Aerospace Science and Technology 147 (2024) 109002

Contents lists available at ScienceDirect

Aerospace Science and Technology

journal homepage: www.elsevier.com/locate/aescte

Mechanism of single-mode panel flutter in low supersonic flow

Liu-qing Ye a, b, Kun Ye b, *, Ke Jin a, Yi-fan Zhang b, Zheng-yin Ye b
a
School of Aerospace Science and Technology, Xidian University, Xi’an, Shaanxi 710126, China
b
School of Aeronautics, Northwestern Polytechnical University, Xi’an, Shaanxi 710072, China

A R T I C L E I N F O A B S T R A C T

Communicated by Dr Qiulin Qu Supersonic panel flutter can be of two possible types: single-mode flutter and coupled-mode flutter. Compared to
coupled-mode flutter, the physical mechanism of single-mode flutter is less studied at present. In order to reveal
Keywords: the inducing mechanism of single-mode flutter, the present paper constructed a reduced-order fluid model by
Aeroelasticity using the system identification and the Auto Regressive with eXogenous input (ARX) model. Coupling the
Panel flutter
reduced-order model (ROM) for unsteady aerodynamics in low supersonic flow with the structural equation, a
Single mode flutter
highly efficient ROM-based aeroelastic model in state space is formulated, then the flutter boundaries are ob­
Computational fluid dynamics
Fluid-structure interaction tained and the stability of aeroelastic modes are investigated by the complex eigenvalue analysis. According to
the physical meaning of relevant parameters in unsteady aerodynamic reduced order model ARX, it is found that
the unsteady characteristic of the flow, specifically the history effects of the unsteady aerodynamic forces, plays a
dominant role in causing the single-mode panel flutter at low supersonic speeds. It is also shown that the higher
modes are indeed weakly unstable in the absence of any structural damping or a viscous boundary layer.

in an enormous number of publications by means of piston theory due to

the fact that the piston theory has the advantage of mathematical
1. Introduction simplicity and relatively good accuracy at high Mach numbers.
Compared to coupled-mode flutter, investigation related to single-mode
Supersonic panel flutter is an aeroelastic instability of a thin plate or flutter is less common, and this type of flutter cannot be investigated by
shell-like structural components of flight vehicles moving at supersonic using the piston theory and requires potential flow theory or more
speed [1–7], which is often encountered in the operation of aircraft and complex aerodynamic theories.
missiles. This dynamic instability phenomenon is a self-excited oscilla­ Single-mode panel flutter was the first observed by Nelson and
tion caused by the interaction of the aerodynamic, elastic and inertia Cunnigham by using the potential flow theory [15]. Some numerical
loads [8–11] and this vibration occurs due to energy transfer from the observations on the single-mode flutter at low supersonic flow are
gas flow to the flexible panel. Unlike wing flutter leading to the sudden summarized by Dowell [16] adopting the same aerodynamic model. In
destruction of the structure, panel flutter usually causes fatigue damage fact, for a long time there was a lack of researches on this type of
of panel structure and decreases the panel’s lifetime. The earliest re­ instability after it was uncovered at low supersonic speeds, and there has
ported structure damage that can be attributed to panel flutter were the been some speculation that this type of flutter may not occur in physical
failure of the German V-2 rockets during World War II. Extensive studies reality and on actual structures. However, the classical experimental
on the panel flutter problem have been conducted Dowell [12], Mei et al. study by Muhlstein and Gaspers [17,18] connected with computational
[13] and McNamara and Friedmann [14] have given the excellent re­ studies by Dowell [16] resolved this issue. Dowell also studied the limit
views of these studies. For supersonic flutter, there are two possible cycle oscillations of single mode and coupled mode flutter. Starting from
qualitatively different instability types, i.e. coupled- or single-mode 2000, Vedeneev [19–27,29] has conducted studies on single-mode
flutter. The first one develops from the interaction of two oscillatory flutter in low supersonic flows. Based on an asymptotic method, Vede­
modes via an aerodynamic coupling, which occurs at high Mach neev [19–21] studied the single-mode flutter at low supersonic flows
numbers. The single-mode flutter arises as a result of negative aero­ theoretically. Applying this method to the case of one-dimensional
dynamic damping on a single modal degree of freedom with a flutter system, Vedeneev [19] investigated the stability of a wide strip plate
frequency in the vicinity of the corresponding eigenfrequency, which in a supersonic inviscid gas flow. The classic low-frequency flutter
occurs at low supersonic speeds. Coupled-mode flutter has been studied

* Corresponding author.
E-mail address: [email protected] (K. Ye).

https://doi.org/10.1016/j.ast.2024.109002
Received 30 September 2023; Received in revised form 18 February 2024; Accepted 18 February 2024
Available online 22 February 2024
1270-9638/© 2024 Elsevier Masson SAS. All rights reserved.
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Nomenclature W nondimensional plate deflection

w plate deflection
a speed of sound x streamwise coordinate
a∞ freestream sound velocity z normal coordinate
D plate stiffness γ air specific heat ratio
E elasticity modulus λ nondimensional dynamic pressure, 2ql3 /MaD
h plate thickness υ Poisson’s ratio
l Plate length ρ air density
Ma Mach number ρs density of plate
na orders of the ROM ρ∞ , U∞ , p∞ , Ma∞ freestream air density, velocity, pressure, Mach
nb orders of the ROM number
N mode number (• ) derivative with respect to τ
N0 aerodynamic pressure (− ) dimensional quantity
p(x, t) aerodynamic pressure
Δp pressure difference across the panel surface Subscripts
q∞ dynamic pressure, ρ∞ U∞2
/2 a aerodynamic
Qj generalized force cr critical
i, j modal indices
qj generalized displacement
s structural
R0 nondimensional in-plane force
t physical time

developing from the interaction of two oscillation modes and single Gordnier and Visbal [30] investigated the flutter characteristics of the
mode flutter which is a consequence of negative aerodynamic damping flexible panel at subsonic and low supersonic speeds. He found some
were detected. Vedeneev [20] extended the work to two-dimensional complex flutter phenomena, which are seldom uncovered with the
system, and then observed the single mode flutter of rectangular plate frame of potential flow theory. In addition, the simulation results based
at supersonic speeds. He obtained the critical condition which can on CFD technology showed that the potential flow theory may not be
determine whether the plate oscillations are damped or amplified. accurate enough to capture all the system details specially at transonic
Vedeneed [21] further estimated the width of the plates to which the and low supersonic speeds if boundary layer effects or shocks are pre­
results in Ref. [19] for single mode flutter are applicable. The above sent. Based on a coupled computational fluid dynamics and computa­
studies conducted by Vedeneev all focused on the linear stability, and in tional structure dynamics (CFD/CSD) method, Shishaeva and Vedeneev
Ref. [22] he investigated nonlinear limit cycle oscillation amplitudes. [31] numerically investigated the nonlinear oscillations of single- and
Then he found that compared with the case of coupled mode flutter, the coupled- mode flutter. The results showed that the oscillation amplitude
amplitude growth rates for single mode flutter are much stronger. This of single-mode flutter is typically several times higher than that for the
characteristic also has been observed by Dowell [8,9]. Using the exact case of coupled-mode flutter. It was pointed out that compared with the
aerodynamic theory and piston theory, Vedeneev [23] numerically classic coupled-mode flutter, the single-mode flutter in low supersonic
investigated the dynamic behaviors of the elastic panel in supersonic flow can be much more dangerous. Considering the turbulent
flows. It was shown that a single-mode flutter developing in the region of boundary-layer effects, Alder [32,33] used the unsteady
low supersonic Mach numbers, which cannot be revealed by the piston Reynolds-averaged Navier–Stokes (URANS) equations in combination
theory, can be detected by the exact aerodynamic theory. The instability with the von-Kármán large deflection theory to conduct transonic and
boundaries of the coupled-mode flutter, which occurs at high Mach low supersonic panel flutter analysis. The results showed that a turbu­
numbers, calculated by the exact aerodynamic theory agree well with lent boundary layer affects the stability boundaries of flat panel exposed
those obtained by the piston theory. Based on the potential flow theory, to subsonic and low supersonic flow in a stabilizing manner. However,
Vedeneev [24] conducted a comprehensive numerical analysis for even though the CFD technology has an obvious advantage in studying
single-mode flutter in low supersonic flow. It was shown that gas density complex phenomenon in single-mode flutter problems, this numerical
has no effect on the occurrence of single-mode flutter, which can method is very computationally intensive and does not allow an
significantly affect the phenomenon of coupled-mode flutter. Vedeneev extensive exploration of the effect of various parameters to the extent.
[25] studied nonlinear development of multi-mode limit cycles caused Moreover, for the CFD numerical method, results for the system ob­
by a single-mode flutter. He also obtained the relations between the tained is in the time domain, and it is very costly to do parametric
amplitudes of the limit cycles and the parameters of the problem, which studies and develop physical insight.
enable one to estimate the risk of the onset of flutter. Dowell [16] and Recently, a CFD-based unsteady aerodynamic reduced-order model
Vedeneev and Bondarev [26,27] studied the influence of boundary layer [34,35], which is a highly efficient tool to calculate aerodynamic forces,
on single-mode flutter. Shishaeva [28] studied the stability of limit cy­ has been widely used for aeroelastic analysis. Dowell [1], Lucia et al.
cles with respect to unsteady flow conditions at transonic and low su­ [36], and Ghoreyshi et al. [37] have presented some overviews of
personic flow speed, and he found that the most dangerous oscillation reduced-order models and their applications on aeroelastic research.
regimes, high-frequency periodic or nonperiodic oscillations, are sup­ Many researchers have successfully revealed the mechanisms of aero­
pressed if the flow acceleration is sufficiently fast. Recently, an experi­ elastic problems by employing the reduced-order model. Using an
ment conducted by Vedeneev [29] has confirmed that single-mode identification method to construct the reduced-order models of unsteady
flutter can occur in real panel structures. aerodynamics for the incompressible flow past a vibrating cylinder,
With the development of computer capacity and computational fluid Zhang et al. [38] studied the physical mechanisms underlying frequency
dynamics (CFD) techniques, many researchers begin to study the panel lock- in vortex-induced vibrations at low Reynolds numbers. The results
flutter problems at transonic and low supersonic Mach numbers by using show that according to different inducing mechanisms, the frequency
aeroelastic solvers based on full Euler or Reynolds-averaged Navier–­ lock-in phenomenon could be divided into two patterns,
Stokes equations. By solving Euler and Navier–Stokes equations, resonance-induced lock-in and flutter-induced lock-in. It provides a

2
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

comprehensive understanding of mechanisms of the frequency lock- in represents net aerodynamic loading acting on the panel surface. N0 is
vortex-induced vibrations. Gao et al. [39] constructed a linear externally applied in-plane load. The definition of Nx can be seen in Refs.
reduced-order model based on the system identification method to [6,7].
investigate the physical mechanism of frequency lock-in phenomenon in Introducing the following nondimensional parameters
transonic buffeting flow. He found that the frequency lock-in transonic √̅̅̅̅̅̅̅̅̅
buffeting flow is not caused by the resonance, but the linear x
ξ = , τs = t
D w
,W = ,λ =
2q∞ l3 ρ l
,μ = ∞
coupled-mode flutter. Quite recently, by coupling the CFD-based l ρhl4 h D ρh
(2)
reduced-order model with the structural motion equation, Li et al. 2 2 4
N0 l Nx l qa l
[40] studied the physical mechanisms of galloping of a square cylinder R0 = , Rx = , qa =
D D Dh
at low Reynolds numbers by using the ROM-based aeroelastic model. It
was found that the instability of the structure mode is the primary cause Eh 3 ∫ 1 (∂W)2
where, Nx is the dimensionless form of Nx ,Nx = 2l2 (1− dξ,Rx =
of the galloping phenomenon because the vibration frequency of the μ2 ) 0 ∂ξ
cylinder is identical to the eigenfrequency of structure mode during ∫ 1 (∂W)2
6 0 ∂ξ dξ
galloping.
In this paper, in order to investigate the physical mechanism of Eq. (1) becomes the dimensionless form
single-mode panel flutter in low supersonic flow, an unsteady aero­ ∂2 W ∂4 W ∂2 W
dynamic reduced-order model based on the system identification + 4 − (Rx − R0 ) 2 = qa (3)
∂τ2s ∂ξ ∂ξ
method and Auto Regressive with eXogenous input is also constructed.
Using the ROM-based aerodynamic model in combination with von- Galerkin’s method is used to solve Eq. (3). Considering the simply
Kármán large deflection theory, a highly efficient aeroelastic model in supported boundary condition, the deflection can be expressed by
state space is formulated, and then the aeroelastic instability boundaries Fourier series expansion
of the panel at low supersonic speeds are calculated by the complex ∑
N
eigenvalue analysis. According to the physical meaning of relevant pa­ W(ξ, τs ) = qi (τs )sin(iπξ) (4)
rameters in unsteady aerodynamic reduced order model ARX and the i=1

model stability analysis, the physical mechanism of single-mode panel Substituting Eq. (4) into Eq. (3), multiplying each term of Eq. (3) by
flutter at low supersonic speeds is studied in detailed. In addition, the sin(jπξ)(j = 1, 2, ⋯, N)and integrating it over the panel length, then Eq.
effects of mass ratio on single-mode flutter characteristics in low su­ (3) can be finally dispersed into a set of ordinary differential equations
personic flow are studied by means of direct coupled CFD/CSD simu­
lation method and ROM-based aeroelastic model. d2 qj (τs ) [ ] ∑N
+ (jπ)4 − R0 (jπ )2 qj (τs ) + 3(jπ)2 qj (τs ) (iπ )2 q2i (τs ) =
d τs2
(5)
i=1
2. Numerical simulation method ∫ 1
λl
− Cp (ξ)sin(jπξ)dξ, j = 1, 2, ⋯, N
2.1. Structural dynamics equations h 0

The right term in Eq. (5) is the generalized aerodynamic force and
Consider a two-dimensional isotropic flat panel of length l and can be denoted as fa,j . The Cp is the aerodynamic pressure coefficient,
thickness h, as shown in Fig. 1. Both edges of this panel are simply which is computed by the CFD solver at each time step.
supported. The upper surface of the panel is exposed to a supersonic
Introducing a set of state variables e = [e1 , ⋯, eN , eN+1 , ⋯, e2N ]T =
airstream flowing along the x direction with air density ρ∞ , flow velocity [ ]T
U∞ , Mach number Ma∞ , and aerodynamic pressure p∞ . It is assumed q1 , ⋯, qN , ∂∂τq1s , ⋯, ∂∂τqNs . Eq. (5) can be rewritten as the state space
that the pressure on the bottom side of the panel remains at the free­ equation as shown
stream value p∞ . The deflection of the panel undergoing cylindrical ⎧
bending is caused by the interaction of elastic, inertia and aerodynamic ⎪ dej
⎪
⎪ = eN+j
⎪
⎨ d τs
forces. Based on Hamilton’s principle, the equation of motion for the
(6)
elastic panel is derived as ⎪
⎪ dej+N [ 2]
∑N
λl
⎪ 4 2
(iπ )2 q2i − fa,j
⎩ dτ = − (jπ) − R0 (jπ) ej − 3(jπ) ej
⎪
h
∂2 w ∂4 w ∂2 w s i=1
ρh + D 4 − (Nx + N0 ) 2 = qa (1)
∂t2 ∂x ∂x Ulteriorly, the above aeroelastic Eq. (6) can be expressed in the
3 following from
where, w is the transverse displacement in the z direction. D = Eh
/12(1 − υ2 ) is the bending stiffness of the panel. ρ is the density of panel, de
= F(e(τs ), f(τs )) (7)
E is the Young’s modulus and υ is the Poisson’s ratio. p(x, t) is the d τs
pressure acting on the upper surface of the panel and qa = p∞ − p(x, t)
where, f(τs ) represents the aerodynamic terms, f = [fa,1 , fa,2 , ..., fa,n ]T .

Fig. 1. Gas flow over an elastic plate on one side. Fig. 2. Block diagram of the models and their relationship.

3
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

2.2. Computation and modeling of unsteady aerodynamics where the elastic plate is located at x = 0 ∼ 1. The total number of mesh
nodes on the wall are 274, with 131 nodes evenly distributed on the
Fig. 2 shows the models used in this paper and their relationship. The elastic panel surface. In present study, inviscid flow is simulated. The
structural model has been introduced in the above Section. For low su­ number of grid cells are 28,224 and the nondimensional height of first
personic flow, there are two models to calculate the unsteady general­ layer is 0.01.
ized aerodynamic loads. The first fluid model is CFD simulation with the
unsteady Euler equations, which is presented in the yellow block. The 2.2.2. Reduced order modeling for unsteady aerodynamics
other fluid model is the ROM with an external input by the system The model reduction for unsteady flow has been an active area of
identification method, which can be seen in the blue block. The building research in recent years [34]. Compared with the high-fidelity CFD
of the ROM is based on the CFD data. Compared with the CFD simula­ simulation, computing efficiency of the ROM method is greatly
tion, the ROM method has a higher efficiency and considerable accu­ increased because it grasps the dominant flow dynamics of the original
racy. Coupling these two different fluid models with the structural system with significant lower orders. Here, the system identification
model respectively, two aeroelastic models are obtained. When we method based on the ARX model is used to construct the reduced-order
connect the block of structural model with that of CFD simulation, the model for the unsteady aerodynamic forces in low supersonic flow. ARX
coupled CFD/CSD simulation strategy is obtained. When the block of model is an AutoRegressive model with eXogenous terms, and it is a
structural model is connected with that of the ROM method, the ROM- time-marching method that predicts the output of a system based on
based aeroelastic model can be built. These two aeroelastic models earlier observations and inputs, represented in the form of a recurrence
will be discussed in Section 3. relation. Owing to its simplicity and easy parameterization, the ARX
model has been widely used for aeroelastic analysis, such as flutter
2.2.1. CFD numerical method analysis at high attack angles, control low design of transonic flutter
In order to clarify the inducing mechanism of single-mode panel suppression, and transonic aeroservoelastic parametrical analysis. For
flutter and avoid the interference of other factors, the inviscid cases are the discrete-time multi-input/multi-output system, the ARX model can
considered. Fluid dynamics computations are performed by using an in- be given as
house solver
∑
na ∑1
nb−
Generalized Fluid-Structure Interaction (GFSI) [41,42]. GFSI flow y(k) = Ai y(k − i) + Bi u(k − i) + e(k) (9)
solver is a finite volume unstructured flow solver which solves the un­ i=1 i=0
steady Euler equations by adopting a cell-centered approach. The un­
steady Euler equations for a bounded control volume Ω with boundary where, y(k) is the vector of system output at the kth step, and u(k) is the
∂Ω are expressed in integral form as follow vector of system input at the kth step. e(k) is the error between the actual
∫ ∫ ∫ ∫ ∫ output and model-predicted output at the kth time step. In our study, the
( )
∂
QdV + Fc Q,Vgrid ⋅ndS = 0 (8) system outputs are the generalized aerodynamic force coefficients and
∂t
Ω ∂Ω
the output-vector is y = [f a,1 , f a,2 , ..., f a,n ]. The system inputs are the
generalized displacements of the structure, i.e. u = [q1 ,q2 ,⋯,qn ]. na and
where, Q = [ρ, ρu, ρv, ρw, e]T is the vector of conservative variables, nb are called the delay orders given by the user. Specifically, na is the
Fc (Q,Vgrid ) is the inviscid flux vector, Vgrid is the velocity of grid number of the previous outputs considered in the model, and nb is the
motion.n is the normal vector of the boundary face of a control volume. total number of the present and previous inputs included in the model.
The advection upstream splitting method (AUSM) is selected for The matrices Ai and Bi are the constant coefficients to be estimated
shock capturing and suppressing numerical oscillations. Dual-time employing the training data from the unsteady CFD solver. Multistep
stepping is employed for time discretization, and lower–upper sym­ input is employed owing to its ease of implementation and broad fre­
metric Gauss–Seidel (LU-SGS) method is used for pseudo-time iteration. quency content. The least squares (LS) method is employed to identify
The code has been parallelized with OpenMP in a globally shared the unknown model parameters. By changing the delay orders of ARX
memory model. Free-slip boundary condition is adopted on the surface model (i.e. na and nb), the model that minimizes the error between the
of walls and Riemann non-reflecting boundary condition is adopted on training data and ROM predictions will be find.
the far field. In this paper, the radial basis function (RBF) based mesh From Eq. (9), it can be found that the system response at given time
deformation method is used in the aeroelastic analysis. The detailed can be expressed as an algebraic series of multiplications and additions.
introductions to calculation methods can be seen in Ref. [41]. In addition, the system response at any time step y(k) is a linear com­
Fig. 3 shows the computational domain and boundary conditions for bination of past inputs u(k − i) and outputs y(k − i). Therefore, the ARX
flow equations. The far-field boundary conditions are set to the left, right model is easy to construct the ROM mathematically. Actually, the
and upper edges. The lower edge is set as the wall boundary condition, modeling process in present study mainly contains two steps: training
under the approximate signal and identification based on the ARX
model. Once the system inputs and outputs are given and the matrices Ai
and Bi are estimated, the construction of the ARX model is completed.
Then, this ARX model can be used in place of the CFD simulation in the
prediction of the unsteady aerodynamic forces in low supersonic flow.
In order to derive the state-space form of aeroelastic model, a state
vector xa (k) consisting of (na + nb − 1) vector state is introduced as
follows

xa (k) = [f a (k − 1), ..., f a (k − na), q(k − 1), ..., q(k − nb + 1)]T (10)
The state-space form for the discrete-time aerodynamic model can be
expressed as
{
xa (k + 1) = A ̃a xa (k) + B
̃a q(k)
(11)
̃ ̃
f a (k) = Ca xa (k) + Da q(k)

where,
Fig. 3. Computational domain and mesh.

4
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 4. Static deflection of panel under different non-dimensional dynamic pressures (a) at Ma=0.9, (b) at Ma=0.95.

⎡ ⎤
A1 A2 ⋯ Ana− 1 Ana B1 B2 ⋯ Bnb− 2 Bnb− 1 coupled with the flow equations, and Eq. (13) shows the detailed pro­
⎢ I
⎢ 0 ⋯ 0 0 0 0 ⋯ 0 0 ⎥
⎥ cedure. Compared to the standard Runge-Kutta method, the rules of
⎢ ⋮
⎢ I ⋯ 0 0 0 0 ⋯ 0 0 ⎥
⎥ small time step can be relaxed to a certain extent for the improved
⎢ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⎥ approximate Runge-Kutta scheme. The more detailed information about
⎢ ⎥
⎢ 0 ⎥
̃
Aa = ⎢
0 ⋯ I 0 0 0 ⋯ 0 0 ⎥ the improved approximate Runge-Kutta scheme can be seen in Ref. [44].
⎢ 0 0 ⋯ 0 0 0 0 ⋯ 0 0 ⎥
⎢
⎢ 0
⎥
⎥ In order to achieve high-order accuracy, the polynomial extrapolations
0 ⋯ 0 0 I 0 ⋯ 0 0
⎢
⎢ 0
⎥
⎥ of the general aerodynamic loads are utilized. The flow field needs to be
0 ⋯ 0 0 0 I ⋯ 0 0
⎢
⎣ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮
⎥
⎦ solved once in each time step. Once the en+1 shown in Eq. (6) is calcu­
0 0 ⋯ 0 0 0 0 ⋯ I 0 lated, the deformation of two-dimensional panel can be obtained from
Eq. (4). Employing the mesh deformation method, the computational
̃a = [ B0
B 0 ⋯ 0 I 0 0 ⋯ 0 ]T mesh is updated. And then the flow field at n + 1 time level can be solved
according to the updated mesh.
̃a = [A1 A2 ⋯Ana− 1 Ana B1 B2 ⋯Bnb− 2 Bnb− 1 ] ⎧
C ⎪ 1
⎪
⎪ f n+0.5 = (3f n− 2 − 10f n− 1 + 15f n )
⎪
⎪ 8
⎪
⎪
̃a = [B0 ]
D ⎪
⎪ f f
⎪
⎪ n+1 = n− 2 − 3f n− 1 + 3f n
⎪
⎪
⎪
In fact, the matrices A1 , A2 , …, Ana− 1 , Ana , B0 , B1 , B2 , …, Bnb− 2 , Bnb− 1
⎪
⎪
⎪ k 1 = F(e n fn)
,
⎪
in the above matrices ⎪
⎪
⎨ ( Δτ s )
̃a , B
A ̃a , C
̃a , D
̃a are the coefficient matrix Ai and Bi in Eq. (9). k2 = F en +
2
k1 , f n+0.5 (13)
⎪
⎪
In order to couple the structural equations, the continue-time form ⎪
⎪
⎪
( Δ τ s
)
⎪
⎪ k = F en + k2 , f n+0.5
for aerodynamic equation is needed. Using methods of bilinear trans­ ⎪
⎪ 3
⎪ 2
⎪
formation [39,40], the discrete-time form of Eq. (11) is converted into ⎪
⎪
⎪ k4 = F(en + Δτs k3 , f n+1 )
⎪
⎪
the continuous-time form, and the model in the state-space form is ⎪
⎪
⎪
constructed as follows ⎩ en+1 = en + Δτs (k1 + 2k2 + 2k3 + k4 )
⎪
6
{
ẋa = Aa xa + Ba q
(12) Actually, in the present study the coupled CFD/CSD simulation
f a = Ca xa + Da q strategy introduced above is mainly realized by our in-house code GFSI.
In fact, the matrices in discrete-time Eq. (11) are not the same as that Here, it should be noted that the nondimensional structural time step Δτs
in continuous-time Eq. (12), and the relationship between them can be must be matched with the nondimensional flow time step Δτa , because
seen in Ref. [43]. the structure and flow equations are nondimensionalized with different
parameters, respectively. According to the nondimensional parameters
3. Models for aeroelasticity in Eq. (2), the structure equations are nondimensionalized. The nondi­
mensional parameters which are used to nondimensionalize the flow
In this Section, the two different aeroelastic models shown in Fig. 2 equations can be seen in the appendix. The relationship between the
will be introduced in detail. The first aeroelastic model is the coupled nondimensional structural time step and the nondimensional flow time
CFD/CSD simulation method for which the unsteady aerodynamic step is shown as follows
forces in aeroelastic Eq. (7) are calculated by CFD simulation technique. √̅̅̅
μ
For the second aeroelastic model, an ARX-based ROM taking the place of Δτs = Δτa ⋅Ma (14)
λ
CFD simulation is used to predict the unsteady aerodynamic loads in
aeroelastic Eq. (7). where, the nondimensional parameters shown in Eq. (14) can be seen in
Eq. (2).
3.1. Coupled CFD/CSD simulation method According to the directly coupled CFD/CSD simulation strategy
introduced in this part, the aeroelastic stability of the two-dimensional
Firstly, we will introduce the coupled CFD/CSD simulation method. flexible panel in low supersonic flow can be analyzed in time domain.
The generalized aerodynamic forces in aeroelastic Eq. (7) are provided
by solving the unsteady Euler equations (8). Using an improved
approximate Runge-Kutta scheme, the structure equations are loosely

5
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

3.2. ROM-based aeroelastic model [ ][ ]

As + Bs Da Cs Bs Ca eε
ẋae = = Aae xae (21)
Ba C s Aa xa
Considering the ARX-based ROM constructed in the Section 2.2.2, an
ROM-based aeroelastic model on the base of the state space is estab­ The ROM-based aeroelastic model of the two-dimensional panel in
lished here, and then the aeroelastic stability of the elastic panel in low low supersonic flow is established. And then the problem of aeroelastic
supersonic flow can be studied in frequency domain. Assuming that at stability is converted into the analysis of complex eigenvalues of Aae . The
the static equilibrium position of the flexible panel, the jth modal process of the aeroelastic analysis is mainly as follows
displacement coefficient is e0,j and the jth modal aerodynamic coeffi­
cient is fa0,j . Lyapunov indirect method is utilized to evaluate the sta­ (1) Use the CFD solver to calculate the modal aerodynamic coeffi­
bility of this equilibrium of the nonlinear aeroelastic system. Lyapunov cient according to the designed input signals.
indirect method analyzes the stability of nonlinear systems by the (2) Use the identification technique to construct the input-output
following way. The nonlinear system is linearized at the selected equi­ difference model (Eq. (11)) in the discrete-time domain, and
librium positions, and then it indirectly analyzes the stability of the then turn it into the continuous-time state space form (Eq. (12)).
nonlinear system by analyzing the properties of solutions of the corre­ (3) Couple the aerodynamic state-space equations (Eq. (12)) and the
sponding linear system. structural state-space equations (Eq. (17)) to establish the ROM-
It is assumed that in the neighbourhood of the static equilibrium based aeroelastic state space Eq. (21).
position, the modal displace coefficient ej and the modal aerodynamic (4) Solve the complex eigenvalues of the state matrix Aae in Eq. (21)
coefficient fa,j consist of two components respectively: the static balance at different Mach numbers and mass ratios, finally the stability of
quantity e0,j , fa0,j and small dynamic disturbance eε,j , faε,j , then the aeroelastic system can be analyzed based on the root loci
{ analysis. The real part of the eigenvalue denotes the growth rate
ej = e0,j + eε,j
(15) of the eigenmode, and a positive one means that the aeroelastic
fa,j = fa0,j + faε,j
system is unstable. On the contrary, the aeroelastic system is
Substituting Eq. (15) into the aeroelastic Eq. (6) and using the Lya­ stable when the real parts of all the eigenvalues are negative. The
punov indirect method, it obtains the linear differential equation as imaginary part of the eigenvalue indicates the circular frequency
shown of the eigenmode.
⎧ deε,j
⎪
⎪
⎪ = eε,j+N 4. Results and discussion
⎪
⎪ d τs
⎪
⎪
⎪
⎪ [ ]
⎪
⎨ deε,j+N ( ) ∑N
( )2 This paper focuses on the physical mechanism of single-mode panel
= − (jπ)4 + R0 (jπ )2 + 3 1 − υ2 (jπ)2 e0,i (iπ)2 eε,j (16)
⎪ d τs i=1
flutter in low supersonic flow. This mechanism is revealed and studied
⎪
⎪
⎪
⎪ [ ] by the ROM-based aeroelastic model established in Section 3.2. The
⎪
⎪
⎪
⎪ ( ) ∑N
( ) λl directly coupled CFD/CSD time domain simulation is used as the veri­
⎪
⎩ − 3 1 − υ2 (jπ)2 2e0,i eε,i (iπ)2 e0,j − faε,j
i=1
h fication of ROM-based method and the supplement to conclusions from
ROM-based aeroelastic model. Dowell in Ref. [8] has pointed out that six
The above equation is further written in the state-space form as Galerkin’s modes are enough for supersonic flutter analysis. Thus, the
follows first six Galerkin’s modes are also adopted in the present study to ensure
{
ėε = As eε + Bs f aε the computational efficiency.
(17)
q = C s e ε + D s f aε
4.1. Validation of the time-domain simulation
where, eε = [eε,1 eε,2 ⋯eε,2N ]T .f ε = [fε,1 fε,2 ⋯fε,2N ]T
[ ] [ ]
0N×N IN×N − λl 0N×N The construction of ROM for unsteady aerodynamic loads is based on
As = ⋅Bs = ⋅Cs = [IN×N 0N×N ]⋅Ds = 0N×N (18) CFD data, and the results of ROM-based aeroelastic model is verified by
MN×N 0N×N h IN×N
direct CFD/CSD simulation method. Therefore, the FSI procedure and
where, the elements in matrix MN×N are given as CFD solver, which is realized by our in-house code GFSI, should be
⎧ ( ) verified before the ROM-based analysis. Owing to fact that the valida­
2 2
tion of FSI procedure involves the validation of CFD solver, the valida­
⎪ 2
⎪ − 6 1 − υ (jπ) e0,i (iπ) e0,j i ∕
⎪
⎪
=j
⎪ [ ]
⎪
⎨ ( ) ∑N tion of the present code is performed based on several flutter cases
εi,j = − (jπ)4 + R0 (jπ )2 + 3 1 − υ2 (jπ)2 e20,i (iπ)2 (19) suggested by Gordnier and Visbal [30]. The non-dimensional parame­
⎪
⎪
⎪
⎪
⎪
i=1
ters used in the computational cases are: μ = 0.1, h/l = 0.002.
⎩ − 6( 1 − υ2 )(jπ)2 e2 (jπ)2 i = j
⎪
At subsonic speed, the instability of the flexible panel usually be­
0,j
haves as static divergence. Fig. 4(a) and Fig. 4(b) show the static de­
This research is mainly focused on the aeroelastic stability of the flections of the panel under different non-dimensional dynamic
initially flat panel, while the initially flat panel corresponds to the static pressures at Ma=0.9 and at Ma=0.95, respectively. The static equilib­
equilibrium position. In this equilibrium condition, the static balance rium positions of the midpoint (ξ = 0.5) of the flexible panel are
quantities of the modal displacement coefficient and the modal aero­ observed. As shown in Fig. 4(a), the upper static equilibrium position
dynamic coefficient are both equal to zero, i.e. e0,j = 0,fa0,j = 0. Thus, and the lower static equilibrium of the panel under given λ are nearly
there is no static component for the aerodynamic loads in ARX-based symmetrical in the case of Ma=0.9. The results presented in Fig. 4(b)
ROM shown in Eq. (12). In other words, the modal aerodynamic coef­ show that in the case of Ma=0.95, there is obvious asymmetric phe­
ficient vector f a in Eq. (12) is equal to the vector f aε shown in Eq. (17). nomenon between the upper static equilibrium position and the lower
Coupling the structural state Eq. (17) and the aerodynamic state Eq. static equilibrium position considering large non-dimensional dynamic
(12), the state equation for the aeroelastic system is obtained as follows pressures. The presents results are in good accordance with the results of
[ ] [
ėε As + Bs Da Cs Bs Ca
][ ]
eε
[ ]
e Gordnier and Visbal [30]. Fig. 5 shows the supersonic flutter cases with
= = Aae s (20) the dynamic instability, and the amplitudes of limit cycle oscillation
ẋa Ba Cs Aa xa xa
(LCO) under given λ at different Mach numbers are investigated. The
Defining the state vector xae = [ee , xa ]T , Eq. (20) can be written as initial condition is q̇1 = 10, which means providing a vertical velocity

6
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

spectrum density (PSD) analysis of the chirp signal. The dominant fre­
quencies of the signal are from 1 to 60. The nondimensional natural
frequency of the panel for jth-order mode is ωj = (jπ )2 , and the range of
dimensionless frequency of first six modes is 1.57~56.55. Thus, it can be
seen in Fig. 6(b) that the designed signal meets the requirement of fre­
quency bandwidth.
We set up the aerodynamic model at Ma = 1.3, μ = 0.1 to verify the
correctness of the ROM introduced in Section 2.2.2. Applying the above
multistep input signals to the flexible panel, the modal aerodynamic
coefficients are calculated by running the CFD solver. By using the
resulting data as the training data, the parameters Ai and Bi of the ARX
model can be identified. In fact, the time cost of the ROM-based analysis
method is mainly focused on the model training. A variety of delay or­
ders (naandnb) have been tried, and finally the ARX model orders na = 7
and na = 6 are adopted as the best fit for the training data. Fig. 8 shows
the comparison of the time history of modal coefficients predicted by the
ROM and calculated by the CFD simulation. It can be seen in Fig. 8 that
the identified results match well with those of direct numerical simu­
lations. In order to conduct the quantitative analysis, a relative error is
Fig. 5. Amplitudes of LCO under different non-dimensional dynamic pressures. defined to evaluate the capacity of the ROM:
∑
L
disturbance to the first mode. The midpoint of the panel is regarded as |yCFD (i) − yiden (i)|
the observation point. With the increase of non-dimensional dynamic Error1 = i=1 (22)
∑
L
pressure, the amplitudes of LCO increase continuously. It can be seen in |yCFD (i)|
Fig. 5 that the present results agree well with Gordnier’s work [30]. All
i=1

test cases shown above have proved that the FSI procedure and CFD where, L is the length of the training signals. yiden represents the vector of
solver have a high numerical precision, and they are appropriate for the identified aerodynamic forces and yCFD represents the vector of aero­
subsequent investigations. dynamic forces calculated by CFD solver. Calculating the Eq. (22), the
errors of different modes are obtained, which are both less than 0.1 %.
4.2. Validation of the ROM Thus, it has been proved that the identification method based on the
ARX model has a high numerical accuracy.
Training signal design is the key in the construction of ROM for The ROMs are then validated in the time domain by comparing the
unsteady aerodynamic forces. An appropriate training signal must responses when the input is Gaussian White noise with those of direct
satisfy the following two requirements: (1) the frequency of the training numerical results. The input signals with Gaussian White noise are
signal should cover the frequencies of the structural modes that need to shown in Fig. 9, and the appropriate frequency bandwidth is considered
be excited; (2) the amplitude of the training signal should not be too shown in Fig. 10. Fig. 11 shows the comparison of the time history of the
large, otherwise it will induce the nonlinear aerodynamic responses and modal aerodynamic coefficients predicted by the ROM and calculated by
the linear modeling method cannot be used. In the present study, a chirp the CFD simulation. It is obviously seen that the ROM predictions
signal with an increasing frequency is designed as the input, as shown in exactly match those computed by the direct numerical simulations.
Fig. 6. Because there are six structural modes which need to be excited,
the training signals are multistep input signals shown in Fig. 7. One of
the basis mode excitations is presented in Fig. 6(a). For the sake of 4.3. Inducing mechanism of single-mode panel flutter in low supersonic
simplicity, the input signals to excite each structural modes are designed flow
as the same except for the amplitude. According to the amplitudes of
each mode response in the actual aeroelastic response, the amplitudes of Based on the ROM-based aeroelastic model constructed in Section
the input signals to excite six structural modes are designed as: W = 4.0, 3.2, the inducing mechanism of single-mode panel flutter in low su­
3.5, 2.0, 1.0, 0.5, and 0.25, respectively. Fig. 6(b) shows the power personic flow will be discussed in detail in this Section. From the non-

Fig. 6. One of the basis mode excitations.

7
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 7. The training signals.

dimensional Eqs. (14) and (21), it can be seen that the non-dimensional panel. In order to better understand the inducing mechanism of single-
control parameters for the aeroelastic system are: Ma, R0 , λ, h /l, and μ. mode flutter, the variation of aeroelastic modal stability at two
Here, the Mach numbers in the range of 1.1~2.8 are observed. Unless different regions for different Mach numbers will be studied in the
otherwise specified, the nondimensional structural parameters are: μ = following text.
0.1, h/l = 0.002, and the initial in-plane load is not considered, i.e. R0 =
0. For different Mach numbers, the reduced order models for unsteady 4.3.2. Model stability analysis
aerodynamic forces are different. Thus, one process of model training is For coupled-type flutter, the system has similar dynamic character­
needed for one Mach number. In the case of a given Mach number, the istics in the Mach region (1.7 ≤ Ma ≤ 2.8) and Ma=2.0 is selected for
critical flutter dynamic pressure λcr can be determined by solving the the modal stability analysis. This paper is mainly focused on the study of
eigenvalues of Aae in the ROM-based aeroelastic model (Eq. (21)) under physical mechanism of single-mode panel flutter, thus the model sta­
different dynamic pressures λ. Then the aeroelastic stability of the bility at Ma=1.1, 1.2, 1.3, 1.4, 1.5, 1.6 are all investigated in detail.
flexible panel in low supersonic flow can be analyzed, and further the Fig. 13 shows the root loci of the aeroelastic system varying with
inducing mechanism of single-mode panel flutter in low supersonic flow nondimensional dynamic pressure λ at Ma = 2.0. The nondimensional
can be revealed. dynamic pressure λ varies from 100 to 800, and the interval is Δλ = 1.
For λ = 100, the results presented in Fig. 13 show that all the eigen­
4.3.1. Aeroelastic stability analysis values lie in the lower half-plane and the position of the panel is stable.
Using the ROM-based aeroelastic model and the CFD/CSD direct Increasing the nondimensional dynamic pressure, the first two eigen­
simulation method, the critical flutter dynamic pressures for all Mach frequencies converge and almost coincide when λ = 608. With further
numbers in the range of 1.1~2.8 have been determined. Fig. 12(a) and increase of λ, the trajectories of the first two modes start to diverge in
Fig. 12(b) show the dynamic pressure boundaries and the frequency direction nearly perpendicular to their trajectories before they converge.
boundaries, respectively. It can be seen that the flutter/frequency Specifically, the trajectory of the first mode moves to the upper side of
boundaries predicted by ROM-based method and calculated by CFD/ the plane, and the second mode moves to the lower side. At λ = 610, that
CSD direct simulation method are both in well agreement with those is, almost immediately after merging, the trajectory of the second mode
given by Wang et al. [2]. Therefore, it validates that these two flutter intersects the real axis and enters the upper half-plane, which means that
computational methods used in the present paper are correct and have a the aeroelastic system becomes unstable. At the same time, it can be
high numerical precision. For the ROM-based flutter analysis method, found that the frequencies of the first and second modes gradually
the time cost is mainly focused on the model training, while only one approach each other. When the dynamic pressure further increases, the
process of model training is needed for one Mach number. However, in first and second modes will eventually be coupled to form a coupled-
order to search for the flutter neutral point, the CFD/CSD direct simu­ type flutter through an aerodynamic coupling. From Fig. 13, it is
lation method needs to compute a lot of different responses with a wide obvious that the flutter frequencies are between the first and second
variety of dynamic pressures for one Mach number. This will lead to the natural frequency, and they are closer to the second natural frequency of
fact that the computational efficiency of the ROM-based method is 1–2 the panel. The eigenfrequencies of the third and higher modes vary only
orders of magnitude higher than of CFD/CSD direct simulation method. slightly and their motion in the complex plane does not lead to insta­
Thus, the ROM-based method is an appropriate and efficient flutter bility. Here, λcr is introduced as the critical flutter dynamic pressure
analysis method for single-mode panel flutter in low supersonic flow. On beyond which the aeroelastic system becomes unstable.
the other hand, as using linear ROM-based method can successfully From Fig. 13, it can be seen that λcr = 610 at Ma=2.0.
predict the stability boundaries for single-mode flutter, we can conclude The roots loci of the aeroelastic mode as the non-dimensional dy­
that single-mode flutter in low supersonic flow mainly depends on the namic pressure λ increases at Ma=1.1 is shown in Fig. 14. The nondi­
linear dynamics instead of the nonlinear characteristics of the unsteady mensional dynamic pressure λ varies in the range of 10~35, and the
aerodynamic forces. interval is Δλ = 1. It can be seen in Fig. 14 that for sufficiently small
As shown in Fig. 12, according to the type of flutter, the Ma-plane can dynamic pressure λ = 10, all the eigenvalues lie in the lower half-plane
be divided into two regions: Region I and Region II. In Region I and it means that the situation of the panel is stable. As the nondi­
(1.1 ≤ Ma ≤ 1.6), the instability of the panel is single-mode flutter, and mensional dynamic pressure λ increases to 21.9, the trajectory of the
the flutter frequency is close to a natural frequency of the panel. The first mode interacts the real axis and enters the upper half-plane. When
single-mode flutter includes first-order single-mode flutter and also the dynamic pressure λ goes beyond 21.9, the real parts of the eigen­
higher-order single-mode flutter. The first order single-mode flutter is values for the first mode becomes positive, which means that the aero­
observed in the range of 1.1 ≤ Ma ≤ 1.3, and the flutter frequency is elastic system becomes unstable. At that time, it should be noticed that
very close to the first natural frequency of the panel. For 1.4 ≤ Ma ≤ 1.6, the frequencies do not approach one another, that is, there is no inter­
the higher order mode instability emerges, and the flutter frequency is action of two oscillation modes as in coupled flutter. Thus, this insta­
very close to the natural frequency of the higher mode. In Region II bility is flutter with a single degree of freedom, which is also called
(1.7 ≤ Ma ≤ 2.8), the instability of the panel is coupled mode panel single-mode flutter. The results presented in Fig. 14 show that for the
flutter. For this type of flutter, the flutter frequency is located between range of dynamic pressures considered here, the trajectories of other
the first natural frequency and the second natural frequency of the modes except the first mode always lie in the lower half-plane, which

8
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 8. Identified results under the training signal compared with those of CFD simulations for Ma=1.3 andμ = 0.1.

means that these modes always maintain stable and their motion does dynamic pressure λ varies in the range of 10~50, and the interval is
not lead to the instability of the flexible. Increasing the nondimensional Δλ = 1. Shown in Figs. 15 and 16, it can be obviously seen that the first-
dynamic pressure λ from 10 to 30 and selecting the interval as Δλ = 1, order single-mode flutter also occurs at Ma=1.2 and Ma=1.3. At
Fig. 15 shows the root loci of the aeroelastic system at Ma=1.2. Fig. 16 Ma=1.2, the real parts of the eigenvalues for the first mode becomes
shows the roots loci of aeroelastic mode at Ma=1.3. The nondimensional positive when the dynamic pressure λ goes beyond 14.5. The first mode

9
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 9. The verification signals.

Fig. 10. One of the basis mode excitation of verification.

loses the stability when λ ≥ 22.7at Ma=1.3. not provided by Davis and Bendiksen [45] and Alder [30], it is possible
For 1.4 ≤ Ma ≤ 1.6, the single-mode flutter occurs at higher modes. that the flutter mode at Ma=1.6 has been misidentified in these litera­
Increasing the nondimensional dynamic pressure λ from 10 to 90 and tures due to the improper flutter boundary calculation method and
selecting the interval as Δλ = 1, Fig. 17 shows the root loci of the inadequate simulation time. This is because the time-domain flutter
aeroelastic system at Ma=1.4. As shown in Fig. 17, it can be seen that the calculations are usually initiated by the first mode velocity disturbance,
second mode is the first to become unstable and it loses the stability near and sometimes it takes quite a long time for the unstable higher-order
λ = 83.7. From Fig. 17, it is clearly can be seen that there is no obvious mode become noticeable. As one of the scenarios, Fig. 20 shows the
coupling between the aeroelastic modes. Then it can be defined that this time domain response of the panel for Ma=1.6 and λ = 400and Fig. 21 is
stability is single-mode flutter, and the single-mode flutter occurs in the the corresponding time-frequency distribution. The point observed is
higher mode. The results presented in Fig. 17 show that for the range of 3/4 chord point (x/l = 0.75) and the initial disturbance is provided by
dynamic pressures considered here, the real parts of eigenvalues for the first mode velocity q̇1 = 50. It is assumed that the simulation
other modes except the second mode are always negative, and it means calculation is terminated when the total nondimensional time reaches 8
that the oscillations of these modes are always damped. The motions of (involves 8000 time steps), which is shown in dashed line. From Fig. 20,
these modes are incapable of causing the instability of the flexible panel. it will make a judgement that this aeroelastic system is stable under the
Fig. 18 shows the root loci of the aeroelastic system varying with given dynamic pressure according to the damping rate of the response
nondimensional dynamic pressure λ at Ma = 1.5. The nondimensional curve (or the converging trend). However, it can be seen in Fig. 20 that
dynamic pressure λ varies in the range of 10~160, and the interval is Δλ the panel is actually unstable if the calculation time is long enough.
= 1. Increasing the nondimensional dynamic pressure λ from 10 to 260 Comparatively, if the ROM-based flutter analysis method is adopted, the
and selecting the interval as Δλ = 1, Fig. 19 shows the root loci of the unstable mode of the aeroelastic system can be instantly obtained shown
aeroelastic system at Ma = 1.6. From Fig. 18~19, it can be seen that the in Fig. 19. Therefore, compared with the CFD/CSD direct simulation
single-mode flutter occurs at the third mode for Ma = 1.5, and the method, the ROM-based method even has a higher accuracy in pre­
single-mode flutter takes place at the fifth mode for Ma = 1.6. What’s dicting flutter boundaries for higher-order single-mode flutter in low
more, the critical flutter dynamic pressures are λcr = 127.8 at Ma = supersonic flow to a certain degree. In fact, the point that in
1.5and λcr = 219.9 at Ma = 1.6. time-domain simulation the result depends on the initial perturbation,
In fact, besides Wang et al. [2] and the author in the present paper, was explored in detail by Shishaeva et al. [47]. Moreover, she also
Davis and Bendiksen [45], Alder [30] and Bhatia and Beran [46] have analyzed nonlinear development of limit cycles caused by different
also calculated the instability boundaries in low supersonic flow. These initial perturbations, and possibilities to move from one limit cycle to
instability boundaries provided by different researchers agree well the other.
except for the case of Ma=1.6. For Ma=1.6, the critical dynamic pres­ In fact, the definitive experiment on single degree of freedom flutter
sure determined by Wang et al. [2] and author in the present paper agree at low supersonic Mach numbers was done at NASA Ames Research
well with those given by Bhatia and Beran [46], but much lower than Center [18] and also discussed by Dowell [48]. In the above experiment
that given by Davis and Bendiksen [45] and Alder [30]. Davis and they showed that single degree of freedom flutter only occurs in the first
Bendiksen [45] and Alder [30] mainly used the time-domain CFD/CSD mode and not in the higher modes. They also showed that single degree
method to predict the instability boundaries. Wang et al. [2] has pointed of freedom flutter is strongly influenced by a turbulent boundary layer.
out that as the detailed calculation procedures of flutter boundaries are Both of these conclusions were further confirmed by Dowell [49] based

10
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 11. Comparisons of the time responses from ROM and those from CFD simulations Ma=1.3 and μ = 0.1.

upon computations using a shear layer aerodynamic theory. As or even eliminate flutter altogether in the higher modes. Considering the
explained in the above book [48] and Ref. [49], single degree of freedom Figs. 17–19 (the cases of 1.4 ≤ Ma ≤ 1.6), it can be seen that the ratio of
flutter is due to negative aerodynamic damping. The effect of a turbulent the real to imaginary parts of the aeroelastic eigenvalues is much smaller
boundary layer or structural damping is to offset the negative inviscid for higher modes and greater compared to this ratio for mode 1. This
aerodynamic damping and thus significantly raise the flutter boundary ratio is a non-dimensional measure of aerodynamic damping and can be

11
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 12. Flutter characteristics of simply supported 2D panel in supersonic flow.

Fig. 13. Eigenvalue loci of the aeroelastic system with the first six modes in the Fig. 15. Eigenvalue loci of the aeroelastic system with the first six modes in the
complex plane for Ma = 2.0, μ = 0.1 and 100 ≤ λ ≤ 800. Points 1 and 2 are the complex plane for Ma = 1.2, μ = 0.1 and 10 ≤ λ ≤ 30. Points 1 and 2 are the
eigenvalue for λ = 100 and λ = 800, respectively. Point 3 shows the frequencies eigenvalue for λ= 10 and λ = 30, respectively. Point 3 shows the frequencies in
in a vacuum. a vacuum.

Fig. 14. Eigenvalue loci of the aeroelastic system with the first six modes in the Fig. 16. Eigenvalue loci of the aeroelastic system with the first six modes in the
complex plane for Ma = 1.1, μ = 0.1 and 10 ≤ λ ≤ 35. Points 1 and 2 are the complex plane for Ma = 1.3, μ = 0.1 and 10 ≤ λ ≤ 50. Points 1 and 2 are the
eigenvalue for λ= 10 and λ = 35, respectively. Point 3 shows the frequencies in eigenvalue for λ= 10 and λ = 50, respectively. Point 3 shows the frequencies in
a vacuum. a vacuum.

compared to typical values of positive structural damping that are than that for higher modes. The positive value of the ratio of the real to
typically of the order of 0.01 or greater. At 1.4 ≤ Ma ≤ 1.6, the imaginary part of the aeroelastic eigenvalue corresponds to the negative
single-mode flutter occurs at higher modes. Taking the case of Ma = 1.4 aerodynamic damping. It means that the negative aerodynamic damping
for example, Fig. 22 show the comparisons of the ratios of the real to is much smaller for higher modes compared to this ratio for mode 1.
imaginary parts of the aeroelastic eigenvalues for different modes. The Thus, it is unlikely that these higher modes will be seen to flutter in a
results presented in Fig. 22 show that with the dynamic pressure varying single degree of freedom in an experiment. Note that the results shown
from 10 to 1000, the positive values of the ratios of the real to imaginary in Figs. 17–19 and 22 indicate these higher modes are indeed weakly
part of the aeroelastic eigenvalue for the mode 1 are significantly larger unstable in the absence of any structural damping or a viscous boundary

12
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 20. Time domain response.

Fig. 17. Eigenvalue loci of the aeroelastic system with the first six modes in the
complex plane for Ma = 1.4, μ = 0.1 and 10 ≤ λ ≤ 90. Points 1 and 2 are the
eigenvalue for λ= 10 and λ = 90, respectively. Point 3 shows the frequencies in
a vacuum.

Fig. 21. Time-frequency distribution.

relevant parameters in ARX model are introduced first. For the unsteady
aerodynamic reduced order model ARX used here, the corresponding
input delay order nb items have specific and distinct physical meanings,
which are different from those of the corresponding output delay order
na items. Specifically,
∑ 1
The corresponding nb items (i.e. nb− i=0 Bi u(k − i) in Eq. (9)) represent
taking into account panel vibration informations in modeling of un­
steady aerodynamic forces and they correspond to the derivative of vi­
Fig. 18. Eigenvalue loci of the aeroelastic system with the first six modes in the bration quantity with respect to time. For example, when nb = 1 is
complex plane for Ma = 1.5, μ = 0.1 and 10 ≤ λ ≤ 160. Points 1 and 2 are the
selected, it means that the effect of vibrating displacement of the panel is
eigenvalue for λ= 10 and λ = 160, respectively. Point 3 shows the frequencies in
considered for modeling the unsteady aerodynamic forces; if nb = 2, it
a vacuum.
indicates that not only the vibrating displacement but also the vibration
velocity of the panel is taken into account; for nb = 3, besides the
displacement and velocity, the unsteady aerodynamic modeling will
consider the vibration acceleration of the panel. And so on. The bigger
the value of nb, the higher the orders of derivatives of vibration quan­
tities with respect to time in the aerodynamic model. In the ARX-based
∑
aerodynamic model, the corresponding na items (i.e. na i=1 Ai y(k − i) in
Eq. (9)) relate to the unsteady characteristics of the flow, specifically the
history effects of the unsteady aerodynamic forces. As shown in Eq. (9),
it can be found that in the case of na = 1, the calculation of aerodynamic
force at the current time will consider the aerodynamic force at the last
time. When the condition of na = 2 is adopted, it means that the aero­
dynamic force at the last time and the time before that will be taken into
account for estimating the unsteady aerodynamic force at the current
time. And so on. Generally speaking, the larger the value of the output
delay order na, the longer the history of development of flow field needs
Fig. 19. Eigenvalue loci of the aeroelastic system with the first six modes in the to be traced for calculating the unsteady aerodynamic forces with ARX
complex plane for Ma = 1.6, μ = 0.1 and 10 ≤ λ ≤ 260. Points 1 and 2 are the model. From the above discussion, it can be seen that an ARX-based
eigenvalue for λ= 10 and λ = 260, respectively. Point 3 shows the frequencies in aerodynamic model with na = 0 and nb ∕ = 0 is a quasi-steady aero­
a vacuum. dynamic model.

layer. In fact, the role of the structural damping and rapid stabilization 4.3.4. Physical mechanism analysis
of higher modes was explored by Vedeneev [50]. In large-amplitude For the ARX-based ROM, choosing the appropriate out/input delay
case, the situation is different as shown by Shishaeva et al. [47]. orders is particularly important for accurately predicting flutter
boundaries. The results shown in Fig. 12 indicate that using the ROM-
4.3.3. Physical meanings of relevant parameters in ARX model based method to calculate the stability boundaries can obtain satisfac­
In order to further make clear the physical mechanism of single- tory results. Eq. (23) is used to evaluate the error for calculating the
mode panel flutter in low supersonic flow, the physical meanings of critical flutter dynamic pressures between the ROM-based method and

13
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 22. Comparisons of the ratios of the real to imaginary parts of the aeroelastic eigenvalues for different modes at Ma=1.4.

the CFD-based method. Table 1 shows the out/input delay orders that selecting the corresponding optimal input/output delay orders
adopted in Fig. 12 and the corresponding errors for different Mach (na = na optimal, nb = nb optimal) in the ARX model at different Mach
numbers. In fact, na and nb presented in Table 1 are the optimal output numbers obtains satisfactory prediction results. It can be seen in Fig. 23
delay order (na optimal) and the optimal input delay order (na optimal), that critical dynamic pressure predicted in the case of na = 0, nb =
which minimize the error between the ROM and the CFD-based model. nb optimalat every Mach number is much higher than that calculated by
Take Ma = 1.1 for example, only when na = 11 and nb = 6, the critical using the optimal input/output delay orders. In other words, using the
flutter dynamic pressure predicted by the ROM-based aeroelastic model ARX-based aerodynamic model with na = 0, nb = nb optimal to predict
is the closest to that given by CFD-based method. From Table 1, we can the instability boundaries will lead to a huge error. Also, the results in
find an interesting phenomenon. In the case of couple-mode flutter Fig. 23 show that with the increase of the input delay order nb, the
(1.7 ≤ Ma ≤ 2.8), the values of optimal output delay orders are all equal critical flutter dynamic pressures calculated by the ARX-based aero­
to zero, i.e. na = 0. It means that for coupled-mode flutter, the ARX- dynamic model with na = 0, nb ∕ = 0 decreases slightly, but still much
based ROM with na = 0 have satisfied prediction precision. According higher than those with na = na optimal, nb = nb optimal. It should be
to the physical meanings of the corresponding input/output delay order noted that the results from ARX model with na = 0 are very close to that
items introduced in the above section, it can be concluded that a quasi- from piston theory. It means that a quasi-steady aerodynamic model,
steady aerodynamic model is precise enough to calculate the flutter which even considers much information of panel vibration, still cannot
boundaries for coupled-mode flutter. As is known to all, the quasi-steady conduct the aeroelastic analysis of single-mode panel flutter in low su­
piston theory is widely used for coupled-mode flutter analysis. For the personic flow. The instability boundaries of single-mode flutter calcu­
case of single-mode flutter (1.1 ≤ Ma ≤ 1.6), the results presented in lated using the quasi-steady aerodynamic model will lead to inaccurate
Table 1 show that the optimal output delay order na are all greater than
zero, i.e. na > 0. In addition, the value of na almost increases as Mach
number decreases. These results indicate that in order to precisely pre­
dict the flutter boundaries of single-mode flutter in low supersonic flow,
the unsteady characteristics of the flow, or more specifically, the history
effects of the unsteady aerodynamic forces should be considered in
aerodynamic model.
|λcr − λcr ROM |
(23)
CFD
Error2 =
|λcr CFD |
Fig. 23 shows the influence of delay orders in ARX model on the
critical flutter dynamic pressures at different Mach numbers in the case
of single-mode panel flutter. Using Ma=1.1 as an example to explain the
line legends presented in Fig. 23. According to Table 1, na = na optimal,
nb = nb optimal at Ma=1.1 means na = 11,
nb = 6. Then na = 0, nb = nb optimal at Ma=1.1 represents na = 0,
nb = 6. In accordance to the introduction of the physical meanings of
the corresponding input/output delay order items in the above section,
na = 0 means the unsteady characteristics of the flow, specifically the
history effects of the unsteady aerodynamic forces, are not considered in
Fig. 23. The influence of delay orders on the critical flutter dynamic pressures
the ARX-based aerodynamic model. From Fig. 12, it has been known in the case of single-mode panel flutter.

Table 1
The optimal out/input delay orders and the corresponding errors under different Mach number.
Single-mode panel flutter Coupled-mode panel flutter

Ma 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 2.0 2.2 2.4 2.6 2.8
na optimal 11 8 6 1 1 1 0 0 0 0 0 0 0
nb optimal 6 7 7 6 3 4 3 3 3 3 3 3 3
Error2/% 2.0 2.0 1.9 1.5 1.2 1.3 1.2 1.1 0.9 0.7 0.7 0.5 1.3

14
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

results. To correctly predict the flutter boundary of single-mode panel

flutter in low supersonic flow, the unsteady aerodynamic model used
must take into account the unsteady characteristics of the flow, specif­
ically, the history effects of the unsteady aerodynamic forces. This
finding helps explain why the piston theory is not applicable at low
supersonic Mach numbers, because it is a quasi-steady aerodynamic
model which loses accuracy for single-mode flutter in low supersonic
flows.
To quantify the importance of the unsteady characteristics of the
flow in predicting the flutter boundaries of single-mode flutter in low
supersonic flow, Eq. (24) is used to calculate the error caused by not
considering the unsteady effects of aerodynamic forces. In Eq. (24),
λcr(na optimal nb optimal) denotes the critical flutter dynamic pressure ob­
tained by employing the optimal input/output delay orders. Selecting
the output delay order as zero (na = 0) and considering the optimal Fig. 24. Eigenvalue loci of the aeroelastic system with the first six modes in the
input delay order (nb = nb optimal), the critical flutter dynamic pressure complex plane for Ma = 1.1, μ = 0.1 and 5 ≤ λ ≤ 1000. Points 1 and 2 are the
λcr(na 0 nb optimal) is calculated according to the ARX-based aerodynamic eigenvalue for λ = 5 and λ = 1000, respectively. Point 3 shows the frequencies
in a vacuum.
model. It can be seen that the absolute error value at Ma=1.6 is mini­
mum, and the maximum absolute error appears at Ma=1.2. The mini­
mum absolute error at Ma=1.6 caused by not considering the unsteady mode flutter, which actually develops, and yields another coupled-
effects of aerodynamic forces reaches 99.53 %. For Ma=1.2, neglecting type instability, which in fact is not present. However, the unsteady
the unsteady effects of aerodynamic forces will lead to the fact that the aerodynamic model (i.e. the ARX-based ROM with the delay orders na =
absolute error of instability boundaries of single-mode flutter in low na optimal, nb = nb optimal) with considering the unsteady character­
supersonic flow even reaches 1569.9 %. istics of the flow can not only precisely predict the instability boundaries
⃒ ⃒ of the panel at low supersonic speeds, but also accurately capture the
⃒λcr(na 0 nb optimal) − λcr(na optimal nb optimal) ⃒ single-mode panel flutter. Thus, we can conclude from the results that
Error3 = ⃒ ⃒ (24)
⃒λcr(na optimal nb optimal) ⃒ the single-mode panel flutter at low supersonic speeds is mainly induced
by the unsteady characteristics of the flow (specifically the history ef­
In order to illustrate that the single-mode panel flutter at low su­ fects of the unsteady aerodynamic forces). And that is why the single-
personic speeds is mainly induced by the unsteady characteristics of the mode flutter in low supersonic flow cannot be detected by the piston
flow (specifically the history effects of the unsteady aerodynamic theory or other quasi-steady aerodynamic theory. In fact, Figs. 13-19
forces), we will further use the quasi-steady aerodynamic model to show frequency loci similar to those obtained in Vedeneev [24]. What’s
conduct the aeroelastic modal stability analysis of the panel in low su­ more, the limitation of the quasi-static theories to the coupled-mode
personic flow. The quasi-steady aerodynamic model used here is the flutter was also obtained by Vedeneev [24]. In the present paper, this
ARX-based ROM with the delay orders na = 0,nb = nb optimal. The ROM-based analysis did not reveal anything new, but since this analysis
corresponding optimal input delay orders (i.e. nb optimal) at different is not restricted to flat panels/2D problems, it could be a power tool for
Mach numbers are shown in Table 1. For comparative analysis, we still future studies of more complex problems.
investigate the variation of aeroelastic modes at Ma=1.1 and Ma=1.6
according to the Section (4.3.2. Modal stability analysis). Fig. 24 and 4.3.5. Effect of the mass ratio
Fig. 25 show the root loci of the aeroelastic system varying with The mass ratio μ defined in the present paper reflect the density
nondimensional dynamic pressure λ at Ma=1.1 and Ma=1.6, respec­ characteristics of the aeroelastic system, and the detailed expression of
tively. Obviously, the results in accordance with the quasi-steady aero­ μcan be seen in Eq. (2). When the density of structural is fixed, the air
dynamic model (i.e. the ARX-based ROM with the delay orders na = 0, density of the flow above the panel increases with the increase of the
nb = nb optimal) presented in Figs. 24~25 differ sharply from the re­ mass ratio μ. Thus, the large value of mass ratio corresponds to the high
sults calculated by using the unsteady aerodynamic model (i.e. the ARX- density of the air, and this is equivalent to a situation in which the
based ROM with the delay orders na = na optimal, nb = nb optimal), aircraft flies at a lower altitude. Conversely, the small value of mass ratio
which are shown in Figs. 14 and 19. It can be seen in Figs. 24~25 that corresponds to a situation in which the aircraft flies at a higher altitude.
within the framework of the quasi-steady aerodynamic model, single- When the air density is fixed, the variation of the mass ratio μrepresents
mode flutter cannot be observed at all. The variations of aeroelastic the change of the density of structure. It is necessary to study the effect
modes at Ma=1.1 and Ma=1.6 show the similar characteristics, and then of mass ratio μon the single-mode panel flutter in low supersonic flow.
the case of Ma=1.1 will be taken as an example to introduce these The Mach number of 1.1, 1.2,1.3 1.4, 1.5 and 1.6 are selected and the
variations. As shown in Fig. 24, the real parts of the eigenvalues for the range of mass ratio chosen here is μ = 0.0001~0.20.
first mode becomes positive when the dynamic pressure λ goes beyond Fig. 26 and Fig. 27 show the effects of mass ratios on the critical
290.5. The trajectories of all the other modes lie in the lower half-plane, flutter dynamic pressures and flutter frequencies, respectively. It can be
and their motions will not lead to the instability of the panel. The critical seen that the mass ratio has a significant effect on the flutter boundaries
flutter dynamic pressure λcr = 290.5 predicted here agree with that of single-mode panel flutter in low supersonic flow. Specifically, the
shown in Fig. 23. From Fig. 24, it should be noted that at λ = 290.5 a critical dynamic pressure will monotonically increase with the increase
coupled-type flutter develops, with the 1st and 2nd modes involved, but of the mass ratio. As the mass ratio goes to zero, the critical flutter dy­
it has no physical meaning. In fact, the type of instability presented in namic pressure is likely to go to a constant value. The mass ratio has
Fig. 24 are also the coupled-mode panel flutter. In addition, the critical almost no influence on the flutter frequencies. From Fig. 26(a), it can be
flutter dynamic pressures shown in Fig. 24 are in accordance with those seen that the minimum critical flutter dynamic pressure (or the lowest
presented in Fig. 23, and it is previously known that they are incorrect. aeroelastic stability boundary) occurs at Ma=1.2, instead of Ma=1.1.
Generally, it can be found that at the low supersonic speeds, the quasi- This finding is consistent with that obtained by Alder [32].
steady aerodynamic model (i.e. the ARX-based ROM with the delay Take the case of Ma=1.1 as an example, the effects of mass ratio μon
orders na = 0, nb = nb optimal) becomes not only quantitatively the aeroelastic mode stability are investigated by using the ROM-based
incorrect but also qualitatively incorrect. It “does not see” the single- aeroelastic model, as shown in Figs. 28 and 29. It can be seen that the

15
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

Fig. 25. Eigenvalue loci of the aeroelastic system with the first six modes in the complex plane for Ma = 1.6, μ = 0.1 and 10 ≤ λ ≤ 500. Points 1 and 2 are the
eigenvalue for λ= 10 and λ = 500, respectively. Point 3 shows the frequencies in a vacuum.

Fig. 26. Dynamic pressure boundaries of simply supported 2D panel in supersonic flow.

Fig. 27. Frequency boundaries of simply supported 2D panel in supersonic flow.

variation of the mass ratio μ will not change the instability type of single- Considering the cases of different mass ratios shown in Figs. 28 and 29, it
mode panel flutter in low supersonic flow. Specifically, assuming that also can be seen that at the low supersonic speeds, the quasi-steady
the instability type of the flexible panel in low supersonic flow is single aerodynamic model (i.e. the ARX-based ROM with the delay orders na =
mode flutter at one mass ratio μ. If the mass ratio μ is changed, single- 0, nb = nb optimal) “does not see” the single-mode flutter, which
mode flutter still occurs when the aeroelastic system loses its stability. actually develops, and yields another coupled-type instability, which in

16
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

relevant parameters in unsteady aerodynamic reduced order model

ARX, the inducing mechanism of single-mode flutter in low supersonic
flow is revealed. The effects of mass ratio on the flutter boundaries,
flutter frequencies and instability types of the aeroelastic system in low
supersonic flow are also investigated. The coupled CFD/CSD simulation
is also performed as a supplement to time-domain responses. Main
conclusions can be summarized as follows:
(1) The flutter boundaries in low supersonic flows given by ROM-
based method agree well with those by the CFD-based direct simula­
tion method. Thus, the more efficient ARX-based ROM can be used for
modeling unsteady aerodynamic forces in low supersonic flows.
(2) The mass ratio has a significant effect on the flutter boundaries of
single-mode panel flutter in low supersonic flow. Specifically, the crit­
ical dynamic pressure will monotonically increase with the increase of
Fig. 28. Eigenvalue loci of the aeroelastic system with the first six modes in the
the mass ratio. The mass ratio has almost no influence on the flutter
complex plane for Ma = 1.1, μ = 0.01 and 10 ≤ λ ≤ 600. Points 1 and 2 are the
frequencies. The variation of the mass ratio will not change the insta­
eigenvalue for λ= 10 and λ = 600, respectively. Point 3 shows the frequencies in
a vacuum.
bility type of single-mode panel flutter in low supersonic flow. If the
mass ratio is changed, single-mode flutter still occurs.
(3) The ROM-based analysis method has been proved to be a highly
efficient flutter analysis method which allowed qualitative analysis of
the flow unsteadiness, modal composition, etc. In this particular prob­
lem this analysis did not reveal anything new, but since this analysis is
not restricted to flat panels/2D problems, it could be a power tool for
future studies of more complex problems.

CRediT authorship contribution statement

Liu-qing Ye: Writing – original draft, Validation, Software, Meth­

odology, Investigation, Formal analysis, Data curation, Conceptualiza­
tion. Kun Ye: Writing – review & editing, Visualization, Supervision,
Software, Resources, Methodology, Investigation, Formal analysis, Data
curation, Conceptualization. Ke Jin: Validation, Methodology, Investi­
gation, Formal analysis, Data curation. Yi-fan Zhang: Validation,
Fig. 29. Eigenvalue loci of the aeroelastic system with the first six modes in the
Investigation, Data curation. Zheng-yin Ye: Investigation, Formal
complex plane for Ma = 1.1,μ = 0.2 and 10 ≤ λ ≤ 600. Points 1 and 2 are the
analysis.
eigenvalue for λ= 10 and λ = 600, respectively. Point 3 shows the frequencies in
a vacuum.
Declaration of competing interest
fact is not present. Thus, it adds to the proof that the unsteady effects of
aerodynamic forces play a significant role in causing single-mode panel The authors declare that they have no known competing financial
flutter in low supersonic flow. The unsteady characteristics of aero­ interests or personal relationships that could have appeared to influence
dynamic forces are the necessary conditions for the occurrence of single- the work reported in this paper.
mode flutter in low supersonic flow. This is the main reason why the
piston theory or other quasi-steady aerodynamic theory cannot detect Data availability
the single-mode flutter. In our further research work, we will study
deeply the instability characteristics of each order model when the Data will be made available on request.
single-mode flutter occurs in low supersonic flow.

5. Conclusions Acknowledgments

In this paper, an ARX-based ROM for low supersonic flow is con­ The authors would like to acknowledge the funding support of this
structed by using CFD technology and the system identification method. research by National Natural Science Foundation of China (NSFC) (No.
Coupling with the structural equation, a highly efficient ROM-based U2341241, No. 12272306, No. 52175510, No. 12072246), and the 111
flutter analysis method is formulated, and then the flutter boundaries project of China (no. B17037). The authors would like to thank Professor
are obtained and the stability of aeroelastic modes are investigated by Earl H. Dowell for his helpful comments on a draft of this paper and
the complex eigenvalue analysis. According to the physical meaning of having thoroughly revised the draft of this paper.

Appendix A. Nondimensional parameters used for nondimensionalizing the flow equations

The flow equations are nondimensionalized by using the following nondimensional parameters:
Δτa = Laref∞ Δt, x = x ∗ /Lref , y = y ∗ /Lref , z = z ∗ /Lref
/ / /
Vref = V∞ , u = u ∗ Vref , v = v ∗ Vref , w = w ∗ Vref

Tref = T∞ , μref = μ∞ , ρref = ρ∞

17
L.-q. Ye et al. Aerospace Science and Technology 147 (2024) 109002

/ / /
T = T ∗ Tref , μ = μ ∗ μref , ρ = ρ ∗ ρref

where, the reference length Lref is the plate length in the present paper.

References [26] V.V. Vedeneev, Single-mode plate flutter taking the boundary layer into account,
Fluid Dyn. 47 (3) (2012) 417–429.
[27] V. Bondarev, V. Vedeneev, Influence of the viscous boundary layer perturbations
[1] E.H. Dowell, K.C. Hall, Modeling of fluid-structure interaction, Annu. Rev. Fluid
on single-mode panel flutter at finite reynolds numbers, J. Fluid Mech. 852 (2018)
Mech. 33 (1) (2001) 445–490.
578–601.
[2] G. Wang, H. Zhou, H.H. Mian, Numerical analysis on modal stability characteristics
[28] A. Shishaeva, V.V. Vedeneev, A. Aksenov, G. Sushko, Transonic panel flutter in
of 2D panel flutter at low supersonic speeds, J. Fluids Struct. 103 (2021) 1–19.
accelerating or decelerating flow conditions, AIAA J. 56 (3) (2018) 997–1010.
[3] W. Tian, T. Zhao, Y.S. Gu, Z.C. Yang, Nonlinear flutter suppression and
[29] V.V. Vedeneev, S.V. Guvernyuk, A.F. Zubkov, M.E. Kolotnikov, Experimental
performance evaluation of periodically embedded nonlinear vibration absorbers in
observation of single mode panel flutter in supersonic gas flow, J. Fluids Struct. 26
a supersonic FGM plate, Aerosp. Sci. Technol. 121 (2022) 107198.
(5) (2010) 764–779.
[4] J.J. Tang, S.B. Wu, M. Habibi, M. Safarpour, H.E. Ali, Flutter analysis of multi-
[30] R.E. Godnier, M.R. Visbal, Development of a three-dimensional viscous aeroelastic
directional functionally graded sector poroelastic disks, Aerosp. Sci. Technol. 140
solver for nonlinear panel flutter, J. Fluids Struct. 16 (4) (2002) 497–527.
(2023) 108481.
[31] A. Shishaeva, V. Vedeneev, A. Aksenov, Nonlinear single-mode and multi-mode
[5] Z.P. Zhang, Y.W. Wang, W. Zhang, Temperature- and moisture-dependent
panel flutter oscillations at low supersonic speeds, J. Fluids Struct. 56 (2015)
aeroelastic stability of graphene platelet reinforced nanocomposite lattice
205–223.
sandwich plates subjected to supersonic flow, Aerosp. Sci. Technol. 138 (2023)
[32] M. Alder, Development and validation of a fluid-structure solver for transonic
108348.
panel flutter, AIAA J. 53 (12) (2015) 3509–3521.
[6] L.Q. Ye, Z.Y. Ye, Effects of shock location on aeroelastic stability of flexible panel,
[33] M. Alder, Nonlinear dynamics of prestressed panels in low supersonic turbulent
AIAA J. 56 (9) (2018) 1–13.
flow, AIAA J. 54 (11) (2016) 3632–3646.
[7] L.Q. Ye, Z.Y. Ye, K. Ye, J. Wu, Aeroelastic stability and nonlinear flutter analysis of
[34] J.Y. Yang, W.W. Zhang, Forced response analysis of the rotor blade rows with the
viscoelastic heated panel in shock-dominated flows, Aerosp. Sci. Technol. 117
ROM-based aeroelastic model, Aerosp. Sci. Technol. 139 (2023) 108366.
(2021) 1–22.
[35] L. Kai, J.Q. Kou, W.W. Zhang, Unsteady aerodynamic reduced-order modeling
[8] E.H. Dowell, Nonlinear oscillations of a fluttering plate, AIAA J. 4 (7) (1966)
based on machine learning across multiple airfoils, Aerosp. Sci. Technol. 119
1267–1275.
(2021) 1–13.
[9] E.H. Dowell, Nonlinear oscillations of a fluttering plate II, AIAA J. 4 (10) (1967)
[36] D.J. Lucia, P.S. Beran, W.A. Silva, Reduced-order modeling: new approach for
1856–1862.
computational physics, Prog. Aerosp. Sci. 40 (2004) 51–117.
[10] A. Voß, T. Klimmek, Parametric aeroelastic modeling, maneuver loads analysis
[37] M. Ghoreyshi, A. Jirásek, R.M. Cummings, Reduced order unsteady aerodynamic
using CFD methods and structural design of a fighter aircraft, Aerosp. Sci. Technol.
modeling for stability and control analysis using computational fluid dynamics,
136 (2023) 108231.
Prog. Aerosp. Sci. 71 (2014) 167–217.
[11] N. Tsushima, K. Saitoh, K. Nakakita, Structural and aeroelastic characteristics of
[38] W.W. Zhang, X.T. Li, Z.Y. Ye, Y.W. Jiang, Mechanism of frequency lock-in in
wing model for transonic flutter wind tunnel test fabricated by additive
vortex-induced vibrations at low Reynolds numbers, J. Fluid Mech. 783 (2015)
manufacturing with AlSi10Mg alloys, Aerosp. Sci. Technol. 140 (2023) 108476.
72–102.
[12] E.H. Dowell, Panel flutter: a review of the aeroelastic stability of plates and shells,
[39] C.Q. Gao, W.W. Zhang, X.T. Li, Y.L. Liu, J.G. Quan, Z.Y. Ye, Y.W. Jiang, Mechanism
AIAA J. 8 (3) (1970) 385–399.
of frequency lock-in in transonic buffeting flow, J. Fluid Mech 818 (2017)
[13] C. Mei, K. Abel-Motagaly, R. Chen, Review of nonlinear panel flutter at supersonic
528–561.
and hypersonic speeds, Appl. Mech. Rev. 52 (10) (1999) 321–332.
[40] X.T. Li, Z. Lyu, J.Q. Kou, W.W. Zhang, Mode competition in galloping of a square
[14] J.J. McNamara, P.P. Friedmann, Aeroelastic and aerothermoelastic analysis in
cylinder at low Reynolds number, J. Fluid Mech. 867 (2019) 516–555.
hypersonic flow: past, present, and future, AIAA J. 49 (6) (2011) 1089–1122.
[41] K. Ye, Y.F. Zhang, Z. Chen, Z.Y. Ye, Numerical investigation of aeroelastic
[15] H.C. Nelson, H.J. Cunnigham, Theoretical Investigation of Flutter of Two-
characteristics of grid fin, AIAA J. 60 (5) (2022) 3107–3121.
Dimensional Flat Panels with One Surface Exposed to Supersonic Potential Flow,
[42] K. Ye, Z.Y. Ye, C.N. Li, W. Jie, Effects of the aerothermoelastic deformation on the
1956. NACA Report no. 1280.
performance of the three-dimensional hypersonic inlet, Aerosp. Sci. Technol. 84
[16] E.H. Dowell, Generalized aerodynamic forces on a flexible plate undergoing
(2019) 747–762.
transient motion in a shear flow with an application to panel flutter, AIAA J. 9 (5)
[43] A.V. Oppenheim, R.W. Schafer, Discrete-Time Signal Processing, 3rd ed., Pearson
(1971) 834–841.
Higher Education, 2010.
[17] L. Muhlstein, P.A. Gaspers, An Experimental Study of the Influence of the
[44] W.W. Zhang, Y.W. Jiang, Z.Y. Ye, Two better loosely coupled solution algorithms
Turbulent Boundary Layer On Panel Flutter, NASA TN D-4486, 1968.
of CFD based aeroelastic simulation, Eng. Appl. Comp. Fluid. 1 (4) (2007)
[18] P.A. Gaspers, L. Muhlstein, D.N. Petroff, Further Experimental Results On the
253–262.
Influence of the Turbulent Boundary Layer On Panel Flutter, NASA TN D-5798,
[45] G. Davis, O. Bendiksen, Transonic panel flutter, in: Proceedings of the 34th
1970.
Structures, Structural Dynamics and Materials Conference, American Institute of
[19] V.V. Vedeneev, Flutter of a wide strip plate in a supersonic gas flow, Fluid Dyn. 40
Aeronautics and Astronautics, 1993.
(5) (2005) 805–817.
[46] M. Bhatia, P. Beran, Transonic panel flutter predictions using a linearized stability
[20] V.V. Vedeneev, High-frequency flutter of a rectangular plate, Fluid Dyn. 41 (4)
formulation, AIAA J. 55 (10) (2017) 3499–3516.
(2006) 641–648.
[47] A. Shishaeva, A. Aksenov, V.V. Vedeneev, The effect of external perturbations on
[21] V.V. Vedeneev, High-frequency plate flutter, Fluid Dyn. 41 (2) (2006) 313–321.
nonlinear panel flutter at low supersonic speed, J. Fluids Struct. 111 (2022)
[22] V.V. Vedeneev, Nonlinear high-frequency flutter of a plate, Fluid Dyn. 42 (5)
103570.
(2007) 858–868.
[48] E.H. Dowell, Aeroelasticity of Plates and Shells, Leyden, Noordhoff International
[23] V.V. Vedeneev, Numerical investigation of supersonic plate flutter using the exact
Publish, 1975.
aerodynamic theory, Fluid Dyn. 44 (2)) (2009) 314–321.
[49] E.H. Dowell, Aerodynamic boundary layer effects on flutter and damping of plates,
[24] V.V. Vedeneev, Panel flutter at low supersonic speeds, J. Fluids Struct. 29 (2012)
J. Aircr. 10 (12) (1973) 734–738.
79–96.
[50] V.V. Vedeneev, Effect of damping on flutter of simply supported and clamped
[25] V.V. Vedeneev, Limit oscillatory cycles in the single mode flutter of a plate, J. Appl.
panels at low supersonic speeds, J. Fluids Struct. 40 (2013) 366–372.
Math. Mech. 77 (2013) 257–267.

18