J. Wind Eng. Ind. Aerodyn.

155 (2016) 11–22

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

Aeroelastic analysis of bridge deck flutter with modified implicit

coupling method
Namhun Lee a, Hyungro Lee b, Chung Baek a, Seungsoo Lee a,n
a
Department of Aerospace Engineering, Inha University, 100 Inharo, Nam-gu, Incheon 22212, Republic of Korea
b
Numerical Data Application Division, National Institute of Meteorological Science, Korea Meteorological Administration, 33 Seohobuk-ro, Seogwipo-si,
Jeju-do 63568, Republic of Korea

art ic l e i nf o a b s t r a c t

Article history: An effective time-domain aeroelastic framework for bridge deck flutters is presented based on a modified
Received 14 July 2015 implicit coupling algorithm with grid deformation techniques. The grid deformation is accomplished by
Received in revised form radial basis function interpolation as well as by the rigid movement of the initial grid. In this paper, for
24 February 2016
computational efficiency, a coupling frequency control technique is adopted for the implicit coupling
Accepted 23 April 2016
algorithm. To verify the time-domain aeroelastic framework by using the grid deformation technique, the
Available online 20 May 2016
vortex-induced vibration of the cylinder and H-section bridge deck flutter are computed, and the results
Keywords: are compared with published results. The effect of the coupling frequency with the grid deformation
Bridge deck flutter technique is presented for the flutter analysis of the Great Belt East Bridge suspension girder section.
Fluid–structure interaction (FSI)
& 2016 Elsevier Ltd. All rights reserved.
Computational fluid dynamics (CFD)
Aeroelasticity
Deforming grid

1. Introduction the preliminary design stage, a sectional model (or a taut strip
model) test is usually carried out in order to assess the aero-
Current trends with respect to increasing the sizes of vessel dynamic stability of the bridge deck section. After testing, a
sizes are largely due to the economy and efficiency associated with detailed full aeroelastic bridge model test is often carried out to
large ships. Consequently, newly constructed bridges are required predict the three-dimensional (3D) aerodynamic effects and the
to have longer spans to minimize the risk of vessel collision and to impact of complex surroundings such as suspension-bridge cables.
have large navigational clearances. However, with the increasing However, wind tunnel experiments for flutter analysis are not only
bridge spans being employed, aeroelastic stabilities such as flutter expensive, but also time-consuming.
and galloping are likely to occur. Of these aeroelastic stabilities, Two types of numerical methods may be used to determine
flutter is the most catastrophic. When the wind speed exceeds a flutter characteristics such as the critical wind speed. One is the
certain speed known as the “critical flutter speed,” the non-linear linearized method of flutter derivatives, and the other is direct
interaction of the aerodynamic, inertial, and structural forces of simulation. The flutter derivative methods are the most commonly
the bridge induces unstoppable and uncontrollable structural used approaches to obtain the aeroelastic behavior of the bridge
deflections, eventually resulting in the total structural failure of (Scanlan and Tomko, 1971). In these methods, the aerodynamic
forces and moment coefficients are assumed to be linear combi-
the bridge. After the infamous collapse of the Tacoma Narrows
nations of steady aerodynamic coefficients and aerodynamic
Bridge in 1940, bridge engineers have been required to demon-
derivatives of the bridge (Toshio Miyata, 2003). The aerodynamic
strate that their bridge designs have sufficient flutter margins.
derivatives are calculated from the force and moment responses of
One of the successful methods for bridge flutter analysis is
the heaving and torsional motions. The critical flutter speed can be
based on wind tunnel testing with a dynamically scaled model.
estimated from the stability analysis of the linearized equations of
The wind tunnel test results are directly applied to the design
motion. Larsen and Walther (1997) presented a method to deter-
process. In addition, the wind tunnel tests should enhance the
mine the critical flutter speed by using this method. They used a
understanding of the aeroelastic motions of the bridge, and thus
numerical code based on discrete vortex simulation (DVS) for the
provide the means for the “calibration” of analytical procedures to
computation of the aerodynamic responses. However, interactions
be used in connection with design calculations (Larsen, 1993). In between aerodynamic and structural dynamics can be nonlinear
when there is significant deformation of the bridge. The linearized
n
Corresponding author. Tel.: þ 82 32 860 7358; fax: þ 82 32 860 5401. methods cannot take into account these nonlinear interactions
E-mail address: [email protected] (S. Lee). beyond linear superposition (Wu et al., 2013).

http://dx.doi.org/10.1016/j.jweia.2016.04.010
0167-6105/& 2016 Elsevier Ltd. All rights reserved.
12 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

On the other hand, the direct simulation methods calculate movement. The implicit method is more accurate, while the
the aeroelastic responses by using coupled fluid and structural explicit method is more efficient. Care should be taken to ensure
analyses. The flutter speed can be computed from the dynamic conservation on the fluid–structure interface when the parti-
behavior of the bridge deck condition. The flutter condition is tioned approach is chosen.
defined when an oscillation of increasing amplitude is obtained. To correctly study the aeroelastic behavior of a bridge deck near
The direct simulation methods are more computationally the water surface or the ground, automatic grid generation is
intensive than the flutter derivative methods. However, with required to generate the grid system over the bridge deck. It is well
recent advances in computer technology and numerical analy-
known that the ground effect significantly changes the aero-
sis, numerical methods using computational fluid dynamics
dynamic forces and moments of bodies near the ground. When the
(CFD) and computational structural dynamics (CSD) are alter-
bridge deck moves relative to the water surface or the ground, the
native analysis tools that provide a reduced cost and time to
analyze the fluid–structure interaction (FSI). One of the main lower part of the outer boundaries should be fixed to in order to
advantages of the direct simulation methods is that the account for the ground.
unsteady aerodynamic forces and wind actions of the bridge can The grid system can be completely regenerated every time by
be obtained directly from the aeroelastic behavior, unlike the using transfinite interpolation (TFI) (Byun and Guruswamy,
flutter derivative methods. In addition, they can simulate 1998) or can be obtained by deforming the initial grid system.
situations that are more complicated by the interaction of The linear spring analogy (Batina, 1990) or torsional spring
aeroelastic nonlinearities that are due to the large deformation, analogy (Farhat et al., 1998) can be used to obtain the grid
material nonlinearities, or complex damping properties (Wu system automatically. While TFI is efficient and fast, it can only
et al., 2013). In addition, the direct simulation methods are be applied on structured grids. The linear spring analogy
cheaper and faster than the wind tunnel test. The results that replaces the grid edges inside the computational domain with a
were obtained based on these methods can be found in early network of springs, and the new grid points are determined as
studies (Selvam and Govindaswamy, 2001; Braun and Awruch, equilibrium points. Greater control over deformation can be
2003; Frandsen, 2004).
achieved with the addition of a torsional spring system. Spring
The numerical approaches to the FSI problems can be cate-
analogy methods can be used for both the structured grid and
gorized as two types of approaches: the monolithic approach and
the unstructured grid. However, these methods are known to be
the partitioned approach. The monolithic approach combines the
fluid and structural equations into a single formulation. The pro- inadequate for highly deforming grid systems. On the other
minent feature of this approach is that conservation can be hand, the radial basis function (RBF) interpolation method
maintained. Moreover, it has improved time-accuracy and stability (Rendall and Allen, 2008) can create a good quality grid system
properties. However, it is difficult to formulate this approach in a from the initial grid points. In this approach, the deformed grid
single numerical form. The partitioned approach is more common. quality or the deformation efficiency can be controlled by the
Coupled fluid and structural systems are solved sequentially by number of surface grid points used to determine the inter-
using existing CFD and CSD solvers to determine the converged polation coefficients.
fluid and structural coupling solutions. The advantages of this In this paper, we propose an efficient analysis framework for
approach are computational efficiency and simplicity of imple- the aeroelastic simulation of bridge deck flutter. This framework
mentation. These advantages come from the modularity used in uses the implicit coupling method for fluid–structural coupling
this approach. and the free oscillation method for determining the critical wind
The coupling of fluid and structural systems using the par- speed. An incompressible Reynolds averaged Navier–Stokes
titioned approach can be achieved either with an implicit
(RANS) solver based on a finite volume method (FVM) is used for
(strong) coupling method or with an explicit (loose) coupling
flow analysis, while the dynamic equations of the bridge deck
method. Fig. 1 depicts these two coupling methods of the par-
section are used to determine the dynamic response. To reduce the
titioned approach. The implicit coupling method iteratively
determines a coupled solution at each time step. On the other computational burden of coupling aerodynamic and structural
hand, the explicit coupling method seeks the coupled solution solvers, a modified implicit coupling method is adopted. The RBF
by solving the dynamic equations with a time-delayed fluid interpolation method is used to automatically generate a grid
force and the fluid equations with a time-delayed geometric system around the bridge deck.

Fig. 1. Partitioned coupling approach: (a) implicit coupling, (b) explicit coupling.
 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22 13

2. Formulations and implementation approximate factorization-alternate direction implicit (AF-ADI)

method, is used to discretize the fictitious time derivative term
2.1. Fluid solver of the equations. Detailed information of the code can be found
in Lee and Lee's (2011) paper.
An in-house two-dimensional (2D) incompressible RANS solver A deforming grid technique is adopted for the automatic gen-
with two-equation turbulence models (Lee and Lee, 2011) are eration of a grid system inside the flow field in order to predict the
chosen for flow analyses around bridge sections. The modified flow induced by the flutter motion of the bridge. In this study, we
artificial compressibility method (Turkel, 1987) is adopted to solve use an RBF interpolation method. Rigid movement of the grid
the incompressible RANS equations using a time marching system is used to verify the suitability of the RBF interpolation
method. An integral form of the non-dimensionalized governing method by comparing the computational results with the RBF
equations can be written as: method and the rigid movement method.
Z Z Z Z Z A good quality grid can be created from the deformed surface
d d ! !
Im Q dV þ Γ p Q dV þ b dS ¼
F Un b dS þ SdV;
F Un ð1Þ grid by using the RBF interpolation method (Rendall and Allen,
dt V dτ V δV δV V
2008). This grid deformation method is based on a multivariate
where Q is the primitive solution vector and I m ¼ Diag interpolation using RBFs. Moreover, it can be applied without any
  ! !
0 1 1 1 1 1 . F and F v are the inviscid vector and viscous geometric constraints. For the efficient construction of the inter-
flux vector, respectively. Further, S is the source term due to the polation function, a data reduction (boundary point coarsening)
turbulence models. n b denotes the outward normal vector to the scheme, that is based on a greedy algorithm proposed by Schaback
surface of the computational cell. The solution vector and the flux and Wendland (2000), is used. Furthermore, the singular value
vectors are defined by decomposition is used to solve the linear equations.
2 3 2 3 2 3 2 3 Fig. 2 shows the initial multi-block grid system over an H-
V 0
p ! b7 0 section bridge deck and the deformed grid system, which results
6 6 b b 7
6 7
6u7 6 u V þpi 7 6 τxx i þ τyx j 7 6 7
607 from the rotation as well as the translation of the bridge deck.
6 7 6 ! b7 7 6 7 6 7
6v7 ! 6 6 b
6 v V þ pj 7 ! 6 τxy i þ τyy j
b 7
7 607 During the deformation, the outer boundaries of the grid system
6 7   6 7
Q ¼ 6 7; F ¼ 6 6
7
! 7 v 6; F ¼ 6
ν ν
7
7 ; S ¼ 6 7; remain fixed. Fig. 2 indicates that the final grid system maintains a
6T 7 6 T V 7 6 m
þ t
∇T 7 607
6 7 6 7 6 Pr Prt
7 6 7
6 s1 !
6 s1 7 6 S1 7 good grid quality even with a large rotational displacement.
4 5 V 7 6 ðν þ σ ν Þ∇s 7 4 5
4 5 4 m s1 t 15 Thomas and Lombard (1978) showed that the geometric eva-
s2 ! S2
s2 V ðνm þ σ s1 νt Þ∇s2 luation of a computational cell of a deforming grid is not sufficient
ð2Þ to ensure that a uniform flow is a solution to the Navier–Stokes
! ^ equations. The geometric conservation law is adopted to alleviate
where p, V ¼ iu þ ^jv, and T are the pressure, velocity vector, and this problem.
temperature, respectively. In addition, s1 and s2 are the turbulent
variables, and σ s1 and σ s2 are the turbulent model constants. The 2.2. Structural solver
total stress tensor, τij is defined by
  The sectional deformation of the bridge deck section is small
∂ui ∂uj
τij ¼ ν þ ; ð3Þ compared to the lateral, vertical, and rotational displacements of
∂xj ∂xi
the bridge deck. Furthermore, the lateral displacement of the
where ν is the kinematic viscosity. Here, the kinematic viscosity is bridge deck is quite small compared to the others. Therefore, the
defined by the sum of the molecular kinematic viscosity, νm and structural dynamic behavior of the bridge deck section can be
the turbulent kinematic viscosity, νt . The preconditioning matrix, determined from a 2-degree-of-freedom (DOF) linear oscillator.
Γ p (Turkel, 1987) is defined by The matrix form of the equations of motion can be written as
2 3
1=β 0 0 0 0 0 Mr€ þ Cr_ þ Kr ¼ F; ð5Þ
6 α u=β 1 0 0 0 0 7
6 u 7
6 7 where M, C, and K denote the mass, damping, and stiffness
6 αu v=β 0 1 0 0 0 7
Γp ¼ 66 0
7; ð4Þ matrices, respectively. In addition, r and F are the displacement
6 0 0 1 0 07 7 and force vectors, respectively.
6 7
4 0 0 0 0 1 05  " #
y Fy
0 0 0 0 0 1 r¼ ; F¼ ; ð6Þ
α Mα
where αu and β are the preconditioning factor and the artificial
compressibility, respectively. In this study, Menter's k-ω SST (shear where y and α are the vertical and rotational displacements,
stress transport) model (Menter, 1994) and Coakley's q–ω model respectively. In addition, F y and M α are the lift and pitching
(Coakley, 1983) are used to simulate turbulent flow. moment, respectively. For the numerical integration of the equa-
The numerical algorithm used in this study is based on the tions of motion, we use the Newmark-β method (Newmark, 1959),
FVM. For the discretization of the inviscid flux vector, Roe's which is a second-order time integration method that is widely
approximated Riemann solver (Roe, 1981), in conjunction with used because of its stability.
the Monotone Upstream-Centered Scheme for Conservation
Laws (MUSCL) extrapolation of van Leer (1979), is used. Van 2.3. Coupling procedure
Albada's limiter is also adopted to obtain a higher spatial
accuracy while maintaining the total variation diminishing In this study, the implicit coupling method is adopted to
(TVD) property. The derivatives of the solution vector are analyze the FSI problems. Generally, the implicit coupling
computed at the cell interfaces by applying the gradient theo- method provides reliable aerodynamic/structure coupled solu-
rem over an auxiliary cell. These derivatives are used to com- tions by exchanging data once every sub-iteration for each
pute the viscous flux vector, which is equivalent to the second- physical time step. At a given time instant, the forces and
order central difference method on a regular grid. Moreover, a moments that are integrated on the surface of the structure are
dual time stepping method, in conjunction with the transferred to the structural solver from the CFD solver. After
14 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

Fig. 2. Grid deformation results by using RBF interpolation.

3. Numerical results

3.1. Validations

3.1.1. Vortex-induced vibration of the elastically mounted cylinder

As a first validation problem, numerical simulations of the
vortex-induced vibration (VIV) of an elastically mounted cylinder
in a uniform flow are presented. Because the numerical simula-
tions of the VIV require a direct interaction of the flow and
structure, this problem can demonstrate the validity and accuracy
of our method. The frequency of vortex shedding from the cylinder
depends on the Reynolds number. When the vortex shedding
frequency fv and the structural vibration frequency fn coincide, the
cylinder oscillates with an increasing amplitude.
Anagnostopoulos and Bearman (1992) studied the lock-in
phenomenon experimentally. Moreover, a number of computa-
tional studies (Schulz and Kallinderis, 1998; Dettmer and Peric,
2006) of this experiment have been conducted. In Fig. 4, the
experimental setup of Anagnostopoulos and Bearman is depicted.
A cylinder that is mounted on two leaf springs vibrates in the
Fig. 3. Diagram of the implicit coupling approach with the coupling frequency transverse direction of the fluid flow as the vortices are shed. Fig. 5
control parameter. depicts the schematics of the experiment. This experiment can be
modeled as a 1-dimensional (1D) mass-spring-damper (m–K y –C y )
system.
the structural solver computes the structural deformation that is After non-dimensionalization, the equations of motion can be
due to the aerodynamic forces and moments, the deformed written as
geometry of the structure and the rates of deformation are
y€ þ 2ζω y_ þ ω 2 y ¼ nC l ðt Þ; ð7Þ
transferred to the CFD for re-meshing. If the tolerances are
satisfied, the coupling procedure can be moved to the next time where the non-dimensional mass ratio n, the natural frequency ω ,
level. Otherwise, the coupling procedure continues to the next and the damping ratio ζ are defined by
sub-iteration. However, this coupling procedure is computa- rffiffiffiffiffiffi
ρ D2 l Df CY Ky
tionally intensive when the FSI problem includes a turbulent n¼ 1 ; ω ¼ 2π n ; ζ ¼ pffiffiffiffiffiffiffiffiffiffi; f n ¼ : ð8Þ
2m U1 2 mK Y m
computation with a deforming grid.
For computational efficiency, we introduce the parameter nfsi In Eq. (7), C l ðt Þ is the unsteady lift coefficient computed by the
to control the frequency of the data exchange in the sub-iteration CFD solver.
The numerical simulations are performed over a range of the
loop (Lee et al., 2014). Fig. 3 presents a diagram of the modified
Reynolds number from 90 to 120. Figs. 6–9 show the simulation
implicit coupling method with the coupling frequency control
results. It is found that the flows are laminar for this range of
parameter. From Fig. 3, if nfsi ¼3, for example, the data exchange
Reynolds numbers, and that steady vortex shedding occurs. Fig. 6
occurs every three sub-iterations. Using this coupling frequency shows the fast Fourier transform (FFT) of the lift coefficient time
control parameter, the FSI computation can be performed more histories for the fixed cylinder and the elastically mounted cylin-
efficiently. Note that when nfsi ¼ 1, the method becomes implicit, der in terms of the vortex shedding frequency. The computations
and when nfsi is the maximum number of sub-iterations, the are performed at a Reynolds number of 112. From the figure, it is
method becomes explicit. found that the vortex shedding frequency changes because of the
 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22 15

Fig. 6. Frequency response of the transverse force, Re¼112.

Fig. 4. Experimental setup (Anagnostopoulos and Bearman, 1992).

Fig. 7. Comparison of the displacement according to the grid deforming technique.

are in agreement with each other. In particular, the maximum

displacements and the Strouhal numbers computed with both
grids agree up to four significant figures, and are 0.3612 and
0.1633, respectively.
Fig. 8 shows the relative displacement Y=D of the oscillations of
the elastically mounted cylinder and the frequency ratio f v =f n with
Fig. 5. Dynamic model of the elastically mounted cylinder. respect to the Reynolds number. The computational results indi-
cate that the lock-in occurs for Reynolds numbers ranging from
oscillation of the cylinder. At this Reynolds number, the vortex 102 to 112. For reference, the Reynolds numbers computed in this
shedding frequency of the elastically mounted cylinder is close to example are listed on the right side of the figure. The lock-in can
the natural frequency of the cylinder. As a result, this frequency be identified by the sudden jumps in the amplitude of the oscil-
change leads to the lock-in phenomena. lation. Moreover, during the lock-in, the frequency ratio becomes
To verify the present grid deforming method, two computa- nearly one. The figure shows plots of the experimental and com-
tional grids are used: a rigid grid and a deformed grid. The rigid putational results. All of the numerical results underestimate the
grid moves as the cylinder moves because of the aerodynamic displacement during the lock-in. In addition, the upper Reynolds
force. The deforming grid is generated from the surface grid points number of the lock-in region is higher in the experiment than in
that move with the cylinder and the outer fixed boundary points. the computational results. This discrepancy comes from the 3D
Fig. 7 presents the displacement responses obtained using both effects of the experiment. Dettmer and Peric (2006) noted that
grids. The Reynolds number is 108, and the results of the two grids Anagnostopoulos and Bearman (1992) did not install the end plate
16 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

Fig. 8. Amplitudes and frequencies of the oscillating cylinder.

Fig. 10. Dynamic model of the elastically mounted H-section bridge deck.

onset instability of H-section bridge decks (Hübner et al., 2001;

Dettmer and Peric, 2006; Ge and Xiang, 2008). As a second ver-
ification problem, we simulate the flutter of an H-section bridge
deck at a specific Reynolds number, and compare the results with
the numerical results obtained by Dettmer and Peric (2006).
For the numerical simulation of the flutter problem of the H-
section bridge deck, a simplified 2-DOF mass-spring system with
Fig. 9. Maximum directional force of the oscillating cylinder with respect to the translational and rotational DOFs is adopted. Fig. 10 presents the
Reynolds number. schematic of the dynamic model and the geometry of the H-
section bridge deck (Table 1).
at the lower end of the cylinder; thus, their experimental results In Table 2, the material properties used in the computation are
included 3D effects. presented, as given by Hübner et al. (2001). In this computation,
Fig. 9 shows the maximum lift (lateral force) and drag for each the Reynolds number, which is based on the width of the bridge
Reynolds number. The numbers in the symbols correspond to the deck B, is 1500. From Fig. 10, the width of the bridge deck B was set
indices listed in Fig. 8. As shown in Fig. 9, the maximum lift and to 12 m.
drag increase drastically at the low end of the lock-in region. As Fig. 11 illustrates time histories of the relative vertical dis-
the Reynolds number increases in the lock-in region, the max- placement Y=B and rotational angle α. The maximum relative
imum lift and drag decrease gradually (pt. 4–9). However, beyond vertical displacement and the maximum rotational angle are 0.059
the lock-in region, the drag decreases while the lift remains nearly and 53.6°, respectively, and they differ by about 17% and 6%,
unchanged (pt. 10–13). respectively, when compared to the numerical results obtained by
Dettmer and Peric (2006).
3.1.2. Flutter of the H-section bridge deck The reduced frequency of translation and rotation, ko which is
H-section shapes are widely adopted as bridge decks because obtained using the FFT, is 1.35, as shown in Fig. 12. This result
they are relatively easy to make and maintain. However, the H- indicates that the dominant motion of the flutter in this example is
section bridge deck is aerodynamically unstable; thus, special care rotation. The reduced natural frequencies and the nondimensional
should be taken in designs to prevent aerodynamic instability. form of the natural frequency, which is calculated from the
There have been many studies concerning the prediction of the material properties of the bridge deck, are ky;n ¼0.98 and
 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22 17

kα;n ¼1.51, respectively. The flutter frequency is closer to the “dynamic (flutter)” state. A simplified mass-spring system with
rotational natural frequency than to the vertical natural frequency. translational and rotational DOFs is used to compute the dynamic
Fig. 13 shows the aeroelastic responses of the bridge deck and behavior of the bridge section.
the flow patterns (vorticity distribution) around the bridge deck. For the computational analysis, an H-type grid is generated,
In Fig. 13(c–f), the blue regions indicate zones of negative vorticity which has a far-field length and height equal to 30B and 15B,
and red regions indicate zones of positive vorticity. When the respectively. There are about 80,000 cells, which are subdivided
bridge deck reaches the maximum amplitude of the rotation, the into eight blocks. The normal spacing next to the wall is 0.0001B;
pitching moment exerted on the bridge deck acts against the thus, the dimensionless wall distance y þ is maintained close to
rotation of the bridge deck. As can be seen in Fig. 13(a), the one at a Reynolds number (Re ¼ U 1 B=ν) of 1:0  105 . The k–ω SST
pitching moment curve (or phase curve, cm  cl response) generally and the q–ω model are employed to compute the turbulent visc-
shows unstable slopes, except where the direction of rotation osity. From the experimental setup, a parameter of the intensity of
changes. turbulent flow of 7.5% is considered. A nondimensional time step
Δt  of 0.05 is used for both the stationary and dynamic compu-
tations. The number of sub-iterations for the dual time stepping is
3.2. Flutter of the suspension span girder of the Great Belt East
set to 100.
Bridge
3.2.1. Stationary results
In this section, the computational results obtained for the
The stationary aerodynamic analysis is performed for angles of
cross-section of the suspension span girder of the Great Belt East
incidence α ranging from  10° to 10° with intervals of 5°. The
Bridge (GBEB) are considered. The GBEB is a suspension bridge in deforming grid using RBF interpolation is used to construct the
Denmark with a main span of 1624 m between two main towers. grid with respect to the angle of incidence. From the unsteady
Various wind tunnel test data (Reinhold et al., 1992; Poulsen et al., computation, the mean aerodynamic coefficients, cd , cl , and cm are
1992; Larsen, 1993), as well as computational results (Larsen and calculated by averaging the time history.
Walther, 1997; Frandsen and McRobie, 1998; Taylor et al., 1999;
Bruno and Khris, 2003) are available for comparison purposes. The
geometric and material properties of the bridge girder are shown
in Fig. 14 and Table 3, respectively. In this study, computational
analyses are performed for both the “stationary” state and the

Table 1
Properties of the elastically mounted cylinder.

Natural frequency Non-dimensional mass ratio Damping ratio

(fn, Hz) (n ¼ ρ1 D2 l=2m) (ζ)

7.016 0.00427 0.0012

Table 2
Properties of the elastically mounted H-section bridge deck and the fluid around
the bridge deck.

Fluid properties Structural properties of the bridge deck

ρ μ u1 Re m Iα Ky Kα
(kg/m3) (Ns/m2) (m/s) (kg) (kg m2) (N/m) (N m)

1.25 0.1 10 1500 3000 25,300 2000 40,000 Fig. 12. Frequency responses of the directional displacement of the H-section
bridge deck.

Fig. 11. Displacement histories of the H-section bridge deck: (a) vertical displacement, (b) rotational angle.
18 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

Fig. 13. Aeroelastic responses and flow patterns: (a) pitching moment curve, (b) displacement hysteresis for one flutter period T, (c) vorticity contour at T ¼0, (d) vorticity
contour at T ¼ 0.25, (e) vorticity contour at T ¼0.5, and (f) vorticity contour at T ¼ 0.75.

Fig. 14. Geometry of the Great Belt East Bridge: (a) sectional model (Reinhold et al., 1992), (b) dynamic model of the GBEB girder section (Selvam and Govindaswamy, 2001).

Table 4 shows a summary of the results for the averaged static Table 3
force coefficients and the Strouhal number of the flow for the Properties of the GBEB bridge girder section.
GBEB suspension span for an angle of incidence of zero. The
Mass (kg/m) Mass Moment fy (Hz) f α ðHzÞ
computed results are in good agreement with the results obtained
of Inertia
in the previous studies. Fig. 15 shows a comparison of the averaged (kg m2)
static force coefficients between the computed results and the
wind tunnel tests for different angles of incidence. The results 22:7  103 2:47  106 0.099 0.272
obtained from the present computations show good correlation
with the wind tunnel experiments. However, the computed drag
coefficients are smaller than the experimental results for most
angles of incidence. The difference in the drag coefficients results
from the difference between the experimental model and the Table 4
Summary of the stationary aerodynamic results (a ¼ 0 3 ).
numerical model. As can be seen in Fig. 14, the side railings and
crash barrier were attached in the experimental model; however, Method Model cd cl St
they are not included in the numerical model. Note that there are
some differences between the results of the k–ω SST and the q–ω Reinhold et al. Wind tunnel Section 0.08 þ 0.01 0.109–
(1992) test 0.158
models. Larsen and DVM pot. þ ω 0.06 þ 0.06 0.100–
Walther 0.168
3.2.2. Dynamic results (1997)
To determine the critical flutter speed, unsteady computations Taylor and DVM pot. þ ω 0.05 þ 0.07 0.16–0.18
Vezza (1999)
of the bridge girder section are performed for various inflow
Selvam and FEM NSE les 0.06  0.34 0.168
velocities U 1 , where the bridge section is allowed to move Govindas-
because of the interaction between the flow and the structure. The wamy (2001)
critical flutter speed is calculated using the free oscillation pro- Present FVM RANS (k–ω 0.05 þ 0.03 0.160
cedure suggested by Selvam and Govindaswamy (2001). In this SST)
RANS (q–ω) 0.05  0.01 0.160
procedure, the dynamic computations start with an initial
 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22 19

Fig. 15. History of the aerodynamic coefficient with respect to the angle of incidence: (a) drag coefficient, (b) lift coefficient, and (c) moment coefficient.

Fig. 16. Comparison of the displacement according to the grid deforming technique and coupling frequency: (a) rigid movement, (b) RBF interpolation.

perturbation angle of incidence of 1.8°. Depending on the stability

at the freestream speeds, this initial perturbation will decay or
amplify.
For efficient calculations, convergence test is performed in
terms of nfsi at the flutter condition U 1 ¼ 80 m/s. Computations
are done with a rigid grid, as well as the deforming grid, using RBF
interpolation. The k–ω SST model is employed in these compu-
tations. The results presented in this paper are computed on a
single node cluster with a 3.0 GHz CPU and 8 GB RAM. Fig. 16
shows the rotational angle responses with respect to the fre-
quency of the coupling for the two different grids. Some dis-
crepancies exist between the converged rotational angle respon-
ses. The amplitudes of the fifth rotational period are 11.7° (rigid
grid) and 12.5° (RBF interpolation). However, the rotational angle
responses remain the same for different nfsi values. The con-
vergence test indicates that an nfsi of five provides accurate
solutions.
Fig. 17 and Table 5 summarize the details of the convergence
study. When a coupling frequency of five is chosen (nfsi ¼ 5), the
computational efficiency is 2.0 with the rigid grid, and 2.5 with the
RBF interpolation. Here, the computational efficiency is defined by Fig. 17. Comparison of the computation time with respect to the coupling
the ratio to the fully implicit (nfsi ¼1) calculation time. Note that frequency.
the computational efficiency of the explicit coupling is 3.2 with the
rigid grid, and 3.7 with the RBF interpolation. Nevertheless, the An aeroelastic analysis is performed to determine the critical
difference in the results with an nfsi of one is less than 1%. Based flutter speed. Computations are performed with freestream velo-
on the convergence test, an nfsi of five is chosen for the rigid grid cities ranging from 50 m/s to 80 m/s with an interval of 10 m/s.
and for the deforming grid to ensure the accuracy and efficiency of These freestream velocities correspond to the reduced velocities of
the solution. 5.93, 7.12, 8.30, and 9.49, respectively. Fig. 18 shows the rotational
20 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

angle responses at these reduced velocities. The rotational for the reduced velocities of 8.06 and 8.53. The critical reduced
amplitude decreases when the freestream velocity is below the velocities are then computed with the interpolation, and listed in
critical flutter speed, while it increases when the freestream Table 6. The critical velocity predicted in this study is in good
velocity is above the critical flutter speed. agreement with that of the wind tunnel tests.
The total damping ratio can be determined from the structural
time response (Fig. 18) by adopting the logarithmic decrement
method. From the logarithmic decrement, δ, the total damping
ratio, ζ total can be defined as
δ
ζ total ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi: ð8Þ
ð2π Þ2 þ δ
2

By calculating the total damping ratio, we can determine the

stability of the bridge deck. For example, the negative values of the
damping ratio indicate that the bridge is in the flutter condition.
Fig. 19 presents the total damping ratio of the torsional oscil-
lation at each reduced velocity. From Fig. 19, the critical reduced
velocity is estimated to be around 8.50. For the accurate prediction
of the flutter speed, two additional computations are carried out

Table 5
Summary of the error with respect to the grid deforming method and coupling
frequency.

Deforming method nfsi Error (%) Deforming method nfsi Error (%)

Rigid movement 1 – RBF interpolation 1 –

5 0.5 5 0.6
10 1.3 10 1.3
Fig. 19. Critical velocity computation.

Fig. 18. Comparison of the displacement according to the reduced velocity V  and turbulence model: (a) V  ¼ 5:93, (b) V  ¼ 7:12, (c) V  ¼ 8:30, and (d) V  ¼ 9:49.
 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22 21

Table 6
Summary of the results of the critical flutter speed analysis.

Method Flutter speed analysis V crit V  (m/s)

Larsen (1993) Wind tunnel tests (Section model) Flutter derivative 8.66 73
Larsen and Walther (1997) DVM, Pot. þ ω (Forced oscillation) Flutter derivative 8.78 74
Frandsen and McRobie (1998) NSE/Lam (Free oscillation) Point computation 5.93 50
Selvam and Govindaswamy (2001) NSE/LES (Free oscillation) Growth/Decay rate 7.71–8.54 65–72
Braun and Awruch (2003) NSE/LES (Free oscillation) Growth/Decay rate 8.18 69
Flutter derivative 8.66 73
Present RANS/k–ω SST (Free oscillation) Total damping ratio 8.57 72.3
RANS/q–ω (Free oscillation) Total damping ratio 8.53 71.9

the RBF interpolation, presents a more efficient method for

achieving time-accurate solutions.

Acknowledgments

This work was supported by Inha University and a National

Research Foundation of Korea (NRF) Grant funded by the Korean
Government (2011-0029094).

References

Anagnostopoulos, P., Bearman, P.W., 1992. Response characteristics of a vortex-

excited cylinder at low Reynolds numbers. J. Fluids Struct. 6.1, 39–50.
Batina, J.T., 1990. Unsteady Euler airfoil solutions using unstructured dynamic
meshes. AIAA J. 28.8, 1381–1388.
Braun, A.L., Awruch, A.M., 2003. Numerical simulation of the wind action on a long-
span bridge deck. J. Braz. Soc. Mech. Sci. Eng. 25.4, 342–363.
Bruno, L., Khris, S., 2003. The validity of 2D numerical simulations of vortical
structures around a bridge deck. Math. Comput. Model. 37, 795–828.
Byun, C., Guruswamy, G.P., 1998. A parallel, multi-block, moving grid method for
aeroelastic applications on full aircraft. AIAA, 98–4782.
Fig. 20. Variation of rotational and heave frequencies with reduced velocity. Coakley, T.J., 1983. Turbulence modeling methods for the compressible Navier–
Stokes equations. AIAA Pap., 83–1693.
Dettmer, W., Peric, D., 2006. A computational framework for fluid-rigid body
Fig. 18 also shows that as the reduced velocity increases, the interaction: finite element formulation and applications. Comput. Methods
rotational frequency decreases. The variations of the rotational and Appl. Mech. Eng. 195, 1633–1666.
heaving frequencies are plotted as a function of the reduced Farhat, C., Degand, C., Koobus, B., Lesoinne, M., 1998. Torsional springs for two-
dimensional dynamic unstructured fluid meshes. Comput. Methods Appl.
velocity in Fig. 20. Note that the rotational frequency decreases as Mech. Eng. 163, 231–245.
the freestream velocity increases, but the heaving frequency Frandsen, J., McRobie, A., 1998. Comparison of numerical and physical models for
bridge deck aeroelasticity. IABSE Symp., Kobe, Japan
increases near the onset of flutter. Furthermore, the heaving fre-
Frandsen, J.B., 2004. Numerical bridge deck studies using finite elements. Part I:
quency becomes identical to the rotational frequency at the point flutter. J. Fluids Struct. 19, 171–191.
of flutter instability, which was reported by Robertson et al., Ge, Y.J., Xiang, H.F., 2008. Computational models and methods for aerodynamic
flutter of long-span bridges. J. Wind. Eng. Ind. Aerodyn. 96, 1912–1924.
(2003). Hübner B., Walhorm E., Dinkler D., 2001. Strongly coupled analysis of fluid-
structure interaction using space-time finite elements. In: Proceedings of the
2nd European Conference on Computational Mechanics, Cracow, Poland.
Larsen, A., 1993. Aerodynamic aspects of the final design of the 1624 m suspension
4. Concluding remarks bridge across the Great Belt. J. Wind. Eng. Ind. Aerodyn. 48, 261–285.
Larsen, A., Walther, J.H., 1997. Aeroelastic analysis of bridge girder sections based
on discrete vortex simulations. J. Wind. Eng. Ind. Aerodyn. 67, 253–265.
This paper presents an effective time-domain aeroelastic fra- Lee, H., Lee, S., 2011. Convergence characteristics of upwind method for modified
mework for long span bridges. An incompressible RANS solver is artificial compressibility method. Int. J. Aeronaut. Space Sci. 12.4, 318–330.
Lee N., Lee S., Cho H., Kwak J.Y., Shin S.J., 2014. Computational analysis for flapping
used for the aerodynamic analysis, and the rigid body equations of wing by coupled CFD and CFD solutions. In: Proceedings of the 29th Congress of
motion are employed for the structural analysis. The present the International Council of the Aeronautical Sciences, Petersburg, Russia.
aerodynamic and structural analyses are coupled with a modified Menter, F.R., 1994. Two-equation Eddy-viscosity turbulence models for engineering
applications. AIAA J. 32.8, 1598–1605.
implicit coupling approach, which can control the coupling fre- Miyata, T., 2003. Historical view of long-span bridge aerodynamics. J. Wind. Eng.
quency for efficient aeroelastic computations. To verify this fra- Ind. Aerodyn. 91, 1393–1410.
Newmark, N.M., 1959. A method of computation for structural dynamics. J. Eng.
mework, computations of the bridge deck flutter of the GBEB
Mech. Div. 85.3, 67–94.
suspension girder section are performed. In particular, the critical Poulsen, N.K., Damsgaard, A., Reinhold, T.A., 1992. Determination of flutter deri-
flutter speed is evaluated with the free oscillation method. The vatives for the Great Belt Bridge. J. Wind. Eng. Ind. Aerodyn. 41.44, 153–164.
Reinhold, T.A., Brinch, M., Damsgaard, A., 1992. Wind-tunnel tests for the Great Belt
present computational results with a rigid grid and deforming grid Link. In: Larsen, A. (Ed.), Aerodynamics of Large Bridges. Balkema, Rotterdam.
with RBF interpolation agree well with the experimental results, as Rendall, T.C.S., Allen, C.B., 2008. Unified fluid–structure interpolation and mesh
well as the other computational results. In addition, the results are motion using radial basis functions. Int. J. Numer. Methods Eng. 74 (10),
1519–1559.
demonstrated that the modified implicit coupling method, in Roe, P.L., 1981. Approximate Riemann solvers, parameter vectors, and difference
conjunction with the efficient grid-deforming algorithm based on schemes. J. Comput. Phys. 43, 357–372.
22 N. Lee et al. / J. Wind Eng. Ind. Aerodyn. 155 (2016) 11–22

Robertson, I., Sherwin, S.J., Bearman, P.W., 2003. Flutter instability prediction Thomas, P.D., Lombard, C.K., 1978. The geometric conservation law a-link between
techniques for bridge deck sections. Int. J. Numer. Methods Fluids 43, finite-difference and finite-volume methods of flow computation on moving
1239–1256. grids. AIAA Pap., 78–1208.
Scanlan, R.H., Tomko, J.J., 1971. Airfoil and bridge deck flutter derivatives. J. Eng. Turkel, E., 1987. Preconditioned methods for solving the incompressible and low
Mech. Div. 97.6, 1717–1737. speed compressible equations. J. Comput. Phys. 72, 277–298.
Schulz, K.W., Kallinderis, Y., 1998. Unsteady flow structure interaction for incom- van Leer, B., 1979. Towards the ultimate conservative difference scheme, V. A sec-
pressible flows using deformable hybrid grids. J. Comput. Phys. 143, 569–597. ond order sequel to Godunov's method. J. Comput. Phys. 32, 101–136.
Selvam, R.P., Govindaswamy, S., 2001. Aeroelastic Analysis of Bridge Girder Section Wu, T., Kareem, A., GE, Y., 2013. Linear and nonlinear aeroelastic analysis frame-
Using Computer Modelling. University of Arkansas. works for cable-supported bridges. Nonlinear Dyn. 74.3, 487–516.
Schaback, R., Wendland, H., 2000. Adaptive greedy techniques for approximate
solution of large RBF systems. Numer. Algorithms 24 (3), 239–254.
Taylor, I.J., Vezza, M., 1999. Analysis of the wind loading on bridge-deck sections
using a discrete vortex method. In: Larsen, A., et al. (Eds.), Wind Engineering
into the 21st Century. Balkema, Rotterdam.