Numerical study on the effect of shape modification to the flow around

circular cylinders

Kai Zhanga,b , Katsuchi Hiroshia,∗, Dai Zhoub,c , Hitoshi Yamadaa , Zhaolong Hanb,d
a Department of Civil Engineering, Graduate School of Urban Innovation, Yokohama National University, Yokohama,
2408501, Japan
b School of Naval Architecture, Ocean and Civil Engineering, Shanghai Jiao Tong University, Shanghai, 200240, China
c State Key Laboratory of Ocean Engineering, Shanghai Jiao Tong University, No. 800, Dongchuan Road, Shanghai,

200240, China
d Cullen College of Engineering, University of Houston, Houston, Texas, 77004, USA

Abstract
Shape modification has been one of the effective ways of manipulating the flow past bluff bodies.
Aiming at exploration of the flow control mechanisms of shape modified circular cylinders, a series of
numerical simulations are conducted at the Reynolds number of 5000. The modifications adopted in
the current paper could be divided into two categories, the 2-dimensional cross-sectional modifications
(polygonal and ridged) and 3-dimensional span-wise modifications (linear wavy, sinusoidal wavy and
O-ringed). By exploiting the numerical data obtained from the Large Eddy Simulations, several as-
pects, including the wake flow properties, the aerodynamic forces, the flow instabilities and span-wise
correlations are compared in detail. The two wavy cylinders are found to modify the flow wake signi-
ficantly, leading to considerable reduction in aerodynamic forces. The polygonal and ridged cylinders,
however, are of no such effect in aerodynamic force mitigation. The O-rings only have slight effect
on the flow wake and the forces. As for the flow instabilities, the modifications adopted in this paper
barely affects the Kármán vortex shedding frequency. The shear layer frequency, however, differs from
case to case. Particularly, in the cases of the two wavy cylinders, disparity exists in the shear layer
frequency at the node and saddle. This is explained by the boundary layer properties at separation.
While recognizing the insufficiency in the span-wise length of the cylinders, the span-wise correlations
of the aerodynamic forces are inspected. It is found that the lift forces of the 3-dimensional cylinders
are generally less correlated than that of the 2-dimensional ones.
Keywords: Circular Cylinders; Shape modification; Wake flow; Drag reduction; Shear layer
Instability; Correlation coefficient.

1. Introduction

Flow around circular cylindrical structures such as high rise buildings, bridge cables and overhead
transmission lines, is a common occurrence in various wind engineering applications. In most cases the
elusive aerodynamic forces acting upon the cylinders induce undesired effects such as structural noise
5 and fatigue. Especially in the case where the natural frequency meets the vortex shedding frequency,
large amplitude vibrations could occur on the structures.
Owing to the engineering significance, flow control has been an active and patenting field of re-
search for several decades. Previous investigations in this area have been fruitful in that not only a
variety of methods have been proposed, but some of the basic flow control mechanisms explained.
10 A comprehensive summary was presented in the annual review by Choi et al. [1]. Often, the control
approaches could be categorized into active methods, which requires energy input, such as rotary [2–4],
blowing/suction [5, 6], distributed forcing [7], etc., and passive control methods in which no external
energy is supplied but modifications on the structure are required, such as surface roughness [8, 9],

∗ Corresponding author. Email Address: katsuchi@ynu.ac.jp

Preprint submitted to Journal of Wind Engineering and Industrial Application 27th February 2016
 dimples [10], grooves [11–13], span-wise waviness [14, 15], helix [16, 17], wake splitter plates [18], etc. In
15 some researches, passive and active approaches are even combined [19, 20].
Despite the abundance of these flow control approaches, only a few of them are applicable to the
engineering applications. Particularly for the cable structures, active control approaches are excluded
since the implementation of the energy supplement apparatuses is difficult. Direct wake modifications
like splitter plates are also not suitable because this configuration could not attend to the flow coming
20 from all directions. On the contrary, circular cylinders with shape modifications are easy to fabricate
and install. Moreover, these methods are omni-directional so that they are effective irrespective of the
direction of the incoming wind. In the following paragraphs, we present a selective retrospect on the
documented flow control approaches or those which have the potential to be used for cable structures.
The aerodynamic means for controlling the flow wake and the aerodynamic forces by shape or
25 surface modification could be classified into two major categories, based on their phenomenological
mechanisms. The first is the boundary layer control. A laminar-to-turbulence transition takes place in
the boundary layer at the critical Reynolds number and empowers it with larger momentum against
the adverse pressure gradient. In this way the separation is delayed and the drag crisis is engendered.
The idea of making the boundary layer turbulent has bred several flow control methods such as surface
30 roughness, dimples, axial grooves and ridges, polygon, etc. The effect of surface roughness to the
aerodynamic forces has been investigated by a number of researchers. Achenbach [8] concluded from
the wind tunnel experiments that the drag crisis could be made earlier by the introduction of surface
roughness. However, this gain is accompanied with the unfavorable fact that the rougher the cylinder
surface is, the smaller the reduction in the drag coefficients is through critical regime, and the higher the
35 Cd is in the post-critical regime. Thus, the effectiveness of the surface-roughened cylinder is valid only
in limited range of Reynolds numbers. Inspired by the dimpled golf balls [21], Bearman and Harvey [10]
investigated the effect of dimples on the surface of a circular cylinder. It was found that with a similar
value of k/D(D is the diameter of the cylinder and k is the sand height in the roughened cylinder or
the depth of the dimples on the dimpled cylinder), the dimpled one induces a larger reduction in the
40 drag than the sand-roughened cylinder, although, at a slightly higher Re. Besides, the drag coefficient
in the trans-critical Re is substantially lower than that of the sand-roughened one. Kimura et al. [11]
explored the drag reduction effect of a single groove on a circular cylinder. It was found that the
flow near the wall is energized by this groove and thusly travels slightly farther against the adverse
pressure gradient before separation, leading to the favorable effect of drag mitigation. He further
45 concluded that the position of the groove is important and the effect of the dimples on the golf balls is
considered to be the same as that of the grooves. Apparently, since a single groove could not handle
the complications in practical applications, a regular distribution of grooves is instead required. Such
configurations have been considered by Yamagishi and Oki [12], who investigated the shape effect of
the grooves on a circular cylinder. Little disparity could be observed in the critical Reynolds number
50 between the triangle-shaped and arc-shaped cylinders, but the latter is better at drag mitigation in the
trans-critical regime. The same authors further studied the effect of the number of the grooves [13] and
came to the conclusion that increasing the number of the grooves would reduce the critical Reynolds
number. In a similar fashion, ridges on the cylinder are also capable of drag mitigation. To reduce the
wind load on the insulated wire, Matsumura et al. [22] studied the cylinders with mountainous surfaces
55 consisted of a number of ridges. It was revealed that, similar to the grooved ones, as the number of
the ridges increases, both the wind speed at which the drag coefficient begins to drop and the wind
speed at which the coefficient shows a minimum shift to the region of lower wind speed. Moreover,
the minimum value of the drag coefficient decreases.
So far, the surface/shape modifications mentioned above are mostly applied on the cross-sections
60 to achieve a retreated separation of the boundary layer. The second category of modifications, as
opposed to the first, is usually performed in the axial direction. One of the benefits in doing this is
to force the 3-dimensionlization of the flow separation. The research group led by Bearman has been
active in this field for several years. Bearman et al. [23] investigated the effect of sinusoidal waviness
installed at the stagnation face of a rectangular cylinder. Substantial drag reduction was achieved,
65 accompanied by significant attenuation of the Kármán vortex behind the bluff body. El-Gammal et
al. [24] then introduced this idea of span-wise sinusoidal perturbation method (SPPM) to control the
vortex-induced vibrations in a plate girder bridge. In the subsequent work, Owen et al. [25] studied

2
 a circular cross-sectional body with a sinuous axis and a circular cylinder with hemispherical bumps
attached. Again significant drag reduction was obtained for these two configurations. An elegant flow
70 visualization of the sinuous axis cylinder was also provided by the same group [26]. Another kind of
circular wavy cylinder, which is configured by keeping the axis straight and varying the local diameter,
has also been investigated intensively. Ahemd and Baysmuchmore [14] experimentally examined this
shape as early as the year of 1992. After him, the research team led by Lam from Hong Kong
Polytechnic University conducted a series of systematic investigations focusing on the flow physics and
75 the shape optimization of the sinusoidal wavy cylinders, with Reynolds number spanning from 100 to
50,000 [15, 27–30]. Some other researchers [31–33] also contributed to enhance the understanding of this
particular shape. Based on these literature, the most salient features of this shape are summarized
as follows. (i) The wavy surface leads to the formation of a 3 dimensional shear layer, which is
more stable than the 2 dimensional one; (ii) The vortex formation length is elongated by this more
80 stable shear layer, resulting in reduced drag force; (iii) The vortex shedding is attenuated, leading
to mitigated lift force. The attachment of O-rings onto the cylinder surface has also been reported
efficacious in reducing the aerodynamic forces. Different from the wavy cylinders, in which the gain
of force mitigation could be attained even at low Reynolds number flows, in both works by Lim and
Lee[34] and Nakamura and Igarashi[35], the effectiveness of the O-rings was only reported in relatively
85 higher Reynolds numbers of around 104 . Moreover, if the grooves were engraved in the longitudinal
direction, the desired effect of aerodynamic force reduction could also be achieved, as was revealed
by Lim and Lee[36]. One common observation between the O-ringed cylinders and the longitudinally
grooved cylinder is the elongated vortex formation regions, which was considered by Zdravkovich [37]
as the governing factor for high effectiveness of flow control. It appears that axial modifications of
90 these kinds are also able to alter the wake flow in a beneficial way, however, the profound cause of
which is less clear.
If a third category must be added to accommodate the industrially favorable morphology of the
cylinder with helical strakes [16, 37, 38], then we would call it a combination of the previous two, since
modifications are performed both in the cross-section and the span-wise direction. However, while
95 these strakes do serve well as countermeasures for vortex induced vibration, a by-product is the large
drag force that arises from them. Protrusions in smaller size, such as wires, fillets, strips, as well as the
engraved grooves, could also be arranged in a helical fashion, although with uneven performance in flow
control. Lee and Kim[17] studied the flow wake structure of a circular cylinder wrapped with 3 helical
wires and reported the suppression of Kármán vortex along with elongated vortex formation region.
100 From these observations it could be inferred that the aerodynamic forces were reduced. However,
Klessel and Georgakis [39] did not find satisfactory flow control effect in his experiment of a cylinder
wrapped with 2 helical fillets. Kiguchi et al. [40] investigated a new type of power transmission line
with helical grooves and proved its efficacy in drag reduction in the environment of rain. Moreover, a
cylinder with patterned indented surface, originally developed as countermeasure for the stability of
105 cable in both dry and rain environment[41], was also found to produce low drag [42].
As mentioned above, considerable efforts have been devoted to the manipulation of flow around a
circular cylindrical body. A comprehensive understanding of these control methods is of great signi-
ficance if they are to be used in practical applications. Despite the fact that the previous researchers
have conducted intensive researches to clarify the flow control mechanisms, a thorough understanding
110 of which, however, is far from complete. With the recent advancement of the computer hardware
as well as the sophisticated numerical schemes, it is tempting to use Computational Fluid Dynamics
(CFD) [43] as a tool to study the fluid physics associated with flow control approaches we have re-
viewed. The CFD simulation of flow around circular cylinders, either in the laminar regimes [44–46] or
turbulence [47, 48], has been proven successful in modelling the fluidic phenomenon and advantageous
115 in examining the flow details. Particularly, Large eddy simulation (LES), which resolves the large scale
flow and models the small, is capable of reproducing many of the flow-related physics both spatially
and temporally, at an affordable cost.
In the present work, we numerically investigate the cross-flow around six kinds of circular cylinders,
accommodating both cross-section and span-wise modifications, by the means of Large eddy simulation.
120 Owing to the limited computation resources, the simulated Reynolds number is kept at 5000. At this
Re the flow is already highly turbulent and is subjected to a number of flow instabilities. The results

3
 are compared in terms of the wake properties, the aerodynamic forces, the flow instabilities, the span-
wise correlation, etc. It is anticipated that this numerical work could throw some light upon the flow
control mechanisms of some of the shape modified cylinders as well as the flow details brought about
125 by these modifications.

2. Case setup

2.1. Geometries of the cylinders

(a) unmodified (b) polygonal (c) ridged

(d) linear wavy (e) sinusoidal wavy (f) O-ringed

Figure 1: Geometries of the cylinders

In order to investigate the effects of the flow control, we produce six types of shape modified
cylinders, which are shown in Fig. 1. One may refer to Fig. 2 for the information of coordinate
130 systems. The modifications are performed on the base circular cylinder (Fig. 1(a)) with the diameter
of D = 1m and height of H = 6.28D. The polygonal cylinder (Fig. 1(b)) is designed to be 20-faceted
and circumscribed by the base cylinder. The local radius of the ridged cylinder (Fig. 1(c)) is controlled
by r (θ) /D = 0.5 + 0.025 |sin (12 · πθ/180)|, which translates to 24 ridges around the circumference
and 0.025D in height. Both the polygonal and ridged cylinders have been seen services in the real
135 applications, especially the overhead transmission lines [22]. As a matter of fact, such shapes are more
favorable than the surface treatment such as dimples or indents since the latter becomes less effective
in the case of transmission lines with smaller diameter. As for the two wavy cylinders (Fig. 1(d), 1(e)),
we use the same nomenclatures as Lam and Lin [30], i.e., node is the place where the local diameter
maximizes and saddle is the place where it minimizes. The diameter varies linearly or sinusoidally
140 with the axial coordinate, as their names suggest. Both wavy cylinders have an average diameter of
D, wavelength of λ/D = 2.093 and amplitude of a/D = 0.1. This configuration is similar to the
WY-A4 model in Lam and Lin [30]. Although the wavy cylinders are seldom seen in civil engineering
applications, a similar undulated elliptic geometry, which is inspired by the whisker of the harbor
seals, have been used as a low-power, low-cost flow sensor by Beem et al. [49]. It is hoped that by
145 getting to know more about these wavy geometries, the aerodynamically beneficial shapes could be
innovated to be used in more ocassions. Three O-rings, which could be regarded as the joints of the
segmented pipes, are installed on the base cylinder with an interval of p/D = 1.57, measured from one
ring center to another (Fig. 1(f)). The width of the ring is ∆s/D = 0.2, and the height is h/D = 0.1.
To facilitate our discussion, we still refer the ringed part to node and the plane part to saddle. The
150 first 3 cylinders are termed 2 dimensional cylinders since they could be created by stretching the

4
 modified cross-sections straightly into the third dimension. The remaining are termed 3 dimensional
ones because the modifications are done in the axial direction. By studying the aerodynamics of
the modified cylinders of different types in one single work, we can not only appreciate the different
flow control mechanisms they exhibited, but also the recgonize the respective pros and cons of each
155 method. These modified cylinders differ slightly in their equivalent diameters, however, the same
reference diameter of D is used in the calculation of Reynolds number and force coefficients for all the
six cases.

2.2. Governing equations and turbulence model

The current work is done in the frame of Large Eddy Simulation. The LES equations are derived
160 from the classical time-dependent filtered incompressible Navier-Stokes equation:
∂ui
=0 (1)
∂xi
   
∂ui ∂ui uj 1 ∂p ∂ ∂ui ∂uj
+ =− + ν + + τij (2)
∂t ∂xj ρ ∂xi ∂xj ∂xj ∂xi
τij = ui uj − ui uj (3)
where ν is the kinematic viscosity, u and p are the filtered velocity and pressure respectively. Eq. 3
is termed the sub-grid scale(SGS) stress which requires modeling. The SGS stress τij is expressed
165 according to the Boussinesq approximation with the introduction of a turbulent eddy viscosity νt :
2
τij − kt δij = −2νt S ij (4)
3
 
1 ∂ui ∂uj
S ij = + (5)
2 ∂xj ∂xi
in which kt = τkk /2 is the SGS turbulent kinetic energy and S ij is the rate of strain tensor computed
directly from the resolved scales.
Methods differ in the modeling of the turbulent eddy viscosity νt [50]. Lysenko et al. [51] compared
170 the performance of the Smagorinsky SGS model with the dynamic k-equation model in the open-source
CFD software OpenFOAM. Although the results of both models had experimental data backed on,
we found that the results obtained by the latter more convincing. In the current work the dynamic k-
equation model is adopted. Instead of assuming the local equilibrium of the production and dissipation
of the sub-grid turbulent energy, as is done in the Smagorinsky model [52], in dynamic k-equation model
175 [50] an exact balance equation for kt is derived:
 
∂kt ∂ ∂ ∂kt
+ (uj kt ) = P + (ν + νt ) − (6)
∂t ∂xj ∂xj ∂xj

P = 2νt S ij S ij ,  = C kt1.5 ∆−1 (7)

where the SGS viscosity is given by νt = Ck ∆kt0.5 . The model coefficients C and Ck are dynamically
computed as part of the solution based on the Germano identity [53] with test filter ∆ b = 2∆ by the
least square minimization procedure proposed by Lilly [54].

180 2.3. Numerical aspects

The finite-volume based open-source code OpenFOAM [55] is used in parallel to carry out the
simulations in the present work. Originally developed as high-end C++ class libraries for a broad range
of fluid dynamic applications, OpenFOAM has quickly gained popularity among both the academia and
the industry. The canonical problem of flow around a circular cylinder has also been seen in several
185 publications by using the OpenFOAM toolbox [51, 56–58]. In the present research, the transient
solver for incompressible flow, pimpleFoam, is selected to solve this pressure-velocity coupled system.
The distinctive feature of the pimple algorithm is that it combines the PISO (Pressure Implicit with

5
 Splitting of Operator) and SIMPLE (Semi-Implicit Method for Pressure Linked Equations) methods
to improve the numerical stability when a larger time-step is employed. Exploiting this merit, the
190 restriction of the time-step imposed by the maximum Courant number (maxCo) [59] could be relaxed,
leading to a reduction in the total computing time.
A schematic of the coordinate system and the planar view of mesh is presented in Fig. 2. The
circular cylinders are placed in the center of a circular computation domain, which has a radial exten-
sion of 20D. The curvilinear O-type mesh is used with 160 × 160 grids in the cross section (the exact
195 number of grids on the circumference is slightly adjusted in the polygonal and ridged cylinder cases
to evenly mesh each of the sides or ridges). The radial grid points are clustered in the vicinity of the
cylinder with an expansion factor of 1.038 in avoidance of abrupt change of the grid size, and the first
grid is placed at around 1.9 × 10−3 D away from the cylinder wall. This will be proven sufficient to
well resolve the boundary layer, since the maximum y + value in all the six cases are found to be below
200 unity. 96 grids are prescribed in the span-wise direction, which extends as long as 6.28D. The total
number of control volumes thusly mounts up to around 2.45 × 106 . The time-step is firstly decided by
the condition maxCo< 2 (where maxCo stands for the maximum courant number) in a pre-run and
then kept fixed during the time integration. In general, statistics are compiled over periods of at least
200 non-dimensional seconds, or about 40 vortex shedding cycles to obtained the mean flow variables.
205 The current mesh resolution is chosen after the mesh dependency test, as will be presented in the next
subsection. The inlet boundary is assigned with a constant velocity u∞ = 1m/s in the x direction.
Periodicity is enforced in the span-wise direction of the cylinder. The zero-gradient condition is used
for the outlet. The cylinder wall is set to be non-slip.

(a) Coordinate system (b) Schematic of mesh in the xy domain

Figure 2: Schematic of the coordinate system and planar view of mesh.

The time integration is performed using the second order implicit backward scheme. Gauss lim-
210 itedLinear, one of the TVD (Total Variation Diminishing) schemes, is employed for the convection
differencing. The diffusion and pressure terms are discretized by the Gauss linear scheme.
Up to now, both the computational domain and the governing equations are discretized, a set of
algebraic equations are thusly formed, awaiting to be solved by the matrix solvers. The generalized
geometric-algebraic multi-grid (GAMG) method, together with the Gauss-Seidel smoother is used to
215 solve the pressure equations. The PBiCG method, abbreviated for the preconditioned bi-conjugate
gradient solver for asymmetric matrices is used to solve the velocity and k equations.

2.4. Mesh dependency test

CFD simulations are carried out on computation domains that are properly meshed. The mesh
resolution should be delicately chosen so that it could yield satisfactory result without posing excessive
220 burden on the computational resources. Restricted by the computer resource, in the current work the
mesh dependency test is done for the unmodified case only. The shape modified cylinders will employ
a similar mesh density. The results of the mesh dependency test are shown as follows.

6
 Table 1: Mesh dependency test

NO. Domain Mesh maxCo Cd Cl r.m.s.

A1 40D × 6.28D 160 × 160 × 96 2 1.005 0.149
A2 40D × 6.28D 180 × 180 × 96 2 0.985 0.121
B1 40D × 3.14D 160 × 160 × 48 1 0.996 0.143
B2 40D × 3.14D 160 × 160 × 48 2 1.021 0.153
B3 40D × 3.14D 200 × 200 × 48 2 0.977 0.123

Table 1 presents the five cases that are used in the mesh dependency test. Case A1 and A2 have
a span-wise length of 6.28D, while the cases titled B are performed with half of the span-wise length
225 of case A to save computation time. It was revealed by Kravechenko and Moin [60] and Breuer [61]
that doubling the cylinder span from πD to 2πD while maintaining the same axial resolution would
yield similar results in terms of aerodynamic forces and mean wake properties. This conclusion grants
us the rightful reason to make direct comparisons between Case As and Bs. With the same resolution
in the span-wise direction, case A1, A2 and B3 are distinguished by their increasing grid points in
230 the cross-section. The effect of time-step is compared by B1 and B2, in which the time-steps in each
of the cases are decided by setting the maximum Courant number to be 1 and 2. The statistics of
each case are gathered for 200 non-dimensional seconds after a 100 non-dimensional seconds pre-run.
Three papers taken as reference in this section are the experimental work by Norberg [62] and two
LES works by Kim [63] and Doolan [57]. Note that the last paper is dedicated to a slightly larger
235 Reynolds number of 5600.

0.8 0.5
A1
A2
0.6 B1
0.4 B2
B3
0.4 Noberg Exp
Kim LES
0.3 Doolan LES
0.2
umean urms
0 A1 0.2
A2
-0.2 B1
B2
B3 0.1
-0.4 Noberg Exp
Kim LES
Doolan LES
-0.6 0
0 1 2 3 4 5 6 0 1 2 3 4 5 6
x/D x/D
(a) Mean stream-wise velocity (b) Fluctuating stream-wise velocity

Figure 3: Mean stream-wise velocity and r.m.s. fluctuating velocity along wake center line

While the drag force coefficients are found to decrease as the cross-sectional mesh is refined, the
difference between case A1 and B3 is only 3%. The effect of span-wise length is again proved of minor
significance, based on case A1 and B2. From case B1 and B2 it is concluded that the effect of time-step
is also trivial. The resulted time-step for case A1 is ∆t = 0.013. It is noted that a time-step of around
240 ∆t = 0.01 was used in several cross cylinder flow studies at similar Reynolds numbers (e.g., ∆t = 0.01
for Lee et al. [58]’s DES at Re = 5000, ∆t = 0.0172 for Doolan [57]’s LES at Re = 5600).
The mean stream-wise velocity profiles are shown in Fig. 3 to Fig. 5. It could be observed from
these figures that case A1, although possessing the relatively coarse mesh resolution, produces fare
agreement with the existing literature. The results from the finer meshed cases deviate only slightly
245 from that of A1, and this deviation does not improve the results to better agreement with the literature.
Less satisfying result is obtained for the r.m.s. fluctuating velocity. Although the general trend has
been reproduced, its value is smaller compared to that in the literature. The relatively lower value
of urms may have its origin in the turbulence model we used. Lysenko et al. [51] employed high
spatial resolution (300×300 in the cross-section plane) for his LES of flow around circular cylinder at
250 Re = 3900 with the dynamic k-equation turbulence model, the value of urms was also under-predicted.
It would seem that the prediction of urms is more challenging than umean by this turbulence model.

7
 1.6 0.6
A1
A2
B1
1.2 0.5 B2
B3
Noberg Exp
0.4 Kim LES
0.8
umean urms 0.3
0.4 A1
A2
B1
0.2
B2
0 B3
0.1
Noberg Exp
Kim LES
Doolan LES
-0.4 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
y/D y/D
(a) Mean stream-wise velocity (b) Fluctuating stream-wise velocity

Figure 4: Mean stream-wise velocity and r.m.s. fluctuating velocity at x/D = 1

1.2 0.5
A1
A2
1 B1
0.4 B2
B3
0.8 Noberg Exp
Kim LES
0.6 0.3
umean urms
0.4 A1 0.2
A2
0.2 B1
B2
B3 0.1
0 Noberg Exp
Kim LES
Doolan LES
-0.2 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
y/D y/D
(a) Mean stream-wise velocity (b) Fluctuating stream-wise velocity

Figure 5: Mean stream-wise velocity and r.m.s. fluctuating velocity at x/D = 2

Based on the above statements, it could be inferred that the case A1 is not perfectly converged;
however, the discrepancy of A1 compared with other refined cases is small. Particularly, case A1
qualifies an accurate prediction of flow variables such as mean velocity, aerodynamic forces, etc. The
255 mesh dependency test for the other cylinders is omitted here because of the restricted computational
resource. Since all the cases are performed under the same Reynolds number of 5000, and we are
particularly interested in the comparison between the cases, it is decided that the mesh resolution
used in case A1 be applied to each of the modified cylinders.

3. Results and discussion

260 3.1. Wake properties

The mean streamlines of the unmodified, polygonal and ridged cylinders are presented in Fig. 6.
The snapshots are taken at the mid-span of the cylinders and rendered with the mean stream-wise
velocity. Figures below are the zoom views of the boundary layer separation locations (red box region),
in which the red and blue lines are distinguished by the sign of the mean stream-wise velocity. The
265 wake of each cylinder is mainly composed of a pair of symmetric vortex loops, the extent of which is
usually referred to as the recirculation length. An observant reader might also notice that in Fig. 6(c),
small vortices are generated in between the ridges; this is more clearly shown in the zoom view. These
small vortical structures are engendered by the local separations at the top of these ridges. It is
expected that these ridges would generate a turbulent boundary layer that is more resistant to the
270 adverse pressure gradient, thus delaying the separation and cause an earlier drag crisis. However,
as will be shown later, at current Reynolds number, this phenomenon does not manifest. It could

8
 (a) unmodified (b) polygonal (c) ridged

Figure 6: Averaged streamline of the unmodified, polygonal and ridged cylinders

be concluded that without the exception of the ridged cylinder, the boundary layers in these three
cylinders all separate at around θ ≈ 90◦ (Hereinafter, θ is measured from the stagnation point of the
cylinder).
275 Owing to the three-dimensionality of the geometries, it is difficult to present the wake shapes
for the wavy and ringed cylinders with streamline plots. Alternatively, Fig. 7 presents the contours
of umeanx = 0, which delineate the reverse flow regions. It should be noted this contour cuts the
each of the vortex loops into halve, thus the enclosed region is smaller than the wake presented in
Fig. 6. Nevertheless, we use them to qualitatively sketch the wake size. The concentric circles in Fig. 7
280 represent the cross-sections of the node and saddle planes. It is clear from these plots that the wakes
feature distinctively at the node and the saddle sections for all these three cylinders. The wake at
the node section soon converges to the confluence point; however, at the saddle section it undergoes
a slight expansion in width before it shrinks and converges. Besides, in the case of the two wavy
cylinders, the wake lengths at the saddle are noticeably longer than that at the node. This finding is
285 in agreement with Lam et al. [28]. The disparities in the wake shapes originate from the difference
in the boundary layer separation angle. At present we roughly regard the intersection of the contour
of umeanx = 0 and the cylinder boundary as the point of separation. Using θ = 90◦ as the reference
point the separation angle at the node and saddle could be clearly distinguished. Intuitively, at the
saddle the flow separates before θ = 90◦ , the separated free shear layer possesses an outward trend,
290 resulting in a wider, longer wake; at the node the separation point lies beyond θ = 90◦ , the two free
shear layers from both side of the cylinder moves towards each other, a narrower, shorter wake is thusly
engendered. A more precise approach to locate the separation point, which gives the same conclusion,
will be given in the latter part of this section.
More clearly, the mean stream-wise velocity on the wake center lines are depicted in Fig. 8. The
295 unmodified cylinder is presented in each of these figures for comparison. The results of a sinusoidal
wavy cylinder with λ/D = 2.273 and a/D = 0.091 at Re = 6000 by Lam et al. [28] are also cited in
Fig. 8(c). The differences in the shape parameters, Reynolds number as well as the experimental setup
might contribute to the deviations that are observed in the velocity profiles between our results and
the cited ones. However, as will be shown later, qualitative agreement is achieved. For the polygonal
300 and ridged cylinders, the velocity distribution deviate from the unmodified cylinder only slightly. As
for the 3-dimensional modifications, some interesting features could be noticed. Firstly, as mentioned
in the previous section, the recirculation length (defined as the point at which umeanx = 0) of the wavy
cylinders at the saddle plane is longer than the node plane, and both are considerably longer than the
unmodified cylinder. Another aspect worthy of noting is that the back-flow velocity, indicated by the
305 absolute value of minimum of each velocity profile, at the node is significantly slower than that at the

9
 0.6 0.6
0.4 0.4
0.2 0.2
y 0 y 0
-0.2 -0.2
-0.4 node
-0.4 node
saddle saddle
-0.6 -0.6
-0.5 0 0.5 1 1.5 2 2.5 -0.5 0 0.5 1 1.5 2 2.5
x x

(a) linear wavy cylinder (b) sinusoidal wavy cylinder

0.6
0.4
0.2
y 0
-0.2
-0.4 node
saddle
-0.6
-0.5 0 0.5 1 1.5 2 2.5
x

(c) O-ringed cylinder

Figure 7: The contour of umeanx = 0 of the 3 dimensional cylinders

saddle. These findings is also manifested in the cited data of Lam et al. [28].

0.8 0.8

0.6 0.6
0.4
0.4
0.2
umean

umean

0.2
0
0
-0.2
unmodified unmodified
-0.2 polygonal -0.4 saddle
ridged node
-0.4 -0.6
0 1 2 3 4 5 6 0 1 2 3 4 5 6
x/D x/D
(a) 2 dimensional cylinders (b) linear wavy cylinder
0.8 0.8
0.6 0.6
0.4
0.4
0.2
umean

umean

0.2
0
unmodified 0
-0.2 saddle
node unmodified
-0.4 saddle (Lam2004) -0.2 saddle
node (Lam2004) node
-0.6 -0.4
0 1 2 3 4 5 6 0 1 2 3 4 5 6
x/D x/D
(c) sinusoidal wavy cylinder (d) O-ringed cylinder

Figure 8: Profiles of mean stream-wise velocity in the wake center line.

For the 3 dimensional geometries, it is also meaningful to check the flow profiles in the cross-

10
 sectional planes. Fig. 9 presents the evolution of the mean stream-wise velocity in the yz planes
at several downstream locations for the two wavy cylinders and the O-ringed cylinder. It could be
310 observed that velocity profiles undergo considerable changes as the position shifts to the downstream
locations. In the case of the two wavy cylinders, the mean flow structures in the immediate wakes are
characterized by a pair of thin and undulated shear layers. The undualation is periodic as a result
of the wavy geometry. Besides, it could also be observed that the relative width of the wake in the
span-wise direction, indicated by the width betweeen the two shear layers, varies with the downstream
315 locations. At X/D = 0.6, the wake width at the node is slightly wider than that at the saddle.
However, when it comes to X/D = 1.5, and more clearly at X/D = 2.0, the opposite is true. This
could be easily explained with the aid of Fig. 7, as the shear layers expand starting from a relatively
narrower width at the saddle and contract starting from wider width at the node. The shear layers
grow thicker as they travel further in the streamwise direction, as is shown at X/D = 3.0, and the
320 undulations become diminishingly small. As for the O-ringed cylinder, the shear layers are strongly
distorted in the vicinity of the O-rings, and the wake width at the node is smaller than the other
locations. Unlike the wavy cylinders, the effect of the O-rings is lessly felt at X/D = 2.0, indicating
the smaller wake length of the cylinder.

(a) linear wavy cylinder

(b) sinusoidal wavy cylinder

(c) O-ringed cylinder

Figure 9: Mean streamwise velocity in the yz planes for the 3 dimensional cylinders

By plotting the boundary layer profiles the separation point could be identified. Depicted in Fig. 10

11
325 is the boundary layer profiles at different angular locations of the unmodified cylinder. ubl is the mean
velocity tangential to the cylinder wall and l is the distance from the surface. The boundary layer
is observed to accelerate before θ ≈ 70◦ . Between θ ≈ 70◦ ∼ 90◦ the velocity is retarded and the
boundary layer thusly thickens swiftly. At θ = 90◦ the reverse flow has already taken place, indicating
the occurrence of separation. The reverse flow is more clearly shown at θ = 100◦ .
330 Between θ = 70◦ and 90◦ the boundary layer profile shifts from an L shape to an S shape, which is
an evidence of the existence of the inflection point (∂ 2 u/∂l2 = 0). It is thus decided that the boundary
layer separation point ((∂u/∂l)y=0 = 0) lies between this range [64]. A detailed inspection (not shown
here for brevity) reveals that a more precise value of separation angle for the unmodified cylinder is
θsep = 88◦ . This boundary layer profiling method is applied to find the separation points for the other
335 cases.The separation angle θsep together with the recirculation length Lr /D are summarized in Table
2. Significant difference could be found in θ between the nodes and saddles of the wavy cylinders. The
separation angle at the ring node is left blank since the current method does not succeed in finding a
precise value for it. As a matter of fact, it has been shown by Nakamura and Igarashi [35] that the

° 90° 100 °
° 80
70
°
60
0.1
° 30
50 40
0.08 50
60
70
°
40

80
0.06 90
100
l
30 °

0.04

0.02

0
-0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
ubl

Figure 10: Boundary layer profiles of the unmodified cylinder

Table 2: Recirculation lengths and separation angles of the six cases

line wavy sine wavy ringed

unmodified polygonal ridged
node saddle node saddle node saddle
Lr /D 1.9 2.0 2.15 2.15 2.34 2.38 2.64 1.83 1.90
θsep 88◦ 85◦ 87◦ 99◦ 79◦ 92◦ 82◦ − 87◦

3.2. Aerodynamic forces

340 Flow induced forces on a structure consist of two major components: pressure force normal to
the surface of the structure and shear force tangential to the surface. In our calculation the latter is
neglected since at Re = 5000, the viscous shear force is much smaller compared to the pressure force.
The integration of the normal pressure stresses over the structure results in the flow-induced aero-
dynamic forces, which could be decomposed into the drag in the along-flow direction and the lift in

12
345 the cross-flow direction. The sectional drag and lift forces of a circular cylinder could be calculated
by:
Z 2π n
X
Fd = p · nx · r · dβ = p · nx · r · ∆β (8)
0 0
Z 2π n
X
Fl = p · ny · r · dβ = p · ny · r · ∆β (9)
0 0

in which p is the pressure on the cylinder surface, nx and ny are the wall normal vectors in the x
and y directions, r is the local radius of the cylinder, n is the number of grids along the cylinder
350 circumference and ∆β = 2π/n.
It will be proven useful to refer p, Fd and Fl to their non-dimensional forms Cp , Cd and Cl , as are
defined by Eq. 10 to Eq. 12:
p − p0
Cp = (10)
ρU02 /2
n
Fd 1X
Cd = 2 = Cp · nx · ∆β (11)
ρU0 D/2 2 0
n
Fl 1X
Cl = 2 = Cp · ny · ∆β (12)
ρU0 D/2 2 0
355 where p0 stands for the reference pressure at infinite upstream, ρ is the density of the fluid and U0 is
the free stream velocity of the flow.
The span-wise averaged aerodynamic force coefficients of the six cases are summarized in Table 3.
Compared to the unmodified cylinder, the polygonal, ridged and O-ringed cylinders fail in the attempt
to reduce the aerodynamic forces. However, both the drag and lift coefficients are appreciably reduced
360 in the two wavy cylinders. Especially in the case of the sinusoidal wavy cylinder, 10% reduction in Cd
and 70% reduction in the Cl rms are achieved. The cause for this favorable effect will be explained in
the paragraphs that follow.

Table 3: Drag and lift force coefficients of the six cases

unmodified polygonal ridged line wavy sine wavy ringed

Cd 1.005 0.99 1.035 0.938 0.890 1.041
Cl r.m.s. 0.147 0.125 0.120 0.057 0.039 0.146

It is also meaningful to inspect the spatial and temporal evolution of the aerodynamic forces, as
shown in Fig. 11 ∼ Fig. 16. Each of these figures is composed of two parts. The color maps in the
365 upper part are the spatial-temporal distributions of Cd and Cl and the figures below are their axially
averaged values. Additionally, the blue lines shown in the Cd figures are the temporally averaged values
calculated at the local instants. The flatness of this line is an indication that the time integration
length adopted in the current simulations is enough to converge the drag forces, as well as other flow
quantities, to their mean values.
370 For the 2-dimensional cylinders, i.e., unmodified, polygonal and ridged cylinders, little regularity
could be observed in the drag force coefficients. While an average value of about 1 dominates the
figures, local maximum and minimum spots scatter throughout the figures. The Cl plots are mainly
consisted of repeated filaments of red and blue, with each taking a band width of around 2.5 non-
dimensional seconds, which represents the periodic lift force acting upon the cylinder. Similar to the
375 drag force coefficients, the Cl color maps also show certain degree of span-wise variation, which is an
evidence for the 3-dimensionality of the wake flow.
Situation differs for the axially modified cylinders. Significant span-wise variation could be observed
in the Cd plots of the two wavy cylinders: Cd reaches its maximum at the node and falls to the minimum
at the saddle. This finding is in agreement with that of Ahmed et al. [14]. Recall the fact that in
380 the wavy cylinders the boundary layer separation is delayed at the node. The higher value of Cd at

13
 (a) Cd (b) Cl

Figure 11: Aerodynamic force coefficients of the unmodified cylinder.

(a) Cd (b) Cl

Figure 12: Aerodynamic force coefficients of the polygonal cylinder.

(a) Cd (b) Cl

Figure 13: Aerodynamic force coefficients of the ridged cylinder.

(a) Cd (b) Cl

Figure 14: Aerodynamic force coefficients of the linear wavy cylinder

the node seems to be in contradictory with the previous knowledge that delaying separation is one
way to mitigate the drag force, as is the case of the roughened cylinder, dimpled cylinder, etc. This
paradox serves as an evidence for the different flow control mechanisms exhibited by the two kinds of
modifications. The Cl of these two wavy cylinders, however, does not show regular span-wise variation

14
 (a) Cd (b) Cl

Figure 15: Aerodynamic force coefficients of the sinusoidal wavy cylinder

(a) Cd (b) Cl

Figure 16: Aerodynamic force coefficients of the O-ringed cylinder

385 as Cd does. As for the cylinder with O-rings, 3 sets of double lines, representing locally low Cd and
Cl values, cut the plots at the intersections of the rings and the cylinder surface. This is owing to
the sudden protrusion of the rings. For any location on the surface the identity n2x + n2y + n2z = 1
holds, where nx , ny and nz stands for the three components of the surface normal vector. At the
intersections of the rings and the cylinder, the abrupt increase in nz significantly reduces the other
390 two components, leading to lower Cd and Cl , according to Eq. 11 and Eq. 12. It should be noted that
the existence of non-zero nz plays only minor role in aerodynamic force mitigation of the two wavy
cylinders. Taking the linear wavy cylinder, whose nz = 0.19, for example, the expected Cd or Cl would
0.5
scale as (1 − nz ) = 0.98 (2% reduction), while the actual reduction of 6.2% is achieved, as shown in
Table 3.
395 Since the drag and lift force coefficients are calculated from Cp , it is instructive to have a look
at the pressure distribution along the cylinder circumference. Depicted in Fig. 17 are the mean Cp
of different cylinders and at different characteristic locations. For all the cases, the highest pressure
coefficient locates at θ = 0◦ . It then decreases sharply, giving rise to the positive pressure gradient
which accelerates the boundary layer flow. The pressure reaches its minimum at around θ = 70◦ ∼ 80◦
400 and then it goes through a slight increase. In this range of pressure recovery, the boundary layer is
retarded by the adverse pressure gradient, which consequently leads to the separation of the boundary
layer, as shown in Fig. 10. The pressure coefficient remains almost constant after about θ = 90◦ .
Distinctive features could also be noted in Fig. 17. Firstly, although the Cp of the polygonal and
ridged cylinders follow a similar trend with the unmodified cylinder globally, a zig-zag distribution of
405 the pressure coefficients could be seen in the leeward side (Fig. 17(a)), and the oscillations in the zig-
zags exacerbate as θ increases. As the boundary layer travels downstream from the stagnation point, it
encounters the ridges or the corners repeatedly. The drastic changes in the geometry curvature foster
repeated local separation followed by local reattachment, giving rise to the oscillatory pattern of the
pressure distribution. After θ = 90◦ , as the boundary layer has completely separated, the pressure
410 coefficients remains almost unchanged, with their values being close to that of the unmodified cylinder.
Such features have also been identified by Yamagishi and Oki [12], as is cited in Fig. 17(a). In their
work, a cylinder with 32 triangular grooves at Re = 1 × 105 was studied, and drag reduction was
observed, whereas in the current paper, despite the distinctive feature of the pressure distribution in

15
 1 unmodified 1 unmodified
polygonal node of line-wavy
0.5 ridged saddle of line-wavy
groove (Yamagishi) 0.5

0 0
Cp Cp
-0.5
-0.5
-1
-1
-1.5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
θ θ
(a) 2 dimensional cylinders (b) linear wavy cylinder

1 unmodified 1 unmodified
node of sine-wavy node of O-ringed
saddle of sine-wavy 0.5 saddle of O-ringed
0.5
unmodified (Lam) unmodified (Nakamura)
node (Lam) 0 node (Nakamura)
0 saddle (Lam) saddle (Nakamura)
Cp Cp -0.5
-0.5
-1
-1 -1.5
-1.5 -2
0 30 60 90 120 150 180 0 30 60 90 120 150 180
θ θ
(c) sinusoidal wavy cylinder (d) O-ringed cylinder

Figure 17: Mean pressure coefficients of the characteristic positions of the cylinders

the frontal part of the cylinders, the integrated values of which are almost identical with the unmodified
415 circular cylinder. The main difference between our and the cited results lies in the rear part of the
cylinders, where, owing to the Reynolds number difference, the pressure predicted by the current paper
is remarkably lower than that of Yamagishi and Oki [12]. It is also due to this difference that we fail
to achieve satisfactory drag reduction in the present study.
As for the two wavy cylinders (Fig. 17(b), Fig. 17(c)), results are presented at the node and saddle
420 characteristic sections. The results of the WY-A4 model (λ/D = 2.273, a/D = 0.091) from the work
of Lam and Lin [30] are also presented in Fig. 17(c) for comparison. Owing to the combined effect of
Re, shape parameters, turbulence model, etc., their predicted drag was slightly higher than ours, as
is manifested in the lower Cp in the rear part of the cylinder. However, generally, their results lend
support to the conclusions that will be made in the current paper. Firstly, the minimum value of Cp
425 appears at smaller θ at the saddle than that at the node. This is associated with the disparity in the
separation angle at saddle and node as mentioned previously. Without being rigorous, we deduce that
the forwarded appearance of the minimum Cp results in an forwarded commencement of the adverse
pressure gradient, hence the separation at the saddle is made prior to that at the node. Besides,
although the base pressure at the node and saddle converge to the same value, it is noticeably larger
430 than the unmodified cylinder, which means less suction on the rear side. This fact is the direct cause
of the drag reduction exhibited by the two wavy cylinders.
A more illustrative way to explain the drag reduction mechanism is provided here. Considering
again Eq. 11, the product Cp · nx could be considered as the contribution of each angular locations
to the total drag, as sketched in Fig. 18. We see that as explained earlier, the drag mitigation of the
435 wavy cylinder is mainly achieved in the rear part of the cylinder. Besides, it could be observed that in
the range θ = 30◦ ∼ 70◦ , the Cp at the saddle contributes more negatively to the total drag than that
at the node. It is thus clear that the pressure difference in this region is responsible for the disparity

16
 of Cd at the node and saddle, as observed in Fig. 14(a) and Fig. 15(a).

1.2
unmodified cylinder
node of sine-wavy
0.8 saddle of sine-wavy

0.4
Cp⋅cosθ
0

-0.4

-0.8
0 30 60 90 120 150 180
θ

Figure 18: Angular contribution of Cp to Cd

In the case of the cylinder with O-rings (Fig. 17(d)), while the mean pressure distribution at the
440 plane part is the same with the unmodified cylinder, it appears that the Cp trough at the ringed part
is much delayed than that at the saddle part. Besides, on the rings, the pressure recovery after Cp the
bottom is much weaker, which also means a weaker adverse pressure gradient. This might be the cause
for the difficulty in finding the separation point on the node of the O-ringed cylinder in Table 2. The
delayed separation at the ring has also been identified by Nakamura and Igarashi [35] at Re = 20900,
445 as is cited in Fig. 17(d). In the current paper, the mean pressure in the rear side of cylinder at both
locations is also identical to that of the unmodified cylinder. This is contrasted by the successful
drag reduction in the cited work, in which the pressure on the unringed part recovers significantly in
the rear of the cylinder. Among the many different parameters, such as O-ring shape, size, pitching
distance, between our study and the cited work, we cautiously attribute the distinctive difference in
450 the pressure distribution to the lower Re we have studied, as this effect has been proven critical in
previous studies on the O-ringed cylinders [34, 35]. It was revealed by Nakamura and Igarashi [35] that
the drag reduction could not be achieved until the Re exceeds 10,000. Lim and Lee [34] also found out
that Re plays an important role in the effectiveness of the O-rings, although the size of which in their
study was much smaller. The Re studied in the current work falls far short of those in the mentioned
455 studies. Actually, as Nakamura [65] further explained in a domestic journal, the separation of the
boundary layer at the ringed location is significantly delayed, which gives rise to the shrinked wake.
This promotes the formation two transverse vortices (in the coordinate system of the current paper,
the vortices rotate around the y axis) on both sides of the rings. These vortices obstruct the shedding
of Kármán vortices and thus reduce the aerodynamic forces. Compared with the work of Nakamura
460 [35, 65], the configuration of the O-rings in our work is smaller in size, but closer in pitching distance.
Some features, such as the delayed separation, shrinked wake at the ringed locations are observed
in the present paper. However, the two transverse vortices could not be visualized, thus the vortex
shedding in the unringed part is left unaffected. With further increasing the Reynolds number, it is
anticipated that the transverse vortices could emerge and grow in intensity and finally help to mitigate
465 the drag and lift forces. Further investigations are under consideration to confirm this conjecture.
We are now in position to analyse and compare the flow control mechanisms provided by the
different modifications we have undertaken. As could be concluded, one common characteristic of the
successful practices of drag mitigation is the higher pressure at the rear part of the cylinder. However,
it is achieved by different mechanisms. As far as the current paper is concerned, the polygonal, ridged
470 and grooved cylinders achieve this goal by energizing the boundary layer and prompting the drag
crisis. The wavy cylinders modify the separated free shear layers and thus stablize the flow wake.
The installation of the O-rings produces a pair of transverse vortices on both sides of the rings and
suppresses the Kármán vortices. The effectiveness of the polygonal, ridged and the O-ringed cylinders
could only be manifested when the Re is higher than a critical value. The wavy cylinders, one the other
475 hand, are capable of drag reduction at a larger range of Re. However, because of their span-wisely
3 dimensional shapes, the difficulty in manufacture of this kind of cable might inhibit its wide use in

17
 engineering.

3.3. Flow instabilities

In the context of circular cross flow, there are mainly two kinds of flow instabilities that attract the
480 attention of engineers and researchers. The first one, known as Kármán vortex, exists in the wake of
various kinds of bluff bodies, and is the source to many of the undesired flow induced phenomena such
as excessive aero/hydro-dynamic forces, vortex induced vibration, flow induced noise, etc. Kármán
vortex originates from the absolute instability in the flow wake [66] and features a distinctively sharp
peak in the power spectrum of the fluctuating flow variables. The value of this peak frequency is
485 decided by the Strouhal number (St = fkv D/U ), which, in the case of a circular cylinder at sub-
critical Reynolds number, is around 0.20 ∼ 0.21. Compared to the Kármán vortex, the knowledge of
this second instability, commonly known as the Kelvin-Helmholtz instability or shear layer instability,
is less established. Taking place in the free shear layer, the K-H vortices develop in the same fashion as
those observed in the mixing layers, except that the length of the layer is restricted by the stream-wise
490 extent of the recirculation length [67]. Bloor [68] was credited for being the first to detect the frequency
of the K-H instability as early as the year of 1964. Up to now, controversy still exists regarding the
onset Reynolds number for K-H instability, the corresponding frequency [69], and the cause for its
intermittency [70].

3.3.1. The Kármán vortex

495 By performing Fast Fourier Transformation to the time history of the lift force coefficients the
Kármán vortex frequency could be identified, as depicted in Fig. 19.
The power spectrum density of the six cases resemble each other in that they are all characterized
by one sharp peak corresponding to the Kármán vortex shedding frequency. It is easily concluded that
the fkv of the different cylinders lies at a close, if not the same, value of around 0.20 to 0.21. It seems
500 that the absolute instability in the wake, which gives rise to the Kármán vortex street, is insensitive
to these shape modifications.
One more doubt exists in that whether the local shedding frequency at different axial locations in
the 3 dimensional cylinders would differ between each other, since the local Reynolds number varies
with the corresponding diameter. This conjecture is negated by an examination of the flow variables
505 at different span-wise locations, as will be presented in the next subsection.

3.3.2. Kelvin-Helmholtz instability

The transverse component of velocity Uy of the unmodified cylinder near the free shear layer zone is
presented in Fig. 20. A global frequency could be expected from both time traces owing to the periodic
shedding of Kármán vortex. However, intermittently, some high frequency oscillations could also be
510 recognized at certain instant, say t ≈ 50, 70, 120, 180, etc. These high frequency packets correspond
to the Kelvin-Helmholtz instability as this subsection is about to address. The power spectrum are
calculated from these velocity oscillograms to give a better perception of the K-H instability, as shown
in Fig. 21. It could be observed that apart from the Kármán vortex frequency at fkv ≈ 0.21, another
peak fsl , representing the K-H instability, could also be discerned. While the Kármán vortex frequency
515 is characterized by the distinctively sharp peak in the power spectra, the K-H frequency, by contrast,
seems to be consisted of multiple peaks clustered between fsl ≈ 1.1 to fsl ≈ 1.5. This blunt peak is
thought to be attributed to the small variations in the separation point of the boundary layer during
the process of vortex shedding[68].
Prasad et al. [71] conducted a comprehensive investigation into the shear layer instability behind a
520 circular cylinder. In their experiments, measurements were carried out with hot wire probes fixed at
locations near the outer edge of the shear layer. They found out that the instability was at some time
barely perceptible, while at others had involved into discrete shear layer vortices. They deduced that
the intermittency was a result of the onset point of the K-H instability transitioning randomly past
the probed location. However, in the current work, we observe in our flow visualization animation that
525 the shear layer at some instants behaves quite stable before it involves into the Kármán vortex, and at
some other instants are subjected to the K-H instability. We present the above animation in the form
of its two characteristic clips. Fig. 22 shows the velocity magnitude fields of two time instants. The left

18
 10
-2
10-2
fkv=0.205 fkv=0.209 fkv=0.202
10
-4 10-4 10
-4

-6 10-6 10-6
PSD 10 PSD PSD
-8
10-8
10 10-8
10-10 10-10
10-10
0.1 1 f 10 0.1 1 f 10 0.1 1 f 10

(a) unmodified (b) polygonal (c) ridged

-2
fkv=0.21 10-4 fkv=0.21 10 fkv=0.202
10-4
10 -6
10-4
10-6 10-8 10-6
PSD PSD10-10 PSD
10-8 10-8
10-12
10-14 10-10
10-10
10-16 10-12
0.1 1 f 10 0.1 1 f 10 0.1 1 f 10

(d) linear wavy (e) sinusoidal wavy (f) O-ringed

Figure 19: Power spectrum density of the lift force coefficients

0.15

Uy
0

-0.15
0 25 50 75 100 125 150 175 200

0.15

Uy
0

0 25 50 75 100 125 150 175 200

t

Figure 20: Oscillogram of Uy near the free shear layer of the unmodified cylinder. Up: x/D = 0.6, y/D = 0.6; below:
x/D = 0.6, y/D = 0.7.

10-4 fkv 10-4 fkv

-6 fsl -6 fsl
10 10
-8 -8
PSD 10 PSD 10
10-10 10-10
10-12 10-12
0.1 1 f 10 0.1 1 f 10

Figure 21: Power spectrum density of the velocity oscillograms. Left: x/D = 0.6, y/D = 0.6; right: x/D = 0.6,
y/D = 0.7.

figure shows no evidence of K-H instability while in the right the flapping waves in the lower cylinder
shoulder clearly indicates that the instability is taking place. This comparison indicates that rather
530 than the problem of the probe detection, the intermittency of the shear layer instability might be a

19
 nature of its own. Our conjecture is supported by a recent computational investigation by Rai [70],
who proposed that the principal cause of the intermittency is the strengthening of shear-layer vortices
via stretching during certain time periods.

Figure 22: The stable shear layer(left) and the K-H instability in the shear layer(right).
535 The velocity power spectra of the surface modified cylinders are presented in Fig. 23. The power
spectra of the polygonal cylinder resembles that of the circular cylinder, for the existence of a broad
and blunt peak at fsl ≈ 1. The ridged cylinder, however, generate a sharper peak at fsl ≈ 1.18. This
might be ascribed to the fixed separation point at the top of ridge near θ ≈ 90◦ .
As for the 3 dimensional cylinders, the Kármán vortex frequency at the node and saddle sits at the
540 same value with that obtained from the lift force coefficient. However, the instability of the free shear
layer behaves differently at different span-wise locations. The sinusoidal wavy cylinder, for example,
at the node the free shear layer frequency is around fsl = 0.9, while the value is around 1.3 at the
saddle. The general conclusion that fsl is higher at the saddle than that at the node could be made
to the wavy cylinders. We now propose to explain the cause of this difference.

10-4 10-4
10-6
10-6 10-6 10-6
10-8 10-8
10-8 10-8
PSD

PSD

PSD

PSD

10-10 10-10
10-10 10-10
10-12 10-12
10-12 10-12
10-14 10-14
0.1 1 f 10 0.1 1 f 10 0.1 1 f 10 0.1 1 f 10

(a) polygonal cylinder (b) ridged cylinder (c) node of linear wavy (d) saddle of linear wavy
-6 10-4
10 10-4
10-6 10-6
10-8 10-6
10-8
10-10 10-8
PSD

PSD

PSD

PSD

10-8
10-10 10-12 10-10
10 -10 10-14 10-12
10-12 10-16 10-14
0.1 1 f 10 0.1 1 f 10 0.1 1 f 10 0.1 1 f 10

(e) node of sine wavy (f) saddle of sine wavy (g) node of O-ringed (h) saddle of O-ringed

Figure 23: Power spectrum of the y component of velocity in the free shear layer area

545 On the dimensional basis, the free shear layer frequency will scale with a characteristic velocity
and a characteristic length scale in the form, fsl ∝ Uc /lc . While Uc is taken proportional to the free
stream velocity U∞ by the previous researchers, there seems to be multiple choices for lc [68, 71, 72].
One of the reasonable choices is the boundary layer thickness at separation. In the current research,
owing to the well resolved cylinder vicinity, Uc and lc are selected to be the velocity and thickness of
550 boundary layer at separation, as shown in Fig. 24. At both the node and the saddle, the boundary layer
separates at a close, if not the same, velocity. However, in the two wavy cylinder cases the boundary

20
 layer thickness at the node is considerably larger than that at the saddle. It is thus clear that at the
node, the characteristic frequency, which scales with Usep /l, is smaller than that at the saddle. This
should explain the shear layer frequency difference we observe in Fig. 23.

0.1 0.1
saddle 79°° saddle 82°°
0.08 node 99 0.08 node 92

0.06 0.06
l l
0.04 0.04

0.02 0.02

0 0
0 0.4 0.8 1.2 0 0.4 0.8 1.2
Usep Usep

(a) linear wavy (b) sinusoidal wavy

Figure 24: Boundary layer profiles at separation of the wavy cylinders

555 3.4. Span-wise correlation coefficients

Span-wise correlation of the flow variables, such as the lift force coefficients, is an important meas-
urement of the 3-dimensionality of the flow. It is also one of the factors that affects the vortex induced
vibrations. Gabbai et al. [73] stated that the absence of the two-dimensionality in the geometry of
the bluff-body might lead to significant reductions in the vortex shedding correlation length. It is
560 thus interesting to inspect the differences in the correlations of these shape modified cylinders. The
correlation coefficient is defined by:
E [(z1 − µz1 ) (z2 − µz2 )]
ρ (z1 , z2 ) = (13)
σz1 σz2
in which µx and σx are the mean and standard deviation of the variable x. Here z1 and z2 could
represent the force coefficients at any two different span-wise locations.
Taking the cylinder bottom end as the reference point, the Cd and Cl correlation coefficients ρ (0, z)
565 of the six cylinders are shown in Fig. 25. ρ (0, 6.28) = 1 is a reflection of the periodic boundary condition
used in the span-wise direction, which means that only half of the cylinder length could be used for
correlation coefficient analysis. The Cd correlation appears rather different from case to case. The
sinusoidal wavy cylinder is best correlated in drag, its correlation coefficient staying as high as 0.6.
The Cd correlation of the unmodified and polygonal cylinders, however, decrease sharply as the axial
570 separation increases.
It is more physically meaningful to inspect the correlation coefficients of the lift forces, since it is
related to the vortex shedding process along the span-wise direction. Norberg [74] pointed out that
at large axial separation the lift correlation coefficients is expected to vanish because of the highly 3-
dimensionlized vortex shedding. However, in the current work the lift forces are quite well correlated.
575 This is particularly true for the 2 dimensional geometries, of which the correlation coefficients remain
above 0.85. This observation indicates that the major vortices are shedding in an almost 2-dimensional
fashion along the axial length. Lee et al. [58] investigated the span-wise pressure correlation of circular
cylinders with varying aspect ratios numerically. It was found that it is not until the axial length reaches
16D (periodicity used for the span-wise boundary condition) that a plausible prediction in terms of
580 span-wise correlation could be established. Clearly, in the current simulation the span-wise length falls
far short of that requirement. Nevertheless, some notable features could still be observed. Considerably
lower correlation coefficients are observed in the two wavy cylinders. It could be anticipated that the
correlation length of an ideally long wavy cylinder would be smaller than that of an unmodified cylinder.
If this conjecture holds true, then in addition to the mitigated aerodynamic forces, the wavy cylinders
585 present another merit that a rigid cylinder with less correlated lift force is more resistant to the vortex
induced vibration. The Cl correlation coefficient of the O-ringed cylinder stays as high as the 2-D

21
 1 1

0.8 0.95

0.6 0.9

0.4 0.85
ρ(0,z) ρ(0,z)
0.2 0.8

0 0.75 unmodified
unmodified polygonal
polygonal ridged
-0.2 ridged 0.7 linear wavy
linear wavy sinusoidal wavy
sinusoidal wavy O-ringed
O-ringed
-0.4 0.65
0 1 2 3 4 5 6 0 1 2 3 4 5 6
z/D z/D

(a) Cd (b) Cl

Figure 25: Span-wise correlation coefficients of the force coefficients

geometries. Although 3-dimensionality is introduced into the body, it seems that the effect of these
local perturbations to the span-wise correlations is small.

4. Conclusion

590 With the aim of studying in detail the flow properties and exploring the flow control mechanisms,
the flow past shape modified circular cylinders at subcritical Reynolds number of 5000 was investigated
by the LES method. The modifications were performed in both cross-section and span-wise direction.
A mesh dependency test was done for the unmodified cylinder to validate the adequacy of the mesh
resolution. The main findings of the current work are summarized as follows.
595 (1) The wakes of the 2-dimensional cylinders, i.e., the unmodified, polygonal and ridged cylinders
were similar in that the separation angle, recirculation length, etc., were close. The wakes behind the
wavy cylinders exhibited significant difference. At the node section the separation was delayed, leading
to a narrower and shorter wake and at the saddle the wake was wider and longer, due to the forwarded
separation. This difference had a strong impact on the other flow variables such as aerodynamic forces,
600 shear layer instabilities, and so on. The O-ringed cylinder only affected the wake locally.
(2) The two wavy cylinders, amongst the six cases, gave rise to considerable drag and lift mitigation.
The underlying mechanism for this reduction is the elongated vortex recirculation length, which leads
to a larger value of base pressure. This flow control mechanism is anticipated to be effective in higher
Reynolds number regimes. The difference of Cd between the node and the saddle sections was found
605 to be attributed mainly to the pressure difference between θ = 30◦ ∼ 70◦ . The polygonal and ridged
cylinders failed to produce the desired effect of drag crisis, mainly due to the low Reynolds number in
this study. The effect of the rings on the aerodynamic forces was also minor.
(3) The wakes of the six cases were all governed by the Kármán vortex with nearly the same
frequency. The frequency of the absolute instability was barely affected by the shape modifications.
610 Difference between the cases existed in the shear layer instability, the cause of which was explained by
the boundary layer properties at separation. It was also observed that the intermittency of the shear
layer instability seemed to be a nature of its own, rather than the detection problem as was proposed
by the previous researchers.
(4) The correlation coefficients were found high due to the small aspect ratio. A general conclusion
615 could be made that in the case of lift force, the 2-dimensional cylinders were better correlated than
the 3-dimensional ones.
In view of the above, it is concluded that the flow wake properties were significantly affected in
a beneficial way by the span-wise waviness. Although cylinders with waviness are seldomly seen in
the practical applications, it has been proven by the previous researches as well as the current work
620 that it has the potential to counteract the excessive aerodynamic forces of a cable at a large range

22
 of Reynolds numbers. However, the dynamic response, such as vortex induced vibration, rain-wind
induced vibration, dry galloping, etc., of this particular shape is still not understood. Future researches
should address these aspects. The modifications made in the cross-section were barely felt by the flow
wake, although they are commonly used in industrial applications such as overhead transmission lines.
625 Simulations with higher Reynolds number are in need to reproduce the effect of these modifications.
This article has adopted a rather arbitrary arrangement of the O-rings, resulting a disappointing flow
control effect. In future simulations, optimal arrangement of the O-rings should be sought to achieve
satisfactory control of the flow.

Acknowledgement

630 Support from the Major Program of the National Natural Science Foundation of China (No.
51490674) and National Natural Science Foundation of China (No. 51278297), Doctoral Disciplinary
Special Research Project of Chinese Ministry of Education (No.20130073110096), are acknowledged.

References

[1] H. Choi, W.-P. Jeon, J. Kim, Control of flow over a bluff body, Annual Review of Fluid Mechanics
635 40 (2008) 113–139.
[2] P. Poncet, Vanishing of mode B in the wake behind a rotationally oscillating circular cylinder,
Physics of Fluids 14 (6) (2002) 2021–2023.
[3] S. Mittal, B. Kumar, Flow past a rotating cylinder, Journal of Fluid Mechanics 476 (2003) 303–
334.
640 [4] J.-W. He, R. Glowinski, R. Metcalfe, A. Nordlander, J. Periaux, Active control and drag optim-
ization for flow past a circular cylinder: I. oscillatory cylinder rotation, Journal of Computational
Physics 163 (1) (2000) 83–117.
[5] J. H. Fransson, P. Konieczny, P. H. Alfredsson, Flow around a porous cylinder subject to continu-
ous suction or blowing, Journal of Fluids and Structures 19 (8) (2004) 1031–1048.
645 [6] N. Fujisawa, G. Takeda, N. Ike, Phase-averaged characteristics of flow around a circular cylinder
under acoustic excitation control, Journal of Fluids and Structures 19 (2) (2004) 159–170.
[7] J. Kim, H. Choi, Distributed forcing of flow over a circular cylinder, Physics of Fluids (1994-
present) 17 (3) (2005) 033103.
[8] E. Achenbach, Influence of surface roughness on the cross-flow around a circular cylinder, Journal
650 of Fluid Mechanics 46 (02) (1971) 321–335.
[9] Y. Nakamura, Y. Tomonari, The effects of surface roughness on the flow past circular cylinders
at high reynolds numbers, Journal of Fluid Mechanics 123 (1982) 363–378.
[10] P. Bearman, J. Harvey, Control of circular cylinder flow by the use of dimples, AIAA Journal
31 (10) (1993) 1753–1756.
655 [11] T. Kimura, M. Tsutahara, Fluid dynamic effects of grooves on circular cylinder surface, AIAA
Journal 29 (12) (1991) 2062–2068.
[12] Y. Yamagishi, M. Oki, Effect of groove shape on flow characteristics around a circular cylinder
with grooves, Journal of Visualization 7 (3) (2004) 209–216.
[13] Y. Yamagishi, M. Oki, Effect of the number of grooves on flow characteristics around a circular
660 cylinder with triangular grooves, Journal of Visualization 8 (1) (2005) 57–64.
[14] A. Ahmed, B. Bays-Muchmore, Transverse flow over a wavy cylinder, Physics of Fluids A: Fluid
Dynamics (1989-1993) 4 (9) (1992) 1959–1967.

23
 [15] K. Lam, Y. Lin, Effects of wavelength and amplitude of a wavy cylinder in cross-flow at low
reynolds numbers, Journal of Fluid Mechanics 620 (2009) 195–220.
665 [16] T. Zhou, S. M. Razali, Z. Hao, L. Cheng, On the study of vortex-induced vibration of a cylinder
with helical strakes, Journal of Fluids and Structures 27 (7) (2011) 903–917.

[17] S.-J. Lee, H.-B. Kim, The effect of surface protrusions on the near wake of a circular cylinder,
Journal of Wind Engineering and Industrial Aerodynamics 69 (1997) 351–361.
[18] J.-Y. Hwang, K.-S. Yang, Drag reduction on a circular cylinder using dual detached splitter plates,
670 Journal of Wind Engineering and Industrial Aerodynamics 95 (7) (2007) 551–564.

[19] Y. Bao, J. Tao, Active control of a cylinder wake flow by using a streamwise oscillating foil,
Physics of Fluids (1994-present) 25 (5) (2013) 053601.
[20] N. Udovidchik, J. F. Morrison, Investigation of active dimple actuators for separation control,
AIAA Paper 3190 (2006) 2006.

675 [21] P. W. Bearman, Golfball aerodynamics, Aeronautical Quarterly 27 (1976) 112–122.

[22] T. Matsumura, T. Yura, N. Kikuchi, T. Matsumoto, H. Takeuchi, K. Iwasaki, Y. Tominaga,
Development of low wind-pressure insulated wires, Furukawa Electric Review (2002) 39–44.
[23] P. W. Bearman, J. C. Owen, Reduction of bluff-body drag and suppression of vortex shedding by
the introduction of wavy separation lines, Journal of Fluids and Structures 12 (1) (1998) 123–130.

680 [24] M. El-Gammal, H. Hangan, P. King, Control of vortex shedding-induced effects in a sectional
bridge model by spanwise perturbation method, Journal of Wind Engineering and Industrial
Aerodynamics 95 (8) (2007) 663–678.
[25] J. C. Owen, P. W. Bearman, A. A. Szewczyk, Passive control of viv with drag reduction, Journal
of Fluids and Structures 15 (3) (2001) 597–605.
685 [26] J. Owen, A. Szewczyk, P. Bearman, Suppression of karman vortex shedding, Physics of Fluids
(1994-present) 12 (9) (2000) S9–S9.
[27] K. Lam, F. Wang, J. Li, R. So, Experimental investigation of the mean and fluctuating forces of
wavy (varicose) cylinders in a cross-flow, Journal of Fluids and Structures 19 (3) (2004) 321–334.

[28] K. Lam, F. Wang, R. So, Three-dimensional nature of vortices in the near wake of a wavy cylinder,
690 Journal of Fluids and Structures 19 (6) (2004) 815–833.
[29] K. Lam, Y. Lin, Drag force control of flow over wavy cylinders at low reynolds number, Journal
of Mechanical Science and Technology 21 (9) (2007) 1331–1337.

[30] K. Lam, Y. Lin, Large eddy simulation of flow around wavy cylinders at a subcritical reynolds
number, International Journal of Heat and Fluid Flow 29 (4) (2008) 1071–1088.
695 [31] C.-Y. Xu, L.-W. Chen, X.-Y. Lu, Large-eddy simulation of the compressible flow past a wavy
cylinder, Journal of Fluid Mechanics 665 (2010) 238–273.

[32] S.-J. Lee, A.-T. Nguyen, Experimental investigation on wake behind a wavy cylinder having
sinusoidal cross-sectional area variation, Fluid Dynamics Research 39 (4) (2007) 292–304.
[33] K. Kleissl, C. Georgakis, Shape modification of bridge cables for aerodynamic vibration control,
700 Advances and Trends in Structural Engineering, Mechanics and Computation (2010) 16.
[34] H.-C. Lim, S.-J. Lee, Flow control of a circular cylinder with O-rings, Fluid Dynamics Research
35 (2) (2004) 107–122.

24
 [35] H. Nakamura, T. Igarashi, Omnidirectional reductions in drag and fluctuating forces for a circular
cylinder by attaching rings, Journal of Wind Engineering and Industrial Aerodynamics 96 (6)
705 (2008) 887–899.
[36] H.-C. Lim, S.-J. Lee, Flow control of circular cylinders with longitudinal grooved surfaces, AIAA
Journal 40 (10) (2002) 2027–2036.
[37] M. M. Zdravkovich, Review and classification of various aerodynamic and hydrodynamic means
for suppressing vortex shedding, Journal of Wind Engineering and Industrial Aerodynamics 7 (2)
710 (1981) 145–189.

[38] J. Vandiver, S. Swithenbank, V. Jaiswal, H. Marcollo, et al., The effectiveness of helical strakes
in the suppression of high-mode-number viv, in: Offshore Technology Conference, Offshore Tech-
nology Conference, 2006.
[39] K. Kleissl, C. Georgakis, Comparison of the aerodynamics of bridge cables with helical fillets and
715 a pattern-indented surface, Journal of Wind Engineering and Industrial Aerodynamics 104 (2012)
166–175.

[40] N. Kikuchi, Y. Matsuzaki, T. Yukino, H. Ishida, Aerodynamic drag of new-design electric power
wire in a heavy rainfall and wind, Journal of Wind Engineering and Industrial Aerodynamics
91 (1) (2003) 41–51.
720 [41] Y. Miyata, H. Yamada, T. Hojo, Experimental study on aerodynamic characteristics of cables
with patterned surface, Journal of Structural Engineering A 40 (1994) 1065–1076.
[42] T. Hojo, T. Miyata, H. Yamada, T. Fujiwara, Wind-resistant design of cables for the tatara bridge,
IABSE REPORTS (1998) 51–56.
[43] J. D. Anderson, J. Wendt, Computational fluid dynamics, Vol. 206, Springer, 1995.

725 [44] Z. Han, D. Zhou, J. Tu, Laminar flow patterns around three side-by-side arranged circular cylin-
ders using semi-implicit three-step taylor-characteristic-based-split (3-tcbs) algorithm, Engineer-
ing Applications of Computational Fluid Mechanics 7 (1) (2013) 1–12.
[45] J. Tu, D. Zhou, Y. Bao, C. Fang, K. Zhang, C. Li, Z. Han, Flow-induced vibration on a circular
cylinder in planar shear flow, Computers & Fluids 105 (2014) 138–154.

730 [46] T. He, Semi-implicit coupling of CS-FEM and FEM for the interaction between a geometrically
nonlinear solid and an incompressible fluid, International Journal of Computational Methods
12 (05) (2015) 1550025.
[47] M. Breuer, A challenging test case for large eddy simulation: high reynolds number circular
cylinder flow, International Journal of Heat and Fluid Flow 21 (5) (2000) 648–654.

735 [48] A. Travin, M. Shur, M. Strelets, P. Spalart, Detached-eddy simulations past a circular cylinder,
Flow, Turbulence and Combustion 63 (1-4) (2000) 293–313.
[49] H. Beem, M. Hildner, M. Triantafyllou, Calibration and validation of a harbor seal whisker-
inspired flow sensor, Smart Materials and Structures 22 (1) (2013) 014012.

[50] C. Fureby, G. Tabor, H. Weller, A. Gosman, A comparative study of subgrid scale models in
740 homogeneous isotropic turbulence, Physics of Fluids (1994-present) 9 (5) (1997) 1416–1429.
[51] D. A. Lysenko, I. S. Ertesvåg, K. E. Rian, Large-eddy simulation of the flow over a circular
cylinder at reynolds number 3900 using the openfoam toolbox, Flow, Turbulence and Combustion
89 (4) (2012) 491–518.

[52] J. Smagorinsky, General circulation experiments with the primitive equations: I. the basic exper-
745 iment, Monthly Weather Review 91 (3) (1963) 99–164.

25
 [53] M. Germano, U. Piomelli, P. Moin, W. H. Cabot, A dynamic subgrid-scale eddy viscosity model,
Physics of Fluids A: Fluid Dynamics (1989-1993) 3 (7) (1991) 1760–1765.
[54] D. K. Lilly, A proposed modification of the germano subgrid-scale closure method, Physics of
Fluids A: Fluid Dynamics (1989-1993) 4 (3) (1992) 633–635.

750 [55] H. G. Weller, G. Tabor, H. Jasak, C. Fureby, A tensorial approach to computational continuum
mechanics using object-oriented techniques, Computers in Physics 12 (6) (1998) 620–631.
[56] D. A. Lysenko, I. S. Ertesvåg, K. E. Rian, Large-eddy simulation of the flow over a circular
cylinder at reynolds number 2×104 , Flow, Turbulence and Combustion 92 (3) (2014) 673–698.

[57] C. Doolan, Large eddy simulation of the near wake of a circular cylinder at sub-critical reynolds
755 number, Engineering Applications of Computational Fluid Mechanics 4 (4) (2010) 496–510.
[58] A. H. Lee, R. L. Campbell, S. A. Hambric, Coupled delayed-detached-eddy simulation and struc-
tural vibration of a self-oscillating cylinder due to vortex-shedding, Journal of Fluids and Struc-
tures 48 (2014) 216–234.

[59] R. Courant, K. Friedrichs, H. Lewy, On the partial difference equations of mathematical physics,
760 IBM Journal of Research and Development 11 (2) (1967) 215–234.
[60] A. G. Kravchenko, P. Moin, Numerical studies of flow over a circular cylinder at ReD=3900,
Physics of Fluids (1994-present) 12 (2) (2000) 403–417.

[61] M. Breuer, Numerical and modeling influences on large eddy simulations for the flow past a
circular cylinder, International Journal of Heat and Fluid Flow 19 (5) (1998) 512–521.
765 [62] C. Norberg, LDV-measurements in the near wake of a circular cylinder, ASME Paper No.
FEDSM98-521.

[63] S. E. Kim, Large eddy simulation of turbulent flow past a circular cylinder in subcritical regime,
AIAA Paper 1418 (2006) 9–12.
[64] H. Schlichting, K. Gersten, K. Gersten, Boundary-layer theory, Springer Science & Business Me-
770 dia, 2000.
[65] H. Nakamura, Suppression of fluctuating lift on a circular cylinder by attaching cylindrical rings,
Journal of Fluid Science and Technology 6 (6) (2011) 1036–1050.
[66] P. Monkewitz, L. Nguyen, Absolute instability in the near-wake of two-dimensional bluff bodies,
Journal of Fluids and Structures 1 (2) (1987) 165–184.
775 [67] S. Rajagopalan, R. Antonia, Flow around a circular cylinderstructure of the near wake shear layer,
Experiments in Fluids 38 (4) (2005) 393–402.
[68] M. S. Bloor, The transition to turbulence in the wake of a circular cylinder, Journal of Fluid
Mechanics 19 (02) (1964) 290–304.
[69] S. I. Green, Fluid vortices: fluid mechanics and its applications, Vol. 30, Springer Science &
780 Business Media, 1995.

[70] M. M. Rai, A computational investigation of the instability of the detached shear layers in the
wake of a circular cylinder, Journal of Fluid Mechanics 659 (2010) 375–404.
[71] A. Prasad, C. H. Williamson, The instability of the shear layer separating from a bluff body,
Journal of Fluid Mechanics 333 (1997) 375–402.

785 [72] M. Unal, D. Rockwell, On vortex formation from a cylinder. part 1. the initial instability, Journal
of Fluid Mechanics 190 (1988) 491–512.

26
 [73] R. Gabbai, H. Benaroya, An overview of modeling and experiments of vortex-induced vibration
of circular cylinders, Journal of Sound and Vibration 282 (3) (2005) 575–616.
[74] C. Norberg, Fluctuating lift on a circular cylinder: review and new measurements, Journal of
790 Fluids and Structures 17 (1) (2003) 57–96.

27