Journal of Fluids and Structures 74 (2017) 413–426

Contents lists available at ScienceDirect

Journal of Fluids and Structures

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

Nonlinear aerodynamic and energy input properties of a

twin-box girder bridge deck section
Zhitian Zhang a, *, Xianxiong Zhang a , Yongxin Yang b , Yaojun Ge b
a
Wind Engineering Research Center, Hunan University, Changsha 410082, China
b
State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, ShangHai 200092, China

highlights

• Motion amplitude-dependent nonlinear aeroelastic properties of a twin-box girder is investigated.

• A critical amplitude is found to cause to flow pattern fully detached.
• Hysteresis loops of dynamic load coefficients are presented.
• Energy trapping properties are obtained in terms of dimensionless coefficients.
• Energy properties show that flutter instability necessitate motion coupling between vertical and torsional motions, and that the
symmetricity of the two motions has be identical.

article info a b s t r a c t
Article history: By means of computational fluid dynamics (CFD), the nonlinear aeroelastic properties of
Received 20 December 2016 a bridge deck configuration is investigated in this study in terms of amplitude-dependent
Received in revised form 23 May 2017 flutter derivatives and indicial functions. The results are partially compared with experi-
Accepted 22 June 2017
mental results. It shows that the concerned properties exhibit significant dependence on
Available online 20 July 2017
the motion amplitudes. Moreover, based on flutter derivatives, the nonlinear aerodynamic
properties can be divided into two groups: the group with torsional amplitudes less or
Keywords:
Bridge equal than 10◦ , and the one with amplitudes larger than 10◦ . Flow patterns around the
Non-linear section of the two groups differ substantially; one group remains an overall streamlined
Aerodynamic pattern with locally distributed vortices and detached flow, while the other shows fully
Indicial function detached flow with large vortices emerging and developing drastically. Dynamic load co-
Hysteresis loop efficients indicate that, as the motion amplitude increases, the smoothness of the hysteresis
Energy property loops decreases, suggesting irregular fluctuations of the loads resulted from signature
turbulence, which becomes progressively prominent. Energy trapping properties derived
from indicial functions are expressed in terms of dimensionless coefficients, of which the
results indicate there is no possibility of single-DOF flutter, and coupling between vertical
and torsional motions is necessary for flutter instability. Moreover, by the analysis of the
phase angles involved in coupling, it is indicated that symmetricity of vertical motion has
to be consistent with that of torsional motion in the event of a coupled flutter.
© 2017 Elsevier Ltd. All rights reserved.

* Corresponding author. Fax: +86 731 88823923.

E-mail address: [email protected] (Z. Zhang).

http://dx.doi.org/10.1016/j.jfluidstructs.2017.06.016
0889-9746/© 2017 Elsevier Ltd. All rights reserved.
414 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

1. Introduction

Aerodynamic instability of long-span bridges has been investigated extensively in recent decades. The attention of
most of the published literatures in this regard, however, has been paid to linear problems, and the aeroelastic stability of
bridge spans is usually judged by flutter thresholds in sorts of design codes. Achievements in aeronautical engineering show
repeatedly that a wind speed beyond the flutter threshold does not always mean catastrophic collapse of the structure. The
post-flutter behaviors have been shown to depend on a combination of material, geometric and aeroelastic nonlinearities,
and to be significant to a structure’s robustness design. In some situations, as indicated in aeronautical engineering, it is the
geometric nonlinearity alone that dictates the evolution of the limit cycle oscillation (LCO) at post-flutter stages. A number of
researchers have paid exclusive attention to this kind of contribution ( Tang et al., 1999; Attar et al., 2003; Attar and Dowell,
2005; Shams et al., 2008; Eskandary et al., 2012; etc.).
Many theoretical models have been developed for airfoils to describe nonlinear aerodynamic properties. The models
generally involve a semi-empirical model. Tran and Petot (1981) brought forward ONERA model, which has obtained
extensive applications to describe dynamic stall of helicopter blades, wind turbines, etc. (Tang and Dowell, 1993, 2004;
Sarkar and Bijl, 2008; Stanford and Beran, 2013); Leishman and Beddoes (1986) developed another semi-empirical model
for description of dynamic stall, where a fairly elaborate representation of the nonstationary attached flow depending on
the Mach number is included. The model of Leishman and Beddoes also has been widely used ( Leishman, 1988; Galvanetto
et al., 2008; etc.); More recently, Larsen et al. (2007) put forward another semi-empirical dynamic stall model for wind
turbine airfoils. In general, semi-empirical methods involve a dynamic wind angle of attack and a backbone function. A
semi-empirical model should be able to reproduce the static values when velocity terms vanish; that means, they degenerate
to corresponding backbone functions.
The above mentioned semi-empirical models were developed for airfoils or helicopter rotors; therefore, they are
inapplicable to bluff bridge sections where far less regularities can be found for the formation, development, and detachment
of the vortices. In the context of bluff bridge deck sections, Diana et al. (2008, 2010) developed a model that could
account for nonlinear effects due to frequency and amplitude. The proposed nonlinear model is a polynomial function of
a dynamic angle of attack. Based on artificial neural network, Wu and Kareem (2011) proposed a nonparametric model
to capture the hysteretic nonlinear behavior of aerodynamic systems. Later, Wu and Kareem (2015) tried another way
based on Volterra theory to describe linear and nonlinear aerodynamic effects. The works of Wu and Kareem (2011, 2015)
address phenomenological models which leave aside explicit physical meanings. Another phenomenological model, based
on nonlinear flutter derivatives, has been exercised more recently by Gao and Zhu (2015) and Ying et al. (2016). The
nonlinear flutter derivatives in this method depend on higher odd-order terms of the motions, and the identification of
these derivatives are based directly on resulted time histories.
Most of the above mentioned nonlinear models concern two issues. First, they address hysteresis loops of load coefficients,
not development of these loops corresponding to different motion amplitudes; second, they address analytical descriptions
of these hysteresis loops, which involves identification of a number of model parameters. In post-flutter simulations,
however, the primary issue is evolution of the energy trapping/dissipating properties with the motion amplitudes, instead
of exact shapes of hysteresis loops. On the other hand, an analytical description of nonlinearities is not always necessary
for computations, especially when the description comes from a group of discrete experimental data, since it does not
have any advantage over the original data in terms of accuracy. Based on this ideology, the present work deals with a
piece-wise linearized method to express straightforwardly the amplitude-dependent energy properties, which does not
pay attention to exact shapes of complicated hysteresis loops. It is believed in this work that energy trapping/dissipating
properties are sufficient to describe precisely development of structural motions. The nonlinear aerodynamic properties of
a deck configuration employed in the Xihoumen suspension bridge is investigated, in terms of amplitude-dependent flutter
derivatives, hysteresis loops, indicial functions (IFs) and energy trapping properties, respectively.

2. Description of nonlinear aeroelasticity

2.1. Energy properties related to an indicial function

Motion-related aerodynamic loads developed on bluff bridge decks have been denoted conventionally in terms of flutter
derivatives, obtained from either wind tunnel tests or from CFD simulations. The self-excited aerodynamic lift and torque
per unit length under sinusoidal motions are given as (Scanlan, 1993, 2000; Katsuchi et al., 1999):

∗ Bα̇
[ ]
1 ḣ ∗h
Lse = ρ U B KH1
2 ∗
+ KH2 + K H3 α + K H4
2 ∗ 2
, (1)
2 U U B

∗ Bα̇
[ ]
1 ḣ h
Mse = ρU B 2 2
KA1 ∗
+ KA2 + K 2 A∗3 α+ K 2 A∗4 , (2)
2 U U B
where ρ = air density; U = wind speed; B = reference width; K = Bω/U is the reduced frequency; ω = circular natural
frequency; Hi∗ (i = 1–4), A∗i (i = 1–4) are flutter derivatives, h and α are the vertical and torsional displacements; ḣ, α̇ are
the derivatives of h and α with respect to time t, respectively.
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 415

18 flutter derivatives in total can be involved when the drag and lateral motions p and ṗ are considered ( Katsuchi et al.,
1999; Chen and Kareem, 2008; etc.). For sake of simplicity, we leave aside in this paper the self-excited drag and lateral-
motion-induced aeroelastic effects.
Eqs. (1) to (2) are self-excited loads expressed in a time–frequency mixed domain. They are not able to be employed
directly in a time-domain analysis. However, the flutter derivatives can be used to identify a group of IFs, which are applicable
in time-domain. The concept of indicial function (IF) derives from the classic airfoil theory in description of the transient
evolution of the aerodynamic lift due to an abrupt change of the status (velocity or wind angle of attack), as
1
L (s) = ρ U 2 BCL′ α0 ϕ (s) , (3)
2
where ϕ (s) is the indicial lift-growth function; CL′ = dCL /dα ; s = Ut /B is a dimensionless time. Eq. (3) indicates that when
the wind angle of attack is changed, the corresponding aerodynamic loads would not develop immediately, but instead,
undergo a transient evolution process before the steady value is attained. The IF for thin airfoils was firstly determined by
Wagner, with the limiting characteristics ϕ (0) = 0.5 and ϕ (∞) = 1. Theodorsen (1934) provided discrete numerical values
of ϕ (s) for a flat plate and Jones (1940) offered an approximation as

ϕ (s) ≈ 1 − 0.165e−0.0455s − 0.335e−0.3s . (4)

In bridge aerodynamics, the IF can be, if desired, altered to a more flexible form as
i
∑
ϕ (s) = 1 − ai e−di s , (5)
1

where ai , di are constants to be identified and di > 0. A number of authors tried IFs in time-domain flutter analysis or just
for the description of aeroelastic effects (Caracoglia and Jones, 2003; Borri et al., 2002; Costa and Borri, 2006; Salvatori and
Borri, 2007; Zhang et al., 2011; Miranda et al., 2013; Farsani et al., 2014).
In virtue of the concept of the IFs and the principle of linear superposition, the aerodynamic loads per unit length due to
arbitrary motions can be expressed as
Lse (x, s) = Lseα (x, s) +[∫Lseh (x, s)
s
h′′ (x, σ )
∫ s ]
1 (6)
= ρ U BCL
2 ′
ϕLα (s − σ ) α ′ (x, σ ) dσ + ϕLh (s − σ ) dσ
2 −∞ −∞ B

Mse (x, s) = Mseα (x, s) +[∫

Mseh (x, s)
s
h′′ (x, σ )
∫ s ]
1 (7)
= ρ U B CM
2 2 ′
ϕM α (s − σ ) α ′ (x, σ ) dσ + ϕMh (s − σ ) dσ
2 −∞ −∞ B
where α ′ is the derivative of α with respect to dimensionless time s, h′′ the second derivative of h with respect to s, and x is
the longitudinal coordinate of the bridge deck.
Perform Fourier transforms of (6) and (7) result in the following frequency-domain equations:
{ }
1 ] iK h [
L (x, K ) = ρ U BCL ϕLh (0) + ϕ Lh
2 ′ ′
+ ϕLα (0) + ϕ Lα α ,
′
[ ]
(8)
2 B
{ }
1 ] iK h [
M (x, K ) = ρ U B CM ϕMh (0) + ϕ Mh
2 2 ′′
+ ϕM α (0) + ϕ M α α ,
′
[ ]
(9)
2 B

where the overbar denotes Fourier transform, K = Bω/U, and

i i i
aFAi d2FAi aFAi dFAi
= RFA eiθFA ,
∑ ∑ ∑
ϕFA (0) + ϕ ′FA = 1 − aFAi + 2
− iK (10)
dFAi + K 2 d2FAi + K 2
1 1 1

and
∑i aFAi dFAi
−K 1 d2 +K 2
θFA = arctg FAi
, (11)
∑i ∑i aFAi d2FAi
1− 1 aFAi + 1 d2 +K 2
FAi

[ i i
]2 [ i ]2
 ∑ ∑ aFAi d2 ∑ aFAi dFAi
RFA =
√ 1− aFAi + FAi
+ K . (12)
d2FAi + K 2 d2FAi + K 2
1 1 1
416 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

The indicial functions are able to be obtained from known flutter derivatives. This can be done first with Fourier
transforms Eqs. (1) and (2), as follow:
[ ]
1 )h
L (K ) = ρ U BK 2 2 ∗ ∗
+ iH2 + H3 α ,
∗ ∗
( ( )
iH1 + H4 (13)
2 B
[ ]
1 )h
M (K ) = ρU B K 2 2 2 ∗ ∗
+ iA2 + A3 α .
∗ ∗
( ( )
iA1 + A4 (14)
2 B

Compare expression (13) with (8), (14) with (9) respectively, and notice that the two group of spectrums should be
equivalent, one may get the following connections:

K H1∗ − iH4∗ = CL′ ϕLh (0) + ϕ ′Lh ,

( ) [ ]
(15)
2
H3∗ + iH2 = CL ϕLα (0) + ϕ ′Lα ,
∗ ′
( ) [ ]
K (16)

K A1 − iA4 = CM ϕMh (0) + ϕ ′Mh ,

∗ ∗ ′
( [ ) ]
(17)

K 2 A∗3 + iA∗2 = CM
′
ϕM α (0) + ϕ ′M α .
( ) [ ]
(18)
According to Eqs. (8) and (9), if suppose α (x, t ) = α0 sin (ωt ) and h (x, t ) = h0 sin (ωt + φ), then the aerodynamic lift
and torque would be
[ ]
1 K
L (x, t ) = ρ U 2 BCL′ RLα α0 |sin θLα | sin (ωt + θLα ) + RLh h0 |cos θLh | cos (ωt + φ + θLh ) , (19)
2 B
[ ]
1 K
M (x, t ) = ρ U B CM RM α α0 |sin θM α | sin (ωt + θM α ) + RMh h0 |cos θMh | cos (ωt + φ + θMh ) .
2 2 ′
(20)
2 B
Thus the work done by the lift and torque over a single period can be integrated as
2π [ ]
ω
∫
1 K
WLh = L (t ) dh (t ) = ρ U 2 BCL′ π RLα α0 h0 |sin θLα | sin (θLα − φ) + RLh h20 |cos θLh | cos (θLh ) , (21)
0 2 B
2π [ ]
ω
∫
1 K
WM α = M (t ) dα (t ) = ρ U 2 B2 CM′ π RM α α02 |sin θM α | sin θM α + RMh h0 α0 |cos θMh | cos (φ + θMh ) . (22)
0 2 B
Eqs. (21) and (22) provide the energy input properties of a group of IFs, and it is noted that the phase angles θLα , θLh , θM α , θMh ,
and φ altogether determine the sign of the work done by aerodynamic loads, negative or positive.

3. CFD model

The deck configuration of Xihoumen bridge, a super-long suspension bridge in China with a main span of 1650 m, is
investigated in this study. Fig. 1 shows the picture of this bridge and the layout of its twin-girder deck configuration. The total
width B is 36.1 m; single-girder-width W is 15.05 m; slot-width G is 6.0 m; girder-height H is 3.524 m. The 2D computational
domain is shown in Fig. 2. A constant velocity inlet is set at the left boundary, while a pressure outlet at the atmospheric
pressure has been imposed on the right side boundary. Both the upper and the lower boundaries have been defined as slip
walls. Surfaces of the deck section are defined as non-slip. Relative to the deck width B, the upwind computational length is
10B and the downwind length is 25B. The height of the computational domain is 12B, which ensures a small enough blocking
ratio, H /(12B) ≈ 0.008.
The finite-volume method is adopted in the CFD algorithm. Numerical investigation of the concerned kind of nonlinear
aeroelasticity, namely, amplitude-dependent, requires a computational strategy that is appropriate for simulations of deep
dynamic stall, and a kernel issue in this task is turbulence modeling. Over the recent years, many turbulence models have
been tried in such kind of simulations. The overall results have shown that the class of k-ω models can lead to a good
agreement with experiments, and can find a reasonable balance between computational costs and simulating accuracy.
Among them, Wang et al. (2010, 2012) have shown that SST k-ω model can predict experimental data with reasonable
accuracy, especially for down-stroke phases; Gharali and Johnson (2013) used SST k-ω model for dynamic stall simulation
of a NACA0012 airfoil, and the results showed a good agreement of the dynamic lift coefficients with experimental results
(Lee and Gerontakos, 2004); More recently, Zanotti et al. (2014) demonstrated good performance of the Hellsten k-ω model
in deep dynamic stall simulations. In view of this, the shear stress transportation method (SST k-ω) is used to model the
turbulence viscosity in this work. The incoming flow has been set at a turbulence intensity of 2%, and the ratio of turbulence
viscosity to molecular viscosity is 5.0.
The CFD computations are conducted with the commercial code ANSYS Fluent CFX, at the newly founded National
Supercomputer Center in Changsha, China. A hybrid meshing technique is adopted to discretize the computational domain.
In the nearby region around the section, where moving grids are used to adapt the section’s motions, unstructured gridding
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 417

Fig. 1. Xihoumen suspension bridge: (a) a photograph after construction; (b) slotted bridge deck configuration.

Fig. 2. Sketch of the computational domain.

Fig. 3. Meshing around the deck and around the tip of faring.

is adopted. For the remaining field far away from the section, structured gridding is adopted. At the end of every time step,
the domain of moving grids is re-meshed based on the ALE dynamic gridding technique. Enhanced wall treatment is applied
to the boundary layer around the deck section. A second-order implicit scheme and a second-order upwind scheme are
used to solve the Reynold-averaged NS equations respectively for advancing in time and space. The coupling of pressure and
velocity is treated with SIMPLEC method. A small number of 10−6 is set for the criterion of convergence, and the maximum
number of iteration for every time step is 20.
The deck section in computation is scaled down to 1:80, the same as that used for wind tunnel tests. After a number of
inspections of independence on temporal and spatial discretization, the final grid number is determined to be about 237 400.
Gridding details around the section are shown in Fig. 3. Fig. 4 shows the non-dimensional heights of the first layer of cells
around the section, Y + . It is noticed that all Y + values corresponding the two Reynolds numbers are below 2.0, and most Y +
values are less than 1.
418 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

Fig. 4. Y + values.

Forced vibration method is used in CFD to identify the flutter derivatives. The inputs of the forced vertical and torsional
motions (see Fig. 5) are respectively as follows:

h (t ) = h0 sin (2π fh t ) , (23)

α (t ) = Φ sin (2π fα t ) , (24)

where h0 and Φ are vertical and torsional motion amplitude, respectively.

The motion frequency is fixed at 3 Hz, and the vertical and torsional motions are treated separately. Amplitudes of the
vertical motions include 6 cases from 2 to 15 cm, and those of torsional motions include 15 cases from 2◦ to 30◦ . The CFD
simulations cover a range of reduced velocity U /fB from 1.5238 to 19.8095.

4. Nonlinear aerodynamic and energy properties

4.1. Flutter derivatives

Flutter derivatives are able to be identified according to motion histories and resulted aerodynamic load histories. Fig. 6
shows the 4 flutter derivatives relating to torsional motions, which are vital to flutter stability of bluff bridge decks. Authors
also conducted wind tunnel tests, and the results are also included in Fig. 6 for comparison. It is noticed that, while the flutter
derivatives differ always among different motion amplitudes, the same trend is shared by A∗3 and H3∗ and no substantial
distinctions is observed among them. It is derivatives A∗2 and H2∗ , which dominate respectively the torque and lift due to the
torsional velocity α̇ , that divide the curves into two bunches. Curves with torsional amplitude α equal or less than 10◦ come
under the first bunch while those with α equal or greater than 12◦ come under the other. Still, in the range of low reduced
velocity, these two bunches of curves exhibit no obvious distinctions. However, from a turning point to beyond, the curves
begin to split into different bunches. The bunch of relative smaller motion amplitudes decreases monotonously in A∗2 with
the increase of U /fB, and shows an obvious trend in H2∗ turning from negative to positive. The bunch of larger amplitudes, on
the contrary, exhibits in A∗2 a trend turning from negative to positive and shows no turning point in H2∗ as reduced velocity
increases.
The demarcation between torsional amplitudes 10◦ and 12◦ can also be seen with the streamlines patterns around the
section, as shown in Figs. 7 and 8. Both cases begin with overall streamlined patterns at zero dynamic wind angle of attack.
With increase of the dynamic wind angle of attack, a small scale vortex begins to emerge at the trailing edge in the 10◦ case
and then detaches and travels downstream. Throughout the motion history this vortex seems to be quite local, and shows
no obvious trend to travel upstream and develop in size. In contrast, the situation in the 12◦ case is much different, where
a vortex emerges at the trailing edge, transfers upstream and develops drastically in size, detaches from the surface of the
girder near the edge of slot, and finally travels downstream. Another more significant difference is a vortex formation at
the leading edge in the case of 12◦ , and this leading edge vortex develops substantially in size and finally detaches from the
deck surface and travels downstream. In general, the two cases differ from each other substantially, with one exhibits overall
streamlined flow pattern and the other exhibits fully detached.
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 419

Fig. 5. Forced vibrations of the section.

Fig. 6. Flutter derivatives relating to torsional motions: (a) A∗2 ; (b) A∗3 ; (c) H2∗ ; (d) H3∗ .

4.2. Hysteresis loops

The aerodynamic lift, torque, and drag during the motion are defined in terms of aerodynamic coefficients as follows:

L = 0.5ρ U 2 BCL [α (t )] , (25)

420 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

Fig. 7. Streamlines patterns during half a motion cycle (upstroke), Φ = 10◦ , U /fB = 20.

Fig. 8. Streamlines patterns during half a motion cycle (upstroke), Φ = 12◦ , U /fB = 20.

M = 0.5ρ U 2 B2 CM [α (t )] , (26)

D = 0.5ρ U BCD [α (t )] ,
2
(27)
where CL , CM , CD are respectively aerodynamic lift, torque, and drag coefficients. They are functions of dynamic wind angle
of attack α (t ). The relations between aerodynamic load coefficients and α (t ) are able to be obtained with CFD results. Some
typical cases are shown in Figs. 9–11 in terms of hysteresis loops.
Major trends reflected by these loops are able to be observed. First, with the increase of U /fB from 2.857 to 20, shapes of
the loops become progressively flattened and the area enclosed in a loop shrinks drastically. This is a trend logically indicated
by the quasi-steady theory since large U /fB values imply situations close to a steady state. Second, as the motion amplitude
Φ increases, the smoothness of the hysteresis loops decreases, denoting a form of irregular fluctuation. This kind of irregular
fluctuation is not a fixed, periodical phenomenon, but instead exhibits stochastic characteristics and differs always among
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 421

Fig. 9. Dynamic lift coefficient CL : (a) U /fB = 2.857, Φ = 10◦ ; (b) U /fB = 2.857, Φ = 12◦ ; (c) U /fB = 2.857, Φ = 20◦ ; (d) U /fB = 2.857, Φ = 30◦ ; (e)
U /fB = 20, Φ = 10◦ ; (f) U /fB = 20, Φ = 12◦ ; (g) U /fB = 20, Φ = 20◦ ; (h) U /fB = 20, Φ = 30◦ .

Fig. 10. Dynamic torque coefficient CM : (a) U /fB = 2.857, Φ = 10◦ ; (b) U /fB = 2.857, Φ = 12◦ ; (c) U /fB = 2.857, Φ = 20◦ ; (d) U /fB = 2.857, Φ = 30◦ ; (e)
U /fB = 20, Φ = 10◦ ; (f) U /fB = 20, Φ = 12◦ ; (g) U /fB = 20, Φ = 20◦ ; (h) U /fB = 20, Φ = 30◦ .

Fig. 11. Dynamic drag coefficient CD : (a) U /fB = 2.857, Φ = 10◦ ; (b) U /fB = 2.857, Φ = 12◦ ; (c) U /fB = 2.857, Φ = 20◦ ; (d) U /fB = 2.857, Φ = 30◦ ; (e)
U /fB = 20, Φ = 10◦ ; (f) U /fB = 20, Φ = 12◦ ; (g) U /fB = 20, Φ = 20◦ ; (h) U /fB = 20, Φ = 30◦ .
422 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

Fig. 12. Dynamic lift coefficient CL within 4 motion periods, Φ = 30◦ : (a) U /fB = 2.857, (b) U /fB = 20.

Fig. 13. CWM under single-DOF harmonic torsional motion.

different motion periods (see Fig. 12), as a result of the signature turbulence formed in the flow field around the section.
Third, while the aerodynamic lift and torque coefficients increase overall linearly with the motion amplitude Φ , the drag
coefficient CD increases drastically in an obvious nonlinear style.

4.3. Energy trapping properties

Indicial functions are identified from CFD obtained flutter derivatives with relations expressed in Eqs. (15) to (18).
Parameters relating to the indicial functions are listed in the appendix tables, where 3 terms are kept for each indicial
function. Resorting to Eqs. (21) and (22), energy trapping properties are able to be computed.
The energy trapping properties, including one related to the aerodynamic lift and another to torque, are defined in
dimensionless forms as follows:
h20
[ ]
WLh h0
CWL = = CL π RLα α0 |sin θLα | sin (θLα − φ) + RLh K 2 |cos θLh | cos (θLh ) ,
′
(28)
0.5ρ U 2 B2 B B
[ ]
WM α h 0
CWM = = C ′
π R M α |sin θ M α | sin θM α + R Mh K | cos θMh | cos (φ + θ Mh ) , (29)
0.5ρ U 2 B2 α02
M
α0 B
where CWL is the energy trapping coefficient by aerodynamic lift, and CWM the one by torque.
Fig. 13 plots CWM when single-DOF harmonic torsional motion is considered, where all cases of amplitude exhibit a
negative value, showing that aeroelastic effects in these situations would dissipate the structure’s kinetic energy. This is
in consistent with A∗2 as shown in Fig. 6, of which the negative values indicate a positive aerodynamic damping. Fig. 14
shows CWL when single-DOF harmonic vertical motion is considered. As the cases of single-DOF torsional motion, negative
CWL is resulted from all motion amplitudes throughout the concerned range of U /fB.
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 423

Fig. 14. CWL under single-DOF harmonic vertical motion.

Fig. 15. CWM under torsional–vertical coupled 2-DOF harmonic motion, Φ = 2◦ , φ = 0◦ .

Negative energy trapping properties in all single-DOF cases exclude the possibility of single-mode flutter, and occurrence
of flutter of this bridge necessitate coupling of the vertical and torsional motions. This is shown by the coefficient CWM , plotted
in Fig. 15, where it is noticed as the increase of h0 /B, CWM shifts progressively from negative to positive. Another interesting
phenomenon noticed in Fig. 15 is that, with the coupling of vertical motion, it is in the relative lower part of U /fB that CWM
first enters into the positive zone. This by no means, however, indicates flutter instability is more likely to occur at low wind
speeds. That is because coupling of vertical and torsional motions would not happen until the wind speed reaches a sufficient
high threshold. However, in what situations can this kind of coupling occur is out of the range of this work.
Fig. 16 plots the results when phase angles between the coupled torsional and vertical motions are included. It is found
that, for all phase angles regardless of positive or negative, it is the case of zero phase angle that provides the maximum
energy trapping coefficients. That means, in the event of a flutter instability, the vertical and torsional displacement of a
specific position in the bridge deck would reach its maximum or minimum values simultaneously. This phenomenon well
explains what have been observed in wind tunnel tests; that is, symmetric torsional motions in flutter couple only with
symmetric vertical motion and, vice versa, asymmetric torsional couple only with asymmetric vertical motions.

5. Concluding remarks

Nonlinear aerodynamic properties of a bridge deck configuration used in the Xihoumen suspension bridge is investigated
in this work. The nonlinearity is expressed in piecewise linearized forms in terms of indicial functions. How this group of
discrete, linearized aerodynamic properties are able to be employed in time-domain aerodynamic simulation of the whole
structure will be addressed in the future. The following observations are made based on the presented results: (i) The
aerodynamic properties are found to be notably nonlinear, depending on the motion amplitudes. An obvious demarcation,
10◦ in torsion, is found to divide the properties into two groups. Flutter derivatives A∗2 and H2∗ differ significantly between
these two groups. (ii) Streamline patterns around the section indicate that, despite locally distributed vortices, the flow
424 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

Fig. 16. CWM under coupled 2-DOF harmonic motion with phase angles, Φ = 2◦ .

Table 1
Coefficients of the indicial function ϕLα .
Amplitude Φ a1 a2 a3 d1 d2 d3
2◦ 65.488 −173.838 109.593 0.0802 0.0744 0.0700
4◦ −74.050 61.690 13.606 0.0750 0.0700 0.0902
6◦ 0.189 4.286 −3.227 0.2000 0.0502 0.0798
8◦ 0.091 8.798 −7.562 2.0600 0.0502 0.0602
10◦ 0.906 −8.381 8.732 0.0702 0.0602 0.0502
12◦ −48.482 27.701 22.086 0.0602 0.0502 0.0702
14◦ 190.021 9401.612 −9590.291 0.1000 0.1106 0.1104
16◦ 0.429 −587.485 588.401 0.2700 0.0628 0.0626
18◦ 475.325 0.253 −474.190 0.0604 0.5000 0.0606
20◦ −493.633 0.258 494.749 0.0636 0.4800 0.0634
22◦ −3.629 4.781 0.194 0.0712 0.0502 0.4800
24◦ 6.192 −5.051 0.171 0.0502 0.0648 0.3700
26◦ 0.187 8.579 −7.468 0.3300 0.0502 0.0602
28◦ 7.285 −6.084 0.102 0.0502 0.0602 0.7400
30◦ 0.116 −6.288 7.490 1.0000 0.0602 0.0502

Table 2
Coefficients of the indicial function ϕM α .
Amplitude Φ a1 a2 a3 d1 d2 d3
2◦ 0.187 7.130 −6.488 0.6700 0.0542 0.0602
4◦ 0.189 −6.086 6.730 0.6800 0.0602 0.0538
6◦ 7.172 0.196 −6.527 0.0540 0.6800 0.0602
8◦ −3.661 0.191 4.328 0.0602 0.8400 0.0504
10◦ −9.129 9.816 0.252 0.0602 0.0552 1.2100
12◦ 2.024 0.397 −1.318 0.0520 2.0600 0.1000
14◦ 3.100 0.336 −2.375 0.0616 2.0600 0.0900
16◦ 2.452 0.328 −1.720 0.0556 2.0600 0.0900
18◦ 5.527 −4.773 0.259 0.0502 0.0602 2.0600
20◦ 0.229 5.560 −4.792 2.0600 0.0502 0.0602
22◦ 5.618 −4.845 0.257 0.0502 0.0602 2.0600
24◦ −4.973 0.281 5.747 0.0602 1.9600 0.0502
26◦ 0.278 8.630 −7.850 1.9300 0.0534 0.0602
28◦ 0.214 4.497 −3.712 1.5100 0.0504 0.0636
30◦ 21.043 0.219 −20.264 0.0574 1.3000 0.0602

patterns remain overall streamlined for cases in the lower-amplitudes group. In contrast, cases in the higher-amplitudes
group are much different, where a vortex emerges, transfers, and develops drastically, and finally detaches from the surface of
the girder and travels downstream, making the overall flow pattern fully detached. (iii) As U /fB increases, hysteresis loops of
load coefficients become progressively flattened and enclosed areas shrink. As the motion amplitude increases, smoothness
of the loops decreases, indicating irregular load fluctuations due to signature turbulence. (iv) The energy trapping properties
are provided in terms of dimensionless coefficients. The results show that all single-DOF cases result in negative energy
coefficients; however, coupled motions could lead to positive coefficients, showing that flutter instability of the concerned
 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426 425

Table 3
Coefficients of the indicial function ϕLh .
Amplitude h0 /B a1 a2 a3 d1 d2 d3
0.044 −1.126 0.243 2.293 0.1400 2.0600 0.0502
0.111 0.282 −2.360 3.497 2.0600 0.0900 0.0532
0.155 3.903 0.277 −2.758 0.0502 2.0600 0.0800
0.222 6.359 0.302 −5.205 0.0502 2.0600 0.0668
0.266 0.316 −3.240 4.402 2.0600 0.0800 0.0516
0.333 9.974 0.275 −8.922 0.0502 0.4300 0.0628

Table 4
Coefficients of the indicial function ϕMh .
Amplitude h0 /B a1 a2 a3 d1 d2 d3
0.044 0.295 6.401 −5.920 0.3100 0.0502 0.0626
0.111 −8.662 0.281 9.168 0.0602 0.3500 0.0502
0.155 0.295 9.334 −8.841 0.3300 0.0502 0.0602
0.222 0.189 −7.254 7.873 0.5600 0.0602 0.0502
0.266 0.262 8.787 −8.250 0.3900 0.0502 0.0602
0.333 1.785 16.835 −17.836 0.1400 0.0502 0.0602

bridge necessitates coupling between vertical and torsional motions. Further, influences of the phase angles indicate that
the symmetricity of the two coupled motions has to be consistent in the event of a flutter instability.

Appendix

See Tables 1–4.

References

Attar, P.J., Dowell, E.H., 2005. A reduced order system ID approach to the modelling of nonlinear structural behavior in aeroelasticity. J. Fluids Struct. 21,
531–542.
Attar, P.J., Dowell, E.H., Tang, D.M., 2003. A theoretical and experimental investigation of the effects of a steady angle of attack on the nonlinear flutter of a
delta wing plate model. J. Fluids Struct. 17, 243–259.
Borri, C., Costa, C., Zahlten, W., 2002. Non-stationary flow forces for the numerical simulation of aeroelastic instability of bridge decks. Comput. Struct. 80,
1071–1079.
Caracoglia, L., Jones, N.P., 2003. Time domain vs. frequency domain characterization of aeroelastic forces for bridge deck sections. J. Wind Eng. Ind. Aerodyn.
91, 371–402.
Chen, X., Kareem, A., 2008. Identification of critical structural modes and flutter derivatives for predicting coupled bridge flutter. J. Wind Eng. Ind. Aerodyn.
96, 1856–1870.
Costa, C., Borri, C., 2006. Application of indicial functions in bridge deck aeroelasticity. J. Wind Eng. Ind. Aerodyn. 94, 859–881.
Diana, G., Resta, F., Rocchi, D., Argentini, T., 2008. Aerodynamic hysteresis: wind tunnel tests and numerical implementation of a fully nonlinear model for
the bridge aeroelastic forces. In: Proceedings of the AWAS 08, Jeju, Korea.
Diana, G., Rocchi, D., Argentini, T., Muggiasca, S., 2010. Aerodynamic instability of a bridge deck section model: Linear and nonlinear approach to force
modeling. J. Wind Eng. Ind. Aerodyn. 98, 363–374.
Eskandary, K., Dardel, M., Pashaei, M.H., Moosavi, A.K., 2012. Nonlinear aeroelastic analysis of high-aspect-ratio wings in low subsonic flow. Acta Astronaut.
70, 6–22.
Farsani, H.Y., Valentine, D.T., Arena, A., Lacarbonara, W., Marzocca, P., 2014. Indicial functions in the aeroelasticity of bridge deck. J. Fluids Struct. 48, 203–215.
Galvanetto, U., Peiro, J., Chantharasenawong, C., 2008. An assessment of some effects of the nonsmoothness of the Leishman-Beddoes dynamic stall model
on the nonlinear dynamics of a typical aerofoil section. J. Fluids Struct. 24, 151–163.
Gao, G.Z., Zhu, L.D., 2015. A novel nonlinear mathematical model of the unsteady galloping force on 2:1 rectangular section. In: 14th International Conference
on Wind Engineering, Porto Alegre.
Gharali, K., Johnson, D.A., 2013. Dynamic stall simulation of a pitching airfoil under unsteady freestream velocity. J. Fluids Struct. 42, 228–244.
Jones, R.T., 1940. The unsteady lift on a wing of finite aspect ratio, NACA Report 681, US National Advisory Committee for Aeronautics, Langley, VA, USA.
Katsuchi, H., Jones, N.P., Scanlan, R.H., 1999. Multimode coupled flutter and buffeting analysis of the Akashi-Kaikyo bridge. ASCE J. Struct. Eng. 125 (1),
60–70.
Larsen, J.W., Nielsen, S.R.K., Krenk, S., 2007. Dynamic stall model for wind turbine airfoils. J. Fluids Struct. 23, 959–982.
Lee, T., Gerontakos, P., 2004. Investigation of flow over an oscillating airfoil. J. Fluid Mech. 512, 313–341.
Leishman, J.G., 1988. Validation of approximate indicial aerodynamic functions for two-dimensional subsonic flow. J. Aircr. 25, 917–922.
Leishman, J.G., Beddoes, T.S., 1986. A semi-empirical model for dynamic stall. J. Am. Helicopter Soc. 34, 3–17.
Miranda, S., Patruno, L., Ubertini, F., Vairo, G., 2013. Indicial functions and flutter derivatives: A generalized approach to the motion-related wind loads. J.
Fluids Struct. 42, 466–487.
Salvatori, L., Borri, C., 2007. Frequency and time-domain methods for the numerical modeling of full-bridge aeroelasticity. Comput. Struct. 85, 675–687.
Sarkar, S., Bijl, H., 2008. Nonlinear aeroelastic behavior of an oscillating airfoil during stall-induced vibration. J. Fluids Struct. 24, 757–777.
Scanlan, R.H., 1993. Problematics in formulation of wind-force models for bridge decks. J. Eng. Mech. 119, 1353–1375.
Scanlan, R.H., 2000. Motion-related body force functions in two-dimensional low-speed flow. J. Fluids Struct. 14, 49–63.
Shams, Sh., Sadr Lahidjani, M.H., Haddadpour, H., 2008. Nonlinear aeroelastic response of slender wings based on Wagner function. Thin-Walled Struct. 46,
1192–1203.
Stanford, B., Beran, P., 2013. Direct flutter and limit cycle computations of highly flexible wings for efficient analysis and optimization. J. Fluids Struct. 36,
111–123.
426 Z. Zhang et al. / Journal of Fluids and Structures 74 (2017) 413–426

Tang, D.M., Dowell, E.H., 1993. Nonlinear aeroelasticity in rotorcraft. Math. Comput. Modelling 18, 157–184.
Tang, D.M., Dowell, E.H., 2004. Effects of geometric structural nonlinearity on flutter and limit cycle oscillations of high-aspect-ratio wings. J. Fluids Struct.
19, 291–306.
Tang, D., Dowell, E.H., Hall, K.C., 1999. Limit cycle oscillations of a cantilevered wing in low subsonic flow. AIAA J. 37, 364–371.
Theodorsen, T., 1934. General theory of aerodynamic instability and the mechanism of flutter. NACA Report 496, US Advisory Committee for Aeronautics,
Langley, VA, USA.
Tran, C.T., Petot, D., 1981. Semi-empirical model for the dynamic stall of airfoils in view of the application to the calculation of responses of a helicopter
blade in forward flight. Vertica 5, 35–53.
Wang, S., Ingham, D.B., Ma, L., Pourkashanian, M., Tao, Z., 2010. Numerical investigations on dynamic stall of low Reynolds number flow around oscillating
airfoils. Comput. & Fluids 39, 1529–1541.
Wang, S., Ingham, D.B., Ma, L., Pourkashanian, M., Tao, Z., 2012. Turbulence modeling of deep dynamic stall at relatively low Reynolds number. J. Fluids
Struct. 33, 191–209.
Wu, T., Kareem, A., 2011. Modeling hysteretic nonlinear behavior of bridge aerodynamics via cellular automata nested neural network. J. Wind Eng. Ind.
Aerodyn. 99, 378–388.
Wu, T., Kareem, A., 2015. A nonlinear analysis framework for bluff-body aerodynamics: A Volterra representation of the solution of Navier–Stokes equations.
J. Fluids Struct. 54, 479–502.
Ying, X.Y., Xu, F.Y., Zhang, Z., 2016. Study on numerical simulation and mechanism of soft flutter of a bridge deck. In: ACEM 16, Jeju, Korea, 2016.
Zanotti, A., Nilifard, R., Gibertini, G., Guardone, A., Quaranta, G., 2014. Assessment of 2D/3D numerical modeling for deep dynamic stall experiments. J. Fluids
Struct. 51, 97–115.
Zhang, Z.T., Chen, Z.Q., Cai, Y.Y., Ge, Y.J., 2011. Indicial functions for bridge aero-elastic forces and time-domain flutter analysis. ASCE J. Bridge Eng. 16,
546–557.