Você está na página 1de 6

242 9th International Conference on Hydrodynamics

October 11-15, 2010 Shanghai, China

2010, 22(5), supplement :242-247


DOI: 10.1016/S1001-6058(09)60201-8

Numerical simulation of cavity flow induced noise


by LES and FW-H acoustic analogy

Nan Zhang *, Hong-cui Shen, Hui-zhi Yao


China Ship Scientific Research Center, Wuxi, China
*
E-mail: zn_nan@sina.com

ABSTRACT: The predictions of cavity flow and flow-induced advancing computational aeroacoustic (CAA)
noise are two important and complex issues in fluid-acoustic methods.
coupling field. Numerical studies for these issues are performed
in the paper by large eddy simulation (LES) and FW-H
acoustic analogy. Firstly, the wall pressure fluctuations of plate, It was shown that a greater understanding of the flow
foil, shutter hole are computed and compared with field and acoustic field generated by grazing flow past
experimental results. The robustness of large eddy simulation a cavity has been gained over the past ten years. In
in unsteady flow calculation is analyzed. Secondly, the addition, CFD is becoming a more reliable prediction
calculation of a 2-D cavity flow are accomplished. The power tool for this flow field. Many CFD analyses rely on
spectrum of pressure fluctuations is compared with measured
data and the vorticity distribution is analyzed. Finally, the flow
DES or LES to correctly simulate the shear layer in
induced noises of two 3D cavities are predicted. The computed turbulent flow conditions and predict the flow induced
results are compared with experimental data of Large noise accurately[1-5]. As more confidence is gained in
Circulation Channel in CSSRC. It shows that the numerical the use of CFD as a methodology for the prediction of
prediction method in the paper is credible. such complicated flow-acoustic coupling phenomenon,
more researchers are using this method to study the
KEY WORDS: cavity flow; flow induced noise; wall pressure
effect of control devices and cavity/body interactions.
fluctuations; Large eddy simulation; FW-H equation

1 INTRODUCTION Flow induced noise is a serious problem in many


engineering applications. It can cause human
The cavity flow belongs to a basic class of flows with discomfort and influence quiet operations of vehicles.
self-sustaining oscillations. In industrial practice, the In ship applications, the sound generated by marine
cavity-type oscillation is undesirable from the propellers, hydrofoils, and even transitional and
perspective of inducement of structure vibration and turbulent boundary layers can induce ambient
fatigue, generation of noise and drastic increase in concerns and decrease the work efficiency of sailors.
drag on the body. There were lots of discoveries and researches about
flow induced noise in aeroacoustics. However, in
Many computational studies focus on the computation hydroacoustics, it is short of an intensive investigation
of the cavity flow field with little attention given to for the problem.
the acoustic field surrounding the cavity. Exceptions
do exist, including several studies aimed at computing In the paper, the cavity flow and flow induced noise
the acoustics from low and subsonic Mach number are studied by the large eddy simulation (LES) with
flows past cavities. The cavity flow simulations that dynamic Smagorinsky subgrid model and FW-H
focus on the near field are equally important to the acoustic analogy with Kirchhoff integral. We aim at
development of a noise prediction capability for cavity establishing a suitable numerical method to predict the
flows; for, if the near-field flow is not simulated flow induced noise of cavity in water.
properly, one cannot hope to predict the acoustic field
accurately. In this sense, advances in computational
fluid dynamic (CFD) modeling go hand-in-hand with
9th International Conference on Hydrodynamics 243
October 11-15, 2010 Shanghai, China

2 COMPUTATIONAL METHOD normal vector that points into the fluid.

2.1 Large eddy simulation (LES) The far field solution of FW-H equation can be
written as the following:
The basic assumptions of LES are that: (1) transport is G ⎡ ρ (U + U n ) ⎤
largely governed by large-scale unsteady flow and 4π p 'T ( x , t ) = ∫ ⎢ 0 n 2 ⎥
dS
⎣ r (1 − M r ) ⎦ ret
f =0

these structures can be computationally resolved; (2)


small-scale flow features can be undertaken by using ⎡ ρ U ( rM r + c0 ( M r − M 2 )) ⎤ (8)
+∫ ⎢ 0 n ⎥ dS
appropriate subgrid scale turbulence models. In LES, f =0
⎣ r 2 (1 − M r )3 ⎦ ret
the motion is separated into small and large eddies, G 1 ⎡ Lr ⎤ ⎡ L − LM ⎤
4π p ' L ( x , t ) = ∫ ⎢ 2⎥
dS + ∫ ⎢ 2 r dS
this separation is achieved by means of a low-pass c0 = r (1 − M ) f = 0 r (1 − M ) 2 ⎥
f 0
⎣ r ⎦ ret ⎣ r ⎦ ret
filter. The filter function, G ( x, x' ) , implied here is
1 ⎡ Lr ( rM r + c0 ( M r − M 2 )) ⎤
then: + ∫ f =0 ⎢⎣
c0 r 2 (1 − M r )3 ⎥ dS
⎦ ret
(9)
⎧1 / V , x ' ∈V (1)
G ( x, x ') = ⎨ G ⎡K K K ⎤
⎩0, x 'otherwise 4π p 'Q ( x , t ) = ∫ ⎢ 21 + 22 + 33 ⎥dV (10)
f >0 c r cr r ⎦
Filtering the Navier-Stokes equations, one obtains ⎣
∂ρ ∂ We can computed the far field solution on a Kirchhoff
+ ( ρu i ) = 0 (2) control surface to avoid the volume integral.
∂t ∂xi
∂ ∂ ∂ ∂σ ij ∂p ∂τ ij 2.3 Numerical method
( ρu i ) + ( ρu i u j ) = (μ )− −
∂t ∂x j ∂x j ∂x j ∂xi ∂x j
The differential equations are discretized by finite
(3) volume method. Bounded central difference scheme is
where σi j is the stress tensor due to molecular applied. The velocity-pressure coupling is based on
viscosity and τ ij is the subgrid-scale stress. PISO algorithm and algebraic multigrid method is
employed to accelerate the solution convergence. The
time step is 1×10-5s and y+≈1.
In the paper, we adopt four subgrid-scale stress
models: 3 RESULTS AND DISCUSSION
• Smagorinsky Model (SL)
• Dynamic Smagorinsky Model (DSL) 3.1 Wall pressure fluctuations
• Wall-Adapting Local Eddy-Viscosity Model (WALE)
• Dynamic Kinetic Energy Model (KET) 3.1.1 Plate & foil

2.2 FW-H acoustic analogy

Ffowcs Williams and Hawkings utilized the


generalized function theory to obtain the classic
equation that has become associated with their names.
The FW-H equation can be written as the following
inhomogeneous wave equation:
1 ∂ 2 p ' ( x, t )
− ∇ 2 p ' ( x, t )
c 2
∂t 2 Fig.1 Plate model and sensor position
2
∂ ∂
= [( ρ 0U n )δ ( f )] − [Li δ ( f )] + ∂ Tij H ( f )
[ ] The wall pressure fluctuations (WPF) are important
∂t ∂x i ∂x i ∂x j features of turbulence. Turbulent wall pressure
(4) fluctuations are actually a source of noise. Knowledge
where of the wall pressure field induced by a turbulence
U i = [1 − ( ρ / ρ 0 )]vi + ( ρui / ρ 0 )
boundary layer remains a subject of fundamental
(5)
importance for many practical applications, especially
Li = Pij nˆ j + ρu i (u n − v n ) (6) for underwater acoustics. In order to demonstrate the
[ ]
Tij = ρuiu j + ( p − p0 ) − c02 ( ρ − ρ0 ) δ ij − τ ij (7)
capability of LES in predicting unsteady flow, the
wall pressure fluctuations are computed by LES in the
Tij is Lighthill stress tensor, δ ( f ) is Dirac delta paper. The WPF spectrums of plate in two velocities
are numerical simulated to determine the suitable
function, H is Heaviside function, u is fluid velocity, v
subgrid stress model. And we computed the WPF
is body surface velocity, c is velocity of sound, n is a
244 9th International Conference on Hydrodynamics
October 11-15, 2010 Shanghai, China

spectrum of foil by the suitable model. Fig.1 shows


the plate model and sensor position. The plate has an
overall length L of 1.1m and breadth of 0.39m.
Experimental investigations including boundary layer
and wall pressure fluctuations have been made in
anechoic wind tunnel of CSSRC at velocity 18m/s.
The computational domain is discretized using 1.35
million hexahedral cells. Measured and predicted
WPF spectrums are shown in Figures 2~5. The
reference pressure in air is Prefair=2×10-5Pa.

Fig.5 Comparison of measured and computed WPF spectrum


(KET model)

These computations are used to investigate the


performances of the four subgrid models in predicting
the unsteady characteristics of turbulent flow. We can
find that the differences of the computed results with
different subgrid models are very clear. It shows that
the computed WPF magnitude and spectrum shape
with DSL model is better than that with other models.
Generally, the DSL model predicts WPF more
Fig.2 Comparison of measured and computed WPF spectrum accurate than other models.
(SL model)
So we used the DSL model to compute the WPF
spectrum of a foil in three velocities (12m/s, 18m/s,
32m/s) . The geometry of foil and sensor position are
shown in Fig. 6. The foil has an overall length L of
1.1m and breadth of 0.39m. Experiments have been
carried out in anechoic wind tunnel of CSSRC. The
computational domain is discretized using 1.56
million hexahedral cells. Fig.7-10 show the computed
results including flow and WPF spectrum.

Fig.3 Comparison of measured and computed WPF spectrum


(DSL model)

Fig.6 Foil model and sensor position

Fig.4 Comparison of measured and computed WPF spectrum


(WALE model)
Fig.7 Computed flow field
9th International Conference on Hydrodynamics 245
October 11-15, 2010 Shanghai, China

3.1.2 Shutter-hole

We have investigated the wall pressure fluctuations of


shutter hole in water to identify the source of noise in
near field. Experiments including flow field and wall
pressure fluctuations have been made in the Large
Circulation Channel of CSSRC, which are suitable for
validation of numerical results. Figure 11 give the
shutter hole model and sensor position. The shutter
hole is arranged on a axisymmetric body.
Measurements were carried out at velocity V=5m/s.
Predicted and Measured WPF spectrums are shown in
Fig.8 Comparison of measured and computed WPF spectrum Figures 12. The reference pressure in water is
(V=12m/s)
Prefwater=1 × 10-6Pa. The computational domain is
discretized using 4.19 million hexahedral cells.

Fig.11 Shutter hole model and sensor position

Fig.9 Comparison of measured and computed WPF spectrum


(V=18m/s)

Fig.12 Comparison of measured and computed WPF spectrum

Fig.10 Comparison of measured and computed WPF spectrum In the frequency scope of 100Hz to 10kHz, the
(V=32m/s) predicted accuracy is 1.6dB to 6.0dB. The predicted
WPF spectrum is fairly good compared with
The predicted overall shape and the magnitude of the experiment data. The flow around shutter hole is very
WPF spectrum is fairly good compared with complex, so the capability of LES is validated.
experiment data. In the frequency scope of 100Hz to
10KHz, the predicted accuracy is 1.5dB to 5.5dB at 3.2 Flow induced noise
32m/s, 1.2dB to 8.4dB at 18m/s, 1.1dB to 13.3dB at
12m/s. It shows that the computed results agree better 3.2.1 2-D cavity
with measured data in high velocity than that in low
velocity. And the computed results agree better with Unsteady flow past a cavity may create both
measured data in low frequency than that in high broadband and tonal noise. The formation and
frequency. The characteristics of the unsteady flow behavior of a shear layer and its subsequent
with obvious pressure gradient are better captured by interaction with the fluid in the cavity drives the noise
LES approach. production. Lafon [6] studied cavity flow noise of a 2D
cavity. Fig. 13 shows the geometry of the model.
246 9th International Conference on Hydrodynamics
October 11-15, 2010 Shanghai, China

Experiments were carried out in a wind tunnel at the triggers the next vortex shedding. The rapid changes
Institut AeroTechnique, Saint-Cyr-l’Ecole. Flow in the flow past the cavity and its adjacent walls are
velocity is 62.8m/s. M=0.183. the primary sources to generate sound.

Fig.13 2-D cavity model and sensor position

Comparisons of measured and computed SPL


spectrum are shown in Figure 14(a)~(b).

Fig.15 Vorticity distributions at different stages

3.2.2 3-D cavity

We have studied the various flow induced noise of


five 3-D cavities which were arranged in a submerged
body. Experiments have been carried out in the Large
Circulation Channel of CSSRC at velocity V=5m/s.
Figure 16 presents the geometries of two models.
Figure 17 give the computed flow fields.

Fig.14 Comparisons of measured and computed SPL spectrum

Predicted overall shape and magnitude of the SPL


spectrum is well compared with experiment data. The
two oscillations modes are reflected accurately. The
computational accuracy of frequency and magnitude
are in 7% to 10% in comparison with measurement.
Fig.16 3-D cavity models
Computed vorticity distributions of different stages
are shown in Figure 15. The cavity flow is
characterized by violent ejections of vortices from the
cavity with length scales comparable with the cavity
dimensions rather than the thickness of the boundary
layer. The ejected vortex admits the free stream fluid
to enter the cavity impinging on the downstream face
of the cavity and triggers another vortex at the cavity
leading edge. The vortex grows in the cavity until it
completely fills the cavity and again gets ejected and Fig.17 Computed flow field
9th International Conference on Hydrodynamics 247
October 11-15, 2010 Shanghai, China

of noise measurements and calculations are presented


in Figs. 18-19. The trend in computed result is
consistent with that in measurement. The spectrum is
decreased rapidly as the frequency increases. For
sound pressure magnitude, the difference of computed
results and test data is 2dB~8dB. The comparison
between the prediction and the model test
measurement shows fairly good agreement. It
indicates that the complex flow field and flow induced
noise of cavity can be well simulated, and the
simulation method for flow induced noise of cavity is
established.

Fig.18 Validation of predicted flow induced noise of model A 4 CONCLUSIONS

The prediction of flow-induced noise is an important


and complex issue in fluid-dynamic acoustics field.
The hybrid CFD/acoustic methods or CAA methods
are refined for the acoustic field prediction of cavity
flow. In the paper, numerical studies for this problem
is performed by large eddy simulation and FW-H
acoustic analogy. It shows that the numerical
prediction method in the paper is credible. It is
obvious that more work is needed in this area. The
cavity noise prediction algorithms must resolve the
shear layer behavior well. As such, the prediction of
this noise becomes more feasible with continued
advance in computational capabilities. The
Fig.19 Validation of predicted flow induced noise of model B development of computational models also requires
excellent experimental data. The experimental cavity
flow research should be dominated by the studies of
near-field flow oscillation and far-field noise.

REFERENCES
[1] Ching Y L. Computation of Low Speed Cavity Noise[C].
42nd Aerospace Sciences Meeting, Nevada, U.S.A., 2004.
(AIAA–2004–0680)

[2] Mendonca F, Allen R, Charentenay J, et al. CFD prediction


of narrowband cavity acoustics at M=0.85[G]. AIAA Paper
2003-3303. 2003.

[3] Allen R, Mendonca F. DES validations of cavity acoustics


over the subsonic to supersonic range[G]. AIAA Paper
2004-2862. 2004.

[4] Lai H, Luo K H. Large-Eddy Simulation and Control of


Cavity Aeroacoustics[C]. Conference on Turbulence and
Interactions TI2006, Porquerolles, France, May 29 – June 2,
2006.

[5] Chen X X, Sandham N D, Zhang X. Cavity Flow Noise


Predictions[R]. Report No. AFM-07/05, University of
Fig.20 Vorticity distribution of body with cavity Southampton, February 2007.

The computational domain is discretized using 3.75 [6] Lafon P, Caillaud S, Devos J P, et al. Aeroacoustical
million hexahedral cells for model A and 3.41 million coupling in a ducted shallow cavity and fluid/structure
hexahedral cells for model B. The computed vorticity effects on a steam line. Journal of Fluids and structures,
2003, 18: 695-713.
distributions are shown in figure 20. The comparison

Você também pode gostar