1 Experimental investigation on aerodynamic noise and flow structures of a

2 vibrissa-shaped cylinder
3 Guanjiang Chen, Xiao Liu, Bin Zang, and Mahdi Azarpeyvand
4 Faculty of Engineering, University of Bristol, BS8 1TR, Bristol, United Kingdom
5 (Dated: November 20, 2023)
6 The noise mitigation effect of bio-inspired geometries has attracted growing attention from both
7 research and industry, such as the vibrissa-shaped cylinder derived from the harbor seal. Experi-
8 ments were conducted to investigate the far-field noise and the near-field wake of the flow past a
9 vibrissa cylinder, a circular cylinder, and an elliptical cylinder at Re = 3.6 × 104 , in the subcritical
10 flow regime. The frequency characteristic of the far-field acoustic pressure and the near-field veloci-
11 ties are analyzed. The mean and fluctuating velocities, dominant flow modes from proper orthogonal
12 decomposition in both vertical and horizontal planes as well as the time-frequency behaviour of the
13 dominant flow structures from wavelet transform are also presented to better understand the wake
14 dynamics and the direct relation of these flow structures and development to the far-field noise.
15 The vibrissa cylinder reduces the overall sound pressure level by 13.2 dB and 8.3 dB compared with
16 the circular and the elliptical cylinders, respectively, with a remarkable attenuation of the tonal
17 peak associated with vortex shedding. From the detailed velocity measurements in multiple wake
18 planes, it is clearly observed that vortex shedding of the vibrissa cylinder is weaker in strength and
19 significantly less coherent in the spanwise direction than the other two cylinder cases, accompanied
20 by more transient changes for the dominant flow structures. The results also reveal the distinct
21 flow behaviour behind the nodal and saddle planes of the vibrissa cylinder, further contributing to
22 this three dimensional vortex shedding. Consequently, the power spectral density of the tonal peaks
23 associated with the vortex shedding in both near-field velocities and far-field acoustic pressure are
24 attenuated, leading to a lower noise level. Understanding the detailed flow dynamics of the vib-
25 rissa cylinder will provide useful insights into more efficient bio-inspired cylinder designs in noise
26 mitigation and wake control.

27 I. Introduction
28

29 Flow past a cylinder has always attracted research interests as a common problem in engineering applications, such
30 as train pantographs, automotive axles, wind turbine towers, etc. Over the last few decades, numerous investigations
31 have been conducted on understanding the aerodynamic performance of the flow past a circular cylinder [1–6]. The
32 force coefficients and the vortex shedding suppression of other shaped cylinders were also reported, including the
33 elliptical cylinder [7–9], the wavy cylinder [10–12], and the vibrissa cylinder [13, 14]. The flow dynamics of the
34 immersed cylinder is closely related to the Reynolds number, Re = U∞ D/ν based on the circular cylinder diameter
35 D, the freestream velocity U∞ , and the kinematic viscosity ν. Depending on the Reynolds number, the flow past a
36 cylinder could be divided into the subcritical regime (400 < Re < 3×105 ), the critical regime (3×105 < Re < 3×106 ),
37 the supercritical regime (Re > 3 × 106 ) [15, 16]. The unsteady aerodynamic forces acting on the cylinder vary across
38 the different flow regimes. In addition to the unsteady forces, the flow separation and vigorous fluctuations give
39 rise to sound generation from the cylinder. More recently, increasing attention has been focused on understanding
40 aerodynamic noise generation and its mitigation strategies since noise has a negative impact on the community health.
41 The aerodynamic noise of flow past a cylinder, also known as the aeolian tones, was first studied by Strouhal,
42 differing from the vibration noise. A series of studies, such as the ones by Gerrard [17], Phillips [18] and Revell et
43 al. [19], established the analytical methods to predict the sound intensity of the aeolian tones from the fluctuating
44 force experiences by the cylinder, based on the Lighthill integration methods [20–22]. The frequency of primary tone
45 was consistent with the frequency of the lift force fluctuation and the noise in general displayed a dipole pattern.
46 Etkin et al. [23] conducted both the theoretical analysis and experiments to examine the noise characteristics of
47 the flow past a circular cylinder. They showed a discrepancy between the predicted and experimental results and
48 suggested that the influence of experimental condition on the noise generation is more complicated and hence, more
49 precise unsteady force results together with the knowledge of the flow structure coherence along the span are needed
50 for accurate noise prediction.
51 To further understand the noise generation mechanisms and sources of the flow past a cylinder, correlation between
52 the far-field noise and the near-field flow has been carefully examined. Alemdaroglu et al. [24] carried out an
53 experimental investigation of the flow past a cylinder using far-field acoustic pressure measurements and hot-wire
54 velocity measurements. By studying the coherence between the acoustic pressure and the wake velocity, they found
55 that there is an intense level of near-to-far-field coherence at fundamental vortex shedding frequency in the subcritical
 2

56 regime, and the coherence diminished in the critical regime. Then, in the supercritical regime, it increased again
57 to be as high as that in the subcritical regime. Moreover, the far-field acoustic pressure was highly related to the
58 surface boundary layer condition, especially the point of boundary layer separation on the cylinder. The results were
59 consistent with the experimental works of Fujita et al. [25, 26], which analyzed the coherence between the far-field
60 and the near-field pressure. They [25] also suggested that the aeolian tone decreased in the critical Re range because
61 of the reduction of the near-field pressure fluctuations and enhanced three-dimensional vortex shedding. Oguma et al.
62 [27] also experimentally evaluated the noise source of the flow past a cylinder, using cross-correlation analysis of the
63 pressure around the cylinder and the acoustic pressure in the far field. The sources with high cross-correlation levels
64 were both near the point of boundary layer separation at two sides of the cylinder and in the near wake. These noise
65 source regions identified were later confirmed by Zheng et al. [28] and Zamponi et al. [29], using the velocity-acoustic
66 pressure cross-correlation analysis and the beamforming array method, respectively.
67 The noise source regions implied that the flow separation and subsequent vortex shedding of the cylinder played a
68 dominant role in the generation of aeolian tone and hence, key to achieve effective noise mitigation is to control the
69 boundary layer separation and the dynamics of vortex-shedding very close to the cylinder. Fujisawa et al. [30] added
70 longitudinal grooves to the cylinder and attributed the noise reduction to the downstream shift of flow separation and
71 weaker velocity fluctuations. More recently, studies have demonstrated the ability of a bio-inspired vibrissa cylinder
72 to modify the vortex shedding pattern behind the cylinder [31]. Hanke et al. [13] first established an idealized vibrissa
73 cylinder model (see Fig. 1 (c) for the shape description) and simulated the flow past it. Compared with the circular
74 cylinder, the vibrissa cylinder showed more three-dimensional vortex shedding with obviously weaker lift fluctuations
75 and smaller drag coefficients. Jie and Liu [32] carried out large eddy simulations (LES) on the flow past a cylinder and
76 reported an approximate 79% reduction in the root-mean-square (rms) of the fluctuating lift for a vibrissa cylinder
77 when compared to the circular baseline. By changing the wavelength, the vibrissa cylinder could achieve a reduction
78 of the lift fluctuation by approximately 58% to 93% against the circular counterpart [14]. Dunt et al. [33] numerically
79 investigated the flow past vibrissa cylinders with different inlet flow angles. They also observed significantly lower lift
80 fluctuation of 52% to 94% compared to that of the circular cylinder.
81 The wavy morphology of the vibrissa cylinder alters the vortex shedding structures and frequency, which is its major
82 influence on the flow field. The LES results of Hanke et al. [13] demonstrated a wavy separation line along the spanwise
83 direction, which could also be seen in the work of Jie and Liu [32]. The spectral analysis of fluctuating component of
84 the lift from Jie and Liu [32] showed a low-magnitude hump near the Strouhal number, St = f D/U∞ = 0.2 (where f
85 is the frequency, D is the circular cylinder diameter, and U∞ is the freestream velocity), for the vibrissa cylinder. The
86 proper orthogonal decomposition (POD) modes indicated non-synchronous vortex shedding processes in the saddle
87 and nodal planes, revealing that the vortex shedding of the vibrissa cylinder was more three-dimensional and less
88 dominant than the circular cylinder case. The dynamic mode decomposition (DMD) analysis conducted by Wang
89 and Liu [34] on the velocity field results confirmed that there existed discrepancy in the fundamental vortex shedding
90 frequencies along the saddle and nodal planes of the vibrissa cylinder, which were at St = 0.23 and 0.30, respectively.
91 This frequency inconsistency resulted in weaker vortex shedding than the circular cylinder case. Furthermore, Yoon et
92 al. [35] compared the near-field flow structures between the vibrissa and elliptical cylinders. Their LES showed that
93 additional vorticity (i.e., more complex turbulent structures) close to the surface of the vibrissa cylinder contributed
94 to the more three-dimensional vortex shedding of vibrissa cylinder. Chu et al. [36] conducted LES on the flow
95 past a vibrissa cylinder and made spectral proper orthogonal decomposition (SPOD) analysis. The SPOD results
96 identified the three-dimensionality of the vortex shedding of the vibrissa cylinder and four vortex shedding patterns
97 were captured.
98 According to previous studies, the vibrissa cylinder could modify the vortex shedding behaviour close to the cylin-
99 der, thus likely modifying the major source of the aeolian tone and mitigating the generated noise. To the author’s
100 knowledge, investigations into the aeroacoustic performance of the flow past a vibrissa cylinder and its correlation
101 with the flow field has remained limited, and thus detailed examinations are still needed to better understand the
102 consequences of these modifications to the vortex shedding process on the generated noise, which is the major inspi-
103 ration and focus of the present work. Besides, the vortex shedding structures of the vibrissa cylinder and its general
104 wake dynamics are worthwhile to be further studied, especially the flow field along the spanwise plane is rarely pre-
105 sented, but could disclose useful information on the spanwise coherence of flow structures. In the present work, both
106 aeroacoustic and particle image velocimetry (PIV) experiments are conducted on the flow past a vibrissa cylinder,
107 a circular cylinder, and an elliptical cylinder. The studied Reynolds number is in the subcritical regime, where the
108 transition to turbulence happens in the separated shear layers [16, 37]. The spectral analysis is carried out on both
109 the far-field acoustic and near-field velocity results. The flow around the three cylinders is studied in detail through
110 time-averaged and rms velocity distributions and POD. Finally, the correlation between the noise and velocity results
111 are discussed to reveal the noise reduction mechanism of the vibrissa cylinder compared to the other two cylinders.
112 The layout of this paper is as follows: the experimental setup is shown in Sec. II. The results are then presented
113 and discussed in Sec. III. Finally, Sec. IV summaries the effects of the vibrissa cylinder on the flow and aeroacoustic
 3

114 results.

FIG. 1. Schematics of (a) circular cylinder, (b) elliptical cylinder, (c) vibrissa cylinder.

115 II. Experiment setup

116

117 Detailed noise and flow field measurements were performed in the wind tunnel facilities at the University of Bristol
118 [38–40] to both the far-field acoustics and the near-field velocity distribution of flow past circular, elliptical and vibrissa-
119 shaped cylinders. The free-stream velocity was maintained at U∞ = 25m/s, which is equivalent to a Reynolds number
120 of Re = 3.6 × 104 based on the circular cylinder diameter. In this section, the cylinder models, the noise and the flow
121 measurement set-up are introduced.

122 A. Cylinder models

123 The schematics of the circular cylinder, the elliptical cylinder, and the vibrissa cylinder are shown in Fig. 1, and
124 detailed geometrical parameters are listed in Table I. A circular cylinder with a diameter D = 22 mm is used as the
125 baseline case. An elliptical cylinder with a major axis of be = 15.7 mm and a minor axis of ae = 7.7 mm is also
126 studied as a comparison case since the cross-section of the vibrissa cylinder is an ellipse. The structure of the vibrissa
127 cylinder, shown in Fig.1(c), has two dominant ellipses with two axes pairs (av , bv ) and (Av , Bv ), respectively, where
128 bv and Bv are the two major axes and av and Av are the two minor axes. The wavelength λ defines the distance
129 between two ellipses, hence the separation between the nodal and saddle planes. Moreover, the two ellipses are offset
130 with reference to the vertical plane by inclination angles, denoted as α and β. the vibrissa cylinder employed in the
131 present study has a wavelength of λ = 27.5 mm. The nodal ellipse (section C-C in Fig. 1 (c)) of the vibrissa cylinder
132 has a minor axis of Av = 7.26 mm and a major axis of Bv = 17.82 mm, while the saddle ellipse (section D-D in
133 Fig. 1 (c)) has a minor axis ae = 8.58 mm and a major axis bv = 14.3 mm. The offset angles are determined as
134 α = 15.27◦ and β = 17.6◦ , respectively. Following the approach used by Jie and Liu [32], the parameters of the
135 elliptical and vibrissa cylinder are chosen such that the three cylinders have the same volume, with identical spanwise
136 length of 500 mm.

TABLE I. Geometrical parameters of the cylinders.

D λ ae be α β Av Bv av bv
mm mm mm mm degree degree mm mm mm mm
Circular 22 - - - - - - - - -
Elliptical - - 7.7 15.7 - - - - - -
Vibrissa - 27.5 - - 15.27 17.6 7.26 17.82 8.58 14.3
 4

137 B. Noise measurement setup

138 The schematic of the far-field acoustic measurement in the aeroacoustic wind tunnel is shown in Fig. 2. Two
139 side plates was flush-mounted to the open-jet nozzle with exit dimensions of 500 mm (width) and 775 mm (height)
140 to maintain two-dimensional flow past the cylinders. The cylinder was positioned 350 mm downstream the nozzle
141 exit at the mid position in the y direction, where the free-stream flow remained uniform in its potential core. The
142 point of origin, x = y = z = 0, was defined at the centre of the cylinder along the mid-span. Directly above and
143 1.75 m away from the mid-span of the cylinder (i.e., z = 0), a far-field microphone array was placed to collect
144 the far-field acoustic signals. The array consists of 19 G.R.A.S 40PL microphones, giving a polar angle range from
145 θ = 40◦ to 130◦ , where smaller θ indicates upstream angles towards the nozzle exit. All the measurements were
146 sampled simultaneously using National Instrument PXI-4499 acquisition modules, at a sampling rate of 215 Hz and
147 a sampling duration of 16 seconds. The collected time series of data were then evaluated by the “Pwelch” function ′
148 in Matlab to yield the power spectral density (PSD) of the far-field acoustics, as 10 log(Φp′ p′ /pref 2 ), where p is
149 the far-field acoustic pressure fluctuations, Φp′ p′ is the output of the “Pwelch” function and pref = 20 µPa is the
150 reference pressure. After calibration using G.R.A.S 42AA Pistonphone, the uncertainty of the far-field acoustic
151 pressure fluctuation measurements are determined to be ± 1.5 dB/Hz for a 95% confidence interval, consistent with
152 previous works [41–43].

FIG. 2. Experiment setup for the far-field acoustic measurement of the flow past cylinders.

153 C. Particle image velocimetry measurement setup

154 The flow field around the cylinder was obtained using two-dimensional, two-component particle image velocimetry
155 (PIV) in the low turbulence closed-circuit wind tunnel, as shown in the schematic of Fig. 3 (a). a Litron LDY-303
156 Nd:YLF high-speed laser produced a thin laser sheet to illuminate the region of interest behind the cylinder. The
157 emmited light from the DEHS seeding particles were captured by a Photron FASTCAM Mini WX 100 camera with
158 a resolution of 2048 × 2048 pixels2 and 14 bit image depth. A total of 2000 image pairs were collected for each
159 measurement plane, at a sampling frequency of 840 Hz. The particle-images were analyzed using a multi-pass cross-
160 correlation algorithms with an initial interrogation window of 128 x 128 and a final window of 32 × 32 pixels2 and a
161 50% overlap in LaVision DaVis software. The uncertainty of the velocity from PIV measurements is approximately
162 3%, which is comparable to similar studies [44–46]. The flow fields in the vertical planes (xy-planes) and the horizontal
163 planes (xz-planes) were measured to illustrate the flow structures around the cylinders, whose schematic diagrams are
164 in Fig. 3 (b-c). Note that the coordinate system is defined identical to that of the far-field acoustic measurement, i.e.,
165 the center of the cylinder at the mid-span is set as the origin. For the circular and the elliptical cylinders, the vertical
166 plane corresponds to the mid-span plane at z/D = 0. The horizontal planes were chosen to reveal the flow structures
 5

167 and developments close to the shear layer, which were located at y/D = 0.5 and y/D = 0.36, respectively. For the
168 vibrissa cylinder, two vertical planes along both the saddle and nodal planes were measured as shown in Fig. 3 (c),
169 and its horizontal plane was defined at y/D = 0.36, similar to that of the elliptical cylinder. For the present study, x,
170 y, and z represent the streamwise, cross-stream, and spanwise directions, with corresponding velocities u, v, and w.

(a) (b) y

y
x
z Saddle
(c)
z
Nodal
x

CMOS Camera

FIG. 3. (a) Experiment setup for the PIV measurement; Schematic diagram of (b) vertical and (c) horizontal planes in PIV
measurements.

171 III. Results and Discussions

172 The experimental results are presented and discussed in this section. Firstly, the power spectral density (PSD) of
173 the far-field acoustic pressure is presented, and the far-field noise directivity pattern is analyzed. The time-averaged
174 and rms (root-mean-square) characteristics of the obtained velocity fields provide basic information on the flow
175 development past the differently-shaped cylinders. These results are followed by proper orthogonal decomposition
176 (POD) and wavelet analyses to provide a more complete picture of the vortex shedding behaviour of the circular,
177 elliptical and vibrissa cylinders and reveal the frequency characteristics. Finally, the PSD of the velocity fields are
178 presented and discussed to allow a better understanding of the noise reduction mechanism of the vibrissa cylinder.

179 A. Far-field noise results

180 The far-field acoustic pressure PSD at θ = 90◦ is shown in Fig. 4. Recall that θ = 0◦ points to the upstream
181 towards the nozzle, while θ = 180◦ points to the downstream. The frequency is converted to the Strouhal numbers,
182 St = f D/U∞ , based on the circular cylinder diameter D. Overall, the primary tonal peak is well captured for the
183 circular cylinder at the vortex shedding frequency of St = 0.19, while for the elliptical cylinder, the Strouhal number
184 of the vortex shedding frequency shifts to St = 0.37, consistent with the reported results [47–49]. the acoustic pressure
185 PSD decreases with the Strouhal number, and the broadband components of the three cylinders at higher Strouhal
186 number of St ≥ 1 coincide with each other due to the similar background noise. At the vortex shedding frequency,
187 the tonal peak of the PSD protrudes approximately 30 dB/Hz above the broadband components for both circular and
188 elliptical cylinders, and the peak magnitude for the elliptical one is 7.2 dB/Hz lower than that of the circular cylinder.
189 Intriguingly, the vibrissa cylinder does not show any prominent tones, though a minor bump can be observed near
190 St = 0.26, approximately 2.9 dB/Hz above the broadband component, indicating a strong noise reduction of the
191 vibrissa cylinder. As will be shown and discussed in Section. III C and III D, this bump is also related to the weakened
192 vortex shedding in the wake of the vibrissa cylinder. In addition, only the circular cylinder shows noticeable peaks at
193 the first and second harmonics of St = 0.38 and 0.57, contributing to higher overall far-field noise level.
194 The overall sound pressure levels (OASPL) of the threeR cylinder cases are presented in Fig. 5 (a), which are
195 calculated by integrating the pressure fluctuation as 10 log (Φp′ p′ /pref 2 )df from f = 100 Hz to 9000 Hz, where f
196 is the frequency. Figure 5 (b) shows the peak magnitude of the acoustic pressure PSD at the fundamental vortex
197 shedding frequencies of the three cylinders at different angular positions. Note that for the vibrissa cylinder, the
 6

Circular
60 Elliptical
Vibrissa

) [dB/Hz]
40

ref
2
/p
p'p'
10 log (
20

0
-1 0
10 10

f D U¥
/

FIG. 4. Far-field acoustic pressure power spectral density of the circular, elliptical and vibrissa cylinders measured at θ = 90◦ .

198 magnitude is taken at the hump of St = 0.26. The variation of OASPL within the range of polar angles investigated
199 is relatively small for all three cylinders, agreeing with the dipolar pattern expected for the far-field noise of flow past
200 a cylinder [50, 51]. The vibrissa cylinder has very limited effect on modifying the directivity pattern of the far-field
201 acoustics. Nevertheless, comparing the magnitude of the OASPL for the three cylinders, on average, the elliptical
202 cylinder is 4.9 dB lower than that of the circular cylinder. For the vibrissa cylinder, the OASPL is considerably lower,
203 with a reduction of 13.2 dB and 8.3 dB compared to the circular and elliptical cylinders, respectively. As seen from
204 the far-field acoustic spectra in Fig. 4 and the tonal peak magnitude in Fig. 5 (b), the significant noise reduction
205 ability of the vibrissa cylinder can essentially be attributed to its mitigation of vortex shedding tones. The variation
206 of magnitude of the primary tonal peaks with different angles for the three cylinders is similar to those of the OPASL,
207 implying that the tonal peak at the fundamental vortex shedding frequency accounts for a significant proportion of
208 the overall noise level.

75 90 105 75 90 105
(a) (b)
60 120 60 120
45 135 45 135
q q
30 150 30 150

15 165 15 165

0 180 0 180
50 60 70 80 90 20 30 40 50 60 70 80
Circular Circular
Elliptical OASPL [dB] Elliptical PSD peak [dB/Hz]
Vibrissa Vibrissa

FIG. 5. Far-field acoustic directivities, including (a) OASPL, (b) PSD at the fundamental vortex shedding frequency.

209 B. Time-averaged and root-mean-square flow quantities

210 The flow fields shown in Fig. 6 provide first-hand knowledge of the wake pattern and near-field velocity fluctuations
211 along the vertical planes, including the time-averaged and rms velocities as well as the turbulent kinetic energy
212 (TKE). Note that when presenting the time-averaged and rms velocities in Fig. 6, the upper and lower part of each
213 sub-figure represents the u- and the v-velocity component, respectively, as these distributions are symmetry or anti-
214 symmetry about the centerline y/D = 0. The time-averaged velocities have a symmetry pattern with regard to the
215 centerline for the u component, and an anti-symmetry pattern for the v component because the flow separates from
 7

216 both sides of the cylinder, contributing equally to the time-averaged flow field. The time-averaged streamwise (u)
217 velocity shows a recirculation region behind the cylinder, which is about 1.4D long for the circular case, whereas the
218 elliptical cylinder shows a much shorter recirculation length of 0.5D. On the other hand, the vibrissa cylinder shows
219 a more complex time-averaged flow field, displaying different recirculation lengths in the nodal and saddle planes.
220 The shorter recirculation along the nodal plane has a length of approximately 0.7D and the longer one in the saddle
221 plane extends to about 1.7D, indicating distinct flow developments and significant flow structure variations along the
222 spanwise direction of the vibrissa cylinder.

(a1) (b1) (c1) (d1)

(a2) (b2) (c2) (d2)

(a3) (b3) (c3) (d3)

V_rms

Circular Elliptical Vibrissa-nodal Vibrissa-saddle

FIG. 6. Time-averaged velocities (1st row), rms velocities (2nd row), and TKE (3rd row) distributions in the mid plane of
(a1-a3) the circular cylinder case and (b1-b3) the elliptical cylinder case, and in (c1-c3) the nodal and (d1-d3) saddle planes of
the vibrissa cylinder case. Note that in the first two rows, the upper and lower parts of each figure represent the u component
and the v component respectively.

223 The rms velocities show symmetry patterns with regard to y/D = 0 for both the u- the v- velocity components,
224 as seen in Fig. 6 (a2-d2). For the circular and the elliptical cylinders, the higher urms regions occur in the shear
225 layer immediately downstream of the cylinders, and then the regions extend to the centerline, due to the growth
226 of the shear layer instability and the subsequent large-scale vortex shedding in the wake. The elevated vrms are
227 associated with the alternating upward and downward flows caused by the vortex shedding. As a result, the higher
228 vrms regions can be found immediately after the recirculation region where the vortex shedding becomes dominant and
229 are located near the centerline. The vibrissa cylinder shows much weaker rms velocities compared with the circular
230 and elliptical cases in both nodal and saddle planes, suggesting possibly a weaker vortex shedding process in the near
231 wake. Moreover, the vrms velocity component in the nodal plane undergoes first a ‘contraction’ before expanding
 8

232 downstream. The convergent-divergent behaviour as reported in Ref. [32], instead of a monotonic growth seen in the
233 circular and elliptical cases, indicates that the undulating geometry of the vibrissa causes a different vortex shedding
234 pattern and stabilizes the velocity fluctuations in the nodal plane.
′ ′
235 The TKE, calculated by 0.5((u ′ )2 + (v ′ )2 ), reflects the combined effects of velocity fluctuations u and v , as shown
236 in Fig. 6 (a3-d3). Consistent with the rms velocity results, the vibrissa cylinder shows a much smaller TKE magnitude
237 than those of circular and elliptical cylinders, particularly along the nodal plane, implying weakened vortex shedding
238 and induced velocity fluctuations. Moreover, the wake regions with higher TKE magnitudes are much smaller for
239 the elliptical and the vibrissa cylinders than that of the circular case. For the circular and elliptical cylinders, the
240 distribution of the TKE is significantly influenced by the velocity fluctuations induced by the strong and coherent
241 vortex shedding, hence corresponding more closely to the vrms distribution. On the other hand, both the urms and
242 vrms magnitudes of the vibrissa cylinder remain comparable due to weakened vortex shedding. As a result, the TKE
243 distribution of the vibrissa cylinder resembles a combination of the urms and vrms distribution.
244 The three-dimensionality of the flow structures in the cylinder wake could be demonstrated by the time-averaged
245 and rms velocity distributions in the horizontal planes, close to the separated shear layers at y/D = 0.5 for the circular
246 cylinder and at y/D = 0.36 for the elliptical and the vibrissa cylinders, as shown in Fig. 7. Due to the larger-scale
247 vortices, the time-averaged streamwise (u) velocity of the circular cylinder is affected more by the alternating flow,
248 resulting in lower values than that of the elliptical cylinder case, as shown in Fig. 7 (a1, b1). The u/U∞ immediately
249 downstream of the circular cylinder is close to zero and then gradually recovers along the streamwise direction to
250 approximately u/U∞ = 0.75 at x/D = 5. On the other hand, the leeward region of the elliptical cylinder shows a
251 much greater u/U∞ . In the elliptical cylinder case, a decrease in the streamwise velocity can be observed at a further
252 downstream location of x/D = 1.5 and then quickly recover back to free-stream at approximately x/D = 3, which
253 agrees well with the notably reduced recirculation region in the vertical plane in Fig. 6. While both circular and
254 elliptical cylinders retain relatively uniform u velocity distributions along the spanwise direction, the vibrissa cylinder
255 again exhibits distinct flow developments in the nodal and saddle regions. Close to the leeward of the vibrissa cylinder,
256 the streamwise velocity in the saddle region shows low u/U∞ , affected by the wake recirculation similar to the circular
257 cylinder, whereas the counterpart in the nodal region keeps a relatively uniform u/U∞ = 1, resembling the flow
258 development behind the elliptical case. The significant differences between the two regions indicate three-dimensional
259 flow mixing will take place between the nodal and saddle regions, which can be reaffirmed by the fact that the u
260 velocity magnitude becomes increasingly comparable along the span as the flow is convected downstream.
261 Although the vortex shedding is expected to be highly two-dimensional,i.e., uniform along the spanwise direction,
262 for the circular and the elliptical cylinders, the spanwise velocity distribution, w/U∞ , in Figs. 7 (a2-c2) shows a minor
263 extent of spanwise flow very close to the leeward of the cylinders, indicating the three-dimensionality of the vortex
264 shedding process. The vortex shedding of the elliptical cylinder appears to be more two-dimensional than that of
265 the circular case because the streamwise velocity near the circular cylinder leeward is lower, which is subsequently
266 more susceptible to small flow disturbances in the spanwise direction. This phenomenon is also demonstrated in
267 studies [32, 52]. Moreover, the side plates of the nozzle extension may contribute to the spanwise flow as well.
268 It is worthwhile to mention that in this paper the two-dimensional vortex shedding refers to the vortex shedding
269 structure being uniform and coherent along the spanwise direction, while the three-dimensional vortex shedding
270 displays noticeable spanwise variations. For the vibrissa cylinder, the wavy shape naturally induces three-dimensional
271 flows in the spanwise direction, as seen in Fig. 7 (c1, c2). It can be clearly observed from the w/U∞ distributions
272 that relatively strong spanwise flow initiates at x/D = 1.5 downstream and is generally moving and converging from
273 the nodal plane to the saddle plane, resulting in a symmetry distribution with regard of the saddle plane.
274 Figure 7 (a3-c3, a4-c4) shows the distribution of urms /U∞ and wrms /U∞ in the horizontal planes for all three
275 cylinders. Though the magnitude of wrms is apparently smaller than the urms , the distributions of urms and wrms are
276 similar, given that their fluctuations are mostly dominated by the separated shear layers and the subsequent vortex
277 shedding. For the circular cylinder, the high urms and wrms velocity regions are located between 1 < x/D < 3,
278 where the large-scale vortex shedding begins to emerge in the wake. This region moves upstream in the elliptical
279 cylinder with smaller rms velocity magnitudes due to weaker vortex shedding than that of the circular case. The
280 vortex shedding of the vibrissa cylinder is even weaker and its urms and wrms distributions are different from the
281 other two cases. Apart from for the vortex shedding, the strong three-dimensional flow gives rise to intense velocity
282 fluctuations, causing high rms velocity magnitudes near the saddle region of the vibrissa cylinder.
283 To further assess the influence of the three-dimensional vortex shedding and understand the flow development
284 behind the vibrissa cylinder, the urms /U∞ and wrms /U∞ velocity distributions in the y/D = 0 horizontal plane are
285 shown in Fig. 8. The high rms velocity regions are located near x/D = 2, and undulate along the spanwise direction
286 according to the wavy shape of the vibrissa cylinder, corroborating the three-dimensional vortex shedding. The wrms
287 distribution, as expected, shows periodic variations in the spanwise direction because of the velocity fluctuation of
288 nodal-to-saddle flow, consistent with the w/U∞ distributions. The y/D = 0 plane results of the elliptical cylinder
289 and vibrissa cylinders are not presented as they reveal similar flow structures as the shear layer plane distributions in
 9

(a1) (b1) (c1)

(a2) (b2) (c2)

(a3) (b3) (c3)

(a4) (b4) (c4)

Circular (y/D=0.5) Elliptical (y/D=0.36) Vibrissa (y/D=0.36)

FIG. 7. Time-averaged u velocities (1st row), w velocities (2nd row), rms u velocities (3rd row), and rms w velocities (4th row)
distributions in the y/D = 0.5 plane of (a1-a4) the circular cylinder case, and the y/D = 0.36 planes of (b1-b4) the elliptical
cylinder case and (c1-c4) the vibrissa cylinder case.

290 Fig. 7 (a3-a4, b3-b4).

291 It is evident the vibrissa cylinder could suppress the two-dimensional vortex shedding in the near-wake and weaken
292 the related velocity fluctuations, which can help to reduce the far-field noise [53–55]. Nevertheless, it remains unclear
293 of the detailed dynamics and development of the vortex shedding process in the wake of the vibrissa cylinder and
294 whether the minor bump in the far-field acoustic pressure PSD at St = 0.26 indeed corresponds to the vortex shedding
295 phenomenon. Therefore, the following sections will present further analyses from the proper orthogonal decomposition
296 and wavelet analysis, in order to reveal the organisation of the coherent vortex structures.
 10

(a) (b)

FIG. 8. (a) Rms u velocity, and (b) rms w velocity distributions in the horizontal plane y/D = 0 of the vibrissa cylinder case.

297 C. Proper orthogonal decomposition of the near wake

298 The proper orthogonal decomposition (POD) analyses is applied on the velocity fields in both vertical and horizontal
299 planes to identify the energetic flow structures for all three cylinders. The analysis makes use of the snapshot POD
300 methods and the detailed description of the POD method could be found in [11, 56–58]. Though there have been
301 numerous POD studies on flow past cylinders, fewer have shown the coherent flow structures along the spanwise plane,
302 not to mention from the vibrissa-shaped cylinder.

0.4
Circular
Elliptical
Vibrissa-nodal
0.3 Vibrissa-saddle
i

0.2
i
/

0.1

0
1 2 3 4 5 6 7 8 9 10
modes

FIG. 9. Eigenvalue distribution of the first 10 POD modes in vertical planes

The normalized eigenvalues (λi / λi , where λi represents the eigenvalue of the ith POD mode) are associated with
P
303

304 the energy fraction of the corresponding modes with regard to the total fluctuating energy of the flow field. Figure 9
305 shows the normalised eigenvalue distribution of the first 10 POD modes in the vertical planes. The first two POD
306 modes for both the circular and elliptical cylinders have similar energy fractions, hence are considered as a mode pair,
307 which likely indicates the convection behaviour of large-scale coherent structures in the flow field [59–62]. Furthermore,
308 the first two modes accounts for 68% and 60% of the total fluctuating energy for the circular and elliptical cylinders,
309 respectively, corroborating the fact that strong and coherent vortex shedding plays a dominant role in the near-wake
310 flow development of the two cylinders. On the contrary, the energy fractions associated with the first two POD modes
311 of the vibrissa cylinder just account for 15% and 33% for the nodal and saddle planes. Moreover, it can be observed
312 from Fig. 9 that higher POD modes of the vibrissa cylinder contain more energy proportions than their circular and
313 elliptical counterparts, suggesting the loss of large-scale coherent structures for the vibrissa cylinder in both vertical
314 planes.
315 Figure 10 shows the power spectral density of temporal coefficients (10log(Φkk )) of the first three modes in vertical
 11

50 50 50
(a) (b) (c) Circular
Elliptical
40 40 40 Vibrissa-nodal
Vibrissa-saddle
30 30 30

10log(kk)
10log(kk)

10log(kk)
20 20 20

10 10 10

0 0 0
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
f D/U¥ f D/U¥ f D/U¥

FIG. 10. Spectrum of temporal coefficients of (a) mode 1, (b) mode 2, (c) mode 3 in vertical planes.

316 planes, displaying the energy-frequency information of the associated modes. The other modes are not presented
317 since the energy fractions associated with the modes are considerably lower. The spectra of the first two modes are
318 very similar, confirming that they form a POD mode pair, consistent with the eigenvalue results in Fig. 9. Primary
319 tonal peaks are captured at St = 0.21 and St = 0.36 for the first two modes of the circular and elliptical cylinders,
320 respectively, comparable to the tonal peaks identified from the far-field acoustic pressure PSD shown in Fig. 4,
321 revealing these modes are related to the strong and coherent vortex shedding. The differences in the tonal peak
322 frequencies between the far-field acoustic pressure PSD and the POD temporal coefficients spectra can be attributed
323 to the differences in the experimental conditions at the two experimental facilities. However, this is considered to be
324 within the acceptable range as based on a survey of the literature, minor variations of the vortex shedding frequencies
325 have been reported in the previous studies [47, 63], and both frequencies in the present work are associated with the
326 coherent vortex shedding in the subcritical regime [12, 28, 36]. Interestingly, the first two modes of the vibrissa cylinder
327 shows a peak, though broader than those of the circular and elliptical cylinders, at Strouhal number of St = 0.26 in
328 the saddle plane. Recall that a minor hump at similar Strouhal number was observed from the far-field PSD result
329 of the vibrissa cylinder (see Fig. 4), therefore, it can be inferred from the temporal coefficient spectra that vortex
330 shedding is likely to take place in the wake of the vibrissa cylinder along the saddle plane. Indeed, more evidence
331 of vortex shedding behind the vibrissa cylinder can be seen in the POD modes, both in saddle and nodal planes, as
332 will be shown next in Fig. 11. However, only the temporal coefficient spectra associated with POD mode 3 of the
333 vibrissa cylinder in the nodal plane exhibits a low-magnitude peak at St = 0.25, while no obvious tonal peaks can be
334 observed in the first two modes. With reference to the much weaker fluctuating velocity fields presented earlier for the
335 vibrissa cylinder, it can be argued that unlike the large-scale coherent vortex shedding found in circular and elliptical
336 cylinders, the vortex shedding of the vibrissa cylinder is more three-dimensional and sufficiently weak in strength so
337 that it is evenly spread into the first few POD modes in the nodal plane rather than being the most energetic flow
338 structures captured by the first mode pair.
339 Figure 11 shows POD modes 1 and 3 in the vertical mid-plane (z/D = 0) of the circular and the elliptical cylinders
340 as well as in the saddle and nodal planes of the vibrissa cylinder, where ψ is the magnitude of the modes. The POD
341 mode 2 is not presented as it contains similar information as first mode (i.e., a mode pair). Note that for each mode,
342 the maximum absolute magnitude of the mode is normalised to unity. Moreover, similar to the velocity field results,
343 the upper and lower part of each sub-figure is plotted with the u- and the v-velocity component, respectively. The
344 first modes of all three cylinders, as seen in Fig. 11 (a1-d1), exhibit the flow structures related to vortex shedding,
345 agreeing well with the previous studies [60, 64, 65]. The u-velocity component is anti-symmetric, and the v-velocity
346 component is symmetric about the centerline (y/D = 0). Both velocity components consist of alternating positive
347 and negative regions, corresponding to the presence of energetic coherent structures. These coherent structures, i.e.,
348 vortical structures from the vortex shedding, forms close to the leeward of the cylinders and subsequently grows in size
349 along the flow direction. It is also worthwhile to compare the location where the first alternating structure appears
350 in the POD modes of each cylinder. Clearly, the vortex shedding initiates closest to the cylinder for the elliptical
351 case, while a noticeable delay in the formation of vortical structures can be observed in the nodal plane of the vibrissa
352 cylinder. The circular cylinder and the saddle plane of the vibrissa cylinder show similar downstream distances.
353 The alternating coherent structures are also well captured in the POD mode 3 in the nodal plane of the vibrissa
354 cylinder, which means that its energy related to vortex shedding is more evenly distributed in the first few POD
355 modes, consistent with the observation from the temporal coefficient spectra shown in Fig. 10. For the circular and
356 elliptical cylinders, the POD mode 3 distribution is symmetric for the u-velocity component and anti-symmetric for
 12

mode 1

(a1) (b1) (c1) (d1)

u component

v component

mode 3

(a2) (b2) (c2) (d2)

u component

v component

Circular Elliptical Vibrissa-nodal Vibrissa-saddle

FIG. 11. POD mode 1 (1st row) and mode 3 (2nd row) in the mid plane of (a1, a2) the circular cylinder case and (b1, b2) the
elliptical cylinder case, and in the (c1, c2) nodal and (d1, d2) saddle planes of the vibrissa cylinder case. The upper and lower
parts of each figure represent the u component and the v component respectively.

0.4
Cylinder
Elliptical
Vibrissa
0.3
i

0.2
/
i

0.1

0
1 2 3 4 5 6 7 8 9 10
modes

FIG. 12. Eigenvalue distribution of the first 10 POD modes in horizontal planes

357 the v-velocity component about the centerline, which might be the symmetrical vortical structures [66] related to the
358 first harmonic of the vortex shedding [58, 64].
359 The normalised eigenvalue distribution associated with the first 10 POD modes in horizontal planes is shown in
360 Fig. 12 for all three cylinders. Note that the horizontal plane is taken close to the cylinder shear layers, which is at
361 y/D = 0.5 for the circular cylinder and at y/D = 0.36 for the elliptical and the vibrissa cylinders. Similar to the
362 results from the vertical planes, the first two POD modes of the circular and elliptical cylinders contain over 30% of
363 the total fluctuating energy of the flow, thus playing a dominant role in the near-wake flow development. However,
 13

364 their proportions to the total energy drop significantly compared to their counterparts in the vertical plane, shown
365 in Fig. 9. The decrease can be explained by the more extensive three-dimensional flows along the horizontal planes,
366 making the high-order POD modes more energetic, as discussed in Sec. III B. Similar reasoning applies to the vibrissa
367 cylinder - with stronger three-dimensional flow in the horizontal plane, the energy will be more evenly distributed to
368 high-order POD modes than that of the vertical plane, which is confirmed by the eigenvalue distributions.

(a) (b) (c) Circular

50 50 50
Elliptical
Vibrissa
40 40 40

10log(kk)

10log(kk)
10log(kk)

30 30 30

20 20 20

10 10 10

0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
f D/U¥ f D/U¥ f D/U¥

FIG. 13. Spectrum of temporal coefficients of (a) mode 1, (b) mode 2, (c) mode 3 in horizontal planes.

369 Figure 13 shows the power spectral density of temporal coefficients associated with the first three POD modes in
370 horizontal planes. Consistent with the results in the vertical planes, the temporal coefficients from other POD modes
371 are omitted as they contains a very small fraction of the total energy. For the circular cylinder and elliptical cylinders,
372 the first three modes are all related to their vortex shedding with tonal peaks at St = 0.21 and St = 0.36. The
373 presence of vortex shedding peaks in all three modes, similar to the temporal coefficient spectra in the saddle plane
374 of the vibrissa cylinder (see Fig. 10), which means the three-dimensionality of the vortex shedding is captured by the
375 first few POD modes. Also expected is the temporal coefficient spectra of the vibrissa cylinder case displaying a hump
376 at St = 0.25 in POD modes 1 to 3, corresponding to the three-dimensional and weakened vortex shedding process.
377 Figure 14 shows the first three POD modes in the horizontal shear-layer planes for all three cylinders. Here, a
378 single u-velocity component is selected as it is sufficient to highlight the key observations from these POD modes.
379 The POD mode 1 of the circular and elliptical cylinders exhibit clear vortex shedding structures, as indicated by the
380 alternating positive and negative regions as shown in Figs. 14 (a1) and (b1). The vortical structures retain a relatively
381 uniform shape along the spanwise direction. The POD mode 1 of the vibrissa cylinder on the other hand sees the
382 alternating coherent structures only occurring behind the saddle region, as shown in Figs. 14 (c1). It can be inferred
383 from the POD mode 1 that the vortex shedding in the nodal plane is weaker even when compared with that in the
384 saddle plane, which is essentially a consequence of the spanwise nodal-to-saddle flow as well as the potential phase
385 differences between the two planes introduced by wavy shape of the cylinder.
386 The POD mode 2 of the circular and the elliptical cylinders provides similar observations as their POD mode 1, while
387 the alternating coherent structures becomes less clear in the POD mode 2 of the vibrissa cylinder in Fig. 14 (c2), which
388 can possibly be explained by the presence of significant velocity fluctuations caused by the three-dimensional flow
389 structures. Subsequently, the alternating coherent structures re-emerges in POD mode 3. Furthermore, according
390 to velocity field results presented in Figs. 6 and 7, the near-wake of the elliptical cylinder tends to be more two-
391 dimensional than that of the circular case. Therefore, the two-dimensional coherent vortex shedding (i.e., uniform in
392 spanwise direction) is contained primarily in the first two energetic POD modes for the elliptical cylinder rather than
393 distributed over the first three POD modes for the circular cylinder, while the POD mode 3 of the elliptical cylinder
394 corresponds more to the three-dimensional vortex shedding process.
395 From the thorough POD analysis, it is evident that both the energy distribution of the dominant POD modes as
396 well as the POD modes themselves of the vibrissa cylinder significantly differ from those of the circular and elliptical
397 counterparts. Moreover, based on spectral analysis of the temporal coefficients and the corresponding POD modes,
398 vortex shedding is likely to occur in the wake along both nodal and saddle planes; however, such vortex shedding is
399 highly three-dimensional and much weaker in strength, which is directly correlated to the suppression of primary tone
400 of the cylinder in the far-field.
 14

(a1) (b1) (c1)

mode 1

(a2) (b2) (c2)

mode 2

(a3) (b3) (c3)

mode 3

Circular (y/D=0.5) Elliptical (y/D=0.36) Vibrissa (y/D=0.36)

FIG. 14. POD mode 1 (1st row), mode 2 (2nd row), and mode 3 (3rd row) for the u component in the y/D = 0.5 plane
of (a1-a3) the circular cylinder case, and in the y/D = 0.36 planes of (b1-b3) the elliptical cylinder and (c1-c3) the vibrissa
cylinder cases.

401 D. Time-Frequency analysis

402 1. Wavelet analysis

403 The POD results reveal the presence of vortex shedding in the near-wake of the vibrissa cylinder. However, the
404 ‘near disappearance’ of the primary tone associated with the fundamental vortex shedding frequency suggests that
405 the vortex shedding process of the vibrissa cylinder can be less uniform both in space (different in the saddle and
406 nodal planes) and in time. To examine the latter requires the study of the frequency characteristics localized in
407 time. Therefore, the wavelet analysis was then conducted on the near-field velocities to provide knowledge of the flow
408 development over time.
409 Figure 15 (a1-d1) shows the scalogram of the v-velocity wavelet transform at a downstream location of x/D = 3,
410 y/D = 0.35 in the mid-planes (z/D = 0) of the circular cylinder and elliptical cylinder cases, and in the nodal and
411 saddle planes of the vibrissa cylinder case. A continuous wavelet transform with ‘Morse’ analytical wavelet function,
412 a typical symmetry parameter of 3 and a time-bandwidth of 60 [67] is applied to the velocity time-series. The regions
413 of large magnitude of the wavelet coefficient, |Wv |, appears to concentrate around the vortex shedding frequencies
414 for both circular and elliptical cylinders. Since larger magnitude denotes more flow energy contents associated with
415 the frequency at a given time instance, this is expected as continuous vortex shedding takes place in the near-wake
416 of the cylinders. However, the regions of large coefficient magnitudes are considerably more wide spread across the
417 frequencies for the vibrissa cylinder, especially in the nodal plane.
418 In order to understand the temporal consistency of the most energetic flow structures for the three different cylinders,
 15

(a1) (a2) (b1) (b2)

0.3 0.3

fD/U¥
fD/U¥
0.2 0.2

0.1 0.1

0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5

Time [s] Time [s]

(c1) (c2) (d1) (d2)

0.3 0.3

fD/U¥
fD/U¥

0.2 0.2

0.1 0.1

0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5

Time [s] Time [s]

FIG. 15. (a1-d1) Wavelet transform scalogram and (a2-d2) time history of the frequency for the maximum coefficient magnitude
at x/D = 3, y/D = 0.35 point in the mid plane of (a1, a2) the circular cylinder and (b1, b2) elliptical cylinder cases, and in
the (c1, c2) nodal and (d1, d2) saddle planes of the vibrissa cylinder case.

FIG. 16. Probability of occurrence of the maximum wavelet transform coefficient at x/D = 3, y/D = 0.35 for (a) circular
cylinder, (b) elliptical cylinder, (c) nodal plane and (d) saddle plane of the vibrissa cylinder. The percentage indicates the
probability of occurrences within the Stm ± 0.013 threshold.

419 i.e., to assess whether vortex shedding is constantly present in the flow field assuming that it is the most energetic and
420 unsteady flow structures, the Strouhal number at which the maximum coefficient (Stm ) is found at each time instance
421 are extracted. The frequency-time histories of the maximum coefficients are shown in Figs. 15 (a2-d2), illustrating
422 the changes of the dynamics of the flow in time. Firstly, it is useful to mention that the mean Stm of the maximum
423 coefficients are at Stm = 0.21, 0.35, 0.26, and 0.24 for the circular, the elliptical, the nodal and saddle planes of the
424 vibrissa cylinder, respectively, which agree very well with the vortex shedding frequencies of the respective cylinders
425 (see both the far-field acoustic PSD and the POD temporal coefficients spectra in Figs. 4 and 10), reaffirming that the
426 vortex shedding is the most energetic flow development in the cylinder near-wake. Secondly, taking the probability
427 density function of these time histories with reference to the St, it can be seen from Fig. 16 that while both circular
428 and elliptical cylinders show a relatively constant St centering around the vortex shedding Strouhal number, the
429 nodal and saddle planes of the vibrissa cylinder carry a full range of Strouhal numbers from St = 0.1 to 0.36, which
430 the maximum coefficients occur. Taking St = Stm ± 0.013 as a threshold, only 1% of the time instances sees the
431 oscillation of the St of the maximum coefficients exceeding the threshold, while it is approximately 9% for the elliptical
432 cylinder. Remarkably, the percentage increases to 55% and 85%, respectively, for the saddle and nodal planes of the
433 vibrissa cylinder. Moreover, unlike the circular and elliptical cylinders, there is no consistent vortex shedding for the
 16

434 vibrissa cylinder throughout the measurement duration, as seen by the deviation of the maximum coefficients in time,
435 implying highly transient, three-dimensional, and weak vortex shedding.

436 2. Near-field velocity power spectral density

437 To further reveal the energy-frequency information of the near-wake flow, the power spectral density of the velocity
438 fluctuation are presented at different streamwise locations from x/D = 1 to x/D = 5 at (y/D = 0.5, z/D = 0) point in
439 the circular cylinder case, and at (y/D = 0.36, z/D = 0) point in the elliptical cylinder case, and at nodal and saddle
440 points in y/D = 0.36 plane of the vibrissa cylinder case, as shown in Fig. 17. In general, the broadband components
441 u velocity PSD (PSDu ) and w velocity PSD (PSDw ) are comparable regardless of the cylinder shape. The PSDu in
442 the saddle plane of the vibrissa cylinder has higher broadband components than other cases at x/D = 1, which is
443 consistent with the urms results shown in Fig. 7 (c3) and may be attributed to the intense velocity fluctuations of
444 the three-dimensional flow in the shear layer. On contrary, the PSDu and PSDw in the nodal plane of the vibrissa
445 exhibits very low values at x/D = 1, increasing along the streamwise direction due to the flow mixing with the
446 saddle region. The primary tonal peaks associated with the fundamental vortex shedding frequency are only visible
447 in the PSDu spectra. These peak Strouhal number are at St =0.21, 0.35, 0.24 for the circular, the elliptical, and
448 the vibrissa cylinders case, similar to those in the far-field acoustic and POD results. It is worthwhile to mention
449 that the tonal peak first occurs in the saddle plane of the vibrissa cylinder case at x/D = 3 rather than x/D = 1
450 for the circular and elliptical cylinders, corroborating a delayed vortex shedding process. Moving from x/D = 3 to
451 further downstream of x/D = 5, the tonal peak reduces because the vortex shedding gradually weakens in the flow
452 direction. The higher tonal peak magnitudes of the circular and elliptical cylinders reaffirm that their vortex shedding
453 and associated velocity fluctuations are stronger than the vibrissa case.

(a1)-20 (b1) -20 (c1) -20

-25 -25 -25
10log(Fu'u'/U¥2 ) [dB/Hz]

10log(Fu'u'/U¥2 ) [dB/Hz]

10log(Fu'u'/U¥2 ) [dB/Hz]
-30 -30 -30
-35 -35 -35
-40 -40 -40
-45 -45 -45
-50 -50 -50
-55 -55 -55
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
f D/U¥ f D/U¥ f D/U¥
(a2)-20 (b2) -20 (c2) -20
Circular
-25 -25 -25 Elliptical
10log(Fw'w'/U2¥) [dB/Hz]

10log(Fw'w'/U2¥) [dB/Hz]

10log(Fw'w'/U2¥) [dB/Hz]

-30 Vibrissa-nodal
-30 -30
Vibrissa-saddle
-35 -35 -35
-40 -40 -40
-45 -45 -45
-50 -50 -50
-55 -55 -55
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
f D/U¥ f D/U¥ f D/U¥
x/D=1 x/D=3 x/D=5

FIG. 17. (a1-c1) u velocity PSD and (a2-c2) w velocity PSD with different x/D positions at (y/D = 0.5, z/D = 0) point in
the circular cylinder case, and at (y/D = 0.36, z/D = 0) point in the elliptical cylinder case, and at nodal and saddle points
in y/D = 0.36 plane of the vibrissa cylinder case.

454 The PSDu and PSDw along the span at x/D = 3 are presented in Fig. 18 for all three cylinders, to disclose the
455 spanwise variations of the frequency characteristics of the flow. The distributions of PSDw are similar to those of
456 PSDu but with much smaller magnitudes. The relatively constant and uniform PSDu distribution along the span of
457 the circular and elliptical cylinders indicate that their vortex shedding processes are high-correlated in the spanwise
458 direction (i.e., more two-dimensional). For the vibrissa cylinder, both PSDu and PSDw are comparable in magnitudes,
459 demonstrating again the highly three-dimensional vortex shedding process. Such three-dimensional and weaker vortex
 17

(a1) (b1) (c1)

(a2) (b2) (c2)

Circular (y/D=0.5) Elliptical (y/D=0.36) Vibrissa (y/D=0.36)

FIG. 18. u velocity PSD (1st row) and w velocity PSD (2nd row) at x/D = 3 along the span in the y/D = 0.5 plane of
(a1, a2) the circular cylinder case, and in the y/D = 0.36 planes of (b1, b2) the elliptical cylinder and (c1, c2) the vibrissa
cylinder cases.

460 shedding of the vibrissa cylinder hence has direct influence on the significant reduction of the tonal peak as well as
461 the overall sound pressure levels of the far-field acoustics.

462 IV. Conclusions

463 This work presents a detailed experimental study on the far-field acoustics and the near-field flow dynamics of a
464 vibrissa cylinder to uncover the physical mechanisms for noise suppression. Two more cylinders, a cicrular and an
465 elliptic, have also been measured as reference and comparison to the vibrissa cylinder. All three cylinders are designed
466 to have an equivalent volume at a given spanwise length. The far-field acoustics were captured by a microphone array
467 and the near-field flows were quantified by time-resolved particle image velocimetry. Relating the acoustic and flow
468 field results, several major concludions can be drawn from the present investigation:
469 (1) The vibrissa cylinder reduces the far-field noise considerably, especially for the primary tonal peak associated
470 with the fundamental vortex shedding frequency. Instead of a clear tone, the vortex shedding noise manifests as
471 a minor hump, barely exceeding the broadband component. Consequently, the OASPL of the vibrissa cylinder is
472 approximately 13.2 dB and 8.3 dB lower than that of the circular and the elliptical cylinders, respectively.
473 (2) The time-averaged and rms velocities results and POD modes in both vertical and horizontal planes show that
474 although vortex shedding are present in the wake of all three cylinders, the vibrissa cylinder has a much weakened and
475 more three-dimensional vortex shedding with a significantly smaller energy fraction to the total fluctuation energy in
476 the flow, compared to the circular and elliptic cases. Furthermore, there is a clear reduction in the size of the wake
477 region and the intensity of the velocity fluctuations for the vibrissa and elliptical cylinders, correlating to the reduced
478 far-field noise.
479 (3) From the wavelet analysis, the temporal variations of the dominant flow structures in the wake of the cylinders
480 are examined. For both the circular and elliptical cylinders, the maximum wavelet coefficient are primarily associated
481 with the fundamental vortex shedding process whereas the frequencies are much more wide spread for the vibrissa
482 cylinder. This is particular true for the nodal plane where no dominant frequency is observed over the sampling
 18

483 duration. Additionally, the velocity PSD demonstrates that the characteristics of the velocity spectra are different
484 between the the nodal and saddle planes of the vibrissa cylinder. Together with the nodal-to-saddle flow patterns
485 observed from the spanwise velocity fields, they contribute to a three-dimensional, less coherent vortex shedding both
486 in temporal and spatial domains behind the vibrissa cylinder, leading to the marked reduction of the far-field noise
487 of vibrissa case.

488 Acknowledgements

489 The first author’s PhD grant is funded by the China Scholarship Council (CSC)-University of Bristol (UoB) PhD
490 Scholarship.
491

492 Declaration: The authors declare no conflict of interest for the present work.

493 [1] C. Norberg, Effects of Reynolds number and a low-intensity freestream turbulence on the flow around a circular cylinder,
494 Tech. Rep. ISSN 02809265 (Chalmers University, Goteborg, Sweden, Technological Publications, 1987).
495 [2] C. Norberg, Fluctuating lift on a circular cylinder: review and new measurements, Journal of Fluids and Structures 17,
496 57 (2003).
497 [3] P. Parnaudeau, J. Carlier, D. Heitz, and E. Lamballais, Experimental and numerical studies of the flow over a circular
498 cylinder at Reynolds number 3900, Physics of Fluids 20, 085101 (2008).
499 [4] A. Capone, C. Klein, F. Di Felice, and M. Miozzi, Phenomenology of a flow around a circular cylinder at sub-critical and
500 critical Reynolds numbers, Physics of Fluids 28, 074101 (2016).
501 [5] R. Maryami, S. Ali, M. Azarpeyvand, and A. Afshari, Turbulent flow interaction with a circular cylinder, Physics of Fluids
502 32, 015105 (2020).
503 [6] R. Maryami, M. Azarpeyvand, A. Dehghan, and A. Afshari, An experimental investigation of the surface pressure fluctu-
504 ations for round cylinders, Journal of Fluids Engineering 141 (2019).
505 [7] M. J. E. Yazdi and A. B. Khoshnevis, Experimental study of the flow across an elliptic cylinder at subcritical Reynolds
506 number, The European Physical Journal Plus 133, 533 (2018).
507 [8] D. Kumar, M. Mittal, and S. Sen, Modification of response and suppression of vortex-shedding in vortex-induced vibrations
508 of an elliptic cylinder, International Journal of Heat and Fluid Flow 71, 406 (2018).
509 [9] S. H. Nam and H. S. Yoon, Effect of the wavy geometric disturbance on the flow over elliptic cylinders with different aspect
510 ratios, Ocean Engineering 243, 110287 (2022).
511 [10] Y. Lin, H. Bai, M. M. Alam, W. Zhang, and K. Lam, Effects of large spanwise wavelength on the wake of a sinusoidal
512 wavy cylinder, Journal of Fluids and Structures 61, 392 (2016).
513 [11] H. Bai, B. Zang, and T. New, The near wake of a sinusoidal wavy cylinder with a large spanwise wavelength using
514 time-resolved particle image velocimetry, Experiments in Fluids 60, 1 (2019).
515 [12] H. Bai, M. M. Alam, N. Gao, and Y. Lin, The near wake of sinusoidal wavy cylinders: Three-dimensional POD analyses,
516 International Journal of Heat and Fluid Flow 75, 256 (2019).
517 [13] W. Hanke, M. Witte, L. Miersch, M. Brede, J. Oeffner, M. Michael, F. Hanke, A. Leder, and G. Dehnhardt, Harbor seal
518 vibrissa morphology suppresses vortex-induced vibrations, Journal of Experimental Biology 213, 2665 (2010).
519 [14] W.-L. Chen, X.-W. Min, and Y.-J. Guo, Performance of seal vibrissa-inspired bionic surface in suppressing aerodynamic
520 forces and vortex shedding around a circular cylinder, Ocean Engineering 260, 112032 (2022).
521 [15] D. J. Tritton, Physical fluid dynamics (Springer Science & Business Media, 1997) Chap. 3, pp. 21–34.
522 [16] E. Houghton and P. Carpenter, eds., Aerodynamics for engineering students (Todd Green, 2017) Chap. 3, pp. 173–180.
523 [17] J. Gerrard, Measurements of the sound from circular cylinders in an air stream, Proceedings of the Physical Society. Section
524 B 68, 453 (1955).
525 [18] O. Phillips, The intnesity of Aeolian tones, Journal of Fluid Mechanics 1, 607 (1956).
526 [19] J. Revell, R. Prydz, and A. Hays, Experimental study of aerodynamic noise vs drag relationships for circular cylinders,
527 AIAA Journal 16, 889 (1978).
528 [20] M. Lighthill, On sound generated aerodynamically I. General theory, Proceedings of the Royal Society of London. Series
529 A. Mathematical and Physical Sciences 211, 564 (1952).
530 [21] M. Lighthill, On sound generated aerodynamically II. Turbulence as a source of sound, Proceedings of the Royal Society
531 of London. Series A. Mathematical and Physical Sciences 222, 1 (1954).
532 [22] N. Curle, The influence of solid boundaries upon aerodynamic sound, Proceedings of the Royal Society of London. Series
533 A. Mathematical and Physical Sciences 231, 505 (1955).
534 [23] B. Etkin, G. Korbacher, and R. T. Keefe, Acoustic radiation from a stationary cylinder in a fluid stream (aeolian tones),
535 The Journal of the Acoustical Society of America 29, 30 (1957).
536 [24] N. Alemdaroālu, J. Rebillat, and R. Goethals, An aeroacoustic coherence function method applied to circular cylinder
537 flows, Journal of Sound and Vibration 69, 427 (1980).
 19

538 [25] H. Fujita, H. Suzuki, A. Sagawa, and T. Takaishi, The aeolian tone characteristics of a circular cylinder in high Reynolds
539 number flow, in 5th AIAA/CEAS Aeroacoustics Conference and Exhibit (1999) p. 1849.
540 [26] H. Fujita, H. Suzuki, A. Sagawa, and T. Takaishi, The Aeolian tone and the surface pressure in high Reynolds number
541 flow, in 6th Aeroacoustics Conference and Exhibit (2000) p. 2002.
542 [27] Y. Oguma, T. Yamagata, and N. Fujisawa, Measurement of sound source distribution around a circular cylinder in a uniform
543 flow by combined particle image velocimetry and microphone technique, Journal of Wind Engineering and Industrial
544 Aerodynamics 118, 1 (2013).
545 [28] C. Zheng, P. Zhou, S. Zhong, and X. Zhang, Experimental investigation on cylinder noise and its reductions by identifying
546 aerodynamic sound sources in flow fields, Physics of Fluids 35, 035103 (2023).
547 [29] R. Zamponi, F. Avallone, D. Ragni, and S. van der Zwaag, On the aerodynamic-noise sources in a circular cylinder coated
548 with porous materials, in 28th AIAA/CEAS Aeroacoustics 2022 Conference (2022) p. 3042.
549 [30] N. Fujisawa, K. Hirabayashi, and T. Yamagata, Aerodynamic noise reduction of circular cylinder by longitudinal grooves,
550 Journal of Wind Engineering and Industrial Aerodynamics 199, 104129 (2020).
551 [31] M. R. Lekkala, M. Latheef, J. H. Jung, A. Coraddu, H. Zhu, N. Srinil, B.-H. Lee, et al., Recent advances in understanding
552 the flow over bluff bodies with different geometries at moderate Reynolds numbers, Ocean Engineering 261, 111611 (2022).
553 [32] H. Jie and Y. Z. Liu, Large eddy simulation and proper orthogonality decomposition of turbulent flow around a vibrissa-
554 shaped cylinder, International journal of heat and fluid flow 67, 261 (2017).
555 [33] T. Dunt and J. A. Franck, Flow over a seal whisker inspired geometry at swept back angles, in AIAA AVIATION 2022
556 Forum (2022) p. 3982.
557 [34] S. Wang and Y. Z. Liu, Flow structures behind a vibrissa-shaped cylinder at different angles of attack: Complication on
558 vortex-induced vibration, International Journal of Heat and Fluid Flow 68, 31 (2017).
559 [35] H. S. Yoon, S. H. Nam, and M. I. Kim, Effect of the geometric features of the harbor seal vibrissa based biomimetic
560 cylinder on the flow over a cylinder, Ocean Engineering 218, 108150 (2020).
561 [36] S. Chu, C. Xia, H. Wang, Y. Fan, and Z. Yang, Three-dimensional spectral proper orthogonal decomposition analyses of
562 the turbulent flow around a seal-vibrissa-shaped cylinder, Physics of Fluids 33, 025106 (2021).
563 [37] C. Zhang, M. Sanjosé, and S. Moreau, Aeolian noise of a cylinder in the critical regime, The Journal of the Acoustical
564 Society of America 146, 1404 (2019).
565 [38] Y. Mayer, H. Jawahar, M. Szőke, S. Ali, and M. Azarpeyvand, Design and performance of an aeroacoustic wind tunnel
566 facility at the University of Bristol, Applied Acoustics 155, 358 (2019).
567 [39] G. Chen, X. Liu, B. Zang, and M. Azarpeyvand, The effect of the splitter plate on the aeolian tone mitigation, in 28th
568 AIAA/CEAS Aeroacoustics 2022 Conference (2022) pp. 3094–1–11.
569 [40] X. Liu, G. Chen, B. Zang, and M. Azarpeyvand, An experimental study of aerodynamic noise of the wavy and vibrissa
570 shaped cylinders, in 28th AIAA/CEAS Aeroacoustics 2022 Conference (2022) pp. 2982–1–15.
571 [41] B. Zang, Y. D. Mayer, and M. Azarpeyvand, Experimental investigation of near-field aeroacoustic characteristics of a
572 pre-and post-stall NACA 65-410 airfoil, Journal of Aerospace Engineering 34, 04021080 (2021).
573 [42] A. Celik, J. L. Bowen, and M. Azarpeyvand, Effect of trailing-edge bevel on the aeroacoustics of a flat-plate, Physics of
574 Fluids 32 (2020).
575 [43] Y. D. Mayer, B. Zang, and M. Azarpeyvand, Aeroacoustic investigation of an oscillating airfoil in the pre-and post-stall
576 regime, Aerospace Science and Technology 103, 105880 (2020).
577 [44] L. Li, P. Liu, Y. Xing, and H. Guo, Experimental investigation on the noise reduction method of helical cables for a circular
578 cylinder and tandem cylinders, Applied Acoustics 152, 79 (2019).
579 [45] W.-L. Chen, X.-W. Min, D.-L. Gao, A.-X. Guo, and H. Li, Experimental investigation of aerodynamic forces and flow
580 structures of bionic cylinders based on harbor seal vibrissa, Experimental Thermal and Fluid Science 99, 169 (2018).
581 [46] Y. Li, X.-n. Wang, Z.-w. Chen, and Z.-c. Li, Experimental study of vortex–structure interaction noise radiated from
582 rod–airfoil configurations, Journal of Fluids and Structures 51, 313 (2014).
583 [47] F. M. White, Fluid mechanics (McGraw-Hill, 1990) Chap. 5, pp. 316–319.
584 [48] K. Lyons, C. T. Murphy, and J. A. Franck, Flow over seal whiskers: Importance of geometric features for force and
585 frequency response, PloS ONE 15, e0241142 (2020).
586 [49] C. Zheng, P. Zhou, S. Zhong, and X. Zhang, On the cylinder noise and drag reductions in different Reynolds number
587 ranges using surface pattern fabrics, Physics of Fluids 35 (2023).
588 [50] J. Hardin and S. Lamkin, Aeroacoustic Computation of Cylinder Wake Flow, AIAA journal 22, 51 (1984).
589 [51] T. F. Geyer, Experimental evaluation of cylinder vortex shedding noise reduction using porous material, Experiments in
590 Fluids 61, 1 (2020).
591 [52] S. Wang and Y. Liu, Wake dynamics behind a seal-vibrissa-shaped cylinder: a comparative study by time-resolved particle
592 velocimetry measurements, Experiments in Fluids 57, 1 (2016).
593 [53] H. Bai, Z. Lu, R. Wei, Y. Yang, and Y. Liu, Noise reduction of sinusoidal wavy cylinder in subcritical flow regime, Physics
594 of Fluids 33 (2021).
595 [54] B. Chen, X. Yang, G. Chen, X. Tang, J. Ding, and P. Weng, Numerical study on the flow and noise control mechanism of
596 wavy cylinder, Physics of Fluids 34 (2022).
597 [55] C. Kato, A. Iida, Y. Takano, H. Fujita, and M. Ikegawa, Numerical prediction of aerodynamic noise radiated from low
598 Mach number turbulent wake, in 31st Aerospace Sciences Meeting (Reno, NV, 1993) p. 145.
599 [56] M. Ben Chiekh, M. Michard, M. Guellouz, and J. Béra, Pod analysis of momentumless trailing edge wake using synthetic
600 jet actuation, Experimental Thermal and Fluid Science 46, 89 (2013).
 20

601 [57] H. K. Jawahar, R. Theunissen, M. Azarpeyvand, and C. R. I. da Silva, Flow characteristics of slat cove fillers, Aerospace
602 Science and Technology 100, 105789 (2020).
603 [58] G. Chen, B. Zang, and M. Azarpeyvand, Numerical investigation on aerodynamic noise of flow past a cylinder with different
604 spanwise lengths, Physics of Fluids 35 (2023).
605 [59] R. Perrin, M. Braza, E. Cid, S. Cazin, P. Patrick, C. Mockett, T. Reimann, and F. Thiele, Coherent and turbulent process
606 analysis in the flow past a circular cylinder at high Reynolds number, Journal of Fluids and Structures 24, 1313 (2008).
607 [60] J. Weiss, A tutorial on the Proper Orthogonal Decomposition, in 2019 AIAA Aviation Forum (Dallas, Texas, United
608 States, 2019) pp. 1–21.
609 [61] L. Feng, J. Wang, and C. Pan, Proper Orthogonal Decomposition analysis of vortex dynamics of a circular cylinder under
610 synthetic jet control, Physics of Fluids 23, 014106 (2011).
611 [62] B. Noack, K. Afanasiev, M. Morzyński, G. Tadmor, and F.Thiele, A hierarchy of low-dimensional models for the transient
612 and post-transient cylinder wake, Journal of Fluid Mechanics 497, 335 (2003).
613 [63] N. D. Katopodes, Free-surface flow (Butterworth-Heinemann, 2019) Chap. 5, pp. 385–386.
614 [64] L. Kourentis and E. Konstantinidis, Uncovering large-scale coherent structures in natural and forced turbulent wakes by
615 combining PIV, POD, and FTLE, Experiments in Fluids 52, 749 (2012).
616 [65] X. Ma, Hierarchical Galerkin and non-linear Galerkin models for laminar and turbulent wakes, Ph.D. thesis, Brown Uni-
617 versity (2001).
618 [66] C. Xia, Z. Wei, H. Yuan, Q. Li, and Z. Yang, Pod analysis of the wake behind a circular cylinder coated with porous
619 media, Journal of Visualization 21, 965 (2018).
620 [67] J. M. Lilly and S. C. Olhede, Generalized morse wavelets as a superfamily of analytic wavelets, IEEE Transactions on
621 Signal Processing 60, 6036 (2012).