See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/289347277

3D CFD Simulation of Vortex-induced Vibration of Cylinder

Article  in  International Journal of Offshore and Polar Engineering · September 2011

CITATIONS READS
9 464

3 authors, including:

Fabio Saltara J.I.H. Lopez

University of São Paulo Universidade Presbiteriana Mackenzie
20 PUBLICATIONS   482 CITATIONS    8 PUBLICATIONS   14 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

MICRO AXIAL HYDRAULIC TURBINE PROJECT WITH EMPHASIS ON RUNNER AND WICKET GATES PROFILES CALCULATION View project

Stability calculations View project

All content following this page was uploaded by J.I.H. Lopez on 14 August 2017.

The user has requested enhancement of the downloaded file.

International Journal of Offshore and Polar Engineering (ISSN 1053-5381) http://www.isope.org/publications
Copyright © by The International Society of Offshore and Polar Engineers
Vol. 21, No. 2, June 2011, pp. 1–6

3D CFD Simulation of Vortex-induced Vibration of Cylinder

F. Saltara
Department of Mechanical Engineering, University of São Paulo, São Paulo, Brazil
A. D’Agostini Neto
VOITH Hydro, São Paulo, Brazil
J. I. H. Lopez
Department of Mechanical Engineering, Mackenzie Presbyterian University, São Paulo, Brazil

The 3D flow around a circular cylinder free to oscillate transversely to the free stream was simulated using Computational
Fluid Dynamics (CFD) and the Spalart-Allmaras Detached Eddy Simulation (DES) turbulence model for a Reynolds number
Re = 104 . Simulations were carried out for a small mass-damping parameter m∗  = 000858, where m∗ = 33 and  = 00026.
We found good agreement between the numerical results and experimental data. The simulations predicted the high observed
amplitudes of the upper branch of vortex-induced vibrations for low mass-damping parameters.

INTRODUCTION A better alternative for practical engineering simulations seems

The simulation of the vortex-induced vibration (VIV) phe-
nomenon is a subject that has been defying CFD experts. The
vortex wake of the high Reynolds number flow around bluff bod-
ies is in general turbulent and 3D, requiring huge computational
resources and fine meshes to accurately capture its physical behav-
ior. As was pointed out by Shur et al. (2005), even when the
geometry suggests that 2D simulations should be successful, as is
the case with the flow around a circular cylinder, the lack of 3D
effects generates results that are not in 2D experiments. For this
reason, we can hardly expect 2D RANS simulations to produce
meaningful results for higher Re numbers.
Due to the difficulties, the vast majority of CFD VIV simula-
tions of the flow around circular cylinders that can be traced in the
literature are concerned with low Reynolds flows. For Re < 180
the wake is laminar and 2D, simplifying the numerical approach
and avoiding the need of expensive hardware. But, as pointed out
by Williamson and Govardhan (2004), these simulations in gen-
eral predict peak vibration amplitudes around 0.6D (where D is
the cylinder diameter) when experiments for higher Re produce
amplitudes well over 1.0D. The usefulness of low Re simulations
is arguable, once the flow around oil risers and other structures is
in general in the range 104 < Re < 106 .
Some of the few 3D simulations of the VIV phenomenon for
higher Re numbers can be found in the works of Dong and Kar-
niadakis (2005) and Dong et al. (2006). The first work presents
results of DNS simulations of the flow past a stationary cylinder
and a cylinder undergoing forced harmonic oscillations using a
spectral element method at Re = 10000; the second work presents
similar DNS simulations and PIV measurements of the flow past
a stationary cylinder at Re = 4000 and 10000. Results of the sim-
ulations are in close agreement with the experiments, but the use
of the spectral element method to perform DNS required large
computational resources.
Received September 21, 2010; revised manuscript received by the editors
January 27, 2011. The original version was submitted directly to the
Journal.
KEY WORDS: Vortex shedding, vortex-induced vibration, elastic cylin-
der, riser, CFD.
to be to rely on Large Eddy Simulation (LES) or hybrid RANS-
LES methods, such as Detached Eddy Simulation (DES). Accord-
ing to Shur et al. (2005), these methods are better suited for the
simulation of the flow around bluff bodies because they are less
dissipative than RANS, and so do not damp the development of
3D turbulent structures in the wake. For this reason we decided
to carry out simulations of the 3D flow around a circular cylin-
der free to vibrate in the transverse direction using the Spalart-
Allmaras DES model, from Spalart (2000), for a Reynolds num-
ber Re = 104 .
It can be argued that the VIV of a cylinder is strongly affected
by the vibrations in the streamwise direction, and that 2 degrees
of freedom VIV simulations would be more interesting for practi-
cal applications than simulations with the cylinder free to vibrate
only in the transverse direction. When experimentally studying
2 degrees of freedom VIV for mass parameters m∗ < 60, Jau-
vtis and Williamson (2004) found what they called a super-upper
branch of cylinder response, with the streamwise vibration deeply
affecting the modes of vortex shedding and the transverse vibra-
tions, which resulted in cross-flow VIV amplitudes near 1.5D.
Dahl et al. (2007) found through PIV visualizations that when the
inline and transverse natural frequencies are close to the 2:1 ratio,
the motion of the structure near resonance is strongly affected,
and the 2P and 2S modes of vortex shedding seen in transverse-
only oscillations are replaced by more complex patterns consist-
ing of vortex triplets and quintuplets that cause high-amplitude
harmonics in the lift force.
The study of 2 degrees of freedom VIV is more relevant, from
an engineering point of view, than the transverse-only VIV, but it
is a more complex subject. For this reason we decided to focus
the present simulations on the simpler problem of transverse-only
VIV in order to validate CFD procedures and acquire a better
assessment of mesh requirements, turbulence models and compu-
tational resources, and to leave 2 degrees of freedom VIV simu-
lations for future work.
So, for our 3D simulation of the VIV response of a cylinder free
to vibrate in the transverse direction at Re = 104 we used a low
mass-damping parameter, with m∗  = 000858, where m∗ = 33
and  = 00026. The mass parameter m∗ is equal to the mass of
the cylinder divided by the mass of displaced fluid, m∗ = m/md ,
2 3D CFD Simulation of Vortex-induced Vibration of Cylinder
Fig. 1 Cylinder free to vibrate in transverse direction
where md = L
D2 /4,  is the density of the fluid, and L is
The turbulent stresses are related to the rate of deformation tensor
through the eddy viscosity:
t = fv1 ˜ (7)
The eddy viscosity distribution is given by the solution of the
1-equation turbulence model of Spalart and Allmaras, modified
to be a hybrid RANS-LES model capable of solving the largest
turbulent eddies. Hybrid RANS-LES models are generally called
Detached Eddy Simulation (DES) models, and simulations with
such models are believed to be less costly than LES simulations.
According to Spalart (2000), the transport equation for the eddy
viscosity is given by:

2
˜ ūj 
˜ ˜
the length of the cylinder. Our results were compared with the + = cb1 S̃ ˜ −cw1 fw +
t xj d˜
experimental results of Khalak and Williamson (1997) for the  
same values of m∗ and . 1   ˜ cb  ˜  ˜
+ +˜ + 2 (8)
xj xj xj xj
NUMERICAL METHOD
The model is closed through the following constants and auxiliary
The commercial code FLUENT was adapted for the VIV sim- relations:
ulations through a User Defined Function (UDF). The code has
a Finite-Volume CFD algorithm with the ability to solve the cb1 = 01355! cb2 = 0622! cv1 = 71! = 2/3 (9)
unsteady flow around a body with prescribed motion through
cb1 1 + cb2
moving mesh techniques. Although the code has only 1st-order cw1 = + ! cw2 = 03! cw3 = 2! " = 041 (10)
time accuracy for moving mesh problems, we managed to recover "2
the 2nd-order accuracy through the UDF. The function also cal-   16
#3 # 1 + cw6 3
culates the forces exerted by the flow on the body, which is con- fv1 = 3 ! fv2 = 1 − ! fw = g (11)
# + cv31 1 + #fv1 g 6 + cw6 3
sidered a spring-mass-damper system (Fig. 1).
With the forces calculated through the pressure and viscous ˜ ˜
#= ! g = r + cw2 r 6 − r! r= (12)
stress distributions, Eq. 1 is solved for the body acceleration:  S̃"2 d˜2


˜ 1  ūi  ūj
mÿt + c ẏt + kyt = Fy t (1) S̃ = S + fv2 ! S= 2&ij &ij ! &ij = − (13)
"2 d˜2 2 xj xi
With the calculated acceleration, Eqs. 2 and 3 allow the calcula- d˜ = mind' CDMS  (14)
tions of the body displacement and velocity:
Eq. 14 is the fundamental equation of the DES modification pro-
1 posed by Spalart (2000). In this equation, d is the distance of the
yt + t = yt + ẏtt + ÿtt 2 (2)
2 computational cell to the nearest wall, and  is the largest dimen-
sion of the cell. CDMS is a constant equal to 0.65. When the size
3 1
ẏt + t = ẏt + ÿtt − ÿt − tt (3)
2 2
The velocity of the body obtained through Eq. 3 is used as a new
boundary condition on the body surface for the following time
step. The incompressible, unsteady and turbulent flow is solved
through a fractional step method for the velocity-pressure cou-
pling, assuring that the velocity field satisfies continuity:
 ūj
=0 (4)
xj
Central differences are used for the diffusive and convective fluxes
of the transport equations, assuring 2nd -order space accuracy. The
momentum equation for a turbulent flow is given by the Reynolds
Averaged Navier-Stokes (RANS) equation:


of the computational cell becomes small, the destruction term in
the eddy viscosity transport equation (Eq. 8) is increased, with
the effect of lowering the turbulent diffusion. The model behaves
as a subgrid model, allowing the development of larger eddies. If
cells are large, the normal RANS behavior is recovered.
SIMULATION PARAMETERS
The simulations were carried out with a mesh with 615,360
cells. The flow domain extended 10D upstream, down and above
the cylinder and 20D downstream, with a 5D span in the cylinder
length direction. The length of the cylinder circumference was
divided in 160 cells. There were 48 cells along the span of the
cylinder. Mesh was coarsened outside the wake region in order
to lower computational requirements. A general view of the mesh
can be seen in Fig. 2.
ūi  ūj ūi   p̄   ūi  
The nearest cells to the wall of the cylinder have a height of
+ =− +  − ui uj (5) 0.001D. A view of the discretization in the boundary layer region
t xj xi xj xj
can be seen in Fig. 3.
According to Wilcox (1994), the turbulent stresses in Eq. 5 are In Fig. 4 we can see the boundary conditions. On the front,
given by the Boussinesq approximation: top and bottom boundaries a free-stream condition was imposed,

 with a uniform velocity U and the eddy viscosity equal to the
   ūi  ūj molecular viscosity. The values of the velocity U , diameter D
−ui uj = t + (6)
xj xi and fluid properties  and  were chosen in order to have a
International Journal of Offshore and Polar Engineering, Vol. 21, No. 2, June 2011, pp. 1–6
Fig. 2 General view of mesh with one side of domain, cylinder
wall and outflow boundary
3
The cylinder was considered initially at rest. Simulations were
carried out for a total of 8000 time steps (around 40 periods of
vortex-shedding), and all statistics were calculated using the sec-
ond half of the total simulation time in order to avoid influence
of the initial conditions.
A main issue addressed by our simulations is related to the
development of transition in the wake. This is a problem aris-
ing when DES is used for simulations of bluff body flows, as
can be seen in the works of Haase et al. (2009) and Shur et al.
(1996). For Re = 104 we have a laminar separation flow. DES
treats the near-wall region through the RANS model, and for a
laminar separation flow the low Reynolds damping functions are
used to assure the development of the laminar boundary layer.
When using the Spalart-Allmaras DES model, Shur et al. (1996)
proposed 2 approaches to trigger the development of wake turbu-
lence: To use a zero inlet eddy viscosity and
˜ a non-zero arbitrary
initial condition, or to use a low value of inlet eddy viscosity. We
chose the second approach, setting = ˜  at the inlet.
RESULTS
First of all simulations with a fixed cylinder were carried out to
validate the proposed methodology. Very good agreement between
experiments and simulations was obtained. A summary of the
results can be seen in Table 1.
We compared the mean drag coefficient CDMean with the results
from Roshko (1961), and the random mean square value of the
lift coefficient CLrms and the Strouhal number St with results from
Norberg (2003).
The time histories of the drag and lift coefficients CD and CL
as a function of the nondimensional time Ut/D can be seen in
Fig. 5.
A visualization of the flow around the fixed cylinder can be
seen in Fig. 6. The picture shows isosurfaces of the Q parame-
ter given by Q = 05&2 − S 2 , where & is the modulus of the

Fig. 3 Detail of mesh near cylinder wall, showing boundary layer

region Simulation Experiments
CDMean 1.18 1.10–1.20 Roshko (1961)
Reynolds number Re = 104 . On the cylinder wall there is no tur-
CLrms 0.46 0.41 Norberg (2003)
bulence, and the velocity is given by the body velocity calculated
St 0.21 0.20 Norberg (2003)
through Eq. 3. Neumann boundary conditions were employed on
the outflow boundary, and periodic conditions were employed on Table 1 Comparison between simulation and experimental
the side boundaries located at the extremes of the domain span. resultstab1
The mesh moved as a rigid body with the cylinder, and an Arbi-
trary Lagrangian-Eulerian (ALE) formulation was used to calcu- 2
late convective fluxes. In this type of formulation, the mass fluxes CD
on the surface of the computational cells are calculated using rel- 1.5 CL
ative velocities.
A nondimensional time step Ut/D = 0025 was used, which 1

means that there were around 200 time steps per period of vortex-
0.5
shedding.
0

-0.5

-1

-1.5

-2
0 50 100 150 200
Ut/D

Fig. 5 Time histories of drag coefficient CD and lift coefficient

Fig. 4 Boundary conditions used for simulation CL for fixed cylinder case
4 3D CFD Simulation of Vortex-induced Vibration of Cylinder
Fig. 6 Visualization of wake of flow around fixed cylinder, show-
ing isosurfaces of Q = 05
Fig. 7 Contours of turbulent viscosity ratio t / for Ut/D = 200
and z/D = 0 for stationary cylinder case
vorticity and S is the norm of the strain-rate tensor. The visualiza-
tion of the Q parameter is useful to verify if the mesh refinement
allowed the capture of the vortex structures of the wake.
A plot of the turbulent viscosity ratio t / for Ut/D = 200
and z/D = 0 can be seen in Fig. 7. We can see the development
U 2
U
CD
6
CL
Y
-2
-4
0 50 100 150 200
Ut/D
Fig. 9 Time histories of drag coefficient CD , lift coefficient CL
and nondimensional displacement Y for reduced velocity Vr = 65
the mesh region solved through the LES approach. The regions
are defined by Eq. 14, with the turbulent length scale of the DES
Spalart-Allmaras model given by the distance to the nearest wall d
in the RANS region and given by CDMS  in the LES region. The
RANS region is entirely concentrated near the cylinder wall and
independent of time. This is due to the simplicity of the Spalart-
Allmaras model, which employs the distance to the nearest wall
d as the RANS turbulent length scale, resulting in regions defined
solely by the mesh refinement. If we had used a DES 2-equation
turbulence model such as the k − . SST, the RANS turbulent
length scale would be defined using the time-dependent k and .
values, resulting in time-varying RANS and LES regions.
After the simulations of the flow around the fixed cylinder were
carried out, we decided to simulate the cylinder free to oscillate
transversely to the free-stream. We carried out a series of simula-
tions in the range of reduced velocities 2 ≤ Vr ≤ 14. The reduced
velocity is related to the free-stream velocity; the diameter of the
cylinder and the natural frequency in water of the spring-mass-
damper system through:
2
Vr = = (15)
of the wake turbulence after the laminar separation. In Fig. 8 we .n D k/m + md D
can see the mesh region solved through the RANS approach, and
For each reduced velocity it was possible to obtain not only the
lift and drag time histories, but also the time history of the nondi-

Khalak&Williamson(1997)
Simulations
1

0.8

Yamp 0.6

0.4

0.2

0
0 2 4 6 8 10 12 14 16
Vr

Fig. 8 Mesh region solved by RANS approach (black painted Fig. 10 Amplitude of vibration Yamp plotted as a function of
cells) and by the LES approach (white painted cells) reduced velocity Vr
International Journal of Offshore and Polar Engineering, Vol. 21, No. 2, June 2011, pp. 1–6 5

8 200
CDmean - Khalak&Williamson (1997)
7 CDmax - Khalak&Williamson (1997) 175
CDmean - Simulations
CDmax - Simulations 150
6

125
5
θ (°) 100
4
75
3
50

2
25

1
0
3 4 5 6 7 8 9 10 11 12

0
Vr
0 2 4 6 8 10 12 14 16
Vr Fig. 13 Phase angle between lift force coefficient and displace-
ment of cylinder as a function of reduced velocity
Fig. 11 Maximum and mean drag coefficient plotted as a function
of reduced velocity Vr
4

fs/fn
mensional displacement of the cylinder Y = y/D. In Fig. 9 we can 3.5
Stationary Cylinder Shedding Frequency / fn
see the time histories of lift, drag and displacement of the cylin- 3
der for Vr = 65, the reduced velocity that produced the highest
amplitude of vibration. Through the analysis of the time histories 2.5

for each simulation it was possible to plot the amplitude of vibra- f/fn 2
tion and vortex-shedding frequency as a function of the reduced
velocity. In Fig. 10, we can see the plot of the nondimensional 1.5

amplitude of vibration Yamp = yamp /D. 1

In Figs. 11 and 12 we can respectively see the results for
maximum and mean drag coefficient and maximum and rms lift 0.5

coefficient.
0
From the results of the amplitude of vibration, we can see that 0 2 4 6 8 10 12 14 16
Vr
it was possible to capture the high amplitude of vibration char-
acteristic of the so-called upper branch. From the experiments of Fig. 14 Nondimensional frequency of vortex-shedding fs /fn plot-
Khalak and Williamson (1997), the highest obtained amplitude ted as a function of reduced velocity Vr: fs = frequency of vortex-
was Yamp = 108 for Vr = 58. In the numerical simulations, the shedding. fn = system natural frequency .n /2

highest obtained amplitude was Yamp = 106, for Vr = 65.

The main discrepancies between experiments and simulations
ment. We can see from Fig. 13 that in the simulations the phase
are related to the overprediction of values of amplitude of dis-
placement, mean and maximum drag coefficient and rms and max-
imum lift coefficient for Vr > 60. Another discrepancy is related
to the phase angle between lift coefficient and cylinder displace-
5 we obtained amplitudes around 1.0D for 55 ≤ Vr ≤ 70, after the
CLrms - Khalak&Williamson (1997)
CLmax - Khalak&Williamson (1997)
4 CLrms - Simulations
CLmax - Simulations
3
angle 2 shifted from 2  0 to 2  180 for a reduced velocity
Vr = 55. Khalak and Williamson (1999) found that this shift was
related to the transition between the higher amplitude Yamp  10D
of the upper branch and the lower amplitude Yamp  06D of the
lower branch, and it occurred for Vr  60, but in the simulations
phase angle shift. From the amplitude results of the simulations,
we would expect to detect the phase angle shift for Vr  65–70.
According to Williamson and Govardhan (2004), together the
upper and lower branch comprise the so-called lock-in region,
where the frequency of vortex-shedding is captured by the natural

0
0 2 4 6 8 10 12 14 16
Vr
Fig. 12 Maximum and rms lift coefficient plotted as a function of Fig. 15 Flow visualization for reduced velocity Vr = 65, showing
reduced velocity Vr isosurfaces of Q = 05 colorized by z-vorticity
 6 3D CFD Simulation of Vortex-induced Vibration of Cylinder
frequency of the system. In Fig. 14, we can see that the lock-in
region of the simulations extends from Vr = 5 to Vr = 11.
In Fig. 15 we can see a flow visualization for the reduced veloc-
ity Vr = 65. We can see that the flow is characterized by what
seems to be a 2P mode of vortex-shedding.
In Fig. 15, we used the z component of the vorticity to colorize
the isosurfaces of Q. The black vortices correspond to negative z-
vorticity, and the gray vortices correspond to positive z-vorticity.
Looking to the vortices that are near the body, it is apparent that
there is a 2P mode of vortex-shedding.
CONCLUSIONS
3D simulations of the flow around a circular cylinder free
to vibrate transversely to the free-stream using DES turbulence
modeling were able to predict reasonably well the main fea-
tures of the VIV phenomenon for a low mass-damping parame-
ter m∗  = 000858 and Reynolds number Re = 104 . Although the
simulations produced some overprediction of the amplitudes of
displacement, maximum and mean drag coefficient and maximum
and rms lift coefficient for Vr > 60, in general the simulation
results are in good agreement with the experimental results of
Khalak and Williamson (1997). The results are appealing because
the majority of VIV CFD simulations that can be found in the
literature are related to low Re numbers and 2D flows.
A possible explanation for the differences between experiments
and simulations can be related to the mesh refinement and tur-
bulence model. As a continuation for this research, the use of a
simpler model like the Smagorinsky LES model with a finer mesh
can be an alternative.
ACKNOWLEDGEMENTS
This research was sponsored by the Foundation of Support to
Scientific Research of the São Paulo State in Brazil, FAPESP,
under Grant 2005/03050-2.
REFERENCES
Dahl, JM, Hover, FS, and Triantafyllou, MS (2007). “Resonant
Vibrations of Bluff Bodies Cause Multivortex Shedding and
High Frequency Forces,” Phys Rev Letters, Vol 99, October.
View publication stats
Dong S, and Karniadakis, GE (2005). “DNS of Flow past a Sta-
tionary and Oscillating Cylinder at Re = 10000,” J Fluids and
Struct, Vol 20, pp 519–531.
Dong S, Karniadakis GE, Ekmekci A, and Rockwell, D (2006).
“A Combined Direct Numerical Simulation-Particle Image
Velocimetry Study of the Turbulent Near Wake,” J Fluid Mech,
Vol 569, pp 185–207.
Haase, W, Braza, M, and Revell, A (2009). “DESider—A Euro-
pean Effort on Hybrid RANS-LES Modelling,” Notes on Numer
Fluid Mech and Multidisciplinary Design, Vol 103.
Jauvtis, N, and Williamson, CHK (2004). “The Effect of Two
Degrees of Freedom on Vortex-induced Vibration at Low Mass
and Damping,” J Fluid Mech, Vol 509, pp 23–62.
Khalak, A, and Williamson, CHK (1997). “Fluid Forces and
Dynamics of a Hydroelastic Structure with Very Low Mass and
Damping,” J Fluids and Struct, Vol 11, pp 973–982.
Khalak, A, and Williamson, CHK (1999). “Motions, Forces and
Mode Transitions in Vortex-Induced Vibrations at Low Mass-
Damping.“ J Fluids and Struct, Vol 13, pp 813–851.
Norberg, C (2003). “Fluctuating Lift on a Circular Cylinder:
Review and New Measurements,” J Fluids and Struct, Vol 17,
pp 57–96.
Roshko, A (1961). “Experiments on the Flow Around a Circular
Cylinder at Very High Reynolds Number,” J Fluid Mech, No
10, pp 345–356.
Shur, M, Spalart, PR, Strelets, M, and Travin, A (1996). “Navier-
Stokes Simulation of Shedding Turbulent Flow past a Circular
Cylinder and a Cylinder with a Backward Splitter Plate,” Third
ECCOMAS CFD Conf, Paris, John Wiley & Sons, pp 676–682.
Shur, M, Spalart, PR, Squires, KD, Strelets, M, and Travin, A
(2005). “Three Dimensionality in Reynolds-Averaged Navier-
Stokes Solutions Around Two-Dimensional Geometries.“ AIAA
J, Vol 43, pp 1230–1242.
Spalart, PR (2000). “Strategies for Turbulence Modeling and Sim-
ulations,” Int J Heat and Fluid Flow, No 21, pp 252–263.
Wilcox, DC (1994). Turbulence Modeling for CFD, DCW, Inc.
Williamson CHK, and Govardhan, R (2004). “Vortex Induced
Vibrations,” Ann Rev Fluid Mech, No 36, pp 413–455.