NASA-CR-205218
,, /
/
Time-Domain Impedance BoundaryConditions
for ComputationalAeroacoustics
Christopher K. W. Tam and Laurent Auriault
Reprinted from
AIAA
Journal
Volume34, Number5, Pages917-923
A publication of the
American Instituteof Aeronauticsand Astronautics, Inc.
370 L'Enfant Promenade, SW
Washington, DC 20024-2518
AIAA JOURNAL
Vol. 34, No. 5, May
1996
Boundary
Conditions
Aeroacoustics
Time-Domain
Impedance
for Computational
Christopher
Florida
State
K. W. Tam*
University
and Laurent
Tallahassee,
Auriault
Florida
t
32306-3027
It is an accepted practice in aeroaeousties to characterize the properties of an acoustically treated surface by a
quantity known as impedance. Impedance is a complex quantity. As such, it is designed primarily for frequencydomain analysis. Time-domain boundary, conditions that are the equivalent of the frequency-domain impedance
boundary, condition are proposed. Both single frequency and model broadband time-domain impedance boundary
conditions are provided. It is shown that the proposed boundary conditions, together with the linearized Euler
equations, form well-posed initial boundary value problems. Unlike ill-posed problems, they are free from spurious
instabilities that would render time.marching computational solutions impossible.
Nomenclature
ao
k
M
= speed of sound
= total wave number
=Mach number
p
R
= pressure
= acoustic
Ro
Re( )
t
= artificial mesh Reynolds
= real pan
= time
u, v. w
v
v.
= velocity
= velocity
= velocity
(positive
= acoustic
X
resistance
number
p = Zv.
components
vector
component normal to impedance
pointing to the surface)
reactance
At
Ax
= impedance
= wave number
variables)
= time step
= mesh size
p
f2. w
= gas density
= angular frequency
Z
a, _
definitely preferred over a more demanding microscopic
description
of the actual phenomenon.
In the aeroacoustics
community,
it is
an accepted practice to characterize
the macroscopic
properties of
an acoustically
treated surface by a single quantity Z called the
impedance.
The impedance
is defined as the ratio of the acoustic
pressure p to the acoustic velocity component
normal to the treated
surface v_ (positive when pointing into the surface). That is,
I.
components
(Fourier
(1)
Impedance is a complex quantity, Z = R - iX (e -i_" dependence
is assumed). The use of a complex quantit3' is needed to account
for the damping and phase shift imparted on the sound waves by
the acoustically
treated surface. The acoustic resistance
R and the
acoustic reactance X are generally frequent3' dependent. They also
vary with the intensity of the incident sound waves and the adjacent
mean flow velocity. These quantities are usually measured empirically, although some semi-empirical
formulas are available for their
estimates,
provided the construction
of the panels is sufficiently
simple. Figure 3 shows a typical set of measured
resistance
and
reactance data for a 6.7%-perforate
treatment
panel at low sound
intensity given in Ref. I. An important feature is that R is positive
and does not vary much with frequency. On the other hand, X can
be both posi five or negative depending on the frequency. The dependence of X on frequency can be represented
by a simple analytical
surface
transform
Introduction
OWADAYS, acoustic treatment is invariably used on the inside surface of all commercial aircraft jet engines for fan noise
reduction. Acoustic treatment panels, when properly tuned, are extremely effective noise suppressorsJ
Because of structural integrity
requirements,
the treatment panels are usually of the Helmholtz resonator type. The damping mechanism of these types of panels are
gener'a]ly attributed to the dissipation
associated with the oscillatory, jets formed at the mouths of the Helmholtz resonators 2-4 (see
Fig. 1). These jets are induced by the pressure fluctuations accompanying the passage of acoustic disturbances
over the surface of the
treatment panels.
The flow and acoustic fields around the Helmholtz resonators of
Fig. 1 Schematic diagram of the oscillatory, jets induced by incident
acoustic waves at the mouths of the Helmholtz resonators of a treatment
panel.
the treatment panels are exceedingly complicated,
especially when
there is a mean flow adjacent to the panel. Figure 2 shows schematically the oscillatory jets at the mouths of the Helmholtz oscillators or
cavities induced by pressure waves in the presence of a mean flow as
observed by Baumeister
and Rice 3"_ in a water channel simulation.
..al--
Mean flow
cS_.,,
................,.
_._,
For engineering
purposes,
a gross macroscopic
description
of
the effects of the treatment panels on the incident acoustic waves is
Received Aug. 18, 1995; revision received Jan. 18, 1996; accepted for
publication Jan. 29, 1996. Copyright © 1996 by Christopher K. W. Tam and
Laurent Auriault. Published by the American Institute of Aeronautics and
Astronautics. Inc., with permission.
"Professor, Department of Mathematics. Associate Fellow AIAA.
*Graduate Student, Department of Mathematics.
Fig. 2 Schematic diagram of the oscillatory
waves in the presence of a mean flow.
917
jets induced
by sound
918
TAM
o
e4
i
i
i
AND
AURIAULT
X
i
o
o
c_
Y
;<
oo
o.
7
acoustical]
pancl
treated
"7,
Fig.
o
4
Sound
field
adjacent
to an acoustically
treated
panel.
"t
0.0
2.0
tO
frequency
Fig.
3
Ref.
of the
Dependence
treatment
panel
on
resistance
frequency
1; o, reactance
and
at low
3.0
4.0
(kHz)
and
reactance
sound
of a 6.7%-perforate
intensity
(no
flow)
given
in
e, resistance.
expression. For instance, the data in Fig. 3 are accurately
by a two parameter formula,
X/pao
= (X_l/w)
+ Xtw
represented
(2)
The parameters X-x and X: are found by mean-least-square
fit to
be -13.48
and 0.0739, respectively, where w is measured in kiloradian/second.
Impedance boundary, condition (I) is basically a boundary condition established for frequency-domain
analysis. As it is, it cannot
be used in time-domain computation. The primary objective of this
paper is to derive a suitable equivalent of the impedance
boundary
condition
in the time domain. Such a boundary
condition has not
been considered
before. Time-domain
impedance
boundary condition would allow the use of the newly developed computational
aeroacoustics
methods for the solution of duct acoustics and turbomachinery
noise problems. One significant advantage
of timedomain methods over frequency domain methods is that broadband
noise problems can be handled relatively easily, almost without extra
effort. For broadband noise problems, frequency domain methods
are computationally
intensive and laborious.
Time-domain
problems can be solved only if they are well posed.
One of the requirements of well-posedness
is that the mathematical problem is stable and dependent continuously
on the initial
and boundary data. In this paper, it will be shown that the timedomain impedance
boundary conditions
developed do not lead to
ill-posed mathematical
problems. There are no spurious unstable solutions. Results of direct numerical simulations
using these newly
developed time-domain
impedance boundary conditions
are found
to agree well with analytical solutions.
In the presence of a mean flow, the standard formulation
of the
impedance boundary condition s is known to give rise to a spurious
unstable solution of the Kelvin-Helmholtz
type. 6 A general proof
of the existence of such instability is provided here. Because of
this instability, time-domain
methods cannot be successfully
implemented. An alternative way of prescribing
the impedance condition,
that would not lead to an ill-posed initial boundary value problem,
is proposed. A time-domain
solution is possible when the proposed
impedance
boundary condition is used.
It is easy to verify by direct substitution
that Eq. (3) or (,t)
yields the frequency-domain
impedance
boundary condition
p =
Zu,(Z
= R - iX) for a sound field of single frequency Q. The
reason why boundary
condition (3) should not be used when X is
positive is that it will lead to spurious unstable solutions. In other
words, Ec1. (3) will give rise to ill-posed initial boundary value problems for X > 0. The same reason applies to the restriction imposed
on boundary condition (4).
B.
WelI.Posedness
equations)
A.
Single-Frequency
Impedance
Boundary
Time-Domain
Impedance
Boundary
Time-Domain
Condition
Condition
Let us first consider the case in which the sound
field consists of a
single frequency f2 (f2 > 0). That is, the pressure and velocity fields
of the sound waves have time dependence,
at t _ oo, of the form
p(x. t) = Re[/3(x)e -ica] and v(x, t) = Re[_,(x)e-if2:].
Suppose the
resistance and reactance of a treatment panel at angular frequency f2
is R and X; then a suitable set of time-domain
boundary condition
at the surface of the panel is
X
<0,
Op
0-7 =
ROUn
at
- Xf2v.
(3)
p=Ru,+----
(2 at
(4)
Boundary
Condition
are
ap
at
(5)
= _V.
v
(6)
Let f(ot, _, w) be the Fourier-Laplace
transform
f(x, z, t). The functions f and f are related by
f(ct, ,8. co) = (2rr)----_
x exp[--i(ctx
x exp[i(tzx
f(x,z,
of a function
t)
+ _z -- wt)] dt dx dz
+/_z
-- wt)] dcoda
By applying Fourier-laplace
easy to find that the solution
wave condition at y _ oo is
(7)
d/_
(8)
transforms to Eqs. (5) and (6), it is
that satisfies radiation or outgoing
[_]=m[(co.,ll)½/_]exp[ik<_:-l)½y]
k ---_ (Q,2 .-I- fl2)1/2
_.) =
o.)lk
(9)
and v is the velocity
component
in the y direction (note that v = -o.).
The branch cuts of the
function (&2 - 1) I/'- are taken to be 0 _< arg(_: - 1) I/2 _< rr; the
left (right) equality is to be used if & is real and positive (negative).
The branch cut configuration
in the _ plane is shown in Fig. 5. The
Fourier-Laplace
transforms of boundary conditions (3) and (4) (in
dimensionless
form with ,_) as the scale of impedance)
are
X < 0,
/3 = [-R + (iXf2./k_)]_
(I0)
X>_O,
_={-R+(i_kX/f2)]_
(11)
Substitution of EcI. (9) into Eqs. (10) and (l 1) leads to the following
dispersion relations:
co-
X Or.
X>_O,
Impedance
av
-- = -vp
0t
whel_
II.
of T'tme-Domain
To show that Eq. (3) or (4) leads to well-posed initial boundary
value problems, we will consider a plane treatment panel adjacent
to a sound field as illustrated in Fig. 4. For simplicity, we will let
the surface of the panel be the x-z plane. In terms of dimensionless
variables with L (a typical length of the problem) as the length scale,
ao as the velocity scale, L/ao as the time scale, Po (the ambient
gas density) as the density scale, and _a o as the pressure scale,
the acoustic field equations (the linearized momentum
and energy
i X_
X <0.
+R_=_
(rift-
1)½
(12)
k
TAMAND AURIAULT
919
Im(_)
t Im(_)
branch
cut
a)
G
;i
,
A
....
27
Re(h)
Re(¢.7))
Ira(g)
Fig. 5 Branch cut configuration for (&2
1)l/: = 1
_(¢_ + ¢2)-
--
1)l/:.
Argument of (,_,1 _
H
Re(g)
q__A
a)
A
b)
Fig. 7 Map of the upper half _ plane in the g plane: a) if: plane and
b) g plane.
A Re(1)
r_flcctcd
wavc_.
soLmd . i
_mentl
x
/---0
x=0
g
C
E
b)
Fig. 6 Map of the upper half_
f plane.
Fig. 8
plane in thef
&
x > 0,
-
1.8m
(_ - 1)½
plane: a) & plane and b)
kX.
i--_-w = -R
(13)
Initial boundary
value problems involving governing equations
(5) and (6) and boundary conditions
(3) or (4) are stable and well
posed if dispersion relations (12) and (13) have no solutions in the
upper half & plane. To prove that this is the case, let the left-hand side
of Eqs. (12) and (13) be denoted
by f(&)
and g (&). respectively:
_2
=
f(_)
, + RS_
(14)
kX.
i--_-w
_g
(15)
(d3 - 1):
g(_)
=
(_-
1)½
Figures 6 and 7 show the maps of the upper half cb plane in the f
and g planes. The shaded region represents
the image points of the
upper half & plane. Since the right-hand side of Eq. (12) is purely
ima_nary
and negative, and the shaded region in the f plane does
not include the negative ima_nary
axis, no value of & in the upper
half & plane can satisfy dispersion
relation (12). Also, the righthand side of Eq. (13) is real and negative so that it will not lie in
the shaded region of Fig. 7. Thus, dispersion
relation (13) has n6
solution in the upper half t3 plane. Therefore, Eqs. (3) and (4) would
lead to well-posed problems. On the other hand. it is easy to show
if boundary condition
(3) is used when X is positive or boundary
condition (4) is used when X is negative, there will be unstable roots
associated with dispersion relations (12) and (13). In these cases,
there _-il] be spurious unstable solutions and the initial boundary
value problems are ill posed.
Normal incidence impedance tube-
C.
Numerical Implementation
and Results
To illustrate the effectiveness
of time-domain
impedance
boundary conditions (3) and (4), we will apply them to the numerical
simulation of the standing wave pattern of the normal-incidence
impedance tube problem. 1'7-9 The impedance
tube is designed to
measure the impedance
of an acoustic liner sample. The sample is
placed at one end of a long tube as shown in Fig. 8. A single frequency acoustic wave train is introduced at the open end. The sound
waves are reflected off the surface of the acoustically
treated panel.
The inciden_ and reflected waves form a standing wave pattern. By
measuring the standing wave ratio of the pressure envelope and the
relative phase of the incident and reflected waves (from the position
of the first minimum of the pressure envelope) the impedance of the
treatment panel can be determined)
._
Now let us consider an impedance tube of length 1.8 m. The speed
of sound at room temperature
is 340 m/s. To ensure there are at least
seven mesh points per wavelength
in the computation
for sound
frequency up to 4 kHz, we will divide the impedance
tube into 150
mesh spacings yielding Ax = 0.012 m. In the rest of this paper, Ax
is used as the length scale in all of the computations
(i.e., L = Ax).
We will assume that the treatment
panel is the 6.7% perforate of
Fig. 3. The equations governing
the acoustic field inside the tube
are the one-dimensional
version of Eqs. (5) and (6). They are
Lru o
The time-domain
boundary
right end of the tube, is
X < O,
condition
to be applied
3p
3u
(9-'7= Rc_t --Xftu
at x = 0, the
(17a)
XOu
X>O,
p=Ru+----
f2 at
(17b)
920
TAM AND
The incident
sound
AURIAULT
wave will be taken to be
ii,,q
At the left end of the tube, the radiation boundary condition is to be
enforced. Such radiation boundary
condition has been formulated
by Tam. m It may be written in the form
L
0
,
at[;]=-_x[;]-t-[ll]2_sin[i2(x-t)]
(18)
.
The nonhomogeneous
term in Eq. (18) is from the incident acoustic
waves.
We will use the seven-point stencil dispersion-relation-preserving
(DRP) time-marching
scheme m"t l for the numerical solution of the
impedance tube problem. The DRP scheme was designed for highquality numerical
solution of aeroacoustics
problems. It is highly
accurate even when as few as seven mesh points per wavelength are
used in the computation.
The discretized form of the right-hand side
of Eq. (16) is
±
(,)
=--j=-3a)Pt+J--_-)Z-_
K_,)
3
= -j=-3
(")
ajut+_
I
1
-_
,
_
d.u (")
=_ 3,,aj t+j
3
(°)
1_"'3 ajpt÷j=-
- 150.0
At
P Jr
_=[)
KI"-''
(21)
LLJt
dition (18). Similarly, for the bounda_,
points on the right end of
the tube, the expressions
for K_ ) and L_ ) are obtained by backward
differencing
the right-hand side of (16). Artificial selective damping
terms similar to that of Eq. (19), but with reduced stencil size, t° are
again added.
To enforce the impedance
boundary
condition (3) or (4) or (17)
at £ = 0, we note that cOp at and _u/at are given by Eq. (16). On
eliminating
the time derivative
terms by Eq. (16), the impedance
boundary condition at e = 0 becomes
Fig. 9 Spatial distribution of the pressure envelope in a normalincidence impedance tube at 1600 Hz frequency: _,
time-domain
solution (DRP scheme) and .... , exact (frequency-domain) solution.
i
X >__O,
p = Ru-
,
- 15o.o
X ap
--_
f2 0x
(23)
upon discretizing Eq. (22) or (23), at time level (n + 1) by backward
differences,
an algebraic equation involving the ghost value p(j"+_)
,
I
-_00.0
,
,
,
,
I
-50.(:
)
Fig. 10 Spatial distribution of the pressure envelope in a normalincidence impedance tube at 3000 Hz frequency.: _,
time-domain solution (DRP scheme) and .... , exact (frequency-domain) solution.
is obtained. This equation provides the formula by which the ghost
value may be found. For instance, ifEq. (22) is used, we find
,
Ra_-----fkj__L__,
,.,,
.._u,,
,____ a, p,
)
(24)
In carrying out the time-marching DRP algorithm, the computation foUows the following procedure. After the solution at time level
n has been found, the values of u and p are updated to the next time
level (n + I) by Eq. (21) at every point on the grid except the ghost
point. The ghost value pl _+ II is calculated
by Eq. (24). With the
ghost value found, the calculation
proceeds to the next time level
(n + 2). The entire process is then repeated.
Figure 9 shows a comparison of the time-domain
solution of the
pressure envelope along the full length of the impedance
tube calculated by the DRP scheme and the exact frequency-domain
solution
at a frequency of 1.6 kHz. At this frequency X is negative so that
impedance boundary condition (3) was used in the numerical solution. Figure 10 shows a similar comparison
at a frequency
of 3.0
kHz. In this case X is positive. Accordingly,
impedance
boundary
condition (4) was employed. The time-domain
solutions compare
well with the exact frequency-domain
results. The peak values and
their locations, which are important quantities
in impedance
tube
experiments,
are accurately calculated.
These simulations
suggest
that time-domain solutions involving impedance boundaries are feasible and accurate. Moreover, the proposed time-domain
impedance
boundary
condition, indeed, leads to well-posed
initial boundary
value problems.
Broadband
Time-Domain
Boundary
Condition
Three-Parameter Broadband Model
Ill.
(22)
,
X
0u
_x = R_x x + i2Xu
•
O0
(20)
Coefficients
by may be found in Ref. I0. This time-marching
algorithm
applies to all points on the computation
grid; £ =
0, -I, -2 .....
-150. For the last three points on the left-hand side
of the computation
domain; i.e., £ = - 150, - 149, - 148, K_ ") and
L(.) are obtained by applying backward difference to the spatial
derivative terms on the right-hand side of radiation boundary con-
X <0,
,
(19)
where subscript g is the spatial index; e = 0,-1,
-2 .....
-150,
and superscript n denotes the time level. The coefficients
aj and
d_ are provided in Ref. I0. The last terms on the right-hand sides
of Eqs. (19) and (20) are the artificial selective damping terms} 2
The artificial selective damping
terms are designed to remove any
spurious short wavelength numerical waves inside the computation
domain. It can be shown that the damping terms do not affect the
part of the solution that has wavelength
longer than seven mesh
spacings. In the present computation,
R, = 20 has been used.
The impedance
boundary
condition
(3) or (4) or (17) is to be
imposed at e = 0. Here, we will follow Ref. 13 and introduce
a
ghost value of pressure Pt at t = 1. The ghost value Pt is to be
chosen so that both the governing equation (16) and the impedance
boundary condition
are satisfied at the boundary
point e = 0. The
addition of a ghost value at £ = 1 allows Eq. (19) to be applied to
- 147 < t < -2, whereas Eq. (20) is used only for- 147 < e < -3.
To march in time, the DRP scheme uses the following four-timelevel algorithm:
I. Plt
•
-50.0
X
oi
6
["I '°÷''= "I'°' +
-100.0
A.
Impedance
Both the resistance and impedance of an acoustic treatment
panel
of the Helmholtz resonator type are frequency dependent.
Figure 3
shows measured data typical of such panels, as reported in Ref. 1.
It is seen that over the frequency range of interest,
1-3 kHz. the
variation in the resistance is small. However, there is a significant
TAM
AND
AURIAULT
921
change in the reactance. The measured data of Fig. 3 can be closely
approximated
by analytical formulas of the form
R = Ro,
X = (X_l/w)
+ Xjoo
, lm(a_)
H
(25)
where Ro(Ro > 0), X-l(X-l
< 0), and Xl(Xt
> 0) are parameters. The values of these parameters are determined
by mean-leastsquare fit to the data. The impedance
corresponding
to this threeparameter model is
Z = Ro - i[(X_t/w)
+ Xico]
-k
a)
Re(_)
+k
Imff'3
(26)
LH
Now we propose the use of the following
boundary, condition:
by,
0"-'7= R_-_t
time-domain
Op
- X__v.
+ Xl
impedance
02vn
_t---i"
(27)
h is straightforward
to show by assuming time dependence
of the
form e -i'°' that Eq. (27) is equivalent
to the frequency-domain
impedance
boundary
condition with the impedance
given by Eq.
(26). But before Eq. (27) is applied to any large-scale computation,
it is prudent to show that this boundary condition
would not give
rise to spurious instabilities.
B. Stability of Three-Parameter Time-Domain Impedance
Bounda_" Condition
Let us return to the boundary value problem of Fig. 4. It will be
assumed that the time-domain
impedance boundary condition (27)
is to be imposed at the x-z plane. The Fourier-Laplace
transform
of Eq. (27) is
A
Fig. 11 Map of the upper half ca plane in the F plane: a) ca plane and
b) F plane.
(28)
-co_ = [-,oRo +i(x_, + o_x_)]_.
Substitution
G
b)
of Eq. (9) into Eq. (28) results in the dispersion
relation
Equation
equations:
(31)
may
be
casted
into
the
following
system
of
(.,O2
(off - k")_
,xhere the branch
+ wR_ - iXlw
cuts of the function
2 = iX-i
(29)
(w2 - k:) _a are to be taken
such that 0 < arg(w-" - k'-) l/'- < rr; the left (right) equality
used if w is real and positive (negative). Let
-- ( "wz_ k2)½ + wRy, - iX_w 2
In Eq. (33) the term (Ou/Ox)o
backward difference according
ing Eq. (33) in time, we find
(33)
has been replaced by a seven-point
to the DRP scheme. Upon discretiz-
(30)
that is, the left-hand side of Eq. (29). By tracing over the contour
ABCDEFGH
in the upper half co plane, it is easy to establish that
the upper half w plane is mapped into the shaded region in the F
plane as shown in Fig. 11. The mapped region does not include the
negative imaginary
axis. Now X_t, on the right-hand
side of Eq.
(29), is negative. This means that no value of co in the upper half w
plane would satisfy dispe.-'sion relation (29). Thus, the solutions of
the initial boundary value problems are stable.
Numerical Implementation and Comparison with Exact Solution
We will consider an initial value problem associated
with the
normal-incidence
impedance
tube (see Fig. 8) to illustrate how the
time-domain
impedance
boundary
condition
(27) can be implemented. With respect to a fixed point in space, a transient acoustic
pulse produces a broadband pressure field in the frequency domain.
Thus, for a time-dependent
acoustic pulse problem, the broadband
impedance boundary condition should be used at the right terminal
of the tube.
C.
Equation (16) is the governing equation of the problem. Now,
at the impedance
surface, i =0, boundary condition
(27) may be
rewritten in the following form after eliminating Op/at by the second
equation
-z: +,_.,,oo+x
,uo
is to be
CO2
F(w)
-ru°l
1)1)
3
+ At E bj
jffi0
1
_l
a?uj
-Rov,
i + X_luo
-- j=-6
(34)
form of
But the value uo(, + 1) can also be found from the discretized
governing equation (16), namely, Eq. (21). The value u o(n+ 1) calculated by either Eq. (21) or Eq. (34) must be the same. This provides
the condition for determining
the ghost value p_t"). The explicit expression
p<._-
for pl "_ is
1
boaSZAt
+ At
_j'-o
(-
u oc.+,_+,.h._
-- Atbo
)=t
S."aj
jr-5
pj
]
(35)
/
of (16):
dt 2 = _
We will rewrite
Let
-
7x
this equation
_,-
_'-T["
as a first-order
+ X__u,,
system
(31)
in time.
u = 0.
duo
ro = -dt
The boundary condition on the left end of the tube is still _ven by
Eq. (18) except that the nonhomogeneous
terms should be omitted
as there is no incoming wave.
We will assume that the disturbance is generated by the following
initial conditions at t = 0:
(32)
p = exp[-0.0044(x
+ 83.333):]
cos[0.444(x
+ 83.333)
(36)
922
TAM
AND
AUR/AULT
Y
¢_i
_"
M
zero vel0ci_ layer
vortex sheet
o
treatment
panel
o
I
?
,
,
,
,
I
-150.0
,
,
i
Fig.
I
,
-50.0
-100.0
0.0
14
adjacent
Schematic
to an
led to an unstable
Fig.
12
normal-incidence
tion
waveform
Pressure
(DRP
of
tube
impedance
scheme)
and
the
incident
at t = 60.1:
, exact
....
acoustic
pulse
inside
time-domain
_,
the
solu-
solution.
O
e_
diagram
acoustically
showing
treated
solution.
a postulated
panel
zero-velocity
in the presence
The unstable
solution
of a mean
layer
flow.
is of the Kelvin-
Helmholtz type arising from the vortex sheet interface between the
mean flow and the zero-velocity
fluid layer. In standard duct acoustics analysis using frequency-domain
approach,
this instability
is
either not mentioned or totally ignored. _ We will now show that
the use of boundary condition (37) always gives rise to an unstable
solution.
The linearized momentum
and energy equations governing
the
sound field superimposed
on a uniform mean flow of Mach number
M in the x direction (see Fig. 14) are
o O
Q"o
av
?
av
= -vp
aS + M_
(38)
o
J
?
,
,
,
Fig.
13
tion
(DRP
,
°
,
--100.0
Pressure
normal-incidence
,
I
-_0.0
waveform
impedance
scheme)
and
....
i
,
,
,
°
--_.0
of
the
tube
, exact
reflected
acoustic
at t = 140.1:
_,
0.0
pulse
inside
time-domain
the
solu-
at+M
+V.v=0
By applying the Fourier-Laplace
it is easy to find that the solution
condition at y --+ _ is
(39)
transform to Eqs. (38) and (39),
that satisfies the outgoing wave
solution.
Here, the same dimensionless
variables as in Sec. II.C are used. This
choice of initial conditions
ensures that the center of the acoustic
spectrum of the incident wave at the surface of the acoustic treatment
panel has a frequency
of 2 kHz and a spectrum half-width of 0.5
kHz. In this example, the values of the dimensionless
parameters of
Eq. (27) are obtained by fitting Eq. (25) to the data of Fig. 3.
Figure 12 shows the computed pressure distribution
at time t =
60.1. At this time, the left half of the initial pulse is about to exit the
computation
domain, whereas the right half of the pulse is about to
impinge on the surface of the treatment panel. The dotted curve is
the exact solution. Figure 13 shows the reflected pulse propagating
away from the impedance
boundary
of the tube at t = 140.1. The
amplitude of the reflected pulse is considerably
smaller than that of
the incident pulse. Part of the acoustic energy is dissipated during the
reflection process off the impedance
surface. The exact frequencydomain solution is represented by the dotted curve. There is excellent
agreement between numerical results and the exact solution. This is
true in both the wave amplitude and phase. This example provides
further confidence in the use of time-domain
impedance
boundary
conditions.
IV.
A.
Impedance
Traditional
Boundary
of a Subsonic
Condition
in the Presence
Mean Flow
Model
In jet cannes,
the acoustic treatment panel is always placed next
to a mean flow. Traditionally, 5 the impedance boundary condition in
the presence of a mean flow is formulated with the assumption of the
existence of a very thin zero-velocity
fluid layer at the surface of the
treatment panel. At the interface of the zero-velocity
fluid layer and
the mean flow, the condition of continuity of particle displacement is
used. In dimensionless
variables, the frequency-domain
impedance
boundary condition,
after simplification,
may be written as
--iwp
+ Map = -i_Zu,
Ox
(37)
where M is the mean flow Mach number, and x is in the direction
of the mean flow.
In an extensive numerical study of the normal modes of a duct
with treatment panels, Tester _ found that boundary
condition (37)
__AL
_ /y]
(4o)
where (3 = co - Ma, k = (¢x: + _:)1/:, and _ = Colk. The branch
cuts of the function (&2 _ I)tD are the same as those stipulated just
after Eq. (9).
Substitution
of Eq. (40) into the Fourier-Laplace
transform
of
boundary condition (37) yields the following dispersion relation:
&2
(_
o[
- I)½
_-Z_ = --_MZ
(41)
Now lettheleft-hand
sideof Eq. (41)be denotedby f(_):
, + Z_
=
(42)
(&z _ 1)
Figure 15 shows the map of the upper half _ plane in the f plane
(shaded region) for X > 0. Since a can be positive or negative,
there will always be values of a for which the point representing
the right side of Eq. (41) lies in the shaded region of the f plane.
Therefore,
there will always be an unstable solution. If X is negative. a similar mapping procedure will show that there is always an
unstable solution.
The existence of a Kelvin-Helmholtz-type
instability renders the
boundary value problem ill posed for the time-domain
solution.
The instability is, however, nonphysical. As pointed out before, its
origin is in the postulate of a vortex sheet discontinuity right next
to the impedance boundary. In reality, no such vortex sheet exists
in the flow. It is an exaggerated idealization of a zero-thickness
boundary layer. The pertinent point to remember is that the concept
of impedance
is just a gross macroscopic description of the effect
of a treatment panel on the sound field. One could include the effect
of the mean flow in the definition of the impedance directly without
having to introduce its effect through the kinematic property of the
vortex sheet discontinuity. If this is done, there is no need to assume
the existence of a vortex sheet interface. A well-posed time-domain
impedance boundary condition may then be derived. This possibility
is discussed in the next subsection.
TAMAND AURIAULT
It can be easily shown that solutions of governing equations (38)
and (39) and boundary condition (44) or (45) are stable. The same is
true with boundary condition (27). That is, by omitting the fictitious
vortex sheet discontinuity
that is postulated in the traditional formulation of the impedance boundary condition, the resulting initial
boundary
value problems are well-posed and hence can be solved
by time-domain
methods.
H
A
G
I
1
a)
A
Rc(_)
V.
Im(7 )
-7,
For
._o(7)
b)
Fig. 15 Map of the upper half _ plane in the.f plane (X > 0): a) &
plane and b)f plane.
B.
Direct Interaction Model
The experimental studies of Baumeister and Rice 3"4 indicate that,
for Helmholtz resonator
type treatment panels, the jet flows at the
mouths of the resonators
or cavities interact directly with the mean
flow outside. This is illustrated in Fig. 2. There is no zero-velocity
fluid layer nor any vortex sheet discontinuity. When the boundary.
layer over the treatment panel is, indeed, thin, the jet flows at the
mouths of the cavities are directly coupled to the mean flow. When
this is the case, it would be more desirable not to introduce the
mean flow effect through the kinematic condition of a vortex sheet
discontinuity, but consider a direct interaction model by lumping
the mean flow effects on the definition of impedance. What this
amounts to is to define the impedance in exactly the same form as
Eq. (I) even in the presence of a mean flow: i.e.,
p = Zv,
(43)
where Z = R - iX is the impedance. Here Z is a function of the
mean flow. That is, R and X are flow Mach number dependent.
Equation (43) is to replace Eq. (37). The relationship
between R,
X, and the mean flow Mach number must be found experimentally
at this time. Further theoretical
and experimental
validation of this
suggestion would certainly be useful and needed.
It is easy to show by the mapping technique that the dispersion
relation arising from Eqs. (38) and (39) and boundary condition (43)
does not give rise to an unstable solution. In other words, they form
well-posed initial boundary value problems.
C. Single-Frequency and Three-Parameter Broadband
Time-Domain Impedance Boundary Condition
If impedance boundary condition (43) is accepted instead of (37),
then the single-frequency
time-domain
impedance boundary condition discussed in Sec. II and the three-parameter broadband model
impedance boundary condition (27) can again be used. Specifically,
for sound waves of a single frequency f2, we have the time-domain
impedance boundary condition
x < o,
X >_ O,
o.__[p
= ROy. _ Xnv.
Ot
923
Ot
X Or,
p = Rv. + ----
(44)
(45)
Helmholtz
Concluding
resonator
type
Remarks
of acoustic
treatment
panels.
impedance
is merely a macroscopic
representation
of the aguegated effects of the numerous microscopic
cavity flows that take
place in the surface region of the panel. In the presence of a mean
flow, the actual flowfield in this region is extremely
complicated.
So far, it appears that no experiment
has been performed
to provide
an adequate understanding
of the microscopic cavity flowfields and
their cumulative
effects. This is, perhaps, not too surprising,
for
it is very difficult to make accurate time-dependent
measurements
in the small confined space of the cavities. Without such knobqedge of the flowfield, the impedance
of a panel can only be found
empirically.
Recently, computational
aeroacoustics
methods have made impressive advances. It seems, therefore, that a feasible alternative
to study the microscopic
cavity flowfields is to use direct numerical simulations.
Such simulations,
when properly carded out, could
shed light on the physical processes
that lead to the damping and
phase shifting of the acoustic wave field adjacent to the panel surface. When perfected,
such time-domain
simulations
might even
offer a way to determine the resistance
and reactance
of acoustic
treatment
panels
from first principles.
Acknowledgment
This work was supported
Grant NAG 1-1776.
by NASA
Langley
Research
Center
References
Motsinger, R. E., and Kraft, R. E., "Design and Performance of Dun
Acoustic Treatment,"Aeroacoustics of Flight Vehicles: Theory and Practice.
NASA RP-1258, Aug. 1991, Chap. 14.
2Groeneweg, J. E, "Current Understanding of Helmholtz Resonator Arrays as Duct Boundary Conditions" BaMc Aerodynamic Noise Research,
NASA SP-207, July 1969, pp. 357-368.
3Baumeister, K. J., and Rice, E. J., "Visual Study of the Effect of Grazing
Flow on the Oscillatory Flow in a Resonator Orifice;" NASA TM X-3285.
Sept. 1975.
4Baumeister, K. J., and Rice. E. J., "Flow Visualization in Long Neck
Helmholtz Resonators with Grazing Flow," NASA TM X-73400, July 1976.
5Eversman, W. E., *'Theoretical Models for Duct Acoustic Propag'ation
and Radiation," Aeroacoustics of Flight Vehicles: Theory and Practice.
NASA RP-1258, Aug. 1991. Chap. 13.
6Tester, B. J., *'The Propagation and Attenuation of Sound in Lined Ducts
Containing Uniform or Plug Flow," Journal of Sound and Vibration, Voi.
28. No. 2, 1973, pp. 151-203.
7Zorumski, W. E., and Tester, BJ., "Prediction of the Acoustic Impedan_
of Duct Liners," NASA TM X-73951, Sept. 1976.
SLippert, W. K.R., "The Practical Representation of Standing Waves in
an Acoustic Impedance Tube:" Acoustica, Vol. 3, No. 3, 1953, pp. 153-160.
9MelIing, T. H., "An Impedance Tube for Precision Measurement of
Acoustic Impedance and lnsertation Loss at High Sound Pressure Level."
Journal of Soundand Vibration. Vol. 28, No. I. 1973, pp. 23-54.
retain, C. K. W., "Computational Aeroacoustics: Issues and MethodsY
AIAA Journal, Vol. 33, No. 10. 1995, pp. 1788-1796.
llTam. C. K. W., and Webb, J. C., "Dispersion-Relation-Preserving Finite Difference Schemes for Computational Acoustics:' Journal of Computational Physics, Vol. 107. Aug. 1993, pp. 262-281.
12Tam, C. K. W., Webb. J. C., and Dong, Z., "A Study of the Short Wave
Components in Computational Acoustics,"JournalofComputationalAcoustics. Vol. 1, March 1993, pp. 1-30.
13Tam, C. K. W., and .Dong, Z., "'Wall Boundary Conditions for
High-Order Finite Difference Schemes in Computational Aeroacoustics." Theoretical and Computational Fluid Dynamics, Vol. 6, Oct. 199-.'.
pp. 303-322.