Journal of Sound and Vibration 400 (2017) 154–166

Contents lists available at ScienceDirect

Journal of Sound and Vibration

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

Location of aerodynamic noise sources from a 200 kW

vertical-axis wind turbine
Fredric Ottermo a,n, Erik Möllerström a,b, Anders Nordborg c, Jonny Hylander a,
Hans Bernhoff b
a
The Rydberg Laboratory for Applied Sciences, Halmstad University, PO Box 823, SE-301 18 Halmstad, Sweden
b
Division for Electricity, Department of Engineering Sciences, Uppsala University, PO Box 534, SE-751 21 Uppsala, Sweden
c
Sound View Instruments, Hoby Gård, 27636 Borrby, Sweden

a r t i c l e in f o abstract

Article history: Noise levels emitted from a 200 kW H-rotor vertical-axis wind turbine have been mea-
Received 22 August 2016 sured using a microphone array at four different positions, each at a hub-height distance
Received in revised form from the tower. The microphone array, comprising 48 microphones in a spiral pattern,
9 February 2017
allows for directional mapping of the noise sources in the range of 500 Hz to 4 kHz. The
Accepted 27 March 2017
produced images indicate that most of the noise is generated in a narrow azimuth-angle
Handling Editor: M.P. Cartmell
range, compatible with the location where increased turbulence is known to be present in
the flow, as a result of the previous passage of a blade and its support arms. It is also
Keywords: shown that a semi-empirical model for inflow-turbulence noise seems to produce noise
Vertical-axis wind turbine
levels of the correct order of magnitude, based on the amount of turbulence that could be
H-rotor
expected from power extraction considerations.
Noise
Microphone array & 2017 Elsevier Ltd All rights reserved.
Beamforming

1. Introduction

With the current rapid increase in the number and size of wind-power installations worldwide, it is important to
consider different environmental aspects of this expansion. Wind power shows a potential of supplying a major fraction of
the global energy demand [1], and such a penetration will lead to an increased number of people living near wind turbines.
Thus, aspects such as low noise and appealing aesthetics are likely to be key factors for this expansion to acquire general
acceptance. This motivates the evaluation of new wind-power concepts with respect to these aspects, especially noise
performance.
Vertical-axis wind turbines (VAWTs) have been proposed as an alternative to the more common horizontal-axis wind
turbines (HAWTs). Overshadowed by the commercial success of the HAWT design, which is now a big industry, the VAWT
concept has several features that still make them interesting to study. The VAWTs typically have fewer moving parts and a
generator located at ground level, which could ultimately lead to higher availability and lower maintenance costs [2].
Additionally, due to a lower tip-speed ratio (TSR), the VAWT concept has been anticipated to allow lower noise levels.
Noise from operating wind turbines can be divided into aerodynamic and mechanical noise. Aerodynamic noise is of
broadband character and originates from various complex flow phenomena when the air flows around the turbine. Me-
chanical noise originates from the relative motions of various mechanical components. For modern turbines, the

n
Corresponding author.
E-mail address: fredric.ottermo@hh.se (F. Ottermo).

http://dx.doi.org/10.1016/j.jsv.2017.03.033
0022-460X/& 2017 Elsevier Ltd All rights reserved.
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 155

Nomenclature S Compressible Sears function

S0 Reference area of 1 m2
A Cross-section area of turbine, m2 u Wind speed, m/s
At Cross-section area of turbulent volume, m2 U Local velocity over the airfoil, m/s
c Airfoil chord length, m vblade Blade velocity, m/s
c0 Speed of sound, m/s α Measurement-position angle
Cp Power coefficient β 1 − M2
DL Directivity factor Φe Angle specifying the retarded observer
f frequency, Hz position
I Turbulence intensity ρ0 Density of air, kg/m3
k Local wave number θ Blade azimuth angle
ke Wave number of the energy-containing wa- Θe Angle specifying the retarded observer
velength scale position
K kc/2 LFC Low-frequency correction
ℓ Turbulence length scale, m SPL Sound-pressure level, dB or dBA, reference
L Blade span, m value 2 × 10−5 Pa
M Local Mach number, U /c0 SWL Sound-power level, dBA, reference value
P Power, W 10−12 W
r Source-to-observer distance, m TKE Turbulent kinetic energy, m2/s2
re Retarded source-to-observer distance, m TSR Tip-speed ratio

aerodynamic noise is generally dominant [3]. The VAWT design allows for the drive train to be located at ground level,
which may further limit mechanical noise propagation [2].
Extensive research has been presented regarding noise from wind turbines (see, e.g., [4–7]). Most of this research focused
on noise from HAWTs, motivated by the large number of installed HAWTs and reported annoyances at some installations. In
[4], it was shown that, for a modern HAWT, most of the noise is created close to the blade tip when the blade travels
downwards (toward the receiver). This aerodynamic noise was identified as turbulent-boundary-layer trailing-edge (TBL-
TE) noise, and the result is an amplitude-modulated characteristic swishing, due to the directivity of the trailing-edge noise
and convective amplification. This mid-frequency phenomenon (400–1000 Hz) has been found to be the most annoying [3].
Interaction between inflow turbulence and the airfoil leading edge also generates noise, which generally dominates the low-
frequency part of the HAWT noise spectrum [8].

1.1. VAWT noise generation

The generated aerodynamic noise generally increases with the local speed of the blade. The VAWTs usually have lower
TSR than HAWTs. Lower levels of aerodynamic noise might then be expected for a VAWT due to the relatively low blade
velocity compared to HAWTs. However, there are important differences between these designs with respect to noise
generation. First, VAWTs encounter highly unsteady flow properties and varying angles of attack (e.g., dynamic stall at low
TSR [9]). Second, at the downstream half of the rotation, the blades of a VAWT pass the wake of the blades at the upstream
half [10–12]. The turbulence levels at the downstream half of a VAWT are expected to be much larger than the turbulence of
the flow ahead of the turbine.
Tonal components, mainly harmonics of the blade-passage frequency, are generally expected to be present in the VAWT
noise spectrum due to the unsteady blade loading [13]. However, for large VAWTs, the blade-passage frequency and a major
part of its harmonics fall outside the audible frequency range.

1.2. VAWT noise prediction

A few recent studies have specifically considered noise prediction in the context of VAWTs. In [13], the noise char-
acteristics of a model VAWT were investigated and the applicability of different noise-prediction models was examined.
Both the harmonic content and the broadband content were modeled. In [14,15], two-dimensional (2D) vortex methods
were used to simulate aerodynamic noise from VAWTs. Both studies indicate lower noise levels compared to HAWTs. The
VAWT noise studies based on 2D CFD setups were presented in [16,17], the latter including experimental validation using a
small VAWT at very low TSR. In [18], noise predictions using a semi-three-dimensional large-eddy simulation were per-
formed, and [19] used an unsteady three-dimensional (3D) inviscid panel method to predict parts of the noise spectrum.
These studies consider small turbines and/or simplified geometries, which limit their applicability for large VAWTs.
Initial noise-emission measurements of the 200 kW VAWT considered in this study were presented in [20], where it was
suggested that the noise from VAWTs of this size is likely to be of a different origin than the trailing-edge noise dominant for
large HAWTs.
156 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

1.3. Aim of the current study

Location of the aerodynamic noise sources (i.e., quantifying where in the rotor volume the noise originates) is of great
interest, as it allows for deeper understanding of the underlying mechanisms of the noise. Location of the noise sources from
HAWTs has been studied in [4], using a microphone array and beamforming. In [13], microphone-array measurements were
performed on a down-scaled VAWT in a wind tunnel, indicating that, at low TSR, the dominant noise sources are found in
the upstream half of the rotation (interpreted to be due to dynamic stall), while, at higher TSR, the dominant sources were
found in the downstream half. Microphone-array measurements on full-scale VAWTs have not previously been performed.
This paper investigates the location of the aerodynamic noise sources of a 200 kW VAWT, using an array of 48 micro-
phones. The resulting images are used to draw conclusions regarding the mechanism behind the dominant noise source. The
study may help in distinguishing which measures are likely to be most effective for reducing noise generation for an H-rotor
VAWT of this size.

2. Theory

The semi-empiric model for airfoil self-noise proposed in [21] has been commonly used for HAWT noise predictions
[22,23]. This semi-empirical model is based on static angle-of-attack measurements, and good applicability to the HAWT
case can be expected, whereas the VAWT case is more uncertain due to the unsteady flow. However, at nominal TSR, the
dynamic-stall effect is small [9], suggesting that static models might have a chance to apply reasonably well. In support for
this hypothesis, static lift-and-drag data do apply reasonably well to the VAWT for performance calculations, despite the
unsteady aerodynamics [24]. The different self-noise mechanisms modeled in [21] were applied to a small VAWT setup in
[13], and an experimental comparison was made. The TBL-TE noise and the laminar-boundary-layer vortex-shedding (LBL-
VS) noise were the dominant self-noise contributions in the mid-frequency range. These mechanisms were found be
compatible with the measured noise levels up to 1 kHz, provided that an inflow-turbulence noise model was added and
tuned in (the inflow-turbulence noise then dominated this part of the spectrum). Above 1 kHz, the LBL-VS model was found
to overestimate the noise, and it was speculated that a laminar boundary layer might not always form in the unsteady flow.
For larger turbines (higher Reynolds numbers), the boundary layer will be turbulent, and the LBL-VS contribution is less
important [21]. In [20] the TBL-TE noise model was applied to the current VAWT geometry. The predicted levels were far
below the measured values, suggesting that inflow-turbulence noise is likely to play a dominant role for noise from a VAWT
of this size as well.
The flow of a small two-bladed VAWT was mapped experimentally in [11], and in Fig. 17 of that work, a significant
amount of turbulence is generated by the blades and struts (support arms). The velocity fluctuations appear to be domi-
nated by the tip vortices, but a significant amount of turbulence is also seen to be generated by the struts and tower,
apparently, a lot more than the contribution from the trailing-edge vorticity. The tip-vortex trajectories expand with the
widening of the flow and do not seem to cross the paths of the blades behind, as they move from the most downstream
position toward the most upwind position. However, the turbulence generated by the tower and struts is observed to collide
with the downstream blade path. This means that turbulent inflow might be a potential noise source for VAWTs, irre-
spective of the atmospheric properties.

2.1. Inflow-turbulence noise modeling

A semi-empirical model for inflow-turbulence noise has been used in the context of HAWTs [8,22,25,26] for inflow
turbulence produced by the atmospheric boundary layer. In [8], the model was compared to an extensive set of experi-
mental data, accompanied with an accurate determination of the inflow-turbulence characteristics. The model was found to
predict the measured noise levels reasonably well in the mid-frequency range (which will be the range of interest here).
According to this model, the sound-pressure level in dB, in one-third-octave bands, is [8]:
⎛ ρ 2 c 2ℓL (k/k e )3 ⎞
LFC
SPL = 10log10⎜⎜ 0 2 M3u2I 2 D ⎟ + 78.4,
0
2 7/3 1 + LFC L ⎟
⎝ 2re (1 + (k/k e ) ) ⎠ (1)

with SI units for all parameters, where ρ0 is the density of air, c0 is the speed of sound, ℓis a turbulence length scale, L is the
blade span (segment), re the retarded distance to the observer, M is the Mach number of the local velocity over the airfoil, u
is the wind speed, I is the turbulence intensity, k = 2πf /U is the local wavenumber, where f is the frequency, U is the local
velocity over the airfoil, ke = 0.75/ℓ is the wavenumber of the energy-containing wavelength scale of the turbulence, and
LFC is a low-frequency correction given by:

S2MK
2 ⎛ 2π K 1 ⎞−1
LFC = 10 , S2 = ⎜ 2 + ⎟ ,
β2 ⎝ β 1 + 2.4K /β 2 ⎠ (2)
2
where K = kc/2 with c being the airfoil chord length, and β = 1 − M . Moreover, S approximates the compressible Sears
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 157

Fig. 1. Definition of the angles Θe and Φe, and the retarded distance re for a source at the leading edge of a vertical blade. The angles are calculated with
respect to the retarded observer position.

function [27]. The directivity factor used is that of a translating dipole, which is applicable for low frequencies [21,22]:

sin2Θesin2Φe
DL ≈ .
(1 + M cosΘe )4 (3)

The angles Θe and Φe are indicated in Fig. 1 for a source at the leading edge of the airfoil. The retarded observer position is
calculated using the free-stream wind speed (see [13]), and the angles Θe and Φe are calculated with respect to the retarded
position. As can be seen in [23] for the case of TBL-TE noise, the low-frequency directivity pattern applies reasonably well at
wavelengths of the order of the chord, which is what we have here. The local Mach number M and the local wind speed U
are calculated using a simple flow model where the flow is unperturbed by the turbine. Doppler shift is also neglected; for
the velocities encountered here (at most about 717 m/s in the direction toward the observer), intensities and frequencies
shift by only 0.5 dB and 5%, respectively [28].
The model is sensitive to the turbulence length scale ℓ. For the case of atmospheric turbulence in the context of HAWTs,
the integral length scale of the turbulence has been commonly used [22]. Here, we assume that most inflow turbulence is
due to the turbine itself, and the corresponding length scale is not obvious, but the chord length (∼1 m) or blade thickness
(∼0.2 m) appears as a reasonable guess, or possibly the larger structures of the turbine (∼10 m). Moreover, the model
assumes a homogeneous spectrum of frequencies, whereas it is reasonable to expect that certain frequencies might
dominate for the turbulence generated by the turbine.
The turbulence intensity is defined as [29]:

u′2
I= ,
u (4)

where u′ is the velocity fluctuation (the component along the flow), and the bar indicates time-averaged value. The tur-
bulence kinetic energy can be related to I, assuming isotropic turbulence, according to [29]:
1 3
TKE =
2
( 2
)
u′2x + u′2y + u′2z = (Iu)2 .
(5)

This can be used to estimate the energy content of the velocity fluctuations within a volume of air.

2.2. Sound-power level

The sound-power level (SWL) of the noise source can be estimated from a directional SPL measurement according to
[30]:
⎛ 4π r 2 ⎞
SWL = SPL + 10log10⎜ ⎟,
⎝ S0 ⎠ (6)
2
where S0 is a reference area of 1 m , and r is the distance from the SPL observation to the source. A spherically homogeneous
noise distribution is assumed in Eq. (6). Note that a microphone array enables directional sound measurements, with good
suppression of ground-reflected sound, for example.
158 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

2.3. Beamforming

The phase differences between the microphone signals of a microphone array hold information about the positioning of
the sources. By defining a scanning grid (i.e., a set of imposed source positions) a map of source strengths may be de-
termined by requiring them to, as closely as possible, reproduce the actual phase differences between signals of the mi-
crophone array. This process is known as beamforming [31]. The scanning grid may be 2D or 3D, and may also be defined to
follow some emitting object. We use a stationary scanning grid, as we are primarily interested in how the sources are
distributed along the trajectories of the blades. Additionally, a 2D grid is considered sufficient for this study, despite the 3D
motion of the turbine. This is motivated by the fact that it is still possible to unambiguously pinpoint the source positions
along the revolution, due to an off-symmetry-axis measurement position. The measurement of the small VAWT in [13] also
used a 2D scanning grid (though sometimes curved). The symmetric measurement position used in [13] made it difficult to
distinguish between sources at the front and back half of the rotor, but this was compensated for by creating separate source
maps for different ranges of blade positions.
In the present case, the scanning grid is set to a plane that intersects the hub and is perpendicular to the line between the
microphone array and the hub (see Fig. 4 below). This means that source locations may be up to 18 m away from the
scanning grid, amounting to 32% of the distance from the microphone array to the scanning grid. At these points the sources
will be less well localized. For the flat scanning grid used for most of the analysis in [13], the sources miss the scanning grid
by up to 44% of the distance to the array; however, the accuracy in the source localization was still deemed sufficient for the
analysis.
Standard ways of improving the accuracy of the microphone-array beamforming include diagonal deletion of the cross-
spectral matrix [32], and deconvolution techniques (e.g., DAMAS [33]). In the beamforming process, the cross-spectral
matrix captures the phase differences between the different microphone signals. Many types of noise, for example wind-
induced noise, are not coherent between the microphones; the random phase differences cause the noise to only contribute
to the diagonal of the matrix (the auto powers). Consequently, the source maps are typically greatly improved by elim-
inating these diagonal elements. The deconvolution process aims at undoing the spreading of point sources (e.g., addition of
side lobes) that naturally occurs in the classical beamforming process due to a finite number of microphones.

3. Setup and observations

3.1. Wind-turbine system

The VAWT we consider in this study is a 200 kW H-rotor, referred to as the T1 turbine. The turbine was designed and
erected in 2010 by the company Vertical Wind AB in collaboration with Uppsala University and is located at Thorsholm
(56°56′29”N 12°30′38”E) just outside of Falkenberg, Sweden. The T1 turbine has a direct-drive permanent-magnet syn-
chronous generator mounted at the bottom of the tower and connected to the rotor by a steel shaft. The rotor consists of
three 24-m-long straight blades that are connected to the shaft by two struts each (see Fig. 2). Both blades and struts are
made of fiberglass. The blades are fixed, and the variable speed of the turbine is used to control the stall effect so that the
rated power can be attained between the rated wind speed and the cut-out wind speed [34,35]. The T1 turbine is a first
prototype, and the noise level was not considered during design. Properties of the T1 turbine can be seen in Table 1.

3.2. Experimental setup

The microphone array comprises 48 microphones positioned along six spiral arms. The array diameter is about 1.5 m.
Microphone positions have been calculated with the help of the equation for Fermat's spiral, resulting in an irregular
microphone configuration (as opposed to a periodic one, e.g., a rectangular grid). Similar multiple-spiral-arm patterns has
been observed to produce high side-lobe suppression [36].
Class 1 BSWA 1/4” phase-matched array microphones have been used. Microphone signals were recorded with two
synchronized Alesis HD24 hard-disk recorders, at the sampling rate of 44 100 Hz. The array was positioned 39.5 m away
from the center of the wind turbine and directed toward the top of the tower. The entire arrangement can be seen in Fig. 3.
The T1 turbine is situated in a plain field with a heavily trafficked freeway 750 m to the southwest. Interfering noise from
the freeway was minimized as the recordings were performed during late evening.

3.3. Data processing

Each recording took about 20 s, representing about five revolutions of the rotor blades. The calculated source maps
represent the average sound radiation during these five laps. Source maps were generated using Acoular software [37,38]
with a basic delay-and-sum algorithm in the frequency domain and diagonal deletion to remove wind-generated micro-
phone noise. Diagonal deletion also reduced adverse effects of background noise from the nearby freeway, which might be
explained by loss of coherence as the sources are distant and obscured by elevations in the landscape, forcing the sound to
travel along different paths [32]. A subsequent DAMAS deconvolution removed side-lobe effects and improved image
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 159

Fig. 2. The 200 kW VAWT considered in this study, called the T1 turbine. The turbine is in Falkenberg, Sweden.

Table 1
Properties of the T1 turbine.

Rated power 200 kW

Turbine diameter 26 m
Hub height (incl. foundation) 40 m
Blade length 24 m
Blade chord 0.4–0.9 m
Swept area 624 m2
Cut-in wind speed 4 m/s
Rated wind speed 12 m/s
Cut-out wind speed 25 m/s
Survival wind speed 60 m/s
Rotational speed 16–33 rpm
Nominal TSR 3.8
Power regulation Stall
Blade/strut material Fiberglass composite

resolution. Sound-pressure-level spectra and turbine sound-pressure levels were produced by integrating over the source
maps.

3.4. Observations

Recordings were performed at four positions, all at the same distance of 39.5 m from the turbine center but at different
angles α to the wind direction (see Fig. 4). In Table 2, wind speeds, measurement angles, and ambient turbulence intensity
for all four positions can be seen. The data originates from an anemometer at a 40 m measurement mast located less than
100 m to the southwest of the turbine. The 10 m wind-speed values are calculated from the 40 m values, using a power-law
wind profile with exponent 0.17.
The measurements were performed between 22:30 and 23:30 on 2015-06-15. The wind direction was close to western.
The turbine was operating at the lower limit rotational speed, staying within 15.8–16.2 rpm during the entire operational
time. The humidity was just above 100%, the temperature was around 11 °C and the pressure was steady at 101.5 kPa during
the entire measurement.
160 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

Fig. 3. The microphone array used throughout this work. The array comprises 48 microphones in a spiral pattern.

Fig. 4. Side view (a) and view from above (b) of the measurement setup. The scanning grid orientation with respect to the measurement position is shown,
and the blade azimuth angle θ and measurement-position angle α are defined.
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 161

Table 2
Wind speed, measurement-position angle, and measured atmospheric turbulence intensity for the measurement positions. The wind direction angle is
relative to the measurement-position direction.

Position Wind speed at hub height (m/s) Wind speed at 10 m (m/s) Measurement angle α Turbulence intensity

1 4.7 3.7 31° 0.20

2 6.0 4.7 126° 0.11
3 5.4 4.2 218° 0.12
4 5.0 3.9 303° 0.13

Fig. 5. One-third-octave bands in dB for position 2. The turbine location is indicated by the overlaid CAD model. The 2D scanning grid is located at the hub,
perpendicular to the hub-observer line. The range of the dB scale is 10 dB for all bands, and the upper limit has been adjusted for each band.
162 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

Fig. 6. The 800 Hz octave, measured at the four positions. The scale is in dB. The turbine location, with the blade-tip trajectories also shown, is indicated by
the overlaid CAD model.

4. Results and discussion

4.1. Measurements

Fig. 5 shows one-third-octave-band source maps from the measurement at position 2. It is evident that, at low fre-
quencies (around 800 Hz), the main noise sources are located at a very specific azimuth angle of the blade. At higher
frequencies, the sources are distributed over a wider range of blade azimuth angles, seemingly along the trajectories of the
blade-strut joints (i.e., where the blade attaches to the struts). The upper limit of the dB scale varies for the different bands
in Fig. 5, the levels at low frequencies being generally larger (see Fig. 8 below). The distinct pattern of the higher-frequency
images is of help when mapping the scanning grid onto the turbine CAD-model image. Once the CAD-model image is
inserted, it is possible to approximately conclude where, along the revolution, the different noise sources are located. For
this, however, we need to assume that the bilateral symmetry of the turbine is also reflected in the source distribution, an
assumption which appears to be compatible with the images. Sources at the upper half of the blade are typically accom-
panied by mirrored sources at the lower half of the blade, which help distinguish between the front and back half of the
revolution. In some cases, two different interpretations of the source locations may be conceivable, but only one remains
after also considering another measurement position (see, e.g., Fig. 6 below).
For a comparison between the different measurement positions, Figs. 6 and 7 show octave bands at 800 Hz and 3150 Hz
respectively. The location of the noise sources seen in Figs. 6 and 7 are consistent between the positions. The lower-fre-
quency source, for example, is located at the same blade azimuth angle in all four positions. The azimuth angle can be
extracted by varying the blade azimuth angle in the overlaid perspective-corrected CAD model and comparing to the source
locations in Fig. 6, a procedure that results in an azimuth angle of θ = 88°. This is where the blade travels toward the flow
(see Fig. 4); hence, the relative wind speed is close to maximal.
The noise spectrum in one-third-octave bands, extracted from integrating the intensities in the source maps, is shown in
Fig. 8 for positions 1–4. A-weighting has been applied to obtain the levels in dBA. It is clear from the spectra and from Fig. 6,
that the levels at position 1 are substantially lower, especially at lower frequencies, which indicate directivity. It is also clear
that the lower-frequency part of the spectrum dominates the overall noise level.
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 163

Fig. 7. The 3150 Hz octave, measured at the four positions. The scale is in dB. The turbine location, with the blade-strut-joint trajectories also shown, is
indicated by the overlaid CAD model.

Fig. 8. The one-third-octave band spectra in dBA for positions 1–4. The levels in (a) are derived from integrating the source maps, and in (b), they are
estimated using the turbulent-inflow model as given in Eq. (1).

Based on these results and the following argument, it is reasonable to assume that the lower-frequency noise is a result
of turbulent inflow on the blade leading edge due to turbulence created by the turbine itself. It has been argued in [20] that
TBL-TE noise is not likely to be relevant for this VAWT. Furthermore, the position of the lower-frequency source appears to
be within a narrow region where indeed high turbulence can be expected, originating predominantly from the blade-strut
joint of the previous blade passage, as seen in Fig. 17 in [11]. This is compatible with the presence of a higher-frequency
164 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

noise contribution that is observed at the precise position of the blade-strut joint, suggesting that turbulence is likely to be
created there. The model for turbulent-inflow noise in Eq. (1) indicates a strong dependence on the local wind speed, which
is indeed maximal at the observed source position. Finally, considering the ambient inflow turbulence, this contribution is
present all along the revolution and cannot explain the observed narrowness of the source region.

4.2. Noise prediction using the inflow-turbulence model

It is possible to make an order-of-magnitude check using the model in Eq. (1) to attain an indication regarding whether
the turbulent-inflow hypothesis described above is reasonable. The extra turbulence produced by the struts and joints must
be accompanied by a loss in converted power. If the struts were (hypothetically) removed, we expect the power coefficient
to be at least of the order of so-called full-Darrieus turbines (egg-beater design, without struts), which have been reported to
achieve Cp ¼0.4 or slightly better [39]. Indeed, ignoring drag due to the presence of struts, the H-rotor is expected to allow
for better aerodynamic performance than the full Darrieus [24,40]. For the full-Darrieus turbine, the blade works in the stall
region for parts of the cross-section area, decreasing the contribution from these parts. Since the power coefficient of the T1
turbine is measured to peak at about Cp ¼0.33 [41], we may assume a power-coefficient loss of ΔCp = 0.07 due to the
presence of the struts. The corresponding power loss at hub-height wind speed u ¼6 m/s amounts to:
1
ΔP = ΔCpρ0 Au3 ∼ 5.7 kW,
2 (7)
3 2
using air density ρ0 = 1.2 kg/m and rotor cross-section area A = 26 × 24 m . The extra kinetic energy ΔE contained within a
volume V due to the turbulence is ΔE = ρ0 V TKE . Assuming that the power loss of Eq. (7) is contained in an extra amount of
air turbulence that is convected with the flow, the corresponding turbulence intensity may be calculated from Eq. (5):

1 2 1 2
I= TKE = ΔP /(ρ0 V̇ ) ∼ 0.40.
u 3 u 3 (8)

Here, the volume flow rate is V̇ = At U , where the cross-section area At of the turbulent volume convecting downstream is
estimated from the pictures to be 2  10 m2 at the position where noise is emitted, which is where we are interested in the
turbulence intensity. (Shortly after the blade-strut joint, where the turbulence is expected to be created, the area is smaller
and the turbulence intensity larger.) The local velocity is estimated as U ≈ u + vblade = 28 m/s since most of the turbulence is
created when the blade moves toward the wind. Furthermore, for Eq. (8) to hold, it is assumed that each blade effectively
contributes to the power loss according to Eq. (8) during one-third of the revolution. Note that the turbulence intensity
estimated in Eq. (8) is much larger than the ambient turbulence intensities given in Table 2, which appears reasonable given
the measurements in [11].
Using this value of the turbulence intensity (which we assume to be valid for the other wind speeds as well), it is possible
to estimate the corresponding emitted sound spectrum in the directions of the measurement positions using Eq. (1). The
turbulence length scale is taken to be ℓ = 0.2 m , and the blade span where sound is emitted is estimated from the pictures to
be L ¼10 m. Guided by the source maps, the sound is modeled to originate from two regions: one in the upper region and
one in the lower region of the blade. The sound is further modeled to be emitted during a 5 m distance per blade around the
azimuth angle θ = 88° (again estimated from the pictures). The result of this calculation is shown in Fig. 8 (b).
Comparing the measured levels in Fig. 8(a) to the calculated levels in Fig. 8(b), we note that the overall levels seem to
match reasonably well for positions 2–4. It should be noted, that the calculated levels are sensitive to the choice of ℓ, which
is not exactly known and has been tuned in. However, the current choice appears perfectly reasonable, as it coincides with a
length scale present in the system (the blade thickness).
For both the measured and modeled spectra, the levels at position 1 are clearly below the levels at the other positions.
However, the modeled-spectra levels at position 1 are substantially lower than the measured ones (apart from the 1000 Hz
band). The reason that position 1 experiences lower levels in the model is the directivity factor DL in Eq. (3), which, at
position 1, will be suppressed as Φe is very close to zero. However, the uncertainty in this angle is probably big, due to
varying wind direction, for example. Therefore, we cannot expect to see such sharp directional cancellations as predicted by
this simple model. Another consideration is that the ambient turbulence intensity was larger during the measurement at
position 1, as seen in Table 2. The noise produced by ambient turbulence is relevant at a wider range of blade azimuth
angles, even if the levels are generally low. However, if the directivity factor suppresses noise due to self-generated tur-
bulence, the noise contribution due to the ambient turbulence might also become relevant.
The measured spectra in Fig. 8 indicate spectral peaks. No peaks are present in the modeled spectra, as these are based
on a homogeneous spectrum of frequencies of the inflow turbulence. The spectral peaks roughly seem to be harmonics of a
fundamental frequency at about 750 Hz, which could be the Strouhal frequency of some characteristic length of the blade-
strut attachment construction. Assuming a Strouhal number of 0.2, the corresponding length scale would be about 7 mm.
(Within the present Reynolds number regime, the Strouhal number 0.2 is appropriate for a cylinder [42]; we use this
number for simplicity to obtain an indication of the length scale.) In any case, as the blade passes through turbulence that
was created within the blade-strut system just a single second ago, the turbulence spectrum is likely to be non-
homogeneous.
 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166 165

Table 3
Sound-pressure level and sound-power level as derived from the microphone-array measurements for the T1 turbine.

Position SPL (dBA) SWL (dBA)

1 33.4 79.2
2 38.3 84.1
3 38.4 84.2
4 35.0 80.8

4.3. Emitted sound-power level

In Table 3, the overall A-weighted sound-pressure levels derived from integrating the source maps are shown, along with
sound-power levels calculated from Eq. (6). Positions 2 and 3 gave the highest levels, which can be expected, as the wind speeds
were higher for those recordings, combined with favorable directivity. The sound-power levels for these positions are about
84 dBA for wind speeds in the range 4.2–4.7 m/s (at 10 m height). This is significantly lower than 93.1 dBA for 5 m/s that was
measured for the T1 turbine in [20], where the standard method for noise-emission measurements was used. The microphone-
array measurements performed here produce source maps at frequencies down to 500 Hz, as lower frequencies are not accu-
rately resolved; therefore, being based on source-map integration, the levels in Table 3 lack a certain contribution to the total
noise level. To resolve lower frequencies, the size of the microphone array would have to be increased.

4.4. Implications

The order-of-magnitude check performed above seems to indicate that the noise produced by the T1 turbine might be
due to turbulence that is created by the turbine itself, which in turn degrades the power-conversion performance of the
turbine. Therefore, mitigating the turbulence production in the blade-strut system is likely to both enhance the power factor
and lower the overall noise level. The creation of turbulence at the strut attachment point is supported by the measure-
ments in [11]. Furthermore, power loss due to struts is discussed in [39], a retrospective of the development of the full-
Darrieus VAWT at Sandia National Laboratories during the 1970s and 1980s. Each blade of their test-bed VAWT consisted of
five pieces with different chords joined together, as it was not possible to manufacture a blade with a continuously varying
chord. By fairing the joints, the maximum power factor was reported to increase from 0.41 to 0.43. Additionally, adding
struts to the full Darrieus, which was needed to maintain structural integrity for some designs, was reported to degrade
performance. Presence of struts and joints can, according to [39], easily reduce the power output by 20%. The turbine
performance seems to be very sensitive to adding extra structure on top of the ideal airfoil shape.
Based on these considerations, it would be very interesting to study the effect of changing the smoothness of the blade-
strut joints of the T1 turbine, regarding both the power factor and noise emission. It appears likely that smoothing the joints
would be the best way to reduce the total noise level produced by an H-rotor VAWT of this size.

5. Conclusions

Mapping the noise sources of the T1 turbine with a microphone array shows that the dominant part of the noise ori-
ginates from a blade azimuth angle of about 88° with respect to the wind direction, which is close to the position where the
local flow velocity is maximal. A comparison between the source maps produced in this study and images found in [11],
where the near-wake flow pattern was mapped for a small H-rotor VAWT, suggests that the noise might be a result of
turbulent inflow on the blade leading edge, the turbulence being created during the previous blade-strut-system passage,
probably close to the blade-strut joints. A higher-frequency lower-intensity part in the spectrum is clearly localized to the
blade-strut joints. This source is extended over a wider range of azimuth angles. The presence of this higher-frequency part
strengthens the hypothesis that a substantial amount of turbulence is created precisely at the blade-strut joint.
An order-of-magnitude calculation, using a semi-empirical model for inflow-turbulence noise, indicates that the power
loss that can be expected due to the presence of the struts could be responsible for the observed noise-level spectra. The
measured spectra, however, contains spectral peaks not present in the calculation, due to the assumption of an inflow of
homogeneous turbulence. The observed spectra seem to indicate harmonics with the fundamental frequency of 750 Hz,
which could be the Strouhal frequency of some typical length scale of the joint.
A practical conclusion of this work is that the turbine efficiency and noise-emission levels are likely to benefit from
smoothing the blade-strut joints.

Acknowledgments

Stiftelsen Olle Engkvist Byggmästare is acknowledged for financing this project. This work was conducted within the
STandUP for ENERGY strategic research framework.
166 F. Ottermo et al. / Journal of Sound and Vibration 400 (2017) 154–166

References

[1] K. Marvel, B. Kravitz, K. Caldeira, Geophysical limits to global wind power, Nat. Clim. Change 3 (2) (2013) 118–121.
[2] S. Eriksson, H. Bernhoff, M. Leijon, Evaluation of different turbine concepts for wind power, Renew. Sustain. Energy Rev. 12 (5) (2008) 1419–1434.
[3] D. Bowdler, H. Leventhall, Wind Turbine Noise, Multi-Science Pub, 2011.
[4] S. Oerlemans, P. Sijtsma, B. Méndez López, Location and quantification of noise sources on a wind turbine, J. Sound Vib. 299 (4–5) (2007) 869–883.
[5] G. van den Berg, Effects of the wind profile at night on wind turbine sound, J. Sound Vib. 277 (4–5) (2004) 955–970.
[6] K. Bolin, Wind turbine noise and natural sounds: masking, propagation and modeling (Ph.D. thesis), KTH, Marcus Wallenberg Laboratory MWL, 2009.
[7] E. Pedersen, Human response to wind turbine noise: Perception, annoyance and moderating factors (Ph.D. thesis), University of Gothenburg, 2007.
[8] S. Buck, S. Oerlemans, S. Palo, Experimental validation of a wind turbine turbulent inflow noise prediction code, in: Proceedings of the 22nd AIAA/
CEAS Aeroacoustics Conference, 2016, p. 2953.
[9] E. Dyachuk, A. Goude, Simulating dynamic stall effects for vertical axis wind turbines applying a double multiple streamtube model, Energies 8 (2)
(2015) 1353–1372.
[10] C. Simão Ferreira, The near wake of the vawt: 2d and 3d views of the vawt aerodynamics (Ph.D. thesis), TU Delft, Delft University of Technology, 2009.
[11] G. Tescione, D. Ragni, C. He, C.S. Ferreira, G. van Bussel, Near wake flow analysis of a vertical axis wind turbine by stereoscopic particle image
velocimetry, Renew. Energy 70 (2014) 47–61.
[12] S. Shamsoddin, F. Porté-Agel, Large eddy simulation of vertical axis wind turbine wakes, Energies 7 (2) (2014) 890–912.
[13] C. Pearson, Vertical axis wind turbine acoustics (Ph.D. thesis), Cambridge University, 2014.
[14] A. Iida, A. Mizuno, K. Fukudome, Numerical simulation of aerodynamic noise radiated form vertical axis wind turbines, in: Proceedings of the 18
International Congress on Acoustics, 2004.
[15] H. Dumitrescu, V. Cardos, A. Dumitrache, F. Frunzulica, Low-frequency noise prediction of vertical axis wind turbines, Proc. Romanian Acad. 11 (1)
(2010) 47–54.
[16] M. Mohamed, Aero-acoustics noise evaluation of H-rotor Darrieus wind turbines, Energy 65 (2014) 596–604.
[17] J. Weber, S. Becker, C. Scheit, J. Grabinger, M. Kaltenbacher, Aeroacoustics of Darrieus wind turbine, Int. J. Aeroacoust. 14 (5–6) (2015) 883–902.
[18] M. Ghasemian, A. Nejat, Aero-acoustics prediction of a vertical axis wind turbine using large eddy simulation and acoustic analogy, Energy 88 (2015)
711–717.
[19] R. Williams, J. Rocha, E. Matida, F. Nitzsche, Assessment of surface-based aeroacoustic noise from blades of a vertical-axis wind turbine, in: Pro-
ceedings of ASME 2014 International Mechanical Engineering Congress and Exposition (2014) Paper No.: V013T16A003.
[20] E. Möllerström, F. Ottermo, J. Hylander, H. Bernhoff, Noise emission of a 200 kW vertical axis wind turbine, Energies 9 (1) (2016) 19.
[21] T.F. Brooks, D.S. Pope, M.A. Marcolini, Airfoil self-noise and prediction, NASA Reference Publication, Hampton, VA 1218 (1989).
[22] P. Moriarty, P.G. Migliore, Semi-empirical aeroacoustic noise prediction code for wind turbines, NREL/TP-500-34478, National Renewable Energy
Laboratory, Golden, CO, 2003.
[23] S. Oerlemans, J. Schepers, Prediction of wind turbine noise and validation against experiment, Int. J. Aeroacoust. 8 (6) (2009) 555–584.
[24] I. Paraschivoiu, Wind turbine design: with emphasis on Darrieus concept, Presses inter Polytechnique, 2002.
[25] M.V. Lowson, Assessment and prediction of wind turbine noise, Flow Solutions Ltd., Rep 92/19, 1993.
[26] R. Amiet, Acoustic radiation from an airfoil in a turbulent stream, J. Sound Vib. 41 (4) (1975) 407–420.
[27] R.W. Paterson, R.K. Amiet, Acoustic radiation and surface pressure characteristics of an airfoil due to incident turbulence, NASA Contractor Report
2733, 1976.
[28] A. Dowling, J. Ffowcs Williams, Sound and Sources of Sound, Ellis Horwood Limited, Chichester, UK, 1983.
[29] H.K. Versteeg, W. Malalasekera, An introduction to computational fluid dynamics: the finite volume method, Pearson Education, 2007.
[30] International Electrotechnical Commission, IEC 61400-11: Wind turbine generator systems-part 11: Acoustic noise measurement techniques, Edition
3, IEC, Switzerland, 2012.
[31] D.H. Johnson, D.E. Dudgeon, Array signal processing: concepts and techniques, Prentice-Hall, Englewood Cliffs, NJ, 1993.
[32] P. Sijtsma, Acoustic beamforming for the ranking of aircraft noise, National Aerospace Laboratory Report No. NLR-TP-2012-137.
[33] T.F. Brooks, W.M. Humphreys, A deconvolution approach for the mapping of acoustic sources (damas) determined from phased microphone arrays, J.
Sound Vib. 294 (4) (2006) 856–879.
[34] S. Eriksson, J. Kjellin, H. Bernhoff, Tip speed ratio control of a 200 kW VAWT with synchronous generator and variable DC voltage, Energy Sci. Eng. 1
(3) (2013) 135–143.
[35] J. Kjellin, S. Eriksson, H. Bernhoff, Electric control substituting pitch control for large wind turbines, J. Wind Energy 2013 (2013) 4. Article ID 342061.
[36] Z. Prime, C. Doolan, A comparison of popular beamforming arrays, in: Proceedings of ACOUSTICS 2013, Victor Harbor.
[37] E. Sarradj, G. Herold, Acoular – Open-Source-Software zur Anwendung von Mikrofonarrayverfahren [Acoular – Open-source software intended for
microphone-array arrangements], DAGA 2016 Aachen, 2016.
[38] Acoular - Acoustic testing and source mapping software, 〈http://www.acoular.org/〉, (accessed 15 August 2016).
[39] H.J. Sutherland, D.E. Berg, T.D. Ashwill, A retrospective of VAWT technology, Sandia Report No. SAND2012-0304, 2012.
[40] W. Tjiu, T. Marnoto, S. Mat, M.H. Ruslan, K. Sopian, Darrieus vertical axis wind turbine for power generation I: assessment of Darrieus vawt con-
figurations, Renew. Energy 75 (2015) 50–67.
[41] E. Möllerström, F. Ottermo, A. Goude, S. Eriksson, J. Hylander, H. Bernhoff, Turbulence influence on wind energy extraction for a medium size vertical
axis wind turbine, Wind Energy 19 (11) (2016) 1963–1973.
[42] E. Achenbach, E. Heinecke, On vortex shedding from smooth and rough cylinders in the range of Reynolds numbers 6 × 103 to 5 × 106 , J. Fluid Mech.
109 (1981) 239–251.