2-D URANS Vs - Experiments of Flow Induced Motion Softw o Circular Cylinders in Tandem With Passive Turbulence Control For 30,000oreo105,000

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Ocean Engineering 72 (2013) 429440

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

2-D URANS vs. experiments of ow induced motions of two circular


cylinders in tandem with passive turbulence control
for 30,000 oReo105,000
Lin Ding a,b, Michael M. Bernitsas b,c,d,n, Eun Soo Kim b,c
a

College of Power Engineering, Chongqing University, Chongqing 400044, China


Marine Renewable Energy Laboratory, Dept. of Naval Architecture & Marine Engineering, University of Michigan, 2600 Draper Road, Ann Arbor,
MI 48109-2145, United States
c
Department of Mechanical Engineering, University of Michigan, MI, United States
d
CTO of Vortex Hydro Energy, Ann Arbor, MI, United States
b

art ic l e i nf o

a b s t r a c t

Article history:
Received 10 January 2013
Accepted 2 June 2013
Available online 15 August 2013

The ow induced motions (FIM) of two rigid circular cylinders, on end linear-springs, in tandem are
studied using two-dimensional Unsteady Reynolds-Averaged Navier-Stokes (2-D URANS) simulations
veried by experimental data. Passive turbulence control (PTC) is being used in the Marine Renewable
Energy Laboratory (MRELab) of the University of Michigan to enhance FIM of cylinders in the VIVACE
(Vortex Induced Vibration for Aquatic Clean Energy) Converter to increase its efciency and power
density in harnessing marine hydrokinetic energy. Simulation is performed using a solver based on the
open source CFD tool OpenFOAM, which solves continuum mechanics problems with a nite-volume
discretization method. The simulated Reynolds number range for which experiments were conducted in
the MRELab is 30,000 oReo 105,000, which falls in the TrSL3 regime (Transition in Shear Layer), where
the shear layers are fully saturated and consequently lift is high. The amplitude and frequency results are
in excellent agreement with experimental data showing the initial and upper branches in VIV, transition
from VIV to galloping, and galloping. Vortex structures are studied using high-resolution imaging from
the CFD results showing typical 2S structure in the initial branch and both 2P+2S and 2P in the upper
branch of VIV. In the galloping branch, amplitudes of 3.5 diameters are reached before the channel stops
are hit.
& 2013 Elsevier Ltd. All rights reserved.

Keywords:
Two cylinders
URANS
Flow induced motions
Vortex induced vibrations
Galloping
Passive turbulence control
VIVACE Converter
Surface roughness
Hydrokinetic energy

1. Introduction
Elastically mounted, rigid, circular cylinders exposed to uid ow
perpendicular to their axis experience ow induced motions (FIM)
excited by the alternating vortices shed in the cylinder wake and
forming the von Krmn street. Vortex shedding occurs over the
entire range of Reynolds numbers (Re) with the exception of very
low Reo40, the Tritton (1977) transitions region (200oReo400),
and the laminar to turbulent ow transition. The cylinder would be
excited to signicant amplitudes when the frequency of the vortex
shedding mode locks onto the vibration frequency, thus synchronizing the natural frequency and the excitation frequency. For a smooth
or rough cylinder, the oscillatory lift forces on the body lead to
vortex-induced vibration (VIV). When the cylinder is not rotationally

n
Corresponding author at: Department of Naval Architecture & Marine Engineering, University of Michigan, Ann Arbor, MI 48109-2145, United State. Tel.: +1 734
764 9317; fax: +1 734 936 8820.
E-mail addresses: linding@cqu.edu.cn (L. Ding),
michaelb@umich.edu (M.M. Bernitsas).

0029-8018/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.oceaneng.2013.06.005

symmetric, for example by using turbulence stimulation strips,


galloping may be induced as shown experimentally (Chang et al.,
2011; Kim et al., 2011; Lee and Bernitsas, 2011). VIV and galloping
are the most commonly observed FIM phenomena. A comprehensive review of research on VIV can be found in the article by
Williamson and Govardhan (2004).
FIM is typically treated as a destructive phenomenon because
of the fatigue damage it may cause. The effective control of vortex
shedding is important in engineering applications. Unlike previous
efforts to alter vortex shedding and suppress the occurrence of
FIM, Bernitsas et al. (2008) and Lee and Bernitsas (2011) have been
successful in utilizing this potentially disastrous phenomenon
to generate power with the VIVACE (Vortex-Induced Vibration
for Aquatic Clean Energy) Converter. The VIVACE Converter is a
hydrokinetic power generating device invented by Bernitsas and
Raghavan in 2005 (Bernitsas and Raghavan, 2009) and further
developed by the Marine Renewable Energy Laboratory (MRELab)
at the University of Michigan (Bernitsas et al., 2009; Lee et al.,
2010, 2011; Lee and Bernitsas, 2011; Raghavan and Bernitsas,
2011). The simplest form of VIVACE is a single cylinder suspended

430

L. Ding et al. / Ocean Engineering 72 (2013) 429440

by springs with a power-take-off (PTO) system. It can harness


hydrokinetic energy from ocean and river currents as slow as
0.4 m/s 0.8 knots (Chang et al., 2011). The goal of the VIVACE
team is to enhance the oscillation amplitude and maximize the
hydrokinetic energy converted to mechanical energy in the oscillating cylinder. One way to improve the performance of VIVACE is
to use multiple cylinders as would be the case in multi-blade
propellers or windmills. Two rigid circular cylinders in tandem
mounted on end linear-springs with passive turbulence control
(PTC) to enhance FIM are studied in this paper.
Roughness on the cylinder can effectively change the ow
properties. Extensive literature is available on using roughness to
alter FIM of cylinders on springs. There are different roughness
parameters that affect ow-induced motion, such as roughness
location, roughness height, and roughness coverage (Chang et al.,
2011; Park et al., 2012). PTC was introduced in the MRELab to
enhance cylinder FIM and extract more hydrokinetic energy from
uid ows. PTC consists of selectively located surface roughness
with thickness on the order of the boundary layer thickness; and
depending on its location it can induce galloping, hard galloping,
weak suppression, or strong suppression as shown in the FIM-toPTC Map (Park et al. 2012). With the application of PTC, cylinder FIM
can be enhanced. In addition, back-to-back VIV and galloping are
achieved. The maximum power density of a single-cylinder VIVACE
(349 W/m3) was amplied 1.38 times in comparison to that of
VIVACE with a smooth surface cylinder (253 W/m3) at ow speed
U 1.45 m/s (Chang et al., 2011). Amplitudes as high as 2.7 diameters
have been achieved by using passive turbulence control (Chang
et al., 2011; Kim et al., 2011; Raghavan and Bernitsas, 2008). The
effects of PTC were studied in detailed by Chang et al. (2011) and
Park et al. (2012).
To further improve the power density of VIVACE, multiple
cylinder systems are investigated experimentally in the MRELab.
Multiple cylinder systems are used in many applications in civil,
offshore, aeronautical engineering, etc. The interference between
cylinders strongly depends on the arrangement of cylinders and
their orientation with respect to the free stream (Zdravkovich,
1997b). Two-cylinder systems have been studied the most because
they are the simplest multi-cylinder arrangement (Assi et al., 2006;
King and Johns, 1976; Sumner et al., 2000; Zdravkovich, 1985, 1987).
For two cylinders in tandem, the downstream cylinder is subjected
to high level of turbulence generated from the upstream cylinder in
addition to impingement of Krmn-size shed vortices. Most of
studies performed in the past on two-cylinder arrangements were
on smooth cylinders. Moreover, in most studies, the cylinders were
xed or at very low Reynolds number (Borazjani and Sotiropoulos,
2009). FIM of two-cylinders with surface roughness (PTC) for high
Re has been studied only by the MRELab to the best of the authors
knowledge (Kim et al., 2011).
In this paper, two rigid PTC-cylinders in tandem mounted on endsprings are simulated using two-dimensional Unsteady ReynoldsAveraged Navier-Stokes (URANS) equations with the SpalartAllmaras
one-equation turbulence model. The ow is simulated in the range of
30,000oReo105,000, which falls in the high-lift TrSL3 regime, and
for which experiments were conducted in the MRELab. TrSL stands
for Transition in Shear Layer and 3 indicates the third region where
the shear layer is fully saturated resulting in stronger vortices, shorter
formation length, and highest lift (Zdravkovich, 1997a). There are
numerous studies of using URANS for simulation of ow past a
circular cylinder. From the published literature, URANS results of the
Strouhal number agree very well with other numerical and experimental results. Lift and drag coefcient CFD results at low Reynolds
numbers (Wanderley et al., 2008) also agree well with experiments.
Researchers mostly apply URANS at low Reynolds number. Applications at higher Re show that prediction for Re412,000 is still a
challenging task for URANS. Prediction is even poorer near the drag

crisis (Catalano et al., 2003). As explained by Wu et al. (2011), the rst


manifestation of failure lies in the fact that for Re410,000 the
separation point is not predicted properly. Specically, CFD using
2-D URANS predicts that the separation point hardly oscillates around
901 while experimental data show that it oscillates around 811 in
laminar ow with amplitudes as much as 5101. This is a most
important characteristic of ows past a circular cylinder. It is also a
local property of the ow as opposed to integral ow properties such
as the Strouhal number and the lift/drag forces. Some integral
properties are easier to predict as integration lters local errors.
With proper modeling of PTC, however, 2-D URANS simulations
exhibit several of the salient local features of the ow resulting is
excellent agreement with experiments as proven by Wu et al.
(2011). They developed a CFD code based on OpenFOAM to solve
the problem of a single cylinder with PTC. They showed that the
presence of PTC results in very good agreement between experiments and CFD simulations up to Re135,000 for which experimental data were available from tests in the MRELab. Without PTC
such agreement was limited to Re 10,00012,000 (Wanderley
et al., 2008; Wu et al., 2011) when 2-D URANS is used.
Thus, the code developed by Wu et al. (2011) for a single
cylinder in FIM and in this paper for two cylinders in tandem
predict very well the experimentally measured data including
vortex streets, transition from VIV to galloping, and shear layer
oscillation. Consequently, the developed tool can be used with
condence to predict ow properties that are more challenging to
measure experimentally at such high speeds and turbulence levels.
In the present study, the FIM of two rigid circular cylinders, on end
linear-springs, in tandem are studied using 2-D URANS simulations
veried by experimental data. The objective of this study is to
establish the capability of a numerical tool to simulate the VIVACE
system with two PTC-cylinders in FIM and investigate the system
parameter effects on the cylinder dynamics. The physical model and
running parameters are presented in Section 2. In Section 3, the
numerical approach and grid generation are described. The simulation
results of amplitude and frequency for the two PTC-cylinders are
shown in Sections 4 and 5, respectively. Numerical results are
compared with experiments conducted in the Low Turbulence Free
Surface Water (LTFSW) Channel of the MRELab. Vortex structures of
four typical cases are discussed in Section 6. Conclusions are presented
at the end based on the analysis of amplitude and frequency response
and vortex structures.

2. Physical model
The physical model considered in this paper consists of two
oscillatory systems as depicted in Fig. 1. The elements of each
oscillatory system are a rigid circular cylinder of diameter D and
length L, two supporting linear springs of stiffness K, and the

Fig. 1. Schematic of the physical model.

L. Ding et al. / Ocean Engineering 72 (2013) 429440

system damping c due to friction. Two cylinders arranged in


tandem are constrained to oscillate in the y-direction, which is
perpendicular to the ow velocity direction (x). The center-tocenter distance d, between the two cylinders is set at 2D. Two
straight roughness strips are attached to the surface of each
cylinder symmetrically, one on each side of the cylinder (Chang
et al., 2011). The angle PTC, is measured in degrees from the
forward stagnation point in the corresponding ideal ow. The
coverage provided by each sand-strip is 161.
In the present study, simulations are veried by experimental
measurements of the ow induced motion of two circular cylinders with PTC in tandem. The system parameters in the 2-D
URANS simulation are the same as those used in the corresponding experiments in the MRELab, as listed in Tables 1 and 2. The
stiffness of the springs and the system damping are measured
using a series of free-decay tests in air, where linear viscous
damping was assumed. All the experiments were conducted in the
LTFSW Channel located in MRELab. Details on the LTFSW Channel
are provided by Bernitsas et al. (2009).
The test-section of the channel is 1 m wide and 0.8 m deep. The
ratio of cylinder diameter D to channel depth is about 12%. The ratio
of cylinder length L to channel width w is nearly 1. Analysis of four
potential blockage effects: (a) side-to-side blockage, (b) top-tobottom blockage (c) free-surface effect, and (d) bottom-boundary
effect are discussed by Chang et al. (2011). The last two are studied
in detail in Raghavan (2007) and Raghavan et al. (2009).
Passive turbulence control (PTC) is being used in the MRELab of
the University of Michigan to enhance FIM of cylinders in the

Table 1
Nomenclature.
Apeaks
Ca
Cd
Cl
D
K
L
P
Re
St
T
T1,n 1/fn,water
U
U nair U/(fn,airD)
U nwater U/(fn,waterD)
cstructure
charn
c cstrucure+charn
D
p
fn,water K=mosc ma =2
p
fn,air K=mosc =2
fosc
K
md
ma Camd
mosc
mn mosc/md
P
W
y(t)
y+
PTC
z

t
v
~

431

Table 2
Physical model parameters.
Item
Diameter
Length
Oscillating system mass
Spring const.
Damping ratio of system
Damping
Natural freq. in water
Natural freq. in air
Mass ratio
Added mass coef.
Displaced mass
Added mass

D [m]
L [m]
mosc [kg]
K [N/m]

c [N s/m]
fn,water
fn,air
m*
Ca
md [kg]
ma [kg]

First cylinder

Second cylinder

0.0889
0.91441
9.5121
758.11
0.0161
2.7274
1.1246
1.4209
1.6774
1
5.6707
5.6707

0.0889
0.9144
9.5756
726.84
0.017
2.8434
1.0989
1.3866
1.6886
1
5.6707
5.6707

VIVACE Converter to increase its efciency and power density in


harnessing marine hydrokinetic energy. The strips with roughness
designation P60 have been used as PTC for the research in this
study. All modeling parameters of PTC are dened in Fig. 2 (Chang
et al., 2011). The strips are attached running along the entire
length of the cylinder parallel to the cylinder axis. Waterproof
sandpaper strip is cut into specic width which covers 161 of the
surface of the circular cylinder. The strip thickness is about equal
to the thickness of the boundary layer and affects profoundly FIM.
The FIM-to-PTC Map developed by Park et al. (2012) shows the
effect of selective surface roughness in the form of strips on the
FIM of circular cylinders. Table 3 shows the details of the roughness strip P60 used in this study.

3. Mathematical and numerical modeling


Mean amplitude of the peaks
Added mass coefcient
Drag coefcient
Lift coefcient
Cylinder diameter
Spring constant
Cylinder length
Thickness of sand paper
Reynolds number
Strouhal number
Total thickness of PTC
Natural period in water for the 1st cylinder
Mean ow velocity
Reduced velocity in air
Reduced velocity in water
Structural damping
Added damping to harness energy
Total damping of system
Center-to-center distance of cylinders
System natural frequency in water
System natural frequency in air
Oscillating frequency of cylinder
Average height of sandpaper grit
Displaced uid mass
Added mass
Oscillating system mass
Mass ratio
Pressure
Channel width
Displacement of cylinder
Nondimensional rst grid spacing
PTC placement angle
Damping ratio of system
Angular coverage of strip
Turbulent eddy viscosity
Kinematic molecular viscosity
Intermediate working variable
Density of the uid

In this section, the mathematical modeling for the uid


dynamics and the two oscillators is provided rst. The integration
scheme, the computational domain, the grid generation, and the
computational time are presented as well.
3.1. Governing equations
The mathematical model consists of the uid dynamics equations, the turbulence model for the uid, and the body dynamics
equations. Those are described in the following subsections.
3.1.1. Fluid dynamics
In the present study, two-dimensional URANS ow simulations
are performed by developing a solver built into the open source CFD
tool OpenFOAM to predict ow properties past two circular cylinders with PTC. The cylinders are rigidly supported by two end linearsprings and allowed a single degree of freedom motion transversely
to the ow direction. OpenFOAM is a collection of C++ library
subroutines that are developed for solving continuum mechanics
problems with the nite-volume discretization method. The ow is
assumed to be two-dimensional and unsteady, and the uid is
incompressible. The uid ow is modeled using the Unsteady
Reynolds-Averaged Navier-Stokes (URANS) equations together with
the one-equation SpalartAllmaras (SA) turbulence model. The
basic URANS equations are:
U i
0
xi

U i

1 p

U U

2Sij ui uj
xj i j
xi xj
t

where is the molecular kinematic viscosity and Sij is the mean

432

L. Ding et al. / Ocean Engineering 72 (2013) 429440

Fig. 2. Conguration of the passive turbulence control (PTC) on the cylinder (Chang et al., 2011).

3.1.3. Oscillator dynamics


The dynamics of the two oscillators is modeled by the classical
linear oscillator model

Table 3
PTC Parameters (P60 sand paper).
Item
Strip placement angle
Angular coverage of strip
Sand paper thickness
Average grit height
Total thickness of strip

PTC [degree]
[degree]
P [mm]
k [mm]
T P+k [mm]

First cylinder

Second cylinder

20
16
0.587
0.26
0.847

30
16
0.587
0.26
0.847

strain-rate tensor.


1 U i U j
Sij

2 xj
xi

and Ui is the mean ow velocity vector. The quantity ij ui uj is


known as the Reynolds-stress tensor. In order to solve the URANS
equations for the mean-ow properties of the turbulence ow, the
Boussinesq eddy-viscosity approximation is employed to relate the
Reynolds-stress to the mean velocity gradients as

mosc y cy_ Ky f t

where mosc is the total oscillating mass of cylinder and attachments including 1/3 of the spring mass, c is the linear viscous
damping, and K is the linear spring constant.
It should be noted, however, that there is signicant difference
between the mathematical modeling of damping in Eq. (9) and the
real physical damping in the oscillators used in the experiments.
This difference is more pronounced in low oscillator speeds. Using
extensive system identication, the damping model in the physical
oscillators was found by Lee et al. (2011) to be
4

f uSTEP jy_ n jthres y_ n uSTEP thres jy_ n j ak f nk

10

k1

3.1.2. Turbulence model


The SpalartAllmaras (SA) turbulence model is a one-equation
model, which solves a transport equation for the kinematic eddy
viscosity. This model has been shown to give acceptable results for
a wide variety of situations and is known for its stability. Several
modications of the SA model exist (Allmaras et al., 2012; Aupoix
and Spalart, 2003; Edwards and Chandra, 1996), but the original
model (Spalart and Allmaras, 1994) is employed in this work. In
the SpalartAllmaras model, the turbulent eddy viscosity is computed from

where the velocity threshold is thres 0.001, uSTEP is the unit stepfunction, y_ n is a symbolic representation of the nonlinear static
dependence of the friction force upon the current velocity, and ak
is a coefcient determined experimentally as explained by Lee
et al. (2011). This damping model is capable of predicting well the
VIV response even in low oscillator velocity for the virtual damper/
spring VIVACE system in the experiments (Lee and Bernitsas,
2011). Lee et al. (2010) also showed that at low oscillation speeds
discrepancies exist between experiments conducted with real
springs/dampers and experiments conducted with a virtual system using only linear viscous damping. Low oscillator speeds exist
at the beginning of the initial branch in VIV and near the end of
VIV in the desynchronization range. This is observed also in the
results in this paper since the experiments were conducted with
springs/dampers while the CFD oscillator modeluses the classical
linear viscous damping model in Eq. (9).

t ~ f 1

3.2. Integration scheme

ui uj 2t Sij

where the t is turbulence eddy viscosity.

where
f 1

3
3 c31

6
7

~ is an intermediate working variable of the turbulence model and


obeys the following transport equation
 2
 


~
~
~
1
~
~ ~
uj
cb1 S~ ~ cw1 f w

~
cb2
d
t
xj
s xj
xj
xi xi
8
Additional denitions of functions and constants are given by
Spalart and Allmaras (1994). The trip terms ft1 and ft2 are turned
off and the trip-less initial condition (Shur et al., 1996; Travin
et al., 2000) for ~ , which was successfully used in earlier work for a
single circular cylinder (Wu et al., 2011), is used in this study.

A second-order Gauss integration scheme with a linear interpolation for the face-centered value of the unknown is used for the
divergence, gradient, and Laplacian terms in the governing equations. The second-order backward Euler method is adopted for
time integration. Thus, the numerical discretization scheme gives
second order accuracy in space and time. A pressure implicit with
splitting of operators (PISO) algorithm is used for solving momentum and continuity equations together in a segregated way. The
equations of motion for the two cylinders are solved using a
second-order mixed implicit and explicit time integration scheme.
3.3. Computational domain
The computational domain is 52D  9D for the two PTC-cylinders.
As shown in Fig. 3, the entire domain includes ve boundaries:
inow, outow, top, bottom, and the two cylinder walls. The distance
between the inlet boundary and the center of 1st cylinder, lup, is set

L. Ding et al. / Ocean Engineering 72 (2013) 429440

Fig. 3. Computational domain.

at 25D. The downstream length of the domain, ldown, is also set at


25D. The inow velocity is considered as uniform and constant
velocity. At the outow boundary, a zero gradient condition is
specied for velocity. The bottom condition is dened as a wall boundary to match the experimental conditions. In the present numerical
study, the free surface is simplied by modeling it as a wall.
A moving wall boundary condition is applied for the cylinders when
the cylinders are in FIM. For the roughness strips, due to the
specically modied surface geometry, a wall function type boundary condition is used for vt and ~ in order to account for the effect of
surface roughness. Thus, the separation point can be predicted accurately during the calculation. In addition, similar to the trip-less
initial condition for the one-cylinder simulation (Wu et al., 2011), the
uid domain is divided into two regions: (a) from the upstream inlet
to the center of the 1st cylinder, a zero value is applied for the eddy
viscosity, and (b) a nonzero value is used for the downstream-half of
the 1st cylinder through the 2nd cylinder to the outlet of the ow
domain. The nonzero value is set equal to the molecular eddy
viscosity for all the simulations in the present study. The water
properties for testing and simulations are also shown in Fig. 3.
The body and channel boundary conditions in the numerical
model match the physical model conditions as described in
Section 2 with the exception of the free surface, which is modeled
by a wall.

433

As shown in Table 4, the three grids produce similar results.


Thus, in the present work, the medium grid resolution for the two
PTC-cylinders was selected as well. A close-up of the medium grid
is shown in Fig. 4.
In the present work, the ow is simulated in the range
30,000 oReo 105,000, which falls in the high-lift TrSL3 regime,
and for which experiments were conducted in the MRELab, where
TrSL indicates Transition in Shear Layer (Zdravkovich, 1997a). In
these experiments, galloping was observed and the maximum
amplitude reached was 2.8D, where the safety stops were placed
(Kim et al., 2011). In those cases, in the CFD simulations, large
mesh deformations occur with the cylinders undergoing galloping.
In order to minimize the mesh deformation, a dynamic mesh
technique of topological change was used in the present study.
Comparing Fig. 5 with Fig. 4, when the cylinders are in FIM, the
2D  2D square, which is part of the grid, is moving up and down
with the cylinder. The cell layers, which are located at the top or
bottom of each square, are removed when the mesh is compressed
and added when the mesh is expanded. Thus, there is little
deformation in the mesh when the cylinders undergo large FIM.

Table 4
Grid resolution study (Re 30,000).
Grid (central square:
circumferential  radial)

Coarse (180  40)


Medium (240  70)
Fine (360  100)

Cd

Cl

St

1st

2nd

1st

2nd

1st

2nd

1.029
1.039
1.038

0.060
0.065
0.067

0.287
0.299
0.298

0.537
0.561
0.559

0.152
0.152
0.150

0.152
0.152
0.150

3.4. Grid generation


Two-dimensional, structured, computational grids were generated for all cases using the Gambit grid generating software. The
grid domain size is 52D  9D. The distance between the downstream boundary edge and the center of the 2nd cylinder is 25
times the cylinder diameter. This is to ensure that the results of the
numerical model are accurate and that the conditions at the ow
outlet are close to the assumed conditions. The distance from the
upstream boundary to the center of the 1st cylinder is also set at
25D. The computational domain in the vicinity of each cylinder is a
2D  2D square where the grid density for the near-wall region is
enhanced to solve for high resolution in ow properties. For the
cylinder with PTC, the standard rough wall function is used to
account for the effect of surface roughness. Due to the nature of
the wall-function for the roughness model used in this study, the
near-wall grid-spacing was selected to produce a y+ between 30
and 70, depending on the Reynolds number.
In order to determine the overall grid resolution to achieve a
convergent and accurate solution in reasonable computationaltime, three different grid densities were considered. In earlier
work, a similar grid sensitivity study was conducted and the
medium grid was successfully used to simulate a single cylinder
with PTC in FIM (Wu et al., 2011). In this paper the grid sensitivity
study was conducted using three different grid densities for two
stationary PTC-cylinders. The grid parameters and selected results
are listed in Table 4, where Cd is the time-average value of the drag
coefcient, Cl is the average value of the absolute values of the lift
coefcient peaks, and St is the Strouhal number.

Fig. 4. Close-up of the medium resolution grid for 2 cylinders with PTC.

Fig. 5. Close-up of the grid for two PTC-cylinders in FIM.

434

L. Ding et al. / Ocean Engineering 72 (2013) 429440

Table 5
Computational time.
Re (104)

10

Computational time (h) 62 96 180 240 487 523 690


Hits channel
Simulated real time (s) 20 20
20
20
20
20
20
boundaries
Time step
Automatic time step adjustment (maximum Courant
number is 0.2)

3.5. Computational time


Table 5 provides information on the computational time used
in the CFD simulations as one-processor equivalent with reference
to simulated real time. The processor used was an AMD Opteron
64-bit cluster. The operating system was Red Hat Linux. The
memory used was 3 GB.

4. Amplitude ratio results


In earlier work, it was shown that FIM can be enhanced to
achieve back-to-back VIV and galloping by introducing PTC (Chang
et al., 2011; Wu et al., 2011). For a single cylinder with PTC, the
amplitude exceeds three diameters and the synchronization range
remains open-ended due to facility limitations. Results within the
capability of the LTFSW Channel show more than doubling of the
synchronization range compared to that of VIV of a smooth cylinder.
The present study aims at modeling and simulating numerically the ow and cylinder dynamics for two rigid PTC-cylinders in
tandem supported by linear springs in a steady uniform ow in a
uid domain similar to the test section of the LTFSW Channel.
Cylinder oscillations are constrained to the direction perpendicular
to the ow and the cylinder axis. A series of simulations are
conducted for validating the responses of the two cylinders
undergoing ow induced motion. The numerical simulations use
the values of the system parameters used in the model tests (Kim
et al., 2011). The Reynolds number range is 30,000 oReo105,000
which is in the high lift TrSL3 regime, the corresponding reduced
velocity ranges are 3.84 oU nwater o 13.45 for the 1st cylinder and
3.93 oU nwater o 13.77 for the 2nd cylinder. In this section, the
amplitude response of the two cylinders is discussed. The simulation results are compared with the experimental data derived in
the LTFSW Channel of the MRELab (Kim et al., 2011). In the present
study, both cylinders start from the neutral position with zero
initial velocity and displacement. The amplitude, Apeaks, of each
cylinder is calculated by averaging the absolute values of the 60
highest positive or negative peaks.
4.1. First (upstream) cylinder
The amplitude ratios (Apeaks/D) for the numerical study and
experimental data for the 1st cylinder are plotted in Fig. 6. Within
the test range of experiments and simulations, ve regions are
observed in the amplitude ratio curve.
(a) Reo 30,000: No FIM takes place in this range experimentally
or numerically.
(b) 30,000 oReo40,000: This is the initial branch in VIV. FIM
using simulations starts at Re30,000 (U nair 3.04, U nwater
3.84) and the amplitude ratio vs. U=U nwater /Re follows closely
the experimental data with one exception. Specically, the
initial branch is initiated numerically (Re30,000) earlier than
in the experiments (Re40,000). This is attributed to the
difference between the mathematical damping model in the

Fig. 6. Amplitude ratio of the 1st cylinder with PTC.

numerical simulations in this paper and the actual physical


damping model in the experimental apparatus, which is
modeled more accurately by Eq. (10).
(c) 40,000oReo80,000: This is the upper branch in VIV. In the
amplitude curve for 40,000oReo80,000 (5.12oU nwater o10.25),
the URANS results follow closely the upper branch of the
experiments. The amplitude increases steadily as the velocity
increases for 40,000 oReo80,000 and the amplitude ratio
increases from 0.89 to 1.40. For Reynolds numbers less than
10,000, typical VIV response consists of an initial branch
followed by a constant amplitude upper branch and a lower
branch (Williamson and Govardhan, 2004, 2008). For higher
Reynolds numbers, following the initial branch, is a strong
upper branch increasing in amplitude and overtaking the
lower branch nearly completely prior to desynchronization
(Bernitsas et al., 2008, 2009).
(d) 80,000 oReo95,000: This is the region of transition from VIV
to galloping. For cylinders with PTC, transition to galloping
was successfully initiated at U nwater 10.25 that is back-to-back
with VIV (Chang et al., 2011) instead of the typical U nwater 20.
Fig. 6 shows this rapid rise in amplitude for Re480,000
(U nwater 410.25). In this region, both forcing mechanisms coexist as is further explained in Section 6.
(e) Re495,000: This is the galloping region. By the end of
the experimental range, U nwater 13, the amplitude ratio continues to increase and approaches a maximum value of 2.86
for the 1st cylinder. In the range of transition from VIV to
galloping and the galloping range, the agreement between
CFD calculations and experimental data is excellent. In
the experiments, the maximum amplitude ratio is about
2.797 occurring at Re104,356 (U nwater 13.37) for the 1st
cylinder.
4.2. Second (downstream) cylinder
For the 2nd cylinder, PTC is applied at 7 301 as shown in
Table 3. The amplitude ratio results are shown in Fig. 7. FIM results
calculated by CFD fall into one of ve branches as was observed in
the 1st cylinder: no FIM branch, the initial branch of VIV, the
upper branch of VIV, transition from VIV to galloping, and
galloping.
(a) Reo30,000: No FIM takes place in this range experimentally
or numerically.

L. Ding et al. / Ocean Engineering 72 (2013) 429440

(b) 30,000 oReo 40,000: In this initial branch of VIV, the 2nd
cylinder has nearly zero amplitude with an amplitude ratio of
less than 0.1 at Re30,000 (U nwater 3.93). This is also observed
in the corresponding experiments.
(c) 40,000 oReo 80,000: At Re 40,000 (U nwater 5.24) where the
upper branch in VIV begins, the amplitude of oscillation
increases sharply and agrees well with the experiments until
Re 42,300. At the rst part of the upper branch, that is for
40,000 oReo 56,400
(5.24oU nwater o7.40),
experimental
results show a drop in the amplitude of the 2nd cylinder to
nearly zero. Simulation cannot predict this phenomenon. Past
this discrepancy at the beginning of the upper branch, agreement between CFD and experiments is very good. The amplitude ratio increases at a relatively slow rate picking up from
0.80 at Re40,000 (U nwater 5.24) and reaching 1.39 around
Re 80,000 (U nwater 10.49).
(d) 80,000 oReo 95,000: Next comes the transition from VIV to
galloping, a range that has hardly been studied in the literature and is discussed further in Section 6 based on vortex
structures. The amplitude increases rapidly for U nwater 410.49.
(e) Re 495,000: A maximum value of 3.5 in amplitude ratio is
reached in galloping at U nwater 13.31, which is higher than the
maximum value of 2.76 measured experimentally for the 2nd
cylinder. This is due to the fact that in the CFD simulations the
free surface was replaced by a wall. In the experiments, as
energy is converted from hydrokinetic to mechanical, the two
cylinders create a dam effect thus lowering the water level
above the 2nd cylinder. That limits the achievable amplitude
experimentally, which is observed as a plateau in the experimental results in Fig. 7. The safety-stops are placed on both
sides of the mean position with a distance of around 2.8 times
the diameter in the experiments in the MRELab. Consequently,
the cylinder would hit the safety stops and limit the travel
when it was undergoing galloping. In CFD, simulation would
stop when the distance between the bottom wall boundary
and the center of each cylinder would reach one diameter,
which is the distance between the bottom side of the 2D-by2D square grid of higher resolution for near wall calculations.

435

5. Frequency ratio results


The simulation records for each run and for each cylinder are
processed using Fast Fourier Transform (FFT). Thus, the frequency
of oscillation is calculated and the frequency ratio is plotted versus
reduced velocity U nwater , Reynolds number Re, and ow velocity U
for the 1st PTC-cylinder in Fig. 8 and for the 2nd PTC-cylinder in
Fig. 9. The frequency of oscillation for each cylinder is nondimensionalized by the corresponding system natural frequency
in water, fn,water. The results are compared with the experimental
data from the LTFSW Channel (Kim et al., 2011).
5.1. First (upstream) cylinder
As shown in Fig. 8, the frequency ratio curve exhibits variations
as FIM transitions between branches similar to the experimental
results.

Therefore, in both simulations and experiments, the limits of


the tools for analysis are reached as expected for the case of
galloping. It should be reminded that galloping is an instability
phenomenon which stops only with the collapse of the structure
unless stops or higher damping are imposed.

(a) Reo 30,000: No FIM takes place in this range experimentally


or numerically.
(b) 30,000 oReo40,000: The major harmonic frequency in the
VIV initial branch is higher in the numerical simulations than
in the experiments due to the viscous damping model as
explained in Section 4 on the basis of the response amplitude.
Specically, in the numerical model only the linear viscous
damping is modeled while the physical model exhibits a very
complex viscous model see Eq. (10) as identied by Lee et al.
(2011). As a result, the experimental initial branch starts later
at Re40,000. There is a small increase in frequency ratio
around Re40,000 (U nwater 5.12) numerically matching the
experimental jump.
(c) 40,000 oReo80,000: The large jump of frequency observed
in the experiments at Re40,000 indicates the oscillation of
the 1st cylinder transitions from the VIV initial branch to the
VIV upper branch. In the upper branch, simulations and
experiments match very closely. As the Re increases from
40,000 to 60,000 (U nwater 5.127.69), the frequency ratio of
the 1st cylinder decreases from 1.20 and reaches 1.03. After
Re 60,000 (U nwater 7.69), frequency ratio stabilizes around
1.05 and the curve shows a nearly constant slope, with the
oscillation frequency of the 1st cylinder being very close to the
system natural frequency. This good agreement between
experiments and simulations is attributed to the following
two facts.

Fig. 7. Amplitude ratio of the 2nd cylinder with PTC.

Fig. 8. Frequency ratio of the 1st cylinder with PTC.

436

L. Ding et al. / Ocean Engineering 72 (2013) 429440

cylinder and one due to its own vortex shedding. The following
observations can be made regarding the ve regions of FIM.

Fig. 9. Frequency ratio of the 2nd cylinder with PTC.

i. The classical linear viscous damping model used in the


simulations matches well with the physical damping model
because the velocity of oscillations is not near zero. Thus,
the damping dynamic memory effect and the nonlinear
static damping effect are small compared to the linear
viscous damping term as identied by Lee et al. (2011). This
was further veried by Lee and Bernitsas (2011) where
experimental data with physical springs and dampers were
compared to experimental data with virtual springs and
dampers emulated by a controller. The virtual system
provided an oscillator which matched perfectly the mathematical model on the linear oscillator.
ii. The amplitude of oscillation in the upper branch remains
below 1.5D and thus the cylinder is not close to the free
surface experimentally which numerically has been
replaced by a wall. The effect of this discrepancy does not
come into play until Re100,000 as shown in Figs. 6 and 7
when the amplitude experimental data start exhibiting a
plateau.
(d) 80,000 oReo95,000: As the Reynolds number reaches about
80,000 (U nwater 10.25), a small jump in the frequency ratio
occurs right at the point of switching from the VIV upper
branch to the transition region from VIV to galloping.
(e) Re4 95,000: The frequency ratio reduces at a relatively slow
rate after the oscillation mode transition into the galloping
branch has occurred, and then its value remains in the vicinity
of 1. In the experimental results, the frequency ratio of the 1st
cylinder slowly rises with the increase of ow velocity and
drops around the transition between the upper branch and the
galloping branch, and then increases again. The frequency
ratio holds around 1 in the galloping branch. In summary, the
simulation results of the oscillation frequency for the 1st
cylinder are similar with the experimental data.

5.2. Second (downstream) cylinder


In Fig. 9, the frequency ratio fosc/fn,water for the 2nd cylinder is
plotted along with experimental results for comparison. The
motion of the 2nd cylinder is affected by the upstream cylinder
and exhibits unique response which is veried both numerically
and experimentally. The FFT of the 2nd cylinder (see Figs. 11 and
12) shows two frequencies in the response of the 2nd cylinder, one
due to the oscillations and wake frequency of the upstream

(a) Reo30,000: No FIM takes place in this range experimentally


or numerically.
(b) 30,000 oReo40,000: As shown in Fig. 7, the amplitude ratio
of the 2nd cylinder is low for Re30,000 (U nwater 3.93) for the
same reasons as those discussed regarding the 1st cylinder.
The numerical frequency ratio of the 2nd cylinder remains
obviously higher than that in the experiment and almost the
same value as the 1st cylinder, which is shown in Fig. 9. This
difference in general reduces as FIM moves into the upper
branch where the cylinder speed is higher and thus the
discrepancy between the physical damping model in Eq. (10)
and the mathematical linear damping model in Eq. (9)
weakens.
(c) 40,000 oReo80,000: In the numerical simulation results, the
frequency ratio of the 2nd cylinder follows the experimental
results trend. For reduced velocity 5o U nwater o7, the simulated
frequency ratio of the 2nd cylinder follows the same trend but
over-predicts the experimentally measured value by about
515%. As shown in Fig. 9, for the 2nd cylinder, a prominent
drop occurs in both curves of numerical data and experimental
results around U nwater 7.
(d) 80,000 oReo95,000: At Re80,000 (U nwater 410.49), the VIV
to galloping transition occurs. The frequency ratio gradually
drops to about one at the beginning of galloping.
(e) Re495,000: In the galloping range the frequency ratio is very
close to 1 and the results of simulation and experiments are
nearly identical.

6. Near-wake structures
The 2-D URANS results of amplitude and frequency response
for two PTC-cylinders match well with experiments. The amplitude and frequency response are closely related to the vortex
dynamics and wake pattern. Actually, amplitude and frequency are
integral properties of the uidstructure dynamics in the sense
that the pressure is integrated to give a force to which the cylinder
responds. Typically, integrals reduce error compared to nonintegral properties such as pressure distribution or location of
the separation point. Thus, it is harder for a URANS code on
cylinder uid dynamics to predict accurately local properties such
as vorticity and pressure distribution than it is to predict integral
properties such as Strouhal number, drag and lift forces, or
amplitude and frequency of response. A very important local
property is that of the vorticity distribution which results in vortex
structures in the near-wake. The vortex structures around the two
PTC-cylinders in FIM are presented and discussed in this section.
In the numerical and experimental results presented by Wu
et al. (2011) and Chang et al. (2011), the near-wake structures and
mode transition for one PTC-cylinder in FIM were discussed and
the salient features of the ow in the different branches of VIV and
galloping were achieved numerically. For one cylinder in FIM, the
transition between branches is accompanied by vortex pattern
change and the vortex pattern is stable when the cylinder is in a
branch (Wu et al., 2011).
It should be reminded here, that the reason for this successful
numerical prediction of the experimental results lies in the
application of the turbulence stimulation in the form of the PTC.
Specically, 2-D URANS results for a stationary smooth cylinder
match well basic integral experimental results, such as Strouhal
number, and drag and lift coefcients, for Reo10,000. For a
smooth cylinder in VIV this agreement between experiments
and CFD extends to Re about 12,000 (Wanderley et al., 2008; Wu

L. Ding et al. / Ocean Engineering 72 (2013) 429440

et al., 2011). The failure of agreement for Re4 12,000 can be traced
to the inaccurate prediction of a very important local property for
ows past a cylinder stationary or in FIM. That is the point of
separation of the ow and its oscillation as vortices shed in an
alternating manner. Specically, the separation point in laminar
ow (Reo 300,000) is located at 811 and oscillates around it up to
75101. For Re410,000, 2-D URANS methods fail to predict that
motion correctly. Typically, the separation point for Re4 10,000 is
predicted by 2D-URANS to be stationary at 901. With the addition
of the PTC in the experiments and in the 2-D URANS simulations,
the location of the separation point is predetermined resulting in
accurate prediction of the separation point. That resulted in very
good agreement between simulations and experiments in Wu
et al. (2011) for Reynolds numbers at least up to 135,000 for which
experimental results were available for a single PTC-cylinder in
FIM. This successful agreement extended not only to integral
properties but also local properties such as the vortex near-wake
structures. This agreement is also evident in the results in this
paper for two PTC-cylinders in FIM.
For the two PTC-cylinders in tandem cases, the upstream
cylinder (1st cylinder) has great inuence on the motion and
vortex shedding of the downstream cylinder (2nd cylinder), and
the vortex pattern becomes more complex than in the single
cylinder cases. The simulation results of four typical Reynolds
numbers, which correspond to the VIV initial branch, upper
branch, transition from VIV to galloping, and galloping branch,
are presented in this section. The vortex patterns for two PTCcylinders at Re 30,000, Re 59,229, Re93,074, and Re 100,000
are shown in Figs. 1013, respectively. The displacement ratio and
its FFT analysis for each cylinder are shown in Figs. 1012 as well.
6.1. Reynolds number of 30,000 (initial VIV branch)
As shown in Fig. 10, the 2S mode of vortex shedding can be
clearly observed for the 1st cylinder. Here, 2S indicates two single
vortices shed per cycle. Two vortices are shed from the 1st cylinder
per cycle of oscillation, one by the top shear layer and another one
by the bottom shear layer. When the two vortices move downstream and cross into the domain of the 2nd cylinder, the clockwise rotating vortex passes right above the 2nd cylinder and the
counter-clockwise vortex passes below it. This phenomenon,
due to the specic spacing between the two cylinders, causes
the vorticity from the 1st cylinder to absorb the same-rotation
vorticity from the 2nd cylinder preventing formation of large von
Krmn vortices forming behind the 2nd cylinderthus suppres-

437

sing its FIM. Shed vortices of the 1st cylinder allow only generation of small scale and very weak vortices in the 2nd cylinder.
In addition, the motion of the 1st cylinder has a strong regular
form, which can be observed in the displacement ratio curves and
FFT analysis in Fig. 10. The displacement of the 2nd cylinder is very
small, with average value of the 60 maximum peaks about 0.1D
and a maximum displacement of about 0.13D. Therefore, the
motion of the 2nd cylinder is almost suppressed. For the cases in
the VIV initial branch, the suppression of the 2nd cylinder was also
observed in the experiments. Visualization of the near wake vortex
structures using CFD has helped understand and explain this
phenomenon.
6.2. Reynolds number of 59,229 (upper VIV branch)
The time sequence of vortex shedding is shown in Fig. 11. In the
vortex structure of the near-wake of the 1st cylinder, two modes of
vortex shedding are observed in the simulation results: (a) When
the two PTC-cylinders move in opposite direction (out of phase), a
2P vortex pattern is observed behind the 1st cylinder, where 2P
means two pairs of vortices shed per cycle; (b) When the two
cylinders move in phase, the vortex mode of the 1st cylinder is 2P
+2S. The vortex pattern of the 1st cylinder switches between these
two modes over time. Thus, the motion of the 2nd cylinder
inuences the vortex shedding of the 1st cylinder. For the 2nd
cylinder, the 2P vortex pattern is shown in the simulation results.
The upstream vortices directly and closely interact with the
downstream cylinder. As can be seen in the displacement ratio
curves in Fig. 11, the motion of the 1st cylinder shows a periodic
pattern while small displacement is observed in certain cycles for
the 2nd cylinder. The reason for the small displacement in those
cycles is that the vortex development of the 2nd cylinder is
weakened by the shed vortices from the 1st cylinder, which is
similar to the phenomenon of the 2nd cylinder at Re30,000. The
drop-off in displacement of the 2nd cylinder is associated with the
variation of the phase difference between 1st and 2nd cylinder,
which means the relative position of the two cylinders changes
periodically from out-of-phase to in-phase. At the same time, the
vortex structure of the 1st cylinder switches between 2P and 2P
+2S. Relatively large difference of oscillation frequencies between
1st and 2nd cylinder could be found in the FFT analysis of the
displacement ratio. Three peaks appear in the result of FFT analysis
for the 2nd cylinder as shown in Fig. 11. The frequency values of
these three peaks are close to each other. The frequency of the
highest peak is larger than those of other two peaks. The

Fig. 10. Vortex structures, displacement history, and frequency spectrum in the initial VIV branch at Re 30,000. (Tn,1 0.889, where Tn 1/fn,water and Tn,1 is for the 1st
cylinder).

438

L. Ding et al. / Ocean Engineering 72 (2013) 429440

Fig. 11. Vortex structures, displacement history, and frequency spectrum in the upper VIV branch at Re 59,229.

Fig. 12. Vortex structures, displacement history, and frequency spectrum in the VIV-to-galloping transition region at Re93,074.

frequency of the peak in the middle, which has the smallest


amplitude among these three peaks, equals to the one of the FFT
analysis for the 1st cylinder.
6.3. For Reynolds number of 93,074 (VIV to galloping transition)
As shown in Fig. 12, both cylinders shed vortices following the
2P+2S mode. By the preceding analysis in Section 4, the two PTCcylinders are in the region of transition from VIV to galloping. There

is variation in the vortex shedding structure for the 1st cylinder:


The 2P+2S pattern is observed in most cycles; occasionally though,
one additional vortex is shed during the upward travel. That is, a
cycle-to-cycle variation in shedding exists. For the 2nd cylinder, the
vortex pattern is hard to identify as the shed vortices are strongly
disrupted and modied by the vortices shed by the upstream
cylinder. In the displacement ratio curves in Fig. 12, the amplitudes
have strong and weak values and the maximum displacement
reaches 3D in some cycles for both cylinders.

L. Ding et al. / Ocean Engineering 72 (2013) 429440

439

Fig. 13. Vortex structures in galloping at Re 100,000.

During transition from VIV to galloping, several FIM features


change. We have observed and discussed changes from the typical
VIV amplitudes of about 1-2 diameters to higher values and also
changes in the frequency ratio. There is another important phenomenon to be observed and studied in Fig. 12, which has hardly
been studied in the literature: it is the transition from the VIV
driving mechanism to the galloping driving mechanism. The discussion on the driving mechanisms of FIM is presented at the end of
this section as it is better understood by comparing Figs. 1113.

vortices cannot be developed and stay synchronized with the


cylinder motion. That is, some vortices increase the lift force as
they are in phase with the cylinder motion and some decrease the
lift force as they are out of phase with the cylinder motion. In fully
developed galloping the shear layer motion is in synchronization
with the galloping instability motion.

6.4. For Reynolds number of 100,000 (fully developed galloping)

One degree of freedom ow induced motions, transverse to a


uniform ow, of two rigid circular cylinders, mounted on end
linear-springs, in tandem were studied using 2-D URANS simulations veried by experimental data. The range of Reynolds numbers for which experimental data were collected in the MRELab
was 30,000 oReo105,000, which falls in the high-lift TrSL3
regime. Typical 2-D URANS results on smooth circular cylinders,
stationary or in VIV, are valid up to Reynolds number Re10,000
12,000. In earlier work, it has been shown that passive turbulence
control in the form of selectively distributed surface roughness
results in very good agreement between 2-D URANS and experiments for single cylinder FIM. PTC was used in this paper and
proved to be the key factor in achieving agreement between
experimental and CFD simulations. The following conclusions
can be drawn from the results presented in this paper.

Fig. 13 shows the vortex structures for the two cylinders in fully
developed galloping at Re100,000. For the 1st cylinder and there
are 8 vortices shedding in one oscillation cycle. The vortex pattern
for the 2nd cylinder is not easy to identify as the vortex shedding
is severely affected by the 1st cylinder. The number of vortices and
their shedding mode is of secondary importance in galloping as
explained next.
6.5. FIM driving mechanism
The VIV driving mechanism is solely based on the oscillatory lift
resulting from vortex shedding. The mode of vortex shedding
whether 2S, 2P, or 2P+2S has frequency locked onto the frequency
of oscillation of the cylinder. Thus, the oscillatory excitation is in
synchronization with the body motion a condition similar to linear
resonance at each frequency of oscillation as the ow velocity
changes within the synchronization range. In galloping, the driving
mechanism is not based on the alternating vortices but on the lift
instability caused by negative damping due to the lift force induced
by the geometric asymmetry of the circular cylinder due to the
turbulence stimulation. The transition from the VIV mechanism to
the galloping mechanism can be observed by comparing Fig. 11
(VIV) to Fig. 12 (VIV-to-galloping transition) to Fig. 13 (fully
developed galloping). In Fig. 11, the vortex shedding mode is in
synchronization with the cylinder oscillation. In Fig. 12, as the
amplitude of oscillation increases, the number of vortices shed per
cycle increases resulting in more complex modes. The vortex
shedding frequency is several times higher than the vortex-mode
frequency, which is still in synchronization with the cylinder
oscillations. In Fig. 12, it can also be observed that the shear layer
motions follow the cylinder oscillations as expected. Their role in
inducing oscillatory lift in synchronization with the cylinder motion
increases and becomes the dominant driving mechanism in Fig. 13
where vortices no longer shed in modes synchronized with the
cylinder oscillations. Vortices shed in less complex modes as the
cylinder amplitude increases and more complex modes with more

7. Conclusions

1. An effective method was developed to handle large-amplitude


FIM response. Large mesh deformations occur when the
cylinders undergo FIM in the form of VIV or galloping. In order
to minimize the mesh deformation, a dynamic mesh technique
of topological change was implemented.
2. The amplitude-ratio results are in excellent agreement with
experimental data showing the initial and upper branches in
VIV, transition from VIV to galloping, and galloping for the two
PTC-cylinders. The discrepancy observed at the initiation of FIM
in the initial branch of VIV was justied based on the difference
between the mathematical damping model implemented in the
simulations and the actual physical damping model at low
cylinder oscillatory velocity.
3. The frequency results are in excellent agreement with experimental data also showing the initial and upper branches in VIV
with back-to-back galloping for the two PTC-cylinders.
4. Integral properties of FIM such as the Strouhal number and lift/
drag forces are easier to predict using 2-D URANS. Such
methods fail in predicting local features of ow past cylinders
in FIM for Re4 10,000 and particularly the complex motion of
the separation point which is a key and unique feature in
cylinder ows. With the proper implementation of PTC, the

440

5.
6.

7.

8.

9.

L. Ding et al. / Ocean Engineering 72 (2013) 429440

location of the separation point is known a priori resulting in


very good agreement between experiments and simulations.
An important local ow property is the vorticity generation
which results in complex vortex structures. These were studied
using high-resolution imaging from the CFD results.
For Re 30,000, in the initial VIV branch, the typical 2S vortex
structure is shown for the 1st cylinder.
For Re30,000, in the initial VIV branch, the 2nd cylinder
motion is almost suppressed and simulations explain this
phenomenon for center-to-center spacing between the cylinders equal to two diameters.
For Re59,229, which is in the range of the VIV upper branch,
both 2P and 2P+2S patterns are observed for the 1st cylinder,
while the vortex structure for the 2nd cylinder is only 2P. The
upstream vortices shedding from the 1st cylinder, directly and
closely interact with the downstream cylinder.
The vortex structure simulation is most helpful in understanding and demonstrating the differences between the driving
hydrodynamic mechanism in VIV and galloping as well as the
coexistence of the two mechanisms in the transition region
from VIV to galloping.
In galloping, amplitude of 3.5 diameters is achieved numerically in good agreement with experimental results. The ow
domain limits are reached and the stops in the ow channel are
hit thus limiting experimental testing.

Acknowledgements
The following support is gratefully acknowledged: (a) DOE
contract DE-EE0003644 to Vortex Hydro Energy with subcontract
to the University of Michigan, (b) ONR grant N00014-08-1-0601
to the University of Michigan, Program Manager Kelly Cooper.
(c) Specialized Research Fund for the Doctoral Program of Higher
Education of China (Grant No. 20120191130003) and the China
Scholarship Council for Lin Ding.
References
Allmaras, S.R., Johnson, F.T., Spalart, P.R., 2012. Modications and clarications for
the implementation of the SpalartAllmaras turbulence model. In: Seventh
International Conference on Computational Fluid Dynamics. 913 July 2012, Big
Island, Hawaii.
Assi, G.R.S., Meneghini, J.R., Aranha, J.A.P., Bearman, P.W., Casaprima, E., 2006.
Experimental investigation of ow-induced vibration interference between
two circular cylinders. J. Fluid Struct 22 (67), 819827.
Aupoix, B., Spalart, P.R., 2003. Extensions of the SpalartAllmaras turbulence model
to account for wall roughness. Int. J. Heat Fluid Flow 24 (4), 454462.
Bernitsas, M.M., Ben-Simon, Y., Raghavan, K., Garcia, E.M.H., 2009. The VIVACE
converter: model tests at high damping and Reynolds Number around 105. J.
Offshore Mech. Arct. Eng-Trans. ASME 131, 1.
Bernitsas, M.M., Raghavan,K., 2009. Fluid Motion Energy Converter. United States
Patent and Trademark Ofce Patent #7, 493, 759 B2 Issued on February 24,
2009.
Bernitsas, M.M., Raghavan, K., Ben-Simon, Y., Garcia, E.M.H., 2008. VIVACE (vortex
induced vibration aquatic clean energy): a new concept in generation of clean
and renewable energy from uid ow. J. Offshore Mech. Arct. Eng-Trans. ASME
130, 4.
Borazjani, I., Sotiropoulos, F., 2009. Vortex-induced vibrations of two cylinders in
tandem arrangement in the proximity-wake interference region. J. Fluid Mech
621, 321364.

Catalano, P., Wang, M., Iaccarino, G., Moin, P., 2003. Numerical simulation of the
ow around a circular cylinder at high Reynolds numbers. Int. J. Heat Fluid Flow
24 (4), 463469.
Chang, C.-C., Kumar, R.A., Bernitsas, M.M., 2011. VIV and galloping of single circular
cylinder with surface roughness at 3.0  104Re1.2  105. Ocean Eng. 38 (16),
17131732.
Edwards, J.R., Chandra, S., 1996. Comparison of eddy viscosity-transport turbulence
models for three-dimensional, shock-separated ow elds. AIAA J 34 (4),
756763.
Kim, E.S., Bernitsas, M.M., Kumar, R.A., 2011. Multi-cylinder ow-induced motions:
enhancement by passive turbulence control at 28,000 oReo120,000. In:
Proceedings of the OMAE. 1924 June 2011, Rotterdam, the Netherlands,
#44397, pp. 249260.
King, R., Johns, D.J., 1976. Wake interaction experiments with two exible circular
cylinders in owing water. J. Sound Vib 45 (2), 259283.
Lee, J., Chang, C.-C., Xiros, N.I., Bernitsas, M.M., 2010. Integrated power take-off and
virtual oscillator system for the VIVACE Converter: VCK system identication.
In: ASME 2009 International Mechanical Engineering Congress and Exposition,
1319 November 2009, Lake Buena Vista, FL, United states, PART A, pp. 393
399.
Lee, J.H., Bernitsas, M.M., 2011. High-damping, high-Reynolds VIV tests for energy
harnessing using the VIVACE converter. Ocean Eng. 38 (16), 16971712.
Lee, J.H., Xiros, N., Bernitsas, M.M., 2011. Virtual damper-spring system for VIV
experiments and hydrokinetic energy conversion. Ocean Eng. 38 (56), 732747.
Park, H., Bernitsas, M.M., Kumar, R.A., 2012. Selective roughness in the boundary
layer to suppress ow-induced motions of circular cylinder at
30,000o Reo 120,000. J. Offshore Mech. Arct. Eng 134 (4), 041801.
Raghavan, K., 2007. Energy Extraction from a Steady Flow Using Vortex Induced
Vibration. Ph.D. Thesis. Dept. of Naval Architecture & Marine Engineering,
University of Michigan.
Raghavan, K., Bernitsas, M.M., 2008. Enhancement of high damping VIV through
roughness distribution for energy harnessing at 8  103 o Reo 1.5  105. In:
27th International Conference on Offshore Mechanics and Arctic Engineering.
913 June 2008, pp. 871882.
Raghavan, K., Bernitsas, M.M., 2011. Experimental investigation of Reynolds
number effect on vortex induced vibration of rigid circular cylinder on elastic
supports. Ocean Eng 38 (56), 719731.
Raghavan, K., Bernitsas, M.M., Maroulis, D.E., 2009. Effect of bottom boundary on
VIV for energy harnessing at 8  103 o Reo 1.5  105. J. Offshore Mech. Arct.
Eng-Trans. ASME 131 (3), 113.
Shur, M., Spalart, P., Strelets, M., Travin, A., 1996. Navier-Stokes simulation of
shedding turbulent ow past a circular cylinder and a cylinder with backward
splitter plate. In: Desideri, J.A., Hirsch, C., LeTallec, P., Pandol, M., Periaux, J.
(Eds.), Proceedings of the 1996 Third ECCOMAS Computational Fluid Dynamics
Conference, Paris, France, pp. 676682.
Spalart, P.R., Allmaras, S.R., 1994. A one-equation turbulence model for aerodynamic ows. Recherch Aerospatiale 1, 521.
Sumner, D., Price, S.J., Paidoussis, M.P., 2000. Flow-pattern identication for two
staggered circular cylinders in cross-ow. J. Fluid Mech. 411, 263303.
Travin, A., Shur, M., Strelets, M., Spalart, P., 2000. Detached-eddy simulations past a
circular cylinder. Flow Turbul. Combust. 63 (14), 293313.
Tritton, D.J., 1977. Physical Fluid Dynamics. Van Nostrand Reinhold, New York.
Wanderley, J.B.V., Sphaier, S.H., Levi, C., 2008. A Numerical Investigation of Vortex
Induced Vibration on an Elastically Mounted Rigid Cylinder. In: 27th International Conference on Offshore Mechanics and Arctic Engineering, 1520 June
2008, Estoril, Portugal, pp. 703711.
Williamson, C.H.K., Govardhan, R., 2004. Vortex-induced vibrations. Annu. Rev.
Fluid Mech. 36, 413455.
Williamson, C.H.K., Govardhan, R., 2008. A brief review of recent results in vortexinduced vibrations. J. Wind Eng. Ind. Aerodyn. 96 (67), 713735.
Wu, W., Bernitsas, M.M., Maki, K., 2011. RANS simulation vs. experiments of ow
induced motion of circular cylinder with passive turbulence control at
35,000 oReo 130,000. In: ASME 2011 30th International Conference on Ocean,
Offshore and Arctic Engineering, 1924 June 2011, Rotterdam, Netherlands,
pp. 733744.
Zdravkovich, M.M., 1985. Flow induced oscillations of two interfering circular
cylinders. J. Sound Vib 101 (4), 511521.
Zdravkovich, M.M., 1987. The effects of interference between circular cylinders in
cross ow. J. Fluid Struct 1 (2), 239261.
Zdravkovich, M.M., 1997a. Flow Around Circular Cylinders Volume 1: Fundamentals. Oxford University Press, England.
Zdravkovich, M.M., 1997b. Flow Around Circular Cylinders Volume 2: Applications.
Oxford University Press, England.

You might also like