Dependence of square cylinder wake on Reynolds number

Honglei Bai, and Md. Mahbub Alam

Citation: Physics of Fluids 30, 015102 (2018);

View online: https://doi.org/10.1063/1.4996945
View Table of Contents: http://aip.scitation.org/toc/phf/30/1
Published by the American Institute of Physics
 PHYSICS OF FLUIDS 30, 015102 (2018)

Dependence of square cylinder wake on Reynolds number

Honglei Bai1,a) and Md. Mahbub Alam2,b)
1 Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong, China
2 Institute for Turbulence-Noise-Vibration Interaction and Control, Shenzhen Graduate School, Harbin Institute

of Technology, Shenzhen, China and Digital Engineering Laboratory of Offshore Equipment, Shenzhen, China
(Received 20 July 2017; accepted 14 December 2017; published online 5 January 2018)

The wake of a square cylinder is investigated for Reynolds number Re < 107 . Two-dimensional
(2D) laminar simulation and three-dimensional (3D) large-eddy simulation are conducted at
Re ≤ 1.0 × 103 , while experiments of hotwire, particle image velocimetry, and force measurements
are carried out at a higher Re range of 1.0 × 103 < Re < 4.5 × 104 . Furthermore, data covering a
wide Re range, from 100 to 107 , in the literature are comprehensively collected for discussion and
comparison purposes. The dependence on Re of the recirculation bubble size or vortex formation
length, wake width, shear-layer transition, time-mean drag force, and Strouhal number is discussed
in detail, revealing five flow regimes, each having distinct variations of the above parameters. With
increasing Re, while the streamwise recirculation size enlarges at Re < 50 (steady flow regime), the
vortex formation length reduces at 50 < Re < 1.6 × 102 (laminar flow regime), remains unchanged
at 1.6 × 102 < Re < 2.2 × 102 (2D-to-3D transition flow regime), and decreases at 2.2 × 102 < Re
< 1 × 103 (shear layer transition I regime), approaching asymptotically a constant at Re > 1.0 × 103
(shear layer transition II regime). Meanwhile, the wake width decreases with Re in the laminar flow
regime, grows in 2D-to-3D transition and shear layer transition I regimes, and levels off in the shear
layer transition II regime. The narrowest wake width is identified in the 2D-to-3D transition flow
regime, corresponding to a minimum time-mean drag force and a largest Strouhal number. With
increasing Re, the shear-layer transition length rapidly declines in the shear layer transition I regime
where the transition occurs downstream of the trailing corner of the cylinder. On the other hand, it
slowly tapers off in the shear layer transition II regime where the transition takes place upstream of
the trailing corner. An extensive comparison is made between the dependence on Re of a circular
cylinder wake and a square cylinder wake, with their distinct natures highlighted. Published by AIP
Publishing. https://doi.org/10.1063/1.4996945

I. INTRODUCTION around the square cylinder has received less attention in the
literature.1,6
Bluff-bodies are widely seen in practical engineering
With increasing Reynolds number Re (=U ∞ D/ν, where
applications such as high-rise buildings, bridge piers, heat
U ∞ is the incoming free stream velocity, D is the cylinder
exchangers, and offshore platforms. The flow past a bluff
diameter or width, and ν denotes the kinematic viscosity of
body involves boundary-layer development, flow separation,
fluid), the wake of a circular cylinder behaves distinctively due
shear-layer instability and transition, vortex shedding, vortex
to occurrences of different instabilities, change in the flow sep-
dynamics, and so on.1 Due to the relative simplicity and preva-
aration, shift in transition position, etc. Consequently, changes
lent occurrence in engineering applications, a slender cylinder
crop up in time-mean and fluctuating fluid forces, shear layer
with either circular or square cross section has been consid-
characteristics, vortex dynamics, and wake structures.1,7–9 At
ered as the basic and representative model for the bluff-bodies.
Re < ∼50, the flow around a circular cylinder is laminar
The former is featured by its cross section with a continuous
and steady, with a pair of recirculation bubbles symmetrically
and finite curvature, where the flow separation point is oscil-
generated directly downstream of the cylinder. Laminar vor-
latory and may occur over a segment of the surface. On the
tex shedding is detected at 50 < Re < 200, yielding a von
contrary, the latter has a cross section with sharp corners of an
Kármán vortex street of alternating opposite-sign vortices.
infinite curvature, where the flow separation location is station-
With an increase in Re in this regime, the vortex formation
ary. The difference in the nature of the flow separation results
length drops, and the Strouhal number (St = fs D/U ∞ , where fs
in significantly distinct features of the flow structures and fluid
is the vortex shedding frequency) increases. The cylinder wake
forces between the circular and square cylinder cases.2–5 How-
changes from two-dimensional (2D) to three-dimensional (3D)
ever, compared with that around the circular cylinder, the flow
in a transition Re ≈ 200–260. The three-dimensionality in
the wake, together with the formation length and St, is fur-
a) Email: ther enhanced as Re increases from 260 to 103 . At 1.0 ×
[email protected] and hl [email protected]
b) Author to whom correspondence should be addressed: [email protected] and 103 < Re < 2.0 × 105 , the transition to turbulence in the
[email protected] shear layer occurs, resulting in reductions in the formation

1070-6631/2018/30(1)/015102/19/$30.00 30, 015102-1 Published by AIP Publishing.

015102-2 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

length and St. Although different features in the cylinder time-averaged characteristics, downstream evolutions, and
wake are highlighted, the time-mean drag is more or less Reynolds stress distribution. St declines gradually at Rec2 <
constant in this Re range.10 A critical regime is identified Re < 8.0 × 103 and then levels off with increasing Re.21,23,29
at 2.0 × 105 < Re < 4.0 × 105 where the transition in the Fluid forces acting on the square cylinder largely vary
shear layer approaches the separation, the time-mean drag when Re is changed, in view of the internal coherence between
plunges, St jumps, and separation bubble forms on the cylin- the flow structures and fluid forces. In the laminar regime (i.e.,
der surface. The separation bubble bursts, and the transition at Re < Rec1 ), time-mean drag coefficient C d exponentially
to turbulence occurs in the boundary layer at Re > 4.0 × 105 , declines with Re.13,15,19 In the turbulent regime (Re > Rec2 ),
i.e., post-critical regime. The characteristics of the flow at dif- C d however progressively increases at Rec2 < Re < 2 × 104 ,20
ferent Re ranges are very crucial. Without this information, it asymptotically approaching a constant at Re > 2.0 × 104 .29
is very difficult to analyze and understand the wake of two or Meanwhile, fluctuating lift force can be measured as the flow
more cylinders.11,12 becomes unsteady at Re ≥ 50, increasing with Re at 50 < Re
Similar to that of the circular cylinder, the wake struc- < 2.0 × 104 15,29 and remaining almost constant at Re > 2.0 ×
ture of a square cylinder exhibits distinct characteristics with 104 .29 Fluctuating drag force exhibits similar behavior to that
increasing Re. When Re > 50,13 the flow around the square of the fluctuating lift.29
cylinder is unsteady, characterized by alternating vortex shed- While Re-dependent intrinsic features of a circular cylin-
ding and a von Kármán vortex street. The unsteady flow der wake are well documented, there has been no systematic
separates from the trailing corners of the square cylinder at investigation on those of a square cylinder, including vor-
Re < 120 but from the leading corners at Re > 120.14,15 The tex formation length, wake width, and shear-layer transition
flow remains laminar as Re is lower than a critical Reynolds length, as well as on the correlation between these parame-
number Rec1 = 150–200 depending on the freestream turbu- ters and the fluid forces. This work aims to investigate the near
lence intensity, blockage ratio, cylinder aspect ratio, etc.14,16–18 wake of a square cylinder for Re < 107 , based on our numerical
At this Rec1 , the so-called mode A instability and the transition simulations and experimental measurements (at Re = 60–4.5
from 2D to 3D flows occur.18 Mode A is featured by spanwise × 104 ) as well as published data from the literature. Due to the
vortex dislocation, streamwise vortex loops, and a relatively limitation of the wind tunnel, the numerical simulation is done
large spanwise wavelength (around 5.2D) of the streamwise for Re = 60–103 . Typical flow structures and wake character-
vortices.17,18 St increases with Re (<Rec1 ).19–21 With a further istic parameters of the square cylinder are discussed in detail
increase in Re, the so-called mode B instability is identified and compared extensively with those of the circular cylinder.
at another critical Reynolds number Rec2 = 190–250,14,16–18 Sections II and III present the details of numerical simula-
which is again dependent on the factors such as turbu- tions and experimental measurements, respectively. Results
lence intensity, blockage, and aspect ratio. A reduced St was and discussion are given in Sec. IV. This work is summarized
observed in the transition regime at Rec1 < Re < Rec2 . In in Sec. V.
mode B, the spanwise wavelength of the streamwise vortices
is much smaller, around 1.2D, than that in mode A.17,18 In II. NUMERICAL SIMULATIONS
addition, Robichaux et al.14 based on the Floquet instability
A. Laminar simulation
analysis observed a third mode S at Re ≈ 200. Mode S has a
spanwise wavelength of around 2.8D and a shedding period Two-dimensional (2D) simulations are conducted for the
twice that of the base flow. unsteady laminar flow around a square cylinder at Re = 60,
The flow around the square cylinder becomes turbulent at 100, and 120. The 2D incompressible Navier-Stokes (N-S)
Re > Rec2 and has been extensively investigated both experi- equations are solved on structural quadrilateral grids using the
mentally and numerically. The behavior of the separated shear finite-volume method (FVM) in ANSYS Fluent. The conti-
layers and ensuing recirculation region above the side sur- nuity and N-S equations governing the flow can be written,
face of the square cylinder was studied by Lyn and Rodi respectively, in the Einstein convention as follows:
(Re = 2.14 × 104 ),22 Brun et al. (Re = 2.0 × 104 –3.0 × 105 ),23 ∂ui
and Minguez et al. (Re = 2.14 × 104 )24 using laser-Doppler = 0,
∂xi
velocimetry; by Brun et al. (Re = 5.0 × 102 –2.0 × 103 ),23
Minguez et al. (Re = 2.14 × 104 ),24 and Cao and Tamura ∂ui ∂ui uj 1 ∂p ∂ 2 ui
+ =− +ν , i ∈ {1, 2},
(Re = 2.2 × 104 )25 using large eddy simulation (LES); and by ∂t ∂xj ρ ∂xi ∂xj ∂xj
Trias et al. (Re = 2.2 × 104 )26 using direct numerical simulation where ui are the velocity components along the corresponding
(DNS). Both the large-scale von Kármán vortex street in the Cartesian coordinates x i (here, x 1 and x 2 denote the streamwise
near wake and small-scale Kevin-Helmholtz vortical structures x- and cross-stream y-directions, respectively), p is the fluid
in the separating shear layers were well captured in these inves- pressure, ρ is the fluid density, and ν is the kinematic viscosity
tigations, either based on flow visualizations or energy spectra of the fluid.
of point-wise velocities. Brun et al.23 and Minguez et al.24 The computational domain is a rectangle with the size of
also observed the occurrence of Kevin-Helmholtz vortex pair- (Lxu + Lxd ) × Ly = (12D + 24D) × 24D [cf. Fig. 1(a) and Table I],
ing in the shear layer from the leading corners. Furthermore, where Lxu and Lxd are the distance between the cylinder center
Lyn et al. (Re = 2.14 × 104 ),27 Saha et al. (Re = 8.7 × 103 and 1.8 and the upstream inlet and downstream outlet, respectively, L y
× 104 ),28 and Sohankar (Re = 1.0 × 103 –5.0 × 106 )29 inves- is the lateral distance, and D is the cylinder width. The domain
tigated the near-wake structures, including their phase- and parameters are similar to those used by Sohankar et al.19
015102-3 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 1. Computational domain: (a) x-y plane view and (b) perspective view. Flow is from left to right in (a) and along the x-direction in (b). The cylinder
width is D.

and larger than those used by Sharma and Eswaran.30 See N-S equations are solved on structural hexahedral grids using
also the work of Zheng and Alam31 for the effect of the the FVM in ANSYS Fluent. The filtered continuity and N-S
computational domain on the cylinder wake. As shown in equations are given as follows:
Fig. 1(a), quadrilateral meshes are generated in the compu- ∂ ūi
tational domain. With the first grid nearest setting 0.01D away = 0,
∂xi
from the cylinder surface, the grids are stretched progressively
∂ ūi ∂ ūi ūj 1 ∂ p̄ ∂ 2 ūi ∂τij
with a factor of 1.02 along both the x- and y-directions. The + =− +ν − , i ∈ {1, 2, 3},
total number of the grids is 45 600. A uniform streamwise ∂t ∂xj ρ ∂xi ∂xj ∂xj ∂xj
velocity (i.e., U ∞ ) is imposed at the upstream inlet bound- where ūi are the filtered velocity components along x i (here,
ary, while a pressure-outlet boundary condition is used at the x 1 , x 2 , and x 3 denote the streamwise x-, cross-stream y- and
downstream outlet boundary. The no-slip condition is applied spanwise z-directions, respectively), p̄ is the filtered pressure,
to the cylinder surface. The lateral sides are treated as slip sides and τ ij are the subgrid-scale stresses defined as τij = ui uj − ūi ūj .
using symmetric conditions. The Smagorinsky-Lilly model is used presently, q with the tur-
The velocity-pressure coupling in the governing equations
bulent eddy viscosity modeled as νt = (Cs ∆)2 2S̄ij S̄ij , where
is based on the algorithm of semi-implicit method for pressure
∂ūj
 
linked equations-consistent (SIMPLEC).32 The second-order C s is a constant, ∆ is the cell volume, and S̄ij = 21 ∂ūi
∂xj + ∂xi
and second-order upwind differencing schemes are applied is the rate-of-strain tensor. C s = 0.1 for all the present simula-
for the spatial discretization of pressure and momentum, tions, as chosen for circular and square cylinder near-wakes,
respectively. The temporal discretization is done using the e.g., by Lin et al.33,34
second-order implicit differencing scheme. The computation The computational domain (cf. Fig. 1) is chosen to be
is advanced by a non-dimensional time step ∆t ∗ = ∆tU ∞ /D (Lxu + Lxd ) × Ly × Lz = (12D + 24D) × 24D × 10D for the
= 0.01, yielding Courant-Friedrichs-Lewy (CFL) numbers < cases of Re = 1.6 × 102 –2.6 × 102 , where the flow around a
1.33 In this paper, the superscript “*” denotes normaliza- square cylinder is transitional and has a relatively large span-
tion by D and/or U ∞ , without otherwise specified. To cal- wise wavelength,14,16–18,20 and (Lxu + Lxd ) × Ly × Lz = (8D +
culate the statistics such as time-mean and root-mean-square 16D) × 16D × 4D for the cases of Re = 5.0 × 102 and 1.0
(rms) values of the drag force, a sampling time of 20 vortex- × 103 .29,34 The grid distribution in the x-y plane is similar to
shedding periods is adopted after the flow becomes statistically that of the 2D simulation [Fig. 1(a)], while it is uniform along
stationary. the spanwise z-direction [Fig. 1(b)]. The nearest grid to the
B. Large-eddy simulation cylinder surface is 0.01D away, ensuring Y+ ≈ 1 as required
by the computation of turbulent flows. The grids are refined
Three-dimensional (3D) large-eddy simulations (LES) are with an increase in Re. The grid numbers are 45 600 × 72
conducted at Re ≥ 1.6 × 102 (Table I). The three-dimensional and 62 000 × 48 for the cases of Re = 1.6 × 102 –2.6 × 102
and Re = 5.0 × 102 –1.0 × 103 , respectively (Table I). The
TABLE I. Details of the numerical simulations.
average mesh size is 0.0189D in the xy-plane and 0.1389D
Computational domain Grid in the spanwise z-direction at the lower Re range (Re = 1.6
Re (Lxu +Lxd )×Ly ×Lz size Model × 102 –2.6 × 102 ), while it is 0.0062D in the xy-plane and
0.0833D in the spanwise z-direction at the higher Re range
60, 100, 120 (12D + 24D) × 24D 45 600 2D laminar
160, 175, 182,
(Re = 5.0 × 102 and 103 ). The grid resolution [(0.0833–
(12D + 24D) × 24D × 10D 45 600 × 72 3D LES 0.1389)D] along the spanwise z-direction is fine enough to
190, 220, 260
500, 1000 (8D + 16D) × 16D × 4D 62 000 × 48 3D LES capture the unsteady behavior in the spanwise z-direction as
recommended by Mankbadi and Georgiadis.35 The pressure
015102-4 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

outlet condition is used at the outlet boundary, and the no- TABLE II. Time-mean drag coefficient C d and Strouhal number St of a
slip condition is used over the cylinder surface. Symmetric square cylinder.
conditions are used on the lateral sides, while the periodic
Re Investigations St Cd Remark
condition is used along the spanwise direction. The initial
condition used in our 3D LES is the imposition of a uni- 100 Present 0.146 1.48 2D laminar
form flow U ∞ with a turbulent intensity of 0.1%. The com- Sohankar et al.19 0.135-0.15 1.39-1.5 2D laminar
putational domain [(Lxu + Lxd ) × Ly × Lz = (12D + 24D) Yoon et al.15 ... 1.43 2D laminar
× 24D × 10D] used in the present simulations is larger Sharma and Eswaran30 0.15 1.49 2D laminar
than that [(8.5D + 12.5D) × 18D × 6D and (6D + 18D) 120 Present 0.153 1.46 2D laminar
Sharma and Eswaran30 0.155 1.47 2D laminar
× 10D × 6D] used in the previous DNS work.16,20 Saha
Yoon et al.15 ... 1.4 2D laminar
et al.16 and Sohankar et al.20 examined the effect of L z on
160 Present 0.162 1.43 3D LES
the flow, varying L z from 6D to 10D. The effect was found to Sharma and Eswaran30 0.16 1.47 2D laminar
be negligible. Therefore, L z = 10D in our LES simulation is Luo et al.18 0.15 ... Exp.
long enough to capture the mode A instability. 175 Present 0.15 (0.161)a 1.37 (1.44)a 3D LES
The Kolmogorov length scale lυ = (ν 3 /ε)1/4 , where ν is the Saha et al.16 0.159 1.53 3D DNS
kinematic viscosity of the fluid and ε is the energy dissipation 182 Present 0.152 (0.161)a 1.38 (1.44)a 3D LES
rate defined by ε = urms
3 /L, where L is the characteristic length Luo et al.18 0.149 ... Exp.
scale in the cylinder wake.36 l v = (0.0078–0.0532)D, given the 190 Present 0.154 (0.159)a 1.38 (1.44)a 3D LES
maximum urms = (0.4–0.65)U ∞ and vortex formation length Luo et al.18 0.152 ... Exp.
L f = (2.1–1.0)D = L at the Re range of 1.6 × 102 –103 . On the 220 Present 0.157 1.42 3D LES
Luo et al.18 0.162 ... Exp.
other hand, the average mesh size is 0.0062D for Re = 5 × 102
260 Present 0.158 1.44 3D LES
and 103 and 0.0189D for Re = 1.6 × 102 –2.6 × 102 in the
Luo et al.18 0.164 ... Exp.
xy-plane, and 0.0833D for Re = 5 × 102 and 103 and 0.1389D 500 Present 0.136 1.94 3D LES
for Re = 1.6 × 102 –2.6 × 102 along the spanwise z-direction. Sohankar et al.20 0.126 1.87 3D DNS
Therefore, the present mesh size can capture about 10 times the 1000 Present 0.125 2.21 3D LES
Kolmogorov length scale. The subgrid modeling in the present Okajima21 0.123 ... Exp.
LES is correctly made for the small scales in the near wake of
In the transition flow, Re = 175-190, two values are given, corresponding to the laminar
the square cylinder. (–) and turbulent flows, respectively.
The algorithm of pressure implicit splitting of operators
(PISO) is adopted for the velocity-pressure coupling. The
second-order and bounded central differencing schemes are
III. EXPERIMENTAL TESTS
applied to the spatial discretization of pressure and momentum,
respectively, while the bounded second-order implicit differ- Experiments of particle image velocimetry (PIV),
encing scheme for the temporal discretization. The time step hotwire, and force measurements are conducted in an open-
is ∆t ∗ = 0.01, yielding sufficiently small CFL numbers <1 in loop wind tunnel, which has a test section of 0.3 m × 0.3 m ×
the computational domain. The sampling time of 20 periods of 1.5 m (width × height × length). The square cylinder model,
vortex shedding is used for the calculation of statistics. More made of aluminum, spans horizontally across the entire width
details on the LES of bluff-body near-wakes can be found in, of the test section. Three cylinder models of different widths
e.g., the work of Lin et al.33,34 Since the details of the numerical (i.e., D = 5 mm, 15 mm, and 25 mm) are used. The cylinders
aspects in the present work are similar to those in our previous of D = 5 mm and 15 mm are used for Re < 3.0 × 104 , yielding
work on the square cylinder wake,31,34 the convergence analy- the minimum aspect ratio of 20 and the maximum blockage
sis of the numerical simulations was not repeatedly conducted of 5% in the test section. In order to enlarge Re to 4.45 × 104 ,
in the present work. As shown in Table II, the time-mean drag the cylinder of D = 25 mm was employed, corresponding to
coefficient (C d ) and Strouhal number (St) values at different aspect and blockage ratios of 12% and 8.3%, respectively. No
Re from the present work are in good agreement with those blockage correction is performed for the obtained results. The
from the literature. For example, C d = 1.46 and 1.94 at Re = oncoming velocity is changed from 3.6 m/s to 24 m/s in the
120 and 500, respectively, from the present work are compara- tests, corresponding to Re = 1.4 × 103 –4.45 × 104 . The turbu-
ble to those from the literature, i.e., C d = 1.4–1.47 at Re = 120 lent intensity for this range of velocity was less than 0.5% in
from the work of Yoon et al.15 and Sharma and Eswaran30 and the absence of the test model.
C d = 1.87 at Re = 500 from the work of Sohakar et al.20 The A Dantec PIV system (maximum triggering frequency is
present St = 0.153 (Re = 120) and 0.125 (Re = 103 ) from simu- 727 Hz for double frames) is deployed to measure the instanta-
lations are very close to St = 0.155 (Re = 120) from the work of neous velocities in the xy-plane in the near wake of the square
Sharma and Eswaran30 and 0.123 (Re = 103 ) from the work of cylinder. Flow is seeded uniformly with olive oil particles
Okajima.21 Furthermore, our previous LES work 34 on the near of about 1 µm in diameter, generated from a smoke gener-
wake of a square cylinder was conducted at a relatively high ator (TSI 9307-6). The flow is illuminated by two overlapping
Re = 2.2 × 104 , overlapping the present experimental tests. laser sheets from pulse lasers (Litron LDY 304-PIV, Nd: YLF).
C d and St from the work of Lin et al.34 are 2.16 and 0.132, Each laser pulse lasts 10 ns and has a maximum energy out-
respectively, in good agreement with those (C d = 2.14 and put of 30 mJ. The repeating rate of the lasers is 300 Hz, and
St = 0.133) from the present experiments. the interval between the two lasers is set to be 100 µs for
015102-5 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 2. Schematics illustrating (a) PIV

and hotwire, and (b) force measure-
ments. Flow is from left to right (i.e.,
along the x-direction).

Re = 1.4 × 103 –6 × 103 and 50 µs for Re > 6 × 103 . A charge- in Fig. 2(b), the cylinder model is horizontally cantilever-
coupled device (CCD) camera (Phantom V641, double frames, supported by a fixed frame placed outside of the tunnel. The
2560 × 1600 pixels), placed perpendicular to the laser sheets, free end of the cylinder is about 2 mm away from the tunnel
is used to capture the particle images. A Dantec Flowmap pro- wall. The load cell is tightly bolted in between two machine-
cessor is employed to synchronize the laser illumination and polished steel blocks. To avoid effects of the tunnel vibration
image capturing. As shown in Fig. 2, the PIV measurement on the force measurements, the cylinder model and its support
zone, located at the mid-span of the cylinder and symmetric are detached from the wind tunnel wall. The natural frequency
about y = 0, has a size of 4D × 4D. There is a shaded area under- ( fn ) of the test model is about 400 Hz in the present mounting.
neath the cylinder due to the shielding of the cylinder itself, There is a large separation between fn and the vortex-shedding
as the illumination is from top to bottom. About 900 pairs of frequency fs (=98–130 Hz), avoiding resonance and associated
images are captured for each Re considered. The algorithm of vibration. Calibration of the load cell is made in situ, which
cross correlation is adopted to calculate the velocity vectors, exhibits linearity between applied forces and output voltages
with an interrogation window of 32 × 32 pixels and 50% over- in the load cell range tested (not shown). The load cell signal
lapping in both the directions. More details of the algorithm is low-pass filtered at 1.0 kHz and sampled at 2.5 kHz using a
can be found in the work of Raffel et al.37 The spatial reso- 16-channel A/D converter. The sampling period is 45 s for each
lution of the velocity vectors is 0.04D in both the directions, run, corresponding to 4200–5600 vortex-shedding cycles.
and the temporal resolution of velocity fields is 0.0033 s cor-
responding to the repeating rate of 300 Hz. The experimental
uncertainty in determining velocities is estimated to be less IV. RESULTS AND DISCUSSION
than 3% for the present PIV system.
A. Typical flow structures
A single-wire hotwire probe (Dantec 55P01), mounted
on a computer-controlled three-dimensional traverse mecha- Figures 3–6 present the typical flow structures, in terms
nism, is used to measure the streamwise fluctuating velocity, of instantaneous spanwise vorticity (ωz∗ ) or vortical structures
u, in the near wake of the cylinder. Figure 2(a) illustrates the detected using the Q-criterion,38 in the near wake of the cylin-
hotwire measurement zone (1.5D × 0.4D) residing right above der as Re varies from Re = 1.0 × 102 to 1.0 × 103 . The
the upper side of the cylinder. The domain (1.5D × 0.4D) ωz∗ -distributions at Re = 1.0 × 102 [Fig. 3(a)] are from the two-
was selected for identifying the transition point in the shear dimensional numerical simulation, while the ωz∗ -distributions
layer. The measurement zone is a grid with the resolutions and the Q-criterion-based vortical structures at Re = 1.6 ×
of ∆x/D = 0.02 and ∆y/D = 0.016 in the x- and y-directions, 102 –1.0 × 103 [Figs. 3(b)–3(d) and 4–6] are from the three-
respectively. The hotwire (Pt-10% Rh), having a diameter of dimensional LES. The flow structures exhibit distinct features
5 µm and a length of around 2 mm, is operated at a constant at different Re. To facilitate the following discussion, the flow
temperature mode with an over-heat ratio of 1.8. The hotwire structures are grouped into three flow regimes, namely, lami-
signal is low-pass filtered and sampled at 1 kHz and 2–3 kHz nar flow regime (Fig. 3, Re ≤ 1.6 × 102 ), transition flow regime
(depending on Re), respectively, using a 16-channel analogue- (Figs. 4 and 5, 1.6 × 102 < Re < 2.6 × 102 ), and turbulent (or
to-digital (A/D) converter (NI PCI-6143). The sampling time post-transition) flow regime (Fig. 6, Re ≥ 2.6 × 102 ).
lasts for 30–60 s (depending on Re) for each record, which is In the laminar flow regime, the iso-contours of ωz∗
equivalent to 3700–5600 vortex-shedding cycles. The hotwire- [Figs. 3(a) and 3(b)] display two sets of opposite sign vor-
measured u is visually inspected to determine the occurrence tices arranged in an alternating fashion, exhibiting a typical
of shear layer transition. Moreover, the Strouhal number (St) feature of a von Kármán vortex street. The two-dimensional
is obtained from the power spectral density function E u of u at nature of this flow or parallel shedding of spanwise vortices is
x * = 1.0, y* = 0.9 which is done using a fast Fourier transform manifested by the iso-surfaces of ωz∗ (=±0.35) [Fig. 3(c)] and
algorithm. of Q (=2) [Fig. 3(d)], which are uniform along the spanwise
The time-mean drag force acting on the cylinder is mea- direction. A similar observation was made by Luo et al.17 in a
sured using a load cell (Model no MN020-10N). As shown dye flow visualization test at Re = 1.2–1.5 × 102 .
015102-6 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 3. Distributions of instantaneous

spanwise vorticity (ωz∗ ) in the square
cylinder wake. (a) Re = 1 × 102 , CFD
and [(b)–(d)] 1.6 × 102 , CFD. Iso-
contours of ωz∗ = ±0.35 in (a) and
(b) are denoted by solid and dotted lines.
Iso-surfaces in (c) are for ωz∗ = ±0.35
and those in (d) are for Q = 2.

FIG. 4. Evolution of the vortical structures at Re = 1.82

× 102 : (a) t * = to∗ , (b) to∗ + 46, (c) to∗ + 49, (d) to∗ + 52,
(e) to∗ + 60, and (f) to∗ + 70. Iso-surfaces correspond to
Q = 2 and are colored by streamwise vorticity (ωx∗ ).
015102-7 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 5. Evolution of the vortical structures at Re = 2.2

× 102 : (a) t * = to∗ , (b) t * = to∗ + 2, (c) t * = to∗ + 6, and
(d) t * = to∗ + 16. Iso-surfaces correspond to Q = 2 and are
colored by streamwise vorticity (ωx∗ ).

The transition from two-to three-dimensionality occurs in the first transition from two- to three-dimensionality. Stream-
the near wake, with mode A instability appearing at the criti- wise vortices having a relatively large spanwise wavelength of
cal Rec1 = 1.5–2.0 × 102 .14,16,17,20 Williamson1 for a circular (3–4)D are generated in the near wake of the circular cylin-
cylinder wake observed that the mode A instability occurs at der. St drops as the two-dimensional (laminar) vortex shedding

FIG. 6. Distributions of instantaneous spanwise vorticity (ωz∗ ) in the square cylinder wake. (a) Re = 2.6 × 102 , CFD and [(b)–(d)] 1.0 × 103 , CFD. Iso-contours
of ωz∗ = ±0.35 in (a) and (b) are denoted by solid and dotted lines. Iso-surfaces in (c) and (d) are for Q = 2 and colored by ωx∗ .
015102-8 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

changes to the three-dimensional (turbulent) vortex shedding. higher than that at the lower Re = 1.82 × 102 [Figs. 4(b) and
Figure 4 presents the evolution of the vortical structures in the 4(c)]. As a result, more streamwise vortices are generated near
square cylinder wake at Re = 1.82 × 102 to illustrate the devel- the cylinders [Fig. 5(b)]. These newly generated streamwise
opment of mode A instability. At the time instant to∗ [Fig. 4(a)], vortices develop in size with time [Figs. 5(c) and 5(d)], leading
the iso-surfaces display the staggered pattern of a von Kármán to the mode B instability. The spanwise wavelength of these
vortex street downstream, largely similar to that observed in streamwise vortices is about 1.2D, consistent with that experi-
Fig. 3(d) at Re = 1.6 × 102 , indicating the laminar or parallel mentally obtained by Luo et al.17 The streamwise vortices are,
vortex shedding from the cylinder. However, a small waviness however, less regular than those in mode A, perhaps due to the
along the spanwise direction (e.g., vortex 1) is discernible at strong interactions with the primary spanwise vortices.
a certain distance downstream of the cylinder [Fig. 4(a)]. The The critical Reynolds numbers associated with modes A
presence of the waviness on the primary spanwise vortices and B for a circular cylinder, obtained by Williamson,7 are
causes the deformation of the vortex sheets, which is likely to Rec1 ≈ 1.9 × 102 and Rec2 = 2.3–2.6 × 102 , respectively, which
be associated with the onset of mode A instability.18,39 Subse- are higher than those for a square cylinder. However, as noted
quently, streamwise vortices of opposite sign are generated in by Luo et al.,17 the vortical structures in modes A and B for
an alternating pattern along the spanwise direction, as seen at a circular cylinder wake appear similar to those for a square
time instant to∗ + 46 [Fig. 4(b)]. The spanwise vortical struc- cylinder wake, indicating similar generation mechanisms.
tures are further distorted, exhibiting irregular features along In the post-transition flow regime where mode B succeeds
the spanwise direction. Notable deformation of the vortex sheet (Fig. 6), with increasing Re, (i) the regularity of the streamwise
(of vortex 2) and shedding (of vortex 3) can now be seen. Ini- vortices degrades, (ii) the structure of spanwise vortices dete-
tiated from the separated shear layers and associated with the riorates, (iii) the streamwise separation between two succes-
primary spanwise vortices, the streamwise vortices grow in sive spanwise vortices elongates, and (iv) the wake including
size and evolve downstream, as indicated by the sequential spanwise and streamwise vortices becomes more turbulent.
plots in Fig. 4. Vortex loops may be formed [e.g., Fig. 4(e)],
but they are not self-sustaining rather decay subsequently after B. Vortex formation length and wake width
several vortex-shedding cycles. The spanwise wavelength of It is important to quantify the vortex formation length
the streamwise vortices is around 5D [Fig. 4(f)], close to (L f ) and wake width (W ) of a bluff-body wake, in view of
that (∼5.2D) observed by Luo et al.17 in their experimental their close connections with the base pressure and thus the
tests. The presence of streamwise vortices on both upper and drag force acting on the bluff body.8,41 The former is defined
lower sides of a primary vortex tube is regular, following the as the streamwise distance between the cylinder center and
spanwise wavelength. Dislocations of the spanwise vortices the shear layer rolling position identified as the point of the
were observed in the time histories of the streamwise fluctu- maximum (peak) streamwise fluctuating velocity (urms ) in the
ating velocities measured by a single hotwire in the transition wake,42 while the latter is defined as the transverse separation
flow regime,17 which is possibly due to the phase variation of between the two peaks in the urms -contours.43 Figure 7 presents
the large-scale coherent structures or to the merging of the sec- urms contours at different Re, with the definitions of L f and W
ondary streamwise vortices in the spanwise direction.16 Luo [Fig. 7(a)]. Notably, the urms -contours exhibit distinct charac-
et al.18 studied the flow around a square cylinder at Re = 50–4.0 teristics, dependent on Re. First, the streamwise location of the
× 102 , focusing on the transition phenomenon in the near wake. peaks shortens with increasing Re. It is internally connected
The transition Re for the onset of mode A instability was found to the shear layer transition, which will be discussed later.
to be Rec1 = 1.6 × 102 . The mode A had a spanwise wavelength Second, the lateral separation between the two peaks decays
of 5D, with the large-scale secondary vortices from the top and between Re = 120 and 182 and then grows for Re > 182, which
bottom being antiphase, which is consistent with the present will be made clear later. Third, vorticity is distributed largely
observations. Park and Yang40 noted the effects of rounding in the wake and less in the shear layers for Re ≤ 182 but is
the sharp corners of the square cylinder on the instabilities in distributed both in the shear layers and wake for Re > 182.
the near-wake. Fourth, a secondary peak is observed over the cylinder surface
With further increasing Re, the mode B instability occurs (x * ≤ 0.5) at Re ≥ 5 × 102 [Figs. 7(c)–7(f)], more conspicuous
at the second transition of the circular cylinder wake where at Re = 5 × 102 and 1.0 × 103 [Figs. 7(c) and 7(d)] but less at
the spanwise wavelength of streamwise vortices reduces to Re = 1.4 × 103 and 3.7 × 103 [Figs. 7(e) and 7(f)]. The sec-
(0.8–1)D and St jumps.1 The critical Re for the onset of mode ondary peak is ascribed to the reattachment of shear layers on
B instability in the square cylinder wake is Rec2 = 1.8–2.5 × the side surfaces. Figure 8 shows instantaneous vorticity struc-
102 .14,16,17,20 Figure 5 presents the evolution of the vortical tures at different Re. Evidently, the shear layers pass over the
structures in the square cylinder wake at Re = 2.2 × 102 to cylinder at a low Re = 100 [Fig. 8(a)]. However, the reattach-
illustrate the development of mode B instability. At the time ment of shear layers upon the trailing corner of the cylinder can
instant to∗ [Fig. 5(a)], the vortical structures exhibit typical be seen at Re = 5 × 102 and 1 × 103 [Figs. 8(b) and 8(c)]. When
features of mode A, i.e., the large-scale streamwise vortex Re is increased, the reattachment shifting upstream occurs on
loops having a relatively large spanwise wavelength (∼5D). the side surface [Fig. 8(d)]. The observation is consistent with
Meanwhile, a deformation of the vortex sheet associated with the secondary peak appearing in urms contours [Figs. 7(c)–
delayed shedding from the cylinder can be observed in the 7(f)]. The urms magnitude decreases with increasing Re, as
separated shear layer. It seems that the deformation of the vor- the reattachment location moves upstream, from the trailing
tex sheet at this higher Re = 2.2 × 102 [Fig. 5(a)] is much corner to the side surface [Figs. 8(d)–8(f)]. The reattachment
015102-9 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 7. Iso-contours of streamwise fluctuating velocities

urms (a) Re = 1.2 × 102 , CFD; (b) 1.82 × 102 , CFD;
(c) 5.0 × 103 , CFD; (d) 1.0 × 103 , CFD; (e) 1.4 × 103 ,
PIV; (f) 3.7 × 103 , PIV. White dotted-dashed curve in
panel (a) denotes ū= 0.

of the alternating shear layers upon the side surfaces of the the peak at Re > 1.5 × 103 [Figs. 9(d)–9(f)] is less energetic
cylinder is one of the reasons of the small urms magnitude and the power spectrum is less noisy. Indeed, an increase in Re
[Fig. 7(f)]. causing a shift in the shear layer transition upstream makes the
In order to provide further evidence of the above informa- reattachment changing from the corner to the side surface. For
tion, the typical power spectral density functions of fluctuating Re ≥ 5.0 × 102 , small peaks at 2St and/or 3St are discernible
lifts are presented in Fig. 9. Different from the Reynolds num- in Figs. 9(c) and 9(d) which are the signature of alternating
bers in Figs. 7 and 8, a few additional Reynolds numbers reattachment of the shear layers on the trailing corner or side
are chosen for the results in Fig. 9. There is only one pre- surface (depending on Re), as discussed earlier.
dominant peak at St = 0.146 in the power spectral density Figure 10 presents the dependence of Lf∗ on Re. Based on
function at Re = 1.0 × 102 [Fig. 9(a)], indicating the lami- the spatial resolution (0.04D) in the PIV measurements at Re
nar vortex shedding [Fig. 8(a)]. However, two dominant peaks > 1.0 × 103 , the uncertainty in Lf∗ is less than 2%. The stream-
(St = 0.152 and 0.161) are identified in the power spectral den- wise length (Lr∗ ) of the recirculation region is also included.15
sity function at Re = 1.82 × 102 [Fig. 9(b)]. The higher and Furthermore, for a comparison purpose, Lf∗ for the circular
lower Strouhal numbers correspond to the two-dimensional
and three-dimensional flows, respectively. Again, a single cylinder wake, reproduced from the work of Zdravkovich44
peak is discernible at Re > 5.0 × 102 [Figs. 9(c)–9(f)]. The and Norberg,45 is incorporated. Lf∗ dips rapidly from 5 to 2.2
peak is nevertheless strong, and the spectrum is very noisy at with increasing Re for 50 < Re < 1.6 × 102 , as indicated by
Re = 5.0 × 102 [Fig. 9(c)], indicating an alternating shear-layer the least-squares fitted line Lf∗ = 884.5Re−1.37 + 1.27. Mean-
reattachment at the corner [Figs. 8(b) and 8(c)]. On the other while, Lr∗ decreases with an increase in Re at 50 < Re < 1.6
hand, due to the reattachment on the side surface [Fig. 8(d)], × 102 ,14,15,30 echoing the present observations, a decreasing
015102-10 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 8. Instantaneous spanwise vorticity (ωz∗ ) at (a) Re

= 1.0 × 102 , (b) 5.0 × 102 , (c) 1.0 × 103 , and (d) 3.7 × 103
showing shear layer (a) overshooting, [(b) and (c)] reat-
tachment upon the trailing corner, and (d) reattachment
on the side surface.

Lf∗ with increasing Re in this Re range (Fig. 10). Note that Lr∗ of longitudinal vortex loops initiated by the mode A instability
increases from 0.6 to 3.5 with increasing Re (<50), where the plays a role in delaying the rolling of primary spanwise vor-
flow around the square cylinder is steady.14,15,30 At 1.6 × 102 tices,17 which results in Lf∗ not further reduced with increasing
< Re < 2.2 × 102 (transition regime) where mode A occurs, Lf∗ Re at the transition regime (Fig. 10). Robichaux et al.14 noted
stops dropping but fluctuates with Re (Fig. 10). The presence an approximately constant length of the recirculation region at

FIG. 9. Typical power spectral density (PSD) functions

of the fluctuating lift from numerical simulations [(a) Re =
1.0 × 102 ; (b) 1.82 × 102 ; (c) 5.0 × 102 ] and of the stream-
wise fluctuating velocity from hotwire measurements
[(d) 1.5 × 103 ; (e) 6.16 × 103 ; (f) 1.4 × 104 ].
015102-11 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 10. Variations with Re of the formation length

(L f * ). Flow regime classification is marked, and cor-
responding flow structures are sketched on top of the
figure.

1.6 × 102 < Re < 2.2 × 102 . At the upper limit of the transitional At Re > 2.0 × 104 , Lf∗ approaches asymptotically a constant
regime, the mode B instability takes place (Fig. 5). Sohankar of around 1.2, higher than that of the square cylinder wake.
et al.20 observed an increased intensity of secondary vortices Interestingly, the variation in Lf∗ with Re at Re = 2.6 × 102 –1.5
along with an enhanced intensity of the von Kármán vortices at × 103 differs between the square and circular cylinder wakes.
2.5 × 102 < Re < 5.0 × 102 . This may provide an explanation This difference can be ascribed to the distinct natures of the
of why Lf∗ declines rapidly from 2.2 to 1.1 with increasing Re flow separation over the square and circular cylinders. At this
at 2.2 × 102 < Re < 1.0 × 103 where intermittent reattachment Re range, the flow separation occurs at the leading edge of the
of the shear layer persists. At Re > 1.0 × 103 , the turbulent square cylinder [Figs. 8(b) and 8(c)], the separation line being
wake of the square cylinder is fully developed and the shear straight and time-independent (Fig. 6). On the contrary, the
layer reattachment is steady, where Lf∗ becoming less sensitive flow separation line over the surface of the circular cylinder is
to Re reaches an asymptotic value, i.e., Lf∗ ≈ 1. The variation oscillatory, following the rhythm of the primary vortex shed-
in Lf∗ with Re (>2.2 × 102 ) can be described by the power law ding and undulated by the streamwise vortical structures.20,46
Furthermore, the reverse flow in the wake of the circular cylin-
of Lf∗ = 8 × 104 Re−2 + 1.04 (Fig. 10).
der has an influence on the streamwise vortices.7 Therefore,
Now a classification of the flow can be provided based on
the increase in Lf∗ with Re for the circular cylinder may result
the variations in the flow structures and Lf∗ as well as Lr∗ . Here
from a strong interaction between the streamwise vortices and
five flow regimes are identified, namely, steady flow (Re < 50),
the flow separation.
laminar wake (2D flow) with alternating shedding (50 < Re <
The dependence on Re of the wake width W * is presented
1.6 × 102 ), transition from two-to three-dimensionality (1.6 ×
in Fig. 11. The spatial resolution in the PIV measurements
102 < Re < 2.2 × 102 ), shear layer transition I (2.2 × 102 <
at Re > 1.0 × 103 leads to an uncertainty of 2.2% in W * .
Re < 1.0 × 103 , alternating reattachment of shear layers at the
The data at Re < 1.6 × 102 and Re > 2 × 102 are separately
corner and transition occurring at Lt∗ > 0.5), and shear layer
used in the least-squares power law curve fitting. W * with Re
transition II (Re > 1 × 103 , reattachment of shear layers on the
decreases from 1.46 at Re = 60 to 0.95 at Re = 1.6 × 102
side surface and transition occurring at Lt∗ < 0.5). Robichaux
gradually in the laminar regime, following W ∗ = 4 × 10 4 Re
et al.14 and Yoon et al.15 observed that the flow separates from
+ 1.63 and then increases progressively in the post-transition
the trailing corners at Re < 120 but from the leading corners at
regime, following W ∗ = 14Re 0.85 + 1.15 (Re > 1.8 × 102 ).
Re > 120. The flow regimes have been marked and sketched
W * approaches asymptotically a constant of 1.15 at Re > 2 ×
at the top of Fig. 10, distinguishing the variation in Lf∗ .
104 , based on the fitted curve. The minimum W * , around 0.95,
For the circular cylinder wake, Lf∗ reduces rapidly with is attained in the transition regime. It is well known that W * is
increasing Re in the laminar regime at Re < 1.6 × 102 , similar linked to the fluid force acting on a bluff body, the larger the
to the Lf∗ behavior of the square cylinder (Fig. 10). However, W * , the lower the base pressure.47 Robichaux et al.14 observed
in the post-transition regime, L f increases gradually at 2.6 × a linear increase in the base pressure coefficient with Re at 70
102 < Re < 1.5 × 103 , reaches a maximum of 2.3 at Re = 1.5 < Re < 2.2 × 102 , in line with the linear decrease of W * with
× 103 , and then decays progressively to 1.2 at Re = 2.0 × 105 . Re in the same Re range (Fig. 11).
015102-12 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

velocity u/ū at (x * , y* ) = ( 0.1, 0.15), (0.25, 0.25), and (0.8,

0.35) for Re = 3.0 × 103 [see Fig. 2(a)]. The data presented in
Fig. 12 are chosen from a group of hotwire measurements at
different Re. Obviously, the variation in u/ū is regular and peri-
odic at (x * , y* ) = ( 0.1, 0.15), suggesting that the shear layer
is laminar at this location [Fig. 12(a)]. When the measurement
location is moved to (x * , y* ) = (0.25, 0.25), u/ū variation dis-
plays the occurrence of spikes and bursts in a random fashion
[Fig. 12(b)], indicating the onset of the transition in the shear
layer. The time history of streamwise fluctuating velocity in
Fig. 12(b) indeed shows the transition from the laminar flow
(at t * = 0–300) to turbulent flow (at t * = 300–700). The signals
in the former and latter ranges of t * bear similar character-
istics to those measured at (x * , y* ) = ( 0.1, 0.15) and (0.8,
0.35), respectively, which suggests that the transition to tur-
bulent occurs at (x * , y* ) = (0.25, 0.25). The transition feature
in Fig. 12(b) is quite typical in the hotwire data measured at
(x * , y* ) = (0.25, 0.25). The shear layers become fully turbu-
lent at (x * , y* ) = (0.8, 0.35) where more spikes and bursts
appear [Fig. 12(c)]. The power spectral density (PSD) func-
FIG. 11. Dependence of the wake width (W * ) on Re for a square cylinder.
tions of the signals are presented in Fig. 13. A profound peak
can be detected at f * = 0.136 for the laminar signals. How-
ever, this is not the case for the transition and post-transition
C. Shear layer transition
signals; instead, the peak is broad, embedded within the
A shear layer separating from the square cylinder bridges noise.
the boundary layer developing over the cylinder surface and The auto-correlation function ρuu (τ) of the streamwise
rolled-up vortices evolving downstream. Figure 12 presents fluctuating velocity is evaluated to further distinguish the lam-
typical time histories of the hotwire-measured streamwise inar, transition, and post-transition in the shear layer, where

FIG. 12. Time histories of the streamwise fluctuating

velocity u/ū measured by hotwire at Re = 3 × 103
(a) (x * , y* ) = ( 0.1, 0.15); (b) (x * , y* ) = (0.25, 0.25);
(c) (x * , y* ) = (0.8, 0.35). The initial time is arbitrary.
The inset in each panel denotes the measurement position
(dotted cross) of hotwire relative to the square cylinder.
015102-13 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

the cylinder center and the location for the onset of transition
in the shear layer.48 In the experimental tests, the location for
the onset of shear layer transition is determined through care-
ful visual inspection of the hotwire-measured u signals and
autocorrelations, while in the numerical simulations subgrid
turbulent viscosity ratios are used with a threshold of 0.04.
The uncertainty in L t from the hotwire measurements (Re >
1.0 × 103 ) is 0.01D based on the spatial resolution in the mea-
surement positions. Figure 15 presents the dependence of Lt∗
on Re. The circular cylinder case, reproduced from the work
of Zdravkovich,44 is included for the sake of comparison. An
increase in Re leads to Lt∗ declining exponentially, which is
described as Lt∗ = 1.6 × 102 Re−0.81 (Re > 1.6 × 102 ). While
Lt∗ ≈ 2.0–2.7 in the transition regime (1.6 × 102 < Re < 2.2 ×
102 ), Lt∗ = 0.5 at Re = 1 × 103 . The reduction rate slows down
at Re > 1 × 103 . The reduction in Lt∗ with Re is in line with the
FIG. 13. The power spectral density (PSD) functions of the streamwise reduction in L f * with Re (Fig. 10). Interestingly, Lt∗ > 0.5 in
fluctuating velocities at (x * , y* ) = ( 0.1, 0.15), (0.25, 0.25), and (0.8, transition regime I but Lt∗ < 0.5 in transition regime II, explain-
0.35) corresponding to laminar, transition, and post-transition, respectively.
ing why the shear layer reattachment occurs at the corner and
Re = 3 × 103 .
side surface for the transition regimes I and II, respectively. The
dotted-dashed line for the circular cylinder is obtained from
τ is the time-delay. The autocorrelation results are as pre- the hotwire-measured data by Bloor,48 Bloor and Gerrard,49
sented in Fig. 14 for the fluctuating velocity in Fig. 12. It and Gerrard.50 Lt∗ for the circular cylinder rapidly decreases
can be seen that ρuu (τ) for (x * , y* ) = ( 0.1, 0.15) exhibits at Re < 5.0 × 102 , levels off at 5.0 × 102 < Re < 1.0 × 103 ,
a clear oscillation, with amplitude decaying slowly with τ * , declines almost linearly at 1.0 × 103 < Re < 3.0 × 104 , and
which again indicates that the signal u is regular and periodic. becomes less sensitive to Re at 4.0 × 104 < Re < 8.0 × 104 .
That is, the flow is laminar at this point. However, ρuu (τ) for The almost unchanged Lt∗ observed at 5.0 × 102 < Re < 1.0
(x * , y* ) = (0.25, 0.25) and (0.8, 0.35) is small compared to × 103 is ascribed to the delayed rolling-up of the shear lay-
that for (x * , y* ) = ( 0.1, 0.15). The periodicity in ρuu (τ) with ers. Overall, Lt∗ is larger for the circular cylinder than for the
τ * cannot be identified at τ * > 90 (Fig. 14). ρuu (τ) at τ * < square cylinder at Re > 3.0 × 102 , with the largest departure
70, however, displays two periodicities, the longer and lower from each other occurring at Re ≈ 2.0 × 103 .
periods corresponding to von Kármán shedding and Kevin-
Helmholtz vortices. The ρuu (τ) amplitude associated with D. Mean drag force
Kevin-Helmholtz vortices is weaker at (x * , y* ) = (0.25, 0.25)
The time-mean drag coefficient is defined as C d =
than at (x * , y* ) = (0.8, 0.35). The difference explains transition
F d /(0.5ρDLz U∞2 ), where F is the time-mean drag force mea-
d
and post-transition occurring at the former and latter points,
sured by the force balance at Re ≥ 1.0 × 103 or estimated from
respectively.
The position of the shear layer transition can be quantified
as a length scale L t defined by the streamwise distance between

FIG. 14. The auto-correlation ρuu (τ) functions of the streamwise fluctuating
velocities at (x * , y* ) = ( 0.1, 0.15), (0.25, 0.25), and (0.8, 0.35) corresponding
to laminar, transition, and post-transition, respectively. τ denotes the time
delay. Re = 3 × 103 . FIG. 15. Dependence of L t on Re.
015102-14 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 16. Dependence of C d on Re.

the numerical simulation results at Re ≤ 1.0 × 103 , L z is the becoming dominant leads to the decrease in C d . In the transi-
span of the square cylinder, and ρ is the fluid density. Shown tion regime (1.6 × 102 < Re < 2.2 × 102 ) where L f * is almost
in Fig. 16 is the relationship between C d and Re. To provide invariant (Fig. 10), C d reaches a minimum of around 1.44 at
an overview of the dependence of C d on Re, the published Re = 182, complementing a minimum W * (Fig. 11). Beyond
data for a wide range of Re = 1–107 are comprehensively col- the transition regime, C d increases rapidly for 2.2 × 102 < Re
lected in Fig. 16. Meanwhile, the dependence of C d on Re for < 1.0 × 103 and then levels off at Re > 2.0 × 103 , approaching
a circular cylinder, reproduced from the work of Tritton,10 is a constant value of around 2.21. The rapid increase in C d at
included in Fig. 16 for a comparison purpose. Furthermore, 2.2 × 102 < Re < 1.0 × 103 is connected to a rapid drop in
Table II displays a quantitative comparison of C d at different L f * (Fig. 10) and a progressive increase in W * (Fig. 11). Fol-
Re from 102 to 103 . The present numerical and experimental lowing almost invariant L f * and W * for Re > 1.0 × 103 , C d
data are in good agreement with those from the literature at becomes more or less constant. The relationship between C d
60 ≤ Re ≤ 4.0 × 104 , providing a validation to the present and Re for Re > 2.2 × 102 can be described by the power law,
simulations and force measurements. C d = −8.6 × 102 Re−1.27 + 2.21.
It is worthy providing a synthesis discussion of the rela- Figure 17 presents typical time histories of the drag coef-
tionship between C d and Re, connecting the wake character- ficients for the laminar, transition, and turbulent regimes. The
istic parameters L f * and W * . These are the parameters having signal for the laminar flow [Fig. 17(a)] features a rather regular
a close connection with C d . Generally, a shorter L f * or a periodicity and amplitude, indicating that the laminar vortices
larger W * corresponds to a higher C d and vice versa.8,44,47,48,51 alternatively shed from the cylinder. On the contrary, those
As shown in Fig. 16, C d for the square cylinder exponen- for the transition [Fig. 17(b)] and turbulent flows [Fig. 17(c)]
tially decreases in the laminar flow regime (at Re < 1.6 × behave distinctively. First, the time history in Fig. 17(b) is
102 ). The relationship between C d and Re can be presented indicative of the growth of the mode A instability in the near
as C d = 13.2Re−0.85 + 1.2 (as indicated by the solid line in wake. It can be seen that rather regular periodic oscillation
Fig. 16), obtained using least-squares curve fitting of the data of C d , which is associated with the two-dimensional lami-
at Re < 1.6 × 102 . Though C d at Re < 1.6 × 102 is expressed nar vortex shedding [Fig. 18(a)], is reached at t * = 90–140.
by a single equation, the decrease of C d with Re is attributed However, due to the onset of mode A instability, the peri-
to different physics at Re < 50 and 50–1.6 × 102 . In the steady odic oscillation of C d changes into intermittent packets of
flow regime (at Re < 50), a pair of symmetric recirculation oscillation at t * > 140, with the transition from a packet of
bubbles (laminar eddies) forms downstream of the cylinder, a high-amplitude oscillation to that of a low-amplitude oscil-
elongating in the streamwise direction with an increase in lation. That is, the drag force fluctuates with a relatively large
Re.30 The decrease of C d with increasing Re in this regime magnitude as the wake is two-dimensional [e.g., t * = 90–140,
is attributed to the elongation of the bubbles accompanied by Figs. 17(b) and 18(a)] but with a relatively small magnitude as
an increased base pressure.30 On the other hand, alternating the wake is three-dimensional [e.g., t * = 200–250, Figs. 17(b)
von Kármán vortex shedding takes place at 50 < Re < 1.6 × and 18(b)]. As such, two different C d in the transition regime
102 (unsteady laminar flow) with both L f * and W * shortening (Table II) are presented, calculated separately for the laminar
with Re (Figs. 10 and 11). While the shortening of W * tends to and turbulent states, based on the examination of the time his-
decrease C d , that of L f * does the opposite. The influence of W * tories of the drag force. A similar observation was made in the
015102-15 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

FIG. 17. Typical time histories of drag coefficient C d for

the (a) laminar flow (Re = 100), (b) transitional flow (Re =
1.82 × 102 ), and (c) turbulent flow (Re = 5 × 102 ). The
data are from numerical simulations.

previous DNS work.20 Our simulation revealed that the inter- frequencies obtained in the power spectra of the lift signals
mittent packets of oscillation appeared in a random manner up (Fig. 9).
to t * = 1000 examined. Second, the drag force signal for the It is of interest to compare the square cylinder wake with
turbulent flow [Fig. 17(c)] exhibits the representative turbu- the circular cylinder, in terms of the dependence of C d on Re. It
lence fluctuation, echoing the turbulent wake comprising both can be seen that C d for the circular cylinder decreases at Re <
large- and small-scale flow structures (Fig. 6). 2.0 × 102 . A similar trend is observed for the square cylinder at
The power spectral density (PSD) functions of the drag the same Re range. However, with further increasing Re (>2.0
signals are presented in Fig. 19. Clearly, dominant peaks × 102 ), C d for the circular cylinder maintains the declination,
are observed for the three states of the flow. For the lam- albeit at a smaller rate, until Re = 2.0 × 103 , which is opposite
inar flow (Re = 100), the peak is detected at f * = 0.295, to that in the square cylinder counterpart. C d for the circular
while for the turbulent flow (Re = 500) it is identified at cylinder does not change much; C d = 1.1–1.2 for Re = 2.0
f * ≈ 0.272. As expected, for the transitional flow (Re = 182), × 103 –1.0 × 105 . The so-called critical flow regime appears
twin peaks are detected at f * = 0.303 and 0.32, indicating at Re = 2.0 × 105 –4.0 × 105 where C d drops suddenly. This
the turbulent and laminar states, respectively. As expected, is ascribed to the postponed flow separation and formation of
the peaks from the drag signals occur at twice the dominant separation bubbles, both linked to the boundary layer transition

FIG. 18. Iso-surfaces of instantaneous spanwise vortic-

ity (ωz∗ = ±0.35) corresponding to (a) t * = 140 and (b) 290,
respectively, in Fig. 17(b). Red and blue colors denote
positive and negative ωz∗ , respectively. Re = 1.82 × 102 .
015102-16 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

unsteady flow at 50 < Re < 1.6 × 102 can be represented by

the solid curve St = 3.8Re 1 + 0.18. There is a scattering of
St data at the transitional regime.
The present and collected data at 2.2 × 102 < Re <
1.0 × 103 are used to fit into a curve St = 0.54Re 0.23 . At
1.0 × 103 < Re < 2.0 × 105 , St augments following St =
1.9Re 0.73 + 0.134 (dotted-dashed line). Though covering a
wide range of Re = 1.0 × 103 –5.0 × 106 , the LES data by
Sohankar 29 appear slightly lower than those by others; there-
fore, his data are not included in the curve fitting for Re > 1.0
× 103 . For the circular cylinder wake, St with Re increases
progressively to a maximum of 0.22 at Re = 1.0 × 103 , indi-
cating an increase in the shedding frequency of the dominant
vortices. However, with further increasing Re from 1.0 × 103
to 1.0 × 105 , St declines slowly from 0.22 to 0.2 due to the
shear layer transition. Noticeable vortex shedding is difficult
to be detected in the cylinder near wake as the flow goes into
the critical regime at Re > 1.0 × 105 , which is indicated by
FIG. 19. The power spectral density (PSD) functions of drag forces for the
laminar flow (Re = 100), transitional flow (Re = 182), and turbulent flow
the dotted line in Fig. 20.
(Re = 500).
F. Global view of the flow dependence on Re
from laminar to turbulent.39 The critical regime is not observed It is worth summarizing the global flow features at dif-
in the square cylinder wake as the flow separation is fixed on ferent flow regimes at a glance. Table III lists the prominent
the sharp corners and thus insensitive to Re. Beyond Re = 4.0 characteristics and dependence of different parameters on Re.
× 105 , C d for the circular cylinder climbs gradually and then The up and down arrows in the table, respectively, indicate
levels off with an increase in Re. increase and decrease with Re, while the horizontal arrow
denotes approaching asymptotic value. In the steady flow
E. Strouhal number regime (Re < 50), generated immediately downstream of the
Figure 20 presents the relationship between St and Re. St square cylinder is a pair of symmetric recirculation bubbles,
from the present numerical simulations is estimated using the which are elongated with increasing Re, corresponding to the
fluctuating lift force, while that from the present experimen- declining C d . No vortex shedding is identified at Re < 50. In the
tal tests are calculated using the hotwire measured fluctuating laminar wake regime (50 < Re < 160), vortex shedding is how-
velocities. The present numerical and experimental data are ever observed, and the unsteady wake is two-dimensional. The
in general consistent with those from the literature (see also flow separation occurs from the trailing corners for 50 < Re <
Table II). With increasing Re at 50 < Re < 1.0 × 103 , St 120 and from the leading corners for 120 < Re < 160. Both the
increases and attains a maximum at Re ≈ 2.0 × 102 before vortex formation length (Lf∗ ) and wake width (W ∗ ) decreases
declining from 0.16 to 0.12 at 2.0 × 102 < Re < 1.0 × 103 . with an increase in Re, accompanied by a reduction in C d
The increase of St from ≈0.11 to 0.16 with Re in the laminar and a rise in St. The transition from two- to three-dimensional

FIG. 20. Variation in St with Re.

015102-17 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

TABLE III. Dependence of flow parameters on Re at different flow regimes. The up and down arrows, respectively, indicate increase and decrease with Re, while the horizontal arrow denotes approaching asymptotic value.

Separation from the leading corners, alternating shear layer

Separation from the leading corners, alternating shear layer

reattachment on the trailing corner, transition downstream
and vice versa in the flow occurs in the transition regime (160

and separation from the leading corner (50 < Re < 160)
< Re < 220). The primary spanwise vortex shedding is accom-

Steady flow, no separation, and two-dimensional wake

reattachment on the side surface, transition upstream

separation from the trailing corner (50 < Re < 120),

of the trailing corner, and turbulent vortex shedding

of the trailing corner, and turbulent vortex shedding

Primary vortex shedding, streamwise vortex loops,
panied by the streamwise vortex loops with a large spanwise

Laminar vortex shedding, two-dimensional wake,

wavelength of around 5D. Two different values of C d and St
are detected, with the higher C d (higher St) and lower C d
(lower St) corresponding to the two-dimensional laminar and
three-dimensional turbulent flow, respectively. In the transition
regime, Lf∗ is virtually unvaried, W ∗ and C d reach their minima,

and three-dimensional wake

and St attains its maximum. The cylinder wake becomes tur-
symmetric recirculation,

bulent at Re > 220. Shear layer transition I is recognized at 220

< Re < 1.0 × 103 , where the laminar shear layer reattaches on
the trailing corner and then the transition occurs downstream
Major features

of the trailing corner.52–58 Lt∗ decreases exponentially with Re,

together with the decreasing Lf∗ and St; however, W ∗ increases
progressively, associated with a rising C d ,59 with an increase
+ 0.134 (↑) and St → 0.134 in Re. Finally, shear layer transition II is noted at Re > 1.0
× 103 as the shear layer reattaches on the side surface due
to the occurrence of laminar to turbulent shear layer transition
0.165

St = −1.9Re−0.73
St = 0.54Re−0.23
St = −3.8Re−1

upstream of the trailing corner. With increasing Re, all the flow
+ 0.18 (↑)

parameters (i.e., Lf∗ , W ∗ , Lt∗ , C d , and St) approach their cor-

(↓)
...
St

St = 0.14

responding constants in this regime. For example, C d → 2.21

and St → 0.134.

V. SUMMARY
C d = −8.6 × 102 Re−1.27
1.53
Cd = 13.2Re−0.85

C d = 13.2Re−0.85

The wake of a square cylinder is extensively investigated

C d → 2.21

for Re < 107 . The dependence on Re of the streamwise length

+ 2.21(↑)
+ 1.2 (↓)
+ 1.2(↓)

C d = 1.37
Cd

of steady recirculation bubble Lr∗ , vortex formation length Lf∗ ,

wake width W * , shear layer transition length Lt∗ , time-mean
drag coefficient C d , and Strouhal number St are documented.
Their dependence corresponds to five flow regimes, namely,
steady flow (Re < 50), laminar flow (50 < Re < 1.6 × 102 ),
Lt∗ = 1.6 × 102 Re−0.81

Lt∗ = 1.6 × 102 Re−0.81

Lt∗ = 1.6 × 102 Re−0.81

two- to three-dimensional transition flow (1.6 × 102 < Re <

(↓) and Lt∗ → 0

2.2 × 102 ), shear layer transition I (2.2 × 102 < Re < 1.0 ×
(↓)

(↓)
...

...
Lt∗

103 ), and shear layer transition II (Re > 1.0 × 103 ). A discus-
sion is made on the comparison of the square cylinder results
obtained presently with the circular cylinder results from the
literature.
At Re < 50, the flow around a square cylinder is steady,
W ∗ = −4 × 10−4 Re

W ∗ = −14Re−0.85

characterized by a pair of counter-rotating recirculating bub-

W ∗ → 1.15
W ∗ ≈ 0.95
+ 1.63 (↓)

+ 1.15 (↑)

bles downstream of the cylinder. No alternating von Kármán

W∗

...

vortex shedding occurs from the cylinder. With increas-

ing Re, Lr∗ grows rapidly from 0.6 to 3.5 and C d declines
gradually.
At 50 < Re < 1.6 × 102 , the unsteady laminar flow occurs,
Lf∗ = 8 × 104 Re−2
Lf∗ = 884.5Re−1.37
Lr∗ = 0.6 3.5

with the alternating von Kármán vortex shedding from the

Lf∗ → 1.04
1.27 (↓)

+ 1.04 (↓)
Lf∗ ≈ 2.2
Lf∗ or Lr∗

cylinder. Lf∗ dips rapidly from 5.0 to 2.2 with Re. A circular
(↑)

cylinder has a similar trend in this Re range. W * decreases from

+

1.46 at Re = 60 to 0.95 at Re = 1.6 × 102 . The variation in C d

with Re is similar to that in W * in view of internal coherence.
C d declines slowly as Re is increased to 1.6 × 102 . However,
Re > 1 × 102
220 < Re
< 1 × 103
Re ranges

160 < Re
Re < 50

50 < Re
< 160

< 220

St grows from 0.11 to 0.16 when Re is increased from 50 to

1.6 × 102 .
At 1.6 × 102 < Re < 2.2 × 102 , mode A and mode B
instabilities prevail, resulting in the transition from two-to
transition II
Shear layer

Shear layer
transition I
transition

three-dimensionality in the cylinder wake. Streamwise vor-

2D to 3D
Laminar
Steady

tices come into being with a spanwise wavelength of 5D in

wake
flow

mode A and 1.2D in mode B. Lf∗ remains unchanged, W * and

015102-18 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

C d reach their minima, and St becomes the maximum at this 8 A. Roshko, “On the drag and shedding frequency of two-dimensional bluff
transition flow regime. bodies,” NACA Technical Note 3169, 1954.
9 A. Roshko, “On the wake and drag of bluff bodies,” J. Aeronaut. Sci. 22,
At 2.2 × 102 < Re < 1.0 × 103 , shear layer transition 124 (1955).
I is observed, where the transition in the shear layer occurs 10 D. J. Tritton, Physical Fluid Dynamics, 2nd ed. (Oxford University Press,

downstream of (Lt∗ > 0.5) the trailing corner of the cylinder. Lf∗ Oxford, England, 1988).
11 Y. Zhou, S. X. Feng, M. M. Alam, and H. L. Bai, “Reynolds number effect
drops again rapidly from 2.2 to 1.1 with Re increasing from 2.2
on the wake of two staggered cylinders,” Phys. Fluids 21, 125105 (2009).
× 102 to 1.0 × 103 . On the contrary, Lf∗ for the circular cylinder 12 M. M. Alam, “The aerodynamics of a cylinder submerged in the wake of

varies oppositely at this Re range, increasing progressively another,” J. Fluids Struct. 51, 393 (2014).
13 A. Sohankar, C. Norberg, and L. Davidson, “Low-Reynolds-number flow
with Re. Lf∗ is greater for the circular cylinder than for the
around a square cylinder at incidence: Study of blockage, onset of vortex
square cylinder at Re > 3.0 × 102 . With increasing Re from shedding and outlet boundary condition,” Int. J. Numer. Methods Fluids 26,
2.2 × 102 to 1.0 × 103 , W * and C d increase from their minima, 39 (1998).
14 J. Robichaux, S. Balachandar, and S. P. Vanka, “Three-dimensional Floquet
St decreases from 0.16 to 0.12, and Lt∗ drops from 2.0 to 0.5. On
instability of the wake of square cylinder,” Phys. Fluids 11(3), 560 (1999).
the other hand, Lt∗ for the circular cylinder reduces for Re = 2.2 15 D. Yoon, K. Yang, and C. Choi, “Flow past a square cylinder with an angle
× 102 –7 × 102 and levels off for Re = 7 × 102 –1.0 × 103 . of incidence,” Phys. Fluids 22, 43603 (2010).
At Re > 1.0 × 103 , shear layer transition II is identified. 16 A. K. Saha, G. Biswas, and K. Muralidhar, “Three-dimensional study of

flow past a square cylinder at low Reynolds numbers,” Int. J. Heat Fluid
The transition in the shear layer occurs upstream of (Lt∗ <
Flow 24(1), 54 (2003).
0.5) the trailing corner of the cylinder, and the shear layer 17 S. C. Luo, Y. T. Chew, and Y. T. Ng, “Characteristics of square cylinder
alternately reattaches on the side surface of the cylinder. Lf∗ wake transition flows,” Phys. Fluids 15(9), 2549 (2003).
18 S. C. Luo, X. H. Tong, and B. C. Khoo, “Transition phenomena in the wake
asymptotically approaches a constant value of about Lf∗ ≈ 1.0.
of a square cylinder,” J. Fluids Struct. 23(2), 227 (2007).
On the contrary, Lf∗ for the circular cylinder reaches its max- 19 A. Sohankar, L. Davidson, and C. Norberg, “Numerical simulation of

imum of 2.3 at Re = 1.5 × 103 and then decreases gradually unsteady flow around a square two-dimensional cylinder,” in Proceedings of
12th Australasian Fluid Mechanics Conference (The University of Sydney,
to 1.2 at Re = 2.0 × 105 . Both W * and C d increasing with Re Australia, 1995), p. 517.
become almost constant at Re > 1.0 × 104 , with C d ≈ 2.21 at 20 A. Sohankar, C. Norberg, and L. Davidson, “Simulation of three-

Re > 1.0 × 104 . On the other hand, C d for the circular cylinder dimensional flow around a square cylinder at moderate Reynolds numbers,”
becomes almost a constant of 1.2 at 1.0 × 103 < Re < 2.0 × Phys. Fluids 11(2), 288 (1999).
21 A. Okajima, “Strouhal numbers of rectangular cylinders,” J. Fluid Mech.
105 that is followed by a sudden dip at 2.0 × 105 < Re < 5.0 × 123, 379 (1982).
105 and recovers at Re > 5.0 × 105 . No drag crisis flow regime 22 D. A. Lyn and W. Rodi, “The flapping shear layer formed by flow separation

is identified for the square cylinder. Lt∗ decreases more rapidly from the forward corner of a square cylinder,” J. Fluid Mech. 267, 353
for the square cylinder than the circular cylinder at this Re (1994).
23 C. Brun, S. Aubrum, T. Goossens, and P. Ravier, “Coherent structures and
range, approaching an asymptotic value at Re > 1.0 × 104 and their frequency signature in the separated shear layer on the sides of a square
4.0 × 104 for the circular and square cylinders, respectively. St cylinder,” Flow, Turbul. Combust. 81, 97 (2008).
having a slow recovery at 1.0 × 103 < Re < 4.0 ×103 becomes 24 M. Minguez, C. Brun, R. Pasquetti, and E. Serre, “Experimental and high-

order LES analysis of the flow in near-wall region of a square cylinder,” Int.
∼0.134 for Re > 4.0 ×103 . St for the circular cylinder reaches
J. Heat Fluid Flow 32, 559 (2011).
the maximum of about 0.22 at Re ≈ 1.0 × 103 but declines to 25 Y. Cao and T. Tamura, “Large-eddy simulations of flow past a square
about 0.2 at Re = 1.0 × 105 , followed by a sharp rise at the cylinder using structural and unstructured grids,” Comput. Fluids 137, 36
drag crisis flow regime. (2016).
26 F. X. Trias, A. Gorobets, and A. Oliva, “Turbulent flow around a square

cylinder at Reynolds number 22,000: A DNS study,” Comput. Fluids 123,

87 (2015).
ACKNOWLEDGMENTS 27 D. A. Lyn, S. Einav, W. Rodi, and J.-H. Park, “A laser-Doppler velocimetry

This work is supported by the National Natural Science study of ensemble-averaged characteristics of the turbulent wake of a square
cylinder,” J. Fluid Mech. 304, 285 (1995).
Foundation of China (NSFC) through Grant Nos. 11672096 28 A. K. Saha, K. Muralidhar, and G. Biswas, “Experimental study of flow
and 11302062. The authors would like to thank Mr. Y. F. Lin past a square cylinder at high Reynolds numbers,” Exp. Fluids 29, 553
for his contribution to the mesh generation and computational (2000).
29 A. Sohankar, “Flow over a bluff body from moderate to high Reynolds
setup and Mr. Q. T. Zhang for his work on the experiments.
numbers using large eddy simulation,” Comput. Fluids 35(10), 1154
(2006).
1 C. 30 A. Sharma and V. Eswaran, “Heat and fluid flow across a square cylinder
H. K. Williamson, “Vortex dynamics in the cylinder wake,” Annu. Rev.
Fluid Mech. 28, 477 (1996). in the two-dimensional laminar flow regime,” Numer. Heat Transfer, Part
2 D. Sumner, S. J. Price, and M. P. Paidoussis, “Flow-pattern identification A 45, 247 (2004).
31 Q. Zheng and M. M. Alam, “Intrinsic features of flow past three square
for two staggered circular cylinders in cross-flow,” J. Fluid Mech. 411, 263
(2000). prisms in side-by-side arrangement,” J. Fluid Mech. 826, 996 (2017).
3 M. M. Alam, H. Sakamoto, and Y. Zhou, “Determination of flow configu- 32 J. P. Van Doormaal and G. D. Raithby, “Enhancements of the SIMPLE

rations and fluid forces acting on two staggered circular cylinders of equal method for predicting incompressible fluid flows,” Numer. Heat Transfer
diameter in cross-flow,” J. Fluids Struct. 21, 363 (2005). 7(2), 147 (1984).
4 M. M. Alam, Y. Zhou, and X. W. Wang, “The wake of two side-by-side 33 Y. F. Lin, H. L. Bai, M. M. Alam, W. G. Zhang, and K. Lam, “Effects

square cylinders,” J. Fluid Mech. 669, 432 (2011). of large spanwise wavelength on the wake of a sinusoidal wavy cylinder,”
5 M. M. Alam, H. L. Bai, and Y. Zhou, “The wake of two staggered square J. Fluids Struct. 61, 392 (2016).
34 Y. F. Lin, H. L. Bai, and M. M. Alam, “The turbulent wake of a square prism
cylinders,” J. Fluid Mech. 801, 475 (2016).
6 Y. Zhou and M. M. Alam, “Wake of two interacting circular cylinders: with wavy faces,” Wind Struct. 23(2), 127 (2016).
35 M. R. Mankbadi and N. J. Georgiadis, “Examination of parameters affecting
A review,” Int. J. Heat Fluid Flow 62, 510 (2016).
7 C. H. K. Williamson, “Three-dimensional wake transition,” J. Fluid Mech. large-eddy simulations of flow past a square cylinder,” AIAA J. 53(6), 1706
328, 345 (1996). (2015).
015102-19 H. Bai and M. M. Alam Phys. Fluids 30, 015102 (2018)

36 H. Tennekes and J. L. Lumley, A First Course in Turbulence (The MIT Press, 48 M. S. Bloor, “The transition to turbulence in the wake of a circular cylinder,”

Cambridge, Massachusetts, and London, England, 1994), 15th printing. J. Fluid Mech. 19(2), 290 (1964).
37 M. Raffel, C. Willert, and J. Kompenhans, Particle Image Velocimetry. 49 S. M. Bloor and J. H. Gerrard, “Measurements on turbulent vortices in a
A Practical Guide (Springer, 1998). cylinder wake,” Proc. R. Soc. A 294, 319 (1966).
38 J. C. R. Hunt, A. A. Wray, and P. Moin, “Eddies, stream and convergence 50 J. H. Gerrard, “Experimental investigation of separated boundary layer

zones in turbulent flows,” in Proceedings of Summer Program (Center for undergoing transition to turbulence,” Phys. Fluids 10(supplement), 98
Turbulence Research, Stanford University, 1988), p. 193. (1967).
39 X. Mao and H. M. Blackburn, “The structure of primary instability modes 51 S. Balachandar, R. Mittal, and F. M. Najjar, “Properties of the mean recircu-

in the steady wake and separation bubble of a square cylinder,” Phys. Fluids lation region in the wakes of two-dimensional bluff bodies,” J. Fluid Mech.
26, 074103 (2014). 351, 167 (1997).
40 D. Park and K. S. Yang, “Flow instability in the wake of a rounded square 52 A. K. De and A. Dalal, “Numerical simulation of unconfined flow past a

cylinder,” J. Fluid Mech. 793, 915 (2016). triangular cylinder,” Int. J. Numer. Methods Fluids 52, 801 (2006).
41 J. H. Gerrard, “The mechanics of the formation region of vortices behind 53 A. P. Singh, A. K. De, V. K. Carpenter, V. Eswaran, and K. Muralidhar,

bluff bodies,” J. Fluid Mech. 25, 401 (1966). “Flow past a transversely oscillating square cylinder in free stream at low
42 O. M. Griffin, “A note on bluff body vortex formation,” J. Fluid Mech. 284, Reynolds numbers,” Int. J. Numer. Methods Fluids 61, 658 (2009).
217 (1995). 54 T. Tamura and T. Miyagi, “The effect of turbulence on aerodynamic forces
43 O. M. Griffin and S. E. Ramberg, “The vortex street wakes of a vibrating on a square cylinder with various corner shapes,” J. Wind Eng. Ind. Aerodyn.
cylinders,” J. Fluid Mech. 66, 729 (1974). 83, 135 (1999).
44 M. M. Zdravkovich, Flow Around Circular Cylinders. Volume 1 (Oxford 55 C. Norberg, “Flow around rectangular cylinders: Pressure forces

University Press, 1997). and wake frequencies,” J. Wind Eng. Ind. Aerodyn. 49, 187
45 C. Norberg, “LDV-measurements in the near wake of a circular cylinder,” (1993).
in Advances in Understanding of Bluff Body Wakes and Vortex-Induced 56 S. C. Luo, M. G. Yazdani, Y. T. Chew, and T. S. Lee, “Effects of incidence and

Vibration (ASME, Washington, D.C., 1998), paper no. FEDSM98–521. afterbody shape on flow past bluff cylinders,” J. Wind Eng. Ind. Aerodyn.
46 K. Yokoi and K. Kamemoto, “Initial stage of a three-dimensional vortex 53, 375 (1994).
structure existing in a two dimensional boundary layer separation flow 57 T. Igarashi, “Drag reduction of a square prism by flow control using a small

(observations of laminar boundary separation over a circular cylinder by rod,” J. Wind Eng. Ind. Aerodyn. 69-71, 141 (1997).
flow visualization),” JSME Int. J., Ser. II 35, 189 (1992). 58 R. D. Blevins, Flow-Induced Vibration, 2nd ed. (Krieger Publishing
47 H. Nakaguchi, K. Hasimoto, and S. Muto, “An experimental study of aero- Company, 2001).
dynamic drag on rectangular cylinders,” J. Jpn. Soc. Aeronaut. Eng. 16, 1 59 M. M. Alam and Y. Zhou, “The turbulent wake of an inclined cylinder with

(1968). water running,” J. Fluid Mech. 589, 261 (2007).