AIAA JOURNAL

Vol. 55, No. 9, September 2017

Statistical–Empirical Modeling of Airfoil Noise Subjected to

Leading-Edge Serrations

Till M. Biedermann∗
Düsseldorf University of Applied Sciences, 40474 Düsseldorf, Germany
Tze Pei Chong†
Brunel University London, Uxbridge, England UB8 3PH, United Kingdom
Frank Kameier‡
Düsseldorf University of Applied Sciences, 40474 Düsseldorf, Germany
and
Christian O. Paschereit§
Berlin Technical University, 10623 Berlin, Germany
DOI: 10.2514/1.J055633
With the objective of reducing the broadband noise from the interaction of highly turbulent flow and airfoil leading
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

edge, sinusoidal leading-edge serrations were investigated as an effective passive treatment. An extensive aeroacoustic
study was performed to determine the main influences and interdependencies of factors, such as the Reynolds
number, turbulence intensity, serration amplitude, and wavelength as well as the angle of attack on the noise-
reduction capability. A statistical–empirical model was developed to predict the overall sound pressure level and noise
reduction of a NACA 65(12)-10 airfoil with and without leading-edge serrations in the range of chord-based Reynolds
numbers of 2.5 × 105 ≤ Re ≤ 6 × 105 . The observed main influencing factors on the noise radiation were quantified in
a systematic order for the first time. Moreover, significant interdependencies of the turbulence intensity and the
serration wavelength, as well as the serration wavelength and the angle of attack, were observed, validated, and
quantified. The statistical–empirical model was validated against an external set of experimental data, which is shown
to be accurate and reliable.

Nomenclature Θ = polar angle, deg

A = amplitude of leading-edge serrations, mm λ = wavelength of leading-edge serrations, mm
AOA = angle of attack (equal to nondimensional vertical ρ0 = fluid density, kg ⋅ m−3
displacement z∕H), deg ω = angular frequency, s−1
C = airfoil chord length, mm
c = sound velocity, m ⋅ s−1
d = maximum airfoil thickness, mm I. Introduction
f = frequency, Hz
H
Ma
=
=
nozzle height, mm
Mach number
R ECENT research has firmly established sinusoidal leading-
edge (LE) serrations as an effective passive treatment to reduce
the broadband noise of an airfoil when exposed to a highly turbulent
OASPL = overall sound pressure level, dB flow. A reduction in the overall sound pressure level of up to
R = observer distance in the far field, m ΔOASPL
7 dB and the sound pressure level reductions ΔSPL >
Re = chord-based Reynolds number 10 dB in the relevant frequency region could be achieved [1–4].
S = airfoil span, mm Although different hypotheses on the noise-reduction mechanism
SPL = sound pressure level, dB were proposed before, they have hitherto not been comprehensively
Tu = turbulence intensity, % verified. In general, three mechanisms could be responsible for the
U0 = freestream velocity, m ⋅ s−1 reduction in the broadband noise. First is the reduced spanwise
x = local streamwise (longitudinal) coordinate, mm correlation coefficients as a result of incoherent response times of the
y = local anti-streamwise (transversal) coordinate, mm incoming turbulence; second, a reduction of the acoustic sources as
z = local vertical coordinate, mm manifested in the reduction in pressure fluctuation at the serration
ΔOASPL = overall sound pressure level reduction, dB peak; and third, a reduction of the streamwise turbulence intensity
ΔSPL = sound pressure level reduction, dB due to the converging flow within the serration gaps [5,6].
Up to now, research on the effect of LE serrations focused either on
the noise-reduction capability or on the aerodynamic performance of
Received 28 August 2016; revision received 11 February 2017; accepted
for publication 28 March 2017; published online Open Access 21 June 2017.
the airfoil itself. The effect of sinusoidal LE on the lift and drag forces
Copyright © 2017 by Till Biedermann and Tze Pei Chong. Published by the has been analyzed experimentally, numerically, and through the use
American Institute of Aeronautics and Astronautics, Inc., with permission. All of flow pattern visualization [7,8]. Moreover, a numerical study to
requests for copying and permission to reprint should be submitted to CCC at optimize the serration design to improve the aerodynamic forces on
www.copyright.com; employ the ISSN 0001-1452 (print) or 1533-385X the airfoil was presented [9]. Of particular importance is the
(online) to initiate your request. See also AIAA Rights and Permissions correlation between the aerodynamic flow behaviors and
www.aiaa.org/randp. aeroacoustic noise-reduction mechanisms. In general, the incoming
*Doctoral Researcher, Institute of Sound and Vibration Engineering; till. turbulence amplifies the surface pressure fluctuations close to the
[email protected]. Student Member AIAA. airfoil LE, which then radiate into broadband noise [10,11].
†
Senior Lecturer, Department of Mechanical, Aerospace and Civil
Engineering; [email protected]. Member AIAA. Recently, there have been many studies using high-fidelity
‡
Professor, Institute of Sound and Vibration Engineering; frank.kameier@ numerical flow simulation to provide a physical insight of the noise-
hs-duesseldorf.de. reduction mechanisms by the serration [12–14]. These studies show
§
Professor, Institute of Fluid Dynamics and Technical Acoustics; oliver. that the surface pressure fluctuation and the far-field noise on a
[email protected]. serrated leading edge are decorrelated by the serrated LEs. In
3128
 BIEDERMANN ET AL. 3129

particular, the noise source at the midregion of the oblique edge number Re, turbulence intensity Tu, serration amplitude A∕C,
becomes ineffective across the mid- to high-frequency range. The serration wavelength λ∕C, and angle of attack (AOA). However, up to
serration could cause a significant decrease in the surface pressure now, these parameters have been investigated independently, and
fluctuations around the tip and midregions of the serration and only little effort was made to analyze them as an interrelated system
subsequently reduce the broadband noise level. Another noise- of factors with respect to the noise reduction. This serves as
reduction mechanism is attributed to the phase interference and motivation for the current work, where a comprehensive statistical–
destruction effect between the serration peak and the midregion of the empirical model has been developed with the aim to describe the
oblique edge. Accordingly, the serration root could still remain noise radiation of serrated LEs as an interrelated system involving the
effective in the noise radiation. Interestingly, a small modification of aforementioned five influencing parameters. Note that the current
the serration root has been found to be able to further reduce the LE model does not predict the acoustic spectral characteristics for
noise level [15]. The converging nature of the serration could also individual airfoil of serrated leading edges. Rather, the target value
generate a nozzle effect to accelerate the flow within and reduce the describing the noise radiation and noise reduction by the LE
level of turbulence intensity before the fluid–structure interaction serrations is defined as the overall sound pressure level (OASPL).
near the stagnation points. Analytical work also begins to emerge that
generalizes Amiet’s theory of leading-edge noise to calculate the II. Experimental Setup
airfoil response function subjected to serrated LEs of different In the current study, a cambered NACA 65(12)-10 airfoil was
serration wavelengths and amplitudes and the far-field noise chosen due to its relevance in the real-life application such as the
radiation [16]. Although the analytical model can predict the acoustic stator vanes or axial fan blades. As shown in Fig. 1, the airfoil has a
power spectral densities that match reasonably well with the chord length of C
150 mm and a span width of S
300 mm. The
experimental results [1], the requirement of the iterative solving airfoil geometry consists a removable frontal part (0 < x∕C < 0.3)
procedure to calculate the gust response function of the appropriate
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

that allows various LE serration profiles to be attached. Once attached

order makes it not straightforward to use. This paper aims to to the rear part main body, the serrations give the appearance that they
generalize the airfoil noise subjected to LE serrations by developing a are cut into the airfoil’s main body. Therefore, the maximum chord is
statistical–empirical model. always held constant at C
150 mm for all the configurations [2].
Several parameters have been found to influence the effectiveness The serration geometries are predominantly defined by their
of noise reduction by LE serrations, which include the Reynolds amplitude (chordwise peak-to-trough value) and wavelength

Fig. 1 NACA 65(12)-10 airfoil showing the main body, the reattachable leading edges, and the definitions and dimensions of the various geometrical
parameters.
 3130 BIEDERMANN ET AL.

Table 1 Coefficients of lift, drag, and lift-to-drag of Tu, but it was not included as a parameter to be investigated in the
ratio at z∕H
0 and Re
250;000 current noise-modeling analysis.
Type A∕C λ∕C CL CD CL ∕CD To determine Tu, a one-dimensional hot-wire probe was placed
30 mm downstream of the nozzle exit, which coincides with the
BSLN — — — — 0.637 0.0342 18.62
airfoil leading-edge tip when installed. Tu was measured without a
A29λ26 0.19 0.17 0.550 0.0336 16.37
A22λ18 0.12 0.12 0.540 0.0338 15.95 mounted airfoil, but with the turbulence grids and side plates
A35λ18 0.23 0.12 0.566 0.0356 15.91 installed. The mean velocity U0 and Tu profiles were recorded at 106
A22λ34 0.12 0.23 0.521 0.0330 15.79 locations over the whole nozzle exit area.
A35λ34 0.23 0.23 0.564 0.0325 17.36 The velocity range of investigation was 10 ≤ U0 ≤ 60 m ⋅ s−1 in
A12λ26 0.08 0.17 0.539 0.0335 16.06 steps of ΔU0
10 m ⋅ s−1 . All measurements were repeated once to
A45λ26 0.30 0.17 0.497 0.0330 15.07 reduce the statistical spread and to reduce the uncertainty. After
A29λ7.5 0.19 0.05 0.478 0.0363 13.16 ensuring a uniform turbulence distribution in the measurement plane,
A29λ45 0.19 0.30 0.586 0.0309 18.97
Tu for the present study was determined as the average value across the
plane. The distance of the airfoil’s leading edge to the nozzle exit
remains the same for all the turbulence grids. The turbulence level near
the airfoil’s LE will be shown to be isotropic. Figure 2 demonstrates
that the measured turbulent energy spectra of the fluctuating velocity
(spanwise peak-to-peak value). Both parameters were normalized by agree well with the turbulence model of von Kármán and Liepmann for
the airfoil chord length throughout the paper. The shape of the LE longitudinal isotropic turbulence as per Eq. (1). The correction
serrations is designed according to a sinusoidal curve, and the NACA function of Rozenberg [20] in Eq. (2) was applied to the turbulence
65(12)-10 profile was extruded along the line of this curve. An model to correct the turbulent energy in the high-frequency region
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

important feature of the current design is the semicyclic shape of the close to the Kolmogorov scale. u 02 is the velocity fluctuation, Λuu is the
serration tips, as depicted in Fig. 1. integral length scale, K x is the streamwise wave number, and Kη is a
The angle of attack α refers to the geometrical angle between the parameter that controls the slope of the high-frequency rolloff. The
horizontal axis and the chord line of the NACA 65(12)-10 airfoil. As range of chord-based Reynolds number investigated in this study is
shown in Fig. 1, the nondimensional ratio between the vertical LE tip 2.5 × 105 ≤ Re ≤ 6 × 105 . The lower limit of the Reynolds number
displacement (z) and the height of the nozzle outlet (H) is also was determined by the minimum freestream velocity where the
equivalent to α. As shown in Table 1, an experimental investigation isotropic condition of Tu can still be established:
into the aerodynamic performance of the airfoil with a baseline LE in
a closed wind tunnel produces a lift coefficient CL
0.64 at
u 02 Λuu 1
α
0 deg. The airfoil was attached to the side plates extending from ΦLuu ω
⋅ (1)
both sides of the nozzle outlet. πU0 1  K2x Λ2uu
A pivot-mounted insert of the side plates facilitates accurate
rotation of the airfoil between α
10 deg. Because of the


2 
−9 Kx
relatively low value of H in the current nozzle, measurement beyond GKolm:
exp ⋅ (2)
α
10 deg was not attempted. Note that no correction of the free 4 Kη
jet deflection was applied in the current study. Therefore, it is more
effective to use the nondimensional quantity of z∕H to represent the A statistical–empirical model describing the acoustic response,
angle alignment between the incoming mean flow and the airfoil as a function of five influencing parameters, will be presented in
leading edge instead of the geometrical angle α. Sec. IV. Because of the parameters chosen for the design of
The noise experiments took place at the open-jet wind tunnel of the experiments (DOE), which will be discussed in Sec. III, a total of
aeroacoustic facility at Brunel University London. The exit nozzle, 10 LE sections were investigated. These include one configuration
which has a dimension of 100 × 300 mm, is situated inside a semi- with a straight LE to serve as the baseline case (BSLN). To conduct
anechoic chamber (4.0 × 5.0 × 3.4 m). It can produce a typical free-field measurements of the airfoil–gust interaction (AGI) noise,
turbulence intensity of between 0.1 and 0.2% [2,17]. The maximum the airfoil was held by side plates to keep the airfoil in place and to
jet velocity is about 80 m ⋅ s−1 . To generate elevated turbulence keep a two-dimensional flow profile [2]. Noise measurements at the
intensities Tu at the freestream, several turbulence grids of different aeroacoustic wind tunnel were made by a single Pico CoulomB
mesh size M and bar diameter d were used. As per the criteria 0.5 in. prepolarized ICP condenser microphone at polar angles of
suggested by Laws and Livesey [18], all the turbulence grids are Θ
90 deg at a vertical distance of 0.95 m from the LE of the
biplane square meshes with a constant ratio between the mesh size airfoil at midspan (Fig. 3). The acoustic data were recorded at a
and the bar diameter (M∕d
5). Using the turbulence prediction sampling rate of 40 kHz, where the measurement time was set to
model by Aufderheide et al. [19], which is based on the work of Laws 20 s. In the spectral analysis, Hamming windows were used for
and Livesey [18], five different turbulence grids that were predicted to windowing at 50% overlap by using a block size of 1024, yielding a
generate Tu in the range of 2.1 and 5.5% were manufactured. The frequency resolution of Δf
43 Hz and resulting in a number of
integral length scale of the turbulent eddies was found to be a function 1718 averaged blocks for the generated spectra. All the power

Fig. 2 Comparison of the normalized turbulence energy spectra between the theory (Liepmann) and experiment at 300;000 ≤ Re ≤ 600;000 and
Tu
3.9%. The streamwise location of the measurement coincides with the airfoil leading edge.
 BIEDERMANN ET AL. 3131

80
BL, Re 2.5e5, 71.7 dB
Serr, Re 2.5e5, 63.2 dB
70 w/o, Re 2.5e5, 54.1 dB
BL, Re 6.0e5, 93.6 dB
Serr, Re 6.0e5, 88.1 dB
60
w/o, Re 6.0e5, 74.4 dB

50

SPL, dB/Hz
40

30

20

10

0
103 104
f, Hz
Fig. 4 Far-field narrowband spectra at Re
250;000 and 600,000 with
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

baseline LE (BL), serrated A45λ7.5 LE (Serr), and background (“w∕o”)

Fig. 3 Schematic of the experimental setup. noise. Tu
5.5% and AOA
0 deg. OASPL is indicated in the box.

spectral densities (PSDs) presented in this study have a 1 Hz without the presence of an airfoil in the free jet, but still with both the
frequency bandwidth. turbulence grid and side plates installed, produces a significant lower
The frequency range (f1 ≤ f ≤ f2 ) chosen for the analysis in the sound pressure level compared to the cases when the airfoil is present.
overall sound pressure level (OASPL) is set to f1
300 Hz and Moreover, the sound pressure level in the case of the airfoil with a
f2
10 kHz. The decision on f1 is due to the consideration of the serrated LE (A45λ7.5 case) is always above the background noise,
cutoff frequency of the anechoic chamber, and f2 is decided upon especially in the intermediate frequency region of interest between
the exclusion of the possible influences by the airfoil self-noise from 300 Hz and 5 kHz, where the main contribution of the noise reduction
the trailing edge, which is not related to AGI noise. It is known is expected to come from the serrations:
that the OASPL could be very sensitive to the choice of f1 . p
A careful sensitivity study was performed to examine the change u 02
in OASPL and ΔOASPL with regard to the different values of f1 . Tuu
(3)
U
The sensitivity study demonstrates that both the OASPL and
ΔOASPL are reasonably unaffected at f1 ≥ 200 Hz. Therefore, the
B. Amiet’s Flat-Plate Comparison
current choice of f1
300 Hz should be able to characterize the
AGI noise accurately in the OASPL analysis. Another sensitivity Figure 5 shows the nondimensionalized far-field sound pressure
study was also performed to the statistical model, which also level spectra for both the baseline and serrated cases. The SPL is
confirms that the most dominant parameters will remain unaffected scaled with the fourth power of the freestream velocity Uo, whereas
as long as f1 ≥ 200 Hz. the frequency is scaled with the airfoil semispan and Uo . The collapse
of the spectra demonstrates that the AGI noise can be accurately
A. Measurement Environment
scaled with U4o especially for the baseline LE case. A slight deviation
can be observed for the serrated LE case, but generally the scaling law
The range of jet speeds under investigation is 25 ≤ Uo ≤ 60 m ⋅ s−1 , can still be applied in this case. This particular velocity dependency is
corresponding to Reynolds numbers based on the airfoil chord length consistent with the Amiet flat-plate model [21].
of 2 × 5 ⋅ 105 ≤ Re ≤ 6 × 105 , respectively. As shown in Table 2, the The Amiet model [21] was also used to validate the AGI noise
minima and maxima corresponding to each of the five influencing produced by a baseline, straight LE airfoil in the current setting.
parameters are defined. Preliminary measurements were performed
Amiet’s model was modified slightly by taking into account the
at the extreme flow conditions before the main acoustic study to
consideration of the airfoil thickness according to Gershfeld [22]:
ensure that the background noise of the wind-tunnel facility is well
below the AGI noise. At the minimum and maximum fan speed, the
  

acoustic spectra of the background noise, the baseline airfoil, and the 2b 2Λuu Γ1∕3 2 K^ 2x −ω⋅d
PSDω
Tu2 ρ20 U40 exp
airfoil with the A45λ7.5 serrated LE (expected to produce the largest πc 3πR Γ5∕6 1 K^ 2x 7∕3 2U0
reduction in the AGI noise level [2]) were measured. The acoustic p
results are shown in Fig. 4. The angle of attack was chosen to be at ω π Γ5∕6
where K^ x
− and ke
(4)
0 deg, and Tu at the vicinity of the LE is at the maximum of 5.5%. The U0 ke Λuu Γ1∕3
narrowband spectra in Fig. 4 demonstrate that the background noise
where Λuu is the longitudinal integral length scale of the turbulence;
Table 2 Nondimensional DOE levels of the different Tu is the turbulence intensity; R is the observer distance; b is the
factors of interest airfoil semispan; d is the airfoil thickness; and K^ x is the normalized
longitudinal wave number. Λuu and Tu were measured
Parameter −αDOE −1DOE 0DOE 1DOE αDOE independently. The model takes into account the cross-power
xNondim −2.378 −1.0 0.0 1.0 2.378 spectral density of the surface pressure on the airfoil caused by the
Re 250,000 351,422 425,000 498,578 600,000 turbulence.
Tuu, % 2.08 3.07 3.79 4.51 5.50 Figure 6 shows the comparison of the power spectral density of the
ASerr ∕C 0.080 0.144 0.190 0.236 0.300
λSerr ∕C 0.050 0.122 0.175 0.228 0.300
far-field noise for the baseline, straight LE at three different flow
z∕H −0.128 −0.054 0.000 0.054 0.128 velocities between the theory and experimental results. There is
reasonably good agreement with Amiet’s flat-plate model for the
 3132 BIEDERMANN ET AL.

An increase of the parameters inevitably leads to an exponential

rise of the necessary measurement trials (MTs). According to the
n-permutation in Eq. (5), analyzing a system with five parameters (k)
and varying the parameters on five levels each (n) will result in 3125
trials. This represents a hardly manageable experimental volume.
Instead, applying the statistical design of experiments (DOE)
approach as per Eq. (6) could lead to a significant reduction of the
experimental volume to 43 trials without a significant loss of
information on the system behavior. This approach keeps the
experimental volume manageable and facilitates the detailed analysis
of multiple parameters with a reasonably high accuracy:

MTn-per
nk
3125 (5)

Fig. 5 Nondimensionalized SPL spectra for the baseline and serrated

LE at different flow velocities. The airfoil is set at z∕H
0 with MTDOE−CCD
2k  2 ⋅ k  1
43 (6)
Tu
3.79%. Note that the SPL for the serrated LE is shifted by 30 dB.

A. Design of Experiments Methodology

The objective of the experimental modeling is the ability to
describe the defined experimental space by means of functions that
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

take into account of all the influencing parameters of significance

[Eq. (7)]. For this purpose, the response variables (RVs) have to be
defined to act as target values of the regression functions. The
coefficients are determined in accordance to the chosen set of
influencing parameters (IPs):
( )( i
1..3
Xn
Xn
RVi
f IPj  IPj  
2
IPj IPjk  j
1..5 (7)
j
1 k
1 k
1..4

The design of experiments methodology is based on the definition

of an experimental space for a setup that consists a full factorial core
[−1 : : :  1], star points [−α : : :  α] that label the upper and lower
experimental boundaries, and a central point [0], which is defined
as the experimental adjustment where all the parameters are on
Fig. 6 Comparison between Amiet’s flat-plate theory and the their intermediary values [23–26]. Based on this experimental
experiment results at Tu
3.8%, z∕H
0, and Λuu
5.8 mm. composition of the DOE methodology, the analytical statistic gathers
the population from a subset. A circumscribed central composite
design (CCD) was chosen as the appropriate experimental design.
mid-requency range. The underprediction at the high frequencies is Circumscribed CCDs are characterized by statistical properties, such
mainly due to the dominance of the trailing-edge self-noise. In case of as the orthogonality or rotatability [27].
the higher flow velocity, a larger discrepancy between the An experimental design is defined as rotatable if the variance of
measurement and prediction is observed at the low frequencies. This the probability distribution is a function of the distance between the
can be attributed to the larger influence of the wind-tunnel noise. At the star point and the central point, and not of the direction, as is the case
more extreme geometrical setting of the remaining parameters (e.g., with orthogonality. Given a set of points within the experimental
increased angle of attack), however, the comparison between the
space at a constant distance to the central point, the rotatable design
measured noise and Amiet model becomes less accurate. This is
shows a consistent accuracy in the prediction for all the points. With
mainly due to the increasing mismatch between Amiet’s flat-plate
regard to the statistical analysis, this property is highly
model and the NACA 65(12)-10 airfoil with a significant LE curvature
advantageous [23]. On the contrary, the orthogonal designs are
at large angle of attack.
advantageous because they can avoid the confounding of the
effects. This enables the determination of all the regression
coefficients independently [28,29].
III. Statistical–Empirical Modeling Technique In general, the α-values (star point locations) are higher than the
Before the modeling, it is important to first identify meaningful coordinates of the central core (αDOE > 1) and thus represent the
target values for the present study. A set of five parameters, namely limits of the experimental space. Consequently, each factor is varied
the Reynolds number Re, turbulence intensity Tu, serration as a combination of the five nondimensional levels [α, 1, 0, −1,
amplitude A∕C, serration wavelength λ∕C, and angle of attack (z∕H), −α]. A special design is the combination of both the properties in
were selected for the analysis. The main objective of the present work orthogonality and rotatability. Because the requirements of
is to develop a statistical–empirical model to describe the orthogonality are not completely grantable while simultaneously
independent effects and the interdependencies among the five guaranteeing the rotatability, this design is defined as pseudoor-
influencing parameters on the AGI noise reduction. For this purpose, thogonal and rotatable. It combines the advantages of both properties,
the statistical design of experiments (DOE) approach was used. especially because the resulting confounding is of negligible
When analyzing a defined physical experimental space by varying magnitude (less than 0.02%).
several influencing parameters, the classical method would be to vary As already described, the design of experiments approach is
one of the parameters, while the others remain constant. This limited to describing the experimental space of interest by the
procedure will then be repeated for each parameter of interest (raster functions of first and second order as well as linear interdependencies
method). This might be an easy and effective method to describe the between the single influencing parameters [Eq. (7)]. To choose a
influence of these parameters on a certain response variable with a valid model, preliminary investigations are necessary to ensure that
high accuracy, as long as the number of parameters is small, and the the system satisfies these conditions. With this purpose, all the five
interdependencies between the parameters are disregarded. influencing parameters were analyzed individually in a preliminary
 BIEDERMANN ET AL. 3133

study, and their effects on the target values were evaluated carefully. OASPLBL gives the overall sound pressure level reduction ΔOASPL
This analysis, in combination with the defined experimental design, [Eq. (10)]:
results in the test matrix as shown in Table 2, which also includes the

upper and lower parameter settings. The total number of p
ΔOASPL
OASPLBL − OASPLSerr
20 ⋅ log BL (10)
measurement point is 43, in addition to 16 repetitions for the central p Serr
point to define a system-characteristic statistical spread and to
guarantee the desired statistical features. The trials of the strategically
planned experiment were performed in a random order, and they were
repeated twice to obtain the average values. This procedure is to IV. Selection of Key Noise Results
secure the reduction and elimination of unknown and uncontrollable Generally, as shown in Fig. 7a, the serration amplitude is the main
quantities. Additionally, the analysis of the statistical significance factor in reducing the broadband noise. At Re
425;000,
allows the elimination of parameters with impacts on the response Tu
3.79%, and z∕H
0, the LE serration is found to be the
variable that are smaller than the statistical spread. most effective in frequency range between 850 and 3500 Hz
(indicated by the vertical dashed lines in Fig. 7a), where an average
B. Response Variables
sound pressure level reduction of up to ΔSPL ≈ 10 dB is achieved by
the largest serration amplitude (A∕C
0.3).
The response variables (RVs) can be described by means of all The turbulence intensity by itself was found to be another
influencing parameters in the first and second order as well as the important factor that can dictate the level of broadband noise
interdependencies between the influencing parameters [Eq. (7)]. reduction. The narrowband spectra subjected to Tu
2.08, 3.79, and
Defining the response variables is a crucial part of evaluating the 5.5% under influencing parameters of Re
425;000, z∕H
0,
experimental data. They are expected to describe the system with the λ∕C
0.175, and A∕C
0.19 are plotted in Fig. 7b. It can be seen
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

necessary accuracy. This study focuses on the overall sound reduction that the ΔSPL increases with increasing level of Tu. This effect is
of serrated LE compared to a baseline LE and does not predict the SPL especially distinct in the intermediate frequency range of
at a particular frequency. Consequently, the response variables of approximately 800 Hz to 4 kHz. It is important to note that the
interest are limited to the OASPL. To define a sound pressure reduction frequency range corresponding to the scenario when ΔSPL > 0
level, information on both the baseline and the serrated LE is necessary. broadens when Tu increases. Increasing Tu from low to intermediate
It is important to note that single-microphone measurements were values causes an increase of the upper frequency limit where noise
performed; hence, information on the sound directivity, overall sound reduction is still effective. A high Tu also leads to an increase of the
power levels, and sound power reduction is not available. noise reduction toward the low frequencies. More specifically, the
However, the dependencies of the sound generation itself are also level of broadband noise radiation for the baseline LE, and thus the
of interest because it facilitates the analysis of the influence of each OASPLBL , increases with the level of Tu. Consequently, the noise-
case on the reduction independently. As shown in Eq. (8), the noise reduction capabilities of the serrated LE are also the most effective at
produced by a baseline LE is a function of the Reynolds number, the elevated Tu condition.
turbulence intensity, and angle of attack. In the case of serrated LE, The noise reduction due to the serrated LE should increase linearly
additional influences of the serration wavelength and amplitude must with frequency (at mid- to high frequencies) until it is masked by the
be taken into consideration [Eq. (9)]: self-noise at high frequency, and little reduction should be expected at


 low frequencies. These characteristics can also be reproduced in the
p z current experimental results. Figure 8 shows the ΔSPL spectra for a
OASPLBL
20 ⋅ log BL → OASPLBL
f Re; Tu; (8) number of cases involving different serration wavelengths and
pref H
amplitudes as well as the turbulence intensities and Reynolds

 number. Note that the frequency is normalized by use of the free
p Serr stream velocity and the baseline chord length, and thus can be
OASPLSerr
20 ⋅ log → OASPLSerr expressed in terms of the Strouhal number. When the serration
pref

 amplitude is fixed but with different serration wavelengths (Fig. 8a),
A λ z the spectra demonstrate a linear increase of ΔSPL from mid- to high

f Re; Tu; ; ; (9)
C C H frequencies until reaching ΔSPL ≈ 12 dB for the one with the
smallest serration wavelength. After reaching this peak, the ΔSPL
Note that pref
2 × 10−5 Pa, and the frequency range of interest is begins to drop at higher frequency due to the prominence of the self-
between 300 Hz and 10 kHz. Subtracting the OASPLSerr from the noise radiation. As shown in Fig. 8b, when the serrated leading edge

60 90
BL, 84.5 dB Re = 425,000
55 A/C = 0.08, 83.6 dB 80 Tu λ/C = 0.175
50 A/C = 0.19, 80.3 dB A/C = 0.19
A/C = 0.30, 78.6 dB 70 z/H = 0
45
SPL, dB/Hz

60
SPL, dB/Hz

40
50
35
A/C
40
30
Re = 425,000
25 λ/C = 0.175 30
Tu = 3.79% Tu = 2.08 %, ΔOASPL = 4.18 dB
20 20 Tu = 3.79 %, ΔOASPL = 4.19 dB
z/H = 0
Tu = 5.50 %, ΔOASPL = 5.45 dB
15 10
103 104 103 104
a) f, Hz b) f, Hz
Fig. 7 Spectra for the baseline (BL, straight) and serrated airfoil (dashed) at a) different serration amplitudes (A∕C) and b) different Tu. Spectra in b) are
shifted by 0, 15, and 30 dB.
 3134 BIEDERMANN ET AL.

15 15
A29λ7.5 U = 25.6 m/s
10 A29λ26 U = 43.5 m/s
10
A29λ45 U = 61.4 m/s
ΔSPL, dB/Hz

ΔSPL, dB/Hz
5 5

0 0

-5 -5
U = 43.5 ms-1
A29λ26
Tu = 3.8%
Tu = 3.8%
-10 -10
10-1 100 101 102 103 10-1 100 101 102 103
a) f c0/U b) f c0/U
15 15
A12λ26 Tu = 2.1%
10 A29λ26 10 Tu = 3.8%
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

A45λ26 Tu = 5.5%

ΔSPL, dB/Hz
ΔSPL, dB/Hz

5 5

0 0

-5 -5
U = 43.5 ms-1 U = 43.5 ms-1
Tu = 3.8% A29λ26
-10 -10
10-1 100 101 102 103 10-1 100 101 102 103
c) f c0/U d) f c0/U
Fig. 8 ΔSPL spectra at different a) serration wavelength λ, b) Reynolds number Re, and c) serration amplitude A, and d) turbulence intensity Tu.

is subjected to different flow velocities, it produces the ΔSPL spectral serrations. The diagram (Fig. 10b) shows that, in contrast to the
shape that is very similar to those presented by Narayanan et al. [1] response variables of the serrated noise in the absolute value of
and Chaitanya et al. [3] (i.e., the effective frequency range OASPL (Fig. 10a), the most dominant factor affecting the level of
underpinning the ΔSPL will increase with increasing flow velocity). broadband noise reduction is the serration amplitude. The Reynolds
Figure 8c shows the influence of the serration amplitude on the ΔSPL number, previously the strongest enhancing factor for the absolute
spectra, which demonstrates that larger levels of noise reduction can OASPL, seems to become less important for the sound reduction
be achieved with a larger serration amplitude. This observation is capability. Moreover, an increased influence of the serration
consistent with the results reported by Narayanan et al. [1]. Finally, wavelength on the sound reduction is apparent in a linear and
Fig. 8d shows the influence of the freestream turbulence intensity on quadratic form. In general, the most significant dependencies of the
the noise reduction, which is consistent with the earlier results. overall sound pressure level reduction (ΔOASPL) demonstrated in

A. General System Information

All four response variables were analyzed with the design of
experiment methodology described earlier. Figure 9 shows the
comparison of the experimentally observed and the regression-
predicted OASPL values produced by a serrated LE. The diagonal
line represents a perfect match between the observed and predicted
OASPL. Figure 9 shows that the measured OASPL for the serrated
airfoil have an excellent agreement with the model, resulting in a
standard deviation of 0.15–0.17%.
The radiated noise by a baseline LE was also analyzed at different
Reynolds numbers, turbulence intensities, and angles of attack. Note
that the influence parameters of serration amplitude and serration
wavelength are not included in the noise prediction for the baseline
airfoil. The statistical spread rises slightly, however, when defining
the ΔOASPL as uncertainties of the baseline and the serration
prediction accumulate. For the first time, a ranking of the main factors
and the interdependencies by means of their influence on the
broadband noise reduction is presented. The Pareto diagrams in
Fig. 10 show the enhancing (greater than 0) and suppressing Fig. 9 Check of fit for the statistical–empirical model. Plot of observed
(less than 0) effects of the influence parameters on the target values. vs predicted values of the OASPL level produced by a serrated LE. There
The ΔOASPL characterizes the sound reduction capability of the LE are 59 measurement points.
 BIEDERMANN ET AL. 3135

Fig. 10 Pareto diagrams. Ranking of enhancing and suppressing, linear (L) and quadratic (Q) effects. Indicated level of statistical significance (p
5%).
Response variable for a) OASPL (serration) and b) ΔOASPL.
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

this study are consistent with the findings of previous studies radiation from a baseline airfoil. This in turn facilitates an increase of
[2,3,5,30–35]. the noise-reduction capability when a serrated LE is used. However, it
The resulting sound pressure level for both the straight and serrated is important to note that the effects of the individual parameters on the
LEs, as well as the corresponding sound reduction, can be predicted overall noise reduction in Fig. 11 cannot be attributed to the serrated
with respect to the different influence parameters by the regression LEs only. This is because different levels of Reynolds number, Tu,
functions in Table 3. The model predicts the response variables by and angle of attack (z∕H) can also affect the baseline straight LE.
taking into account the statistically significant factors, where factors, Thus, a more detailed analysis of the noise reduction by the serration
whose influence is smaller than the statistical spread, can be added to requires an independent analysis of the noise radiation by the baseline
the error term. This helps reducing the complexity of the regression and serrated airfoil, respectively.
functions and keeps the number of variables to a minimum.
The intermediate effect on the influencing parameters within the B. Interdependency of Serration Wavelength and Turbulence
experimental space on the overall noise reduction is plotted in Fig. 11. Intensity (λ∕C ⋅ Tu)
The serration amplitude has the highest intermediate effect with an A significant effect identifiable in the response variable ΔOASPL
almost linear relationship between the A∕C and ΔOASPL, before was found to be an interdependency of the serration wavelength and
reaching an asymptotic level when the A∕C is increased further. The the turbulence intensity (λ∕C ⋅ Tu), as can be seen in the Pareto
serration wavelength λ∕C shows a nonlinear behavior where the diagram (Fig. 10b). At low Tu, small serration wavelengths are
optimum is achieved at intermediate wavelength, beyond which the needed to achieve a high level of noise reduction, as exhibited by the
noise-reduction capability is weakened considerably. The predicted red color region in Fig. 12. Because Tu is related to the integral length
profile for the influence of the turbulence intensity Tu exhibits a large scale Λuu of the incoming gust, large serration wavelengths are
level of noise reduction at high Tu levels. On the contrary, at a band of expected to be less effective in the decorrelation effects especially if
low Tu, a low level of noise reduction is predicted. This is in the incoming gust is characterized by small turbulent eddies.
agreement with the measurements shown in Fig. 7b, where a high Tu Previous investigations suggested that serration wavelengths should
is identified to be able to cause a high level of broadband noise be small to achieve good level of noise reduction, although in general

Table 3 Functions of response variables defined as linear combination

Term OASPLBL , dB OASPLSerr , dB ΔOASPL, dB

Constant 1.116E  01 6.780E  00 2.123E  00
ReL 1.478E − 04 ⋅ Re 1.437E − 04 ⋅ Re 4.002E − 06 ⋅ Re
ReQ −1.049E − 10 ⋅ Re2 −9.310E − 11 ⋅ Re2 −1.175E − 11 ⋅ Re2
TuL 1.234E  01 ⋅ Tu 1.491E  01 ⋅ Tu −2.426E  00 ⋅ Tu
TuQ −1.281E  00 ⋅ Tu2 −1.472E  00 ⋅ Tu2 1.922E − 01 ⋅ Tu2
A∕CL — — −3.800E  01 ⋅ A∕C 4.994E  01 ⋅ A∕C
A∕CQ — — 3.249E  01 ⋅ A∕C2 −6.538E  01 ⋅ A∕C2
λ∕CL — — 1.352E  01 ⋅ λ∕C −1.835E  00 ⋅ λ∕C
λ∕CQ — — 3.531E  01 ⋅ λ∕C2 −5.787E  01 ⋅ λ∕C2
z∕HL −1.142E  01 ⋅ z∕H −1.357E  01 ⋅ z∕H 4.658E  00 ⋅ z∕H
z∕HQ −5.152E  01 ⋅ z∕H2 −5.638E  00 ⋅ z∕H2 −4.574E  01 ⋅ z∕H2
Re ⋅ Tu 6.323E − 07 ⋅ Re ⋅ Tu 1.715E − 07 ⋅ Re ⋅ Tu 4.608E − 07 ⋅ Re ⋅ Tu
Re ⋅ A∕C — — 2.327E − 05 ⋅ Re ⋅ A∕C −1.876E − 05 ⋅ Re ⋅ A∕C
Re ⋅ λ∕C — — −2.221E − 05 ⋅ Re ⋅ λ∕C 1.709E − 05 ⋅ Re ⋅ λ∕C
Re ⋅ z∕H −8.150E − 06 ⋅ Re ⋅ z∕H 1.574E − 05 ⋅ Re ⋅ z∕H −2.389E − 05 ⋅ Re ⋅ z∕H
Tu ⋅ A∕C — — −2.349E  00 ⋅ Tu ⋅ A∕C 1.951E  00 ⋅ Tu ⋅ A∕C
Tu ⋅ λ∕C — — −4.268E  00 ⋅ Tu ⋅ λ∕C 3.847E  00 ⋅ Tu ⋅ λ∕C
Tu ⋅ z∕H 3.069E  00 ⋅ Tu ⋅ z∕H 2.115E  00 ⋅ Tu ⋅ z∕H 9.546E − 01 ⋅ Tu ⋅ z∕H
A∕C ⋅ λ∕C — — 2.290E  01 ⋅ A∕C ⋅ λ∕C −2.521E  01 ⋅ A∕C ⋅ λ∕C
A∕C ⋅ z∕H — — 6.987E  00 ⋅ A∕C ⋅ z∕H −1.649E  01 ⋅ A∕C ⋅ z∕H
λ∕C ⋅ z∕H — — −4.926E  01 ⋅ λ∕C ⋅ z∕H 4.524E  01 ⋅ λ∕C ⋅ z∕H
 3136 BIEDERMANN ET AL.
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

Fig. 11 Intermediate impact of the influencing parameters on the ΔOASPL, including the error band. The horizontal blue band indicates average noise
reduction by the use of serrated LEs.

the impact of the serration wavelength is not as dominant as the compared to the low turbulent case for the baseline airfoil. The
serration amplitude [2,3,5]. The interdependency in Fig. 12 shows serrated LEs respond more sensitively at high Tu for the OASPL
that the optimum serration wavelength highly depends on the regardless the level of the serration wavelength, where up to 4.8 dB in
incoming Tu level. Low to intermediate Tu support the previous ΔOASPL could be achieved. For the low Tu case, however, there is a
findings that a smaller serration wavelength is more desirable. wider spread of the noise-reduction efficiency among the different
However, at high Tu, serration wavelengths of intermediate values serration wavelengths, where a low serration wavelength tends to
are far more effective in reducing the OASPL, as shown in Fig 12. achieve higher noise reduction especially in the low-frequency region
This agrees with the finding of a recent work [3], where the optimum of 300 Hz ≤ f ≤ 2 kHz. This supports the interdependency
serration wavelength is found to be twice the size of the incoming described previously because the level of noise reduction differs only
turbulent structure in the form of the integral length scale Λuu . An slightly when varying the serration wavelength at high Tu; at low or
optimal set of Tu and λ∕C leads to a reduction of the fluctuating intermediate Tu, large serration wavelengths become ineffective in
acoustic pressure of about 53% to that produced by the baseline the noise reduction. Moreover, in the low-frequency region, the noise
airfoil. reduction for the intermediate to large serration wavelengths mainly
To gain a deeper insight in the underlying principles, the acoustic takes place at frequencies f ≤ 4 kHz. However, in the high Tu case,
spectra at Re
400;000, A∕C
0.2, and z∕H
0 have been
analyzed in more detail. Figure 13 shows the influence of the
serration wavelength at Tu
5.5% (Fig. 13a) and at Tu
3.2% 50
(Fig. 13b) and compares them to the baseline case. In general, high Tu=5.5%
Re = 400,000
turbulent inflow causes an OASPL that is about 4.5 dB higher 40 A/C = 0.2
z/H = 0
SPL, dB/Hz

30

20 BSLN; OASPL = 84.28 dB

λ/C = 0.3; OASPL = 80.57 dB
10 λ/C = 0.2; OASPL = 79.92 dB
λ/C = 0.1; OASPL = 79.41 dB
λ/C = 0.05; OASPL = 79,50 dB
0
a) 103 104

50
Tu=3.2%
40 Re = 400,000
A/C = 0.2
SPL, dB/Hz

z/H = 0
30

20 BSLN; OASPL = 79.74 dB

λ/C = 0.3; OASPL = 77.57 dB
10 λ/C = 0.2; OASPL = 76.38 dB
λ/C = 0.1; OASPL = 75.40 dB
λ/C = 0.05; OASPL = 75.45 dB
0
103 104
Fig. 12 Influence of interdependency between the serration wavelength b) f, Hz
(λ∕C) and turbulence intensity Tu on the ΔOASPL. Other influencing Fig. 13 Narrowband spectra for the different serrations wavelength at
factors remain on intermediate levels (Re
425;000, A∕C
0.19, a) Tu
5.5%, and b) Tu
3.2%. The remaining parameters remain on
z∕H
0). constant, intermediate levels at Re
400;000, z∕H
0, A∕C
0.2.
 BIEDERMANN ET AL. 3137

the noise reduction occurs up to f
6 kHz, albeit the comparatively 50

BSLN; OASPL = 80.99 dB
low energy level in the high-frequency region prevents significant λ/C = 0.3; OASPL = 78.04 dB
changes in the overall sound pressure level. 40 λ/C = 0.2; OASPL = 77.43 dB
λ/C = 0.1; OASPL = 76.54 dB
C. Interdependency of Angle of Attack and Serration Wavelength λ/C = 0.05; OASPL = 76.64 dB
(z∕H ⋅ λ∕C) 30

SPL, dB/Hz
In addition to the interdependency between the turbulence intensity
and serration wavelength described previously, another interdepend- 20
ency between the angle of attack and the serration wavelength
(z∕H ⋅ λ∕C) was also found to be significant, as shown in Fig. 14. In
general, the level of noise reduction by the serrated LEs is the 10
maximum at zero angle of attack, which is in agreement with other Re = 400,000
authors [2,3]. However, at a large negative z∕H (or AOA), small Tu = 3.7%
0 A/C = 0.2
serration wavelengths are needed to achieve a reasonably large noise
z/H = 0
reduction (red color region), whereas serration wavelengths of
intermediate dimensions are preferable at large positive z∕H. The -10
underlying principle of this interdependency could be related to the 103 104
specific semicyclic shape of the LE serrations relative to the stagnation f, Hz
point of the incoming flow. In the case of zero angle of attack, the
Fig. 15 Narrowband spectra produced by the baseline and serrated LEs
serration wavelength that can achieve the largest noise reduction is of different serration wavelength at Re
400;000, Tu
3.7%,
defined by how well it can decorrelate the spanwise coherence of the A∕C
0.2, and z∕H
0.
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

turbulence eddies and how efficient it can facilitate a “nozzle” effect to

accelerate the flow from the serration tip to the serration root and
reduce the Tu level. At a negative z∕H, the projected area of the upper However, this effect is neglected in the present study because the
surface of the LE will cause a significant impingement upon the lower and upper frequency limits for the OASPL analysis are taken
incoming gusts. In this case, the use of small wavelength serrations is a at 300 Hz and 10 kHz, respectively. Within this frequency range, the
logical choice. noise-reduction capability usually improves when the serration
This is because a small serration wavelength will cause many three- wavelength reduces.
dimensional undulations on the upper surface of the LE, which will The impact of the serration wavelength on the low-frequency
maintain the serration effect to achieve the interaction noise reduction. region can be examined in Fig. 16, where the noise reduction ΔSPL
In the case of a positive z∕H, the incoming flow will naturally of a serrated LE with the intermediate (λ∕C
0.1) and a small
impinge on the lower surface of the LE. However, the planar serration wavelength (λ∕C
0.05) are compared with each other as a
geometry at the lower surface of the serrated LE means that the three- function of the inflow velocity and frequency. Above f
1 kHz, the
dimensional undulation can no longer be achieved by using a small noise reduction is significantly higher for the small serration
serration wavelength. Instead, a larger serration wavelength is wavelength (Fig. 16b). However, as indicated by the circled black
preferable to avoid the direct impingement between the incoming region in Fig. 16a, the level of noise reduction is higher for the
gusts and the LE geometry. intermediate wavelength in a frequency range of 500 Hz
Of particular interest is the impact of this interdependency ≤ f ≤ 1.2 kHz. As a result, the ΔOASPL is slightly higher for the
(z∕H ⋅ λ∕C) on the narrowband spectral characteristic. It has intermediate λ∕C
0.1 case.
already been known that the smallest serration wavelength does not Figure 17 shows the SPL spectra produced by the baseline and
necessarily lead to a maximum noise reduction across the whole serrated LEs of a slightly larger serration wavelength (λ∕C
0.175)
frequency range [2]. A first indicator for the effect of small serration at different angles of attack AOA (or z∕H). At first glance, it is clear
wavelengths on the noise can be found in the high-frequency region that the largest level of noise reduction occurs at z∕H > 0 and across
above 10 kHz (Fig. 15), where all the different serration the widest frequency range (Fig. 17a). The level and frequency range
wavelengths actually lead to noise increase. It is clear that the largest of the noise reduction are lessened as the z∕H decreases. However, a
noise increase occurs at the smallest serration wavelengths. closer examination shows that the sensitivities of the SPL and

Fig. 14 Influence of interdependency between the serration wavelength and angle of attack (λ∕C ⋅ z∕H) on the ΔOASPL. Other influencing factors
remain on the intermediate levels (Re
425;000, A∕C
0.19, Tu
3.8%).
 3138 BIEDERMANN ET AL.

Fig. 16 Contours of ΔSPL as a function of f and Uo. The serrated LEs have the common influence parameters of A∕C
0.2, z∕H
0, and Tu
3.7%,
but at different a) λ∕C
0.1 and b) λ∕C
0.05.

60 60 60
Re = 425,000 Re = 425,000 Re = 425,000
Tu = 3.8 % Tu = 3.8 % Tu = 3.8 %
50 λ/C =0.175 50 λ/C =0.175 50 λ/C =0.175
A/C = 0.19 A/C = 0.19 A/C = 0.19
SPL, dB/Hz

SPL, dB/Hz

SPL, dB/Hz
40 40 40
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

30 30 30

20 20 20
BSLN z/H = 0.128; OASPL = 84.16 dB BSLN z/H = 0; OASPL = 84.56 dB BSLN z/H = -0.128; OASPL = 83.99 dB
Serr z/H = 0.128; OASPL = 80.03 dB Serr z/H = 0; OASPL = 80.44 dB Serr z/H = -0.128; OASPL = 81.18 dB
10 10 10
103 104 103 104 103 104
f, Hz b) f, Hz f, Hz
a)
c)
Fig. 17 Narrowband spectra (300 Hz < f < 10 kHz) for the noise produced by the baseline LE (solid lines) and serrated LE (dotted lines) at
a) z∕H
0.128, b) z∕H
0, and c) z∕H
−0.128.

OASPL to the z∕H are different between the baseline and serrated the absolute level of noise radiation (SPL and OASPL) and the level
LEs. For the baseline LE, the lowest OASPL is achieved at z∕H < 0, of noise reduction (ΔSPL and ΔOASPL) reasonably accurately.
which is mainly due to the low level of noise radiation in the However, up to now, the model is based on a data set of 59
frequency range of 2 ≤ f ≤ 4 kHz. The largest level of OASPL measurement trials, including the measurement of the central point
produced by the baseline airfoil, on the other hand, is achieved at for 17 times in a randomized order to describe the statistical spread
z∕H
0. The noise level produced at z∕H > 0 lies in between. For of the model. To improve the stability of the developed model,
the serrated LE, the lowest level of the OASPL is achieved at additional data points were incorporated from an independent
z∕H > 0. As the z∕H slowly decreases, the level of OASPL previous study, which took place under the same measurement
increases. This contradictory behavior between the baseline and conditions, but with different leading-edge serrations and flow
serrated LEs for the noise radiation should be taken into parameters [2]. The measurement results from the previous study are
consideration when examining the interdependency (z∕H ⋅ λ∕C) combined with the current model to increase the total measurement
for the ΔSPL or ΔOASPL. points to 285. This improved model serves as a database to predict the
In Fig. 17c, when the airfoil is set at z∕H
−0.128, there is noise radiation by the serrated LE at various configurations as well as
little noise reduction at 1.2 ≤ f ≤ 3.8 kHz because of the opposite the noise reduction when compared with the baseline airfoil. The
trends in SPL produced by the baseline airfoil (reduction in the additional measurement results are based on eight serration designs
SPL level) and serrated airfoil (increase in the SPL level), (see the test matrix for additional serration parameters in Fig. 18a),
respectively. At f > 3.8 kHz, the serrated LE even causes a which were tested in a velocity range of 20 ≤ U0 ≤ 60 m ⋅ s−1 , or
significant noise increase, which is not due to the experimental Reynolds number of 200;000 ≤ Re ≤ 600;000, respectively. The
error because we have retested it many times. Rather, the presence turbulence level was varied by interchanging three grids that yield
of the serration wavelength actually facilitates crossflow from the Tu
3.2, 3.7, and 5.5%. The angle of attack was altered from
projected upper surface of the LE, through the serration air gaps −0.102 ≤ z∕H ≤ 0.128, or −8 ≤ AOA ≤ 10 deg, respectively.
and exits the lower surface of the LE. This particular fluid– The extended model was found to fit well to the initially developed
structure interaction causes the noise to increase at high frequency algorithm, as indicated by the regression curve for the OASPL in
that will otherwise be absence in a baseline LE. This conjecture is Fig. 18b (initial reference model in shown in Fig. 9). The fit of
supported by the clear trend from other results where the noise regression shows a good match of high order when comparing
between the observed and predicted values for the OASPL by the
increase at high frequency will gradually cease to exist when the
serrated LE.
angle of attack increases.
If the original model (Fig. 10a) is compared with the refined one
(Fig. 18a), the dependencies are found to be the same order and
D. Model Refinement magnitude for the OASPL and ΔOASPL. When examining Fig. 18a,
Complementary measurements were carried out in the outer the modified interdependency plot between Reynolds number and
regions of the defined experimental space to test the stability of the λ∕C for the OASPL of serrated airfoil remains almost unaffected by
statistical model, especially at the extreme settings of the influencing the additional data points, although it is expected to be more reliable
parameters. The measurement results were found to fit well to the due to the increased data pool. This is manifested in Fig. 19, where the
model, although an increase of prediction uncertainty was observed blue circles indicate the underlying set of data on which the model
with multiple factors on the extreme levels, which statistically was established. Despite a significant increase of the amount of
represents a large distance between the central point and the experimental data in Fig. 19b, the shape and magnitude of the
measurement locations. In general, the model was found to predict interdependency remain almost constant.
 BIEDERMANN ET AL. 3139

Fig. 18 Check of model validity for serrated LE noise (OASPLSerr ): a) Pareto diagram for extended model (including test matrix), and b) observed vs
predicted values of OASPLSerr comprising 285 measurement points.
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

Fig. 19 Comparison of interdependency between the (Re ⋅ λ∕C) for a) original model, and b) improved model by the use of additional data points (circles)
obtained from [2].

To conclude, the additional data pool in the modified model has a

negligible impact on the main factors and the reliability of the original
model, which uses less data points. Furthermore, all the
interdependencies remain very similar, although there are slight
changes in the magnitude as a result of the additional data pool.
Therefore, the results presented in this section confirm that the
original statistical–empirical model is still stable and reliable.

E. Polyoptimum of Noise Reduction and Noise Radiation

The purpose of the current system is to reduce the broadband LE
noise caused by the interaction between the high turbulent inflow
conditions and the LE. Therefore, in addition to the main focus on
achieving a high level of noise reduction as defined by the relative
Fig. 20 Weighted functions of the single analyzed response variables.
difference between airfoils with straight and serrated LEs, it is also
desirable to produce low absolute magnitude of the overall sound
pressure. An algorithm has been developed to define a polyoptimum
of the radiated noise and the noise-reduction capability by the Figure 21 shows the polyoptimum of noise radiation and noise
serrated LEs. reduction with serrated LEs. The first row shows the dependencies on
Of further interest is the qualitative information on the impact of the noise radiation (OASPL) where the Reynolds number effect is
the different influencing parameters on the polyoptimum. As shown clearly the most influential. It is then followed by Tu and A∕C. λ∕C
in Fig. 20, the independent response variables OASPLSerr and and z∕H, on the other hand, can be considered as the secondary
ΔOASPL were weighted with an emphasis on the reduction of the importance. The effects of the influencing parameters on the noise
OASPL. This means that the OASPL is weighted linearly between 50 reduction (ΔOASPL) are shown in the second row of Fig. 21. With
and 70 dB, where 50 dB equals an acceptability of 100%. In contrast, the exception of Tu, the rest of the influencing parameters exert the
the ΔOASPL accounts for 0% acceptability at 2 dB with a slope of 2 reverse trend between the OASPL (for the serrated LEs) and the
until reaching the desired optimum of 10 dB overall noise reduction ΔOASPL. For example, an increase of Reynolds number would
(100% acceptability). increase the OASPL. As a result, the ΔOASPL will reduce. A
 3140 BIEDERMANN ET AL.
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

Fig. 21 Polyoptimum of the OASPL (top row) and the noise-reduction capability ΔOASPL (center row), as defined by the function of acceptability
(bottom row and Fig. 20).

maximum noise reduction of ΔOASPL
7.5 dB, while maintaining C
150 mm and a span of S
450 mm. The authors forced a
a relatively low noise radiation of OASPL
53.8 dB, is reached at bypass transition of the boundary layer from laminar to turbulent by
the minimum Reynolds number, Tu, and λ∕C in combination with the tripping tapes to prevent the production of the laminar instability
the maximum A∕C and z∕H. As shown in the third row in Fig. 21, the tonal noise [3]. At elevated level of freestream turbulence, the LE
contribution of each of the parameters to the polyoptimum, which is noise is considered to be the dominant noise source. Therefore in this
defined as the acceptability, spreads over large margins. To achieve a case the boundary layer tripping can be assumed to have no influence
minimum absolute level of noise radiation while maintaining a high on the radiated noise [36,37]. The tests in ISVR were performed by
noise-reduction capability of the serrations, one could use the the use of serrated sinusoidal LEs, defined by the amplitude, with a
algorithm of the polyoptimum to optimize the effective degrees of peak-to-trough ratio of 2h and wavelength λ. Note that there is a
freedom when other design parameters are fixed. difference in the definition of the serration parameters, where ISVR
adopted the “same wetted area” principle. This means that, for the
F. Model Validation with External Data same serration amplitude A
2h, the serration peak would extend
To validate the statistical–empirical model developed in this paper, the initial airfoil chord length by h, giving an overall chord of
the predicted OASPL and ΔOASPL are compared with external data (C  h). Accordingly, the serration root would, retracted by h, give
obtained independently in the aeroacoustic wind tunnel at the an overall chord of (C − h).
Institute of Sound and Vibration Research (ISVR), University of The turbulence intensities at the ISVR were generated at Tu
2.5
Southampton [3]. The model was scaled in accordance with the and 3.2%, and the incoming flow velocities are U0
20, 40, and
different experimental conditions (e.g., microphone measurement 60 m ⋅ s−1 . The difference in distance of the far-field microphone
locations) before the comparison was made. The airfoil used in the location is corrected by use of the monopole scaling law according to
ISVR is the same type (NACA 65(12)-10) with a chord length Eq. (11):


 

 R1   R 

SPLR2 
SPLR1 −20 ⋅ log → ΔSPL
20 ⋅ log 1 
 (11)
R2  R2

Fig. 22 Validation of the current model (lines) with external experimental data (symbols) provided by ISVR [3] at different a) amplitudes and
b) wavelengths. Predicted OASPL (solid lines) and ΔOASPL (dotted lines).
 BIEDERMANN ET AL. 3141

where R1 and R2 are the absolute distances between the source The current model has not yet considered additional influencing
and the observer (measurement location) at a polar angle of parameters such as the serration curvature and the curvature angle of
Θ
90 deg. Differences in the span were also compensated by a the airfoil leading edge, which could otherwise expand the model to
linear scaling. Twelve measurement points were analyzed at zero other airfoil geometries. This gap provides an incentive for future
angle of attack and Tu
2.5%. The Reynolds numbers were work to improve the robustness and fidelity of the current model.
matched at Re
394;000 and 624,000. The serration amplitudes
are varied by 0.1 < A∕C < 0.35, and the serration wavelengths are
varied by 0.05 < λ∕C < 0.25. References
Applying the specific boundary conditions of the ISVR test rig to [1] Narayanan, S., Chaitanya, P., Haeri, S., Joseph, P., Kim, J. W., and
the current model yields the predictions of the OASPL for both the Polacsek, C., “Airfoil Noise Reductions Through Leading Edge
baseline and serrated airfoils, which are shown in Fig. 22. It is clear Serrations,” Physics of Fluids, Vol. 27, No. 2, 2015, Paper 025109.
that excellent agreement has been achieved between the predicted doi:10.1063/1.4907798
and measured data. The overall noise reduction ΔOASPL also [2] Chong, T. P., Vathylakis, A., McEwen, A., Kemsley, F., Muhammad, C.,
demonstrates a good agreement with the predictions, although with a and Siddiqi, S., “Aeroacoustic and Aerodynamic Performances of an
slightly larger discrepancy due to the accumulated errors in the Aerofoil Subjected to Sinusoidal Leading Edges,” 21st AIAA/CEAS
Aeroacoustics Conference, AIAA Paper 2015-2200, 2015.
OASPL radiated by both the baseline and serrated airfoils.
doi:10.2514/6.2015-2200
The OASPL reduces when the serration amplitude increases, as [3] Chaitanya, P., Narayanan, S., Joseph, P., Vanderwel, C., Kim, J. W., and
predicted by the model. The influence of the serration wavelength Ganapathisubramani, B., “Broadband Noise Reduction Through
shows a different behavior. The predicted data underline a decreasing Leading Edge Serrations on Realistic Aerofoils,” 21st AIAA/CEAS
ΔOASPL with increasing serration wavelength (Fig. 22b). This trend Aeroacoustics Conference, AIAA Paper 2015-2202, 2015.
contradicts with the ISVR’s experimental findings, which show that doi:10.2514/6.2015-2202
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

the ΔOASPL increases slightly with the serration wavelength. The [4] Hersh, A. S., Sodermant, P. T., and Hayden, R. E., “Investigation of
discrepancies between the predicted and measured values are up to Acoustic Effects of Leading-Edge Serrations on Airfoils,” Journal of
1.2 dB at the largest serration wavelength. Altogether, the current Aircraft, Vol. 11, No. 4, 1974, pp. 197–202.
doi:10.2514/3.59219
statistical model can still be regarded as a robust tool for the
[5] Lau, A. S., Haeri, S., and Kim, J. W., “The Effect of Wavy Leading
predictions of the AGI broadband noise subjected to serrated LEs. Edges on Aerofoil–Gust Interaction Noise,” Journal of Sound and
Vibration, Vol. 332, No. 24, 2013, pp. 6234–6253.
doi:10.1016/j.jsv.2013.06.031
V. Conclusions [6] Chen, W., Qiao, W., Wang, L., Tong, F., and Wang, X., “Rod-Airfoil
An experimental aeroacoustic study was performed to quantify the Interaction Noise Reduction Using Leading Edge Serrations,” 21st
AIAA/CEAS Aeroacoustics Conference, AIAA Paper 2015-3264, 2015.
effects of five influencing parameters on the airfoil–gust interaction doi:10.2514/6.2015-3264
broadband noise of a NACA 65(12)-10 airfoil and the noise reduction [7] Custodio, D., “The Effect of Humpback Whale-Like Leading Edge
achieved by the serrated leading edges. For the statistical–empirical Protuberances on Hydrofoil Performance,” M.Sc. Thesis, Worcester
modeling, the design of experiments (DOE) technique was used to State Univ., Worcester, MA, Dec. 2007.
reduce the experimental volume to a manageable amount to gain [8] Melo De Sousa, J., and Camara, J., “Numerical Study on the Use of a
information on the interdependencies of each influencing parameter Sinusoidal Leading Edge for Passive Stall Control at Low Reynolds
and to develop a prediction tool that describes the overall noise Number,” 51st AIAA Aerospace Sciences Meeting, AIAA Paper 2013-
radiation. The model, initially based on 59 measurement points, was 0062, Jan. 2013.
validated to be accurate. It is then further stabilized by extensive doi:10.2514/6.2013-62
[9] Gawad, A. F., “Utilization of Whale-Inspired Tubercles as a Control
additional data set. It shows an accurate performance at settings close Technique to Improve Airfoil Performance,” Transaction on Control
to the defined central point of the experimental space and is only and Mechanical Systems, Vol. 2, No. 5, 2013, pp. 212–218.
slightly less accurate in the outer regions of the predefined setting [10] Paterson, R. W., and Amiet, R. K., “Acoustic Radiation and Surface
ranges. When the predicted results are compared with the external Pressure Characteristics of an Airfoil due to Incident Turbulence,”
data that were acquired in a separate experimental setting, the NASA CR-2733, Sept. 1976.
excellent agreement indicates that a robust and reliable statistical– [11] Staubs, J. K., “Real Airfoil Effects on Leading Edge Noise,” Ph.D.
empirical model has been developed in this study. The aeroacoustic Dissertation, Virginia State Univ., Blacksburg, VA, May 2008.
results allow the current paper to reach the following conclusions. [12] Kim, J. W., and Haeri, S., “An Advanced Synthetic Eddy Method for the
1) A clear ranking and quantification of the influencing Computation of Aerofoil–Turbulence Interaction Noise,” Journal of
Computational Physics, Vol. 287, April 2015, pp. 1–17.
parameters, where the Reynolds number Re and the freestream flow doi:10.1016/j.jcp.2015.01.039
turbulence intensity Tu are the main contributors to the broadband [13] Kim, J. W., Haeri, S., and Joseph, P. F., “On the Reduction of Aerofoil–
noise emissions. On the other hand, the serration amplitude A∕C, Turbulence Interaction Noise Associated with Wavy Leading Edges,”
followed by the Reynolds number, and the serration wavelength λ∕C Journal of Fluid Mechanics, Vol. 792, April 2016, pp. 526–552.
would represent the main factors for an effective broadband noise doi:10.1017/jfm.2016.95
reduction. [14] Turner, J. M., and Kim, J. W., “Aeroacoustic Source Mechanisms of a
2) Identification of a significant interdependence of the serration Wavy Leading Edge Undergoing Vortical Disturbances,” Journal of
wavelength and the freestream turbulence intensity (λ∕C ⋅ Tu) with Fluid Mechanics, Vol. 811, Jan. 2017, pp. 582–611.
doi:10.1017/jfm.2016.785
regard to the overall noise-reduction capability. This feature could be
[15] Chaitanya, P., Narayanan, S., Joseph, P., and Kim, J. W., “Leading Edge
linked to the characteristic size of the incoming gust relative to the Serration Geometries for Significantly Enhanced Leading Edge Noise
size of the serration wavelength. Reductions,” 22nd AIAA/CEAS Aeroacoustics Conference, AIAA
3) Identification of a significant interdependence of the angle of Paper 2016-2736. 2016.
attack and the serration wavelength (z∕H ⋅ λ∕C) with regard to the doi:10.2514/6.2016-2736
overall noise-reduction capability. This characteristic behavior could [16] Lyu, B., Azarpeyvand, M., and Sinayoko, S., “Noise Prediction for
be assigned to the three-dimensional effects when the flow is Serrated Leading-Edges,” 22nd AIAA/CEAS Aeroacoustics Conference,
approaching the airfoil and the location of the stagnation points for AIAA Paper 2016-2740. 2016.
the mean flow. The mechanism that causes an increased level of noise doi:10.2514/6.2016-2740
radiation at the low serration wavelength has also been suggested. [17] Vathylakis, A., Chong, T. P., and Kim, J. H., “Design of a Low-Noise
Aeroacoustic Wind Tunnel Facility at Brunel University,” 20th AIAA/
4) An algorithm to achieve the polyoptimum of low-absolute level CEAS Aeroacoustics Conference, AIAA Paper 2014-3288, 2016.
of noise radiation as well as high-level of noise reduction has been doi:10.2514/6.2014-3288
developed. This will serve as a first step toward practical applications [18] Laws, E. M., and Livesey, J. L., “Flow Through Screens,” Annual
to optimize the effective degrees of freedom in the serration design Review of Fluid Mechanics, Vol. 10, 1978, pp. 247–266.
process. doi:10.1146/annurev.fl.10.010178.001335
 3142 BIEDERMANN ET AL.

[19] Aufderheide, T., Bode, C., Friedrichs, J., and Kozulovic, D., “The Dissertation, Technical Univ. Dresden (TU Dresden), Dresden,
Generation of Higher Levels of Turbulence in a Low-Speed Cascade Germany, Nov. 2004.
Windtunnel by Pressurized Tubes,” Proceedings of 11th World [30] Narayanan, S., Joseph, P., Haeri, S., and Kim, J. W., “Noise Reduction
Congress on Computational Mechanics, 5th European Conference on Studies from the Leading Edge of Serrated Flat Plates,” 20th AIAA/
Computational Mechanics and 6th European Conference on CEAS Aeroacoustics Conference, AIAA Paper 2014-2320, 2014.
Computational Fluid Dynamics, Vol. 6, edited by E. Onate, J. Oliver, doi:10.2514/6.2014-2320
and A. Huerta, Barcelona, Spain, 2014, pp. 7521–7532. [31] Hansen, K. L., “Effect of Leading Edge Tubercles on Airfoil
[20] Rozenberg, Y., “Modélisation Analytique du Bruit Aérodynamique à Performance,” Ph.D. Thesis, School of Mechanical Engineering, The
Largebande des Machines Tournantes: Utilisation de Calculs Moyennés Univ. of Adelaide, Adelaide, Australia, July 2012.
de Mécanique des Fuides,” Dissertation, Ecole Centrale de Lyon/ [32] Clair, V., Polacsek, C., Le Garrec, T., Reboul, G., Gruber, M., and
Laboratoire de Mécanique des Fluides et d’Acoustique, Lyon, France, Joseph, P., “Experimental and Numerical Investigation of Turbulence-
Dec. 2007. Airfoil Noise Reduction Using Wavy Edges,” AIAA Journal, Vol. 51,
[21] Amiet, R. K., “Acoustic Radiation from an Airfoil in a Turbulent Stream,” No. 11, 2013, pp. 2695–2713.
Journal of Sound and Vibration, Vol. 41, No. 4, 1975, pp. 407–420. doi:10.2514/1.J052394
doi:10.1016/S0022-460X(75)80105-2 [33] Gruber, M., “Airfoil Noise Reduction by Edge Treatments,” Ph.D.
[22] Gershfeld, J., “Leading Edge Noise from Thick Foils in Turbulent Thesis, Faculty of Engineering and the Environment, Institute of Sound
Flows,” Naval Surface Warfare Center Rept. NSWCCD72-TR-2003/ and Vibration Research ISVR, Univ. of Southampton, Southampton,
099, West Bethesda, MD, 2003. England, U.K., Feb. 2012.
[23] Adam, M., Statistische Versuchsplanung und Auswertung (DoE Design [34] Roger, M., Schram, C., and Santana, L. de., “Reduction of Airfoil
of Experiments), Univ. of Applied Sciences Dusseldorf, Dusseldorf, Turbulence-Impingement Noise by Means of Leading-Edge Serrations
Germany, 2012, pp. 37–60. and/or Porous Material,” 19th AIAA/CEAS Aeroacoustics Conference,
[24] Clementi, S., Fernandi, M., Baroni, M., Decastri, D., Randazzo, G. M., AIAA Paper 2013-2108, 2013.
and Scialpi, F., “Mauro: A Novel Strategy for Optimizing Mixture doi:10.2514/6.2013-2108
Downloaded by 80.82.77.83 on September 11, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.J055633

Properties,” Applied Mathematics, Vol. 3, No. 10, 2012, pp. 1260–1264. [35] Clair, V., Polacsek, C., Le Garrec, T., Reboul, G., Gruber, M., and
doi:10.4236/am.2012.330182 Joseph, P., “Experimental and Numerical Investigation of Turbulence-
[25] Biedermann, T., “Aerofoil Noise Subjected to Leading Edge Serration,” Airfoil Noise Reduction Using Wavy Edges,” AIAA Journal, Vol. 51,
M.Sc. Thesis, Univ. of Applied Sciences Düsseldorf, Düsseldorf, No. 11, 2013, pp. 2695–2713.
Germany, Sept. 2015. doi:10.2514/1.J052394
[26] Biedermann, T., Chong, T. P., and Kameier, F., “Statistical–Empirical [36] Oerlemans, S., and Migliore, P., “Aeroacoustic Wind Tunnel Tests of
Modelling of Aerofoil Noise Subjected to Leading Edge Serrations and Wind Turbine Airfoils,” 10th AIAA/CEAS Aeroacoustics Conference,
Aerodynamic Identification of Noise Reduction Mechanisms,” 22nd AIAA Paper 2004-3042, 2004.
AIAA/CEAS Aeroacoustics Conference, AIAA Paper 2016-2757, 2016. doi:10.2514/6.2004-3042
doi:10.2514/6.2016-2757 [37] Geyer, T., Sarradj, E., and Giesler, J., “Application of a Beamforming
[27] Kleppmann, W., Versuchsplanung—Produkte und Prozesse Technique to the Measurement of Airfoil Leading Edge Noise,”
Optimieren, 9th ed., Hanser, Munich, 2016, pp. 209–233. Advances in Acoustics and Vibration, Vol. 2012, No. 3, 2012, pp. 1–16.
[28] Siebertz, K., van Bebber, D., and Hochkirchen, T., Statistische doi:10.1155/2012/905461
Versuchsplanung, Springer, Berlin, 2010, pp. 3–24, Chap. 1.
[29] Haase, D., “Ein Neues Verfahren zur Modellbasierten Prozessoptimierung L. Ukeiley
auf der Grundlage der Statistischen Versuchsplanung am Beispiel Associate Editor
eines Ottomotors mit Elektromagnetischer Ventilsteuerung (EMVS),”