Accelerated vibro-acoustics of porous domains via a

novel coupled multiscale finite element method

A. Sreekumar 1 , S. P. Triantafyllou 2 , F.-X. Bécot 3 , F. Chevillotte 3 , L. Jaouen 3

1 University of Nottingham, Faculty of Engineering,

Centre for Structural Engineering and Informatics, UK

s
2 National Technical University of Athens, School of Civil Engineering,

20 ng
Institute for Structural Analysis and Aseismic Research, Greece
e-mail: savtri@mail.ntua.gr

20 di
3 Matelys - Research Lab
7 rue des Maraı̂chers Bâtiment B, Vaulx-en-Velin, 69120, France

e
Abstract SD ce
The use of foam materials as a means of improving the sound absorption and transmission properties of
o
structures, especially at low frequencies, has been receiving considerable attention over the past years. Un-
fortunately, the underlying complex material domain that may include solid inclusions, renders numerical
-U r
simulation using the standard FEM a computationally taxing task. Within the taxonomy of multiscale sim-
A yp

ulation methods, the Coupling Multiscale Finite Element Method (CMsFEM) has been recently introduced
as a means of reducing computational costs for the consolidation analysis of highly heterogeneous domains.
In this work, we originally apply the CMsFEM framework to the Biot theory of elastic wave propagation
in air-saturated porous media. The resulting numerical scheme allows inclusions of air and other potential
M ar

sources of heterogeneity at the mesoscopic scale. The performance and accuracy of the method, along with
applications to materials with inclusions are explored using an in-house MATLAB code.
IS in
im

1 Introduction

1.1 Overview
el

The vibro-acoustic performance of porous materials is controlled by structural and visco-inertial-thermal

Pr

dissipation effects of the solid skeleton and pore-fluid, respectively. In several cases the solid skeleton is
quite stiff and the elastic effects may be neglected under the rigid skeleton assumption. In such cases, the
classical Helmholtz theory [1] can be applied to compute pore-fluid pressures. This theory however cannot
capture the significant resonance effects that are manifested due to the deformability of the solid skeleton. In
such cases the modified Biot equations [2] have to be employed.
The Biot theory provides a phenomenological model that predicts the behaviour of waves propagating inside
a fully-saturated porous material at the meso-scale. Considering that the wavelengths under examination are
much larger than the average pore diameter, a periodically repeating unit cell, representative of the entire
domain (also known as an RVE) is constructed. Effective material parameters are obtained over this RVE
through homogenization schemes [3].
In vibro-acoustics, all dissipative effects are captured using complex parameters. Structural dissipation is
accounted for by the loss factor η˜s (ω). Fluid dissipation effects can be either viscous or thermal in character
at the microscopic level. The dynamic density ρ̃eq (ω) and dynamic bulk-modulus K̃eq (ω) capture these
effects.

1
2 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

The model, as originally formulated by Biot [4, 5] is a u − U formulation, i.e., the primary field variables
are the solid-skeleton and pore-fluid displacements. Hence, using a standard finite element approach to
discretize and solve the governing equations results in 4 degrees of freedom (DOFs) per node in 2-D and 6
DOFs per node in 3-D. An alternate mixed u − p formulation was proposed in [6], where the primary field
variables are solid-skeleton displacements and pore-fluid pressures, respectively. In this case, only 3 DOFs
per node in 2-D and 4 DOFs per node in 3-D are required hence resulting in a significant reduction to the
corresponding computational costs. However this alternate formulation requires a reworking of the original
Biot parameters.
The formalism developed in [7], overcomes this limitation by providing a robust methodology to decouple
the computation of the dissipation parameters from the rest of the terms. This allows one to insert any

s
chosen dissipation model into the equations. This is desirable, as estimation of macroscopic parameters for

20 ng
porous materials is a challenging task. For instance, when one does not have all the parameters required in
a six parameter semi-phenomenological dissipation model (the Johnson, Champoux, Allard, Lafarge model
(JCAL)) [8, 9, 10], one can instead choose a model for which all required parameters are indeed available,

20 di
e.g. the one parameter Delaney-Bazley-Miki model (DBM) [11, 12].
Upscaling techniques are used in numerical methods to drive down computational costs incurred when mod-

e
elling physical phenomena exhibited by porous and composite materials.

SD ce
o
-U r
A yp
M ar
IS in
im

Figure 1: Motivation for multiscale approaches

el

Such materials have observably different scales as illustrated in Fig. 1. They can include complicated pore
Pr

geometries created by fluid-solid skeleton interfaces, and varying material properties, e.g, elastic moduli,
porosity, permeability. These variations can have a measurable impact on the behaviour of the material
under consideration, at all scales. It is evident that numerical approaches that can account for these vari-
abilities in micro-structural configurations can provide valuable insights into mechanical and vibro-acoustic
characterizations.
The transfer matrix method (TMM) [13] is used to model acoustic wave propagation in layered media. These
can include solid, porous and fluid layers. Matrices are used to describe the wave propagation through each
medium and the coupling constraints at layer interfaces. Semi-infinite flat surfaces are assumed for all layers.
A significant shortcoming of this method is its inability to account for heterogeneous non-planar configu-
rations. Alternative methods, relying on double porosity theory [14, 15] were proposed to model anechoic
wedges and heterogeneous porous composites, respectively. It is however, still insufficient to model complex
shapes with varied heterogeneous meso-scale inclusions. Such involved material layouts necessitate the use
of numerical techniques such as the Finite Element Method (FEM).
Classical FE approaches require an explicit resolution of all complex fine-scale morphologies. This can
be prohibitively expensive even when studying static behaviour. Dynamic or spectral approaches used for
V IBRO - ACOUSTIC MODELLING AND PREDICTION 3

Table 1: Material parameters used in governing equations

Parameter Description
E elastic Young’s modulus
ν Poisson’s ration
ηs (ω) structural loss factor
ρ̃(ω) modified Biot density
ρ̃eq (ω) dynamic mass density
K̃eq (ω) dynamic bulk modulus
γ̃(ω) coupling factor

s
20 ng
studying evolving phenomena, e.g., consolidation and sound absorption, further exacerbate this problem,
especially in the field of acoustic topology and shape optimization problems [16, 17].

20 di
Heterogeneous multiscale methods seek to address this by defining coarse-elements that cluster their own
set of fine elements. The coarse-element is understood as a generalization of the notion of RVEs. The
more typical notion of being representative of the entire domain is recovered in the periodically repeating

e
limit. The fine-scale heterogeneities are directly mapped to the coarse-scale through a set of multiscale basis
functions. SD ce
o
1.2 Governing Equations
-U r
In the following, we employ the mixed u − p formulation adopted by [2] and [7]. The momentum balance
A yp

equations for both phases are expressed as:

div(σ s ) + ω 2 ρ̃(ω)u = −γ̃(ω)∇p (1a)

∆p p
M ar

+ ω2 = ω 2 γ̃(ω)div(u), (1b)
ρ̃eq (ω) K̃eq (ω)
IS in

where σ s = D̃(ω)ε(u) denotes the in-vacuo stress tensor.

The elastic constitutive tensor D̃(ω) depends on the Young’s modulus E, the Poisson’s ratio ν, and the
im

structural loss factor ηs (ω), i.e.,

D̃(ω) ≡ D̃(E, ν, ηs (ω)). (2)
el

The terms ρ̃(ω) and γ̃(ω) represent a modified Biot density and coupling factor, respectively.
These equations are efficiently solved in the frequency domain. The material parameters used are sum-
Pr

marized in Table 1. The (˜·) symbol denotes the complex-valued nature of those parameters. There is a
parametric dependence of the complex-valued parameters involved, on the angular frequency ω. This is
consistent with experiments, as dissipation effects are frequency dependent.

1.3 Finite Element Formulation

Following the weak formulation proposed in [2] the following coupled system of equations is obtained, i.e.,
     u
K̃ − ω 2 M̃ −C̃ u f
= p (3)
−ω 2 C̃T H̃ − ω 2 Q̃ p f
4 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

Figure 2: Schematic diagram of a flat poroelastic layer in an impedance tube with rigid wall backing. The
material is subject to a normal incidence plane acoustic wave.

s
20 ng
where the state matrices have the following forms:
Z

20 di
K̃ = BT D̃(ω)B dΩ linear elastic stiffness matrix
ZΩ

M̃ = NT ρ̃(ω)N dΩ mass matrix

e
Z Ω

SD ce
o H̃ = (∇Np )T
Z Ω
1
1
ρ̃eq (ω)
(∇Np ) dΩ permeability matrix

Q̃ = NTp Np dΩ compressibility matrix

Ω K̃eq (ω)
-U r
Z
T
A yp

C̃ = NT γ̃(ω)∇Np dΩ coupling matrix.

Ω

Eq. 3 reduces to the simpler Helmholtz equivalent fluid representation in the limit of rigid skeletons or
infinitely soft skeletons (in the latter case however, one would end up with a different expression of the mass
M ar

density). The boundary conditions are first specified for a flat layer subject to impedance tube conditions, as
shown in Fig. 2:
IS in

u = 0, on Γ3 : bonded conditions (4a)

uy = 0, on Γ2 ∪ Γ4 : sliding conditions (4b)
im

p = p0 , on Γ1 : free end conditions (4c)

The constraints will change with the nature of the medium the poroelastic is coupled with (e.g., elastic solid,
el

fluid, poroelastic, rigid-wall etc.). For the study of absorption in poroelastic materials, there is no external
load. As a result, it is sufficient to have a zero vector in place of the load vectors f u and f p . The free-end
Pr

pressure constraint is a simplification of more complicated impedance-type Robin boundaries.

2 A heterogeneous multiscale method for vibro-acoustics

The Coupled Multiscale Finite Element Method [18], has been developed to drive down computational costs
when dealing with mechanical phenomena in highly heterogeneous two-phase porous media. In principle, to
accurately resolve heterogeneities, a very fine mesh discretization is required, often leading to prohibitively
expensive computations. The CMsFEM alleviates this problem by using a second, coarser, mesh; each coarse
element clusters its own portion of the underlying fine mesh. Fine-scale details are mapped onto the coarse
mesh through an upscaling procedure using numerically derived multiscale basis functions. The solution
procedure is finally performed at the coarse scale, thereby significantly reducing computational complexity.
In this work, the Biot (u−p) formulation in the ω-domain, as described in Section 1.2 is originally introduced
within the multiscale framework to achieve these objectives.
V IBRO - ACOUSTIC MODELLING AND PREDICTION 5

s
20 ng
20 di
e
SD ce
o
-U r
Figure 3: Visualizing multiscale basis functions for a heterogeneous domain
A yp

2.1 Evaluation of multiscale basis functions

M ar

The heterogeneous domain shown in Fig. 3 comprises nM = 36 coarse nodes and nMel = 25 quadrilateral
coarse elements. Each coarse element contains nm = 81 fine nodes and nmel = 64 fine elements. In each
coarse element a random distribution of elastic properties is assumed sampled from a normal distribution as
IS in

shown in Fig. 3. We restrict our discussion to the 2-D case, however, generalisation to 3-D is straightforward.
Assuming a generalized representation for the abstract weak form of the governing equations:
im


Find (u, p) ∈ V × W :

a(u, v) = F(v), ∀v ∈ V, (5)
el


b(p, w) = G(w), ∀w ∈ W,
Pr

the multiscale basis functions are evaluated through the solution of the piece-wise continuous homogeneous
version of Eq. 5 over each coarse element domain ΩM(α) , α = 1, . . . nMel for the solid and fluid phase,
respectively, as follows: 

 Solid Phase

Find u ∈ V(Ω
M(α) ) :
(6)
a(u, v) = 0, ∀v ∈ V(ΩM(α) )



u|∂Ω(M(α)) = ū,


Fluid Phase

Find v ∈ W(Ω
M(α) ) :
(7)
b(p, w) = 0, ∀w ∈ W(ΩM(α) )



p|∂Ω(M(α)) = p̄,

where V and W represent the space of all trial and test functions for the solid and fluid phases, respectively.
6 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

The terms ū and p̄ denote the prescribed displacement and pressure fields over the Dirichlet boundary. The
solutions u and p are the static modes of the coarse element and are equivalent to the multiscale basis
functions Nui and Npi , i = 1, . . . 4.
The prescribed boundaries ū and p̄ can influence the shape of the basis and are therefore critical to the
accuracy of the method. Assuming the barycentric property Nu,p i (xj ) = δij , where xj , j = 1, . . . 4 denote
coarse-nodal coordinates, these prescribed kinematical constraints are broadly classified into (a) linear and
(b) periodic boundary conditions.
Linear boundary conditions often result in coarse elements that over-estimate the stiffness of the underlying
micro-structure particularly in the case of highly oscillatory material coefficients. Periodic boundary con-
ditions relax this restriction although necessitating the periodicity of the underlying mesh hence restricting

s
their applicability to periodic domains. Such an approach is not valid when encountering randomly defined

20 ng
material distributions or dealing with more flexible polygonal/polyhedral RVE definitions.
Oscillating boundary conditions have been shown to alleviate the aforementioned issues [19, 20]. In these,
a reduced version of the governing Eqs. 6 and 7 is solved over the boundary under consideration hence ac-

20 di
counting for the heterogeneities along the boundary. Furthermore, it can be easily shown that linear bound-
aries arise naturally as a limit case when dealing with homogeneous material definition along the boundary.

e
In the case where, the coarse and fine length scales approach each other, resonance errors start to propagate
SD ce
[20]. This problem is alleviated by evaluating the basis over a larger domain that encapsulates ΩM(α) ; this
approach is known as an oversampling strategy.
o
The discretized matrix forms of Eqs. 6 and 7 are now expressed:
(
-U r
K̃αm (ωk )uαm = {∅} , on KM(α)
A yp

, I = 1 . . . nM , J = 1, . . . ndim , k = 1 . . . nfreq (8)

uS = ūIJ , on ∂KM(α)

(
H̃αm (ωk )pαm = {∅} , on KM(α)
M ar

, I = 1 . . . nM , J = 1, . . . ndim , k = 1 . . . nfreq (9)

pS = p̄IJ , on ∂KM(α)
where ndim = 2 or 3, depending on the dimension of the problem and nfreq denotes the number of frequency
IS in

steps. The multiscale basis functions are iteratively evaluated for each frequency.
im

2.2 Upscaling procedure

 T
The RVE specific vectors of nodal displacements uαm = umx , umy and pressures pαm are associated
el

with the corresponding coarse-element field variables through the following Eqs.:
Pr

uαm(i) = Num(i) uM(α) (10a)

p
pαm(i) = Nm(i) pM(α) , (10b)

where uαm(i) and pαm(i) denote the displacement and pressure vectors for the ith fine-element in the αth
p
element. The arrays Num(i) and Nm(i) represent the multiscale basis functions mapping the αth coarse-element
nodal displacements uM(α) and pressures pM(α) to the fine-scale, respectively.
Collecting the contributions from each fine-element, Eq. (10) can be expressed over the entire RVE:

uαm = Num uM(α) (11a)

pαm = Npm pM(α) , (11b)
V IBRO - ACOUSTIC MODELLING AND PREDICTION 7

Using Eq. 3, the governing equations at each frequency ωk are expressed for each micro-element as
" el,α #( ) " u el,α #
K̃m(i) (ωk ) − ωk2 M̃el,α (ω k ) − C̃ el,α
(ω k ) uα f
2 el,α
m(i)
T el,α
m(i)
2 el,α
m(i)
α = m(i)
p el,α (12)
−ωk C̃ (ωk ) H̃ (ωk ) − ωk Q̃ (ωk ) pm(i) f
m(i) m(i) m(i) m(i)

Substituting the micro to macro mapping relations from Eq. (10) into Eq. (12), and multiplying the first
row-set of equations by Nu T and second row-set of equations by Np T , the following equation is obtained:
" #  " u el #
K̃el
M (α),m(i) − ω 2 M̃el
k M (α),m(i) − C̃ el
M (α),m(i) uM (α),m(i)
fM (α),m(i)
= p el , (13)
−ωk2 C̃el T H̃el − ω 2 Q̃el p fM (α),m(i)

s
M (α),m(i) M (α),m(i) k M (α),m(i) M (α),m(i)

20 ng
where K̃el el el el el
M(α),m(i) , M̃M(α),m(i) , H̃M(α),m(i) , Q̃M(α),m(i) , and C̃M(α),m(i) correspond to the fine-element state
matrices mapped onto the coarse element nodes and assume the following form

20 di
el,α u
K̃el uT
M(α),m(i) = Nm(i) K̃m(i) Nm(i) (14a)
el,α u
M̃el uT
M(α),m(i) = Nm(i) M̃m(i) Nm(i) (14b)

e
pT el,α p
H̃el
SD ce
o M(α),m(i) = Nm(i) H̃m(i) Nm(i)

Q̃el
pT el,α
M(α),m(i) = Nm(i) Q̃m(i) Nm(i)
p
(14c)
(14d)
el,α p
C̃el uT
M(α),m(i) = Nm(i) C̃m(i) Nm(i) (14e)
-U r
In Eqs. (13) and (14) the dependence of the state matrices on (ωk ) is omitted for brevity. Similarly, the
A yp

forcing terms assume the following form

u el T u el,α
fM(α),m(i) = Num(i) fm(i) (15)
M ar

p el pT p el,α
fM(α),m(i) = Nm(i) fm(i) (16)
IS in

for the nodal forces and outflows, respectively.

On the basis of the principle of energy equivalence between the coarse element and its fine-scale discretiza-
im

tion, the following relations between the micro-macro transition state matrices and vectors and their local
el
Pr
8 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

coarse element representations are established:

nmel
X
K̃el
M(α) = K̃el
M(α),m(i) (17a)
i=1
nmel
X
M̃el
M(α) = M̃el
M(α),m(i) (17b)
i=1
nmel
X
H̃el
M(α) = H̃el
M(α),m(i) (17c)

s
i=1
nmel
X

20 ng
Q̃el
M(α) = Q̃el
M(α),m(i) (17d)
i=1
nmel
X

20 di
C̃el
M(α) = C̃el
M(α),m(i) (17e)
i=1

e
nmel
X
u el u el
fM(α) = fM(α),m(i) (17f)
SD ce
o
p el
fM(α) =
i=1
nmel
X p el
fM(α),m(i) (17g)
i=1
-U r
The local coarse element state matrices and vectors shown in Eqs. (17) can be assembled over the coarse
A yp

domain using standard assembly operations, i.e.,

nMel
KM = A Kel
M(α) , (18a)
M ar

α=1
nMel
MM = A Mel
M(α) , (18b)
IS in

α=1
nMel
HM = A Hel
M(α) , (18c)
α=1
im

nMel
QM = A Qel
M(α) , (18d)
α=1
nMel
el

CM = A Cel
M(α) , (18e)
α=1
nMel
Pr

u u,el
fM = A fM(α) , (18f)
α=1
nMel
p p,el
fM = A fM(α) . (18g)
α=1

Hence, the upscaled global governing equations assume the following form:
    u 
K̃M − ωk2 M̃M −C̃M uM f
2 T 2 = pM (19)
−ωk C̃M H̃M − ωk Q̃M pM fM ,
| {z }| {z } | {z }
Z̃M XM FM

where the unknown field vectors uM and pM denote the coarse-nodal displacements and pressures, respec-
tively.
V IBRO - ACOUSTIC MODELLING AND PREDICTION 9

2.3 Solution procedure at the coarse scale

The coarse-scale governing equations Eq. (19) are solved repeatedly over each frequency step.

XM (ωk ) = Z̃M (ωk )−1 FM , for k = 1 : nfreq (20)

The complex impedance at each frequency can now be computed at the incident surface:

p|Γi
Z̃n (ωk ) = − , (21)
vn |Γi

s
where p|Γi and vn |Γi denote the average pressure and average normal fluid velocity at the incident surface

20 ng
Γi , respectively. The average normal fluid velocity is computed as follows:

vn |Γi = jωk Un , (22a)

P

20 di
Fg |Γi
Un = , (22b)
Γi

e
Fg = Z̃M (ωk )XM (ωk )/ωk2 , (22c)

SD ce
where Zair represents the impedance of air. The sound absorption coefficient α(ωk ) is finally computed in
terms of the complex frequency dependent impedance Z̃n (ωk ) as follows:
o
 
 Z̃ (ω ) − Z 2
-U r
 n k air 
α(ωk ) = 1 −   . (23)
A yp

 Z̃n (ωk ) + Zair 

2.4 Downscaling
M ar

While the SAC is computed at the coarse scale in Eq. (23). It is also possible to compute it from the fine-scale
information, to offer a more detailed account of the underlying heterogeneities.
IS in

The fine-scale displacements and pressures at each time-step can be evaluated from the solution of Eq. (20) by
employing the following down-scaling procedure. The coarse element-wise displacements and pressures are
im

first extracted at the desired frequency steps from uM (ωk ) and pM (ωk ), respectively. These values are now
stored in the vectors of coarse-nodal displacements uM(α) and pressures pM(α) , α = 1 . . . nMel , respectively.
The displacements and pressures associated with the ith fine-element in the αth coarse-element / RVE can
el

be evaluated using Eq. (10). The ith fine-element Fg information is also recovered with the same mapping.
Now the SAC can be computed from all the fine-scale information.
Pr

3 Numerical Examples
Three benchmarks have been examined to investigate the validity of the proposed multiscaling solution
procedure for a) equivalent fluid rigid motionless skeleton models and b) elastically deformable Biot solid
skeleton models, when subjected to plane wave normal incidence acoustical excitation. Melamine foam
and non-dissipative air are used in all cases. The macroscopic material parameters of melamine foam are
summarized in Table 2. The Sound Absorption Coefficient (SAC) of these configurations are computed
through the MsFEM over the frequency range 20Hz ≤ f ≤ 5500Hz. These results are compared to the
corresponding TMM computations.
The efficiency of the MsFEM is measured by comparing its performance against the corresponding FEM.
The FEM operates over the associated global fine mesh. Computational times taken to (A) evaluate the
multiscale basis functions and (B) perform the solution procedure are recorded and averaged over three runs.
Assembly and upscaling of state matrices are included within (A). Similarly, the downscaling operations are
included within (B).
10 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

Table 2: Melamine foam macroscopic parameters

Parameter Name Value

static air flow
σ 104 N · s · m−4
resistivity
φ open porosity 0.99
high frequency limit of
α∞ 1.01
dynamic tortuosity
viscous characteristic
Λ 9.8 × 10−5 m
length

s
thermal characteristic

20 ng
Λ0 1.96 × 10−4 m
length
static thermal
k00 4.75 × 10−9 m2
permeability

20 di
E Young’s modulus 1.6 × 105 N · m2
ν Poisson’s ration 0.44

e
structural damping
ηs 0.1
SD ce
o ρ1
coefficient,(loss factor)
mass-density 8 kg · m−3

The acceleration provided by the MsFEM is demonstrated through computing speedup:

-U r
A yp

tFEM
speedup = , (24)
tMsFEM
where tFEM and tMsFEM denote the total times taken by each method, respectively.
M ar

3.1 Foam layer with rigid backing

IS in

This example is intended to verify the accuracy of the proposed method against the equivalent fluid and Biot
im

models. An 80 mm thick foam layer is placed inside an impedance tube of side 80 mm. The configuration
is subjected to sliding boundaries on the sides and a rigid wall backing at the rear. The 2-D geometry is
discretized with [8 × 8] coarse quadrilateral elements with each coarse element clustering an underlying
[2 × 2] fine quadrilateral mesh. The 3-D geometry is similarly discretized with [8 × 8 × 8] coarse hexahedral
el

elements with each coarse element clustering an underlying [2×2×2] fine hexahedral mesh. The coarse-scale
discretizations are illustrated in Fig. 4 .
Pr

The sound absorption coefficients for the rigid skeleton and elastic skeleton assumptions as evaluated by the
MsFEM and the TMM are shown in Figs. 5a and 5b, respectively. Near exact correspondence between the
MsFEM and the TMM is observed for all cases.
The averaged computational times and speedup offered by the MsFEM over the FEM in the case of the
elastic skeleton assumption is provided in Table 3. A significant speedup of 15.66 is obtained for the 3-D
discretizations.

3.2 Foam layer with air backing

This example is provided to examine the accuracy of porous-fluid coupling. A 40 mm melamine foam layer
with a 40 mm air backing is placed in an impedance tube of side 80 mm. The sides are subjected to sliding
boundaries. Both a 2-D and a 3-D model are examined.
The 2-D geometry is discretized with [10 × 10] coarse quadrilateral elements with each coarse element
clustering an underlying [2 × 2] fine quadrilateral mesh. The 3-D geometry is similarly discretized with
V IBRO - ACOUSTIC MODELLING AND PREDICTION 11

s
20 ng
20 di
(a) 2-D mesh (b) 3-D mesh

Figure 4: Coarse-scale meshes (a) 8 × 8 quadrilateral coarse-elements with 2 × 2 quadrilateral fine-elements

e
per coarse-element and (b) 8 × 8 × 8 hexahedral coarse-elements with 2 × 2 × 2 hexahedral fine-elements
per coarse-element.
SD ce
o
-U r
A yp
M ar
IS in
im

(a) Rigid skeleton model (b) Elastic skeleton model

el

Figure 5: Sound absorption coefficients computed for a 80mm single melamine foam layer with rigid back-
ing, through 2-D and 3-D MsFEM models and TMM models.
Pr

Table 3: Computation time taken (in seconds) and speedup offered by the MsFEM over FEM for a elastic
skeleton description of a melamine foam with rigid backing

MS Basis Solving and

Total time Speedup
Computation times Downscaling times
FEM 0.20 16.82 17.01
2-D 3.19
MsFEM 0.87 4.45 5.32
FEM 4.08 803.67 807.73
3-D 15.66
MsFEM 8.93 42.66 51.59
12 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

s
20 ng
(a) 2-D mesh (b) 3-D mesh

20 di
Figure 6: Coarse-scale meshes (a) 10×10 quadrilateral coarse-elements with 2×2 quadrilateral fine-elements
per coarse-element and (b) 10 × 10 × 10 hexahedral coarse-elements with 2 × 2 × 2 hexahedral fine-elements

e
per coarse-element.
SD ce
o
-U r
A yp
M ar
IS in

(a) Rigid skeleton model (b) Elastic skeleton model

Figure 7: Sound absorption coefficients computed for a 40mm melamine foam layer with 40mm air backing
im

through 2-D and 3-D MsFEM models and TMM models.

el

[10 × 10 × 10] coarse hexahedral elements with each coarse element clustering an underlying [2 × 2 × 2] fine
hexahedral mesh. The coarse-scale discretizations are illustrated in Fig. 6. The sound absorption coefficients
Pr

for the rigid skeleton and the elastic skeleton assumptions as evaluated by MsFEM and TMM are shown in
Figs. 7a and 7b, respectively.
Once again, near exact correspondence between the MsFEM and TMM is observed in all cases. The averaged
computational times and speedup offered by the MsFEM over the FEM in the case of the elastic skeleton
assumption is provided in Table 4. A significant speedup of 19.04 is obtained for the 3-D discretizations.

3.3 Melamine foam layer with meso-scale perforation

This example is provided to illustrate the ability of the MsFEM to adequately handle mesoscale inclusions.
An 80 mm foam layer with a square perforation of side 40 mm is placed in an impedance tube of side
80mm and length 400mm. The configuration is subjected to rigid wall backing and sliding boundaries on
the sides. In this case, to account for reflection, scattering, dispersion and edge effects introduced by the
mesoscale perforation, the entire impedance tube geometry is modelled with impedance mismatch values
being computed at the extreme left end.
The 2-D geometry is discretized with [40×8] coarse quadrilateral elements with each coarse element cluster-
V IBRO - ACOUSTIC MODELLING AND PREDICTION 13

Table 4: Computation time taken (in seconds) and speedup offered by the MsFEM over FEM for a elastic
skeleton description of a melamine foam with air backing

MS Basis Solving and

Total time Speedup
Computation times Downscaling times
FEM 0.35 16.53 16.89
2-D 2.34
MsFEM 1.82 5.40 7.23
FEM 14.46 1592.07 1606.52
3-D 19.04
MsFEM 27.46 56.89 84.36

s
20 ng
20 di
e
SD ce
o
(a) 2-D mesh (b) 3-D mesh
-U r
Figure 8: Coarse-scale meshes (a) 40×8 quadrilateral coarse-elements with 2×2 quadrilateral fine-elements
A yp

per coarse-element and (b) 40 × 8 × 8 hexahedral coarse-elements with 2 × 2 × 2 hexahedral fine-elements

per coarse-element.
M ar

ing an underlying [2 × 2] fine quadrilateral mesh. The 3-D geometry is similarly discretized with [40 × 8 × 8]
coarse hexahedral elements with each coarse element clustering an underlying [2 × 2 × 2] fine hexahedral
IS in

mesh. The coarse-scale discretizations are illustrated in Fig. 8 .

The sound absorption coefficients for the rigid skeleton and elastic skeleton assumptions as evaluated by
im

MsFEM and TMM are shown in Figs. 9a and 9b, respectively.

The 3-D MsFEM computations for both the rigid and elastic skeleton models agree with the TMM porous
composite model [15] , accounting for dynamic radiation correction and neglecting the pressure diffusion
el

effect. The SAC at the quarter wavelength resonance for the elastic model is slightly overestimated in relation
to the TMM model.
Pr

The 2-D MsFEM computations are compared against the TMM mixing law (ML) [15]. While the trend
is replicated correctly, the MsFEM is found to slightly overestimate the SAC. This can be attributed to the
differences in the cross-sectional geometry. The 3-D MsFEM and TMM result can be recovered from the
402
ML procedure when the mesoscale porosity is computed for a 3-D cross-section, i.e., φm = 80 2 = 0.25
. However, to facilitate ML comparisons with the 2-D MsFEM model, a 2-D cross-section is considered,
resulting in a mesoscale porosity φm = 40 80 = 0.5. The comparison proves meaningful only in the case of
low resistive materials, like Melamine foam.
The averaged computational times and speedup offered by the MsFEM over the FEM in the case of the
elastic skeleton assumption is provided in Table 5. The FEM procedure in case of 3-D discretizations in
the elastic skeleton assumption, proves very expensive, as illustrated by the total times. The MsFEM proves
invaluable in accelerating computations here, offering a speedup of 28.94.
14 PRELIMINARY PROCEEDINGS OF ISMA2020 AND USD2020

s
(a) Rigid skeleton model (b) Elastic skeleton model

20 ng
Figure 9: Sound absorption coefficients computed for a 80mm melamine foam layer with a square mesoscale
air perforation of side 40mm through 2-D and 3-D MsFEM models and TMM models.

20 di
Table 5: Computation time taken (in seconds) and speedup offered by the MsFEM over FEM for a elastic
skeleton description of a melamine foam with mesoscale perforation

e
SD ce
o MS Basis
Computation times
Solving and
Downscaling times
Total time Speedup

FEM 1.29 51.81 53.10

2-D 3.22
MsFEM 5.20 13.96 19.16
-U r
FEM 28.28 14,118.37 14,146.65
A yp

3-D 28.94
MsFEM 106.88 381.89 488.77

4 Concluding remarks
M ar

We discuss the particulars of the application of the Multiscale Finite Element Method (MsFEM) to vibro-
IS in

acoustic problems in porous media. This method is shown to significantly drive down computational costs
while retaining the fidelity of the standard FEM. Three benchmark examples are examined to establish the
im

validity of the proposed methodology in its application to coupled multilayered porous segments and porous
composite materials. Transmission loss problems as well as the case of oblique incidence and diffused field
excitations are further extensions to be investigated.
el

Acknowledgements
Pr

This work has been carried out under the auspices of the grant: “European industrial doctorate for advanced,
lightweight and silent, multi-functional composite structures—N2N.” The N2N project is funded under the
European Union’s Horizon 2020 Research and Innovation Programme under the Marie Skłodowska-Curie
Actions Grant: 765472.

References
[1] R. Fitzgerald, “Helmholtz equation as an initial value problem with application to acoustic propaga-
tion,” The Journal of the Acoustical Society of America, vol. 57, no. 4, pp. 839–842, 1975.

[2] N. Atalla, R. Panneton, and P. Debergue, “A mixed displacement-pressure formulation for poroelastic
materials,” Journal of Acoustical Society of America, vol. 104, pp. 1444–1452, 1998.
V IBRO - ACOUSTIC MODELLING AND PREDICTION 15

[3] W.-Y. Jang, W.-Y. Hsieh, C.-C. Miao, and Y.-C. Yen, “Microstructure and mechanical properties of
alporas closed-cell aluminium foam,” Materials Characterization, vol. 107, pp. 228 – 238, sep 2015.
[Online]. Available: http://www.sciencedirect.com/science/article/pii/S1044580315002594

[4] M. A. Biot, “Theory of propagation of elastic waves in a fluid-saturated porous solid. i low-frequency
range,” Journal of the Acoustical Society of America, vol. 28, pp. 168—-178, 1956.

[5] M. A. Biot, “Theory of propagation of elastic waves in a fluid-saturated porous solid. ii higher-
frequency range,” Journal of the Acoustical Society of America, vol. 28, p. 179–191, 1956.

[6] N. Atalla, M. A. Hamdi, and R. Panneton, “Enhanced weak integral formulation for the mixed (u,p)

s
poroelastic equations,” Journal of Acoustical Society of America, vol. 109, pp. 3065–3068, 2001.

20 ng
[7] F.-X. Bécot and L. Jaouen, “An alternative biot’s formulation for dissipative porous media with skeleton
deformation,” The Journal of the Acoustical Society of America, vol. 134, pp. 4801–4807, 2013.

20 di
[8] K. J. Johnson D. L. and D. R., “Theory of dynamic permeability and tortuosity in fluid-saturated porous
media,” J. Fluid Mech., vol. 176, pp. 379–402, 1987.

e
[9] Y. Champoux and J.-F. Allard, “Dynamic tortuosity and bulk modulus in air-saturated porous media,”
SD ce
Journal of applied physics, vol. 70, no. 4, pp. 1975–1979, 1991.

[10] A. J.-F. Lafarge D., Lemarinier P. and T. V., “Dynamic compressibility of air in porous structures at
o
audible frequencies,” J. Acoust. Soc. Am., vol. 102(4), pp. 1995–2006, 1997.
-U r
[11] Y. Miki, “Acoustical properties of porous materials-modifications of delany-bazley models,” Journal of
A yp

the Acoustical Society of Japan (E), vol. 11, no. 1, pp. 19–24, 1990.

[12] M. Delany and E. Bazley, “Acoustical properties of fibrous absorbent materials,” Applied acoustics=
Acoustique appliqu = Angewandte Akustik, vol. 3, no. 2, pp. 105–116, 1970.
M ar

[13] J. Allard and N. Atalla, Propagation of sound in porous media: modelling sound absorbing materials
2e. John Wiley & Sons, 2009.
IS in

[14] F.-X. Bécot, L. Jaouen, and E. Gourdon, “Applications of the dual porosity theory to irregularly shaped
porous materials,” Acta acustica united with Acustica, vol. 94, no. 5, pp. 715–724, 2008.
im

[15] F. Chevillotte, L. Jaouen, and F.-X. Bécot, “On the modeling of visco-thermal dissipations in heteroge-
neous porous media,” The Journal of the Acoustical Society of America, vol. 138, no. 6, pp. 3922–3929,
2015.
el

[16] J. S. Lee, Y. Y. Kim, J. S. Kim, and Y. J. Kang, “Two-dimensional poroelastic acoustical foam shape
Pr

design for absorption coefficient maximization by topology optimization method,” The Journal of the
Acoustical Society of America, vol. 123, no. 4, pp. 2094–2106, 2008.

[17] S. Chen, W. Zhu, and Y. Cheng, “Multi-objective optimization of acoustic performances of

polyurethane foam composites,” Polymers, vol. 10, no. 7, p. 788, 2018.

[18] H. Zhang, Z. Fu, and J. Wu, “Coupling multiscale finite element method for consolidation analysis
of heterogeneous saturated porous media,” Advances in water resources, vol. 32, no. 2, pp. 268–279,
2009.

[19] M. Buck, O. Iliev, and H. Andrä, “Multiscale finite elements for linear elasticity: oscillatory boundary
conditions,” in Domain Decomposition Methods in Science and Engineering XXI. Springer, 2014, pp.
237–245.

[20] H.-W. Zhang, J.-K. Wu, J. Lü, and Z.-D. Fu, “Extended multiscale finite element method for mechanical
analysis of heterogeneous materials,” Acta mechanica sinica, vol. 26, no. 6, pp. 899–920, 2010.