NUMERICAL METHODS IN

ACOUSTICS

SESSIONS
Coupled-field Finite Element/ Spherical Harmonic Analysis
for Close-Packed Arrays
J. B Blottman III, A. J. Kalinowski
Naval Undersea Warfare Center, Newport, Newport RI 02841, U.S.A.

Acoustic projectors assembled in an array experience an interaction effect as a result of the coupling of their individual radiated
powers through the acoustic medium. A technique is formulated to predict the performance of an array that combines a finite
element model of a single transducer with an analytic description of the surrounding fluid and neighboring transducers in the
array by employing spherical harmonic expansions. The approach has been applied to evaluate a simple array of electrically
driven piezoelectric Class IV flextensional transducers. Results are compared to solutions using a coupled finite element-
boundary element method. [Work sponsored by Office of Naval Research, Code 321SS, Jan Lindberg Program Officer]

INTRODUCTION where Ajnm are coefficients to be determined, hn krj

A hybrid numerical method is proposed in which a
single transducer finite element model can be
combined with an analytic description of the
surrounding fluid to model a complete array. This
method takes into account multiple scattering among
the elements and is not limited to low ka . This
technique permits fast design of notional arrays
without the inherent remeshing and additional
computations required of traditional numerical
methods, for each change in array geometry.
(1 )
( )
is the spherical Hankel function of the first kind, and
Ynm is the surface spherical harmonic functions.
Substitution into the total pressure sum yields the
harmonic expanded pressure; however, each sphere is
defined in terms of different local coordinate systems.
To formulate the equations in a consistent, global
coordinate system, an addition theorem must be
applied to write spherical waves in local coordinate
systems as a series of spherical waves in some
reference coordinate system [2].

FORMULATION The surface velocity distribution is also expanded in

Suppose we have M transducers, surrounded by
virtual spheres of fluid and submerged in an infinite
fluid medium. The objective is to relate the sphere
surface velocities expanded as spherical harmonics to
the sphere surface pressures expanded as spherical
harmonics. We follow the work of Scandrett and
Canright [1]. Assume a time dependence of e-iw t for all
quantities. The total pressure at a point x is given by
r r M
r
P( x ) = po ( x ) + p j ( x ) ,
j =1
r j in the coordinate system of that sphere.
where p j (x ) is the total pressure from sphere j due to
r Consider a finite element approach to analyzing the
direct radiation and rigid body scattering, and p0 (x ) is
the incident pressure. The time independent, pressure
field external to the spheres is given by the solution to
the Helmholtz equation, subject to the boundary
r
conditions on the virtual sphere surfaces. Each p j (x )
can be expanded as a series of spherical harmonics,
n
terms of spherical harmonics. The pressure on the
surface is related to the normal component of surface
velocity by the continuity boundary condition, and the
velocity is then written in matrix form as a function of
the A coefficients, {V } = [B]{A}. Similarly, the
pressure on the surface of each virtual sphere is related
to the Ajl coefficients in matrix form {P} = [H ]{A}.
Finally, the pressure anywhere in the fluid medium,
identified by r0, is due to the superposition of pressure
contributions from all the virtual spheres. It is
expanded in terms of spherical harmonics and the
same set of coefficients Ajnm because they are defined
for the expansion of the pressure due to a given sphere
harmonic behavior of a piezoelectric transducer
submerged in a fluid. The formulation results in a
system of equations that can be written:
[KUU ]- w 2 [M ] {K }{U} = {F} .
UE
{K UE }
T
KEE fE Q

( )
p j rj ,q j , j j = A
n =0 m =- n
jmn ( )
h(n 1) Yn m q j ,j j , To combine this formulation with equations for the
fluid [2], divide the exterior wetted surface of the

SESSIONS
structure into regions or patches which coincide with
the faces of the finite element discretization. The
acoustic loading can be related to the surface pressures
by {F} = - [C][R]{P}, where [R] is a diagonal matrix
made up of patch areas and Cij is known as the
compatibility matrix which relates nodal DOF to
wetted surface patches. Similarly, the average patch
velocities are expressed as, {V } = - iw [C] {U }.
T
forming a system of equations that represent the
coupled structural-acoustic behavior of all the spheres,
{V } = {G} + [Z ]{P}. For M identical spheres, the
vector {G} is constructed by duplicating the vector
{E} M times and multiplying by the applied potential
for each sphere to form a column vector. The matrix
[Z] is constructed by diagonalizing a column vector
consisting of M copies of the matrix [Q]. Finally,

combining expressions for {V} and {P} yields:

Define [D] = [KUU ]- w 2 [M ]. After expanding the
finite element system, multiply first equation by
([B] - [Z ][H ]){A} = {G}.
[C]T[D]-1 and apply the loading and surface velocity
The matrices [B] and [H] are the result of the array
expressions to obtain:
equations and depend only on the array geometry. The
matrix [Z] and the vector {G} result from the structural
{V } = iw [C ]T [D]-1 {KUE }f E + iw [C]T [D]-1 [C][R]{P}.
equations. Given the applied potentials for all the
spheres in an array, the A coefficients may be solved
This equation relates the surface patch velocities {V}
for. Once the coefficients are known, The pressure
to the surface patch pressures {P} for an applied
may be computed anywhere in the fluid.
potential; FE. It is the boundary condition to apply to
the acoustic array of virtual spheres. APPLICATION
Boundary elements applied to the surface of the
transducer represent the surrounding fluid and provide The formulation is implemented for a pair of class IV
an implicit transfer of surface pressures and normal flextensional. The numerical model was developed in
velocities to spherical surface fluid pressures and Matlab with links to the Navys CHIEF BIE code
particle velocities on the virtual sphere. The BIE and the ATILA FE code. Each transducer is
provides A and B matrices relating surface pressure electrically driven at ka = 1 and kd = 3 , where a is the
and velocities, semi-major axis. The geometry is depicted in the FEM
model inset. A traditional FE-BE solution is included
A11 A12 P o B11 B12 V o for comparison. Good agreement is found.
A = ,
21 A22 Pi B21 B22 V i

where the superscripts in [A] and [B] denote the inner

and outer surfaces. Substituting the boundary
condition on the inner surface into the BIE expression,
the outer surface velocity harmonics are solved for in virtual spheres
terms of a set of A and B , matrices and the surface
pressure harmonics.

{V }= -[B]{P }- [A]F
o o
E

To combine the structural formulation with the array

formulation, the boundary condition must be expressed
in terms of a spherical harmonic decomposition of the
surface pressure and velocity fields. In matrix form,
{} { }
{P} = [Y ] Ps and {V } = [Y ] V s , where the lth
s
component of {P } is Pnm, and Y jl = Yn q j , f j .
m
( ) FIGURE 1. Far Field Pressure of Two Flextensional Array

Multiplying from the left by [Y]T, solving for {Vs} and REFERENCES
defining new {E} and [Q] coefficient matrices produce
1. Scandrett CL, Canright DR, J. Acoust. Soc. Am.
{V }= -{E}F - [Q]{P }.
S
E
S
vol.90(1), 1991, pp. 589595.
2. Blottman JB, Kalinowski AJ, Proc. Oceans 2000
The structural spherical harmonic relation is combined
MTS/IEEE, ISBN 0-7803-65554-2
with the spherical harmonic array formulation by

SESSIONS
 Coupled finite element/boundary element method
for acoustic radiation and scattering

A. Laviea and C. Vanhilleb

aUniversit dArtois, LAMH - Equipe "Modlisation et Acoustique", Technoparc Futura, 62400 Bthune, France
e-mail: lavie@fsa.univ-artois.fr
bESCET. Universidad Rey Juan Carlos. C/. Tulipan, s/n., 28933 Mostoles, Madrid, Spain

e-mail: c.vanhille@escet.urjc.es

Acoustic radiation and scattering by a complex structure can be modelized using a coupling between finite element method (FEM)
and boundary element method (BEM). This coupling between ATILA (FEM code) and EQI (BEM code) is efficient in the low or
intermediate frequency range. Specific techniques have been developed in the case of the steady-state problem: frequency interpo-
lation, decomposition of the fields in Fourier series, frequency derivative approach of the resonances. In the time-domain problem,
efficiency of the Helmholtz Kirchhoff integral equation (HKIE) for scattering by axisymmetric rigid bodies is discussed.

INTRODUCTION number and D the influence distance) in the exponential

Modeling of acoustic radiation and scattering by solid
elastic structure immersed in an infinite medium is of
main interest. The fluid-structure problem can be solved
by coupling two methods: the FEM (elastic body beha-
viour) and the BEM (acoustic medium behaviour). In
the case of the steady-state problem and in order to in-
crease efficiency of this numerical tool, some improve-
ments have been implemented and are presented. Next,
scattering in time-domain using the HKIE and its nume-
rical instabilities are discussed.
STEADY-STATE PROBLEM
term of the free-space Green's function:
ikD
1 e
g ( r, r' ) = ------ ---------- D = r r' (1)
4 D
By extracting a quantity close to kD from the exponen-
tial, a very slow frequency variation coefficient is obtai-
ned. So, a linear frequency interpolation can be carried
out at a calculation frequency situated between two sto-
rage frequencies. The original coefficient is then recove-
red after multiplying by the sinusoidal oscillating term.
This technique is very accurate when the storage fre-
quencies are 0 Hz and the top-frequency of the spec-
trum, given by the "/4 criterion" associated with the
quadratic isoparametric discretization[2].

Frequency interpolation Fourier series

Although the BEM is a very efficient numerical techni-
que for acoustic analysis for a single frequency, it may
lose its advantage for a multi-frequency run. The main
reason is that the integrals of the coefficients of the sys-
tem are frequency dependent. For each different fre-
quency, all the components in the coefficient matrices
and vectors need to be recalculated.
A way to decrease this multi-frequency difficulty is to
use a linear frequency interpolation technique[1]. The
idea is to eliminate the oscillation of the coefficients ma-
trices with frequency. This oscillation is due to the im-
portant variation of the kD term (where k is the wave
For scattering problems, most practical targets are axi-
symmetrical. Axisymmetric modeling can not be used
directly when the acoustic wave strikes the body at obli-
que incidence. To avoid three-dimensional modeling,
the pressure and the displacements are presented in Fou-
rier series on azimutal angle. Only the generatrix of the
body is covered by the mesh. The computation is carried
out up to convergence for every term of the series. By
using this method, one decreases the central processor
unit time, memory, disk storage, and the numerical ero-
sion. This is equivalent to the enlargement of the upper
bound of the frequency range[3].

SESSIONS
 Frequency derivative approach
Acoustic scattering by elastic bodies is usually analyzed
in terms of resonances. The scattered far-field pressure
ps can be split into the resonance pr and the background
pb components. At small or moderate reduced frequency
x, pb varies slowly with frequency. Considering an iso-
lated resonance xr and a small frequency shift x, we ha-
ve:
p s ( , , x r + x ) p s ( , , x r )
p r ( , , x r + x ) p r ( , , x r )
x p r ( , , x )
x -20
x = xr
where and are the spherical coordinates and the last
term is obtained from a Taylor expansion. For an isola-
ted resonance, we assume that the resonance component
can be presented as:
p r ( , , x r ) = F ( , )G ( x ) (3)
where F(,) is the angular pattern and G(x) the varia-
tion of the resonance component with frequency. This
technique can be extended to surface displacement.
Then, the identification of the resonances is achieved in
two steps:
- after calculation of the backscattered form function,
geometrical accidents are located;
- the above technique is applied for two frequencies very
close from the selected accident and provide spatial dis-
tribution of the surface displacement of the resonance
component.
For some targets (especially cylindrical shell bounded
by hemispherical endcaps), A and S0 wave resonances
have been identified[4].
TIME-DOMAIN PROBLEM
For scattering by axisymmetric rigid bodies in time-do-
main, the HKIE has been implemented in the EQI co-
de[5]. The space discretization is based on the space
finite element, i.e., isoparametric 3-node elements with
quadratic variation. To achieve the time discretization,
two techniques are developed: the time finite difference
method (TFDM) and the time finite element method
(TFEM).
Numerical model is estimated for scattering by a rigid
sphere (radius a). The incident pressure is given by:
t kz
p inc = p 0 sin ( t kz ) 1 cos -----------------
2 5. A. Lavie, B. Dubus and A. ElGhaouty, Proceedings of the
where z is the distance to the source, t is the time and
is the angular frequency. The normalized total pressure
is computed on the front side of the surface target versus
the reduced time (c sound velocity) with ct/a=0.119
and ka=4.2.(cf. Figure 1)
2
TFDM
1.5 TFEM
1
0.5
p/p0
0
-0.5
-1
(2)
-1.5
5 10 15
ct/a 20 25 30
FIGURE 1. Pressure on the front side of the rigid sphere
Analytical and TFEM curves practically coincide up to
ct/a=20. TFDM generates a fictitious wave and provides
wrong results for 5<ct/a<10. For the two discretizations
(more pronounced with TFDM), a strong fictitious wave
which diverges with time is observed. This instability
could be the corollary in the time-domain of the irregu-
lar frequencies in the frequency-domain.
CONCLUSION
Using FEM and BEM is very attractive to deal with
acoustic radiation and scattering. In order to improve the
numerical behaviour of the EQI code in the steady-state
problem, specific techniques have been developed. Mo-
reover, for scattering by elastic target, a method to iso-
late and identify resonances has been proposed. In the
time-domain problem, development of time-domain in-
tegral formulations free from instability is under study.
REFERENCES
1. G.W. Benthien, Technical Report 1323, Naval Ocean Sys-
tems Center, San Diego (1989).
2. C. Vanhille and A. Lavie, Acustica, 84, 884-893 (1998).
3. A. Lavie and B. Dubus, Proceedings of the 4th French
Congress on Acoustics, Teknea Ed., 2, 801-804 (1997).
4. B. Dubus, A. Lavie and N.D. Veksler, J. Acoust. Soc. Am.,
102, 3523-3529 (1997).
(4)
5th French Congress on Acoustics, Presses Polytechniques et
Universitaires Romandes, 129-132 (2000).
SESSIONS
 Numerical Modelling of some Problems in Nonlinear
Acoustics
C. Campos-Pozueloa, C. Vanhilleb and B. Dubusc
a
Instituto de Acstica, C.S.I.C. Serrano, 144 28006 Madrid, Spain. ccampos@ia.cetef.csic.es
b
E.S.C.E.T. Universidad Rey Juan Carlos. Tulipn, s/n. 28933 Mstoles, Madrid, Spain. c.vanhille@escet.urjc.es
c
IEMN, UMR CNRS 8520, Dpartement ISEN, 41 Boulevard Vauban, 59046 Lille Cedex, France.
Bertrand.Dubus@isen.fr

Some recent developments in numerical nonlinear acoustics are presented. First a three dimensional perturbation approach based
on the finite-element method is described. This procedure can predict the propagation of acoustic fields produced by sources of
arbitrary geometry as well as the pressure distribution inside a three dimensional cavity including boundary layer absorption. Its
main limitation is due to its range of validity limited to waves of finite but moderate amplitude. We then describe a second
approach: a numerical model for nonlinear waves and weak shocks in thermoviscous fluids based in a time-domain finite-
difference algorithm. This algorithm does not present any practical limitations about the amplitude of the wave but it is referred
to one-dimensional problems. Some future trends are also commented.

INTRODUCTION density, u is the displacement, and B are the
viscosity and bulk viscosity respectively, and and
Nonlinear acoustic phenomena are of practical are characteristic constants of the fluid. This state
interest since the 1930s, and particularly today in equation reduces to the ideal gas state equation for
applications such as: industrial use of high power = 0 and = , where is the specific heat ratio.
ultrasound, sonar, acoustic microscopy, medical
ultrasound and non-destructive testing. In these
applications, the need to account simultaneously for Second order three-dimensional solution
the combined effects of nonlinearity with absorption
and geometrical characteristics of the system creates a By combining Eqs. (1), (2) and (3) a second order
very hard analytical task. In the last years numerical three-dimensional equation for the pressure is obtained.
modelling becomes an important and useful tool for To solve this equation a perturbative method is applied.
solving this kind of problems [1]. Two recent The solution is assumed to be the addition of two
numerical developments concerning other problems in terms, the linear solution, pl, and a second order
nonlinear acoustics are presented in this paper. correction, p2, being p2 << pl . All the terms of third or
higher orders are neglected. Therefore,
2 2
NUMERICAL APPROXIMATIONS p p0 pl
2 , where c0 is the low-amplitude
0 c0 0 c02
Fundamental equations velocity of sound. Considering the particular case of an
initially generated harmonic wave a linear spatial
We consider nonlinear waves in homogeneous equation is obtained for the second order correction
thermoviscous fluids. In order to obtain the governing which can be solved by classical methods. A numerical
nonlinear wave equations in Lagrangian coordinates, solution based on the finite-element method is
the isentropic state equation of Tait-Kirkwood and proposed to solve three-dimensional acoustical
equations expressing the conservation of mass and problems of arbitrary geometry [3]. The effect of the
momentum are considered [2]. boundary layer has been modelled by considering a

p+ (1) complex impedance at the wall of the cavity.
=
p0 + 0
0   Time domain finite-difference solutions
= u (2)
From Eqs. (1) to (3) a fully nonlinear one-
   
u2
2 u 1   u (3) dimensional wave equation for the displacement is
0 2 = p + + +
t t B 3 t written without truncation. A time-domain numerical
where p is the pressure, t the time, is the density of approach based on a new finite-difference algorithm is
the fluid, p0 is the ambient pressure, 0 is the ambient developed to solve this problem (SNOW-AC). The

SESSIONS
complete solution is carried out in the time domain, wave is strongly distorted: the changes are very abrupt;
i.e., all the harmonic components are obtained by only an asymmetry between rarefaction and compression
one solution. Time periodic, pulsed and any other zones appears. The pressure nulls vary within a zone
excitation conditions can be considered. Since the fluid that takes almost all the tube: there is not a real node
is at complete rest at the outset, the transitory phase is for the pressure; it is a quasistanding wave. In fact,
completely modelled. The model can support any small the pressure wave has formed a shock that propagates
attenuation parameters without any stability or from the emitter to the reflector (from t = 0 to 0.5
convergence problem. Cases from linear to strongly times a period) and vice versa (from t = 0.5 to 1 times
nonlinear behaviour can be studied [4]. a period).
Pressure (Pa) 4
x 10
0
RESULTS 6
0.1
5

Second order three-dimensional solution 0.2 4

0.3 3

Number of periods
The numerical development is applied to several 0.4 2
cases and results are compared with experimental data. 0.5 1
The model was tested in free-field conditions and for
0.6 0
three-dimensional standing waves. Results referred to
the near field of a directional transducer showed a very 0.7 -1

good agreement with experimental data [3] confirming 0.8 -2

the validity of the model for progressive waves. In Fig. 0.9 -3

1 an axisymmetric three dimensional cavity excited by 1 -4
a transducer of complex structure is simulated. 0 2 4
Spatial coordinate (m)
6 8
-3
-3
Pressure (Pa)
x 10
x 10
0
a) 1600

1400 FIGURE 2. The steady state pressure amplitude as a

5
function of time and space at the resonance frequency
Axial coordinate (m)

1200

1000
10 800

600
CONCLUSIONS
15 400
200 Modelling the nonlinear field of actual transducers
0 0.005 0.01 0.015 0.02 0.025 0.03
0 and real cavities opens up new possibilities in design
x 10
-3
Radial coordinate (m)
Pressure (Pa)
for industrial processing where high intensity effects
0 100 are important and nonlinear behaviour cannot be
b)
80
neglected. However, the propagation of finite
5
amplitude waves through fluids involves, besides the
Axial coordinate (m)

60
nonlinear distortion of the waveform, well described
10
40
by the algorithms presented here, other associated
nonlinear effects such as acoustic radiation pressure,
15 20
acoustic streaming and cavitation in liquids. Future
0 studies should focus to solve these problems.
0 0.005 0.01 0.015 0.02 0.025 0.03
Radial coordinate (m)
FIGURE 1. Pressure field distribution in an REFERENCES
axisymmetric 3-D cavity. a) Fundamental b) Second
harmonic 1. Ginsberg, J.H. and Hamilton, M.F. in Nonlinear
Acoustics, Ed. by Hamilton, M.F. and Blackstock, D. T.
Time domain finite-difference solutions (Academic Press), (1998)
2. R. T. Beyer, Nonlinear acoustics in Physical Acoustics,
The numerical model for nonlinear waves and weak ed. W. P. Mason (Academic Press, New York) Vol. II-
shocks in thermoviscous fluids is applied to the Part B, pp 231-332 (1965)
3. Campos-Pozuelo, C., Dubus, B. and Gallego-Jurez, J.A.,
analysis of one-dimensional cavities and plane waves J. Acoust. Soc. Am., 106, 91-101 (1999)
propagation. Fig. 2 shows the steady state pressure 4. Vanhille, C. and Campos-Pozuelo, C., J. Acoust. Soc.
amplitude (function of time and space) for an air-filled Am. 109 2660-2667 (2001)
rigid-walled tube of length c0 /(2 f ) (f is the excitation
frequency) and an acoustic Mach number M=0.1. The

SESSIONS
 Mid-Frequency Analysis Using a CMS Approach

J. M. Cuschieri and T. Fremiot

Center for Acoustics and Vibration, Department of Ocean Engineering,

Florida Atlantic University, Boca Raton, Florida, USA.

In analyzing the behavior of a structure/system, the low frequency response can be readily determined using deterministic
approaches. The high frequency behavior can be estimated using statistical based solutions, most common of which being
SEA. In between the high and low frequencies, there is a region the mid-frequency region for a lack of a better description,
where the response of the structure is very much controlled by the structure modal behavior, and slight variations in the
physical characteristics and parameters of the structure can create significant differences in the response. General statistical
approaches, such as SEA, do not capture the modal behavior and therefore will not closely represent the behavior of the
structure. Therefore an approach based on deterministic methods with the introduction of statistical variations is required.
The question is to how efficiently introduce the statistical variations. Using an approach developed by Mace and Shorter,
based on the component mode synthesis (CMS) method and the perturbation of modes, an analysis is performed of a T-beam
structure for which response results are available in the literature. CMS based results using perturbations of the structure
physical characteristics or the modes are computed and compared to the available results.

INTRODUCTION estimate the behavior of a T-Beam structure, for which

The response of a structure is generally predicted or
estimated using deterministic Finite Element Analysis
(FEA) or statistical based energy analysis such as
Statistical Energy Analysis (SEA). When the frequency
increases and wavelengths become shorter, the FEA
model requires a high mesh resolution, making the model
computationally demanding. Besides the increase in
computation time, the response also becomes
increasingly sensitive to perturbations in the physical
properties. The use of a deterministic FEA to
characterize the typical behavior of a structure may not
be appropriate at mid to high frequencies. SEA is an
approach to estimate the general behavior of a structure
typically associated with high frequencies. To emphasize
the sensitivity of the response with the physical
properties, SEA employs a statistical description of the
structure. The energy flow is calculated from the
responses of an ensemble of similar structures. This
ensemble is defined in terms of the uniform probability
of the local natural frequencies within a bandwidth.
In the mid-frequency range, a technique that
combines the benefits of FEA and SEA is necessary. In
this paper the Component Mode Synthesis (CMS)
technique that uses substructuring similar to SEA, but
where the response is predicted using FEA, is used to
response results are available in the literature.
Furthermore, perturbations in the response of the
structure are introduced to simulate the influence of
variations in the structural properties. Statistical
qualities from these results are superimposed on the
predicted behavior to give an indication of the expected
response variations.
COMPONENT MODE SYNTHESIS
Structures often consist of various connected
components or subsystems. The global behavior of a
structure can be described in terms of the behavior of
subsystems. This approach is the basis of the Component
Mode Synthesis (CMS). Hurty [1], introduced the CMS,
defining three types of modes: rigid-body, constraint and
normal modes, where the rigid-body modes represent the
component displaced without deformation. The
constraint modes are defined by producing a unit
displacement on each redundant constraint in turn, with
all other constraints fixed. The normal modes are the
modes of vibration of the substructure with fixed
coupling nodes.
The analysis of the individual subsystems is first
carried out to obtain the subsystem stiffness and mass
matrices, from which the three types of modes are

SESSIONS
obtained. The procedures used to formulate the different
modes for each subsystem and to assemble the
subsystems to form a reduced-order model of the original
system are explained in [1]. Craig and Bampton [2]
provide a simplification of the CMS method by
describing the global behavior of a structure in terms of
the behavior of the subsystems and treat all interface
degrees of freedom together. Hence, the displacements of
the subsystems are only composed of constraint modes
and normal modes.
Having obtained the normal modes, accurate results
can be obtained using fewer normal modes than the full
compliment of modes. A compact formulation, and a
less expensive calculation is thus possible.
In order to include moderate uncertainties in the
response of the T-Beam an ensemble is defined in terms
of perturbations of the local natural frequencies [5]. The
distributions of the local natural frequencies are assumed
to be Gaussian. Coefficients of variation and correlation
quantify the relationship between the perturbations of the
local natural frequencies.
A typical result from the CMS analysis for the power
ratio (ratio of dissipated to input power) for leg A is
shown in figure 2, which includes a typical result and 2
standard deviations of the predicted response.
100
M cP ow
90 M ean - 2 S D
M ean + 2 S D
80

T-BEAM STRUCTURE 70

Power ratio leg A

60

A previous series of studies in the area of structure- 50

borne noise [3, 4] considered a T-beam for which both 40
experimental and prediction results have been obtained.
30
The T-beam exhibits a modal behavior up to a few kHz,
hence making it ideal for a mid-frequency analysis. 20

Fig. 1 shows the specifications of the T-Beam structure. 10

(a)
0
0 100 200 300 400 500 600 700 800 900 1000
F requency (H z)

Fig. 2. Power ratio with statistical variations, leg A.

Fig. 1. Properties of the T-Beam Structure.
Applying the CMS approach to the T-beam, the
structure is divided into three subsystems. A direct finite
element program is used to generate the mass and
stiffness matrices for each subsystem. Each subsystem is
made of 16 elements. The elements degrees of freedom
are the axial and transverse displacements, and rotation.
The global T-beam has 48 elements and 49 nodes. The
total number of degrees of freedom is 147.
ENERGY FLOW FROM CMS
The results from a fixed interface Component Mode
Synthesis are used as inputs in the calculations of power
inputs, subsystem energies, and power dissipated. The
energy flow is presented in term of Energy Influence
Coefficients. The results from the power input and ratios
analysis are compared with available data.
CONCLUSION
A technique, which combines Finite Element (FE)
analysis with statistical concepts appropriate for mid
frequencies, is presented and applied to a T-Beam
structure.
REFERENCES
[1] W. C. Hurty 1965 American Institute of Aeronautic and
Astronautics Journal 3(4) 678-685. Dynamics analysis of
structural systems using component modes.
[2] R. R. Craig and Bampton 1968 American Institute of
Aeronautic and Astronautics Journal 6, 1312-1319.
Coupling of substructures for dynamic analysis.
[3] J. M. Cushieri, C. Burroughs. G. P. Caroll, May 1994 Proc.
Noisecon94, 541-544. Evaluation of structure-borne noise
prediction techniques.
[4] J. M. Cushieri, C. Burroughs. G. P. Caroll, April 1998 Proc.
Noisecon98. Evaluation of structure-borne noise prediction
techniques Review.
[5] B. R. Mace and P. J. Shorter 1997 Proceedings of the 5th
International Conference on Sound and Vibration. Energy
flow models from finite elements: an application to three
coupled plates.

SESSIONS
 Evaluation of the Earmuff-Earcanal System Using Finite
Element Method for Grazing for Impulsive Noise Incidence

F. Vergaraa, S. N. Y. Gerges a and R. S. Birchb

a
Department of Mechanical Engineering, Federal University of Santa Catarina (UFSC), Florianpoli s, SC, Brazil
b
Impact Research Centre, University of Liverpool, Liverpool, England

The effectiveness of hearing protectors for high impulse noise levels remains the subject of research with objective testing
techniques using acoustic test fixtures. Low cost and quick methods to investigate the performance of a protector to the final
qualification are desirable. This work examines the use of finite element method (FEM) and some of the phyical characteristics
of the grazing impulsive sound pressure incidenceto quantify the hearing protector characteristics.

INTRODUCTION and c=1500 m/s) whose thickness was 2 mm. The

In situations where there are high amplitude impulsive
sorces is recommended the use of the hearing
protection. Impulsive noise is dangerous to hearing
because the human auditory system is unable to sense
the peak level accurately in relation to the steady state
noise levels. The most significant physical factors of
an impulse noise are normally: Peak Pressure Level,
Rise Time and Duration of main pressure pulse (non
reverberante waves). The external ear comprises the
pinna, auditory canal and tympanum. A canal length
between 2737 mm and diameter of 7,5-9,0 mm at the
tympanum are normal dimensions for the average
adult. The passive earmuff, may be regarded as s low
pass filter against impulsive noise. The finite element
method can be used to solve the acoustics wave
propagation problems in the time domain and serves
as a quick and low cost tool for the study the effect
the noise reduction in the precense of the impulsive
noise [1,2].
foam was considered as a fluid and porous material to
five differents earmuffs: Protector 1 (=1,21 kg/m3;
c=343 m/s), Protector 2 (=98 kg/m3; c=320 m/s),
Protector 3 (=98 kg/m3; c=320 m/s, porosity 0,9,
resistivity 25000 Ns/m4 and structural factor 7,9),
Protector 4 (=950 kg/m3; c=1050 m/s, porosity 0,5,
resistivity 25000 Ns/m4 and structural factor 7,9) and
Protector 5 (=950 kg/m3, c=1050 m/s). The coupling
between the protector and the auditory canal was a
cushion simulated as a soft rubber material (=950
kg/m3; c=1050 m/s). The cavity of air is treated as a
homogeneous medium (=1,21 kg/m3 and c=343
m/s). Boundary conditions for the canal model were
imposed at the walls and end of the canal. An
impedance value of 108 Ns/m5 was imposed at the
position representing the tympanum, a value of 104
Ns/m5 was used to represent the impedance of the
walls of the auditory canal [1].

NUNMERICAL SIMULATION
The model is a finite element mesh that considers an
earmuff hearing protector coupled to the auditory
canal, inside of the shock tube whose internal
diameter is D (145 mm) and length is 2D (290 mm)
with air impedance (density =1,21 kg/m3 and sound
velocity c=343 m/s), the thickness of the tube is 2,5
mm being plastic material (=1200 kg/m3 and c=1500
m/s). The protector was modelled as curved type with
dimensions 55 mm x 100 mm and the auditory canal
was projected as a simples straight rectangular two
dimensional of 30 mm x 8 mm (see Figure 1). The FIGURE 1. Mesh of the shock tube and earmuff-earcanal
system applying grazing pressure incidence.
protector shell was defined as a plastic (=1200 kg/m3

SESSIONS
The sound excitation at the entrance of the shock
tube, the experimental and numerical distribution of
sound pressure at the eardrum position are showed in
the Figure 2. Initially, were made some previous
simulation tests considering the five earmuffs
protectors and an grazing ideal impulse with three
differents rise time, namely: short (0,5 ms), medium
(3 ms) and long (6 ms) and maintaining constant the
peak amplitude (1 Pa) and the total duration (8,5 ms).
The results obtained from this simulatios are showed
in Table 1. Also, the grazing sound presure incidence
was imposed as an impulse obtained from the
experimental tests from the shock tube system, being
possible to generate and to controll a plane wave FIGURE 2. Impulsive excitation, numerical and
pulse and his physical properties [2]. The finite experimental response pressure at the eardrum position.
element modelling was carried out controlling the
Newmark parameters and chosing an appropriate time
step (343 m/s x 7,2 s = 2,5 ms) [3].

Table 1. Physical factors of the three ideal impulsive noise: short, medium and long.
Pulse Rise Time [ms] Duration Time [ms] Peak Pressure [Pa]
Short Medium Long Short Medium Long Short Medium Long
Excitation 0,50 3,00 6,00 8,50 8,50 8,50 1,00 1,00 1,00
Protector 1 2,50 3,88 6,88 4,50 6,25 8,88 1,06 0,80 0,77
Protector 2 2,38 3,88 6,88 4,13 6,25 9,00 0,85 0,65 0,66
Protector 3 2,38 3,88 6,88 4,13 6,25 9,00 0,85 0,65 0,66
Protector 4 3,00 4,50 7,25 5,63 7,50 10,13 0,39 0,36 0,33
Protector 5 3,00 4,50 7,25 5,63 7,50 10,13 0,39 0,36 0,33

DISCUSSION AND CONCLUSIONS model. This work is a part of a research project in

In the three cases of ideal pulses it can be observed
that when the rise times of the excitation impulse are
increased the rise times of sound pressure response, at
the eardrum position, are increased. The peak of the
impulse pressure is delayed to the short, medium and
long pulses when is used the earmuff protection in the
auditory system. The hearing protector impulse noise
attenuation, at the peak, increases when are changed
the physical properties of the material, principally the
acoustics impedance (density and sound velocity) of
the foam material. The numerical acoustics modes
obtained using three differents absorbent materials
(air, foam and rubber) inside the earmuff cavity were:
235,6; 245,3 and 118,3 Hz, respectively. The
frequencies associated with the oscillations of the
pulse, inside the air cavity, varying between 222,22-
275,6 Hz of the protector 1, 235,29-307,69 Hz of the
protector 2 and 3, and 181,81-228,57 Hz of the
protector 4 and 5. The hearing protector that gives the
better attenuation corresponding to earmuff with high
material density. The measured time history sound
pressure inside the earmuff has the same form to the
predicted FEM values and the values get closer
changing the earmuff material properties, in spite of
that the headband force are not considered in the FEM
progress where some geometries of earmuff
protectors are studied, for example, the volume and
the shape of the surfaces, others material properties of
the protectors, the high levels of sound pressure and
differents angles of incidence.
REFERENCES
1. F. Vergara, Modelagem Numrica e Medies Acsticas
de Protetores Auditivos para Rudos Impulsivos , Master
Dissertation, 1999.
2. F. Vergara, S. N. Y. Gerges, R. S. Birch, Fi nite
Element Method (FEM) Model for the Hearing
Protector Noise Attenuation for Impulsive Noise,
InterNoise 2000, Nice, 2000.
3. J. P. Coyette, Application of Finite Element and
Boundary Element Models to Transient Acosutic
Problems, IX International Conference on Vehicle
Structural Mechanics and CAE, Michigan, 1995.

SESSIONS
 


 
 
 



!"
#!
$

%'&)(+*,-'.0/123+&54
6798
:
;
<
=?>A@B79C DE;GFIH J H :
H @
>A=
7LKM=
;
N
@OFIH 79:
PBH J QB@RS@OFIHTJ ;GUWV DYX5>[Z0\E]
^798
;
PB<`_a8
b`cedefhgi]
jM79kGU@B7TFIH 798?l@m
mG]
fCYn
op
m
q`fh79@OFIN
@B;G]rs@B79ts8
;Gu
cts8
J v wFIPB<
:
b
@B79H xy@B8
N
z
{ J |T> }{ >A<GU{ N
@
~ 5i5 5I55LB55T5L BaB T 5G5

5  a T5`L  aA 
aLa) 55S5 BB
 B SLOL5 5 SG~LO 55 5+ 55B B5 B5 B5O 5 BMBS O 5OEL555 )O M M 5
S BB55 T 55B5S T 5s5O+5) 5B sO+ 5B B SG~+Oi 55 5+ 5O 5 TO
5 5555S 5y 5 5 55 BT5 5  T 55555Oi 5' 9S BB5 5O0L55B5S
 55 5)555 ' 5 5 B M55 5T5
aL 55 55` B5 T 5 5 5M5G' ' B M 5
555 55IS OO
AA Dx

i 
 B
9BG [5B
B 5T[
E s ik-+11 s ik +1 s ik++11
[5
YOO B[ 5
EWB Sh+s[BO E B t k +1

Discrete time
BGMOBM[ Y [B WaOT
a s v ik-+1122 v ik++11 22 v ik++31 22
T Y O B
9OO[ LO[ E 9  BB 
t k +1 2 Dt

O[ BBL9 L BEW OEBTBO[ B Y
E+
Y0+ tk
BB
S[ i
O
E O9OT  O[O
 BG k
s ik s ik+1
W

BTS T TGMW[O B+BG BT9O `

9G[M O s i-1

WG
M
iWGOO BEM

[O[ G9 WBLGBG O Discrete space

TO[ Bs
9OO[ LO9 G
O OOOO BTBBBB !#"#$&%('*) 5
aL BBB 5O 5 
O O 5 BG
OO[ G
B G 55 5 S 55 5i5 
BEWO 5
Y [5Ts9 T[
O
Sh+9O W59S "T
E s`
B9W
Dx
GWB
O
[ B W Sh+ 9O BhE 2
T[B
OB`OY[B9
`aG9O 09 Me B T s ik-+114 s ik++114 s ik++31 4

OsG 'E B
9 G 50
O
) " G t k +1
Dt

BB 
GBO[ G OB 5T[  9G G 9

t k +1 2
t k +3 4
s ik-+1142 s ik++1142 s ik++31 42 2

 tk
t k +1 4

O a M G T
BT B99I `OY[ B B
B
k k k
s i-14 s i+14 s i+ 34

TWB GT i 9 O sB
BY B [


t k +1
BM
O9 Ti+ OT BG O9 G
iWE[ G
 
t k +3 4 vik-+1324 vik +3 4 v ik++1324 Dt
2
[ 5
O O[ 
Ih
G 5 BB
 t k +1 2
v ik-+1124 vik +1 4 v ik++1124
t k +1 4

 Y

G tk
!#"#$,+'-G5 +95 55L5O 55T 5


 
   



B S 5iT 55I 5+50 S
iMT[5OBh`OY[59 Oy hY9B`a )Ga B
B T
O"
9OO[[ W
`
"O
9OO[TBBB
G
2 1 2 2 1 0/ . 4

O[Y `

9BBG 0G
OBBOBO


 3

  3  
3

B A@ B A@



 




 






:<; = > B A @ 9 8 7 6 C A@ ?  6 C A@ ?


ED

C



 
?





   
? 5 ?


 
   

 
     Y + 
BO[ GW
 Y B BEMG O9` TBBsO [





WGT O[B aTG
Y OY[
 [ B


B
O9' OB B BO[G'
TWY9BaT

+ T B9O+[ 
`G iO `
9 [ G T

 
O T
E  9 WG[ O[B 4 B

SESSIONS
 1 1 1

0.75 0.75 0.75

v_x [arbitrary units]

v_x [arbitrary units]

v_x [arbitrary units]

0.5 0.5 0.5

0.25 0.25 0.25

0 0 0

-0.25 -0.25 -0.25

-0.5 -0.5 -0.5

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
x [cm] x [cm] x [cm]

$" !
" 6
1 1 1

0.75 0.75 0.75

v_x [arbitrary units]

v_x [arbitrary units]

v_x [arbitrary units]

0.5 0.5 0.5

0.25 0.25 0.25

0 0 0

-0.25 -0.25 -0.25

-0.5 -0.5 -0.5

0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
x [cm] x [cm] x [cm]

% "
" 7

8
5 &
( 9 # 9 ! ) 5 0.0003

) 9 9 0.0002

) &
v_x [arbitrary units]

0.0001
" # 0

-0.0001
&
-0.0002




(
) (

  


( 2 ) (

(


( 2 )
 

* 2 4
( 2 4 -0.0003
0 10 20 30 40
* 
) 
* t [us]
0.0003

* 3 # 0.0002

&
v_x [arbitrary units]

0.0001

( 2 4

* 2 4


(
( 0





( 2 )

( 2 )


( 2 )
( 2 ) &


(

 ( 2 4 ( 2 4


-0.0001
) )
-0.0002
7 # -0.0003
0 10 20 30 40
# t [us]

( * * " # % <
4 & : & = !
( ;
( ) ?
"
& ' ( ) ( ) "
) ) ?

* 4

# 6 ! )
"
* ( @ 

   & #!& 4 = ( A A 5
&
) * ) B / - &      !+ + & = ) ( A A ;
& )
% < = * &      !+ % & ) ; C 4 ( A A :
! 4 , D E F D - G ! & !"!# 
$ & $* &
5 & > ) ) A : ( A A :

SESSIONS
 Numerical modelling of low-frequency sound propagation over
poro-elastic ground
C. Madshus, A.M. Kaynia and F. Lvholt
Geomechanics Division, Norwegian Geotechnical Institute, N-0806 Oslo, Norway

The long distance outdoor propagation of sound is affected by meteorological conditions, topography, vegetation and ground
properties. This paper focuses on the last effect by using data collected from an extensive series of air blast tests in Norway as
well as a two-dimensional numerical model. The model consists of an acoustic medium over a layered poro-elastic ground
subjected to plane acoustic waves. From the numerical solution of the coupled wave equations in the two media expressions are
derived for the acoustic and seismo-acoustic impedances. The paper presents a brief introduction of the air blast tests, typical
recorded signals and numerically derived impedance values.

10
INTRODUCTION

p (Pa)
0

This paper deals with long-range outdoor -10

propagation of low frequency impulsive sound, as 0.04

v (mm/s)
typically generated by air blasts and firings from heavy 0
artillery. Such sources may disturb neighbours -0.04

indirectly, more through the vibration and rattling in 1 2 3 4

their houses induced by the impulsive sound pressure,
than directly by the impulsive sound itself [1].
Since human perception of whole body vibration
covers frequencies down to 1 Hz [2], and the funda-
mental natural frequency of buildings is typically in the
range 1-10 Hz, tools for planning and controlling blast
and shooting activity must account for sound propa-
gation at low frequencies. At these frequencies
meteorology, topography, and ground conditions may
affect the propagation differently than at higher
frequencies [3].
This paper focuses on the interaction between air
pressure and ground at low frequencies, where
mechanisms other than the usually studied porous
effects [4], may be important. For developing
numerical models in this new area it is vital to have
access to experimental data as a guide to see the
controlling mechanisms and for model validation.
EXPERIMENTAL BASIS
To study the impulsive sound propagation, four
large-scale experiments known as Blast propagation
over forest Norwegian Trials were conducted in the
period 1994 to 1997, during both summer and winter
and under various weather conditions. More than 800
spherical, 1, 8 and 64 kg charges of C4 explosive were
detonated at 2 m above ground. Overpressure was
recorded from ground level up to 32 m height at a
number of stations at distances from 200 m to 17 km
from the charges. Selected stations were instrumented
to study the interaction between air pressure and
ground response. Additionally, meteorological data and
ground conditions were recorded during the tests (see
[5] for more details). Typical recorded overpressure
and vertical particle velocity of ground surface are
shown in Fig. 1.
time (s)
FIGURE 1. Typical recorded sound overpressure and
vertical particle velocity of ground surface
The figure shows that the ground vibration has two
parts, one primary part that appears as a direct response
to the air pressure. The other part is a longer lasting
tail of vibration that consists of Rayleigh-type waves
generated by the air pressure as it propagates over the
ground on it way towards the observation point.
Analyses of the energy transfer from air to ground and
loss mechanisms in the ground have shown that the
energy in this tail is accumulated over about the past
500 m of wave propagation. The ground response is
dominated by the vibration induced directly by the air
pressure wave. Vibration generated at the charge
position and propagated in the ground is too weak
relative to this.
The dominant frequency of the primary ground
response varies from about 2 Hz for the largest charges
and longest distances to about 30 Hz for the shortest
distances and smallest charges, however there is a large
scatter in this relation among the various shots. The
dominant frequency of the air pressure follows the
same pattern but is slightly higher.
The ground in the test areas consists of sand and silt
deposits. Their shear wave velocities increase with
depth, making the Rayleigh waves dispersive in a way
that they are slower than the sound speed for the higher
frequencies and coincide or exceed the sound speed for
the lowest frequencies. This is a typical situation for
this type of sediments. At low frequencies the tail of
propagating Rayleigh wave ground vibration will thus
interfere with the primary air-pressure-induced ground
response and tend to increase it substantially. Thus, for
low frequencies, the ground will no longer be locally
reacting.

SESSIONS
 This phenomenon has been studied by deriving the Incident
Zaa=pa/va pa

apparent acousto-seismic impedance of the ground, wave Zag=pa/vg vp va v

g

defined as Zag = P()/V(), where P() and V() are

the Fourier transforms for the air pressure and ground
velocity respectively, both determined from a time- Air

Reflected
wave
windowed portion of the recorded signals covering
only the part where there is significant air pressure X
P1-wave
fluctuation. Figure 2 plots the magnitude of the Layered
porous S-wave
apparent impedances as defined above, calculated from medium Z
about 80 shots. From each shot the impedance value at P2-wave

the dominant frequency, where the determination is

best, is plotted. FIGURE 3. Schematics of analytical model

Figure 4 shows variation of the absolute value of Zag

1.00E+6

as a function of both frequency and shear wave

Zag [Pa/(m/s)]

velocity of the ground. The very small impedance

appearing at the velocity matching the sound speed is
1.00E+5

an important contribution of this model. Research is

underway to calibrate and incorporate this model in an
existing calculation tool for sound and vibration
1.00E+4

prediction.

2 3 4 5 6 7 8 910 20 30 40
Frequency (Hz)

FIGURE 2. Apparent acousto-seismic impedance

By including shots with varying charge weight and

distance, the frequency range from about 2 Hz to 20 Hz
has been covered. The figure demonstrates clearly the
effect of the interaction with the propagating Rayleigh
waves and the non-local behaviour on the apparent
impedance. Around 20 Hz the impedance takes a value FIGURE 4. Variation of absolute value of Zag
as expected from locally reacting based models [4].
For lower frequencies, however, it breaks the trend ACKNOWLEDGEMENTS
from these models and drops drastically towards the
lowest frequencies. This behaviour is essential for the The authors gratefully acknowledge the financial
ground effect on propagation of low frequency sound, supports from the Norwegian Defence Construction
and needs to be properly accounted for in numerical Service (NDCS) and the Norwegian Research Council
models. to carry out the present research work.

NUMERICAL MODELLING
Prediction of outdoor sound propagation is often
based on the assumption that the ground acts as a
locally reacting surface represented by its acoustic
impedance (e.g. [3]). Representation of the energy
absorption of the ground by locally-reacting models
may fail to realistically account for the interaction
between sound pressure and surface wave in ground.
Theoretical studies have shown that large ground
vibration is generated when the sound speed is close to
the surface wave velocity in the ground. To address
this issue, a 2-D numerical model has been developed
for the reflection of acoustic plane waves (Fig. 3) at the
surface of a layered poro-elastic (Biot type) halfspace.
The solution of the coupled wave propagation
equations in the two media is used to derive the
acoustic impedance and the acousto-seismic impedance
[6].
REFERENCES
1. Madshus C and Nilsen, N.I (2000) Low frequency
vibration and noise from military blast activity
prediction and evaluation of annoyance, Proc.
Internoise 2000
2. ISO 2631-2 Mechanical vibration and shock
Evaluation of human exposure to whole-body
vibration- Part 2: Vibration in buildings
3. Hole L.R., et al. (1998)Measurements and simulations
of low frequency impulse noise and ground vibration
from airblast. J. Sound Vib. 214, 309-324
4. Attenborough, K. (1985).Acoustical impedance models
for outdoor ground surfaces, J. Sound Vib. 99, 521-544
5. Guice,B. et al. (1998) Impulsive noise measurements in
a forest during summer and winter conditions. Noise
Cont. Eng. J. 46, 185-189
6. Kaynia, A.M. and L.R.Hole (2000). Impedance
consideration from wave reflection at acoustic/porous
interface, Proc. Internoise 2000

SESSIONS
Diffuse Field Transmission Loss of Double Walls Lined with
Heterogeneous Porous Materials
F. C. Sgarda, N. Atallab, X. Olnya
a
Laboratoire des Sciences de l'Habitat, DGCB URA CNRS 1652, Ecole Nationale des Travaux Publics de l'Etat,
69518 Vaulx-en-Velin CEDEX, FRANCE.
b
Groupe d'Acoustique de l'Universit de Sherbrooke, Department of Mechanical Engineering, Univ. de Sherbrooke,
Sherbrooke, QC,J1K2R1,Canada

In several industries (aeronautics, automobile, etc), high performance acoustic insulating lightweight materials are often
demanded for protecting people against extraneous airborne noise. This paper investigates a new solution that could match this
requirement. The work concentrates on the diffuse field transmission loss of double wall barriers involving heterogeneous
materials. The heterogeneous material is made up of porous materials in which macro solid pieces or air cavities are inserted.
This material is then sandwiched between an elastic plate and a septum that are embedded in an infinite baffle. The numerical
model is based on a finite element formulation for the different components. The coupling between the inclusions (solid or
acoustic) and the porous patches is accounted for naturally through the use of appropriate Biot's formulations. This paper presents
the theory behind the technique. Numerical results will be presented during the oral presentation

INTRODUCTION unit normal vector external to the bounding surface

W of the domain of interest.
Recently, several studies have revealed the benefit of
performing air holes in porous material to improve
their absorption efficiency [1,2]. Transmission loss
through heterogeneous materials is also a major issue
in all industries. Indeed, it is important to predict how
the isolation performance of systems involving porous
materials is affected by the presence of air cavities
such as holes carrying electric wires or plumbing pipes
for instance. In addition, one may wonder if the
transmission loss of double wall barriers lined with
porous material can be improved or at least not
deteriorated, keeping constant or decreasing the weight
of the system, if heterogeneities such as air cavities or
solid inclusions are added in the porous material. This
FIGURE 1. Configuration of the problem
paper investigates the diffuse field transmission loss of
an heterogeneous porous material sandwiched between
The weak integral form for an elastic material with
an elastic plate and a septum that are embedded in an
infinite baffle. The proposed model is based on a finite bounding surface We and volume We reads classically:
element Biot-Allard's formulation for the porous
patches and classical finite element formulation for the (1)
solid and acoustic inclusions. In this paper, the theory
of the proposed approach is presented. Numerical
results will be shown in the oral presentation. u
e
is the displacement vector, du
e
refers to its
e e
THEORY admissible variation, s and e are the stress and strain
tensors of the elastic material. For a septum the first
The geometry of the problem is depicted in Fig.1. It term involving the strain tensor in Eq(1) vanishes and
consists of an heterogeneous porous material e
s .n becomes the external force applied on the septum.
sandwiched between an elastic plate and a septum. An
incident diffuse field in medium 1 excites the system The weak integral form for an acoustic domain with
which is supposed to be baffled. Exterior fluid loading bounding surface Wa and volume Wa is also classical:
acting on the plate or the septum is neglected. The
heterogeneous material is built from rectangular three
dimensional patches either made from a homogeneous (2)
porous material modeled using Biot-Allard's theory,
from a solid elastic material or air. In the following, a pa is the acoustic pressure, dpa refers to its admissible
temporal dependency ejwt is assumed and n denotes the variation, ra and ca denote the medium density and the
sound speed, respectively.

SESSIONS
Two different integral forms for the poroelastic
material are used depending whether solid or acoustic
inclusions are inserted into the porous material. The
classical weak integral form associated to the porous
material has been given in [2]. It reads:
(3)
Wp and Wp refer to the porous-elastic domain and its
bounding surface. u and p are the solid phase
displacement vector and the interstitial pressure in the
poroelastic medium, respectively. du and dp refer to
~ s and es are
their admissible variation, respectively. s
the in-vacuo stress and strain tensors of the porous
material. s~ s is related to the total stress tensor of the
~ ~
~ s = s t + f( 1 + Q
material s t by the relation: s R )p 1 . f
stands for the porosity, ~r is a modified density given
by ~r = ~r11 - ~r12 ~r22 where ~r 22 , ~r11 and ~r12 are modified
~ 2
Biot's density accounting for viscous dissipation. Q is
an elastic coupling coefficient between the two
~
phases, R may be interpreted as the bulk modulus of
the air occupying a fraction f of the unit volume
( ~ ~
)
aggregate, ~g = f ~r12 ~r22 - Q R is a coupling factor.
Eq(3) is particularly suited for treating the coupling of
porous-air patches interface. Coupling conditions read:
(4)
where U is the fluid phase macroscopic displacement
f
[ ~ ~
vector. Since U = ~ 2 p - r~12 r~22 u and f 1 + Q R 1
r22w
] Biot's formulations in order to avoid the calculation of
for most of the porous materials, eq(4) simplifies to
(5)
Eq(5) indicates that in the coupled formulation
obtained by adding eq(2) and (3), surface terms vanish.
Only the continuity of pressures needs to be ensured.
A modified form of Eq(3) has been presented in [3]:
(6)
Eq(6) is used to deal with the coupling of porous-solid
patches interface. Indeed,
(7)
and surface terms vanish for the coupled formulation
(eq(1)+(6)). Only the continuity of displacements
needs to be ensured. In all cases, no calculation of
coupling matrices is needed which saves considerable
time during the assembly process.
The diffuse field excitation is considered as a
combination of freely propagating plane waves with
equal amplitude, no two of which are traveling in the
same direction and with propagation vector pointing
toward the plate [4]. The system equations are solved
for each incident plane wave with incidence angle
(q,j) and complex amplitude pinc, to yield the
corresponding acoustic power radiated by the plate or
the septum in medium i, Pi,(q,j) (i=1,2). The diffuse
field radiated in medium i is then obtained through an
integration over the half space. Pi,(q,j) can be rewritten
in terms of the radiation impedance operator of the
plate or the septum in medium i (see [5] for details).
The diffuse field incident power P inc is given by
d
S p inc p / 2r1c1 where S is the surface of the plate and the
diffuse field transmission loss by TLd = -10 log t d where
td = P d2 / Pinc
d
.
CONCLUSION
The diffuse field transmission loss of double wall
barriers lined with porous media with added trapped
solid heterogeneities has been predicted from a 3D
numerical model wherein each porous patch is
modeled using Biot-Allard poroelasticity equations.
The coupling between the porous material and the solid
or air inclusions is accounted for using two appropriate
coupling matrices. Numerical results will be shown
during the oral presentation.
REFERENCES
[1] Olny, X., Absorption acoustique des milieux poreux simple et
double porosit. Modlisation et validation exprimentale, PhD, Insa
de Lyon, 1999, 281p.
[2] Atalla, N., Sgard, F.C., Olny, X. and Panneton R., Acoustic
absorption of macro-perforated porous materials, accepted to Journal
of Sound and Vibration, 2000.
[3] Rigobert, S., Atalla, N. and Sgard, F., Numerical modeling of
porous-elastic materials using hierarchical elements , 5me Congrs
Franais d'acoustique, Lausanne, Suisse, Sept.3-6, 2000.
[4] Pierce, A.D., Acoustics, an introduction to its physical principles
and applications, Mcgraw-Hill, New York, 1981, 678p.
[5] Sgard, F.C., Atalla, N., Journal of the Acoustical Society of
America, 108(6), 2865-2872 (2000).
SESSIONS
 Numerical Homogenization Techniques Applied to Active
Fiber Composites
E. Lenglet a,b, A.C. Hladky-Hennion b and E. Deletombe a
a
Solid and Damage Mechanics Department, Office National dEtudes et de Recherche Aerospatiales, 5 blvd Paul
Painlev, 59045 Lille Cedex, France
b
Acoustics Laboratory, Institut dElectronique et de Microlectronique du Nord (UMR CNRS 8520), 41 blvd
Vauban, 59046 Lille Cedex, France

Abstract : The reduction of noise and vibration in rotor blades is possible by incorporating active fiber composites. The large
strains that they could bear compared to massive piezoelectric stacks makes them useful as skin actuators. But the fiber is
smaller than the wavelength, which makes it hard to model if one wants to numerically design CFAs patches. So one of the
solution can be to homogenize the composite piezoelectric material.
The existing analytical homogenization techniques can be classified into two categories : the homogenization techniques that
include or not the transformation factors. A numerical homogenization method has been developed, based on the ATILA finite
element code. The composite coefficients are obtained by modeling a representative 3D mesh of a geometrical period of the
material (one fiber plus matrix, between two interdigitated electrodes) under various solicitations. By applying appropriate
conditions, it is indeed possible to determine the mechanical, piezoelectric and dielectric constants of the homogenized
equivalent single volume element. The results obtained with the analytical methods are compared with the numerical results for
different piezoelectric fiber volume fractions. The comparison allows to emphasize the advantages and the drawbacks of each
method.
In conclusion, an application to a 1-3 piezocomposite material is presented.

INTRODUCTION HOMOGENIZATION

The composite materials have been used for many
years in aerospace because they give a good
compromise between weight and strength. More
recently, active materials such as active fiber
composites have been introduced in the rotor blades to
compensate the strain caused by aerodynamic loads
acting in the rotor system. Thus the vibrations and the
noise are reduced.
The composites are made of PZT fibers embedded in
epoxy matrix with PZT particles. Fibers are aligned in-
plane. The interdigitated electrodes have alternating
polarity (Figure 1).
Numerical Homogenization
The aim of homogenization is to calculate efficient
values for elastic, piezoelectric and dielectric constants
( 11 constants ).
For sake of simplicity, in this paper, only two
analytical homogenization techniques are considered :
Benveniste [1] and Mori-Tanaka [2] methods.
A numerical homogenization method is developed
using the finite element code ATILA [3]. Only a
quarter of the Representative Volume Elementary
(RVE) is considered (figure 2). By applying
appropriate initial conditions, it is possible to
determinate all the efficient constants for the composite
material. Table 1 presents the applied boundary
conditions to obtain s11, d31 et T11.

FIGURE 1. Active fiber composite A C

To avoid a lot of experiments, it is useful to model

their behaviors. But due to the size of the fibers which A C
is much smaller than the blades size, composites b
B a
homogenization techniques are used. B

b
FIGURE 2. Quarter of the Representative Volume
Elementary

SESSIONS
Table 1. Applied conditions In that case, the corresponding results can be
A A B B C C referenced as exact results. But meshing the whole
s11 Ux =cst Ux =0 Uy =cst Uy =0 Uz =cst Uz =0 structure is quite complex and need a lot of computing
Ux =cst Ux =0 time. Therefore, it is possible to use the different
d31 Uy =cst Uy =0 Uz =cst Uz =0
=1 =0
Uy =cst Uy =0
homogenization techniques and then classical relations
T11 Ux =cst Ux =0
=1 =0
Uz =cst Uz =0 [6] to calculate FFVS and TVR when a plate is
immersed in water. Figure 4 shows that the different
Two shear coefficients (s66 et s44) are obtained using homogenization techniques give results in good
propagation technique and the use of Christoffel tensor agreement.
[4-5]. 180

175

170

Validations 165

FFVS [ dB ]
160

155 Benveniste
150 Mori & Tanaka
First a PZT5A fiber embedded in epoxy is 145
Numerical
Exact
considered. The fiber section is circular. By varying 140

135

the diameter of the fiber, efficient coefficients of the 130

0 50000 100000 150000 200000 250000 300000 350000 400000

materials are numerically obtained and are compared frequency [ Hz ]

on figure 3 with the coefficients obtained using various -174

0 50000 100000 150000 200000 250000 300000 350000 400000

analytical methods. It shows a good agreement -176

Benveniste
Mori & Tanaka
between the curves, thus validating the model. -178
Numerical
Exact
TVR [ dB ] -180

c33eff/c33p
-182

1,0
-184

-186
0,8

-188

-190
0,6
mori frequency [Hz]
benveniste
0,4
numrique FIGURE 4. Variations of the FFVS and of the TVR versus
frequency, using various homogenization techniques
0,2

0,0
CONCLUSION
0 0,2 0,4 0,6 0,8 1

volume fraction
A numerical homogenization technique has been
e33eff/e33p

1,0
developed and has given results in good agreement
with previous analytical models. It has been applied
0,8 with success to a 1-3 piezocomposite. Then, losses
have to be taken into account in the model, for
0,6
mori
benveniste
comparison with actual case.
0,4 numrique

0,2
REFERENCES
0,0 1. Benveniste, Y., and Dvorak, G.J., J. Mech. Phys.
0 0,2 0,4 0,6 0,8 1

volume fraction Solids 40, 1295-1312 (1992).

FIGURE 3. Variation of the efficient coefficients c33 et e33
as a function of the volume fraction of the PZT5A fiber in the
REV.
Application to a 1-3 piezocomposite
The 1-3 piezocomposite considered is made of
PZT5A bars embedded in epoxy. It is immersed in
water. The materials are assumed to be without losses.
First, it is possible to mesh the whole unit cell, with the
fiber and the epoxy and then to calculate the Free Field
Voltage Sensitivity (FFVS) and the Transmitting
Voltage Response (TVR) [6].
2. Dunn, M.L., and, Taya, M., Int. J. Sol. Struct. 30,
161-175 (1993).
3. ATILA, Users Manual, Institut Superieur
dElectronique du Nord, Acoustics Laboratory
(1997).
4. Dieulesaint, E., and Royer, D., Ondes lastiques dans
les solides, edited by Masson and Cie, publisher,
Paris 1974, pp. 203-249.
5. Langlet, P., Hladky-Hennion, A.C., and Decarpigny,
J.N., J. Acoutst. Soc. Am. 98, 2792-2800 (1995).
6. Hladky-Hennion, A.C., and Decarpigny, J.N., J.
Acoust. Soc. Am. 94, 621-635 (1993).

SESSIONS
 High-Order Accurate Numerical Schemes for the Parabolic
Equation in Complex Domains
E. Flouria , J. A. Ekaterinarisa and J. S. Papadakisa
a Foundation for Research and Technology Hellas (FORTH), Institute of Applied and Computational Mathematics
(IACM), P.O. Box 1527,71110 Heraklion, Crete, GREECE
Fast and efficient, high-order accurate methods for the numerical solution of the narrow angle parabolic equation for underwater
sound propagation are developed. The space-like derivatives are evaluated using higher-order accuracy. Implicit numerical schemes,
which are second- or higher-order accurate in time-like marching and fourth-order accurate in the space-like direction are presented.
The efficiency of various numerical methods for time-like marching is evaluated for Cartesian-type meshes. Furthermore, finite-
difference schemes for complex domains are developed where the numerical solution is obtained using irregular meshes and curvilinear
coordinate transformations.

CARTESIAN COORDINATES derivative of the term Rk using the fourth or higher-order

The objective of this paper is development of explicit compact schemes. For explicit methods, the allowable
and implicit high-order finite difference methods for the step is restricted by numerical stability. Therefore, for
linear, narrow angle parabolic equation non-parallel algorithms implicit schemes presented in the


following section are more preferable.
ur r z  u  uzz  1



where r z  ik0  n z  r  1
2 and i 2k0 .
2

This equation is discretized using finite-difference ap-

Implicit Schemes
proximations of the derivatives for both the time-like, r,
Second-order accurate, implicit marching in the time-
and space-like, z, direction. Fourth-order accurate dis-
like, n , direction is obtained with the following Crank-
cretizations in the space-like direction may be obtained
Nicolson (CN-2) method:
by using compact finite differences f j   1  10 f j   f j  1
 2 f j  f j 1  . Compact differentiation schemes  r n  r

x2  j  1
12
f 1 n
U U 3
2  k 2 
R R
of various orders are analyzed in Ref. [1] and it is shown k
that achieve better resolution in wavespace than their ex-
plicit counterparts. These schemes offer an additional Third-order accuracy is obtained by the Adams-Moulton
advantage compared to their explicit counterparts, such (AM-3) scheme at the expense of an additional evaluation
as the explicit, fourth-order accurate f j   f j 2  of the residual term as follows:
12x2 
1

16 f j  1  30 f j  16 f j
1  f j
2  formula, because they ob-  5r n
1 2r n r n 

tain high order of formal accuracy by using a narrower
stencil. Compact methods accomplish these at the ex-
pense of the implied tridiagonal matrix inversion.
1 is evaluated
Explicit Schemes
Development of explicit schemes for the advancement
of the numerical solution in the time-like direction is
of interest because of recent advances in parallel archi-
tectures. Third-order, explicit time-like marching can
be obtained with the following Adams-Bashforth (AB3)
scheme:
r 
Ukn
1
U Un  R  1
4
12  k
R R
3 k 12 k
Both CN-2 and AM-3 implicit schemes, require tridiag-
onal matrix inversion of the left-hand-side implicit oper-
ator when the second derivative, uzz of Rkn
using second-order accurate, three-point stencil, centered
finite differences. Higher order of accuracy in z can be
only obtained at the expence of the increase in bandwidth
of the implicit operator by using explicit high-order cen-
tered formulas.
Using a compact differentiation formula for the sub-
stitution of the second-order derivative, uzz , in CN-2 and
AM-3 yields the following fourth-order accurate in the


Un  23Rnk  16Rkn  1
 5Rnk  2  2 space-like direction implicit schemes CN-2/4 and AM-
12 3/4, respectively.



where Rnk  U  Uzz  nk . In addition, third- or higher- CN-2/4, O  r  3  z  5  , implicit scheme:
order accurate Runge-Kutta (RK) methods may be used
for time-like advancement. For both, the AB3 and RKn al Ukn
11  blUkn
1  alUkn

1
1

methods, it is straight forward to evaluate the second arUkn 1  brUkn  arUkn
1 5

SESSIONS




where al 1  h 
12 z  2  , bl 10  h
10  where a D, b zz  r  D, c 2z D, d
24 z   , ar 1  h
 12 z  2  , br 10  h 10 
2 2z z D, e z D, with D r  zz.
2

24 z  2  , and

h
r 2  . In many applications, it is suffcient to use an equally
AM-3/4, O  r  4  z  5  , implicit scheme: spaced mesh with straight lines along the time-like, or ,
direction and stretched, curvilinear grid lines to represent
al Ukn 1 
1
blUkn
 alUkn 11 a1rUkn 1  b1rUkn
1
a bottom with irregular shape.
Then, the
simpler transfor-

mation of coordinates, r  , r z  , is sufficient
 a1rUkn 1  a2rUkn  11  b2rUkn  1  a2rUkn  11 6
and the transformed equation is:




1  h  12 u a1 u  b1u  c1u 9

z  , bl 10  h 10

where al 2

24 z  , with h
2 5r

12  , a 1  h 1 

with a1 r , b1 zz  r  r , c1 2z r .
1r
12

z 2
 , b 1r 10  h

1 10  24 z 2
 , with
h1
2r 3  , and a2r h2
 12 z2  , b2r h2 10  Explicit high-order schemes may be used for (8) and
24 z2  , with h2  r 12  . (9). In curvilinear coordinates, however, grid stretching
Finally, using compact finite-differences for the first is often used to concentrate grid points at the region of
derivative  ur  nk  1  4  ur  nk   ur  nk 1 r
3 n 1
 uk  unk  1  and
the bottom with an irregular shape. The allowable step
the corresponding formula for the substitution of second of explicit schemes may become too restrictive because
derivative, obtain the implicit scheme, CN-4/4, which is of grid cells with large aspect ratio and/or small spacing
fourth-order
accurate close to the bottom. As a result, for stretched meshes

in both directions. implicit schemes for time marching are preferred.
CN-4/4, O  r  5  z  5  , implicit scheme:

al Ukn 1 
1
blUkn 1
 alUkn 1
1 a1rUkn 1  b1rUkn

Implicit Schemes
 a1rUkn 1  a2rUkn  11  b2rUkn  1  a2rUkn  1
1 7 The following second-order accurate in both time- and
space-like directions CN implicit scheme is used for the
where al 3  h1 
12h2, bl 30  10h
1  24h2, with solution of (9).
h1
r, h2 r z  2 , and a1r 4 h1  12h2  , b1r
8 5h1  12h2  , a2r 3  h1  12h2, b2r 30  10h1  al Ukn 11  blUkn 1  cl Ukn 11




24h2.  brUkn  crUkn 1  O   3   3 
arUkn 1 10 



Similar to the nonlinear case of Ref. [2], in (5) and where

a l b 1  2c 1

 4, b l 1  0  5a 1
 c 1  , c l
(6) high-order of accuracy for the second derivative of  b1  2c1  4,
a r  b 1  2c 1 
 4, b r 1  0  5a 1 
both, the implicit operators and the right-hand-side resid- c1  , and cr b1  2c1  2.
ual terms is obtained at no extra computational cost, be- For the numerical solution of the fully transformed
cause compact space differention has been used in the equation (8) a new variable w u is introduced. A
derivation of these schemes. Furthermore, in (7) fourth box method is applied and the derivatives along are first
order of accuracy is obtained in both directions while the evaluated at the n 1  2 level
and then transferred back to
n 1 2
implicit operator retains tridiagonal structure. the nodes using fk fkn  fkn 1  2. The solution is
obtained by block tridiagional matrix inversion as:


 A vnk  11   B vkn 1   C vkn 11 VRHS 11 
CURVILINEAR COORDINATES
Numerical solutions of underwater sound propagation where vk  Uk  Wk  , the two by two blocks A  B  C are
n n n T

in complex domains are obtained
using curvilinear trans- given by:

formations of coordinates r z  ,
r z  . These 0 c 4

 A  1 2 0    B  0
b11 b12
transformations map the physical domain r z  , which is  12 
1 
unequally spaced and irregular, to an equally spaced or- b11 1  a 2  2e, b12  b 2  d,  A   C ,
and
thogonal grid referred to as the (  ) computational do- VRHS  V r
1  V r2  T , with V r1  4c Wkn 1  1 
n 1
2  4e  Uk  2  d  Wk  4 Wk 1  2eUk
main. The numerical solution is performed in the com- a n b n c n
and V r2
putational domain using unweighted finite-difference for-
 2 Uk  1  Wk  2 Uk 1 .
1 n n 1 n
mulas. The metrics of the transformation r  z etc. re-
late the physical and computational domain forms of the
equation. The derivatives in the transformed variables


and are ur u r  u r , uzz u zz  u z  2 
REFERENCES

2
2u z z  u z   u zz . Therefore in (  ) equa- 1. S. K. Lele, J. Comp. Phys. 103, 16, (1992).
tion (1) transforms as:

2. J. A. Ekaterinaris, J. Comp. Phys. 156, 272, (1999).
u au  bu  cu  du  eu 8

SESSIONS
 Dynamic Regimes of Three-Wave Interaction of Ultrasonic
Beams in TeO2
V.Grimalskya, L.Ilchenkob, G.Burlakc, S.Koshevayaa,c
a
National Institute for Astrophysics, Optics, and Electronics, Puebla 72000, Pue., Mexico.
b
Kiev National University, Radiophysical Faculty, Kiev 03127, Ukraine
c
Autonomous University of Morelos, Cuernavaca 62210, Mor., Mexico

The results of experimental and theoretical investigations of dynamic phenomena occurring under the three-wave parametric
interaction of acoustic beams in a crystal TeO2 are presented. The interacting waves are fast longitudinal ultrasonic wave (pump,
f ~100 1000 MHz) and two slow counter propagating transverse ones. Under ~ 5 times exceeding the threshold of parametric
generation, a complex spatial-temporal dynamics was observed, both regular and chaotic. The wave interaction was simulated by
coupled equations for slowly varying amplitudes of interacting waves with a transverse inhomogeneity. The spatial-temporal
distributions of wave amplitudes within the crystal were very complex in all directions and they qualitatively described the
experimental data. A reflection from end-walls reduced the threshold for dynamic regimes. The dynamic effects of spontaneous
generation of transversely non-uniform parametric bullet-like solitary pulses occurred there under simulations.

INTRODUCTION create a regime of running waves. The duration of

In a strongly anisotropic crystal TeO2, the three-
wave parametric interaction of fast longitudinal
ultrasonic wave (pump, f ~100 1000 MHz) with two
slow counter propagating shear ones is possible due to
elastic nonlinearity, and an influence of generation of
higher harmonics is unessential [1]. Interacting waves
propagate along [110] axis. The velocities of waves are
v1 = 4.21105 cm/s (L) and v2 = 0.616105 cm/s (S) [2]
Lower, the results of experimental and theoretical
investigations of dynamic phenomena occurring under
the parametric interaction of acoustic beams are
presented. The transverse cross-section of the pump
was ~ 10-2 cm2. Under ~ 5 times exceeding the
threshold of parametric generation, a complex spatial-
temporal dynamics took place, both regular and
chaotic. This regime is featured by complex angular
distribution of wave amplitudes.
Matching conditions for three-wave interaction are.
1 2 ,3
1 = 2 + 3; k1 = k2 k3; k1 = ; k2,3 = (1)
v1 v2
where (1, k1), (2, k2), (3, k3) are frequencies and
wave numbers for longitudinal wave, co-propagating
and counter propagating shear ones, respectively.
EXPERIMENTAL RESULTS
Experimental investigations were done in the
crystal TeO2 of 1 cm length. The pump wave was
excited by piezo transducer (LiNbO3) at one of the
end-walls, the another end-wall had an absorber, to
pump pulses was 2 ms. The measurements of angular
and amplitude characteristics of parametrically excited
acoustic waves were done by means of non-collinear
optical Bragg diffraction. Both a temporal dependence
of the power of parametrically excited waves under the
fixed position of probe optical beam and the angular
distribution under the fixed time moment (fixed
coordinates within the crystal) were obtained.
Frequencies of parametrically excited waves were
measured by means of optical heterodyning.
Parametric generation occurred some time after an
input of the pump pulse into the crystal. The input
threshold power level was 2 3 mW and it decreased
with the growth of frequency 200 500 MHz, because
of an influence of acoustic diffraction. Under the pump
levels several times exceeding the threshold, the
modulation instability took place, and a complex
angular and temporal distribution of parametrically
excited waves occurred. In a Fig.1, the oscillograms of
parametrically excited wave and corresponding angular
distributions of intensity are given. Under great values
of input power, the modulation became noise-like.
Crystals TeO2 possess very strong crystalline
anisotropy, and small angular deviations of pump wave
propagation from the [110] axis lead to essential
changes of group velocities of shear waves [2]. Thus,
the diffraction effects are essential for those waves.
Two-dimensional distribution of acoustic field,
both for pump and for signal ones, was also
investigated. That was done for a case of weakly non-
collinear interaction. The distributions were the
longitudinal strip-like structures for pump wave and
more complex ones for shear waves, like a
combination of longitudinal and transverse strips.

SESSIONS
 experimentally. Under essential input amplitudes, the
chaotic dynamics occurred.

Fig.1. Temporal dependencies of the power (oscillograms) in

the fixed point of crystal and corresponding angular
distributions of intensity for forward shear wave. From left to
right: small exceeding pump power level under the threshold;
periodic modulation; chaotic modulation under large
exceeding (the rate of scanning was 200 s/unit).

Fig.2. Spatial distributions of intensities of interacting waves

NUMERICAL SIMULATIONS |U1,2,3|2 for a fixed moment of time and temporal dependence
for |U3|2 in the fixed point obtained in simulations.
The three-wave interaction of acoustic beams with
the one-dimensional transverse inhomogeneity is Interesting phenomena manifested under
described by coupled equations for wave amplitudes: simulations when wave reflection from end-walls was
present (R 0). Moving bullet-like structures appeared
U 1 U 1 iv 2U 1
+ v1 + 1U 1 + 1 = 1U 2U 3 ; within the crystal (Fig.3). The thresholds for observing
t z 2 k 1 x 2 dynamic phenomena can be essentially reduced in a
presence of reflection.
U 2 U 2 iv 2U 2
+ v2 + 2U 2 + 2 = 2U 1U 3 * ; (2)
t z 2 k 2 x 2
U 3 U 3 iv 2U 3
v2 + 2U 3 + 2 = 3U 1U 2 *
t z 2 k 3 x 2
Here U1,2,3, are slowly varying amplitudes of
longitudinal and shear waves. Analogous equations are
for corresponding reflected waves U6,5,4. Z-axis is Fig.3. Bullet-like structures (distribution of reflected wave
directed along [110] one, x is normal to z; t is time. |U4(z,x)|2 for two time moments) in a presence of reflection.
Boundary conditions are: U1(z=0,x,t) = F1(t)1(x)
+ R11U6(z=0,t); U6(z=Lz,x,t) = R12U1(z=Lz,x,t); Pointed above dynamic phenomena under three-
U2(z=0,x,t) = F2(t)2(x) + R21U5(z=0,x,t); wave acoustic interaction have analogues in nonlinear
U5(z=Lz,x,t) = R22U2(z=Lz,x,t); U3(z= Lz,x,t) = optics, where coupled three-wave solitons and
R31U4(z= Lz,x,t); U4(z=0,x,t) = R32U1(z=0,x,t). The modulation instability of wave beams are observed
pump wave was input with a constant amplitude and [3,4]. The dynamic equations describing interaction of
Gaussian-like transverse distribution. The system was optical beams (so-called type II [3]) coincide with the
excited by a small short pulse of the signal wave U2. Eq.(2). Therefore, phenomena investigated here are
For approximation of (2), difference schemes were similar to ones known in nonlinear optics.
used; U1, U6 were calculated for time moments t=m*;
U2,3,4,5 were calculated for t = (m+0.5)*, where m is REFERENCES
integer, is the temporal step. The operator
1. L.Ilchenko and Yu.Oboznenko, Sov. Phys. Solid State
factorization was used.
21, 1979, 1648-1652 (1979).
Simulations for the case without reflection (U4,5,6 = 2. E.Dieulesaint and D.Royer, Elastic Waves in Solids,
0) demonstrated that under exceeding the certain Wiley, N.Y., 1980, 400 pp..
threshold, the dynamic effects occurred. The spatial- 3. Beam Shaping and Control with Nonlinear Optics/ Ed. by
temporal distributions of wave amplitudes within the F.Kajzar and R.Reinisch, Plenum Press, N.Y., 1998, 475 pp.
crystal are very complex in all directions and they were 4. S.Severini, C.Sibilia, and M.Bertolotti, J. Opt. Soc. Am.
multi-hump and strip-like (Fig.2), as it was observed B17, 580-585 (2000).

SESSIONS
 Numerical Models to Estimate the Noise Impact of a
City Kennel
S. Luzzi a, M. Toderib
a
Ordine degli Ingegneri della Provincia di Firenze, Via della Scala 91, 50123 Firenze, Italy
b
Dip. Meccanica e Tecnologie Industriali Universit di Firenze, Via S.Marta 3, 50139 Firenze, Italy

The paper refers to a case study carried out for the city doghouse in Pistoia and defines the methodological
approach to estimate the noise impact of a medium-sized kennel by adapting the algorithms provided for in
the ISO 9613-2 concerning the attenuation and damping of noise during propagation outdoors.

CONTEXT AND APPROACH Within these intervals, there are different levels of
acoustic emissions and the maximum is during
The insertion of a kennel in a city atmosphere is one lunchtime and during visits. Acoustic surveying has
of the sources of acoustic pollution deserving of been organized during five campaigns of noise
specific study. This paper defines the measurement with times of observation
methodological approach to estimate the noise corresponding to the various emission
impact deriving from the enlarging of the municipal characterizations. Measurements have been taken
kennel of Pistoia (Italy) and the consequent acoustic around the same type of boxes planned for the "new
pollution of the area. The system of sources kennel". The equivalent levels at a 1 m distance
occupies an area of approximately 4300 m2. and the from the perimeter of the building during the high
surrounding area is principally destined for rural use, emission interval, are represented in figure 1.
although a motorway access is planned in proximity
of it. The limits of noise according to the
classification of the territory are 55 dB(A) in Point Leq dB(A) S3 S6
nocturnal period and 65 dB(A) in diurnal period. S1 79.7
Moreover the noise bands of respect adjacent to S2 76.6
street infrastructures have to be considered. The S2 S5
S3 82.1
kennel must offer reception to the captured dogs and
S4 56.2
cats, up to a maximum of 96 animals. Beyond the
S5 55.3
sanitary and veterinary service buildings the S1 S4
S6 56.6
structure is composed of 8 modules. Each module
contains 12 boxes, with annexed exercise areas of
exercise of approximately 2.4 m2 per box. Dogs are FIGURE 1. Emission from Box Unit
the only primary sources of meaningful emissions.
The modular system of boxes with the dogs inside is
considered a significant source of noise for the Characteristics of the dogs emission in amplitude
propagation model. Starting from these and frequency have been carried out through noise
considerations about sources we have carried out measurements in the proximity of boxes
acoustic surveys. The standard day in the kennel is accommodating different dogs, getting in this way a
divided into four intervals, shown in table 1. complete spectral image of the phenomenon.
Analysis in 1/3 octave bands of the measured levels
Table 1 Standard kennel day shows that the dogs emissions are concentrated in
the bands from 315 Hz to 2500 Hz with maximum
time activities values in the 500, 800 and 1000 Hz bands.
06.00 08.00 sleeping time Measurements of the residual level have been
08.00 11.00 resting time into the boxes or open air time carried out in emplacements homogeneously
11.00 12.00 lunch time distributed along the perimeter, privileging the
directions of propagation from the more meaningful
12.00 20.00 resting time into the boxes or open air time
sources towards the closer receivers.
20.00 06.00 sleeping time

SESSIONS
 NOISE SOURCES AND MODEL
The mathematical definition of the problem is
obtained from the International Standard ISO 9613-
2 [3]. Noise sources can be grouped according to
criteria that depend on conditions and on the type of
source. For the studied kennel, a group of point
sources is identified as a single source placed in
"central" position. Model DOGSA (Doghouse
Outdoors General Sound Attenuation) considers the
new layout of the kennel like a single source,
composed of a series of point sources (single boxes).
The main receivers are identified in the residential
buildings called R01 and R02, linearly and
acoustically closest to the sources. In figure 2 the
model of propagation towards the main receivers is
represented.
Starting from these results, it is possible to predict
the absolute and differential levels of noise due to
the presence of the kennel. The Barking dogs of
different race and dimension dogs could determine
corrective factors, due to the presence in the
propagating emissions of single-band, impulsive or
low frequency components. DOGSA model is ready
to accept those corrections. The planning of an
acoustic barrier with characteristics and dimensions
capable of producing a wished attenuation, as
defined by ISO 9613-2, has become a part in model
DOGSA. In figure 3 the insertion loss of a 2 m high
barrier is represented.

FIGURE 3: Insertion loss of a barrier

CONCLUSIONS

For the work of the Pistoia kennel the proposed uses

of the territory and the noise classification were
FIGURE 2: Principal front of transmission given. A study of the sources represented by
different types and compositions of boxes has been
According to ISO 9613-2, the calculation algorithm, performed and the frequency characterization of the
considers as a first approximation level: the dogs emissions has been analysed. The structure of
attenuations due to geometrical divergence, air the buildings and the related characteristics of sound
absorption, ground effect and other miscellaneous propagation outdoors have been studied. This project
effects. Moreover the screening attenuation of a has generated and tested a strategy for determining
barrier, inserted between source and receiver, has the suitability of an urban area to accommodate a
also been tested. Using original software, simulated kennel. Model DOGSA is a mathematical
values of noise in proximity of the receivers have instrument adaptive to various contexts
been computed. Each value consists of the characterized by barking dog emissions in proximity
attenuated emissions emerging from the respective to sensitive receivers.
source system added to the measured residual levels
in the four intervals of activity of the kennel. Table REFERENCES
2 shows the result of these simulations.
1. L.L. Beranek, Noise and vibration control,
Table 2 Simulations Report McGraw Hill, New York, 1971, pp. 164-191.
2. C.M. Harris, Handbook of noise control,
McGraw Hill, New York, 1979, chapter 3.
(only new kennel)
simulated level

3. ISO 9613-2: 1996 Acoustics: Attenuation of

(only existing
receiver level
LAeq dB(A)

LAeq dB(A)

LAeq dB(A)

LAeq dB(A)
GLOBAL
esistente)
emission

sound during propagation outdoors- Part 2

ricettore

LEVEL
interval

level

4. S. Luzzi, R. Calonaci, R. Bellomini , Approcci

metodologici e modelli per la classificazione
acustica del territorio,Atti del XXVIII
R01 A 82.1 52.6 57.4 58.6
B 77.0 52.4 52.3 55.4 Convegno Nazionale Associazione Italiana di
C 68.8 52.6 44.1 53.2 Acustica, Trani, 2000
night 66.2 50.7 40.5 51.1 5. N. Campo, P. Rissone, M. Toderi, Adaptive
R02 A 82.1 53.7 60.2 61.1 pyramid tracing a new technique for rooms
B 77.0 53.7 54.3 57.0
C 68.8 52.7 46.1 53.6
acoustics, Applied Acoustics, Vol 61-2, pp. 199-
night 66.2 47.8 42.5 48.9 221, 2000

SESSIONS
 Modeling the magnetic noise of a DC electric motor

M. Furlana, M. Boltearb, M. Petrib

a
R&D, ISKRA Avtoelektrika d.d., Polje 15, 5290 empeter pri Gorici, Slovenia
b
University of Ljubljana, Faculty of Mechanical Engineering, Akereva 6, 1000 Ljubljana, Slovenia

The sound radiation coming from a permanent-magnet DC electric motor, where the magnetic noise tends to dominate, was
investigated using FEM/BEM. Due to the nature of the three-times-coupled problem an electromagnetic-structural-acoustic
numerical model was set up to predict the acoustic field. In the model set-up procedure, the first stage was to calculate the
magnetic forces that excite the structure of the motor using the FEM. In the second stage, the exciting magnetic forces were
applied to the structural model where, the harmonic analysis was also carried out using the FEM. The last phase was to model
the acoustics where the BEM, rather than the computationally more demanding FEM, was applied. All the numerical models
were prepared in 3-D. Finally, to evaluate the numerical model the computational results were compared with the vibration and
acoustic measurements and a reasonable agreement was found.

electric motor, are used as an input for the BEM

INTRODUCTION acoustic model.

The demand for a quiet human environment has

influenced almost every producer of noisy machinery,
including the manufacturers of DC electric motors.
Consequently acoustic noise has become one of the
most important factors that influences the acceptability
of a DC electric motor for customers in a variety of
industrial applications. Improved computational
FIGURE 1. Prediction of magnetic noise in electric
facilities have recently resulted in the possibility of a
machines
more quantitative description of the electric machinery
noise.
Magnetic Forces Calculation
The aim of this paper is to show a step-by-step
approach in order to predict the acoustic noise radiated For the magnetic force calculation in the
from DC electric motors. The analysis was performed investigated DC electric motor, two 3-D FEM models
on a permanent-magnet DC electric motor of 0.6kW, were built up, the only difference between the models
intended to drive an electro-hydraulic power steering was in the rotor skewing angle. One rotor had no
system. The results obtained with the FEM/BEM are skewing and the other had a skewing of one rotor slot.
compared to experimental data. Figure 2 shows both FEM models, with and without
the skewing.

METHOD OF NUMERICAL ANALYSIS

The prediction of magnetic noise is normally based
on a three-times-coupled electromagnetic-mechanical-
acoustic numerical model [1-3]. Figure 1 shows a
typical model of this kind. To calculate the magnetic
forces that excite the structure of the motor, the
electromagnetic model was developed. Applying the
magnetic forces to the structural model we can
calculate the magnitude and frequencies of the
resulting vibrations. The results of the vibrations, FIGURE 2. 3-D FEM electromagnetic models
represented by velocities on the exterior surface of the

SESSIONS
 Analyzing the resulting magnetic force acting on
each magnet we found their amplitude and phase
during the rotation. The variation in the magnetic force
can be decomposited by the discrete Fourier transform
method. Here only the first five harmonics of the
magnetic forces were calculated. As the rotor has
twenty slots these harmonics are the 20th, 40th, 60th,
80th and 100th. As the experimental investigation of
noise and vibrations was conducted for a constant
rotation speed of 3200 rpm the exciting magnetic
forces were calculated for the corresponding loading
conditions
FIGURE 4. 3-D BEM mesh on the left, acoustic field
Structural Dynamic Response instantaneous pressure on the right
To calculate the structural dynamic response of the
investigated electric motor a 3-D FEM model was
developed, see Figure 3. The structural response of
NOISE AND VIBRATIONS
both motors was calculated using harmonic analysis. From the vibration and sound pressure spectra
The analysis includes the first four harmonics of the measured for both motors, peaks at rotor slot
magnetic excitation forces. As both motors operated at harmonics were found. This phenomena shows that the
a rotation speed of 3200 rpm, the excitation frequency magnetic noise, which is related to the rotor-slot
of the 20th harmonic was 1067 Hz. harmonics, strongly dominates. The measured results
of the sound pressure and vibrations are shown in
Table 1.

FIGURE 3. Vibration response of the motor due to
magnetic forces of the 60th harmonic
Figure 3 shows the mechanically deformed motor
resulting due to the 60th harmonic of the magnetic
forces. The results of the measured and calculated
displacements for the collating point are presented in
Table 1.
Acoustic Field
The acoustic field is calculated with the BEM,
where the outer surface of the investigated motor is
discretized, see Figure 4. Table 1 contains the results
of the sound pressure level, both measured and
calculated, at a distance of 10cm from the motors
surface.
CONCLUSIONS
The presented investigation shows a step-by-step
procedure for predicting the acoustic field of a DC
electric motor when the magnetic noise dominates. The
results obtained with both the structural and acoustic
models give reasonable agreement with the
measurements. For a more accurate analysis, however,
improvements need to be made to the structural model.
REFERENCES
1 Wang C and Lai JCS Vibration analysis of an induction
motor, Journal-of-Sound-and-Vibration, vol.Januar 1999
2 Lai JCS and Wang C Prediction of noise radiated from
induction motor, Sixth International Congress on Sound
and Vibration, pp.2449-55, 1999.
3 Verdyck D and Belmans RJM An acoustic model for a
permanent magnet machine: modal shapes and magnetic
forces, IEEE-Transactions-on-Industry-Applications,
vol.30, no.6, pp.1625-31, 1994.

Table 1. Calculated and measured results of sound pressure and vibration displacement
20th harm., 1067Hz 40th harm., 2133Hz 60th harm., 3200Hz 80th harm., 4267Hz
p[mPa] x[m] p[mPa] x[m] p[mPa] x[m] p[mPa] x[m]
No Meas. 7.8 1.810-8 8.9 2.610-9 5.7 2.210-9 4.9 9.610-10
skewed Calc. 0.44 1.910-9 25.1 2.210-8 4.4 1.210-8 2.5 4.610-9
Meas. 3.5 1.3010-9 1.6 8.710-11 1.5 3.610-10 / /
Skewed
Calc. 1.1 1.710-8 1.7 1.110-8 1.2 1.310-8 0.1 7.910-9

SESSIONS
The Lattice Boltzmann Model and its Application to Acoustics
J. M. Buicka , M. A. Neala, J. A. Cosgrovea and D. M. Campbella
Department of Physics and Astronomy, The University of Edinburgh, J.C.M.B., The Kings Buildings,
Mayfield Road, Edinburgh EH9 3JZ, UK.
The lattice Boltzmann model is a novel numerical technique for simulating fluid motion. Recently the technique has been extended and
applied to the study of acoustical problems. In this paper we discuss the LBM approach to simulating fluid flow and acoustical prop-
agation, describing the general approach as well as specific details; these include the pressure being introduced into the formulation
through an equation of state, and the implementation of compressible and incompressible models and their relative advantages. The
application of the LBM is illustrated with examples demonstrating the flexibility of the technique. Sound propagation in an unbound
medium is considered and the results demonstrate the ability of the LBM in non-linear acoustics. Simulations of air flow through a
brass players lips and in the supply channel, or groove, of an organ are also presented.

INTRODUCTION which drives the distribution functions towards their local

The lattice Boltzmann model (LBM) has recently
developed as an alternative numerical method for
studying fluid flow [1]. It has also been applied to a
number of acoustical problems [2-4]. The ability of
the LBM in simulating both fluid flows and acoustical
wave propagation makes it ideally suited for studying
acoustical phenomena where there is a combination and
possible interaction between flow and acoustic fields.
THE LATTICE BOLTZMANN MODEL
equilibrium values f i at a rate determined by the relax-
ation time . The form of f i will be discussed below. The
fluid density and momentum are given by
fi and u fi ei (3)
i i
respectively. The equations of motion for such a system
can be derived using a Chapman-Enskog expansion [5]
and are seen to depend on the form of f i . The form of
the equilibrium distribution function, f i determines the
properties of the fluid being simulated. It is expressed as a
function of the local fluid density and local fluid velocity
up to second order:

fi ti  Ai  Bi e  u  Ci  e  u
2  Di u  u
(4)
Here we discuss the general features of the LBM.
The simulations are performed on a regular grid in D- where the values of ti
Ai
Bi and Di are selected de-
dimensions where each grid site is connected to b neigh- pending on the grid being used and the fluid properties
bouring sites by a link in the direction of the vector which are required. The basic fluid properties which
ei where i 0
1

b  1 if b is odd and i 1
2

b must be satisfied by the collision term are conservation
if b is even. The grid is denoted dDqb, for example of fluid mass and momentum, isotropy of the fluid
d2q7 is a two-dimensional hexagonal grid with 7 links and Galilean invariance. For the d2q7 grid this gives:
  

(0,0), ( 1,0) and ( 1/2, 3  2); and d3q18 is a three- Ai 1
Bi 4
Ci 8
Di  2  i and ti 1  2 for
dimensional cubic grid with six links to the nearest- i 0 and 1  12 otherwise. With this formulation for the
  
neighbours ( 1,0,0),(0, 1,0) and (0,0, 1) and twelve equilibrium distribution function the equations of motion

links to the next-nearest-neighbours along the links ( 1, of the fluid can be found using a multiscale Chapman-
    
1, 0), ( 1, 0, 1,) and (0, 1, 1). At each grid site Enskog expansion [5]. This results in the incompressible,
x there is an ensemble of b distribution functions, f i  x
t
, fully viscous, Navier-Stokes equation up to second order
which can be though of as describing the particle popula- accuracy. The pressure, p is determined by the equation
tion along each of the links. These distribution functions of state for an ideal gas: cs  p 
1  2 . This approach
are evolved according to the Boltzmann equation is limited to low Mach numbers and also to flows where
f i  x  ei
t  1
 f i  x
t
i  x
t

(1) the fluid compression is small with density variations
where the left-hand side of equation (1) represents
streaming of the distribution functions. The collision
term on the right-hand side the BGK [1] approximation:
of only a few percent [2,3]. An alternative approach is
to consider the additional conservation of energy and
the flux conditions of momentum and energy [4]. This
results in the Euler equations, accurate to first order, for
 1 a gas with speed of sound cs  p 
1 2 , where now ,
i  f i  f i

(2) the ratio of specific heats, is a free parameter which can

SESSIONS

4.0
 (a)
 t=T
2.0  t=5T
t=9T
Velocity ( ms )
1
0.0
2.0
4.0
0.0 
10.0 
20.0 
30.0
40.0

x (cm)
FIGURE 1. (a) Non-linear wave propagation. (b) Air flow in an organ groove. (c) Flow through a brass players lips.
be selected in the simulation. This approach has been
verified in situations [4] with a large density variation.
LATTICE BOLTZMANN SIMULATIONS
The two approaches described above [2,3] and [4] give
different models which can be applied to different situa-
tions depending on the form of the problem being inves-
tigated (Mach number, relative compression, accuracy of
solution, ideal/general fluid). Here we present a number
of simulation results obtained using the former approach
[2,3] where the greater accuracy is advantageous due to
the combination of acoustical and flow phenomena.
Figure 1(a) shows the velocity profile over one wave-
length of a 154dB, 1kHz sound wave propagating in air
at times T , 5T and 9T after it is produced, where T
is the wave period. Such a wave has a relatively small
density variation; however, its amplitude is large enough
for a shock front to develop. Figure 1(b) shows a two-
dimensional simulation of the flow which develops in the
groove or air supply of an organ pipe with one hole open.
The open hole was width 2mm to simulate the flow re-
sistance of an organ pipe. The length of the groove is
960cm and x = 0 corresponds to the open hole, the in-
flow of the air supply is 0.5 ms  1 . Figure 1(b) shows
qualitative agreement with PIV measurements [6]. Fig-
ure 1(c) shows the flow and vorticity computed for a jet
emerging from between a static lip model of a brass play-
ers lips. The minimum gap between the lips is 2mm and
the Reynolds number for the flow between the lips has a
maximum value of 1800.
CONCLUSIONS
The LBM has been described with specific emphasis
on its application to acoustics. Two approaches were de-
tailed and their relative merits discussed. Finally a num-
5 (cm/s) (b)
8
y (cm)
4
2

0
0 x (cm) 5
ber of simulations were presented relating to non-linear
sound wave propagation; flow in the channel of an organ
pipe; and flow between a brass players lips.
ACKNOWLEDGEMENTS
Financial support of the EU-CRAFT Project BRST-
CT98-5247 and EPSRC UK is gratefully acknowledged.
We are also grateful to the following organ manufactur-
ers: Werksttte fr Orgelbau Mhleisen GmbH, Leon-
berg, Germany; Manufacture dOrgues Muhleisen, Stras-
bourg, France; Orgelbau Wegscheider, Dresden, Ger-
many; Christian Scheffler Orgelwerkstatt Sieversdorf
(Frankfurt/Oder), Germany; Marcussen & Son, Orgel-
byggeri A/S, Aabenraa, Denmark; Orgelbau Schumacher,
Baelen, Belgium; Pels-dHondt Orgelbouw BVBA,
Herselt, Belgium; Fratelli Ruffatti Pipe organ builders,
Padova, Italy; Gerhard Grenzing, Papiol (Barcelona),
Spain; Oficina e Escola de Organasia, Ltd., Esmoriz
(Porto), Portugal; Didier Grassin, London, U.K. and
Pcsi Orgonapit Manufaktra KFT, Pcs, Hungary.
REFERENCES
 1 S. Chen and G. D. Doolen, Ann. Rev. Fluid Mech.
30, 329364 (1998).
 2 J. M. Buick, C. A. Greated and D. M. Campbell, Eu-
rophys. Lett. 43, 235240 (1998).
 3 J. M. Buick, C. L. Buckley, C. A. Greated and J.
Gilbert, J. Phys. A 33, 39173928 (2000).
 4 Y. Guangwu, C. Yaosong and H. Shouxin, Phys. Rev.
E 59, 454459 (1999).
 5 U. Frisch et al., Complex Syst. 1, 649707 (1987).
 6 J. M. Buick, J. A. Cosgrove, D. M. Campbell and C. A.
Greated, Optical methods applied to the measurement of
acoustic phenomena to appear in Optical Methods and
Data Processing in Heat and Fluid Flow, Professional
Engineering Publishing, 2001.
SESSIONS
 Application of a topology map for the automatic generation of
coupled vibro-acoustic wave models
B. van Hal, W. Desmet, P. Sas and D. Vandepitte
Department of Mechanical Engineering, K.U. Leuven, Celestijnenlaan 300B, 3001, Heverlee, Belgium

The wave based method (WBM) is an alternative approach to the finite element method for the steady-state response analysis of
coupled vibro-acoustic systems. Based on a brief review of the WBM, the need for the incorporation of topological information
is emphasized in order to allow the automation of the wave model generation and to prevent model errors. The topology map is
introduced to fulfill this task.

1 INTRODUCTION the system mathematically

The wave based method (WBM) [1] is a deterministic cavity : (
+ k2 ) p = j ! q; in V
prediction tool for the steady-state analysis of coupled,
p = p; on
p
vibro-acoustic (VA) systems. The method is based on the j @p = v ; on
v (1)
indirect Trefftz method [2]. Unlike the approximation ! @n n
j @p = p ; on
Z
functions of the finite element method (FEM) [3], the ! @n Z
approximation functions satisfy the governing dynamic j @p = j ! w; on

! @n s
equations exactly. Consequently, the WBM exhibits bet-
ter convergence properties than the FEM [1]. plate : ( d44 kb4 w
) = :::
dx 0
Until recently, all wave models were constructed in
(r ) + Dp ; on
s
F
D F
an ad-hoc way, such that for each VA system a separate  (2)
program was written. Now, the wave models are gener- w =0
dw at w;
ated automatically for uncoupled structural and acous- dx = 0 0

tic systems, based on problem specific information only.

To improve the accessibility of the WBM, the automated
wave model generation must be extended for coupled VA
systems, similar to the FEM. Unfortunately, the determi-
nation of the coupling contributions to the wave model
is not straight forward (see section 2). The application
of the topology map simplifies this task and thereby re-
duces the prossibility of model errors.
2 MATHEMATICAL BACKGROUND
2.1 Problem Definition
The WBM is applicable to many VA systems with a va-
riety of coupling principles. Here the basic ideas of
the WBM and the topology map are presented for a 2-
dimensional (2D) coupled, VA system (Fig. 1). It con-
sists of the acoustic cavity V , enclosed by the bound-
aries
p ,
v and
Z with respectively prescribed pres- w(x0 )
with thepLaplace operator
, the acoustic wave number
k, j = 1, the ambient density , the circular exci-
tation frequency ! , the derivative in the outward nor-
@
mal direction @n , the plate bending wave number k b ,
the plate modulus D and the Dirac delta function .
2.2 Response Approximation
The pressure approximation p^ consists of a sum of wave
functions a , satisfying the homogeneous Helmholtz
equation (1), extended by one particular solution p^ q for
the acoustic line-source q
p( r)  p^(r) = Pna=1 a (r)pa + p^q (r) (3)
with the unknown wave contribution factors p a . The fol-
lowing relation for the displacement approximation w ^ is
proposed
 w^(x ) = P4sP=1 s (x )ws + w^F (x )+
0 0 0

(4)
n
sure p, normal velocity vn and impedance Z and by the + ^ ( ) + w^q (x )
a=1 wa x pa
0 0

plate
s with clamped boundary conditions (BCs). The
r
system is loaded by the acoustic line-source q at q and
r
the structural line-force F at F . The pressure p at andr
r
the normal plate displacement w at s form the steady-
state dynamic response. The following set of coupled,
dynamic equations with corresponding BCs describes
where s are the four independent structural wave func-
tions, satisfying the homogeneous plate equation (2),
with the corresponding unknown wave contribution fac-
tors ws , where w ^a and w^q are the particular solutions
for the pressure loading and where w ^ F is the particular
solution for the line-force F .

SESSIONS
 L G w,q (L)
F W s
w BC: W v VA:V- W BC: G w,q (0)
G w,q (0) x' W v
s

rF p BC: W Z V
rs
r BC: W W s BC: G w,q (L)
V p
W p
q
y rq
FIGURE 2. Topology Map

x W Z The matrix contributions are related to the topology

FIGURE 1. 2-Dimensional Acoustic Problem map as follows. The boundaries (e.g. BC : w; (0))
and the interface (VA :V
s ) determine the evalua-
2.3
Wave Model tion points or the integration domains. The functions,
which are evaluated or integrated, result from the appli-
The approximations p^ and w
^ satisfy the dynamic equa- cation of the BCs to the wave functions and particular
tions but violate the BCs. The application of a weighted solutions of the coupled domains (see e.g. Eq. (6)). The
residual formulation of the acoustic BCs and the direct arrows on the topology map indicate the domains, which
evaluation of the structural BCs results in the discrete are coupled either directly or indirectly. E.g. boundary
wave model [1] BC : w; (0) is directly coupled to domain
s and indi-

Ass Csa  w =  fss + gsa  (5) rectly coupled to domain V via the interface VA :V
s .
Cas Aaa + Caa p gsa + faa + gaa If the VA system contains many components, the
visualization of the topology loses transparency. The
with the uncoupled structural matrix A ss and vector f ss , topology map is therefore captured by three topology
the structural-acoustic coupling matrix C sa and vector matrices (one for each building block), which are di-
gsa, the uncoupled acoustic matrix A aa and vector faa, rectly applicable for the automated generation of wave
the acoustic-structural coupling matrix C as and vector models.
gas and the acoustic back-coupling matrix C aa and vec-
tor gaa . The vectors w and p contain the unknown w s 4 CONCLUSIONS
and pa . The formulae for the matrix contributions are The construction of wave models for VA systems con-
too extended to present here. E.g. the acoustic back- taining many components requires the incorporation of
coupling matrix in a least-squares formulation is the systems topology, visualized by the topology map.
Z  j @ c  !

This allows an automated model generation and reduces
j T
Caa = ! @n +j!w^ ac ! @n j!w^ aT d

@
(6) the risk of model errors. Practically, the topology map

s is stored in matrices, which are directly applicable.

with the complex conjugate  c and the transpose  T . ACKNOWLEDGEMENTS

The vectors  w
and ^ a contain the wave functions  a
The research of B. van Hal is financed by a scholarship
^a.
and the particular solutions w
of the Institute for the Promotion of Innovation by Sci-
ence and Technology in Flanders (IWT). W. Desmet is a
3 TOPOLOGY MAP Postdoctoral Fellow of the Fund for Scientific Research
The simple 2D problem shows already the complexity - Flanders (FWO).
of the wave model generation. Especially, the coupling
contributions to the model are not straight forward as REFERENCES
C
is demonstrated by aa . The automation of the model 1. Desmet, W., A wave based prediction technique for
generation requires therefore the incorporation of the coupled vibro-acoustic analysis, Katholieke Universiteit
topology of the model building blocks (domains, bound- Leuven, division PMA, Ph.D thesis, 1998.
aries and interfaces). The topology, graphically repre-
2. Trefftz, E., Ein Gegenstuck zum Ritschen Verfahren, in
sented by the topology map (Fig. 2), describes the re- Proceedings of the 2nd International Congress on Ap-
lations between these building blocks. The geometrical plied Mechanics, Zurich, 1926, 131-137.
properties of domains and boundaries form the basis for
3. Zienkiewicz, O.C. and Taylor, R.L., The Finite Element
the topology map construction, where the description of
Method - Vol. 1: Basic formulation and linear problems,
interfaces appears as by-product (e.g. the VA-coupling McGraw-Hill, London, 1977.
VA :V
s ).

SESSIONS
 The Numerical Prediction of Airfoil Trailing Edge Noise

E. Manohaa, C. Delahaya, S. Redonneta, P. Sagauta, I. Marya

S. Ben Khelilb and P. Guillenb
a
Department of CFD and Aeroacoustics
b
Department of Applied Aerodynamics
ONERA, BP 72, 92322 Chtillon, France
eric.manoha@onera.fr

The numerical prediction of the aerodynamic noise radiated by an isolated airfoil is performed using a CAA hybrid method
combining a compressible three-dimensional Large Eddy Simulation (LES) and an acoustic method providing the far field
noise[1]. The LES exhibits a superimposed pressure fluctuation field which presents the qualitative and quantitative features of
the TE noise generated by the acoustic scattering of the turbulent boundary layers (TBL) convected on both airfoil sides (broad-
band noise) and by the mechanism of alternated vortex shedding generated by the TE bluntness (narrow band component). Due
to the strong radial stretching of the LES computational grid, which acts as a low-pass filter, this acoustic field cannot radiate in
the far field. Consequently, the LES must be relayed by an acoustic propagation method. The most readily available methods
are integral methods such as the Kirchhoff integration and the Ffowcs Williams-Hawkings (FW-H) equation. Results are in
good qualitative and quantitative agreement with experimental data.

LARGE EDDY SIMULATION velocity in [2]), the Mach number is 0.205 and the
Reynolds number based on Uo and C is 2.86millions.
The NACA 0012 airfoil has a C= 0.6096m chord and a The airfoil incidence is 5. Fig. 2 shows instantaneous
blunted TE of thickness H=2.5mm (0.4% of C), with Mach number and flow streamlines at the trailing edge,
reference to an airfoil noise experiment conducted at with the onset of vortex shedding at the TE.
NASA in 1980 [2]. In this experiment, the airfoil had a
span SEXP=0.46m (or 75% of C), whereas the present Z

LES computational domain has a spanwise extent SLES

Y X

0.22
0.215
0.21

representing only 3.3% of C, as obtained by replication

0.225

0.23
0.205
0.21

of a 2D curvilinear structured grid (Fig. 1).

0.24

0.255
0.205 0.2

0.2 0.18
0.205
0.185

0.19 0.2

0.6 0.195

6 0.5

0.4
4
0.3

0.2
2
0.1

0 0 FIGURE 2 : Aerodynamic results. Left : instantaneous

z

-0.1
-2
-0.2 Mach number. Right : Flow streamlines at the TE,
-0.3
-4
-0.4 instantaneous (above) and time-averaged (below).
-6 -0.5

-0.6
-6 -4 -2 0 2 4 6 -0.25 0 0.25 0.5 0.75
x x

Figure 3 shows wall pressure spectra on both airfoil

0.003 sides near the TE, showing a narrowband component
0.002
-0.001
(due to the vortex shedding) emerging out of a
wideband continuum (generated by TBLs). Spectra are
0.001 -0.0011

0 -0.0012

compared (without corrections) to experimental data

z

-0.001 -0.0013

-0.002
from [2]. Levels are slightly overestimated (3 dB), as is
-0.0014

-0.003
-0.0015
the vortex shedding frequency, which is explained by
0.608 0.61
x
0.612 0.6094
x
0.6096 0.6098 the excessive slenderness of the simulated TBLs,
probably due to unsufficient grid refinement in the
FIGURE 1 : Computational grid.
transition region. Fig. 3 also displays the rapidly
decreasing spanwise coherences of the wall pressure
The whole grid is made of 1.76 million points. The
fluctuations, showing that the turbulent structures
upstream flow velocity is U o= 69.45m/s (maximal

SESSIONS
simulated near the wall seem to be smaller than the the spanwise coherence shown on Fig. 4 (right),
3.3%-chord span of the LES domain. probably induced by the limited span of the LES
domain.

FIGURE 3 : Wall pressure fluctuations. Left : spectra. FIGURE 5 : Instantaneous pressure fluctuations.
Right : Spanwise coherences. Left : LES data. Right : Kirchhoff data.

Since wall pressure spanwise coherences (Fig. 3) are

Figure 5 (left) shows instantaneous pressure fluctuations
assumed reliable, a second noise prediction was
in the LES domain, with an acoustic concentric wave
achieved using the Ffowcs Williams-Hawkings method
pattern generated at the TE, the wavelength of which
(dipole surface term only) from the wall pressure data
corresponds to the vortex shedding mechanism. Due to
inside the LES domain (span SLES.). Then the noise from
the strong stretching of the LES computational grid,
a real airfoil with a span SEXP was estimated by applying
which acts as an acoustic low-pass frequency filter, this
a factor S EXP/SLES on the resulting PSD, meaning a
acoustic field cannot radiate farther than a half-chord
decorrelated summation of spanwise periodic domains.
from the body. Consequently, the LES must be relayed
by an acoustic propagation method to correctly simulate Figure 6 compares, without correction or normalisation,
this estimation with the noise measured in [2]. It is
the farfield noise.
interesting to notice that experiment and simulation
differ in the same way as wall pressure spectra (Fig. 3),
NOISE PREDICTIONS which may confirm the validity of the acoustic method.

The Kirchhoff method has been firstly used for noise

prediction from a control surface enclosing the airfoil
(Fig. 4, left), on which it was verified that the pressure
field was purely acoustic. Spanwise coherences on this
control surface (Fig. 4, right) suggested that the acoustic
field was quasi-2D there, so the pressure data on the
surface were constructed by replication, along the SEXP
span of a realistic airfoil (75% of C in [2]), of the LES
data simulated in the median plane of the domain.

j0 +1
j0
z0

z Trailing edge
FIGURE 6 : Noise spectra at two chords above the TE.
L

j0 +1
y Direct comparison of experimental data and numerical
Airfoil
j0
x
simulation using the FW-H method.
2-layer
z0 L=L
integration Dom
ain
surface #1 n
Dom
ain
#2

FIGURE 4 : Kirchoff control surface. Left : surface
construction. Right : Spanwise coherences.
Figure 5 (right) shows a map of the acoustic pressure
field computed using the Kirchhoff method. However,
the noise computed at two chords above the TE was
found to exceed experimental data by more than 15 dB,
which was explained by an artificial overestimation of
REFERENCES
1. Manoha E., Delahay C., Ben Khelil S., Guillen P., Sagaut
P. and Mary Y. Numerical prediction of the unsteady flow
and radiated noise from a 3D lifting airfoil, AIAA Paper
2001-2133, 7th CEAS/AIAA Aeroacoustics Conference,
28-30 May, 2001.
2. Brooks T.F. and Hodgson T.H. Prediction and comparison
of trailing edge noise using measured surface pressures,
J. of Sound and Vibration, Vol. 78 (1), pp. 69-117, 1981.

SESSIONS
 Diffraction of an acoustic wave by a plate in a uniform
flow: a numerical approach
A.S. Bonnet-Ben Dhia, E. Luneville and J.-F. Mercier
Laboratoire de Simulation et de Modelisation des phenomenes de Propagation,
ENSTA, URA 853 du CNRS, Paris, France

Abstract Diffraction problem

We study the diffraction of an acoustic wave by a rigid plate

We are interested in the diffracted field when the plate
in the presence of a uniform flow in a duct. Contrary to
is submitted to a time harmonique incident wave. If
prior analytical studies, using Wiener-Hopf techniques and
we eliminate the velocity in Euler equations, the re-
thus restricted to semi-infinite plates, we use a finite ele-
sulting pressure is to regular comparing to experimen-
ments method, allowing the study of finite plates. To take
tal observations. In the case of a semi-infinite plate,
into account irrotational perturbations induced by the lead-
many authors considered that a vortex sheet S is in-
ing edge of the plate, a vortex sheet is introduced behind
duced by the incident acoustic field behind the plate
the plate. In order to get well suited radiation conditions
[1,2]. A pressure formulation of the diffraction prob-
at infinity, the problem is split in two standard diffraction
lem is unable to take into account such vortex sheet,
problems, thanks to the introduction of a semi-analytic rep-
contrary to a velocity potential formulation. Therefore
resentation of the vortex sheet. The amplitude of the vor-
we consider that the acoustic velocity is irrotational
tex sheet is determined by imposing a finite velocity near
everywhere excepting on the wake. We introduce the
the leading edge (Kutta condition). The key point of the
velocity potential such that v = in /S. Using
method is to get the singular coefficient of the acoustic field
Euler equations in harmonic time regime eit where
near the leading edge. This approach leads to an efficient
is the frequency of the incident wave, the velocity po-
finite elements method, and numerical computations are
tential is found to satisfy the convective wave equation
presented. This method has been extended to the case of
in /S:
two plates, parallel or aligned, is also considered. 2 2
HM () = (1 M 2 ) 2
+ 2
+ 2ikM + k 2 = 0.
x y x
MODEL
M = V /c is the Mach number, c the sound velocity in
Geometry and equations the fluid at rest, and k = /c is the acoustic frequency.
The incident wave inc must solve the equations with-
out the plate. We choose to take a propagative duct
y
y=h D mode [3]. Then the total field is sought in the form
= inc + d .
B- + B+
V S Thanks to a change of variables and unknowns, the
inc
x=-L
-
x=L
equations of acoustics in a uniform flow are trans-
- R +
formed into the equations in a medium at rest. More-
y=0 x=-R x=d x=R x
over across the vortex sheet S, we use the continuity
of the pressure p and of the normal component of the
displacement u (defined by DuDt = v ). Then the diffrac-
Figure 1: Geometry of the problem.
tion problem reads:
A uniform subsonic flow of velocity V is imposed in the
( + 2 )d = 0 in /S,
duct D. A finite plate is placed horizontally in the
d
y = 0 on D,
flow in y = , considered thin enough not to disturb d
= yinc on ,
the flow. The velocity v of the acoustic perturbation
y h i


and the acoustic pressure field p are supposed to satisfy [d ]S = Fe iM x
and y
d
= 0.
S
the linearized isentropic Euler equations in = D\. (1)

SESSIONS
F is an unknown constant, determined by applying the
Kutta-Joukowski condition at the trailing edge T (x =
L, y = ), that is the velocity v has to be bounded.
The classical Helmholtz equation is recovered but with
a different acoustic frequency = k/ 1 M 2 and
with a longer plate of length 2L/ 1 M 2 .
We should impose some radiation condition to en-
sure the well-posedness of problem 1. Because of the
presence of a vortex sheet, it is hard to find some well-
suited radiation condition. To overcome this difficulty
we will eliminate the vortex sheet.
Vortex sheet elimination
We introduce the vortex sheet function S satisfy-
ing Eq. 1 with F = 1 and y = 0 on . Then
S
the diffracted field may be sought in the form d =
a + F S , where a , called the acoustic part of the
diffracted field, satisfies a problem without vortex
sheet. Although S can be obtained from Wiener-
Hopf approach (jone) in the case of a semi-infinite
plate, such construction fails for the case of a finite
plate. Thus we need to introduce the pseudo solu-
tion S = H(x d)(y)ei M x , where H is the Heaviside
function with 0 < d < L. is calculated such that S
is solution of Eq. 1 with F = 1 excepting the bound-
ary condition on the plate. Thus we need to introduce
the correction function c defined as S = S + c .
Therefore we are led to determine a and of c , sat-
isfying two diffraction problems without vortex sheet
and with transmission conditions on in x = d (see
Fig. 1).
SPECTRAL REPRESENTATIONS
Exterior domains
In this part, will denote as well as c . Although
the resolution of Eq. 1 involves a numerical approach,
analytical solutions in the form of spectral represen-
tations can be found in two parts: far from the plate,
thanks to a cartesian decomposition can be expressed
as a sum of the functions exp(in x) cos ny
h . Similarly
near the end edges of the plate, we use a cylindrical
decomposition in the balls B based on the functions
J n2 (r) cos n
2 . Consequently we are allowed to set an
equivalent problem in a bounded domain surrounding
the plate (see Fig. 1) R = {(x, y) ; |x| < R}.
Singularity coefficients
The Kutta-Joukowski condition imposes the velocity to
be bounded close to the leading edge of the
plate. We
have found the asymptotic behavior + rcos 2 as
r 0, where the singularity coefficient at the leading
edge is given by:
Z
+ = C() (, ) cos d.
B 2
Thus F is solution of + F c = 0 (if c 6= 0) where
(respectively c ) is the singularity coefficient associated
to (c ). Thanks to such spectral representations, the
diffraction problem can be set in a bounded domain:
we have proved the well-posedness of this problem [4]
and thus is adapted to a numerical resolution.
NUMERICAL SCHEME
The acoustic field a and the correcting field c are
sought in the bounded domain R \B , where finite
elements can be introduced. We have chosen Lagrange
finite elements of type P1 or P2 and the spectral rep-
resentations are limited to the first ten terms. On
Figure 2: Total field = inc + a + F S .
Fig. 2 is represented the total field for M = 0.87 and
k = 1. With L = h = 4, the incident wave length
is +
1 = 29.3 and the downstream diffracted wave
length is 1 = 0.9 whereas the wakes wave length
is S = 5.4. The Kutta-Joukowski condition leads to
the value F = 0.008 + i 1.6.
REFERENCES
1. D.S. Jones. Aerodynamic sound due to a source
near a half plane. Journal of the institute of Math-
ematics and its Applications, 9:114122, 1972.
2. Mahmood-Ul-Hassan and A. D. Rawlins. Two
problems of waveguides carrying mean fluid flow.
Journal of Sound and Vibration, 216(4):713738,
1998.
3. V. Pagneux. Propagation acoustique dans les
guides a section variable et effets decoulement.
PhD thesis, Universite du Maine, 1996.
4. A.S. Bonnet-BenDhia, L. Dahi, E. Luneville and
V. Pagneux. Acoustic Diffraction by a Plate in a
Uniform Flow. Submitted to M3AS.
SESSIONS
 A complex modes analysis for the resolution of finite-element
poroelastic problems
O. Dazela, F. Sgarda , C.-H. Lamarqueb and N. Atallac
a
LASH-DGCB URA CNRS 1652, F-69518 Vaulx-en-Velin, France
b LGM-DGCB URA CNRS 1652, F-69518 Vaulx-en-Velin, France
c GAUS, Mech. Eng., Univ. de Sherbrooke, QC, Canada

The research of efficient methods to improve the resolution of poroelastic finite-element problems is of major importance for the
numerical-acoustic community. The modal resolution of fu; Pg finite-element poroelastic problems [1] is restricted by the frequency
non-linearity of the associated eigenvalue problem. In this paper, a new modal technique inspired from the complex modes has been
elaborated to overcome this difficulty.

INTRODUCTION THEORY
The following equation is considered
Multi-layer poroelastic structures are used to improve
noise control in many engineering fields such as aeronau- D x = 0; (3)
tics, automobile, etc. Atalla et al. gave an exact fu; Pg
formulation of Biots equations of poroelasticity for an where x 2 C d (R; C n ), d 2 N and D is a differential oper-
harmonic motion at pulsation [1]: ator generally non linear from F (R; C n ) into itself. It is
assumed that D can be written in the following form

2 3 d
e ] + ( j)2 [M
[K e] e]
[C     D x = [Mi ]x(i) + Rd x; (4)
4 5 u Fs
e]
[H = : i=0
e ]t
[C e]
[Q P Ff
2 where [Mi ] is a real symmetric (n; n) matrix , x (i) is the
(1)
Each matrix (except [ Ke ]) represented with the symbol ith derivative of x and R d is an operator, called remainder
e is equal to the product of a complex function of the at order d, from C d (R; C n ) into F (R; C n ). The first part
real parameter (called dependant function) and a real of the right hand side is said to be the expansion at order
symmetric matrix. This formulation is very relevant, nev- d. It is further assumed that det [M d ] 6= 0.
ertheless systems may be still large sized. Therefore there The idea of the exposed approach is to find the modes of
is a need to find out numerical techniques in order to im- the expansion at order d. X 2 F (R; N nd ) is then intro-
prove the solving process. Classical modal analysis can duced and assembled as:
not be applied to poroelastic problems because of the fre-  t
X= x (d 1) ::: x(1) x : (5)
quency non-linearity of the associated eigenvalue prob-
lem. This property is induced by the frequency depen- One can then build the generalized eigenvalue problem of
dance of dependant functions and the selected fu; Pg for- size nd:
mulation. Nevertheless, one can express each dependant [A]X = [B]X
(1)
(6a)
e ]) :
matrix with a Taylor expansion (e.g. for [ M 2 3
[Md 1 ] [Md 2 ]


[M0 ]
6
6 [Md 2 ]


[M0 ] [0] 7
7
[A ] = 6 .. .. .. .. 7 (6b)
d 4 . . . . 5
e]=
[M mi ( j)i [Mint ] + O(d +1 ); (2) [M0 ] [0]


[ 0]
i=0
2 3
[Md ] [0]


[0]
6
6 [0] [M d 2 ]


[M0 ] 7
7
where mi are the coefficients of the Taylor expansion of [B] = 6 .. .. .. .. 7 : (6c)
4 . . . . 5
the dependant function. The purpose of this paper is to
propose a new modal analysis approach based on an ex-
[0] [M0 ]


[0 ]

tension of the complex modes technique to calculate the Eigenvalues and eigenvectors of equation (6a) can then
vibro-acoustic behavior of a porous material. be computed. [] is a matrix where the i th column is the

SESSIONS
corresponding eigenvector of equation (6a). Let then []
be the (n; nd ) sub-matrix of [] dealing with the n last
rows of []. Let (si ,Xi ) and (s j ,X j ) be two solutions of
equation (6a), then one has:
8(i; j) si 6= s j ) X j t [B]Xi = X j t [A]Xi = 0: (7)
The calculated modes are now used to build an approxi-
mation of the solution of a forced harmonic problem. The
idea is to use the property that in the low-frequency range,
the solution has a modal behavior and it can then be rele-
vant to project the initial problem on a modal sub-family
in order to reduce the size of the problem to solve. Forced
harmonic problem can be expressed as
e ()xe jt = fe jt ;
D (8)
where D e verify eq. (4). The idea is to express this prob-
lem in the generalized State space in order to find the
modal contributions z. This is interesting when R d is not
significant and then a decoupling can be obtained. This
wont be the case of the application to poroelastic mate-
NUMERICAL IMPLEMENTATION
The case of a single porous material of dimensions
0.15m*0.10m*0.05m bonded onto a rigid wall is studied
here. The properties of the material are given in Table
1. This material is submitted to a rigid piston motion of
amplitude x0 and its lateral edges are free. The frequency
range of interest is [0; 500Hz]. The Reference result is
given by a commercial software Nova Mecanum. The
mesh is chosen to insure the convergence of the solution
(4218 dofs). To propose a selection process , one can
suggest to increase the ratio = maxc
of the maximum
imaginary part of the selected eigenmodes to the maxi-
mum pulsation of the spectrum of excitation. The mean
square velocity along y axis which is the indicator that
has the most difficulty to converge is considered. Two
choices of are considered. The first one ( = 2) leads
to a 84 dofs problem and the second one ( = 3) leads
to a 204 dofs whose results are in an excellent agreement
with the reference solution.
55

rials; nevertheless, since X = []z , x = []z, one can Mean square velocity (dB) /y
50

prove that the solving of equation (8) is equivalent to the 45

one of
e ()[]z = f:
D (9)
40

35

The multiplication of eq. (9) by [] t yields: 30

Reference (4218 dofs)
=3 (204 dofs)
=2 (84 dofs)
te 25
[ ] D ()[]z = [] f:
t 100 200 300 400 500
(10) Frequency

A sub-family [] of rank m of [] is then elected in order FIGURE 1. Studied case

to reduce significantly the size of the problem to solve.
The approximate x of x is

x() = []() (11) CONCLUSION

with () solution of the size m problem In this paper a modal analysis method based on an ex-
tension of the complex modes has been designed. The
te response for a single porous material has been computed
[ ] D ()[]z = [] f:
t
(12)
and compared with the results obtained by a classical res-
olution of the finite element system. The approach allows
Sub-family [] comprises modes whose imaginary part
for a significant reduction of the size of the systems to
of the eigenvalue is lower than a critical pulsation c de-
solve. The robustness of the method needs to be investi-
pending on the excitation.
gated in the case of a wide range of porous materials and
for multi-layered configurations.
Table 1. Sample characteristics

h (kPa:sm 4) (m) 0 (m) REFERENCES

0.95 25 1.4 93.2 93.2
s (kg=m3 ) N (kPa) x0 (m) 1. Atalla, N., Panneton, R., and Debergue, P., Journal of
600 42 0 0.05 10 8 Acoustical Society of America, 104,1444-1452 (1998) A
mixed displacement-pressure formulation for poroelastic
materials.

SESSIONS
 A simplified finite element model for the vibroacoustic
behavior of a porous coated plate
S. Rigoberta, N.Atallab, F.Sgarda
a
Laboratoire des Sciences de lHabitat, DGCB/URA CNRS 1652, rue Maurice Audin
69518 Vaulx-en-Velin Cedex, France.
b
Groupe Acoustique de lUniversit de Sherbrooke, Universit de Sherbrooke
Dpartement de Gnie Mcanique, 2500 Boulevard de lUniversit, Sherbrooke, Qubec, J1K2R1, Canada.

In this paper, a simplified finite element model dedicated to the configuration of a porous coated plate is presented. The
originality of this model is that all damping phenomena in the porous coating are accounted for and especially those related to the
motion of the fluid phase of the porous material. The present approach is based on the use of a multilayer plate element for the
modeling of the plate together with the in vacuo solid phase of the porous material. Viscous and thermal effects in the poroelastic
coating are taken into account by introducing a frequency dependent damping coefficient. This simplified model is used to model
a porous coated plate in free field and proves to give satisfactory results.

INTRODUCTION fails to give satisfactory results when these latter

The configuration of a porous coated plate has been
widely studied in the aim to investigate the effect of a
poroelastic layer on the vibrations of an elastic
structure. Numerical methods such as the finite
element method (FEM) can be used to model such
configurations subjected to various kinds of loading
and boundary conditions. However, it may lead to an
important number of degrees of freedom when porous
materials are involved. In consequence, it is relevant to
develop simplified finite elements models dedicated to
the configuration of a porous coated. Dauchez[1]
recently proposed a model based on the principle of an
equivalent elastic plate accounting for the elastic plate
and the solid phase of porous materials. However,
viscous or thermal effects in the poroelastic layer,
related to the motion of the fluid phase, were
neglected. As a result, the model proposed by Dauchez
dissipation mechanisms become predominent. In this
paper, a simplified F.E. model dedicated to the
configuration of an elastic plate coated by a porous
material is presented. In this approach, all dissipation
mechanisms in the multilayer are taken into account. In
the following, the theory is first presented. Considering
the multilayered structure in free field, the results given
by the simplified model are then compared to the
predictions of a complete FE model.
THEORY
The present approach starts from the weak
formulation for the coupled system composed of an
elastic material and a porous material. The weak {u,P}
formulation is used for the description of the
poroelastic domain[2]. One writes:

(u ) : (u )d 2 u u d + ( ) ( )
~ 2 u s u s d +
~ s u s : s u s d
( ) (1)
el el el el el el
n u dS
el el
el
el el p p p / el

~ ~
Q
2
~
2
2
( ) Q
( ( ))
p p ~ pp d ~ + 1 + ~ p u d 1 + ~ pdiv u d = 0
s s

p 22 R R p R p

where el, el and uel are the stress and strain tensor poroelastic coefficients. The first and the second
related to the elastic domain and its displacement volume integrals in equation (1) are related to the
~ s and s are the stress
vector, el is its mass density. deformation and kinetic energy of the elastic domain.
The third and the fourth volume integrals are related to
and strain tensor related to the solid phase of the the deformation and kinetic energy of the solid phase
porous material in vacuo. us and p stand for the of the porous material. The integrals in the dotted box
displacement vector of the solid phase and the pressure are associated to the fluid phase of the porous material
and ~
in the pores, ~ 22 are the complex dynamic mass and to the coupling between the two phases of the
density of the solid and fluid phase respectively. porous material. In the simplified model, these terms
denotes the porosity, ~ , Q ~ and ~ are complex are not calculated explictly. They are accounted for by
R
a modification of the plate structural damping as

SESSIONS
mentioned below. The volume integrals in equation (1)
related to the plate (elastic domain) and the solid phase
of the porous material in vacuo are discretized using 8-
node quadratic multilayer plate elements. The damping
coefficient of the layer corresponding to the elastic
plate is modified in order to take into account the
dissipation of energy by viscous and thermal effects in
the porous coating. This new value for , denoted eq,
is obtained as indicated in the following. The first step
consists in a characterization procedure of the
multilayer. Namely, each of the two subdomains are
modeled with the finite element method. The
dissipated powers in each layer are computed. The
modified damping coefficient eq is then defined at any
frequency by the formula:
Ppldiss + Ppovisq + Ppoth
eq =
Pplreac
Where Peldiss and Pelreac are the dissipated and reactive
power in the plate. Ppovisq and Ppoth are the power
dissipated by viscous and thermal effects in the porous
layer.
RESULTS
In this section, a 1mm thick simply supported
aluminum plate is considered. It is coated with a 2cm
thick mineral wool layer. The porous coating is bonded
on the plate and its edges are free as well as its rear
face. The lateral dimensions of the multilayer are
0.35m*0.22m. The characteristics of the materials are
given in table 1. The excitation is a 1N amplitude
point force normal to plate. The mean square velocity
of the plate, denoted <Vz>, is the vibratory indicator
(2) of interest. The multilayer is in free field.

Table 1. Characteristics of the materials

Material E (kPa) s (kg/m3) (kN.s/m4) (m) (m)

Aluminum 6.9*107 0.33 0.1 2700 - - - - -
Mineral wool 42 0 0.05 600 0.95 1.4 25 93.2 93.2

Figure 1 represents the predictions for <Vz> obtained
with the simplified model in the frequency band
[10Hz;500Hz]. This model uses a 12*8 Quad8 element
mesh for the multilayer. The results are compared to
those obtained with a F.E. model modeling both the
free field. In the oral presentation, the present approach
is tested when the multilayer is coupled to a fluid
cavity.
140
Meqn square velocity (dB ref. 1m/s)

plate and the poroelastic layer with non coincident

130
mesh[3]. The characteristics of this latter model,
referred to as complete F.E. model are : a 15*10 120

Quad4 plate element mesh for the plate; the porous 110
material, modeled with hierarchical poroelastic
100
elements[3], is discretized with a mesh comprising
3*2*1 elements. The interpolation order is equal to 6 90 plate without coating
complete F.E. model
in the solid phase and 4 in the fluid phase of the porous simplified model
80
material.
A very good agreement is found between the complete 70
0 50 100 150 200 250 300 350 400 450 500
F.E. model and the simplified model below 200Hz. Frequency (Hz)
Beyond that frequency, the mass effect due to the
porous coating is depicted with less accuracy than with FIGURE 1: Mean square velocity of the plate
the complete F.E. model. One notes that the strong
absorption of the plate vibrations at higher frequencies REFERENCES
due to the importance of viscous effects is predicted
with an acceptable precision. [1] Dauchez, N., Sahraoui, S., Atalla, A. Dissipation
mechanism in a porous layer bonded onto a plate
CONCLUSION (submitted), 1999.
[2] Atalla, N. Panneton, R., Debergue, P. A mixed
displacement-pressure formulation for poroelastic materials,
This paper presented a simplified F.E. model J.A.S.A. 104, 1444-1452 (1998).
dedicated to the configuration of a porous coated plate [3] Rigobert, S., Sgard, F., Atalla, N. Numerical modeling
and accounting for all the dissipation phenomena in the of multilayered structures including porous materials :
multilayer. An acceptable agreement with a complete hierachical elements and non coincident meshes, in
F.E. modeling was observed when the multilayer is in Proceedings of Euronoise2001, Patras, Greece, January
2001.

SESSIONS
Integral formalism and finite element method applied to sound
synthesis by physical modeling
J. Bensoam, N. Misdariis, C. Vergez and R. Causs
Ircam, CNRS UMR 9912 1, place I. Stravinsky, 75004, Paris
To extend the possibilities of the sound synthesis software Modalys developed at Ircam (http://www.ircam.fr/) for musical applications
and research, we analyze the dynamic behavior of any elastic body under arbitrary boundary conditions. The Green formalism applied
to the continuum mechanics leads to integral solutions which incorporate directly the boundary conditions. These solutions are
discretized by using the finite element method. Numerical calculation is done in two step. The first deals with calculating Green
elementary tensors proper to each system by modal decomposition. These tensors represent the dynamic signature of the body and
need to be calculated only once. The second part, much faster, is a convolution of this signature with the external influences applied
on the structure.

THE BOUNDARY-VALUE PROBLEM
Within the framework of the linear theory of elasti-


city, we study an elastic region bounded by a surface
ter discretization of the equations of the movement. The
numerical linear system representing the balance of the
 
forces is first of all written classically :

 
/ One imposes on the sur-

AU F (1)

 
n d , n d 0.
face of Neumann, n , the forces hi r;t while on the sur- The matrix A  
K iD 2 M gathers the matri-
face of Dirichlet displacements ui r;t are supposed to be

ces of stiffness K, damping D and mass M of the free
known. The forces imposed inside are characterized by
 
structure; the vector F represents the nodal forces and
the function fi r t (see fig. 1). the vector U displacements at the nodes of the network.
 
We want to calculate the displacement field u r;
and the stress field i j r; in the elastic region accord-
The boundary conditions (displacements U imposed on
the surface of the body) are then taken into account by
ing to the boundary conditions (f, h and u ). However, eliminating judiciously some degrees of freedom. To do
when one studies the problem of the interactions between so, one carries out a partition of the vector U in the fol-
structures, it is also necessary to know the forces h and lowing way
displacements u on surface which are not prescribed
by the boundary conditions. U 
U1 U
U2  (2)

where U1 represents the displacement of the nodes which

FINITE ELEMENT RESOLUTION belong to the Dirichlet surface d (see fig. 1). Develop-
ing the equilibrium equations (1), one obtains U2 , the un-
In the literature concerning the finite element method, known displacements and F1 reactions on the Dirichlet
surface
 A U
the problem with inhomogenous conditions on the sur-
face of Dirichlet is rarely studied [4], [5]. In this case, the U2 A221 F2 (3)
A A F   A  U
21
boundary conditions are integrated in calculation only af- 1
F1 12 22 2 11 A12 A221 A21
Is this solution obtained with numerical methods
equivalent to the discretization of the one obtained with
h the formalism of continuum mechanics? We show on the
n one hand, that it is possible to get to an explicit formula-
tion of this continuous solution and on the other hand that
its discretization leads to the numerical solution (3).

f

GREENS FORMALISM
d u
FIGURE 1. Region bounded by a surface  
n d
To establish the correspondence between the numeri-
cal finite elements method and the continuous formalism

SESSIONS
we are interested in the Greens formalism which al-
 P11 S  A
1
11 
A12 A221 A21 S111
where A  is the matrix of equation (1). The coefficient
11
lows us to deal with the boundary-value problem without
being obliged to solve integrals equations [1], [4]. The
identity principle of Maxwell-Betty applied to the dy- Gi j of the Green matrix represents the displacement of the
namic problems of linear elasticity in an region makes node i excited at node j by a unit force. The coefficient



it possible to represent the field of displacement at a point
in the form
Ti j gives the force which acts on the Dirichlet surface at
node i for the same unit excitation at node j. Taking into
  
r E

u r  
G r  r  f r  dV
account these expressions we show that the discretisation

 G  r  r  h  r  dS
T  r  r  u  r  dS
k
(4) of continuous equations (4) and (7) leads to the numer-
k i 0 i 0 0
k k ical solutions (3). We can thus conclude that the finite

n i 0 i 0 0 d i 0 i 0 0
element method is to the discrete domain what the Green
The elementary solution G r  r  indicates the displace- k
i 0
formalism is to the continuum mechanics.
ment, at position r, in the direction i, due to a unit force1 Practically, we must calculate the inverse of the matrix

 
fk applied in the direction ek 2 at a fixed point r0 . The
notation Tk r0 r , associated with the elementary solu-
A22 of which we know now that it represents the numeri-
cal tensor of Green. We carry out this operation by modal
tion G, indicates the force at the point r which is applied decomposition and solve the eigenvalue problem in the
on the surface of normal unit n due to the same unit complex field to take damping into account. Solutions (3)
force fk . are finaly transposed into the temporal domain to obtain
The couple (displacement G and force T) satisfies ho- time dependent results.
mogeneous boundary conditions on surfaces of Dirichlet
d and Neumann n :
    r CONCLUSION

Gki r r0
T r r 
k
0

r d
n  r
0

(5)


i 0 0 r 0 (6) The choice of the finite element method is interesting

Within this framework, the calculation of the force h r 

in our purpose because numerical calculus can be divided
i s into two parts. The first deals with calculating elementary
which is applied on the surface simply gives
  
tensors specific to each system by modal decomposition.
h r  
T r  r  f r  dV

 T r  r  h r  dS
P  r  r  u  r  dS
  a The tensors represent the dynamic signature of the body
i s i (7) 0 s a 0 0 and need to be calculated only once. The second part,
a a

  
n i 0 s a 0 0 d i 0 s a 0 0 much faster, is a convolution of this signature with the

where we stated P r  r  C T  r  r  n r  . The

a k
i 0 s abkl i lo
solutions u and h of our problem are thus obtained in the
continuous domain according to the elementary tensors
G, T and P which must still be given.
Tensors discretization
external influences and the boundary conditions.
0 s b 0 Futhermore, even if we deal with a numerical method
by discretising space into finite elements, the Green for-
malism allows us to obtain analytic solutions for the
time part of the signal. It follows that a good quality is
achieved in predicting the transients which are known to
be primordial in perception of realistic sound.

In the finite elements basis the tensors G, T and P REFERENCES

are represented respectively by the matrices G, T and P.
These are obtained by discretization of the virtual power 1. M. Bonnet, Equations intgrales et lments de frontire
principle applied to the elementary solution. By preser- CNRS ditions/Eyrolles, 1995.
ving the partition of the degrees of freedom of the equa- 2. M. Bruneau, Manuel dacoustique fondamentale Hermes,
tion (2), one can find3 1998.

G 
0 0
0 A221  T
S0 11
1

S111 A12 A221
0


3. A.C.Erigen, E.S Suhubi, Elastodynamics Academic Press,
1975, Volume II-linear theory.
4. M. A.Hamdi, Rayonnement acoustique des structures Ey-
rolles, 1988, chapter 9.
5. J. F. Imbert, Analyse des structure par lments finis
1
2
fk r 
 r  r e
0 0 k
Cepadues Ed., 1984, pp. 111-112.
e : unit vector in the direction k
k 6. L. Schwartz, Mthodes mathmatiques pour les sciences
3 S is the normalization of the interpolated functions on Dirichlet sur- physiques Hermann, 1979.
face d

SESSIONS
 Full Band-Gap of Sonic Crystals Composed of A Periodic
Array of Acrylic Cylinders in Air
Numerical Predictions and Experimental Observations
T. Miyashita, C. Inoue and K. Sakata
Department of Electronics and Informatics, Ryukoku University, 520-2194 Ohtsu, Japan
E-Mail: miya@rins.ryukoku.ac.jp

Abstract A sonic crystal is formally an acoustic version of a photonic crystal. They are artificial crystals composed of a
periodic array of scatterers imbedded in the host material, and expected to have full frequency band-gaps where the wave cannot
propagate at all. Sonic crystals have been, however, shown not to be an acoustic replica of photonic crystals by clarifying the
correspondence between the sound wave and the electromagnetic wave in the two-dimensional space. First, we have predicted
that sonic crystals composed of a periodic array of solid cylinders in air have full band-gaps with a sufficiently large filling ratio,
developing a FDTD method to analyze numerically acoustic wave propagation in a finite periodic structure. Next, according to
the prediction, we have constructed a two-dimensional array of 10 11 acrylic cylinders of a radius of 10.225 mm and a lattice
constant of 24.0 mm in air. We have observed a full band-gap between 7.2 kHz and 9.4 kHz with a 35 dB transmission ratio using

a burst wave. We have constructed also a sonic waveguide, and observed a good wave-confinement. Namely, the amplitude ratio of
the guided wave to the leakage has been about 30 dB.

INTRODUCTION finite difference time domain (FDTD) method [7,8] us-

We are interested in the realization of a sonic crys-
tal which is formally an acoustic version of a photonic
crystal for electromagnetic waves [1]. A sonic crystal
is an artificial crystal composed of a periodic alignment
of acoustic scattering materials imbedded in the uniform
host material as shown in Fig. 1(a), and expected to have
full frequency band-gaps, in which any acoustic wave
cannot propagate in the crystal [2-5]. Efficient acous-
tic waveguides, especially bending waveguides, as well
as acoustic directional couplers, are to be realized in the
sonic crystals as shown in Fig. 1(b) just like photonic
crystal ones [6].
ing the perfectly matched layer (PML) [9] for absorb-
ing boundaries. Next, we show experimental observa-
tions of a wide and deep full-band-gaps of the crystals,
and further we show experimental evidences that sound
waves are well confined in a sonic crystal waveguide.
SONIC AND PHOTONIC CRYSTALS
We notice the correspondence relationships in the
wave equations and in the physical parameters be-
tween two-dimensional sound waves and electromag-
netic waves [8], and classify the artificial crystals in
the parameter space according to the correspondence as
shown in Table 1. Here Z and c denote, respectively,
Table 1. Correspondences between sonic crystals and photonic
crystals and their classification.
Class Example Z, c
Air
Sonic I Z 1, c 1
Agar
 

GaAs
Photonic I TM Z c 1
Air



(a) Sonic crystal (b) Bending waveguide

Acrylic resin
FIGURE 1. Sonic crystal and bending waveguide made of a Sonic II Z  1, c  1
Air
periodic array of cylindrical scatterers.
Air
Photonic II TE Z c 1
GaAs



First, we show some interesting possibilities of two-

dimensional sonic crystals, whose parameters belong to
an oppositely different group from those of well-known characteristic-impedance ratio and wave-velocity ratio
photonic crystals, from numerical analyses developing a of the scatterers to the host material. The first-type sonic

SESSIONS
 Normalized Transmission [dB]
crystals which correspond to the well-known photonic -10
0

crystals of GaAs scatterers in air should be composed -20

(
)

Full Band-Gap *

of fluid (or solid) scatterers imbedded in a solid host -30

-40
[100]
medium. The second ones should be composed of solid -50
[110]

1.5 2 2.5 3 3.5

scatterers imbedded in a fluid host medium. Although Normalized Wavelength [/a] %
& '

the latter type has a preferable structure for sound waves, FIGURE 4. Experimental transmission characteristics of a
it is not usual for photonic crystals. In this report, we sonic crystal made of acrylic cylinders in air
have investigated the latter type.
NUMERICAL FULL-BAND-GAP OF EXPERIMENTAL OBSERVATION OF A
SOLID SCATTERERS IN AIR SONIC CRYSTAL WAVEGUIDE
Numerically we find parameter combinations with We have constructed also a waveguide in a sonic crys-
which the second-type sonic crystals, e.g., a periodic ar- tal, and observed an evidence of good wave-confinement
ray of acrylic-resin cylinders in air, have a full band-gap as shown in Fig. 5. Namely, the amplitude ratio of the
as shown in Fig. 2. The gray region in the figure shows guided and output wave to the side-leakage has been
about 30 dB in the full band-gap of the sonic crystal.
0.6
Filling Ratio r2/a2


Frequency [kHz]
+ , -

4 5 6 7 8
8

9 10 11 12


0.5
Guided Wave
3

2
4

Leaked Wave


Amplitude [V]
1.5
Full Band-Gap
, 6

0.4 1
2

1
/

1 1.5 2 2.5 3 3.5 4

Normalized Wavelength /a

 

0.5

FIGURE 2. Full band-gap map of a sonic crystals. 7

0
0.3 0.4 0.5 0.6 0.7 0.8
5

Normalized Frequency [a/]

the full band-gap as a function of filling ratio of the scat-
terers and the normalized wavelength. Two numerical
examples of the wave-transmission ratio versus wave-
length are shown in Fig. 3. The solid lines are for the
plane wave incident in the [100] direction of the crystal,
and the broken ones for the wave incident in the [110]
direction. Frequency band in which the transmission ra-
tio is less than 25 dB independent of the direction of
wave incidence is called full band-gap in this report.
Normalized Transmission Ratio [dB]
FIGURE 5. Experimental evidence of wave-confinement in a
sonic crystal waveguide.
CONCLUSION
We have experimentally realized a sonic crystal made
of acrylic cylinders in air and also a sonic-crystal wave-

guide based on their numerical predictions with FDTD
method. They have a practical and wide full-band-gap
and a good wave-confinement. Useful applications of
these crystals will be the next subjects.

Normalized Transmission Ratio [dB]

0 0

 

ACKNOWLEDGMENTS
 

-10 -10

 

 

-20 -20
full band gap
This work is partially supported by a Grant-in-Aid for
 

 

-30 -30
 




-40
[100] 


-40
#
[100] Scientific Research (B) from the Ministry of Education,
[110] [110]
1 1.5 2 2.5
Normalized Wavelength /a







3 3.5
!

4 1 1.5 2
Normalized Wavelength /a

"
2.5


3 3.5
!

4 Science, Sports and Culture of Japan.

(a) Filling ratio: 0.283 (b) Filling ratio: 0.567 REFERENCES
FIGURE 3. Transmission characteristics of a periodic array of 1. E.Yablonovitch, J. Opt. Soc. Am. B 10 283 (1993).
2. E. N. Economou and M. M. Sigalas, Phys. Rev. B 48 13434
acrylic cylinders in air. (1993).
3. M. S. Kushwaha, P. Halevi, L. Dobrzynski and B. Djafari-
Rouhani, Phys. Rev. Lett. 71 2022 (1993).
4. F. R. Montero de Espinosa, E. Jimenez and M. Torreset, Phys.
EXPERIMENTAL FULL-BAND-GAP OF Rev. Lett. 80 1208 (1998).
5. J. V. Sanchez-Perez, D. Caballero, R. Martinez-Sala, C. Rubio,
ACRYLIC CYLINDERS IN AIR J. Sanchez-Dehesa, F. Meseguer, J. Llinares and F. Galvez,
Phys. Rev. Lett. 80 5325 (1998).
According to the numerical predictions, we have con- 6. S. Fan, A. Mekis, S.G. Johnson and J.D. Joannopoulos, Manip-
ulating Light with Photonic Crystals, in Nanoscale Linear and
structed a two-dimensional array of 10 11 acrylic $

Nonlinear Optics, edited by M. Bertolotti, C.M. Bowden and C.

cylinders of a radius of 10.225 mm with a lattice con- Sibilia, Melville, New York, 2001, pp. 57-76.
7. K. S. Yee, IEEE Trans. Antennas Propagat. 14 302 (1966).
stant of 24.0 mm in air. We have observed a full band- 8. T. Miyashita and C. Inoue, to be published in Jpn. J. Appl. Phys.
gap between 7.2 kHz and 9.4 kHz with a 35 dB trans- 40 (2001).


9. J. -P. Berenger, J. Computational Phys. 114 185 (1994).

mission ratio using a burst wave, as shown in Fig. 4.

SESSIONS
 Perceptually optimized time-varying wavelet packet
decomposition and its applications in acoustic signal
processing

Alexey Petrovsky

Computer Science Department, Belarusian State University of Informatics and Radioelectronics

6, P. Brovky St., 220027 Minsk, Belarus, e-mail: apetrows@bas-net.by

We proposed the method of adaptation wavelet packet (WP) decomposition on each frame of signal such that perceptual
entropy and, ultimately, the bit rate are minimized. The objective of the tree adaptation process, therefore, is to construct a
minimum cost subband decomposition by maximizing the minimum masking threshold in wavelet domain (which is limited by
the perceptual entropy) in every subband. For audio coding, for example, a complexity-constrained tree adaptation procedure is
shown to yield a basis requiring the fewest bits for perceptually transparent coding for a given algorithm complexity or a given
embedded processor architecture and temporal resolution.

INTRODUCTION {ws}s=1,2,3,. The entropy (Shannons entropy) of each

The known best basis tree methodologies for
adapting the WP decomposition to signal properties
are typically formulated in terms of Shannon entropy
and other perceptually blind statistical measures [1].
The questions of perceptually motivated filter selection
and WP decomposition construction are central to
successful application of WP analysis in audio coding
algorithms [2,3]. Several WP audio algorithms have
successfully employed time-invariant WP tree
structures that mimic the ears critical band frequency
resolution properties.
Application of the entropy concept to audio signal
coding algorithms, in particular, to reduction of
musical noise generated by spectral subtraction
method, can be stated as follows. The noise can be
reduced in loudness or even made completely
inaudible in the signal by a phenomenon known as
auditory masking. The given noise is localized in the
frequency domain by using the critical band analysis
based on the psychoacousticaly motivated multirate
filterbank (time-invariant analysis wavelet). Through a
multiscale approach like the wavelet transform, the
time entropy is able to easily detect the relevant scales
(frequencies) and allows to compare various scales
(frequencies) in order to differentiate various cases.
WAVELET TIME ENTROPY AT SCALE
S (WTES) CALCULATION
The WP is applied on the input signal. The result is
some different sets of wavelet coefficients; each one,
corresponding to a different scale. The scales are noted
scale is then computed and corresponds, respectively,
to the approximate frequency ranges by
log 2 ( min 0 ) s log 2 ( max 0 ) , (1)
where min and max are min and max values of the
frequency interval, 0 is a interesting us frequency.
Explicitly, the wavelet coefficients can be written as
summations. The filters h and g are taken from [4]
w1,i = hi ,k w
~ , w = h
k 2 ,i i,k g k ,k1 w~ k1 ,... (2)
k k k1
Thus, we define WTEs by the equation
,
WTE s = ws ,k w s ,i ln ws ,k w s ,i (3)
k i i
where the ws are the wavelet coefficients at scale s.
PERCEPTUALLY ADAPTED WP FOR
THE GIVEN PERFORMANCE
We present a reconfigurable adaptive WP tree
derived via dynamic algorithm transforms (DATs).
The principle behind DAT is to define parameter of
input audio signals (subband entropy) and output
coded sequences (subband rate) for the given
performance. In other hands, DAT techniques is to
construct a minimum cost subband decomposition of
wavelet transform by maximizing the minimum of the
masking threshold (limited by the perceptual entropy)
in every subband for the given algorithm complexity
and temporal resolution.
The pruning algorithm based on the Johnsons
function (perceptual entropy) [5] as a cost function
which is computed for the each nodes of WP tree scale
in the wavelet domain. Suppose the split decision as

SESSIONS
J(zi s), where s is decomposition scale, and zis denotes
the ith node in the level s as:
J ( z is ) = log 2 (SMR( z is )) , (4)
s ,i
where SMR is a relation between WP coefficients ws,i
and perceptual threshold T(zis) in the node zis:
SMR = ws ,i
c T (z is ) K (z is )
s
where K(zi ) is number of the WP coefficients in node
zis, |a| is rounding a to the nearest integer, it is needed
to escape the overflow, coefficient c is defined by
quantizer type, T(zi s) is perceptual threshold in the
node zi s [6]. The analysis filters are time-invariant and
obtained from the family of Daubechies wavelets.
Thus, the WP pruning algorithm is to search for the
best bases from the all possible WP decomposition can
be stated as follows:
Let: N is a length of the input signal frame, log2N is a
tree depth corresponding to the critical bands tree
structure, zis is a current node, and z2is+1, z2i+1s+1 are the
children nodes of zi s, WTEs is a wavelet time entropy
for the scale s (3), C is a given algorithm complexity.
Step 1: Set split decision for all the nodes zis = YES,
s = 0, and define WTEs.
Step 2: s=s+1. Define WTEs, current computational
complexity cs and cost function J(zi s) (4).
IF s > log2N, THEN GOTO step 5;
IF WTEs > WTEs-1, THEN HALT and GOTO step 5;
IF cs > C, THEN HALT and GOTO step 5.
Step 3: FOR each node i at the scale s,
IF the split decision of parent of zis is YES, THEN
DO the pruning as follows,
IF J(zis)< J(z2is+1)+ J(z2i+1s+1), THEN zis=NO
ELSE IF J(zis)==0 THEN zis=NO
ELSE zis=YES.
Step 4: GOTO step 2.
Step 5: The optimal WP tree structure is found (see
Figure 1). Go to processing the next input signal
frame. Set s=s - 2, define WTEs and GOTO step 2.
Frequency
Frequency
Time
FIGURE 1. Dynamic WP tree decompositions depending on the input signal with a tilling time-frequency maps.
The advantages of the above algorithm are
summarized as follows: pruning method is a top-down
method, and the WP pruning can be viewed as a split
process, i.e. we have the temporal construction WP
tree for each signal frame that is ideal decision for real
time processing implementation.
(5)
CONCLUSION
Thus, adaptive wavelet analysis for audio signal
coding purposes is only interesting if psychoacoustic
information is considered in the WP decomposition
scale. Due to the lack of selectivity of wavelet filter
banks, psychoacoustic information is computed in the
wavelet domain, which makes possible to achieve
minimum bit rate for transparent audio coding.
Daubechies wavelets provide better results and coder
performance improves as the wavelets order is 20. The
low bit rates (64 Kbps) coding algorithm based on this
method was designed.
REFERENCES
1. Coifman R., Wickerhauser M., IEEE Trans. Inform.
Theory, vol. 38, Mar. 1992. pp.712-718.
2. Painter T., Spanias A., The proc. Intern. conf. DSP97,
Creece 1997.pp. 179-208.
3. Ruiz Reyes N., Rosa Zurera M., and et al., IEEE Nordic
Signal proc. symposium, Sweden 2000. pp.331-334.
4. Vetterli M., Kovacevic J., Wavelets and Subband
coding. Engle-wood Cliffs, NJ, Prentice-Hall, 1995.
5. Johnston J.D., IEEE Trans. On Select. Areas Commun.,
vol. 6, Feb. 1988, pp. 314-323.
6. Petrovsky Al., Perovsky A.A., AES 110th Convention,
pre-print 5298, 12-15 May, Amsterdam, The
Netherlands, 2001. 8 p.
Frequency
Time Time
SESSIONS
 2-D unstructured numerical simulation for acoustic pulse
scattering at water-immersed steel plate.
Lamb and Scholte-Stonneley waves investigation.
L.P. Derbesse(a) , P. Voinovich(b) , A. Merlen(c) , P. Pernod(a)

(a)
IEMN-DOAE, Groupe Electronique Acoustique, Ecole Centrale de Lille, UMR CNRS 8520, B.P 69, 59651
Villeneuve dAscq cedex, France
(b)
Soft-Impact Ltd., P.O. Box 33, 194156 St.Petersburg, Russia
(c)
Laboratoire de Mcanique de Lille, URA CNRS 1441, USTL1, Cit scientifique, 59655 Villeneuve dAscq cedex,
France

A numerical method for transient acoustic scattering on 2D elastic bodies has been developed and successfully compared to
experimental results obtained by ultra high speed shadowgraph visualizations carried out with ultrashort, spark-generated
acoustic pulses. This model universally includes elastic solids and liquids. The equations of motion are written in terms of
displacements and displacement velocities for control volumes constructed about the nodes of a triangular unstructured grid. The
latter conveniently supports internal boundaries separating sub-domains of different elastic properties.
This method describes the mechanisms of diffraction upside and inside the plate and particularly the creation of the surface (A)
wave from the bending Lamb wave (A0) .A double Fourier Transform in space and time applied to the displacements of a plate
surface provides the dispersion curves and the amplitudes and velocities of each particular waves. In addition, a selective filtering
in (f,k) domain and a 2D inverse Fourier Transform was used to select specific diffracted waves and identify the corresponding
features in the experiments. Comparison with the respective experiment reveals a good agreement in such features as excitation
of Lamb waves (An, Sn ; 0 n 2 ) and the Scholte Stonneley surface wave (A).

INTRODUCTION allow complex geometries and possible local grid

A specific and original numerical method [1] for the
simulation of transient acoustic scattering on 2D elastic
bodies is presented. This approach was developed in
order to provide a theoretical basis for interpretation of
experimental results obtained by ultra high speed
shadowgraph visualization carried out with ultrashort
refinement. A triangular grid generation is provided
automatically for a wide class of multi-connected
domains. The control volumes are constructed about
the grid nodes and bounded by medians of the
triangular grid elements. The momentum equation is
written for the control volumes V in an integral form:

acoustic pulses [2].
A post-processing of the numerical results has been
implemented to quantitatively register the dispersion
curves of the Lamb waves.
In the first section, the theoretical considerations of
the numerical method is discussed. In the second
section, numerical and experimental results are
presented and compared for the study of an immersed
plate of constant thickness. A post-processing is
applied to give a space-time representation of various
Lamb waves modes.
NUMERICAL METHOD
The numerical method is based on a direct
simulation of transient motion in an elastic medium[1].
The unstructured grid approach has been selected to
dV = df
d 2U i
ik k (1)
dt 2
Ui being displacement in x or y direction,
- density, ik - components of the stress tensor,
fk x or y area projections of the control volumes
faces.
An explicit integration in time is applied to solve this
equation. In computations, the domain is partitioned
into subdomains to accommodate substances of
different elastic properties. The subdomains are
coupled ensuring equality of displacements and
displacement velocities normal to the separating
boundaries.

SESSIONS
 Ao So
A So

a) b) Ao

FIGURE 1: Numerical simulation and experimental visualization of acoustic scattering at elastic plate

Wave number(mm -1)

Position 1,5

1
Ao
0,5
time So S1
a) b) A1 S2 MHz.mm
0 2 4 6 8
FIGURE 2:Time history and frequency wave number representation for displacement on plate

FREQUENCY-WAVE NUMBER determine the spatio-temporal shape of each mode, in

ANALYSIS
The experimental method is based on a combination
of quasi-Dirac spark source and high-speed
shadowgraph visualization.[2]
Results obtained by computing sound waves radiated
from an aluminium plate of 1 mm thickness loaded
with water on each side are presented in Fig 1.a. They
can be compared to experimental result of fig.1.b. The
numerical simulation was performed using a rather
coarse grid containing 8 internal nodes across the plate.
In addition to the incident wave and the specular
reflexion, the photograph shows a qualitative
representation of the shape of the re-radiation of the
So, Ao Lamb modes and Scholte-Stonneley A wave[3].
A good qualitative agreement with numerical results
can be observed in this case.
Figure 2.a) shows the time history of the plates
response to excitation at the boundary between plate
and water and can therefore be used to measure waves
amplitudes and velocities
In order to determine precisely the reemitted waves,
a 2D Fourier transform was applied.
The results are presented in fig. 2.b in form of
amplitude versus frequency and wave number. This
provides directly the dispersion curves and points out
the various modes propagating inside the plate. Owing
to this presentation, individual Lamb waves may be
resolved.
Five differents Lamb modes are clearly identified:
A0, S0, A1, S1, S2. This analysis indicates that all
these modes are present in both numerical and
experimental results.
A selective filtering in (f,k) domain and a 2-D
inverse Fourier Transform can be performed to
the simulation picture as well as experimental results.
An example of this processing applied for the selection
of the S1 Lamb mode is presented in Figure 3.
Position
time
FIGURE 3: Time history of S1 Lamb mode
CONCLUSION
The considered numerical method allows a good
physical understanding of the transient patterns and the
generation conditions of various Lamb waves modes
(An , Sn ; 0 n 2 )
This method can be extended and applied to many
other fields where transient elastic and acoustic waves
are involved, and to more complex geometries.
REFERENCES
1. Voinovich P.A.et al., "Numerical simulation of wave
scattering at shock-loaded metallic plates and shells in
water , 23 Int. Symp. On Shock Waves, Fort Worth,
Texas, USA, July 22-27 2001
2. Ahyi A., Pernod P., Latard V., Merlen A., Uberall H.,
J.Acous.Soc.Am, 104, p.2727,(1998).
3. Derbesse L. et al., IEEE WCU, pp.573-576, October
2000.

SESSIONS