Sie sind auf Seite 1von 7

Numerical simulation of vortex-induced vibration of a

circular cylinder at low mass and damping with


different turbulent models
Wei Lia, b, Jun Lia, Shengyu Liua
a

School of Naval Architecture and Ocean Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
Hubei Key Laboratory of Naval Architecture and Ocean Engineering Hydrodynamics, Huazhong University of Science and
Technology, Wuhan 430074, China
E-mail address: hustliw@hust.edu.cn (Wei Li), hustleejun@hust.edu.cn (Jun Li)

AbstractDue to the great damage to widely utilized flexible


structures in ocean engineering, vortex-induced vibration (VIV)
of such long flexible marine structures is still a hot issue that
needs more theoretical research, and CFD techniques become
gradually indispensable to study the VIV problem. In this paper,
two-dimensional Reynolds-averaged Navier-Stokes (RANS)
equations are adopted to investigate transverse VIV of elastically
mounted rigid cylinder with low mass-damping, and two typical
turbulent models are applied to solve the RANS equations: RNG
k- model and SST k- model. By comparing the cylinder
displacement response and vortex shedding modes of three
different response branches, analysis of differences between two
turbulence models are presented. The numerical results indicate
that SST k- model is more appropriate for VIV of the elastically
mounted rigid cylinder. Subsequently, other hydrodynamic
coefficients obtained by SST k- model are discussed and
compared with previous research in detail. This investigation
provides theoretical evidence for the numerical simulation of
VIV of marine riser in engineering application.
Keywordsvortex-induced vibration (VIV); vortex shedding;
RANS; turbulent model; riser

I.

INTRODUCTION

Vortex-induced vibration (VIV) involved in a great many


fields of engineering, especially in subsea pipelines, flexible
risers and other marine structures. As flexible materials are
increasingly utilized in the deep-sea, VIV of such crucial
structures with low mass-damping renews much attention to
research recently, which produced a large number of
fundamental studies [1,2,3,4,5].
The case of an elastically mounted rigid cylinder
constrained to oscillate transversely to a free stream is one of
the most foundational research in the field of VIVs. And
Williamsons group have made a significant contribution in
this respect with a series of classical experimental studies
[6,7,8,9], where the relevant conclusion of the experiments
could be summarized as follows: for an elastically mounted
rigid cylinder with low mass-damping, three distinct branches
of amplitude response for the transverse oscillation were
observed, which were denoted the initial branch, the upper
branch and the lower branch, respectively. The transition

978-1-4799-3646-5/14/$31.00 2014 IEEE

between different branches were associated with two phase


angles: total phase and vortex phase, which were defined as
the phase angle between relevant hydrodynamic coefficient and
the cylinder response. Additionally, the modes of vortex
formation were also well present in their papers. Those
conclusions presented in Williamsons research have a
significant value for the future studies.
Recently, with the fast development of computer,
numerical calculation become one of most important methods
to solve the VIV problems and many researchers come to
utilize the computational fluid dynamics (CFD) techniques to
do the modeling, which primarily include three different
computational approaches to solve the turbulent properties of
the flow: Reynolds-averaged Navier-Stokes (RANS), Direct
Numerical Simulation (DNS) and Large Eddy Simulation
(LES). Since the RANS is robust enough and can acquire a
simulated result with relative accuracy under much less timeconsuming compared with other two approaches, it has a wide
range of application into this field. Pan and Cui [10] used
RANS code to simulate a two-dimensional numerical model
based on the experiment by Khalak and Williamson [6]. The
vortex modes and transition between different branches were
consistent with the experimental results, but the response
amplitudes obtained in the simulation were much lower than
the experimental results, especially the maximum response
amplitude occurred in only one timestamp, which seems
irregular compared with the experimental results. Meanwhile it
was pointed out that the loss of the spanwise correlation and
the low value of mass-damping may bring about the random
disturbance and characteristics of the vortex-shedding process
with the vortex shedding.
It is well-known that the SST k- model and RNG k-
model, basing on different hypothesis about the flow field, are
the most common turbulent models in solving RANS
equations. The main objective of this paper is to assess a
relatively accuracy turbulent model for an elastically mounted
rigid cylinder with low mass-damping constrained to oscillate
transversely. Then the numerical simulations based on the
turbulent model chosen are presented for our case. The
physical parameters are referred to the experiment of Khalak

and Williamson[6]: m* = 2.4 and m* = 0.013 ; Meanwhile


the reduced velocity ( U * = U inlet / ( f n D ) ) is changed from 2.0 to
13.9 with increasing velocity, and the corresponding Reynolds
number is from 1700 to 11600. During the whole numerical
process, the results are always compared with the experimental
data provided by Khalak and Williamson [6].

II. THEORETICAL FORMULATION


The unsteady incompressible RANS equations can be
written as follow:

ui
=0
xi

(1)

2 Sij u j ' ui '


( ui ) + ( ui u j ) = +
t
x j
xi xi

(2)

Where u, p represents the time-average value of the velocity


and pressure, respectively; is the molecular viscosity; Sij is
the mean stress tensor; and u j ' ui ' is the Reynolds stress tensor,
which all can be solved by means of an Newtonian model as
follow:

u u j
u j ' ui ' = t i +

x j xi

2
u
k + t i ij
3
xi

(3)

Here, the eddy viscosity t is given by the turbulence model;


ij is the Kronecker delta; and the turbulent kinetic energy
k can be represented as

u 'u ' 1
k = i i = u ' 2 + v' 2
2
2

(4)

Meanwhile, considering a riser oscillating transversely to a


free stream, to simplify the flow into two dimensions, strategy
known as the strip theory approach is utilized to combine
with the CFD method, in which 2-D CFD models first are
carried out for a number of strips along the riser and loads are
applied to the structure later for the dynamic analysis.
Furthermore, each strip can be regarded as an elastically
mounted rigid cylinder and analyzed as a linear spring-massdamper vibration system of single freedom. Fig.1 illustrates the
simplification process.


y+

4
2 CL
2
y + * y =
*
m*
U
U

(5)

Where U* is the reduced velocity; is the system structural


damping ratio; m* is the mass ratio; y is the nondimensional
transverse cylinder displacement; and CL is the lift coefficient.
Once the vortex shedding frequency was fully locked onto
the oscillation frequency of cylinder, which referred to the
lock-in region. Following formulas prevailed in the relevant
research as [11, 12]

y = A*sin ( ex t )

(6)

CL = CL 0 sin (ex t + )

(7)

CL = CLv cos (ex t ) + CLa sin (ex t )

(8)

CLv = CL 0 sin , CLa = CL 0 cos

(9)

= arctan ( CLv / CLa )

(10)

Where CL0 CLv CLa are the amplitude of the lift coefficient and
its components in phase with velocity and acceleration,
respectively; ex is the oscillation frequency and represented
as the phase angle between the lift coefficient and response
displacement.
By Parkinson [13], the response amplitude ratio
( A* = Ymax / D ) and the frequency ratio ( f * = f ex / f n ) can be
derived as
A* =

1 CLv f n U

4 3 m* f ex f n D

(11)
1

2 2
1 C
U
*
La
f = 1 +

3 * *
2 m A f n D

(12)

These equations originated from the linearization with energy


balance between the structure and the surrounding fluid. Once
the response amplitude ratio and frequency ratio were obtained,
CLv and CLa could be calculated by the above two equations,
as well as the phase angle .
Furthermore, by Lighthill [14], the total fluid force F acting
on the cylinder can be separated into a potential force
component Fpotential, given by the potential added-mass force,
and a vortex force component Fvortex that is due to the
dynamics of what is called the additional vorticity. Similarly
following equations were set up after nondimensional disposal:
Cvortex = CL 0 C potential

(13)

Cpotential could be derived as:


Fig. 1. the simplification process of a 3D riser

Thus the nondimensional motion equations generally used


to express the VIV of the cylinder in the transverse y-direction
could be

C potential = 2 3

A*

(U

/ f *)

(14)

Besides, by using vortex coefficient the nondimensional


equation of motion can be written as follow:


y+

4
2 Cvortex
2
sin (ex + vortex )
y + * y =
*
m* + ma
U
U

(15)

Where ma is the potential added-mass corresponding to the


potential coefficient; is the system structural damping
ratio that containing the potential added-mass; and vortex is the
phase angle between the vortex coefficient and response
displacement, which could be solved by the similar method
given above.
To obtain the nondimensional response displacement for
the transverse oscillation, the Newmark- approach is utilized
to solve the above differential motion equation, and the force
coefficient is obtained by solving the two-dimensional RANS
equations.
III. GRID MODEL
The whole computational field in this investigation is
discretized as shown in Fig.2(a), which is a rectangle with 20D
in transverse and 30D in lengthways. The inflow boundary of
the domain is located on the left hand side of the cylinder at a
distance of 10D from the center of the cylinder, and
consequently the right side of the domain the outflow boundary
is defined. The symmetry boundary is located 10D away from
the center of the cylinder in upward and downward directions
respectively, and the no-slip condition is located on the
cylinder surface. The hybrid meshing is used, shown in
Fig.2(b), and stretching of the mesh is performed to achieve a
fine resolution of the region closed to the cylinder surface.
This mesh domain is large enough to enable obtaining
oscillation of the wake down-stream of the cylinder. The first
points of the mesh away from the cylindrical face are located
where y + 0.5 for each reduced velocity considered.

(b)

10D

10D

20D

(c)
Fig. 2. Computational field and grid model

y
10D
x

U0
(a)

Grid independence tests are performed with three cases of


different total grids, under the condition of reduced velocity U*
can be 6.0, where the response amplitude ratio A* and
frequency ratio f* can be obtained. Relevant testing results,
including comparing with the experimental ones, are listed in
Table I.
TABLE I.

NUMERICAL RESULTS OF RESPONSE AMPLITUDE AND


FREQUENCY WITH DIFFERENT GRID RESOLUTIONS
( m* = 2.4 m = 0.013 U * = 6.0 )
*

Wall
grid
120
120
120

Nearfield
grid
6000
6000
6000

Total
grid

A*
num

A*
exp

f*
num

f*
exp

8260
10620
11460

0.556
0.570
0.573

0.583
0.583
0.583

1.09
1.15
1.17

1.23
1.23
1.23

From Table I, it shows that there are good consistence


between the results of response amplitude and frequency

obtained by different grid resolutions, which means the


simulation in this research has good grid independence. Thus,
latter investigation is based on the second grid scheme which
makes the simulation more efficient.
IV. RESULTS AND DISCUSSION
Before presenting results, some numerical parameters need
to be specified. The nondimensional time step, U t / D , is set
to be 0.005 in the calculations. For each time step, a reduction
of nonlinear residuals for the discrete momentum equations is
required, and a two orders of magnitude of nonlinear residuals
of discrete momentum equations is established for the
reduction, which is exactly same with Guilmineau and
Queutey [15]. The divergence of the velocity field is decreased
as 10-5.
A. Comparing between SST k- model and RNG k- model
Two typical turbulent models were adopted to solve the
RANS equations: SST k- model and RNG k- model. By
using of a k- formulation, SST k- model has an advantage in
near wall treatment and it could be used as a Low-Re
turbulence model without any extra damping functions. On the
other hand, based on the standard k- model, the eddy viscosity
is modified and the rotating flows are taken into consideration
in RNG k- model, which made it much more appropriate to
simulate the flow field around a circular cylinder. Theoretically
SST k- model seems much appropriate to the problem of flow
around circular cylinder. In order to figure out a better one,
VIV of an elastically mounted rigid cylinder under designated
typically reduced velocities are investigated. The response
amplitude ratio A*, obtained by different turbulent models, as a
function of reduced velocity U* are given in Fig.3. Also the
results are compared with the experimental ones, which are
denoted by solid circular.

(a) SST k- model

(b) RNG k- model


Fig. 3. Amplitude responses with different turbulent model ( m* = 2.4 )

Fig.3 shows that both two turbulent models can get three
response branches. From Fig.3(a), it is obvious that transition
between the initial branch to upper branch happened around
where reduced velocity U* being 3.5, and when reduced
velocity U* reaches 5.2, the lower branch occurred. Also the
maximum amplitude ratio reached 0.747 in the upper branch,
while it was 0.642 in the lower branch. After the reduced
velocity U* reach 11.0, the response amplitude of cylinder
becomes much lower than before, which corresponds to the
lock-out region.
However, for the result of RNG k- model, the transition
timestamp between the initial branch to upper branch is a bit
delayed, and transition between the upper branch to lower
branch seems not quiet instinctive. Moreover, the maximum
amplitude ratio is 0.713, which is lower than one obtained by
SST k- model. Although the response amplitudes are all less
than the experimental ones for both two turbulent models,
which may due to the simplification of two-dimensional and
other factor caused by turbulence model, the comparison
results still indicate that SST k- model agreed better with the
experimental results than that of RNG k- model.
Besides, according to Govardhan and Williamson [16], the
modes of vortex formation differ between each response
branch. A 2S mode (two single vortices shed per cycle) turned
up in the initial branch, while a 2P mode (two pair vortices
shed per cycle) is observed in both the upper and lower
branches, merely the intensity of the second vortex of each pair
in the upper branch differ from the other branch.
Corresponding to some typical reduced velocity above,
Contour maps of vortices in different timestamp during a
shedding cycle are shown in Fig.4-5.

y/D

y/D

y /D

y /D

-1
-1

-1

-1

-2

-2

-2

10

-2

x/D
0

-3

10

x/D

y/D

y/D

y/D

-1

-1

-2

-2

-2

10

-3

10

x/D

10

-2

y /D

-1
1

y/D

10

-2

-3

-3

10

10

y /D

-1

-1

-1

-2

-2

-3

10

10

x/D

(b) U = 5.0
3

10

-2

10

10

x/D

-1

-1

-2
2

-1

-2

-2

-3

10

10

10

y /D

-1

-1

-2

-2

-3

10

(c) U * = 7.5
Fig. 4. Vorticity magnitude contours in a vortex shedding period (SST k-)

y /D

-1

-1

x/D

10

-2

10

-2

x/D

Then, when the reduced velocity U* comes to 5.0, which


represents the upper response branch, a typical 2P vortex
formation mode is successfully simulated by SST k- model in
Fig.4(b), which is completely in accord with the experimental
result by Williamsons. This is due to the good behavior of
SST k- model in adverse pressure gradients and separating
flow. However, the vortex formation mode acquired by RNG
k- model is totally not consistent with the experimental one.

x/D

x/D

In the case of reduced velocity U* being 3.0, which referred


to the initial branch. The mode of vortex formation acquired by
SST k- model in Fig.4(a) seems like not a 2S mode but a P+S
mode (two pair vortices shed on one side while one single
vortex shedding on the other during per cycle), which is
considered as a transient, unsteady-state pattern. While a 2S
vortex formation mode obtained by RNG k- model is shown
in Fig.5(a), from where the vortex follows a pattern similar to
what is found in a classical Von Karman Street.

x/D

Fig. 5. Vorticity magnitude contours in a vortex shedding period (RNG k-)

(c) U = 7.5

x/D
3

-1

x/D

y/D

y/D

x/D

x/D

y/D

10

-1

y /D

y /D

y/D

y/D

-2

-2

x/D

(b) U = 5.0

-2

x/D

-1

-1

x/D

y /D

-2

x/D

y /D

y/D

-2

-3

-1

-3

10

x/D

(a) U * = 3.0

-3

x/D

y /D

-1

x/D
0

10

-1

-3

(a) U * = 3.0

10

x/D

x/D

y /D

-3

x/D

10

Finally, when the reduced velocity U* is 7.5, which


represents the lower response branch, it seems that both two
turbulent models do not simulate 2P mode successfully in the
relevant figures.

Consequently, compared to RNG k- model, the maximum


response amplitude obtained by SST k- model agreed better
with the experimental result, and the 2P wake mode in the
upper branch was successfully simulated by SST k- model,
which indicated SST k- model is more appropriate for VIV of
the elastically mounted rigid cylinder. Thus more discussion
about VIV will be based on SST k- model subsequently.
B. Response frequency and force coefficient
Generally, besides the response amplitude ratio, response
frequency ratio is also a significant parameter to assess the
numerical results. Here, the response frequency ratio f*,
obtained by SST k- model, as a function of reduced velocity
U* is given in Fig.6, which also compared with the
experimental ones.
Fig.6 shows that the response frequency in the lower
branch agrees well with the experimental ones. In the lock-in
region, the oscillation frequency of cylinder fex separates from
the vortex shedding frequency fst, and the frequency ratio goes
smoothly from 1.03 to 1.29. On the other hand, when the
cylinder is out of lock-in region, the oscillation frequency fex
equals the vortex shedding frequency fst , which is similar to
the previous research.
Fig.7 presents the lift and drag coefficients for some
selected cases, as well as the nondimensional displacement
response. At U * = 3.0 (Fig.7(a)), time traces of force
coefficients are extremely irregular, and it is distinct that the
multi-frequency vibrations occurred, which is consistent with
the experimental result in the initial branch region. At U * = 5.0 ,
it turns into the lock-in region, time traces of force coefficients
are regularly periodic and the amplitude of the lift coefficient
also becomes larger. At U * = 7.5 , the force coefficient decrease
a bit due to the lower branch, and it has an obvious decrease
when the reduced velocity U* equals 11.0, which indicates that
the system have turned into the lock-out regime.

Fig. 6. Response frequency ratio responses with U ( m* = 2.4 SST k-)

(a)

(b)

(c)

(d)

Fig. 7. Time traces of cylinder displacement, lift and drag coefficient


( m* = 2.4 , SST k-)

C. Vortex coefficient and vortex phase


As mentioned earlier, once the values of A* and f* were
obtained, the lift coefficient CL, vortex coefficient Cvortex, total
phase angle , and the vortex phase angle vortex can all
deduced through relevant equations. Here omitting the process
of calculate, the corresponding force coefficients and phase
angles are directly presented at Fig.8.
From Fig.8 we can figure out that in the initial branch the
vortex phase vortex is closed to 0D , which means that vortex
coefficient Cvortex and potential coefficient Cpotential are in
phase, so in the initial branch the total lift coefficient CL
achieves a large value. Then in the upper branch although
Cvortex and Cpotential are out of phase because of vortex closing
to 180D , the large response amplitude in this branch causes a
large potential force, which makes Cpotential much larger than
Cvortex , and still leading to a large total lift coefficient CL .
Finally, Cvortex and Cpotential are quite comparable and still out
of phase in the lower branch, so CL becomes quite diminished.
Furthermore, from the former research, there is a distinct jump
to both two phase angle: the total phase jumps at the
transition between initial-upper branches, while the vortex
phase vortex jumps at the transition between upper-lower
branches. The results from Fig.8 quite coincide with
Govardhan and Williamson [16].

[7]

[8]

[9]

[10]
*

Fig. 8. Relevant force coefficients and phase angles varing with U


( m* = 2.4 SST k-)

[11]

V. SUMMARY AND CONCLUSIONS


In this paper, two-dimensional RANS equations were used
to calculate VIV of an elastically mounted rigid cylinder at low
mass-damping constrained to oscillate transversely, and two
typical turbulent models were adopted and compared to solve
the RANS equations: RNG k- model and SST k- model. The
comparison results between two turbulent models show
significant differences between two turbulent models: the
maximum response amplitude obtained by SST k- model
agreed better with the experimental result than that of RNG k-
model, and the 2P wake mode in the upper branch was
successfully simulated by SST k- model.
Subsequently, transverse VIV for our case is extensively
investigated based on SST k- model. The response frequency,
hydrodynamic coefficients, and relevant phase angles are in
good agreement with the experimental results on the whole.
These results suggest that SST k- model is valid and effective
for the VIV of cylinder with low mass-damping.
With the fast development of computer and computational
technique, the CFD method becomes much indispensable to
study in the ocean engineering. The present research provided
theoretical evidence for the computation of VIV of marine riser
in engineering application. And the 3-D RANS issues are
essential to be discussed in the further research.

REFERENCES
[1]
[2]

[3]

[4]

[5]

[6]

P. W. Bearman, Circular cylinder wakes and vortex-induced vibrations,


Journal of Fluids and Structures, vol 27, pp. 648-658, 2011
P. Catalano, M. Wang and G. Iaccarino, Numerical simulation of the
flow around a circular cylinder at high Reynolds numbers, International
Journal of Heat and Fluid Flow, vol 24, pp. 463-469, 2003
D. M. F. Gao and C. G. Mingham, Numerical and experimental
investigation of turbulent flow around a vertical circular cylinder, 20th
International Offshore and Offshore and Polar Engineering Conference
Proceedings, pp. 639-643, 2010
A. Elbanhawy and A. Turan, On two-dimensional predictions of
turbulent cross-flow induced vibration: Forces on a cylinder and wake
interaction, Flow, Turbulence and Combustion, pp. 199-224, 2010
C. M. Larsen and K. H. Halse, Comparison of models for vortex
induced vibrations of slender marine structures, Marine Structures, vol
10, pp. 413-441, 1997
A. Khalak and C. H. K. Williamson, Dynamics of a hydroelastic
cylinder with very low mass and damping, Journal of Fluids and
Structures, vol 10, pp. 455-472, 1996

[12]

[13]
[14]
[15]

[16]

C. H. K. Williamson, Advances in our understanding of vortex


dynamics in bluff body wakes, Journal of Wind Engineering and
Industrial Aerodynamics, pp. 3-32, 1997
A. Khalak and C. H. K. Williamson, Motions, forces and mode
transitions in vortex dynamics in bluff body wakes, Journal of Wind
Engineering and Industrial Aerodynamics, vol 13, pp. 813-851, 1999
R. Govardhan and C. H. K. Williamson, Modes of vortex formation
and frequency response of a freely vibrating cylinder, Journal of Fluid
Mechanics, pp. 85-130, 2000
Z. Y. Pan, W. C. Cui and Q. M. Miao, Numerical simulation of vortexinduced vibration of a circular cylinder at low mass-damping using
RANS code, Journal of Fluids and Structures, vol 23, pp. 23-37, 2007
J. G. Wissink and W. Rodi, Numerical study of the near wake of a
circular cylinder, International Journal of Heat and Fluid Flow, vol 29,
pp. 1060-1070, 2008
P. Catalano, M. Wang and G. Iaccarino, Experimental and numerical
studies of the flow over a circular cylinder at Reynolds numbers,
International Journal of Heat and Fluid Flow, vol 24, pp. 463-469, 2008
G. V. Parkinson, Mathematical models of flow-induced vibrations of
bluff bodies, Flow-Induced Structural Vibrations, pp. 81-127, 1974
J. Lighthill, Fundamentals concerning wave loading on offshore
structures, Journal of Fluid Mechanics, pp. 667-681, 1986
E. Guilmineau and P. Queutey, Numerical simulation of vortexinduced vibration of a circular cylinder with low mass-damping in a
turbulent flow, vol 19, pp. 449-466, 2004
R. Govardhan and C. H. K. Williamson, Mean and fluctuating velocity
fields in the wake of a freely-vibrating cylinder, Journal of Fluid and
Structures, vol 15, pp. 85-130, 2000

Das könnte Ihnen auch gefallen