Trita Ave - 2013 04

Statistical energy analysis and variational principles
for the prediction of sound transmission in

multilayered structures
Mathias Barbagallo
Doctoral Thesis
Stockholm, Sweden 2013

The Marcus Wallenberg Laboratory for Sound and Vibration Research
Department of Aeronautical and Vehicle Engineering
Postal address
Visiting address
Royal Institute of Technology Teknikringen 8
MWL/AVE
Stockholm
SE-100 44 Stockholm
Sweden
Contact
Email: matbar@kth.se
Akademisk avhandling som med tillst

and av Kungliga Tekniska Hogskolan i Stockholm
framlaggs till offentlig granskning for avlaggande av teknologie doktorsexamen tisdagen
den 5:e mars 2013, kl 13:15 i sal F3, Lindstedtsvagen 26, Kungliga Tekniska Hogskolan,
Stockholm.
TRITA-AVE-2013:04
ISSN-1651-7660
ISBN-978-91-7501-648-1
c Mathias Barbagallo, 2013
!
ii
Abstract
Multilayered structures have many application in industry and society: they have
peculiar properties and serve a variety of purposes, like structural support, thermal
insulation, vibrational and acoustic isolation. This thesis concerns the prediction of
sound transmission in multilayered structures. Two problems are herein investigated:
the transmission of energy through structures and the transmission of energy along
structures. The focus of the analysis is on the mid to high frequency range. To predict
sound transmission in these structures, statistical energy analysis (SEA) is used.
SEA models are devised for the prediction of the sound reduction index for two kinds
of multilayered structures, double-walls used in buildings and trim-panels in vehicles;
the double-walls comprise an air cavity in between flat plasterboard or glass plates,
whereas the trim-panels a porous layer in between curved aluminium and rubber layers.
The SEA models are based upon the wave-types carrying energy. The novelty in
these SEAs is an element describing the waves in the air cavity, or in the porous
layer, fully coupled to the mass-impeded external layers. Compared to measurements,
the proposed SEA performs well: for double-walls, it performs better than previous
models; for trim-panels, it is an original result. The parameters of the new SEA
element, such as modal density, are derived from the coupling equations describing the
fully coupled waves. For double-walls, these equations are derived via Newtons laws.
For trim-panels, a variational approach based upon a modified Hamiltons principle
valid for non-conservative systems is preferred, because it is a powerful machinery
for deriving equations of motion and coupling conditions of a medium as complex as
the porous layer. The modified Hamiltons principle for non-conservative systems is
based upon a self-adjoint functional analogous to the Lagrangian, inspired by Morse
and Feshbachs construction. A self-adjoint variational principle for Biots equations in
the displacement formulation is devised. An equivalent mixed formulation is obtained
changing the coordinates of the displacement formulation via Lagrange multipliers.
From this mixed formulation, the Lagrangian for a porous material with a limp frame
is derived, which yields the continuity of the total displacement of the porous layer.
Lagrange multipliers help to obtain the correct coupling functionals between a porous
material and a solid. The Lagrange multipliers introducing the continuity of the frame
and the solid displacements equal the traction of the in-vacuo frame, thus disappearing
if the latter is limp. Measurements to gather material parameters for a Biot model of
the porous layer have been conducted.
The effects of spatial energy decay in the transmission along structures predicted by
SEA is studied: a major effect is the increased relevance of indirect coupling loss factors
between SEA elements. This may jeopardize the usefulness of SEA at higher frequencies.
vi
Papers contained in the Doctoral thesis

This thesis consists of the following papers:
1. S. Finnveden and M. Barbagallo. A cavity-wall element for the statistical energy
analysis of the sound transmission through double walls, to be submitted to Applied
Acoustics.
2. M. Barbagallo and S. Finnveden. A self-adjoint variational principle for anisotropic
viscoelastic Biots equations. International Journal of Engineering Science 2013;
63, pp 71-83.
3. M. Barbagallo and S. Finnveden. Statistical energy analysis of the sound
transmission through layered panels using a variational formulation of the porous
material, to be submitted.
4. S. Finnveden, N.-E. Horlin and M. Barbagallo. Characterization of the in vacuo
viscoelastic material properties of porous foams used in vehicles, submitted to
Journal of the Acoustical Society of America.
5. M. Barbagallo and S. Finnveden. Characterisation of a generic trim-panel: sound
reduction index and material parameters.
6. M. Barbagallo and S. Finnveden. Spatial energy decay and indirect couplings in
statistical energy analysis, Proceedings of ISMA 2010.
Division of work between the authors of the papers

1. Finnveden developed the theory and wrote the paper with Barbagallo.
2. Barbagallo performed the analysis and wrote the paper, supervised by Finnveden.
3. Barbagallo performed the analysis and wrote the paper, supervised by Finnveden;
Liu provided plots of the dispersion curves.
4. Barbagallo did the measurements with the first rig and participated in the setting
up of the second rig, developed and used by Horlin; Barbagallo wrote a preliminary
Mid-mod report for the measurements; Finnveden wrote the paper and did the
inverse estimation.
5. Barbagallo did the measurements and wrote the report, supervised by Finnveden.
6. Barbagallo performed the analysis and wrote the paper, supervised by Finnveden.
vii
Acknowledgments
I started this PhD to learn new things: my most sincere gratitude goes to everybody
who helped me achieving this goal.
This work was supported by the Swedish Research Council (621-2005-5754) and the
European Commission (Mid-Mod, Grant Agreement No.: 218508): the funding is
acknowledged.
A huge thank you is for my supervisor Svante Finnveden: thank you for having guided
me in my PhD, answered my questions, introduced me to all the fascinating topics
covered in this thesis and in our chats; I feel very lucky to have had the chance to study
these topics and your support has been fundamental. Thank you Mats
Abom, your
support in the last part of my project has been precious. Thank you to all the staff in
MWL, I learnt many things from you, during courses, meetings, setting up experiments
or simply just talking; I would like to thank especially Nils-Erik Horlin, Ulf Carlsson
and Danilo Prelevic. Thanks also to the people I met within Mid-Mod.
Thanks to all past and present PhD students in MWL for the nice discussions, for the
help you gave me, for your presentations about your works, for the fun during MWL
trips, for the beers together, for the exhausting innebandy matches. In particular, I
would like to thank Christophe, Eskil and Karl: I met you on my first day at MWL and
you have been a constant help for all sort of things during all these years; thank you
Hao for the discussions and the work together.
Thank you my mom and dad because you have always supported me in my studies and
because I knew I could count on you. Thank you Franci for all the love and for having
shared these years with me.
Contents
I
Overview and Summary
1 Overview
1.1 Transmission through multilayered structures
1.2 The variational approach . . . . . . . . . . . .
1.3 Measurements . . . . . . . . . . . . . . . . . .
1.4 Transmission along long structures . . . . . .
1.5 Organisation of the thesis . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
4
8
9
9
9
2 Statistical energy analysis

11
2.1 The rationale of SEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 SEA and indirect couplings . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Variational principles
3.1 Premise . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Hamiltons principle . . . . . . . . . . . . . . . . . . . .
3.2.1 Two interrogatives on Hamiltons principles . . .
3.2.2 Self-adjoint operators . . . . . . . . . . . . . . . .
3.3 Fluid formulations . . . . . . . . . . . . . . . . . . . . .
3.4 A self-adjoint variational principle for Biots equations .
3.4.1 Couplings to other media . . . . . . . . . . . . . .
3.4.2 Equivalent fluid theory . . . . . . . . . . . . . . .
3.5 Variational principle for a trim panel . . . . . . . . . . .
3.5.1 The trim-coupled waves . . . . . . . . . . . . . .
3.5.2 Comment on the variational statements . . . . . .
3.6 Wave form and modal density of the trim-coupled waves
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
15
15
16
18
18
19
21
23
24
25
27
28
29
4 SEA of a trim panel

31
4.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 Conclusions
35
Bibliography
36
II
41
Appended Papers
xi
Part I
Overview and Summary
Chapter 1
Overview
Multilayered structures have many applications in industry and society. The fuselage
of an airplane, the floor of a truck, the door of a car, the wall of a building are
examples of such structures. These structures have peculiar properties and serve a
variety of purposes, such as structural support, thermal insulation, vibrational and
acoustic isolation. The characterisation of their properties is an on-going challenge for
engineers: the physics of these structures is usually complex to understand and to model.
As their name suggests, these structures are made of layers of different materials. The
combination of the properties of each layer and the way they interact with each other
define what a multilayered structures is useful for. How the interaction between layers
functions and which consequences this interaction has on the wished use of the panel
are the interrogatives motivating and driving a lot of research, including this thesis.
Common materials used as layers are wood, chipboard, plaster, rubber, metal, cloth,
porous materials, mineral wools, honeycomb structures. Some of the layers may often
be shaped with particular geometries, such as curved plates and extruded metal profiles.
Porous materials often serve as layers because of their appealing properties: they are
light, they dissipate well vibrations and they are thermal isolators.
This thesis is concerned with sound transmission in multilayered structures employed in
buildings and vehicles. Two kinds of multilayered structures are studied: double-walls
used in buildings and trim-panels used in vehicles. The studied double-walls comprise
flat plasterboards separated by an air cavity and a double-glass window, while the trimpanels a curved aluminium layer and a rubber layer separated by a layer of porous
material.
Two kinds of problems are investigated: first, the transmission of energy through the
structures - e.g., the airborne sound insulation capacity of a double-wall; second, the
transmission of energy along long structures - e.g., the propagation of waves along the
fuselage of an airplane.
The focus of the analysis is on the mid-frequency and high-frequency ranges. The
CHAPTER 1. OVERVIEW
distinction into high, mid or low frequency ranges may be explained in two fashions,
related to each other. First, if the wavelength of the acoustic disturbance is small
(large) compared to the dimension of the structure, one speaks of high (low) frequency.
Second, whether the statistics of the problem should be included or not in the analysis.
At low frequencies, nominally identical structures, e.g. the doors of a car from the same
production line, behave similarly, having similar vibroacoustic responses: statistics is
not necessary. On the contrary, at higher frequencies the responses change dramatically
and statistics must be included in the analysis. Thus, depending on frequency, a
deterministic or a statistical approach should be preferred and, consequently, different
mathematical methods may be employed. For instance, standard finite elements can be
used at low frequencies for a deterministic analysis, while energy methods can be used
at high frequencies for a statistical analysis. The mid-frequency range is particularly
challenging because it is too high for deterministic approaches and too low for statistical
ones; as a number of recently funded EU-projects shows, the research interest in this
frequency range is high. Most of the work in this thesis has, in fact, been funded by one
of these projects, Mid-Mod.
The title of the thesis is briefly explained. Statistical energy analysis is perhaps the most
common high frequency and energetic approach and is the principal method of choice
in this work. Variational principles enter in the work because they are the theoretical
foundation of part of the analysis: a modified Hamiltons principle also valid for nonconservative systems is used to study the trim-panel.
1.1
Transmission through multilayered structures
Multilayered structures may be employed to acoustically isolate a target environment

from a sound source. Think of a wooden double-wall separating a bedroom and a room
with a stereo playing. The power emitted by the stereo impinges on the external layer
of the structure: some power is reflected and some is transferred to the other layers.
The layers are typically constituted by different materials with different geometries.
The energy flows through them via a number of channels. The energy transported
by these channels may be structure-borne or air-borne, may be due to resonant or
non-resonant motion, may exist at low frequencies or at high frequencies only. The
resonant-energy channels are associated to different types of propagating waves. These
wave-types depend on the excitation, on the layers and on their boundary and coupling
conditions. Once the energy has flown through the whole structure, it is radiated into
the target environment. To achieve a good sound insulation, most of the energy should
be dissipated within the double-wall, reflected back into the room with the stereo and
little radiated into the bedroom. There is a lot of interesting physics happening here.
The airborne sound transmission of a multilayered structure can be quantified via the
sound reduction index (SRI). The SRI of a structure depends on its vibroacoustic
4
CHAPTER 1. OVERVIEW
234(+*56(' .!751
!"
!"
"
!"
%!
!"
!"
&'()*(+,- ./01
!"
Figure 1.1: Dispersion curves of a multilayered panel, described in Ref. [57]. Numerical
dispersion curves from waveguide finite element method [42] (black dots). Measured
SRI of a multilayered panel taken from Ref. [57] plotted over the dispersion curves (full
curve with squares). The frequencies at which new waves start propagating correspond
to changes in the SRI, as evidenced by the vertical dashed lines.
response, which in turn depends on the waves propagating within it: as illustrated in Fig.
1.1, the frequencies at which new waves cut-on, i.e. start propagating, correspond to
changes in the SRI. The wave motion of multilayered structures can be complex: each
layer may support many waves, the coupled substructures may support other waves,
waves may cut-on at specific frequencies. Propagating waves transport energy across
the structure and, eventually, the SRI depends on them. The understanding of this
energy flow is salient and fascinating. Papers 1 and 3 deal with this.
In the literature there are many works on techniques to predict the SRI of multilayered
structures: Paper 1 and 3 respectively give an overview for double-wall panels and
for multilayered panels containing a porous layer. It is wished that any mathematical
approach calculating the SRI of a structure has the following properties, besides giving
a correct prediction. First, it should provide insights into the physics of the structure.
Second, it should be able to identify and rate the contributions of the various energy
channels to the SRI. Third, the statistical variability of real-life structures should be
taken into account. Fourth, it should provide all this in a short calculation time.
Standard deterministic methods may not easily meet all four properties: they may
not help understanding the physics, unless post-processing routines are implemented;
the rating of the various contribution to the SRI is not immediate; their usefulness
is questioned by the statistical variation of, e.g., the properties of the layers and the
inexact knowledge of boundary and coupling conditions; their computational cost is
high, especially when frequency dependant parameters and porous materials are involved
[32]. On the contrary, statistical energy analysis (SEA) may, in principle, comfortably
meet all four properties.
5
CHAPTER 1. OVERVIEW
SEA is a statistical method for predicting the vibroacoustic response of built-up
structures at higher frequencies, intrinsically accounting for their statistical variation;
SEAs rationale and its underlaying assumptions are presented in Ref. [53]; criteria for
stating the applicability of SEA to complex structures are still object of researches, such
as Ref. [11]. SEA is based upon a set of elements describing the vibroacoustic response
of the structure and satisfying the statement of energy conservation; the statement is
expressed via a set of linear equations, which is solved for the energies of the elements
in a blink of an eye on any computer. The elements may correspond to wave-types,
rather than to substructures: this sometimes improves the performance of SEA, in
terms of the predicted result and of the first two properties above. The identification
of the elements governing the response is immediate. More details on SEA are given in
Chapter 2. For now it suffices to say that very important parameters for SEA are the
wavenumbers of these wave-types. The relation between wavenumbers and frequencies is
termed dispersion relation; for a coupled structure an example is given by Fig. 1.1. The
importance of dispersion relations is twofold. First, their understanding may suggest a
set of SEA elements describing the vibroacoustic response. Second, the wavenumber of a
wave-type is related to its modal density, which is a parameter important for SEA. The
dispersion relations may be obtained in at least three ways. First, they may be found
in textbooks for uncoupled wave-types, such as bending waves in plates or longitudinal
waves in rods (e.g., Ch. 3 of Ref. [38]). Second, they may be calculated from numerical
deterministic methods, such as the waveguide finite element method (WFEM), Ref. [25].
Third, they may be obtained from analytical analyses, perhaps simplifying part of the
problem to obtain simple results (e.g., see Section 3.5.1). In this thesis, all these method
are employed (Papers 1 and 3).
The WFEM is a deterministic tool convenient for describing structures with a complex
cross-section and constant properties along the other direction(s). Its main outputs are
the dispersion relations of the propagating waves, see Fig. 1.1. In Papers 1 and 3,
the WFEM is used for diagnostic analyses of the dispersion curves of double-walls and
trim panels. Upon the understanding of their dispersion curves, the wave-types carrying
energy through the structure are identified, thus setting the basis for an SEA.
Despite its appealing features, the potentialities of SEA with multilayered structures
have not been fully explored. In this thesis, original SEAs for this class of structures
are derived. In Refs. [13, 14, 12] Craik and Smith use SEA for calculating the SRI of
double-walls with and without added absorption. The cavity with added absorption is
modelled as an air cavity with a higher damping loss factor; the damping loss factor
is estimated via a reverberation-time measurement, which, however, decays too quickly
An alternative to Craik and Smiths work for double-walls without added absorption is
shown in Fig. 1.2. The figure presents the set of SEA elements devised in Paper 1, for
the same structures as those used by Craik et al.. The novelty is the cavity-wall element,
modelling the waves in the air cavity fully coupled to the walls, when the motion of the
walls is mass-impeded only. In Paper 3, a similar approach is used for trim panels
CHAPTER 1. OVERVIEW
Figure 1.2: SEA for a double-wall filled with air, see Paper 1. The arrow define different
channels for energy flow: A, quasi-plane cavity-wall waves
free acoustic waves; B,
oblique cavity-wall waves
free acoustic waves; C, free acoustic waves
free acoustic
waves (mass law, non-resonant); D, free acoustic waves
wall bending waves; E, quasiplane cavity-wall waves
bendin waves; F, oblique cavity-wall waves
wall bending
waves; G, wall bending waves
wall bending waves.
comprising porous layers: SEA elements modelling the waves in the porous layer fully
coupled to the mass-impeded external layers are sought; these waves are named trimcoupled waves. The porous material is described by an equivalent fluid theory, derived
from Biot theory assuming a limp frame; the input parameters to these theories can
be measured, as shown in Ref. [37] and in Paper 5. The author thinks this is the first
SEA based upon either Biots or equivalent fluid theories for modelling porous layers in
multilayered structures.
An additional incentive for more research in this field comes from Ref. [36]: after
having compared many existing models for sound transmission in double-walls, Hongisto
concludes that the versatility of most of these models is very limited. Paper 1 is relevant
also in this regard: the proposed SEA performs well with three types of buildingconstruction double-walls, two aluminium double-walls and one double-glass window.
The SEA model in Fig. 1.2 needs parameters for the wave-types in each SEA element.
All elements but the two cavity-wall wave elements are standard SEA elements and the
expressions of their SEA parameters are known. The equations describing the cavity-wall
waves in double-walls are derived in Paper 1 via Newtons law and the calculation of the
SEA parameters follows. The equations describing the trim-coupled waves in the trim
panel are more difficult to derive than the cavity-wall waves, because a porous material
is generally more complex to model than air. The trim-coupled waves are described by
the coupling equations between the external layers and the porous layer; these equations
are derived in Paper 3 and Chapter 2. To derive them, a more powerful machinery than
Newtons law is preferred, due to the complexity of the involved media: Hamiltons
variational principle, a sound machinery to derive equations of motions and coupling
7
CHAPTER 1. OVERVIEW
conditions for complex structures (see, e.g., Ref. [51] on the usefulness of variational
principles for dealing with complex couplings).
1.2
The variational approach
In this thesis a variational approach is used to derive the equations describing the trimcoupled waves in the trim-panel. This is done within the framework of Hamiltons
principle: the Lagrangians of the various layers, and any constraint functional, are
summed: this sum is required to be stationary for the true coupled motion of the trimpanel. Hence, the equations of motion of the layers and their coupling equations follow.
Once the trim-coupled wave equations are obtained, their wavenumber, their modal
density and their wave form are calculated and used in an SEA for sound transmission
(Chapters 3, 4 and Paper 3). In this work the variational approach also helps: to obtain
equivalent formulations of a medium when its variable(s) of response is (are) changed; to
better understand complicated couplings, such as those of porous materials; to manage
coupling conditions when simplified theories are derived from more complex ones: e.g.,
the coupling conditions of a porous material, first described by the three Biots waves
then by one equivalent fluid wave.
The layers in the trim-panel are non-conservative, because they are constituted either
of viscoelastic materials, or of intrinsically lossy materials, like porous materials.
Hamiltons principle can, however, be used with conservative systems only; hence,
alternative solutions are needed to cope with the non-conservative layers of the trimpanel.
The literature offers some solutions [54, 47]; one particularly interesting is sometimes
called Morse and Feshbachs construction [48]. The original system under study is
considered together with its mirror system. The mirror system is in all aspects
identical to the original, except that it has negative losses. Intuitively, the losses
of the original system go to the mirror system, thus making the global system
conservative. Therefore, Hamiltons principle can be applied to the global system
and used to derive the equations of motion of the (non-conservative) original system
and its boundary/coupling conditions. Mathematically speaking, the mirror system is
the adjoint of the original and the global system is self-adjoint.
Morse and Feshbachs construction has already been used in vibroacoustics [30, 49, 20,
25] and this work continues this tradition. Before using the construction for the trimpanel, a self-adjoint variational principle for the porous layer described by viscoelastic
Biots equations is defined, since such principle was missing in the literature. The
principle is presented in Paper 2 and summarised in Chapter 3.
8
CHAPTER 1. OVERVIEW
1.3
Measurements
Various measurements on the trim-panel and its components are taken. This is a
necessary step to obtain data, both to input to numerical models and, then, to check
their performance, e.g. calculating the SRI. The measurements taken on the trimpanel are: 1) SRI of the trim-panel following ISO 15186 - 1:2000 (Paper 3); 2) static
flow resistivity of the porous materials following ISO 9053:991 (Paper 5); 3) frequency
dependant viscoelastic parameters of the in-vacuo frame of the porous materials (Paper
4); 4) static Youngs modulus and Poissons ratio of the frame of the porous material,
not in-vacuo (Paper 5); 5) shear modulus of the rubber plate (Paper 5).
The measurements follow ISO standards or published procedures [26]. The exception
is measurement 3, which presents original experimental rigs to estimate the viscoelastic
properties of porous materials, using an inverse estimate procedure. The reader can
refer to Papers 3, 4 and 5 for more details on these measurements: Part I of this thesis
does not specifically summarise them.
1.4
Transmission along long structures
The motivation of the work on SEA for the vibroacoustic transmission along structures
many wavelengths long comes from two observations in the literature, Refs. [22, 27].
These works contain problems that can be handled by SEA. Under certain conditions,
however, SEA fails where it is most expected to work: at higher frequencies. An
explanation may be the violation of one SEA hypothesis, i.e., the requirement that
the energy density in an SEA element remains constant. If an element is manywavelengths long, however, spatial energy decay may occur and this SEA hypothesis
is violated. Then, the neat form of the SEA equations stating energy conservation
has to be modified to account for the so-called indirect couplings between non-adjacent
elements. Consequently, the convenience and the applicability of SEA are questioned.
Paper 6 investigates this problem and Chapter 2 summarises some of its conclusions.
1.5
Organisation of the thesis
The next chapters summarise the most relevant results of this thesis. Chapter 2 presents
SEA, introducing its rationale and its founding hypotheses; it is related to Papers 1, 3
and 6. Chapter 3 introduces Hamiltons principle, illustrates a self-adjoint variational
principle for porous materials, and shows how to derive the equations of the trim-coupled
waves; it is related to Papers 2 and 3. The calculated modal density is used in Chapter
4 in the SEA of the whole trim-panel, which calculates its SRI; it is related to Paper 3
(and uses the data collected in the measurements in Papers 4 and 5). Chapter 5 draws
conclusions. Part 2 contains the appended papers.
9
CHAPTER 1. OVERVIEW
10
Chapter 2
Statistical energy analysis
In this chapter statistical energy analysis (SEA) is presented. The first section illustrates
its rationale and its underlying assumptions. The second section investigates the
consequences of energy decay in a structure, by studying the form of SEA equations and
its variables, relating them to power injection measurements and deterministic analyses.
2.1
The rationale of SEA
SEA is an approximate method to estimate steady state ensemble, space and frequency
band averages of vibroacoustic energy in built-up structures at somewhat higher
frequencies [53]. Today, there seems to be an accord, or at least a hope, in vehicle
industry and other parties interested in vibroacoustic predictions that SEA works, if
only the frequency is high enough - as is evidenced by the public homepages of the
ongoing EC-funded projects Mid-Mod and Mid-Frequency.
The governing equations in an SEA express the law of energy conservation for elements,
each describing in a frequency band one kind of response in a substructure of the whole
structure. The formulation relies on two postulates. First, the rate of energy dissipated
in an element is assumed proportional to the energy in that element. Second, the
rate of energy flow between two directly coupled elements is assumed proportional to
the difference of the elements modal power potentials, being the energies times the
expected frequency spacing between their natural frequencies [53]. This is the law of
coupling power proportionality (CPP). It is a very powerful potential flow model for
the conduction of vibroacoustic energy, similar to Fouriers law for heat conduction and
Ohms law for electrical conduction.
The constant of proportionality in the CPP law, the so-called conductivity, is often
calculated by a one way method, where the action of one element onto the other elements
is investigated. In doing so, field characteristics of the first element are defined, such
as the amplitudes of vibroacoustic waves impinging on a junction; the amplitude in a
diffuse field or the amplitudes of modes proper to the uncoupled element. Often the
11
CHAPTER 2. STATISTICAL ENERGY ANALYSIS

calculations are simplified by considering elements that are extended away from the
junctions towards infinity. The conductivity evaluated by a one way method describes
a potential flow and can be used in a proper SEA model only if it is a symmetric
function of the elements properties in such a way that it is equally evaluated when field
characteristics in the first and in the second element are prescribed. This symmetry
relation is termed SEA reciprocity [53] or consistency [45] relation. It is demonstrated
for numerous one way calculations. Similar to one way methods in nature, but not in
mathematical formulation, are perturbation methods [23, 41]
The first postulate is an established engineering approximation; however, the second
postulate, the CPP law, is useful only sometimes. To date it has been exactly
demonstrated for the rate of energy flow between two conservatively coupled, randomly
excited, oscillators [56] and, also, for the ensemble averaged energy flow between two
one-dimensional structures, with random properties, in the limit of vanishing spatial
damping decay within the elements, [19], see also [44]. In contrast, for three oscillators
[58] and for the ensemble averaged of three one-dimensional structures [19], the CPP
law is true for weak coupling only. For general structures, the range of usefulness of
the CPP law is not established, though, slowly appears a common understanding of the
criteria for a valuable SEA model. One example is provided in a article by Le Bot and
Cotoni [11], stating that, for each element, each junction and each frequency band, SEA
is valid if: i) The number of natural frequencies is large; ii) The modal overlap factor is
large (it is the ratio of resonances 3-dB bandwidth to their expected frequency spacing);
iii) The strength of coupling is small; iv) The spatial decay of energy in a wave travelling
across an element is small. The first three criteria are for many engineering structures
better fulfilled as frequency increases while the opposite is true for the fourth criterion.
In this thesis, the work related to the transmission through multilayered panels is
founded upon the identification of a good set of SEA elements, which catches well
the physics of the structure and which can be used to obtain a good prediction of the
SRI at mid-frequencies (Chapters 4 and Papers 1, 3); e.g, Fig. 1.2 shows the set of
elements used for double-walls, which is similar to that used for trim-panels except for
the cavity-wall waves element, substituted by the trim-coupled waves element. The work
related to transmission along long structures investigates what happens to SEA if the
fourth assumption fails (next section 2.2 and Paper 6).
2.2
SEA and indirect couplings
The fundamental equation in SEA expresses the law of energy conservation. This
law can also be expressed based on calculated or measured response using the power
injection method (PIM). Thus, for a given excitation, the input power is evaluated and
the energies of the elements are linearly related to this input power. The procedure
is repeated for excitation in all elements and an energy distribution (ED) model is
12

devised [34]. Then, when the ED model is inverted, an SEA-like model results.
Like SEA models, an SEA-like model relates the energy stored in the elements to
the input powers. Though, as stressed by Fredo in Ref. [28], the term SEA-like
signifies that the energy balance is applied for the individual and deterministic case,
not for the ensemble averaged energy balance for random structures, as SEA attempts.
Nevertheless, the dependant variables in an SEA-like and a proper SEA model are equal
when the frequency- spatial- ensemble-averaged responses of the single specimen and of
the ensemble are similar, which is most likely when the modal overlap factors are large,
see e.g. [55].
An SEA-like model is, with respect to the precision of the measurements or calculations
upon which it is based, an exact statement of energy conservation. Comparing the
equations defining such a model with those that would apply for a proper SEA model,
two major differences are apparent. First, the consistency relation is not fulfilled, though
there are evidences for that it approximately applies for higher frequencies, when there
are many resonances in the frequency bands and the elements modal overlap factors
are large. Specifically, the consistency relation is fulfilled for a homogeneous structure
if the frequency and space averaged point mobilities of the elements equal those for the
corresponding infinite elements [40] (and later, equally, [45]). The second difference is
that the rate of energy flow to an element is not governed by the CPP law. Thus, in
this mathematical clothing, there appears energy flows between elements that are not
physically connected; this artefact is termed indirect coupling [45, 46], or tunnelling
[35, 27, 17]. In the literature, there are different names for models that include indirect
couplings, e.g., advanced SEA [35], extended SEA [27] or quasi-SEA [45]. In some
cases such models might be useful. They cannot, however, be used in general predictive
SEA software, as it is not practical to calculate all the necessary coupling routines that
consider all the possible arrangements of connections in large structures.
The presence of indirect couplings does not, of course, describe an energy flow between
two elements that are not physically connected but signifies that the energy flow between
two elements depends on the response in other elements and, consequently, that CPP
does not properly describe the flow of energy in the system. In the literature tunnelling
is demonstrated to happen in the regime of strong coupling and small modal overlap
factors [19, 27]. In particular, if a structures response is defined by one global mode
indirect couplings cannot be neglected [45].
In Ref. [46] Mace investigates the importance of indirect couplings; he notes that for
increasing modal overlap factors, the indirect coupling loss factors asymptote to zero
and concludes that a proper SEA model can be formulated. The study in Paper 6
confirms that with increasing modal overlap factors the indirect coupling loss factors
decreases but so does also the direct coupling loss factors and the net effect is, in
fact, an increasing tunnelling. Consequently, proper SEA models are not useful, for
the investigated structures, at higher frequencies. Previously, Fredo found increasing
tunnelling in a three-plate structure with increasing frequency ([27], paper IV) and
13

Finnveden found a consistent overestimation of the energy flow in a three-pipe structure
at very high frequencies [22]; these observations are the basis for the study on the high
frequency limitation of standard SEA done in Paper 6.
In Paper 6, the vibrations of long structures with many elements in series are described
by SEA-like and proper SEA models. The vibrations considered in the deterministic
SEA-like model are excited by a large number of uncorrelated point forces, thus
simulating rain-on-the-roof excitation. Most of the calculations are made in the high
modal overlap regime where the frequency averaged response of an ensemble and an
individual structure should be similar. This surmise is supported by a few Monte Carlo
runs considering elements with random properties. The investigated structures are very
long compared to the wavelength and are well connected, meaning that a substantial
part of the energy coming into a junction is transmitted. The applications in mind
are structures such as ventilation ducts, pipelines and rib-stiffened shell structures such
as found in trains, aircraft and ships. Besides of being long and well connected these
systems often have piece wise constant properties along the structure and support a
numerable set of waves propagating in this direction. Therefore, each section is as a set
of parallel one-dimensional structures, being independent channels of energy flow. Not
to obscure the general picture and to facilitate extensive averaging based on Monte Carlo
simulations, simple one-dimensional structures consisting of rods in series are considered.
Thus, it is possible to highlight the findings on the effects of spatial energy decay on
tunnelling and the resulting failure of predictive standard SEA at high frequencies.
The investigations in Paper 6 are restricted to one-dimensional systems. Two
considerations, however, can be asserted; first, it is argued that two- and threedimensional systems with long mean free paths [11] might probably suffer the same
consequences of spatial energy decay; second, one-dimensional systems are actually
relevant: structures such ventilation ducts, pipelines, rib-stiffened panels that are
common in trains, ships and aircrafts, are long and show piece-wise constant properties
along one direction, thus supporting a numerable set of waves propagating in this
direction [22]; therefore, each section can be modelled as a set of parallel one-dimensional
structures, each being an independent channel of energy flow.
14
Chapter 3
Variational principles
Variational principles are a fascinating topic that has occupied a lot time in the writing
of this thesis. The first two sections of this chapter are dedicated to explain Hamiltons
principle to a reader not accustomed to it; they also explain the technique to adapt it to
non-conservative systems. The following three sections present the variational principles
defined in this thesis. The last section uses the equations of the trim-coupled waves,
derived from a simplified analysis based upon variational principles, to derive their wave
form and modal density. This chapter is related to Papers 2 and 3.
3.1
Premise
During the last football European championship, Italy won against England at penalties;
one of the finest ones was Pirlos chip penalty. Very few people would be able to score
like that, but many would be able to correctly guess which trajectory took the ball,
choosing between line A and line B in Fig. 3.1 (it is line A1 !). What matters now is that
line A intuitively illustrates a principle of nature, i.e. Hamiltons variational principle:
line A is the trajectory that minimises the difference of the potential and kinetic energies
of the ball, from the moment Pirlo kicks it to the moment it ends behind the back of the
English goalkeeper. On the contrary, line B does not minimise that quantity (a similar
and more formal example is in Ch. 19 of Ref. [18]).
1
Pirlo probably exploited the Magnus effect, neglected here; however, the ball used in EURO 2012
was presumably engineered to be less sensitive to it after the complaints in the 2010 World Cup.
Figure 3.1: Pirlo (left) beating the English goalkeeper. The trajectory of the ball is A.
15
CHAPTER 3. VARIATIONAL PRINCIPLES
3.2
Hamiltons principle
The motion of a body is described by its boundary value problem, i.e., by a partial
differential equation of motion in its domain and by conditions on its boundary; if time
is a variable, initial values should also be prescribed.
One way to get equations of motion is Newtons second law. This is often done in
text books, as in Cremer and Heckl for waves in solids an in Pierce for waves in air
(Ch. 3 in Ref. [38] and Ch. 1 in Ref. [50]). The advantage of this approach is that
for simple problems it is straightforward to apply with basic mathematic skills. The
limitations of this approach are the following. First, this approach does not directly
yield the boundary conditions of the medium, but only the equation in the domain,
thus leading to an incomplete description. Second, the dimensions of the domain do not
explicitly appear in the equation of motion. Third, if the motion of the medium is given
by various wave-types, or if various media are coupled, it may become cumbersome or
unfeasible. Fourth, it uses vectors, thus depending on the system of coordinates2 .
Another way to describe the motion of a body is Hamiltons variational principle. This
principle is based upon a scalar quantity called Lagrangian and upon the machinery of
calculus of variations. The Lagrangian is defined as the difference of strain and kinetic
energies; mathematically, it is a functional, i.e., a function of functions: energy is a
function of displacement, which itself is a function. The Lagrangian of a system is
expressed using variables of response at choice, e.g., particle displacement or pressure in
a fluid. Hamiltons principle requires that the integral between the times t1 and t2 of the
Lagrangian, i.e. the action, is stationary for the true motion of the system: the functions
for which the first variation of the action equals zero are sought or, equivalently, the
functions that make the action stationary are sought (in fact, it is not relevant whether
the action has a minimum, a maximum or a saddle, see e.g. Ch. 19 [18] or p 83 [59]).
Searching for these functions is the field of calculus of variation3 . The functions that
make the action stationary in its domain are the Euler-Lagrange equations, or equations
of motion. The functions that make the action stationary on in its boundary are the
boundary conditions, usually distinguished in essential and natural conditions. The
essential conditions involve the variable of response itself, the natural conditions its
higher spatial derivatives.
This variational approach is often used when the medium to be described is complicated.
For instance, the equations of motion of shells and of porous materials are derived with
this approach [4, 8]. A difficulty of this approach is that it requires more advanced
mathematical skills, i.e. calculus of variations. Another difficulty is due to an intrinsic
limitation of Hamiltons principle: it can only be applied to conservative systems, i.e.,
systems in which the forces can be obtained from scalar potentials: e.g., the stresses
2
Fifth, the author has always had a hard time using it!
The search for the functions making a functional stationary is conceptually similar to the search
for the point of stationarity of a function, i.e., to the point in which its first derivative equals zero.
3
16

in a body are derived from the strain energy that is a potential. The possibility to
derive a force from a scalar potential is granted by the self-adjointness of the operators
describing the system under study. Conservative systems are indeed described by selfadjoint operators, but non-conservative systems are not: e.g., the operator describing
viscoelastic systems is not self-adjoint.
However, there exist workarounds to use a variational approach with non-conservative
systems too. Internal losses, e.g. viscoelasticity, are dealt as follows: Hamiltons
principle is applied to the conservative elastic system, its equations of motion are derived
and then viscoelastic forces are directly inserted into the equations of motion. For elastic
and viscoelastic theories, this is called by Biot isomorphism [9, 10]: the elastic operators
are simply substituted by the viscoelastic ones. From Newtons laws point of view, this
is perfectly fine: mass times acceleration equals any force, conservative or not (see, e.g.,
p 19-7 Ch. 19 Ref. [18]). Viscous losses need a different workaround: a pseudo-potential
dissipation functional is used. This functional is originally introduced by Rayleigh [52]
and, among others, used by Biot to derive his equations for porous materials [8]. So,
Hamiltons principle is applied to the non-viscous conservative medium; then, the viscous
losses are added ad-hoc via the pseudo-potential (see Eqs. (15) in Paper 2). Notable
is Maximovs approach in Ref. [47]: considering interacting mechanical and heat fields,
he can establish energy conservation in dissipative media and, thus, define a generalised
variational principle; see also Ref. [3].
An even more interesting solution, at least from an engineering point of view, is Morse
and Feshbachs construction [48]. It is beautiful because it allows the use of the
machinery of Hamiltons principle also with non-conservative systems, by building a
self-adjoint functional analogue to the Lagrangian that can be the basis of a self-adjoint
variational principle. The functional is said to be an analogy to the Lagrangian because
a Lagrangian refers to a real system, related to nature, while the Morse and Feshbachs
construction is rather artificial. The functional constructed via this procedure is, in fact,
a bilinear functional of the generalised coordinates of the original system and its adjoint
system; the adjoint system is in all aspects identical to the original, except that it has
negative losses, i.e., operators adjoint to the original ones (more on this later). It is an
elegant procedure, but comes at the sacrifice of certain amount of reality in some of the
intermediate results. The final results are, in fact, realistic and useful: the first variation
of this constructed functional in respect to the adjoint coordinates yields the equations
making the original system stationary; hence, it is an analogy to the Lagrangian, but
for brevity, the term Lagrangian is herein used.
Self-adjoint variational principles that are stationary for true motion have appealing
properties. To quote Fung and Tong on p 313 in Ref. [29]: basic minimum principles
exist, which are among the most beautiful of theoretical physics. Ref. [31] on p
370 considers them as starting points for new formulations of mechanical structures.
They are excellent concepts for deriving equations of motions and natural coupling
conditions, see e.g. [8, 43, 15, 2, 25, 20, 51]. They provide the theoretical foundation to
17

derive simplified theories for: (visco)elastic media, e.g. condensing a three-dimensional
medium to two- or one-dimension (see p 45 of Ref. [33]); complex structures, e.g.
describing a specific kind of motion only, like the radial-axial vibrations of fluid-filled
pipes [21]. On p 313 in Ref. [29] it is added that the computational methods of solution
of variational problems are the most powerful tools for obtaining numerical results in
practical problems of engineering importance; e.g., variational principles constitute a
sound basis for new finite element formulations, as indicated by Refs. [25] and [59] in
pp. 76-83. Finally, in a variational formulation, the variables of response are chosen
at will and there are systematic methods for coordinate transformations; such powerful
methods are used in the present work, specifically in Papers 2, 3 and in the rest of this
chapter.
The next two subsections comment on two interrogatives on Hamiltons principle and
on the mathematical definition of adjoint operators.
3.2.1
Two interrogatives on Hamiltons principles
On p 43 of Ref. [7], Berdichevsky answers to two fascinating interrogatives4 raised by

Hertz: is Hamiltons principle supposing intentions in nature, i.e., to make a certain
functional stationary? And about the functional: does it have any physical meaning
at all? Berdichevsky provides neat explanations to both interrogatives on pp 43-44 of
Ref. [7] (see also Ch. 19 of Ref. [18]). The answer to the first question relies on
quantum mechanics, is a bit complicated and the reader may refer to the textbook. The
answer to the second question follows from relativity theory and is simpler to grasp: the
Lagrangian is the total energy in the Newtonian limit of the observers time frame, a
limit that is valid for velocities much lower that the speed of light. The Lagrangian has
a clear meaning, it is not a meaningless quantity that happens to be stationary.
3.2.2
Self-adjoint operators
The operator DA that is adjoint to D satisfies the following relation

v DA u dt,
u D v dt
(3.1)
for any function u and v. If D DA , then D is self-adjoint. For instance, if u and v

describe displacement, viscous losses have Dv d dt, which is an operator that is not
self-adjoint, i.e.,
u Dv v dt
v Dv u dt.
(3.2)
The field of variational principle may likely trigger in a student philosophical thoughts, which
typically have difficult answers, if any: the author was happy to find fairly understandable explanations
to two such interesting interrogatives, although beyond the scope of this thesis.
18

Its adjoint operator is DA
d dt, which characterises a system with negative losses,
as expected from the discussion above.
Self-adjoint operators are very important because they are related to three fundamental
properties in acoustics. First, the principle of reciprocity is demonstrated starting from
the self-adjointess of the operators: e.g., the reciprocity of the Greens functions is
based upon the assumption that the Greens function is self-adjoint (see the note on p
199 in Ref. [50]). Second, if an operator is self-adjoint, then it is symmetric. Third, if
an operator is symmetric, then the matrixes used in computer implementation can be
solved with efficient schemes.
3.3
Fluid formulations
In this section Hamiltons principle and the variational machinery are applied to a fluid.
First, the Lagrangian in the displacement formulation is defined; second, via a change of
coordinates the Lagrangian is expressed in the pressure formulation. This analysis and
the obtained results may facilitate the understanding of next section, where the same
machinery is applied to a porous material.
Time-harmonic motion at angular frequency is considered; accordingly, all variables
in this chapter are expressed in this form,
1
veit v e it ,
2
1 A it
v e
va e it ,
2
v
vA
(3.3a)
(3.3b)
where v is a variable of response, v is its complex amplitude, v is its complex conjugate

and the upper index a is introduced so that va
vA . Below, cartesian tensor notation
is used, a comma in the subscript indicates a partial derivation.
The strain and kinetic potential5 densities of a fluid described by its displacement are6
W
a,e e e
Ui,i
R Ui,i ,
Uia,e 2 e Uie ,
(3.4)
the superscript e indicates that these quantities describe a fluid: U e is the fluid
e is the fluid bulk modulus and e the fluid density. The Lagrangian is
displacement, R
defined as the difference between these two potentials:
L
dV.
(3.5)
V
5
They are termed potentials and not simply energies for the same reason that the Lagrangian is
in reality an analogy to the Lagrangian: they are artificial constructed bilinear functionals, from which
stresses and inertia forces can be derived.
6
Notice that only the original variables and the complex conjugate of the adjoint ones appear in W
and K: if a variational principle for Eq. (3.5) is true, so it is for the conjugate variables.
19

Its first variation is stationary for the true motion; varying the adjoint displacement,
the equations of motions and boundary integrals of the original system are obtained:
e U e
R
i,ii
B
2 e Uie
0;
(3.6a)
e U e dS
ni Uia,e R
i,i
ni Uia,e pe dS
0.
(3.6b)
The derivation of the pressure formulation is based upon the invariance of the Lagrangian
to coordinate transformations. In order to switch from the displacement coordinate to
the pressure coordinate, Lagrange multipliers are employed. The Lagrange multipliers
e and a,e introduce the fluid dilatation ee Ui,i as an additional variational parameter
e
to eliminate the fluid displacement from L . A new functional is defined:
Le
e
Ui,i
a,e ee
V
a,e
Ui,i
e ea,e
dV.
(3.7)
The original and adjoint fluid displacements and dilatations are varied independently,
obtaining the following relations:
e
e ee ;
R
a,e
2 e Uie ,
e,i
a,e
,i
e ea,e ;
R
2 e U a,e .
(3.8)
(3.9)
It follows that e equals minus the pressure of the fluid,

e
pe .
(3.10)
Using Eqs. (3.8)-(3.10) the fluid displacements, dilatations and Lagrange multipliers are
eliminated from Le ; integrating by parts all the terms featuring a second-order spatial
derivative, the functional Le is rewritten as
Le
We
K e dV
Fe
pa,e
where
Fe
ni pa,e
S
1 e
p
2
e ,i
1 e
p
Re
dS
S
pa,e
,i
ni pa,e
,i
1 e
p
2
e ,i
1
2 e
dV
pe dS,
F e.
(3.11)
(3.12)
where ni is the normal pointing outwards. Varying the adjoint pressure in Le , the
equations of motions and boundary integrals follow:
1 e
p
e
R
1
2 e
Be
S
pe,ii
ni pa,e
,i
0,
1
2 e
(3.13a)
pe dS
S
ni Uia,e pe dS
0.
(3.13b)
Eqs. (3.6b) and (3.13b) are equal. Thus, a fluid described by the Lagrangian in Eq.
(3.11) does not need the fluid-structure coupling functional to coupled to solids or,
rather, this functional is not added ad-hoc, but results from the change of coordinates:
20

it is F e in Eq. (3.12). The fluid, therefore, looks at the surrounding world in the same
fashion regardless of the chosen variable of response.
The functional F e is further defined: it is an analogy to twice the work at the boundary.
Integration by parts shows that
Fe
We
K e dV.
(3.14)
Substituting this result into Eq. (3.11), it follows that

Le
We
K e dV
L.
(3.15)
Eqs. (3.14) and (3.15) explain the change of sign of the strain and kinetic potentials in
Eq. (3.11) compared to Eq. (3.5).
3.4
A self-adjoint variational principle for Biots

equations
In a seminal work Biot formulates a theory for poro-elastic materials that models their
macroscopical behavior by considering them as homogenized two-phase media, i.e., the
solid elastic frame and the compressible fluid in the pores [8]. For porous materials
without dissipation, Biot gives expressions of the strain and the kinetic energies; their
difference defines the Lagrangian. Biot also introduces a Rayleigh-like dissipation
pseudo-potential modeling the viscous drag [52], p 315 (see Eqs. (15) in Paper 2).
Consequently, via variational calculus he devises the EulerLagrangeRayleigh equations
as functions of the frame and the fluid displacements. These equations are also referred
to as the displacement u, U -formulation. The employment of a dissipation pseudopotential is a common technique to handle viscous losses. Paper 2 presents a short
overview on notable articles using it. Herein, Morse and Feshbachs construction is
instead used. In the following, the main results for Biots theory are reported; the
complete derivation and analysis is found in Paper 2.
A variational principle needs potentials; the postulate of a strain energy functional is a
standard step in the derivation of variational principles, see e.g. pp 331 and 380 in Ref.
[29]. Besides, demonstrations of Hamiltons principle rely, at least implicitly, on the
fact that the strain energy density is a potential. The variational principle for harmonic
motion, a modified Hamiltons principle also valid for non-conservative media, reads
L
K dV
0;
(3.16)
L is called Lagrangian, W and K strain and kinetic potential densities. These potentials
depend on the frame and fluid displacements, respectively ui and Ui ; accordingly, the
21

frame strain %ij and the fluid dilatation e are defined as
%ij
1
ui,j
2
uj,i ;
Ui,i .
(3.17)
W and K are, respectively,

W
ijkl %kl
%aij D
ij e
%aij Q
e
ea R
ij %ij .
ea Q
(3.18)
and
K
uai 2 sc
j
ij u
uai 2 cij Uj
Uia 2 fc
ij Uj
Uia 2 cij uj ;
(3.19)
ijkl is the viscoelastic modulus tensor of the frame, Qij the potential coupling tensor, R
D
the bulk modulus of the fluid in the pores, sc
fc
cij tensors of equivalent densities
ij ,
ij and
for the frame, fluid and the coupling between them, accounting for mass, viscous and
inertial effects; details on the Biot material parameters and on the functionals can
be found in Paper 2. The potentials in Eqs. (3.18) and (3.19) are clearly self-adjoint.
Applying Eq. (3.16) Biots equations are retrieved, together with the boundary integrals.
Biots equations were originally expressed using the frame and the fluid displacements.
Refs. [6] and [5] derive a mixed u, p -formulation, employing the frame displacement
and the pore fluid pressure as field variables, thereby reducing the number of variables
from six to four. This formulation is derived via a variable substitution in Biots
equations and a subsequent weak formulation of these equations. Herein, the derivation
of the mixed formulation is based upon the invariance of the Lagrangian to coordinate
transformations. In order to switch from u, U coordinates to u, p coordinates,
Lagrange multipliers are employed. Specifically, to eliminate the dependance of L on
the fluid displacement, the fluid dilatation e is introduced as an additional variational
parameter. A new functional is defined:
L
C,
(3.20)
where C expresses a constraint via the Lagrange multipliers and a

a e
Ui,i
ea
a
Ui,i
dV.
(3.21)
Similarly to Eqs. (3.8), in Paper 2 it is shown that
h p,
(3.22)
where s is the fluid stress, h the porosity and p the fluid pressure. The fluid
displacements, the fluid dilatations and the Lagrange multipliers can be eliminated from
22

L, as done going from Eq. (3.7) to Eq. (3.11). Eventually, L reads
L
V
ijkl uk,l
uai,j D
uai,j ij p
uai,j
ij Q
kl
Q
uk,l
pa ij ui,j dV
uai 2 ij uj
pa
h2
p
pa,i
h2 fc
ij
p,j
2
F,
(3.23)
where
ij
sc
ij
cik fc
clj ;
kl
ij
h cik fc
kj
ij
Q
;
fc
fc
ik
kj
ni pa,j
h2 fc
ij
p dS.
2
ij .
(3.24)
The surface functional F in Eq. (3.23) is given by

ni pa
F
S
h2 fc
ij
p,j dS
2
(3.25)
L is self-adjoint and defines a variational principle:

L
0,
(3.26)
from which the equations of motion in the mixed formulation and their boundary
integrals follow.
3.4.1
Couplings to other media
Interesting insights to the results obtained for a porous material are gained looking at
the boundary integrals resulting from the first variations of the Lagrangians.
The boundary integrals for the original system obtained from Eq. (3.16) are
ni
uaj
ij dS
ni Uia s dS
0,
(3.27)
where
ij is the stress of the frame and s is the stress of the fluid. This equation is
satisfied either by prescribing the displacements, or by having free stresses, or by a
combination of the two. The boundary integrals obtained from the Lagrangian in the
mixed formulation in Eq. (3.26) are identical, see Appendix B of Paper 2. This result
is expected because the Lagrangian is invariant to changes of coordinates. In other
words, the way the porous material looks at the surrounding world does not depend
on the choice of coordinates. A specific coupling can thus be studied in any of the
formulations, the one that is most convenient (see e.g. Section 3.5).
The boundary integrals of the popular weak form by Atalla in Ref. [5] differ from Eq.
(3.27). That weak form, of course, satisfies the known coupling conditions to solids, fluids
and other porous materials, but it does so in a different fashion than the displacement
formulation. Atallas weak form couples naturally to solids and porous materials, i.e.,
23

without the need of any coupling integral; it needs additional fluid-structure coupling
integral to couple to fluids. It might be argued that on the boundary Atallas weak form
behaves more like a solid, than a porous material.
Two explanations to the difference between the self-adjoint variational principle and the
weak form are given. First, many weak forms may correspond to a set of equations of
motion, whereas it exists only one functional that is stationary for that set. Second,
the weak form is often derived from a set of equations of motion, thus containing
information only on the domain of the medium. The variational principle, instead,
contains information both on the domain and on its boundary. This difference is relevant
if the variational principle has boundary terms, like F in Eq. (3.25), which may not
easily obtained via a weak formulation.
3.4.2
Equivalent fluid theory
Besides Biots theory, another common theory for porous materials is equivalent fluid
theory: the porous material is modelled as a fluid with complex bulk modulus and
density. The porous material is thus described by one compressional wave, instead of
two compressional waves and one shear wave as in Biots theory.
Two equivalent fluid theories exist, one for rigid porous frames, one for limp porous
frames (e.g., see p 251 Ref. [1]); this thesis employs the latter because it better suits
porous layers attached to vibrating solid layers, like in the trim panels under study (p
253 Ref. [1]). Equivalent fluid theory for limp frame is conveniently derived from the
mixed formulation: the first two terms in Eq. (3.23) are the strain potential relative to
the stress of the frame in-vacuo, which equals zero for a limp frame. Thus, the functional
for such a limp porous material is7 :
uai 2 ui
Lpor
V
pa
h2
p
R
pa,i
h2
p,i
2 fc
a p
pa dV
(3.28)
where is the divergence of the frame displacement. The first variation of the Lagrangian
Lpor in Eq. (3.28) is stationary for the true motion of the system, yielding the equation
of motion and the boundary integral of the equivalent fluid. Taking the variation of the
adjoint coordinates, it follows
uai
2 ui
uai ni p dS
pa
p,i dV
V
ni pa,i
h2
p dS
fc 2
h2
p
R
h2
p,ii
fc 2
ui,i dV
0.
(3.29)
From the fundamental lemma of calculus of variations, the equations of motion of

the limp frame and of the fluid follow from the first two functionals in Eq. (3.29),
7
For the slightly different notation in the remaining of this chapter, see Paper 3.
24

respectively:
2 ui p,i 0,
h2
h2
p
p,ii ui,i
R
fc 2
(3.30a)
0.
(3.30b)
The divergence of the frame is eliminated from the fluid equation of motion, obtaining
the equation of motion of an equivalent fluid, Eq. (3.31):
R
p,ii
h2
2 L p
0;
fc
,
2 fc h2
(3.31)
The two boundary integrals in Eq. (3.29) may be collected and rewritten:
ni ut,a
i p dS
0,
(3.32)
hUi .
(3.33)
uti is the total displacement defined as

uti
h ui
This boundary integral contains the (variation of the) normal total displacement, which
is the continuous quantity when the equivalent fluid couples to solids, fluids or other
porous materials; notice that neither the normal frame displacement nor the normal fluid
displacement equals the normal solid displacement, which might be counterintuitive.
The author could not obtain the same boundary integral calculating the weak forms of
Eqs. (3.30) or Eq. (3.31).
3.5
Variational principle for a trim panel
The trim panels studied in this thesis are made of two external layers, one of aluminium
and one of rubber, with a porous layer in between them; its cross-section is shown in
Fig. 3.2. To start, a modified Hamiltons variational principle for the trim panel is
defined. Then, in Subsection 3.5.1 the model of the trim panel is simplified assuming
that its solid parts are limp. Upon variational calculus, the equations of motion of the
(limp) porous layer and of the (limp) external layers follow, together with the coupling
equations between them. These coupling equations are the sought result: they describe
the trim-coupled waves. This wave-type constitutes an SEA element, as expounded in
next chapter.
The Lagrangian of the coupled trim-panel is defined summing the Lagrangians of the
three layers and the functionals defining the constraints on the displacements of the three
layers. These constraints are introduced in a variational form via Lagrange multipliers;
their actual expressions are identified at the end of this section. The Lagrangian of the
25
Figure 3.2: From top to bottom, the aluminium plate, the porous layer and the rubber
plate.
coupled trim-panel, Ltot , reads

Ltot
Lpor
Lp 1
Lp 2
V,
(3.34)
where Lpor is the Lagrangian of the porous material in any of the three formulations
presented above, Lp1 and Lp2 are the Lagrangians of the external layers, T is the
functional expressing the continuity of the solid displacements, V is the functional
expressing the continuity of the normal total displacement in the porous material and
the normal displacements in the external layers. The variational principle based upon
Ltot is:
Ltot 0,
(3.35)
where indicates the first variation of Ltot . A sketch of the trim-panel of Fig. 3.2
in a cartesian system of coordinates is given in Fig. 3.3. Lp1 and Lp2 are detailed in
Paper 3 and the reader may refer to that; in the following, the focus is on the constraint
functionals.
The constraints are introduced in a variational form, using Lagrange multipliers. The
functional T reads
2
1
l 1
l
Sl
wipl
l
ui
a,p
i
pl uai
wia,pl
dSl ,
(3.36)
l
where pi l is a Lagrange multiplier and a,p
its adjoint; these Lagrange multipliers are
i
26
Figure 3.3: Coordinate systems used in the article. Volume V is occupied by the porous
layer, which is connected to the rubber layer at surface S1 , and to the aluminium layer
at surface S2 . The local normal vectors point outwards of the structures. The global
normal vector is aligned with the positive x2 axis.
vectors because T is (an analogy to) work on the boundary Sl . The functional V reads
2
1
l 1
l
Sl
a,pl ni uti
ni wipl
pl ni ut,a
i
ni wia,pl
dSl ,
(3.37)
where pl is a Lagrange multiplier and a,pl its adjoint.

In Paper 3, the actual values of the Lagrange multipliers are evaluated. The values
of the Lagrange multipliers in Eqs. (3.36) and (3.37) are identified by calculating the
variation in Eq. (3.35), with the multipliers as independent parameters. For plate p1 it
follows:
ijkl %kl , on S1 ;
p1 p;
pi 1
ijt
pij nj
nj D
(3.38)
ijkl is the stiffness of the frame in vacuo and, thus, p1 is identified as minus the
where D
i
traction of the frame in-vacuo: if the frame of the porous material is limp, the constraint
introduced by the functional T in Eq. (3.36) disappears. Similar results are obtained
for p2 and pi 2 for plate p2 , and also apply to the adjoint Lagrange multipliers.
Thus, when the frame is limp, the continuity of the solid displacements disappears,
accommodating the decrease of the number of waves in the model of the porous
materials, from three Biot waves to one equivalent fluid wave.
In Paper 3 these results on the Lagrange multipliers are derived using the displacement
formulation, because it allows for simpler calculations. The results are valid also for the
mixed formulation, in light of the comments in Section 3.4.1; they are also valid for a
limp porous material and they are used in the next subsection.
3.5.1
The trim-coupled waves
In the SEA for the sound transmission through the trim-panel presented in the next
chapter and derived in Paper 3, two elements describe the bending waves in the external
27

layers modelled as plates (Fig. 1.2). In this and in the next section, these bending waves
are neglected and the focus is on the elements accounting for the trim-coupled waves,
substituting the cavity-wall waves in Fig. 1.2.
The equations describing the trim-coupled waves are devised for a porous material that
has a limp frame, described by the Lagrangian in Eqs. (3.28). The trim-coupled waves
propagate in the equivalent fluid and are fully coupled to the external plates. If the
wavelength of the trim-coupled wave is much larger than the free wavelength of the
bending wave in a plate, then the motion of the plate is mass-impeded and, equivalently,
its material is assumed limp. The viscoelastic forces of plate pl are thus neglected and
its Lagrangian only comprises kinetic energy. The functional T in Eq. (3.36) disappears
because the Lagrange multiplier i equals zero for a limp porous material.
The equations of motion and the coupling equations for the original system are retrieved
from Ltot by taking the variations with respect to the adjoint generalised coordinates;
the full analysis is in Paper 3. The equations resulting from the fundamental lemma of
calculus of variations are: the equation of motion of the equivalent fluid, Eq. (3.31), the
equations of motion of the forced limp plates, the coupling equations of the trim-coupled
waves,
p
d p1 p,2 , on S1 ;
p d p2 p,2 , on S2 ,
(3.39)
where
p1
m p1
;
d L
p 2
m p2
,
d L
(3.40)
and d is the distance between the two plates, mp1 and mp2 are the mass per unit areas
of the plates. Eqs. (3.39) satisfy the continuity of the normal total displacement of the
porous layer and of the normal displacements of the external layers.
3.5.2
Comment on the variational statements
In this thesis, it is shown that if the variational statement of the coupling between two or
more media is properly set-up, no additional, ad-hoc coupling functional is needed. By
properly it is meant that: first, the Lagrangians of all media are included; second, the
functional expressing constraints are also included; third, the Lagrangians have proven
to be invariant to coordinate transformations both in the domain and on its boundary
(see discussion in Section 3.4.1).
The essential conditions may be either imposed in the statement, as in finite elements
procedures and in Paper 2, or introduced in a variational form using Lagrange
multipliers, as above and in Paper 3 (see p 43 in Ref. [39]). In this work, the latter
method shows its usefulness by revealing what happens to the condition on the continuity
of the solid displacements when equivalent fluid theory for limp frame is derived from
Biots theory, Eq. (3.38).
28
3.6
Wave form and modal density of the trimcoupled waves
Eqs. (3.39) are the sought equations describing the trim-coupled waves. The solution
p x, y, z to Eq. (3.31) can be decomposed as (x x1 , y x2 , z x3 )8 :
p x, y, z
g x, z f y
where the function g x, z describes a wave field along x

modal field along y. It follows that
2
g x, z
g x, z
2 g x, z
2
x
z2
2
f y
2 f y
0,
y2
(3.41)
z, while the function f y a
0;
(3.42a)
(3.42b)
where
kL2
2 ;
;
cL
kL
cL
R
h2 L
(3.43)
kL is the wavenumber of the equivalent fluid wave and cL its wavespeed. A wave solution
to f y in Eq. (3.42b) is a linear combination of sines and cosines. Solving the boundary
value problem given by the equation of motion in Eq. (3.42b) and the coupling conditions
in Eqs. (3.39), the following transcendental eigenvalue problem is obtained:
1
p 1 p 2 d
p 1
sin d
p2 d cos d
0.
(3.44)
Eq. (3.44) can be solved for d at each angular frequency , since p1 and p2 are
frequency dependant. The eigenvalue problem in Eq. (3.44) is approximately solved by:
0
2fdw
;
cL
2fr
.
cL
(3.45)
The frequency fdw corresponds to the typical double-wall resonance, which can be
expressed as
cL
1
1
fdw
;
(3.46)
2d p1 p2
the frequencies fr have half-wavelengths that are integer fractions of the plate distance,
fr
r cL
;
2d
N.
(3.47)
When Eq. (3.44) is solved in MATLAB using the fzero function, the real parts of 0
and r are used as initial values.
Two groups of waves are identified: one corresponds to the quasi-plane waves starting
8
The remaining of this section uses scalars instead of zeroth-order tensors, as in Paper 3 where this
section is taken from.
29

to propagate a fdw (r
0); one corresponds to the oblique coupled waves starting to
propagate at fr (r N). To each mode r described by r , corresponds a set of waves
with wavenumber r propagating in the x z plane; r is obtained via Eq. (3.43).
Finally, the pressure pr x, y, z for each mode r reads:
pr x, y, z
p sin r y
dp2 r cos r y
ir x cos z sin
(3.48)
where p is the complex amplitude of the wave and is the angle of propagation of the
wave. The trigonometric functions in the parentheses describe the modes along y, while
the exponential functions the waves travelling along x z. The function fzero yields as
output the real part of r ,
(3.49)
r Re r .
For all
r , the number of modes in a large two-dimensional structure below the angular
frequency is asymptotically given by (p 302 in Ref. [38])
2r S
.
4
Nr
The asymptotic modal density in x
nr
(3.50)
z is calculated as [38]:
2r u
2r l S
,
4 u l
(3.51)
where u and l are the upper and lower frequency of a frequency band having bandwidth
u l .
The most important results are the modal density in Eq. (3.51) and the pressure wave
in Eq. (3.48): both are used in the SEA of the trim-panel in the next chapter, where the
two groups of waves are used in the SEA of trim-panel as two separate SEA elements.
Results consistent to Eqs. (3.39) and (3.51) are obtained in Paper 1 for the coupled
waves in the cavity of a building construction double-wall, suggesting that these two
kinds of structures may be studied in a similar fashion. In that paper, the analysis is
done starting from Newton second law. A variational approach is, however, preferred
with the trim-panel to deal with a more complex medium - the porous layer.
30
Chapter 4
SEA of a trim panel
The first step in an SEA is the identification of a set of elements. As anticipated in
Section 1.1, the SEA of the sound transmission through double-walls in Fig. 1.2 derived
in Paper 1 sets the basis for the SEA of the trim panel1 . To calculate the SRI of the
trim-panel, the SEA mimics a SRI measurement: the trim-panel is placed in a rigid wall
between a reverberant room and an anechoic room; a diffuse sound field is generated in
the reverberation room via a distributed source. Thus, the SEA elements corresponding
to the wave-types carrying the energies from the reverberation room to the anechoic
room are: 1, an element describing the acoustic waves in the reverberation room; 2, an
element describing the bending waves in the aluminium plate; 3, an element describing
the quasi-plane coupled trim-plane wave cutting-on at fdw , Eq. (3.46); 4, an element
describing the oblique trim-coupled waves, each cutting-on at fr , Eq. (3.47); 5, an
element describing the bending waves in the rubber plate; 6, an element describing the
acoustic waves in the anechoic room. The wave form of the trim-coupled waves and
their modal density are calculated in the previous chapter; Paper 3 shows how to derive
the SEA parameters describing internal and coupling losses of all elements.
4.1
Results
In Ref. [16], Doutres et al. derive a criteria for assessing the validity of equivalent
fluid theory for limp frames, based on a parameter called frame stiffness influence (FSI).
The FSI expresses the ratio of two wave speeds, i.e., the wave speed of the wave in
frame in-vacuo and the wave speed of the equivalent fluid wave. A porous material can
be considered limp if the FSI is below a threshold, which in Ref. [16] is determined
simulating 256 porous materials.
The SEA model derived in the previous section is used to calculated the SRI of two
trim-panels, a car-floor and a truck-floor; the material parameters and details on the
1
The SEA elements of the trim panel can also be identified from the dispersion curves of the waves
propagating along the x z plane: this is done in Paper 3 but omitted in this introduction.
31
CHAPTER 4. SEA OF A TRIM PANEL

2
Wavenumber [1/m]
10
10
10
10
10
Wavenumber [1/m]
10
10
10
10
10
Frequency [Hz]
Figure 4.1: Real parts of the wavenumbers of the uncoupled porous layer in the car-floor
(top) and in the truck-floor (bottom). Fast wave (dashed line), slow wave (dot-dashed
line), equivalent fluid wave (full line).
SRI measurement are in Papers 4 and 5. The porous layer in the car-floor is limp
according to the FSI, whereas that in the truck-floor is not.
In Fig. 4.1, the dispersion curves of the uncoupled porous layers in the two floors are
shown; the real wavenumbers of the fast and slow waves are calculated using Allards
expressions on p 121 in Ref. [1], while the real wavenumber kL of the uncoupled
equivalent fluid wave using Eqs. (3.43). The two figures show that when equivalent
fluid theory works, its wave has a wavenumber close to the fast Biot waves.
The SEA defined in the previous section is applied to the car-floor and the truck-floor. In
Figs. 4.2-4.3, the SRI predicted by SEA is plotted and compared to SRI measurements.
In Fig. 4.2, SEA shows to perform generally well with the car-floor. SEA catches fairly
well the double-wall resonance frequency, around 260 Hz. The trend of the SRI between
260 Hz and 1000 Hz is also well caught. At frequencies below 260 Hz, SEA prediction
is poor: this problem is also encountered with building construction double-walls in
Ref. [24]. Fig. 4.2 also shows a reduced SEA that only employs three elements, i.e.,
the acoustic waves in the two rooms and the quasi-plane trim-coupled waves. In the
frequency range of interest, these three elements are enough to calculate the SRI, thus
highlighting their relevance.
In Fig. 4.3, SEA shows to give a rather incorrect prediction of the SRI of the truck-floor,
but for the trend between 500 Hz and 1000 Hz which is quite correct. It seems that if
the resonance frequency in the 500 Hz third-octave frequency band was caught and not
underestimated, the SEA prediction would give the correct level of the SRI between 500
Hz and 1000 Hz. It might be argued that the equivalent double-wall resonance frequency
corresponding to the dip in the SRI curve also depends on the frame stiffness, which is
neglected in the equivalent fluid model. The dispersion curves in Fig. 4.1 show that the
equivalent fluid wave likely has a lower stiffness than the fast wave: accordingly, in Fig.
4.3 the resonance frequency around 500 Hz is underestimated.
32
60
50
SRI [dB]
40
30
20
10
10
10
Frequency [Hz]
Figure 4.2: SRI curves, car-floor. Measurement (full line), SEA prediction (dashed line),
reduced SEA prediction (circles), mass-law prediction (dash-dotted line). Note that the
SEA and reduced SEA predictions are almost identical.
70
60
SRI [dB]
50
40
30
20
10
10
10
Frequency [Hz]
Figure 4.3: SRI curves, truck-floor. Measurement (full line), SEA prediction (dashed
line), mass-law prediction (dash-dotted line).
33
34
Chapter 5
Conclusions
This thesis investigates sound transmission in multilayered structures in the mid to
high frequency range, using statistical energy analysis and variational principles. The
multilayered structures of interest are building construction double-walls, double-glass
windows and vehicle trim-panels with a porous layer. The main scientific contributions
of this thesis are now summarised.
The problem of transmission through double-walls and trim-panels is throughly
investigated, from theory to simulations to experiments. The SEAs devised for
predicting their SRI are novel. All SEA elements refer to wave-types, permitting a
deeper understanding of the physics. Regarding double-walls, the new SEA performs
better than older ones [13, 14, 12] and proves to be versatile. Regarding trim-panels, the
proposed SEA is the first one that models the porous layer as an equivalent fluid derived
from Biot theory (to the knowledge of the author). One element of the SEA for the
trim-panel corresponds to the waves in the trim fully coupled to the external layers; the
equations describing these waves and their SEA parameters are derived. These equations
are derived using a variational principle, being a convenient machinery for media as
complex as the porous layer. The variational approach shows how to reduce coupling
conditions when going from Biot theory to equivalent fluid theory. The non-conservative
layers in the trim panel require the definition of self-adjoint variational principles valid
for dissipative systems: this is achieved following Morse and Feshbachs construction.
The self-adjoint variational principle for Biots equations is novel and may be seen as
a continuation to Biots original approach, based upon Hamiltons principle. Despite
skepticism towards Morse and Feshbachs construction [3, 7, 54], this thesis shows its
advantages by deriving original, sound and useful results for porous materials, fluids and
multilayered panels. The original study on energy transmission along long structures
questions the applicability of standard SEA, setting the basis for future research: at high
frequencies the statement of energy conservation as interpreted by SEA is incomplete
because indirect couplings become relevant.
35
CHAPTER 5. CONCLUSIONS
36
Bibliography
[1] J.F. Allard and N. Atalla. Propagation of Sound in Porous Media. Wiley, 2nd
edition, 2009.
[2] G. Altay and M. C. Dokmeci. On the equations governing the motion of an
anisotropic poroelastic material. Proceedings of the Royal Society A, 462:2373
2396, 2006.
[3] G. Apostolakis and G. F. Dargush. Mixed variational principles for dynamic
response of thermoelastic and poroelastic continua. International Journal of Solids
and Structures, 50:642650, 2013.
[4] R. N. Arnold and G. B. Warburton. Flexural vibrations of the walls of thin
cylindrical shells having freely supported ends. Proceedings of the Royal Society of
London. Series A. Mathematical and Physical Sciences, 197(1049):238256, 1949.
[5] N. Atalla, M. A. Hamdi, and R. Panneton. Enhanced weak integral formulation for
the mixed (u,p) poroelastic equations. Journal of the Acoustics Society of America,
109(6), June 2001.
[6] N. Atalla, R. Panneton, and P. Debergue. A mixed displacement-pressure
formulation for poroelastic materials. Journal of the Acoustics Society of America,
104(3), September 1998.
[7] V. L. Berdichevsky. Variational principles of continuum mechanics, volume I.
Fundamentals. Springer-Verlag, 2009.
[8] M. A. Biot. Theory of propagation of elastic waves in a fluid-satured porous solid. 1.
low-frequency range. Journal of the Acoustics Society of America, 26(182):168178,
1956.
[9] M. A. Biot. Generalized theory of acoustic propagation in porous dissipative media.
Journal of the Acoustical Society of America, 34:12541264, 1961.
[10] M. A. Biot. Mechanics of deformation and acoustic propagation in porous media.
Journal of Applied Physics, 33(4), 1962.
[11] A. Le Bot and V. Cotoni. Validity diagrams of statistical energy analysis. Journal
of Sound and Vibration, 329:221235, 2010.
37
BIBLIOGRAPHY
[12] R. J. M. Craik. Non-resonant sound transmission through double walls using
statistical energy analysis. Applied Acoustics, 64:325341, 2003.
[13] R. J. M. Craik and R. S. Smith. Sound transmission through double leaf lightweight
partitions, part i: airborne sound transmission. Applied Acoustics, 61(223-245),
2000.
[14] R. J. M. Craik and R. S. Smith. Sound transmission through lightweight parallel
plates. part ii: structure-borne sound. Applied Acoustics, 64:247269, 2000.
[15] F. dellIsola, A. Madeo, and P. Seppecher. Boundary conditions at fluid-permeable
interfaces in porous media: A variational approach. International Journal of Solids
and Structures, 46:31503164, 2009.
[16] O. Doutres, N. Dauchez, J.-M. Genevaux, and O. Dazel. Validity of the limp model
for porous materials: A criterion based on the biot theory. Journal of the Acoustical
Society of America, 122(4):20382048, 2007.
[17] F. J. Fahy. Statistical energy analysis: A critical overview. Philosophical
Transaction of the Royal Society of London, 346(1681):431447, 1994.
[18] R. P. Feynman, R. B. Leighton, and M. Sands. The Feynman lectures on physics,
volume 2. Addison Wesley, 1964.
[19] S. Finnveden. Energy flows within a three element structure with a statistical
description of the design parameters. Proceedings of the Inter-Noise, 90:945948,
1990.
[20] S. Finnveden. Exact spectral finite element analysis of stationary vibrations in a
rail way car structure. Acta Acustica, 2:461482, 1994.
[21] S. Finnveden. Simplified equations of motion for the radial-axis vibrations of fluid
filled pipes. Journal of Sound and Vibration, 208(5):685703, 1997.
[22] S. Finnveden. Statistical energy analysis of fluid-filled pipes. IUTAM symposium
on Statistical Energy Analysis, pages 289300, 1999.
[23] S. Finnveden. A quantitative criterion validating coupling power proportionality in
statistical energy analysis. Journal of Sound and Vibration, 330(1):87109, 2011.
[24] S. Finnveden and M. Barbagallo. A cavity-wall element for statistical energy
analysis of double walls. Technical Report TRITA-AVE 2012:02 ISSN 1651-7660,
AVE/MWL, 2012.
[25] S. Finnveden and M. Fraggstedt. Waveguide finite elements for curved structures.
Journal of Sound and Vibration, 312, 2008.
[26] M. Fraggstedt. Vibrations, damping and power dissipation in car tyres. PhD thesis,
AVE-SCI, Teknikringen 8 10044 Stockholm, 2008.
38
BIBLIOGRAPHY
[27] C. Fredo. Statistical energy analysis and the individual case.
Department of Applied Acoustic - Chalmers University, June 1995.
PhD thesis,
[28] C. Fredo. Sea-like approach for the derivation of energy flow coefficients with a
finite element model. Journal of Sound and Vibration, pages 654666, 1997.
[29] Y. C. Fung and P. Tong. Classical and Computational Solid Mechanics. World
Scientific, 2001.
[30] G. M. Gladwell. A variational formulation of damped acousto-structural vibration
problem. Journal of Sound and Vibration, 4(2), 1966.
[31] H. Goldstein. Classical Mechanics. Addison Wesley, 2nd edition, 1980.
[32] P. Goransson. Convergence aspects in fnite element solutions of multilayered,
heterogenous porous materials. In Proceedings of SAPEM, 2011.
[33] J.-L. Guayder. Vibration in Continuos Media. ISTE, London, UK, 2006.
[34] J.-L. Guayder, C. Boisson, and C. Lesueur. Energy transmission in finite coupled
plates, part i: theory. Journal of Sound and Vibration, 81(1):8192, 1982.
[35] K. H. Heron. Advanced statistical energy analysis. Philosophical Transaction of
the Royal Society of London, 346(1681):501510, 1994.
[36] V. Hongisto. Sound insulation of double walls - comparison of existing prediction
model. Acta Acustica united with Acustica, 92:6178, 2006.
[37] L. Jaouen, A. Renault, and M. Deverge. Elastic and damping characterizations of
acoustical porous materials: Available experimental methods and applications to a
melanine foam. Applied Acoustics, 69:1129 1140, 2008.
[38] E. E. Ungar L. Cremer, M. Heckl. Structure-Borne Sound. Springer-Verlag, 2nd
edition, 1988.
[39] C. Lanczos. The variational principle of mechanics. Dover, 1949.
[40] R. S. Langley. A general derivation of the statistical energy analysis equations for
coupled dynamic systems. Journal of Sound and Vibration, 135:499508, 1989.
[41] R. S. Langley. A derivation of the coupling loss factors used in statistical energy
analysis. Journal of Sound and Vibration, 141(2):207219, 1990.
[42] H. Liu, M. Barbagallo, and S. Finnveden. Wave motion and sound transmission
loss of double walls filled with porous materials. Proceedings of ISMA, 2012.
[43] S. L. Lopatnikov and A. H.-D. Cheng. Variational formulation of fluid infiltrated
porous material in thermal and mechanical equilibrium. Mechanics of Materials,
34:685704, 2002.
39
BIBLIOGRAPHY
[44] B. R. Mace. The statistical energy analysis of two continuous one-dimensional
subsystems. Journal of Sound and Vibration, 166:429461, 1993.
[45] B. R. Mace. Statistical energy analysis, energy distribution models and system
modes. Journal of Sound and Vibration, 264:391409, 2003.
[46] B. R. Mace. Statistical energy analysis: coupling loss factor, indirect coupling and
system modes. Journal of Sound and Vibration, 279:141170, 2005.
[47] G. A. Maximov. A generalized variational principle for dissipative hydrodynamics
and its application to Biots theory for the description of a fluid shear relaxation.
Acta Acustica united with Acustica, 96:199207, 2010.
[48] P. M. Morse and H. Feshbach. Methods of Theoretical Physics. McGraw-Hill, 1953.
[49] P. M. Morse and K. U. Ingard. Theoretical Acoustics. Princeton University Press,
1986.
[50] A. D. Pierce. Acoustics: An introduction to its physical principle and applications.
Acoustical Society of America, 1989.
[51] A. D. Pierce. Variational formulations in acoustic radiation and scattering. Physical
Acoustics, 22, 1993.
[52] Lord J. W. S. Rayleigh. Theory of Sound, volume Two. Dover, 1945.
[53] R.G. DeJong R.H. Lyon. Theory and Application of Statistical Energy Analysis.
Butterworth-Heinemann, 2nd edition, 1995.
[54] F. Riewe. Nonconservative lagrangian and hamiltonian mechanics. Physical Review
E., 53(2):18901899, 1996.
[55] A. W. M. Brown R.S. Langley. The ensemble statistics of the band-averaged energy
of a random system. Journal of Sound and Vibration, 275(847-857), 2004.
[56] T.D. Scharton and R.H. Lyon. Power flow and energy sharing in random vibration.
Journal of the Acoustical Society of America, (43):13321343, 1967.
[57] M. Villot, C. Guigou, and L. Gagliardini. Predicting the acoustical radiation
of finite size multi-layered structures by applying spatial windowing on infinite
structures. Journal of Sound and Vibration, 245(3):433455, 2002.
[58] J. Woodhouse. An approach to the theoretical background of statistical energy
analysis applied to structural vibration. Journal of the Acoustical Society of
America, (69):16951709, 1981.
[59] O. C. Zienkiewicz and R. K. Taylor. The Finite Element Method - Its basis &
fundamentals. McGraw-Hill, 6th edition, 1967.
40
Part II
Appended Papers
41
Paper 1
!
A Cavity-Wall Element for the Statistical Energy Analysis of the Sound

Transmission through Double Walls
S. Finnveden1*, M. Barbagallo2
Marcus Wallenberg Laboratory for Sound and Vibration, KTH
Teknikringen 8, 10044, Stockholm, Sweden.
1) svantef@kth.se , +4687909139; 2) matbar@kth.se
* corresponding author
Abstract
The wave motion within a cavity between two flexible walls is first investigated
numerically. The results then form the basis for a new SEA formulation in which each
element describe one kind of coupled cavity-wall wave motion. This formulation
obsoletes the non-resonant transmission path commonly used in SEA of sound
transmission and compared to classical formulations it improves results at frequencies
around and a bit above the double wall resonance. The new formulation is compared
to three sets of measurements found in the literature showing fair agreements.
!
Introduction
Double walls are frequently used in buildings, vehicles and aircrafts and continue to
be of interest for research, see, e.g., Refs [1-10], and the reviews on different
computational methods in Refs. [11-12].
This work concerns Statistical Energy Analysis (SEA), which is a convenient
approach for the prediction of the frequency- and space-averaged vibroacoustic
energy within elements [13-14]. The governing equation in SEA expresses the
conservation of energy based on the coupling power proportionality (CPP) law. This
asserts that the rate of energy flow from one element to another is proportional to the
difference in their modal power, which is the vibroacoustic energy in a frequency
band times the expected difference between natural frequencies. It is presumed that
the structure is a member of an ensemble of similar structures having random
properties and SEA attempts to predict the averaged energy across this ensemble. It is
furthermore presumed that the responses in the elements, at least in principle, may be
synthesized by sets of similar local modes having resonances in the frequency band of
interest, or by free waves in diffuse fields. These archetypical responses, or templates,
form the basis for the evaluation of parameters in the energy equations and are
!
"!
sometimes useful for evaluating response variables such as the mean square velocity
and mean square sound pressure.
An SEA requires the division of the system under study into a number of SEA
elements. An element may represent the total response of a substructure or the
response of one kind of motion in a substructure; a beam is typically represented by
four SEA elements that describe the motion of two orthogonal transverse waves, a
rotational wave and a quasi-longitudinal wave [13, Sect 7.1]. In accordance, earlier
works on double walls and fluid filled cylinders described the out of plane structural
waves in each substructure by one element and the acoustic waves in the contained
fluid as another [1-2, 15-17]. In a waveguide approach, Finnveden first observed the
wave motion along fluid filled cylinders and then proposed that the elements should
attribute the vibroacoustic energy of the different waves that could travel in the
composite structure [18]. Thus, in a lower frequency regime the predominantly fluid
element should attribute the flexibility of the pipe wall and the structural element
should attribute the associated fluid inertia so that the elements represented different
kinds of waves in the composite structure. This waveguide approach is used in the
present work where, first, the waves that travels in a plate-fluid-plate structure are
investigated and, then, new SEA elements are formulated that represent the energy of
the different kinds of waves in the composite.
The original SEA model for double walls was presented by Price and Crocker
[15], based on their earlier model for the sound transmission through single walls
[19]. This single-wall SEA model consists of three elements: two rooms in which
there are diffuse sound fields and in between them a plate performing resonant
vibrations. The constant of proportionality in the CPP law is defined by Smiths
wonderful result that the radiation of sound from a mode in a structure to an infinite
acoustic space is reciprocal to the modes reception of vibroacoustic energy from a
diffuse sound field [20]. This defines the resonant path: free waves in a room to
resonant vibration of plate to free waves in the receiving room. Additionally, Crocker
and Price introduced indirect coupling into SEA. In a lower frequency range where
the flexural waves in the plate are subsonic, its response to an acoustic forcing is
defined partly by the resonant modes, and partly by modes that have a length scale
comparable to the acoustic wave length and that have resonances at frequencies much
below the considered frequency. The response of these latter modes is impeded by
their modal mass and they radiate well into the other room. This non-resonant
"!
response is not modelled by an SEA element but the corresponding flow of energy is
described by an indirect coupling from room to room.
Crocker and Prices double-wall SEA model adds to the single-wall model a wall
and a cavity in between the walls; the radiation of sound into the cavity is described
equally as the radiation into a semi-infinite space. It appears as if the Crocker and
Prices double-wall is still the basis for SEA of double walls; structural couplings and
other complicating effects need of course be added when handling engineering
structures. Also, one commercial SEA program adds an indirect coupling between the
rooms and two indirect couplings describing energy flow from a room to the plate
adjacent to the other room.
Craik and Smith systematically investigate the Crocker and Prices model
comparing calculation results to measurements of different partitions [1-2]. They also
include structural coupling between the walls. In Ref. [5] Craik further improves the
SEA prediction upon the argument that in a lower frequency regime where the cavity
waves are plane, the forced wall motion should be sonic and therefore the radiation
efficiency associated with the indirect coupling exceeds the value of unity that was
presumed in earlier works.
Similar arguments are used in the present work where it is noticed that the wallcavity-wall wave that resembles the plane acoustic cavity wave is supersonic,
decreasing to sonic speed as frequency increases. The new approach differs from the
one in earlier works in two respects: 1) The elements and their modal density and
modal energy are defined by the waves that can travel in the plane of the composite
partition; 2) The wall-cavity-wall waves are directly coupled to the rooms so there are
no indirect couplings in the new model.
After a review of the SEA of double walls in Section 2, the wave motion of
double wall structures is investigated with the help of a waveguide finite element
model (WFEM) in Section 3. This analysis leads to a new formulation of an SEA
element for the coupled cavity-wall motion. Section 4 defines couplings in the
employed SEA model. Section 5 compares the results of this model to measurements
of building construction walls [1], aluminium double walls [4] and a double glass
window [3] showing a fair agreement. Finally, Section 6 concludes the work.
"!
2. Statistical energy analysis of double walls, a review

This section reviews the classical SEA of double walls as defined by Crocker and
Price [15] and Craik and Smith [1]. Much of the procedures and definitions are used
also in the new SEA model presented in Sect. 4 and 5.
!"#$%&'$()*+,-.*/,01$
Consider an element i excited by an external source providing a steady state
i
vibroacoustic power Pin( ) in a frequency band of width "! . A statistical energy
analysis of its response is built upon the energy conservation principle, stating that the
power injected into element i by external sources equals the power dissipated within
(i, j )
i
this element, Pd( ) , plus the net power transmitted to other elements Pc :
(i )
(i )
Pin = Pd
!P
j #i
(i, j )
c
.
(1)
A common engineering approximation for the dissipated power accounting for linear
losses, proportional to the vibroacoustic energy, is:
(i )
Pd
= $ i ! Ei ,
(2)
where ! is the frequency, $i is the damping loss factor and Ei is the vibroacoustic
energy of element i in the considered frequency band.
In SEA, the net power between two directly connected elements is often called
coupling power and is proportional to the difference in the energy per mode of the
two elements. This coupling power proportionality is herein formulated as
(i , j )
Pc
= C
(i , j )
(ei % e j ),
(3)
where the modal power is given by
ei = Ei ni ; ni = "N i "! ,
(4)
ni is the modal density, "N i is the number of modes in the considered frequency
band and C
(i , j )
is the modal power conductivity, or in any given context, simply the
conductivity. The conductivity is a non-dimensional parameter defined by Eq. (3),

(i, j )
being related to the coupling loss factor $ c
as:
C ( i , j ) = $c(i , j ) ! ni .
(5)
In possibly all practical applications of SEA, the conductivity is equated to its

travelling wave estimate, evaluated for one junction at a time, disregarding the rest of
"!
the system by extending the receiving element(s) to infinity. Thus, the conductivity is
given by
C
(i, j )
(i , j )
= Pc
ei
(i , j )
(6)
M j !0
Moreover, the CPP in Eq. (3) describes a potential flow and consequently:
(i , j )
= C
( j ,i )
(7)
There are numerous applications showing that the conductivity defined by Eqs (3) and
(6) obeys the law (7) including Refs [20-26].
Given the definition of the modal power in Eq. (4), the dissipated power is
rewritten:
(i )
Pd
= M i ei ,
(8)
where M is the non-dimensional modal overlap factor defined as
M i = " i # ni .
(9)
Prior to the identification of conductivities, modal overlap factors and input powers,
an SEA requires the identification of the elements describing the response of the
system under study. In the present work, the elements are defined by the wave types
that are supported by the double wall structure, as is further discussed in Section 3.
The next subsection presents the classical SEA for double walls, where each SEA
element corresponds to a physical substructure.
!"!#$%&#'(#)#*'+,-.#/)--#
Consider a partition of size S = a $ b , consisting of two thin-walled plates separated
by a distance d. For both plates the coincidence frequency, at which the flexural
wavelength of the plate w equals the acoustic wavelength, occurs at a rather high
frequency
2
fc =
c
2%
mw
,
Bw
(10)
where c is the speed of sound in air, m w is the plates mass per unit area and Bw is its
bending stiffness. In a lower frequency regime, much below the coincidence
frequency, the impedances of the plates to acoustic forcing are predominantly of mass
character. At such low frequencies, the double-wall resonance is mainly determined
by the masses of the walls and by the stiffness of the air within the cavity. London
"!
defines the double-wall resonance for walls that are heavy compared to the fluid mass
in the cavity as [27]:
f dw =
2! d
(11)
where w = mw / " 0 d and " 0 is the density of air. At frequencies lower than f dw
the plates move in phase and the Sound Reduction Index (SRI) is approximately given
by the field-incidence mass law ([14], pg 101):
RML
2
!
! mtot# " "
= 10 log # 1 + #
$ $ $ 5; mtot = m1 + m2 + " 0 d .
#
2" 0 c & $
%
%
&
(12)
At frequencies above the double wall resonance, Price and Crocker define in Ref.
[15] five SEA elements, each describing the response in one of the substructures: 1)
sending room; 2) plate 1; 3) cavity; 4) plate 2; 5) receiving room. The sound
transmission from a room through a plate to the cavity, or vice versa, has two
contributions. One is the direct resonant transmission: the sound field excites the
resonances of the plate that in turn radiates into the cavity. The other is the indirect
non-resonant transmission of the mass-impeded modes of the plate, described by the
mass law; this second transmission path no longer exists above the coincidence
frequency.
Craik and Smith, in addition to the SEA elements employed by Price and
Crocker, add structural coupling. Measurements and calculations of the SRI of double
wall partitions are presented in References [1-2] where the plates are made of
plasterboards or chipboard; they are joined by wooden frames, modelled as Euler
beams that are erected in the vertical direction at a given distance; the plates are
nailed to the frames.
Fig. 1 illustrate results from Craik and Smiths model [1] where the measured and
calculated SRI for a double wall with a cavity depth of 100 mm is shown; the double
wall resonance occurs at 87 Hz, while the cavity depth equals half an acoustic
wavelength at 1.7 kHz and the coincidence frequency is in the 3.15 kHz third-octave
band. (The construction is defined by the data in Table 1 and should be the same as in
Ref. [1]). As it may be seen, Price and Crockers model, as interpreted by Craik and
Smith, performs well at lower frequencies, where the partition is modelled as one
single plate, even though the mode count is rather low at these frequencies. The
model is also good at higher frequencies where the acoustic wavelength is not very
!
"!
much longer than the cavity depth. At intermediate frequencies there are consistent
errors that are large just above the double wall resonance and decrease as frequency
increases. Most disturbing is the trend of the SRI at the intermediate frequencies
where the measurements show an increase of perhaps 12 dB per octave and the model
an increase of a bit more than 6 dB. This indicates that the SEA model doesnt capture
the physics of the sound transmission and might, therefore, not only give incorrect
results but might also indicate incorrect trends for measures applied to improve a
construction.
The double wall resonance is not observed in the measured SRI, whereas there is
a plateau extending approximately one octave above this frequency. The dip in the
160 Hz band, seen also in the measurements where the cavity depth is 50 mm and 150
mm, might be caused by an increase in the plate mobility. (The distance between the
beams is a bit more than half a plate wavelength which would correspond to the first
set of resonance if the boundary conditions were in between simply supported and
clamped.)
!"#$%&'()*+,$+-./0(&1$2-)$3-&(+$+45(,678400$9-,(-&$
In Reference [5] Craik improves the non-resonant coupling loss factor between a
room and the cavity upon the argument that the wall-vibrations forced by a plane
acoustic wave in the cavity radiates as if the considered frequency was the
coincidence frequency, since the acoustic waves in the cavity are sonic. As seen in
Fig. 1, this significantly improves the agreement between model and measurements at
intermediate frequencies.
The SEA for double walls in References [1, 15] is herein the basis for the
analysis, except that a cavity-wall element replaces the old cavity element The next
section first investigates the wave motion in a double wall and then identifies the
elements of the new SEA model.
3. Waves propagating in a double wall and the associated SEA elements

The structure of interest consists of an air cavity in between two plates. The plates are
parallel to the x-y plane and are separated by a distance d in the z-direction. The
motion is harmonic of the form of
(
u ( x, y, t ) = Re (
p( x, y, z, t ) = Re
w
2 p! ( x, y, z, ! ) e
2 u! w ( x, y, ! ) e
"!
" i!t
" i!t
)
(13)
!
where p is the sound pressure in the fluid and uw is the displacement in the zdirection of plate w. The equations of motion for this structure are
" 2 p! + k a2 p! = 0; ka = ! c ,
(14)
Bw " 4 u! w $ mw ! 2 u! = # w p! , #1 = $1; # 2 = 1.
(15)
The plate velocities equal the fluid velocity at their interfaces and it follows that
% 0 ! 2 u! w =
& p! ( x, y , z = z w , ! )
; z1 = 0; z2 = d .
&z
(16)
The set of equations (14)-(16) are first solved numerically. Then, based on the
knowledge gained, the equations are approximated and employed to define the SEA
elements used for the analysis of the transport of vibroacoustic energy through double
wall partitions.
!"#$%&'()*+,-$,.,-/0*0$12$34($+1&5-(6$+,7*3/$8,--$'13*1.$
The waveguide finite element method is a versatile tool for investigations of wave
motion in structures that have constant material and geometrical properties along one
direction [28-30]. Using the WFEM, the motions dependence on the cross sectional
coordinates is approximated with FE polynomial shape functions. Upon the
application of standard FE procedures follows a set of coupled ordinary differential
equations describing wave propagation and decay along the structure.
Fig. 2 shows the FE mesh in the x-y plane used to model the 100 mm building
construction double wall in Ref. [1], the data are defined in Table 1. The plasterboards
are described by three-node quadratic deep shell elements and the fluid by nine-node
Lagrange type elements [29-30]. All motion in the y direction is blocked so the model
is, in effect, two-dimensional. Fig. 2 also shows the mesh used to depict the wave
motion in the x-z plane.
Fig. 3 shows the dispersion curves for this double wall. The two straight curves at
the bottom are the uncoupled quasi-longitudinal waves in the two plates; these have
slightly different wavenumbers as the plates have slightly different characteristics.
The two straight lines at the top are the in vacuo flexural waves in the plates. The
calculated wavenumbers for the flexural waves are close to those for the uncoupled
plates, also at frequencies below the double wall resonance (87 Hz), which is perhaps
surprising. The waveforms, however, exhibit coupling between the plates and
between the plates and the fluid, as Fig. 4 shows. At low frequencies one of the
coupled flexural waves is anti-symmetric and the other is symmetric, whereas the
!
"!
plate amplitudes are in both cases roughly equal. Above the double wall resonance
each wave is localised to either of the plates while the amplitude of the other plate
decreases rapidly with frequency.
The waves that cut-on, or start propagating, at 87 Hz and 1.7 kHz are
predominantly fluid waves that have supersonic wave speeds along the structure
decreasing towards sonic speed as frequency increases. The waveforms of the
predominantly two fluid wave types are shown in Fig. 5 and Fig. 6. In the former, the
fluid motion is almost plane with some motion of the walls, which decreases with
frequency. In the latter, the fluid moves while there is almost no motion of the walls.
The behaviour of these two fluid waves is evidently different. It is significant for an
SEA that the fluid waves do not exist below their cut-on frequencies and that the
almost plane wave is supersonic, or near sonic, in the plane of the plates, so the
associated wall motion radiates well.
!"#$%&'()*+*,-)*.($.+$/01$'2'3'()4$
The WFEM has revealed the waves that carry energy along the double wall. It has
shown that the plates flexural and quasi-longitudinal waves are almost as in vacuo. It
has shown the existence of supersonic, almost plane, waves having large motion both
in the plates and in the fluid; these cavity-wall waves might have been neglected in
previous SEA of double walls. Finally, it has shown oblique fluid waves involving
very little wall motion.
In an infinite homogenous structure, these waves would propagate independently.
In a finite structure that is not perfectly homogenous, the waves are coupled. If,
however, the structure is large, the waves might be weakly coupled and may
characterize SEA elements. The SEA model defined upon these arguments is
illustrated in Fig. 7; the elements are defined below and the couplings are defined in
Section 4.
3.2.1 Plate elements

The flexural plate motion is, a bit above the double wall resonance, almost as if in
vacuo, and, accordingly, for each plate one SEA element can be defined that is
weakly coupled to the rest of the structure. The mode count and modal density for
plate w are given by [13, p140]
S k w2
;
Nw =
4!
"!
(17)
nw =
S kw
# N # kw
=
,
#kw #!
2" cg .w
(18)
$%&'&!
kw =
! m w Bw ; c g . w = # ! # k w ;
(19)
kw is the flexural wavenumber and c g . w is the group velocity. Eq (18), together with
the damping loss factor, defines the elements modal overlap factors. The
conductivities, quantifying the energy flow to the rooms and to the cavity-wall
element, are defined by the radiation efficiencies, calculated by Leppington et al [31],
as is discussed in Section 4.1.
3.2.2 Room elements and beam elements
The SEA model mimics a measurement in a transmission loss suit: there is a source in
the source room and sound is eventually transferred to the receiving room. These
large 3-D fluid cavities have modal densities and losses that are so large that their
actual values do not influence the calculated SRI.
The vertical beams that support the double wall structure are modelled as pointconnected Euler beams, the details are found in Ref. [1].
3.2.3 Cavity-wall element below the coincidence frequency
The cavity-wall waves are supersonic and therefore at frequencies below the
coincidence frequency, the plates bending stiffness, the first term in Eq. (15), is not
significant. Eqs (15) and (16) then define mixed Robin boundary conditions for the
fluid:
w d
# p!
= %$ w p! , z = z w .
#z
(20)
A solution to Eqs. (14) that describes the wave motion in the x-y plane and fulfils the
boundary conditions (20) is given by
p! = p! 0 ( sin ( r z + 2 ( r d cos ( r z ) e
i & r ( x cos ' + y sin ' )
(21)
where any value of the angle ' is admissible,
& r2 = ka2 % ( r2 ,
(22)
and ( r is one of the numerable solutions to the transcendental eigenvalue problem
(1 % 1 2( r d )sin ( ( r d ) + ( 1 + 2 ) ( r d cos ( ( r d ) = 0 .
2
"#
(23)
It is straight forward to solve Eq. (23) numerically; however, analytical solutions are
more descriptive, which helps the analyst. If 1 >> 1, 2 >> 1 , Londons formula (11)
for the double wall resonance is useful, this is the case for all structures studied in this
work. The eigenvalues ! r are then well approximated by
!0 =
2" f dw
1
=
c
d
! r = 2" f r / c = r" d ; r = 1, 2, ... .
(24)
(25)
Consequently, for each eigenvalue ! r there is a two-dimensional wave field in the

x-y plane with a given mode shape or wave form in the z-direction, as defined by Eq.
(21). Each of these wave fields are here used to define an SEA element. These
elements have mode counts that, in analogy with Eq. (17), are given by
N r = S # r2 4"
(26)
where # r is defined by Eq. (22) and it is implicit that the mode count is zero if
# r2 < 0 . At the cut on frequencies, the group velocity is zero and the modal density
defined by Eq. (18) is infinite [28]. This equation is therefore not useful and instead
the modal density, defined for a frequency band of width %$ = $u & $l , is given by
nr = ( N r ($u ) & N r ($l ) ) %$ = S (# r2 ($u ) & # r2 ($l ) ) 4" %$ .
(27)
3.2.3 Cavity-wall element at and above the coincidence frequency

Eq. (23) is derived upon the assumption flexural motion of the walls is sub sonic,
which is not true at and above the coincidence frequency. At such high frequencies, it
is seen in Figs. 5 and 6 that there is very little wall motion and, therefore, the cavitywall element becomes a conventional SEA cavity element defined for rigid walls. The
approximate solution (25) applies also above coincidence as it equals the solution for
blocked wall motion. With rigid walls, instead of the eigen solution (21), we have
! 0 = 0 . If, however, the double wall frequency is much lower than the coincidence
frequency, this difference wont affect the modal density (27) much. The transition of
the cavity-wall element into a cavity element will, however, affect the conductivities.
In particular, when the element is defined for rigid walls there is no direct transfer of
energy between the element and the rooms, as is discussed in the next section.
4. Couplings in the SEA double wall model

The SEA elements used in the analysis of double walls are identified in the previous
section. The structural coupling between the walls via the beams (coupling path G in
!
""
Fig. 7) is described as in Ref [1]. The indirect mass impeded coupling path through
the entire structure, which exists below and around the double wall resonance,
coupling path C, is defined by the mass law (12). All other conductivities are defined
by the power radiated from a vibrating surface to an acoustic volume, as is described
in what follows.
4.1 Radiation efficiency
Leppington et al [31] calculate the sound radiation into an infinite acoustic space from
a rectangular simply supported thin walled plate, which is inserted into an infinite
baffle and has its vibration velocity defined by the modes having resonances in a
frequency band while these have, on average, equal vibration amplitude. The radiated
sound power is given by
Prad = ! o c " S v! ,
2
where " is the radiation efficiency and v!
(28)
is the plates temporal and spatial mean
square vibration velocity in the transverse direction.The radiation efficiency is often

expressed as a function of frequency. The analysis in Ref. [31] is, however, made in
the wavenumber domain considering that the radiating surfaces vibrational waves
have wavenumber ks and the fluid waves k f . Thus, Leppingtons radiation efficiency
[31, Eqs. (7.6), (7.7) and (7.11)] is given by
!
c ( a + b)
2# #
" $# + 1 %
+
1 ) ln &
'
( 2
2
*, # > 1
2
2
#
%
1
+
,
#
%
1
.
( 2$ # k f a b # % 1
((
a%
$
" (# ) = /(
& 0.5 % 0.15 ' k f a , # = 1
b,
+
(
1
2 % 2
(
1
%
#
, # < 1
(0
(29)
$%&'&! a and b !('&!)%&!*+,&!-&./)%*0!*1!)%()! S = a b 0! a & b 0!(.,!
# = ks k f .
(30)
!"#$%&''$()$*))+$&,-$.&''$()$/&01(23.&''$/)45'1,6$
The conductivity is equated as in Eq. (6) based on the radiated power (28) :
C (i , j ) =
2
c " (i , j ) mi v!i
,
i d
ei
"#
(31)
!
where ei is the modal power associated with the wave motion considered and the
quantity mi v!i2 ei is defined by the wave form associated with these waves. For the
flexural vibration of plate w , radiating into an acoustic volume, R ,
! ( w,R ) = ! ( k w / ka )
(32)
and
mw v!w 2
ew = nw
(33)
where nw is the modal density of plate w . Thus, the conductivity for the plate - room
coupling (coupling path D in Fig 7) is defined.
The current work follows the practice in Refs. [15, 1] and models the radiation
into the cavity as if the plate was radiating into an infinite space. The conductivity for
the coupling between a plate and the plane cavity-wall element (coupling path E in
Fig. 7) at frequencies below the cut-on of the first oblique cavity mode is therefore
also defined by Eqs. (31) and (33). At frequencies above the cut-on, the sum of the
conductivities for path E and F is likewise defined by Eqs. (31) and (33) while they
are here set proportional to the modal density of the plane and oblique cavity-wall
modes respectively. (Thus, it is assumed that all mode to mode coupling loss factors
are equal.) Consequently, the conductivity between plate w and cavity wall element r
is defined as:
( w, r )
c!
( w, r )
n w nr
; nrtot =
w d nrtot
!n
(34)
The radiation efficiency ! ( w ,r ) from the plate to the cavity is given by the radiation
efficiency ! ( k w / k r ) in Eq. (29) times the multipliers suggested by Craik and Smith
in Fig. 5 of Ref. [1].
4.3 Cavity-wall to room coupling
To evaluate the conductivity between a cavity-wall element and a room, first the
mean square vibration velocity of the wall is related to the modal power of a cavitywall element; then, the radiation efficiency from the cavity-wall element to the room
is defined.
4.3.1 Wall vibration in cavity-wall waves
In the cavity-wall element at frequencies below coincidence all strain energy is in the
fluid, whereas the kinetic energy is both in the walls and in the fluid. The strain
!
"#
!
energy density in an acoustic volume is e p = p!
2! 0c . For resonant modes, the

2
averaged kinetic energy equals the averaged strain energy density, and the total
acoustic energy in the cavity-wall element is 2 e p V , where e p is the strain
V
energy averaged over the volume V. Thus, the modal power potential for a cavity-wall
element r is given by
er =
S p! 0
! ( sin "
d
4 n r ! 0c
z + 2 " r d cos " r z ) d z ,

2
(35)
if wave type r is propagating and zero otherwise; the factor 1/4 comes from
averaging the strain energy in a diffuse wave field in the x-y plane. The integral in Eq.
(35) could be solved analytically, while it is here evaluated by quadrature at no
numerical cost. From Eqs. (20) and (21) it follows that the spatially and temporally
squared averaged vibration velocity of plate w associated with cavity-wall element r
is:
v!
2
w,r
p! 0
(sin " r z w + 2 " rd cos " r z w )

2
4 (# ! 0 d w )
(36)
It follows that
2
m w v! w,r
e r
( sin " r z w + 2 " r d cos " r z w )

.
d
2
2
# S d w ! ( sin " r z + 2 " r d cos " r z ) d z
0
nrc
(37)
For heavy walls, Londons formula applies and the quantity " 0d is small. A
series expansion then provides an approximate plane wave cavity-wall to room
conductivity, while this is still work in progress.
4.3.2 Sound radiation from wall vibrations in cavity-wall waves
The ratio of ms vibration velocity of the wall to the modal power of the cavity-wall
waves is calculated above; to evaluate the conductivity (31), the radiation efficiency
for these waves is also required. It is typically assumed that the radiation efficiency of
a limp wall forced by a diffuse sound field equals unity [1, 15, 19]. Craik, however,
argues in Ref. [5] that plane acoustic waves between two walls have sonic speed.
Consequently he defines the radiation efficiency by its value at coincidence, and, as
seen in Fig 1, the SEA prediction is improved. The results for fully coupled fluidstructure motion in Fig. 3 show that Craiks surmise is almost correct: the cavity-wall
waves are supersonic, asymptotically approaching sonic speed as frequency increase.
"#
!
The radiation efficiency is thus given by ! < 1 in Eq. (29), while it should not exceed
the maximum value for sonic wave speed given by Eq. (29) for ! = 1 .
At and above coincidence, the cavity pressure no longer excites the mass impeded
wall motion, but the resonant modes only. At these higher frequencies, the importance
of the resonant wall vibrations increase, the cavity-wall waves turn into cavity waves
with blocked walls and the room to cavity-wall coupling vanishes. A dodge to
account for the two latter effects is to multiply the radiation efficiency associated with
this coupling with the factor (1 # " ( w,R ) ) or zero, whichever is larger. Consequently,
the radiation efficiency from cavity wall element r to room R is given by
"
(r ,R)
r
= min (" ( k r / k a ) , " (1) ) $ max (1 # " ( k w / k a ) , 0) .
(38)
Upon this result, the conductivity for the coupling of a cavity wall element and a
room is
2
C
2
where the factor m w v! w ,r
( r ,R )
(r,R )
= %0 c S " r
m w v! w ,r
er
(39)
e r is defined in Eq (37).
All the parameters of the cavity-wall element needed for the SEA of a double
wall are now detailed: its scheme is illustrated in Fig. 7.
!"#"#$%&''()*$&)$*+($,-./-*/&)$(00/1/()12$
The radiation efficiencies for a plate and a cavity-wall element, both radiating into a
room, are different. The waves defining the cavity-wall element are supersonic,
approaching sonic speed as frequency increases. The radiation efficiency for a plate,
on the other hand, approach as frequency increases its maximum value at coincidence
and decreases towards unity at even higher frequencies. Fig. 8 describes this peculiar
behaviour.
5. Comparison to measurements
The performance of the proposed SEA model is evaluated for three kinds of double
wall structures found in the literature: i) three building constructions [1], ii) two
aluminium structures [4] and iii) a double glass window [3]. Tables 1-3 detail the
properties of the investigated structures.
The quantity of interest is the SRI which for large rooms with sufficient damping
is given by
"#
R = 10 log (1 ! ) ; ! =
( M e )receiver room
Pt
A ereceiver room
= 2
=
.
2
Pinc ( ka S 8 " ) esource room S esource room
(40)
Pinc !!$%!&'(!)*+(,!-*.$/0!*/&*!&'(!)1,&$&$*/!*2!1,(1! S !3! Pt !!$%!&'(!)*+(,!
&,1/%.$&&(4!&*!&'(!,(-($5$/0!,**.!+'(,(!&'(!(67$518(/&!19%*,)&$*/!1,(1!!$%! A :!!
!"#$%&'()*'+&$,-$.*)/(01$
The damping loss factors are required to calculate the modal overlap factors in Eq. (9)
and have a strong influence on the results. It is, nevertheless, often difficult to retrieve
damping values from the quoted articles. Some considerations are needed.
The damping loss factors of the rooms are based on their equivalent absorption
areas, which are so large that their precise values are irrelevant for the calculated SRI.
In Reference [1], measurements of the reverberation time for the 100 mm cavity
are reported. In the absence of any other information, it is here assumed that the
damping loss factors for the plane and oblique cavity-wall elements are equally
defined by this reverberation time. It is, furthermore, assumed that the damping loss
factors for the 50 mm and the 150 mm cavities are given by the one for the 100 mm
cavity in inverse proportion to the cavity depth. The arguments for this assumption
are that most of the damping arises from viscous forces and heat conduction at the
walls and frames and from air pumping at the interface of these structural
components. Therefore, the losses might be independent of cavity depth in which case
the loss factor is inversely proportional to cavity depth. The reverberation time is
measured in situo and is therefore defined by the total losses including both damping
and coupling losses. The latter might have been estimated by the SEA model and
subtracted from the total losses though this was not made here. There is no
information about the cavity damping in Refs. [3-4].
Wall-loss measurements are provided in Refs. [1, 4]; Ref. [1] presents these
measurements for one of the plasterboard walls forming a part of the 50 mm double
wall. The information on damping loss is consequently incomplete and, furthermore,
do not distinguish between total and internal damping losses. The SEA predictions
that correspond to the measurements in references must therefore be based on some
speculations; Table 1-3 defines the current authors estimates.
"#
!
!"#"$%%&'()*(+,%-.+/01'-0(.+/%
Fig. and Figs. 9-10 present the sound reduction indexes of the three building
construction double walls; the results from the new SEA model are compared to those
from the old model and to the measurements in Ref. [1].
The agreement in the low frequency region, below, around and an octave above
the double wall resonance, is good but somewhat erratic, probably so as the mode
count is rather low for most of the elements. There is, however, a consistent
overestimation of the SRI in the 160 Hz band, which might be caused by the
increased mobility of the plates because of resonances. The proposed SEA model
performs quite well at the coincidence frequency in the 3.15 kHz band, in particular
for the 100 mm cavity. The calculation results presented here and those by Craik and
Smith [] should be identical in the high frequency range. This is not exactly so, which
might be caused by difficulties in reading the data. In the frequency region between
the double wall resonance and coincidence, the agreements between the new model
and the measurements are excellent for the 50 mm and the 100 mm cavities and quite
good for the 150 mm cavity The increase with frequency of the calculated SRI is
generally correct and, given the expected uncertainty of an SEA in general, and of the
loss factors in particular, the results are better than warranted.
5.2.2 Aluminium double wall
Reference [4] presents measurements of the SRI of aluminium double walls, with
cavity depths of 125 mm and 250 mm. These measurements are compared to SEA
results in Figs. 11-12. For both constructions, the calculated SRI is too large at
frequencies below the coincidence whereas around this frequency the agreement is
good. It is interesting to note that the cut-on of oblique cavity modes, around 700 Hz
and 1.4 kHz, hardly influence the SRI.
Reference [4] claims that there is no flanking transmission in the measurement
setup. It is, however, very difficult to completely avoid flanking transmission and
leakage and it might be that such coupling paths restricts the SRI below 60 dB. It is
also plausible that there is some direct coupling between the plates if the fluid near
field extends from one plate to the other, as would be the case at low frequencies and
frequencies not much below the coincidence frequency. To mimic such imperfections
a minute coupling loss between the plates is added to the SEA model: ! c = 10 "5 . Figs
11-12 show that this minor modification increases the agreement between
measurements and calculations.
!
"#
5.2.3 Glass window

Ref. [3] presents measurements and calculations of the SRI of a double glass window,
which are compared to results from the proposed SEA. Fig. 13 shows that at low and
middle frequencies the agreement is rather satisfying, whereas a bit below and above
the coincidence frequency the SEA prediction is quite wrong.
One reason for this might be that the constant, frequency independent, loss factor
is not correct at all frequencies. It is also plausible that there is some direct coupling
between the glasses caused by flanking transmission or fluid near field transmission.
It turns out that if the damping of the glass plates is reduced and a small coupling loss
is added, the SEA prediction considerably improves as Fig. 13 shows.
5.3 A reduced SEA model
In the mid-frequency region, between the double wall frequency and either the cut-on
frequency for the oblique cavity waves or the coincidence frequency, whichever
comes first, the sound transmission is dominated by the room to cavity-wall to room
path (coupling paths A in Fig 7). This is illustrated in Fig 14 where the results from a
three element SEA model are compared to those of the full model.
5.3.1 A simplified formula for the transmission coefficient
Based on the reduced SEA model, a formula for the transmission efficiency of double
walls is derived upon the assumptions that the walls have equal properties. It is
moreover assumed that the walls are heavy compared to the contained air and that
cavity response is damping controlled. In mathematical terms it is thus assumed that
1 = 2 = << 1; C ( r ,R1) = C ( r ,R 2) << M r
(41)
From the heaviness of the walls it follows that ( ! 0 d ) << 1 . Now, calculating the
transmission efficiency from the SEA model and evaluating the resulting expression
to leading order of the small parameters in Eq. (41), it follows that
# =
0.23 c 5 a TR
=
d 6 24 $ 4
!
%
a"
&
" a 4 # 0.5 & 0.15 $

b
d 4 % r ( ka d )
(42)
The SRI thus increases with 15 dB per octave, with 12 dB per mass doubling and with
3 dB per loss factor doubling. The formula for the SRI of double walls taught in
fundamental courses in noise and vibration control, e.g., [32-33] has the same
dependence on mass but predicts an increase of the SRI with 12 dB octave and is
!
"#
independent of damping. The SRI in Fig 1 and Fig 14 increases about 13 dB/octave.
This is a typical characteristic shown by all the double walls herein investigated (the
T
reverberation time R measured by Craik and Smith decreases approximately with
1.3 dB/octave). Fig 14 shows the SRI for the 100 mm building construction double
wall calculated with the full SEA model, the reduced SEA model and the simplified
formula for the transmission coefficient.
The building constructions considered herein are well described by the
approximations (41); however, for cavities that have response that is not controlled by
( r , R 1)
( r , R 2)
= C
damping but by coupling loss, such that M r ! C
, it is possible to
derive the following approximated formula:
" =
! ka a
22 ( ka d )
(43)
6. Conclusions
The present article develops the classical SEA model for double walls [1-15]. The
investigated structures have wall masses that are large compared to the mass of the
contained fluid, i.e., >> 1 , and the coincidence, when uncoupled flexural plate
waves and acoustic waves have equal wave length, is at a rather high frequencies.
Ref. [1] shows that the old SEA model and measurements compare poorly in a midfrequency range, from the double wall resonance to a bit below the coincidence. The
new model improves mid-frequency results, while at frequencies below and above
this region it equals the old model.
First, the waveguide FEM is used as a diagnostic tool to find the waves that
travels along the fully coupled wall cavity wall system. It is seen that there are no
waves in the cavity at frequencies below the double-wall resonance. At frequencies
above this resonance there are quasi-plane fluid waves that involve quite large, mass
impeded, wall motion; these are the cavity-wall waves. At even higher frequencies,
when the cavity depth is larger than half the acoustic wave length, there are oblique
cavity waves. The walls in-plane and flexural waves are almost as for the walls in
isolation. At lower frequencies there are, however, fluid near fields that decay away
from one wall towards the other. These near fields couple the walls motion and an
improved low frequency model, of particular interest for shallow cavities, would
describe this coupling. Similarly, as the frequency approaches the coincidence
frequency, the near fields should extend further away from the walls and there might
be a direct wall to wall coupling. Neither of these improvements was attempted here.
"#
It was, however, observed that a minute ad hoc wall to wall coupling improved the
SEA predictions at frequencies somewhat below the coincidence.
The new SEA model is depicted in Fig. 7. It contains two acoustic elements,
describing the source and receiver rooms, and two plate elements, describing flexural
vibrations in the walls. For the building constructions examples, it also contains a
beam element describing the frames erected between the walls. These five elements
and the couplings between them are modelled as in Ref. [1].
The modes that correspond to the cavity-wall waves define one SEA element for
the cavity; the other element is defined by the oblique cavity waves. Below the
double-wall resonance there is no transmission through the cavity but only the mass
impeded room to room coupling (path C).
The new element formulation obsoletes the non-resonant transmission paths used
in earlier double-wall formulations. Instead there is a direct coupling between the
acoustic volumes and the resonant cavity-wall modes, which depends on the cross
sectional wave form. The conductivity also depends on the radiation efficiency; it is
unity at the double-wall resonance (as in Refs. [1]) and then increases towards the
value at sonic wave speed (as in Ref. [5]). Similarly, the modal density is high at the
resonance and then deceases towards the one for a two-dimensional cavity. These
factors explain why the reduction index does not have a distinct minimum at the
double-wall resonance but instead have a rather constant value in a frequency region,
which extends roughly an octave above this resonance.
The new SEA model is compared to three sets of measurements found in the
literature: a building construction double wall [1], an aluminium double wall [4] and a
double glass window [8]. The SEA agrees with Craik and Smiths measurements [1];
in fact, the agreement for the 50 mm and 100 mm double-walls is much better than
warranted by the uncertainty in the damping loss estimates. On the other hand, the
result for the 150 mm building construction is only quite good as are the results for
the aluminium double walls and the glass window. This might be explained by
uncertainty in the damping estimates. Also, if a minute direct coupling between the
plates is introduced, predictions are improved at frequencies a bit below the
coincidence, at which the SRI is very high, which either suggests in-accuracies in the
measurements or, more likely, the need for further improvements of the SEA model.
Finally, this successful derivation emphasises that an SEA element is an element
of response that need not be localised to one substructure of the whole structure. The
"#
successful application of SEA to a new kind of structure, therefore, requires

diagnostic measurements and diagnostic calculations so that the proper elements can
be identified.
Acknowledgement
This work was supported by the Swedish Research Council (621-2005-5754) and the
European Commission (Mid-Mod,Grant Agreement No.: 218508). The first author
thanks Magnus Lundin for stimulating discussions.
References
1. R.J.M. Craik, R.S. Smith, Sound transmission through double leaf lightweight
partitions, part I: airborne sound transmission, Applied Acoustics 61 (2000)

223-245.
2. R.J.M. Craik, R.S. Smith, Sound transmission through lightweight parallel
plates. Part II: structure-borne sound, Applied Acoustics 61 (2000) 247-269.

3. M. Villot, C. Guigou and L. Gagliardini Predicting the acoustical radiation of
finite size multi-layered structures by applying spatial windowing on infinite

structures. Journal of Sound and Vibration 286 (2005) 817-847.
4. V. Hongisto, M. Lindgren, R. Helenius, Sound insulation of double walls an
experimental parameter study, Acta Acoustica united with Acustica 88 (2002)

904-923.
5. R.J.M. Craik, Non-resonant sound transmission through double walls using
statistical energy analysis, Applied Acoustics 64 (2003) 325-341.

6. J. Brunskog, The influence of finite cavities on the sound insulation of double-
plate structures, Journal of the Acoustical Society of America 117 (2005)

3727-3739.
7. J. Wang, T.J. Lu, J. Woodhouse, R.S. Langley, J. Evans, Sound transmission
through lightweight double-leaf partitions: theoretical modelling, Journal of

Sound and Vibration 286 (2005) 817-847.
8. J-D Chazot, J-L Guayder, Prediction of transmission loss of double panels
with a patch-mobility method, Journal of the Acoustical Society of America

121 (2007) 267-278.
"#
9. Dijckmans, Vermeir, Lauriks, "Sound transmission through finite lightweight
multilayered structures with thin air layers", JASA 128(6), 2010.

10. Diaz-Cereceda, Poblet-Puig, Rodrigues-Ferran, "The finite layer method for
modelling the sound transmission through double walls", JSV 331, 2012
11. Hongisto, "Sound insulation of double panels - comparison of existing
prediction models", Acta Acustica united with Acustica 92, 2006.

12. A. Pellicier, N. Trompette, A review of analytical methods, based on the wave
approach, to compute partitions transmission loss, Applied Acoustics 68

(2007) 1192-1212.
13. R.H. Lyon, R.G. DeJong, Theory and application of statistical energy analysis,
Butterworth-Heinemann, 2nd edition, 1995.

14. Craik, R.J.M., Sound transmission through buildings using statistical Energy
Analysis, Gower press, 1996, ISBN 056607572-5.

15. A.J. Price, M.J. Crocker, Sound transmission through double panels using
statistical energy analysis, Journal of the Acoustical Society of America 47

(1970) 683-693.
16. F. J. Fahy J. Sound and Vibration 13, 171-194. Response of a cylinder to
random sound in the contained fluid.

17. M. P. Norton and A. Pruiti 1991 Applied Acoustics 33, 313-336. Universal
prediction schemes for estimating flow-induced industrial pipeline noise and

vibration.
18. S. Finnveden 1999 in F. J. Fahy and W. G. Price (Eds.) IUTAM symposium
on Statistical Energy Analysis, 289-300. Statistical energy analysis of fluidfilled pipes. Kluwer Academic Publishers.
19. M.J. Crocker, A.J. Price, Sound transmission using statistical energy analysis,
Journal of Sound and Vibration 9 (1968) 469-486.

20. P.W. Smith, Response and radiation of structural modes excited by sound,
Journal of Acoustical Society of America 34 (1961) 640-647.

21. R. H. Lyon and G. Maidanik (1962) Journal of the Acoustical Society of
America 34, 623-639. Power flow between linearly coupled oscillators.
""
22. R. S. Langley. A derivation of the coupling loss factors used in statistical
energy analysis. Journal of Sound and Vibration 141, 207-219 (1990).

23. S. Finnveden. Ensemble averaged vibration energy flows in a three element
structure. Journal of Sound and Vibration 187, 495-529 (1995).

24. P.J. Shorter, R.S. Langley. On the reciprocity relationship between direct field
radiation and diffuse reverberant loading. Journal of the Acoustical Society of

America 117, 85-95 (2005).
25. A. Le Bot. Derivation of statistical energy analysis from radiative exchanges.
Journal of Sound and Vibration 300, 763-779 (2007).

26. S. Finnveden. 2011 Journal of Sound and Vibration 330, 87-109 A
quantitative criterion validating coupling power proportionality in statistical

energy analysis.
27. A. London. Transmission of Reverberant Sound through Double Walls.
Journal of the Acoustical Society of America 22(2) (1950) 270-279.

28. C-M Nilsson and S. Finnveden 2007 Journal of Sound and Vibration 305,
641-658. Input power to waveguides calculated by a finite element method.

29. C-M Nilsson and S. Finnveden 2008 Journal of Sound and Vibration 310, 58-
76. Waves in thin-walled fluid-filled ducts with arbitrary cross sections.

30. S. Finnveden and M. Fraggstedt 2008 Journal of Sound and Vibration 312,
644-671.Waveguide Finite Elements for Curved Structures.

31. F. G. Leppington, E.G. Broadbent, K.H. Heron, The acoustic radiation
efficiency of rectangular panels, Proceedings of the Royal Society London A

382 (1982) 245-271.
32. Fahy, F. J. (2001) Foundations of Engineering Acoustics. Academic Press,
London.
33. Bies, D. A. and Hansen, C. H. (2003) Engineering Noise Control, Theory and
Practise, (3rd edn), Spoon Press, London.
"#
60
50
SRI [dB]
40
30
20
10
10
10
Frequency [Hz]
$%&'()!*+!,-'./!()/'01%-.!%./)2!-3!4'%5/%.&!0-.61(%01%-.!7%18!9!09:%1;!/)<18!-3!
*==>>+!!!?)96'()>).16@!6-5%/!5%.)!A*BC!D5/!>-/)5@!/-11)/!5%.)!A*BC!E)3!AFB@!!!!!
/968G/-1!5%.)C!.)7!>-/)5@!/968)/!5%.)+!
"#
0.1
0.1
0.08
0.08
Z (m)
Z (m)
0.12
0.06
0.06
0.04
0.04
0.02
0.02
0
0.008
0
0
0.01
10
15
X (m)
Y (m)
Fig. 2 Waveguide FE mesh for the building construction with 100 mm cavity depth.
Left, mesh used for calculations; Right, mesh used for displaying wave forms.
"#
Fig. 3 Dispersion curves for a double wall. Calculated with WFEM (large dots);
calculated for uncoupled substructures: bending waves (full curves), acoustic waves
(dashed curve); longitudinal waves in plates (dotted curves). The wave that starts
propagating around 87 Hz is the cavity-wall wave and the one around 1700 Hz is the
first oblique cavity waves. Curves for bending and longitudinal waves are double
because the plates are slightly different.
!
"#
12!
x [m]
0.5!
1!
x [m]
1.5
x [m]
0.5
1.5
x [m]
Fig. 4. Displacements of the fluid and the plates in flexural waves. Top left 2.1 Hz;
top right 2.9 Hz; bottom left, 116 Hz; bottom right, 136 Hz;
"#
Fig. 5a. Displacements of the fluid and the plates in cavity-wall wave just above the
cut-on at 87 Hz.
0.12
0.1
Z (m)
0.08
0.06
0.04
0.02
0.2
0.4
0.6
0.8
1.2
1.4
X (m)
$%&'!()'!*%+,-./01023+!45!360!5-7%8!.28!360!,-.30+!%2!/.9%3:;<.--!<.90!.3!=">!?@'!
!
"#
Fig. 6. Displacements of the fluid and the plates in oblique cavity wave just above the
cut-on at 1700 Hz.
"#
Fig. 7. Proposed SEA model for a double wall. The couplings are identified with
arrows. A) Cavity-wall element to room, Eq. (39) for r = 0 ; B) Oblique cavity
element to room, Eq. (39) for r ! 1 ; C) Room to room coupling defined by the masslaw, Eq. (12); D) Plate to room, Eq. (32); E) Plate to cavity-wall element, Eq. (34) for
r = 0 ; F) Plate to oblique cavity element, Eq. (34) for r ! 1 ; G) Plate to frames, Ref
[1].
"#
Fig. 8. Radiation efficiency, up until the coincidence frequency, as a function of the

wavenumber in the structure kb and the wavenumber in air ka . The full circle
indicates the cavity-wall wave and the solid line its evolution with frequency. The full
triangle indicates the flexural wave in a plate and the dashed line its evolution with
frequency.
"#
Fig. 9. SRI of a 50 mm building construction double wall. Measurement (full curve)

and calculation (dotted curve) both from Ref [1]; new SEA model (dashed line).
"#
Fig. 10. SRI of a 150 mm building construction double wall. Measurement (full
curve) and calculation (dotted curve) both from Ref [1]; new SEA model (dashed
line).
""
Fig. 11. SRI of a 125 mm aluminium double wall. Measurement (full curve) [4]; new
SEA model (dotted line); new SEA model with a plate to plate coupling loss factor
!c = 10"5 (dashed line).
"#
Fig. 12. SRI of a 250 mm aluminium double wall. Measurement (full curve) [4]; new
SEA model (dotted line); new SEA model with a plate to plate coupling loss factor
!c = 10"5 (dashed line).
"#
Fig. 13 SRI of a 12 mm double glass window. Measurement (full curve) [8],

calculation (dotted curve) [8]; new SEA model (dashed lines): A) wall loss factor
!w = 0.05 (as suggested in Ref. [8]); B) !w = 0.005 ; C) !w = 0.005 and with a plate
to plate coupling loss factor !c = 10"3 .
"#
Fig. 1 SRI for a 100 mm building construction double wall. Measurement (dashed
line) 5, proposed SEA model (thick line), reduced SEA model (thin line) simple
formula Eq. (38) (squared line). It is strange that Eq. (38) is 5dB out.
!
"#
Table 1 Building construction data

The superscripts 1 and 2 refer to plate 1 and plate 2.
Plasterb
Plasterboard 1,2
Chipboard 1,
oard 1,2
d = 0.1
plasterboard 2
d = 0.05
Room absorption area [m2]
Cavity internal loss factor [-]
Plate dimension [m]
Plate internal loss factor [-]
Plate Youngs modulus [GPa]
Plate Poissons ratio [-]
d = 0.15
50
From 1
50
From 1
50
From 1
4x3x0.0
4x3x0.0125 1,2
4x3x0.018 1
125 1,2
From 1
From 1
0.05
1.49 1
2.12 2
2.12 2
2.22 2
0.2 1,2
0.2 1,2
0.3 1
2.52
4x3x0.0125 2
2.52
0.2 2
Plate density [kg/m3]
710 1
710 1
678 1
806 2
806 2
801 2
Beam (quantity) [-]
10
10
10
Beam dimension [m]
3x0.05x
3x0.05xd
3x0.05xd
Beam internal loss factor [-]
0.015
0.015
0.015
Beam Youngs modulus [GPa]
7.64
8.24
6.94
1,2
Number of nails per beam [-]
11
Beam density [kg/m3]
380
11
1,2
475
"#
11 1, 6 2
469
Table 2 Aluminium
Guessed values are marked with a star.
50
Cavity depth [m]
0.125; 0.250
0.0176*
Plate dimension [m]
2.250x1.105x0.002
From 4

20
0.28
7800
Plate coupling loss factor [-]
1e-5
Table 3 Glass
Guessed values are marked with a star.
50
Cavity depth [m]
0.012
0.013*
Plate dimension [m]
1.48x1.23x0.004
0.05 (or 0.005*)
62
0.22
2500
Plate coupling loss factor [-]
1e-3
"#
Paper 2
!
International Journal of Engineering Science 63 (2013) 7183
Contents lists available at SciVerse ScienceDirect
International Journal of Engineering Science

journal homepage: www.elsevier.com/locate/ijengsci
A self-adjoint variational principle for anisotropic viscoelastic

Biots equations
Mathias Barbagallo, Svante Finnveden
The Marcus Wallenberg Laboratory for Sound and Vibration Research (MWL), Royal Institute of Technology (KTH), 10044 Stockholm, Sweden
a r t i c l e
i n f o
Article history:
Received 14 February 2012
Received in revised form 21 June 2012
Accepted 25 June 2012
Available online 17 October 2012
Keywords:
Porous materials
Biots equations
Variational principles
Dissipative systems
a b s t r a c t
A variational principle for anisotropic viscoelastic Biots equations of motion is presented.
It is based upon an extended Hamiltons principle, also valid for dissipative systems. Using
this principle, a functional analogous to the Lagrangian is defined, starting from Biots variational formulation based on frame and fluid displacements. Then, a mixed displacement
pressure formulation is presented, which reduces the number of variables of response from
six to four. The corresponding functional analogous to the Lagrangian is derived making
full use of variational calculus. The derived functionals are self-adjoint and stationary for
true motion.
! 2012 Elsevier Ltd. All rights reserved.
1. Introduction
In a seminal work Biot (1956) formulates a theory for porous elastic materials that models their macroscopical behavior
by considering them as homogenized two-phase media, i.e., the solid elastic frame and the compressible fluid in the pores.
For porous materials without dissipation, Biot gives expressions of the strain and the kinetic energies; their difference defines the Lagrangian. Biot also introduces a Rayleigh-like dissipation pseudo-potential modeling the viscous drag (Rayleigh,
1945, p. 315). Consequently, via variational calculus he devises the EulerLagrangeRayleigh equations as functions of
the frame and the fluid displacements. These equations are also referred to as the displacement (u, U)-formulation or
BiotAllards model (see Allard & Atalla, 2009). Also, Biot (1961, 1962) presents a fully anisotropic formulation.
The employment of a dissipation pseudo-potential is a common technique to handle non-conservative motion within the
framework of calculus of variations and Hamiltons principle (Riewe, 1996). As porous materials are intrinsically dissipative,
this technique appears in several works. Lopatnikov and Cheng (2002, 2004) use it to describe porous materials starting from
microscopical quantities. dellIsola, Madeo, and Seppecher (2009) use it to derive boundary conditions for fluid-permeable
interfaces in porous materials. Bedford and Drumheller (1979) use it to include microinertial effects in Biots theory. Dazel,
Brouard, Depollier, and Griffiths (2007) and Dazel and Tournat (2010) use it to define an alternative displacement formulation. Maximov (2010) combines it with Onsagers variational principle to define a generalized variational principle, which is
based upon the conservation of energy of interacting mechanical and heat displacement fields, thereby including fluid shear
relaxation in Biots theory.
Variational principles such as Hamiltons principle are based upon a self-adjoint functional that is stationary for the true
motion of the system (see e.g., Altay & Dkmeci, 2006; Fung & Tong, 2001). Zienkiewicz and Taylor (1967, pp. 7683) call
such principles natural variational principles. The variational statements mentioned above are based upon functionals that
Corresponding author. Tel.: +46 8 790 91 39.
E-mail addresses: matbar@kth.se (M. Barbagallo), svantef@kth.se (S. Finnveden).
0020-7225/$ - see front matter ! 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijengsci.2012.06.019
72
M. Barbagallo, S. Finnveden / International Journal of Engineering Science 63 (2013) 7183
do not have these mathematical properties, i.e., they are neither self-adjoint nor stationary for true motion (e.g., Biot, 1956;
Dazel et al., 2007; Lopatnikov & Cheng, 2002, 2004; Maximov, 2010). Moreover, a Rayleigh-like dissipation functional cannot
describe the constitutive relations for viscoelastic materials that are characterized by tensorial relaxation functions (see e.g.,
Finnveden & Fraggstedt, 2008; Fung & Tong, 2001).
Self-adjoint variational principles that are stationary for true motion have appealing properties. To quote Fung and Tong
(2001, p. 313): basic minimum principles exist, which are among the most beautiful of theoretical physics. Goldstein
(1980, p. 370) considers them as starting points for new formulations of mechanical structures. They are excellent concepts
for deriving equations of motions and natural coupling conditions (see e.g., Altay & Dkmeci, 2006; Biot, 1956; dellIsola
et al., 2009; Finnveden, 1994; Lopatnikov & Cheng, 2002, 2004). Fung and Tong (2001, p. 313) add that the computational
methods of solution of variational problems are the most powerful tools for obtaining numerical results in practical problems of engineering importance; e.g., variational principles constitute a sound basis for new finite element formulations,
as indicated by Finnveden and Fraggstedt (2008) and Zienkiewicz and Taylor (1967, pp. 7683). Finally, in a variational formulation, the variables of response are chosen at will and there are systematic methods for coordinate transformations; such
powerful methods are used in the present work.
Morse and Feshbach (1953, p. 298) describe a natural (albeit rather artificial) variational principle for the analysis of nonconservative motion of a viscously damped oscillator. They say, The dodge is to consider, simultaneously with the system
having the usual friction, a mirror image system with negative friction into which the energy goes which is drained from the
dissipative system. In this way the total energy is conserved and we have an invariant Lagrange function at the sacrifice of a
certain amount of reality in some of the intermediate results (see also Riewe, 1996; Schechter, 1967). The approach appears
a few times in the vibroacoustics literature for more complex problems. Gladwell (1966) employs it for acousto-structural
problems, with the fluid motion described by its displacement. Morse and Ingard (1986, pp. 252256) apply it to a granular
solidfluid medium, where the fluid is viscously impeded in a rigid porous solid. Finnveden extends the principle to cover
also viscoelastic materials having stressstrain relations described by relaxation functions: Finnveden (1994) uses a simple
relaxation function for beam structures, while Finnveden and Fraggstedt (2008) use a general relaxation function when formulating new finite elements.
Inspired by Morse and Feshbachs approach, this work presents a self-adjoint variational principle for dissipative porous
media that is stationary for viscoelastic Biots (1961) equations. This extended Hamiltons principle is based upon a functional that is an analogy to the Lagrangian for the displacement formulation.
Atalla, Panneton, and Debergue (1998) and Atalla, Hamdi, and Panneton (2001) derive a mixed (u, p)-formulation, employing the frame displacement and the pore fluid pressure as field variables, thereby reducing the number of variables from six
to four. This formulation is derived via a variable substitution in Biots equations and a subsequent weak formulation of these
equations. Herein, starting from the analogy to the Lagrangian for the displacement formulation, via a coordinate transformation, the Lagrangian for the mixed formulation is derived. The new functional is self-adjoint, symmetric and stationary for
true motion.
The article is organized as follows. In Section 2 linear viscoelastic solid and Biots theory are recalled; the theories are then
combined and a self-adjoint variational principle for dissipative porous materials is formulated, which is the articles most
important theoretical result. The analogy to the Lagrangian requires twice as much ink as the Lagrangian for conservative
motion. This inconvenience disappears for time-harmonic motion, as seen in Section 3. The power of variational methods
is then used for a variable transformation resulting in the new mixed (u, p)-formulation. It is expressed by tensor algebra
and, to facilitate numerical implementation, in Section 4 it is also expressed by matrix algebra; this may be the articles most
important practical result. Dazel, Sgard, Becot, and Atalla (2008) report that it is difficult to identify strain and kinetic energies of the fluid and solid phase in Atalla et al.s (2001) formulation: in Section 4 this observation is commented and the variational approach is applied to a fluid, exemplifying what derived earlier. Finally, Section 5 concludes the article.
2. Anisotropic elastic and viscoelastic solids and porous materials
The constitutive relations and the expressions for kinetic and strain energies of viscoelastic and elastic solids and of porous materials are first recalled. Then, the equations of motions of a porous material are derived both employing Biots original approach and an extended Hamiltons principle. The analysis in this section is conducted in the time domain. The
formulation for time-harmonic motion is less complex, as shown in Section 3.
2.1. Elastic and viscoelastic solids
2.1.1. Stressstrain relation for viscoelastic and elastic solids
The linear relation between the second-order strain tensor
!ij
1
ui; j uj; i :
2
!ij and the first order displacement tensor ui in a solid is

1
Cartesian tensor notation with the summation convention is employed; a comma in the subscript indicates a partial spatial
derivative. For linear viscoelastic materials, the stress tensor rrij and the strain tensor are related by (Fung & Tong, 2001, chap. 15)
"
rrij xi ; t Drijkl !rkl xi ; t $
%1
Grijkl xi ; t % s
@ !rkl
xi ; s ds;
@s
73
where the superscript r indicates a linear viscoelastic solid, xi refers to the cartesian axes and Grijkl is the tensorial relaxation
function of the material. The admissible functions in Eq. (2) and their mathematical requirements are discussed by Gurtin
and Sternberg (1962).
Three possible choices for finding stressstrain relations in accordance to Eq. (2) are referenced. One choice is Biots hidden coordinates theory (see Biot, 1954; Fung & Tong, 2001, chap. 13). In irreversible thermodynamical systems the number
of observed variables is, in general, small compared to the total number of degrees of freedom; Biot derives a relationship
between observed variables such as stress and strain in which the influence of the hidden coordinates appears in the hereditary character of the material. Another choice is the augmented Hookes law presented by Dovstam (1995), which is based
upon hidden coordinates . Finally, a favorite of the authors is the fractional derivative model proposed by Bagley and Torvik
(1983), based upon the differential form of the stressstrain relation (see Gurtin & Sternberg, 1962, Section 4). These models
may be used to fit measurements of tensorial relaxation functions. Examples of such measurements are provided by
Dovstam (2000) and Jaouen, Renault, and Deverge (2008).
For elastic materials Eq. (2) is simplified to the generalized Hookes law,
rrij xi ; t Grijkl xi ; 0!rkl xi ; t Drijkl !rkl xi ; t;

where
Drijkl
is the elastic modulus tensor of the material.
2.1.2. Strain energy for elastic solids

In elasticity, it is common to define a strain energy functional W, which is a potential having the property that
@W
rij :
@ !ij
Therefore, for a linear elastic solid Wr is given by
Wr
1
2
rrij d!rij !rij Drijkl !rkl ;
plus a constant that is of no relevance.

The postulate of a strain energy functional is a standard step in the derivation of variational principles (Fung & Tong, 2001,
pp. 331, 380). Besides, demonstrations of Hamiltons principle rely, at least implicitly, on the fact that the strain energy density is a potential such that Eq. (4) is true.
2.2. Biots theory for elastic porous materials
In this subsection Biots derivation of the equations of motions is recalled. Anisotropy is included (see also Biot, 1961,
1962; Hrlin & Gransson, 2010). For a conservative, isotropic and elastic porous material, Biot postulates the total strain
and kinetic energies. The strain energy density for anisotropic elastic porous materials is given by (Biot, 1956, Eq. (2.6))
2W !ij rij es;
where rij is the solid stress in the frame, e is the fluid dilatation defined as
e $ U i;i :
Ui is the fluid displacement and s is a stress variable introduced by Biot as
s $ %hp:
h is the porosity of the material and p is the pore fluid pressure. For elastic and conservative materials, the strain energy is a
potential and the stresses are given by
rij
@W
;
@ !ij
@W
:
@e
Biot postulates the stressstrain relations as
rij Dijkl !kl Q ij e;

s Re Q ij !ij ;
10a
10b
where Dijkl is the elastic modulus tensor of the solid phase, Qij is the dilatation coupling tensor and R is the effective fluid
compressibility. The extension to anisotropic conservative materials of Biots kinetic energy density reads (Biot, 1956):
_
_
2K qs u_ 2i qf U_ 2i % u_ i % U_ i qsf
ij uj % U j ;
11
74
where the over dot indicates a time derivative, qsf

ij is the mass coupling tensor between the solid and the fluid. The actual
densities of the solid and fluid phases are given by:
qs 1 # hqs0 ;
qf hqf0 ;
12
qs0 and qf0 are the intrinsic densities of the solid and of the fluid.
The Lagrangian of the conservative system is defined as
L%
W # K dV;
13
where V is the volume occupied by the porous material.

Porous material are dissipative; part of the dissipation is due to the viscous drag, which is introduced by Biot via a Rayleigh-like dissipation pseudo-potential density:
2Rd u_ i # U_ i bij u_ j # U_ j ;
14
where bij is the viscous drag tensor. Dissipation is also caused by viscoelasticity: since Biots variational procedure cannot
include it, viscoelasticity is left to the next subsection. Biot (1961, 1962) introduces viscoelasticity not in the functionals
but in the equation of motion, based upon the isomorphism between elastic and viscoelastic theories.
Eventually, the (u, U) EulerLagrangeRayleigh equations and boundary integrals are derived from
Z Z
@Rd
dui dt dV 0;
@ u_ i
Zt
Zt ZV
@Rd
d L dt
dU i dt dV 0:
_
t
t
V @Ui
L dt
15a
15b
The functionals in Eq. (15) are neither self-adjoint nor stationary for true motion; besides, they cannot include viscoelasticity.
In the next subsection, a variational principle for anisotropic dissipative porous materials that has these mathematical
properties and that can include viscoelasticity is defined.
2.3. A variational principle for viscoelastic porous materials
The constitutive relations of a viscoelastic porous material are defined:
rij Dijkl !kl Qij e;

s Re Qij !ij :
16a
16b
Dijkl is the viscoelastic tensor operator for the solid phase and is defined in a similar fashion as Eq. (2); similar tensorial relaxation functions also define the dilatation coupling operator Qij and the equivalent fluid compressibility operator R.
The extended Hamiltons principle herein presented is a variational principle that can handle dissipative systems and that
is based upon constructed self-adjoint bilinear functionals. These functionals are called strain and kinetic potentials. Then, an
analogy to the Lagrangian can be defined.
Generally, the operator DA that is adjoint to D satisfies the following relation
uDv dt
v DA u dt;
17
for any function u and v. If the Fourier transforms of u and v exist, the operator DAijkl that is adjoint to Dijkl in Eq. (16a) is given
by Finnveden and Fraggstedt (2008):
DAijkl u %
#Gijkl xi ; s # t
@ !kl
xi ; s ds:
@s
18
A variational principle needs a potential. The strain potential density is defined by the following obviously self-adjoint
bilinear functional:
c !A Dijkl !kl !ij DA

4W
ijkl
ij
!Akl
"
!Aij Qij e !ij QAij eA eA Re eRA eA eA Qij !ij eQAij
# $
!Aij ;
19
where !Akl and eA are the solid strain and fluid dilatation of the adjoint system. It equals in all aspects the original system,
except that its constitutive relations are adjoint to those in Eqs. (16):
"
rAij DAijkl !Akl QAij eA ;
20a
s R e
20b
QAij
# $
!Aij :
c has the property that:

The strain potential density W
75
Z !
c dt 1
W
2
"
rij d!Aij rAij d!ij sdeA sA de dt
1
2
!
c
c
c
c
@W
@W
@W
@W
A
A
d
!
d
!
de
de
dt:
ij
@eA
@e
@ !Aij ij @ !ij
21
c is therefore a potential, similar to Eq. (4). Instead of the kinetic energy plus the dissipation pseudo-potential in Eqs. (11)
W
and (14), a kinetic potential density including the viscous drag is defined:
!
"
!!
"
!
""
b q u_ i u_ A q U_ i U_ A % u_ A % U_ A qsf u_ j % U_ j % 1 uA % U A bij u_ j % U_ j % ui % U i bij u_ A % U_ A :
2K
s
f
i
i
i
i
ij
i
j
j
i
2
22
The proposed variational principle, or extended Hamiltons principle, states that
b
L dt d
Z Z
t
c%K
b dV dt 0:
W
23
The functional b
L is an analogy to the Lagrangian and is a scalar quantity that is conserved for damped motion; for brevity, the
term Lagrangian is used.
The variational procedure implied by Eq. (23) is made explicit:
b
L dt
Z Z
c
c
c
c
b
b
b
b
b
@W
@W
@W
@W
@K
@K
@K
@K
@K
d!ij A d!Aij
du_ i % A du_ Ai %
dU_ i %
dU_ Ai %
dui
de A deA %
@ !ij
@e
@e
@ u_ i
@ui
@ !ij
@ u_ i
@ U_ i
@ U_ Ai
!
b
b
b
@K
@K
@K
dU i %
dU Ai dV dt:
% A duAi %
@U i
@ui
@U Ai
t
24
Integrations by parts and the symmetry of the stress and strain tensors yield (Fung & Tong, 2001, p. 333):
b
L dt %
Z Z
duAi
"
rij; j % qs ui qsfij uj % U j % bij u_ j % U_ j dV dt Bb s %
Z Z
dui
t
V
t
V
Z Z
!
"
!
""
i % qsf u
bf
b s;A %
_ j % U_ j dV dt B
dU Ai s;i % qf U
%
U
bij u_ Aj % U_ Aj
dV dt B
j
j
ij u
ij
t
V
Z Z
!
"
A % qsf u
A % bij u_ A % U_ A dV dt B
b f;A 0:
Aj % U
%
dU i sA;i % qf U
i
ij
j
j
j
t
rAij; j % qs uAi qsfij uAj % U Aj
"
25
Inside the brackets, Biots equations are recognized. The boundary terms in Eq. (25) are given by:
bs
B
bf
B
Z Z
Z t ZS
t
nj duAi rij dt dS;

ni dU Ai s dt dS;
Z Z
b s;A
B
nj dui rAij dt dS;
Z tZ S
b f;A
B
ni dU i sA dt dS:
t
26a
26b
The positive direction of the normal vector nj points outward the surface S and the same convention is used throughout.
Applying the fundamental lemma of calculus of variations to Eq. (25), the EulerLagrange equations of the solid and fluid
phase of the original and adjoint systems follow; those of the original system are retrieved varying the adjoint generalized
coordinates and appear in the parentheses in the first and third line of Eq. (25); they equal those derived from Eq. (15) and
those by Biot (1956) and by Allard and Atalla (2009).
The functional b
L is self-adjoint and stationary for true motion. It defines a variational principle in Eq. (23) that applies for
general viscoelastic motion. Eq. (23) is the most important theoretical result of this work.
3. Harmonic motion of anisotropic viscoelastic porous materials
The Lagrangian b
L requires almost twice as much ink as the conservative Lagrangian and dissipation pseudo-potential. This
disadvantage disappears for time-harmonic motion as is shown in what follows. Firstly, the time-harmonic Lagrangian in the
displacement formulation is defined. Then, the time-harmonic Lagrangian in the mixed formulation is obtained via a transformation of coordinates.
3.1. A variational principle for harmonic motion
Time-harmonic motion at angular frequency x is considered; accordingly, the frame displacements are given by
$
1 # ixt
~i e u
~ &i e%ixt ;
u
2
$
1 # A ixt
~i e u
~ai e%ixt ;
uAi u
2
ui
27a
27b
76
~ i is the complex amplitude of the frame displacement, u

~ !i is the complex conjugate and the upper index a is introwhere u
! A "!
~ ai u
~ i . All other variables are expressed in a similar time-harmonic form. The time average of the kinetic
duced so that u
potential density in Eq. (22) is expressed by
1
T
T=2
#T=2
b dt 1 K K ! ;
K
8
28
where T is the period time and K reads
where
e a 2 ~ fc e
e a 2 ~c ~
~ ai x2 q
~a 2 ~ c e
~ sc
~
Ku
ij uj ui x qij U j U i x qij U j U i x qij uj ;
q~ scij qs dij # q~ cij ;
q~ fcij qf dij # q~ cij ;
q~ cij qsfij #
bij
ix
29
30
and dij is Kroneckers delta. The time average of the strain potential density in Eq. (19) is expressed by (see Appendix A for
further details):
1
T
T=2
c dt 1 W W ! ;
W
8
#T=2
31
where W reads
e ~ ~a e ~ ~a e ~
e ijkl !~kl ~!a Q
~aij D
W !
ij ij e e R e e Q ij !ij :
e ijkl may include viscoelasticity and is given by

The complex rigidity tensor D
e ijkl
D
#1
#ix Gijkl seixs ds:
32
33
e ijkl is a tensor analogy to the use of a complex Youngs modulus. The complex fluid compressibility
For any given frequency D
e ij are defined in a similar fashion.
e and the anisotropic complex dilatation coupling parameter Q
parameter R
It suffices to consider the first term on the right-hand sides of Eqs. (28) and (31), since the functional built by the second
term is just the complex conjugate of the functional built by the first term, and if the variation of one of these is zero so is the
other. Also, the common factor 18 in these equations is dropped for brevity (Finnveden & Fraggstedt, 2008).
Upon this basis, the variational principle for harmonic motion reads
dL d
W # K dV 0;
34
L is called Lagrangian, W and K strain and kinetic potential densities; four times their real part defines the time-averaged
c and K
b.
potentials W
3.2. Mixed (u, p)-formulation
The Lagrangian is invariant to coordinate transformations. In order to switch from (u, U) coordinates to (u, p) coordinates,
Lagrange multipliers are employed. Specifically, to eliminate the dependance of L on the fluid displacement, the fluid dilatation is introduced as an additional variational parameter. A new functional is defined:
!L ' L # C;
35
where C expresses a constraint via the Lagrange multipliers k and ka
Z #
V
#
$$
e i;i k ~ea # U
ea
ka ~e # U
dV:
i;i
36
Varying the fluid dilatations in Eq. (35), it follows that
e ij ~!ij ;
e ~e Q
kR
e ij !~a :
e ~ea Q
ka R
ij
37
Varying the fluid displacements it follows that
e
~ cij u
~ j x2 q
~ fc
k;i x2 q
ij U j 0;
ea
~ cij u
~ aj x2 q
~ fc
ka;i x2 q
ij U j 0:
38
For time-harmonic motion, Eqs. (37) and (38) respectively equal the constitutive relation of the fluid-phase and its equation
of motion in the displacement formulation, Eqs. (10b), (25). Thus,
k ' s:
39
77
Upon this basis, the fluid displacements, the fluid dilatations and the Lagrange multipliers are eliminated from !L using Eqs.
(37)(39). Eventually, !
L reads
!L
e ijkl u
~ ai; j D
~ k;l " u
~ ai;j
u
where
~ sc
ij
!
2 fc
2
e ij Q
e kl
^ ij
h q
Q
h
a ~ ~
a~ ~
!
~ k; l " u
~p
~a;i
~ ai x2 q
~j " p
~a p
~
~
~
~ ij u
u
u
c
c
p
p
p
u
ij i;j dV F;
i; j ij
e
e
x2 ;j
R
R
!
e ij
Q
c~ij h q q "
;
e
R
~c
~ cik ^ fc
kl lj ;
~ cik ^ fc
kj
q~ ij q " q q q
q~ fcik q^ fckj dij :
40
41
The surface functional F! in Eq. (40) is given by
F! "
~a
ni p
2 fc
^ ij
h q
x2
~; j dS "
p
~a; j
ni p
2 fc
^ ij
h q
x2
~ dS:
p
42
Its origin is explained. Eq. (36) contains a spatial derivative of the fluid displacements Ui; therefore, the elimination of Ui via
Eq. (38) yields a second-order spatial derivative of the Lagrange multipliers. For !
L not to include second-order derivatives, an
!
integration by parts is made: this operation yields F.
!
L is self-adjoint and defines a variational principle. Differences to the weak forms by Atalla et al. (1998, 2001) and Hrlin
and Gransson (2010) are commented in next section.
4. Additional remarks
In this section some additional remarks and analyses are presented. First, the derived functionals are compared to results
in the literature. Second, for a fluid the coordinate is transformed from displacement to pressure to illustrate some of the
results. Third, the functional !L is expressed in matrix algebra for isotropic porous materials ready for computer
implementation.
4.1. Discussion of Section 3
The Lagrangian for time-harmonic motion L is derived in Section 3.1 and expresses the difference between strain and kinetic potentials. Eq. (34) produces the EulerLagrange equations and the boundary integrals (compare to Eq. (25) and to:
Allard & Atalla, 2009; Biot, 1956; Panneton & Atalla, 1996). Then, in Section 3.2 Lagrange multipliers are employed to transform the coordinates of L from (u, U) to (u, p), thus obtaining the Lagrangian in the mixed formulation !L. The corresponding
EulerLagrange equations are derived in Appendix B and equal those by Hrlin and Gransson (2010) and, for an isotropic
material, those by Atalla et al. (1998).
!
L differs in form from the mixed weak forms in the literature (e.g., Atalla et al., 1998, 2001; Hrlin & Gransson, 2010). !L is
symmetric in all terms, whereas the earlier weak forms have asymmetric surface terms; these surface terms often disappear
when couplings to other media are considered, thus yielding a symmetric weak form (Atalla et al., 2001; Hrlin & Gransson,
2010). Symmetric formulations have great numerical and theoretical advantages: computational cost is reduced, numerical
stability is gained, more numerical algorithms are available, generalized forces are easily identified, the principle of reciprocity is directly evident.
The Lagrangian !
L of the here proposed mixed formulation includes volume and surface functionals. In a finite element
formulation the surface functionals need to be included except at interfaces between elements that have equal material
properties, at which the elements surface integrals cancel. When studying boundary and couplings conditions and equations
of motion, the variation of the volume functional in !L produces, as in Eq. (25), volume integrals, in which the equations of
motion could be recognized, and surface integrals. These surface integrals together with the variation of the surface function! define coupling and boundary conditions. This is studied in Appendix B. There it is shown that, once essential
als in !
L, i.e., F,
conditions are imposed, the surface integrals define natural coupling conditions, which are those that are derived by Allard
and Atalla (2009) and Atalla et al. (2001). In Appendix B it is also shown that the Lagrangians in the two formulations satisfy
these known boundary and coupling conditions in the same fashion. This is the main practical difference between the here
proposed Lagrangians and the weak forms in the literature, which require investigations of these conditions on a case by case
basis for each formulation (see e.g., Atalla et al., 2001; Panneton & Atalla, 1996).
!
L expresses the difference between strain and kinetic potentials when C 0, Eq. (35). If the fluid displacements and dilatations in L are expressed by Eqs. (38) and (37), the strain and kinetic potential densities read:
e ijkl u
~ ai; j D
~ ai; j
~ k; l " u
W u
2
~ ai x2 q
~a; i
~u
~i p
Ku
2
e ij Q
e kl
h
Q
~ k; l p
~;
~a p
u
e
e
R
R
^ fc
h q
ij
x2
~; j :
p
43a
43b
78
The difference of the two potentials does not express the Lagrangian: the terms containing both pressure and displacement
are missing and the correct equations of motion cannot be retrieved. The missing terms appear if Eq. (35) is employed.
Dazel et al. (2008) present the expressions of the energies in the mixed formulation in a porous material, distinguishing
between solid and fluid phases . This distinction is beyond the scope of the present article. However, the functional !
L is consistent with the expressions by Dazel et al. (2008).
In the next section, via the same variational machinery the coordinates of a fluid are transformed from displacement to
pressure.
4.2. Fluid Lagrangian
When the motion of a fluid is described by its displacement, the strain and kinetic potential densities in Eqs. (32) and (29)
in Section 3.1 simplify to
ee U
ee ;
e a;e R
We U
i;i
i;i
e e;
e a;e x2 qe U
Ke U
i
i
44
e e is the fluid displacement, R

e e qe c2 and qe the fluid denthe superscript e indicates that these quantities describe a fluid: U
sity. The Lagrangian is defined as the difference between these two potentials:
Le
W e $ K e dV:
45
Its first variation is stationary for the true motion. Now, Lagrange multipliers ke and ka,e are employed to transform the coordinates from displacement to pressure; the fluid dilatation is introduced as an additional variational parameter to eliminate
the fluid displacement from Le . A new functional is defined:
Z ! !
"
!
""
e e ke ~ea;e $ U
e a;e dV:
ka;e ~ee $ U
i;i
i;i
!Le Le $
46
The original and adjoint fluid displacements and dilatations are varied independently, obtaining the following relations:
e e ~ee ;
ke R
ke;i
e e ~ea;e ;
ka;e R
ka;e
;i
e e;
$x q U
i
2 ~e
It follows that
47
e a;e :
$x q U
i
2 ~e
48
~e :
ke & $p
49
Using Eqs. (47)(49) the fluid displacements, dilatations and Lagrange multipliers are eliminated from !Le; integrating by
parts all the terms featuring a second-order spatial derivative, the functional !
Le is rewritten as
!Le
$
!eK
! e dV F!e
$W
The surface functional F!e reads
F!e $
~a;e
ni p
~e dS $
p
e ;i
x2 q
&
Z %
1 e
1
~a;e p
~a;e
~ p
~e;i dV F!e :
$p
p
;i
2
e
ee
xq
V
R
~a;e
ni p
;i
x2 q e
~e dS;
p
50
51
which is similar to F! in Eq. (42). F!e is an analogy to twice the work at the boundary. Integration by parts shows that
F!e 2
!e$K
! e dV:
W
52
Substituting this result into Eq. (50), it follows that
!Le
!e$K
! e dV:
W
53
Eqs. (52) and (53) explain the change of sign of the strain and kinetic potentials in Eq. (50) compared to Eq. (45). The potentials of the fluid-phase of the porous material switch signs too in the mixed formulation, as the fourth and fifth terms in Eq.
(40) show. This switch also happens to the Lagrangian of a fluid within a cylinder derived by Finnveden (1997).
4.2.1. Fluid couplings to other media
The extended Hamiltons principle is applied to Le and !
Le to derive the EulerLagrange equations and the boundary integrals: via these integrals the couplings to other media are studied. The variation of the adjoint fluid displacement in Le yields,
in analogy to Eq. (25)
ee U
e e x2 qe U
e e 0;
R
i;ii
i
Z
Z
a;e
e
e R
e a;e p
ee U
e e dS $ ni d U
~e dS 0:
B
ni d U
i;i
i
i
S
54a
54b
79
The variation of the adjoint pressure in !Le yields:
1 e
1
~ 2 p
~e 0;
p
e
e
x
qe ;ii
R
Z
Z
1
e a;e p
! e % ni dp
~a;e
~e dS % ni d U
~e dS 0:
B
p
;i
i
2 e
xq
55a
55b
Notice that Eqs. (54b) and (55b) are equal.

~e couples to another fluid with the same material properties there is no need for F!e . If, however, it
If a fluid described by p
couples to another fluid that has a different density or if it couples to a porous material (see Appendix B), the surface func~ ri , the
tionals need be evaluated for both media. Finally, if the fluid is connected to a solid described by its displacement u
kinematic condition that the displacements in the direction normal to the interface are equal need be imposed. Thus, on
the interface it holds that
e e ni
ni U
i
~e;i
p
qx
e
~ ri ;
ni u
e a;e ni
ni U
i
~a;e
p
;i
qe x2
~ a:r
ni u
i :
56
The surface functional F!e is part of the Lagrangian and is needed to couple a fluid to a solid. F!e directly arises from the variational machinery and makes the boundary integrals in the displacement and pressure formulation equal. Appendix B de! defined in Eq. (42).
rives similar results for F,
4.3. Functional L for isotropic media
If the porous material is isotropic, the functional !
L simplifies. The following quantities are introduced:
~ij bdij ;
e ij Q dij ;
Q
b
q~ sfij q12 dij ;
!
"
!
"
b
b
q~ cij q12 %
dij ;
q~ scij q11
dij ;
ix
ix
!
"
b
d
q~ fcij q22
dij ;
q^ fcij # ij b $ ;
ix
q22 ix
57a
57b
57c
where q12, q11 = qs % q12 and q22 = qf % q12 are introduced in accordance with Biots (1956) notation. Accordingly, the
parameters in Eq. (41) become:
"
q12 % b=ix Q
%
d ;
q22 b=ix R ij
q % b=ix2
q~ ij qt dij q11 b=ixdij 12
d :
q22 b=ix ij
c~ij cdij h
58a
58b
Matrix algebra is more convenient than tensor algebra for computer implementation. Thus, the displacement vector
u = [u v w] is substituted for the displacement tensor ui, where u, v and w are the three components for i = 1, 2, 3 respectively.
The engineering strain vector is introduced:
! !x !y !z !xy !xz !yz 'T ;
59
where
@u
@v
@w
!y ;
!z ;
;
@x
@z
@y
@u @ v
@u @w
!xz ;
;
@y @x
@z @x
!x
!xy
60
!yz
@w @ v
:
@y @z
61
The frame dilatation is also introduced:
n !x !y !z :
It follows that !
L in Eq. (40) simplifies to:
!L
F! %
!
2
2
Q2
h
a
aT 2
ah
aT
a
!
$ rp cn p p n dV F;
! D! % n
n % u x qt u % p
p rp #
R
R
q22 ixb x2
aT
aT
62
63a
nT #
h
$ pa rp rpa p dS:
q22 ixb x2
63b
e ijkl so that the engineering stress vector r = D!. Eq. (63) are symmetric and
The entries of the stiffness matrix D are given by D
ready to be implemented in a computer.
80
5. Conclusion
A variational principle for anisotropic viscoelastic Biots equations is presented. This principle is based upon a constructed
self-adjoint functional, stationary for true motion, which is an analogy to the Lagrangian. This approach is inspired by Morse
and Feshbachs (1953) construction for a viscously damped oscillator, and adds to other works in vibroacoustics that are similarly inspired (e.g., Finnveden, 1997, 1994; Finnveden & Fraggstedt, 2008; Gladwell, 1966; Morse & Ingard, 1986).
The (analogy) to the Lagrangian is first defined as a functional of frame and fluid displacements in the time domain, where
relaxation functions describe viscoelastic constitutive relations. Then, time-harmonic motion is considered, for which the
hereditarian characteristics of the material are described with complex rigidities, which simplifies the notation considerably.
A mixed frame displacementfluid pressure (u ! p) formulation is derived via a coordinate transformation. It follows, more
or less automatically, upon the introduction of fluid dilatation as an additional variable with the help of Lagrange multipliers.
The EulerLagrange equations that follow upon these formulations agree with earlier works (e.g., Atalla et al., 1998; Biot,
1956; Hrlin & Gransson, 2010).
The natural boundary conditions for a porous material require no frame stress and no fluid stress. This applies equally for
the displacement formulation and the mixed formulation; so, any analysis on couplings applies equally for both formulations. This appears to be an original result as previous analyses are made on a case by case basis for the various formulations
(see e.g., Atalla et al., 2001; Panneton & Atalla (1996)).
The strength of variational principles is further exemplified by changing the generalized coordinates of a fluid from displacement to pressure. The pressure Lagrangian contains volume and surface terms and needs no additional coupling functional for
coupling to a solid. This approach might help handling not obvious couplings in other cases of fluidstructure interaction.
A functional analogous to the Lagrangian is a scalar quantity that is conserved for damped motion and expresses the difference
of strain and kinetic potentials. For Biots theory, these potentials are consistent with expressions of energies by Dazel et al.
(2008). Previous works on variational formulations of Biots equation often employs a Rayleigh-like pseudo potential to describe
the viscous fluid-structure coupling (e.g., Bedford & Drumheller, 1979; Biot, 1956; Dazel et al., 2007; dellIsola et al., 2009; Lopatnikov & Cheng, 2002, 2004). The present work offers another possibility to define a variational principle for materials described
by Biots equation. It applies equally for the displacement (u ! U) and mixed (u ! p) formulations. It attributes not only viscosity
but also viscoelasticity. Most important, it employs functionals that are self adjoint and are stationary for true motion.
It is occasionally critiqued that Morse and Feshbachs construction sacrifices a certain amount of reality (e.g., Riewe, 1996;
Schechter, 1967). Here, however, it is demonstrated that this construction can provide useful results.
Acknowledgments
This work was supported by the Swedish Research Council (621-2005-5754) and the European Commission (Mid-Mod,
Grant Agreement No.: 218508). The stimulating discussions with Peter Gransson, Nils-Erik Hrlin and Hao Liu have been
most helpful.
Appendix A. Time averages of the strain potential
The first two terms of Eq. (19) are considered.
b !A Dijkl !kl !ij DA

P
ijkl
ij
!Akl
"
1
T
T=2
!T=2
P is given by:
!Aij Dijkl !kl !ij DAijkl !Akl
"$
1
dt P P& :
8
&
!
"
~kl e!ixs ds dt
Gijkl t ! s !ix!
!T=2
!1
%Z 1
&
Z T=2
!
"
1
!i
x
!~ij e t
!Gijkl s ! t ix~!akl eixs ds dt
2T !T=2
%Zt t
&
Z T=2
1
a
i
x
t
!i
x
t
~kl Gijkl t ! se!ixs!t ds dt
!~ij e e
!ix!
2T !T=2
!1
%
&
Z T=2
Z 1
1
!i
x
t
i
x
t
!~ij e e
!ix!~akl Gijkl s ! teixs!t ds dt
2T !T=2
t
%Z 1
&
Z
1 a T=2
~
!ij
!ixGijkl seixs ds !~kl dt
2T
!T=2
!1
&
Z T=2 %Z 1
1
~
!ij
!ixGijkl seixs ds !~akl dt
2T
!T=2
!1
1 ae
1 e
e ijkl !
~kl ~!ij D ijkl !~akl !~aij D
~kl :
!~ij D ijkl !
2
2
1
2T
T=2
~!aij eixt
%Z
A:1
b is calculated:
The time average of P
A:2
A:3
81
The third equality follows from the fact that the tensorial relaxation function Gijkl is zero for negative arguments, as the moe ijkl is given by the Fourier transform of Gijkl,
tion is causal. The complex rigidity modulus D
e ijkl
D
"1
"ixGijkl seixs ds:
A:4
Appendix B. Validation of the proposed formulations

A variational principle, such as d!L = 0 where !L is defined by Eq. (40), may be the basis of a finite element formulation as
described in Zienkiewicz and Taylor (1967, pp. 7683). That statement also leads to equations of motion and boundary conditions (e.g., Fung & Tong, 2001, p. 310). Boundary conditions are either essential or natural. Essential conditions may either
express the continuity of the variables of response, or express an additional requirement on them; e.g., in porous materials
the conservation of volume flow is one such requirement. Natural conditions express conditions involving derivatives of the
field variables. In a finite element procedure, essential conditions expressing continuity are fulfilled in the assembly process,
those expressing additional constraints may be introduced with Lagrange multipliers; once all essential conditions are imposed, natural conditions are automatically satisfied.
The equations and conditions for porous materials are known from earlier works (e.g., Allard & Atalla, 2009; Atalla et al.,
1998; Biot, 1956; Hrlin & Gransson, 2010; Panneton & Atalla, 1996). The variational procedure is now used, instead, to
validate the proposed variational statements. First, the displacement formulation is investigated. The constraint on volume
flow is introduced via Lagrange multipliers. Then, the EulerLagrange equations and boundary integrals for a single homogenous porous material are derived in the mixed formulation; it results in that the boundary integrals equal those in the displacement formulation. This result is probably new in this context. A discussion on the couplings in the mixed formulation
concludes the appendix.
B.1. Couplings in the displacement formulation
A variational principle yields the natural coupling conditions once all essential conditions are imposed. If a porous material is structurally attached to another media, e.g., a solid, the frame displacement is continuos across the interface. The continuity of the volume flow is also required. This constraint is typically expressed via the continuity of the normal total
~ ti across the interface, where u
~ ti is defined as
displacement u
e i:
~i h U
~ ti % 1 " hu
u
B:1
dL Lm T 0;
B:2
The variational statement expressing the coupling between a porous material and another media reads:
~ ti introduced via Lagrange multipliers, b

where Lm is the Lagrangian of the coupled media and T expresses the continuity of ni u
and ba in the following.
If a porous material is attached to a solid, the essential conditions are the continuity of the frame and solid displacement
~ ri , and of normal total and solid displacements at the interface. The latter condition is introduced as
u
T ps "
! t
"
~ ri dS "
~i " u
ni ba u
! a;t
"
~i " u
~ a;r
ni b u
dS:
i
B:3
e e , the essential condition on the conservation of volIf a porous material is coupled to a fluid described by its displacement U
i
ume flow is introduced as
T pf "
Z
#
$
#
$
e e dS " ni b u
e a;e dS:
~ ia;t " U
~ ti " U
ni ba u
i
i
B:4
Z
#
$
#
$
t;2
~
~ a;t;1
~ a;t;2
~ t;1
ni ba u
"
u
ni b u
"u
dS
"
dS:
i
i
i
i
B:5
If two porous materials are structurally connected and are not separated by a jacket, one essential condition expresses the
continuity of the frame displacements; the other expresses the continuity of normal total displacement:
T pp "
Thus, Eq. (B.2) yields the EulerLagrange equations and, imposing the essential conditions on the continuity of the field variables, the natural coupling conditions follow. For solidporous couplings, the natural condition expresses the balance of the
total stress and the solid stress; for fluidporous couplings, it expresses the balance of the pore and fluid pressures, and the
balance of the total stress and fluid stress; for porousporous couplings, it expresses the balance of the two total stresses, and
of the two pore pressures. It appears that the Lagrange multiplier b (ba) equals minus the (adjoint) pore pressure.
82
B.2. EulerLagrange equations and boundary integrals in the mixed formulation

The EulerLagrange equations and the boundary integrals are derived from the functional !
L, Eq. (40). Proceeding as in Eq.
(25), the variations of the adjoint frame displacement and pressure in !L lead to the time-harmonic EulerLagrange equations
in the (u, p)-formulation:
r~ 0ij;j x2 q~ ij u~j c~ij p~; j 0;
^ fc
hq
ij
x2
~; ij
p
h
c~ ~
~ # ij u
p
i; j 0:
e
h
R
B:6
r~ 0ij is the in-vacuo frame stress tensor, which is also expressed in terms of the total stress tensor r~ tij :
!
e e
Q Q
Q
r~ 0ij De ijkl # ije kl ~!kl r~ tij h dij eij p~;
R
r~ tij r~ ij ~sdij :
B:7
The variation of !
L also leads to the boundary integrals:
! sf
B
Z
2 fc
!
"
^ ij
h q
~ aj r
~a; j
~ 0ij cij p
~ dS # ni dp
~ dS 0:
ni du
p
2
S
B:8
! sf is calculated summing the surface integrals resulting from F! Eq. (42), and those from the integrations by parts of the volB
ume integrals in !L. Eqs. (B.6) equal those by Hrlin and Gransson (2010) (for isotropic media see Atalla et al. (1998)).
Eq. (B.8) is further simplified when the pressure gradient is evaluated. This is done utilizing the equation of motion of the
fluid phase in the displacement formulation, here rewritten as
2 fc
^ ij
h q
x2
ei hq
~;j h U
~
~ cik q
^ fc
p
kj uj :
B:9
Employing also Eqs. (B.7), Eq. (B.8) can be rewritten as:
~ ij dS
~ aj r
n i du
e a s dS 0:
ni d U
i
B:10
The two terms in Eq. (B.10) resemble the typical solid and fluid boundary integrals, like in the displacement formulation as
shown in Eq. (26). Therefore, expressing Eq. (B.8) as Eq. (B.10) shows that the two formulations have equal boundary integrals. Therefore, there is no need to repeat the investigation of the couplings for the proposed mixed formulation.
B.3. Couplings in the mixed formulation
As concluded above, it is unnecessary to repeat the investigation of the couplings for the proposed mixed formulation.
From a numerical point of view, F! is not conveniently evaluated since it contains the gradient of the pressure on the surface
S. This difficulty is overcome by summing F! to the constraint functional needed for each coupling, Eqs. (B.3)(B.5): for solid
couplings,
F! T ps #
Z
!
"
!
"
~a h q
~
~ aj h q
~
^ fc
^ fc
~ cik q
~ cik q
ni p
ni u
kj dij uj dS #
kj dij p dS:
B:11
for fluid couplings (see also Eq. (51)),
F! F!e T pf #
Z
! !
"
"
! !
"
"
~a h q
~ aj h q
~
~
^ fc
^ fc
~ cik q
~ cik q
ni p
ni u
kj dij # dij uj dS #
kj dij # dij p dS:
S
B:12
for unjacketed and bonded porous material couplings,

Z
Z
! !
"
!
""
! !
"
!
""
2
2
~ j dS # ni u
~ dS:
~ a h1 q
~ aj h1 q
~ c;1
^ fc;1
^ fc;2
~ c;1
^ fc;1
^ fc;2
F!1 F!2 T pp # ni p
q~ c;2
q~ c;2
u
p
ik qkj dij # h
ik qkj dij
ik qkj dij # h
ik qkj dij
S
B:13
For isotropic media the term within brackets in Eqs. (B.11)(B.13) simplifies to:
#
$
#
$
!
"
q12 # b=ix
qf
~ cik q
^ fc
d :
h q
1 dij h
kj dij h
q22 b=ix
qf # q12 b=ix ij
B:14
Eqs. (B.11)(B.13) only depend on frame displacement and pore pressure. When these functionals are included in the
Lagrangian, only essential conditions expressing the continuity of the field variables are left to be imposed. For the three
investigated cases, these conditions respectively impose the continuity of frame and solid displacements (solid), pore and
external pressures (fluid), frame displacements and pore pressures (porous material).
83
B.4. Conclusion of Appendix B

In this appendix the Lagrangian in the mixed formulation !L is validated by deriving the EulerLagrange equations and by
investigating boundary and coupling conditions. The derived EulerLagrange equations equal the equations of motion in the
literature. The boundary integrals of the two formulations are equal, as required by the invariance of the Lagrangian. Therefore, any coupling investigation valid for one formulation also applies to the other.
References
Allard, J., & Atalla, N. (2009). Propagation of sound in porous media (2nd ed.). Wiley.
Altay, G., & Dkmeci, M. C. (2006). On the equations governing the motion of an anisotropic poroelastic material. Proceedings of the Royal Society A, 462,
23732396.
Atalla, N., Hamdi, M. A., & Panneton, R. (2001). Enhanced weak integral formulation for the mixed (u,p) poroelastic equations. Journal of the Acoustics Society
of America, 109.
Atalla, N., Panneton, R., & Debergue, P. (1998). A mixed displacementpressure formulation for poroelastic materials. Journal of the Acoustics Society of
America, 104.
Bagley, R. L., & Torvik, P. J. (1983). Fractional calculus A different approach to the analysis of viscoelastically damped structures. AIAA Journal, 21, 741748.
Bedford, A., & Drumheller, D. S. (1979). A variational theory of porous media. International Journal Solids Structures, 15, 967980.
Biot, M. A. (1954). Theory of stressstrain relations in anisotropic viscoelasticity and relaxation phenomena. Journal of Applied Physics, 25, 13851391.
Biot, M. A. (1956). Theory of propagation of elastic waves in a fluid-saturated porous solid: 1. Low-frequency range. Journal of the Acoustics Society of America,
26, 168178.
Biot, M. A. (1961). Generalized theory of acoustic propagation in porous dissipative media. Journal of the Acoustical Society of America, 34, 12541264.
Biot, M. A. (1962). Mechanics of deformation and acoustic propagation in porous media. Journal of Applied Physics, 33.
Dazel, O., Brouard, B., Depollier, C., & Griffiths, S. (2007). An alternative Biots displacement formulation for porous materials. Journal of the Acoustics Society
of America, 121, 35093516.
Dazel, O., Sgard, F., Becot, F.-X., & Atalla, N. (2008). Expressions of dissipated powers and stored energies in poroelastic media modeled by u, U and u, P
formulations. Journal of the Acoustics Society of America, 123, 20542063.
Dazel, O., & Tournat, V. (2010). Nonlinear Biot waves in porous media with application to unconsolidated granular media. Journal of the Acoustical Society of
America, 127, 692702.
dellIsola, F., Madeo, A., & Seppecher, P. (2009). Boundary conditions at fluid-permeable interfaces in porous media: A variational approach. International
Journal of Solids and Structures, 46, 31503164.
Dovstam, K. (1995). Augmented hookes law in frequency domain. A three-dimensional, material damping formulation. International Journal of Solids and
Structures, 32, 28352852.
Dovstam, K. (2000). Augmented Hookes law based on alternative stress relaxation models. Computational Mechanics, 26, 90103.
Finnveden, S. (1994). Exact spectral finite element analysis of stationary vibrations in a rail way car structure. Acta Acustica, 2, 461482.
Finnveden, S. (1997). Spectral finite element analysis of the vibration of straight fluid-filled pipes with flanges. Journal of Sound and Vibration, 199, 125154.
Finnveden, S., & Fraggstedt, M. (2008). Waveguide finite elements for curved structures. Journal of Sound and Vibration, 312.
Fung, Y. C., & Tong, P. (2001). Classical and computational solid mechanics. World Scientific.
Gladwell, G. M. (1966). A variational formulation of damped acousto-structural vibration problem. Journal of Sound and Vibration, 4.
Goldstein, H. (1980). Classical mechanics (2nd ed.). Addison Wesley.
Gurtin, M. E., & Sternberg, E. (1962). On the linear theory of viscoelasticity. Archive for Rational Mechanics and Analysis, 11, 291356.
Hrlin, N.-E., & Gransson, P. (2010). Weak, anisotropic symmetric formulation of Biots equations for vibro-acoustic modelling of porous elastic material.
International Journal for Numerical Methods in Engineering, 84, 15191540.
Jaouen, L., Renault, A., & Deverge, M. (2008). Elastic and damping characterizations of acoustical porous materials: Available experimental methods and
applications to a melanine foam. Applied Acoustics, 69, 11291140.
Lopatnikov, S. L., & Cheng, A. H.-D. (2002). Variational formulation of fluid infiltrated porous material in thermal and mechanical equilibrium. Mechanics of
Materials, 34, 685704.
Lopatnikov, S. L., & Cheng, A. H.-D. (2004). Macroscopic Lagrangian formulation of poroelasticity with porosity dynamics. Journal of the Mechanics and Physics
of Solids, 52, 28012839.
Maximov, G. A. (2010). A generalized variational principle for dissipative hydrodynamics and its application to Biots theory for the description of a fluid
shear relaxation. Acta Acustica United with Acustica, 96, 199207.
Morse, P. M., & Feshbach, H. (1953). Methods of theoretical physics. McGraw-Hill.
Morse, P. M., & Ingard, K. U. (1986). Theoretical acoustics. Princeton University Press.
Panneton, R., & Atalla, N. (1996). Numerical prediction of sound transmission through finite multilayer systems with poroelastic materials. Journal of the
Acoustical Society of America, 100, 346354.
Rayleigh, L. J. W. S. (1945). Theory of sound volume two. Dover.
Riewe, F. (1996). Nonconservative Lagrangian and Hamiltonian mechanics. Physical Review E, 53, 18901899.
Schechter, R. S. (1967). The variational method in engineering. McGraw-Hill.
Zienkiewicz, O. C., & Taylor, R. K. (1967). The finite element method Its basis fundamentals (6th ed.). McGraw-Hill.
Paper 3
!
Statistical energy analysis of the sound transmission through layered panels

based on a variational formulation of the porous material
Mathias Barbagallo, Svante Finnveden , Hao Liu
The Marcus Wallenberg Laboratory for Sound and Vibration Research (MWL), Royal Institute of Technology (KTH), 10044
Stockholm, Sweden
Abstract
The sound reduction index of layered panels comprising porous layers in between two layers made of rubber
and aluminium is found with statistical energy analysis. The porous layer is modelled via an equivalent
fluid theory for limp frames, derived from Biots theory. The elements of the statistical energy analysis
correspond to wave-types carrying energy along the layered panel. Two of these elements refer to waves
appearing in the equivalent fluid having boundary condition that describe the fully coupled reaction of the
external layers. The wave form and the modal density of these trim-coupled waves is obtained studying
the interaction of the three layers via a variational approach. The variational formulation helps devising
the correct coupling conditions between a solid and a porous material, which is first modelled using Biots
theory with three waves, then using equivalent fluid theory with one wave.
Keywords: multilayered structures, double walls, porous materials, Biots equations, statistical energy
analysis, variational principle
1. Introduction
The airborne sound transmission of multilayered structures can be quantified via their sound reduction
index (SRI). The SRI of a structure depends on its vibroacoustic response, which in turn depends on the
waves propagating within it: as illustrated in a recent conference paper by Liu et al. [26], the frequencies
at which new waves start propagating correspond to substantial changes in the SRI (see Fig. 1). The wave
motion of multilayered structures can be complex: each layer may support many kinds of waves, the coupled
substructures may support other waves, these waves may cut-on at specific frequencies. Each propagating
wave transports energy across the structure. The amount of transported energy determines the SRI. The
understanding of these channels of energy transport is salient and fascinating.
This work deals with the prediction of the SRI of multilayered structures comprising porous layers, such
as trim-panels used in the vehicle industry, and aims at providing insights to their physics.
The literature offers a variety of approaches to predict the SRI of such structures. In the following
overview the focus is on works involving porous materials, either modelled by Biots equations ([5] and Ch.
6 in Ref. [1]), or by equivalent fluid theories ([29], [12], Ch. 11 in Ref. [1]). Lauriks et al. employ the transfer
matrix formalism to calculate the SRI through an infinitely extended construction [22]. This formalism is
reprised in Allard and Atallas book to also include porous materials [1]. A publication by Khurana et al.
shows how to use a modified displacement formulation to describe transversely isotropic porous materials
by transfer matrices, providing good agreement between their theory and the measurements of the surface
impedance of porous layers backed by a rigid wall [20]. Lee and Xu present a modified transfer matrix method
using 1-D theory to model Kundts tubes measurements of porous materials [23]. Rhazi and Atalla propose
Corresponding author, Tel. +46 8 790 91 39.
Email addresses: matbar@kth.se (Mathias Barbagallo), svantef@kth.se (Svante Finnveden), haoliu@kth.se (Hao Liu)
234(+*56(' .!751
!"
!"
"
!"
%!
!"
!"
&'()*(+,- ./01
!"
Figure 1: Dispersion curves of a multilayered panel, described in Ref. [36]. Numerical dispersion curves from waveguide finite
element method [26] (black dots). Measured SRI of a multilayered panel taken from Ref. [36] plotted over the dispersion
curves (full curve with squares). The frequencies at which new waves start propagating correspond to changes in the SRI, as
evidenced by the vertical dashed lines.
a simple method to account for size effects in the transfer matrix method, which improves the calculated
sound radiation but does not attribute resonant effects [32]. Finite element models (FEMs) natively account
for the finiteness of the partition. Panneton and Atalla present FEMs of multilayered structures comprising
porous materials, both using Biot equations and the rigid-frame equivalent fluid model [30]. Rigobert et al.
use the displacement-pressure formulation of Biot equations to define hierarchical FEs and their couplings to
elastic solids and fluids [33]; see also Schneider et al.s work [34]. Legault and Atalla investigate an analytical
periodic approach for structure-borne transmission for multilayered structures in aircrafts [24, 25]. Liu et
al. combine a waveguide finite element method with a Rayleigh-Ritz procedure to calculate the SRI of
trim-panels, which is an interesting procedure because it presents the propagating waves in the structure
and is very efficient [26]. To identify the contributions to the SRI of the different energy channels, Doutres
and Atalla present a method valid for normal incidence and for transmission problems only [10] (see also the
measurements in Ref. [11]). Other notable contributions for calculating sound transmission in multilayered
structures are those of: Tanneau et al., who use a boundary element method; [35]; Chazot and Guyader, who
extend the patch-mobility method [8]; Bolton et al., who analytically solve the boundary value problem of
the layered media using wave solutions [6]; Fernandez et al., who employ fuzzy-structure theory to account
for the variation introduced by porous layers in FEM [13]; Ghinet et al., who use statistical energy analysis
to predict the SRI of curved sandwich panels comprising layers of graphite/epoxy and rigid foams - but not
poro-elastic layers [17].
It is wished that any numerical approach calculating the SRI of a structure has the following properties,
besides giving a correct prediction. First, it should provide insights into the physics of the structure. Second,
it should be able to identify and rate the contributions of the various energy channels to the SRI. Third,
the statistical variability of real-life structures should be taken into account. Fourth, it should provide all
this in a short calculation time.
Standard deterministic methods may not easily meet all four properties: they may not help understanding
the physics, unless post-processing routines are implemented; the rating of the various contribution to
the SRI is not immediate; their usefulness is questioned by the statistical variation of, e.g., the ambient
2
Figure 2: SEA for a double-wall filled with air, Ref. [15]. The arrow define different channels for energy flow: A, quasi-plane
cavity-wall waves
free acoustic waves; B, oblique cavity-wall waves
free acoustic waves; C, free acoustic waves
free
acoustic waves (mass law, non-resonant); D, free acoustic waves
wall bending waves; E, quasi-plane cavity-wall waves
bending waves; F, oblique cavity-wall waves
wall bending waves; G, wall bending waves
wall bending waves.
temperature, the properties of the layers and the inexact knowledge of boundary and coupling conditions;
their computational cost is high, especially when frequency dependant parameters and porous materials are
involved [18]. On the contrary, statistical energy analysis (SEA) may, in principle, comfortably meet all four
properties. SEA is a statistical method for predicting the vibroacoustic response of built-up structures at
higher frequencies, intrinsically accounting for their statistical variation; SEAs rationale and its underlaying
assumptions are presented in Ref. [31]; criteria for stating the applicability of SEA to complex structures are
still object of researches, such as Ref. [7]. SEA is based upon a set of elements describing the vibroacoustic
response of the structure and satisfying the statement of energy conservation; the statement is expressed via
a set of linear equations, which is solved for the energies of the elements; the computation time is often very
short. The elements may correspond to wave-types, rather than to substructures: this sometimes improves
the performance of SEA, in terms of the predicted result and of the first two properties above (see Ref.
[15]). The identification of the elements governing the response is immediate.
Despite its appealing features, the possibilities of SEA with multilayered structures comprising porous
layers have not been explored. This work aims at deriving an SEA for this class of structures. In Ref. [9]
Craik et al. use SEA for calculating the SRI of double-walls with and without added absorption. The cavity
with added absorption is modelled as an air cavity with a higher damping loss factor; the damping loss factor
is estimated via a reverberation-time measurement, which, however, decays too quickly. An alternative to
Craik and Smiths work for double-walls without added absorption is shown in Fig. 2. The figure presents
the set of SEA elements devised in Ref. [15], for the same structures as those used by Craik et al.. The
novelty is the cavity-wall element, modelling the waves in the air cavity fully coupled to the walls, when the
motion of the walls is mass-impeded. Herein, a similar approach is used: SEA elements modelling the waves
in the porous layer fully coupled to the mass-impeded external layers are sought; these waves are named
trim-coupled waves. The porous material is described by equivalent fluid theory, derived from Biot theory
assuming a limp frame; the input parameters to these theories can be measured, as shown in Ref. [19]. The
authors think that this is the first SEA based upon either Biot or equivalent fluid theories for modelling
porous layers in multilayered structures.
The structures under study are single-curved trim-panels having a porous layer in between a rubber layer
and an aluminium layer; they may be seen as a first approximations to a car or truck floor. In Section 2 a
functional that is stationary for the true motion of the coupled trim-panel is defined; the porous layers is
described by Biot theory and the external layers as solids. This functional is then simplified, by assuming that
3
the porous frame is limp and that the aluminium and rubber layers also are limp, a valid assumption when
describing the waves within the trim panel at frequencies below the coincidence frequency. A variational
procedure similar to Hamiltons principle is applied to the simplified functional. The equations of motion
of the equivalent fluid, of the mass-impeded external layers and the coupling equations describing the trimcoupled waves follow. When the model describing the porous layer is simplified from Biots to equivalent
fluid theory, one coupling condition between the porous layer and the external layer has to disappear: a
systematic and original procedure to eliminate it is provided by the variational principle. Upon this basis,
in Section 3 the wave form and the modal density of the trim-coupled waves are derived and used in an
SEA for predicting the sound reduction index of the trim panel. The SEA of the whole panel in Section 4,
also accounting for the radiation of the bending waves in the external layers, is similar to that shown in Fig.
2, except that the cavity-wall elements are substituted by the trim-coupled wave elements.In Section 5 the
calculated SRI is compared to laboratory measurements and in Section 6 conclusions are drawn.
2. A variational principle for the trim-panel
The multilayered structures under study are trim-panels made of two external layers, one of aluminium
and one of rubber, with a porous layer in between them; its cross-section is shown in Fig. 3. To start, a
modified Hamiltons variational principle for the trim panel is defined. Then, in Subsection 2.4 the model of
the trim panel is simplified assuming that its solid parts are limp. Upon variational calculus, the equations
of motion of the (limp) porous layer and of the (limp) external layers follow, together with the coupling
equations between them. These coupling equations are the sought result: they describe the trim-coupled
waves. This wave-type constitutes an SEA element, as expounded in Section 4.
Hamiltons variational principle needs self-adjoint functionals and, therefore, conservative systems [16].
To apply it to the non-conservative layers in the trim-panel, self-adjoint functionals analogue to the Lagrangians are built inspired by Morse and Feshbachs construction [27]: the original system is considered
together with its adjoint [4]. The adjoint is a mirror system, identical to the original one except that it
has negative losses. The two systems together constitute a conservative system, which can be the basis of a
variational procedure similar to Hamiltons principle. In Ref. [4], it is shown how functionals analogous to
the Lagrangian can be constructed for Biots equations, both in their displacement and mixed formulations;
the main results from that work are used in the next subsection.
The (analogy to the) Lagrangian of the coupled trim-panel is defined summing the (analogies to the) Lagrangians of the three layers and the functionals defining the constraints on the displacements of the three
layers. These constraints are introduced in the variational form via Lagrange multipliers; their actual expressions are identified in Section 2.3. The Lagrangian of the coupled trim-panel, Ltot , reads
Ltot
Lpor
Lp1
Lp 2
V,
(1)
where Lpor is the Lagrangian of the porous material, Lp1 and Lp2 are the Lagrangians of the external layers,
T is the functional expressing the continuity of the solid displacements, V is the functional expressing the
continuity of the normal total displacement in the porous material and the normal displacements in the
external layers. The variational principle based upon Ltot is [4]:
Ltot
0,
(2)
where indicates the first variation of Ltot . In the following subsections the functionals in Eq. (1) are
presented. Time-harmonic motion at angular frequency is considered and the time dependance is eit . A
sketch of the trim-panel in Fig. 3 is given in Fig. 4.
2.1. Porous material layer
In this work, the porous layer is modelled first with Biot theory and then with equivalent fluid theory
for limp frames: the values of the Lagrange multipliers in T and V are obtained using the former, the
trim-coupled waves are devised with the latter.
4
Figure 3: From top to bottom, the aluminium plate, the porous layer and the rubber plate.
Figure 4: Coordinate systems used in the article. Volume V is occupied by the porous layer, which is connected to the rubber
layer at surface S1 , and to the aluminium layer at surface S2 . The normal vectors point outwards of the structures, aligned
with the x2 axis.
The Lagrangian of a porous material described by Biots equations expressed in the displacement formulation reads [4]:
Lpor
K dV,
(3)
where V is the domain of the porous material, W is the strain potential density and K is the kinetic potential
density. In this section, cartesian tensor notation is used, a comma in the subscript corresponds to a partial
derivation. The frame and fluid displacements are respectively ui and Ui ; accordingly, the frame strain !ij ,
the frame dilatation and the fluid dilatation e are defined as
1
ui,j uj,i ;
ui,i ;
e Ui,i .
(4)
!ij
2
W and K are defined as
W
!aij Dijkl !kl
a Q e
uai
uai
2 sc
ui
ea R e
2 c
ea Q ,
Uia
Ui
2 fc
Ui
(5a)
Uia
2 c
ui ,
(5b)
Dijkl is the dynamic moduli tensor of the frame, Q the potential coupling parameter, R the bulk modulus
of the fluid, c an equivalent density accounting for viscous and inertial coupling between frame and fluid,
sc and fc equivalent densities of the frame and fluid, accounting for inertial and viscous effects:
b
;
sc
1 h s0 c ;
fc hf0 c ,
(6)
i
where 12 is the mass coupling between the solid- and fluid-phase, b is the viscous drag, h is the porosity,
s0 and f0 are the intrinsic densities of the frame and of the fluid, respectively; for more details on these
parameters see Refs. [4, 5]. The superscript a refers to the complex conjugate of the generalised variables
of the system adjoint to the original; more details on the Morse and Feshbachs construction for Biots
equations are found in Ref. [4]. The functionals W and K - and all functionals in the following - are bilinear
functionals of the original and adjoint generalised coordinates. They are self-adjoint and can be the basis
of a variational principle.
The frame and fluid stresses, respectively ij and hp, are introduced,
c
12
ij
Dijkl !kl
hp
Re
Q e ij ;
(7a)
Q ,
(7b)
ij is Kroneckers delta. The total stress of the porous material is defined as

t
ij
ij
hp ij .
(8)
In Ref. [4] a mixed formulation function of frame displacement and fluid pressure is derived. This
formulation is equivalent to the displacement formulation. It reads:
Lpor
V
a
!aij Dijkl !kl
pa dV
Q2
uai 2 ui
pa
h2
p
R
pa,i
h2
p,i
fc
(9)
where
sc
c 2
;
fc
c
fc
Q
.
R
(10)
ni pa,i
h2
p dS1
2 fc
(11)
ni pa,i
h2
p dS2 ,
2 fc
The surface functional F in Eq. (9) is given by

n i pa
F
S1
n i pa
S2
h2
p,i dS1
2 fc
h2
p,i dS2
2 fc
6
S1
S2
where ni is the normal outward pointing vector, here aligned with x2 .

Equivalent fluid theory for limp frame is conveniently derived from the mixed formulation: the first two
terms in Eq. (9) are the strain potential relative to the stress of the frame in-vacuo, which equals zero for a
limp frame. Thus, the functional for such a porous material is:
uai 2 ui
Lpor
V
pa
h2
p
R
pa,i
h2
p,i
2 fc
a p
pa dV
F.
(12)
For a limp frame, the Lagrangians in Eqs. (3), (9) and (12) are equivalent.
2.2. Continuity conditions
The continuity between the displacements of the frame and of the external layer is introduced in the
variational form via Lagrange multipliers. The functional T reads
2
l
Sl
l 1
wipl
l
ui
a,p
i
pl uai
wia,pl
dSl ,
(13)
where wipl and its adjoint are first-order tensors describing the displacement of the external layer pl , pi l is
a Lagrange multiplier and ai i ,pl its adjoint; these Lagrange multipliers are vectors because T is (an analogy
to) work at boundary Sl .
The continuity between the normal total displacement of the porous material and the normal displacements of the external layers is also introduced via Lagrange multipliers. The total displacement uti is
uti
h ui
hUi .
(14)
The functional V reads

2
l
Sl
l 1
a,pl ni uti
ni wipl
pl ni ut,a
i
ni wia,pl
dSl ,
(15)
where pl is a Lagrange multiplier and a,pl its adjoint.

2.3. Lagrange multipliers
The values of the Lagrange multipliers in Eqs. (13) and (15) are identified using the variational principle
in Eq. (2). The functional Lpor is expressed with Eqs. (3); in this subsection, the displacement formulation
of Biots equations is preferred over the mixed formulation because it allows for simpler calculations. The
chosen independent variational parameters are Uia and uai . Thus, for plate p1 it follows:
Ltot
S1
Uia ni
hp
h p1 dS1
S1
uai ij nj
p1 1
h ni
pi 1 dS1
0.
(16)
Applying the fundamental lemma of calculus of variations, the following results are obtained. From the first
integral, p1 is obtained:
p1 p on S1 .
(17)
Eliminating p1 from the second integrals, it follows that
pi 1
t
ij
pij nj , on S1 ;
(18)
which defines the other Lagrange multipliers pi 1 . It is further noticed that

pi 1
t
ij
pij nj
7
ijkl )kl , on S1 ,
nj D
(19)
ijkl is the stiffness of the frame in vacuo and, thus, p1 is identified as minus the traction of the
where D
i
frame in-vacuo: if the frame of the porous material is limp, the constraint introduced by the functional T
in Eq. (13) disappears; to the authors knowledge, this results has not been obtained before. Similar results
are obtained for p2 and pi 2 when considering plate p2 ,
p2
pi 2
p;
t
ij
ijkl %kl , on S2 ;
nj D
pij nj
(20)
These results also apply to the adjoint Lagrange multipliers.

All Lagrange multipliers can now be eliminated from the contraint functionals using the derived values.
These results equally apply if the porous material is modelled using the displacement formulation, the mixed
formulation or, for a limp frame, equivalent fluid theory.
2.4. Equations of motion and coupling conditions for the simplified trim panel
In the SEA for the sound transmission through the trim-panel derived in the next sections, two elements
describe the bending waves in the external layers modelled as plates. In this and in the next section,
these bending waves are neglected and the focus is on the elements accounting for the trim-coupled waves,
substituting the cavity-wall waves in Fig. 2.
The equations describing the trim-coupled waves are devised for a porous material that has a limp frame,
described by the Lagrangian in Eqs. (12). The trim-coupled waves propagate in the equivalent fluid and are
fully coupled to the external plates. If the wavelength of the trim-coupled wave is much larger than the free
wavelength of the bending wave in a plate, then the motion of the plate is mass-impeded and, equivalently,
its material is assumed limp. The viscoelastic forces of plate pl are thus neglected and its Lagrangian only
comprises kinetic energy,
2 mpl wa,pl wpl dSl ,
Lpl
(21)
Sl
where mpl is the mass per unit area the plate, Sl is the two-dimensional domain of the plate, wpl and its
adjoint are the zeroth-order tensors of the normal displacement,
wipl ni ;
w pl
wia,pl ni .
wa,pl
(22)
The frequency range in which the external layers can be considered limp plates is known examining the
dispersion curves of the coupled trim-panel, as discussed in Section 4.1.
For a limp frame the constraint functional T in Eq. (13) disappears, Eqs. (19)-(20). The constraint
functional V in Eq. (15) is expressed as
2
pa u t
w pl
p ut,a
wa,pl
dSl ,
(23)
Sl
l 1
where scalars are used to describe the total displacement,

ut
h u2
hU2
h u i ni
hUi ni ,
(24)
since this coupling only occurs along the x2 axis. The Lagrangian in Eq. (1) for the trim-panel with limp
solid parts is consequently given by:
uai 2 ui
Ltot
V
2
ni
pa
Sl
l 1
h2
p
R
pa,i
h2
p,i
fc
pa,i
pa
h2
p,i
2 fc
h2
p
fc
a p
pa dV
dSl
2 mpl wa,pl wpl dSl

l 1 Sl
2
1
l 1
pa u t
w pl
p ut,a
Sl
wa,pl
dSl .
(25)
Eventually, the equations of motion and the coupling conditions for the original system are retrieved by
taking the variations with respect to the adjoint generalised coordinates. Before doing so, the adjoint
generalised coordinates are collected in Eq. (25) and integrations by parts are taken when their derivates
appear in any of the volume functionals. Eq. (25) becomes
Ltot
V
uai
2 ui
h2
p
R
pa
p,i dV
V
pa u t
wp1 dS1
h2
p,ii
fc
wa,p1
2 mp1 wp1 dS1
wa,p2
2 mp2 wp2 dS2 .
dV
S1
S1
pa u t
wp2 dS2
S2
(26)
S2
The first variation of Ltot in Eq. (26) is stationary for the true motion of the original and adjoint systems.
The following equations result from the fundamental lemma of calculus of variations. From the variation of
uai in the volume integral, the Euler-Lagrange equation of the limp frame follows,
p,i .
2
ui
(27)
Form the variation of pa in the volume integral, the Euler-Lagrange equation of the equivalent fluid follow,
R
p,ii
h2
2 L p
0;
fc
,
2 fc h2
(28)
where the frame displacement was eliminated using Eq. (27). From the variations of wa,p1 and wa,p2 , the
equations of motion of the forced external plates follow,
2 mp1 wp1 , on S1 ;
2 mp2 wp2 , on S2 .
(29)
Finally, from the variation of pa in the surface integrals, the coupling equations between the porous material
and the external plates follow
1
p,2
2 L
w p1
1
p,2
2 L
0, on S1 ;
w p2
0, on S2 .
(30)
To obtain Eq. (30) the fluid displacement was eliminated using the equation of motion of the fluid phase,
obtained taking the variation of the adjoint fluid displacement in Eq. (3),
h Ui
h2
fc 2
p,i
c
ui ,
fc
(31)
and then the frame displacement was eliminated using Eq. (27).The plate displacements in Eqs. (30) are
now eliminated using Eqs. (29), upon which mixed boundary conditions for the equivalent fluid follows:
p
d p1 p,2 , on S1 ;
where
p 1
mp1
;
d L
p2
d p2 p,2 , on S2 ,
mp 2
,
d L
(32)
(33)
and d is the distance between the two external layers. Eqs. (32) are the sought equations describing the
coupling between the porous layer modelled by equivalent fluid theory and the limp external plates; they
are used in the next section.
Eqs. (32) satisfy the continuity of the normal total displacement of the porous layer and the normal
displacements of the external layers. As shown in Appendix A, this continuity condition follows from the
first variation of Eq. (12): the produced boundary integral expresses the (virtual) work of the pore pressure
on the variation of the total normal displacement, which has not been explicitly stated before.
9
3. Wave form and modal density of the trim-coupled waves

In this section, the wave form and the modal density of the propagating trim-coupled waves are derived,
both needed by the SEA of the trim panel. In the remaining of the article, tensor notation is abandoned in
favour of vector notation, more suitable to computer implementation.
The solution p x, y, z to Eq. (28) can be decomposed as (x x1 , y x2 , z x3 ):
p x, y, z
g x, z f y ,
where the function g x, z describes a wave field along x

It follows that
(34)
z, while the function f y a modal field along y.
2
g x, z
g x, z
2 g x, z
2
x
z2
2
f y
2 f y
0,
y2
2
0;
(35a)
(35b)
where
kL2
2 ;
;
cL
kL
R
,
h 2 L
cL
(36)
kL is the wavenumber of the equivalent fluid wave and cL its wavespeed. A wave solution to f y in Eq.
(35b) is a linear combination of sines and cosines. Solving the boundary value problem given by the equation
of motion in Eq. (35b) and the boundary conditions in Eqs. (32), the following trascendental eigenvalue
problem is obtained:
1
p1 p 2 d
p1
sin d
p2 d cos d
0.
(37)
Eq. (37) can be solved for d at each angular frequency , since p1 and p2 are frequency dependant. The
eigenvalue problem in Eq. (37) is approximately solved by:
2fdw
;
cL
2fr
.
cL
(38)
The frequency fdw corresponds to the double-wall resonance, which (if pl

fdw
cL
2d
1) can be expressed as
1
.
p2
1
p1
(39)
At frequency fr , half of the wavelength is an integer fraction of the plate distance,

fr
r cL
;
2d
N.
(40)
When Eq. (37) is solved in MATLAB using fzero, the real parts of 0 and r are used as initial values.
Two groups of waves are identified: one corresponds to the quasi-plane waves starting to propagate a
fdw (r
0); one corresponds to the oblique coupled waves starting to propagate at fr (r N). To each
mode r described by r , corresponds a set of waves with wavenumber r propagating in the x z plane; r
is obtained via Eq. (36). Finally, the pressure pr x, y, z for each mode r reads:
pr x, y, z
p sin r y
dp2 r cos r y
ir x cos z sin
(41)
where p is the complex amplitude of the wave and is the angle of propagation of the wave. The trigonometric
functions in the parentheses describe the modes along y, while the exponential functions the waves travelling
along x z plane. These results are similar to those obtained for building construction double-walls enclosing
10
an air cavity studied in Ref. [15], suggesting that these two kinds of structures may be studied in a similar
fashion. The function fzero yields as output the real part of r ,
Re r .
(42)
For all
r , the number of modes in a large two-dimensional structure below the angular frequency is
asymptotically given by (p 302 in Ref. [21])
2r S
.
4
Nr
The asymptotic modal density in x
(43)
z is calculated as [21]:
nr
2r u
2r l
4 u l
(44)
where u and l are the upper and lower frequency of a frequency band having bandwidth u l .
For the rest of the article, the most important results are the form of the pressure wave in Eq. (41) and the
modal density in Eq. (44): both are needed for the SEA of the trim-panel in the next section.
4. Statistical energy analysis
A statistical energy analysis is founded upon the statement of energy conservation between the elements
of the vibroacoustic response of the system under study [31]. The first step in an SEA is the identification
of the elements. In the next subsection, the dispersion curves of a trim-panel are investigated and a set of
SEA elements is presented in Subsection 4.2.
4.1. Dispersion curves of the trim-panel
The dispersion curves of the waves propagating along the x z plane in a trim-panel are shown in Fig.
6. They are calculated via a waveguide finite element method (WFEM) and via analytical solutions. The
WFEM uses the mesh shown in Fig. 5 and models the porous layer with full Biot theory and the plates
with Mindlin theory; details on the formulation are found in Ref. [26]. The trim-panel is defined in Ref. [3],
where it is referred to as a car-floor. The analytical wavenumber
0 of the trim-coupled waves is computed
with Eqs. (36); the analytical wavenumbers of the Biot waves are given on p 121 of Ref. [1]; the analytical
wavenumbers of the uncoupled bending waves in the external layers and for the coupled full structure are
given on p 55 of Ref. [21].
The WFEM calculates a wave cutting-on around 300 Hz. This wave asymptotes to the uncoupled fastwave and it is, therefore, identified as the coupled fast-wave. The wavenumber
0 cutting-on at fdw is very
close to the coupled fast-wave.
0 is also smaller than the wavenumber of the bending waves in the external
layers: this validates the limp assumption on the external layers, at least till 1250 Hz. The coupled fastwave has a high imaginary part around cut-on, due to the high damping in those frequency bands. At lower
frequencies, the trim-panel acts as a single layer with an equivalent stiffness. As frequency increases, the
wavenumbers of the bending waves in the coupled external layers are well approximated by the wavenumbers
of the uncoupled layers.
4.2. SEA elements
To calculate the SRI of the trim panel, the SEA model mimics a SRI measurement: the trim panel is
placed in a rigid wall between a reverberant room and an anechoic room; a diffuse sound field is generated
in the reverberation room.
The SEA for the sound transmission through the trim panel is inspired by that of a double-wall, Fig.
2, except that the trim-coupled wave elements substitute for the cavity-wall elements. The analysis of the
dispersion curves in Fig. 6 confirms that the SEA elements corresponding to the wave-types carrying energy
from the reverberation room to the anechoic room can be taken as: 1, an element describing the acoustic
11
0.02
0.015
0.01
0.005
0
2
8
3
x 10
Figure 5: Two-dimensional (x z) mesh used by the WFEM. The nodes on the thick lines at the bottom and at the top model
the external layers, while the other nodes model the porous layer.
12
10
Wavenumber || [1/m]
10
10
10
10
10
Frequency [Hz]
Figure 6: Real part of the wavenumbers of the coupled trim-panel, calculated by analytical formulae and by WFEM. Bending
waves in the external layers: WFEM (dashed lines) and analytical uncoupled (thin lines with black triangles); the upper two
lines refer to rubber layer, the lower two lines to aluminium layer. Bending waves of the trim-panel modelled as an equivalent
single layer: analytical (line with empty squares). Coupled fast Biot wave: WFEM (full line). Uncoupled fast Biot wave: see p
121 Ref. [1] (dot-dashed line). Trim-coupled wave:
0 (line with full circles). The dotted lines, obtained with WFEM, indicate
waves with an imaginary part 1.6 times larger than their real part.
waves in the reverberation room; 2, an element describing the bending waves in the aluminium layer; 3, an
element describing the quasi-plane trim-coupled wave cutting-on at fdw , Eq. (39); 4, an element describing
the oblique trim-coupled waves, each cutting-on at fr , Eq. (40); 5, an element describing the bending waves
in the rubber layer; 6, an element describing the acoustic waves in the anechoic room.
The statement of energy conservation for element i is:
Piin
Pid
c
,
Pi,j
(45)
c
where Piin is the input power to element i, Pid its dissipated power, Pi,j
the net coupling power between
elements i and j. In this SEA, all elements have input power equal to zero, except element 1 that has unit
input power. Eq. (45) is expressed for all elements thus obtaining a linear system of equations, where the
unknowns are the modal energies of the elements; for element i, its modal energy ei is defined as the ratio
of its total energy Eit and its modal density ni . The dissipated power is calculated as
Pid
Mi ei ;
Mi
i ni
(46)
where Mi is the modal overlap factor of element i, ni its modal density and i its damping loss factor. The
coupling power is calculated as
c
Pi,j
Ci,j ei
ej ;
Ci,j
ij ni
ji nj ,
(47)
where Ci,j is the conductivity of elements i and j and ij is the corresponding coupling loss factor. Ci,j can
also be calculated using a travelling-wave estimate,
Ci,j
tx
Pi,j
ei
13
,
Ci,j Mj
(48)
tx
is the transmitted power from i to j, when j is extended towards infinity (hence, Mj
).
where Pi,j
All modal densities, modal overlap factors and conductivities are needed to calculate the elements modal
energies. The modal densities of elements 3 and 4 are given by Eq. (44) for r 0 and r 1, respectively;
the conductivities and the modal overlap factor are derived in the next subsections. The parameters of all
remaining elements are defined in Ref. [15].
4.3. Conductivities and modal overlap factor of the trim-coupled wave

First, the expression of the conductivity between bending waves in a plate and free acoustic waves in a
semi-infinite space is recalled (see, e.g., Ref. [15]); then, the conductivities between bending waves in a layer
and trim-coupled waves, and between trim-coupled waves and free acoustic waves are derived. Finally, the
modal overlap factor of the trim-coupled wave is given.
The conductivity between the bending wave in layer pl and the free acoustic wave in volume R is obtained
via a travelling wave estimate, Eq. (48). Two equivalent expressions are given (see Ref. [15] for details):
CR,l
0 c0 R,l S
np l
;
mpl
CR,pl
0 c0 R,l S
npl 2 Epkl
,
mpl Ept l
(49)
where npl is the modal density of the external layer pl , R,l is the radiation efficiency between acoustic
volume R (see Section 4.3.1) and layer pl , Epkl is the kinetic energy of layer pl and Ept l its total energy. The
conductivity between the bending wave in layer pl and the trim-coupled wave r has an expression similar to
the previous,
np l
Cl,r 0 c0 l,r S
,
(50)
mp l
the only difference being the value of radiation efficiency pl ,r , see Section 4.3.1. The conductivity between
the free acoustic wave in volume R and the trim-coupled wave r can be calculated in a fashion similar to
the one in Eq. (49),
k
nr 2
Elr
CR,r 0 c0 R,r S
qlr ;
qlr
(51)
t ,
mpl
Elr
k
t
where Elr
and Elr
are respectively the kinetic energies of plate pl due to the trim-coupled wave r and the
total energy of the trim-coupled wave. The out-of-plane displacement of limp plate pl due to the coupled
wave r is given by Eq. (29), here rewritten for convenience,
pr x, y, z
,
2 mp l
wlr
with
(52)
1 for p1 and 1 for p2 . The mean-square spatial-averaged velocity

vlr
i wlr
which, using Eqs. (41) and (52), for plate p1 , at y

v1r
and for plate p2 , at y
2
vlr
is calculated as
(53)
d, equals
1 2 sin r d
p
4
d r p2 cos r d
2 mp1
0, equals
(54)
v2r
1 2 d r p2
p
2 .
4
2 mp2
(55)
k
The kinetic energy Elr
is consequently:
k
Elr
1 pl
m S vlr
2
14
(56)
while the strain energy Ers is (see Eq. (41)):

Ers
1
1
V
2 L c2L
p2
1
p 2
S
2 4 L c2L
sin r y
d r 2 cos r y
dy.
(57)
t
The total energy Elr
can be calculated as twice the potential energy Ers and, thus, the ratio qlr is expressed
using Eqs. (56) and (57):
k
Elr
qlr
.
(58)
2 Ers
4.3.1. Radiation efficiency

The radiation efficiencies R,l , l,r and R,r are calculated using Leppingtons formula, expressed as a
function of the real pats of the wavenumbers of the radiating waves and of the receiving wave. The reader
may refer to Ref. [15].
4.4. Modal overlap factor of the trim-coupled wave
The modal overlap factor of the trim-coupled wave r is calculated as:
Mr
nr r .
(59)
The damping loss factor of the trim-coupled wave needs to be estimated. In Ref. [28] it is shown that a
complex wavenumber of a lightly damped wave can be expressed as
r
r
r
r
r
,
2cg,r
(60)
where cg,r is the group velocity of the undamped trim-coupled wave. The group velocity for an undamped
wave is calculated as [14]
cg,r
.
(61)
r
The following relation holds
r
2cg,r
Im r ,
(62)
from which it results that the damping loss factor is:

r
2 Im r cg,r
.
(63)
A good approximation for r to extract its imaginary part is given by Eq. (38) using Eq. (36).
5. Results
First, the validity of equivalent fluid theory for limp frame is discussed, then the prediction of the SRI
by the SEA presented in the previous section is compared to laboratory measurement.
5.1. Validity of limp theory
In Ref. [12], Doutres et al. derive a criteria for assessing the validity of equivalent fluid theory for limp
frames, based on a parameter called frame stiffness influence (FSI). The FSI expresses the ratio of two wave
speeds, i.e., the wave speed of the wave in in-vacuo frame and the wave speed of the equivalent fluid wave.
A porous material can be considered limp if the FSI is below a threshold, which in Ref. [12] is determined
simulating 256 porous materials.
15
Wavenumber [1/m]
10
10
10
10
10
Wavenumber [1/m]
10
10
10
10
10
Frequency [Hz]
Figure 7: Real parts of the wavenumbers of the uncoupled porous layer in the car-floor (top) and in the truck-floor (bottom).
Fast wave (dashed line), slow wave (dot-dashed line), equivalent fluid wave (full line).
The SEA model derived in the previous section is used to calculated the SRI of two trim-panels, a carfloor and a truck-floor; the material parameters and details on the SRI measurement are in Ref. [3]. The
porous layer in the car-floor is limp according to the FSI, whereas the one in the truck-floor is not.
In Fig. 7, the dispersion curves of the uncoupled porous layers in the two floors are shown; the real
wavenumbers of the fast and slow waves are calculated using Allards expressions on p 121 in Ref. [1], while
the real wavenumber kL of the uncoupled equivalent fluid wave using Eqs. (36). The two figures show that
when equivalent fluid theory works, its wave has a wavenumber close to the fast Biot waves.
5.2. SEA results
The SEA defined in the previous section is applied to the car-floor and the truck-floor. In Figs. 8-9, the
SRI predicted by SEA is plotted and compared to SRI measurements.
In Fig. 8, SEA shows to perform generally well with the car-floor. SEA catches fairly well the doublewall resonance frequency, around 260 Hz. The trend of the SRI between 260 Hz and 1000 Hz is also well
caught. At frequencies below 260 Hz, SEA prediction is poor: this problem is also encountered with building
construction double-walls in Ref. [15]. Fig. 8 also shows a reduced SEA that only employs three elements,
i.e., the acoustic waves in the two rooms and the quasi-plane trim-coupled waves. In the frequency range of
interest, these three elements are enough to calculate the SRI, thus highlighting their relevance.
In Fig. 9, SEA shows to give a rather incorrect prediction of the SRI of the truck-floor, but for the trend
between 500 Hz and 1000 Hz which is quite correct. It seems that if the resonance frequency in the 500 Hz
third-octave frequency band was caught and not underestimated, the SEA prediction would give the correct
level of the SRI between 500 Hz and 1000 Hz. It might be argued that the equivalent double-wall resonance
frequency corresponding to the dip in the SRI curve also depends on the frame stiffness, which is neglected
in the equivalent fluid model. The dispersion curves in Fig. 7 show that the equivalent fluid wave likely
has a lower stiffness than the fast wave: accordingly, in Fig. 9 the resonance frequency around 500 Hz is
underestimated.
6. Conclusions
This article presents a statistical energy analysis for trim-panels that comprise two external layers enclosing a porous layer. One SEA element describes the trim-coupled waves, i.e. waves in the porous layer
fully coupled to the mass-impeded external layers. The equations describing these waves are derived using a
16
60
50
SRI [dB]
40
30
20
10
10
10
Frequency [Hz]
Figure 8: SRI curves, car-floor. Measurement (full line), SEA prediction (dashed line), reduced SEA prediction (circles),
mass-law prediction (dash-dotted line). Note that the SEA and reduced SEA predictions are almost identical.
70
60
SRI [dB]
50
40
30
20
10
10
10
Frequency [Hz]
Figure 9: SRI curves, truck-floor. Measurement (full line), SEA prediction (dashed line), mass-law prediction (dash-dotted
line).
17
modified Hamiltons principle also valid for non-conservative media. To start, a functional analogous to the
Lagrangian is defined for the coupled trim panel. This is used to study the couplings between the porous
layer described by Biots theory and the two external layers are initially described as general solid media.
The constraints on the continuity of the displacements of the three layers are introduced via Lagrange multipliers: those introducing the continuity of the displacements of the frame and of the external layers equal the
traction of the frame in vacuo. For a limp frame this continuity condition disappears, thus accommodating
the reduction of the waves in the porous layers when going from Biot theory to equivalent fluid theory; this
result is not found in the classical literature [1] and the authors could not derive it starting from Atallas
weak form in Ref. [2].
The trim-coupled waves are calculated assuming a limp frame, which is valid for some materials. These
waves are furthermore calculated modelling the external layers as limp, which is valid since the trim-coupled
waves are faster than the free bending waves in the external layers. The Lagrangian of the coupled trim
panel is thus simplified, assuming that its solid parts are limp. The porous material is now modelled via
equivalent fluid theory, which is described by a novel Lagrangian. The external layers only have kinetic
energy. Upon standard variational calculus applied to this simplified Lagrangian, the equations describing
the trim-coupled waves are derived. These waves are described by two groups of wavenumbers: the first
corresponds to those waves cutting-on at the equivalent double-wall resonance frequency, the second to
those waves cutting-on at frequencies having half of the wavelength that is an integer fraction of the plate
distance. Knowing their wavenumbers, it is possible to calculate their wave form and modal density. A
SEA mimicking a SRI measurement is defined, having elements corresponding to the wave-types carrying
energy: two elements refers to the two groups of trim-coupled waves. This SEA predicts well the SRI of the
trim-panel whose porous layer is limp according to the criteria proposed by Doutres et al. in Ref. [12]. The
SEA prediction for the trim-panel whose porous layer does not fulfil this criteria is poor, but it might be
argued that a model accounting for the frame stiffness would give better results. The performance of the
SEA might be improved with a theory for the porous layer that includes more waves, e.g. full Biot theory.
The wave approach to SEA requires a deeper understanding of its physics, which is welcomed as it
highlights the elements of the response of the trim-panel. SEA per se has many advantages. It easily
permits to identify the most important channel of energy flow, which for the studied trim-panel is the
resonant transmission between the two rooms and the first group of trim-coupled waves. It takes into
account the statistical variability of any real-life structure, such as the trim-panel used in vehicle industry
studied in this work. It provides results in a blink of an eye on any computer. This article adds SEA to the
literature on numerical methods for predicting the SRI of multilayered panels with porous layers. It is the
authors opinion that the possibilities of SEA for these structures have not been fully explored yet.
Acknowledgements
This work was supported by the Swedish Research Council (621-2005-5754) and the European Commission (Mid-Mod, Grant Agreement No.: 218508).
Appendix A. Equivalent fluid formulation
The first variation of the Lagrangian in Eq. (12) is stationary for the true motion of the system, yielding
the equation of motion and the boundary integral of the equivalent fluid. Taking the variation of the adjoint
coordinates, it follows that
uai
2 ui
uai ni p dS
pa
p,i dV
ni pa,i
V
2
h
p dS
fc 2
18
h2
p
R
0.
h2
p,ii
fc 2
ui,i dV
(A.1)
From the fundamental lemma of calculus of variations, the equations of motion of the limp frame and of the
fluid follow from the first two functionals in Eq. (A.1), respectively:
2 ui
2
p,i
0,
(A.2a)
h
p
R
h
p,ii
fc 2
ui,i
0.
(A.2b)
The divergence of the frame displacement is eliminated from Eq. (A.2b) and the equation of motion of
an equivalent fluid follow, Eq. (28). The gradient of the pressure at the boundary is eliminated using the
equation of motion of the fluid phase, obtained taking the variation of the adjoint fluid displacement in Eq.
(3),
c
h2
p
h
U
h
ui .
(A.3)
,i
i
fc 2
fc
Thus, the two boundary integrals in Eq. (A.1) may be collected and rewritten as:
ni ut,a
i p dS
0,
(A.4)
with the relation h Q

1 h. This boundary integrals contains the (variation of the) normal total displaceR
ment, which is the continuous quantity when the equivalent fluid couples to solids, fluids (see the formulation
in Ref. [4]) or other porous materials.
References
[1] J.F. Allard and N. Atalla. Propagation of Sound in Porous Media. Wiley, 2nd edition, 2009.
[2] N. Atalla, M. A. Hamdi, and R. Panneton. Enhanced weak integral formulation for the mixed (u,p) poroelastic equations.
Journal of the Acoustics Society of America, 109(6), June 2001.
[3] M. Barbagallo. Statistical energy analysis and variational principles for multilayered panels. PhD thesis, SCI-AVE,
Teknikringen 8 10044 Stockholm, March 2013.
[4] M. Barbagallo and S. Finnveden. A self-adjoint variational principle for anisotropic viscoelastic Biots equations. International Journal of Engineering Science, 2012.
[5] M. A. Biot. Theory of propagation of elastic waves in a fluid-satured porous solid. 1. low-frequency range. Journal of the
Acoustics Society of America, 26(182):168178, 1956.
[6] J. S. Bolton, N.-M. Schiau, and Y. J. Kang. Sound transmission through multi-panel structures lined with elastic porous
materials. Journal of Sound and Vibration, 191:317347, 1996.
[7] A. Le Bot and V. Cotoni. Validity diagrams of statistical energy analysis. Journal of Sound and Vibration, 329:221235,
2010.
[8] J.-D. Chazot and J.-L. Guayder. Transmission loss of double panels filled with porogranular materials. Journal of the
Acoustical Society of America, 126:30403048, 2009.
[9] R. J. M. Craik and R. S. Smith. Sound transmission through double leaf lightweight partitions, part i: airborne sound
transmission. Applied Acoustics, 61(223-245), 2000.
[10] O. Doutres and N. Atalla. Acoustic contributions of a sound absorbing blanket placed in a double panel structure:
Absorption versus transmission. Journal of the Acoustical Society of America, 128:664671, 2010.
[11] O. Doutres and N. Atalla. Experimental estimation of the transmission loss contributions of a sound package placed in a
double wall structure. Applied Acoustics, 72:372379, 2011.
[12] O. Doutres, N. Dauchez, J.-M. G
enevaux, and O. Dazel. Validity of the limp model for porous materials: A criterion
based on the biot theory. Journal of the Acoustical Society of America, 122(4):20382048, 2007.
[13] C. Fernandez, C. Soize, and L. Gagliardini. Sound-insulation layer modelling in car computational vibroacoustics in the
medium-frequency range. Acta Acustica united with Acustica, 96:437444, 2010.
[14] S. Finnveden. Evaluation of modal density and group velocity by a finite element method. Journal of Sound and Vibration,
173:113124, 2004.
[15] S. Finnveden and M. Barbagallo. A cavity-wall element for statistical energy analysis of double walls. Technical Report
TRITA-AVE 2012:02 ISSN 1651-7660, AVE/MWL, 2012.
[16] Y. C. Fung and P. Tong. Classical and Computational Solid Mechanics. World Scientific, 2001.
[17] S. Ghinet, N. Atalla, and H. Osman. The transmission loss of curved laminates and sandwich composite panels. Journal
of the Acoustical Society of America, 118:774790, 2005.
[18] P. G
oransson. Convergence aspects in fnite element solutions of multilayered, heterogenous porous materials. In Proceedings
of SAPEM, 2011.
19
[19] L. Jaouen, A. Renault, and M. Deverge. Elastic and damping characterizations of acoustical porous materials: Available
experimental methods and applications to a melanine foam. Applied Acoustics, 69:1129 1140, 2008.
[20] P. Khurana, L. Boeckx, W. Lauriks, P. Leclaire, O. Dazel, and J. F. Allard. A description of transversely isotropic sound
absorbing porous materials by transfer matrices. Journal of the Acoustical Society of America, 125(2):915921, 2009.
[21] E. E. Ungar L. Cremer, M. Heckl. Structure-Borne Sound. Springer-Verlag, 2nd edition, 1988.
[22] W. Lauriks. The acoustic transmission through layered systems. Journal of Sound and Vibration, 155:125132, 1992.
[23] C.-M. Lee and Y. Xu. A modified transfer matrix method for prediction of transmission loss of multilayer acoustic
materials. Journal of Sound and Vibration, 326:290301, 2009.
[24] J. Legault and N. Atalla. Numerical and experimental investigation of the effect of structural links on the sound transmission of a lightweight double panel structure. Journal of Sound and Vibration, 324:712732, 2009.
[25] J. Legault and N. Atalla. Sound transmission through a double panel structure periodically coupled with vibration
insulators. Journal of Sound and Vibration, 329:30823100, 2010.
[26] H. Liu, M. Barbagallo, and S. Finnveden. Wave motion and sound transmission loss of double walls filled with porous
materials. Proceedings of ISMA conference, 2012.
[27] P. M. Morse and H. Feshbach. Methods of Theoretical Physics. McGraw-Hill, 1953.
[28] C.-M. Nilsson and S. Finnveden. Input power to waveguides calculated by a finite element method. Journal of Sound and
Vibration, 305:641658, 2007.
[29] R. Panneton. Comments on the limp frame equivalent fluid model for porous media. Journal of the Acoustical Society of
America, 122(6):EL217EL222, 2007.
[30] R. Panneton and N. Atalla. Numerical prediction of sound transmission through finite multilayer systems with poroelastic
materials. Journal of the Acoustical Society of America, 100:346354, 1996.
[31] R.G. DeJong R.H. Lyon. Theory and Application of Statistical Energy Analysis. Butterworth-Heinemann, 2nd edition,
1995.
[32] D. Rhazi and N. Atalla. A simple method to account for size effects in the transfer matrix method. Journal of the
Acoustical Society of America, 127(2):EL30EL36, 2010.
[33] S. Rigobert, F. C. Sgard, and N. Atalla. A two-field hybrid formulation for multilayeres involving poroelastic, acoustic,
and elastic materials. Journal of the Acoustical Society of America, 115(6):27862797, 2004.
[34] S. Schneider and P.-O. Mattei. Sound transmission through multilayered shells. Experiments and modelling. Acta Acustica
united with Acustica, 93:5762, 2007.
[35] O. Tanneau, P. Lamary, and Y. Chevalier. A boundary element method for porous media. Journal of the Acoustical
Society of America, 120:12391251, 2006.
[36] M. Villot, C. Guigou, and L. Gagliardini. Predicting the acoustical radiation of finite size multi-layered structures by
applying spatial windowing on infinite structures. Journal of Sound and Vibration, 245(3):433455, 2002.
20
Paper 4
!
!"#$#!%&$'(#%')*+),+%"&+'*+-#!.)+-'/!)&0#/%'!+1#%&$'#0+2$)2&$%'&/+
),+2)$)./+,)#1/+./&3+'*+-&"'!0&/4+
#$%&'(")*&&$(+(&,-".*/0123*4"563/*&-"7%'8*%0"9%3:%;%//<"
7=>-"?(3<&%@'*A%/"%&+"B(8*A/("2&;*&((3*&;"
CD5-"'8("E<F%/"G&0'*'@'("<H"D(A8&</<;F"
#21!II"JJ-"#'<A48</K"
#L(+(&"
#5/%$#!%+
B*0A<(/%0'*A"M3<M(3'*(0"<H"M<3<@0"K%'(3*%/0"'FM*A%//F"@0(+"*&"$(8*A/(0"%3("(0'*K%'(+"H3<K"
K(%0@3(K(&'0"%&+"*&$(30("H*&*'("(/(K(&'"M3<A(+@3(0N"D8("K(%0@3(K(&'0"%3("'%4(&"*&"&(%3"
$%A@@K"%&+"A<$(3"%"3%'8(3":3<%+"H3(O@(&AF"3%&;(P",I"5Q"R"!"45QN"D83(("4*&+0"<H"3(0M<&0("
%3("K(%0@3(+"%&+"'8*0"%//<L0"(0'*K%'(0"<H"'L<"$*0A<(/%0'*A"K<+@/*"%&+"M3<$*+(0"0<K("
*&H<3K%'*<&"<&"'8("+(;3(("<H"K%'(3*%/"%&*0<'3<M*A*'FN")*30'-"H3(O@(&AF"*&+(M(&+(&'"
M3<M(3'*(0"%3("(0'*K%'(+"L8(3("+*00*M%'*<&"*0"+(0A3*:(+":F"A<&0'%&'"/<00"H%A'<30N"D8(&-"
H3%A'*<&%/"+(3*$%'*$("K<+(/0"%3("%+%M'(+"'8%'"A%M'@3("'8("$%3*%'*<&"L*'8"H3(O@(&AF"<H"
0'*HH&(00"%&+"+%KM*&;N"D8("%;3((K(&'":('L((&"K(%0@3(+"%&+"A%/A@/%'(+"3(0M<&0("M3<$*+(0"
0<K("A<&H*+(&A("*&"'8("M3(0(&'(+"M3<A(+@3(0N"G'"*0-"8<L($(3-"&<'(+"'8%'"*H"'8("K<+(/"
A%&&<'"H*'"'8("K(%0@3(K(&'0"L(//-"'8("*&$(30("M3<A(+@3("F*(/+0"%"A(3'%*&"+(;3(("<H"
%3:*'3%3*&(00"'<"'8("M%3%K('(3"(0'*K%'*<&N"
"
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
!
"D8*0"%3'*A/("*0"*&'(&+(+"H<3"'8("0M(A*%/"*00@("<&"S<3<@0"K%'(3*%/0"
"0$%&'(HT4'8N0("
!"
"
!"#!$%&'()*%!'$###
#$%"&'()(*$+,$-"+,..,/'0(1",$"*$1"23%+,/(,&$"&."4,53&*/&')(,/"3%)2&$)%",)"(6%"+%.,$,(,&$"&."(6%"
7*(%3,*0)"*$+")(3'/('3%)")('+,%+8"9."(6%)%":%3%";$&:$<"5%((%3"+%),-$)"/&'0+"5%".&'$+"51"
&2(,7,=,$-"23&/%+'3%)"0%*+,$-"(&"/6%*2%3"*$+">',%(%3"23&+'/()8"?6%$<"%8-8<"4%6,/0%)"7,-6("
6*4%"*")7*00%3"%$4,3&$7%$(*0",72*/("*)"(6%1"/&'0+"5%"7*+%"0,-6(%3":6%$"(6%,3"*/&')(,/*0"
2%3.&37*$/%":*)",723&4%+8"
"
9$/3%*),$-01"4%6,/0%"*$+"/&$)(3'/(,&$",$+')(3,%)"*3%",$(3&+'/,$-"@$%:A"7*(%3,*0)"
7*+%"&.<"%8-8<"2&017%3)<"3'55%3<":&&+"*$+".*53,/)8"#$%"2'32&)%"&."(6%)%"7*(%3,*0)"7*1"5%"
(&"23&4,+%"7%/6*$,/*0"+*72,$-"*$+<".&3"7&)("7*(%3,*0)<"+*72,$-",)"0*3-%)("*("(6%"
.3%>'%$/,%)"*$+"(%72%3*('3%)"*(":6,/6"(6%,3"23&2%3(,%)"/6*$-%)"7&)(">',/;01"B8"?6')<"*$1"
23%+,/(,&$"/*$"&$01"5%")'//%)).'0",."(6%".3%>'%$/1"+%2%$+%$("4,)/&%0*)(,/"23&2%3(,%)"&."(6%"
7*(%3,*0)",$4&04%+"*3%";$&:$8"#$%"$&(&3,&')01"+,..,/'0("-3&'2"&."7*(%3,*0)"*3%"2&3&')"%0*)(,/"
7*(%3,*0)<"/&77&$01"+%)/3,5%+"51"C,&(A)"%>'*(,&$)!DE8?6%")(*(%"&."*3(",$"/6*3*/(%3,=,$-"
4,)/&%0*)(,/"2&3&')"7*(%3,*0)"')%+",$"4%6,/0%",$+')(31",)":%00"+%)/3,5%+",$"*"(6%),)F<"*"3%4,%:"
*3(,/0%G"*$+"*";%1$&(%"0%/('3%H8""
"
?6%"7*$'.*/('3,$-"&.".&*7)"*$+".,53&')"7*(%3,*0)"')%+",$"4%6,/0%"(3,7"/&72&$%$()"
,)")'/6"(6*(".&*7)"7*1"$&("5%",)&(3&2,/":6,0%",("7,-6("5%"(6*("(3*$)4%3)%",)&(3&2,/"7*(%3,*0"
7&+%0)":&'0+"+&<")%%<"I%."E<"J6*28"BK8"L%4%3(6%0%))<"7&)("7%*)'3%7%$("23&/%+'3%)"*3%"
&$%D*M,*0<"-,4,$-",$.&37*(,&$"*5&'("&$%")(,..$%))"2*3*7%(%3GDN8"O&3%&4%3<"(6%"
7%*)'3%7%$()"*3%"&.(%$">'*),")(*(,/"&3"*3%"5*)%+"&$"(6%".3%>'%$/1"*$+"5*$+:,+(6"&."*".%:"
3%)&$*$/%)":6,0%"(6%"7*(%3,*0)"*3%"4,)/&%0*)(,/"*$+"(6%,3"23&2%3(,%)"*3%"/6*$-,$-"&4%3"*"
53&*+".3%>'%$/1"5*$+8"O&)("&.(%$"%M2%3,7%$(*0"+*(*"*3%"%4*0'*(%+"5*)%+"&$"(6%"
*))'72(,&$"(6*("(6%"*,3":,(6,$"(6%".&*7"+&%)"$&("),-$,.,/*$(01",$.0'%$/%"(6%"&5)%34%+".3*7%"
!"
"
#$%&$'(")*&+"&+",-./%/.0-1"-(2(3"4&2("53"&'"6-7%&$'"89".-0$:3"+*$:+"%*/%"%*-"/&;"*/+"/'"-<<-7%"$'"
%*-"<;/#-=+"#$%&$'1"+--"/0+$">-<("?("""
"
)*-"<$00$:&'2",-+7;&.-+"#-/+@;-#-'%"A;$7-,@;-+"%*/%"$B-;7$#-"+$#-"$<"%*-"
,&+/,B/'%/2-+"$<"7@;;-'%"A;$7-,@;-+("C;$7-,@;-+",-B-0$A-,".D">&7-3"EF;/'++$'"-%(/0(":/+"
%*-"+%/;%&'2"A$&'%"<$;"%*-":$;GHI3HH(")*@+3"%*-"#-/+@;-#-'%+":-;-"#/,-"&'"'-/;"B/7@@#"/',"
&'B-;+-"4J"A;$7-,@;-+"/;-"@+-,"%$"<&',"%*-"K.-+%L"#/%-;&/0",/%/(")*-"7@;;-'%":$;G"/,,-,"
-M7&%/%&$'+"%*/%"A;$B$G-,".$%*"7$#A;-++&$'/0"/',"+*-/;"#$%&$'"+$"%*/%"&%":/+"A$++&.0-"%$"
-+%&#/%-"%:$"-0/+%&7"#$,@0&3":*&0-"2/&'&'2"+$#-"&'<$;#/%&$'"$'"%*-"/'&+$%;$A&7&%D"$<"%*-"
#/%-;&/0("N"<&;+%";&2"/77$#A0&+*-,"%*&+":&%*"/'"/+D##-%;&7"+-&+#&7"#/++3":*&7*"&'+A&;-,"%*-"
+&#&0/;""@+-"$<"/'"/+D##-%;&7"+-&+#&7"#/++"&'"O@-'7/"/',"EF;/'++$'=+"'@#-;&7/0"
&'B-+%&2/%&$'+HP("N"+-7$',";&2"@+-,".$%*"*$;&Q$'%/0"/',"B-;%&7/0"-M7&%/%&$'+("8'":*/%"<$00$:+3"
6-7%&$'"88",-+7;&.-+"%*-"#-/+@;-#-'%";&2("6-7%&$'"8883"A;-+-'%+"%*-"&'B-;+-"<&'&%-"-0-#-'%"R4JS"
A;$7-,@;-"<$;"A/;/#-%-;"-+%&#/%&$'"/',"%*-",&<<-;-'%"7$'+%&%@%&B-"#$,-0+"%*/%"/;-"@+-,("
6-B-;/0"%;/'+<-;"<@'7%&$'":-;-"#-/+@;-,"&'"/";/%*-;".;$/,"<;-T@-'7D"./',"RPI"UQ"V"H"GUQS"
/',"&%":/+"%*-;-<$;-"A$++&.0-"%$"<&%"<;-T@-'7D",-A-',-'%"B&+7$-0/+%&7"#$,-0+(""6-7%&$'"89"
A;-+-'%+"%*-";-+@0%+"/',"6-7%&$'"9"7$'70@,-+"%*-"/;%&70-(""
"
!!"#$%&'()%$%*+#)!,#&*-#+%'+#'&$./%'#
40$$;"#/%+"@+-,"&'"%;@7G+"/',"7/;+"$<%-'"7$'+&+%"$<"A$;$@+"A$0D@;-%*/'-"<$/#"7$B-;-,"$'"
%*-"%$A"+&,-".D"/"!"#$%&'#%"("#/,-"$<";@..-;(")*-"*-/BD"0/D-;"/,*-;-+",&;-7%0D"%$"%*-"
A$0D@;-%*/'-"7$;-",@;&'2"%*-"#/'@</7%@;&'2"A;$7-++3"+$"%*-;-"&+"'$"/,,&%&$'/0"/,*-+&B-"
&'B$0B-,("8'"B-*&70-+3"<0$$;"#/%+"*/B-"B/;D&'2"%*&7G'-++",@-"%$"%*-"7$#A0-M"2-$#-%;D"$<"%*-"
<0$$;+"/',"%*-D"/;-"'$;#/00D"70/#A-,"/%"%*-"/%%/7*#-'%"A$&'%+"$<"<0$$;"#$@'%-,"
!"
"
#$%&'(#)*+",%-."/,",#/*,+"01"023*,"/)4"01"4225,*#'"'3/,*&-"-26#5,+"7.&-."-./)8#,"*.#"
(#-./)&-/3"'52'#5*&#,9:;"<2,*"2=*#)+"*.#5#"&,")2"/4.#,&6#"0#*7##)"*.#"=3225"(/*,"/)4"*.#"
(#*/3",%5=/-#;">25"6&052/-2%,*&-"'5#4&-*&2),"&*"&,"*.#"(/*#5&/3"'52'#5*&#,"2="*.#"-2('2,&*#"
=2/(?".#/61"3/1#5"*./*"&,"2="&)*#5#,*;"
"
@)"*.#"'52A#-*"!"#!$#%,&('3&=&#4"=3225"(/*",'#-&(#),"7&*."4&(#),&2),"2="9;B"C"9;B"
(D"/)4"-2),*/)*"*.&-E)#,,"7#5#"(/)%=/-*%5#4;""F.#",2%)4"*5/),(&,,&2)"32,,"2="8#)#5&-"*5&("
'/)#3,"(/4#"2="/"(/*"/)4"/",&)831"-%56#4"/3%(&)%(",.#33"7/,"(#/,%5#4+"/,"7&33"0#"
4#,-5&0#4"&)"=25*.-2(&)8"5#'25*,;"F72"*5%-E"(/*,"/)4"2)#"-/5"(/*"7#5#"-2),&4#5#4"/)4"
*.#,#"./4",3&8.*31"4&==#5#)*"*.&-E)#,,#,"7.&-."7#5#"/02%*"DG"((;"F#,*",/('3#,"7#5#"-%*"2%*"
2="*.#,#",'#-&(#),"/)4"*.5##"*#,*,"7#5#"(/4#H"IJ"F.#"=2/(,K"=327"5#,&,*&6&*1"7#5#"
(#/,%5#4"&)"*.#"2)#?/C&/3"5&8"4#,-5&0#4"&)"L#=;"!M"NJ"F.#".#/61"3/1#5,K",.#/5",*&==)#,,"7#5#"
(#/,%5#4"&)"*.#"5&8"4#,-5&0#4"&)"L#=;"9!+"I''#)4&C"OM"PJ"F.#"=2/(,K"=5/(#",*&==)#,,"7#5#"
#,*&(/*#4"01"/"(#/,%5#(#)*"'52-#4%5#"4#,-5&0#4"&)"7./*"=23327,;"
"
>25"6&05/*&2)"*#,*&)8+",/('3#,"7#5#"-%*"2%*+",##">&8;"9;"F.#",/('3#,"7#5#"(2%)*#4"
2)"/",*##3"*/03#"7.&-."&)"*%5)"7/,"(2%)*#4"2)"/)"#3#-*52"41)/(&-",./E#5;"@)"*.#"=&5,*"5&8"
*.#5#"7/,"2)"*2'"2="*.#".#/61"3/1#5"/",(/33",*##3"(/,,"*./*"'5262E#4"D?Q"(2*&2)"7.&3#"&)"
*.#",#-2)4"5&8"*.#"#C-&*/*&2)"7/,"02*."&)"*.#"6#5*&-/3"/)4".25&R2)*/3"4&5#-*&2),+"/,",.27)"&)"
>&8;"D;"F.#"(#/,%5#(#)*,"7#5#"(/4#"&)"*.#"6/-%%("-./(0#5",.27)"&)">&8;":;"F.#"/&5"&)"*.#"
-./(0#5"7/,"#6/-%/*#4",2"*./*"*.#"'5#,,%5#"&),&4#"7/,"/''52C&(/*#31"9"S"2="/*(2,'.#5&-"
'5#,,%5#;"@)"*.#"=&5,*"(#/,%5#(#)*"5&8+"4#'&-*#4"*2"*.#"3#=*"&)">&8;"D+"*72"*5/),=#5"=%)-*&2),"
7#5#"(#/,%5#4H"6#5*&-/3"/--#3#5/*&2)"2="*.#"*/03#+"2)"7.&-."*.#",/('3#"&,"(2%)*#4+"*2"*.#"
6#5*&-/3"/--#3#5/*&2)"/*"*.#"*72"/--#3#52(#*#5,",.27)"&)"*.#"=&8%5#;"@)"*.#",#-2)4"5&8+"*.5##"
*5/),=#5"=%)-*&2),"7#5#"(#/,%5#4H"TU?UJ"6#5*&-/3"*/03#"6#32-&*1"*2"6#5*&-/3"6#32-&*1"/*"*.#"*2'"
2="*.#",/('3#M"TV?UJ".25&R2)*/3"*/03#"6#32-&*1"*2"6#5*&-/3"6#32-&*1"/)4"TV?VJ".25&R2)*/3"*/03#"
!"
"
#$%&'()*")&"+&,(-&.)/%"#$%&'()*"/)")+$")&0"&1")+$"2/30%$4"5+$"#$,)('/%"2/30%$"#$%&'()*"6/2"
/#$,/7$8"(."1&9,"0&2()(&.24"
"
5+$"1(,2)"3$/29,$3$.)",(7"1/(%$8"2(.'$:"8$20()$"%/,7$"$11&,)2:"3$/29,$3$.)2"/.8"
'/%'9%/)(&.2"6&9%8".&)"/7,$$4"";.$",$/2&."1&,")+(2"'&9%8"<$")+/)"&.%*")6&",$20&.2$2"6$,$"
&<2$,#$8"6+(%$:"(.")+(2"$11$')(#$%*")6&=8(3$.2(&./%"2$)"90:")+$,$"/,$")+,$$"19.8/3$.)/%"
3&8$2"&1"#(<,/)(&."/.8"()"6/2")+$,$1&,$"8(11('9%)")&"'&,,$%/)$"&<2$,#$8"1,$>9$.'($2"6()+"
3&8$"2+/0$24"?.&)+$,",$/2&."'&9%8"<$"%/'@"&1"0,$'(2(&."(.")+(2"$A0$,(3$.)/%"2$)904"5+$"
)+(,8"/.8"3&2)"%(@$%*",$/2&."(2")+/)")+$"/''$%$,&3$)$,2"6$,$",/)+$,"%/,7$"BC4D"7E"/.8")+$(,"
3&)(&."3(7+)"+/#$"<$$."/11$')$8"<*"%&'/%"1%$A(<(%()*"&1")+$",9<<$,"&,")+$"(30$8/.'$"&1")+$"
$%$'),('"%$/82:"6+('+"/,$",/)+$,"2)(114""
"
"5+$"2$'&.8",(7"6/2"'&.2),9')$8"29'+")+/)"/%%"8(11('9%)($2")+/)"'&9%8"<$"(3/7(.$8"
6$,$",$3&#$84"5+92:")+$"/2*33$),('/%%*"0%/'$8"2)$$%"3/22"6/2",$3&#$8"/.8"(.2)$/8")+$"
2+/@$,"1(,2)"$A'()$8")+$"2),9')9,$"(.")+$"#$,)('/%"8(,$')(&."/.8")+$."(.")+$"+&,(-&.)/%"
8(,$')(&.4"5+$")/<%$"&."6+('+")+$"2/30%$"(2"3&9.)$8"6/2"2920$.8$8"(."20,(.72"/.8")+$"
2+/@$,"/')$8"#(/"/",&84"?%2&:")&",$89'$")+$"(.1%9$.'$"&1")+$",9<<$,"%/*$,:"/"#$,*"2)(11"/.8"%(7+)"
BF4G"7E"'&30&2()$"6/2"0%/'$8"&.")+$")&04"5+$"8(2/8#/.)/7$"6()+")+(2"2$)"90"(2")+/)"1&/3"
2),$22"(.")+$"+&,(-&.)/%"8(,$')(&."'/92$8"<*"<$.8(.7"&1")+$",9<<$,"'/..&)"<$"&<2$,#$84"?%2&"
)+$"3$/29,$3$.)2",$>9(,$"3&,$")(3$"/2")+$"%/2$,"+/2")&"<$",$0&2()(&.$8"3/.*")(3$2"/.8"
)6&"$A'()/)(&.2"/,$".$$8$84"5+$"7,$/)"/8#/.)/7$"(2")+/)"6$".&6"+/8"7&8",$0$/)/<(%()*"/.8"
8/)/")+/)"'&9%8"<$"1())$8")&"'/%'9%/)(&.24"
!"
"
!!!"#$%&'($%&!)%*#$+,(*-#
%"#&./#012/34/#5/6.78#
#$"%$&'()'"*(+,'-.('"/0)".)'-"1+"2%$-"13'"(%4%-%1%')"+2"13'"2+05")05*6')7"83.)9":(+0-":0$-"
&%:(01%+$)"%$"13'"2(';.'$,<"(0$4'"=>"?@"A"B"C?@"/'('"'D,%1'-"0$-"5'0).('-"%$"13'"(%4"0$-"
13'")'1.*"/0)")%5.601'-"%$"0$"EF")+21/0('7"83'",06,.601%+$)"/'('"('G(.$".$1%6"0").22%,%'$16<"
4++-"2%1"1+"5'0).('5'$1)"/'('"2+.$-7"H0+.'$"'17067"('*+(1"1301").,3"%$&'()'"*(+,'-.(')"0('"
&'(<"-'50$-%$4I"0$-"13%)"*(+&'-"1+":'"(%431"/3'$".)%$4"0",+55'(,%06"EF"*0,J04'7"K%13"
3+5'":('/'-")+21/0('BI"1301",+.6-":'"10%6+('-"1+"13'"*(+:6'5"01"30$-9":+13",+5*.101%+$)"
0$-"5+-'6%$4"/'('9"3+/'&'(9",+$&'$%'$16<"50-'7"83'"'5*6+<'-")*',%06"=GL";.0-(01%,"
M04(0$4'"1<*'"'6'5'$1)"0('"-+,.5'$1'-"%$"N'27"BI7"83'"5')3"-'*%,1'-"%$"E%47"O"%)".)'-"%$"
5+)1",06,.601%+$)P"%1"30)"!"D"!"'6'5'$1)"%$"13'"2+057""#",+0()'"5')3"/%13"Q"D"Q"2+05"
'6'5'$1)"0$-"0"2%$'"5')3"/%13"B="D"B="2+05"'6'5'$1)"/'('"06)+"'5*6+<'-7"""
"
"83'"+*1%5%@01%+$"/0)"50-'"/%13"13'"R0160:"(+.1%$'"S6);,.(&'2%175T"/3%,3"%)"
&'()01%6'"0$-"'22%,%'$1":.1"(';.%(')"0"4++-")10(1"0**(+D%501%+$"0)"%1"2%$-)"0"6+,06"+*1%5.57"
8+"1301"'$-9"13'",+0()'"=GL"5+-'6"/0)"2%()1".)'-"1+"2%$-"2(';.'$,<"%$-'*'$-'$1"(%4%-%1%')"2+("
0$"%)+1(+*%,"501'(%06P"13%)"1++J"0"5%$.1'"+(")+"+$"0")10$-0(-"UV7"W'&'(06"('G(.$)"/'('"50-'"
1+"'$).('"1301"13'"(').61"/'('"%$"13'"&%,%$%1<"+2"0"46+:06"5%$%5.57"83'"%$%1%06"&06.')"2+("13'"
2%()1",06,.601%+$"0('"-'2%$'-":<"13'"1(0$)2'("2.$,1%+$":'1/''$"&'(1%,06"5+1%+$"+2"13'"1+*"
*601'"0$-"13'"&'(1%,06"-%)*60,'5'$1"+2"13'"10:6'7"83'"(')+$0$,'"2(';.'$,<")''$"5+)1",6'0(6<"
%$"13%)"2.$,1%+$9"0(+.$-"BQI"?@"%$"E%47"I9"%)"1+"0"20%("0**(+D%501%+$"4%&'$":<""
"
f =
1
2!
AE
""
h ( m f + mc )
!"
"
XBY"
#$%&%" E ""'(")*+,-.("/*0+1+("*2"3$%"2*4/5" A ""'("3$%"(4/61%.("7&*(("(%73'*,41"4&%45" h ""'("'3("

$%'-$35" m f ""'("$412"3$%"/4(("*2"3$%"2*4/"4,0" m c ""'("3$%"3*341"/4(("*2"3$%"&+88%&"4,0"3$%"(3'22"
3*6"6143%9":$%"&%(+13("2&*/"3$%"2'&(3"*63'/';43'*,"&+,"(%&<%0"4("','3'41"<41+%("3*"/*&%"
7*/61'743%0"/*0%1(9":$%"2',41"2&%=+%,7>"0%6%,0%,3"?@A"/*0%1"#'3$"%1%/%,3"0'/%,('*,("
1%(("3$4,"?"//"#4("',"BCDB"*63'/';%0"*<%&"4",'-$3"*,"4"(34,04&0"EF9""
"
:$%"*63'/';43'*,"&*+3',%"/','/';%("3$%"#%'-$3%0"/%4,"(=+4&%"%&&*&5" Wi G"
Wi = Re ( X im ! X ic ) + Im ( X im ! X ic )
"
X im "5"
#$%&%" X im ""'("4"/%4(+&%0"4,0" X ic "'("4"7417+143%0"3&4,(2%&"2+,73'*,9""J2"3$%"&%143'<%"%&&*&"#4("

/','/';%05"3**"/+7$"%/6$4('("#*+10"8%"6+3"*,"4,3'@&%(*,4,7%("4,0"*,"K+,',3%&%(3',-L"
64&3("*2"3$%"3&4,(2%&"2+,73'*,5"#$'1%"'2"3$%"48(*1+3%"%&&*&"'("+(%0"*,1>"3$%"&%(*,4,7%("
0%3%&/',%"3$%"&%(+13M"3$%"#%'-$3%0"%&&*&"',"N=9"HBI"6&*<'0%("4"7*/6&*/'(%9"""
"
:$%"*63'/';43'*,"7*,('0%&("3$%"(+/"*2"411" Wi "*<%&"411"7*,('0%&%0"2&%=+%,7'%("4,0"4"
-**0"7+&<%"2'3"&%1'%("*,"4"-**0"7$*'7%"*2"2&%=+%,7>"84,0"4,0"2&%=+%,7>"(%64&43'*,9"O%&%5"4"
1*-4&'3$/'7"(647',-"#4("+(%0"4("3$'("6&*<'0%("%=+41"#%'-$3"3*"0'22%&%,3"*734<%"84,0("4,0"
41(*"3*"0'22%&%,3"&%(*,4,7%(5"'2"3$%>"$4<%"%=+41"1*(("2473*&(9"P66&*Q'/43%1>"BCC"2&%=+%,7'%("
#%&%"',71+0%0"',"3$%"6&*7%0+&%9":$%"1*-4&'3$/'7"2&%=+%,7>"#%'-$3',-"(%&<%0"3$%"7+&&%,3"
6+&6*(%M"*3$%&"#%'-$3',-("#*+10"6*(('81>"8%"/*&%"+(%2+1"',"*3$%&"4661'743'*,("4,0"'3"'("
&%7*//%,0%0"3*"7*/64&%"0'22%&%,3"#%'-$3',-"2+,73'*,("3*"7*,2'&/"3$%"&*8+(3,%(("*2"3$%"
6&*7%0+&%(9"
!"#$%&'()(*()+,#-,./()%&'#
J,"R'*3"3$%*&>5"3$%"7*,(3'3+3'<%"&%143'*,("4&%""
!"
"
HBI"
! ij = ( Cijkl + Qij Qkl R ) " kl + Qij e; s = R e + Q " ii #"
"
$%&"
'()*)"+()",-+.+/-,"/0"1)2/,)1"/,""3)24"564"7()"8*)0),+"'-*9":-,0/1)*0"1)2-*;.+/-,".+"
:-,0+.,+#"/4)4#"<)*-#"2=>/1"8-*)"8*)00>*)".,1"?@4"$%&"0/;8=/2A"+-"+()":-,0+/+>+/B)"*)=.+/-,"2-*".,"
)=.0+/:"0-=/1#""
! ij = Cijkl " kl 4"
"
$C&"
"
D-*":-;8>+)*"/;8=);),+.+/-,"/+"/0":-,B),/),+"+-")E8*)00"?@4"$C&">0/,F";.+*/E".=F)G*."
! = D " "#"
"
$H&"
'()*)" " "/0"+()"),F/,))*/,F"0+*./,"B):+-*#"'(/:("/,"+'-".,1"+(*))"1/;),0/-,0"/0"F/B),"GA"

" = !#" x
"
" y # xy "$ ; " = !#" x " y " z # xy # xz # yz "$

T
"x =
$u
$ ux
$u
, etc; # xy = y + x , etc.
$x
$x
$y
""
$6&"
! ""/0"+()"0/;/=.*=A"1)2/,)1"),F/,))*/,F"0+*)00"B):+-*# D ""/0"+()"0A;;)+*/:"*/F/1/+A";.+*/E".,1"
u = !# u x
"
uy
u z "$ "/0"+()"2*.;)"1/08=.:);),+"/,"."I.*+)0/.,":--*1/,.+)"0A0+);4""
7()" ;-0+" :-;8=)E" ;.+)*/.=" ;-1)=" :-,0/1)*)1" ()*)" /0" +()" +*.,0B)*0)" /0-+*-8/:#"
'()*)"+()"B)*+/:.="A!1/*):+/-,"/0"+()"+*.,0B)*0)"1/*):+/-,4"7()"*/F/1/+A";.+*/E"/0"+(),"/,"J".,1"
%"1/;),0/-,0"F/B),"GA"
"
! D11
%
D = % D12
% 0
#
D12
D22
0
! D11
%
%
0 "
%
&
0 &; D = %
%
Gxy &$
%
%
#%
!"
"
D12
D13
D22
D12
D11
Gxy
sym.
Gxz
0 "
0 &&
0 &
& ""
0 &
0 &
&
Gxy $&
$K&"
#$%&'"()*"+,-"./012"&3*041"3$2*2"0*2"()%*"5032*40/".0*05232*&"016"&47"41"8,-"941"(0:3'"(4;2"0&"
D13 , D11 and Gxz "0*2"*2/0326<=">)*"01"4&)3*).4:"5032*40/'")1/?"3@)".0*05232*&"0*2"122626"016"
$2*2"3$2?"0*2"30A21"0&"3$2"B)%1CD&"5)6%/%&' E '"016"&$20*"5)6%/%&'" G '""
"
D11 = D22 =
E
1 +!
!
!
#1 +
% 1 " 2!
!
E
"
; Gxy = Gxz = Gyz = G "'"
$ ; D12 = D13 =
1 " 2! 1 +!
&
9E<"
@$2*2"F)4&&)1D&"*034)"4&"C4;21"G?"
!=
"
E
" 1 "="
2G
9!<"
!"#$%&'()*+&,%'%,-#
#$2"5032*40/&"41;2&34C0326"0*2";4&:)2/0&34:H"&22"2=C="3$2"27:2//213"3273"41"I$0.32*"J8")("
*2(2*21:2"JK=">)*"&%:$"5032*40/&"3$2"&3*2&&"62.216&")1"3$2"&3*041"$4&3)*?"016"4("3$2";0*40G/2&"
)("*2&.)1&2"0*2"62(4126"0&"(%1:34)1&")("3452'"3$2"5)6%/4" C, R and Q 41"LM="98<"&$)%/6"G2"
4132*.*2326"0&").2*03)*&'"62(4126"0&":)1;)/%34)1&")("3$2"&3*041";2/):43?"@43$"0..*).*4032"
*2/07034)1"(%1:34)1&JN'"JK,JE="#$2".*2&213"@)*A":)1&462*&"3452"$0*5)14:"5)34)1"016"3$2"
5)6%/4"0*2"3$21":)5./27";0/%26"010/?34:0/"(%1:34)1&")("(*2M%21:?="#$2&2"(%1:34)1&":011)3"
G2":$)&21"0*G43*0*4/?"G%3"5%&3"G2"&%:$"3$03"3$2*5)6?1054:"*2/034)1&"016"3$2"/0@")("
:0%&0/43?"0*2"1)3";4)/0326=""#$4&"4&"&)"()*"3$2"(*0:34)10/"62*4;034;2"5)62/J!,+O'"+J="P2*2'"3$2"
()%*,.0*05232*"()*5"4&"%&26"&)"3$03'"2=C='"B)%1CD&"5)6%/%&'" E ""4&'"()*"$0*5)14:"3452"
62.21621:2")("3$2"()*5" ei# t '"C4;21"G?""
"
E ( f ) = E0 F ( f ) ; F ( f ) =
!"
"
1 + ( i f f1 )
1 + ( i f f2 )
"'"
9JO<"
$%&'&# f #()#*%&#+'&,-&./0#1+#2(3'4*(1.#4.5# E0 , f1 , f 2 and ! 4'&#64'47&*&')#*%4*#4'&#8(2&.#30#

*%&#(.2&')&#6'1/&))9#:#)(7(;4'#+'4/*(1.4;#715&;#5&)/'(3&)#*%&#)%&4'#)*(++.&))#1+#*%&#%&420#
;40&'!<9##
!"#$%&'()*'$
+#$*,-$./012$/3456-37-$83$2.9:5-$;/<0.=/832$
=&4)-'&7&.*)#$&'&#+(')*#*4>&.#*1#(.2&)*(84*&#*%&#(.+;-&./&#1+#*%&#/1.*4(.&5#4('#1.#*%&#
*'4.)+&'+-./*(1.)#3&*$&&.#*43;&#4.5#)476;&#2(3'4*(1.)9#?.#*%&#24/--7#/%473&'#)%1$.#*1#
*%&#;&+*#(.#@(89#AB#*%&#4('6'&))-'&#/1-;5#3&#'&5-/&5#*1#466'1C(74*&;0#!D#1+#4*71)6%&'(/#
6'&))-'&9#@(89#E#)%1$)#*%'&&#*'4.)+&'#+-./*(1)#7&4)-'&5#(.#4('#4.5#(.#24/--7#(.5(/4*(.8#*%4*#
*%&#4('B#(.5&&5B#%4)#4.#(.+;-&./&9##?.#4('#*%&#5476(.8#()#)17&$%4*#;4'8&'#4.5#*%&#+'&,-&./0#
)6&/*'4#4'&#)%(+*&5#*1$4'5)#;1$&'#+'&,-&./(&)9#F%&#6'&)&./&#1+#4('#)%1-;5#.1*#5&/'&4)&#*%&#
)*(++.&))#4.5#(*#()#*%&'&+1'&#/1./;-5&5#*%4*#(*#(./'&)&)#*%&#466&'&.*#(.&'*(4#4))1/(4*&5#$(*%#
*%&#13)&'2&5#71*(1.9#F%&#'&)1.4./&)#)&&.#(.#@(8#E#5&/'&4)&)#30#4#3(*#71'&#*%4.#!"#DB#)1B#
+1'#*%&#(.2&)*(84*&5#+147#*%&#(.&'*(4#(./'&4)&)#30#G"#DH#1'#71'&#(+#*%&#4('#4;)1#(./'&)&)#*%&#
466&'&.*#)*(++.&))#1+#*%&#)476;&)B#4)#()#/1./(&243;&I9##
>#$?0-@6-37A$/3B-:-3B-3=$98B-5$
F%&#()1*'16(/#715&;B#$(*%#/1.)*4.*#+'&,-&./0#(.5&6&.5&.*#715-;(#4.5#;1))#+4/*1'B#()#*%&#
;&4)*#/176;&C#74*&'(4;#715&;9#F%&#'(8(5(*0#74*'(C#()#*%&.#8(2&.#30#J,)9#KIL#4.5#KML#(.#$%(/%#
E = E0 (1 + i" ) ; G = G0 (1 + i" )
B#
K!!L#
$%&'&#" ##()#*%&#;1))#+4/*1'9#N6*(74;#24;-&)#1+#*%&#*%'&&#64'47&*&')# E0 , G0 and " #4'&#+1-.5#

)-/%#*%4*#*%&#5(++&'&./&#3&*$&&.#7&4)-'&5#4.5#/4;/-;4*&5#*'4.)+&'#+-./*(1.)#()#7(.(7(O&5#
!"#
#
#$$%&'()*"+%"+,-".-(*,+()*"()"/01"2341""5(*1"6"$%78#&-"+,-"7#*)(+9'-"%:"+,-"7-#;9&-'"
+&#);:-&":9)$+(%);"+%"+,%;-"$#<$9<#+-'".(+,"+,-"%8+(7#<"8#&#7-+-&;":%9)'"()"=#><-"!":%&"
?=&9$@";#78<-"!A"!;+"&9)B1"=,-"#*&--7-)+"(;":#(&">9+":#&":&%7"8-&:-$+1"C"+D8($#<":-#+9&-"()"
;9$,"#"$#;-"(;"+,#+"+,-"'#78()*"(;"%E-&-;+(7#+-'A"#;"(;"$<-#&<D";--)">D"$%78#&()*"7-#;9&-'"
#)'"$#<$9<#+-'"&-;%)#)$-;"#)'"#)+(F&-;%)#)$-;1"C"$<-E-&-&"%8+(7(G#+(%)"&%9+()-"(;")--'-'"+%"
#E%('"+,-"-::-$+"+,#+A"(:"+,-"7%'-<"$#))%+"$#8+9&-"+,-":&-09-)$D"'-8-)'-)$-A"#";7%%+,"
$9&E-"8&%E('-;"+,-"<-#;+";09#&-"-&&%&1"""
"
=%"%E-&$%7-"+,-"#88#&-)+"%E-&-;+(7#+(%)"%:"'#78()*A"'(::-&-)+".-(*,+()*"
:9)$+(%);".-&-"9;-'1"H+".#;"#<;%"-;;#D-'"+%"8&-F$%)'(+(%)"+,-"7-#;9&-'";8-$+&#".(+,"<%."
%&'-&"7%'-<;A":%9)'">D"+,-"I#+<#>";D;+-7"('-)+(:($#+(%)"+%%<>%J1"K%)-"%:"+,-;-"#++-78+;"
8&%E('-'"#)D";(*)(:($#)+"(78&%E-7-)+;1"=,-")97-&($#<"-J8-&(7-)+;"&-E-#<-'A",%.-E-&A"+,#+"
+,-"7()(797"%:"+,-"-&&%&":9)$+(%)"234"(;"09(+-";,#<<%."#)'A".,(<-"+,-"-;+(7#+-;":%&"L%9)*M;"
7%'9<9;".-&-";+#><-A";%7-.,#+"'(::-&-)+"&-;9<+;":%&"+,-"<%;;":#$+%&".-&-":%9)'"(:"'(::-&-)+"
;+#&+"E#<9-;".-&-"9;-'1"N8%)"+,(;">#;(;A"+,-"<%;;":#$+%&".#;"-;+(7#+-'"'(&-$+<D":&%7"+,-"
7-#;9&-'";8-$+&#A";--"+,-"C88-)'(J1"=,(;"<%;;":#$+%&"-;+(7#+-"#)'"+,-"L%9)*M;"7%'9<9;"#)'"
O%(;;%)M;"&#+(%"-;+(7#+-;":&%7"+,-":(&;+"&9)".-&-"+,-)"9;-'"#;"()(+(#<"E#<9-;"+%"#";-$%)'"&9)1"
=,-"%8+(7#<"8#&#7-+-&;"#&-":%9)'"()"=#><-"!":%&"?=&9$@";#78<-"!A"3)'""&9)B1"=,-"&-;9<+;"#&-"
;,%.)"()"5(*;1"P"#)'A".,(<-"+,-"-&&%&":9)$+(%)"234"()"+,-";-$%)'"&9)",#;"#";<(*,+<D"<#&*-&"
E#<9-A"(+"(;"#88#&-)+"+,#+"+,-"<%;;":#$+%&"-;+(7#+-"(;">-++-&1"
"
=,-"+.%";+#*-"8&%$-'9&-".#;"#<;%"#88<(-'":%&"+,-";+(::-&"?=&9$@";#78<-"3B"#)'"+,-"
E-&D"79$,";%:+-&"?$#&";#78<-BQ";--"5(*;1"R"#)'"S1"=,-"&-;9<+;":%&"=&9$@";#78<-"3"#88-#&"+%">-"
>(#;-'">D"<%.":&-09-)$D"&(88<-;"+,#+"7(*,+",#E-">--)"$#9;-'">D"&-;%)#)$-;"()"+,-"&(*"()"
.,($,"+,-";-+"98".#;"7%9)+-'1"H+"(;";--)"()"5(*1"SA"#)'"+%"#"<-;;-&"'-*&--"()"5(*;1"PA"+,#+"+,-"
!!"
"
$%&'(&)*+#,)-&.&)-&)/#01-&2#$3,2#/1#*3./(%&#/4&#3..3%&)/#,)*%&35&#1$#/4&#$130#5/,$$)&55#35#
$%&'(&)*+#,)*%&35&56#
#
7&8&%32#01%&#*10.2&9#03/&%,32#01-&25#:&%&#3//&0./&-;#,)*2(-,)<#3#*(=,*#
5+00&/%,*#01-&2;#3#$(22#/%3)58&%5&#,51/%1.,*#01-&2#3)-#01-&25#:4&%&#/4&#-,$$&%&)/#01-(2,#
:&%&#3**10.3),&-#:,/4#-,$$&%&)/#2155#$3*/1%56#>1)&#1$#/4&5&#01-&25#.%18,-&-#5,<),$,*3)/#
,0.%18&0&)/5#3)-;#/4&%&$1%&;#/4&+#322#$&22#$1%#?*@430A5#%3B1%;#:4,*4#5/3/&5#/43/#1$#322#
.155,=2&#512(/,1)5#/4&#2&35/#*10.2,*3/&-#541(2-#438&#.%&*&-&)*&""6#C4&#%3B1%#,5#3#5(=D&*/,8&#
,)5/%(0&)/#5,)*&;#1$#*1(%5&;#:,/4#01%&#.3%30&/&%5#,)#/4&#1./,0,B3/,1)#.%1*&55#/4&#%&5(2/5#
:&%&#52,<4/2+#=&//&%6#>&8&%/4&2&55;#/4&#*(%%&)/#,)8&5/,<3/1%5#412-#/43/#/4&#,51/%1.,*#01-&2#
(5,)<#1)&#2155#$3*/1%#,5;#/4&#E=&5/F#$%&'(&)*+#,)-&.&)-&)/#01-&2;#$1%#/4&#/4%&&#,)8&5/,<3/&-#
03/&%,3256##
!"#$%&'(&)*+#,&-&),&).#/0,&1#
C4&#$%3*/,1)32#-&%,83/,8&#01-&2#G!HI#*3)#-&5*%,=&#/4&#,)*%&35&#1$#5/,$$)&55#:,/4#$%&'(&)*+#
/43/#,5#3..3%&)/#,)#J,<5#K#3)-#L6#C4&#5,0.2&5/#01-&2#,5#,51/%1.,*#:,/4#.%1.1%/,1)32#$%&'(&)*+#
-&.&)-&)*&6#C4(5;#M1()<A5#01-(2(5#3)-#/4&#54&3%#01-(2(5#3%&;#35#,)#N'6#G!HI;#<,8&)#=+#
E ( f ) = E0 F ( f ) #3)-# G ( f ) = G0 F ( f ) 6#C4,5#01-&2#,5#/4(5#-&$,)&-#=+#$,8&#.3%30&/&%5O#
E0 , G0 , f1 , f 2 and ! 6##
#
J,<6#!H#541:5#/4&#&5/,03/&5#1$#M1()<A5#01-(2(5#$1%#C%(*@#530.2&#!#,)-,*3/,)<#3#
%3/4&%#23%<&#83%,3/,1)#:,/4#$%&'(&)*+6#J,<56#!!P#!Q#*10.3%&#/4&#/%3)5$&%#$()*/,1)5#*32*(23/&-#
:,/4#/4&#$%3*/,1)32#01-&2#/1#/4&#0&35(%&-#1)&56#C4&#3<%&&0&)/5#$1%#C%(*@#530.2&#!#3)-#$1%#
/4&#R3%#530.2&#3%&#)&3%#.&%$&*/6#R1)5&'(&)/2+;#,/#,5#322#=(/#,0.155,=2&#/1#,0.%18&#/4&#
3<%&&0&)/#:,/4#01%&#*10.2,*3/&-#03/&%,32#01-&25;#35#:35#8&%,$,&-#=+#)(0&%,*32#
&9.&%,0&)/56#S)+#$(%/4&%#*1)*2(5,1)#1)#/4&#=&5/#03/&%,32#01-&2#0(5/#/4&%&$1%&#=&#=35&-#1)#
!"#
#
$%&'#()*%&$+,(%)-#'./.-#/+()'0#12#$'+34&'$'),3#%*#$%&'#,&+)3*'&#*4)5,(%)3#5%6'&()/#+#
7+&/'&#*&'84')52#&+)/'.##9:'#+/&''$'),#1',;'')#$'+34&'$'),3#+)0#5+7547+,(%)#&'347,3#(3#
7'33#<'&*'5,#*%&#9&45=#3+$<7'#>.#?,#(3#<%33(17'#,%#/+()#+#3%$';:+,#1',,'&#+/&''$'),#;(,:#+#
$%&'#5%$<7(5+,'0#$+,'&(+7#$%0'7#14,#+)2#%<,($(@+,(%)#&'347,#;%470#1'#1(+3'0#12#,:'#7%;#
*&'84')52#&(<<7'3#()#,:'#$'+34&'$'),3.#9:'#3'+&5:#*%&#+#1',,'&#$%0'7#*%&#,:(3#$+,'&(+7#(3#
,:'&'*%&'#)%,#5%)5740'0.#
!"#$%!#&'()*+#
9:'#5+7547+,(%)#&'347,3#<&'3'),'0#3%#*+&#+&'#+77#1+3'0#%)#<7+)'A3,&+()#>AB#$%0'73.#9:'#
/'%$',&2#%*#,:'#3,&45,4&'#+)0#,:'#'C5(,+,(%)#+&'#345:#,:+,#,:(3#3:%470#1'#+#/%%0#$%0'7.#9%#
6'&(*2#,:(3#<&'$(3'-#+#*477#"B#5+7547+,(%)#;+3#+73%#$+0'.#?)#,:(3#$%0'7-#,:'#0(3<7+5'$'),3D#
0'<')0')5'#%)#,:'#!"5%%&0()+,'#(3#0'35&(1'0#12#+#*%4&,:#%&0'&#<%72)%$(+7.#?)#E(/#!F#,:'#
5+7547+,'0#,&+)3*'&#*4)5,(%)3#+&'#5%$<+&'0-#3:%;()/#$()4,'#0(**'&')5'3.#9:'#"AB#$%0'7#
3:%;3#+#1(,#$%&'#0+$<()/G#(,#(3#+#;''#1(,#3%*,'&#()#,:'#:(/:#*&'84')52#&'/(%)#+)0#(,#2('703#+#
3,+,(5#H%4)/D3#$%04743#,:+,#(3#FI#3$+77'&#;:(7'#,:'#3,+,(5#3:'+&#$%04743#(3#!#I#7+&/'&.#9:'3'#
0(**'&')5'3#+&'#<&%1+172#5+43'0#12#,:'#7+5=#%*#<&'5(3(%)#()#,:'#()6'&3'#<&%5'04&'G#+#:(/:'&#
<&'5(3(%)#;%470#&'84(&'#$%&'#$'+34&'$'),3#%*#+#:(/:'&#84+7(,2.#9:'#0+,+#()#E(/.#!F#+)0#()#
9+17'#>#()0(5+,'#,:+,#,:'#<7+()#3,&+()#>AB#$%0'7#(3#34**(5('),#,%#/',#,:'#1'3,#'3,($+,'3#%*#
$+,'&(+7#0+,+#1+3'0#%)#$'+34&'$'),3#%*#,:'#=()0#+)0#84+7(,2#<&'3'),'0#:'&'.#
,"#-./-0123./2#
9:(3#+&,(57'#<&'3'),3#<&%5'04&'3#*%&#'3,($+,()/#6(35%'7+3,(5#<&%<'&,('3#%*#*%+$3#5%$$%)72#
43'0#()#6':(57'#()043,&2.#9:'#$'+34&'$'),3#+&'#,+=')#%)#,:&''#3+$<7'3#$+0'#%*#*%+$#+)0#+#
:'+62#7+2'&#$+0'#%*#&411'&-#3()5'#,:(3#;+3#:%;#,:'#$+,'&(+73#;'&'#0'7(6'&'0#*&%$#,:'#
!"#
#
$%&'(%)*'+,+-#.&#+,),&*#$,%/'+,$,&*/0#123)2#%+,#&4*#5+,/,&*,6#2,+,0#*2,#5+4),6'+,#14+7/#
8'/*#%/#1,99#(4+#/%$59,/#$%6,#4(#(4%$#4&9:-#
#
;2,#$,%/'+,$,&*#+3<#3/#59%),6#3&#%#=%)''$#)2%$>,+#/4#*2%*#*2,#$4*34&#3/#&4*#
3$5,6,6#>:#)4&*%3&,6#%3+-#;14#73&6/#4(#,?)3*%*34&#%+,#'/,6#/4#*2%*#>4*2#/2,%+#%&6#
)4$5+,//34&%9#$4*34&#)%&#>,#4>/,+=,6-#@&#,%+9:#+3<#%))4$593/2,6#*23/#13*2#%#+3<36#$%//#>'*#
*2,#$,%/'+,$,&*#5+,)3/34&#1%/#&4*#/'((3)3,&*-#.&/*,%60#%#)9,%&,+#+3<#1%/#6,/3<&,60#123)2#
2%6#*2,#6+%1>%)7#4(#+,A'3+3&<#*14#,?)3*%*34&/-#@&#3&=,+/,#BC#5+4),6'+,#1%/#'/,6#*4#(3&6#*2,#
D>,/*E#$%*,+3%9#5%+%$,*,+/0#12,+,#D>,/*E#3&63)%*,/#*2,#/$%99,/*#63((,+,&),#>,*1,,&#
$,%/'+,6#%&6#)%9)'9%*,6#*+%&/(,+#('&)*34&/-#.*#1%/#&4*,6#*2%*#/4$,#)4&/36,+%*34&/#%+,#
+,A'3+,6#12,&#)244/3&<#1,3<2*3&<#('&)*34&#%&6#(+,A',&):#+%&<,#(4+#*2,#$,%/'+,$,&*/-###
#
B3+/*0#%#(+,A',&):#3&6,5,&6,&*#3/4*+453)#$%*,+3%9#$46,9#1%/#%5593,6-#.*#<%=,#%#A'3*,#
<446#(3*#*4#$,%/'+,$,&*/-#;2,#,/*3$%*,/#(4+#F4'&<G/#$46'9'/#1,+,#+4>'/*#>'*#*2,#6%$53&<#
,/*3$%*,/#1,+,#/,&/3*3=,#*4#/*%+*#=%9',/#(4+#*2,#45*3$3H%*34&#5+4),//-#.*#3/#/'<<,/*,6#*2%*#
3&6,5,&6,&*#,/*3$%*,/#4(#6%$53&<#>%/,6#4&#+,/4&%&),G#>%&6136*2#%+,#'/,6-#I,+2%5/#3*#
14'96#>,#%#<446#36,%#*4#3&)4+54+%*,#/')2#,/*3$%*,/#3&*4#*2,#1,3<2*3&<#('&)*34&-#
#
@#(4'+#5%+%$,*,+#(+%)*34&%9#$46,9#1%/#*2,&#%**,$5*,6#*4#6,/)+3>,#*2,#$%*,+3%9/G#
(+,A',&):#6,5,&6,&),-#;2,#/3$59,/*#$46,90#%&#3/4*+453)#$46,9#3&#123)2#*2,#$46'93#2%=,#
*2,#/%$,#(+,A',&):#6,5,&6,&),0#<%=,#/')2#<446#+,/'9*/#*2%*#%&:#('+*2,+#/4523/*3)%*34&#1%/#
6,,$,6#3$54//3>9,-#.*0#*2,+,(4+,0#%55,%+/#%/#3(#*2,#/*'63,6#(4%$/#2%=,#&,%+#3/4*+453)#
5+45,+*3,/#*24'<2#$4+,#,9%>4+%*,#$,%/'+,$,&*/#%+,#&,,6,6#*4#+,/49=,#*23/#A',/*34&-#.*#3/#
(3&%99:#&4*,6#*2%*#3(#*2,#)4$5'*%*34&%9#$46,9#)%&&4*#6,/)+3>,#*2,#$,%/'+,$,&*/0#,-<-0#3(#*2,#
$%*,+3%9#$46,9#3/#&4*#%))'+%*,0#*2,#3&=,+/,#5+4),6'+,#3/#+3/7:#%&6#*2,#,/*3$%*,/#2%=,#*4#>,#
/'554+*,6#>:#3&6,5,&6,&*#,/*3$%*,/#<3=,&#>:0#,-<-0#/*%*3)#$,%/'+,$,&*/#4+#*2,#%&%9:/3/#4(#
*2,#/:/*,$#+,/4&%&),/-#
!"#
#
!"#$%&'()*(+($,$%&#'()%*+,#-+')&.(//0#'123*4/&5-&#)%&#6(+*7&'3#8*99:,,:*3#.*+#)%&:+#,(77*+)#*.#)%&#
8*//';*+'):<&#=+*>&1)#8=?@!A"BA#CDEF?DGF#?#D:5?.+&H(&310#<:;+*?'1*(,):1#9*5&//:3-#)**/,I#
'35#'/,*#6,2:/#J:35;&+-#.*+#9&',(+&9&3)#&K7&+):,&L#
!..($)/0-M)#.+&H(&31:&,#:3#'#3&:-%;*+%**5#*.#'3#:,*/')&5#+&,*3'31&N#)%&#+&1&7)'31&#:,#-:<&3#;0#
H (! ) =
C
##
!! " ! 2
2
r
O!@P#
4%&+&# C ##:,#'#1*3,)'3)N#5&.:3&5#;0#)%&#9*5'/#.*+1&#'35#)%&#9*5'/#9',,N#'35#
!! r2 = !r2 (1 + i#r ) ##:,#)%&#1*97/&K#,H('+&5#+&,*3'31&#.+&H(&310L#E.#)%&#3&:-%;*+%**5#

.+&H(&31:&,#'35#)%&#1*++&,7*35:3-#+&1&7)'31&,#'+&#1*//&1)&5#:3#)%&#<&1)*+,# ! #'35# H N#*.#
/&3-)%# L N#)%&#/&',)#,H('+&#&,):9')&#*.# !! r2 ##:,#-:<&3#;0#)%&#D')/';#1*99'35#
x = [ H , " ones( L,1)] \ ( H "*! "^2); #
4%&+&# x (1) = !! r2 ; x ( 2 ) = C #L###
1(2(1($"(31. J. D. Ferry Viscoelastic properties of polymers (John Wiley & Sons,1961, NewYork) Chap. 2,
pp. 21-40.
2. M.A. Biot Theory of propagation of elastic waves in fluid saturated solid. Part 1 low frequency
range J. Acoust. Soc. Am. 26, 168-178 (1956).
3. J. Allard, N. Atalla Propagation of sound in porous media, 2nd ed. (John Wiley & Sons, 2009,
New York) Chap 6 pp. 111-135 and Chap 10 pp. 213-241.
!"#
#
O!QP#
4. Remi Guastavino Elastic and acoustic characterisation of anisotropic porous media (Ph.D.
thesis, the MWL, KTH, 2008) pp. 1-19.
5. Luc Jaouen, Amelie Renault Mickael Deverge Elastic and damping characterization methods of
acoustic porous materials: Available experimental methods and application to melamine foams
(review article) Applied Acoustics 69, 1129-1140 (2008).
6. Luc Jaouen Characterisation of acoustic and elastic parameters of porous media, Keynote
lecture presented at the Symposium of the Acoustics of Poro-elastic materials (SAPEM, Ferrara,
2011).
7. A Renault, L Jaouen, F Sgard Characterisation of elastic parameters of acoustical porous
materials from beam bending vibrations, J. of Sound and Vibration 330, 1950-1963 (2011).
8. O. Doutres, N. Dauchez, J.-M. Genevaux and G. Lemarquand, On the use of a loudspeaker for
measuring the viscoelastic properties of sound absorbing materials J. Acoust. Soc. Am. 124,
Pages: EL335-EL340 (2008)
9. O. Danilov, F. Sgard, X. Olny On the limits of an in vacuum model to determine the
mechanical parameters of isotropic poroelastic materials J. of Sound and Vibration 276, 729-754
(2004).
10. H.J Rice and P. Gransson Adynamical model of light fibrous materials. Int. J. of Mechanical
Science 41, 561-579 (1999)
11. P. Gransson and R. Guastavino Identification of a general orthotropic, viscoelastic model of a
porous foam in Remi Guastavino Elastic and acoustic characterisation of anisotropic porous
media (Ph.D. thesis, the MWL, KTH, 2008) Appendix E.
12. J. Cuenca and P. Gransson Inverse estimation of the elastic and anelastic properties of the
porous fram of anisotropic open-cell foams. J. Acoust. Soc. Am. 132, 621-629 (2012)
13. A. Geslain, O. Dazel, J.-P. Groby and S. Sahraoui, Influence of staic compression on mechanical
parameters of acoustic foams J. Acoust. Soc. Am. 130, 818-825 (2011)
!"#
#
14. Martin Fraggstedt Vibrations, damping and power dissipation in car tyres (Ph.D. thesis, the
MWL, KTH, 2008) Appendix E. Accessed the 2 October 2012 at
http://www.kth.se/sci/institutioner/ave/avd/mwl/contact/associate-professor-svante-finnveden-dsc-tech-1.64947 )
15. S. Finnveden and M. Fraggstedt Waveguide Finite Elements for Curved Structures, J. of Sound
and Vibration, 312, 644-671 (2008).
16. N.-E. Hrlin, P. Gransson Weak, anisotropic symmetric formulation of Biots equations for
vibro-acoustic modelling of porous elastic material, International Journal for Numerical Methods
in Engineering 84, 1519-1540 (2010).
17. Y.C. Fung, P. Tong Classical and Computational Solid Mechanics (World Scientific
Publishing, Singapore, 2001) Chap 13 pp. 428-455 and pp. 487-491.
18. Mathias Barbagallo and Svante Finnveden. A self adjoint variational principle for anisotropic
viscoelastic Biots equation (Accepted for publication in the Int. Journal for Engineering
Science, 2012) Doi: 10.1016/j.ijengsci.2012.06.019
19. R.L. Bagley and P.J. Torvik, Fractional calculus a different approach to the analysis of
viscoelastically damped structures American Institute of Aeronautics and Astronautics Journal
21, 741-748 (1983).
20. T Pritz Five-parameter fractional derivative model for polymeric damping materials, Journal of
Sound and Vibration 265, 935-952 (2003).
21. M. E. Gurtin and E. Sternberg, On the linear theory of viscoelasticity, Archive for Rational
Mechanics and Analysis, 11, 291-356 (1962).
22. Encyclopaedia Britannica: http://www.britannica.com/EBchecked/topic/424706/Ockhams-razor.
Accessed 26 Sept 2012
#
!"#
#
!"#$%&'
#
#
$%&'(#)%*+#!,#!-)#
[ kPa ] !
[ "] #
# [ "] #
$ s !# kg/m 3 "$ #
!/0#
0120#
0134#
/513#
!/3#
0130#
0135#
/513#
$%&'(#)%*+#3#
547#
01!"#
0132#
5713#
89%#)%*+#
3"15#
01!7#
0134#
4717#
%&.#
$%&'(#)%*+#!,#3.6#
%&.#
#
$9:;<#!1#=9)<%*9;#69)9#>?%#>%<@&<.'A#*.6<B<.6<.)#*-?)%?B*'#+?6<;-1#
#
#
#
!
E0 [ kPa ]
#
G0 [ kPa ] # # f1 ##CDEF#
f 2 #C(DEF#
% ##C#G#F#
$%&'(#)%*+#!## !35#
5712#
!"0#
313/#
014H5#
$%&'(#)%*+#!,# !20#
5/14#
303#
!1H3#
01537#
$%&'(#)%*+#3## 43/#
!"4#
240#
31"!0#
0145H#
89%#)%*+##
!31H#
32"#
310H#
014H5##
2I#+?6<;#
3310#
#
$9:;<#3#J-?)%?B*'#>%9')*?.9;#6<%*K9)*K<#+?6<;#
!"#
#
!"#
#
!"#$%&'%'"()*+%,-.$"&/#%
$%&'()#*+#,(%-#./0)(#1234%45%3&#26#72(2'4#62/-#/38#/#9)/:0#./0)(#-/8)#26#(';;)(+#<)65=#.23&#
4/-7.)=#>"#--#?#!@#--#?#!#--=#'4)8#%3#(%&#*#23.0+#A%&95=#4-/..#4/-7.)=#!B#--#?#!@#--#?#!#
--#'4)8#%3#;259#(%&4+#,9)#9)%&95#!#:/(%)4#;)5C))3#4/-7.)4#;'5#%4#26#59)#2(8)(#26#!@#--+#
#
$%&+##!/#D)/4'()-)35#(%&=#6%(45#/55)-75#
#
$%&+##!;#D)/4'()-)35#(%&=#4'11)446'.#(%&#
#
$%&'()#E+#F/1''-#19/-;)(#/38#%345('-)35)8#.23&#4/-7.)#52#59)#(%&95#
#
$%&'()#G+#D)8%'-#-)49#'4)8#%3#-245#1/.1'./5%234+#H%(1.)4=#328)4I#.%3)4=#).)-)35#)8&)4+#,9)#
527#./0)(#-28).4#59)#45%66#/38#.%&95#12-724%5)=#59)#3)?5#./0)(#-28).4#59)#9)/:0#./0)(#/38#59)#
4%?#.2C)45#./0)(4#-28).#59)#62/-+#
#
$%&'()#@+#D)/4'()8#5(/346)(#6'315%234#62(#5('1J#4/-7.)#*+#K2.%8#.%3)4=#%3#:/1''-I#8/49)8#
.%3)4=#%3#/%(+#LMNMO=#92(%P235/.#)?1%5/5%23#Q#92(%P235/.#()47234)I#LFNFO=#:)(5%1/.#)?1%5/5%23#N#
:)(5%1/.#()47234)I#LMNFO=#92(%P235/.#)?1%5/5%23#Q#:)(5%1/.#()47234)+##
#
$%&'()#R+#,(/346)(#6'315%234#62(#5('1J#4/-7.)#*+#K2.%8#.%3)4=#-)/4'()8#%3#:/1''-I#8/49)8#
.%3)4=#1/.1'./5)8#C%59#%425(27%1#6()S')310#%38)7)38)35#-28).=#4))#,/;.)#*=#*45#('3+#<)&)384#
/4#%3#$%&+#@+#
#
!"#
#
$%&'()#*+,#-(+./0)(#0'.12%3./#03(#2('14#/+567)#",#837%9#7%.)/:#5)+/'()9#%.#;+1''5<#9+/=)9#
7%.)/:#1+71'7+2)9#>%2=#%/32(36%1#0()?').1@#%.9)6).9).2#539)7:#/))#-+A7)#":#!.9#('.,#B)&).9/#
+/#%.#$%&,#C,#
#
$%&'()#*A,#D=+/)#30#2(+./0)(#0'.12%3./#03(#2('14#/+567)#",#837%9#7%.)/:#5)+/'()9#%.#;+1''5<#
9+/=)9#7%.)/:#1+71'7+2)9#>%2=#9+2+#%.#-+A7)#":#!.9#('.,#B)&).9/#+/#%.#$%&,#C,#
#
$%&'()#E,#F+&.%2'9)#30#2(+./0)(#0'.12%3./#03(#2('14#/+567)#!,#837%9#7%.)/:#5)+/'()9#%.#
;+1''5<#9+/=)9#7%.)/:#1+71'7+2)9#>%2=#9+2+#%.#-+A7)#",#B)&).9/#+/#%.#$%&,#C,#
#
$%&'()#G,#F+&.%2'9)#30#2(+./0)(#0'.12%3./#03(#H+(#/+567),#837%9#7%.)/:#5)+/'()9#%.#;+1''5<#
9+/=)9#7%.)/:#1+71'7+2)9#>%2=#9+2+#%.#-+A7)#",#B)&).9/#+/#%.#$%&,#C,#
#
$%&'()#"I,#J/2%5+2)9#K3'.&L/#539'7'/#03(#-('14#/+567)#",#M)+7#6+(2#30#539'7'/:#'66)(#7%.)/<#
%5+&%.+(@#6+(2#:#73>)(#7%.)/,#837%9#7%.)/:#0(+12%3.+7#539)7:#9+2+#+/#%.#-+A7)#!<#9+/=)9#7%.)/:#
0()?').1@#%.9)6).9).2#539)7:#9+2+#+/#%.#-+A7)#":#!.9#('.,#
#
$%&'()#""+,#F+&.%2'9)#30#2(+./0)(#0'.12%3./#03(#-('14#/+567)#",#837%9#7%.)/:#5)+/'()9#%.#
;+1''5<#9+/=)9#7%.)/:#1+71'7+2)9#>%2=#%/32(36%1#0(+12%3.+7#539)7:#/))#-+A7)#!,#B)&).9/#+/#%.#
$%&,C,#
#
$%&'()#""A,#D=+/)#30#2(+./0)(#0'.12%3./#03(#2('14#2(%5#",#837%9#7%.)/:#5)+/'()9#%.#;+1''5<#
9+/=)9#7%.)/:#1+71'7+2)9#>%2=#%/32(36%1#0(+12%3.+7#539)7:#/))#-+A7)#!,#B)&).9/#+/#%.#$%&,C,#
#
!"#
#
#$%&'(")!*"+,%-$.&/("01".',-21('"1&-3.$0-2"10'".'&34".'$5"!*"607$/"7$-(28"5(,2&'(/"$-"9,3&&5:"
/,2;(/"7$-(28"3,73&7,.(/"<$.;"$20.'0=$3"1',3.$0-,7"50/(78"2((">,?7("!*"@(%(-/2",2"$-"#$%*A*"
"
#$%&'(")B*"+,%-$.&/("01".',-21('"1&-3.$0-2"10'"3,'"2,5=7(*"607$/"7$-(28"5(,2&'(/"$-"9,3&&5:"
/,2;(/"7$-(28"3,73&7,.(/"<$.;"$20.'0=$3"1',3.$0-,7"50/(78"2((">,?7("!*"@(%(-/2",2"$-"#$%*A*"
"
#$%&'(")C*"+,%-$.&/("01".',-21('"1&-3.$0-2"10'">'&34"2,5=7(")"3,73&7,.(/"<$.;"$20.'0=$3"
1',3.$0-,7"50/(78"2((">,?7("!*"607$/"7$-(28"BDE"50/(7:"/,2;(/"7$-(28"!DE"50/(7*""
"
"
!!"
"
##
#
$%&'()#*+#,(%-#./0)(#1234%45%3&#26#72(2'4#62/-#/38#/#9)/:0#./0)(#-/8)#26#(';;)(+#<)65=#.23&#
4/-7.)=#>?#--#@#!A#--#@#!#--=#'4)8#%3#(%&#*#23.0+#B%&95=#4-/..#4/-7.)=#!C#--#@#!A#--#@#!#
--#'4)8#%3#;259#(%&4+#,9)#9)%&95#!#:/(%)4#;)5D))3#4/-7.)4#;'5#%4#26#59)#2(8)(#26#!A#--+#
#
#
#
#
#
#
#
#
#
$%&+##!#E)/4'()-)35#(%&4+#<)65#6%(45#/55)-75F#B%&95=#4'11)446'.#(%&#
#
!"#
#
#
#
$%&'()#*+#,-.''/#.0-/1)(#-23#%245('/)25)3#672&#4-/86)#57#50)#(%&05+#
0.07
0.065
0.06
0.055
Y (m)
0.05
0.045
0.04
0.035
0.03
0.025
0.02
0.025
0.03
0.035
0.04
0.045
0.05
0.055
0.06
X (m)
#
$%&'()#"+#9)3%'/#/)40#'4)3#%2#/745#.-6.'6-5%724+#:%(.6)4;#273)4<#6%2)4;#)6)/)25#)3&)4+#=0)#
578#6->)(#/73)64#50)#45%??#-23#6%&05#.7/874%5);#50)#2)@5#6->)(#/73)64#50)#0)-A>#6->)(#-23#50)#
4%@#67B)45#6->)(4#/73)6#50)#?7-/+#
!"#
#
10
H!H
V!V
Transfer function (dB rel 1)
H!V
0
!5
!10
!15
1
10
10
10
Frequency (Hz)
#
#
$%&'()#"*#+),-'().#/(,0-1)(#1'02/%30-#13(#/('24#-,567)#8*#937%.#7%0)-:#%0#;,2''5<#.,-=).#
7%0)-:#%0#,%(*#>?@?A:#=3(%B30/,7#)C2%/,/%30#D#=3(%B30/,7#()-630-)<#>E@EA:#;)(/%2,7#)C2%/,/%30#@#
;)(/%2,7#()-630-)<#>?@EA:#=3(%B30/,7#)C2%/,/%30#D#;)(/%2,7#()-630-)*##
#
!"#
#
10
H!H
V!V
H!V
0
!5
!10
!15
1
10
10
10
Frequency (Hz)
#
#
$%&'()#"*#+(,-./)(#/'-01%2-.#/2(#1('03#.,456)#7*#826%9#6%-).:#4),.'()9#%-#;,0''4<#9,.=)9#
6%-).:#0,60'6,1)9#>%1=#%.21(25%0#/()?')-0@#%-9)5)-9)-1#429)6:#.))#+,A6)#7:#7.1#('-*#B)&)-9.#
,.#%-#$%&*#C*#
!"#
#
10
H!H
V!V
H!V
0
!5
!10
!15
1
10
10
10
Frequency (Hz)
#
#
$%&'()#"*+#,(*-./)(#/'-01%2-.#/2(#1('03#.*456)#7+#826%9#6%-).:#4)*.'()9#%-#;*0''4<#9*.=)9#
6%-).:#0*60'6*1)9#>%1=#%.21(25%0#/()?')-0@#%-9)5)-9)-1#429)6:#.))#,*A6)#7:#!-9#('-+#B)&)-9.#
*.#%-#$%&+#C+#
#
!"#
#
V!V
Angle(Transfer function) (radians)
H!H
!1
!2
!3
!4
H!V
!5
!6
!7
!8
!9
1
10
10
10
Frequency (Hz)
#
#
$%&'()#*+,#-./0)#12#3(/402)(#2'453%140#21(#3('56#0/789)#:,#;19%<#9%4)0=#7)/0'()<#%4#>/5''7?#
</0.)<#9%4)0=#5/95'9/3)<#@%3.#</3/#%4#A/+9)#:=#!4<#('4,#B)&)4<0#/0#%4#$%&,#C,#
#
!"#
#
10
V!V
H!H
5
H!V
0
!5
!10
!15
!20
1
10
10
10
Frequency (Hz)
#
#
$%&'()#*+#,-&.%/'0)#12#/(-.32)(#2'.4/%1.3#21(#/('45#3-678)#!+#918%0#8%.)3:#6)-3'()0#%.#
;-4''6<#0-3=)0#8%.)3:#4-84'8-/)0#>%/=#0-/-#%.#?-@8)#A+#B)&).03#-3#%.#$%&+#C+#
#
!"#
#
#
#
10
V!V
H!H
H!V
!5
!10
!15
!20
1
10
10
10
Frequency (Hz)
#
#
$%&'()#*+#,-&.%/'0)#12#/(-.32)(#2'.4/%1.3#21(#5-(#3-678)+#918%0#8%.)3:#6)-3'()0#%.#;-4''6<#
0-3=)0#8%.)3:#4-84'8-/)0#>%/=#0-/-#%.#?-@8)#A+#B)&).03#-3#%.#$%&+#C+#
#
#
!"#
#
2.5
x 10
Youngs modulus (Pa)
1.5
0.5
0
1
10
10
10
Frequency (Hz)
#
$%&'()#"*+#,-.%/0.)1#23'4&5-#/31'6'-#73(#8('9:#-0/;6)#"+#<)06#;0(.#37#/31'6'-=#';;)(#6%4)->#
?/0&%40(@#;0(.#=#63A)(#6%4)-+#B36%1#6%4)-=#7(09.%3406#/31)6=#10.0#0-#%4#80C6)#D>#10-E)1#6%4)-=#
7()F')49@#%41);)41)4.#/31)6=#10.0#0-#%4#80C6)#"=#D41#('4+#
#
#
#
!"#
#
10
H!H
V!V
H!V
0
!5
!10
!15
1
10
10
10
Frequency (Hz)
#
#
$%&'()#**+,#-+&.%/'0)#12#/(+.32)(#2'.4/%1.3#21(#5('46#3+789)#*,#:19%0#9%.)3;#7)+3'()0#%.#
<+4''7=#0+3>)0#9%.)3;#4+94'9+/)0#?%/>#%31/(18%4#2(+4/+9#710)9;#3))#5+@9)#",#A)&).03#+3#%.#
$%&,B,#
#
#
!"#
#
"
V!V
Angle(Transfer function) (radians)
H!H
!1
!2
!3
!4
H!V
!5
!6
!7
1
10
10
10
Frequency (Hz)
"
#$%&'("))*+",-./("01"2'.3/1('"1&342$03/"10'"2'&45"2'$6")+"708$9"8$3(/:"6(./&'(9"$3";.4&&6<"
9./-(9"8$3(/:"4.84&8.2(9"=$2-"$/02'0>$4"1'.42.8"609(8:"/(("?.*8("@+"A(%(39/"./"$3"#$%+B+"
"
"
!!"
"
10
V!V
H!H
5
H!V
0
!5
!10
!15
!20
1
10
10
10
Frequency (Hz)
#
#
$%&'()#*+,#-.&/%0'1)#23#0(./43)(#3'/50%2/4#32(#0('56#0(%7#+,#829%1#9%/)4:#7).4'()1#%/#;.5''7<#
1.4=)1#9%/)4:#5.95'9.0)1#>%0=#%420(2?%5#3(.50.9#721)9:#4))#@.A9)#+,#B)&)/14#.4#%/#$%&,C,#
,#
#
!"#
#
10
V!V
H!H
H!V
!5
!10
!15
!20
1
10
10
10
Frequency (Hz)
#
#
$%&'()#*!+#,-&.%/'0)#12#/(-.32)(#2'.4/%1.3#21(#4-(#3-567)+#817%0#7%.)39#5)-3'()0#%.#:-4''5;#
0-3<)0#7%.)39#4-74'7-/)0#=%/<#%31/(16%4#2(-4/-7#510)79#3))#>-?7)#@+#A)&).03#-3#%.#$%&+"+#
#
!"#
#
10
!5
!10
!15
1
10
10
10
Frequency (Hz)
#
$%&'()#*+,#-.&/%0'1)#23#0(./43)(#3'/50%2/4#32(#6('57#4.89:)#*#5.:5':.0)1#;%0<#%420(29%5#
3(.50.:#821):=#4))#6.>:)#?,#@2:%1#:%/)4=#!AB#821):C#1.4<)1#:%/)4=#?AB#821):,##
#
!"#
#
Paper 5
!
Characterisation of a generic trim panel: sound reduction index and

material parameters technical report
Mathias Barbagallo, Svante Finnveden
The Marcus Wallenberg Laboratory for Sound and Vibration Research (MWL), Royal Institute of Technology (KTH), 10044
Stockholm, Sweden
Abstract
Measurements and data to characterise a trim-panel are presented. The trim panel consists of a aluminium
layer and a rubber layer with a porous layer in between them. First, the measurement of the sound reduction
index is described. Then, the parameters to describe the porous layer via Biots theory or equivalent fluid
theory are provided. Finally, the viscoelastic properties of the rubber layer are measure.
Keywords: multilayered structure, porous material, rubber, measurement
1. Introduction
Laboratory measurements of structures have a twofold importance for numerical models describing their
vibroacoustic response. First, measurements provide the necessary inputs to models describing the structural
and geometrical properties of the structure under-study. Second, measurements are used to check the
accuracy of a model.
This report presents measurements on two generic trim panels, which are multilayered structures composed of a common aluminium shell with single curvature, to which two different composites are attached;
the composites consist of a rubber layer attached to a porous material. These panels may be seen as first
approximations of a car-floor and a truck-floor ; Figure 1 shows the composite of the truck-floor attached to
the aluminium panel.
In Section 2, the sound reduction index measurement is presented. In Section 3, the parameters that are
needed to model the porous material with Biot or equivalent fluid theory are given and the flow resistivity
measurement is described. In Section 4, the measurement to estimate the viscoelastic properties of the
rubber plate is presented.
All measurements have been taken at The Marcus Wallenberg Laboratory for Sound and Vibration
Research (MWL) at the Royal Institute of Technology (KTH) in Stockholm.
2. Sound reduction index measurement
The sound reduction index (SRI) measurement follows the ISO 15186 - 1:2000, Measurement of sound
insulation in buildings and of building elements using sound intensity.
Figure 2 illustrates the set-up of the measurement. The specimens are first mounted in a wooden frame
and then installed in a rigid wall separating the anechoic room and the reverberation room in MWL. The
aluminium panel faces the former room and the rubber mat the latter, see Figures 2-4.
The specimens are mounted in the wooden frame via pins nailed to the frame and clamping the aluminium
panel, Figure 5; damping material is used between the aluminium panel and the pins, to avoid contact and
Corresponding author, Tel. +46 8 790 91 39.
Email addresses: matbar@kth.se (Mathias Barbagallo), svantef@kth.se (Svante Finnveden)
Figure 1: From left to right, the aluminium panel, truck-floor porous material and the rubber mat. The aluminium panel and
the composites in both floors have the same length and height of 900 mm x 920 mm. The panel has a radius of curvature of 3
m and a thickness of 1.5 mm.
rattling noise; the gap between the wooden frame and the aluminium panel is sealed using tape, Figure 5;
the gap between the wooden frame and the rubber mat is sealed using mastic, Figure 6. The positions of
the metal pins is shown in Figure 7. The specimen mounted in the wooden frame is inserted in the rigid
wall; tape and mineral wool are used to seal the gap between the wooden frame and the rigid wall to avoid
unwanted sound leakage.
In the anechoic room an intensity probe operated by a person measures the sound intensity level (probe
model 2260 used with a Larson Davis Real time analyzer 2900); in the reverberation room a microphone
on a rotating stand measures the space-averaged sound pressure level (half-inch microphone, together with
a PC running SpectraPLUS); the sound pressure is generated by four loudspeakers driven by white noise
generated by a B&K Noise generator type 1405 and amplified by a NAD 310. Both measurements are
done in third-octave bands. The cut-off frequency of the anechoic room is 80 Hz and its dimensions are
7.00 m
5.95 m
5.80 m; the dimensions of the reverberation room are 6.21 m
7.86 m
5.05 m.
The intensity probe allows measurements up to 5 kHz, this limit being set by the distance of the two
microphones. The quality factor of the intensity measurement, i.e. the difference between intensity and
pressure levels, is acceptable till 5 kHz for the truck-floor and 2.5 kHz for the car-floor; at higher frequencies
the assumption of no reflections in the anechoic room may be violated due to the measurement procedure,
e.g., by the reflections on the person measuring intensity (Figure 4 shows the disposition of the equipment
used to measure intensity). The field in the reverberant room is diffuse between 50 Hz and 10 kHz. The SRI
measurements are considered accurate between 80 Hz and 2.5 kHz for the car-floor and 80 Hz and 5 kHz
for the truck-floor. In these frequency ranges, the repeatability of the measurement is reliable: the same
result is obtained if the plate is mounted, dismounted and remounted, and also if the scanning of intensity
is performed in different fashions.
The space-averaged sound pressure level in third-octave bands Lp is acquired via SpectraPLUS installed
on a PC; the space-averaged sound intensity level in third-octave bands LIn is measured via a Larson-Davis
real-time analyzer. The sound reduction index SRI is calculated using the formula provided in the standard:
SRI
Lp
LIn .
(1)
The SRIs of the two floors are reported in Table 1. The measured SRIs have similar trends to those of
typical building-construction double-walls presented in Ref. [2].
3. The porous layer
The parameters needed to model the porous layer both with Biot theory and with equivalent fluid theory
are given. The former theory includes the strain energy of the frame of the porous material: its in-vacuo
2
Figure 2: Sketch of the measurement set-up.
Figure 3: View of the multilayered structure from the reverberation room, where a rotating microphone measures the spatialaveraged rms pressure.
Figure 4: View of the multilayered structure from the anechoic room: the the time-averaged intensity is measured by a person
using an intensity probe.
Figure 5: Pins clamping the aluminium panel; the aluminium panel is in the lower half, the wooden frame is in between the
two dashed lines and the top half is the rigid wall. The tape seals the gap between the wooden frame and the rigid wall.
Figure 6: White mastic sealing the gap between the rubber mat and the wooden frame.
Figure 7: Location of the metal pins on the aluminium panel (valid for both panels). Distance a: 40.6 cm; Distance b: 9.6 cm;
Distance c: 31.3 cm; Distance d: 39 cm.
viscoelastic properties are estimated in Ref. [3] and the reader can refer to that for the description of the
measurement and of the inverse estimation procedure, and for the estimated parameters (also Ref. [1]). In the
following subsection, the flow resistivity measurement is described. Then, the values for material thickness,
density, tortuosity, viscous and thermal lengths, porosity are provided. Before taking measurements on
a sample of porous material, it may be advised to gently squeeze the sample to break any closed pore
containing air bubbles, so that Biot theory may be applicable.
3.1. Flow resistivity
The flow resistivity experiment is based upon the assumption that DArcy law governs the static fluid
flow in the porous material; besides, in these experiments the materials are considered isotropic (at KTH
it was developed a technique to measure the flow resistivity third-order tensor, but the dimensions of
the provided specimens prevents three-dimensional measurements [5]). The measurement follows the ISO
9053:991: acoustics - material for acoustical application - determination of aiflow resistance, 1991. The set
up is shown in Figure 8; for further details, see reference [5]. The following instruments are used:
the pressure sensor SwemaAir 300 measures the pressure drop in the laminar flow element, and the
flow speed is estimated (Art. No. 757501, Serial: 389739);
the flow passes through the laminar element Meriam 50MJ10-12;
the differential pressure transmitter Furness controls model FCO332-3W measures the pressure drop
across the sample (Serial: 0405132-8);
a caliper, precise till the second decimal digit, measures the length of the samples.
Cylindrical samples are cut from the specimens, Figure 10. The porous materials have to be detached
from the rubber plate. In addition, both porous materials have a thin plastic film closing the cells on the
surface: this film is removed by hand with a cutter. These two modifications lead to uneven surfaces, which
may cause measurement errors. Several samples from these specimens are measured.
All measurements are performed at a laminar flow velocity of 5 mm s (corresponding to 76.5P a measured
by the SwemaAir), which is verified to yield a linear flow in the laminar flow element. Finally, the formula
to calculate the flow-resistivity is given by:
pL
(2)
v
where p is the pressure across the sample (measured by Furness), L is the length of the sample and v is the
velocity across the laminar flow element (measured by the SwemaAir).
Table 2 shows the mean values of the flow resistivity of the two floors, based on five samples per specimen;
the standard deviations of the truck-floor is 4.16 P a s m2 and that of the car-floor is 921 P a s m2 . These
variations between samples of the same specimen may be due to inhomogeneities in the material itself: in
fact, inside the porous materials there are large pores with a diameter ranging from 1 till 8 mm, in random
position, of various shapes, which surely affect the result of the measurement.
3.2. Remaining parameters
Table 2 also presents more parameters of the porous layers. The densities are measured with a ruler and
a scale, respectively precise till the second and the first decimal digits. The other parameters are estimated
via educated guesses.
4. Rubber plate
The shear rig measurement provides the dynamic shear modulus and the damping loss factor of the
heavy layer; Figure 11 shows a photo of the set-up and Figure 12 shows a sketch of it [4]. The measurement
proceeds as follows, quoting Fraggstedt [4]: A shaker is mounted to a small magnesium piece through the
means of a stinger. The magnesium piece is attached to two small rubber samples which in turn are mounted
5
Table 1: Sound reduction index measurement in third-octave bands
f [Hz]
80
100
125
160
200
250
315
400
500
630
800
1000
1250
1600
2000
2500
3150
4000
5000
car-floor [dB]
4.3
14.4
14.8
14.2
15.4
12.9
12.8
19.7
28.5
34.1
38.3
42.9
40.7
40.4
38.7
38.2
-
truck-floor [dB]
16.3
20.6
21.7
20.8
22.1
21.5
22.4
22.2
19.3
23.1
32.8
38.8
42.2
43.0
44.9
47.1
51.6
53.3
54.3
Figure 8: Flow resistivity measurement sketch.
Table 2: Specific porous materials data. indicates direct measurements and its absence indicates educated guesses.
Symbol
Quantity
Static flow resistivity [P a s m2 ]
Porous material density [kg m3 ]
Porosity [ ]
Tortuosity [ ]
Viscous length [m]
Thermal length [m]
Thickness [m]
6
truck-floor
5.52 104
57.2
0.95
1.05
37 10 6
110 10 6
2.24 10 2
car-floor
1.87 104
47.7
0.957
1.2
49 10 6
100 10 6
2.12 10 2
Figure 9: Measurement set-up: material sample (a), laminar flow element (b), SwemaAir (c), Furness (d), flow income (e).
Figure 10: Various samples. The car-floor porous layer is the second from left and the truck-floor layer is the fourth.
Table 3: General porous material data
Symbol
Pr
s
P0
0
Parameter
Dynamic viscosity [kg m s]
The Prandtl number
Ratio of specific heats
Air pressure at reference temperature [P a]
Density of fluid in pores [kg m3 ]
Value
18.4 10 6
0.71
1.4
1.01 105
1.205
Table 4: Porous material parameters
Symbol
q0
q0
Parameter
Kinetimatic viscosity
Inverse thermal diffusivity
Viscous static permeability
Thermal static permeability
Formula
0
Pr

2 8
Table 5: Porous material elastic frame data
E0 [kP a]
G0 [kP a]
f1 [Hz]
f2 [kHz]
[-]
[-]
truck-floor
414
717
209
4.67
0.411
0.306
car-floor
22
47
220
4.51
0.410
0.402
Figure 11: Photo of the shear rig.
Figure 12: Sketch of the shear rig.
to an aluminium fork-like arrangement which is rigid for the consider load. When the force is applied, the
rubber is sheared and a vertical force is transferred to the aluminium fork. A force transducer is mounted
between the fork and a big concrete block. To measure the displacement of the magnesium piece a small
accelerometer is used. To keep the rubber in place it is squeezed between the fork blocks where sandpaper
is used to prevent it from slipping. The arrangement can be viewed can be viewed as a simple mass-spring
system where the modal mass is given by the moving mass of the shaker, the pushrod, the accelerometer
(and part of the electrical cord), the magnesium piece, the sandpaper and approximately half of the sheared
rubber. The spring is given by the geometric and stiffness properties of the rubber sample. The following
instruments are employed:
Mini Shaker exciting the magnesium piece: B&K Type 4810 (SN 647049);
Accelerometer: B&K Type 4375 (SN 152577);
Force transducer: Kistler Type 9067 (SN 349818);
Siglab acquisition station model 20-42;
Charge amplifier: B&K Type 2635 (SN 1571336, 638484);
A caliper, precise till the 2nd decimal digit, measures the dimensions of the samples.
Via Siglab, the transfer function H between the force transducer and the accelerometer is measured. Knowing
H, the measured shear modulus G is obtained as
2 t H
,
(3)
2A
where t and A are the thickness and the area of the rubber sample, respectively.
The estimation of the material data is based on the assumption that the fractional Kelvin-Voigt model
may, for a harmonic time dependence of the form eit , describe the dynamic shear modulus of the heavy
layer:
i

G
G0 1
,
(4)
0
where G0 is the static shear modulus, and 0 and describe the frequency dependence. These parameters
are fitted in a least square sense to the measured data using the MATLAB routine fminsearch. The object
function that is minimised is
G
#
f
1
f
Re G
w Im G
(5)
where the function f1 weights different octaves in a similar fashion and w balances difference between the
real and imaginary parts (w = 3 for the two floors). Table 6 shows the parameters for the Kelvin-Voigt
model extracted from the procedure for the two floors. Figures 13 and 14 show the extracted shear modulus
and loss factor of the truck floor. The lower frequency limit of the measurements is 50 Hz, due to a rig
resonance; the upper limit is 400 Hz, due to the appearance of the resonances of the rig; the extrapolation of
the curve is nonetheless possible till 700 Hz, since after the dip at 400 Hz the curve visibly continues with the
same trend. Note, fractional derivative models are a bit volatile if the frequency region is too small: one can
exchange a large for a small 0 or vice versa. Some caution is therefore called upon when extrapolating
the data outside the frequency range of the measurements.
provides the shear modulus, while the ratio of its imaginary and real parts the damping
The real part of G
loss factors.
Acknowledgements
Christophe Van der Kelen is acknowledged for the help provided in the flow resistivity measurement.
Peter Goransson is acknowledged for the educated guesses of the porous layers parameters in Table 2.
9
Table 6: Rubber mats data
G0 [P a]
0 [rad s]
[-]
Density [kg m3 ]
Damping loss factor [ ]
Thickness [m]
truck-floor
3.3598 106
1.8768 103
0.2135
1821.5
0.12
3.3 10 3
car-floor
1.1545 106
0.75971 103
0.3418
2144
0.22
2.0 10 3
12
x 10
10
Shear modulus [Pa]
2
2
10
10
[Hz]
Figure 13: Shear modulus of the truck-floor, measured (solid line) and estimated (dashed line). Real part top lines, imaginary
part bottom lines.
0.3
Loss factor []
0.2
0.1
0.1
0.2
10
10
[Hz]
Figure 14: Loss factor of the truck-floor, measured (solid line) and estimated (dashed line).
10
References
[1] M. Barbagallo. Statistical energy analysis and variational principles for multilayered panels. PhD thesis, SCI-AVE,
Teknikringen 8 10044 Stockholm, March 2013.
[2] S. Finnveden and M. Barbagallo. A cavity-wall element for statistical energy analysis of double walls. Technical Report
TRITA-AVE 2012:02 ISSN 1651-7660, AVE/MWL, 2012.
[3] S. Finnveden, N.-E. H
orlin, and M. Barbagallo. Characterization of the in vacuo viscoelastic material of porous foams used
in vehicles. Journal of the Acoustical Society of America, 2012.
[4] M. Fraggstedt. Vibrations, damping and power dissipation in car tyres. PhD thesis, AVE-SCI, Teknikringen 8 10044
Stockholm, 2008.
[5] P. G
oransson, R. Gustavino, and N.-E. H
orlin. Measurement and inverse estimation of 3d anisotropic flow resistivity for
porous materials. Journal of Sound and Vibration, 327:354367, 2009.
11
Paper 6
!
Spatial energy decay and indirect couplings in statistical

energy analysis
M. Barbagallo 1 , S. Finnveden 2
The Marcus Wallenberg Laboratory for Sound and Vibration Research (MWL), KTH,
100 44 Stockholm, Sweden
1
2
e-mail: matbar@kth.se
e-mail: svantef@kth.se
Abstract
Spatial energy decay within elements affects the validity of SEA. This is particularly significant for chains
of similar long well-connected structures such as ventilation ducts, fluid-filled pipes and rib-stiffened plates
found in ships, aircraft and railway cars. The effects of spatial energy decay on the high frequency response
of one-dimensional well-connected elements are herein studied by comparing calculations by an SEA, a
spectral finite element method and an SEA-like model. An SEA only includes direct coupling loss factors
(CLFs); conversely, an SEA-like model also contains indirect CLFs. At high frequencies, the spatial energy
decay increases and SEA overestimates the energies in all elements away from the excitation. Moreover,
the indirect CLFs in the SEA-like model have to be considered when evaluating the energy flows, as the
accumulated spatial decay from the excitation to the observed point increases. Thus, SEA cannot predict the
high frequency response of similar long well-connected elements and alternative formulations are needed.
Introduction
Statistical energy analysis (SEA) is an approximate method to estimate steady state ensemble, space and
frequency band averages of vibroacoustic energy in built-up structures at somewhat higher frequencies [1, 2].
Today, there seems to be an accord, or at least a hope, in vehicle industry and other parties interested in
vibroacoustic predictions that SEA works, if only the frequency is high enough - as is evidenced by the
public homepages of the ongoing EC-funded projects MID-MOD [3] and MID-FREQUENCY [4].
The governing equations in an SEA express the law of energy conservation for elements, each describing, in
a frequency band, one kind of response in a substructure of the whole structure. The formulation relies on
two postulates. First, the rate of energy dissipated in an element is assumed proportional to the energy in that
element. Second, the rate of energy flow between two directly coupled elements is assumed proportional
to the difference of the elements modal power potentials, being the energies times the expected frequency
spacing between their natural frequencies [1]. This is the law of Coupling Power Proportionality (CPP). It is
a very powerful potential flow model for the conduction of vibroacoustic energy; similar to Fouriers law for
heat conduction and Ohms law for electrical conduction.
The constant of proportionality in the CPP law is often calculated by a one way method, where the action of
one element onto the other elements is investigated. In doing so, field characteristics of the first element are
defined, such as the amplitudes of vibroacoustic waves impinging on a junction; the amplitude in a diffuse
field or the amplitudes of modes proper to the uncoupled element. Often the calculations are simplified by
considering elements that are extended away from the junctions towards infinity. The conductivity evaluated
by a one way method describes a potential flow and can be used in a proper SEA model only if it is a sym-
metric function of the elements properties in such a way that it is equally evaluated when field characteristics
in the first and in the second element are prescribed. This symmetry relation is termed SEA reciprocity [1]
or consistency [5] relation. It is demonstrated for numerous one way calculations, e.g. [refs]. Similar to one
way methods in nature, but not in mathematical formulation, are perturbation methods [6, 7].
The first postulate is an established engineering approximation; however, the second postulate, the CPP law,
is useful only sometimes. To date it has been exactly demonstrated for the rate of energy flow between two
conservatively coupled, randomly excited, oscillators [9] and, also, for the ensemble averaged energy flow
between two one-dimensional structures, with random properties, in the limit of vanishing spatial damping
decay within the elements, [10], see also [11]. In contrast, for three oscillators [13] and for the ensemble
averaged of three one-dimensional structures [10], the CPP law is true for weak coupling only. For general
structures, the range of usefulness of the CPP law is not established, though, slowly appears a common understanding of the criteria for a valuable SEA model. One example is provided in a recent article by Le Bot
and Cotoni [14], stating that, for each element, each junction and each frequency band, SEA is valid if: i)
The number of natural frequencies is large; ii) The modal overlap factor is large (it is the ratio of resonances
3-dB bandwidth to their expected frequency spacing); iii) The strength of coupling is small; iv) The spatial
decay of energy in a wave travelling across an element is small. The first three criteria are for many engineering structures better fulfilled as frequency increases while the opposite is true for the fourth criterion.
The present article is concerned with this high frequency limitation of SEA.
The fundamental equation in SEA expresses the law of energy conservation. This law can also be expressed
based on calculated or measured response using the power injection method (PIM). Thus, for a given excitation, the input power is evaluated and the elements energies are linearly related to this input power. The
procedure is repeated for excitation in all elements and an energy distribution (ED) model is devised [15].
Then, when the ED model is inverted, an SEA-like model results. Like SEA models, an SEA-like model
relates the energy stored in the elements to the input powers. Though, as stressed by Fredo in reference [16],
the term SEA-like signifies that the energy balance is applied for the individual and deterministic case, not for
the ensemble averaged energy balance for random structures, as SEA attempts. Nevertheless, the dependant
variable in an SEA-like and a proper SEA model are equal when the frequency- spatial- ensemble-averaged
responses of the single specimen and of the ensemble are similar, which is most likely when the modal overlap factors are large, see e.g. [17].
An SEA-like model is, with respect to the precision of the measurements or calculations upon which it is
based, an exact statement of energy conservation. Comparing the equations defining such a model with
those that would apply for a proper SEA model, two major differences are apparent. First, the consistency
relation is not fulfilled, though there are evidences for that it approximately applies for higher frequencies,
when there are many resonances in the frequency bands and the elements modal overlap factors are large.
Specifically, the consistency relation is fulfilled for a homogeneous structure if the elements frequency and
space averaged point mobilities equal those for the corresponding infinite elements [8] (and later, equally,
[5]). The second difference is that the rate of energy flow to an element is not governed by the CPP law.
Thus, in this mathematical clothing, there appears energy flows between elements that are not physically
connected; this artefact is termed indirect coupling [5, 20], or tunnelling [18, 19, 25]. In the literature, there
are different names for models that include indirect couplings, e.g., advanced SEA [18], extended SEA [19]
or quasi-SEA [5]. In some cases such models might be useful. They cannot, however, be used in general
predictive SEA software, as it is not practical to calculate all the necessary coupling routines that consider
all the possible arrangements of connections in large structures.
The presence of indirect couplings does not, of course, describe an energy flow between two elements that
are not physically connected but signifies that the energy flow between two elements depends on the response
in other elements and, consequently, that CPP does not properly describe the flow of energy in the system.
In the literature tunnelling is demonstrated to happen in the regime of strong coupling and small modal overlap factors [10, 19]. In particular, if a structures response is defined by one global mode indirect couplings
cannot be neglected [5].
In reference [20] Mace investigates the importance of indirect couplings; he notes that for increasing modal
overlap factors, the indirect coupling loss factors asymptote to zero and concludes that a proper SEA model
can be formulated. The present study confirms that with increasing modal overlap factors the indirect coupling loss factors decreases but so does also the direct coupling loss factors and the net effect is, in fact, an
increasing tunnelling. Consequently, proper SEA models are not useful, for the investigated structures, at
higher frequencies. Previously, Fredo found increasing tunnelling in a three-plate structure with increasing
frequency ([19], paper IV) and Finnveden found a consistent overestimation of the energy flow in a threepipe structure at very high frequencies [21]; these observations are the basis for the present study on the high
frequency limitation of standard SEA.
In what follows the vibrations of long structures with many elements in series are described by SEA-like and
proper SEA models. The vibrations considered in the deterministic SEA-like model are excited by a large
number of uncorrelated point forces, thus simulating rain-on-the-roof excitation. Most of the calculations
are made in the high modal overlap regime where the frequency averaged response of an ensemble and an
individual structure should be similar. This surmise is supported by a few Monte Carlo runs considering
elements with random properties. The investigate structures are very long compared to the wavelength and
are well connected, meaning that a substantial part of the energy coming into a junction is transmitted.
The applications in mind are structures such as ventilation ducts, pipelines and rib-stiffened shell structures
such as found in trains, aircraft and ships. Besides of being long and well connected these systems often
have piece wise constant properties along the structure and support a numerable set of waves propagating in
this direction. Therefore, each section is as a set of parallel one-dimensional structures, being independent
channels of energy flow. Not to obscure the general picture and to facilitate extensive averaging based
on Monte Carlo simulations, simple one-dimensional structures consisting of rods in series are considered.
Thus, it is possible to highlight the findings on the spatial energy decays effect on tunnelling and the resulting
failure of predictive standard SEA at high frequencies.
The article is structured as follows. Section 2 presents the theoretical background, defining proper SEA and
SEA-like models, the energy decay factor and the energy flow between two elements. Section 3 presents
the class of structure that is investigated and explains the numerical methods used in the analysis: a spectral
finite element (SFE) method calculates the energy inputs and responses for the PIM, from which an SEA-like
model is devised. Section 4 compares the deterministic predictions of the SFEM and SEA-like models to the
statistical ones of proper SEA. Finally, Section 5 draws the conclusions.
Theory
An energy flow model is a statement of energy conservation: it relates the energy in the substructures of a
structure to the input power; that is to say, an energy flow model describes the distribution of input power in
a structure. In matrix form, an energy flow model can be represented as:
LE = Pin ,
(1)
where is the angular frequency, E is the total energies vector, Pin is the input powers vector and the
matrix L contains the damping loss factor and the coupling loss factor (CLF), respectively accounting for
the internal losses and the coupling losses within the structure.
Statistical energy analysis is an energy flow model for the ensemble with further assumptions, such as the
coupling power proportionality (CPP) and the consistency relation; besides, SEA directly provides averaged
results, whereas an energy flow model does not, in general, being applied to the individual deterministic
case [19]. On the other hand, an energy distribution (ED) directly expresses the subsystem energies as linear
functions of the input power; it can be devised via the power injection method (PIM). Eventually, an ED
model can be inverted, thus yielding an energy flow model.
This Section further defines these models, being central to this study.
2.1
SEA models
Statistical energy analysis is a potential energy flow model, based upon the subdivision of a system in l
elements; an element may be either a physical substructure or a component of the vibroacoustic response of
such structure, such as the rotational wave in a beam. In matrix form, SEA can be represented as:
B
e = Pin ,
(2)
where the (l, l) matrix B accounts fo the losses due to couplings and internal damping; the (l, 1) vector
e
contains the modal energies (or modal power potential [W]), i.e. ei = Ei /ni , where Ei is the total energy
of element i and ni is its asymptotic modal density. The response of an element is described by its modal
energy, which is, therefore, the unknown in an SEA: consequently, the energy density has to remain constant
within an element. Besides, SEA provides frequency- space- and ensemble-averaged results, this averaging
process being intrinsic in its rationale. The coefficients of the matrix B are given by:
Bii = Mi +
Cij ; Bij = Cij , i "= j,
(3)
Cij = ij ni ,
(4)
j!=i
where Mi is the modal overlap factor of element i, and Cij is the conductivity between elements i and j;
the modal overlap factor describes the dissipative losses in an element (see Subsection 2.3); the conductivity
Cij , on the other hand, describes the coupling between elements i and j, and it is defined as:
where ij is the coupling loss factor (CLF) between elements i and j. In an SEA model, modal densities and
CLFs are related by the SEA reciprocity [1] or the consistency relation [5],
ni ij = nj ji ,
(5)
which yields a symmetric matrix B. The power between two elements, or the coupling power Pc , is based
upon the coupling power proportionality (CPP), stating that the energy between two elements flows from
that with an higher modal energy to that with a lower one; this, in fact, makes SEA a potential energy flow
model. The energy flow between two elements is thus proportional to the difference in their modal energies,
Pcij = Cij (
ei ej ) .
(6)
Hence, the coupling between elements i and j depends only on these two elements. The CPP is exactly
fulfilled by two randomly excited oscillators, regardless of the coupling strength [9]; on the contrary, this
result cannot be extend to three-element systems unless they are weakly coupled [13, 12]. For more complex
structures, the validity of CPP depends on the assumption of weak coupling [12, 11, 14]. For the purpose
of this research, it is nonetheless enough to define as proper SEA an SEA model fulfilling the consistency
relation and the CPP; such proper SEA model, therefore, only features direct CLFs, whilst ignoring indirect
CLFs, i.e., couplings between not physically connected elements [5]. On the contrary, SEA models also
accounting for indirect CLFs while retaining the consistency relation, such as quasi-SEA [5], advanced SEA
[18] or extended SEA [19], are not herein investigated. The next Subsection explains the role of indirect
CFLs.
2.2
SEA-like models
In an energy distribution (ED) model, energies and input powers in the substructures of a whole structure are
related by the energy influence coefficient matrix A,
APin = E.
(7)
An ED model only requires the assumptions of linearity and uncorrelated excitations [5]; it can be devised,
numerically or experimentally, via the PIM. The PIM consists of injecting a known amount of power in
each substructure and, then, of measuring the responses of all of them. Eventually, the energy influence
coefficients can be retrieved solving the linear system in Equation (7).
Energy distribution models and energy flow models represent two views on the same issue: the relation
between energies and input powers. Henceforth, it is assumed that if the matrix L in Equation (1) is calculated
by inverting the energy influence matrix A, an SEA-like model is defined. In this article the term SEA-like
is, indeed, used to address a deterministic energy flow model devised by a numerical PIM; the term SEA-like
stresses its similarity to proper SEA: the dependant variables in an SEA-like and a proper SEA model are,
in fact, equal when the averaged responses of the single specimen and those of the ensemble are similar; an
SEA-like model, however, features indirect CLFs and the CPP and the consistency relation are unsatisfied.
Indirect CLFs, of course, do not refer to actual energy flows between not directly coupled subsystems; they
rather evidence that the energy flow between two subsystems also depends on other subsystems; they can
also be interpreted as mathematical artifacts appearing as the statement of energy conservation is forced into
an SEA-like clothing.
In the analysis presented in Section 3, there is no distinction between a substructure in an SEA-like model
and an element in a proper SEA model (see Subsection 3.3.3); herein, these two terms are exchangeable.
2.3
The energy decay factor
The modal overlap factor M is a non-dimensional number quantifying the dissipated power Pd in a system
as
Pd = M e,
(8)
depending on the damping loss factor , on the modal density n and the angular frequency ,
M = n.
(9)
The modal density describes the expected number of natural frequencies in a frequency band; the asymptotic
modal density of a one-dimensional system is [24]
n1D =
L
,
cg
(10)
where L is the length of the system and cg is the group velocity of the waves in this system; hence, the modal
overlap factor can be expressed as:
L
M = n =
.
(11)
cg
It is commonly understood that proper SEA provides useful results at higher frequencies where the modal
overlap factor is larger than unity; on the other hand, it is also understood that the energy decay should
be limited [14, 19, 23, 25, 26]. The magnitude of damping, travelled distance [25] and higher frequencies
combine to augment energy decay. Accordingly, after a distance x = L the energy density e(x) has decayed
as
e(x = L) = e0 exp (L/cg ) = e0 exp (M ) = e0 exp (D),
(12)
where e0 = e(x = 0). The argument of the exponential in Equation (12),
D = M =
L
,
cg
(13)
is named energy decay factor: in one-dimensional systems, a large modal overlap factor yields a large energy
decay factor. Via this quantity, the effects of spatial energy decay on the proper SEA of one dimensional
systems are investigated in Section 4.
In two- and three-dimensional systems the scenario is slightly different; for homogeneous and isotropic
structures, the two- and three-dimensional asymptotic modal densities are, respectively [24]:
S
2V
;
n
=
,
3D
4c2g2
2c30
n2D =
(14)
where S and V are the surface and the volume of the systems, cg2 and c0 are the group velocities for bending
waves in a two dimensional structure and for air respectively. Hence, for two- and three-dimensional systems
a large modal overlap factor does not imply a large energy decay factor.
2.4
Power and energy flow
The time averaged power is defined as:
W = !{F v },
(15)
W j = W ij W jk .
(16)
where ! extracts the real part of F times v , which respectively are the complex time rms force and the
complex time rms conjugate velocity; if they represent the values at the interface between two substructures,
Equation (15) expresses the flow of energy between them.
Proper SEA and SEA-like models, on the other hand, do not employ Equation (15) to calculate the energy
flow. Proper SEA provides an approximate version of the energy flow, the coupling power in Equation (6),
which is useful, as said, in the regime of weak coupling [27, 11, 10] or, for more complex structures, when Le
Bot and Cotonis criteria are fulfilled [14]. Calculating the energy flow via an SEA-like model is, however,
very cumbersome; in fact, the matrix L in Equation (1) splits the energy flowing into a substructure, into
a number of contributions dependant either on direct CLFs or on indirect CLFs. As a matter of fact, for
complex systems it might be very difficult, or even impossible, to trace the direction of all direct and indirect
transmission paths.
On the contrary, the net energy flow is usually feasible to determine. Assuming that substructure j is directly
connected to substructures i and k only and that energy flows from i to j to k, the net energy flow to
substructure j based on Equation (15) is:
The net energy flow calculated via the SEA-like model includes all direct and indirect couplings (1 and N 1
respectively) to substructure j; it is expressed as:
j =
N
!
i=1
(ij Ei ji Ej ) .
(17)
The net energy flow to element j calculated via proper SEA and based on Equation (6) is:
P j = Pcij Pcjk .
(18)
Methods
First, the class of structure under study is presented; preliminarily, a single specimen is discussed, then an
ensemble is defined. Second, the averages taken on such a class of structures are explained. Eventually, the
numerical methods employed in the analysis are detailed.
Table 1: Geometrical data of the rods.

Rod
1
2
3
4
5
6
7
8
9
10
Length [m]
0.1622
3.0742
0.3050
2.2073
0.2770
1.5936
0.2094
2.9762
0.2756
3.0544
Height [m]
0.0363
0.6882
0.0683
0.4942
0.0620
0.3568
0.0469
0.6663
0.0617
0.6838
Breadth [m]
0.0562
0.0314
0.0640
0.0674
0.0571
0.0603
0.0597
0.0457
0.0562
0.0368
Table 2: Structural data of the rods.

Youngs modulus [GPa]
Poissons ratio [-]
Density [kg/m3 ]
Damping loss factor [-]
3.1
210
0.3
7850
0.05
Class of structures
The class of structures under study is a chain of long one-dimensional systems supporting longitudinal vibrational only (i.e., rods), Figure 3.1. From a wider perspective, this might be seen as part of a chain of parallel
one-dimensional systems, each system being an independent channel of energy; such configuration can, in
fact, model long structures showing piece-wise constant properties along one direction [21].
A particular instance of this class of rods is a chain of ten rods with nominal dimensions as in Table 1; the
rods have different lengths and cross-sections, whilst sharing the same material properties, Table 2. In the 20
Hz - 20 kHz frequency range, the interesting characteristics of such specimen are: a) the transmission factors
are rather high, hence much energy is transmitted at the interfaces, Figure 2-a; b) the energy decay factor is
around unity (maximum average value 1.9) for a single rod, whilst larger than unity for the sum of all rods
(maximum value 20), Figure 2-b; c) above 8 kHz, the strengths of the couplings are weak both according
to the measures by Smith [27] and Finnveden [10], times a factor 2 as in References [6, 11], Figure 2-c;
d) the mode counts of the rods are suitable for SEA calculations [14], Figure 2-d. When a unit point force
excites rod 1 in its middle point, Figure 3 shows the total energies of rods 1 and 4 and the net energy flow in
rod 4, as calculated with an exact SFEM (see Subsection 3.3); these energies are the two most significative
quantities considered in the results Section, Section 4.
This instance of a chain of rods can be the representing specimen of an ensemble of similar structures; an
ensemble is thus defined by perturbing the nominal lengths of this specimen with a normal distribution of
standard deviation 0.01.
3.2
Averages
In order to compare deterministic results to statistical predictions, frequency- space- and ensemble-averages
based on Monte Carlo simulations, are performed on the total energies and on the input powers of the rods,
these quantities being calculated by the spectral finite element method (see Subsection 3.3). First, the narrow
band energies are frequency averaged in third-octave bands. Second, the spatial averaging is obtained by
"$A
"$'
"$%
"
K
J-*);/0@*.G/0FG.=7)01!5
"$C
! % & ' # A B C D
E-=*)FG.*01!5
.
%"
!#
!"
#
"
"$# !
# !"
()*+,*-./012345
@
%"
'
?7@*0.7,-=01!5
67,89:-;0<=)*-;=>01!5
H)G-<I:<<:7-0.7*FF:.:*-=01!5
Figure 1: The chain of 10 rods under study; the outer rods are subject to rigid wall boundary conditions; the
first rod is excited with unit point force in its middle position. In Section 4 the excitation is, however, rainon-the-roof, as explained in Subsection 3.3: a number of longitudinal unit point forces as the one depicted in
the figure are employed.
&
%
!
"$# !
# !"
()*+,*-./012345
!#
!"
#
"
"$# !
# !"
()*+,*-./012345
Figure 2: Characteristics of the chain of rods under study. Figure (a): transmission efficiency at the nine
interfaces of a 10 rods system, Equation (21). Figure (b): energy decay factor as function of frequency,
Equation 13; average value of a 10 rods system (full line), sum of a 10 rod system (dotted line). Figure (c):
coupling strength measures as function of frequency: Finnvedens criterion [10] (full line), Smiths criterion
[27] (dotted line); below 1 the systems are weakly coupled. Figure (d): mode count of the 10 rods as function
of frequency, formula from reference [24], Table V.1.
exciting the system with a large number of uncorrelated longitudinal unit point forces, thus simulating rainon-the-roof [24]. Third, the ensemble average is achieved by calculating the mean response of the ensemble
presented in Subsection 3.1. This averaging process continues till it converges. Convergence is obtained
when the relative standard error (RSE) of the averaged energies is smaller than a threshold (0.1). The RSE
is defined as the ratio between the standard error of the mean (SEM) and the sample mean, whilst the SEM
is defined as the ratio between the sample standard deviation and the square root of the number of iterations.
The sample standard deviation and the sample mean are calculated via an online algorithm by Knuth [28].
Herein, the sample mean represents the frequency- spatial- and ensemble- average of the total energy of a
rod. In particular, each realisation of this sample corresponds to the third octave band averaged total energy,
for a particular specimen of the ensemble and for a particular position of the point force.
3.3
Numerical methods
Three numerical methods are used to predict the total energies and the energy flows: 1) an exact spectral
finite element model (SFEM) [22], 2) an SEA-like model and 3) a proper SEA model.
4,)(5./067/()89/!/:3
!&"
!!""
!!&"
!%""
#
;<=)(/067/()89/!)!!%/>3
!"
'()*+),-./0123
!"
!""
&"
"
!&"
!"
'()*+),-./0123
!"
Figure 3: Deterministic calculation of the energies of the rods. Figure (a): total energies in rods 1 (dashed
line) and 4 (full line) predicted by SFEM. Figure (b): net energy flow into rod 4 predicted by SFEM, Equation
(16).
3.3.1
Spectral finite element
An exact spectral finite element method (SFEM, [22]) is used to calculate exact predictions of the total
energies of the rods and of the input powers; exact is relative to the numerical accuracy of the computer and
to the linear model considered. The SFEM is, consequently, the datum which proper SEA and SEA-like
models are compared to. Post-processing the SFEM outputs allows to retrieve the exact net energy flows,
according to Equation (16). Equation (15) is applied by using the force and velocity at the node between the
considered rods. The SFEM total energies and input powers are averaged as explained in Subsection 3.2;
then, the net energy flows are calculated from the averaged quantities.
3.3.2
SEA-like
The numerical PIM utilises the averaged energies and input powers of the SFEM to devise an SEA-like
model. Each experiment of the PIM consists of exciting one rod with rain-on-the-roof; when one experiment
converges (Subsection 3.2), the energy responses in all rods are calculated; then, the next rod is excited.
Eventually, the energy influence coefficients in matrix A in Equation (7) can be retrieved: each column
of matrices E and Pin corresponds to one experiment of the PIM; eventually, the net energy flows can be
evaluated via Equation (17).
3.3.3
Proper SEA
The elements of the proper SEA model are identified with the nominal rods presented in Section 3.1; the data
used in the proper SEA model, Equation (2), are the nominal dimensions of the chain of rods in Table 1. The
modal densities and the modal overlap factors are calculated via Equations (10) and (11), respectively. The
conductivities in Equation (4) are obtained via their travelling wave estimate, a one way method,
Cij =
Pcij
ei
"
Cij
Mj
(19)
In particular, the following Equation is employed:

Cij =
ij
,
(2 ij )
(20)
!%"
!$"
+,-./0()12(.-34(!(5*
!!""
!!#"
!!&"
!!%"
!!$"
!#"" (
!!
"
!"
'!&()!*
!"
!"
Figure 4: Total energy of rod 1 (dotted line) and of rod 4 (not dotted line); both SFEM prediction (full line)
and proper SEA (dashed line), as function of the energy decay factor D14 .
where the transmission coefficient is defined as [31]:

=!
12
+ 2
"2 ,
(21)
and is the ratio of the the cross-sections of the areas of the receiving and transmitting rod.
The proper SEA model utilises the SFEM prediction of input power, in order to avoid biased results due to
the differences between deterministic and statistical predictions of the input power (see Equation (130) [24]).
Results
This section compares the SFEM, SEA-like and proper SEA predictions of the total energies of the rods and
for their net energy flows. The aim is to highlight the effects of spatial energy decay on tunnelling and on
proper SEA; the system under study is the chain of rods presented in Section 3, when the first rod is excited
with rain-on-the-roof, Figure 3.1. The attention is, in particular, focused on rods 1 and 4; the predictions of
the total energies via the SFE and via proper SEA are compared, whilst when analysing the net energy flows
the predictions of SEA-like are also considered. These quantities are plotted as function of the energy decay
factor D, defined in Equation (13): in particular, the sum of the energy decay factors from the excitation
#
(rod 1) till the point of observation (rod 4) is employed, that is D14 4i=1 Di ; the energy decay factor
is function of frequency only, ranging from 100 Hz till 20 kHz: in fact, Equation (13) utilises the nominal
parameters of the rods.
4.1
Total energies
Figure 4 presents the total energy of rod 1 and rod 4. The energy in rod 1 is correctly predicted by proper
SEA: the rain-on-the-roof excitation, indeed, injects a constant amount of energy in this rod, thus maintaining
its energy density constant regardless of the magnitude of the energy decay factor. As the sum of the energy
decay factors from the excitation to the point of observation increases, the proper SEA prediction for rod
4, however, deviates more than 50 dB from that of the SFEM; in fact, the extremely high damping in rod 4
(Figure 5) causes spatial energy decay.
("
'"
-./01*+23*1045*!0!!$,
&"
%"
#"
$"
!"
"
!!"
!$"
!#" *
!!
!"
"
)!%*+!,
!"
!"
Figure 5: Net energy flow into rod 4 (full) and power dissipated within rod 4 (squares), Equation (8), calculated by the SFEM, as function of the energy decay factor D14 .
4.2
Net energy flow
In order to study the effects of spatial energy decay on tunnelling, all the coupling loss factors defined
by the SEA-like model involving rod 4 are plotted in Figure 6. Direct and indirect coupling loss factors
approximately fulfil the consistency relation at higher energy decay factor, which is expected due to the
proportionality between that and the modal overlap factor. Besides, it is evident that all CLFs decrease as the
energy decay factor increases [20]. The net energy flows into rod 4 calculated by SFEM, SEA-like and proper
SEA are presented in Figure 7. As the sum of the energy decay factors from the excitation to the point of
observation is larger than unity, proper SEA overestimates the net energy flow. On the other hand, the SFEM
and the SEA-like models provide similar predictions: the SFEM yields the exact net energy flow and the
SEA-like model achieves the same result by including all the indirect energy paths. Similar considerations
are valid for Figure 8, where the net energy flow into rod 6 is considered.Finally, Figure 9 shows the ratio
of the indirect energy flows to the direct energy flow, as calculated by the SEA-like model. The two energy
flows with the highest relative magnitudes tend to cancel each other as the energy decay factor increases:
in fact, it yields P14 P24 . The indirect energy flows having the highest magnitude are typically those
ones involving the rods with the highest energy density, rods 1 and 2 indeed. Rod 1, in fact, is excited with
rain-on-the-roof, thus not experiencing spatial energy decay; accordingly, the direct energy flow to rod 2
is high. Moreover, at higher frequencies all indirect energy flows alternate between positive and negative
values, thus cancelling each other and yielding an extremely small net value of transmitted energy (Figure
9). Proper SEA, on the contrary, excluding the indirect CLFs misses this cancellation; eventually, this might
explain the large errors in proper SEA.
Conclusions
This study was motivated by two observations in the literature, the increasing tunnelling in a three-plate
structure with increasing frequency in reference [19], and the consistent overestimation of the energy flow
in a three-pipe structure at very high frequency in reference [21]. In the present work, it is indicated that
the spatial energy decay might probably cause these phenomena, as it increases the tunnelling effect and
jeopardises the validity of proper SEA. The objective of this article is, in fact, twofold. On one side, the
consequences of spatial energy decay on direct and indirect coupling loss factors are investigated. On the
other side, the validity of proper SEA at higher frequencies where spatial energy decay might be consistent
"
!#"
*+,-(./-0123-!)
!&"
!%"
!$"
!!""
!!#"
!!
!"
"
'!&(!)
!"
!"
Figure 6: Coupling loss factors to rod 4, direct CLFs (thick lines) and indirect CLFs (thin lines), as function
of the energy decay factor D14 . Full lines refer to ij , whereas dashed lines to ji . These CLFs are obtained
via a numerical PIM and Equation (1).
is questioned. To accomplish this, deterministic and statistical calculations of the response of a chain of
one-dimensional well-connected rods subject to longitudinal vibrations only are compared. A spectral finite
element method calculates the exact values of the energies, which are used as reference; these results are
then compared to the predictions of proper SEA and by an SEA-like model devised by a numerical PIM.
The simulations show that at higher frequencies energy decays and, thus, proper SEA overestimates the
total energy of a rod away from the force excitation; this result agrees with literature [14, 25]. Moreover,
as the spatial decay of energy becomes more significant, the indirect energy flows between not physically
connected rods have to be considered in order to correctly predict the net energy flow. As a matter of fact,
by considering the indirect CLFs and their relative indirect energy flows, the SEA-like model yields correct
results, whereas proper SEA does not. Therefore, the CPP no longer accurately describes the energy flow
in the structure and, accordingly, proper SEA is unsuitable with a large energy decay factor or, as here for
one-dimensional structures, with a large modal overlap factor. These findings contrast with the conclusions
in reference [20], where it is claimed that a proper SEA model can be devised in the regime of infinite modal
overlap factors, since indirect CLFs asymptote to zero. On the other hand, the present results agree with
the considerations on the cancellation of indirect energy flows by Finnveden and Fredo [12, 19], where it
is found that, for three-element structures, P23 P13 ; here, it is found that, for instance, P14 P24 ,
when considering rod 4. It might be actually argued that proper SEA overestimates the energies of the rods
because it excludes these cancellations, by neglecting the indirect CFLs. Eventually, it is opined that a global
measure of the energy decay factor is more important than a local one in order to determine the applicability
of proper SEA; as a matter of fact, the energy decay factor of a single system plays a minor role: the energy
decay factors of the studied rods are always smaller than unity, whereas the sum of them is not. Accordingly,
a practical criterion to verify the usefulness of proper SEA in one-dimensional systems is proposed: if the
sum of the energy decay factors from the excitation point to the point of observation exceeds unity, energy
decay and tunnelling occur, thus jeopardising proper SEA.
These investigations are restricted to one-dimensional systems. Two considerations, however, can be asserted; first, it is argued that two- and three-dimensional systems with long mean free paths [14], might
probably suffer the same consequences of spatial energy decay; second, one-dimensional systems are actually relevant: structures such ventilation ducts, pipelines, rib-stiffened panels that are common in trains,
ships and aircrafts, are long and show piece-wise constant properties along one direction, thus supporting
a numerable set of waves propagating in this direction; therefore, each section can be modelled as a set of
parallel one-dimensional structures, each being an independent channel of energy flow.
("
'"
-./01*+23*1045*!0!!$*6,
&"
%"
#"
$"
!"
"
!!"
!$"
!#" *
!!
!"
"
)!%*+!,
!"
!"
Figure 7: Net energy flow into rod 4 predicted by proper SEA (dashed line), SFEM (full line) and SEA-like
(dots), as function of the energy decay factor D14 .
$"
%"
+,-./()01(/.23(!.!!&(4*
&"
"
!&"
!%"
!$"
!#"
!!"" (
!!
!"
"
'!$()!*
!"
!"
Figure 8: Net energy flow into rod 4 predicted by proper SEA (dashed line), SFEM (full line) and SEA-like
(dots), as function of the energy decay factor D16 .
Besides the evidences in references [19, 21], personal communication with Doctor Ulf Orrenius of Bombardier confirms that to date proper SEA fails with long structures, where spatial energy decay, in fact,
happens. More complex SEA models accounting for indirect CLFs might possibly overcome this failure,
e.g., advanced SEA [18], extended SEA [19] or quasi-SEA [5]. These models cannot, however, be used in
general predictive SEA software, as it is impractical to derive all the necessary indirect CLFs accounting for
all the possible connections in large structures. Therefore, proper SEA maintains its usefulness and this is
not herein questioned; an alternative formulation of the modal energy accounting for spatial energy decay is,
however, necessary for successfully employing proper SEA with long structures.
Acknowledgement
This work is funded by the Swedish Research Council (621-2005-5754) and the European Commission
(Mid-Mod, Grant agreement no: 218508)
$*
!+,-./012/.-34/15.-67/38+,9
!$*
!#*
!&*
!(*
!"**
!"#
!$#
!#% !#&
!:7;
!#'
!#(
!#) !#"*
Figure 9: Ratio of all indirect energy flows to rod 4 and the direct actual energy flow between rods 3 and
4 (SEA-like); Pij indicates the energy flow between rods i and j. Black: positive power, White: negative
power. The results are plotted for D14 5.
References
[1] R. H. Lyon, R. G. DeJong Theory and Application of Statistical Energy Analysis ButterworthHeinemann, 2nd edition, 1995
[2] R. J. M. Craik Sound transmission through buildings using statistical energy analysis. Gower, England.
1996
[3] MID-MOD URL: http://www.mid-mod.eu/ (retrieved 2010-05-27)
[4] MID-FREQUENCY URL: http://www.midfrequency.org/ (retrieved 2010-05-27)
[5] B. R. Mace Statistical Energy Analysis, energy distribution models and system modes Journal of Sound
and Vibration, 264:391-409, 2003
[6] S. Finnveden A Quantitative Criterion Validating Coupling Power Proportionality in Statistical Energy
Analysis TRITA AVE 2006:97
[7] R. S. Langley A derivation of the coupling loss factors used in statistical energy analysis Journal of
Sound and Vibration, 141:207-219, 1990
[8] R. S. Langley A general derivation of the statistical energy analysis equations for coupled dynamic
systems Journal of Sound and Vibration, 135:499-508, 1989
[9] T. D. Scharton, R. H. Lyon Power flow and energy sharing in random vibration Journal of the Acoustical
Society of America, 43:1332-1343, 1967
[10] S. Finnveden Energy flows within a three element structure with a statistical description of the design
parameters Proceedings of the Inter -Noise 90, 945948, 1990
[11] B. R. Mace The Statistical Energy Analysis of Two Continuous One-Dimensional Subsystems Journal
of Sound and Vibration, 166:429-461, 1993
[12] S. Finnveden Ensemble averaged vibration energy flows in a three-element structure Journal of Sound
and Vibration, 187:495-529, 1995
[13] J. Woodhouse An approach to the theoretical background of statistical energy analysis applied to structural vibration Journal of the Acoustical Society of America, 69:1695-1709, 1981
[14] A. Le Bot, V. Cotoni Validity diagrams of statistical energy analysis, Journal of Sound and Vibration,
329:221-235
[15] J. L. Guyader, C. Boisson, C. Lesueur Energy transmission in finite coupled plates, part I: theory
Journal of Sound and Vibration, 81:81-92, 1982
[16] C. Fredo SEA-like approach for the derivation of energy flow coefficients with a finite element model
Journal of Sound and Vibration, 199:654-666, 1997
[17] R. S. Langley, A. W. M. Brown The ensemble statistics of the band-averaged energy of a random system
Journal of Sound and Vibration, 275:847-857,2004
[18] K. H. Heron Advanced Statistical Energy Analysis Philosophical Transaction of the Royal Society of
London, 1681:501-510
[19] C. Fredo Statistical energy analysis and the individual case PhD Thesis, Department of Applied Acoustic - Chalmers University, 1995
[20] B. R. Mace Statistical energy analysis: coupling loss factor, indirect coupling and system modes Journal
of Sound and Vibration, 279:141-170
[21] S. Finnveden Statistical Energy Analysis of Fluid-Filled Pipes IUTAM symposium on Statistical Energy
Analysis, 289-300, 1999
[22] S. Finnveden Exact spectral finite element analysis of stationary vibrations in a rail way car structure
Acta Acustica, 2:461-482, 1994
[23] V. Cotoni, R.S. Langley, M.R.F. Kidner Numerical and experimental validation of variance prediction
in the statistical energy analysis of built up systems Journal of Sound and Vibration, 288:701-728, 2005
[24] L. Cremer, M. Heckl, E.E. Ungar Structure-Borne Sound 2nd edition, Springer-Verlag, 1988
[25] F. J. Fahy Statistical energy analysis: a critical overview Philosophical Transaction of the Royal Society
of London, 346:431-447, 1994
[26] F. J. Fahy, A. D. Mohammed A study of uncertainty in applications of sea to coupled beam and plate
systems, part 1: computational experiments Journal of Sound and Vibration, 158:45-67, 1992
[27] P. W. Smith Statistical models of coupled dynamical systems and the transition form weak to strong
coupling Journal of the Acoustical Society of America, 65:695-698, 1979
[28] D. E. Knuth The art of computer programming Volume 2, Seminumerical algorithms, 3rd edition,
Addison-Wesley
[29] R. S. Langley A derivation of the coupling loss factors used in statistical energy analysis Journal of
Sound and Vibration, 141:207-219, 1990
[30] R. H. Lyon, G. Maidanik Power flow between linearly coupled oscillators Journal of the Acoustical
Society of America, 34:623-639
[31] B. R. Mace Wave reflection and transmission in beams Journal of Sound and Vibration, 97:237-246,
1984

Trita Ave - 2013 04

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Trita Ave - 2013 04

Uploaded by

Copyright:

Available Formats

Statistical energy analysis and variational principles

for the prediction of sound transmission in

Stockholm, Sweden 2013

Akademisk avhandling som med tillst

Papers contained in the Doctoral thesis

Division of work between the authors of the papers

Overview and Summary

2 Statistical energy analysis

4 SEA of a trim panel

Transmission through multilayered structures

Multilayered structures may be employed to acoustically isolate a target environment

The variational approach

Transmission along long structures

Organisation of the thesis

The rationale of SEA

CHAPTER 2. STATISTICAL ENERGY ANALYSIS

SEA and indirect couplings

CHAPTER 2. STATISTICAL ENERGY ANALYSIS

CHAPTER 2. STATISTICAL ENERGY ANALYSIS

CHAPTER 3. VARIATIONAL PRINCIPLES

CHAPTER 3. VARIATIONAL PRINCIPLES

CHAPTER 3. VARIATIONAL PRINCIPLES

Two interrogatives on Hamiltons principles

On p 43 of Ref. [7], Berdichevsky answers to two fascinating interrogatives4 raised by

The operator DA that is adjoint to D satisfies the following relation

for any function u and v. If D DA , then D is self-adjoint. For instance, if u and v

CHAPTER 3. VARIATIONAL PRINCIPLES

where v is a variable of response, v is its complex amplitude, v is its complex conjugate

CHAPTER 3. VARIATIONAL PRINCIPLES

It follows that e equals minus the pressure of the fluid,

CHAPTER 3. VARIATIONAL PRINCIPLES

Substituting this result into Eq. (3.11), it follows that

A self-adjoint variational principle for Biots

CHAPTER 3. VARIATIONAL PRINCIPLES

W and K are, respectively,

where C expresses a constraint via the Lagrange multipliers and a

Similarly to Eqs. (3.8), in Paper 2 it is shown that

CHAPTER 3. VARIATIONAL PRINCIPLES

The surface functional F in Eq. (3.23) is given by

L is self-adjoint and defines a variational principle:

Couplings to other media

CHAPTER 3. VARIATIONAL PRINCIPLES

Equivalent fluid theory

From the fundamental lemma of calculus of variations, the equations of motion of

CHAPTER 3. VARIATIONAL PRINCIPLES

uti is the total displacement defined as

Variational principle for a trim panel

CHAPTER 3. VARIATIONAL PRINCIPLES

coupled trim-panel, Ltot , reads

CHAPTER 3. VARIATIONAL PRINCIPLES

where pl is a Lagrange multiplier and a,pl its adjoint.

The trim-coupled waves

CHAPTER 3. VARIATIONAL PRINCIPLES

Comment on the variational statements

CHAPTER 3. VARIATIONAL PRINCIPLES

Wave form and modal density of the trimcoupled waves

where the function g x, z describes a wave field along x

CHAPTER 3. VARIATIONAL PRINCIPLES

CHAPTER 4. SEA OF A TRIM PANEL

CHAPTER 4. SEA OF A TRIM PANEL

CHAPTER 4. SEA OF A TRIM PANEL

A Cavity-Wall Element for the Statistical Energy Analysis of the Sound

2. Statistical energy analysis of double walls, a review

$%&'&! a and b !('&!)%&!+,&!-&./)%0!*1!)%()! S = a b 0! a & b 0!(.,!

Pinc !!$%!&'(!)+(,!-.$/0!/&!&'(!)1,&$&$/!2!1,(1! S !3! Pt !!$%!&'(!)*+(,!