Zhang 2019

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Aeroacoustics Conferences 10.2514/6.

2019-2717
20-23 May 2019, Delft, The Netherlands
25th AIAA/CEAS Aeroacoustics Conference

A study of shear-layer corrections and a tensioned fabric


wall for the localization of sound sources in wind tunnel

Jun Zhang,1 and Xunnian Wang 2


China Aerodynamics Research and Development Center, Mianyang, Sichuan, 621000, China

Junlong Zhang 3
China Aerodynamics Research and Development Center, Mianyang, Sichuan, 621000, China

Con Doolan4, Jeoffrey Fischer5 and Danielle Moreau 64


University of New South Wales, Sydney, NSW, 2032, Australia
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Shear layer causes phase and amplitude distortion of sound during the propagation of
sound from the inside to the outside of open-jet flow. The phase of sound must be corrected
for accurate localization of sound sources using a microphone array. In this study, a new
shear-layer correction method is presented, and the difference between the accuracy of
various correction methods is analyzed. The results show that when the Mach number is
smaller than 0.3 and the measurement angle ranges from 40 to 140 degrees, the relative
difference of phase correction values using various methods is trivial in the two-dimensional
case. However, the relative difference of phase-correction values between 2D and 3D
correction methods exceeds 5% under shallow measurement angles and high flow speeds.
Experimental studies were performed in CARDC and UNSW respectively. An inflow speaker,
two closely located out-of-flow microphones and an array were used to measure the phase
distortion and the location of the sound source. The measured data of the phase of sound
agrees well with calculation values, and the stream-wise drift of source locations caused by the
shear layer is compensated for by using any shear-layer correction method. A wall-mounted
speaker and an array were used to study the effect of a tensioned fabric wall on the distortion
of sound. The results show that the spectral broadening of sound is suppressed and the
amplitude of sound is greatly attenuated by the fabric compared to the open-jet flow case. For
sound source localization, the fabric just resembles an infinitely thin shear layer. The location
of the source is recovered with shear-layer corrections.

I. Nomenclature
αΘm = the measurement angle in the X-Z plane
αΘ = the convection angle in the X-Z plane
αΘ’ = the emission angle in the X-Z plane
αΘ0 = the refraction angle in the X-Z plane
Θm = the measurement angle in the X-Y plane
Θ = the convection angle in the X-Y plane
Θ’ = the emission angle in the X-Y plane
Θ0 = the refraction angle in the X-Y plane
Rm = the distance from the source position to the observer position
Rt = the half-width of the nozzle
h = the distance from the source position to the projection plane

1
Research fellow, Key Laboratory of Aerodynamic Noise Control, CARDC, jzhang@nudt.edu.cn, AIAA member.
2
Professor, State Key Laboratory of Aerodynamics, CARDC, skla_cardc@126.com, AIAA member.
3
PhD student, Key Laboratory of Aerodynamic Noise Control, CARDC, jlzhang@cardc.cn, AIAA member.
4
Professor, Mechanical and Manufacturing School, UNSW, c.doolan@unsw.edu.au, AIAA senior member
5
Research Associate, Mechanical and Manufacturing School, UNSW, jeoffrey.fischer@unsw.edu.au, AIAA member
6
Lecturer, Mechanical and Manufacturing School, UNSW, d.moreau@unsw.edu.au, AIAA senior member

Copyright © 2019 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
II. Introduction
Aeroacoustic measurements are often performed in open-jet wind tunnels. The shear layer lies between the core
flow and ambient stagnant flow, which will cause reflection, refraction, and scattering of sound [1-3]. It’s preferred
to put a microphone array outside of the flow to avoid flow induced noise. In this way, sound waves will pass through
the open-jet shear layer before reaching out-of-flow microphones. This study is focused on the localization of sound
sources in an open-jet wind tunnel. To obtain accurate estimates of the location and sound power of a noise source,
the shear layer effects must be corrected.
Miles [4] and Ribner [5] studied the propagation of sound through two overlapping flows with different speeds.
They established a useful relationship of pressure amplitude when sound crosses flow boundary. Amiet[6-7] derived
a two-dimensional (2D) shear-layer correction formulae for circular open-jet flow based on geometric acoustic theory.
A zero-thickness shear layer that followed the nozzle-lip line was assumed in Amiet’s model. Amiet’s method can
correct both the amplitude and the phase of sound, and it has been widely used since 1975. As mentioned, the method
is strictly only suitable for 2D, where the microphone and the source lie on the same plane. Later, Schlinker [8]
extended Amiet’s 2D correction to “2.5D” and the effect of an off-center source was considered. In Schilinker’s model,
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

a circular wind tunnel nozzle shape was assumed, which explains the reason why this method was called 2.5D. Morfey
[9] presented an analytical solution based on geometrical ray tracing. Due to the complexity of the expression, his
method is less widely used as Amiet’s. More recently, Koop [10] proposed a 2D simplified flow model (SFM) for the
correction of sound phase for any type of shear layer, e.g., spatially curved shear layer (due to the sway of flow),
which was extended to full 3D by Porteous et al. [11] and by Zhang et al. [12] independently. The extended Koop’s
model can only correct the source-receiver delay time, which means the amplitude correction is not supported. Sarradj
[13] proposed a fast ray casting method that considers the thickness of shear layer, and found that the thickness
contribution was not important. To simplify the delay time calculation, Orlemans [14] and Sijtsma [15] presented the
so-called average Mach number correction method (AM), which was believed to be accurate up to Mach number
Ma=0.2. AM assumes the space between source and microphone are immersed in an averaged flow, whose speed
equals the core-flow speed times the ratio of source-to-shear-layer and source-to-receiver distance. A very simple
formula to estimate the source-position shift was given by Padois [16], which was named as “M-H” method by Hansen
et al. [17]. Very recently, Amiet’s method was extended by Delfs [18] to full 3D, who followed the work of Schlinker
[8] in the derivation. However, Delfs’ work is not formally published.
As mentioned above, many shear-layer correction methods have been proposed to correct the phase or/and
amplitude distortion of sound, but few of them are satisfactory for 3D rectangular flow, for example, the open-jet flow
from 5.5m by 4m aeroacoustic wind tunnel (FL-17) in China Aerodynamics Research and Development Center
(CARDC). In this study, a new 3D shear layer angle correction method is presented. And the relative difference of the
accuracy and calculation speed of various existing shear-layer correction methods are discussed. The results are
meaningful for aeroacoustic test conducted in open-jet or closed jet wind tunnel.

III. Shear-layer correction methods

Shear-layer correction methods can be generally divided into two categories, i.e., frequency-domain methods and
time-domain methods. Time-domain methods, such as those use solution of the linearized Euler equations (LEE) [19]
or CAA [20] based solution is very time consuming. For large wind tunnels and large microphone arrays, the
computational cost is so large that super computer might be required. Amiet [7] found that if the wave length of sound
is larger than the characteristic dimension of nozzle, e.g., the frequency of sound larger than 100Hz for FL-17 wind
tunnel, there’s no obvious difference between using frequency-domain or time-domain methods. Therefore, in this
work, only frequency-domain methods are addressed, such as Amiet’s method, ray tracing and some empirical models
(AM, SFM and M-H). In the following, a new 3D shear layer angle and amplitude correction method for rectangular
nozzle are presented. In this paper, the angle correction is derived independently, while the amplitude correction
follows the work of Delfs[18].

A. 3D shear-layer correction
A schematic of 3D shear-layer correction is shown in Fig. 1. Cartesian coordinate system is used. The flow speed
v is parallel to the X axis, and the direction of flow is from left to right. A planar shear layer A’B’C’D’ is located
between the mean flow and ambient flow. A sound source S is situated in the mean flow. O is in the observation plane
ABCD. The red line denotes the actual ray path, the blue line denotes the line-of-sight ray path and the green line
denotes the emission ray path without flow. α denotes the angle between an acoustic ray and the Z axis projected onto
the X-Z plane. αΘ’ is the angle between the emission ray SE’ and the Z axis, αΘm is the angle between the line-of-sight

2
ray path SO and the Z axis, and αΘ is the angle between the actual ray SJ and the Z axis. Θ denotes the angle between
the ray and X axis projected onto the X-Y plane. Θ’ is the angle between the emission ray and the X axis, Θm is the
angle between the line-of-sight ray path EO’ and the X axis, and Θ is the angle between the actual ray and the X axis.
Zero-degree angle denotes the upstream direction. The blue line SE’ represents the emission ray path without flow
and the red line SJO represents the actual ray path crossing the sheared flow. SJO’ is the projection of SJO in the
horizontal plane CDEF, which contains the ray crossing point J and is perpendicular to the shear layer plane. SE=h
represents the distance from S to CDEF. Rt is the distance from S to shear layer and Rd is the distance from shear layer
to O.
Z
A B
O Measurement plane
A’ B’
Shear-layer plane
O’
D α Θ0 C
Rd
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Θ0 J
Nozzle D’ E’
C’
Rt
ΘΘmΘ’ E F

αΘ
αΘm Y h
U αΘ’
X
S

Fig. 1 Sketch of the 3D shear-layer correction model

Assumptions of an infinitely thin shear layer and planar wavefront are made in the current derivation. Assume S
coincides with the origin of the coordinate system, and the coordinate of observation point O is (dx, dy, dz), then the
line-of-sight distance from S to O is , where d , 1 / is implied.
Based on the geometry shown in Fig.1, the following equations are derived:
dx 2  dy 2
sin  m  (1)
Rm
dy
sin  m  (2)
dx  dy 2
2

EJ Rt sin  1
sin     = (3)
SJ  Rt sin    h
2 2
 dz sin  
2

1  
 dy 
Similarly,
1
sin   '  (4)
2
 dz sin  ' 
1  
 dy 
1
sin  0  (5)
2
 dz sin 0 
1  
 dy 

3
Following the method of derivation of 2D convection relationship [18], in the 3D case, we have
sin   '  sin  '
tan   (6)
sin   '  cos  '  M
Based on Snell’s law, the following relationship is reached:
1 c0 ct
M   (7)
sin   '  cos   M sin  0  cos 0
'

Where, ct and c0 denote the sound speed in the core flow and ambient flow respectively. M is the Mach number of
flow. Combing Eq. (6) and (7), the following relationship is established:

2
 
 c0 ct  1  sin 2  0
   sin 2  0  cos 2  0  (8)
 1 dz 2 sin 2  0
M  1
 sin   '  cos  ' 
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

dy 2
And from the geometry shown in Fig.1, the following relationships are given
Rt  Rd
tan  m  (9)
Rt  tan 1   Rd  tan 1 0
Rt  Rd
dx  (10)
tan m
Once the locations of the source, shear layer and observation point are given, i.e., variables c0/ct, M, dy, dz, Rt and
Rd are known, the above nonlinear equations can be solved iteratively using Newton’s method. However, to speed up
the calculation process, the following steps are followed instead:
1) for an arbitrary input value Θ’, αΘ’ is solved using Eq. (4);
2) given Θ’ and αΘ’, Θ is solved using Eq. (6);
3) given Θ’ and αΘ’, Θ0 is solved using Eq. (8);
4) given Θ and Θ0, Θm and dx are solved using Eq. (9) and (10);
5) input another Θ’, Θm and dx are obtained accordingly following steps from 1) to 4).
Therefore, for an arbitrary observer location Θm and dx, Θ’, Θ and Θ0 can be solved using linear interpolation.
Although the above process is not mathematically strict, if samples of data collection Θ’ = f (Θm, dx) are large enough,
the precision of angle calculation will be satisfactory. The source-to-receiver delay time (SRDT) for the open-jet flow
is given as follows:
Rt ct Rd c0
t   (11)
sin  sin  ' sin 0 sin  0
'

The present angle-correction method extends Amiet’s 2D angle-correction method to 3D with the consideration
of the change of angle in the X-Z plane. Once angle-correction values are obtained, the amplitude correction can easily
be calculated. For pressure-amplitude correction, we directly use the formula obtained by Delfs [18], which is given
as follow:
2
 2 
h J M    cos  0  sin 0  cos  0 1  M  cos 0  
2 2 2
PC2
 (12)
PM2 zO J 0 
4  2  cos 2  0 
Where, Pc denotes the corrected pressure at point O’ and PM denotes the measurement pressure at point O. JM and JO
are auxiliary variables.
z
 2  1  M  cos 0   cos2 0 , zO*  O
2
(13)
h

 
h 2 sin 0 sin  0
1  1  M 2  cos 2     2  1  M 2  cos    M  cos 2  
2
J0  (14)

 2  cos 2  0
 0   0  0

J M  h 2 sin  0 sin  0  C1C2  C3C4  (15)

4
1  1  M 2  cos 2  0 z *
O  1 sin 2  0
C1   (16)
   sin   cos  
32 32
2
 cos 2  0 2
0
2
0

C2 
2

 z  1 sin 
*
O
2
0
(17)
   sin   cos  
32 32
2
 cos  0
2 2
0
2
0

 1  M  cos 0  z  1 sin  sin 


 2
 M  cos  * 

C3  cos  0  
0 O 0 0
 32 
(18)
    sin   cos  
32
  2  cos 2 0
2
0
2
0 

C4 
1  M  cos  2
0 M

 z  1 cos 
*
O 0
(19)
  cos    sin   cos  
32 32
2 2 2 2
0 0 0
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

The above algebraic expressions are in analytical form. Thus, using the strategies given above, the equations can
be solved very quickly to obtain angle and amplitude correction values.

B. Ray tracing
In an inhomogeneous flow field with a velocity v(x, t), the wavefront of a point source is no longer spherical.
Assume a point xp is located on the wavefront, and the motion speed of xp can be given by
dx p
 v  xp ,t   n xp ,t  c  xp  (20)
dt
Where n(xp, t) is the unit vector normal to the wavefront, and c(xp) is the local speed of sound. The ray path is given
by the trajectory of xp, which could be traced to find out the propagation of sound. Normally, the ray tracing equations
are written as a system of ordinary differential equations (ODE):
dx p c 2 s
 v (21)
dt 
ds
 s  c  s     v    s    v (22)
dt
Where, s and Ω are auxiliary values, and s is called the “slowness vector”. Between these variables, the following
relations hold:
n 
s ,   1 v  s , s  (23)
c  v n c
For irrotational and isothermal flow with constant speed of sound, the first two terms on the right side of Eq. (22) are
zero. The ODEs can be solved using standard solver with adaptive step-size control, e.g., the 4th order Runge-Kutta
method. The sound-pressure amplitude along a ray path can be estimated using the Blokhintzev invariant [21], which
is written as:
2
pray  
x p  c  v  n  dAray
 const (24)
20 c02
Here, pray is the pressure at a point xp along the ray and dAray is the cross-sectional area of the ray tube. Ordinary ray
tracing is very time consuming as it needs to iteratively solve ODEs to make sure that the ray lands precisely on the
observation point (microphone sensor). However, the calculation process can be greatly improved by using ray casting
or spatial interpolation [13]. The mathematical expression of other classical shear-layer correction methods, such as
Amiet’s method and average Mach number method, can be found in Ref. [6-7] and [14-15] respectively. Interested
readers can refer to those literatures.

IV. Numerical calculations

A. Correction of SRDT
In the following, SRDT is calculated based on the geometry of FL-17 wind tunnel in CARDC [22]. The dimension
of the nozzle of FL-17 is 5.5m×4m. Assume a point source is located at the central part of open-jet flow, and the
coordinate of the source is xs=[0, 0]. The distance from source to microphone is 7m, and the measurement angle Θm
ranges from 40o to 140o. The measurement angle is defined as the angle between source-microphone direction and the

5
upstream direction. An infinitely thin shear layer is located between the source and far-field microphones, which
extends from the nozzle lip line to the downstream collector. The source, the shear layer and microphones are in the
same plane. Therefore, shear-layer correction is performed in 2D. The distance from source to shear layer is 2m, and
the distance from shear layer to the out-of-flow microphone is 5m. In the present calculation, the flow speed is 70m/s,
which is most frequently used speed for FL-17 wind tunnel.
The calculation results of SRDT are presented in Fig.2a. The blue and black solid line represent the average Mach
number method and Amiet’s method respectively. The red dashed and green dashed line represent the simplified flow
method and ray tracing respectively. We chose these four methods in the calculation, because they are much faster
than time-domain methods and can easily be integrated into beamforming algorithm. As can be seen from the figure,
four curves are nearly identical.

35
AM
Amiet
SFM
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

30 RT
t, ms

25
(b) AM  Amiet
 100
Amiet
(a)
20
40 60 80 100 120 140
m
, deg

SFM  Amiet RT  Amiet


(c) Amiet
 100 (d) Amiet
 100

Fig. 2 SRDT calculations according to different 2D shear-layer-correction methods. AM stands for average
Mach number method, Amiet stands for Amiet’s 2D method, SFM stands for the simplified flow method, and
RT stands for ray tracing. U0=70m/s, Rt=2m, and Rd=5m. a) the calculated SRDT values; b), c) and d) the
relative difference of various shear-layer correction method and Amiet’s method.

Then, Amiet’s method is chosen as a reference, and the relative difference between other methods and Amiet’s
method is calculated as below:
SLCi  Amiet
rel _ diff  %    100 (25)
Amiet
Where, ref_diff stands for relative difference in percentage, and SLCi stands for a specified shear-layer correction
method. Figure 2b-d shows the relative difference of SRDT given by various shear-layer-correction methods. Amiet’s
method is taken as a reference. SRDT are calculated under different flow Mach numbers and measurement angles.
From these results, it can be concluded that when the Mach number is smaller than 0.3 and the measurement angle
ranges from 40 to 140 degrees, the relative difference of SRDT for various shear-layer correction methods is trivial

6
(<1%) in the two-dimension case. Specifically, the relative difference between SFM and Amiet can be ignored
(<0.1%). Porteous et al. mathematically proved that SFM is equivalent to Amiet under 2D [11]. It should be noted
that the precision of RT and AM are quite similar when a zero-thickness shear layer is assumed. AM is a very simple
and useful empirical method for the correction of sound phase, e.g., the position-shift of source. However, amplitude
correction is not available in AM or SFM model.
Table 1. Computational cost for various shear-layer correction methods (s)
Array type AM Amiet SFM RT
Linear array 0.007 0.011 0.010 1.50
Planar array 0.009 0.062 0.106 18.33
In table 1, the computational cost for various shear-layer correction methods are presented. The calculation speed
of shear-layer-correction algorithm is important for on-line data processing, because sometimes it will take longer
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

than other modules in beamforming program, e.g., cross-spectrum calculation. The calculation is based on a single
source and a 11-sensor linear array or a 143-sensor planar array. A personal computer with 8G memory and four
2.7GHz i7 processor is used. An impression of the speed of various shear-layer correction methods is given by the
above results. In terms of calculation speed, AM is the fastest and RT is the slowest among these methods.

B. Correction of pressure amplitude


Results of pressure-amplitude correction are shown in Fig.3, both for a linear array and a planar array. For the
linear array, the correction is performed under 2D using Amiet’s method, which means the source, the shear layer and
the microphone lie in the same plane. The source-to-microphone distance equals four times the source-to-shear-layer
distance. It is assumed that the sound speed inside and outside of the flow remains unchanged. As shown in Fig.3a,
the value of amplitude correction increases as Mach number increases. Amplitude correction in the upstream direction
(for measurement angles larger than 90o) is larger than that in the downstream direction, which can be explained by
the reflection of sound energy in the upstream direction [23]. When the Mach number is larger than 0.2, the amplitude
correction exceeds 2dB in the upstream direction. Therefore, the shear layer effects must be included, especially for
the extrapolation of wind tunnel test data.
6
M=0.1
M=0.2
M=0.3
20log(Pc/Pm), dB

4
M=0.4

0
(a) (b)
-2
40 60 80 100 120 140
m
, deg

Fig. 3 Amplitude correction values for a linear array (left) and a planar array (right).
Figure 3b shows the amplitude-correction results for a planar array using the 3D correction method presented in
section III. The radius of the array is 1.5m, and 144 microphone sensors (denoted by dots) are randomly distributed
on the array surface. This planar array is used very often in CARDC during wind tunnel tests. In the calculation, the
distance from source to array surface is 4m, and the array center is located 2m upstream relative to the source. The
distance from source to shear layer is 2m, and the flow speed equals 80m/s (from left to right). As can be seen from
Fig.3b, the amplitude correction exceeds 2dB in the upstream-half array. To accurately estimate the source power,
amplitude correction can’t be ignored especially for large wind tunnels and high flow speeds. In the past, some
researchers suggested that amplitude correction was not important for beamforming [24], but that point should be
argued based on the present calculations.

7
C. Difference between 2D and 3D correction method
In the following, the difference between 2D and 3D correction is given. Calculation parameters are selected to be
the same as in section B. The relative difference of SRDT calculation is given in Fig.4a, where the 2D correction is
taken as a reference. As shown in the figure, the relative difference of SRDT increases as the measurement angle
deviates from 90 degrees. The relative difference is larger for a larger flow Mach number. In the upstream direction,
the maximum relative difference of 2D and 3D correction reaches 5%, which will have an impact on noise source
localization. Indeed, the phase of sound or SRDT is important for delay-and-sum beamforming [25]. In Fig.4b,
differences of amplitude correction between 2D and 3D method are shown. When the Mach number is larger than 0.2
and the observation point moves to the upstream or downstream locations, the maximum difference of amplitude
correction exceeds 1.0dB. Therefore, it is preferable to use a 3D correction method instead of a 2D method, especially
for far-field noise testing performed in large wind tunnels.
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

0.4 -8 0.4 1 1.2


4
-4 0 1

0.3 0.3 0.8


Mach number

Mach number

0.8
-6
0.6
0.6
0.4
0.2 0.2
0.4
0.2
0.2
-2 2
0.1 0.1 (b) 0
(a)
40 60 80 100 120 140 40 60 80 100 120 140
m
, deg m
, deg

Fig. 4 Contour plot of the difference between 3D and 2D method on SRDT (left) and amplitude (right)
correction, where 2D method is taken as a reference. a) relative difference of delay-time calculation (%). b)
difference of amplitude correction (dB).

V. Experiments

A. Experimental study of shear layer corrections using an inflow speaker


Experimental study of shear-layer correction was performed in the wind tunnel located at CARDC. This tunnel is
the pilot aeroacoustic tunnel (PAT) of FL-17. PAT is a return-flow wind tunnel with air temperature control equipment
(water cooling). The length of the test section is 1.5m, and the maximum flow speed is 100m/s (without model). PAT
is surrounded by an anechoic chamber, and the cut-off frequency of the chamber is 100Hz. The dimension of the
nozzle is 0.55m×0.4m. The nozzle is defined as the opening of the contraction section, where the air is exhausted into
the anechoic chamber. The background noise level is about 75.6dB(A) (80m/s), which makes PAT very suitable for
aeroacoutic tests. During the experiment, an inflow VIFA 1’’ speaker was used as the noise source. To avoid flow-
speaker interaction noise, the speaker was installed inside of a streamlined skull. A window is located on one side of
the skull which permits the emission of sound. The window was covered with a grid fabric to avoid flow-generated
cavity noise, and the size of the window was 50mm by 30mm. Two out-of-flow microphones (also called microphone
pair in the following) were used to measure SRDT. The position of microphone pair can be moved upstream or
downstream along a slide track. An out-of-flow microphone array was used to identify the location of noise sources.
39 GRAS microphones were installed on the microphone array. The array pattern was designed following Underbrink
[24]. The sampling rate was 102.4kHz and the duration of data acquisition was 30s.

8
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Fig. 5 Experimental setup of shear-layer correction using an inflow speaker as noise source. The experiment
was performed in PAT wind tunnel in CARDC. Time delay from the speaker to microphone was measured
using a microphone pair and the position of noise source was identified using a microphone array.

Nozzle

U0
0.5m
Jet center line
Source

0.07m
0.67m Shear layer

0.3m m Traverse direction


m
1 2
Microphone pair

Fig. 6 Schematic of the determination of refraction angle θ0 using a microphone pair (plan view). The red line
represents the actual ray path, the blue line represents the line-of-sight ray path and the green line represents
the emission ray path without flow, and the dashed line represents the wavefront.
As shown in Fig.6, SRDT is calculated using the following steps:
1) ensemble average the transient pressure signal to reduce the influence of background flow noise;
2) obtain the delay time from source to microphones dt0 and dt1 using cross correlation technique;
3) calculate μ1, using the formula given by Schlinker [8], . Where f is the frequency of
sound and l is the distance between two microphones m1 and m2;
4) calculate the refraction angle θ0, using the formula ;
5) after θ0 and are obtained, the emission angle θ’ is calculated using Eq. (7).
6) calculate the experiment determined SRDT:
1 R R 
Tdelay   t  d  (25)
c0  sin  ' sin  0 

9
7) calculate the theoretical SRDT based on the 2D Amiet’s method.
Delay time, ms

, deg
0
(a) U0=30m/s (d) U0=30m/s
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

2.4
2kHz,Exp.
4kHz,Exp.
2.3 8kHz,Exp.
Amiet2D theory
Delay time, ms

2.2
, deg
0

2.1

2
(b) U0=50m/s (e) U0=50m/s
1.9
60 80 100 120 140
m
, deg
Delay time, ms

, deg
0

(c) U0=70m/s (f) U0=70m/s

Fig. 7 Experimental data and theoretical predictions of SRDT (left column) and of the corresponding
refraction angle (right column). The source, the microphone pair and the shear layer lie on the same plane.
And the theoretical correction value is calculated using Amiet’s 2D method.

Experimental and theoretical value of SRDT (left column) and of the corresponding refraction angle (right column)
are shown in Fig.7. The source, the microphone pair and the shear layer lie on the same plane. The theoretical
correction value is calculated using Amiet’s 2D method. The experimental SRDT agrees well with the theoretical
calculation when the flow speed is smaller than 30m/s. As flow speed increases, the experimental data becomes more
scattered. This could be explained by the fact that the signal-to-noise ratio (SNR) decreases as background flow noise

10
increases. According to Lighthill’s theory, the jet noise level follows the 5th to 6th order law of flow speed. The acquired
data are more noisy as the flow speed goes higher, which causes the degradation of the accuracy of SRDT calculation.
More accurate experimental results can be achieved using advanced background-noise cancellation technique instead
of time-domain averaging. From the experimental results, it can be concluded that variation of the frequency of sound
has no obvious effect on shear layer correction. If the source-receiver distance is large enough, said larger than 3 times
of the wave length (above 1kHz), the planar-wave assumption in Amiet’s theory holds.

In-flow speaker position


Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Y, m

Fig. 8 Identified source positions using beamforming. “+” stands for source positions obtained without
shear-layer correction, and other markers stand for source positions with various correction methods. The
color of marker represents flow speeds, and the size of maker represents the frequency of sound.
In Fig.8 the identified source positions are shown. The color of marker represents flow speeds, and the size of
maker represents the frequency of sound. Red, green and blue marker color corresponds to 30m/s, 50m/s and 70m/s
respectively. And small, medium and large marker size corresponds to 2kHz, 4kHz and 8kHz respectively. Source
positions are obtained by recording the coordinate of the maximum-value point in the noise map. The marker “+”
stands for identified source positions without shear layer correction. As flow speed increases, the identified source
position shifts further to the downstream (from left to right). Obviously, this is caused by the convection of flow. All
shear-layer correction methods as shown are capable of correcting source-location shift in the stream-wise direction.
However, none of them can correct the cross-stream shift of source positions. The physics behind the cross-stream
shift of source location is beyond the scope of this study. Among these methods, M-H, AM and SFM are not
mathematically strict, but they are fast and have enough accuracy for SRTD calculation for small wind tunnels under
low speeds. Taking M-H method as an example, it only considers the convection of flow and ignores the refraction
effect. This indicates that the refraction effect is weak for small tunnels under low flow speeds (see Fig. 8). It should
be noted that amplitude correction is not available for M-H, AM and SFM method.
In Table 2, noise source maps of the inflow speaker are given. The green “x” marker denotes the physical source
position. In the 3rd column, the noise map without shear-layer correction is given. The results once again proved that
convection effects mainly cause the position-shift of noise sources. All the above shear-layer correction methods can
compensate for the position shift.

11
Table 2. Noise-source maps obtained using various shear-layer correction methods, U0=70m/s
SLC
2kHz 4kHz 8kHz
Mtd.

w/o C
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

AM

3D/C

SFM

RT

12
B. Noise source localization using a wall-mounted speaker and tensioned fabric wall
An experimental study of noise source localization was performed in UAT wind tunnel in University of New South
Wales (UNSW Australia). The UAT is an open-jet wind tunnel where the test section is surrounded by a
3m×3.2m×2.15m anechoic chamber. The dimension of the nozzle is 0.455m×0.455m. More information about UAT
can be found in Ref. [26]. The UAT was upgraded in 2018, as a result, much higher flow speed and lower turbulence
level can be achieved.

Traverse Microphone array

Tensioned fabric wall Nozzle


Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Wall-mounted speaker

Sound waves

Fig. 9 Noise source localization using a wall-mounted speaker and a tensioned fabric wall. The experiment
was performed in the UAT wind tunnel in UNSW.

A picture of the experimental setup is given in Fig.9. An ordinary fabric that connects both the nozzle and the
traverse was used to mimic the Kevlar wall used in closed-test-section wind tunnel [27]. The dimension of the fabric
is 0.6mm×460mm×1000mm. 8 wires on the upper and the lower edge of the fabric were connected to the ceiling beam
and floor grids of the anechoic chamber, and wires were tensioned to provide extra support for the fabric. To avoid
damaging the fabric, all edges of the fabric were sewn together using wires. The fabric was stretched by moving the
traverse downstream incrementally. The vibration of tensioned fabric was checked by gradually increasing the flow
speed. It was found that when the flow speed was lower than 20m/s, the vibration of the fabric can be ignored. A
serrated plate was attached to the trailing edge of the fabric to mitigate flow separation noise. A 5’’ wall-mounted
cone speaker was served as noise source. A diameter 30mm hole on the endplate at the position of speaker permits the
propagation of sound. A 1m×1m GRAS microphone array was used to identify the location of noise sources. There’s
a 5kHz electrical tone generated by the driven motor of wind tunnel, which will interfere with the speaker sound.
Therefore, the volume of speaker was tuned to achieve high SNR. The distance from speaker to fabric is 0.46m and
the distance from fabric to array surface is 0.5m. NI-PIXe 4499 DAQ cards and Labview® software were used for
data acquisition. During the test, the sampling rate was 65536Hz, and the duration of acquisition was 32s.
Noise maps of wall-mounted speaker without or with fabric wall are shown in Fig.10. The green marker “x”
denotes the position of speaker source and the rectangle denotes the position of endplate. The flow speed was 20m/s,
and the frequency of sound was 4kHz. The results show that without shear layer correction, the identified source
position was not aligned with the speaker position, no matter a fabric wall is installed or not. And the amount of
position drift is about the same with or without fabric wall, which can be explained by the fact that the free-jet shear
layer and the fabric wall can be treated as a boundary separating the ambient stagnant flow and core flow. Although
the growth rate of the thickness of a free-jet shear layer and a boundary layer on the fabric wall is different, it has been
proved that the thickness has no obvious effect on shear-layer correction [13]. After shear-layer correction is applied
(SFM), the identified source position agrees well with speaker position. However, it should be noted that the sound
pressure level is greatly reduced (about 12dB at 4kHz) by the fabric wall. Therefore, to accurately measure the pressure
amplitude of sound, the attenuation of sound caused by the fabric needs to be considered [27].

13
(a)w/o fabric, no correction (b)w fabric, no correction
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

(c)w/o fabric, SFM Corr. (d)w fabric, SFM Corr.

Fig. 10 Noise maps of a wall-mounted speaker without or with fabric wall. The rectangle denotes the
position of endplate, and green marker “x” denotes the position of the speaker source. u0=20m/s, f=4kHz.

-5
Insertion Loss, dB
PSD, dB/Hz

-10

-15

-20
Present fabric (b)
(a) Kevlar 120, Devenport et al. 2013
-25
3 4
10 10
f, Hz

Fig. 11 The attenuation of sound caused by the tensioned fabric. a) The received PSD of white noise
with/without fabric. b) The static insertion loss of present tensioned fabric over 1/3 octave frequency band.

The attenuation of sound caused by the tensioned fabric is given in Fig.11. Power spectrum density of the received
sound from the central microphone on the array surface without(w/o) and with fabric wall are shown in Fig.11a. And
the speaker was driven with white noise signal. The result shows that the fabric attenuates sound across the whole
spectrum band. Define the insertion loss (IL) as:
P2
ILdB  10 lg A2
PB

14
Where, PA and PB represent the integrated sound pressure without and with fabric wall respectively. The insertion loss
over the 1/3 octave frequency band is given in Fig.11b. The blue curve represents the experiment data with present
fabric and the black curve represents the fitted curve given by Devenport et al. (2013) [27] using a Kevlar 120 fabric.
The present fabric is much thicker than Kevlar 120 (0.6mm vs. 0.08mm). The result shows that the insertion loss of
present fabric is much larger than that of Kelvar 120. Therefore, to improve the SNR of aeroacoustic test, a thinner
and stronger fabric is preferred to be used for the wall of closed test section.

(a)w/o fabric (b)w fabric


PSD, dB/Hz

PSD, dB/Hz
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

Fig. 12 Spectrum broadening of tonal noise without/with fabric wall. The wall-mounted speaker is
located 0.55m downstream of the nozzle, and the frequency of sound is 12.5kHz.

The spectrum broadening of tonal noise is shown in Fig.12. Spectrum broadening reduces the amplitude of tonal
sound and shifts its energy to the adjacent frequency component. According to Vincent and Gabard [28], the spectrum
broadening stems from the combination of the spatial scattering of sound due to the refraction of waves propagating
through shear-layer vortexes, and two Doppler shifts induced by the motion of the vortex relative to the source and of
the observer relative to the vortex. Spectrum broadening is harmful for out-of-flow test of rotor noise, e.g., propeller
noise, as the amplitude and frequency of the tonal part of rotor noise changes during the propagation from inside to
the outside of open-jet flow. As can be seen from Fig.13, the spectrum broadening is mitigated when a tensioned fabric
is used, which could be explained by the fact that the vortex generation and growth from the trailing edge of wind-
tunnel nozzle is suppressed by the fabric. Therefore, it’s preferred to perform rotor noise test in a Kevlar-walled wind
tunnel, as less spectrum broadening and higher flow speeds can be achieved.

VI. Conclusion
A shear layer exists between the core flow and ambient stagnant flow for open-jet wind tunnels, which causes
phase and amplitude distortion of sound during the propagation of sound from the inside to the outside of the jet.
Therefore, the phase and amplitude of sound needs to be corrected for accurate localization and quantification of noise
sources using microphone array. In this study, a geometric acoustics (GA) based 3D shear-layer correction method is
proposed, which can be used for rectangular-nozzle wind tunnels. And the relative difference between present method
and various existing shear-layer correction methods is analyzed. Numerical calculations are performed. It’s found that
when the Mach number is smaller than 0.3 and the measurement angle ranges from 40 to 140 degrees, the difference
of phase correction is trivial for various shear-layer correction methods in the two-dimensional case. However, in the
three-dimensional case, the maximum relative difference of phase correction exceeds 5% and that of amplitude
correction reaches 1dB compared to the two-dimensional case. The amplitude correction over the upstream-half array
surface will exceed 2dB for large wind tunnels when the Mach number is larger than 0.2.
Experimental studies were conducted in CARDC and UNSW respectively. An inflow speaker and an out-of-flow
microphone pair were used to measure the phase of sound (SRDT). And a 0.72m diameter GRAS array were used to
identify the location of noise source. The measured SRDT values agree well with theoretical prediction values, and
the location of the speaker source is recovered after shear-corrections been applied. Errors between the identified and
the physical source location are studied using present 3D method and various existing methods. It was found that all
shear-layer correction methods can correct the stream-wise drift of source location. However, none of the them can
correct the cross-stream drift of source location, which might be induced by the local cross flow. A wall-mounted
speaker and an out-of-flow array were used to assess the effect of a tensioned fabric wall on the localization of sound

15
source. It was found that the amplitude of sound is greatly attenuated, while the spectrum broadening was suppressed
compared to the open-jet flow case. Based on the experimental findings, we recommend to perform rotor noise test in
a Kevlar-walled wind tunnel, since less spectrum broadening and higher flow speeds can be achieved.

Acknowledgments

The study was supported by National Natural Science Foundation of China (NO.11504417) and China Scholarship
Council. Thanks Thomas Geyer from BTU for helpful discussion about shear-layer correction methods. Thanks,
Omear Saeed, Manuj Awasthi, Jiang Chaoyang and Arthur Tan for helpful discussion during the preparation of the
experiment in UNSW.

References
[1] Wang, Y. G., Yang, J. S., and Jia, Q., et al., “An improved correction method for sound source drift in a jet flow and its
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

application to a wind tunnel measurement,” Acta Acustica united with Acustica, Vol. 101, No. 3, 2015, pp.642, 649.
[2] Sulaiman, Z., “Effect of open-jet shear layers on aeroacoustic wind tunnel measurements,” M.S. Thesis, Delft University of
Technology, Delft, Netherlands, 2011.
[3] Bahr, C. J., Zawodny, N. Z., and Liu, F., et al., “Shear-layer correction validation using a non-intrusive acoustic point
source,” in 16th AIAA/CEAS aeroacoustics conference, Stockholm, Sweden, AIAA paper 2010-3735.
[4] Miles, J. W., “On the reflection of sound at an interface of relative motion,”. J. Acoust. Soc. Am., Vol. 29, No. 2, 1957, pp.
226, 228.
[5] Ribner, H. S., “Refraction, transmission, and amplification of sound by a moving medium,” J. Acoust. Soc. Am., Vol. 29, No.
4, 1957, pp. 435, 441.
[6] Amiet, R. K., “Correction of open jet wind tunnel measurements for shear layer refraction,” the 2nd AIAA Aero-Acoustics
Conference, Hampton, VA., 1975, AIAA paper 75-0532.
[7] Amiet, R. K., “Refraction of sound by a shear layer,” Journal of Sound and Vibration, Vol. 58, No.4, 1978, pp.467, 482.
[8] Schlinker, R. H., Refraction and scattering of sound by a shear layer,” NASA CR-3371, 1980.
[9] Morfey, C. L., and Joseph, P. F., “Shear layer refraction corrections for off-axis sources in a jet flow,” Journal of Sound and
Vibration, Vol. 239, No. 4, 2001, pp.819, 848.
[10] Koop, L., and Ehrenfried, K., “Investigation of the systematic phase mismatch in microphone-array analysis,” in 11th
AIAA/CEAS Aeroacoustics Conference, Monterey, California, AIAA paper 2005-2962.
[11] Porteous, R., Geyer, T., and Moreau, D., Doolan, Con., “A correction method for acoustic source localisation in convex
shear layer geometries,” Applied Acoustics, Vol. 130, 2018, pp. 128, 132.
[12] Zhang, J., Chen, P., and Zhang, J. L., et al., “A shear layer phase shift correction method based on simplified ray model,”
Journal of Aerospace Power, Vol. 33, No. 10, 2018, pp.2458, 2464.
[13] Sarradj, E., “A fast ray casting method for sound refraction at shear layers,” in 22nd AIAA/CEAS Aeroacoustics Conference,
Lyon, France, AIAA paper 2016-2762.
[14] Oerlemans, S., “Detection of aeroacoustic sound sources on aircraft and wind turbines,” Ph.D Dissertation, University of
Twente, 2009.
[15] Sijtsma, P., “Experimental techniques for identification and characterisation of noise sources,” NLR-TP-2004-165.
[16] Padois, T., Prax, C., and Valeau, V., “Numerical validation of shear flow corrections for beamforming acoustic source
localisation in open wind-tunnels,” Applied Acoustics, Vol. 74, 2013, pp. 591, 601.
[17] Hansen, C., Doolan, C., and Hansen, K., “Wind Farm Noise: Measurements, Assessment, and Control,”1st ed., Wiley Series
in Acoustics, Noise and Vibration, 2017.
[18] Delfs, J. W., “Gliederung der vorlesung ’methoden der aeroakustik,”, class notes, 2015.
[19] Ewert, R., Kornow, O., Tester, B., Powles, C., Delfs, J., and Rose, M., “Spectral broadening of jet engine turbine tones”, In
14th AIAA/CEAS aeroacoustic conference, AIAA Paper 2008-2940

16
[20] Jiao, J., Delfs, J. W., and Dierke J., “Towards CAA based acoustic wind tunnel corrections for realistic shear layers,” In 21st
AIAA/CEAS aeroacoustic conference, Dallas, TX, AIAA paper 2015-3278.
[21] McDonald, B. E., and Andrew, A. P., “Nonlinear progressive wave equation for stratified atmospheres,” J. Acoust. Soc. Am.
Vol. 130, No. 5, 2011, pp.2648, 2653.
[22] Wang, X. N., Zhang, J., and Chen, P. et al., “An introduction of CARDC 5.5m by 4m anechoic wind tunnel and the
aeroacoustic tests,”. in 3rd International Symposium on Fluid-Structure Sound Interaction and Control, Tokyo, Japan, 2017.
[23] Jiao, J., “Aeroacoustic wind tunnel correction based on numerical simulation,” Ph.D Dissertation, TU Braunschweig, 2017.
[24] Muller, T. J., “Aeroacoustic Measurements,” 2nd ed., Springer press, 2002, Chap. 2.
[25] Fischer, J. and Doolan, Con., “Improving acoustic beamforming maps in a reverberant environment by modifying the cross-
correlation matrix”, Journal of Sound and Vibration, Vol. 411, No. 17, 2017, pp.129, 147.
[26] Awasthi, M., Rowland, J., and Doolan, C., “Two-step hybrid calibration of remote microphones,” J. Acoust. Soc. Am., Vol.
144, No. 5, 2018, pp. EL477, 483.
Downloaded by CARLETON UNIVERSITY on March 28, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2019-2717

[27] Devenport, W., Ricardo, B., and Aurelien, B. et al., “The Kevlar-walled anechoic wind tunnel,” Journal of Sound and
Vibration, Vol. 332, No. 17, 2013, pp.3971, 3991.
[28] Vincent, C., and Gabard, G., “Spectrum broadening of acoustic waves by convected vortices,” J. Fluid. Mech., Vol. 841,
2018, pp.50, 80.

17

You might also like