Proceedings of Acoustics 2012 - Fremantle 21-23 November 2012, Fremantle, Australia

Prediction of noise from a wing-in-junction flow using

computational fluid dynamics

Con J. Doolan, Jesse L. Coombs, Danielle J. Moreau, Anthony C. Zander and

Laura A. Brooks
School of Mechanical Engineering, University of Adelaide, Australia

ABSTRACT
The leading edge turbulence interaction noise model of Amiet was extended to incorporate span-wise variations in
flow properties and integration with modern computational fluid dynamics codes. The present implementation of the
leading edge noise model was validated against experimental data in the literature. To demonstrate the use of the ex-
tended leading edge noise model, the flow and noise from a wing-in-junction test case was simulated numerically.
Noise was calculated using flow data from different upstream positions to illustrate the importance of choosing the
most appropriate turbulence data for noise prediction. The effect of span-wise discretisation on the acoustic predic-
tion was shown and a study of the noise contributions from each span-wise part of the wing was performed. This
showed that the upper part of the wing produced the most noise. Thus, any noise mitigation strategies should be con-
centrated in this area for maximum effect.

flows may vary considerably across the radius of a wind tur-

INTRODUCTION bine blade as it passes through different regions of the atmos-
pheric boundary layer. It is therefore necessary to develop
The interaction of turbulent flow with an airfoil creates un- new prediction methodologies that can incorporate span-wise
steady lift, which is a source of broadband noise. This type variations in flow properties. Moreover, it is advantageous to
of noise, known as turbulence leading edge interaction noise, combine such a technique with modern computational fluid
is important for many applications, such as wind turbine ro- dynamics (CFD) that can predict the mean flow properties for
tors encountering turbulent gusts, helicopter rotors interacting use as an input to the noise model.
with turbulence from preceding blade wakes and gas turbine
compressor blades that pass through areas of turbulent flow
either from the atmosphere or from wakes created by other
rotors or stators in compressor cascades. It is therefore im-
portant that the physics controlling this type of noise is
understood completely and methods are available for engi-
neers to accurately predict it.

Figure 1 illustrates the essential components of turbulence Figure 1. Side view of an airfoil encountering a turbulent
leading edge interaction noise (known as leading edge noise flow field. When the turbulent eddies interact with the airfoil
hereafter). Turbulent eddies with characteristic length scale leading edge, unsteady lift is produced, which is a source of
L (created by atmospheric shear, a wall boundary layer or by noise.
other components upstream of the airfoil) are convected to-
ward an airfoil by the flow. When the eddies reach the lead- There are several methods of numerically simulating turbu-
ing edge, the random velocity fluctuations in the flow induce lent flow about an airfoil. However, the only practical
an unsteady pressure over the surface of the airfoil. The un- method that can be used for engineering design is the solution
steady pressure distribution creates unsteady lift that, by the of the Reynolds-averaged Navier-Stokes (RANS) equations.
theory of Curle (1955), creates noise. This is a different This is because other methods such as Large Eddy Simula-
mechanism to that found at the trailing edge (Moreau et al., tion (LES) or Direct Numerical Simulation (DNS) require
2011), where turbulent eddies created in the boundary layer high computational resources and solution times, making
produce noise via an edge diffraction process. In this paper, them impractical for use in situations where multiple design
only leading-edge interaction noise is considered. solutions must be evaluated in a realistic time. For leading
edge noise considerations, the use of RANS in the design
Most semi-analytical techniques that are used for the predic- process is to quantify and influence the flow properties up-
tion of leading-edge noise are based on the theory of Amiet stream of an airfoil. For example, RANS can be used to cal-
(1975). This is a very useful methodology and has been used culate the flow properties in an air-conditioning duct in order
by many researchers (Paterson & Amiet, 1982; Roger & to evaluate the noise created by a fan located within it.
Moreau, 2004); however, it was derived for cases where
homogeneous turbulence encounters an airfoil or wing so the The aim of this paper is to present a technique that links
turbulence properties do not vary across the span. In many RANS flow solutions with a semi-analytical leading-edge
situations, this is not the case. For example, turbulent in- noise prediction method. It is complementary to other meth-

Australian Acoustical Society Paper Peer Reviewed 1

21-23 November 2012, Fremantle, Australia Proceedings of Acoustics 2012 - Fremantle

ods currently under development at the University of Adel- fluctuating lift to noise. The response function has solutions
aide for other airfoil self noise mechanisms, such as trailing for low and high frequencies, as detailed by Amiet (1975).
edge noise (Doolan et al., 2010). Additionally, this paper Mωb π
extends the leading-edge noise model so that it is able to The low frequency solution is valid when < and
Uβ
2
accommodate span-wise variations in turbulent flow proper- 4
ties the high frequency solution is valid for all other frequencies.
The response function itself is lengthy to document, hence
The paper is structured as follows. The leading edge noise the reader is referred to Amiet (1975) for complete details.
model is presented, including a description of how it can be
used with a RANS flow solution and how to take into ac- Using RANS data as an input
count span-wise variations in the flow. The noise model is
then validated against published leading edge noise data for When using a RANS simulation to provide turbulent inflow
an airfoil placed in homogeneous turbulence to show that the data for noise calculations, a method is needed to provide
described implementation of the model is accurate. Finally, turbulence intensity and length scale to the noise model from
the entire model is used on the complex flow demonstration the computed flow field.
test case of a wing-in-junction flow. Here, a RANS flow
solution is presented and used to predict the noise in the far- Turbulence intensity (TI) can be calculated using (Wilcox,
field. 2006)

LEADING EDGE NOISE MODEL 2 k

TI = (3)
Theory of Amiet 3U
2

When unsteady flow (such as turbulence) encounters the

where k is the turbulent kinetic energy of the flow provided
leading edge of an airfoil, the airfoil experiences a fluctuating
lift response, a side effect of which is a small component of by the RANS simulation and U is the mean local velocity.
the flow energy is radiated as sound to the far-field. Amiet
The turbulent integral length scale L can be determined using
(1975) has derived a model describing this sound generation
(Wilcox, 2006)
process. This theory is based on a derivation of the cross-
power-spectral density of surface pressure on the airfoil sur- 3/2
face to far-field sound that uses the techniques of Curle * k
(1955). Following this derivation gives the power spectral
L=C (4)

density for far-field noise, G pp , generated by the interaction
of turbulence with the leading edge of an airfoil. where  is the turbulent dissipation rate provided by the
RANS simulation. The constant C* is needed to relate the
turbulence parameters to the length scale. Traditionally, a
value of C*=0.09 is used for this purpose (Wilcox, 2006).
⎛ z ρ U ⎞ ⎛ ω b ⎞ | L |2 φ (ω )l (ω )
2 2
However, when this value is used with the RANS flow data,
G pp (x, ω ) = 2d a 02 0 ⎜
⎝ σ ⎠ ⎝ c0 ⎟⎠ ww y the turbulent length scale is much too small (much smaller
than the boundary layer height), which is not realistic as it is
(1) expected that the maximum turbulent length scale should be
similar to the boundary layer height (Pope, 2000). The under
prediction of the length scale can be traced to an over predic-
where x = (x , 0, z ) is the observer position vector with
a a tion of turbulent dissipation by the RANS flow model. Based
respect to an origin at the mid-chord and mid-span of the on a careful comparison of experimental flat plate boundary
airfoil and with xa the chord-wise distance, ya the span-wise layer measurements and numerical simulation (a separate and
distance and za the perpendicular distance (relative to the as yet unpublished study), a value of C*=10 was found to
airfoil) to the observer. Other symbols are defined as fol- give the most realistic estimates of turbulent length scale and
lows: d is the half-span, b is the half-chord, ρ0 is the ambient this will be used for the demonstration of the noise prediction
density of the fluid, U 0 is the mean free stream flow velocity, method in this paper. Development of accurate turbulent
length scale estimation methods from RANS flow data is an
ω =2π f is the angular frequency (f is the frequency in Hz) on-going area of research in this project.
and c0 is the speed of sound of the ambient fluid. The sym-
Discretisation of the airfoil for noise calculations
bols φww (ω ) and l y (ω ) relate to the vertical velocity
The incoming flow will be non-uniform in many situations
turbulence spectrum and span-wise correlation length scale
encountered in engineering design, such as wind turbines,
respectively. In this work, the Karman spectrum is used to
propellors, ducted fans and for the present test case of a
define these quantities, full details of which are documented
wing-in-junction flow. Figure 2 shows an idealised general
in Amiet (1975). The symbol σ is the far-field corrected
example of span-wise varying turbulent flow with mean local
distance, defined as velocity U encountering an airfoil. Figure 2 also shows a
non-uniform turbulent length scale (L) across the span and a
σ = xa + β (ya + za ) similar graph could be drawn showing a varying turbulent
2 2 2 2
(2)
kinetic energy or turbulence intensity.

with the compressibility term defined as β = 1− M ,

2 The theory of Amiet was developed for uniform turbulent
inflow and must be modified to take into account span-wise
where M is the Mach number of the flow (M = U0/c0). The varying turbulent conditions. The method proposed here is to
symbol L represents an airfoil response function that relates discretise the span into many small segments of width dy, as

2 Australian Acoustical Society

Proceedings of Acoustics 2012 - Fremantle 21-23 November 2012, Fremantle, Australia

shown in Figure 2, and assume that the flow conditions are 70

uniform across each strip. The noise from the airfoil is then a
65
sum of contributions from each strip. If the airfoil is discre-
tised into N strips, dy = 2d/N, and the total noise is 60

55

Spectrum Level (dB)

50

2d ⎛ za ,i ρ 0U 0 ⎞ ⎛ ωb ⎞
N
2 2 U = 60 m/s

∑ N ⎜⎝
45
G pp ,N = ⎟⎠ ⎜⎝ c ⎟⎠ | L | φww ,i (ω )l y ,i (ω )
2

σi
2
i=1
40
0

35
(5)
30

25

20
500 1000 1500 2000 2500 3000
Frequency (Hz)
Figure 3. Comparison of the theory of Amiet with experi-
mental results (Paterson & Amiet, 1977). The data shown are
a 1 Hz bandwidth power spectrum, referenced to 20 µPa.

WING-IN-JUNCTION TEST CASE

Numerical Details
Figure 2. Airfoil encountering a turbulent flow with mean To demonstrate the leading-edge noise prediction method on
local velocity U that has varying properties across its span. a complex flow case, a wing-in-junction flow was simulated
The figure depicts the variation of turbulent integral length numerically. The flow geometry is shown in Figure 4. The
scale (L) across the airfoil’s span. A strip of width dy is airfoil (wing) shown in the figure has a NACA 0012 profile
shown and this is used as part of the noise prediction method. and was set at zero angle of attack. The wing has an aspect
ratio of unity and a chord of c = 69 mm. A right handed co-
Validation of Amiet’s Theory
ordinate system (for the flow results) was defined with its
The experimental results of Paterson and Amiet (1977) are origin at the intersection of the airfoil leading edge and flat
used to validate the present implementation of Amiet’s lead- plate, with x the stream-wise, y the span-wise, and z the
ing edge noise model, for the case of uniform span-wise tur- crossflow directions respectively. The Reynolds number of
bulent flow. Here, a NACA 0012 airfoil with 0.23 m chord the flow (based on chord length 69 mm and free steam ve-
and 0.53 m span was placed in a homogeneous turbulent locity 38 m/s) was 1.75 × 105.
stream at zero angle of attack. The test case selected for
comparison had a mean flow speed of 60 m/s and a turbu- The flow was treated as incompressible and solved using the
lence intensity of 3.9%. The integral turbulence length scale OpenFOAM code (Weller et al., 1998) using the Semi-
L was 30 mm. Noise was measured using a microphone 2.25 Implicit Method for Pressure-Linked Equations algorithm.
m directly overhead the mid-chord of the airfoil. Full details The RANS equations were solved using the realisable k-ε
of the tests can be found in Paterson and Amiet (1977). turbulence model for closure. The inlet boundary condition of
the domain was set to be identical to that of a turbulent boun-
For validation against theory, the high frequency airfoil re- dary layer with the same height as the airfoil chord. A no-
sponse function was used. The Karman turbulence model was slip boundary condition was applied to the floor and airfoil.
also used in the calculation. The results are compared in The sides of the domain had zero-normal-gradient boundary
Figure 3 where theory predicts the experimental spectrum conditions applied to them, while the roof was an inlet type
reasonably well. Above approximately 1500 Hz, the experi- boundary with a constant velocity equal to the free stream (38
mental results do not match the model. This is because (as m/s).
discussed in Paterson and Amiet (1977)) the signal-to-noise
ratio at these frequencies was poor. Linear interpolation schemes were used throughout, as was a
second-order accurate linear scheme for the discretisation of
gradient terms. The velocity field divergence terms were
discretised using an interpolation scheme in which traditional
linear-upwind and linear interpolation schemes are blended to
stabilise solutions while maintaining second-order behaviour.
The turbulent kinetic energy and dissipation fields divergence
terms were discretised using a first-order accurate upwind
scheme. All Laplacian terms were discretised with the sec-
ond-order accurate linear scheme with explicit non-
orthogonal correction. Finally, explicit non-orthogonal cor-
rection was performed when calculating surface-normal
gradient terms.

Australian Acoustical Society 3

21-23 November 2012, Fremantle, Australia Proceedings of Acoustics 2012 - Fremantle

Figure 6. Contours of normalised velocity magnitude

Figure 4. Computational domain. Flow is from lower on the x-y plane at z = 0.
left (Inlet) to upper right (Outlet). Units on axes are Figure 6 shows contours of computed normalised velocity
normalised by airfoil thickness T. magnitude on the x-y plane at y = 0. The computational data
shows the flow impacting upon the leading edge, accelerating
A grid refinement study was performed where the number of around the tip and expanding into the wake. Non-
cells were increased until the flow solution became invariant dimensionalised turbulent kinetic energy about the wing is
and the drag coefficient changed by less than 1.05%. The shown in Figure 7. The turbulent kinetic energy is non-
final mesh used in this study contained 12,058,082 cells. The uniform upstream of the leading edge. Similarly, turbulent
computations were performed on a 24-core workstation using dissipation (Figure 8) is non-uniform upstream of the leading
the Linux operating system. The flow simulation required edge of the wing. Therefore, this variation in mean proper-
approximately one week of (wall-clock) time to produce a ties upstream of the leading edge must be taken into account
converged result. when computing noise.
Flow Results

Figure 7. Contours of normalised turbulent kinetic

energy on the x-y plane at z = 0.

Figure 5. Iso-contours of the second invariant of the

velocity gradient tensor Q about the wing-in-junction
flow, coloured by velocity magnitude. Flow is from
left to right.
The flow field about the wing-in-junction test case is summa-
rised in Figure 5. Here and in the flow results that follow, the
reference velocity used for non-dimensionalisation was set to
Uref = 38 m/s. In Figure 5, iso-contours of the flow parameter
Q (the second invariant of the velocity gradient tensor, which
is a scalar measure of the flow strain rate) show the main
flow structures about the wing. At the junction of the wing
and floor, a vortex forms around the leading edge of the
wing. There is a trace of a trailing vortex structure along the Figure 8. Contours of normalised turbulent dissipation
chord at the base of the wing while large, well-defined struc- on the x-y plane at z = 0.
tures are formed along the floor in the wake. Well-defined
vortex structures form over the tip of wing and into the wake, Amiet’s theory of leading-edge noise, used as a basis for the
which evolve and merge with additional trailing vortices to noise methodology presented here, is based on flow turbu-
form a complex wake from the tip region. lence properties measured in the absence of an airfoil. That
is, turbulence intensity and length scale are measured in a
wind tunnel prior to the installation of a model. However,

4 Australian Acoustical Society

Proceedings of Acoustics 2012 - Fremantle 21-23 November 2012, Fremantle, Australia

the effect of solid surfaces will alter the turbulence intensity Acoustic Results
and dissipation in their close vicinity and care must be taken
when collecting flow data upstream of the leading edge for Before the acoustic calculations can be performed, an under-
noise calculations. In order to investigate this effect, data standing of the effect of span-wise discretisation must be
were collected at two locations upstream of the leading edge. obtained. The effect of the number of strips (N) on the root-
These data are summarised in Figure 9 and will be discussed mean-square (rms) of the predicted acoustic signal was inves-
in the following paragraphs. tigated by calculating the power spectral density of the acous-
tic pressure from the wing-in-junction flow using Equation
Data were collected along a line that extended from the floor (5). In each case, the observer or virtual microphone location
to a position equal to the span of the airfoil at two stream- was placed at (xa ya za)= (0 0 600) mm directly over the
wise locations upstream of the leading edge. One location centre of the wing surface (90 degrees to the chord line). For
was chosen to be very close to the leading edge (x/c = various values of N, the predicted spectrum was integrated
-0.0032) where the flow is strongly influenced by the leading over the frequency range of 100 Hz to 20 kHz to obtain an
edge. The second location was chosen to be relatively far overall unweighted sound pressure level. The values of N
from the leading edge (x/c = -0.203) where the flow was un- used for this study were N = [1 5 50 100 500]. A value of N
affected by the wing and was similar to the flow entering the = 1 means that the span-wise varying flow upstream of the
computational domain. leading edge was averaged over the span. Similarly, N > 1
indicates that N equispaced strips were created, and the flow
Figure 9(a) shows the mean stream-wise velocity profiles at properties upstream of the leading edge were averaged over
the two stream-wise locations. The effect of the wing is ob- each strip and used in the acoustic calculation.
vious in this case. At the upstream location, the velocity
profile resembles that of the inlet profile, which is a turbulent Figure 10 displays the results of the investigation of the effect
boundary layer. Close to the leading edge, the flow has of number of strips on acoustic prediction. The figure shows
slowed considerably and is no longer similar to that of a the percentage change in solution (the unweighted overall
boundary layer. sound pressure level) versus number of strips using the nu-
merical flow data taken at the upstream location (x/c =
In Figure 9(b) turbulent kinetic energy profiles are shown at -0.203). The percentage change in solution was calculated by
the two stream-wise locations. In this case, the profiles are dividing change in the integrated pressure level from one
similar in shape, but the magnitude of the data close to the solution to the next as N increased and expressing it as a per-
leading edge is higher than that computed upstream. This is cent. The acoustic solution is quite sensitive to the number of
because the turbulence model detects a wall and increases the strips used when N < 100. However, when N > 100, the solu-
production of turbulence to form a boundary layer. Even tion appears to be unaffected, indicating the numerical
though the mean flow velocity is low, the turbulent kinetic method has become grid insensitive and converged. Thus,
energy is high, thus the overall effect of these competing the acoustic calculations presented here will use a value of N
parameters on noise generation is unclear without the use of = 100, which corresponds to a strip width of dy = 0.69 mm
an acoustic model. It is also interesting to note the produc- for the wing-in-junction test case.
tion of what is termed here as a “secondary” shear layer
along the floor of the domain upstream of the boundary layer. Predicted leading edge noise spectra are shown in Figure 11.
This is clearly identified in the computational data as a peak These spectra were obtained using Equation (5) using the
in the turbulent kinetic energy data over 0 ≤ y/c ≤ 0.1. This data presented in Figure 9 and at both stream-wise locations,
region of higher turbulence intensity will produce more noise i.e. those labelled “Upstream” (x/c = -0.203) and “Leading
than would otherwise be expected from a flow without the edge” (x/c = -0.0032). The results clearly show the import-
secondary shear layer. ance of choosing the location to sample the in-coming flow
correctly. The spectrum calculated using the leading edge
Figure 9(c) compares the turbulent dissipation at the leading (x/c = -0.0032) flow data is approximately 35 dB
edge and upstream from it. Similar to the turbulent kinetic below the spectrum calculated using the upstream flow data
energy results of Figure 9(b), dissipation is higher closer to (x/c = -0.203). The reason for this discrepancy can be traced
the leading edge due to the high production of turbulence to differences in the mean flow data, specifically the differ-
induced by the presence of the wall. The secondary shear ences in turbulent integral length scale (Figure 9(e)) and
layer is also noticeable in the dissipation results. mean velocity (Figure 9(a)).

The turbulent kinetic energy profiles of Figure 9(b) were

converted to turbulence intensity profiles using Equation (3),
for use with the leading-edge noise model (Equation (5)).
The results of this conversion are shown in Figure 9(d). The
turbulence intensity profiles are identical in shape to the tur-
bulent kinetic energy. The turbulence intensity is higher at
the floor and varies in a non-uniform manner to the tip.
Close to the leading edge, the effect of the wing tip is noticed
in the results, where turbulence intensity increases as y/c
approaches unity.

Turbulent integral length scale profiles were computed from

the computational data and are shown in Figure 9(e). There
are considerable differences between the two stream-wise
locations and each will have a significant effect on the noise
calculation.

Australian Acoustical Society 5

21-23 November 2012, Fremantle, Australia Proceedings of Acoustics 2012 - Fremantle

1 1
Upstream Upstream
Leading edge Leading edge
0.8 0.8

0.6 0.6

y/c
y/c

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.002 0.004 0.006 0.008 0.01 0.012
2
U/U k/U0
0

(a) Mean velocity (b) Turbulent kinetic energy

1 1
Upstream Upstream
Leading edge Leading edge
0.8 0.8

0.6 0.6
y/c

y/c

0.4 0.4

0.2 0.2

0
10
4
10
3
10
2
10
1
10
0
10
1 0
0 1 2 3 4 5 6 7 8 9
3
c/U0 TI (%)

(c) Dissipation (d) Turbulence intensity

1
Upstream
Leading edge
0.8

0.6
y/c

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
L/c

(e) Integral length scale

Figure 9. Flow property profiles measured upstream of the leading edge of the wing. Solid line (labelled “Upstream”)
indicates a position x/c = -0.203 upstream of the leading edge, at a position where the flow variables are unaffected by
the presence of the wing; dashed line (labelled “Leading edge”) indicates a position x/c = -0.0032 upstream of the lead-
ing edge.

6 Australian Acoustical Society

Proceedings of Acoustics 2012 - Fremantle 21-23 November 2012, Fremantle, Australia

20 40
0<y/c<0.1
0.1<y/c<0.2
0.2<y/c<0.3
30 0.3<y/c<0.4
15 0.4<y/c<0.5
0.5<y/c<0.6
0.6<y/c<0.7
% change in solution

20 0.7<y/c<0.8
0.8<y/c<0.9
0.9<y/c<1.0
10 Total

PSD (dB/Hz)
10

5 0

10
0

20

5
0 100 200 300 400 500 30 2
N 10 10
3
10
4

Frequency (Hz)

Figure 10. Effect of the number of strips (N) on the

acoustic solution expressed as a percentage change in Figure 12. Prediction of acoustic power spectral den-
overall unweighted sound pressure level. sity (PSD) for the wing-in-junction test case (Up-
stream) showing the contribution from sections of the
40
Upstream span each of width equal to 10% of the chord.
Leading edge
30
Figure 12 shows the predicted acoustic power spectral den-
20 sity along with contributions from various sections of the
span in order to link the flow physics to noise generation
10 more directly. The span was divided into equal widths of
PSD (dB/Hz)

10% chord and acoustic power spectral density calculated for

0 each. These data are shown in Figure 12 and are compared
with the acoustic power spectral density for the entire wing
10 (labelled “Total”). Most of the noise is being produced from
the upper part of the wing, where y/c > 0.5. In this region of
20 the flow field, the mean velocity is high and the turbulence
length scales are large. As discussed earlier in this section,
30
the radiated sound is very sensitive to these parameters and
explains why the upper part of the wing is responsible for
40 2
10 10
3
10
4
most of the noise generation. Thus, any noise control strat-
Frequency (Hz) egies for these types of flow should take this into consider-
ation. When a boundary layer on a surface encounters an
Figure 11. Comparison of acoustic power spectral den- object connected to it, it is the flow in the upper region of the
sities (PSD) using numerical flow data from two up- boundary layer that is responsible for most of the noise pro-
stream locations. Solid line (labelled “Upstream”) in- duction. In the present case where the boundary layer is of
dicates a position x/c = -0.203; Dashed line (labelled the same height as the span, any noise mitigation strategies
“Leading edge”) indicates a position x/c = -0.0032. should involve either shape modifications or turbulence con-
trol in the upper region of the wing.
Observing Equation (5), the sound pressure level ( G pp ) is
SUMMARY AND CONCLUSIONS
proportional to U2 and the product of φ (ω ) and l y (ω ) ,
ww A leading edge noise model has been extended to include the
the turbulence velocity spectrum and span-wise turbulence effects of span-wise variations in flow properties as well as
correlation length scale, respectively. Both the turbulence integration into RANS-based numerical flow solutions. The
spectrum and correlation length scale are functions of L benefits of these extensions to the original leading edge noise
(Amiet, 1975), resulting in acoustic pressure having an L2 model are that they can be used for modern engineering de-
dependence. Hence, reductions in U and L will result in a sign that uses CFD methods to calculate the sometimes com-
large decrease in acoustic pressure in the far field. Note that plex, non-uniform turbulent flows encountering airfoils.
turbulence intensity increases near the leading edge (Figure
9(d)), compared with the upstream location. As the turbu- The implementation of Amiet’s leading edge noise model
lence velocity spectrum is only linearly related to turbulence was validated successfully against some experimental data
intensity (Amiet, 1975), the increase due to this effect is from the literature. The use of the extended leading edge
small compared with the reduction due to velocity and turbu- model was demonstrated against a wing-in-junction flow test
lence length scale, hence the overall sound pressure level will case as it contains complex non-uniform flow.
reduce.
Predicted noise spectra were obtained using flow from two
upstream locations to investigate the effect of the wing on the
on-coming flow. As the noise model was originally derived
using the undisturbed turbulence field as an input, it is ex-

Australian Acoustical Society 7

21-23 November 2012, Fremantle, Australia Proceedings of Acoustics 2012 - Fremantle

pected that the noise results obtained using the upstream (x/c
= -0.203) data will be closer to that expected in reality.
However, this may not be the case and careful experimental
measurements are needed to validate these predictions.

An investigation of the contribution of each part of the span

to the far-field noise was performed. The upper 50% of the
wing leading edge contributed most of the noise and this can
be explained in terms of the flow properties. It is concluded
that any noise mitigation strategies for wing-in-junction
flows must concentrate on the upper regions of the boundary
layer where flow velocity is high and integral turbulence
lengths scales are large.

ACKNOWLEDGEMENTS
This work has been supported by the Australian Research
Council under linkage grant LP110100033 ‘Understanding
and predicting submarine hydrofoil noise’.

REFERENCES

Amiet, R. (1975) Acoustic radiation from an airfoil in a tur-

bulent stream. Journal of Sound and Vibration 41, 407-420.

Curle, N. (1955) The influence of solid boundaries on aero-

dynamic sound. Proc. Roy. Soc. London A231, 505-514.

Doolan, C. J., Albarracin Gonzalez, C., & Hansen, C. H.

(2010) Statistical estimation of turbulent trailing edge noise.
Proceedings of the 20th International Congress on Acoustics,
Sydney.

Moreau, D. J., Brooks, L. A., & Doolan, C. J. (2011) Broad-

band trailing edge noise from a sharp-edged strut. The Jour-
nal of the Acoustical Society of America 129, 2820-2829.

Paterson, R. & Amiet, R. (1982) Noise of a model helicopter

rotor due to ingestion of isotropic turbulence. Journal of
Sound and Vibration 85, 551-577.

Paterson, R. & Amiet, R. (1977) Noise and surface pressure

response of an airfoil to incident turbulence. J. Aircraft 14,
729-736.

Pope, S. B. (2000) Turbulent Flows. Cambridge University

Press, UK.

Roger, M. & Moreau, S. (2004) Broadband self-noise from

loaded fan blades. AIAA Journal 42, 536--544.

Weller, H. G., Tabor, G., Jasak, H., & Fureby, C. (1998) A

tensorial approach to CFD using object orientated techniques.
Computers in Physics 12, 620-631.

Wilcox, D. C. (2006) Turbulence Modeling for CFD. DCW

Industries, USA.

8 Australian Acoustical Society