You are on page 1of 41

ARTICLE IN PRESS

Progress in Aerospace Sciences 39 (2003) 425–465

Membrane wing aerodynamics for micro air vehicles


Yongsheng Lian, Wei Shyy*, Dragos Viieru, Baoning Zhang
Department of Mechanical and Aerospace Engineering, University of Florida, 231 Aerospace Building P.O. Box 116250,
Gainesville, FL 32611-6250, USA

Abstract

The aerodynamic performance of a wing deteriorates considerably as the Reynolds number decreases from 106 to 104 :
In particular, flow separation can result in substantial change in effective airfoil shape and cause reduced aerodynamic
performance. Lately, there has been growing interest in developing suitable techniques for sustained and robust flight of
micro air vehicles (MAVs) with a wingspan of 15 cm or smaller, flight speed around 10 m=s; and a corresponding
Reynolds number of 104 –105 : This paper reviews the aerodynamics of membrane and corresponding rigid wings under
the MAV flight conditions. The membrane wing is observed to yield desirable characteristics in delaying stall as well as
adapting to the unsteady flight environment, which is intrinsic to the designated flight speed. Flow structures associated
with the low Reynolds number and low aspect ratio wing, such as pressure distribution, separation bubble and tip
vortex are reviewed. Structural dynamics in response to the surrounding flow field is presented to highlight the multiple
time-scale phenomena. Based on the computational capabilities for treating moving boundary problems, wing shape
optimization can be conducted in automated manners. To enhance the lift, the effect of endplates is evaluated. The
proper orthogonal decomposition method is also discussed as an economic tool to describe the flow structure around
a wing and to facilitate flow and vehicle control.
r 2003 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
2. Theoretical and computational models for coupled fluid–structure interactions . . . . . . . . 429
2.1. Fluid solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
2.1.1. Governing equations for fluid flows . . . . . . . . . . . . . . . . . . . . . . . 429
2.1.2. Fluid flow solvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
2.1.3. The geometric conservation law . . . . . . . . . . . . . . . . . . . . . . . . . 430
2.1.4. Laminar–turbulent transition . . . . . . . . . . . . . . . . . . . . . . . . . . 431
2.2. Moving grid technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
2.2.1. Perturbation and transfinite interpolation methods . . . . . . . . . . . . . . . 431
2.2.2. Master/slave concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
2.3. Membrane structural solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
2.3.1. Mooney–Rivlin model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
2.3.2. Principle of virtual work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
2.3.3. Solution of the dynamic equations . . . . . . . . . . . . . . . . . . . . . . . 435

*Corresponding author. Tel.: +1-352-392-0961; fax: +1-352-392-7303.


E-mail addresses: wei-shyy@ufl.edu (W. Shyy).

0376-0421/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0376-0421(03)00076-9
ARTICLE IN PRESS
426 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Nomenclature DF skin friction drag


DP form drag
a angle of attack e span efficiency factor
AR aspect ratio I1 ; I2 ; I3 first, second, and third invariants of the Green
c chord length deformation tensor
C Green deformation tensor L lift force
c1 ; c2 membrane material property parameters S second Piola–Kirchhoff stress tensor
CL lift coefficient u chordwise velocity
CD drag coefficient U freestream velocity
CD;F skin friction drag coefficient v vertical velocity
CD;P form drag coefficient, or pressure drag coeffi- w spanwise velocity
cient x chordwise distance from the leading edge
cp pressure coefficient z distance from the root
D drag force Z half wing span

2.4. Coupling between fluid solver and structural solver . . . . . . . . . . . . . . . . . . . 436


3. MAV wing: aerodynamics and structural response . . . . . . . . . . . . . . . . . . . . . . . 437
3.1. Computational background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
3.2. Rigid wing aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
3.2.1. Vortical flow structure of the rigid wing . . . . . . . . . . . . . . . . . . . . . 438
3.2.2. Laminar boundary layer separation on rigid wing . . . . . . . . . . . . . . . . 439
3.2.3. Unsteady phenomena at high angle of attack . . . . . . . . . . . . . . . . . . 442
3.3. Membrane wing dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.3.1. Time synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.3.2. The geometric conservation law . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.3.3. Self-initiated membrane vibration . . . . . . . . . . . . . . . . . . . . . . . . 445
3.3.4. Comparison between membrane and rigid wings . . . . . . . . . . . . . . . . 446
4. Wing shape optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
4.1. Optimization and computational framework . . . . . . . . . . . . . . . . . . . . . . . 449
4.2. Rigid wing optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
4.3. Flexible wing optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
5. Endplate effects on MAV aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
6. Reduced-order representation using proper orthogonal decomposition . . . . . . . . . . . . 454
7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462

1. Introduction conventional flight vehicles, MAV gains certain favor-


able scaling characteristics, including low inertia and
Micro air vehicles (MAVs) with a maximal dimension reduced stalling speed [85] (defined as the minimum
of 15 cm or less and a flight speed of 10 m=s are of speed at which the wing can produce sufficient lift force
interests to both military and civilian applications. for the flight). However, in terms of aerodynamics,
Equipped with a video camera or a sensor, these vehicles control, range and maneuverability, many outstanding
can perform surveillance and reconnaissance, targeting, issues are still present. First, due to its low flight speed
and bio-chemical sensing at a remote or otherwise and small dimensions, MAV flies in a low Reynolds
hazardous location [85]. With the rapid progress made number regime (104 –105 ), which is often accompanied
in structural and material technologies, miniaturization by laminar boundary layer separation, transition, and
of power plants, communication, visualization, and low lift-to-drag ratio [85]. Second, the low aspect ratio
control devices, several groups, e.g., Grasmeyer and wing creates strong vortical flow structures, such as tip
Keennon [26], and Ifju et al. [36] have developed vortices [53,52], which have twofold impacts on the lift
successful MAVs. As its size is reduced from the force. On the one hand, tip vortices lower the lift force
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 427

by reducing the effective angle of attack [2,85]; on the chordwise battens [35]. A membrane material is bonded
other hand, tip vortices provide additional lift force by to the spar and battens. The membrane wing has a span
creating a low-pressure zone [52,68]. Third, the low of 15:2 cm; a root chord of 13:7 cm; a mean chord of
aspect ratio wing is quite susceptible to rolling instabil- 9:4 cm; and a wing area of 160 cm2 : The wing has a
ities, which becomes even serious because of the 7.5% camber at the root which decreases to 2% at the
existence of tip vortices. Fourth, the fluctuations in tip. In this configuration, the maximum camber is
wind speed, which can be comparable to MAV’s flight located at 27% chord at the root.
speed, make both the instantaneous flight Reynolds One advantage for the membrane wing is that it can
number and angle of attack vary substantially [85]. facilitate passive shape adaptation, which results in
In the development of MAVs, there exist two delayed stall [85,110]. Fig. 2, adopted from Waszak et al.
approaches, one utilizing flapping wings and the other [110], compares the lift curves versus angles of attack for
employing fixed wings. In the flapping wing area, rigid and membrane wings with configurations similar to
pioneering work has been done by Weis-Fogh [111], that shown in Fig. 1. Under modest angles of attack,
and Lighthill [57]. Recent works, both in experiments both rigid and membrane wings demonstrate similar lift
and simulations, are documented by Chasman and characteristics with the stiffer wings having slightly
Chakravarthy [10], DeLaurier [14], Smith [91], higher lift coefficient. However, it is clear that the
Dickenson et al. [15], Ellington [19], Jones and Platzer membrane wings stall at higher angles of attack than
[43–45], Katz [48], Kamakoti et al. [47], Liu and the rigid wing. This aspect is a key element in enhancing
Kawachi [60], Vest [105], Vest and Katz [106], and the stability and agility of MAVs. Some aspects of low
Wang [109]. A recent review of the characteristics aspect ratio, low Reynolds number rigid wing aero-
of both flapping wings and MAVs has been given by dynamics have been presented by Torres and Mueller
Shyy et al. [85]. The spectrum of animal flight with [102]. However, there are important differences between
flapping wing is presented in an interesting article by rigid and membrane wings. At low angles of attack the
Templin [97]. aeroelastic wings behave like rigid wings with similar
This review will focus on the three-dimensional aspect ratio. The lift curve slope in Fig. 2 is approxi-
aerodynamics associated with fixed wing MAVs. More mately 2.9 with the prop pinned. The lift curve slopes of
specifically, we focus on the recent progress made in similar rigid wings studied in [102] at comparable
employing flexible membrane wings for MAV applica- Reynolds number and aspect ratio (Re ¼ 7  104 ;
tions. Earlier reviews on this topic, largely on the two- AR ¼ 2) are approximately 2.9 as well. However, these
dimensional flow, can be found in Refs. [85,88]. wings have stall angles between 12 and 15 : The stall
Fig. 1 shows a 15 cm MAV with a flexible wing angles of the flexible wings are between 30 and 45 and
designed by Ifju and coworkers [36]. The vehicle is are similar to that of much lower aspect ratio rigid wings
powered by an electric motor. It has a total mass of 40 g (AR ¼ 0:5–1.0). However, the very low aspect wings
and a flight speed between 20 and 40 km=h for about exhibit lower lift curve slopes of 1.3 to 1.7. Hence, the
15 min while carrying a video camera and transmitter. flexible wings appear to combine the desirable perfor-
The membrane wing consists of unidirectional carbon mance of rigid wings with higher and lower aspect
fiber prepreg laminate forming a leading edge spar and ratios, by exhibiting the stall behavior similar to rigid
wings with aspect ratio 0.5–1.0, and the lift generating
capability of rigid wings with aspect ratio 2.0 [110].

Fig. 2. Lift coefficient versus angle of attack (from Waszak


Fig. 1. 15 cm MAV with membrane wing. et al. [110]).
ARTICLE IN PRESS
428 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Another advantage of a flexible membrane wing is that interactions between the fluid and structure is desirable
a flexible wing can adapt to the wind gust and provide to provide more detailed insight into the underlying
a smoother flight platform. This last aspect has been physics. Aeroelastic analysis methods coupling compu-
verified by wind tunnel experiments [86]. tational fluid dynamics (CFD) solvers with computa-
The low Reynolds number condition presents numer- tional structural dynamics (CSD) solvers provide an
ous challenges on MAVs because of the concomitant effective tool for the membrane wing study. Although
susceptibility of their lift surfaces to flow separation and both CFD and CSD have achieved great success
low lift-to-drag ratio. In the range of Reynolds number individually, a coupled computation involving both is
of 104 –106 ; complex flow phenomena often take place on still a challenging task [82]. To date, the study of
the upper wing surface, such as laminar boundary layer membrane and fluid flow interactions is limited. Jackson
separation, transition, and reattachment. Valuable and Christie [38] adopt a three-dimensional potential
information in these regards is reviewed by Lissaman flow-based solver and a membrane wing model to
[58] and Tani [96], as well as addressed in several analyze the aeroelastic behavior of marine sails. Smith
proceedings [66,67]. Active control approaches with and Shyy [93] and Shyy et al. [89] present a computa-
microelectronicsmechanical systems (MEMS) for MAVs tional approach to model the interaction between a two-
are discussed by Gad-el-Hak [22]. Lian and Shyy [52] dimensional flexible membrane wing and surrounding
have studied the flow separation on the MAV wing and viscous flows. A combined numerical and experimental
its impacts on aerodynamic performance. analysis of flow over a two-dimensional flexible sail is
The low aspect ratio wing exhibits clear vortical conducted by Lorillu and Hureau [62]. Recently, Lian
structures. Consequently, it has a higher angle of attack et al. [54] have proposed a dynamic membrane model to
because tip vortices provide additional lift force by perform coupled fluid and structure computations
creating low-pressure zones, which is similar to delta between a three-dimensional membrane wing and a
wings. However, the low aspect ratio also increases the viscous fluid flow. In their work, the flow equations are
induced drag. More importantly, the vortical flow causes the full Navier–Stokes equations for the incompressible
rolling instability of the small vehicle, especially when fluid flow, the structural equations consider the large
the vortex strength is not equal on the two sides. In this strain of the structure and therefore nonlinear. A further
aspect, relevant experimental work is done by Pelletier improved computational capability is developed later by
and Mueller [74], and numerical investigations are Lian and Shyy [52] by adopting a more efficient time
reported by Lian and Shyy [52] and Viieru et al. [107]. marching flow solver to compare the membrane and
Unlike a rigid wing, a membrane wing exhibits self- rigid wings performance.
initiated unsteady response even under a steady state This paper reviews the recent progress in aerody-
freestream. Such vibrations and associated shape namics of MAVs with a special interest in membrane
changes impact the wing aerodynamics, which, in turn, wings. We will examine the following topics:
affects the membrane dynamics. For example, Waszak
et al. [110] have conducted wing tunnel experiments and * Computational strategy for fluid and membrane wing
found that the membrane wings exhibit vibration at interaction. In low Reynolds number condition.
Oð102 Þ Hz under a steady free stream. Similar observa- * Aerodynamics of a membrane and corresponding
tion has also been reported numerically by Lian and rigid wings.
Shyy [52]. Such reciprocal action results in a fluid and * Optimization of a membrane wing.
structure interaction problem. * Endplate effects on the low aspect ratio MAV wing.
Other effects have also been conducted toward * POD analysis of MAV aerodynamics.
understanding the MAV performance. The reduced-
order representation of the three-dimensional, time Each of these areas will be discussed in detail with
dependent flow solutions based on the proper orthogo- examples. The rest of the paper is organized as follows:
nal decomposition (POD) approach for the membrane In Section 2, a summary of the computational strategy
aerodynamics is analyzed by Schmit et al. [81] and for the coupled fluid and structure interactions is given,
Zhang et al. [114]. Such techniques can be extremely which includes the governing equations for the structure
valuable to offer an economic description of the salient and fluid flow, a moving grid technique, and a coupling
features of the MAV aerodynamics as well as to strategy between the two field equation solvers.
facilitate the development of flow and vehicle control Section 3 focuses on the recent investigation on MAV
strategies. wing aerodynamics. The rigid wing performance is
To optimize the vehicle performance, it is critical to evaluated first for the comparative purpose, with a
understand the interplay between the structural- and focus on aerodynamic characteristics associated with
aero-dynamics. Experiments have shed light on the the low aspect ratio and low Reynolds number wing,
interaction mechanics between fluid and structure; including tip vortex, leading edge separation, and the
however, a numerical capability for treating the intrinsic resulting unsteady phenomenon. Issues associated with
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 429

computations of the fluid and structure interaction, such Steady flow computations have achieved significant
as time synchronization and geometric conservation law success in their development and applications, however,
[100] are assessed. The pressure distribution, shape little attention has been given to the systematic analysis
change, and the resulting effective angle of attack of of unsteady computations [80,82]. Several factors are
the membrane wing are assessed in comparison to those responsible for this situation. First, steady CFD is still
of the rigid wing. the mainstream aerodynamic analysis, and the techni-
Section 4 reviews shape optimization of a flexible ques used to enhance the performance of steady CFD
membrane wing with the gradient-based method. Since a computations may not be suitable to unsteady computa-
direct optimization of a membrane wing requires tions. For example, in steady computations, a variety of
substantial computational resource and time, a rigid techniques have been proposed to eliminate the low
wing is used as a surrogate. The membrane wing frequency transients to accelerate the convergence to
performance is then evaluated based on the surrogate steady state; however, low frequency phenomena are
model. intrinsic to many fluid and structure problems, and such
Section 5 assesses the impact of an endplate, with the techniques will be inappropriate to the unsteady analysis
goal of enhancing the lift by reducing the vortex effect. [82]. Furthermore, the applicability of the turbulence
In Section 6, a POD analysis is applied to investigate model to unsteady flows is not well understood [80].
the aerodynamics of a flexible membrane wing. This However, there are numerous approaches that can be
method has the potential to provide an efficient adopted for unsteady flow computations. Pulliam [75]
description of the dynamics of the fluid and structure has incorporated subiteration to improve time accuracy
interaction problems. of conventional implicit schemes. Jameson [39] and
Rumsey et al. [80] have developed a dual time stepping
technique in the context of multigrid methodology to
2. Theoretical and computational models for coupled enhance the efficiency and accuracy of traditional
fluid–structure interactions factored schemes. Issa [37], and Oliveira and Issa [72]
propose the Pressure-Implicit with Splitting of Opera-
In this section, the elements of developing a computa- tors (PISO) method to improve the convergence.
tional capability to handle fluid and membrane interac- For relevant information of these and some of the
tions are discussed. First, the fluid flow solver and most frequently adopted approaches, see Shyy and
physical models of the fluid dynamics are presented. The Mittal [87].
Navier–Stokes equations for the incompressible flow on Two of the most popular three-dimensional Navier–
curvilinear coordinates are employed. A brief introduc- Stokes equations for incompressible flows in curvilinear
tion is also given to the geometric conservation law [100] coordinates employ the so-called SIMPLE method
in the context of fluid and structure interaction [73,84] and the PISO method [37,72,98]. Both belong
problems. Under low Reynolds number conditions, the to the pressure-based approach [84] which devises an
flow is prone to separate which is usually accompanied artificially derived pressure (correction) equation by
by a quick transition from laminar flow to turbulent manipulating the mass continuity and momentum
flow. A transition model for general practical applica- equations.
tions is not yet available, a brief of current used
laminar–turbulent transition models is offered. We will
review the moving grid techniques used for both the 2.1.1. Governing equations for fluid flows
coupled fluid–structure problem and the shape optimi- The three-dimensional Navier–Stokes equations for
incompressible flow in curvilinear coordinates [84] in
zation process. We then discuss the structural solver,
with a special attention to a hyperelastic nonlinear strong conservative form can be written
dynamic membrane model. @U @V @W
þ þ ¼ 0; ð1Þ
@x @Z @z
2.1. Fluid solver

During flight, the membrane wing undergoes shape @ð JruÞ @ðrUuÞ @ðrVuÞ @ðrWuÞ
change under the external forces; meanwhile, its shape þ þ þ
@t @x @Z @z
variation affects the pattern and structure of its
@hm i
surrounding fluid flow. The membrane wing vibration ¼ ðq11 ux þ q12 uZ þ q13 uz Þ
@x J
is observed both experimentally [110] and numerically
@hm i
[54]. To understand the membrane wing dynamics, it is þ ðq21 ux þ q22 uZ þ q23 uz Þ
desirable to have computational capabilities to accu- @Z J
rately capture the transient behavior of fluid flow and @hm i
þ ðq31 ux þ q32 uZ þ q33 uz Þ
structural dynamics. @z J
ARTICLE IN PRESS
430 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

  they are defined as the following:


@ @ @
 ð f11 pÞ þ ð f21 pÞ þ ð f31 pÞ
@x @Z @z ’ þ f12 ðv  yÞ
U ¼ f11 ðu  xÞ ’ þ f13 ðw  z’Þ;
þ G1 ðx; Z; zÞ:J; ð2Þ ’ þ f22 ðv  yÞ
V ¼ f21 ðu  xÞ ’ þ f23 ðw  z’Þ;
’ þ f32 ðv  yÞ
W ¼ f31 ðu  xÞ ’ þ f33 ðw  z’Þ; ð7Þ
@ð JrvÞ @ðrUvÞ @ðrVvÞ @ðrWvÞ where, for example, x’ is the grid velocity component in
þ þ þ
@t @x @Z @z the x-direction which can be approximated by the first-
h
@ m i order Euler method,
¼ ðq11 vx þ q12 vZ þ q13 vz Þ
@x J x  x0
@hm i x’ ¼
Dt
; ð8Þ
þ ðq21 vx þ q22 vZ þ q23 vz Þ
@Z J
where Dt is the time step and the superscript 0 refers to
@hm i
the previous time level. Similarly, the values of the other
þ ðq31 vx þ q32 vZ þ q33 vz Þ
@z J two components y; ’ and z’ can be estimated.
 
@ @ @
 ð f12 pÞ þ ð f22 pÞ þ ð f32 pÞ
@x @Z @z 2.1.2. Fluid flow solvers
þ G2 ðx; Z; zÞ:J; ð3Þ The SIMPLEC method [103] is a variant of the
SIMPLE method originally proposed by Patankar [73]
in Cartesian coordinates, and its extension to curvilinear
@ð JrwÞ @ðrUwÞ @ðrVwÞ @ðrWwÞ coordinates is discussed by Shyy [84] to accommodate
þ þ þ
@t @x @Z @z the complex geometry computation. The SIMPLEC
h
@ m i
¼ ðq11 wx þ q12 wZ þ q13 wz Þ method uses coordinated under-relaxations for the
@x J momentum and pressure corrections to improve the
@hm i
convergence which is intrinsically slow in the original
þ ðq21 wx þ q22 wZ þ q23 wz Þ
@Z J SIMPLE method. The SIMPLEC method has also been
@hm i used to solve the moving boundary problems [51,89].
þ ðq31 wx þ q32 wZ þ q33 wz Þ The PISO method is a non-iterative method to handle
@z J
  the pressure–velocity coupling which splits the process
@ @ @
 ð f13 pÞ þ ð f23 pÞ þ ð f33 pÞ of solution into a series of predictor and corrector steps.
@x @Z @z
The PISO method is generally more efficient for
þ G3 ðx; Z; gÞ:J; ð4Þ transient flow computations [37,98].
To solve the Navier–Stokes equations, proper bound-
where u; v; and w are the velocity components in the
ary conditions are required. At the interface between
x-, y-, and z-directions, respectively, m is the viscosity,
fluid and solid structure, the no-slip condition is applied,
G1 ; G2 ; and G3 are the body force components in the x-,
which requires that the fluid velocity at the membrane
y-, and z-directions per unit volume. The coefficients
surface match the surface velocity of the structure. These
appearing in the equations are expressed as follows:
velocity boundary conditions are provided by comput-
2
q11 ¼ f11 2
þ f12 2
þ f13 ; ing the displacement of the solid structure.
q12 ¼ q21 ¼ f11 f21 þ f12 f22 þ f13 f23 ;
2.1.3. The geometric conservation law
2 2 2
q22 ¼ f21 þ f22 þ f23 ; When body-fitted curvilinear coordinates are used in
q13 ¼ q31 ¼ f11 f31 þ f12 f32 þ f13 f33 ; the computation, a transformation maps a physical flow
2 2 2 region ðx; y; zÞ onto a computational space ðx; Z; zÞ where
q33 ¼ f31 þ f32 þ f33 ;
the conservation laws are carried out. To facilitate the
q23 ¼ q32 ¼ f31 f21 þ f32 f22 þ f13 f23 : ð5Þ coordinate transformation, the Jacobian matrix J is
introduced. The Jacobian matrix is defined as in Eq. (6).
The Jacobian transformational matrix is defined as the
The determinant of J represented a volume element in
follows:
the transformed coordinates. In moving grid problems,
@ðx; y; zÞ the computational grid needs to be updated with time;
J¼ : ð6Þ meanwhile, the Jacobian J needs to be updated
@ðx; Z; zÞ
simultaneously. Specific procedures are required to
In the governing equations, U is the flux through the compute the effective value of J; otherwise, errors arise
control surface whose normal direction is x; V and W due to the inconsistency in evaluating geometric
are the fluxes through Z and z control surfaces. When the quantities and the adding of non-physical quantities.
governing equations are considered under a moving grid As pointed by Thomas and Lombard [100], in this
framework, the grid velocities should be considered; Jacobian updating process, the following geometric
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 431

conservation law (GCL) needs to be satisfied, of the subject will be redistributed along the grid line,
Z Z which connects aeroelastic surface to the outer bound-
d
J dx dZ dz ¼ ðr Ws Þ dx dZ dz; ð9Þ ary. This simple method works quite well for single
dt V V
block and modest deflection; however, when applied to
where V is the volume bounded by a close surface S; Ws multiblock grid arrangements, extensive intervention is
denotes the local velocity of the boundary surface S: often needed. The spring analogy method is first
Thomas and Lombard [100] propose to evaluate J while proposed for unstructured grid [4], and later extended
maintaining the geometric conservation law by setting to structured grid by Robinson et al. [79]. This method
r ¼ 1; V ¼ 0 from the continuity equation, i.e., regards all grids as connected by springs, each with the
Jt þ ð Jxt Þx þ ð JZt ÞZ þ ð Jzt Þz ¼ 0: ð10Þ stiffness inversely proportional to the spacing between
target vertex and its neighbors. Compared with other
Implication of the GCL with fluid-structure interaction currently used grid regeneration method, the spring
problems has been discussed by Shyy et al. [89]. analogy method needs more memory and CPU time.
Direct transfinite interpolation by Eriksson [21] is a
2.1.4. Laminar–turbulent transition popular method because of its better efficiency for
A reliable computation of the boundary layer transi- structured grid. Hartwich and Agrawal [32] also propose
tion remains one of the most challenging demands in the Master/Slave concept to expedite the grid regenera-
aerodynamics. Since the transitional separation bubbles tion process and minimize the user intervention.
and their losses play an important role in the pressure
and drag distributions, an accurate representation of 2.2.1. Perturbation and transfinite interpolation methods
both laminar and turbulent separated flow is critical For small amplitude displacement of the moving
when drag prediction needs to be accurately predicted. boundary, change of shape occurs only on one face of
However, the fundamental fluid-dynamics problem of the computational block. For such a case, a simple but
transition to turbulence has resisted simple phenomen- efficient one-dimensional perturbation method [55,78]
ological description. There is still no comprehensive can efficiently regenerate the grid. The one-dimensional
theory of transition. perturbation method obeys the following formula:
The range of existing transition prediction methods
extends from simple empirical relationships through xnew
i ¼ xold
i þ Siold ðxnew
s  xold
s Þ; ð11Þ
those based on parallel and linear stability, like the eN where xi represents the location of interior grid point, xs
methods [18,76], or linear or nonlinear parabolized represents the location of grid point on a boundary, and
stability equations (PSE) [33], or the low-dimensional S represents the normalized arc length along the radial
models [77] to direct numerical simulation like large- mesh line measured from the outer domain, which is
eddy simulation techniques. given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2.2. Moving grid technique Sil¼1 ðxlþ1  xl Þ2 þ ð ylþ1  yl Þ2 þ ðzlþ1  zl Þ2
Si ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi:
For problems involving interactions between fluid Snl¼1 ðxlþ1  xl Þ2 þ ð ylþ1  yl Þ2 þ ðzlþ1  zl Þ2
flows and moving structures, the geometry of the solid ð12Þ
object is not known a priori. It is necessary to track in
the course of computation, the geometry, the field To use this one-dimensional perturbation method, one
equations, and the interfacial condition to ensure that all needs to know the displacements of the two ending
requirements are met. To facilitate the solution of such points. The positions of the interim points are solely
moving boundary problems, the moving grid technique determined by the displacements of these ending points.
(or dynamic grid technique) is employed to adjust the Such displacement can be considered as the source of a
grid dynamically along with the geometric updates. It is perturbation. Intuitively, a perturbation will spread
desirable but difficult to develop an automated regrid- throughout the whole domain while exerting more effect
ding procedure to ensure that the new grid not only on nearer points. Based on Eq. (11), a more complicated
matches the geometric changes but also maintains 3-stage perturbation method, analogous to the transfi-
satisfactory characteristics such as non-overlapping, nite interpolation (TFI) method [21], is proposed for
smooth and not excessively skewed. The moving grid three-dimensional problems treated by the single-block
technique can also be used by shape optimization [53]. approach. Unlike the original TFI method, these
For moving grid computations, several alternative perturbation methods consider the original grid dis-
approaches have been suggested to treat grid redistribu- tribution, and hence can preserve the grid quality.
tions. Schuster et al. [83] use an algebraic shearing Take a two-dimensional case as an example. In Fig. 3,
process in their static aeroelastic analysis of a fighter a rectangle undergoes a shape change. To regenerate the
aircraft; the basic idea is to assume that the displacement grid, the perturbation method proceeds as a 2-stage
ARTICLE IN PRESS
432 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

where DI ¼ ð1  Si ÞDP þ Si DQ; DJ ¼ ð1  Sj ÞDS þ


Sj DT; Si and Sj are the normalized arc length along
mesh lines PQ and ST; respectively. Displacements in
the other direction can be computed similarly.
After the first stage, the four corner points have
matched their new locations; however, the interim edges
do not necessarily match their final positions. The
second stage is to match the interim edges to their final
edge locations. The difference between the interim and
final edge locations are taken to be the source of
perturbation. In this case, since the perturbations in
both directions are dependent, the contributions from all
individual directions are added [See Fig. 3(c)]. For three-
dimensional problem, the 3-stage process is adopted by
Lian et al. [55].

2.2.2. Master/slave concepts


The aforementioned perturbation method can be
implemented efficiently for single block and small
displacement problems. However, when a multi-block,
structured grid or large interface displacement is
involved, this method needs to be enhanced by other
techniques. The Master/Slave concept acts as a useful
approach to maintain grid quality and to prevent
potential grid cross-over.
Although unstructured grid has been used to solve
fluid dynamics problems by different researchers [4,28],
here, our attention is limited to structured grid only. For
a multi-block structured grid, for simplicity, CFD
software often requires point-matched grid block inter-
faces. A basic requirement for regenerating grid in the
moving boundary problems is to maintain the point-
matched interface. While this updating process is easily
satisfied by those block surfaces that coincide with the
body surface, the remaining block surfaces need to be
Fig. 3. Schematic of the perturbation method for moving grid:
handled with due care. In order to use the perturbation
(a) ABCD changes to A0 B0 C0 D0 ; (b) matching corner points;
method, one needs to specify the new positions of the
(c) matching edge points.
off-body surface points, or, at least, the vertices of such
block surfaces.
process. The first stage is to match corner points; that is, When an object changes its shape/position, the master
to move the corner point A to its new position A0 ; and B points, which are located at the body surface, move first,
to B0 ; etc. [See Fig. 3(b)]. Once the corner points match and then affect the distribution of the slave points, i.e.,
their new positions, interim edges are generated based the off-body points. While the task of redistributing
on Eq. (11). To adjust the position of an internal point interior grid can be challenging, in practice, it is more
O; effects from the four corner points need to be difficult to move the vertices of each block if a point-to-
accounted for in a coordinated manner. In this point match between two abutting blocks without
approach, P0 ; Q0 ; S 0 ; and T 0 ; which connect O0 to its overlap is required. If two abutting blocks have
boundary, will be first adjusted. The displacements congruent interfaces, such as the interface of block 2
of the four edge points in, e.g., x-direction and block 3 in Fig. 4(a), once the corner vertices are
ðDP; DQ; DS; DTÞ can be calculated with Eq. (11), and determined, the edge points and the interior points can
the movement of O0 is evaluated as a combination of be uniquely settled based on Eq. (11). However, when
these displacements. Specifically, one can adopt the the two neighboring grid blocks do not share identical
following expression: interfaces, such as that of block 1 and block 2 shown in
Fig. 4(a), this procedure can cause discontinuity at the
absðDIÞDI þ absðDJÞDJ interface, because the regriding procedures at the inter-
DO0 ¼ ; ð13Þ
maxðabsðDIÞ þ absðDJÞ; 106 Þ face are based on different stencils for different blocks.
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 433

(a) (b)

(c) (d)

(e)
Fig. 4. Illustration of moving grid scheme: (a) initial geometry; (b) side view of the grid; (c) new shape; (d) distribution of the new grid;
(e) same test but with high stiffness.

Slaving the vertices of a grid block to their master points space ði; j; kÞ into a one-dimensional data structure, for
can avoid creating undesirable grid discontinuities. each grid point it has an identification number l defined
The Master/Slave coupling is based on the distance by
between an off-body point and a surface point (Master
l ¼ boffðbnÞ þ i þ imax  ð j  1Þ þ imax  jmax  ðk  1Þ;
point). The distance is given by
ð15Þ
rj ¼ ½ðxv  xb Þ2 þ ð yv  yb Þ2 þ ðzv  zb Þ2 0:5 ;
j~ ð14Þ
where boffðbnÞ is the identification number of the first
where the subscript v represents off-body vertex, and, b; point in block bn in the one-dimensional array, imax ; jmax ;
the body surface point. For each off-body vertex, the are dimensions in i and j direction, respectively. In this
nearest body surface point is defined as its master point. way a slave point is associated with its master point by
To simplify the connection between the master and slave storing its master’s identification number into its
points, one can convert the three-dimensional parameter address.
ARTICLE IN PRESS
434 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Once the relation between a master point and a slave The studies of a two-dimensional membrane with
point is established, the next step is to determine how the fixed leading and trailing edges were originally con-
slave point moves according to the influences from its ducted by Voelz [108], Thwaites [101], and Nelsen [69].
master point. Intuitively, the nearer vertex will have These works consider inextensible membranes sur-
more effect (displacement) than those far away. Also, to rounded by steady, two-dimensional, irrotational flows.
avoid causing a lower surface to cross over its opposite As a consequence of the inextensible assumption, when
face when the movement is relatively large, one needs to the camber and the incidence angle are small, this
scale the movement. A simple but effective formula is problem is linearized and expressed compactly in a non-
expressed as follows: dimensional integral form
Z
x* s ¼ ~
~ x* m  ~
xs þ yð~ xm Þ: ð16Þ CT 1 d2 ð y=aÞ=dx2 dð y=aÞ
1 dx ¼ ; ð20Þ
2 0 2pðx  xÞ dx
Subscripts m and s represent master and slave,
respectively, and the tilde ðBÞ indicates the new where yðxÞ defines the membrane profile as a function of
position. y is the decay function; here, the Gaussian x coordinate, a is the incidence angle, and CT is the
distribution function is used, given by tension coefficient. This equation, referred to as the ‘‘sail
y ¼ expfb min½500; dv=ðe þ dmÞ g; ð17Þ equation’’ by Newman [70], completely defines the
linearized theory of inextensible membrane wings in
where steady, inviscid flow fields. Based on the work of Voelz
dv ¼ ðxv  xm Þ2 þ ð yv  ym Þ2 þ ðzv  zm Þ2 ; ð18Þ [108], the sail equation has been successfully solved
analytically and numerically. Newman [70] has given a
dm ¼ ðx* m  xm Þ2 þ ðy* m  ym Þ2 þ ð*zm  zm Þ2 : ð19Þ comprehensive review of the earlier works related to
membrane wing aerodynamics.
Coefficient b affects the stiffness. A larger b causes the The development of a three-dimensional membrane
block to behave more like a rigid body (See Fig. 4(e)). model introduces several complicating factors not
More details about master and slave concept can be encountered in a two-dimensional analysis. First, unlike
found in Refs. [32,55]. the ‘‘sail equation’’, the tension in a three-dimensional
membrane is no longer a single constant value but is best
2.3. Membrane structural solver described by a state of biaxial tension along the lines of
principal stresses [38]. Second, the geometric and
The deformations of membrane structures are often material properties vary along the spanwise direction,
large and thus inherently nonlinear. Qualitative descrip- which have to be described in details. Third, if one of the
tions of behavior of thin membranes are often more principal tensions vanishes, the membrane will be
complex than those of three-dimensional bodies. Until compressed and wrinkled, which further complicates
recently, their analysis has primarily relied on trial and the analysis. Considering the complexity of the three-
error. Green and Adkins [27] are the first to derive a dimensional problem, simplifications are made in the
general non-linear theory for membranes. The mem- early works in this area.
brane theory has two fundamental assumptions because Oden and Sato [71] have presented a finite element
of their thinness [41]. First, the membrane tension method for the analysis of large displacements and finite
stiffness is much larger than its bending stiffness, so the strains in elastic membranes of general shape. In their
stress couple effect can often be neglected. Second, the work the static equilibrium of an inflated membrane is
ratio of thickness to radius of the curvature is small, considered. A static analysis can be considered as a
which effectively decouples the strain-displacement special case of a dynamic analysis in which the inertia
relation from the curvature tensor. Practically speaking, effect is included. Works on the dynamic analysis of
there exist two approaches to model the response of a membranes before 1991 are reviewed by Jenkins and
membrane structure. The first approach is to idealize a Leonard [41]; most recent works are reviewed by Jenkins
plate or shell structure as a membrane away from its [40]. Verron et al. [104] have performed a combined
boundaries, in which the stress couples are neglected in numerical and experimental study of the dynamic
its interior region. By formally equating the plate inflation of rubber-like membrane. Under compression,
stiffness to zero, this membrane model can be described the flexible membrane experiences wrinkle, this topic is
.
by the Foppl–von K!arm!an theory [36,41]. In the second studied in the work of Ding et al. [16] in which they
approach, the structure cannot sustain stress couples perform a numerical study of partial wrinkled mem-
anywhere. The second approach is adopted by Lian et al. branes. The stability of fluid flow past a membrane is
[54] because it is appropriate for MAV computations. reported by Thaokar and Kumaran [99].
In the following, further comment on membranes Motivated by their interests in MAVs, Lian et al. [54]
will be given with an emphasis on their aerodynamic propose a three-dimensional membrane model to
applications. account for the membrane dynamics. The membrane
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 435

material considered obeys the hyperelastic Mooney– respectively, have the following forms
Rivlin model [65]. They use the Green–Lagrange strain 2 3
C11 ðtÞ C12 ðtÞ 0
tensor for the description of large strains. The dynamic 6 7
response of such a membrane is described by a system of CðtÞ ¼ 4 C12 ðtÞ C22 ðtÞ 0 5;
second-order time dependent equations. A finite element 0 0 C33 ðtÞ
2 3
method with triangular element for spatial discretization S11 ðtÞ S12 ðtÞ 0
is utilized by them. For time integration, they adopt the 6 7
SðtÞ ¼ 4 S12 ðtÞ S22 ðtÞ 0 5: ð25Þ
implicit Wilson-y method. In the following a brief review
of their membrane model is given. 0 0 0
Therefore, the hydrostatic pressure is determined by the
condition that S33 ¼ 0;
2.3.1. Mooney–Rivlin model
For an initially isotropic membrane, Green and p ¼ 2ðc1 þ c2 I1  c2 l23 Þl23 ; ð26Þ
Adkins [27] show that there exits a strain energy
where l3 is defined by
function W which can be expressed as the following
form pffiffiffiffiffiffiffiffiffiffiffiffi hðtÞ
l3 ¼ C33 ðtÞ ¼ ; ð27Þ
h0
W ¼ W ðI1 ; I2 ; I3 Þ; ð21Þ in which hðtÞ and h0 are the membrane thickness in the
deformed and undeformed configurations, respectively.
where I1 ; I2 ; and I3 are the first, second, and third
invariants of the Green deformation tensor C. If the 2.3.2. Principle of virtual work
material is incompressible, namely, I3 =1, then the strain Consider an elastic membrane, which in its initial state
energy is a function of I1 and I2 only. The following occupies a finite region in space. The position of the
linear form is useful in the study of a membrane region can be specified by an orthogonal coordinate
system XYZ which is referred to as global coordinates.
W ¼ c1 ðI1  3Þ þ c2 ðI2  3Þ; ð22Þ In global coordinates, the general form of the principle
of virtual work in the Lagrangian description is given by
Z Z
where c1 and c2 are the two material parameters. A . dV þ
dUT ðtÞr0 UðtÞ dE : S dV
material that obeys Eq. (22) is known as the Mooney– V0
Z
V0
Rivlin material [65], and the model is the Mooney–
 dUT ðtÞPðtÞ dS ¼ 0; 8@U; ð28Þ
Rivlin model. The Mooney–Rivlin model is one @V0
of the most frequently employed hyperelastic models
because of its mathematical simplicity and relatively where V0 and @V0 are, respectively, the volume and the
good accuracy for reasonably large strains (less than boundary surface of the undeformed membrane, r0 is
150%) [65]. the membrane density, UðtÞ is the displacement vector in
For an initially isotropic membrane, the general global coordinates, dUðtÞ is the virtual of vector U, the
stress–strain relationship is written as superscript T refers to the transverse of a vector or a
.
tensor, UðtÞ is the acceleration vector, P(t) is the
generalized external surface force, E is the Green–
@W
S ¼ pC1 þ 2 ; ð23Þ Lagrange strain tensor, S is the second Piola–Kirchhoff
@C
stress tensor, and dE : S is the scalar product of two
tensors. In a rectangular Cartesian coordinates, the
where S is the second Piola–Kirchhoff stress tensor, p is scalar product is defined by
the hydrostatic pressure. If the membrane is incompres-
sible, then the stress–strain relation can be further dE : S ¼ dEij Sji ; ð29Þ
simplified as follows: with summation on the repeated indices.

S ¼ pC1 þ 2½ðc1 þ c2 I1 ÞI  c2 C ; ð24Þ 2.3.3. Solution of the dynamic equations


The governing equation for the membrane response
where I is 3-by-3 identity matrix. In the membrane under external load is
theory the stress field is essentially treated as two- . þ Fint ¼ Fext ;
MDðtÞ ð30Þ
dimensional, and the stress normal to the deformed
membrane surface is negligible in comparison with the where M is a positive definite mass matrix, which
stresses in the tangent plane [71]. With this considera- remains constant, DðtÞ represents the nodal displace-
tion, the deformation and stress matrices, CðtÞ and SðtÞ; .
ment vector, DðtÞ is the nodal acceleration vector, Fint is
ARTICLE IN PRESS
436 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

the internal force due to the membrane deformation, but their CFD solver and CSD solver are solved on
and Fext is the external load. different grid systems. The fluid and the structural
Either an implicit or an explicit scheme can be governing equations are regarded as one single system of
adopted to follow the time history of the system of time-dependent equations in a pseudo-time.
equations. The explicit scheme is computationally To better handle different characteristics of the flow
economic for each time step and requires less storage. and structural domains, most of the researchers solve
In some cases, its efficiency can be further improved by the fluid and structural equations separately and then
special techniques, such as lumping technique of the couple them through a synchronization process. Smith
mass matrix [3,115]. However, its time step needs to be and Shyy [94] have adopted a combined simulations
small in order to satisfy the stability requirement. As far treating the membrane using the finite element ap-
as the implicit scheme is concerned, a matrix system is proach, and the potential fluid flow using the moving
updated more than once per time step to account for the grid approach. Smith and Shyy [94], Gordnier and
nonlinear effect, which requires more computations per Fithen [24], Kalro and Tezduyar [46], Zwaan and
time step and more storage. Generally speaking, implicit Prananta [116], and Lian et al. [52,54] have coupled
algorithms are effective for many structural dynamics the fluid and structural solvers with a subiteration
problems in which the response is controlled by a small approach. The fluid solver and the structural solver
number of low frequency modes, which explicit algo- function independently on their own computational grid
rithms are efficient for wave propagation problems in with their own time step, subiteration between these two
which the contribution of intermediate and high solvers is employed at each time step to avoid phase lag.
frequency structural modes to the response is important Alternatively, Farhat et al. [30,28] formulate the fluid
[95]. Among the available techniques, the implicit, and structure interaction problem as a three-field
unconditionally stable Wilson-y method [112] is used problem: the fluid, the structure, and the dynamic mesh
by Lian et al. [54,52]. The Wilson-y method has that is represented by a pseudo-structural system. In
controllable algorithmic numerical dissipation; it is a their work, the arbitrary Lagrangian Eulerian (ALE)
one-step method, and it yields second-order accuracy, finite volume schemes are employed for the fluid
which makes it easier to code and behaves well in fluid/ dynamics equations, finite element method is applied
structure interaction system. Since the Wilson-y method for the structure dynamics equations. Partitioned
is dissipative, the used time step is smaller than that procedures and staggered algorithms are adopted for
required by the overall accuracy criterion to avoid the solution of coupled fluid and structure interaction
excessive dissipation. For y > 1:37; the method is problems in the time domain.
unconditionally stable, usually y ¼ 1:4 is employed. In fluid-structure interaction computations, the mov-
For details and illustration of this model and its ing grid technique can be fruitfully employed to
computations, we refer to Lian et al. [51,54]. automatically regenerate the new CFD grid after the
shape change. In addition, since the grid distributions
2.4. Coupling between fluid solver and structural solver between the fluid and structure domains are often
different, interpolations are needed to allow the in-
When a fluid flow passes a flexible structure, the formation to be exchanged. Smith et al. [92] have
structure changes its shape due to the external forces evaluated suitable methods to transfer information
from the fluid flow; meanwhile, the shape change affects between CFD and CSD grids. For example, a thin plate
the fluid flow around the structure. This leads to a fluid interpolation method is adopted by Lian et al. [54,52]
and structure interaction problem. Usually, the fluid as the interface to exchange information between the
problem and the structural problem have different fluid solver and structure solver. The thin plate spline
mathematical and numerical properties. Depending on interpolation is a global interpolation method which is
the problem complexity (linear problem or nonlinear invariant with respect to rotation and translation, and
problem, simple or complex geometry, small-scale or hence is suitable for the interpolation on moving
large scale problem), available software, and computing surfaces. A one-dimensional version of the function is
platform, numerous techniques have been developed to
provide coupled CFD and CSD methods. One approach X
N
HðxÞ ¼ ai jx  xi j2 log jx  xi j; ð31Þ
is to treat the flow and structural equations as one i¼1
monolithic system of equations [5] and solve the system
of equations simultaneously on a ‘‘monolithic’’ grid. where HðxÞ is the displacement, ai is the undetermined
However, there are multiple time scales present in the coefficient, N is the number of structural nodes on the
fluid and structure systems, and a unified treatment may interface being monitored, and xi ’s are their locations. It
not be efficient to handle the computational stiffness. can be seen from Eq. (31) that the interpolated function
Liu et al. [59] have solved the flow filed and the HðxÞ has continuous first-order derivative. Once these
structural field simultaneously by a fully implicit method coefficients ai are determined based on the structural
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 437

grid locations, the displacement vector defined on the


computational structure dynamics (CSD) grid D is
mapped to the CFD grid via a matrix G
dX ¼ GD; ð32Þ
where dX is the corresponding displacement on the CFD
grid. Similarly the surface forces can also be mapped
from the CFD grid to the CSD grid. z
Root
Upon updating the aerodynamic forces in the
structural equations and providing the new boundary
conditions to the fluid solver, the temporal development
between the aerodynamic and structural equations can
be coordinated. In order to conduct time-accurate x Batten 3
solutions for the fluid-membrane dynamics, iteration
should be done at each time instant. A step-by-step Batten 2
coupled algorithm is described by Lian [51].
Batten 1

3. MAV wing: aerodynamics and structural response Fig. 5. Schematic wing with six battens.

As shown in Fig. 1 the membrane wing consists of a


leading edge spar and six chordwise battens which are
made of unidirectional carbon fiber prepreg laminate.
A membrane material is bonded to the spar and battens.
While the membrane is flexible which does not sustain
bending moment, the carbon fiber is rigid which
provides support for the membrane and can sustain
bending moment. To fully model the membrane wing,
ideally, one needs to model both the rigid battens and
the flexible membrane. Two principal difficulties arise by
doing that. First, since the membrane does not bear
bending moment, it has three degree of freedoms at each
node, while the batten, if modeled as a beam, has five
degree of freedoms at each node. Special treatment is
required to treat the interface point which belongs to
different regimes and has different degree of freedoms.
Second, the membrane is much thinner than the batten, Fig. 6. Computational set-up and boundary conditions.
the mass matrix associated with these two different
materials makes the assembled mass matrix ill-condi-
tioned. Given these factors, Lian et al. [54,52] treat the Shyy [52] to assess the sensitivity of the computed results
batten as a special membrane material with larger to the wing location and the type of boundary
density; the density ratio between the batten and the conditions imposed to the outer flow boundary. The
membrane is three. basic requirement for the wing location is to make the
outer boundaries sufficiently far away from the mem-
3.1. Computational background brane so that they do not have much impact on the
computed results. The relative position of the wing is
In the study of Lian and coauthors [54,52,51] and shown in Fig. 6, in which c represents the chord length at
Viieru et al. [107], only the membrane wing is considered the root. At the surface ABCD a zero gradient boundary
while ignoring the fuselage and propellers. A schematic condition is imposed for the velocity components; all the
geometry of the wing is shown in Fig. 5; the shaded area other boundaries are assigned a Dirichlet type boundary
shown is according to the fuselage area which does not condition. Since the freestream velocity is parallel to the
change the shape during the flight. The physical problem chord direction and no propeller is modeled, only half of
under consideration is the flow over a wing in an the domain is computed, and a symmetric mapping can
unbounded domain. Based on the free stream velocity of be applied to the other half domain.
10 m=s; the root chord Reynolds number is 9  104 : A grid sensitivity test is done by Lian and Shyy [52]
Numerous computations are performed by Lian and where three grid systems have been systematically
ARTICLE IN PRESS
438 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

chosen, ranging from 1:8  105 nodes at the coarse level illustrate the key issues in the fluid and flexible structure
and 2:3  106 nodes at the fine level. The coarse grid has interaction problem.
41 points in the chord direction and 31 points in the
spanwise direction. The grid distribution at the coarse 3.2. Rigid wing aerodynamics
level around the wing is shown in Fig. 7. Following Lian
and Shyy [52], the results from the fine grid are used as 3.2.1. Vortical flow structure of the rigid wing
the reference. The comparisons among these three grid Tip vortices exist on a finite wing due to the pressure
systems are summarized in Table 1. The coarse grid difference between the upper and lower wing surface.
solution under-predicts the lift-to-drag ratio ðCL =CD Þ by The tip vortex establishes a circulatory motion which
0.5%, and the difference in the form drag force is less presents over the wing surface and exerts great influence
than 0.7%. A similar conclusion can be drawn for the on the wing aerodynamics. Specifically, tip vortex
lift force, L: As expected, the skin friction drag, DF ; is increases the drag force. The total drag coefficients for
sensitive to the grid density; the coarse mesh over- a finite wing at subsonic speeds can be written as [2]
estimates DF by almost 10%. An intermediate grid,
which has 6:7  105 nodes and more than doubles the CL2
CD ¼ CD;P þ CD;F þ ; ð33Þ
surface points on the wing comparing to the coarse grid, peAR
gives an improved prediction on the skin friction drag. where CD;P is the drag coefficient due to pressure, CD;F is
From Table 1 it can be seen that the contribution of the the drag coefficient due to skin friction, e is the span
friction drag to the total drag is much smaller than the efficiency factor which is less than one, AR is the aspect
form drag. This is also valid at higher angle of attack ratio, and CD;i ¼ CL2 =peAR is the induced drag coeffi-
where the form drag is the dominant factor in the total cient due to the existence of tip vortices. Eq. (33)
drag. Since truly grid independent results are difficult to demonstrates that the induced drag varies as the square
attain at such low Reynolds number condition, the of the lift coefficient; at high angles of attack, the
intermediate grid is utilized by Lian et al. [51,52] for the induced drag can be a substantial portion of the total
rigid wing simulation while use the coarse grid to drag. Furthermore, Eq. (33) illustrates that as AR is
decreased, the induced drag increases. The MAV wing
presented by Ifju et al. [36,35] has a low aspect ratio of
1.4; therefore, it is important to investigate the tip vortex
effects on the wing aerodynamics. In general, the tip
vortex effects are two-folded:
* Tip vortex causes downwash which decreases
the effective angle of attack and increases the drag
force [2].
* Tip vortex forms low-pressure region on the top
surface of the wing, which provides additional lift
force [68].

Fig. 8 shows tip vortices around the wing surface


together with the streamlines at an angle of attack of 39
[52]. The vortical flow is usually associated with a low-
pressure zone as shown in Fig. 9. The pressure drop
further strengthens the swirl by attracting more fluids
toward the vortex core; meanwhile, the pressure
Fig. 7. Computational grid distribution near the wing. Coarse decreases correspondingly in the vortex core. The
grid with 1:8  105 points. low-pressure region created by the vortex generates

Table 1
Grid refinement on computed aerodynamic performance

GRID DP ðNÞ LðNÞ DF ðNÞ CL =CD


5
Coarse 1:8  10 6:24E  2 5:10E  1 9:70E  3 7.06
Intermediate 6:7  105 6:16E  2 5:08E  1 9:22E  3 7.16
Fine 2:3  106 6:20E  2 5:04E  1 8:80E  3 7.10
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 439

even though they cover a small area and are of modest


strength. The flow is attached to the upper surface and
follows the chord direction. A low-pressure region near
the tip, caused by the vortical structure there, is
observed. This pressure drop can be seen from
Fig. 11(a) where the spanwise pressure coefficient on
the upper wing surface at x=c ¼ 0:4 is plotted. At a ¼ 6
the spanwise pressure is almost uniform on the upper
wing surface, and the tip vortex causes the pressure drop
to occur at approximate 90% of the half span from the
root. Fig. 11 is illustrative in regard to pressure
distributions versus the vortical structures. They are
not indicative of the total level of the pressure force.
Vortices strengthen with the increase of angle of
attack. At a ¼ 27 ; as shown in Fig. 10, tip vortices
develop a strong swirl motion while entraining the
surrounding flow. The low-pressure area increases as
the angle of attack becomes higher. In Fig. 11(a), the
pressure drop moves along the spanwise direction
towards the root, and now occurs at 75% from the
root. At lower angles of attack, the vortex core position
Fig. 8. Streamlines and vortices for rigid wing at a ¼ 39 : The shows a linear relation with the incidence. This relation
vortical structures are shown on selected planes.
disappears at higher angles of attack when the flow is
separated on the upper surface. For example, at a ¼ 45 ;
the flow is separated at the leading edge, and the low-
pressure zone covers more than 40% of the wing surface,
which helps to maintain the increase in lift force. At
a ¼ 51 ; considerable spanwise velocity component is
seen and the flow is separated from most of the upper
surface (Fig. 10). The lift force by the vortices is not
enough to maintain the increase of the lift coefficient
with the angle of attack, and stall occurs. Tip vortices
also affect the pressure distribution at the lower surface;
however, the effected region is mainly located at the tip.
Most regions on the lower surface are not affected by the
tip vortices. From Fig. 11(b) it can be seen that the
pressure is almost uniform in the spanwise direction
even at high angles of attack. One should also note that
Fig. 11 indicates that the moment experiences substan-
tial variations as the angle of attack changes.

3.2.2. Laminar boundary layer separation on rigid wing


The laminar boundary layer, especially under low
Reynolds number conditions, is prone to separate under
Fig. 9. Pressure distribution around the rigid wing in the cross an adverse pressure gradient. This separation is usually
sections with streamlines at angle of attack of 39 : followed by a quick transition from laminar to turbulent
flow. If the adverse pressure gradient is not strong, the
resulting turbulent flow, by obtaining energy through
additional lift. Toward downstream, the pressure re- entrainment, may reattach to the surface and form an
covers to its ambient value, the swirling weakens, the attached turbulent boundary layer. Such a phenomenon
diameter of the vortex core increases, and the vortex is known as a laminar separation bubble.
core losses its coherent structure. Ever since its first observation by Jones [42], many
In Fig. 10 evolution of the vortical structure with experimental investigations on laminar separation bub-
increasing angle of attack is visualized. The pressure ble have been conducted, as reviewed by Young and
distribution on the upper surface is also presented in the Horton [113]. Typically, a separation bubble has
same figure. At a ¼ 6 ; the tip vortices are clearly visible the following characteristics. Just downstream of the
ARTICLE IN PRESS
440 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Fig. 10. Evolution of flow pattern for rigid wing versus angles of attack (from left to right, top to bottom, 6 ; 15 ; 27 ; and 51 Þ:

separation point, there is a ‘‘dead-air’’ region, where the number is slightly higher than 5  104 : In the following,
recalculating velocity is significantly less than the free- detailed flow structures around a rigid wing reported by
stream velocity and the flow is virtually stationary. The Lian and Shyy [52] are presented. This information is
separated flow forms a free-shear layer, which is highly helpful to gain more insight into the MAV aerody-
unstable. As suggested by Young and Horton [113], the namics. By contrasting the detailed fluid flow around the
separated shear layer quickly undergoes transition from rigid and membrane wings, one can better evaluate the
laminar to turbulent flow; the turbulent flow may MAV technologies. The bubble structures for different
reattach to the surface behind a vortical structure and angles of attack are demonstrated in Fig. 12. Repre-
form a turbulent boundary layer. Hence, a separation sentative instantaneous streamlines at the root section
bubble forms. The dynamics of a laminar separation on the rigid wing are plotted. At this moment, the
bubble depends on the Reynolds number, the pressure spanwise velocity component is ignored to make a clear
distribution, the geometry, the surface roughness, and representation of these separated structures. At a low
the freestream turbulence. A rough rule given by angle of attack of 6 ; the flow primarily attaches to the
Carmichael [8] is that the Reynolds number based on surface and a tiny separation bubble is observed at the
the freestream velocity and the distance from the lower surface near the leading edge. Beginning at 45%
separation point to the reattachment point is approxi- chord from the leading edge, a weak recirculation zone
mately 5  104 : On the one hand, if the Reynolds is seen on the upper surface, which produces a
number is less than 5  104 ; an airfoil will experience reattachment length of x=c ¼ 0:5: The maximum reverse
separation without reattachment; on the other hand, a velocity is less than 0.5% of the freestream velocity. The
long separation bubble will occur if the Reynolds normalized chordwise velocity contours at different
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 441

-1

-2
cp

-3

α=51
-4 α=45
o
o
α=27
α=15o
-5 α=6o

-6
0 0.2 0.4 0.6 0.8 1
(a) z/Z

0.5

-0.5

-1
cp

-1.5 α=51o Fig. 12. Streamlines at different angles of attack for rigid wing.
α=45
α=27o
From top to bottom 6 ; 15 ; 27 ; and 51 : Shown are single
-2
α=15o span shorts of the individual time dependent flows.
o
-2.5 α=6

-3
Carmichael. The reattachment point corresponds to a
-3.5
0 0.2 0.4 0.6 0.8 1 rapid increase in the surface pressure. The reattachment
(b) z/Z point is highly unstable, which moves back and forth.
Fig. 11. Spanwise pressure coefficient distributions at x=c ¼ 0:4 At a ¼ 27 ; the reverse flow component of the
for rigid wing at different angles of attack: (a) cp at upper separation bubble reaches a maximum mean velocity
surface; (b) cp at lower surface. of 0:49U: Under such a situation, Crompton and Barrett
[11] have shown that the shear layer is particularly
energetic since most of the shear layer is turbulent. Even
though the flow on the upper surface near the root has
angles of attack are demonstrated in Fig. 13. Under the separated from a large portion of the surface; however,
present circumstance, a considerable portion of the the lift force still increases with the angle of attack. Two
shear layer remains laminar [23]. reasons might lie behind this phenomenon. First, tip
With the increase of incidence, the separation point vortices generate suction near the tip area (Fig. 11)
moves forward toward the leading edge; by gaining which provides additional lift. Second, even though flow
energy from the shear layer, the maximal reverse velocity separates near the root, it still attaches to the surface in
becomes larger. At the angle of attack of 15 ; as shown the other region. When the angle of attack is increased
in Fig. 12, the separation occurs at 39% of the chord further, at a ¼ 51 ; massive separation occurs on most
from the leading edge, which is followed by a long ‘dead- of the upper surface; vortex shedding is observed near
air’ zone before it reaches a maximal reverse velocity of the root (Fig. 14).
0:26U: In the ‘dead-air’ zone, the stationary reverse flow For low aspect ratio wings, tip vortices have
does not have much effect on the pressure distribution, considerable contributions to the lift force. This scenario
which is primary determined by the wing curvature. The is quite similar to that for the delta wings [25]. In his
shear layer reattachment happens at 89% of the chord. numerical study, Lian [51] has observed that the low
Based on the freestream velocity and the distance aspect ratio wing suffered less from separation. As seen
between the separation and reattachment points, the from Fig. 15, the lift coefficient keeps a linear relation-
Reynolds number is approximately 4:5  104 which is ship with the angle of attack even at very high angles of
slightly smaller than the Reynolds number suggested by attack. Experiments by Torres and Mueller [102] on low
ARTICLE IN PRESS
442 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Fig. 13. Normalized chordwise velocity u=U contours at root


for rigid wing. From top to bottom the angles of attack are 6 ;
15 ; 27 ; and 51 Þ:

Fig. 14. Vortex shedding at a ¼ 51 :

aspect ratio wing have similar findings. However, neither


fuselage nor propeller is included in the analysis by
Lian [51].
2.2
3.2.3. Unsteady phenomena at high angle of attack
2
Vortex shedding and boundary layer separation occur
at large angle of attack, which introduce unsteadiness to 1.8
the aerodynamic performance. Cummings et al. [12] 1.6
have reported that the unsteady computations predict
1.4
much lower lift coefficients than the steady computa-
L

tions at large angle of attack. Lian and Shyy [52] 1.2


C

perform steady computations when the incidence is less 1


than 15 while conducting both steady and unsteady Coarse Grid
0.8 Intermediate Grid
computations when the angle of attack is larger than
that. Interestingly, the difference between the steady 0.6
state computations and the time-averaged unsteady 0.4
computations are small even at large angle of attack in
0.2
which unsteady phenomenon such as vortex shedding 0 10 20 30 40 50 60
Angle of Attack
are prominent. The differences in lift coefficient and lift-
to-drag ratio are found to be less than 1%. Fig. 16 shows Fig. 15. Lift coefficient versus angle of attack for rigid wing.
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 443

2.5 The pressure distributions at the root section are also


Time Averaged
Steady compared, where separation usually appears first due to
the large camber in the present MAV wing design [36].
2 Fig. 17 shows that at a=6 the time averaged pressure
coefficient matches closely with the steady state result.
This finding is consistent with Figs. 12 and 13 where a
very weak recirculation is seen at this angle of attack.
CL

1.5
However, at a ¼ 15 ; clear difference is shown after
x=c ¼ 0:6 which approximately corresponds to the
location of the maximal reverse velocity (Fig. 13). The
1 time averaged value yields a smooth pressure distribu-
tion; the variation in the steady result is apparent from
the recirculation zone. The difference between steady
0.5 and unsteady computations is also shown from velocity
0 10 20 30 40 50 60
Angle of Attack (degree) distribution. At a ¼ 6 ; as seen from Fig. 18, there is
almost no difference in the chordwise velocity distribu-
Fig. 16. Comparison of rigid wing CL for steady and unsteady
tion. While at a ¼ 15 ; there is little difference in the
computations.
leading edge region, clear difference appears after the
0.8
separation bubble (Fig. 19).
Time Averaged
Steady
0.6
Boundary layer profile at x/c=0.14 for α=6o
0.4 0.022
Time Averaged
Steady
0.2 0.0215

0.021
cp

0.0205
-0.2

0.02
y

-0.4

0.0195
-0.6

0.019
-0.8
0 0.2 0.4 0.6 0.8 1
(a) x/c 0.0185

0 0.018
Time Averaged 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Steady u/U
-0.2 Boundary layer profile at x/c=0.80 for α=6
o

0.022
Time Averaged
-0.4 Steady
0.021

-0.6 0.02
cp

-0.8 0.019

0.018
y

-1

0.017
-1.2
0.016
-1.4
0 0.2 0.4 0.6 0.8 1
(b) 0.015
x/c

Fig. 17. Comparisons of cp on rigid wing at the root for steady 0.014
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
and unsteady computations: (a) a ¼ 6 ; (b) a ¼ 15 : u/U

Fig. 18. Comparisons of chordwise velocity profiles on rigid


the results with steady and unsteady computations. wing for steady and time-averaged unsteady computations,
Noticeable difference appears only when the angle of at angle of attack of 6 ; at two chordwise locations, both at
attack becomes sufficiently high. the root.
ARTICLE IN PRESS
444 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Boundary layer profile at x/c=0.19 for α=15o 0.14


0.022
Time Averaged
Steady 0.12
0.0215
0.1
0.021

Displacement (cm)
0.08
0.0205
0.06
0.02
y

0.04
With Synchronization
0.0195
Without Synchronization
0.02
0.019
0
0.0185
-0.02
0 0.05 0.1 0.15 0.2
0.018
0 0.2 0.4 0.6 0.8 1 1.2 1.4 Time (Second)
u/U
Boundary layer profile at x/c=0.80 for α=15o Fig. 20. Effects of time synchronization on the displacement of
0.022
Time Averaged
a trailing edge point for membrane wing at a ¼ 6 :
Steady
0.021

0.02 40

0.019 30
With Synchronization
Without Synchronization
0.018
y

20
Vertical velocity (cm/s)

0.017
10
0.016
0
0.015
-10
0.014
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
u/U -20
Fig. 19. Comparisons of chordwise velocity profile on rigid
wing for steady and time-averaged unsteady computations, -30
0 0.05 0.1 0.15 0.2
at angle of attack of 15 ; at two chordwise locations, both at Time (Second)
the root.
Fig. 21. Effects of time synchronization on vertical velocity of
a trailing edge point for membrane wing at a ¼ 6 :

3.3. Membrane wing dynamics

3.3.1. Time synchronization matches well with that with synchronization at the very
Before presenting the membrane wing results, some early stage. However, the lagging errors accumulate with
key issues in coupled membrane–fluid interactions are time which eventually result in much larger displacement
first addressed. Specifically, time synchronization be- than that with synchronization. Gordnier and Visbal
tween fluid and structure computations, and the [25] also report the importance of subiteration. The
geometric conservation law in regard to the moving velocity and pressure histories of the same node are
grid method are stressed. shown in Figs. 21 and 22, respectively. Before the
According to the design of Ifju and coworkers [35,36], computation diverges, the velocity without synchroniza-
the membrane has a uniform thickness of 0:2 mm; and tion is more than two times higher than that with
with a density of 1200 kg=m3 : Lian et al. [54,52] have synchronization. The high velocity is believed to be
simulated the fluid and structure interaction by integrat- associated with the sudden pressure drop shown in
ing the fluid solver and the structure solver. To Fig. 22, which causes the wrinkle of the membrane and
demonstrate the importance of the synchronization the failure of the structural solver.
between fluid and structural solvers, Fig. 20 depicts the
displacement history of a trailing edge node. The 3.3.2. The geometric conservation law
computation without synchronization fails to continue. The geometric conservation law [100] is critical when
The displacement history without synchronization the moving grid technique is employed. Fig. 23(a) shows
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 445

1 the time history of L=D for the membrane wing at


0
a ¼ 6 : Without satisfying the geometric conservation
law, spikes are observed in the course of computation.
-1
With Synchronization
However, the computations are better regulated once the
-2
Without Synchronization geometric conservation law is enforced as in Fig. 23(b).
Farhat et al. [29] argue that the geometric conservation
-3 law is a necessary condition to maintain the stability of a
cp

-4
scheme utilized in moving boundary problems. Lian and
Shyy [52] do not come across instability issues in their
-5 study. Nevertheless, the computations behave erratically
-6
when the geometric conservation law is not enforced.

-7
3.3.3. Self-initiated membrane vibration
-8
0 0.05 0.1 0.15 0.2 Earlier works in membrane wing [86,93] are based on
Time (Second) the consideration that the membrane is assumed
Fig. 22. Effects of time synchronization on pressure distribu-
massless and there is no time dependent movement in
tion on a trailing edge point for membrane wing at a ¼ 6 : the steady freestream. Experiments [110] have shown
that the membrane experiences high frequency vibra-
tion, on the order of 100 Hz: By adopting a dynamic
membrane model, computations by Lian et al. [54,52]
show that the membrane wing also exerts high frequency
vibration in the steady freestream. Fig. 24 demonstrates
the displacement of the trailing edge with time. The
maximal displacement during that period occurs near
the tip. The displacement is normalized by the maximal
camber which is about 0:9 cm: To investigate the
vibration frequency, Lian and Shyy [52] choose a point
between batten 1 and batten 2 on the trailing edge as an
example. Its deflection history is shown in Fig. 20, and
the estimated dominated frequency is around 120 Hz
(Fig. 25). Analyses at other points also show the
frequencies around 120 Hz which is comparable to the
experimental frequency of 140 Hz: [110] The experi-
mental conditions and the detailed wing configuration
(a) between the study by Lian and Shyy [52] and those
reported by Waszak et al. [110] are not identical. Hence,
9
Membrane Wing
8.5 Time Averaged
0.14
8 t=0
-3
0.12 t=3 × 10
-3
7.5 t=7 × 10
0.1 t=10 × 10-3
7
Normalized displacement

t=13 × 10-3
D

0.08 -3
C /C

6.5 t=18 × 10
L

6 0.06

5.5 0.04

5 0.02
4.5 0
4
0 0.05 0.1 0.15 0.2 -0.02
(b) Time (sec)
-0.04
Fig. 23. Effect of the geometric conservation law on CL =CD for 0 0.2 0.4 0.6 0.8 1
z/Z
membrane wing computation at a ¼ 6 : (a) not satisfying the
geometric conservation law; (b) satisfying the geometric Fig. 24. Time history of trailing edge displacement for
conservation law. membrane wing at a ¼ 6 : The camber at the root is 0:90 cm:
ARTICLE IN PRESS
446 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

-3
x 10 v/U contour at t=3.0 × 10 -3
8
0

7 0.000746

2 -0.00284
0.000746
6
0.0115
4
Power Spectum

Chordwise (cm)
0.0115

4 6
0.0366

3 8

2 0.000746
10
0.0223
1
12 0.0223

0
0 50 100 150 200 250 300 350 400
Frequency (Hz) -6 -4 -2 0 2 4 6
Spanwise (cm)
Fig. 25. Spectrum analysis of the trailing edge point vibration -3
v/U contour at t=7.0 × 10
for membrane wing at a ¼ 6 : 0
-0.000772
-0.000772
2
the qualitative agreement between computation and 0.00438
-0.000772
experiment in this regard is deemed satisfactory. Liu [61] 4
has reported that, under a typical wing gust situation,
Chordwise (cm)

the energy is mainly located in the low frequency range 6 0.00696 -0.000772

of several Hz or lower. The membrane fluctuates in a 0.00696

time scale much faster than either the vehicle control 8 -0.0137
scale or the expected wing gust time scale; hence, the 0.0121 -0.000772
membrane fluctuation is not expected to cause sensitive
10
response to the vehicle. -0.0111

When the membrane surface vibrates, the velocity on -0.0137 -0.0085


12 -0.00592
the surface is no longer zero. A considerable velocity
component is observed at the surface. The vertical -6 -4 -2 0 2 4 6
velocity contour at two different time instants are Spanwise (cm)
plotted in Fig. 26. Fig. 26(a) corresponds to the early Fig. 26. Normalized vertical velocity contour at two different
stage when the membrane wing moves up under the time instants for membrane wing at a ¼ 6 on the upper
aerodynamic forces. At the trailing edge, the vertical surface.
velocity is about 2% of the freestream velocity. A
negative velocity component is also observed near the
leading edge, because at the leading edge the pressure at Table 2
the upper wing surface is higher than that at the lower Aerodynamic coefficients comparison between membrane wing
wing surface of the same point. Fig. 26(b) corresponds and rigid wing
to the time of maximal displacement. A negative velocity Membrane wing Rigid wing
near the root indicates that the membrane goes down.
6 CL =CD 7.05 7.06
CL 0.530 0.532
3.3.4. Comparison between membrane and rigid wings
With the same freestream condition, as shown in
15 CL =CD 3.94 3.88
Table 2, the flexible wing exhibits slightly less lift CL 0.920 0.954
coefficient than the rigid one at a ¼ 6 : The difference
in CL =CD is also small. At higher angle of attack of 15 ;
the membrane wing generates a lift coefficient about 2%
less than the rigid wing; however, its CL =CD is 1.5% decreases the lift force by reducing the effective angle of
larger than the rigid one. These differences lie behind the attack of the membrane wing; on the other hand,
high frequency of the membrane vibration. Under the it increases the lift force by increasing the camber.
external force, the membrane wing changes its shape. The numerical finding of Lian and Shyy [52] shows
This shape change has two effects. On the one hand, it that membrane and rigid wings exhibit comparable
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 447

aerodynamic performance before stall limit, which is 10


Rigid Wing
also experimentally observed by Waszak et al. [110]. Membrane Wing
9.5
Fig. 27 shows the time averaged vertical displacement
of the trailing edge. The displacement is normalized by 9
the maximal camber of the wing. Due to the membrane

Angle of Attack
deformation, the effective angle of attack of the 8.5
membrane wing is less than the rigid wing. The spanwise
8
angles of attack between rigid and membrane wings
under the same flow condition and with identical 7.5
geometric configurations are shown in Fig. 28. In
Fig. 28(a), the rigid wing has an incidence of 6 at the 7

root and monotonically increases to 9:5 at the tip; the 6.5


membrane wing shares the same angles of attack with
the rigid wing in 36% of the inner wing; however, the 6
0 0.2 0.4 0.6 0.8 1
effective angle of attack toward the tip is less than that (a) Spanwise z/Z
of the rigid wing. At the tip, the angle of attack of the
membrane wing lowers by about 0:8 : Fig. 28(b) 19
Rigid Wing
compares the angle of attack at a ¼ 15 ; which shows Membrane Wing
18.5
that the effective angle of attack of the membrane wing
is about 1 less than that of the rigid wing at the tip. The 18
Angle of attack (degree)
17.5

17
0.15
Normalized displacement by maximal camber

16.5

16

0.1 15.5

15
0 0.2 0.4 0.6 0.8 1
(b) Spanwise z/Z

0.05 Fig. 28. Time averaged spanwise angle of attack for membrane
wing: (a) a ¼ 6 ; (b) a ¼ 15 :

Rigid part
reduced effective angle of attack causes the decrease in
0
0 0.2 0.4 0.6 0.8 1 the lift force.
(a) z/Z The shape change of the membrane wing has another
effect; the pressure difference between upper and lower
0.25
wing surfaces inflates the membrane wing’s camber,
which is shown in Fig. 29. Two airfoil shapes at different
Normalized time averaged displacement

0.2 spanwise positions are plotted together with the


corresponding rigid wing shape. The camber increase
is more visible in the outer wing than the inner wing.
0.15 This is consistent with Fig. 27 where the maximal
trailing edge displacement occurs near the tip. As
expected, the increase is larger at a ¼ 15 than that at
0.1
a ¼ 6 :
The movement of the wing surface affects the overall
0.05 pressure distribution. A comparison of time averaged
Rigid part pressure contours for rigid and membrane wings is
shown in Fig. 30. As expected, the differences are mostly
0
0 0.2 0.4 0.6 0.8 1 located at the trailing edge where large movement is
(b) z/Z
observed. In the spanwise direction, as shown in Fig. 31,
Fig. 27. Averaged displacement of the membrane wing trailing the time averaged membrane wing has little difference
edge: (a) a ¼ 6 ; (b) a ¼ 15 : between two wings.
ARTICLE IN PRESS
448 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Spanwise z/Z=0.40 z/Z=0


2 0.8
o
α=6 Time Averaged Membrane
Rigid Wing
1.8 α=15o 0.6
Rigid wing
1.6
0.4
1.4
Airfoil shape (cm)

0.2
1.2

cp
0
1

0.8 -0.2

0.6 -0.4

0.4 -0.6

0.2
-0.8
0 0.2 0.4 0.6 0.8 1
0 x/c
0 0.2 0.4 0.6 0.8 1
x/c z/Z=0.37
0.6
Spanwise z/Z=0.60 Time Averaged Membrane
2.4 Rigid Wing
α=6o 0.4
2.2 α=15o
Rigid wing 0.2
2

1.8 0
Airfoil shape (cm)

cp

1.6
-0.2
1.4

1.2 -0.4

1 -0.6
0.8
-0.8
0.6 0 0.2 0.4 0.6 0.8 1
x/c
0.4 z/Z=0.69
0 0.2 0.4 0.6 0.8 1
x/c 0.1
Time Averaged Membrane
Rigid Wing
Fig. 29. Time averaged airfoil shapes at different spanwise 0
positions for membrane wing at a ¼ 6 :
-0.1

-0.2
cp

Pressure Coefficient Contour


0 -0.3
0.0363
-0.5 -0.4
2 -0.299 -0.5
-0.567 -0.567 -0.5

-1.24
-0.5
-0.634
4 -0.567
-1.1
Chordwise (cm)

-1.17
-0.433 -0.433 -0.6
0 0.2 0.4 0.6 0.8 1
6 -0.0308 x/c
-0.299
-0.299
Fig. 31. Chordwise cp comparison at different spanwise
8 locations between time averaged membrane wing and steady
-0.232 -0.232
rigid wing at a ¼ 6 :
10

-0.165 -0.165
12 -0.165 4. Wing shape optimization
-0.165
-0.165

-6 -4 -2 0 2 4 6 It is desirable not only to understand the pertinent


Spanwise (cm)
physical phenomena of the membrane wing dynamics,
Fig. 30. Time averaged cp contours for membrane wing (left) but to improve its performance based on improved
and cp contours for steady rigid wing (right) at a ¼ 6 : understanding. The optimization of a membrane wing
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 449

introduces multiple challenges. First, coupled, time 2 o


(x1,y1)
dependent simulations of viscous fluid flow and flexible o (x2,y2)
structure are very expensive. Second, an efficient and
1.5
automatic grid regeneration based on the perturbed o
shape is essential. Adopting the rigid wing as a surrogate (x0,y0)

model for membrane wing optimization, Lian et al. [53] 1


optimize the rigid wing under a given flow condition, (x3,y3)

y(cm)
and evaluate the performance of a flexible wing with the o

optimized shape. The surrogate approach has also been 0.5


employed in [7,34,49,64].

0
4.1. Optimization and computational framework

The objective in Lian et al. [53] is to improve the lift- -0.5


0 0.2 0.4 0.6 0.8 1
to-drag ratio of a membrane wing. The baseline (a) x/C
membrane wing shape selected is an existing design
[36] devised from solutions provided by XFOIL [18] and 2 O
O
empirical modification based on flight tests [36]. XFOIL O 2
provides two-dimensional fluid flow solutions based on 1
1.5
coupled thin layer and inviscid flow models. A schematic
baseline wing shape is shown in Fig. 5. The three Design point 3
carbon-fiber-made battens are labeled. The shading area 1
O

corresponds to the fuselage. The fuselage is made of


y(cm)

rigid carbon fiber prepreg cloth which does not deform


during the flight. A flexible latex membrane covers 0.5
Baseline Shape
the top surface which changes its shape under aero- Perturbd Shape
dynamic forces. Only the wing shape is considered by
0
Lian et al. [53].
The baseline wing has a variable spanwise camber.
The nominal camber of the baseline design is 7.5% at -0.5
0 0.2 0.4 0.6 0.8 1
the root, and 2% at the tip. The wing has a maximal (b) x/C
chord length of 13:7 cm; a mean chord of 9:4 cm; and a
span of 15:2 cm: C is used to indicate the chord length at Fig. 32. A cross section of membrane wing along a batten:
(a) straight lines indicating how convexity constraint is applied;
different span stations, and B for the half span. The
(b) effect of y coordinate perturbation at a point. The design
angle of attack is defined in reference to the root points are located at 10%, 27%, and 77% of the chord.
geometry. At the design point, the MAV flies with a
speed of 10 m=s and an angle of attack of 6 : The
objective is to maximize the lift-to-drag ratio at such a
flight condition. At such a condition the root chord Constraint 1 requires that the lift coefficient be no less
Reynolds number is 9  104 : than that of the baseline wing. Constraint 2 maintains
The six design variables chosen by Lian et al. [53] are the convexity of the airfoil (see Fig. 32(a)) so to
the y-coordinates at six points on the surface. Three are eliminate obviously inappropriate shapes, only one of
located at the root whose positions are approximately at the six constraints is shown for illustration purpose.
10%, 27%, and 77% of the chord, as seen in Fig. 32. Constraint 3 gives the lower and upper bound of the
The other three are located at Batten 2 with the same design variables whose values are listed in Table 3.
relative chord positions. In the work of Lian et al. [53] the optimization
The optimization problem investigated by Lian et al. procedure begins with the baseline design by calling the
[53] can be formulated as follows: Navier–Stokes equations solver. The optimizer then
perturbs one or more design variables for a potential
Maximize CL =CD declining direction. After the perturbation, a new
Subject to computational grid is generated with the moving grid
1 : CL XCL baseline ; technique discussed. Based on the design variables
ð34Þ values, a thin plate spline (TPS) interpolation is used
Y1 þ y1  y0 Y2 þ y2
2 : Convexity constraint : X ; to calculate the y coordinate at other points. With TPS,
x1  x0 x2  x0
the interpolated surface has a second-order continuity
3 : YiL pYi pYiU ; i ¼ 1; 6: of the geometry, and, as shown in Fig. 32(b), the
ARTICLE IN PRESS
450 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Table 3
Design variables bounds and their optimal values at 6

Parameter Initial Lower Upper Optimal values with Optimal values with
design bound bound 600 analysis iteration 1000 analysis iteration

X1 0.00 0.40 0.00 0.258 0.259


X2 0.00 0.40 0.00 0.400 0.400
X3 0.00 0.20 0.20 0.152 0.139
X4 0.00 0.20 0.20 0.100 0.095
X5 0.00 0.10 0.20 0.082 0.068
X6 0.00 0.10 0.20 0.135 0.121
Analysis 97 108
Cycle 6 6
CL 0.530 0.529 0.529 0.529
CL =CD 7.06 7.55 7.55

Number of analyses
with convexity violations 13 22

interpolated surface passes through all the control The optimization is set at the angle of attack of 6 : To
points. probe the performance of the optimized shape at other
angles of attack, the aerodynamic characteristics for
4.2. Rigid wing optimization different angles of attack are shown in Fig. 37. Overall,
the improvement is more substantial at the modest
The baseline shape chosen by Lian et al. [53] has the angles. The spanwise L=D distributions at different
largest camber at the root, about 7.5% of the chord angles are shown in Fig. 38. Consistently, the improve-
length; it decreases to 2% at the tip. The analysis of the ment is largely realized from lowering the form drag.
baseline wing shows that the large camber near the root
can cause flow recirculation near the root. The 4.3. Flexible wing optimization
optimization reduces the camber near the root while
increasing the camber near the tip, as seen in Fig. 33. With the surrogate optimization output, the mem-
The optimized shape has a more uniform camber, brane wing performance is evaluated based on the
gradually decreasing from 4.8% at root to 4% at tip. coupled fluid-structure interaction model [54]. To
This modification results in the elimination of the investigate the spanwise force distribution at different
recirculation in the root region. Fig. 34 shows the time instants, The overall aerodynamics of the mem-
streamwise velocity profiles of both baseline and brane wings, for both baseline and optimized shapes, is
optimized rigid wings, in a location near the root on summarized in Fig. 39. Consistent with the rigid wing
the pressure side. It demonstrates that the reverse flow case, shown in Fig. 37, the optimized shape improves
associated with the baseline wing shape has now been L=D consistently with more substantial improvement at
eliminated in the optimized shape. Fig. 35 illustrates the modest angles of attack. Furthermore, between the rigid
spanwise pressure distributions of both wings. At the and membrane wings, the membrane wing varies less in
root, the baseline wing yields a cross-over pressure L=D versus the angle of attack, indicating that it can
distribution at the leading edge, which decreases the offer a more steady flight behavior than the rigid wing.
total lift force. The optimized shape eliminates the cross- This final finding is consistent with that observed in
over. In addition, the adverse pressure gradient on flight test [36].
the lower surface is smoother for the optimized shape
than for the baseline; this helps explain why the
optimized wing does not experience flow separation in 5. Endplate effects on MAV aerodynamics
that region.
The spanwise L=D distribution is depicted in Fig. 36. Tip vortex induces downwash movement that reduces
The aerodynamic improvement is largely realized in the the effective angle of attack. When a wing has a
70% of the inner wing. With a reduced camber in the relatively low aspect ratio, such as that employed by
root region, the optimized shape has the largest impact the MAV, the induced drag by the tip vortex is relatively
there. With a reduced camber in the root region, the large and therefore the aerodynamic performance of the
drag coefficient decreases noticeably there, which leads vehicle is deteriorated. Viieru et al. [107] have reviewed
to higher overall lift-to-drag ratio. the use of the endplate to help probe the tip vortex
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 451

Span station z/B=0 0.6


2
Baseline Baseline
Optimized 0.8 Optimized
1.5
1

1 1.2

y(cm)
y(cm)

1.4
0.5

1.6
0
1.8

-0.5
0 0.2 0.4 0.6 0.8 1 2
x/C -2 0 2 4 6 8 10
Streamwise velocity (m/s)
Span station z/B=0.40
2 Fig. 34. Comparison of streamwise velocity profile at the root
Baseline
1.8 Optimized between baseline and optimized shape.
1.6

1.4
inner wing by reducing the camber near the root and
1.2 increasing the camber around the tip. There is no
y(cm)

1 significant improvement near the tip. As illustrated in


0.8 Fig. 36, the lift to drag ratio drops substantially in the
tip region, apparently due to the tip vortex. Tip vortex
0.6
causes a circular movement that modifies the pressure
0.4 distribution on the wing surface. If the wing aspect ratio
0.2 is low, it can affect a large portion of the surface area.
0
Based on the lift-to-drag ratio profile, it is estimated that
0 0.2 0.4 0.6 0.8 1 the tip vortex affects 30% of the wing surface. Fig. 8 also
x/C
graphically illustrates tip vortex around the wing tip.
Span station z/B=0.80
2 Methods to reduce the induced drag by reducing the
Baseline tip vortex effects are described in literature and
Optimized
1.8 confirmed by actual applications of them to aircraft
wing design [50]. Viieru et al. [107] investigate the effect
1.6 of the endplate on wing aerodynamic performances as a
method to assess the induced drag. In their work, the
1.4
flow over a rigid wing at angles of attack of 6 ; 15 ; and
y(cm)

27 ; with and without the endplate, is studied. For all


1.2
the cases the endplate geometry is the same. Also for an
1 angle of attack of 6 a higher endplate is attached to the
wing tip in order to study the influence of endplate
0.8 geometry on the overall aerodynamic performance. The
endplate geometry studies by Viieru et al. [107] is shown
0.6 in Fig. 40. Two endplates with different height but
0 0.2 0.4 0.6 0.8 1
x/C otherwise identical shape, are investigated by Viieru et al.
[107]. The length of both endplates is 3:96 cm represent-
Fig. 33. Comparison between optimized shape and baseline
shape at different span positions. The camber decreases from ing 29.8% of the root chord. The height of the shorter
4.8% at the root to 4% at the tip. endplate is 1 cm while the height of the higher one is
2:4 cm:
Wing tip vortices are caused by the pressure difference
effects. The investigation is facilitated by solving the between the high-pressure region below the wing surface
Navier–Stokes equations around a rigid wing with a and the low-pressure region above the wing. This
root chord Reynolds number of 9  104 : induces a circulatory motion over the wing tip, which
As reviewed above, shape optimization shows that the affects the wing aerodynamics. The tip vortices down-
improvement is obtained largely within the 70% of the stream of the wing induce a small secondary downward
ARTICLE IN PRESS
452 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Span station z/B=0 8.5


-0.4 Optimized
Optimized Baseline
Baseline 8
-0.3
7.5
-0.2
7

cl /cd
-0.1
Cp

6.5
0
6
0.1
5.5
0.2
5
0 0.2 0.4 0.6 0.8 1
0.3 z/B
0 0.2 0.4 0.6 0.8 1
x/C 0.7
Optimized
Span station z/B=0.40 0.65 Baseline
-0.4
Optimized 0.6
Baseline
-0.3 0.55

0.5
-0.2 l
0.45
c

-0.1 0.4
Cp

0.35
0
0.3

0.1 0.25

0.2
0.2 0 0.2 0.4 0.6 0.8 1
z/B
0.1
0.3 Optimized
0 0.2 0.4 0.6 0.8 1
Baseline
x/C 0.09
Span station z/B=0.80
-0.3 0.08
Optimized
-0.25 Baseline
0.07
-0.2
d
c

-0.15 0.06

-0.1
0.05
Cp

-0.05

0 0.04

0.05
0.03
0 0.2 0.4 0.6 0.8 1
0.1 z/B

0.15
Fig. 36. Comparisons of spanwise aerodynamic coefficients
0.2 distribution between the baseline and optimized wings at
0 0.2 0.4 0.6 0.8 1
x/C a ¼ 6 :

Fig. 35. Comparisons of pressure coefficients on baseline and


optimized wing at different spanwise locations at a ¼ 6 : pressure region on the upper wing surface. The effective
local angle of attack in those regions is increased
movement that drags the surrounding air around them. compared with the same region of the wing without
The rotational energy is shed as vorticity into the wake endplate and as a result the lift increases.
of a finite-span lifting wing. The endplate leads to a reduction of the downwash
The endplate deters the flow from the high-pressure component in the region on the wing surface affected by
region on the lower wing surface to reach the low- the endplate. The increase in lift near the wing tip when
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 453

Rigid Wing Performance 10.5


9 Optimized
Baseline Baseline
Optimized 10

8.5
9.5

8 9

d
c /c
8.5
D

l
C /C

7.5
L

7 7.5

7
6.5
6.5
0 0.2 0.4 0.6 0.8 1
6 (a) z/B
3 4 5 6 7 8 9
αo
7
Rigid Wing Performance Optimized
0.36 Baseline
Baseline
Optimized
6.5
0.34

0.32 6

0.3
d
c /c

5.5
l

0.28
CL

0.26 5
0.24
4.5
0.22

0.2
4
0 0.2 0.4 0.6 0.8 1
0.18 (b) z/B
3 4 5 6 7 8 9
o
α
Fig. 38. Comparisons of spanwise lift-to-drag ratio between the
Rigid Wing Performance baseline and optimized rigid wings at different angles of attack:
0.06
Baseline (a) a ¼ 3 ; (b) a ¼ 9 :
Optimized
0.055
tion [52]. The low-pressure region associated with the
0.05
vortex core can be better visualized in Fig. 41.
0.045 The circulatory motion can be estimated by inspecting
the tangential velocity intensity pinffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Fig. ffi 42. The
D

0.04 tangential velocity is defined as: v2 þ w2 ; where v


C

0.035
and w are the vertical and lateral velocity components.
In the case of the wing without endplate an increase in
0.03 tangential velocity can be observed near the wing tip
indicating a stronger circulatory motion than in the case
0.025
of the wing with the endplate. This observation is
0.02 supported by the low-pressure vortex core region. The
3 4 5 6 7 8 9
αo tangential velocity increases in the region between the
two tip vortices where the induced velocities are summed
Fig. 37. Comparisons of the baseline and optimized rigid wings up. The presence of the endplate increases the vortex
at different angles of attack.
radius as can be seen in Fig. 42. In the case of the wing
without the endplate the vortex is more intense and the
the endplate is present is also linked with an increase in low-pressure core can be observed even further away
pressure on the lower surface of the wing behind the form the wing.
endplate. The tip vortex generates a strong circulatory Fig. 43 visualizes the pressure coefficient in the
motion at the wing tip, which is usually associated with a spanwise direction at x=C ¼ 0:44 for an angle of attack
low-pressure zone reflected in pressure coefficient varia- of 6 and with different endplate geometries. One can
ARTICLE IN PRESS
454 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Membrane Wing Performance Y


8.5
Baseline
Optimized

8 Z X

7.5
CL /CD

6.5

6
3 4 5 6 7 8 9
αo
Endplate
Membrane Wing Performance
0.34 Fig. 40. Schematic of endplate geometry and location.
Baseline
Optimized
0.32
created by the presence of the endplate affects more than
0.3 50% of the wingspan.
As the angle of attack is increased to 15 the tip
0.28
vortices strength increases and so does the low-pressure
region on the upper wing surface generated by the tip
L

0.26
C

vortex as shown in Fig. 44. The impact of the endplate


0.24 seem more confined to the immediate neighborhood.
The overall increase in lift for the wing with endplate is
0.22
diminished.
0.2 In Fig. 45 the spanwise pressure coefficients with and
without endplate at an angle of attack of 27 are plotted.
0.18
3 4 5 6 7 8 9 The wing tip vortex strength increases further compar-
αo ing with the previous cases and the area affected by the
Membrane Wing Performance vortex also increases and moves closer to the wing root.
0.055
Baseline
The high-pressure region on the lower wing surface
Optimized generated by the presence of the endplate decreases as
0.05 the tip vortex radius and strength increases. Conse-
quently, more fluid is driven from the area affected by
0.045
the endplate. Due to the same effect, the low-pressure
0.04
zone associated with vortex core increases and over-
comes the effect of the endplate.
D
C

0.035 The endplate can reduce the downwash, and therefore


increases the effective angle of attack and the lift.
0.03 However, as the angle of attack becomes substantial, the
wing tip vortex is of higher strength and can no longer
0.025 clearly affect the vortex structure to improve lift.
Furthermore, drag also increases along with the end-
0.02 plate.
3 4 5 6 7 8 9
o
α

Fig. 39. The optimized membrane wing performance at


different angles of attack. 6. Reduced-order representation using proper orthogonal
decomposition
observe that on the upper surface of the wing the
influence of the endplate is limited to a region near the POD captures intrinsic structures of a three-dimen-
tip; the rest of the wing remains unaffected. Meanwhile, sional, time dependent field via a statistical analysis of
on the lower surface of the wing, the high-pressure area the ensemble of data. By extracting a set of basis
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 455

Fig. 41. Pressure contours on rigid wing at different cross sections for a ¼ 6 with and without endplate, streamlines are also plotted:
(a) without endplate; (b) with higher endplates.

functions of the spatial variables that have maximum the research of turbulence [63]. Because it can handle a
energy content, POD effectively captures the principal large amount of data and highlight the main character-
structures and also greatly reduces the data storage. In istics, this technique has been widely investigated in
fluid dynamics, POD was first employed by Lumley in numerous flow regimes and geometries. For example,
ARTICLE IN PRESS
456 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Fig. 42. Tangential velocity contours on rigid wing at different cross sections for a ¼ 6 with and without endplate, streamlines are
also plotted: (a) without endplate; (b) with higher endplates.

Deane and Mavriplis [13] have used it to study the separated airfoil can be captured by a small set of
dynamics in separated flow past thick airfoils and shown ordinary differential equations. Hall et al. [31] have
that the vortex shedding past the airfoil can be under- reported an application of their POD formulation to a
stood in terms of the interaction of pairs of modes, and two-dimensional aeroelastic configuration. Silva and
the gross features of the weakly turbulent flow past the Bartels [90] have also reported their application of this
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 457

0.75 1.5

AoA = 6 deg AoA = 15 deg


0.7 1.4

0.65 1.3

0.6 1.2
without endplate
with shorter endplate
0.55 1.1
with higher endplate
-Cp

-Cp
without endplate
0.5 1 with shorter endplate

0.45 0.9

0.4 0.8

0.35 0.7

0.3 0.6
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
(a) z/Z (a) z/Z

0.2
0.5

AoA = 6 deg AoA = 15 deg


0.1 0.4

0.3
0
without endplate
0.2
with shorter endplate
-0.1 with higher endplate
0.1
without endplate
-Cp

-0.2
-Cp

0 with shorter endplate

-0.3 -0.1

-0.2
-0.4

-0.3
-0.5
-0.4

-0.6
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 -0.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
(b) z/Z (b) z/Z
Fig. 43. Spanwise pressure distribution at x=c ¼ 0:44 and
a ¼ 6 : (a) upper wing surface; (b) lower wing surface. Fig. 44. Spanwise pressure distribution at x=c ¼ 0:44 and
a ¼ 15 : (a) upper wing surface; (b) low wing surface.

technique to model the transonic flutter of the AGARD In a recent effort by Zhang et al. [114], POD is
445.6 wing. Epureanu et al. [20] have applied it in the extended to investigate a moving boundary problem.
frequency domain to obtain a reduced-order model of Their investigation focuses on both velocity and
the flow in a turbomachinery cascade and shown that pressure fields around the wings at angles of attack of
the reduced-order model with only 25-degrees of free- 6 and 27 : Overall, three-dimensional, time dependent
dom can accurately predict the unsteady response of the solutions for both membrane and rigid wings at 95 time
full system with approximately 1  105 degrees of steps are collected for the POD calculations.
freedom. Liberzon et al. [56] have used it to study the Zhang et al. [114] have chosen a maximal of 95
turbulent boundary layer and demonstrated that the first snapshot in their study. Their goal is not to get correct
modes accurately recover the instantaneous velocity time averaged values, but to check how consistent can
fields. Linear combination of the first three POD modes POD reproduce time dependent CFD results with
of the vorticity fluctuations can reproduce the coherent moving boundaries. Though only a few snapshots are
patterns of vorticity in the near-wall region. Cazemier [9] needed for the first several eigenmodes, more snapshots
and Ahlman et al. [1] have successfully used POD to are needed to ensure an accurate computation of higher
investigate the flow structures in cavity flow. These order, lower energy eigenmodes. Although the POD
studies show that POD can efficiently identify the analysis based on the 95 time steps solution is
systematic structure hidden in random fluctuations. insufficient for producing a correct time averaged value,
The recent development of this new technique has been it is capable of reproducing the CFD data base used. We
reviewed by Beran and Silva [6], and by Dowell and summarize the results reported by Zhang et al. [114] in
Hall [17]. the following.
ARTICLE IN PRESS
458 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

2.8
For POD analysis on the membrane wing, the flow
AoA = 27 deg
2.6
quantities are first interpolated from non-stationary grid
to a fixed, uniformed grid. Fig. 46 shows the comparison
2.4 of streamlines based on the interpolated velocities and
the original CFD results. Based on the interpolated
2.2
results, the POD analysis is calculated and the capability
of capturing the main structures of the flow field is
-Cp

2
demonstrated. The first three eigenmodes for 3D at
1.8
without endplate different spanwise locations are shown in Fig. 47. For
with shorter endplate
this case, Zhang et al. [114] have found that the first
1.6 eigenmode captures more than 99.8% of the total kinetic
energy. The second eigenmode, capturing the most
1.4
fluctuation energies, presents a large vortex structure
1.2
above the membrane wing and a small vortex structure
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
below it. This means that the largest velocity fluctua-
(a) z/Z
tions take place in those areas. The third eigenmode is
1.2 orthogonal to the first two, and it captures smaller
AoA = 27 deg vortices with less energy. Fig. 47 shows the first three
1
eigenmodes at two different spanwise locations. Due to
0.8 the membrane shape and the low aspect ratio of the
wing, the eigenmodes are quite different at different
0.6
locations.
0.4 without endplate To expand the investigation, the first three eigen-
modes for the flow fields with a ¼ 27 at various
-Cp

with shorter endplate


0.2
spanwise locations are shown in Fig. 48. It can be seen
0 that except for the first eigenmode, all others are
different from those for a ¼ 6 : This indicates, as
-0.2
expected, that the flow structures changes substantially
-0.4
when the angle of attack increases from 6 to 27 : Also,
the energy fractions for the first eigenmode is about
-0.6
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 99.56%, which implies the fluctuating energy at a ¼ 27
(b) z/Z is larger than that at a ¼ 6 :
Fig. 45. Spanwise pressure distribution at x=c ¼ 0:44 and
In order to better understand the POD representation
a ¼ 27 : (a) upper wing surface; (b) lower wing surface. for the membrane wing, Zhang et al. [114] also employ
the POD procedure for the rigid wing, which has the
same shape as the undeformed membrane wing. The
corresponding eigenmodes and eigenvalues for the rigid
wing flow fields are shown in Fig. 49. For both angles of
attack of 6 and 27 ; the first eigenmode shows no
vortex structures. Compared to the membrane wing
cases, the second eigenmode for the rigid wing has
different characteristics, indicating higher-order modes
exhibiting bigger differences.
Figs. 50 and 51 show how lift and drag vary when
(a) more eigenmodes are used for the reconstruction. Due to
the errors in the lift and drag, the ratio of lift over drag
also shows fluctuations. Since the viscous drag results
from the membrane wing are not easy to compute from
the fixed uniform grid, and since the main part of the
drag comes from the form drag, the L=D data in Zhang
et al. [114], excludes the consideration of the viscous
drag. Despite oscillations, the lift, drag and their ratio
(b) are close to the original values for the rigid wing when 28
Fig. 46. Streamlines on a cross surface of the membrane wing, eigenmodes are adopted for the reconstruction, despite
at the 25th time step, z=Z ¼ 0:23; and a ¼ 6 : (a) interpolated some oscillations (Fig. 52). For the membrane wing, the
streamlines; (b) original CFD results. relative error for L=D is larger (Fig. 53). This means that
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 459

(a)

(b)

(c)
Fig. 47. Eigenmodes for 3D membrane wing at different spanwise positions and a ¼ 6 : (a) the first eigenmode: left, z=Z ¼ 48:7%;
right, z=Z ¼ 66:8%. (b) the second eigenmode: left: z=Z ¼ 48:7%; right, z=Z ¼ 66:8%. (c) the third eigenmode: left: z=Z ¼ 48:7%;
right: z=Z ¼ 66:8%.

(a)

(b)

(c)
Fig. 48. Eigenmodes for 3D membrane wing at different spanwise positions and a ¼ 27 : (a) The first eigenmode: left: z=Z ¼ 48:7%;
right: z=Z ¼ 66:8%. (b) The second eigenmode: left: z=Z ¼ 48:7%; right: z=Z ¼ 66:8%. (c) The third eigenmode: left: z=Z ¼ 48:7%;
right: z=Z ¼ 66:8%:

the moving boundary makes the POD representation of viewed. The membrane wing exhibits self-
aerodynamics less accurate. Although Zhang et al. [114] initiated vibration even in the steady freestream. To
have used 16 eigenmodes to reconstruct the L=D shown accurately simulate the mutual interaction between
in Fig. 53, there are still some noticeable deviations. the flexible membrane structure and its surrounding
viscous flow, coupled fluid and structure computations
are performed. In terms of the computational require-
7. Conclusions ments for the coupled fluid and structure interactions,
time synchronization between the fluid solver and
Recent research in MAV wing aerodynamics, shape structural solver significantly reduces the error caused
optimization, and reduced-order representation is re- by phase lag, and the geometric conservation law helps
ARTICLE IN PRESS
460 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Timestep 25
0.227

0.226

0.225
(a)

L
0.224

0.223

0.222
(b) 0 20 40 60 80 100
Used POD modes
(a) Lift.

Timestep 25
0.0252

0.025

(c)
0.0248
Fig. 49. Streamline plots of the first three POD eigenmodes for
3D rigid wing flow fields at 48.7% spanwise for a ¼ 27 : (a) the
D

first eigenmode; (b) the second eigenmode; (c) the third 0.0246
eigenmode.

0.0244

maintain consistency in the context of the moving grid 0.0242


technique. 0 20 40 60 80 100
Used POD modes
In regard to the MAV aerodynamics, it seems clear
that tip vortices play an important role for the low (b) Drag.
aspect ratio wing. They reduce the effective angle while Fig. 50. Lift and drag versus numbers of eigenmodes for rigid
bringing low-pressure regions which provide additional wing at a ¼ 6 : Solid line: POD results; Dotted line: CFD
lift force. The low Reynolds number flow on the upper results.
wing surface is prone to separate due to the adverse
pressure gradient. The separation first occurs near the
root section because of the larger camber there. Tip
vortices and separated flow cause unsteady effects on the It is shown that the CFD-based optimization can be
wing. However, the aerodynamic coefficient differences conducted in a methodical manner by using the rigid
in the aerodynamic coefficients are small for the nominal wing as the surrogate model for the membrane wing.
large angles of attack considered here. Compared to the baseline, the optimization leads to
The membrane wing vibrates even in a steady lower camber near the root and higher camber near the
freestream; this vibration lowers the effective angle of tip, while still leaving the camber slightly higher at the
the membrane wing compared to the rigid wing. root than the tip. The lift-to-drag ratio improves over a
Typically, the membrane wing maintains lower effective range of angles of attack. However, at large angles of
angles of attack than the rigid wing of the same baseline attack, the improvement with the optimized shape
configuration. However, the shape inflation causes the diminishes, the L=D of the membrane wing varies less
membrane wing to exhibit a larger camber than the rigid with changes in the angle of attack, indicating that it can
one. Before reaching the stall limit, the time-averaged lift offer a more steady flight behavior than the rigid wing.
and drag of the membrane wing seem to be comparable The improvement is mostly seen within 70% of the inner
to that of the rigid wing. Both computational and wing. The lift of the optimized wing remains the same as
experimental studies indicate that under steady state the baseline wing, even though its camber is reduced
freestream, the membrane vibrates with Oð100Þ Hz: at the root. The improvement in aerodynamics of the
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 461

Timestep 25 Timestep 25
0.25 9.18

0.2
9.16

9.14
0.15
9.12
0.1
9.1

L/D
L

0.05 9.08

9.06
0
9.04
-0.05
9.02
-0.1
0 20 40 60 80 100 9
0 20 40 60 80 100
Used POD modes
(a) Used POD modes
(a) Lift .
Timestep 25 Timestep 50
0.025 10.5

0.02

0.015
10
0.01
L/D
D

0.005

0
9.5
-0.005

-0.01

-0.015
0 20 40 60 80 100 9
0 20 40 60 80 100
Used POD modes
(b) Used POD modes
(b) Drag.
Fig. 52. L=D versus numbers of eigenmodes at different time
Fig. 51. Lift and drag versus numbers of eigenmodes at the steps for rigid wing at a ¼ 6 : (a) at 25th time step; (b) at 50th
25th time step for membrane wing at a ¼ 6 : Solid line: POD time step.
results; Dotted line: CFD results.

optimized wing is largely realized via reduced form drag, of an arbitrary flow field on a number of eigenmodes is
recirculation, and better pressure distributions. verified. Furthermore, the first several eigenmodes can
The endplate increases lift by preventing the flow from clearly elucidate the main flow features, such as vortex
leading from the lower wing surface to the upper wing structures. Compared with the rigid wing, POD for
surface. The endplate effectively reduces the downwash, membrane wing has a slower convergence rate, and
increases the effective angle of attack and consequently needs more eigenmodes for an accurate reconstruction.
increases the lift. The effectiveness of the endplate One needs substantial number of eigenmodes to
diminishes as the angle of attack increases due to adequately reproduce the aerodynamic data, lift and
stronger wing tip vortex. However, a higher endplate drag, for the membrane wing.
can improve its performance. The endplate behaves as a Much progress has been made in this exciting research
vertically placed airfoil, and the generated additional area. With the continuous improvement in simulation,
force increases the drag. materials, fabrication, and measurement techniques, as
POD can be employed to represent the detailed well as the fast development in micro-systems, signifi-
computational results for both membrane and rigid cant potential exists to further advance the micro air
wings. By analyzing the energy fractions and the L2 vehicle as a vehicle platform for performing numerous
norms between the original CFD and the POD- missions, and as a technology platform for evaluating
reconstructed results, the convergence of the projection system.
ARTICLE IN PRESS
462 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

Timestep 25 [4] Batina J. Unsteady Euler airfoil solutions using unstruc-


9.7
tured dynamic meshes. AIAA Paper 1989-0115.
9.6 [5] Bendiksen OO. A new approach to computational
aeroelasticity. AIAA Paper 91-0939-CP, 1991.
9.5
[6] Beran PS, Silva WA. Reduced-order modeling: new
9.4 approaches for computational physics. AIAA Paper
2001-0853.
9.3
[7] Burgee S, Giunta AA, Narducci R, Watson LT,
L/D

9.2 Grossman B, Haftka RT. A coarse grained parallel


9.1
variable-complexity multidisciplinary optimization para-
digm. Int J Supercomput Appl High Performance
9 Comput 1996;10:269–99.
8.9 [8] Carmichael BH. Low Reynolds number airfoil survey.
NASA CR-165803, 1981.
8.8 [9] Cazemier W. Proper orthogonal decomposition and low
8.7 dimensional models for turbulent flows. University
0 20 40 60 80 100
Library Groningen, Groningen, 1997.
(a) Used POD modes [10] Chasman D, Chakravarthy S. Computational and
experimental studies of asymmetric pitch/plunge flap-
Timestep 50
9.4 ping—the secret of biological flyers. AIAA Paper 2001-
0859.
9.3 [11] Crompton MJ, Barrett RV. Investigation of the separa-
tion bubble formed behind the sharp leading edge of a flat
9.2
plate at incidence. J Aerospace Eng 2000;214:157–76.
[12] Cummings RM, Morton SA, Siegel SG, Bosscher S.
Numerical prediction and wind tunnel experiment for a
9.1
pitching unmanned combat air vehicle. AIAA Paper
L/D

2003-0417.
9
[13] Deane AE, Mavriplis C. Low-dimensional description of
the dynamics in separated flow past thick airfoils. AIAA J
8.9
1994;32:1222–7.
[14] DeLaurier JD. An aerodynamic model for flapping wing
8.8
flight. Aeronautical Journal 1993;97:125–30.
[15] Dickinson MH, Lehmann F, Sane SP. Wing rotation
8.7 and the aerodynamic basis of insect flight. Science
0 20 40 60 80 100
(b) Used POD modes 1999;284:1954–60.
[16] Ding H, Yang B, Lou M, Fang H. New numerical
Fig. 53. L=D versus numbers of eigenmodes at different time method for two-dimensional partially wrinkled mem-
steps for membrane wing at a ¼ 6 : (a) at 25th time step; (b) at branes. AIAA J 2003;41:125–32.
50th time step. [17] Dowell EH, Hall KC. Modeling of fluid-structure
interaction. Annu Rev Fluid Mech 2001;33:445–90.
[18] Drela M, Giles MB. Viscous-inviscid analysis of transonic
and low Reynolds number airfoils. AIAA J 1987;25:
Acknowledgements
1347–55.
[19] Ellington CP. The aerodynamics of hovering insect flight,
The present work is supported by US Air Force. We I. The quasi-steady analysis. Philos Trans R Soc London
have benefited from interacting with Dr. Peter Ifju and Ser A 1984;35:1–15.
his students, and Dr. Raphael Haftka. [20] Epureanu B, Hall K, Dowell E. Reduced-order models of
unsteady viscous flows in turbomachinery using viscous-
inviscid coupling. J Fluids Struct 2001;15:255–73.
[21] Eriksson LE. Generation of boundary-confirming grids
References around wing-body configurations using transfinite inter-
polation. AIAA J 1982;20:1313–20.
.
[1] Ahlman D, Soderlund F, Jackson J, Kurdila A, Shyy W. [22] Gad-el-Hak M. Micro-air-vehicles: can they be controlled
Proper orthogonal decomposition for time-dependent lid- better. J Aircraft 2001;38:419–29.
driven cavity flows. Numer. Heat Transfer Part B: [23] Gault DE. An investigation at low speed of the flow over
Fundam 2002;42:285–306. a simulated flat plate at small angles of attack using pitot
[2] Anderson Jr. JD. Introduction to flight, 3rd ed. New static and hot-wire probes. 1957, NACA TN-3876.
York: McGraw-Hill; 1989. [24] Gordnier RE, Fithen R. Coupling of a nonlinear finite
[3] Bathe KJ. Finite element procedures. Englewood, NJ: element structural method with a Navier–Stokes solver.
Prentice-Hall; 1996. Comput Struct 2003;81:75–89.
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 463

[25] Gordnier RE, Visbal MR. Development of a three- systems. Comput Methods Appl Mech Eng
dimensional viscous aeroelastic solver for nonlinear panel 2000;190:321–32.
flutter. J Fluids Struct 2002;16:497–527. [47] Kamakoti R, Berg M, Ljungqvist D, Shyy W. A
[26] Grasmeyer JM, Keennon MT. Development of the black computational study for biological flapping wing flight.
widow micro air vehicle. AIAA Paper 2001-0127. Trans Aeronaut Astronaut Soc Republic China
[27] Green AE, Adkins JE. Large elastic deformations. 2000;32:265–79 [Keynote Paper].
Oxford: The Clarendon Press; 1960. [48] Katz J. Low-speed aerodynamics: from wing theory to
[28] Farhat C, Geuzaine P, Brown G. Application of a three- panel methods. San Francisco, CA: Mc-Graw-Hill; 1979.
field nonlinear fluid–structure formulation to the predic- [49] Knill DL, Giunta AA, Baker CA, Grossman B, Mason
tion of the aeroelastic parameters of an F-16 fighter. WH, Haftka RT, Watson LT. Response surface methods
Comput Fluids 2003;32:3–29. combining linear and euler aerodynamics for supersonic
[29] Farhat C, Geuzaine P, Grandmont C. The discrete transport design. J Aircraft 1999;36:75–86.
geometric conservation law and the nonlinear stability [50] La Roche U, Palffy S. Wing grid, a novel device for
of ALE schemes for the solution of flow problems on reduction of induced drag on wings. ICAS 1996,
moving grids. J Comput Phys 2001;174:669–94. Sorrento, Italy.
[30] Farhat C, Lesoinne M. Two efficient staggered algo- [51] Lian Y. Membrane and adaptively-shaped wings for
rithms for the serial and parallel solution of three- micro air vehicles. Ph.D. Dissertation, University of
dimensional nonlinear transient aeroelastic problems. Florida, Gainesville, FL, 2003.
Comput Methods Appl Mech Eng 2000;182:499–515. [52] Lian Y, Shyy W. Three-dimensional fluid-structure
[31] Hall KC, Thomas JP, Dowell EH. Proper orthogonal interactions of a membrane wing for micro air vehicle
decomposition technique for transonic unsteady aero- applications. AIAA Paper 2003-1726.
dynamic flow. AIAA J 2000;38:1853–62. [53] Lian Y, Shyy W, Haftka R. shape optimization
[32] Hartwich PM, Agrawal S. Method for perturbing multi- of a membrane wing for micro air vehicles. AIAA Paper
block patched grids in aeroelastic and design optimiza- 2003-0106.
tion applications. AIAA Paper 1997-2038. [54] Lian Y, Shyy W, Ifju P, Verron E. A computational
[33] Herbert T. Parabolized stability equations. Annu Rev model for coupled membrane-fluid dynamics. AIAA
Fluid Mech 1997;29:245–83. Paper 2002-2972.
[34] Hutchison MG, Unger ER, Mason WH, Grossman B, [55] Lian Y, Steen J, Trygg-Wilander M, Shyy W. Low
Haftka RT. Variable-complexity aerodynamic optimiza- Reynolds number turbulent flows around a dynamically
tion of a high-speed civil transport wing. J Aircraft shaped airfoil. Comput Fluids 2003;32:287–303.
1994;31:110–6. [56] Liberzon A, Gurka R, Hetsroni G. Vorticity character-
[35] Ifju PG, Ettinger S, Jenkins D, Martinez L. Composite ization in a turbulent boundary layer using PIV and POD
materials for micro air vehicles. SAMPE J 2001;37:7–12. analysis. 4th International Symposium on Particle Image
[36] Ifju P, Jenkins D, Ettinger S, Lian Y, Shyy W, Waszak Velocimetry, Germany, September 17–19, 2001.
RM. Flexible-wing-based micro air vehicles. AIAA Paper [57] Lighthill MJ. Hydromechanics of aquatic animal propul-
2002-0705. sion. Annu Rev Fluid Mech 1969;1:413–45.
[37] Issa RI. Solution of the implicitly discretised fluid flow [58] Lissaman PBS. Low-Reynolds-number airfoils. Annu
equations by operator-splitting. J Comput Phys Rev Fluid Mech 1983;15:223–39.
1985;62:40–65. [59] Liu F, Zhu Y, Tsai HM, Wong ASF. Calculation of wing
[38] Jackson PS, Christie GW. Numerical analysis of three- flutter by a coupled fluid-structure method. J Aircraft
dimensional elastic membrane wings. AIAA J 2001;38:334–42.
1987;25:676–82. [60] Liu H, Kawachi K. A numerical study of insect flight.
[39] Jameson A. Time dependent calculations using multigrid, J Comput Phys 1998;146:124–56.
with applications to unsteady flows past airfoils and [61] Liu HT. Unsteady aerodynamics of a wortmann wing at
wings. AIAA Paper 1991-1596. low Reynolds number. J Aircraft 1992;29:532–9.
[40] Jenkins CH. Nonlinear dynamic response of membranes: [62] Lorillu O, Hureau J. Numerical and experimental
state of the art–update. Appl Mech Rev 1996;49:S41–8. analysis of two-dimensional separated flows over a
[41] Jenkins CH, Leonard JW. Nonlinear dynamic response of flexible sail. J Fluid Mech 2002;466:319–41.
membranes: state of the art. Appl Mech Rev [63] Lumley JL. The structure of inhomogeneous turbulent
1991;44:319–28. flows. In: Yaglom AM, Tararsky VI, editors. Atmo-
[42] Jones BM, Stalling JR. Aeronaut Soc 1938;38:747–70. spheric turbulence and radio wave propagation. Moscow:
[43] Jones KD, Platzer MF. An experimental and numerical Nauka; 1967. p. 166–78.
investigation of flapping-wing propulsion. AIAA Paper [64] Mason BH, Haftka RT, Johnson ER, Farley GL.
1999-0995. Variable complexity design of composite fuselage frames
[44] Jones KD, Platzer MF. Flapping-wing propulsion for a by response surface techniques. Thin Wall Struct
micro air vehicle. AIAA Paper 2000-0897. 1998;32:235–61.
[45] Jones KD, Platzer MF. Experimental investigation of the [65] Mooney M. A theory of large elastic deformation. J Appl
aerodynamic characteristics of flapping-wing micro air Phys 1940;11:582–92.
vehicles. AIAA Paper 2003-0418. [66] Mueller TJ, editor. Proceedings of the Conference on
[46] Kalro V, Tezduyar TE. A parallel 3D computational Low Reynolds Number Aerodynamics. University of
method for fluid-structure interactions in parachute Notre Dame, Notre Dame, IN, 1989.
ARTICLE IN PRESS
464 Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465

[67] Mueller TJ, editor. Proceedings of the Conference on [87] Shyy W, Mittal R. Solution methods for the incompres-
Fixed, Flapping and Rotary Wing Vehicles at Very Low sible Navier–Stokes equations. In: Johnson R, editor.
Reynolds Numbers. University of Notre Dame, Notre Handbook of fluid dynamics. Boca Raton, FL: CRC
Dame, IN, 2000. Press; 1998. p. 31-1 to 31-33.
[68] Mueller TJ, DeLaurier JD. Aerodynamics of small [88] Shyy W, Smith RW. Computation of Laminar Flow and
vehicles. Annu Rev Fluid Mech 2003;35:89–111. Flexible Structure Interaction. In: Hafez M, Oshima K,
[69] Nelsen JN. Theory of flexible aerodynamics surfaces. editors. Computational fluid dynamics review. New
J Appl Mech 1963;30:435–42. York: Wiley; 1995. p. 777–96.
[70] Newman BG. Aerodynamics theory for membrane and [89] Shyy W, Udaykumar HS, Rao MM, Smith RW.
sails. Progress Aerospace Sci 1987;24:1–27. Computational fluid dynamics with moving boundaries.
[71] Oden JT, Sato T. Finite strains and displacements of Washington, DC: Taylor & Francis; 1996.
elastic membranes by the finite element method. Int J [90] Silva WA, Bartels RE. Development of reduced-order
Solids Struct 1967;3:471–88. models for aeroelastic analysis and flutter prediction using
[72] Oliveira PJ, Issa RI. An implicit PISO algorithm for the the CFL3DV6.0 Code. AIAA Paper 2002-1596, 2002.
computation of buoyancy-driven flows. Numer Heat [91] Smith MJ. Computational considerations of an Euler/
Transfer Part B 2001;40:473–93. Navier–Stokes aeroelastic method for a hovering rotor.
[73] Patankar SV, Spalding DB. A calculation procedure for J Aircraft 1996;33:429–34.
heat, mass and momentum transfer in three-dimensional [92] Smith MJ, Hodges DH, Cesnik CES. Evaluation of
parabolic flows. Int J Heat Mass Transfer 1972;15: computational algorithms suitable for fluid–structure
1787–806. interactions. J Aircraft 2000;37:282–94.
[74] Pelletier A, Mueller TJ. Low Reynolds number aero- [93] Smith RW, Shyy W. Computational model of flexible
dynamics of low-aspect-ratio, thin/flat/cambered-plate membrane wings in steady laminar flow. AIAA J
wings. J Aircraft 2000;37:825–32. 1995;33:1769–77.
[75] Pulliam T. Time accuracy and the use of implicit [94] Smith RW, Shyy W. Incremental potential flow based
methods. AIAA Paper 93-3360-CP, 1993. membrane wing element. AIAA J 1997;35:782–8.
[76] Reed HL, Saric WS. Linear stability theory applied [95] Subbaraj K, Dokainish MA. A survey of direct time-
to boundary layers. Annu Rev Fluid Mech 1996;28: integration methods in computational structural dy-
389–428. namics—II. Implicit methods. Comput Struct 1989;32:
[77] Rempfer D. Low-dimensional modeling and numerical 1387–401.
simulation of transition in simple shear flows. Annu Rev [96] Tani I. Low-speed flows involving bubble separations.
Fluid Mech 2003;35:229–65. .
In: Kuchemann D, l. Sterne HG, editors. Progress in
[78] Reuther J, Alonso JJ, Rimlinger MJ, Jameson A. aeronautical science, vol. 5. New York: Pergamon; 1964,
Aerodynamic shape optimization of supersonic aircraft p. 70–103.
configuration via an adjoint formulation on distributed [97] Templin RJ. The spectrum of animal flight: insects to
memory parallel computers. Comput Fluids 1999;28: pterosaurs. Progress Aerospace Sci 2000;36:393–436.
675–700. [98] Thakur S, Wright J, Shyy W. STREAM: a computational
[79] Robinson BA, Batina JT, Yang HTY. Aeroelastic fluid dynamics and heat transfer Navier–Stokes solver.
analysis of wings using the Euler equations with a Theory and applications. Streamline Numerics, Inc., and
deforming mesh. J Aircraft 1991;28:781–8. Computational Thermo-Fluids Laboratory, Department
[80] Rumsey CL, Sanetrik MD, Biedron RT, Melson ND, of Mechanical and Aerospace Engineering Technical
Parlette EB. Efficiency and accuracy of time-accurate Report, Gainesville, Florida, 2002.
turbulent Navier–Stokes computations. Comput Fluids [99] Thaokar RM, Kumaran V. Stability of fluid flow past a
1996;25:217–36. membrane. J Fluid Mech 2002;472:29–50.
[81] Schmit R, Glauser M, Gorton S. Low dimensional tools [100] Thomas PD, Lombard CK. Geometric conservation law
for flow-structure interaction problems: application to and its application to flow computations on moving grids.
micro air vehicles. AIAA Paper 2003-0626. AIAA J 1979;17:1030–7.
[82] Schuster DM, Liu DD, Huttsell LJ. Computational [101] Thwaites B. Aerodynamics theory of sails—Part I. Proc R
aeroelasticity: success, progress, challenge. AIAA Paper Soc 1961;A261:402–42.
2003-1725. [102] Torres GE, Mueller TJ. Aerodynamic characteristics of
[83] Schuster D, Vadyak J, Atta E. Static aeroelastic analysis low aspect ratio wings at low Reynolds number. In:
of fighter aircraft using a three-dimensional Navier– Mueller TJ, editor. Fixed and flapping wing aerody-
Stokes algorithm. J Aircraft 1990;27:820–5. namics for micro air vehicle applications, vol. 195.
[84] Shyy W. Computational modeling for fluid flow and Reston, VA: AIAA; 2001. p. 191–213.
interfacial transport. Amsterdam, The Netherlands: [103] Van Doormaal JP, Raithby GD. Enhancements of the
Elsevier; 1994. SIMPLE method for predicting incompressible fluid
[85] Shyy W, Berg M, Ljungqvist D. Flapping and flexible flows. Numer Heat Transfer 1985;67:147–63.
wings for biological and micro vehicles. Process Aero- [104] Verron E, Marckmann G, Peseux B. Dynamic inflation of
space Sci 1999;35:455–506. non-linear elastic and viscoelastic rubber-like membranes.
[86] Shyy W, Jenkins DA, Smith RW. Study of adaptive shape Int J Numer Mech Eng 2001;50:1233–51.
airfoils at low Reynolds number in oscillatory flow. [105] Vest NS. Unsteady aerodynamic model of flapping wings.
AIAA J 1997;35:1545–8. AIAA J 1996;34:1435–40.
ARTICLE IN PRESS
Y. Lian et al. / Progress in Aerospace Sciences 39 (2003) 425–465 465

[106] Vest MS, Katz J. Aerodynamic study of a flapping wing, [112] Wilson EL, Farhoomand I, Bathe KJ. Nonlinear
micro-UAV. AIAA 1999-0994. dynamic analysis of complex structures. Earthquake
[107] Viieru D, Lian Y, Shyy W. Investigation of tip vortex on Eng Struct Dyn 1973;1:241–52.
aerodynamic performance of a micro air vehicle. AIAA [113] Young AD, Horton HP. Some results of investigation of
Paper 2003-3597. separation bubbles. AGARD CP 1966;4:779–811.
[108] Voelz K. Profil und Luftriebeines Segels. ZAMM [114] Zhang BN, Lian Y, Shyy W. Proper orthogonal
1950;30:301–17. decomposition for three-dimensional membrane wing
[109] Wang ZJ. Vortex shedding and frequency selection in aerodynamics. AIAA Paper 2003-3917.
flapping flight. J Fluid Mech 2000;410:323–41. [115] Zienkiewicz OC, Taylor RL. The finite element
[110] Waszak RM, Jenkins NL, Ifju P. Stability and control method: solid and fluid mechanics dynamics and
properties of an aeroelastic fixed wing micro aerial non-linearity, vol. II. New York: MacGraw-Hill;
vehicle. AIAA Paper 2001-4005. 1989.
[111] Weis-Fogh T. Quick estimates of flight fitness in hovering [116] Zwaan RJ, Prananta BB. Fluid/structure interaction in
animals, including novel mechanisms for lift production. numerical aeroelastic simulation. Int J Non-Linear Mech
J Exp Biol 1973;59:169–230. 2002;37:987–1002.

You might also like