AIAA JOURNAL

Vol. 59, No. 10, October 2021

Reduced-Order Nonlinear Damping Model: Formulation

and Application to Postflutter Aeroelastic Behavior

X. Q. Wang,∗ Pengchao Song,† and Marc P. Mignolet‡

Arizona State University, Tempe, Arizona 85287
and
P. C. Chen§
ZONA Technology, Scottsdale, Arizona 85258
https://doi.org/10.2514/1.J059804
Recent studies of postflutter limit-cycle oscillations (LCOs) have suggested the presence/effects of a nonlinear
structural damping in addition to other potential sources of nonlinearity. Such a nonlinearity occurs, for example, for
structures with linear viscoelastic properties when their responses are in the nonlinear geometric regime. The present
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

effort focuses on such situations; first on developing a structural reduced-order model (ROM) which could be used in
aeroelastic analyses. Adopting a linear Kelvin–Voigt constitutive model in the undeformed configuration, the ROM
governing equations are obtained and found to be of a generalized Van der Pol–Duffing form. A nonintrusive
identification approach is next developed to determine the parameters of these governing equations from a structural
finite element model constructed in a commercial software. Finally, the effects of this nonlinear damping on
postflutter response are analyzed on the Goland wing assuming a linear aerodynamic model. It is found that the
nonlinearity in the damping can stabilize the unstable aerodynamics and lead to finite amplitude limit-cycle
oscillations, even when the stiffness-related nonlinear geometric effects and aerodynamic nonlinearities are
neglected. The dependence of the LCO amplitude and frequency on the parameters of the Kelvin–Voigt model is
analyzed to provide insight into this nonlinearity.

Nomenclature α0 = Rayleigh damping coefficient, 1∕s

β0 = Rayleigh damping coefficient, s
b0 = body force vector with respect to the unde-
formed configuration; physical coordinates δij = Kronecker delta
C = fourth-order elasticity tensor; physical coor- ρ0 = density in the undeformed configuration
dinates Ω0 = structure domain in the undeformed
D = fourth-order dissipation tensor; physical configuration
coordinates
D
1
2
3
ij , Dijl , Dijlp
= linear, quadratic, and cubic damping coeffi-
cients; modal coordinates I. Introduction
E = Green strain tensor; physical coordinates
F = deformation gradient tensor; physical coor-
dinates
A LARGE number of investigations (e.g., Refs. [1–7]) have
focused in the past on the limit-cycle oscillations (LCOs)
observed during flight for certain configurations of the F-16. It is
Fi = ith force from external excitation; modal now generally well accepted that this behavior is a postflutter event,
coordinates but the source of the stabilization mechanism (i.e., the physical
K
1
2
3
ij , K ijl , K ijlp
= linear, quadratic, and cubic stiffness coeffi- process that transforms the unstable response at flutter into a finite
cients; modal coordinates amplitude LCO) remains unclear: purely aerodynamic phenomena
K
2 
3
ijl , K ijlp
= identified quadratic and cubic stiffness coef- have been proposed while other suggestions involve the appearance
ficients; modal coordinates of structural nonlinearity. The latter nonlinearity could be localized
M = mass matrix; modal coordinates and/or global and originate in stiffness and/or damping properties.
q
t = response vector; modal coordinates Localized nonlinearities include, in particular, freeplay (stiffness
S = second Piola–Kirchhoff stress tensor; physi- nonlinearity) and friction (damping nonlinearity). In fact, friction
cal coordinates was proposed as such a nonlinear damping in Ref. [1], and the
t = time response of an aeroelastic system with friction was investigated in
U
n = modal basis function detail in Refs. [8–10]: both theoretically [8,9] and experimentally
u = displacement vector; physical coordinates [10]. Although stabilization of the unstable response postflutter was
demonstrated, it was also shown that the LCO is subcritical, as
opposed to the supercritical LCO often observed in practice.
Presented as Paper 2016-1795 at the 15th Dynamics Specialists Conference,
San Diego, CA, January 4–8, 2016; received 20 June 2020; revision received 1
Global nonlinearities are associated in particular with “large”
March 2021; accepted for publication 20 April 2021; published online 19 July motions (i.e., nonlinear geometric effects), which are well recognized
2021. Copyright © 2021 by X. Q. Wang, Pengchao Song, Marc P. Mignolet, and to modify the “stiffness” structural properties but may also induce
P. C. Chen. Published by the American Institute of Aeronautics and Astronautics, nonlinearity in the damping terms in viscoelastic materials. This
Inc., with permission. All requests for copying and permission to reprint should effect does not appear to have been studied in the context of post-
be submitted to CCC at www.copyright.com; employ the eISSN 1533-385X to flutter LCO, and thus is the focus of this investigation to complement
initiate your request. See also AIAA Rights and Permissions www.aiaa.org/ the array of available structural nonlinearities models. The effort
randp. proceeds as in Ref. [11]. More specifically, a novel nonlinear damp-
*Research Scientist, SEMTE, Faculties of Mechanical and Aerospace ing model is first formally derived within the framework of finite
Engineering. Senior Member AIAA.
†
Postdoctoral Fellow, SEMTE, Faculties of Mechanical and Aerospace deformation viscoelasticity. This derivation is achieved in a reduced-
Engineering. order modeling (ROM) format consistent with the one used for the
‡
Professor, SEMTE, Faculties of Mechanical and Aerospace Engineering; nonlinear stiffness effects (see Ref. [12] for a review) and, more
[email protected]. Associate Fellow AIAA (Corresponding Author). specifically, the approach with dual modes originally proposed in
§
President. Associate Fellow AIAA. Ref. [13] (see Refs. [14–17] for some applications). Finally, a first
4144
 WANG ET AL. 4145

assessment of the capabilities of this model to induce LCO is carried 1

out on a simple wing model, i.e., the Goland wing. This study is Eij 
Fki Fkj − δij  (2)
2
conducted in the absence of all other structural or aerodynamic
nonlinearities, and thus will shed light on the effects and symptoms
of this nonlinearity. Note that a parallel investigation [18] has con- Note in the preceding equation and in the remainder of this paper
sidered the application of this nonlinear damping model to the F-16 that summation is implied over all repeated indices.
behavior. In the reference configuration, the equations of motion of the
The formulation of the proposed nonlinear structural damping structure are [22,23]
model in the context of the nonlinear ROM is described in Sec. II.
Section III presents the finite element structural and aeroelastic model ∂

F S   ρ0 b0i  ρ0 u i for X ∈ Ω0 (3)
of the test application, i.e., the heavy Goland wing [19,20]. The ∂X k ij jk
capability of the proposed nonlinear structural damping to bound
the otherwise unstable aeroelastic response is finally analyzed where S denotes the second Piola–Kirchhoff stress tensor; ρ0
X is
in Sec. IV. the density in the reference configuration; and b0 is the vector of body
forces, all of which may depend on the coordinates X i . Finally, Ω0 is
the domain occupied by the structure in the reference configuration.
II. Nonlinear Structural Reduced-Order Modeling To complete Eq. (3), it remains to specify the constitutive behavior
A. Section Overview of the material, i.e., relate the second Piola–Kirchhoff stress tensor S
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

to the Green strain tensor E. Since a damping/dissipation model is the

The focus of this section is primarily on the derivation of a non-
goal of the present derivation, an elastic behavior is not sufficient;
linear structural damping model that is consistent with the larger than
rather, a viscoelastic constitutive model is necessary. Specifically, a
linear deformations that are expected during LCO events. In Sec. II.B,
Kelvin–Voigt model will be assumed (see Refs. [22,24] for more
this model is developed from finite deformation viscoelasticity in a
general models) so that
reduced-order format consistent with mode-based aeroelasticity
analyses (e.g., see discussion of Sec. III.B). The identification of
the parameters of this model from a finite element representation of Sij  Cijkl Ekl  Dijkl E_ kl (4)
the structure performed in a commercial software (e.g., Nastran,
Abaqus) is next considered. Identifying the stiffness parameters has where C and D are the fourth-order elasticity and dissipation tensors,
been achieved in prior publications (e.g., see Refs. [12,13,21] and and the former satisfies the symmetry conditions
Sec. II.C), but the corresponding effort for the damping parameters is
novel. To this end, a parallel between the parameters of the nonlinear Cijkl  Cjikl  Cijlk Cijkl  Cklij (5)
damping model and the nonlinear stiffness coefficients associated
with the geometric nonlinearity is first demonstrated (see Sec. II.D), and the positive definiteness property
and the identification of the nonlinear damping model based on this
parallel is formulated. Aij Cijkl Akl > 0 (6)

B. Nonlinear Geometric Viscoelastic Modeling for any nonzero second-order tensor A. To ensure that dissipation
As stated in the Introduction (Sec. I), the framework for the takes place at all times for all possible deformation velocities, it is
derivation of the nonlinear structural damping model is finite defor- required here that
mation viscoelasticity. To proceed, let the position vector of a point of
the structure be denoted by X in the reference configuration and by x E_ ij Dijkl E_ kl > 0 (7)
in the deformed one so that the displacement vector is u  x − X;
see Fig. 1. That is, D must be positive definite.
Define next the deformation gradient tensor F of components Fij A reduced-order model of the preceding problem can be developed
as by expressing the displacement field ui
X; t as an expansion over M
∂xi ∂u basis functions U
n
i
X that do not change with time and automati-
Fij   δij  i (1) cally satisfy the geometric boundary conditions (assumed here to be
∂Xj ∂X j
homogenous). That is,
where δij denotes the Kronecker symbol. The deformations of the
structure will be quantified by the Green strain tensor E, the compo- ui
X; t  qn
tU
n
i
X (8)
nents of which are
where qn
t are the time-dependent generalized coordinates. Intro-
ducing this representation in Eq. (1) leads to

∂U
n
Fij  δij  qn i
(9)
∂Xj

where the dependences of qn on time and U
ni on the position X have

been suppressed for notational simplicity. Combining Eqs. (9) and (2)
leads to the following expression for the components of the Green
strain tensor:
" #
1 ∂U
n ∂U
n
j 1 ∂U
n
m
k ∂U k
Eij  qn i
  qn qm (10)
2 ∂Xj ∂X i 2 ∂X i ∂Xj

Differentiating this expression with respect to time and recombin-

Fig. 1 Reference and deformed configurations [23]. ing terms lead to
 4146 WANG ET AL.

2 3 Z
∂U
n
∂U
n ∂U
m ∂U
n
p
r ∂U r
1
E_ ij  q_ n 4 i 
j 5 D^
2 ^
2
mnp  Dmpn 
i
Dijkl dX (22)
2 ∂X j ∂X i Ω0 ∂X j ∂X k ∂Xl
2 3
and
1 4∂U
n ∂U
m
∂U
m
∂U
n
k 5
 q_ n qm k k
 k (11) Z
2 ∂Xi ∂X j ∂X i ∂Xj ∂U
m ∂U
p ∂U
l
n
v ∂U v
D
3
mpln 
i i
Djkrs dX (23)
Ω0 ∂X j ∂X k ∂Xr ∂X s
The derivation of the corresponding ROM governing equations
stems from imposing Eq. (3). Note, however, that this equation In carrying out the final assembly of the ROM equations, it should
cannot be satisfied at every point X because the representation of be recognized that the contributions of the terms K
2 mnl qn ql and
Eq. (8) is finite-dimensional (finite number of generalized coordi-
K
2 q q
mln l n can be combined with each other, and similarly for other
nates). To resolve this issue, a Galerkin approach is adopted in which
cubic terms. This process leads to the equivalent expression
the residual of Eq. (3) after imposing Eq. (8) is required to be
orthogonal to the basis functions U
ni
X. This process leads to the _  K
1
fm
q; q 
2 qn ql  K
3 qn ql qp  D
1
mn qn  K mnl mnlp mn q_n
ROM governing equations
 D
2 _ n  D
3
mjn qj q _n
mlpn ql qp q (24)
Mmn q n  fm
q; q
_  Fm (12)
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

where
where
8
Z >
<0 for l < n

2
2
2
Mmn  ρ0 U
m
n
i Ui dX (13) K
2  K mnn  12 K^ mnn  K^ nnm for l  n (25)
Ω0
mnl
>
: K
2  K
2  K^
2  K^
2  K^
2 for l > n
mnl mln mnl nlm lnm

are the components of the mass matrix; Fm are the modal forces
8
resulting from both body forces and surface tractions; and, finally, >
> 0 unless p ≥ l ≥ n
> K
3
> for p  l  n
>
< mnnn
_  D
1
fm
q; q mn q_ n  D
2 _ n  D
3
mjn qj q _ n  K
1
mlpn ql qp q mn qn
3
K mnlp

3
3
3
 K mnll  K mlnl  Kmlln for p  l > n (26)
>
> K
3
3
3
 K
2
3 >
> mpll  K mlpl  K mllp for p > l  n
mnl qn ql  K mnlp qn ql qp (14) >
: 2K
3  2K
3  2K
3
mnlp mpln mlpn for p > l > n
where
recognizing the properties
Z " #
1 ∂U
n ∂U
n ∂U
m
K
1
r s
mn  Cjkrs δij  K
3
3
3
3
i
mnlp  K nmlp  K mnpl  K lpmn
dX (27)
Ω0 2 ∂Xs ∂Xr ∂X k
Z
∂U
n
m
r ∂Ui resulting from Eqs. (5) and (18), and
 Cikrs dX (15)
Ω0 ∂Xs ∂X k 8
< 0
3 for p < l
D
3  Dmlln for p  l (28)
1 ^
2 mlpn
:
3
K
2 ^
2
mnl  K mnl  K nlm (16) Dmlpn  D
3
mpln for p > l
2

where The final resulting ROM governing equations are thus

Z Mmn q n  K
1 
2 qn ql  K
3 qn ql qp  D
1

mn qn  K
∂U
m ∂U
n
p
r ∂U r mnl mnlp mn q_n
K^
2 ^
2
mnp  K mpn 
i
Cijkl dX (17)
Ω0 ∂X j ∂X k ∂X l  D
2 _ n  D
3
mln ql q _ n  Fm
mlpn ql qp q (29)

Z
1 ∂U
m ∂U
n ∂U
l
p
v ∂U v Some properties of the tensors Mmn , K
1
2
3
1
2
mn , K mnl , K mnlp , Dmn , Dmln ,
K
3
mnlp 
i i
Cjkrs dX (18)
2 Ω0 ∂X j ∂X k ∂X r ∂Xs and D
3
mlpn that reflect those of the elasticity and dissipation tensors
have already been stated in this paper, e.g., Eq. (27). In addition, the
in which the symmetry properties of Eq. (5) have been used to following symmetry properties can be recognized:
regroup similar terms. Assuming that similar symmetry properties
hold for the dissipation tensor, i.e., Mmn  Mnm K
1
1
mn  K nm D
1
1
mn  Dnm (30)

Dijkl  Djikl  Dijlk Dijkl  Dklij (19) and

one obtains D
3
3
3
3

mnlp  Dnmlp  Dmnpl  Dlpmn (31)
Z
∂U
n
m
r ∂U i With regard to positive definiteness, a property involving the
D
1
mn  Dikrs dX (20)
Ω0 ∂X s ∂Xk tensors K
1
2
3
mn , K mnl , and K mnlp has been proved in Ref. [25]. For a
similar condition relating to the damping terms, introduce the tangent
D
2 ^
2 ^
2 damping matrix D
T
q of elements
mln  Dmnl  Dnlm (21)

where D
T
1
2
3

mn  Dmn  Dmln ql  Dmlpn ql qp (32)
 WANG ET AL. 4147

Since the tensor D is positive definite, dissipation takes place columns of which are the modes U
j . From the above equations, the
continuously, and thus one expects that D
T
q is a positive definite coefficients K
2 
3 
3
ijl , K ijjl , and K ijll can be estimated by assuming that
matrix. Indeed, note that the linear stiffness coefficients have been obtained first, i.e., based on
the similar projection of the linear finite element stiffness matrix.
q_ m D
T
mn q_ n  q_ m D
1
mn q _ n  q_ m D
2 _ n  q_ m D
3
mln ql q _n
mlpn ql qp q To complete the identification of the reduced-order model, it remains
Z to evaluate the coefficients K
3ijlu for j ≠ l, j ≠ u, and u ≠ l. They can
 E_ ik Dikrs E_ rs dX ≥ 0 (33) be evaluated from the knowledge of K
T
Ω0 iu corresponding to a displace-
ment field that involves both basis functions j and l, i.e., of the form of
where the last equality results from introducing the expressions of ui  qj Ui
j  ql U
l
T
i . Then, K iu is given, for u > l > j, by
Eqs. (20–23) into the left-hand side and performing algebraic manip- h i

T
ulations. K iu  ΨT K^
T Ψ
The positive definiteness of the tangent damping matrix D
T
q h
iu
i
for any vector q can be further translated into the positive definiteness  K iu  K iju qj  K
2

1
2 
3 
3 2 
3 2
ilu ql   K ijlu qj ql  K ijju qj  K illu ql
of the extended matrix
2 3 (41)
D
1 0.5D~
2
DB  4 5 (34) in which no summation on j and l applies, and all terms are known
0.5D~
2T D~
3
except K
3
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

ijlu .
It should be noted that the preceding identification procedure does
where the M × M2 rectangular array D~
2 and M2 × M2 square matrix not identify the parameters K
2
3
mnp or K mnps but rather K 
2
mnp and
D~
3 are obtained through a reshaping of the third- and fourth-order
3
K mnps . From the standpoint of computing the response of the struc-
tensors D
2 and D
3 . These operations are achieved as follows: ture using the reduced-order model, this is, however, fully appropri-
ate because the latter parameters are the ones that appear in Eq. (29).
D~
2
2
mJ  Dmnp with J 
n − 1M  p (35)
D. Nonintrusive Identification of the Damping Parameters: A Parallel
and The final step of the nonlinear reduced-order modeling effort is the
evaluation of the linear and nonlinear damping parameters D
1 mn ,
D~
3
3
IJ  Dmsnp with I 
m − 1M  s and D
2 
3
mln , and Dmlpn . In that regard, a strategy similar to the one carried
J 
n − 1M  p (36) out in Sec. II.C for the identification of the stiffness parameters could
be construed. It would involve imposing velocity fields and determin-
ing the possibly time-dependent forces necessary on a structure
C. Nonintrusive Identification of the Stiffness Parameters exhibiting a Kelvin–Voigt viscoelastic constitutive relation. This
Equations (15–18) provide direct expressions for all stiffness approach was not pursued here because it would require more
parameters of the reduced-order model parameters in terms of the expensive transient solutions and would require an appropriate han-
basis functions U
m
i
X and the geometrical and material properties
dling (i.e., setting up the computations so that they vanish or evalu-
of the structure, e.g. ρ0 , Cijkl , Ω0 , etc. Although these equations can ating them accurately) of the inertia and stiffness terms.
Rather, an alternative approach was followed that relies on the
indeed be used efficiently [26,27], they require the details of the finite
similarities between damping and stiffness parameters. Indeed, compar-
element mesh and formulation. When using a commercial finite
ing Eqs. (15) and (20), it is seen that D
1
1
mn and K mn are identically
element software, it is much more convenient to proceed in an
computed from the dissipation and elasticity tensors Dijkl and Cijkl ,
indirect or nonintrusive manner by relating the parameters to be
determined to nonlinear static finite element solutions. A convenient respectively. Thus, if one was to carry out the identification effort of
approach to performing this task is to impose well-chosen specific Sec. II.C with the numerical values of a fictitious elasticity tensor
fields, determine the nodal forces required to achieve these displace- matching those of the dissipation tensor (note that their units are differ-
ments [21] or the tangent stiffness matrix [16], project these data on ent), one would find that the identified linear stiffness matrix matches the
the basis, and finally match these results with their ROM counter- linear damping one. Since the properties assumed here of the dissipation
parts. The tangent stiffness version [16] of the aforementioned strat- tensor match those of the elasticity tensor, this computation is mean-
egy was used here and is accordingly described in the following. ingful. This correspondence is in fact similar to the one observed in
The first set of imposed displacements is proportional to one Ref. [24] between the damping and stiffness matrices of a linear model.
particular basis function: the jth one, say, i.e., ui  qj U
j i . The
The overall plan for the identification of the damping parameters is
corresponding ROM tangent stiffness matrix can then be written as thus as follows:
(no sum on j) 1) Perform the identification of the stiffness parameters D K
1 mn ,

2 
3
D K mnp , and D K mnps using the elasticity tensor Cijkl numerically

1
K
T
1 
2 
2 
3 
3 
3 2
iu  K iu  K iju  K iuj qj  K ijju  K ijuj  K iujj qj (37)
equal to Dijkl . The parameters D K mn , D K
2
mnp , and D K 
3
mnps will be
referred to as the pseudostiffness parameters because their units are
not consistent with stiffnesses.
Since the elements K
2 
3
ijl and K ijlp are zero unless p ≥ l ≥ j, the 2) Determine the damping parameters D
1
2
mn , Dmln , and D 
3
mlpn from
preceding equation is equivalent to
1
2
3
the pseudostiffness parameters D Kmn , D K mnp , and D K mnps .
Although the preceding approach is straightforward for the linear
K
T T ^
T
iu  Ψ K Ψiu  K
1 
2 
3 2
iu  K iju qj  K ijju qj j<u (38) terms and yields, as suggested earlier in this paper,

K
T T ^
T
iu  Ψ K Ψiu  K
1 
2 
3 2
iu  2K iuu qu  3K iuuu qu ju (39) D
1
1
mn  D K mn (42)

additional discussion is necessary for the quadratic and cubic parameters

K
T T ^
T Ψ  K
1  K
iu  Ψ K

2 
3 2
iuj qj  K iujj qj j>u (40)
D
2 
3
iu iu
mln and Dmlpn . Indeed, the connectivity between stiffness and damp-

where K^
T denotes the finite element tangent stiffness matrix corre- ing parameters is not between D
2 
2 ^
2
mln and D K mnp but rather between Dmln

2
sponding to the imposed displacements and Ψ is the modal matrix the ^
and D K mnp . Specifically, comparing Eqs. (17) and (22), one obtains
 4148 WANG ET AL.

D^
2 ^
2
mnp  D K mnp (43)

Similarly, one find from Eqs. (18) and (23) that

D
3
3
mpln  2D K mpln (44)

Thus, to use the aforementioned correspondences, it is necessary to first

determine D K^
2 
2
3 
3
mnp from D K mnp and, similarly, D K mnps from D K mnps by
“inverting” Eqs. (25) and (26). This operation cannot be achieved
uniquely, but an optimum solution has recently been proposed [28]
and is briefly described in the next section. As a preliminary, however,
note that the positive definiteness of the matrix DB [Eq. (34)] implies
[using Eqs. (42–44)] the same property for the matrix
2 3

1
2
K K
~
D KB 
4D D
5 (45)
~
2T 2D K~
3
DK
Fig. 2 Goland wing structural model.
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

~
2 ~
3
where D K and D K are the reshaping of the third- and fourth-order
tensors D K^
2 and D K
3 as in Eqs. (35) and (36) consistently with the stiffeners. Another set of such bars connects the top and bottom
derivation of [25]. surfaces in addition to shear panels (CQUAD4 Nastran elements)
placed along both the span and the chord at each intersection of the
E. Nonintrusive Identification of the Damping Parameters: Optimum plate elements. The Young’s modulus and Poisson ratio were taken as
Decomposition 1.4976 × 109 psf and 0.3333 for all elements. The geometrical dimen-
It was proposed in Ref. [28] that the decomposition of the identi- sions of the plate, rod, and shear elements are given in Table 1. Finally,
fied stiffness coefficients be carried out so that the following the tip store has a mass of 22.498 slug and zero mass moments of inertia,
holds true: except the one along the span direction that equals 50.3396 slug ⋅ ft2 .
1) The parameters D K^
2
3
ijl and D K ijlp should only be a function of
The store is connected to the tip of the wing by rigid connections (RBE3
the identified coefficients D Kabc and D K
3

2
abcd , with a, b, and c (a, b, c,
Nastran element) at six nodes of the tip.
and d) selected only from the mode numbers i, j, and l (i, j, l, p). A linear modal analysis of the Goland wing (SOL 103 in Nastran)
2) Matrix D KB corresponding to the M selected modes or any yielded the first natural frequencies given in Table 2. It is seen that the
subset of these modes is positive definite, or as close as possible to it. first mode is a spanwise bending mode, whereas the second one is the
Accordingly, the approach of Ref. [28] proceeds in steps, resolving first spanwise torsion mode. Moreover, modes 1–4 are transverse,
first the indeterminacy on all distinct two-mode coefficients (i.e., whereas mode 5 is in-plane; it is a lead–lag mode.
^
2 ^
2 ^
2 ^
2
3
3
3
3 A linear Rayleigh damping was assumed to represent the small-
D K ijj , D K iij , D K jij , D K jii , D K iiij , D K iijj , D K ijij , and D K ijjj ) by
amplitude dissipation in the wing; i.e., the damping matrix of the full
enforcing that they satisfy Eqs. (16) and (25) and lead to a maximum finite element model was expressed as
of the lowest eigenvalue of the matrix D KB corresponding to the two
modes i and j > i.
Next, the indeterminacy on all distinct three-mode coefficients DFE;lin  α0 MFE  β0 K
1
FE (46)
(e.g., D K^
2
ijl ) was similarly resolved by enforcing that they satisfy
Eqs. (16) and (25–27) and lead to a maximum of the lowest eigen- where MFE and K
1FE are the finite element mass and linear stiffness
value of the matrix D KB corresponding to the three distinct mode matrices. The coefficients α0 and β0 were selected as α0  2.565 ×
numbers l > j > i. 10−1 and β0  1.338 × 10−4 s, which lead to damping ratios of
Finally, the indeterminacy on all distinct four-mode coefficients 1.279%, 0.797%, 0.608%, 0.644%, and 0.654% for the first five

3 transverse modes.
D ijlp was again resolved by enforcing that they satisfy Eqs. (16) and
K
(25–27) and lead to a maximum of the lowest eigenvalue of the matrix
B. Nonlinear Structural Model: Basis Functions
D KB corresponding to the four distinct mode numbers i, j, l, and p.
Note that the aforementioned optimum decomposition does not A key aspect of constructing nonlinear reduced-order models of
necessarily guarantee that the final matrices D KB and DB are positive structures is the selection of the basis. Since a linear behavior is
definite as theoretically expected. The lack of this property may not expected for small enough motions, this basis should include all of
be detrimental, i.e., when negative damping is along deflections that the linear modes that would be included in a linear analysis. This is,
are not observed on the structure but ensuring positive definiteness is however, not sufficient, as demonstrated in past investigations (e.g.,
desirable and can be achieved as a final step following the iterative see Ref. [12]), because these modes typically do not accurately
process described in Ref. [28]. capture the “membrane” (in-plane) motions that occur in large

Table 1 Dimensions of finite elements

III. Goland Wing and Its Nonlinear ROM
A. Linear Structural Model Elements Length, ft Width, ft Thickness, ft
The Goland wing will be used in the next section to assess the Plates 2 2 0.0155
effects of damping nonlinearity on post-LCO behavior. The Goland Shear spanwise (midwing) 2 0.33334 0.0889
wing is a standard aeroelastic model; it is a rectangular wing of Shear spanwise (edges) 2 0.33334 6 × 10−4
constant thickness, an aspect ratio equal to five, and exhibits a tip Shear chordwise 2 0.33334 0.0347
store as shown in Fig. 2 (see Ref. [29] for full definition). Rods Length, ft Area, ft2
The Nastran finite element model is composed of four-node square Spanwise (midwing) 2 0.1496
plate elements (CQUAD4 Nastran elements): two rows of 10 such Spanwise (edges) 2 0.0416
elements on the top of the wing and a similar layout on its bottom. Chordwise 2 0.0422
These elements are complemented by a grid of bar elements (CROD Thicknesswise 0.33334 8 × 10−4
Nastran elements) aligned both spanwise and chordwise to model the
 WANG ET AL. 4149

Table 2 Natural frequencies aerodynamic forces. It was obtained here through a rational function
of the Goland wing approximation of the generalized aerodynamic forces in the reduced
Mode no. Natural frequency, Hz frequency domain; see Refs. [30,31]. Accordingly, the aerodynamic
forces on the wing could then be rewritten directly in terms of the
1 1.690
2 3.051 structural generalized coordinates q
t as
3 9.172
4 10.834 q∞ c2 q c
5 11.258 FAero
t  − A2 q − ∞ A1 q_ − q∞ A0 q − q∞ Da ξ (49)
V 2∞ V∞
6 16.260
7 22.845
8 26.318
where the time-dependent variables ξ are referred to as the aerody-
9 29.183 namic lag modes and satisfy the first-order differential equations

V
ξ_ − Ea q_ − ∞ Ra ξ  0 (50)
c
displacements. This observation has led to the introduction of the
“dual” modes to complement the basis [12,13,16]. These modes are In the preceding equations, A0 , A1 , and A2 are real m × m
constructed to capture the displacements not represented by the linear matrices. Furthermore, na is the number of aerodynamic modes
modes when these modes are excited. To this end, a series of loadings assumed and the real matrices Da , Ra , and Ea are of respective
of the full finite element model is first determined that would induce dimensions m × na, na × na, and na × m. Combining Eqs. (49) and
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

in the linear case displacement fields that are exactly linear combi- (50) with the structural equations of motion provides a complete
nations of one or two linear modes. The nodal forces, denoted as F
m
FE modeling of the aeroelastic system.
for loading case m, are of the forms The selection of the number of lag modes necessary was deter-
mined by comparing the flutter conditions predicted by the original
F
m
m
1
j
FE  αj KFE U (47) frequency-domain and time-domain [Eqs. (49) and (50)] aeroelastic
models at the specified Mach number of M∞  0.70.
and This comparison was carried out with the first m  9 linear modes
and gave close agreement with na  13 aerodynamic lag modes,
F
m
m
1
j
FE  αj KFE U  α
m
1
l
l KFE U (48) leading to a flutter altitude of 11,275 ft and a frequency of 1.975 Hz.
As could be expected, it was found that the flutter mode involved
primarily the first bending and first torsion linear modes and, in fact,
to induce displacements along mode j (U
j ) only [Eq. (47)] or modes mostly the first bending, as can be confirmed by the closeness of the
j and l only [Eq. (48)] in the linear case. flutter frequency with the natural frequency of the first mode. At
When the forces of Eqs. (47) and (48) are applied statically to the altitudes below the flutter one, the aeroelastic response grows in time
nonlinear finite element model, the computed displacements are no in an oscillatory manner, with frequencies shown in Table 3.
longer exactly linear combinations of the linear modes; and the out-
of-basis components, or residuals, of these displacement fields pro-
D. Dissipation Modeling
vide the raw data for the extension of the basis. More specifically, the
dual modes are obtained through a proper orthogonal decomposition To complete the characterization of the Goland wing, it remains to
of the ensemble of residuals. specify a dissipation model and, in particular, a dissipation tensor that
In the present effort, 20 values (half positive and half negative) of will be used as a basis for the nonlinear structural damping model.
the scaling factors α
m
j were used for each combination of modes. Before addressing the latter, it should be recognized that dissipation
The values were chosen so that the displacement data induce tip does not only originate from viscoelastic effects; other phenomena
transverse displacements of the wing ranging from about 0.1% to (e.g., friction as considered in Refs. [8–10]) may be present. To
15% of span, and the data are distributed in this range as uniformly as account globally and (in the absence of a well-defined mechanism)
possible. linearly for such effects, a damping term proportional to the modal
The combinations of modes (i.e., the values of j and l) were chosen mass matrix was first added to Eq. (14) as
to capture a broad range of expected motions. Since the flutter
involves primarily modes 1 and 2, the combinations focused on those _  αMmn q_ n  D
1
fm
q; q mn q_ n  D
2 _ n  D
3
mjn qj q _n
mlpn ql qp q
two modes and possible contributions of the three next ones. That is,
the combinations considered were
j; l 
1; 1, (1, 2), (1, 3), (1, 4),  K
1
2
3
mn qn  K mnl qn ql  K mnlp qn ql qp (51)
(1, 5), (2, 2), (2, 3), (2, 4), and (2, 5). The proper orthogonal
decomposition approach led to 12 dominant eigenvectors selected Consider next the dissipation tensor modeling. Since it exhibits the
as dual modes and added to the nine linear modes to form the basis. same properties as the elasticity tensor (see Eqs. (5–7)), one can
Note that these dual modes only exhibit in-plane x and y displace- define it to be isotropic, orthotropic, etc.; see Ref. [22] for definitions.
ments. The 9 linear and 12 dual modes ROM will be referred as ROM In the present investigation, it is simply assumed that this tensor has
9L12D in the sequel. the same form as the elasticity tensor but scaled. That is,

C. Aeroelastic Model Dijkl  γCijkl (52)

The modeling of the aerodynamic forces acting on the Goland
wing during constant-speed level flight was achieved in the modal Table 3 Postflutter
domain (i.e., with the structural response expressed as in Eq. (8)) aeroelastic response
using the ZONATechnology code ZAERO. The first step of the linear frequencies at different
modeling process assumed a harmonic motion of the structure pro- altitudes
portional to each of the m structural modes in turn and determined the
ensuing harmonic pressure field on the structure. This field was then Altitude, ft Frequency, Hz
integrated to yield the matrix of generalized aerodynamic forces. 4,000 1.851
This representation of the aerodynamic forces in the frequency 6,000 1.881
8,000 1.911
domain is particularly convenient for the steady-state and flutter 10,000 1.945
analyses of linear structures; but, nonlinear structural models (such 11,000 1.962
as those of Sec. II) require a time-domain representation of the
 4150 WANG ET AL.

where γ is a positive constant. It can then be shown from Eqs. (20–23)

that all coefficients D
1
2
3
mn , Dmjn , and Dmlpn are also proportional to γ.
Then, denoting

D
1 
1
mn  γ D mn ; D
2 
2
mjn  γ Dmjn ; and D
3 
3
mlpn  γ Dmlpn (53)

where D
1 
2 
3

mn , Dmjn , and Dmlpn are independent of the choice of γ, i.e.,
correspond to the case γ  1 in Eq. (52). They are thus characterized
only by the finite element model and the basis functions. Equa-
tion (51) then becomes

_  αMmn q_ n  γD
1

fm
q; q 
2 
3
mn  Dmjn qj  Dmlpn ql qp q
_n
 K
1
2
3
mn qn  K mnl qn ql  K mnlp qn ql qp (54)
Fig. 4 Spanwise Ty static displacements at node A under a uniform
This damping model appears effectively as a “nonlinear Rayleigh” pressure, ROM, and Nastran nonlinear solutions for varying pressure
damping model defined by its two parameters α and γ. levels.
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

E. Stiffness and Damping Coefficients’ Estimation and ROM Valida-

tion
The stiffness and pseudostiffness coefficients were identified by
the imposed displacement-tangent stiffness matrix approach of
Sec. II.C [16]. The pseudostiffness coefficients were then trans-
formed to their damping counterparts using the optimum decompo-
sition of Sec. II.E, with the extended matrix DB rendered positive
definite with the algorithm of [28]. This step concluded the construc-
tion of the nonlinear reduced-order model.
A validation of this ROM was first carried out by comparing
nonlinear static responses predicted from it with their counterparts
from the finite element model. Of particular interest in this com-
parison were loadings that would trigger the first bending and first
torsion modes of the wing, which are the primary components of the
flutter mode and the ensuing LCO. The ROM deflections were
computed by solving static equations (12) and (14) by Newton–
Raphson.
This comparison is shown for point A (see Fig. 2) in Figs 3–5 for a Fig. 5 Lead–lag Tx static displacements at node A under a uniform
pressure, ROM, and Nastran nonlinear solutions for varying pressure
uniform pressure on the wing and in Figs 6–8 for a pressure on it levels.
which is uniform positive ahead (toward leading edge) of the midline
of the wing and uniform negative on the other side (toward the trailing
edge). This pressure will be referred to as “step uniform.” The
comparison includes pressures spanning the linear range and extend-
ing well into nonlinear behavior.
An excellent matching of the transverse and spanwise displace-
ments (Figs. 3 and 4) is observed for the uniform loading, even for
large deflections (25% of span). The corresponding lead–lag dis-
placement is poorly captured, but it is very small: about 20 times
smaller than the spanwise.

Fig. 6 Transverse Tz static displacements at node A under a step uni-

form pressure, ROM, and Nastran nonlinear solutions for varying pres-
sure levels.

The torsional response is much smaller: in both the step uniform

loading but also the LCO, owing to the notably higher stiffness in
torsion. Nevertheless, the finite element predictions are well matched
by their ROM counterparts for both the transverse and lead–lag
Fig. 3 Transverse Tz static displacements at node A under a uniform displacements, which are the largest two. The ROM spanwise pre-
pressure, ROM, and Nastran nonlinear solutions for varying pressure diction, which is extremely small, appropriately matches its finite
levels. element counterpart.
 WANG ET AL. 4151
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

Fig. 7 Spanwise Ty static displacements at node A under a step uniform Fig. 9 Plot of damping parameter α versus γ.
pressure, ROM, and Nastran nonlinear solutions for varying pressure
levels.

Fig. 10 Amplitude of LCO as a function of altitude for various

damping coefficients γ. Curve NL stiff corresponds to stiffness nonlinear-
Fig. 8 Lead–lag Tx static displacements at node A under a step uniform
ity only, whereas curve γ
10−3 only was obtained without stiffness
pressure, ROM, and Nastran nonlinear solutions for varying pressure
nonlinearity.
levels.

0.001 s until convergence on the amplitude of response, which

IV. LCO Results occurred after 20,000–40,000 time steps; see Fig. 11 for typical time
The focus of the Goland wing LCO investigation was on 1) assess- histories. Note the very harmonic behavior of the LCO response on
ing the potential of nonlinear damping to induce LCO, 2) analyzing Fig. 11b.
this behavior, as well as 3) studying the interplay between the non- First and foremost, the results of Fig. 10 demonstrate that LCO
linear stiffness effects and their damping counterparts. In this regard, occurs, for this Goland wing, through either or both nonlinearity (NL)
it must be recognized that varying the factor γ while keeping all other in damping and in stiffness. When the damping parameter is small
parameters constant will affect the flutter altitude, decreasing it as γ is (e.g., for γ  10−5 s), its effects are small and the LCO is effectively
increased. As expected, it was observed in such conditions that dictated by the stiffness nonlinearity; see Fig. 10, where the curve
increasing γ first decreases the amplitude of postflutter LCO; and corresponding to γ  10−5 s is coincident with the one with γ  0 s
then it eliminates flutter altogether at that altitude as it corresponds to (“NL stiff only”). As γ increases, the amplitude of LCO decreases
increasing the overall structural damping. slowly, with this change most visible at higher altitudes where the
To obtain a more rich perspective on the effects of γ, its variations stiffness nonlinearity is weaker, given that the LCO amplitudes are
were accompanied by opposite changes in the parameter α so that the smaller. At low altitudes (4000 ft), the change in LCO amplitude
flutter altitude remained constant; see Fig. 9 for the plot of the remains nearly zero when γ increases from γ  10−5 s to γ  10−4 s,
corresponding parameter α versus γ. Then, the observed effects of even though it is about 10% at higher altitudes. For large enough γ
changes in γ truly depict the influence of nonlinearity vs linearity in ( 5 × 10−4 s here), the damping nonlinearity becomes dominant
the structural damping. and sharp decreases in LCO amplitude occur throughout the altitude
The results of this investigation are summarized in Fig. 10. They range. In fact, the effect of stiffness nonlinearity may then become
were obtained by a time integration of the equations of motion very small; compare the curves “γ  10−3 s” and “γ  10−3 s only,”
[Eqs. (12), (49), (50), and (54)] within MATLAB using a New- which correspond to the same value of γ but with stiffness non-
mark-β algorithm combined with a fixed-point iterative strategy to linearity on and off, respectively.
solve the nonlinear algebraic equations at each time step. This From the time traces of Fig. 11b, the LCO frequency could be
process started from near-zero initial conditions with a time step of identified; and it is shown in Fig. 12 as a function of altitude for
 4152 WANG ET AL.

a)
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

b) Fig. 13 Ratio of LCO generalized coordinates of dominant modes:

Fig. 11 Time history of transverse displacement of point A at altitude of a) mode 1/mode 2, and b) mode 1/mode 4. Mode 1 is bending mode, mode
10,000 ft with γ
10−3 s obtained with both nonlinearities present [NL 2 is torsional mode, and mode 4 is bending–torsional coupled mode.
damping (damp.) and stiffness (stiff.)], stiffness nonlinearity only (NL
stiffness only), and damping nonlinearity only (NL damping only): a) full
time history, and b) zoomed-in converged LCO. which the LCO frequency increases in comparison to its value
for γ  5 × 10−4 s.
The changes in frequency observed in Fig. 12 are actually rooted in
the relative importance of the bending and torsion modes in the LCO.
To clarify, shown in Fig. 13a is the ratio of the amplitudes of LCOs of
the generalized coordinates of modes 1 (bending) and 2 (torsion). As
the altitude decreases and γ increases, this ratio typically increases,
indicating that there is a relative increase in the bending motions as
compared to the torsional ones. Since the bending motions are softer
than the torsional ones, this increase in bending is reflected in the
LCO by a lowering of the frequency, as seen in Fig. 12 in most cases.
The only conditions for which the results of Fig. 13a do not explain
the trends of Fig. 12 are for γ increasing from 5 × 10−4 s to 10−3 s and
altitudes around 4000 ft. There, the increase in γ leads to an increase
in the first mode (see Fig. 13a) but the LCO frequency is also
increased and not decreased (see Fig. 12). This behavior is not due
to modes 1 and 2 but actually results from an increased participation
of higher-frequency (stiffer) mode 4;, see Fig. 13b. Thus, the behavior
of the LCO frequency actually reflects the changes in participation to
the LCO response of the various modes.
Fig. 12 Frequency of LCO as a function of altitude for various damping
coefficients γ. NL stiff curve corresponds to stiffness nonlinearity
only, whereas γ
10−3 s only curve was obtained without stiffness non- V. Conclusions
linearity. The focus of the present investigation was on the formulation,
development, and a preliminary assessment of a nonlinear structural
damping model that is consistent with a linear viscoelastic material
different values of γ. For most cases, it is found that the LCO undergoing large deformations. The model was developed in a
frequency decreases with decreasing altitude and with increasing reduced-order modeling format assuming a Kelvin–Voigt constitu-
value of γ. The exception to this behavior occurs for γ  10−3 s tive relation expressed in the undeformed configuration and led to
(the largest value of this coefficient used in the investigation), for governing equations for the generalized coordinates in the form of
 WANG ET AL. 4153

Van der Pol–Duffing equations. These equations, which generalize Amplitude,” International Forum on Aeroelasticity and Structural
similar reduced-order models obtained for elastic structures, are Dynamics 2005, Vol. 1, 2005, Paper IF-022.
parametric, i.e., involve a series of coefficients that are functions of [10] Kingsbury, D. W., Agelastos, A. M., Dietz, G., Mignolet, M. P., Liu,
the material properties, the geometry of the structure, and the basis D. D., and Schewe, G., “Limit Cycle Oscillations of Aeroelastic Systems
with Internal Friction in the Transonic Domain–Experimental Results,”
functions used to represent the response. An existing identification 46th Structures, Structural Dynamics, and Materials Conference,
strategy of the stiffness parameters in such reduced-order models was AIAA Paper 2005-1914, 2005.
next modified to permit the determination of the linear and nonlinear https://doi.org/10.2514/6.2005-1914
damping coefficients. [11] Song, P., Wang, X. Q., Mignolet, M. P., and Chen, P. C., “A Reduced
The application of particular interest here for the assessment of the Order Model-Based Nonlinear Damping Model: Formulation and
nonlinear damping model is the postflutter response of wings and the Application to Post Flutter Aeroelastic Behavior,” AIAA Science and
occurrence of limit-cycle oscillations. The Goland wing was selected Technology Forum and Exposition (SciTech 2016), AIAA Paper 2016-
as the structure of interest for this validation because of the public 1795, Jan. 2016.
availability of its structural finite element model and of a commer- https://doi.org/10.2514/6.2016-1795
[12] Mignolet, M. P., Przekop, A., Rizzi, S. A., and Spottswood, S. M., “A
cially developed time-domain model of its aerodynamics.
Review of Indirect/Non-Intrusive Reduced Order Modeling of Non-
The identification of the nonlinear damping coefficients was per- linear Geometric Structures,” Journal of Sound and Vibration, Vol. 332,
formed from the Goland wing full finite element model and permitted No. 10, 2013, pp. 2437–2460.
the assessment of the nonlinear structural damping model to induce https://doi.org/10.1016/j.jsv.2012.10.017
LCO from flutter. It was found that the nonlinear damping can indeed [13] Kim, K., Radu, A. G., Wang, X. Q., and Mignolet, M. P., “Nonlinear
lead to LCOs of amplitudes that can range from below 1% of span to Reduced Order Modeling of Isotropic and Functionally Graded Plates,”
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804

significant fractions of span. These results suggested the possibility International Journal of Non-Linear Mechanics, Vol. 49, March 2013,
to calibrate the nonlinear damping model from experimentally pp. 100–110.
observed LCO amplitudes. Finally, the frequency of the LCO https://doi.org/10.1016/j.ijnonlinmec.2012.07.008
responses was also analyzed and was observed to vary relatively [14] Wang, X. Q., Khanna, V., Kim, K., and Mignolet, M. P., “Nonlinear
Reduced Order Modeling of Flat Cantilevered Structures: Challenges
little unless at very low altitudes, being primarily driven by the and Remedies,” ASCE Journal of Aerospace Engineering (in press).
relative magnitudes of the bending and torsional LCO motions. [15] Perez, R., Wang, X. Q., and Mignolet, M. P., “Nonlinear Reduced Order
Models for Thermoelastodynamic Response of Isotropic and FGM
Panels,” AIAA Journal, Vol. 49, No. 3, 2011, pp. 630–641.
Acknowledgments https://doi.org/10.2514/1.J050684
[16] Perez, R. A., Wang, X. Q., and Mignolet, M. P., “Non-Intrusive Struc-
This work was sponsored by the U.S. Air Force Office of Scientific tural Dynamic Reduced Order Modeling for Large Deformations:
Research under the Small Business Technology Transfer Research Enhancements for Complex Structures,” Journal of Computational
program (contract number: FA9550-15-C-0006). The contract mon- and Nonlinear Dynamics, Vol. 9, No. 3, 2014, Paper 031008.
itor was Michael Kendra. The authors would like to thank Charles https://doi.org/10.1115/1.4026155
Denegri and Vin Sharma of the SEEK Eagle Office for their valuable [17] Wang, X. Q., Perez, R., Mignolet, M. P., Capillon, R., and Soize, C.,
suggestions throughout the period of this project. “Nonlinear Reduced Order Modeling of Complex Wing Models,” 54th
Structures, Structural Dynamics and Materials Conference, AIAA
Paper 2013-1520, April 2013.
https://doi.org/10.2514/6.2013-1520
References [18] Chen, P. C., Zhang, Z., Zhou, Z., Wang, X. Q., and Mignolet, M. P.,
[1] Chen, P. C., Sarhaddi, D., and Liu, D. D., “ Limit Cycle Oscillation “Limit Cycle Oscillation Prediction for Aircraft with External Stores,”
Studies of a Fighter with External Stores,” 39th AIAA/ASME/ASCE/ International Forum on Aeroelasticity and Structural Dynamics
AHS/ASC Structures, Structural Dynamics, and Materials Conference (IFASD 2019), June 2019.
and Exhibit, AIAA Paper 1998-1727, April 1998, pp. 259–266. [19] Goland, M., and Luke, Y., “The Flutter of a Uniform Wing with Tip
https://doi.org/10.2514/6.1998-1727 Weights,” Journal of Applied Mechanics, Vol. 15, No. 1, 1948, pp. 13–20.
[2] Denegri, C. M., Jr., “Limit Cycle Oscillation Flight Test Results of a https://doi.org/10.1115/1.4009753
Fighter with External Stores,” Journal of Aircraft. Vol. 37, No. 5, 2000, [20] Eastep, F. E., Khot, N. S., Beran, P. S., Zweber, J. V., and Snyder, R. D.,
pp. 761–769. “Investigation of Shock-Induced LCO of a Wing/Store Configuration
https://doi.org/10.2514/2.2696 using the Transonic Small Disturbance Method,” Proceedings of the
[3] Farhat, C., Geuzaine, P., and Brown, G., “Application of a Three-Field 23rd Congress of the International Council on the Aeronautical Scien-
Nonlinear Fluid-Structure Formulation to the Prediction of the Aero- ces (ICAS), Sept. 2002, pp. 445.1–445.10.
elastic Parameters of an F-16 Fighter,” Computers and Fluids. Vol. 32, [21] Muravyov, A. A., and Rizzi, S. A., “Determination of Nonlinear
No. 1, 2003, pp. 3–29. Stiffness with Application to Random Vibration of Geometrically Non-
https://doi.org/10.1016/S0045-7930(01)00104-9 linear Structures,” Computers and Structures, Vol. 81, No. 15, 2003,
[4] Norton, W. J., “Limit Cycle Oscillation and Flight Flutter Testing,” pp. 1513–1523.
Proceedings of the 21st Annual Symposium, Soc. of Flight Test Engi- https://doi.org/10.1016/S0045-7949(03)00145-7
neers, Lancaster, CA, 1990, pp. 3.4-3.4-12. [22] Fung, Y. C., and Tong, P., Classical and Computational Solid Mechan-
[5] Parananta, B. B., Kok, J. C., Spekreijse, S. P., Hounjet, M. H. L., and ics, World Scientific, River Edge, NJ, 2001.
Meijer, J. J., “Simulation of Limit Cycle Oscillation of Fighter Aircraft at https://doi.org/10.1142/4134
Moderate Angle of Attack,” National Aerospace Lab./NLR TP 2003- [23] Bonet, J., and Wood, R. D., Nonlinear Continuum Mechanics for Finite
526, Amsterdam, 2003, pp. 1–38, https://reports.nlr.nl/xmlui/bitstream/ Element Analysis, 2nd ed., Cambridge Univ. Press, Cambridge,
handle/10921/644/TP-2003-526.pdf?sequence=1&isAllowed=y England, U.K., 2010.
[retrieved 2021]. https://doi.org/10.1017/CBO9780511755446
[6] Pasiliao, C. L., “Characterization of Aero-Structural Interaction Flow- [24] Kuether, R. J., Troyer, K. L., and Brake, M. R., “Time Domain Model
Field Physics,” Aerospace Flutter and Dynamics Council Fall Meeting, Reduction of Linear Viscoelastic Finite Element Models,” Sandia National
Oct. 2012. Lab. Rept. SAND2016-5154C, Albuquerque, NM, 2016, https://www.
[7] Sharma, V. K., and Denegri, C. M., “Time Domain Aeroelastic Solution osti.gov/servlets/purl/1368881 [retrieved 19 June 2020].
Using Exact Aerodynamic Influence Coefficients and Nonlinear Damp- [25] Mignolet, M. P., and Soize, C., “Stochastic Reduced Order Models for
ing,” Proceedings of the International Forum on Aeroelasticity and Uncertain Geometrically Nonlinear Dynamical Systems,” Computer
Structural Dynamics, Paper IFASD-2013-29D: 1-11, 2013. Methods in Applied Mechanics and Engineering, Vol. 197, Nos. 45–
[8] Choi, G. G., Agelastos, A. M., Mignolet, M. P., and Liu, D. D., “Effects 48, 2008, pp. 3951–3963.
of Internal Friction on the Dynamic Behavior of Aeroelastic Systems,” https://doi.org/10.1016/j.cma.2008.03.032
45th Structures, Structural Dynamics, and Materials Conference, [26] Capiez-Lernout, E., Soize, C., and Mignolet, M. P., “Computational
AIAA Paper 2004-1591, April 2004. Stochastic Statics of an Uncertain Curved Structure with Geometrical
https://doi.org/10.2514/6.2004-1591 Nonlinearity in Three-Dimensional Elasticity,” Computational Mechan-
[9] Choi, G. G., Agelastos, A. M., Mignolet, M. P., and Liu, D. D., “On the ics, Vol. 49, No. 1, 2012, pp. 87–97.
Impact of Internal Friction on Flutter Onset and Limit Cycle Oscillations https://doi.org/10.1007/s00466-011-0629-y
 4154 WANG ET AL.

[27] Capiez-Lernout, E., Soize, C., and Mignolet, M. P., “Post-Buckling Journal of Aircraft, Vol. 41, No. 6, 2004, pp. 1315–1326.
Nonlinear Static and Dynamical Analyses of Uncertain Cylindri- https://doi.org/10.2514/1.404
cal Shells and Experimental Validation,” Computer Methods in Applied [30] Karpel, M., “Design for Active Flutter Suppression and Gust Alleviation
Mechanics and Engineering, Vol. 271, April 2014, pp. 210–230. Using State-Space Aeroelastic Modeling,” Journal of Aircraft, Vol. 19,
https://doi.org/10.1016/j.cma.2013.12.011 No. 3, 1982, pp. 221–227.
[28] Wang, X. Q., Mignolet, M. P., and Soize, C., “Structural Uncertainty https://doi.org/10.2514/3.57379
Modeling for Nonlinear Geometric Response Using Nonintrusive [31] Karpel, M., “Extension to the Minimum-State Aeroelastic Modeling
Reduced Order Models,” Probabilistic Engineering Mechanics, Vol. 60, Method,” AIAA Journal, Vol. 29, No. 11, 1991, pp. 2007–2009.
April 2020, Paper 103033. https://doi.org/10.2514/3.10832
https://doi.org/10.1016/j.probengmech.2020.103033
[29] Beran, P. S., Khot, N. S., Eastep, F. E., Snyder, R. D., and Zweber, J. V., C. Pettit
“Numerical Analysis of Store-Induced Limit-Cycle Oscillation,” Associate Editor
Downloaded by Turkish Aerospace Industries on December 9, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J059804