3rd Ankara International Aerospace Conference


Feb 22, 2014 (7 years and 8 months ago)




Mehmet Çavuş

Istanbul Technical University
Istanbul, Turkey

Ergin Arslan

Istanbul, Turkey

Esra Sorgüven

Yeditepe University
Istanbul, Turkey

Aydın Mısırlıoğlu

Istanbul Technical University
Istanbul, Turkey


MSc student in Aerospace Engineering, Email: cavusme@itu.edu.tr

R&D Specialist, Email: ergin.arslan@arcelik.com

Assistant Professor in Mechanical Engineering, Email: sorguven@yeditepe.edu.tr


Associate Professor in Astronautical Engineering, Email: misirli@itu.edu.tr

The aim of this paper is to present a computational
method to predict the aeroacoustic noise in ducted
fans. Here, the method is applied to an axial fan
running at 1980 rpm in a circular duct. Two
commercial softwares, FLUENT and LMS
SYSNOISE, are employed for this purpose.
Unsteady flow analysis is performed with Large Eddy
Simulation (LES) using FLUENT and then passed to
LMS Sysnoise in order to compute the acoustic
radiation. The numerically obtained acoustic field
around the duct is compared with the experimental
data and fairly good agreement is observed.

Aeroacoustic noise generation and propagation
attracts ever increasing interest of both academic
and industrial community [2, 3]. The control of the
aeroacoustic noise is a principal concern in different
industries such as aerospace, automotive, and home
appliances. Usually, the most significant contribution
to the noise level comes from fans. Especially in
home appliances, the small and medium sized axial
and radial fans are the dominant noise emitters [7].
Recently, with the exponential growth in the
computer technology, computational methods in
aeroacoustics became a powerful tool for noise
prediction [6, 10]. This paper presents a
computational aeroacoustics method, which involves
the coupling of a flow solver and a wave-equation
solver. With the help of CFD, the time-dependent
turbulent flow data is obtained. Depending on
turbulent fluctuations, noise sources in the flow field
are computed. The propagation of the emitted noise
is computed via Helmholtz equation in interior and
exterior domains. Thus, the sound pressure level
distribution in both near and far fields is obtained.
A major advantage of this method is that, the whole
sound pressure level spectrum is calculated directly.
There is a rich literature concerning the prediction of
the discrete noise at the blade passing frequency or
the broadband noise [8, 9]. However, there is a lack
of a method, which can predict the sound pressure
level (SPL) at every frequency in the range of
interest as a continuous curve. The originality of the
presented method arises from the capability of
predicting the SPL distribution over the whole
frequency range of interest. SPL at the blade
passing, cavity resonance and structural resonance
frequencies are predicted with one computation.
The numerical computations are performed in mainly
two steps. In the first step, the turbulent flow is
computed with Large Eddy Simulation (LES). LES is
a good compromise for the sake of the accuracy and
CPU-time in comparison with Direct Numerical
Simulation (DNS) and Reynolds Averaged Navier
Stokes Equations (RANS) [4]. The commercial flow
solver FLUENT is employed for this task.
In the second step, the time dependent flow data is
fed into the aeroacoustic module where the wave
equation is solved and the noise propagation is
computed. Structural modal analysis is also
performed and taken into account. Sound
propagation is computed by the LMS SYSNOISE.
Not only the tonal noise radiation, representing Blade
Passing Frequency (BPF) and its harmonics, but
also the excitation at the cavity and structural
resonance frequencies are investigated. As a result,
the sound pressure and intensity levels are predicted
AIAC-2005-092 Cavus & Arslan

in near and far field. In order to validate the
computational method, numerical results are
compared with experimental data. Both the
aerodynamic and the aeroacoustic steps of the
method are validated.
In this paper, the method is applied to an axial fan,
which is running in a circular duct. Reynolds number
based on the velocity and the chord length at the tip
is about 72,000 and the tip Mach number is about
0.04. Thus, a turbulent, incompressible flow is
The comparison of numerical results with
experiments indicates that the proposed
methodology inherits a high accuracy and potential
to apply to more complicated systems.

The CAD model of the investigated test case is
shown in figure 1. The circular duct has an inner
diameter of 156 mm and a length of 960 mm. The
outer diameter of the fan is 125 mm. The axial fan
with a tip chord length of 77.5 mm, is running at 1980
rpm, which corresponds to a blade passing
frequency of 132 Hz. A motor, which is placed
upstream the fan, is driving the fan.

Figure 1: The CAD model of the test case

To predict the aeroacoustic signal in the far field,
first, the flow induced acoustic sources are computed
with a transient CFD analysis. Next, the computed
acoustic sources are transferred to an acoustic
module to compute the propagation of sound waves.
In the following, the CFD and aeroacoustic models
are described.
Computational Fluid Dynamics (CFD):
The finite
volume method based commercial flow solver
FLUENT is employed for the CFD analysis. FLUENT
is capable of solving unsteady, incompressible flows
on unstructured grids with different turbulence model
[14]. For Computational Aeroacoustics (CAA), LES is
preferred to Direct Numerical Simulation (DNS) and
to Reynolds Averaged Navier-Stokes (RANS)
simulation [5]. With the available computational
sources today, DNS is restricted to low Reynolds
number flows with simple geometries since it
requires very fine grid resolution and therefore very
high CPU-time [1]. RANS method is the fastest
approach, however fails in accuracy due to the fact
that it is based on averaging the flow variables [4].
LES is capable of resolving large scales of the flow,
which are more efficient than the small ones in
generating sound [11]. Thus, LES is the most
appropriate method for CAA.
The fan motion is modeled with the sliding mesh
approach, which is implemented in FLUENT. This
model is chosen in order to simulate the relative
motion between the rotating fan and the stationary
motor appropriately. The result of a steady Multiple
Reference Frames (MRF) simulation is employed as
an initial guess for the unsteady sliding mesh
simulation in order to speed up the convergence.
The time discretization scheme is second order and
implicit. The pressure-velocity coupling is calculated
with the SIMPLE algorithm. Bounded central
differencing scheme is used for discretization of the
convection. The boundary conditions employed in
the simulations are total pressure at the inlet and
static pressure at the outlet, which are equal to
101325 Pa. The subgrid scale model employed in
LES is the Smagorinsky-Lilly model, which is an
algebraic turbulence model.
Two simulations are performed with different
computational grids, having different resolutions. The
grid of the first simulation consists of approximately
570.000 control volumes, which is coarse for an
LES. However, it is aimed to establish a basis for
modeling flow-induced noise in complex geometries
with this simulation. The second simulation involves
a finer grid with about 2.2×10
control volumes. The
simulation with the coarse grid results in
dimensionless wall distance, y
, values as about 20,
whereas the fine grid decreases the y
values to
about 5. The figure below shows the surface grid on
the fan of the second simulation.

Figure 2: The computational grid on the fan (totally
about 2.2×10
control volumes)

The time step is set to match the required
aerodynamic and acoustic time resolutions. The time
step is 10
s, which corresponds to a rotation of 1.2
degrees. A complete revolution of the fan is
performed each 300 time steps which takes
approximately 10 hours with 8 processors of SGI

Ankara International Aerospace Conference
AIAC-2005-092 Cavus & Arslan

Origin 2000 for the coarse grid simulation. It extends
to about 40 hours of CPU-time for the fine grid.
When the statistically steady-state condition is
reached, acoustic data sampling is started. This
typically takes five complete fan revolutions.
In the following, results of the LES are presented.
Figure 3 shows the instantaneous pressure
distribution on the suction and pressure sides of the
fan. Pressure contours on the blades are not axi-
symmetric, which is due to the asymmetric
positioning of the motor upstream the fan. The
contours especially on the pressure side exhibit
vortical structures, which are time-dependent. The
pressure fluctuation on any point on a blade is about
80 Pa.

Figure 3: Instantaneous pressure distribution on both
sides of the fan

Figure 4: Instantaneous vorticity distribution along
the pipe cross-section (upper: coarse grid, lower: fine

The instantaneous vorticity distribution along the
pipe cross-section is shown in figure 4 to point out
the dependency on grid resolution. The upper
snapshot is from the simulation with coarse grid
whereas the lower snapshot belongs to the
simulation of fine grid. Both simulations result in a
similar vorticity range. However, the vorticity levels
are still high downstream the fan in the fine grid
case. Thus, turbulent structures are not dissipated
immediately as in the coarse grid case; but carried
further with the flow.
The aeroacoustic modeling is
performed with the aeroacoustic module of the
vibroacoustic solver LMS Sysnoise. Sysnoise is
capable of solving wave equation in interior and
exterior domains with different discretization
techniques like Boundary Element Method (BEM)
and Finite Element Method (FEM) [13].
The input for aeroacoustic module is time-dependent
pressure and velocity data, which is obtained from
the CFD solution. The flow data from FLUENT is
used to calculate the acoustic source term, which is
on the right hand side of the wave equation. The
aeroacoustic grid, on which the wave equation is
discretized, is much more uniform and coarser than
the CFD grid. The interpolation between two grids is
performed by an in-house developed post processor.
The turbulent flow around a ducted axial fan induces
three types of acoustic sources: monopoles, dipoles,
and quadrupole [12]. In this test case, the dipoles are
the dominant sources since the fan blades are thin
and the flow is subsonic. Therefore, only the dipoles
are taken into account. On the fan blades, the
dipoles are rotating, which are responsible for the
tonal noise at the Blade Passing Frequency (BPF)
and its harmonics. The contribution of the rotational
dipoles is computed based on the force fluctuations
on one blade.
In order to model the interior and the exterior
domains simultaneously, the Multi-Domain BEM
analysis is performed. The analysis consists of two
models, which are the Direct BEM Interior, and the
Direct BEM Exterior models. Both models are linked
at the openings of the duct, through a fluid-fluid
coupling. The coupling automatically satisfies the
boundary condition at the openings, which is
equivalent to ambient pressure boundary condition.
The boundary condition applied on the duct surface
is the rigid wall boundary condition. The stationary
dipole sources on the duct surface are defined as
discrete sound sources on the nodes of the acoustic
An additional coupled structure - fluid model is set up
in order to simulate the tonal fan noise at the BPF
and its harmonics and the dipole structural
contribution. In this coupled model, the inputs to
Sysnoise are the force fluctuations on a fan blade
and the structural modes of the duct from a structural
modal analysis.
To finalize the aeroacoustic computation, the results
of the Multi-Domain BEM model and the coupled
structure - fluid model are superposed. The achieved
aeroacoustic solution involves:


Ankara International Aerospace Conference
AIAC-2005-092 Cavus & Arslan

• Tonal fan noise at BPF and its harmonics
• Excitation of the cavity modes
• The dipole structural contribution
Structural Modal Analysis:
The experimental setup
does not include any structural attachments from the
fan to the duct directly. The structural contribution
arises only from the dipoles and is very weak
compared to the fan source and to the dipolar direct
In order to identify the natural frequencies of the
duct, which is made of steel, structural modal
analysis is performed and validated. The
experimental and numerical results are in good
agreement as shown in figure 5.

100 200 300 400 500 600 700 800
Frequency [Hz]
FRF [m/s^2/N]
Figure 5: Comparison between experimental and
numerical results of structural modal analysis

Acoustical Modal Analysis:
In order to determine the
natural frequencies of the duct cavity, an additional
analysis is performed with the LMS Sysnoise and it
is experimentally validated. In the numerical model, a
multi domain BEM model is set up and the natural
frequencies of the duct cavity are excited by a
discrete monopole source. The strength of the
source is specified arbitrarily. The experimental and
numerical results are given in figure 6.
100 200 300 400 500 600 700 800
Frequency [Hz]
SPL [dB] ref = 20 uPa

Figure 6: Comparison between experimental and
numerical results of acoustical modal analysis

Two sets of acoustic measurements are performed
in the anechoic room. First, sound pressure levels at
different microphone positions in the acoustic far
field around the duct are measured in narrow band
spectrum. Secondly, the sound pressure and
intensity are mapped in 1/3 octave band frequencies
in order to determine the directivity of sound on a
measurement field. The sketch of the experimental
setup for intensity measurement is illustrated in
figure 8. The duct is placed at the center of the
measurement field and the acoustic data is obtained
by an intensity probe, which collects pressure data
on the nodes of the measurement field.
Unfortunately, the experimental setup does not allow
the probe to scan the under hood of the system
because of the supporting boxes as shown in figure

Figure 7: SPL measurement setup

Figure 8: Sketch of the sound intensity measurement

The numerical and the experimental color maps of
the sound pressure level distribution in 1/3 octave
band are presented in the figures below. These
figures designate the directivity of sound at 125 Hz
and 250 Hz. The distributions in both numerical and

Ankara International Aerospace Conference
AIAC-2005-092 Cavus & Arslan

experimental maps are similar. The discrepancies
can be attributed to the experimental setup. In the
numerical analysis, only the fan and the duct are
taken into account. The wooden box, supporting the
circular duct, is neglected in the computations. Since
the reflections due to the box are not considered, the
numerical results exhibit a perfect symmetry around
the duct. However, the reflective surfaces under the
duct cause asymmetry and higher noise levels.

Figure 9: Numerical sound pressure level distribution
at 125 Hz. (1/3 Octave band)

Figure 10: Experimental sound pressure level
distribution 125 Hz (1/3 Octave band)
Figure 11:Numerical sound pressure level distribution
at 250 Hz (1/3 Octave band)

Figure 12: Experimental sound pressure level
distribution at 250 Hz (1/3 Octave band)

Figure 13: The experimental and numerical sound
pressure level distribution (coarse grid) at a point in
far field (narrow band, ∆f=2 Hz)


Ankara International Aerospace Conference
Figure 13 represents the sound pressure level
distribution in the narrow band at a far field point,
with a distance of about 0.75m from the center of the
fan. The numerical curve sketched in this figure is a
AIAC-2005-092 Cavus & Arslan

result of the LES with the coarse grid. The sound
pressure level of the tonal noise at the BPF (132 Hz)
is predicted with a high accuracy by the numerical
model. The first cavity mode is at 162 Hz, and the
second mode is at 324 Hz. Both cavity modes are
visible in both numerical and experimental data. The
tendency of both curves are similar, however there is
a deviation in the amplitudes. The numerical SPL
values are below the experimental values.
The amplitude of the numerical SPL-curve
decreases for frequencies higher than 350 Hz
rapidly. This is caused due to the insufficient
resolution of turbulent structures in the flow. Since
the spatial filter size in LES is too large to capture
the small-sized, high frequent turbulent structures,
the accuracy in noise prediction at high frequencies

Figure 14: The experimental and numerical sound
pressure level distribution (fine grid) at a point in far
field (narrow band, ∆f=2 Hz)

The acoustic result of the LES with the fine grid
shows a better agreement with the experimental
data, as seen from figure 14. Finer grid resolution
enables the simulation of a greater range of turbulent
scales, and results in a higher accuracy for a large
range of frequencies.
The SPL spectrum shows peaks at
• The BPF (132 hz)
• The first cavity mode of the duct (162 Hz)
• The second cavity mode of the duct (324 hz)
• The third cavity mode of the duct (486 Hz)
No peaks at the structural resonance frequencies are
visible since the duct is made of steel and the
acoustic sources are too weak to excite the

Flow induced noise generated by an axial fan
running in a circular duct is predicted. The time-
dependent turbulent flow is simulated with FLUENT
and the aeracoustic field is computed with the LMS
SYSNOISE. The comparison between the numerical
and the experimental data shows a good agreement,
and indicates that the numerical method can be
applied to more complex geometries.

[1] Andersson, N., Eriksson L.E., Davidson L.,
Investigation of An Isothermal Mach 0.75 Jet and Its
Radiated Sound Using Large Eddy Simulation and
Kirchoff Surface Integration, Int. J. of Heat And Fluid
Flow, Vol. 26, p: 393-410, 2005.
[2] Maaloum, A., Koudri, S., Rey R., Aeroacoustic
Performance Evaluation of Axial Flow Fans Based
on the Unsteady Pressure Field on the Blade Surface,
Applied Acoustics, Vol. 65, p: 367-384, 2004.
[3] Envia, E., Wilson, A., Huff, D.L., Fan Noise: A
Challenge to CAA, Int. J. of Computational Fluid
Dynamics, Vol. 18, p: 471-480, August 2004.
[4] Sorgüven, E., A Computational Aeroacoustic Method
Using Large Eddy Simulation And Acoustic Analogy,
Phd thesis, University of Karlsruhe, Germany, 2004.
[5] Magagnato, F., Sorgüven E., A., Gabi M., Far Field
Noise Prediction by LES and FWH Analogy, 9

AIAA/CEAS Aeroacoustics conference and exhibit,
South Carolina, 2003.
[6] Tournour, M., El Hachemi, Z., Read, A., Barone F.,
Investigation of The Tonal Noise Radiated By
Subsonic Fans Using The Aeroacoustic Analogy,
Proceedings of Fan Noise Symposium, CETIM
Senlis, 2003.
[7] Chung, K., Kim, K.Y., Na S.U., Development of Fan
Design Program for High Performance and Low
Noise Fan, Proceedings of Fan Noise Symposium,
CETIM Senlis, 2003.
[8] Cho, Y., Moon, Y.J., Discrete Noise Prediction of
Variable Pitch Cross-Flow Fans by Unsteady Navier-
Stokes Computations, Journal of Fluids Engineering,
Vol. 125, p: 543-550, 2003.
[9] Hanson, D., Broadband noise source studies for a fan
with coupled rotor/stator, 9
Aeroacoustics conference and exhibit, Hilton Head,
South Carolina, 12-14 May 2003.
[10] Mendonça, F., Allen, R., Charentenay, J., Lewis, M.,
Towards Understanding LES and DES for Industrial
Aeroacoustic Predictions, International Workshop on
‘LES for Acoustics’, DLR Göttingen, Germany,
[11] Mankbadi, R., Review of Computational
Aeroacoustics in Propulsion Systems, Journal of
Propulsion and Power 15 (4), p: 504-512, 1999.
[12] Goldstein, M.E., Aeroacoustics, McGraw-Hill Book
Company, New York, 1976.
[13] LMS Sysnoise Rev 5.6: Computational
Vibroacoustics, User’s Manual, and LMS
[14] Fluent 6.2, User’s Guide, FLUENT Inc.

Ankara International Aerospace Conference