ArticlePDF Available

Abstract and Figures

To accurately characterise the noise measured in the vicinity of wind farms, outdoor microphones must be adequately protected from the wind. A standard 90 mm windshield is appropriate for measurements in light winds; however, as the wind speed increases, wind-induced pressure fluctuations contribute to the measured sound pressure level, leading to erroneous data. Three alternative secondary windshields have been developed and tested in an outdoor environment and evaluated for their ability to allow low frequency noise and infrasound measurements to be obtained in the presence of wind. Performance evaluation is facilitated through analysis of high resolution spectra as well as analysis of the coherence between microphones with different windshields under various meteorological conditions. This enables a distinction to be made between noise originating from sources such as a wind farm and wind-induced noise. The effect of the microphone location with respect to the ground surface has also been investigated for frequencies up to 100 Hz.
Content may be subject to copyright.
Identication of low frequency wind turbine noise using secondary
windshields of various geometries
1)
Kristy Hansen
a)
, Branko Zajamšek
b)
and Colin Hansen
c)
(Received: 15 November 2013; Revised: 25 March 2014; Accepted: 25 March 2014)
To accurately characterise the noise measured in the vicinity of wind farms, out-
door microphones must be adequately protected from the wind. A standard
90 mm windshield is appropriate for measurements in light winds; however, as
the wind speed increases, wind-induced pressure uctuations contribute to the
measured sound pressure level, leading to erroneous data. Three alternative sec-
ondary windshields have been developed and tested in an outdoor environment
and evaluated for their ability to allow low frequency noise and infrasound mea-
surements to be obtained in the presence of wind. Performance evaluation is fa-
cilitated through analysis of high resolution spectra as well as analysis of the
coherence between microphones with different windshields under various mete-
orological conditions. This enables a distinction to be made between noise orig-
inating from sources such as a wind farm and wind-induced noise. The effect of
the microphone location with respect to the ground surface has also been investi-
gated for frequencies up to 100 Hz. © 2014 Institute of Noise Control Engineering.
Primary subject classication: 14.5.4; Secondary subject classication: 71.1.1
1 INTRODUCTION
Characterisation of wind turbine noise poses many
challenges due to the complicated nature of the sound
source, the large distances from source to receiver and
the demand for measurements to be carried out in
windy conditions. It is important to understand each
of these variables when analysing measurements taken
in the vicinity of a wind farm. In addition, to ensure
measurement accuracy, it is necessary to develop wind-
shield designs that can be used for a range of wind
speeds associated with wind farm operation.
For modern wind turbine designs, aeroacoustic noise
sources are generally dominant
1
. However, mechanical
noise can become signicant if there is a fault within
the gearbox or generator or if the blades or tower are ex-
cited into resonant vibration and radiate noise. In this
case, the character of the noise is often tonal. There
are three main mechanisms responsible for aeroacoustic
noise generation and the resulting sound spectra can be
divided into overlapping frequency ranges
2
. Infrasound
is produced when the airow is slowed down by the
presence of the wind turbine tower, rapidly changing
the angle of attack of the air on the blade
3
. It is charac-
terised by tonal components at the blade-pass frequency
and har monics. Broadband low frequency noise gener-
ation has been attributed to aerodynamic loading uc-
tuations which are caused by interaction between
inow turbulence and the rotating turbine blades
4
. The
level of inow turbulence varies with atmospheric con-
ditions and also with the locations of turbines relative to
the wake of other turbines. Higher frequency noise is
produced by the interaction between the turbulent
boundary layer on the blade and the blade trailing
edge
5
. Both inow turbulence noise and trailing edge
noise are broadband in nature
2,4
.However,inow tur-
bulence is considered to be the most signicant aeroa-
coustic noise mechanism for frequencies below a few
hundred hertz
6
and trailing edge noise is the main
mechanism for frequencies from 400 to 1000 Hz
7
. Var-
iations in the noise contributions as a function of time
have been attributed to wind shear
2
, directivity
8
and
wind speed and direction variations. The wind shear,
wind speed variations and yaw error cause changes in
the blade loading and in the worst case, can lead to dy-
namic stall
9
. This is one explanation for the enhanced
1)
This is the rst paper published in NCEJ on the special
topic of Wind Turbine Noise.
a)
School of Mechanical Engineering, the University of
Adelaide, North Tce., Adelaide 5005, AUSTRALIA;
email: kristy.hansen@adelaide.edu.au.
b)
School of Mechanical Engineering, the University of
Adelaide, North Tce., Adelaide 5005, AUSTRALIA;
email: branko.zajamsek@adelaide.edu.au.
c)
School of Mechanical Engineering, the University of
Adelaide, North Tce., Adelaide 5005, AUSTRALIA;
email: colin.hansen@adelaide.edu.au.
69Noise Control Engr. J. 62 (2), March-April 2014
amplitude modulation or thumping noise heard at
residences
7
.
At a typical residential location near a wind farm, the
wind turbine noise spectrum is biased towards lower
frequencies due to propagation effects
10
. Attenuation
of sound in air and attenuation due to reection from
a grass-covered ground both decrease with frequency,
and low frequencies are poorly absorbed by other obsta-
cles as well. In addition, a house can behave as a low-
pass lter, since the walls of a residence selectively
block mid to high frequency noise. Hence, the determi-
nation of the levels of low frequency noise and infra-
sound at a residence is important when investigating
possible causes of annoyance.
It has been shown that the dominant source of turbu-
lent pressure uctuations experienced by an outdoor
microphone is the intrinsic turbulence in a ow rather
than the uctuating wake behind the windscreen
11
.
Since the wind velocity turbulence spectrum is heavily
weighted to low frequencies, wind-induced noise is
higher at low frequencies
12
. Consequently, it can be dif-
cult to distinguish between wind turbine noise and
wind-induced noise. This issue can be partially
addressed by using a windshield, which reduces the at-
mospheric turbulence incident on the microphone. The
effectiveness of the windshield can be further enhanced
by using a secondary windshield which is separated
from the primary windshield by a layer of air. The layer
of air provides a region for viscous dissipation to re-
duce the turbulence generated behind the rst wind-
shield layer
11
.
The reduction in turbulence provided by a multi-layer
windshield design was demonstrated for outdoor mea-
surements in a large eld by Ref. 13. These researchers
placed a hot-wire on the inside of the windshield and no-
ticed a signicantreductioninwindspeedandassociated
uctuations. Reference 14 found that the wind-induced
noise on the microphone with a doub le-layer windshield
was signicantly lo wer. The reduction was greater than
15 dB at infrasonic frequencies compared to a micro-
phone with a 70 mm wind shield. Substantial reductions
in wind-induced noise w ere also observed across the fre-
quency spectrum for a wind speed of 6 m/s.
Reference 15 species use of a layered windshield
design for turbine sound power measurements, where
the secondary layer is hemispherical and the micro-
phone is mounted to a circular board/plate. However,
the layered windshield concept can also be incorporated
into a spherical design. The advantage of spherically
shaped windshields is that at low frequencies, the pres-
sure uctuations at the front and rear of the sphere are
of opposite sign and thus cancellation occurs between
these contributions, reducing the wind-induced noise
at the centre of the sphere
12
. This cancellation would
not occur for downward-travelling sound waves in a
hemispherical windshield design and hence the sound
eld at the centre of the hemisphere would be different.
The layered windshield design can also be achieved by
using an underground box
16
, where the lid is replaced
with a layer of acoustic foam, which is level with the
ground to avoid the generation of turbulent eddies. This
design minimises exposure to wind-induced noise and
eliminates wake-induced turbulence.
The three layered-windshield designs discussed
above were compared by analysing the coherence be-
tween microphone measurements using the hemispher-
ical and box windshields, hemispherical and spherical
windshields and the box and spherical windshields for
various wind conditions. It is reasonable to assume that
the wind-induced noise at each microphone would be
uncorrelated with that at the other microphones due to
differences in the secondary windshield geometry and
differences in location, which would alter the turbulent
interaction. Hence, it was deduced that a high coher-
ence could indicate that the noise was associated with
the operation of the wind farm. The coherence analysis
was used to complement a high resolution narrowband
analysis to identify potentially audible low-frequency
noise not induced by the wind. A comparison between
noise spectra from the different microphones was made
to determine if the 6 dB correction specied in Ref. 15
could be applicable to the other windshield congura-
tions at low frequencies.
2 METHODOLOGY
2.1 Field Measurements
Outdoor measurements were carried out for 6 days at
a residence located approximately 1 km north of the
nearest turbine of a South Australian wind farm, which
is made up of 37 operational turbines, each with a rated
power of 3 MW. The microphones were located approx-
imately 5 m from the residence and as far as possible
(~10 m) from the nearest trees, which were around
5 m in height. Some small bushes and shrubs were lo-
cated within a few metres of the microphones. Time se-
ries data were acquired using a National Instruments
data acquisition device with a sampling rate of
10,240 Hz for a continuous sequence of 10-minute
samples. All microphones attached to this device were
G.R.A.S. type 40AZ with 26CG preampliers having
a noise oor of 16 dBA and a low frequency limit of
0.5 Hz. The noise oor of the system was 27 dBA as
measured in the anechoic chamber at the University of
Adelaide, which has a background noise level of less
than 5 dBA. Although the A-weighted noise oor was
relatively high, it was sufciently low for the purpose
of the cur rent measurements. This is shown in Sec. 3.3,
70 Noise Control Engr. J. 62 (2), March-April 2014
where the measured outdoor levels are greater than
10 dB above the noise oor for frequencies greater than
0.7 Hz. The local wind speed and direction were mea-
sured concurrently at 1.5 and 10 m using a Davis Vantage
Vue and a Da vis Vantage Pro weather station, respec-
tiv ely, capable of measuring to an accuracy of 0.4 m/s.
The 10 m mast was a low cost TV antenna mast slightly
modied to suit the w eather station and to facilitate a fast
set-up. The weather stations w ere located around 20 m
from the residence in an open eld and are pictured in
Fig. 1(a). The wind data were av eraged o v er 10-minute
sample periods.
All three outdoor microphones were equipped with
90 mm-diameter windshields as well as secondary
windshields of various congurations. One microphone
was positioned at ground level, another was mounted at
a height of 1.5 m, and the third was located under-
ground inside a small plywood box.
The microphone at ground level pictured in Fig. 1(b)
was taped horizontally at the centre of a 1 m diameter
aluminium plate of 3 mm thickness and covered by
both a primary and secondary windshield as speci ed
in Ref. 15. The secondary windshield consisted of a
16 mm layer of acoustic foam, covered by a layer of
SoundMaster acoustic fur and supported by a thin-wire
steel frame of outer diameter 450 mm. The windshield
was riveted to the aluminium plate and secured with a pin.
The microphone pictured in Fig. 1(d) was mounted
at a height of 1.5 m using a star-dropper to minimise
wind noise interference associated with the more con-
ventional method of tripod mounting. Strakes were in-
corporated into the mount design to minimise vortex
shedding and associated noise and these were designed
according to Ref. 17. The microphone was tted with a
secondary spherical windshield which was attached to a
thin-wire steel frame of outer diameter 450 mm as
shown in Fig. 1(d). The windshield materials were iden-
tical to those used for the hemispherical secondary
windshield described above.
The underground microphone pictured in Fig. 1(c)
was located in a 120 mm 120 mm 280 mm ply-
wood box with an acoustic foam lid, 50 mm thick.
The acoustic foam had a pore size of 20 pores per inch.
The top of the lid was ush with the surrounding
ground to minimise the for mation of eddies that would
generate extraneous noise. The microphone was
mounted horizontally on a custom-made shelf, which
incorporated a hemispherical groove covered with a
3 mm layer of rubber. This method of locating a micro-
phone in an underground box to minimise wind noise
was conceived by Ref. 16.
2.2 Wind Shield Insertion Losses
The anechoic chamber at the University of Adelaide,
which has dimensions 4.79 m 3.90 m 3.94 m, was
used to measure the inser tion loss of various micro-
phone windshield congurations. The set-up for these
measurements is shown in Fig. 2. The porous alumin-
ium oor plates of the anechoic chamber were covered
with 19 mm thick medium-density breboard to simu-
late outdoor conditions at the eld measurement loca-
tion. Two JBL 1000 W subwoofer loudspeakers were
located in one corner of the anechoic chamber and a
G.R.A.S. type 40AZ microphone with 26CG preampli-
er was located in the opposite corner at a distance of
4 m from the speakers. This was connected to a NI
9234 data acquisition device which transferred data to
a computer. The loudspeakers emitted white noise
which was generated using a B&K noise generator type
1405 and low-pass ltered at 200 Hz. The signal was
amplied using a Peavey GPS 3400 1800 W amplier
a
c
d
b
Fig. 1(a) Davis weather stations at 1.5 and 10 m, (b) hemispherical windshield, (c) box
windshield and (d) spherical windshield.
71Noise Control Engr. J. 62 (2), March-April 2014
and the input voltage to the loudspeakers was maxi-
mised to increase the sound output at low frequencies.
The overall level of the noise emitted from the loud-
speakers was 100 dB. Signals were analysed with fre-
quency resolution of 0.1 Hz.
The results are shown in Fig. 3 where the green
curve represents the noise level measured using a mi-
crophone with a primar y windshield only, which was
placed on the oor. This measurement was repeated
10 times to determine the associated error. The standard
deviation in noise level for this conguration was calcu-
lated as a function of frequency and is shown by the
grey shaded region which has been overlaid on the
green curve. The blue curve shows results for the mi-
crophone in exactly the same position but with a sec-
ondary windshield in addition to the primary one. The
microphone with the spherical windshield was located
at 700 mm above the oor, since the top section of
the existing mount has this height. The box was located
on the oor as it was believed that there would be neg-
ligible difference in the results for low frequencies if it
were mounted under the oor. In all cases, the red curve
represents the results with the windshield off, where the
microphone position is identical to that with the wind-
shield on. The black curve is the noise level measured
in the anechoic chamber with the speakers and ampli-
er switched off.
Results indicate that at frequencies above 4 Hz there is
good agreement between the measured sound pressure
lev els with a primary windshield only compared to a pri-
mary and secondary windshield. There are some minor
differencesintherangeof20 to40Hzwhicharemost
signicant for the comparison betw een results for the mi-
crophone on the ground and the microphone with the
windshield off. It is possible that sound wa v es at these
frequencies interact differentl y with the different mount-
ing surfaces and this is currently under investigation. At
frequencies below 4 Hz, the performance of the different
secondary windshields varies slightly. The variation in
performance could be attributed to random air mo v ement
in the anechoic chamber which is induced by vibration of
the speaker cone and which is clearl y noticeable.
3 RESULTS
3.1 Overall Noise Levels
The data collected at night time are the main focus of
this paper as there was some noise from farming
machinery during the day. Nevertheless, daytime data
are plotted for comparative purposes. The overall
unweighted noise levels over the frequency range of
0200 Hz are plotted in Fig. 4 as a function of time
for the 6-day analysis period. These results were cal-
culated by applying a 6th order low-pass Butterworth
lter (with a cut off frequency of 200 Hz) to the raw
time data and then nding the root mean square for
each 10-minute measurement. There was negligible
difference between the results for this low frequency
range and results calculated over the frequency range
of 01000 Hz, indicating that the measured linear
spectra are dominated by low frequency noise.
Thewindspeedanddirectionat1.5 and10marein-
cluded in the gure to show how these variables affect
the overall noise levels. Note that the measurement loca-
tion w as 1 km from the nearest turbine in a direction be-
tw een N and NNW from the wind farm. It can be seen
that the ov e rall unweighted noise le vels increased with in-
creased wind speed but that the wind direction did not sig-
nicantly affect the results. It should be noted, however,
that downwind conditions only occurred during the
Fig. 2Schematic diagram of the set-up for insertion loss measurements.
72 Noise Control Engr. J. 62 (2), March-April 2014
daytime and that these daytime results should be viewed
with caution due to possible farming machinery noise.
The wind farm capacity factor is also plotted, and in
general this increases with wind speed as expected.
However, there are some exceptions to this trend and
these could be attributed to the fact that the wind tur-
bine closest to the residence may not have been gener-
ating the same amount of power as those much further
away. In addition, the wind characteristics close to the
residence may have varied signicantly from those at
hub height for these instances. Comparison between
the capacity factor and measured noise levels indicates
that peaks in noise level are generally consistent with
increased wind farm power output.
Comparisons between the overall unweighted sound
pressure level measured using different windshields
reveal that the general trends are the same. However,
the box windshield microphone consistently measured
a lower sound pressure level and the hemispherical and
spherical windshield microphones gave the most similar
results. Differences between the results are attributed to
interaction between atmospheric turbulence and the
windshields which is inuenced by windshield geome-
try and mounting location. The windshield geometry
also affects the wake characteristics, and thus the
wind-induced noise associated with the wake is different
for the hemispherical and spherical windshields and
non-existent for the box windshield. As mentioned
in Sec. 1, this component of wind-induced noise is
expected to be less important than that associated with
atmospheric turbulence. Further insight into the differ-
ences in noise level measured using different windshield
Fig. 3Windshield insertion loss for the different windshields used for measurements. Narrow
band analysis with 0.1 Hz resolution.
73Noise Control Engr. J. 62 (2), March-April 2014
congurations can be gained by inspection of the fre-
quency spectrum results, as discussed in Sec. 3.2.
To predict when wind turbine noise may have been
most noticeable, an atmospheric stability plot was con-
structed. The value of m is determined from Eqn. (1),
which was proposed by Ref. 18 and used in Ref. 19.
m ¼
log v
h
=v
ref
ðÞ
log h=h
ref
ðÞ
ð1Þ
This equation shows a relationship between the stabil-
ity factor, m, the wind speed v
h
at height h and the refer-
ence wind speed v
ref
at a reference height h
ref
which is
governed by the atmospheric stability. Stability classes
corresponding to various ranges for m are outlined in de-
tail in Ref. 20 where it was also shown that the contrast
between ambient noise and wind turbine noise was
greatest under stable conditions. Figure 5 shows the var-
iation of stability with time over the 6-day measurement
11pm 7am 11pm 7am 11pm 7am 11pm 7am 11pm 7am 11pm 7am
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Clock time
m factor
m
factor (
h
ref
= 1.5 m,
h
= 10 m)
m
factor (
h
ref
= 1.5 m,
h
= 174 m)
stable
neutral
unstable
18-19 Apr16-17 Apr
17-18 Apr 19-20 Apr 20-21 Apr 21-22 Apr
Fig. 5Stability factor m plotted over the measurement period where stability classes are
indicated with dashed lines.
11pm7am 11pm7am 11pm7am 11pm7am 11pm7am 11pm7am
0
0.25
0.5
0.75
1
50
60
70
80
90
Capacity factor
L
eq
(dB re 20 Pa) 0.8-200Hz
11pm7am 11pm7am 11pm7am 11pm7am 11pm7am 11pm7am
0
2
4
6
8
no
W
S
E
N
Wind Velocity (m/s) Wind Direction
L
eq
Hemi (dB)
L
eq
Box (dB)
L
eq
Sph (dB)
Capacity factor (%)
V
wind@1.5m
(m/s)
V
dir@1.5m
V
wind@10m
(m/s)
V
dir@10m
16-17 Apr 18-19 Apr 19-20 Apr 20-21 Apr 21-22 Apr
17-18 Apr
Clock Time
Fig. 4Unweighted overall equivalent sound pressure level (L
eq
) 0.8 to 200 Hz for three
microphones with a hemispherical, box and spherical windshield, respectively. The wind
speed and direction at 1.5 and 10 m are also shown in green and black, respectively.
74 Noise Control Engr. J. 62 (2), March-April 2014
period. The m factor was calculated using the weather
station data from heights of 1.5 and 10 m close to the mi-
crophone locations as well as SODAR data for the hub
height, which is 174 m higher than the residence. The
minimum velocity allowed for the valid use of Eqn. (1)
was 0.1 m/s to ensure that a reasonable value for the m
factor could be calculated.
The results for the two different calculations of the
m-factor are reasonably consistent, which shows that
while 10 m wind speed measurements are not suf-
ciently accurate for noise emission estimates
20
it is still
possible to gain an insight into atmospheric stability
characteristics from these data. It can be seen that con-
ditions were the most stable on the sixth night and
hence this was selected as a time period for more de-
tailed analysis. For comparison, the third night was cho-
sen, as the conditions were least stable on this night.
A comparison between the overall unweighted, A-
weighted and G-weighted sound pressure levels for
the 6-day measurement period is shown in Fig. 6 for
the microphone in the hemispherical windshield. The
A-weighting is outlined in many regulations and stan-
dards for specifying allowable levels produced by wind
farms (see for example Ref. 21). The background level
as measured in the anechoic chamber has been sub-
tracted from the measured A-weighted level and the re-
sultant value was not allowed to be lower than the
instrumentation noise oor of 27 dBA. For some of
the night time measurements, this noise oor proved
to be too high and actual levels would be less than
27 dBA. It is evident that the overall A-weighted noise
level exceeds 50 dBA at some instances between the
night time hours of 11 pm7 am. The range of mea-
sured levels is consistent with other data for the same
location
22
. Despite the occur rence of such high noise
levels in a quiet rural area at night time, the wind farm
is deemed compliant at this location. This is shown by
the regression curve determined by Ref. 22 which is
below the guideline limit specied in Ref. 21. The
G-weighting is an rms level and is intended to reect
human perception of infrasonic noise
23
. For the mea-
surement period, the maximum G-weighted level was
approximately 70 dBG, which is not considered to be
signicant for human perception
23
. However, the per-
ception thresholds specied in Ref. 23 are based on
steady pure tones and there is evidence to suggest that
11pm7am 11pm7am 11pm7am 11pm7am 11pm7am 11pm7am
0
0.25
0.5
0.75
1
40
50
60
70
80
90
Capacity factor Overall SPL (dB re 20
µ
Pa)
11pm7am 11pm7am 11pm7am 11pm7am 11pm7am 11pm7am
0
2
4
6
8
no
W
S
E
N
Wind Velocity (m/s) Wind Direction
L
eq
(dB)
LA
eq
(dB)
LG
eq
(dB)
Capacity factor (%)
V
wind@1.5m
(m/s)
V
dir@1.5m
V
wind@10m
(m/s)
V
dir@10m
16-17 Apr 17-18 Apr 18-19 Apr 19-20 Apr 20-21 Apr 21-22 Apr
Clock Time
Fig. 6Comparison between the overall unweighted, A-weighted and G-weighted sound pressure
levels for the hemispherical windshield. The wind speed and direction at 1.5 and 10 m are
also shown in green and black, respectively.
75Noise Control Engr. J. 62 (2), March-April 2014
the rms perception levels for strongly amplitude modu-
lated noise may be much lower.
The variation of the A-weighted and G-weighted
levels in response to changes in wind speed reects a
direct relationship between these variables. There are
some peaks in the A-weighted level that do not coincide
with an increase in wind speed, but these mainly occur
during the day and may thus be attributed to a noise
source other than the wind farm. At coincident times,
peaks are also observed in the G-weighted levels, sug-
gesting that this other noise source also produces infra-
sound above ambient levels. A low capacity factor is
associated with several of these peaks which supports
the idea that they may not be related to the wind farm.
3.2 Third-Octave Spectra
To establish the contribution of various frequencies
to the overall unweighted noise level, third-octave spec-
tra were plotted for the three microphones as shown in
Fig. 7. Two times were selected for each of the least sta-
ble night [Figs. 7(a) and (b)] and the most stable night
[Figs. 7(c) and (d)], respectively. In the former case,
the times were selected randomly and in the latter case,
they were chosen to show when the measurements
appeared to be most affected by the wind farm noise.
The reference threshold for human hearing of steady
tonal noise according to Ref. 24 is also shown in the
gures, even though this does not represent the thresh-
old for amplitude modulated noise. A summary of the
wind conditions and wind farm capacity factor is pro-
vided in Table 1. It can be seen that the conditions on
the two nights of measurement varied signicantly.
Inspection of Figs. 7(a) through (d) reveals peaks in
the 25, 50 and 80 Hz third-octave bands, which exist in
Table 1Summary of wind conditions and wind
farm capacity factor for selected mea-
surement times.
Date/time Wind speed
at 1.5 m (m/s)
Wind
direction
at 1.5 m
Capacity
factor (%)
Stability
factor
18/4, 23:04 2.7 ESE 83 0.26
19/4, 1:04 3.8 ESE 89 0.20
21/4, 23:24 0.2 WSW 52 1.52
22/4, 0:34 0.7 WSW 85 0.65
1 10 100
30
40
50
60
70
80
90
18 April 23:04
1 10 100
30
40
50
60
70
80
90
19 April 1:04
1 10 100
30
35
40
45
50
55
60
21 April 23:24
Frequenc
y
(Hz)
1 10 100
30
35
40
45
50
55
60
22 April 0:34
SPL
(1/3)
(dB re 20
µPa)
Frequency (Hz)
Hemi
Box
Sph
noise floor
ISO 389-7
V
1.5 m
= 2.7 m/s
V
1.5 m
= 0.2 m/s
V
1.5 m
= 2.0 m/s
V
1.5 m
= 3.8 m/s
a
c
d
b
Fig. 7One-third octave spectral plots showing signicant variations between measurements with
identical eld set-ups on different nights. Results are presented for microphones with a
hemispherical, box and spherical windshield and the Ref. 24 curve is shown for
comparison.
76 Noise Control Engr. J. 62 (2), March-April 2014
each gure. The relative sound pressure levels are sim-
ilar but are consistently higher for the measurement
times where the wind farm had a high power output
[Figs. 7(a) and (b)]. This suggests that these noise
sources are associated with the wind farm and this is
supported by the fact that peaks would not be expected
in a third-octave spectrum for a quiet rural area. This is
further reinforced by results from measurements at a
different residence located approximately 3 km from
the nearest turbine in the same wind farm.
Third-octave spectra measured at this second resi-
dence for zero wind speed at 1.5 m height at the resi-
dence (but signicant wind at hub height) are shown
in Fig. 8(a) and indicate that in the 50 Hz third-octave
band, the sound pressure level increased by 25 dB when
the wind farm was operating compared to when it was
shut-down. Peaks in the 1.6 and 2.5 Hz third-octave
bands can be observed both when the wind farm was
operating and when it was shut down. In the former
case, the peaks in the infrasonic range are much higher.
It is probable that the peaks observed when the wind
farm was shut down can be attributed to either the Hal-
lett wind farm, which is located 60 km away or to blade
and/or tower resonances, excited by air ow past them.
According to the narrowband plot in Fig. 8(b), the
peaks are spaced at 0.8 Hz, which also corresponds to
a possible blade-pass frequency for the wind turbines
at Hallett. The Hallett wind turbines are Suzlon
2.1 MW, which have a rotational speed of 15.1
17.7 rpm
25
.
Returning to Fig. 7, it should be mentioned that dif-
ferent y-axis scales have been used for Figs. 7(a) and (b)
compared to Figs. 7(c) and (d). This is due to the rela-
tively higher levels of sound pressure level in the
20 Hz third-octave band and below in Figs. 7(a) and
(b). The higher sound pressure level in these infrasonic
third-octave bands appears to be directly related to a
higher wind speed at 1.5 m for these measurements.
Wind-induced noise associated with atmospheric turbu-
lence is expected to be dominant at frequencies below
20 Hz according to measurements of the spectral turbu-
lent energy distribution in the atmospheric boundar y
layer
26
. This indicates that the effectiveness of the sec-
ondary windshields is limited at these frequencies. It
can be seen that the box windshield conguration has
the lowest infrasonic levels in Figs. 7(a) and (b), which
implies that this windshield performs best in windy
conditions. This is consistent with the fact that the
wind-induced turbulence is expected to be lower sub-
surface. The spherical windshield measures a lower
infrasound level compared to the hemispherical wind-
shield on the windiest night, as shown in Fig. 7(b).
According to these results and the results shown in
Fig. 4, it can be seen that the spherical windshield is
at least as effective as the hemispherical design. A
larger data set would need to be analysed to determine
if the spherical windshield consistently measured lower
levels of infrasound in windy conditions.
Referring back to the peaks which were identied in
Fig. 7, it is evident that for the 25 Hz third-octave band,
the results obtained using different windshield cong-
urations vary by up to 6 dB. It is possible that this tone
originates from several sources and hence reinforce-
ment of the signals occurs at some positions and cancel-
lation at others. On the other hand, for the peaks in the
1.6 and 4 Hz third-octave bands, results differ by less
than 1 dB. In addition, levels vary by less than 2 dB
for the peak in the 50 Hz third-octave band for the
results shown in Figs. 7(a) and (c). The increased levels
for this third-octave band in Figs. 7(b) and (d) for the
1 10 100
20
25
30
35
40
45
50
55
60
SPL
(1/3)
(dB re 20
µ
Pa)
ON
OFF
ISO 389-7
1 2 4 6 810 20
0
10
20
30
40
50
60
70
SPL (dB re 20
µ
Pa)
ON
OFF
a
b
Frequency (Hz) Frequency (Hz)
Fig. 8(a) Third-octave spectra and (b) narrowband spectra (0.1 Hz resolution) for a second
residence located approximately 3 km from the nearest turbine in the same wind farm. The
wind farm was operating at 68% of maximum capacity for the ON case.
77Noise Control Engr. J. 62 (2), March-April 2014
spherical windshield results are attributed to electrical
noise as discussed in Sec. 3.3. The results from mea-
surements using the hemispherical and box windshield
congurations which were not affected by electrical
noise show that the 50 Hz peaks would be audible to a
person with normal hearing according to Ref. 24, which
species the threshold of hearing for free-eld tonal
noise. The degree of perceptibility would be even
greater than shown in Fig. 7 for amplitude modulated
noise due to the high associated crest factors.
Further observation of Figs. 7(c) and (d) reveals that
the peak levels in the 25 and 50 Hz third-octave bands
are as high as 15 dB above the adjacent frequency bin
levels. This is greater than for the results with the higher
associated wind speed shown in Figs. 7(a) and (b). The
peaks in the infrasonic frequency range imply the pres-
ence of sound at the blade-pass frequency and this is dis-
cussed in more detail in Sec. 3.3.
3.3 Narrowband Analysis
Although analysis of the third-octave band levels
provides a good overview of the variation in sound
pressure level with frequency, more detailed informa-
tion is lost. Hence, a narrowband analysis was carried
out with a frequency resolution of 0.1 Hz. The results
are plotted in Figs. 9 and 10 for the worst case of ampli-
tude modulation (21/4 at 23:24) and for a low stability
case (18/4 at 23:04), respectively. The gures are sep-
arated into the infrasound region in (a) and the low-
frequency region in (b).
In Fig. 9(a), peaks in the infrasound region are visible
at 1.6 Hz up to around 9.6 Hz, and these are spaced at a
frequency of 0.8 Hz. For the Vesta V90-3 MW wind tur-
bines at the nearby wind farm, the nominal rotational
speed is 16.1 rpm
27
, which corresponds to a blade-pass
frequency of 0.8 Hz. Hence, these peaks appear to be
harmonics of the blade-pass frequency. Interestingly,
the blade-pass frequency itself is indistinguishable
which may be related to wind-induced noise being too
high in level at this frequency. The narrowband analysis
reveals a signicant amount of information in Fig. 9(b)
that was not visible on the third-octave plot. The peaks
that were identied in the third-octave plots are now vis-
ible as broadband humps. Also, secondary peaks that
have a spacing corresponding to the blade-pass fre-
quency are evident throughout the spectrum, especially
on the broadband humps. These spectral characteristics
imply the presence of tonal peaks which have been am-
plitude modulated at the blade-pass frequency. The ap-
pearance of the tonal peaks as broadband humps is
related to the presence of closely-spaced side-bands
which are associated with the amplitude modulation.
The narrowband plots are not compared to the
threshold of perception for free-eld tonal noise pro-
vided in Ref. 24 since this curve is more relevant for
third-octave levels. The peak noise levels are a function
of the frequency resolution and decrease as the resolu-
tion is increased. Hence comparison would lead to the
erroneous conclusion that the low frequency noise close
to 50 Hz is not audible. However, it was shown in
Sec. 3.2 that noise in the 50 Hz third-octave band is
1 2 4 6 8 10 20
0
20
40
60
20 30 50 80
0
10
20
30
40
50
Frequency (Hz)
SPL (dB re 20
µ
Pa)
hemi
box
sph
noise floor
a
b
Fig. 9Narrowband spectra (0.1 Hz resolution) for amplitude modulated case, 21st of April at
23:24. Results are presented for microphones with a hemi spherical, box and spherical
windshield.
78 Noise Control Engr. J. 62 (2), March-April 2014
in fact perceptible and the high degree of amplitude
modulation, which is indicated by the large amplitude
side-bands adjacent to the main tonal peaks, would
increase the degree of perceptibility. Reference 28
suggests that a 10 dBA penalty should be applied
where both tonality and amplitude modulation are
identied.
Comparison between the results for the different
windshield congurations in Fig. 9 indicates that there
are slight variations in the peak sound pressure level
but the position of the peaks with respect to frequency
is consistent. The results for the microphones with the
hemispherical and spherical windshields show the best
agreement, with some minor discrepancies in the value
of the amplitude for some peaks and side-bands. The
level of broadband infrasound measured by the box
windshield microphone is slightly higher than for the
other windshield congurations for this measurement,
making the blade-pass har monic peaks less distinct.
On the other hand, the broadband level measured by
the microphone in the box windshield at 23.3 Hz is
lower compared to the other windshields and hence
the peak is 6 dB lower. At 43.4 Hz, the peak in sound
pressure measured using the box windshield is 5 dB
higher compared to the other windshield congura-
tions. These differences could be caused by reinforce-
ment of the signals at some positions and cancellation
at others as discussed in Sec. 3.2. In addition, the differ-
ent microphones may have experienced different local
air movements as a result of localised wind interaction
with ground surface roughness.
It is also evident that the sound pressure level mea-
sured with the hemispherical windshield is not consis-
tently 6 dB higher than the other measurements.
Hence, it is probable that positive reinforcement be-
tween the incident and reected waves at low frequen-
cies occurs not only at the ground, but also at a height
of 1.5 m. This indicates that the pressure-doubling ef-
fect would be relevant for both microphone receivers.
However, in the context of a human receiver of height
~ 1.52 m, the 6 dB correction specied in Ref. 15
for sound power measurements near a wind turbine is
not relevant for measurements of low frequency noise
at a residence.
It should be noted that all measured peaks exceed the
instrumentation noise oor by at least 10 dB. Between
16 and 20 Hz, the anechoic chamber at the University
of Adelaide performs poorly and hence these results
do not reect the true noise oor of the instrumentation.
Figure 10 is a plot of the sound pressure level vs. fre-
quency for a low stability case (18/4 at 23:04) involving
microphone measurements with the hemispherical, box
and spherical windshield congurations. Peaks at the
blade-pass frequency are still discernible in Fig. 10(a),
but are much smaller relative to adjacent levels. This
is attributed to increased levels of broadband infra-
sound associated with the higher wind speed. The rea-
son that the peaks are visible at higher levels than
those in Fig. 10(a) is that the wind farm was producing
more power at this time. It should be noted that the
peaks can be seen most clearly for the box windshield
results, which highlights the effectiveness of this
1 2 4 6 8 10 20
0
20
40
60
80
20 30 50 80
0
10
20
30
40
50
SPL (dB re 20
µ
Pa)
hemi
box
sph
noise floor
a
b
Frequency (Hz)
Fig. 10Narrowband spectra (0.1 Hz resolution) for low stability case, 18th of April at 23:04.
Results are presented for microphones with a hemispherical, box and spherical
windshield.
79Noise Control Engr. J. 62 (2), March-April 2014
conguration for infrasound measurements at higher
wind speeds.
Similar to what was observed in Fig. 9(a), the peak at
23.3 Hz in Fig. 10(a) is 6 dB lower for measurements
using the box windshield in comparison with the other
windshield congurations. For all windshield cong-
urations, the level of this peak is lower for this case
and there no longer exist side-bands spaced at the
blade-pass frequency. It can be seen in Fig. 10(a) that
there are two main peaks at 44.8 and 49.6 Hz rather
than the continuous peaks spaced at the blade-pass fre-
quency visible in Fig. 9(a) for this frequency range.
This indicates that amplitude modulation is not signi-
cant for this case. This is most likely due to the lower
degree of atmospheric stability associated with this case
compared to the higher stability case shown in Fig. 9.In
addition, the wind direction measured at the residence
was cross-wind, whereas the wind direction for the
results shown in Fig. 9 had a downwind component.
The box windshield predicts the highest noise levels
for these peaks if the electrical noise at 50 Hz is ignored
for the spherical windshield. It is believed that this
50 Hz problem was caused by contact between the mi-
crophone and its aluminium mount and the peak was
observed in other measurements to varying degrees.
3.4 Coherence
Coherence is used to determine the existence or oth-
erwise of a relationship between data sets for the differ-
ent windshield congurations. High coherence between
microphones in different locations would be expected
for noise originating from a wind farm, especially if
the meteorological conditions did not vary signicantly
during each 10-minute measurement period. Low co-
herence would be expected for wind-induced noise
since the level at one location is not closely related to
that at another location and turbulence varies over the
measurement period.
To improve the accuracy of the coherence method, a
6th order Butterworth low-pass lter (with a cut off fre-
quency of 100 Hz) was used to maximise the signal-to-
noise ratio. Coherence plots with a frequency resolution
of 0.1 Hz are shown in Figs. 11 and 12 for the infra-
sonic and low-frequency ranges. These gures show
the coherence between microphone measurements us-
ing the hemispherical and box windshields, hemispher-
ical and spherical windshields and the box and
spherical windshields. In Fig. 11, the coherence is high
for the harmonics of the blade-pass frequency as well as
for the peaks around 23, 27, 45 and 75 Hz and the ad-
jacent secondary peaks spaced at the blade-pass fre-
quency of 0.8 Hz. The best coherence is achieved
between the hemispherical and spherical windshields
up to a frequency of 55 Hz.
In general, the signals are less coherent for the data
shown in Fig. 12, especially in the infrasonic range.
This indicates that wind-induced noise dominates the
signal at these frequencies, and an increase in sound
level at the blade-pass frequency cannot be distin-
guished. The signals are coherent for the low frequency
peaks that were identied in the narrowband analysis
for this case. A low coherence is evident between the
spherical windshield and the other congurations for
1 2 4 6 8 10 20
0
0.2
0.4
0.6
0.8
1
Coherence
20 30 50 80
0
0.2
0.4
0.6
0.8
1
Hemi & Box
Hemi & Sph
Box & Sph
a
b
Frequenc
y
, Hz
Fig. 11Coherence (0.1 Hz resolution) between microphone measurements using the
hemispherical, box and spherical windshields on 21st of April at 23:24.
80 Noise Control Engr. J. 62 (2), March-April 2014
the peak at 50 Hz, conrming that this is due to electri-
cal noise.
4 CONCLUSIONS
For the different windshield congurations investi-
gated in this study, there is good agreement between
measurements at low frequencies, particularly for stable
atmospheric conditions. Hence, it has been shown that
each of the secondary windshields can successfully
measure wind turbine noise in windy conditions.
Under stable atmospheric conditions, at a location
1 km north from the nearest turbine in the wind farm,
the harmonics of the blade-pass frequency are clearly
visible in the infrasonic spectral plots. In addition, tonal
peaks are evident in the spectra during night-time con-
ditions, which is unusual for a rural environment. Since
these tonal peaks are amplitude modulated at the blade-
pass frequency for some measurements, particularly
during stable conditions, it is clear that the noise source
originates from the wind farm. Comparison with the
threshold of hearing for free-eld tonal noise
24
indi-
cates that noise in the 50 Hz third-octave band would
be audible to a person with normal hearing. For
instances where noise in this third-octave band is ampli-
tude modulated, there would be an increased likelihood
of audibility.
For the measurement of infrasound in the presence
of wind, the box windshield conguration showed the
most promising results. Also, the most conservative
results for the peaks in sound pressure level above
40 Hz were obtained using this windshield. Investigation
of the inuence of microphone mounting conguration
on the results shows that below 100 Hz, there are no con-
sistent differences in the spectral results. Hence, it may
be deduced that at low frequencies, pressure doubling
could occur both at the ground and at 1.5 m, due to the
large wavelengths involved. Therefore a 6 dB correction
for a microphone located on a hard ground seems irrele-
vant for measurements of low frequency noise at a resi-
dence where the receivers are located in the height range
for which pressure doubling occurs.
5 ACKNOWLEDGEMENTS
Financial support from the Australian Research
Council, Project DP120102185, is gratefully acknowl-
edged. The authors also extend their thanks to the me-
chanical and electrical workshop staff at the
University of Adelaide. We are also grateful to the rural
residents in South Australia who participated in this
study.
6 REFERENCES
1. S. Wagner, R. Bareiss and G. Guidati, Wind Turbine Noise,
Springer Verlag, (1996).
2. G.P. van den Berg, The beat is getting stronger: the effect of
atmospheric stability on low frequency modulated sound of
wind turbines, Noise Notes, 4(4), 1540, (2005).
3. C.J. Doolan, D.J. Moreau and L.A. Brooks, Wind turbine
noise mechanisms and some concepts for its control, Acous-
tics Australia, 40(1), 713, (2012).
4. H.H. Hubbard and K.P. Shepherd, Aeroacoustics of Large
Wind Turbines, J. Acoust. Soc. Am., 89(6), 24952508,
(1991).
1 2 4 6 8 10 20
0
0.2
0.4
0.6
0.8
1
Coherence
20 30 50 80
0
0.2
0.4
0.6
0.8
1
Hemi & Box
Hemi & Sph
Box & Sph
Frequenc
y
, Hz
Fig. 12Coherence (0.1 Hz resolution) between microphone measurements using the
hemispherical, box and spherical windshields on 18th of April at 23:04.
81Noise Control Engr. J. 62 (2), March-April 2014
5. S. Oerlemans, P. Sijtsma and B. Mendez-Lopez, Location and
quantication of noise sources on a wind turbine, J. Sound
Vibr., 299, 869883, (2007).
6. H. Møller and C.S. Pedersen, Low frequency noise from large
wind turbines, J. Acoust. Soc. Am., 129(6), 37273744,
(2011).
7. UK Renewable, Wind turbine amplitude modulation: research
to improve understanding as to its cause and effect, (2013).
8. S. Oerlemans and J.G. Schepers, Prediction of wind turbine
noise and validation against experiment, Aeroacoustics, 8(6),
555584, (2009).
9. A. Laratro, M. Arjomandi, R. Kelso and B. Cazzolato, A dis-
cussion of wind turbine interaction and stall contributions to
wind farm noise, Journal of Wind Engineering and Industrial
Aerodynamics, 127,110, (2014).
10. G. Leventhall, A review of published research on low fre-
quency noise and its effects, Defra Publications, London,
(2003).
11. S. Morgan and R. Raspet, Investigation of the mechanisms of
low-frequency wind noise generation outdoors, J. Acoust. Soc.
Am., 92(20), 11801183, (1991).
12. R. Raspet, J. Webster and K. Dillon, Framework for wind
noise studies, J. Acoust. Soc. Am., 119(2), 834843, (2005).
13. H. Imaizumi and Y. Takahashi, Improvement of wind noise re-
duction performance of microphone-enclosed type wind-
screen, InterNoise10, (2010).
14. H. Tachibana, H. Yano and A. Fukushima, Assessment of
wind turbine noise in immission areas, 5th International Con-
ference on Wind Turbine Noise, (2013).
15. Wind turbine generator systems part 11: acoustic noise mea-
surement techniques, International standard IEC 61400-11, In-
ternational Electrotechnical Commission, (2012).
16. K. Betke, M. Schults von Glahn and O. Goos Measurement of
infrasound emitted by wind turbines, DEWEK, 207210,
(1996), (in German).
17. Engineering Sciences Data Unit (ESDU), Wind Engineering
(uid mechanics, external ow), Volume 3, London, (1977).
18. D. Kühner, Excess attenuation due to meteorological inu-
ences and ground impedance, AcusticaActa Acustica, 84,
870883, (1998).
19. TA-Luft, Erste Allgemeine Verwaltungsvorschrift zum Bundes-
Immissionsschutzgesetz Technische Anleitung zur Reinhal-
tung der Luft, (First general directive to the federal immission
protection actTechnical guideline for clean air), (1986), (in
German).
20. G.P. van den Berg, Effects of wind prole at night on wind tur-
bine sound, J Sound Vibr., 277, 955970, (2003).
21. Wind farms: environmental noise guidelines, South Australian
Environmental Protection Authority (EPA), Adelaide, (2009).
22. Marshall Day, Waterloo wind farm noise complaint, Rp005
2010277ML, (2012).
23. Acoustics Frequency-weighting characteristic for infrasound
measurements, International standard ISO 7196, International
Organization for Standardization, (1995).
24. Reference threshold of hearing under free-eld and diffuse-
eld listening conditions, International standard ISO 389-7,
International Organization for Standardization, (2005).
25. Suzlon technical documentation, n.d., V iew ed 12 No v ember 2013,
http://aurecon.webfoundryza.com/assets/les/vredenburgDSR/
Appendices/Appendix%20G/Turbine%20Types/Enclosure%20A%
20Technical%20Data_S88%20(WD00121-05-00).pdf.
26. J.C. Kaimal, J.C. Wyngaard, Y. Izumi and O.R. Coté, Spectral
characteristics of surface layer turbulence, Quarterly Journal
of the Royal Meteorological Society, 98, 563589, (1972).
27. Vestas product brochure, n.d., V iew ed 17 August 2013, http://www.
ceoe.udel.edu/windpow er/docs/ProductbrochureV90_3_0_UK.pdf.
28. Draft NSW Planning Guidelines: Wind Farms, NSW Gov-
ernment, Sydney, (2011).
82 Noise Control Engr. J. 62 (2), March-April 2014
... A typical measurement setup included a microphone that was positioned at 1.5 m above ground level (except H1 where a ground level microphone was used) and protected using a secondary windshield with a diameter of 450 mm (See Hansen et al. [28] for details). The microphone was typically positioned at least 10 m away from the residence and surrounding vegetation to minimize façade reflections and wind-induced vegetation noise. ...
... At all measurement locations, acoustic data were acquired using a Bruel and Kajer LAN-XI Type 3050 data acquisition system with a sampling rate of 8,192 Hz and a G.R.A.S type 40 AZ microphone with a 26CG preamplifier, which has a noise floor of 16 dB(A) and a flat frequency response down to 0.5 Hz. Further details of the experimental setup are described in [9,28]. ...
Article
Amplitude modulation (AM) is a characteristic feature of wind farm noise and has the potential to contribute to annoyance and sleep disturbance. Detection, quantification and characterisation of AM is relevant for regulatory bodies that seek to reduce adverse impacts of wind farm noise and for researchers and wind farm developers that aim to understand and account for this phenomenon. We here present an approach to detect and characterise AM in a comprehensive and long-term wind farm noise data set using human scoring. We established benchmark AM characteristics, which are important for validation and calibration of results obtained using automated methods. We further proposed an advanced AM detection method, which has a predictive power close to the practical limit set by human scoring. Human-based approaches should be considered as benchmark methods for characterising and detecting unique noise features.
... The use of ground-positioned microphones and (double) wind-screens are common practice nowadays to measure the noise immission of wind turbines. Although wind-induced microphone 4 noise can be significantly decreased in the high and middle frequency range in this way, the SNR remains inadequate at low and infrasound frequencies; reductions of wind-induced microphone noise are only a few decibels [5], [6]. A main problem is that the efficiency of a wind screen at low and infrasound frequencies is limited by its dimension and thus remains unable to shield the noise production of large turbulent eddies. ...
... The air flow encounters a stagnation point in front of the barrier, will accelerate near the top and faces an adverse pressure gradient behind it, leading to flow separation. Making the barrier porous allows 6 part of the flow to pass through. In this way, the stagnation point in front of the barrier is eliminated and the adverse pressure gradient behind the barrier is reduced. ...
Article
Wind-induced microphone noise complicates infrasound measurements considerably. A wind-shielding dome for signal-to-noise ratio improvement of acoustic pressure infrasound frequencies was designed and tested. The semi-spherical shape aimed at maximizing the pressure averaging of large atmospheric turbulent eddies while keeping the structure reasonably compact. The insertion loss of the dome was measured in a semi-anechoic chamber (in absence of flow) and showed nearly full transparency in the low frequency range. In an outdoor test, wind turbine infrasound was simultaneously measured with an uncovered and a dome-covered low-frequency microphone under different wind speeds and turbulence intensities. Largest improvements of the signal-to-noise ratio were measured at high mean wind speeds for frequencies down to 0.5 Hz. The dome allowed to clearly identify the infrasonic tonal components of wind turbines that were otherwise completely covered by the wind-induced microphone noise even at low mean wind speeds. The use of the dome thus opens possibilities for more accurately measuring infrasonic immissions from e.g. wind turbines.
... In addition to using a secondary wind screen of practical size, the microphone can be placed as close as possible to the ground, where the atmospheric turbulent fluctuations and vortex shedding about the wind screen are minimal. Commonly used secondary wind screens are of hemispherical or spherical shape, where the former is used for measurement at ground level and the latter for measurements at a height of 1.5 m [6]. Data presented in the ANSI/ASA S1.18-2010 standard [7] indicate that measuring at ground level is advantageous if the signal of interest is dominated by low frequencies. ...
... At all measurement locations, outdoor acoustic measurements were made using two microphones that were protected using a hemispherical or spherical type of wind screen. These wind screen designs are described in more detail in Hansen et al. [6]. At locations H1, H3 and H4, data were recorded using a National Instruments 9234, 24-bit data acquisition system with a sampling frequency of 10240 Hz and G.R.A.S. type 40 AZ 1/2 inch microphones with 20 CG G.R.A.S. pre-amplifiers. ...
Article
Full-text available
In the measurement of wind farm noise, it is standard practice to mount outdoor microphones at a height of 1.5 m. On the other hand, measurements at this height can be affected by wind-induced noise, which has the potential to mask the noise of interest, particularly at low and infrasonic frequencies. Therefore, to minimise wind-induced noise, it is advantageous to measure on or below the ground, where the wind speed is close to zero. However, results from measurements taken at any height other than 1.5 m must be interpreted with caution, due to different interference effects between direct and ground-reflected waves at each location. This investigation explores the feasibility of using a prediction model based on Nord2000 algorithms to correct the 1/3-octave sound pressure level measured at ground level to obtain a representative value for a height of 1.5 m. The model takes into account phase changes due to the difference in travel-time for the direct and reflected rays and finite ground impedance, multiple source contributions and incoherence due to turbulence. The focus is on propagation distances greater than 2 km, where limited validation of existing propagation models has been attempted previously. Comparison is made between the model and measurement results obtained at four locations near a wind farm, where microphones were mounted at a height of 1.5 m and at ground level. A lack of agreement between measurements and the model indicates that the efficient and practical correction method considered here is not feasible for long-range wind farm measurements. Thus, it is recommended that wind farm noise is measured at both 1.5 m (for mid- to high-frequency noise) and at ground level (for low-frequency noise, which is more affected by wind).
... At location H1, the outdoor data measured at a height of 1.5 m were not available, and thus data measured at ground level were used. Although the noise measured using a microphone mounted at ground level and 1.5 m is not exactly the same [14], the use of 1.5 m data was still reliable, particularly for AM quantification. The agreement between AM detection results in both data sets was high (accuracy = 0.82, F1-score = 0.78, AUC = 0.9) (See Supplementary Fig. S5 for details). ...
Preprint
Full-text available
The large-scale expansion of wind farms has prompted community debate regarding adverse impacts of wind farm noise (WFN). One of the most annoying and potentially sleep disturbing components of WFN is amplitude modulation (AM). Here we quantified and characterised AM over one year using acoustical and meteorological data measured at three locations near three wind farms. We found that the diurnal variation of outdoor AM prevalence was substantial, the nighttime prevalence was approximately 2 to 5 times higher than the daytime prevalence. On average, indoor AM occurred during the nighttime from 1.1 to 1.7 times less often than outdoor AM, but the indoor AM depth was higher than that measured outdoors. We observed an association between AM prevalence and sunset and sunrise. AM occurred more often at downwind and crosswind conditions. These findings provide important insights into long term WFN characteristics that will help to inform future WFN assessment guidelines.
... The microphone noise floor is also a good approximation of the complete recording system noise floor given that the acquisition device noise floor is below 0 dB(A). The outdoor microphone was mounted at a height of 1.5 m inside a spherical secondary windscreen with a diameter of 450 mm (Hansen et al., 2014). The outdoor microphone was positioned at least 20 m away from the residence and at least 10 m away from the surrounding vegetation to minimise façade reflections and wind-induced vegetation noise, respectively. ...
Article
Full-text available
This paper presents results from a one-year study of indoor annoyance and self-reported sleep times for two participants located near different wind farms. Continuous measurements of outdoor and indoor noise and meteorological conditions were taken at each location for the duration of the study. In at least 50% of the annoyance recordings, participants described noise as “swish” or “swoosh.” Furthermore, the majority of the annoyance recordings occurred at nighttime and in the early morning. The third quartile of A-weighted indoor sound pressure level [SPL(A)], between 27 and 31 dBA, was associated with an 88% increased probability of annoyance compared to the lowest reference quartile, which was between 12 and 22 dBA [odds ratio and 95% confidence intervals, 7.72 (2.61,22.8), p < 0.001]. The outdoor SPL(A) was also predictive of annoyance but only between 40 and 45 dBA. The outdoor prevalence of amplitude modulation (AM), defined as the percentage of time that AM was detectable by an algorithm for each annoyance period, was also associated with annoyance. Self-reported sleep efficiency (time spent asleep relative to time in bed available for sleep) was significantly associated with nighttime annoyance (β = −0.66, p = 0.02) but only explained a small fraction of the variance (R² = 5%).
... There were very occasional exceedances of the night-time DEFRA criteria at this site, but the percentage of exceedances was no greater than at two other locations with no wind turbines within 10 kilometres. Hansen et al. [64] measured wind farm noise in the 50 Hz third octave which was well above the DEFRA limit. ...
Article
This manuscript describes a range of technical deliberations undertaken by the authors during their work as members of the Australian Government’s Independent Scientific Committee on Wind Turbines. Central to these deliberations was the requirement upon the committee to improve understanding and monitoring of the potential impacts of sound from wind turbines (including low frequency and infrasound) on health and the environment. The paper examines existing wind turbine sound limits, possible perceptual and physiological effects of wind turbine noise, aspects of the effects of wind turbine sound on sleep health and quality of life, low-frequency noise limits, the concept of annoyance including alternative causes of it and the potential for it to be affected by low-frequency noise, the influence of amplitude modulation and tonality, sound measurement and analysis and management strategies. In so doing it provides an objective basis for harmonisation across Australia of provisions for siting and monitoring of wind turbines, which currently vary from state to state, contributing to contention and potential inequities between Australians, depending on their place of residence.
... There have been studies with respect to measuring wind turbine infrasound and low frequency noise, in relation to the attenuation of wind directly on the microphone; various size windscreens are identified in the US ANSI/ASA Standard S12.9-2016/Part 7 [60]. Hansen and others have undertaken investigations of double layered wind screens [61][62][63][64][65], for low frequency and infrasound wind turbine measurements. ...
Article
Full-text available
An issue exists around the world of wind farms that comply with permit conditions giving rise to noise complaints. Approval limits are normally expressed in A-weighted levels (dB(A)) external to residential receivers. The distance from the wind farm to residential receivers can result in difficulty in establishing the dB(A) contribution of the wind farm, as the overall noise includes background noise that can provide masking of the wind turbine noise. The determination of the ambient background at a receiver location (without the influence of the wind farm) presents challenges, as the background level varies with the wind and different seasons throughout the year. On-off testing of wind farms does not normally occur at high wind farm output and limits this approach for acoustic compliance testing of a wind farm. The use of a regression analysis method developed more than 20 years ago is questioned. Anomalies with respect to compliance procedures and the regression method of analysis based on real-world experience are discussed.
... The outdoor microphone was mounted at a height of 1.5 m and protected using a spherical secondary windscreen with a diameter of 450 mm. Details of the construction of this windscreen are provided in Hansen et al. [12]. The outdoor microphone was typically positioned at least 20 m away from the residence and at least 10 m from surrounding vegetation to minimise façade reflections and wind-induced vegetation noise, respectively. ...
Article
Full-text available
The presence of amplitude modulation (AM) in wind farm noise has been shown to result in increased annoyance. Therefore, it is important to determine how often this characteristic is present at residential locations near a wind farm. This study investigates the prevalence and characteristics of wind farm AM at 9 different residences located near a South Australian wind farm that has been the subject of complaints from local residents. It is shown that an audible indoor low-frequency tone was amplitude modulated at the blade-pass frequency for 20% of the time up to a distance of 2.4 km. The audible AM occurred for a similar percentage of time between wind farm percentage power capacities of 40 and 85%, indicating that it is important that AM analysis is not restricted to high power output conditions only. Although the number of AM events is shown to reduce with distance, audible indoor AM still occurred for 16% of the time at a distance of 3.5 km. At distances of 7.6 and 8.8 km, audible AM was only detected on one occasion. At night-time, audible AM occurred indoors at residences located as far as 3.5 km from the wind farm for up to 22% of the time.
Article
The large-scale expansion of wind farms has prompted community debate regarding adverse impacts of wind farm noise (WFN). One of the most annoying and potentially sleep disturbing components of WFN is amplitude modulation (AM). Here we quantified and characterised AM over one year using acoustical and meteorological data measured at three locations near three wind farms. We found that the diurnal variation of outdoor AM prevalence was substantial, the nighttime prevalence was approximately 2 to 5 times higher than the daytime prevalence. On average, indoor AM occurred during the nighttime from 1.1 to 1.7 times less often than outdoor AM, but the indoor AM depth was higher than that measured outdoors. We observed an association between AM prevalence and sunset and sunrise. AM occurred more often at downwind and crosswind conditions. These findings provide important insights into long term WFN characteristics that will help to inform future WFN assessment guidelines.
Preprint
Full-text available
Amplitude modulation (AM) is a characteristic feature of wind farm noise and has the potential to contribute to annoyance and sleep disturbance. This study aimed to develop an AM detection method using a random forest approach. The method was developed and validated on 6,000 10-second samples of wind farm noise manually classified by a scorer via a listening experiment. Comparison between the random forest method and other widely-used methods showed that the proposed method consistently demonstrated superior performance. This study also found that a combination of low-frequency content features and other unique characteristics of wind farm noise play an important role in enhancing AM detection performance. Taken together, these findings support that using machine learning-based detection of AM is well suited and effective for in-depth exploration of large wind farm noise data sets for potential legislative and research purposes.
Article
Full-text available
Sound from wind turbines involves a number of sound production mechanisms related to different interactions between the turbine blades and the air. An important contribution to the low frequency part of the sound spectrum is due to the sudden variation in air flow which the blade encounters when it passes the tower: the angle of attack of the incoming air suddenly deviates from the angle that is optimized for the mean flow. Hitherto, low-frequency sound from wind turbines has not been shown to be a major factor contributing to annoyance. This seems reasonable as the blade passing frequency is of the order of one hertz where the human auditory system is relatively insensitive. This argument, however, obscures a very relevant effect: the blade passing frequency modulates well audible, higher-frequency sounds and thus creates periodic sound: blade swish. This effect is stronger at night because in a stable atmosphere there is a greater difference between rotor averaged and near-tower wind speed. Measurements have shown that additional turbines can interact to further amplify this effect. Theoretically the resulting fluctuations in sound level will be clearly perceptible to human hearing. This is confirmed by residents near wind turbines with the same common observation: often late in the afternoon or in the evening the turbine sound acquires a distinct 'beating' character, the rhythm of which is in agreement with the blade passing frequency. It is clear from the observations that this is associated to a change toward a higher atmospheric stability. The effect of stronger fluctuations on annoyance has not been investigated as such, although it is highly relevant because a) the effect is stronger for modern (that is: tall) wind turbines, and b) more people in Europe will be living close to these wind turbines as a result of the growth of wind energy projects.
Article
Full-text available
Morgan and Raspet [J. Acoust. Soc. Am. 92, 1180–1183 (1992) ] performed simultaneous wind velocity and wind noise measurements and determined that the wind noise spectrum is highly correlated with the wind velocity spectrum. In this paper, two methods are developed for predicting the upper limits of wind noise pressure spectra from fluctuating velocity spectra in the inertial range. Lower limits on wind noise are estimated from two theories of the pressure fluctuations that occur in turbulence when no wind screen or microphone is present. Empirical results for the self-noise of spherical and cylindrical windscreens in substantially nonturbulent flows are also presented. Measurements of the wind velocity spectra and wind noise spectra from a variety of windscreens are described and compared to the theoretical predictions. The wind noise data taken at the height of the anemometer lies between the upper and lower limits and the predicted self-noise is negligible. The theoretical framework allows windscreen reduction to be evaluated in terms of the turbulent inflow properties and establishes practical upper limits on wind noise reduction for varying wind conditions.
Article
Full-text available
The aerodynamic noise production mechanisms of modern horizontal axis wind turbines are reviewed. An engineering analysis of the time and frequency scales from three noise sources, leading edge turbulence interaction noise, trailing edge noise and blade-tower interaction noise is presented. The analysis shows that noise sources are present from low-frequencies (1-4 Hz) to over 500 Hz for a representative wind turbine. The results of the analysis are used to explain amplitude modulation observed during noise measurements at a European wind farm. Daytime noise measurements close to a South Australian wind farm are also presented that show amplitude modulation. The paper concludes with a description of conceptual ideas for the control of wind turbine noise.
Article
Wind noise reduction is of great importance to perform exact measurements of low frequency noises, especially under windy condition, and this might be an essential technique for evaluation or assessment in the development of wind energy. The authors developed a windscreen that the microphone with a 90 mm spherical windscreen of sound level meter is set up inside a thin iron frame, being covered with urethane foam, and approximate 20 dB attenuation of the wind noise at frequencies from 3 to 10 Hz was observed in outdoor measurements. One of the current subjects on the windscreen is to improve both the reduction performance of wind noise and the portability. In the paper, we show noise reduction performance of a modified windscreen, in which the microphone with the spherical windscreen is covered additionally with wind-shielding mesh for fruit farming.
Article
A semi-empirical prediction method for trailing edge noise is applied to calculate the noise from two modern large wind turbines. The prediction code only needs the blade geometry and the turbine operating conditions as input. Using detailed acoustic array and directivity measurements, a thorough validation of the predictions is carried out. The predicted noise source distribution in the rotor plane (as a function of frequency and observer position) shows the same characteristics as in the experiments: due to trailing edge noise directivity and convective amplification, practically all noise (emitted to the ground) is produced during the downward movement of the blades, causing an amplitude modulation of broadband aerodynamic blade noise at the blade passing frequency ('swish'). Good agreement is also found between the measured and predicted spectra, in terms of levels and spectral shape. For both turbines, the deviation between predicted and measured overall sound levels (as a function of rotor power) is...
Article
Simultaneous measurement of wind noise and the instantaneous wind speed were performed for bare and screened microphones outdoors. Analysis of these measurements demonstrates that the dominant source of pressure fluctuations at the microphone outdoors is the intrinsic turbulence in the flow. This is in contrast to the results of measurements performed in low‐turbulence environments by Hosier and Donavan [Natl. Bur. Stand. Rpt. NBSIR79‐1599 (Jan. 1979)] and by Strasberg [J. Acoust. Soc. Am. 83, 544–548 (1988)]. For low‐turbulence conditions the fluctuating wake of the screen is the dominant noise source. This finding has important implications for windscreen design for outdoor measurements since the principles described by Hosier and Donavan apply only to low‐turbulence conditions.