Progress in Aerospace Sciences 35 (1999) 205—334

Nonlinear aeroelastic analysis of airfoils:

bifurcation and chaos
B.H.K. Lee *, S.J. Price
, Y.S. Wong
Aerodynamics Laboratory, Institute for Aerospace Research, National Research Council, Ottawa,
Ontario, Canada K1A 0R6

Department of Mechanical Engineering, McGill University, Montreal, Quebec, Canada H3A 2K6
Department of Mathematical Sciences, University of Alberta, Edmonton, Alberta, Canada T6G 2G1
Received 17 July 1998

Contents

1. Introduction 207 5. Bifurcation and chaos of airfoils with structural

2. Nonlinearities encountered in aeroelastic nonlinearities 248
behavior of aircraft structures 208 5.1. Stability and bifurcation analysis 248
2.1. Structural nonlinearities 208 5.2. Cubic nonlinearity 257
2.2. Aerodynamics nonlinearities 217 5.3. Freeplay with preload 271
3. Formulation of the nonlinear aeroelastic 5.4. Hysteresis nonlinearity 286
equations for airfoils 221 5.5. Forced oscillations 296
3.1. Equations with structural nonlinearities and 6. Bifurcation and chaos of airfoils with
subsonic aerodynamics 221 aerodynamic nonlinearities 300
3.2. Equations with nonlinear aerodynamics 228 6.1. Transonic flows 300
4. Solution techniques of equations with 6.2. Dynamic stall 308
structural nonlinearities 239 7. Gust effects: airfoils in longitudinal atmospheric
4.1. Finite difference scheme 239 turbulence 319
4.2. Runge—Kutta scheme 240 8. Conclusions 326
4.3. Describing function 242 References 328
4.4. Analytical techniques 245 Appendix 332

Nomenclature

a non-dimensional distance from airfoil C aerodynamic lift coefficient

 *
mid-chord to elastic axis C pitching moment coefficient
+
b airfoil semi-chord c chord
C generalized damping DOF degree of freedom

* Corresponding author. Tel.: 001 613 998 3401; fax: 001 613 998 1281; e-mail: [email protected].

0376-0421/99/$ — see front matter  1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 3 7 6 - 0 4 2 1 ( 9 8 ) 0 0 0 1 5 - 3
206 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

e total energy u velocity in the x-direction

F generalized force of ith mode u gust velocity

G(m) structural nonlinearity in plunge v velocity in the y-direction
G Gaussian white noise w velocity in the z-direction

h plunge displacement X system variable vector
J Jacobian matrix X system equilibrium point
#
K generalized stiffness of ith mode x non-dimensional distance from elastic axis to
G a
K,K linear stiffness in pitch and plunge center of mass
? K
k reduced frequency based on semi-chord y variable vector
("ub/º) a pitch angle of airfoil
k reduced frequency based on chord ("uc/º) a pitch angle amplitude of limit cycle oscillation
 
¸ turbulence scale a dynamic stall angle
"
LCO limit cycle oscillations a static stall angle
1
¸ non-dimensional turbulence scale ("¸/b) a unsteady decay parameter
 5
M Mach number b,b coefficients of cubic spring in pitch and plunge
? K
M generalized mass of ith mode d pitch angle for the central region of the freeplay
G
M(a) structural nonlinearity in pitch stiffness, see Fig. 2b
m airfoil mass e ,e constants in Wagner’s function
 
P externally applied force U gust velocity PSD

PDD probability density distribution U non-dimensional gust velocity PSD
 
PSD power spectral density U white noise PSD

p pressure Wagner’s function
Q externally applied moment mode shape of ith mode
G
Q amplitude of applied moment j eigenvalue

q generalized coordinates k airfoil/air mass ratio ("m/nob)
G
R response amplitude of pitch motion o density
r response amplitude of plunge motion p gust velocity variance

r radius of gyration about elastic axis p non-dimensional gust velocity variance
?  
S static moment about elastic axis l parameter
¹ temperature q non-dimensional time ("ºt/b)
TSD transonic small disturbance t viscosity
t time u frequency
º free stream velocity u,u natural frequencies in pitch and plunge
? K
º linear flutter speed uN frequency ratio ("u /u )
* K ?
º mean free stream velocity m non-dimensional plunge displacement ("h/b)
º* non-dimensional velocity ("º/bu ) m plunge amplitude of limit cycle oscillation
? 
º*, º* non-dimensional bifurcation speeds t ,t constants in Wagner’s function
   
º* non-dimensional velocity ("º /bu ) f,f viscous damping ratios in pitch and plunge
? ? K
º* non-dimensional linear flutter speed
*

Abstract

Different types of structural and aerodynamic nonlinearities commonly encountered in aeronautical

engineering are discussed. The equations of motion of a two-dimensional airfoil oscillating in pitch and
plunge are derived for a structural nonlinearity using subsonic aerodynamics theory. Three classical
nonlinearities, namely, cubic, freeplay and hysteresis are investigated in some detail. The governing equations
are reduced to a set of ordinary differential equations suitable for numerical simulations and analytical
investigation of the system stability. The onset of Hopf-bifurcation, and amplitudes and frequencies of limit
cycle oscillations are investigated, with examples given for a cubic hardening spring. For various geometries
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 207

of the freeplay, bifurcations and chaos are discussed via the phase plane, Poincaré maps, and Lyapunov
spectrum. The route to chaos is investigated from bifurcation diagrams, and for the freeplay nonlinearity it is
shown that frequency doubling is the most commonly observed route. Examples of aerodynamic nonlineari-
ties arising from transonic flow and dynamic stall are discussed, and special attention is paid to numerical
simulation results for dynamic stall using a time-synthesized method for the unsteady aerodynamics. The
assumption of uniform flow is usually not met in practice since perturbations in velocities are encountered in
flight. Longitudinal atmospheric turbulence is introduced to show its effect on both the flutter boundary and
the onset of Hopf-bifurcation for a cubic restoring force.  1999 Elsevier Science Ltd. All rights reserved.

1. Introduction

Aeroelasticity is a multi-disciplinary field of study dealing with the interaction of inertia,

structural and aerodynamic forces. Classical theories assume linear aerodynamics and structures,
and the problem reduces to the solution of a set of linear equations that can readily be programmed
for the computer. For many decades, the classical approach has been successful in providing
approximate estimates of aircraft response to gust, turbulence and external excitations. The flutter
boundaries are often quite accurately predicted when compared to flight test results. However,
when the airspeed increases to high subsonic or transonic Mach numbers, linear aerodynamics
usually give insufficiently accurate results, an example of which is the transonic dip that linear
aerodynamics fail to detect. Also, flow separation and shock oscillations can introduce phenomena,
such as limit cycle oscillations which classical aeroelasticity is unable to handle. Nonlinear
aerodynamic effects are more difficult to analyze since the fluid motion is governed by equations
where analytical solutions are practically non-existent. Usually, we have to resort to numerical
techniques to solve the aerodynamic equations. Coupling them to the structural motion can be
achieved using a numerical time marching scheme. The computation is usually extremely
involved and time-consuming, and often only a specific aircraft configuration or flight condition is
analyzed.
Structural nonlinearities arise from worn hinges of control surfaces, loose control linkages,
material behavior and various other sources. Aging aircraft and combat aircraft that carry heavy
external stores are more likely to be influenced by effects associated with nonlinear structures. This
type of nonlinearity can be treated as a concentrated nonlinearity, and usually can be approxi-
mated by one of the three classical structural nonlinearities, namely, cubic, bilinear and hysteresis.
In this review, emphasis is placed on structural nonlinearities since they can be analyzed more
readily using existing theories in nonlinear system dynamics. Stability, bifurcation and chaos are
amongst the subjects commonly encountered in nonlinear aeroelasticity where tools developed to
study nonlinear dynamics problems can readily be applied. In order to limit the complexity of the
aeroelastic system with structural nonlinearities, we consider a two-dimensional airfoil oscillating
in pitch and plunge using subsonic aerodynamics where approximate expressions for the force and
moment are available. A two-degree-of-freedom system is of particular interest when we deal with
binary flutter, which offers a good physical insight into the instability problem. Extending
the analysis to more than two modes is straightforward, but the algebra is considerably more
complex.
208 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

This review gives an introductory exposure to theoretical developments in nonlinear aeroelastic-

ity. In the discussion on analytical solutions, the emphasis is placed on structural nonlinearities,
since more theoretical studies have been carried out and published in the open literature on this
subject. Section 2.1 discusses the types of structural nonlinearities often used to represent aircraft
structures, and introduces the cubic, bilinear and hysteresis springs. The types of aerodynamic
nonlinearities associated with transonic Mach numbers, flow separation and viscous effects are
given in Section 2.2. The equations of motion of a two-dimensional airfoil with structural nonlinear
restoring forces are derived in Section 3.1. These equations are reduced to a set of ordinary
differential equations for convenience in numerical simulation and analytical investigation. The
equations of motion with nonlinear aerodynamics are discussed in Section 3.2.1 for transonic flows
where the full potential, Euler and Navier—Stokes formulations are briefly discussed. The methodo-
logy used to analyze separated flows in dynamic stall is given in Section 3.2.2. The solution
techniques for airfoil motion in subsonic flows are presented in Section 4.1, where the details of
a finite difference scheme are given. The Runge—Kutta method is outlined in Section 4.2, while the
describing function technique, whereby the nonlinear spring is replaced by an equivalent linear
spring for harmonic motion is derived in Section 4.3. Analytical techniques where the airfoil motion
is periodic are derived in Section 4.4. Stability and bifurcation analyses are carried out in Section
5.1, and examples of limit-cycle-oscillations, bifurcations and chaos for the cubic, bilinear and
hysteresis springs are discussed in Sections 5.2, 5.3 and 5.4. Examples of forced oscillations are
shown in Section 5.5, and the only example available in the literature is for a cubic restoring force.
Examples of bifurcation and chaos of airfoil motion with aerodynamic nonlinearities are given in
Section 6.1 for transonic flows and Section 6.2 for dynamic stall. Section 7 considers the effect of
longitudinal atmospheric turbulence on the dynamics of an airfoil with a hardening cubic struc-
tural nonlinearity showing the flutter speed is advanced by the presence of the gust, while the Hopf
bifurcation is postponed. Finally, some concluding remarks on the subject of nonlinear aeroelastic-
ity and analysis techniques are given in Section 8.

2. Nonlinearities encountered in aeroelastic behavior of aircraft structures

2.1. Structural nonlinearities

The assumption of structural linearity is frequently made so that the available standard
computational methods can be used to determine the divergence and flutter characteristics of
aerodynamic surfaces. Linear theory predicts the magnitude of dynamic pressure or flight velocity
above which the system under consideration becomes unstable and the motion grows exponenti-
ally in time. However, aircraft structures often exhibit nonlinearities that affect not only the flutter
speed, but also the characteristics of the motion itself. Little effort has been devoted to nonlinear
effects and this subject remained relatively unexplored until the 1950s when Woolston et al.
[120,121] and Shen [96] initiated analog and numerical studies on this subject. An understanding
of the nonlinear behavior of the system is crucial to the efficient and safe design of aircraft wings
and control surfaces.
The structure of an aircraft is very complicated and consequently it is virtually impossible to
know the nature of any nonlinearity from knowledge of the behavior of the material and parts of
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 209

which it is made. An excellent review of some possible structural nonlinearities and their effect on
aeroelastically induced vibrations is given by Brietbach [18,19]. Structural nonlinearities may be
classified as being either distributed or concentrated. In general, distributed structural nonlineari-
ties are governed by elastodynamic deformations that affect the whole structure. Concentrated
nonlinearities, on the other hand, act locally and are commonly found in control mechanisms or
in the connecting parts between wing, pylon, engine or external stores. In this review, we shall
consider only concentrated nonlinearities and investigate their effects on the aeroelastic behavior of
aerosurfaces.
A thin wing or propeller blade which is being twisted will most likely behave as a cubic
hardening spring which becomes stiffer as the angle of twist increases. If buckling occurs, its effects
can be approximated by a softening spring whose stiffness decreases when the displacement is
increased. This may be of importance to panel flutter. Kinetic heating at high Mach numbers can
produce large reductions in structural stiffness. Depending on the temperature and initial condi-
tions, the nonlinearity can be a hardening or softening spring type. The force versus displacement
curves are shown in Fig. 1 for both types of cubic nonlinearities. Cubic nonlinearities in one-degree-
of-freedom (DOF) mechanical and electrical systems can often be represented by a Duffing’s
equation which has been the subject of investigation for many years. The two classical books by
Stoker [99] and Hayashi [43] give an excellent account of the behavior of Duffing’s equation, and
more recent studies by Ueda [112] deal with chaotic characteristics of this equation. Experi-
mentally, a cubic hardening spring for a one DOF system can easily be demonstrated in the
classroom by vibrating a thin steel strip fastened at one end and driven magnetically by a small coil
at the other. Under the correct conditions, the jump phenomenon in Duffing’s equation can be
simulated.
In power-operated control systems and spring-tab systems, the most common cause for non-
linear behavior is backlash in, say, a control linkage. This produces a flat spot nonlinearity which
possesses a force—displacement characteristic of the type shown in Fig. 2a. For small displacements
the spring offers no resistance to the movement of the control surface. A modified form of the flat

Fig. 1. Force versus displacement curve for (a) cubic hardening spring, (b) cubic softening spring.
210 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 2. Force versus displacement curve for (a) flat spot without preload, (b) general bilinear spring.

Fig. 3. Aileron hinge moment of a glider versus hinge angle, symmetrical loading (from Ref. [18]).

spot arises if the spring has a preload shown in Fig. 2b. The flat spot can also be replaced by
a spring of different stiffness located not at the equilibrium position, but at a certain displacement
from it. Fig. 3 shows experimental data of the aileron hinge moment [18] of a glider versus hinge
angle for the case of static moments symmetrically acting in the sense opposite to the regular
operation of the aileron system. This behavior can be adequately represented by the model shown
in Fig. 2a.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 211

Certain types of spring tab systems that use preload springs encounter a stiffness nonlinearity
which is important in flutter prediction. This nonlinearity can be deliberately introduced so that
the pilot operates the main control surface directly when the airloads are small, while for large
airloads, he moves it indirectly by deflecting a tab. The result is that the main surface stiffness
coefficient has one value when its displacement is less than a certain amount, and a smaller value at
larger displacements. It behaves like a softening spring, though one in which there is a sudden drop
in stiffness at a certain displacement rather than one like a cubic softening spring in which there is
a continuous decrease in stiffness as the displacement increases. A schematic of a simple spring tab
system and the force—deflection curve are shown in Figs. 4 and 5.
Hinges of aircraft control surfaces often exhibit a freeplay with preload characteristics. The
preload depends on the aerodynamic forces acting on the control surfaces for given flight

Fig. 4. Sketch of a spring tab system (from Ref. [18]).

Fig. 5. Schematic force deflection diagram of a pre-tensioned spring tab (from Ref. [18]).
212 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

conditions. The F/A-18 aircraft wing folds at about 1/4 span measured from the wing tip. Over
years of usage, there is a decrease in stiffness and it was observed from flight tests that with
light-to-intermediate weight outboard stores and with wing tip missiles off, the aircraft was subject
to lightly damped low-frequency limit cycle oscillations (LCO). There is also another nonlinearity
at the outboard leading-edge flap hinge that may cause flap oscillations. The locations of these two
hinges on the aircraft wing are shown in Fig. 6. Lee and Tron [66] investigated limit cycle
oscillations using a describing function technique. Ground testing of the aircraft was carried out
using a static test rig. A typical curve of the hinge moment versus wing-fold rotation is shown
in Fig. 7. A small amount of hysteresis is present, but Lee and Tron [66] neglected this in their
investigation and treated the nonlinearity as a bilinear spring. Here K is the design stiffness and

d is the amount of freeplay. Fig. 8 shows a typical ground test moment—displacement curve for the
outboard leading-edge hinge. Again a small hysteresis is detected and the nonlinearity from static
tests is a slightly non-symmetrical freeplay without preload when the small amount of hysteresis
present is neglected.
Since the flight conditions determine the amount of preload, Lee and Tron [66] used different
flap settings and calculated the loads corresponding to one particular test point at M"0.95 and
altitude of 7000 ft above sea level using a transonic small disturbance aerodynamic code. They
presented the normalized wing-fold stiffness for three flap settings according to the flap schedule as
shown in Table 1.
The equivalent stiffness (to be discussed in Section 4.3) is given in Fig. 9 for the wing-fold hinge.
For the leading-edge flap, they used four flap settings and also obtained a set of equivalent stiffness
curves.
Another simple cause of nonlinearity is solid friction. A control surface restrained only by solid
friction will ideally have a force—velocity characteristic like that shown in Fig. 10a. In practice, it

Fig. 6. Schematic of locations of wing-fold and outboard leading-edge flap hinges on the CF-18 wing (from Ref. [66]).
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 213

Fig. 7. Moment versus wing-fold hinge rotation of the CF-18 wing (from Ref. [66]).

Fig. 8. Moment versus leading-edge flap rotation of the CF-18 wing (from Ref. [66]).

will probably behave more like Fig. 10b. If there is both friction and backlash, we have a hysteresis
nonlinearity shown in Fig. 11.
A hysteresis nonlinearity is characterized by a force or moment which increases linearly with
displacement until a value is reached at which a jump occurs, after which the system is again linear.
214 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Table 1
Flap settings

Case Trailing-edge Aileron angle (AIL) (deg) Leading-edge

flap angle (TEF) (deg) flap angle (LEF) (deg)

a 1 !4 2
b 1 !2 !3
c 1 4 0

Fig. 9. Normalized equivalent wing-fold stiffness vs. amplitude ratio at flap-aileron settings: curve (a) AIL"!4°,
LEF"2°; curve (b) AIL"!2°, LEF"!3°; curve (c) AIL"4°, LEF"0°; TEF is fixed at 1° in all cases, A is the
amplitude of LCO and d is the dimensional of the freeplay (from Ref. [66]).

On the return path, a corresponding jump occurs at another value of the force or moment. The
hysteresis introduces damping, and the time variations of displacement for a one-DOF mechanical
vibration system usually shows damping to vary with amplitude of oscillations for moderately
large amplitudes. However, for small amplitudes the system oscillates on a line through the box
with a linear spring constant and the hysteretic damping becomes zero. For very large amplitudes,
the effect of the hysteresis is small and can be neglected in the limit when the ratio of the width of
the hysteresis box d to the oscillation amplitude approaches zero.
Another effect of hysteresis on the structural properties of a vibrating system is to introduce
an effectively weaker spring as in the case of a freeplay. At amplitudes of oscillation for which the
system passes completely through the hysteresis box, the frequency of oscillation is less than
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 215

Fig. 10. Schematic of force versus velocity for a solid friction nonlinearity: (a) ideal, (b) practical.

Fig. 11. Notations used for a hysteresis nonlinearity.

the frequency at low amplitudes where the system is linear. For very large amplitudes, the effective
stiffness is again approximately linear.
Many structural systems are composed of ductile materials and are assembled in such a way that
they exhibit a hardening hysteresis behavior under cyclic loading. An example of a hardening
system can be found in riveted and bolted structures where slipping connections provide a major
contribution to the overall damping within the structure. In general, this type of hardening
hysteresis behavior occurs in the large deformation of composite structural systems where second-
ary structural elements contribute to the response only after the primary structural elements have
216 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

yielded or slipped at the connections. Miller [80] gave a physical model of a hysteresis by con-
sidering a structural assemblage consisting of linear springs with linear stiffnesses and a Coulomb
friction element with a slip force.
For sweepable wing and sweepable wing mounted stores like the F-111 aircraft, large non-
linearities can exist because a considerable number of joints (with possible freeplay) and bear-
ings (with freeplay and friction) are necessary. A view of the wing sweep-underwing
stores alignment mechanical system [22] for static ground test is shown in Fig. 12. This can
be represented schematically by Fig. 13 using a combination of springs and dampers. At low
amplitude, the store yaw motion involves the pylon and wing fore and aft stiffness (the pylon
behaves as if it was clamped to the wing). When a certain amplitude at which the friction preload in
the pylon bearings is exceeded, there is a relative motion of the pylon with respect to the wing, and
also the control rod stiffness becomes important after the backlash is exceeded. Hysteresis
measurements were made for the store yaw, and typical static test results for an inboard store are
shown in Fig. 14.
Basically, the concentrated nonlinearities can be classified roughly into three types. The aero-
elastic response can be quite different and these will be the subject of investigation in this
review.

Fig. 12. Schematic view of the wing sweep-underwing stores Fig. 13. Idealization of the system for Fig. 12 for
alignment system (from Ref. [22]). wing and store yaw motion (from Ref. [22]).
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 217

Fig. 14. Measured hysteresis curve of in-board store yaw deflections (from Ref. [22]).

2.2. Aerodynamic nonlinearities

When the flow velocity is large, compressibility effects are important and they can have a
pronounced influence on the aeroelastic characteristics of the airfoil response compared to that
using incompressible theory. We shall consider a few situations in which aerodynamic nonlineari-
ties are important in their interaction with the structural dynamics of the airfoil.
The first is associated with the presence of shock waves in transonic flows. The flow is assumed to
be inviscid and separation does not occur. In this situation, the unsteady forces generated by the
motion of the shock wave have been shown to destabilize single degree-of-freedom airfoil pitching
motion and affect the bending-torsional flutter by lowering the flutter speed at the so called
transonic dip regime. Ashley [4] pointed out that the shock waves located on the upper and lower
wing surfaces move periodically with large phase lags with the oscillatory airfoil motion. Their
motions are the predominant factor in the anomalies observed when the Mach number approaches
unity. He quoted an example when the flutter boundary can be 25% lower in dynamic pressure
than its conventional shock-free counterpart. The distance traversed by the shocks on the airfoil
surfaces depends on the frequency and can be quite large compared with the airfoil motion.
A number of computer codes are available to calculate the unsteady airloads at transonic flow
conditions. They range from transonic small disturbance formulation [8,9], full potential [49],
to the Euler equations [11]. Fig. 15 shows the shock wave time history for plunging oscillation of
a NACA 64A006 airfoil at M"0.85 and reduced frequency based on the chord k "0.5 and 2

calculated by Lee [62] using a 2D transonic small disturbance aerodynamics code. At k "0.5, we

see a peculiar shock behavior occurs. During part of the oscillation cycle, the shock disappears only
218 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 15. Shock wave position for plunging oscillation of a NACA64A006 airfoil at M"0.85, reduced frequency k "0.5

and 2 (from Ref. [62]).

to reappear at a later time. The forces generated are nonlinear and usually a number of harmonics
are present.
When viscous effects are considered, flow separation can occur due to shock-boundary layer
interaction. This can cause single DOF flutter, control surface buzz and buffeting. Lee [67] studied
shock oscillations on a rigid supercritical airfoil and showed the growth of the shock induced
separation bubble with increase in angle-of-attack a for given Mach numbers. For small incidence,
the separation reattaches on the airfoil surface and the bubble size grows with a. The shock can also
induce trailing-edge separation and the two separation regions can merge to produce a fully
separated flow behind the shock as shown in Fig. 16. The unsteady air loads are quite large and
have characteristic frequencies that can be near those involved in flutter. Lee [67] proposed
a feedback mechanism for self-sustained shock motion on a 2D supercritical airfoil based on
experimental measurements. Later, Lee et al. [68] carried out a theoretical analysis using a wave
propagation approach to demonstrate the coupling between the unsteady Kutta waves generated
at the airfoil trailing edge and the shock oscillation. The unsteady shock motion occurs at a discrete
frequency that can excite the structural vibration modes resulting in limit cycle flutter for an elastic
structure. This type of shock-induced excitation can be calculated theoretically using a viscous-
inviscid interactive method [26] or a solution of the Navier—Stokes equations [50]. For an
oscillating airfoil, the motion of the shock waves and the wake are illustrated in Fig. 17 for an 18%
circular arc airfoil studied in the NASA LaRC Transonic Dynamics Tunnel [12]. For a particular
test condition, small separation regions appear at the foot of the weak shock and at the trailing
edge on the lower surface. As the shock strength increases with increase in a at a latter part of the
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 219

Fig. 16. Model of self-sustained shock oscillations (from Ref. [67]).

Fig. 17. Sketch of transonic shock-boundary layer oscillation on circular-arc airfoil (from Ref. [12]).

oscillation cycle, these two regions merge and the upper surface shock wave weakens producing
a flow structure similar to the earlier one on the lower surface.
Aerodynamic nonlinearities associated with flow separation at low speed are found in dynamic
stall of helicopter blades or propeller blades. The events of dynamic stall are illustrated in Fig. 18
220 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 18. Dynamic stall from Navier—Stokes simulation showing streamlines, equi-vorticity lines and pressure distribution
at k "0.25 (from Ref. [79]).


taken from a numerical simulation using the Navier—Stokes equations [79]. At low angle-of-attack,
the flow is attached but leading-edge separation begins as a reaches a certain critical angle. The
leading-edge vortex moves downstream along the airfoil surface, and after traversing a certain
distance from the leading edge, lifts off from the surface and finally convects away from the trailing
edge. During the downstroke cycle, the flow reattaches on the upper surface starting from the
trailing edge and moves upstream while separation occurs on the lower surface following events on
the upper surface during the upstroke. The empirical method using wind tunnel data [32—34] is
widely used although Navier—Stokes solvers [122] are available at much higher computational
costs. The use of reliable NS solvers to analyze stall characteristics is of great help since time-
synthesization methods using experimental data have limited use and required large amounts of
loop data to cover a wide range of conditions, such as, angle of attack, Mach number, amplitude of
oscillations, frequency, etc. Also, the transition from attached to separated flow using the empirical
models is usually not accurately determined.
Another type of aerodynamic nonlinearities arises from the formation of wing tip vortices. This
was investigated by Strganac et al. [100] for a finite wing and the flow modeling is illustrated in
Fig. 19. The vortices oscillate with time and their strengths depend on the static angle of attack and
the amplitude of airfoil motion. This type of flow can readily be computed using an unsteady vortex
lattice method. The wing and wake are modeled as a vortex lattice. The position of the wing
portion, called the bound vortex lattice, is specified and there is a finite pressure jump across it. The
position of the wake portion, referred to as the free-vortex lattice, is not specified but is force-free
and is predicted as part of the solution. The aerodynamic loading is determined by calculating the
pressure jump across each individual element in the bound vortex. This model has been used to
study vortex dominated flows coupled to the equations of motion of the airfoil structure. However,
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 221

Fig. 19. Flowfield predicted by the unsteady vortex-lattice method (from Ref. [100]).

for very large amplitude motions when the roll-up vortices are formed alternatively on the upper
and lower surfaces of the wing, the phase lag with respect to the wing oscillation can be very large.
Very few studies on this topic have been carried out, although some investigations have been
conducted for delta wings at high incidence and pitch amplitudes.

3. Formulation of the nonlinear aeroelastic equations for airfoils

3.1. Equations with structural nonlinearities and subsonic aerodynamics

A simple and efficient method to investigate the dynamic response of a wing structure is to use
the concept of generalized coordinates [77]. Consider a cartesian coordinate system x, y and z fixed
on a thin wing as illustrated in Fig. 20. The displacement of the wing can be expressed in terms of
a set of generalized coordinates q (t) as
G
'
z(x, y, t)" (x, y)q (t) (1)
G G
G
where (x, y) is the mode shape of the ith mode, and I is the number of modes required to
G
adequately represent z(x, y, t) in the form of a series. I may vary from two to three for a simple
beam structure to a few dozens for complicated aircraft configurations. The dynamic aeroelastic
equations governing the response of q (t) to external forces is given in generalized coordinates
G
M q( #C qR #K q "F (2)
G G G G G G G
for the ith mode. The dots denote differentiation with respect to time, and M , C , K and F are the
G G G G
generalized mass, damping, stiffness and force, respectively. The fact that the equation does not
222 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 20. Schematic of a wing showing the coordinate system.

contain any other general coordinates q is due to the characteristic ‘normal’ property of the natural
G
modes. Actually, a set of terms ' O C qR should be included on the left-hand side of Eq. (2)
H G GH H
since this ‘normal’ property does not always apply to the system of damping forces acting on the
wing, but these modal coupling forces are usually small enough to be ignored.
The generalized mass M has the same magnitude of a mass which when moving with the velocity
G
qR has the same kinetic energy as the whole system moving with the velocity qR . It can be written
G G G
as

M"
G  m G dA , (3)

where m is the mass per unit area of the wing, and the integration is taken over the whole wing
surface area A. The generalized stiffness K is the stiffness of the spring which, when displaced from
G
its unstrained position by q , has the same potential energy as the actual system when displaced by
G
q . If we approximate the wing by a beam of length equal to, say, the span s, we can write K as
G G G
follows:

  
d 
K " EI G dx , (4)
G dx
Q
where EI is the flexural stiffness of the beam and " (x) in this case. There are other ways to
G G
represent the stiffness, for example, using a lumped parameter structural idealization.
The generalized damping coefficient C can be considered as the rate of the damper which, when
G
moved at a velocity qR dissipates energy at the same rate as the whole system of damping forces and
G
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 223

pressures acting on and within the system when moving with a velocity of qR . We can express
G G
C in terms of a local viscous damping pressure per unit velocity p (x, y) in counter phase with the
G BG
velocity qR as follows:
G G

C"
G O pBG  dA .
G
(5)

Finally, the generalized force F is a single force which, when moved through a displacement dq ,
G G
does the same amount of work as all the external forces and pressures acting on the system after
displaced by a distance dq . In terms of the pressure p(x, y, t) acting at a point (x, y), it can be
G G
written as

F (t)"
G  p(x, y, t) G
dA . (6)

In matrix formulation, Eq. (2) can be written as

[M] +q̈,#[C] +q ,#[K] +q,"+F, , (7)
where [M], [C] and [K] are the mass, damping and stiffness matrices. +F, is the aerodynamic force
vector. For a nonlinear structure, we replace the stiffness term [K] +q, by a nonlinear term +Q(q),.
Eq. (7) is useful when we couple a three-dimensional wing motion to an aerodynamics code and
use a time integration technique to solve the structural motion. The system we shall investigate in
detail in later sections is a two-dimensional airfoil oscillating in pitch and in plunge. We shall
express the dynamic equations of motion in a slightly different form which will be useful in later
sections.
Fig. 21 gives the symbols used in the analysis of a two-degree-of-freedom airfoil motion. The
plunge deflection is denoted by h, positive in the downward direction, and a is the pitch angle about
the elastic axis, positive nose up. The elastic axis is located at a distance a b from the midchord,
F
while the mass centre is located at a distance x b from the elastic axis. Both distances are positive
?
when measured towards the trailing edge of the airfoil. The aeroelastic equations of motion for
linear springs have been derived by Fung [31]. For nonlinear restoring forces such as those
considered in Section 2.1, the coupled bending-torsion equations for the airfoil can be written as
follows:
mh® #Sa( #C h #GM (h)"p(t) , (8)
F
Sh® #I a( #C aR #M M (a)"r(t) , (9)
? ?
where the symbols m, S, C , I and C are the airfoil mass, airfoil static moment about the elastic
F ? ?
axis, damping coefficient in plunge, wing mass moment of inertia about elastic axis, and torsion
damping constant, respectively. GM (h) and MM (a) are the nonlinear plunge and pitch stiffness terms,
and p(t) and r(t) are the forces and moments acting on the airfoil, respectively.
In Section 2.1 we have discussed the various types of structural nonlinearities. In this review, we
shall consider only the three common ones, namely, cubic, freeplay and hysteresis nonlinearities.
Consider the torsion degree-of-freedom nonlinearity for the time being, we use the same repres-
entation of a cubic spring as that of Woolston et al. [120,121] and Lee and LeBlanc [64]. We write
224 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 21. Two-degree-of-freedom airfoil motion.

MM (a)"K (a#b a), where K is the linear spring constant and b is a constant less than unity for
? ? ? ?
the type of problems they investigated. For the freeplay and hysteresis, MM (a)"K a, and K is
? ?
specified as a piecewise function of a. For the plunge degree-of-freedom, we can write down similar
expressions for GM (h) simply by replacing a with h. We shall discuss these functions in greater detail
in Section 5.
Define m"h/b, K "K , x "S/bm, u "(K /m), u "(K /I ), r "(I /mb), f "C /
K F ? K K ? ? ? ? @ K F
2(mK ), f "C /2(I K ). Eqs. (8) and (9) can be written in non-dimensional form as [65]
F ? ? ? ?


uN uN  1 P(q)b
m#x a#21 m# G(m)"! C (q)# , (10)
? K º* º* nk * mº

x 1 1 2 Q(q)
? m#a#2 ? a# M(a)" C (q)# , (11)
r º* º* nkr + mºr
? ? ?
where G(m)"GM (h)/K and M(a)"M M (a)/K .
K ?
In Eqs. (10) and (11), º* is a non-dimensional velocity defined as

º
º*" (12)
bu
?
and uN is given by

u
uN " K , (13)
u
?
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 225

where u and u are the uncoupled plunging and pitching modes natural frequencies, and the
K ?
 denotes differentiation with respect to the nondimensional time q defined as

ºt
q" (14)
b

C (q) and C (q) in Eqs. (10) and (11) are the lift and pitching moment coefficients, respectively. For
* +
incompressible flow, Fung [31] gives the following expressions for C (q) and C (q):
* +
C (q)"n(m!a a#a)#2n +a(0)#m(0)#[!a ]a(0), (q)
* F  F
O
#2n

 (q!p) [a(p)#m(p)#(!a )a(p)] dp ,
 F
(15)

C (q)"n(#a ) +a(0)#m(0)#[!a ]a(0), (q)

+  F  F
O
#n(#a )
 F


(q!p) +a(p)#m(p)#(!a )a(p), dp
 F

n n n
# a (m!a a)!(!a ) a! a , (16)
2 F F  F 2 16

where the Wagner function (q) is given by

(q)"1!t e\eq!t e\eq (17)

 
and the constants t "0.165, t "0.335, e "0.0455 and e "0.3 are obtained from Jones [53].
   
P(q) and Q(q) are the externally applied forces and moments, respectively.
For unforced oscillations, Eqs. (10) and (11) were solved by Lee and Desrochers [65] and Price
et al. [88,89] using a finite difference scheme which will be discussed in Section 4.1. Due to the
existence of the integral terms in the integro-differential equations (10) and (11), it is cumbersome to
integrate them numerically. A simpler set of equations was derived by Lee et al. [69], and they
introduced four new variables

O O
w "
  e !e (t!p)
 a(p) dp, w "
  e !e (t!p)
 a(p) dp ,
(18)
O O
w "
  e !e (t!p)
 m(p) dp, w "
  e !e (t!p)
 m(p) dp .

The equations given by Lee et al. [69] can be written in general form for cubic, freeplay or
hysteresis nonlinearities as

c m#c a#c m#c a#c m#c a#c w #c w #c w #c w

             


uN 
# G(m)"f (q) , (19)
º*
226 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

d m#d a#d a#d a#d m#d m#d w #d w #d w #d w

             


1 
# M(a)"g(q) (20)
º*

where G(m) and M(a) are nonlinear functions of m and a, respectively. The coefficients c ,

c ,2, d ,2, d are given in Appendix A. f (q) and g(q) are functions of initial conditions and terms
  
in the Wagner function. In the absence of external forcing, i.e. P(q)"Q(q)"0, they are given by

 
2 1
f (q)" !a a(0)#m(0) (t e e!eq#t e e!eq ) , (21)
k 2 F  

(1#2a ) f (q)
g(q)"! F . (22)
2r
?
If P(q) and Q(q)O0, f (q) and g(q) are given in Eqs. (108) and (109). After introducing a variable
vector X"(x , x ,2, x )2 defined as
  
x "a, x "a, x "m, x "m, x "w , x "w , x "w , x "w , (23)
           
Eqs. (19) and (20) can be written as a set of eight first-order ordinary differential equations,

X "f (X) . (24)

In terms of vector components, Eq. (24) can be expressed as

x "x , x "(c H!d P)/(d c !c d ), x "x , x "(!c H#d P)/(d c !c d) ,

                
x "x !e x , x "x !e x , x "x !e x , x "x !e x , (25)
               
where


uN 
P"c x #c x #c x #c x #c x #c x #c x #c x # G(x )!f (q) ,
                º* 
(26)


1 
H"d x #d x #d x #d x #d x #d x #d x #d x # M(x )!g(q) .
                º* 

Eq. (24) can be integrated numerically once the initial conditions a(0), a(0), m(0), m(0) are given.
These equations are easier to solve than Eqs. (10) and (11), especially when we investigate the
system behavior analytically.
Assuming the existence of the third and fourth derivatives, Alighanbari and Price [1] expressed
Eqs. (10) and (11) as a set of fourth-order differential equations. Eight initial conditions are required
in this approach since a fourth-order solution requires a(0), a(0), m(0) and m(0) to be speci-
fied in addition to a(0), a(0), m(0) and m(0) for a set of second-order differential equations.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 227

According to Alighanbari and Price’s [1] method, both sides of Eqs. (10) and (19) are differenti-
ated twice and the following two equations are obtained:
m m'4#m a'4#m m#m a#m m#m a#m m#m a
       


uN 
#m m#m a# [N (m)]"0 (27)
  º* K

n m'4#n a'4#n m#n a#n m#n a#n m#n a

       


1 
#n m#n a# [N (a)]"0 . (28)
  º* ?

The coefficients m and n in Eqs. (27) and (28) are given in the appendix. N (m) and N (a) are given
G G K ?
as
N (m)"G(m)#(e #e )G(m)#e e G(m) , (29)
K   
N (a)"M(a)#(e #e )M(a)#e e M(a) . (30)
?   
The nonlinear functions G(m) and M(a) given in Eqs. (19) and (20) for the three types of
concentrated structural nonlinearities we shall discuss in details can be written as follows using the
symbols defined in Figs. 2 and 11:

Cubic spring:
M(a)"b #b a#b a#b a , (31)
   
where b , b , b , and b are constants.
   
Bilinear spring:


M #a!a for a(a ,
 D D
M(a)" M #M (a!a ) for a 4a4a #d , (32)
 D D D D
M #a!a #d(M !1) for a #d(a .
 D D D
Hysteresis:


a!a #M for a(a ; a increasing ,
D  D
a#a !M for a'!a ; a decreasing ,
D  D
M for a 4a4a #d; a increasing ,
M(a)"  D D (33)
!M for !a 4a4!a !d; a decreasing ,
 D D
a!a !d#M for a'a #d; a increasing ,
D  D
a#a #d!M for a(!a !d; a decreasing .
D  D
Here we give the expressions for M(a) in the pitch degree of freedom. Similar expressions for G(m)
for the plunge motion can be written by replacing a with m. Note that M(a) does not exist if M(a) is
228 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

defined by Eq. (32) for the bilinear spring. To overcome this difficulty, M(a) can be expressed as an
analytical function given in Eq. (167).

3.2. Equations with nonlinear aerodynamics

3.2.1. Transonic flows

A number of mathematical models are available to compute nonlinear unsteady transonic
aerodynamic loads. The article by Edwards and Malone [25] reviews recent developments in the
field of computational methods for unsteady transonic aerodynamics with aeroelastic applications.
The first computational model, the LTRAN2 code, was proposed by Ballhaus and Goorjian [5] in
which the alternating direction implicit (ADI) algorithm was used for the solution of the two-
dimensional low-frequency transonic small disturbance (TSD) equation. The three-dimensional
TSD equation in Cartesian coordinates can be expressed in the following form [9]:
jf jf jf jf
# # # "0 , (34)
jt jx jy jz
where x, y and z denote the non-dimensional physical coordinates in the streamwise, spanwise and
vertical directions, respectively, and
f "!A !B ,
 R V
f "E #F #G ,
  V  V  W (35)
f " #H ,
 W  V W
f " .
 X
Here is the velocity potential and the coefficients in Eq. (35) are defined as
A"M, B"2M, E "1!M ,

F "![3!(2!c)M]M/2, G "!M/2, H "!M , (36)
  
where c is the specific heat ratio. The TSD formulation has been widely used to compute unsteady
transonic flows about isolated wings and complete aircraft configurations. The XTRAN3S [17],
ATRAN3S [40], and CAP-TSD [8] computer simulation codes are essentially based on the TSD
formulation given in Eq. (34), and a time integration numerical procedure is used in conjunction
with an ADI or approximate factorization (AF) algorithm. A time linearization procedure for
solving the TSD equation has also been applied by Wong and Lee [118] to develop the UST3D
code, in which the velocity potential is expressed as
(x, y, z, t)" Q(x, y, z)#e S(x, y, z, t) and e;1 , (37)
where the steady component Q satisfies the nonlinear, mixed elliptic-hyperbolic differential
equation, and the unsteady component S is calculated from the linear time-dependent differential
equation whose coefficients are determined from the steady solution of Q. Generally speaking, in
the TSD formulation, we assume that the flow is isentropic, irrotational and a small perturbation of
the steady uniform flow in the x-direction. The TSD theory, however, has been modified to include
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 229

entropy and vorticity effects, so that it can be used to deal with flowfields with stronger shock
waves [9]. For some test cases, it was found that numerical simulations from the TSD model are
almost as accurate as those obtained by the Euler equations.
A second formulation for unsteady transonic calculations is based on the full potential (FP)
model. The FP model also assumes the flow to be inviscid and the shock wave to be weak. Let
U denote the velocity potential, then the governing equation for the FP formulation in conserva-
tion form is given by
o #(oU ) #(oU ) #(oU ) "0 , (38)
R VV WW XX
where the density o satisfies the unsteady Bernoulli equation:


c#1 A\
o" 1# M(I!U!U!U!2U . (39)
2 V W X R
The FTRAN3 code developed by Hounjet [47] and the USTF3 program by Iosgai and Suetsugu
[49] are based on the FP formulation. In using the FP equation for transonic flows, Steinhoff and
Jameson [98] have shown that the equation can admit non-unique solutions due to the entropy
generation within the shocks violating the isentropic assumption. To circumvent the non-unique
solution, Osher et al. [83] and Whitlow et al. [116] proposed an entropy correction method that
satisfies the approximations for the FP equation. The method is used to rule out the non-physical
expansion shocks and to accurately track the sonic conditions.
Until a decade ago, unsteady transonic flow computations for aeroelastic problems were mostly
carried out using the TSD and FP formulations. Numerical methods for solving unsteady Euler
and Navier—Stokes (NS) equations were available, but they have not been widely used in practice
because these procedures are very costly. However, with the recent advances in computer techno-
logy, numerical simulation codes using these accurate formulations have been developed for
aeroelastic analysis because of their ability to model complex flow phenomena, such as strong
shocks, vortices and flow separations.
The NS equations in a Cartesian coordinate system can be expressed as follows:
j j j j
(Q)# (F!F )# (G!G )# (H!H )"0 , (40)
jt jx  jy  jz 
where the conserved variables Q and the inviscid flux vectors F, G, H are defined by





o ou ov ow
ou ou#p ouv ouw
Q" ov , F" ouv , G" ov#p , H" ovw . (41)
ow ouw ovw ow#p
oe oeu#pu oev#pv oew#pw
The total energy is denoted by e, where

 
p u#v#w
e" # . (42)
(c!1)o 2
230 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

The corresponding viscous flux vectors are denoted by F , G , H and they are given as follows:
  



0 0
q q
VV VW
F" q , G" q ,
 VW  WW
q q
VX WX
uq #vq #wq !q uq #vq #wq !q
VV VW VX V VW WW WX W
(43)


0
q
VX
H" q .
 WX
q
XX
uq #vq #wq !q
VX WX XX X
The components of the shear stress tensor and the heat flux vector are defined by




2 ju jl jw 2 jv ju jw 2 jw ju jv
q " t 2 ! ! , q " t 2 ! ! , q " t 2 ! ! ,
VV 3 jx jy dz WW 3 jy jx dz XX 3 dz jx jy




ju jl jw ju jv jw
q "t 2 ! "q , q "t 2 ! "q , q "t 2 ! "q ,
VW jy jx WV VX jx jz XV WX jz jy XW

j¹ j¹ j¹
q "!k , q "!k , q "!k . (44)
V jx W jy X jz

where ¹ is the temperature and t, k denote the coefficient of the viscosity and the thermal
conductivity, respectively. In Eq. (41), p is the pressure and u, v, and w are the velocity components
in the x, y, and z directions, respectively. For viscous flows with mild separation, the NS
formulation can be simplified to the thin-layer NS equations. A further simplification can be made
when the viscous effects are small and can be neglected. The Euler equations can be obtained from
Eq. (40) by setting the viscous flux vectors F , G , H to zero. Unlike the TSD and FP formulations
  
where the governing equation is a single equation, the Euler and NS formulations lead to a system
of five equations in three-dimensional problems. Consequently, the resulting discrete approxima-
tions from the Euler and NS will be five times larger than those obtained by TSD and FP. As
a result, a significant increase in computing power is required to deal with simulations based on
these equations.
To analyse aeroelastic problems, the normal procedure is to solve the structural and aerodyna-
mic equations simultaneously. The methodology is quite involved and depends on the particular
approach used to solve the aerodynamic equations. We consider in this review only a simple
example useful for solving the coupled TSD equations following the approach given by Borland
and Rizzetta [16]. In solving the unsteady equations, we assume the wing section to be thin. The
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 231

boundary conditions for the flowfield are: "0 far upstream; # "0 far downstream; "0
V  X
far above and below the wing; "0 far from the wing tip and at the wing root; * "0 and
W X
*( # )"0 across the trailing edge vortex sheet in the wake region defined by z"0 for
V R
x'x , and * indicates a jump across the wake. The initial conditions (x, y, z, 0)"g(x, y, z) and
2#
(x, y, z, 0)"h(x, y, z) are specified for the particular problem to be considered.
R
On the wing surface,
$
"S$#S$ for x 4x4x , y4y , z"0$ , (45)
X V R *# 2# 2'.
where the superscript $ refers to the airfoil upper and lower surface, and S(x, y, t) denotes the wing
shape. The solution to the system of structural equations of motion is coupled to the aerodynamic
solution through the boundary conditions on the wing surface. We first determine the actual
time-dependent deformed shape and motion as the sum of contributions from the static rigid, static
flexible, dynamic rigid, and dynamic flexible components of the structure. The pressure distribution
and integrated forces on this deformed shape are then calculated in a time dependent fashion. The
response is determined from these computed forces using the structural equations of motion. The
wing surface boundary conditions for static aeroelastic analysis are given by


x 4x4x ,
+ $
,"+S$0'%'" ,#+S$$*#6' *# ,, *# 2# (46)
X V V y4y ,
2'.
where S 0'%'" can contain contributions from the local slope of the airfoil section coordinate
$
V
definition on the upper and lower surface z$ , the geometrical angle of attack, the static rigid twist

distribution, and static control surface deflections. The other component S$$*#6' *# , which is the same
V
for the upper and lower surface, is defined as

+S$$*#6' *#,"[U ] +q, , (47)

V 
where [U ] is the matrix of streamwise slopes of the natural vibration modes given at the

aerodynamic control points, and +q, is the matrix of static values of the generalized coordinates.
For dynamic aeroelastic analysis, the wing surface boundary conditions are given by

+ $
,"+S$0'%'",STEADY#+(S$#S$)RIGID ,UNSTEADY#+(S$#S$)FLEXIBLE,UNSTEADY . (48)
X V V R V R
The first term is identical to the corresponding term for static aeroelastic analysis in Eq. (46). The
second term accounts for rigid-body motion such as pitch and plunge, and the last term deals with
the time-dependent elastic motion. It can be written as
1
+(S$#S$)FLEXIBLE,UNSTEADY "[U ] +q,# [U ] +qR , . (49)
V R  º 
The generalized coordinates +q, are determined from the equation of motion given by Eq. (7). Here,
we shall consider the equation for a linear structure and the procedure can be extended to
a nonlinear one.
For a static solution, Eq. (7) can be written as
[K]+q,"+F, . (50)
232 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Once the static solution of the generalized coordinates is determined, Eq. (47) is used to recompute
the flowfield, and the process is repeated until the solution converges.
In using Eq. (7) to solve the dynamic response, Borland and Rizetta [16] adopted a simple
central difference scheme with a constant time step *¹. The first and second time derivatives of the
generalized coordinate are written as

 
qL>!qL\
+q ," , (51)
2*¹

 
qL>!2qL#qL\
+q̈," , (52)
(*¹)

where n denotes the time step. Substituting Eqs. (51) and (52) into the structural equations of
motion, we have a simple implicit relation for the generalized coordinates at the (n#1)th time
given by the following:



 
1  1 \
+qL>," M# C +FL,
*¹ 2*¹



  
 
1  1 \ 1 
# M# C 2 M!K +qL,
*¹ 2*¹ *¹



  

 
1  1 \ 1 1 
# M# C C! M +qL\, (53)
*¹ 2*¹ 2*¹ *¹

The time-marching scheme can proceed once the initial conditions, such as, displacement and
velocity of the generalized coordinates are specified.
For higher order aerodynamics codes, the coupling with the structural motion can be handled
similarly but the procedure is more complex. Guruswamy [41] developed a computer code
ENSAERO to calculate unsteady aerodynamics loads, and he coupled the aerodynamic computa-
tions with the linear structural equations for aeroelastic applications. The ENSAERO is based on
Euler/NS equations with an algebraic model to compute the turbulence effects. Another computer
program, ENS3DAE, which is also based on Euler/NS formulation has been developed by
Schuster et al. [93]. The ENS3DAE is coupled with a linear structural optimization program
ASTROS (Automated STRuctural Optimization System) to compute aeroelastic responses [82].
The various computational models based on TSD, FP, Euler and Navier—Stokes formulations
have been successfully used to investigate the aeroelastic response of isolated wings and complete
aircraft configurations.

3.2.2. Dynamic stall

An aerodynamic nonlinearity with considerable potential for producing interesting dynamic
behavior, and possibly chaotic motion is dynamic stall. There are many practical aeronautical
applications where airfoils are required to operate under stalled conditions; including aircraft
operating at high angles of attack and helicopter blades in the retreating part of the cycle.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 233

Dynamic stall of an airfoil, with its associated flow separation and reattachment, is an extremely
complex phenomenon, and a detailed explanation of the dynamic stall process is not attempted
here. The interested reader is referred to Ericsson and Reding [29,30], where a detailed discussion
of the physics of dynamic stall is given. Instead, a semi-empirical model is described in this review
since this type of formulation of the nonlinear aerodynamics has been used by a number
of researchers in their investigations of airfoil dynamics under stall conditions. There are a number
of different empirical models of dynamic stall available, for example: the quasi-steady model of
Ericsson and Reding [28]; the Tran and Petot [109] model developed at the Office National
d’Ëtudes et de Recherches Aérospatiale, ONERA; the Beddoes [10] model, later extended by
Leishman and Beddoes [73]; or the Gangwani [32,34] and Bielawa et al. [15] model developed at
the National Aeronautics and Space Administration, NASA. Some of these models are reviewed by
Reddy and Kaza [92].
The semi-empirical ONERA model [109] assumes that the aerodynamic forces can be written as
a function of the variables describing the motion of the airfoil. These functions are obtained by
curve fitting expressions to available experimental data. Although the results of this model give
good agreement with experimental data for harmonically oscillating airfoils, no attempt is made in
this approach to model the effects of any of the physical flow phenomena associated with dynamic
stall. As shown by Tang and Dowell [102] when the flow over the airfoil remains attached, the
ONERA model is equivalent to Theodorsen [107] unsteady aerodynamics. Recently, Tang and
Dowell [104] extended the ONERA model, and employed it for the analysis of a low aspect ratio
three-dimensional wing. Good comparison between theory and experimental results was obtained
for the case of pure pitch motion. A number of different authors have made use of the ONERA
model to investigate the aeroelastic response of stalled airfoils. For example, Tran and Falchero
[110] used it to investigate the response of a helicopter blade in forward flight. Dunn and Dugundji
[24] also employed the ONERA model in their investigation of the nonlinear aeroelastic behavior
of rectangular graphite/epoxy cantilevered wings when subject to dynamic stall. Also, Tang and
Dowell [101—106] have made extensive use of the ONERA model to investigate the aeroelastic
response of stalled non-rotating helicopter blades.
The Beddoes [10] and Leishman and Beddoes [73] model of dynamic stall uses an indicial
response and superposition method, incorporating corrections to account for the effect of com-
pressibility. The nonlinear effects of trailing edge separation are implemented using Kirchhoff
theory, enabling the force and moment characteristics to be related to the trailing edge separation
point. Extensive validation of this model is given by Leishman and Beddoes [73] for a NACA 0012
airfoil.
The method described in this review is taken from Price and Keleris [90] who used a model
based on the work of Gangwani [32,34] and Bielawa et al. [15]. This model has the advantage that
it is formulated in the time domain, and the aerodynamic forces can be calculated easily from the
pitch and pitch-rate of the airfoil. However, one drawback of this model is that it accounts for the
pitch motion only; thus, it is restricted to cases where there is no heave.

3.2.2.1. The semi-empirical dynamic stall model. Dynamic stall occurs after the angle of attack of an
airfoil, a, has exceeded the static stall angle, a , and when the leading edge vortex breaks away from

the airfoil. In order to accurately model dynamic stall it is important to be able to predict the onset
of three major events: the formation and shedding of a leading edge vortex, the arrival of this vortex
234 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

at the trailing edge, and the reattachment of flow over the airfoil. The model described in this
review, developed by Gangwani [32] and Bielawa et al. [15] is illustrated in Fig. 18, and involves
several analytical expressions, which are simple mathematical models of these main dynamic stall
events. These expressions include numerous unknown coefficients, which are determined by curve
fitting data from wind tunnel tests on oscillating airfoils. Since this model gives the aerodynamics in
the time domain, the main parameters required to predict the pitching moment C are a and the
+
pitch-rate, both of which are simple to define physically and mathematically. Furthermore, the
model does not require the frequency of oscillation to be known, which is advantageous for
the present analysis because of the difficulty of defining the instantaneous frequency of an airfoil
undergoing quasi-periodic or chaotic motion.
In order to account for the time history of the airfoil motion, an effective angle of attack,
obtained using Duhamel’s integral, is defined as

O ja
a "a(0) (q, M)#
# !  
jp !
(q!p, M) dp , (54)

where a(0) is the initial angle of attack, and is Wagner’s function corrected for compressibility
!
effects [10]. As suggested by Beddoes [10], is given approximately by
!
(q, M)"(1!0.165 exp (!0.0455b)q!0.335 exp (!0.3b) q)/b , (55)
!
where b"(1!M. The difference between the geometric angle of attack, a, and a is defined as
#
the unsteady decay parameter, a . This interpretation of a is strictly correct only when the flow is
5 5
attached. However, as suggested by Gangwani [32] it is also used to predict certain dynamic stall
events, and to approximate the aerodynamic loads even when the airfoil is stalled.
To predict the instantaneous angle of attack at which dynamic stall occurs, a a semi-empirical
"
relationship is assumed between a and the static stall angle a , the pitch-rate, and a . Linearizing
" 1 #
this expression about a gives
1
a "(1#e#C A #C a ) a , (56)
"  "  5+ 1
where A and a are the airfoil pitch-rate and the unsteady decay parameter, respec-
" 5+
tively, evaluated at the point of dynamic stall, and e, C and C are empirically determined
 
constants.
After dynamic stall occurs the leading edge vortex is convected downstream over the upper
surface of the airfoil, and its strength and distance from the airfoil strongly influence the aerody-
namic loads. Its main effect is to increase the negative (nose down) C , which reaches a maximum
+
when the vortex arrives at the trailing edge. A semi-empirical relationship is used to predict the
nondimensional time required for the vortex to travel from the leading to the trailing edges of the
airfoil, q , given by
2
1.0
q " , (57)
2 C A #C a
 "  "
where C and C are empirically determined constants.
 
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 235

The instantaneous angle of attack at which the flow reattaches, a is obtained using the same
0
method as for predicting a , and an appropriate expression is
"
a "(1!e#C A #C a ) a , (58)
0  "  5+ 1
where e is the same as in Eq. (56), and C and C are new constants.
 
The aerodynamic pitching moment, C , is then approximated as
+
C "C (a!*a )#a *a #*C , (59)
+ +1    +
where


a
*C "P A#P a #P #P "a "#P d #P *a #P a A q , (60)
+   5  a  5      " " +
1
*a "d a , (61)
  1
º (t!t )
q " " , (62)
+ b


0, for a4a ,
1


a
!1 , for a (a4a ,
a 1 "
d " 1 (63)




 
a q 
"!1 1! + , for 04q 4q ,
a q + 2
1 2
0, for q 'q .
+ "


0, for a4a ,
1


a
!1 , for a (a4a ,
a 1 "
1


a
d " "!1 , for 04q 4q , (64)
 a + 2
1


a a!a
"!1 0 , for a 4a4a ,
a a !a 0 2
1 2 0
0, for a(a .
0

The parameter a is the static moment coefficient curve slope at zero angle of attack.

As shown in Eq. (59), C is modeled as the sum of the static moment coefficient, C , at a shifted
+ +1
angle of attack, a!*a , plus the static moment curve slope, a , multiplied by the incremental
 
236 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

angle of attack, *a , plus an incremental moment coefficient, *C . The term *a , which is

‚ + 
non-zero for stalled flow only, accounts for dynamic stall and flow reattachment, a is the value of
2
a when the vortex reaches the trailing edge, and t is the time at which dynamic stall occurs. The
"
term *C accounts for the effect of the shed leading edge vortex, and the constants, P 2 P , are
+  
determined by curve fitting wind tunnel data to the above expressions using a least-squares
technique.

3.2.2.2. Modification and verification of the dynamic stall model. Price and Keleris [90] made two
simple modifications to the dynamic stall model described above. The first modification was to
restrict the domains of a and a such that they are greater and less than a , respectively. It can be
" 0 1
argued that a should always be greater than a due to the boundary layer improvements as a is
" 1
increasing [29]. Similarly, it can be argued that a must always be less than a since dynamic effects
0 1
are present as a is decreasing, and they will destabilize the boundary layer and delay reattachment
of the flow. However, by introducing these assumptions the model is restricted to cases where stall
is not induced by local shocks, i.e. approximately M must be 40.6.
The second modification was made to ensure that C remain piece-wise continuous. Often the
+
model predicted that the shed vortex did not reach the trailing edge before a had decreased below
a . In other words, the two conditions, q (q and a(a , existed simultaneously, which results in
0 + 2 0
discontinuous functions for C and d and (see Eqs. (59) and (64)). To avoid this, the expressions for
+ 
d and *C were modified for the cases where a(a and q (q , to give
 + 0 + 2


a
d " "!1 [1!exp (K (q !q ))] . (65)
 a  + 2
1
a
*C "P A#P a #P #P "a "#P d #P *a
+   5 a  5    
1 (66)
#P a A q [1!exp (K (q !q ))] ,
 " " +  + 2
where K "0.3 is a new constant determined by comparing the modified expressions with wind

tunnel data. The modified terms are those which are supposed to be zero when the vortex detaches
from the leading edge and reach a certain value as the flow reattaches after the vortex has been
shed. These terms are multiplied by [1!exp (K (q !q ))], which ensures that they tend to
 + 2
appropriate values as q becomes increasingly larger than q .
+ 2
Eighteen sets of data measured by Gray and Liiva [38] for a NACA 0012 airfoil at a Mach
number M"0.6 and Reynolds number Re"6.2;10 were used to obtain the empirical constants.
The nondimensional frequency of forced oscillation (k"ub/º) was in the range
0.0444k40.256, the amplitude of oscillation was 2.5°4a47.5°, and the mean angle of attack
varied from 0.0°4a 410.0°, thus, ensuring that the model would be valid over a wide range of

parameters. The test data were curve fit to the semi-analytical expressions for C using the
+
‘‘Amoeba’’ subroutine from Press et al. [87] and a least-squares method. The coefficients obtained
are: P "!1.757, P "!0.08255, P "0.006019, P "!0.06206, P "0.04894, P "!0.7773,
     
P "!5.964, C "54.93, C "10.52, C "5.420, C "0.4884, C "16.65, C "!8.669 and
      
e"!0.0139.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 237

Fig. 22. Comparison of the actual moment coefficient data obtained from Gray and Liiva [38], —, with the moment
coefficient predicted by the dynamic stall Bielawa et al. [15] model 2, Re"6.2;10, M"0.6. (a) k"0.045, a "7.39°,

a"4.94°; (b) k"0.165, a "0.20°, aN "5.49°; and (c) k"0.129, a "7.62°, a"5.28° (from Ref. [90]).
 

Some typical comparisons between measured C and the values predicted by the modified
+
dynamic stall model are shown in Fig. 22 for three different non-dimensional frequencies, ampli-
tudes and mean angles of attack. These comparisons show neither the best nor worst agreement
between the experimental data and the dynamic stall model (a complete comparison for all of the
experimental data is given by Keleris [54]). As shown in Fig. 22, although the comparison is far
from excellent in terms of the magnitude of C , the model does give a reasonable representation of
+
the main events occurring in the dynamic stall process.

3.2.2.3. Solution of the aeroelastic equations. Price and Keleris [90] studied a rigid NACA 0012
airfoil, flexibly mounted in pitch in subsonic flow. No heave motion is allowed, which corresponds
to a system much stiffer in heave than in pitch. The oscillations are not self excited, an externally
applied sinusoidal torque continuously drives the system, and it is the response of the airfoil to this
applied torque which they examined. However, the results obtained are potentially of some
significance to those interested in self excited flutter. For example, the results indicate the type of
complexities which can be obtained in the dynamics of an airfoil executing two-degree-of-freedom
flutter when large pitch amplitude oscillations occur. Furthermore, the results presented are
directly applicable to an airfoil executing single degree-of-freedom flutter in pitch.
Considering an airfoil with freedom to oscillate in pitch only, then the equation of motion is
given from Eq. (11) as


f 1 2 Q(q)
r a#2 ? a# a " C (q)# , (67)
? º* º* nk + mº
238 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

where C (q) is the aerodynamic moment coefficient taken about the elastic axis, and Q(q) the
+
externally applied torque about the elastic axis which drives the system.
To solve the aeroelastic equations, Houbolt’s [46] implicit finite difference method was used,
giving a at the nth time step as follows [90]:

a "Z /Z , (68)
L  
where




a 1 5 3 a
Z "C  #Q #a # 
 +L a L a L\ (*q) *q a
   (69)



4 3 a 1 1 a
!a #  #a #  ,
L\ (*q) 2*q a L\ (*q) 3*q a
 
2 11 a a
Z " # #  , (70)
 (*q) 6*q a a
 
and

2r · r 2
a "r , a " ? ? , a " ? , a " . (71)
 ?  º*  º*  nk

The following recursive relationships were used to obtain the pitch rate, A , and the unsteady decay
L
parameter, (a ) , at each time step [15]:
5
(*a)
A" L, (72)
L (*q)
L
and

(a ) "x #y , (73)
5L L L
where

x "x exp (!0.0455 (1!M) (*q) )#0.165 (a !a ), (74)

L L\ L L L\
y "y exp (!0.3 (1!M) (*q) )#0.355 (a !a ). (75)
L L\ L L L\
Houbolt’s method requires C at any time step, n, in order to calculate a at the same time step.
+L L
However, because C depends on a at the same time, a recursive predictor-corrector procedure
+L L
must be implemented. First, C is calculated at time step n based on a . Next, this value of C is
+L L\ +L
used to calculate the predictor value of a , and based on this C is recalculated. Finally, the
L +L
recalculated C is used to find the corrector value of a . If the absolute difference between the
+L L
predictor and the corrector values is within a specified tolerance (approximately 1.0;10\), then
the corrector value is accepted. If the absolute difference is greater than the tolerance the entire
procedure is halted. Convergence of this corrector-predictor method was investigated extensively,
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 239

as was the convergence of the whole numerical procedure, and is discussed in some detail by
Keleris [54]. In all cases it was ensured that the time step was sufficiently small such that a further
reduction had no effect on the amplitude of motion, when the motion was periodic, or on the
character of the attractor when it was chaotic.

4. Solution techniques of equations with structural nonlinearities

4.1. Finite difference scheme

To solve Eqs. (10) and (11), Houbolt’s [46] finite difference method can be used. This has been
shown to be more efficient than higher order finite difference schemes [52], yet still have good
accuracy. The derivatives at time q#*q are replaced by backward difference formulas using values
at three previous times. For example
a(q#*q)"[2a(q#*q)!5a(q)#4a(q!*q)!a(q!2*q)]/*q , (76)
a(q#*q)"[11a(q#*q)!18a(q)#9a(q!*q)!2a(q!2*q)]/6*q , (77)
and similar expressions can be written for m(q#*q) and m(q#*q). Hence, in difference form, Eqs.
(10) and (11) can be expressed, after considerable algebra [65], as
PM a(q#*q)#PM m(q#*q)"XM #¹ (m) , (78)
   K
PM a(q#*q)#PM m(q#*q)"XM #¹ (a) , (79)
   ?
where PM 2 PM are coefficients depending on the airfoil parameters and the constants in the
 
Wagner’s function. ¹ (a) and ¹ (m) contain functions of the nonlinear structures. These are long
? K
algebraic terms and are given in Lee and Desrochers [65], Lee and LeBlanc [64], and Price et al.
[88] for the freeplay and cubic nonlinearities.
As shown in Eqs. (15) and (16), the aerodynamic forces and moments depend on two integrals:


O
I (q)" j(p) e\C O\N dp ,

 (80)


O
I (q)" j(p) e\C O\N dp ,


where
j(p)"a(p)#m(p)#(1/2!a ) a(p) . (81)
F
These integrals have to be evaluated at each time step, and in order to reduce the amount of
computations, Lee and Desrochers [65] derived a recurrence formula using Simpson’s rule and
obtained

 
*q 9j(q#*q)#19j(q) e\C O!5j(q!*q) e\C O
I (q#*q)"e\C O I (q)# (82)
  24 #j(q!2*q) e\C O
and a similar expression for I (j) after replacing e by e in the above equation.
  
240 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Houbolt’s finite difference scheme requires values of a and m at times q!2*q, q!*q and q in
order to determine the respective values at q#*q. Hence, at time q"0, a starting procedure is
required. Using Taylor’s series the following can be written:

*q
a(!*q)"a(0)!*qa(0)# a(0)#O(*q) , (83)
2

*q
a(*q)"a(0)#*qa(0)# a(0)#O(*q) , (84)
2

with similar expressions for a, a, m, m and m. The initial conditions a(0), a(0), m(0) and m(0) are
known and the higher derivatives up to fourth order are required. They can be obtained from Eqs.
(10) and (11) and are given by

   
aL(0) ½L
"[Q]\  , (85)
mL(0) ½ L

where Q , Q , Q , Q , ½L and ½L can be found in Lee and Desrochers [65], Lee and LeBlanc
    
[64] and Price et al. [88]. For the next step, Houbolt’s scheme can be used since a and m at
q"!*q, 0 and *q are known. The accuracy of the numerical scheme is 0(*q) at each step while
Eq. (83) and (84) limit the accuracy to 0 (*q). A starting accuracy higher than 0(*q) is not
necessary since Jones and Lee [52] showed that the error per cycle in the numerical scheme is
(2n/u) 0(*q). They also found that a step size of 1/256 of the period of the highest frequency
component yields sufficiently accurate results.

4.2. Runge—Kutta scheme

The aeroelastic equations given by Eq. (24) are formulated as a set of first-order ordinary
differential equations, and a number of numerical integration methods are available to solve this
initial-value problem. Lee et al. [70] used a fourth-order Runge—Kutta scheme to integrate the
system of equations for given initial conditions in their study of cubic restoring forces.
The airfoil motion from Eq. (24) is given by

dX
"f (X, q) . (86)
dq

Replacing the differentials dX and dq by finite increments *X and *q, Eq. (86) becomes

*X"f (X, q) *q . (87)

Denoting XL and XL> as eight-component vectors at time step n and n#1, we can write

XL>"XL#*X . (88)
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 241

The derivative f (X, q) is evaluated at specified values of X and q which depends on the numerical
scheme selected. As pointed by Press et al. [87], numerical schemes vary in complexity, accuracy
and efficiency as well. The fourth-order Runge—Kutta scheme is commonly used and has been
found to be sufficiently accurate for solving engineering problems.
To implement the Runge—Kutta method, the right-hand side of Eq. (86) is evaluated four times
for each time step *q: once at the initial point, twice at trial midpoints, and once at a trial endpoint.
Multiplying these derivatives by *q, we obtain four different increments for the next time step. The
procedure can be represented by the following equations:
*X "f (XL, qL) *q , (89)



*X *q
*X "f XL#  , qL# *q , (90)
 2 2


*X *q
*X "f XL#  , qL# *q , (91)
 2 2
*X "f (XL#*X , qL#*q) *q , (92)
 
*X *X *X *X
X L>"XL# # # # #0(*q) . (93)
6 3 3 6
In Eqs. (89)—(93), *X to *X are eight-component vector increments, and qL is the nth time step.
 
These equations can be written in terms of vector components, and from Eqs. (25) and (26), the
corresponding set of equations for Eq. (89) is given as
*x "xL *q, *x "(c HL!d PL)*q/(d c !c d ), *x "xL *q ,
           (94)
*x "(!c HL#d PL)*q/(d c !c d), *x "(xL!e xL ) *q ,
         
*x "(xL!e xL ) *q, *x "(xL!e xL ) *q, *x "(xL!e xL ) *q ,
           
where
PL"c xL#c xL#c xL#c xL#c xL
         


uN 
#c xL#c xL#c xL# G(xL )!f (qL) ,
      º* 

HL"d xL#d xL#d xL#d xL#d xL

         


1 
#d xL#d xL#d xL# M(xL )!g(qL) . (95)
      º* 

In Eqs. (94) and (95), *x is the jth component of the vector increment *X , the superscript
H 
n indicates that the associated values of the system variables x 2 x , P, H and independent
 
variable q are evaluated at the nth time step. Replacing *x , xL and qL in the above two
G G
equations with *x , (xL#*x /2) and (qL#*q/2) respectively, we can get a set of equations in
G G G
242 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

component form for Eq. (90). In a similar manner, Eqs. (91) and (92) can be evaluated. Substituting
into Eq. (93) we obtain xL> 2 xL\ .
 
From the definitions of w to w (see Eq. (18)), the initial values of w to w are equal to zero
   
while the other four variables a, a, m, m can take on arbitrary values. The initial conditions of the
system can be expressed as
X(0)"X"[x (0), x (0), 2 , x (0)]2"[a(0), a(0), m(0), m(0), 0, 0, 0, 0]2, (96)
  
where a(0), a(0), m(0), m(0) are the initial values of pitch displacement, pitch velocity, plunge
displacement, and plunge velocity, respectively.
Given initial values of the system variables, the starting procedure of the Runge—Kutta method is
straightforward. The formulas for the first-step integration can be obtained by simply replacing XL
and qL in Eqs. (89) and (90) by X or X(0) and q"0. The Runge—Kutta method is an explicit
and stable numerical procedure, and the only input parameter required is the time step *q.

4.3. Describing function

The describing function technique, sometimes referred to as the harmonic balance method, is
a method of obtaining an equivalent linear system such that traditional linear aeroelastic methods
of analysis can be employed. This method is essentially the same as the first approximation of
Kryloff and Bogoliuboff [58].
The first example of the use of the describing function method in aeroelastic systems known to
the present authors is by Shen [96], who analyzed a wing control-surface flutter with nonlinear
structural stiffnesses. The nonlinear elements chosen included freeplay and hysteretic structural
restoring forces and moments. A favorable comparison was made between results obtained from
the describing function method and those previously obtained by Woolston et al. [120,121] using
an analog computer. Johnson [51] extended the analysis to the next higher approximation and
used the term sinusoidal analysis. Brietbach [18,19] also used the describing function method to
assess the effect of various structural nonlinearities on the aeroelastic response of airfoils. Lauren-
son and Trn [60] used this method to investigate the aeroelastic response of missile control surfaces
containing either one or two structural nonlinearities. The describing function results were
compared with those obtained from numerical simulations showing excellent agreement. Lauren-
son and Trn [60] also demonstrated that for systems with structural nonlinearities, limit cycle
oscillations can be obtained at velocities below the linear flutter boundary. Both Brietbach [18,19]
and Lee [72] showed how the describing function method could be extended for use on systems
with multiple structural nonlinearities. Ueda and Dowell [111] employed the describing function
method for the case of aerodynamic nonlinearities in the analysis of a typical section subject to
transonic aerodynamics. They also compared their describing function results with those obtained
from numerical integration, verifying the describing function method of solution. Comparisons
between results obtained from a describing function method and numerical integration are also
given by Tang and Dowell [101,102] for the aeroelastic analysis of non-rotating helicopter blades,
and by Yang and Zhao [123], Zhao and Yang [125] and Liu and Zhao [75] in an investigation of
a typical section subject to nonlinear structural restoring forces. Dunn and Dugundji [24] used the
harmonic balance method, combined with the ONERA stall model, to investigate the aeroelastic
response of a three-dimensional wing. A Fourier analysis was employed to extract harmonics from
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 243

the ONERA aerodynamics, then the describing function method was used in conjunction with
a Newton—Raphson technique to solve the resulting nonlinear Rayleigh—Ritz equations.
Price et al. [89] analyzed a two-degree-of-freedom airfoil motion with either bilinear or cubic
structural restoring moments in pitch. As previously shown by Lee and Tron [66], such nonlineari-
ties are representative of loose control surface hinges. A describing function for a bilinear stiffness is
given by Lee and Tron [66], obtained using the method of Laurenson et al. [61]. However, this
describing function is applicable only where there is no preload. If a preload is applied, as in the
general case presented by Price et al. [89], then the airfoil does not oscillate about the zero mean
position. Instead the airfoil pitch motion is of the form
a(q)"B#A sin uq , (97)
where A and B are constants. To account for this offset, a dual-input describing function technique
must be used, as discussed by Gelb and Vander Velde [35], which results in a mean component for
the describing function. A similar study of a dual-input describing function technique is also
presented by Kim and Lee [55] for a flexible airfoil with a freeplay nonlinearity where the airfoil is
modeled as finite beam elements.
The dual-input describing function is given by
N"N #N sin uq#N cos uq . (98)
 !
Considering only the fundamental components of the restoring moment response, the following
expressions for N , N and N are obtained:
 !

 
1 L 1 L
N " M(a) d(uq), N " M(a) sin uq d(uq) ,
2n  n
\L \L
and (99)


1 L
N " M(a) cos uq d(uq) .
! n
\L
Evaluating the above integrals for the bilinear stiffness with M(a) given in Eq. (32), Price et al. [89]
obtained
N "A[(M #M d/2)/A!(c#b)/2!(1!M )g(c)#(1!M )g(b)] ,
 D D D (100)
N "A[1#(1!M )( f (c)!f (b))] and N "0 ,
 D !
where


1/n(sin\ x#x(1!x)) for "x"(1 ,
f (x)" !1/2 for x4!1 ,
1/2 for x51 ,
(101)


1/n(x sin\ x#(1!x)) for "x"41 ,
g(x)"
"x"/2 for "x"'1
and c"(a !B)/A and b"(a #d!B)/A.
D D
244 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

For a cubic nonlinearity given by Eq. (31), the describing function is given by [89]
N "B[b /B#b #b B#b B# A(b /B#3b )] ,
      
N "A[b #2b B#3b B# b A] , (102)
     
and
N "0 .
!
For a hysteresis nonlinearity of the type shown in Fig 11, Shen [96] derived the describing
function using the equations in dimensional form similar to those given by Eqs. (8) and (9).
Using the notations in this review, we can reproduce Shen’s [96] results, and for the sake of
brevity, consider only a hysteresis nonlinearity in the pitch degree-of-freedom. We replace
GM (h)"K h ,
F (103)
MM (a)"K (a)a ,
?
where K is a constant and K is a function of a. We express a in a form similar to Eq. (97) setting
F ?
B"0 as follows:
a"AeGSR . (104)
Retaining only the fundamental harmonics, Shen [96] obtained
K (a)"I uAK (A) . (105)
? ? ? ?
Here u is the reference frequency and K (A) is the describing function. When the describing
? ?
function is real, Eqs. (8) and (9) are a set of linear equations with different natural frequency u and
?
stiffness K which depend on the amplitude of oscillation. By specifying A, we can proceed to
?
compute the airfoil motion. When the describing function is complex, our equivalent linear system
will have structural damping introduced which is a function of the amplitude.
For a "0 in Fig. 11, Shen [96] derived the following expressions for K (a):
D ?
K (a)"!M #(2M /d)A cos ut, 0(ut(n/2 ,
?  
K (a)"!M , n/2(ut(n/2# ,
? 
K (a)"M !(2M /d)A cos ut, n/2# (ut(3n/2 ,
?  
K (a)"M , 3n/2(ut(3n/2# (106)
? 
K (a)"!M #(2M /d)A cos ut, 3n/2# (ut(2n ,
?  
where
"sin\(d/A)4n/2 .
The first harmonic now consists of both sine and cosine terms and is given by
K (a)"KeGSR\R , (107)
?
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 245

where
K"(K#K ), t"tan\(K /K )
A Q Q A
and
K "!4M cos /n#(2M /d)A[1! /n#sin 2 /n] ,
A  
K "!4M sin /n#(2M /d)A sin 2 /n .
Q  
The presence of structural damping is apparent from the imaginary part of K (a).
?
Once the describing function has been obtained, the method of solution is relatively straightfor-
ward. Using the non-dimensional equations (10) and (11), Price et al. [89] replaced the nonlinear
term M(a) by the appropriate describing function, given by Eqs. (100) or (102) for bilinear and cubic
nonlinearities, respectively. The describing function depends on the amplitude of oscillation A and
B, and thus, an iterative approach is required. First, it should be realized that there is no steady
external moment, aerodynamic or otherwise, acting on the airfoil, hence, there can be no steady
component to the restoring moment, or N must be equal to zero. Hence, in the iterative procedure
a value of A is initially assumed, then setting N "0 the value of B obtained, and the equivalent
linear stiffness given by the describing function is now known. From this point on the equations are
solved using standard linear aeroelastic techniques, in this case the º—g method [31] was employed
to determine the required value of º to give simple harmonic motion (which will be of amplitude
equal to the assumed value of A). The above procedure is repeated for different values of A, and the
variation of A and B with º is then obtained. For the hysteresis nonlinearity, the describing
function given by Eqs. (107) is to be used with Eqs. (8) and (9), but they can be changed to be
compatible with the non-dimensional equations (10) and (11).

4.4. Analytical techniques

Analytic solutions in nonlinear aeroelastic problems are only possible for special cases. We shall
consider structural nonlinearities where the restoring force or moment term can be represented by
an analytic function. The case of forced oscillating was studied by Lee et al. [69]. The governing
equations are given by Eq. (24), and to make the analysis general, the forcing terms are included in
Eqs. (21) and (22) such that
2 P(q)b
f (q)" ((!a )a(0)#m(0))(t e e\CO#t e e\CO)# (108)
k  F     mº
and
(1#2a ) Q(q)
g(q)"! F f (q)# . (109)
2r mºr
? ?
We assume that the external applied forces and moments are sinusoidal, and, without loss in
generality, the excitation is only applied in the pitch degree of freedom. In this case, P(q)"0 and we
write
Q(q)"Q sin uq , (110)

246 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

and let
F"Q /mºr . (111)
 ?
At sufficiently large values of q when transients are damped out and steady-state solutions are
obtained, f (q)P0. For sinusoidal external excitation forces, we assume the plunge and pitch
motions together with the expressions for w defined in Eq. (18) to be of the following form:
m(q)"a (q) cos (uq)#b (q) sin(uq), a(q)"a (q) cos(uq)#b (q) sin(uq) ,
   
w (q)"a (q) cos(uq)#b (q) sin(uq), w (q)"a (q) cos(uq)#b (q) sin(uq) , (112)
     
w (q)"a (q) cos(uq)#b (q) sin(uq), w (q)"a (q) cos(uq)#b (q) sin(uq) .
     
Here, a and b (i"1, 2, 2 , 6) are assumed to be slowly varying function in q. The amplitudes of
G G
m(q) and a(q) are written as
r"a#b (113)
 
and
R"a#b . (114)
 
From Eq. (112), we can obtain expressions for various time derivatives. Assuming the second
derivatives are small and can be neglected, we obtain a system of 12 first-order nonlinear
differential equations after matching the coefficients of cos(uq) and sin(uq). The resulting system in
a matrix notation is given by
AX"Y(X) , (115)
where
a y (x)
 
b y (x)
 
) )
X" and Y(X)" . (116)
) )
a y (x)
 
b y (x)
 
Here A is a matrix whose elements depend on the airfoil parameters and the structural nonlinearity.
If the structural nonlinearity, such as a freeplay or hysteresis, can be represented by a simple
polynominal expression, then G(m) and M(a) can be used to evaluate A and Y(X). Lee et al. [69]
consider cubic nonlinear functions of the form
G(m)"m#b m , (117)
K
and
M(a)"a#b a . (118)
?
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 247

The nonlinear terms m(q) and a(q) can be expressed as

m(q)" r[a (q) cos(uq)#b sin(uq)]#higher harmonics in 3uq, . . . ,
   (119)
a(q)" R[a (q) cos(uq)#b sin(uq)]#higher harmonics in 3uq, 2 .
  
In deriving the 12 first-order nonlinear differential equations, we substitute Eq. (119) for the
nonlinear terms m(q) and a(q), and assume that the higher harmonic terms are small and can be
neglected. The same procedure can be applied for nonlinear terms in the form mL(q) and aL(q) where
n is an integer 52. For the case where the nonlinear functions G(m) and M(a) are given in Eqs. (117)
and (118), the resulting matrix A in Eq. (115) can be expressed as

 
H 0
A" . (120)
0 I
Here I is an 8;8 identity matrix given by
1 0 2 0
0 1 2 0
I" . (121)
. . 2 .
0 0 2 1
The elements of H are functions of c , c , 2 , d , d , 2 , and the vector components
   
y (x), 2 , y (x) in Eq. (116) are functions of c , c , 2 , d , d , 2 , a , a , 2 , b , b , 2 , F, u, r, R.
         
They are long algebraic expressions given by Lee et al. [69].
Since the interest is in the harmonic solutions of Eq. (24) subject to an external excitation
g(q)"F sin(uq), the periodicity condition is enforced by requiring the coefficients a and b to be
G G
constants. Here, a , b , 2 , a , b are the equilibrium points for the system given in Eq. (115), and
   
they are determined by setting X"0 and solving for Y(X)"0. The harmonic solutions of the
dynamical system given in Eq. (115) are identified by a , b , 2 , a , b . In deriving the expressions
   
for the equilibrium points, Eq. (115) is first solved for a , b , 2 , a , b , which are then substituted
   
into the expressions for a , b , a , and b . Lee et al. [69] derived a frequency—amplitude relation-
   
ship given by
D R!D F"0 (122)
 
where D and D are functions of c , c , 2 , d , d , 2 , r, R, u and are given in their paper. The
     
amplitudes R and r are related by the expression
R"k r#k r#k r, (123)
  
where k , k , and k are functions of c , c , 2 , d , d , 2 , u.
      
Eqs. (122) and (123) give the relation between r and R with u. Lee et al. [69] showed that the
resulting equation after evaluating all the coefficients and substituting into the above two equa-
tions give a polynomial of degree 11 in r. Once r is computed, then R can be determined from Eq.
(123). For certain values of u, both equations admit multi-valued amplitudes.
If a more accurate analytical prediction is required, the expressions given in Eq. (112) can include
terms associated with higher harmonic. For example, the first two expressions for m(q) and a(q) in
248 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Eq. (112) can be replaced by

m(q)"a (q) cos(uq)#b (q) sin(uq)#c (q) cos(kuq)#d (q) sin(kuq) , (124)
   
a(q)"a (q) cos(uq)#b (q) sin(uq)#c (q) cos( juq)#d (q) sin( juq) , (125)
   
with similar expressions for w , 2 , w . The values of k and j in the above equations will be
 
determined according to the form of the nonlinear functions. If the nonlinear functions G(m) is
G(m)"m#b m , (126)
K
and Eq. (118) is used for M(a), we should take k"2 and j"3. The value for jO2 because the first
part of the approximation in uq when cubed gives a higher harmonic in 3uq and not 2uq.
Substituting the modified expressions which include the higher harmonic terms into Eq. (24), we
obtain a linear system given by Eq. (115) but with a higher dimension.
The procedure presented here can also be applied to a self-excited system, in which f (q) and g(q)
in Eqs. (108) and (109) are equal to zero. Lee et al. [70] derived the amplitude-frequency relation for
a self-excited system in which G(m) is a linear function G(m)"m and M(a) is a nonlinear function
given by Eq. (118). Substituting the expressions given in Eq. (112) into Eq. (24), and after
considerable algebraic simplifications, the amplitude of the pitch and plunge motions are obtained,
and they are expressed as
R"f (u)$(f (u) , (127)
 
and
r"A(u)R . (128)
Here, f , f , and A are functions of c , c , 2 , d , d , 2 , u, and can be obtained from Lee et al. [70].
     
Unlike the case for the system subject to a sinusoidal driving force F sin(uq), we do not have
a reference value for the frequency u for the self-excited system. Consequently, the amplitude
relations given in Eqs. (127) and (128) are not complete, and we need to derive an additional
relation for u. Lee et al. [70] discussed various methods to estimate u. One possible approach is to
apply the center manifold theory to Eq. (24). By rewriting in its normal form, a linear equation
which relates the frequency of the limit cycle oscillations can be derived. Using this equation which
provides the value for u, and Eqs. (127) and (128), the amplitudes of the limit cycle oscillations for
the self-excited system can be determined analytically. The details on the application of the center
manifold theorem is given in Liu et al. [76] and an example which demonstrates the accuracy of the
analytical predictions will be presented in Section 5.2.

5. Bifurcation and chaos of airfoils with structural nonlinearities

5.1. Stability and bifurcation analysis

5.1.1. Linear stability analysis of autonomous systems

In Eq. (24) the terms f (q) and g(q) for unforced motion can be neglected if a(0);1 and m(0);1,
or q<1. Lee et al. [70] considered the system response at large times when transients are damped
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 249

out (that is, q<1). The equations governing the airfoil motion are then a set of autonomous
differential equations, and the vector components for X and f(X) are given by Eqs. (25) and (26)
with f (q)"g(q)"0.
To investigate the stability of the nonlinear system, we first apply a linearization technique and
then examine the stability using a linear analysis. This approach has been in use for a long time, and
the justification has been given by Poincaré [84] more than a hundred years ago. It is well known
that the flow of the nonlinear dynamical system in the neighborhood of the singularity (i.e. the
equilibrium point of the vector field f ) is topologically equivalent to the flow of the linearized
system around that fixed point. A dynamical system is said to be structurally stable if small
disturbances introduced to the system lead to approximately the same results as those obtained in
the absence of those disturbances.
Denoting X to be a fixed point of the system, we define y(q) to be a small perturbation about X
# #
given by

y(q)"X(q)!X . (129)
#
To investigate whether the perturbation decays or grows, a differential equation for y(q) is obtained
by differentiating Eq. (129) with respect to q to give

d
y" (X!X )"X"f(X)"f(X #y) . (130)
dq # #

Expanding f(X) about X using Taylor’s series, we obtain

#
jf(X)
f(X #y)"f(X )# y#O(y) , (131)
# # jX X
#

where O(y) denotes a second order term in y and can be neglected. Since X is a fixed point,
#
f(X )"0, we obtain
#
jf(X)
y" y"Jy , (132)
jX X
#

where J"jf(X)/jX" X# denotes the Jacobian matrix, and Eq. (132) is linear in y. If the fixed point is
non-degenerate, then det JO0, and there exists a real and non-singular matrix T such that

j

j
T\JT"" "  . (133)
( \
j

Using a linear transformation, y"TZ, we obtain a simple system in Z, such that

Z"" Z . (134)
(
250 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

The solution for the above system is given by Z "C exp(j q), for i"1, 2, 2 , 8, where C is
G G G G
a constant. If all eigenvalues of J are non-zero and have non-zero real part for the complex
conjugate eigenvalues, and if all the real part of j(0, Z(q)P0 as qPR, the perturbation y(q)
decays, and the system is said to be stable. On the other hand, if the real part of j'0, y(q) will grow
exponentially with q, and the system is unstable.
Another approach to analyze the stability of the nonlinear system is to use the stable and
unstable manifold theorem [20,42]. The nonlinear system given in Eq. (24) can be rewritten as
X"AX#G(X) , (135)
where A is a constant 8;8 matrix of which all eigenvalues have non-zero real parts. The term AX
represents the linear part of the system and G(X) represents the nonlinear part. The stable and
unstable manifold theorem deals with the autonomous system in the form given by Eq. (135). If the
nonlinear term G(X) is smooth, and
#G(X)#
lim "0 , (136)
#X#P0 #X#
then, in the neighborhood of the fixed point, there exists stable and unstable manifolds W

and W with the same dimensions as the stable and unstable manifolds E and E of the linear
  
system
y"Ay , (137)
in which E and E are tangent to W and W . Thus, the stability can be analyzed by determining
   
the sign of the eigenvalues of A following the approach used to investigate the linearized system
given in Eq. (132).

5.1.2. Nonlinear analysis

In using a linear stability analysis, we assume that all eigenvalues of J in Eq. (132) and all
eigenvalues of A in Eq. (137) have non-zero real part. If this condition is not satisfied, i.e., either we
have a zero eigenvalue j"0 or a pair of purely imaginary eigenvalues j"$iu, then a nonlinear
analysis is needed to determine the stability of the dynamical system. An important tool in the
nonlinear analysis of dynamical system is to use the theorem of center manifolds, which can be used
to separate the influence of the stable and unstable manifolds. A detailed discussion on the
theoretical background and applications of the center manifold is given by Gukenheim and
Holmes [39], Verhulst [115], Beyn [14], Wiggins [117], and Kuznetsov [59]. Here, we present
only the basic idea and describe the method to analyze the bifurcation of dynamical systems
associated with one parameter d. Eq. (24) can be rewritten as
X"f(X, d) , (138)
where d3R, X3R, and the system is known as the co-dimension one bifurcation problem. By
changing the parameter d, the asymptotic system behavior may switch from one state to another.
For certain values of d, one type of invariant set may lose its stability and a new type of invariant
set may be created which takes over the stability. This is generally referred to as a bifurcation. If the
Jacobian matrix has a real eigenvalue which crosses zero as d changes, we have a turning point or
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 251

saddle-node bifurcation. If J has a pair of complex conjugate eigenvalues which cross the
imaginary axis, we have a Hopf-bifurcation.
The center manifold theory can be applied to analyze the behavior of the dynamical system in
nonlinear aeroelasticity. The main attractive feature of the center manifold is that it can be used to
lower the dimensionality of the system under investigation, so that the dynamics of a high-
dimensional system can be studied in a low-dimensional system.
For the autonomous system given in Eq. (138), we introduce a parameter d such that

1 1
" (1!d) , (139)
º* º*
*

where º* is constant and equal to the linear flutter speed. By substituting the expression for º*
*
given in Eq. (138) into Eq. (139), the system of equations can be rewritten as

X"AX#BdX#(1!d)F(X) ,

d"0 , (140)

where

   
A A B 0
A"   , B"  , (141)
A A 0 0
 

and

0 1 0 0 0 0 0 0 1 0 0 0
a a a a a a a a 1 0 0 0
A "     , A "     , A " ,
 0 0 0 1  0 0 0 0  0 0 1 0
a a a a a a a a 0 0 1 0
       
(142)
!e 0 0 0 0 0 0 0

0 !e 0 0 b b b b
A "  , B "     .
 0 0 !e 0  0 0 0 0

0 0 0 !e b b b b
    

The first two terms in Eq. (140) represent the linear part of the dynamical system, and the nonlinear
part is represented by F(X) given by

F(X)"(0, f , 0, f , 0, 0, 0, 0)2, (143)

 
252 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

where



 
1  uN 
f "j c M(x )!d G(x ) ,
  º*   º* 
* *



 
1  uN 
f "!j c M(x )!d G(x ) , (144)
  º*   º* 
* *
j"c d !d c .
   
The coefficients a , b are defined as follows:
GH GH
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) , (145)
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
a "j(d c !c d ), a "!j(d c !c d ) ,
         
b "0, b "!jd c , b "0, b "jc d ,
       
b "0, b "jd c , b "0, b "!jc d ,
       
where c , d , 2 , c , d are given in appendix A with º* replaced by º* . The functions G(x ) and
    * 
M(x ) are the plunge and pitch degree-of-freedom structural nonlinearities (see Eqs. (10) and (11))

and x "m, x "a. Here, M(a) and G(m) are assumed to contain only nonlinear functions in a and
 
m, respectively. When linear terms are included, some modifications are needed in which b , b ,
 
b , b become non-zero. The details can be found in Liu et al. [76].
 
5.1.3. Hopf-bifurcation point
We denote (X(d), d) to be a stationary branch of Eq. (138), and assume that at d"d the Jacobian

matrix f (X(d ), d ) has a simple eigenvalue iu , u O0 with eigenvector P #iP , and no other
6      
eigenvalue of the type iku , k"0, 2, 3, 2 . If the condition

d[Re(j)]
O0 , (146)
dd d


is satisfied, then there exists a smooth branch of periodic solutions. A Hopf-point can be regarded
as the transition point where a stable equilibrium solution changes to an oscillatory solution [94].
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 253

Let q be a vector such that q3R, and satisfies the following condition:

q2P "0, and q2P "1 , (147)

 
where P and P are the eigenvectors. The condition expressed in Eq. (146) is satisfied if and only if
 
(X(d ), d , P , P , u )3R is a regular solution of the following equation:
    


f (X, d)
f (X, d)P #uP
V  
T(X, d, P , P , u)" f (X, d)P !uP "0 . (148)
  V  
q2P

q2P !1

The Hopf-point is determined by setting d "0 in Eq. (140) and the value of º* is determined
 *
from the condition that the eigenvalues of A consist of a pair of imaginary eigenvalues $iu .

From the eigenspace of A, a transformation matrix P is determined, such that

0 u

!u 0

b c
!c b
P\ AP" "J . (149)
a

a

a

a


The characteristic equation of A has two pairs of complex eigenvalues $iu , b$ic, and four

real eigenvalues a , 2 , a . By introducing a new variable Z3R, such that Z"P\X, the system
 
given in Eq. (140) can be rewritten as

Z"JZ!d(P\ BP)Z#(1!d)P\F(PZ) ,

d"0 . (150)

The standard form of Eq. (150) can be expressed as

z "u z #f (z , z , 2 , z , d), z "!u z #f (z , z , 2 , z , d) ,

             
z "bz #cz #f (z , z , 2 , z , d), z "!cz #bz #f (z , z , 2 , z , d) ,
             
z "a z #f (z , z , 2 , z , d), z "a z #f (z , z , 2 , z , d) , (151)
             
z "a z #f (z , z , 2 , z , d), z "a z #f (z , z , 2 , z , d), d"0 .
             
254 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

By the application of the center manifold principle, and after solving a set of algebraic equations,
the above nine-dimensional system can be reduced to the following two-dimensional system:
u "u u #g (u , u , d) ,
      (152)
u "!u u #g (u , u , d) ,
     
where g and g contain the nonlinearity terms of u , u , d. Since the center manifold is an invariant
   
manifold, the system behavior of the eight-dimensional equations given in Eq. (138) can be studied
by analyzing the two-dimensional system given in Eq. (152).
As an example, the center manifold technique was applied by Lee et al. [70] to analyze the
bifurcation of the system given in Eq. (138) with k"100, x "0.25, r "0.25, u"0.2. They found
? ?
that the stationary points loose the hyperbolicity property when a pair of complex conjugate
eigenvalues crosses the imaginary axis. From the solution of Eq. (152), we obtain a stationary
solution for d(0, and the solution looses its stability and switches to a limit-cycle oscillation when
d'0. These solutions are not dependent on the value of the initial condition a(0). This transition
also confirms that we have a supercritical Hopf-bifurcation. The dynamical behavior of the system
defined by Eq. (152) agrees with that computed numerically from Eq. (138) using a fourth-order
accurate Runge—Kutta scheme.
A further simplification can be obtained by rewriting the two-dimensional center manifold
equations first in normal form so that they can be reduced to a single first-order differential
equation. When this equation is expressed in polar coordinates, Liu et al. [76] derived a relation for
the frequency expressed as a function of d as follows:
u"u #Cd , (153)

where C is a constant depending on the system parameters and is determined by the transformation
from the center manifold equations to the normal form. Substituting the frequency determined
from the above equation into Eqs. (127) and (128), the amplitudes of limit cycle oscillations (LCO)
for the self-excited system can be computed.

5.1.4. Analysis of non-autonomous systems

We consider the system to be subject to a sinusoidal excitation, and without loss in generality,
the excitation given in Eqs. (108) and (109) is applied only in the pitch degree of freedom (i.e.,
P(q)"0). In this case, f (q)"0 and g(q)"F sin uq, and the system of equations expressed in Eq.
(24) is non-autonomous. Time-dependent systems are more complicated than autonomous sys-
tems, since the solutions depend upon both X and q instead of solely on X. Moreover, the concept
of a fixed point is no longer available. Consequently, the linear stability analysis and the center
manifold theory discussed above cannot be directly applied to examine the stability and bifurcation
of the non-autonomous system. However, a method can be derived whereby a non-autonomous
system can be rewritten as an autonomous system.
When an external excitation F sin(uq) is included, Eq. (20) can be expressed as


1 
d m#d a#d a#d a# M(a)#d m#d m#d w #d w #d w #d w
    º*          
"F sin(uq) . (154)
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 255

The resulting system of first-order differential equations comprising of Eqs. (19) and (154) leads to
a non-autonomous system due to the presence of the external excitation forcing. However, it is easy
to verify that the solution for the following system:
y#uy"0 ,
y(0)"0 , (155)
y(0)"uF
is given by y(q)"F sin(uq). Eq. (154) can be rewritten in the form:


1 
d m#d a#d a#d a# M(a)#d m#d m#d w
    º*    

#d w #d w #d w !y"0 ,
     
y#uy"0 . (156)
The system of first-order differential equations given by Eqs. (19) and (156) becomes an
autonomous system. In this case, X3R, where
X"(a, a, m, m, w ,w , w , w , y, y )2 . (157)
   
The stability and bifurcation analysis presented earlier can be applied to the new system.
Another useful result in the stability analysis for non-autonomous systems is due to Poincaré
[115]. For a system given by
X"AX#B(q)X#f(q, X) , (158)
if A is a constant matrix with all eigenvalues having a negative real part, B(q) is a matrix with the
property

lim #B(q)#"0 , (159)

O
and the vector function f(q, X) is continuous in q and X, and is Lipschitz continuous in X in the
neighborhood X"0, while satisfying
#f (q, X)#
lim "0 , (160)
#x#P0 #X#
uniformly in q, then the solution X"0 is asymptotically stable. On the other hand, if the matrix
A has at least one eigenvalue with a positive real part, then the solution is unstable.

5.1.5. Numerical analysis of stability and bifurcation — AUTO program

Another approach to analyze stability and bifurcation of dynamical systems is by numerical
methods. Alighanbari and Price [1] investigated Eqs. (27) and (28) using the AUTO software
package [23]. Bifurcation diagrams with stable and unstable branches were obtained, and the types
of bifurcations were verified by calculating the Floquet multipliers.
256 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

A brief description of the method is given here. AUTO can perform a limited bifurcation analysis
of algebraic systems, or of a system of ordinary differential equations given in Eq. (138). The
differential equations are approximated by the method of orthogonal collocation [86] at m Gauss
points with piecewise polynomials; in AUTO m is an integer between 2 and 7. More precisely,
defining an N-point mesh on each time interval, then for any general mesh-point j the Lagrange
basis polynomials are introduced, which are defined by

K q!q i
w (q)" “ H>IK , q "q #m (q !q ) , (161)
HG q !q H>GK H H> H
I IO H>GK H>IK
where j"0, 1, 2 , N!1 and i"0, 1, 2 , m. The collocation method consists of finding a summa-
tion of the form

K
p (q)" w (q)X , (162)
H HG H>GK
G
such that it satisfies the differential equation at zeros of the mth degree Legendre polynomial
relative to each subinterval

p (z )"f (p (z ), v), i"1, 2 , m, j"0, 1, 2 , N!1 , (163)

H HG H HG
where v is a parameter such as air speed. With the above choice of basis polynomials the
continuous solution X(q) is approximated by X and X at q and q , respectively.
H H>GK H H>GK
Therefore the procedure required to take one step along a solution branch consists of solving
a system of nonlinear algebraic equations, which is done by Newton or Newton—Chord iteration.
AUTO also computes the approximate Floquet multipliers by applying a standard eigenvalue
routine to an approximation of the linearized Poincaré map. The stability of periodic solutions is
determined by the Floquet multipliers of the linearized equations. One of these multipliers is
always equal to #1, and if the remaining multipliers lie inside the unit circle of the complex plane
the periodic solution is stable. Loss of stability of the periodic solution, accompanied by a bifurca-
tion, occurs when one of the Floquet multipliers exits from the unit circle as velocity is varied.
Eqs. (27) and (28) can be written in equivalent first-order form as

[A]+X,"[B]+X,#+F, , (164)

where +F, and +X, are vectors given by

+F,"+0, 0, 0, 0, 0, 0, !(u/º*)N (m), !N (a)/º*,2 (165)

K ?
and

+X,"+x , x , x , x , x , x , x , x ,2 . (166)
       
In Eq. (166), x "m, x "a, x "m, x "a, x "m, x "a, x "m and x "a. [A] and
       
[B] are 8;8 sparse matrices given in Alighanbari and Price [1].
Once the structural nonlinearity G(m) and M(a) are specified, a bifurcation analysis can be
performed using AUTO. Alighanbari and Price [1] used a third-order rational function to
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 257

approximate a freeplay nonlinearity in the pitch degree of freedom and expressed the restoring
moment as

c #a(c #a(c #c a))

M(a)"     (167)
1#a(c #a(c #c a))
  
where c , 2 , c are constants. The nonlinearity in the plunge degree of freedom G(m) can be
 
represented by a similar expression.

5.2. Cubic nonlinearity

Cubic nonlinearities in one-degree-of-freedom mechanical and electrical systems can often be

represented by a Duffing’s equation that has been the subject of investigation for many years. The
dynamic response of a coupled 2-DOF system with cubic nonlinearities was investigated analyti-
cally and numerically by Wong et al. [119] who showed that the amplitude-frequency response
curve has a much more complex structure compared to a 1 DOF system. The amplitude-frequency
relation changes from a cubic equation for a 1 DOF system to a polynomial of degree nine for
a 2 DOF system. The coupled Duffing’s equations were further investigated by Gong et al. [36]
who showed that harmonic, quasi-periodic and chaotic motions can exist for system parameters
that correspond to those commonly used to analyze aeroelastic behavior of aircraft structures. The
first attempt to study the effects of a cubic structural nonlinearity in aeroelasticity was carried out
by Woolston et al. [120,121] using an analog computer. They analyzed a 2 DOF system for hard
and soft springs in the torsional degree of freedom of a pitch and plunge system.
Lee and LeBlanc [64] analyzed numerically a 2 DOF airfoil motion with a cubic nonlinearity in
the pitch degree of freedom. They investigated the effects of initial pitch displacement on the flutter
boundaries of soft and hard springs, as well as the amplitudes of pitch and plunge motion of limit
cycle oscillations for various system parameters using Eqs. (78) and (79). The cubic nonlinearity is
represented by Eq. (31), and similar to Woolston et al. [120,121] b and b were set to zero. In Eqs.
 
(19) and (20), G(m)"m and M(a)"a#b a after setting b "1 and b "b for the pitch
?   ?
nonlinearity. The elastic axis of the airfoil was placed at the 1/4 chord point (that is a "!1/2),

f "f "0 and r was kept constant at 0.5. Other properties of the airfoil, such as k, x , u and
K ? ? ?
b were varied in their studies.
?
The deviation from linearity of the restoring moment is shown in Fig. 23 where the term a#b
?
a is plotted against a for three values of b up to $3. At the largest value of b "3, the
? ?
contribution of the cubic term is about 37% at a"20°, approximately 20% at a"15° and can be
neglected for a(5°. For values of a+10—15° where linear aerodynamic theory can be used (Eqs.
(15) and (16)), the nonlinear term can be considered small and the system is dominated by the linear
stiffness.

5.2.1. Flutter boundaries for soft and hard springs

For a soft spring, b is negative, Lee and LeBlanc [64] investigated the effects of initial conditions
?
by a numerical simulation study using the finite difference scheme given in Section 4.1. The linear
flutter velocity º* was used as a reference velocity and was determined numerically by setting the
*
nonlinear term b to zero.
?
258 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 23. Effects of b on nonlinear moment M(a)"a#b a (from Ref. [64]).

? ?

Fig. 24 shows flutter boundaries obtained by Lee et al. [70] for !22°(a(0)(to 22° and
a(0)"0.0, 1.15, 1.72 and 2.29°/unit time while keeping m(0)"m(0)"0. When a large value of a(0)
is used, the transient solution at small values of time is in error since the linear aerodynamics in the
formulation given in Section 3.1 becomes invalid. As long as the steady-state solution decays to
zero or reaches an amplitude of less than approximately 10—15° in the case of limit cycle
oscillations, the results are acceptable. However, if the solution is chaotic, the initial transients may
affect the solution at large values of time, and we should avoid using large values of a(0) and limit
a(0) to less than approximately 10—15°. The motion is unstable to the right of the flutter boundary.
The airfoil parameters are: k"100, u"0.2, r "0.5, a "!1/2, x "0.25 and b "!3.0. The
?  ? ?
destabilizing effect of a soft spring is illustrated in this figure which shows that flutter can be
induced at a velocity º* below the linear flutter velocity º* . This effect increases with increasing
*
a(0). The flutter boundary curve for a(0)"0 is symmetrical about the axis a(0)"0. For the other
three values of a(0), the flutter boundaries are not symmetrical because an initial positive velocity is
given to the airfoil.
Fig. 25 shows the destabilizing effect of initial a(0) on flutter boundaries for !3(a(0)(
3°/unit time at a (0)"0, 5, 10, 15° for the same values of k, b , x , r and u. As expected, the curve
? ? ?
for a(0)"0° is symmetric about the axis a(0)"0, while for a(0)'0, the flutter boundaries are
displaced downwards and the amount of the shift increases with a(0).
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 259

Fig. 24. Flutter boundary (a(0) vs. º*/º* ) for a soft spring; u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0
* F ? ? ? K
and b "!3 (from Ref. [70]).
?

Fig. 25. Flutter boundary (a(0) vs. º*/º* ) for a soft spring; u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0
* F ? ? ? K
and b "!3 (from Ref. [70]).
?

The flutter boundaries for !0.6(m(0)(0.6 corresponding to m(0)"0.0, 0.01, 0.02, 0.03 and
0.04 are shown in Fig. 26 for the same values of k, b , x , r and u. The criterion for selecting the
? ? ?
ranges of m(0) and m(0) is similar to that for a(0) and a(0), that is, the amplitudes of plunge and pitch
motions are sufficiently small for the linear aerodynamics to be applicable. The flutter boundary for
m(0)"0.0 is symmetric with respect to the m(0)-axis while for other values of m(0) it is highly
non-symmetric. The value of m(0) at the maximum º*/º* moves downwards in the negative m(0)
*
260 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 26. Flutter boundary (m(0) vs. º*/º* ) for a soft spring; u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0
* F ? ? ? K
and b "!3 (from Ref. [70]).
?

Fig. 27. Flutter boundary (m(0) vs. º*/º* ) for a soft spring; u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0
* F ? ? ? K
and b "!3 (from Ref. [70]).
?

direction. Each of the curves for which m(0)O0 intersects with the flutter boundary for m(0)"0 in
the region of negative m(0).
Fig. 27 gives the flutter boundaries for !0.04(m(0)(0.04 at m(0)"0.0, 0.2, 0.4, 0.6 for the
same values of k, b , x , r and u. The curves cross each other in the region of negative m(0).
? ? ?
The flutter boundary is affected by the system parameters such as k, b , u and x . These effects
? ?
have been studied by Lee and LeBlanc [64] and Fig. 28, taken from their report, shows the effects
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 261

Fig. 28. Effect of k on flutter boundary (a(0) vs. º*/º* ) for u"0.2, a "!0.5, x "0.25, r "0.5 and f "f "0 (from
* F ? ? ? K
Ref. [64]).

of airfoil/air mass ratio on the flutter boundaries. This figure shows that increasing k has
a destabilizing effect, and that the boundary curves are all displaced more towards the left for the
larger values of k. The effect is more pronounced as the value of b decreases.
?
Increasing the distance between the centre of mass and the elastic axis has a stabilizing effect,
and results from Lee and LeBlanc [64] at k"250 show the flutter boundaries to move closer to
the linear flutter speed as x is increased. In the same report, the effect of u was investigated and the
?
results show that as the uncoupled natural frequency for plunging motion approaches that of
the pitching motion, the flutter boundaries move closer to the linear flutter boundary and the
destabilizing effect becomes smaller. At u"1.2, it was shown [64] that the flutter boundaries are
virtually independent of the coefficient b .
?
The system stability near equilibrium points can be presented in a plot similar to a bifurcation
diagram. For b (0, a subcritical Hopf-bifurcation [108] occurs at º*/º*"1. In Fig. 29, a(0) is
? *
plotted against º*/º* for a(0)"m(0)"m(0)"0 with b "!3.0, k"100, u"0.2, x "0.25 and
* ? ?
r "0.5. The dotted lines for º*/º*41.0 is the flutter boundary. For values of a(0) bounded by
? *
these two curves, the motion is stable. The inset of four figures showing the trajectories of the pitch
oscillation are plotted in the phase plane (a vs a). They are obtained at º*/º*"0.94, 0.98, 1.02,
*
1.06 and a(0)"10, 7.5, !0.2 and !0.2°, respectively. The system converges to the equilibrium
points (a"a"0) for º*/º*(1.0 and diverges away from the unstable equilibrium points
*
(a"a"0) for º*/º*'1.0. The Hopf-bifurcation point is located at º*/º*"1.
* *
For positive values of b (hard spring), divergent flutter is not encountered. Instead, the flutter
?
boundaries (e.g. a(0) vs. º*/º*) for all b coalesce into a straight line at the linear flutter speed. This
* ?
was shown by Woolston et al. [120,121] and numerically by Lee and LeBlanc [64] for a cubic
262 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 29. Subcritical Hopf-bifurcation for a soft spring with b "!3 (from Ref. [70]).
?

nonlinearity in the pitch degree of freedom and various airfoil parameters. To the right of this
boundary, the oscillations maintain a self-limited amplitude which is independent of the initial
angular displacement a(0).
For a given a(0), the time it takes for the airfoil motion to reach a steady value depends on the
velocity ratio º*/º* . If a solution is required very close to the flutter boundary, a large number of
*
cycles of oscillations have to be computed. Using the same airfoil parameters as those for a soft
spring (that is, k"100, u"0.2, x "0.25 and r "0.5), Lee and LeBlanc [64] showed that at
? ?
b "0.3 it takes approximately 250 cycles for the solution to decay to practically zero at
?
º*/º*"0.9992 (0.08% less than linear flutter speed) with initial amplitude a(0)"10° and
*
a(0)"m(0)"m(0)"0. At smaller values of a(0)"1 and 3°, it requires approximately the same
number of cycles for the oscillation amplitude to reach zero. In the LCO region where º*/º*'1,
*
the time it takes to reach steady state depends on the initial conditions. At º*/º*"1.0008 (0.08%
*
greater than linear flutter speed), the solution takes 130 cycles at a(0)"10° to reach a constant
amplitude, compared to 190 cycles at a(0)"1°. Increasing the value of º*/º* increases the
*
convergence rate. At º*/º*"1.0024, steady pitch amplitudes are reached after 60 and 100 cycles
*
for a(0)"10 and 1°, respectively. Eq. (24) has been used to repeat the Lee and LeBlanc [64]
investigation and identical results were obtained. It was found that when larger values of b were
?
used, the time it takes to decay to a zero value (º*/º*(1) or reach a constant value (º*/º*'1)
* *
is considerably shorter than those at b "0.3 shown in Lee and LeBlanc’s [64] report.
?
Lee and LeBlanc [64] concluded from a number of case studies for different airfoil parameters
that there is no noticeable change in the amplitude of the pitch motion when the airfoil/air mass
ratio is varied, while the plunge amplitude increases with an increase in that ratio. Increasing the
distance between the centre of mass and the elastic axis results in a larger pitch amplitude while the
plunge amplitude gets smaller. As the ratio of the uncoupled natural frequencies of the plunge to
pitch motion approaches and exceeds unity, there is an increase in the pitch amplitude while
a much larger drop in the plunge amplitude is detected.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 263

Fig. 30. Subcritical Hopf-bifurcation for a hard spring with b "3 (from Ref. [70]).
?

Similar to Fig. 29 the stability behavior of the airfoil near equilibrium points can be presented in
a bifurcation diagram. For b "3, k"100, u"0.2, r "0.5, a "!1/2 and x "0.25, Fig. 30
? ? F ?
shows a plot of the pitch amplitude a against º*/º*. This figure is obtained by varying a(0) only,
 *
while keeping a(0)"m(0)"m(0)"0. For º*/º*(1, the solution is stable for all initial displace-
*
ment a(0). The solutions in the phase plane (a vs. a) at º*/º*"0.94, a(0)"10° and
*
º*/º*"0.98, a(0)"10° are shown in the inset. The phase curve spirals into the origin
*
(a"a"0) for all a(0), showing the solution to be stable for º*/º*(1. For º*/º*'1, the two
* *
insets of the phase diagram at º*/º*"1.02 and 1.04 for a(0)"0.2° show the trajectories to spiral
*
away from the origin and becomes periodic with constant amplitude. The solid line shows the
amplitude of pitch motion a obtained numerically using Eq. (24). This value is independent of

initial displacement a(0) since all solution with different a(0) will eventually reach a limit-cycle state.
The transition from a stable equilibrium solution to limit cycle oscillation is known as a supercriti-
cal Hopf-bifurcation [108].

5.2.2. System behavior near equilibrium points

For various values of º*/º* with k"100, x "0.25, r "0.5 and u"0.2, the eigenvalues of the
* ? ?
Jacobian matrix at equilibrium points can be calculated using Eq. (132). The solution of the
Jacobian matrix for a given value of º*/º* has eight eigenvalues. There are two pairs of conjugate
*
eigenvalues (modes I and II denoted by dashed and solid lines, respectively) and their real and
imaginary parts are plotted against º*/º* in Fig. 31. The other four eigenvalues have zero
*
imaginary parts and they do not represent oscillatory motion.
On examining a typical time series of a from numerical integration of Eq. (24) shown in Fig. 32,
we see that the pitch motion can be adequately represented by a single mode. We can denote the
pitch motion by a"a e(j#ij )q where a is a constant, j and j are the real and imaginary parts of
  
the exponential coefficient j. Superimposed in Fig. 31 are results obtained from numerical
simulation for b "3. The initial conditions are a(0)"1° (for º*/º*(1) and a(0)"0.1° (for
? *
264 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 31. Variation of j with º*/º* for the pitch motion with u"0.2, k"100, a "!0.5, x "0.25, r "0.5,
* F ? ?
f "f "0, and b "3 (from Ref. [70]).
? K ?

Fig. 32. Time series of pitch at º*/º*"0.8 with u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0, and
* F ? ? ? K
b "3 (from Ref. [70]).
?

º*/º*'1) with a(0)"m(0)"m(0)"0. The imaginary part j is calculated by the following

*
equation j "2n(n !n )/(q !q ) where time q and q correspond to cycle numbers n and n ,
       
respectively. The real part j is determined by j "(log a !log a )/(q !q ) where a and
      
a are the amplitudes of pitch angles corresponding to q and q .
  
The maximum values of a and a are kept below 10° in the numerical simulation. The first 
  
to 3 cycles in the time series are used to calculate j and j .
 
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 265

The analytical and numerical simulation results are in excellent agreement for mode II for
º*/º*'1. However, for º*/º*(1, the results from numerical simulation have some scatter
* *
about the analytical solution in j only.
P
Time series of m (Fig. 33) show that for º*/º*(1 more than one mode is present. We can write
*
m as the sum of two modes as m"  m ejq . The two modes can be obtained by decomposing the
L L
time series using wavelets. The real and imaginary parts of j can then be solved and their values are
plotted in Fig. 34 which shows fair agreement with the solution from the Jacobian matrix. It is

Fig. 33. Time series of plunge motion at º*/º*"0.8 with u"0.2, k"100, a "!0.5, x "0.25, r "0.5, f "f "0,
* F ? ? ? K
and b "3 (from Ref. [70]).
?

Fig. 34. Variation of j with º*/º* for the plunge motion with u"0.2, k"100, a "!0.5, x "0.25, r "0.5,
* F ? ?
f "f "0, and b "3 (from Ref. [70]).
? K ?
266 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

difficult to obtain good accuracy of j when only a few cycles of the time series are available. This is
especially true for mode I which decays to practically zero in less than three cycles. The results
show the coalescence of the two modes and after a short transient period, only the mode II is
present for º*/º '1. In the region 0.974º*/º 41, the use of wavelets to separate the two
* *
modes becomes increasingly difficult as the two frequencies approach each other. Experience [71]
with wavelets shows that j is difficult to obtain accurately when frequency separation between the

two modes is not large.
In Fig. 35 the eigenvalues of mode II for the pitch motion together with j from numerical
simulations shown in Fig. 31 are re-plotted in the real and imaginary plane for º*/º*"0.97 to
*
1.08. The filled circle symbols denote numerical results while the solid lines represent solutions
from the eigenvalues of the Jacobian matrix. We can see that the real part of the complex
conjugates increases with º*/º* and passes through zero at º*/º*"1.0 where a Hopf-bifurca-
* *
tion occurs. With further increase in º*/º*, the real part of the conjugates switches sign and
*
becomes positive, changing the stable equilibrium state to an unstable one.

5.2.3. Flutter velocity and angular frequency at Hopf-Bifurcation point

By solving Eq. (148), Lee et al. [70] determined the Hopf-bifurcation point and found the linear
flutter velocity and angular frequency at this point. These two quantities have been calculated for
various airfoil parameters and compared with numerical simulation. The results show that up to
the fourth decimal point there were no noticeable differences detected.
Fig. 36 illustrates the variation of the linear flutter velocity and angular frequency at the
Hopf-point as functions of the frequency ratio u for k"100, x "0.25 and r "0.5. The flutter
? ?
velocity decreases with u until u"0.8 is reached and increases from then on with increasing u.
The angular frequency increases rapidly with u initially until a maximum is reached at u"1.1 and
then decreases gradually. The effect of airfoil—air mass ratio on flutter velocity and frequency shows

Fig. 35. Variation of conjugate eigenvalues with º*/º* for pitch motion in the j-plane for u"0.2, k"100, a "!0.5,
* F
x "0.25, r "0.5, f "f "0 and b "3 (from Ref. [70]).
? ? ? K ?
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 267

Fig. 36. Variation of flutter velocity and frequency at the Hopf-point for k"100, a "!0.5, x "0.25, r "0.5 and
F ? ?
f "f "0 (from Ref. [70]).
? K

the linear flutter velocity increases almost linearly with k while the frequency decreases monotoni-
cally. The effect of the distance between the center of mass and the elastic axis shows that
º* decreases and u increases with increases in x . The radius of gyration about the elastic axis has
* ?
the effect of increasing the flutter velocity while at the same time causes a decrease in frequency for
increasing r .
?

5.2.4. Amplitude and frequency of limit cycle oscillations

The amplitudes of pitch and plunge motion of LCO have been calculated for various values of
º*/º* from Eqs. (127) and (128) by Lee et al. [70] using various methods to determine the
*
frequency u. The approximate methods they proposed depend on u, and for some cases the
agreement with numerical simulation is rather poor. The method by Liu et al. [76] using the center
manifold theory gives a much better match between analytical and numerical computations for all
values of u and airfoil parameters. In Fig. 37, the frequency u is plotted against º*/º* for u"0.2,
*
k"100, x "0.25, r "0.5 and b "3. The open circle symbol denotes u obtained from numerical
? ? ?
simulation using Eq. (24) while the solid line represents results using Eq. (153). The two methods
give extremely close results and we can draw similar conclusions by comparison with other values
of u.
Figs. 38 and 39 show the variations of the amplitudes of pitch and plunge LCO with º*/º* for
*
u"0.2, k"100, x "0.25, r "0.5 and b "3. The solid line is obtained using Eqs. (127) and
? ? ?
(128), and the numerical simulation results are given by the open circles. We observe that the two
methods give almost identical results, and other values of u also compare favorably.

5.2.5. Chaotic oscillations

The investigations carried out by Lee and LeBlanc [64] and Lee et al. [70] assumed the cubic
nonlinearity is small compared to the linear stiffness term in the equations governing the airfoil
268 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 37. Variation of u with º*/º* for pitch motion in post-Hopf-bifurcation for u"0.2, k"100, a "!0.5,
* F
x "0.25, r "0.5, f "f "0 and b "3 (from Ref. [76]).
? ? ? K ?

Fig. 38. Variation of pitch amplitude with º*/º* in post-Hopf-bifurcation for u"0.2, k"100, a "!0.5, x "0.25,
* F ?
r "0.5, f "f "0 and b "3 (from Ref. [76]).
? ? K ?

motion. For hard spring, a Hopf-bifurcation occurs at º*/º *"1 and chaotic motion was not
*
detected for the airfoil and spring parameters used. However, chaos was observed by Price et al.
[89] and Zhao and Yang [125] and they studied cases where the nonlinearity dominates over the
linear term in the restoring force or moment.
In the Price et al. [89] investigation, the nonlinear equations for the airfoil plunge and pitch
response were solved using both the finite difference and describing function methods for a range of
airfoil parameters similar to those used by Lee and LeBlanc [64]. Only results for uN "0.2, k"100,
200, a "!1/2, x "0.25, r "0.5, and f "f "0 were presented in their paper.
F ? ? ? @
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 269

Fig. 39. Variation of plunge amplitude with º*/º* in post-Hopf-bifurcation for u"0.2, k"100, a "!0.5, x "0.25,
* F ?
r "0.5, f "f "0 and b "3 (from Ref. [76]).
? ? K ?

Fig. 40. Bifurcation diagram of a cubic nonlinearity for u"0.2, k"100, a "!1/2, x "0.25, r "0.5, f "f "0,
F ? ? ? @
b "0, b "0.1, b "0, b "40 (from Ref. [89]).
   

Fig. 40 shows a typical bifurcation diagram of the pitch response for a cubic nonlinearity as
a function of velocity obtained using both the finite difference method and the describing function
technique. In this example, the constants in M(a) given by Eq. (31) is chosen as follows: b "b "0,
 
b "0.1/rad and b "40.0/rad. It is seen that the nonlinear term dominates over the linear stiffness
 
term. The finite difference results show the value of a when a"0. If two points occur at one
velocity this suggests that the motion is period-one, and four points suggests period-two. Hence, for
270 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

period-one motion the value of a shown in the figure represents the magnitude of the LCO. It
should be realized that the finite difference solution may possibly be dependent on the particular set
of initial conditions for the airfoil. Thus, this bifurcation diagram is particular to the set of initial
conditions a(0)"7.0°, a(0)"m(0)"m(0)"0.
The results presented in the figure obtained using the describing function method show the value
of a as a function of air speed; the describing function solution does not require the initial

conditions to be specified, and thus the describing function solution shown is good for any set of
initial conditions, not just those used in the numerical simulation.
It can be seen that in this case both the finite difference and describing function solutions indicate
a supercritical Hopf-bifurcation at º*/º *"0.22, giving a period-one LCO. Furthermore, for
*
0.224º*/º *40.5 approximately, the describing function and finite difference methods predict
*
essentially the same magnitude of LCO motion. However as º*/º * increases, the describing
*
function method gives an increasingly smaller prediction for the magnitude of the LCO motion
compared with the finite difference solution.
At º*/º *"0.76 the finite difference solution undergoes a further bifurcation to give period-two
*
motion, and a final bifurcation is obtained at º*/º *"0.83 giving another periodic solution. The
*
period-one motion is symmetrical about the a"0 axis. In the region 0.764º*/º *40.83
*
approximately, this symmetry is lost. Because the describing function used in this analysis assumes
period-one motion, it is not capable of predicting the bifurcations at higher velocities.
In Fig. 41a, a bifurcation diagram for k"200 with the rest of the airfoil parameters similar to
those in Fig. 40 is shown. An even stronger nonlinear spring (b "b "0, b "0.01/rad and
  
b "50.0/rad) is considered in this case. In the finite difference integration, the initial conditions

were set at a(0)"3.0°, a(0)"m(0)"m(0)°"0. In the region 0.464º*/º *40.49, approxim-
*
ately, the motion is chaotic. A Poincaré map constructed at º*/º *"0.475 shown in Fig. 41b
*
shows there is some ‘‘structure’’ indicating the motion is not random but most probably chaotic
[81]. Price et al. [89] computed the Lyapunov spectrum at different velocities for the bifurcation
diagram of Fig. 41a. In those cases where the bifurcation diagram indicated periodic motion, all of
the Lyapunov exponents were either negative or zero, in agreement with the bifurcation diagram.
However, for º*/º *"0.47, in the middle of the apparently chaotic region of the bifurcation
*

Fig. 41. (a) Bifurcation diagram of a cubic nonlinearity; (b) Poincaré map for º*/º*"0.475, u"0.2, k"200,
*
a "!1/2, x "0.25, r "0.5, f "f "0, b "0, b "0.01, b "0, b "50 (from Ref. [89]).
F ? ? ? @    
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 271

diagram shown in Fig. 41a, a positive Lyapunov exponent of 0.01 was calculated, indicating that
the system is ‘‘mildly’’ chaotic.

5.3. Freeplay with preload

Lee and Desrochers [65] studied only nonlinearities in the pitch degree of freedom. The elastic
axis of the airfoil was placed at the  chord location (that is, a "!0.5), r and x were kept
 F ? ?
constant at 0.5 and 0.25, respectively. Furthermore, they considered a bilinear spring (shown in
Fig. 2) with M "0, that is, the spring stiffness vanishes inside the freeplay d. The two properties of

the airfoil being varied were k and uN . The effects of preload and freeplay were investigated by
varying the values of M , a and d. In their report sixteen cases were considered and they
 
investigated the effect of varying the preload for constant freeplay, effect of freeplay on constant
preload, and effect of k and uN for different freeplays and preloads.
The intention of Lee and Desrochers’ [65] investigation was to re-exam the results obtained by
Woolston et al. [120, 121] using a digital instead of an analog computer. Their emphasis was on the
influence of the geometry of the freeplay on flutter boundaries and the amplitudes of the plunge and
pitch motions in the limit cycle oscillation regions. At the time of their investigation, chaotic
behavior of nonlinear aeroelastic systems was not well researched and they have not encountered
chaos or might even have inadvertently missed detecting it by not choosing the airfoil and freeplay
parameters where chaotic motion can exist. Unlike the results given by Woolston et al. [120,121],
Lee and Desrochers [65] detected islands of LCO in the damped oscillation region of the flutter
diagram.
To determine the flutter boundary, Eqs. (78) and (79) are solved for given initial conditions. In
their report, Lee and Desrochers [65] varied only the initial pitch displacement a(0) while keeping
a(0), m(0) and m(0) zero. In their procedure, they determined the linear flutter speed º * first, and
*
this is equivalent to solving the problem for M "d"a "0. In the nonlinear case, once a(0) is
 
specified, a value of º* is selected and a and m are obtained by the time marching finite difference
scheme. For the type of structural nonlinearity considered, the solution is divergent for º*'º * ,
*
and the nonlinear divergent flutter speed is the same as º * for all the cases they considered. They
*
observed that for a(0)'M decreasing º* below º * results in limit-cycle flutter. The oscillation is
 *
self-excited and maintains a constant amplitude which is self-limited. The boundaries between
divergent and limit-cycle flutter in their examples are vertical lines at º*/º *"1 for values of a(0)
*
ranging from !10° to 20°. The choice of a(0) should be such that the steady-state pitch and plunge
amplitudes are small enough for linear aerodynamics to be used. This point was discussed in
Section 5.2. Throughout their study, the value of the time step *q was taken to be 1/128 of the
shorter period of the two coupled modes of oscillation of the airfoil in the absence of aerodynamic
forces. In numerical time-integration schemes, Bathe and Wilson [7] pointed out that the ampli-
tude decays due to numerical errors and is dependent on *q. The value of *q used in Lee and
Desrochers’ study [65] is sufficiently small to give good accuracy in determining the flutter
boundaries while ensuring the computation time is not excessive [52].
As º* decreases away from º * , a value will be reached where any further decrease will result in
*
damped oscillations of the airfoil. Boundaries can be identified in the a(0) versus º*/º * plots
*
separating the regions of limit-cycle flutter with the stable regions where the airfoil motion decays
to its equilibrium condition from the initial displacement after the transients die out.
272 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 42. Flutter boundary for u"0.2, k"100, a "0.25°, d"0.5°, and M "0.25° (from Ref. [65]).
 

Fig. 42 shows the flutter boundary obtained by Lee and Desrochers [65] for k"100, uN "0.2,
M "0.25°, a "0.25° and d"0.5°. There are pockets in the damped oscillation region where the
 
airfoil oscillates with constant amplitude. These LCO regions are determined using a binary search
complemented with linear grid scans. This by no means assures that all such regions, especially the
small ones, have been identified, but those that are found can be considered to be quite accurately
defined.
Time series showing the behavior of a and m in the various regions inside the divergent flutter
boundary were given by Lee and Desrochers [65]. In the main LCO region which lies between
0.8254º*/º *41 in Fig. 42, the airfoil transients damp out fairly rapidly and the oscillations are
*
mainly sinusoidal and no noticeable harmonics are detected. Inside a LCO pocket, a strong
harmonic is always present.
Comparison of results for a freeplay d"0.5° and three values of M "a "0.25, 0.5 and 1°
 
shows the region of limit-cycle flutter decreases with increasing values of the preload. Another
observation is the non-symmetry of the boundaries between decaying oscillation and limit-cycle
flutter with initial displacement a(0). For values of initial displacement less than the preload, the
system moves on the linear part of the moment displacement curve shown in Fig. 2, and at
velocities below the linear flutter speed, the system is stable. Increasing a(0) will have a destabilizing
effect, but this is only restricted to the LCO regions. The variations of the LCO amplitudes a and

m with speed ratio º*/º * show that they are independent of the initial displacement a(0) for this
 *
freeplay and the three values of M investigated.

The pockets of LCO in the damped oscillation region are only observed for the smallest value of
M "0.25° and are not detected when M increases to 0.5 or 1° for the freeplay considered.
 
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 273

To investigate the effect of freeplay on the flutter boundary, Lee and Desrochers [65] considered
three cases where uN "0.2, k"100, M "a "0.5° and d"0.25, 0.5 and 1°. Decreasing the
 
freeplay decreases the LCO region and moves the limit-cycle flutter boundary closer to the
divergent flutter boundary. In the limit, as the freeplay tends to zero, the two flutter boundaries
coincide at º*/º *"1 which is to be expected since the system is acted on by linear spring forces.
*
Because of the preload, the flutter boundaries are not symmetrical about a(0)"0.
The LCO amplitudes a and m are independent of the initial displacement a(0), but for given
 
speed ratio, higher amplitudes are obtained when d increases. It was found by Lee and Desrochers
[65] that the flutter boundary for M "a "0.25° and d"0.5° when compared to that for
 
M "a "0.5° and d"1° is identical if the vertical scale for a(0) for the former is multiplied by
 
a factor of 2, which is the ratio of the two values of preload. Similarly, the amplitudes a and m in
 
the two cases differ also by a factor of two. From the limited results of the two cases considered,
they concluded that for a particular combination of preload and freeplay, similar results are
obtained for other values of preload and freeplay if the ratio of preload to freeplay is kept the same.
The flutter boundary curves can be made identical if the vertical scales are multiplied by factors
equal to the ratio of the preloads. Also, the amplitudes of the plunge and pitch oscillations differ
from those with other values of preload and freeplay by the same factors.
The effect of airfoil—air mass ratio was carried out for uN "0.2, M "a "0.5°, d"0.5° with
 
k"50, 100 and 250. It was found that the boundaries for limit-cycle flutter move to-
wards º*/º *"1 for increasing k. Comparison of the curves for a shows that the amplitude
* 
of pitch motion in the region between the limit-cycle and divergent flutter boundaries practic-
ally does not change with k. On the other hand, m is found to increase with k. At uN "0.8,

a decreases slightly with increasing k while the decrease in m is much larger. This behavior
 
where k has an effect on the plunge degree of freedom but with no noticeable effect on the pitch
motion is also observed for a cubic nonlinearity in the restoring moment reported by Lee and
LeBlanc [64].
Lee and Desrochers [65] also investigated the effects of uncoupled plunge to pitch natural
frequency ratio uN on flutter boundaries. When uN is close to unity, the LCO pockets disappear for
the freeplay geometry they investigated, but large irregularities at the limit-cycle flutter boundary
are detected. The amplitudes a and m also vary with uN . For the lowest value of uN "0.2
 
considered in their study, larger oscillatory motions in both degrees of freedom are observed than
those at uN "0.8.
Price et al. [88] extended the study of Lee and Desrochers [65] and investigated the dynamics of
the airfoil motion in greater details. They gave results for the case uN "0.2, k"100, M "0,

a "0.25°, d"0.5° and M "0.25°. The results they obtained are similar to Fig. 42 and they
 
reached similar conclusions that the existence of pockets of LCO was very dependent on the
freeplay and preload conditions. They carried out power spectral densities (PSDs) of the time series
and typical example are shown in Fig. 43. In Fig. 43a the pitching motion in one of the LCO
islands, at º*/º *"0.79 is presented. The motion is dominated by two frequencies at 0.335u and
* ?
0.67u . Although there are a number of other very distinct frequency peaks in the spectrum, their
?
amplitudes are small compared to the fundamental and first harmonic. The plunge motion (not
shown) has one significant peak only, at a frequency of approximately 0.34u which is close to the
?
pitch frequency and probably they are the same if the frequency can be resolved more accurately. In
the main region of the LCO motion both the pitch and plunge motion have only one dominant
274 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 43. Power spectral density of the pitch motion of the airfoil, u"0.2, k"100, a "0.25°, d"0.5°, M "0.25°,
 
a(0)"7.5°: (a) º*/º*"0.79 and (b) º*/º*"0.85 (from Ref. [88]).
* *

frequency, e.g., at º*/º *"0.85 both the plunge and pitch motion frequencies are 0.48 u . Fig. 43b
?
shows the PSD for the pitch motion.
Fig. 42 represents only a two-dimensional section through a five-dimensional boundary between
stable oscillations and LCO; the five dimensions being º*/º *, a(0), a(0), m(0) and m(0). A two-
*
dimensional section obtained by Price et al. [88] which is similar to that shown in Fig. 42 but with
a(0)"0.229b/º deg/s is shown in Fig. 44. Once again pockets of LCO are obtained below the
main LCO boundary, but the shape of both these islands and the main boundary are very different
to that shown in Fig. 42. An interesting feature in Fig. 44 is the multitude of small islands of LCO
for 0.794º*/º *40.84, approximately. If a different two-dimensional section through the
*
five-dimensional boundary is taken, the stability boundary as a function of a(0) and a(0) for
constant m(0), m(0) and º*/º * is presented in Fig. 45 for º*/º *"0.83, showing a very complex
* *
boundary between LCO and stable motion. Indeed, based on the totality of results they obtained,
Price et al. [88] concluded that the five-dimensional boundary is in fact a continuous geometric
shape with a number of protuberances that spiral out from the main body; furthermore, the
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 275

Fig. 44. Stability boundary for the airfoil as a function of initial pitch displacement, u"0.2, k"100, a "0.25°,

d"0.5°, M "0.25°, a(0)"0.229 b/º deg./s, m(0)"m(0)"0 (from Ref. [88]).


Fig. 45. Stability boundary for the airfoil as a function of a(0) and a(0) for constant velocity, u"0.2, k"100, a "0.25°,

d"0.5°, M "0.25°, m(0)"m(0)"0, º*/º*"0.83 (from Ref. [88]).
 *
276 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

apparent islands that appear in Figs. 42 and 44 are due to taking two-dimensional sections through
this five-dimensional body.
Price et al. [88] carried out a second example for uN "0.2, k"100, M "0, a "0.25°, d"0.5°
 
and M "0.0°. Note this differs from the previous example only in the value of M and in this
 
case there is no preload in the freeplay. For this set of parameters it was found that for
some velocities below the linear flutter speed, no matter how long the simulation was allowed
to run, the time histories never reached a steady-state condition (see, e.g., Fig. 46). It was
suspected that this may indicate chaos, and this was investigated initially by forming PSDs of the
time traces, a typical example of which is shown in Fig. 47. The spectrum is typical of chaotic
motion where the spectrum is mainly broadband without sharp dominant frequency peaks [81].
To add further evidence to the existence of chaos, phase-plane plots and Poincaré sections were
obtained; typical examples of which are shown in Fig. 48, for the data of Figs. 46 and 47. The
phase-plane section shown in Fig. 48a is typical of a ‘‘two-well potential’’ and is indicative of chaos.
This is very similar to the phase-plane plot obtained by Tang and Dowell [101] for their analysis
with a freeplay nonlinearity. The Poincaré section shown in Fig. 48b has a very distinct structure
that lends further evidence to the existence of chaotic motion; this is in contrast to the results
of Tang and Dowell [101] where the Poincaré section was fairly random and showed no
distinct structure. Although the results presented cannot be said to conclusively prove the
motion to be chaotic, the time histories, PSDs, phase-plane section, and Poincaré sections are all
strongly indicative of the existence of chaos, and thus, Price et al. [88] concluded that the motion is
indeed chaotic.

Fig. 46. Time history of airfoil pitch motion, º*/º*"0.3, u"0.2, k"100, a "0.25°, d"0.5°, M "0.0°, a(0)"7.0°,
*  
a(0)"m(0)"m(0)"0 (from Ref. [88]).
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 277

Fig. 47. Power spectral density of airfoil pitch motion, º*/º*"0.3, u"0.2, k"100, a "0.25°, d"0.5°, M "0.0°,
*  
a(0)"7.0°, a(0)"m(0)"m(0)"0 (from Ref. [88]).

Fig. 48. (a) Phase-plane, and (b) Poincaré map (for m"0 and m5!0.02); º*/º*"0.3, u"0.2, k"100, a "0.25°,
* 
d"0.5°, M "0.0°, a(0)"7.0°, a(0)"m(0)"m(0)"0 (from Ref. [88]).


Numerous simulations of the type discussed above were completed, over a wide range of
velocities, to determine the range of this chaotic motion; these are best illustrated via a bifurcation
diagram as shown in Fig. 49. This diagram shows, as a function of º*/º * , the value of a when
*
a"0. The significance of the bifurcation diagram is as follows. If at a particular º*/º * the system
*
is stable, then a single point is obtained, e.g., º*/º *40.13, approximately. If the motion is an
*
LCO with one frequency, then two points are obtained (0.144º*/º *40.22, approximately),
*
278 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 49. Bifurcation diagram showing a for a"0; u"0.2, k"100, a "0.25°, d"0.5°, M "0.0°, a(0)"7.0°,
 
a(0)"m(0)"m(0)"0 (from Ref. [88]).

and an LCO with two frequencies gives four points, etc. However, for some velocities a large
number of points are obtained (giving what appears to be almost a vertical line on the bifurcation
diagram) indicating chaos; examples of this can be seen in Fig. 49 for 0.294º*/º *40.33,
*
approximately. The bifurcation diagram shown in Fig. 49 suggests that the route to chaos is via
period-doubling.
Using a large number of bifurcation diagrams for different initial values of a(0), a map showing
the boundaries of the different types of motion was obtained and is presented in Fig. 50. It is
apparent that the initial value of a(0) has much less effect in this case than that obtained with
preload, as suggested by Figs. 42 and 44. Price et al. [88] have not investigated in detail the effect of
changing a(0), m(0), m(0), and did not elaborate on the manner in which the regions of chaotic
motion are affected by these parameters.
In a later study, Price et al. [89] considered a more general freeplay by considering a bilinear
spring. The properties of the nonlinearity is the same as that used to obtain the previous figures
except that M was not set to zero and a describing function was also used to analyse the airfoil

motion following the method outlined in Section 4.3.
Using the following airfoil and freeplay properties: uN "0.2, k"100, M "0.5°, a "0.25°,
 
d"0.25°, and M "0.05/rad., Price et al. [89] compared the limit cycle amplitude obtained from

the finite difference scheme and the describing function technique. Fig. 51 shows the pitch
amplitude plotted against º*/º * for a(0)"!1°, a(0)"m(0)"m(0)"0. This is a typical bifurca-
*
tion diagram where the results from the finite difference scheme are given for the value of a when
a"0. For period-one motion the value of a shown in the figure represents the magnitude of the
limit cycle oscillation. It should be realised that the finite difference solution may possibly be
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 279

Fig. 50. Approximate stability boundaries for the airfoil as a function of initial pitch displacement, u"0.2, k"100,
a "0.25°, d"0.5°, M "0.0° (from Ref. [88]).
 

Fig. 51. Comparison of limit cycle amplitude obtained from finite difference and describing function methods for the
bilinear nonlinearity; u"0.2, k"100, a "0.25°, d"0.5°, M "0.25°, M "0.05/rad.; E finite difference method for
  
a(0)"!1.0°, a(0)"m(0)"m(0)"0; — describing function method (from Ref. [89]).

dependent on the particular set of initial conditions for the airfoil. Thus, this bifurcation dia-
gram is particular to the particular set of initial conditions Price et al. [89] chose for their study.
The finite difference solution predicts that the initial loss of stability at º*/º *"0.75 is via
*
a subcritical Hopf bifurcation and, furthermore, it results in a period-two motion. This is followed
280 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

by a re-stabilization at º*/º *"0.8 and finally by the onset of period-one motion at

*
º*/º *"0.83. Thus, it is clear that there is a region of period-two LCO for velocities less than
*
those required for the main body of period-one LCO. Although not immediately apparent, closer
examination of Fig. 51 shows that the finite difference solution is not symmetric about the a"0
axis; this is due to the small preload on the airfoil in this case.
The describing function solution (see Section 4.3) gives two values of A (A and A ) each value of
 
º*/º * ; furthermore, because there is a preload, in this case B is nonzero. Hence, what is presented
*
in the figure is B#A and B#A . Although not proven because of the non-analytical nature of
 
the bilinear nonlinearity, after comparison with the finite difference solution presented in Fig. 51
and other results presented in Fig. 52, it is apparent that the larger value of A shown in Fig. 51
represents a stable LCO, while the smaller value represents an unstable LCO. The unstable LCO
gives a dividing line between stable and unstable motion: for a(0) less than this line the subsequent
motion tends back to the stable equilibrium condition, and for a(0) greater than this line the
subsequent motion tends towards the stable LCO.
The finite difference and describing function solutions presented in Fig. 51 do not agree on
the value of º*/º * at which instability first occurs, but once the period-one motion commences
*
the describing function method gives excellent agreement with the finite difference solution. As the
amplitude of oscillation increases, the effective nonlinearity of the bilinear stiffness decreases and,
hence, better agreement is obtained between the finite difference and describing function methods
at higher values of º*/º * .
*
If the finite difference solution is repeated for the same airfoil with exactly the same nonlinearity
as used for the results of Fig. 51, but with different initial conditions for the airfoil, then it is found
that whether or not LCO occur is very dependent on the particular initial conditions. This is
indicated in Fig. 52a where a two-dimensional section of the ‘‘basin of attraction’’ for LCO is
shown. The complete basin of attraction is five-dimensional, the dimensions being a(0), a(0), m(0),
m(0) and º*/º * , but the two-dimensional section shown is for constant a(0)"m(0)"m(0)"0.
*
Similar to the discussion in Fig. 42, there are apparent ‘‘islands’’ of LCO below the main body of

Fig. 52. Two-dimensional section of the basin of attraction for LCO for the bilinear nonlinearity obtained from the finite
difference method; u"0.2, k"100, a "0.25°, d"0.5°, M "0.25°, M "0.05/rad, a(0)"m(0)"m(0)"0°. (a) general
  
view; (b) enlarged view for a smaller region of a(0) and º*/º* ; E describing function solution (from Ref. [89]).
*
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 281

LCO for some values of a(0); it is one of these islands of instability which gives the period-two
motion for 0.754º*/º *40.83 shown in Fig. 51. This is in agreement with the observations from
*
Fig. 42 with a free-play nonlinearity. For both the free-play and bilinear nonlinearities the LCO in
the islands of instability is period-two; hence, it is not surprising that the agreement between the
finite difference and describing function solutions in these islands is not as good as elsewhere.
As shown in Fig. 52a, for small initial values of a(0), typically in the range !0.5°4a(0)40.5°
instability is not obtained for º*/º *41. Qualitatively, this is in agreement with the unstable
*
branch of the describing function solution presented in Fig. 51 where the describing function
solution from Fig. 51 is superimposed on the results. An enlarged view of the results for small a(0) is
shown in Fig. 52b. The solutions B!A and B!A are presented in addition to B#A and
  
B#A which are shown in Fig. 51. As shown, there is reasonable, but not exact, agreement

between the unstable LCO branch of the describing function method and the finite difference
solution. Quite possibly the agreement between these two solutions for the unstable LCO motion
could be improved if the finite difference solution were given for other values of a(0), m(0) and m(0).
All of the results presented so far by Lee and Desrochers [65] and Price et al. [88] have been for
zero structural damping, but it is of interest to see what the effect of structural damping is on the
aeroelastic response of the airfoil. Fig. 53 shows three bifurcation diagrams for the airfoil with
a bilinear nonlinearity with the following parameters: uN "0.2, k"100, M "!0.0025°,

a "0.25°, d"0.5°, and M "0.01/rad with initial conditions a(0)"3°, a(0)"m(0)"m(0)"0.
 

Fig. 53. Bifurcation diagrams showing a for a"0 as a function of º*/º* for u"0.2, k"100, a "!0.25°, d"0.5°,
* 
M "!0.0025°, M "0.01/rad, a(0)"3°, a(0)"m(0)"m(0)"0. (a) f "f "0.02; (b) f "f "0.05; (c) f "f "0.1
  ? K ? K ? K
(from Ref. [89]).
282 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Three different values of structural damping f and f were used. The bifurcation diagram shown in
? K
Fig. 53a with 2% critical structural damping for both f and f is very similar to that with zero
? K
structural damping. It is seen in the range of velocity 0.324º*/º *40.53 there is a very large
*
number of points in the bifurcation diagram at any particular velocity, suggesting non-periodic
motion. Interestingly, as the velocity is increased beyond this nonperiodic region and approaches
the divergent flutter condition, the airfoil motion goes from being non-periodic (probably chaotic)
to period-four, then period-two and finally period-one. Increasing the structural damping f and f
? K
to 5% of critical reduces the regions of apparently chaotic motion as shown in Fig. 53b, and for
10% of critical damping it is clear that the airfoil motion is now periodic as shown in Fig. 53c.
Thus, it seems that a reasonably large amount of structural damping can eliminate the chaotic
motion of the airfoil which is present for small amounts of structural damping.
Price et al. [89] concluded that where the LCO is period-one there is good agreement between
the finite difference and describing function solutions. However, by its very nature, the describing
function solution used here is not capable of predicting the higher order periodic motion unless
modifications are made following the approach by Johnson [51] who took into consideration the
next higher harmonic term.
Alighanbari and Price [1] carried out a bifurcation analysis of the airfoil using the same airfoil
parameters and freeplay properties in the earlier studies by Price et al. [88,89]. They used Eqs. (27)
and (28) and approximated the freeplay by a third-order rational curve fit given in Eq. (167). The
freeplay they considered has the properties M "a "0.25°, M "0 and d"0.5°. This gives values
  
of c "0.00021, c "0.9277, c "!134.7957, c "5954.619, c "!121.2787, c "6414.885 and
     
c "1064.4611 for the constants in Eq. (167). The rest of the airfoil parameters are the same as

those used to derive Fig. 42. The nonlinear equations of motion have been analyzed for different
cases using both the Runge—Kutta method and AUTO [23].
The fixed points of the system can be evaluated by solving X"0 in Eq. (138). By using the
curve-fit results for the freeplay, Alighanbari and Price [1] showed that the fixed point occurs at
a"!0.0012° and m"0.0001º*. The difference from the true fixed point at a"m"0 is small
and is due to the error in the approximated third-order rational curve fit.
Stability of the fixed point is analyzed by linearizing the dynamic equations of the system about
this point. The eigenvalue analysis of the linearized equations shows that the system is stable for
º*/º *(0.98 and unstable for higher velocities. Two purely imaginary eigenvalues at
*
º*/º *"0.98 are an indication of a Hopf-bifurcation. For the exact freeplay the Hopf-bifurcation
*
occurs at º*/º *"1; again the difference is small and is due to the approximate curve fit of the
*
true freeplay.
A bifurcation diagram, evaluated using AUTO [23], showing both the stable and unstable
solutions is presented in Fig. 54. In this diagram the horizontal axis is the bifurcation parameter
º*/º * and the vertical axis is the maximum value of the pitch angle a. For a stationary solution
*
a is plotted and for a periodic solution the maximum values of a(q) during each period of oscillation
are presented. The origin is a stable fixed point until º*/º *"0.98, point 1, which is near the
*
linear flutter speed. As predicted by the eigenvalue analysis, at point 1 a Hopf-bifurcation makes
the origin unstable and an unstable period-one limit cycle solution begins showing that the
Hopf-bifurcation is subcritical. The branch changes its direction at a limit point 2, and finally, this
unstable period-one solution becomes stable at point 3, with amplitudes increasing to infinity when
º* approaches the linear flutter velocity. At point 3 a Floquet multiplier is equal to !1; this
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 283

Fig. 54. (a) Bifurcation diagrams showing both stable and unstable solutions for pitch motion obtained using AUTO,
(b) expanded view of partial region from (a). u"0.2, a "0.25°, d"0.5°, M "0.25°. E stable fixed point; 䊊 unstable
 
fixed point; — stable limit cycle, - - - unstable limit cycle (from Ref. [1] ).

indicates a period doubling bifurcation. A period-two branch also starts at point 3 (branch 3-3) but
does not lead to any stable solutions. However, another unstable branch, 3-2, starts at point 3, is
attracted to branch 2-3 and joins it at point 2.
Other periodic solutions are obtained by starting the process from known stable limit cycles,
points 4 and 5, evaluated using the Runge—Kutta method for the same nonlinearity. Two separate
islands of period-two and period-four stable and unstable solutions are found, as shown in Fig. 54a
and b. Since some solutions separate from the main body of periodic oscillations have been found,
there is the possibility that even more solutions, stable or unstable, may exist but have not been
detected in this analysis. The corresponding bifurcation diagram for plunge motion is shown in
Fig. 55. As can be seen, islands of periodic solutions also exist for plunge motion, but the limit
cycles are confined to being either period-one or period-two, unlike the pitch response where
period-four motion was also obtained.
Fig. 56 represents the bifurcation diagrams obtained using the Runge—Kutta method for the
same parameters as Fig. 54, along with the results given by AUTO previously presented in Fig. 54.
The results presented from the Runge—Kutta method show the value of a when a"0 for the
particular initial conditions given by a(0)"8° and a(0)"m(0)"m(0)"0. Both methods predict
essentially the same magnitude of stable limit cycle motion. However, in order to completely
determine all possible branches of the stable solution the Runge—Kutta method should be repeated
with a large number of different initial conditions and the total set of results considered. In the
range of air-speed around º*/º *"0.78, the Runge—Kutta results indicate two points at each
*
º*/º * ; it should be noted that both points correspond to only one limit cycle, and this suggests
*
period-two motion. AUTO also predicts a period-two limit cycle, but it shows only one point
which is the maximum amplitude of the oscillation.
The bifurcation diagrams obtained for this nonlinearity are in good agreement with the finite
different results previously presented by Price et al. [88] for the true freeplay nonlinearity. Using
the approximate equation for the freeplay, complete branches of stable and unstable solutions
could be determined using Alighanbari and Price’s [1] analysis, and the type of bifurcations could
also be determined via calculation of the Floquet multipliers. The results given by Price et al. [88]
284 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 55. Bifurcation diagrams showing both stable and unstable solutions for plunge motion obtained using AUTO.
u"0.2, a "0.25°, d"0.5°, M "0.25°. E stable fixed point; 䊊 unstable fixed point; — stable limit cycle, - - - unstable
 
limit cycle (from Ref. [1] ).

Fig. 56. Comparison of the bifurcation diagrams obtained using Runge—Kutta method and AUTO. u"0.2, a "0.25°,

d"0.5°, M "0.25°. * Runge—Kutta results; E AUTO (stable fixed point); 䊊 AUTO (unstable fixed point); — AUTO

(stable limit cycle); - - - AUTO (unstable limit cycle) (from Ref. [1] ).

show only stable branches of the solution and the specific type of bifurcation could not be
inferred.
Alighanbari and Price [1] also considered a freeplay nonlinearity without preload given
by M "M "0, a "0.25°, and d"0.5°. This gives values of c "!0.00422, c "1.6164,
    
c "!194.6997, c "7436.942, c "!143.1963, c "8207.7659 and c "!175.107. There are
    
now three fixed points obtained by solving the equation X"0. They are: a"0.691°,
m"!0.00603º*; a"0.309°, m"!0.00270º*; and a"0.500°, m"!0.00436º*. The stabi-
lity of these fixed points can be analyzed by linearizing the equations of motion about them. The
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 285

Fig. 57. (a) Bifurcation diagrams showing both stable and unstable solutions for pitch motion using AUTO, (b) expanded
view of branch 6-7-8, (c) expanded view 12—13. u"0.2, a "0.25°, d"0.5°, M "0.25° (from Ref. [1]).
 

eigenvalue analysis of the linearized equations shows that the first and second fixed points are
stable for velocities below º*/º *"0.529 and that they are unstable for higher velocities.
*
However, the third fixed point is unstable for º*/º *50.
*
Fig. 57 shows the bifurcation diagram constructed using AUTO for this nonlinearity. Three fixed
points are shown in Fig. 57a, at low velocities two of these are stable and one unstable. As predicted
by the eigenvalue analysis increasing º*/º * causes both of the stable fixed points to become
*
unstable via Hopf-bifurcations at points 1 and 2, and the other fixed point remains unstable.
Unstable periodic solutions are initiated at these bifurcation points which indicates a subcritical
type of instability. However, these branches of periodic solutions do not lead to any further stable
solutions, and after suggesting some bifurcations these branches cease to exist at points 3 and 4
of Fig. 57a.
Other branches are found by initiating the procedure from either of the two stable periodic
solutions, given by points 5 and 12, which were obtained from the Runge—Kutta method. Starting
from point 5 the amplitude of the stable periodic solution increases and approaches infinity as
º*/º * approaches unity. However, by decreasing º*/º * the system undergoes more bifurca-
* *
tions. In an attempt to clarify the results this part of the solution, branch 6-7-8-9, is presented
separately in Fig. 57b. The first bifurcation occurs at point 6; one of the Floquet multipliers crosses
the unit circle at #1 and the periodic solution does not merely become unstable, instead it
286 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

disappears entirely indicating a saddle node bifurcation [13]. The unstable solution originating
from point 6 stabilizes after a limit point at 7. The results shown in Fig. 57b indicate two bifurcation
points on the branch 7-8-9. Crossing the unit circle at !1 the Floquet multipliers show a period
doubling bifurcation at point 8. An unstable branch of period-four motion starts from this point,
branch 8-10, indicating a subcritical (indirect) type of period doubling at point 8. The main branch,
7-8-9, which is unstable between 8 and 9, is restabilized after another period doubling bifurcation at
point 9; contrary to the previous bifurcation at point 8, a new stable branch of period-four motion
starts from point 9 showing a supercritical (direct) period doubling. This new stable branch
of period-four motion, becomes unstable after a saddle-node bifurcation at point 11. The period-
four branches of 8-10 and 9-11 are also connected through other unstable solutions, as shown in
Fig. 57b.
The Runge—Kutta results also give another stable periodic solution at º*/º *"0.7. Starting
*
from this point, labelled 12 in Fig. 57a, more branches of stable and unstable periodic solutions are
evaluated; these are presented in Fig. 57c. The branch 12-13-14 shows inverse and direct period
doubling bifurcations, labeled 13 and 14, respectively, at the same air speeds as branch 7-8-9 shown
in Fig. 57b. The period-four branches 13-15 and 14-16 are also connected via unstable solutions as
shown in Fig. 57c.
The results presented in Fig. 57 show that a large number of stable periodic solutions, with no
stable limit cycles, are given for 0.36(º*/º *(0.44. On the other hand there is no possibility of
*
divergent flutter in this velocity range since º*/º *(1. These multiple unstable solutions indicate
*
the possibility of chaotic oscillations.
Alighanbari and Price [1] have shown for those cases where the limit cycles are stable
Runge—Kutta integration and AUTO give virtually identical results, both in terms of the velocities
at which the bifurcations occur and the magnitude of the limit cycles. However, only AUTO is
capable of predicting unstable solutions. By using periodic solutions obtained from the
Runge—Kutta numerical integration as starting solutions for AUTO many stable and unstable
periodic solutions are detected, resulting in extremely complex bifurcation diagrams.

5.4. Hysteresis nonlinearity

The earliest attempt to study hysteresis effects in aeroelastic systems was carried out by
Woolston et al. [120,121] using an analog computer to determine flutter of a two-dimensional
airfoil oscillating in 2- and 3-DOF with incompressible indicial aerodynamics. Shen and Hsu [95]
and Shen [96,97] studied the approximate behavior of nonlinear flutter problems by the describing
function method and re-examined a few examples of a wing-control surface flutter with nonlinear
structural stiffness previously analyzed by Woolston et al. [120,121]. They also considered the
bending-torsion-control surface flutter of a two-dimensional airfoil-flap system in incompressible
flow. The describing function method was also used by Breitbach [18,19] to analyse aircraft
structures with hysteresis. Little numerical simulation work has been carried out except for the
recent study by Chan [21].
Fig. 11 shows a symmetrical structural hysteresis nonlinearity that can be represented by the
superposition of two free-plays. For increasing a, the notations follow that used to describe
a freeplay given in Fig. 2b. For decreasing a, the freeplay starts at !a . The loop has a flat-spot
$
with width d between a 4a4a #d and !a 4a4!a !d. We can generalize the hysteresis
$ $ $ $
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 287

by assigning an arbitrary slope to the flat-spot. In Chan’s [21] analysis, the slope was set to zero,
and to preserve symmetry, the line joining the mid-points of the flat-spots on the upper and lower
branches of the loop must pass through the origin of the moment versus displacement diagram.
The equations of motion of the airfoil are given in Eqs. (19) and (20). With the initial values a(0),
a(0), m(0), m(0) specified, these two equations were integrated numerically by Chan [21] using
a fourth-order Runge—Kutta scheme. Only a hysteresis nonlinearity in the pitch degree of freedom
was considered and the investigation was limited to an airfoil with a "!1/2, r "0.5, x "0.25.
F ? ?
The other airfoil parameters and hysteresis geometry were varied according to Table 2.
A typical displacement curve a in the pitch degree-of-freedom versus nondimensional time q is
shown in Fig. 58 for the hysteresis considered in Case 13 of Table 2. The value of º*/º* is 0.875
*
which gives an amplitude of pitch oscillation of 3.2° approximately. Only five cycles of the curve are
shown from time between 2000 and 2400 where the motion has long reached steady state. The
moment curve is also shown and it is in phase with the displacement. The constant values in
the M(a) curve corresponds to those instances when the displacement reaches the flat-spots on the
hysteresis loop in the directions of increasing and decreasing a respectively. Since d"1°, we see
that the discontinuities in the moment curve occur between 0°(a(1.0° and !1.0°(a(0°
when projected on the displacement curve.
Fig. 59 shows the flutter boundary for Case 13 in Table 2. The initial conditions are
a(0)"m(0)"m(0) and a(0) varies between the range $20°. This value of a(0) is probably too large
and at small values of time may invalidate the linear aerodynamic assumptions of Section 3.1.
The remarks made in Section 5.2 concerning a(0) also apply here.

Table 2
Case studies from Chan [21]

Case k u a (deg) d (deg) M (deg) M /d º*

   *
1 50 0.2 0.0 2.5 1.25 0.5 4.525
2 50 0.2 0.5 1.5 1.25 0.833 4.525
3 50 0.2 1.0 0.5 1.25 2.5 4.525
4 50 0.8 1.0 0.5 1.25 2.5 3.074
5 100 0.2 0.0 2.5 1.25 0.5 6.285
6 100 0.2 0.25 0.5 0.25 0.5 6.285
7 100 0.2 0.5 1.5 1.25 0.833 6.285
8 100 0.2 1.0 0.5 1.25 2.5 6.285
9 100 0.8 0.0 2.5 1.25 0.5 4.114
10 100 0.8 0.5 1.5 1.25 0.833 4.114
11 100 0.8 1.0 0.5 1.25 2.5 4.114
12 100 0.2 0.0 0.5 0.25 0.5 6.285
13 100 0.2 0.0 1.0 0.5 0.5 6.285
14 100 0.2 !0.125 0.5 0.125 0.25 6.285
15 100 0.2 !0.25 1.0 0.25 0.25 6.285
16 250 0.2 0.0 2.5 1.25 0.5 9.710
17 250 0.2 0.5 1.5 1.25 0.833 9.710
18 250 0.2 1.0 0.5 1.25 2.5 9.710
19 250 0.8 1.0 0.5 1.25 2.5 5.962
288 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 58. Time variation of a and M(a) with q for Case 13 in Table 2 (from Ref. [21]).

Fig. 59. Flutter boundary for a hysteresis structural nonlinearity for Case 13 in Table 2 (from Ref. [21]).

The grid size used in obtaining the flutter boundary is *(º*/º*)"0.01 and *a(0)"1°. A time
*
step *q"0.025 is used in all the numerical simulations. Other values were tried but this time step
was chosen based on efficiency and accuracy [52]. This type of flutter boundary diagram is similar
to those shown for a freeplay nonlinearity. Note that the diagram is symmetrical about a(0)"0.
Comparisons with Woolston et al. [120] results show the flutter boundary to be similar, but
the numerical simulations give more details on the boundary curves and the existence of
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 289

isolated pockets of LCO inside the main flutter boundary which was not observed by Woolston
et al. [120].
The grid used to generate Fig. 59 is reasonably small and most of the LCO regions should have
been identified with sufficient accuracy. To ensure that none has been missed, we re-examined the
region 0.754º*/º*40.9 and !20°(a(0)40° and used a finer grid to locate the LCO regions
*
as shown in Fig. 60. The spacings are *(º*/º*)"0.00125 and *a(0)"0.2°. Comparisons with
*
Fig. 59 shows that the coarser grid gives sufficiently accurate results.
The flutter boundary shows that for º*/º*'1 the oscillations become divergent. The divergent
*
rate usually increases with increasing values of º*/º*. Fig. 61 shows a typical time series for the
*
pitch DOF at º*/º*"1.001 and a(0)"1.0° for Case 13 in Table 2. Flutter is gradual and not
*
explosive in character like those for higher values of º*/º*.
*
In the LCO region we show in Fig. 62 a typical time series at º*/º*"0.95 and a(0)"10°. The
*
transients die out in approximately 2 cycles and the pitch angle reaches an amplitude of 7.5°. The
signal is seen to be close to a sinusoid with a dominating fundamental and possibly very weak
higher harmonics for some values of º*/º* approaching the convergent region of the flutter
*
boundary. This shows that the assumption of a dominating frequency in the harmonic balance or
describing function technique is a good approximation in this LCO region.
Inside one of the pockets between 0.734º*/º*40.85 and $20°4a(0)4$15°, the charac-
*
teristics of the LCO are different from those in the main LCO region. The amplitudes are smaller
and the oscillation is periodic but the presence of higher harmonics is clearly noticeable. This has
been shown to be the case from power spectral density plots of the time series. Fig. 63 gives a time
series for º*/º*"0.775 and a(0)"!17.5°. Steady state is reached after approximately 2 cycles
*
and the amplitude is lower than those in the main LCO region with an amplitude of 2.5° in this
case. In this region, the assumption in Shen’s [96] harmonic balance analysis is not good and
higher approximations such as sinusoidal method describe by Johnson [51] should be used.

Fig. 60. Enlarged view of the flutter boundary of Fig. 59 (from Ref. [21]).
290 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 61. Typical time series showing divergent oscillations (from Ref. [21]).

Fig. 62. Typical time series showing limit cycle oscillations in the main LCO region (from Ref. [21]).

The final time series we show in Fig. 64 corresponds to a condition inside the convergent region
where º*/º*"0.675 and a(0)"12°. It takes approximately 8—10 cycles for the motion to decay
*
to stationary values. The rate of decay depends on the velocity ratio º*/º*.
*
A plot of the pitch amplitude versus velocity ratio is shown in Fig. 65. Notice that at
º*/º*"0.81, there is a small decrease in the amplitude curve. On examining the curve closely,
*
we see that there are actually two curves for the range 0.7354º*/º*40.81 and 0.814
*
º*/º*41.0. It is interesting to note that the amplitude is independent of initial condition a(0).
*
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 291

Fig. 63. Typical time series showing limit cycle oscillations in a LCO pocket (from Ref. [21]).

Fig. 64. Typical time series showing convergent oscillations (from Ref. [21]).

If we compare this figure with the results from a harmonic balance analysis such as that given by
Shen [96], we found the shape of the curves to be similar. In Shen’s [96] results, a was used instead

of a(0) in plotting the flutter boundary curve. The reader should be cautioned that for
º*/º*'0.975, the amplitude a is greater than 10—15° which is the limit where linear aerodyna-
* 
mics theory should be used.
The frequency u is plotted against velocity ratio in Fig. 66. The two regions 0.7354º*/º*
*
40.81 and 0.814º*/º*41.0 detected in the previous figure are quite distinct and the variations
*
of u with velocity ratio are very small for 0.7354º*/º*40.81. Similar to the amplitude, the
*
frequency is independent of initial conditions a(0). In the range 0.814º*/º*41.0, u increases
*
with velocity ratio quite rapidly.
292 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 65. Pitch amplitude versus velocity ratio for Case 13 in Table 2 (from Ref. [21]).

Fig. 66. Frequency versus velocity ratio for Case 13 in Table 2 (from Ref. [21]).

From the case studied by Price et al. [89] for a freeplay with preload having geometrical
properties given by a "0.25°, d"0.5°, and M "0, we can construct a non-symmetrical hyster-
 
esis loop where a is 0.25° when M "0. By nonsymmetrical we mean that the linear arm of the
 
hysteresis loop for small M(a) does not pass through the origin ‘0’. We see from Price et al. [89]
results that there are regions of chaotic behavior at approximately 0.25(º*/º*(0.33 and
*
0.46(º*/º*(0.49 for !10°4a(0)410° with a(0)"m(0)"m(0). If we construct hysteresis
*
loops based on that freeplay by increasing M , we can study the effect of M on the types of
 
nonlinear phenomena that are observed, namely, convergent motions, LCO and chaos. In Fig. 67
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 293

Fig. 67. Chaotic regions for a nonsymmetrical structural hysteresis nonlinearity (from Ref. [21]).

we plot M /d versus º*/º* and choose a(0)"7° for all the numerical simulations. For fixed
 *
values of M /d and º*/º* a power spectral density plot from the time series was computed. From
 *
the shape of the power spectra, the oscillation was considered chaotic if no distinct or dominating
peaks are detected in an otherwise broadband spectrum (see, for example, Price et al. [89]). The
figure shows that a hysteresis nonlinearity with small values of M /d for this particular set of airfoil

parameters enlarges the regions of chaotic motion beyond those for the freeplay studied by Price
et al. [89]. This figure was obtained by choosing a region 04M /d40.13 and 0.14º*/

º*40.75. A grid was constructed using *(º*/º*)"0.05 and *(M /d)"0.005, and at each grid
* * 
point the motion was identified from the power spectral density plot to be either convergent, LCO
or chaotic. We see that chaos is mainly confined to 04M /d40.13 and 0.24º*/º*40.56
 *
using a rather coarse velocity increment in the present studies. This range is larger than that
reported by Price et al. [89] for a freeplay and as M increases from 0 to higher values, the two

chaotic regions in the freeplay merged. For larger values of M /d'0.03, chaos is encountered only

at º*/º*"0.55 and 0.6 for limited values of M /d. No numerical simulations were carried out
* 
beyond M /d'0.13 and it is suspected that chaos may not be found when the value of M /d for
 
the hysteresis loop is sufficiently large. For º*/º*'0.15, the figure shows that the decaying
*
oscillation region decreases with M /d. When º*/º* reaches values greater than 0.45 the motion is
 *
mainly of the LCO type except for the chaotic cases. There are, however, a few conditions where the
decaying solution was found and this cannot be explained.
No other values of initial conditions aside from a(0)"7° were used in the computations. The
dependence on initial conditions was not investigated and we may expect the chaotic regions to
vary with a(0), m(0) and their derivatives.
To investigate the effects of airfoil parameters, we vary a , d, u, M and k. In order that
 
the hysteresis loop is symmetrical about the line M(a)"ka, a and M cannot be varied
 
independently.
294 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 68. Effect of flat-spot on flutter boundary (from Ref. [21]).

Fig. 68 shows the effect of d on the flutter boundaries for Cases 5, 7 and 8 in Table 2. The
frequency ratio is kept at 0.2 and M "1.25, d"0.5, 1.5 and 2.5 while a varied in order that M
  
can be kept constant. For d"0.5 (M /d"2.5) which has the smallest flat-spot (Case 8), the flutter

boundary curve shows a typical behavior of the variations of the boundary with initial conditions
a(0) while keeping a(0)"m(0)"m(0)"0. At a velocity ratio equal to one, the boundary is vertical
with increasing a(0) until it reaches a value of 2.5°. It then moves in the direction of decreasing
º*/º* with a small initial slope that increases rapidly until º*/º* reaches approximately 0.93
* *
when the boundary line becomes almost vertical. The pitch amplitude increases from 2.25° at
º*/º*"0.93 to 10° at º*/º*"0.98 while the corresponding plunge amplitude changes from
* *
0.156 at º*/º*"0.93 and increases rapidly to a value of 0.375 at º*/º*"0.98.
* *
Increasing the value of d to 1.5 (M /d"0.833), we have for Case 7 a flutter boundary that

resembles Case 8. There is a small indent at º*/º*"1 and a(0) approximately 2°. The conver-
*
gent/LCO boundary is almost vertical at º*/º* approximately 0.855 and a(0)"7°.
*
For Case 5 where d"2.5° and M /d"0.5 the boundary has a more complex structure than

those with smaller values of the flat-spot. The boundary between divergent and convergent
oscillations remains at a(0)"$2° as in the last case. The indent observed at d"1.5° grows to
º*/º*"0.9, and the point where the convergent/LCO boundary starts to become almost vertical
*
has now moved to º*/º*"0.825 and a(0)"6°.
*
If the loop is not symmetrical, Chan [21] compared Case 6 flutter boundary with that for
Case 8 (d"0.5°) and showed that off-setting the hysteresis by 0.25° and changing M to be 0.25

(M /d"0.5 vs. 2.5 for Case 8) the flutter boundary becomes non-symmetrical. The boundary

between divergent and convergent oscillation near º*/º*"1 is very small, and the almost vertical
*
boundary separating decaying oscillation and LCO moved to º*/º*"0.825 instead of 0.93. The
*
plunge amplitude is small except near º*/º*"1 where comparisons are difficult since the slope of
*
the amplitude—velocity curve is very large. The pitch amplitude is similar for the two cases in the
range 0.93(º*/º*41.
*
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 295

Fig. 69. Effect of k on flutter boundary (from Ref. [21]).

Fig. 69 shows the effect of k on the flutter boundary for Cases 1, 5, 16. At this value of d where the
value of the flat spot is the largest used in the numerical simulations (d"2.5° and M /d"0.5) we

found that at the lowest value of k"50 where the airfoil is the lightest, the flutter boundary is
similar to that at k"100 (Case 5). The difference lies mainly in the vertical portion of the
convergent/LCO boundary for a(0)'6° where the lower k results show a smoother boundary.
Chan [21] found that the pitch amplitude for k"50 is slightly below that at k"100 for
º*/º*(0.9. The plunge amplitude is noticeably lower and the difference can be as large as 25%
*
at º*/º*"0.825. Increasing k to 250, the flutter boundary is much more complex where decaying
*
oscillations exist in narrow regions extending into the LCO regions. The boundary between
decaying oscillations and LCO for a(0)'6° decreases slightly for the heavier airfoil. The ampli-
tudes of pitch motion show three distinct regions. For 0.854º*/º*41, the curve is similar to
*
those for the two lower values of k except that it has slightly higher values. Between
0.764º*/º*40.85, the amplitude is slightly higher than those extrapolated from the curve
*
between 0.854º*/º*41. The curve between 0.7054º*/º*40.76 shows a small increase with
* *
º*/º* initially, followed by a decrease as º*/º* approaches 0.76. The plunge amplitudes [21]
* *
also show the existence of these three regions and the amplitudes are very large, varying between
0.6 and 0.75 for the three regions, which are much greater than those for the other two values of k. It
is found that for the heavier airfoil, the plunge is the more dominant DOF. The time series from
these different regions especially those at the smaller values of º*/º* usually show the presence of
*
higher harmonics with fairly large amplitudes.
Fig. 70 shows the effect of u on the flutter boundary for Case 7 and 10 where d"1.5°. Increasing
u alters the bulge which starts at a(0)"2° and moves it to a different location at º*/º*"0.92
*
and a(0)"9°. Above a(0)"9° the width of the LCO region is slightly larger at u"0.8 where
º*/º* has moved from 0.875 to 0.865. The pitch amplitude curve [21] is slightly higher at the
*
lower frequency ratio while the plunge amplitude is again very large at u"0.2 compared to those
at u"0.8 (e.g. 0.2875 compared with 0.0188 at º*/º*"0.87).
*
296 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 70. Effect of u on flutter boundary (from Ref. [21]).

For the largest value of the flat spot at d"2.5°, Cases 5 and 9 show that increasing the frequency
ratio changed the flutter boundary substantially. At the lower frequency ratio, the boundary curve
between convergent and LCO is not vertical but has a wavy profile. There exists also a pocket of
LCO when 2.5(a(0)(6° and there is a bulge with a maximum peak extending to º*/º*"0.97.
*
At u"0.8, these features are not present and the flutter boundary is quite smooth. The pitch
amplitudes are larger for the smaller frequency ratio, the difference being greater at the lower values
of º*/º*. The plunge amplitudes show large reductions by increasing u to 0.8. For example, at
*
º*/º*"0.85, m decreases from 0.406 to 0.025 when u increases from 0.2 to 0.8. This difference
* 
increases with increasing º*/º*.
*
For the smallest flat-spot studied in Cases 8 and 11 for d"0.5°, the lower frequency ratio shows
the flutter boundary to be fairly smooth. The boundary between divergent/convergent flutter is
a vertical line at º*/º*"1 and terminates at a(0)"2.5°. It decreases gradually with º*/º* until
* *
at a value of 0.93 it becomes almost a vertical line where º*/º* hardly changes with a(0).
*
Increasing u to u"0.8, Chan [21] showed that the LCO region is divided into 2 regions with one
between 0.9854º*/º*41 and the other starting at a(0)"9° and º*/º*"0.92 which asymp-
* *
totically reaches the left hand boundary of the first region at º*/º* approximate 0.985.
*

5.5. Forced oscillations

In the previous sections, the airfoil motion has been analysed for three common types of
structural nonlinearities in the absence of external excitation forces. In other words, we considered
only self-excited oscillatory motions. When P(q) and Q(q) in Eqs. (10) and (11) are not zero, the
motion is dependent on the forcing frequency u of the applied force or moment. In aeroelasticity,
forced motion has not been extensively studied, and even the simpler case of a coupled nonlinear
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 297

mechanical system has not been analysed thoroughly. Some recent studies by Lee et al. [69] and
Gong et al. [36] show the complexity of such coupled mechanical systems.
A numerical simulation of the forced oscillation of a two-dimensional airfoil was carried out by
Lee et al. [69] for incompressible aerodynamics with cubic nonlinear restoring forces of the form
given by G(m)"m#b m and M(a)"a#b a in Eqs. (19) and (20). Lee et al. [69] used the
K ?
following airfoil parameters: x "0.2, r "1, º*"1, u"0.8944, f "0.05, f "0, b "0.1, and
? ? ? K ?
b "0.125. These were chosen to coincide with a set of differential equations that the authors have
K
previously studied [119].
It is useful to have some indication of the airfoil response in the absence of aerodynamic forces
for comparison purposes. In this case, the governing equations become a coupled system of forced
Duffing’s equations. Fig. 71 shows the complex structure of the amplitude response curve R of a(q)
in degrees plotted against the excitation frequency u for F"1.0, where F is defined in Eq. (111).
Using the analysis given in Section 4.4, Lee et al. [69] found that at u"1.5 there are seven
equilibrium points. Once these values are computed, the stability can be determined from the sign
of the real part of all the eigenvalues, and only three equilibrium points are stable according to the
linear stability analysis. Gong et al. [36] showed that for a certain range of system parameters, the
forced couple oscillators admit both stable harmonic and stable quasi-periodic motions at the same
frequency value.
A detailed numerical study was also carried out for this particular example to demonstrate some
of the features of the coupled system. The existence of multi-valued harmonic solutions is
confirmed by a numerical simulation using Eqs. (24) for different initial conditions.
To investigate the effects of aerodynamics in Eqs. (19) and (20), Lee et al. [69] choose a "!1/2
F
and k"25. Other values of a and k can be used since there are no restrictions on the parameters in
F
the analysis. In order for the linear aerodynamics to be valid, the value of the forcing function F has
to be kept small so that the pitch angle is less than 10—15° while at the same time the plunge motion
m is within the range for Eqs. (15) and (16) to be applicable. Using these airfoil parameters, the

Fig. 71. Pitch amplitude response curve of a two-degree-of-freedom system with F"1.0 (from Ref. [69]).
298 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

linear flutter speed is obtained by setting the nonlinear stiffness terms to be zero, and its value º* is
*
found to be 2.423. Two values of º* were used by Lee et al. [69], one at 41% and the other at 95%
of º*, that is, º*"1 and 2.3019 approximately.
*
The a-amplitude—frequency curve is shown in Fig. 72 for º*"1 and F"0.03. The case without
consideration of aerodynamics is shown in Fig. 73. Both figures show the presence of two peaks at

Fig. 72. Pitch amplitude response curve of a two-degree-of-freedom system for k"25, º*"1 in a moving medium
(from Ref. [69]).

Fig. 73. Pitch amplitude response curve of a two-degree-of-freedom system for k"25, º*"1 in a stationary medium
(from Ref. [69]).
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 299

u approximately 0.86 and 1.07. These frequencies can be estimated from Eqs. (19) and (20) since by
decoupling the two equations, we can show that u "1/º* and u "u u by definition. For
? K ?
º*"1 and u"0.8944 in this example, we obtain u "0.89 and u "1 which are close to the
K ?
observed values in the figures. The numerical simulation results are also included in the figure and
they are denoted by the open circles. It is seen that good agreement is obtained. Comparison with
Fig. 71 shows that the complex jump condition is not encountered. However, when the forcing
amplitude F is increased to larger values, these jumps are detected, but the results are invalid since
the amplitudes are very large and violated the linear aerodynamics assumptions.
Increasing º* to 0.95º*, Fig. 74 shows the disappearance of one of the modes. It is seen from
*
this figure that the peak occurs at u+0.43. For the uncoupled modes, the natural frequencies u
?
and u are estimated to be 0.4344 and 0.3885, respectively for º*"2.3019 and u"0.8944. The
K
forcing amplitude in this case is F"0.002 which is very small. However, this is expected since only
a weak external forcing to the airfoil is required in order to maintain a harmonic motion of fairly
large amplitude as the linear flutter speed is approached. Also, near the flutter boundary,
coalescence of the two modes accounts for the observation of only one peak in Fig. 74.
For the cubic nonlinearity studied by Lee et al. [69], only harmonic motions were observed.
However, if the ratio b /b is large so that the cubic term dominates over the linear term in Eq. (31)
 
while keeping b "b "0 to obtain the form of the nonlinearity studied by Lee et al. [69], the
 
motion may be more complex. The discussion in Section 5.2 on cubic restoring force gives examples
of chaotic motions from Price et al. [89] and Zhao and Yang [125] where they used large values of
b /b in their studies. For other nonlinearities, such as the freeplay, it is possible that complex
 
motions or chaos may be encountered for small values of the external forcing. Further studies are
needed to confirm this even though results from Section 5.3 show that a freeplay generates many
examples where bifurcation and chaos are detected. The method described by Gong et al. [36] on

Fig. 74. Pitch amplitude response curve of a two-degree-of-freedom system for k"25, º*"0.95 º* in a moving
*
medium (from Ref. [69]).
300 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

the use of the Poincaré map for analysing bifurcation in system response to periodic forcing may be
extended to more general nonlinearities other than the cubic spring, and possibly with the inclusion
of aerodynamic terms in the governing equations.

6. Bifurcation and chaos of airfoils with aerodynamic nonlinearities

6.1. Transonic flows

In this section, we consider two examples of LCO caused by transonic aerodynamic nonlineari-
ties. The first occurs in the presence of shock oscillation in an inviscid flow, and was studied by
Kousen and Bendiksen [56,57]. They investigated a two-degree-of-freedom airfoil motion for
a linear spring, and later extended the analysis to a structural freeplay nonlinearity. The other
example is from Edwards [26] who considered a linear structure with attached and separated flow
behind the shock wave.
Using a two DOF typical section model for bending-torsion motion, Kousen and Bendiksen
[56] coupled the airfoil motion to an unsteady Euler solver to study transonic freestream Mach
number effects. Unlike the transonic small disturbance formulation, Euler equations allow larger
amplitude shock motion. However, neglecting viscous effects at large amplitude airfoil oscillation
may not yield accurate results when significant flow separation takes place.
From Eq. (40), the NS equations can be reduced to the Euler equations by dropping the viscous
terms. Bendiksen and Kousen [11] expressed the two-dimensional equations in integral form as
follows:
j
jt  Q dx dy# (F dy!G dx)"0 , (168)

where Q is given by Eq. (41) without the w term, that is,


o
ou
Q" , (169)
ov
oe

and F and G for the Euler equations in two dimensions can be written as



o(u!x ) o(v!y )
R R
ou(u!x )#p ou(v!y )
F " R , G " R . (170)
ov(u!x ) ov(v!y )#p
R R
oe(u!x )#pu oe(v!y )#pv
R R
In these equations, Q is the vector of independent variable, F and G are the x- and y-direction
vectors of mass, momentum, and energy flux, respectively. For a perfect gas, the energy can be
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 301

written as

p 1
e" # (u#v) , (171)
(c!1)o 2

where c is the ratio of specific heats. Eqs. (168)—(170) represent a two-dimensional Euler formula-
tion in which the velocity vector is defined as the instantaneous velocity at the moving boundary
given by (x, y, t). In this formulation, the velocity components u and v in the x- and y-directions are
replaced by (u!x ) and (v!y ), respectively. Note that for a fixed grid system, x "y "0. These
R R R R
equations are solved by Bendikson and Kousen [11] on a nondeformable C-mesh of quadrilateral
elements rigidly attached to the airfoil. Eq. (168) is applied to each cell (i, j) of the mesh and we have
a system of ordinary differential equations written as

d
(S Q )#Q !D "0 , (172)
dt GH GH GH GH

where S and Q is the area and net flux out of the (i, j) cell, respectively. D is a dissipative term
GH GH GH
added to prevent odd—even coupling in the mesh, and to suppress spurious oscillations. Eq. (172) is
integrated forward in time using a five-stage Runge—Kutta scheme. Non-reflecting boundary
conditions are used in the far field, and flow tangency boundary conditions are evaluated at the
instantaneous position on the airfoil surface.
The airfoil motion is computed from Eq. (7) of Section 3.1 without the viscous damping term and
the structure is assumed to be linear. Taking the Laplace transform and expressing it in finite
difference form, we obtain

1 *R

1
q (t )"q (t )cos(u *t)# qR (t )sin(u *t)# F (¹)sin[u (t!¹)]d¹ , (173)
P L> P L P u P L P u P P
P P 
where *t is the time step and the generalized force F (t ) is written in difference form as
P L>


*F
F (t )"F (t )# P *t . (174)
P L> P L *t
L\
Kousen and Bendiksen [56] studied a NACA64A006 airfoil with a "!0.2, x "0.2,
F ?
r "0.5385, u"0.3433 and k"10. In the computations, a steady flowfield was first calculated
?
around the airfoil. The airfoil was then forced in pitch at an amplitude of either 0.1 or 4° for three to
five cycles. Forcing was then stopped and the airfoil allowed to come to a steady state condition.
Generally, a time history containing 40—60 cycles were computed. The velocity range º* was
varied for M"0.25, 0.6, 0.8, 0.85, 0.87 and 0.9.
LCOs were observed only in the transonic range. Fig. 75 shows a bifurcation diagram of the
pitch LCO for four M at 0.8, 0.85, 0.87 and 0.92. These are supercritical Hopf-bifurcations and the
amplitudes grow monotonically from zero at the bifurcation point. Results for the two initial forced
amplitudes show similar results for the limiting amplitude away from the linear flutter point, which
coincides with the Hopf-bifurcation point. From the bifurcation diagram, we note that the linear
flutter velocity increases with increasing Mach number. We can imagine an increase in º* at
302 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 75. Bifurcation diagram for transonic cases (from Ref. [56]).

constant M to be equivalent to a decrease in airfoil chord or the pitch natural frequency. Also limit
cycle amplitudes generally decrease with increasing M.
An interesting phenomenon occurs when the torsional spring is given an initial pretwist. The
pretwist was applied such that the torsional spring would have a zero moment about the elastic
axis at a nonzero angle of attack a . Using the system parameters given above, and choosing

M"0.85 and º*"1.85, cases were computed with different values of a . When a (3.75°, the
 
airfoil approaches a limit cycle centered around a nonzero, positive angle of attack. Slightly below
4° in a , significant harmonics are observed in the airfoil response. At a "4°, the response in pitch
 
is shown in Fig. 76 and the oscillations are chaotic. At a "4.75°, Fig. 77 shows the motion to

settle to some type of LCO behavior, but the amplitude is not constant and the mean angle of
attack is !0.6°. When a "5°, Fig. 78 shows the oscillations to decay to a stable offset of !0.6°.

Since the pretwist is 5°, this means that the computations predict the airfoil will twist 5.6°
downward into a strong stable solution. The pressure distributions on the airfoil show the presence
of multiple shocks on the lower surface and a large supersonic pocket on the upper surface
extending almost to the trailing edge, thus producing the strong pitch-down moment.
In a later paper, Kousen and Bendiksen [57] used the same formulation but studied a freeplay
nonlinearity in the torsional DOF. The airfoil parameters are the same as Kousen and Bendiksen
[56] except that k"60. They considered a symmetric freeplay without preload, having a value of
a "!1°, M "0° and d"2°. Only one Mach number at 0.87 was studied. For a linear spring, the
 
phase-plane plot for a at º*"2 shows a slow approach towards LCO and the amplitude is small
around 0.07 rad. Changing to a freeplay nonlinearity, Fig. 79 shows at º*"1.5 a distorted
character indicating the presence of higher frequency components. The initial forcing of the airfoil
was 0.1°.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 303

Fig. 76. Pitch amplitude for M"0.85, º*"1.85, a "4.0° (from Ref. [56]).


Fig. 77. Pitch amplitude for M"0.85, º*"1.85, a "4.75° (from Ref. [56]).


Fig. 80 shows a bifurcation diagram for the freeplay nonlinearity. Initial forcings at the two
values of 0.1 and 4° give similar results, and a case was tested at º*"1.7 which showed the
resulting limit cycle amplitudes differ by less than 2%. The bifurcation curve has a discontinuity at
º*"0.5. The lack of torsional rigidity in the freeplay prevents the formation of pitch limit cycle
304 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 78. Pitch amplitude for M"0.85, º*"1.85, a "5.0° (from Ref. [56]).


Fig. 79. Pitch phase plane for a freeplay º*"1.5 (from Ref. [57]).

there. Therefore, at reduced velocities where the resulting pitch LCO amplitude might be less
than 1°, the system is unable to respond in that manner. The linear flutter point has dropped
significantly from about 1.9 without freeplay to around 0.5 with freeplay. In this respect, the
freeplay is highly destabilizing.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 305

Fig. 80. Freeplay bifurcation diagram (from Ref. [57]).

When viscous effects are present, transonic shock-induced flow separation can cause self-excited
shock oscillations. This phenomenon was discussed in Section 2.2 as a possible mechanism
responsible for control surface buzz and wing buffeting. For moderate to strong shocks travelling
a significant distance along the airfoil, flow separation is most likely to occur. Many attempts have
been carried out using NS solvers for aeroelastic problems, but the computations are so demanding
that very few configurations have been analyzed. Those that were investigated were mainly on
the topic of dynamic stall [122]. The interactive boundary layer modeling IBLM proposed by
Edwards [26] provides an alternative to the NS direct computation of flows involving viscous
shear layers.
Edwards used the TSD equations [9] given in Section 3.2.1 to compute the inviscid flow. The
reader should be reminded that the TSD equation is suitable only for weak shocks and hence
Edwards’ method should be restricted to weak-shock boundary-layer interaction where the
resulting flow separation is not considered massive. The physical model is illustrated schematically
in Fig. 81. Edwards [26] separated the flow into a region for the inner viscous boundary layer and
an outer inviscid region. The superscripts ‘‘i’’ and ‘‘v’’ denote the inviscid and viscous regions
respectively.
Starting from the leading edge of the airfoil, the boundary layer is approximated by the turbulent
boundary layer on a flat plate. The description of the boundary layer solution is very involved and
requires a lengthy discussion of the boundary layer equations and the various approximations used
in solving those equations. We shall not go into the details of Edwards’ method [26], but refer the
reader to his earlier investigation.
The boundary layer equations are solved in a quasi-steady manner using a set of ordinary
differential equations in the x-direction for the momentum thickness h, shape factor HM , and
306 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 81. Sketch of shock-boundary layer interaction (from Ref. [26]).

entrainment coefficient C . From these quantities, the displacement thickness d* is calculated. The
#
approach used by Edwards [26,27] utilizes the inverse boundary layer method. The inner and
outer solutions are coupled through the boundary conditions on the wing and wake. The boundary
conditions (Eq. (45)) become
$
"S$$S$#d* for x 4x4x , y4y , z"0$ , (175)
X V R *# 2# 2'.
and on the wake we have
D "D(d* ) for x'x , z"0$ (176)
X V 2#
and D( # )"0 for x'x , z"0$.
V R 2#
The coupling method is developed based on the observation that for transonic flow, the flowfield
is unsteady, displaying oscillating shocks and separating and reattaching boundaries. The interac-
ting boundary layer method is thus regarded as a simulation of two dynamic systems, the outer
inviscid flow and the inner viscous flow, whose coupling requires special treatment to ensure that
the coupling error between the two systems is minimised.
Edwards [26] demonstrated his method on a NACA 0012 airfoil and an 18% thick circular arc
airfoil to show the onset of self-excited shock-induced oscillation for perfectly rigid airfoils. Of more
interest is his study on wing flutter models.
The first wing flutter model is the AGARD Standard Aeroelastic Configuration [124] (AGARD
445.6) which was tested in the NASA LaRC Transonic Dynamics Tunnel. It is a semispan model
having a quarter-chord sweep angle of 45°, taper ratio of 0.66 and is wall mounted to the wind
tunnel. The wing had a NACA 65A004 airfoil section, and is modeled structurally using the first
four natural vibration modes, with natural frequencies ranging from 9.6 Hz for the first bending
mode to 91.54 Hz for the second torsion mode. Flutter calculations for this wing at Mach numbers
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 307

below and close to unity show that viscous modeling is required to achieve acceptable accuracy. In
this region of M close to unity (subsonic M), calculations show evidence of small amplitude LCO
behavior. At M"0.96 and dynamic pressure Q"0.75 psi, Fig. 82 shows nonlinear response
features from the time variation of the wing tip displacement z . The early portion of the response

shows positive damping of the flutter mode and a higher flutter frequency. The damping of the
flutter mode decreases as the amplitude decays to approximately 0.12 in peak-to-peak, where stable
LCOs persist. The initial portion of the time series can be considered as transients. This LCO
behavior was further studied by sequentially increasing the dynamic pressure between computed
runs from Q"0.5 to 0.81 psi. The resulting tip deflection time history is shown in Fig. 83 where
eleven computer runs were carried out. The dynamic pressure was incremented as indicated in steps
between restarted runs. For Q40.6 psi the response is damped and for Q"0.70 psi small
neutrally stable oscillations are seen. With Q increased to 0.78 psi slowly divergent oscillations
develop and with further increase to 0.81 psi the divergent oscillations grow with increased
negative damping until the amplitude reaches approximately 0.12 in peak-to-peak. The growth of
the oscillations then quenches and it appears that a limit cycle condition will again develop.
Further calculations are needed to fully establish this feature, but these have not been carried out

Fig. 82. Calculated AGARD 445.6 wing tip response in heavy gas for M"0.96 and Q"0.75 psi (from Ref. [26]).

Fig. 83. Calculated AGARD 445.6 wing tip response in heavy gas for M"0.96 and increasing dynamic pressure (from
Ref. [26]).
308 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

by Edwards [26]. This phenomenon of LCO is only observed at M"0.96 which was the highest
M calculated. At this value of M and for the wing motions considered, the flow is fully attached
since the wing is thin. Computations show significant regions of near sonic flow develop adjacent to
the wing upper and lower surfaces as the wing oscillates. This is an example of nonlinear
aerodynamics with attached flow interacting with the linear structure to produce LCO.
The second model is a typical business jet configuration also tested in the NASA LaRC
Transonic Dynamic Tunnel. The wing has a taper ratio of 0.29 and a mid-chord sweep of 23°. The
airfoil thickness varies from 13% at the symmetry plane to 8.5% at the wing tip. Six natural
vibration modes were included in the calculations, with frequencies varying from 4.3 to 62.7 Hz.
The wing is thicker than the AGARD 445.6 and the requirement for viscous modeling extends to
lower Mach numbers.
At M"0.888 and at the flutter dynamic pressure of 79 lb/ft, Fig. 84 shows two transient
responses indicating the existence of LCO. The first figure shows a wing tip initial displacement
of about 7 in and the motion decays to LCO with amplitude of 5—6 in peak-to-peak. The second
curve shows a growth from a small initial displacement to the LCO amplitude. This agrees with
experimental observations on tests carried out at NASA LaRC. At the experimental flutter
conditions for this M, the model undergoes constant amplitude wing oscillations with an amplitude
of slightly less than one tip chord (6.3 in) peak-to-peak. Calculations show that the flow over the
wing was intermittently separating and reattaching on the outbroad upper and lower surfaces.

6.2. Dynamic stall

In addition to the airfoil parameters that we have discussed in previous sections, there are
a number of other parameters which affect the aeroelastic response of an airfoil undergoing forced
oscillations under dynamic stall conditions, such as, the nondimensional frequency of oscillation k,
the magnitude of the applied torque, and the initial value of the airfoil pitch angle, a . Bearing in

mind that the airfoil response must be evaluated numerically it is clear that the effect of only
a limited set of parameters can be considered.
Price and Keleris [90] considered only a one degree-of-freedom airfoil motion in pitch. Assum-
ing a sinusoidal externally moment Q/mº"Q sin kq is applied to the airfoil, they studied the

effect of k, º* and Q , and presented results for k"100, r "0.5, a "!0.5, and zero structural
 ? F
damping. The product of k and º* gives the ratio of the forcing frequency, u, to the structural
natural frequency in pitch, u , hence, this product is also of interest. Appropriate values of Q
? 
should be of the same order of magnitude as the aerodynamic moment term, 2C (q)/nk, and in the
+
range of k and C they considered, 2C (q)/nk lies between 10\ and 10\, so values of Q were
+ + 
chosen to be of order 10\. The results presented may conveniently be divided into two categories
corresponding to what Price and Keleris [90] referred to as low and high frequencies of oscillation,
and these are discussed separately in the following two sections. The examples given by Price and
Keleris [90] are merely a limited subset of a more complete study carried out by Keleris [54].

6.2.1. High-frequency response

The results in this section are for º*"25.2 and k"0.088, corresponding to u/u +2.2; hence,
?
the forcing frequency is higher than the airfoil natural frequency. Some typical time histories of
a for four different Q are presented in Fig. 85, and the corresponding phase-plane plots are shown

 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 309

Fig. 84. Calculated limit cycle response for a business jet wing flutter model: M"0.888, Q"79 psf., Re "1.14;10

(from Ref. [26]).

Fig. 85. Time histories of the airfoil response; a "9.76°, º*"25.20 and k"0.088. (a) Q "0.550;10\, (b) Q "
  
0.640;10\, (c) Q "0.675;10\, (d) Q "0.73;10\ (from Ref. [90]).
 
310 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 86. Phase plane sections of the airfoil response; a "9.76°, º*"25.20 and k"0.088. (a) Q "0.550;10\,
 
(b) Q "0.640;10\, (c) Q "0.675;10\, (d) Q "0.73;10\ (from Ref. [90]).
  

in Fig. 86. Fig. 85a shows a periodic limit cycle, with frequency corresponding to the forcing
frequency of the applied torque Q ; this period-one motion is represented via a single curve in the

phase plane plot of Fig. 86a. Upon increasing Q a period-two response is obtained, as shown in

Figs 85b and 86b, and a further increase in Q yields the period-four response shown in Figs 85c

and 86c. However, yet a further increase in Q yields what appears to be a non-periodic response for

a (see Figs. 85d and 86d).
This type of behavior is illustrated in Fig. 87 for a wide range of Q where a bifurcation diagram

of a versus Q is presented. The bifurcation diagram shows, for each value of Q the local value of
 
a when a"0; hence, for a period-one response, two points appear on the bifurcation diagram,
corresponding to the minimum and maximum a. For the period-two motion, the bifurcation
diagram contains four points, while for Q "0.73;10\, the bifurcation diagram contains an

extremely large number of points, suggesting that the airfoil motion may possibly be chaotic.
The bifurcation diagram shows that the amplitude of a generally increases as Q is increased,

and at Q +5.75;10\ the response undergoes a period doubling bifurcation. The response

undergoes another period doubling bifurcation at Q +6.5;10\, and after this bifurcation the

system quickly becomes unstable. It eventually restabilizes into a period-two oscillation at
Q +9.2;10\. The period-doubling was confirmed via a Fourier analysis of the time traces.

Although the results presented above are indicative of a chaotic response, because of the
nonanalytic nature of the system it is not possible to prove that the response is chaotic using
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 311

Fig. 87. Bifurcation diagram of the airfoil pitch motion; a "9.76°, º*"25.20, k"0.088 (from Ref. [90]).


standard tools, such as calculating the Lyapunov exponents. Furthermore, it is difficult to make
any conclusions concerning the route to chaos. However, the bifurcation diagram shown in Fig. 87
suggests very strongly that the route to chaos is via a period doubling cascade. Further evidence for
this can be gained from an examination of the Poincaré sections for values of Q close to the

apparent boundary between chaotic and periodic motion. In Fig. 88a, the Poincaré section for
Q "6.90;10\ (near the onset of chaos) is shown. The general shape of the attractor, consisting

of four short lines, indicates that the response follows a marginally unstable period-four attractor.
If the period-four attractor were stable, then the Poincaré section should consist of four distinct
points, and indeed, this is what was obtained with Q "6.75;10\, but the results are not shown

here. However, in Poincaré section for a slightly higher value of Q , shown in Fig. 88b, the four

part structure of the attractor, evident in Fig. 88a, has evolved into a two part attractor. The two
pairs of curves of the previous attractor have grown to overlap each other so that only two
independent curves can be distinguished. Although it may appear that these curves are one-
dimensional, if a small part of the attractor is enlarged then it is evident that there is considerable
internal structure to the attractor, as shown in Fig. 88c. The continued evolution, as Q is increased,

of the original four part attractor into a chaotic attractor can be seen from Fig. 88d, where
although four distinct regions of the attractor are apparent the attractor is typical of that obtained
with a chaotic system.

6.2.2. Low-frequency response

All the results in this section are for a "8.9°, º*"21.0227 and k"0.044; this corresponds to

a frequency ratio u/u "0.925, hence, the forcing frequency is much closer to the natural frequency
?
u than for the results presented previously. A bifurcation diagram of a as a function of Q shown in
? 
Fig. 89. Fig. 89a shows that the system undergoes two period doubling bifurcations, the first at
312 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 88. Poincaré sections of the airfoil response; a "9.76°, º*"25.20, and k"0.088. (a) Q "0.690;10\,
 
(b) Q "0.700;10\, (c) an enlarged view of the micro-structure for the attractor in (b), (d) Q "0.730;10\ (from
 
Ref. [90]).

Fig. 89. (a) Bifurcation diagram for the airfoil response for 0.1;10\(Q (1.18;10\, (b) expanded view of (a) for

0.98;10\(Q (1.10;10\. a "8.9°, º*"21.0227, k"0.044 (from Ref. [90]).
 
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 313

Q +0.20;10\ and the second at Q +0.88;10\. Furthermore, a region of possibly chaotic

 
motion can be seen to exist in the approximate range 1.00;10\(Q (1.09;10\.

The more detailed bifurcation diagram of Fig. 89b shows that as Q increases above 0.98;10\

the well-defined lines, characteristic of periodic motion, begin to spread out or diffuse into one
another. Furthermore, at Q +1.003;10\ and 1.0062;10\, the system appears to undergo

complicated bifurcations. This bifurcation diagram suggests that the system is possibly chaotic or
quasi-periodic for approximately Q 51.0062;10\, and that the transition from periodic to

chaotic or quasi-periodic behaviour is a gradual decrease in the stability of the periodic oscillations.
After Q exceeds approximately 1.079;10\, the response returns to a stable period-two oscilla-

tion. Within the apparently chaotic region there are smaller regions where the response is periodic,
for example near Q "1.058;10\ and 1.065;10\.

To further assess the possibility of chaos, Fourier spectra, phase plane plots and Poincaré
sections were constructed at a number of values of Q . Three typical Poincaré sections are shown in

Fig. 90. The Poincaré section for Q "1.00;10\, Fig. 90a, has four groups of points indicating

that the response is a period-four oscillation. However, since the groups are not exact points, this
period-four oscillation is not perfectly stable. The Poincaré section for Q "1.006;10\, Fig. 90b

shows that the four groups of points have diffused into fine lines, and an attractor is beginning to
emerge. The structure of this attractor is clearly visible in the Poincaré section for Q "1.02;10\

(Fig. 90c). This attractor is not as one-dimensional as Fig. 90c suggests, and an enlarged view of

Fig. 90. Poincaré sections of the airfoil response, a "8.9°, º*"21.0227, and k"0.044: (a) Q "1.000;10\,
 
(b) Q "1.006;10\, (c) Q "1.020;10\ (from Ref. [90]).
 
314 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 91. Long-term behavior of the airfoil response, a "8.9°, º*"21.0227, k"0.044 and Q "1.020;10\ (from
 
Ref. [90]).

certain regions indicates that the attractor varies in thickness with definite boundaries defining its
shape.
To better appreciate the behavior illustrated in Fig. 90c the long-term response of the system for
Q "1.02;10\ is presented in Fig. 91. Here, discrete values of a at a"0 are plotted for each

period of oscillation of Q during which they exist. Thus, a stable period-one type oscillation would

appear as horizontal line, and a period-two motion as two horizontal lines, etc. However, the
long-term behaviour of the system shown in Fig. 91 is much more complicated than this. Initially
the response is a high-order periodic oscillation, and it remains in this state for approximately its
first 2500 periods, then the response goes through approximately 500 periods where it is either
quasi-periodic or chaotic, followed by a return to a periodic oscillation that remains stable for
approximately the next 5000 periods. This periodic response then suddenly becomes unstable and
the response undergoes frequent transitions from high-order periodic phases to quasi-periodic or
chaotic phases. These high-order periodic phases appear in a seemingly random fashion and can
remain stable for up to several hundred periods. On the other hand, the quasi-periodic or chaotic
phases are usually very short, on average lasting only about ten periods before the response
restabilizes into a new and different periodic phase. After approximately 8000 periods of this
unstable response, the total number of elapsed periods being of order 17 000, the system stabilizes
into a period-one oscillation lasting for approximately 400 periods. This then becomes unstable
(not shown in this figure) and the more complicated behaviour returns.
The behavior illustrated in Fig. 91, particularly the time scale of the chaotic behaviour and the
marginal stability of the response, suggests, at least superficially, that the response is intermittently
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 315

chaotic [13]. However, to give a more definitive assessment of this, and to see if the present system
shares any characteristics with classical types I, II or III intermittency the Poincaré data is exa-
mined in more detail in the form of first and second return maps.
Type I intermittency involves the destabilization of a periodic trajectory. Its main characteristic
is that a laminarization channel exists in the first return map, the response enters this channel and
with successive iterations moves through it. While the response is in the channel it is reasonably
stable, and therefore the behavior appears periodic. The length of time required for the response to
move through the channel depends on the width of the channel, however, Bergé et al. [13] indicate
that there will be some very short laminar phases and that there is a definite maximum laminar
phase length which is system dependant. After the system leaves the channel it moves to another
part of the phase plane where the attractor is chaotic, giving a chaotic outburst in the response. The
system remains in the chaotic region of the phase plane until it is mapped back to the beginning of
the channel.
To the authors’ knowledge no examples of type II intermittency, either numerical or experi-
mental, have been observed and so this type of intermittency will not be considered here.
Type III intermittency also has a laminarization channel, but in this case it is found in the second
return map, and it is due to the first return map crossing the identity line at a slope slightly less than
!1. While the system is in this channel the response is periodic, as in type I intermittency. An
important characteristic of this type of intermittency is that the approach of the chaotic outburst is
signalled by the growth of a subharmonic oscillation which increases in amplitude. The character-
istic probability distribution of the average length of time for the laminar phase suggests that there
is a definite minimum length for the laminar phases and that some laminar phases can last for
a very long time.
The first and second return maps for the present system with Q "1.02;10\ are shown in

Fig. 92. Unfortunately, these return maps do not appear to show any of the characteristics
described previously for classical types I or III intermittency. Furthermore, the long-term behavior
of the system, as shown in Fig. 91, is not typical of these types of intermittency. Firstly, the
long-term response shows regions of both very long and very short periodic phases, suggesting
a combination of types I and III intermittency, and secondly, after each chaotic burst the system
does not return to the same periodic oscillation but instead tends to a new periodic state.
However, if the first return map of Fig. 92a is examined in more detail, as shown in Fig. 93, then
some indication of the type of intermittency can be obtained. The internal structure of the attractor,
shown in Fig. 93, consists of short lines or small groups of points; different combinations of these
points corresponding to different periodic solutions. The quantized structure of the attractor
implies a periodic response because it consists of a finite number of groups of points, but, since
these groups of points have a two-dimensional structure the periodic responses are unstable. An
example of the implications of the internal structure of the attractor can be seen from the following.
According to Fig. 91, after approximately 16 500 periods the system begins a marginally stable
period-one oscillation that persists for approximately 4000 periods. In order for a marginally stable
period one oscillation to exist the attractor of the first return map must intersect the identity line
at a slope slightly greater than !1 or slightly less than #1, otherwise successive iterations of the
map will quickly move away from the intersection point and the period-one oscillation will be
unstable. The orientation of the attractor for the first return map, shown in Fig. 92a, suggests that
the slope of the attractor at the intersection point is significantly less than !1, therefore, no stable
316 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 92. (a) The first return map and (b) the second return map of the Poincaré data; Q "1.020;10\, a "8.90°,
 
º*"21.0227 and k"0.044 (from Ref. [90]).

Fig. 93. The micro-structure of the first return map of the Poincaré data; Q "1.020;10\, a "8.90°, º*"21.0227
 
and k"0.044 (from Ref. [90]).

period-one oscillation should exist. However, as seen in Fig. 93, the microscopic structure of the
attractor makes it possible for two of the internal structures of the attractor to intersect the identity
line at slopes that result in a marginally stable period-two oscillation. The two internal structures
are so close to the identity line that, on the normal scale of the Poincaré map, this marginally stable
period-two oscillation appears as a period-one oscillation. Hence, by examining the Poincaré data
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 317

at the microscopic level, it is clear that the internal structure of the attractor plays an important
role in the marginal stability of this type of chaotic response. However, how this particular internal
structure develops is still unclear.
Lee and LeBlanc [63] also used the same approach to study forced oscillation of a 2-DOF airfoil
with attached and separated flows. Only small plunge amplitudes were considered and a linear
theory was used to predict the aerodynamic loads due to plunge motion.
The equations of motion are the same as those given by Eqs. (10) and (11). In addition to C ,
+
an empirical expression for the normal force coefficient C at dynamic stall was also obtained.
,
Both the moment and normal force of the two degree-of-freedom system consist of two terms:
one from the nonlinear pitch motion using Bielawa et al.’s [15] time synthesization technique
and a linear term using indicial lift and moment functions given by Mazelsky and Drischler
[78]. This is a very crude approximation since in a linear analysis where the airfoil
motion is considered small, the superposition of C and C for pitch and plunge motions
+ ,
derived separately is permissible in determining the total moment and force coefficients.
However, when the aerodynamic loads are nonlinear functions of the displacements, these
coefficients have to be determined for combined motions in the two degrees of freedom.
The approach used by Lee and LeBlanc [63] in modifying the Bielawa et al. [15] expressions
for C and C should be treated as an approximation to an otherwise extremely complex flow
+ ,
phenomenon.
Lee and LeBlanc [63] used the same airfoil parameters as those investigated by Price and Keleris
[90]. Different values of u, u , and forcing frequency u were considered. The viscous damping for
?
the plunge and pitch motion was set to zero. In the limit when u is infinite, the motion degenerates
to that for a 1-DOF system. For forced oscillation in the pitch DOF, the reduced frequency was
kept constant and u/u was varied. This implies that º* has to be adjusted accordingly so that k
?
is constant. In one example given by Lee and LeBlanc [63], response curves were computed for
different forcing Q ranging from 1;10\ to 25;10\ for the NACA 0012 airfoil at a "7.48°,
 
k"0.165 and M"0.6. In the regions where transition from attached to separated flow occurs,
there are breaks in the response curves. The flow is highly unsteady and the time series do not reach
constant amplitudes after a large number of cycles of oscillation. This is due to the modeling in the
aerodynamics where the variations of the attached and separated flow aerodynamic coefficients
with displacements have discontinuities in slopes. It was found in this particular example that for
the larger values of u/u ('2) and Q ('8;10\) investigated, the time series do not reach a
? 
steady state and the motion is likely chaotic.
Lee and LeBlanc [63] also studied the response by holding Q constant and varied the driving

frequency for different values of u . The curves show the same characteristics as those for constant
?
k and varying u/u . For the NACA 0012 airfoil at a "7.48°, the response curves for a 2 DOF
? 
system with small plunge motions are similar to those for a 1DOF system. Typical response curves
are shown in Fig. 94. For small values of u/u the flow is attached and transition to separated flow
?
occurs at u/u +1. As u/u increases to values greater than 1.6, the flow becomes attached again.
? ?
For the largest value of u"10, a steady condition is reached after approximately 20 cycles of
oscillation. Decreasing u results in failure to achieve a steady state condition for u/u '1.6 even
?
for a very large number of cycles. The flow remains attached but the amplitudes of pitch and plunge
motions are scattered within the shaded regions shown by curves 2 and 3. The motion is probably
chaotic, but Lee and LeBlanc [63] did not purse this subject any further. They concluded for the
318 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 94. Variation of the pitch amplitude with u/u for an airfoil oscillating in pitch and plunge. a "7.48°, f "64 Hz
?  ?
and Q "0.5;10\ (from Ref. [63]).


airfoil parameters and type of motion they studied, a small plunge motion increases the pitch
amplitude slightly for attached flow, while the opposite is true for separated flow.
In a recent paper, Li and Fleeter [74] investigated periodic and chaotic motions of an airfoil
experiencing dynamic stall using a semi-empirical dynamic stall model by Gormont [37]. This
relatively simple model captures the nonlinear behavior generated by the dynamic stall and is used
to investigate bifurcations and routes to chaos.
A 2D airfoil was used and the dynamic equations are expressed in a set of nondimensional
first-order differential equations (a four dimensional autonomous system in state space) and
integrated numerically using a fourth-order Runge—Kutta scheme. The airfoil parameters are:
k"60, r "0.69, u"1.26, a "!0.5, x "0.5, and airfoil thickness-to-chord ratio"12%. The
?  ?
static stall angle a was 9°. The initial displacement a(0) was 12° and they defined a nondimensional

velocity ºM "º*(k/2.
Fig. 95 shows a bifurcation diagram of a/a versus ºM . The torsion response amplitude is taken at

which the phase plane trajectories cross the zero velocity axis as a function of the nondimensional
air speed. Transients have decayed prior to the construction of this bifurcation diagram. A subcriti-
cal Hopf-bifurcation occurs at ºM "0.3038, and the airfoil motion is periodic between ºM "0.3038
and 0.3452. The motion becomes unstable when ºM is further increased and bifurcates into
a quasi-periodic oscillation. This does not lead to chaos, but instead a reverse secondary Hopf-
bifurcation changes the motion back to periodic again. Beyond ºM "0.4607, another secondary
Hopf-bifurcation occurs and further increase in velocity results in chaotic response. For the airfoil
parameters investigated by Li and Fleeter [74], only the quasi-periodic route to chaos was
observed. They also presented some other results for a different k"80 at the same u"1.26 and
for k"60 at u"1.6. The value of velocity where chaos occurs changes with u for the same k and
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 319

Fig. 95. Bifurcation diagram: amplitude of torsion displacement vs. airspeed ºM (k"60, u"1.26) (from Ref. [74]).

there is only one secondary Hopf-bifurcation to chaos. Increasing k but keeping u fixed at 1.26,
there is little change in ºM at the onset of chaos, and the bifurcation diagram look very similar.
The route to chaos is different from that reported by Price and Keleris [90] since frequency
doubling and intermittency were not observed by Li and Fleeter [74], but it should be remembered
that Li and Fleeter [74] considered an unforced two degree-of-freedom system. Also, the empirical
aerodynamics is different and it is not certain whether Bielawa et al. [15] formulation of C and
,
C will make any difference to Li and Fleeter’s [74] results.
+

7. Gust effects: airfoils in longitudinal atmospheric turbulence

One idealization made in the previous sections is that the airfoil is assumed to be free of
any sources of perturbation, which in reality are always present. This raises, from both a practical
and theoretical point of view, the question of the presence of random noise introduced via
atmospheric turbulence. The problem of stochastic fluctuations has recently been reviewed in light
of the understanding of chaotic behavior and related nonlinear dynamics. It has been shown that
random noise can produce bifurcation phenomena and organized behavior which have no analog
in their deterministic counterparts. These phenomena have been designated as ‘‘noise-induced
transitions [45]’’. They are a product of the interplay between stochastic fluctuations and nonlinear
dynamics.
Atmospheric turbulence models can generally be categorized within two different approaches
[6]. The methods associated with a discrete gust representation are usually of a deterministic
nature. On the other hand, continuous turbulence methods allow a stochastic perspective. In this
review, the interest lies in the effect of nonlinearities from a dynamics perspective; thus, a continu-
ous stochastic approach is required. A deterministic gust model was used to study nonlinear
320 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

helicopter blade response, and the interested reader can refer to the work of Tang and Dowell
[105].
In two-dimensional flow atmospheric turbulence is composed of vertical and longitudinal
components. The vertical component acts as an external forcing function and does not alter
fundamentally the low-dimensional dynamics of the otherwise deterministic airfoil response. On
the contrary, the longitudinal turbulent field excitation, also known as head-on turbulence, is
parametric as it acts on the airspeed. It is in this context that the stochastic system behavior departs
from its deterministic counterpart and induces new bifurcation phenomena. Only the effect of
longitudinal atmospheric turbulence is considered here.
In this review, the highlights of Poirel and Price [85] investigations are summarized. The
spectral content of turbulence can be provided by either the von Karman or Dryden models, which
are presently the two most widely accepted models. The Dryden model was used by Poirel and
Price [85] since it is easier to handle mathematically. The one-sided PSD Dryden longitudinal gust
representation is given as [44]
U (u)"p 2(¸/nº )/(1#(¸u/º )) , (177)
 
and is presented in non-dimensional form in Fig. 96 for the particular scales of turbulence Poirel
and Price [85] considered. p in the above equation is the gust velocity variance. It can be shown

that the scale of turbulence, ¸, divided by the mean free-stream velocity, º , is equal to the
correlation time of the random excitation.
Only attached flow conditions with small amplitude oscillations are considered, hence, the
aerodynamics is linear. The unsteady aerodynamics is modeled, assuming incompressible inviscid
flow, via Duhamel’s integral and the two-state representation of Wagner’s function. This is usually
referred to as arbitrary-motion theory. Classically, it considers motion in the structural degrees of
freedom only, where the downwash at the three-quarter chord position represents the effective
downwash on the airfoil [31].
Here, arbitrary-motion theory is extended to the case of a random time-varying airspeed based
on the work of van der Wall and Leishman [114]. They show that arbitrary-motion theory
compares very well with Isaacs’ [48] theory for the case of periodic fore-aft movements of an airfoil,
along with pitch and heave oscillations. In fact, the only difference appears to be attributable to the
limited number of states used in the approximate representation of Wagner’s function. Note that
Isaacs [48] considered only periodic fore-aft and pitch oscillations. His theory was generalized by
van der Wall [113] to include heave motion as well and can be considered exact for the preceding
fluid assumptions.
It is important to note that fore-aft movement of an airfoil in a uniform stream is not physically
the same as fluctuations of the free-stream velocity. In the first case the airflow is uniform along the
chord, whereas a chordwise velocity gradient exists for a varying freestream velocity. This
difference affects both the non-circulatory force and moment, as well as the bound vorticity of the
circulatory terms. However, the free vortex sheet in the airfoil wake is the same in both cases. van
der Wall and Leishman [114] showed that fore-aft movement aerodynamics is an approximation
of the unsteady free-stream problem, which is valid only for small frequencies. The chordwise
uniformity of the airflow assumption is considered here and is extended to the case of random
fluctuations. This is supported by the atmospheric turbulence velocity spectrum being concen-
trated in the low frequency range. It is thus assumed that the aerodynamic forces induced by
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 321

Fig. 96. Non-dimensional PSD of the Dryden longitudinal gust for different scales of turbulence; closed-form solution
for p "1.0 (from Ref. [85]).
 

random variations of the free-stream velocity, and accounting for arbitrary pitch and heave
motions, can be accurately modelled by the following:

  
R d (t!s)
!¸(t)"!nob[h® #ºQ a#ºaR !ba ä]!2nobº N (t) (0)! N (s) ds ,
   ds

(178)

and

M (t)"nob[ba h® !b(1/2!a )ºaR !b(1/2!a )ºQ a!b (a#1/8)ä]

C?  F  

  
R d (t!s)
#2nob(a #1/2)º N (t) (0)! N (s) ds , (179)
   ds

where N (t)"hQ #ºa#b(1/2!a )a. The dots denote differentiation with respect to t.
 
The first terms in the lift and moment expressions represent the noncirculatory forces, associated
with fluid inertia. The circulatory forces, given by the second terms, model the effect of the bound
vorticity and the shed wake, convecting downstream at velocity º(t)"º #u (t).

In order to obtain a numerical solution via Runge—Kutta integration the equations of motion
must be expressed in state space form, giving a seventh order system. Four states represent the
structural degrees-of-freedom. Two of the other states are associated with the lag terms of the
aerodynamics, which appear after reformulation of the integral in the circulatory force. This leads
322 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

to an additional second order ordinary differential equation, given by


R d (t!s)
N (s) ds"Z(t)ZQ (t) , (180)
 ds    

where Z® (t)#(e #e )(º /b)ZQ (t)#e e (º /b)Z(t)"N (t).
    
The final state of the seventh-order system models the Dryden longitudinal gust, in the time
domain, it is transformed as

uR (t)"!u (t)º /¸#p (2º /n¸U )G (t) , (181)

    
where G is Gaussian white noise, whose intensity is defined by a single-sided PSD of magnitude

one, i.e., U "1. The Gaussian white noise is generated at each time step of the integration as

discussed by Poirel and Price [85].
Poirel and Price [85] generated results using a fourth-order Runge—Kutta numerical scheme.
Bearing in mind the stochastic nature of the process, they paid special attention to the size of the
time step and number of iterations. The factors that determine the magnitude of the time step
are numerical stability and noise correlation time, ¸/º . Accordingly, the time step is taken as the
smallest of either 1/128 of the lowest uncoupled natural frequency of the airfoil or 1/5 of the noise
correlation time. A very large number of iterations is necessary to ensure a smooth probability
density distribution, PDD. This is particularly important when examining the dynamics of the
system. For the work presented here 4;10 to 8;10 iterations were used.
Before presenting results showing the effect of longitudinal atmospheric turbulence on the flutter
and post-flutter response, the dynamics of the deterministic (non-excited) airfoil will be discussed.
The cases presented are for an airfoil with the following nondimensional parameters: u"0.6325,
x "0.25, r "0.5, k"100 and a "!0.5, the nonlinearity considered is a cubic restoring
? ? 
moment in pitch with b "b "0.0 and b "400 (see Eq. (31)).
  
For comparison purposes, bifurcation diagrams of the airfoil response as the airspeed is
increased, for both the nonexcited and excited cases, are presented in Fig. 97. At this point,
however, only the nonexcited case will be discussed. The vertical axis shows both the traditional
amplitude of oscillation and the variance of the airfoil pitch motion. At low air speeds the
bifurcation diagram indicates a stable equilibrium point, which becomes unstable for air speeds
º* 54.31. Here º* "º /bu is defined in Eq. (12). The point º*"4.31 is the nondimensional
? *
flutter speed and corresponds to a supercritical Hopf-bifurcation (at higher airspeeds the equilib-
rium position is unstable, but a stable limit cycle oscillation, LCO, exists). The flutter speed and
Hopf-bifurcation point are uniquely determined by either the amplitude of the pitch oscillation or
its variance. In general terms, a bifurcation is defined as being where there is a qualitative change in
the topological structure of the dynamic behavior [2]. In a deterministic system this definition
corresponds exactly to the point where one of the Lyapunov exponents vanishes, and thus, it is
associated with a critical slowdown of the dynamics. In this case, linearization of the system about
the fixed point gives a complex conjugate pair of eigenvalues, whose real parts go to zero at the
flutter speed. Although not presented here it was found that in the vicinity of the bifurcation point
the reduced frequency of the LCO is approximately 0.18.
Since the main objective of this section is to examine the stochastic (randomly excited) case, it is
relevant to describe the dynamics of the deterministic system in terms of a probability density
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 323

Fig. 97. Bifurcation diagram of the airfoil response for the excited and non-excited cases; p "1.0, ¸ "5.0 (from
  
Ref. [85]).

distribution (PDD) which is presented in Fig. 98. Note that the vertical axis shows the probability
density multiplied by the variance of the response. This representation shows two qualitatively
different regions of dynamical behavior; before and after the Hopf-bifurcation. Prior to the
Hopf-bifurcation the response is characterized by a flat distribution, which indicates zero variance
or no dynamics (for a pure PDD plot there would be a Dirac delta function centred at zero pitch
angle). The LCO response after the Hopf-bifurcation is characterized by a ‘‘crater-like’’ shape, the
peaks of which correspond to the most probable value that the pitch angle will take, and thus, to
the amplitude of oscillation. Hence, as demonstrated in Fig. 98, the variance and location of the
peaks of the PDD are two significant measures of the system dynamics.
The PDD multiplied by the variance of the stochastic pitch response as a function of mean
free-stream velocity is presented in Fig. 99 (this is the equivalent of Fig. 98 but for the ‘‘excited
case’’); in these results the nondimensional gust variance is p "p /(bu )"1.0 and the
   ?
nondimensional scale of turbulence ¸ "¸/b"5.0. There are now three distinct regions of

qualitatively different dynamic behaviour, separated by two critical airspeeds, º*+4.05 and

º*+4.45. This suggests that the response goes through two bifurcations instead of one. The first

region, for velocities 4º* , is characterized by a flat distribution indicating no dynamics. It is only

in the third region, for velocities 5º* , that the response demonstrates fully developed periodic

oscillations and we recognize the double-peaked crater-like shape of the PDD representation
which characterizes the LCO motion. The pitch angles at the peaks of the PDD are analogous to
the amplitude of the LCO motion of the deterministic case. The second region, between º* and

º* , is denoted by single peak with some diffusion about its mean, which occurs at zero pitch angle.

Therefore, the most probable value that the pitch angle will take during this motion is zero.
324 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 98. Bifurcation diagram of the airfoil PDD response for the nonexcited case (from Ref. [85]).

A second observation is that the flutter speed and Hopf-bifurcation are apparently no longer
coincident. It is suggested that º* defines the stochastic flutter speed, as the equilibrium state

becomes dynamically unstable at this velocity. Flutter bears its origin from a linear approach, and
thus, can be observed with a linearized system. Because of the stochastic nature of the paramet-
rically excited system it is difficult to pinpoint exactly the point of instability of the linear airfoil
using a numerical time integration method. This is because the closer we are to the instability the
lower the damping of the slow variable, thus, locally exceeding the maximum value permitted by
the simulation — even though global statistical measures such as mean and variance are within an
acceptable range. However, the flutter point can still be determined within an acceptable margin of
error. To that extent, numerical solutions of the system with different magnitudes of the nonlinear
stiffness coefficient, 504b 4800 gave exactly the same flutter speed, confirming that this is

a linear phenomenon.
The definition of the stochastic Hopf-bifurcation point is somewhat more ambiguous and leaves
room for interpretation. Although the spectrum of Lyapunov exponents has not been calculated
for the second bifurcation, intuitively it is suggested that this second topological change in the
dynamics is not associated with a vanishing of the exponents or a critical slowdown of the
dynamics. This is contrary to deterministic bifurcation phenomenon, and is purely a product of
the interaction between the stochastic and nonlinear natures of the system. In stochastic nonlinear
dynamical systems, this bifurcation is called a P-bifurcation [3] (P for Phenomenological). The first
bifurcation is termed a D-bifurcation (D for Dynamical). Poirel and Price [85] argued that the
P-bifurcation represents the stochastic Hopf-bifurcation and occurs at º* , i.e., where the PDD

 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 325

Fig. 99. Bifurcation diagram of the airfoil PDD response for the excited case; p "1.0, ¸ "5.0 (from Ref. [85]).
  

changes from being single- to double-peaked, and also where fully developed periodic oscillations
start to appear in the time history.
A final observation concerns the exact location of the two bifurcation airspeeds. This is best seen
from the results of Fig. 97, where the vertical axis shows the variance and most probable value of
the pitch state. For this particular set of parameters, interpolation of the variance results suggests
the stochastic flutter point to be at º*"4.05, while the most probable value suggests that the

stochastic Hopf-bifurcation occurs at º*"4.45. Also presented in Fig. 97 for comparison is the

nonexcited deterministic case. Compared with the nonexcited case, the effect of the longitudinal
random gust is to advance the flutter speed and to postpone the onset of the LCO. Advancement of
the flutter point can be explained in part due to the quadratic airspeed term. The longitudinal gust
increases the average magnitude of the dynamic pressure as defined by oºM /2 or o(º #u )/2, so

that the dynamic pressure associated with the mean airspeed at which flutter occurs is lower than
for the non-excited case. The presence of the random component in the linear airspeed also
influences the location of the flutter point, but may advance or retard it, as discussed by Prussing
and Lin [91]. In their work quasi-steady aerodynamics was employed, and the effect of the
quadratic noise term was neglected, also white noise was assumed.
Some results showing the effects of gust intensity and frequency are presented in Fig. 100, where
the two bifurcation airspeeds are plotted as a function of gust variance for three different scales of
326 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

Fig. 100. Bifurcation diagram of the mean-free stream velocities as a function of the gust variance for three scales of
turbulence (from Ref. [85]).

turbulence. Zero gust variance corresponds to the deterministic case, giving unique flutter and
Hopf-bifurcation airspeeds, º*"4.31. As the variance is increased, the region of single peak PDD
*
becomes larger for all three cases. However, the second bifurcation point, º* , seems to depart from

the nonexcited case at a lower rate than the first bifurcation. The effect of scale of turbulence, which is
a measure of the frequency content, is perhaps more interesting. For equivalent gust variances, the
two bifurcation airspeeds approach each other as the bandwidth of the gust is increased. This
explained by noting that the effective portion of the excitation on the system, defined by the range of
natural frequencies or time scales of the deterministic airfoil relative to the gust bandwidth, decreases
for a larger total bandwidth. Fig. 96 shows that the area under the PSD curves for reduced fre-
quencies lower than 0.18, which is the LCO frequency, decreases for smaller scales of turbulence.
Considering a more practical perspective to the behavior and significance of the single-peak
region it is noted that this region has no analog in the deterministic case, as opposed to the two
extreme regions. It cannot be approximated by either an equilibrium point, nor a limit cycle
oscillation. As mentioned earlier although dynamical behavior is observed between º* and º* , the
 
most probable value that the pitch (or plunge) variable takes during the motion is zero. This is of
significance when considering structural fatigue limitations.

8. Conclusions

Classical theory in aeroelasticity assumes linear structures and aerodynamics in studying the
interaction between the inertia, structural and aerodynamic forces. Many experiments, whether
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 327

performed in wind tunnels or on flight test of aircraft, show phenomena that are not predictable
from linear theory, examples of which are limit cycle oscillations and random-like responses of
aero-surfaces that are self-excited or occur in the presence of air turbulence and gusts.
This review introduces examples of nonlinearities, both in the structural and aerodynamic forces,
and suggests methods to deal with them. The handling of the structure is simplified by considering
only concentrated nonlinearities represented by the three classical types, namely, the cubic, bilinear
and hysteresis restoring force of a nonlinear spring. The study is further simplified by considering
only two-degree-of-freedom oscillations of an airfoil immersed in a subsonic flow field. Multi-
degree-of-freedom oscillations introduce more complexities in the mathematical formulation, but
the fundamental ideas of LCO, bifurcations and chaos can adequately be investigated from
a binary system oscillating in pitch and plunge. The sensitivity to initial conditions in nonlinear
aeroelasticity can only be studied from numerical integration of the equations of motion. The
describing function technique assumes the airfoil motion to be harmonic, and is useful to predict
the airfoil amplitudes when LCO have already set in. This method has the advantage that it can
handle a complex aircraft configuration with a large number of vibration modes by essentially
considering an equivalent linear system where tools routinely used by aircraft designers can
compute flutter boundaries and response amplitudes. Analytical methods presented in this review
for analyzing bifurcations can be used to determine the amplitudes and frequencies of supercritical
post-Hopf-bifurcations, but they are limited to those cases where the structural nonlinearities can
be represented by simple analytical functions, such as a cubic spring. However, much can be
learned from the analytical analysis since it gives good physical insight into the bifurcation process.
Most of the analysis in this review deals with a single nonlinearity in the pitch degree of freedom.
Only one example is given for a coupled cubic nonlinearity in the plunge and pitch motions. Even
in the absence of aerodynamics, the equations of motion describing the coupled Duffing’s oscil-
lators show an extremely complex jump condition. There are infinite possibilities, and perhaps
unknown complexities, of the airfoil motion that may occur when various types of nonlinearities
are coupled. Also, the three basic nonlinearities can possess a large number of geometrical
configurations. This, together with the fact that a number of parameters are needed to define the
airfoil, makes a complete parametric study of the simple two-degree-of-freedom system practically
impossible. In addition, four initial conditions specifying the plunge and pitch displacements and
their time derivatives are required to uniquely define the problem. The examples in this review on
concentrated structural nonlinearities are therefore incomplete, but leave opportunities for the
interested reader to explore more complex airfoil motions that will be encountered with various
coupled nonlinearities, airfoil parameters and initial conditions.
Aerodynamic nonlinearities are only briefly discussed in this review. This is due to the fact that
analysis can only be carried out using complex aerodynamic codes that require large amount of
computing time, and hence there are few examples in the literature that the authors can draw on to
give a detailed and in-depth review of this subject. However, the basic solution techniques are given
in this review. Because of the long computing time required to execute a coupled aerodynamics and
structural dynamics code, only limited cases of interest can be studied.
Numerical simulations can isolate interesting phenomena so that they can be studied individ-
ually. This is unlike experiments in practical situations where a number of system phenomena may
occur simultaneously, thus making the problem much more complex and difficult to handle.
However, we must be prepared to deal with complex physical aeroelastic problems, and the
328 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

methodologies commonly used to handle linear problems fail in parametric identification of

nonlinear systems. Extending the methods developed for nonlinear mechanical systems to aeroelas-
tic systems is a challenging topic for aeroelasticians. The demand for reliable tools to predict
aeroelastic characteristics of modern aircraft that fly in the high subsonic and transonic speed
regimes has not been successfully met since transonic flights began nearly a quarter of a century
ago. Aging and military aircraft carrying heavy external stores potentially face problems arising
from nonlinearities in the structure. Learning the source of LCO, bifurcations and chaos can
possibly lead to the discovery of methods to control these undesirable phenomena. This will result
in more comfortable rides for passengers, better aircraft handling characteristics, and longer
aircraft life span by reducing structural fatigue and other damaging effects associated with LCO.

Acknowledgements

The authors would like to acknowledge the support they received from the Natural Sciences and
Engineering Research Council of Canada.

References

[1] Alighanbari H, Price SJ. The post-Hopf-bifurcation response of an airfoil in incompressible two-dimensional flow.
Nonlinear Dyn 1996;10:381—400.
[2] Argyris J, Faust G, Haase M. An exploration of chaos — an introduction for natural scientists and engineers, Texts
on computational mechanics, vol. VII. Amsterdam: North-Holland, 1994.
[3] Arnold L, Jones C, Mischaikow K, Raugel G. Dynamical systems, Lectures given at the 2nd session of the
Centro Internazionale Matematico Estivo. Berlin: Springer, 1995.
[4] Ashley H. On the role of shock waves in the ‘‘sub-transonic’’ flutter phenomenon. AIAA 79-0765, AIAA/
ASME/ASCE/AHS 20th Structures, Structural Dynamics, and Materials Conf, 4—6 April MO: St. Louis, 1979.
[5] Ballhaus WF, Goorjian PM. Implicit finite-difference computations of unsteady transonic flows about an airfoil.
AIAA J. 1977;15:1728—35.
[6] Barnes TJ. Overview of the activities of the ad hoc committee of international gust specialists. AGARD-R-798,
1994.
[7] Bathe KJ, Wilson EL. Numerical methods in finite element analysis. Englewood Cliffs, NJ. Prentice-Hall,
1976.
[8] Batina JT. Unsteady transonic small-disturbance theory including entropy and vorticity effects. J Aircraft
1989;26:531—8.
[9] Batina JT. Unsteady transonic algorithm improvements for realistic aircraft configurations. J Aircraft
1989;26:131—9.
[10] Beddoes TS. A synthesis of unsteady aerodynamic effects including stall hysteresis. Vertica 1976;1:113—23.
[11] Bendiksen OO, Kousen KA. Transonic flutter analysis using the Euler equations. AIAA 87-0911-CP, AIAA
Dynamics Specialists Conf, Monterey, CA, April 1987.
[12] Bennett RM, Eckstrom CV, Rivera JA Jr, Dansberry BE, Farmer MG, Durham MH. The benchmark aeroelastic
models program — description and highlights of initial results. AGARD-CP-507, 1992:25.1—11.
[13] Bergé P, Pomeau Y, Vidal C. Order within chaos, New York: Wiley, 1986.
[14] Beyn WJ. Numerical methods for dynamical systems. In: Light W, editor. Advances in numerical analysis. Oxford:
Oxford Science Publication, 1991.
[15] Bielawa RL, Johnson SA, Chi RM, Gangwani ST. Aeroelastic analysis for propellers. NASA CR 3729, 1983.
[16] Borland CJ, Rizzeta DP. Nonlinear transonic flutter analysis. AIAA J 1982;20:1606—15.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 329

[17] Borland CJ, Rizzetta DP, Yoshihara H. Numerical solution of three-dimensional unsteady transonic flow over
swept wings. AIAA J 1982;20:340—7.
[18] Breitbach EJ. Effect of structural nonlinearities on aircraft vibration and flutter. Presented at the 45th Structures
and Materials AGARD Panel Meeting. AGARD Report 665. Voss, Norway, 1977.
[19] Breitbach EJ. Flutter analysis of an airplane with multiple structural nonlinearities in the control system. NASA
TP 1620, 1980.
[20] Carr J. Applications of centre manifold theory. New York: Springer, 1981.
[21] Chan Y. Numerical simulation of a two-dimensional airfoil with a hysteresis nonlinearity. National Research
Council Canada, NRC summer student project Report IAR-97-1. September 1997.
[22] De Ferrari G, Chesta, L, Sensburg O, Lotze A. Effects of nonlinearities on wing-store flutter. Mathematical
modeling of linear and non-linear aircraft structures. AGARD-R-687, 1980.
[23] Doedel EJ, Kernevez JP. AUTO: software for continuation and bifurcation problems in ordinary differential
equations. Applied Mathematics Report. California Institute of Technology, 1986.
[24] Dunn P, Dugundji J. Nonlinear stall flutter and divergence analysis of cantilevered graphite/epoxy wings. AIAA
J 1992;30:153—62.
[25] Edwards JW, Malone JB. Current status of computational methods for transonic unsteady aerodynamics and
aeroelastic applications. AGARD-CP-507. March 1992:1.1—24.
[26] Edwards JW. Transonic shock oscillations and wing flutter calculated with an interactive boundary layer coupling
method. NASA Technical Memorandum 110284, NASA Langley Research Center, Hampton, Virginia, 1996.
[27] Edwards JW. Transonic shock oscillations calculations with a new interactive boundary layer coupling method.
AIAA Paper 93-0777, 1997.
[28] Ericsson LE, Reding JP. Dynamic stall at high frequency and large amplitude. J Aircraft 1980;17:136—42.
[29] Ericsson LE, Reding JP. Fluid mechanics of dynamic stall. Part I: unsteady flow concepts. J Fluids Struct
1988;2:1—33.
[30] Ericsson LE, Reding JP. Fluid mechanics of dynamic stall. Part II: prediction of full scale characteristics. J Fluids
Struct 1988;2:113—43.
[31] Fung YC. An introduction to the theory of aeroelasticity. New York: Dover, 1993.
[32] Gangwani ST. Prediction of dynamic stall and unsteady airloads for rotor blades. J Am Helicopter Soc
1982;27:57—64.
[33] Gangwani ST. Synthesized airfoil data method for prediction of dynamic stall and unsteady airloads. Proc Annual
Forum of the AHS. Washington, DC: American Helicopter Society, 1983.
[34] Gangwani ST. Synthesized airfoil data method for prediction of dynamic stall and unsteady airloads. Vertica
1984;8:93—118.
[35] Gelb A, Vander Velde WA. Multiple-input describing functions and nonlinear system design. New York:
McGraw-Hill, 1968.
[36] Gong L, Wong YS, Lee BHK. Dynamics of a coupled system of Duffing’s equations. In Dynamics of continuous,
discrete and impulsive systems, vol. 4. Waterloo: Watam Press, 1998:99—119.
[37] Gormont RE. A numerical model of unsteady aerodynamics and radial flow for application to helicopter rotors.
USAAMRDL-TR-72-67, 1973.
[38] Gray L, Liiva J. Two dimensional tests of airfoils oscillating near stall. vol II, Data Report, USAAVLABS
Technical Report 68-13B, USAAMRDL, Ft. Eustio, VA, 1968.
[39] Guckenheimer J, Holmes P. Nonlinear oscillations, dynamical systems, and bifurcation of vector fields. Berlin:
Springer, 1993.
[40] Guruswamy GP, Goorjian PM, Merrit FJ. ATRAN3S: an unsteady transonic code for clean wings. NASA
TM-86783, 1985.
[41] Guruswamy GP. ENSAERO — A multidisciplinary program for fluid/structural interaction studies of aerospace
vehicles. Comput Systems Engng 1990;1:237—56.
[42] Hartman P. Ordinary differential equations, 2nd ed. Boston: Birkäuser, 1982.
[43] Hayashi C. Nonlinear oscillations in physical systems. New York: McGraw Hill, 1964.
[44] Hoblit FM. Gust loads on aircraft: concepts and applications. AIAA Education Series. Washington: AIAA,
1988.
330 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

[45] Horsthemke W, Lefever R. Noise-induced transitions; Theory and applications in physics, chemistry and biology.
Berlin: Springer, 1984.
[46] Houbolt JC. A recurrence matrix solution for the dynamic response of elastic aircraft. J Aeronaut Sci
1950;17:540—50.
[47] Hounjet MHL. A field panel/finite difference method for potential unsteady transonic flow. AIAA J
1985;23:537—45.
[48] Isaacs R. Airfoil theory for flows of variable velocity. J Aeronaut Sci 1945:13:113—7.
[49] Isogai K, Suetsugu K. Numerical simulation of transonic flutter of a supercritical wing. NAL Report TR-726T,
National Aerospace Laboratory, Japan, 1982.
[50] Isogai K. Numerical simulation of shock-stall flutter of an airfoil using Navier-Stokes equations. J Fluids Struct
1993;7:595—609.
[51] Johnson EC. Sinusoidal analysis of feedback-control systems containing nonlinear elements. Trans AIEE
1952;71:169—81.
[52] Jones DJ, Lee BHK. Time marching numerical solution of the dynamic response of nonlinear systems. Aeronauti-
cal Note NAE-AN-25, NRC No. 24131, National Research Council Canada, 1985.
[53] Jones RT. The unsteady lift of a wing of finite aspect ratio. NACA Rept 681, 1940.
[54] Keleris JP. Nonlinear dynamics of an airfoil forced to oscillate in dynamic stall. M Engng Thesis. McGill
University, Montreal, Canada, 1994.
[55] Kim SH, Lee I. Aeroelastic analysis of a flexible airfoil with a freeplay non-linearity. J Sound Vib 1996;193:823—46.
[56] Kousen KA, Bendiksen OO. Nonlinear aspects of the transonic aeroelastic problem. Proc AIAA/ASME/ASCE/
AHS 29th Structures, Structural Dynamics and Materials Conf, Williamsburg, Virginia, 18—20 April 1988:760—9.
[57] Kousen KA, Bendiksen, OO. Limit cycle phenomena in computational transonic aeroelasticity. AIAA-89-1185-
CP, 1989.
[58] Kryloff N, Bogoliuboff N. Introduction to nonlinear mechanics (Solomon Lifschitz, Trans.). Princeton: Princeton
University Press, 1947.
[59] Kuznetsov YA. Elements of applied bifurcation THEORY. Berlin: Springer, 1995.
[60] Laurenson RM, Trn RM. Flutter analysis of missile control surfaces containing structural nonlinearities. AIAA
J 1980;18:1245—51.
[61] Laurenson RM, Hauenstein AJ, Gubser JL. Effects of structural nonlinearities on limit-cycle response of
aerodynamic surfaces. AIAA 86-0899, 1986.
[62] Lee BHK. A study of transonic flutter of a two-dimensional airfoil using the U-g and p-k methods. Aeronautical
Rept. LR-615, NRC No. 23959, National Research Council Canada, 1984.
[63] Lee BHK, LeBlanc P. Forced oscillation of a two-dimensional airfoil with nonlinear aerodynamic loads.
Aeronautical Report LR-617, National Research Council of Canada, 1986.
[64] Lee BHK, LeBlanc P. Flutter analysis of a two-dimensional airfoil with cubic nonlinear restoring force.
Aeronautical Note NAE-AN-36, NRC No. 25438, National Research Council of Canada, 1986.
[65] Lee BHK, Desrochers J. Flutter analysis of a two-dimensional airfoil containing structural nonlinearities.
Aeronautical Report LR-618, NRC No. 27833, National Research Council of Canada, 1987.
[66] Lee BHK, Tron A. Effects of structural nonlinearities on flutter characteristics of the CF-18 aircraft. J Aircraft
1989;26:781—6.
[67] Lee BHK. Oscillatory shock motion caused by transonic shock-boundary-layer interaction. AIAA J 1990; 28:942—4.
[68] Lee BHK, Murty H, Jiang H. Role of Kutta waves on oscillatory shock motion on an airfoil. AIAA
J 1994;32:789—96.
[69] Lee BHK, Gong L, Wong YS. Analysis and computation of nonlinear dynamic response of a two-degree-of-
freedom system and its application in aeroelasticity. J Fluids Struct 1997;11:225—46.
[70] Lee BHK, Jiang LY, Wong YS. Flutter of an airfoil with a cubic nonlinear restoring force, AIAA Paper 98-1725,
39th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conf, 20—23 April, Long
Beach, CA, 1998.
[71] Lee BHK, Wong YS. Neural network parameter extraction with application to flutter signals. J Aircraft
1998;35:165—8.
[72] Lee CL. An iterative procedure for nonlinear flutter analysis. AIAA J 1986;24:833—40.
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 331

[73] Leishman JG, Beddoes TS. A semi-empirical model for dynamic stall. J Am Helicopter Soc 1989;34:3—17.
[74] Li X, Fleeter S. Dynamic stall generated airfoil oscillations including chaotic responses. AIAA 97-1022, 38th
AIAA/ASME/ASCE/AHS Structures, Structural Dynamics and Materials Conf, Kissimmee, FL, April, 1997.
[75] Liu JK, Zhao LC. Bifurcation analysis of airfoils in incompressible flow. J Sound Vib 1992;154:117—24.
[76] Liu L, Wong YS, Lee BHK. Application of the center manifold theory in nonlinear aeroelasticity.
CEAS/AIAA/ICASE/NASA Langley International Forum on Aeroelasticity and Structural Dynamics. William-
sburg, VA, June 1999.
[77] Loring SJ. General approach to the flutter problem. SAE Trans 1941;49:345—56.
[78] Mazelsky B, Drischler JA. Numerical determination of indicial lift and moment functions for a two-dimensional
sinking and pitching airfoil at mach numbers 0.5 and 0.6. NACA TN 2739, 1952.
[79] Mehta UB. Dynamic stall of oscillating airfoil, AGARD-CP-227. Unsteady Aerodynamics. 1977:23.1—32.
[80] Miller RK. The steady state response of systems with hardening hysteresis. Paper ASME 77-DET-71, 1977.
[81] Moon FC. Chaotic and fractal dynamics, New York: Wiley, 1992.
[82] Neil DJ, Johnson EH, Canfield R. ASTROS-A multidisciplinary automated structural design tool. J Aircraft
1990;27:1021—7.
[83] Osher S, Hafez M, Whitlow W Jr. Entropy condition satisfying approximations for the full potential equation of
transonic flow. Math Comput 1985;44:1—29.
[84] Poincaré H. Les méthodes nouvelles de la méchanic célèste. Paris, 1893.
[85] Poirel DC, Price SJ. Post-instability behavior of a structurally nonlinear airfoil in longitudinal turbulence.
J Aircraft 1997;34:619—26.
[86] Prenter PM. Splines and variational methods. New York: Wiley, 1975.
[87] Press WH, Flannery BP, Teukolsky SA, Vetterling WT. Numerical recipes, The art of scientific computing.
Cambridge, New York, 1986.
[88] Price SJ, Lee BHK, Alighanbari H. Post instability behaviour of a two-dimensional airfoil with a structural
nonlinearity. J Aircraft 1994;31:1395—401.
[89] Price SJ, Alighanbari H, Lee BHK. The aeroelastic response of a two-dimensional airfoil with bilinear and cubic
structural nonlinearities. J Fluids Struct 1995;9:175—93.
[90] Price SJ, Keleris JP. Non-linear dynamics of an airfoil forced to oscillate in dynamic stall. J Sound Vib
1996;194:265—83.
[91] Prussing JE, Lin YK. A closed-form analysis of rotor blade flap-lag stability in hover and low-speed forward flight
in turbulent flow. J Am Helicopter Soc 1983;28:42—6.
[92] Reddy TSR, Kaza KRV. A comparative study of some dynamic stall methods. NASA TM-88917, 1987.
[93] Schuster DM, Vadyak J, Atta EJ. Flight loads prediction methods for flighter aircraft. WRDC-TR-89-3104, 1989.
[94] Seydel R. From equilibrium to chaos. New York: Elsevier, 1988.
[95] Shen SF, Hsu CC. Analytical results of certain nonlinear flutter problems. J Aeronaut Sci 1958;25:136—7.
[96] Shen SF. An approximate analysis of nonlinear flutter problems. J Aerosp Sci 1959;26:25—32, 45.
[97] Shen SF. Author’s reply to: Remarks on Analytical results of certain nonlinear flutter problems. J Aerosp Sci
1959;26:52—3.
[98] Steinhoff J, Jameson A. Multiple solutions of the transonic potential equation. AIAA J 1982;20:1521—5.
[99] Stoker JJ. Nonlinear vibrations in mechanical and electrical systems. New York: Interscience Publishers,
1950.
[100] Straganac TW, Mook DT, Mitchum MV. The numerical simulation of subsonic flutter. AIAA 87-1428, AIAA 19th
Fluid Dynamics, Plasma Dynamics and Lasers Conf, Honolulu, Hawaii, 8—10 June 1987.
[101] Tang DM, Dowell EH. Flutter and stall response of a helicopter blade with structural nonlinearity. J Aircraft
1992;29:953—60.
[102] Tang DM, Dowell EH. Comparison of theory and experiment for non-linear flutter and stall response of
a helicopter blade. J Sound Vib 1993;165:251—76.
[103] Tang DM, Dowell EH. Experimental and theoretical study for nonlinear aeroelastic behaviour of a flexible rotor
blade. AIAA J 1993;31:1133—42.
[104] Tang DM, Dowell EH. Experimental investigation of three-dimensional dynamic stall model oscillating in pitch.
J Aircraft 1995;32:1062—71.
332 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334

[105] Tang DM, Dowell EH. Nonlinear response of a non-rotating rotor blade due to a periodic gust. J Fluids Struct
1996;10:721—42.
[106] Tang DM, Dowell EH. Nonlinear rotor aeroelastic analysis with stall and advanced wake dynamics. J Aircraft
1997;34:679—87.
[107] Theodorsen T. General theory of aerodynamic instability and the mechanism of flutter. NACA Report 496, 1935.
[108] Thompson JMT, Stewart HB. Nonlinear dynamics and chaos. Chichester: Wiley, 1986.
[109] Tran CT, Petot D. Semi-empirical model for the dynamic stall of airfoils in view of the application to the
calculation of responses of a helicopter blade in forward flight. Vertica 1981;5:35—53.
[110] Tran CT, Falchero D. Application of the ONERA dynamic stall model to a helicopter blade in forward flight.
Vertica 1982;6:219—39.
[111] Ueda T, Dowell EH. Flutter analysis using nonlinear aerodynamic forces. J Aircraft 1984;21:101—9.
[112] Ueda Y. Steady motions exhibited by Duffing’s equation: a picture cook of regular and chaotic motions. In:
Holmes PJ, editor. New approaches to nonlinear problems in dynamics. Philadelphia: SIAM, 1980:311—22.
[113] Van der Wall BG. The influence of variable flow velocity on unsteady airfoil behavior. DLR-FB 92-22, 1992.
[114] Van der Wall BG, Leishman JG. On the influence of time-varying flow velocity on unsteady aerodynamics. J Am
Helicopter Soc 1994;39:25—36.
[115] Verhulst F. Nonlinear differential equations and dynamical systems. Berlin: Springer, 1990.
[116] Whitlow W Jr, Hafez MH, Osher SJ. An entropy correction method for unsteady full potential flows with strong
shocks. AIAA Paper 86-1768-CP, 1986.
[117] Wiggins S. Introduction to applied nonlinear dynamical systems and chaos. New York: Springer, 1990.
[118] Wong YS, Lee BHK. Development of a three-dimensional unsteady transonic aerodynamics computer code for
flutter analysis. Proc 17th Congress of the Int Council of the Aeronautical Sciences Stockholm, Sweden,
1990;1:19—29.
[119] Wong YS, Lee BHK, Gong L. Dynamic response of a two-degree-of-freedom system with a cubic nonlinearity. 3rd
Int Conf on Computational Physics. Chung Li, Taiwan, 1995.
[120] Woolston DS, Runyan HL, Byrdsong TA. Some effects of system nonlinearities in the problem of aircraft flutter.
NACA TN 3539, 1955.
[121] Woolston DS, Runyan HL, Andrews RE. An investigation of effects of certain types of structural nonlinearities on
wing and control surface flutter. J Aeronaut Sci 1957;24:57—63.
[122] Wu JC, Kaza KRV, Sankar LN. Technique for the prediction of airfoil flutter characteristics in separated flows.
J Aircraft 1989;26:168—77.
[123] Yang ZC, Zhao LC. Analysis of limit cycle flutter of an airfoil in incompressible flow. J Sound Vib 1988;123:1—13.
[124] Yates EC Jr. AGARD standard aeroelastic configurations for dynamic response. Candidate Configuration I.
— Wing 445.6. NASA TM 100492, 1987.
[125] Zhao LC, Yang ZC. Chaotic motions of an airfoil with nonlinear stiffness in incompressible flow. J Sound Vib
1990;138:245—54.

Appendix

The coefficients of Eqs. (19) and (20) are given as follows:

1 a uN 2
c "1# , c "x ! F , c "2f # (1!t !t ),
 k  ? k  K º* k  

1#2(1/2!a ) (1!t !t ) 2
c " F   , c " (t e #t e ),
 k  k  

2
c " [(1!t !t )#(1/2!a ) (t e #t e )] ,
 k   F  
 B.H.K Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334 333

2 2 2
c " t e [1!(1/2!a )e ], c " t e [1!(1/2!a )e ], c "! t e,
 k  F   k  F   k 

2 x a 1#8a
c "! t e , d " ?! F , d "1# F,
 k    r kr  8kr
? ? ?
f 1!2a (1#2a ) (1!2a ) (1!t !t )
d "2 ? # F! F F  ,
 º* 2kr 2kr
? ?
(1#2a ) (1!t !t ) (1#2a ) (1!2a ) (t e #t e )
d "! F  ! F F   ,
 kr 2kr
? ?
(1#2a ) (1!t !t ) (1#2a ) (t e #t e )
d "! F   , d "! F   ,
 kr  kr
? ?
(1#2a )t e [1!(1/2!a )e ] (1#2a )t e [1!(1/2!a )e ]
d "! F   F  , d "! F  F  ,
 kr  kr
? ?
(1#2a )t e (1#2a ) t e
d " F  , d " F 
 kr  kr
? ?
The coefficients of Eqs. (27) and (28) are given below:



1 a 1 2(1!t !t ) u
m "1# , m "x ! F , m "(e #e ) 1# #   #21 ,
 k  ? k    k k K º*


1 (1/2!a ) a
m " #2(1!t !t ) F #(e #e ) x ! F ,
 k   k   ? k



1 2(e #e !t e !t e ) u
m "e e 1# #      #21 (e #e ) ,
  k k K   º*


a (e #e )#2(1!t !t ) 2(1/2!a ) (e #e !t e !t e )
m "e e x ! F #    # F     ,
  ? k k k


2e e u e e 2e e (1/2!a ) 2(e #e !t e !t e )
m "  #2e e 1 , m "  #   F#     ,
 k   K º*  k k k



2e e a 1 a#1/8
m "0, m "   , n " x ! F , n " 1# F ,
  k  ? k r  kr
? ?


a 1 2(1/2#a ) (1!t !t )
n "(e #e ) x ! F ! F   ,
   ? k r kr
? ?
334 B.H.K. Lee et al. / Progress in Aerospace Sciences 35 (1999) 205—334


a#1/8 1 (1/2!a )!2(1/2#a ) (1/2!a ) (1!t !t )
n "(e #e ) 1# F #21 # F F F   ,
   kr ? º* kr
? ?


e e a 2(e #e ) (1/2#a ) 2(1/2#a ) (t e #t e )
n " x! F !   F# F   ,
 r ? k kr kr
? ? ?

  
a#1/8 21 (1/2!a ) (1!2(1/2#a ))
n "e e 1# F #(e #e ) ?# F F
  kr   º* kr
? ?
2(1/2#a ) [!1#t #t #(1/2!a ) (t e #t e )]
# F   F   ,
kr
?
!2e e (1/2#a )
n "  F ,
 kr
?

 
21 (1/2!a ) [1!2(1/2#a )] 2(1/2#a ) [t e #t e !(e #e )]
n "e e ?# F F # F     ,
   º* kr kr
? ?
!2(1/2#a )e e
n "0, n " F  .
  kr
?