Marine Structures 22 (2009) 142153

Contents lists available at ScienceDirect

Marine Structures
journal homepage: www.elsevier.com/locate/
marstruc

Numerical simulation of ow around a smooth circular

cylinder at very high Reynolds numbers
Muk Chen Ong a, *, Torbjrn Utnes b, Lars Erik Holmedal a, Dag Myrhaug a,
Bjrnar Pettersen a
a
b

Department of Marine Technology, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway
SINTEF IKT Applied Mathematics, NO-7465 Trondheim, Norway

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 14 March 2008
Received in revised form 4 July 2008
Accepted 8 September 2008

High Reynolds number ows (Re 1  106, 2  106 and 3.6  106,
based on the free stream velocity and cylinder diameter) covering
the supercritical to upper-transition ow regimes around a twodimensional (2D) smooth circular cylinder, have been investigated
numerically using 2D Unsteady Reynolds-Averaged NavierStokes
(URANS) equations with a standard high Reynolds number k  3
turbulence model. The objective of the present study is to evaluate
whether the model is applicable for engineering design within
these ow regimes. The results are compared with published
experimental data and numerical results. Although the k  3 model
is known to yield less accurate predictions of ows with strong
anisotropic turbulence, satisfactory results for engineering design
purposes are obtained for high Reynolds number ows around
a smooth circular cylinder in the supercritical and upper-transition
ow regimes, i.e. Re > 106. This is based on the comparison with
published experimental data and numerical results.
2008 Elsevier Ltd. All rights reserved.

Keywords:
Numerical models
Cylinder
Turbulent ow
High Reynolds number

1. Introduction
One of the classical problems in uid mechanics is the ow around a circular cylinder. This
represents an idealized bluff body ow which is of great interest for a wide range of engineering
applications, such as hydrodynamic loading on marine pipelines, risers, offshore platform support legs,

* Corresponding author.
E-mail address: muk.c.ong@ntnu.no (M.C. Ong).
0951-8339/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marstruc.2008.09.001

M.C. Ong et al. / Marine Structures 22 (2009) 142153

143

etc. Many of these engineering applications are often subject to ow conditions corresponding to very
high Reynolds number (Re UND/n) ows with typical values of O(106)O(107). This covers the
supercritical (3.5  105 < Re < 1.5  106) to transcritical (Re > 4  106) ow regimes. A detailed denition of the ow regimes is given by Sumer and Fredse [22]. Here UN is the free stream velocity; D is the
cylinder diameter; and n is the kinematic viscosity. These very high Reynolds number ow conditions
are hard and expensive to achieve in an experimental setup requiring appropriate experimental
facilities, minimizing human and instrument errors during measuring hydrodynamic quantities, etc.
Therefore an attractive alternative is to use Computational Fluid Dynamics (CFD) to obtain the essential
hydrodynamic quantities needed for engineering design. For example, Schulz and Meling [20] used
a multi-strip method to analyze the ow-structure interaction of long exible risers. This was achieved
by solving the two-dimensional (2D) Unsteady Reynolds-Averaged NavierStokes (URANS) equations
(using a SpalartAllmaras turbulence model) in conjunction with a nite element structural dynamic
response model. A number of individual 2D CFD simulations of cross sections along the riser were
combined with a full 3D structural analysis to predict overall vortex-induced vibration (VIV) loads and
displacement of the riser. They showed that a relatively modest number of sections could be used to
capture multi-modal VIV responses in long risers. Chaplin et al. [4] compared laboratory measurements of the VIV of a vertical model riser in a stepped current with predictions obtained with 11
different numerical models. It was shown that empirical-based models were better in predicting crossow displacements than the CFD-based models. This might be due to uncertainties in CFD turbulence
modeling and the modeling technique of the vortex-shedding interacting with dynamic response of
the structure (see Ref. [4]). Thus, more computational and experimental research on high Reynolds
number ows over a rigid cylinder section is necessary in order to gain better understanding on
dynamic responses of slender marine structures.
To date, not many numerical simulations have been performed to predict very high Reynolds
number ows (Re > 106) around a smooth circular cylinder due to the complexity of the ow. Direct
Numerical Simulation (DNS) of ows at such very high Reynolds numbers is not presently possible
because of the high demand on the computational resources. Among the few numerical results
reported in the open literature (for Re > 106) are those of Catalano et al. [3] and Singh and Mittal [21].
Catalano et al. [3] applied 3D Large Eddy Simulation (LES) with wall modeling as well as URANS using
the standard high Reynolds number k  3 model of Launder and Spalding [12] with wall functions, for
0.5  106 < Re < 4  106. Singh and Mittal [21] performed their studies for 100 < Re < 1 107 using a 2D
LES method. Catalano et al. [3] mainly focused on the cases of Re 0.5  106 and 1 106 when the drag
coefcient is recovering from the drag crisis (sudden loss of drag at Re 2  105). Their numerical
results captured the delayed boundary layer separation and reduced drag coefcients correctly right
after the drag crisis. They concluded that the LES results were considerably more accurate than the
URANS results at Re w 1 106. However, they also commented that the LES results became less
accurate compared with the experimental data at higher Reynolds numbers due to their insufcient
grid resolution. Singh and Mittals [21] main objective was to investigate a possible relationship
between the drag crisis and the instability of the separated shear layer. Their computations were able to
capture the sudden reduction in drag coefcient close to the critical Re. Even though their study
primarily focuses on the ow in the subcritical regime (300 < Re < 3  105), they also presented some
results for the ow beyond the supercritical ow regime, i.e. Re > 106.
The standard high Reynolds number k  3 model has been incorporated into most commercial CFD
codes. When used in conjunction with wall functions, it is generally taken as being computationally
less expensive than LES and DNS. Nevertheless, the model has been well documented for several
shortcomings, especially in the subcritical ow regime where the drag crisis occurs. Franke et al. [7]
and Tutar and Hold [23] evaluated numerically the detailed experiments of Cantwell and Coles [2] at
Re 1.4  105. Franke et al. [7] applied URANS with the standard high Reynolds number k  3 model of
Launder and Spalding [12]; Tutar and Hold [23] used both the standard high Reynolds number k  3
model and non-linear k  3 models. Their results were mainly obtained for the ow in the subcritical
ow regime at the start of the drag crisis. Both studies concluded that the k  3 models give an inaccurate prediction of ows with strong anisotropic turbulence. Catalano et al. [3] presented timeaveraged drag coefcients for Re 1 106, 2  106 and 4  106, Strouhal number for Re 1 106 and
mean pressure distribution for Re 1 106 using URANS with a standard high Reynolds number k  3

144

M.C. Ong et al. / Marine Structures 22 (2009) 142153

model. Most of their results were given for Re 1 106, but for Re 2  106 and 4  106 (in the uppertransition regime) they only presented the time-averaged drag coefcients without discussing the
applicability of the model in this upper-transition ow regime. Their results appear to yield satisfactory
agreements with experimental data.
The main objective of the present study is to evaluate whether the standard high Reynolds number
k  3 model is applicable for engineering applications in the supercritical and upper-transition ow
regimes (after the drag crisis). The ows around a 2D smooth circular cylinder at Re 1 106, 2  106
and 3.6  106 are investigated numerically, and these results are compared with available experimental
data and the numerical results reported by Catalano et al. [3] and Singh and Mittal [21].
2. Mathematical formulation
2.1. Flow model
The Reynolds-averaged equations for conservation of mass and momentum are given by

vui
0
vxi

(1)

 
0 0
vui
vu
1 vP
v2 u vui uj
v 2i 
uj i 
r vxi
vt
vxj
vxj
vxj

(2)

where i, j 1, 2. Here x1 and x2 denote the horizontal and vertical directions, respectively; u1 and u2 are
the corresponding mean velocity components; u0i u0j is the Reynolds stress component where u0i denotes
the uctuating part of the velocity; P is the dynamic pressure; and r is the density of the uid.
The Reynolds stress component, u0i u0j , is expressed in terms of a turbulent viscosity nT and the mean
ow gradients using the Boussinesq approximation,

u0i u0j

nT

vui vuj

vxj vxi

2
 kdij
3

(3)

where k is the turbulent kinetic energy and dij is the kronecker delta function.
A standard high Reynolds number k  3 turbulence model (see e.g. Refs. [12,17]) is used in the
present study. This model has been applied previously on vortex-shedding ow by Majumdar and Rodi
[14]. The k and 3 equations are given by:

vk
vk
v nT vk

uj
vt
vxj
vxj sk vxj

v3
v3
v nT v3

uj
vt
vxj
vxj s3 vxj

!
vui vuj vui

3
vxj vxi vxj

vT

C1 vT
k

!
32
vui vuj vui

 C2
vxj vxi vxj
k

(4)

(5)

where nT Cm k2 =3. The following standard model coefcients have been adopted: (C1 1.44,
C2 1.92, Cm 0.09, sk 1.0, s3 1.3).
2.2. Numerical solution procedure, computational domain and boundary conditions
The Reynolds-averaged equations for conservation of mass and momentum, in conjunction with
a standard high Reynolds number k  3 model, are solved by using a Galerkin nite element method. A
Segregated Implicit Projection (SIP) solution algorithm proposed by Utnes [25] is used for the time-step

M.C. Ong et al. / Marine Structures 22 (2009) 142153

145

solution. Here this numerical method is 1st order in time and 2nd order in spatial discretization. A
detailed description of the method is given by Utnes [25].
The geometric size of the rectangular computational domain and the boundary conditions imposed
for all simulations are shown in Fig. 1. The size of the whole computational domain is 27D  14D with
the cylinder in the centre of the vertical plane. The upper and lower boundaries are located at
a distance 7D from the centre of the cylinder; this ensures that these boundaries have no effect on the
ow around the cylinder. The ow inlet is located 7D upstream from the centre of the cylinder, and the
ow outlet is located 20D downstream from the centre of the cylinder. These distances are sufcient to
eliminate the far eld effects on the ow upstream and downstream of the cylinder. The boundary
conditions used for the numerical simulations are as follows:
(i) Uniform ow is specied at the inlet with u1 UN , u2 0. The free stream inlet turbulence values
for kinetic energy k 3=2Iu UN 2 and turbulent dissipation 3 Cm k3=2 =0:1L, proposed
by Tutar and Hold [23], have been imposed based on a turbulent intensity Iu u01 =UN of 0.8%
and a non-dimensional turbulent length scale (L/D) of 0.0045.
(ii) Along the outow boundary, u1, u2, k and 3 are specied as free boundary conditions in a nite
element context. This means that a traction-free velocitypressure boundary condition is applied
for u1, u2 and P (see Refs. [9,24] for details), while the ux is set equal to zero for k and 3. Along the
upper and lower boundaries, u1, k and 3 are free, while u2 is set equal to zero.
(iii) No-slip condition is applied on the cylinder surface with u1 u2 0.
(iv) Standard near-wall conditions are applied for k and 3 near the cylinder wall (see e.g. Ref. [17]) as

u2
;
k p
Cm

3 Cm3=4

k3=2
khp

(6)

where hp is the radial distance between the rst node and the wall, k 0.41 is the von Karman
constant, and u* is the wall friction velocity obtained from the logarithmic (log) law. It is well
known that the log law




ln 9d

(7)

is applicable for d  30, where d hpu*/n, u utan/u*, and utan tangential velocity to the wall.
In the viscous sublayer

u d

for

d  5

(8)

u1 = free, u2 = 0, k = free, = free

u1 = U
u2 = 0
k = 3 (IuU)2
2
Ck3/2
=
0.1L

7D

x2

u1 = 0, u2 = 0
x1

7D

20D

7D
u1= free, u2 = 0, k = free, = free

Fig. 1. The size of the computational domain and the imposed boundary conditions.

u1 = free
u2 = free
k = free
= free

146

M.C. Ong et al. / Marine Structures 22 (2009) 142153

Table 1
Details of the nite element mesh used and the results of grid convergence study.
Mesh

Nodes

Elements

Nc

Nt

CD

CLrms

St

M1
M2
M3
M4

27,874
38,766
48,706
67,034

27,512
38,340
47,600
66,480

304
360
400
480

76
90
100
120

0.4594
0.4586
0.4573
0.4573

0.0806
0.0767
0.0766
0.0765

0.2823
0.2899
0.3052
0.3052

Note: Nc number of nodes in the cylinder circumferential direction; and Nt number of nodes in cylinder wall normal
direction.

It is not possible to know d at the rst node normal to the cylinder wall a priori. Therefore the wall
law is given by the log law for d  30, by u d for d  5, and by a weighted average (linear
interpolation) of these two wall laws in the buffer zone, i.e. 5 < d < 30 (see e.g. Ref. [6]):

u 

kw 
ln 9d

1
 

1w
d

for 5 < d < 30

(9)

where w (d  5)/25. One should note that this blending method is sometimes sensitive to the grid
resolution; hence, trial and error of grid adjustment is required in order to avoid non-realistic results.
Stretching of the mesh is performed to achieve a ne resolution of the region close to the cylinder
surface. A grid convergence study has been performed for the ow at Re 3.6  106 with 4 sets of
meshes as shown in Table 1. Fig. 2 shows that the computations with the meshes M3 (48,706 nodes)

0.468
0.464

M1

M2

CD

0.46

M3

M4

0.456
0.452
0.448
2

x104

Number of Nodes
0.086

M1

CLrms

0.082

0.078

M2

M3

M4

0.074

0.07
2

Number of Nodes

x104

Fig. 2. Grid convergence study for CD and CLrms with respect to the number of nodes in the computational domain, see also Table 1.

M.C. Ong et al. / Marine Structures 22 (2009) 142153

147

and M4 (67,034 nodes) yield insignicant changes of results in terms of time-averaged drag coefcient
(CD) and root-mean-square value of uctuating lift coefcient (CLrms), i.e. the same result for CD up to
the fourth decimal and 0.13% difference in CLrms. Therefore, mesh M3 is considered to give sufcient
grid resolution. Fig. 3 shows the computational mesh M3. For mesh M3, the radial distance to the rst
node from the cylinder surface is 0.044% of the cylinder diameter. A non-dimensional time step (Dt) of
0.001D/UN is used; and the simulation is run for 250 non-dimensional time units (D/UN). d varies
from 0 to 87 depending on the local skin friction, as shown in Fig. 4. Furthermore, in order to ensure the
time step convergence, the ow has also been computed with a reduced Dt of 0.0005D/UN using mesh
M3. The deviation of CD and CLrms from the simulation with Dt 0.001D/UN, which are 0.04% for CD and
0.39% for CLrms. Hence it is concluded that Mesh M3 with Dt 0.001D/UN gives a sufcient numerical
accuracy. This is also used for the present computations at Re 1 106 and 2  106.

Fig. 3. The computational mesh M3 with 48,706 nodes and 47,600 elements.

148

M.C. Ong et al. / Marine Structures 22 (2009) 142153

100

80

60

40

20

0
0

30

60

90

120

150

180

Fig. 4. The variation of d around the cylinder. q is the peripheral angle of the cylinder measured clockwise from the stagnation
point.

3. Results and discussion

The computations have been performed at Re 1 106, 2  106 and 3.6  106, covering the supercritical to upper-transition ow regimes. The objective is to evaluate the applicability of using a standard high Reynolds number k  3 model for engineering computations of ow around a smooth circular
cylinder in the supercritical and the upper-transition ow regimes. Therefore, essential hydrodynamic
quantities, such as CD, CLrms and St, have been predicted and compared with published experimental
data [1,8,10,11,18,19,28] and numerical results [3,21]. Here St fD=UN is the Strouhal number, where f
is the vortex-shedding frequency. The comparisons are shown in Table 2 for Re 1 106 and 3.6  106.
Generally the present predictions are within the range of the experimental data and in reasonable
agreement with the published numerical results. For Re 1 106, the predicted CD is 0.5174; and it lies
between those predicted by 3D LES: CD 0.31 [3] and 2D LES: CD 0.591 [21]. Fig. 5 shows the values of
CD as a function of Reynolds number. Small discrepancies between the present results and the URANS
results reported by Catalano et al. [3] are seen. The present computed CD decreases slightly as the
Reynolds number increases, whereas the URANS results reported by Catalano et al. [3] exhibit a slight
increase of CD. This might be caused by different implementations of the wall function. The numerical
results (except those by Ref. [21]) exhibit a smaller variation with Reynolds number for Re > 106 than
the experimental data that generally shows an obvious increase of CD in the supercritical ow regime.
However, it appears that URANS with the standard high Reynolds number k  3 model is able to give
satisfactory predictions of commonly used hydrodynamic quantities in engineering design (i.e. CD,
CLrms and St) in the supercritical and the upper-transition ow regimes, i.e. Re > 106.
Relf [16] and Wieselsberger [27] inferred that the periodic vortex shedding should cease due to the
chaotic state of free shear layers when the ow becomes turbulent upstream of the separation. This
Table 2
Numerical and experimental results at Re 1 106 and 3.6  106.
CD

CLrms

St

1  106 (supercritical regime)

Present simulation
Catalano et al. [3] 3D LES
Catalano et al. [3] URANS
Singh and Mittal [21] 2D LES
Published experimental data

0.5174
0.31
0.41
0.591
0.210.63

0.0901

0.030.15

0.2823
0.35

0.180.50

3.6  106 (upper-transition regime)

Present simulation
Catalano et al. [3] URANS Re 4  106
Published experimental data

0.4573
0.46
0.360.75

0.0766

0.060.14

0.3052

0.170.29

Re

M.C. Ong et al. / Marine Structures 22 (2009) 142153

149

1.8
Present simulations
Catalano et al.(2003) 3-D LES
Catalano et al.(2003) URANS
Singh and Mittal(2005) 2-D LES
Wieselsberger(1923)
Roshko(1961)
Achenbach(1968)
James et al.(1980)
Schewe(1983)

1.6
1.4

CD

1.2
1
0.8
0.6
0.4
0.2
0
105

106

107

Re
Fig. 5. Time-averaged drag coefcient as a function of Reynolds number.

was indeed noted by Cincotta et al. [5], Loiseau and Szechenyi [13] and Schewe [19] in the supercritical
regime. Roshko [18], Cincotta et al. [5] and Schewe [19] measured the frequency of vortex-shedding
beyond the supercritical regime, i.e. Re 3.5  106 to 6  106. In this range of Reynolds number, they
discovered a narrow spectral peak of vortex-shedding frequency, showing that the periodic vortexshedding motion reappeared. Fig. 6 shows an example of contours of the non-dimensional vorticity
magnitude uD=UN for Re 3.6  106 at the non-dimensional time of 250 D/UN, predicted by the
present numerical model. It appears that the present model is able to produce the vortex-shedding
motion with a periodic vortex street qualitatively.
2  at Re 1 106
Fig. 7 shows the predicted mean pressure distribution Cp pc  pcN =0:5rUN
compared with the experimental data (Re 1.2  106) from Warschauer and Leene [26], the 3D LES
results and the URANS results by Catalano et al. [3], and the 2D LES results by Singh and Mittal [21].
Here pc is the static pressure at the peripheral angle of the cylinder, q, measured clockwise from the
stagnation point; and pcN is the static pressure of the ow at innity. The present URANS results
qualitatively capture the trend of the experimental data and the LES results reported by both Catalano
et al. [3] and Singh and Mittal [21], but the discrepancy with the experimental data is larger. The
discrepancy between the present URANS results and the URANS results by Catalano et al. [3] might be
due to the different implementations of the wall function. The present URANS results, the URANS
results by Catalano et al. [3] and the 2D LES results by Singh and Mittal [21] show a moderate overprediction of the negative Cp at the back half of the cylinder. It is well known that the ow in the
separation point region has strong pressure gradients, and that it is very difcult to model accurately.
Fig. 7 also shows that the 3D LES results by Catalano et al. [3] agree well with the measurements around

Fig. 6. Instantaneous non-dimensional vorticity contours of ow around a smooth circular cylinder for Re 3.6  106 at the nondimensional time of 250 D/UN. 40 contour levels from uD/UN 520 to uD/UN 520 are plotted.

150

M.C. Ong et al. / Marine Structures 22 (2009) 142153

1.5
Present simulation
Catalano et al.(2003) 3-D LES

Catalano et al.(2003) URANS

Singh and Mittal (2005) 2-D LES

0.5

Warschauer and Leene (1971) at Re=1.2x106

Cp

-0.5
-1
-1.5
-2
-2.5
-3

20

40

60

80

100

120

140

160

180

Fig. 7. Mean pressure distribution on the cylinder at Re 1  106.

the entire cylinder surface. This may show the limitation of using 2D URANS models for turbulent ow,
as effects from the spanwise secondary ow are not considered in the 2D simulation (see e.g. Ref. [15]).
It is interesting to observe that for the higher Reynolds number ow at Re 3.6  106, the predicted Cp
in Fig. 8 shows a better agreement with the experimental data from Achenbach [1] than for Re 1 106
shown in Fig. 7. The reason might be that at Re 3.6  106 the boundary layer ahead of the separation
point at one side of the cylinder is completely turbulent, while at Re 1 106 the boundary layers at
both sides of the cylinder are partly laminar and partly turbulent (see e.g. Ref. [22], Fig. 1.1) for

1.5
Present simulation

Achenbach(1968)

0.5
0

Cp

-0.5
-1
-1.5
-2
-2.5
-3

20

40

60

80

100

120

140

160

Fig. 8. Mean pressure distribution on the cylinder at Re 3.6  106.

180

M.C. Ong et al. / Marine Structures 22 (2009) 142153

151

a detailed description of the ow around a circular cylinder in steady ow. Hence the URANS with the
standard high Reynolds number k  3 model, for which a fully developed turbulent ow is assumed, is
expected to be more capable of predicting the ow at Re 3.6  106 (upper-transition ow regime)
than at Re 1 106 (supercritical ow regime). From Fig. 8 an approximate 30% under-prediction of
the negative Cp is observed at the back half of the cylinder due to the difculty in modeling the strong
pressure gradient accurately.
2 , where s is the tangential wall shear
The predictions of the skin friction coefcient (Cf s=0:5rUN
stress) are shown in Fig. 9 for Re 1 106, 2  106 and 3.6  106. In the same gure, the 3D LES results
by Catalano et al. [3], obtained at Re 0.5  106, 1 106 and 2  106, are plotted for comparison. The
present predictions show a similar shape of the Cf distribution around the cylinder surface as the 3D
LES results; the maximum magnitude of Cf decreases as the Reynolds number increases, which agrees
qualitatively with the 3D LES results by Catalano et al. [3]. It is observed that the present predictions
yield a lower maximum magnitude of Cf than the 3D LES results for the same Reynolds number. Further
discussion of this discrepancy of results is given in the next paragraph.
Fig. 10 shows the predicted Cf at Re 2  106 and 3.6  106 together with the 3D LES results at
Re 2  106 and the experimental data by Achenbach [1] at Re 3.6  106. Even though the 3D LES
results are not for the same Reynolds number as the experimental data, Fig. 9 shows that the predicted
Cf distributions do not change signicantly with Reynolds number neither for the present simulations
nor for the 3D LES simulations by Catalano et al. [3]. Therefore, the comparison shown in Fig. 10 seems
reasonable. Based on these observations, it appears that the Cf distribution obtained from the present
simulations agrees better with the experimental data than those by Catalano et al. [3]. However,
Catalano et al. [3] commented that their grid resolution was not sufcient in their work when they
compared their results with the experimental data in the high Reynolds number regimes. This might
explain that the present predictions show a better agreement with the experimental data. It is also
noted that both the present model and the 3D LES model by Catalano et al. [3] over-predict Cf on the
front half of the cylinder (i.e. before the separation) compared with the experimental data. The Cf
distribution on the back half of the cylinder is generally under-predicted by the present model
compared with the experimental data. Furthermore, it is also observed that the predicted time-averaged separation angle (qs) at Re 3.6  106 is 114 , which is in good agreement with the measured of
115 reported by Achenbach [1].
Overall, the URANS with the standard high Reynolds number k  3 model appears to give satisfactory predictions of the ow around a 2D smooth circular cylinder in the range Re 1 106 to
3.6  106. This is based on comparing the results with the published experimental data and numerical

0.016
Re = 3.6 x 106
Re = 2 x 106
Re = 1 x 106
Catalano et al.(2003),Re = 2 x 106
Catalano et al.(2003),Re = 1 x 106
Catalano et al.(2003),Re = 0.5 x 106

0.012
0.008

Cf

0.004
0
-0.004
-0.008
0.016
-0.016

60

120

180

240

300

360

Fig. 9. Comparison of skin friction distribution around the cylinder between the present computations and the 3D LES simulations
by Catalano et al. [3].

152

M.C. Ong et al. / Marine Structures 22 (2009) 142153

0.016
0.012

Re = 3.6 x 106
Re = 2 x 106
Catalano et al.(2003),Re = 2 x 106

0.008

Achenbach(1968),Re = 3.6 x 106

s=114

Cf

0.004
0
-0.004
-0.008
-0.012
-0.016

60

120

180

240

300

360

Fig. 10. Distribution of the skin friction coefcient on the cylinder in the upper-transition regime.

results. The results of the present study are encouraging for CFD-based engineering applications, e.g.
modeling of vortex-induced vibration responses of marine pipelines and risers together with a strip
theory approach to obtain 3D hydrodynamic loads, because the URANS with the standard high Reynolds number k  3 model requires less computational effort compared with LES and DNS for
Re > 1 106 (beyond the supercritical ow regime). Furthermore, its accessibility is high as it is
available in most of the commercial CFD software packages.
4. Conclusions
The ow around a 2D smooth circular cylinder has been computed for very high Reynolds numbers,
covering the supercritical to upper-transition ow regime, using the 2D URANS in conjunction with
a standard high Reynolds number k  3 model. Although it has been shown earlier that this model gives
less accurate predictions of ow with strong anisotropic turbulence, the present study shows that for
engineering design purposes it gives satisfactory qualitative agreements with the published experimental data and numerical results in the supercritical and upper-transition ow regimes, i.e. Re > 106.
However, more experimental data are required beyond the supercritical ow regime, especially
velocity and Reynolds stress prole measurements, in order to perform a more detailed validation
study of the model. In the meantime the present study should be reliable and useful as an engineering
assessment tool for design work.
Acknowledgements
This work has received nancial and computing support from Marine CFD- a project funded by The
Research Council of Norway. The rst author was the recipient of a research fellowship offered by The
Research Council of Norway.
References
[1] Achenbach E. Distribution of local pressure and skin friction around a circular cylinder in cross-ow up to Re 5  106. J
Fluid Mech 1968;34(4):62539.
[2] Cantwell B, Coles D. An experimental study of entrainment and transport in the turbulent near wake of a circular cylinder.
J Fluid Mech 1983;139:32174.
[3] Catalano P, Wang M, Iaccarino G, Moin P. Numerical simulation of the ow around a circular cylinder at high Reynolds
numbers. Int J Heat Fluid Flow 2003;24:4639.

M.C. Ong et al. / Marine Structures 22 (2009) 142153

153

[4] Chaplin JR, Bearman PW, Cheng Y, Fontaine E, Graham JMR, Herfjord K, et al. Blind predictions of laboratory measurements
of vortex-induced vibrations of a tension riser. J Fluids Structures 2005;21:2540.
[5] Cincotta TT, Jones GW, Walker WW. Experimental investigation of wind induced oscillation effects on cylinders in two
dimensional ow at high Reynolds numbers. National Aeronautics and Space Administration, NASA TMX 57779, paper 20.
120.35; 1966.
[6] Davies JT. Turbulence phenomena. London: Academic Press; 1972.
[7] Franke R, Rodi W, Schonung B. Analysis of experimental vortex shedding data with respect to turbulence modeling. In:
Proceedings of the 7th turbulent shear ow symposium; 1989. p. 24.4.124.4.5. Stanford, USA.
[8] Fung YC. Fluctuating lift and drag acting on a cylinder in a ow at supercritical Reynolds numbers. J Aeronautical Sci 1960;
27:80114.
[9] Gresho PM, Sani RL. Incompressible ow and the nite element method. West Sussex, England: John Wiley & Sons Ltd.;
1999.
[10] James WD, Paris SW, Malcolm GV. Study of viscous cross ow effects on circular cylinders at high Reynolds numbers. AIAA
J 1980;18:106672.
[11] Jones GW. Unsteady lift forces generated by vortex shedding about large stationary and oscillating cylinders at high
Reynolds numbers. In: ASME proceedings of the symposium on unsteady ow; 1968. Philadelphia, 68-FE-36.
[12] Launder BE, Spalding DB. Mathematical models of turbulence. London: Academic Press; 1972.
[13] Loiseau H, Szechenyi E. Experimental analysis of lift on a xed cylinder subjected to cross ow at high Reynolds numbers.
La Recherche Aerospatiale 1972:27991. No. 5.
[14] Majumdar S, Rodi W. Numerical calculation of turbulent ow past circular cylinder. In: Proceedings of the 7th turbulent
shear ow symposium; 1985. p. 3.1325. Stanford, USA.
[15] Mittal R, Balachandar S. Effect of three-dimensionality on the lift and drag of nominally two-dimensional cylinders. Phys
Fluids 1995;7:184165.
[16] Relf EF. On the sound emitted by wires of circular cross section when exposed to an air current. 6th series. Philosophical
Magazine 1921;42:1736.
[17] Rodi W. Turbulence models and their application in hydraulics. A state-of-the-art review. IAHR Monograph Series. 3rd ed.
Rotterdam, Netherlands: A.A. Balkema; 1993.
[18] Roshko A. Experiments of the ow past a circular cylinder at very high Reynolds number. J Fluid Mech 1961;10:34556.
[19] Schewe G. On the force uctuations acting on a circular cylinder in cross ow from subcritical up to transcritical Reynolds
numbers. J Fluid Mech 1983;133:26585.
[20] Schulz KW, Meling TS. Multi-strip numerical analysis for exible riser response. In: Proceedings of the 23rd international
conference on offshore mechanics and arctic engineering; 2004. OMAE2004-51186, Vancouver, Canada.
[21] Singh SP, Mittal S. Flow past a cylinder: shear layer instability and drag crisis. Int J Numer Meth Fluids 2005;47:7598.
[22] Sumer BM, Fredse J. Hydrodynamics around cylindrical structures: advanced series on ocean engineering vol. 12.
Singapore: World Scientic; 1997.
[23] Tutar M, Hold AE. Computational modeling of ow around a circular cylinder in sub-critical ow regime with various
turbulence models. Int J Numer Meth Fluids 2001;35:76384.
[24] Utnes T. Two-equation (k, 3) turbulence computations by the use of a nite element model. Int J Numer Meth Fluids 1988;
8:96575.
[25] Utnes T. A segregated implicit pressure projection method for incompressible ows. J Comp Physics 2008;227:2198211.
[26] Warschauer KA, Leene JA. Experiments on mean and uctuating pressures of circular cylinders at cross ow at very high
Reynolds numbers. In: Proceeding of the international conference on wind effects on buildings and structures; 1971. p.
30515. Tokyo, Japan.
[27] Wieselsberger C. Results of aerodynamic research at Gottingen (in German), II Lieferung; 1923.
[28] Zdravkovich MM. Flow around circular cylinders, vol. 1: fundamentals. New York: Oxford University Press; 1997.