Journal of Fluids and Structures 61 (2016) 274–294

Contents lists available at ScienceDirect

Journal of Fluids and Structures

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

Experimental investigation on wake characteristics behind

a yawed square cylinder
X. Lou a, T. Zhou a,n, Y. Zhou b, H. Wang c, L. Cheng a
a
School of Civil, Environmental and Mining Engineering, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009,
Australia
b
Shen Zhen Postgraduate School, Harbin Institute of Technology, PR China
c
School of Civil Engineering and Architecture, Central South University, Chang Sha, PR China

a r t i c l e in f o abstract

Article history: The wake vortical structures of a square cylinder at different yaw angles to the incoming
Received 12 August 2015 flow (α ¼ 0°, 15°, 30° and 45°) are studied using a one-dimensional (1D) hot-wire vorticity
Accepted 23 November 2015 probe at a Reynolds number (Re) of about 3600. The results are compared with those
obtained in a yawed circular cylinder wake. The Strouhal number (StN) as well as the mean
Keywords: drag coefficient (C DN ), normalized by the velocity component normal to the cylinder axis,
Yawed square cylinder follow the independent principle (IP) satisfactorily up to α ¼ 40°. Using the phase-
Vortex shedding averaging analysis, both the coherent and the remaining contributions of velocity and
Phase-average method vorticity are quantified. The flow patterns of the coherent spanwise vorticity (ωz ) display
obvious Kármán vortex streets and their maximum concentrations decrease as α increa-
ses. Similar phenomena are also shown in the coherent contours of the streamwise (u)
and transverse (v) velocities as well as the Reynolds shear stress (uv). The contours of the
spanwise velocity (w) and Reynolds shear stress (uw), however, experience an increasing
trend for the maximum concentrations with increasing yaw angle. These results indicate
an enhancement of the three-dimensionality of the wake and the reduction of vortex
shedding strength as α increases. While general similarities to the wake behind a yawed
circular cylinder are found in terms of flow features, some differences between the two
wakes at different yaw angles are highlighted.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction

One of the main concerns about cylindrical structures immersed in a fluid flow is that the alternate shedding of vortices
from the structures and hence the resultant time dependent forces may lead to structure vibration. With increasing use of
slender structures of a square cross-section in engineering, significant research on square cylinder wakes has been con-
ducted (e.g. Lyn et al., 1995; Nakagawa et al., 1999; Sohankar et al., 1999; Saha et al., 2000, 2003; Oudheusden et al., 2005;
Ozgoren, 2006; Tong et al., 2008; Sheard et al., 2009). In general terms, the description for the flow around a circular
cylinder is also believed to be valid for flow around a square cylinder at Reynolds numbers (Re ¼ U 1 d=ν, where U 1 is the
free streamwise velocity, d is the diameter of the cylinder and ν is the kinematic viscosity of the fluid) less than 100
(Sohankar et al., 1999). However, at high Re, the sharp corners on the square cylinder play a significant role in the evolution

n
Corresponding author.
E-mail address: tongming.zhou@uwa.edu.au (T. Zhou).

http://dx.doi.org/10.1016/j.jfluidstructs.2015.11.017
0889-9746/& 2015 Elsevier Ltd. All rights reserved.
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 275

of flow instabilities and other flow characteristics. The wake of a square cylinder is characterized by fixed separation points,
which are distinct from the continuous-curvature cross-section with oscillating separation points, typically represented by
the circular cylinder (Alam et al., 2011). A numerical study on the wake of a square cylinder by Saha et al. (2003) showed
that at Re of about 175, the three-dimensional structure is formed with intermittent large-scale irregularities as a result of
vortex merging in the spanwise direction or due to the phase variations within the large coherent structures. The values of
the Strouhal number and the time-averaged drag coefficient are closely associated with each other over a given range of
Re ¼ 150–500, reflecting the spatial structure of the wake. Besides, these findings were also in good agreement with the
three-dimensional modes of transition that are well-known in the circular cylinder wake. Through the two-dimensional
(2D) and three-dimensional (3D) unsteady flow simulations past a square cylinder with zero incidence at moderate Rey-
nolds numbers (Re ¼ 150–500), Sohankar et al. (1999) concluded that the transition from the stable 2D laminar shedding
flow to 3D flow appears between Re¼150 and 200. The Strouhal number decreases and the mean drag coefficient increases
with Reynolds number, which is opposite to the circular cylinder. The difference is attributed to the fact that the separation
always occurs at the upstream corners of the square cylinder and it is independent of time, whereas the circular cylinder has
separation points varying with time. A linear stability analysis was employed successfully to analyze the 3D transition of a
square cylinder wake by Blackburn and Lopez (2003) and Blackburn and Sheard (2010). The former found a quasi-periodic
intermediate wavelength mode between two synchronous instability modes with long and short spanwise wavelengths,
which may be a general scenario for three-dimensional instabilities of the time-periodic wakes generated by two-
dimensional bluff bodies. The latter examined the effect of symmetry breaking of the vortex street wake on quasi-periodic
secondary instabilities, where the symmetry is broken by a small fixed rotation of the square cylinder about its axis. It is
found that at a small rotation angle of 7.5°, the quasi-periodic mode is replaced by a subharmonic one which starts to
bifurcate from the two-dimensional base state at moderate rotation angles of 12–26°.
Nakagawa et al. (1999) carried out an experimental study with a laser Doppler velocimetry on the turbulence around a
square cylinder at Re¼3000. The time-averaged and phase-averaged statistics showed that the turbulent intensities on the
centerline of the wake reach their maxima near the stagnation points of the recirculation region. The contributions from the
coherent component of both turbulent kinetic energy and Reynolds shear stress to the time-averaged ones are dominant,
compared with those from the incoherent component. Moreover, the kinetic energy of the free stream flow is transferred
mainly to the coherent structure near the cylinder and the coherent vortices feed the incoherent turbulence at the
downstream region. Saha et al. (2000) conducted hot-wire measurements in the wake of a square cylinder at Reþ8700 and
17,625, showing that the peak on the streamwise velocity spectrum becomes broader along the streamwise direction due to
the energy transported from the time-averaged streamwise velocity to the small-scale structures. The peak of the transverse
velocity spectrum is observable over the whole measured range. However, the broadening spectrum in the far wake
indicates the decay of the coherent structures in the flow field and an evolution towards the fully three-dimensional state.
Using a digital particle image velocimetry (PIV), experiments on a square cylinder wake in an open-circuit wind tunnel for
550 rRe r3400 were conducted by Ozgoren (2006) to characterize instantaneous as well as phase-averaged signals of both
velocity and vorticity in the wake. It was found that the Strouhal number and the wake patterns were dependent on the
cross-section of the cylinder as well as the Reynolds number. Owing to the fixed separation points of a square cylinder, the
flow separation and recirculation regions had an increasing effect on the length scales in both the streamwise and the
transverse directions, i.e., the increase of the size of the wake region and the distance between oppositely signed vortices.
The results also indicated that predominant features of instantaneous and phase-averaged vortices were closely related.
Therefore, both of them can be used as representative characteristics of the flow.
As flow approaches a cylinder at a yaw angle α (the angle between the flow direction and the plane which is perpen-
dicular to the cylinder axis, so α ¼ 0° corresponds to the cross-flow case while α ¼ 90° corresponds to the axial flow case),
the flow velocity component in the cylinder axial direction would affect the vortex shedding characteristics. For a yawed
circular cylinder, it has been found that over a certain range of cylinder yaw angle the Strouhal number and the drag
coefficient, normalized by the velocity component perpendicular to the cylinder, i.e. StN ¼ f 0 d=U N and C DN ¼ F x =(1/2ρU 2N d),
where f 0 is the vortex shedding frequency, F x is the in-line force and U N ¼ U 1 cos α is the velocity component normal to the
cylinder axis, are the same as those when the cylinder encounters a normal incidence flow. This is generally known as the
Independence Principle (IP), the cross flow principle or the cosine law. The validity of IP in steady flow past a yawed circular
cylinder has been investigated extensively both experimentally (Surry and Surry, 1967; King, 1977; Ramberg, 1983; Thakur
et al., 2004; Zhou et al., 2009, 2010; Franzini et al., 2013; Gallardo et al., 2014) and numerically (Marshall, 2003; Lucor and
Karniadakis, 2004; Zhao et al., 2009), even though there are some discrepancies with the angle range over which IP is valid.
In an experimental study on a stationary yawed circular cylinder (α¼0–60°), Ramberg (1983) concluded that the vortex
shedding frequency was greater than that expected from the IP since the measured Strouhal number of the yawed cylinder
was larger than the theoretical value when taking the yaw angle into consideration, which invalidates the IP. Marshall
(2003) investigated the wake dynamics of a yawed circular cylinder in a quasi-two-dimensional idealization with a per-
turbation analysis. It was found that the instability of the vortex street and the breakdown of IP at large yaw angles for fully
three-dimensional flows might be led by the cross-stream vortex sheets and the streamwise flow deficit. Lucor and Kar-
niadakis (2004) numerically studied a yawed circular cylinder by comparing forces, cylinder responses and reduced velocity
for different yawed angles. Their results indicated that for large angle of attack, such as α ¼  60° and 70°, the vortex
shedding angle is smaller than the yaw angle and the IP is not valid. The base pressure is lower and hence the drag
coefficient is higher than the value predicted by the IP. Similarly, a direct numerical simulation for flow past a circular
276 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

cylinder with a yaw angle ranging between 0° and 60° was conducted by Zhao et al. (2009), who found that IP applied very
well to the mean drag coefficient and the Strouhal number over the yaw angles considered.
Apart from the investigations on the validity of IP for stationary circular cylinders mentioned previously, the validity of IP
for vibrating circular cylinders was also studied by Franzini et al. (2013). It seemed that IP is more favorably validated for the
vibrating cylinder wakes than the stationary cylinder wakes. Evidences so far have shown that the failure of IP at large yaw
angles is due to the influence of the axial velocity component on the cylinder boundary layer separation and wake formation
process as well as the three-dimensionality of the wake flow (Zhao et al., 2009; Gallardo et al., 2014). However, there has
been no study on the validity of IP and the effect of cylinder yaw angles on vortical structures for square cylinder wakes. This
information is significant in evaluating the dynamic forces and the vibrating frequency of the yawed square cylinders, which
may trigger failure and damage of the structures if the vortex shedding frequency is close to the natural frequency of the
structure. Therefore, the present study aims to furnish information on the above two aspects. The underlying mechanism for
the failure of the IP will be explored by examining the wake flow features, such as vortex shedding characteristics,
hydrodynamic forces and coherent flow structures. Similar studies on inclined circular cylinder wakes using a 3D multi-
hotwire vorticity probe have been reported previously (Zhou et al., 2009, 2010). Considering the limitations of the spatial
resolution of the 3D vorticity probe, a 1D vorticity probe was used to measure the spanwise vorticity (or the transverse
vorticity by rotating the vorticity probe 90°) in the present study at a Reynolds number of 3600. This allows us to have an
improved spatial resolution of the probe. Apart from the vorticity probe, an extra X-probe is required to provide reference
signals for investigating the large-scale structures. Data analysis is conducted using the phase-averaged method (Kiya and
Matsumura, 1985, 1988) by decomposing the turbulence signals into the time-averaged, coherent and incoherent compo-
nents. This technique has also been used by Matsumura and Antonia (1993) and Zhou et al. (2003, 2010) to investigate the
wake structures behind circular cylinders. With this method, the contributions from the coherent structures to the velocity
and vorticity variances, and Reynolds stresses and their dependence on cylinder yaw angle can be quantified.

2. Experimental system and technique

2.1. Experiment setup

The experiments were conducted in a blower type wind tunnel with a test section of 0.38 m (width)  0.25 m (height) 
2 m (length). Five aluminum square cylinders with the same cross-section of 12.7 mm  12.7 mm were aligned horizontally
at yaw angles α¼ 0°, 15°, 30°, 45° and 60° in the test section and supported rigidly by plasticizers at both ends. Therefore all
cylinders have the same projected length of 0.38 m across the wind tunnel. Due to the apparent three-dimensionality effect
at large yaw angles, which makes the application of Taylor’s frozen hypothesis questionable, reported results at α¼60° will
be limited to the spectrum only in the paper. The free stream velocity U 1 was 4.27 m=s, corresponding to a Reynolds
number of about 3600. Such a moderate Reynolds number was chosen under the consideration of minimizing the spectral
attenuation of the vorticity probe without losing the representativeness of Reynolds number ranges. Besides, Okajima
(1982) showed that the aerodynamic characteristics of a rectangular cylinder are relatively insensitive to Reynolds number.
Especially for a square cylinder wake, the Strouhal number revealed a slight and continuous change around a constant value
over a wide range of Re between 100 and 20,000.
Sketches of the experimental setup are shown in Figs. 1 and 2. The coordinate system is defined such that the x-axis is in
the flow direction, the y-axis is orthogonal to both the flow direction and the cylinder axis, and the z-axis is parallel to the
cylinder axis. The spanwise and transverse vorticity components were measured using a 1D hot-wire vorticity probe across
the wake. The measurement locations were at x/d ¼ 10, 20 and 40. Only results at x/d ¼ 10 will be presented hereafter due
to the limitation of the paper length. The vorticity probe (Fig. 1) consists of an X-probe (hot wires a and b) straddled by a pair
of parallel hot wires (wires c and d). Another X-probe, fixed at y/d ¼ 4–7, was used to provide reference signals for phase-
averaging analysis. Significant effort was made to keep the reference probe at the same streamwise location and as close as
possible (about 3 mm) in the spanwise direction to the vorticity probe so that phase difference between the reference signal
and the signal measured by the vorticity probe could be minimized. The separation between the two parallel wires c and d,
and that between the two inclined wires a and b in the X-probe are 1.5 mm. The hot wires were etched from Wollaston (Pt-
10% Rh) wires with an active length of about 200dw , where dw (  2.5 μm) is the wire diameter. The wires were operated
with in-house constant temperature circuits at an over-heat ratio of 1.5. The probes were calibrated at the centerline of the
tunnel against a Pitot-static tube connected to a MKS Baratron pressure transducer (least count ¼0.01 mm H2O) before and
after each set of the experiments. Angle calibration was performed for the two X-probes over 7 20° in order to get the
effective angle of each wire. While the included angle of each X-wire was around 110°, the effective angles of the two yawed
wires for the vorticity probe and the reference probe were 32° and 43°, 42° and 32°, respectively. Low-pass filtered signals at
a cutoff frequency of 5200 Hz were sampled at a frequency (f s ) of 10,400 Hz using a 16 bit A=D converter. The record
duration was 30 s, corresponding to about 1200 vortices being shed from the cylinder.
Aerodynamic forces on the cylinder at various yaw angles were measured in the wind tunnel using a 6-component load
cell (Gamma F/T 9105-TIF, ATI Industrial Automation). One end of the cylinder was mounted stably to the load cell which
was bolted tightly between two rigid flat plates. The other end of the cylinder was cut with an angle so that it was parallel to
the wind tunnel walls. The gap between the cylinder end surface and the tunnel side wall was about 3–5 mm to minimize
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 277

Fig. 1. Sketches of the spanwise vorticity probe and reference probe arrangement (a) Photo of the vorticity probe, (b) Probe side view, (c) Probe font view
and (d) Sketch of the wake and probe arrangement

the influence of the horseshoe vortex. The sketch of the experimental setup for drag force measurement is shown in Fig. 2
for an intuitive view. Considering the low magnitude of the forces at low Reynolds numbers and the sensitivity of the load
cell, forces were measured at Re¼7700. This should be plausible as the force coefficients are independent of Reynolds
number in the subcritical Reynolds number range. The instantaneous signals of the forces acting on the length of the
cylinder encountered the wind flow were recorded with a frequency of 1000 Hz for 60 s.
Experimental uncertainties for velocity, drag coefficient and vortex shedding frequency measurements were estimated
based on the errors in hot wire calibration, frequency resolution in FFT (Fast Fourier Transform) window size, cylinder yaw
angle and the load cell measurements. The uncertainty for mean velocity U was about 7 2% while the uncertainties for u0 , v0
and w0 , where the superscript prime denotes the root-mean-square (rms) values, were about 74%, 7 5% and 75%,
respectively. The uncertainty of the load cell for the force measurement was about 1.5%. Using propagation of errors, the
uncertainties for Strouhal number and drag coefficient in the present study were estimated to be about 6% and 4%,
respectively.
The spanwise vorticity ωz and transverse vorticity ωy (measured by rotating the probe 90°) were calculated from the
measured velocity signals via.

∂v ∂u Δv Δu
ωz     ; ð1Þ
∂x ∂y Δx Δy
278 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Fig. 2. Sketch of force measurement setup (top view).

∂u ∂w Δu ∂w
ωy     ; ð2Þ
∂z ∂x Δz ∂x
where Δu is the streamwise velocity difference measured by the parallel wires c and d; Δv and Δw are the velocity dif-
ferences between two consecutive readings in the time series measured by the X-wire with separation Δx, which is cal-
culated based on Taylor's hypothesis given by Δx ¼  U c =f s . The convection velocity U c at the vortex center can be identified
with the location of the maximum spanwise vorticity, which was estimated as 0.75, 0.76, 0.79 and 0.86U 1 for different yaw
angles (Table 1). Besides, the velocity gradients Δν/Δx and Δw=Δx were discretized using a central difference scheme to the
time series, e.g. Δν/Δx ¼ (vi þ 1  vi  1 )=(2Δx) to avoid phase shift with that of Δu=Δy (Wallace and Foss, 1995). The
uncertainty for the separation Δy between the two parallel wires c and d is about 74%. In addition, referring to the
uncertainties in measuring u, ν and w, the uncertainty in measuring the vorticity components ωy and ωz using Eqs. (1) and
(2) was estimated to be about 78%.

2.2. Phase-averaged method

The near and intermediate wake regions of the square cylinder are characterized by periodic Kármán vortex streets. This
periodicity persists even for the cylinder with yaw angle of 45° (See Fig. 5). A number of methods have been used to detect
the large-scale vortex structures, such as the spanwise vorticity-based technique (Hussain and Hayakawa, 1987), window
average gradient method (Zhang et al., 2000) and phase-averaging techniques (Cantwell and Coles, 1983; Kiya and Mat-
sumura, 1985, 1986). Cantwell and Coles (1983) studied the entrainment and transport in the near wake of a circular
cylinder using a fly hot-wire technique. Velocity vector patterns and contours of vorticity and Reynolds stresses were
examined using phase-average technique conditioned on the pressure signals measured on the cylinder at 65° from the
front stagnation point. The same phase-averaging method as that of Cantwell and Coles (1983) was also used to study the
properties of the mean recirculation region in the wake generated by circular, elliptic and square cylinders and a normal flat
plate by Balachandar et al. (1997) and to study the low frequency unsteadiness in the wake of a normal flat plate by Najjar
and Balachandar (1998). In the latter paper, three-dimensionality characteristics of the wake flow behind a normal plate
were investigated through numerical simulations at Re ¼ 250 and it was concluded that a balanced interaction between the
spanwise and streamwise vortices was required for the generation of a three-dimensional periodic wake. The authors also
conjectured that the distortion of the spanwise vorticity and the formation of the streamwise vorticity were not perfectly
synchronized, and thus a low-frequency cycle was observed in the wake due to this imbalance.
To avoid phase jitter due to the variations in strength, shape, streamwise spacing and lateral location of the vortices, Kiya
and Matsumura (1985) and Hussain and Hayakawa (1987) used a conditional signal obtained at the measurement location.
A sharp peak in the spectrum of v was detected first and then was used to bandpass filter the v signal to provide the phase
for averaging. This phase-averaging method was applied successfully to study momentum and heat transfer in the inter-
mediate regions of cylinder wakes (e.g. Matsumura and Antonia, 1993; Zhou and Yiu, 2006). It was also used to study the

Table 1
The vortex shedding frequency f 0 , drag coefficient C D , convection velocity U c , and wavelength λ for different cylinder yaw angles.

α (°) f 0 (Hz) CD C DN U c =U 1 λ=d λ=dð cos αÞ

0 44.2 1.80 1.80 0.75 5.73 5.73

15 43.5 1.74 1.86 0.76 5.87 5.67
30 41.3 1.40 1.87 0.79 6.43 5.57
45 36.1 0.98 1.95 0.86 8.01 5.66
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 279

yawed circular cylinder wakes (e.g. Zhou et al., 2010). The feasibility of this phase-averaging method in the far wake, where
spatial jitter is expected to be more severe than that in the near and intermediate wake regions, was studied by Antonia and
Britz (1989). It was concluded that, based on the filtered transverse velocity signal, this phase-averaging method led to
essentially the same conditional flow patterns as obtained by other methods of detecting the organized motion. In the
present study, the strong periodicity of the flow in the near wake of a bluff body at a fixed frequency allows the use of the
phase-averaging analysis. Detailed explanation of the phase-averaged method can be found in Matsumura and Antonia
(1993), Zhou and Yiu (2006) and Zhou et al. (2003, 2010). Briefly, the v signal measured by the reference X-probe was
bandpass filtered with a central frequency set at the vortex shedding frequency f 0 , which can be obtained from the energy
spectrum of v. Using a fourth-order Butterworth filter, both the low- and high-pass frequencies are set as f 0 in order to
preserve the identity of the large-scale vortices. Two phases (ϕ) of particular interest are identified on the filtered velocity
signals vf , namely,
t t 1; i
ϕ¼π ; t 1;i r t r t 2;i ; ð3Þ
t 2;i  t 1;i

t  t 2;i
ϕ¼π þ π; t 2;i r t r t 1;i þ 1 ; ð4Þ
t 1;i þ 1 t 2;i

where t 1;i and t 2;i correspond to the instants when dvf =dt 4 0, vf ¼ 0 and vf =dt o 0, vf ¼ 0, respectively. The interval between
the two phases is set as 0.5f 0 by compressing or stretching and is further divided into 30 equal intervals. The measured
velocity signals of u, v and w from the vorticity probe, as well as the calculated vorticity signals ωz and ωy are phase-
averaged at each y-location. The difference between the local phase and the reference phase is used to determine the phase
shift between the signals. The phase ϕ, ranging from  2π to þ 2π, can be interpreted in terms of streamwise distance and
ϕ ¼ 2π corresponds to the average vortex wavelength λ( ¼ U c =f 0 ). With the difference between the phase of the local v signal
and that of the reference νf signal, the vorticity, velocity and Reynolds shear stress contours can be produced in the (ϕ,
y )-plane (Hereafter the superscript asterisk denotes normalizations by U 1 and/or d).
Following Hussain and Reynolds (1970), an instantaneous turbulence quantity B can be decomposed as the sum of a
time-averaged component B, a coherent (or periodic) component β~ and a remainder βr , viz.,
B ¼ B þ β~ þ βr : ð5Þ
The periodic fluctuation β~ reflects the large-scale coherent structures while the remainder βr refers to the incoherent
structures (Hussain, 1986). As pointed out by Hussain (1986), this separation of the total flow field is conceptually precise,
but operationally non-unique: incoherent turbulence does not consist of only fine-scale turbulence, as is generally pre-
sumed, but may contain large-scale irrotational (perhaps even vortical but irrelevant) motions. The phase-averaged
structure is an average of successive structures at the same phase. Therefore, the structures of the same mode and para-
meter size are the coherent structure and the departure of each instantaneous realization from the average phase denotes
the incoherent turbulence (Hussain, 1986). In the following discussion, we will consistently use remainder or the remaining
component to represent the incoherent turbulence.
The coherent and the remaining contributions to the Reynolds stresses and the vorticity variances can be obtained in
terms of the structural average. Specifically, for each wavelength, the phase-averaged structure begins at sample j1 (cor-
responding to ϕ ¼  π) before ϕ ¼ 0 and ends at sample j2 (corresponding to ϕ ¼ π) after ϕ ¼ 0. The structural average for
contribution of the coherent structures to velocity or vorticity variances, denoted by a double overbar, is defined by
1 Xj2
β~ :
β~
2 2
¼ ð6Þ
j1 þj2 þ 1  j1

The value of j1 ( ¼ j2 ) is 20 so the value of j1 þ j2 þ 1 is approximately the average vortex shedding frequency. Comparably,
2
Eq. (6) can also be used to calculate the contribution from the remaining structures (〈 βr 〉). The structural average 〈 β 〉 of the
2

velocity or vorticity variances may be derived from Eq. (5), viz.,

〈 β 〉 ¼ 〈 β~ 〉 þ 〈 βr 〉
2 2 2
ð7Þ

3. Results and discussion

3.1. Mean and rms velocity and vorticity distributions across the wake

The mean streamwise velocity (U) was measured by both the two single wires and the X-wire. In Fig. 3(a), the mean
velocity distributions across the wake for different yaw angles averaged from the two parallel-probes are presented.
Generally the distributions at all angles are symmetric about the centerline regardless of small discrepancies due to
experimental uncertainties. The velocity defect, which is the difference between the freestream velocity U 1 and the
measured local mean velocity U, is the largest for α¼0°. As α increases, the velocity defect decreases, showing a decrease in
280 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Fig. 3. Velocity and vorticity profiles for different yaw angles.

 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 281

the wake region, which is in the same trend as in a yawed circular cylinder wake (Zhou et al., 2009). However, a wider wake
width is observed in the present square cylinder wake than that in the circular cylinder wake (Zhou et al. 2009), which is in
accordance with a smaller base pressure and hence larger drag coefficient C D , as well as a lower Strouhal number for a
square cylinder (showed in the subsequent section).
The spanwise mean velocity distributions W=U 1 in the present study also experience a significant increase with
increasing yaw angle and yet remain symmetric around the wake centerline in Fig. 3(b). They are less than 0.05 for both
α ¼ 0° and 15°, but increase to 0.12 and 0.16 for α ¼ 30° and 45°, respectively, at the wake centerline. The latter two values
are larger than those found in yawed circular cylinder wakes (Zhou et al., 2009), which are 0.075 and 0.085 for α ¼ 30° and
45°, respectively. As the spanwise velocity W provides a measure of flow three-dimensionality (Alam and Zhou, 2008), a
larger magnitude in W implies a higher degree of three-dimensionality. It was also suggested that the spanwise velocity in
general acts to impair the two-dimensionality of the wake and hence the vorticity strength (Marshall, 2003; Alam and Zhou,
2008). The present result indicates that the three-dimensionality is more apparent in yawed square cylinder wake than that
in yawed circular cylinder wake. As the velocity gradients ∂U=∂y and ∂W=∂y are involved in the mean streamwise vorticity
ωx (  ∂V=∂z  ∂W=∂y) and spanwise vorticity ωz (  ∂V=∂x  ∂U=∂y), the dependence of U and W on cylinder yaw angle as
shown in Fig. 3(a and b) illustrates a decreased mean spanwise vorticity and an increased streamwise vorticity as α increases
considering the small magnitude of V (results of V are not shown here).
The root-mean-square values (denoted by a superscript prime) of the three velocity components for different yaw angles
are shown in Fig. 3(c–e). All the u=U 1 distributions show double peaks at about y/d ¼ 7 0.8 except for α ¼ 45°. For the u=U 1
distribution at ¼15°, the deviation from that at ¼0° is negligible. When α increases to 45°, u=U 1 decreases significantly
with a decay ratio of about 1.5. Because the separated shear layers are entrained alternately across the centerline, the
variation of v=U 1 reveals a single peak for all yaw angles. For αr 30°, the decrease in the magnitude of v=U 1 is not as
significant as that of u=U 1 . With a further increase of α to 45°, there seems a sudden decrease in both v=U 1 and u=U 1 ,
which is consistent with that of the yawed circular cylinder wakes (Zhou et al., 2009). Similar phenomenon can be observed
from the w=U 1 distribution in Fig. 3(e) where the magnitude of w=U 1 for other angles only experiences a slight decrease as
α varies from 0° to 30°, but a sudden drop appears at α ¼ 45°. These results may imply a sudden change in flow structures
when α changes from 30° to 45°.
The spanwise and transverse vorticity components can be calculated based on the measured velocity signals via Eqs.
(1) and (2). Similar to the distributions of u=U 1 and v=U 1 , a decreasing trend is also observed in the magnitude of ω0z d=U 1
and ω0y d=U 1 as α increases, even though the change from α ¼ 0° to 15° is not significant. This is consistent with the results
for a yawed circular cylinder reported by Zhou et al. (2009), who showed that the increase of yaw angle results in a decrease
of the three vorticity components. Whereas the rms values of ωy show symmetric distribution about the centerline as can be
seen from Fig. 3(g), an asymmetry is observed in the variation of ω0z in Fig. 3(f) and both demonstrate comparable mag-
nitude. A similar asymmetric distribution of is observed at α ¼ 60° (results are not shown here) as well, which may be
attributed to the special wake structures behind a square cylinder at large yaw angles. This might be of interest for future
study using optical method, such as three-dimensional PIV. It needs to be noted that as the measurements of ωz and ωy
involve the evaluation of velocity derivatives, which should be replaced by the finite differences, the spatial resolution of the
probe due to spectral attenuation will always be a concern (Wyngaard, 1969; Antonia et al., 1996). Using the method
proposed by Antonia et al. (1996), the spectral attenuation of the vorticity probe in measuring the rms vorticity values can
be corrected if local isotropy is assumed. When α increases from 0° to 45°, the spectral attenuation of the probe in measuring
the velocity gradients ∂u=∂y and ∂w=∂z improves as the Kolmogorov length scale η on the wake centerline increases from
 1=4
0.11 mm to 0.13 mm, which is defined as η ¼ v3 =hϵi , where hϵi is the mean energy dissipation rate and can be estimated
using Taylor's hypothesis. It is estimated that the root-mean-square values of the spanwise and transverse vorticity have
been underestimated by about 28–22% for α ¼ 0° to 45°. Therefore, the reduction of ω;z d=U 1 and ω;y d=U 1 as α increases
should represent a genuine decrease of the vorticity. The above results on the reduction of the rms values for both velocity
and vorticity components with the increase of α are related to the instability of the wake vorticity and the decrease of the
spanwise vortex strength.
It also needs to be noted that the calculation of ωy and ωz invokes Taylor's frozen hypothesis when evaluating ∂w=∂x and
∂v=∂x. For any variable in turbulence, the total derivative is equal to zero if frozen hypothesis is used. Then the above velocity
gradients can be quantified as
∂v ∂v V ∂v W ∂v
¼   ð8Þ
U U∂t ∂x U ∂y U ∂z

∂w ∂w V ∂w W ∂w
¼   : ð9Þ
U U∂t ∂x U ∂y U ∂z

Assuming that the flow is entirely in the x-direction, then the velocity gradient ∂α=∂x can be replaced by  ∂α=ðU∂t Þ,
where α represents either v or w. Whereas this simplification is acceptable for quasi-two-dimensional flows, it may
introduce errors when the flow is three-dimensional. From Fig. 3(b), it can be seen that the spanwise mean velocity W=U 1
are about 12% and 16% around the centerline for α¼30° and 45°, respectively, indicating the enhanced 3D effect. Considering
the errors caused by using Taylor's hypothesis, the uncertainties in measuring Δy and the velocity components and the
282 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

spectral attenuation on velocity gradients, it is estimated that the errors in the measured vorticity components ωy and ωz are
about 30%.
It is pertinent to provide some more discussion on the application of Taylor's hypothesis. It is widely accepted that if the
velocity gradients in the streamwise direction need to be quantified from single point measurements, e.g. using hot wires,
Taylor’s hypothesis has to be invoked. To assess the reliability of this hypothesis, extensive studies have been reported in the
literatures so far, for example, Cenedese and Romano (1991) for highly turbulent flow, Zaman and Hussain (1981) and Mi
and Antonia (1994) for turbulent jet flows, Hussain and Hayakawa (1987) and Dahm and Southerland (1997) for turbulent
wakes, Uddin et al. (1997) and Dennis and Nickels (2008) for turbulent boundary layers, and Schlipf et al. (2010) for
atmospheric boundary layer. It is true that Taylor’s hypothesis will be well satisfied if the turbulent field remain unchanged,
i.e. completely “frozen”, and is just advected by the mean flow velocity. Previous studies found that Taylor's hypothesis
works satisfactorily in many situations outside the original assumption. For example, using LDA technique, Cenedese and
Romano (1991) found that even at high turbulent levels Taylor’s hypothesis is substantially correct for small separations,
especially when examining the large-scale turbulent structures, which travel with smaller velocity than the mean flow
velocity. The validity of Taylor’s hypothesis in wall turbulence was also studied by Uddin et al. (1997) at Re ¼ 2613 and it was
concluded that Taylor's hypothesis can be used with confidence for a turbulent boundary layer with a constant convection
velocity along the whole region since the eddies in the outer portion of the boundary layer were convected more slowly
than the mean velocity, which is in accordance with the finding of Krogstad et al. (1997) who showed that different
wavenumbers or scales of vortex structures convect at different velocities. The hypothesis only breaks down in the region
very close to the wall. Dennis and Nickels (2008) studied the applicability of Taylor's hypothesis in constructing long
meandering vortex structures reported by Hutchins and Marusic (2007) in a turbulent boundary layer where the mean-
dering long structures were realised using a rake of hot wires by invoking Taylor’s hypothesis. In the study by Dennis and
Nickels (2008), the velocity fields constructed using Taylor’s approximation were compared with those obtained using time
resolved PIV in the turbulent boundary layer. A visual examination of the plots of the streamwise velocity fields showed a
striking similarity, although some differences in the large-scale structures were also evident. The vast majority of the dif-
ferences occurred in the latter half of the field. A quantitative analysis was performed by correlating the appropriate spatial
and the Taylor fields. It was found that there was a strong correlation between the two fields and this correlation decreased
approximately linearly with projection distance. It was proposed that a significant proportion of the error between the two
fields was due to small-scale motion which was not maintained in time, and as such, was a source of consistent error. A
method of Fourier filtering was used to filter out the small scale structures. The correlation was found to increase sign-
ificantly when the filtered fields were used. They concluded that Taylor's approximation can be used to a greater extent if it
is only the large-scale structures that are of interest (as in the present study). However, care must be taken since excessive
filtering will distort the actual structures and lead to erroneous results.
The time traces of the transverse velocity v signals, as well as the spanwise vorticity ωz signals for α ¼ 0° and 45° obtained
at about y=d ¼ 0.5 (corresponding to the vortex center) are depicted in Fig. 4. The v signals show a large-scale quasi-periodic
variation with a relatively fixed frequency in both cases, indicating the occurrence of the Kármán vortices even though the
periodicity at α ¼ 45° is not as regular as that at α ¼ 0°. Based on the conditional analysis, Zhou and Antonia (1993) found a
correspondence between the vortex center of the spanwise vortices (negative sign), which is located at the maximum
vorticity concentration, and the zero values of the v signals at positive dv=dt. Therefore, the zero values on the v signals with
positive dv=dt, as marked by the dotted vertical lines in Fig. 4, are identified with the possible Kármán vortex centers. At
α ¼ 45°, vortex shedding seems to be suppressed in the wake due to the less often occurrence of positive dv=dt on the v
signals. Within each shedding period, the vorticity signals at α ¼ 0° fluctuate much stronger than that at α ¼ 45°. This result
can also be verified by the lower peak magnitude on the energy spectra compared with that at α ¼ 0° (shown later in Fig. 5).
The fluctuation of the ωz signals is much more significant than that of the v signals, which means that the vorticity signals
represent mainly the small-scale structures which are more intermittent (Zhou et al., 2010).

3.2. Dependence of vortex shedding frequency and drag coefficient on cylinder yaw angle

The transverse velocity v is more sensitive to the organized structure than other velocity components. Therefore, the
spectra of v measured on the wake centerline for various cylinder yaw  angles
2 are chosen here and shown in Fig. 5. The
R1
power spectral density functions ϕv is defined such that 0 ϕv ðxÞdx ¼ v=U N , which is obtained by calculating FFT of the
velocity signals. For convenience of viewing and comparing the spectra at different yaw angles, the spectra for α ¼ 15°, 30°,
45° and 60° have been shifted downward by one, two, three and four orders, respectively. An apparent peak can be found on
each spectrum for αr 45°, denoting the vortex shedding frequency f 0 . When the frequency f is normalized by d and U N , i.e.
f d=U N , the normalized peak frequency f 0 d=U N on the spectrum corresponds to the Strouhal number StN . According to Dutta
et al. (2003), the dimensionless flow fields in the wakes of square cylinders are practically independent of Reynolds number
because the separation points are fixed at the sharp corners facing the incoming flow. The values of St in their study were in
the range of 0.141  0.143 for different Re and different section orientations. Saha et al. (2000) also gave the respective
Strouhal numbers of 0.144 and 0.142 at Re ¼ 8700 and 17,625 for a square cylinder in uniform wind flow. Considering the
different Re, aspect ratios of the model and the experimental setup, Ozgoren (2006) obtained a Strouhal number of 0.134 at
Re ¼ 3400 with a square cylinder in a water tunnel. In the present study, the values of are 0.132, 0.134, 0.142 and 0.152 for
α ¼ 0°, 15°, 30° and 45°, respectively. The present result of at α ¼ 0° agrees well with that reported previously for the square
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 283

Fig. 4. Time traces of the fluctuating transverse velocity and the spanwise vorticity measured at about y=d ¼ 0.5 (corresponding to vortex center) for yaw
angles (a) α ¼ 0°; (b) α¼ 45°.

Fig. 5. Spectral density functions of the spanwise velocity ϕv for different yaw angles. Spectra for α Z 15° are shifted by one order for convenient viewing.

cylinder wake. The peak region on the spectrum is sharp and narrow at α ¼ 0°. As α increases, the height of the peak
decreases and the width of the peak region broadens significantly, which denote a reduction of vortex shedding intensity
and a coexistence of the vortex structures with a size comparable to the dominant Kármán vortex structures. This trend is
consistent with that found in the yawed circular cylinder wake (Surry and Surry, 1967; Zhou et al., 2009; Zhao et al., 2009).
In fact, when the yaw angle is large enough, there will be a strong interaction between the spanwise and the streamwise
vortical structures. The shedding of the streamwise vorticity sheets wrap around spanwise vortices and induce a local
284 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Fig. 6. Comparison of and C DN for different yaw angles normalized by St0 and C D0 .

spanwise velocity component, resulting in the instability of the Kármán vortex that becomes constricted and distorted
(Lucor and Karniadakis, 2004). In addition, there is no apparent peak on ϕv at α ¼ 60°, indicating that there is no vortex
shedding at this angle due to the enhancement of the three-dimensional effect of the flow (at which case, the ratio W=U 1
would be around 19% on the centerline of the wake). This result seems to support our extrapolation in Fig. 9.
The comparison of the Strouhal number at different α is shown in Fig. 6, with the ratio StN/St0 against the yaw angles,
where St0 represents the Strouhal number at α ¼ 0°. Owing to the lack of St results for yawed square cylinder wakes, only
related results from the circular cylinder wakes are presented for comparison. For the validity of IP, the ratio StN/St0 should
be equal to 1. Taking the experimental uncertainty into consideration, a tolerance of around 7 6% is anticipated. Results in
Fig. 6 show that for α ¼ 0°, 15°, 30°, the ratio StN/St0 is close to one, supporting the IP with the tolerance of uncertainty.
However, the ratio (  1.15) at α ¼ 45° deviates significantly from 1, which is too big to be attributed to the experiment
uncertainty. The present results, therefore, indicate that the IP is valid for a yawed square cylinder, at least in terms of vortex
shedding frequency, when the yaw angle is smaller than about 40°. This result is in agreement with that found for a yawed
circular cylinder wake (Surry and Surry, 1967; Zhou et al., 2009).
The drag coefficients C DN at different yaw angles, normalized by the value at α ¼ 0°, i.e. C DN =C D0 , are also included in
Fig. 6. Some results of drag coefficient for the square cylinder wake reported previously are 2.2 at Re ¼ 2800 (Davis and
Moore, 1982), 1.97 at Re¼3000 (Kim et al., 2004), 2.13 at Re ¼ 8700 (Saha et al., 2000) and 2.1 at Re ¼ 21,400 (Lyn et al.,
1995). The present drag coefficient at α ¼ 0° (shown in Table 1) is slightly lower than the above values, which could be due
to the imperfect ‘sharp’ edges of the square cylinder used in the present experiment. It can be seen from Table 1 that C D
decreases dramatically with the increase of yaw angle. However, when U N is used for normalization, i.e. C DN ¼ C D =cos2 α, the
deviation of C DN at 15° and 30° from that at α ¼0° is rather small. The ratio of C DN =C D0 is very close to 1 as required by the IP
except at α ¼ 45°, where a discrepancy of 8% is observed. Considering the uncertainties of the load cell (1.5%) in force
measurement, as well as the uncertainties in the cylinder yaw angle, the result of the mean drag coefficient seems to
support the IP more favorably than the Strouhal number for the square cylinder wake.

3.3. Phase-averaged vorticity fields

Prior to the discussion on phase-averaging results, it is relevant to examine if flow reattachment occurs or not, especially
at large yaw angles. In a previous study on unsteady turbulent wake of a rectangular cylinder in channel flow by Nakagawa
et al. (1998), flow visualization at Re¼ 3000 was conducted and it was found that the flow separated at the leading edge and
reattached only when the ratio of width=height Z 2 due to the increased after-body length. In the present study, the
width=height ratio of the cross-section are 1, 1.04, 1.15 and 1.41 for α¼ 0°, 15°, 30° and 45°, respectively. Therefore, there
should be no reattachment observed. To confirm this, flow visualizations at Re ¼ 3600 using smoke technique were also
conducted at α¼0° and 45° and the snapshots are shown in Fig. 7. It can be seen that at α ¼ 0°, the streaklines at the outmost
of the vortices are quite smooth and the vortex formation length is about 1.5d and the contour-rotating vortices in the wake
can be seen clearly. No reattachment is found in the image. At α ¼ 45°, the vortex formation length seems to be enlarged till
about 3.5d and no reattachment is found either. The vortex structures are deformed. The outmost of the vortices are not
smooth but rather there are smaller scale vortices in the mixing layers. Actually even for the rectangular cross-section with
width/height Z 2, Okajima (1982) concluded that the steady reattachment was impossible at moderate Reynolds numbers.
At higher Reynolds numbers, the separated flows detached themselves suddenly from the cylinder surfaces and led to a
wider wake.
Two critical points, namely, foci and saddle points, are important in describing the characteristics of bluff body wake.
These points can be identified from the phase-averaged sectional streamlines (Fig. 8), which are denoted by a plus ( þ ) and a
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 285

1.5d
3.5d

Fig. 7. Smoke wire flow visualization at Re ¼ 3600 for a yawed square cylinder. (a) α ¼ 0°; (b) α ¼ 45°.

Fig. 8. Phase-averaged sectional streamlines at different yaw angles. The dashed lines represent diverging separatrix and the saddles and vortex centers are
marked by cross and plus signs.

cross (), respectively. The diverging separatrix are also shown with thick dashed lines in the figure. The streamlines reveal
apparent Kármán vortex streets for all yaw angles. With the increase of α, both the vortex centers and the saddle points
approach towards the centerline. The iso-contours of the phase-averaged coherent and remaining vorticity ω~ z and ω2 zr for
different yaw angles are presented in Fig. 9. For convenience of visualization and illustration, smoothing is applied for all the
contours. The convection velocity U c for different cylinder yaw angles (Table 1) are estimated with the values of U þ u~ at the
vortex centers. Also included in Table 1 are the normalized wavelength (λ) of the vortical structures for different yaw angles,
which is the streamwise separation between successive vorticity peaks. Seeing from Table 1, λ=d is equal to 5.73 at α ¼ 0°,
which is close to the value of 5.8 found by Lyn et al. (1995) in the near-wake region of a square cylinder and larger than that
(λ=d ¼ 4.16) found for the circular cylinder wake by Zhou et al. (2010). As λ represents the distance that a vortex travels over
one cycle, the ratio between the wavelength of the square and the circular cylinder wakes, i.e. 5.73/4.16 ¼ 1.38, should be
approximately equal to the inversed ratio of the Strouhal numbers of these two types of cylinder wakes, i.e.
0.195=0.132 ¼ 1.4. Lyn et al. (1995) concluded that the relatively larger streamwise length scales implied smaller average
streamwise gradients, which may lead to slower recovery of the mean streamwise velocity U. Meanwhile the larger
streamwise length scale is also consistent with a larger wake width, as mentioned previously, which will result in the slower
recovery as well since a wider wake requires more fluid to be entrained to contribute the larger recirculation area (Alam
et al., 2011). In the present study, the wavelength increases monotonically with the increase of α. The value of λð cos αÞ is
approximately constant for all the angles if a factor of cos α is multiplied by λ. This result indicates that the spanwise
vortices shed from the yawed cylinder are parallel to the cylinder axis.
The coherent spanwise vorticity ω~ z contours in Fig. 9(a–d) display the well-known Kármán vortex street, which are
similar to the shape of the vortex contours in the near wake of a circular cylinder. The foci ( þ ) on the streamlines coincide
well with the vortex centers (the maximum concentration of ω~ z ) on the vorticity contours. The present result at α ¼ 0°
agrees well with the trend shown by Lyn et al. (1995) at x=d ¼ 8, where a maximum contour value of 0.8 was found. Note
that the solid lines represent the positive value and the dashed lines represent the negative values in all the figures. The
maximum concentration of ω~ z at α ¼ 0° is  0.65 (the negative value of the vorticity contours represents the clockwise
rotated vorticity above the centerline), which is the largest among all the cases and in accordance with the pronounced peak
on the spectrum shown in Fig. 5. For α ¼ 15° and 30°, the maximum values of ω~ z decrease to  0.55 and  0.39, respectively.
However, when α increases further to 45°, this value is only  0.14 with a decay ratio of 4.6, which is larger than that of a
yawed circular cylinder (Zhou et al., 2010), of which the maximum ω~ z contour values of 0.8, 0.8, 0.6 and 0.4 for α ¼ 0°, 15°,
30° and 45°, respectively, were reported. With the increase of α, the vortex center shifts towards the wake centerline,
resulting in the reduction of the wake region. The numerical results of a circular cylinder yawed at α ¼ 45° from Zhao et al.
(2009) indicated that, after the streamwise flow passes the cylinder, some of the fluid slides along the spanwise direction for
286 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Fig. 9. Phase-averaged coherent and incoherent spanwise vorticity contours at different yaw angles. ω~ z (contour interval: a, 0.1; b, 0.1; c, 0.05; d, 0.02). ω2
zr
(contour interval: e, 0.5; f, 0.5; g, 0.2; h, 0.2). The dashed lines represent diverging separatrix and the saddles and vortex centers are marked by cross and
plus signs.

Fig. 10. Comparison of decay of the peak spanwise vorticity (ω~ z ) and vorticity circulation Γ* at different yaw angles for square and circular cylinder wakes.
The dash lines denote fitting curves with second order of polynomial. The error bars are included based on the experimental uncertainties in vorticity
measurements.

a certain distance. Once the spanwise vortex is shed from the cylinder surface, the part of flow at the center of the principal
spanwise vortex, which points to the spanwise direction, is convected to the incoming flow direction. This results in the
dispersion of the large-organized vortical structures.
To examine the decay of the spanwise vorticity as α increases, the maximum values of ω~ z at different cylinder yaw angles
are shown in Fig. 10. Also included in the figure are the vorticity circulations Γ  , which are calculated by integrating the
measured spanwise vorticity over an area enclosed by the contours as shown in Fig. 9(a–d),
Z
Γ  ¼ ωz dA: ð10Þ
A

The error bars are estimated based on the errors in measuring the spanwise vorticity and selecting the integration areas.
It needs to be noted that a cutoff value for vorticity has to be applied so that relatively distinct vorticity contours can be
obtained for integration (Lyn et al., 1995). Three cutoff values, namely, 20%, 30% and 40% of the maximum ω~ z contours, are
used. For different cutoff vorticity, the variation of the circulation with respect to the yaw angles shows the same trend. The
circulation values decrease by about 60%, 62% and 54% when the yaw angle increases from 0° to 45° for the 20%, 30% and 40%
cutoff vorticity, respectively. For the same angle, when the cutoff value is increased by 10%, the spanwise vorticity circulation
decreases by about 15%. As the 30% cutoff gives more regular vorticity contour than that of 20%, the former has been used in
the present study. Results obtained in yawed circular cylinder wakes, which are inferred from Fig. 7 in Zhou et al. (2010)
using the same cutoff vorticity, are also included in Fig. 10. The circulations for the square cylinder wake are 1.10, 1.06, 0.87
and 0.42 for α ¼ 0°, 15°, 30° and 45°, respectively, showing a larger decay rate than that in the circular cylinder wake.
Compared with the decay of the coherent spanwise vorticity, the decay of the circulation with respect to the increase of α is
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 287

Table 2
Maximum magnitude of spanwise vorticity and its circulation for square cylinder (SC, present study) and circular cylinder (CC, Zhou et al., 2010) at different
cylinder yaw angles.

α(°) ω~ zmax Γ

SC CC SC CC

0 0.65 0.80 1.10 1.24

15 0.55 0.80 1.06 1.22
30 0.39 0.60 0.87 1.14
45 0.14 0.40 0.42 0.64

Fig. 11. Phase-averaged coherent and incoherent transverse vorticity contours at different yaw angles. ω~ y (contour interval: a, 0.04; b, 0.04; c, 0.04; d, 0.04).
ω2
yr (contour interval: e, 0.2; f, 0.2; g, 0.2; h, 0.2).

more significant for both the circular and square cylinder wakes. An extrapolation with second order polynomial fitting
curve is included in Fig. 10, which confirms our anticipation that the shedding of the large-scale structures will disappear
completely when α ¼ 55–60°, consistent with the energy spectrum at α¼ 60° (Fig. 5). The decrease of the spanwise vorticity
circulation with the increase of yaw angle reflects the decrease of the vorticity strength as well. The normalized values of the
spanwise vorticity and the circulation for both the square and circular cylinder wakes at different yaw angles are listed in
Table 2.
The remaining spanwise vorticity contours ω2 zr in Fig. 9(e  h) display a lower decay rate than the coherent spanwise
vorticity contours, of which the maximum values are 5.1, 4.5, 3.5 and 2.3 for α ¼ 0°, 15°, 30° and 45°, respectively. These
values are much larger than that of the maximum ω~ z contours, suggesting that more contributions to the spanwise vorticity
component come from the small-scale turbulent structures rather than those of the large-scale ones. The magnitudes of ω2 zr
are also much larger than those obtained previously in yawed circular cylinder wake, indicating that the small-scale vortices
in the square cylinder wake are much stronger than that in circular cylinder wakes. In addition, the main axes of the ω2 zr
contours have a tendency to run along the diverging separatrix. Compared with the reduction ratio (1.54) of the maximum
ω2
zr contours for a yawed circular cylinder by Zhou et al. (2010), the ratio (2.22) in the present study is larger. This result is
also consistent with that of the decay of the coherent vorticity, implying that the cylinder yaw angles affect both the
coherent and the remaining structures more in square cylinder wake than that in the circular cylinder wake. Obviously, the
reduction of both the coherent and remaining vortex structures indicates that vortex shedding is weakened significantly
when the yaw angle is changed from 30° to 45°.
The coherent contours of the transverse vorticity ω~ y (Fig. 11) display lower strength than that of ω~ z and almost lose
coherence. With the increase of α, the contours become more organized though the maximum concentration keeps prac-
tically constant. Given the fact that the total magnitude of the measured transverse vorticity is decreasing with increasing
yaw angle, the coherent structures of the transverse vorticity are actually enhanced. The ω2 yr contours in Fig. 11
(e–h) are also less organized than the remaining spanwise vorticity contours. The reduction ratio (2.22) of the maximum
value from α ¼ 0° to 45° is the same as that of ω2 zr , which is higher than that (1.6) for a circular cylinder wake (Zhou
et al., 2010).
288 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

3.4. Phase-averaged velocity and Reynolds shear stress

As indicated in the conclusions by Ozgoren (2006), the phase-averaged streamwise velocity u~  showed that the rolled-
up flows originally starting from the upstream side of the cylinder were conveyed downstream, together with the corre-
sponding Kármán vortex. Eventually they formed a counter-rotating flow pair. Similar features can also be traced for the u~ 
patterns in the present study. The u~  contours depicted in Fig. 12(a–d) are anti-symmetric about y ¼ 0 and symmetric about
ϕ ¼ 0. The maximum concentration occurs at about y ¼ 7 1.2, which is in accordance with the double peaks on the
distribution (shown later in Fig. 14) of the contributions from coherent structures to the streamwise velocity component.
Compared with the results of the circular cylinder wake by Zhou et al. (2010) that the maximum concentration of u~  existed
at about y ¼ 0.75, the present results verify again that the square cylinder has a wider wake region than the circular
cylinder wake. There is only a small variation in u~  as α varies from 0° to 30°, reduced by about 40% from 0.19 to 0.12.
However, the decrease in the maximum u~  concentration becomes much more apparent with a reduction of about 70%
when the yaw angle changes from 30° to 45°, which is about twice that found in the yawed circular cylinder wake (Zhou et
al., 2010).
In agreement with the previous phase-averaged transverse velocity results (Matsumura and Antonia, 1993; Ozgoren,
2006), both positive and negative concentrations of the v~  contours presented in Fig. 12(e–h) appear in a staggered pattern.
Specifically, v~  contours are anti-symmetric about ϕ ¼ 0 and the maximum value occurs on the centerline. The maximum
contour values of v~  are at least about 50% higher than those of u~  for the same cylinder yaw angle, indicating that the
coherent structures contribute more to the transverse velocity component than that of the streamwise one, especially at
large yawed angles. With the increase of α, the maximum values of v~  decrease apparently from 0.33 to 0.08 when α
increases from 0° to 45°, which corresponds well with the coherent contribution shown in Fig. 14. Similar to the decay of u~ 
when α changes from 30° to 45°, the significant decay of v~  is reflected not only by a high decay ratio of 1.25, but also by the
distortion of the contour patterns, indicating the irregular vortex shedding at α ¼ 45°. Compared with the yawed circular
cylinder wake (Zhou et al., 2010), it can be seen that the decay of the maximum v~  contours in the yawed square cylinder
wake is much faster as α increases. This may be attributed to the fact that the wake of a square cylinder has a more
anisotropic nature and a higher value of turbulent intensity, which can be reflected by the maximum value of the v fluc-
tuation on the centerline as well (Durao et al., 1988). Comparing the contours of u~  and v~  in both the square and the circular
cylinder wakes, it seems that the contours in the former wake are less well organized and stretched more along the
diverging separatrix than that in the latter wake at large yaw angles, supporting our former conclusion that vortex struc-
tures deform much faster in a square cylinder wake than that in a circular cylinder wake when α increases.
It can be seen from Fig. 12(i–l) that the maximum magnitude of the w ~  contours is much smaller than that of the u~  and

v~ contours due to the less coherent contributions to w ~ . Similar to the contours of u~  , the contour plots of w

~  display anti-
symmetry about the vortex center but with opposite signs. The slight increase of the maximum concentration and the more
organized contour pattern in w ~  as α increases indicate the generation of the secondary axial organized structures, i.e. the

Fig. 12. Phase-averaged velocity contours at different yaw angles. u~  (contour interval: a, 0.02; b, 0.02; c, 0.012; d, 0.004). v~  (contour interval: e, 0.05; f,
~  (contour interval: i, 0.005; j, 0.005; k, 0.005; l, 0.005). The saddles and vortex centers are marked by cross and plus signs.
0.05; g, 0.05; h, 0.01). w
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 289

Fig. 13. Phase-averaged coherent and incoherent shear stress contours at different yaw angles. u~  v~  (contour interval: a, 0.005; b, 0.0025; c, 0.002; d,
0.0002). ur vr (contour interval: e, 0.002; f, 0.0025; g, 0.0013; h, 0.001). u~  w
~  (contour interval: i, 0.0005; j, 0.0005; k, 0.0005; l, 0.0005). The dashed lines
represent diverging separatrix and the saddles are marked by cross signs.

streamwise vorticity. This cross-stream vorticity component, according to Marshall (2003), is shed from both top and
bottom sides of the cylinder and wrap around the spanwise vorticity, which may lead to the breakdown of the IP at large
yaw angles due to the enhancement of the three-dimensionality in the wake.
The contours of the coherent and remaining Reynolds shear stress for different cylinder yaw angles are presented in
Fig. 13. The u~  v~  contours reveal the well-established clover-leaf pattern, which are nearly anti-symmetric about vortex
centers. Similar to the results for the yawed circular cylinder wake (Zhou et al., 2010), the u~  v~  contours in the present study
are almost zero at the position of vortex centers and present a local extrema in each quadrant. In a reference frame
translating at U c , the clover-leaf pattern (Zhou and Yiu, 2006; Zhou et al., 2010) of the u~  v~  contours from the coherent
velocity field associates with the vortical motion, e.g., for vortex ω~ z o 0, u~  4 0 and v~  o0 in quadrant 1 yields u~  v~  o 0.
When α increases from 0° to 30°, the maximum contour value of u~  v~  decreases by about 66%. Moreover, this magnitude
decreases by about 95% when α increases further to 45°. Referring to Fig. 13(a–c), the u~  v~  contours keep a regular shape
until α ¼ 30°. When α increases to 45°, the contours display an obvious distortion in shape and an apparent reduction in the
wake width. Compared with the u~  v~  contours in the circular cylinder wake (Zhou et al., 2010), the decay of the maximum
contour values is much faster for the square cylinder wake. In Fig. 13(e–h), the ur vr contours display a typical shape with the
main body and the ‘leg’ (Lyn et al., 1995). The ‘leg’ is almost perpendicular to the main body, but parallel to the main body of
the upstream ur vr contours. The ur vr contours show peaks at the saddle points and are stretched with the main axis being
along the diverging separatrix when α r 30°. At α ¼ 45°, the ur vr contours lose their regular shapes and become distorted
and do not appear to cross the centerline. The maximum value of the ur vr contours also decreases as α increases, with a
much smaller decay rate than that of u~  v~  . The double vortex centers at y ¼ 7 1.0 on the ur vr contours for α r 30° are in
correspondence with the contribution variations shown in Fig. 14. This double-peak structures and their evolution resemble
that depicted by Lyn et al. (1995) and Nakagawa et al. (1999) for the square cylinder wake, with the high ur vr regions being
observed near the saddle points. In terms of the maximum concentration for all contours discussed above, a faster decay
with the yaw angle can be found for the square cylinder wake than that for the circular cylinder wake, which means that
cylinder yaw angles have a stronger influence on the wake structure of a square cylinder than that of a circular cylinder.
The coherent contours of the Reynolds shear stress u~  w ~  in Fig. 13(i–l) show much lower strength than u~  v~  . The
maximum concentration of u~  w ~  increases by about 67% as α increases from 0° to 45°. The u~  w
~  contours display negative in
 
most wake regions, which can be explained with Fig. 12 that the contours of u~ and w ~ demonstrate similar patterns yet
opposite signs. The higher maximum concentration and the more organized pattern at α ¼ 45° than that at α ¼ 0° reflect the
generation of the secondary vortex at large yaw angles.

3.5. Contributions from the coherent structures to Reynolds stresses and vorticity variances

The contributions from the coherent and remaining structures to Reynolds stresses and vorticity variances can be
obtained in terms of their structural average according to Eqs. (6) and (7). The time-averaged values, coherent and
290 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Fig. 14. Time-averaged Reynolds stresses and spanwise vorticity variance with coherent and incoherent contributions at different yaw angles. (a–d),
streamwise velocity; (e–h), transverse velocity; (i–l), spanwise velocity; (m–t), Reynolds shear stress.
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 291

remaining components at different cylinder yaw angles are shown in Figs. 14 and 15. When phase-averaging is applied, care
has been taken to make sure that the coherent and remaining contributions together should be no less than 95% of the time-
averaged values so that the flow has been properly represented. For the streamwise velocity, the main contribution comes
from the remaining component, especially on the wake centerline, where the coherent component is nearly zero. This result
agrees well with the previous findings (Matsumura and Antonia, 1993) about the u2 distribution in the wake of a circular
cylinder. The distributions are symmetric about y ¼ 0 and the time-averaged variations display twin peaks for α r 30°,
which corresponds to the rms values shown in Fig. 3. For α ¼ 0°, the coherent contribution is about 50% of the remaining one
at y ¼ 7 1.0 and reduces to 35% and 24% for α ¼ 15° and 30° respectively. When α ¼ 45°, the coherent contribution reduces
to less than 10% of the remaining one, indicating that the large-scale structures begin to dissipate significantly at this yaw
angle. This value is much smaller than that in the yawed circular cylinder wake (30%) obtained by Zhou et al. (2010),
implying that the coherent structures in the circular cylinder wake is stronger. As mentioned previously, the reduction of the
coherent structures may be related with the increase in the spanwise velocity, which is generally regarded as the dete-
rioration of the two-dimensionality of the flow. As for the transverse velocity component, the coherent structures contribute
more to v2 at α ¼ 0° over the wake range of 1.0 ry r1.0. As α increases from 0° to 30°, the remaining contribution
increases and the coherent contribution decreases. Particularly, the dominant contribution to v2 comes from the remaining
component for α ¼ 45°, which is more than 90%. This result manifests that with the increase of α to about 45°, the large-
scale organized structures nearly break down and their contribution becomes much weaker. Generally the coherent con-
tribution to v2 is greater than that to the other Reynolds stresses as well as the spanwise vorticity variance. This pro-
nounced contribution reflects the enhancement of v~  due to the primarily anti-symmetrical arrangement of the counter-
rotating vorticity of Kármán street (Zhou et al., 2002). Compared with those for a circular cylinder wake (Zhou et al., 2010), a
faster decay from the coherent structures in the contributions to u2 and v2 for the square cylinder wake can also be
observed in Fig. 14, which is consistent with the variation of the velocity contours shown in Fig. 12. As discussed in the
previous section, the coherent contribution to w2 is rather small and this can be seen clearly from Fig. 13(i–l). More than
90% contribution to w2 comes from the remaining structures. A small increase in the coherent contribution is observed as α
increases to 45°, which corresponds to the variation of the w ~  contours.
For the contributions to Reynolds shear stress u v in Fig. 14(m–p), the distribution of ur vr is anti-symmetric about y ¼ 0,
   

which is in agreement with the characteristics of the contours shown in Fig. 13. For α ¼ 0°, the contribution to u v from the
 
coherent component u~ v~ accounts for as much as 50% at y ¼ 71.0, beyond which the remaining contribution still
dominates u v . Overall the remaining contribution to u v is larger than the coherent contribution for other angles. From
 

the wake of a square cylinder at Re ¼ 21,400, Lyn et al. (1995) also found that the coherent contribution to the shear stress
u v is not totally negligible, but significantly smaller than the remaining contribution. With the increase of α from 15° to
45°, the proportion of the coherent contribution decreases significantly to almost zero, which corresponds well with the
magnitudes of the Reynolds shear stress contours mentioned previously. The magnitude of u w in Fig. 14(q–t) is much

Fig. 15. Time-averaged spanwise and transverse vorticity variances with coherent and incoherent contributions at different yaw angles. (a–d), spanwise
vorticity; (e–h), transverse vorticity.
292 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

smaller than that of the Reynolds normal stresses. More than 90% contribution comes from the remaining structures. Unlike
the anti-symmetric distribution of the Reynolds shear stress u v , u w displays more negative signs along the wake. This
result can also be observed in the contour plots (Fig. 13), where the maximum concentration of u~  w ~  at around y ¼1.0
increases when α varies from 0° to 45°.
From Fig. 15(a–d), it can be seen that the remaining contribution dominates the vorticity variance ω2 z , i.e. the ratios of
2
the coherent contribution ω~ z to the vorticity variance ω2z on the centerline are 6%, 3.2%, 3% and 1% for α ¼ 0°, 15°, 30° and
45°, respectively. This result reflects the fact that vorticity is mainly resided with the small-scale structures (Wallace and
Foss, 1995), as illustrated by the features of the coherent and remaining vorticity contours (Fig. 9). With the increase of α, the
intensity of the spanwise vorticity decreases, resulting in the reduction of the coherent contributions. Both Mansy et al.
(1994) and Zhou et al. (2010) have shown that the reduction of the spanwise vortices will result in an increase in the
streamwise vorticity at large yaw angle, leading to the impairment on the two dimensionality of the flow. Unfortunately, this
could not be checked in the present study due to the lack of the streamwise vorticity data. Compared with the spanwise
vorticity component, the dominated contribution to ω2 y also comes from the remaining structures. Similar to the results
2
from a circular cylinder (Zhou et al., 2010), the ratio ω~ y =ω2 y is almost negligible for all the angles, implying that the
transverse vorticity is mainly associated with the small-scale structures.

4. Conclusions

The wake vortical structures and the drag coefficient of a yawed square cylinder are investigated in a wind tunnel at Re of
3600 using a 1D hot-wire vorticity probe and a load cell. The effect of cylinder yaw angle (α¼0–45°) on vortex strength,
vortex shedding frequency and drag coefficient is examined. Quantitative comparisons with those obtained in circular
cylinder wakes are made to demonstrate the similarities and differences between the two types of wakes. The main con-
clusions are summarized below:

1. The rms values of the velocity components, as well as the spanwise and transverse vortices decrease as the yaw angle α
increases, especially for α ¼ 45°. The significant decrease in the rms values on the centerline indicates the reduction of
vortex strength.
2. As α increases from 0° to 45°, the peak region on energy spectra becomes wider and the peak height decreases, reflecting
the dispersion of the turbulent energy over the shedding frequency. At α¼60°, vortex shedding is completely suppressed
as there is no peak discernable on the spectrum. This result is in apparent difference with that obtained in a circular
cylinder wake (Zhao et al. 2009), indicating again that vortices decay much faster in the former than that in the latter,
which is believed to be related with the stronger three-dimensionality effect in the former.
3. Given the experimental uncertainty of 6%, the Strouhal number keeps constant when α r40°, indicating that the IP is
approximately valid over this angle range. This result is also supported by the measured drag coefficients at different yaw
angles, which show a departure from IP by about 8% at α ¼45°.
4. Based on the phase-average analysis, the coherent contours of the spanwise vorticity ω~ z display the apparent Kármán
vortex streets. The vortex concentration shows no much difference for αr30°. However, this value reduces by more than
70% when α increases to 45°, indicating the weakening of the spanwise vorticity strength.
5. The contours of the velocity components also exhibit regular vortex patterns. The maximum concentrations of the u~  and
v~  contours decrease with the increase of α. Especially at α ¼ 45°, the contours of u~  and v~  have lost the regular shape and
reflected a pronounced reduction in vortex strength. However, for the spanwise velocity component w ~  , the magnitude of
the contours increases with the increase of α, which is responsible for the enhancement of the three-dimensionality of
the wake flow.
6. The present study shows similarity between yawed square and circular cylinder wakes in terms of the wake properties,
though it has been found that the decay of the vortex strength with yaw angle in the square cylinder wake is faster than
that in the circular one due to stronger three-dimensionality effect in the former. As the aerodynamic characteristics of a
square cylinder wake are relatively insensitive to Reynolds number, the results obtained in the present study can be
applicable for a wide Reynolds number range.

Acknowledgment

The authors sincerely acknowledge the financial support from Australian Research Council, Australia through ARC Dis-
covery Projects Program Grant no. DP110105171. X. Lou would also like to express his sincere thanks to the China Scho-
larship Council and the University of Western Australia for the CSC Scholarship and the top-up scholarship provided to
support his study in UWA.
 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294 293

References

Alam, M.M., Zhou, Y., 2008. Strouhal numbers, forces and flow structures around two tandem cylinders of different diameters. Journal of Fluids and
Structures 24, 505–526.
Alam, M., Zhou, Y., Wang, X.W., 2011. The wake of two side-by-side square cylinders. Journal of Fluid Mechanics 669, 432–471.
Antonia, R.A., Britz, D., 1989. Phase-averaging in the turbulent far-wake. Experiments in Fluids 7, 138–142.
Antonia, R.A., Zhu, Y., Shafi, H.S., 1996. Lateral vorticity measurements in a turbulent wake. Journal of Fluid Mechanics 323, 173–200.
Balachandar, S., Mittal, R., Najjar, F.M., 1997. Properties of the mean recirculation region in the wake of two-dimensional bluff bodies. Journal of Fluid
Mechanics 351, 167–199.
Blackburn, H.M., Lopez, J.M., 2003. On three-dimensional quasiperiodic floquet instabilities of two-dimensional bluff body wakes. Physics of Fluids 15,
57–60.
Blackburn, H.M., Sheard, G.J., 2010. On quasiperiodic and subnarmonic floquet wake instabilities. Physics of Fluids 22 (No. 031701), 1–4.
Cantwell, B., Coles, D., 1983. An experimental study of entrainment and transport in the turbulent near wake of a circular cylinder. Journal of Fluid
Mechanics 136, 321–374.
Cenedese, A., Romano, G.P., 1991. Experimental test of Taylor's hypothesis by L.D.A in highly turbulent flow. Experiments in Fluids 11, 351–358.
Dahm, W.J.A., Southerland, K.B., 1997. Experimental assessment of Taylor's hypothesis and its applicability to dissipation estimates in turbulent flows.
Physics of Fluids 9 (7), 2101–2107.
Davis, R.W., Moore, E.F., 1982. A numerical study of vortex shedding from rectangles. Journal of Fluid Mechanics 116, 475–506.
Dennis, D.J.C., Nickels, T.B., 2008. On the limitations of Taylor's hypothesis in constructing long structures in a turbulent boundary layer. Journal of Fluid
Mechanics 614, 197–206.
Dutta, S., Muralidhar, K., Panigrahi, P.K., 2003. Influence of the orientation of a square cylinder on the wake properties. Experiments in Fluids 34, 16–23.
Durao, D.F.G., Heitor, M.V., Pereira, J.C.F., 1988. Measurements of turbulent and periodic flows around a square cross-section cylinder. Experiments in Fluids
6, 298–304.
Franzini, G.R., GonÇalves, R.T., Meneghini, J.R., Fujarra, A.L.C., 2013. One and two degree-of-freedom vortex-induced vibration experiments with yawed
cylinders. Journal of Fluids and Structures 42, 401–420.
Gallardo, J.P., Andersson, H.I., Pettersen, B., 2014. Turbulent wake behind a curved circular cylinder. Journal of Fluid Mechanics 742, 192–229.
Hussain, A.K.M.F., Hayakawa, M., 1987. Education of large-scale organized structures in a turbulent plane wake. Journal of Fluid Mechanics 180, 193–229.
Hussain, A.K.M.F., Reynolds, W.C., 1970. The mechanism of an organized wave in turbulent shear flow. Journal of Fluid Mechanics 41, 1–241.
Hussain, A.K.M.F., 1986. Coherent structures and turbulence. Journal of Fluid Mechanics 173, 303–356.
Hutchins, N., Marusic, I., 2007. Evidence of very long meandering features in the logarithmic region of the turbulent boundary layers. Journal of Fluid
Mechanics 579, 1–28.
Kim, D.H., Yang, K.S., Senda, M., 2004. Large eddy simulation of turbulent flow past a square cylinder confined in a channel. Computers & Fluids 33, 81–96.
King, R., 1977. Vortex excited oscillations of yawed circular cylinders. Journal of Fluid Engineering 99, 495–502.
Kiya, M., Matsumura, M., 1985. Turbulence structure in intermediate wake of a circular cylinder. Bulletin of Japan Society of Mechanical Engineers 245,
2617–2624.
Kiya, M., Matsumura, M., 1988. Incoherent turbulence structure in the near wake of a normal plate. Journal of Fluid Mechanics 190, 343–356.
Krogstad, P.A., Kaspersen, J.H., Rimestad, S., 1997. Propagation velocity of perturbations in turbulent channel flow. Physics of Fluids A 10, 949–957.
Lucor, D., Karniadakis, G.E., 2004. Effects of oblique inflow in vortex-induced vibrations. Flow, Turbulence and Combustion 71, 375–389.
Lyn, D.A., Einav, S., Rodi, W., Park, J.H., 1995. A laser-doppler velocimetry study of ensemble-averaged characteristics of the turbulent near wake of a square
cylinder. Journal of Fluid Mechanics 304, 285–319.
Matsumura, M., Antonia, R.A., 1993. Momentum and heat transport in the turbulent intermediate wake of a circular cylinder. Journal of Fluid and
Mechanics 250, 651–668.
Marshall, J.S., 2003. Wake dynamic of a yawed cylinder. Journal of Fluid Engineering 125, 97–103.
Mansy, H., Yang, P.M., Williams, D.R., 1994. Quantitative measurements of three-dimensional in the wake of a circular cylinder. Journal of Fluid and
Mechanics 270, 277–296.
Mi, J., Antonia, R.A., 1994. Corrections to Taylor's hypothesis in a turbulent circular jet. Physics of Fluids 6, 1548–1552.
Najjar, F.M., Balachandar, S., 1998. Low-frequency unsteadiness in the wake of a normal flat plate. Journal of Fluid and Mechanics 370, 101–147.
Nakagawa, S., Nitta, K., Senda, M., 1999. An experimental study on unsteady turbulent near wake of a rectangular cylinder in channel flow. Experiments in
Fluids 27, 284–294.
Okajima, A., 1982. Strouhal numbers of rectangular cylinders. Journal of Fluid Mechanics 123, 379–398.
Oudheusden, B.W., Scarano, F., Hinsberg, N.P., Watt, D.W., 2005. Phase-resolved characterization of vortex shedding in the near wake of a square-section
cylinder at incidence. Experiments in Fluids 39, 86–98.
Ozgoren, M., 2006. Flow structure in the downstream of square and circular cylinders. Flow Measurement and Instrumentation 17, 225–235.
Ramberg, S.E., 1983. The effect of yaw and finite length upon the vortex wakes of stationary and vibrating circular cylinders. Journal of Fluid and Mechanics
128, 81–107.
Saha, A.K., Biswas, G., Muralidhar, K., 2003. Three-dimensional study of flow past a square cylinder at low Reynolds numbers. International Journal of Heat
and Fluid Flow 24, 54–56.
Saha, A.K., Muralidhar, K., Biswas, G., 2000. Experimental study of flow past a square cylinder at high Reynolds numbers. Experiments in Fluids 29, 553–563.
Schlipf, D., Trabucchi, D., Bischoff, O., Hoffsass, M., 2010. Testing of frozen turbulence hypothesis for wind turbine applications with a scanning LIDAR
system. In: Proceedings of the 15th International Symposium for the Advancement of Boundary Layer Remote Sensing, No. 15.
Sheard, G.J., Fitzgerald, M.J., Ryan, K., 2009. Cylinders with square cross-section: wake instabilities with incidence angle variation. Journal of Fluid and
Mechanics 630, 43–69.
Sohankar, A., Norberg, C., Davidson, L., 1999. Simulation of three-dimensional flow around a square cylinder at moderate Reynolds numbers. Physics of
Fluids 11, 288–306.
Surry, D., Surry, J., 1967. The effect of yaw on the Strouhal number and other wake properties of circular cylinders at subcritical Reynolds numbers.
Technical Report, Institute for Aerospace Studies, University of Toronto.
Thakur, A., Liu, X., Marshall, J.S., 2004. Wake flow of single and multiple yawed cylinders. Journal of Fluid Engineering 126, 861–870.
Tong, X.H., Luo, S.C., Khoo, B.C., 2008. Transition phenomena in the wake of an inclined square cylinder. Journal of Fluids and Structures 24, 994–1005.
Uddin, A.K.M., Perry, A.E., Marusic, I., 1997. On the validity of Taylor's Hypothesis in wall turbulence. Journal of Mechanical Engineering Research and
Development 19-20, 57–66.
Wallace, J.M., Foss, J.F., 1995. The measurement of vorticity in turbulent flows. Annual Review of Fluid Mechanics 27, 469–514.
Wyngaard, J.C., 1969. Spatial resolution of the vorticity meter and other hot-wire arrays. Journal of Scientific Instruments (Journal of Physics E) 2, 983–987.
Zaman, K.B.M.Q., Hussain, A.K.M.F., 1981. Taylor hypothesis and large-scale coherent structures. Journal of Fluid Mechanics 112, 379–396.
Zhang, H.J., Zhou, Y., Antonia, R.A., 2000. Longitudinal and spanwise vortical structures in a turbulent near wake. Physics of Fluids 12, 2954–2964.
Zhao, M., Cheng, L., Zhou, T., 2009. Direct numerical simulation of three-dimensional flow past a yawed circular cylinder of infinite length. Journal of Fluids
and Structures 25, 831–847.
294 X. Lou et al. / Journal of Fluids and Structures 61 (2016) 274–294

Zhou, T., Razali, S.F., Zhou, Y., Chua, L.P., Cheng, L., 2009. Dependence of the wake on yaw of a stationary cylinder. Experiments in Fluids 46, 1125–1138.
Zhou, T., Wang, H., Razali, S.F., Zhou, Y., Cheng, L., 2010. Three-dimensional vorticity measurements in the wake of a yawed circular cylinder. Physics of
Fluids 22 (No. 015108), 1–15.
Zhou, T., Zhou, Y., Yiu, M.W., Chua, L.P., 2003. Three-dimensional vorticity in a turbulent cylinder wake. Experiments in Fluids 35, 459–471.
Zhou, Y., Antonia, R.A., 1993. A study of turbulent vortices in the near wake of a cylinder. Journal of Fluid Mechanics 253, 643–681.
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.
Zhou, Y., Zhang, H.J., Yiu, M.W., 2002. The turbulent wake of two side-by-side circular cylinders. Journal of Fluid Mechanics 458, 303–332.