You are on page 1of 30

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

net/publication/272318599

Characterisation and analysis of flow over two side by side cylinders for different
gaps at low Reynolds number: A numerical approach

Article  in  Physics of Fluids · June 2014


DOI: 10.1063/1.4883484

CITATIONS READS

28 1,050

2 authors, including:

K. Supradeepan
BITS Pilani, Hyderabad
37 PUBLICATIONS   94 CITATIONS   

SEE PROFILE

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

POD Analysis on low Reynolds number flow past bluff bodies View project

All content following this page was uploaded by K. Supradeepan on 17 March 2016.

The user has requested enhancement of the downloaded file.


Characterisation and analysis of flow over two side by side cylinders for different gaps
at low Reynolds number: A numerical approach
K. Supradeepan and Arnab Roy

Citation: Physics of Fluids (1994-present) 26, 063602 (2014); doi: 10.1063/1.4883484


View online: http://dx.doi.org/10.1063/1.4883484
View Table of Contents: http://scitation.aip.org/content/aip/journal/pof2/26/6?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in


Flow induced vibration of two rigidly coupled circular cylinders in tandem and side-by-side arrangements at a low
Reynolds number of 150
Phys. Fluids 25, 123601 (2013); 10.1063/1.4832956

Numerical simulation of vortex-induced vibration of two circular cylinders of different diameters at low Reynolds
number
Phys. Fluids 25, 083601 (2013); 10.1063/1.4816637

Numerical simulation of vortex-induced vibration of a square cylinder at a low Reynolds number


Phys. Fluids 25, 023603 (2013); 10.1063/1.4792351

Low Reynolds number flow over a square cylinder with a splitter plate
Phys. Fluids 23, 033602 (2011); 10.1063/1.3563619

Linear shear flow over a square cylinder at low Reynolds number


Phys. Fluids 17, 078103 (2005); 10.1063/1.1953727

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
PHYSICS OF FLUIDS 26, 063602 (2014)

Characterisation and analysis of flow over two side


by side cylinders for different gaps at low Reynolds
number: A numerical approach
K. Supradeepana) and Arnab Royb)
Department of Aerospace Engineering, Indian Institute of Technology Kharagpur,
Kharagpur 721302, West Bengal, India
(Received 24 September 2013; accepted 3 June 2014; published online 18 June 2014)

Numerical simulations were performed for two-dimensional viscous incompressible


flow past two stationary side by side circular cylinders at Reynolds number (Re) 100
by varying centre to centre distance between the cylinders from 1.1 to 8 times the
diameter (D) of a cylinder. The incompressible Navier-Stokes equations were solved
using a finite volume method. Five different flow regimes were observed in the flow
as the distance between the cylinders was increased systematically. An attempt has
been made to characterise the different flow regimes using different numerical tools
like Proper Orthogonal Decomposition (POD), λ2 criterion, instantaneous streamwise
normal stress produced due to periodic variation of flow, frequency spectrum analysis
and phase diagrams. All these tools confirm the clear existence of five different
regimes in the flow when the gap between the cylinders is varied over the above
stated range. C 2014 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4883484]

I. INTRODUCTION
Flow over a pair of circular cylinders arranged side by side is a problem of fundamental interest
in the fluid dynamics domain because of the complexities involved in strong vortex-vortex interaction
phenomenon which arises due to their close proximity. Such interactions lead to modification of
the lift and drag coefficients acting on the cylinders. The time variation of these forces lead to flow
induced vibrations. This problem has enormous engineering importance. Structural design of such
cylindrical bodies needs sound understanding of the flow physics. The difficulties get compounded
when two or more cylinders are critically close to each other in which case the interactions are
stronger and often aperiodic. The interaction does not only depend on the distance between the
cylinders but also on the Reynolds number of the flow. Flow over two side by side cylinders has
been explored both numerically and experimentally in the past by many researchers. However, a
complete understanding of such flow has not yet been achieved.
Numerical and experimental investigation of flow over two side by side cylinders has been
reported by Ding et al.,1 Bearman and Wadcock,2 Chang and Song,3 Inoue et al.,4 Kang,5 Liu et al.,6
Sumner et al.,7 Williamson,8 Xu et al.,9 Yoon and Yang,10 Zdravkovich,11 and many other contem-
porary works. Most of the existing research focuses on the variation of lift and drag coefficients and
classification of the flow into various wake patterns and categories with varying Reynolds number
and gap between cylinders. POD has been increasingly utilized in the recent times to identify the
coherent structures and energetic modes in a flow field. Also, POD is being used to obtain reduced
order model of complex flow problems. POD analysis of low Reynolds number flow past a square
cylinder is reported by Borggaard et al.12 and Hay et al.13 and circular cylinder by Sengupta et al.14
Solution of Navier-Stokes equations in unstructured grid using finite volume method has become

a) Electronic mail: supradeepan@gmail.com


b) Electronic mail: arnab@aero.iitkgp.ernet.in

1070-6631/2014/26(6)/063602/28/$30.00 26, 063602-1 


C 2014 AIP Publishing LLC

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-2 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

popular in recent times especially for solving flow past complex geometries. Consistent Flux Recon-
struction for unstructured grids (CFRUNS) is one such recently developed scheme which has been
applied to low Reynolds number multi-cylinder flow simulation as reported by Harichandan and
Roy.15 The present study focuses on flow simulations involving an improved version of the original
CFRUNS scheme. This improved scheme is used to study the various flow regimes observed in the
flow over two side by side circular cylinders for various centre to centre distances at Re = 100.
For the present study several tools have been used to analyse the complex features of the flow
field and to identify different flow regimes depending on the gap existing between two cylinders.
Alternate shedding of vortices from the cylinders and their subsequent interaction produces time
varying loads on them. Dominant frequencies have been obtained from the spectra of time history of
force coefficients using FFT. The swirling and shearing regions in the vicinity of the cylinders and
their wake have been studied using λ2 criterion. POD modes have been obtained to study coherent
structures and energetic modes in the flow field. Stresses due to periodic variation of flow field have
been investigated. Phase space representation has been utilised to analyse the dynamic behavior of
the flow field.

II. GOVERNING EQUATIONS


The equations governing two-dimensional viscous incompressible isothermal flow are the con-
tinuity equation and momentum equations, which together constitute the Navier-Stokes equations.
The nondimensional conservative form of equations is given as follows:
Continuity equation:
∂u ∂v
+ = 0. (1)
∂x ∂y
Momentum equations:
x-component:
 
∂u ∂u 2 ∂uv ∂p 1 ∂ 2u ∂ 2u
+ + =− + + , (2)
∂t ∂x ∂y ∂x Re ∂ x 2 ∂ y2
y-component:
 
∂v ∂uv ∂v 2 ∂p 1 ∂ 2v ∂ 2v
+ + =− + + 2 . (3)
∂t ∂x ∂y ∂y Re ∂ x 2 ∂y
In the above equations, the nondimensional velocity components are u and v defined along x
and y directions, respectively. The ratio of pressure to density is represented by p and Reynolds
number by Re. Body forces are assumed to be zero in the above equations.

III. NUMERICAL SCHEME


The governing equations (1)–(3) are numerically solved using a finite volume Consistent Flux
Reconstruction (CFR)16 scheme for unstructured grid (CFRUNS).15 This scheme uses cell centered
primitive variables. A second order Adams-Bashforth time marching scheme is used. The convective,
diffusive, and pressure terms are all discretised using central difference scheme. This makes the solver
a formal second order accurate scheme. In the present study, the derivatives are calculated based on
least square procedure for gradient reconstruction as outlined by Mavriplis17 instead of the Taylor
series based approximation which is used by Harichandan and Roy15 and Roy and Bandyopadhyay.16
This enhances the robustness of the solver especially for unstructured mesh when used in complex
geometries as reported by Mavriplis.17 Geometrical description of the flow domain is provided in
Fig. 1.
Details of the initial and boundary conditions are provided as follows.
Free stream velocity and pressure has been prescribed as initial condition for all cells in the flow
domain, i.e., u = u∞ , v = v∞ , p = p∞ . This physically means that the body is suddenly immersed

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-3 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 1. Computational domain.

into an uniform flow at t = 0. The values of nondimensional freestream velocity components and
pressure are prescribed as u∞ = 1.0, v∞ = 0, p∞ = 0.
The following boundary conditions have been prescribed at the various boundaries of the flow
domain:
At the inflow boundary, an uniform velocity profile is assumed and zero normal pressure gradient
is applied: u = u∞ , v = v∞ , ∂∂ px = 0.
Zero shear boundary condition is applied for the velocity field at the top and bottom boundaries
along with zero normal pressure gradient. ∂u ∂y
= 0, v = v∞ , ∂∂ py = 0.
Convective boundary condition is applied at the outlet for the velocity components along with
Dirichlet free stream pressure: ∂φ ∂t
+ Uc ∂φ
∂x
= 0, p = p∞ , where φ = u, v. The value of Uc , which
is the convective velocity, has been set equal to the nondimensional streamwise velocity which is
equal to unity.
No slip boundary condition is applied on the body surface for velocity components along with
zero normal pressure gradient: u = 0, v = 0, ∂∂np = 0.

IV. RESULTS AND DISCUSSIONS


In the present work, a second order spatio-temporal accurate CFRUNS scheme is used to
simulate flow over two side by side circular cylinders at Re = 100. To understand the physics of
the flow field, various tools like streamlines, vorticity (ω) contours, λ2 criterion, POD eigenmode
contours, stress contours, history of force coefficients, and phase plots were used. The present flow
solver has been validated for shear driven cavity flow and flow over a single circular cylinder at Re
= 100 and 200 before applying it to solve flow past two cylinders.

A. Lid driven cavity flow


Lid driven cavity flow is a classical problem to validate a viscous flow solver. In this problem,
a square cavity is bounded by static walls on three sides. The top wall of the cavity moves with a
constant velocity along the positive x-direction. This flow problem is simulated at Re = 100 with
150 node points located on each walls of the cavity. The total number of cells used to discretise the
domain is 44402. Suitable grid clustering has been used near the walls. The time marched numerical
solution reaches a steady state with residual convergence limit of 10−5 . The x-component of velocity
profile is shown in Fig. 2 which is taken along the vertical line passing through the centre of the

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-4 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 2. x-component of velocity (u) at the mid vertical plane at steady state.

cavity and the y-component of velocity profile is shown in Fig. 3 which is taken along the horizontal
line passing through the centre of the cavity. They compare reasonably well with results reported by
Ghia et al.18

B. Unconfined flow over a single circular cylinder


Accurate computation of flow past a circular cylinder is a challenging problem due to its inherent
intricacies like shear layer instability, vortex shedding, convection and diffusion of vortices in the

FIG. 3. y-component of velocity (v) at the mid horizontal plane at steady state.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-5 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 4. Streamlines for flow past a single circular cylinder at Re = 100.

wake, etc. Prescribing suitable boundary conditions is essential in producing acceptable levels of
accuracy. Thus this problem can be treated as an important validation for any viscous flow solver. In
the present study, unconfined flow past circular cylinder at Re = 100 and 200 has been computed. The
simulations have been carried out with 76776 triangular elements in the flow domain of which 200
elements are taken on the body surface. This grid was chosen after a systematic grid independence
study was conducted with three different grids. The top and bottom boundaries are located at 10 D
distance from cylinder centre on either sides. The inlet is at a distance of 10 D and the outlet at 25
D from the centre of the cylinder. A constant nondimensional time step of 0.0005 has been used for
the computations which satisfies both the CFL and cell Reynolds number based stability criteria for
all the flow field cells. At each time step, the pressure Poisson equation is iteratively solved with
a convergence criteria of 10−5 with the maximum number of iterations limited to 5000. After the
initial transient, the flow develops gradually with increase in the wake length behind the cylinder.
Once the wake length is sufficiently large, a global instability evolves in the cylinder wake which
leads to shedding of vortices behind the cylinder. The shed vortices are convected downstream of
the cylinder and smoothly convect out of the domain. This is facilitated by the convective boundary
condition imposed at the outflow boundary which minimises the generation of wave reflections from
that boundary. Figure 4 shows the streamlines and Fig. 5 shows vorticity contours for Re = 100 at
nondimensional time (t) = 200. The minimum vorticity (ωmin ) level used is −1.0 and maximum
vorticity (ωmax ) level used is 1.0 with a difference of 0.154 between levels (ω). Henceforth this is
represented as (ωmin , ωmax , ω) ≡ (−1.0, 1.0, 0.154).
Alternate shedding of vortices from the cylinder surface leads to periodic oscillation of lift and
drag coefficients as shown in Figures 6 and 7. The results provided in Table I comprise of mean
and fluctuating components of lift and drag coefficients and Strouhal number (St) of the present
simulations and those reported by Braza et al.,19 Meneghini et al.,20 Ding et al.,1 and Harichandan
and Roy.15 The present results compare well with those reported in literature.
Proper orthogonal decomposition of these flow fields was carried out to study the coherent
structures. The method of snapshots outlined by Sirovich21 has been used in the present study. The
velocity field u(x), at a given time t can be expressed as

N
u (x, t) = an (t) ψn (x) , (4)
n=1

where ψ n (x) is the POD eigenmode and an (t) is the corresponding temporal coefficient. The number
of snapshots (N) used to extract the POD modes has a strong influence on the basis function. In order
to ensure that the properties are statistically stationary, a large number of snapshots have been used.
The dependence of the basis function on the number of snapshots is tested by the concept of residual

FIG. 5. Vorticity contours for flow past a single circular cylinder at Re = 100, (ωmin , ωmax , ω) ≡ (−1.0, 1.0, 0.154).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-6 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 6. Time history of lift coefficient for a single circular cylinder at Re = 100.

as given in Eq. (5). The residual concept outlined by Lee et al.22 is used to study the convergence of
the first eigenmode:

 N +1 
η (N ) = ψ (x) − ψ1N (x) d x, (5)
1


where ψ1Nis the first eigenmode calculated with N number of snapshots. The convergence of
the basis function for first eigenmode for streamwise velocity component u is shown in Fig. 8.
Desired convergence is obtained when the number of snapshots is over 400. The same approach

FIG. 7. Time history of drag coefficient for a single circular cylinder at Re = 100.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-7 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

TABLE I. Lift, drag coefficient, and Strouhal number for flow around a single circular cylinder at Re = 100 and 200.

Lift coefficient Drag coefficient Strouhal number


(Cl) (Cd) (St)
Re 100 Re 200 Re 100 Re 200 Re 100 Re 200

Braza et al.19 ± 0.25 ± 0.75 1.366 ± 0.015 1.40 ± 0.05 0.160 0.200
Meneghini et al.20 ... ... 1.37 ± 0.010 1.30 ± 0.05 0.165 0.196
Ding et al.1 ± 0.287 ± 0.659 1.356 ± 0.010 1.38 ± 0.05 0.166 0.196
Harichandan and Roy15 ± 0.278 ± 0.602 1.352 ± 0.010 1.352 ± 0.010 0.161 0.192
Present result ± 0.275 ± 0.652 1.360 ± 0.010 1.42 ± 0.05 0.165 0.198

is followed for other cases to ensure statistically stationary flow properties. Figure 9 shows the
energy content in different POD modes. It is observed that first fifteen modes capture 99.99% of
the energy. The validation of POD computation is carried out by reconstructing the instantaneous
fields by considering the first seven modes which capture 99.9% of energy. The actual data in
Fig. 10 is represented well by the instantaneous reconstructed data shown in Fig. 11, which exhibits
the efficacy of the POD procedure followed in the present study.
The first six POD mode structures obtained from the streamwise velocity component for flow
past single square cylinder at Re = 100 are shown in Fig. 12. The present results have good agreement
with results reported by Borggaard et al.12 and Hay et al.13
Figure 13 shows the first, third, fifth, and seventh modes of the streamwise velocity component
for flow past single circular cylinder at Re = 100. After obtaining satisfactory results for single
square and circular cylinder, flow past two side by side cylinders was computed and POD modes
were analysed. The time variation of the POD coefficients and their phase portraits are shown in
Figures 14 and 15. They depict periodic behaviour of the flow.
Vorticity contours cannot be used to discriminate between swirling and shearing regions in a
flow field. For this purpose, the λ2 criterion proposed by Jeong and Hussain23 is used, where λ2 is the
discriminant for complex eigenvalues of the velocity gradient matrix. For 2D velocity vectors, λ2 is
 2  
given by Vollmers24 as λ2 = ∂∂ux + ∂v ∂y
− 4 ∂u ∂v
∂x ∂y
− ∂u ∂v
∂y ∂x
. In the swirling region, λ2 is negative
whereas it is positive in the shearing zones. Figure 16 distinguishes the shearing and swirling regions
of the flow. The darker regions represent higher swirl and lighter regions represent higher shear. In
a given flow the time varying component φ can be split in to the mean φ, a periodic component φ̃,
and a random component φ  . The present study is confined to laminar flow, therefore the random
component due to turbulent flow can be neglected and thus any time varying component can be
represented as φ = φ + φ̃. Due to the periodicity in the flow, stress arises in the flow field which

FIG. 8. Convergence of the first eigenmode for single circular cylinder.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-8 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 9. Energy content of POD modes of streamwise velocity field for flow past circular cylinder at Re = 100.

FIG. 10. Actual instantaneous data.

FIG. 11. Reconstructed instantaneous data based on the first seven modes.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-9 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 12. First six POD modes of the streamwise velocity field for flow past square cylinder at Re = 100.

was discussed by Saha et al.25 For example, instantaneous streamwise normal stress due to the flow
periodicity can be written as ũ ũ = (u − u)2 . The time mean value (u) is obtained using sufficiently
large time history such that the statistical mean has a convergence of 10−3 . Typical contours of
instantaneous streamwise normal stress for flow past a circular cylinder is shown in Fig. 17. The
minimum stress (ũ ũ min ) level used is 0.01 and maximum stress (ũ ũ max ) level used is 0.2 with a
difference of 0.01 between levels (ũ ũ). Henceforth, this is represented as (ũ ũ min , ũ ũ max , ũ ũ) ≡
(0.01, 0.2, 0.01).

C. Unconfined flow over two side by side circular cylinders


In order to perform a detailed investigation of interference effects between two side by side
cylinders, simulations were performed using the flow domain shown in Fig. 1 by varying the centre
to centre distance between the cylinders (T) from 1.1 to 8 times the diameter of a cylinder. It is

FIG. 13. First, third, fifth, and seventh POD modes of the streamwise velocity field for flow past single circular cylinder at
Re = 100.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-10 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 14. Time dependent POD coefficients for single circular cylinder.

FIG. 15. Phase portraits of POD time dependent coefficients.

FIG. 16. Contours of λ2 for flow past a circular cylinder.

known that at Re approximately and greater than 190, a single circular cylinder wake exhibits three-
dimensional and transitional behavior. Three dimensionality in the wake is expected to occur at a
lower threshold Reynolds number for side by side cylinder configurations because of the enhanced
effective length scale in the cross stream direction and strong interaction effects. Therefore, all

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-11 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 17. Contours of instantaneous streamwise normal stress for flow past a circular cylinder (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01,
0.2, 0.01).

simulations in the present study were confined to Re = 100 so that the flow field could be assumed
to be by and large two-dimensional even for close proximity and strong interaction cases.

1. Single bluff body periodic regime


Figure 18 shows the vorticity contours for T = 1.1 D. Figures 19(a) and 19(b) show a comparison
between the lift and drag time histories for the two cylinders. Time averaged lift coefficient acting
on the upper cylinder is 1.6 and that for lower cylinder is −1.6. The lift coefficient at any instant on
the upper cylinder is positive whereas the lower cylinder is negative. As shown in Fig. 19(c), the lift
and drag acting on the upper cylinder are nearly in-phase, whereas for lower cylinder, they are nearly
anti-phase. The frequency of both lift and drag are identical for both the cylinders. The gap between
the cylinders is so small that the fluid barely reaches 50% of the free stream velocity at the centre of
the gap. Time variation of POD coefficients is shown in Fig. 20. Figure 21 shows the first and third
POD modes. These contours have a close resemblance with those of a single circular cylinder. This
pattern is observed till T = 1.3 D. Figure 22 shows the λ2 contours. It is evident that the shear layer
fuses together in the leading edge of the pair of cylinders and appears similar in nature to that of a
single circular cylinder. Figure 23 shows the instantaneous streamwise normal stress ũ ũ. The region
of concentrated stress on the upper half do not merge or interact with those in the lower half. This
indicates that there is no interaction between the upper and lower vortex streets and their topology
bear close resemblance to that of a single circular cylinder.

2. Aperiodic regime
In this flow regime, the two cylinders cease to shed as a single bluff body. They shed separately,
but the gap between the two cylinders is such that the vortex shedding phenomenon is strongly
coupled and nonlinear in nature. The strong interaction between the shed vortices is exhibited
through the vorticity contours shown in Fig. 24. Due to the lack of periodic behavior in the flow
field, the force coefficients have been monitored for fairly long time of 1000 units which was
sufficient to attain statistical stationarity. The time history of lift and drag coefficients are shown in
Fig. 25. Time variation of POD coefficients is shown in Fig. 26. Figure 27 shows the first and the
third POD modes. The first POD mode shows a strong energy concentration close to the base and

FIG. 18. Vorticity contours for T = 1.1 D, (ωmin , ωmax , ω) ≡ (−2.0, 2.0, 0.166).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-12 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 19. Time history of force coefficients for T = 1.1 D: (a) lift coefficient of both cylinders; (b) drag coefficient of both
cylinders; and (c) superimposed lift and drag coefficients for each cylinder.

gap region of two cylinders and in their near wake. This strong interaction between vortices and the
aperiodic pattern of force coefficients is observed from T = 1.4 D to T = 2.2 D. Figure 28 shows that
the energy captured in the first mode is approximately 50% and the other modes individually capture
less than 10% of the energy. Also it takes more than 100 modes to capture 99.9% of energy. This
indicates that the energy content is distributed across large number of modes. Figure 29 shows the λ2
contours. The upstream half of the gap region between two cylinders exhibits dark regions in the λ2
plot which indicates the initiation of two vortex streets from the respective cylinders. Subsequently,
due to their interaction, an oscillating shear layer is formed in the downstream half of the gap region
and further. Closer observation of the gap region flow dynamics over long time periods confirms that
the oscillating shear layer phenomenon is intermittent and aperiodic. The neighbouring flow field,
including the wake region is perturbed irregularly by the oscillating shear layer. This is likely to be
the reason behind aperiodic behavior of the flow. Consequently, aperiodic forces are experienced by
the cylinders. The instantaneous streamwise normal stress contours shown in Fig. 30 demonstrates
that the stress field in the downstream region loses symmetry and becomes quite disorganised.

3. Anti-phase synchronised regime


This regime is observed for T = 3.2 D to T = 7.9 D. In this regime, the positive vortex of the
upper cylinder and the negative vortex of the lower cylinder are synchronously shed. As a result
of this shedding pattern a set of six vortices forming a hexagonal pattern convect into the wake as

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-13 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 20. Time dependent POD coefficients for T = 1.1 D.

FIG. 21. First and third POD modes for T = 1.1 D obtained from the streamwise velocity field.

FIG. 22. Contours of λ2 for T = 1.1 D.

shown in Fig. 31. There is fairly strong interaction between the vortex streets in the near wake, which
weakens further downstream. The time averaged lift generated by each cylinder is small due to the
comparatively weak interaction. As expected, the upper cylinder generates a small positive lift and
lower cylinder generates a small negative lift in time averaged sense as shown in Fig. 32(a). The drag
acting on the two cylinders as shown in Fig. 32(b) are comparable. Since the shedding is anti-phase,

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-14 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 23. Instantaneous streamwise normal stress contours for T = 1.1 D (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01, 0.2, 0.01).

FIG. 24. Vorticity contours for T = 1.5 D, (ωmin , ωmax , ω) ≡ (−2.0, 2.0, 0.166).

FIG. 25. Time history of force coefficients for T = 1.5 D.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-15 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 26. Time dependent POD coefficients for T = 1.5 D.

FIG. 27. First and third POD modes for T = 1.5 D obtained from the streamwise velocity field.

FIG. 28. Energy content of POD modes of streamwise velocity field for T = 1.5 D.

the lift coefficients of both the cylinders are also found to be anti-phase as shown in Fig. 32(a). As
shown in Fig. 32(c), the lift and drag acting on the upper cylinder are nearly anti-phase, whereas for
lower cylinder, they are found to be nearly in-phase. The frequency of both lift and drag are identical
for both the cylinders. Time variation of POD coefficients is shown in Fig. 33. Figure 34 shows the
first and the third POD modes. The reason for anti-phase synchronised pattern can be well explained

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-16 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 29. Contours of λ2 for T = 1.5 D.

FIG. 30. Contours for instantaneous streamwise normal stress for T = 1.5 D (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01, 0.2, 0.01).

FIG. 31. Vorticity contours for T = 3.5 D, (ωmin , ωmax , ω) ≡ (−2.0, 2.0, 0.166).

from these mode shapes. The mode shapes have close similarity with those of a single circular
cylinder as far as their outer structures are concerned. In the gap region, the modal structures merge.
Skewing of the modal structures occurs in the far wake region. This interference pattern leads to the
anti-phase shedding phenomenon. The λ2 contours also exhibit anti-phase synchronised pattern as
seen in Fig. 35 and shear layer interaction is considerably reduced. The hexagonal pattern is clearly
visible in the λ2 plot. Streamwise normal stress contours show strong interaction and merging of
stress regions in the central downstream portion of the wake as seen in Fig. 36. Considerable affinity
exists between positive and negative vortices in this region. However, inspite of interaction, the
vortex structures do not mix and merge but retain their identity as is evident from the λ2 contours.
There is a close structural similarity between the 1st POD mode and the instantaneous streamwise
normal stress field.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-17 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 32. Time history of force coefficients for T = 3.5 D: (a) lift coefficient of both cylinders; (b) drag coefficient of both
cylinders; and (c) superimposed lift and drag coefficients for each cylinder.

4. In-phase synchronised regime


This regime is observed for T ≥ 8.0 D. In this flow regime, the vortex shedding follows an
in-phase synchronised behavior. In-phase synchronised shedding means that when a vortex is shed
from the upper half of the upper cylinder, simultaneously a vortex is shed from the upper half of
the lower cylinder, as seen in Fig. 37. For the in-phase synchronised regime, the limiting phase
difference has been considered as 20% with regard to the history of lift coefficient. For T = 8.0 D,
the phase difference is found to be marginally less than 20% as seen in Fig. 38(a). It is anticipated that
at larger gaps, the phase difference would decrease monotonically. Phase differences in the range of
20 to 50% (where 50% represents perfect anti-phase) are categorised under anti-phase synchronised
regime. It is evident from Fig. 38(b) that the time averaged drag coefficients for the two cylinders
are close to each other. In Fig. 38(c) it is observed that the harmonic of drag coefficient which occurs
at twice its fundamental frequency is nearly in-phase with the lift for the upper cylinder whereas the
fundamental frequency of drag is nearly in-phase with the lift for the lower cylinder. It is interesting
to observe that the drag harmonic at T = 8.0 D has become considerably stronger in comparison to
its weak manifestation at T = 3.5 D as seen in Fig. 32(c). Figure 39 shows the time variation of POD
coefficients. Figure 40 shows the first and the third POD modes. These mode shapes are similar to
two independent single cylinders. Even at large distance downstream of the cylinders no perceivable
interaction effect is visible through the vorticity contours and POD modes. The λ2 contours and

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-18 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 33. Time dependent POD coefficients for T = 3.5 D.

FIG. 34. First and third POD modes for T = 3.5 D obtained from the streamwise velocity field.

FIG. 35. Contours of λ2 for T = 3.5 D.

the streamwise normal stress contours as shown in Figs. 41 and 42 respectively reflect the same
observation.

5. Transformation regime
There exists a region between the aperiodic regime and anti-phase synchronised regime where
the shedding pattern transforms from anti-phase to in-phase synchronised as shown in Fig. 43 . When

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-19 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 36. Contours for instantaneous streamwise normal stress for T = 3.5 D (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01, 0.2, 0.01).

the centre to centre distance varies between 2.3 D and 3.1 D, the shedding initiates with an anti-phase
shedding pattern but this pattern is highly unstable and it reaches a stable in-phase shedding pattern
within a short time which is a function of the centre to centre distance. As T increases, the anti-phase
shedding pattern gains stability which can be observed from the history of force coefficients of T
= 2.7 D (Fig. 44) and T = 3.1 D (Figures 45(a)–45(d)). At T = 3.1 D, the anti-phase shedding
pattern is weakly unstable and hence it takes fairly long time of the order of 150 nondimensional
units for the transformation. This transformation can be observed from the vorticity contours as
shown in Fig. 43. The lift and drag coefficients also undergo perceivable changes on account of
this transformation. At the end of the transformation, the force coefficients gradually settle down to
the periodic limit cycle behavior as seen from Figures 45(c) and 45(d). On closer examination of
the history of lift coefficients shown in Fig. 45(a), it is observed that the phase difference between
lift coefficients of the two cylinders reduces gradually during the transformation stage. Once the
transformation is complete, they are found to be in-phase (Fig. 45(c)). The opposite behavior is
observed for drag (Figures 45(b) and 45(d)). As shown in Fig. 45(e), the lift and drag acting on
the upper cylinder are nearly anti-phase, whereas for lower cylinder, they are nearly in-phase. The
frequency of both lift and drag are identical for both the cylinders. Unlike the in-phase synchronised
behavior observed beyond T = 8.0 D, the vortices shed from the top and bottom cylinders merge
together. This leads to stretching and skewing of vortices in the wake region. Time variation of POD
coefficients for T = 3.1D is shown in Fig. 46. It is evident from the POD modes that there is virtually
no interaction between modes close to the body but they do interact some distance downstream in
the wake. The first POD mode shown in Fig. 47 exhibits skewing of modal structures at about 8 to
10 cylinder diameters downstream in the wake. The POD refers to the period when the switch over
from anti-phase to in-phase has already occurred and the flow has stabilised in the later phase. The
λ2 contours shown in Fig. 48 exhibit very little shear layer interaction. The instantaneous streamwise

FIG. 37. Vorticity contours for T = 8.0 D, (ωmin , ωmax , ω) ≡ (−2.0, 2.0, 0.166).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-20 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 38. Time history of force coefficients for T = 8.0 D: (a) lift coefficient of both cylinders; (b) drag coefficient of both
cylinders; and (c) superimposed lift and drag coefficients for each cylinder.

normal stress contours shown in Fig. 49 display considerable mixing, merging, and skewing at about
3 D–6 D distance downstream of the cylinders which weakens further downstream.

D. Frequency spectrum analysis


Figure 50 shows the frequency spectrum obtained from the FFT of lift coefficient for flow
past two side by side cylinders placed at different gaps. The spectrum for T = 1.1 D shows the
most dominating peak at St = 0.11 which corresponds to the fundamental frequency. A couple
of harmonics are visible at St = 0.22, 0.33, and 0.44, respectively. The dominating peak gets
substantially weaker in terms of power content at T = 1.5 D. It occurs at a slightly higher frequency
of St = 0.14. A broadband spectrum with large number of weaker frequency peaks disposed on
both sides of the dominant frequency peak are visible in the spectrum. It shows that the energy
content of the flow is distributed over a wide range of frequencies. At T = 3.1 D, the dominant
frequency attains a still higher value of St = 0.18 with higher harmonics visible at St = 0.36 and
0.54. The dominant frequency regains its sharp structure at this gap. At T = 3.5 D, the dominant
frequency remains at St = 0.18 with a harmonic visible at St = 0.36. At T = 8.0 D, the dominant
peak drops marginally to St = 0.17 with weaker harmonics visible at St = 0.34 and 0.51. In terms
of power content, the strongest dominant frequency peak is found at T = 1.1 D and the weakest at T

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-21 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 39. Time dependent POD coefficients for T = 8.0 D.

FIG. 40. First and Third POD modes for T = 8.0 D obtained from the streamwise velocity field.

FIG. 41. Contours of λ2 for T = 8.0 D.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-22 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 42. Contours for instantaneous streamwise normal stress for T = 8.0 D (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01, 0.2, 0.01).

FIG. 43. Vorticity contours for T = 3.1 D, (ωmin , ωmax , ω) ≡ (−2.0, 2.0, 0.166).

FIG. 44. Time history of drag coefficients for T = 2.7 D.

= 1.5 D. The power content of dominant peak for T = 3.1 D, 3.5 D, and 8.0 D are comparable. As
expected, the power content of the dominant frequency peak correlates strongly with the existence of
periodicity in the flow field. Consequently a weak frequency peak and broadband frequency spectrum
is observed in the aperiodic regime while distinct dominant frequency peaks are observed in the other
regimes.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-23 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 45. Time history of force coefficients for T = 3.1 D: (a) lift coefficient of both cylinders, 20<t<150; (b) drag
coefficient of both cylinders, 20<t<150; (c) lift coefficient of both cylinders, 200<t<260; (d) drag coefficient of both
cylinders, 200<t<260; and (e) superimposed lift and drag coefficients for each cylinder.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-24 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 46. Time dependent POD coefficients for T = 3.1 D.

FIG. 47. First and third POD modes for T = 3.1 D obtained from the streamwise velocity field.

FIG. 48. Contours of λ2 for T = 3.1 D.

E. Phase portrait
Phase portraits are obtained by monitoring u and v values at three spatial locations in the near
wake and gap regions. Station 1 (S1) is located at 2 D distance downstream from the centre of
the top cylinder, station 2 (S2) is the mid point of the line joining the centres of the two cylinders
and station 3 (S3) is located at 2 D distance downstream of station 2. Figures 51(a)–51(c) are the
phase portraits for T = 1.1 D at stations 1, 2, and 3, respectively. It is clear from Figures 51(a)

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-25 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 49. Contours for instantaneous streamwise normal stress for T = 3.1 D (ũ ũ min , ũ ũ max , ũ ũ) ≡ (0.01, 0.16, 0.01).

FIG. 50. FFT of the history of lift coefficient: (a) T = 1.1 D, (b) T = 1.5 D, (c) T = 3.1 D, (d) T = 3.5 D, (e) T = 8.0 D.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-26 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

FIG. 51. Phase portraits for different values of T at three different locations S1, S2, and S3: (a) T = 1.1 D, at S1; (b) T =
1.1 D, at S2; (c) T = 1.1 D, at S3; (d) T = 1.5 D, at S1; (e) T = 1.5 D, at S2; (f) T = 1.5 D, at S3; (g) T = 3.1 D, at S1,
40<t<200; (h) T = 3.1 D, at S2, 40<t<200; (i) T = 3.1 D, at S3, 40<t<200; (j) T = 3.1 D, at S1, 200<t<400; (k) T = 3.1 D,
at S2, 200<t<400; (l) T = 3.1 D, at S3, 200<t<400; (m) T = 3.5 D, at S1; (n) T = 3.5 D, at S2; (o) T = 3.5 D, at S3; (p) T
= 8.0 D, at S1; (q) T = 8.0 D, at S2; and (r) T = 8.0 D, at S3.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-27 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

and 51(c) that the change in v component of velocity is very high, i.e., it varies between +1.2
and −1.2. This is a significantly large range when compared to other cases. At station 2, for T =
1.1 D, the flow is aligned in the direction of the inflow and the flow is restricted such that the u
velocity component attains only about 50% to 60% of the free stream. This is because the shear
layers between the cylinders occupy the entire gap and therefore constrain the gap flow. This also
suppresses the generation of vortices from cylinder surfaces in the gap region. Figures 51(d)–51(f)
do not exhibit any closed loop because of the aperiodicity in the flow corresponding to T = 1.5
D. As shown earlier in Figures 45(a) and 45(b), for T = 3.1 D, the history of lift and drag coeffi-
cients keeps changing till t ≈ 200 is reached. Figures 51(g)–51(i) also show that large number of
loops are formed in the phase space and the loops finally get attracted towards a trajectory which
resembles that of a single cylinder. After approximately 200 nondimensional time steps, the flow
stabilises to a periodic one and subsequently it executes regular loops in the phase space, as seen from
Figures 51(j)–51(l). Once the flow stabilizes, at station 2, the value of u remains almost constant,
however the v component oscillates weakly about zero mean. On further increase in T, the in-
teraction between vortices and shear layers produced by the two cylinders weakens but still the
velocity components exhibit some variations which indicates the mutual interference as shown in
Figures 51(m)–51(o) for T = 3.5 D. Figures 51(p)–51(r) show the phase diagrams for T = 8.0 D
where increasing symmetry is observed in all the three cases which indicates that there is insignifi-
cant interference between the two cylinders. From Figures 51(h), 51(k), and 51(n), it is evident that
the flow is accelerated in the gap between the cylinders because of venturi effect. The u component
of velocity in the gap varies from an approximate average value of 22% to 40% above freestream as
T increases from 1.5 D to 3.5 D. In Fig. 51(q), the flow is 9% to 10% higher than the free stream
value. On further increase in gap, the velocity asymptotically approaches that of the free stream
value.

V. CONCLUSIONS
A finite volume incompressible Navier-Stokes solver has been developed based on a second
order spatio-temporal accurate scheme and it is applied to flow problems on unstructured grids. The
scheme has been validated by computing lid driven cavity flow and flow over single circular cylinder
at Re = 100 and 200. Further the scheme is used to study and identify the different flow regimes
observed for flow past two side by side circular cylinders for centre to centre distance varying from
1.1 D to 8.0 D. Five different flow regimes have been identified in the present study. The flow
is analysed using POD, λ2 criterion, instantaneous streamwise normal stress, frequency spectrum
analysis, and phase-space representations. All these tools confirm the existence of the various flow
regimes. Detailed analysis of the flow reveals that when the centre to centre distance is varied between
1.1 D and 1.3 D, flow past the pair of cylinders behaves like a single bluff body flow with augmented
cross flow dimension. For centre to centre distance varying between 1.4 D and 2.2 D, there is a
concentration of the most energetic POD mode structures in the gap and near wake region of the
cylinders. The force coefficients and phase portraits show uncorrelated data which depicts aperiodic
behavior. In the 2.3 D to 3.1 D centre to centre gap range, the concentration of energetic mode
structure close to the cylinder weakens and transformation from anti-phase to in-phase synchronised
shedding takes place. Time required to reach a periodic limit cycle behavior of the force coefficients
increases substantially. In 3.2 D-7.9 D centre to centre gap range, the energy concentration in the
gap region and near wake region further weakens giving rise to POD modes which are topologically
similar to those of a single circular cylinder. Anti-phase vortex shedding is observed in this gap
range. On further increase of the gap the vortex shedding behavior of each cylinder asymptotically
approaches that of a single cylinder. In periodic regimes, the lift and drag coefficients of the upper
cylinder are nearly anti-phase except in single bluff body periodic regime because this is the only
regime where the cylinder sheds purely from one half and gap flow is virtually inhibited. For the
lower cylinder, the force coefficients are nearly in-phase except for the single bluff body periodic
regime for the same reason as stated earlier for the upper cylinder. An interesting extension of the
present work would be to reconstruct reduced order models of the various flow regimes identified in
the study.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
063602-28 K. Supradeepan and A. Roy Phys. Fluids 26, 063602 (2014)

ACKNOWLEDGMENTS
This work has been supported by the Associate Node for CFD at Indian Institute of Technol-
ogy, Kharagpur funded by Aeronautical Research and Development Board, Ministry of Defence,
Government of India.
1 H. Ding, C. Shu, K. S. Yeo, and D. Xu, “Numerical simulation of flows around two circular cylinders by mesh-free least
square-based finite difference methods,” Int. J. Numer. Methods Fluids 53, 305–332 (2007).
2 P. W. Bearman and A. J. Wadcock, “The interaction between a pair of circular cylinders normal to a stream,” J. Fluid Mech.

61, 499–511 (1973).


3 K.-S. Chang and C.-J. Song, “Interactive vortex shedding from a pair of circular cylinders in a transverse arrangement,”

Int. J. Numer. Methods Fluids 11(3), 317–329 (1990).


4 O. Inoue, W. Iwakami, and N. Hatakeyama, “Aeolian tones radiated from flow past two square cylinders in a side-by-side

arrangement,” Phys. Fluids 18(4), 046104 (2006).


5 S. Kang, “Characteristics of flow over two circular cylinders in a side-by-side arrangement at low Reynolds numbers,”

Phys. Fluids 15(9), 2486–2498 (2003).


6 K. Liu, D. J. Ma, D. J. Sun, and X. Y. Yin, “Wake patterns of flow past a pair of circular cylinders in side-by-side

arrangements at low Reynolds numbers,” J. Hydrodyn., Ser. B 19(6), 690–697 (2007).


7 D. Sumner, M. D. Richards, and O. O. Akosile, “Two staggered circular cylinders of equal diameter in cross-flow,” J.

Fluids Struct. 20(2), 255–276 (2005).


8 C. H. K. Williamson, “Evolution of a single wake behind a pair of bluff bodies,” J. Fluid Mech. 159, 1–18 (1985).
9 S. J. Xu, Y. Zhou, and R. M. C. So, “Reynolds number effects on the flow structure behind two side-by-side cylinders,”

Phys. Fluids 15(5), 1214–1219 (2003).


10 D.-H. Yoon and K.-S. Yang, “Characterization of flow pattern past two spheres in proximity,” Phys. Fluids 21(7), 073603

(2009).
11 M. M. Zdravkovich, “Review review of flow interference between two circular cylinders in various arrangements,” J. Fluids

Eng. 99(4), 618–633 (1977).


12 J. Borggaard, A. Hay, and D. Pelletier, “Interval-based reduced-order models for unsteady fluid flow,” Int. J. Numer. Anal.

Model. 4, 353–368 (2007).


13 A. Hay, J. T. Borggaard, and D. Pelletier, “Local improvements to reduced-order models using sensitivity analysis of the

proper orthogonal decomposition,” J. Fluid Mech. 629(1), 41–72 (2009).


14 T. K. Sengupta, V. K. Suman, and N. Singh, “Solving Navier-Stokes equation for flow past cylinders using single-block

structured and overset grids,” J. Comput. Phys. 229(1), 178–199 (2010).


15 A. B. Harichandan and A. Roy, “Numerical investigation of low Reynolds number flow past two and three circular cylinders

using unstructured grid CFR scheme,” Int. J. Heat Fluid Flow 31(2), 154–171 (2010).
16 A. Roy and G. Bandyopadhyay, “A finite volume method for viscous incompressible flows using a consistent flux

reconstruction scheme,” Int. J. Numer. Methods Fluids 52(3), 297–319 (2006).


17 D. J. Mavriplis, “Revisiting the least-squares procedure for gradient reconstruction on unstructured meshes,” AIAA Paper

2003-3986, 2003.
18 U. Ghia, K. N. Ghia, and C. T. Shin, “High-Re solutions for incompressible flow using the Navier-Stokes equations and a

multigrid method,” J. Comput. Phys. 48(3), 387–411 (1982).


19 M. Braza, P. Chassaing, and H. Ha Minh, “Numerical study and physical analysis of the pressure and velocity fields in the

near wake of a circular cylinder,” J. Fluid Mech. 165(3), 79–130 (1986).


20 J. R. Meneghini, F. Saltara, C. L. R. Siqueira, and J. A. Ferrari, Jr., “Numerical simulation of flow interference between

two circular cylinders in tandem and side-by-side arrangements,” J. Fluids Struct. 15(2), 327–350 (2001).
21 L. Sirovich, “Turbulence and the dynamics of coherent structures. I - Coherent structures. II - Symmetries and transforma-

tions. III - Dynamics and scaling,” Q. Appl. Math. 45, 561–571 (1987).
22 S. B. Lee, W. Kang, and H. J. Sung, “Organized self-sustained oscillations of turbulent flows over an open cavity,” AIAA

J. 46(11), 2848–2856 (2008).


23 J. Jeong and F. Hussain, “On the identification of a vortex,” J. Fluid Mech. 285(2), 69–94 (1995).
24 H. Vollmers, “Detection of vortices and quantitative evaluation of their main parameters from experimental velocity data,”

Meas. Sci. Technol. 12(8), 1199–1207 (2001).


25 A. K. Saha, K. Muralidhar, and G. Biswas, “Vortex structures and kinetic energy budget in two-dimensional flow past a

square cylinder,” Comput. Fluids 29(6), 669–694 (2000).

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
203.110.243.21 On: Fri, 20 Jun 2014 17:40:51
View publication stats

You might also like