Journal of Fluids and Structures 43 (2013) 179–199

Contents lists available at ScienceDirect

Journal of Fluids and Structures

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

Flow around four cylinders arranged in a square configuration

X.K. Wang a,n, K. Gong a, H. Liu b, J.-X. Zhang b, S.K. Tan c
a
Maritime Research Centre, Nanyang Technological University, 639798 Singapore, Singapore
b
Department of Engineering Mechanics, School of Naval Architecture, Ocean and Civil Engineering, Shanghai Jiao Tong University,
Shanghai 200240, China
c
Nanyang Environment and Water Research Institute, School of Civil and Environmental Engineering, Nanyang Technological University,
639798 Singapore, Singapore

a r t i c l e i n f o abstract

Article history: This paper presents an experimental study of the flow around four circular cylinders
Received 13 November 2012 arranged in a square configuration. The Reynolds number was fixed at Re ¼ 8000, the
Accepted 16 August 2013 pitch-to-diameter ratio between adjacent cylinders was varied from P/D ¼2 to 5 and the
Available online 1 October 2013
incidence angle was changed from α ¼ 01 (in-line square configuration) to 451 (diamond
Keywords: configuration) at an interval of 7.51. The flow field was measured using digital Particle
Four cylinders Image Velocimetry (PIV) to examine the vortex shedding characteristics of the cylinders,
Flow interference together with direct measurement of fluid dynamic forces (lift and drag) on each cylinder
Strouhal number using a piezoelectric load cell. Depending on the pitch ratio, the flow could be broadly
Force coefficients
classified as shielding regime (P/Dr 2), shear layer reattachment regime (2.5 r P/Dr 3.5)
Particle Image Velocimetry (PIV)
and vortex impinging regime (P/DZ 4). However, this classification is valid only in the
case that the cylinder array is arranged nearly in-line with the free stream (α E01),
because the flow is also sensitive to α. As α increases from 01 to 451, each cylinder
experiences a transition of vortex shedding pattern from a one-frequency mode to a two-
frequency mode. The flow interference among the cylinders is complicated, which could
be non-synchronous, quasi-periodic or synchronized with a definite phase relationship
with other cylinders depending on the combined value of α and P/D. The change in vortex
pattern is also reflected by some integral parameters of the flow such as force coefficients,
power spectra and Strouhal numbers.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction

The study of flow around circular cylindrical structures is of both fundamental and practical significance. A great deal of
research work has been carried out in order to understand the classical problem of an isolated (or single) circular cylinder in
cross flow, see the extensive reviews by Williamson (1996), Norberg (2003), Zdravkovich (2003) and Bearman (2011). On the
other hand, in many cases, cylindrical structures are arranged in groups (or arrays) in applications for offshore structures,
chimneys, power lines, heat exchanger tubes, etc. Due to mutual interference, the flow around multiple cylinders is usually a
much more complicated, and thus is less well studied and understood than the case of an isolated cylinder.
The presence of neighboring cylinders significantly affects the wake flow patterns, force coefficients, vortex shedding
frequencies, as well as vortex-induced vibration (VIV). Two cylinders arranged in various configurations (side-by-side,
tandem, or staggered), being the simplest example of multiple-cylinder arrays, have attracted considerable attention over

n
Corresponding author. Tel.: þ65 67906619; fax: þ65 67906620.
E-mail address: cxkwang@ntu.edu.sg (X.K. Wang).

0889-9746/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfluidstructs.2013.08.011
180 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

the last two decades and hence are now much better understood, as reviewed by Sumner (2010), in which about 150
technical papers were compiled.
The number of cylinders is obviously a critical parameter in determining the level of complexity in flow interference.
As compared to the two-cylinder configuration, much less attention has been paid to the case of even larger number of
cylinders, such as the three- or four-cylinder arrays, see for example the review by Khalifa et al. (2012) regarding the state-
of-art research progresses on this topic. In particular, a four-cylinder array arranged in the square configuration is a
fundamental element in offshore structures (e.g., semi-submersible platform), pipe bundles and tube banks. The geometry
of the four-cylinder array is set by the spacing between the cylinders (which is typically expressed as the ratio between the
cylinders' centre-to-centre pitch and the cylinder diameter, P/D, thereafter abbreviated as the pitch ratio) and the
orientation angle of the cylinder array relative to the oncoming flow (α). When the cylinder's aspect ratio (AR ¼L/D, where
L is the length or span of the cylinder, and D is the cylinder diameter) is sufficiently large (AR Z8 according to previous
finding, e.g., Lam and Zou (2010)) that the flow can be treated as two-dimensional (2D), and is governed by three most
important variables: (i) pitch ratio (P/D); (ii) incidence angle (α); and (iii) Reynolds number (Re), which is defined as
Re¼UD/ν, where U is the free-stream velocity and ν is the kinematic viscosity of fluid. To date, there are only a handful of
reported experimental and numerical studies on the four-cylinder array, which are summarized in Table 1. The published
studies, focusing on either the in-line square (α¼01) or the diamond (α¼451) configuration, revealed that the flow around
four cylinders is far more complicated than the single- or two-cylinder counterparts. The interference among the cylinders
in close proximity to one another may give rise to flow separation, reattachment, shear-layer instability and vortex
impingement, as well as VIV (Zhao and Cheng, 2012). Clearly, the principle of superposition cannot be applied to this highly
nonlinear situation.
The four-cylinder array in a square configuration was first investigated by Sayers (1988, 1990), who measured the force
coefficients and vortex shedding frequencies for different pitch ratios and incidence angles at Re ¼3  104. Our present
understanding of this flow configuration is largely attributed to the long-term research efforts by Lam and co-authors over
the past 20 years (e.g., Lam and Lo, 1992; Lam and Fang, 1995; Lam et al., 2003a, 2003b; Lam et al., 2008; Lam and Zou,
2010). Most of the published experiments were conducted either in the laminar regime (Re ¼100–200, mainly for flow
visualization study) or in the subcritical regime (Re ¼103–104, for measurements of pressure, velocity, drag and lift). On the
other hand, it is noted that during the past several years, some numerical simulations have been carried out on the four-
cylinder array at α¼01 or 451. Except for the LES study by Lam and co-authors (Lam and Zou 2007, 2009; Zou et al., 2008) at
Re¼1.5  104 and the FEM study by Zhao and Cheng (2012) at Re¼103–2  104, most numerical studies were limited to
relatively low Reynolds numbers (Re o300), e.g., Farrant et al. (2000), Lam et al. (2008), Esfahani and Be Hagh (2010),
Lam and Zou (2010), Tong et al. (2011), and Zou et al. (2011). Based on a flow visualization study conducted on the four-
cylinder array at α ¼01 (in-line square configuration), Lam and Lo (1992) classified the flow behavior into three distinct types
of interference as a function of P/D: namely the shielding, reattachment and impinging regimes, which are shown in Fig. 1.
Because of the complexity of the multiple-cylinder wake, the vortex shedding frequency (f), typically expressed in non-
dimensional form as the Strouhal number, St¼ fD/U, may vary significantly across the wake. Lam et al. (2003b) showed that
when P/D is less than 1.7, the flow around a four-cylinder array is biased (or deflected) behind the downstream cylinders,
which is characterized by a narrow wake of higher frequency and a wide wake of lower frequency, as compared to that of a
single cylinder. The different vortex shedding frequencies would be appropriately associated with individual shear layers

Table 1
Summary of published studies on four cylinders in a square configuration.

Researchers Method Techniquea Re P/D α (interval)

Sayers (1988, 1990) Exp. Pressure, HWA 3  104 1.1–5 0–1801 (151)
Lam and Lo (1992) Exp. Pressure, FV 2.1  103 1.28–5.96 0–451 (151)
Lam and Fang (1995) Exp. Pressure 1.28  104 1.26–5.8 0–451 (151)
Lam et al. (2003a) Exp. LIF, PIV 200 4 0–451 (51)
Lam et al. (2003b) Exp. Force, LIF 200, 800 (LIF) 1.69–3.83 0–451 (151)
2.25–4.5  104 (Force)
Lam and Zou (2007, 2009); Zou et al. (2008) Exp. LDA, PIV 1.1–2  104 1.5–5 01
Lam et al. (2008) Exp. LIF 100, 200 1.6–5 01
Present Exp. Force, PIV 8  103 2–5 0–451 (7.51)
Farrant et al. (2000) Sim. BEM (2D) 200 3, 5 01, 451
Lam and Zou (2007, 2009); Zou et al. (2008) Sim. LES (3D) 1.5  104 1.5, 3.5 01
Lam et al. (2008) Sim. FVM (2D, 3D) 100, 200 1.5, 3.5 01
Esfahani and Be Hagh (2010) Sim. LBM (2D) 100 1.5–4.5 01
Lam and Zou (2010) Sim. FVM (2D, 3D) 200 1.6–5 01
Tong et al. (2011) Sim. FVM (2D, 3D) 270 2–6 01
Zou et al. (2011) Sim. FVM (2D, 3D) 200 1.2–5 451
Zhao and Cheng (2012) Sim. FEM (2D) 103–2  104 3 0–451 (151)

a
FV—flow visualization; HWA—hot-wire anemometry; LDA—laser Doppler anemometry; LIF—laser-induced fluorescence; PIV—Particle Image
Velocimetry; BEM—Boundary Element Method; FEM—Finite Element Method; FVM—Finite Volume Method; LBM—Lattice Boltzmann Method; LES—Large
eddy simulation.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 181

Fig. 1. Typical flow patterns around four cylinders in the in-line square configuration: (a) shielding; (b) reattachment; and (c) impinging. A schematic
sketch modified from Lam et al. (2008).

rather than individual cylinders, as suggested by Sumner et al. (2000) on two cylinders in a staggered arrangement. Several
experimental studies using PIV and LIF (Lam and Fang, 1995; Lam et al., 2003a, 2003b; Lam and Zou, 2007, 2009; Zou et al.,
2008) confirmed the distinct flow regimes as a function of P/D (though the exact values of P/D to delineate the different
regimes may vary among those studies), as well as corresponding changes in the force coefficients and vortex shedding
frequencies (Lam and Fang, 1995; Lam et al., 2003b). Furthermore, the incidence angle strongly influences both the vortex
shedding frequency (Sayers, 1990) and the flow pattern (Lam et al., 2003a).
Nevertheless, due to the limited data available on the four-cylinder array, a number of important issues have yet to be
clarified on the flow interference among the cylinders, as well as on the correlation between the flow patterns and the fluid
dynamic forces (lift and drag) on the cylinders. For instance, most of the previous studies obtained the force coefficients by
integrating the pressure distribution on the cylinder surface (Sayers, 1988, 1990; Lam and Fang, 1995). The only study in the
literature that directly measured the spanwise averaged forces is that of Lam et al. (2003b). However, only one cylinder was
instrumented with a load cell, and hence simultaneous measurements of the forces on the four cylinders were not realized.
Moreover, quantitative information on the flow around multiple-cylinder arrays is still rather scarce, particularly in the
subcritical regime of direct practical importance. As commented by Sumner (2010), the fluid behavior determined from
measured quantities, such as the vortex shedding frequency by spectral analysis of either fluctuating forces on the cylinder
or the velocity signal at a particular point in the wake, is prone to misinterpretation particularly when the study was
conducted without the benefit of accompanying flow visualization. Therefore, Particle Image Velocimetry (PIV) has been
employed in this study, since it is an effective technique in providing whole-field, instantaneous velocity and vorticity
distributions. To date, there are only a handful of PIV studies on the four-cylinder array, namely Lam et al. (2003a), Lam and
Zou (2007, 2009) and Zou et al. (2008), in which, however, the fluid forces on the cylinders were not measured. Moreover,
because most of the previous studies were focused on either α¼01 or 451, the sensitivity of the flow structure and force
coefficients to very small change in the incidence angle is not yet well understood. These considerations motivate the
present study on the four-cylinder array at a subcritical Reynolds number (Re ¼8000). We will examine the combined
effects of pitch ratio over the range P/D¼2–5 and incidence angle over the range α¼ 0–451 (with an interval of 7.51) on the
flow patterns, Strouhal numbers, power spectra, mean and fluctuating forces. Each cylinder in the cylinder array was
instrumented with a piezoelectric load cell for concurrent direct force measurement. The objective of this study is to provide
further insight into the vortex formation and interactions for multiple cylinders. The data collected will serve as a
benchmark for validation of computational fluid dynamics (CFD) codes.

2. Experimental set-up and methodology

The experiments were performed in a re-circulating rectangular open channel located at Maritime Research Centre,
Nangyang Technological University, with a test section of 5 m  0.3 m  0.45 m (length  width  height). The channel
bottom and the two side walls of the test section were made of glass to allow for optical access. The free-stream velocity was
uniform to within 1.5% across the test section, and the turbulence intensity in the free stream was well below 1%.
Fig. 2(a) shows a sketch of the experimental set-up. The arrangement of the four-cylinder array in the square
configuration is depicted in Fig. 2(b), in which the cylinders are labeled as 1–4 (denoted as Cy-1, Cy-2, Cy-3 and Cy-4).
The cylinder models, made of smooth, transparent, acrylic rod with an outer diameter of D¼20 mm, were suspended
vertically in the water channel and subjected to the cross flow. The origin of the coordinate system was located at the center
point of the four-cylinder array, with x, y and z denoting the streamwise, transverse and spanwise directions, respectively. In
this experiment, the free-stream velocity was in x-direction, which was also the positive drag direction, while the positive
lift was in y-direction. The pitch ratio was varied as P/D ¼2, 2.5, 3, 3.5, 4 and 5. The lower end of each cylinder had a small
clearance of  2 mm (0.1D) from the surface of the bottom end plate to encourage two dimensionality of the flow (Slaouti
and Gerrard, 1981). The upper end of each cylinder was fixed to a separate load cell, which was in turn mounted onto a top
plate that could be rotated in the anti-clockwise direction to the desired angle of incidence (α). The end plates,
manufactured from acrylic plate of 10 mm in thickness, were circular in shape (diameter ¼280 mm) with a sharp edge.
The cylinder array was always positioned in the middle of the test section to minimize the effects from the two side walls of
the channel. The angle of incidence was varied from α¼01 to 451 at an interval of 7.51. Therefore, a total of 42 cases were
182 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

Fig. 2. Schematic diagram of the experimental set-up: (a) side view; (b) top view.

considered in the present study. In order to eliminate the free-surface effects, the top plate was perforated with four circular
apertures of 22 mm in diameter, leaving a small gap of  1 mm when the cylinder was slotted through.
The blockage ratio per cylinder was about 6.7%, which is similar to that in Lam et al. (2003a, 2003b) and Lam and Zou
(2009). However, the total blockage of the four-cylinder array would vary up to several times (e.g., 2 times in the case of
α¼ 01, and 3 times in the case of α¼451) the single-cylinder value, which would yield an under-prediction of the oncoming
velocity. Therefore, care was paid to ensure that the oncoming (free-stream) velocity in every case was maintained constant
at U¼0.4 m/s, corresponding to Re¼8000. During the experiments, the water surface elevation was aligned with the top
end plate, so the immersed length of the cylinders was kept constant at H¼200 mm. This yielded an aspect ratio of 10,
which was large enough to ensure a nominally 2-dimensional (2D) flow along the majority of the central span, see for
example by West and Apelt (1993) on a single cylinder and Lam and Zou (2010) on a four-cylinder array. Thus, all the
measurements were carried out at the horizontal, mid-span plane of the submerged part of the cylinders, where the end
effects were minimal. The flow field was illuminated with a double cavity Nd:YAG laser light sheet at 532 nm wavelength
(Litron model, power 135 mJ per pulse, duration  5 ns). Sphericels 110P8 hollow glass spheres (neutrally buoyant with a
mean diameter of 13 μm) were seeded in the flow as tracer particles, which offered good traceability and scattering
efficiency. The images were recorded using a 12-bit charge-coupled device (CCD) camera, which had a resolution of
1600  1200 pixels and a sampling rate of 15 Hz.
LaVision Davis software (Version 7.2) was used to process the particle images and determine the velocity vectors. Particle
displacement was calculated using the fast-Fourier-transform (FFT) based cross-correlation algorithm with standard
Gaussian sub-pixel fit structured as an iterative multi-grid method. The processing procedure included two passes, starting
with a grid size of 64  64 pixels, stepping down to 32  32 pixels overlapping by 50%, which resulted in a set of 7500
vectors (100  75) for a typical field. In between passes, the vector maps were filtered by using a 3  3 median filter in order
to remove possible outliers. The number of particles in a 32  32 pixel window was of the order of 10–15, which was
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 183

sufficient to yield strong correlations. The field of view was 200 mm  150 mm, therefore the spatial resolution was
2 mm  2 mm (i.e., 0.1D  0.1D). The uncertainty in the instantaneous velocities (u and v) was estimated to be about 2% for
the present set-up. Based on the velocity distribution, the instantaneous spanwise vorticity (ωz ¼Δv/Δx  Δu/Δy) was
calculated using the least squares extrapolation scheme. The uncertainty in ωz was about 10%. A more detailed description of
the PIV post-processing procedure and uncertainty analysis was given in Wang and Tan (2008).
A load cell (piezoelectric Kistler Model 9317B) was mounted between each cylinder and the turntable plate to measure
the fluid dynamic forces, drag (FD) and lift (FL), on the cylinder (integrated over the immersed span of the cylinder). This type
of load cell, having the advantage of high response, resolution and stiffness, has been used to measured the forces on
cylinders either stationary (Lam et al., 2003b) or undergoing VIV (Wang et al., 2012, 2013). The amplified output signal was
captured with a National Instruments D/A card at a sampling rate of 1000 Hz (which was at least two orders of magnitude
greater than the vortex shedding frequency). A signal exceeding 150 vortex shedding cycles in length may be considered to
be stationary in the wake of a cylinder (Sakamoto et al., 1987). In the present study, the duration of recording for each test
case was about 200 s, corresponding to about 240–1600 cycles of vortex shedding (with a frequency range of 3–8 Hz). Then,
the mean and root-mean-square (RMS) values of drag and lift coefficients of each cylinder were calculated. Through a
number of repeated measurements on a single circular cylinder of the same diameter, the uncertainty in the mean drag
coefficient (C D ) was determined to be within 0.5%. The data for the single cylinder measured under the same Reynolds
number of Re¼8000 served as the benchmark reference: C D ¼1.1, C D0 ¼0.08, C L0 ¼0.14 and St ¼0.19. In general, the data are in
good agreement with the reported results on a single cylinder using similar measurement technique (e.g., Tadrist et al.,
1990).

3. Results and discussion

3.1. In-line square configuration (α¼01)

At α ¼01, the four-cylinder array can be regarded as two parallel rows of two cylinders in tandem. Fig. 3 depicts the
effects of pitch ratio (P/D¼ 2–5) on the instantaneous velocity and vorticity distributions measured with PIV. At P/D¼ 2, the
shear layers separated from Cy-1 and Cy-2 are elongated chains of vorticity in Kelvin–Helmholtz (K-H) type of roll-up. They
largely wrap around the downstream cylinders (Cy-4 and Cy-3) and are connected to alternative vortices formed after the
cylinder array. As P/D increases to 3, the separated shear layers are more likely to reattach on the leading surface of the
downstream cylinders, resulting in that no discrete vortices are shed from each cylinder. At P/D¼ 4, they begin to curl up

Fig. 3. A representative snapshot of the instantaneous flow fields around the cylinder array in in-line square configuration (α ¼01) at P/D ¼ (a) 2; (b) 3;
(c) 4; and (d) 5. The velocity vector is plotted with a reference-frame velocity of 0.5U, superimposed with flood contours of normalized spanwise vorticity.
184 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

into discrete vortices in the gaps between the upstream and downstream cylinders, most notably the anti-clockwise vortex
from the lower side of Cy-2. When P/D increases to 5, the vortex formation length, defined as the streamwise distance from
the axis of the cylinder to the core of fully formed vortex, becomes shorter. The vortices are shed alternately from two sides
of each upstream cylinder, and periodically move downstream and impinge on the respective downstream cylinder. These
flow characteristics are consistent with the flow patterns shown in Fig. 1. Depending on the value of P/D, the shear layers
separated from the upstream cylinders produce different interference with the downstream cylinders: (a) shielding pattern
at P/D r2; (b) shear layer reattachment pattern at 2.5 rP/Dr3.5; and (c) vortex impinging pattern at P/DZ4. A similar
observation has been reported for two cylinders in tandem arrangement, for example, Zhou and Yiu (2006). The critical
pitch ratio for the transition from shear layer reattachment regime (in instability mode of symmetric K-H vortices) to vortex
impinging regime (in instability mode of anti-symmetric von Kármán vortices) is 3.5–4, which also agrees with reported
values of 3–5 by earlier researchers.
The different flow patterns lead to corresponding variation of drag and lift coefficients with P/D, as shown in Fig. 4. The
data reported by Lam et al. (2003b) on Cy-1 and Cy-4 are also included for comparison. The two sets of data agree
satisfactorily, especially so for the mean drag and lift coefficients (C D and C L ). The mean drag coefficients of the upstream
cylinders, i.e., C D1 and C D2 (subscript i denotes the i-th cylinder), attain their maxima at the smallest pitch ratio (P/D¼2),

Fig. 4. Variation of force coefficients on each cylinder versus P/D in the case of in-line square configuration (α ¼01): (a) mean drag coefficient (C D ); (b) RMS
drag coefficient (C D0 ); (c) mean lift coefficient (C L ); and (d) RMS lift coefficient (C L0 ).
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 185

which is about 25% higher than the corresponding value of C D0 E1.1 for a single cylinder. The downstream cylinders (Cy-3
and Cy-4), on the other hand, always show a lower mean drag because they are immersed in the wake of the upstream
cylinders. While C D3 and C D4 increase monotonically from 0.2 to 0.8, C D1 and C D2 decrease slightly until P/D ¼4, and
thereafter maintain a value close to that of a single cylinder. Fig. 4(b) reveals that there is a net repulsive force between the
upper (C L 40) and lower (C L o0) rows of cylinders, most notably at small pitch ratios. However, a closer examination
reveals that the values of C D , C D0 or C L0 for the upper row and lower row of cylinders (i.e., Cy-1 vs. Cy-2, and Cy-3 vs. Cy-4)
may not perfectly coincide especially in the case of small pitch ratios: for instance, at P/D ¼2 or 2.5, C D3 is appreciably higher
than C D4 . In addition, the C L values are not exactly symmetric about the x-axis. This asymmetric behavior in the force
coefficients is attributed to (i) possible misalignment of the cylinders (since a perfect alignment is hard to achieve during the
experiments); and (ii) bi-stable flow pattern that has been observed for the two side-by-side cylinders (Sumner et al., 1999)
and for a four-cylinder array (Lam et al., 2003b). For example, Sumner et al. (1999) experimentally studied the flow past two
side-by-side cylinders in the ranges of Re ¼500–3000 and P/D¼1–5, and observed four vortex shedding patterns: a single
bluff-body pattern for 1rP/D o1.2, a biased pattern for 1.2rP/Dr2.2, and anti-phase and in-phase synchronized patterns
for 2.2 oP/D r4.5. It is established that a narrow wake would induce a higher drag on the cylinder than a wide wake.
Furthermore, Alam et al. (2003) showed that in the bi-stable regime, the flow between the cylinders may spontaneously flip
over from one side to the other, and may also be biased (deflected) steadily to one side. The bi-stable patterns for the present
four-cylinder array are evident in the time histories of force signals as will be shown later.
The RMS drag (C D0 ) and lift (C L0 ) coefficients plotted against P/D are shown in Fig. 4(c) and (d), respectively. The amplitude
of C D0 and C L0 for the downstream cylinders (Cy-3 and Cy-4) can be several times higher than that of the upstream cylinders
or a single cylinder. At P/D r3.5 the values of C D0 for the upstream and downstream cylinders are nearly constant at about

Fig. 5. Time history of dynamic drag (CD) and lift (CL) coefficients on each cylinder in the case of in-line square configuration (α¼ 01) at P/D¼ (a) 2; (b) 3;
(c) 4; and (d) 5.
186 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

0.03 and 0.075, respectively, but they rise noticeably with further increase in P/D. This is especially true for the two
downstream cylinders (Cy-3 and Cy-4), with an extraordinarily sharp increase near the critical pitch ratio P/DE3.5. It is
noted that the magnitude of C D0 and C L0 is considerably lower than that reported by Lam et al. (2003b), but the overall
variation trend with P/D for these two data sets is similar.
The time histories of dynamic drag (CD) and lift (CL) coefficients on each cylinder are shown in Fig. 5 for different pitch
ratios (P/D¼2, 3, 4 and 5). As P/D increases, the CL signals change from a chaotic or quasi-periodic pattern (at P/D¼2 and 3),
to a periodic pattern (at P/D¼4 and 5) with a much higher magnitude of fluctuation and an almost identical frequency.
Obviously, the increased periodicity of fluctuation in the latter is related to regular shedding of vortices from the upstream
cylinders (Cy-1 and Cy-2) and subsequent impinging on the downstream cylinders (Cy-3 and Cy-4). Furthermore, the
concurrent signals on the four cylinders allow for an analysis of the relationship among them. The CL signals for the
cylinders change from a non-synchronous or quasi-periodic pattern at small pitch ratios (P/D ¼2 or 3), to a synchronized
pattern with a definite phase relationship at large pitch ratios (P/D ¼4 or 5). Most notably at P/D ¼5 (Fig. 5(d)), CL3 and C L4
are obviously in anti-phase (with a phase lag of 1801); CL1 and C L2 , although much less sinusoidal than CL3 and C L4 , are also
roughly anti-phase. This is in accordance with the vortex pattern shown in Fig. 3(d) that the anticlockwise vortices shedding
from the lower side of Cy-1 (and Cy-4) and the clockwise vortices shedding from the upper side of Cy-2 (and Cy-3) are
approximately at the same streamwise location. At P/D¼ 4, a similar anti-phase relationship is found between CL3 and C L4 ,
but to a lesser degree. At even smaller pitch ratios (P/D ¼2 or 3), on the other hand, the signals become quasi-periodic with
non-constant frequencies and amplitudes, so that the phase lag between them varies with time. However, at P/D¼2 one can
even observe the anti-phase relationship between CL3 and C L4 sometimes (most notably at t¼2–3 s in Fig. 5(a)), whereas at
P/D ¼3 the signals become completely non-synchronous (Fig. 5(b)). This observation is supported by the PIV results shown
in Fig. 3(b) that no clear discrete vortices are formed in the wake of the downstream cylinders at this pitch ratio.
Fig. 5 also shows that CD is always more chaotic than CL, which is ascribed to the competition between the anti-
symmetric von Kármán vortices and the symmetric K-H vortices shed from the upstream cylinders and their subsequent
interactions with the downstream cylinders. Being symmetric about the cylinder centerline, the K-H vortices at a frequency
about 1–2 orders higher than that of von Kármán vortex shedding at Re ¼103–104 (Rajagopalan and Antonia, 2005), do not
affect the spectral characteristics of the lift, but would result in super-harmonic components on the drag (e.g., CD1 and C D2 ).
In each case, CD1 and C D2 are non-stationary and may not coincide with each other; instead they intermittently exhibit the
“flip-over” behavior with a time-scale that can be several times longer than vortex shedding. Similar behavior is found for
CD3 and C D4 , which are more chaotic with a considerably larger magnitude of fluctuation. The flip-over phenomenon would
be responsible for the asymmetric behavior of the mean drag and lift coefficients shown in Fig. 4.
It is therefore helpful to perform spectral analysis of the lift signal, in order to quantitatively determine the frequencies of
vortex shedding. Fig. 6 shows the power spectral density (PSD) of lift coefficients on the cylinders at different pitch ratios (P/D¼
2–5). In general, the spectral peaks for each cylinder become stronger as P/D increases. For the upstream cylinders (Cy-1 or
Cy-2), a prominent peak at f¼ 3.7 Hz (corresponding to St ¼0.185) is observed in the spectra for P/D Z4, but when P/D o3.5
there are no obvious peaks. For Cy-3 or Cy-4, on the other hand, peaks are always discernible even at the smallest pitch ratio
of P/D ¼2. At moderate pitch ratios (2.5 rP/Dr3.5), however, the spectra are rather broad-banded (or lack of well-defined

Fig. 6. Power spectrum of lift coefficients on the cylinders at different pitch ratios in the case of in-line square configuration (α ¼ 01): (a) Cy-1 and Cy-2;
(b) Cy-3 and Cy-4.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 187

peaks), indicating that the interactions are complex and intermittent. Additionally, note that the peak frequencies for Cy-3
and Cy-4 at P/D¼ 2.5 or 3 do not coincide, which is likely attributed to the bi-stable phenomenon, since the narrow wake
and the wide wake are associated with different vortex shedding frequencies (Afgan et al., 2011). When the pitch ratio is
larger still (P/DZ4), the peak for Cy-3 or Cy-4 becomes sharp and prominent at the same frequency as that of the upstream
cylinders, suggesting periodic shedding of vortices from the upstream cylinders and subsequent impinging on the downstream
cylinders.
The force coefficients and Strouhal numbers for the case of P/D ¼3.5 are respectively listed in Table 2 and the left column
of Table 3, together with published data at similar gap ratios (P/D¼3.4–4). In general, the mean drag coefficient (C D ) and
Strouhal number (St) agree well with the literature, whereas the RMS lift coefficient (C L0 ) shows a large discrepancy from
those reported by Lam et al. (2003b) and Lam and Zou (2009). This discrepancy is, however, not surprising, considering that
there is a wide scatter of documented C L0 values even on a single cylinder, see Norberg (2003). The data scatter is attributed
partly to the large variation in flow conditions (such as Reynolds number, free-stream turbulence level, aspect ratio and
blockage ratio of the cylinder), and partly to the difference in measurement techniques used in these studies.

3.2. Diamond configuration (α ¼451)

This section presents the results for the four-cylinder array set at the diamond configuration (α¼451). In relative to the in-
line square configuration (α¼01), the four-cylinder array has been rotated 451 in the anti-clockwise direction, resulting in Cy-3
being located directly downstream of Cy-1 in the streamwise direction (x-axis), whilst Cy-2 and Cy-4 arranged side-by-side in
the transverse direction (y-axis).
Fig. 7(a)–(d) shows a typical flow field for the diamond configuration (α¼451) at P/D ¼2, 2.5, 3 and 4, respectively. The
flow patterns are completely different from those of α¼01. Due to the sandwich effects imposed by the two flanking
cylinders (Cy-2 and Cy-4), the flows around Cy-1 at small pitch ratios are characterized as high-velocity jets, which first
converge behind Cy-1 and then diverge in front of Cy-3. At P/D¼2, the flows behind Cy-1 show distinct inward deflection,
resulting in a narrower and shorter wake than that of an isolated cylinder. The length of vortex formation from Cy-1 is very
short (about 1D from the trailing edge of the cylinder), and discrete vortices are evident in its wake. The significant smaller
size of vortices shed from Cy-1 implies higher frequency of vortex shedding, which agrees with the force measurement data
(to be shown later). The shed vortices periodically impinge on Cy-3, or in impinging regime, which is in contrast to the
shielding regime observed for the in-line square configuration (α¼ 01) at similar pitch ratios. When P/D increases to 2.5
(Fig. 7(b)), the gap flows become stronger but with a smaller curvature. Consequently, the shear layers are more elongated in
the streamwise direction, accompanied by a longer vortex formation length and a longer streamwise wavelength of the shed
vortices than the case of P/D¼2. At P/DZ3, the flow interference between Cy-1, Cy-2 and Cy-4 decreases further, as
indicated by the fact that the vortex shedding from each of the three cylinders approaches that of a single cylinder in terms
of vortex formation length and vortex shedding frequency.
Fig. 8 shows the variation of lift and drag coefficients on each cylinder as a function of P/D for the four-cylinder array at
α¼451. As compared to the case of α¼01, the coefficients on each cylinder (particularly Cy-1, Cy-2 and Cy-4) are less
sensitive to P/D. Cy-2 and Cy-4, being geometrically symmetric about the flow direction (x-axis), also display a slightly

Table 2
Comparison with published mean drag, mean and RMS lift coefficients on four cylinders arranged in in-line square configuration (α¼ 01).

Researchers L/D C D1 C D2 C D3 C D4 C L1 C L2 C L3 C L4 C ′L1 C ′L2 C ′L3 C ′L4

Sayers (1988, 1990) 4 1.27 0.38

Lam and Fang (1995) 3.58 1.35 1.36 0.59 0.59
Lam et al. (2003b) 3.4 1.19 0.25 0.038 0.075 0.06 0.38
Lam and Zou (2009) 3.5 1.28 1.27 0.70 0.69 0.66 0.62 1.22 1.25
Present 3.5 1.27 1.13 0.43 0.4 0.076 -0.085 -0.066 0.054 0.038 0.037 0.164 0.186

Table 3
Comparison with published Strouhal numbers on four cylinders arranged in: in-line square configuration (α¼ 01)—left column; and diamond configuration
(α¼ 451)—right column.

α¼ 01 α ¼451
Researchers L/D
St1 St2 St3 St4 St1 St2 St3 St4

Sayers (1988, 1990) 4 0.186 0.177 0.2 0.19

Lam and Lo (1992) 3.51 0.175 0.15 0.292 0.257 0.143 0.24
Lam et al. (2003b) 3.4 0.20 0.167 0.28/0.21 0.27/0.22 0.27
Lam and Zou (2009) 3.5 0.196 0.192 0.192 0.196
Present 3.5 0.196 0.171 0.23 0.23 0.35/0.153 0.23
188 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

Fig. 7. A representative snapshot of the instantaneous flow fields around the cylinder array in diamond configuration (α ¼451) at P/D¼ (a) 2; (b) 2.5;
(c) 3; and (d) 4.

asymmetric behavior in the force coefficients similar to the case of α¼01, which is likely due to the flip-over pattern in the
time histories of dynamic force signals (to be shown later). The data for Cy-1, Cy-2 and Cy-3 reported in Lam et al. (2003b) are
also included, which are in good agreement with the present study, especially C D and C L . Since Cy-3 is located in the wake of
Cy-1, C D3 is consistently lower than C D1 . On the other hand, C D2 and C D4 are always higher than C D1 within the measurement
range (P/D¼2–5). Therefore, the mean drag coefficients follow the relationship: C D3 oC D1 oC D2 EC D4 . The relatively high
mean drag for the two flanking cylinders (Cy-2 and Cy-4) is attributed to fluid acceleration formed in the gaps between each
flanking cylinder and the upstream cylinder. Similar trend was reported by Lam et al. (2003b). They also showed that at even
smaller pitch ratios, C D2 is lower than C D1 , because the biased flow would be formed at P/Do1.7.
Furthermore, a few notable observations can be made on the RMS coefficients. Firstly, for a given P/D,
0 0 0 0 0 0 0 0
C D1 oC D2 EC D4 oC D3 , and C L1 oC L2 EC L4 oC L3 . Secondly, while the variation of C D0 and C L0 for Cy-1, Cy-2 and Cy-4 is
0 0 0
relatively insensitive to P/D, C D3 and C L3 (particularly C L3 ) increase steadily with P/D within the measurement range. This
0
trend is in contrast to that for the α¼01 case, where C L3 experiences a sharp increase at about P/D ¼3.5 due to flow
transition from reattachment regime to impinging regime. In addition, Cy-3 for the diamond configuration (α ¼451) is
0
always impinged by periodic vortex shedding from Cy-1, and hence the values of C L3 at the small-to-moderate pitch ratios
(P/D r3.5) are significantly higher than those of α ¼01.
Fig. 9 shows the time histories of CD and CL on each cylinder for the diamond configuration (α¼451) at different pitch ratios.
Similar to the α¼01 case, the flip-over phenomenon is always observed for the cylinders arranged side-by-side (i.e., Cy-2 and
Cy-4 in this configuration), and the lift on each cylinder generally becomes more regular with increasing P/D. However, even at
the smallest pitch ratio of P/D¼ 2 (Fig. 9(a)), CL3 already oscillates quite regularly at a frequency of about 8 Hz (much higher
than the corresponding value of  4 Hz for a single cylinder in the same flow), whereas the oscillations of CL1, CL2 or CL4 are less
regular but still roughly at the same frequency as CL3. Obviously, at this pitch ratio there is no clear phase relationship among
the four CL signals. On the other hand, a closer check of the CD signals shows that CD2 and CD4, oscillating at the same frequency
of  8 Hz, are almost anti-phase to each other. These observations agree well with the flow patterns shown in Fig. 7(a) that
during one cycle of vortex shedding from Cy-1, the anticlockwise vortex and the clockwise vortex more or less impinge on the
leading surfaces of Cy-4 and Cy-2, respectively, which take place in an alternate manner (with a phase lag of 1801). As P/D
increases to 3 (Fig. 9(b)), the CL signals become more periodic but their phase differences are not definite (or time-variant);
meanwhile, CD2 and CD4 are less periodic and the phase relationship between them also becomes unclear, indicating that the
interference between Cy-1, Cy-2 and Cy-4 is decreased. At P/D¼4 and 5 (Fig. 9(c) and (d)), all the CL signals oscillate at an
identical frequency of about 4 Hz (similar to an isolated cylinder) with phase differences that are almost invariant with time.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 189

Fig. 8. Variation of force coefficients on each cylinder versus P/D in the case of diamond configuration (α¼ 451): (a) C D ; (b) C D0 ; (c) C L ; and (d) C L0 .

Fig. 10 shows the power spectra of the dynamic lift coefficient on each cylinder at different pitch ratios. A single well-
defined peak is found in the spectrum of Cy-1: the peak frequency decreases as P/D increases, i.e., from 8 Hz at P/D¼2,
to 4 Hz at P/D¼5, which is highlighted by the thick line. On the other hand, the spectra for the other three cylinders (Cy-2, Cy-3
and Cy-4) may have two distinct peaks, e.g., Cy-2 and Cy-4 at P/Dr2.5, and Cy-3 at P/DZ2.5. Similar two-peak character was
reported by Lam and Lo (1992) and Lam et al. (2003b), which was attributed to the unsteady wake of Cy-1 due to the sandwich
effects of Cy-2 and Cy-4. However, the present study suggests that the two distinct peaks are more likely due to periodic
shedding and mutual interaction of the vortices. Even at the smallest pitch ratio (P/D¼2), vortex shedding from Cy-1 is very
regular and so is the lift fluctuation. The difference between the two frequencies is relatively large when P/D is small. As P/D
increases, the two peaks gradually merge into one. The presence of two spectral peaks suggests the existence of two different
shedding processes in these cases. The high-frequency component (as indicated by the thick line) for Cy-2, Cy-3 and Cy-4
roughly equals to the frequency of Cy-1 at the same pitch ratio. This is consistent with the PIV results that Cy-2, Cy-3 and Cy-4
are more or less affected by periodic vortex shedding from Cy-1 that is located relatively upstream. The low-frequency peak (as
indicated by the dashed line) on Cy-2, Cy-3 and Cy-4, on the other hand, is likely associated with the vortex shedding from each
190 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

Fig. 9. Time history of dynamic drag (CD) and lift (CL) coefficients on each cylinder in the case of diamond configuration (α ¼451) at P/D¼ (a) 2; (b) 3;
(c) 4; and (d) 5.

individual cylinder itself. A closer examination of the spectra for Cy-1 at very small pitch ratios (e.g., P/D¼2.5) shows that just
like Cy-2 or Cy-4, a peak, albeit rather obscure, is still discernible at about 5 Hz, as highlighted by the black arrow in Fig. 10(a).
This suggests that the flow interference is significant when the cylinders are in close proximity to one another. Although Cy-3 is
impinged upon by the vortices shed from Cy-1, the effect of vortex shedding from Cy-2 and Cy-4 is also ‘felt’ by Cy-1.
The vortex shedding characteristics can be inferred from the amplitude of the spectral peaks as well. The amplitude of
the primary (higher) peak for each cylinder increases with P/D, indicating that vortex shedding becomes more periodic and
regular. However, when the secondary (lower) peak exists, there is a substantial decrease in the amplitude of the primary
peak probably due to the competition between the two different vortex shedding modes. For example, the low-frequency
(secondary) peaks for Cy-2 and Cy-4 at P/D¼2 or 2.5 are even more energetic than the high-frequency (primary) peaks.
As shown in Fig. 10(a), the primary peaks for P/D¼2.5 are about 6.9 Hz. Therefore, the temporal evolution of vortex
formation and shedding for this case is depicted in Fig. 11 by a time series of PIV snapshots sampled at a frequency of 15 Hz
(more than twice the vortex shedding frequency). It is clearly revealed that the formation and shedding of vortices in each
row are quite periodic and regular. For Cy-1, each subsequent snapshot corresponds to one new vortex shed alternately from
the two sides of the cylinder (namely, B1 at t¼ 0 s, A1 at t ¼ð1=15Þ s, B2 at t ¼ð2=15Þ s, A2 at t¼ ð3=15Þ s, and B3 at
t ¼ð4=15Þ s), confirming that the shedding frequency is approximately 7.5 Hz (in agreement with the actual value 6.9 Hz).
The first four snapshots (Fig. 11(a)–(d)) roughly depict one complete cycle of vortex shedding from Cy-2 or Cy-4, since the
locations of C3 and D2 in Fig. 11(d) are almost the same as those of C2 and D1 in Fig. 11(a) (similarly for H2 and G2 in Fig. 11
(d) vs. H1 and G1 in Fig. 11(a)). Therefore, the vortex shedding frequency for Cy-2 or Cy-4 can be estimated to be about 5 Hz,
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 191

Fig. 10. Power spectra of dynamic lift coefficients on each cylinder at different pitch ratios in the case of diamond configuration (α¼ 451): (a) Cy-1; (b) Cy-2;
(c) Cy-3; and Cy-4. The thick line represents the fundamental (higher) frequency, whereas the dashed line indicates the secondary (lower) frequency.

because the duration for these four snapshots is ð3=15Þ s. Similarly, we can estimate the vortex shedding frequency of Cy-3
by observing the evolution of vortices in its wake (e.g., E1 and F1). It is evident that E1 is synchronized with the lower row of
vortices from Cy-4 (i.e., H1 and H2), whereas F1 is synchronized with the upper row of vortices from Cy-2 (i.e., C1 and C2).
However, instead of one-to-one synchronization, a single vortex, E1 (or F1), is formed and shed in association with two
successive vortices in the same row, namely H1 and H2 (or C1 and C2). Consequently, the vortex shedding frequency of Cy-3
is half that of Cy-2 or Cy-4, namely 2.5 Hz. This frequency is also discernible in the lift spectrum for Cy-3 at P/D¼2.5 (Fig. 10
(c)) as the low-frequency component. This time series of instantaneous snapshots also reveal that the vortices are only
apparent within the cylinder array or immediately downstream of the cylinders, and quickly become indiscernible further
downstream due to strong merging amongst themselves and/or interaction with the cylinders.
Fig. 12 shows the variations of St with P/D for the cylinder array at α¼451, together with those reported by Lam and Lo
(1992) and Lam et al. (2003b). The data for each cylinder are grouped into the “lower” and “upper” branches that correspond
to the low- and high-frequency components, respectively. As seen in Fig. 12(a), the data for the upper branch of Cy-3
192 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

Fig. 11. A time series of PIV snapshots showing the instantaneous flow fields around the cylinder array in diamond configuration (α ¼451) at P/D¼ 2.5 and
t ¼ (a) 0 s; (b) (1/15) s; (c) (2/15) s; (d) (3/15) s; and (e) (4/15) s. The vortices shed from each row (either side of the cylinder) are denoted in chronological
order of genesis, e.g., A1, A2,… from the upper side of Cy-1, see inset table.

Fig. 12. Variation of Strouhal number (St) on each cylinder versus P/D for the cylinder array in diamond configuration (α ¼ 451): (a) Cy-1 and Cy-3; (b) Cy-2
and Cy-4.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 193

(St3_upper) collapse well with the upper branch of Cy-1 (St1_upper), forming a single curve. The curve decreases sharply from
StE0.4 over the range of P/Dr3.5 and thereafter gradually approaches the corresponding value of StE0.19 for a single
cylinder. However, the lower branches of Cy-1 and Cy-3 have different characteristics. While St1_lower is located at the low-P/D
end with a relatively high value (StE0.25), St3_lower occurs in the high-P/D range and increases slightly but steadily from StE0.1
to 0.15.
The results for the two flanking cylinders (Cy-2 and Cy-4) are shown in Fig. 12(b), indicating that the data collapse is good
except for the upper branch at the smallest pitch ratio (P/D ¼2), namely St2_upper ¼0.41 vs. St4_upper ¼ 0.34. The difference in
St values is due to the bi-stable flow pattern in the combined wake of the four-cylinder array at very small pitch ratios.
For Cy-2 and Cy-4, the lower branch is very short (P/Dr2.5) and considerably lower than the upper branch.
Generally, the present data agree well with those of Lam and Lo (1992) and Lam et al. (2003b). However, some
discrepancies can be found particularly in terms of the length of the lower branch. Specifically, Lam and Lo (1992) reported a
significantly longer lower branch than the present study and Lam et al. (2003b), which may be attributed to the difference in
the methodology used to calculate the frequencies, namely from spectral analysis of velocity signal in Lam and Lo (1992) vs.
lift force signal in the other two studies. On the other hand, all upper branches of Cy-2, Cy-3 and Cy-4 collapse quite
satisfactorily.
The force coefficients for the cylinder array in diamond configuration (α ¼451) at P/D ¼3.5 are listed in Table 4 together
with published data at similar pitch ratios (P/D ¼3.4–4), whereas the St data are given in the right column of Table 3. There
is an obvious lack of experimental data, particularly the lift coefficient (both mean and RMS), on this flow configuration.
It can also be seen that there is a large discrepancy in the values of C D3 and St, which is again attributed to the large variation
in flow conditions and measurement techniques in these studies.

3.3. Effects of incidence angle (α¼ 0–451)

The variations of St with P/D for the four-cylinder array arranged at α¼ 151 and 301 are presented in Fig. 13 to examine the
effects of the incidence angle. Also included are the data reported by Lam et al. (2003b). For the case of α¼151, some
difference between the two studies is evident at P/Do3.5: while Lam et al. (2003b) showed a large scatter of St values for
each cylinder, the present results indicate an absence of reliable spectral peaks in the lift signal. At large pitch ratios
(P/DZ3.5), however, distinct peaks are found for each cylinder and collapse well on an identical frequency of St E0.19. For
α¼301 (Fig. 13(b)), the present study shows the appearance of the lower branch for Cy-3, which is considerably shorter and

Table 4
Comparison with published mean drag, mean and RMS lift coefficients on four cylinders in diamond configuration (α ¼451).

Researchers L/D C D1 C D2 C D3 C D4 C L1 C L2 C L3 C L4 C ′L1 C ′L2 C ′L3 C ′L4

Sayers (1988, 1990) 4 1.09 0.44

Lam and Fang (1995) 3.58 0.993 1.295 0.648 1.303
Lam et al. (2003b) 3.4 1.11 1.45 0.47  0.021 0.057 0.025 0.113 0.326 0.784
Present 3.5 1.24 1.42 0.89 1.34 0.026 -0.037 0.041 0.038 0.057 0.124 0.341 0.143

Fig. 13. Variation of Strouhal number (St) on each cylinder versus P/D for the cylinder array at α¼ (a) 151; (b) 301.
194 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

lower than that in Lam et al. (2003b). On the other hand, the St data for Cy-1, Cy-2 and Cy-4 and the upper branch of Cy-3
collapse well into a single curve. The curve decreases sharply at the small-to-moderate pitch ratios (P/D r3.5) and thereafter
asymptotes to the single-cylinder value of StE0.19.
The sensitivity of St to the incidence angle (α) can be deduced by comparing Fig. 13 (α¼151 and 301) with Fig. 6 (α¼01)
and Fig. 10 (α¼451). When the four-cylinder array is in in-line or slightly rotated configuration (α¼01 or 151), it may be
considered as two rows of two cylinders in tandem, i.e., Cy-3 and Cy-4 located downstream of Cy-2 and Cy-1, respectively.
When P/D is small, there are no data points due to the weakening or absence of vortex shedding activity. When P/D is large,
St for each cylinder approaches that of a single cylinder. Therefore, only a single value of St (or an identical vortex shedding
frequency) is observed for all cylinders. At large incidence angles (α¼301 and 451), by contrast, a more complex behavior is
seen due to stronger interference among the cylinders. Specifically, when α is large, each cylinder tends to be associated

Fig. 14. A representative snapshot of the instantaneous flow fields for the cylinder array at P/D ¼2.5 and α ¼ (a) 01; (b) 7.51; (c) 151; (d) 22.51; (e) 301;
(f) 37.51; and (g) 451.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 195

with two distinct values (or two branches) of St, initially on Cy-3 only in the case of α¼301, and then on every cylinder in the
case of α¼451. Also notable is that the appearance of distinct St values (corresponding to periodic vortex shedding) occurs at
much smaller P/D, as compared to the case of small incidence angle.
A similar two-St behavior is reported Sumner et al. (2000) and Hu and Zhou (2008) on two cylinders in a staggered
arrangement over a wide range of pitch ratios (1.5rP/Dr4) and incidence angles (101rα r881). The high- and low-St are
attributed to the vortex shedding from the upstream and downstream cylinders, respectively. The relatively higher values of
St for the upstream cylinder arise from increased base pressure due to the presence of the downstream cylinder. The
downstream cylinder, on the other hand, has a lower St because it is more or less immersed in the wake of the upstream
cylinder and is thus subjected to a lower oncoming flow velocity. When the incidence angle is small (αr101), however, the
vortex shedding from the downstream cylinder is synchronized with that of the upstream cylinder and consequently only
one St value is identifiable (Xu and Zhou, 2004). Similar feature is observed for the four-cylinder array at small incidence
angles (α r151). Therefore, vortex shedding from the moderately spaced four-cylinder array can be roughly classified into
two modes depending on α. The first is the one-peak (or single-frequency) mode, since each cylinder has a single (but may
not be identical) frequency. This mode is found when the cylinder array is nearly in-line with the incident flow (αr151). The
other is the two-peak (or two-frequency) mode at relatively large incidence angles (22.51 rαr451), in which stronger
vortex shedding and more complex interaction take place in the gaps between the cylinders.
Fig. 14(a)–(g) shows the effects of varying the incidence angle (α) on the instantaneous flow field around the cylinder
array at a fixed pitch ratio (P/D ¼2.5). It is obvious that the flow is very sensitive to α. At α ¼01 (in in-line square
configuration), the flow is in the reattachment regime. When α is increased to 7.51, the cylinder array can be considered
as two rows of cylinders in slightly staggered arrangement, and Cy-4 and Cy-3 are respectively at a “higher” position in
relative to Cy-1 and Cy-2. The flow structure is similar to the case of α ¼01 in that the downstream cylinders (Cy-4 and
Cy-3) are largely engulfed by the shear layers separated from the upstream ones (Cy-1 and Cy-2), but some difference
can be found between the two cases. At α ¼01 the shear layer reattachment is intermittent. At α ¼7.51, on the other
hand, the upper shear layers separated from Cy-1 and Cy-2 steadily reattach respectively onto the leading surface of
Cy-4 and Cy-3, whereas the lower shear layers pass by from below Cy-4 and Cy-3 (i.e., no reattachment). When α is

Fig. 15. Power spectra of lift coefficient on Cy-1 (black line) and transverse velocity in its wake (red line) for the cylinder array at P/D ¼2.5 and α ¼ (a) 01;
(b) 7.51; (c) 151; (d) 22.51; (e) 301; (f) 37.51; and (g) 451. The velocity signal at the point of (x1, y1) ¼ (1.5D, -0.5D) (inset) was retrieved for analysis.
(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
196 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

further increased, the downstream cylinders are no longer located in the wake of the upstream cylinders. Instead they
displace the upstream-cylinder wake and redirect the flows into the gap regions. At α ¼151 or 22.51 the upper shear
layers from Cy-1 and Cy-2 deflect towards the cylinder centerline with a much shorter length, and reattach on the
lower-left surface of Cy-4 and Cy-3. Consequently, flow-induced separation occurs, as evident by the discrete vortices
periodically shed from the lower surface of Cy-4 and Cy-3. At α Z 301, a jet flow is formed between Cy-1 and Cy-4 (as
well as between Cy-2 and Cy-3) and periodic vortex shedding occurs from each cylinder. The flow interference
is basically of anti-phase type, such as that between the upper shear layer of Cy-1 and the lower shear layer of Cy-4, as
well as that between the upper shear layer of Cy-2 and the lower shear layer of Cy-3. In addition, the interference
between Cy-1 and Cy-3 becomes significant at α¼37.51 or 451, since the vortices shed from Cy-1 largely impinge on Cy-3.

Fig. 16. Variation of the total force coefficients on the cylinder array as a function of α at P/D¼ (a) 2; (b) 2.5; (c) 3; (d) 3.5; (e) 4; and (f) 5.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 197

This is in contrast to the case of small incidence angles (e.g., α¼01 or 7.51) where Cy-3 is free of direct interference with the
wake of Cy-1.
The process of vortex formation and shedding from the cylinders is strongly dependent on the incidence angle (α). As α
increases from 01 to 451, vortex shedding generally becomes stronger as evident by (i) shorter vortex formation length from
each cylinder; (ii) smaller size and spacing of the shed vortices; and (iii) higher magnitude of spanwise vorticity. These flow
patterns are, however, quite different from those in Lam et al. (2003a), which is the only published paper that investigated
the sensitivity of incidence angles (α ¼0–451) on flow structure of the four-cylinder array. The disagreement is, however, not
unexpected due to the significant discrepancy in flow parameters between the two studies, that is, P/D¼4 and Re¼200 in
Lam et al. (2003a) vs. P/D ¼2.5 and Re¼ 8000 in the present study.
Fig. 15 compares the power spectra of lift coefficient on Cy-1 (black line) and those of transverse velocity in its wake (red
line) for the cylinder array with P/D¼ 2.5 at different incidence angles (α¼01–451). In each case, the two spectra agree well
up to 7.5 Hz (half the PIV's sampling rate of 15 Hz), suggesting that the prominent peaks in the lift spectra indeed
correspond to vortex shedding from the cylinder. The spectral peak is not obvious for αr22.51 (Fig. 15(a)–(d)), which is also
consistent with Fig. 14 that periodic vortex shedding from Cy-1 can only be observed at large incidence angles (αZ301).
In the case of α¼451 (Fig. 15(g)), one can observe that except for the prominent peak at about 7 Hz, there is an additional
peak at about 5 Hz in the lift spectrum (as highlighted by the black arrow). This additional peak is attributed to the
reciprocal effects of vortex shedding from Cy-2 and Cy-4 as already explained in Fig. 10, and hence does not exist in the
velocity spectrum.
Fig. 16 shows the variation of mean drag and lift coefficients as a function of α for the four-cylinder array as a whole
group at various pitch ratios. For each P/D, the total mean drag coefficient (C D ) increases with α initially before
attaining a local maxima within the range of α ¼0–451. As P/D increases, the position for peak C D shifts leftward (or
lower α). The maximum values of C D , however, remain roughly constant for P/D ¼2–4, but increase notably at the
largest pitch ratio (P/D ¼5). A similar convex profile is observed for C D0 with the maxima also shifting to lower α with
increasing P/D, but the peak position does not match with that of C D for each case. The variation of the total mean lift
coefficient (C L ), on the contrary, always features a concave shape occurred at α E 10–201, indicating that there is always
a negative mean lift on the cylinder array. The minimum value of C L ¼ 1.5 is found at P/D ¼2.5 and α ¼151 (Fig. 16(b)).
The negative C L is believed due to the high-velocity jet flows formed between the cylinders, which are directed
downward. The shape of C L0 curve, however, varies rather drastically with P/D, from a concave at P/D r 2.5 to a convex
at P/D Z 3. With the increase of P/D, the magnitude of C L0 increases with the peak position progressively shifted to lower
α. The maxima of C L0 ¼1.3 is achieved at α ¼01 and P/D ¼ 5.
The variation of the total mean drag and lift coefficient (C D and C L ) with P/D at different incidence angles is plotted in
Fig. 17. The behavior of C D could be broadly grouped into three categories depending on the value of the incidence angle (α).
When the incidence angle is small (α¼01, 7.51), C D remains roughly constant until P/D r3.5 and thereafter increases with
P/D, due to flow transition from reattachment regime to impinging regime. At moderate incidence angles (α¼151, 22.51),
C D increases monotonically and almost linearly with P/D. At large incidence angles (α¼ 301, 37.51, 451), by comparison,
C D remains roughly constant at a value slightly more than four times that of a single cylinder (C D0 ). Fig. 17(b) indicates that
at α¼01 or 451, C L is nearly zero because the flow is basically symmetric about x-axis. At α¼151, very significant negative lift
forces (with a magnitude of C L   1.5) are observed for all pitch ratios, with the magnitude slightly decreasing with the
increase of P/D. Similar negative lift forces are observed for the case of α¼ 301, but only occur at small to moderate pitch

Fig. 17. Variation of the total force coefficients on the cylinder array as a function of P/D at different incidence angles: (a) total mean drag coefficient (C D )
and (b) total mean lift coefficient (C L ).
198 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199

ratios (P/D r3.5) with a lower magnitude than that of α¼ 151. These results confirm the conclusion by Lam and Fang (1995)
that a slight change in the incidence angle (7151) might significantly affect the lateral (or transverse) structural vibration of
the four-cylinder array.

4. Concluding remarks

The flow around a four-cylinder array arranged in a square configuration at Re ¼8000 was investigated extensively for 6
pitch ratios (P/D¼ 2–5) and 7 incidence angles (α¼0–451), making 42 combinations in total. The fluid dynamic drag and lift
forces on the cylinders were measured using piezoelectric load cells installed on each cylinder, in conjunction with flow
measurements using digital Particle Image Velocimetry (PIV) to quantitatively visualize the vortex formation, shedding and
interaction process. The pitch ratio and the incidence angle have important combined effects on the flow around the four
cylinders either as a whole group or individually. The major findings are summarized as follows.
In the case of in-line square configuration (α¼01), the flow patterns could be broadly divided into three types with
reference to P/D, namely, shielding regime (P/Dr2), reattachment regime (2.5 rP/Dr3.5), and impinging regime (P/DZ4).
This is consistent with the previously published findings. The downstream cylinders generally experience a lower mean
drag, but assume remarkably higher RMS drag and lift (as high as 6–7 times in the vortex impinging regime) than the
upstream cylinders. Also, there exists a repulsive force between the upper row and lower row of cylinders at small P/D.
However, the above classification is valid only when the cylinder array is arranged nearly in-line with the incident free
stream, since the flow characteristics are very sensitive to α as well. Based on the total mean drag coefficient (C D ) on the
cylinder array as a whole, three categories can be identified with respect to α: (I) at small-α (01, 7.51), C D remains roughly
constant until P/Dr3.5, and thereafter increases sharply with P/D due to flow transition to vortex impinging regime; (II) at
moderate-α (151, 22.51), C D increases monotonically with P/D; and (III) at large-α (301, 37.51, 451), C D remains roughly
constant. For Categories I (small-α) and III (large-α), the total mean lift on the cylinder array is about zero (C L E0), because
the flow is basically symmetric about the x-axis. For Category II (moderate-α), on the other hand, significant negative lift
(with a peak magnitude of C L E  1.5) is found particularly at small and moderate pitch ratios due to the high-velocity jets
formed within the cylinder array that are directed downward.
The results show that not only do the cylinders shed vortices individually, but interactions amongst themselves and with
the cylinders also occur. In the case of very small incidence angle (αE01), the vortex shedding patterns experience a
discontinuous jump near the critical gap ratio for flow transition from reattachment regime to impinging regime. As a result,
the spectrum of lift on the downstream cylinders always has a peak at St E0.19, while for the upstream cylinders the peak
can only be observable at large enough pitch ratios (P/D 43.5). Moreover, the St behavior exhibits two different modes
depending on the value of α, namely, a one-peak mode at small- and moderate-α, and a two-peak mode at large-α. In the
two-peak mode, the high- and low-frequency peaks are attributed to vortex shedding from the upstream and downstream
cylinders, respectively. As α increases from 01 to 451, vortex shedding from the cylinders generally becomes stronger as
evident by (i) shorter vortex formation length from each cylinder; (ii) smaller size and spacing of the shed vortices; and (iii)
higher magnitude of spanwise vorticity.
Furthermore, the correlation between the fluid forces and flow patterns is explored in detail to gain more insight into the
flow mechanism. The concurrently-measured dynamic forces on the cylinders show that the flow interference is complex,
which, depending on the combination of α and P/D, could be non-synchronous, quasi-periodic or synchronized with a
definite phase relationship with other cylinders. In the case of inline square configuration (α¼01) or diamond configuration
(α¼451), vortex shedding from all the four cylinders is fully synchronized only at very large pitch ratios (e.g., P/D Z5), with
an almost constant and identical frequency, as well as with a definite phase relationship with one another. At P/D o4, by
contrast, vortex shedding is generally non-synchronized. However, at the smallest pitch ratio (P/D¼2), some “partial”
synchronization occurs, for example, anti-phase between CL3 and C L4 for the inline square configuration (α¼01), and anti-
phase between CD2 and C D4 for the diamond configuration (α¼451).

Acknowledgment

Funding supports from the Singapore National Research Foundation (NRF) through the Competitive Research Programme
(CRP) Project No. NRF-CRP5-2009-01 and Maritime and Port Authority (MPA) of Singapore on this project are gratefully
acknowledged.

References

Afgan, I., Kahi, Y., Benhamadouche, S., Sagaut, P., 2011. Large eddy simulation of the flow around single and two side-by-side cylinders at subcritical
Reynolds numbers. Physics of Fluids 23, 075101.
Alam, M.M., Moriya, M., Sakamoto, H., 2003. Aerodynamic characteristics of two side-by-side circular cylinders and application of wavelet analysis on the
switching phenomenon. Journal of Fluids and Structures 18, 325–346.
Bearman, P.W., 2011. Circular cylinder wakes and vortex-induced vibrations. Journal of Fluids and Structures 27, 648–658.
Esfahani, J.A., Be Hagh, A.R.V., 2010. A Lattice Boltzmann simulation of cross-flow around four cylinders in a square arrangement. In: Proceedings of ASME
2010 10th Biennial Conference on Engineering Systems Design and Analysis, ESDA2010 4, pp. 109–115.
 X.K. Wang et al. / Journal of Fluids and Structures 43 (2013) 179–199 199

Farrant, T., Tan, M., Price, W.G., 2000. A cell boundary element method applied to laminar vortex-shedding from arrays of cylinders in various
arrangements. Journal of Fluids and Structures 14, 375–402.
Hu, J.C., Zhou, Y., 2008. Flow structure behind two staggered circular cylinders. Part 1. Downstream evolution and classification. Journal of Fluid Mechanics
607, 51–80.
Khalifa, A., Weaver, D., Ziada, S., 2012. A single flexible tube in a rigid array as a model for fluidelastic instability in tube bundles. Journal of Fluids and
Structures 34, 14–32.
Lam, K., Fang, X., 1995. The effect of interference of four equispaced cylinders in cross flow on pressure and force coefficients. Journal of Fluids and
Structures 9, 195–214.
Lam, K., Gong, W.Q., So, R.M.C., 2008. Numerical simulation of cross-flow around four cylinders in an in-line square configuration. Journal of Fluids and
Structures 24, 34–57.
Lam, K., Li, J.Y., Chan, K.T., So, R.M.C., 2003a. Flow pattern and velocity field distribution of cross-flow around four cylinders in a square configuration at a
low Reynolds number. Journal of Fluids and Structures 17, 665–679.
Lam, K., Li, J.Y., So, R.M.C., 2003b. Force coefficients and Strouhal numbers of four cylinders in cross flow. Journal of Fluids and Structures 18, 305–324.
Lam, K., Lo, S.C., 1992. A visualization study of cross-flow around four cylinders in a square configuration. Journal of Fluids and Structures 6, 109–131.
Lam, K., Zou, L., 2007. Experimental and numerical study for the cross-flow around four cylinders in an in-line square configuration. Journal of Mechanical
Science and Technology 21, 1338–1343.
Lam, K., Zou, L., 2009. Experimental study and large eddy simulation for the turbulent flow around four cylinders in an in-line square configuration.
International Journal of Heat and Fluid Flow 30, 276–285.
Lam, K., Zou, L., 2010. Three-dimensional numerical simulations of cross-flow around four cylinders in an in-line square configuration. Journal of Fluids and
Structures 26, 482–502.
Norberg, C., 2003. Fluctuating lift on a circular cylinder: review and new measurements. Journal of Fluids and Structures 17, 57–96.
Rajagopalan, S., Antonia, R.A., 2005. Flow around a circular cylinder-structure of the near wake shear layer. Experiments in Fluids 38, 393–402.
Sakamoto, H., Haniu, H., Obata, Y., 1987. Fluctuating forces acting on two square prisms in a tandem arrangement. Journal of Wind Engineering and
Industrial Aerodynamics 26, 85–103.
Sayers, A.T., 1988. Flow interference between four equispaced cylinders when subjected to a cross flow. Journal of Wind Engineering and Industrial
Aerodynamics 31, 9–28.
Sayers, A.T., 1990. Vortex shedding from groups of three and four equispaced cylinders situated in a cross flow. Journal of Wind Engineering and Industrial
Aerodynamics 34, 213–221.
Slaouti, A., Gerrard, J.H., 1981. An experimental investigation of the end effects on the wake of a circular cylinder towed through water at low Reynolds
numbers. Journal of Fluid Mechanics 112, 297–314.
Sumner, D., Wong, S.S.T., Price, S.J., Païdoussis, M.P., 1999. Flow behavior of side-byside circular cylinders in steady cross-flow. Journal of Fluids and
Structures 13, 309−338.
Sumner, D., 2010. Two circular cylinders in cross-flow: a review. Journal of Fluids and Structures 26, 849–899.
Sumner, D., Price, S.J., Païdoussis, M.P., 2000. Flow-pattern identification for two staggered circular cylinders in cross-flow. Journal of Fluid Mechanics 411,
263–303.
Tadrist, H., Martin, R., Tadrist, L., Seguin, P., 1990. Experimental investigation of fluctuating forces exerted on a cylindrical tube (Reynolds numbers from
3000 to 30 000). Physics of Fluids A 2, 2176–2182.
Tong, F., Cheng L., Zhao, M., Chen, X.B., 2011. Three dimensional numerical simulation of flow around four circular cylinders in an in-line square
configuration. In: Proceedings of the 6th International Conference on Asian and Pacific Coasts (APAC 2011), Hong Kong, China.
Wang, X.K., Hao, Z., Tan, S.K., 2013. Vortex-induced vibrations of a neutrally buoyant circular cylinder near a plane wall. Journal of Fluids and Structures 39,
188–204.
Wang, X.K., Su, B.Y., Tan, S.K., 2012. Experimental study of vortex-induced vibrations of a tethered cylinder. Journal of Fluids and Structures 34, 51–67.
Wang, X.K., Tan, S.K., 2008. Near-wake flow characteristics of circular cylinder close to a wall. Journal of Fluids and Structures 24, 605–627.
West, G.S., Apelt, C.J., 1993. Measurements of fluctuating pressures and forces on a circular cylinder in the Reynolds number range 104 to 2.5  105. Journal
of Fluids and Structures 7, 227–244.
Williamson, C.H.K., 1996. Vortex dynamics in the cylinder wake. Annual Review of Fluid Mechanics 28, 477–539.
Xu, G., Zhou, Y., 2004. Strouhal numbers in the wake of two inline cylinder. Experiments in Fluids 37, 248–256.
Zdravkovich, M.M., 2003. Flow around circular cylinders, vol. 1: Applications. Oxford University Press, Oxford, UK.
Zhao, M., Cheng, L., 2012. Numerical simulation of vortex-induced vibration of four circular cylinders in a square configuration. Journal of Fluids and
Structures 31, 125–140.
Zhou, Y., Yiu, M.W., 2006. Flow structure, momentum and heat transport in a two-tandem-cylinder wake. Journal of Fluid Mechanics 548, 17–48.
Zou, L., Lin, Y.-F., Lam, K., 2008. Large-eddy simulation of flow around cylinder arrays at a subcritical Reynolds number. Journal of Hydrodynamics (Series B)
20, 403–413.
Zou, L., Lin, Y.-F., Lu, H., 2011. Flow patterns and force characteristics of laminar flow past four cylinders in diamond arrangement. Journal of Hydrodynamics
(Series B) 23, 55–64.