Você está na página 1de 14

European Journal of Mechanics B/Fluids 40 (2013) 5063

Contents lists available at SciVerse ScienceDirect

European Journal of Mechanics B/Fluids


journal homepage: www.elsevier.com/locate/ejmflu

Sound of fluids at low Mach numbers


Young J. Moon
Computational Fluid Dynamics and Acoustics Laboratory, School of Mechanical Engineering, Korea University, Seoul, 136-701, Republic of Korea

article

info

Article history:
Available online 9 February 2013
Keywords:
Low subsonic flow
Turbulent flow noise
LES/LPCE hybrid method

abstract
The sound of fluid at low Mach number is a special research area that poses diverse applications not only
in aerodynamics but also in bio-medical or biological fluids. The related Mach numbers are in the order
of O(102 ) or even less and therefore the compressibility effects are substantially low but still play an
important role in many aspects. A hybrid method of splitting the hydrodynamic field and the acoustic field
is of our present interest and attention is given to the linearized perturbed compressible equations (LPCE).
In this paper, the linearized perturbed compressible equations are reviewed with some discussion on the
acoustic source term, DP /Dt. A few selected applications of aerodynamic noise and bio-fluid sound are
demonstrated by the present hybrid method.
Crown Copyright 2013 Published by Elsevier Masson SAS. All rights reserved.

1. Introduction
The sound of fluids at low Mach numbers is often encountered
in many practical aerodynamic applications such as ground transportation vehicles, ventilation ducts and jets, etc. A typical flow
speed of automobiles, for example, is in the range of 50100 km/h
(or M = 0.040.08) and flows are mostly turbulent and at moderately high Reynolds numbers. The computation of low-subsonic
turbulent flow noise is, however, a difficult task because the noise
sources are highly localized in the turbulent boundary layer near
the wall or in the wake, while the acoustic wavelengths far exceed
the hydrodynamic length scales. In this case, a direct numerical
simulation (DNS) employing the full compressible NavierStokes
equations becomes very difficult and expensive, coping with the
fact that a long-time computation is often required to represent
the turbulence statistics, i.e. the noise sources.
The sound in bio-medical or biological fluids is also in the range
of very low Mach numbers. For example, the vocal fold of the human larynx [1], or an insect flapping wings [2] produces the sound
by periodically disturbing the flow with body oscillating at the frequency range of 50200 Hz. The associated Mach number can be
figured as M = Ub /co = (Lc f )/co , where Ub is the moving speed of
the body, Lc the distance of travel, f the oscillating frequency of the
body, and co the speed of sound. For a biological body with length
scale of 12 cm, the associated Mach numbers are in the range of
M = 0.00150.012.
For computing sound of fluids at such low Mach numbers,
numerical difficulties lie not only in scale disparity between the

Correspondence to: Korea University, Department of Mechanical Engineering,


136-701 Seoul, Republic of Korea.
E-mail address: yjmoon@korea.ac.kr.

hydrodynamic field and the acoustic field but also in stiffness


of the system of the governing equations. In regard to this, a
hybrid approach has been sought as an alternative. This hybrid
method is based on a hydrodynamic/acoustic splitting technique
proposed by Hardin and Pope [3]. The hydrodynamic flow field is
solved by the incompressible NavierStokes equations, while the
acoustic field is computed by the perturbed Euler equations with
acoustic source obtained from the incompressible NavierStokes
equations. The idea of splitting the hydrodynamic part and the
acoustically perturbed part from the full compressible flow field is,
however, not a straightforward task because of the physics coupled
between these two fields. It has been found [4,5] that an unstable
vortical mode can easily be excited by the non-linear terms in
the perturbed momentum equations when the source terms are
improperly treated; for example, either lack of physical diffusion or
lack of grid resolution of the perturbed vorticity (
= u ). Here
the prime denotes an instantaneously perturbed quantity from the
incompressible state.
To avoid such vortical instability, the linearized perturbed
compressible equations (LPCE) [5] are formulated by eliminating
the terms related to the generation of the perturbed vorticity.
The details of the linearized perturbed compressible equations
(LPCE) are reviewed in Section 2. In present formulation, a material
derivative of the hydrodynamic pressure (DP /Dt) is derived
as an acoustic source in the linearized perturbed compressible
equations. At low Mach numbers, the material derivative of the
hydrodynamic pressure (scaled by itself) in the near field of the
incompressible flow is found very closely related to the dilatation
rate of the compressible counterpart, because the flow speed
relative to the speed of sound is substantially low that any thermal
effect during processes is nearly negligible. The near field of the
compressible flow computed by the direct numerical simulation
and that by the present hybrid method with acoustic source are

0997-7546/$ see front matter Crown Copyright 2013 Published by Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.euromechflu.2013.02.002

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

examined in Section 3 with discussion on the process of sound


generation.
In Sections 4 and 5, a few selected applications of aerodynamic
noise and bio-fluid sound are demonstrated by the present hybrid
method.

The present LES/LPCE hybrid method is based on a hydrodynamic/acoustic splitting method [3], in which the total flow
variables are decomposed into the incompressible and perturbed
compressible variables as,

(1)

p(
x, t ) = P (
x, t ) + p (
x, t ).
The incompressible variables represent hydrodynamic flow field,
while acoustic fluctuations and other compressibility effects are
resolved by perturbed quantities denoted by ( ).
The hydrodynamic turbulent flow field is first solved by incompressible LES. The filtered incompressible NavierStokes equations
are written as,
(2)

(3)

(4)

Here, is the mean radius of the grid cell (computed as cubic root
of its volume), Sij is the strain-rate tensor.
After a quasi-periodic stage of hydrodynamic field is attained,
the perturbed quantities are computed by the linearized perturbed
compressible equations (LPCE) [5]. A set of the linearized perturbed
compressible equations is written as,

+ (U ) + 0 ( u ) = 0
t
u
1
+ (u U ) + p = 0
t
0

+ b1

fi+2 fi2
41x

fi+1 2fi + fi1


1x2
fi+2 2fi + fi2
,
+ b2
41x2

(8)

(9)

where 1 = 1/3, 2 = 2/11, a1 = 14/9, b1 = 1/9, a2 = 12/11,


and b2 = 3/11.
Practically, when using a high order scheme to the stretched
meshes, numerical instability is encountered due to numerical
truncations or failure of capturing high wave-number phenomena.
Thus, a tenth-order spatial filtering (cut-off wave number, k1x
2.9) proposed by Gaitonde et al. [7] is applied every iteration to
suppress the high frequency errors that might be caused by grid
non-uniformity. For the far-field boundary condition, an energy
transfer and annihilation (ETA) boundary condition [8] with buffer
zone is used for eliminating any reflection of the out-going waves.
The ETA boundary condition is easily facilitated with a rapid
grid stretching in a buffer-zone and the spatial filtering which is
damping out waves shorter than grid spacing. So, if a buffer-zone
has grid spacing larger than out-going acoustic wave length, the
wave can be successfully absorbed by the ETA boundary condition.

The compressibly perturbed field was originally calculated by


the perturbed compressible equations (PCE), obtained by subtracting the incompressible NavierStokes equations from the full compressible NavierStokes equations.
The perturbed compressible equations [4] are written as


+ (u ) + ( u ) = 0
t
u
1
DU
1
+ (u )u + (u )U + p +
= fvis
t

Dt

p
+ (u )p + p( u ) + (u )P
t

DP
=
+ ( 1) q
Dt

p
DP
+ (U )p + P ( u ) + (u )P = .
t
Dt

21x

2.2. Linearized perturbed compressible equations (LPCE)

where the grid-resolved quantities are denoted by () and the unknown sub-grid tensor Mij is modeled as
2


Mij = U
i Uj Ui Uj = 2(Cs ) |S |Sij .

fi+1 fi1

2 fi
1 + fi + 2 fi+1 = a2

2.1. LES/LPCE hybrid formulation

U j
=0
xj

U i
+ 0 (U i U j )
0
t
xj

U i
U j
P

+ 0
+
=
0 Mij ,
xi
xj xj
xi
xj

spatially discretized with a sixth-order compact finite difference


scheme [6] and integrated in time by a four-stage RungeKutta
method. For example, the first and second derivatives with respect
to x are implicitly calculated with a five-point stencil, i.e.

1 fi
1 + fi + 1 fi+1 = a1

2. Computational methodology

(x, t ) = 0 + (x, t )
(x, t ) = U (x, t ) + u (x, t )
u

51

(5)
(6)
(7)

The left hand side of LPCE represents effects of acoustic wave


propagation and refraction in an unsteady, inhomogeneous flow,
while the right hand side only contains an acoustic source term,
which is projected from the incompressible LES flow solution. It
is interesting to note that for low Mach number flows, the total
change of the hydrodynamic pressure, DP /Dt is only considered as
the explicit noise source term.
The filtered incompressible NavierStokes equations are solved
by an iterative fractional-step method (Poisson equation for pressure), whereas the linearized perturbed compressible equations
are solved in a time-marching fashion. To avoid excessive numerical dissipations and dispersions errors, the governing equations are

(10)
(11)

(12)

), fvis
where D/Dt = / t + (U
is the perturbed viscous force

vector, and q represent thermal viscous dissipation and heat flux


vector, respectively. At low Mach numbers, the perturbed viscous
forces can be approximated as

uj
ui
2 uk
+

ij
(13)
xj
xi
3 xk
by assuming viscosity
= 0 (=constant), and and q are ex-

fvis
,i = 0

xj

pressed as

uj
uk
2 ul
=
+

jk
xk
xj
3 xl

p
qj = k
.
xj

uk
,
xk

(14)

(15)

Since perturbed variables are residuals of the total variables


with incompressible components subtracted, they represent not
only the acoustic fluctuations but also the other compressibility

52

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

effects such as coupling effects between the hydrodynamic flow


and the perturbed field. One particular component of the perturbed
variables related to the consistency of the acoustic solution
is perturbed vorticity (
= u ), a non-radiating vortical
component generated in the PCE system. This fluctuating quantity
becomes unstable for various reasons and generates unwanted
errors in acoustic calculations [4].
Here, attention is given to identify the terms associated with
production and diffusion of the perturbed vorticity in its transport
processes. The perturbed vorticity transport equations, derived by
taking a curl on Eq. (11) with mathematical identities and the
incompressible NavierStokes equations employed, are written as

+ (u )

)u + (
+
( u )]
[(
)u] [(u )

I-a
I-b
II-a
II-b

II

( p) + Fvis ,


III

(16)

IV

where Fvis = (
fvis
0 2 U )/ .
In Eq. (16), one can clearly see that perturbed vorticity is
generated and diffused by source terms on the right hand side
through: (i) coupling effects between the hydrodynamic vorticity
and the perturbed velocities (terms I and II), (ii) entropy field
(term III), and (iii) viscous force (term IV). Term I is related to
the three-dimensional effect of vortex stretching: stretching of
hydrodynamic vorticity by perturbed velocities (term I-a) and
stretching of perturbed vorticity by total velocities (term I-b). Term
II represents a more direct coupling between the hydrodynamic
vorticity and the perturbed velocities. The convective effect of
hydrodynamic vorticity by perturbed velocity is represented by
term II-a, whereas term II-b is related to the dilatation rate effect.
Term III is not so important for low Mach number, non thermallydriven flows and term IV only provides physical diffusion to
the perturbed vorticity. In the previous study [5], it was shown
that term II-a is the most dominant source term that generates
perturbed vorticity and term II-b is considered less important at
low Mach numbers.
It is interesting to note that perturbed vorticity is not a radiating acoustic quantity but a convecting hydrodynamic vortical.
Its physical meaning represents modification of the hydrodynamic
vorticity through interactions between the hydrodynamic vorticity and the velocity fluctuations. At low Mach numbers, the magnitude of the perturbed vorticity is small but, if falsely resolved,
it becomes self-excited and grows to affect the acoustic solution.
Since term II-a is related to the gradient of hydrodynamic vorticity
, perturbed vorticity usually appears at the edge of the hydrodynamic vorticity and its length scale is similar to (or sometimes
smaller than) the hydrodynamic vortical scale. Therefore, acoustic
grid resolution must carefully be handled in calculation.
By neglecting the second-order, non-linear terms such as (
u

)u , the original PCE, Eqs. (10)(12) can be re-written as


+ (U ) + 0 ( u ) = 0
t
u
1
+ (u U ) + p
t
0
DU
1
u ) (
= (
U )
+ fvis
0 Dt
0

(17)

(18)

p
+ (U )p + P ( u ) + (u )P
t

DP
=
+ ( 1) q

(19)

Dt

)u + (u )U = (u U ) +
with mathematical identity, (U

u ) + (
(
U ). Since the left hand side of Eq. (18) does not
generate any vortical component, only the right hand side terms
are responsible for the generation of perturbed vorticity. The first
u and
two terms,
U correspond to the dominant source
terms (terms I and II) in the perturbed vorticity transport equations
and the last two terms are associated with the entropy and viscous
effects (terms III and IV).
To show the Mach number dependence of each term, the perturbed momentum and energy equations, Eqs. (18)(19) are combined into a convective wave equation, neglecting the viscous and
thermal effect terms and then a Mach number scaling is conducted.
The hydrodynamic variables are scaled by their free stream values:
2
0 , U U , and P U
. For the perturbed variables,
a Mach number expansion approach [9,10] is employed; for example, u = U + Mu(1) + M 2 u(2) + M 3 u(3) + . So, the perturbed
velocity, u Mu(1) and from the linear acoustics, p ( c )u
and ( /c )u . The time is also scaled by l/c , where l is a
reference length scale and c is the speed of sound.
The resulting convective wave equation is written as

2 p
) p + U p
+
(
U
t2
t
t

O(M )
O(M 2 )

P 2

p +
0

P
P
u
P + (u )
+
( u )
t
t
t

O(M 3 )

u )} +
P {(
U ) + (

O(M 4 )


DP
.
t Dt

DU
0 Dt

+ (u U )

(20)

O(M )

3 (1) 2
3
Each term has the order of c
u /l (or c
U /l2 ) multiplied by a Mach number to the power denoted in Eq. (20). It is
clearly shown that the terms responsible for the generation of perturbed vorticity (i.e. the right hand side in Eq. (18)) have a Mach
number dependency O(M 4 ), whereas the leading-order terms
are O(M ). It is also interesting to note that the only explicit
acoustic source term, DP /Dt on the right hand side of Eq. (20) has
the same order as the first term in the convective wave equation,
2 p / t 2 .
Now, it is evident that the first two terms on the right hand
side of Eq. (18) are not so responsible for sound generation at low
Mach numbers and thereby one can exclude these to suppress
the generation of perturbed vorticity. The third term related to a
momentum correction to the perturbed mass can also be neglected
at low Mach numbers. The last term (perturbed viscous force)
is not necessary any more because there is no generation and
diffusion of perturbed vorticity. With the thermal terms neglected
in the perturbed energy equation, a set of linearized perturbed
compressible equations (LPCE) now read Eqs. (5)(7).
Because a curl of the linearized perturbed momentum equations, Eq. (6) yields

= 0,
t

(21)

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 1. Instantaneous 1

D
Dt

Fig. 2. Instantaneous

53

contours (compressible NS eqs., ReD = 150, M = 0.2).

1 DP
P Dt

contours (incompressible NS eqs., ReD = 150).

the LPCE prevents any further changes (generation, convection,


and decaying) of perturbed vorticity in time. In fact, the perturbed
vorticity could generate self-excited errors, if
is not properly
resolved with the acoustic grid. Hence, the evolution of the
perturbed vorticity is pre-suppressed in LPCE, deliberating the fact
that the perturbed vorticity has little effects on noise generation,
particularly at low Mach numbers. For hybrid methods [5,11], this
is an important property that ensures consistent, grid-independent
acoustic solutions. The derivation of LPCE with detailed discussion
on the characteristics of the perturbed vorticity can be found in
Ref. [5].

3. Acoustic source
The aerodynamic noise at low Mach numbers is often dominated by vortex interactions with solid walls. When a vortex interacts with the solid body, its strength changes in time and as a
consequence, the circulation in the neighboring fluids is altered
and so are the local streamlines. The time-varying streamlines are
directly connected to the pressure change in time and space and
therefore a so-called vortex sound is produced.
A generation of dipole tone from a circular cylinder is, for example, due to an alternating formation of vortex behind the cylinder and therefore circulation around the cylinder oscillates in time.
When vortex is formed at the upper side, a negative circulation
around the cylinder lowers the stagnation point at the frontal face
of the cylinder with creation of positive lift force. At a certain spe-

cific time interval, the flow field and lift force are at the opposite
phase when the vortex is formed at the lower side.
D

The instantaneous dilatation rate ( 1 Dt ) contours in the near


field during one period of dipole sound generated from the cylinder
at ReD = 150 and M = 0.2 is shown in Fig. 1. This is computed by
solving the full compressible NavierStokes equations. The upper
and lower four figures represent respectively the processes of
expansion and compression over the upper surface of the cylinder.
It is clearly noticeable that the rate of change of the dilatation
rate acts as a sound source in the near field, and Fig. 3 shows
the generation and propagation of the sound wave generated in
the near field by the snapshots of the two-level contours of the
dilatation rate, positive (red) and negative (blue).
The same physics of sound generation can also be represented
by the rate of change of the hydrodynamic pressure experienced
), which is computed
by a material element of the fluid (i.e. P1 DP
Dt
by solving the incompressible NavierStokes equations for the
contours
same flow condition (ReD = 150). The instantaneous P1 DP
Dt
in the near field during one period of dipole sound generation
are shown in Fig. 2. One can note a good resemblance between
these two total derivatives of the density and the hydrodynamic
pressure, both scaled by itself and plotted with the same contour
levels. It is also clearly noticeable in Fig. 4 that the rate of change
of P1 DP
acts as a sound source in the near field. The propagation of
Dt
the sound produced by the rate of change of P1 DP
is well shown by
Dt
the instantaneous pressure fluctuation contours, computed by the
linearized perturbed compressible equations with acoustic source,
DP /Dt.

54

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 3. Instantaneous 1 Dt contours; positive (red) and negative (blue) (compressible NS eqs., ReD = 150, M = 0.2). (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)

Fig. 4. Instantaneous pressure fluctuation p contours (linearized perturbed compressible eqs. (LPCE) with acoustic source DP /Dt from the incompressible NS eqs.,
ReD = 150, M = 0.2).

If one is asked where and when the sound is originated, it can


be estimated at an instant by a point where the dilatation rate and
both the linear strain rates in x and y directions are null. During the
repeated process of expansion and compression, the rate of change
of the linear strain rates are coupled with that of the dilatation rate,
via conservation of mass. The condition of null for the dilatation
rate and the linear strain rates represents an inflection point, i.e. a
material point (or line in 3D) that takes only translation in flow
motions and therefore, pressure or density changes most rapidly
during a transient development of the flow field.
D
The locations satisfying the null condition (i.e. 1 Dt = 0,
u
x

= 0, and

v
y

= 0) are marked in circles on the snapshots of


the dilatation rate contours in Fig. 3. In the top four figures, one
can note a process of expansion along the trace of the marked
circles on the right, whereas the emission of the compression
wave can be traced along those on the left. The bottom four
figures also show the formation and emission of the sound wave
at opposite phase along the traces of the marked circles. The
same process of sound generation can also be traced in Fig. 4,
which shows the instantaneous pressure fluctuations p perturbed
from the incompressible hydrodynamic pressure P, computed by
the linearized perturbed compressible equations with acoustic
source, DP /Dt acquired from the incompressible NavierStokes
solutions. The marked circles found in compressible flow solutions
can be found in the incompressible flow solutions with condition
satisfying P1 DP
= 0, Ux = 0, and Vy = 0. Along the marked circles,
Dt
the sound generation process of expansion and compression as
well as their emissions can also be well traced in the figures.

4. Applications on aerodynamic noise


4.1. Trailing-edge noise
This case considers a flow (Uo = 20 m/s) over the flat plate at
zero angle of attack (experiment described in Ref. [12]). The plate
has a chord length of c = 10 cm with thickness h = 0.03c and
span L = 3c. The Reynolds number of the flow based on the chord
length, Rec is 1.3 105 and the Mach number, M is 0.06. This free
stream Mach number is considerably low, as far as capturing the
compressibility effects are concerned.
For incompressible large eddy simulation, an o-type grid is employed to treat four rounded-corners of the leading and trailing
edges. The computational domain is set to r = 10c and a spanwise
extension is chosen as 3% of the plate chord with flow periodicity
assumed at the side boundaries. The computational domain consists of 657 201 21 (about 2.8 millions) points in x, y, and z and
is divided into 32 blocks for parallel computations. A minimal grid
+
size for x and y is 0.0005c (or 1x+
min = 1ymin 3), while a uniform
grid spacing of 0.0015c (or 1z + 15) is used in the spanwise direction. The computation is conducted with 1t = 1 106 s for
400,000 iterations (or 0.4 s).
The boundary layer is triggered approximately at x = 0.2c by
the leading-edge separation bubble and becomes turbulent downstream towards the trailing-edge of the plate. It was found that the
thickness of the boundary layer, is 1.12h at x = 0.2c from the
trailing-edge and the turbulent Reynolds number, Re is approximately 230. The iso-surfaces of the second invariant property of the
velocity gradients (Q = 200) clearly show the noise sources near

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

55

Fig. 5. Instantaneous Q iso-surfaces around the trailing-edge (left); wall pressure fluctuations along the plate (right).

Fig. 6. Instantaneous DP /Dt contours on the acoustic (top) and hydrodynamic (bottom) grids (left); instantaneous pressure fluctuation field (right).

the trailing-edge (see Fig. 5(left)), i.e. convecting turbulent eddies


within the boundary layer and the vortex shedding at the trailingedge. Fig. 5(right) also shows the wall pressure fluctuations monitored along the plate from A to I (A: x = 0.2c, I: the rear face of the
trailing-edge). One can notice the leading-edge separation, convection of turbulent eddies, and the vortex shedding at the trailingedge in the time history of the wall pressure fluctuations.
The flat plate self-noise is now computed by the linearized
perturbed compressible equations. Fig. 6(left) shows the noise
sources near the trailing-edge, i.e. the acoustic source, DP /Dt
computed by LES interpolated onto the acoustic grid using bilinear
interpolation. The acoustic grid (347 247) with minimal normal
spacing at the wall five times larger than that of the hydrodynamic
grid allows the same time step used in LES (i.e. 1t = 1 106 s)
for the LPCE computation.
An instantaneous pressure fluctuation field (1p = (P + p )
(P + p )) around the plate in Fig. 6(right) clearly shows the radiation of the dipole tone generated by the vortex shedding at
the trailing-edge. The acoustic wavelength of the tone is close to
/c = 2.5, corresponding to the frequency St = 0.2 at M = 0.06.
Besides, the figure shows other high frequency waves being emanated from the trailing-edge as well as from the shear-layer reattachment point. There will also be the waves diffracted at the
leading and trailing-edge of the plate, and all of these will contribute in part to the far-field noise measured at the microphone
location.
In order to predict the far-field SPL spectrum, a computational
procedure described in Ref. [13] is followed. Since the microphone
is located at 20c from the plate, the 2D acoustic field computed
by the LPCE for the domain of 10c needs to be extrapolated to
20c and also to be corrected for 3D spectral pressure. Finally,
the 3D spectral pressure radiated by the simulated span h needs
to be corrected for the total span 100h (or 3c) employed in the
experiment. This procedure requires information on the spanwise

Fig. 7. Sound pressure level spectrum at r = 20c vertically away from the midchord of the plate; computation (blue), experiment (black). (For interpretation of
the references to colour in this figure legend, the reader is referred to the web
version of this article.)

coherence function of the surface pressure, (z ) in the most


dominant noise source region, i.e. the trailing-edge of the plate.
The spanwise coherence length of the surface pressure, Lc () is
then calculated by a Gaussian law, (z ) = exp{(z /Lc ())2 }. The
largest value of Lc () is approximately estimated as 7h at St = 0.2
but in most cases, Lc () is below h [13].
The far-field SPL spectrum for the actual span 3c is now
compared in Fig. 7 with the measured data of the Ecole Centrale de
Lyon [14]. The numerical results are signal-processed by applying a
hanning window function with the sampling frequency of 50 kHz,

56

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 8. Directivity patterns of 1prms at r = 20c for different Strouhal numbers.

the block length of 0.04 s, and the number of averages of 10. The
agreement is found excellent, especially for the match of the tonal
peak (peak level deviation is 2.7 dB), its spectral broadening, as
well as the other broadband part. This comparison indicates that
not only the noise sources but also their turbulence statistics are
well captured by the incompressible LES, while the propagation,
scattering, and diffraction of the acoustic waves around the plate
are accurately computed by the LPCE.
The directivity patterns at r = 20c are also presented in Fig. 8
for various Strouhal numbers (or ratios of the plate chord length
to the acoustic wavelength). At vortex shedding frequency (St =
0.2 or c / = 0.4), it represents a clear dipole. As the Strouhal
number increases or the acoustic wavelength becomes shorter
than the chord length, the waves diffracted at the leading and
trailing-edge of the plate are well captured; the directivity pattern
changes to a finger-like shape. It is worth noting that the first two
plots of St = 0.2 and 0.4 are consistent with what is expected
from analytical modeling based on zero-thickness assumption, as
shown for instance in the study of Roger and Moreau [15]. At
higher Strouhal numbers, the directivity pattern departs from the
analytical results, essentially by showing a secondary beaming
around 45 and 30 at St = 1 and 2, respectively. This could be
attributed to the plate thickness.
4.2. Porous trailing-edge
A porous treatment to the same flat plate for reduction of
trailing-edge noise is demonstrated, imposing a porous surface

with porosity of = 0.25 to a small, selected area of the trailingedge (2h upstream from the edge, with a plenum inside, see
Fig. 9), where the vortex shedding and eddy scattering produce a
dipole sound. The porous surface has a thickness of = 0.001c
and is characterized by non-dimensionalized permeability, K =
KU0 / c = 1 103 for case 1, 1 102 for case 2, and 1 101
for case 3. The porous flow is often modeled by employing the Ergun equation,

P s =

CE
U s + 0 U s U s
K

(22)

where CE denotes a dimensionless Ergun coefficient or Forchheimer constant, which is dependent on porosity and pore structure. The Forchheimer constant is, however, set to zero in the
present computation because the pore-level Reynolds number
based on the permeability and the transpiration velocity averaged
in time and space, ReK = Ut K 1/2 / turns out to be less than unity.
First, an xt plot of the wall pressure fluctuations is examined
along the plate and in the wake region. As shown in Fig. 10(left),
the solid trailing-edge exhibits distinct, regularly-spaced pressure
marks (1tUo /h 5) near the trailing-edge which corresponds
to the vortex shedding frequency at St 0.2. This is obviously
the noise source for producing the tone. With the porous surface
(K = 1 102 ), however, the strips of pressure marks are broken
into pieces (see Fig. 10(right)) by local blowing and suction of the
flow in the plenum.
The influence of the porous surface is more clearly explained
in Fig. 11 that the correlation length of the pressure fluctuations

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

57

Fig. 9. Depiction of porous surface (left); schematic of turbulent flow over a flat plate with porous trailing-edge (plenum inside) (right).

Fig. 10. Comparison of xt plot of wall pressure fluctuation at the mid-span near the trailing-edge; solid (left) and porous (right).

Fig. 11. Spatial correlation Rpp contours of wall pressure fluctuations (left); Rpp along the mid-span (right).

is substantially reduced in the streamwise direction, i.e. a significant reduction in the size of the dipole noise source. The spatial
correlation of the wall pressure fluctuations, Rpp along the midspan of the plate clearly indicates that the streamwise correlation length does not exceed 0.01c, i.e. 1% of the chord length with
K = 1 102 . The uncorrelated pressure field is resulted from
the transpiration velocity along the porous surface, non-uniformly
distributed in both the streamwise and the spanwise directions.
As discussed before, the far-field acoustics for the actual span
is directly related to the spanwise coherence length of the noise
source. Fig. 12 shows the spectrally-decomposed spanwise coherence lengths Lc at x = 0.02c. The most prominent reduction of

the spanwise coherence length is observed at St = 0.21, while


at other frequencies no noticeable difference or even slightly increased coherence lengths are found with the porous treatment.
Thereby, the tonal noise at St = 0.21 is expected to be much reduced compared to that of the solid case.
Finally, the PSD spectra of the far-field noise are compared in
Fig. 13 for different permeabilities. It is found that for K = 1
102 , the dipole peak at St = 0.21 is significantly reduced by 13 dB,
while there is no significant noise reduction with others. This is
due to the fact that with large permeability (e.g. K = 1 101 ),
the fluid flow in porous medium encounters almost negligible
resistance and therefore the plate with porous trailing-edge is

58

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 12. Spanwise coherence length of wall pressure fluctuations for solid and
porous trailing-edges.

Fig. 13. Comparison of PSD spectra at r = 20c for porous trailing-edges with
different permeabilities.

regarded as a plate with a shortened chord. In contrast, with small


permeability (e.g. K = 1 103 ), the porous surface behaves like
a solid surface so that the wall pressure fluctuations as well as the
PSD of the far-field acoustics are almost identical to those for the
solid case. Only within a certain range of permeability (e.g. K =
1 102 ), however, the porous surface provides a mechanism of
non-uniformly distributed transpiration velocity along the lower,
upper, and back-end surfaces and subsequently allows pressure
fluctuations to be modified along the porous surface.
5. Applications on bio-fluid sound
5.1. Human larynx
An experimental study has been conducted at the department
of Phoniatrics and Pediatric Audiology of the University hospital
Erlangen, Germany, to measure the vibrating surface of the human hemilarynx. Fig. 14 shows the human hemilarynx with a tube
inserted to blow air from the compressor and the optical measurement system: glass plate at the glottal midline, vocal fold and

brass calibration cube on the left side of the plate, glass prism,
and high-speed camera on the right. The surface grid evolution
of the human hemilarynx measured during one period of motion is monotonically interpolated in time and space: raw geometry(left); first interpolation(middle); second interpolation(right)
(see Fig. 15).
A 2D computational domain of flow and sound is configured
to include the hemilarynx vibrating at a fundamental frequency
of 140 Hz within a vocal track, 28 times of the vocal fold length
(L = 12 mm). Fig. 16 shows the moving grids at four different time
intervals in one period of the vocal fold motion at the mid-plane.
The first two figures are at the closure phase of the vocal fold,
whereas the last two are at the opening. The case considered here
corresponds to a flow rate of 0.0004 m3 /s with pressure difference
of 2.88 kPa, and in terms of non-dimensional quantities, the
corresponding flow condition is at ReL = 840 and M = 3.1 103 .
Fig. 17 shows the instantaneous vorticity contours of the flow
within the vocal track at the same time intervals as the moving
grids. A periodic disturbance introduced by the mucosal wave motion of the hemilarynx continuously makes the gap flow pulsating
at 140 Hz and as a result, the vortices are shed downstream the vocal track with a specific scale. The corresponding acoustic field is
computed with the same moving grids by the linearized perturbed
compressible equations with acoustic source acquired from the INS
solutions. As shown in Fig. 18, the top five figures are the instantaneous acoustic fields at closure of the vocal fold, whereas the remaining figures are at the opening phase. One can clearly note that
the sound waves generated in the human hemilarynx are very subtle to the vocal fold motion at each stage.
The computed 2D acoustics monitord at 11L upstream the
hemilarynx are compared in Fig. 19(left) with the experimental
data measured at the University Hospital Erlangen. Considering
the fact that the computed sound wave is a 2D solution, it depicts
well the basic characteristics of the sound produced by the human
hemilarynx. The main vocal sound at fundamental frequency of
140 Hz is well resolved by the present 2D model, whereas the
high frequency waves observed in the experiment is substantially
simplified as a discrete tone with frequency corresponding to the
scale shown in the vorticity pattern. From the present result, it is
worth to note that there exits an interesting correlation among the
hemilarynx motion, the vortical flow structure within the vocal
track, and the corresponding acoustic field. A similar correlation
can also be noted in Fig. 18 (right) on the computed sound wave
from the larynx vibrating with the same fundamental frequency.
The difference of the sound wave profile between the hemilarynx
and the larynx is attributed by the Coanda effect of the pulsating jet
in the gap, which was clearly depicted in the computed flow field
of the larynx.
A computation of the sound wave at this low Mach number
(i.e. M < O(103 )) can hardly be accessible with the compressible
NavierStokes equations and for this reason, the present hybrid
formulation may be considered as a viable tool for computing
diverse bio-medical fluid and sound applications at such low Mach
numbers. The full 3D surface geometry of the human larynx is
now undertaken by the present LES/LPCE hybrid method, and
one can expect more complex vortical structures produced by the
three-dimensional surface motions of the human larynx and their
associated sound waves in wider range of frequency scales, as
observed in the experiment.
5.2. Bumblebee
The unsteady flow and acoustic characteristics of the flapping
wing are numerically investigated for a two-dimensional model
of bumblebee at hovering and forward flight conditions. In this
study, the time-dependent flow and acoustic fields are computed

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

59

Fig. 14. Measurement of 3D surface of the human hemilarynx vibrating at the fundamental frequency (left); schematic of optical measurement system (right).

Fig. 15. Monotonic interpolation of the hemilarynx surface grid.

Fig. 16. Moving grids at four different time intervals in one period of the vocal fold motion at the mid-plane.

Fig. 17. Instantaneous vorticity contours at four different time intervals in one
period of the vocal fold motion; top two at closure, bottom two at opening.

for a prescribed flapping wing motion which mimics the real wing
kinematics employed by superposition of the pitching and heaving
motions, sometimes referred to as a figure-eight motion [16].

Fig. 20(left) illustrates the flapping motion of a two-dimensional


elliptic wing (chord length c and thickness d = 0.1c). The unsteady
motion is replicated by a wing motion of two strokes (down and
up), and each stroke consists of three stages (transverse, tangential,
and rotational motions).
Here, all the specific parameters of the flapping wing are based
on Bombus terrestris, bumblebee [17,18]; chord length (c ) is 0.8 cm,
wing span (R) which is the distance from base of the wing to tip
is 1.7 cm, beat frequency (f ) is 170 Hz, and stroke amplitude ( )
is 150, defined as the angle swept out by the leading edge from
dorsal reversal (start of downstroke) to ventral reversal (start of
upstroke) in the mean stroke plane.
For a computational domain of circle (extended to r = 500c),
the incompressible NavierStokes equations are solved with moving hydrodynamic grid (401 181, see Fig. 20(right)), to compute

Fig. 18. Instantaneous pressure fluctuation contours at ten different time intervals in one period of the vocal fold motion; top five at closure, bottom five at opening.

60

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 19. Sound wave profiles at 11L upstream the vocal fold; red (hemilarynx/compt.), blue (larynx/compt.), black (hemilarynx/exp.), L: vocal fold base length. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 20. Flapping motion of an elliptic wing (downstroke by hollow, upstroke by filled) (left); moving grids at transverse and tangential motions (right).

the flow of a wing flapping at a non-dimensional beat frequency,


St = fc /c0 = 0.004. The sound field is then computed by LPCE on
the moving acoustic grid (251 91), with minimum grid spacing
five times that of the hydrodynamic grid. Note that all the variables
investigated here are non-dimensionalized by the speed of sound
c0 , chord length c, and air density 0 .
For hovering case, the flapping angle is set to 0 and the advance ratio defined by the ratio of the flight speed to the mean
flapping velocity of the wing J = U /(2 fR) [18] is also set to
0. The Reynolds number based on the maximum translational
velocity, Rec = Umax c /air is 8800 and Mach number M = Umax /c0
is 0.0485, where c0 is the speed of sound. Fig. 21 shows the time
variations of the drag and lift coefficients of the flapping wing. It
is indicated that the mean drag coefficient (averaged over 10 periods) is nearly 0, while the mean lift coefficient is about 0.66. From
the definition of CL = FL /0.50 Umax cR, one can calculate the lift
force on a three-dimensional wing, employing the properties of the
bumblebee [17]. In this study, the maximum translational velocity
Umax of the bumblebee is 17 m/s so that the lift force generated by
a pair of 3D wing is about 3.11 102 N. This value is large enough
to support the weight of bumblebee, 8.63 103 N, although the
three-dimensional effects neglected in this study could partially
reduce the lift force.
Now the computed sound fields of the flapping wing in hovering motion are presented in Fig. 22. The result indicates that the
flapping wing sound is generated by two different basic mechanisms. First, a dipole sound is generated by a transverse motion
of the wing (two on the left). Due to the fact that the dipole axis
changes its direction from downstroke to upstroke, a drag dipole
is generated at wing beat frequency (St = fc /c0 = 0.004), while
the lift dipole is produced at 2f (i.e. St = 0.008), similar to the
drag and lift coefficients. Hence, the flapping wing sound is directional, as shown in Fig. 23. The sound pressure level (SPL) peak

Fig. 21. Time history of drag and lift coefficients (hovering).

corresponding to the lift dipole (St = 0.008) is not present or weak


at 0 and 180, while the wing beat frequency (drag dipole) is also
not present at 90 and 270. At other angles, both the drag and lift
dipoles clearly exhibit their peaks. This result is similar to the previous observation by Sueur et al. [19], indicating that the wing beat
frequency is most dominant in front, whereas the second harmonic
is most appreciable at sides.
Another sound source is associated with the vortex edgescattering during tangential motion of the wing. In Fig. 22(two
on the right), one can identify the sound waves (bracketed) at

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

61

Fig. 22. Instantaneous pressure fluctuation contours around the wing in hovering motion; flow fields representing the associated sound sources: (first and second) wing
loading by transverse motion and (third and fourth) vortex edge-scattering during tangential motion.

Fig. 23. Sound pressure level spectra around a hovering insect ( = 0 and = 0; U = 0 m/s) at r = 100c and every 45 position.

150c175c from the center with wavelengths () observed as


41c and 48c at t /T = 0.5 and 1, respectively. Considering the
wave speed c0 (=340 m/s = 250c /T ), the travel time of the
waves is estimated as 0.60.7 (i.e. 1t /T = 150/250175/250).
So, it is figured that these waves were generated during t /T =
8/109/10 and 3/104/10 at each stroke. Now, one can note
that the flow fields at t /T = 9/10 and 4/10 clearly exhibit the
vortical structures that are responsible for producing the dipole
sound during tangential motion of the wing. The vortices in the
shear layer emanated from the leading-edge scatter at the trailingedge of the wing and generate waves radiating perpendicularly to
the wing. It is also found that the frequencies of these waves are
close to St (=fc /c0 = c /) = 1/48 0.02 and 1/41 0.024.
These frequencies of dipole tones generated at the trailing-edge
agree fairly well with the theory of shear layer instability [20]: the
frequency of the shear layer breaking-off is calculated as 0.021 with
St = 0.017ML / , where ML = 0.044 is a local free stream Mach
number and = 0.035 is the momentum thickness normalized

by the chord length at t /T = 9/10, for example. Finally, one can


note in the spectrum that the SPL peaks are multiples of the wing
beat frequency with comparable amplitudes (see Fig. 23). This
frequency composition closely resembles the buzz sound of fly
measured by Sueur et al. [19].
In order to mimic the forward flight of bumblebee, all the
properties are kept constant, except the advance ratio J = U /
(2 fR) and stroke plane angle . The advance ratio is generally
used to determine the free stream velocity U . It ranges from 0
(hovering) to 0.6 (fast flight) [21,22], and an intermediate value of
J = 0.3 corresponding to U = 4.5 m/s is considered in this study.
The stroke plane angle is then determined by the force balance
condition [21]; the mean thrust must be equal to mean drag for
a flight at constant speed. In the presence of free stream velocity
(J = 0.3), it is found that the mean drag coefficient is nearly 0 at
= 40. At this angle, the thrust force generated by the flapping
wing is almost the same as the drag force caused by the forward
flight of bumblebee. So, the stroke plane angle is determined as

62

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063

Fig. 24. Sound pressure level spectra around a hovering insect ( = 40 and = 0.3; U = 4.5 m/s) at r = 100c and every 45 position.

= 40, which is very similar to the real bumblebee [18] for the
flight speed at U = 4.5 m/s.
Due to the free stream effect, the vortices shed from the leading and trailing-edge of the wing during transverse motion are not
developed as symmetric as for the hovering case and so are the induced velocity fields. Therefore, these vortices cannot self-propel
away from the wing but rather remain in the stroke paths. Besides,
the ratio between the free stream velocity and the maximum translational velocity of the wing is close to 0.26 and so the convection
effect is quite weak. As a result, the vortices drifting around the
flapping wing encounter complex wing-vortex interactions. When
compared with the hovering case, this clear distinction in vortical
flow structure is expected to change the aerodynamic sound characteristics for the forward flight case.
The sound fields for the flapping wing in forward flight are investigated by comparing the sound spectra in Fig. 24. Similar to the
hovering case, the transverse motion of the dipolar axis results in
drag (St = 0.004) and lift dipoles (St = 0.008). It is, however,
important to note that the directivity change is not as clear as that
at hovering. The dominant frequency does not vary significantly
and both the drag and lift dipoles exhibit their peaks with comparable amplitudes, regardless of directions. One may also note that
the dipole tones generated at the trailing-edge (St = 0.02 and
0.024) are not as distinct as for the hovering case (Fig. 23). These are
largely due to the prominent interactions between the wing and
the vortices, being considered as a discernible difference in acoustic feature between hovering and forward flight. This indicates that
the radiation pattern and frequency composition can change with
flight conditions and it is expected that these could be used as some
biological functions such as communication, territory defense, and
echolocation.
6. Conclusions
The present LES/LPCE hybrid method has efficiently predicted
the low-subsonic, turbulent flow noise, with accuracy confirmed

by comparison of the far-field sound pressure level with the experiment. The present method with modeling of flow in porous
medium has been extended for reduction of the trailing-edge
noise via porous material at the same low-subsonic, turbulent
flow condition. For applications of bio-medical and biological-fluid
sound, capacity and future potential of the method has been well
demonstrated by the present study. The flow and sound in biomedical and biological applications is a unique research field which
requires a very sophisticated analysis tool to emulate very weak
compressibility effects; for example, sound of blood flows in the
circulatory system or sound of airway flows in the respiratory system. It is also shown that at low Mach number, the acoustic source
represented by a material derivative of the hydrodynamic pressure
(i.e. DP /Dt) scaled by itself in the incompressible NavierStokes solution is very closely related to the dilatation rate of the compressible counterpart, at any instant, and so is the rate of change of P1 DP
Dt
to that of the dilatation rate.
Acknowledgments
The author would like to thank Prof. Roger, M. at the Ecole
Centrale de Lyon and Prof. Doellinger, M. at the University Hospital
Erlangen for providing experimental data for the flat plate and the
human larynx, respectively, and students, Dr. Seo, J.H., Dr. Bae Y.M.,
Mr. Jo, Y.W. for their contributions during graduate study at Korea
University.
References
[1] Y.M. Bae, Y.J. Moon, Computation of phonation aeroacoustics by an INS/PCE
splitting method, Comput. Fluids 37 (2008) 13321343.
[2] Y.M. Bae, Y.J. Moon, Aerodynamic sound generation of flapping wing, J. Acoust.
Soc. Am. 124 (1) (2008) 7281.
[3] J.C. Hardin, D.S. Pope, An acoustic/viscous splitting technique for computational aeroacoustics, Theor. Comput. Fluid Dyn. 6 (1994) 323340.
[4] J.H. Seo, Y.J. Moon, Perturbed compressible equations for aeroacoustic noise
prediction at low mach numbers, AIAA J. 43 (2005) 17161724.

Y.J. Moon / European Journal of Mechanics B/Fluids 40 (2013) 5063


[5] J.H. Seo, Y.J. Moon, Linearized perturbed compressible equations for low mach
number aeroacoustics, J. Comput. Phys. 218 (2006) 702719.
[6] S.K. Lele, Compact finite difference schemes with spectral-like resolution,
J. Comput. Phys. 103 (1992) 1642.
[7] D. Gaitonde, J.S. Shang, J.L. Young, Practical aspects of high-order numerical
schemes for wave propagation phenomena, Internat. J. Numer. Methods Engrg.
45 (1992) 18491869.
[8] N.B. Edgar, M.R. Visbal, A general buffer zone-type non-reflecting boundary
condition for computational aeroacoustics, AIAA-Paper 2003-3300.
[9] J.H. Park, C.D. Munz, Multiple pressure variables methods for fluid flow at all
Mach numbers, Internat. J. Numer. Methods Fluids 49 (2005).
[10] S.A. Slimon, M.C. Soteriou, D.W. Davis, Development of computational
aeroacoustics equations for subsonic flows using a Mach number expansion
approach, J. Comput. Phys. (2000) 159.
[11] R. Ewert, W. Schrder, Acoustic perturbation equations based on flow
decomposition via source filtering, J. Comput. Phys. 188 (2003) 365398.
[12] Y.J. Moon, J.H. Seo, Y.M. Bae, M. Roger, S. Becker, A hybrid prediction method
for low-subsonic turbulent flow noise, Comput. Fluids 39 (2010) 11251135.
[13] J.H. Seo, Y.J. Moon, Aerodynamic noise prediction for long-span bodies, J. Sound
Vib. 306 (2007) 564579.

63

[14] M. Roger, S. Moreau, A. Gudel, Vortex-shedding noise and potential-interaction noise modeling by a reversed sears problem, AIAA-Paper 2006-2607.
[15] M. Roger, S. Moreau, Back-scattering correction and further extensions of
Amiets trailing-edge noise model. part 1: theory, J. Sound Vib. 286 (2005)
477506.
[16] Z.J. Wang, Two dimensional mechanism for insect hovering, Phys. Rev. Lett. 85
(2000) 22162219.
[17] T. Weig-Fogh, Quick estimates of flight fitness in hovering animals, including
novel mechanism for lift production, J. Exp. Biol. 59 (1973) 169230.
[18] R. Dudley, C.P. Ellington, Mechanics of forward flight in bumblebees, I,
kinematics and morphology, J. Exp. Biol. 148 (1990) 1952.
[19] J. Sueur, E.J. Tuck, D. Robert, Sound radiation around a flying fly, J. Acoust. Soc.
Am. 118 (2005) 530538.
[20] W.K. Blake, Mechanics of Flow-Induced Sound and Vibration, Vol. 1, Academic
Press, Orlando, 1986.
[21] M. Sun, J.H. Wu, Aerodynamic force generation and power requirements in
forward flight in a fruit fly with modeled wing motion, J. Exp. Biol. 206 (2003)
30653083.
[22] A.R. Ennos, The kinematics of aerodynamics of the free flight of some diptera,
J. Exp. Biol. 142 (2002) 4985.

Você também pode gostar