s11071-012-0648-z

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Nonlinear Dyn (2013) 71:159–173

DOI 10.1007/s11071-012-0648-z

O R I G I N A L PA P E R

An analytical and experimental investigation into limit-cycle


oscillations of an aeroelastic system
Abdessattar Abdelkefi · Rui Vasconcellos ·
Ali H. Nayfeh · Muhammad R. Hajj

Received: 15 July 2012 / Accepted: 12 October 2012 / Published online: 24 October 2012
© Springer Science+Business Media Dordrecht 2012

Abstract We perform an analytical and experimental 1 Introduction


investigation into the dynamics of an aeroelastic sys-
tem consisting of a plunging and pitching rigid airfoil Predicting the flutter speed and ensuing complex non-
supported by a linear spring in the plunge degree of linear responses of aeroelastic systems is of particular
freedom and a nonlinear spring in the pitch degree of importance in many applications [1–7]. In the case of
freedom. The experimental results show that the onset aerospace vehicles, it is desired to expand the flight
of flutter takes place at a speed smaller than the one boundary by increasing the flutter speed. On the other
predicted by a quasi-steady aerodynamic approxima- hand, for aeroelastic energy harvesters, it is desired to
tion. On the other hand, the unsteady representation generate power at the lowest possible speeds [8–13].
of the aerodynamic loads accurately predicts the ex- In the latter application, modeling the subsequent re-
perimental value. The linear analysis details the dif- sponses, such as limit-cycle oscillations, secondary
ference in both formulation and provides an expla- Hopf bifurcations, and other complex phenomena, is
nation for this difference. Nonlinear analysis is then also required. Flutter in a pitching and plunging aeroe-
performed to identify the nonlinear coefficients of the lastic system is the result of the coupling of the two de-
pitch spring. The normal form of the Hopf bifurcation grees of freedom by the aerodynamic loads. As such,
is then derived to characterize the type of instability. the modeling of these aerodynamic loads impacts the
It is demonstrated that the instability of the considered prediction of the instability and flutter speed and fre-
aeroelastic system is supercritical as observed in the quency. Their modeling also impacts the ability to ac-
experiments. curately predict the frequency content and amplitudes
of the subsequent motions, such as limit-cycle oscilla-
Keywords Experimental identification · tions in the nonlinear regime.
Aeroelasticity · Normal form · Hopf bifurcation In general, the aerodynamic loads used for predict-
ing the onset of flutter are determined from the quasi-
steady approximation [14–16]. However, in a recent
A. Abdelkefi · A.H. Nayfeh · M.R. Hajj ()
experiment on a two-dimensional pitching and plung-
Department of Engineering Science and Mechanics,
MC 0219, Virginia Polytechnic Institute and State ing rigid aeroelastic system supported by springs, we
University, Blacksburg, VA 24061, USA have observed that flutter takes place at a speed smaller
e-mail: mhajj@vt.edu than the one predicted by modeling the aerodynamic
loads with the quasi-steady approximation. This ap-
R. Vasconcellos
Laboratory of Aeroelasticity, University of Sao Paulo, proximation does not account for wake effects. Con-
São Carlos, Brazil sidering this fact, we took the approach of investi-
160 A. Abdelkefi et al.

gating whether an unsteady aerodynamic formulation The aeroelastic tests were carried out using an
that includes wake effects on the aerodynamic loads open-circuit wind tunnel with a 520 mm × 515 mm
would resolve this discrepancy and would predict the test section. Signals of the rotational motion of the
observed responses, particularly the flutter speed and elastic axis were collected using an encoder. An ac-
frequency. Furthermore, we aim to identify all param- celerometer was used to measure the plunge motion.
eters of the mathematical model of this experimental Motions of the wing were induced by applying ini-
system that predict its linear and nonlinear responses. tial displacements in the plunge degree of freedom.
We then perform a nonlinear analysis using the normal This process was repeated starting with a low speed of
form of the Hopf bifurcation (flutter) of the dynamics about 8 m/s. It was observed that all disturbances with
of this system to identify its nonlinearity. the same initial displacements would decay for speeds
In Sect. 2, the experimental set up is introduced. less than 10.9 m/s. Above this speed, limit-cycle os-
In Sect. 3, the equations of motion of the aeroelastic cillations (LCO) were observed.
system are presented. Aerodynamic models based on
the quasi-steady representation and an unsteady for-
mulation are described in Sect. 4. In Sect. 5, we per- 3 Analytical model
form linear analysis to determine the flutter speed and
frequency. In Sect. 6, and based on numerical inte- We model the wing section as an aeroelastic system
grations, we identify the nonlinear coefficients of the constrained to move with two degrees of freedom,
namely the plunge (h) and pitch (α) motions, as shown
pitch spring and compare the numerically obtained re-
in Fig. 2. The equations of motion governing this sys-
sponses with those obtained experimentally. We also
tem are [17, 18]
derive the normal form of the Hopf bifurcation, and
     
use it to obtain analytical expressions for the steady- mT mw xα b ḧ ch 0 ḣ
state plunge and pitch motions and to determine the +
mw xα b Iα α̈ 0 cα α̇
type of the ensuing instability. Conclusions are pre-
sented in Sect. 7.     
kh 0 h −L
+ = (1)
0 kα (α) α M
2 Experimental setup
where mT is the total mass of the wing with its sup-
port structure, mw is the wing mass alone, Iα is the
The experimental apparatus has been designed to sim- mass moment of inertia about the elastic axis, b is the
ulate a wing section in a two-dimensional incompress- semi-chord length, and xα is the nondimensional dis-
ible flow. The airfoil consists of an aluminum rigid tance between the center of mass and the elastic axis.
wing mounted vertically at the 1/4 chord point from The viscous damping forces are described through the
the leading edge by using an aluminum shaft. At the coefficients ch and cα for plunge and pitch, respec-
edges, the shaft is connected with bearings to the sup- tively. Table 1 gives values of all of the parameters
port of the plunge mechanism, which is a bi-cantilever used in the following analysis. These parameters were
beam made of two steel leaf springs. The distance be- determined using the geometric dimensions mass and
tween the two cantilever beams is 20.3 cm. The chord material of the wing. The damping coefficients were
wise center of gravity is located at the mid-chord. The determined using the log-decrement approach. In ad-
torsional spring consists of a steel leaf spring inserted dition, L and M are the aerodynamic lift and moment
tightly into a slot in the main shaft at the bottom of the about the elastic axis. The two spring forces for plunge
wing section. The free end of the leaf spring is placed and pitch are represented by kh h and kα α, respectively.
into a support that allows for freeplay variations. For In this experimental setup, the sole structural source of
the cases considered in this paper, the support is to- nonlinearity is the concentrated pitch nonlinearity. For
tally closed, therefore the freeplay is zero. A schematic the analysis, this nonlinearity is modeled by quadratic
and pictures of the experimental setup and pitch spring and cubic terms in the torsional spring coefficient; that
mechanism are shown in Figs. 1(a), 1(b) and 1(c), is,
respectively. The physical properties of the wing are
listed in Table 1. kα (α) = kα0 + kα1 α + kα2 α 2 (2)
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 161

Fig. 1 (a) A schematic of


the experimental setup,
(b) the experimental setup,
and (c) the pitch spring
mechanism

Table 1 Typical section


parameters of the Span Wing span (m) 0.6
aeroelastic wing b Wing semi-chord (m) 0.0325
a Position of elastic axis relative to the semi-chord −0.5
ρp Air density (kg/m3 ) 1.204
mw Mass of the wing (kg) 1.0662
mT Mass of wing and supports (kg) 3.836
Iα Moment of inertia about the elastic axis (kg m2 ) 0.0004438
cα Pitch damping coefficient (kg m2 /s) 0.0115
ch Plunge damping coefficient (kg/s) 0.011
kα0 Stiffness in pitch (N/m) 0.942
kh Stiffness in plunge (N/m) 895.10
xα Nondimensional distance between center of gravity and elastic axis 0.5

4 Representation of the aerodynamic loads and


     
1 2 1
The lift and moment on a wing section, as derived by Mα = πρb ba ḧ − U b
2
− a α̇ − b + a α̈
2
2 8
Theodorsen [19], are given by  
1
+ 2πρb2 U a + QC (4)
L = πρb2 [ḧ + U α̇ − ba α̈] + 2πρU bQC (3) 2
162 A. Abdelkefi et al.
 
ch + 2πρbU 2(1 − a)πρb2 U
D=
−2π(a + 12 )ρb2 U cα + a(2a − 1)πρb3 U
(10)
 2

kh0 2πρbU
K= (11)
0 kα0 − 2π( 12 + a)ρb2 U 2

Multiplying (7) from the left with the inverse M −1


of M, we obtain

Fig. 2 Schematic of an aeroelastic system under uniform air- q̈ = D ∗ q̇ + K ∗ q − M −1 Q(q, q) − M −1 N (q, q, q)


flow
(12)

where U is the freestream velocity, C is the Theo- where D ∗ = −M −1 D and K ∗ = −M −1 K.


dorsen function, and Introducing the following state variables:
 
Q = U α + ḣ + α̇b
1
−a (5) ⎤ ⎡ ⎤

2 X1 h
⎢ X2 ⎥ ⎢ ḣ ⎥
X=⎢ ⎥ ⎢ ⎥
⎣ X3 ⎦ = ⎣ α ⎦ (13)
is the effective angle of attack. In the frequency do-
main, the Theodorsen function is written as X4 α̇

C(k) = F (k) + iG(k) (6) we rewrite (12) as

where k = Uω is the reduced frequency of harmonic os-        


Ẋ2 X2 X1 0
cillations, ω is the oscillation frequency, and F and G = D∗ + K∗ − M −1
Ẋ4 X4 X3 kα1 X32
are functions of the Hankel and modified Bessel func-  
0
tions. We consider two representations of the aerody- − M −1 (14)
namic loads, namely, the quasi-steady and unsteady kα2 X33
approximations.
Additionally, we have
4.1 Quasi-steady approximation
Ẋ1 = X2
(15)
In the quasi-steady approximation, the Theodorsen Ẋ3 = X4
function C is set equal to unity. This assumption ne-
glects any lag between the unsteady oscillations and its Combining (14) and (15), we obtain the following
effect on the aerodynamic loads, thereby limiting the
state-space form:
modeling to small k values. Setting C = 1, we rewrite
the governing equations of motion as
Ẋ = B(U )X + Nq (X, X) + Nc (X, X, X) (16)
M q̈ + D q̇ + Kq + Q(q, q) + N (q, q, q) = 0 (7)
where
where

⎡ ⎡ ⎤
      0 0
q=
h
Q=
0
=
0 ⎢Nq2 ⎥ ⎢Nc2 ⎥
α kα1 α 2
N
kα2 α 3
(8) Nq (X, X) = ⎢
⎣ 0 ⎦
⎥ and Nc (X, X, X) = ⎢
⎣ 0 ⎦

 
mT + πρb2 Sα − aπρb3 Nq4 Nc4
M= (9)
Sα − aπρb3 Iα + π( 18 + a 2 )ρb4 (17)
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 163
 τ
are, respectively, quadratic and cubic vector functions ∂Q(σ )
= Q(0)φ(τ ) + φ(τ − σ ) dσ (21)
of the state variables and 0 ∂σ
⎡ ⎤
0 1 0 0 Using integration by parts, we rewrite (21) as
⎢ K ∗ D∗ K ∗ D∗ ⎥  τ
⎢ 12 ⎥
B(U ) = ⎢ 11 11 12
⎥ ∂φ(τ − σ )
⎣ 0 0 0 1 ⎦ Lc = Q(τ )φ(0) + Q(σ ) dσ (22)
∗ ∗ ∗ ∗ 0 ∂σ
K21 D21 K22 D22
The Sears approximation to φ(τ ) is given by [20]
The matrices K ∗ and D ∗ depend, respectively, on
2
U and U . The matrix B(U ) has a set of four eigen- φ(τ ) ≈ c0 − c1 e−c2 τ − c3 e−c4 τ (23)
values λi , i = 1, 2, 3, 4. These eigenvalues are com-
plex conjugates (λ2 = λ̄1 and λ4 = λ̄3 ). The real parts where c0 = 1, c1 = 0.165, c2 = 0.0455, c3 = 0.335,
of these eigenvalues correspond to the damping coeffi- and c4 = 0.3.
cients and the imaginary parts are the coupled frequen- Using the Sears and Pade approximations, Edwards
cies of the aeroelastic system. The solution of the lin- et al. [21] expressed (22) as
ear part is asymptotically stable if all of the real parts
 2
of the λi are negative. On the other hand, if one or U
more of the real parts are positive, the linearized sys- Lc = (c0 − c1 − c3 )Q(t) + c2 c4 (c1 + c3 ) x̄
b
tem is unstable. The speed at which one or more eigen-
+ (c1 c2 + c3 c4 )U x̄˙ (24)
values have zero real parts, corresponds to the onset of
the linear instability and is termed the flutter speed Uf where x̄ and x̄˙ are two augmented variables in the state
if the imaginary part is different from zero.
space. They are related to the system variables by the
following second-order differential equation:
4.2 Unsteady formulation
U2 U U
Taking into account the unsteady effects of the flow, x̄¨ = −c2 c4 2 x̄ − (c2 + c4 ) x̄˙ + α
b b b
one can represent the aerodynamic loads for arbitrary  
time-dependent motion in the form [20, 21] 1 ḣ
+ − a α̇ + (25)
2 b
L = πρb [ḧ + U α̇ − ba α̈]
2
 +∞ Using (21) and (24), we express the lift and moment
+ 2πρU b C(k)f (ω)eiωt dω (18) by
−∞

and L = πρb2 [ḧ + U α̇ − ba α̈] + 2πρU b(c0 − c1 − c3 )Q


      + 2πρU 3 c2 c4 (c1 + c3 )x̄
1 2 1
M = πρb ba ḧ − U b
2
− a α̇ − b + a α̈
2
+ 2πρU 2 b(c1 c2 + c3 c4 )x̄˙ (26)
2 8
   +∞
1 and
+ 2πρb U a +
2
C(k)f (ω)eiωt dω
2 −∞      
1 1
(19) Mα = πρb2 ba ḧ − U b − a α̇ − b2 + a 2 α̈
2 8
 
where 1
+ 2πρb2 U a + (c0 − c1 − c3 )Q
 +∞ 2
 
f (ω) = Q(t)e−iωt dt (20) 1
−∞ + 2πρbU 3 a + c2 c4 (c1 + c3 )x̄
2
 +∞  
Using the Wagner function φ(t) = −∞ C(k)ik e
ikt dk 1
+ 2πρb2 U 2 a + (c1 c2 + c3 c4 )x̄˙ (27)
and the convolution theorem, one obtains [20, 21] 2
 +∞
We note here that the first two terms on the right-
Lc = C(k)f (ω)eiωt dω
−∞ hand sides of (26) and (27) are the same as the quasi-
164 A. Abdelkefi et al.

steady terms except that the second terms in the un- Substituting (26) and (27) into (1) and augmenting
steady formulation are one-half those in the quasi- the result with (25), we obtain the following general
steady approximation. The third and fourth terms in- form of the equations of motion:
clude the contributions of the augmented variables,
which model the unsteady effects. It is also clear that M1 p̈ + D1 ṗ + K1 p + Q1 (p, p) + N1 (p, p, p) = 0
the relative contributions of the last two terms in com-
parison to the contributions of the quasi-steady terms (28)
depend on U , b, and Q. Of particular importance is
the dependence on the mid-chord b. where

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
h 0 0
⎢ ⎥ ⎢ 2⎥ ⎢ 3⎥
p = ⎣α ⎦ Q1 = ⎣ kα1 α ⎦ N1 = ⎣ kα2 α ⎦ (29)
x̄ 0 0
⎡ ⎤
mT + πρb2 mxα b − aπρb3 0
⎢ ⎥
M1 = ⎣ mxα b − aπρb3 Iα + π( 18 + a 2 )ρb4 0 ⎦ (30)
0 0 1
⎡ ⎤
ch + 2πρbU (c0 − c1 − c3 ) (1 + (c0 − c1 − c3 )(1 − 2a))πρb2 U 2πρU 2 b(c1 c2 + c3 c4 )
⎢ ⎥
D1 = ⎣ −2π(a + 12 )ρb2 (c0 − c1 − c3 )U cα + ( 12 − a)(1 − (c0 − c1 − c3 )(1 + 2a))πρb3 U −2πρb2 U 2 (a + 12 )(c1 c2 + c3 c4 ) ⎦
− b1 a− 1
2 (c2 + c4 ) Ub
(31)
⎡ ⎤
kh 2πρbU 2 (c0 − c1 − c3 ) 2πρU 3 c2 c4 (c1 + c3 )
⎢ ⎥

K1 = ⎣ 0 kα0 − 2π( 12 + a)ρ(c0 − c1 − c3 )b2 U 2 −2πρbU 3 (a + 12 )c2 c4 (c1 + c3 ) ⎥ (32)

2
0 − Ub c2 c4 Ub2

Multiplying (28) from the left with the inverse M1−1 we express (33) as
of M1 , we obtain ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
Ż2 Z2 Z1 0
⎣ Ż4 ⎦ = D1∗ ⎣ Z4 ⎦ + K1∗ ⎣ Z3 ⎦ − M −1 ⎣ kα1 X 2 ⎦
1 3
p̈ = D1∗ ṗ + K1∗ p Ż6 Z6 Z5 0
⎡ ⎤
− M1−1 Q1 (p, p) − M1−1 N1 (p, p, p) (33) 0
− M1−1 ⎣ kα2 X33 ⎦ (35)
0
where D1∗ = −M1−1 D1 and K1∗ = −M1−1 K1 .
In terms of the state variables Additionally, we have

⎡ ⎡ ⎤ ⎤ Ż1 = Z2
h
Z1
⎢ Z ⎥ ⎢ ḣ ⎥ Ż3 = Z4 (36)
⎢ 2⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢ Z3 ⎥ ⎢ α ⎥ Ż5 = Z6
⎢ ⎥ ⎢ ⎥
Z=⎢ ⎥=⎢ ⎥ (34)
⎢ Z4 ⎥ ⎢ α̇ ⎥ Combining these equations, we obtain the following
⎢ ⎥ ⎢ ⎥
⎢ Z5 ⎥ ⎢ x̄ ⎥ form:
⎣ ⎦ ⎣ ⎦
Z6 x̄˙ Ż = B1 (U )Z + N1q (Z, Z) + N1c (Z, Z, Z) (37)
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 165

where
⎡ ⎤
0 1 0 0 0 0
⎢ K ∗ D∗ ∗ ∗ ∗ ∗ ⎥
⎢ 111 111 K112 D112 K113 D113 ⎥
⎢ 0 0 0 1 0 0 ⎥
B1 (U ) = ⎢
⎢ K ∗ D∗ ∗ ∗ ∗ ∗ ⎥

⎢ 121 121 K122 D122 K123 D123 ⎥
⎣ 0 0 0 0 0 1 ⎦

K131 ∗
D131 ∗
K132 ∗
D132 ∗
K133 ∗
D133
⎡ ⎤ ⎡ ⎤
0 0
⎢ N1q2 ⎥ ⎢ N1c2 ⎥
⎢ ⎥ ⎢ ⎥
⎢ 0 ⎥ ⎢ 0 ⎥
N1q (Z, Z) = ⎢ ⎥ and N1c (Z, Z, Z) = ⎢ ⎥
⎢ N1q4 ⎥ ⎢ N1c4 ⎥
⎣ 0 ⎦ ⎣ 0 ⎦ Fig. 3 Variation of one of the complex-conjugate eigenvalues,
which cross the imaginary axis, with the freestream velocity (all
0 0
speeds are in meters per second) using the quasi-steady approx-
(38) imation (red dashed line) and the unsteady formulation (black
solid line) (Color figure online)
are quadratic and cubic vector functions of the state
variables, which describe the structural nonlinearity.
The matrix K1∗ depends on U , U 2 , and U 3 and the velocity exceeds a critical value (flutter speed), with
matrix D1∗ depends on U and U 2 . It is clear that, in the the other eigenvalues remain in the left-half plane. The
unsteady formulation, the size of B1 (U ) has increased results of Fig. 3 show that the flutter speeds predicted
because the aerodynamic loads are treated as an addi- with the quasi-steady and unsteady formulations are
tional degree of freedom through the use of the aug- quite different. Another way to predict the flutter speed
mented variables. The matrix B1 (U ) has a set of six is to study variation of the effective damping with the
eigenvalues λi , i = 1, 2 . . . , 6. The first four eigenval- freestream velocity. The results are shown in Figs. 4(a)
ues are complex conjugates (λ2 = λ1 , λ4 = λ3 ), which and 4(b) using the quasi-steady and unsteady formula-
are directly related to the pitch α and plunge h mo- tions, respectively. It follows from these figures that
tions. The real parts of these eigenvalues correspond the damping changes from negative to positive at U =
to the damping coefficients and the imaginary parts 14.35 m/s and U = 10.90 m/s for the quasi-steady
are the coupled frequencies of the aeroelastic system. and unsteady formulations, respectively. We compare
The other eigenvalues λ5 and λ6 are related to the aug- these predicted flutter speeds and corresponding fre-
mented state variable x̄; they are real and negative. The quencies with those obtained experimentally in Ta-
solution of the linear part is asymptotically stable if all ble 2. Clearly, the predicted flutter speed based on the
of the real parts of the λi are negative. On the other unsteady formulation is very close to the experimen-
hand, if one or more of the real parts are positive, the tally obtained values 10.89 m/s ≤ Uf ≤ 10.92 m/s,
linearized system is unstable. The speed for which one whereas the predicted flutter speed based on the quasi-
or more eigenvalues have zero real parts corresponds steady approximation is approximately 30 % higher
to the onset of the linear instability and is termed the than the experimental values. Furthermore, the flut-
flutter speed, Uf . As noted above, the relative con- ter frequency predicted with the unsteady formulation
tribution of the unsteady effects to the aerodynamic is much closer to that obtained experimentally than
loads and, thus, to the system’s stability depends on the one predicted with the quasi-steady formulation.
the wing’s geometry. Therefore, it is concluded that the quasi-steady ap-
proximation fails to accurately predict the experimen-
tally obtained flutter speed and frequency.
5 Linear analysis: flutter speed and frequency
The eigenvalues of the linearized equations of the 6 Nonlinear analysis
quasi-steady and unsteady formulations were calcu-
lated using the parameter values in Table 1. Variations 6.1 Identification of the nonlinear torsional spring
of these eigenvalues with the freestream velocity are coefficient
shown in Fig. 3. It is clear that two complex-conjugate To identify the coefficients of the quadratic and cubic
eigenvalues cross the imaginary axis as the freestream terms kα1 and kα2 in the pitch spring, we use the un-
166 A. Abdelkefi et al.

Fig. 4 Variation of the effective damping, which changes sign, with the freestream velocity: (a) quasi-steady approximation and
(b) unsteady representation of the aerodynamic loads

Table 2 Comparison
between the theoretically Quasi-steady approximation Unsteady formulation Experimental results
and experimentally
obtained flutter speed and Flutter frequency (Hz) 2.78 2.59 2.63
frequency Uf (m/s) 14.35 10.902 10.89 < Uf < 10.92

Fig. 5 Variations of the (a) plunge and (b) pitch RMS amplitudes using the unsteady formulation when U/Uf = 1.25

steady formulation to model the aerodynamic loads. of the plunge and pitch RMS amplitudes. The result is
The curves in Figs. 5 and 6 show variations of the kα1 = 3.95 Nm and kα2 = 107 Nm.
plunge and pitch RMS amplitudes with kα1 and kα2
for U/Uf = 1.25 and U/Uf = 1.4, respectively. In 6.2 Validation of aerodynamic model and identified
both figures, we note that increasing the coefficient coefficients
of the quadratic nonlinearity results in an increase in
The curves in Figs. 7 and 8 show a comparison of the
the RMS amplitudes of the plunge and pitch. On the time histories of the pitch and plunge obtained from
other hand, increasing the coefficient of the cubic non- the experiments and numerical integration of the gov-
linearity results in a decrease in the RMS amplitudes erning equations using the unsteady formulation to
of both of the plunge and pitch. Inspecting these fig- model the aerodynamic loads for U/Uf = 1.25 and
ures, we estimate the values of the nonlinear coeffi- U/Uf = 1.4, respectively. The plunge seems to be
cients that yield the experimentally measured values harmonic for both freestream velocities. On the other
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 167

Fig. 6 Variations of the (a) plunge and (b) pitch RMS amplitudes using the unsteady formulation when U/Uf = 1.4

Fig. 7 Experimental and numerical time histories of the (a) plunge and (b) pitch obtained using the unsteady formulation when
U/Uf = 1.25

Fig. 8 Experimental and numerical time histories of the (a) plunge and (b) pitch obtained using the unsteady formulation when
U/Uf = 1.4

hand, there is more than one frequency in the pitch. are in good agreement with the experimental measure-
Furthermore, the oscillations in both motions have a ments.
nonzero mean value, indicating nonzero static posi- Comparisons of the time histories of the experi-
tions. Using the values presented above shows that mentally measured pitch and plunge with those ob-
the oscillations predicted with the unsteady approach tained by numerical integration of the governing equa-
168 A. Abdelkefi et al.

Fig. 9 Experimental and numerical time histories of the (a) plunge and (b) pitch obtained using the quasi-steady approximation when
U/Uf = 1.25

Fig. 10 Experimental and numerical time histories of the (a) plunge and (b) pitch obtained using the quasi-steady approximation
when U/Uf = 1.4

Fig. 11 Bifurcation diagrams of the (a) plunge and (b) pitch RMS amplitudes obtained experimentally (dots) and using numerical
integration (dashed lines) based on the unsteady formulation

tions using the quasi-steady approximation to model tude and frequency of the oscillations. This is expected
the aerodynamic loads are shown in Figs. 9 and 10 from the linear analysis because the flutter frequency
for U/Uf = 1.25 and U/Uf = 1.4, respectively. The obtained with the quasi-steady approximation is larger
results show a clear discrepancy in both of the ampli- than the experimental flutter frequency.
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 169

Fig. 12 Bifurcation diagrams of the (a) plunge and (b) pitch RMS amplitudes obtained experimentally (dots) and using numerical
integration (dashed lines) based on the quasi-steady approximation

Figures 11 and 12 show bifurcation diagrams for 6.3 Normal form of Hopf bifurcation: Supercritical
both of the plunge and pitch obtained using the un- instability
steady and quasi-steady representations of the aero-
dynamic loads, respectively. Figure 11 shows good To determine the type of instability, we derive the
agreement between numerical predictions based on the normal form [22–24] of the dynamics of the sys-
unsteady formulation and experimental data. On the tem near flutter. To this end, we add the perturba-
other hand, Fig. 12 shows a small discrepancy between tion term 2 σU Uf to the flutter speed and obtain
numerical predictions based on the quasi-steady ap- U = Uf + σU 2 Uf . The matrix B1 (U ) is then ex-
proximation and experimental data particularly in the pressed as the sum B1 (Uf ) + σU 2 B2 (Uf ) where
plunge amplitude, as shown in Fig. 12(a).

⎡ ⎤
0 0 0 0 0 0
⎢ ⎥
⎢ 0 e 1 Uf f2 Uf + 2f3 Uf2 e 2 Uf f4 Uf + 2f5 Uf2 + 3f6 Uf3 e3 Uf + 2e4 Uf2 ⎥
⎢ ⎥
⎢0 ⎥
⎢ 0 0 0 0 0 ⎥
B2 (Uf ) = ⎢
⎢ 0 e 5 Uf

⎢ f7 Uf + 2f8 Uf2 e 6 Uf f9 Uf + 2f10 Uf2 + 3f11 Uf3 e7 Uf + 2e8 Uf ⎥
2

⎢ ⎥
⎢0 0 0 0 0 0 ⎥
⎣ ⎦
0 e 9 Uf f12 Uf + 2f13 Uf2 e10 Uf f14 Uf + 2f15 Uf2 + 3f16 Uf3 e11 Uf + 2e12 Uf2

and the fi and ei are, respectively, coefficients asso- and letting P be the matrix whose columns are the
ciated with K1∗ and C1∗ based on like powers of Uf . eigenvectors of the matrix B1 (Uf ) corresponding to
Rewriting (37) as the eigenvalues ±j ω1 −μ1 , ±j ω2 , −μ3 , and −μ4 , we
define a new vector Y such that Z = P Y and rewrite
(39) as
Ż = B1 (Uf )Z + 2
σU B2 (Uf )Z P Ẏ = B1 (Uf )P Y + 2
σU B2 (Uf )P Y
+ N1q (Z, Z) + N1c (Z, Z, Z) (39) + N1q (P Y, P Y ) + N1c (P Y, P Y, P Y ) (40)
170 A. Abdelkefi et al.

Multiplying (40) from the left with the inverse P −1 Y6 = Y61 (T0 , T2 ) + 2
Y62 (T0 , T2 ) + 3
Y63 (T0 , T2 )
of P , we obtain  
+O 4 (49)
−1
Ẏ = J Y + 2
σU GY + P N1q (P Y, P Y )
where Tn = n t. In terms of the Tn , the time derivative
−1
+P N1c (P Y, P Y, P Y ) (41) is expressed as

where J = P −1 B1 (Uf )P is a diagonal matrix whose d ∂ ∂


= + 2
+ · · · = D0 + 2
D2 + · · · (50)
elements are the eigenvalues ±j ω1 − μ1 , ±j ω2 , −μ3 , dt ∂T0 ∂T2
and −μ4 of B1 (Uf ) and G = P −1 B2 (Uf )P . We note
that Y2 = Y1 and Y4 = Y3 and hence we rewrite (41) in Substituting (46)–(50) into (42)–(45), using the afore-
component form as mentioned scalings, and equating coefficients of like
powers of , we obtain four equations for the Ymn . We

6 note that Y31 is the only non-decaying of the first-order
Ẏ1 = j ω1 Y1 − μ1 Y1 + 2
σU G1i Yi + Q̂1 (Y, Y ) terms. Therefore, the equations at different orders re-
1 duce to
+ Ĉ1 (Y, Y, Y ) (42) Order ( )


6
D0 Y31 − j ω2 Y31 = 0 (51)
Ẏ3 = j ω2 Y3 + 2
σU G3i Yi + Q̂3 (Y, Y )
1 Order ( 2 )
+ Ĉ3 (Y, Y, Y ) (43) 2
D0 Y12 = j ω1 Y12 + q11 Y31
2
+ q12 Y31 Y 31 + q13 Y 31

6
(52)
Ẏ5 = −μ3 Y5 + 2
σU G5i Yi + Q̂5 (Y, Y )
1 2
D0 Y32 = j ω2 Y32 + q31 Y31
2
+ q32 Y31 Y 31 + q33 Y 31
+ Ĉ5 (Y, Y, Y ) (44) (53)

6
D0 Y52 = −μ3 Y52 + q51 Y31
2
+ q52 Y31 Y 31 + q53 Y 31
2
Ẏ6 = −μ4 Y6 + 2
σU G6i Yi + Q̂6 (Y, Y )
(54)
1
2
+ Ĉ6 (Y, Y, Y ) (45) D0 Y72 = −μ4 Y62 + q61 Y31
2
+ q62 Y31 Y 31 + q63 Y 31
(55)
where the Q̂i (Y, Y ) and Ĉi (Y, Y, Y ) are, respectively,
bilinear and tri-linear functions of the components Order ( 3 )
of Y .
The damping coefficient μ1 is expressed as 2 μ1 . D0 Y33 − j ω2 Y33
On the other hand, μ3 and μ4 are scaled to be O( ).
= −D2 Y31 + σU G33 Y31 + c3 Y31
2
Y 31 + 2q31 Y31 Y32
To compute the normal form of the Hopf bifurcation
of (42)–(45) near U = Uf , we follow Nayfeh [22–24] + q32 Y31 Y 32 + q32 Y 31 Y32 + 2q33 Y 31 Y 32
and seek a third-order approximate solution in the
form + q14 Y31 Y12 + q15 Y31 Y 12 + q16 Y 31 Y12
+ q17 Y 31 Y 12 + q18 Y31 Y52 + q19 Y 31 Y52
Y1 = Y11 (T0 , T2 ) + 2
Y12 (T0 , T2 ) + 3
Y13 (T0 , T2 )
  + q20 Y31 Y62 + q21 Y 31 Y62 + cc + NST (56)
+O 4 (46)
Y3 = Y31 (T0 , T2 ) + 2
Y32 (T0 , T2 ) + 3
Y33 (T0 , T2 ) where NST stands for terms that do not produce sec-
  ular terms and cc stands for the complex conjugate of
+O 4 (47) the preceding terms.
Y5 = Y51 (T0 , T2 ) + 2
Y52 (T0 , T2 ) + 3
Y53 (T0 , T2 ) The solution of (51) is expressed as
 
+O 4 (48) Y31 = A(T2 )ej ω2 T0 (57)
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 171

Substituting (57) into (52)–(55) and solving the result- where a is the amplitude and γ is the phase shift of the
ing equations, we obtain periodic solution. Equation (64) has the three equilib-
rium solutions
j q11 j q12 
Y12 = A2 e2j ω2 T0 + AA
ω1 − 2ω2 ω1 −4βr
a = 0 and a = ±
j q13 2 Ner
+ A e−2j ω2 T0 (58)
ω1 + 2ω2
The solution a = 0 corresponds to the trivial solu-
−j q31 2 2j ω2 T0 j q32 tion. The other two solutions are nontrivial. The triv-
Y32 = A e + AA
ω2 ω2 ial solution is asymptotically stable when βr < 0 or
j q33 2 −2j ω2 T0 βr = 0 and Ner < 0; and unstable when βr > 0 or
+ A e (59) βr = 0 and Ner > 0. The nontrivial solutions exist
3ω2
q51 q52 when βr Ner < 0; they are stable (supercritical Hopf
Y52 = A2 e2j ω2 T0 + AA bifurcation) when βr > 0 and Ner < 0 and unsta-
2j ω2 + μ3 μ3
ble (subcritical Hopf bifurcation) when βr < 0 and
q53 2
+ A e−2j ω2 T0 (60) Ner > 0.
−2j ω2 + μ3 For the parameters presented in Table 1, the real
q61 q62 parts of β and Ne are
Y62 = A2 e2j ω2 T0 + AA
2j ω2 + μ4 μ4
q63 2
βr = 0.254σU (66)
+ A e−2j ω2 T0 (61)
−2j ω2 + μ4 Ner = 0.029kα1
2
− 0.00774kα2 (67)
Substituting (57)–(61) into (56) and eliminating the As expected, the type of Hopf bifurcation depends on
terms that lead to secular terms, we obtain the complex- the nonlinear coefficients kα1 and kα2 . Using the non-
valued normal form linear normal form, we express the dynamic ampli-
tudes of the LCO for the plunge (Ah ) and pitch (Aα )
D2 A = βA + Ne A2 A (62)
as

where β = σU G33 and
Ah = a P [1, 3]2r + P [1, 3]2i (68)
j q32 q31 j q32 q 32 2j q33 q 33 
Ne = c3 + − − Aα = a P [3, 3]2r + P [3, 3]2i (69)
ω2 ω2 3ω2

+
j q14 q12 j q15 q 12
− +
j q16 q11 where [.]r and [.]i denote the real and imaginary parts,
ω1 ω1 ω1 − 2ω2 respectively, and P [j, k] denotes the component (j ,k)
j q17 q 13 q18 q52 q19 q51 of the matrix P . For the considered system parameters,
− + + the plunge and pitch dynamic amplitudes are given by
ω1 + 2ω2 μ3 2j ω2 + μ3
q20 q62 q21 q61 Ah = 0.01109a and Aα = 0.0598a (70)
+ + (63)
μ4 2j ω2 + μ4
It follows from (58)–(61) that, beyond flutter, the
The effects of the quadratic and cubic nonlinearities quadratic nonlinearity produces a static deflection in
kα1 and kα2 on the system are expressed through Ne . both of the pitch and plunge, as observed in our exper-
Expressing A in the polar form A(T2 ) = 12 aeiγ (T2 ) iments. The predicted static deflections are given by
and separating the real and imaginary parts in (62), we 
obtain the following real-valued normal form of the j q12 jq j q32
hs = P [1, 1] − P [1, 2] 12 + P [1, 3]
Hopf bifurcation: 4ω1 4ω1 4ω2
j q 32 q52
1
a  = βr a + Ner a 3 (64) − P [1, 4] + P [1, 5]
4 4ω2 4μ5

1 q62
γ  = βi + Nei a 2 (65) + P [1, 6] a2 (71)
4 4μ6
172 A. Abdelkefi et al.

Fig. 13 Bifurcation diagrams of the (a) plunge and (b) pitch RMS amplitudes obtained using the analytical solution (solid lines),
using numerical integration (dashed lines) based on the unsteady formulation, and experimentally (dots)


j q12 jq j q32 two-dimensional typical section. The databases for the
αs = P [3, 1] − P [3, 2] 12 + P [3, 3]
4ω1 4ω1 4ω2 identification are generated from experimental mea-
j q 32 q52 surements, periodic responses in the pitch containing
− P [3, 4] + P [3, 5] more than one frequency, obtained at speeds above the
4ω2 4μ5
 flutter speed. The concentrated nonlinearity is mod-
q62 eled by quadratic and cubic nonlinear terms in the
+ P [3, 6] a2 (72)
4μ6 pitch stiffness. Two different representations for the
For the considered system parameters, the plunge and aerodynamic loads are used: the quasi-steady repre-
pitch static amplitudes are given by sentation and an unsteady formulation. The results
show that, for this setup, the quasi-steady representa-
hs = 0.0000617a 2 kα1 and αs = −0.002267a 2 kα1 tion fails to correctly predict the flutter frequency and
speed. On the other hand, the unsteady representation,
(73)
which takes into consideration the wake effects, yields
In Fig. 13, we compare analytical results based on excellent agreement with the experimentally obtained
the normal form of the Hopf bifurcation with predic- flutter frequency and speed. Based on numerical inte-
tions obtained with integration of the governing equa- grations, we have identified the nonlinear coefficients
tions using the unsteady formulation to model the un- of the pitch spring. The representative model and iden-
steady aerodynamic loads and experimental data. We tified coefficients are validated by comparing results
note the agreement of the three plots near bifurcation. of numerical integration of the governing equations
On the other hand, at higher speeds, the normal form with the experimentally obtained responses. Further-
fails to predict the pitch and plunge LCOs. However, more, to characterize the type of instability, we deter-
there is good agreement between the experimental re- mine the normal form of the Hopf bifurcation. It shows
sults and the numerical integrations. Finally, we note that the Hopf bifurcation is supercritical, which is val-
that the normal form is that of a supercritical Hopf bi- idated experimentally.
furcation because the real part of the effective nonlin-
earity is negative, which is validated experimentally.
References

1. Dowell, E.H., Tang, D.: Nonlinear aeroelasticity and un-


7 Conclusions
steady aerodynamics. AIAA J. 40, 1697–1707 (2002)
2. Gilliat, H.C., Strganac, T.W., Kurdila, A.J.: An investiga-
We have performed a nonlinear analysis to identify tion of internal resonance in aeroelastic systems. Nonlinear
the concentrated nonlinearity in the pitch spring of a Dyn. 31, 1–22 (2003)
An analytical and experimental investigation into limit-cycle oscillations of an aeroelastic system 173

3. Raghothama, A., Narayanan, S.: Non-linear dynamics of a 14. Strganac, T.W., Ko, J., Thompson, D.E., Kurdila, A.J.:
two-dimensional air foil by incremental harmonic balance Identification and control of limit cycle oscillations
method. J. Sound Vib. 226, 493–517 (1999) in aeroelastic systems. In: Proceedings of the 40th
4. Liu, L., Wong, Y.S., Lee, B.H.K.: Application of the centre AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dy-
manifold theory in non-linear aeroelasticity. J. Sound Vib. namics, and Materials Conference and Exhibit (1999).
234, 641–659 (2000) AIAA Paper No. 99-1463
5. Abdelkefi, A., Vasconcellos, R., Marques, F., Hajj, M.R.:
15. Ko, J., Strganac, T.W., Kurdila, A.J.: Adaptive feedback
Bifurcation analysis of an aeroelastic system with concen-
linearization for the control of a typical wing section with
trated nonlinearities. Nonlinear Dyn. 69, 57–70 (2012)
6. Abdelkefi, A., Vasconcellos, R., Marques, F., Hajj, M.R.: structural nonlinearity. Nonlinear Dyn. 18, 289–301 (1999)
Modeling and identification of freeplay nonlinearity. 16. Gilliatt, H.C., Straganac, T.W., Kurdilla, A.J.: An investiga-
J. Sound Vib. 331, 1898–1907 (2012) tion of internal resonance in aeroelastic systems. Nonlinear
7. Vasconcellos, R., Abdelkefi, A., Marques, F., Hajj, M.R.: Dyn. 31, 1–22 (2003)
Representation and analysis of control surface freeplay 17. Fung, Y.C.: An Introduction to the Theory of Aeroelastic-
nonlinearity. J. Fluids Struct. 31, 79–91 (2012) ity. Dover, New York (1993)
8. Erturk, A., Vieira, W.G.R., De Marqui, C., Inman, D.J.: On 18. Dowell, E.H.: A Modern Course in Aeroelasticity. Kluwer,
the energy harvesting potential of piezoaeroelastic systems. Dordrecht (1995)
Appl. Phys. Lett. 96, 184103 (2010)
9. Abdelkefi, A., Nayfeh, A.H., Hajj, M.R.: Modeling and 19. Theodorsen, T.: General theory of aerodynamic instability
analysis of piezoaeroelastic energy harvesters. Nonlinear and the mechanism of flutter. Technical Report NACA 496
Dyn. 67, 925–939 (2012) (1935)
10. Abdelkefi, A., Nayfeh, A.H., Hajj, M.R.: Design of 20. Bisplinghoff, R.L., Ashley, H.A., Halfman, R.L.: Aeroelas-
piezoaeroelastic energy harvesters. Nonlinear Dyn. 68, ticity. Dover, New York (1996)
519–530 (2012) 21. Edwards, J.W., Ashley, A.H., Breakwell, J.V.: Unsteady
11. Abdelkefi, A., Nayfeh, A.H., Hajj, M.R.: Enhancement of aerodynamic modeling for arbitrary motions. AIAA J. 17,
power harvesting from piezoaeroelastic systems. Nonlinear 365–374 (1979)
Dyn. 68, 531–541 (2012) 22. Nayfeh, A.H.: Perturbation Methods. Wiley, New York
12. Souza, V.C., Anicezio, M., De Marqui, C., Erturk, A.: En-
(1973)
hanced aeroelastic energy harvesting by exploiting com-
bined nonlinearities: theory and experiment. J. Smart 23. Nayfeh, A.H.: Introduction to Perturbation Techniques.
Mater. Struct. 20, 094007 (2011) Wiley-Interscience, New York (1981)
13. Abdelkefi, A., Hajj, M.R., Nayfeh, A.H.: Sensitivity anal- 24. Nayfeh, A.H., Balachandran, B.: In: Applied nonlinear dy-
ysis of piezoaeroelastic energy harvesters. J. Intell. Mater. namics. Wiley Series in Nonlinear Science. Wiley, New
Syst. Struct. (2012). doi:10.1177/1045389X12440752 York (1994)

You might also like