of a baseline design
c chord of airfoil cross-section
c
o
chord of the baseline design
C
m.a.c
pitching moment coefcient
d distance between the wing centroidal axis and elastic
axis
e fractional location of the airfoil shear center, positive
aft with respect to the aerodynamic center
h wall thickness of the main torsional-carrying struc-
ture
h
o
wall thickness of the baseline design
h
r
wall thickness at wing root
G modulus of rigidity
J torsional constant of the airfoil cross-section about
elastic axis
J
o
torsional constant of baseline design
mg wing weight distribution in spanwise direction
T torsional moment
V airow velocity
q dynamic pressure ( rV
2
/2)
D thickness taper ratio ( h
tip
/h
root
)
a
o
rigid angle of attack
a
e
elastic angle of attack.
b 1D
r air density
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 90
(1a) into (1b), we get
d
dy
GJ
da
e
dy
_ _
qc
2
eaa
e
qc
2
eaa
o
qc
2
C
m:a:c
N
z
mgd (1c)
The right hand side of Eq. (1c) represents the disturbing external
forcing function. Since we are dealing with the wing torsional
instability (i.e. an eigenvalue problem), only the homogeneous
part of the general solution of Eq. (1c) is of interest. Therefore, we
consider solution of the differential equation of the left hand side,
i.e.:
GJa
0
e
0

1
2
rV
2
c
2
eaa 0 (1d)
which must be satised everywhere over the domain [0, b]. For a
cantilevered lifting surface, the boundary conditions are a(0) 0
and a
0
(b) 0, where the prime denotes differentiation with
respect to y. Here a denotes the elastic angle of attack. The
various variables and parameters are dened in the nomenclature.
The problem of solving Eqs. (1ad) under the specied boundary
conditions is called a self-adjoint boundary value problem, or a
SturmLiouville problem [17]. Details of the main characteristic
features of such problems are given in Appendix A.
For thin-walled, multi-cell construction, the total structural
mass M, and the torsional constant J(y), can be determined from
the expressions
M f
1
_
b
0
chydy (2)
Jy f
2
c
3
hy (3a)
where f
1
and f
2
are shape factors that depend on the shape of the
airfoil section, number of interior cells, and the ratios between the
shear web thicknesses and the main wall thickness h(y). For
example, in the case of two-cell wing section (see Fig. 1b) the
torsional constant has the mathematical form [7,9]
J 4:0
a
20
A
2
1
a
12
A
2
a
10
A
2
2
a
12
a
10
a
20
a
10
a
20
_ _
(3b)
where a
10
represents a line integral
_
ds=h for cell wall 10, and
a
12
and a
20
the line integrals for the other interior web and outside
wall of cell 2. A
i
is the enclosed area of the ith cell and A the total
enclosed area ( A
1
+A
2
). Since a
10
, a
20
, a
12
are proportional to the
chord/thickness ratio (c/h) and A
i
is proportional to c
2
, the
torsional constant is directly proportional to c
3
h, with f
2
in
Eq. (3a) being the proportionality factor.
2.1. Baseline design
It is convenient to non-dimensionalize all variables and
parameters with respect to a baseline design having uniform
stiffness and mass distributions with the same material proper-
ties, airfoil section, type of construction, shape factors f
1
and f
2
,
and same ow properties as well. The various dimensionless
quantities that are described by the (^:) notation are dened in
Table 1. Therefore, dividing by the corresponding baseline design
parameters, the governing differential equation Eq. (1d) takes the
following dimensionless form:

^
ha
0

0

^
V
2
^
b
2
^ c
_
_
_
_
a 0 (4)
which must be satised everywhere over the domain [0, 1]. For a
cantilevered lifting surface, the boundary conditions to be
satised are a(0) 0 and a
0
(1) 0, where the prime here denotes
differentiation with respect to y. The dimensionless expressions
for the total mass and the torsional constant are, respectively,
given by
^
M ^ c
_
1
0
^
hd^ y (5)
^
J ^ c
3
^
h (6)
which depend on a specic thickness distribution.
3. Optimization problem formulation
In formulating an optimization, three principal phases must be
considered:
denition and measure of the design objectives;
denition of the design constraints;
denition of the design variables and preassigned parameters.
In the present study the objective function is represented by
maximization of the divergence speed while maintaining the
structural mass constant. Divergence provides a useful measure
for the general stiffness level of the wing structure. In order to
formulate a practical simplied model, pursue feasible optimiza-
tion, and reduce the dimensionality of the design space, some of
ARTICLE IN PRESS
Fig. 1. Aeroelastic model of a rectangular lifting surface.
Table 1
Denition of dimensionless quantities
Airfoil chord ^ c c=co
Semi-span of lifting surface ^
b b=bo
Length of the kth portion ^
b
k
b
k
=bo
Torsional rigidity ^
J GJ=GJ
o
Wall thickness h

h/h
o
Thickness of kth portion h

k
h
k
/h
o
Structural mass M

M/M
o
Air velocity ^
V Vcobo

rea=2GJ
o
_
Spatial coordinate y y/b
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 91
the design parameters that affect wing divergence shall be given
values similar to those of the baseline design. They are selected to
be the type of airfoil section (i.e. a is xed), type of material of
construction, the number of cells constituting the main wing
cross-section, and provided similar location of the elastic axis
with respect to the aerodynamic center (i.e. e is xed). Same ight
altitude is also postulated (i.e. r is xed). Therefore, the remaining
variables that are subjected to change during the optimization
process are the spanwise variation of the main wall thicknesses
h(y), chord length c, and the wing semi-span b. The nal
mathematical form of the optimization model can be cast as the
following:
Find the design variable vector ~x h; c; b that
minimize the objective function F~x
^
V
div
(7a)
subject to the mass equality constraint
^
M 1:0 (7b)
and the side constraints ~x
l
p~xp~x
u
(7c)
where V

div
is the normalized divergence speed (see Table 1) and~x
l
and ~x
u
are, respectively, the lower and upper bounds that restrict
the values of the design variables. The thickness of the wing skin
is usually restricted by the available standard sheet thicknesses.
Its lower bounds may also be determined from the consideration
of wall instability that might happen by local buckling.
The above optimization problem may be thought of as a search
in the design space for a point corresponding to the minimum
value of the objective function and such that it lies within the
region bounded by subspaces representing the constraint func-
tions. It can be solved by a number of iterative techniques, such as
the sequential search methods [18,19], which have received a
great deal of attention and are the most promising. A method that
has a wide applicability in engineering applications is the penalty
function method [18]. Other algorithms for solving global
optimization problems may be classied into heuristic methods,
which nd only the global optimum with high probability, and
methods that guarantee a global optimum with some accuracy.
The simulated annealing technique [19] and the genetic algo-
rithms [20] belong to the former type, where analogies to physics
and biology in approaching the global optimum are utilized. The
most important methods of the second type are the branch and
bound methods [18], in which the design space is split repeatedly
by branching into smaller and smaller parts.
In the present study we select the unconstrained formulation
using the interior penalty function technique to nd the needed
optimal solution of the proposed optimization problem. Powells
non-gradient algorithm is chosen for the generation of the search
vectors [18].
3.1. Continuous model
The functional behavior of the objective function, the diver-
gence speed, will be examined rst by considering the case of a
rectangular wing having a linear wall thickness distribution in the
spanwise direction:
^
h^ y
^
h
r
1 b^ y (8)
where the subscript r identies the root section. Introducing the
transformation x (1by), and substituting for h

in Eq. (4), the


following second-order differential equation is obtained:
x
d
2
a
dx
2

da
dx
l=b
2
a 0; Dpxp1 (9)
where l
^
V
^
b=

^ c
^
h
r
_
. The general solution of Eq. (9) is given by
[9,17]
ax D
1
J
o
2l
b

x
p
_ _
D
2
Y
o
2l
b

x
p
_ _
a1 0 and a
0
D 0 (10)
where D
1
and D
2
are constants of integration, and J
o
and Y
o
are the
zero-order Bessel functions of the rst and second kind,
respectively. Applying the boundary conditions and considering
the non-trivial solution, where D
1
60 and D
2
60, the resulting
transcendental equation takes the form
J
o
2l
b
_ _
Y
0
o
2l
b

D
p
_ _
Y
o
2l
b
_ _
J
0
o
2l
b

D
p
_ _
0 (11)
which is to be solved for the dimensionless parameter l. The
lowest root corresponds to the wing divergence speed, V

div
.
Fig. 2 shows the developed level curves of the objective
function, V

div
in the (b

D) design space while keeping the total


structural mass at a value equal to that of the baseline design, i.e.
M

1.0. It is seen that V

div
is a well-behaved function and
continuous in the design variables b

and D. For a given value of b

,
the divergence speed increases as the thickness taper ratio D
decreases. It is also observed that, for a xed value of D, a slight
decrease in the wingspan would result in an appreciable increase
in the divergence speed. However, care ought to be taken to not
violate aerodynamic performance requirements of the ight
vehicle. Therefore, maintaining the wingspan constant at a value
equal to that of the baseline design (b

1), Fig. 3 depicts the level


curves V

div
in the (Dh

r
) design space. It is seen that for a known
taper of the wall thickness, V

div
monotonically increases with the
root thickness, h

r
, with the penalty of increasing the total
structural weight.
For the purpose of completeness of the general solution we
consider the case with thickness tapering ratios exceeding unity,
i.e. D41.0, where the transcendental equation for determining V

div
takes the form:
J
o
2l
b

D
p
_ _
Y
0
o
2l
b
_ _
Y
o
2l
b

D
p
_ _
J
0
o
2l
b
_ _
0 (12)
The nal results showing variation of both the divergence speed
and the associated root thickness while maintaining both mass
ARTICLE IN PRESS
Fig. 2. Level curves of V

div
in the (b

D) design space; M

1.0. Rectangular wings


with linear thickness distribution.
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 92
and span at constant values (M

1.0) are given in Fig. 4. The


optimal root thickness, h

r
, is seen to be sharply decreasing with
increasing taper ratio D because of the imposed mass equality
constraint. Good wing designs are seen to have chosen lower limit
of the thickness taper, where a value of V

div
1.715 is reached at
D 0.1. This corresponds to a value of the root thickness h

r
1.82
and an optimization gain of about 9.18% measured relative to the
baseline design value of p/2, i.e. (1.7151.5708)/1.5708. For D41.0,
it is seen that the wing torsional stability is degraded consider-
ably, and therefore, such a case is not of practical interest.
4. Optimization of wings with stepped thickness distribution:
piecewise model
Some authors [8,9] have addressed the problem of divergence
instability of rectangular wings made of two-piecewise uniform
portions. In those studies, however, the continuity requirements
at the joining boundary of the inboard and outboard portions
were not adequately satised. Maalawi [21] presented an
analytical model for buckling analysis of exible columns. The
study showed that the use of piecewise model in structural
optimization gives excellent results and can be promising for
similar applications.
ARTICLE IN PRESS
Fig. 3. V

div
-level curves in the Dh

r
design space (b

1.0).
Fig. 4. Variation of V

div
and h

r
with the wall thickness taper D for constant mass
and span (M

1.0, b

1.0).
Fig. 5. Rectangular lifting surface constructed from piecewise panels.
Fig. 6. Denition of the state variables for the kth portion.
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 93
Fig. 5 shows a rectangular lifting surface, which is constructed
from uniform piecewise portions, each of which has different
torsional rigidity and length. For the kth portion shown in Fig. 6,
Eqs. (1ad) reduces to
a
00

reaV
2
c
2
2GJ
k
a 0; y
k
pypy
k1
(13)
Dening a dimensionless local spanwise coordinate y (yy
k
/b
o
),
one gets
a
00
l
2
k
a 0; 0p yp
^
b
k
(14a)
where
l
k

^
V^ c

^
J
k
_ (14b)
It is to be noticed here that the prime in Eq. (14a) denotes
differentiation with respect to y, instead of y. Its general solution is
given by:
a y A cosl
k
y B sinl
k
y (15a)
where A and B are constants of integration. The dimensionless
internal torsional moment can be obtained by differentiating
Eq. (15a), and multiplying by the dimensionless torsional rigidity:
T y
^
J
k
a
0
l
k
^
J
k
A sinl
k
y B cosl
k
y (15b)
Applying the boundary conditions at stations k and k+1 (refer to
Fig. 6):
At y 0; a
k
A; T
k
Bl
k
^
J
k
At y
^
b
k
; a
k1
A cosl
k
^
b
k
B sinl
k
^
b
k
(16a)
T
k1
l
k
^
J
k
A sinl
k
^
b
k
B cosl
k
^
b
k
(16b)
Applying the transfer matrix method [25] by eliminating the
coefcients A and B, the state variables at both stations are related
by the following matrix relation:
a
k1
T
k1
_ _

C
k
S
k
l
k
^
J
k
l
k
^
J
k
S
k
C
k
_

_
_

_
a
k
T
k
_ _
(17a)
or
fU
k1
g E
k
fU
k
g (17b)
S
k
sinl
k
^
b
k
and C
k
cosl
k
^
b
k
(17c)
It is now possible to compute the state variables progressively
along the wingspan by applying continuity requirements
among the portions interconnecting boundaries. Therefore, the
state variables at the extreme boundaries are related by the
equation
fU
N1
g EfU
1
g (18a)
where
E E
N
E
N1
. . . E
3
E
2
E
1
(18b)
Applying the appropriate boundary conditions at the xed and
free ends, that is a
1
T
N+1
0, Eq. (18a) takes the form
a
N1
0
_ _

E
11
E
12
E
21
E
22
_ _
0
T
1
_ _
(18c)
Considering the non-trivial solution of Eq. (18c), i.e. T
1
60, the
resulting transcendental equation, also named characteristic
equation, is given by
E
22
0 (18d)
which can be directly solved for the critical velocity at which
torsional divergence instability occurs. The explicit form of
Eq. (18d) for the case of a two-panel wing is given in Eq. (21a)
of Section 4.1. In the subsequent sections the functional behavior
of the implicit function V

div
with the selected design variables will
be thoroughly investigated by considering several cases of study
of lifting surfaces built of different number of portions. The
dimensionless torsional constant and total mass can be shown to
have the forms
^
J
k
^ c
3
^
h
k
(19a)
^
M ^ c

N
k1
^
h
k
^
b
k
(19b)
For a wing composed of only one portion, the solution of Eq.
(18d) yields the lowest eigenvalue l
1
b

1
p/2. Therefore, the
dimensionless divergence velocity is
^
V
div

p
2
_ _

^
J
1
_
^ c
^
b
1

p
2
_ _

^ c
^
h
1
_
^
b
1
(20a)
Eq. (20a) shows that the main design variables that have a
signicant effect on the present aeroelastic optimization problem
are the wing chord, span, and wall thickness. Actually, the
equation may be thought of as an explicit function describing
the critical divergence speed in terms of aforementioned vari-
ables. It is seen that V

div
increases monotonically with the square
root of the chord and wall thickness and decreases with the semi-
span, which is a natural expected behavior. For M

1.0, which
means that the optimized wing has the same total mass of its
baseline deign, the divergence speed can be expressed by the
relation
^
V
div

p
2
^
b
3=2
(20b)
A slight decrease in the wing semi-span of about 7% results in
11.4% increase in the divergence speed. In conclusion, instead of
treating V

div
as an implicit function, it may be possible to choose
prescribed values for V

div
and two of the variables (h

, b

, ^ c) and
solve Eq. (18d) numerically for the remaining unknown variable,
which must also satisfy Eq. (20a). These novel mathematical
consequences will be implemented for lifting surfaces composed
of several numbers of piecewise portions.
4.1. Rectangular wing constructed from two portions
Applying the same procedure as before, the derived character-
istic equation (18d) for calculating the divergence velocity can be
shown to have the following compacted form:
tanl
1
^
b
1
tan l
2
^
b
2

^
J
1
_
^
J
2
_
(21a)
where
l
1

^
V^ c

^
J
1
_ ; l
2

^
V^ c

^
J
2
_ ; and
^
b
1

^
b
2

^
b (21b)
It is seen that the velocity function V

is implicitly expressed in
terms of the variables
^
J
1
,
^
J
2
, b

1
, b

2
, and ^ c. Keeping the planform
geometry the same as the baseline design, i.e. ^ c
^
b 1, and using
Eq. (19a), Eq. (21a) takes the form
tan
^
V
^
b
1

^
h
1
_
_
_
_
_
_
_tan
^
V
^
b
2

^
h
2
_
_
_
_
_
_
_

^
h
1
^
h
2

_
(21c)
ARTICLE IN PRESS
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 94
which is to be solved for the lowest root corresponding to the
critical ight speed, V

div
, at which divergence occurs. Actually
V

div
is implicitly expressed in terms of the variables (h

1
, b

1
)
and (h

2
, b

2
). One of the span length variables can be discarded by
considering the semi-span to be equal to that of the baseline
design, i.e. b

1
+b

2
1.0. It is virtually possible to have very large
values of V

div
, which provides more stiffened wing structures. This,
unfortunately, results in increasing the total structural weight,
which has a harmful effect on the overall performance character-
istics of the aircraft design. To avoid this problem, the mass
equality constraint described by the relation
^
M 1:0
^
h
1
^
b
1

^
h
2
^
b
2
(22)
must be imposed on the solutions of Eq. (21c), which further
reduces the number of effective design variables to only two,
namely: (h

1
, b

1
) or (h

2
, b

2
). It is noticed here that the chord of the
optimized wing is taken to be the same as that of the baseline
design (Fig. 7).
Fig. 8 shows variation of V

div
with span length of the inboard
portion, b

1
, for different values of the main wall thickness h

1
while
maintaining constant mass, i.e. M

1.0. It is seen that V

div
is a
well-behaved function and continuous in the selected design
variables. Good patterns with V

div
4p/2 lie in the favorable range
1.0ph

1
p2.0 and 0.25pb

1
p0.75. The optimal design point was
found to be (h

1
, b

1
) (1.25, 0.7) and (h

2
, b

2
) (0.417, 0.3) having
the maximum value of V

div
1.68, which represents about 7%
increase above that of the uniform baseline design (p/2). The point
of intersection of a constant-h

1
curve with the horizontal line
V

div
p/2 represents the critical maximum length of the inboard
portion, after which the divergence velocity sinks rapidly to
undesirable lower limits. In conclusion, it is recommended that
the portion with higher torsional rigidity be located near the wing
root. However, abrupt variation in the stiffness would make the
value of V

div
less than the reference value p/2. A moderate
variation is thus preferable for good wing designs. Applying the
developed mathematical concept of treating the objective func-
tion as if it were an explicit function in the design variables, the
optimum zone encompassing the global optima has been
produced and depicted in Fig. 9. The feasible domain must be
bounded by the imposed mass equality constraint described by
the curved line h

1
b

1
1.0. It may be of interest to notice here that
the developed contour lines in the neighborhood of the optimal
solution have airfoil-like-shaped geometry as that of common
subsonic wing sections.
4.2. Wings constructed from several portions
In the previous section, we have claried the existence of
global optima for a lifting surface composed of two-piecewise
portions. In this section, we apply the penalty function method
(see Appendix B) to nd the optimum solutions for wings
constructed from several portions. Results have been achieved
by maximization of the wing divergence speed, while maintaining
the chord, span, and total mass at values equal to those of the
ARTICLE IN PRESS
Fig. 7. A rectangular wing composed of two portions.
Fig. 8. Variation of V

div
with b

1
for different values of h

1
.
Fig. 9. Optimum zone containing the global optima in b

1
h

1
design space.
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 95
baseline design. One of the design variables is discarded by using
the mass equality constraint. The lower and upper limiting values
imposed on the dimensionless wall thickness were taken to be 0.1
and 2.0, respectively. Non-negativity constraints were also
imposed on the lengths of the individual portions. Table 2
summarizes the obtained results for different cases.
It is seen that the overall torsional stability of the wing
structure can be improved substantially as the number of portions
increases. As a general observation, the wall thickness and length
of the individual panels composing the wing decrease as we move
from the inboard panels toward the outboard ones.
If one further considers the needed cost of machine tooling or
assembling connections, a compromise has to be made on
whether to increase the number of panels composing the wing
up to some limiting value.
5. Conclusions
A novel mathematical approach to the aeroelastic optimization
of a wing-type structure has been presented. The objective
function has been measured by maximization of the divergence
speed while maintaining the total structural mass at a constant
value. The selected design variables encompass the main wall
thickness, chord, and the wing semi-span. Exact analytical
solutions for both continuous and discrete models are given
within the context of linear elasticity and aerodynamic strip
theories. The developed mathematical models ensured the
attainment of global optimality, where the objective function
can be imagined to be explicitly expressed in the selected design
variables. The imposed mass equality constraint was used to
discard one of the design variables, reducing the dimensionality of
the optimization model by one. The given mathematical con-
sequences have proved the possibility of solving the associated
eigenvalue problem in any one of the design variables for a given
value of the divergence speed. Detailed results are presented and
discussed, where useful design trends have been recommended
for optimum patterns with either linear or stepped thickness
distribution along the entire length of the wing. In all, the model
has succeeded in arriving at the needed optimal wing designs
having maximum torsional stability without violating the perfor-
mance requirements imposed on the total structural weight.
Current investigations include incorporation of the chord dis-
tribution by considering wings having trapezoidal and elliptical
geometry as well. Swept wings, both metallic and composite, are
also under study.
Acknowledgments
The authors would like to thank Professor S. Ragab of the
Engineering Science and Mechanics Department, Virginia Tech for
his valuable discussions and help. Support from the Ministry of
Higher Education and the National Research Center in Cairo, Egypt
through Grant SAB 1386/2006 is gratefully acknowledged.
Appendix A. The boundary value problem
Consider the homogeneous, ordinary second-order differential
equation governing torsional instability of unswept subsonic
wings (refer to Eq. (1ad)):
d
dy
GJy
d
dy
ay
_ _
qRyay 0; 0oyob (A.1)
subject to the linear homogeneous boundary conditions:
a0 0 and a
0
b 0 (A.2)
where
q
1
2
rV
2
and Ry eac
2
(A.3)
where c(y) is the wing chord and GJ(y) is its torsional rigidity. GJ(y)
and R(y) are positive and continuous functions of y on [0, b] and
GJ(y) is differentiable. The prime symbol (U)
0
denotes differentia-
tion with respect to y.
The problem of solving Eq. (A.1) under the boundary conditions
(A.2) is called a self-adjoint boundary value problem, or a
SturmLiouville problem [17], in honor of the French mathema-
ticians Charles Sturm and Joseph Liouville, who investigated such
problem in 1830s. The adjoint differentiable operator of the
second order, L, is dened as
L
d
dy
GJ
d
dy
_ _
qR (A.4)
and Eq. (A.1) takes the form L(a) 0, which is called a self-adjoint
differential equation of second order if for any two functions a
1
(y)
and a
2
(y), that are twice differentiable and satisfy Eq. (A.2), we
have [22,23]:
_
b
0
a
2
La
1
a
1
La
2
dy 0 (A.5)
Self-adjointness implies certain mathematical symmetry and can
be ascertained through integrations by parts with due considera-
tions of the boundary conditions. Expanding Eq. (A.4) to GJ(d
2
/
dy
2
)+(GJ)
0
(d/dy)+qR, self-adjointness condition writes:
_
b
0
a
2
La
1
a
1
La
2
dy

_
b
0
fa
2
GJa
00
1
GJ
0
a
0
1
qRa
1
a
1
GJa
00
2
GJ
0
a
0
2
qRa
2
gdy

_
b
0
GJa
2
a
00
1
a
1
a
00
2
GJ
0
a
2
a
0
1
a
1
a
0
2
dy

_
b
0
d
dy
GJa
2
a
0
1
a
1
a
0
2
dy GJa
2
a
0
1
a
1
a
0
2
j
b
0
0
which proves Eq. (A.5).
ARTICLE IN PRESS
Table 2
Optimal wing designs with different number of panels (M

1,
^
b ^ c 1)
N (h

k
, b

k
), k 1,2,yN V

div
Gain
a
(%)
1 (1, 1) 1.5708 0.0
2 (1.25, 0.7), (0.417, 0.3) 1.680 7.01
3 (1.3547, 0.514), (0.846, 0.2785), (0.3325, 0.2075) 1.7063 8.63
4 (1.395, 0.43), (1.041, 0.248), (0.605, 0.192), (0.205, 0.13) 1.717 9.33
5 (1.4505, 0.25), (1.2545, 0.25), (0.9825, 0.215), (0.5525, 0.17), (0.1725, 0.115) 1.721 9.50
a
Gain (V
div
p/2)/(p/2), where p/2 is the baseline design value.
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 96
The values of q for which the above SturmLiouville problem,
described by Eq. (A.1), has a non-trivial solution a(y)60 are called
eigenvalues. The corresponding solutions are called the eigen-
functions. Let q
i
be an eigenvalue. We multiply Eq. (A.1) by the
corresponding eigenfunction a
i
(y) and integrate over [0, b]:
0
_
b
0
a
i
GJa
0
i

0
q
i
Ra
i
dy
GJa
i
a
0
i
j
b
0

_
b
0
GJa
0 2
i
q
i
Ra
2
i
dy
q
i

_
b
0
GJa
0 2
i
dy
_
b
0
Ra
2
i
dy
(A.6)
Since GJ(y) and R(y) are positive, and a
i
(y) is not constant
(otherwise it would be zero), we conclude that q
i
40. That is, all
the eigenvalues are positive. Now we try to nd a solution of
Eq. (A.1) that satises only the rst boundary condition a(0) 0
for a xed value of q; we denote this solution by a(y,q), i.e.
a0; q 0 (A.7)
As q changes, a(y,q) also changes, but nevertheless continues to
satisfy Eq. (A.7). The function a(y,q) can be represented in terms of
a power series in q, and hence is an analytic function of q for all
values of q. Differentiating with respect to y, then substituting for
y b, we obtain the polynomial
Dq a
0
b; q 0 (A.8)
Therefore, D(q) is a known function of one variable q. Every value
of q for which D(q) 0 is obviously an eigenvalue of our problem,
since for such value Eqs. (A.7) and (A.8) are satised simulta-
neously, i.e. both of the conditions in Eq. (A.2) are met. Thus, the
problem of existence of eigenvalues reduces to a study of the roots
of D(q). By using this fact, it can be shown that the problem has an
N set of real eigenvalues, which can be written as an increasing
sequence:
q
1
oq
2
oq
3
o. . . oq
n1
oq
n
o. . . (A.9)
of positive real numbers with a limiting value q
n
+Nas n-N.
Appendix B. The interior penalty function technique
In this method the original objective function F~x is augmen-
ted with terms, called penalty terms, such that as ~x approaches a
constraint surface, one term increases indenitely. Since the
algorithm seeks to minimize the value of the objective function, it
will try not to penetrate any constraint surface. Thus all
constraints are taken into consideration by representing them
by penalty terms in the objective function expression. The most
commonly used interior penalty function [18] is cast in the form
F~x; r F~x r

M
j1
1
G
j
~x
(B.1)
where F~x; r is the modied objective function G
j
~x is the Jth
constraint function and r is a multiplier. A sequence of
unconstrained minimization problems is solved with successively
decreasing values of r. The MATLAB optimization Toolbox [24]
offers routines that implement the interior penalty function
method via a built-in function named fminsearch.
References
[1] Ashley H. On making things the best-aeronautical uses of optimization. J
Aircraft 1982;19(1):528.
[2] Harvey HP, Johnson RR. Application of numerical optimization to the design of
wings with specied pressure distribution. NACA CR3238, February 1980.
[3] Rohl PJ, Mavris DN, Schrage DP. Combined aerodynamic and structural
optimization of a high-speed civil transport wing. AIAA paper-95-1222,
1995.
[4] Edwin L. Application of practical optimization techniques in the preliminary
structural design of a forward-swept wing. In: Proceedings of the second
international symposium on aeroelasticity and structural dynamics, Aachen,
West Germany, 1985.
[5] Grossman B, Strauch GJ, Eppard WM, Gurdal Z, Haftka RT. Integrated
aerodynamic/structural design of sail-plane wing. In: Proceedings of the
AIAA/AHS/ASEE aircraft systems design and technology meeting conference,
Dayton, OH, USA, October 2022, 1986.
[6] Rao SS. Optimization of airplane wing structures under taxiing loads.
J Comput Struct 1987;26(3):46979.
[7] Bisplinghoff RL, Ashley H. Principles of aeroelasticity. New York: Wiley; 1962.
[8] Flax HH. Aeroelasticity and utter in high speed problems of aircraft and
experimental methods. In: Donavan HF, Lawrence HR, editors. High speed
aerodynamics and jet propulsion, vol. VIII. Princeton: Princeton University
Press; 1961.
[9] Petre A. Theory of aeroelasticity: statics (in Romanian). Bucharest: Publishing
House of the Romanian Academy of Sciences; 1966.
[10] Librescu L, Thangjitham S. Analytical studies on static aeroelastic behavior of
forward-swept composite wing structures. J Aircraft 1991;28(2):1517.
[11] Librescu L, Song O. On the static aeroelastic tailoring of composite aircraft
swept wings modeled as thin-walled beam structures. J Compos Eng
1992;2(5):497512.
[12] Karpouzian G, Librescu L. Comprehensive model of anisotropic aircraft wings
suitable for aeroelastic analysis. J Aircraft 1994;31(3):70312.
[13] Done GT. Past and future progress in xed and rotary wing aeroelasticity.
Aeronaut J 1996;100(97):26979.
[14] Schuster DM. Inverse method for computation of structural stiffness
distributions of aeroelastically optimized wings. In: Proceedings of the
AIAA/ASME/ASME/ASCE/AHS structures, structural dynamics and materials
conference, Washington, DC, 1993; p. 20207.
[15] Butler R, Lillico M, Banerjee JR, Guo S. Optimum design of high aspect ratio
wing subject to aeroelastic constraints. In: Proceedings of the AIAA/ASME/
ASCE/AHS structures, structural dynamics and materials conference, New
York, 1995. p. 55866.
[16] Layton JB. Aeroelastic and aeroservoelastic tailoring with geometry variables
for minimizing the gust response of a cantilevered nite span wing. In:
Proceedings of the AIAA/ASME/ASCE/AHS structures, structural dynamics and
materials conference, New York, 1996. p. 1086103.
[17] Edwards E, Penney P. Differential equations and boundary value problems:
computing and modeling. Englewood Cliffs, NJ: N.J. Pearson, Prentice-Hall;
2004.
[18] Vanderplaats GN. Numerical optimization techniques for engineering design
with applications. New York: McGraw-Hill; 1984.
[19] Belegundy AA, Chruptla TR. Optimization concepts and applications in
engineering. Englewood Cliffs, NJ: Prentice-Hall; 1999.
[20] Zalzala AM, Fleming PJ. Genetic algorithms in engineering systems. Engle-
wood Cliffs, NJ: Prentice-Hall; 1997.
[21] Maalawi KY. Buckling optimization of exible columns. Int J Solids Struct
2002;39:586576.
[22] Meirovitch L. Principles and techniques of vibrations. Englewood Cliffs, NJ:
Prentice-Hall; 1997.
[23] Tolstov GP. Fourier series. New York: Dover; 1976.
[24] Venkataraman P. Applied optimization with MATLAB programming. New
York: Wiley; 2002.
[25] Pestel EC, Leckie FA. Matrix methods in elastomechanics. New York: McGraw-
Hill; 1963.
ARTICLE IN PRESS
L. Librescu, K.Y. Maalawi / Thin-Walled Structures 47 (2009) 8997 97

You might also like