Sunteți pe pagina 1din 13

SCIENCE CHINA

Physics, Mechanics & Astronomy


Article

February 2013 Vol.56 No.2: 277289

Special Topic: Fluid Mechanics

doi: 10.1007/s11433-012-4982-4

Numerical aerodynamic analysis of bluff bodies at a high Reynolds


number with three-dimensional CFD modeling
BAI YuGuang1*, YANG Kai1, SUN DongKe1, ZHANG YuGuang1, KENNEDY David2,
WILLIAMS Fred2 & GAO XiaoWei1
1

State Key Laboratory of Structural Analysis for Industrial Equipment, Faculty of Vehicle Engineering and Mechanics,
Dalian University of Technology, Dalian 116023, China;
2
Cardiff School of Engineering, Cardiff University, Cardiff CF24 3AA, Wales, UK
Received July 2, 2012; accepted September 5, 2012; published online January 21, 2013

This paper focuses on numerical simulations of bluff body aerodynamics with three-dimensional CFD (computational fluid
dynamics) modeling, where a computational scheme for fluid-structure interactions is implemented. The choice of an appropriate turbulence model for the computational modeling of bluff body aerodynamics using both two-dimensional and
three-dimensional CFD numerical simulations is also considered. An efficient mesh control method which employs the mesh
deformation technique is proposed to achieve better simulation results. Several long-span deck sections are chosen as examples
which were stationary and pitching at a high Reynolds number. With the proposed CFD method and turbulence models, the
force coefficients and flutter derivatives thus obtained are compared with the experimental measurement results and computed
values completely from commercial software. Finally, a discussion on the effects of oscillation amplitude on the flutter instability of a bluff body is carried out with extended numerical simulations. These numerical analysis results demonstrate that the
proposed three-dimensional CFD method, with proper turbulence modeling, has good accuracy and significant benefits for
aerodynamic analysis and computational FSI studies of bluff bodies.
bluff body, aerodynamic analysis, fluid-structure interaction, three-dimensional CFD modeling, flutter
PACS number(s): 02.70.-c, 47.11.Df, 47.27.E-, 47.27.nb
Citation:

Bai Y G, Yang K, Sun D K, et al. Numerical aerodynamic analysis of bluff bodies at a high Reynolds number with three-dimensional CFD modeling.
Sci China-Phys Mech Astron, 2013, 56: 277289, doi: 10.1007/s11433-012-4982-4

In wind engineering, investigations of flows around bluff


bodies have attracted wide attention. This challenging aeroelasticity problem has been studied extensively through
wind tunnel tests or numerical simulations [15]. In the last
decade progress has been made to find a computational alternative to partly replacing physical wind tunnel tests
which may be influenced by unpredictable factors [6] (e.g.,
incoming flow properties, model geometrical fidelity and
measurement complexity). CFD has been accepted as a potentially powerful method for investigating various wind-

*Corresponding author (email: baiyg@dlut.edu.cn)


Science China Press and Springer-Verlag Berlin Heidelberg 2013

induced vibrations of bluff bodies.


In the case of wind-induced vibrations of bluff bodies,
such as bridge decks, the combination of wind turbulence
excitation and aeroelastic effects can lead to new phenomena which are not always fully understood [4]. Especially for
high turbulent wind occurring close to the ground, the wind
gusts act more as sudden transient excitations than as stationary excitation. Computational fluid-structure interaction
(FSI) research in this field has been recognized as an efficient tool. It has been the subject of many recent engineering applications.
Tamura [7], Tamura & Itoh [8] and Tamura & Ono [9]
presented the state-of-the-art for computational FSI research
phys.scichina.com

www.springerlink.com

278

Bai Y G, et al.

Sci China-Phys Mech Astron

in the field of wind engineering. In previous work, similar


numerical simulations were already conducted with an
NACA airfoil [10,11]. Bai et al. [11] proposed a CFD
method with a computational scheme for FSI and provided
an effective assessment of aerodynamic and aeroelastic
performance of airfoil at high Reynolds numbers. However,
the dynamics and the aeroelasticity of a bluff body are quite
different from that of an airfoil. For example, most bridge
deck sections, except very streamlined ones, behave like
bluff bodies and the airflow is essentially separated downstream. Scanlan and Tomko [12] showed conclusively that,
though helpful, the Unsteady Airfoil Theory [13] has very
distinct limitations in cases of bluff bodies (e.g., aerodynamic flutter derivatives calculated even for streamlined
bridge deck sections can show limited resemblance with
those of a symmetric airfoil). A very successful two-dimensional CFD method for bluff bodies is the discrete vortex
method (DVM) as implemented by researchers such as
Walther & Larsen [14] and Taylor & Vezza [15]. These FSI
studies employed the classical grid-free method to investigate the flows with moving boundaries and it is easy to program. This has significant benefits in terms of efficiency but
has been criticized as a viscous flow model because it cannot be readily extended to three-dimensional flows. Recently, CFD software, e.g., Fluent and CFX, have become
widely used, but sometimes the precision is unsatisfactory
due to the difficulty of solving FSI problems in simulations
of flows around blunt bodies [16,17]. The block-iterative
coupling approach has been successfully applied to couple
simple fluid and solid mechanics behaviors [18]. Bai et al.
[11] have extended this method to study two-dimensional
FSI problems successfully with turbulence modeling,
boundary layer treatment and other features needed in airfoil aeroelasticity. This approach combines the dynamic
structure analysis program system DDJ-W [19,20] developed at the Dalian University of Technology with a solver
of the commercially available CFD code CFX. Whether this
approach is efficient for bluff bodies will be investigated in
this paper.
Flows around long-span bridge decks are highly turbulent, unsteady and three dimensional. The type of turbulence
model used is important for the computational modeling of
the bridge deck FSI. Bai et al. [11] used the k- turbulence
model to compute the flutter derivatives of a two- dimensional airfoil and investigated the effects of turbulence on
them. In addition to large eddy simulation (LES) or Reynolds averaged Navier-Stokes (RANS) simulation, Detachededdy Simulation (DES) is a novel turbulence model [2123]
which is promising due to its ability to explicitly resolve
turbulent structures for massively separated high Reynolds
number flows around bluff bodies. Its behavior is similar to
LES but computationally cheaper, being closer to unsteady
RANS in terms of required CPU time. This paper is concerned with the choice of an appropriate turbulence model
for the computational modeling of bluff body aerodynamics.

February (2013) Vol. 56 No. 2

Three-dimensional wind flow past three bridge deck sections is investigated and simulations yield the aerodynamic
force coefficients and flutter derivatives obtained from simulating the motion-induced aerodynamic forces when the
deck cross sections oscillate within an incompressible flow
with a high Reynolds number. In addition, the effects of
oscillation amplitude on the flutter instability of them are
studied.

1 Turbulence modeling
Most engineering flows being studied are turbulent, thus we
have to consider how to represent or model the effects of
turbulence in simulating these flows. One issue for researchers is to make appropriate choice of models for particular flows [24]. Though the Navier-Stokes equations can
describe a turbulent flow including all the turbulent eddy
details, the computational cost of direct numerical simulation (DNS) is huge. There are mainly other three kinds of
turbulence models: RANS (Reynolds averaged NavierStokes Equations), LES, and DES (= hybrid RANS/LES)
[21,25].
1.1

RANS models

In most engineering situations, it is the average velocity,


pressure, etc. that are of interest, and the details of all the
turbulent eddies are not required. RANS turbulence model
may strike the balance between computational efficiency
and accuracy in simulating the flow regime [25].
1.1.1 The standard k- model
The transport equations for the turbulent kinetic energy k
and its dissipation rate are given by the following equations in which the five constants needed are given the values
of k 1.0, 1.3, C 0.09, C1 1.44, and C2

1.92 [26].


(k)
( kui )

t
xi
xi
k

k
Gk Yk ,

xi

(1)


( )
( ui )

t
xi
xi


G Y

xi

(2)

with the density of the fluid taken as a constant, the


previously undefined terms are now defined.
The production of turbulent kinetic energy Gk is computed consistently with the Boussinesq hypothesis from
Gk t S 2 ,

(3)

where S is the modulus of the mean rate-of-strain tensor,


defined as:

Bai Y G, et al.

Sci China-Phys Mech Astron

Finally, the turbulent viscosity t is computed as:

S 2 Sij Sij
with

t
Sij

1 ui u j

2 x j xi

(4)

The turbulent viscosity t is computed from

t C

The dissipation term in eq. (1) is Yk , while the


production and dissipation terms in eq. (2) are given by
G C1

2 k 500
, 2 .
0.09 y y

2 max

(5)

Gk , Yk C2

2
k

0.09, and 1 [26]. The transport equation for k has


the same form as in the standard k- model. For the specific
dissipation rate , this is

xi

Gk = min ( t S , 10 k ) , Yk .
2

(8)

The specific dissipation is related to the dissipation


by ( / k ) ; the production and dissipation equation
terms of are
G

G.
t k

(9)

The cross term in the dissipation terms of is


D 2(1 F1 ) ,2

1 k
,
xi xi

(10)

where ,2 is a constant and F1 tanh(14 ) is a blending


function with

500
4k
k
1 min max
, 2 ,
2
0.09

,2 D y

1
F1 / ,1 (1 F1 ) / ,2

(14)

k ,1 1.176, k ,2 1.0,

,1 2.0, and ,2 1.0.


LES models

ui
0,
xi

(15)

ij p ij
u i u i u j

,
t
x j
x j xi x j

(16)

where filtered quantities are denoted by an overbar. ij is


the filtered molecular viscosity stress tensor and ij is the
sub-grid scale stress tensor resulting from the filtering operation: ij ui u j ui u j . Eqs. (15) and (16) are only correct for constant filter width. This means that for spatially
varying filter width, the spatial commutation errors that
generate supplementary terms in the equations are neglected
[27], as is commonly done in practically-oriented work. The
sub-grid scale stress tensor is modeled by
1
3

ij kk ij 2 t Sij ,

(17)

where t is the sub-grid eddy viscosity and Sij is the

(11)

where D max[2 (1 / ,2 )(1 / )(k / x j )( / x j ), 1010]


and ,2 1.168 .

(7)

The production and dissipation terms of turbulent kinetic


energy are

1
,
F1 / k ,1 (1 F1 ) / k ,2

The filtered Navier-Stokes equations for a constant-density


fluid are

t

G Y D .

xi

with 1 0.075, 2 0.0828,

1.2

( )
( ui )
t
xi

(13)

F1 1 (1 F1 ) 2 ,

1.1.2 The k- shear stress transport (SST) model


The constants needed below are given the values of

(12)

The model factors , k , are interpolated as:

(6)

k
1
,

max[(1 / ),( SF2 / a1 )]

where a1 0.31 and F2 tanh(22 ) is a blending function with

279

February (2013) Vol. 56 No. 2

rate-of-strain tensor calculated with the filtered velocity


components. The sub-grid eddy viscosity has to be further
modeled. The isotropic part of the stresses is not modeled
but added to the pressure term.
More extensive details of LES models are not given in

280

Bai Y G, et al.

Sci China-Phys Mech Astron

this paper because its calculations are too costly.

vector of structure velocities by ; a = u n1 , p n1

1.3 DES (hybrid RANS/LES) models

1.3.1 The DES k- model


The transport equations for k and of the realizable k-
model are used to model the eddy viscosity in the RANS
zones and to model the sub-grid viscosity in the LES zones.
In the hybrid formulation, the dissipation term in eq. (1)
is computed from
Yk k 3/ 2 / des ,

February (2013) Vol. 56 No. 2

(18)

contains the variables from the fluid domain; b =


n1 , n1 contains the variables from the structural
domain. The field variables at the previous time step n are
assumed to be known, and are not reflected in eq. (19). The
two equation sets are fully coupled.
Eq. (19) can be solved by the block-iterative method [18].
By using any available solvers for each of the two parts of
eq. (19) in conjunction with the block-Gauss-Seidel iterative
algorithm, the Navier-Stokes equations are solved first for a
and then for b. The iteration scheme can be written as:

where des min( rke , les ), rke k 3/ 2 / , les Cdes and


max( x, y, z ), the maximum grid spacing. The

Cdes , as in the previous model.


The k- SST model, which belongs to RANS models, is
employed for two-dimensional CFD modeling in this paper,
because it allows direct integration through the boundary
layer. Also, it has been successfully applied to numerical
simulations of two-dimensional airfoil aerodynamics [11].
The DES k- SST model is employed in this paper for
three-dimensional CFD simulations because it has better
accuracy than RANS and much lower cost of computation
than LES [21].

2 Numerical algorithm for FSI


The computation of nonlinear FSI problems requires the
simultaneous solution of the strongly coupled fluid and
structural equations of motion. In partition methods [18],
the coupled problem is computed with a solution procedure
where the fluid and structure are separated and exchange
data in every time step or iteration of the coupling algorithm.
The governing equations for the FSI come from both the
corresponding flow and structural analyses [11]. All these
equations can be treated as time and space dependent and
need to be discretized before solution methods are applied.
The discretized incremental Navier-Stokes and structural
equations can be expressed as:
N(a, b) = 0, S(b) = f(a),

(19)

where a and b are the field vectors consisting of the unknowns at the time step n+1 currently being solved for. Let
us denote the discretized vector of velocities in the fluid by
u, the corresponding pressures by p, the discretized displacement vector in the structure by and the discretized

(20)

(21)

S b(i 1) f a (i 1) ,

standard value of Cdes is 0.65.


1.3.2 The DES k- SST model
This model is based on the k- SST model. The dissipation
term of the turbulent kinetic energy is modified into eq. (18),
where des min( k , les ) , k k 1/ 2 / and les

N a (i 1) , b(i ) 0,

where i is the iteration counter, and it converges linearly.


For generality, eqs. (20) and (21) are both treated as nonlinear. Therefore linearization methods like the Newton-Raphson method or the Picard (fixed point) method
must be used. For the latter, the linearization iteration is

a((ij1)1) f a((ij) 1) , b(i ) , b((ij1)1) G a(( ij) 1) , b((ij) 1) ,

(22)

where j is the linearization iteration counter. The two layers


of iteration i and j in eq. (22) address both the field coupling
and the non-linearity. With global convergence checked at
every time step, the solution obtained should be identical to
that given by the direct coupled solution to eq. (19). Note
that the two iterations can be mixed, and when merged, the
equivalent iteration is

a ( k 1) F a ( k ) , a ( k ) , b( k 1) G a ( k 1) , b( k ) ,

(23)

where k is the merged iteration counter. Solvers in the form


of eq. (23) can be found in existing CFD and structural
analysis codes.
This paper uses a self-owned three-dimensional dynamic
structure analysis code DDJ-W, which is combined with a
solver of the commercially available finite volume code
CFX to conduct FSI problems. The accuracy of this way
will be compared to that of the popular commercial software system ANSYS workbench in this paper.

3 Structure examples and numerical modeling


3.1

Structure examples

Three long-span bridge deck cross sections are taken as


examples in the present paper. They are named as sections
G1G3. Their geometrical features are shown in Figure 1,
in which B is the chord length. Every model has two degrees of freedom, namely vertical translation h and rotation
about its centre.
Section G1 can be treated as a streamlined structure, but

Bai Y G, et al.

Sci China-Phys Mech Astron

281

February (2013) Vol. 56 No. 2

Figure 2

Rigid-plane-based mesh deformation for a cylinder.

rotate in the rigid plane which contains the shear centre response of the cylinder. Grids falling in the buffer region are
updated with mesh movements that are interpolated from
those at R R1 and at R R2 . Bai et al. [11] improved
this algorithm through adding a wake zone in the buffer
region ( R1 R R2 ) and got better computed results for
airfoils than those without wake zone, as shown in Figure 3.
The algorithm is summarized as:

Figure 1 The three long-span bridge deck sections G1G3 used. They all
have a vertical axis of symmetry.

G2 and G3 are typical bluff bodies with sharp edges. Particularly, the section G3, which has infamous aerodynamic
instability because it is the prototype of the Tacoma Narrows Bridge in the USA, was destroyed in 1954 by a steady
wind with the small velocity of 20 m/s. So it is necessary to
investigate these structures.
3.2

xS xS 0 TS xS 0 ,

R2 R

TS xS 0 ,
xS xS 0
R
2 R1

x x ,
S0
S

with R R1 and a buffer region with R1 R R2 . Fluid


grids falling in the rigid region are assumed to translate and

R1 R R2 ,

(24)

R R2 ,

TS xS ,

(25)

where

R sqrt xS2 0 yS2 0 ,


cos z
TS sin z
0

Three-dimensional CFD modeling

3.2.1 Mesh control method


Most failed simulations are caused by mesh failure, so an
efficient and robust mesh algorithm is crucial to FSI studies.
In addition to the question of coupling the fluid and structural analyses, mesh movement is an important issue in FSI
studies. Rewriting the fluid equations in the arbitrary Lagrangian Eulerian (ALE) formulation allows one to move
the mesh arbitrarily. In fact, this paper uses a mesh control
method which employs the mesh deformation technique
[11].
The mesh algorithm can be conveniently described
through a circular cylinder, see Figure 2, which only has the
three degrees of freedom of heave, shove, and pitch, respectively. The structural coordinate system differs from that of
the fluid domain. R0 is the radius of the cylinder and mesh
deformation is performed only in the cylindrical region with
R R2 . This region is further divided into a rigid region

R R1 ,

sin z
cos z
0

0
0 .
1

The left-hand column in Figure 4 shows the inner region


of the two-dimensional mesh for the three sections respectively. It is cylindrical, is centered on the deck section, and
has radius R2 16 B . The G1 meshes used for the rigid,
wake and remaining buffer region are, respectively, (594
88), (64160) and (10439), where the first and second
numbers are the number of hexahedral cells in, respectively,
the tangential and radial directions. Hence there are 66568

Figure 3

Mesh geometry.

282

Bai Y G, et al.

Sci China-Phys Mech Astron

cells, of which 44368 are in the rigid region and 10240 are
in the wake. Two-dimensional meshes for the other two
deck sections were similar to that for G1, and the cell numbers of the two-dimensional models for the other sections
are, respectively 65100 and 72728 (see the left-hand column
in Figure 4).
The main difference between two-dimensional and
three-dimensional CFD modeling is the model thickness
(i.e., perpendicular to the sections shown in Figure 1).
Meshes along the thickness direction of G1 are shown in
Figure 5. This influences the number of cells in the mesh
significantly. Many CFD methods use a very small thickness or a two-dimensional model because of the limitations
of computation capacity and uncertainty about accuracy.
For example, increasing the thickness of 0.1 m of the
two-dimensional model of section G3 to 1 m for the
three-dimensional model leads to a huge number of cells
(4089792), which require parallel computing and much
CPU time. The right-hand column of Figure 4 shows threedimensional meshes for the three sections when they all
have thickness 1 m. The numbers of cells for the three sections are respectively: 3695552, 3601600, and 4089792.

February (2013) Vol. 56 No. 2

3.2.2 Boundary layer treatment


The viscous boundary layer over the structure surface is
well resolved by a fine mesh with the overall y less than
2, which eliminates the need for a boundary layer treatment
and the sub-viscous layer is resolved by the meshes [11].
The y values of the three sections for the present CFD
simulations are given in the Appendix.
3.2.3 Parallel processing introduction
Parallel computing with 32 processors was used for the
three-dimensional numerical simulations and the fluid domain was divided into 32 blocks, with the computations for
each block delegated to one of the 32 processors and with
one processor used as the master and the rest as the slaves.
During the iteration, the master and the slave nodes performed the mesh deformation for their own blocks using the
calculated structural responses. Solutions for both the
structural and the fluid analyses according to eq. (23) were
synchronized at each iteration loop.

Results and discussion

4.1

Flows around fixed sections

This analysis is to simulate the flow and aerodynamic forces


developed on fixed deck sections. Especially for G2, different angles of attack, 10, 8, 4, 0, 4, 8, 10, are
chosen. The computed aerodynamic forces of drag D, lift L
and moment M are expressed in the conventional
non-dimensional forms [28].
Cd

Figure 4 Enlargements of the inner regions for the two-dimensional


meshes (left-hand column) and the three-dimensional meshes (right-hand
column) used, with the densely populated circle being the rigid region.

Figure 5

Meshes along the thickness direction.

D
1
U 2 Bl
2

, Cl

L
1
U 2 Bl
2

, Cm

M
1
U 2 B 2 l
2

, (26)

where is the fluid density, taken as 1.185 kg/m 3 ; U is


the flow velocity; the stream flows from the front to the rear
of the sections for all of the simulations; for consistency
with the wind tunnel tests [28], the examples have a high
Reynolds number of 105, with U 1.545 m/s and B 1
m; Cd , Cl and Cm are the drag, lift, and moment coefficients, respectively; and l is the thickness of the deck sections (the thickness is 0.1 m for the two-dimensional models
and 1 m for the three-dimensional models). The time increment used was T = 0.002 s (i.e., approximately a fiftieth
of the time unit 0.14B/U).
At each time step the section surface pressure distribution was computed and integrated along the contour to form
the time traces of drag, lift, and moment. For example, Figure 6 shows partial simulated time traces for the aerodynamic force coefficients obtained from two-dimensional and
three-dimensional CFD simulations of section G1 with angle of attack 0. Figure 7 shows the computed aerodynamic

Bai Y G, et al.

Sci China-Phys Mech Astron

283

February (2013) Vol. 56 No. 2

Figure 6 Time traces showing evolution with the time step of the aerodynamic force coefficients for the fixed deck section G1 at 0 angle of
attack: (a) two-dimensional CFD simulation; (b) three-dimensional CFD
simulation.

force coefficients Cd , Cl and Cm of section G2 for different angles of attack using two-dimensional () and
three-dimensional () CFD modeling. Clearly there are
many differences between the two-dimensional and
three-dimensional values of section G2. Table 1 shows the
comparison of drag coefficient values among the present
CFD method, wind tunnel test and DVM method, and it is
obviously that the present three-dimensional CFD method
has better accuracy than two-dimensional ones.
Though most wind tunnel test results could not be obtained, it is reasonable to anticipate that the values using
three-dimensional CFD would be more accurate than those
using two-dimensional ones and that three-dimensional
CFD simulations have important practical significance for
blunt bodies.
The features of the flow field at different parts can be
shown visually, which is another advantage of three-dimensional CFD simulations. For example, the three-dimensional
wake flows for the three deck cross sections are shown pictorially in Figure 8. Hence it can be seen that section G1 has
the best aerodynamic stability, while section G3 has the
worst. Such visualization of wake through three-dimensional CFD simulations is of direct benefit for aerodynamic
analysis of structures.

Figure 7 Computed aerodynamic force coefficients versus angle of attack for the five generic deck sections: = three-dimensional CFD; =
two-dimensional CFD.

Table 1
section

Comparisons drag coefficient using different method for Tacoma

The present computed results


(two/three dimensional)
0.279/0.296

Wind
tunnel test
0.3

DVM
method
0.27

Figures 911 show the pressure contours of the three


sections (i.e., the upper part from two-dimensional simulations and the lower part from three-dimensional simulations). For the streamlined section G1, as shown in Figure 9,
the head contours are nearly the same but the downstream
ones have significant differences. For sections G2 and G3,
as shown in Figures 10 and 11, pressure distributions from
three-dimensional simulations are quite different from those
from two-dimensional simulations. From the contour of the
lower half of Figure 10, it can be found that around section
G2, the flow reattachment occurs at the middle of the deck,
and a large vortex region can be found at the rear part,
where the section has a concave angle. For the two kinds of
contours for section G3, though the flow separations in the
front have both been observed, the flow reattachment region

284

Bai Y G, et al.

Sci China-Phys Mech Astron

February (2013) Vol. 56 No. 2

Figure 10

Pressure contour of section G2.

Figure 11

Pressure contour of section G3.

Figure 8 Pictures of the three-dimensional wake flow for the three sections G1G3.

Figure 9

Pressure contour of section G1.

and the variation of vortex are obviously different. Therefore, only three dimensional numerical simulations can be
used to implement qualitative analysis of bluff body aerodynamics.
4.2

Computation of flutter derivatives

In order to evaluate the safety of a long-span bridge against


flutter instability, it is very important to accurately obtain
the flutter derivatives of the bridge deck. Usually, the flutter
analysis is carried out by using flutter derivatives obtained

from wind tunnel experiments on a scaled model of the


bridge deck [1,12]. There are two conventional methods for
flutter derivative identification. One is the free vibration
method [1,12,28], which is based on analysis of the variations of the apparent damping ratios and natural mechanical
frequencies of the structure when placed in free flow. The
alternative forced vibration method uses identification of
the motion-related fluid forces exerted on the structures, and
a formulation of the motion induced aerodynamic forces L
and moment M has been proposed which is suitable for
cross-sections in cross wind bending and twisting motion
[12]. Hence there has been much discussion on how to identify flutter derivatives accurately through wind tunnel tests,
e.g., that of Jones et al. [29].
The free vibration approach is easy to deploy in wind
tunnel laboratories, but suffers from poor quality, since
buffeting and vortex shedding factors may be mixed in ref.
[6]. In contrast, the forced vibration approach has strongly

Bai Y G, et al.

Sci China-Phys Mech Astron

and clearly defined input and output signals and so gives


flutter derivatives of high quality. However it depends on
actuation systems, which are expensive and hard to build.
For the CFD technique, the forced vibration method is
generally more convenient than the free vibration one, and
Larsen and Walther [28] proposed the expansions:

h
B
h
K 2 A3* K 2 A4* , (27)
M U 2 B 2 KA1* KA2*
U
U
B

h
B
h
K 2 H 3* K 2 H 4* , (28)
L U 2 B KH1* KH 2*
U
U
B

where K B /U is the dimensionless reduced frequency


of the motion; 2f ; f is the forced vibration frequency; h and h are the vertical cross wind motion and its
time derivative; and are the section rotation (twist)
in degrees and its time derivative; and A*j and H *j
( j 1, 2, 3, 4) are the flutter derivatives, which in general
are functions of K. The time increment T used was still
equal to 0.002 s. Another parameter needed is the dimensionless reduced velocity U * U /fB , for which this paper
uses the six values of 2, 4, 6, 8, 10 and 12 [28] when computing the dimensionless reduced frequency K.
Assume forced vibration of the form:

h(t ) h0 exp(it ), (t ) 0 exp(it ).

(29)

The motion induced forces are also assumed to be harmonic, with identical but a phase shift relative to the
motion. Replacing exp(i ) by ( cos i sin ) to determine the flutter derivatives from eqs. (27) and (28) gives

February (2013) Vol. 56 No. 2

A1*
A3*

285

M h (t ) sin
M (t ) sin
, A2*
,
2
2
h0 K U B
0 K 2 U 2 B 2

M (t ) cos
M (t ) cos
, A4* h 2
,
2
2 2
h0 K U 2 B
0 K U B

L (t ) sin
L (t ) sin
H h 2
, H 2* 2
,
2
h0 K U
0 K U 2 B

(30)

*
1

H 3*

L (t ) cos
L (t ) cos
, H 4* h 2
,
2
2
h0 K U 2
0 K U B

where L and M are, respectively, the section lift and


section moment caused by the forced twisting vibration; and
Lh and M h are, respectively, the section lift and section
moment caused by the forced vertical vibration. Thus,
A1* , A4* , H1* and H 4* can be computed from the forced
vertical motion, and A2* , A3* , H 2* and H 3* can be computed from the forced twisting motion.
To identify the flutter derivatives of the three sections for
zero angle of attack, forced motion simulations were conducted using the driving signal amplitudes of eq. (29) with
h(t ) 0.05 B sin t and (t ) 3sin t [28]. However the
flutter derivatives A4* and H 4* are not presented below
because wind tunnel results [28] are not available for them.
Each individual simulation was run for enough time increments for the simulated time traces of L and M to be
stable. The analysis of the simulations involved the leastsquares fitting of a sinusoid to the simulated L and M
time traces. An example of this procedure is shown in Figure 12, obtained for section G1 with 0=3 and U*=6.
Figures 1315 show the values computed by the proposed

Figure 12 Three-dimensional simulated motion-induced aerodynamic force time traces (----) and corresponding sinusoidal least-squares fit (solid curve),
for section G3 with 0=3 and U*=6.

286

Bai Y G, et al.

Sci China-Phys Mech Astron

February (2013) Vol. 56 No. 2

Figure 13 Comparison of the three-dimensional () and two-dimensional () CFD computed values for the flutter derivatives of section G1 with wind
tunnel test results () , computed values from ANSYS workbench () and the curve obtained via DVM.

Figure 14 Comparison of the three-dimensional () and two-dimensional () CFD computed values for the flutter derivatives of section G2 with wind
tunnel test results () , computed values from ANSYS workbench () and the curve obtained via DVM.

Bai Y G, et al.

Sci China-Phys Mech Astron

February (2013) Vol. 56 No. 2

287

Figure 15 Comparison of the three-dimensional () and two-dimensional () CFD computed values for the flutter derivatives of section G3 with wind
tunnel test results () and the curve obtained via DVM.

two-dimensional () and three-dimensional () CFD simulations. The results given by DVM (solid lines), ANSYS
workbench () and from the wind tunnel test () are also
given for comparison, noting that the wind tunnel test results are incomplete, and for section G3 none are available
for A1* and A3* . The meshes and parameters used by
ANSYS workbench are the same as those of the present
CFD simulations. However, ANSYS has its own mesh deformation algorithm which is different from that of sect.
3.2.1.
It can be seen that the present three-dimensional CFD
simulations mostly give better results than the DVM method
and ANSYS workbench, though sometimes all results are in
good agreement. For A2* , which is well-known to be a critical parameter for flutter [29], the present three-dimensional
CFD method has obtained exact results when compared to
wind tunnel results. It can be concluded that no matter if the
structure belongs to streamline body or bluff body, the present three-dimensional CFD method gives better predictions
for flutter derivatives of bluff bodies, which have relatively
poor aerodynamic stabilities than the DVM method and
ANSYS workbench.
4.3 Forced vibration amplitude influences on the flutter derivatives

Noda et al. [30] focused on the effects of forced vibration


amplitudes on the flutter derivatives of a thin rectangular
cylinder, and found that flutter derivatives identified using
small forced vibration amplitudes are quite different from
those using large forced vibration amplitudes, especially for
A2* and H 2* . Bai et al. [11] investigated this effect on an
airfoil, but found A2* and H 2* were not affected strongly

until the maximum amplitudes that Noda et al. [30] had


used had been reached. Here two increasing amplitudes are
chosen to study this effect which is related to dynamic stall
[11].
The computed values of the flutter derivatives for section
G2 with the twist amplitudes, 0 8 and 0 12 , are
plotted in Figure 16. The results of 0 3 are also shown
for comparison. All these results are from three-dimensional
CFD simulations.
There are not obvious differences between them, though
the maximum amplitudes that Noda et al. [30] used were
reached. It can be concluded that although there are inherent
flow separation around some bluff bodies like the deck section G2 used here, amplitude effects on the flutter derivatives are not significant if the amplitudes are not large
enough.

5 Conclusions
The main aim of the current study is to investigate the bluff
body aerodynamics with three-dimensional CFD modeling
at a high Reynolds number and proper turbulence models.
The results presented are encouraging and demonstrate the
accuracy of the proposed three-dimensional CFD method. It
has been shown that the present three-dimensional CFD
method is an effective numerical tool for evaluating the
aerodynamic stability instability of bluff bodies.
Three long-span bridge deck cross sections were investigated, using both two-dimensional and three-dimensional
CFD modeling. Comparisons of flutter derivatives were
given by the CFD method, along with those given by the
DVM method, the commercial software system ANSYS

288

Figure 16

Bai Y G, et al.

Sci China-Phys Mech Astron

February (2013) Vol. 56 No. 2

Influence of increasing forced vibration amplitudes. Computed results are shown for A*j and H *j , ( j 2, 3) for 0 = 3 (); 8 (); and 12

(). Wind tunnel test results () and the curve obtained via DVM are also shown for comparison.

workbench, and by the wind tunnel tests, and showed that


the present three-dimensional CFD method is better overall
than the DVM method and ANSYS workbench, especially
for bluff bodies with relatively poor aerodynamic stability.
There are some differences between the two-dimensional
and three-dimensional CFD simulation results in the computed aerodynamic force coefficients of section G2 for different angles of attack. The features of the flow field at different parts of the long-span deck cross section can be displayed pictorially via the three-dimensional CFD simulations and the results show that section G1 has the best aerodynamic stability, while section G3 has the worst. Finally,
an investigation with different forced vibration amplitudes
shows that amplitude effects on the flutter derivatives of
bluff bodies are not significant if the amplitudes are not
large enough.
Being treated as a hybrid RANS/LES model, DES was
successfully applied in the three-dimensional CFD simulations of bluff body with different aerodynamic stability, and
has shown significant benefits in efficiency and accuracy.
So it can be widely used to investigate engineering structure
aeroelasticity when turbulence effect is also included.

Appendix

This work was supported by the National Natural Science Foundation of


China (Grant No. 11172055) and the Foundation for the Author of National Excellent Doctoral (Grant No. 2002030).
1
2

7
8
9

The values of y+ obtained from both two-dimensional and


three-dimensional CFD simulations can be seen in Table A1.

10

11
Table A1 The values of y+ obtained from both two-dimensional and
three-dimensional CFD simulations

Section
Two-dimensional CFD
Three-dimensional CFD

G1
1.7655
1.8236

G2
1.6397
1.7214

G3
1.6027
1.9574

12
13

Nakamura Y. Recent research into bluff-body flutter. J Wind Eng Ind


Aerodyn, 1990, 33(1-2): 110
Mansor S, Passmore M A. Estimation of bluff body transient aerodynamics using an oscillating model rig. J Wind Eng Ind Aerodyn,
2008, 96(6-7): 12181231
Manzoor S, Hemon P, Amandolese X. On the aeroelastic transient
behaviour of a streamlined bridge deck section in a wind tunnel. J
Fluid Struct, 2011, 27: 12161227
Minh N N, Miyata T, Yamada H, et al. Numerical simulation of wind
turbulence and buffeting analysis of long-span bridges. J Wind Eng
Ind Aerodyn, 1999, 83(1-3): 301315
Larosea G L, DAuteuilb A. On the Reynolds number sensitivity of
the aerodynamics of bluff bodies with sharp edges. J Wind Eng Ind
Aerodyn, 2006, 94(5): 365376
Sarkar P P, Caracoglia L, Haan F L, et al. Comparative and sensitivity study of flutter derivatives of selected bridge deck sections: Part 1.
Analysis of inter-laboratory experimental data. Eng Struct, 2009, 31:
158169
Tamura T. Reliability on CFD estimation for wind-structure interaction problems. J Wind Eng Ind Aerodyn, 1999, 81: 117143
Tamura T, Itoh Y. Unstable oscillation of rectangular cylinder at various mass ratios. J Aerospace Eng, 1999, 12(4): 136144
Tamura T, Ono Y. LES analysis on aeroelastic instability of prisms in
turbulent flow. J Wind Eng Ind Aerodyn, 2003, 91: 18271846
Le Maitre O P, Scanlan R H, Knio O M. Estimation of the flutter derivatives of an NACA airfoil by means of Navier-Stokes simulation. J
Fluid Struct, 2003, 17: 128
Bai Y G, Sun D K, Lin J H, et al. Numerical aerodynamic simulations of a NACA airfoil using CFD with block-iterative coupling and
turbulence modeling. Int J Comput Fluid D, 2012, 26(2): 119132
Scanlan R H, Tomko J J. Airfoil and bridge deck flutter derivatives. J
Eng Mech Div ASCE, 1971, 97: 17171737
Fung Y C. An Introduction to the Theory of Aeroelasticity. New

Bai Y G, et al.

14

15

16

17

18

19

20

21

Sci China-Phys Mech Astron

York: Dover, 1993


Walther J H, Larsen A. Two dimensional discrete vortex method for
application to bluff body aerodynamics. J Wind Eng Ind Aerodyn,
1997, 67: 183193
Taylor I, Vezza M. Calculations of the flow field around a square
section cylinder undergoing forced transverse oscillations using a
discrete vortex method. J Wind Eng Ind Aerodyn, 1999, 82: 271291
Delgadillo J A, Rajamani R K. Computational fluid dynamics prediction of the air-core in hydrocyclones. Int J Comput Fluid D, 2009,
23(2): 189197
El Gharbi N, Absi R, Benzaoui A, et al. An improved near-wall
treatment for turbulent channel flows. Int J Comput Fluid D, 2011,
25(1): 4146
Matthies H G, Steindorf J. Partitioned but strongly coupled iteration
schemes for nonlinear fluid-structure interaction. Comput Struct,
2002, 80: 19911999
Lin J H, Sun D K, Sun Y, et al. Structural responses to non-uniformly
modulated evolutionary random seismic excitations. Commun Numer
Methods Eng, 1997, 13: 605616
Sun D K, Zhi H, Zhang W S, et al. Highly efficient and accurate buffeting analysis of complex structures. Commun Numer Methods Eng,
1998, 14: 559567
Spalart P R, Jou W H, Strelets M, et al. Comments on the Feasibility
of LES for Wings, and on a Hybrid RANS/LES Approach. Advances
in DNS/LES. Colombus: Greyden Press, 1997. 137147

22

23

24

25
26

27

28

29

30

February (2013) Vol. 56 No. 2

289

Strelets M. Detached eddy simulation of massively separated flows.


In: 39th AIAA Aerospace Sciences Meeting and Exhibit. Reno:
AIAA, 2001
Spalart P R, Squires K D. The Status of Detached-Eddy Simulation
for Bluff Bodies. Direct and Large Eddy Simulation V. Berlin:
Springer, 2004
Nallasamy M. Turbulence models and their applications to the prediction of internal flows: A review. Comput Fluid, 1987, 15(2): 151
194
Wilcox D C. Reassessment of the scale-determining equation for advanced turbulence models. AIAA J, 1988, 26(11): 12991310
Versteeg H K, Malalasekera W. An Introduction to Computational
Fluid DynamicsThe Finite Volume Method. Harlow: Longman
Scientific & Technical, 1995
Ghosal S, Moin P. The basic equations for large eddy simulation of
turbulent flows in complex geometry. J Comput Phys, 1995, 118:
2437
Larsen A, Walther J H. Discrete vortex simulation of flow around
five generic bridge deck sections. J Wind Eng Ind Aerodyn, 1998,
77-78: 591602
Jones N P, Scanlan R H, Li Q C. Discussion on measuring flutter
derivatives for bridge sectional models in water channel. ASCE J
Eng Mech, 1996, 122(4): 389390
Noda M, Utsunomiya H, Nagao F, et al. Effects of oscillation amplitude on aerodynamic derivatives. J Wind Eng Ind Aerodyn, 2003,
91(1-2): 101111

S-ar putea să vă placă și