You are on page 1of 22

Journal of Fluids and Structures 27 (2011) 1–22

Contents lists available at ScienceDirect

Journal of Fluids and Structures


journal homepage: www.elsevier.com/locate/jfs

The effect of air–water interface on the vortex shedding


from a vertical circular cylinder
Jungsoo Suh 1, Jianming Yang, Frederick Stern 
IIHR – Hydroscience and Engineering, University of Iowa, Iowa City, IA 52242-1585, USA

a r t i c l e in fo abstract

Article history: The flow past an interface piercing circular cylinder at the Reynolds number
Received 29 December 2009 Re = 2.7  104 and the Froude numbers Fr= 0.2 and 0.8 is investigated using large-eddy
Accepted 26 August 2010 simulation. A Lagrangian dynamic subgrid-scale model and a level set based sharp
interface method are used for the spatially filtered turbulence closure and the air–water
Keywords: interface treatment, respectively. The mean interface elevation and the rms of interface
Air–water interface fluctuations from the simulation are in excellent agreement with the available
Vortex shedding experimental data. The organized periodic vortex shedding observed in the deep flow
Turbulence is attenuated and replaced by small-scale vortices at the interface. The streamwise
Large-eddy simulation
vorticity and the outward transverse velocity generated near the edge of the separated
Level-set method
region, which enforces the separated shear layers to deviate from each other and
Cylinder
restrains their interaction, are primarily responsible for the devitalization of the
periodic vortex shedding at the interface. The lateral gradient of the difference between
the vertical and transverse Reynolds normal stresses, increasing with the Froude
number, is the main source of the streamwise vorticity and the outward transverse
velocity at the interface.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

The flow past a circular cylinder has been an important topic in fluid dynamics for a long time because of its wide
applications in engineering and its abundant flow physics including separation, reattachment, vortex shedding, etc. One
prominent feature of this flow is that it presents very different vortical structures in different ranges of Reynolds numbers
(Re, definition deferred to Section 3.2). For extensive discussions on the single-phase flow past a circular cylinder, see
Williamson (1996) and Zdravkovich (2003a, 2003b) among others.
Unlike its single-phase counterpart, the flow past an interface piercing cylinder has received much less attention,
in spite of its importance in various applications including offshore structures and surface vessels. Although a few
experimental and numerical studies on the flow around a surface-piercing cylinder are available in the literature, the
detailed hydrodynamics is not well understood yet. In general, the free surface adds great complexities to the flow due to
the generation of waves in various forms and their interaction with the body and vortices, the air–water interfacial effects
like bubble entrainment and surface tension, and the three-dimensional flow separation.
There were several experimental studies on the flow past an interface piercing cylinder with different configurations
and flow conditions. The literature survey of these studies will follow the order of increasing Re, as the Reynolds number is

 Corresponding author. Tel.: + 1 319 335 5215; fax: + 1 319 335 5238.
E-mail address: frederick-stern@uiowa.edu (F. Stern).
1
Current affiliation: Nuclear Engineering & Technology Institute, Korea Hydro & Nuclear Power Co., Daejeon, Republic of Korea.

0889-9746/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfluidstructs.2010.09.001
2 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

one of two major parameters of this flow (the other is the Froude number, Fr, and the definition is deferred to Section 3.2
too). Lang and Gharib (2000) studied the wake behind a surface-piercing cylinder for a clean and contaminated free surface
at low Reynolds numbers ranging from 350 to 460 and Froude numbers below 1.0, to investigate the effect of varying free
surface conditions. For the clean surface case, a large surface deformation was observed due to vortex filaments normal to
the free surface. As a result of the high surface curvature, a zigzag-pattern vortex pair of the surface-parallel component
was generated. When the free surface was contaminated by surfactants, the surface was no longer stress-free and, like
from a solid boundary, a horseshoe vortex was generated from the surface. Then the surface-normal vortices from the
vortex shedding were connected by the horseshoe vortex. Hence, the free surface condition changed by surfactants
provided a mechanism to make the vertical vorticity to align with the free surface. Instead of a vertical cylinder, Vlachos
and Tellionis (2008) studied the free surface effect on the vortex shedding from an inclined circular cylinder at Re = 1700
and 6000, and Fr =0.3, 0.65 and 1.06 using hydrogen-bubbles and particle flow visualizations. The inclination angle was
defined as positive when the upper part of the cylinder was inclined in the flow direction. For a zero inclination angle,
vortex shedding and vortex reconnection normal to the free surface were suppressed near the free surface at a high Froude
number (Fr = 1.06). Also, it was also found that the positive inclination of the cylinder counteracted the free surface effect
by recovering vortex shedding while a less-organized wake was observed near the free surface for the negative inclination.
Akilli and Rockwell (2002) investigated the near wake of a surface piercing cylinder in shallow water at Re = 10 052 using a
combination of visualization marker and a technique of high-image-density particle image velocimetry. They observed
significant distortion of the free surface due to a horizontal vortex induced by upward ejection of fluid through the center
of a Karman vortex. It was noted that the vorticity and Reynolds stress were broadly distributed in the transverse direction
and digressed from the symmetric plane as approaching the free surface, whereas they were concentrated in a narrow
region and directed to the symmetric plane near the bed. Inoue et al. (1993) conducted experimental studies on two cases,
in one case Re =27 000 and the Froude number Fr =0.8, and the other case Re = 29 000 and Fr =1.0. The measurement of
velocity showed that the periodic vortex shedding was apparent in the deep flow, whereas the periodic component of
fluctuations was reduced and random fluctuations of higher frequency were more prominent near the free surface.
There were also a few numerical investigations on the flow past an interface piercing cylinder. Kawamura et al. (2002)
studied the cases at Re = 27 000 and Fr =0.2, 0.5 and 0.8 using large-eddy simulation (LES) based on a Smagorinsky subgrid-
scale (SGS) model. At a small Froude number, the amplitude of surface waves was small and the influence on the wake was
negligible. On the other hand, surface waves became steeper and wave–wake interaction was also stronger at a large
Froude number. Numerical simulation predicted significant surface fluctuations inside the recirculation zone immediately
after the surface wave crest. In addition, they were able to visualize the attenuation of vortex shedding near the free
surface. Similarly, Yu et al. (2008) studied flows past a free surface piercing cylinder at Froude numbers up to Fr =3.0 and
Reynolds numbers up to Re = 1  105 using LES based on a Smagorinsky SGS model and a volume-of-fluid (VOF) method.
They also observed the attenuation of organized vortex shedding at the free surface. This effect was stronger at a larger
Froude number, but it was reduced as the Reynolds number increased. The mean overall drag coefficient increased with the
Reynolds number, but decreased along with the Froude number. Yu et al. (2009) also studied the near field sound due to
the free surface piercing cylinder.
Although the above numerical simulations presented some discussions on the effect of the free surface on the vortex
shedding behind the cylinder, the mechanism driving the attenuation of vortex shedding near the interface has not been
studied in detail yet. In this work, complex unsteady features including the boundary layer separation and transition to
turbulence, as well as the interface effects, were fully considered in a LES of the flow past an interface piercing cylinder at
Re = 27 000 and Fr = 0.8. A Lagrangian dynamic SGS model, which allows the dynamic modeling of eddy viscosity for
inhomogeneous flows in complex configurations, was used to handle the laminar boundary layer separation and transition
to turbulence in the wake at this subcritical Reynolds number. Moreover, the two-phase interface treatment was
performed in a sharp fashion using the level-set and ghost-fluid methods as a means to accurately capture the non-
stationary and sometimes mildly breaking air–water interface at this Froude number.

2. Computational method

The mathematical model and numerical method used in the present study are extensions for the orthogonal curvilinear
coordinates of CFDShip-Iowa version 6, a sharp interface Cartesian grid solver for two-phase incompressible flows, recently
developed at IIHR (Yang and Stern, 2009) and summarized below. The present solver has been fully verified and validated
during the course of code development using several benchmark test cases, which will be reported in a separate work.

2.1. Mathematical model

The governing equations in this study are the unsteady, three-dimensional, incompressible Navier–Stokes equations,
written in the orthogonal curvilinear coordinate system. The continuity equation is given as follows:

rðiÞ½ui  ¼ 0, ð1Þ
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 3

where ui (i= 1, 2, 3) is the velocity in the orthogonal coordinate xi direction and


 
1 @ J
rðiÞ½ ¼ , ð2Þ
J @xi hi

following Pope (1978). The Jacobian of the coordinate transformation is defined as J=hihjhk, and hi ¼ @xi =@xi with xi a
Cartesian coordinate.
The momentum equation is written as follows:
   
@ui 1 1 @p tij tij
þ rðjÞ½ui uj  rðjÞ½tij  ¼  þ gi þHj ðiÞ uj uj  Hi ðjÞ ui uj  , ð3Þ
@t r r @xðiÞ r r
where r is the density, p the pressure, t the time, and gi the gravity vector in the xi direction. In addition,

1 @hi
Hi ðjÞ ¼ ð4Þ
hi hj @xj

and @xðiÞ ¼ hi @xi following Pope (1978). The viscous stress tensor tij is defined as follows:
 
@ui @uj
tij ¼ m þ ui Hi ðjÞuj Hj ðiÞ þ 2ul Hi ðlÞdij , ð5Þ
@xðjÞ @xðjÞ

where m is the dynamic viscosity and dij is the Kronecker delta function.
The air–water interface has to be solved as a part of the solution. In this study, the interface is tracked as the zero level
set of a signed distance function, f, or the level-set function, by solving the following advection equation:

@f @f
þ ui ¼ 0: ð6Þ
@t @xðiÞ

In addition, the reinitialization of the level-set function is required to keep f as a signed distance function in the course of
interface evolution (Sussman et al., 1994).
The density and viscosity are defined according to the level-set function and sharp jumps of the fluid properties occur at
the interface. In this study, the density maintains its jump whereas viscosity is smoothed over a transition band across the
interface (Yang and Stern, 2009).
To handle the fully inhomogeneous turbulence in this study, the Lagrangian dynamic subgrid-scale (SGS) model based
on Sarghini et al. (1999) is adopted as it averages the model coefficient along the flow pathline (Meneveau et al., 1996).
In the LES, the small dissipative eddies are modeled by the SGS model whereas the large, energy carrying, eddies are
resolved by the spatially filtered Navier–Stokes equations. Hence, Eq. (3) can be rewritten as in the following form:
   
@u i 1 1 @p t ij t ij
þ rðjÞ½u i u j  rðjÞ½t ij rðjÞ½t~ ij  ¼  þgi þ Hj ðiÞ u j u j  t~ ij Hi ðjÞ u i u j  t~ ij ð7Þ
@t r r @xðiÞ r r
with t ij ¼ mS ij and t~ ij ¼ nt S ij with nt the turbulent eddy viscosity, respectively. Note an effective total viscosity cannot be
defined using m þ rnt because r is discontinuous across the interface in the present study (Yang and Stern, 2009). Hereafter
the filtering sign for LES will be dropped for simplicity.

2.2. Numerical method

A finite-difference method is used to discretize the governing equations on a non-uniform staggered orthogonal grid, in
which the contravariant velocity components ui, uj, uk are defined at centers of cell faces in the xi , xj , xk directions,
respectively. All other variables are defined at cell centers. Time advancement of the present study is based on the semi-
implicit four-step fractional step method by Choi and Moin (1994). The diagonal diffusion terms are advanced with the
second-order Crank–Nicholson method and the other terms by the second-order explicit Adams–Bashforth method. The
pressure Poisson equation is solved to enforce the continuity equation.
The convective terms are discretized using the fifth-order Hamilton–Jacobi Weighted-ENO (HJ-WENO) scheme (Jiang
and Shu, 1996). The other terms are discretized by the second-order central difference scheme. The pressure Poisson
equation is solved using a semi-coarsening multigrid solver from the HYPRE library (Falgout et al., 2006). In general, the
Poisson solver is the most expensive part of the whole algorithm and uses about 80–85% of the total computation time.
Both the level-set and reinitialization equations are solved using the third-order TVD Runge–Kutta scheme (Shu and
Osher, 1988) for time advancement and the fifth-order HJ-WENO scheme (Jiang and Shu, 1996) for spatial discretization.
These equations are solved for the cells in a narrow band about a few grid-cell width to reduce computational overhead
(Peng et al., 1999). An additional damping function is added to the right-hand side of the level-set advection equation to
drive the level-set function toward the desired smooth solution and minimize reflection at the outer boundary (Vogt and
Larsson, 1999).
4 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

3. Verification and validation

3.1. Grid and domain

The computational domain is shown in Fig. 1. The set-up followed the previous numerical studies (Kawamura et al.,
2002; Yu et al., 2008). The total height is 6D with D the cylinder diameter; and the portions of air and water are 2D and 4D,
respectively. The distance from the center of the cylinder to the outer boundary is 20D including a buffer zone for damping
wave reflections.
Three body-fitted cylindrical grids, coarse (C), medium (M), and fine (F), were used to investigate the influence of grid
resolution. The grid parameters are listed in Table 1. The grid points were clustered near the cylinder surface to capture the
boundary layer and flow separation. The grid close to the interface was also refined to resolve the interface deformations.
In order to investigate the effect of computational domain size, an additional complementary simulation with doubled
height (12D with 4D and 8D for air and water portions, respectively) using grid C1 ðNr  Ny  Nz ¼ 128  128  256Þ was
performed.
For simplicity, a Cartesian coordinate system is used for the discussion of results. The origin is the center of the cylinder
at the still free surface level, and the x, y, and z axes are defined in the streamwise, transverse, and vertical direction,
respectively. The instantaneous Cartesian velocity components u, ~ v,
~ and w~ are defined correspondingly. U, V, and W and u,
v and w indicate the mean velocity and velocity fluctuation of each Cartesian component, respectively.

3.2. Simulation conditions

In the present study, all variables are nondimensionalized with respect to D and the uniform inflow velocity U1 . Hence,
two dimensionless parameters, the Reynolds number and the Froude number, can be defined as
U1 D
Re ¼ , ð8Þ
n
with n the kinematic viscosity of the liquid phase, and
U1
Fr ¼ pffiffiffiffiffiffi : ð9Þ
gD
In this study, the flow at Re = 2.7  104 and Fr =0.8 is investigated. The Froude number is related to the deformation of the
interface and the present value 0.8 corresponds to a moderate interface deformation. In addition, a complementary case of
Fr = 0.2 is performed using the medium grid, to study the effect of the Froude number on the flow field near the interface
and only partial results, as stated in the text and figures, from the Fr = 0.2 case are discussed herein.
On the cylinder wall, the no-slip condition was imposed while the slip condition (i.e., zero velocity in the normal
direction to the face and the Neumann condition for the other directions) was used at the bottom (water side) and the top

Fig. 1. Schematic diagram of the computational domain and coordinate system.

Table 1
Grid parameters and hydrodynamic forces on an interface piercing cylinder.

Grid Nr  Ny  Nz CD C rms
L y+

C 128  128  128 0.959 0.195 1.927


M 256  128  128 0.983 0.216 0.966
F 512  128  128 0.984 0.220 0.485

Kawamura et al. (2002) 99  161  33 0.97 0.24


J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 5

(air side) of the computational domain. The radial outer boundary was divided into in- and outflow at y ¼ 903 and 2701, in
which y is the tangential angle starting from the downstream direction. A Dirichlet boundary condition was used for the
inflow boundary with uniform streamwise velocity U1 ¼ 1 and a convective boundary condition (@ui =@t þU1 ð@ui =@nÞ ¼ 0)
was used for the outflow boundary (Breuer, 1998).

3.3. Verification and validation

Fig. 2 shows the time histories of the drag and lift coefficients from the medium grid, along with the running mean of
the drag coefficient. The drag and lift coefficients are defined as
Drag Lift
CD ¼ 1
, CL ¼ 1
: ð10; 11Þ
r 2
2 L U1 DH
2
r
2 L U1 DH

Note these coefficients are based on the still water depth H= 4D and the water density rL in order to be consistent with the
previous studies (Kawamura et al., 2002; Yu et al., 2008).
The convergence of the running mean from the time history of CD was used to define the statistically stationary state.
The criterion for this convergence is that the fluctuations of the running mean are less than 1% of the mean drag coefficient
C D , which was ensured for all cases run in this study. For instance, the simulation on the medium grid reached statistically
steady state around t = 80. After the flow was converged, flow field data covering 16 vortex shedding cycles were collected
for statistics. Here the cycle is defined using the Strouhal number St ¼ fD=U1 with f the frequency and a typical Strouhal
number in this flow is St = 0.2. Fig. 3 shows the FFT of the CD and CL on the medium grid. For the FFT of CL, the highest peak is
around 0.2, corresponding to the Karman vortex shedding. The FFT of CD has a broad range of frequency content.

Fig. 2. Time histories of the drag and lift coefficients. Dotted line is the running mean of the drag coefficient.

1
10

0 CD
10
CL

-1
Magnitude

10

-2
10

-3
10

-4
10
0 1 2 3
St

Fig. 3. FFT of the drag and lift coefficients from the medium grid.
6 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

C D and C rms
L from three cases are listed in Table 1 with the fine grid results from Kawamura et al. (2002). Both C D and
C rms
L show monotonic convergence using refinement ratio rG =2. Table 2 tabulates results from grid studies for C D and C rmsL
following the procedure given by Xing and Stern (2010). For C D , the order of accuracy is pG = 2.39 with grid uncertainty
UG =0.09%. For C rms
L , the order of accuracy is pG = 4.19 and the uncertainities are much lower compared to the mean drag
coefficient with UG = 0.01%.
The mean streamwise velocities at x = 4.5 and y= 0 from the simulation are compared with the available experimental
data (Inoue et al., 1993) and the fine grid solution by Kawamura et al. (2002) in Fig. 4. The present results are in good
agreement with the measurement by a hot-film anemometer (Inoue et al., 1993) and the computation by Kawamura et al.
(2002). The decrease of streamwise velocity near the interface is clearly observed. The effect of grid resolution on the mean
velocity is also shown in this figure. The differences between the results from the three grids are not evident. The effect of
computational domain size is not significant either. Since the simulation results between different grids are very similar
and the amount of data from the fine grid is very large for post-processing, numerical results from the medium grid will be
presented in this study unless otherwise mentioned.
Profiles of the mean interface elevation and the rms of the interface fluctuations at two transverse planes are shown in
Fig. 5. They are in very good agreement with the experimental data (Inoue et al., 1993). However, the present simulation
under-predicted the depression of the interface in the near-wake region behind the cylinder. It also confirms the difference
between computational results from different grids or domain size is not significant.
Contours of the mean interface elevation are shown in Fig. 6, side by side with the previous measurement (Inoue et al.,
1993). The generation of bow wave in front of the cylinder, a large surface cavity behind the cylinder, and the diverging
Kelvin wake are evident. The simulation result is in good agreement with the experiment on the overall magnitude and
location of interface waves and the expansion angle of Kelvin wave, although the size of large depressed region behind the
cylinder was slightly under-predicted. The feature that the slope from the bow wave crest to the depressed region is almost
constant is preserved (Kawamura et al., 2002).
Fig. 6 also shows the rms of the interface fluctuations, along with that from the previous experiment. A good agreement
between the simulation and the experiment is observed on the overall distribution of the fluctuation intensity and the
location of its peak value. Most of the interface fluctuations are observed in the immediately downstream of the constant
slope region between the crest and the cavity. The peak fluctuation is observed near the edge of the flat bottom of the
surface cavity. In addition to this region, the simulation result gives some slight fluctuations in front of the cylinder,
probably due to the presence of the necklace vortices. Nevertheless, the bow wave in front of the cylinder and the constant
slope region between the crest and the cavity retain a fixed wave shape with negligible fluctuations. But remarkable
interface fluctuations are generated in the near-wake surface cavity region.

Table 2
Grid studies for C D and Crms
L .

Parameter e21 e32 RG PG UG (%)

CD  0.004  0.021 0.190 2.39 0.09


Crms
L  0.0013  0.0238 0.055 4.19 0.01

Grid C1
1 Grid C
Grid M
Grid F
0
Experiment
Kawamura
-1
z

-2

-3

-0.5 0 0.5 1 1.5


U

Fig. 4. Vertical profiles of the mean streamwise velocity at x= 4.5 and y= 0.


J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 7

0.3 0.3

0.2 0.2

0.1 0.1
hrms hrms
0 0

-0.1 -0.1 hmean


hmean
-0.2 Grid C 1 -0.2 Grid C 1
Grid C Grid C
-0.3 Grid M -0.3 Grid M
Grid F Grid F
-0.4 Experiment -0.4 Experiment
-0.5 -0.5
0 1 2 3 4 5 6 1 2 3 4 5
y y
Fig. 5. Profiles of the mean interface elevation and the rms of the interface fluctuations: (a) x =0.9 and (b) x= 2.0.

LES x=0.9 x=2.0 LES x=0.9 x=2.0

0.03 -0.15

0.0
0.01 5
0. 1
5

0.15
0.05
-0.30

0.03
-0.15

0.01

Experiment Experiment

Fig. 6. Comparison of the mean interface elevation (left) and the rms of the interface fluctuations (right) around the cylinder at Fr= 0.8.

4. Results and discussion

4.1. Instantaneous flow

4.1.1. Air–water interface


Fig. 7 shows the instantaneous air–water interfaces from the simulations. For the Fr =0.8 case, the bow waves in front of
the cylinder and the Kelvin waves displaced away from the cylinder by the separation region are clearly observed. The
wavelength of the Kelvin wave system approximately corresponds to the theoretical value, i.e. 2pFr2  4. The rough
surface, immediately up- and downstream of the cylinder, indicates the existence of vortical structures below the interface
(Sarpkaya, 1996). On the other hand, the upstream bow waves, the cavity region in the downstream of the cylinder and the
Kelvin waves are almost unnoticeable in the Fr =0.2 case.
In addition, the mean interface elevation and the rms of the interface fluctuations from the Fr = 0.2 case are shown in
Fig. 8. The height of the wave crest in front of the cylinder and the depth of the cavity region behind the cylinder are much
reduced and a nearly flat interface is observed. The location of the region with the most intensive interface fluctuations is
quite similar to the Fr= 0.8 case, but the rms value of the interface fluctuations is greatly reduced.

4.1.2. Vorticity field


In order to assess the effect of the interface on the cylinder wake, contours of the instantaneous vertical vorticity at
different horizontal planes and the interface are presented in Fig. 9. Organized vortex shedding, which is very similar to
8 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

Fig. 7. Instantaneous interface elevation at Fr =0.8 (top) and Fr= 0.2 (bottom).

LES: Fr=0.2 x=0.9 x=2.0 LES: Fr=0.2 x=0.9 x=2.0

0.
01 - 0. 0
1

-0.05
0.005

0.015

0.05 0.02
5

Fig. 8. The mean interface elevation (left) and the rms of the interface fluctuations (right) around the cylinder at Fr =0.2.

that from an infinitely long cylinder in a single phase flow, is observed in the deep flow. In addition to the large vortex
shedding, smaller vortices on the shear layers are also observed. The shear layers separated from two sides of the cylinder
start becoming unstable via the Kelvin–Helmholtz instability due to the velocity difference within the shear layer and
generate the small-scale vortices. As the interface is approached, vortices appear over a larger range of wake. At the
interface, the organized large-scale vortex shedding is no longer present and only small-scale vortices are observed mostly
at the edge of the separated region. The shear layers from the two sides of the cylinder digress from each other and no
longer interact as in the deep flow. Moreover, the distribution of the vortical structures at the interface approximately
coincides with the region of high rms fluctuations in Fig. 6(left), which indicates that there is a strong correlation between
the swirling motion of the vortical structures and the interface fluctuations.
In addition, contours of the instantaneous vertical vorticity from Fr= 0.2 are shown in Fig. 10. Necklace vortices, which
were observed in front of the cylinder for higher Froude numbers, are not seen in this case. On the interface, vortices are
present in an area smaller than the Fr= 0.8 case and large-scale vortices affected by the vortex shedding are observed.
Hence the attenuation of vortex shedding at the interface is not very significant for a lower Froude number case. This is
consistent with Vlachos and Tellionis’s (2008) observation that the increase of Froude number results in the suppression of
vortex shedding at the interface.
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 9

2 2

1 1

0 0
y

y
-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x
2 2

1 1

0 0
y

y
-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x

Fig. 9. Instantaneous vertical vorticity at the interface and horizontal planes for Fr = 0.8: (a) on the interface; (b) z =  0.5; (c) z=  1; and (d) z=  3.5.
Contour interval is 1.2.

2 2

1 1

0 0
y

-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x
2 2

1 1

0 0
y

-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x

Fig. 10. Instantaneous vertical vorticity at the interface and horizontal planes for Fr = 0.2: (a) on the interface; (b) z =  0.5; (c) z =  1; and (d) z=  3.5.
Contour interval is 1.2.

4.1.3. Vortical structures


To further identify large-scale flow structures, isosurfaces of the second invariant of the velocity gradient tensor rv
(Hunt et al., 1988) from both cases are shown in Fig. 11. At Fr= 0.8, the dominant coherent structures are the elongated
quasi-vertical vortices away from the interface, which are originated from the shear-layer instability. As the interface is
approached, the structures are inclined to the outward transverse direction. On the other hand, the prevalent vortical
10 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

Fig. 11. Instantaneous vortical structures identified by the second invariant of the velocity gradient tensor Q= 2.0: (a) Fr =0.8 and (b) Fr= 0.2.

structures inside the separated region are small and parallel to the interface. Most of these structures are located slightly
below the interface, but some structures exist at the interface, which indicates the rotation and deformation of the
interface. This observation is consistent with previous work (Yu et al., 2008). However, the figure for the Fr =0.2 case shows
that the vortical structures parallel to the interface diminish due to the reduced interface deformation and vertical vortices
can maintain their direction even near the interface at a smaller Froude number.

4.2. Hydrodynamic forces

Fig. 12 shows the distribution of sectional C D and CrmsL along the cylinder. The magnitude of C D increases to its local
maximum value at the interface due to the enhanced momentum mixing and turbulence generation, which is consistent
with (Yu et al., 2008). The sectional Crms
L clearly shows the attenuation of vortex shedding at the interface. At Fr = 0.2, vortex
shedding is stronger, especially, near the interface; and the enhanced mixing near the interface and the attenuation of
vortex shedding appear in a narrower region than the larger Froude number case.

4.3. Mean flow

4.3.1. Phase-averaged vorticity


In order to investigate the usefulness of mean flow analysis for the present time-dependent flow, vertical vorticities
from two different depths were phase averaged with the deep-level vortex shedding frequency. Fig. 13 shows the phase-
averaged vertical vortices at two planes, z =  0.5 and 3.5, and at two phase angles, jA and jA þ T=2, where T is the vortex
shedding frequency at the deep region. The phase difference between (a), (c) and (b), (d) is T/2. While the shear layer
pattern in the deep region is in opposite direction between (c) and (d), the change of the vorticity distribution close to the
interface is not significant, especially for the separated shear layer immediately downstream of the cylinder. Hence the
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 11

-1 Grid C
cLrms Grid M cD

z
Grid F
Fr=0.2
-2

-3

0 0.2 0.4 0.6 0.8 1 1.2 1.4


cD ,cLrms

Fig. 12. Sectional drag and lift coefficients.

2 2

1 1

0 0
y

-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x
2 2

1 1

0 0
y

-1 -1

-2 -2
-2 0 2 4 6 -2 0 2 4 6
x x

Fig. 13. Phase-averaged vertical vorticity at horizontal planes. Contour interval is 1.2: (a) at z=  0.5 and phase jA ; (b) at z =  0.5 and phase jA þ T=2;
(c) at z =  3.5 and phase jA ; and (d) at z=  3.5 and phase jA þ T=2.

effect of vortex shedding frequency on the flow structures near the interface is convinced to be small, and mean flow
analysis might be helpful for the analysis of flow field near the interface.

4.3.2. Boundary layer


Fig. 14 shows the friction and pressure coefficients
tw
Cf ¼ 1 ð12Þ
r 2
2 L U1

and
pp1
Cp ¼ 1 ð13Þ
2rL U1
2
12 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

0.03 1
z=-0.5 Exp.(Re=2.0104)
z=-2 z=-0.5
z=-3.5 0.5
0.02 z=-2
z=-3.5
0

Cp
0.01
Cf

-0.5

0
-1

-0.01 -1.5
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Angle from stagnation point (divided by π) Angle from stagnation point (divided by π)

Fig. 14. Force coefficients on the cylinder surface: (a) friction coefficient and (b) pressure coefficient.

-0 . 0 9

8
-0 . 1
0.09

0.6

0.5
y

0.4

0.3
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
x

Fig. 15. Close-up view of the mean interface elevation around the cylinder boundary.

on the cylinder wall at three depths. The boundary layer separation, indicated by a change of Cf from positive to negative, is
moved from f ¼ 0:462p in the deep flow to f ¼ 0:514p near the interface. The wall pressure distributions in the deep flow
are in good agreement with the single phase experimental data (Norberg, 1992) of Re = 2.0  104 up to the separation point,
but the present simulation yields higher pressure values than the single phase experiment at the rear part of the cylinder.
While the pressure in the separated region of the single phase flow shows a slightly reduced value than that at the
separation point, continuously increasing values are observed in the deep flow in the present simulation. In addition, the
pressure distribution near the interface is quite different. The pressure gradient along the cylinder is significantly
decreased due to the negative interface elevation slope near the cylinder. The adverse pressure gradient downstream is
also lessened by the reduced upstream pressure gradient and responsible for the displacement of the separation point
further downstream near the interface. Fig. 15 is the close-up view of the interface elevation near the cylinder boundary.
Along the cylinder circumference, a steeper slope of the interface elevation is observed inside the boundary layer than that
of the outside. Immediately after the position of an angle p=2 from the front stagnation point, the interface elevation starts
to increase inside the boundary layer while it is still decreasing outside. The location where the wave elevation starts to
increase is consistent with the boundary layer separation point. This rising interface elevation along the cylinder
downstream made the pressure in this region keep increasing, which is shown in Fig. 14.

4.3.3. Separation pattern


Fig. 16 shows the separation pattern of the mean flow. The separated shear layer was visualized approximately using
the isosurfaces of the stagnation Cp near the interface immediately before the separation, assuming that the flow outside
the separation region is inviscid (Kandasamy et al., 2009). Inside the separated region, the mean vortex core lines are extracted
using the vortex core identification technique (Sujudi and Haimes, 1995). The mean vertical vortices (V 1) are due to the vortex
shedding in the deep flow. As the interface is approached, these vortices are inclined and attached on the cylinder wall. In
addition, two kinds of vortex core lines are observed inside the separated region near the interface: (a) the mean streamwise
vortices located near the edge of the separation region immediately under the interface (V 2); (b) a V-shaped mean vortex
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 13

Fig. 16. Mean separation pattern with the vortex core lines.

0.8

0.6

0.4
U

0.2 Interface
z=-1.5
0 z=-2.5
z=-3.5
-0.2 Exp. 1 (Re=3900)
Exp. 2 (Re=3900)
-0.4 Exp. (Re=1.4×105)

0 2 4 6 8 10
x

Fig. 17. Mean streamwise velocity on the centerline in the wake.

observed inside the separated region (V 3). V 3 is inclined and approaches to V 2 as it goes downstream. V 3 is located near the
high gradient region of the mean streamwise velocity where a significant amount of turbulent kinetic energy is produced.

4.3.4. Mean velocity


Fig. 17 shows the mean streamwise velocity U from the Fr = 0.8 case on the centerline for several different vertical
locations, along with the single-phase measurements at Re= 3900 and 1.4  105, respectively (Cantwell and Coles, 1983;
Ong and Wallace, 1996; Lourenco and Shih, 1993). To the best knowledge of the authors, there is no velocity measurement
available in the literature on the single phase flow past a cylinder at a similar Reynolds number as the present case. Hence
we will compare the present simulation results with the available experimental data in the subcritical Reynolds number
regime. The far-wake centerline streamwise velocity for single-phase flow shows independence on Re because the
measurement profiles (x 4 4) from two different Reynolds numbers approach almost same value (U=U1  0:8). On the
other hand, the near-wake profile varies significantly. The recirculation zone immediately after the cylinder for the high
Reynolds number flow is significantly smaller than in the low Reynolds number case. The near-wake profile of the mean
centerline streamwise velocity away from the interface (z=  3.5) from the present simulation falls between the
experiments of lower and higher Reynolds numbers, whereas the far-wake profile approaches the same value as all
the experiments. In the present simulation, the recirculation region increases substantially as the profile approaches the
interface. The streamwise length of the recirculation region at the interface (x= 0.5–3.56) is more than triple of the deep
flow recirculation zone (x =0.5–1.33). Its far-wake velocity magnitude is also remarkably smaller than the deep flow region.
In addition, the small positive streamwise centerline velocity zone is observed to exist immediately downstream of the
cylinder (see the top-left close-up view of the interface centerline velocity), which means a complicated interface topology
with the presence of nodal and saddle points (Kandasamy et al., 2009).
14 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

The mean streamwise velocities in the wake for several vertical locations and the interface are compared in Fig. 18,
along with the experiments at Re =3900 and 1.4  105, respectively (Cantwell and Coles, 1983; Lourenco and Shih, 1993).
The mean streamwise velocity profiles for the horizontal zones away from the interface (z =  1.5 to  3.5) located between
two experimental profiles of lower and higher Reynolds numbers (e.g., x= 1.06). However, a significantly different profile
with increased wake width is obtained for the mean streamwise velocity at the interface. The profiles downstream
(x =2.02) show that the mean streamwise velocity at the interface wake still maintains similar values as upstream (x= 1.06),
although the mean streamwise velocities in the deep flow are noticeably recovered. Hence the wake at the interface is
greatly expanded in both the streamwise and transverse directions. In the far wake, the interface still has a significant
effect on the wake (not shown). Consequently, velocity recovery in the wake region at the interface is substantially slower
than that in the deep flow.
Fig. 18(b) also shows that the pattern of the mean transverse velocity profiles at the interface is quite different from that
in the deep flow. The amplitude of the profiles decreases as the interface is approached from the deep level but increases
again near the interface. Also, it is noticed that the sign of the mean transverse velocity at the interface is inverted from
that in the deep flow. Whereas the mean transverse velocity points toward the symmetric plane in the deep flow, the
surface wave still spreads the wake away from the symmetric plane with reversed mean transverse velocities. Note the
location of the maximum magnitude of the outward mean transverse velocity is changed from y= 1.1 in the cross-stream
plane x= 1.06 to y=  1.5 in the plane x = 2.02. Also, the outward mean transverse velocity becomes very small immediately
below the interface (z =  0.5), indicating it is significant at the interface only and derived by the streamwise vorticity
generated near the interface. The outward mean transverse velocity observed at the edge of the separated region is
primarily responsible for the attenuation of the Karman vortex shedding at the interface. It restricts the two shear layers
from interacting with each other and thus generating organized vortex shedding. The mean transverse velocity profile from
the Fr = 0.2 case is also shown in Fig. 18. The mean transverse velocity profiles for the deep level are almost the same as for
the larger Froude number case. However, significantly reduced outward transverse velocity is observed at the interface,
which indicates the large transverse velocity observed at Fr = 0.8 is from the effect of Froude number.
To see the mean flow pattern at the interface and its relation with the separated region, contours of the mean Cartesian
velocity components at the interface overlapped with the mean pathlines are shown in Fig. 19. The pathline pattern

2
x=1.06 x=1.06

1 0

0 Exp. (Re=1.4 ×105)


Interface
x=2.02
U

z=-0.5
z=-1.5
-1 -1 z=-2.5
Exp. (Re=3900) x=2.02 z=-3.5
Exp. (Re=1.4 ×105)
Interface
-2 z=-0.5
z=-1.5
z=-2.5
z=-3.5
-3 -2
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
y y

x=1.06

Exp. (Re=1.4 ×105)


Interface
V

z=-0.5
z=-1.5
-1 z=-2.5
x=2.02 z=-3.5

-2
-3 -2 -1 0 1 2 3
y

Fig. 18. Mean velocity at two locations in the wake: (a) streamwise velocity at Fr= 0.8; (b) transverse velocity at Fr = 0.8; and (c) transverse velocity
at Fr= 0.2.
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 15

2 2
y

y
1 1

0 0
-2 0 2 4 -2 0 2 4
x x

2 2

y
y

1 1

0 0
-2 0 2 4 -2 0 2 4
x x

Fig. 19. Contours of the mean velocity at the interface: (a) streamwise velocity for Fr = 0.8; (b) transverse velocity for Fr= 0.8; (c) vertical velocity for
Fr= 0.8; and (d) transverse velocity for Fr= 0.2.

showing the separation at the interface is an open type. Immediately outside the front separated region, the mean
streamwise velocity changes significantly in the transverse direction. In addition, the outward transverse velocities noted
in Fig. 18 are observed in the same region. Note the mean vertical velocity is negligible inside the separated region
although there are significant interface fluctuations in this region. Contours of the mean transverse velocity at the interface
from the Fr= 0.2 case are also shown in Fig. 19. As expected, the outward mean transverse velocity is reduced and only
appears near the cylinder and the width of the wake is also reduced at a smaller Froude number.
Fig. 20 shows contours of the mean streamwise velocity at two cross-stream (yz) planes in the near wake. The interface
position and the location of the cylinder are marked in the figure, and only half of the domain is shown. At the x = 1 plane,
the width of the wake increases significantly near the interface, whereas the width is almost constant in the deep flow.
Since the x= 1 plane is in the recirculation region, negative mean streamwise velocities are observed in the deep flow and
near the interface. However, there is no negative streamwise velocity at the narrow wake region around z= 1. Hence, the
width of the recirculating zone increases as the interface is approached. At the plane x= 2.5, a negative streamwise velocity
is still observed near the interface, whereas the velocities in the deep flow increase to positive values, which confirms the
slower velocity recovery in the wake at the interface. At this cross-stream plane, the width of the wake is substantially
increased near the interface and in the deep flow. The narrowest wake region locates at a slightly lower position (z =  1.1)
in this plane. This observation is consistent with Kawamura et al. (2002).

4.3.5. Mean vorticity


The mean streamwise vorticity is responsible for the attenuation of the vortex shedding and the significantly increased
wake width near the interface. In addition, the large outward mean transverse velocity near the interface shown in Fig. 18
is induced by the streamwise vorticity near the interface. The mean streamwise vorticity is shown in Fig. 20(b). The
locations of the vortices and the shapes of wakes correlate well, indicating the effect of secondary swirl of the vortical
structures on the wake structures (Kawamura et al., 2002). At the x= 1 plane, three vortices with alternative rotation
directions are observed behind the cylinder, which are responsible for the complicated serpentine wake pattern. In
addition, a pair of strong counter-rotating vortices is observed across the interface around y=  1. This vortex pair is
responsible for the increased wake width and the large outward transverse velocity near the interface. Weaker vortices are
also observed in the wake of both the air and water regions. On the other hand, at the x =2.5 plane, vortices are only seen
near the interface and in the air region.
Note that the mean transverse vorticity component is significant near the interface only. Since the magnitude of the
mean transverse vorticity is larger than that of the mean streamwise vorticity at the interface between the two surface-
parallel vorticity components, the mean transverse vorticity is more responsible for the interface fluctuations. Contours of
the mean transverse vorticity at the interface are shown in Fig. 21. High vorticity regions near the cylinder and the Kelvin
waves are due to the change of interface elevation. In addition, significant mean transverse vorticity is observed inside the
separated region. The location of this region matches well with the region of high interface fluctuations, which indicates
that the mean transverse vorticity is responsible for the interface fluctuations.
The vertical distributions of the mean vertical vorticity component in the deep flow and the air region are due to the
Karman vortex shedding and the shear-layer instability. However, its distribution near the interface is inclined outward in
16 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

1 1 1 1

0 0 0 0

-1 -1 -1 -1
z

z
-2 -2 -2 -2

-3 -3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
y y y y

1 1 1 1

0 0 0 0

-1 -1 -1 -1
z

z
-2 -2 -2 -2

-3 -3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
y y y y

Fig. 20. Contours of the mean flow at two cross-stream planes: (a) streamwise velocity at x= 1.0; (b) streamwise vorticity at x = 1.0; (c) transverse vorticity
at x = 1.0; (d) vertical vorticity at x = 1.0; (e) streamwise velocity at x = 2.5; (f) streamwise vorticity at x =2.5; (g) transverse vorticity at x =2.5; and
(h) vertical vorticity at x = 2.5. Contour intervals are 1.0 except for streamwise vorticity, a contour interval 0.5 is used.

2
y

0
-2 0 2 4
x

Fig. 21. Contours of the mean transverse vorticity at the interface. Contour interval is 1.0.

the transverse direction, due to the outward mean transverse velocity generated near the interface. The location of the high
magnitude vorticity matches with the high gradient region of the mean streamwise velocity.

4.4. Reynolds stresses

Fig. 22 shows contours of the Reynolds stresses at two cross-stream planes in the near wake. For the streamwise
Reynolds normal stress uuuu, at the x= 1 plane, peak values are produced near the interface and along the separation region
in the deep flow where the mean streamwise velocity is recirculated. Unlike the mean streamwise velocity, uuuu does not
show a two-dimensional distribution in the deep flow. Note that large uuuu is observed where the mean streamwise velocity
gradients are very high (see Fig. 20), which indicates a region of high turbulent kinetic energy production (uul uuk Ul,k ).
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 17

1 1 1 1

0 0 0 0

-1 -1 -1 -1
z

z
-2 -2 -2 -2

-3 -3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
y y y y

1 1 1 1

0 0 0 0

-1 -1 -1 -1
z

z
-2 -2 -2 -2

-3 -3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
y y y y

Fig. 22. Distribution of the Reynolds stresses at two cross-stream planes: (a) uuuu at x= 1.0; (b) vuvu at x =1.0; (c) wuwu at x =1.0; (d) uuvu at x= 1.0; (e) uuuu at
x =2.5; (f) vuvu at x= 2.5; (g) wuwu at x= 2.5; and (h) uuvu at x = 2.5. Contour intervals are 0.025 for uuuu and vuvu and 0.01 for wuwu and uuvu.

In addition, uuuu is increased locally near the edge of the separated region at the interface, due to the increased velocity
gradient by the outward transverse velocity generation. Outside the separation region, uuuu increases as the interface is
approached; and the behavior of uuuu changes following the interface position near the interface. However, the interfacial
behavior in the separation region is different. First, uuuu decreases as the interface is approached. Then it increases slightly
in the region with substantial interface fluctuations (in the x= 1 plane) or continuously decreases (in the x= 2.5 plane).
Relatively uniform distribution of uuuu inside the separated region at x =1 is probably owing to the enhanced mixing by the
complex streamwise vorticity inside the separated flow region noted in Fig. 20.
For the transverse Reynolds normal stress vuvu, only one peak is observed in the deep flow near the symmetry plane
while double peaks are observed near the interface. The single peak in the deep flow is due to the vortex shedding by the
interaction of the shear layers separated from both sides of the cylinder. The behavior of vuvu near the interface shows a
similar trend with uuuu. In the separated flow region, vuvu decreases as it approaches the interface.
The vertical Reynolds normal stress wuwu has a large magnitude at the interface and is directly related with the interface
fluctuations, which result in the significant streamwise vorticity and outward transverse velocity at the interface. In
addition, additional peaks are observed near the recirculation zone in the deep flow. Akilli and Rockwell (2002) observed a
strong upwelling motion of fluid through the centers of the mean vortices at the bottom region in their shallow water
experiment. wuwu at this deep region in the present work is also owing to the up- and downward flow through the vortex
centers during the periodic vortex shedding. The feature of wuwu near the interface is very similar to that of uuuu.
Fig. 22 also shows contours of the Reynolds shear stress uuvu at two cross-stream planes in the near wake. At the interface,
this Reynolds stress term has an insignificant magnitude in the wake region because the shear layers from both sides deviate
from each other and the interaction between them is restrained. Yu et al. (2008) also observed reduction of shear stress near the
interface. Similarly, the Reynolds shear stresses related to the vertical velocity (uuwu and vuwu, distributions are not shown here)
are one order of magnitude smaller and negligible at the interface comparing to those in the deep flow.
Fig. 23 shows contours of the vertical Reynolds normal stress wuwu at the interface. Note the region of high gradients of
wuwu near the edge of the separated region matches very well with the locations of the high outward mean transverse
velocity in Fig. 18. The gradients of vuvu at the interface (not shown) are not so large as wuwu. Hence, the lateral gradient of
the difference between vuvu and wuwu, which is proportional to the Froude number, is mostly responsible for the streamwise
18 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

2 0.01

y
0.03
1 0.08

0
-2 0 2 4
x

2
y

1
3
0.0 0.01

0
-2 0 2 4
x

Fig. 23. Distribution of the Reynolds normal stress wuwu at the interface: (a) Fr= 0.8 and (b) Fr= 0.2.

vorticity and presumably the mean outward transverse velocity at the interface in the present study. As a consequence, in a
case with a significant Froude number, the formation of the separated fluctuating region behaves as an obstacle, which
accelerates the flow around it (Vlachos and Tellionis, 2008).
Contours of wuwu at the interface for the Fr= 0.2 case are also shown in Fig. 23. The distribution area and magnitude are
much reduced comparing to the Fr= 0.8 case. The lateral gradients of wuwu are small for most of the region except near the
cylinder.

4.5. Vorticity transport

Longo et al. (1998) explained the physical mechanism for the mean streamwise vortices generated in solid-interface
juncture flow using the vorticity transport equation, which can be derived by taking the curl of the time-averaged Navier–
Stokes equation. For a steady flow of constant density, the equation can be written as
 
@Ox @Ox @Ox
U þV þW ðAÞ
@x @y @z
 
@U @U @U
¼ Ox þ Oy þ Oz ðBÞ
@x @y @z
 2 
@ Ox @2 Ox @2 Ox
þn þ þ ðCÞ
@x2 @y2 @z2
 
@ @uuvu @uuwu
þ  ðDÞ
@x @z @y
2
@
þ ðvuvuwuwuÞ ðEÞ
@y@z
 2 
@ @2
þ  vuwu, ðFÞ ð14Þ
@z2 @y2
where Ox , Oy , and Oz are the streamwise, transverse, and vertical components of the mean vorticity, respectively. Term (A)
in Eq. (14) represents the material derivative of the mean streamwise vorticity. The first term of term (B) is the vorticity
amplification by the streamwise stretching, while the other terms provide vortex-line bending effects. Term (C) suggests
the vorticity damping by the viscous diffusion, and the other terms (D), (E), and (F) are the vorticity production by
inhomogeneities in the Reynolds stress field (Launder and Rodi, 1983). In addition, there are other terms from the density
discontinuity in the present study, but it is expected negligible except close to the interface.
The dominant terms in the mean streamwise vorticity transport equation at the cross-stream plane x =1 are the y and z
components of terms (B) and (E) shown in Fig. 24. Two terms by the vortex bending are of a similar magnitude, but of
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 19

30 30 7
24 24 5.6
1 18 1 18 1 4.2
12 12 2.8
6 6 1.4
0 0 0
-6 -6 -1.4
0 -12 0 -12
0 -2.8
-18 -18 -4.2
-24 -24 -5.6
-30 -30 -7
-1 -1 -1

z
z
z

-2 -2 -2

-3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0
y y y

5 5 5
4 4 4
1 3 1 3 1 3
2 2 2
1 1 1
0 0 0
-1 -1 -1
0 -2
0 -2 0 -2
-3 -3 -3
-4 -4 -4
-5 -5 -5
-1 -1 -1
z

z
z

-2 -2 -2

-3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0
y y y

5 5 2
4 4 1.6
1 3 1 3 1 1.2
2 2 0.8
1 1 0.4
0 0 0
-1 -1 -0.4
0 -2
0 -2 0 -0.8
-3 -3 -1.2
-4 -4 -1.6
-5 -5 -2
-1 -1 -1
z

z
z

-2 -2 -2

-3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0
y y y
Fig. 24. Dominant source terms for the mean vorticity at the cross-stream plane x= 1.0 for Fr =0.8: (a) y component of term (B) for streamwise vorticity;
(b) z component of term (B) for streamwise vorticity; (c) term (E) for streamwise vorticity; (d) z component of term (B) for transverse vorticity; (e) term
(E) for transverse vorticity; (f) term (F) for transverse vorticity; (g) term (F) for vertical vorticity; (h) z component of term (B) for vertical vorticity; and
(i) term (E) for vertical vorticity.

opposite signs, so the total effect of them is canceled out. Therefore, the remaining (E) term @2 =@y@zðvuvuwuwuÞ is the main
production mechanism of the mean streamwise vorticity. The vertical and transverse gradients of the difference between
vuvu and wuwu are responsible for the generation of the streamwise vorticity near the interface and presumably cause the
outward transverse velocity at the interface.
The dominant terms for the mean transverse vorticity at the cross-stream plane x= 1 are the z component of terms (B),
(E), and (F). The z component of term B, Oz ð@V=@zÞ, is from the bending of vertical vorticity, which is dominant in the deep
20 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

flow due to the vortex shedding and shear-layer instability, by the outward mean transverse velocity generation near the
interface. In other words, the swirling motion of the vortex shedding and shear-layer instability in the deep flow are
changed to interface fluctuations near the interface via vortex bending. The other sources for the mean transverse vorticity
are terms (E) and (F), ð@2 =@x@zÞðwuwuuuuuÞ and ð@2 =@x2 @2 =@z2 ÞðuuwuÞ, respectively.
The dominant terms for the vertical vorticity component are also shown in Fig. 24. In the deep flow, term (F),
ð@2 =@y2 @2 =@x2 ÞðuuvuÞ is the dominant source for the mean vertical vorticity. Hence, the present analysis using the vorticity
transport equation confirms that the shear stress is primarily responsible for the Karman vortex shedding and shear-layer
instability shown in the deep flow. The dominant terms for the mean vertical vorticity near the interface are the z
component of terms (B) and (E). The z component of term (B), Oz ð@W=@zÞ, is the vortex stretching of the vertical vorticity.
Term (E), ð@2 =@x@yÞðuuuuvuvuÞ, is from the anisotropy between the streamwise and transverse Reynolds normal stresses.
For the Fr =0.2 case, the streamwise vorticity and the dominant sources in Eq. (14) at x =1.0 are shown in Fig. 25.
Compared with the Fr = 0.8 case, the vorticity strength is much reduced, supporting the fact that the outward transverse
velocity is driven by the streamwise vorticity. The strength of the dominant sources for the streamwise vorticity is also
greatly reduced. Same as the Fr = 0.8 case, the y and z components of terms (B) and (E) are dominant sources for the
streamwise vorticity. Similarly, two terms from vortex bending are canceled. Hence, the term (E) is the main mechanism
for the streamwise vorticity, but its strength is greatly reduced.
Interestingly, for a flat interface with negligible fluctuations, which corresponds to a zero or very low Froude number
case, Walker (1997) suggested that the outward mean transverse velocity at the interface, or the origin of the surface
current, is owing to the transverse gradient of the Reynolds-stress anisotropy, @=@yðwuwuvuvuÞ, near the interface. Then in
Hong and Walker (2000), a set of Reynolds-averaged Navier–Stokes equations for free-surface flows were derived to
extend the order-of-magnitude analysis in Walker (1997) to high Froude numbers. And their results of two cases for a
turbulent jet issuing from a circular nozzlepof ffiffiffiffiffiffidiameter d in parallel beneath a free surface at a uniform velocity Ue, one at
Fr = 1.0 and the other at Fr = 8.0 (Fr ¼ Ue = gd), showed the Reynolds stress anisotropy is smaller in the higher Froude
number case than that in the lower Froude number case, but the much larger free-surface fluctuations in the former case
compensate its weaker Reynolds-stress anisotropy and result in even stronger surface current.
In the present study, instead of the Reynolds-averaged Navier–Stokes equations, we adopt the Reynolds-averaged
vorticity transport equations, following the lines of our previous study (Longo et al., 1998). But the vorticity generation
terms due to the presence of air–water interface are ignored in the analysis for simplicity. It should be noted that, although
the attenuation of the vortex shedding at the interface can be attributed to the outward mean transverse velocity, as shown
in the previous section, it can also be connected to the mean streamwise vorticity as well and the mechanism behind this
phenomenon can be explained using the mean streamwise vorticity via the mean vorticity transport equation. For the
small Froude number case, we obtain results consistent with Walker’s order-of-magnitude analysis of the momentum
equations, as both analyses give the identical term as the main mechanism for the outward mean transverse velocity and
the interface fluctuations are small. However, for the larger Froude number case, our results show a higher Reynolds stress
anisotropy as well as a stronger surface current; on the other hand, it is reasonable to expect a higher contribution to the
vorticity generation from the interface fluctuations in the larger Froude number case. Therefore, although the order-of-
magnitude analysis method using the Reynolds-averaged Navier–Stokes equations in Hong and Walker (2000) is very
general and useful, their conclusions on the origin of surface current in submerged horizontal turbulent jet spreading could
be questionable when extended to other flows, such as the turbulent wakes from a surface-piercing cylinder as studied

15 15 1.5
12 12 1.2
1 1 9 1 9 1 0.9
6 6 0.6
3 3 0.3
0 0 0
-3 -3 -0.3
0 0 -6
0 -6
0 -0.6
-9 -9 -0.9
-12 -12 -1.2
-15 -15 -1.5
-1 -1 -1 -1
z

-2 -2 -2 -2

-3 -3 -3 -3

-2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
y y y y
Fig. 25. Streamwise vorticity and dominant source terms for Fr= 0.2 at x = 1.0: (a) streamwise vorticity at x= 1.0; (b) y component of term (B) for
streamwise vorticity; (c) z component of term (B) for streamwise vorticity; and (d) term (E) for streamwise vorticity. Contour level of streamwise vorticity
is 0.5.
J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22 21

herein. It has to be pointed out that the two Froude numbers considered in the present study are still relatively low,
although both have non-negligible interface fluctuations. Cases with much higher Froude numbers and terms due to
interface fluctuations in the Reynolds-averaged equations have to be considered for a more complete spectrum of sources
of surface current, which are beyond the scope of the present study.

5. Conclusions

Large-eddy simulation of the flow past an interface piercing circular cylinder at Re = 27 000 and Fr =0.2 and 0.8 has been
performed using a level-set/ghost-fluid method for the sharp interface treatment of air–water interface and a Lagrangian
dynamic SGS model for the dynamic modeling of the eddy viscosity in inhomogeneous complex flows. At this subcritical
Reynolds number, the latter has some particular advantages over the usual Smagorinsky SGS model for the realistic
prediction of the laminar separation from the cylinder and the transition to turbulence in the wake. The excellent
agreement between the numerical results and the available experimental data demonstrates the accuracy of the
simulation in this work.
The present study shows vortical structures are significantly changed at the interface. In the deep flow, organized
periodic vortex shedding is observed. Near the interface, the organized vortex shedding disappears and small structures
inclined along the interface are observed. The two shear layers deviate from the symmetric vertical plane such that there
are no more direct interactions between them. The flow field is also remarkably changed near the interface. Separation is
enhanced due to the reduced adverse pressure gradient by the negative interface elevation slope along the cylinder. The
distribution of the mean velocity near the interface is also significantly altered from that in the deep flow. In addition, the
size of the separated region is substantially increased in the streamwise and transverse directions. The streamwise
vorticity and outward transverse velocity generated at the edge of the separated region are responsible for the increased
width of the separated region and the attenuation of vortex shedding near the interface. The lateral gradient of the
difference between the vertical and transverse Reynolds normal stresses, increasing with the Froude number, is
responsible for the streamwise vorticity and outward transverse velocity generated at the interface. Turbulence kinetic
energy is also significantly generated at the edge of separated flow region.
The present work provides discussions on the interaction between the vortical structures and the interface, using one
case with experimental data available in the literature. As a first step toward the deep understanding of the detailed
hydrodynamic mechanism, this study is limited with regard to the range of Reynolds and Froude numbers considered and
the neglect of terms related to the fluctuations of the interface position in the Reynolds-averaged vorticity transport
equations. The effect of the Reynolds and Froude number variation and the inclusion of additional terms due to the
interface fluctuations in the Reynolds-averaged equations warrants further study for a more complete understanding of
vortex interface interactions.

Acknowledgments

This work was supported by research Grants N00014-01-1-0073 and N00014-06-1-0420 from the Office of Naval
Research (ONR), with Dr Patrick Purtell as the program manager. The simulations presented in this paper were performed
at the Department of Defense (DoD) Supercomputing Resource Centers (DSRCs) through the High Performance Computing
Modernization Program (HPCMP). We are grateful to Professor Nobuhiro Baba at Osaka Prefecture University, Japan for
providing us with the experimental data figures.

References

Akilli, H., Rockwell, D., 2002. Vortex formation from a cylinder in shallow water. Physics of Fluids 14, 2957–2967.
Breuer, M., 1998. Large eddy simulation of the subcritical flow past a circular cylinder: numerical and modeling aspects. International Journal for
Numerical Methods in Fluids 28, 1281–1302.
Cantwell, B., Coles, D., 1983. An experimental study on entrainment and transport in the turbulent near wake of a circular cylinder. Journal of Fluid
Mechanics 136, 321–374.
Choi, H., Moin, P., 1994. Effects of the computational time step on numerical solutions of turbulent flow. Journal of Computational Physics 113, 1–4.
Falgout, R.D., Jones, J.E., Yang, U.M., 2006. The design and implementation of HYPRE, a library of parallel high performance preconditioners. Numerical
Solution of Partial Differential Equations on Parallel Computers, vol. 51. Springer-Verlag, pp. 267–294.
Hong, W.-L., Walker, D.T., 2000. Reynolds-averaged equations for free-surface flows with application to high-Froude-number jet spreading. Journal of
Fluid Mechanics 417, 183–209.
Hunt, J., Wray, A., Moin, P., 1988. Eddies, stream, and convergence zones in turbulent flows. In: Proceedings of the CTR Summer Program. Center for
Turbulence Research, Stanford, CA, pp. 193–208.
Inoue, M., Bara, N., Himeno, Y., 1993. Experimental and numerical study of viscous flow field around an advancing vertical circular cylinder piercing a
free-surface. Journal of the Kansai Society of Naval Architects, Japan 220, 57–64.
Jiang, G.-S., Shu, C.-W., 1996. Efficient implementation of weighted ENO schemes. Journal of Computational Physics 126, 202–228.
Kandasamy, M., Xing, T., Stern, F., 2009. Unsteady free surface wave-induced separation: vortical structures and instabilities. Journal of Fluids and
Structures 25, 343–363.
Kawamura, T., Mayer, S., Garapon, A., Sørensen, L., 2002. Large eddy simulation of a flow past free surface piercing circular cylinder. ASME Journal of Fluids
Engineering 124, 91–101.
22 J. Suh et al. / Journal of Fluids and Structures 27 (2011) 1–22

Lang, A.W., Gharib, M., 2000. Experimental study of the wake behind a surface-piercing cylinder for a clean and contaminated free surface. Journal of Fluid
Mechanics 402, 109–136.
Launder, B.E., Rodi, W., 1983. The turbulent wall jet—measurements and modeling. Annual Review of Fluid Mechanics 15, 429–459.
Longo, J., Huang, H.P., Stern, F., 1998. Solid/free-surface juncture boundary layer and wake. Experiments in Fluids 25, 283–297.
Lourenco, L.M., Shih, C., 1993. Characteristics of the plane turbulent near wake of a circular cylinder. A particle image velocimetry study. Data taken from:
Kravchenko, A.G., Moin, P., 2000.Numerical studies of flow over a circular cylinder at ReD =3900. Physics of Fluids 12, 403.
Meneveau, C., Lund, C.S., Cabot, W.H., 1996. A Lagrangian dynamic subgrid-scale model of turbulence. Journal of Fluid Mechanics 319, 353–385.
Norberg, C., 1992. Pressure forces on a circular cylinder in cross flow. In: IUTAM Symposium on Bluff-Body Wakes, Dynamics and Instabilities, Göttingen,
Germany, pp. 275–278.
Ong, L., Wallace, J., 1996. The velocity field of the turbulent very near wake of a circular cylinder. Experiments in Fluids 20, 441–453.
Peng, D., Merriman, B., Osher, S., Zhao, H., Kang, M., 1999. A PDE-based fast local level set method. Journal of Computational Physics 155, 410–438.
Pope, S.B., 1978. The calculation of turbulent recirculating flows in general orthogonal coordinates. Journal of Computational Physics 26, 197–217.
Sarghini, F., Piomelli, U., Balaras, E., 1999. Scale-similar models for large-eddy simulations. Physics of Fluids 11, 1596–1607.
Sarpkaya, T., 1996. Vorticity, free surface, and surfactants. Annual Review of Fluid Mechanics 28, 83–128.
Shu, C.-W., Osher, S., 1988. Efficient implementation of essentially non-oscillatory shock-capturing schemes. Journal of Computational Physics 77,
439–471.
Sujudi, D., Haimes, R., 1975. Identification of swirling flow in 3-D vector fields. AIAA Paper 95-1715.
Sussman, M., Smereka, P., Osher, S., 1994. A level set approach for computing solutions to incompressible two-phase flow. Journal of Computational
Physics 114, 146–159.
Vlachos, P.P., Tellionis, D.P., 2008. The effect of free surface on the vortex shedding from inclined circular cylinders. ASME Journal of Fluids Engineering
130, 021103.1–021103.9.
Vogt, M., Larsson, L., 1999. Level set methods for predicting viscous free surface flows. In: Proceedings of the Seventh International Conference on
Numerical Ship Hydrodynamics, Nantes, France, pp. 2.4.1–2.4.19.
Walker, D.T., 1997. On the origin of the ‘surface current’ in turbulent free-surface flows. Journal of Fluid Mechanics 339, 275–285.
Williamson, C.H.K., 1996. Vortex dynamics in the cylinder wake. Annual Review of Fluid Mechanics 28, 477–539.
Xing, T., Stern, F., 2010. Factors of safety for Richardson extrapolation. ASME Journal of Fluids Engineering 132, 061403.1–061403.13.
Yang, J., Stern, F., 2009. Sharp interface immersed-boundary/level-set method for wave–body interactions. Journal of Computational Physics 228,
6590–6616.
Yu, G., Avital, E.J., Williams, J.J.R., 2008. Large eddy simulation of flow past free surface piercing circular cylinders. ASME Journal of Fluids Engineering 130,
101304.1–101304.9.
Yu, G., Avital, E.J., Williams, J.J.R., 2009. Computation of flow and near sound fields of a free surface piercing circular cylinders. Journal of Computational
Acoustics 17, 365–382.
Zdravkovich, M.M., 2003a. Flow Around Circular Cylinders: A Comprehensive Guide through Flow Phenomena, Experiments, Applications, Mathematical
Models, and Computer Simulations, Fundamentals, vol. 1. Oxford University Press, New York, NY.
Zdravkovich, M.M., 2003b. Flow Around Circular Cylinders: A Comprehensive Guide through Flow Phenomena, Experiments, Applications, Mathematical
Models, and Computer Simulations, Applications, vol. 2. Oxford University Press, New York, NY.

You might also like