Buffeting Noise Computations for an Open Side Window


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


Page 1 of



In this paper, side window buffeting analysis of
the Sedan car model is studied computationally.
Flow over an open side window in car exhibits
similar characteristics as the flow over an open
cavity. Hence, a grazing flow simulation over an
overhang cavity is done as a benchmark. The
benchmark results, frequency and sound pressure
levels of feedback and resonance modes, matched
well with the available experimental data. The
analysis is done in two parts. First a steady state
solution is obtained using Reynolds Averaged
Navier Stokes (RANS), and then the computed
flow field information is used as input for Non
Linear Acoustic Solver (NLAS). After validation,
the buffeting spectrum generated from flow over
open front side window of generic Sedan car
model is analyzed. Vortex generation at the root
of A-pillar, its propagation and growth
downstream and its breaking mechanism at the B-
pillar are well captured. The obtained frequency
of buffeting at the location of the driver's ear and
the corresponding sound pressure level (SPL) are
as expected when compared to general Sedan car
side window buffeting spectra. The acoustics
computations have been performed on the
supercomputer Eka in 70 hrs real time on 15
million cells mesh using 128 processors. The
scalability studies have also been carried out for
the acoustic calculations.

Keywords: Cavity noise, Buffeting, Feedback
resonance, Helmholtz resonance, CFD
Buffeting is a type of cavity noise when the entire
passenger compartment acts as a cavity in case of
an open sunroof or a side window. (Rossister,
1964, Ota et al., 1994). The unstable shear layer
generated at the upstream edge of the cavity is the
cause of the cavity noise. Disturbances (in the
form of vortices) are created at the front edge of
the opening (in the present case the A-pillar) and
are convected along the flow. When they strike
the rear end (the B-pillar) a pressure wave is
generated which propagates inside and outside
the passenger compartment (the cavity). When
the wave reaches the front end it triggers another
shedding of the disturbance. This periodic
occurrence results in the shear layer generating a
specified buffeting frequency. In automobiles this
frequency is very low (~ 25Hz) but the intensity
is very high (> 100 dB) (Sovani et al., 2003). This
fluctuation is felt by the passengers as a pulsating
force which can be very troublesome. Hence, it is
important to keep buffeting in view while
designing an automobile and also to understand
the underlying physics of cavity noise.

In the past, ample amount of work is done to
understand aerodynamic noise generated by flow
over open cavities. Rossister, 1964, had done a
pioneering work, where he did extensive

Buffeting Noise Computations for an Open Side Window

Ashish Singh
Deepanshu Rajvanshi

CRL Ltd,
CRL Ltd,
Pune, Maharashtra, India.
Pune, Maharshtra, India.



Kishor Nikam
CRL Ltd,
Pune, Maharashtra, India.

Proceedings of the 37th National &4th International Conference on Fluid Mechanics and Fluid Power
December 16-18,2010,IIT Madras,Chennai,India.
FMFP10 - AM- 13

Page 2 of



experiments to understand the physics of noise
sources emanating from flow over open cavities.
It was observed that apart from random pressure
fluctuations, discrete tones related to feedback
loop and resonance with cavity mode were
present for different length to depth ratio of
cavity. Another example is the case of flow over
overhang cavity (Henderson, 2004), where both,
feedback and cavity mode, are seen to be present
at the same time. Cavity modes occur when the
frequency of periodic oscillation matches with
Helmholtz cavity mode (Howe, 1997). The sound
pressure level exceeds 110 decibels due to these
dominating modes. The noise produced by
overhang cavity at low speeds has resemblance
with aerodynamic noise generated by wheel bay,
weapon bay, sunroof and side window buffeting
depending upon different free stream conditions.

In case of automobiles, the car compartment acts
as a cavity and these modes emit temporally
varying forces of pressure fluctuations of high
intensity and low frequency, which is in turn
known as buffeting. In automotive engineering,
buffeting refers to an undesirable noise. This
causes passenger discomfort which is the primary
cause of investigation of buffeting. Similar to
overhang cavity noise level, buffeting noise
emanating from open sunroof or open side
window of a passenger car may exceed 110 dB.

Ota et al., 1994, performed an early CFD
simulation of sunroof buffeting (a type of cavity
noise), where a 2D flow-field along the symmetry
plane of a passenger car was considered. Another
early 3D sunroof buffeting study was carried out
by Ukita et al., 1997, on a larger model size. With
increase in the computational capacity, more
detailed CFD models have been studied. Karbon
et al., 2002, did sunroof buffeting studies using
complex vehicle models that included detailed
representation of the passenger compartment.
Recently, Ashish et al., 2008, performed 2D
CFD/CAA simulations for general sedan car
sunroof buffeting, and, detailed the number of
modes of oscillations based on Rossister’s
formulation. Also, the focus was to modify rear
end of sunroof for noise suppression rather than
improving the front end. Significant improvement
is shown in noise levels by avoiding vortex
breakdown at rear end.

Sovani et al., 2003, and Chang-Fa, 2004,
performed a number of simulations to study side
window buffeting and the effect of various
parameters like velocity, yaw angle, receiver
location and turbulence models on the peak
frequency and SPL. Their study found out that
CFD predictions of buffeting SPL and frequency
matched well with the experimental results. SPL
was predicted within an accuracy of 4 dB and the
frequency within 1 Hz. Moreover, the trends of
buffeting frequency and SPL with the variation of
other parameters also matched well with the
experimental results.

Conducting experiments for acoustic noise
measurements for full car model like SEDAN Car
are highly cost and time intensive. To overcome
experimental set up challenges, researchers are
now using Computational Aero Acoustics (CAA)
approach in conjunction with Computational
Fluid Dynamics (CFD) to simulate aerodynamic
noise and sound pressure level. CAA simulation
requires time dependent computations on highly
dense grid near sound sources as well as in the
computational domain. These simulations are
highly compute intensive which is provided by
High Performance Computing (HPC) platform
like CRL’s supercomputer facility, EKA.

The purpose of the present study is to perform
computational aeroacoustic simulations with high
accuracy and high end scalability.

This study goes through baseline validation first
(which increases the confidence in CAA
methodology), for which the test case (Fig. 1a) is
selected from NASA’s 4
CAA workshop in
2004. The detailed experimental data is available
from Henderson, 2004, to compare the results. In
the present paper, comparison is done with the
work of Henderson, 2004.

Page 3 of



After baseline validation, the present paper
investigates side window buffeting noise. Sedan
car vehicle configuration is considered in which
front side window is kept completely open. The
methodology for predicting the buffeting
frequencies and the corresponding sound pressure
levels (SPL) is discussed first followed by a
discussion of meshing considerations and solver
settings. The simulations were carried out using
CFD++ and CAA++. A speed up study is also
done for CAA simulations to quantify effects of
parallel computing on simulation time.

The overhang cavity as a benchmark problem is
shown in Fig. 1a. The dimensions are taken to be
similar as used in experiments (Henderson,
2004). Flow is fully turbulent over the cavity with
inlet free stream velocity of 50m/s. Three receiver
points, at the center of left, right and bottom
cavity wall (Fig. 1b) are considered to validate
frequency spectrum with experimental results.


Fig. 1: (a) Benchmark problem at 4
workshop: Overhang Cavity (b) Receiver

Overhang cavity is the practical example of
producing discrete tones related to longitudinal
waves (fluid dynamic oscillation) and transverse
waves (Helmholtz resonance) for low Mach
number flows. The frequency of fluid dynamic
oscillations is approximated by Rossister, 1964,

￿￿￿￿￿￿￿￿￿￿ kMnLUfs

Page 4 of




Page 5 of



(represented by '). Statistical mean (RHS) can be
obtained by steady RANS simulations whereas
the equation can be solved for perturbations
present in LHS of the equation. NLAS, like
CFD++, uses a finite-volume spatial
discretization, in which a continuous integral of
the conservation variable fluxes over an arbitrary
control volume is replaced via a discrete
representation of the flux across each face.
Spatial derivatives, used to compute both viscous
and inviscid flux terms, are first computed via a
least-squares approximation at the element or
control-volume vertices. Then a polynomial
representation of solution is constructed, whose
integral value matches that of the initial
(piecewise constant) cell volume data.
The smallest wavelength structures are modeled
and so there is no coupling or feedback from the
highest frequency acoustics waves, but large
scale vortical structures can be resolved directly
and those larger scales will interact with any
acoustic waves in the system.
The time step for NLAS computation is set to be
, which provides sufficient stability on
existing mesh. This time step is not small enough
to resolve all higher frequencies; however, it is
acceptable for this kind of problem where first
few peaks corresponding to longitudinal and
transverse waves are dominant. The reduced
domain for NLAS solver is decomposed into 4
parts to run it in parallel on 4 processors. Pressure
signals are recorded at specified receiver
locations and Fast Fourier Transform (FFT) tool
is used to get frequency spectrum from simulated
data. Baseline pressure history is shown in Fig. 4
a-d for all three receiver locations. The Sound
Pressure Level (SPL) is calculated using
following relation
￿￿ ref
Where, p is the perturbation amplitude and p
20e-06 Pa.
The frequency and SPL are compared with the
work of Henderson, 2004. Frequency and SPL,
obtained from pressure history (Fig. 4a) are
compared in Table 1. Figures 4 b-d shows the
noise spectrum obtained at right left and bottom
centers respectively.





Page 6 of



Fig. 4: (a) Pressure History (b) Spectra at right
center (c) Spectra at left center (d) Spectra at
bottom center

Table 1: Comparison of baseline results with

In the present study, first tone at 1678 Hz is very
close to experimental prediction, 1624 Hz, but
SPL is over predicted by 15-20 dB, while
compared with test data (Henderson, 2004). The
reason for this over prediction in SPL may be
attributed to simulation of a practical 3D problem
with 2D approach where relaxation effects in the
span, coupled with physically spherical spreading
of acoustic wave radiation causes a higher
attenuation of these wave energies. There are no
discrete tones observed corresponding to
feedback loop resonance and cavity resonance
(Helmholtz mode) but a single tone around 1678
Hz is numerically obtained. In experiments,
frequency of these tones was found to be 1624
and 1504 Hz respectively (at all receiver
locations). It can be attributed that in present
numerical analysis, both higher tones, feedback
and cavity resonance, are coinciding at 1678 Hz.
Henderson, 2004 also stated in his work that the
first cavity mode of this kind of cavity occurs at
3016 Hz which is based on the assumption that
acoustic wavelength is four times the cavity
depth. This can be obtained from the Helmholtz
Eq. (2). In the same context, the second mode of
cavity can be assumed half of the 3016 Hz
(1503). Obtained frequency 1678 Hz (for cavity
mode) is closer to the second mode 1504 Hz in
experiments. Hence frequencies in current
analysis (1678 Hz) are very close to experimental
value (1624 Hz, 1504 Hz) for feedback and
cavity resonance. Two separate peaks
corresponding to two different tones may be hard
to predict numerically if they are resonating with
each other.
Only one vehicle configuration is considered in
which the side window corresponding to the A-
pillar is kept completely open as shown in Fig 5.
Inside features of the car, e.g. chairs and human
dummies are not considered here. So the model is
kept as simple as possible to behave like a hollow
cavity. The methodology for predicting the
buffeting frequencies and the corresponding
sound pressure levels (SPL) is discussed first
followed by a discussion of meshing
considerations and solver settings. The
simulations were carried out using CFD++, a
finite volume based solver.

Fig.5: Sedan Car model with front window open

Frequency (
SPL (dB)
1624 118

Page 7 of



In the present study, the total mesh size is 15
million cells without boundary layer mesh. The
vehicle is placed inside a parellopiped
computational domain (as shown in Fig 6). The
inlet and outlet of the domain are at an
approximate distance of 3 and 6 body lengths
from the front and rear face of the model
respectively. This is chosen so that the incoming
flow is developed to a good extent and the eddies
downstream of the vehicle are captured well. The
passenger compartment of the car is connected to
the external flow via the open window fluid layer.
The vehicle surface is meshed in detail (Fig 7 a,
b) to capture the development of the flow.
Especially, the areas of A-pillar, open window
and B-pillar are modeled to a great detail as
shown by Fig 7 b-c. The cell edge length in these
areas is around 3 mm, sufficient enough to
resolve the wavelength of interest. This spacing
insures at least 5 cells per wavelength. In the
passenger compartment the edge length is around
15 mm. The model does not include underhood
and underbody details as its effect on the
buffeting frequency is assumed to be negligible.
Fig 7a shows the surface over the model. Fig 7b
shows the area where the mesh is refined in order
to capture the shear layer phenomena. Fig 7c is
the enlarged view of Fig 7b.

Fig. 6: Computational domain with boundary

(a) Surface mesh of the model
(b) Refined mesh near the open window

(c) Magnified view of the selected area in
Fig.7: Meshing details
The two equation cubic k-
turbulence model is
used for the steady RANS closure. This case was
run on 64 processors for 500 iterations in which
Side wall
Top wall
Car body

Page 8 of



convergence (of the order of 10
) was achieved.
The approximate turnaround time for this
simulation was around 8 hrs. The steady state
data obtained is then used to initialize the
unsteady runs (NLAS). For the NLAS run, the
time step chosen was 0.0004s. This transient
simulation used 128 parallel processors and the
turnaround time was around 70 hrs.

A.Steady State
The pressure contours in the form of coefficient
of pressure (Cp) for the steady state simulation
are shown in Figs 8 a, b. Cp distribution is as per
expectation but in the absence of any
experimental data it is difficult to comment more
on it.

(a) Front view

(b) Rear view
Fig. 8: Pressure Contours over the car

Figures 9 (a-f) shows the development of the
vortex at the A-pillar region of the open window,
its gradual convective growth and finally it’s
breaking on the B-pillar. Pictures are shown at
cut section planes placed 0.2 m apart. It is
observed that the inception of the instability at the
lower end of A-pillar contains very high energy
in comparison to others generating from length of
A-pillar. Root cause of buffeting is the breaking
mechanism of this instability in the form of high
energy vortex hitting at the B-pillar, emission of
pressure waves (acoustic waves) at B-pillar, their
resonance with different harmonics of Helmholtz
mode of car cavity, and new instability generation
at the A-pillar again to continue the process.




Page 9 of




Fig.9: Convection and development of the
vortex, cut planes with an interval of 0.2 m, a-f

Vortex breakdown at B-pillar is captured in more
detail in figure 10.

Fig.10: Vortex breaking at B-pillar
B.Transient Analysis
NLAS solver is used for transient CAA
simulations. Pressure data is recorded at every
time step and FFT is applied on the acoustic
pressure obtained at receiver location. The 12

octave band produced spectrum of SPL vs.
frequency as shown in Fig 11. The peak SPL,
124dB corresponding to 28 Hz is as expected
when compared to available data.

Fig. 11: SPL vs. frequency plot showing peak
SPL and the corresponding frequency

All simulations have been performed in parallel
on the supercomputer Eka. Figure 12 shows ideal
and actual speed up curves for 64 to 512 CPUs.
In the next plot (figure 13) the parallel efficiency
F~ 28 Hz, SPL ~ 124 dB

Page 10 of



of the solver is shown for Eka architecture. For
the defined range of CPUs, speed up is almost
parallel to ideal with over 90% efficiency for
entire CPU range.

Fig. 12: Ideal and actual speed up of CFD++ on
Eka architecture

Fig. 13: Parallel efficiency of CFD++ on Eka
Computational Aero Acoustic validation is done
using overhang cavity problem. Feedback and
Helmholtz modes are captured at the same
frequency with less than 3% error when
compared to experiments. Sound Pressure Levels
were over predicted and the root cause is
attributed to 2D simulation of a practical 3D
turbulent problem. Furthermore, side window
buffeting analysis of the Sedan car model using
CFD and CAA has been investigated in this
paper. The methodology, corresponding grid
resolution, solver settings were discussed in
detail. Predicted buffeting frequency and SPL are
closer to buffeting levels of general sedan car
model (28 Hz, 124 dB). Parallel simulations of
CFD and CAA methods accelerated the
turnaround time for complete CFD/CAA analysis.
Full transient simulation of 15 million mesh with
sufficient time resolution took around 70 hrs on
128 processors to generate the noise spectrum.

We are thankful to Metacomp Technologies for
their coherent support in this work. Also, we
would like to thank Computational Research
Laboratories for providing Supercomputing
facility, Eka.
Ashish Singh, Saravana Kumar, J. S. Rao., 2008.
Numerical Analysis of Sunroof Buffeting, SAE
paper 2008-28-0059.

Batten, P, Ribaldone, E, Casella, M,
Chakravarthy, S, 2004. Towards a Generalized
Non-Linear Acoustics Solver, AIAA 2004-3001,
AIAA/CEAS Aeroacoustics Conference.

CFD++ User Manual, Version 7.5, 2008.
Metacomp Technologies, USA.

Chang-Fa An, Alaie S.M, Sovani S.D, Scislowicz
M.S and Kanwerdip Singh., 2004. Side Window
Buffeting Characteristics of an SUV, SAE paper,
0 100 200 300 400 500 600
Speed up
Number of CPUs
Speedup Curve
Ideal Speed Up
Actual Speed Up
0 100 200 300 400 500 600
Parallel Efficiency
Number of CPUs
Parallel Efficiency vs No. of CPUs
Parallel Efficiency

Page 11 of



Henderson, B., 2004. Problem 2 of Category 5 –
Sound Generation by Viscous Flows, 4

Computational Aero Acoustics (CAA) Workshop
on Benchmark Problems, NASA/CP-2004-

Hendriana D, Sovani, S.D and Schiemann M.K.,
2003. On Simulating Passenger Car Side Window
Buffeting, SAE paper, 2003-01-1316.

Howe, M.S., 1997. Edge, Cavity and Aperture
Tones at Very Low Mach Numbers, J. Fluid
Mech., vol.330, pp. 61-84.

Karbon, K., Kumarasamy, S., Singh, R., 2002.
Applications and issues in automotive
computational aeroacoustics, 10th Annual
Conference of the CFD Society, of Canada,
Windsor, Canada, June 9-11.

Nelson, P. A., Halliwell, N. A., and Doak, P. E.,
1981. Fluid Dynamics of Flow Excited
Resonance, Part I: Experiment, Journal of Sound
and Vibration, vol. 78 (1), pp 15-38.

Nelson, P. A., Halliwell, N. A., and Doak, P. E.,
1983. Fluid Dynamics of Flow Excited
Resonance, Part II: Flow Acoustic Interaction,
Journal of Sound and Vibration, vol. 91(3), pp

Ota, D.K., Chakravarthy, S.R., Becker, T., and
Sturzenegger, T., 1994. Computational study of
resonance suppression of open sunroofs, Journal
of Fluids Engineering, Vol. 116, pp. 877-882.

Rossister, J.E., 1964. Wind Tunnel Experiments
of the Flow over Rectangular Cavities at
Subsonic and Transonic speeds, Aero. Res.
Counc. R & M., 3438.

Ukita, T., China, H., and Kanie, K., 1997.
Analysis of vehicle wind throb using CFD and
flow visualization, SAE paper 970407.