Desmet PDF

KATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT TOEGEPASTE WETENSCHAPPEN

DEPARTEMENT WERKTUIGKUNDE
AFDELING PRODUCTIETECHNIEKEN,
MACHINEBOUW EN AUTOMATISERING
Celestijnenlaan 300B – B-3001 Heverlee (Leuven), Belgium
A WAVE BASED PREDICTION TECHNIQUE

FOR COUPLED VIBRO-ACOUSTIC ANALYSIS
Promotoren : Proefschrift voorgedragen tot

Prof. dr. ir. P. SAS het behalen van het doctoraat
Prof. dr. ir. D. VANDEPITTE in de toegepaste wetenschappen
door
Wim DESMET
98D12 December 1998

KATHOLIEKE UNIVERSITEIT LEUVEN
FACULTEIT TOEGEPASTE WETENSCHAPPEN
DEPARTEMENT WERKTUIGKUNDE
AFDELING PRODUCTIETECHNIEKEN,
MACHINEBOUW EN AUTOMATISERING
Celestijnenlaan 300B – B-3001 Heverlee (Leuven), Belgium
A WAVE BASED PREDICTION TECHNIQUE

FOR COUPLED VIBRO-ACOUSTIC ANALYSIS
Jury : Proefschrift voorgedragen tot

Prof. dr. ir. R. Govaerts (voorzitter) het behalen van het doctoraat
Prof. dr. ir. P. Sas in de toegepaste wetenschappen
Prof. dr. ir. D. Vandepitte
Prof. dr. ir. G. De Roeck door
Prof. dr. ir. J.P. Coyette (UCL)
Prof. P. Göransson (KTH Stockholm) Wim DESMET
Prof. dr. W. Lauriks
Prof. dr. ir. G. Degrande
UDC 534.1 December 1998

© Katholieke Universiteit Leuven - Faculteit Toegepaste Wetenschappen
Arenbergkasteel, B-3001 Heverlee (Belgium)
Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd en/of
openbaar gemaakt door middel van druk, fotokopie, microfilm, elektronisch of op
welke andere wijze ook zonder voorafgaande schriftelijke toestemming van de
uitgever.
All rights reserved. No part of the publication may be reproduced in any form, by
print, photoprint, microfilm or any other means without written permission from the
publisher.
D/1998/7515/51
ISBN 90-5682-151-2
DANKWOORD
Een doctoraat maak je nooit alleen. Dat heb ik gelukkig ook zelf mogen
ervaren. Daarom wil ik graag iedereen bedanken die rechtstreeks of
onrechtstreeks heeft bijgedragen tot dit werk.
In de eerste plaats wil ik mijn beide promotoren oprecht bedanken voor de

steun en het vertrouwen, die ze mij doorheen al de jaren gegeven hebben.
Professor Sas wil ik van harte bedanken voor de ruimte die ik gekregen heb
om me wetenschappelijk uit te leven, waarbij hij me met zijn inzicht en
interesse de juiste sturing en motivering wist te geven, en voor de vele
kansen die hij me gaf om dit werk aan een internationaal publiek voor te
leggen.
Mijn gevoel van tevredenheid en voldoening over dit werk heb ik ook in
grote mate te danken aan professor Vandepitte. Hij was steeds bereid om
met zijn klare en kritische kijk mee te denken aan dit project. Met zijn vele
constructieve opmerkingen en nuttige suggesties, zijn enthousiasme en zijn
aanmoedigingen heeft hij een belangrijke hand in dit werk. Voor dit alles,
van harte dank.
Tevens wil ik professor De Roeck bedanken omdat hij de tijd heeft willen
vrijmaken om deze tekst grondig door te nemen en mee vorm te geven.
Furthermore I would like to thank professor Göransson, not only for reading
and evaluating my work as a member of the jury, but also for our interesting
and pleasant co-operation during the BRAIN project. Ook professor
ii
Coyette, professor Lauriks en professor Degrande wil ik bedanken voor hun

betoonde interesse als leden van de jury.
Furthermore I would like to express my gratitude to professor Bolton and

professor Bernhard from Purdue University for introducing me to the
interesting world of poroelastic materials during my stay at the Ray W.
Herrick laboratories. Hierbij wil ik het fonds “Prof. R. Snoeys Foundation”
bedanken voor de financiële ondersteuning van dit studieverblijf, alsook het
- toen nog - Nationaal Fonds voor Wetenschappelijk Onderzoek voor het
toekennen van een vierjarig onderzoeksmandaat.
Ook een welgemeend woordje van dank aan alle mensen van de afdeling
PMA, en in het bijzonder van de ‘modale groep’, voor de wetenschap-
pelijke, technische en administratieve ondersteuning en vooral voor de
aangename werksfeer.
Verder zou ik alle vrienden willen bedanken voor de vele plezante, en vaak
memorabele, bele(u)venissen, waardoor de herinneringen aan de voorbije
jaren steeds heel levendig zullen blijven.
Tot slot wil ik mijn familie bedanken. Hun vele bemoedigende woorden
tijdens de soms moeilijke momenten hebben me veel deugd gedaan. Hierbij
wil ik vooral mijn ouders bedanken ... voor alles!
Heel in het bijzonder wil ik Leen bedanken voor haar liefdevol begrip en
haar geduldige steun, die ervoor gezorgd hebben dat ik dit werk tot een goed
einde heb kunnen brengen. Aan haar wil ik dan ook graag dit werk op-
dragen.
ABSTRACT
This dissertation presents a new, wave based prediction technique for the
steady-state dynamic analysis of coupled vibro-acoustic systems. The
technique is based on a Trefftz approach, in which the field variables are
expanded in terms of structural and acoustic wave functions, which are
homogeneous solutions of the governing dynamic equations, along with
some particular solution functions. In this way, the dynamic equations are
exactly satisfied, irrespective of the contributions of the wave functions to
the field variable expansions. These contributions result from a weighted
residual or a least-squares formulation of the boundary conditions.
It follows from the application of this methodology for various types of
coupled vibro-acoustic problems that accurate predictions can be obtained
with wave models, which are not only significantly smaller, but also
computationally less demanding than corresponding element based models.
Due to this enhanced computational efficiency, the practical frequency
limitation of the proposed technique is substantially higher than for the
existing techniques. In this way, the wave based prediction technique
enables accurate predictions in the mid-frequency range, for which the
computational efforts, involved with element based models, become
prohibitively large.
iv
EEN GOLFGEBASEERDE VOORSPELLINGS-
TECHNIEK VOOR GEKOPPELDE VIBRO-
AKOESTISCHE ANALYSE
Beknopte samenvatting
Dit proefschrift stelt een nieuwe, golfgebaseerde voorspellingstechniek voor

voor de dynamische regimerespons van gekoppelde vibro-akoestische
systemen. De techniek is gebaseerd op een Trefftz-benadering, waarbij de
veldveranderlijken worden beschreven met structurele en akoestische
golffuncties, die homogene oplossingen zijn van de dynamische vergelij-
kingen, aangevuld met de nodige particuliere oplossingsfuncties. Op deze
manier is er exact voldaan aan de dynamische vergelijkingen, ongeacht de
bijdrages van de golffuncties tot de benaderingen voor de veldverander-
lijken. Deze golffunctiebijdrages worden bepaald aan de hand van een
gewogen-residu of een kleinste-kwadraten formulering van de randvoor-
waarden.
Uit de toepassing van deze methodiek voor een aantal representatieve vibro-
akoestische problemen is gebleken dat een grote nauwkeurigheid bekomen
wordt met golfmodellen, die niet alleen van beduidend kleinere dimensie
zijn dan elementgebaseerde modellen, maar tevens een kleinere reken-
kundige belasting vergen. Bijgevolg wordt de limiet van de beschikbare
rekencapaciteit pas bereikt bij veel hogere frequenties dan bij de huidige
technieken. Hierdoor biedt de nieuwe techniek een geschikte manier om ook
in het middenfrequentiegebied nauwkeurige voorspellingen te maken.
vi Samenvatting van het proefschrift
1. Inleiding en doelstelling van het onderzoek

belang van vibro-akoestische voorspellingstechnieken
Wegens de steeds hoger wordende eisen van de klant en onder druk van een
sterk competitieve markt wordt de ontwerpingenieur geconfronteerd met een
steeds toenemend aantal ontwerpcriteria, die vaak tegenstrijdige
doelstellingen vooropstellen. De industriële wereld is tot het besef en de
overtuiging gekomen dat het inschakelen van adequate simulatietechnieken
in alle stadia van het ontwerp- en ontwikkelingsproces een noodzakelijke
vereiste is om de complexe ontwerpopdrachten doelgericht en efficiënt aan
te pakken. Deze technieken laten toe zowel inzicht te verwerven in de
gevoeligheden van de ontwerpcriteria voor de verschillende ontwerp-
parameters alsook het tijds- en kostenverslindende bouwen en testen van
prototypes tot een minimum te beperken.
Mede door de voortdurend scherper wordende wettelijke reglementeringen
inzake geluidsemissie en menselijke blootstelling aan lawaai krijgen de
akoestische eigenschappen van een product een steeds hogere prioriteit
binnen het geheel van ontwerpcriteria. Vaak worden deze akoestische
eigenschappen bepaald door structuurgeluid, waarbij akoestische druk-
golven veroorzaakt worden door de dynamische verplaatsingen van een
elastische structuur. Gezien de drukgolven ook een mechanische belasting
vormen voor de structuur, zijn deze golven niet alleen het gevolg maar ook
mede de oorzaak van de structuurtrillingen. Vooral voor systemen met een
of meerdere akoestische caviteiten, omsloten door een dunwandige
elastische structuur met geringe massa, is deze vibro-akoestische koppeling
niet te verwaarlozen.
Om inzicht te verwerven in de invloed van parameters zoals geometrie,
randvoorwaarden en structurele en akoestische demping op dit complexe
koppelingsmechanisme en om tot een efficiënt ontwerp van
geluidsreducerende maatregelen te komen, zijn betrouwbare vibro-
akoestische voorspellingstechnieken noodzakelijk. De geluidsniveaus in de
passagiersruimte van wagens of vliegtuigen zijn sterk afhankelijk van de
wisselwerking tussen de drukgolven in deze ruimte en de dynamische
respons van de wagencarrosserie of de dunwandige rompstructuur. De
vibro-akoestische koppelingseffecten zijn bepalend voor de geluids-
transmissie doorheen machine-omkapselingen en dubbelwandige structuren
zoals dubbelwandige glaspanelen. Dit zijn slechts enkele van de vele
toepassingen waarbij vibro-akoestische koppeling een dominante rol speelt
in de akoestische eigenschappen. Met het toenemend gebruik van
Samenvatting van het proefschrift vii
lichtgewicht composietmaterialen, zeker in de auto-, luchtvaart- en ruimte-

vaartindustrie, zullen deze vibro-akoestische koppelingseffecten trouwens
nog aan belang winnen.
beperkingen van de huidige voorspellingstechnieken

Tot op heden zijn de voornaamste deterministische voorspellingstechnieken
voor akoestische en structuurdynamische analyses gebaseerd op de eindige-
elementen- en randelementenmethode.
In deze methodes worden de akoestische en structurele domeinen of hun
randoppervlakken opgedeeld in kleine elementen. Binnen elk element
worden de dynamische veranderlijken beschreven door een aantal
eenvoudige vormfuncties. De bijdrages van deze vormfuncties tot de
oplossingsbenadering worden bepaald aan de hand van een integraal-
formulering van de dynamische vergelijkingen en de randvoorwaarden van
het beschouwde continuumprobleem. Gezien de vormfuncties binnen elk
element slechts een benaderende oplossing kunnen genereren, dient een
voldoende aantal elementen gebruikt te worden om de benaderingsfout
binnen aanvaardbare grenzen te houden.
Voor de voorspelling van de regimerepons ten gevolge van een
tijdsharmonische excitatie is de excitatiefrequentie een bepalende factor
voor het aantal elementen en de daaruit voortvloeiende modelgroottes.
Naarmate de frequentie van de systeemexcitatie verhoogt, verhoogt ook de
ruimtelijke variatie van de dynamische respons, waardoor een fijnere
elementopdeling vereist is. Dit resulteert in steeds groter wordende
modellen, waardoor de vereiste rekentijden eveneens toenemen met de
frequentie. Door deze zware rekenkundige belasting zijn de eindige-
elementen- en randelementenmethode praktisch slechts bruikbaar tot een
bepaalde frequentiegrens. Voor hogere frequenties worden de modellen
dermate groot dat het opstellen en oplossen ervan, zelfs met de enorme
rekenkracht van de moderne computers, onnoemelijk veel rekentijd vergt.
In vergelijking met ongekoppelde akoestische of structuurdynamische
problemen is de praktische bruikbaarheid van elementgebaseerde technieken
voor gekoppelde vibro-akoestische problemen beperkt tot beduidend lagere
frequenties. Dit is voornamelijk toe te schrijven aan de volgende factoren.
• Gekoppelde vibro-akoestische modellen zijn aanzienlijk groter, gezien
een structureel en een akoestisch probleem simultaan opgelost dienen te
worden.
• De numerieke oplossingsmethodes voor gekoppelde vibro-akoestische
modellen zijn minder efficiënt, daar de gekoppelde modellen niet langer
symmetrisch zijn, althans niet in hun meest gebruikte vorm, waarbij de
viii Samenvatting van het proefschrift
akoestische druk en de structurele verplaatsing de dynamische

veldveranderlijken zijn.
• Het verlies aan modelsymmetrie is tevens de oorzaak van de aanzienlijk
lagere efficiëntie van modelreductietechnieken zoals de modale expansie
techniek.
Doelstelling
Het enige momenteel beschikbare alternatief is gebruik te maken van
probabilistische voorspellingstechnieken, die gebaseerd zijn op statistische
energie analyse of vermogenstroomanalyse. Deze technieken zijn veel
minder rekenintensief, maar laten enkel toe een ruimte- en frequentie-
gemiddelde schatting te maken van de dynamische respons. Gezien deze
technieken een voldoende hoge modale densiteit van het beschouwde
dynamische systeem veronderstellen, zijn ze enkel geschikt voor
hoogfrequente dynamische analyses. Tussen het laagfrequente toepassings-
gebied van de elementgebaseerde technieken en het hoogfrequente
toepassingsgebied van de probabilistische technieken ligt echter nog een
breed frequentiegebied, waarvoor nog geen afdoende voorspellings-
technieken beschikbaar zijn.
Voor heel wat industriële toepassingen is er nochtans nood aan
deterministische technieken die in dit frequentiegebied een nauwkeurige
voorspelling toelaten. In antwoord op dit hiaat stelt dit proefschrift een
nieuwe deterministische techniek voor, die in staat is betrouwbare
gekoppeld vibro-akoestische voorspellingen te maken voor een veel breder
frequentiegebied dan de bestaande elementgebaseerde technieken.
2. Methodiek van de voorspellingstechniek
2.1. Inleiding
De nieuwe voorspellingstechniek is gebaseerd op de indirecte Trefftz-

methode voor het oplossen van continuumproblemen. In deze methode
worden de veldveranderlijken beschreven door vormfuncties, die exact
voldoen aan de vergelijkingen van het beschouwde continuumprobleem. De
bijdrages van de vormfuncties tot de uiteindelijke oplossing worden bepaald
aan de hand van een integraalformulering van de randvoorwaarden.
Het grote voordeel van een dergelijke vormfunctiekeuze is dat er enkel een
benaderingsfout wordt gemaakt op de randvoorwaarden. Hierdoor kan een
goede voorspellingsnauwkeurigheid bekomen worden met aanzienlijk
Samenvatting van het proefschrift ix
kleinere modellen dan met overeenkomstige eindige-elementenmodellen die

gebruik maken van vormfuncties, die niet exact voldoen aan de
vergelijkingen. De indirecte Trefftz-methode vereist echter wel dat een
volledig set van vormfuncties gedefiniëerd kan worden, waarvoor de
convergentie naar de exacte oplossing verzekerd is. Voor verschillende
types continuumproblemen zijn dergelijke volledige vormfunctiesets reeds
beschikbaar, waarvoor de theoretische convergentie bewezen is. Hun
praktische implementatie leidt echter tot slecht-geconditioneerde modellen,
wat de praktische convergentie sterk verstoort of zelfs onmogelijk maakt.
Dit verklaart waarom de indirecte Trefftz-methode tot op heden nog niet
uitgebouwd is tot een volwaardig modelleringsalternatief voor de element-
gebaseerde technieken.
Eén van de belangrijke verdiensten van het in dit proefschrift voorgestelde
onderzoek is dat nieuwe, volledige vormfunctiesets gedefiniëerd zijn, die
aanleiding geven tot deterministische modellen, die, ondanks hun slechte
conditie, heel nauwkeurige voorspellingen toelaten van de dynamische
regimerespons van gekoppelde vibro-akoestische systemen.
2.2. Basisprincipes
2.2.1. Probleemdefinitie
De basisprincipes van de voorspellingstechniek worden besproken aan de

hand van een tweedimensionaal gekoppeld vibro-akoestisch probleem, zoals
getoond in figuur 1.
Figuur 1 : tweedimensionaal gekoppeld vibro-akoestisch probleem

x Samenvatting van het proefschrift
Het randoppervlak Ωa van een gesloten akoestische caviteit V bestaat uit vier
deeloppervlakken. Op de deeloppervlakken Ωp, Ωv en ΩZ zijn,
respectievelijk, gekende druk-, normaalsnelheids- en normaal-
impedantieverdelingen opgelegd, terwijl het deeloppervlak Ωs bestaat uit
een dunne, vlakke plaat.
De caviteit bevat een fluïdum met dichtheid ρ0 en geluidssnelheid c. Het
materiaal van de vlakke plaat met dikte t heeft een dichtheid ρs, een Poisson
coëfficiënt ν, een elasticiteitsmodulus E en een verliesfactor η.
Op plaatpositie rF(xF’) wordt in de normaalrichting een externe lijnkracht F
aangelegd, terwijl een externe akoestische lijnbron q aangelegd wordt op
caviteitspositie rq(xq,yq). Beide excitaties hebben een harmonisch tijds-
verloop met frequentie ω.
In regimetoestand is de akoestische druk p in de caviteit bepaald door de
Helmholtzvergelijking
∇ 2 p( r ) + k 2 . p( r ) = − jρ 0ωq.δ ( r ,r q ), (1)
waarbij k(=ω/c) het akoestisch golfgetal is en waarbij δ een Dirac delta

functie voorstelt. De normaalverplaatsing w van de plaat is bepaald door de
dynamische plaatvergelijking
F p( rs )
∇ 4 w( rs ) − kb4 . w( rs ) = .δ ( rs , rF ) + , (2)
D D
waarbij het buigingsgolfgetal kb en de buigstijfheid D van de plaat als volgt

gedefiniëerd zijn,
ρ tω 2 Et 3 (1 + jη )
kb = 4 s en D = . (3)
D 12(1 − ν 2 )
De akoestische randvoorwaarden voor het beschouwde gekoppelde vibro-

akoestische probleem zijn
j ∂p( r )
r ∈ Ω p : p( r ) = p ( r ), r ∈Ω v : = vn ( r ),
ρ 0ω ∂n
Samenvatting van het proefschrift xi
j ∂p( r ) p( r ) j ∂p( r )
r ∈Ω Z : = , r ∈Ω s : = jωw( r ), (4)
ρ 0ω ∂n Z (r ) ρ 0ω ∂n
waarbij p , vn en Z respectievelijk de opgelegde druk-, normaalsnelheids-

en normaalimpedantieverdelingen voorstellen. De normaalverplaatsing van
de plaat is éénduidig bepaald door op de beide randposities twee structurele
randvoorwaarden op te leggen.
2.2.2. Benadering voor de veldveranderlijken
Voor de caviteitsdruk p(x,y) wordt een benaderende oplossing p(x, y)

voorgesteld van de vorm
na
p ( x , y ) ≈ p( x , y ) = ∑ p a . Φ a ( x , y ) + pq ( x , y ) . (5)
a=1
Elke akoestische golffunctie Φ a (x, y) is een homogene oplossing van de

Helmholtzvergelijking (1),
− j ( k xa . x + k ya . y )
Φ a ( x, y) = e (6)

waarbij 2
k xa + k ya
2
= k2. (7)
De functie pq (x, y) is een particuliere oplossing voor de niet-homogene

Helmholtzvergelijking (1). Hiervoor wordt de analytische uitdrukking
gebruikt voor het drukveld ten gevolge van een akoestische lijnbron in vrije-

veld omstandigheden.
Voor de normaalverplaatsing w(x’) van de plaat wordt een benaderende

oplossing w(x’) voorgesteld van de vorm
4 na
w( x’) ≈ w( x’) = ∑ ws .Ψ s ( x’) + ∑ pa . wa ( x’) + w F ( x’) + wq ( x’). (8)
s=1 a=1
De structurele golffuncties Ψ s (x’) zijn vier linear onafhankelijke, homo-

gene oplossingen van de vierde-orde plaatvergelijking (2),
xii Samenvatting van het proefschrift

Ψ s ( x’) = e − j
s
. k b . x’
, ( s = 1..4) . (9)
De functie w F (x’) is een particuliere oplossing voor de lijnkrachtterm in

het rechterlid van de plaatvergelijking. Hiervoor wordt de analytische

uitdrukking gebruikt voor het verplaatsingsveld van een oneindige plaat ten
gevolge van een lijnkrachtexcitatie in de normaalrichting van de plaat.

De functie wq (x’) en de functies w a (x’) zijn particuliere oplossingen voor
die delen van de akoestische drukterm in het rechterlid van de plaat-
vergelijking, die overeenstemmen met, respectievelijk, de functie pq (x, y)
en de akoestische golffuncties Φ a (x, y) in de voorgestelde oplossing (5).
Hiervoor kunnen de analytische integraaluitdrukkingen gebruikt worden

voor het verplaatsingsveld van een oneindige plaat ten gevolge van deze
drukbelastingen. In het geval van een vlakke plaat kunnen voor de functies
w a (x’) echter uitdrukkingen geformuleerd worden, die evenredig zijn met
de akoestische golffuncties. Hoewel deze uitdrukkingen geen directe
fysische betekenis hebben, worden ze verkozen boven de fysisch betekenis-
volle integraaluitdrukkingen omwille van hun rekenkundig voordeel.
Met deze oplossingsbenaderingen wordt het gekoppeld vibro-akoestisch
continuumprobleem omgevormd tot een benaderend, discreet probleem,
waarbij de onbekende golffunctiebijdrages pa en ws in de voorgestelde
oplossingen (5) en (8) bepaald moeten worden. Een belangrijke eigenschap
van deze oplossingen is dat ze steeds exact voldoen aan de dynamische
vergelijkingen (1) en (2), ongeacht de waardes van de onbekenden pa en ws.
2.2.3. Gekoppeld vibro-akoestisch golfmodel
Gezien de voorgestelde oplossingsbenaderingen (5) en (8) exact voldoen aan

de dynamische vergelijkingen, zijn de onbekende golffunctiebijdrages pa en
ws uitsluitend bepaald door de randvoorwaarden van het gekoppeld vibro-
akoestisch probleem.
structurele randvoorwaarden
Voor het beschouwde tweedimensionale probleem zijn de structurele rand-
voorwaarden gedefiniëerd op beide randposities van de plaat en kunnen als
dusdanig gebruikt worden. Deze randvoorwaarden resulteren in vier
vergelijkingen in de 4+na onbekende golffunctiebijdrages,
Samenvatting van het proefschrift xiii
w s 
[ Ass Csa ].   = { fs }. (10)
î pa 
akoestische randvoorwaarden
Gezien enkel matrixmodellen van eindige grootte numeriek opgelost kunnen
worden, dienen de akoestische randvoorwaarden (4), die gedefiniëerd zijn
op een oneindig aantal posities op het randoppervlak van de caviteit,
benaderd te worden door een eindig aantal vergelijkingen. Hiervoor zijn
twee types benaderende integraalformuleringen voorgesteld.
1. gewogen-residu formulering
In een gewogen-residu formulering wordt, op basis van een gewichts-
functie ~
p , het gewogen gemiddelde van de foutfuncties (residu’s) op de
akoestische randvoorwaarden ten gevolge van de voorgestelde
− j ∂~
p
oplossingsbenaderingen nul gesteld,
j ∂p

~
∫ ρ ω ∂n .( p − p ). dΩ + ∫ p.( ρ ω ∂n − vn ). dΩ +
Ωp 0 Ωv 0
(11)
~ j ∂p p ~ j ∂p
∫ p.( ρ ω ∂n − Z ). dΩ + ∫ p.( ρ ω ∂n − jωw ). dΩ = 0.
ΩZ 0 Ωs 0
Door elke akoestische golffunctie Φ a in (5) als gewichtsfunctie te

gebruiken in deze gewogen-residu formulering, worden de akoestische
randvoorwaarden benaderend omgezet tot na vergelijkingen in de 4+na
onbekende golffunctiebijdrages,
[Cas ]
( GR )  s 
{ }
w
( GR ) ( GR )
Aaa + Caa .   = fa( GR ) , (12)
î pa 
( GR )
waarbij de (naxna) deelmatrix Aaa [ ]
symmetrisch is.
2. kleinste-kwadraten formulering
De kleinste-kwadraten formulering is gebaseerd op een functionaal Fa,
die een maat is voor de totale kwadratische benaderingsfout op de
akoestische randvoorwaarden,

xiv Samenvatting van het proefschrift
2
2 j ∂p
Fa = ∫ β p . p − p . dΩ + ∫ β v . − vn . dΩ +

Ωp Ωv
ρ 0 ω ∂n
(13)
2 2
j ∂p p j ∂p
∫ β Z . ρ ω ∂n − Z . dΩ + ∫ β s . ρ ω ∂n − jωw . dΩ .
ΩZ 0 Ωs 0
De parameters β p, β v, β Z en β s laten toe het relatief belang van de

verschillende types randvoorwaarden vast te leggen, alsook de
dimensionele homogeniteit van de verschillende functionaaltermen te
bewaren.
Door deze functionaal te minimaliseren met betrekking tot elke
akoestische golffunctiebijdrage pa, worden de akoestische randvoor-
waarden benaderend omgezet tot na vergelijkingen in de 4+na onbekende
golffunctiebijdrages,
[Cas ]
( KK )  s 
{ }
w
( KK )
Aaa .   = fa( KK ) , (14)
î pa 
[
( KK )
waarbij de (naxna) deelmatrix Aaa ]
Hermitiaans is.
gekoppeld vibro-akoestisch golfmodel

De combinatie van de structurele randvoorwaarden (10) met de gewogen-
residu formulering (12) of de kleinste-kwadraten formulering (14) van de
akoestische randvoorwaarden resulteert in een gekoppeld vibro-akoestisch
golfmodel met 4+na vergelijkingen in de 4+na onbekende golffunctie-
bijdrages pa en ws.
2.2.4. Convergentie
In de oplossingsbenaderingen (5) en (8) kan slechts een eindig aantal, na,

akoestische golffuncties van type (6) gebruikt worden, hoewel een oneindig
aantal golffuncties voldoet aan voorwaarde (7) voor de golfgetalcompo-
nenten kxa en kya. De hamvraag is bijgevolg of er uit de oneindige verzame-
ling van golffuncties een eindige selectie kan gemaakt worden, zodanig dat
de oplossingsbenaderingen, in de limiet voor na→∝, convergeren naar de
exacte oplossing.
Samenvatting van het proefschrift xv
In hoofdstuk 2 van dit proefschrift is aangetoond dat, voor het geval van een
convexe caviteit, een convergerende oplossing bekomen wordt door gebruik
te maken van de akoestische golffuncties met golfgetalcomponenten
a .π a .π a .π a .π
( k xa , k ya ) = ( 1 ,± k 2 − ( 1 ) 2 ) en ( ± k 2 − ( 2 ) 2 , 2 ) , (15)
Lx Lx Ly Ly
waarbij a1,a2=0,±1,±2,... en waarbij Lx en Ly de afmetingen zijn van een

omschrijvende rechthoek rond het convexe caviteitsdomein.
Hoewel het wiskundig bewijs nog ontbreekt, is er geen tegenvoorbeeld
gevonden voor de stelling dat, voor het geval van een concave caviteit, deze
opgesplitst kan worden in een aantal deelcaviteiten en dat een converge-
rende oplossing bekomen wordt door, in elke deelcaviteit, gebruik te maken
van golffuncties met golfgetalcomponenten van type (15), waarbij Lx en Ly
de afmetingen zijn van een omschrijvende rechthoek rond de deelcaviteit, en
door bijkomende continuïteitsvoorwaarden voor druk en normaalsnelheid op
te leggen op de grensoppervlakken tussen de verschillende deelcaviteiten.
Op basis van een aantal fysische overwegingen is in hoofdstuk 2 tevens een
vuistregel opgesteld omtrent de waardes voor a1 en a2, waarbij de oneindige
reeks van golfgetalcomponenten (15) afgebroken kan worden, om een goede
voorspellingsnauwkeurigheid te bekomen. In dit opzicht is tevens gebleken
dat een gewogen-residu formulering van de randvoorwaarden een betere
convergentiesnelheid oplevert dan een kleinste-kwadraten formulering.
2.3. Eigenschappen
2.3.1. Vergelijking met de eindige-elementenmethode
De golfgebaseerde voorspellingstechniek biedt een aantal belangrijke

voordelen ten opzichte van de eindige-elementenmethode.
• Gezien de benaderingen voor de veldveranderlijken exact voldoen aan de

dynamische vergelijkingen, wordt er enkel een benaderingsfout gemaakt
op de randvoorwaarden. Hierdoor zijn golfmodellen van een beduidend
kleinere dimensie dan overeenkomstige eindige-elementenmodellen.
• Gezien de ruimtelijke afgeleiden van de golffuncties eveneens golf-
functies zijn met een zelfde ruimtelijk verloop, kunnen de afgeleide
grootheden, zoals de akoestische deeltjessnelheid en de structurele
xvi Samenvatting van het proefschrift
spanningen, met een zelfde graad van ruimtelijk detail benaderd worden
als de akoestische druk en de structurele verplaatsing. Dit is voordelig
voor de convergentiesnelheid, vooral voor gekoppelde vibro-akoestische
problemen, waarbij het effect van het fluïdum op de structuur druk-
gecontroleerd is, maar waarbij het effect van de structuur op het fluïdum
snelheidsgecontroleerd is.
• Modelverfijning kan op een efficiënte manier gebeuren. Om de model-
nauwkeurigheid in te schatten, moeten enkel de benaderingsfouten op de
randvoorwaarden nagegaan worden. Daarenboven moeten bij het
toevoegen van golffuncties aan de benaderingen voor de veldverander-
lijken enkel de bijkomende matrixelementen berekend worden, terwijl de
oorspronkelijke elementen in een golfmodel behouden blijven. In de
eindige-elementenmethode vereist een globale verfijning van het
elementennet een volledig nieuwe modelberekening.
De golfgebaseerde voorspellingstechniek heeft ook een aantal nadelige

eigenschappen in vergelijking met de eindige-elementenmethode.
• Gezien de golffuncties complexe functies zijn, die in het volledige of

toch minstens in grote deelgebieden van het continuumdomein gedefi-
niëerd zijn, zijn golfmodellen volledig bevolkt met complexe matrix-
elementen.
• Door de impliciete frequentieafhankelijkheid van de golffuncties kunnen
de matrices in een golfmodel niet ontbonden worden in frequentie-
onafhankelijke deelmatrices. Dit heeft voor gevolg dat een golfmodel
volledig herberekend moet worden voor elke frequentie en dat de
natuurlijke frequenties en modevormen niet berekend kunnen worden
aan de hand van een standaard eigenwaardenprobleem.
• De numerieke integraties voor de berekening van de matrixelementen in
een golfmodel vereisen een beduidend groter aantal integratiepunten en
bijgevolg ook een groter aantal rekenkundige bewerkingen. Dit is ener-
zijds te wijten aan het globale en sterk oscillerende karakter van de
golffuncties, en anderzijds aan het feit dat de slechte conditie van een
golfmodel een hoge integratienauwkeurigheid vereist.
• De eindige-elementenmethode heeft een heel ruim toepassingsgebied,
gezien de convergentie van een eindige-elementenmodel reeds bekomen
wordt met eenvoudige vormfunctiebenaderingen. De convergentie van
een golfmodel vereist daarentegen de definitie van een volledig set van
golffuncties en van geschikte particuliere oplossingen.
Samenvatting van het proefschrift xvii
2.3.2. Vergelijking met de randelementenmethode
Net zoals de randelementenmethode vereist de golfgebaseerde voorspel-

lingstechniek de definitie van een particuliere oplossing (kernfunctie) en
resulteert de techniek in kleine, maar complexe, volledig bevolkte en
frequentie-afhankelijke modellen.
Naast deze gemeenschappelijke eigenschappen, heeft de golfgebaseerde
voorspellingstechniek ook een aantal voordelige eigenschappen.
• In heel wat gevallen bevat een integraalformulering, ttz. een gewogen-

residu of een kleinste-kwadraten formulering, van de randvoorwaarden
slechts enkelvoudige randintegralen. In de randelementenmethode
daarentegen vereist een integraalformulering de rekenkundig duurdere
berekening van dubbele randintegralen. Enkel een collocatieformulering
geeft aanleiding tot enkelvoudige randintegralen, wat rekenkundig
weliswaar efficiënter is, maar resulteert in een kleinere convergentie-
snelheid dan een integraalformulering.
• In tegenstelling tot de directe randelementenmodellen zijn golfmodellen
voor ongekoppelde (akoestische) problemen symmetrisch of Hermi-
tiaans.
• De golfgebaseerde voorspellingstechniek vereist geen singuliere
integraalberekeningen, noch speciale numerieke bewerkingen om de
éénduidigheid van de oplossing te vrijwaren.
• Bij een globale verfijning van een randelementennet moet een volledig
nieuw randelementenmodel berekend worden. Zoals eerder reeds
aangegeven, moeten bij de verfijning van een golfmodel echter enkel de
bijkomende matrixelementen berekend worden, terwijl de oorspronke-
lijke elementen behouden blijven.
• Na het oplossen van een golfmodel naar de onbekende golffunctie-
bijdrages, volgen de benaderingen voor de veldveranderlijken uit
eenvoudige functie-evaluaties. In de randelementenmethode bekomt men
deze benaderingen pas na bijkomende integraalberekeningen.
De golfgebaseerde voorspellingstechniek heeft ook een aantal nadelige

eigenschappen in vergelijking met de randelementenmethode.
• De convergentie van een randelementenmodel wordt reeds bekomen met

eenvoudige vormfunctiebenaderingen, terwijl de convergentie van een
golfmodel de definitie van een volledig set van golffuncties vereist.
xviii Samenvatting van het
proefschrift
• Ieder matrixelement in een golfmodel vereist een integraalevaluatie over
het volledig randoppervlak van het continuumdomein, terwijl de
integratiegebieden voor de matrixelementen in een randelementenmodel
telkens beperkt zijn tot een klein deel van het randoppervlak.
• Een golfmodel voldoet niet impliciet aan de Sommerfeld stralings-
voorwaarde, die van toepassing is voor problemen met een oneindig
uitgestrekt continuumdomein.
2.4. Numeriek voorbeeld
Het volgende voorbeeld illustreert de nauwkeurigheid en de convergentie

van de golfgebaseerde voorspellingstechniek. Figuur 2 toont een twee-
dimensionaal gekoppeld vibro-akoestisch systeem, waarbij een deel van het
randoppervlak van de caviteit bestaat uit een dunne, vlakke plaat met inge-
klemde randen. De tussenschotten met hoogte 0.5*Ly en alle andere delen
van het randoppervlak van de caviteit zijn star.
De caviteit (Lx1+Lx2+Lx3=1.5 m, Ly=1 m) is gevuld met lucht (ρ0=1.225 kg/m ,
3
c=340 m/s) en de aluminium plaat (E=70.10 N/m , ρs=2790 kg/m3, ν=0.3,

9 2
t=0.002 m, η=0) wordt geëxciteerd door een eenheidslijnkracht op positie

x0’=0.5 m. De excitatiefrequentie is ω=2π.200 Hz.
Figuur 2 : tweedimensionaal gekoppeld vibro-akoestisch systeem

met een concave caviteit
Samenvatting van het proefschrift xix
Om een convergerend golfmodel te bekomen, wordt de concave caviteit

opgesplitst in drie deelcaviteiten, zoals aangegeven in figuur 2. De
benaderingen van de drukvelden in deze convexe deelcaviteiten worden
beschreven met akoestische golffuncties, waarvan de golfgetalcomponenten
van type (15) bepaald worden op basis van de afmetingen Lx1xLy, Lx2xLy en
Lx3xLy van de omschrijvende rechthoekige domeinen.
Aan de hand van de inklemmingsrandvoorwaarden van de plaat en de
gewogen-residu formulering van de akoestische randvoorwaarden is een
golfmodel opgesteld voor het bepalen van de bijdrages van 4 structurele en
540 akoestische golffuncties in de benaderingen voor de caviteitsdruk en de
normaalverplaatsing van de plaat.
Figuur 3 toont de resulterende plaatverplaatsing en de normaalverplaatsing
van de lucht aan het grensoppervlak tussen de plaat en de caviteit. Op deze
figuur is duidelijk te zien dat de inklemmingsvoorwaarden van de plaat en
de continuïteitsvoorwaarde voor de structurele en akoestische normaal-
verplaatsing nauwkeurig voorgesteld zijn.
Figuur 3 : ogenblikkelijke normaalverplaatsing w van de plaat (volle lijn) en

akoestische normaalverplaatsing wa (+) aan het plaat-caviteit
grensoppervlak bij 200 Hz
Figuur 4 toont het resulterende drukveld in de caviteit. Het feit dat de

contourlijnen loodrecht staan op de starre delen van het randoppervlak van
de caviteit en dat ze continu verlopen over de grensoppervlakken tussen de
verschillende deelcaviteiten, toont aan dat ook de overige akoestische rand-
voorwaarden nauwkeurig voorgesteld zijn.
xx Samenvatting van het proefschrift
Figuur 4 : ogenblikkelijke caviteitsdruk bij 200 Hz
Gezien alle randvoorwaarden nauwkeurig voorgesteld zijn en gezien de

benaderingen voor de veldveranderlijken exact voldoen aan de dynamische
vergelijkingen, mag men besluiten dat de voorgestelde techniek in staat is
een heel nauwkeurige voorspelling te maken van de dynamische regime-
respons van het beschouwde vibro-akoestische systeem.
Figuur 5 : structurele en akoestische convergentiecurves bij 200 Hz

Samenvatting van het proefschrift xxi
Figuur 5 toont de relatieve benaderingsfout ∆w op de normaalverplaatsing

op de plaatpositie x’=0.25m (∇) en de relatieve benaderingsfouten ∆p op de
druk op drie caviteitsposities ((x,y)=(0.2m,0.2m) (+), (x,y)=(1m,0.5m) (o) en
(x,y)=(1.3m,0.8m) (*)) in functie van het aantal golffuncties, dat gebruikt
wordt voor de benaderingen van de veldveranderlijken. Deze figuur
illustreert dat de voorgestelde set van akoestische golffuncties toelaat om,
ondanks de slechte conditie van de resulterende golfmodellen, tot een
convergerende oplossing te komen.
3. Toepassing voor twee- en driedimensionale gekoppel-

de vibro-akoestische problemen
3.1. Toepassingsgebied
Hoofdstukken 3 en 4 van dit proefschrift beschrijven hoe de golfgebaseerde

voorspellingstechniek toegepast kan worden voor verscheidene types twee-
en driedimensionale gekoppelde vibro-akoestisch problemen.
Vooreerst worden vibro-akoestische problemen behandeld, waarbij een deel
van het randoppervlak van een tweedimensionale, gesloten caviteit bestaat
uit een vlakke plaat of uit een agglomeraat van cilindrische schaalsecties.
Voor dergelijke problemen is de benadering voor de caviteitsdruk -
eventueel na opsplitsing van de caviteit in een aantal (convexe)
deelcaviteiten - gebaseerd op een set van akoestische golffuncties, waarvan
de golfgetalcomponenten afhangen van het akoestisch golfgetal en van de
afmetingen van een omschrijvende rechthoek rond de (deel)caviteit (zie
(15)). De benadering voor de normaalverplaatsing van een vlakke plaat is
gebaseerd op 4 structurele golffuncties (zie (9)), terwijl de benaderingen
voor de radiale en tangentiële verplaatsingen van cilindrische schaalsecties
gebaseerd zijn op 6 structurele golffuncties. Uit verschillende toepassings-
voorbeelden van dergelijke vibro-akoestische problemen is gebleken dat
deze structurele en akoestische golffuncties, aangevuld met de nodige
particuliere oplossingsfuncties, toelaten om heel nauwkeurige voorspel-
lingen te maken van de dynamische regimerespons.
Vervolgens is de golfgebaseerde voorspellingsmethodiek toegepast voor het
modelleren van het dynamisch gedrag van poreuze isolatiematerialen. In
tegenstelling tot een niet-viskeus fluïdum, waarin slechts één longitudinaal
golftype kan optreden, kunnen in een poreus medium drie golftypes
optreden; twee longitudinale golftypes en één transversaal golftype.
Bijgevolg worden de verplaatsingen en spanningen in zowel de vaste faze
xxii Samenvatting van het proefschrift
als de fluïdumfaze van een tweedimensionaal poreus medium benaderd met

drie golffunctiesets van type (15), elk gebaseerd op één van de drie
golfgetallen. In de toepassingsvoorbeelden is ondermeer aangetoond dat de
aanname van een ruimtelijk onafhankelijke normaalimpedantie voor een
poreuze dempingslaag in een caviteit tot aanzienlijke modelleringsfouten
kan leiden.
Om de geluidsafstraling van een gekoppeld vibro-akoestisch systeem in een
oneindig uitgestrekt akoestisch domein te kunnen voorspellen, wordt op een
eindige afstand van het systeem een bijkomend randoppervlak gedefiniëerd,
waarop de specifieke akoestische impedantie ρ0c als normaalimpedantie
opgelegd wordt. Op deze manier kan het afgestraald geluidsveld benaderend
gemodelleerd worden als een drukveld in een gesloten caviteit, dat
beschreven wordt met een golffunctieset van type (15).
Tenslotte worden vibro-akoestische problemen behandeld, waarbij een deel
van het randoppervlak van een driedimensionale, gesloten caviteit bestaat
uit een vlakke plaat. De caviteitsdruk wordt op een volledig gelijkaardige
manier benaderd als bij tweedimensionale problemen. De akoestische
golffuncties bevatten nu echter wel drie golfgetalcomponenten, waarbij -
naar analogie met het tweedimensionale set (15) - telkens twee compo-
nenten omgekeerd evenredig zijn met de overeenkomstige afmetingen van
een omschrijvend rechthoekig prisma rond de caviteit en waarbij de derde
component volgt uit het feit dat aan de homogene Helmholtzvergelijking
voldaan moet zijn. De benadering voor de structurele normaalverplaatsing
vereist nu eveneens een golffunctieselectie, aangezien een oneindig aantal
lopende buigingsgolven met golfgetal kb en niet-lopende golven met
golfgetal jkb kunnen optreden in een vlakke plaat. Opnieuw wordt een
convergerende oplossing bekomen door twee golffunctiesets van type (15)
te selecteren, elk gebaseerd op één van de twee golfgetallen.
3.2. Vergelijking met de eindige-elementenmethode
De ontwikkeling van de golfgebaseerde voorspellingstechniek heeft als doel

om via een verhoogde rekenkundige efficiëntie nauwkeurige voorspellingen
te kunnen maken voor een veel breder frequentiegebied dan mogelijk met de
bestaande technieken. Daarom zijn voor verscheidene gekoppelde vibro-
akoestische problemen de nauwkeurigheid en de bijhorende rekenkundige
belasting van de nieuwe techniek en van de eindige-elementenmethode met
elkaar vergeleken. Uit deze vergelijkingen is duidelijk gebleken dat de
doelstelling bereikt is. Het volgende voorbeeld illustreert dit.
Samenvatting van het proefschrift xxiii
Figuur 6 toont een driedimensionale caviteit, waarvan een deel van het
randoppervlak bestaat uit een vlakke plaat met ingeklemde randen. Alle
andere delen van het randoppervlak van de caviteit zijn star. Het systeem
wordt geëxciteerd door een tijdsharmonische normaalkracht op de plaat.
Figuur 6 : driedimensionaal gekoppeld vibro-akoestisch systeem
Voor het geval van een aluminium plaat (E=70.109 N/m2, ρs=2790 kg/m3,
ν=0.3, t=0.002 m, η=0), waarbij de caviteit (Lx=1.5 m, Ly=0.5 m, Lz=1 m)
met lucht (ρ0=1.225 kg/m3, c=340 m/s) gevuld is, toont figuur 7 de directe
frequentieresponsiefuncties, die bekomen worden met een golfmodel met
314 golffuncties (volle lijn) en met eindige-elementenmodellen (streepjes-
lijn) met 1849 (a), 12586 (b) en 39991 (c) knoopveranderlijken.
Deze figuur illustreert duidelijk het effect van de opdeling van een domein
in kleine elementen. Door het benaderen van de veldveranderlijken met
eenvoudige vormfuncties, wordt de ‘stijfheid’ van het systeem overschat.
Dit heeft voor gevolg dat de resonantiefrequenties systematisch overschat
worden en dat deze overschatting, voor een gegeven elementennet, groter
wordt naarmate de frequentie stijgt. Uit deze figuur blijkt tevens dat, door
het verfijnen van het elementennet, de eindige-elementenresultaten
geleidelijk convergeren naar de resultaten van het golfmodel, dat een bedui-
dend kleinere dimensie heeft.
Dat de nieuwe techniek, in vergelijking met de eindige-elementenmethode,
een hoge nauwkeurigheid oplevert met niet alleen beduidend kleinere, maar
ook rekenkundig efficiëntere modellen, mag blijken uit figuur 8. Deze
figuur toont de relatieve benaderingsfouten ∆w en ∆p op, respectievelijk, de
normaalverplaatsing van een punt van de vlakke plaat en de druk in een punt
van de caviteit in functie van de vereiste rekentijd.
xxiv Samenvatting van het proefschrift
Figuur 7 : directe frequentieresponsiefunctie

(golfmodel (volle lijn) met 314 golffuncties, EE-modellen (streepjeslijn)
met 1849 (a), 12586 (b) en 39991 (c) knoopveranderlijken)
Merk hierbij op dat de rekentijden voor de golfmodellen (volle lijn) zowel

de tijd voor het berekenen als voor het oplossen van het model omvatten.
Voor de eindige-elementenmodellen (streepjeslijn) worden enkel de
oplossingstijden aangegeven, gezien deze modellen niet voor elke frequentie
opnieuw berekend worden, maar samengesteld worden uit frequentie-
Samenvatting van het proefschrift xxv
onafhankelijke deelmatrices. Merk ook op dat de golfgebaseerde voorspel-

lingstechniek geïmplementeerd is in een MATLAB-omgeving, terwijl de
eindige-elementenresultaten bekomen zijn met het MSC/NASTRAN-
pakket. Door de nieuwe techniek eveneens in een efficiëntere programmeer-
omgeving te implementeren zal de verhoogde rekenkundige efficiëntie, naar
alle waarschijnlijkheid, nog meer uitgesproken zijn.
Figuur 8 : structurele (links) en akoestische (rechts) convergentiecurves

bij 60 Hz (+) en 180 Hz (∇) (golfmodellen (volle lijn), EE-modellen (streepjeslijn))
4. Besluiten
Dit proefschrift stelt een nieuwe, golfgebaseerde voorspellingstechniek voor
voor de dynamische regimerespons van gekoppelde vibro-akoestische
systemen. De techniek is gebaseerd op de indirecte Trefftz-methode,
waarbij, in tegenstelling tot elementgebaseerde technieken, de structurele en
akoestische domeinen niet meer opgedeeld worden in kleine elementen,
maar elk in hun geheel beschouwd of in een klein aantal deeldomeinen
opgedeeld worden. De benaderingen voor de veldveranderlijken in elk
(deel)domein worden beschreven door structurele en akoestische
xxvi Samenvatting van het proefschrift
golffuncties, die homogene oplossingen zijn van de dynamische

vergelijkingen, aangevuld met de nodige particuliere oplossingsfuncties. Op
deze manier is er steeds exact voldaan aan de dynamische vergelijkingen,
ongeacht de bijdrages van de golffuncties tot de benaderingen voor de
veldveranderlijken. Deze golffunctiebijdrages worden bepaald aan de hand
van een gewogen-residu of een kleinste-kwadraten formulering van de
randvoorwaarden.
De voornaamste beperking van de indirecte Trefftz-methode is dat een
volledig set van vormfuncties gedefiniëerd moet worden, waarvoor de
convergentie naar de exacte oplossing verzekerd is. Voor sommige types
continuumproblemen zijn dergelijke volledige vormfunctiesets reeds
beschikbaar. Hun praktische implementatie leidt echter tot slecht-gecondi-
tioneerde modellen, wat de praktische convergentie sterk verstoort of zelfs
onmogelijk maakt. Dit verklaart waarom de indirecte Trefftz-methode tot op
heden nog niet uitgebouwd is tot een volwaardig modelleringsalternatief
voor de elementgebaseerde technieken. Eén van de belangrijke verdiensten
van het voorgestelde onderzoek is dat volledige golffunctiesets gedefiniëerd
zijn, die aanleiding geven tot deterministische modellen, die, ondanks hun
slechte conditie, heel nauwkeurige voorspellingen toelaten van de
dynamische regimerespons van gekoppelde vibro-akoestische systemen.
Uit de toepassing van de golfgebaseerde voorspellingstechniek voor een
aantal representatieve vibro-akoestische problemen is gebleken dat hoge
nauwkeurigheden bekomen worden met golfmodellen, die niet alleen van
beduidend kleinere dimensie zijn dan elementgebaseerde modellen, maar
tevens een kleinere rekenkundige belasting vergen. Bijgevolg wordt de
limiet van de beschikbare rekencapaciteit pas bereikt bij veel hogere
frequenties dan bij de huidige technieken. Hierdoor biedt de nieuwe
techniek - hoewel nog een aantal onderzoeksstappen nodig zijn, vooral met
betrekking tot de algemene toepasbaarheid en de automatisatie van het
modelleringsproces - een geschikte manier om ook in het middenfrequentie-
gebied nauwkeurige voorspellingen te maken, waarvoor de bestaande
technieken praktisch ontoereikend zijn.
NOMENCLATURE
The main symbols, used in this dissertation, are listed below.

2
A first Lamé constant (N/m )
b viscous coupling factor (kg/m3s)
c speed of sound (m/s)
D plate bending stiffness (Nm)
es solid phase volumetric strain (-)
E modulus of elasticity (N/m2)
E1 modulus of elasticity of the bulk solid phase (N/m2)
E2 modulus of elasticity of the bulk fluid phase (N/m2)
Ese contribution of a poroelastic dilatational wave function (-)
f normal force distribution (N/m2)
F normal force (2D: N/m, 3D: N)
Fa acoustic functional
G Green’s kernel function
h porosity (-)
H 0( 2) zero-order Hankel function of the second kind
j imaginary unit
k acoustic wavenumber (1/m)
k1 poroelastic dilatational wavenumber (1/m)
k2 poroelastic dilatational wavenumber (1/m)
kb structural bending wavenumber (1/m)
kt poroelastic rotational wavenumber (1/m)
xxviii
mn plate bending moment per unit length (N)

M bending moment per unit length in a cylindrical shell section (N)
n normal vector
n normal direction
na number of acoustic wave functions
ne number of poroelastic dilatational wave functions
ns number of structural wave functions
nw number of poroelastic rotational wave functions
2
N shear modulus (N/m )
p acoustic pressure (N/m2)
~
p acoustic weighting function
pa acoustic wave function contribution (N/m2)
pq particular solution function for the acoustic pressure, due to an
acoustic source distribution (N/m2)
q acoustic volume velocity distribution (1/s)
Q shear force per unit length in a cylindrical shell section (N/m)
Qa acoustic volume velocity (2D : m2/s, 3D : m3/s)
Qn generalised plate shear force per unit length (N/m)
r position vector (m)
R radius of a cylindrical shell section (m)
Rφ residual error function, associated with a boundary condition for
dynamic quantity φ
s fluid stress (N/m2)
t plate thickness (m)

time (s)
u solid phase displacement vector (m)
un normal solid phase displacement (m)
us tangential solid phase displacement (m)

U fluid phase displacement vector (m)

Un normal fluid phase displacement (m)

v velocity vector (m/s)

v circumferential displacement of a cylindrical shell section (m)
vn normal velocity (m/s)
V acoustic domain
Vp poroelastic domain
w normal structural displacement (m)
w~ structural weighting function
xxix
wa particular solution function for the normal displacement, due to the

pressure loading, associated with an acoustic wave function (m)
wF particular solution function for the normal displacement, due to a

normal mechanical force (m)
wq particular solution function for the normal displacement, due to the
pressure loading from an acoustic source distribution (m)
ws structural wave function contribution (m)
Wsw contribution of a poroelastic rotational wave function (-)
Z impedance (kg/m2s)
βp pressure weighting parameter in an acoustic functional
β s, β v velocity weighting parameters in an acoustic functional
βZ impedance weighting parameter in an acoustic functional
γn direction, normal to a surface boundary
γs direction, tangent to a surface boundary
ΓmQ structural boundary with prescribed generalised shear force and
bending moment distributions
Γs boundary of a structural surface
Γwθ structural boundary with prescribed normal translational and
rotational displacement distributions
Γwm structural boundary with prescribed normal displacement and
bending moment distributions
δ Dirac delta function (1D : 1/m, 2D : 1/m2, 3D : 1/m3)
∆ϕ relative prediction error on dynamic quantity ϕ
ε fluid phase volumetric strain (-)
ε‘ geometrical structure factor (-)
η material loss factor (-)
θn normal rotational displacement (rad)
λ acoustic wavelength (m)
λb structural bending wavelength (m)
λc measure for the fluid-structure coupling interaction (-)
ν Poisson coefficient (-)
ρ0 ambient mass density of the fluid (kg/m )
3
ρ1 3
bulk density of the solid phase (kg/m )
ρ2 bulk density of the fluid phase (kg/m3)
ρs structural mass density (kg/m3)
σ flow resistivity (Ns/m4)
σi singular value
xxx
σn solid normal forces per unit poroelastic material area (N/m2)

τns solid shear forces per unit poroelastic material area (N/m2)
ϕ angular position (rad)
Φa acoustic wave function
Ψe poroelastic dilatational wave function
Ψs structural wave function
Ψω poroelastic rotational wave function
ω circular frequency (rad/s)
ωs solid phase rotational strain (-)
Ωa acoustic boundary surface
Ωc boundary surface between two subdomains
Ωp acoustic boundary surface with a prescribed pressure distribution
Ωp1 poroelastic boundary surface with prescribed normal fluid, normal
solid and tangential solid displacement distributions
Ωp2 poroelastic boundary surface with prescribed fluid stress, normal
and shear solid stress distributions
Ωs elastic structural surface
Ωv acoustic boundary surface with a prescribed normal velocity
distribution
ΩZ acoustic boundary surface with a prescribed normal impedance
distribution
Ω∝ acoustic boundary surface at infinity
^ approximation
−
prescribed value
| | amplitude
Im( ) imaginary part
Re( ) real part
* complex conjugate
H
complex conjugate transpose
T
transpose
∇

gradient vector
∇2 Laplace operator
TABLE OF CONTENTS
DANKWOORD i
ABSTRACT iii
NEDERLANDSE SAMENVATTING v
NOMENCLATURE xxvii
TABLE OF CONTENTS xxxi
1. INTRODUCTION AND STATE-OF-THE-ART IN COUPLED

VIBRO-ACOUSTIC MODELLING 1
1.1. Introduction 1
1.1.1. Fluid-structure interaction 1
1.1.2. Importance of (vibro-acoustic) numerical modelling 3
1.1.3. Scope of the dissertation 4
1.2. Formulation of a coupled vibro-acoustic problem 5
1.3. Existing numerical prediction techniques 11

1.3.1. Basic concepts and properties of FEM and BEM 12
xxxii
1.3.1.1. FEM 12
1.3.1.2. BEM 22
1.3.1.3. FEM versus BEM for acoustic problems 30
1.3.2. Coupled FE/FE models 31
1.3.2.1. Interior problems 31
1.3.2.2. Exterior problems 36
1.3.3. Coupled FE/BE models 39
1.3.3.1. Coupled FE/direct BE model 39
1.3.3.2. Coupled FE/indirect BE model 40
1.3.4. Coupled FE/FE models versus coupled FE/BE models 40
1.3.5. Limitations of coupled models 42
1.4. Recent advances in numerical prediction techniques 48

1.4.1. Motivation 48
1.4.2. Advances in element based prediction techniques 49
1.4.3. Trefftz approach 52
1.4.3.1. Trefftz method 52
1.4.3.2. Dynamic stiffness method 56
1.4.3.3. T-element method 59
1.4.4. Wave based prediction technique 61
1.4.4.1. Basic principles 61
1.4.4.2. Outline of the dissertation 62
2. METHODOLOGY OF THE WAVE BASED PREDICTION

TECHNIQUE 65
2.1. Introduction 65
2.2. Problem definition 67
2.3. Field variable expansions 70

2.3.1. Acoustic pressure expansion 70
2.3.2. Structural displacement expansion 71
2.4. Coupled vibro-acoustic wave model 73

2.4.1. Integral formulation of the boundary conditions 73
2.4.1.1. Weighted residual formulation 74
2.4.1.2. Least-squares formulation 78
2.4.2. Model properties 80
2.5. Convergence requirements 82

2.5.1. Introductory considerations 82
2.5.2. Rectangular continuum domain 84
2.5.3. Non-rectangular continuum domain 88
xxxiii
2.5.4. Wave model implementation 93
2.6. Practical implementation 104

2.6.1. Wave function scaling 104
2.6.2. Numerical integration 106
2.6.3. Wave function truncation 107
2.7. Numerical condition and convergence 109

2.7.1. Condition 109
2.7.2. Convergence 113
2.7.2.1. A posteriori convergence assessment 113
2.7.2.2. Some validation examples 114
2.7.2.3. Weighted residual versus least-squares model 122
2.7.2.4. Comparison with an existing complete function set 124
2.8. Comparison with FEM and BEM 128
2.9. Conclusions 130
3. APPLICATION OF THE WAVE BASED PREDICTION

TECHNIQUE FOR TWO-DIMENSIONAL COUPLED VIBRO-
ACOUSTIC PROBLEMS 133
3.2. Interior coupled vibro-acoustic problems with flat structural

components 134
3.2.1. Vibro-acoustic reciprocity 134
3.2.2. Comparison with coupled FE/FE models 136
3.3. Interior coupled vibro-acoustic problems with curved structural

components 142
3.3.1. Problem definition 142
3.3.1.1. Dynamic equations for circular cylindrical shell structures 142
3.3.1.2. Coupled vibro-acoustic problem 145
3.3.2. Field variable expansions 146
3.3.2.1. Acoustic pressure expansion 146
3.3.2.2. Structural field variable expansions 147
3.3.3. Coupled vibro-acoustic wave model 150
3.3.4. Validation examples 151
3.3.4.1. Cavity-backed circular cylindrical shell section 151
3.3.4.2. Fluid-loaded circular cylindrical shell 153
xxxiv
3.3.5. Comparison with coupled FE/FE models 156
3.4. Interior coupled vibro-acoustic problems with poroelastic insulation

material 165
3.4.1.1. Dynamic equations 166
3.4.1.2. Boundary conditions 171
3.4.3. Wave model 174
3.4.4. Validation example 178
3.4.5. Accuracy of a spatially invariant impedance model 181
3.5. Exterior coupled vibro-acoustic problems 187
4. APPLICATION OF THE WAVE BASED PREDICTION

TECHNIQUE FOR THREE-DIMENSIONAL COUPLED
VIBRO-ACOUSTIC PROBLEMS 193
4.2. Uncoupled plate problems 194

4.2.2. Field variable expansion 197
4.2.3. Weighted residual wave model 198
4.2.4. Convergence requirements 204
4.2.5. Some validation examples 211
4.2.6. Comparison with an existing complete function set 215
4.3. Uncoupled interior acoustic problems 217

4.3.2. Field variable expansion 218
4.3.3. Weighted residual wave model 219
4.3.4. Convergence requirements 220
4.3.5. Validation example 225
4.4. Coupled vibro-acoustic problems 227

4.4.3. Coupled vibro-acoustic wave model 237
4.4.4. Performance comparison with coupled FE/FE models 240
4.4.5. Modal analysis of a double-panel partition 251
xxxv
5. CONCLUSIONS AND FUTURE DEVELOPMENTS 259
5.2. Future developments 263
APPENDIX A. FINITE ELEMENT METHOD FOR

UNCOUPLED ACOUSTIC PROBLEMS 265
APPENDIX B. FINITE ELEMENT METHOD FOR

UNCOUPLED STRUCTURAL PROBLEMS 297
APPENDIX C. DIRECT BOUNDARY ELEMENT METHOD

FOR UNCOUPLED ACOUSTIC PROBLEMS
335
APPENDIX D. COUPLED FE/FE AND COUPLED FE/BE

MODELS 351
APPENDIX E. WAVE PROPAGATION IN FLUID-

SATURATED POROELASTIC MEDIA 371
APPENDIX F. COUPLED VIBRO-ACOUSTIC BEHAVIOUR

OF DOUBLE-PANEL PARTITIONS 391
REFERENCES 433
xxxvi
1. INTRODUCTION AND STATE-OF-THE-ART IN
COUPLED VIBRO-ACOUSTIC MODELLING
1.1. Introduction
1.1.1. Fluid-structure interaction
Whenever an elastic structure is (partly) in contact with a fluid, the

structural vibrations and the acoustic pressure field in the fluid are
influenced by the mutual vibro-acoustic coupling interaction: the force
loading on the structure, caused by the acoustic pressure along the fluid-
structure interface, influences the structural vibrations, while at the same
time the acoustic pressure field in the fluid is also sensitive to the structural
vibrations along the fluid-structure interface. The strength of this vibro-
acoustic coupling interaction is largely dependent on the geometry of the
structure and the fluid domain as well as on the fluid and structural material
properties and on the frequency of the dynamic disturbances. Depending on
the strength of the mutual interaction, vibro-acoustic systems may be
classified into uncoupled and coupled systems.
In uncoupled vibro-acoustic systems, the mutual vibro-acoustic coupling
interaction between the structural and the fluid components is very weak
and may be neglected in the analysis of the dynamic behaviour of such
systems. In this way, only a one-way interaction must be considered,
yielding two types of dynamic problems for uncoupled vibro-acoustic
systems. In the first type, the structural components are regarded as
2 Chapter 1
independent external excitations for the fluid components, in that the

vibrations of the structural components generate acoustic pressures in the
fluid components, whose force loading effects on the exciting structural
vibrations are negligible. In the second type of dynamic problems, the fluid
components act as independent force excitations for the structural
components, in that the pressure distributions along the fluid-structure
interfaces, caused by the acoustic waves in the fluid, generate vibrations in
the structural components, whose effects on the exciting pressure
distributions are negligible. In the latter, the exciting pressure distributions
are usually referred to as ‘blocked’ pressure distributions to indicate that the
pressure components, radiated by the structural vibrations, are negligible
and that the considered pressure distributions correspond with the ones that
would occur when the structural components are rigid (‘blocked’).
Considering only a one-way interaction between the structural and fluid
components is a reasonable assumption for many vibro-acoustic systems,
especially when an elastic structure with a high stiffness is surrounded by a
low-density fluid. An engine block, surrounded by air, is an example of an
uncoupled vibro-acoustic system, in which only the effect of the structure on
the fluid is important, since the radiated pressure field, induced by the
engine block vibrations, has a negligible effect on the engine block
vibrations. The analysis of the vibrations of window panes in houses,
induced by adjacent traffic noise, is an example of an uncoupled vibro-
acoustic problem, in which the effect of the structural vibrations on the
acoustic pressure excitation is negligible.
In coupled vibro-acoustic systems, the mutual vibro-acoustic coupling
interaction between the structural and the fluid components is no longer
negligible and all components must be regarded as parts of one coupled
system, instead of considering the structural components as independent
excitations for the fluid components or vice versa. Strong coupling effects
may occur, for instance, when an elastic structure is submerged in a high-
density fluid, as it is the case, for example, for submarines, or when a thin,
light-weight structure is in contact with even a low-density fluid, as it is the
case, for example, for electrodynamic loudspeakers.
ATALLA and BERNHARD (1994) defined a dimensionless measure λc to
indicate the strength of the mutual coupling interaction in a vibro-acoustic
system,
ρ 0c
λc = , (1.1)
ρ s tω
Introduction and state-of-the-art in coupled vibro-acoustic modelling 3
where ω is the circular frequency of a time-harmonic structural or acoustic

excitation of the system, t is a characteristic thickness of the structure, ρs is
the mass density of the structure, ρ0 and c are the mass density and sound
speed of the fluid. Based on this coupling measure λc, a criterion may be
defined to assess the importance of the vibro-acoustic coupling interaction:
systems with large values of λc (> 1) should be regarded as coupled vibro-
acoustic systems, whereas systems with small values of λc (<< 1) may be
regarded as uncoupled vibro-acoustic systems. However, the coupling
measure λc is not fully comprehensive, especially with respect to the size of
the fluid domain, and the associated criterion is not always sufficient to
denote a vibro-acoustic system as coupled or uncoupled. For instance, the
dynamic behaviour of an elastic structure enclosing a small acoustic cavity
can be strongly influenced by the fluid pressure, even when the cavity is
filled with a low-density fluid. This is the case, for example, for the
interaction between the vibrations of the double wall aircraft fuselage
structure and the acoustic pressure in the enclosed fluid and, to some extent,
for the interaction between the vibrations of the car body and the acoustic
pressure in the interior of a car.
In this framework, it must be noticed that an ambiguous definition of vibro-
acoustic coupling may be found in literature. An elastic structure is
sometimes said to ‘couple well with a fluid’ to indicate its strong one-way
interaction with the fluid, i.e. its ability to efficiently radiate acoustic energy
into the far-field1, and not to indicate a strong mutual vibro-acoustic
coupling interaction, as defined above and as will be used throughout this
dissertation.
1.1.2. Importance of (vibro-acoustic) numerical modelling
With the steadily increasing customer demands and competitive nature of

the market, engineers face the challenging but complex problem of meeting
the ever expanding but often conflicting design criteria. To accomplish this
difficult task, the industry has become aware and convinced that a major
1
an important difference between the radiation efficiency of vibrating structures
and the mutual coupling interaction between an elastic structure and a fluid
concerns their frequency dependence: the radiation efficiency usually increases
for increasing frequency (see e.g. FAHY (1985)), whereas the strength of the
mutual coupling interaction usually decreases with increasing frequency, as
indicated e.g. by the coupling measure λc in (1.1)
4 Chapter 1
imperative for any advance beyond the very time-consuming and expensive
prototype testing approach is to apply appropriate predictive engineering
methods in all stages of the design process (see e.g. SORGATZ (1995,1996)).
Over the last few years, customer demands regarding acoustic performance,
along with the tightening of the legal regulations on noise emission levels
and human exposure to noise, have made the acoustic behaviour into an
important criterion in many design problems. In the automotive and
aerospace industry, for instance, the passengers’ acoustic comfort has
become an important commercial asset. This affects the efforts in reducing
the weight of cars and aircraft, which are mainly motivated by potential fuel
savings but quite often induce substantial noise and vibration levels. In this
framework, vibro-acoustic coupling phenomena are yet important and will
become even more important in the course of the next decades with the
introduction of new light-weight materials such as composites (see e.g.
CHARRON et al. (1995), SORGATZ (1996)).
In order to incorporate the vibro-acoustic criteria in the design process, there
is a strong need for numerical prediction tools that provide the necessary
insight in the physical phenomena which govern the (coupled) vibro-
acoustic behaviour of complex systems and that allow a reliable evaluation
of different design alternatives (see e.g. FAVRE (1998)).
1.1.3. Scope of the dissertation
The numerical prediction of the dynamic behaviour of uncoupled vibro-

acoustic systems may be performed in a two-step procedure. When the
effect of the fluid pressure on the structural vibrations is negligible, the first
step consists of calculating the structural vibrations, induced by the given
external force excitations, without taking into account the acoustic pressure
loading. In the second step, the acoustic pressure field in the fluid is
calculated, using the structural vibrations, obtained in the first step, as
velocity boundary conditions. When the effect of the structural vibrations on
the fluid pressure field is negligible, the first step consists of calculating the
fluid pressure field, induced by the given external acoustic sources,
disregarding the flexibility of the structural components. In the second step,
the structural vibrations are calculated, using the pressure distributions
along the fluid-structure interfaces, obtained in the first step, as external
force excitations.
The numerical prediction of the dynamic behaviour of coupled vibro-
acoustic systems cannot be split into a structural and an acoustic
subproblem, due to the strong mutual coupling interaction between the

structural and fluid components. For such coupled vibro-acoustic problems,
presented in their general form in section 1.2., several numerical prediction
techniques are already available. Due to the large amount of computational
efforts and memory resources, involved with the application of these
techniques, their use is practically restricted to low-frequency dynamic
analyses, as discussed in section 1.3. Therefore, there is a strong need for
alternative prediction techniques, which can be used for the analysis of the
dynamic behaviour of coupled vibro-acoustic systems in a much larger
frequency range. In this framework, some recent advances in numerical
prediction techniques, outlined in section 1.4., exhibit some promising
features, but still need a vast amount of research to achieve a mature and
versatile prediction technique for coupled vibro-acoustic problems.
Motivated by the current lack of efficient prediction tools for coupled vibro-
acoustic problems, a wave based prediction technique has been developed.
Throughout the discussion of its methodology and its application for two-
and three-dimensional coupled vibro-acoustic problems, presented in this
dissertation, it will become apparent that this technique provides an
adequate way to comply with the challenge of extending the applicability of
numerical prediction techniques towards higher frequencies.
1.2. Formulation of a coupled vibro-acoustic problem

For modelling purposes, a fundamental distinction should be made between
interior and exterior coupled vibro-acoustic systems, depending on whether
the acoustic domain is bounded or unbounded.
In an interior coupled vibro-acoustic system, the fluid is comprised in a
bounded acoustic domain V, of which the boundary surface Ωa contains an
elastic structural surface Ωs (Ωa=Ωs ∪ Ωp ∪ Ωv ∪ ΩZ), as shown in figure 1.1.
Figure 1.1 : interior coupled vibro-acoustic system

6 Chapter 1
In an exterior coupled vibro-acoustic system, the fluid is comprised in an

unbounded acoustic domain V, which is the space between a boundary
surface Ωa, containing an elastic structural surface Ωs, and a boundary
surface Ω∝, located at infinity, as shown in figure 1.2.
Figure 1.2 : exterior coupled vibro-acoustic system
Although any type of elastic structure could be part of a coupled vibro-

acoustic system, the elastic structure is represented as a ‘surface’ component
since, according to (1.1), significant vibro-acoustic coupling effects only
occur for thin shell-type elastic structures, which have a small thickness
dimension.
The goal of a coupled vibro-acoustic model is the prediction of the dynamic

quantities of the system, i.e. the fluid pressure field in the acoustic domain
and the displacement field of the elastic structure. In order to accomplish
this goal, a full mathematical formulation of the acoustic and structural
dynamic behaviour is required. The set of dynamic equations and boundary
conditions that govern the fluid pressure and structural displacement in
interior and exterior coupled vibro-acoustic systems are given below. The
dynamic quantities are confined to the steady-state response fields for

external structural and acoustic excitations with harmonic time dependence
e jωt , which will be suppressed throughout ( j = −1 ).
fluid pressure field
The steady-state pressure p (small dynamic perturbation to an ambient

reference state with no mean fluid flow) at any point with generalised co-
ordinate r in a homogeneous, inviscid, irrotational fluid in domain V is
governed by the linear Helmholtz wave equation
∇ 2 p( r ) + k 2 p( r ) = − jρ 0ωq( r ), r ∈ V , (1.2)
where ∇ 2 is the Laplace operator, k (=ω/c) is the acoustic wavenumber of

the longitudinal waves in the fluid, ω is the circular frequency, c is the phase
speed of the longitudinal waves, ρ0 is the ambient fluid density and q(r) is a
distribution of acoustically applied forces (volume velocity per unit volume)
in domain V. The derivation of the Helmholtz equation is given in appendix
A.
When the pressure field is known in the entire acoustic domain, all other
acoustic quantities can be derived from this pressure field. For instance, the
fluid velocity vector is proportional to the gradient of the pressure,
j
v (r ) = ∇p( r ). (1.3)
ρ 0ω
Since the second-order Helmholtz equation (1.2) describes one longitudinal

wavetype, the pressure field in V is uniquely defined, if one boundary
condition is specified at each point on the boundary surface Ωa of domain V.
For interior coupled vibro-acoustic systems (see figure 1.1), four different
types of boundary conditions may occur:
• imposed pressure (Dirichlet or essential boundary condition) :
p( r ) = p ( r ), r ∈ Ω p , (1.4)
where p( r ) is a prescribed pressure function;

8 Chapter 1
• imposed normal velocity (Neumann or natural boundary condition) :
j∂ p( r )
vn (r ) = = v n ( r ), r ∈Ω v , (1.5)
ρ 0 ω ∂n
where vn ( r ) is a prescribed normal velocity function (n denotes the di-

rection, normal to the boundary surface);
note that a rigid boundary wall ( vn ≡ 0 ) is a special case of this boun-
dary condition;
• normal impedance boundary condition (mixed boundary condition) :
the effect of a boundary layer of insulation material is often represented

by a normal impedance function Z (r), which relates the pressure to the
fluid velocity, normal to the boundary,
jZ ( r ) ∂p( r )
p( r ) = Z ( r ). vn ( r ) = , r ∈Ω Z ; (1.6)
ρ 0ω ∂ n
• normal displacement continuity at the fluid-structure interface

(Neumann or natural boundary condition):
the structural displacement component wn, normal to the fluid-structure

coupling interface, must be identical to the normal fluid displacement,
1 1 ∂p( r )
wn ( r ) = vn ( r ) = , r ∈Ω s . (1.7)
jω ρ 0ω 2 ∂n
The same types of boundary conditions as defined in (1.4)-(1.7) for the

interior case, may occur at the boundary surface Ωa of exterior coupled
vibro-acoustic systems (see figure 1.2). In addition to these boundary
conditions, the Sommerfeld radiation condition (SOMMERFELD (1949)) must
be satisfied at the boundary surface Ω∝, located at infinity, in order to ensure
that all acoustic waves propagate freely towards infinity and that no
reflections occur at this boundary,
∂p( r )
lim r . ( + jkp( r )) = 0 . (1.8)
r →∞ ∂r
structural displacement field
As mentioned before, the elastic structure is assumed to be a shell-type

structure, which has a small thickness dimension. For this type of structures,
the displacement field is usually characterised in terms of the displacement
components of the shell middle surface, which may be described in a local
co-ordinate system (s1,s2,n), attached to the middle surface (see figure 1.3).
The co-ordinate directions s1 and s2 are located in the plane of the middle
surface, while n represents the direction normal to the middle surface with
positive orientation away from the fluid domain.
Figure 1.3 : shell co-ordinate system
The corresponding middle surface displacement components ws1, ws2 and wn

are governed by the following type of linear dynamic equations,
~
 ws 1 ( r )   f s 1 ( r )   0 
[ ] [ ]   ~   
( Ls − ω 2 M s ). w s 2 ( r ) =  fs 2 ( r ) +  0 , r ∈Ω s .

(1.9)
 w ( r )   ~f ( r )   p( r )
î n  î n  î 
[ ]

Ls is a (3x3) matrix of differential operators, governing the elastic and

damping forces in the shell structure and the elements of the (3x3) matrix
[ ]

M s represent the inertial parameters of the structure. The first term in the
10 Chapter 1
right-hand side of (1.9) represents the distribution of mechanically applied

forces (force per unit area) on the structure. The second term represents the
fluid pressure loading. Since the fluid has no viscosity, the fluid pressure
exerts only a force excitation in the normal direction. Note that, although
assumed in (1.9) for the sake of simplicity, the elastic structure needs not to
be completely in contact with the fluid.
The dynamic equations (1.9) describe the different wavetypes in the elastic
structure. With s being the number of wavetypes, the structural displacement
field is uniquely defined, if s boundary conditions are specified at each point
of the boundary Γs of surface Ωs. A general formulation of these boundary
conditions is
 ws 1 ( r ) 
[ ]  
Ls,b . ws 2 ( r ) = {g s ( r )}, r ∈Γ s ,

(1.10)
 w (r ) 
î n 
[ Ls,b ]

where is a (sx3) matrix of differential operators, governing the

appropriate boundary conditions and {g s (r)} is a (sx1) vector of corres-
ponding prescribed functions.
In the above formulation, two basic types of coupled vibro-acoustic systems

(interior and exterior) have been considered. In both types, only one side of
the elastic structure is in contact with a fluid. Vibro-acoustic systems, in
which both sides of the elastic structure are in contact with a fluid, may be
regarded as special cases of the two basic types.
When an elastic structure, of which both sides are in contact with a fluid, is
part of a closed boundary surface Ωa, there are two separate acoustic
domains: a bounded domain, interior to Ωa, and an unbounded domain,
exterior to Ωa. For each of these acoustic domains, a Helmholtz equation of
type (1.2) and boundary conditions of type (1.4)-(1.8) may be formulated.
The structural displacement field is governed by the dynamic equations of
type (1.9), in which the pressure loading term in the right hand side is no
longer the one-sided fluid pressure but the pressure difference between both
sides of the structure. The structural boundary conditions are still formulated
in the form of (1.10).
When the elastic structure, of which both sides are in contact with a fluid, is
part of an open boundary surface Ωa, there is only one unbounded acoustic
domain. This type of vibro-acoustic systems is a special type of exterior

system. As indicated in figure 1.4, both sides of the open boundary surface
(solid line) may be regarded as two separate boundary surfaces, which form
one closed boundary surface (dashed line). The dynamic equations and
boundary conditions are similar to those of a regular exterior system. Only
the pressure loading term in the right-hand side of the structural dynamic
equation (1.9) becomes the pressure difference between both sides of the
structure.
Figure 1.4
1.3. Existing numerical prediction techniques

For most coupled vibro-acoustic problems, the solution of the partial
differential equations (1.2) and (1.9) in combination with the boundary
conditions (1.4)-(1.8) and (1.10) cannot be found in a closed analytical
form, due to the complexity of the geometry and the boundary conditions.
Therefore, an approximation of the exact solution is looked for by
transforming the mathematical model into a set of approximating (algebraic)
equations, which are amenable to numerical solution procedures.
This section doesn’t aim at giving an exhaustive list of different numerical
methods that can provide an approximate solution for coupled vibro-
acoustic problems, but focuses on the two most commonly used numerical
methods in acoustics and structural dynamics: the finite element method
(FEM) and the boundary element method (BEM).
In view of a comparison with the newly developed wave based prediction
technique later on in the dissertation, subsection 1.3.1. outlines the basic
concepts and properties of the existing finite element and boundary element
method. Subsections 1.3.2. and 1.3.3. describe how a finite element model
of the structural part of a coupled vibro-acoustic problem can be combined
with respectively a finite element and a boundary element model of the
acoustic part. Both types of coupled models are compared in 1.3.4. and the
main limitations of these coupled models are indicated in subsection 1.3.5.
12 Chapter 1
1.3.1. Basic concepts and properties of FEM and BEM
1.3.1.1. FEM
The finite element method is the most commonly used numerical prediction
technique for solving engineering problems, which consist of finding the
distribution of one (or several) field variable(s) in a continuum domain,
governed by an appropriate (set of) partial differential equation(s) and
boundary conditions.
The finite element method is based on two concepts:
• transformation of the original problem into an equivalent integral
formulation (weighted residual or variational),
• approximation of the field variable distributions and the geometry of the
continuum domain in terms of a set of shape functions, which are locally
defined within small subdomains (‘finite elements’) of the continuum
domain.
These concepts and the subsequent properties of the finite element method,
together with its application for uncoupled vibro-acoustic problems, are
briefly discussed below. For a detailed discussion, the reader is referred to
e.g. ZIENKIEWICZ (1977), BATHE (1982) and HUGHES (1987).
weighted residual approach

The substitution of any set of functions, proposed as solution for the field
variables within the continuum domain and on the boundary surface, into
the mathematical formulation of the continuum problem yields some error
functions (‘residuals’) on both the partial differential equations and the
boundary conditions. The equivalent weighted residual formulation of a
continuum problem defines the exact field variable solutions as those
2
functions, whose residuals are orthogonal with respect to any weighting
function.
In the weighted residual method, an expansion in terms of some prescribed
trial or shape functions is proposed as solution for each field variable. In this
way, the problem of determining the distribution of the field variables in a
continuum domain is converted into a discrete problem of determining the
contributions of each shape function to the solution expansion. Since the
shape functions span only a limited function space, the weighted residual
method usually provides only an approximation of the exact solution. The
2
provided that the weighting function is bounded and uniquely defined within the
continuum domain and on the boundary surface
contributions of the shape functions are obtained from an alleviated form of

the weighted residual formulation, in that the resulting residuals of the
proposed field variable expansions are orthogonalized with respect to a set
of weighting functions, equal in number to the number of shape functions.
The Galerkin weighted residual method, in which the prescribed shape
functions are also used as weighting functions, is a commonly used selection
process, since it leads, for several - but not all - continuum problems, to a
symmetric matrix equation in the unknown shape function contributions.
The finite element method can be considered as a special case of the
(Galerkin) weighted residual method in that the concept is applied using
shape functions, that are only locally defined. The continuum domain V is
discretized into a number of small subdomains Ve (‘finite elements’) and
nodes are defined at some particular locations in each element. Within each
element domain Ve, each field variable φ is approximated as a solution
expansion φ in terms of a set of prescribed shape functions Ni(φ ) (i=1..nφ),

that are only defined within the considered element domain,
nφ
(φ )
φ (r ) ≈ φ (r ) = ∑ Ni r ∈Ve .

( r ). ai , (1.11)
i=1
For each node, one or more degrees of freedom are specified, such that the
total number of nodal degrees of freedom equals nφ, the total number of
shape functions. Each specified degree of freedom is usually related to the
field variable φ and its spatial derivatives. Each shape function Ni(φ ) is
defined such that it has a unit value for a degree of freedom, specified at one
of the element nodes, and that is zero for all other nodal degrees of freedom.
In this way, each contribution ai to the solution expansion (1.11) represents
the value of a degree of freedom, specified at one of the element nodes.
variational approach
Several continuum problems may also be reformulated as an equivalent
variational problem. In the latter, the exact distributions of the field
variables are defined as those functions that make a functional stationary
with respect to small changes to these functions. The functional is a scalar
quantity that results from integrating a function over the continuum domain
and from integrating another function over the boundary surface. Both
14 Chapter 1
integration functions are obtained by applying some problem dependent

operators on the field variable distributions and their spatial derivatives3.
In the Rayleigh-Ritz method, the solution for each field variable is sought as
an expansion in terms of some prescribed trial or shape functions. The
contributions of the shape functions to the solution expansion are
determined such that the functional of the considered continuum problem,
evaluated using the proposed solution, is stationary, i.e. the functional value
remains unaltered for each infinitesimal variation on one of the shape
function contributions. This method yields always a symmetric matrix
equation in the unknown shape function contributions and corresponds with
the Galerkin weighted residual method.
The finite element method can be considered as a special case of the
Rayleigh-Ritz method in that the solution for each field variable is sought in
terms of shape functions, that are locally defined within each element (see
(1.11)).
convergence
The prediction accuracy, obtained from a finite element model, depends
mainly on the number of elements and on the nature of the prescribed shape
functions. In this framework, it is important that the accuracy systematically
improves as the number of elements increases, so that the proposed solution
expansion eventually converges towards the exact solution. To ensure this
convergence, some conditions must be satisfied by the shape functions and,
for the case of a weighted residual formulation, also by the weighting
functions.
A necessary condition for the solution expansion to be convergent is the
condition of completeness. Whether a finite element model is based upon a
variational or a weighted residual formulation, the nodal degrees of freedom
result always from a set of equations, in which the coefficients are defined
in integral forms. All terms in the integration functions comprise the field
variables and their higher-order spatial derivatives. Each of these integration
terms, evaluated using the exact field variable distributions, reaches a
constant value within any infinitesimally small region in the continuum
domain. Therefore, a necessary convergence condition upon the element
shape functions is that each of the integration terms, evaluated using the
approximating field variable expansions of type (1.11), can also reach a
constant value within each element. In a similar way, a completeness
3
the functional has often a physical meaning (e.g. the kinetic and potential energy
in the considered system and the work done by non-conservative forces)
condition is imposed upon the weighting functions for the case of a

weighted residual formulation. Since the order of the highest weighting
function derivative is usually smaller than those of the field variables, the
completeness condition upon the shape functions is more severe than the
one upon the weighting functions. Therefore, a weighted residual
formulation is usually transformed into its weak form by using integration
by parts. This reduces, on the one hand, the order of the highest-order field
variable derivatives in the integration functions, which results in a relaxation
of the completeness condition upon the shape functions. On the other hand,
this order increases for the weighting functions, which usually results in the
same completeness condition upon the weighting functions and the shape
functions.
A second convergence consideration concerns the element compatibility. In
order to be able to evaluate the various integrals in a variational or weighted
residual formulation, shape functions (and weighting functions), which
result in terms in the integrals becoming infinite, should be avoided. To
achieve this, the elements should be compatible in that, for each field
variable, the derivatives of order s, being the highest-order field variable
derivatives in the integral form, exist for the proposed field variable
expansion within each element and that its derivatives of order s-1 are
continuous along the interelement boundaries. In a similar way, a
compatibility condition applies for the weighting functions for the case of a
weighted residual formulation. Elements, that yield a solution expansion
with continuous derivatives of order s-1 but discontinuous higher-order
derivatives along the interelement boundaries, are denoted as elements with
Cs-1-continuity.
For a finite element discretization that consists of conforming elements, i.e.
elements that satisfy the completeness and compatibility conditions, the
solution expansion converges monotonically to the exact solution. For a
finite element discretization that consists of nonconforming elements, i.e.
elements that satisfy the completeness condition but not the compatibility
condition, its - not necessarily monotonic - convergence can still be ensured,
provided that the discretization passes a ‘patch test’, which is basically a
completeness test on an assemblage of incompatible elements.
geometrical description
Since the differentiation and integration of a polynomial function are very
easy, the construction of a finite element model becomes simple and
straightforward, if the element shape functions are obtained from a set of
polynomial functions. For this type of finite element discretizations, some
16 Chapter 1
specific requirements may be derived from the above mentioned

convergence considerations. If the highest-order spatial derivatives of a field
variable in the variational or weighted residual formulation are of order s,
the necessary condition of completeness is satisfied, if the set of
polynomials functions, used for the construction of the shape functions for
this field variable, contains at least all terms up to order s. Moreover, to get
a discretization with Cs-1-continuous elements or at least a discretization, for
which the convergence is ensured, the faces of the element boundary
surfaces must usually be located in ‘planes’, having a constant value for one
of the Cartesian co-ordinates or one of the normalized area or volume co-
ordinates. As a result, a convergent discretization with polynomial shape
functions is usually confined to rectangular and triangular elements for two-
dimensional problems and rectangular prism and tetrahedral elements for
three-dimensional problems. For most continuum problems, however, the
continuum domain has a geometrically complex shape, which cannot be
discretized exactly into an assemblage of these geometrically simple
elements. Consequently, a finite element discretization for such problems
induces not only an approximation error on the field variable distributions,
but also on the geometrical description. To keep the geometrical
discretization error within acceptable levels, a discretization is often
constructed using the concept of parametric mapping, which enables the use
of elements with more complex, distorted geometries.
The above mentioned geometrically simple elements, denoted as ‘parent’
elements, constitute the starting point of the mapping concept. In these
parent elements, for which convergence is ensured, the field variables are
described in terms of polynomial shape functions, which are defined in a
local element co-ordinate system4 (Cartesian, normalized area or normalized
volume co-ordinate system). By distorting such a parent element, together
with its local element co-ordinate system, an element with a distorted
geometry and a local, now curvilinear, element co-ordinate system is
obtained. In a similar way as for the field variable description, the geometry
of the distorted element, i.e. the relationship between the global Cartesian
and the local curvilinear co-ordinates of each point in the distorted element
domain, is described as an expansion of some prescribed shape functions.
4
with respect to the convergence of a finite element discretization, it is desirable
that the element shape functions are spatially isotropic or geometrically invariant,
which means that the proposed expansions for the field variables within each
element are independent of the orientation of the local element co-ordinate
system
For some points in the element5, their desired global co-ordinates are
specified. By defining each shape function as a (polynomial) function in the
local co-ordinate system with a value of unity at one of the specified points
and zero at the others, each shape function contribution in the geometrical
expansion corresponds with the desired global Cartesian co-ordinates of one
of the specified points. Elements with the same shape functions describing
their geometry and field variable distributions are commonly used and
denoted as isoparametric elements. Provided that some limitations on the
amount of distortion are not violated, the parametric mapping concept yields
a one-to-one correspondence between the global Cartesian and the local
curvilinear co-ordinate systems. Based on this co-ordinate transformation,
the integration functions in the variational or weighted residual formulation
may be transformed from the global Cartesian to the local curvilinear co-
ordinates. Due to the simple geometry of the parent elements, the integration
limits of the resulting integrals, which could be complicated when defined in
the global Cartesian co-ordinates, become constant in the local co-ordinates.
Although not automatically ensured, the convergence properties of the
parent discretization are usually preserved in the mapped discretization,
especially for isoparametric discretizations. An illustration of this mapping
concept for defining distorted fluid elements is given in appendix A.
application for uncoupled vibro-acoustic problems

As already mentioned in section 1.1.3., the numerical prediction of the
dynamic behaviour of uncoupled vibro-acoustic systems may be performed
in two steps, each of which involves either an uncoupled structural or an
uncoupled acoustic problem. The main features of the application of the
finite element method for uncoupled structural and uncoupled interior
acoustic problems are briefly presented below. A detailed discussion is
given in appendices A and B.
As already mentioned in section 1.2., the steady-state dynamic response of

an uncoupled thin elastic shell structure may be characterised by the
distribution of three field variables, i.e. the displacement components of the
shell middle surface in the two orthogonal directions, located in the plane of
the middle surface, and in the direction, normal to the middle surface. In a
finite element discretization, it is common practice to model a curved shell
structure as an assemblage of small, flat plate sections.
5
the specified points usually coincide with (some of) the nodes of the element and
their number equals the number of prescribed shape functions
18 Chapter 1
According to Kirchhoff’s thin plate theory, the normal displacement

component in a flat plate is independent from the in-plane displacement
components. For a finite element approximation of these components to be
convergent, it is sufficient that, on the one hand, a polynomial expansion is
used for the normal displacement component, which is complete up to the
second order within each parent element and which ensures C1-continuity
along the interelement boundaries and that, on the other hand, polynomial
expansions are used for the in-plane displacement components, which are
complete up to the first order within each parent element and which ensures
C0-continuity along the interelement boundaries.
The most commonly used elements are rectangular6 and triangular plate
elements. The polynomial expansions for the in-plane displacement
components within these elements are based on the two nodal in-plane
translational degrees of freedom, defined at each corner point of the
element. The polynomial expansion for the normal displacement component
within the elements is usually based on the normal translational degree of
freedom and the two in-plane rotational degrees of freedom, defined at each
1
corner point node. Although the latter expansion violates the C -continuity
condition, the convergence of a discretization with these plate elements is
usually ensured. In view of assembling the different plate elements into a
shell structure, the element shape functions and the subsequent equations in
the unknown nodal degrees of freedom must be transformed from the local
element co-ordinate systems to a global Cartesian co-ordinate system. To
facilitate this transformation, the nodal rotational degrees of freedom around
the co-ordinate axis, normal to the element surfaces, are usually added to the
set of unknown degrees of freedom, although they do not contribute to the
field variable expansions.
The resulting equations in the unknown nodal translational and rotational
degrees of freedom ai may be written in terms of a matrix equation7,
([ K ] + jω [ C] − ω 2 [ M ]). {ai } = {Fi } . (1.12)
The back-substitution of the solution vector {ai } into the expansions for the
6
or their mapped quadrilateral plate elements
7
when all plate elements, meeting at a certain node, are co-planar and have the
additional normal rotational degrees of freedom, some dedicated numerical
treatment (see e.g. ZIENKIEWICZ and TAYLOR (1991)) must be introduced to
circumvent the singularity of this matrix equation
normal and the in-plane displacement components yields the approximation

of the steady-state dynamic response of a thin elastic shell structure. The
matrices [ K ] , [ C ] and [ M ] are the stiffness, damping and mass matrix of
the shell structure. The prescribed translations and rotations, imposed on
(part of) the boundary, are a priori assigned to the corresponding boundary
degrees of freedom. Hence, the latter are not comprised in the vector {ai } .
The excitation vector {Fi } comprises the contributions from these a priori
assigned nodal degrees of freedom and from the external forces and
moments, applied on the shell structure and along (parts of) its boundary.
The steady-state dynamic response of an uncoupled interior acoustic system

may be characterised by the distribution of one field variable, i.e. the
acoustic pressure distribution, which is governed by the Helmholtz equation
(1.2) and boundary conditions (1.4)-(1.6). For a finite element
approximation of the pressure to be convergent, it is sufficient that a
polynomial pressure expansion is used, which is complete up to the first
order within each parent element and which ensures C0-continuity along the
interelement boundaries. Linear tetrahedral and linear hexahedral fluid
elements are commonly used elements. The polynomial expansion for the
pressure within their parent elements is based on the pressure degree of
freedom, defined at each corner point node, and satisfies both convergence
conditions.
The resulting equations in the unknown nodal pressure degrees of freedom
are written in terms of a matrix equation of the same form as (1.12). To
preserve the analogy with a structural finite element model, the matrices
[ K ] and [ M ] in an acoustic finite element model are called the acoustic
stiffness and acoustic mass matrix. The ‘stiffness’ matrix [ K ] is, however,
an inverse mass or mobility matrix, relating the pressure to an acceleration.
The ‘mass’ matrix [ M ] is a compressibility matrix, relating the pressure to
a displacement. Matrix [ C ] in an acoustic finite element model is the
acoustic damping matrix, which results from the impedance boundary
conditions (1.6). The prescribed pressure values in the pressure boundary
condition (1.4) are a priori assigned to the corresponding boundary degrees
of freedom. Hence, the latter are not comprised in the vector {ai } . The
acoustic excitation vector {Fi } comprises the contributions from the a priori
assigned nodal degrees of freedom and from the external acoustic sources
and the imposed velocity boundary conditions (1.5).
20 Chapter 1
properties
The application of the finite element method for dynamic structural and
acoustic problems offers several advantageous properties.
• Since the shape functions are only locally defined within an element,
[ K ] , [ C] and [ M ] are sparsely populated, banded matrices. Moreover, for
uncoupled structural and uncoupled acoustic problems, which can be cast
into a variational formulation, the resulting matrices are symmetric. Due
to these matrix properties, the solution vector of unknown nodal values
can be obtained from efficient matrix solvers.
• Since the shape functions are independent of frequency, the stiffness
matrix [ K ] and the mass matrix [ M ] have frequency independent
coefficients. This allows the use of standard eigenvalue solvers for the
calculation of the undamped natural frequencies and mode shapes of
structural and acoustic system.
• Since the shape functions have real values, the coefficients of the
stiffness matrix [ K ] and the mass matrix [ M ] are real. The coefficients
of the acoustic damping matrix [ C ] are usually complex, since the
prescribed impedance Z in (1.6) is usually complex.
• The numerical calculation of the matrix coefficients is very easy and
straightforward. The involved numerical integrations are usually
performed with a simple Gauss quadrature rule with only a few
integration points per element (see e.g. ZIENKIEWICZ (1977)).
The modelling concepts, especially the use of low-order polynomial shape

functions, induce also some disadvantageous properties of the finite element
method.
• The dynamic response of a structural or acoustic system results from a
complex mechanism of wave propagation in the structural or acoustic
continuum domain. The resulting spatial variation of the oscillatory
dynamic response is mainly determined by the spatial distribution of the
external excitation and its frequency contents (LANGLEY (1990)). The
latter is caused by the fact that the wavelengths of structural and acoustic
waves depend on the frequency.
Since the (low-order) polynomial functions, used for the description of
the dynamic field variables, can only represent a restricted spatial
variation, a large number of elements is required to accurately represent
the oscillatory wave nature of the dynamic response. To obtain an
acceptable level of prediction accuracy for structural displacements or
acoustic pressures, a general rule of thumb states that at least 10

elements per structural or acoustic wavelength are required (see e.g.
THOMPSON and PINSKY (1995)). Hence, according to this rule of thumb,
the number of elements should increase as the frequency increases, since
structural and acoustic wavelengths decrease as the frequency increases.
Therefore, the larger the continuum domain and the higher the excitation
frequency, the larger becomes the model size. An illustration of the
dependence of the prediction accuracy on the number of elements is
given in appendix A.
As a result, the application of the finite element method for real-life
engineering problems usually involves large model sizes and requires a
large amount of memory and computational efforts for constructing and
solving the model8. LE SALVER (1998) indicated, for example, the large
amount of computer resources, involved with the use of finite element
models in the acoustic design process of a car body. The calculation of
the transfer function in a frequency range up to 200 Hz between the
force, applied at one attachment position of the powertrain unit to the car
body, and the acoustic pressure at the driver’s ear position in a passenger
car, required 10000 seconds of CPU time and 200 megabytes of memory
on a C90 CRAY computer. Bearing in mind the iterative nature of a
design process, these resources are already at the limit of acceptable
computational cost, although the finite element models did not yet
include the trim panel damping treatment of the car body, nor the vibro-
acoustic coupling effects between the car body and the acoustic cavity.
Due to these large model sizes, model size reduction techniques are often
applied, as will be discussed in section 1.3.5.
• The prediction accuracy for derived secondary field variables is smaller
than for primary field variables. Predictions of dynamic quantities such
as structural stresses and strains or fluid velocities are obtained by
derivating the prediction results of the primary field variables, i.e. the
structural displacement components or the acoustic pressure. Since the
latter are usually expressed in terms of polynomial expansions, which are
complete up to a certain order, the derived variables are expressed in
terms of polynomial expansions, which are only complete up to a lower
order. As a result, these lower-order polynomial expansions represent a
8
due to the advantageous matrix properties of a finite element model, the
construction of the model takes only a minor part of the computational effort;
most of the effort is spent on solving the model for the unknown nodal degrees of
freedom
22 Chapter 1
smaller spatial variation of the derived secondary variables than the

expansions of the corresponding primary variables.
In a linear hexahedral fluid element, for example, the pressure is expressed
in terms of a polynomial expansion, which is complete up to the first
order, while the fluid velocity vector, which is proportional to the spatial
gradient of the pressure (see (1.3)), is no longer expressed in terms of a
complete linear expansion. However, there is no physical justification for
the subsequent smaller spatial variation of the fluid velocity predictions,
since the pressure and fluid velocity of an acoustic wave have the same
spatial variation. Consequently, the prediction accuracy of the fluid
velocity is smaller than the accuracy of the pressure for a given finite
element discretization. Moreover, for low-order polynomial expansions
of the primary field variables, the subsequent expansions for the derived
secondary variables become often discontinuous at the interelement
boundaries. This is the case, for example, for the fluid velocity in linear
hexahedral fluid elements.
An illustration of the accuracy decrease for derived secondary variables is
given in appendix A.
• Since the element domains in a finite element discretization have a finite
volume and since only models of finite size are amenable to numerical
solution algorithms, the finite element method is, in principle, restricted
to engineering problems, which are defined in bounded continuum
domains. However, the modelling concepts of the finite element method
can still be used for solving problems with unbounded domains, as will
be discussed in section 1.3.2.2.
1.3.1.2. BEM
Over the last two decades, the boundary element method has become a
valuable modelling alternative for several types of engineering problems,
especially for problems involving unbounded domains.
The boundary element method is based on the direct or indirect boundary
integral formulation of the considered problem. These formulations relate
the distributions of the field variables in the continuum domain to the
distribution of some problem-related boundary variables on the boundary
surface of the domain. In this way, the boundary element method follows a
two-step procedure. In the first step, the distributions of the boundary
variables are determined. In the second step, the field variables in any point
in the continuum domain are obtained from the boundary integral
formulation, using the boundary surface results of the first step.
Similar to the finite element method, the boundary element method is based
on two concepts for solving the first-step boundary problem:
• transformation of the boundary problem into a collocational or varia-
tional formulation;
• approximation of the boundary surface geometry and the boundary
variables in terms of a set of shape functions, which are locally defined
within small subsurfaces (‘boundary elements’) of the boundary surface.
These concepts and the subsequent properties of the boundary element
method, when applied for acoustic problems, are briefly discussed below.
For a detailed discussion of the boundary element method in general, the
reader is referred to e.g. BANERJEE and BUTTERFIELD (1981), BREBBIA et al.
(1984), CISKOWSKI and BREBBIA (1991). A detailed discussion of the
application of the boundary element method for acoustic problems is given
in appendix C.
direct and indirect boundary integral formulations for acoustic problems

The direct boundary integral formulation (SCHENCK (1968)) relates the
9
pressure in any point of an acoustic field, that satisfies the homogeneous
Helmholtz equation (1.2) and, for exterior acoustic problems, the
Sommerfeld radiation condition (1.8), to the pressure and normal velocity
distribution on the boundary surface of the acoustic domain. For exterior
problems, the pressure p(r) in any point of an unbounded acoustic domain V
(see figure 1.2) is expressed as
∂G( r , ra )
C( r ). p( r ) = ∫ [ p( ra ). + jρ 0ωvn ( ra ). G( r , ra )]. dΩ ( ra ) , (1.13)
Ωa
∂n
where p( ra ) and vn ( ra ) are the pressure and normal velocity distribution

on the boundary surface Ωa. The direction n, normal to the boundary surface
Ωa, has a positive orientation into the acoustic domain V. The function
G( r , ra ) is the Green’s kernel function, which satisfies the inhomogeneous
Helmholtz equation
9
when external acoustic sources are applied to the fluid, the total pressure field,
governed by an inhomogeneous Helmholtz equation, is considered as the
superposition of the free-field pressure, which is the response to the acoustic
sources in absence of the actual boundary surface, and the scattered field, which
is governed by the corresponding homogeneous Helmholtz equation
24 Chapter 1
∇ 2 G( r , ra ) + k 2 G( r , ra ) = −δ ( r − ra ) (1.14)
and the Sommerfeld radiation condition, as defined in (1.8). Hence, the

Green’s kernel function represents the free-field pressure in any point r due
to an acoustic point source in ra. For three-dimensional problems, the
Green’s kernel function is
− jk r − r
a
e
G( r , ra ) = (1.15)
4π r − ra
For two-dimensional problems, it is
j
G( r , ra ) = − . H 0( 2) ( k r − ra ) , (1.16)
4
where H 0(2) is the zero-order Hankel function of the second kind.

For positions r, located in the unbounded domain V, the coefficient C(r) in
(1.13) is 1 and for positions r in the interior domain, enclosed by the
boundary surface Ωa, the coefficient C(r) is 0. For positions r on the
boundary surface Ωa, the coefficient C(r) represents the exterior solid angle,
expressed as a fraction of 4π, of the boundary surface Ωa at the surface
position r. At surface positions where the normal direction is uniquely
defined, this coefficient equals 1/2.
The term ‘direct’ indicates that the boundary variables, i.e. the pressure and
normal velocity distributions on the boundary surface, have a direct physical
meaning.
The direct boundary integral formulation for interior acoustic problems is
similar to the exterior formulation (1.13). Since the direct boundary integral
formulation requires that the boundary surface Ωa is closed, it can only
represent either an interior or an exterior pressure field, but not a combined
interior/exterior field.
Acoustic problems with an open boundary surface Ωa or combined

interior/exterior acoustic problems, in which an acoustic domain exists on
both sides of a closed boundary surface Ωa, can be reformulated in terms of
the indirect boundary integral formulation (FILIPPI (1977)). This
formulation relates the pressure in any point of an acoustic field, that
satisfies the homogeneous Helmholtz equation (1.2) and the Sommerfeld

radiation condition (1.8), to the distributions of a single layer potential
σ ( ra ) and a double layer potential µ ( ra ) on the boundary surface Ωa,
∂G( r , ra )
p( r ) = ∫ [ µ ( ra ). ∂n
− σ ( ra ). G( r , ra )]. dΩ ( ra ), ( r ∈V \ Ω a ).
Ωa
(1.17)
The single layer potential is the difference in normal pressure gradient
between both sides of the boundary surface Ωa,
∂p( ra+ ) ∂p( ra− )

σ ( ra ) = − . (1.18)
∂n ∂n
This potential function can be regarded as a distribution of monopole

sources on the boundary surface. The double layer potential is the pressure
difference between both sides of the boundary surface Ωa,
µ ( ra ) = p( ra+ ) − p( ra− ) (1.19)
and represents a distribution of dipole sources on the boundary surface. The

positions ra+ and ra− indicate the boundary surface positions at the positive
and negative side of the normal direction n. The function G( r , ra ) is the
same Green’s kernel function as used in the direct integral formulation (see
(1.15) and (1.16)).
direct boundary element method

The direct boundary element method follows a two-step procedure, based on
the direct boundary integral formulation (1.13) for interior or exterior
acoustic problems with a closed boundary surface. In the first step, the
pressure and normal velocity distributions on the boundary surface are
determined. In the second, post-processing step, the pressure in any point of
the interior or exterior acoustic domain is obtained from the direct boundary
integral formulation, using the surface results from the first step.
The determination of the two boundary variables, i.e. the pressure and the
normal velocity distributions on the boundary surface, is based on the same
modelling concepts, as used in the finite element method. The boundary
surface is discretized into a number of small subsurfaces (‘boundary
26 Chapter 1
elements’) and some nodes are defined at particular locations in each

element. In a similar way as defined in (1.11) for the finite element method,
the solutions for the two boundary variables within each boundary element
are approximated in terms of a set of prescribed shape functions. In this
way, the problem of determining the boundary variable distributions is
turned into a discrete problem of finding the shape function contributions,
which represent the pressure values pi and normal velocity values vni at the
discrete node locations on the boundary surface.
To determine the unknown nodal degrees of freedom, the direct boundary
integral formulation (1.13) and the boundary conditions (1.4)-(1.6) are
transformed into a weighted residual or a variational formulation. Since
these formulations as well as the direct boundary integral formulation are
integral forms, this procedure involves the numerical evaluation of double
surface integrals. Since this is a computationally expensive procedure, the
nodal pressure and normal velocity values are usually obtained from a
collocational formulation (KOOPMANN and BENNER (1982), SEYBERT et al.
(1985)). In the latter formulation, the direct boundary integral formulation
(1.13) is evaluated for each nodal position on the boundary surface, using
the discretized expressions for the surface pressure and normal velocity10.
When the boundary element discretization11 consists of na nodes, this yields
na algebraic equations in the 2na nodal values pi and vni. These equations may
be written in terms of a matrix equation,
[ A]. { pi } = jρ 0ω [ B].{vni } , (1.20)
in which the (naxna) matrices [ A] and [ B] relate the nodal pressure values to
the nodal normal velocity values. Since the boundary conditions (1.4)-(1.6)
impose a prescribed pressure, a prescribed normal velocity or an impedance
relation between both boundary variables, at each node either pi or vni or
their relation is known a priori. The remaining na unknown nodal values
result from the direct boundary element model (1.20). A detailed description
of this model is given in appendix C.
10
the collocational formulation may be regarded as a special case of the weighted
residual formulation, in which Dirac delta functions, located at the boundary
nodes, are used as weighting functions
11
the pressure and normal velocity are usually approximated in terms of the same
shape functions, yielding two degrees of freedom per boundary element node
indirect boundary element method

The indirect boundary element method follows a two-step procedure, based
on the indirect boundary integral formulation (1.17) for acoustic problems
with an open boundary surface or combined interior/exterior acoustic
problems with a closed boundary surface. In the first step, the single and
double layer potentials on the boundary surface are determined. In the
second, post-processing step, the pressure in any point of the acoustic
domain is obtained from the indirect boundary integral formulation, using
the surface results from the first step.
In exactly the same way as in the direct boundary element method, the
solutions for the two boundary variables, i.e. the single and double layer
potentials, within each element in the discretized boundary surface are
approximated in terms of a set of prescribed shape functions. The shape
function contributions to this solutions represent the single layer potentials
σi and double layer potentials µi at the discrete node locations on the
boundary surface.
For the determination of the unknown nodal degrees of freedom, a
collocational approach, as selected in the direct boundary element method,
becomes much more difficult, since it involves the numerical evaluation of
divergent integrals. Therefore, the single and double layer potentials on the
boundary surface are usually obtained from a variational formulation of the
indirect boundary integral formulation (1.17) and the boundary conditions
(1.4)-(1.6), which induces, however, the numerical evaluation of double
surface integrals. Imposing the stationarity condition on the associated
functional, using the discretized expressions for the single and double layer
potentials, yields a symmetric matrix equation for the unknown nodal
values,
~
 E D σ i   fσ 
 DT . =
  ~  . (1.21)
 F  î µ i   f µ 
î 
As in the direct boundary element method, only one boundary value per
node is comprised in the vector of unknowns in (1.21), since the boundary
conditions provide some a priori relations. At the part of the boundary
surface, on which a prescribed pressure is imposed, the double layer
potential is zero. At the part of the boundary surface, on which a prescribed
normal velocity is imposed, the single layer potential is zero. At the part of
the boundary surface, on which a prescribed impedance is imposed, a
relation exists between the single and the double layer potential.
28 Chapter 1
The boundary conditions appear also in an explicit way in the indirect

boundary element model (1.21). The imposed pressure and imposed normal
velocity boundary conditions appear in the vectors fσ
~
{ } ~
and f µ , { }
respectively, while the imposed impedance boundary conditions appear in
the matrices [ D] and [ F ] . A detailed mathematical description of the model
is given by PIERCE and WU (1983).
convergence and geometrical description

In the same way as in the finite element method, the boundary element
discretization of the boundary surface is based on the parametric mapping of
a geometrically simple parent element discretization, in which the boundary
variables are expressed in terms of polynomial shape functions and for
which the convergence towards the exact solution is ensured.
The collocational formulation, used in the direct boundary element method,
doesn’t comprise any derivatives of the boundary surface pressure and
normal velocity. Hence, the convergence is already ensured for parent
discretizations with constant values of the pressure and normal velocity
within each element. The variational formulation, used in the indirect
boundary element method, comprises first-order derivatives of the double
layer potential, but no derivatives of the single layer potential. Hence, parent
elements with a constant single layer potential and a linear expansion for the
double layer potential meet the minimum requirements for convergence.
The most commonly used parent elements in both the direct and indirect
boundary element method are linear triangular and linear rectangular surface
elements. In these elements, the nodes are defined at each corner point of
the element and two degrees of freedom are allocated to each node. This
enables a linear expansion for both boundary variables. In this way,
convergence is definitely ensured, since the boundary variable expansions
exceed the minimum convergence requirements.
properties
The application of the direct and indirect boundary element method for
acoustic problems offers several advantageous properties, compared with
the finite element method.
• In the same way as the pressure field distribution, the distributions of the
derived secondary field variables, such as the fluid velocity, result from
the boundary integral formulations. Since an analytical expression is
available for the Green’s kernel function, the differentiation of the
boundary integral formulations implies no accuracy loss for the velocity

predictions in comparison with the accuracy for the pressure predictions.
• Since only the boundary surface of the acoustic domain is discretized
into elements, the size of direct and indirect boundary element models is
substantially smaller than corresponding finite element models.
• The boundary element method can easily handle problems with
unbounded acoustic domains, since the Sommerfeld radiation condition
is inherently satisfied in the boundary integral formulations.
The modelling concepts of the boundary element method induce also some
disadvantageous properties, compared with the finite element method.
• The boundary integral formulations (1.13) and (1.17) relate the acoustic
quantities at a certain position on the boundary surface to the distribution
of these quantities on the entire boundary surface. As a result, the
matrices in the direct and indirect boundary element models (1.20) and
(1.21) are fully populated. Moreover, the matrices in direct boundary
element models are not symmetric. The computational benefits of the
symmetrical indirect boundary element models are largely counter-
balanced by the calculation of the matrix coefficients, since this involves
the numerical evaluation of double surface integrals, while the non-
symmetric direct boundary element models result from single surface
integrals.
• Since the Green’s kernel function is a complex function, the matrix
coefficients in boundary element models are complex.
• Since the Green’s kernel function depends on the frequency dependent
wavenumber k, the matrix coefficients in boundary element models are
frequency dependent. As a result, a boundary element model does not
naturally lead to an algebraic eigenvalue problem for the extraction of
the natural frequencies and mode shapes of interior acoustic systems.
When natural frequencies are computed, it is usually assumed that the
entire boundary surface is rigid. In this case, the right-hand side of the
direct boundary element model (1.20) becomes identically zero and the
natural frequencies are those frequencies, for which the determinant of
matrix [ A] is zero. Due to the implicit frequency dependence of this
matrix, the search for the natural frequencies is a non-linear problem,
which can be solved using a secant method (ADEYE et al. (1985)). This
procedure is, however, computationally inefficient and prone to failure in
the case of closely spaced natural frequencies. A more advanced
approach is based on the decomposition of the frequency dependent
30 Chapter 1
Helmholtz problem into two subproblems: one consists of a

homogeneous, frequency independent Laplace problem and the other is a
Laplace problem, in which the frequency dependence of the Helmholtz
problem is incorporated as an inhomogeneous right-hand side excitation.
This decomposition leads to an algebraic non-symmetric eigenvalue
problem for the natural frequencies and mode shapes of interior acoustic
systems. For a detailed discussion of this approach, the reader is referred
to COYETTE and FYFE (1990) and ALI et al. (1995).
• A mathematical problem of boundary integral formulations is the non-
uniqueness of the solution for exterior problems with closed boundary
surfaces, when the excitation frequency coincides with an
eigenfrequency of an associated interior problem. Several techniques are
available to circumvent this problem. SCHENCK (1968) proposed to add
some additional equations that force the solution to satisfy the boundary
integral formulation at some points in the interior of the boundary
surface and to solve the resulting overdetermined matrix equation in a
least-squares sense. BURTON and MILLER (1971) proposed the use of a
combination of the original and a derivative form of the boundary
integral formulation, which ensures a unique solution at all frequencies.
• Since the Green’s kernel function and its derivatives exhibit a singularity
when the distance between positions r and ra becomes zero, special
attention must be paid to the numerical evaluation of the corresponding
matrix elements.
1.3.1.3. FEM versus BEM for acoustic problems
Based on the above mentioned properties of the finite element and the
boundary element method, two important features may be identified
regarding their computational loads for solving acoustic problems.
• Since acoustic finite element models are sparsely populated and
symmetric and since their matrix coefficients, which are real and
frequency independent for the stiffness and mass matrices, result from
simple numerical integrations, the construction of these models takes
only a minor part of the total computational effort. Most of the
computational effort is spent on solving the large models for the
unknown degrees of freedom.
• The computational effort for solving acoustic boundary element models
for the unknown boundary degrees of freedom benefits from the small
model sizes. However, the construction of these models requires a
substantial amount of computational effort, since the matrices in acoustic

boundary element models are fully populated, complex, frequency
dependent and, for the case of direct boundary element models, non-
symmetric and since singular integral evaluations are encountered in the
calculation of their matrix coefficients.
As a result, when comparing the total computational loads of both methods,
the boundary element method can hardly compete with the finite element
method for solving interior acoustic problems, as illustrated in appendix C.
Therefore, the finite element method is usually preferred for studying the
acoustic field in complex enclosures12. The strength of the boundary element
method becomes mainly apparent for solving acoustic problems in
unbounded domains, such as the exterior acoustic radiation or scattering
from vibrating structures.
1.3.2. Coupled FE/FE models
For coupled vibro-acoustic problems, the finite element method can be

applied by discretizing both the acoustic domain and the structural domain
into finite elements.
1.3.2.1. Interior problems

For interior coupled vibro-acoustic systems (see figure 1.1), the acoustic and
structural domains are suited for a finite element discretization, since both
domains are bounded. The coupled finite element implementations can be
classified into three major categories, based on the approach used in the
problem formulation : the Eulerian, the Lagrangian and the mixed approach.
• In an Eulerian formulation, the acoustic response is described by a single

scalar function, usually the pressure, while the structural response is
described by the displacement vector (ZIENKIEWICZ and NEWTON
(1969), CRAGGS (1973)). The resulting system equations in the unknown
nodal structural displacement components wi and nodal acoustic
pressures pi are
12
the acoustic field in, for instance, a car cavity is sometimes investigated with
boundary element models, despite their larger computational efforts; this choice
is mainly motivated by the savings in manhours, since a boundary element mesh
for a car cavity can be readily obtained from the existing finite element meshes of
the car body
32 Chapter 1
 Ks Kc  Cs 0 2  Ms 0  wi   Fsi 

 + ω − ω  .   =   . (1.22)
K a  0 Ca  − ρ K T M a  î pi  î Fai 
j
 0   0 c
[ Ks ] and [ K a ] , [ Cs ] and [ Ca ] , [ Ms ] and [ M a ] are the structural and

acoustic stiffness, damping and mass matrices. {Fsi } and {Fai } are the
external nodal structural and acoustic excitation vectors. [ K c ] represents
the fluid-structure coupling interaction. The coefficients in the coupled
stiffness matrix and coupled mass matrix are still frequency independent
but, in contrast with an uncoupled structural or uncoupled acoustic finite
element model, these coupled matrices are no longer symmetric. This is
due to the fact that the force loading of the fluid on the structure is
proportional to the pressure, resulting in a cross-coupling matrix [ K c ] in
the coupled stiffness matrix, while the force loading of the structure on
the fluid is proportional to the acceleration, resulting in a cross-coupling
matrix [− ρ0 KcT ] in the coupled mass matrix. A more detailed
description of the coupled model (1.22) is given in appendix D.
Due to the cross-coupling matrices, the banded, sparsely populated
nature of the matrices in an uncoupled finite element model partly
vanishes in a coupled finite element model. As a consequence, the
efficient equation solvers and eigenvalue solvers for sparse symmetric
matrices can no longer be used and have to be replaced by non-
symmetric solvers, which are more time expensive. For this reason, the
undamped natural frequencies and mode shapes at low frequencies are
often calculated by assuming an incompressible fluid ( [ Ma ] = [ 0] ),
which yields a simpler eigenvalue problem,
([ K ] − ω ([ M ] + [ M~ ])).{w } = {0} ,
s
2
s i (1.23)
~
[ ]
where M is referred to as the added mass matrix,
[ M~ ] = ρ 0 [ Kc ][ Ka ]−1[ Kc ]T . (1.24)
In an alternative Eulerian formulation (EVERSTINE (1981), OLSON and

BATHE (1985)), the fluid is described by the fluid velocity potential ϕ, of

which the gradient vector equals the fluid velocity vector v ,
j
∇p = ∇ϕ .

v=

(1.25)
ρ 0ω
This choice renders the symmetric model,
 Ks 0   Cs − ρ 0 Kc   Ms0  wi   Fsi 

  + jω   −ω 2  .   =  Fai 
 0 −ρ 0 Ka  − ρ 0 K c
T
− ρ 0 Ca  − ρ 0 Ma  î ϕ i  
 0
î jω 
(1.26)
For damped systems, this symmetric formulation is obtained without any
penalty. For undamped systems ( [ Cs ] = [Ca ] = [ 0] ), the use of the fluid
velocity potential yields a symmetric model, which becomes, however,
complex due to the introduction of an artificial damping matrix. This is
disadvantageous in terms of computational effort since solving a non-
symmetric but real model, obtained from an undamped pressure
formulation (see (1.22)), requires less arithmetical operations than
solving a symmetric but complex model, obtained from a velocity
potential formulation (see (1.26)).
• In a Lagrangian formulation, both structural and acoustic responses are

described by their displacement vector (FENG and KIEFLING (1976)).
Since both the force loading of the structure on the fluid as well as the
force loading of the fluid on the structure are proportional to the
structural/fluid acceleration, the coupled system matrices in a Lagrangian
formulation are symmetric. However, the advantage of symmetry can
hardly counterbalance the drawbacks of the formulation. A first
drawback is the substantial increase of the size of the coupled problem,
since the fluid displacement has three vectorcomponents instead of one
scalar value (pressure or fluid velocity potential) for each node of the
acoustic mesh. A second, and more severe, drawback is the existence of
spurious rotational modes. Under the assumption of an inviscid fluid, the
shear modulus of the fluid is zero and hence, the fluid motion is
irrotational. The Lagrangian formulation results from the application of
the variational principle to an energy functional. Since the shear modulus
34 Chapter 1
of an inviscid fluid is zero, the matrix D, linking the fluid stresses to the
fluid strains (σ=D.ε), is singular. This allows zero-energy rotational
deformation modes, which are not excluded from the Lagrangian model
since these spurious modes don’t contribute to the energy functional. In
order to avoid these physically unfeasible modes, an additional
irrotationality constraint of the fluid has to be imposed by a penalty
method (HAMDI et al. (1978), ZIENKIEWICZ and BETTESS (1978),
BELYTSCHKO (1980), OLSON and BATHE (1983)). This is done by adding
an additional rotational energy functional J(α), weighted by a penalty
factor α, to the energy functional,
1
α . J (α ) (u ) = α . ∫ (∇xu ) 2 dVa ,

(1.27)
2 V
a
where Va is the acoustic domain and u ( = v / (jω )) is the fluid displace-

ment vector. The resulting system equations are, in absence of any type
of damping,
 T K ~ ~ ~
 T  Ms 0   F
{ }
0
 [ L]  s ~  [ L ] − ω 2
(α )[ L ]  ~  [ L ] V = ~ 
s
 ~
K a + αKα  M a  
i
î Fa 
 0  0
wi 
with   = [ L ]{Vi } .
î ui 
(1.28)
Based on the coupling conditions at the fluid-structure interface, matrix
[ L] represents the static condensation of all nodal structural
displacement components wi and nodal fluid displacement components ui
in terms of the displacements Vi at the nodes which do not belong to the
[ ]
~ ~
[ ] [ ]
~
fluid-structure interface. K s and K a , M s and M a are the
~
[ ]
~
structural and acoustic stiffness and mass matrices. Matrix K a results [ ]
~
{ } ~
{ }
from the irrotationality constraint. Fs and Fa are the external nodal
structural and acoustic excitation vectors. Only a penalty factor α=∝
ensures an irrotational acoustic response. Since the system equations can
only be solved for a finite value of α, the response vectors and
eigensolutions depend on α and satisfy the irrotationality constraint only

in an approximate way.
To remove the deficiency of spurious rotational modes, BATHE et al.
(1995) have proposed an advanced Lagrangian formulation, using the
displacement, pressure and a ‘vorticity moment’ as acoustic variables.
The two latter variables are associated with element internal variables
and are statically condensed out at element level, so that only the nodal
fluid displacements occur in the global coupled model.
• In mixed formulations, the structural response is described by its

displacement vector and the acoustic response is described by the
pressure and the fluid displacement potential ψ, whose gradient is
proportional to the fluid displacement. The fluid-structure coupling
interaction can be introduced as cross-coupling matrices in the coupled
mass matrix (MORAND and OHAYON (1979)). A more advantageous
mixed formulation introduces the fluid-structure coupling interaction as
cross-coupling matrices in the coupled stiffness matrix (OHAYON (1985),
SANDBERG and GÖRANSSON (1988)). In absence of any type of damping,
the latter formulation is
 K 0 − McT   Ms 0 0  wi   Fsi 

 s     
T  2
 0 0 Ba  − ω 0 Ka 0 . ψ i  =  0  . (1.29)
  
 − Mc Ba − Ma   0 0 0 î pi  î Fai 

[ Mc ]results from the fluid-structure coupling interaction and [ Ba ]

relates the nodal fluid pressures to the nodal fluid displacement
potentials. The system equations are symmetric, no third artificial
damping matrix is introduced and no spurious rotational modes are
present. This mixed formulation has, however, two major drawbacks.
First of all, the size of the coupled problem is substantially larger,
compared to the Eulerian formulation (1.22), since there are two degrees
of freedom for each acoustic node. Secondly, special equation and
eigenvalue solvers are required, since the coupled stiffness matrix is
indefinite and the coupled mass matrix is positive semi-definite.
At present, a comparison of the advantages and drawbacks of the different

symmetric formulations with those of the non-symmetric Eulerian pressure
formulation denotes the latter as the most appropriate prediction technique.
36 Chapter 1
Therefore, most of the available finite elements programs (e.g. SYSNOISE,

MSC/NASTRAN, ANSYS) use the Eulerian pressure formulation for
coupled predictions of interior coupled vibro-acoustic systems.
1.3.2.2. Exterior problems

In exterior coupled vibro-acoustic systems (see figure 1.2), the acoustic
domain is unbounded. Since the element domains in a finite element
discretization are finite in extent and since only models of finite size are
amenable to numerical solution algorithms, the finite element method is, in
principle, restricted to problems with a bounded continuum domain. To
overcome this difficulty, an artificial boundary surface Ωe is introduced at
some finite distance from the boundary surface Ωa. Consequently, the
original acoustic domain is split into a bounded domain V1 between Ωa and
Ωe and an unbounded domain V2 between Ωe and Ω∝ (see figure 1.5).
Figure 1.5 : computational domains for exterior coupled FE/FE models
The bounded domain V1 is suited for a finite element discretization. To get a

well-posed acoustic problem in this domain, an appropriate boundary
condition must be specified at the artificial boundary surface Ωe. In this way,
an exterior problem is dealt with as an approximate interior problem for

which the coupled FE/FE models, described in 1.3.2.1., can be used.
To approximate the Sommerfeld radiation condition at the boundary surface

Ω∝, which ensures that all acoustic waves propagate freely towards infinity
and that no reflections occur at this boundary (see (1.8)), the simplest
approach is to use the specific acoustic impedance of freely propagating
plane waves as impedance boundary condition for the pressure and normal
fluid velocity at the artificial boundary surface Ωe (see e.g. COYETTE
(1992)),
p( r ) = ρ 0 cvn ( r ), r ∈ Ω e . (1.30)
For this boundary condition to be (quasi) non-reflective for the acoustic

waves in an exterior vibro-acoustic system, the distance between the
artificial boundary surface Ωe and the boundary surface Ωa of the system
must be very large. As a result, the bounded domain V1 is still very large,
which yields a very large and computationally expensive finite element
model.
A more advanced approach consists of using ‘damping’ elements at the
artificial boundary surface of the finite element discretization of domain V1
in order to approximately model the absorption of outgoing acoustic waves.
This approach was initially proposed by BAYLISS et al. (1980, 1982) and has
been refined by BOSSUT and DECARPIGNY (1989), PINSKY et al. (1992) and
ASSAAD et al. (1993,1995). A surface with a simple geometrical shape,
usually a sphere, is selected as artificial boundary surface Ωe. The pressure
field in the unbounded domain V2 is approximated in terms of an analytical
multipolar expansion of outgoing wave functions that satisfy the
Sommerfeld radiation condition. The evaluation of this analytical expansion
at the artificial boundary surface Ωe yields the pressure values, which are
imposed at the boundary nodes of the ‘damping’ elements, located at the
artificial boundary surface in the finite element discretization of domain V1.
For ‘monopolar damping’ elements, which are based on a monopolar
pressure expansion, the artificial boundary surface must be located in the
far-field of the exterior pressure at a large distance from the boundary
surface Ωa. For ‘dipolar damping’ elements, the extent of the domain V1 can
be quite limited and becomes attractive for finite element discretization.
The Dirichlet-to-Neumann method (DtN method) is based on an exact
formulation of the boundary condition at the artificial boundary surface Ωe.
38 Chapter 1
KELLER and GIVOLI (1989) proposed an analytical solution for the pressure
field in the unbounded domain V2 that satisfies the Sommerfeld radiation
condition and an arbitrary Dirichlet condition (imposed pressure condition)
on the artificial boundary surface. They used this solution to form a relation
between the pressure and its normal gradient at the artificial boundary
surface Ωe, which yields the exact impedance boundary condition. The finite
element implementation of this exact Dirichlet-to-Neumann impedance
boundary condition has been discussed by HARARI and BLEJER (1995),
THOMPSON (1995) and GIVOLI et al. (1997). Since this method allows the
artificial boundary surface Ωe to be located in the near-field of the exterior
pressure, only a small domain V1 must be discretized, which yields a fairly
small finite element model.
Instead of using an analytical expression for the pressure field in the

unbounded domain V2 and for the derivation of the boundary condition on
the artificial boundary surface Ωe, the pressure field in V2 can also be
approximated numerically through the use of an infinite element
discretization of this unbounded domain (BETTESS and ZIENKIEWICZ
(1977), BETTESS (1992), ALLIK et al. (1995)). Each of the infinite elements
in such a discretization contains a part of the artificial boundary surface Ωe
and is infinitely extended away from the artificial boundary surface. In this
way, the infinite elements span as a single layer around the conventional
finite elements, used for the pressure field in the domain V1. The pressure
within the infinite elements is expressed in terms of shape functions with a
built-in amplitude decay and wave-like variation to model outgoing waves.
To overcome the difficulties which arise with the numerical integrations
involved in the implementation of infinite elements, infinite wave envelope
elements (ASTLEY (1983), ASTLEY and EVERSMAN (1988)) and more
recently mapped infinite wave envelope elements (ASTLEY et al. (1994),
CREMERS et al. (1994), ASTLEY et al. (1998)) have been developed. They
use shape functions similar to those of regular infinite elements but use the
complex conjugates of the shape functions as weighting functions. This
removes all wavelike terms from the element integrals and simplifies the
frequency dependence of the resulting model matrices.
The size of the resulting coupled model, consisting of a finite element model
for the structural domain and for the acoustic domain V1 and an infinite
element model for the acoustic domain V2, is strongly dependent on the
location of the artificial boundary surface Ωe and the subsequent size of
domain V1. The size of domain V1 is in its turn strongly related to the shape
functions, used in the infinite elements in domain V2. The more accurate
these shape functions can represent outwards travelling waves, the smaller
the size of domain V1 can be to get an accurate prediction of the coupled
response in exterior vibro-acoustic systems. In this framework, BURNETT
(1994) has proposed the use of spheroidal co-ordinate systems for the
formulation of the infinite element shape functions. This permits the
required extent of domain V1 to be reduced in the case of slender objects.
1.3.3. Coupled FE/BE models
For coupled vibro-acoustic problems, the finite element method can be

applied for the prediction of the structural response and the direct or indirect
boundary element method can be applied for the prediction of the acoustic
response13.
1.3.3.1. Coupled FE/direct BE model

The coupling of a structural FE model to an acoustic direct BE model was
initially proposed by ZIENKIEWICZ et al. (1977) and has been refined by
several authors (see e.g. WILTON (1978), MATHEWS (1986), EVERSTINE and
HENDERSON (1990)).
The structural and fluid displacement continuity at the fluid-structure
coupling interface is taken into account by a transformation matrix [ T ] ,
which relates the fluid normal velocity values vni at the nodes of the acoustic
BE mesh to the displacement components wi at the nodes of the structural
FE mesh,
{vni } = jω [ T ]. {wi } . (1.31)
Assuming that the whole boundary surface of a coupled vibro-acoustic

system consists of an elastic structural surface, the coupling of a structural
FE model, as formulated in (1.12), to an acoustic direct BE model, as
formulated in (1.20), yields a non-symmetric coupled model,
13
Although, in principle, a structural BE model can also be linked with an acoustic
BE model, this is not commonly done in practice, since, for similar reasons as for
acoustic problems, a dynamic structural BE model can hardly compete with a
structural FE model for bounded structural domains.
40 Chapter 1
 K s + jωCs − ω 2 Ms Lc  wi   Fsi 

 .   =   . (1.32)
 ρ 0ω B(ω ). T A(ω ) î pi  î Fai 
2
The matrix [ Lc ] results from the acoustic pressure loading on the elastic
structure. The derivation of this model is given in appendix D.
Note that, since the direct boundary element method is confined to either
interior or exterior acoustic systems with a closed boundary surface, the
coupled FE/direct BE model can be used for interior or exterior coupled
vibro-acoustic systems with a closed boundary surface.
1.3.3.2. Coupled FE/indirect BE model

For coupled vibro-acoustic systems with an open boundary surface or
combined interior/exterior coupled vibro-acoustic systems, a structural FE
model can be linked with an acoustic indirect BE model.
Assuming that the whole boundary surface of a vibro-acoustic system
consists of an elastic structural surface, the coupling of a structural FE
model, as formulated in (1.12), to an acoustic indirect BE model, as
formulated in (1.21), yields a symmetric coupled model,
 K s + jωCs − ω 2 M s Lc 
 . wi  =  Fsi  .
F (ω ) îµ i  î Fai 
1 (1.33)
 LTc
 ρ 0ω 2 
The coupling matrix [ Lc ] and its transpose result from the acoustic pressure
loading on both sides of the elastic structure and the structural and fluid
displacement continuity at the fluid-structure coupling interface.
A detailed mathematical description of this model is given e.g. by MARIEM
and HAMDI (1987), COYETTE and FYFE (1989), JEANS and MATHEWS
(1990).
1.3.4. Coupled FE/FE models versus coupled FE/BE models
When comparing the coupled FE/FE models, discussed in 1.3.2., and the
coupled FE/BE models, discussed in 1.3.3., in terms of computational
efficiency, two important features may be noticed.
• As already indicated in 1.3.1.3., the smaller size of the acoustic part in a
coupled FE/BE model usually doesn’t result in a higher computational
efficiency for interior problems, since the acoustic matrices in such a

coupled model are, in contrast with a coupled FE/FE model, fully
populated, complex and frequency dependent. The computational
efficiency of coupled FE/BE models may be improved, for instance, by
using a frequency interpolation technique to avoid the construction of the
model matrices at every frequency of interest (SCHENCK and BENTHIEN
(1989)). This induces, however, an additional prediction error.
• In addition, the overall size reduction, obtained by using a boundary
surface discretization instead of a domain discretization for the acoustic
domain, is less pronounced in coupled vibro-acoustic problems than in
uncoupled acoustic problems. This is due to the fact that the structural
part in a coupled model has a significant contribution to the total size of
a coupled model. Although in most coupled vibro-acoustic systems, the
elastic shell structure is comprised in a fairly small surface domain,
while the fluid is comprised in a vast volumetric domain, the size of the
structural part in a coupled FE/FE model is usually comparable to the
size of the acoustic part. This is due to two major reasons. On the one
hand, a finite element discretization uses several displacement
components per structural node, while only one pressure value is needed
per acoustic node. On the other hand, the structural wavelengths at a
given frequency are usually smaller than the corresponding acoustic
wavelengths, at least in the frequency range where vibro-acoustic
coupling effects are significant. As a result, the subsequent higher spatial
variation of the structural response compared to the acoustic response
requires a finer element discretization of the structural domain compared
to the acoustic domain. Only in the vicinity of the fluid-structure
coupling interface, the density of the acoustic domain discretization must
be similar to the one of the structural discretization in order to accurately
represent the near-field effects, induced by the structural displacements.
These features indicate why a coupled FE/FE model has usually a higher
computational efficiency than a coupled FE/BE model for analysing interior
coupled vibro-acoustic problems, such as the interior acoustic fields in cars
and aircraft. On the contrary, for coupled vibro-acoustic problems with an
unbounded acoustic domain, as encountered, for instance, in the design of
loudspeakers or in the analysis of the sound transmission loss of light-
weight partitions, coupled FE/BE models become more efficient, since these
models have the advantageous property of implicitly taking into account the
Sommerfeld radiation condition, whereas this condition can only be satisfied
in an approximate manner in coupled FE/FE models (see 1.3.2.2.).
42 Chapter 1
1.3.5. Limitations of coupled models
In comparison with uncoupled vibro-acoustic problems, the practical use of

existing numerical prediction tools for coupled vibro-acoustic problems is
more limited, since, as discussed below, the involved model sizes and the
associated computational efforts and memory requirements are larger and
the commonly used model size reduction techniques are less efficient.
larger model sizes

The discussion of the finite element method in 1.3.1.1. has already revealed
the necessity of using a large number of degrees of freedom to get an
acceptable level of accuracy. In comparison with uncoupled vibro-acoustic
problems, the involved model sizes and the associated computational efforts
and memory requirements are substantially larger for coupled vibro-acoustic
problems. This is mainly due to the following reasons.
• In order to incorporate the vibro-acoustic coupling effects, the structural
and the acoustic problem must be solved simultaneously, whereas for
uncoupled vibro-acoustic systems, the structural and the acoustic
problem may be solved in a sequential procedure.
• Coupled models are not only larger, but their numerical solution
algorithms have also a smaller computational efficiency. Compared with
an uncoupled structural or an uncoupled acoustic finite element model of
the same size, solving a coupled Eulerian FE/FE model is
computationally less efficient, since such a coupled model is no longer
symmetric and since its bandwidth is substantially larger (see (1.22)).
reduced efficiency of model size reduction techniques

For the uncoupled structural and uncoupled acoustic problems, encountered
in uncoupled vibro-acoustic systems, their model sizes and subsequent
computational efforts may be efficiently reduced by using model size
reduction techniques, such as the modal expansion technique, the
component mode synthesis and the Ritz vector expansion technique. For
coupled vibro-acoustic problems, however, the efficiency of these model
size reduction techniques is significantly reduced, as discussed below.
In the modal expansion technique, the dynamic field variables are expanded
in terms of the modes of the considered system. For element based models,
the contributions of the different modes to the dynamic response become the
unknowns of the modal model instead of the nodal degrees of freedom in the
original model. Although all modes should be used to get the same accuracy
as with the original model14, a relatively small truncated set of modes yields
already a level of accuracy close to the one of the much larger original
model. In this framework, a rule of thumb states that an accurate prediction
of the steady-state dynamic behaviour in a certain frequency range is
obtained by using all modes with natural frequencies, smaller than twice the
upper frequency limit of the considered frequency range. Especially in the
low-frequency range, where modal densities are small, this yields a
significant model size reduction. An illustration of this reduction efficiency
is given in appendix A.
In addition to the reduced model size, the matrices in a modal model often
become diagonal due to the orthogonality properties of the modes. As a
result, when the steady-state response is needed at a large number of
frequencies, the computational effort for constructing the modal base and
then solving the reduced modal model at each frequency of interest is
usually substantially smaller than for directly solving the original model at
each frequency.
The application of the modal expansion technique for coupled vibro-
acoustic models suffers, however, from some severe problems:
• The extraction of the modes from a coupled FE/FE model is much more
time-consuming than from an uncoupled structural or uncoupled acoustic
FE model. Non-symmetric eigenvalue solvers, needed for the mode
extraction from the non-symmetric and large sized Eulerian FE/FE
models (1.22), have a much smaller computational efficiency than the
symmetric eigenvalue solvers for uncoupled FE models, as illustrated in
appendix D. In this way, the mode extraction becomes a computationally
very expensive calculation for many coupled vibro-acoustic problems.
• The mode extraction from coupled FE/BE models with an algebraic
eigenvalue solver is not possible anyway due to the implicit frequency
dependence of the matrix coefficients in the acoustic part of the models.
The modes can only be obtained after introducing some approximations
in the coupled FE/BE models. COYETTE et al. (1989) proposed, for
example, to evaluate the frequency dependent coefficients in matrix
[ F(ω )] in a coupled FE/indirect BE model (1.33) at a particular
frequency ω = ω , which lies somewhere in the considered frequency
14
since the total number of modes equals the total number of degrees of freedom of
the model, no model size reduction would be obtained
44 Chapter 1
range of interest. In this way, the undamped modes of a coupled system

result from the algebraic eigenvalue problem,
([ Ks ] − ω 2 ([ Ms ] + [ M ])).{wi } = {0} , (1.34)
[ ]
where M is referred to as the added mass matrix,
[ M ] = ρ 0 [ Lc ][ F(ω )] −1[ Lc ]T . (1.35)
GIORDANO and KOOPMANN (1995) made the implicit frequency depen-

dence of the acoustic part in a coupled FE/direct BE model (1.32)
explicit by using an approximate Taylor series expansion of the
corresponding acoustic matrices, consisting of frequency independent
matrices, weighted by a certain power of frequency. The resulting model
is then recast in a canonical state space form, which renders an algebraic
eigenvalue problem for the mode extraction.
Due to these problems, involved with the calculation of the modes of a

coupled vibro-acoustic system, the component mode synthesis technique
provides an alternative approach to get a model size reduction. In this
technique, a system is regarded as an assembly of several components and
the dynamic variables of the total system are expanded in terms of the
modes of the different system components, which can be calculated in a
more efficient way than the modes of the total system. Since the system is
divided into several components, some additional constraints must be
specified at the interfaces between the components to enable the calculation
of the component modes. This has an important influence on the resulting
accuracy and efficiency of this model size reduction technique. For a
detailed discussion of this technique, the reader is referred to e.g. CRAIG JR.
(1987) and HEYLEN et al. (1996).
Coupled vibro-acoustic systems are often decomposed into two components,
i.e. the elastic shell structure and the acoustic fluid domain. Several types of
interface boundary conditions between the structural and acoustic
component may be used to determine the component modes. The most
commonly used interface conditions for coupled vibro-acoustic systems are
discussed below.
• WOLF JR. (1977) proposed the expansion of the structural displacement
in terms of the uncoupled structural modes and the expansion of the fluid
pressure in terms of the uncoupled acoustic modes. The uncoupled

structural modes are the modes of the elastic shell structure without fluid
pressure loading at the fluid-structure coupling interface. The uncoupled
acoustic modes are the fluid modes with rigid wall boundary conditions
at the fluid-structure coupling interface. The component mode synthesis
technique, using these uncoupled component modes, has been applied to
several of the coupled FE/FE models, described in 1.3.2. COYETTE and
DUBOIS-PÉLERIN (1994), GÖRANSSON (1992) and CURA et al. (1997)
applied it, for example, for the reduction of, respectively, a coupled
Eulerian pressure model (1.22), a coupled mixed model (1.29) and a
coupled Eulerian velocity potential model (1.26). Since the uncoupled
structural and uncoupled acoustic modes result from a symmetric
eigenvalue problem, the computational effort for constructing a set of
uncoupled modes in a component mode synthesis is much smaller than
for a set of coupled modes with the same size, needed in a modal
expansion. The efficiency of this model size reduction procedure is,
however, significantly smaller than the efficiency of the modal expansion
technique. This is mainly due to the inefficient way, in which the
displacement continuity (1.7) at the fluid-structure coupling interface is
approximated by the uncoupled acoustic modes. Since the uncoupled
acoustic modes are calculated with rigid wall boundary conditions at the
fluid-structure interface, the fluid displacement, normal to this interface,
is zero in each mode. In this way, any combination of the uncoupled
acoustic modes yields a zero normal fluid displacement and violates the
displacement continuity condition. To get an accurate approximation of
the near-field pressure effects in the vicinity of the fluid-structure
coupling interface, which are usually associated with the displacement
continuity, a lot of high-order acoustic modes should be comprised in the
uncoupled modal base, which results in a large size of the modal model.
Consequently, the benefit of a computationally efficient construction of
the modal base is reduced by the smaller model size reduction, obtained
with uncoupled component modes. An illustration of the reduced
efficiency of the model size reduction for this type of component modes
is given in appendix D.
In this respect, it should be noted that a model size reduction can also be
obtained by using uncoupled structural modes for the expansion of the
structural displacement, without reducing the acoustic part of a coupled
model. This is described e.g. by COYETTE and VAN DE PEER (1993) for
the reduction of a coupled FE/indirect BE model.
46 Chapter 1
• To enhance the ability of the acoustic component modes to represent the

displacement continuity and the associated near-field effects, YOUNES
and HAMDI (1997) recently proposed to calculate the acoustic component
modes with an impedance boundary condition at the fluid-structure
coupling interface. Since the efficiency of this type of component mode
synthesis is largely dependent on the applied impedance value, this
approach is still subject of on-going research, mainly looking for a
systematic procedure to determine a proper impedance value.
• Another way to enhance the accuracy of a component mode synthesis has
been described by DANIEL (1980), who extended the fixed-interface
component mode synthesis technique, developed by CRAIG JR. and
BAMPTON (1968) and SZU (1976) for large structural problems, to
coupled vibro-acoustic problems. In this approach, the structural
displacement is still expanded in terms of a set of uncoupled structural
modes. The fluid pressure is expanded in terms of a set of fixed-interface
modes and constraint modes. Fixed-interface modes are acoustic modes
with zero pressure boundary conditions at the fluid-structure coupling
interface. In each of the constraint modes, which are usually determined
with a static condensation technique, one of the previously fixed
interface degrees of freedom is released. In this way, each constraint
mode represents the static pressure field with a unit pressure at the
released interface degree of freedom. ICHIKAWA and HAGIWARA (1994)
refined the method by determining the constraint modes with a dynamic
calculation procedure. Since a large number of constraint modes are
needed to get an acceptable level of accuracy, the resulting reduction of
the model size and the computational effort is fairly small, which makes
this approach of minor practical use for coupled vibro-acoustic problems.
The Ritz vector expansion technique provides a way to circumvent the

problems, associated with the fact that the modes of a system or its
components are independent of the external excitations of the system.
Especially for local excitations, such as structural point forces and acoustic
point sources, the forced responses often peak in the vicinity of the
excitation point and attenuate rapidly away from the excitation. In order to
accurately represent these local spatial patterns with excitation independent
modes, a lot of high-order modes should be taken into account in the modal
base of a modal expansion or component mode synthesis model. This has a
disadvantageous effect on the model size and subsequent computational
effort. The Ritz vectors constitute an orthogonal set of functions, which not
only depend on the dynamic properties of the system but also on the
external excitations. COYETTE (1990), for instance, proposed the use of Ritz
vectors for the expansion of the structural displacement and the fluid
pressure in a coupled Eulerian FE/FE model (1.22). SEYBERT et al. (1993)
expanded the structural displacements in a coupled FE/direct BE model
(1.32) in terms of Ritz vectors. In comparison with a modal expansion
model, a smaller number of Ritz vectors is needed to get an acceptable level
of accuracy, since the Ritz vectors depend also on the external excitations. A
Ritz vector expansion is, however, not commonly preferred to a modal
expansion, since the benefit of the smaller model size is often nullified by
the increased computational effort for calculating an orthogonal set of Ritz
vectors.
limited applicability
As already mentioned in 1.3.1.1., the number of elements in a finite element
or boundary element discretization should increase as the frequency
increases. An increase of the number of elements yields a larger model size
and requires a larger amount of computational effort and memory resources.
As a result, the applicability of the finite element and boundary element
method is practically restricted to a limited frequency range. Above a
certain frequency limit, which depends on the nature of the problem and on
the available computer resources, these prediction methods would require a
prohibitively large amount of computational effort and memory resources to
get an acceptable level of accuracy.
As discussed above, the model sizes and subsequent computational efforts
are larger and the efficiency of model size reduction techniques is smaller
for coupled vibro-acoustic problems than for uncoupled structural and
uncoupled acoustic problems. For these reasons, the practical frequency
upper limit for applying element based prediction techniques for coupled
problems is substantially smaller than for uncoupled problems. Even with
the nowadays available enormous computational power, the huge computa-
tional effort involved in coupled models put a severe practical limitation to
their use for industrial engineering problems. ROOZEN (1992) illustrated, for
example, that an uncoupled finite element model of 550000 degrees of
freedom only allows an accurate study of the vibrational behaviour of a 2m
long fuselage section of a Fokker 50 aircraft at frequencies up to 225 Hz. To
study the coupled vibro-acoustic behaviour of a similar section of a SAAB
2000 aircraft at frequencies up to 120 Hz, GREEN (1992) needed a coupled
finite element model of more than 400000 degrees of freedom.
48 Chapter 1
1.4. Recent advances in numerical prediction techniques
1.4.1. Motivation
As indicated in 1.3.5., the use of element based prediction techniques for the
steady-state dynamic analysis of coupled vibro-acoustic systems is restricted
to a limited frequency range. Above the practically induced frequency limit,
one should resort to alternative prediction techniques such as statistical
energy analysis (see e.g. LYON and DEJONG (1995), DELANGHE (1996)) or
energy flow analysis (see e.g. NEFSKE and SUNG (1989), CHO (1993)).
These prediction techniques are appropriate for high-frequency modelling,
since they are based on the high modal density of a system at high
frequencies. These techniques are, however, probabilistic, in that they only
provide averaged approximations of the global response of a dynamic
system. The statistical approach of these non-deterministic techniques is not
only motivated by the computational efficiency, but also by the fact that, at
high frequencies, the dynamic behaviour is very sensitive to the system
parameters, such as material properties and boundary conditions. Since the
physical values of these parameters are only known with a certain level of
uncertainty, deterministic models, which yield only nominal prediction
results, become useless at high frequencies. In this respect, KOMPELLA and
BERNHARD (1993) measured the vibro-acoustic characteristics of a large
series of nominally identical cars and illustrated the large variations of these
characteristics at high frequencies.
At present, vibro-acoustic prediction techniques suffer from what LEURIDAN
(1995) calls a ‘mid-frequency crisis’: the currently available prediction
techniques are not appropriate for the dynamic analysis in the mid-
frequency range, which lies above the frequency limit of element based
prediction techniques, but where the high modal density assumption of the
probabilistic prediction techniques is not yet valid. However, for many
industrial engineering problems, there is a strong need for deterministic
techniques that provide accurate coupled vibro-acoustic predictions in this
mid-frequency range. As already mentioned in section 1.3.1.1., LE SALVER
(1998) indicated, for example, that even for the uncoupled vibro-acoustic
analysis of the acoustic field, induced by the displacements of a car body
without trim panel damping treatments, the practical frequency limitation
lies around 200 Hz, although coupled vibro-acoustic predictions for a fully
trimmed car are needed up to much higher frequencies (see e.g. LIM and
STEYER (1992)). For the vibro-acoustic analysis of the passengers’ cabin in
propeller aircraft, the application range of deterministic prediction

techniques, which currently covers only the propeller blade pass frequency,
should be extended to a frequency range that covers several harmonics of
this frequency (see e.g. GREEN (1992)).
One of the big challenges in coupled vibro-acoustic modelling is thus the
development of deterministic prediction techniques, which are appropriate
for a larger frequency range and narrow the mid-frequency twilight zone.
In order to meet this challenge, new developments should focus on the

enhancement of the convergence rate, viz. providing accurate predictions
with smaller model sizes. Within the framework of the element based
modelling concept, using locally defined approximate shape functions,
several new developments have been initiated, as briefly described in
section 1.4.2. Some alternative prediction methods, based on the Trefftz
approach, are discussed in section 1.4.3. The main feature of the latter
methods is the use of exact, instead of approximate, solutions of the
governing partial differential equations. Although these recent developments
may offer some beneficial convergence properties, the current state of
development of these methods still encounters some severe limitations and
requires a substantial amount of profound research before these methods
may become valuable modelling alternatives for solving engineering
problems and, in particular, for solving coupled vibro-acoustic problems.
Section 1.4.4. outlines the basic principles of a newly developed wave based
prediction technique, which forms the scope of this dissertation. This
technique is a particular Trefftz-based prediction technique, which, in
contrast with the other Trefftz-type developments, provides an efficient way
to apply the beneficial convergence properties of the Trefftz approach for
general coupled vibro-acoustic problems.
1.4.2. Advances in element based prediction techniques
h-, p- and hp-methods

The main reason for the large size of element based models is the fine
element discretization that is needed for an accurate representation of the
dynamic response (see 1.3.1.1.). The construction of such an element
discretization (mesh generation) is based on the engineering judgement and
experience of the analyst and performed by an automatic mesh generator.
This usually results in a mesh of fairly uniform density, i.e. a mesh in which
most elements have a similar geometrical size and in which the polynomial
50 Chapter 1
shape functions of the parent elements are of the same (low) order.
However, since the spatial variation of the dynamic response often varies
within the considered domain, a fine discretization is only required in those
areas of the domain where the response has a high spatial variation, while a
coarse discretization is allowed in the areas where the response has a low
spatial variation. Hence, a uniform mesh density is not always the most
efficient discretization.
To optimise the discretization, some adaptive mesh refinement techniques
have been developed (JIROUSEK (1986a), SZABO (1986), BABUŠKA (1987)),
based on local error estimators (ZIENKIEWICZ and ZHU (1987)). In these
techniques, the considered domain is initially discretized into a coarse mesh
of fairly uniform density. A local error estimator evaluates the distribution
of the resulting approximation errors over the domain. In areas of large
error, the discretization is refined by dividing the elements into several
smaller elements (h-method) or by increasing the order of the polynomial
shape functions in the elements (p-method) or a combination of both (hp-
method). After some iterations of this procedure, a mesh is obtained, in
which a fine discretization is only used in those areas where it is really
necessary. Recently, these refinement techniques have been successfully
applied for uncoupled acoustic problems (IHLENBURG and BABUŠKA
(1995,1997)) and some dedicated local error estimators (STEWART and
HUGHES (1996)) have been developed. Since these techniques ensure that
there is no waste of computational effort, they could be very useful for the
computationally expensive coupled vibro-acoustic problems. They are,
however, not yet applied for this type of problems, at least not to the
author’s knowledge.
Galerkin least-squares FEM

Most fluids in an acoustic problem are non-dispersive in that the speed of
sound c is independent of the frequency ω, resulting in an acoustic wave-
number k (=ω/c) that is proportional and an acoustic wavelength λ (=2π/k)
that is inversely proportional to the frequency. An entirely dispersion-free
representation of an acoustic wave cannot be obtained from a conventional
Galerkin finite element formulation and this dispersion error gets worse for
increasing frequency.
To reduce the dispersion error, HARARI and HUGHES (1992) and THOMPSON
and PINSKY (1995) applied the Galerkin least-squares finite element
method, developed by HUGHES et al. (1989), for uncoupled two-dimensional
acoustic problems. The method is based on a formulation, in which a least-
squares form of the residual of the governing Helmholtz equation is added
to the conventional Galerkin weighted residual formulation. The least-

squares form contains a mesh parameter, which can be determined such that
the dispersion error in the representation of a plane acoustic wave
propagating in a certain direction is minimised. Several numerical examples
have illustrated that this Galerkin least-squares FEM yields a higher
convergence rate than a conventional Galerkin FEM.
spectral element method

The spectral element method is basically a finite element method, in which
high-order polynomial functions are used as parent element shape functions.
Similar to the construction of Lagrangian elements, the shape functions in a
two- or three-dimensional parent element are products of, respectively, two
or three shape functions of a one-dimensional element. However, in contrast
with Lagrangian elements, the nodes of the generating one-dimensional
elements are no longer equidistant. This method was introduced by PATERA
(1984) for fluid dynamics problems. The application of the method for
acoustic and elastic wave propagation problems is mainly due to Seriani and
his collaborators. They proposed a one-dimensional acoustic element
(PRIOLO and SERIANI (1991)) and a two-dimensional rectangular element
for acoustic (SERIANI and PRIOLO (1991)) and elastic (SERIANI et al. (1992))
wave propagation modelling. In these elements, the nodes of the generating
one-dimensional element are located at the integration points of the
Chebyshev-Gauss-Lobatto quadrature rule. Based on the parametric
mapping concept, PRIOLO and SERIANI (1994) defined some two-dimen-
sional quadrilateral elements for acoustic and elastic wave propagation
modelling in homogeneous media and, later on, for the modelling of
acoustic wave propagation in heterogeneous media (SERIANI and PRIOLO
(1994)) and elastic wave propagation in anisotropic media (SERIANI et al.
(1995)). Several numerical examples have illustrated that the spectral
element method, in comparison with the conventional finite element method
with low-order polynomial shape functions, yields an improved convergence
rate, at least when the exact solutions are sufficiently smooth. Whether this
method yields also an improved convergence rate for three-dimensional
acoustic and (coupled) vibro-acoustic problems, is still an open question.
52 Chapter 1
1.4.3. Trefftz approach
As already indicated in 1.3.1.1., the use of polynomial shape functions in the

finite element method, as well as in the recent developments, discussed in
1.4.2., has two major drawbacks.
• The polynomial functions are no exact solutions of the governing partial
differential equations.
• The prediction accuracy for derived secondary variables is smaller than
for the primary field variables.
These drawbacks would vanish, if the solutions for the field variables are
expressed in terms of exact solution functions of the governing partial
differential equations. In 1926, TREFFTZ utilised already this idea and
proposed an alternative for the Rayleigh-Ritz method, discussed in 1.3.1.1.
Almost half a century later, researchers became interested again in Trefftz’
ideas, when their potential advantages for an efficient use in numerical
prediction techniques were recognised. This has led to a modelling method,
which is commonly denoted as Trefftz method. The dynamic stiffness
method combines the Trefftz approach with the element approach of the
finite element method, but its use is restricted to continuum problems with
some particular geometrical shapes. The T-element method combines both
approaches in a more generally applicable procedure.
This section describes the basic concepts of these three methods and
indicates their current limitations, which explain why, despite the promising
features regarding computational efficiency, these methods still require a
substantial amount of research, before they may become mature and
versatile modelling alternatives.
1.4.3.1. Trefftz method

The original ideas, proposed by TREFFTZ (1926), have led to the
development of the indirect Trefftz method. More recently, these ideas have
served also as the basis for an alternative modelling approach, commonly
denoted as the direct Trefftz method. This subsection briefly discusses both
methods and indicates how the boundary element method, as discussed in
1.3.1.2. for its application to acoustic problems, may be considered as a
special implementation of the Trefftz approach. A more detailed overview
of the Trefftz method is given e.g. by HERRERA (1984a) and by KITA and
KAMIYA (1995).
Indirect Trefftz method

The main feature of the indirect Trefftz method is the approximation of the
solution in terms of functions, which belong to a complete set of functions
that satisfy exactly the governing (homogeneous) equations. In the original
method, proposed by Trefftz, a variational formulation was used, in that the
contributions of the functions to the solution result from a linear system of
equations, imposing a stationarity condition on an energy functional. Since
the proposed solution exhibits only some approximation errors on the
boundary conditions, only some boundary integrals are involved in the
evaluation of the energy functional for the proposed solution. In this way,
the method may offer some computational benefits over the Rayleigh-Ritz
method, in which also some domain integrals are involved in the functional
evaluation. Although the original indirect Trefftz method utilises a
variational formulation, the contributions in the proposed solution can also
be obtained from the approximation of the boundary conditions in a
collocational, a least-squares or a weighted residual scheme. The latter
yields the same system of equations as the originally proposed variational
formulation, when the weighting functions are selected from the same
complete function set as used for the solution.
The completeness of the function set indicates the ability to represent any
possible solution in a given continuum domain. In honour of the originator
of the method, the name of T(Trefftz)-complete functions is adopted for
such sets. Important contributions to the development of a mathematical
basis for the construction of such T-complete function sets are made by
Ismael Herrera and his collaborators. They presented T-complete sets for
two- and three-dimensional problems of Laplace and Helmholtz equations
(HERRERA (1984b)), biharmonic equations (GURGEON and HERRERA
(1981)) and Stokes problems (HERRERA and GURGEON (1982)).
KUPRADZE (1979) proposed an alternative indirect Trefftz formulation15.

Instead of using T-complete functions, he proposed a solution
15
This formulation was discovered independently by several researchers, who gave
it different names: ‘method of fundamental solutions’ (JOHNSTON and
FAIRWEATHER (1984), BOGOMOLNY (1985)), ‘source function method’ (DE MEY
(1978)), ‘discrete singularity method’ (YANO et al. (1983)), ‘superposition
method’ (BURGESS and MAHAJERIN (1984)), ‘modified Trefftz method’
(PATTERSON and SHEIKH (1982,1983)), ‘boundary point method’ (JOHNSON
(1987)), ‘boundary collocation method’ (KOLODZIEJ and KLEIBER (1989)),
‘regular indirect boundary element method’ (WEARING and SHEIKH (1988)).
54 Chapter 1
approximation in terms of fundamental solutions of the governing equations.

Each fundamental solution represents the solution for a pointwise source
function, applied at a certain location in the continuum domain, which is
assumed to be infinitely extended. For the acoustic Helmholtz equation, for
instance, the Green’s kernel function (see 1.3.1.2.) may serve as a
fundamental solution. Since the fundamental solutions are singular at their
source location, Kupradze proposed the use of fundamental solutions, whose
source locations are comprised on an artificial boundary surface outside the
continuum domain of the considered problem. In this way, the approximate
solution exactly satisfies the governing homogeneous equations within the
considered continuum domain and no singularities are involved in the
approximation of the boundary conditions on the actual boundary surface.
In comparison with the classical indirect Trefftz method, this method
requires the determination of some additional parameters, i.e. the shape of
the artificial boundary surface, which comprises the source locations of the
fundamental solutions, as well as its distance from the actual boundary
surface of the considered continuum domain. As indicated e.g. by ZIELINSKI
(1995), these parameters influence the convergence of the solution and
determine the effectiveness of the method. Therefore, the method requires
some additional numerical tests for the determination of these parameters,
before it can be applied for a given problem, and, in its present stage of
development, the method is, in general, less reliable than the classical
indirect Trefftz method using T-complete solution functions.
Direct Trefftz method

The direct Trefftz formulation is based on a weighted residual formulation
of the governing equations, using T-complete functions as weighting
functions. For this particular weighting function selection and by applying
integration by parts, the weighted residual formulation consists only of
boundary domain integrals. As in the boundary element method, the
boundary surface is discretized into small boundary elements and the
solution approximations for the field variables, occurring in the boundary
domain integrals, are expressed in terms of classical (polynomial) shape
functions, defined within each boundary element. The shape function
contributions result from the discretized version of the weighted residual,
now boundary integral, formulation, combined with the boundary conditions
of the considered problem. The term ‘direct’ indicates that the unknowns in
the system of equations are physical quantities on the boundary surface of
the considered continuum domain.
This formulation was firstly presented by CHEUNG et al. (1989) in the two-
dimensional potential problem and extended to the two-dimensional elastic
problem (JIN et al. (1990)), the two-dimensional Helmholtz problem
(CHEUNG et al. (1991)) and the plate bending problem (JIN et al. (1993)).
Relation with the boundary element method

The direct and indirect boundary element method, as discussed e.g. in
1.3.1.2. for their application to acoustic problems, are closely related to the
direct and indirect Trefftz method, respectively.
The main difference between the direct Trefftz and the direct boundary
element method is that the direct Trefftz method uses regular T-complete
weighting functions, whereas the direct boundary element method uses
singular fundamental solutions, such as the acoustic Green’s kernel
function, as weighting functions. Since the direct Trefftz method has only
been applied so far for a small number of problems, a substantial amount of
additional validation examples are needed, before it may become apparent
whether or not this is an attractive, generally applicable alternative for the
boundary element method.
The indirect boundary element method is very similar to the alternative
indirect Trefftz method, proposed by Kupradze. For acoustic problems, for
instance, the indirect boundary element method is based on the
representation of any acoustic pressure field in a certain continuum domain
as the response to a distribution of monopole and dipole sources, located on
the boundary surface of the continuum domain. The main difference with
the alternative indirect Trefftz method is that, in the latter method, the
singularities of the source functions are located outside the continuum
domain.
current limitations
For large and/or geometrically complex continuum domains, a large number
of globally defined T-complete functions is required to obtain a reasonable
accuracy. Despite the current availability of several function sets, for which
a mathematical proof of their completeness exists, their practical
implementations yield solution matrices, which are very ill-conditioned (see
e.g. ZIELINSKI and HERRERA (1987)). This puts a severe practical limitation
on the use of the indirect Trefftz method. Hence, for the indirect Trefftz
method to become a valuable modelling alternative, alternative T-complete
function sets are needed, which can be implemented without practical
condition problems.
56 Chapter 1
1.4.3.2. Dynamic stiffness method

The dynamic stiffness method introduces the Trefftz approach, in which the
field variable solutions are expressed in terms of exact solution functions of
the governing dynamic equations, into the element concept, in which some
field variable-related quantities, specified at some nodes on the element
boundary, are related to some generalised excitation forces at these nodes.
This subsection briefly describes how the dynamic stiffness method yields
the exact solution for continuum problems, whose steady-state dynamic
behaviour is governed by an ordinary differential equation in one
independent spatial variable16. For some particular types of continuum
problems, whose dynamic behaviour is governed by partial differential
equations in several independent spatial variables, the method may yield
some highly accurate solution approximations, by expressing the field
variables in terms of frequency and wavenumber spectra, of which the
components are governed by ordinary differential equations in one
independent variable17.
For several continuum problems, of which the steady-state dynamic

behaviour is governed by ordinary differential equations in one independent
spatial variable, the continuum domain may be discretized into a number of
one-dimensional elements, such that all geometrical and material
discontinuities and external excitations occur only at the interfaces between
the elements. In this way, the field variables within each element are
governed by homogeneous ordinary differential equations and their exact
solutions may be expressed in terms of the finite number18 of homogeneous
solutions of these equations. By constructing appropriate linear
combinations of these homogeneous solutions, nodal shape functions can be
obtained. Each of them represents an exact solution, which has a unit value
for one degree of freedom in one of the two endpoints of the element and in
19
which all other endpoint degrees of freedom are zero. Based on the shape
16
in this case, the dynamic stiffness method is often referred to as ‘exact finite
element method’
17
since the dynamic field variables are represented in terms of frequency and
wavenumber spectra, the dynamic stiffness method is often referred to as
‘spectral element method’, but is not related to the identically called method,
discussed in 1.4.2.
18
equal to the order of the differential equation
19
the total number of endpoint degrees of freedom is equal to the order of the
governing differential equations
function expansions for the field variables within an element, a dynamic

stiffness matrix can be constructed, which exactly relates the nodal field
variables of the element to the nodal excitations, applied at the endpoints of
the element. The dynamic stiffness matrices of the different elements are
then assembled by enforcing the boundary conditions of the considered
problem and by enforcing the compatibility of the nodal field variables and
the nodal and external excitations at the element interfaces. The obtained
global matrix equation may then be solved for the unknown nodal field
variables and the back-substitution of these results into the shape function
expansions yields the exact solution for the considered continuum problem.
FINNVEDEN (1994) illustrated, for example, how the element shape
functions are obtained for the dynamic displacements in a structural beam
element and applied the dynamic stiffness method to calculate the steady-
state dynamic response in a railway car frame structure, which consists of
several structural beams. KARASALO (1994) applied the method to model
the wave propagation in a one-dimensional layered fluid-solid medium.
When a distributed external excitation is applied to the continuum domain,
the ordinary differential equations that govern the steady-state dynamic
variables within the elements are inhomogeneous for at least one element of
the domain. Although the homogeneous nodal shape functions in these
elements are no exact local solutions anymore, they still offer a beneficial
convergence rate compared to the polynomial shape functions, used in a
conventional finite element discretization. LEE and LEE (1997) illustrated
this, for example, for the case of a structural beam with a distributed
dynamic force load.
The dynamic behaviour of general two- and three-dimensional continuum

problems is governed by partial differential equations in one time-variable
and, respectively, two and three independent spatial variables. Provided that
the problem geometry and material properties are piecewise constant in one
spatial variable and that they are invariant in the other spatial variables, the
dynamic stiffness method may be applied for the prediction of the dynamic
behaviour in each piecewise constant part of the continuum domain.
DEGRANDE and DE ROECK (1992), for instance, applied the dynamic
stiffness method for the modelling of the transient wave propagation in a
two-dimensional multilayered system, consisting of infinitely extended
horizontal layers of dry and saturated poroelastic media, and DEGRANDE et
al. (1997) extended the method to include also unsaturated porous layers.
The distributions of the field variables in a certain horizontal surface, i.e. at
a certain vertical position, within a layer are represented in terms of their
58 Chapter 1
double Fourier spectra, which result from successively transforming the

time-variable into the frequency domain and the horizontal spatial variable
into the horizontal wavenumber domain. In this way, the original problem of
finding the field variables at any time and in any horizontal and vertical
position in the continuum domain is turned into the problem of finding the
contributions of any frequency/wavenumber component in the double
Fourier spectra of the field variable distributions in any horizontal surface,
i.e. at any vertical position. The dependence of each contribution on the non-
transformed vertical spatial variable is now governed by a homogeneous
ordinary differential equation, of which the exact solution may be expressed
in terms of the finite number of homogeneous solutions of this equation. By
rearranging these homogeneous solutions into conventional element shape
functions, a dynamic stiffness matrix can be constructed, which exactly
relates the contributions of a certain frequency/wavenumber component in
the double Fourier spectra of the field variable distributions in the two
external surfaces of a horizontal layer to the contributions of this
frequency/wavenumber component in the double Fourier spectra of the
external excitations, applied at both surfaces. The different dynamic
stiffness matrices are then assembled, according to the, now double Fourier
transformed, boundary conditions and the necessary compatibility of the
field variables at the interfaces between the different layers. The obtained
global matrix equation may then be solved for the unknown contributions of
each frequency/wavenumber component in the double Fourier spectra of the
field variable distributions in the two external layer surfaces. The back-
substitution of these results into the associated shape function expansions
yields the exact double Fourier spectra of the field variable distributions in
any horizontal surface within each layer. The numerical implementation of
the method yields highly accurate, but not exact, spectra, since the
continuous Fourier transformations must be replaced by discrete Fourier
transformations.
When all parts of the boundary surface of a finite sized continuum domain
are located in planes, which can be described by a constant value for one of
the independent spatial variables, as it is the case for rectangular plates
(FINNVEDEN (1996), LEE and LEE (1997)) and cylindrical pipes (BILODEAU
and DOYLE (1997), FINNVEDEN (1997)), the dynamic stiffness method
applies in a similar way as for the infinitely extended domains, but the exact
field variable spectra are no longer continuous, but discrete.
1.4.3.3. T-element method

The numerical condition problems, involved with the practical
implementation of the indirect Trefftz method, may be circumvented, to
some extent, by dividing the continuum domain into smaller subdomains
(see e.g. ZIELINSKI and HERRERA (1987)). This phenomenon has led to the
construction of T(Trefftz)-elements, which allow the introduction of the
Trefftz approach into a standard finite element scheme and which can be
used, in contrast with the dynamic stiffness method, for continuum problems
with geometrical shapes of any kind.
In the classical finite element method (see 1.3.1.1.), the field variables are
expanded in terms of shape functions, which are locally defined within the
elements. The conformity of the element discretization is verified a priori
and the ‘equilibrium’ and constitutive relations, expressed by the governing
equations, are then restored approximately in an average integral sense. An
alternative approach, based on the Trefftz methodology, proceeds in exactly
the opposite way. The field variables are expanded in terms of T-complete
functions, which satisfy a priori the governing equations within each
element, while the interelement continuity and the boundary conditions are
enforced in an average integral sense.
The T-element approach was initiated in the late 1970s (JIROUSEK and LEON
(1977), JIROUSEK (1978)) and the subsequent research efforts, mainly
performed by Jirousek and his collaborators, have led to the development of
hybrid-Trefftz elements and least-squares Trefftz elements.
Hybrid-Trefftz elements (see e.g. JIROUSEK (1986b), JIROUSEK and GUEX
(1986)) combine the Trefftz methodology with the concept of hybrid finite
element modelling (see e.g. ATLURI et al. (1983)). Two types of degrees of
freedom are involved in the definition of hybrid-Trefftz elements. The first
type are the contributions of the T-complete functions to the field variable
distributions within the elements. The second type results from the
introduction of auxiliary ‘frame’-functions. These functions are only defined
on the element boundaries and are expressed in terms of conventional
(polynomial) shape functions and associated nodal frame degrees of
freedom. The element formulation results from enforcing the conformity of
the frame functions and the corresponding T-complete function expansions
for the field variable distributions on the element boundaries, as well as
enforcing the interelement conformity and the external boundary conditions
in an average integral sense. The subsequent relation between the nodal
frame degrees of freedom of an element and the contributions of the T-
complete functions allows the elimination of the latter degrees of freedom in
60 Chapter 1
the element formulation. This yields, in exactly the same way as for
conventional finite element formulations, symmetric element matrices,
which relate the nodal frame degrees of freedom to the nodal ‘forces’.
Therefore, the linking of hybrid-Trefftz elements can be accomplished via a
standard element assemblage procedure and T-element discretizations may
also be coupled with conventional finite element discretizations.
The least-squares Trefftz element formulation is based only on the
contributions of the T-complete functions to the field variable distributions
within the element, without the introduction of auxiliary frame degrees of
freedom. ZIELINSKI and ZIENKIEWICZ (1985) indicated how the interelement
conformity and the boundary conditions may be enforced directly in a least-
squares sense. JIROUSEK and WROBLEWSKI (1994) translated this approach
into an element formulation, which is suitable for implementation into a
standard finite element code.
In comparison with conventional finite elements, Trefftz-elements have

some beneficial properties.
• Since the T-complete functions satisfy a priori the governing equations, a
substantially improved accuracy is (often) obtained for a given number
of degrees of freedom.
• The element formulation offers a greater flexibility in choosing the
element geometry, since the integrations, involved in the construction of
the element matrices, are confined to the element boundaries.
• The element formulation is suited for the development of an adaptive p-
method approach, based on very simple error estimators (see e.g.
JIROUSEK (1987), JIROUSEK and VENKATESH (1989)).
• Difficult singularity problems and concentrated or discontinuous loads
may efficiently be handled without troublesome local mesh refinements
by simply adding some special-purpose functions to the solution
expansion (see e.g. PILTNER (1985), VENKATESH and JIROUSEK (1995)).
In its current stage of development, the T-element method suffers, however,

from a severe limitation.
• The efficiency of the currently available hybrid-Trefftz and least-squares
Trefftz element formulations in fulfilling the boundary conditions and
element conformity is strongly dependent on the considered problem.
Hence, for the T-element method to become an attractive alternative for the
conventional finite element method, there is a need for some alternative
element formulations, whose efficiency is less problem-sensitive.
1.4.4. Wave based prediction technique
1.4.4.1. Basic principles

This dissertation proposes a new prediction technique for coupled vibro-
acoustic analysis, which is based on the indirect Trefftz method, discussed
in 1.4.3.1. The steady-state acoustic and structural dynamic field variables
are expanded in terms of a set of functions, which are defined in the entire
domain or at least in large subdomains of the acoustic or structural domain
in the considered coupled vibro-acoustic system. The function set
comprises, on the one hand, structural and acoustic wave functions, which
are exact solutions of the homogeneous part of the governing structural and
acoustic dynamic equations, respectively. The function set comprises, on the
other hand, some particular solution functions to account for the
inhomogeneous terms in the dynamic equations, which arise from the
external excitations and from the acoustic pressure loading on the elastic
structure in the coupled vibro-acoustic system. By selecting this function
set, the problem of determining the steady-state dynamic response in any
point of a coupled vibro-acoustic system is transformed into a problem of
determining the contributions of the acoustic and structural wave functions
to the field variable expansions. Since the proposed function set ensures a
priori that the field variable expansions are exact solutions of the governing
dynamic equations, approximation errors are only involved with the
representation of the boundary conditions. The latter are expressed in an
integral formulation to allow the determination of the wave function
contributions to the field variable expansions.
An important asset of the proposed wave based prediction technique is that a
T-complete function set for the expansion of the structural and acoustic field
variables has been found, which yields, in contrast with the existing T-
complete function sets (see e.g. ZIELINSKI and HERRERA (1987)), a system
model, whose poor numerical condition doesn’t prevent the prediction
results from converging towards the exact, unique problem solution.
Bearing in mind the prophecy by PIERCE (1990), “... there is some historical
precedent to believe that any scheme that incorporates some of the basic
physical understanding of how waves propagate should have some intrinsic
advantage over one based on brute force”, several properties of the
proposed wave based prediction technique are expected to contribute to its
improved convergence rate, compared with the existing prediction
techniques.
62 Chapter 1
• In contrast with the polynomial shape functions in a finite element

discretization, the proposed T-complete function set exactly satisfies the
governing dynamic equations and inherently incorporates the oscillatory
wave nature of the dynamic response. In this way, the major cause of the
large sizes of finite element models (see 1.3.1.) is circumvented.
Moreover, the spatial derivatives of the wave functions, which are used
in the expansions for the derived secondary field variables, such as the
fluid velocity, incorporate an oscillatory nature with the same spatial
period as the wave functions in the primary variable expansions. This is
advantageous for the convergence rate, especially in the case of coupled
vibro-acoustic problems, for which the effect of the fluid on the structure
is pressure controlled, but for which the effect of the structure on the
fluid is velocity controlled.
• The wave based prediction technique has an advantageous property in
common with the boundary element method, in that only some
approximation errors are involved with the representation of the field
variable distributions on the boundary surface of the continuum domains.
However, in contrast with the boundary element method, no singular
integrations are involved with the construction of the prediction model.
• The wave based prediction technique resembles the dynamic stiffness
method, whose beneficial convergence properties have already been
proven (see 1.4.3.2.). However, in contrast with the latter method, which
can only be applied for continuum problems with some particular
geometrical shapes, the wave based prediction technique can handle a
large variety of coupled vibro-acoustic problems.
The applications for numerous two- and three-dimensional vibro-acoustic
systems, discussed throughout this dissertation, indeed confirm the expected
enhancement of the convergence rate, in that the wave based prediction
technique provides highly accurate prediction results with substantially
smaller model sizes and computational efforts than the existing element
based prediction techniques. Consequently, the technique offers an adequate
way to comply with the challenge, indicated in 1.4.1., of developing new
deterministic prediction techniques, which can be used for coupled vibro-
acoustic analyses in larger frequency ranges and which narrow the currently
existing mid-frequency twilight zone.
1.4.4.2. Outline of the dissertation

Chapter 2 gives a detailed description of the methodology of the wave based
prediction technique on the basis of the simple but representative two-
dimensional case of a coupled vibro-acoustic system, in which the fluid in a

bounded cavity domain couples with a flat structural component. The
discussion of the basic principles deals with the representation of the
dynamic field variables in terms of acoustic and structural wave functions
and some particular solutions, with the integral formulation of the boundary
conditions and with the construction of an appropriate T-complete wave
function set. Next, the practical implementation of these principles into a
numerical scheme is presented and the resulting numerical performance is
discussed. Finally, the properties of the resulting wave model are compared
with the properties of finite element and boundary element models.
Chapter 3 addresses the application of the wave based prediction technique
for two-dimensional coupled vibro-acoustic systems. The prediction results
for several illustrative examples, in which an acoustic cavity interacts with
flat as well as curved shell structures, are discussed and compared with the
results, obtained with the existing element based methods. This chapter
indicates also how the wave based prediction technique can be applied for
the modelling of porous insulation material and exterior acoustic radiation.
The application for three-dimensional systems is described in chapter 4. The
first two parts indicate the specific modelling aspects, involved with the
uncoupled dynamic analysis of flat plates and three-dimensional acoustic
cavities. The third part discusses the application of the wave based
prediction technique for the steady-state dynamic analysis of coupled vibro-
acoustic systems, in which part of the boundary surface of the acoustic
domain consists of a flat plate. In this part, the performances of the wave
based prediction technique and the finite element method are compared for
some validation examples. This part presents also an experimental
validation of the wave based prediction technique, applied for the
identification of the low-frequency modes of a double-panel partition, which
is a typical coupled vibro-acoustic system that is often encountered in noise
control engineering20.
Chapter 5 presents the general conclusions and indicates the next steps in
the development of the wave based prediction technique towards a versatile
and generally applicable modelling technique.
20
Appendix F gives a detailed description of the low-frequency modal properties of
double-panel partitions, identified from an analytical model, along with a wave
model, obtained with the proposed wave based prediction technique.
64 Chapter 1
2. METHODOLOGY OF THE
WAVE BASED PREDICTION TECHNIQUE
2.1. Introduction
In the weighted residual and Rayleigh-Ritz methods, used in the finite
element method, the field variables of a continuum problem are expanded in
terms of shape functions, which only satisfy a priori the essential boundary
conditions1. The shape function contributions are then determined by
approximately restoring the ‘equilibrium’ and constitutive relations,
expressed by the governing partial differential equations, in an average
integral sense. The main asset of subdividing the continuum domain into
small finite elements is that convergence can be obtained with simple,
locally defined shape functions. The disadvantage is, however, that the size
of the resulting model may become very large.
In 1926, TREFFTZ proposed an alternative method, in which the field
variables are expanded in terms of shape functions, which satisfy a priori the
governing partial differential equations. The shape function contributions
are then determined by approximately restoring the boundary conditions in
an average integral sense. Since an approximation is only involved with the
boundary conditions, the main asset of this method is the small size of the
resulting model. The disadvantage is, however, that, in order to obtain
1
the essential boundary conditions in acoustic problems are the prescribed pressure
boundary conditions; the essential boundary conditions in structural problems are
the prescribed translational and rotational displacement boundary conditions
66 Chapter 2
convergence, a complete set of globally defined shape functions is required,

i.e. a set of exact solution functions, which is able to represent any possible
field variable distribution in the continuum domain. Important contributions
to the development of a mathematical basis for the construction of such
complete function sets are made by Ismael Herrera and his collaborators.
They presented, for instance, complete function sets for solving steady-state
heat diffusion problems and velocity potential problems for irrotational
incompressible fluid flow, which are governed by the Laplace equation
(HERRERA (1984b)), and for solving static plate bending problems, which
are governed by the biharmonic equation (GURGEON and HERRERA (1981)).
HERRERA (1984b) presented also a complete set for solving problems,
which are governed by the Helmholtz equation. MASSON et al. (1994) used
this function set for solving steady-state acoustic problems and proposed
2
also an alternative function set , consisting of acoustic plane wave functions.
A similar set of structural plane wave functions was used by LANGLEY
(1997) for solving steady-state dynamic bending problems in plates.
Although the convergence for these function sets has been proven
theoretically, their practical convergence is, however, prevented, due to the
ill-conditioning of the involved model matrices. As a result, for the Trefftz
method to become a valuable modelling alternative for the finite element
method, new complete function sets are needed, for which the practical
convergence is ensured.
In this respect, an important breakthrough is obtained with the newly
developed wave based prediction technique, proposed in this dissertation.
Complete function sets have been found, which allow the use of the Trefftz
method for the prediction of the steady-state acoustic and dynamic structural
field variables in coupled vibro-acoustic problems. In contrast with all other
available function sets, the poor condition of the model matrices, involved
with the proposed function sets, no longer prevents the prediction results
from converging towards the exact solution.
This chapter gives a detailed description of the methodology of the wave

based prediction technique on the basis of a simple, but representative two-
dimensional interior coupled vibro-acoustic problem, defined in section 2.2.
Section 2.3. indicates how the acoustic and structural field variables may be
expanded in terms of a set of acoustic and structural wave functions, which
are exact solutions of the homogeneous parts of the governing dynamic
2
although its completeness was not proven
Methodology of the wave based prediction technique 67
equations, and some particular solution functions of the inhomogeneous

equations.
Section 2.4. describes how the contributions of the acoustic and structural
wave functions to the field variable expansions may be obtained from an
integral formulation of the boundary conditions. Two types of integral
formulations are discussed, i.e. a weighted residual and a least-squares
formulation, along with the properties of the resulting coupled vibro-
acoustic models.
For the proposed models to be convergent, a complete set of wave functions
is required, which is able to represent any possible field variable distribution
in the considered continuum domain. Section 2.5. indicates how a new,
complete set is selected from the infinite number of possible wave
functions, which exactly satisfy the homogeneous dynamic equations.
Section 2.6. addresses some practical aspects, involved with the
implementation of the modelling principles into a numerical scheme. This
includes the scaling of the wave functions, some considerations regarding
the involved numerical integrations and some rules for truncating the
infinitely extended wave function set into a finite function set, amenable to
numerical implementation.
Section 2.7. discusses the resulting numerical performance of the wave
based prediction technique. The convergence of the technique, despite the
poor numerical condition of the involved models, is illustrated for several
validation examples and the convergence rates, obtained with the weighted
residual and the least-squares formulations of the boundary conditions, are
compared. This section illustrates also the important breakthrough of the
proposed complete function set. In contrast with the existing function sets,
the numerical prediction results, obtained with the new function set,
converge towards the exact solution.
In section 2.8., the properties of the models, obtained with the wave based
prediction technique, are compared with the properties of finite element and
boundary element models.
2.2. Problem definition

Figure 2.1 shows the considered two-dimensional interior coupled vibro-
acoustic problem. The cavity boundary surface Ωa of an acoustic cavity with
domain V consists of four parts. On parts Ωp, Ωv and ΩZ, prescribed pressure,
normal velocity and normal impedance distributions are imposed,
68 Chapter 2
respectively, while part Ωs consists of a flat, flexible plate with clamped

boundaries ( Ωa = Ω p ∪ Ωv ∪ Ω Z ∪ Ωs ).
The cavity is filled with a fluid with an ambient density ρ0 and speed of
sound c. In one direction, the plate has a finite length L, while it is infinitely
extended in the other direction. The thickness of the plate is t. The plate
material has a density ρs, a Poisson coefficient ν, an elasticity modulus E
and a material loss factor η.
An external normal line force F is applied at position rF(xF’) on the plate,
while an external acoustic line source q is applied at position rq(xq,yq) in the
cavity. Both excitations have a harmonic time dependence with circular
frequency ω.
Figure 2.1 : two-dimensional interior coupled vibro-acoustic system
The steady-state acoustic pressure p at any position r(x,y) in the cavity

domain V is governed by the inhomogeneous Helmholtz equation (see
appendix A)
∇ 2 p( r ) + k 2 . p( r ) = − jρ 0ωq.δ ( r ,r q ), r ∈V , (2.1)
where k(=ω/c) is the acoustic wavenumber and where δ is a Dirac delta

function.
By defining a one-dimensional plate co-ordinate system x’ with origin at

position r0(x0,y0), as shown in figure 2.1, the geometry of the plate may be
described by the position vector
rs ( x , y ) = rs ( x 0 + x’.cos α , y0 + x’.sin α ) = rs ( x’) . (2.2)
The steady-state normal plate displacement w, with positive orientation

away from the acoustic cavity, is governed by the inhomogeneous Kirchhoff
equation for plate bending motion (see appendix B),
d 4 w( x’) F p( x 0 + x’.cos α , y0 + x’.sin α )

4
− kb4 . w( x’) = .δ ( x’, x’F ) + , (2.3)
dx’ D D
where the structural bending wavenumber kb and the plate bending stiffness
D are
ρ tω 2 Et 3 (1 + jη )
kb = 4 s and D = . (2.4)
D 12(1 − ν 2 )
The acoustic boundary conditions for this interior coupled vibro-acoustic

problem are
p( r ) = p ( r ), r ∈Ω p , (2.5)
j ∂p( r )
= vn ( r ), r ∈ Ω v , (2.6)
ρ 0ω ∂n
j ∂p( r ) p( r )
= , r ∈Ω Z , (2.7)
ρ 0 ω ∂n Z (r )
j ∂p( r )
= jωw( r ), r ∈ Ω s , (2.8)
ρ 0 ω ∂n
where p , vn and Z are prescribed pressure, normal velocity and normal

impedance functions, respectively.
The structural clamping boundary conditions are
dw( 0) dw( L )
w( 0) = w( L ) = = = 0. (2.9)
dx’ dx’
70 Chapter 2
Recall that, according to equation (1.1), vibro-acoustic coupling effects are

only important for the case of thin structures in the low- and mid-frequency
range. The fact that, for this case, the effects of shear deformations and
rotational inertia forces on the plate bending motion are negligible, justifies
the use of the Kirchhoff plate theory instead of the more general Mindlin
plate theory. Note also that the restriction to discrete line excitations induces
no loss of generality, since general distributed excitations may be regarded
as superpositions of discrete line excitations.
2.3. Field variable expansions
2.3.1. Acoustic pressure expansion
The steady-state pressure p(x,y) in the cavity is approximated as a solution

expansion p(x, y) ,
na
p ( x , y ) ≈ p( x , y ) = ∑ pa .Φ a ( x, y ) + pq ( x, y) = [Φ a ]{ pa } + pq ( x, y) . (2.10)
a=1
Each function Φ a (x, y) in the (1xna) matrix [Φ a ] is an acoustic wave

function, which satisfies the homogeneous part of the Helmholtz equation
(2.1),
− j ( k xa . x + k ya . y )
Φ a ( x, y) = e (2.11)
with 2
k xa + k ya
2
= k2. (2.12)

The unknown contributions of these acoustic wave functions to the solution
expansion are comprised in the (nax1) vector { pa } .
Function pq (x, y) is a particular solution function for the external acoustic
source term in the inhomogeneous right-hand side of the Helmholtz
equation (2.1). Several mathematical expressions may serve as particular
solution. It is advantageous, however, to select an expression, which is
already close to the physical response of the cavity. Therefore, the particular
solution function is based on the two-dimensional Green’s kernel function,

defined in equation (1.16), since it represents the free-field pressure due to
an acoustic line source,
ρ 0ωq ( 2)
pq ( x , y ) = H 0 ( k. ( x − x q ) 2 + ( y − y q ) 2 ) , (2.13)
4
where H 0(2) is the zero-order Hankel function of the second kind.

According to equation (1.3), the fluid velocity vector is approximated as

v=
j
∇p ≈
 ∂p 
j  ∂x 
 ∂p  =
j
([∂ ][Φ a ]{ pa } + [∂ ] pq )

ρ 0ω ρ 0ω   ρ 0 ω
 ∂y  (2.14)
([ Ba ]{ pa } + [∂ ] pq ) ,
j
=
ρ 0ω
where [∂ ] is a (2x1) vector of gradient operators and where [ Ba ] is a (2xna)

matrix of gradient components of the acoustic wave functions.
From the above definitions, it may be concluded that the proposed acoustic
pressure expansion (2.10) exactly satisfies the Helmholtz equation (2.1), no
matter what the values of the acoustic wave function contributions pa are.
2.3.2. Structural displacement expansion
The steady-state normal plate displacement w(x’) is approximated as a

solution expansion w(x’) ,

4 na
w( x’) ≈ w( x’) = ∑ wsΨ s ( x’) + ∑ pa wa ( x’) + w F ( x’) + wq ( x’)
s=1 a=1 (2.15)
= [Ψ s ]{w s } + [ w a ]{ pa } + w F ( x’) + w q ( x’).
The four structural wave functions Ψ s (x’) in the (1x4) matrix [Ψ s ] are four
linear independent solutions of the homogeneous part of the fourth-order
dynamic plate equation (2.3),
72 Chapter 2
Ψ s ( x’) = e − j
s
. k b . x’
, ( s = 1..4) . (2.16)

The unknown contributions of these structural wave functions to the
solution expansion are comprised in the (4x1) vector {w s } .
Function w F (x’) is a particular solution function for the mechanical force
term in the inhomogeneous right-hand side of equation (2.3). Again, the

physically meaningful expression for the displacement of an infinite plate,
excited by a normal line force F, is selected (JUNGER and FEIT (1972)),
− jF

− jk b x ’− x ’F − k b x ’− x ’F
w F ( x’) = (e − j. e ). (2.17)
4 Dkb3
Each function wa (x’) in the (1xna) matrix [ w a ] is a particular solution

function for the part of the cavity pressure loading term in the inhomoge-
neous right-hand side of equation (2.3), resulting from one of the acoustic

wave functions Φ a . The displacement of an infinite plate, excited by this
distributed pressure, may be used as particular solution function,
w a ( x’) =
−j L 
 ∫ Φ a ( x 0 + ξ.cos α , y0 + ξ.sin α ).(e − jk b x’−ξ − j. e − k b x’−ξ ). dξ 
4 Dk b3  0 

(2.18)
However, due to the flat shape of the considered plate, an expression may be
defined, which is a particular solution function and which is proportional to

the pressure loading function,
− j ( k xa .( x 0 + x ’.cos α ) + k ya .( y 0 + x ’.sin α ))
e
w a ( x’) =
[ ]
. (2.19)
D. ( k xa .cos α + k ya .sin α ) 4 − k b4
Although the latter expression has no direct physical meaning, it is preferred

to expression (2.18), since it is less computationally demanding. Note that
the latter expression can only be used for those acoustic wave functions,
whose acoustic wavenumber components kxa and kya are such that the
denominator in equation (2.19) is non-zero.


In a similar way, function wq (x’) , which is a particular solution function
for the part of the cavity pressure loading term in the inhomogeneous right-

hand side of equation (2.3), resulting from the acoustic line source function
pq , may be expressed as
wq ( x’) =
− j ρ 0 ωq  L ( 2 ) 
 ∫ H 0 ( k. ( x 0 + ξ .cos α − x q ) 2 + ( y0 + ξ.sin α − yq ) 2 ). e − jk b x’−ξ . dξ 
16 Dkb3  0 

ρ 0ωq  L ( 2) x ’−ξ 
 H ( k. ( x 0 + ξ.cos α − x q ) 2 + ( y0 + ξ.sin α − yq ) 2 ). e − k b . dξ 
3 ∫ 0
−
16 Dkb  0 
(2.20)
From the above definitions, it may be concluded that the proposed structural
displacement expansion (2.15) exactly satisfies the dynamic plate equation
(2.3), no matter what the values of the acoustic and structural wave function
contributions pa and ws are.
2.4. Coupled vibro-acoustic wave model
2.4.1. Integral formulation of the boundary conditions
Since the proposed field variable expansions (2.10) and (2.15) satisfy a
priori the governing dynamic equations (2.1) and (2.3), the contributions pa
and ws of, respectively, the acoustic and the structural wave functions to the
field variable expansions are merely determined by the boundary conditions.
The boundary conditions are defined at any position on the boundaries of
the structural and acoustic domains. However, since only finite sized models
are amenable to numerical implementation, the boundary conditions can
only be satisfied in an approximate way. This approximation may be
obtained by reformulating the boundary conditions, such that they are
satisfied in an average integral sense. Two types of integral formulations are
considered, i.e. a weighted residual and a least-squares formulation.
For the considered two-dimensional problem, only the acoustic boundary
conditions must be transformed into an approximate integral formulation.
The structural boundary conditions can be satisfied exactly, since they are
defined at some discrete locations.
74 Chapter 2
2.4.1.1. Weighted residual formulation
acoustic boundary conditions

Depending on the wave function contributions pa and ws in the field variable
expansions (2.10) and (2.15), some approximation errors are induced in the
representation of the acoustic boundary conditions, defined in equations

(2.5)-(2.8). In the proposed weighted residual formulation of these boundary
conditions, the involved residual error functions,
R p ( r ) = p( r ) − p ( r ), r ∈Ω p , (2.21)
Rv ( r ) =
ρ 0ω ∂n

j ∂p( r )
− vn ( r ), r ∈ Ω v , (2.22)
RZ ( r ) =
Rs ( r ) =
ρ 0ω ∂n

j ∂p( r ) p( r )
j ∂ p( r )
ρ 0 ω ∂n
−
Z (r )
, r ∈Ω Z ,
− jωw( r ), r ∈ Ω s ,
(2.23)
(2.24)
are orthogonalised with respect to some weighting functions. By using a

weighting function ~ p for the residual errors functions Rv, RZ and Rs and by
using a weighting function, which is proportional to the normal derivative of
~
p , for the residual error function Rp, the weighted residual formulation is
expressed as
− j ∂~
p ~ ~ ~
∫ ρ ω ∂n . R p . dΩ + ∫ p. Rv . dΩ + ∫ p. RZ . dΩ + ∫ p. Rs . dΩ = 0 . (2.25)
Ωp 0 Ωv ΩZ Ωs
In a similar way as in the Galerkin weighting procedure, used in the finite

element method, the weighting function ~ p is expanded in terms of the same
set of acoustic wave functions Φ a , used in the field variable expansions,
na
~
p( x, y) = ∑ ~pa .Φ a ( x, y) = [Φ a ]{ ~pa } . (2.26)
a=1
The substitution of the field variable expansions (2.10) and (2.15) and the
weighting function expansion (2.26) into the weighted residual formulation
(2.25) yields
{ ~pa }T . ([ Aaa

( WR )
] + [Caa ]).{ pa } + [Cas ].{ws } − { fa } = 0,
( WR ) ( WR ) ( WR )
(2.27)
where T denotes the transpose.

(WR)
The (naxna) matrix Aaa is [ ]
[ Aaa ] = Ω
( WR )
∫ ρ
j
ω
[Φ a ]T {n} T [ Ba ]. dΩ
v +Ω Z +Ω s
0
(2.28)
∫ ρ ω [ Ba ] {n}[Φ a ]. dΩ − ∫ Z [Φ a ] [Φ a ]. dΩ ,
j T 1 T
−
Ωp 0 ΩZ
where the (2x1) vector {n} contains the x- and y-component of the unit
vector, normal to the considered boundary surface.
(WR)
The (naxna) matrix Caa is [ ]
[Caa ] = −Ω∫ jω[Φ a ]T [wa ]. dΩ
( WR )
(2.29)
s
(WR)
and the (nax4) matrix Cas is [ ]
[Cas ] = −Ω∫ jω[Φ a ]T [Ψ s ]. dΩ
( WR )
. (2.30)
s
The (nax1) vector { fa } is

(WR)
{ fa } = { fap } + { fav } + { faZ } + { fas }

( WR ) ( WR ) ( WR ) ( WR ) ( WR )
(2.31)
with the (nax1) vectors

76
{ fap } = ! Chapter 2
"
∫ ρ ω [ Ba ] {n}( pq − p ). dΩ
( WR ) jT
, (2.32)
Ωp 0
{ fav } =
#
∫ [Φ a ] (vn − ρ ω {n} [∂ ] pq ). dΩ
#
( WR ) T T j
, (2.33)
Ωv
0
{ faZ } =
$ $ $
pq
∫ [Φ a ]
j
( WR ) T
( − {n} T [∂ ] pq ). dΩ , (2.34)
ΩZ
Z ρ 0ω
{ fas } =
( WR )
∫ [Φ a ]
T
( jωw F + jωwq −
j
ρ 0ω
{n} T [∂ ] pq ). dΩ . (2.35)
Ωs
Since the weighted residual formulation should hold for any weighting
function ~
p , the expressions between square brackets in equation (2.27)
must be zero. This yields a set of na algebraic equations in the (4+na)
unknown wave function contributions ws and pa,
[Cas ( WR )  s 
] { }
w
( WR ) ( WR )
Aaa + Caa .   = fa( WR ) . (2.36)
î pa 
structural boundary conditions

Since the four structural boundary conditions in equation (2.9) are defined at
discrete locations, they can be used as such. The substitution of the field
variable expansions (2.10) and (2.15) yields the following matrix form of
the structural boundary conditions,
w s 
[ Ass Csa ].   = { fs }. (2.37)
î pa 
The (4x4) matrix [ Ass ] results from the application of the differential
operators, which are involved with the structural boundary conditions in
equation (2.9), on the structural wave functions in the (1x4) matrix [Ψ s ] ,
 Ψ s (0) 
 Ψ ( L) 
 s 
[ ss ]  dΨ s (0)  .
A = (2.38)
 dx’ 
 dΨ s ( L ) 
 dx’ 
%
The (4xna) coupling matrix [ Csa ] results from the application of these
&'
differential operators on the particular solution functions in the (1xna) matrix
[ wa ] ,
 wa ( 0 ) 
(
)
 w ( L) 
 a 
[Csa ] = 
 dw a ( 0 ) . (2.39)
dx’ 
 dwa ( L ) 
 
 dx’ 
The (4x1) vector { fs } * +

results from the application of these differential
,. -/
operators on the particular solution functions w F and wq ,
0 1
 − w F ( 0 ) − w q ( 0) 
 − w ( L) − w ( L) 
2 3
 F q 
 dw F ( 0 ) dw q ( 0 ) 
{ s}  −
f = −  . (2.40)
 dx’ dx’ 
 dw F ( L ) dw q ( L ) 
− − 
î dx’ dx’ 
coupled vibro-acoustic model

The combination of the structural boundary conditions (2.37) and the
weighted residual formulation (2.36) of the acoustic boundary conditions
yields the coupled vibro-acoustic model
[ A ]. î pas  = {b }

( WR )
w ( WR )
(2.41)
78 Chapter 2
[ A ] = Casss   fs 
{ }
A Csa
with ( WR )
 and b ( WR ) =  ( WR )  . (2.42)
( WR )
Aaa + Caa 
( WR ) ( WR )
î fa 
Solving model (2.41) for the unknown wave function contributions ws and pa
and back-substituting the results into the expansions (2.10) and (2.15) yield
the prediction of the steady-state dynamic response of the coupled vibro-
acoustic system.
2.4.1.2. Least-squares formulation

Based on the residual error functions, defined in equations (2.21)-(2.24), a
functional Fa may be defined, which is a measure for the total approximation
error, involved with the representation of the acoustic boundary conditions,
2 2 2 2
Fa = ∫ β p . Rp . dΩ + ∫ β v . Rv . dΩ + ∫ β Z . RZ . dΩ + ∫ β s . Rs . dΩ
Ωp Ωv ΩZ Ωs
(2.43)
The parameters β p, β v, β Z and β s are introduced to restore the homogeneity of
the physical dimensions of the different components in the functional and
they allow to tune the relative importance, attributed to the different
acoustic boundary conditions.
In the least-squares formulation of the acoustic boundary conditions, the
functional Fa is minimised with respect to any acoustic wave function
contribution pa in the acoustic pressure expansion (2.10). This yields a set of
na algebraic equations in the (4+na) unknown wave function contributions ws
and pa,
[Cas ]
( LS )  s 
{ }
w
( LS )
Aaa .   = fa( LS ) . (2.44)
î pa 
(LS)
[ ]
4
The element on row ai and column s of the (nax4) matrix Cas is
 1 ∂Φ ai
* 
[ ]
Cas
ai s

= β s . ∫ (ω w ai −
( LS )

Ωs
2 *
ρ 0 ∂n
).Ψ s . dΩ 


(2.45)
where * denotes the complex conjugate.

(LS)
The element on row ai and column aj of the (naxna) matrix Aaa is [ ]
 
β v  ∂Φ ai ∂Φ a j 
*
[ ]( LS )
Aaa
ai a j
= β p .  ∫ Φ *ai .Φ a j . dΩ  +
 
. ∫
ρ 02ω 2  Ω ∂n
.
∂n
. d Ω 


Ω p  v
 Φ *ai j ∂Φ ai Φ a j
*
j ∂Φ a j 

+ βZ. ∫ ( * + − ). dΩ 
5 6
).(
 ρ 0ω ∂n Z ρ 0ω ∂ n 
Ω Z Z 
 
j ∂Φ a j
*
j ∂Φ ai
*

+ β s . ∫ ( j ωw a i −
*
).( − jωw a j +
*
). dΩ 
 ρ 0ω ∂n ρ 0ω ∂ n 
Ωs 
8
(2.46)
7 { }
The element on row ai of the (nax1) vector fa(LS) is
 
β v  ∂Φ ai ∂pq 
*
{ fa }
9:9
( LS )
= β p .  ∫ Φ *ai .( p − pq ). dΩ  + . ∫ .( j ρ ωv − ). d Ω 
ai   ρ 02ω 2  Ω ∂n
0 n
∂n 
Ω p  v 
 Φ *ai j ∂Φ ai
*
j ∂pq pq 
<

+ βZ. ∫ ( * + − ). dΩ 
; < <
).(
 ρ 0ω ∂n ρ 0ω ∂n Z 
Ω Z Z 
 j ∂Φ ai
*
j ∂pq 
+ β s .  ∫ ( jωwa*i − ).( jωw F + jωwq − ).dΩ 
 ρ 0ω ∂n ρ 0ω ∂n 
Ωs 
(2.47)

The combination of the structural boundary conditions (2.37) and the least-
squares formulation (2.44) of the acoustic boundary conditions yields the
[ A ]. î pas  = {b }

w
( LS ) ( LS )
(2.48)
[ A ] = Casss { }
A Csa   fs 
with ( LS )
( LS ) ( LS )  and b
( LS )
=  ( LS )  . (2.49)
Aaa  î fa 
80 Chapter 2
Solving model (2.48) for the unknown wave function contributions ws and pa
and back-substituting the results into the expansions (2.10) and (2.15) yield
the prediction of the steady-state dynamic response of the coupled vibro-
acoustic system.
In comparison with the weighted residual model (2.41), the least-squares

model (2.48) has two disadvantageous features.
• The convergence depends on the parameters β p, β v, β Z and β s, for which it

is not straightforward to specify a priori some adequate values.
• The convergence of a least-squares model is slower than a corresponding
weighted residual model.
These features will be illustrated in section 2.7.2.3.
2.4.2. Model properties
From the above definitions of the matrices [A ] (WR)

and [A ]
(LS)
in the
proposed coupled vibro-acoustic models (2.41) and (2.48), several model
properties may be identified.
• Since the acoustic and structural wave functions Φ a and Ψ s are defined
in, respectively, the entire acoustic domain V and the entire structural
domain Ωs, the model matrices are fully populated.
• Since the acoustic and structural wave functions Φ a and Ψ s are
complex functions and since they are implicitly dependent on frequency
by virtue of the frequency dependence of their wavenumber components,
the model matrices are complex and frequency dependent and cannot be
partitioned into frequency independent submatrices.
• The model matrices are not symmetric. However, the submatrices Aaa
(WR)
[ ]
[ ]
(LS)
and Aaa are, respectively, symmetric and Hermitian, as proven below.
From the definition formula (2.28), it follows that the element on row ai
[ (WR)
and column aj of the (naxna) matrix Aaa ]
is
∂Φ a j
[ Aaa ]a a
( WR )
i j
= ∫
Ω v +Ω Z +Ω s
ρ
j
0 ω
.Φ a i .
∂n
. dΩ
(2.50)
j ∂Φ ai 1
− ∫ . . Φ a j . dΩ − ∫ . Φ a i . Φ a j . dΩ
Ωp
ρ 0 ω ∂n ΩZ
Z
For any two functions ϕ and ψ, which are sufficiently smooth and non-
singular in a domain V, enclosed by a surface Ωa, Green’s second identity
states that (see e.g. GREENBERG (1998))
∂ψ ∂ϕ
∫ (ϕ . ∂n − ψ . ∂n ). dΩ = ∫ (ϕ . ∇ ψ − ψ . ∇ ϕ ). dV ,
2 2
(2.51)
Ω a V
where the direction n, normal to the boundary surface Ωa, has a positive
orientation away from domain V.
Since the acoustic wave functions Φ ai and Φ a j are smooth, non-
singular solutions of the homogeneous Helmholtz equation, i.e.
∇ 2Φ ai = − k 2 .Φ ai and ∇ 2Φ a j = − k 2 .Φ a j , (2.52)
Green’s second identity implies that
j ∂Φ a j j ∂Φ ai
∫ ( ρ ω .Φ ai . ∂n −
ρ 0ω
.Φ a j .
∂n
). dΩ = 0 . (2.53)
Ω a
0
Since Ωa = Ω p ∪ Ωv ∪ Ω Z ∪ Ωs , it follows from equation (2.53) that
j ∂Φ a j j ∂Φ ai
∫ .Φ ai . . dΩ − ∫ . . Φ a j . dΩ
ρ ω
Ω v +Ω Z + Ω s 0
∂n ρ ω ∂n
Ωp 0 (2.54)
j ∂Φ ai j ∂Φ a j
= ∫ .Φ a j . . dΩ − ∫ . . Φ ai . dΩ .
ρ
Ω v +Ω Z +Ω s 0
ω ∂ n ρ
Ωp 0
ω ∂n
82 Chapter 2
Since
1 1
∫ . Φ a i . Φ a j . dΩ = ∫ . Φ a j . Φ a i . dΩ , (2.55)
ΩZ
Z ΩZ
Z
the comparison between equation (2.50) and equations (2.54) and (2.55)
proves that matrix Aaa [ ]
(WR)
is symmetric, i.e.
[ Aaa ]a a = [ Aaa ]a a
( WR )
i j
( WR )
j i
(∀ ai , a j = 1.. na ) . (2.56)
(LS)
From its element definition (2.46), one can readily see that matrix Aaa [ ]
is Hermitian, i.e.
[ Aaa ]a a = [ Aaa ]a a
*
( LS ) ( LS )
(∀ ai , a j = 1.. na ) . (2.57)
i j j i
Note that, from these properties of the acoustic submatrices, it follows

that the application of the wave based prediction technique for an
uncoupled interior acoustic problem, yields a symmetric or a Hermitian
model.
Note also that, for a three-dimensional coupled vibro-acoustic problem,
in which the structural boundary conditions are no longer defined at
some discrete locations and are also transformed into a weighted residual
or least-squares formulation, the structural submatrix [ Ass ] is also
symmetric or Hermitian, respectively (see chapter 4).
2.5. Convergence requirements
2.5.1. Introductory considerations
Recall from section 2.3.1. that the acoustic pressure is approximated as

= >
na
p ( x , y ) ≈ p( x , y ) = ∑ p a . Φ a ( x , y ) + pq ( x , y ) , (2.58)
a=1
where the wavenumber components kxa and kya of each acoustic wave
function
− j ( k xa . x + k ya . y )
Φ a ( x, y) = e (2.59)
satisfy 2
k xa + k ya
2
= k2. (2.60)
Since equation (2.60) has an infinite number of real and complex solutions,
the key question is whether a set of acoustic wave functions can be selected
from the infinite number of possible wave functions, such that the
subsequent pressure expansion converges towards the exact solution. Note
that, for the considered two-dimensional case, no selection must be made
regarding the structural wave functions, since the dynamic plate equation
(2.3) has only a finite number of homogeneous, linear independent solution
functions, which are all comprised in the displacement expansion (2.15).
Only for three-dimensional problems (see chapter 4), both an acoustic and a
structural wave function selection is needed.
A necessary condition for convergence of the proposed models (2.41) and
(2.48) is that the predicted normal fluid velocity distribution on the
boundary surface Ωa of the considered bounded acoustic domain V
?
converges towards the exact normal fluid velocity distribution, i.e.
∀( x , y ) ∈ Ω a :
 n a jpa ∂Φ a ( x, y )  j ∂ pq ( x , y ) j ∂p( x, y ) (2.61)
lim  ∑ .  + . = . .
ρ
n a →∞ a = 1 0 ω ∂ n  ρ 0 ω ∂ n ρ 0ω ∂n
Since the Helmholtz equation is a second-order partial differential equation,

any pressure field in a bounded domain is uniquely characterised by its
governing Helmholtz equation, in combination with the specification of one
acoustic quantity, i.e. the normal fluid velocity or the pressure, at each
position on the boundary surface of the domain. Hence, since the proposed
pressure expansion (2.58) a priori satisfies the governing Helmholtz
equation, condition (2.61) is also a sufficient condition for convergence.
84 Chapter 2
As a result, the convergence is automatically ensured, if a set of acoustic

wave functions Φ a is used, for which the wavenumber components kxa and
kya satisfy equation (2.60), and which are able to represent a homogeneous
pressure field with an arbitrary normal fluid velocity distribution on the
boundary surface of the considered domain.
Sections 2.5.2 and 2.5.3. prove3 that such a complete set of acoustic wave
functions may be defined for rectangular and for non-rectangular domains,
respectively. Section 2.5.4. indicates how these wave functions are
implemented into a convergent coupled vibro-acoustic wave model.
2.5.2. Rectangular continuum domain
Figure 2.2. shows a rectangular domain with dimensions Lx and Ly, in which
the homogeneous pressure field p(x,y) is requested, whose normal fluid
velocity distribution is characterised by four arbitrary functions v1, v2, v3 and
v4, each of them being defined on one of the four side walls of the
rectangular boundary surface.
Figure 2.2 : general homogeneous pressure field in a rectangular domain
3
although the proofs need some refinements to become entirely sound from a pure
mathematical point of view, they certainly serve as ‘engineering proofs’
The considered problem may be split into four subproblems, each of them
having three side walls with zero normal fluid velocity and one side wall
with a non-zero normal fluid velocity distribution, as shown in figure 2.2.
One of these subproblems consists of determining the pressure field p1(x,y),
satisfying the homogeneous Helmholtz equation
∇ 2 p1 ( x , y ) + k 2 . p1 ( x , y ) = 0 (2.62)
and the boundary conditions
∂p1 ( x ,0 )
= 0, ( x ∈ [ 0, L x ]), (2.63)
∂y
∂p1 ( x, L y )
= 0, ( x ∈ [ 0, L x ]), (2.64)
∂y
j ∂p1 ( 0, y )
− = v1 ( y ), ( y ∈ [ 0, Ly ]), (2.65)
ρ 0ω ∂x
∂p1 ( L x , y )
= 0, ( y ∈ [ 0, L y ]). (2.66)
∂x
By separation of variables, the pressure field
p1 ( x, y ) = X ( x ). Y ( y ) (2.67)
is governed by two ordinary differential equations
d2X
2
+ ( k 2 − λ2 ). X = 0 , (2.68)
dx
d 2Y
2
+ λ2 . Y = 0 . (2.69)
dy
The linear independent solutions of these differential equations yield a

pressure field of the form
p1 ( x, y ) = ( A.cos( λy ) + B.sin( λy )).(C. e − j k − λ . x + D. e j k − λ . x )

2 2 2 2
(2.70)
+ ( E. y + F ).(G. e − jkx + H. e jkx )
86 Chapter 2
where A,...,H are arbitrary constants and λ is an arbitrary non-zero constant.

For the boundary condition (2.63) to be satisfied, these constants must be
defined, such that the equation
k 2 − λ2 . x
λ. B.(C. e − j + D. e j k − λ . x ) + E.(G. e − jkx + H. e jkx ) = 0 (2.71)
2 2
is satisfied at any position x ∈ [0, L x ] . Hence, B=0 and E=0, yielding a

pressure field of the form
p1 ( x , y ) = cos(λy ).( I. e − j k − λ . x + J . e j k − λ . x ) ,
2 2 2 2
(2.72)
where I and J are arbitrary constants and λ is zero or an arbitrary non-zero

constant.
For the boundary condition (2.66) to be satisfied, these constants must be
defined, such that the equation
j. k 2 − λ2 .cos( λy ).( − I. e − j k − λ . L x + J . e j k − λ . L x ) = 0
2 2 2 2
(2.73)
is satisfied at any position y ∈ [0, L y ] . Hence,
I = J . e 2 j k − λ . Lx ,
2 2
(2.74)
yielding a pressure field of the form
p1 ( x, y ) = cos(λy ).( J . e 2 j k − λ . L x . e − j k − λ . x + J . e j k − λ . x ) . (2.75)

2 2 2 2 2 2
For the boundary condition (2.64) to be satisfied, the constants J and λ must
be defined, such that the equation
k 2 − λ2 . Lx − j k 2 − λ2 . x
+ J. e j k − λ .x ) = 0
2 2
λ.sin( λL y ).( J . e 2 j e (2.76)
is satisfied at any position x ∈ [0, L x ] . Since J=0 would yield the trivial
solution p1(x,y)=0, λ is confined to
a .π
sin( λ . L y ) = 0 → λ = 2 ( a 2 = 0,1,...) . (2.77)
Ly
This yields a pressure field of the form
p1 ( x , y ) =
a 2π 2 a 2π 2 a 2π 2
∞ 2 j k 2 −( ) . Lx − j k 2 −( ) .x j k 2 −( ) .x
a πy Ly Ly Ly
∑ pa 2 .cos( 2 ).(e
Ly
.e +e )
a2 = 0
(2.78)
For the remaining boundary condition (2.65) to be satisfied, the constants
pa 2 must be defined, such that the equation
a 2π 2
∞ 2 j k 2 −( ) . Lx
1 a π Ly a πy
v1 ( y ) = ∑ pa 2 . k 2 − ( 2 ) 2 .(1 − e
ρ 0ω a = 0 Ly
− 1).cos( 2 )
Ly
2
(2.79)
a 2πy
is satisfied at any position y ∈ [0, L y ] . Since the functions cos( )
Ly
(a2=0,1,...) are the orthogonal eigenfunctions of the Sturm-Liouville problem
d 2Y dY ( 0 ) dY ( L y )
2
+ λ2 . Y = 0 with = 0, = 0, (2.80)
dy dy dy
the constants pa 2 in equation (2.78) can be determined, such that the series
expansion in the right-hand side of equation (2.79) converges to v1(y) at the
positions y ∈]0, L y [ , where v1(y) is continuous, and to the mean value
0.5(v1(y-)+v1(y+)) at the positions y ∈]0, L y [ , where v1(y) is discontinuous
(see e.g. GREENBERG (1998)).
In a completely similar way, it can be proven that a homogeneous pressure
field p2(x,y), having a normal fluid velocity distribution v2(y) along the side
wall at x=Lx and a zero normal fluid velocity along the three other side
walls, may be expressed as
88 Chapter 2
a 2π 2 a 2π 2
∞ − j k 2 −( ) .x j k 2 −( ) .x
a πy Ly Ly
p 2 ( x, y ) = ∑ pa 2 .cos( 2 ).(e +e ) .(2.81)
a =0 Ly
2
The general forms of the pressure fields p3(x,y) and p4(x,y), shown in figure
2.2, are obtained by interchanging the position variables x and y and the
dimensions Lx and Ly in equations (2.78) and (2.81).
As a result, the requested homogeneous pressure field p(=p1+p2+p3+p4) in a
rectangular domain with dimensions Lx and Ly and with an arbitrary normal
fluid velocity distribution on the boundary surface may be expressed in
terms of the following acoustic wave function set,
∞ ∞
p( x , y ) = ∑ pa 1 .Φ a 1 ( x , y ) + ∑ pa 2 .Φ a 2 ( x , y )
a1 = 0 a2 = 0
(2.82)
∞ − jk ya1 y ∞ − jk xa2 x
= ∑ pa 1 .cos( k xa1 x ). e + ∑ pa 2 . e .cos( k ya 2 y )
a1 = 0 a2 = 0
with
a .π a .π
( k xa 1 , k ya 1 ) = ( 1 ,± k 2 − ( 1 ) 2 ) ,
Lx Lx
(2.83)
a .π a .π
( k xa 2 , k ya 2 ) = ( ± k − ( 2 ) 2 , 2 ) .
2
Ly Ly
Hence, according to the introductory convergence considerations in section

2.5.1., it can be concluded that, for the case of a coupled vibro-acoustic
system with a rectangular acoustic domain, the proposed wave models
(2.41) and (2.48) are convergent, if the acoustic wave functions with
wavenumber components, defined in (2.83), are used in the pressure
expansion (2.58).
2.5.3. Non-rectangular continuum domain
For the case of a coupled vibro-acoustic system with a non-rectangular

acoustic domain, the proposed wave models (2.41) and (2.48) are
convergent, if a set of acoustic wave functions is used, which can represent a
homogeneous pressure field p with an arbitrary normal fluid velocity
distribution on the boundary surface Ωa of the considered non-rectangular

domain V.
For the construction of this complete wave function set, an additional
domain V’ is defined (see figure 2.3.), such that the boundary surface Ωa of
domain V is completely comprised in the boundary surface of domain V’
and that the boundary surface of the combined domain V∪V’ is a
circumscribing rectangle of the boundary surface Ωa.
Figure 2.3 : extension of the continuum domain
If it is possible to define, for any possible homogeneous pressure field p in

domain V, a homogeneous pressure field ψ in domain V’, such that both
fields have the same pressure and normal fluid velocity distribution on the
common boundary surface Ωa, it may be concluded that any homogeneous
pressure field in V is part of a homogeneous pressure field, defined in the
larger, but rectangular domain V∪V’. Since it is proven in the previous
section that, for this rectangular domain, a complete wave function set may
be defined, this wave function set is also able to represent any homogeneous
pressure field in the enclosed, non-rectangular domain V.
To prove the existence of the homogeneous pressure field ψ in domain V’,
an additional domain V’’ is defined (see shaded region in figure 2.4), which
consists of several rectangular subdomains and contains the entire domain
V’ and some part of domain V. Its rectangular subdomains, which are
denoted with number 1 in figure 2.4, are defined, such that two adjacent
90 Chapter 2
sides of their rectangular boundary belong to domain V, while their other

two boundary sides belong to domain V’.
Figure 2.4 : definition of domain V’’
In each of the rectangular subdomains, denoted with number 1, a

homogeneous pressure field ψ1 is considered (see figure 2.5), such that its
pressure and normal fluid velocity distribution on the two adjacent boundary
sides, which belong to domain V, equal the corresponding pressure and
normal fluid velocity distribution of the arbitrary homogeneous pressure
field p, considered in domain V.
Figure 2.5 : initial-value problem

By imposing both pressure and normal fluid velocity continuity on one half
of the boundary surface, instead of imposing one boundary condition on
each position of the entire boundary surface, an initial-value problem is
obtained (see e.g. GREENBERG (1998)), which completely and uniquely
defines each homogeneous pressure field ψ1.
In each of the rectangular subdomains, which are denoted with number 2
and share two adjacent boundary sides with two neighbouring rectangular
domains, denoted with number 1, a homogeneous pressure field ψ2 is
completely and uniquely defined by an initial-value problem, imposing on
the two shared boundary sides the pressure and normal fluid velocity
continuity with the homogeneous pressure fields ψ1, which are previously
defined in those two neighbouring domains. In a completely similar way, a
homogeneous pressure field ψn+1 in each rectangular subdomain, denoted
with number n+1, is defined by an initial-value problem, imposing the
pressure and normal fluid velocity continuity with two neighbouring
homogeneous pressure fields ψn.
Consider now the particular situation of a domain V’’, which is defined,
such that each rectangular subdomain, denoted with number n, shares two
adjacent boundary sides with two rectangular subdomains, denoted with
number n-1, and that its other two adjacent boundary sides are shared with
two rectangular subdomains, denoted with number n+1. For this situation,
the combination of all homogeneous pressure fields, each of them being
defined in one rectangular subdomain, yields a global pressure field, which
is defined in the entire domain V’’ and which is also homogeneous, since the
continuity conditions for both the pressure and the normal fluid velocity
have been imposed on all four boundary sides of each rectangular
subdomain. Moreover, its pressure and normal fluid velocity distribution on
the boundary sides, which belong to domain V, equal the corresponding
pressure and normal fluid velocity distribution of the arbitrary homogeneous
pressure field p, considered in domain V. Hence, in the limit of a
discretization of domain V’’ into infinitesimally small rectangular
subdomains, the domain V’’ becomes identical to domain V’ and the desired
homogeneous pressure field ψ is obtained.
One can readily see that the considered particular domain situation occurs
for any convex domain V, as it is the case, for instance, for the domain,
shown in figure 2.4. However, when the domain V is not convex, it is
possible that the global pressure field is not homogeneous, as it is the case,
for instance, for the domain, shown in figure 2.6(a). This is due to the fact
that the various subdomain pressure fields can only be completely and
92 Chapter 2
uniquely defined without imposing the necessary continuity conditions for

the pressure and normal fluid velocity on the boundary sides, located in the
shaded region in figure 2.6(a). Figure 2.6(b) shows, however, that the
domain V may be decomposed into some smaller subdomains, for each of
which a homogeneous pressure field ψ can be defined. Although its
mathematical proof is still lacking, the author cannot think of a situation,
counterproving the statement that any possible domain can be decomposed
into some smaller subdomains, for each of which a homogeneous pressure
field ψ can be defined4.
Figure 2.6 : domain decomposition
As a result, it may be concluded that, for any two-dimensional coupled

vibro-acoustic system, a convergent wave model is obtained
• by decomposing the acoustic domain into some convex subdomains or at
least into some subdomains, such that an arbitrary homogeneous pressure
field, defined in each subdomain, may be considered as part of a
homogeneous pressure field, defined in a rectangular domain,
circumscribing the subdomain
• and by approximating the pressure field in each subdomain in terms of a
pressure expansion (2.58), in which the complete acoustic wave function
set (2.83), associated with the circumscribing rectangular domain, is
used.
4
note that MORSE and FESHBACH (1953) utilised a similar idea, in that they
proposed to express the mode shapes of a slightly non-rectangular acoustic cavity
in terms of the mode shapes of the circumscribing rectangular cavity, and
MISSAOUI and CHENG (1997) successfully extended this approach for the
prediction of the acoustic properties of general, non-rectangular cavities
2.5.4. Wave model implementation
This section describes how the complete wave function sets, derived in the
previous section, are implemented into the weighted residual and the least-
squares wave models, derived in section 2.4.1. For the sake of simplicity, a
two-dimensional interior coupled vibro-acoustic system is considered (see
figure 2.7), for which complete wave function sets are obtained by
decomposing the acoustic domain into two subdomains and in which the
structural component and the external acoustic excitation occur in one
subdomain only. For the general case of a domain decomposition into
several subdomains, each of them containing a part of the structural
component, the implementation is performed in a completely similar way.
Figure 2.7 : domain decomposition of an interior coupled vibro-acoustic system
field variable expansions

By using the complete wave function set, defined in (2.82) and (2.83), a

convergent approximation of the steady-state pressure p1(x1,y1) in the first
subdomain of the acoustic cavity is obtained, i.e.

p1 ( x 1 , y1 )
n 1,a
= ∑ p1, a .Φ 1, a ( x1 , y1 ) + pq ( x 1 , y1 ) = [Φ 1, a ]{ p1, a } + pq ( x 1 , y1 ) (2.84)
a=1
n 1,a1 n 1,a2
= ∑ p1, a 1 .Φ 1, a 1 ( x 1 , y1 ) + ∑ p1, a 2 .Φ 1, a 2 ( x1 , y1 ) + pq ( x 1 , y1 )
a1 = 0 a2 = 0
94

where pq is a particular solution function, defined in (2.13), and where
Chapter 2
− j. k 1, ya1 . y 1
Φ 1, a 1 ( x 1 , y1 ) = cos( k1, xa 1 . x 1 ). e
(2.85)
− j. k 1, xa2 . x 1
Φ 1, a 2 ( x 1 , y1 ) = e .cos( k1, ya 2 . y1 )
with
a .π a .π
( k1, xa1 , k1, ya 1 ) = ( 1 ,± k 2 − ( 1 ) 2 )
Lx 1 Lx 1
(2.86)
a .π a .π
( k1, xa 2 , k1, ya 2 ) = ( ± k − ( 2 ) 2 , 2 )
2
Ly1 Ly1
and where Lx1 and Ly1 are the dimensions of a rectangular domain,

circumscribing the first cavity subdomain.
A convergent approximation of the steady-state pressure p2(x2,y2) in the
second subdomain of the acoustic cavity is obtained with the expansion
n 2,a
p2 ( x 2 , y2 ) = ∑ p2, a .Φ 2, a ( x 2 , y2 ) = [Φ 2, a ]{ p2, a }
a=1
n 2,a1 n 2,a2
(2.87)
= ∑ p2, a1 .Φ 2, a1 ( x 2 , y2 ) + ∑ p2, a 2 .Φ 2, a 2 ( x 2 , y2 )
a1 = 0 a2 = 0
where
− j. k 2, ya1 . y 2
Φ 2, a 1 ( x 2 , y 2 ) = cos( k 2, xa1 . x 2 ). e
(2.88)
− j. k 2, xa2 . x 2
Φ 2, a 2 ( x 2 , y 2 ) = e .cos( k 2, ya 2 . y 2 )
with
a .π a .π
( k 2, xa 1 , k 2, ya 1 ) = ( 1 ,± k 2 − ( 1 ) 2 ) ,
Lx 2 Lx 2
(2.89)
a .π a .π
( k 2, xa 2 , k 2, ya 2 ) = ( ± k − ( 2 ) 2 , 2 ) .
2
Ly 2 Ly 2
and where Lx2 and Ly2 are the dimensions of a rectangular domain,
circumscribing the second cavity subdomain.
Note that, although the complete wave function sets, defined in any possible
circumscribing rectangular domain, may be used for the pressure
expansions, it is beneficial for the convergence rate of the subsequent wave
model to select, for each subdomain, its smallest circumscribing rectangular
domain, as will be illustrated in section 2.7.2.2.

Since it is assumed that the structural component is entirely comprised in
the boundary surface of the first cavity subdomain, a convergent expansion

for the steady-state normal plate displacement w(x’) is
4 n 1,a
w( x’) = ∑ ws .Ψ s ( x’) + ∑ p1, a . w1, a ( x’) + w F ( x’) + wq ( x’)

s=1 a=1 (2.90)
[ ]{ }
= [Ψ s ]{w s } + w1, a p1, a + w F ( x’) + wq ( x’)
where the structural wave functions Ψ s and the particular solution functions
w F and wq are defined in (2.16), (2.17) and (2.20), respectively. For the
particular solution functions w1,a , expressions of type (2.19) cannot be
used, since the dependence of each acoustic wave function Φ1,a on either

the x1 or the y1 co-ordinate is no longer expressed by a complex exponential,
but by a cosine function. Therefore, the following expression is used in
(2.90),

 p1, a 1 
[ w1, a ]{ p1, a } = [ w1, a ]
w1, a 2 .   (2.91)
î p1, a 2 
1
[ ]
where each function w1,a1 (x’) in the (1x(2(n1,a1+1))) matrix w1,a1 is
( Ba 1 ,1 − kb4 ).cos( k1, xa1 ( x 0 + x’.cos α )) − jk 1, ya ( y 0 + x’.sin α )

w1, a 1 ( x’) = .e 1
D.[( Ba ,1 − kb4 ) 2 + Ba21 , 2 ]

1
Ba , 2 .sin( k1, xa 1 ( x 0 + x’.cos α )) − jk 1, ya ( y 0 + x’.sin α )

− 1 .e 1
D.[( Ba ,1 − kb4 ) 2 + Ba21 , 2 ]
1
(2.92)
and where each function w1,a 2 (x’) in the (1x(2(n1,a2+1))) matrix w1,a 2 is [ ]

96
w1, a 2 ( x’) =
D.[( Ba ,1 − k b ) + Ba 2 , 2 ]
4 2
2
2
.e 2
Chapter 2
( Ba 2 ,1 − kb4 ).cos( k1, ya 2 ( y0 + x’.sin α )) − jk 1, xa ( x 0 + x’.cos α )
Ba 2 , 2 .sin( k1, ya 2 ( y0 + x’.sin α )) − jk 1, xa ( x 0 + x’.cos α )

− .e 2
D.[( Ba ,1 − kb ) + Ba 2 , 2 ]
4 2 2
2
(2.93)
with coefficients Bi,1 and Bi,2 (i=a1,a2),
1
Bi,1 = [( k1, xi .cos α + k1, yi .sin α ) 4 + ( − k1, xi .cos α + k1, yi .sin α ) 4 ],
2
(2.94)
j
Bi, 2 = [( − k1, xi .cos α + k1, yi .sin α ) 4 − ( k1, xi .cos α + k1, yi .sin α ) 4 ].
2
weighted residual model

Due to the decomposition of the cavity domain, some additional acoustic
boundary conditions must be specified, in that the continuity of both the
pressure and fluid normal velocity must be satisfied on the common
boundary interface Ωc12 between both subdomains.
Assuming that the continuity condition of the fluid normal velocity on Ωc12
is imposed on the pressure field p1 in the first cavity subdomain, the
procedure of transforming the acoustic boundary conditions on p1 into a
weighted residual formulation, as outlined in section 2.4.1.1., yields a set of
n1,a algebraic equations,
 ws 
[C1, as
( WR )
A1, aa + C1, aa
( WR ) ( WR ) 
( WR )
p 
] 
{ }
A1, 2  p1, a  = f1(,WR
a .
)
(2.95)
2, a 
(WR)
The (n1,axn1,a) matrix A1,aa is [ ]
[ A1,aa ] = Ω
( WR )
∫
+ Ω Z1 + Ω s + Ω c12
j
ρ 0ω
[ T T
]
Φ 1, a {n1} [∂ 1 ] Φ 1, a . dΩ [ ]
v1
∫ ρ ω ([∂ 1 ][Φ 1, a ]) {n1}[Φ 1, a ]. dΩ − ∫ [ ][ ]

j T 1 T
− Φ 1, a Φ 1, a . dΩ ,
Ωp 0 ΩZ
Z
1 1
(2.96)
and the (n1,axn2,a) matrix A1,2 [ (WR)
] is
[ A1,2 ] = Ω ∫
( WR ) j
ρ 0ω
[
T T
]
Φ 1, a {n2 } [∂ 2 ] Φ 2, a . dΩ . [ ] (2.97)
c12
The (2x1) vectors {n1} and {n 2 } contain, respectively, the x1- and y1-
component of the unit normal vector with positive orientation away from the
first cavity subdomain and the x2- and y2-component of the unit normal
vector with positive orientation away from the second cavity subdomain.
The (2x1) vectors [∂ 1 ] and [∂ 2 ] contain the gradient operators, defined in,

respectively, the (x1,y1) and the (x2,y2) co-ordinate systems.
(WR)
The (n1,axn1,a) matrix C1,aa is [ ]
[C1,aa ] = −Ω∫ jω[Φ 1,a ]T [ w1,a ]. dΩ
( WR )
(2.98)
s
(WR)
and the (n1,ax4) matrix C1,as is [ ]
[C1,as ] = −Ω∫ jω[Φ 1,a ] T [Ψ s ]. dΩ
( WR )
. (2.99)
s
The (n1,ax1) vector { f1,a } is

(WR)
{ f1,a } = { f1,ap } + { f1,av } + { f1, aZ } + { f1,as } + { f1,2 }

( WR ) ( WR ) ( WR ) ( WR ) ( WR ) ( WR )
(2.100)
with the (n1,ax1) vectors

98 Chapter 2
{ f } = Ω∫
( WR ) j
[ ( [
∂ 1 ] Φ 1, a {n1}( pq − p ). dΩ ,
T
])

(2.101)
1, ap
ρ 0ω

p1
{ f1,av } = Ω∫ [Φ 1,a ]T (vn − ρ 0jω {n1}T [∂ 1 ] pq ). dΩ

( WR )
, (2.102)

v1
{ f1,aZ } = Ω∫ [Φ 1,a ]
pq
{n1} [∂ 1 ] pq ). dΩ ,
T j T
( WR )
( − (2.103)
Z ρ 0ω
Z1
{ f1,as } = ∫ [Φ 1,a ]T ( jωwF + jωwq − ρ 0jω {n1}T [∂ 1 ] pq ). dΩ

( WR )
, (2.104)
Ωs
{ f1,2 } = −Ω ∫
( WR ) j
ρ 0ω
T
[T
]
Φ 1, a {n1} [∂ 1 ] pq ). dΩ . (2.105)
c12
The transformation of the acoustic boundary conditions on the pressure field

p2, including the continuity condition of the pressure on Ωc12 , into a
weighted residual formulation yields a set of n2,a algebraic equations,
 ws 
[0 A2,1( WR ) 
( WR )
p 
] 
A2, aa  p1, a  = f 2(,WR )
a . { } (2.106)
2, a 
(WR)
The (n2,axn2,a) matrix A2,aa is [ ]
[ A2,aa ] = Ω
( WR )
∫
+ Ω Z2
j
ρ 0ω
T
[ T
]
Φ 2, a {n 2 } [∂ 2 ] Φ 2, a . dΩ [ ]
v2
j
[ ( [ ])
∂ 2 ] Φ 2, a {n 2 } Φ 2, a . dΩ − ∫
1
[ ] [ ][ ]
T T
− ∫ Φ 2, a Φ 2, a . dΩ
Ω p2 + Ω c12
ρ 0ω Ω
Z
Z2
(2.107)
and the (n2,axn1,a) matrix A2,1 [ (WR)
] is
[ A2,1 ] = Ω∫
( WR ) j
ρ 0ω
[ ( [
∂ 2 ] Φ 2,a {n 2 } Φ 1,a . dΩ .
T
]) [ ] (2.108)
c12
The (n2,ax1) vector { f2,a } is

(WR)
{ f2,a } = { f2,ap } + { f2,av } + { f2,1 }

( WR ) ( WR ) ( WR ) ( WR )
(2.109)
with the (n2,ax1) vectors
{ f } = −Ω∫
( WR ) j
ρ 0ω
[( [
T
])
∂ 2 ] Φ 2, a {n 2 } p. dΩ , (2.110)

2, ap
p2
{ f2,av } = Ω∫ [Φ 2,a ] T vn . dΩ
( WR )
, (2.111)
v2
{ f2,1 } = −Ω ∫
( WR ) j
ρ 0ω
[( [
T
])
∂ 2 ] Φ 2, a {n 2 } pq . dΩ . (2.112)
c12
The substitution of the field variable expansions (2.84) and (2.90) into the
structural boundary conditions (2.9) yields a set of 4 equations,
 ws 
 
[ Ass Cs 1, a ]
0  p1, a  = { fs } , (2.113)
p 
2, a 
where the (4x4) matrix [ Ass ] and the (4x1) vector { fs }

are defined in
(2.38) and (2.40). The (4xn1,a) coupling matrix Cs1,a [ ] results from the
!
application of the differential operators, involved with the structural
boundary conditions, on the particular solution functions in the (1xn1,a)
[
matrix w1,a , ]
"
 w1, a ( 0 ) 
 w ( L) 
 1, a 
[ ]
Cs 1, a =  dw 1,
 dx’ 
a ( 0 ) . (2.114)
 dw ( L ) 
 1, a 
 dx’ 
100 Chapter 2
The combination of the matrix equations (2.95), (2.106) and (2.113) yields
the convergent wave model
 A Cs 1, a 0   ws   fs 
ss
 ( WR ) 
   ( WR ) 
C1, as , aa + C1, aa , 2   p1, a  =  f1, a  .
)
A1( WR ) ( WR )
A1( WR (2.115)
 ) 
p   ( WR ) 
 0 A2( WR
,1
)
A2( WR
, aa  2, a   f 2, a  
Since it follows from the discussion in section 2.4.2. that
[ A1,aa ] = [ A1,aa ] , [ A2,aa ] = [ A2,aa ]

( WR ) ( WR )
T ( WR ) ( WR )
T
, (2.116)
and since it follows from a comparison between (2.97) and (2.108) that
[ A1,2 ] = [ A2,1 ]
( WR ) ( WR )
T
, (2.117)
a weighted residual model for an uncoupled interior acoustic problem is

symmetric, even if a domain decomposition is required for convergence.
least-squares model
When a least-squares approach is used, as outlined in section 2.1.4.2., an
error functional Fa may be defined, which is a measure for the total
approximation error, involved with the representation of the acoustic
boundary conditions in both cavity subdomains,
2 2 2
Fa = ∫ β p . R p1 . dΩ + ∫ β v . Rv 1 . dΩ + ∫ β Z . RZ 1 . dΩ
Ω p1 Ω v1 Ω Z1
2 2 2
+ ∫ β p . Rp2 . dΩ + ∫ β v . Rv 2 . dΩ + ∫ β Z . RZ 2 . dΩ (2.118)
Ω p2 Ω v2 Ω Z2
2 2 2
+ ∫ β s . Rs . dΩ + ∫ β p . R p12 . dΩ + ∫ β v . Rv 12 . dΩ .
Ωs Ω c12 Ω c12
The minimisation of the functional Fa with respect to any acoustic wave

function contribution p1,a in the expansion (2.84) of the pressure field p1
yields a set of n1,a algebraic equations,
 ws 
[C1, as
( LS ) ( LS )
A1, aa

( LS )
p 
]
{ }
A1, 2  p1, a  = f1(,LSa ) . (2.119)
2, a 
#
(LS)
The element on row ai and column s of the (n1,ax4) matrix C1,as is [ ]
 1 ∂Φ 1, ai
* 
[ C1( ,LSas) ] ai s
= β s .  ∫ (ω 2 w1*, ai −

Ω s ρ 0 ∂n
).Ψ s . dΩ  .


(2.120)
(LS)
The element on row ai and column aj of the (n1,axn1,a) matrix A1,aa is [ ]
[ A1,aa ]a a
( LS )
i j
=
   ∂Φ *1, ai ∂Φ 1, a j 
  βv  
β p . Φ . Φ
∫ 1, a 1, a j
*
. d Ω + .
 ρ 2ω 2  ∫ . . d Ω 
 Ω +Ω i ∂ n ∂ n
 p 1 c12  0  Ω v1 + Ω c12 
$ %
 
Φ *1, a j ∂Φ 1, ai Φ 1, a j j ∂Φ 1, a j
*
+ β Z .  ∫ ( * i + ).( − ). dΩ 
ρ 0ω ∂n Z ρ 0ω ∂n
 Ω Z1 Z 
 
j ∂Φ 1, a j
*
j ∂Φ 1, ai
*

+ β s . ∫ ( jωw1, ai −
*
).( − jωw1, a j +
*
). dΩ 
 ρ 0ω ∂n ρ 0ω ∂n 
Ω s 
(2.121)
(LS)
and the element on row ai and column aj of the (n1,axn2,a) matrix A1,2 is [ ]
   ∂Φ *1, ai ∂Φ 2, a j 
[ ]( LS )
A1, 2
ai a j
= − β p .  ∫ Φ 1, ai .Φ 2, a j . dΩ  −
 * βv 
2 2  ∫
.
ρ 0ω  Ω ∂n
.
∂n
. dΩ 
 Ω c12  c12 
(2.122)
(LS)
The element on row ai of the (n1,ax1) vector f1,a is { }
102
&'& Chapter 2
( ) ) )
 Φ *1, ai 
j ∂Φ 1, ai j ∂ pq p q
*
{ } ( LS )
f1, a
ai

= βZ ∫ ( * +
ρ 0 ω ∂n
).(
ρ 0ω ∂n
−
Z
). dΩ 
 Ω Z1 Z 
*
 j ∂Φ 1, ai
*
j ∂pq 

+ β s ∫ ( jωw1, ai −
*
).( jωw F + jωw q − ). dΩ 
 ρ ω ∂n ρ ω ∂n 
Ωs 0 0 
+
 ∂Φ *1, ai   ∂Φ *1, ai ∂pq 
jβ v  − βv 
+  ∫ ∂n . v . d Ω  ∫ . . dΩ  (2.123)
ρ 02ω 2  Ω v
n
ρ 0ω  Ω v + Ω c12
∂n ∂n
1  1 
   
  
+ β p  ∫ Φ 1, ai . p. dΩ  − β p  
∫ Φ 1, ai . pq . dΩ  .
* *
 Ω p1   Ω p1 + Ω c12 
The minimisation of the functional Fa with respect to any acoustic wave

function contribution p2,a in the expansion (2.87) of the pressure field p2
yields a set of n2,a algebraic equations,
 ws 
[0 ( LS )
A2,1

] 
{ }
A2, aa  p1, a  = f 2(,LSa) .
( LS )
p 
(2.124)
2, a 
(LS)
The element on row ai and column aj of the (n2,axn2,a) matrix A2,aa is [ ]
 
Φ *2, a j ∂Φ 2, ai Φ 2, a j j ∂Φ 2, a j
*
[ A2( LS )
, aa ] ai a j
= β Z .  ∫ ( * i +
ρ 0 ω ∂n
).(
Z
−
ρ 0ω ∂n
). dΩ 
 Ω Z2 Z 
   ∂Φ *2, ai ∂Φ 2, a j 
  βv 
+ β p . ∫ Φ 2, ai .Φ 2, a j . dΩ  +
*
. ∫ . . dΩ 
Ω +Ω  ρ 02ω 2  Ω + Ω c12
∂n ∂n

 p 2 c12  v2
(2.125)
(LS)
and the element on row ai and column aj of the (n2,axn1,a) matrix A2,1 is [ ]
   ∂Φ *2, ai ∂Φ 1, a j 
[ ]
A2( LS
,1
) β
= − β p .  ∫ Φ *2, ai .Φ 1, a j . dΩ  − 2 v 2 .  ∫
∂n
.
∂n
. dΩ 
 ρ 0ω  Ω c12
ai a j
 Ω c12 
,
(2.126)
(LS)
The element on row ai of the (n2,ax1) vector f 2,a is { }
-
   
{ } ( LS )
f 2, a
ai
= β p  ∫ Φ 2, ai . p. dΩ  + β p  ∫ Φ 2, ai . pq . dΩ 
 *   *
 Ω p2   Ω c12 
 ∂Φ *2, ai   ∂Φ *2, ai ∂pq 
jβ v  + βv   . (2.127)
+ ∫ ∂n . v . d Ω ∫ . . d Ω
  ρ 02ω 2  Ω c 
n
ρ 0ω  Ω v ∂n ∂n
2  12 
The combination of the matrix equations (2.113), (2.119) and (2.124) yields
the convergent wave model
 A Cs 1, a 0   ws   fs 
 ( ss 
)    ( LS ) 
C1, as , 2   p1, a  =  f 1, a  .
LS )
A1( LS )
, aa A1( LS (2.128)
 ( LS ) ) 
p   ( LS ) 
 0 A2,1 A2( LS
, aa  2, a   f 2, a  
Since it follows from the element definitions in (2.121) and (2.125) that
[ A1,aa ] = [ A1, aa ] , [ A2, aa ] = [ A2,aa ]

( LS ) ( LS )
H ( LS ) ( LS )
H
, (2.129)
where H denotes the complex conjugate transpose, and since it follows

from a comparison between (2.122) and (2.126) that
[ A1,2 ] = [ A2,1 ]
( LS ) ( LS )
H
, (2.130)
a least-squares model for an uncoupled interior acoustic problem is

Hermitian, even if a domain decomposition is required for convergence.
104 Chapter 2
linear independence of the wave functions

When a coupled vibro-acoustic system is excited at a circular frequency ω,
which corresponds with a natural frequency of one of the circumscribing
rectangular domains with side walls, assumed to be perfectly rigid, i.e. when
a1 . π 2 a 2 . π 2
ω = c. ( ) +( ) (i = 1, 2 and a1 , a 2 = 0,1,.. ) , (2.131)
Lx i Lyi
one can readily see from (2.86) and (2.89) that not all functions in the
complete wave function sets are linearly independent. Hence, in order to
prevent the wave models (2.115) and (2.128) from becoming singular at
such a frequency, some wave functions must be eliminated a priori from
each wave function set, such that all remaining wave functions are linearly
independent and that the spans of the original wave function sets are not
reduced.
2.6. Practical implementation
2.6.1. Wave function scaling
It can be seen from the wavenumber definitions in (2.83) that two different
types of wave functions are comprised in the proposed complete wave
function set. In one type of wave functions, the wavenumber components in
both the x- and y-direction are real numbers, which means that the
amplitudes of these wave functions are nowhere larger than 1 (see e.g.
figure 2.8(a)). In the other type, the wavenumber component in one
direction is a real number, while the wavenumber component in the other
direction is a complex number. For this type of wave functions, the
amplitudes are decaying in the direction, associated with the complex
wavenumber component, and they may become significantly larger than 1
(see e.g. figure 2.8(b)). Moreover, the larger is the real wavenumber
component in one direction, the larger becomes the amplitude decay in the
other direction.
As will be illustrated in section 2.7.1., it is important for the numerical
condition of the wave model that all wave functions have amplitudes, not
larger than 1, within their definition domain, i.e. within the entire acoustic
domain or within one of the acoustic subdomains of the considered vibro-
acoustic system.
a1π 2
− j k 2 −( ) .y
a π Lx
Figure 2.8 : real part of cos( 1 x).e for a1=1 (a) and a1=4 (b)
Lx
(Lx=Ly=1m, k=ω/c=2π.500/340m ).
-1
Hence, for the practical implementation of the proposed wave models, some
scaling factors f xa 2 and f ya1 are introduced in the wave function
definitions. When the circumscribing rectangular domain of the considered
acoustic (sub)domain is defined by x ∈ [0, L x ], y ∈ [0, L y ] , the following
wave function scaling is introduced,
− j. k ya1 .( y − f ya1 . L y )
Φ a 1 ( x, y ) = cos( k xa1 . x ). e
(2.132)
− j. k xa2 .( x − f xa2 . L x )
Φ a 2 ( x, y ) = e .cos( k ya 2 . y )
with
1, if Im( k xa 2 ) > 0 1, if Im( k ya 1 ) > 0
f xa 2 =  , f ya1 =  (2.133)
0, if Im( k xa 2 ) ≤ 0 0, if Im( k ya 1 ) ≤ 0
In a similar way, the four structural wave functions, defined in (2.16), are
scaled as follows (x’∈ [0, L]) ,
Ψ s ( x’) = e − j . k b .( x ’− f s . L )
s
, (2.134)
106 Chapter 2
where the scaling factor f2 equals 1, while the factors f1, f3 and f4 are zero.
2.6.2. Numerical integration
It follows from the definitions of the wave model coefficients that the
practical implementation of a wave model requires the numerical evaluation
of boundary surface integrals. The Gauss-Legendre quadrature rule (see e.g.
GERALD and WHEATLEY (1989)) is an accurate numerical integration rule,
which is most commonly used for the practical implementation of finite
element and boundary element models. According to this quadrature rule,
the one-dimensional boundary surface integrals, involved with wave models
for two-dimensional systems, are evaluated as follows,
t2
t 2 − t1 1
∫ f ( t ) dt = (
2
). ∫ f (u )du
t1 −1
(2.135)
t −t  n ~ 
≈ ( 2 1 ).  ∑ w i . f (ui ) ,
2 i=1 
where the function f is evaluated at the quadrature points ui, which are the
roots of the n-th order Legendre polynomial Pn, i.e.
Pn (ui ) = 0, (i = 1.. n) , (2.136)
and where the weighting values w ~ are determined, such that the exact
i
integral is obtained for all polynomial functions of degree 2n-1 or less. The
two-dimensional boundary surface integrals, involved with wave models for
three-dimensional systems, are evaluated as follows,
s2  t 2 (s)  s 2 − s1 1  1 t 2 (u ) − t 1 (u ) 
∫  ∫ f (s,t ). dt  .ds = ( ). ∫  ∫ . f (u, v ). dv du
−1  −1 
s1  t 1 ( s )  2 2
(2.137)
s − s  n 1 n 2 ~ ~ t 2 ( ui 1 ) − t 1 ( ui 1 ) 
≈ ( 2 1 ).  ∑ ∑ w . w . . f ( u 
i1 i 2  .
, v )
2  i1 i 2
2
 i1 = 1 i 2 = 1 
For the implementation of finite element and boundary element models,

only a few quadrature points are needed for the involved numerical
integrations, since the field variable shape functions are simple, locally
defined functions. For the implementation of the proposed wave models,
5
however, a substantially larger amount of quadrature points are needed ,
• since the wave functions are defined in the entire continuum domain, or
at least in large subdomains,
• since the wave functions are highly oscillating functions6, especially
those with large wavenumber components,
• since the wave model coefficients must be calculated with a high
accuracy, due to the poor numerical condition of the wave models, as
will be discussed in section 2.7.1.
2.6.3. Wave function truncation
. /
Recall from sections 2.3 and 2.5 that the steady-state pressure field in the
cavity domain or in each cavity subdomain of a two-dimensional interior
coupled vibro-acoustic system is approximated as an expansion of the form
0
na
p( x , y ) = ∑ pa .Φ a ( x , y ) + pq ( x , y )
a=1
n a1 n a2
− jk ya1 y − jk xa2 x
= ∑ pa1 .cos(k xa1 x ). e + ∑ pa 2 . e .cos( k ya 2 y ) + pq ( x, y )
a1 = 0 a2 = 0
(2.138)
a .π a .π
( k xa 1 , k ya 1 ) = ( 1 ,± k 2 − ( 1 ) 2 ) ,
Lx Lx
with (2.139)
a .π a .π
( k xa 2 , k ya 2 ) = ( ± k − ( 2 ) 2 , 2 ) .
2
Ly Ly
5
instead of using a large number of quadrature points, distributed over the entire
boundary surface integration domain, the boundary surface integrals may be
evaluated by splitting the integration domain into smaller subdomains, so that the
subdomain integrals require a smaller number of quadrature points
6
in view of enhancing the numerical integration efficiency, the potentials of
special-purpose Filon-type quadrature rules for oscillating integration functions
(see e.g. EVANS (1993)) will be explored in the future
108 Chapter 2
Although it has been proven in section 2.5. that this type of solution
expansion converges towards the exact solution in the limit for
na=2(na1+1)+2(na2+1)→∞, the practical implementation of the corresponding
wave model requires a truncation of the complete wave function set, in that
only an expansion, truncated at some finite integer values na1 and na2, is
amenable to numerical implementation.
Although there is no straightforward relation between the truncation values
na1 and na2 and the subsequent prediction accuracy, a rule of thumb for this
truncation may be formulated, based on some physical considerations. Since
the dynamic response of a coupled vibro-acoustic system results from a
complex mechanism of wave propagation in its structural and acoustic
domains, the resulting spatial variation of the dynamic response is strongly
dependent on the structural and acoustic wavelengths. Since the (undamped)
structural wavelength λb is usually much smaller than the acoustic
wavelength λ, at least at frequencies where vibro-acoustic coupling effects
are significant, i.e.
2π 4π 4 Et 2 2π 2πc
λb = =4 << λ = = , (2.140)
kb 3ρ s ω ( 1 − ν )
2 2 k ω
7
the largest spatial variation of the acoustic pressure field usually occurs in
the vicinity of the fluid-structure coupling interface and is strongly related to
the structural wavelength λb.
The smallest spatial periods in the x- and y-direction, associated with the
wave functions in the pressure expansion (2.138), are, respectively,
2Lx 2Ly
λ x,min = , λ y,min = . (2.141)
na1 na 2
Since the wave functions are expressed as cosine functions, the proposed
truncation rule is based on a similar truncation rule, proposed for half-range
Fourier cosine series (see e.g. SPIEGEL (1973)), in that the smallest period of
of the cosine functions in the expansion should not be larger than half the
7
assuming, for instance, that the spatial periods of the external source distribution
in the acoustic cavity or the pressure, normal velocity and normal impedance
distributions, imposed on the cavity boundary surface, are not smaller than the
structural wavelength λb
period of the function to be approximated. An extrapolation of this rule to

the considered wave function expansion yields a truncation rule
λb
λ x,min ≈ λ y,min ≤ , (2.142)
2
which means that the integer values na1 and na2 should be such that
na1 na 2 4 2k
≈ ≥ = b . (2.143)
Lx Ly λb π
Note that this truncation rule is applied for the pressure expansion in any
cavity subdomain, even if the structural component is not part of the
boundary surface of that subdomain. In this way, it is ensured that the
expansions in two adjacent subdomains can represent pressure fields with
similar spatial variations along their common interface, which is necessary
for an accurate approximation of the pressure and normal fluid velocity
continuity at this interface.
It is very hard to assess a priori the prediction accuracy, obtained with a
certain truncated expansion. However, the a posteriori assessment of the
prediction accuracy can be easily performed. Since the dynamic equations
are inherently satisfied by the proposed field variable expansions, it is
sufficient to check the representation of the boundary conditions. Moreover,
if the required accuracy is not yet reached with a certain wave model, a
model refinement is easily obtained. Since the addition of some wave
functions to the field variable expansions is not changing the integration
domains of the boundary surface integrals, involved with the calculation of
the matrix coefficients of a wave model, only the additional rows and
columns, associated with the additional wave functions, must be calculated.
The rows and columns, associated with the wave functions in the original
field variable expansions, remain unchanged.
2.7. Numerical condition and convergence
2.7.1. Condition
As already defined in equations (2.41) and (2.48), a wave model is basically

a matrix equation of type
110 Chapter 2
w 
[ A]  s  = {b} , (2.144)
pa
where [ A] is a fully populated (na+4)x(na+4) matrix. The singular value

decomposition of this matrix takes the form (GOLUB and VAN LOAN (1983))
na + 4
[ A] = [U ][ Σ ][ V ] H = ∑ {Ui }.σ i .{Vi } H , (2.145)
i=1
where [ Σ ] is a diagonal matrix of singular values σi, where {Ui } and {Vi }
are the i-th column vectors of the orthonormal matrices [U ] and [ V ] and
where H denotes the complex conjugate transpose. The solution of matrix
equation (2.144) may then be written as
w s  n a + 4 {Ui } . {b} na + 4
H
β
 = ∑
σi
{Vi } = ∑ i {Vi } . (2.146)
pa  i = 1 i=1 σ i
ZIELINSKI and HERRERA (1987) indicated that the transformation of a

continuum problem into the formulation, proposed by Trefftz, yields a very
ill-conditioned problem, due to the fact that the complete function sets, used
for the field variable expansions, consist of non-orthogonal functions, which
are globally defined in the entire continuum domain. As a consequence, the
condition number of matrix [ A] is large, i.e. the ratio between the largest
and smallest singular value is large, and the singular values decay gradually
towards very small values. Due to these properties, one can readily see from
its definition formula (2.146) that the solution vector is very sensitive to any
small error on the calculation of the coefficients β i, associated with the very
small singular values σi. Since a numerical implementation inevitably
induces some round-off errors, the numerically obtained solution vector of
an ill-conditioned problem is usually not a reliable representation of the
exact mathematical solution vector. However, for some ill-conditioned
problems, the numerical solution vector may be accurate. These problems
are said to be mildly ill-conditioned and satisfy the so-called Picard
conditions (see e.g. VARAH (1979,1983)). These conditions state that
accurate numerical solutions for ill-conditioned problems can only be
obtained,
• if the difference between the amplitudes of two consecutive singular

values, when sorted in a sequence with descending amplitude, becomes
only large at the end of this sequence, i.e. for the very small singular
values,
• and if the coefficients β i are smaller or at least not very much larger than
their associated singular values σi.
To illustrate that the numerical models, obtained with the proposed wave
based prediction technique, satisfy these Picard conditions, the following
coupled vibro-acoustic problem, shown in figure 2.9, is considered.
One part of the boundary surface of an acoustic cavity consists of a flexible

plate with clamped boundaries, while another part of the boundary surface is
treated with a layer of insulation material with normal impedance Z. The
remaining parts of the cavity boundary surface are perfectly rigid. The
coupled vibro-acoustic system is excited by a mechanical line force F,
applied at location x’=x0’ on the plate, in the direction normal to the plate
and with a harmonic time dependence at a frequency ω. A weighted residual
model, as defined in (2.41), is used for the prediction of the steady-state
pressure field p(x,y) and normal plate displacement w(x’), for the case of an
air-filled (ρ0=1.225 kg/m , c=340 m/s) cavity (Lx=1.5 m, Ly=1 m) and an
3
aluminium plate (E=70.109 N/m2, ρs=2790 kg/m3, ν=0.3, t=0.002 m, η=0),

excited by a unit line force at x0’=0.5m with frequency ω=2π.200 Hz. The
112 Chapter 2
normal impedance Z is assumed to be infinitely large, which means that the

insulation layer is considered to be perfectly rigid. The expansion for the
pressure field is based on the complete acoustic wave function set,
associated with the rectangular domain, enclosing the cavity. Figure 2.10
plots the singular values σi and their associated coefficients β i of the wave
model, obtained with the 4 linearly independent structural wave functions
and a truncated set of 150 (na1=44, na2=29) linearly independent acoustic
wave functions.
Figure 2.10 : Picard test of a scaled wave model
Since the decay of the singular value curve is only large for the small
singular values and since the coefficients β i are smaller or at least not
8
substantially larger than the associated singular values, the wave model
satisfies both Picard conditions.
In this respect, it is important that the structural and acoustic wave functions
in the field variable expansions are scaled, as outlined in section 2.6.1.
8
the large decay at the first singular values is due to the fact that the structural
wave functions are not scaled, relative to the acoustic wave functions, so that the
the resulting structural wave function contributions, having the physical
dimension of displacement (m), are some orders of magnitude smaller than the
(dominant) acoustic wave function contributions, having the physical dimension
of pressure (Pa); this relative scaling may be introduced, but is not crucial for the
accuracy of the numerical solution
Figure 2.11 plots the resulting singular values and coefficients β i of the
same wave model, but without wave function scaling.
Figure 2.11 : Picard test of an unscaled wave model
This figure clearly illustrates that the condition of an unscaled wave model
is significantly worse than the condition of the corresponding scaled wave
model, since the ratio between the largest and smallest singular value is
significantly larger. It illustrates also that the first Picard condition is no
longer satisfied, since the decay of the singular value curve is large for all
singular values.
The important consequence of the wave models, satisfying the Picard
conditions, is that the proposed wave based prediction technique, for which
the convergence is theoretically ensured, can be numerically implemented
into a prediction model, whose poor numerical condition is not preventing
the prediction results from converging towards the exact solution. Some
illustrations of this numerical convergence are given in the next section.
2.7.2. Convergence
2.7.2.1. A posteriori convergence assessment
accuracy check
Since the dynamic equations are inherently satisfied by the field variable
expansions, it is sufficient to check the approximation errors, involved with
the representation of the boundary conditions. Hence, the prediction
114 Chapter 2
accuracy, obtained with a certain wave model, can be easily identified from
the visualisation of the predicted field variable distributions on the
boundaries of the structural and acoustic domains.
convergence curve
Another indication of the convergence of a wave model is given by the
convergence curve, which plots the relative prediction error ∆ϕ on a
dynamic quantity ϕ against the number of degrees of freedom of a wave
model. The relative prediction error may be defined as
ϕ (na + 4) − ϕ ex
∆ϕ (na + 4) = , (2.147)
ϕ ex
where ϕ(na+4) is the value for ϕ, predicted by a wave model with 4

structural and na acoustic wave functions, and where ϕex is its exact value.
Since the exact solution is not known, this prediction error is evaluated,
using the value, predicted by a very large wave model, instead of the exact
value ϕex. If a wave model converges towards the exact solution, the
prediction error should decay towards zero for an increasing number of
degrees of freedom.
2.7.2.2. Some validation examples

In this section, the above defined convergence assessment is performed for
two coupled vibro-acoustic systems, one with a convex acoustic domain,
and one, for which the construction of a complete wave function set requires
the decomposition of the acoustic domain into some subdomains.
convex continuum domain

The accuracy of a weighted residual wave model of type (2.41) is illustrated
for the coupled vibro-acoustic system, described in section 2.7.1. and shown
in figure 2.9. By expanding the cavity pressure field in terms of the acoustic
wave function set, associated with a rectangular domain, enclosing the
cavity domain, the theoretical convergence of the wave model is ensured,
since, according to the discussion in section 2.5.3., the convexity of the
cavity domain is a sufficient condition for convergence. Two different cases
are considered.
• In the first case, the plate in the coupled vibro-acoustic system is excited
by a unit line force F with frequency ω=2π.200 Hz. The normal
impedance Z is infinitely large, which means that the insulation layer,
located at x=0 m in the cavity, is perfectly rigid. All other system
properties are already mentioned in section 2.7.1. For this case, a
weighted residual wave model with 4 structural and 300 acoustic wave
functions (na1=89, na2=59) is built.
• In the second case, the frequency of the plate excitation is ω=2π.1000
Hz. The insulation layer has a normal impedance Z=ρ0c(1-2j) kg/m2s. For
this case, a weighted residual wave model with 4 structural and 660
acoustic wave functions (na1=199, na2=129) is built.
The prediction results for the wave model of the first case are visualised in
figures 2.12 and 2.13. The first figure plots the real part of the normal plate
displacement against the real part of the normal fluid displacement along the
fluid-structure coupling interface. The second figure shows the contour plot
of the real part of the cavity pressure field. The imaginary parts of the
prediction results are not shown, since these are negligibly small, due the
absence of any type of structural or acoustic damping.
Figure 2.12 : real part of the normal plate displacement w (solid) and the normal fluid
displacement wa (+) along the fluid-structure coupling interface at 200 Hz
116 Chapter 2
Figure 2.13 : contour plot of the real part of the cavity pressure at 200 Hz
Figure 2.12 clearly illustrates that the clamped boundary conditions of the
plate and the displacement continuity condition at the fluid-structure
coupling interface are accurately represented. The fact that the pressure
contour lines in figure 2.13 are perpendicular to all rigid parts of the cavity
boundary surface, indicates that there is no normal pressure gradient at these
boundary parts and illustrates that the zero normal fluid velocity conditions
are accurately represented.
The prediction results for the wave model of the second case are visualised
in figures 2.14-2.17.
Figure 2.15 : pressure (solid) versus normal fluid velocity, multiplied by Z (+),
along the cavity insulation layer at 1000 Hz
Figure 2.16 : contour plot of the real part of the cavity pressure at 1000 Hz
118 Chapter 2
Figure 2.17 : contour plot of the imaginary part of the cavity pressure at 1000 Hz
Again, it may be concluded from these figures that all structural and
acoustic boundary conditions are accurately represented. Note that the
largest approximation errors occur in the vicinity of those locations, where
some discontinuous boundary conditions are imposed. This is illustrated, for
instance, on figure 2.15, where some small error may be noticed for the
upper end of the impedance boundary condition. This is due to the fact that
this location belongs to two parts of the cavity boundary surface, on which
two different boundary conditions are imposed.
The numerical convergence is also illustrated by the convergence curves,

defined in section 2.7.2.1. Figure 2.18 plots the convergence curves,
obtained with weighted residual wave models for the coupled vibro-acoustic
system, considered in the first case. It illustrates the numerical convergence
for the normal plate displacement predictions at location x’=0.25m and for
the cavity pressure predictions at location (x,y)=(0.5m,0.5m) at two
excitation frequencies, 200 Hz and 500 Hz. The ‘TR’ marks on this figure
indicate the prediction errors for those wave models, for which the acoustic
wave function sets are truncated at the smallest integer values na1 and na2,
satisfying the truncation rule (2.143).
The convergence curves for the normal plate displacement predictions at
200 Hz at location x’=0.25m, shown in figure 2.19, illustrate that it is
advantageous for the convergence rate to select the wave functions,
associated with the smallest rectangular domain, enclosing the considered
continuum domain.
Figure 2.18 : structural (upper part) and acoustic (lower part)

convergence curves at 200 Hz (+) and 500 Hz (∇)
Figure 2.19 : structural convergence curves at 200 Hz, obtained with the acoustic
wave functions, associated with two different enclosing rectangular domains
domain decomposition
Figure 2.20. shows an interior coupled vibro-acoustic system, which is
identical to the system, shown in figure 2.9, except that the cavity boundary
120 Chapter 2
surface is nowhere treated with an insulation layer and that two rigid
bulkheads are positioned in the cavity domain. The bulkheads are assumed
to be infinitesimally thin and their heights are 0.5*Ly.
Figure 2.20 : two-dimensional interior coupled vibro-acoustic system with a non-

convex acoustic domain
According to the procedure, outlined in section 2.5.3., a convergent wave

model may be obtained by decomposing the non-convex acoustic domain
into 3 subdomains and by expanding the pressure fields p1, p2 and p3 in these
subdomains in terms of the acoustic wave function sets, associated with the
enclosing rectangular domains with dimensions Lx1xLy, Lx2xLy and Lx3xLy,
respectively (see figure 2.20).
The case is considered of an air-filled cavity and an aluminium plate with
the same geometrical and material properties as in the previously considered
vibro-acoustic system. The plate is excited by a unit line force at location
x0’=0.5m with a frequency ω=2π.200 Hz. A weighted residual wave model
of type (2.115) is built with 4 structural and 540 acoustic wave functions
(180 wave functions for pressure field p1 (n1,a1=30, n1,a2=60), 210 wave
functions for pressure field p2 (n2,a1=30, n2,a2=60) and 150 wave functions for
pressure field p3 (n3,a1=15, n3,a2=60)).
Again, the accuracy check of the prediction results for this wave model
illustrates the numerical convergence of the proposed wave based prediction
technique. Figure 2.21 plots the real part of the normal plate displacement
against the real part of the normal fluid displacement along the fluid-
structure coupling interface and illustrates the accurate representation of the
clamped boundary conditions of the plate and the displacement continuity

condition at the fluid-structure coupling interface.
Figure 2.22 shows the contour plot of the real part of the cavity pressure
field. Since the contour lines are perpendicular to the rigid parts of the
cavity boundary surface and since they are continuous at the common
interfaces between the considered subdomains, this figure illustrates the
accurate representation of the acoustic boundary conditions.
Figure 2.22: contour plot of the real part of the cavity pressure at 200 Hz
122 Chapter 2
The numerical convergence of the wave based prediction technique, applied

for the considered case, is also illustrated by the convergence curves. Figure
2.23. illustrates the numerical convergence for the normal plate
displacement predictions (∇) at x’=0.25m and for the pressure predictions at
locations in the three cavity subdomains, i.e. (x,y)=(0.2m,0.2m) (+),
(x,y)=(1m,0.5m) (o) and (x,y)=(1.3m,0.8m) (*).
Figure 2.23 : structural and acoustic convergence curves at 200 Hz
2.7.2.3. Weighted residual versus least-squares model
For all cases, considered in the previous section, the numerical convergence
is illustrated for the wave models of type (2.41) and (2.115), which result
from a weighted residual formulation of the acoustic boundary conditions.
For least-squares wave models of type (2.48) and (2.128), which result from
the minimisation of an error functional, the numerical prediction results are
also converging. However, weighted residual models are preferred to least-
squares models, since the former have a substantially higher convergence
rate. This is illustrated for the two coupled vibro-acoustic systems,
considered in the previous section.
For the system, shown in figure 2.9, a convergence comparison is made for
the normal plate displacement predictions at x0’=0.25m, obtained with
weighted residual and with least-squares models. Two force excitation
frequencies are considered, i.e. 200 Hz and 500 Hz, and the insulation layer
in the cavity is assumed to be perfectly rigid. As mentioned in section

2.4.1.2., some values must be specified for the parameters β p, β v, β Z and β s
in the error functional, involved with least-squares models, in order to
restore the homogeneity of the physical dimensions of the different
functional components and to possibly allocate different weights to the
various boundary conditions. For the considered system, the error functional
comprises only the parameters β v and β s, which are identically set to 1, since
both parameters are associated with normal velocity boundary conditions.
The resulting convergence curves, shown in figure 2.24, clearly illustrate the
beneficial convergence rates, obtained with weighted residual models.
Figure 2.24 : structural convergence curves at 200 Hz (∇) and 500 Hz (+)
for weighted residual (solid) and least-squares (dashed) models
For the system, shown in figure 2.20, a similar convergence comparison is

made. For the error functional, involved with least-squares models of this
system, two types of parameter specifications are considered. For the first
type, the pressure and normal velocity boundary conditions are equally
weighted, i.e. β p=β v=β s=1. For the second type, the various boundary
conditions are weighted, such that their functional contributions are
dimensionally homogeneous, i.e. β v=β s=1, β p=1/(ρ0c)2. Figure 2.25 shows
the resulting convergence curves for the normal plate displacement
predictions at x0’=0.25m for a force excitation frequency of 200 Hz.
124 Chapter 2
Figure 2.25 : structural convergence curves at 200 Hz for weighted residual

models(∇) and least-squares models with (o) and without (+) error weights
This figure illustrates that a dimensionally homogeneous parameter

specification in a least-squares model is beneficial for its convergence rate,
but that the latter is inferior to the convergence rate of a weighted residual
model.
2.7.2.4. Comparison with an existing complete function set
As already mentioned in the introductory section 2.1, HERRERA (1984b) has

also defined a complete function set for the Helmholtz equation. He proved
that, for solving two-dimensional acoustic problems in a bounded fluid
domain, the theoretical convergence of a Trefftz model is ensured, if the
pressure field in a bounded acoustic domain is expanded in terms of the
following function set, defined in a polar co-ordinate system (r,θ),
Φ a ( r, θ ) = { J 0 ( kr ), J n ( kr ).cos( nθ ), J n ( kr ).sin(nθ )} ( n = 1, 2,...) , (2.148)
where Jn is the n-th order Bessel function of the first kind.

As it is the case for all practical implementations of Trefftz-type models
performed so far, the poor condition of the Trefftz models, using the
proposed function set (2.148), seriously disturbs the numerical convergence.
MASSON et al. (1994) illustrated, for instance, the numerical convergence

problems, involved with the practical implementation of this function set for
the prediction of the mode shapes of a two-dimensional rectangular cavity
with a perfectly rigid boundary surface. They utilised also an alternative
function set for this problem, consisting of propagating plane wave
functions. However, they couldn’t prove the theoretical convergence for that
function set and its practical implementation suffered also from severe
numerical convergence problems.
An important breakthrough of the wave based prediction technique,
proposed in this dissertation, is that a complete wave function set is found,
which allows the introduction of the approach, proposed by Trefftz, into a
coupled vibro-acoustic prediction model, whose poor condition no longer
prevents the numerical results from converging towards the exact solution.
This section illustrates this breakthrough for an uncoupled acoustic problem,
shown in figure 2.26.
Figure 2.26 : two-dimensional uncoupled acoustic problem
Along one part of the boundary surface of the convex acoustic cavity, a
normal fluid velocity distribution is imposed with a harmonic time
dependence of 200 Hz. All other parts of the boundary surface are perfectly
rigid.
Two types of weighted residual models are considered. The first type
utilises the newly proposed wave functions, associated with the enclosing
rectangular domain with dimensions Lx=1.5m and Ly=1m (see figure 2.26).
126 Chapter 2
The second type utilises the existing function set, defined in (2.148). The
origin of the polar co-ordinate system of this function set is centrally
positioned at (x,y)=(0.75m,0.5m). For the same reasons as outlined in 2.6.1.,
this function set is scaled, such that the largest amplitude of each function
on the cavity boundary surface equals 1.
Figure 2.27 shows the imaginary parts of the pressure predictions at
(x,y)=(1m,0.8m) for the two types of weighted residual models, obtained by
increasing their number of degrees of freedom, i.e. the number of functions,
used for the cavity pressure expansion.
Figure 2.27 : imaginary part of the pressure at (x,y)=(1m,0.8m) at 200 Hz,

obtained with weighted residual models, using the newly proposed (solid)
and the existing (dashed) complete function set
This figure clearly illustrates, once again, that the numerical implementation
of a prediction model, using the newly proposed wave function set, is
convergent. On the contrary, the numerical implementation of a prediction
model, using the existing complete function set, is not convergent, despite
the theoretically proven convergence. Note also that, from a certain number
of degrees of freedom onwards, the prediction models cannot be solved
anymore, since the scaling factors of the high-order Bessel functions
become so large that they are numerically considered to be infinite.
Relaxing the scaling factors or discarding the function scaling put also a
practical limitation on the model size, since the condition then becomes so
poor that the models are numerically considered to be singular.
A verification of the two Picard conditions, discussed in section 2.7.1.,
reveals also the numerical convergence problems, involved with the existing
function set.
Figure 2.28 : Picard test for a weighted residual model, using the newly proposed
(upper) and the existing (lower) complete function set
Figure 2.28 illustrates that a prediction model with the newly proposed wave
function set satisfies the Picard conditions, whereas a model with the
existing function set doesn’t satisfy these conditions: the decay of the
singular value curve is large for all singular values and most of the
128 Chapter 2
coefficients β i are substantially larger than their associated singular values

σi.
2.8. Comparison with FEM and BEM

As indicated in section 1.4.1., one of the big challenges in coupled vibro-
acoustic modelling is the development of a new deterministic prediction
technique, which provides accurate prediction results with an enhanced
computational efficiency, compared with the existing element based
prediction techniques.
In order to assess whether the proposed wave based prediction technique has
the potential for meeting this challenge, this section outlines its advantages
and drawbacks, compared with the finite element and boundary element
method.
FEM
In comparison with the FE method, the wave based prediction technique has
several advantageous properties.
• Since the field variable expansions satisfy a priori the governing dynamic
equations, approximation errors are only involved with the representation
of the boundary conditions. This yields substantially smaller models than
corresponding FE models.
• The expansions for derived secondary field variables, such as fluid
velocity and structural stress, have the same spatial variation as the
primary field variables, i.e. the acoustic pressure and structural
displacement. This is advantageous for the convergence rate, especially
in the case of coupled vibro-acoustic problems, for which the effect of
the fluid on the structure is pressure controlled, but for which the effect
of the structure on the fluid is velocity controlled.
• A model refinement is easily obtained. Firstly, the accuracy of a model is
easily checked, since only the approximation errors of the boundary
conditions must be verified. Secondly, increasing the number of degrees
of freedom, i.e. the number of wave functions in the field variable
expansions, requires only the calculation of the additional matrix
coefficients, associated with the additional wave functions, while the
original model coefficients remain unchanged. For FE models, increasing
the number of degrees of freedom requires the calculation of a
completely new model, at least when a global refinement of the element

discretization is needed.
The wave based prediction technique has also some drawbacks, compared
with the FE method.
• Since the wave functions are complex functions, defined within the
entire continuum domain or at least within large subdomains, the wave
model matrices are fully populated and complex.
• Due to the implicit frequency dependence of the wave functions, the
wave model matrices cannot be decomposed into frequency independent
submatrices, which implies that a wave model must be recalculated for
each frequency of interest and that natural frequencies and mode shapes
cannot be obtained from a standard eigenvalue calculation.
• Due to the global and oscillatory nature of the wave functions and due to
the poor condition of a wave model, the involved numerical integrations
are computationally more demanding.
• The FE method is a widely applicable technique, since the convergence
of a FE model is obtained with simple element shape functions, whereas
the convergence of a wave model requires the definition of a complete
wave function set and appropriate particular solution functions.
BEM
The wave based prediction technique has several properties in common with
the BE method, in that both methods require a particular solution function
(kernel function) and yield small, but complex, fully populated and
frequency dependent models.
Besides their common properties, the wave based prediction method has
several advantageous properties, compared with the BE method.
• In the wave based prediction technique, an integral, i.e. weighted residual

or least-squares, formulation of the boundary conditions requires the
evaluation of single boundary surface integrals9. In the BE method,
single integral evaluations are only obtained by using a non-integral,
collocational formulation, as is commonly used in the direct BE method.
However, the resulting convergence rate is substantially smaller than the
9
at least when a non-integral particular solution function w a can be defined, as in
(2.19), instead of its general integral form (2.18)
130 Chapter 2
convergence rate, obtained with an integral formulation. This faster

convergence of an integral formulation, as is commonly used in the
indirect BE method, can only be obtained at the expense of the
computationally more demanding evaluation of double boundary surface
integrals.
• In contrast with the direct BE method, the wave based prediction
technique yields a symmetric or Hermitian model for uncoupled
(acoustic) problems.
• The wave based prediction technique doesn’t involve singular integral
evaluations.
• A global refinement of the boundary element discretization requires the
calculation of a completely new BE model, whereas the refinement of a
wave model requires only the calculation of some additional matrix
coefficients, as already indicated above.
• Since the BE method follows a two-step procedure, the field variable
predictions within the considered continuum domain result from a post-
processing step, which requires some integral evaluations, whereas, in
the wave based prediction technique, the field variable predictions result
from some simple function evaluations.
The wave based prediction technique has also some drawbacks, compared
with the BE method.
• Due to the discretization of the boundary surface, the element shape

functions in a BE model must only meet some mild convergence
requirements, whereas the wave based prediction technique requires the
definition of a complete wave function set.
• The wave model coefficients result from the evaluation of single
integrals over the entire boundary surface, whereas the BE model
coefficients, at least in the direct BE method, result from the evaluation
of single integrals over small parts of the boundary surface.
• A wave model doesn’t inherently satisfy the Sommerfeld radiation
condition, involved with problems with an unbounded continuum
domain.
2.9. Conclusions
This chapter presents the methodology of a new, wave based prediction
technique, which is based on the modelling approach, initially proposed by
Trefftz. The steady-state dynamic field variables in a coupled vibro-acoustic

problem are approximated in terms of a set of acoustic and structural wave
functions, which are exact solutions of the homogeneous parts of the
governing dynamic equations, and some particular solution functions of the
inhomogeneous equations. In this way, the problem of determining the field
variable distributions in a continuum domain is turned into a discrete
problem of determining the contributions of the wave functions to the field
variable expansions. Since the latter expansions satisfy a priori the
governing dynamic equations, the wave function contributions are merely
determined by the acoustic and structural boundary conditions, which are
transformed into a weighted residual or a least-squares formulation.
Although Trefftz’ modelling approach exists for more than seventy years
and offers several advantages over the element based modelling approach, it
is still not implemented into a valuable modelling alternative for the finite
element and boundary element method. The main reason is that the
transformation of a continuum problem into a Trefftz formulation yields an
ill-conditioned problem. As a result, all attempts, performed so far, to
practically implement a Trefftz-type prediction model, although theoreti-
cally proven to be convergent, suffer from severe numerical convergence
problems. An important breakthrough of the proposed wave based
prediction technique is that a complete wave function set is found, which
allows the introduction of Trefftz’ modelling approach into a coupled vibro-
acoustic prediction model, of which the poor condition no longer prevents
the numerical results from converging towards the exact solution.
The comparison of the properties of the wave based prediction technique
with the properties of the finite element and boundary element method
provides a strong indication for the enhanced convergence rate of the wave
based prediction technique, in that a high prediction accuracy may be
obtained from wave models, which are substantially smaller than
corresponding finite element and boundary element models. In the next
chapters, it becomes apparent that the smaller wave models involve also
smaller computational efforts, so that the proposed technique can be used in
a much larger frequency range than the existing element based prediction
techniques.
132 Chapter 2
3. APPLICATION OF THE WAVE BASED
PREDICTION TECHNIQUE FOR TWO-
DIMENSIONAL COUPLED VIBRO-ACOUSTIC
PROBLEMS
3.1. Introduction
This chapter describes how the prediction methodology, proposed in the
previous chapter, is applied for the modelling of the steady-state dynamic
behaviour of the main components that may be encountered in two-
dimensional coupled vibro-acoustic problems.
Sections 3.2. and 3.3. address the use of the wave based prediction
technique for the modelling of the coupling interaction between the pressure
field in a general, bounded acoustic domain and the dynamic displacements
of, respectively, flat and curved shell structures. Apart from the definition of
appropriate expansions for the dynamic field variables in curved shell
structures, given in section 3.3., the application of the proposed prediction
technique for the considered vibro-acoustic problems has already been
described and validated in the previous chapter. Therefore, both sections
focus mainly on the computational efficiency of the wave based prediction
technique. Based on several validation examples, the beneficial convergence
rate of the prediction technique is demonstrated, in that it provides highly
accurate prediction results with a smaller computational effort, compared
with the finite element method.
134 Chapter 3
The next two sections describe how acoustic damping effects, induced by
the treatment of an acoustic cavity with poroelastic insulation material and
by exterior acoustic radiation, can be included in a coupled vibro-acoustic
wave model. Section 3.4. defines a set of wave functions and describes the
construction of the subsequent wave model for the prediction of the steady-
state dynamic field variables in fluid-saturated, poroelastic insulation
materials. Throughout the validation examples, illustrating the convergence
of the proposed wave model, special attention is paid to the modelling of a
layer of poroelastic insulation material, used as damping treatment in an
acoustic cavity. A porous insulation layer is commonly modelled as a
frequency-dependent, but spatially invariant impedance boundary condition
for the acoustic cavity. To assess the validity of this spatial invariance
assumption, the actual surface impedance of a porous insulation layer is
evaluated, using the accurate prediction results of the proposed wave model.
Section 3.5. addresses the modelling of the acoustic radiation in the
unbounded acoustic domain of exterior coupled vibro-acoustic systems. In a
similar way as discussed in section 1.3.2.2., the modelling approach is based
on confining the unbounded exterior acoustic domain to a bounded domain
by introducing an artificial boundary surface, on which a particular, energy
absorbing, impedance boundary condition is specified.
3.2. Interior coupled vibro-acoustic problems with flat

structural components
The application of the wave based prediction technique for two-dimensional
coupled vibro-acoustic problems with flat structural components has already
been described and validated in the previous chapter. This section addresses
the computational efficiency of the proposed prediction technique. First, an
additional illustration of the prediction accuracy is given through the
verification of the vibro-acoustic reciprocity relation. Next, the new
prediction technique is compared with the finite element method in terms of
the computational efforts, needed for obtaining accurate prediction results.
3.2.1. Vibro-acoustic reciprocity
When the external excitations of a coupled vibro-acoustic system consist of

a mechanical point force, applied in a certain direction at location rs1 in the
structural domain Ωs ( f( rs ) = F1.δ ( rs , rs1 ) ), and an acoustic point source,
Application for two-dimensional coupled vibro-acoustic problems 135
applied at location r2 in the acoustic domain V ( q( r ) = Q2 .δ ( r , r2 ) ), the

following vibro-acoustic reciprocity relation may be formulated (see e.g.
NORRIS and REBINSKY (1993), WYCKAERT et al. (1996))
p( r2 ) v( r )
= − s1 . (3.1)
F1 Q = 0 Q2 F = 0
2 1
This relation states that the acoustic pressure p at location r2, caused by a
unit point force excitation at location rs1, and the structural velocity
component v in the direction of the mechanical point force at location rs1,
caused by a unit point source excitation at location r2, have the same
amplitudes, but opposite phases.
To illustrate the accuracy of the proposed prediction technique, the vibro-
acoustic reciprocity relation is verified for the prediction results of the
following coupled vibro-acoustic system (see figure 3.1). One part of the
boundary surface of a two-dimensional, air-filled (ρ0=1.225 kg/m , c=340
3
m/s) cavity consists of a flexible plate (E=70.10 N/m , ρs=2790 kg/m3,

9 2
ν=0.3, t=0.002 m, η=0) with clamped boundaries. The other boundary

surface parts are perfectly rigid. Two dynamic response spectra are
considered. The first is the spectrum of the steady-state acoustic pressure at
location (x,y)=(Lx/3,Ly/2), caused by a unit normal force excitation F1 at
location x’=2(Lx-0.75*Ly)/3 (see figure 3.1(a)). The second is the spectrum
of the steady-state normal structural velocity at location x’=2(Lx-0.75*Ly)/3
caused by a unit acoustic excitation Q2 at location (x,y)=(Lx/3,Ly/2) (see
figure 3.1(b)).
(a) (b)
Figure 3.1 : structural (a) and acoustic (b) excitations of

a two-dimensional interior coupled vibro-acoustic system
136 Chapter 3
A weighted residual wave model of type (2.41) is built, using 4 structural

and 300 acoustic wave functions, associated with the rectangular domain
LxxLy=1.5m x 1m, circumscribing the acoustic cavity. Figure 3.2 shows the
resulting amplitude spectra of the considered acoustic pressure and
structural velocity. The very good agreement between both spectra
illustrates that the proposed wave model enables an accurate representation
of the vibro-acoustic reciprocity relation (3.1).
Figure 3.2 : vibro-acoustic reciprocity check (solid : pressure

spectrum (∆f=0.5 Hz), +-marks : structural velocity spectrum (∆f=1 Hz))
3.2.2. Comparison with coupled FE/FE models
In order to compare the proposed wave based prediction technique with the
finite element method in terms of accuracy and associated computational
efforts, the performances of weighted residual wave models of type (2.41)
for solving the two-dimensional validation problem, shown in figure 3.1(a),
are compared with the performances of corresponding Eulerian coupled
FE/FE models1 of type (1.22).
1
a comparison with coupled FE/BE models is not considered here, since, as
mentioned in section 1.3.1.3. and illustrated in appendix C, the boundary element
method can hardly compete with the finite element method for solving interior
(vibro-)acoustic problems
model properties
Each element discretization for the considered problem consists of one
three-dimensional layer of 8-noded hexahedral fluid elements (CHEXA) for
the cavity and 4-noded quadrilateral plate elements (CQUAD4) for the plate.
The two-dimensional discretizations are obtained from these three-
dimensional discretizations by imposing additional multipoint constraints,
expressing the invariance in the third dimension. Table 3.1 lists the number
of elements and the corresponding number of unconstrained degrees of
freedom2 (dof) in each coupled FE/FE model. Table 3.2 lists the number of
structural and acoustic wave functions, used in the weighted residual wave
models.
FE/FE 1 FE/FE 2 FE/FE 3

# elem. dof # elem. dof # elem. dof
plate 9 16 18 34 36 70
cavity 216 247 864 925 3456 3577
total dof 263 959 3647
plate 54 106 108 214 135 268
cavity 7776 7957 31104 31465 48600 49051
total dof 8063 31679 49319
Table 3.1 : properties of the coupled FE/FE models
wave 1 wave 2 wave 3 wave 4

dof dof dof dof
plate 4 4 4 4
cavity 50 100 150 200
total dof 54 104 154 204
Table 3.2 : properties of the weighted residual wave models
accuracy comparison
Figures 3.3 and 3.4 show some structural and acoustic frequency response
functions (with a frequency resolution of 1 Hz), obtained with the second
2
each node in the plate discretizations has 2 degrees of freedom, i.e. the normal
translational and the rotational displacement components; each node in the cavity
discretizations has 1 pressure degree of freedom
138 Chapter 3
wave model in table 3.2 and the first three FE/FE models in table 3.1. These
figures clearly illustrate that, by increasing the number of elements in the
plate and cavity discretizations (see figures 3.3(a)→(c), 3.4(a)→(c)), the
finite element results gradually converge towards the results, obtained with
a small wave model.
Figure 3.3 : structural frequency response functions

(normal plate displacement over input force) at the excitation point x’=2Lx’/3
(solid : wave 2 / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
These figures illustrate also the typical approximation errors, involved with
finite element discretizations. As already indicated in chapter 1, the low-
order (polynomial) shape functions in each (parent) element can only
represent a dynamic response with a limited spatial variation.
Figure 3.4 : acoustic frequency response functions

(cavity pressure over input force) at cavity location (x,y)=(2Lx/3, Ly/2)
(solid : wave 2 / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
140 Chapter 3
As a result, the dynamic stiffness is usually overestimated, yielding an

overestimation of the resonance frequencies. Since the structural and
acoustic wavelengths decrease with increasing frequency, the spatial
variation of the dynamic response increases also with frequency, so that, for
a given discretization, the overestimation of the resonance frequencies gets
worse for increasing frequency (see figures 3.3 and 3.4). Note also that this
frequency overestimation is less pronounced for the resonances at 121 Hz
and 179 Hz than for the other resonances. This is due to the fact that most of
the energy at these two resonances is stored in the cavity, while at the other
resonances most of the energy is stored in the plate. Since the acoustic
wavelengths are substantially larger than the structural bending
wavelengths, at least in the considered frequency range, the spatial variation
of the dynamic response around cavity-controlled resonance frequencies is
substantially smaller than the spatial variation around plate-controlled
resonance frequencies.
computational efficiency
As illustrated in figures 3.3 and 3.4, the wave based prediction technique
yields highly accurate prediction results with substantially smaller models
than the finite element method, since an approximation error is induced only
on the representation of the boundary conditions. However, as already
indicated in section 2.8., the beneficial effect of a smaller model size on the
computational effort is partly annihilated by some disadvantageous model
properties: in contrast with coupled FE/FE models, wave models are fully
populated and cannot be decomposed into frequency-independent
submatrices. Therefore, it is more appropriate for convergence comparisons
3
to plot the relative prediction errors, as defined in section 2.7.2.1., against
the corresponding CPU times.
Figure 3.5 shows these convergence curves for the normal plate
displacement at x’=Lx’/3 and for the cavity pressure at (x,y)=(Lx/3,Ly/2) for
two excitation frequencies, i.e. 200 Hz and 500 Hz. The CPU times in this
figure denote the times, needed for the direct response calculations at one
frequency. The CPU times, plotted for the 4 wave models of table 3.2,
include both the times, needed for constructing the models as well as for
solving the resulting matrix equations. Since coupled FE/FE models consist
of frequency-independent submatrices, these submatrices must only be
calculated once and the construction of the models at each frequency
3
the prediction results from a large wave model (4 structural and 500 acoustic
wave functions) serve as ‘exact’ reference solutions
requires only a simple submatrix assembly. As a result, the computational

effort for constructing the models is negligible, compared with the efforts
for solving the (large) models, at least when the dynamic response is
calculated in a large frequency range. Therefore, the CPU times, plotted for
the 4 largest FE/FE models of table 3.2, comprise only the direct solution
times. Note that the wave models are implemented in a MATLAB
environment, while the finite element results are obtained with the
MSC/NASTRAN software package. All calculations are performed on a
HP-C180 workstation (SPECfp95=18.7, SPECint95=11.8).
Figure 3.5 : structural (left) and acoustic (right) convergence curves

at 200 Hz (+) and 500 Hz (∇) (solid : wave / dashed : FE/FE)
This figure illustrates the beneficial convergence rate of the proposed wave
based prediction technique. In comparison with the finite element method,
the new prediction technique provides highly accurate predictions of the
coupled vibro-acoustic response with a substantially smaller computational
effort. Note that the beneficial convergence rate will most likely become
even more apparent, when the prediction technique is implemented in a
more efficient software environment, instead of the currently used
MATLAB environment.
In the next section, as well as in chapter 4, similar convergence comparisons
will be performed for several other two- and three-dimensional validation
examples and will confirm the above illustrated beneficial convergence rate
of the wave based prediction technique.
142 Chapter 3
3.3. Interior coupled vibro-acoustic problems with

curved structural components
This section describes how the wave based prediction technique can be
applied for the steady-state dynamic analysis of two-dimensional coupled
vibro-acoustic systems, in which the elastic structural component consists of
an assemblage of circular cylindrical shell sections. Based on the dynamic
equations for a circular cylindrical shell section, appropriate expansions are
defined for the dynamic field variables, which allow the construction of a
coupled vibro-acoustic wave model. Next, the accuracy and the associated
computational effort of the wave based prediction technique and the finite
element method are compared through their application for two validation
examples.
3.3.1. Problem definition
3.3.1.1. Dynamic equations for circular cylindrical shell structures

Figure 3.6 shows an infinitesimal section of a thin, circular cylindrical shell
with radius R and thickness t.
Figure 3.6 : infinitesimal circular cylindrical shell section

By assuming that there are no axial displacement components, the shell

displacements may be expressed in a two-dimensional co-ordinate system
(ϕ,z), in which the shell middle surface is located at z=0. The radial and
circumferential displacements of the shell middle surface are denoted by w
and v, respectively.
According to Love-Timoshenko’s thin, circular cylindrical shell theory (see
e.g. LEISSA (1993)), the in-plane circumferential strains εϕ and normal
stresses σϕ are defined as
1 dv z d 2 w dv E ( 1 + jη )
εϕ = ( + w) − 2 ( 2 − ) and σ ϕ = εϕ , (3.2)
R dϕ R dϕ dϕ 1−ν 2
where E is the modulus of elasticity, ν is the Poisson coefficient and η is the

loss factor of the shell material. Based on the stress and strain expressions
(3.2), the expressions for the resulting normal force N and bending moment
M, acting on the shell cross section (see figure 3.6), become
t /2
E (1 + jη )t dv
N= ∫ σ ϕ . dz = (
( 1 − ν 2 ) R dϕ
+ w) , (3.3)
−t / 2
t /2
E (1 + jη )t 3 d 2w dv
M= ∫ σ ϕ . zdz = − 12(1 − ν 2 ) R 2 ( dϕ 2 − dϕ ) . (3.4)
−t / 2
The expression for the resulting shear force Q (see figure 3.6) is obtained
from the rotational dynamic equilibrium, in which the rotational inertia is
neglected,
1 dM
Q=− . (3.5)
R dϕ
The dynamic equations, governing the steady-state dynamic displacement

components v and w of the shell middle surface, due to an internal pressure
excitation f with a harmonic time dependence with circular frequency ω (see
figure 3.6), result from the dynamic force balances in both the
circumferential and radial directions,
144 Chapter 3
dN 1 dM
− ρ s tRω 2 v = +
dϕ R dϕ
(3.6)
E (1 + jη )t t2 dwd 2v
t 2 d 3w
= [( 1 + ) + − ],
(1 − v 2 ) R 12 R 2 dϕ 2 dϕ 12 R 2 dϕ 3
1 d 2M
− ρ s tRω 2 w = f . R + −N
R dϕ 2
(3.7)
E (1 + jη )t dv t 2 d 4 w d 3v
= f.R − [ w + + ( − )],
(1 − ν 2 ) R dϕ 12 R 2 dϕ 4 dϕ 3
where ρs is the density of the shell material. Note that, in contrast with flat
plates, the out-of-plane motion in circular cylindrical shells is no longer
decoupled from the in-plane motion.
Adding the first derivative of equation (3.6) to equation (3.7) yields
R 2ω 2 2 d 2M d 2N
(1 − ) N = f . R + + (3.8)
c p2 R dϕ 2 dϕ 2
E (1 + jη )
with c p2 = . (3.9)
ρ s (1 − ν 2 )
Subtracting the first derivative of equation (3.6) from the second derivative
of equation (3.7) yields
d 2N R d 2 f 6 R 3ω 2 1 d 2M 1 d 4M
= − M − + . (3.10)
dϕ 2 2 dϕ 2 c 2p t 2 2 R dϕ 2 2 R dϕ 4
Substituting equation (3.10) into equation (3.8) yields
R 2ω 2 R d 2 f 6 R 3ω 2 3 d 2M 1 d 4M
(1 − ) N = f . R + − M + + . (3.11)
c 2p 2 dϕ 2 c p2 t 2 2 R dϕ 2 2 R dϕ 4
Substituting equation (3.11) into equation (3.10) yields the sixth-order

dynamic equation, governing the bending moment M,
1 d6M 1 R 2ω 2 d 4 M
− − ( 1 + )
2 R dϕ 6 R 2c 2p dϕ 4
6 R 3ω 2 1 Rω 2 d 2 M R 2ω 2 6 R 3ω 2
+( − + ) − ( 1 − ) 2 2 M (3.12)
t 2c p2 2 R 2c p2 dϕ 2 c 2p t cp
R R 2ω 2 d 2 f R d 4 f
= (1 + ) + .
2 c p2 dϕ 2 2 dϕ 4
Hence, once the distribution of the bending moment M is known in the shell
structure, all other field variable distributions are known. The distributions
of the normal force N and the shear force Q result from, respectively, (3.11)
and (3.5). The substitution of the resulting bending moment and normal
force distributions into (3.6) and (3.7) yields the distributions of the
circumferential and radial displacement components v and w.
3.3.1.2. Coupled vibro-acoustic problem

Figure 3.7 shows a two-dimensional interior coupled vibro-acoustic system,
in which part of the boundary surface Ωa of the cavity domain V consists a
circular cylindrical4 shell section Ωs (⊆ Ωa).

with circular cylindrical shell section
4
a general, curved shell structure may be regarded - or at least approximated - as
an assemblage of circular cylindrical shell sections
146 Chapter 3
The shell section has a radius R, an arc length R.Φ and its centre point is
located at r0(x0,y0). An external line force F is applied normal to the shell
section at the angular position ϕ=ϕF, while an external acoustic line source q
is applied at position rq(xq,yq) in the cavity. Both excitations have a harmonic
time dependence with circular frequency ω.
As already defined in section 2.2, the steady-state acoustic pressure p at any
position r(x,y) in V is governed by the inhomogeneous Helmholtz equation
∇ 2 p( r ) + k 2 . p( r ) = − jρ 0ωq.δ ( r , rq ), r ∈V. (3.13)
The acoustic boundary condition on the shell surface Ωs is
j ∂p( r )
= jωw( r ), r ∈ Ωs , (3.14)
ρ 0 ω ∂n
where w is the radial displacement component of the shell middle surface.

The boundary conditions on part Ωa\Ωs are defined by equations of type
(2.5)-(2.7).
The steady-state bending moment M of the shell section is governed by the
dynamic equation (3.12), in which the inhomogeneous excitation term is
defined as ( ϕ ∈ [0 ,Φ ] )
f (ϕ ). R = p( x 0 + R.cos(ϕ 0 − ϕ ), y0 + R.sin(ϕ 0 − ϕ )). R + F.δ (ϕ , ϕ F ) . (3.15)
Since the dynamic equation (3.12) is a sixth-order equation, three boundary

conditions must be specified at both boundaries (ϕ=0,ϕ=Φ) of the shell
section.
3.3.2. Field variable expansions
3.3.2.1. Acoustic pressure expansion

As described in section 2.5., the steady-state pressure field p(x,y) in the
acoustic cavity V is approximated as an expansion p(x, y) ,
na
p ( x , y ) ≈ p( x , y ) = ∑ pa .Φ a ( x, y) + pq ( x, y) = [Φ a ]{ pa } + pq ( x, y) .

(3.16)
a=1

The free-field pressure pq (x, y) due to an acoustic line source (see (2.13))
serves as particular solution function for the acoustic source term in the
inhomogeneous right-hand side of the Helmholtz equation (3.13). The
functions Φ a (x, y) in the (1xna) matrix [Φ a ] are the acoustic wave
functions, associated with the (smallest) rectangular domain, circumscribing
the cavity domain5 (see (2.82), (2.83)). The contributions of these wave
functions to the pressure expansion are comprised in the (nax1) vector { pa } .
3.3.2.2. Structural field variable expansions

The steady-state bending moment M(ϕ) in the circular cylindrical shell
section Ωs is approximated as an expansion M(ϕ ) ,

6 na
M (ϕ ) ≈ M (ϕ ) = ∑ Ms .Ψ s( M ) (ϕ ) + ∑ pa . M a (ϕ ) + M F (ϕ ) + Mq (ϕ )

s=1 a=1 (3.17)

[ ] [ ]
= Ψ s( M ) { Ms } + Ma { pa } + M F (ϕ ) + Mq (ϕ ).

The associated force and displacement components are approximated as
[ ] [ ]
N (ϕ ) ≈ N (ϕ ) = Ψ s( N ) { Ms } + N a { pa } + N F (ϕ ) + N q (ϕ ),

(3.18)
Q(ϕ ) ≈ Q(ϕ ) = [Ψ s(Q ) ]{ Ms } + [Qa ]{ pa } + QF (ϕ ) + Qq (ϕ ),

(3.19)
v(ϕ ) ≈ v(ϕ ) = [Ψ s( v ) ]{ M s } + [ va ]{ pa } + v F (ϕ ) + vq (ϕ ),

(3.20)
w(ϕ ) ≈ w(ϕ ) = [Ψ s( w ) ]{ Ms } + [ w a ]{ pa } + w F (ϕ ) + wq (ϕ ).

(3.21)
structural wave functions

The six structural wave functions Ψ s(M) (ϕ ) in the (1x6) matrix Ψ s(M) are [ ]
six linear independent solutions of the homogeneous part of the sixth-order
5
provided that the cavity domain is convex or at least such that any homogeneous
pressure field in the cavity domain may be considered as part of a homogeneous
pressure field, defined in the circumscribing rectangular domain; if not, the cavity
domain must be decomposed into some subdomains, as described in section
2.5.4.
148 Chapter 3
dynamic equation (3.12),
Ψ s( M ) (ϕ ) = e − jk s ϕ , ( s = 1..6 ), (3.22)
where ks (s=1..6) are the six solutions of the characteristic equation
1 6 1 R 2ω 2 4 6 R 3ω 2 1 Rω 2 2 R 2ω 2 6 R 3ω 2
ks − (1 + ) k s − ( − + ) k s − ( 1 − ) 2 2 = 0.
2R R 2c 2p t 2c p2 2 R 2c p2 c 2p t cp
(3.23)
The contributions of these wave functions to the bending moment expansion
are comprised in the (6x1) vector { Ms } .
According to the relations between the bending moments and the associated
force and displacements components in the shell section, given in section
3.3.1.1., the associated wave functions are defined as (s=1..6)
1 3ks2 ks4 6 R 3ω 2 ( M )
Ψ s( N ) (ϕ ) = ( − + − 2 2 )Ψ s (ϕ ) , (3.24)
R 2ω 2 2R 2R t cp
1−
c p2
jk s ( M )
Ψ s(Q ) (ϕ ) = Ψ s (ϕ ) , (3.25)
R
jks 1
Ψ s( v ) (ϕ ) = [Ψ s( N ) (ϕ ) + Ψ s( M ) (ϕ )] , (3.26)
ρ s tRω 2 R
1 ks2 ( M )
Ψ s( w ) (ϕ ) = [Ψ (N)
s (ϕ ) + Ψ s (ϕ )] . (3.27)
ρ s tRω 2 R
particular solution functions

The particular solution functions for the mechanical force term and the
cavity pressure loading term in the inhomogeneous right-hand side of the
dynamic equation (3.12) are based on the dynamic response of a complete
circular cylindrical shell, due to normal line force F, applied at angular
position ϕ=0.
For this dynamic response, the field variable distributions may be expressed
as (ϕ∈[0,2π])
6 6
M F0 (ϕ ) = ∑ Ms,0 .Ψ s( M ) (ϕ ), N F0 (ϕ ) = ∑ Ms,0 .Ψ s( N ) (ϕ ),

s=1 s=1
6
QF0 (ϕ ) = ∑ M s,0 .Ψ s(Q) (ϕ ),

(3.28)
s=1
6 6
v F0 (ϕ ) = ∑ M s,0 .Ψ s( v ) (ϕ ), w F0 (ϕ ) = ∑ M s,0 .Ψ s( w ) (ϕ ).

s=1 s=1
The six constants Ms,0 (s=1..6) in expressions (3.28) result from the six
boundary conditions
dw F0 ( 2π )
! "
dw F0 ( 0)
w F0 ( 0) = w F0 ( 2π ), v F0 ( 0 ) = v F0 ( 2π ), =

,
dϕ dϕ (3.29)
N F0 ( 0 ) = N F0 ( 2π ), M F0 ( 0 ) = M F0 ( 2π ), QF0 ( 0 ) − QF0 ( 2π ) = F.
# $ % & ' (
The response of a complete cylindrical shell, due to a normal line force F,

applied at angular position ϕ=ϕF, is selected as particular solution for the
normal line force excitation of the shell section in the considered coupled
vibro-acoustic system. Similarly, for the particular solution for the cavity
pressure loading of the shell section in the considered coupled vibro-
acoustic system, the response of a complete cylindrical shell, due to this
pressure loading, is selected.
In the expansion (3.21) of the radial displacement component, for instance,
these particular solution functions are
w F (ϕ ) = w F0 (mod(ϕ − ϕ F , 2π )),
) *
(3.30)
w a (ϕ ) =
+
Φ
∫ ( R.Φ a ( x 0 + R.cos(ϕ 0 − ξ ), y0 + R.sin(ϕ 0 − ξ )). w F0 (mod(ϕ − ξ ,2π ))). dξ ,
,
0
(3.31)
wq (ϕ ) =
-
Φ
∫ ( R. pq ( x 0 + R.cos(ϕ 0 − ξ ), y0 + R.sin(ϕ 0 − ξ )). w F0 (mod(ϕ − ξ ,2π ))). dξ ,
. /
0
(3.32)
150 Chapter 3
where mod(*,2π) denotes the congruence modulo 2π. The particular solution
functions in the other expansions (3.17)-(3.20) are defined in a completely
similar way.
3.3.3. Coupled vibro-acoustic wave model
The contributions pa (a=1..na) and Ms (s=1..6) of the acoustic and structural

wave functions to the solution expansions (3.16)-(3.21) are determined by
the acoustic and structural boundary conditions.
As outlined in section 2.4.1., the structural boundary conditions in two-
dimensional coupled vibro-acoustic problems are defined at discrete
locations and can be used as such. For the considered problem, the six
boundary conditions, defined at both boundaries of the circular cylindrical
shell section, yield a set of 6 equations in the (6+na) unknown wave function
contributions,
Ms 
[ Ass Csa ].   = { fs } . (3.33)
î pa 
The acoustic boundary conditions, defined at any position on the cavity

boundary surface Ωa, must be transformed into an integral formulation.
According to the procedures, described in sections 2.4.1.1. and 2.4.1.2, a set
of na equations in the (6+na) unknown wave function contributions is
obtained from either a weighted residual formulation of the acoustic
boundary conditions,
[Cas ]
( WR ) 
{ }
Ms 
( WR ) ( WR )
Aaa + Caa .   = fa( WR ) , (3.34)
î pa 
or from a least-squares formulation of the acoustic boundary conditions,
[Cas ]
( LS ) 
{ }
Ms 
( LS )
Aaa .   = fa( LS ) . (3.35)
î pa 
The combination of the structural boundary conditions (3.33) and the

weighted residual (3.34) or least-squares (3.35) formulation of the acoustic
boundary conditions yields a coupled vibro-acoustic wave model for the

considered two-dimensional interior coupled vibro-acoustic system.
3.3.4. Validation examples
3.3.4.1. Cavity-backed circular cylindrical shell section

Figure 3.8 shows a line force excited circular cylindrical shell section,
backed by an acoustic cavity. The boundaries of the shell section are
clamped to the rigid side walls of the cavity. The geometrical parameters are
R=Ly=0.707 m, Φ=π/2 rad, t=0.002 m, Lx=1 m and ϕF=π/4 rad. The shell is
made of aluminium (E=70.109 N/m2, ρs=2790 kg/m3, ν=0.3, η=0) and the
cavity is filled with air (c=340 m/s, ρ0=1.225 kg/m3).
Figure 3.8 : cavity-backed circular cylindrical shell section
Due to the clamping of the shell section, the six structural boundary
conditions are
dw( 0 ) dw(Φ )
w( 0) = v( 0) = = w(Φ ) = v(Φ ) = = 0. (3.36)
dϕ dϕ
Since three cavity side walls are rigid, the fluid displacement, normal to
these walls, must be zero. Along the cavity-shell interface, the radial shell
and normal fluid displacement components must be identical, as defined in
(3.14). Based on these structural and acoustic boundary conditions, a
152 Chapter 3
weighted residual wave model is built, using 6 structural wave functions and
52 acoustic wave functions, associated with the rectangular domain LxxLy,
circumscribing the considered cavity (see figure 3.8).
The coupled vibro-acoustic response is calculated for the case of a unit line
force with frequency 230 Hz. Figure 3.9 shows the resulting instantaneous
shell displacement field.
Figure 3.9 : instantaneous shell displacement field at 230 Hz
Figure 3.10 plots the resulting radial shell displacement against the fluid
displacement, normal to the cavity-shell interface. These figures illustrate
that the small wave model yields an accurate representation of the structural
boundary conditions (3.36) and the continuity condition (3.14).
Figure 3.10 : instantaneous radial shell displacement (solid)

and normal fluid displacement (+-marks) at 230 Hz
Figure 3.11 shows the contour plot of the resulting instantaneous cavity
pressure. The fact that these contour lines are (almost) perpendicular to the
rigid cavity side walls illustrates the accurate representation of the acoustic
boundary conditions on these side walls.
Figure 3.11 : contour plot of the instantaneous cavity pressure at 230 Hz
Since all boundary conditions are accurately represented and since the
proposed field variable expansions ensure that the dynamic equations are
always exactly satisfied, it may be concluded that the small wave model
enables an accurate prediction of the steady-state dynamic response of the
considered coupled vibro-acoustic problem.
3.3.4.2. Fluid-loaded circular cylindrical shell

Figure 3.12 shows a line force excited circular cylindrical shell, which is
supported at two positions. The geometrical parameters are R=0.25 m,
Φ=4π/3 rad, Ψ=2π/3 rad, t=0.002 m, Lx=2R, Ly=2R and ϕF=2π/3 rad. The
shell is made of aluminium (E=70.109 N/m2, ρs=2790 kg/m3, ν=0.3, η=0).
The cases are considered of an air-filled (c=340 m/s, ρ0=1.225 kg/m3) and a
water-filled (c=1440 m/s, ρ0=1000 kg/m ) cavity.
3
For modelling purposes, the circular cylindrical shell is considered as an

assemblage of two shell sections, one with arc length R.Φ and one with arc
length R.Ψ (see figure 3.12). In this way, the effects of the supports on the
dynamic response can be modelled as structural boundary conditions. Since
the cylinder consists of two shell sections, twelve instead of six structural
boundary conditions must be satisfied,
154 Chapter 3
Figure 3.12 : fluid-loaded circular cylindrical shell
w1 ( 0 ) = v1 ( 0 ) = w1 (Φ ) = v1 (Φ ) = w 2 ( 0 ) = v 2 ( 0 ) = w 2 (Ψ ) = v 2 (Ψ ) = 0,
dw1 ( 0 ) dw 2 (Ψ ) dw1 (Φ ) dw 2 ( 0)
= , = , M1 ( 0) = M 2 (Ψ ), M1 (Φ ) = M 2 ( 0 ).
dϕ dψ dϕ dψ
(3.37)
The acoustic boundary conditions are defined by the continuity condition
(3.14) for the radial shell displacement and the normal fluid displacement
along the cavity-shell interface. Based on these structural and acoustic
boundary conditions, a weighted residual wave model is built, using 12
structural wave functions (6 per shell section) and 80 acoustic wave
functions, associated with the rectangular domain LxxLy, circumscribing the
considered cavity (see figure 3.12).
The coupled vibro-acoustic response is calculated for a unit line force
excitation at 500 Hz. Figure 3.13 shows the resulting instantaneous shell
displacement fields for both the air-loaded and the water-loaded cylinder. It
can be seen from these displacement fields that, for both cases, the first 10
structural boundary conditions in (3.37) are accurately represented. Figure
3.13 shows also the instantaneous shell displacement field for an uncoupled,
in-vacuo cylinder. The comparison of the three displacement fields in this
figure illustrates the sensitivity of the vibro-acoustic coupling interaction to

the fluid density. As indicated by the coupling measure λc, defined in (1.1),
the air in the cavity, having a low density, hardly influences the shell
displacement field at this fairly high excitation frequency, while the high-
density water in the cavity has still a significant influence on the shell
displacement field. The stronger fluid-structure coupling interaction in the
water-loaded cylinder results also in substantially larger cavity pressure
amplitudes, as shown in figure 3.14.
Figure 3.13 : instantaneous displacement fields at 500 Hz

of the in-vacuo (a), the air-loaded (b) and the water-loaded (c)
circular cylindrical shell
156 Chapter 3
Figure 3.14 : contour plots of the instantaneous cavity pressure fields at 500 Hz
of the air-loaded (left) and the water-loaded (right) circular cylindrical shell
Figure 3.15 plots, for the two fluid-loaded cylinders, the resulting radial
shell displacements against the fluid displacements, normal to the cavity-
shell interface. Figure 3.16 shows the resulting bending moments in those
cylinders. These figures illustrate that the small wave model yields an
accurate representation of the acoustic boundary conditions (3.14) and the
last 2 structural boundary conditions in (3.37).
Since all boundary conditions are accurately represented and since the
proposed field variable expansions ensure that the dynamic equations are
always exactly satisfied, it may be concluded, again, that the small wave
model enables an accurate prediction of the steady-state dynamic response
of the considered coupled vibro-acoustic problem.
3.3.5. Comparison with coupled FE/FE models
In a similar way as presented in section 3.2.2., the accuracy and associated

computational efforts of the wave models for both validation examples,
described in section 3.3.4., are compared with the performances of
corresponding Eulerian coupled FE/FE models.
Figure 3.15 : instantaneous radial shell displacements (solid)

and normal fluid displacements (+-marks) at 500 Hz
of the air-loaded (a) and the water-loaded (b) circular cylindrical shell
158 Chapter 3
Figure 3.16 : instantaneous bending moments at 500 Hz

in the air-loaded (a) and the water-loaded (b) circular cylindrical shell
accuracy comparison
The accuracy of the weighted residual wave model for the cavity-backed
circular cylindrical shell section, described in section 3.3.4.1., is compared
with the accuracy of two coupled FE/FE models. As indicated in section
3.2.2., both element discretizations for the considered two-dimensional
problem consist of one three-dimensional layer of 8-noded hexahedral fluid

elements (CHEXA) for the cavity and 4-noded quadrilateral plate elements
(CQUAD4) for the shell section, in which the invariance in the third
dimension is imposed through additional multipoint constraints. Table 3.3
lists the number of elements and the corresponding number of unconstrained
degrees of freedom6 (dof) in both coupled FE/FE models and the number of
wave functions in the considered wave model.
FE/FE 1 FE/FE 2 wave

# elem. dof # elem. dof dof
shell 20 57 60 177 6
cavity 200 231 1800 1891 52
total dof 288 2068 58
Table 3.3 : properties of the prediction models

for the cavity-backed circular cylindrical shell section
For the air-loaded circular cylindrical shell, described in section 3.3.4.2.,

two coupled FE/FE models are built, in which the shell is discretized into 4-
noded quadrilateral plate elements and the cavity into a combination of 8-
noded hexahedral and 6-noded pentahedral fluid elements (CPENTA). The
model properties of these coupled FE/FE models and the considered wave
model are listed in table 3.4.
FE/FE 1 FE/FE 2 wave
# elem. dof # elem. dof dof
shell 30 87 60 447 12
cavity 300 310 1500 1510 80
total dof 397 1957 92
Table 3.4 : properties of the prediction models

for the air-loaded circular cylindrical shell
Figures 3.17/3.18 and figures 3.19/3.20 show some structural and acoustic
frequency response functions (with a frequency resolution of 1 Hz, except
around the resonance frequencies, where a finer resolution is used), obtained
6
each node in the shell discretizations has 3 degrees of freedom, i.e. the in-plane
and out-of-plane translational and the rotational displacement components; each
node in the cavity discretizations has 1 pressure degree of freedom
160 Chapter 3
with the prediction models for, respectively, the cavity-backed circular

cylindrical shell section (see table 3.3) and the air-loaded circular cylindrical
shell (see table 3.4).
Figure 3.17 : structural frequency response functions (radial shell displacement over
input force) at the excitation point ϕ=π/4 of the cavity-backed circular cylindrical
shell section (thick: wave / thin: FE/FE 1 (upper), FE/FE 2 (lower) (see table 3.3))
These figures confirm the observations from the previous accuracy

comparison, discussed in section 3.2.2. The finite element results gradually
converge towards the prediction results, obtained with a small wave model.
Figure 3.18 : acoustic frequency response functions (cavity pressure over input force)
at location (x,y)=(0.5 m,0.4 m) of the cavity-backed circular cylindrical shell section
(thick: wave / thin: FE/FE 1 (upper), FE/FE 2 (lower) (see table 3.3))
162 Chapter 3
Figure 3.19 : structural frequency response functions (radial shell displacement over
input force) at the excitation point ϕ=2π/3 of the air-loaded circular cylindrical shell
(thick: wave / thin: FE/FE 1 (upper), FE/FE 2 (lower) (see table 3.4))
The discretization errors, involved with finite element models, induce an

overestimation of the resonance frequencies, which gets worse for
increasing frequency. Again, this frequency overestimation is smaller for the
cavity-controlled resonances than for the shell-controlled resonances. In the
considered frequency range of the line force excitation, cavity-controlled
resonances occur around 257 Hz, 348 Hz and 436 Hz in the cavity-backed
circular cylindrical shell section (see figures 3.17 and 3.18) and around 399
Hz in the air-loaded circular cylindrical shell (see figures 3.19 and 3.20).
at location (x,y)=(0.25 m,0.4 m) of the air-loaded circular cylindrical shell (thick:
wave / thin: FE/FE 1 (upper), FE/FE 2 (lower) (see table 3.4))
computational efficiency
To illustrate the beneficial convergence rate of the proposed wave based
prediction technique, the accuracy and associated computational efforts of
three coupled FE/FE models (see table 3.5) and three weighted residual
164 Chapter 3
wave models (see table 3.6) are compared for the cavity-backed circular
cylindrical shell problem.

shell 60 177 100 297 200 597
cavity 1800 1891 5000 5151 25000 25251
total dof 2068 5448 25848
Table 3.5 : properties of the coupled FE/FE models

wave 1 wave 2 wave 3

dof dof dof
shell 6 6 6
cavity 26 78 130
total dof 32 84 136
Table 3.6 : properties of the weighted residual wave models

The relative7 prediction errors of the various prediction models in tables 3.5
and 3.6 are compared with the corresponding CPU times. Figure 3.21 shows
these convergence curves for the radial shell displacement component at the
excitation point ϕ=π/4 and for the cavity pressure at (x,y)=(0.5 m,0.4 m) for
an excitation frequency of 300 Hz. In a similar way as in section 3.2.2., the
indicated CPU times for the wave models include both the times, needed for
constructing and solving the models at one frequency, while the indicated
CPU times for the FE/FE models comprise only the direct solution times.
Again, the wave models are implemented in a MATLAB environment,
while the finite element results are obtained with the MSC/NASTRAN
software package. All calculations are performed on a HP-C180 workstation
(SPECfp95=18.7, SPECint95=11.8).
This figure confirms the observations from section 3.2.2. in that the
proposed wave based prediction technique provides highly accurate
predictions of the coupled vibro-acoustic response with a smaller
7
the prediction results from a large wave model (6 structural and 312 acoustic
wave functions) serve as ‘exact’ reference solutions
computational effort. Note that this beneficial convergence rate for coupled
vibro-acoustic problems with a curved structural component is less
pronounced than for problems with a flat structural component (see figures
3.5 and 3.21). This is mainly due to the fact that the structural particular
solution functions wa , which account for the pressure loading, associated
with the acoustic wave functions in the pressure expansion, can be defined
in a direct form for the case of a flat structural component (see (2.19)),
while these particular solution functions are defined in an integral form for
the case of a curved structural component (see (3.31)), which is
computationally more demanding.
Figure 3.21 : structural (left) and acoustic (right) convergence curves at 300 Hz
(solid : wave (see table 3.6)/ dashed: FE/FE (see table 3.5))
3.4. Interior coupled vibro-acoustic problems with

poroelastic insulation material
Poroelastic insulation materials are widely used in noise control engineering
as damping layers for e.g. aircraft and automobile interior noise control and
to line anechoic chambers. The effect of a poroelastic damping layer is
commonly modelled as a frequency dependent, but spatially invariant
normal impedance boundary condition for the cavity pressure field.
This section describes how the wave based prediction technique can be used
for the prediction of the steady-state dynamic response of fluid-saturated
166 Chapter 3
poroelastic materials. The technique is applied for some validation examples

to illustrate the convergence of the proposed wave model and to assess the
validity of the spatially invariant impedance assumption.
3.4.1.1. Dynamic equations

Fluid-saturated poroelastic materials have two phases, i.e. the solid phase
and the fluid phase, contained within the pores, formed by the solid phase.
In the classical Biot theory (BIOT (1956)), it is assumed that the pores are
homogeneously distributed in the poroelastic material and that the dynamic
deformations in both phases may be described on a macroscopic level, using
the modelling principles, used in isotropic continuum mechanics.
When it is assumed that the dynamic deformations have a harmonic time
dependence with circular frequency ω, the mean macroscopic displacement
of the solid phase in a two-dimensional poroelastic material sample may be
described by the vector field
u x ( x, y ) jωt
u ( x , y, t ) =  . e

(3.38)
î uy ( x, y)
and the mean macroscopic displacement of the fluid phase may be described
by the vector field
U x ( x , y ) jωt
U ( x, y, t ) = 

. e . (3.39)
î U y ( x, y )
According to the Biot theory (see BIOT (1956) and appendix E), these
steady-state dynamic displacement fields are governed by the vector
equations
N . ∇ 2u + ∇[( A + N ). es + Q. ε ] = −ω 2 ( ρ *11 . u + ρ *12 . U ) ,

(3.40)
∇[ Q. es + R. ε ] = −ω 2 ( ρ *12 . u + ρ *22 .U ) ,

(3.41)
where es and ε are, respectively, the volumetric strain in the solid phase,
∂u x ∂u y
es = ∇. u = +

, (3.42)
∂x ∂y
and the volumetric strain in the fluid phase,
∂U x ∂U y
ε = ∇.U =

+

. (3.43)
∂x ∂y
The constant N in (3.40) is the shear modulus,
E1
N= , (3.44)
2(1 + ν )
where E1 is the in-vacuo modulus of elasticity of the bulk solid phase and ν
is its Poisson coefficient. To account for internal friction losses, the modulus
of elasticity of the solid phase can be made complex, E1=Em(1+jη), where Em
is the real, static modulus and η is a loss factor, that accounts for the
mechanical damping, associated with the motion of the solid phase of the
material.
The constant A in (3.40) is the first Lamé constant,
ν . E1
A= . (3.45)
(1 + ν )(1 − 2ν )
The coefficient Q in (3.41) represents the coupling between the volume

change of the solid and that of the fluid, and may be defined as
Q = (1 − h) E2 , (3.46)
where h is the porosity, i.e. the ratio of the fluid volume to the total material
volume, and where E2 is the modulus of elasticity of the bulk fluid phase. To
account for the losses, associated with heat transfer from the fluid within the
pores to the pore walls, this modulus of elasticity can be made complex. An
expression for this complex modulus is given e.g. in BOLTON et al. (1996)
and appendix E.
The constant R in (3.41) relates the fluid stress and strain, and may be
defined as
168 Chapter 3
R = hE 2 . (3.47)
where ρ1 is the bulk density of the solid phase and where ρ2=h.ρ0 is the bulk
density of the fluid phase with ρ0 being the ambient fluid density.
The constants ρ11
*
, ρ12
*
and ρ22
*
in (3.40) and (3.41) are
b
ρ *11 = ρ 11 + , ρ 11 = ρ 1 + ρ a , (3.48)
jω
b
ρ *12 = ρ 12 − , ρ 12 = − ρ a , (3.49)
jω
b
ρ *22 = ρ 22 + , ρ 22 = ρ 2 + ρ a . (3.50)
jω
The constant ρ11 represents the effective mass density of the solid moving in
the fluid and is equal to the real solid mass density ρ1 plus an additional
mass density ρa to account for the inertial coupling forces. These forces
result from the momentum transfer between the solid and the fluid, due to
the pore tortuosity and variation of the pore diameter. The latter effects may
be quantified by the geometrical structure factor ε‘. This factor is equal to
unity, if the pores are straight and uniform (in which case there is no inertial
coupling), and increases as the pores become irregularly constricted and
more tortuous. With this geometrical structure factor ε‘, the inertial coupling
forces, which are proportional to the relative acceleration of both phases,
may be quantified by the inertial coupling parameter ρa (ALLARD et al.
(1989), BOLTON et al. (1996)),
ρ a = ρ 2 (ε ’−1) . (3.51)
Similarly, the effective fluid mass density ρ22 is the sum of the real fluid
mass density ρ2 and the coupling parameter ρa. The constants ρ11
*
, ρ12
*
and
ρ22
*
are the counterparts of ρ11,, ρ12 and ρ22, when allowance is made for the
viscous energy dissipation, resulting from the relative motion between the
solid and the fluid phases of the poroelastic material. The viscous coupling
forces, which are proportional to the relative velocity of both phases, are
quantified by the viscous coupling factor b, which may be related to the
macroscopic flow resistivity σ of the porous material as indicated in

ALLARD et al. (1989), BOLTON et al. (1996) and appendix E.
As indicated in appendix E, two wave equations may be obtained from the

vector equations (3.40) and (3.41). The first equation governs the dilatation
es of the solid phase (see (3.42)),
∇ 4 es + A1 . ∇ 2es + A2 . es = 0 , (3.52)
with
A1 = ω 2 ( ρ *11 R − 2ρ *12Q + ρ *22 P ) / ( PR − Q 2 ) , (3.53)
A2 = ω 4 [ ρ *11ρ *22 − ( ρ *12 ) 2 ] / ( PR − Q 2 ) , (3.54)
where P=A+2N. The fact that equation (3.52) is a fourth-order wave

equation indicates the existence of two types of dilatational waves with
distinct wavenumbers k1 and k2,
A1 A12 A1 A12
k1 = + − A2 , k2 = − − A2 . (3.55)
2 4 2 4
The second wave equation,
∇ 2ω s + kt2 .ω s = 0 , (3.56)
governs the rotational strain ωs in the solid phase,
∂u y ∂u x
ωs = − , (3.57)
∂x ∂y
and indicates the existence of one rotational wave type with wavenumber kt,
ρ *11ρ *22 − ( ρ *12 ) 2

kt = ω . (3.58)
N . ρ *22
170 Chapter 3
Once the distributions es(x,y) and ωs(x,y) of, respectively, the volumetric and
rotational strains in the solid phase of the poroelastic material are known, all
distributions of the steady-state dynamic field variables are completely
defined. As indicated in appendix E, the distributions of the displacement
components of the solid phase result from,
ux = −
1 ∂
A2 ∂x
( 1 ∂ω s
∇ 2es + A1. es + 2
kt ∂y
, ) (3.59)
uy = −
1 ∂
A2 ∂y
( 1 ∂ω s
∇ 2es + A1. es − 2
kt ∂x
, ) (3.60)
and the distributions of the displacement components of the fluid phase

result from,
a ∂
Ux = − 1
A2 ∂x
( ∂e
) ρ * ∂ω s
∇ 2es + A1 . es + a 2 . s − * 12 2
∂x ρ 22 . kt ∂y
. (3.61)
a ∂
Uy = − 1
A2 ∂y
( ∂e
) ρ * ∂ω s
∇ 2es + A1. es + a 2 . s + * 12 2
∂y ρ 22 . kt ∂x
. (3.62)
( ρ *11 R − ρ *12Q) ( PR − Q 2 )
with a1 = , a2 = . (3.63)
( ρ *22Q − ρ *12 R) ω 2 ( ρ *22Q − ρ *12 R)
The stress distributions result from
σ x = ( A + a1Q)es + a 2Q∇ 2es −

2N ∂ 2
A2 ∂x 2
(
∇ 2
e s + A .
1 se + )
2 N ∂ 2ω s
kt2 ∂x∂y
, (3.64)
σ y = ( A + a1Q)es + a 2Q∇ 2es −

2N ∂ 2
A2 ∂y 2
(
∇ 2
e s + A .
1 se − )
2 N ∂ 2ω s
kt2 ∂x∂y
, (3.65)
τ xy = τ yx = −
2N ∂ 2
A2 ∂x∂y
(
∇ 2es + A1. es + 2 (
kt ∂y 2
)
N ∂ 2ω s ∂ 2ω s
−
∂x 2
), (3.66)
s = ( a1 R + Q)es + a 2 R∇ 2es . (3.67)

The stresses σx and σy represent the forces per unit material area, acting in,
respectively, the x- and y-direction on the solid phase, while τxy represents
the shear force per unit material area, acting on the solid phase in the y-
direction in a plane, whose normal vector is oriented in the x-direction. The
fluid stress s represents the force per unit material area, acting on the fluid
phase, and is assumed to be negative in compression.
3.4.1.2. Boundary conditions

Since the dynamic equations (3.52) and (3.56) are, respectively, fourth-order
and second-order equations, the distributions of the volumetric and
rotational strains of the solid phase and, hence, all field variable
distributions in a bounded poroelastic domain Vp are uniquely defined, if
three boundary conditions are specified at each point on the boundary Ωp of
the domain Vp. Two major types of boundary conditions are considered
(Ωp=Ωp1∪Ωp2).
• kinematic boundary conditions :
un ( r ) = un ( r ), r ∈ Ω p1 , (3.68)
us ( r ) = us ( r ), r ∈ Ω p1 , (3.69)
U n ( r ) = U n ( r ), r ∈ Ω p1 , (3.70)
where un , Un and us are prescribed functions for, respectively, the

solid and fluid displacement components, in the direction, normal to the
boundary Ωp1, and the solid displacement component in the direction,
tangent to the boundary Ωp1. These kinematic boundary conditions are
specified, for instance, along the parts of the boundary surface, which are
bonded to a rigid surface ( un ≡0, Un ≡0 and us ≡0).
• mechanical boundary conditions :
σ n ( r ) = σ n ( r ), r ∈ Ω p 2 , (3.71)
τ ns ( r ) = τ ns ( r ), r ∈ Ω p 2 , (3.72)
s( r ) = s ( r ), r ∈ Ω p 2 , (3.73)
172 Chapter 3
where σ n , τ ns and s are prescribed functions for, respectively, the

boundary normal and shear stresses of the solid phase and the boundary
fluid stress. These mechanical boundary conditions are specified, for
instance, along the parts of the boundary surface, which are in contact
with an inviscid fluid in an acoustic domain. In this case, the boundary
conditions for the poroelastic domain may be expressed as (see e.g.
BOLTON et al. (1996)),
σ n ( r ) = −(1 − h ). p( r ), r ∈ Ω p 2 , (3.74)
τ ns ( r ) = 0, r ∈ Ω p 2 , (3.75)
s( r ) = − h. p( r ), r ∈ Ω p 2 , (3.76)
where h is the porosity and where p is the pressure field in the acoustic
domain. In addition, the velocity continuity condition in the normal
direction γn of the common boundary surface between the poroelastic and
acoustic domain serves as boundary condition for the acoustic domain,
j ∂p( r )
= jω [ h.U n ( r ) + (1 − h ). un ], r ∈ Ω p 2 . (3.77)
ρ 0ω ∂γ n
The steady-state volumetric strain es(x,y) in the solid phase of a two-

dimensional poroelastic domain Vp is approximated as a solution expansion

es (x, y) ,
ne
es ( x , y ) ≈ es ( x , y ) = ∑ Ese .Ψ e ( x, y) = [Ψ e ]{Ese } .

(3.78)
e=1
Each function Ψ e (x, y) in the (1xne) matrix [Ψ e ] is a dilatational wave

function, which satisfies the dynamic equation (3.52),
− j ( k xe . x + k ye . y )
Ψ e ( x, y ) = e (3.79)
with 2
k xe + k ye
2
= k12 or k xe
2
+ k ye
2
= k 22 . (3.80)
The contributions of these dilatational wave functions to the solution

expansion for the volumetric strain in the solid phase are comprised in the
(nex1) vector {Ese } .
Similarly, the steady-state rotational strain ωs(x,y) in the solid phase of a
two-dimensional poroelastic domain Vp is approximated as a solution
expansion ω s (x, y) ,

nw
ω s ( x, y ) ≈ ω s ( x, y ) = ∑ Wsw .Ψ ω ( x, y) = [Ψ ω ]{Wsw } .

(3.81)
w =1
Each function Ψ ω (x, y) in the (1xnw) matrix [Ψ ω ] is a rotational wave

function, which satisfies the dynamic equation (3.56),
− j ( k xw . x + k yw . y )
Ψ ω ( x, y ) = e (3.82)
with 2
k xw + k yw
2
= kt2 . (3.83)
The contributions of these rotational wave functions to the solution

expansion for the rotational strain in the solid phase are comprised in the
(nwx1) vector {Wsw } .
Based on the expansions (3.78) and (3.81) for the distributions of the
volumetric and rotational strains in the solid phase of a two-dimensional
fluid-saturated poroelastic material, the associated expansions for the
distributions of the displacement and stress components result from the
substitution of (3.78) and (3.81) into, respectively, (3.59)-(3.62) and (3.64)-
(3.67). For the x-component of the solid displacement, for instance, the
expansion is
[ ]  Ese 
u x ( x, y ) ≈ u x ( x, y ) = Ψ u x  ,

(3.84)
î Wsw 
where the (1x(ne+nw)) matrix Ψ u x [ ] is

[Ψ u ] = [ Leu
x x
[Ψ e ] Lωu x [Ψ ω ] ] (3.85)
174 Chapter 3
with the differential operators
Leu x = −
1 ∂
A2 ∂x
( ) 1 ∂
∇ 2 + A1 , Lωu x = 2
kt ∂y
. (3.86)
3.4.3. Wave model
weighted residual formulation of the boundary conditions

Since the field variable expansions, proposed in the previous subsection,
satisfy a priori the dynamic equations (3.52) and (3.56), the wave function
contributions Ese (e=1..ne) and Wsw (w=1..nw) are only determined by the
boundary conditions, defined in (3.68)-(3.73). Depending on the wave
function contributions, some residual error functions are involved with the
representation of the boundary conditions,
[ ] n
 Ese 
Ru n ( r ) = un ( r ) − un ( r ) = Ψ u ( r )   − un ( r ), r ∈ Ω p1 ,
î Wsw 
(3.87)
R ( r ) = u ( r ) − u ( r ) = [Ψ ( r )] 
E  se
 − us ( r ), r ∈ Ω p1 ,

us s s us (3.88)
î W sw 
R ( r ) = U ( r ) − U ( r ) = [Ψ ( r )] 
E se 
 − U n ( r ), r ∈ Ω p1 ,

Un n n Un (3.89)
î W sw 
R ( r ) = σ ( r ) − σ ( r ) = [Ψ ( r )] 
E se 
 − σ n ( r ), r ∈ Ω p 2 ,

σn n n σn (3.90)
î W sw 
R ( r ) = τ ( r ) − τ ( r ) = [Ψ ( r )] 
E se 
 − τ ns ( r ), r ∈ Ω p 2 ,

τ ns ns ns τ ns (3.91)
î W sw 
 Ese 
Rs ( r ) = s ( r ) − s ( r ) = Ψ s ( r )  [ ]
 − s ( r ), r ∈ Ω p 2 .

(3.92)
î Wsw 
The (1x(ne+nw)) wave function matrices in these residual error functions are
defined as (φ=un, us, Un, σn, τns, s)
[Ψ φ ] = [ Lφe [Ψ e ] Lω
φ [Ψ ω ] ] (3.93)
with the differential operators

Leu n = −
1 ∂
A2 ∂γ n
(
∇ 2 + A1 , Lωun = 2
1 ∂
)
kt ∂γ s
, (3.94)
Leu s = −
1 ∂
A2 ∂γ s
( )
1 ∂
∇ 2 + A1 , Lωu s = − 2
kt ∂γ n
, (3.95)
a ∂
LeU n = − 1
A2 ∂γ n
(
∇ 2 + A1 + a 2
∂
)
∂γ n
, Lω
Un = −
ρ *12 ∂
ρ 12 . kt ∂γ s
* 2
, (3.96)
Lσe n = ( A + a1Q + a 2Q∇ 2 ) −

2N ∂ 2
A2 ∂γ n2
∇ 2
+ (
A1 , Lω
σn = )
2N ∂ 2
kt2 ∂γ n∂γ s
, (3.96)
Leτ ns = −
2N ∂ 2
A2 ∂γ n∂γ s
( ∇ 2 + A1 , Lω )
τ ns =
N ∂2
( −
∂2
kt2 ∂γ s2 ∂γ n2
), (3.97)
Les = ( a1 R + Q + a 2 R∇ 2 ), Lωs = 0, (3.98)
where γn and γs are, respectively, the normal and tangential directions of the
boundary surface.
As proposed in section 2.4.1.1., a weighted residual formulation of the
boundary conditions (3.68)-(3.73) may be obtained from orthogonalising the
residual error functions (3.87)-(3.92) with respect to some weighting
functions,
~ ~ ~
∫ σ n . Run . dΩ + ∫ τ ns . Ru s . dΩ + ∫ s . RU n . dΩ
Ω p1 Ω p1 Ω p1
~ (3.99)
~ ~
− ∫ un . Rσ n . dΩ − ∫ us . Rτ ns . dΩ − ∫ U n . Rs . dΩ = 0,
~
in which the weighting functions φ (φ=un, us, Un, σn, τns, s) are expanded in
terms of the same set of wave functions, used in the field variable
expansions,
~
E 
~
[ ]
φ ( x, y ) = Ψ φ  ~se  .
î Wsw 
(3.100)
By imposing the condition that the weighted residual formulation (3.99)

~
should hold for any weighting function expansion, i.e. for any Ese (e=1..ne)
176 Chapter 3
~
and Wse (w=1..nw), a weighted residual wave model for the unknown wave
function contributions Ese and Wsw is obtained,
[ App ]îWswse  = { f p } .

E
(3.101)
The element on row i and column j of the ((ne+nw)x(ne+nw)) matrix A pp is [ ]

[ App ]ij = ∫Ψ σ n ,i .Ψ un , j . dΩ + ∫Ψ τ ns ,i .Ψ u s , j . dΩ + ∫ Ψ s,i .Ψ U n , j . dΩ
− ∫Ψ un ,i .Ψ σ n , j . dΩ − ∫ Ψ u s ,i .Ψ τ ns , j . dΩ − ∫Ψ U n ,i .Ψ s,i . dΩ .
(3.102)
It follows from the reciprocity relation, discussed in section E.5 of appendix
E, that this matrix is symmetric, i.e.
[ App ]ij = [ App ] ji (∀i, j = 1..( ne + nw )) . (3.103)
The element on row i of the ((ne+nw)x1) vector f p is { }

{ f p }i = ∫ Ψ σ n ,i . un . dΩ + ∫ Ψ τ ns ,i . us . dΩ + ∫ Ψ s,i .U n . dΩ
(3.104)
− ∫ Ψ u n ,i .σ n . dΩ − ∫ Ψ u s ,i .τ ns . dΩ − ∫ Ψ U n ,i . s . dΩ .
wave function selection

For the dilatational and rotational wave functions Ψ e and Ψ ω (see (3.79)
and (3.82)) in the field variable expansions, the following wavenumber
selection is proposed,
 2 2 
( mπ ,± k 2 −  mπ  ), ( mπ ,± k 2 −  mπ  ),
 Lx 1
 Lx  Lx
2
 Lx  
 
( k xe , k ye ) ∈   (3.105)
2 2
  n π  n π  n π  n π 
( ± k12 −   , ), ( ± k 22 −   , ) 
î  Ly  Ly  Ly  Ly 
 2 
 mπ 2  mπ 
2
 
2  nπ  nπ 
( k xw , k yw ) ∈ ( ,± k t −   ),( ± kt −   , ) (3.106)
 Lx  Lx   Ly  L y 
î 
where m and n are integer values (m,n=0,±1, ±2,...). The dimensions Lx and
Ly are the dimensions of a rectangular domain, circumscribing the
considered poroelastic domain, provided that this poroelastic domain is
convex or at least such that any homogeneous dynamic response in the
domain can be regarded as part of a homogeneous dynamic response in a
rectangular domain, circumscribing the domain. If the domain geometry
doesn’t satisfy this condition, the domain must be decomposed into some
subdomains, whose geometries satisfy this condition and for each of which a
wave function selection of type (3.105)-(3.106) is defined (cfr. section 2.5).
Although a mathematical proof is still required, the similarity of the
proposed wavenumber selection with the wavenumber selections, proposed
for acoustic and plate bending problems (see sections 2.5, 4.2 and 4.3),
along with the practical convergence, obtained with the proposed wave
model in its application for various validation problems, provide already a
strong indication of the completeness of the proposed wave function
selection.
coupling with a coupled vibro-acoustic wave model

When a poroelastic material is in contact with the inviscid fluid in the
acoustic domain of a coupled vibro-acoustic system, a wave model of type
(3.101) for the prediction of the steady-state dynamic response in the
poroelastic material, must be linked with the acoustic part of a wave model
of type (2.41) or (2.48) for the prediction of the steady-state dynamic
response of the coupled vibro-acoustic system. This wave model linking is
based on the substitution of the acoustic pressure field expansion in the
residual error functions, associated with the poroelastic boundary conditions
(3.74)-(3.76), and on the substitution of the poroelastic field variable
178 Chapter 3
expansions in the residual error function, associated with the acoustic

boundary condition (3.77).
3.4.4. Validation example
To illustrate the accuracy of the proposed wave model for fluid-saturated

poroelastic materials, the two-dimensional coupled vibro-acoustic problem,
shown in figure 3.22, is considered. One side wall of an air-filled (c=340
m/s, ρ0=1.225 kg/m3) rectangular cavity (Lx=1 m, Ly=0.5 m) consists of an
aluminium (E=70.109 N/m2, ν=0.3, ρs=2790 kg/m3, η=0) plate of thickness
t=0.002 m. The plate has clamped boundaries and is excited by a time-
harmonic normal line force F. The cavity side wall, opposite to the elastic
plate, is treated with a layer of air-saturated polyurethane foam material
(Em= 80.104 N/m2, η=0.265, ν=0.4, ρ1=30 kg/m , h=0.9, ε‘=7.8, σ=25.10
3 3
Ns/m4, Lx’=Lx=1 m, Ly’=0.2 m), which is bonded to a rigid surface. The two
other side walls of the cavity are perfectly rigid.
Figure 3.22 : coupled vibro-acoustic system with poroelastic insulation material
A weighted residual wave model is built for the prediction of the steady-
state dynamic response of this coupled vibro-acoustic system, excited by a
unit normal line force at location x=Lx/4 at 200 Hz. For the expansions of the
poroelastic field variables, 144 dilatational and 72 rotational wave functions
are used. The expansion for the normal plate displacement consists of 4
structural wave functions and the cavity pressure field expansion consists of
60 acoustic wave functions. Note that, for the considered homogeneous
pressure field in the rectangular cavity domain, the acoustic wave functions
can be selected, such that not only the governing Helmholtz equation, but
also the zero normal fluid velocity conditions on both rigid cavity side walls
are exactly satisfied a priori (cfr. section 2.5.2.).
Figures 3.23 and 3.24 show the prediction results for the real and imaginary
parts of the steady-state dynamic displacement fields in the solid and fluid
phase of the foam layer.
Figure 3.23 : real (a) and imaginary (b) parts of the steady-state dynamic
displacement fields in the solid phase of the foam layer (200 Hz)
Figure 3.24 : real (a) and imaginary (b) parts of the steady-state dynamic
displacement fields in the fluid phase of the foam layer (200 Hz)
180 Chapter 3
These figures illustrate that the boundary conditions of type (3.68)-(3.70),

imposing zero normal and tangential solid displacements and zero normal
fluid displacements along the parts of the foam surface, which are bonded to
the rigid surface, are accurately represented.
Figure 3.25 shows the contour plot of the prediction results for the
amplitudes of the solid shear stresses, defined in the x’- and y’- co-ordinate
directions (see figure 3.22). This figure illustrates the accurate
representation of the boundary condition (3.75) along the cavity-foam
interface, which imposes a zero boundary shear stress, due to the inviscid
nature of the air in the cavity.
Figure 3.25 : contour plot of the amplitudes of the steady-state shear stress τx’y’ in the
solid phase of the foam layer (200 Hz)
Figures 3.26 and 3.27 illustrate that the continuity conditions (3.74), (3.76)
and (3.77), imposed on the cavity-foam interface, are accurately
represented. Figure 3.28 shows the real and imaginary parts of the
prediction results for the normal plate displacement and the fluid
displacement component, normal to the cavity-plate interface. This figure
illustrates that the displacement continuity condition along the cavity-plate
interface, as well as the clamped plate boundary conditions are accurately
represented.
Since all governing dynamic equations are exactly satisfied a priori and
since all boundary conditions are accurately represented, it may be
concluded that the proposed wave model enables an accurate prediction of
the steady-state dynamic response of the considered coupled vibro-acoustic
system.
Figure 3.26 : validation of the continuity conditions (3.74) and (3.76), imposed on the
acoustic (solid) and the poroelastic (+-marks) field variables along the cavity-foam
interface (200 Hz)
3.4.5. Accuracy of a spatially invariant impedance model
The effect of a layer of poroelastic insulation material, used as damping

treatment for an acoustic cavity, is commonly modelled as a normal
impedance boundary condition for the pressure field in the cavity. The
impedance value is usually assumed to be frequency dependent, but
spatially invariant.
Since the proposed wave based prediction technique enables an accurate
prediction of the dynamic response of both phases in a poroelastic material,
the damping effect of a poroelastic insulation layer can be modelled without
the use of an impedance model. In this way, the prediction technique allows
the validation of the spatial invariance assumption of an impedance model.
182 Chapter 3
Figure 3.27 : validation of the continuity condition (3.77), imposed on the acoustic
(solid) and poroelastic (+-marks) response along the cavity-foam interface (200 Hz)
Figure 3.28 : real (upper) and imaginary (lower) parts of the steady-state normal
plate displacement (solid) and the fluid displacement component (+-marks), normal
to the cavity-plate interface (200 Hz)
For this validation purpose, the coupled vibro-acoustic system, shown in

figure 3.29, is considered. One side wall of an air-filled (c=340 m/s,
ρ0=1.225 kg/m3) rectangular cavity (Lx=1 m, Ly=0.45 m) consists of an
aluminium (E=70.109 N/m2, ν=0.3, ρs=2790 kg/m3, η=0) plate of thickness
t=0.002 m. The plate has clamped boundaries and is excited by a time-
harmonic normal line force F at location x0=Lx/4. The cavity side wall,
opposite to the elastic plate, is treated with a layer of air-saturated
polyurethane foam material (Em= 80.104 N/m2, η=0.265, ν=0.4, ρ1=30 kg/m3,
h=0.9, ε‘=7.8, σ=25.103 Ns/m4) of thickness Ly’=0.05 m, while the two other
cavity side walls are perfectly rigid. The foam layer is also in contact with a
rectangular air gap, which has two rigid side walls and which couples to
another aluminium plate with clamped boundaries. The two remaining
boundary surfaces of the foam layer are bonded to a rigid surface.
Figure 3.29 : coupled vibro-acoustic system

with an air gap-backed poroelastic foam layer
Three different cases are considered, regarding the backing of the foam
layer :
• a quasi-rigid backing, consisting of a very thin air gap (Ly”=0.001 m) and
a second plate, having an infinite bending stiffness,
• a rigid air gap backing, consisting of an air gap with the same thickness
as the foam layer (Ly’=Ly”=0.05 m) and a second plate, having an infinite
bending stiffness,
184 Chapter 3
• a flexible air gap backing, consisting of an air gap with the same
thickness as the foam layer (Ly’=Ly”=0.05 m) and a second plate, which is
identical to the line force excited plate.
The steady-state dynamic responses for these three foam-backing cases in

the considered coupled vibro-acoustic system are predicted with the
proposed wave based prediction technique. Although not illustrated here, the
accuracy of the used wave model has been identified from a validation of
the boundary condition representation, in a completely similar way as
presented for the validation example in the previous subsection.
The validity of the spatial invariance assumption of the impedance of the
foam layer can be assessed by plotting the amplitudes of the pressure along
the cavity-foam interface against the amplitudes of the normal fluid
velocity, multiplied with a spatially averaged impedance Z ,
 Lx   L x j ∂ p( x , L y ) 
Z =  ∫ p( x, L y ) dx  /  ∫ dx  . (3.107)
 0   0 ρ 0ω ∂y 
The deviation between both curves provides a measure for the spatial
variation of the impedance, since both curves should perfectly match, if the
normal impedance is spatially invariant.
Figure 3.30 shows these validation curves at two excitation frequencies, i.e.
200 Hz and 1000 Hz. For the quasi-rigid foam backing (see figures 3.30
(a,d)) and for the rigid air gap backing (see figures 3.30 (b,e)), the
deviations between both curves and, hence, the spatial variation of the foam
layer impedances are fairly small. For the flexible air gap backing (see
figures 3.30 (c,f)), the spatial variation of the foam layer impedance
becomes significant, especially at low frequencies, where strong vibro-
acoustic coupling effects occur.
To illustrate the approximation errors, involved with using a spatially
invariant impedance model, the actual cavity pressure field at 200 Hz for the
case of the flexible air gap backing (see figure 3.30 (c)) is compared with
the cavity pressure field, obtained from modelling the flexible air gap-
backed foam layer as a spatially invariant impedance
Z = Z r + jZi (3.108)
Figure 3.30 : amplitudes of the pressure (solid) and the normal fluid velocity,
multiplied with the spatially averaged impedance Z (dashed), along the cavity-foam
interface for the case of a quasi-rigid backing (a,d), a rigid air gap backing (b,e) and
a flexible air gap backing (c,f) ((a-c) 200 Hz, (d-f) 1000 Hz)
186 Chapter 3
with
   
 Lx p ( x , L )   Lx p ( x , L ) 
1  1 
∫ Re( j ∂p( x, Ly ) )dx , Zi = L  ∫ Im( j ∂p( x, Ly ) )dx  .

y y
Zr =
Lx  0 x 0
   
 ρ 0ω ∂y   ρ 0ω ∂y 
(3.109)
The value of this impedance is obtained from the highly accurate prediction
results of the wave model. Figure 3.31 shows the resulting cavity pressure
fields and clearly illustrates that the assumption of a spatially invariant
impedance model may induce a substantial approximation error.
Figure 3.31 : contour plots of the amplitudes of the cavity pressure field at 200 Hz in
the coupled vibro-acoustic system with a flexible air gap-backed foam layer (upper)
and the cavity pressure field, obtained from approximating the foam layer as a
spatially invariant impedance boundary condition for the cavity (lower)
3.5. Exterior coupled vibro-acoustic problems

As already discussed in section 1.3.2.2., an exterior coupled vibro-acoustic
problem, involving the acoustic radiation in an unbounded fluid domain,
may be approximated as an interior problem. This is done by confining the
unbounded fluid domain to a bounded domain through the introduction of an
artificial boundary surface, on which a particular, energy absorbing
impedance boundary condition is imposed.
A simple, but commonly used approximation procedure consists of defining
a geometrically simple boundary surface in the exterior domain and using
the specific acoustic impedance ρ0c of freely propagating plane waves as
impedance boundary condition to relate the pressure p to the normal fluid
velocity vn on this boundary surface (see 1.30).
To illustrate this approximation procedure, the two-dimensional problem is
considered of a normal line force excited, cavity-backed plate, which is
baffled in an infinitely extended, rigid wall and which radiates into a semi-
infinite fluid domain (see figure 3.32).
Figure 3.32 : coupled vibro-acoustic problem with an unbounded acoustic domain

188 Chapter 3
The semi-infinite fluid domain is confined to a bounded domain by

introducing a half-cylindrical boundary surface with radius R, on which the
specific acoustic impedance ρ0c is imposed (see figure 3.32). To illustrate
that the proposed wave based prediction technique enables an accurate
prediction of this approximated problem, the following numerical example
is considered.
An aluminium plate (E=70.109 N/m2, ν=0.3, ρs=2790 kg/m3, η=0, t=0.002
m) with clamped boundaries is excited at 1000 Hz by a unit normal line
force F at location x0=Lx/4. The cavity domain (Lx=1 m, Ly=0.5 m) and the
half-cylindrical (R=1.5 m) fluid domain are filled with air (c=340 m/s,
ρ0=1.225 kg/m3). For this case, a wave model is built, which consists of 4
structural wave functions for the expansion of the normal plate
displacement, 60 acoustic wave functions for the pressure field expansion in
the rectangular cavity and 270 acoustic wave functions for the pressure field
expansion in the half-cylindrical fluid domain. The latter wave functions are
associated with the rectangular domain 2RxR, circumscribing the half-
cylindrical fluid domain.
Again, a verification of the boundary condition representation reveals the
high accuracy of the proposed wave based prediction technique. Figure 3.33
shows, for example, that the zero normal fluid displacement condition for
the exterior pressure field along the rigid baffle and the normal displacement
continuity condition along the plate surface are accurately represented.
Figure 3.34 illustrates the high accuracy, obtained for the impedance
boundary condition, imposed on the half-cylindrical boundary surface in the
exterior acoustic domain.
As already indicated in section 1.3.2.2., the efficiency of the above
illustrated approximation procedure for exterior problems is rather low. For
the proposed impedance boundary condition to be a reliable approximation
of the Sommerfeld radiation condition (1.8), i.e. to represent a (quasi) non-
reflective boundary surface for the acoustic waves in the exterior domain,
the artificially introduced boundary surface should be located in the far-
field. The far-field may be characterised by the satisfaction of three criteria
(JUNGER and FEIT (1972)),
λ πl 2
r >> , r >> l, r >> , (3.110)
2π 2λ
where r is the distance from the radiating object, λ is the acoustic

wavelength and l is a characteristic dimension of the radiating object.
As a result, the distance between the artificially introduced boundary surface

and the actual boundary surface of the coupled vibro-acoustic system must
be large (for the considered example, the radius R of the half-cylindrical
boundary surface should be at least 5m). Consequently, the bounded exterior
fluid domain must still be large, which results in large prediction models.
Figure 3.33 : coupled vibro-acoustic response at 1000 Hz

((a) instantaneous exterior pressure in the plane of the baffle, (b) instantaneous
normal fluid displacement in the plane of the baffle (solid line) and normal plate
displacement (+-marks), (c) instantaneous pressure along the cavity-plate interface)
For this reason, one of the next research steps in the development of the
proposed prediction technique will focus on the enhancement of the
modelling efficiency for exterior problems. On the one hand, the use of a
190 Chapter 3
more advanced impedance boundary condition for the artificially introduced

boundary surface, as proposed by KELLER and GIVOLI (1989) and briefly
described in section 1.3.2.2, will be explored. On the other hand, it will be
investigated whether, in a similar way as in the boundary element method,
the introduction of an artificial boundary surface can be avoided. In this
respect, the possibility will be explored of defining an additional wave
function truncation for the pressure field expansion in an unbounded
acoustic domain, such that the Sommerfeld radiation condition is inherently
incorporated in the wave model.
Figure 3.34 : amplitudes of the pressure (solid) and normal fluid velocity,
multiplied by the impedance value ρ0c (+-marks), along the half-cylindrical boundary
surface (1000 Hz)
3.6. Conclusions
This chapter presents the application of the wave based prediction technique
for the modelling of the steady-state dynamic behaviour of the main
components that may be encountered in two-dimensional coupled vibro-
acoustic problems.
Through various validation problems, it is illustrated how the prediction

methodology, proposed in the previous chapter, allows an accurate
prediction of the steady-state dynamic response of two-dimensional coupled
vibro-acoustic systems, involving both flat and curved structural
components. It is also indicated how the prediction methodology can be
applied for the accurate prediction of the steady-state dynamic behaviour of
both phases in fluid-saturated poroelastic materials. As a result, the effect of
of a layer of poroelastic insulation material, used as damping treatment for
an acoustic cavity, for instance, can be assessed with a full dynamic model
of the insulation layer. This avoids the use of an impedance model, for
which it is illustrated that the commonly used assumption of a spatially
invariant impedance may induce some substantial approximation errors. For
exterior problems, involving unbounded acoustic domains, the prediction
technique, at least in its current stage of development, requires the
transformation of the exterior problem into an approximated interior
problem. This is obtained by confining the unbounded acoustic domain to a
bounded domain through the introduction of an artificial boundary surface,
on which an energy absorbing impedance boundary condition is imposed.
In order to assess the computational efficiency of the new wave based
prediction technique in comparison with the existing finite element method,
the accuracy and the associated computational efforts, involved with a wave
model and a corresponding coupled FE/FE model, are compared for various
validation examples. These comparisons reveal the beneficial convergence
rate of the proposed wave based prediction technique, in that highly accurate
predictions are obtained with a substantially smaller computational effort.
This beneficial convergence rate will also be identified in the next chapter,
which addresses the application of the prediction technique for three-
dimensional coupled vibro-acoustic problems.
192 Chapter 3
4. APPLICATION OF THE WAVE BASED
PREDICTION TECHNIQUE FOR THREE-
DIMENSIONAL COUPLED VIBRO-ACOUSTIC
PROBLEMS
4.1. Introduction
This chapter presents the application of the wave based prediction technique
for the steady-state dynamic analysis of three-dimensional coupled vibro-
domain consists of a thin, flat plate.
Section 4.2. indicates some specific modelling aspects, involved with the
use of the prediction methodology, outlined in chapter 2, for the steady-state
dynamic analysis of the out-of-plane bending motion in thin, flat plates. It
describes how a set of structural wave functions, together with a particular
solution function, provide an expansion for the normal displacement of the
plate middle surface, which exactly satisfies the dynamic plate equation.
This section describes also how a convergent, symmetric wave model for
the determination of the wave function contributions in the proposed
solution expansion is obtained from a weighted residual formulation of the
boundary conditions. In contrast with a complete wave function set, recently
proposed by LANGLEY (1997), the poor condition of the proposed wave
model does not prevent the numerical results from converging towards the
exact solution, as illustrated for some validation examples.
194 Chapter 4
Section 4.3 defines a complete set of acoustic wave functions for the
expansion of the steady-state cavity pressure field in three-dimensional
uncoupled interior acoustic systems. Based on this function set, the
construction of a convergent wave model is performed in a completely
similar way as for two-dimensional problems. The accuracy of the resulting
uncoupled acoustic wave models is illustrated for a validation example, for
which the wave model results are compared with experimental results, as
well as prediction results from some finite element models.
Section 4.4. describes how a structural and an acoustic wave model, defined
in the two previous sections, are combined into a convergent wave model
for the steady-state dynamic analysis of three-dimensional coupled vibro-
domain consists of a thin, flat plate. For several validation examples, the
prediction results, obtained from these coupled wave models, are compared
with the results, obtained from corresponding Eulerian coupled FE/FE
models. This comparison demonstrates the beneficial convergence rate of
the wave based prediction technique, in that it provides highly accurate
prediction results with a smaller computational effort than the finite element
method. This section presents also an experimental validation of the wave
based prediction technique, applied for the identification of the modal
properties of a double-panel partition, which is a typical coupled vibro-
acoustic system that is often encountered in noise control engineering.
4.2. Uncoupled plate problems
Figure 4.1. shows a thin, flat plate Ωs, in which an out-of-plane bending
motion1 is induced by an external point2 force F, applied at location rsF(xF,yF)
in the direction, normal to the plate middle surface. The force excitation has
a harmonic time dependence with circular frequency ω. The thickness of the
1
the wave based prediction technique can also be applied for the prediction of the
in-plane motion in flat plates, but this is not described here, since the vibro-
acoustic coupling interaction of a plate with the surrounding fluid influences only
the out-of-plane plate displacement
2
the general case of a distributed force excitation may be regarded as a
superposition of point forces
Application for three-dimensional coupled vibro-acoustic problems 195
plate is t. The plate material has a density ρs, a Poisson coefficient ν, an

elasticity modulus E and a material loss factor η.
Figure 4.1 : uncoupled flat plate problem
According to Kirchhoff’s thin plate theory (see appendix B), the steady-state
displacement w of the plate middle surface, in the direction normal to the
plate surface, is governed by the dynamic equation
F
∇ 4 w( rs ) − kb4 . w( rs ) = .δ ( rs , rsF ), rs ( x, y ) ∈ Ω s , (4.1)
D
where
∂4 ∂4 ∂4
∇4 = + 2 + (4.2)
∂x 4 ∂x 2∂y 2 ∂y 4
and where the plate bending stiffness D and the bending wavenumber kb are
Et 3 (1 + jη ) ρ tω 2
D= , kb = 4 s . (4.3)
12(1 − ν 2 ) D
Since the dynamic equation (4.1) is a fourth-order equation, the

displacement field w(x,y) is uniquely defined, if two boundary conditions
are specified at each point on the plate boundary Γs. By defining the
differential operators
196 Chapter 4
∂
Lθ = − , (4.4)
∂γ n
 ∂2 ∂2 
Lm = − D 2 + v 2  , (4.5)
 ∂γ n ∂γ s 
∂  ∂2 ∂2 
LQ = − D  + ( 2 − v ) , (4.6)
∂γ n  ∂γ n2 ∂γ s2 
where γn and γs are, respectively, the normal and tangential directions of the
plate boundary (see figure 4.1), the following expressions are obtained for
the three major types of boundary conditions that may be specified on the
plate boundary Γs ( Γ s = Γ wθ ∪ Γ mQ ∪ Γ wm ).
w( rs ) = w ( rs ), rs ∈ Γ wθ , (4.7)
Lθ [ w( rs )] = θ n ( rs ), rs ∈Γ wθ , (4.8)
where w and θ n are prescribed translational and rotational displacement

functions. Note that boundary condition (4.7) imposes also a prescribed
∂w
value for the rotational displacement component θ s = − . These
∂γ s
kinematic boundary conditions are specified, for instance, at clamped
plate boundaries.

These boundary conditions impose prescribed functions qn , mn and
mns for the shear forces and the bending and torsional moments along
the boundary. Since only two boundary conditions can be applied, the
prescribed shear force and torsional moment functions are combined into
one generalised shear force function Qn (see appendix B),
Lm [ w( rs )] = mn ( rs ), rs ∈ Γ mQ , (4.9)
∂mns ( rs )
LQ [ w( rs )] = Qn ( rs ) = qn ( rs ) + , rs ∈ Γ mQ . (4.10)
∂γ s
These boundary conditions are specified, for instance, at free plate

boundaries.
• mixed boundary conditions :

At some parts of the plate boundary, one kinematic and one mechanical
boundary condition may be specified. For simply supported plate
boundaries, for instance, the normal displacement w and the bending
moment mn have prescribed values,
w( rs ) = w ( rs ), rs ∈ Γ wm , (4.11)
Lm [ w( rs )] = mn ( rs ), rs ∈ Γ wm . (4.12)
4.2.2. Field variable expansion
The steady-state normal plate displacement w(x,y) is approximated as a

solution expansion w(x, y) ,
ns
w( x, y ) ≈ w ( x, y ) = ∑ ws .Ψ s ( x, y) + w F ( x, y) = [Ψ s ]{ws } + w F ( x, y) . (4.13)

s=1
Each function Ψ s (x, y) in the (1xns) matrix [Ψ s ] is a structural wave

function, which satisfies the homogeneous part of the dynamic equation
(4.1),
− j ( k xs . x + k ys . y )
Ψ s ( x, y ) = e (4.14)
(k xs2 + kys2 )
2
with = kb4 . (4.15)
The contributions of these structural wave functions to the solution

expansion are comprised in the (nsx1) vector {w s } .

Function w F (x, y) is a particular solution function for the external force

term in the inhomogeneous right-hand side of the dynamic equation (4.1).
From all possible mathematical expressions, the physically meaningful
expression for the displacement of an infinite plate, excited by a normal
point force F, is selected as particular solution function (JUNGER and FEIT
(1972)),
198 Chapter 4
w F ( x, y ) = −
jF
[H ( 2) ( 2)
]
0 ( kb r ) − H 0 ( − jk b r )

(4.16)
8 kb2 D
with r = ( x − x F ) 2 + ( y − yF ) 2 (4.17)
and where H 0(2) is the zero-order Hankel function of the second kind.
From the above definitions, it may be concluded that the proposed
displacement expansion (4.13) satisfies a priori the dynamic plate equation
(4.1), no matter what the values of the structural wave function
contributions ws are.
4.2.3. Weighted residual wave model
The contributions ws of the structural wave functions to the solution

expansion (4.13) are determined by the boundary conditions, defined in
(4.7)-(4.12). These boundary conditions may be transformed into a weighted
residual or a least-squares formulation. This section describes only the
construction of a weighted residual model for the determination of the wave
function contributions. The construction of a Hermitian least-squares wave
model follows a completely similar procedure, as outlined in section 2.4.1.2.
However, as it is the case for two-dimensional coupled vibro-acoustic
problems, illustrated in section 2.7.2.3., a least-squares wave model for
uncoupled plate problems has a smaller convergence rate than a
corresponding weighted residual wave model.
weighted residual formulation of the boundary conditions

Depending on the wave function contributions ws in expansion (4.13), some
residual error functions are involved with the representation of the boundary
conditions,
Rw ( rs ) = w( rs ) − w ( rs ), rs ∈( Γ wθ ∪ Γ wm ),

(4.18)
Rθ ( rs ) = Lθ [ w( rs )] − θ n ( rs ), rs ∈Γ wθ ,

(4.19)
Rm ( rs ) = Lm [ w( rs )] − mn ( rs ), rs ∈ (Γ mQ ∪ Γ wm ),

(4.20)
RQ ( rs ) = LQ [ w( rs )] − Qn ( rs ), rs ∈ Γ mQ .

(4.21)
For the sake of model symmetry, which will become apparent later on, some
additional residuals are defined at the corner points of the plate boundary Γs,
at which the normal and tangential directions are not uniquely defined (see
figure 4.2).
Figure 4.2 : normal and tangential directions at a corner point
For each of the nw corner points, which belong to the boundary part Γwθ∪Γwm,
on which kinematic or mixed boundary conditions are specified, the
additional residual is the approximation error for the normal plate
displacement at the corner point location rsc,
Rcw ( rsc ) = w( rsc ) − w ( rsc ), c = 1.. nw .

(4.22)
For each of the nF corner points, which belong to the boundary part ΓmQ, on
which mechanical boundary conditions are specified, the additional residual
is the approximation error for the concentrated corner point force, which is
associated with the discontinuity of the torsional moment,
RcF ( rsc )
(
= LF [ w ] − mns ( rsc+ ) − mns ( rsc− ) )

 ∂ 2 w( r ) ∂ 2 w( rsc ) 
( )

= − D(1 − ν ) + sc+ −  − mns ( rsc+ ) − mns ( rsc− ) , c = 1.. n F .

− − 
 ∂γ n ∂γ s ∂γ n ∂γ s 
(4.23)
In the proposed weighted residual formulation of the boundary conditions,
the above defined residual error functions are orthogonalised with respect to
some weighting functions, which are obtained from applying the differential
operators, involved with the definition of the various residuals, on a
weighting function w~,
200 Chapter 4
~ ~ ~
∫ LQ [ w ]. Rw . dΓ + ∫ Lm [ w ]. Rθ . dΓ − ∫ Lθ [ w ]. Rm . dΓ
Γ wθ + Γ wm Γ wθ Γ wm + Γ mQ
nw nF (4.24)
~. R . dΓ +
− ∫w Q ∑ LF [ w~ ]. Rcw ( rsc ) − ∑ w~(rsc ). RcF ( rsc ) = 0.
Γ mQ c=1 c=1
As proposed in section 2.4.1.1., the structural wave function contributions ws

are determined, such that the weighted residual formulation (4.24) is
satisfied for any weighting function w ~ , which can be expanded in terms of
the same set of structural wave functions Ψ s , used in the displacement
expansion (4.13). This yields a weighted residual wave model of the form
[ Ass ]{ws } = { fs } . (4.25)
The element on row si and column sj of the (nsxns) matrix [ Ass ] is
[ Ass ]si s j = ∫ LQ [Ψ si ].Ψ s j . dΓ + ∫ Lm [Ψ si ]. Lθ [Ψ s j ]. dΓ

Γ wθ + Γ wm Γ wθ
− ∫ Lθ [Ψ si ]. Lm [Ψ s j ]. dΓ − ∫ Ψ si . LQ [Ψ s j ]. dΓ (4.26)
Γ wm + Γ mQ Γ mQ
nw nF
+ ∑ LF [Ψ si ].Ψ s j ( rsc ) − ∑ Ψ si ( rsc ). LF [Ψ s j ]
c=1 c=1
and the element on row si of the (nsx1) vector { fs } is
{ f s } si = ∫ LQ [Ψ si ].( w − w F ). dΓ + ∫ Lm [Ψ si ].(θ n − Lθ [ w F ]). dΓ

− ∫ Lθ [Ψ si ].( mn - Lm [ w F ]). dΓ − ∫ Ψ si .(Qn − LQ [ w F ]). dΓ

Γ wm + Γ mQ Γ mQ
nw
+ ∑ LF [Ψ si ].( w ( rsc ) − w F ( rsc ))

c=1
nF
− ∑ LF [Ψ si ].( mns ( rsc+ ) − mns ( rsc- ) − LF [ w F ])

(4.27)
c=1
model properties
As already indicated in section 2.4.2., matrix [ Ass ] is a fully populated
matrix, whose elements are complex and frequency dependent. The
symmetry of this matrix is proven as follows.
Consider two structural wave functions Ψ s i and Ψ s j , which satisfy the
homogeneous part of the dynamic equation (4.1), i.e.
∇ 2 (∇ 2Ψ s i ) = kb4 .Ψ s i and ∇ 2 (∇ 2Ψ s j ) = kb4 .Ψ s j . (4.28)
Since these wave functions are smooth and non-singular in the plate surface
domain Ωs with boundary Γs, the two-dimensional version of Green’s second
identity (see equation (2.51)) implies that
∫ Ψ si . ∇ (∇ 2Ψ s j ). dΩ =
2
Ωs
(4.29)
∂ (∇ 2Ψ s j ) ∂Ψ si
∫ ∇ Ψ s i . ∇ Ψ s j . dΩ + ∫ Ψ s i . . dΓ − ∫ ∂γ . ∇ Ψ s j . dΓ
2 2 2
Ωs Γs
∂γ n Γs n
After some rearrangements of the integration functions, this expression

becomes
∫ Ψ si . ∇ (∇ 2Ψ s j ). dΩ =
2
Ωs
∂ ∂ Ψsj ∂ 2Ψ s j
2
∫ ∇ Ψ si . ∇ Ψ s j . dΩ + ∫ Ψ si . ∂γ ( 2 + ( 2 − ν ) ). dΓ
2 2
Ωs Γs n ∂γ n ∂γ s2
(4.30)
∂Ψ si ∂ 2Ψ s j ∂ 2Ψ s j ∂ ∂ 2Ψ s j
− ∫ ∂γ .( +ν ). dΓ − ∫ Ψ s i . ((1 − ν ) ). dΓ
Γs n ∂γ n2 ∂γ s2 Γs
∂γ n ∂γ s2
∂Ψ si ∂ 2Ψ s j
− ∫ ∂γ .((1 − ν ) ). dΓ .
Γs n ∂γ s2
By using the differential operators, defined in (4.4)-(4.6), this expression
may be written as
202 Chapter 4
∫ Ψ si . ∇ (∇ 2Ψ s j ). dΩ = ∫ ∇ Ψ si . ∇ Ψ s j . dΩ
2 2 2
Ωs Ωs
1
−
DΓ
∫ (Ψ si . LQ [Ψ s j ] + Lθ [Ψ si ]. Lm [Ψ s j ]). dΓ (4.31)
s
∂ 3Ψ s j ∂Ψ si ∂ Ψ s j
2
−(1 − ν ) ∫ (Ψ si . + . ). dΓ .
Γs ∂γ n∂γ s2 ∂γ n ∂γ s2
Since the plate boundary Γs is a closed curve, the application of integration

by parts on the last line integral term in (4.31) (see e.g. REKTORYS (1977))
yields,
∂ 3Ψ s j ∂Ψ si ∂ Ψ s j
2
−(1 − ν ) ∫ (Ψ si . + . ). dΓ =
Γs ∂γ n ∂γ s2 ∂γ n ∂γ s2
∂Ψ si ∂ Ψ s j ∂ 2Ψ si ∂Ψ s j
2 n +n
1 w F
(1 − ν ) ∫ (
∂γ
. +
∂γ ∂γ
. ). dΓ − ∑ LF [Ψ s j ].Ψ si
Γs s ∂γ n ∂γ s n s ∂γ s D c =1
nw + n F ∂Ψ si ( rsc ) ∂Ψ s j ( rsc ) ∂Ψ si ( rsc ) ∂Ψ s j ( rsc )
+(1 − ν ) ∑ . − . .
c=1 ∂γ n+ ∂γ s+ ∂γ n− ∂γ s−
(4.32)
The substitution of (4.28) and (4.32) into (4.31) yields
k b4 ∫ Ψ s i .Ψ s j . dΩ =
Ωs
1
∫ ∇ Ψ si . ∇ Ψ s j . dΩ − D ∫ (Ψ si . LQ [Ψ s j ] + Lθ [Ψ si ]. Lm [Ψ s j ]). dΓ
2 2
Ωs Γs
∂Ψ si ∂ Ψ s j ∂ 2Ψ si ∂Ψ s j
2
1 nw + nF
(1 − ν ) ∫ ( . + .
∂γ s ∂γ n ∂γ s ∂γ n ∂γ s ∂γ s
). dΓ − ∑ LF [Ψ s j ].Ψ si ( rsc )
D c=1
Γs
 ∂Ψ si ( rsc ) ∂Ψ s j ( rsc ) ∂Ψ s i ( rsc ) ∂Ψ s j ( rsc ) 

nw + n F
+(1 − ν ) ∑ 
 ∂γ + . ∂γ + − −
. −


c=1  n s ∂γ n ∂γ s 
(4.33)
Interchanging the indices i and j in (4.33) yields
k b4 ∫ Ψ s j .Ψ s i . dΩ =
Ωs
1
∫ ∇ Ψ s j . ∇ Ψ si . dΩ − D ∫ (Ψ s j . LQ [Ψ si ] + Lθ [Ψ s j ]. Lm [Ψ si ]). dΓ
2 2
Ωs Γs
∂Ψ s j ∂ 2Ψ si ∂ 2Ψ s j ∂Ψ si 1 nw + n F
(1 − ν ) ∫ ∂γ( .
∂γ n ∂γ s ∂γ s
+ . ). dΓ − ∑ LF [Ψ si ].Ψ s j ( rsc )
Γs s ∂γ n ∂γ s D c=1
nw + nF ∂Ψ s j ( rsc ) ∂Ψ si ( rsc ) ∂Ψ s j ( rsc ) ∂Ψ si ( rsc ) 
+(1 − ν ) 
∑ − 
 ∂γ + . ∂γ + ∂γ −
.
∂γ − 
c=1  n s n s 
(4.34)
Since the left-hand sides of expressions (4.33) and (4.34) are identical, their
right-hand sides are also identical. By equating these right-hand sides and
using the following identity,
∂Ψ si ∂Ψ s j ∂Ψ si ∂Ψ s j ∂Ψ s j ∂Ψ si ∂Ψ s j ∂Ψ si
− − +
∂γ n+ ∂γ s+ ∂γ n− ∂γ −s ∂γ n+ ∂γ +s ∂γ n− ∂γ −s
 ∂Ψ s i ∂Ψ si   + ∂Ψ s j ∂Ψ s j 
=  n x+ + n y+   − n y + n x+ 
 ∂x ∂y   ∂x ∂y 
 ∂Ψ s i ∂Ψ si   − ∂Ψ s j ∂Ψ s j 
−  n x− + n y−   − n y + n x− 
 ∂x ∂y   ∂x ∂y  (4.35)
 ∂Ψ s j ∂Ψ s j   ∂Ψ si ∂Ψ si 
−  n x+ + n y+   − n y+
 + n x+ 
 ∂x ∂y   ∂x ∂y 
 ∂Ψ s j ∂Ψ s j   − ∂Ψ s i ∂Ψ si 
+  n x− + n y−   −ny
 + n −
x 
 ∂x ∂y  ∂x ∂y 
=0,
where (n x+ ,n y+ ) and (n x− ,n y− ) are the unit vectors in the normal directions

γ n+ and γ n− , respectively, the following expression is obtained
204 Chapter 4
nw + n F
∫ (Ψ s j . LQ [Ψ si ] + Lθ [Ψ s j ]. Lm [Ψ si ]). dΓ + ∑ LF [Ψ si ].Ψ s j ( rsc )
Γs c=1
nw + n F
= ∫ (Ψ si . LQ [Ψ s j ] + Lθ [Ψ si ]. Lm [Ψ s j ]). dΓ + ∑ LF [Ψ s j ].Ψ si ( rsc ).
Γs c=1
(4.36)
Since Γ s = Γ wθ ∪ Γ mQ ∪ Γ wm , this expression may be written as
∫ LQ [Ψ si ].Ψ s j . dΓ + ∫ Lm [Ψ si ]. Lθ [Ψ s j ]. dΓ
− ∫ Lθ [Ψ si ]. Lm [Ψ s j ]. dΓ − ∫ Ψ si . LQ [Ψ s j ]. dΓ
Γ wm + Γ mQ Γ mQ
nw nF
+ ∑ LF [Ψ s i ].Ψ s j ( rsc ) − ∑ Ψ si ( rsc ). LF [Ψ s j ]
c=1 c=1
= ∫ LQ [Ψ s j ].Ψ si . dΓ + ∫ Lm [Ψ s j ]. Lθ [Ψ si ]. dΓ
− ∫ Lθ [Ψ s j ]. Lm [Ψ si ]. dΓ − ∫ Ψ s j . LQ [Ψ si ]. dΓ
Γ wm + Γ mQ Γ mQ
nw nF
+ ∑ LF [Ψ s j ].Ψ s i ( rsc ) − ∑ Ψ s j ( rsc ). LF [Ψ si ].
c=1 c=1
(4.37)
It follows from the matrix element definition formula (4.26) that expression
(4.37) proves the symmetry of matrix [ Ass ] , i.e.
[ Ass ]si s j = [ Ass ]s j si (∀si , s j = 1.. ns ) . (4.38)
4.2.4. Convergence requirements
complete wave function set

As outlined in section 2.5., a convergent displacement expansion (4.13) is
obtained, if the structural wave functions Ψ s are selected from a complete
wave function set, which can represent any possible homogeneous
displacement field in the plate domain Ωs, i.e. any possible displacement
field, which satisfies the homogeneous part of the dynamic plate equation
(4.1).
It is proven below that an expansion of type
w( x, y ) = w1 ( x, y ) + w 2 ( x, y )

(4.39)
with
w1 ( x, y ) =

 s π
2
s π
2 
 − j k b2 −  1  . y j k b2 −  1  . y 
∞   Lx 
+ ws 1 , 2 . e
 Lx 
+  (4.40)
s1πx  w s 1 ,1. e 
∑ cos(
L
)
 2 2 
s1 = 0 x  s π   s π 
 − j − k b2 −  1  . y j − k b2 −  1  . y 
w  Lx   Lx  
 s1 ,3 . e + ws 1 ,4 . e 
and
w 2 ( x, y ) =

 s π
2
s π
2 
 − j k b2 −  2  . x j k b2 −  2  . x 
∞
  Ly   Ly  
s 2πy  w s 2 ,1. e + ws 2 ,2 . e +  (4.41)
∑ cos( L ) 2 2 
s2 = 0 y  
2  s 2π 
 
2  s 2π 

− j − kb −  . x j − k − . x 
  b  L  
 Ly   y 
 ws 2 ,3 . e + ws 2 ,4 . e 
can represent any possible homogeneous displacement field in a rectangular

plate domain with dimensions LxxLy.
Since each function in the latter expansion satisfies the homogeneous part of
the dynamic plate equation and since this equation is a fourth-order partial
differential equation, it is sufficient to prove that the latter expansion can
represent a homogeneous displacement field with arbitrary rotational
displacement and generalised shear force distributions along the rectangular
boundary Γs (see figure 4.3).
Hence, it must be proven that the contributions ws1,1, ws1,2, ws1,3, ws1,4, ws2,1, ws2,2,
ws2,3 and ws2,4 in the proposed expansion (4.39) can be determined, such that
206 Chapter 4
Figure 4.3 : general homogeneous displacement field in a rectangular plate
∂w( 0, y )

= θ n 1 ( y ), ( y ∈ [ 0, Ly ]) , (4.42)
∂x
∂w( Lx , y )

= −θ n 2 ( y ), ( y ∈ [ 0, Ly ]) , (4.43)
∂x
∂ 3 w( 0, y ) Qn1 ( y ) ∂ 2θ n1 ( y )

= − (2 − ν ) , ( y ∈ [ 0, Ly ]) , (4.44)
∂x 3 D ∂y 2
∂ 3 w( Lx , y ) ∂ 2θ n 2 ( y )

Qn 2 ( y )
=− + (2 − ν ) , ( y ∈ [ 0, Ly ]) , (4.45)
∂x 3 D ∂y 2
∂w( x ,0 )

= θ n 3 ( x ), ( x ∈ [ 0, Lx ]) , (4.46)
∂y
∂w( x , Ly )

= −θ n 4 ( x ), ( x ∈ [ 0, L x ]) , (4.47)
∂y
∂ 3 w( x,0 ) Qn 3 ( x ) ∂ 2θ n 3 ( x )

= − (2 − ν ) , ( x ∈ [ 0, L x ]) , (4.48)
∂y 3 D ∂x 2
∂ 3 w( x, Ly ) Q (x) ∂ 2θ n 4 ( x )
= − n4 + (2 − ν ) , ( x ∈ [ 0, Lx ]) . (4.49)
∂y 3 D ∂x 2
By defining (s2=0,1,...)
Ws 2 ,1 =
2 2
 s 2π   s 2π  (4.50)
j kb2 −   ( w s , 2 − w s ,1 ) + j − k b − 

2
 L 
 ( w s , 4 − w s , 3 ),
 Ly   y 
2 2 2 2
Ws 2 , 2 =
2 2
 
s π  s π
2 j k b2 −  2  . Lx −j k b2 −  2  . Lx
s π  Ly   Ly 
j k b2 −  2  ( w s 2 , 2 . e − w s 2 ,1 . e )
 Ly 
2 2
s π s π
2 j − k b2 −  2  . L x − j − k b2 −  2  . L x
s π  Ly   Ly 
+ j − kb2 −  2  ( w s 2 , 4 . e − ws 2 , 3 . e )
 Ly 
(4.51)
Ws 2 , 3 =
3 3
 2  2
 2  s 2π     
2  s 2π  
j  kb −    ( w s ,1 − w s , 2 ) +
 j  − kb −    ( w s 2 , 3 − w s 2 , 4 ),
  Ly     Ly  
2 2
   
(4.52)
Ws 2 , 4 =
2 2
3 s π s π
 2 − j k b2 −  2  . L x j k b2 −  2  . L x
 2  s 2π    Ly   Ly 
j  kb −    ( w s ,1 . e
 − w s , 2 . e )
  Ly  
2 2
 
2 2
3 s π s π
 2 − j − k b2 −  2  . L x j − k b2 −  2  . L x
  s π    y 
L  Ly 
+ j  − kb2 −  2   ( w s 2 , 3 . e − ws 2 ,4 . e )
  Ly  
 
(4.53)
the conditions (4.42)-(4.45) become

208 Chapter 4
∞ s πy
∑ Ws 2 ,1 .cos( 2 ) = θ n 1 ( y ) ,
Ly
(4.54)
s2 = 0
∞ s πy
∑ Ws 2 , 2 .cos( 2 ) = −θ n 2 ( y ) ,
Ly
(4.55)
s2 = 0
∞ s πy Q ( y) ∂ 2θ n1 ( y )
∑ Ws 2 , 3 .cos( 2 ) = n 1 − (2 − ν ) , (4.56)
s2 = 0 Ly D ∂y 2
∞ s πy Q ( y) ∂ 2θ n 2 ( y )
∑ Ws 2 , 4 .cos( 2 ) = − n 2 + (2 − ν ) . (4.57)
s2 = 0 Ly D ∂y 2
As mentioned in section 2.5.2., any (piecewise continuous) function, defined

in y ∈]0, L y [ , can be represented as a series expansion in terms of the
s πy
functions cos( 2 ) (s2=0,1,...). Hence, a set of coefficients Ws2,1, Ws2,2, Ws2,3
Ly
and Ws2,4 (s2=0,1,...) can be found, such that conditions (4.54)-(4.57) are
satisfied. The substitution of these coefficients into (4.50)-(4.53) yields a set
of contributions ws2,1, ws2,2, ws2,3 and ws2,4 such that the conditions (4.42)-(4.45)
are satisfied.
In a completely similar way, a set of contributions ws1,1, ws1,2, ws1,3 and ws1,4
may be found, such that the conditions (4.46)-(4.49) are satisfied.
The above defined complete wave function set for plate problems with
rectangular domains allows the construction of a convergent displacement
expansion for any type of uncoupled plate problems. According to the
procedure, described in section 2.5.3., the convergence of a weighted
residual wave model of type (4.25) is ensured
• by decomposing the plate domain into some convex subdomains or at
least into some subdomains, such that an arbitrary homogeneous
displacement field, defined in each subdomain, may be considered as
part of a homogeneous displacement field, defined in a rectangular
domain, circumscribing the subdomain
• and by approximating the displacement field in each subdomain in terms
of an expansion
n s1
w( x, y ) = ∑ ( ws1 ,1.Ψ s1 ,1 ( x, y) + ws 1 ,2 .Ψ s 1 ,2 ( x, y))
!
s1 = 0
(4.58)
n s2
+ ∑ (ws 2 ,1 .Ψ s 2 ,1 ( x, y) + ws 2 ,2 .Ψ s 2 ,2 ( x, y)) + w F ( x, y)
"
s2 = 0
where
− jk ys1 , 1 . y
Ψ s1 ,1 ( x, y ) = cos( k xs 1, 1 . x ). e ,
− jk ys1 , 2 . y
Ψ s1 , 2 ( x, y ) = cos( k xs1, 2 . x ). e ,
(4.59)
− jk xs2 , 1 . x
Ψ s 2 ,1 ( x , y ) = e .cos( k ys 2, 1. y ),
− jk xs2 , 2 . x
Ψ s 2 , 2 ( x, y ) = e .cos( k ys 2, 2 . y ),
with
2
sπ s π
( k xs 1 ,1 , k ys 1 ,1 ) = ( 1 ,± kb2 −  1  ),
Lx  Lx 
2
sπ s π
( k xs 1 , 2 , k ys 1 , 2 ) = ( 1 ,± j k b2 +  1  ),
Lx  Lx 
(4.60)
2
s π s π
( k xs 2 ,1 , k ys 2 ,1 ) = ( ± kb2 −  2  , 2 ),
 Ly  Ly
2
s π s π
( k xs 2 , 2 , k ys 2 , 2 ) = ( ± j k b2 +  2  , 2 ).
 Ly  L y
and where Lx and Ly are the dimensions of the rectangular domain,

circumscribing the considered subdomain.
wave function truncation
In a similar way as described in section 2.6.3., a rule of thumb may be

formulated for the specification of the integer truncation values ns1 and ns2 in
the displacement expansion (4.58).
210 Chapter 4
The smallest spatial periods in the x- and y-direction, associated with the
wave functions, defined in (4.59), are
2Lx 2 Ly
λ x,min = , λ y,min = . (4.61)
ns 1 ns 2
The spatial variation of the steady-state dynamic plate response is strongly

related to the (undamped) bending wavelength
4π 4 Et 2
λb = 4 . (4.62)
3ρ s ω 2 ( 1 − ν 2 )
The proposed truncation rule is based on the requirement that the smallest
period of the cosine wave functions in the displacement expansion should be
not larger than half the undamped bending wavelength, i.e.
λb
λ x,min ≈ λ y,min ≤ . (4.63)
2
As a result, the integer values ns1 and ns2 should be specified, such that3
ns 1 ns 2 4
≈ ≥ . (4.64)
Lx Ly λb

Recall from section 2.5.4. that the linear independence of all
4(ns1+1)+4(ns2+1) wave functions in the truncated set must be verified in
order to avoid the singularity of the wave model. When the excitation
frequency ω corresponds, for instance, with an undamped natural frequency
of the circumscribing rectangular plate domain with simply supported
boundaries, i.e.
s12 s 22 Et 2
ω = π 2( + ) ( s1 , s 2 = 0,1,...) , (4.65)
L2x L2y 12ρ s (1 − ν 2 )
3
note that this truncation rule assumes that the spatial periods of the prescribed
dynamic quantities, specified at the plate boundaries, are not smaller than λb
it follows from (4.60) that not all wave functions are linearly independent
and that some wave functions must be eliminated to restore the linear
independence without reducing the span of the wave function set.
4.2.5. Some validation examples
In this section, the numerical convergence of the proposed weighted residual

wave model (4.25) is illustrated for two uncoupled plate problems, one with
a convex plate domain, and one, for which the plate domain must be
decomposed into two subdomains.
convex
Figure 4.4. shows a convex plate, of which the boundaries are assumed to be
clamped, i.e. with zero translational and normal rotational displacement
components ( w ≡ 0, θ n ≡ 0 ). The considered case consists of an aluminium
plate (E=70.109 N/m2, ρs=2790 kg/m3, ν=0.3, t=0.002 m) with material loss
factor η=0.1, excited by a normal point force F=1 N, applied at (xF,yF)=
(0.25*Lx, 0.25*Ly) at ω=2π.250 Hz. For this case, a weighted residual wave
model is built with 104 structural wave functions (ns1=16,ns2=8), associated
with the enclosing rectangular domain LxxLy=1m x 0.5m.
Figure 4.4 : convex plate with clamped boundaries
Figure 4.5 shows the resulting contour plots of the real and imaginary parts
of the normal plate displacement. This figure clearly illustrates that the
proposed wave model provides an accurate representation of the clamped
boundary conditions. Since the displacement expansion exactly satisfies the
212 Chapter 4
dynamic plate equation, it may be concluded that the proposed wave model
converges towards the exact solution.
Figure 4.5 : contour plots of the real (upper) and imaginary (lower) parts of the
normal plate displacement at 250 Hz.
As described in section 2.7.2.1., the numerical convergence may also be

illustrated by the convergence curves. Figure 4.6 plots these curves for the
amplitudes of the normal displacement predictions at location
(x,y)=(0.4m,0.3m) at three excitation frequencies, i.e. 250 Hz, 500 Hz and
1000 Hz. The ‘TR’ marks on this figure indicate the prediction errors for
those wave models, for which the structural wave function set is truncated at
the smallest integer values ns1 and ns2, satisfying the truncation rule (4.64).
domain decomposition
Figure 4.7 shows a concave plate, whose boundaries are assumed to be
clamped. A convergent wave model may be obtained by decomposing the
plate domain into two convex subdomains and by expanding the
displacement fields w1 and w2 in these subdomains in terms of the structural

wave functions, associated with the circumscribing rectangular domains
with dimensions Lx1xLy and Lx2xLy, respectively (see figure 4.7).
Figure 4.6 : structural convergence curves

at 250 Hz (∇), 500 Hz (+) and 1000 Hz (o)
Figure 4.7 : concave plate with clamped boundaries
In a similar way as described in section 2.5.4., the coupling of the two

subdomain wave models results from the boundary conditions, specified at
214 Chapter 4
the common interface between both subdomains. Since the dynamic plate
equation is a fourth-order partial differential equation, four boundary
conditions are specified at the common interface, i.e. continuity of the
normal displacement, continuity of the normal in-plane rotational
displacement, equilibrium of the generalised shear forces and equilibrium of
the bending moments.
The considered case consists of an undamped aluminium plate (E=70.109
N/m2, ρs=2790 kg/m3, ν=0.3, t=0.002 m, η=0.1), excited by a normal point
force F=1 N, applied at (x1F,y1F)= (0.5*Lx1, 0.3*Ly) with a circular frequency
ω=2π.300 Hz. For this case, a weighted residual wave model is built with 80
structural wave functions (ns1=10,ns2=8), associated with the rectangular
domain Lx1xLy=0.6m x 0.5m, circumscribing the first subdomain, and with
64 structural wave functions (ns1=6,ns2=8), associated with the rectangular
domain Lx2xLy=0.4m x 0.5m, circumscribing the second subdomain.
Figure 4.8 shows the resulting contour plot of the real part of the normal
plate displacement. Again, the accurate representation of the clamped
boundary conditions as well as the subdomain interface boundary conditions
illustrates that the numerical implementation of the proposed wave model
converges towards the exact solution.
Figure 4.8 : contour plot of the real part of the normal plate displacement at 300 Hz
4.2.6. Comparison with an existing complete function set
This section compares the prediction results of the wave model, proposed in
the previous section for solving the convex plate problem, shown in figure
4.4, with the prediction results, obtained with the wave function expansion,
recently proposed by LANGLEY (1997),
2πs 2πs
ns − jk b cos( )x − jk b sin( )y
ns + 1 ns + 1
w( x, y ) = ∑ ws . e .e
s=0
(4.66)
2πs 2πs
ns − k b cos( )x − k b sin( )y
ns + 1 ns + 1
+ ∑ ws . e + w F ( x, y ).

.e
s=0
In order to compare the numerical convergence, involved with the use of

expansions (4.66) and (4.58), several wave models have been constructed
with an increasing number of degrees of freedom, i.e. an increasing number
of wave functions. Figure 4.9 shows the resulting real and imaginary parts
of the normal plate displacement at (x,y)=(0.4m,0.3m) at ω=2π.1000 Hz.
This figure confirms the numerical convergence, involved with the wave
function set, proposed in this dissertation, and confirms also the conclusion
by LANGLEY (1997), in that, despite the theoretical convergence of the
displacement expansion (4.66), its numerical implementation involves
severe convergence problems.
A verification of the Picard conditions, discussed in section 2.7.1., reveals
also the numerical convergence problems, involved with the wave function
expansion (4.66). Figure 4.10 shows that, in contrast with the new wave
function expansion (4.58), a wave model, using the existing wave function
expansion (4.66), doesn’t satisfy the Picard conditions : a substantial decay
of the singular value curve occurs already at large singular values and the
coefficients β i, associated with the small singular values σi, are significantly
larger than these singular values.
216 Chapter 4
Figure 4.9 : real (upper) and imaginary (lower) part of the normal plate displacement
at (x,y)=(0.4m,0.3m) at 1000 Hz, obtained with the new wave function expansion
(4.58) (solid) and with the existing wave function expansion (4.66) (dashed)
Figure 4.10 : Picard test for a wave model, using the new
(upper) and the existing wave function expansion (lower)
4.3. Uncoupled interior acoustic problems
The steady-state pressure field p at any position r(x,y,z), induced by a time-

harmonic external point source excitation q with circular frequency ω,
218 Chapter 4
located at rq(xq,yq,zq) in the three-dimensional cavity domain V of an

uncoupled interior acoustic system, is governed by the inhomogeneous
Helmholtz equation (see appendix A)
∇ 2 p( r ) + k 2 . p( r ) = − jρ 0ωq.δ ( r , rq ), r ∈V . (4.67)
By specifying one boundary condition at each point on the closed boundary

surface Ωa of the cavity domain V, the pressure field is uniquely defined.
Three major types of boundary conditions occur in uncoupled interior
acoustic problems ( Ωa = Ω p ∪ Ωv ∪ Ω Z ),
p( r ) = p ( r ), r ∈Ω p , (4.68)
j ∂p( r )
= vn ( r ), r ∈Ω v , (4.69)
ρ 0ω ∂n
j ∂p( r ) p( r )
= , r ∈Ω Z , (4.70)
ρ 0 ω ∂n Z (r )

impedance functions, respectively.
4.3.2. Field variable expansion
The steady-state pressure p(x,y,z) in the cavity is approximated as a solution

expansion p(x, y, z) ,
na
p( x, y, z ) ≈ p( x , y, z ) = ∑ pa .Φ a ( x, y, z ) + pq ( x, y, z ) = [Φ a ]{ pa } + pq ( x, y, z ).

a=1
(4.71)
Each function Φ a (x, y, z) in the (1xna) matrix [Φ a ] is an acoustic wave
function, which satisfies the homogeneous part of the Helmholtz equation
(4.67),
− j ( k xa . x + k ya . y + k za . z )
Φ a ( x, y, z ) = e (4.72)
with 2
k xa + k ya
2
+ k za
2
= k2. (4.73)
The contributions of these acoustic wave functions to the solution expansion

are comprised in the (nax1) vector { pa } .

Function pq (x, y, z) is a particular solution function for the external acoustic

source term in the inhomogeneous right-hand side of the Helmholtz
equation (4.67). From all possible mathematical expressions, the physically
meaningful Green’s kernel function, defined in (1.15), is selected as
particular solution function,
− jk ( x − x ) 2 + ( y − y ) 2 + ( z − z ) 2
jρ 0ωq e q q q
p q ( x , y, z ) =

. (4.74)
4π ( x − x q ) 2 + ( y − yq ) 2 + ( z − z q ) 2
From the above definitions, it may be concluded that the proposed pressure
expansion (4.71) satisfies a priori the Helmholtz equation (4.67), no matter
what the values of the acoustic wave function contributions pa are.
4.3.3. Weighted residual wave model
The contributions pa of the acoustic wave functions to the solution

expansion (4.71) are determined in a completely similar way as described in
section 2.4.1.1. A weighted residual formulation of the boundary conditions
(4.68)-(4.70) yields a wave model of type
[ Aaa ]{ pa } = { fa } . (4.75)
The (naxna) matrix [ Aaa ] and the (nax1) vector { fa } are defined in (2.28)
and (2.31), in which the integral terms, defined on boundary surface Ωs, are
omitted, since fluid-structure coupling interfaces are absent in uncoupled
acoustic systems.
220 Chapter 4
4.3.4. Convergence requirements
complete wave function set

The definition of a complete three-dimensional wave function set, ensuring
the convergence of the pressure expansion (4.71), is readily obtained as an
extension of the two-dimensional definition, described in section 2.5.
For the case of a rectangular cavity domain with dimensions LxxLyxLz, it is
sufficient to define a set of acoustic wave functions, which satisfy the
homogeneous part of the Helmholtz equation (4.67) and which can represent
a homogeneous pressure field with an arbitrary normal fluid velocity
distribution v1(x,y), v2(x,y), v3(y,z), v4(y,z), v5(x,z) and v6(x,z), along the
rectangular boundary surface (see figure 4.11).
Figure 4.11 : general homogeneous pressure field in a rectangular prism
This problem may be split into three subproblems, each of them having two
opposite side walls with a non-zero normal fluid velocity distribution, while
the four other side walls are perfectly rigid.
One of these subproblems consists of determining the pressure field
p1(x,y,z), satisfying the homogeneous Helmholtz equation
∇ 2 p1 ( x, y, z ) + k 2 . p1 ( x, y, z ) = 0 (4.76)
and the boundary conditions

∂p1 ( 0, y, z ) ∂ p1 ( L x , y , z )
=0 , = 0, ( y ∈ [ 0, L y ], z ∈ [ 0, Lz ]), (4.77)
∂x ∂x
∂p1 ( x ,0, z ) ∂p1 ( x , L y , z )

=0 , = 0, ( x ∈ [ 0, L x ], z ∈ [ 0, Lz ]), (4.78)
∂y ∂y
∂p1 ( x, y, Lz ) ∂p1 ( x, y,0 )
= − jρ 0ωv1 ( x, y ) , = jρ 0ωv 2 ( x, y ),
∂z ∂z (4.79)
( x ∈ [ 0, L x ], y ∈ [ 0, L y ]).
By separation of variables, the pressure field
p1 ( x, y, z ) = X ( x )Y ( y ) Z ( z ) (4.80)
is governed by three ordinary differential equations
d 2X
2
+ α 2. X = 0 , (4.81)
dx
d 2Y
2
+ β 2.Y = 0 , (4.82)
dy
d 2Z
2
+ ( k 2 − α 2 − β 2 ). Z = 0 , (4.83)
dz
where α and β are arbitrary constants.

Solving equation (4.81), subject to the boundary conditions (4.77), implies
that α=a1π/Lx (a1=0,1,...), so that
a πx
X ( x ) = A. cos 1 , ( a1 = 0,1,... ), (4.84)
Lx
where A is an arbitrary constant.

Similarly, solving equation (4.82), subject to the boundary conditions (4.78),
implies that β =a2π/Ly (a2=0,1,...), so that
a πy
Y ( y ) = B.cos 2 ( a 2 = 0,1,...) , (4.85)
Ly
where B is an arbitrary constant.

The general solution of equation (4.83) is
222 Chapter 4
− j ( k 2 −α 2 − β 2 )z j ( k 2 −α 2 − β 2 )z
Z ( z ) = C. e + D. e , (4.86)
resulting in a pressure field
p 1 ( x , y, z ) =
∞ ∞ − jk za1a2 z jk za1a2 z
∑ ∑ cos(k xa 1 x ).cos(k ya 2 y).( pa 1a 2 ,1 . e + pa 1 a 2 , 2 . e ),
a1 = 0 a 2 = 0
(4.87)
with
2
a1π a 2π 2  a1π 
2
a π
k xa 1 = , k ya 2 = and k za1 a 2 = k −   −  2  . (4.88)
Lx Ly  Lx   Ly 
The arbitrary constants pa1a2,1 and pa1a2,2 are determined by the boundary
conditions (4.79),
∞ ∞
∑ ∑ Pa1 a 2 ,1 .cos( k xa 1 x ).cos( k ya 2 y) = v1 ( x, y) , (4.89)
a1 = 0 a 2 = 0
∞ ∞
∑ ∑ Pa 1a 2 ,2 .cos( k xa 1 x ).cos( k ya 2 y) = v 2 ( x, y) , (4.90)
a1 = 0 a 2 = 0
with
k za 1 a 2 − jk za1a2 L z jk z ,mn L z
Pa1 a 2 ,1 = ( p a 1 a 2 ,1 . e − pa1 a 2 , 2 . e ), (4.91)
ρ 0ω
k za1 a 2
Pa1 a 2 , 2 = ( pa 1 a 2 , 2 − pa 1 a 2 ,1 ) . (4.92)
ρ 0ω
The left-hand sides of (4.89) and (4.90) are double Fourier series (half-range
cosine series). For all (piecewise continuous) functions v1(x,y) and v2(x,y),
two sets of constants Pa1a2,1 and Pa1a2,2 can be found, such that both series
converge to v1(x,y) and v2(x,y) (see e.g. BERG and MCGREGOR (1964)).
In a completely similar way, a convergent pressure expansion can be
defined for the two other subproblems.
As a result, the homogeneous pressure field with an arbitrary normal fluid

velocity distribution along the boundary surface of a rectangular cavity may
be expressed as
∞ ∞ − jk za1a2 z
p( x, y, z ) = ∑ ∑ pa1a 2 .cos(k xa1 x ).cos(k ya 2 y). e
a1 = 0 a2 = 0
∞ ∞ − jk ya1a2 y
+ ∑ ∑ pa 3 a 4 .cos(k xa 3 x ). e .cos( k za 4 z ) (4.93)
a3 = 0 a4 = 0
∞ ∞ − jk xa5 a6 x
+ ∑ ∑ pa 5 a 6 . e .cos( k ya 5 y ).cos( k za 6 z )
a 5 = 0 a6 = 0
with
2
a1π a 2π 2  a1π 
2
a π
( k xa 1 , k ya 2 , k za 1 a 2 ) = ( , ,± k −   −  2  ),
Lx Ly  Lx   Ly 
2 2
a π a π a π a π
( k xa 3 , k ya 3 a 4 , k za 4 ) = ( 3 ,± k 2 −  3  −  4  , 4 ), (4.94)
Lx  Lx   Lz  Lz
2
 a 5π  2
 a 6 π  a 5 π a6 π

( k xa 5 a 6 , k ya 5 , k za 6 ) = ( ± k −  
 −  L  , L , L ).
2
 Ly  z y z
This complete wave function set for homogeneous acoustic problems with
rectangular cavity domains allows the construction of a convergent pressure
expansion for any type of uncoupled interior acoustic problem. According to
the procedure, described in section 2.5.3., the convergence of a weighted
residual wave model of type (4.75) is ensured
• by decomposing the cavity domain into some convex subdomains or at
least into some subdomains, such that an arbitrary homogeneous pressure
field, defined in each subdomain, may be considered as part of a
homogeneous pressure field, defined in a rectangular domain,
circumscribing the subdomain
• and by approximating the pressure field in each subdomain in terms of an
expansion
224 Chapter 4
n a1 n a2 n a3 n a4
p( x, y, z ) = ∑ ∑ pa1 a 2 .Φ a 1a 2 ( x, y, z ) + ∑ ∑ pa 3 a 4 .Φ a 3 a 4 ( x, y, z )

a1 = 0 a 2 = 0 a3 = 0 a4 = 0
n a5 n a6
+ ∑ ∑ pa 5 a 6 .Φ a 5 a 6 ( x, y, z ) + pq ( x, y, z )

a 5 = 0 a6 = 0
(4.95)
where
− jk za1a2 z
Φ a 1a 2 ( x , y, z ) = cos( k xa 1 x ).cos( k ya 2 y ). e ,
− jk ya3a4 y
Φ a 3 a 4 ( x, y, z ) = cos( k xa 3 x ). e .cos( k za 4 z ), (4.96)
− jk xa5a6 x
Φ a 5 a 6 ( x, y, z ) = e .cos( k ya 5 y ).cos( k za 6 z ).
and where Lx, Ly and Lz in the definition formula (4.94) of the

wavenumber components are the dimensions of the rectangular domain,
circumscribing the considered subdomain.

Assuming that the spatial periods of the prescribed acoustic quantities,
specified on the cavity boundary surface, are not smaller than the acoustic
wavelength, a truncation rule may be based on the requirement that the
smallest period of the cosine wave functions in the pressure expansion
(4.95) is not larger than half the acoustic wavelength. As a result, the integer
truncation values na1, na2, na3, na4, na5 and na6 should be specified, such that
na1 na2 na 3 na 4 na 5 na 6 2k
≈ ≈ ≈ ≈ ≈ ≥ . (4.97)
Lx Ly Lx Lz Ly Lz π

In order to avoid the singularity of the wave model, the linear independence
of all 2(na1+1)(na2+1)+2(na3+1)(na4+1)+2(na5+1)(na6+1) wave functions in the
truncated set must be verified. When the excitation frequency ω
corresponds, for instance, with an undamped natural frequency of the
circumscribing rectangular cavity with side walls, assumed to be perfectly
rigid, i.e.
2
n π
2
n π n π
2
ω = c.  1  +  2  +  3  ( n1 , n 2 , n 3 = 0,1,...) , (4.98)
 Lx   Ly   Lz 
it follows from (4.94) that not all wave functions are linearly independent
and that some wave functions must be eliminated to restore the linear
independence without reducing the span of the wave function set.
4.3.5. Validation example
To illustrate the accuracy of the proposed acoustic wave model, the pressure
field in the cavity of a simplified car model is predicted, using the measured
structural velocity distribution of the enclosing car body as boundary
conditions to the car cavity. The wave model prediction results are
compared with finite element predictions and with the measured cavity
pressure field.
Figure 4.12 shows the simplified car model. It consists of a stiff beam
structure (rigid body frame) and a subframe, which are connected through
some rubber mounts. The engine excitation is simulated by exciting a mass
with an electrodynamic shaker. The mass is isolated from the subframe by
means of four rubber mounts. Aluminium panels of 3mm thickness are
sealed to the beam structure with silicone to constitute the car cavity.
Figure 4.12 : simplified car model (exploded view)

226 Chapter 4
The steady-state normal structural velocity of the aluminium car body

panels is obtained from scanning the panels with a laser vibrometer on a
dense grid of measurement points. These measurements are used as normal
fluid velocity boundary conditions in both a wave model and a finite
element model for the cavity pressure field.
The car model cavity is discretized into 9600 linear hexahedral fluid
elements, such that the node locations on the cavity boundary surface
coincide with the measurement grid points (see figure 4.13). The resulting
finite element model has 11067 nodal pressure degrees of freedom and a
maximum bandwidth of 340.
Figure 4.13 : finite element discretization of the cavity
The cavity wave model consists of 370 acoustic wave functions (na1=na3=10,
na2=na5=6,na4=na6=5), associated with the circumscribing rectangular domain
with dimensions 1.5m x 0.975m x 0.8m (see figure 4.12). In order to make
sure that the wave model and the finite element model are based on the same
normal velocity input, the spatial velocity distribution, used for the
calculation of the wave model elements, results from a linear interpolation
of the measured data. Note also that the number of wave functions is larger
than required by the truncation rule (4.97), since the spatial period of the
boundary panel displacements is substantially smaller than the acoustic

wavelength in the considered frequency range.
Figure 4.14 compares the pressure spectrum, measured at some position in
the cavity, with the predicted spectra, obtained with the finite element and
the wave model.
Figure 4.14 : cavity pressure spectra and prediction errors
Especially around the natural frequencies of the cavity (see 214 Hz, 219 Hz
and 243 Hz), where the acoustic response is very sensitive to the velocity
input, the substantially smaller wave model yields a better prediction
accuracy than the finite element model. This illustrates the fact that the
convergence rate of the wave based prediction technique benefits from the
fluid velocity expansions, having the same spatial variation as the pressure
expansions.
4.4. Coupled vibro-acoustic problems

This section describes how a structural and an acoustic wave model, defined
in sections 4.2 and 4.3, are combined into a convergent wave model for the
steady-state dynamic analysis of interior coupled vibro-acoustic systems, in
which part of the boundary surface of the acoustic domain consists of a thin,
flat plate.
228 Chapter 4
When a part of the closed boundary surface Ωa of a bounded acoustic

domain V consists of an elastic structural surface Ωs (Ωs ⊆ Ωa) and when a
time-harmonic external point source excitation q with circular frequency ω
is applied at location rq(xq,yq,zq) in the acoustic domain V, the steady-state
pressure field p at any position r(x,y,z) is uniquely characterised by the
inhomogeneous Helmholtz equation
∇ 2 p( r ) + k 2 . p( r ) = − jρ 0ωq.δ ( r , rq ), r ∈V , (4.99)
together with the specification of one boundary condition at each point on

the boundary surface Ωa = Ω p ∪ Ωv ∪ Ω Z ∪ Ωs ,
p( r ) = p ( r ), r ∈Ω p , (4.100)
j ∂p( r )
= vn ( r ), r ∈ Ω v , (4.101)
ρ 0ω ∂n
j ∂p( r ) p( r )
= , r ∈Ω Z , (4.102)
ρ 0 ω ∂n Z (r )
j ∂p( r )
= jωw( r ), r ∈ Ω s , (4.103)
ρ 0 ω ∂n

impedance functions, respectively, and where w is the steady-state
displacement field of the elastic structural surface Ωs in the direction,
normal to this surface.
When the elastic structural surface consists of a thin, flat plate, an
orthogonal co-ordinate system (x’,y’) may be defined in the plane of the
plate middle surface. In this way, the geometry of the plate middle surface
may be described in the global Cartesian co-ordinate system (x,y,z) by using
a bilinear parametrization of the form
 x = u x ( x’, y’) = u x,0 + u x , x’. x’+u x , y’. y’,


∀ r( x, y, z ) ∈ Ω s :  y = u y ( x’, y’) = u y,0 + u y, x’. x’+u y, y’. y’, (4.104)

î z = u z ( x’, y’) = u z ,0 + uz , x’. x’+u z , y’. y’,
where ui,0, ui,x’ and ui,y’ (i=x,y,z) are constant values.

When a time-harmonic external point force excitation F with circular
frequency ω is applied at plate location rF(x’F,y’F) in the direction, normal to
the plate middle surface Ωs, the steady-state normal displacement field w of
the plate middle surface is governed by the dynamic plate equation
∇ 4 w( x’, y’) − kb4 . w( x’, y’) =

F.δ ( r ( x’, y’), rF ( x’F , y’F )) p(u x ( x’, y’), u y ( x’, y’), uz ( x’, y’))
+ .
D D
(4.105)
Note that the positive sign of the pressure loading term in the right-hand
side of this equation indicates that the normal plate displacement is assumed
to have a positive orientation away from the acoustic domain.
For a unique characterisation of the normal displacement field, two
boundary conditions should be specified at each point on the plate boundary
Γs. In a completely similar way as in uncoupled plate problems, the three
major types of plate boundary conditions, defined in equations (4.7)-(4.12),
may occur in coupled vibro-acoustic problems.
Since the steady-state pressure field in a coupled vibro-acoustic system is

governed by the same Helmholtz equation as in uncoupled acoustic systems
(see (4.67) and (4.99)), the same pressure expansion is used, as proposed in
section 4.3. for uncoupled interior acoustic problems. Provided that the
bounded acoustic domain is convex or that its geometry is at least such that
any homogeneous pressure field in the acoustic domain may be considered
as part of a homogeneous pressure field, defined in a rectangular domain,
circumscribing the acoustic domain4, the steady-state pressure field p(x,y,z)
in the acoustic domain is approximated as a solution expansion p(x, y, z) ,
4
if the domain geometry doesn’t satisfy this condition, the acoustic domain must
be decomposed into some subdomains, whose geometries do satisfy this
condition; in a completely similar way, as described in section 2.5.4. for two-
dimensional coupled vibro-acoustic problems, the acoustic part of the coupled
vibro-acoustic wave model results then from a weighted residual or least-squares
formulation of the acoustic boundary conditions (4.100)-(4.103), together with
the interface conditions, specified along the common interfaces of the various
acoustic subdomains
230 Chapter 4
na
p( x, y, z ) ≈ p( x , y, z ) = ∑ pa .Φ a ( x, y, z ) + pq ( x, y, z ) = [Φ a ]{ pa } + pq ( x, y, z ),

a=1
(4.106)
where
na n a1 n a2
− jk za1a2 z
∑ p a . Φ a ( x , y, z ) = ∑ ∑ pa1 a 2 .cos( k xa1 x ) cos(k ya 2 y)e
a=1 a1 = 0 a 2 = 0
n a3 n a4
− jk ya3a4 y
+ ∑ ∑ pa 3 a 4 .cos(k xa 3 x )e cos( k za 4 z ) (4.107)
a3 = 0 a4 = 0
n a5 n a6
− jk xa5 a6 x
+ ∑ ∑ pa 5 a 6 . e cos( k ya 5 y )cos( k za6 z )
a 5 = 0 a6 = 0

and where the function pq (x, y, z) , defined in (4.74), serves as particular

solution function for the external acoustic source term in the inhomogeneous
right-hand side of the Helmholtz equation (4.99). The definition formula for
the wavenumber components (k xa 1 ,k ya 2 , k za 1 a 2 ) , (k xa 3 ,k ya 3 a 4 , k za 4 ) and
(k xa 5 a 6 ,k ya 5 ,k za 6 ) of the acoustic wave functions in (4.107) are given in
(4.94), in which Lx, Ly and Lz are the dimensions of the (smallest) rectangular
domain, circumscribing the considered acoustic domain V.
Compared with the dynamic equation (4.1) for uncoupled plate problems,
the dynamic plate equation (4.105) for coupled vibro-acoustic problems
contains an additional excitation term for the pressure loading effect of the
fluid on the plate. As a result, the expansion for the normal plate
displacement, as proposed in section 4.2 for uncoupled plate problems, must
be extended with some particular solution functions for the fluid pressure
loading. Therefore, the steady-state normal plate displacement w(x’,y’) is

approximated as a solution expansion w(x’, y’) ,
w( x’, y’) ≈
ns na
w( x’, y’) = ∑ ws .Ψ s ( x’, y’) + ∑ pa . wa ( x’, y’) + w F ( x’, y’) + wq ( x’, y’)

s=1 a=1
= [Ψ s ]{w s } + [ w a ]{ pa } + w F ( x’, y’) + w q ( x’, y’).

(4.108)
Provided that the plate domain Ωs is convex or that its geometry is at least
such that any homogeneous displacement field in the plate domain may be
considered as part of a homogeneous displacement field, defined in a
rectangular domain, circumscribing the plate domain5, the structural wave
function expansion in (4.108) is
ns ns1
− jk − jk
cos( k x’s1 x’).  w s1 ,1 . e y’s1 ,1 + w s1 , 2 . e y’s1 , 2 
y’ y’
∑ w s .Ψ s ( x’, y’) = ∑
s= 1 s1 = 0
ns2
− jk − jk
cos( k y’s2 y’).  w s2 ,1 . e x ’s2 ,1 + w s2 , 2 . e x ’s2 , 2  ,
x’ x’
+ ∑
s2 = 0
(4.109)
where the wavenumber components of the structural wave functions are
defined as
2
sπ s π
( k x’s 1 , k y’s 1 ,1 ) = ( 1 ,± kb2 −  1  ),
Lx’  Lx’ 
2
sπ s π
( k x’s 1 , k y’s 1 , 2 ) = ( 1 ,± j kb2 +  1  ),
Lx’  Lx’ 
(4.110)
2
s π s π
( k x’s 2 ,1 , k y’s 2 ) = ( ± kb2 −  2  , 2 ),
 L y’  L y’
2
s π s π
( k x’s 2 , 2 , k y’s 2 ) = ( ± j kb2 +  2  , 2 ),
 L y’  L y’
in which Lx’ and Ly’ are the dimensions of the (smallest) rectangular plate
domain, circumscribing the considered plate domain Ωs.
5
if the domain geometry doesn’t satisfy this condition, the plate domain must be
decomposed into some subdomains, whose geometries do satisfy this condition;
in a completely similar way, as illustrated in section 4.2.5. for uncoupled
problems, the structural part of the coupled vibro-acoustic wave model results
then from a weighted residual or least-squares formulation of the plate boundary
conditions (4.7)-(4.12), together with the interface conditions, specified along the
common interfaces of the various plate subdomains
232 Chapter 4
In a similar way as described in section 4.2.2. for uncoupled plate problems,

the displacement of an infinite plate, excited by a normal point force, serves

as particular solution function w F (x’, y’) for the external point force term in
the inhomogeneous right-hand side of the dynamic plate equation (4.105),
w F ( x’, y’) = −
jF
[H ( 2)
0 ( kb r ) − H 0( 2) ( − jk b r ) ]

(4.111)
8 k b2 D
with r = ( x’− x’F ) 2 + ( y’− y’F ) 2 (4.112)
Each function wa (x’, y’) in the (1xna) matrix [ w a ] is a particular solution

function for the part of the pressure loading term in the inhomogeneous
right-hand side of the dynamic plate equation (4.105), which results from
one of the acoustic wave functions Φ a in the pressure expansion (4.106).
The normal displacement of an infinite plate, excited by this pressure
distribution, may be used as particular solution function,
w a ( x’, y’) =

L x ’  f s 2 (ξ ) 
j  Φ (u (ξ ,ζ ), u (ξ , ζ ), u (ξ ,ζ )). H ( 2) ( k ρ )dζ  dξ
− ∫ ∫ (4.113)
8 kb2 D 0  f (ξ ) 
a x y z 0 b
s1 
L x ’  f s 2 (ξ ) 
j  Φ (u (ξ , ζ ), u (ξ ,ζ ), u (ξ ,ζ )). H ( 2) ( − jk ρ )dζ  dξ
+ ∫ ∫ a x
8 kb2 D 0  f (ξ )
y z 0 b 
s1 
with ρ = ( x’−ξ ) 2 + ( y’−ζ ) 2 (4.114)
and where the functions fs1 and fs2 describe the geometry of the plate
boundary Γs (see figure 4.15).
In a similar way, function w q (x’, y’) in the displacement expansion (4.108),

which is a particular solution function for the part of the pressure loading
term in the inhomogeneous right-hand side of the dynamic plate equation
(4.105), which results from the external acoustic excitation function

pq (x, y, z) in the pressure expansion (4.106), may be expressed as

Figure 4.15 : plate boundary description
wq ( x’, y’) =

L x ’  f s 2 (ξ ) 
j  ξ ζ ξ ζ ξ ζ ρ ζ  dξ
∫ ∫
( 2)
−

p ( u ( , ), u ( , ), u ( , )). H ( k ) d (4.115)
2 
8 k b D 0  f (ξ )
q x y z 0 b 
s1 
L x ’  f s 2 (ξ ) 
j  ξ ζ ξ ζ ξ ζ ρ ζ  dξ .
∫ ∫
( 2)
+ −

p ( u ( , ), u ( , ), u ( , )). H ( jk ) d
8 k b2 D 0  f (ξ ) 
q x y z 0 b
s1 
Due to the flat shape of the considered plate, alternative expressions for the

particular solution functions wa (x’, y’) may be defined, which have no

direct physical meaning, but which are preferred to expressions of type
(4.113), since they don’t involve the surface integral evaluations. Based on
the expression (4.107) for the acoustic wave function expansion, these
alternative particular solution functions may be written in the form
na n a1 n a2
[ wa ]{ pa } = ∑ pa . wa ( x’, y’) = ∑ ∑ pa1 a 2 . wa1 a 2 ( x’, y’) +
!
a=1 a1 = 0 a 2 = 0
(4.116)
n a3 n a4 n a5 n a6
∑ ∑ pa 3 a 4 . wa 3 a 4 ( x’, y’) + ∑ ∑ pa 5 a 6 . wa 5 a6 ( x’, y’).
" #
a3 = 0 a4 = 0 a 5 = 0 a6 = 0
The particular solution functions w a1 a 2 (x’, y’) in (4.116) are defined as

234 Chapter 4
w a1 a 2 ( x’, y’) =
%
 B1, a 1 a 2 .cos( k xa 1 . u x ( x’, y’)).cos( k ya 2 . u y ( x’, y’)) 

 
+ B2, a1 a 2 .cos( k xa1 . u x ( x’, y’)).sin( k ya 2 . u y ( x’, y’)) − jk za1a2 u z ( x’, y’)
 . e
+ B3, a1 a 2 .sin( k xa 1 . u x ( x’, y’)).cos( k ya 2 . u y ( x’, y’))
+ B 
 4, a 1 a 2 .sin( k xa 1 . u x ( x’, y’)).sin( k ya 2 . u y ( x’, y’)) 
(4.117)
where
1 1 1 1 1
B1, a1 a 2 = ( + + + ),
4 D A1, a1 a 2 A2, a 1 a 2 A3, a1 a 2 A4, a 1 a 2
j 1 1 1 1
B2, a 1 a 2 = (− − + + ),
4 D A1, a1 a 2 A2, a 1 a 2 A3, a1 a 2 A4, a1 a 2
(4.118)
j 1 1 1 1
B3, a 1 a 2 = (− + − + ),
4 D A1, a1 a 2 A2, a 1 a 2 A3, a1 a 2 A4, a 1 a 2
1 1 1 1 1
B4, a 1 a 2 = (− + + − ),
4 D A1, a 1 a 2 A2, a 1 a 2 A3, a1 a 2 A4, a1 a 2
with
 2 2 
 ( k xa 1 u x, x’ + k ya 2 u y, x’ + k za1 a 2 u z, x’ )  4
A1, a 1 a 2 =  − kb  ,
+ ( k u + + 2
  xa1 x , y’ k u
ya 2 y, y’ k u )
za 1 a 2 z , y’  
 
 2 2 
 ( − k xa1 u x , x’ + k ya 2 u y, x’ + k za 1 a 2 u z , x’ )  4
A2, a 1 a 2 =  − kb  ,
+( − k u + + 2
 xa 1 x , y’ k u
ya 2 y, y’ k u
za 1 a 2 z, y’  ) 
 
 2 2 
 ( k xa 1 u x , x’ − k ya 2 u y, x’ + k za1 a 2 u z, x’ )  4
A3, a 1 a 2 =  − kb  ,
+(k u − + 2
  xa 1 x, y’ k u
ya 2 y, y’ k u )
za 1 a 2 z , y’  
 
 2 2 
 ( − k xa1 u x , x’ − k ya 2 u y, x’ + k za 1 a 2 u z , x’ )  4
A4, a1 a 2 =  − kb . (4.119)
+( − k u − + 2
 xa 1 x , y’ k u
ya 2 y, y’ k u
za 1 a 2 z, y’  ) 
 
&
The particular solution functions w a 3 a 4 (x’, y’) in (4.116) are defined as

w a 3 a 4 ( x’, y’) =
'
 B1, a 3 a 4 .cos( k xa 3 . u x ( x’, y’)).cos( k za 4 . uz ( x’, y’)) 

 
+ B2, a 3 a 4 .cos( k xa 3 . u x ( x’, y’)).sin( k za 4 . u z ( x’, y’)) − jk ya3a4 u y ( x’, y’)
 . e
+ B3, a 3 a 4 .sin( k xa 3 . u x ( x’, y’)).cos( k za 4 . u z ( x’, y’))
+ B 
 4, a 3 a 4 .sin( k xa 3 . u x ( x’, y’)).sin( k za 4 . u z ( x’, y’)) 
(4.120)
where
1 1 1 1 1
B1, a 3 a 4 = ( + + + ),
4 D A1, a 3 a 4 A2, a 3 a 4 A3, a 3 a 4 A4, a 3 a 4
j 1 1 1 1
B2, a 3 a 4 = (− − + + ),
4 D A1, a 3 a 4 A2, a 3 a 4 A3, a 3 a 4 A4, a 3 a 4
(4.121)
j 1 1 1 1
B3, a 3 a 4 = (− + − + ),
4 D A1, a 3 a 4 A2, a 3 a 4 A3, a 3 a 4 A4, a 3 a 4
1 1 1 1 1
B4, a 3 a 4 = (− + + − ),
4 D A1, a 3 a 4 A2, a 3 a 4 A3, a 3 a 4 A4, a 3 a 4
with
 2 2 
 ( k xa 3 u x , x’ + k ya 3 a 4 u y, x’ + k za 4 u z, x’ )  4
A1, a 3 a 4 =  − kb  ,
 +( k u + + 2
  xa 3 x , y’ k u
ya 3 a 4 y, y’ k u )
za 4 z , y’  
 
 2 2 
 ( − k xa 3 u x, x’ + k ya 3 a 4 u y, x’ + k za 4 u z , x’ )  4
A2, a 3 a 4 =  − kb  ,
+( − k u + + 2
 xa 3 x , y’ k u
ya 3 a 4 y, y’ k u
za 4 z, y’  ) 
 
 2 2 
 ( k xa 3 u x , x’ + k ya 3 a 4 u y, x’ − k za 4 u z, x’ )  4
A3, a 3 a 4 =  − kb  ,
+(k u + − 2
  xa 3 x , y’ k u
ya 3 a 4 y, y’ k u )
za 4 z , y’  
 
 2 2 
 ( − k xa 3 u x , x’ + k ya 3 a 4 u y, x’ − k za 4 u z, x’ )  4
A4, a 3 a 4 =  − k b . (4.122)
+( − k u + − 2
 xa 3 x , y’ k u
ya 3 a 4 y, y’ k u )
za 4 z , y’  
 
(
The particular solution functions w a 5 a 6 (x’, y’) in (4.116) are defined as

236 Chapter 4
w a 5 a 6 ( x’, y’) =
)
 B1, a 5 a 6 .cos( k ya 5 . u y ( x’, y’)).cos( k za 6 . u z ( x’, y’)) 

 
+ B2, a 5 a 6 .cos( k ya 5 . u y ( x’, y’)).sin( k za 6 . u z ( x’, y’)) − jk xa5 a6 u x ( x’, y’)
 . e
+ B3, a 5 a 6 .sin( k ya 5 . u y ( x’, y’)).cos( k za 6 . u z ( x’, y’))
+ B 
 4, a 5 a 6 .sin( k ya 5 . u y ( x’, y’)).sin( k za6 . u z ( x’, y’)) 
(4.123)
where
1 1 1 1 1
B1, a 5 a 6 = ( + + + ),
4 D A1, a 5 a 6 A2, a 5 a 6 A3, a 5 a 6 A4, a 5 a 6
j 1 1 1 1
B2, a 5 a 6 = (− − + + ),
4 D A1, a 5 a 6 A2, a 5 a 6 A3, a 5 a 6 A4, a 5 a 6
(4.124)
j 1 1 1 1
B3, a 5 a 6 = (− + − + ),
4 D A1, a 5 a 6 A2, a 5 a 6 A3, a 5 a 6 A4, a 5 a 6
1 1 1 1 1
B4, a 5 a 6 = (− + + − ),
4 D A1, a 5 a 6 A2, a 5 a 6 A3, a 5 a 6 A4, a 5 a 6
with
 2 2 
 ( k xa 5 a 6 u x , x’ + k ya 5 u y, x’ + k za 6 u z , x’ )  4
A1, a 5 a 6 =  − kb  ,
+(k + + 2
  xa 5 a 6 x , y’
u k u
ya 5 y, y’ k u
za 6 z, y’  ) 
 
 2 2 
 ( k xa 5 a 6 u x , x’ − k ya 5 u y, x’ + k za 6 u z , x’ )  4
A2, a 5 a 6 =  − kb  ,
+ ( k − + 2
  xa 5 a 6 x , y’
u k u
ya 5 y, y’ k u
za 6 z, y’  ) 
 
 2 2 
 ( k xa 5 a 6 u x , x’ + k ya 5 u y, x’ − k za 6 u z , x’ )  4
A3, a 5 a 6 =  − kb  ,
+ ( k + − 2
  xa 5 a 6 x , y’
u k u
ya 5 y, y’ k u
za 6 z, y’  ) 
 
 2 2 
 ( k xa 5 a 6 u x , x’ − k ya 5 u y, x’ − k za 6 u z , x’ )  4
A4, a 5 a 6 =  − kb . (4.125)
+ ( k − − 2
  xa 5 a 6 x , y’
u k u
ya 5 y, y’ k u
za 6 z, y’  ) 
 
From the above definitions, it may be concluded that the proposed acoustic
and structural expansions (4.106) and (4.108) exactly satisfy, respectively,
the Helmholtz equation (4.99) and the dynamic plate equation (4.105), no
matter what the values of the wave function contributions pa and ws are.
4.4.3. Coupled vibro-acoustic wave model
The contributions pa (a=1..na) and ws (s=1..ns) of the acoustic and structural

wave functions to the solution expansions (4.106) and (4.108) are
determined by the acoustic boundary conditions, defined in (4.100)-(4.103),
and by the structural boundary conditions, defined in (4.7)-(4.12). These
boundary conditions may be transformed into a weighted residual or a least-
squares formulation. This section describes only the construction of a
weighted residual model for the determination of the wave function
contributions. The construction of a least-squares wave model follows a
completely similar procedure, as outlined in section 2.4.1.2. However, as it
is the case for two-dimensional coupled vibro-acoustic problems, illustrated
in section 2.7.2.3., a least-squares wave model for three-dimensional
coupled vibro-acoustic problems has a smaller convergence rate than a
corresponding weighted residual wave model.
weighted residual formulation of the structural boundary conditions

Since the structural boundary conditions in the considered coupled vibro-
acoustic problems are completely similar to the boundary conditions in
uncoupled plate problems, the residual error functions, involved with the
representation of the structural boundary conditions, are obtained from the
substitution of the displacement expansion (4.108) into the definition
formula (4.18)-(4.23). As proposed in section 4.2.3. for uncoupled plate
problems, the use of these residual error functions in the weighted residual
formulation (4.24) yields a set of ns algebraic equations in the ns unknown
structural wave contributions ws and the na unknown acoustic wave
contributions pa,
w s 
[ Ass Csa ]  = { fs } . (4.126)
î pa 
The elements of the symmetric (nsxns) matrix [ Ass ] are defined in (4.26).
The element on row si and column aj of the (nsxna) matrix [ C sa ] is
238
Chapter 4
[Csa ]si a j =

∫ LQ [Ψ si ]. wa j . dΓ + ∫ Lm [Ψ si ]. Lθ [ wa j ]. dΓ

− ∫ Lθ [Ψ si ]. Lm [ wa j ]. dΓ − ∫ Ψ si . LQ [ wa j ]. dΓ (4.127)
Γ wm + Γ mQ Γ mQ
nw nF
+ ∑ LF [Ψ si ]. wa j (rsc ) − ∑ Ψ si ( rsc ). LF [ wa j ]
c=1 c=1
and the element on row si of the (nsx1) vector { fs } is
{ f s } si =

∫ LQ [Ψ si ].( w − w F − wq ). dΓ + ∫ Lm [Ψ si ].(θ n − Lθ [ w F + wq ]). dΓ


− ∫ Lθ [Ψ si ].( mn - Lm [ w F + wq ]). dΓ − ∫ Ψ si .(Qn − LQ [ w F + wq ]). dΓ
Γ wm + Γ mQ Γ mQ

nw
+ ∑ LF [Ψ si ].( w ( rsc ) − w F ( rsc ) − w q ( rsc ))
c=1
nF
− ∑ LF [Ψ s i ].( mns ( rsc
+
) − mns ( rsc
-
) − LF [ w F + w q ]).
c=1
(4.128)
weighted residual formulation of the acoustic boundary conditions

Since the acoustic boundary conditions in two- and three-dimensional
coupled vibro-acoustic problems are completely similar, the residual error
functions, involved with the representation of the acoustic boundary
conditions (4.100)-(4.103), are obtained from the substitutions of the
pressure expansion (4.106) and the displacement expansion (4.108) into the
definition formula (2.21)-(2.24). As proposed in section 2.4.1.1. for two-
dimensional coupled vibro-acoustic problems, the use of these residual error
functions in the weighted residual formulation (2.25) yields a set of na
algebraic equations in the ns unknown structural wave contributions ws and
the na unknown acoustic wave contributions pa,
w s 
[Cas Aaa + Caa ]  = { fa } .
pa 
(4.129)
The (naxna) matrices [ Aaa ] and [ Caa ] , the (naxns) matrix [ Cas ] and the
(nax1) vector { fa } are defined in, respectively, (2.28), (2.29), (2.30) and
(2.31).
coupled vibro-acoustic wave model

The coupled vibro-acoustic wave model is obtained from the combination of
the algebraic equations (4.126) and (4.129) into a square set of ns+na
algebraic equations in the ns unknown structural wave contributions ws and
the na unknown acoustic wave contributions pa,
 Ass Csa  w s   f s 
C   =  .
Aaa + Caa  pa 
(4.130)
 as fa 
Recall from section 2.4.2. that this coupled vibro-acoustic wave model is
non-symmetric, since [Csa ]T ≠ [Cas ] and [Caa ]T ≠ [Caa ] . Only the
acoustic and structural submatrices [ Aaa ] and [ Ass ] are symmetric.

A rule of thumb for the specification of the integer truncation values ns1 and
ns2 in the structural wave function expansion (4.109) and the integer
truncation values na1, na2, na3, na4, na5 and na6 in the acoustic wave function
expansion (4.107) can be readily obtained as a three-dimensional extension
of the truncation rule for two-dimensional coupled vibro-acoustic problems,
defined in section 2.6.3., together with the truncation rule for uncoupled
plate problems, defined in section 4.2.4.
Based on the requirement that the smallest spatial periods of the acoustic
and structural wave functions in the pressure and displacement expansions
(4.106) and (4.108) should be not larger than half the (undamped) structural
bending wavelength λb, defined in (4.62), the integer truncation values
should be specified, such that
na1 na 2 na 3 na 4 na 5 na 6 ns 1 ns 2 4
≈ ≈ ≈ ≈ ≈ ≈ ≈ ≥ . (4.131)
Lx Ly Lx Lz Ly Lz L x’ L y’ λb
240 Chapter 4
Recall also from sections 4.2.4. and 4.3.4. that the linear independence of all
ns=4(ns1+1)(ns2+1) structural wave functions Ψs and the linear independence
of all na=2(na1+1)(na2+1)+2(na3+1)(na4+1)+2(na5+1)(na6+1) acoustic wave
functions Φa must be verified a priori, in order to avoid the singularity of the
wave model (4.130).
4.4.4. Performance comparison with coupled FE/FE models
validation examples
In order to illustrate the accuracy of the proposed wave model (4.130) for
three-dimensional coupled vibro-acoustic problems and to compare its
convergence rate with the convergence rate of an Eulerian coupled FE/FE
model (1.22), two validation examples are considered, one with a convex
and one with a concave cavity domain.
Figure 4.16 shows the first validation example. One boundary surface of an
acoustic cavity consists of a flat rectangular (Lx’=0.75 m, Ly’= 0.5 m) plate
with clamped boundaries, while all other cavity boundary surfaces are
perfectly rigid. The air in the cavity has an ambient fluid density ρ0=1.225
kg/m3 and a speed of sound c=340 m/s. The undamped (η=0) aluminium
plate has a thickness t=2.10-3 m, a density ρs=2790 kg/m3, a Poisson
coefficient ν=0.3 and an elasticity modulus E=70.109 N/m2. The acoustic
cavity is comprised in the volume of an enclosing rectangular prism with
dimensions Lx=1.5 m, Ly=0.5 m, Lz=1 m. The coupled vibro-acoustic system
is excited by a time-harmonic mechanical point force F, applied at location
(x’F,y’F)=(2Lx’/3,Ly’/3) on the plate, in the direction normal to the plate.
Figure 4.16 : interior coupled vibro-acoustic system (example 1)

Figure 4.17 shows the second validation example. One boundary surface of
a funnel-type acoustic cavity consists of a flat rectangular (Lx’=1 m, Ly’= 0.5
m) plate with clamped boundaries, while all other cavity boundary surfaces
are perfectly rigid. The fluid properties of the air in the cavity and the
material properties of the plate are the same as in the first validation
example. The dimensions, indicated in figure 4.17, are Lx1=1 m, Lx2=0.6 m,
Ly=0.5 m, Lz1=0.5 m and Lz2=0.3 m. The coupled vibro-acoustic system is
excited by a time-harmonic mechanical point force F, applied at location
(x1,F,y1,F)=(Lx1/4,Ly/4) on the plate, in the direction normal to the plate.
Figure 4.17 : interior coupled vibro-acoustic system (example 2)
coupled vibro-acoustic wave models

For the first validation example, the steady-state cavity pressure p(x,y,z) is
approximated as an expansion of type (4.106), using the acoustic wave
functions Φa, which are associated with the circumscribing rectangular
domain LxxLyxLz (see figure 4.16). The steady-state normal plate
displacement w(x’,y’) is approximated as an expansion of type (4.108),
using the structural wave functions Ψs, which are associated with the
considered rectangular plate domain Lx’xLy’ (=(Lx-0.75*Lz)xLy).
242 Chapter 4
For the second validation example, the acoustic cavity is decomposed into
two subcavities. The expansion for the steady-state pressure p1(x1,y1,z1) in the
first subcavity utilises the acoustic wave functions, which are associated
with the circumscribing rectangular domain Lx1xLyxLz1. The expansion for
the steady-state pressure p2(x2,y2,z2) in the second subcavity utilises the
acoustic wave functions, which are associated with this rectangular
subcavity Lx2xLyxLz2 (see figure 4.17). The expansion for the steady-state
normal plate displacement w(x1,y1) utilises the structural wave functions,
which are associated with the considered rectangular plate domain Lx1xLy.
To illustrate the accuracy of the proposed wave based prediction technique,
the prediction results for the field variable distributions in the second
validation example are shown for the case of a unit normal point force
excitation at 250 Hz.
Figure 4.18 : instantaneous plate displacement (a) and fluid displacement (b), normal
to the plate-cavity interface at 250 Hz
Figure 4.18 shows the instantaneous normal plate displacement and normal
fluid displacement along the plate-cavity interface, obtained from a wave
model, which consists of 72 structural and 444 acoustic wave functions.
This figure illustrates that the clamped plate boundary conditions and the
displacement continuity conditions along the plate-cavity interface are
accurately represented.

in the plane y1=0.25*Ly

in the plane x1=0.25*Lx1
244 Chapter 4
Figures 4.19 and 4.20 show the contour plots of the corresponding
instantaneous cavity pressure in the planes y1=0.25*Ly and x1=0.25*Lx1. Since
the pressure contour lines are perpendicular to the rigid boundary surfaces
of the cavity, the zero normal fluid velocity conditions at these boundary
surfaces are accurately represented. The continuity of the pressure contour
lines at the common interface (z1=Lz1, z2=0) between both subcavities
indicates that the continuity conditions of the pressure and normal fluid
velocity are accurately represented. Since the above figures illustrate that all
structural and acoustic boundary conditions are accurately represented and
since the proposed field variable expansions satisfy a priori the dynamic
equations, it may be concluded that a highly accurate approximation of the
exact coupled vibro-acoustic response is obtained.
comparison with coupled FE/FE models

In order to compare the proposed wave based prediction technique with the
the finite element method in terms of accuracy and associated computational
efforts for three-dimensional coupled vibro-acoustic analysis, the two
considered validation examples are solved with a wave model of type
(4.130) and with several Eulerian coupled FE/FE models of type (1.22).
For both validation examples, the coupled FE/FE models result from a
discretization of the plate into 4-noded quadrilateral shell elements
(CQUAD4) and a discretization of the cavity into 8-noded hexahedral fluid
elements (CHEXA). Tables 4.1 and 4.2 list, for both validation examples,
the number of wave functions in the coupled vibro-acoustic wave models, as
well as the number of elements and the corresponding number of
unconstrained degrees of freedom6 (dof) in each FE/FE model. The tables
indicate also the CPU times, needed for the direct response calculations at
one frequency on a HP-C180 workstation (SPECfp95=18.7, SPECint95
=11.8). The indicated CPU times for the wave models include both the times
for constructing the models as well as for solving the resulting matrix
equations. Since the coupled FE/FE models consist of frequency-
independent submatrices, these submatrices must only be calculated once
and the construction of the models at each frequency requires only a simple
submatrix assembly. As a result, the computational effort for constructing
the models is negligible, compared with the efforts for solving the (large)
6
each node in the plate discretizations has 3 degrees of freedom, i.e. the
translational displacement component, normal to the plate, and the rotational
displacement components around the two orthogonal in-plane co-ordinate axes;
each node in the cavity discretizations has 1 pressure degree of freedom
models, at least when the dynamic response is calculated in a large

frequency range. Therefore, the indicated CPU times for the finite element
models comprise only the direct solution times. Note also that the wave
models are implemented in a MATLAB environment, while the finite
element results are obtained with the MSC/NASTRAN software package.
FE/FE 1 FE/FE 2 FE/FE 3 wave

plate 54 120 216 561 486 1326 56
cavity 1296 1729 10368 12025 34992 38665 258
total dof 1849 12586 39991 314
CPU time 1s 35 s 460 s 80 s
Table 4.1 : properties of the various prediction models for validation example 1
FE/FE 1 FE/FE 2 FE/FE 3 wave

plate 128 315 240 627 960 2691 56
cavity 1920 2421 6528 7579 30780 33825 224
total dof 2736 8206 36516 280
CPU time 1.2 s 8.2 s 130 s 35 s
Table 4.2 : properties of the various prediction models for validation example 2
functions (with a frequency resolution of 1 Hz), obtained with the wave
model and the FE/FE models for the first validation example. These figures
clearly illustrate that, by increasing the number of elements in the plate and
cavity discretizations (see figures 4.21(a)→(c), 4.22(a)→(c)), the finite
element results gradually converge towards the results, obtained with the
small wave model. These figures illustrate also the typical approximation
errors, involved with finite element discretizations. As already indicated in
chapter 1 and illustrated in chapter 3, the low-order (polynomial) shape
functions in each (parent) element can only represent a dynamic response
with a limited spatial variation. As a result, the dynamic stiffness is usually
overestimated, yielding an overestimation of the resonance frequencies.
Since the structural and acoustic wavelengths decrease with increasing
frequency, the spatial variation of the dynamic response increases also with
frequency, so that, for a given discretization, the overestimation of the
246 Chapter 4
resonance frequencies gets worse for increasing frequency (see figures 4.21
and 4.22). Note also that this frequency overestimation is less pronounced
for the resonances at 121 Hz and 179 Hz than for the other resonances. This
is due to the fact that most of the energy at these two resonances is stored in
the cavity, while at the other resonances most of the energy is stored in the
plate. Since the acoustic wavelengths are substantially larger than the
Figure 4.21 : structural frequency response functions (normal plate displacement

over input force) at the excitation point (x’F,y’F)=(2Lx’/3,Ly’/3) of the first validation
example (solid : wave / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
structural bending wavelengths, at least in the considered frequency range,

the spatial variation of the dynamic response around cavity-controlled
resonance frequencies is substantially smaller than the spatial variation
around plate-controlled resonance frequencies. As a result, the discretization
at cavity location (x,y,z)=(2Lx/3,2Ly’/3,Lz/4) of the first validation example (solid :
wave / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
248 Chapter 4
errors and associated frequency overestimations are smaller for cavity-

controlled than for plate-controlled resonances.
functions (with a frequency resolution of 1 Hz), obtained with the wave
model and the FE/FE models for the second validation example. These
Figure 4.23 : structural frequency response functions (normal plate displacement

over input force) at the excitation point (x1,F,y1,F)=(Lx1/4,Ly/4) of the second validation
example (solid : wave / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
figures confirm the observations from the results of the first validation
example, in that the finite element results gradually converge towards the
prediction results, obtained with the small wave model, and that the
overestimation of the resonance frequencies gets worse for increasing
frequency. Note that, in the considered frequency range, only one cavity-
controlled resonance (192 Hz) of this vibro-acoustic system has been identi-
at cavity location (x1,y1,z1)=(Lx1/4,Ly/4,2Lz1/5) of the second validation example (solid
: wave / dashed : FE/FE 1 (a), FE/FE 2 (b), FE/FE 3 (c))
250 Chapter 4
fied. At this resonance, the frequency overestimation is indeed smaller

compared with the other, plate-controlled resonances.
A comparison of the accuracy and the associated computational efforts,
involved with the wave models and the FE/FE models (see CPU times in
tables 4.1 and 4.2) illustrates the beneficial convergence rate of the proposed
wave based prediction technique. In comparison with the finite element
method, the new prediction technique provides highly accurate predictions
of the coupled vibro-acoustic response with a substantially smaller
computational effort. This beneficial convergence rate is also illustrated in
figure 4.25. This figure plots the relative prediction errors, as defined in
section 2.7.2.1., against the CPU times, involved with some structural and
acoustic results from the various prediction models for the first validation
example (harmonic excitations at 60 Hz and 180 Hz). Note also that the
beneficial convergence rate of the wave based prediction technique will
most likely become even more apparent, when the technique is implemented
in a more efficient software environment, instead of the currently used
MATLAB environment.
Figure 4.25 : convergence curves for the normal plate displacement (left) at the
excitation point (x’F,y’F)=(2Lx’/3,Ly’/3) and the cavity pressure (right) at cavity
location (x,y,z)=(2Lx/3,2Ly’/3,Lz/4) of the first validation example
(solid : wave , dashed : FE/FE) (+ : 60 Hz, ∇ : 180 Hz)
These performance comparisons clearly illustrate the potentials of the

proposed wave based prediction technique for meeting the current challenge
in coupled vibro-acoustic modelling, as outlined in section 1.4.1. Since this
new, deterministic prediction technique provides highly accurate prediction
results in a more efficient way than the currently available (element based)
prediction techniques, its practical frequency limitation for coupled vibro-
acoustic problems may be shifted towards substantially higher frequencies.
4.4.5. Modal analysis of a double-panel partition
This section presents an experimental validation of the wave based

prediction technique, applied for the identification of the low-frequency
modal properties of a double-panel partition, which is a typical coupled
vibro-acoustic system that is often encountered in noise control engineering.
The author and some colleagues of the PMA Noise and Vibration research
group have spent a vast amount of research effort on the investigation of the
low-frequency dynamic behaviour of finite-sized double-panel partitions.
Analytical and numerical prediction models, together with several
experimental validations, have provided considerable insight in the effect of
the vibro-acoustic coupling interaction between the panels and the enclosed
cavity on the low-frequency modal properties and sound transmission
characteristics of finite sized double-panel partitions. Based on the
identified modal properties, several active control configurations have been
designed and successfully implemented, yielding a significant improvement
of the low-frequency sound transmission loss.
Appendix F describes the main low-frequency modal properties of double-
panel partitions, identified from an analytical model, along with a wave
model, obtained with the proposed wave based prediction technique. For a
detailed discussion of the numerical and experimental investigations of the
low-frequency dynamic behaviour and sound transmission characteristics of
double-panel partitions, the reader is referred to DESMET et al. (1994) and
DESMET and SAS (1994b, 1995, 1997). The active control implementations
are discussed in SAS et al. (1993, 1995), DESMET and SAS (1994a) and DE
FONSECA et al. (1996)).
experimental set-up
Figure 4.26 shows the considered double-panel partition. Two identical flat
aluminium panels with dimensions Lx1xLy1=1.14 m x 0.73 m are clamped to a
rigid framework with height Lz1=0.15 m. The thickness of both panels is
252 Chapter 4
t=1.5x10-3 m. The double-panel partition is mounted in the upper opening of

a rigid walled enclosure with dimensions Lx2xLy2xLz2=1.215 m x 0.805 m x
1.065 m. A loudspeaker in this enclosure provides the acoustic excitation of
the double-panel partition.
Figure 4.26 : double-panel partition (experimental set-up)
In order to enable the experimental identification of the modal properties of

this double-panel partition, an array of 7x6 microphones with a spacing of
0.127 m x 0.12 m is positioned in the plane z1=3Lz1/4 in the cavity, enclosed
by both panels. The central parts of each panel are instrumented with an
array of 8x6 accelerometers with a spacing of 0.06 m x 0.06 m. These
accelerometer arrays don’t cover the entire surface of the panels in order to
allow a proper identification of the higher-order modes with the given
number of accelerometers. However, due to the central location of the
accelerometers, their mass loading becomes a local effect. To overcome this
problem, dummy masses with the same weight as the accelerometers
(0.0032 kg) are added to both panels with the same spacing, so that the mass
loading is homogeneously spread over the entire panel surfaces. In total, a

mass of 0.69 kg is added to each panel.
coupled vibro-acoustic wave model

A wave model of type (4.130) is constructed for the prediction of the
coupled vibro-acoustic behaviour of the experimental set-up, shown in
figure 4.26. The steady-state normal displacements of both panels are
approximated as expansions of type (4.108), using 60 structural wave
functions, which are associated with the rectangular panel domains Lx1xLy1.
The steady-state pressure in the cavity between both panels is approximated
as an expansion of type (4.106), using 216 acoustic wave functions, which
are associated with the rectangular cavity domain Lx1xLy1xLz1. The steady-
state pressure in the enclosure is approximated as an expansion of type
(4.106), using 594 acoustic wave functions, which are associated with the
rectangular enclosure domain Lx2xLy2xLz2.
The loudspeaker in the enclosure is modelled as an acoustic point source at
location (x2q,y2q,z2q)=(0.4m,0.365m,0.2m). An ambient density ρ0=1.225
kg/m3 and a speed of sound c=340 m/s are used as fluid properties for the air
in the cavity and enclosure. Since the additional mass (0.69 kg) of the
9 2
accelerometers and dummy masses on each aluminium (E=70.10 N/m ,
v=0.3) panel doesn’t contribute to the panel bending stiffness, the mass
loading effect is taken into account by increasing the aluminium density
from ρs=2790 kg/m3 to ρs=3343 kg/m3 (=2790+0.69/(Lx1xLy1xt)).
Figure 4.27 shows the prediction results for the instantaneous structural and
fluid displacements at 100 Hz in the plane y2=0.6375 m. The accurate
representations of the clamped panel boundary conditions, the continuity
conditions of the normal structural and fluid displacements along the panel-
cavity and panel-enclosure interfaces as well as the rigid walled boundary
conditions of the cavity and the enclosure, illustrate the high prediction
accuracy, obtained with the proposed wave model.
modal properties
Due to the implicit frequency dependence of the acoustic and structural
wave functions, the resulting coupled vibro-acoustic wave model is
frequency dependent, so that the predictions of the undamped natural
frequencies and mode shapes of the considered double-panel partition
cannot be obtained from a standard eigenvalue problem. However, by
calculating the forced response of the undamped double-panel partition, the
resulting resonance frequencies and corresponding operational deflection
254 Chapter 4
shapes are very close to the exact modal parameters, provided that the
resonances are well separated from each other.
Figure 4.27 : instantaneous structural and fluid displacement at 100 Hz in plane

y2=0.6375 m for an acoustic point source excitation at location
(x2,y2,z2)=(0.4m,0.365m,0.2m) (exploded view)
Two major classifications can be made regarding the modes of a double-

panel partition (see appendix F). In a first classification, a distinction is
made between panel-controlled and cavity-controlled modes, depending on
whether most of the modal energy is predominantly stored in the panels or
in the enclosed cavity. A second classification is based on the symmetry of a
double-panel partition, in that both panels have the same modal
displacements patterns, which are either in-phase or out-of-phase, relative to
each other. Table 4.3 adopts these two classifications in a comparison of the
predicted and experimentally identified undamped natural frequencies of the
considered double-panel partition. Note that the modal displacement

patterns of both panels in the considered partition are not completely
identical, since the symmetry of this partition is disturbed by the presence of
the enclosure. However, the effect of the enclosure on the panel
7
displacements is fairly small , compared with the effect of the cavity, since
the enclosure volume is substantially larger than the cavity volume.
panel-controlled predicted experimental

(u,v) in-phase (Hz) in-phase (Hz)
out-of-phase (Hz) out-of-phase (Hz)
(1,1) 19.7
34.2
(2,1) 24.3
22.3
(3,1) 38.6
61.4 63.0
(1,2) 40.2
38.6
(2,2) 47.9 48.1
46.6 45.6
(4,1) 57.6 63.7
55.9 59.3
(3,2) 60.8
59.9 58.5
(1,3) 76.6 76.0
79.6 80.7
(4,2) 79.7
78.5
(5,1) 82.4 85.1
92.3 92.9
(2,3) 84.1
82.0 81.8
(3,3) 96.8 97.6
99.8 100.6
(5,2) 103.7 104.5
102.6
(6,1) 111.9 116.8
109.6 109.6
7
except for the panel displacements at frequencies, close to the resonance
frequencies of the uncoupled enclosure
256 Chapter 4
(4,3) 115.1 116.6

113.9 113.0
(1,4) 125.2 123.5
123.0 120.8
(6,2) 133.0 134.4
131.9
(2,4) 132.8 132.8
131.1 129.0
(5,3) 138.7 140.7
137.7 135.9
(3,4) 145.5 146.5
144.2 142.5
(7,1) 145.9 144.5
147.4 151.6
(4,4) 163.4 164.2
162.1 161.0
(6,3) 167.0
168.6
(7,2) 167.5 168.2
166.2
(1,5) 186.2 181.4
187.2 184.0
cavity-controlled predicted experimental
(l,m,n) out-of-phase (Hz) out-of-phase (Hz)
(1,0,0) 169.1 171.8
Table 4.3 : undamped natural frequencies of the double-panel partition

((u,v): number of half wavelengths of the modal panel displacement in the x1- and y1-
directions; (l,m,n) : number of half wavelengths of the modal cavity pressure field in
the x1-, y1- and z1-directions)
Note that the blanks in table 4.3 indicate that those particular modes could
not (clearly) be identified from the experimental frequency response
functions.
Figures 4.28 and 4.29 compare the predicted forced responses8 with the
corresponding, experimentally identified, modal panel displacements and
modal cavity pressure field of the (1,4) out-of-phase panel-controlled mode.
8
the forced responses on these figures are scaled, such that the maximum upper
panel displacement corresponds with the experimentally identified maximum
modal displacement of the upper panel
Figure 4.28 : predicted forced responses of both panels (123.0 Hz) and
experimentally identified modal panel displacements (120.8 Hz)
for the (1,4) out-of-phase panel controlled-mode
Figure 4.29 : predicted forced response of the cavity (123.0 Hz) and
experimentally identified modal cavity pressure field (120.8 Hz)
for the (1,4) out-of-phase panel controlled-mode
This good agreement has also been observed for the other modes, listed in
table 4.3.
By taking into account, for instance, the inherent uncertainties on the
material properties and the (slight) imperfections, involved with the
practical realisation of the clamped panel boundary conditions, it may be
concluded from the above comparisons between the experimental and
numerical results that the proposed wave model provides a reliable
prediction of the low-frequency modal properties of the considered double-
panel partition.
258 Chapter 4
4.5. Conclusions
The first two parts of this chapter describe how the methodology of the
wave based prediction technique, outlined in chapter 2, is applied for the
prediction of the steady-state out-of-plane bending motion in thin, flat plates
and the steady-state pressure field in three-dimensional uncoupled interior
acoustic systems. Complete wave function sets, together with some
particular solution functions, are proposed for the field variable expansions
in both types of uncoupled dynamic problems. The contributions of the
wave functions to these expansions result from symmetric prediction
models, which are based on a weighted residual formulation of the boundary
conditions.
The third part of this chapter indicates how the proposed field variable
expansions and corresponding wave models for uncoupled acoustic and
uncoupled plate problems are modified for the prediction of the steady-state
dynamic behaviour of three-dimensional coupled vibro-acoustic systems, in
which part of the boundary surface of the acoustic domain consists of a thin,
flat plate.
In all three parts, the practical convergence of the proposed prediction
models is illustrated for several numerical as well as experimental validation
problems. In a similar way as performed in the previous chapter for two-
dimensional problems, the accuracy and the associated computational
efforts of the proposed wave models and corresponding FE/FE models are
compared for some three-dimensional coupled vibro-acoustic validation
problems. Again, these comparisons reveal the beneficial convergence rate
of the proposed wave based prediction technique, in that highly accurate
predictions are obtained with a substantially smaller computational effort.
5. CONCLUSIONS AND
FUTURE DEVELOPMENTS
5.1. Conclusions
At present, the finite element and boundary element method are the most
commonly used numerical prediction techniques for solving steady-state
dynamic problems, defined in structural and acoustic continuum domains.
An important implication of the element based modelling concept is that the
involved large model sizes practically restrict the applicability of these
deterministic prediction techniques to a limited frequency range. Above a
certain frequency limit, these methods would require, even with the
nowadays available high-performance computer resources, a prohibitively
large amount of computational effort and memory resources to get an
acceptable level of prediction accuracy.
For the case of coupled vibro-acoustic problems, this practical frequency
limitation is significantly smaller than for uncoupled structural or uncoupled
acoustic problems. This is mainly due to the following reasons.
• Coupled models are substantially larger, since a structural and an
acoustic problem must be solved simultaneously.
• The numerical solution procedure for coupled models is less efficient,
since coupled models, at least for the most commonly used acoustic
pressure/structural displacement formulation, are no longer symmetric.
• The efficiency of model size reduction techniques such as the modal
expansion and component mode synthesis technique is significantly
smaller.
260 Chapter 5
As a result, there exists still a substantial discrepancy between the limited

frequency range, in which element based models can provide accurate
coupled vibro-acoustic predictions, and the significantly larger frequency
range, in which accurate, deterministic predictions are needed for many
industrial engineering problems.
This dissertation presents a new deterministic prediction technique, whose
very promising characteristics regarding computational efficiency for
solving coupled vibro-acoustic problems may provide the requested
breakthrough for narrowing the currently existing frequency discrepancy.
The new wave based prediction technique follows a Trefftz approach and is
based on three major modelling concepts:
• field variable expansions
The steady-state dynamic field variables in the entire - or at least in large
subdomains of the - acoustic and structural continuum domains of a
coupled vibro-acoustic system are approximated in terms of a set of
acoustic and structural wave functions, which are exact solutions of the
homogeneous parts of the governing dynamic equations, and some
particular solution functions of the inhomogeneous equations. In this
way, the governing dynamic equations are exactly satisfied a priori,
irrespective of the contributions of the wave functions to the field
variable expansions.
• integral formulation of the boundary conditions

Since the proposed field variable expansions are exact solutions of the
governing dynamic equations, the contributions of the wave functions to
the field variable expansions are merely determined by the acoustic and
structural boundary conditions. Since only finite sized prediction models
are amenable to numerical implementation, the boundary conditions can
only be satisfied in an approximate way. Therefore, the wave function
contributions result from an approximate integral - weighted residual or
least-squares - formulation of the boundary conditions.
• wave function selection to ensure convergence

Since (most of) the governing dynamic equations are wave equations
with an infinite number of homogeneous solutions, a selection from this
infinite function set must be made for the construction of the field
variable expansions. In this respect, a key issue in the proposed
prediction technique is the definition of an appropriate wave function
Conclusions and future developments 261
selection, which ensures the convergence of the subsequent field variable

expansions towards the exact solutions.
These modelling concepts are successfully applied for various types of

coupled vibro-acoustic problems. In its current stage of development, the
proposed wave based prediction technique enables already the accurate
prediction of the following steady-state dynamic field variable distributions
in coupled vibro-acoustic systems:
• pressure fields in two- and three-dimensional bounded acoustic domains
with arbitrary geometries,
• pressure fields in two- and three-dimensional1 unbounded acoustic
domains, provided that these fields can be accurately approximated by
confining the unbounded domain to a bounded domain through the
introduction of an artificial boundary surface, on which an appropriate
impedance boundary condition is specified,
• normal displacement fields in one- and two-dimensional flat plates with
arbitrary geometries,
• displacement fields in one-dimensional curved structures, which may be
regarded as assemblages of cylindrical shell sections,
• dynamic field variable distributions in two-dimensional fluid-saturated,
poroelastic domains with arbitrary geometries.
It is illustrated through several validation examples that, for all these types
of field variables, the proposed wave based prediction technique yields
accurate numerical prediction results, which converge towards the exact
solutions.
In this way, the proposed wave based prediction technique provides an
important contribution to the advance of the Trefftz approach. Although this
approach exists for more than seventy years and offers several advantages
over the element based modelling approach, it has still not been
implemented into a valuable modelling alternative for the finite element and
boundary element method. The main reason is that the transformation of a
continuum problem into a Trefftz formulation yields an ill-conditioned
problem. As a result, all attempts2 so far to practically implement a Trefftz-
type prediction model, although theoretically proven to be convergent,
1
although not reported in this dissertation, the modelling procedure for three-
dimensional unbounded acoustic domains applies in a completely similar way as
for two-dimensional unbounded domains
2
for acoustic and structural dynamic problems, only very few, unsuccessful
attempts have been performed so far
262 Chapter 5
suffer from severe numerical convergence problems. An important

breakthrough of the proposed wave based prediction technique is that, for
the expansions of the various field variables that may be encountered in
coupled vibro-acoustic problems, complete wave function sets have been
found, which allow the introduction of the Trefftz approach into a coupled
vibro-acoustic prediction model, whose poor condition is no longer
preventing the numerical results from converging towards the exact
solution.
The new prediction technique has an important contribution to the advance

of coupled vibro-acoustic modelling, due to its enhanced computational
efficiency, compared with the existing prediction methods.
In comparison with the finite element method, for instance, the wave based
prediction technique has some beneficial (+) and some disadvantageous (-)
properties regarding computational efficiency.
(+) Coupled vibro-acoustic wave models are substantially smaller than
corresponding finite element models. This is due to the fact that the
proposed field variable expansions satisfy a priori the governing
dynamic equations, so that approximation errors are only involved with
the representation of the boundary conditions.
(+) The expansions for derived secondary field variables, such as fluid
velocity and structural stress, have the same spatial variation as the
primary field variables, i.e. the acoustic pressure and structural
displacement. This is advantageous for the convergence rate, especially
in the case of coupled vibro-acoustic problems, for which the effect of
the fluid on the structure is pressure-controlled, but for which the effect
of the structure on the fluid is velocity-controlled.
(-) Since the wave functions are complex functions, defined within the
entire continuum domain or at least within large subdomains, the wave
model matrices are fully populated and complex.
(-) Due to the implicit frequency dependence of the wave functions, the
wave model matrices cannot be decomposed into frequency independent
submatrices, which implies that a wave model must be recalculated for
each frequency of interest.
(-) Due to the global and oscillatory nature of the wave functions and due
to the poor condition of a wave model, the involved numerical
integrations are computationally more demanding.
It is identified that the effects of the beneficial properties on the
computational efficiency clearly predominate the effects of the
disadvantageous properties. The accuracy and associated computational
Conclusions and future developments 263
efforts, involved with a wave model and a corresponding finite element

model, are compared for several coupled vibro-acoustic validation
examples. These comparisons reveal the beneficial convergence rate of the
proposed wave based prediction technique, in that it provides highly
accurate prediction results with a substantially smaller computational effort
than the finite element method.
In this respect, it is also identified that the weighted residual formulations,
which are proposed for the various types of boundary conditions and which
yield some symmetric submatrices in a coupled vibro-acoustic wave model,
result in a better convergence rate than corresponding least-squares
formulations.
From the discussion of the methodology and the application of the wave
based prediction technique for two- and three-dimensional coupled vibro-
acoustic problems, it may be concluded that the proposed technique offers
an adequate way to comply with the current challenge in coupled vibro-
acoustic modelling. Due to its beneficial convergence rate, the practical
frequency limitation of the proposed technique is significantly larger than
for the existing prediction techniques and results in a significant narrowing
of the currently existing mid-frequency twilight zone.
5.2. Future developments

In view of obtaining a versatile and generally applicable modelling
technique for coupled vibro-acoustic analysis, the next steps in the
development of the wave based prediction technique will focus on the
following aspects.
1. extension of the applicability and mathematical refinement

To enable the application of the wave based prediction technique for a
general coupled vibro-acoustic problem, convergent field variable
expansions will be defined for the following steady-state dynamic field
variable distributions :
• in-plane displacement fields in two-dimensional flat plates with
arbitrary geometries,
• displacement fields in two-dimensional curved structures, which may
be regarded as assemblages of flat, cylindrical and spherical shell
sections,
264 Chapter 5
• pressure fields in two- and three-dimensional unbounded acoustic

domains without confining the domains,
• dynamic field variable distributions in three-dimensional fluid-
saturated, poroelastic domains with arbitrary geometries.
For these new as well as for the already defined field variable
expansions, a mathematically firm definition will be explored for the
continuum domain geometry, in which the convergence of the field
variable expansions is ensured.
2. automation of the modelling procedure

For the technique to be easily accessible without full knowledge of the
mathematical background, its implementation must exhibit a substantial
level of automation. By analogy with the element based methods and
with the Trefftz-element method (cfr. section 1.4.3.3.), the idea will be
explored of using the proposed field variable expansions to define
standard acoustic, structural and poroelastic ‘elements’. This should
allow to automatically generate a wave model for each element, based on
the specification of its geometrical and material properties, and to obtain
a global wave model from an automated assembly of the element wave
models, based on the boundary conditions and external excitation(s).
In addition, it will be investigated whether these element wave models
may be coupled with conventional finite element and boundary element
models. Such a hybrid modelling procedure would allow to combine the
computational efficiency of a wave model for modelling the
geometrically simple parts of a coupled vibro-acoustic system with the
flexibility of finite element and boundary element models for modelling
the geometrically complex parts.
Appendix A. FINITE ELEMENT METHOD FOR
UNCOUPLED ACOUSTIC PROBLEMS
A.1. Helmholtz equation

Acoustic responses in a fluid are usually regarded as small perturbations to
an ambient reference state. The total pressure field p, mass density field ρ
and velocity vector field v at any time t and any position (x,y,z) in the fluid
may then be expressed as
p( x, y, z, t ) = p0 ( x, y, z, t ) + p’( x, y, z, t )
ρ ( x, y, z, t ) = ρ 0 ( x, y, z, t ) + ρ’( x, y, z, t )

(A.1)
v ( x , y, z , t ) = v 0 ( x , y, z, t ) + v ’( x, y, z, t )
where the primed variables represent the acoustic perturbations to the

ambient fields. In a homogeneous fluid which is initially at rest, the ambient
fields are constant in time and position,

p0 ( x , y, z, t ) ≡ p0 ρ 0 ( x, y, z, t ) ≡ ρ 0 v0 ( x , y, z, t ) ≡ 0 . (A.2)
For the derivation of the dynamic equations that govern the acoustic
perturbation fields, an infinitesimal fluid volume dx.dy.dz is considered.
266 Appendix A
The conservation of mass requires that the increase per unit time of the mass
of the fluid volume equals the net mass entering the volume per unit time.
With the latter shown in figure A.1,
Figure A.1 : mass flow for an infinitesimal fluid volume dx.dy.dz
this yields
∂ρ ∂ ( ρv x ) ∂ ( ρv y ) ∂ ( ρv z )
dx. dy. dz = −( + + )dx. dy. dz . (A.3)
∂t ∂x ∂y ∂z
When an external acoustic source is applied in the fluid, an additional mass

flow is induced in the fluid, which must be taken into account in the
expression for the mass conservation. With q being the volume velocity per
unit volume, induced by the external acoustic source,
q( x, y, z, t ) = q 0 ( x, y, z, t ) + q’( x, y, z, t ) , (A.4)
and with the assumption that there is no ambient external flow ( q 0 ≡ 0 ), the
mass conservation law becomes
∂ ( ρ 0 + ρ’)
= ( ρ 0 + ρ’)q’−∇.[( ρ 0 + ρ’)v ’] . (A.5)
∂t
Finite element method for uncoupled acoustic problems 267
The conservation of momentum requires a dynamic force balance in the

three orthogonal directions x, y and z. Figure A.2 shows the dynamic forces
acting on the infinitesimal fluid volume dx.dy.dz in the x-direction, when the
fluid is assumed to be inviscid.
Figure A.2 : dynamic forces acting in the x-direction
The conservation of momentum in the x-direction becomes then
dv x ∂p
ρdx. dy. dz. = − dx. dy. dz , (A.6)
dt ∂x
where the total time derivative of the x-component of the fluid velocity
vector equals
dv x ∂v x ∂v x
= +
∂v ∂v ∂
. v x + x . v y + x . v z = ( + v . ∇ )v x .
(A.7)
dt ∂t ∂x ∂y ∂z ∂t
Together with similar expressions for the dynamic force balances in the y-
and z-direction, the conservation of momentum may be expressed as a
vector equation
( ρ 0 + ρ’)(
∂
+ v ’. ∇)v ’ = −∇( p0 + p’) . (A.8)
∂t
It is usually assumed that acoustic responses in a fluid occur under adiabatic

conditions. This implies the following pressure-density relation,
268 Appendix A
p = f ( ρ ) = C. ρ γ , (A.9)
where C is constant in time and position and γ=cp/cv is the specific heat ratio
between the specific heat coefficients cp and cv at constant pressure and
constant volume, respectively. For air, for example, this ratio γ equals 1.4.
Since this relation applies also to the ambient state, the constant C may be
expressed as
p
C = γ0 . (A.10)
ρ0
The relation (A.9) may also be expressed as a Taylor-series expansion,
df 1 d2 f
p = ( f ρ= ρ ) + ( ).( ρ − ρ 0 ) + ( ).( ρ − ρ 0 ) 2 +... (A.11)
0 dρ ρ=ρ0
2 dρ 2
ρ=ρ0
By using equations (A.1), this yields
γp0 γ (γ − 1)
p’ = . ρ’+ .( ρ’) 2 +... (A.12)
ρ0 2ρ 02
By retaining only the terms with one primed variable in the aforementioned
expressions, the linearized expressions for the conservation of mass, the
conservation of momentum and the pressure-density relation become
∂ρ’
= ρ 0 . q’− ρ 0 . ∇. v ’ , (A.13)
∂t
∂v ’
ρ0. = −∇p’ , (A.14)
∂t
γp
p’ = 0 . ρ’ . (A.15)
ρ0
Time derivation of (A.13) and substitution of (A.14) and (A.15), yields the
linear acoustic wave equation
1 ∂ 2 p’ ∂q’
∇ 2 p’− 2 = −ρ 0 , (A.16)
c ∂t 2 ∂t
where c denotes the phase speed of an acoustic wave in a fluid,
γp0
c= . (A.17)
ρ0
By confining the acoustic field variables to their steady-state values for a

time-harmonic excitation (e.g. p’(x, y, z,t) = p’(x, y, z).e jωt ), the wave
equation (A.16) transforms into the linear Helmholtz equation
∇ 2 p’+ k 2 p’ = − jρ 0ωq’ , (A.18)
where k(=ω/c) is the acoustic wavenumber. The transformation of the

momentum equation (A.14) yields the relation between the steady-state fluid
velocity vector field and the steady-state pressure field,
j
v’= ∇p’ . (A.19)
ρ 0ω
Note that the primes on the acoustic variables in (A.18) and (A.19) are
usually omitted, when there is little possibility of confusing the total field
variables with their acoustic perturbations.
A.2. Weighted residual formulation

The weighted residual concept provides an equivalent integral formulation
of the linear Helmholtz equation (A.18). The concept defines a steady-state
acoustic pressure field in a bounded fluid volume V as a pressure field, for
which the integral equation
~ p + k 2 p + jρ 0ωq ). dV = 0
∫ p (∇
2
(A.20)
V
270 Appendix A
is satisfied for any weighting function ~p , that is bounded and uniquely

defined within volume V and on its boundary surface Ω.
The weighted residual formulation (A.20) may be reformulated as
∂ ~ ∂p ∂ ~ ∂p ∂ ~ ∂p ∂~p ∂p ∂~p ∂p ∂~
p ∂p
∫ [ ∂x ( p ∂x ) + ∂y ( p ∂y ) + ∂z ( p ∂z )]. dV − ∫ ( ∂x ∂x + ∂y ∂y + ∂z ∂z ). dV
V V
+∫ k2~
pp. dV + ∫ jρ 0ω~
pq. dV = 0 . (A.21)
V V

According to the divergence theorem, the integral of the normal component
of a vector field φ , taken over a closed surface Ω, is equal to the integral of
the divergence of the vector field, taken over the volume V, enclosed by the
surface Ω,

∫ (∇.φ ). dV = ∫ (φ. n ). dΩ , (A.22)

V Ω

where n is the unit normal vector with positive orientation away from the
volume V. Application of this theorem to the first integral term in (A.21)
yields the ‘weak form’ of the weighted residual formulation of the linear
Helmholtz equation,
~ 1 ~
∫ (∇p. ∇p). dV − ω ∫ ( pp ). dV =
2
2
c
V V
ρ ω ~ ρ ω~ Ω
. (A.23)
∫ 0
( j pq ). dV − ∫ 0
( j pv . n ). d
V Ω
A.3. FE implementation
Since the second-order Helmholtz equation (A.18) describes one wavetype,
the pressure field in a bounded fluid volume V is uniquely defined, if one
boundary condition is specified at each point on the boundary surface Ω of
domain V. For uncoupled acoustic problems, three types of boundary
conditions may occur ( Ω = Ω p ∪ Ωv ∪ Ω Z ):
p= p on Ω p ,

(A.24)
v. n = v on Ω v , (A.25)

n
p
v. n = = Ap, on Ω Z , (A.26)
Z
where p , vn and Z (or A ) are prescribed pressure, normal velocity and

normal impedance (or normal admittance) functions, respectively.
construction of an uncoupled acoustic finite element model

In the finite element method, the fluid domain V is discretized into a number
of small subdomains Ve (‘finite elements’) and a number of nodes, say ne,
are defined at some particular locations in each element. Within each

element, the distribution of the field variable, i.e. the pressure p, is
approximated as an expansion p in terms of a number, say np, of prescribed
shape functions N ie , which are only defined within the considered element
domain Ve,
n p
p( x, y, z ) ≈ p( x , y, z ) = ∑ Nie ( x, y, z ). ai ( x, y, z ) ∈ Ve . (A.27)
i =1
The contributions ai in these expansions may be determined from the

weighted residual formulation (A.23)1.
For the commonly used linear tetrahedral and linear hexahedral fluid
elements, the nodes are defined at each corner point of their element
volumes and the number of element shape functions is equal to the number
of nodes (np=ne). Each shape function N ie is defined, such that it has a value
of unity at node i of the element and that it is zero at all other element

nodes. In this way, each contribution ai in the pressure expansion (A.27)
represents the pressure approximation pi at node i of the element,
ne
p( x, y, z ) = ∑ Nie ( x, y, z ). pi ( x, y, z ) ∈ Ve . (A.28)
i=1
1
an uncoupled acoustic problem may also be cast into a variational formulation; for
a detailed discussion of this formulation, the reader is referred to e.g. MORAND and
OHAYON (1995)
272 Appendix A
Based on the element shape functions N ie , which are locally defined in one
element Ve, some global shape functions Ni may be constructed, which are
defined in the entire domain V. In each element domain Ve to which node i
belongs, the global shape function Ni is identical to the corresponding
element shape function N ie , while it is zero in all other element domains. In
this way, a global pressure expansion may be defined as
nf

p( x, y, z ) = ∑ Ni ( x, y, z ). pi = [ N ].{ pi } ( x , y, z ) ∈ V , (A.29)
i=1
where nf is the total number of nodes in the discretization, [ N ] is a (1xnf)

vector of global shape functions and { pi } is a (nfx1) vector of unknown
nodal pressure values. The corresponding pressure gradient approximation
becomes then
 ∂p 

 ∂x 
 ∂p 
∇p ≈ ∇p =   = [∂ ].[ N ]. { pi } = [ B]. { pi } , (A.30)
 ∂y 
 ∂p 

 
 ∂z 
where [∂ ] is a (3x1) vector of gradient operators and [ B] is a (3xnf) matrix

of gradient components of the global shape functions.
In a Galerkin weighted residual approach, the weighting function is
expanded in terms of the same set of shape functions as used for the
pressure expansion,
nf
~
p ( x , y, z ) = ∑ Ni ( x, y, z ). ~pi = [ N ].{ ~pi } ( x, y, z ) ∈V , (A.31)

i=1
p = [∂ ].[ N ]. { ~
∇~ pi } = [ B]. { ~
pi } . (A.32)
The concepts of element discretization and shape function definition are

illustrated in figure A.3 for a two-dimensional fluid volume, which is
discretized into linear rectangular fluid elements.
Figure A.3 : (a) FE discretization of a two-dimensional volume,

(b) element shape functions, (c) global shape function

The determination of the nodal pressure values pi in the expansions (A.29)
and (A.30) is based on the substitution of the expansions (A.29)-(A.32) into
the weak form (A.23) of the weighted residual formulation of the Helmholtz
equation.
For the first term in the left hand side of (A.23), this substitution yields
$ $ !
( ) ( )
∫ (∇p. ∇p). dV = ∫  [ B]. { pi } . [ B]. { pi }  . dV
~ ~ T
V V
T   " # (A.33)
= {~
V
( )
pi } .  ∫ [ B] T .[ B] . dV  . { pi } = { ~

pi } .[ K ]. { pi },
T
274 Appendix A
where [. ] T denotes the transpose of a matrix and [ K ] is a (nfxnf) acoustic

‘stiffness’ matrix. The matrix element Kij on row i and column j of this
matrix is
∂Ni ∂N j ∂Ni ∂N j ∂Ni ∂N j

K ij = ∫ ( + + ). dV . (A.34)
V
∂x ∂x ∂y ∂y ∂z ∂z
Recall that the global shape functions Ni and N j (and their spatial
derivatives) have only non-zero values in the domains of those elements, to
which, respectively, node i and node j belong. As a result, the volume
integration in (A.34) may be confined to the integration over the domains of
those elements, to which both node i and node j belong. Since the latter
integration may be regarded as a sum of integrations over each of the
common element domains and since the global shape functions in each of
these element domains are identical to the corresponding element shape
functions, matrix element Kij may be expressed as
m ij  ∂Nie ∂N j ∂Nie ∂N j ∂Nie ∂N j

e e e 
K ij = 
∑  ∫ ∂x ∂x
( + + ). dV  (A.35)
e = 1  Ve ∂y ∂ y ∂z ∂ z 

where mij is the number of elements, to which both node i and node j belong.
Since each node belongs to common elements with only a few, adjacent
nodes, only a few matrix elements Kij are non-zero. This results in a sparsely
populated stiffness matrix [ K ] .
Due to this advantageous matrix property, the practical calculation of the
stiffness matrix can be performed in a very efficient way. By confining the
volume integration in (A.33) to one element domain, one may write
)*) %
{& }
 
{ } 
( [ ]) ( [ ]) e 
T
e T  . pe
∫ ( ∇~
p . ∇p ). dV = ~
pi .
 ∫  [∂ ]. N e
. [∂ ]. N 

. dV
 i
Ve  Ve 
{' } { } [ ] {( }
 
{ } 
[ ] [ ]

pie .  ∫  B e . B e  . dV  . pie = ~
T T T
= ~ pie . K e . pie ,
   
 Ve 
(A.36)
+
[ ]
where N e is a (1xnp) vector of element shape functions and { pie } is a
(npx1) vector of unknown nodal pressure values of the considered element.
The matrix elements in the associated (npxnp) element stiffness matrix K e [ ]
are
∂Nie ∂N j ∂Nie ∂N j ∂Nie ∂N j

e e e
K ije = ∫ (
∂x ∂x
+
∂y ∂y
+
∂z ∂z
). dV . (A.37)
Ve
Note that, since all element shape functions have non-zero values in their
element domain Ve, each element stiffness matrix is now fully populated.
The calculation of the global stiffness matrix [ K ] may now be performed in
a two-step procedure. In the first step, all element stiffness matrices are
calculated. In the second step, each non-zero element Kij of the global
stiffness matrix is obtained, according to (A.35), from a simple addition of
the corresponding entries (A.37) in the appropriate element stiffness
matrices. With an appropriate numbering of the nodes in the FE
discretization, the non-zero entries in the stiffness matrix appear in a narrow
band around the matrix diagonal, yielding a sparsely populated, banded
stiffness matrix.
In a completely similar way, the second term in the left hand side of (A.23)
may be expressed as
1 ,
p. p). dV = −ω 2 { ~
−ω 2 ∫ ( 2 ~
T   1   -
pi } .  ∫  2 [ N ] T .[ N ] . dV  . { pi }
V c V c   (A.38)
= −ω 2 { ~
.
p } .[ M ]. { p }
T
,
i i
where [ M ] is a (nfxnf) acoustic ‘mass’ matrix. As for the global stiffness

matrix, the practical calculation of this sparsely populated, banded global
mass matrix is based on an assemblage of the element mass matrices,
according to
m ij  
Mij = ∫ ( 2 N i N j ). dV = ∑  ∫ ( 2 N ie N ej ). dV  .
1 1
(A.39)

e = 1  Ve c

V c 
276 Appendix A
The first term in the right hand side of (A.23) may be expressed as
 
∫ ( jρ 0ωpq ). dV = { pi }
~

V
∫ 0 ( 
)
 { pi } . {Qi } ,
~ T .  jρ ω [ N ] T q . dV  = ~ T
(A.40)
V
where {Qi } is a (nfx1) acoustic source vector. When the distribution q of

external acoustic sources is confined, for instance, to an acoustic point
source of strength qi , located at node i, the source distribution q is
q( x, y, z ) = qi .δ ( xi , yi , zi ) , (A.41)
where δ is a Dirac delta function at node i. The subsequent source vector

becomes then
 
{Qi } = jρ 0ω  ∫ (qi .[ N ] T .δ ). dV  . (A.42)
V 
Provided that node i is not located on the boundary surface of V, all

components of the (nfx1) source vector are zero, except the component on
row i, which equals jρ0ωqi .
The second term in the right hand side of (A.23) allows the introduction of
the boundary conditions into the FE formulation. Since the integration over
the boundary surface Ω may be regarded as a sum of the integrations over
the subsurfaces Ωv , Ω Z and Ω p and since the normal velocity and normal
impedance boundary conditions (A.25) and (A.26) must be satisfied on,
respectively, Ωv and Ω Z , the second term in the right hand side of (A.23)
may be expressed as
− ∫ ( jρ 0ω~
pvn ). dΩ − ~ / ~ 0 0
∫ ( jρ 0ωpAp). dΩ − ∫ ( jρ 0ωpv . n ). dΩ . (A.43)
Ωv ΩZ Ωp
The substitution of expansion (A.31) into the first term of (A.43) yields
− ∫ ( jρ 0ω~
pvn ). dΩ =
Ωv
  (A.44)
{ pi } .  ∫ ( − jρ 0ω [ N ] vn ). dΩ  = { pi } . {Vni }
~ T  T ~ T
Ωv 
The component on row i of the (nfx1) input velocity vector {Vni } is thus
Vni = ∫ (− jρ 0ωNi vn ). dΩ . (A.45)

Ωv
The boundary surface of an element is the union of all its faces. The
boundary surface of a linear tetrahedral or a linear hexahedral fluid element,
for instance, is the union of, respectively, four and six faces. For these
compatible elements, the value of a global shape function Ni at a certain
element face is only non-zero, if node i is located on the considered element
face. As a consequence, the value of a global shape function Ni at the
boundary surface Ωv and the subsequent vector component Vni are only
non-zero for those nodes that are located on the boundary surface Ωv .
Hence, the practical calculation of the input velocity vector is based on its
component expression
m vi f vie  
 ( N e . v ). dΩ  ,
Vni = − jρ 0ω ∑ ∑ ∫ i n 
(A.46)
e = 1 f = 1 Ω f 
e
where mvi is the number of elements, for which node i is located on their
fvie element faces Ωef , that are part of the boundary surface Ωv . The
prescribed normal velocity at a certain location in an element face is often
specified by a shape function expansion, comparable to the pressure
expansion (A.31),
[ ]{ }
vn ( x, y, z ) = {n} T . N ve
f
. vef ( x, y, z ) ∈ Ω ef , (A.47)
278 Appendix A
where the (3x1) vector {n} consists of the x-, y- and z-components of the
unit vector, normal to the considered element face. The matrix
 N 1e 0 0 1 N nev 0 0 
[ ] f
N ve

= 0 N 1e 0 1 0 N nev 0 

(A.48)

 0 0 N 1e 1 0 0

N nev 

is a (3x3nv) matrix, which contains the pressure shape functions of the nv

nodes, located on the considered element face. In each of these nodes, the x-,
y- and z-component of the fluid velocity is specified, yielding the (3nvx1)
vector
{vef } = {vx 1
T
v y1 vz 1 2 v xnv v yn v }
v zn v . (A.49)
The second term in (A.43) may be expressed as
3   4
− ∫ ( jρ 0ωpAp). dΩ = − jω { pi }
~
 ∫ 0 ( { }
~ T .  ρ A [ N ] T .[ N ] . dΩ  . p
 i (A.50) )
ΩZ Ω Z 
= − jω { ~
pi } .[ C ]. { pi }
T 5
where [ C ] is a (nfxnf) acoustic damping matrix. The matrix element Cij on

row i and column j of this matrix is
Cij = ∫ ( ρ 0 ANi N j )dΩ . (A.51)

ΩZ
As it is the case for the stiffness and mass matrix, the damping matrix is
sparsely populated, since matrix element Cij is only non-zero, if node i and
node j are located on at least one common element face that is part of the
boundary surface Ω Z . The practical calculation of the non-zero matrix
elements is therefore based on the expression
m Zij  
 ( ρ A . N e . N e ). dΩ  ,
Cij = ∑ ∫ 0 i j 
(A.52)
f = 1 Ω f 
e
where mZij is the number of element faces Ωef , on which both node i and
node j are located and that are part of the boundary surface Ω Z . The
specification of the prescribed normal admittance is usually restricted to a
constant value per element face in Ω Z .
The substitution of expansion (A.31) into the third term of (A.43) yields
− ∫ ( jρ 0ω~
6 6
pv . n ). dΩ = { ~
 6 6 
pi } .  ∫ ( − jρ 0ω [ N ] T v . n ). dΩ  = { ~
T
pi } . {Pi }
T
 
Ωp Ωv 
(A.53)
Due to the particular shapes of the global shape functions Ni , the
component on row i of the (nfx1) vector {Pi } ,
7 7
Pi = ∫ ( − jρ 0ωNi v . n ). dΩ , (A.54)
Ωp
is only non-zero, if node i is located on the boundary surface Ω p . Since the

latter expression doesn’t allow the introduction of the prescribed pressure
boundary condition (A.24), this boundary condition enters the finite element
model in a different way, as will be discussed later.
By substituting expressions (A.33), (A.36), (A.40), (A.44), (A.50) and
(A.53) into (A.23), the weak form of the weighted residual formulation of
the Helmholtz equation, including the boundary conditions (A.25) and
(A.26), becomes
8
{ ~pi }T . ([ K ] + jω [ C] − ω 2 [ M ]).{ pi } = { ~pi }T . ({Qi } + {Vni } + {Pi }) . (A.55)
Since the weighted residual formulation should hold for any expansion of
the weighting function, i.e. for any set of shape function contributions { ~
pi }
9
(see (A.31)), a set of nf equations in the nf unknown nodal pressure
approximations pi is obtained,
([ K ] + jω[ C] − ω 2[ M ]).{ p: i } = {Qi } + {Vni } + {Pi } . (A.56)

280 Appendix A
Row i in this matrix equation expresses the weighted residual formulation,

in which the global shape function Ni , associated with node i, is used as
weighting function ~p.
The prescribed pressure boundary condition (A.24) is not yet included in
matrix equation (A.56). This is usually done by directly assigning the
prescribed pressure value at each node location on the boundary surface Ω p
;
to its corresponding nodal unknown pi . When this assignment is done for
<
the n p nodes on Ω p , only na (= n f - n p ) pressure approximations pi are
still unknown. This means that n p equations should be eliminated in matrix
equation (A.56) to have a well-determined set of equations. This is usually
done by eliminating each row in (A.56), that expresses the weighted residual
formulation, in which the global shape function of a node on the boundary
surface Ω p is used as weighting function. The elimination of these
equations, which orthogonalize the error on the pressure predictions in the
region near the boundary surface Ω p with respect to the shape functions in
this region, is motivated by the fact that this prediction error is smaller than
the errors in the other regions of the fluid domain, since the exact pressure
values at the nodes of the boundary surface Ω p are a priori assigned.
By eliminating the appropriate rows in (A.56) and by shifting all terms in
the left hand side of (A.56), which contain the a priori assigned nodal
pressure values, to the right hand side vector, the resulting finite element
model for an uncoupled acoustic problem is obtained,
([ Ka ] + jω[Ca ] − ω 2[ Ma ]).{ pi } = {Fai } , (A.57)
where the (nax1) vector { pi } contains the remaining unknown nodal

pressure approximations and where the acoustic stiffness, damping and mass
matrices [ K a ] , [ Ca ] and [ M a ] are now (naxna) matrices. Since the non-
zero components of {Pi } occur only in the eliminated equations of (A.56),
the (nax1) acoustic force vector {Fai } contains the stiffness, damping and
mass terms in the a priori assigned nodal pressures and the contributions
from the acoustic source vector (A.42) and the input velocity vector (A.45).
convergence and parametric mapping

The highest-order spatial derivatives of the pressure p and the weighting
function ~p in the weighted residual formulation (A.23) are of first order.
Hence, the convergence of an acoustic finite element discretization is
ensured, if the expansions (A.29) and (A.31) for, respectively, p and ~ p
consist of polynomial shape functions, which are complete up to at least the
first order within each element and if they are continuous along the
interelement boundaries (C0-continuity).
The most commonly used elements in acoustic finite element modelling are
linear quadrilateral and linear triangular elements for two-dimensional
problems and linear hexahedral and linear tetrahedral elements for three-
dimensional problems. Their parent elements are shown in figure A.4.
Figure A.4 : linear rectangular (a), linear triangular (b),

linear rectangular prism (c) and linear tetrahedral (d) parent elements
282 Appendix A
The pressure shape functions for the parent elements are defined as follows :
• linear rectangular element (figure A.4 (a))
1
N ie = (1 + ξ i ξ )(1 + η iη ) (i = 1..4) (A.58)
4
• linear triangular element (figure A.4 (b))
N ie = Li (i = 1..3) , (A.59)
where the normalized area co-ordinate system (L1,L2,L3) is related to the

global cartesian co-ordinate system (x,y) :
 L1   b1 c1 a1   x 
  1   
 L2  =  b2 c2 a 2   y  (A.60)
L  2 ∆
î 3 b3 c3 a 3  î 1
where
 1 x 1 y1 
1  
∆ = det  1 x 2 y 2  ,
2 
y 3 
(A.61)
 1 x 3
a1 = x 2 y 3 − x 3 y 2 , b1 = y 2 − y 3 , c1 = x 3 − x 2
The other parameters are obtained from cyclic rotation of the indices.
• linear rectangular prism element (figure A.4 (c))
1
N ie = (1 + ξ i ξ )(1 + η i η )(1 + ζ iζ ) (i = 1..8 ) (A.62)
8
• linear tetrahedral element (figure A.4 (d))
N ie = Li (i = 1..4) , (A.63)
where the normalized volume co-ordinate system (L1,L2,L3,L4) is related

to the global cartesian co-ordinate system (x,y,z) in a similar way as the
relations (A.60) and (A.61) for the normalized area co-ordinate system.
The convergence of a finite element discretization with these element types

is ensured. All these parent elements satisfy the necessary completeness
condition, since the polynomial shape functions comprise at least all terms
up to the first order. The efficiency of triangular and tetrahedral elements
may be noticed in this respect, since their number of nodes equals the
minimum number, required for a complete first-order polynomial
expansion. All parent elements satisfy also the C0-continuity condition, since
the pressure in each element boundary is completely defined by the pressure
degrees of freedom of the nodes in that boundary. For the linear rectangular
element (figure A.4 (a)), for instance, the pressure along the boundary
between nodes 3 and 4 has a linear shape, which is uniquely defined by the
pressure values in nodes 3 and 4. Since in the adjacent rectangular element
the pressure along this boundary has also a linear shape, which is uniquely
defined by the common nodes 3 and 4, the pressure continuity along this
boundary is ensured.
Linear quadrilateral and linear hexahedral elements are obtained from an
isoparametric mapping of, respectively, the linear rectangular and the linear
rectangular prism elements. The geometry of these isoparametric elements,
i.e. the global cartesian co-ordinates of each point in the element, is defined
by a one-to-one correspondence between a point in the parent element and a
point in the mapped element. This relation is expressed as an expansion in
terms of the same shape function as used for the pressure expansion (A.28);
the contributions of each shape function to the expansion are the desired
global co-ordinates of the element nodes,
[ ]
ne
x= ∑ Nie (ξ , η,ζ ). xi = N e {xi }
i=1
[ ]
ne
y= ∑ Nie (ξ, η,ζ ). yi = N e { yi } . (A.64)
i=1
[ ]
ne
z= ∑ Nie (ξ, η,ζ ). zi = N e {zi }
i=1
284 Appendix A
Figure A.5 illustrates this concept for the isoparametric mapping of a linear
rectangular parent element onto a linear quadrilateral element.
Figure A.5 : isoparametric linear quadrilateral element
For this type of isoparametric mapping, the convergence properties of the

parent discretization are preserved in the mapped discretization.
According to the rules of partial differentiation, the spatial derivatives with
respect to the local element co-ordinates ξ, η and ζ may be expressed in
terms of those with respect to the global cartesian co-ordinates x, y and z ,
 ∂N ie   ∂x ∂y ∂z   ∂Ni 
e  ∂Nie 
      
 ∂ξ   ∂ξ ∂ξ ∂ξ   ∂x   ∂x 
 ∂N ie   ∂x ∂y ∂z   ∂Nie   ∂Ni 
e
 =  .
∂η   ∂y 
 = [ J (ξ , η , ζ ) ] . (A.65)
 ∂η   ∂η ∂η  ∂y 
 ∂N ie   ∂x ∂y ∂z   ∂N e   ∂Nie 
 i
 ∂ζ   ∂ζ ∂ζ ∂ζ  î ∂z   ∂z 
î  î 
Each element in the Jacobian matrix [ J ] is a function in the local element

co-ordinates ξ, η and ζ , which is obtained from the mapping relationship
(A.64). Based on the expressions (A.64) and (A.65), the integration
variables in the integrals, which determine the components of the force
vector and the stiffness, mass and damping matrix of the finite element
model (A.57), may be transformed from the global cartesian to the local
element co-ordinates. This co-ordinate transformation yields, for instance,
for an element stiffness matrix, as defined in (A.36),
[ K e ] = ∫  ([∂ ].[ N e ]) .([∂ ].[ N e ]) . dx. dy. dz

T
Ve
(A.66)
(
∫ ∫ ∫  [∂ ].[ N ]) .([∂ ].[ N e ]) det([ J ]). dξ. dη. dζ .
1 1 1
 e
T
=
−1 −1 −1
in which the components of vector [∂ ] , which are the partial differential

operators with respect to the global co-ordinates x, y and z, are expressed in
terms of the partial differential operators with respect to the local element
co-ordinates ξ, η and ζ , using the inverse of the Jacobian matrix (see
(A.65)).
properties of an uncoupled acoustic finite element model

Several properties of the stiffness, mass and damping matrices of an
acoustic finite element model may be derived from their element
expressions.
• As already mentioned before, the matrices are sparsely populated due to

the local character of the shape functions. Moreover, with an appropriate
numbering of the nodes in the discretization, which is automatically
generated in commercial FE programs, the matrices may have a banded
structure, i.e the non-zero matrix elements may be arranged in a narrow
band around the matrix diagonal.
• Since the indices i and j in the element expressions (A.34), (A.39) and
(A.51) may be interchanged, the matrices are symmetrical.
• Since the frequency ω doesn’t occur in (A.34) and (A.39), the elements
of the stiffness and mass matrices are frequency independent. Although
the frequency doesn’t occur in an explicit way in (A.51), the elements of
the damping matrix are usually frequency dependent, due to the
frequency dependence of the prescribed normal impedance (or
admittance).
• Since all element shape functions have real values, the elements of the
stiffness and mass matrices are real, while the elements of the damping
matrix are usually complex, since the prescribed normal impedance (or
admittance) is usually a complex function.
286 Appendix A
modal expansion technique

The predictions for the undamped mode shapes of an interior acoustic
system, in which the entire boundary surface is assumed to be perfectly
rigid, are obtained by discarding the damping matrix [ Ca ] and the external
excitation vector {Fa } in the finite element model (A.57). Since the
stiffness and mass matrices [ K a ] and [ M a ] are independent of frequency,
the mode shape predictions are obtained from the following eigenvalue
problem,
[ K a ]{Φ m } = ω m2 .[ Ma ]{Φ m } ( m = 1.. na ) , (A.67)
where each (nax1) eigenvector {Φ m } represents a mode shape and where

the associated eigenvalue corresponds with the square of the natural
frequency ωm of that mode. Note that, due to the discretization of the
acoustic system, which has an infinite number of degrees of freedom and,
hence, an infinite number of modes, into a system with na degrees of
freedom, only na mode shapes are obtained.
Instead of directly solving the (large) finite element model (A.57) for the
unknown nodal pressure degrees of freedom, an alternative, modal
approach, may be used. In the modal expansion technique, the nodal degrees
of freedom in the (nax1) vector { pi } are expressed in terms of a set of
ma(≤na) modal vectors,
ma
{ pi } = ∑ φ m . {Φ m } = [Φ ]. {φ m } , (A.68)
m=1
where [Φ ] is a (naxma) matrix of modal vectors and where {φm } is a (max1)

vector of modal participation factors.
By substituting the modal expansion (A.68) into the finite element model
(A.57) and by premultiplying both sides of the resulting matrix equation
with the transpose of the modal vector matrix, the following modal model is
obtained,
([ K~ ] + jω[C~ ] − ω [ M~ ]).{φ
a a
2
a m } = {Fa } ,
~
(A.69)
where the (max1) modal excitation vector is
{F~a } = [Φ ] T {Fai } (A.70)
and where the (maxma) modal stiffness, mass and damping matrices are
[ K~a ] = [Φ ]T [ Ka ][Φ ], [ M~ a ] = [Φ ] T [ Ma ][Φ ], [C~a ] = [Φ ] T [Ca ][Φ ] . (A.71)

Due to the orthogonality of the modal vectors with respect to the mass
matrix,
{Φ m } [ Ma ]{Φ m 2 } = 0,
T
1
if m1 ≠ m 2 , (A.72)
the modal mass and modal stiffness matrices are diagonal matrices. By
normalizing the modal vectors, according to
{Φ m } [ Ma ]{Φ m1 } = 1,
T
1
( m1 = 1.. ma ), (A.73)
the modal mass matrix becomes the unity matrix,
1 
=
[ ]
Ma = 
~

 ,

(A.74)
 1
and the diagonal modal stiffness matrix becomes
ω 12 
 > 
[ ] ~
Ka = 
 2 
 . (A.75)

ω ma 
When the finite element model (A.57) has a proportional damping matrix,
i.e. a damping matrix, which may be written as a linear combination of the
stiffness and mass matrix,
288 Appendix A
[Ca ] = α [ K a ] + β[ Ma ] , (A.76)
the modal damping matrix is also a diagonal matrix,
 2ζ 1ω 1 
?
[ ]
~ 
Ca = 


, (A.77)
 2ζ m a ω m a 
with the modal damping ratio’s
αω m β
ζm = + . (A.78)
2 2ω m
Although all modes should be used (ma=na) in the modal expansion (A.68) to
get the same prediction accuracy with the modal model (A.69) as with the
direct finite element model (A.57), a modal model with a relatively small
truncated set of modes yields already a level of accuracy, close to the one of
the much larger direct model. In this framework, a rule of thumb states that
an accurate prediction of the steady-state dynamic behaviour at a certain
frequency ω is obtained by using all the modes, whose natural frequencies
ωm are smaller than 2ω. In this way, a significant model size reduction is
obtained (ma<<na), especially in the low-frequency range, where the modal
densities are small. An illustration of the efficiency of the modal expansion
technique is given in the following section.
A.4. One-dimensional example

problem definition
In this section, the above described finite element method is applied for a
one-dimensional acoustic problem. A tube of length Lx is filled with a fluid
with density ρ0 and speed of sound c. At one end of the tube, a time-
harmonic velocity with amplitude V (displacement amplitude X) and
frequency ω is imposed, while the other end of the tube is rigid walled (see
figure A.6(a)). The steady-state pressure response in the fluid is
approximated using a finite element discretization, which consists of n two-
noded elements of length L=Lx/n, yielding a total of (n+1) nodes (see figure
A.6(b)).
Figure A.6 : one-dimensional acoustic problem (a) and its FE discretization (b)
The exact steady-state pressure in the tube is
ωx 1 ωx
pexact ( x ) = ρ 0 cωX [sin( )+ cos( )], (A.79)
c ωL c
tan( x )
c
in which the harmonic time dependence e jωt is suppressed.
direct finite element model

The pressure within each element i between nodes i and (i+1) is
approximated in terms of two linear shape functions ( x ∈ [x i , x i + 1 ] ),
x − xi x − xi
p( x ) ≈ pi . N ii ( x ) + pi + 1. N ii+ 1 ( x ) = pi .(1 − ) + pi + 1 .( ) , (A.80)
L L
where xi is the location of node i.

According to (A.37) and (A.39), the element stiffness and element mass
matrices for this discretization become,
[ K e ] = −1 / L [ M e ] = c12  L / 6

1/ L −1 / L  L / 3 L / 6
, . (A.81)
1 / L  L / 3
The resulting finite element model becomes

290 Appendix A
 p1   − ρ 0 ω 2 X 
 p   
( 2
) @  2  
[ K a ] − ω [ Ma ] .   = 
  
@
0 
,

(A.82)
î pn + 1   
î 0 
in which the global stiffness and global mass matrices are
 1 / L −1 / L 
 −1 / L 2 / L −1 / L 0 
 
 −1 / L 2 / L −1 / L 
 ABABA 
[ Ka ] =   , (A.83)
 −1 / L 2 / L −1 / L 
 
 0 −1 / L 2 / L −1 / L 
 −1 / L 1 / L 
L / 3 L / 6 
 L / 6 2L / 3 L / 6 0 
 
 
1  CDCDC
L / 6 2L / 3 L / 6

[ Ma ] = 2   . (A.84)
c  
L/6 2L / 3 L / 6
 
 0 L / 6 2L / 3 L / 6 
 L / 6 L / 3
Since the pressure expansion (A.80) enables the representation of a

complete polynomial function of the first order and since the pressure has a
unique value at the interelement boundaries, the convergence of this type of
finite element discretization is ensured.
Figure A.7 compares the finite element predictions (n=20) with the exact
solutions for the case of an air-filled (ρ0=1.225 kg/m3, c=340 m/s) tube of
length Lx=1 m. At the left hand side, a displacement is imposed with
amplitude X=10-3 m. Two excitation frequencies are considered: ω/2π=100
Hz (figures A.7(a,c)) and 500 Hz (figures A.7(b,d)).
Figure A.7 : instantaneous steady-state pressure and displacement

(solid : exact, x-marked : FE (n=20))
To indicate the accuracy of a certain finite element approximation, a global

measure ∆p for the pressure predictions and a global measure ∆v for the
velocity (and displacement) predictions may be defined as follows :
Lx Lx 2
2 ∂p ∂pexact
∫ p − pexact . dx ∫ ∂x
−
∂x
. dx
∆p = 0
Lx
, ∆v = 0
Lx 2
. (A.85)
2 ∂pexact
∫ pexact . dx ∫ ∂x
. dx
0 0
292 Appendix A
An indication of the convergence rate is obtained by plotting the number of

nodes (n+1) of a certain FE discretization against the corresponding
accuracy measures ∆p and ∆v. Figure A.8 shows the resulting curves for the
two considered excitation frequencies.
Figure A.8 : pressure (solid) and velocity (dotted) convergence rates
Figures A.7 and A.8 illustrate some important features of the finite element
prediction for solving steady-state dynamic problems.
• The accuracy, obtained from a given finite element discretization,

decreases, when the excitation frequency increases
The dynamic response of an acoustic system results from a complex
wave propagation mechanism in the fluid. The wavelength λ of a freely
propagating harmonic wave in a fluid is proportional to the speed of
sound c in the fluid and inversely proportional to the frequency ω,
2πc
λ= . (A.86)
ω
Consequently, the spatial variation of the dynamic response, which is

strongly related to the wavelength λ, increases for increasing frequency.
This is clearly illustrated in figure A.7, where the period of the exact
pressure oscillations in the tube corresponds with the fact that Lx≈0.3λ at
100 Hz and Lx≈1.5λ at 500 Hz.
Since the pressure distribution is determined by low-order polynomial
shape functions, a finite element discretization yields only an
approximation with a limited spatial variation. In this respect, a rule of
thumb states that a discretization should have at least 10 elements per
wavelength to keep the discretization error within acceptable limits.
Hence, according to (A.86), the accuracy, obtained from a given
discretization (i.e. a given number of nodes and elements), decreases
with increasing frequency, as illustrated in figure A.8.
• The accuracy for derived secondary variables is smaller than for

primary field variables
According to (A.19), the steady-state fluid velocity and fluid
displacement distributions in an acoustic field are proportional to the
spatial gradient of the pressure distribution. Since the fluid velocity and
displacement in a freely propagating harmonic wave have the same
spatial period as the pressure, the dynamic pressure and velocity (and
displacement) distributions in an acoustic system have similar spatial
variations, as may be seen from the exact solutions in figure A.7.
In a finite element approximation, the velocity and displacement
distributions are derived secondary variables, in that they are expressed
in terms of the first-order derivatives of the pressure shape functions.
Since the order of the derived velocity shape functions is smaller than the
order of the pressure shape functions, the velocity and displacement
approximations have a smaller spatial variation than the pressure
approximation. Since this is in contrast with the above mentioned
physical reality, the accuracy for the derived secondary variables in a
given discretization is smaller than for the primary variables. In the
present example, in which the pressure is approximated as a linear
function within each element, the corresponding velocity and
displacement predictions are constant within each element and
discontinuous at the element boundaries, as shown in figure A.7. The
subsequent accuracy of the velocity predictions in a given discretization
is smaller than the pressure accuracy, as illustrated by the pressure and
velocity convergence rates in figure A.8.
294 Appendix A
modal model
The exact natural frequencies of the modes of the considered acoustic tube,
assuming both tube ends to be perfectly rigid, are
ω m ( m − 1)c
f m, exact = = ( m = 1, 2,..). (A.87)
2π 2Lx
The predictions of the modes and their natural frequencies for the finite
element model (A.82), which results from a discretization of the tube into n
linear elements, are obtained from the eigenvalue problem
[ K a ]{Φ m } = ω m2 .[ Ma ]{Φ m } ( m = 1..(n + 1)) , (A.88)
where each ((n+1)x1) eigenvector represents a predicted mode shape
 p1, m 
 p 
{ m} 
Φ =
 2, m  E , (A.89)
 
î pn + 1, m 
and where its associated eigenvalue represents the squared value of the
predicted natural frequency.
Table A.1. compares the exact and predicted natural frequencies of the first
10 modes of an air-filled (ρ0=1.225 kg/m3, c=340 m/s) tube of length Lx=1
m, which is discretized into n=100 linear elements.
exact (A.87) FE (n=100)
0 Hz 0.000 Hz
170 Hz 170.007 Hz
340 Hz 340.056 Hz
510 Hz 510.189 Hz
680 Hz 680.448 Hz
850 Hz 850.874 Hz
1020 Hz 1021.511 Hz
1190 Hz 1192.399 Hz
1360 Hz 1363.582 Hz
1530 Hz 1535.101 Hz
Table A.1 : exact and predicted natural frequencies

This table illustrates the typical feature of a finite element model regarding
mode extraction: due to the element discretization, the natural frequencies
are systematically overestimated and the absolute and relative overesti-
mations increase for increasing frequency.
By expressing the unknown nodal pressure degrees of freedom in the finite

element model (A.82) in terms of a set of ma modal vectors, as proposed in
(A.68), the modal model (A.69) for the considered acoustic tube becomes
ω 12 − ω 2   φ 1   − ρ 0ω 2 X. p1,1 
 F 0
 G  G 
 .  =  (A.90)


0 ω m a − ω  î φ m a  î − ρ 0ω X . p1, m a 
2 2    2
where the unknowns are the modal participation factors φm in the modal
expansion (A.68).
The modal model (A.90) was solved for the case of an air-filled (ρ0=1.225
kg/m3, c=340 m/s) tube of length Lx=1 m with an imposed displacement at
the left hand side with amplitude X=10-3 m and frequency ω/2π=500 Hz. The
tube discretization consisted of n=100 linear elements. Figure A.9 plots the
resulting accuracies of the pressure and velocity predictions, indicated by
their measures ∆p and ∆v (see (A.85)), against the number of modes ma in
the modal model.
This figure clearly indicates the efficiency of the modal expansion
technique. The steep decrease of the error curves for small values of ma
indicates that a modal model with a relatively small number of modes yields
an accuracy, close to the accuracy, obtained with a direct solution of the
much larger finite element model. Note that the latter accuracy corresponds
with a modal solution, in which all modes are taken into account (ma=n+1).
As mentioned in section A.3, a rule of thumb states that a good accuracy is
obtained by using all the modes, whose natural frequencies are smaller than
2ω. According to this rule of thumb and to table A.1, a modal solution with
the first 7 modes should have an accuracy, close to one of the direct
solution, for an excitation frequency of 500 Hz. This is indeed confirmed by
figure A.9, in which the error measures for this modal model are ‘+’-
marked.
296 Appendix A
Figure A.9 : pressure (solid) and velocity (dashed) convergence rates at 500 Hz
Appendix B. FINITE ELEMENT METHOD FOR
UNCOUPLED STRUCTURAL PROBLEMS
B.1. Dynamic equations for thin, flat plates

According to Kirchhoff’s plate theory, the in-plane and out-of-plane
motions in a thin, flat plate are decoupled and are governed by separate
dynamic equations.
B.1.1. Out-of-plane bending motion
assumptions
• The plate material is elastic and its elastic quantities are governed by
Hooke’s law.
• The plate material is homogeneous and isotropic, with an elasticity
modulus E, a density ρs and a Poisson coefficient ν.
• The plate thickness t is small in comparison with the other dimensions of
the plate.
• The plate displacements are small in comparison with the plate thickness.
• Sections, normal to the middle plate surface, remain normal to this
surface after deformation.
• The stress component σz in the direction, normal to the plate surface, is
negligible.
• There are no stresses in the plate middle surface.
298 Appendix B
• The external loads are applied in the direction, normal to the plate
surface.
stresses and strains

The stresses in an infinitesimal part t.dx.dy, induced by out-of-plane motion
in a flat plate, are shown in figure B.1(a). These stresses induce some
resulting forces and moments per unit length, acting on a plate section,
t/2 t/2
mx = ∫ σ x . z. dz , qx = ∫ τ xz . dz
−t / 2 −t / 2
t /2 t/2
my = ∫ σ y . z. dz , qy = ∫ τ yz . dz (B.1)
−t / 2 −t / 2
t/2 t/2
m xy = ∫ τ xy . z. dz = ∫ τ yx . z. dz = m yx
−t / 2 −t / 2
The positive orientations of these forces and moments are shown in figure
B.1(b).
Figure B.1 : stresses (a) and their resultants (b)

in an infinitesimal part t.dx.dy
Finite element method for uncoupled structural problems 299
Based on the above mentioned assumptions, the displacement components

wx(x,y,z) and wy(x,y,z) in, respectively, the x- and y-direction may be related
to the displacement component wz(x,y) of the plate middle surface (z=0) in
the z-direction,
∂w z ( x, y )
w x ( x , y, z ) = − z , (B.2)
∂x
∂w ( x , y )
w y ( x, y, z ) = − z z . (B.3)
∂y
Relation (B.2) is also illustrated in figure B.2.
Figure B.2 : relation between wx(x,y,z) and wz(x,y)
Based on the latter relations, the strains may be expressed as,
 ∂w x   ∂2 
   2 
εx   ∂x   ∂x 
   ∂w y   ∂2 
εy  =  ∂y  = − z.  ∂y 2  w z = − z.[ Lb ]w z . (B.4)
γ xy     
   ∂w x ∂w y   ∂2 
+ 2 
 ∂y ∂x   ∂x∂y 
300 Appendix B
Equations (B.1) and (B.4), together with Hooke’s law, relate the moments
mx, my and mxy to the normal middle surface displacement wz,
 mx 
 
 m y  = −[ Db ].[ Lb ]w z , (B.5)
m xy 
 
1 ν 0 
3  
[ Db ] =
Et
with ν 1 0 . (B.6)
12(1 − ν 2 )  1−ν 
 0 0 2 
dynamic equation for bending motion

The dynamic equation, which governs the bending motion in a thin, flat
plate, is obtained by expressing the dynamic equilibrium of an infinitesimal
part t.dx.dy of the plate. Figure B.3 shows the acting forces and moments
and the external load p, which is applied in the direction normal to the plate
surface.
Figure B.3 : forces and moments, acting on an infinitesimal part t.dx.dy

Assuming a harmonic time dependence with circular frequency ω, the

steady-state dynamic force equilibrium in the z-direction requires
∂q x ∂q y
p+ + + ρ s tω 2 w z = 0 . (B.7)
∂x ∂y
Neglecting the rotational inertia and all second-order terms, the rotational
dynamic equilibrium around the y-axis requires
∂m x ∂m yx
qx = + . (B.8)
∂x ∂y
In a similar way, the rotational equilibrium around the x-axis requires
∂m y ∂m xy
qy = + . (B.9)
∂y ∂x
Substituting (B.8) and (B.9) into (B.7) yields
 mx 
T 
p + ρ s tω w z + [ Lb ]
2
 my  = 0 . (B.10)
m xy 
 
Substituting (B.5) into (B.10) yields the steady-state dynamic equation for
the bending motion in a thin, flat plate,
p
∇ 4 w z ( x, y ) − kb4 . w z ( x, y ) = , (B.11)
D
where
∂4 ∂4 ∂4
∇ 4 = [ Lb ] [ Db ][ Lb ] =
T
+ 2 + , (B.12)
∂x 4 ∂x 2∂y 2 ∂y 4
and where the plate bending stiffness D and the bending wavenumber kb are
Et 3 ρ s tω 2
D= , kb = 4 . (B.13)
12(1 − ν 2 ) D
302 Appendix B
B.1.2. In-plane motion
stresses and strains

The stresses in an infinitesimal part t.dx.dy, induced by in-plane motion in a
flat plate, are shown in figure B.4(a). These stresses induce some resulting
forces per unit length, acting on a plate section,
t/2 t/2
tx = ∫ σ x . dz , ty = ∫ σ y . dz
−t / 2 −t / 2
(B.14)
t /2 t/2
t xy = ∫ τ xy . dz = ∫ τ yx . dz = t yx .
−t / 2 −t / 2
The positive orientations of these forces are shown in figure B.4(b).
Figure B.4 : stresses (a) and their resultants (b)

in an infinitesimal part t.dx.dy
Since the stresses are invariant in the z-direction, the strains and the
displacement components wx and wy in, respectively, the x- and y-direction
are also invariant in the z-direction. The relation between the strains and the
displacement components wx and wy is
∂ 
 0
 ε x   ∂x 
   ∂  w x  w x 
εy  =  0  
∂y  î w y 
[ ]
= Lp   .
î wy 
(B.15)
τ xy  
  ∂ ∂
 
 ∂y ∂x 
This equation, together with Hooke’s law, relate the forces tx, ty and txy to the
(middle surface) displacement components wx and wy,
 tx 
w x 
 
[ ]
 t y  = t[ D]. L p w 
î y
(B.16)
t xy 
 
1 ν 0 
E  
with [ D] = ν 1 0 . (B.17)
(1 − ν 2 )  1−ν 
 0 0 2 
dynamic equations for in-plane motion

The dynamic equations, which govern the in-plane motion in a thin, flat
plate, is obtained by expressing the dynamic equilibrium of an infinitesimal
part t.dx.dy of the plate. Figure B.5 shows the acting forces and the body
forces bx and by, which are applied in, respectively, the x- and y-direction.
Assuming a harmonic time dependence with circular frequency ω, the
steady-state dynamic force equilibrium in the x-direction requires
∂t x ∂t yx
t. b x + + + ρ s tω 2 w x = 0 . (B.18)
∂x ∂y
In a similar way, the steady-state dynamic force equilibrium in the y-

direction requires
∂t y ∂t xy
t. by + + + ρ s tω 2 w y = 0 . (B.19)
∂y ∂x
304 Appendix B
Figure B.5 : forces, acting on an infinitesimal part t.dx.dy
Substituting (B.16) into (B.18) and (B.19) yields the steady-state dynamic
equations for the in-plane motion in a thin, flat plate,
∂ 2w x 1 − ν ∂ 2w x 1 + ν ∂ w y 1 − ν 2
2
ρ s (1 − ν 2 )ω 2
+ + + b x + wx = 0 ,
∂x 2 2 ∂y 2 2 ∂x∂y E E
(B.20)
∂ wy 1 − ν ∂ wy 1 + ν ∂ wx 1 − ν
2 2 2 2
ρ ( 1 − ν )ω
2 2
+ + + by + s wy = 0 ,
∂y 2 2 ∂x 2 2 ∂x∂y E E
(B.21)
which may be written in matrix form,
bx  2 x  w x 
[ ] [ ]
w T
b  + ρ sω w  + L p [ D] L p w  = 0 . (B.22)
î y î y î y
B.2. Weighted residual formulation
The weighted residual formulation of the dynamic equation (B.11) for the
out-of-plane bending motion may be based on its originating equilibrium
equations (B.7), (B.8) and (B.9). The formulation defines the steady-state
normal displacement field wz(x,y) in a thin, flat plate Ωs as the displacement
field, for which the integral equation
~ ( p + ρ tω 2 w + ∂q x + ∂q y )dΩ
∫w z s z
∂x ∂y
Ωs
(B.23)
∂w~ ∂m ∂m xy ∂w~ ∂m y ∂m xy
+ ∫ z (q −
x
x − ) dΩ + ∫ z (q −
y − ) dΩ = 0
Ω
∂x ∂x ∂y Ω
∂y ∂y ∂x
s s
is satisfied for any weighting function w ~ , for which the function and its
z
first-order spatial derivatives are bounded and uniquely defined within the
plate surface Ωs and on its boundary Γs.
By rearranging some terms, the weighted residual formulation (B.23) may
be written as
~ q ) ∂ (w~ q )
~ ∂ (w z y
∫ w z ( p + ρ s tω w z ). dΩ +
∫ ). dΩ
2 z x + (
Ωs Ωs
∂ x ∂ y
(B.24)
− ∫ ( z( x + )+ z ( + )). dΩ = 0
Ω
∂x ∂x ∂y ∂y ∂y ∂x
s
According to the two-dimensional divergence theorem, the integral of the

normal component of a vector field φ , taken over a closed boundary Γs, is
equal to the integral of the divergence of the vector field, taken over the
surface Ωs, enclosed by the boundary Γs,
∫ (∇.φ ). dΩ = ∫ (φ. n ). dΓ , (B.25)

Ωs Γs
306

where n (n x ,n y ) is the unit normal vector on the boundary with positive
orientation away from the surface. Application of this theorem to the second
Appendix B
integral term in (B.24) yields
~ ~
∫ w z ( p + ρ s tω w z ). dΩ + ∫ w z (q x n x + q y n y ). dΓ
2
Ωs Γs
(B.26)
− ∫ ( z( x + )+ z ( + )). dΩ = 0
Ωs
∂x ∂x ∂y ∂y ∂y ∂x
By rearranging the terms in the third integral term, equation (B.26) may be
written as
~ ~
∫ w z ( p + ρ s tω w z ). dΩ + ∫ wz (q x n x + q y n y ). dΓ
2
Ωs Γs
∂w~ ~
∂
− ∫ ( (mx z ) + ∂ ( m ∂w z )). dΩ
Ωs
∂x ∂x ∂y y ∂y
∂w~ ∂w~ (B.27)
∂ ∂ z )). dΩ
− ∫ ( ( m xy z ) + ( m xy
Ω
∂x ∂y ∂y ∂x
s
~
∂ 2w 2~ 2~
z + m ∂ w z + 2m ∂ w z ). dΩ = 0
+ ∫ (mx
∂x 2
y
∂y 2
xy
∂x∂y
Ωs
Application of the divergence theorem to the third and fourth integral term
in (B.27) yields
~ ~
∫ w z ( p + ρ s tω w z ). dΩ + ∫ w z (q x n x + q y n y ). dΓ
2
Ωs Γs
∂w~ ∂w~
z ( m n + m n ). dΓ −
−∫ ∫ ∂y ( m y n y + m xy n x ). dΓ
z (B.28)
x x xy y
Γs
∂ x Γs
~
∂ 2w 2~ 2~
+ ∫ (mx z + m ∂ w z + 2m ∂ w z ). dΩ = 0
y xy
Ω ∂x 2 ∂y 2 ∂x∂y
s
The first-order derivatives in the boundary integral terms in (B.28),

expressed in the global co-ordinate system (x,y), may be expressed in the
local co-ordinate system (n,s), shown in figure B.6.
Figure B.6 : local co-ordinate system (n,s)
According to the chain-rule for derivatives, one may write
∂w~ ~ ~ ~ ~
z = ∂w z ∂n ∂w
+ z
∂s
= nx
∂w z − n ∂w z , (B.29)
y
∂x ∂n ∂x ∂s ∂x ∂n ∂s
∂w~ ∂ ~ ~
∂n ∂w ∂s ∂w~ ∂ ~
z = wz + z = ny z + nx
w z . (B.30)
∂y ∂n ∂y ∂s ∂y ∂n ∂s
Substituting the latter relations into (B.28) yields
~ ~
∫ w z ( p + ρ s tω w z ). dΩ + ∫ wz (q x n x + q y n y ). dΓ
2
Ωs Γs
∂w~
−∫ z ( m n 2 + m n 2 + 2m n n ). dΓ
x x y y xy x y
Γs
∂ n
∂w~ (B.31)
−∫ z ( m ( n 2 − n 2 ) + ( m − m )n n ). dΓ
xy x y y x x y
Γ
∂ s
s
~
∂ 2w 2~ 2~
z + m ∂ w z + 2m ∂ w z ). dΩ = 0
+ ∫ (mx
∂x 2
y
∂y 2
xy
∂x∂y
Ωs
308 Appendix B
The internal forces and moments qx, qy, mx, my and mxy along the boundary Γs
may be related to the external forces and moments qn, mn and mns, as shown
in figure B.7(a),
qn = q x n x + q y n y
mn = m x n x2 + m y n y2 + 2m xy n x n y (B.32)
mns = m xy (n x2 − n y2 ) + ( m y − m x )n x n y
Substituting the latter relations into (B.31) yields
~
∂ 2w 2~ 2~
~ z + m ∂ w z + 2m ∂ w z ). dΩ
∫ w z ( p + ρ s tω wz ). dΩ +
∫
2
(mx y xy
Ωs Ωs ∂x 2 ∂y 2 ∂x∂y
~ ~
+ ∫w~ q . dΓ − ∂w z m . dΓ − ∂w z m . dΓ = 0
z n ∫ ∂s ns ∫ ∂n n
Γs Γs Γs
(B.33)
Since the boundary Γs is a closed curve, it can be proven (see e.g.
REKTORYS (1977)) that1
~
∂w ~ ∂mns
∫ ∂s mns . dΓ = − ∫ w z ∂s . dΓ .
z (B.34)
Γs Γs
The torsional moment mns may be omitted and its effect may be replaced by
a shear force, which is added to the shear force qn, yielding a generalised
shear force (see figure B.7(b))
∂mns
Qn = qn + . (B.35)
∂s
1 ~ and m are
provided that the boundary Γs is sufficiently smooth and that w z ns
continuous along the boundary

Figure B.7 : (a) boundary forces and moments

(b) generalised shear force Qn
Substituting (B.34) and (B.35) into (B.33) yields the ‘weak form’ of the
weighted residual formulation of the dynamic equation, which governs the
steady-state normal displacement in a thin, flat plate,
~
∂ 2w 2~ 2~
z + m ∂ w z + 2m ∂ w z ). dΩ + ρ tω 2 w ~ w . dΩ
∫ (m x
∂x 2 y
∂ y 2 xy
∂x ∂y
∫ s z z
Ωs Ωs
(B.36)
∂w~
~ ~
+ ∫ w z p. dΩ + ∫ w z Qn . dΓ − ∫ z m . dΓ = 0
Ω Γ Γ
∂n n
s s s
or in matrix form
∫ (([ Lb ]wz ) [ Db ][ Lb ]wz ). dΩ + ∫ ρ s tω w z wz . dΩ

− ~ T 2~
Ωs Ωs
~ (B.37)
~ p. dΩ + ~ ∂w
+ ∫w ∫ w z Qn . dΓ − ∫ ∂ n m n . dΓ = 0
z
z
Ωs Γs Γs
310 Appendix B
The weighted residual formulation of the dynamic equations (B.18) and

(B.19) for the in-plane motion defines the steady-state displacement fields
wx(x,y) and wy(x,y) in a thin, flat plate Ωs as the displacement fields, for
which the integral equation
~ (t. b + ρ tω 2 w + ∂t x + ∂t yx ). dΩ
∫ w x x s x
∂x ∂y
Ωs
(B.38)
~ ( t . b + ρ tω 2 w + ∂t y ∂t xy
+ ∫ w y y s y + ). dΩ = 0
Ωs
∂y ∂x
~ , which are bounded

~ and w
is satisfied for any set of weighting functions w x y
and uniquely defined within the plate surface Ωs and on its boundary Γs.
By rearranging some terms, the weighted residual formulation (B.38) may
be written as
~ ~ (t. b + ρ tω 2 w )). dΩ
∫ (w x (t. bx + ρ s tω wx ) + w
2
y y s y
Ωs
~ t ) ∂ (w ~ t ) ~ t ) ∂ (w
∂ (w ~ t )
∂ (w y y y xy x xy
+ ∫ ( x x
+ ). dΩ + ∫ ( + ). dΩ (B.39)
Ωs
∂x ∂y Ωs
∂x ∂y
~ ∂w~ ~ ∂w~
∂w y ∂w y
− ∫ ( x
tx + ty + ( x
+ )t ). dΩ = 0
Ω
∂x ∂y ∂y ∂x xy
s
Application of the divergence theorem (B.25) to the second and third

integral term in (B.39) yields
~ ~ (t. b + ρ tω 2 w )). dΩ
∫ ( w x (t. bx + ρ s tω wx ) + w
2
y y s y
Ωs
~ (t n + t n ) + w
+ ∫ (w ~ (t n + t n )). dΓ (B.40)
x x x xy y y y y xy x
Γs
~ ~
∂w ~ ~
∂w
∂w y ∂w y
− ∫ ( x tx + ty + ( x + )t xy ). dΩ = 0
Ωs
∂x ∂y ∂y ∂x

Finite element method for uncoupled structural problems
The internal forces tx, ty and txy along the boundary Γs may be related to the
external force vector T(Tx ,Ty ) , as shown in figure B.8,
311
Tx = t x n x + t yx n y
(B.41)
Ty = t xy n x + t y n y
Figure B.8 : boundary forces
Substituting (B.41) into (B.40) yields the ‘weak form’ of the weighted
residual formulation of the dynamic equations, which govern the steady-
state in-plane displacement fields in a thin, flat plate,
~ ∂w~ ~ ∂w~
∂w y ∂w y ~ w +w~ w ). dΩ
− ∫ ( x tx + ty + ( x + )t xy ). dΩ + ∫ ρ s tω 2 ( w x x y y
Ωs
∂ x ∂ y ∂y ∂ x Ωs
~ ~ ~ ~
∫ t (w x bx + w y by ). dΩ + ∫ ( w x Tx + w y Ty ). dΓ = 0
Ωs Γs
(B.42)
or in matrix form
w~  ~ T w
w x  2 w x   x 
[ ] x T
î wy 
[ ]
− ∫ t ( L p  ~ ) [ D] L p  . dΩ + ∫ ρ s tω (  ~   ). dΩ
î wy  î wy  î wy 
Ωs Ωs
(B.43)
w~ T b ~ T T
w
x  x x  x
+ ∫ t (  ~   ). dΩ + ∫  ~   . dΓ = 0
w b w T
Ωs î y î y Γs î y  î y 
312 Appendix B
B.3. FE implementation for thin, flat plates

Since the dynamic equation (B.11) for the steady-state normal displacement
field in a thin, flat plate Ωs is a fourth-order equation, the displacement field
is uniquely defined, if two boundary conditions are specified at each point
on the plate boundary Γs. Three major types of boundary conditions may
occur ( Γ s = Γ s1 ∪ Γ s2 ∪ Γ s3 ):
w z = w z on Γ s 1 , (B.44)
∂w z
= −θ n on Γ s 1 , (B.45)
∂n
where w z and θ n are prescribed translation and rotation functions. Note

that boundary condition (B.44) automatically imposes also a prescribed
∂w z
value for the rotational component . These boundary conditions
∂s
describe, for instance, clamped boundaries.

These boundary conditions impose prescribed functions qn , mn and
mns for the shear forces and the bending and torsional moments along
the boundary. Since only two boundary conditions can be applied, the
prescribed shear force and torsional moment functions are combined,
according to (B.35), to a generalised shear force function Qn ,
∂mns
Qn = q n + on Γ s 2 , (B.46)
∂s
mn = mn on Γ s 2 . (B.47)
These boundary conditions describe, for instance, free boundaries.

• mixed boundary conditions :

At some parts of the plate boundary, one kinematic and one mechanical
boundary condition may be specified. For simply supported boundaries,
for instance, the translation wz and the bending moment mn have
prescribed values,
w z = w z on Γ s 3 , (B.48)
mn = mn on Γ s 3 . (B.49)
construction of a finite element model

The plate surface Ωs is discretized into a number of small subdomains Ωse
(‘finite elements’) and a number of nodes, say ne, are defined at some
particular locations in each element. Within each element, the distribution of
the field variable, i.e. the normal middle surface displacement wz, is
approximated as an expansion w z in terms of a number, say nw, of
prescribed shape functions, which are only defined within the considered
element domain Ωse. For the commonly used triangular and rectangular2
plate elements, the nodes are defined at each corner point of the element and
∂w ∂w
three degrees of freedom (w z ,θ x = − z ,θ y = − z ) are specified per

∂x ∂y
node (nw=3ne). Hence,
ne 3
w z ( x, y ) ≈ w z ( x, y ) = ∑ ∑ N nd
e
( x , y ). and ( x, y ) ∈ Ω se . (B.50)
n =1d =1
The contributions and in these expansions may be determined from the

weighted residual formulation (B.36).
e
Each shape function N nd is defined, such that it has a unit value for degree
of freedom d at node n and that all other nodal degrees of freedom are zero.
In this way, each contribution and in the displacement expansion (B.50)
represents the value of degree of freedom d at node n.
e
Based on the element shape functions N nd , which are locally defined in one
element Ωse, some global shape functions N nd may be constructed, which
2
and their mapped quadrilateral plate elements
314 Appendix B
are defined in the entire plate surface Ωs. In each element domain Ωse to
which node n belongs, the global shape function N nd is identical to the
e
corresponding element shape function N nd , while it is zero in all other

element domains. In this way, a global displacement expansion may be
defined as
ns 3
∑ ∑ N nd ( x, y). and = [ N ]. {wi }

w z ( x, y ) = ( x, y ) ∈ Ω s , (B.51)
n =1d =1
where ns is the total number of nodes in the plate discretization, [ N ] is a

(1x3ns) vector of global shape functions and {wi } is a (3nsx1) vector of
unknown nodal translational and rotational degrees of freedom,
{wi } = {wz 1 }
T
θ x 1 θ y1 w zn s θ xn s θ yn s . (B.52)

According to (B.5), the corresponding approximations for the moments mx,

my and mxy become then
 mx   mx 
   
 m y  ≈  m y  = −[ Db ].[ Lb ][ N ]{wi } = [ Db ].[ Bb ]{wi } , (B.53)
m xy   m xy 
   

with the (3x3ns) matrix

 ∂ 2N ∂ 2 N 12 ∂ 2 N13 ∂ 2 Nns 1 ∂ 2 Nns 2 ∂ 2 Nns 3 
 11 
 ∂x 2 ∂x 2 ∂x 2 ∂x 2 ∂x 2 ∂x 2 

 ∂ 2N ∂ 2 N 12 ∂ 2 N13 ∂ Nns 1 2
∂ Nns 2
2
∂ 2 Nns 3 
[ Bb ] = −  211 
∂y ∂y 2 ∂y 2 ∂y 2 ∂y 2 ∂y 2 
 2
 2 ∂ N 11 ∂ 2 N12 ∂ 2 N 13 ∂ 2 Nns 1 ∂ 2 Nns 2 ∂ 2 N n s 3 
2 2 2 2 2
 ∂x∂y ∂x∂y ∂x∂y ∂x∂y ∂x∂y ∂x∂y 

(B.54)
In a Galerkin weighted residual approach, the weighting function is

expanded in terms of the same set of shape functions as used for the field

variable expansion,
w z { ~i }
~ ( x, y ) = [ N ]. w ( x, y ) ∈ Ω s . (B.55)
The determination of the nodal degrees of freedom wi in the expansions

(B.51) and (B.53) is based on the substitution of the expansions into the
weak form (B.36) of the weighted residual formulation.
For the first integral term in (B.36), this substitution yields

~
∂ 2w ~
∂ 2w ∂ 2w~
∫ (mx z
+ m y
z
+ 2m xy
z
). dΩ =
Ωs ∂x 2
∂y 2 ∂ x ∂y
(B.56)
− ∫ (
 [ B ]{w
 b i} )
~ T . D . B w  . dΩ = − w
[ b ] [ b ]{ i } { ~i }T .[ K b ]. {wi },
Ωs
with the (3nsx3ns) structural stiffness matrix
[ K b ] = ∫ ([ Bb ]T .[ Db ].[ Bb ]). dΩ . (B.57)

Ωs
The second integral term in (B.36) becomes
Ωs
Ωs
∫ ( ρ s tω
2
2~
w z . w z ). dΩ =
( ~
)

∫ ( ρ s tω [ N ]{wi } .[ N ]{wi }). dΩ = ω {wi } .[ Mb ]. {wi },
T 2 ~ T (B.58)
with the (3nsx3ns) structural mass matrix
[ Mb ] = ∫ ( ρ s t.[ N ] T .[ N ]). dΩ . (B.59)

Ωs
316 Appendix B
These structural stiffness and mass matrices are sparsely populated, since
each global shape function N nd has only non-zero values in the domains of
those elements, to which node n belongs. As a result, the practical
calculation of these matrices may be based on an efficient assemblage
procedure of element stiffness and mass matrices, in a similar way as
already discussed in appendix A for acoustic finite element models.
The third integral term in (B.36) becomes
~
∫ (w z . p)dΩ = ∫ (
 [ N ]{w
  )
~ T . p dΩ = w
i} { ~i }T . {Pi } , (B.60)
Ωs Ωs
with the (3nsx1) external load vector
{Pi } = ∫ ([ N ] T p). dΩ . (B.61)

Ωs
The boundary integral terms in (B.36) allow the introduction of the

mechanical boundary conditions (B.46), (B.47) and (B.49),
∂w~
~
∫ w z Qn dΓ −∫ ∂ n m n dΓ =
z
Γs Γs
∂w~ ~
∂w
~ Q dΓ − z m dΓ + ~ Q dΓ −
∫ z n ∫ ∫ z n ∫ ∂n mn dΓ = (B.62)
w w z
n
Γ s2 Γ s2 +Γ s3
∂ n Γ s1 +Γ s3 Γ s1
i { ni } { ni }
{w~ } .( Q − M ) + {w~ } .({Q } − {M }),
T T
i ni ni
with the (3nsx1) vectors
{Qni } = ∫ ([ N ] )
Qn . dΓ ,
T
(B.63)
Γ s2
{Qni } = ∫ ([ N ] )
Qn . dΓ ,
T
(B.64)
Γ s1 + Γ s 3
{ Mni } = 
( 
)
T
∫  {n} [∂ ][ N ] mn  . dΓ ,
T
(B.65)
 
Γ s2 +Γ s3
{Mni } = ∫  ({n} T [∂ ][ N ]) 

T
m n  . dΓ , (B.66)

Γ s1
where the (2x1) vector {n} consists of the x- and y-components of the unit
vector, normal to the plate boundary and where the (2x1) vector [∂ ] consists
of the spatial gradient operators in the x- and y-direction. Note that, as it is
the case for the stiffness and mass matrices, only a few coefficients in the
vectors (B.63)-(B.66) are non-zero, due to the particular shapes of the global
shape functions.

By substituting (B.56), (B.58), (B.60) and (B.62) into (B.36), the weighted
residual formulation becomes
{w~i }T . ([ K b ] − ω 2 [ Mb ]). {wi } =

(B.67)
{w~i }T . ({Pi } + {Qni } − { Mni } + {Qni } − {Mni })

the weighting function, i.e. for any set of shape function contributions {wi}
~ ,

a set of 3ns equations in the 3ns unknown nodal translational and rotational
degrees of freedom wi is obtained,
([ Kb ] − ω 2[ Mb ]).{wi } = {Pi } + {Qni } − {Mni } + {Qni } − {Mni } . (B.68)
Each row in this matrix equation represents the weighted residual

formulation, in which a global shape function N nd is used as weighting
~ .
function w z

The kinematic boundary conditions (B.44), (B.45) and (B.48) are not yet
included in matrix equation (B.68). This is usually done by directly
assigning the prescribed values of the translational and/or rotational degrees
of freedom at each node location on the boundaries Γs1 and Γs3 to the
corresponding nodal unknowns wi . When this assignment is done for nw
nodal degrees of freedom, only n p ( = 3ns − nw ) nodal degrees of freedom
are still unknown. This means that nw equations should be eliminated in
matrix equation (B.68) to have a well-determined set of equations. This is
318 Appendix B
usually done by eliminating the rows in (B.68), which express the weighted
residual formulation, in which the global shape functions, associated with
the a priori assigned nodal degrees of freedom, are used as weighting
functions.
By eliminating the appropriate rows in (B.68) and by shifting all terms in
the left-hand side of (B.68), which contain the a priori assigned nodal
degrees of freedom, to the right-hand side vector, the finite element model
for the bending motion in a thin, flat plate is obtained,
([ Kb ] − ω 2[ Mb ]).{wi } = {Fbi } , (B.69)
where the (npx1) vector {wi } contains the remaining unknown nodal
translational and rotational degrees of freedom and where the stiffness and
mass matrices [ K b ] and [ Mb ] are now (npxnp) matrices. In discretizations
with the above mentioned rectangular and triangular plate elements, the
non-zero coefficients in {Qni } occur only in the eliminated equations of
(B.68). Due to the non-conformity of these discretizations, which will be
discussed later on, some non-zero coefficients in {Mni } occur also in non-
eliminated equations in (B.68). However, these non-zero coeffcients are
usually not taken into account in the right-hand side vector of the finite
element model (B.69). As a result, the (npx1) vector {Fbi } contains the
terms in the a priori assigned nodal degrees of freedom and the contributions
from the vector {Pi } , which results from the external load on the plate, and
from the vectors {Qni } and { Mni } , which result from the forces and
moments, applied on (parts of) the plate boundary.
In order to take into account the effect of the material damping in the plate,
a structural damping matrix [ Cb ] may be introduced in (B.69), yielding a
finite element model
([ Kb ] + jω[Cb ] − ω 2[ Mb ]).{wi } = {Fbi } . (B.70)
For mathematical convenience, a proportional or Rayleigh damping model

is often used, in which the damping matrix is expressed as a linear
combination of the stiffness and mass matrix,
[Cb ] = α [ K b ] + β[ Mb ] . (B.71)
A commonly used damping model, which yields a proportional damping

matrix, is the visco-elastic damping model, in which the damping effect is
modelled by defining a complex elasticity modulus,
E * = E ( 1 + jη ) , (B.72)
where η is the material loss factor.

The highest-order spatial derivatives of the displacement w z and the
weighting function w ~ in the weighted residual formulation (B.36) are of
z
second order. Hence, the convergence of a finite element discretization is
ensured, if the expansions (B.51) and (B.55) for, respectively, w z and w ~
z
consist of polynomial shape functions, which are complete up to at least the
second order within each element and if they are continuous along the
1
interelement boundaries (C -continuity).
The commonly used triangular and quadrilateral plate elements satisfy the
completeness condition, but yield no C1-continuity along their boundaries.
For instance, the rectangular parent element for quadrilateral plate elements,
has 4 nodes with a total of 12 degrees of freedom, as shown in figure B.9.
Figure B.9 : rectangular parent element

320 Appendix B
e
The 12 shape functions N nd of this parent element are
1
N ne1 = (1 + x n x )( 1 + yn y )( 2 + x n x + yn y − x 2 − y 2 )
8
1
N ne2 = x n (1 + x n x ) 2 (1 − x n x )( 1 + yn y ) ( n = 1..4) (B.73)
8
1
N ne3 = yn (1 + x n x )(1 + yn y ) 2 (1 − yn y )
8
From these definitions, it may be seen that the completeness condition is

indeed satisfied. As indicated in figure B.10, the displacement w z and its
∂w z
tangential derivative along a certain element boundary are only
∂s
determined by element shape functions, which are associated with the nodes
∂w z
on that boundary. However, the normal derivative along a certain
∂n
element boundary is not only determined by the element shape functions,
which are associated with the nodes on that boundary, but also by some
other shape functions. Hence, the C1-continuity is not ensured for this
element-type.
Figure B.10 : contributing shape functions

along boundary x=1 between node 2 and node 3
Despite the lack of C1-continuity, a good convergence is usually obtained

from a plate discretization with these rectangular elements. However, for
discretizations with quadrilateral plate elements, which result from a
parametric mapping3, the convergence is usually less good. For a more
3
a discussion of the parametric mapping concept is given in appendix A
detailed discussion of plate elements and their convergence properties, the

reader is referred to e.g. ZIENKIEWICZ and TAYLOR (1991).

Since the dynamic equations (B.20) and (B.21) for the steady-state in-plane
motion in a thin, flat plate Ωs are coupled second-order equations, the
displacement fields wx and wy are uniquely defined, if two boundary
conditions are specified at each point on the plate boundary Γs. Two types of
boundary conditions may occur ( Γ s = Γ s4 ∪ Γ s5 ):
w x = w x on Γ s 4 , (B.74)
w y = w y on Γ s 4 , (B.75)
where w x and w y are prescribed displacement functions.
Tx = Tx on Γ s 5 , (B.76)
Ty = Ty on Γ s 5 , (B.77)
where Tx and Ty are prescribed boundary force functions.
construction of a finite element model

The plate surface Ωs is discretized into a number of small subdomains Ωse
(‘finite elements’) and a number of nodes, say ne, are defined at some
particular locations in each element. Within each element, the distributions
of the field variables, i.e. the displacement wx and wy, are approximated as
expansions w x and w y in terms of a number, say nw, of prescribed shape
functions, which are only defined within the considered element domain Ωse.
For the commonly used triangular and rectangular4 plate elements, the nodes
4
and their mapped quadrilateral plate elements
322 Appendix B

are defined at each corner point of the element and, for each field variable,
one degree of freedom is specified per node. Hence,

ne
w x ( x, y) ≈ w x ( x, y) = ∑ Nie ( x, y). a xi ( x , y ) ∈ Ω se , (B.78)
i=1
ne
w y ( x, y ) ≈ w y ( x, y ) = ∑ Nie ( x, y). a yi ( x, y ) ∈ Ω se . (B.79)
i=1
The contributions axi and ayi in these expansions may be determined from
the weighted residual formulation (B.42).
Each shape function N ie is defined, such that it has a unit value at the
location of node i and that it is zero at all other node locations. In this way,
the contributions axi and ayi in the displacement expansions (B.78) and (B.79)
represent the approximated displacement values in, respectively, the x- and
y-direction at the location of node i.
Based on the element shape functions N ie , which are locally defined in one
element Ωse, some global shape functions Ni may be constructed, which are
defined in the entire plate surface Ωs. In each element domain Ωse to which
node i belongs, the global shape function Ni is identical to the
!! "
corresponding element shape function N ie , while it is zero in all other
element domains. In this way, global displacement expansions may be
defined as
w x ( x , y )  n s w xi 
w ( x , y ) = ∑ Ni ( x, y ). w  = [ N ]. {wi } ( x, y ) ∈ Ω s , (B.80)
î y  i=1 î yi 
##
where ns is the total number of nodes in the plate discretization, [ N ] is a
(2xns) matrix of global shape functions,
$
 N1 0 N2 0 N ns 0 
[ N] =  0 N1 0 N2 0 N n s 
, (B.81)

and {wi } is a (2nsx1) vector of unknown nodal degrees of freedom,

%&%'%(%)% + *
{wi } = {w x 1
% %
Finite element method for uncoupled structural problems
w y1 wx 2 wy2 w xn s w yn s }
T
.
323
(B.82)
,, - .
According to (B.16), the corresponding approximations for the forces tx, ty
,
and txy become then
 tx   tx 
   
[ ] [ ]
 t y  ≈  t y  = t[ D]. L p [ N ]{wi } = t[ D]. B p {wi } , (B.83)
t xy  t xy 
   
//
with the (3x2ns) matrix
 ∂N 1 ∂N 2 ∂N n s 
 0 0 0 
/
 ∂x ∂x ∂x 
∂N 1 ∂N 2 ∂N n s 
[ ] 
Bp =  0
∂y
0
∂y
0
∂y 
. (B.84)
 
 ∂N 1 ∂N 1 ∂N 2 ∂N 2 ∂N n s ∂N n s 
 ∂y ∂x ∂y ∂x ∂y ∂x 
In a Galerkin weighted residual approach, the weighting functions are

expanded in terms of the same set of shape functions as used for the field
variable expansions,
w~ ( x , y )
~ ( x , y ) = [ N ]. {wi }
x ~
w ( x, y ) ∈ Ω s . (B.85)
î y 
The determination of the nodal degrees of freedom in the expansions (B.80)

and (B.83) is based on the substitution of the expansions into the weak form
(B.42) of the weighted residual formulation.
For the first integral term in (B.42), this substitution yields
0 1
~ ~
∂w ~ ~
∂w
∂w y ∂w y
− ∫ ( x tx + ty + ( x + )t xy ). dΩ =
Ωs
∂x ∂y ∂y ∂x
(B.86)
− ∫
Ωs
([ ] )

 i} [ ]
~ T .[ D]. B w  . dΩ = − w
t  B p {w p { i } { ~i }T . K p . {wi } [ ]
324 Appendix B
with the (2nsx2ns) structural stiffness matrix
[K p ] = ∫ [ ] [ ]
 t B T . D . B  . dΩ .
 p [ ] p 
(B.87)
Ωs
The second integral term in (B.42) becomes
2 3
~ w +w~ w ). dΩ =
∫ ρ s tω
2
(w x x y y
Ωs
(B.88)
∫ ρ s tω 2  ([ N ]{wi })
~ T .[ N ] w  . dΩ = ω 2 w
{ i } [ ]
{ ~i }T . M p .{wi },
Ωs
with the (2nsx2ns) structural mass matrix
[ M p ] = ∫ ( ρ st.[ N ]T .[ N ]). dΩ . (B.89)

Ωs
As it is the case for the finite element model for the out-of-plane bending
motion, these stiffness and mass matrices are sparsely populated and their
practical calculation is based on an assemblage of element stiffness and
mass matrices.
The third integral term in (B.42) becomes
~ ~
∫ t (w x bx + w y by ). dΩ =
Ωs
(B.90)
 ~ T . bx  . dΩ = w
∫ 
t (
 [ N ]{w i} )
b  { ~i }T . {Bi },
Ωs  î y 
with the (2nsx1) body force vector
 b 
{Bi } = t ∫ [ N ] T . b  . dΩ
x
. (B.91)

Ωs îy 
The boundary integral term in (B.42) allows the introduction of the

mechanical boundary conditions (B.76) and (B.77),
~ ~
∫ (w x Tx + w y Ty ). dΓ =
Γs
(B.92)
~ ~
∫ ( w x Tx + w y Ty ). dΓ +
~ ~ ~ T
{ }
∫ ( w x Tx + w y Ty ). dΓ = {wi } .( Ti + {Ti })
Γ s5 Γ s4
with the (2nsx1) vectors
 T 
{Ti } = ∫  [ N ] . T  . dΓ
T x
, (B.93)
Γ s5  î y 
 T 
{Ti } = ∫ [ N ] T . T  . dΓ
x
, (B.94)

Γ s4 y 
î 
In plate discretizations with triangular and rectangular elements, only the

global shape functions Ni , which are associated with nodes, located on the
boundary Γs, have non-zero values along the boundary. Hence, only the
global shape functions, which are associated with nodes, located on Γs4 or on
Γs5, yield the non-zero coefficients in the sparsely populated vectors {Ti }
{ }
and Ti , respectively.
4
By substituting (B.86), (B.88), (B.90) and (B.92) into (B.42), the weighted
residual formulation becomes
{w~i }T . ([ K p ] − ω 2 [ M p ]).{wi } = {w~i }T . ({Bi } + {Ti } + {Ti }) . (B.95)
5
the weighting functions, i.e. for any set of shape function contributions
6
{w~i } , a set of 2ns equations in the 2ns unknown nodal translational degrees
of freedom wi is obtained,
([ K p ] − ω 2[ M p ]).{wi } = {Bi } + {Ti } + {Ti } . (B.96)

326 Appendix B
Each row in this matrix equation represents the weighted residual

formulation, in which a global shape function Ni is used as weighting
7
~ is zero, or vice versa.
~ , while the weighting function w
function w x y
The kinematic boundary conditions (B.74) and (B.75) are not yet included in
matrix equation (B.96). This is usually done by directly assigning the
prescribed values of the translational degrees of freedom at each node
location on the boundary Γs4 to the corresponding nodal unknowns wi .
When this assignment is done for nw nodal degrees of freedom, only
n p ( = 2ns − nw ) nodal degrees of freedom are still unknown. This means
that nw equations should be eliminated in matrix equation (B.96) to have a
well-determined set of equations. This is usually done by eliminating the
rows in (B.96), which express the weighted residual formulation, in which
the global shape functions, associated with the nodes on boundary Γs4, are
used as weighting functions.
By eliminating the appropriate rows in (B.96) and by shifting all terms in
the left-hand side of (B.96), which contain the a priori assigned nodal
degrees of freedom, to the right-hand side vector, the finite element model
for the in-plane motion in a thin, flat plate is obtained,
([ K p ] − ω 2[ M p ]).{wi } = {Fpi } , (B.97)
where the (npx1) vector {wi } contains the remaining unknown nodal
degrees of freedom and where the stiffness and mass matrices K p and [ ]
[M p ] are now (npxnp) matrices. In discretizations with rectangular and
triangular plate elements, the non-zero coefficients in {Ti } occur only in the
{ } contains
eliminated equations of (B.96). As a result, the (npx1) vector Fpi
the terms in the a priori assigned nodal degrees of freedom and the
contributions from the vector {Bi } , which results from the body forces in
{ }
the plate, and from the vector Ti , which result from the forces, applied on
(parts of) the plate boundary.
In a similar way as for the finite element model for out-of-plane bending
[ ]
motion, a structural damping matrix C p may be introduced in (B.97), in
order to take into account the effect of the material damping in the plate.
This yields the finite element model
([ K p ] + jω[C p ] − ω 2[ M p ]).{wi } = {Fpi } . (B.98)

The highest-order spatial derivatives of the displacements w x and w y and
the weighting functions w ~ and w ~ in the weighted residual formulation
x y
(B.42) are of first order. Hence, the convergence of a finite element
discretization is ensured, if the expansions (B.80) and (B.85) consist of
polynomial shape functions, which are complete up to at least the first order
within each element, and if they are continuous along the interelement
boundaries (C0-continuity).
Since these convergence conditions are the same as for acoustic finite
element discretizations (see appendix A), the shape functions in the
commonly used triangular and quadrilateral plate elements are the same as
in the triangular and quadrilateral fluid elements. Hence, the convergence of
discretizations with these elements is ensured.
B.4. FE implementation for thin, elastic shell structures

In finite element modelling, it is common practice to model a curved shell
structure as an assemblage of small, flat plate sections. For shell structures
with arbitrary shapes, only triangular plate elements can be used. For single-
curved structures, such as cylinders, a good shell discretization may also be
obtained with flat elements of rectangular or quadrilateral shape.
plate element model for shell modelling purposes

Due to the curved geometry of thin, elastic shell structures, the in-plane and
out-of-plane motions are no longer decoupled. Therefore, the plate element
models for in-plane and out-of-plane motion are combined into one plate
element model for each flat plate section in the shell discretization.
In view of their assemblage into one global shell model, two modifications
are made to the plate element models, compared with their use for flat
plates:
328 Appendix B
• co-ordinate transformation :
Since each plate element model is expressed in its local element co-
ordinate system (xe,ye,ze), in which the plate element is located in the
(xe,ye)-plane, the plate element models must be transformed to one global
co-ordinate system (x,y,z). As shown in figure B.11, the displacement
components of the middle surface of a plate section in each direction of
the local co-ordinate system have the following relation with the
displacement components in the directions of the glocal co-ordinate
system:
w e  w x 
 x 
[ ]
w y e  = R
w e 
e  
. w y 
w 
(B.99)
898:8
î z  î z
T
s1ex s 2e x n ex 
[ Re ] = {s1e }  
T
with s 2e ne =  s1ey s 2e y n ye  (B.100)
s e s 2e z n ze 
 1z
Figure B.11: local element and global co-ordinate systems
• additional nodal degree of freedom :

An additional out-of-plane rotational degree of freedom θ z e is usually
introduced for each node in a plate element in order to facilitate the
assemblage of the plate element models into a global shell model. In this
way, each of the ne nodes in a plate element has 6 degrees of freedom and
;
the approximations for the middle surface displacement components in
each plate section become
w e 
 x 
[ ]{ }
<
e e
w y e  ≈ N se . w se . (B.101)
w e 
î z 
=
The (6nex1) vector {wsee } contains the nodal degrees of freedom,
= =>
expressed in the local element co-ordinate system,
 ase
e
,1

{ }  
? ?A? ?
?B?
? {@
e
w se =   (B.102)
a e 
î se, n e 
with
{asee ,n } = }
T
wx en wye n wz e n θ x e n θ y e n θ z e n ( n: 1.. ne ) (B.103)
The (3x6ne) matrix [ Nsee ] contains the shape functions, used in the
C
element models for the in-plane and the out-of-plane motion of a flat
plate (see (B.50), (B.78) and (B.79)),
[ Nsee ] = [ Nsee,1 e
N se , ne ] (B.104)
 N ne 0 0 0 0 0
with [ e
N se ] 
,n =  0
 0
N ne 0 0 0

0 (n: 1.. ne ) . (B.105)
0 N ne1 N ne2 N ne3 0
 
Note that the last column in the latter matrices (B.105) is the zero vector,
indicating that the additional degrees of freedom have no influence on
the displacement predictions.
330 Appendix B
D E
Based on the co-ordinate transformation (B.99), the nodal degrees of
freedom in each plate element may be transformed to the global co-
ordinate system,
{wsee } = [T e ].{wse } , (B.106)
where the (6nex6ne) transformation matrix T e [ ] is a block diagonal

matrix,
[ ]
 Re

0
T =
e

0
Re
F 


 ,

(B.107)
G
e
 R 0
 R e 
 0
H
and where the (6nex1) vector {wse } contains the nodal degrees of
H HI
freedom, expressed in the global co-ordinate system,
 ase,1 
{ }  
J JKJ'J JJJ
w se =   (B.108)
a e 
î s, n e 
{ } { }
T
with ase, n = w xn w yn w zn θ xn θ yn θ zn ( n: 1.. ne ) . (B.109)
Combining (B.99), (B.101) and (B.106) yields the approximations for the
middle surface displacement components in each plate section, expressed
L
in the global co-ordinate system,
w x 
 
w 
[ ]{ }
w y  ≈ N s . w s ,
e e
(B.110)
î z
where the (3x6ne) matrix N se is [ ]

[ Nse ] = [ Re ] .[ Nsee ][. T e ]
T
. (B.111)
Each of the 18ne components in this matrix is a shape function, which is

defined within the domain of one plate element in the shell discretization
and which is associated with one of the 6ne nodal degrees of freedom for
the displacement approximation in one of the 3 global directions x, y or z.
By combining all the element shape functions, which are associated with
a certain nodal degree of freedom for a certain global displacement
direction and which are defined in each plate element, to which the
particular node belongs, some globally defined shape functions are
M
obtained, yielding
w x 
N
 
w y  ≈ [ N s ]. {w s } , (B.112)
w 
î z
where the (6nsx1) vector {w s } contains all degrees of freedom,
O
expressed in the global co-ordinate system, of the ns nodes in the shell
discretization and where the (3x6ns) matrix [ N s ] contains all globally
defined shape functions.
The determination of the unknown nodal degrees of freedom {w s } is based

on the weighted residual formulations of the dynamic equations for both the
out-of-plane and the in-plane motion in each plate section in the shell
discretization.
For the out-of-plane motion in one plate section Ωse with boundary Γ se ,
P
the weighted residual formulation (B.36) is used, in which the expansions
for the out-of-plane displacement and the weighting function become now
{ } .[ N s ]. {w s }, { }
[ s ] { ~s } .
T ~ = ne T . N . w
wz e = n e w ze
(B.113)
This yields a plate element model for the out-of-plane bending motion
332
e
sb
2 e
sb Q
([ K ] − ω [ M ]).{w } = {P } + {F } + {F } .
s s
e e
sb
e
sb
Appendix B
(B.114)
e
The (6nsx6ns) stiffness and mass matrices K sb e
and M sb [ ] [ ] in this model
are
[ Ksbe ] = Ω∫ ([ Bsb ]T .[ Db ].[ Bsb ]). dΩ { }

with [ Bsb ] = −[ Lb ] n e [ N s ] ,(B.115)
T
se
[ Msbe ] = ρ st Ω∫ 
[ Ns ] n . n

T e
{ }{ }
e T
[ N s ] . dΩ . (B.116)
se
The (6nsx1) generalised force vector {Pse } results from the external load
distribution p on the considered plate section,
{Pse } = ∫ ([ Ns ]T {n e }. p). dΩ . (B.117)

Ω se
The (6nsx1) generalised force vector {Fsbe } results from the prescribed
forces and moments, applied on the part of the plate element boundary Γ se ,
that belongs to the parts of the shell boundary with mechanical boundary
conditions,
{Fsbe } = Γ ∫ ([ N ] {n }. Q ). dΓ
s
T e
n
se ∩ Γ s 2
(B.118)
 
{ }
T

− ∫ 
 {n} T
[∂ ] n e T
[ N ]
s  . mn  . dΓ

Γ se ∩ ( Γ s 1 ∪ Γ s 3 )  
e
The (6nsx1) generalised force vector Fsb { } results from the internal forces
and moments along the plate element boundary,
{Fsbe } = Γ ∫
\Γ s2
([ N ] {n }.Q ). dΓ
s
T e
n
se
(B.119)
 
{ }
T
  {n} T [∂ ] n e [ N s ] . mn  . dΓ
T
− ∫  
Γ se \(Γ s 1 ∪ Γ s 3 )  
R
In a similar way, the weighted residual formulation (B.42) yields a plate
element model for the in-plane motion,
([ K ] − ω [ M ]).{w } = {B } + {F } + {F } .
e
sp
2 e
sp s
e
s
e
sp
e
sp (B.120)
The total model for each plate element in the shell discretization is obtained
S
from a simple addition of the out-of-plane element model (B.114) and the
in-plane element model (B.120),
([ K ] − ω [ M ]).{w } = {P } + {B } + {F } + {F },
e
s
2 e
s s s
e e
s s
e
s
e
(B.121)
with
[ Kse ] = [K sbe ] + [ Kspe ], [ Mse ] = [ Msbe ] + [ Mspe ], (B.122)
{Fse } = {Fsbe } + {Fspe }, {Fse } = {Fsbe } + {Fspe }.
global finite element model
The global finite element model for a shell discretization, which consists of
T
nse plate elements, is obtained from a simple addition of the different plate
element models (B.121),
([ Ks ] − ω 2[ Ms ]).{ws } = {Fs }, (B.123)
[ ] [ ][ ] [ M se ],
n se n se
Ks = ∑ K se , M s = ∑
e=1 e=1
with (B.124)
{Fs } = e∑= 1{ } + e∑= 1{ } + e∑= 1{ }.
n se n se n se
Pse Bse Fse
334 Appendix B
Note that the internal nodal forces and moments vanish in the generalised
force vector { Fs } , since the equilibrium between the different plate
elements requires
∑ {Fse } = {0} .
n se
(B.125)
e=1
In a similar way as for flat plates, the kinematic boundary conditions of the
shell are taken into account by a priori assigning the appropriate values to
the corresponding nodal degrees of freedom. This yields, together with the
introduction of a (proportional) damping matrix, the finite element model
for a thin, elastic shell model,
([ Ks ] + jω[Cs ] − ω 2[ Ms ]).{wi } = {Fsi } , (B.126)
where the vector {wi } contains all unconstrained, i.e. not a priori known,
nodal degrees of freedom in the shell discretization and where {Fsi }
contains the nodal forces and moments, resulting from the external
excitations on the shell and from the forces and moments, applied on the
shell boundaries. Note that, when all plate elements, meeting at a certain
node in the shell discretization, are co-planar, the matrix equation (B.126)
becomes singular, due to the additional out-of-plane rotational degrees of
freedom in that node. To eliminate this singularity, some dedicated
numerical treatments must be introduced. A commonly used treatment
consists of assembling the equations, associated with nodes where elements
are co-planar, in local co-ordinates, which allows to eliminate the
singularities and the associated out-of-plane rotational degrees of freedom.
For a detailed discussion of some alternative treatments, as well as for a
discussion of some special plate elements for axisymmetric shells and some
curved shell elements, the reader is referred to e.g. ZIENKIEWICZ and
TAYLOR (1991).
Appendix C. DIRECT BOUNDARY ELEMENT
METHOD FOR UNCOUPLED ACOUSTIC
PROBLEMS
C.1. Direct boundary integral formulation

Green’s kernel function
The Green’s kernel function
− jk r − r
a
e
G( r , ra ) = (C.1)
4π r − ra
is a function, which satisfies the inhomogeneous Helmholtz equation
∇ 2 G( r , ra ) + k 2 . G( r , ra ) = −δ ( r − ra ) (C.2)
and the Sommerfeld radiation condition
∂G( r , ra )
lim r - ra .( + jkG( r , ra )) = 0 . (C.3)
r - ra →∞ ∂ r - ra
Hence, the Green’s kernel function G(r,ra) represents the free-field pressure
at any position r due to an acoustic point source at position ra. Note that this
336 Appendix C
function becomes singular when the distance r − ra between the field

position and the source position becomes zero.
Green’s second identity

For any two functions ϕ and ψ, which are sufficiently smooth and non-
singular in a domain V, enclosed by a surface Ωs, Green’s second identity
states that (see e.g. GREENBERG (1998))
∂ψ ∂ϕ
∫ (ϕ . ∂n − ψ . ∂n ). dΩ = ∫ (ϕ . ∇ ψ − ψ . ∇ ϕ ). dV ,
2 2
(C.4)
Ω s V
where the direction n, normal to the boundary surface Ωs, has a positive
orientation away from domain V.
direct boundary integral formulation for exterior problems

Consider a closed surface Ωa and two spheres ΩR1 and ΩR2 with radius R1 and
R2 and with centre point at a certain position r, as shown in figure C.1.
Figure C.1
At any position ra in the domain V, bounded by Ωa, ΩR1 and ΩR2 (shaded
region in figure C.1), the Green’s kernel function G(r,ra) is non-singular,
since the point at position r is not located in V, and the function satisfies the
homogeneous Helmholtz equation,
Direct boundary element method for uncoupled acoustic problems 337
∇ 2 G( r , ra ) + k 2 . G( r , ra ) = 0 , ( ra ∈ V , r ∉ V ). (C.5)
The application of the Green’s second identity (C.4) with the above defined
Green’s kernel function as ψ and with a function p, which satisfies the
homogenous Helmholtz equation and the Sommerfeld radiation condition,
∇ 2 p( ra ) + k 2 . p( ra ) = 0 , ra ∈ V , (C.6)
∂p( ra )
lim ra .( + jkp( ra )) = 0 , (C.7)
ra →∞ ∂ ra
as ϕ, yields
∂G( r , ra ) ∂p( ra )
∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra ) = 0 . (C.8)
Ω a + Ω R1 + Ω R2
∂n ∂n
On the spherical surface ΩR1,, the distance r − ra between position r and

any position ra on this sphere is equal to the sphere radius R1 and the normal
direction n, shown in figure C.1, is equal to (-R1). By using a spherical co-
ordinate system (R1,φ,θ), the part of the surface integral in (C.8), taken over
the spherical surface ΩR1,, may be written as
∂G( r , ra ) ∂p( ra )
∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra ) =
Ω R1
∂n ∂n
2π  π  
 ∂ e − jkR1 e − jkR1 ∂p( R1 , φ ,θ )  2
∫ ∫ − p ( R1 , φ , θ ). ( ) + ( ) . R1 sin θ . dθ  . dφ =
0  0 
∂R1 4πR1 4πR1 ∂R1  
2π  π 
e − jkR1
R e − jkR1 ∂p( R1 , φ ,θ )  
∫  ∫  p( R1 , φ ,θ ).(1 + jkR1 ). 4π + 1
4π
.
∂R1
.sin θ . dθ  . dφ
0  0   
(C.9)
Since, in the limit R1→0,
p( ra ) = p( R1 , φ , θ ) → p( r ) , (C.10)
338 Appendix C
expression (C.9) yields
 
∂G( r , ra ) ∂p( ra )
lim  ∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra )
R1 → 0 Ω ∂n ∂n
 R1 
 2π  π sin θ  
= p( r ).  ∫  ∫ dθ  . dφ  (C.11)
  4π  
0 0
= p( r ).
On the spherical surface ΩR2,, the distance r − ra between position r and

any position ra on this sphere is equal to the sphere radius R2 and the normal
direction n, shown in figure C.1, is equal to R2. By using a spherical co-
ordinate system (R2,φ,θ), the part of the surface integral in (C.8), taken over
the spherical surface ΩR2,, may be written as
∂G( r , ra ) ∂p( ra )
∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra ) =
Ω R2
∂n ∂n
2π  π  
 ∂ e − jkR2 e − jkR2 ∂p( R2 , φ ,θ )  2  . dφ =
∫ ∫  p ( R2 , φ , θ ). ( ) − ( )  . R2 sin θ . d θ 
0  0 
∂ R2 4πR2 4 π R2 ∂ R2  
2π  π   e − jkR2 
∂p( R2 , φ , θ )
− ∫  ∫  p( R2 , φ ,θ ) + R2 ( ∂R2
+ jkp( R2 , φ ,θ )) .sin θ . dθ  . dφ
0 0  4π 
(C.12)
Since, according to assumption (C.7), the function p(R2,φ,θ) satisfies the

Sommerfeld radiation condition, and assuming that p vanishes in the limit
for R2→∝, expression (C.12) yields
 
∂G( r , ra ) ∂p( ra )
lim  ∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra ) = 0. (C.13)
R2 →∞ Ω ∂n ∂n
 R2 
Substitution of expressions (C.11) and (C.13) into expression (C.8) yields

∂G( r , ra ) ∂p( ra )
∫ ( p( ra ). − G( r , ra ). ). dΩ ( ra ) = − p( r ). (C.14)
Ωa
∂n ∂n
By defining the direction ν, normal to the closed boundary surface Ωa, with
a positive orientation into the unbounded domain V (ν=-n), the direct
boundary integral formulation for exterior problems is obtained. This
formulation states that, for any pressure field, which satisfies the
homogeneous Helmholtz equation (C.6) and the Sommerfeld radiation
condition (C.7), the pressure at any point r in the unbounded domain V\Ωa is
related to the pressure distribution p(ra) and the normal velocity distribution
vν(ra) on the closed surface Ωa,
∂G( r , ra )
p( r ) = ∫ ( p( ra ). + jρ 0ωG( r , ra ). vν ( ra )). dΩ ( ra ). (C.15)
Ωa
∂ν
When the point at position r is located on the closed surface Ωa, as shown in
figure C.2,
Figure C.2
the surface ΩR1, becomes a hemisphere for R1→0. Provided that the surface
Ωa is smooth, i.e. that the normal direction is uniquely defined at any
position on the surface, expression (C.11) yields
340 Appendix C
 
 ∂G( r , ra ) ∂p( ra ) 
lim
R1 → 0 Ω
∫ ( p( ra ). ∂n − G( r , ra ). ∂n ). dΩ ( ra )
 R1 
 2π  π / 2 sin θ  
= p( r ).  ∫  ∫ dθ  . dφ  (C.16)
   
 0 0 4π
1
= p( r ).
2
Hence, the direct boundary integral formulation for any position r on Ωa

becomes
1 ∂G( r , ra )
p( r ) = ∫ ( p( ra ). + jρ 0ωG( r , ra ). vν ( ra )). dΩ ( ra ). (C.17)
2 Ω
∂ν
a
When the point at position r is not located in the domain V, as shown in

figure C.3,
Figure C.3
the surface ΩR1 is no longer part of the boundary surface of domain V, at

least for R1→0. Hence, the direct boundary integral formulation for any
position r outside V becomes
∂G( r , ra )
0= ∫ ( p( ra ). + jρ 0ωG( r , ra ). vν ( ra )). dΩ ( ra ). (C.18)
Ωa
∂ν
The formulations (C.15), (C.17) and (C.18) for smooth surfaces Ωa may be
combined into the general direct boundary integral formulation for exterior
acoustic problems,
∂G( r , ra )
C( r ). p( r ) = ∫ ( p( ra ). + jρ 0ωG( r , ra ). vν ( ra )). dΩ ( ra ) , (C.19)
Ωa
∂ν
with

1, r ∈ V \ Ωa ,

C( r ) = 0, r ∉ V , (C.20)
1
 , r ∈ Ωa .
î 2
For non-smooth surfaces Ωa, it can be proven that, for positions r at the
edges and corners on the surface, where the normal direction is not uniquely
defined, this coefficient is
 ∂ 
1  1
Ω 
 ∫ ∂ν r − ra
C( r ) = 1 + ( ). d ( r )
a  (C.21)
4π Ω a 
and represents the exterior solid angle, expressed as a fraction of 4π, of the
boundary surface Ωa at the surface position r.
direct boundary integral formulation for interior problems

For interior problems with a bounded domain V, a similar derivation applies
as for exterior problems, but only one spherical surface ΩR1 must be defined,
as shown in figure C.4. The same direct boundary integral formulation
(C.19) is obtained with the normal direction ν, having a positive orientation
into the bounded domain V.
342 Appendix C
Figure C.4
C.2. Boundary element implementation

problem definition
In a general uncoupled acoustic problem, the steady-state pressure field is
governed by the inhomogeneous Helmholtz equation
∇ 2 p + k 2 . p = − j ρ 0 ωq, (C.22)
where q is a distribution of external acoustic sources, applied in the acoustic

domain V. Three types of boundary conditions may occur on the boundary
surface Ωa ( Ωa = Ω p ∪ Ωv ∪ Ω Z ):
p= p on Ω p , (C.23)
v . n = vn on Ω v , (C.24)

v. n =
p
= Ap, on Ω Z , (C.25)
Z
where p , vn and Z (or A ) are prescribed pressure, normal velocity and

normal impedance (or normal admittance) functions, respectively. For
exterior problems, the pressure field must also satisfy the Sommerfeld
radiation condition (C.7) at infinity.
When a source distribution q is applied, the total pressure field p may be

regarded as a superposition of a homogeneous pressure field pa and an
inhomogeneous pressure field pb. The latter pressure field is the free-field
pressure due to the source distribution q, i.e. the pressure field in absence of
the boundary surface Ωa. An analytical solution for this pressure field pb
may be obtained, since the source distribution can be regarded as a
combination of acoustic point sources, for which an analytical Green’s
kernel functions of type (C.1) exist. The homogeneous pressure field pa is
then defined as the solution of the Helmholtz equation
∇ 2 pa + k 2 . pa = 0, (C.26)
with boundary conditions
p a = p − pb on Ω p , (C.27)
j ∂pa j ∂pb
= vn − on Ω v , (C.28)
ρ 0ω ∂ n ρ 0ω ∂n
j ∂pa ∂pb
pa = Z ( ( + )) − pb , on Ω Z . (C.29)
ρ 0 ω ∂n ∂n
For exterior problems, the pressure field pa must also satisfy the Sommerfeld
radiation condition. As a result, a numerical solution procedure is needed
only for the homogeneous subproblem. Therefore, the below discussion of
the boundary element method may be confined to homogeneous problems
without loss of generality.
direct BEM: two-step procedure

The direct boundary element method follows a two-step procedure, based on
the direct boundary integral formulation (C.19) for homogeneous interior or
exterior acoustic problems with a closed boundary surface Ωa. The first step
consists of determining the pressure and normal velocity distribution on the
boundary surface. In the second, post-processing step, the pressure field in
any point of the acoustic domain is calculated, according to (C.15).
For the determination of the pressure and normal velocity distribution on the
closed boundary surface Ωa, the surface is discretized into a number of small
subsurfaces Ωae (‘boundary elements’) and a number, say ne, of nodes are
defined at some particular locations in each element. Within each element,
344 Appendix C

the distributions of the two boundary variables, i.e the pressure p and the
normal velocity vn, are approximated as expansions p and vn in terms of a

number of prescribed shape functions, which are only defined within the
considered element domain Ωae. For the most commonly used types of
triangular and quadrilateral boundary elements, the nodes are defined at
each corner point of the element and, for each boundary variable, one degree
of freedom is specified per node. Hence,
ne
p( ra ) ≈ p( r a) = ∑ Nie (ra ). a pi ( ra ∈ Ω ae ) , (C.30)
i=1
ne
vn ( ra ) ≈ vn ( ra ) = ∑ Nie ( ra ). avi ( ra ∈ Ω ae ) . (C.31)
i=1
Each shape function N ie is defined, such that it has a unit value at the
location of node i and that it is zero at all other node locations. In this way,
the contributions api and avi in the expansions (C.30) and (C.31) represent,
respectively, the approximated pressure and normal velocity values at the
location of node i.
Based on the element shape functions N ie , which are locally defined within
one boundary element Ωae, some global shape functions Ni may be
constructed, which are defined in the entire boundary surface Ωa. In each
element domain Ωae to which node i belongs, the global shape function Ni
is identical to the corresponding element shape function N ie , while it is zero
in all other element domains. In this way, global boundary variable
expansions may be defined as

p( ra ) = [ N ]. { pi } ( ra ∈ Ω a ) ,

(C.32)
v n ( ra ) = [ ] { ni }
N. v ( ra ∈ Ω a ) , (C.33)
where na is the total number of nodes in the boundary surface discretization,

[ N ] is a (1xna) matrix of global shape functions and { pi } and {vni } are
(nax1) vectors of unknown nodal degrees of freedom.
The determination of the unknown nodal degrees of freedom is commonly

based on a collocational scheme. In this scheme, the direct boundary
integral formulation (C.19), using the boundary variable expansions (C.32)

and (C.33), is evaluated for positions r, which correspond with the locations
of the nodes in the boundary surface discretization. For the position of node
b, i.e. r=rb, this yields an equation

[ Ab ]{ pi } = jρ 0ω [ Bb ]{vni } . (C.34)
The coefficient Abi on column i in the (1xna) matrix [ Ab ] is
   
Abi = δ bi .  1 +
1 ∂ 1
Ω  −  N ( r ). ∂G( rb , ra ) . dΩ ( r )
 ∫ (
4π Ω ∂n rb − ra
). d ( ra   ∫
) i a
∂n a 
 a  Ω a 
(C.35)
where the normal direction n has a positive orientation into the acoustic
domain and where δbi is the Kronecker delta, which is 1 for b=i and 0 for
b≠i. The coefficient Bbi on column i in the (1xna) matrix [ Bb ] is
Bbi = ∫ Ni ( ra ). G( rb , ra ). dΩ ( ra ) . (C.36)
Ωa
By writing an equation of type (C.34) for each node b=1..na, the direct
boundary element model is obtained,

[ A]{ pi } = jρ 0ω [ B]{vni } , (C.37)
where the (1xna) matrices [ Ab ] and [ Bb ] (b=1..na) form the rows of the
(naxna) matrices [ A] and [ B] , respectively.
The boundary element model (C.37) is a set of na algebraic equations in the
2na nodal unknowns. However, since the boundary conditions (C.23), (C.24)
and (C.25) impose, respectively, a prescribed pressure, a prescribed normal
velocity or a prescribed impedance relation between both boundary

variables, at each node either the pressure value pi or the normal velocity
value vni or their impedance relation is known a priori. The remaining na
unknown nodal degrees of freedom result from the na equations in the
boundary element model (C.37).
346 Appendix C
In the second, post-processing step, the pressure at any position r in the

acoustic domain V, which is not located on the boundary surface Ωa, is
obtained from the direct boundary integral formulation (C.15), using the
surface results from the first step. In this way, the approximation for the
pressure at a certain position r∈V\Ωa is

p( r ) = [ C ]{ pi } + [ D]{vni }, (C.38)
where the coefficients Ci and Di on the column i in the (1xna) matrices [ C ]

and [ D] are
∂G( r , ra )
Ci = ∫ Ni ( ra ). . dΩ ( ra ) , (C.39)
Ωa
∂n
Di = jρ 0ω ∫ Ni ( ra ). G( r , ra ). dΩ ( ra ) . (C.40)
Ωa
convergence
Since the direct boundary integral formulation (C.19) contains no spatial
derivatives of the boundary pressure and boundary normal velocity, the
convergence of a boundary element discretization is ensured, even if the
expansions (C.32) and (C.33) are discontinuous along the interelement
boundaries (C-1-continuity).
As indicated e.g. by BREBBIA et al. (1984), a family of C-1-continuous
(‘discontinuous’) elements can be defined, for which the nodes are located
within the element and for which the element shape functions are based on
-1
Lagrangian polynomials. The most simple C -continuous elements are
constant elements, which have one node, defined at the centre of the
element, and one constant shape function.
Although the convergence of discretizations with constant elements is
ensured, the convergence rate may be rather slow. Therefore, 3-noded linear
triangular and 4-noded linear quadrilateral elements are commonly used
elements. The nodes of these elements are located at the corner points of the
element and their linear element shape functions allow a C0-continuous field
variable expansion (see appendix A). A commonly used tool to enhance the
convergence rate is the duplication of nodes at the edges and corners of the
boundary surface, where the normal direction to the boundary surface is not
uniquely defined (see figure C.5(a)), and at the intersecting curves between
parts of the boundary surface with different boundary conditions (see figure
C.5(b)). Instead of defining one node, which belongs to m different
elements, m different nodes are defined at the same location, but the global
shape functions, associated with each of these nodes, are only non-zero
within one of the m elements, as illustrated in figure C.6. This allows to take
into account the different boundary conditions, imposed at one location.
Figure C.5 : node duplication at edges and corners (a)

and at boundary condition discontinuities (b)
Figure C.6 : global shape functions for duplicated nodes
Note that the m rows in the boundary element model (C.37), each of them
expressing the direct boundary integral formulation for a position r,
348 Appendix C
corresponding with the location rb of one of the duplicated nodes, may be

written in the form

([ Am1 ] + [ Am 2 ]){ pi } = jρ 0ω[ Bm ]{vni } . (C.41)
The coefficients Bbi on row b and column i in the (mxna) matrices [ Bm ] are
defined by (C.36) and the coefficients Abi1 on row b and column i in the
(mxna) matrix [ Am1 ] are
∂G( rb , ra )
Abi 1 = − ∫ Ni ( ra ). . dΩ ( ra ) . (C.42)
Ωa
∂n
Since all duplicated nodes are located at the same position rb, all rows in
matrix [ Bm ] are identical and all rows in matrix [ Am1 ] are identical.
All coefficients in the (mxna) matrix [ Am2 ] are zero, except the ones, which
belong to the diagonal of the global model matrix [ A] and which all have an
identical value C,
0 0 C 0 0
0 
[ Am2 ] =  0 C
0 
0
(C.43)

0
0 C 0

0
1 ∂ 1
with C = 1+ ∫ ( ). dΩ ( ra ) . (C.44)
4π Ω ∂n rb − ra
a
Due to this particular form, each part of type (C.41) in the boundary element
model (C.37) ensures that the pressure degrees of freedom of all duplicated
nodes, defined at the same location, are identical.
C.3. BEM versus FEM for an interior acoustic problem

In comparison with a finite element model, a boundary element model is
substantially smaller, since only the boundary surface is discretized.
However, it can be seen from the equations (C.35) and (C.36), which define
the coefficients in the matrices [ A] and [ B] , that
• both matrices are fully populated, since all coefficients are non-zero,
• both matrices are non-symmetric, since Abi≠Aib and Bbi≠Bib,
• both matrices are complex and frequency dependent, since the Green’s
kernel function is a complex and frequency dependent function.
As a result, when comparing the computational loads, involved with a finite

element and a boundary element model, the boundary element method can
hardly compete with the finite element method for solving interior acoustic
problems, as illustrated in the following example.
On the sides of a cube with side length L, the following normal velocities
are imposed,
vn ( 0, y, z ) = 0, vn ( x,0, z ) = 0, vn ( x , y,0 ) = 0,
(C.45)
vn ( L, y, z ) = V , vn ( x , L, z ) = 2V , vn ( x, y, L ) = 3V .
The exact pressure field for this interior acoustic example is
jρ 0 cV ωx ωy ωz
pex ( x, y, z ) = (cos( ) + 2 cos( ) + 3 cos( )) . (C.46)
ωL c c c
sin( )
c
Three different FE models with 8-noded rectangular elements and three

corresponding BE models with 4-noded rectangular elements were built for
the case of an air-filled (ρ0=1.225 kg/m , c=340 m/s) cube with side length
3
L=1 m. All models were built with the SYSNOISE software and solved for
the case of V=1 m/s at frequency ω/2π=100 Hz. Table C.1 lists the number
of elements of the different models and the CPU times1, needed on a
Hewlett-Packard-C180 workstation (SPECfp95=18.7, SPECint95=11.8).
1
the CPU times for the BE models include the times, needed for constructing the
model matrices and for solving the model, while the CPU times for the FE
models include only the solution times; this is motivated by the fact that, for the
calculation of the pressure field over a wide frequency range, the stiffness and
mass matrices in an FE model are frequency independent and must be
constructed only once, while the frequency dependent BE matrices must be
constructed at each frequency
350 Appendix C
FEM BEM
n # elem. CPU (s) # elem. CPU (s)
10 1000 1 600 12
20 8000 65 2400 500
30 27000 1020 5400 4420
Table C.1 : number of elements and associated CPU times

(n=number of elements per side)
Figure C.7 plots the relative prediction errors for the pressure at
(x,y,z)=(L/2,L/2,L/2) (solid lines) and at (x,y,z)=(4L/5,4L/5,4L/5) (dashed
lines) against the associated CPU times for the various FE and BE models.
Figure C.7 : convergence rates with FEM (∇) and BEM (o)
This example clearly illustrates the usually much higher efficiency of the
finite element method for solving interior acoustic problems. The assets of
the boundary element method become mostly apparent for exterior acoustic
problems, involving unbounded acoustic domains.
Appendix D. COUPLED FE/FE AND COUPLED
FE/BE MODELS
D.1. Eulerian FE/FE model
D.1.1. Model description
In an Eulerian FE/FE model for an interior coupled vibro-acoustic system,

an acoustic FE model with nodal pressure degrees of freedom is used for the
prediction of the steady-state acoustic pressure in the fluid domain and a
structural FE model with nodal translational and rotational displacement
degrees of freedom is used for the prediction of the steady-state dynamic
displacement of the middle surface of the elastic shell structure.
acoustic FE model
As derived in appendix A, the finite element approximation of the steady-
state pressure p in the bounded fluid domain V of an interior uncoupled
acoustic system is an expansion p in terms of a set of global shape
functions Ni,
na np
p( x, y, z ) = ∑ Ni ( x, y, z ). pi + ∑ Ni ( x, y, z ). pi

i=1 i =1 (D.1)
[ ]
= [ Na ]. { pi } + N p . { pi }, ( x, y, z ) ∈V ,
352 Appendix D
where np is the number of constrained degrees of freedom, i.e. the prescribed

pressure values pi at the np nodes, which are located on the part Ωp of the
boundary surface, on which a prescribed pressure boundary condition is
imposed. The global shape functions, associated with these constrained
degrees of freedom, are comprised in the (1xnp) vector N p . The na [ ]
unconstrained degrees of freedom are comprised in the (nax1) vector { pi }
and their associated global shape functions are comprised in the (1xna)
vector [ N a ] . The resulting finite element model for the unconstrained
degrees of freedom takes the form,
([ Ka ] + jω[Ca ] − ω 2[ Ma ]).{ pi } = {Fa } . (D.2)
The (naxna) matrices [ K a ] , [ M a ] and [ Ca ] are the acoustic stiffness, mass

and damping matrices. The (nax1) vector {Fa } contains the terms in the
constrained degrees of freedom, the contributions from the external acoustic
sources in the fluid domain V and the contributions from the prescribed
velocity input, imposed on part Ωv of the boundary surface. The latter
contributions are expressed as (see (A.44))
∫ ( − jρ 0ω [ N a ]
T
vn ). dΩ , (D.3)
Ωv
where vn is the prescribed normal fluid velocity, with positive orientation

away from the fluid domain V.
structural FE model
As derived in appendix B, the finite element approximations of the steady-
state dynamic displacement components of the middle surface Ωs of an
elastic shell structure in the x-, y- and z-direction of a global cartesian co-
ordinate system are
 w x ( x , y, z ) 

 
w y ( x , y, z ) = [ N s ]. {wi } + [ N w ]. {wi } ,

(D.4)
 w ( x, y, z ) 

î z 
Coupled FE/FE and coupled FE/BE models 353
where the (3xnw) matrix [ N w ] comprises the global shape functions, which
are associated with the nw constrained degrees of freedom, i.e. the prescribed
translational and rotational displacements wi at nodes, which are located on
the part of the shell boundary, on which prescribed translational and/or
rotational displacements are imposed. The ns unconstrained translational and
rotational displacement degrees of freedom are comprised in the (nsx1)
vector {wi } and their associated global shape functions are comprised in
the (3xns) matrix [ N s ] . The resulting finite element model for the
unconstrained degrees of freedom takes the form,
([ Ks ] + jω[Cs ] − ω 2[ Ms ]).{wi } = {Fs } . (D.5)
The (nsxns) matrices [ K s ] , [ Ms ] and [ Cs ] are the structural stiffness, mass

and damping matrices. The (nsx1) vector { Fs } contains the terms in the
constrained degrees of freedom, the contributions from the prescribed forces
and moments, applied on (part of) the shell boundary and the contributions
from the external load p, applied normal to the shell surface Ωs. The latter
contributions are expressed as (see (B.117))
 
( { } )
n se
∑  ∫ [ s]
 
T
N . n e
. p . d Ω (D.6)
e = 1  Ω se


where nse is the number of flat plate elements Ωse in the shell discretization
and where the unit vector, normal to a plate element, is represented in the
{ }
(3x1) vector n e .
coupling of both models

The force loading of the acoustic pressure on the elastic shell structure along
the fluid-structure coupling interface in an interior coupled vibro-acoustic
system may be regarded as an additional normal load. As a result, an
additional term of type (D.6), using the acoustic pressure approximation
(D.1), must be added to the structural FE model (D.5). When it is assumed
that the elastic shell structure is completely comprised in the boundary
surface of the fluid domain, the structural FE model (D.5) modifies to
354 Appendix D
([ Ks ] + jω[Cs ] − ω 2[ Ms ]).{wi } + [ Kc ].{ pi } = {Fsi } . (D.7)
The (nsxna) coupling matrix [ K c ] is
n se  
[ c] ∑  ∫ [ s]
K = − 
e = 1  Ω se
N
T
.(n e
. N { }
[ a ] 
. dΩ

) (D.8)
and the (nsx1) excitation vector {Fsi } is
 
( { }[ ] )
n se
{Fsi } = {Fs } + ∑  ∫ [ Ns ] . n . N p { pi } . dΩ  .
 T e
(D.9)
e = 1  Ω se 
The continuity of the normal shell velocities and the normal fluid velocities
at the fluid-structure coupling interface may be regarded as an additional
velocity input on the part Ωs of the boundary surface of the acoustic domain.
As a result, an additional term of type (D.3), using the shell displacement
approximations (D.4), must be added to the acoustic FE model (D.2). This
modified acoustic FE model becomes
([ Ka ] + jω[Ca ] − ω 2[ Ma ]).{ pi } − ω 2[ Mc ]{wi } = {Fai } . (D.10)
The (naxns) coupling matrix [ Mc ] is
 
{ }
n se
 
[ Mc ] = ∑  ∫  0[ a ]
  ρ N
T
. n e T
. [ s ] 
N  . d Ω (D.11)
e = 1  Ω se 
and the (nax1) excitation vector {Fai } is
 
{ }
n se
{Fai } = {Fa } + ∑  ∫ ρ 0ω 2  [ N a ] . n e 
.[ N w ]{wi } . dΩ  .
T T
(D.12)
e = 1  Ω se
  

A comparison between the coupling matrices (D.8) and (D.11) indicates that
[ Mc ] = − ρ 0 [ K c ]T . (D.13)
Combining the modified structural FE model (D.7) and the modified

acoustic FE model (D.10) yields the Eulerian FE/FE model for an interior
coupled vibro-acoustic system,
 Ks Kc  Cs 0  Ms 0  wi   Fsi 

  + jω   −ω 2  .   =   . (D.14)
 0 Ka  0 Ca  − ρ 0 K c
T
M a  î pi  î Fai 
Note that for the practical calculation of this coupled model, it is convenient
that the acoustic and structural meshes are matching, i.e. that the nodes of
the acoustic and the structural meshes at the fluid-structure coupling
interface coincide. If not, the structural nodal displacement degrees of
freedom must be related to the acoustic nodal pressure degrees of freedom
along the fluid-structure coupling interface through some geometrical
transfer matrices, as proposed e.g. by COYETTE and DUBOIS-PÉLERIN
(1994).
D.1.2. Modal expansion

The predictions for the undamped mode shapes and natural frequencies of a
coupled vibro-acoustic system result from the following right eigenvalue
problem,
Ks Kc   Ms 0 
{Φ c } = ω c2  T
M a {Φ c } (c = 1.. ns + na ) , (D.15)
 1 1 
0 ρ0
Ka  − K c ρ0
   
where each ((ns+na)x1) right eigenvector {Φc } represents a mode shape and
where the associated eigenvalue corresponds with the squared value of the
natural frequency ωc of that mode.
Since, in contrast with uncoupled acoustic and uncoupled structural FE
models, the stiffness and mass matrices in a coupled Eulerian model are no
356 Appendix D
longer symmetric, the above eigenvalue problem is non-symmetric. Hence,

{ }
the left eigenvectors Φc of the associated left eigenvalue problem,
Ks Kc   Ms 0 
{Φ c } T
. 0

1
Ka  { }
 = ω c2 Φ c T .  T
− K c
1 
Ma  , (D.16)
 ρ0   ρ0 
differ from the right eigenvectors {Φc } . However, LUO and GEA (1997)
indicated that, due to the particular matrix relation (D.13), the components,
which correspond with the na acoustic degrees of freedom in each pair of
associated left and right eigenvectors, are identical and that the components,
which correspond with the ns structural degrees of freedom in the left
eigenvectors, are proportional to the corresponding right eigenvector
components with a factor, equal to the associated eigenvalue,
Φ sc  ω c2 .Φ sc 
{ c } Φ  =  Φ  (c = 1.. ns + na ) .
Φ = (D.17)
î ac  î ac 
The Eulerian model (D.14) may be transformed into a modal model by

expanding the nodal degrees of freedom in terms of a set of mc(≤ ns+na)
modes of the coupled vibro-acoustic system,
wi  m c Φ sc 
  = ∑ φ c.  = [Φ ]. {φ c } , (D.18)
î pi  c = 1 î Φ ac 
where [Φ ] is a ((ns+na)xmc) matrix of right eigenvectors and where {φc } is

a (mcx1) vector of modal participation factors. By substituting the modal
expansion (D.18) into the Eulerian model (D.14), in which all equations of
the acoustic part are divided by ρ0, and by premultiplying both sides of the
resulting matrix equation with the transpose of the ((ns+na)xmc) matrix of
corresponding left eigenvectors, the following modal model is obtained,
([ K~] + jω[C~] − ω [ M~ ]).{φ } = {F~} ,

2
c (D.19)
where the (mcx1) modal excitation vector is

 Fsi 
{} [ ]
~ T
F = Φ  1 F 

(D.20)
î ρ 0 ai

and where the (mcxmc) modal stiffness, mass and damping matrices are
Ks Kc   Ms 0 
[ ] [ ]
~
K = Φ 0
T

1 [Φ ], M
Ka 
~
[ ] [ ]
= Φ − K T
T
 c
1 
M a [Φ ],
 ρ0   ρ0 
. (D.21)
Cs 0 
[] [ ]
~
C = Φ 0
T

1
ρ0 

Ca [Φ ].

Due to the orthogonality of the left and right eigenvectors with respect to the
mass matrix,
 Ms 0 
{Φ c } { }
T
. − K T 1 
M a . Φ c 2 = 0, if c1 ≠ c 2 , (D.22)
1  c ρ0
 
the modal mass and modal stiffness matrices are diagonal matrices. By
normalising the eigenvectors, according to
 Ms 0 
{Φ c } { }
T
. − K T 1 
M a . Φ c 1 = 0, ( c1 = 1.. mc ), (D.23)
1  c ρ0
 
the modal mass matrix becomes the unity matrix,
1 
[ ]
~ 
M =  ,

(D.24)
 
 1
and the diagonal modal stiffness matrix becomes

358 Appendix D
ω 12 
~  
[ ]
K =

 . (D.25)
 2 
ω mc 

The efficiency of the model size reduction with this modal expansion
technique is illustrated for a one-dimensional example in section D.1.4.
The fact that the left and right eigenvector result from a non-symmetric
eigenvalue problem, puts, however, a severe practical limitation on the use
of the modal expansion technique for coupled vibro-acoustic problems. In
contrast with the symmetric eigenvalue problems for uncoupled structural
and uncoupled acoustic models, the non-symmetric eigenvalue problems for
coupled models are much more computationally demanding. To illustrate
this, a rectangular acoustic cavity with a length of 1m, a width of 0.5m and a
height of 0.15m was considered. One side of the cavity consists of a
flexible, rectangular plate of 1m x 0.5m with clamped boundaries, while the
five other cavity side walls are perfectly rigid. Three structural
discretizations were built with 4-noded rectangular plate elements
(CQUAD4), six acoustic discretizations were built with 8-noded rectangular
fluid elements (CHEXA8), yielding six discretizations for the coupled
system. Table D.1 lists the number of elements in the structural and acoustic
discretizations and the resulting numbers ns, na and nc of unconstrained
degrees of freedom of the different FE models.
structural acoustic coupled

nxxny ns nxxnyxnz na nc= ns+na
20x12 627 20x12x4 1365 1992
20x12x8 2457 3084
40x24 2691 40x24x4 5125 7816
40x24x8 9225 11916
60x36 6195 60x36x4 11285 17480
60x36x8 20313 26508
Table D.1 : numbers ns,na,nc of unconstrained degrees of freedom

(nx,ny,nz: number of elements in the length, width and height of the acoustic cavity)
The first 10 modes of the considered finite elements models have been
calculated with the MSC/NASTRAN software on a Hewlett-Packard-C180
workstation (SPECfp95=18.7, SPECint95=11.8). The involved CPU-times
are plotted against the corresponding number of unconstrained degrees of

freedom in figure D.1.
Figure D.1. : CPU time for mode extraction from the uncoupled structural (x),
uncoupled acoustic (o) and coupled vibro-acoustic (∇) models
This figure clearly illustrates that a non-symmetric eigenvalue problem,

involved with the mode extraction from coupled Eulerian FE/FE models,
requires a substantially larger amount of computational effort than the
symmetric eigenvalue problems, involved with the mode extraction from
uncoupled structural and uncoupled acoustic FE models. Note also that the
difference between the CPU times for the uncoupled structural and
uncoupled acoustic models illustrates that the computational effort is not
only determined by the model size, but also by the bandwidth of the sparsely
populated stiffness and mass matrices.
D.1.3. Component mode synthesis

A commonly used type of component mode synthesis for coupled vibro-
acoustic problems consists of expanding the structural nodal degrees of
freedom in terms of a set of ms modes of the uncoupled structural system,
i.e. the modes of the elastic shell structure without acoustic pressure loading
along the fluid-structure coupling interface, and expanding the acoustic
nodal degrees of freedom in terms of a set of ma modes of the uncoupled
acoustic system, i.e. the modes of the fluid domain with the fluid-structure
coupling interface assumed to be perfectly rigid.
The structural expansion becomes then
360 Appendix D
ms
{wi } = ∑ φ sm . {Φ su, m } = [Φ su ]. {φ s } , (D.26)
m=1
where each column in the (nsxms) matrix [Φ su ] is a modal vector Φ su,m of { }

the uncoupled elastic shell structure with natural frequency ω s,m ,
[ K s ]. {Φ su, m } = ω s2, m [ Ms ]. {Φ su, m } ( m = 1.. ms ) , (D.27)
and where {φs } is a (msx1) vector of structural modal participation factors.

The acoustic expansion is
ma
{ pi } = ∑ φ am . {Φ au, m } = [Φ au ]. {φ a } , (D.28)
m=1
where each column in the (naxma) matrix [Φau ] is a modal vector

{Φau,m } of the uncoupled fluid domain with natural frequency ω a,m ,
[ K a ]. {Φ au, m } = ω a2, m [ Ma ]. {Φ au, m } ( m = 1.. ma ) , (D.29)
and where {φa } is a (max1) vector of acoustic modal participation factors.

By substituting the component mode expansions (D.26) and (D.28), in
which the uncoupled structural and acoustic modal vectors are normalised
with respect to their corresponding mass matrices, into the Eulerian model
(D.14) and by premultiplying the structural and acoustic part of the resulting
matrix equation with the transpose of, respectively, the structural and the
acoustic modal vector matrix, the following modal model is obtained,
φ
([ K ] + jω[C] − ω 2[ M ]). îφ as  = {F} .

(D.30)
The ((ms+ma)x1) right-hand side vector is

 T 0   Fsi 
{F} = Φ0su

.   . (D.31)
 Φ Tau  î Fai 
The ((ms+ma)x(ms+ma)) modal stiffness matrix is
Λ
[ K ] =  0s

A
Λ a 
, (D.32)
where the (msxms) matrix [ Λ s ] and the (maxma) matrix [ Λ a ] are diagonal
matrices,
ω s2,1 0 ω a2,1 0 
   
[Λ s ] =  [Λ a ] = 

,  , (D.33)
 0 ω s, m s 
2  0 ω a, m a 
2
 
and where the (msxma) matrix [ A] is
[ A] = [Φ su ]T .[ K c ].[Φ au ] (D.34)
The ((ms+ma)x(ms+ma)) modal mass matrix is
 Is 0
[ ]
M =

I a 
, (D.35)
− ρ 0 A
T
where [ I s ] and [ I a ] are, respectively, the (msxms) and (maxma) unity

matrices.
The ((ms+ma)x(ms+ma)) modal damping matrix is
 T 
[C] = Φ su . C0s .Φ su 0

 . (D.36)
 Φ Tau . Ca .Φ au 
In comparison with the modal expansion, discussed in section D.1.2., the

calculation of the modal vectors are much less computationally demanding,
since the modes of the uncoupled structural and uncoupled acoustic systems
362 Appendix D
result from symmetric eigenvalue problems. However, the efficiency of the

component mode synthesis in reducing the size of the original Eulerian
model is substantially smaller. This is mainly due to the fact that the modes
of the uncoupled acoustic system have a zero normal fluid displacement
along the fluid-structure coupling interface. Hence, the displacement
continuity at the fluid-structure coupling interface is violated and the
representation of the near-field effects in the vicinity of the fluid-structure
coupling interface requires a substantial amount of high-order modes in the
acoustic modal base, which results in a slow convergence. An illustration of
this reduced efficiency is given in the next section.
D.1.4. One-dimensional example

A one-dimensional acoustic tube of length Lx is filled with a fluid with
density ρ0 and a speed of sound c. One end of the tube is rigid walled, while
the other end consists of a piston with an infinitesimal area dy.dz and a mass
per unit area ms. The piston is elastically supported with a stiffness per unit
area ks and is excited by a dynamic force F.dydz at frequency ω (see figure
D.2(a)).
Figure D.2 : one-dimensional coupled problem (a) and its acoustic discretization (b)
The exact steady-state pressure response in the fluid and the exact steady-
state piston displacement are
ρ 0 cωF ωx
pexact ( x ) = − cos( ) , (D.37)
ωL x ωL c
ρ 0 cω cos( ) + ( k s − msω 2 ) sin( x )
c c
ωL x
F.sin( )
wexact = c , (D.38)
ωL x ωL
ρ 0 cω cos( ) + ( ks − msω 2 ) sin( x )
c c
in which the harmonic time dependence e jωt is suppressed.
The steady-state pressure response in the fluid is approximated using a finite

element discretization, which consists of n linear fluid elements with (n+1)
nodal pressure degrees of freedom (see figure D.2(b)). The corresponding
acoustic FE model is given in appendix A. The steady-state piston
displacement is the single nodal degree of freedom of the structural FE
model. The Eulerian FE/FE model of the coupled vibro-acoustic system is
 w  F 
 p  0
0  
1   
 k s Kc   ms   
  −ω2   p  = 0 .
Ma   2   
. (D.39)
− ρ 0 K c
T
 0 Ka 

   
î pn + 1  î 0 
The ((n+1)x(n+1)) acoustic stiffness and mass matrices [ K a ] and [ M a ] are

given in equations (A.83) and (A.84) in appendix A. The (1x(n+1)) coupling
matrix [ K c ] is
[ Kc ] = [0 0
0 −1] . (D.40)
The modal expansion technique, discussed in section D.1.2., and the

component mode synthesis technique, discussed in section D.1.3., have been
applied to an Eulerian model (D.39) with n=50 fluid elements for the case of
an air-filled (ρ0=1.225 kg/m , c=340 m/s) tube of length Lx=1 m with a
3
piston (ms=1 kg/m , ks=100000 N/m3), excited by a force F=1 N/m2 at

2
frequency ω/2π=200 Hz. Figure D.3 compares the exact steady-state

pressure and displacement with the corresponding prediction results,
obtained with, on the one hand, an expansion with mc=5 modes of the
coupled system and, on the other hand, an expansion with the single
structural degree of freedom and ma=4 modes of the uncoupled acoustic
system. This figure clearly illustrates that, in comparison with a modal
364 Appendix D
expansion model of the same size, the accuracy with a component mode
synthesis model is significantly smaller, due to the zero fluid displacement
in all uncoupled acoustic modes at the fluid-structure coupling interface
(x=1).
Figure D.3 : instantaneous steady-state pressure (a) and displacement (b) at 200 Hz
(solid: exact, x-marked: modal expansion (mc=5,n=50),
dashed : component mode synthesis (ma=4,n=50))
As a result, the component mode synthesis technique requires a substantially

larger modal base to get an accuracy, comparable with the accuracy of the
modal expansion technique. This is illustrated in figure D.4., which plots the
accuracy of the pressure and displacement predictions, indicated by their

error measures ∆p and ∆v, defined in (A.85) in appendix A, against the
number of modes in a modal expansion and a component mode synthesis
model.
Figure D.4 : pressure (left) and displacement (right) convergence rates

for modal expansion (solid) and component mode synthesis (dashed) models
D.2. Coupled FE/direct BE model

In a coupled FE/direct BE model for interior or exterior coupled vibro-
acoustic systems with a closed boundary surface, an acoustic direct BE
modal with nodal boundary pressure and boundary normal velocity degrees
of freedom is used for the prediction of the steady-state acoustic response on
the closed boundary surface of the fluid domain and a structural FE model
with nodal translational and rotational displacement degrees of freedom is
used for the prediction of the steady-state dynamic displacement of the
middle surface of the elastic shell structure.
acoustic BE model
As derived in appendix C, the boundary element approximations of the
steady-state pressure and normal velocity on the closed boundary surface Ωa
of a fluid domain are expansions p and vn in terms of a set of global shape

functions,
p( ra ) = [ N a ]. { pi }, ( ra ∈ Ω a ),

(D.41)
vn ( ra ) = [ N a ]. {vni }, ( ra ∈ Ω a ),

(D.42)
366 Appendix D
where [ N a ] is a (1xna) matrix of global shape functions, associated with the

na nodes on the boundary surface Ωa and where the (nax1) vectors { pi } and

{vni }

comprise the nodal boundary pressure and boundary normal velocity

degrees of freedom. In absence of any external acoustic source, the resulting
direct BE model takes the form,
[ A]{ pi } = jρ 0ω [ B]{vni } ,

(D.43)
where the (naxna) matrices [ A] and [ B] describe a set of na equations in the

2na nodal degrees of freedom, of which only na degrees of freedom are
unknown, since, for each node, the pressure or the normal velocity or the
impedance relation between both variables are known a priori from the
boundary conditions. Note that the normal direction has a positive
orientation into the fluid domain.
structural FE model
The FE model for an elastic shell structure is derived in appendix B and
briefly summarised in section D.1.1.
coupling of both models

Assuming that the elastic shell structure is completely comprised in the
boundary surface Ωa, the nodes in the BE discretization may be subdivided
into a group of na1 nodes, located on the fluid-structure coupling interface
Ωs, and a group of na2 nodes, located on the remaining part Ωa\Ωs of the
boundary surface (na1+na2=na). Assuming that on the latter part of the
boundary surface a prescribed normal velocity is imposed, the
approximations of the acoustic boundary pressure and normal velocity may
be written as
p( ra ) = [ N a1 ]. { pi 1} + [ N a 2 ]. { pi 2 }, ( ra ∈ Ω a ),

(D.44)
vn ( ra ) = [ N a 1 ]. {vni 1} + [ N a 2 ]. {vni 2 }, ( ra ∈ Ω a ).

(D.45)
The (1xna1) matrix [ N a1 ] contains the global shape functions, associated

with the coupling interface degrees of freedom in the (na1x1) vectors { pi1}

and {vni1} . The (1xna2) matrix [ N a2 ] contains the global shape functions,

associated with the remaining pressure degrees of freedom in the (na2x1)

vector { pi2 } and the prescribed normal velocity degrees of freedom in the
(na2x1) vector {vni2 } .
The force loading of the acoustic pressure on the elastic shell structure along
the fluid-structure coupling interface in an interior or exterior coupled vibro-
acoustic system may be regarded as an additional normal load. As a result,
an additional term of type (D.6), using the acoustic pressure approximation
(D.44), must be added to the structural FE model (D.5), yielding
([ Ks ] + jω[Cs ] − ω 2[ Ms ]).{wi } + [ Lc ].{ pi1} = {Fs } .

! (D.46)
The (nsxna1) coupling matrix [ Lc ] is
 
[ Lc ] = − ∑  ∫ ([ N s ]T . {n e }.[ N a1 ]). dΩ  ,
n se
(D.47)
e = 1  Ω se 
where nse is the number of flat plate elements Ωse in the shell discretization
and where the unit vector, normal to a plate element, is represented in the
{ }
(3x1) vector n e . Note that, since the positive orientations of these normal
vectors are defined, such that a positive pressure load applies in the positive
normal displacement direction, the structural normal vectors have a positive
orientation away from the fluid domain, whereas the acoustic normal
vectors have a positive orientation into the fluid domain. Note also that the
nodal pressure degrees of freedom in { pi2 } do not appear in (D.46), since
"
their global shape functions are zero along the fluid-structure coupling
interface.
Assuming that the structural and acoustic meshes are matching, i.e. that the
acoustic nodes on the fluid-structure coupling interface Ωs coincide with the
structural nodes, the continuity of the normal shell velocities and the normal
fluid velocities at the fluid-structure coupling interface is expressed as
{vni 1} = jω ([Ts ]. {wi } + [Tw ]. {wi }) .

#
(D.48)
Row i1 in the (na1xns) matrix [ Ts ] is defined as

368 Appendix D
{ } .[ N s ( ri 1 )]
T
− ne (i1 = 1.. na 1 ), (D.49)
where the (3x1) vector n e { } represent the structural normal vector of the
plate element, of which one of the nodes coincides with the acoustic node i1,
located at ri1. The negative sign in (D.49) results from the opposite
orientations of the structural and acoustic normal vectors. Row i1 in the
(na1xnw) matrix [ Tw ] is defined in a similar way,
{ } .[ N w ( ri 1 )]
T
− ne (i 1 = 1.. na 1 ). (D.50)
Substitution of (A.48) into the acoustic direct BE model (D.43) yields
 A11 A12   pi 1   B11 (

B12  − ρ 0ω 2 [ Ts ]{wi } + [ Tw ]{wi } ) . (D.51)
$
A  = 
 21 A22  î pi 2   B21 B22  î jρ 0ωvni 2 
$
Combining the modified structural FE model (D.46) and the modified

acoustic BE model (D.51) yields the coupled FE/direct BE model
 K s + jωCs − ω 2 M s Lc 0   wi   Fs 
    
%
 ρ 0ω 2 B11Ts A11 A12   pi 1  =  Fa 1 

%
(D.52)
 ρ 0ω 2 B21Ts A21 A22  î pi 2  î Fa 2 
 
with
{Fa1} = − ρ 0ω 2 [ B11 ][Tw ]{wi } + jρ 0ω [ B12 ]{vni 2 } , (D.53)
{Fa1} = − ρ 0ω 2[ B21 ][Tw ]{wi } + jρ 0ω [ B22 ]{vni 2 } . (D.54)
For the special case of a coupled vibro-acoustic system, in which the whole
boundary surface consists of a closed elastic structural surface, the coupled
FE/direct BE model is
 K s + jωCs − ω 2 M s Lc  wi   Fs 
   =   (D.55)
 ρ 0ω 2 BTs A  î pi  î Fa 
&
with
{Fa } = − ρ 0ω 2 [ B][Tw ]{wi } . (D.56)
370 Appendix D
Appendix E. WAVE PROPAGATION IN FLUID-
SATURATED POROELASTIC MEDIA
This appendix presents a theoretical model for the description of wave

propagation in (two-dimensional) fluid-saturated poroelastic media, based
on the Biot theory (BIOT (1956)), as interpreted by BOLTON et al. (1996).
E.1. Stress-strain relations

Poroelastic materials have two phases, i.e. the solid phase, often referred to
as the frame, and the fluid phase, contained within the pores, formed by the
frame. In the classical Biot theory, it is assumed that the pores are
homogeneously distributed in the poroelastic material and that the dynamic
deformations in both phases may be described on a macroscopic level, using
the modelling principles, used in isotropic continuum mechanics.
When it is assumed that the dynamic deformations have a harmonic time-
dependence with circular frequency ω, the mean macroscopic displacement
of the solid phase in a two-dimensional material sample may be described
by the vector field
u x ( x, y ) jωt
u ( x , y, t ) =  . e (E.1)
î uy ( x, y)
and the mean macroscopic displacement of the fluid phase may be described
by the vector field
372 Appendix E
U x ( x , y ) jωt
U ( x, y, t ) = 

. e . (E.2)
î U y ( x, y )
According to BIOT (1956), the two-dimensional stress-strain relations in

isotropic fluid-saturated poroelastic media may be defined as
∂u x
σ x = 2 N. + A. es + Q. ε , (E.3)
∂x
∂u y
σ y = 2 N. + A. es + Q. ε , (E.4)
∂y
∂u ∂u y
τ xy = τ yx = N ( x + ), (E.5)
∂y ∂x
s = R. ε + Q. es , (E.6)
where σx and σy are the forces per unit material area, acting in, respectively,
the x- and y-direction on the solid phase, where τxy is the shear force per unit
material area, acting on the solid phase in the y-direction in a plane, whose
normal vector is oriented in the x-direction, and where s is the force per unit
material area, acting on the fluid phase. Note that the latter is assumed to be
negative in compression. The terms es and ε in equations (E.3)-(E.6) denote,
respectively, the volumetric strain in the solid phase,
∂u x ∂u y
es = ∇. u = +

, (E.7)
∂x ∂y
and the volumetric strain in the fluid phase,
∂U x ∂U y
ε = ∇.U =

+

. (E.8)
∂x ∂y
The constant N is the shear modulus,
E1
N= , (E.9)
2(1 + ν )
Wave propagation in fluid-saturated poroelastic media 373
where E1 is the in-vacuo modulus of elasticity of the bulk solid phase and ν
is its Poisson coefficient. To account for internal friction losses, the modulus
of elasticity of the solid phase can be made complex, E1=Em(1+jη), where
Em is the real, static modulus and η is a loss factor, that accounts for the
mechanical damping, associated with the motion of the solid phase of the
material.
The constant A is the first Lamé constant,
ν . E1
A= . (E.10)
(1 + ν )(1 − 2ν )
The coefficient Q represents the coupling between the volume change of the
solid and that of the fluid, and may be defined as
Q = (1 − h) E2 , (E.11)
where h is the porosity, i.e. the ratio of the fluid volume to the total material
volume, and where E2 is the modulus of elasticity of the bulk fluid phase. To
account for the losses, associated with heat transfer from the fluid within the
pores to the pore walls, this modulus of elasticity can be made complex
(BOLTON et al. (1996)),
E0
E2 = , (E.12)
 − j 8ωρ 0 ε ’ N pr 
 J1 ( )
2(γ − 1) hσ
1+  . 
 − j 8ωρ 0 ε ’ N pr − j 8ωρ 0 ε ’ N pr 
 J0 ( )
 hσ hσ 
where E0=ρ0c2, ρ0 is the ambient fluid density, c is the ambient speed of

sound, γ is the ratio of the specific heats for the ambient fluid, Npr is the
Prandtl number, ε‘ is the geometrical structure factor, σ is the steady-state
macroscopic flow resistivity and where J0 and J1 are, respectively, the zero-
order and first-order Bessel functions of the first kind.
The constant R relates the fluid stress and strain, and may be defined as
R = hE 2 . (E.13)
374 Appendix E
E.2. Dynamic equations

According to BIOT (1956), the steady-state dynamic force equilibrium in
both phases of an isotropic fluid-saturated poroelastic material may be
expressed as
∂σ x ∂τ yx
+ = −ω 2 ( ρ 1. u x + ρ a .(u x − U x )) + jωb(u x − U x ) , (E.14)
∂x ∂y
∂τ xy ∂σ y
+ = −ω 2 ( ρ 1. u y + ρ a .(u y − U y )) + jωb(u y − U y ) , (E.15)
∂x ∂y
∂s
= −ω 2 ( ρ 2 .U x + ρ a .(U x − u x )) + jωb(U x − u x ) , (E.16)
∂x
∂s
= −ω 2 ( ρ 2 .U y + ρ a .(U y − u y )) + jωb(U y − u y ) , (E.17)
∂y
where ρ1 is the bulk density of the solid phase and where ρ2=h.ρ0 is the bulk
density of the fluid phase. The first terms in the right-hand sides of
equations (E.14)-(E.17) represent the solid or fluid accelerations. The
second terms represent inertial coupling forces, which result from
momentum transfer between the solid and the fluid due to the pore tortuosity
and variation of the pore diameter. The latter effects may be quantified by
the geometrical structure factor ε‘. This factor is equal to unity, if the pores
are straight and uniform (in which case there is no inertial coupling), and
increases as the pores become irregularly constricted and more tortuous.
Hence, the inertial coupling forces, which are proportional to the relative
acceleration of both phases, may be quantified by the inertial coupling
parameter ρa (ALLARD et al. (1989), BOLTON et al. (1996)),
ρ a = ρ 2 (ε ’−1) . (E.18)
The last terms represent viscous coupling forces, which are proportional to
the relative velocity of both phases and may be quantified by the viscous
coupling factor b. This factor may be related to the macroscopic flow
resistivity σ of the porous material (ALLARD et al. (1989), BOLTON et al.
(1996)),
− j 8ωρ 0 ε ’
J1 ( )
j 2ωρ 2ε ’ hσ 1
b= . . .
− j 8ωρ 0 ε ’ − j 8ωρ 0 ε ’ − j 8ωρ 0 ε ’
J0 ( ) 2. J 1 ( )
hσ hσ hσ
1−
− j 8ωρ 0 ε ’ − j 8ωρ 0 ε ’
. J0 ( )
hσ hσ
(E.19)
E.3. Wave equations

By substituting equations (E.3)-(E.6) into the steady-state dynamic
equations (E.14)-(E.17), the latter may be written as two vector equations
N . ∇ 2u + ∇[( A + N ). es + Q. ε ] = −ω 2 ( ρ *11 . u + ρ *12 . U ) ,

(E.20)
∇[ Q. es + R. ε ] = −ω ( ρ *12 . u + ρ *22 .U ) ,

2

(E.21)
where
b
ρ *11 = ρ 11 + , ρ 11 = ρ 1 + ρ a , (E.22)
jω
b
ρ *12 = ρ 12 − , ρ 12 = − ρ a , (E.23)
jω
b
ρ *22 = ρ 22 + , ρ 22 = ρ 2 + ρ a . (E.24)
jω
The constant ρ11 represents the effective mass density of the solid moving in
the fluid and is equal to the real solid mass density ρ1 plus an additional
mass density ρa (see (E.18)) to account for the non-uniform relative fluid
flow through the pores. Similarly, the effective fluid mass density is the sum
of the real fluid mass density and the coupling parameter ρa. The constants
ρ11
*
, ρ12
*
and ρ22
*
are the counterparts of ρ11,, ρ12 and ρ22, when allowance is
made for viscous energy dissipation, resulting from the relative motion
between the solid and the fluid phases of the poroelastic material.
376 Appendix E
The application of the divergence operation to the vector equations (E.20)

and (E.21) yields
∇ 2 ( P. es + Q. ε ) = −ω 2 ( ρ *11. es + ρ *12 . ε ), (E.25)

∇ 2 (Q. es + R. ε ) = −ω 2 ( ρ *12 . es + ρ *22 . ε ), (E.26)
where P=A+2N. When it is assumed that the porous material is isotropic,

equations (E.25) and (E.26) can be written as
Q. ∇ 2ε + ω 2 ρ *12 . ε = − P. ∇ 2es − ω 2 ρ *11 . es , (E.27)

R. ∇ 2ε + ω 2 ρ *22 . ε = −Q. ∇ 2es − ω 2 ρ *12 . es . (E.28)
Solving these equations for ∇2ε and ε yields
( ρ *12Q − ρ *22 P ). ∇ 2 es + ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ]. es

∇ 2ε = , (E.29)
( ρ *22Q − ρ *12 R)
( PR − Q 2 ). ∇ 2es + ω 2 ( ρ *11 R − ρ *12Q). es
ε= . (E.30)
ω 2 ( ρ *22Q − ρ *12 R)
Equating (E.29) and the Laplacian of (E.30) yields
∇ 4 es + A1 . ∇ 2es + A2 . es = 0 , (E.31)
with
A1 = ω 2 ( ρ *11 R − 2ρ *12Q + ρ *22 P ) / ( PR − Q 2 ) , (E.32)
A2 = ω 4
[ ρ *11ρ *22 − ( ρ *12 ) 2 ] / ( PR − Q 2 ) . (E.33)
Equation (E.31) governs the dilatation es of the solid phase of a poroelastic

medium and the corresponding fluid dilatation ε results from (E.30). The
fact that equation (E.31) is a fourth-order wave equation indicates the
existence of two dilatational wave types with distinct wavenumbers k1 and
k2,
A1 A12 A1 A12
k1 = + − A2 , k2 = − − A2 . (E.34)
2 4 2 4
The application of the curl operation to the vector equations (E.20) and
(E.21) yields
N . ∇ 2ω s = −ω 2 ( ρ *11 .ω s + ρ *12 . Ω ), (E.35)

ρ*
Ω = − *12 ω s , (E.36)
ρ 22
where ωs is the rotational strain in the solid phase,
∂u y ∂u x
ωs = − , (E.37)
∂x ∂y
and where Ω is the rotational strain in the fluid phase,
∂U y ∂U x
Ω= − . (E.38)
∂x ∂y
Equations (E.35) and (E.36) may be combined into a second-order wave

equation for the solid rotational strain,
∇ 2ω s + kt2 .ω s = 0 , (E.39)
indicating the existence of one rotational wave type with wavenumber kt,
ρ *11ρ *22 − ( ρ *12 ) 2

kt = ω . (E.40)
N . ρ *22
As will become apparent in the next section, all three wave types, i.e. the
two longitudinal wave types and the transverse wave type, involve coupled
motion in the solid and the fluid.
E.4. Field variable expressions

The substitution of equation (E.30) into the x-components of the vector
equations (E.20) and (E.21) yields
378 Appendix E
∂ 2u x ∂ 2u x ∂es ∂ ( ∇ 2es )
N( + ) + [( A + N + a 1Q ) + a 2 Q ]=
∂x 2 ∂y 2 ∂x ∂x (E.41)
− ω 2 ( ρ *11 . u x + ρ *12 .U x ),
∂es ∂ (∇ 2 es )
[(Q + a1 R) + a2 R ] = −ω 2 ( ρ *12 . u x + ρ *22 .U x ) , (E.42)
∂x ∂x
( ρ *11 R − ρ *12Q) ( PR − Q 2 )
with a1 = , a2 = . (E.43)
( ρ *22Q − ρ *12 R) ω 2 ( ρ *22Q − ρ *12 R)
Solving equations (E.41) and (E.42) for ux and Ux yields
ρ *22 N a 2 ( ρ *22Q − ρ *12 R) ∂ (∇ 2es )

ux = ∇ 2
u x +
ω 2 [( ρ *12 ) 2 − ρ *11 ρ *22 ] ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
(E.44)
[ ρ * ( A + N ) + a1 ( ρ *22Q − ρ *12 R) − ρ *12Q ] ∂es
+ 22 ,
ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
− ρ *12 N a 2 ( ρ *11 R − ρ *12Q) ∂ (∇ 2es )

Ux = ∇ 2u x +
ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
(E.45)
[ − ρ *12 ( A + N ) + a1 ( ρ *11 R − ρ *12Q) + ρ *11Q ] ∂es
+ .
ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
The substitution of the identity
∂  ∂u x ∂u y  ∂  ∂u y ∂u x  ∂es ∂ω s
∇ 2u x =  + −  − = − (E.46)
∂x  ∂x ∂y  ∂y  ∂x ∂y  ∂x ∂y
into equations (E.44) and (E.45) yields

− ρ *22 N ∂ω s a 2 ( ρ *22Q − ρ *12 R) ∂ (∇ 2es )

ux = 2 +
ω [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂y ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
(E.47)
[ ρ * P − ρ * Q + a ( ρ * Q − ρ * R)] ∂es
+ 22 2 12 * 2 1 *22 * 12 ,
ω [( ρ 12 ) − ρ 11ρ 22 ] ∂x
ρ *12 N ∂ω s a 2 ( ρ *11 R − ρ *12Q) ∂ (∇ 2es )
Ux = +
ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂y ω 2 [( ρ *12 ) 2 − ρ *11ρ *22 ] ∂x
(E.48)
[ ρ * Q − ρ * P + a ( ρ * R − ρ * Q)] ∂es
+ 11 2 12 * 2 1 *11 * 12 .
ω [( ρ 12 ) − ρ 11ρ 22 ] ∂x
By using the expressions (E.32), (E.33), (E.40) and (E.43), equations (E.47)
and (E.48) become
ux = −
1 ∂
A2 ∂x
( )
1 ∂ω s
∇ 2es + A1. es + 2
kt ∂y
, (E.49)
a ∂
Ux = − 1
A2 ∂x
( )
∂e ρ * ∂ω s
∇ 2es + A1 . es + a 2 . s − * 12 2
∂x ρ 22 . kt ∂y
. (E.50)
In a completely similar way, the following expressions for the y-components

of the displacement vectors are obtained,
uy = −
1 ∂
A2 ∂y
( )
1 ∂ω s
∇ 2es + A1. es − 2
kt ∂x
, (E.51)
a ∂
Uy = − 1
A2 ∂y
( )
∂e ρ * ∂ω s
∇ 2es + A1. es + a 2 . s + * 12 2
∂y ρ 22 . kt ∂x
. (E.52)
The substitution of expressions (E.49)-(E.52) into equations (E.3)-(E.6)

yields the following expressions for the stress components,
σ x = ( A + a1Q)es + a 2Q∇ 2es −

2N ∂ 2
A2 ∂x 2
(
∇ 2
e s + A .
1 s)
e +
2 N ∂ 2ω s
kt2 ∂x∂y
, (E.53)
σ y = ( A + a1Q)es + a 2Q∇ 2es −

2N ∂ 2
A2 ∂y 2
(
∇ 2
e s + A .
1 se)−
2 N ∂ 2ω s
kt2 ∂x∂y
, (E.54)
380 Appendix E
τ xy = τ yx = −
2N ∂ 2
A2 ∂x∂y
(
∇ 2es + A1. es + 2 (
kt ∂y 2
)
N ∂ 2ω s ∂ 2ω s
−
∂x 2
), (E.55)
s = ( a1 R + Q)es + a 2 R∇ 2es . (E.56)
Since
A1 = k12 + k 22 , A2 = k12 . k 22 , (E.57)
it follows from the above expressions that, for any longitudinal wave with
wavenumber ki (i=1,2), the involved two-dimensional distributions of the
dynamic field variables are
− j ( k i,x . x + k i,y . y )
es ( x , y ) = e with ki2, x + ki2, y = ki2 , ω s ( x, y ) = 0 , (E.58)
1 ∂e 1 ∂e
u x ( x, y ) = − 2 s , u y ( x, y ) = − 2 s , (E.59)
ki ∂ x ki ∂y
a − a k 2 ∂es a − a k 2 ∂es
U x ( x , y ) = − 1 22 i , U y ( x , y ) = − 1 22 i , (E.60)
ki ∂x ki ∂y
2 N ∂ 2es
σ x ( x, y ) = [ A + ( a1 − a 2 ki2 )Q ]. es − , (E.61)
ki2 ∂x 2
2 N ∂ 2es
σ y ( x, y ) = [ A + ( a1 − a 2 ki2 )Q ]. es − , (E.62)
ki2 ∂y 2
2 N ∂ 2 es
τ xy ( x, y ) = τ yx ( x, y ) = − , s( x, y ) = [( a1 − a 2 ki2 ) R + Q ]. es .(E.63)
ki2 ∂x∂y
For any transverse wave with wavenumber kt, the involved two-dimensional
distributions of the dynamic field variables are
− j ( kt ,x . x + kt ,y . y )
ω s ( x, y ) = e with kt2, x + kt2, y = kt2 , es ( x , y ) = 0 , (E.64)
1 ∂ω s 1 ∂ω s
u x ( x, y ) = 2 , u y ( x, y ) = − 2 , (E.65)
kt ∂y kt ∂x
ρ * ∂ω s ρ *12 ∂ω s
U x ( x, y ) = − * 12 2 , U y ( x, y ) = , (E.66)
ρ 22 . kt ∂y ρ *22 . k t2 ∂x
2 N ∂ 2ω s 2 N ∂ 2ω s
σ x ( x, y ) = , σ y ( x, y ) = − , (E.67)
kt2 ∂x∂y kt2 ∂x∂y
N ∂ 2ω s ∂ 2ω s
τ xy ( x, y ) = τ yx ( x, y ) = ( − ), s( x, y ) = 0 . (E.68)
kt2 ∂y 2 ∂x 2
E.5. Reciprocity relation

For any two dynamic field variable distributions, associated with two waves
in a two-dimensional, poroelastic domain Ωp with closed boundary surface
Γp, their boundary surface distributions are such that
∫ (σ n,i . un, j + si .Un, j + τ ns,i . us, j ). dΓ =

Γp
(E.69)
∫ (σ n, j . un,i + s j .U n,i + τ ns, j . us,i ). dΓ ,
Γp
where the indices n and s refer to the normal and tangential directions γn and
γs of the boundary surface (see figure E.1)
Figure E.1 : two-dimensional poroelastic domain
For the case of two longitudinal waves, the proof of relation (E.69) consists
of demonstrating whether
382 Appendix E
 2 N ∂ 2es, i ∂es, j A + ( a1 − a 2 ki2 )Q ∂e s, j 


∫  2 2 − [ ]e  dΓ
∂γ n 
s, i
Γ p  ki k j ∂γ n ∂γ n
2
k j2
 ( a1 − a 2 k 2j )[( a1 − a 2 ki2 ) R + Q ] ∂e s, j 
− ∫  e  dΓ
 2 s , i
∂γ 
Γp k j n 
 2 N ∂ 2es,i ∂es, j   2 N ∂ 2es, j ∂es, i 

+ 
∫  k 2 k 2 ∂γ ∂γ ∂γ   d Γ − ∫  k 2 k 2 ∂γ ∂γ ∂γ  dΓ (E.70)

Γp i j n s s  Γp i j n s s 
 ( a1 − a 2 ki2 )[( a1 − a 2 k j2 ) R + Q ] ∂es, i 

+ 
∫  e  dΓ
∂γ n 
2 s, j
Γp ki
 2 N ∂ 2es, j ∂es,i A + ( a1 − a 2 k j2 )Q ∂e s, i  ?
− ∫  − [ ]e  dΓ = 0
 k 2 k 2 ∂γ 2 ∂γ 2 s , j
∂γ 
Γp i j n n k i n
Since
∂ 2 e s, i ∂ 2 e s, i ∂ 2 es , i
= ∇ 2
e s, i − = − k 2
e
i s, i − , (E.71)
∂γ n2 ∂γ s2 ∂γ s2
∂ 2 e s, j ∂ 2 e s, j ∂ 2 e s, j
= ∇ 2
e s, j − = − k 2
e
j s, j − , (E.72)
∂γ n2 ∂γ s2 ∂γ s2
expression (E.70) becomes
 
2N  ∂ 2 e s , j ∂e s, i ∂ 2es, i ∂es, j ∂ 2es, i ∂es, j ∂ 2 es, j ∂es, i 
2 2  ∫
. ( + − − ) dΓ 
ki k j  Γ ∂γ s ∂γ n ∂γ n ∂γ s ∂γ s
2
∂γ s ∂γ n ∂γ n ∂γ s ∂γ s
2

 p 
 
 es, j ∂es, i es, i ∂es, j ?
+ B.  ∫ ( 2 − 2 ) dΓ  = 0
Γ k ∂γ n k j ∂γ n 
 p i 
(E.73)
with B= P + [ 2a1 − a 2 ( ki2 + k j2 )]Q + ( a1 − a 2 ki2 )( a1 − a 2 k 2j ) R . (E.74)
When both longitudinal waves have the same wavenumber, i.e. ki2 = k j2 =
kα2 (α = 1 or 2) , the application of the two-dimensional version of Green’s
second identity (see equation (2.51)) for the second term in (E.73) yields
 
 e s , j ∂e s , i e s , i ∂e s , j 
B.  ∫ ( 2 − 2 ) dΓ 
Γ k ∂γ n k j ∂γ n 
 p i 
   
B  ∂e s , i ∂es, j  B  
= 2 .  ∫ ( e s, j − e s, i ) dΓ  = 2 .  ∫ ( e s , j . ∇ 2 e s , i − e s , i . ∇ 2 e s , j ) dΩ 
kα  Γ ∂γ n ∂γ n  kα  Ω 
 p   p 
 
B  
= 2 .  ∫ ( − kα . es, j . es, i + kα . es,i . es, j )dΩ  = 0.
2 2
kα  Ω 
 p 
(E.75)
When the two longitudinal waves have different wavenumbers, i.e. ki2 = k12
and k j2 = k 22 , the substitution of (E.57) into (E.74) yields
B = P + ( 2a1 − a 2 A1 )Q + ( a12 − a1a2 A1 + a 22 A2 ) R . (E.76)
Since it follows from (E.32), (E.33) and (E.43) that
R( ρ *11Q − ρ *12 P )
P + ( 2a1 − a 2 A1 )Q = , (E.77)
( ρ *22Q − ρ *12 R)
( a12 − a1a 2 A1 + a 22 A2 ) R =
ρ *11ρ *12QR 2 + ρ *12 ρ *22 PQR − ρ *11ρ *22Q 2 R − ( ρ *12 ) 2 PR 2
= (E.78)
( ρ *22Q − ρ *12 R) 2
R( ρ *22Q − ρ *12 R)( ρ *12 P − ρ *11Q)
= ,
( ρ *22Q − ρ *12 R) 2
the substitution of (E.77) and (E.78) into (E.76) yields
B = 0. (E.79)
384 Appendix E
Hence, it follows from (E.75) and (E.79) that expression (E.73) simplifies to
∂ 2 e s , j ∂e s , i ∂ 2es, i ∂es, j ∂ 2es, i ∂es, j ∂ 2 e s , j ∂e s , i ?

∫ ( + − − )dΓ = 0 .
Γp ∂γ s2 ∂γ n ∂γ n ∂γ s ∂γ s ∂γ s2 ∂γ n ∂γ n ∂γ s ∂γ s
(E.80)
Since the boundary Γp is a closed curve, the application of integration by
parts on the first and third line integral term in (E.80) (see e.g. REKTORYS
(1977)) yields
nc  ∂es, j ( rc ) ∂es, i ( rc ) ∂es, j ( rc ) ∂es, i ( rc ) 

∑  − 
c = 1 ∂γ −s ∂γ n− ∂γ +s ∂γ n+ 
(E.81)
nc  ∂es,i ( rc ) ∂es, j ( rc ) ∂es,i ( rc ) ∂es, j ( rc )  ?
− ∑  −  = 0,
c = 1  ∂γ s
−
∂γ n− ∂γ s+ ∂γ n+ 
where nc is the number of corner points of the boundary, at which the normal
and tangential directions are not uniquely defined (see figure E.2).
Figure E.2 : normal and tangential directions at a corner point
By using the unit normal vectors ( n x+ ,n y+ ) and ( n x− ,n y− ) to express the

spatial derivatives in the local γ n+ -, γ n− -, γ +s - and γ −s -directions at each
corner point in terms of the spatial derivatives in the global x- and y-
directions, expression (E.81) becomes
 ∂es, j ∂es, j − ∂es, i ∂es,i 

 ( − n y− + n x− )( n x + n y− )
 ∂x ∂y ∂x ∂y 
 ∂e ∂e ∂e ∂es,i 
 −( n x− s, j + n y− s, j )( − n y− s, i + n x− )
nc  ∂x ∂y ∂x ∂y  n c !
∑  = ∑ 0 = 0, (E.82)
c = 1 + ∂e s , j + ∂es, j + ∂es, i + ∂es, i  c = 1
 − ( − n + n )( n + ny )
y
∂x x
∂y x
∂x ∂y 
 
 + ∂es, j ∂es, j ∂e s , i ∂ e s, i 
 +(n x + n y+ )( − n y+ + n x+ )
 ∂x ∂y ∂x ∂y 
which proves that relation (E.69) is valid for any two longitudinal waves.
For the case of one longitudinal wave and one transverse wave, the proof of
relation (E.69) consists of demonstrating whether
 2 N ∂ 2es, i ∂ω s, j A + ( a1 − a 2 ki2 )Q ∂ω s, j 
∫  − k 2k 2 ∂γ 2 ∂γ + [ 2
]e s , i
∂γ
 dΓ

Γp  i t n s k t s 
 ρ *12 [( a1 − a 2 ki2 ) R + Q ] ∂ω s, j   2 N ∂ 2es,i ∂ω s, j 

− 
∫  ρ *22 kt2
e s,i
∂γ s 
 d Γ + ∫  2 2  dΓ

Γp Γ p  ki kt ∂γ n ∂γ s ∂γ n 
 N ∂ 2ω s, j ∂ 2ω s, j ∂es, i   2 N ∂ 2ω s, j ∂es,i  ?
+ ∫  ( − )  d Γ + ∫   dΓ = 0
 2 2 ∂γ 2 ∂γ n2 ∂γ s   2 2 
Γ p  ki kt s Γ p  ki kt ∂γ n∂γ s ∂γ n 
(E.83)
The substitutions of (E.71) and
∂ 2ω s, j ∂ 2ω s, j ∂ 2ω s, j
= ∇ ω s, j −
2
= − kt2ω s, j − , (E.84)
∂γ n2 ∂γ s2 ∂γ s2
into (E.83) yield

386 Appendix E
 
2N  ∂ 2es, i ∂ω s, j ∂ 2es,i ∂ω s, j ∂ 2ω s, j ∂es,i ∂ 2ω s, j ∂es, i 
2 2 ∫
( + + + ) dΓ 
ki k t  Γ ∂γ s ∂γ s
2
∂γ n∂γ s ∂γ n ∂γ s ∂γ s ∂γ n∂γ s ∂γ n
2

 p 
 
 ∂ω s, j N ∂es,i ?
+  ∫ (C. es, i + 2 .ω s , j )dΓ  = 0,
Γ ∂γ s ki ∂γ s 
 p 
(E.85)
ρ *22 P − ρ *12Q + a1 ( ρ *22Q − ρ *12 R) − a 2 ki2 ( ρ *22Q − ρ *12 R)
with C = . (E.86)
ρ *22 kt2
Since equation (E.31) implies that
A2 = A1ki2 − ki4 , (E.87)
it follows that
( E .86 ) ρ *22 P − ρ *12Q + a1 ( ρ *22Q − ρ *12 R) − a 2 ki2 ( ρ *22Q − ρ *12 R)

C =
ρ *22 kt2
( E . 43 ) ( ρ *
11 R − 2ρ 12Q + ρ 22 P )ω − ki ( PR − Q )
* * 2 2 2
=
ρ *22 kt2ω 2
(E.88)
( E . 32) ( A − k 2 )( PR − Q 2 ) ( E . 40 ) N ( A − k 2 )( PR − Q 2 )
= 1 i
= 1 i
ρ *22 kt2ω 2 [ ρ *11ρ *22 − ( ρ *12 ) 2 ]ω 4
( E. 33) N ( A − k 2 ) ( E.87 ) N
= 1 i = .
A2 ki2
The substitution of (E.88) into (E.85) yields
 ∂ 2es, i ∂ω s, j ∂ 2es, i ∂ω s, j ∂ 2ω s, j ∂es, i ∂ 2ω s, j ∂es, i 

 + + + 
 ∂γ s2 ∂γ s ∂γ n ∂γ s ∂γ n ∂γ s ∂γ s ∂γ n ∂γ s ∂γ n  ?
2
∫  2  dΓ = 0.
Γ p  kt ∂ω s, j kt2 ∂es, i 
+ e s, i + ω s, j 
 2 ∂γ s 2 ∂γ s 
(E.89)
The application of integration by parts on the first, second and fifth line
integral term in (E.89) yields
nc  ∂es, i ( rc ) ∂ω s, j ( rc ) ∂es, i ( rc ) ∂ω s, j ( rc ) 
∑  − 
c = 1 ∂γ −s ∂γ −s ∂γ s+ ∂γ s+ 
(E.90)
nc  ∂es,i ( rc ) ∂ω s, j ( rc ) ∂es, i ( rc ) ∂ω s, j ( rc )  ?
+ ∑  −  = 0.
c = 1  ∂γ n
−
∂γ n− ∂γ n+ ∂γ n+ 
By expressing the local spatial derivatives in terms of the global spatial

derivatives, expression (E.90) becomes
 ∂es, i ∂ e s, i ∂ω s, j ∂ω s, j 
 ( − n y− + n x− )( − n y− + n x− )
 ∂x ∂y ∂x ∂y 
 ∂ω s, j 
∂e ∂e
 −( − n y+ s,i + n x+ s,i )( − n y+ + ∂ω s, j 
nc 
+ n x )
∂x ∂y ∂x ∂y  n c !
∑  = ∑ 0 = 0, (E.90)
c = 1 − ∂es, i − ∂ e s, i − ∂ω s, j − ∂ω s, j  c=1
 +(n x ∂x + n y ∂y )( n x ∂x + n y ∂y ) 
 
 + ∂ e ∂ e ∂ω ∂ω 
+ n y+ )(n x+ + n y+
s, i s, i s , j s , j
 −( n x ) 
 ∂x ∂y ∂x ∂y 
which proves that relation (E.69) is valid for any pair of one longitudinal
and one transverse wave.
For the case of two transverse waves, the proof of relation (E.69) consists of
demonstrating whether
 
N  ∂ 2ω s, i ∂ω s, j ∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i ∂ 2ω s, j ∂ω s, i 
− 4 ∫ ( − − + ) dΓ 
kt  Γ ∂γ s ∂γ n
2
∂γ n ∂γ n
2
∂γ s ∂γ n
2
∂γ n ∂γ n
2

 p 
 
2N  ∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i ?
+ 4 ∫ ( − )dΓ  = 0.
kt  Γ ∂γ n∂γ s ∂γ s ∂γ n∂γ s ∂γ s 
 p 
(E.91)
388 Appendix E
The substitution of (E.84) into (E.91) yields
 
2N  ∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i ∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i 
4  ∫
( − − + ) d Γ 
kt  Γ ∂γ n ∂γ s ∂γ s ∂γ n ∂γ s ∂γ s ∂γ s ∂γ n
2
∂γ s ∂γ n
2

 p 
 
N  ∂ω s, i ∂ω s, j ?
+ 2  ∫ (ω s, j − ω s, i )dΓ  = 0.
kt  Γ ∂γ n ∂γ n 
 p 
(E.92)
The application of the two-dimensional version of Green’s second identity
for the second term in (E.92) yields
   
N  ∂ω s, i ∂ω s, j  N  
2 ∫
(ω s, j − ω s, i )dΓ  = 2  ∫ (ω s, j ∇ ω s, i − ω s, i ∇ ω s, j )dΩ 
2 2
kt  Γ ∂γ n ∂γ n  kt  Ω 
 p   p 
 
N  
= 2  ∫ ( − kt ω s, j ω s, i + kt ω s, i ω s, j )dΩ  = 0,
2 2
kt  Ω 
 p 
(E.93)
so that expression (E.92) simplifies to
∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i ∂ 2ω s, i ∂ω s, j ∂ 2ω s, j ∂ω s, i ?
∫ ( − − + )dΓ = 0.
Γ p ∂γ n ∂γ s ∂γ s ∂γ n ∂γ s ∂γ s ∂γ s2 ∂γ n ∂γ s2 ∂γ n
(E.94)
The application of integration by parts on the first and second line integral
term in (E.94) yields
nc  ∂ω s, i ( rc ) ∂ω s, j ( rc ) ∂ω s, i ( rc ) ∂ω s, j ( rc ) 
∑  − 
c = 1 ∂γ −
n ∂γ −
s ∂γ n+ ∂γ s+ 
(E.95)
nc  ∂ω s, j ( rc ) ∂ω s, i ( rc ) ∂ω s, j ( rc ) ∂ω s, i ( rc )  ?
− ∑  −  = 0.
c = 1 ∂γ n− ∂γ s− ∂γ n+ ∂γ s+ 
By expressing the local spatial derivatives in terms of the global spatial

derivatives, expression (E.95) becomes
 ∂ω s,i ∂ω s,i ∂ω s, j ∂ω s, j 
 ( n x− + n y− )( − n y− + n x− )
 ∂x ∂y ∂x ∂y 
 
 −( n x+
∂ω s, i + ∂ω s, i + ∂ω s, j + ∂ω s, j 
nc 
+ n y )( − n y + nx )
∂x ∂y ∂x ∂y  nc !
∑  = ∑ 0 = 0, (E.96)
c = 1 − ∂ω s, i − ∂ω s, i − ∂ω s, j − ∂ω s, j  c = 1
 −( − n y ∂x + n x ∂y )( n x ∂x + ny
∂y 
)
 
 + ∂ω s, i ∂ω s, i + ∂ω s, j ∂ω 
+ n x+ + n y+
s , j
 +( − n y )( n x )
 ∂x ∂y ∂x ∂y 
which proves that relation (E.69) is valid for any two transverse waves.
As a result, it may be concluded from (E.82), (E.90) and (E.96) that relation
(E.69) is valid for any two waves in a two-dimensional poroelastic domain.
390 Appendix E
Appendix F. COUPLED VIBRO-ACOUSTIC
BEHAVIOUR OF DOUBLE-PANEL PARTITIONS
F.1. Introduction
Double-panel partitions often provide an efficient solution for noise control
engineering applications due to their high sound transmission loss,
combined with a fairly low specific weight. Examples include machinery
sound encapsulations, double glazing windows, partition walls in buildings,
movable walls in recording studios and aircraft fuselages. However, the
sound transmission loss of double-panel partitions rapidly deteriorates
towards the low frequency range, at which it can even fall short of a single
panel of the same total mass. Since in many practical applications double-
panel partitions are exposed to significant low-frequency dynamic loads,
active and/or passive control measures are often required to enhance the
low-frequency sound transmission loss. With respect to an adequate and
efficient design of these measures, reliable prediction models are very useful
tools to get a good insight in the physical phenomena that govern the
dynamic behaviour of double-panel partitions.
Prediction models for the sound transmission of double-panel partitions with
infinitely extended panels go back to the work of BERANEK and WORK
(1949), who studied the sound transmission of normally incident plane
waves through unbounded double-panel partitions. With some additional
modifications for obliquely incident waves and improvements with respect
to the panel flexibility (MULHOLLAND et al. (1967), SHARP (1978),
TROCHIDIS and KALAROUTIS (1986)), these simple prediction models have
392 Appendix F
gained wide acceptance in practical engineering, especially in building

acoustics (VINOKUR (1996)).
For the modelling of the high-frequency sound transmission of finite-sized
double-panel partitions, considerable simplifications still can be made using
Statistical Energy Analysis (WHITE and POWELL (1965), PRICE and
CROCKER (1970), CRAIK and WILSON (1995), MANNING and HEBERT
(1995)). In the low-frequency range, however, the modal behaviour of the
panels and the enclosed cavity can no longer be neglected. The author and
some colleagues of the PMA Noise and Vibration research group have spent
a vast amount of research effort on the investigation of the low-frequency
dynamic behaviour of finite-sized double-panel partitions. Analytical and
numerical prediction models, together with several experimental validations,
have provided considerable insight in the effect of the vibro-acoustic
coupling interaction between the panels and the enclosed cavity on the low-
frequency modal properties and sound transmission characteristics of finite-
sized double-panel partitions. Based on the identified modal properties,
several active control configurations have been designed and successfully
implemented, yielding a significant improvement of the low-frequency
sound transmission loss.
This appendix describes the main low-frequency modal properties of
double-panel partitions, identified from an analytical model, along with a
wave model, obtained with the proposed wave based prediction technique.
Some experimental validations of these properties are given in chapter 4.
For a detailed discussion of the numerical and experimental investigations
of the low-frequency dynamic behaviour and sound transmission
characteristics of double-panel partitions, the reader is referred to DESMET
et al. (1994) and DESMET and SAS (1994b, 1995, 1997). The active control
implementations are discussed in SAS et al. (1993, 1995), DESMET and SAS
(1994a) and DE FONSECA et al. (1996)).
F.2. Prediction models for double-panel partitions
F.2.1. Problem definition
A typical double-panel partition is shown in figure F.1. A rectangular fluid-

filled cavity is enclosed by two identical flexible panels and four rigid side
walls, to which the edges of both panels are clamped. The dynamic variables
of interest are the steady-state pressure field in the cavity and the steady-
Coupled vibro-acoustic behaviour of double-panel partitions 393
state normal displacement fields of both panels due to a mechanical point

force, applied normal to one of the panels.
Figure F.1 : double-panel partition
For a harmonic time dependence e jωt of the applied force and using a
Cartesian co-ordinate system (x,y,z) (see figure F.1), the steady-state
acoustic pressure p in the inviscid, homogeneous fluid in the rectangular
cavity is governed by the Helmholtz equation
∂2 ∂2 ∂2
( + + + k 2 ). p( x, y, z ) = 0 , (F.1)
∂x 2
∂y 2
∂z 2
with k being the acoustic wavenumber (k=ω/c), ω the circular frequency and
c the speed of sound in the fluid. Due to the mechanical point force F,
applied at position (xF,yF), and the cavity pressure loading on the panel,
located in plane z=0 and further denoted as ‘lower panel’, its steady-state
normal displacement w1, having a positive orientation in the positive z-
direction, is governed by the dynamic equation
∂4 ∂4 ∂4 F.δ ( x − x F , y − y F ) p( x, y,0)
( + 2 + − k b4 ). w1 ( x, y ) = − .
∂x 4
∂x ∂y
2 2
∂y 4 D D
(F.2)
For the other panel, located in plane z=Lz and further denoted as ‘upper
panel’, its steady-state normal displacement w2 with positive orientation in
the positive z-direction, is governed by the dynamic equation
394 Appendix F
∂4 ∂4 ∂4 p( x, y, Lz )
( + 2 + − kb4 ). w 2 ( x, y ) = . (F.3)
∂x 4
∂x ∂y
2 2
∂y 4 D
The panel bending stiffness D and the structural bending wavenumber kb are
E (1 + jη )t 3 ρ s tω 2
D= , kb = 4 , (F.4)
12(1 − ν 2 ) D
where E is the elasticity modulus of the panel material, η is the material loss
factor, ν is the material Poisson coefficient, ρs is the material density and t is
the panel thickness.
Since both panels are clamped, the boundary displacements as well as the
in-plane gradients, normal to the boundaries, must be zero
(0 ≤ x ≤ L x , 0 ≤ y ≤ L y ) ,
∂wi ( 0, y ) ∂wi ( Lx , y )
wi ( 0, y ) = 0, wi ( Lx , y ) = 0, = 0, = 0, i = 1,2 ,
∂x ∂x
∂wi ( x,0 ) ∂wi ( x, Ly )
wi ( x,0) = 0, wi ( x, Ly ) = 0, = 0, = 0, i = 1,2 . (F.5)
∂y ∂y
Since there is no fluid displacement, normal to the four rigid side walls of
the cavity, the cavity pressure must satisfy the boundary conditions
(0 ≤ x ≤ L x , 0 ≤ y ≤ L y ,0 ≤ z ≤ Lz ) ,
∂p( 0, y, z ) ∂p( Lx , y, z ) ∂p( x,0, z ) ∂p( x, Ly , z )

= 0, = 0, = 0, = 0. (F.6)
∂x ∂x ∂y ∂y
Since the normal fluid displacement equals the normal structural

displacement at the panel-cavity interfaces, the cavity pressure must also
satisfy the boundary conditions (0 ≤ x ≤ L x , 0 ≤ y ≤ L y ) ,
∂p( x , y,0 ) ∂p( x, y, Lz )

= ρ 0ω 2 w1 ( x, y ), = ρ 0ω 2 w 2 ( x, y ), (F.7)
∂z ∂z
where ρ0 is the ambient fluid density.

F.2.2. Analytical model
An analytical prediction model for double-panel partitions has been built,

based on the modal coupling theory of DOWELL et al. (1977), as interpreted
by FAHY (1985).
The basic idea of the analytical prediction model is the approximation of the
dynamic structural and acoustic field variables in terms of the uncoupled
structural and uncoupled acoustic modes, i.e. the in-vacuo panel modes Ψ s
and the acoustic modes Φ a of the cavity, of which all six side walls are
assumed to be rigid,
w1 ( x, y ) ≈ ∑ Ws( 1) .Ψ s ( x, y ) and w 2 ( x, y ) ≈ ∑ Ws( 2) .Ψ s ( x, y ) , (F.8,9)

s s
p( x, y, z ) ≈ ∑ Pa .Φ a ( x, y, z ) . (F.10)
a
These modal expansions exactly satisfy the structural boundary conditions,

defined in (F.5), as well as the acoustic boundary conditions, defined in
(F.6). On the contrary, the displacement continuity conditions at the panel-
cavity interfaces (see (F.7)) cannot be satisfied, since the rigid-walled cavity
modes Φ a have no fluid displacement, normal to the cavity walls. In order
to take into account the fluid-structure coupling interaction, the effects of
the panel displacements are no longer expressed as boundary conditions to
the cavity, but as ‘external’ acoustic excitations of a rigid-walled cavity.
These acoustic sources are confined to infinitesimally thin layers, located at
the panel-cavity interfaces. Consequently, the Helmholtz equation (F.1)
modifies to
∂2 ∂2 ∂2 ω2
( + + + ). p( x, y, z ) =
∂x 2 ∂y 2 ∂z 2 c 2 (F.11)
2ρ 0ω 2 [ w1 ( x, y ).δ ( z,0) − w 2 ( x, y ).δ ( z, Lz )].
To determine the modal participation factors Pa ,Ws(1) and Ws(2) in the field
variable expansions (F.8)-(F.10), these expansions are substituted in the
modified Helmholtz equation (F.11) and the structural equations (F.2) and
(F.3). Since the analytical model is used to identify the modal properties of a
double-panel partition, the damping η and the external force excitation F are
396 Appendix F
disregarded. Since each uncoupled acoustic mode Φ a satisfies the

homogeneous part of (F.11), when the frequency ω equals its associated
uncoupled eigenfrequency ω a , and since each uncoupled structural mode
Ψ s satisfies the homogeneous parts of (F.2) and (F.3), when the frequency
ω equals its associated uncoupled eigenfrequency ω s , the substitutions
yield
ω 2 − ω a2
∑ . Pa .Φ a ( x , y, z )
a c2 (F.12)
= 2ρ 0ω [ ∑Ψ s ( x, y ).(Ws( 1) .δ ( z,0 ) − Ws( 2) .δ ( z, Lz )],
2
s
∑ ms (ω s − ω ). Ws( 1) .Ψ s ( x, y) = − ∑ Pa .Φ a ( x, y,0) ,
2 2
(F.13)
s a
∑ ms (ω s − ω ). Ws .Ψ s ( x, y) = ∑ Pa .Φ a ( x, y, Lz ) ,
2 2 ( 2)
(F.14)
s a
where ms is the panel mass per unit area ( ms = ρs .t ). By multiplying both

sides of (F.12) with an arbitrary uncoupled acoustic mode shape and
integrating them over the cavity volume, the summation over all acoustic
modes in the left-hand side reduces to one single mode contribution, due to
the orthogonality of the uncoupled modes. Consequently, the participation
factor Pa of each uncoupled acoustic mode results from
(ω 2 − ω a2 ). Λ a . Pa = ρ 0 c 2ω 2 [ ∑ (Ws( 1) . Cas
( 1)
− Ws( 2) . Cas
( 2)
)] ( ∀a ). (F.15)
s
In a completely similar way, the participation factors Ws(1) and Ws(2) of each
uncoupled structural mode result from
ms (ω s2 − ω 2 ). Λ s . Ws( 1) = − ∑ Pa . Cas
( 1)
(∀s ), (F.16)
a
ms (ω s2 − ω 2 ). Λ s . Ws( 2) = ∑ Pa . Cas
( 2)
(∀s ). (F.17)
a
(1) (2)
The modal coupling factors Cas and Cas are defined as
Lx Ly
= ∫ ∫ Φ a ( x, y,0).Ψ s ( x, y)dxdy ,
( 1)
Cas (F.18)
0 0
Lx Ly
= ∫ ∫ Φ a ( x, y, Lz ).Ψ s ( x, y)dxdy ,
( 2)
Cas (F.19)
0 0
and express the geometrical matching between the uncoupled acoustic and
structural modes at their common interfaces.
The uncoupled modes of the rectangular cavity are
lπx mπy nπz

Φ a ( x, y, z ) = cos( ).cos( ).cos( ), (l, m,n = 0,1,....) . (F.20)
Lx Ly Lz
For the uncoupled modes of the clamped rectangular panels, simple

analytical expressions do not exist, but the expressions for the modes of a
simply supported rectangular panel are good approximations as long as the
length in one panel direction is not substantially larger than the other
(MITCHELL and HAZELL (1987)),
uπx vπy
Ψ s ( x, y ) ≈ sin( ).sin( ), (u, v = 1, 2,....) . (F.21)
Lx Ly
With these mode shape expressions, the modal coupling factors become
 L x Ly uv [( −1)l + u − 1][( −1) m + v − 1]

( 2)  , l ≠ u and m ≠ v.
( 1)
Cas = ( −1) n . Cas =  π2 (l 2 − u 2 )( m 2 − v 2 )

î 0, l = u and / or m = v.
(F.22)
This expression indicates the selectivity of the coupling, due to the
symmetry of the double-panel partition, since both l+u and m+v must be
odd, in order to have a non-zero coupling factor. Based on this selectivity,
the uncoupled modes may be split into four independent groups of mutually
interacting modes, depending on the number of half wavelengths in each
direction, as listed in table F.1. A panel mode with an odd number of half
wavelengths in the x- and y-direction, for example, only couples with the
acoustic modes with an even number of half wavelengths in these directions.
398 Appendix F
panel mode (u,v) cavity mode (l,m,n)

(o,o,n) (e,e,n) (o,e,n) (e,o,n)
(o,o) 0 * 0 0
(e,e) * 0 0 0
(o,e) 0 0 0 *
(e,o) 0 0 * 0
(1)
Table F.1 : modal coupling factors Cas = ( −1) n .Cas
(2)
(*: non-zero; o/e/n : odd/even/arbitrary number of half wavelengths)
With the mode shape expressions (F.21) and (F.22), the modal scale factors
Λa and Λs in (F.15)-(F.17) become
L x L y Lz
L x L y Lz
Λa = ∫ ∫ ∫ Φ a ( x, y, z )dxdydz =
2
, (F.23)
0 0 0 2i
Lx Ly
L x Ly
Λs = ∫ ∫Ψ s ( x, y)dxdy ≈
2
, (F.24)
0 0
4
where i (=0,1,2 or 3) is the number of non-zero components in the mode

annotation (l,m,n) of the considered uncoupled cavity mode.
By introducing the variable λ and the factors φa, such that
jω
λ=− and − jρ 0ωφ a = Pa (∀a) , (F.25,26)
c
the algebraic equations (F.15)-(F.17) can be written in the matrix form
([ K~] + λ.[C~] + λ .[ I ]).{X} =

2
~
 K ~ ~
0 0 0 C1 C2  0 
 Ia 0
 a  ~  
2 
I s 0  . { X} = {0}.
~
 0 Ks 0  + λ . C 3 0 0  + λ . 0
 ~ ~
C 
 0 0 K s   4 0 0   0
0 I s 
(F.27)
Assuming that the number of considered uncoupled acoustic modes in the
expansion (F.10) are confined to na and that the number of uncoupled panel
modes in (F.8) and (F.9) are confined to ns, the solution vector { X} contains
the (na+2ns) modal participation factors,
 φa 
 
{ X} = Ws(1)  . (F.28)
W ( 2) 
î s 
The matrices [ I a ] and [ I s ] are, respectively, the (naxna) and (nsxns) unitary
~
[ ] ~
matrices. The (naxna) matrix K a and the (nsxns) matrix K s are diagonal [ ]
matrices,
 0  0
[ ]
~ 1 
Ka = 2 
c
ω a2 ,
 [ ]
~ 1 
Ks = 2 
c
ω s2 .
 (F.29)
 0   0 
The elements on row a and column s (a=1..na, s=1..ns) in the (naxns) matrices
[ ]
~ ~
[ ]
C1 and C2 are, respectively,
[ ]as [ ]
( 1) ( 2)
~ c. Cas ~ c. Cas
C1 =− , C2 = . (F.30)
Λa as Λa
The elements on row s and column a (s=1..ns, a=1..na) in the (nsxna) matrices
[ ]
~ ~
[ ]
C3 and C4 are, respectively,
[C~3 ]sa = cρ. m0 .sC. Λass , [C~4 ]sa = − cρ. 0m.sC. asΛ s .
( 1) ( 2)
(F.31)
The predictions for the undamped natural eigenfrequencies and mode shapes
of the double-panel partition can be obtained from the following eigenvalue
problem,
 0 I  X X 
− K
~ ~ .   = λ .   . (F.32)
 −C  î Y  î Y
400 Appendix F
The resulting eigenvalues consist of pairs of imaginary values with opposite

signs. Each of the (na+2ns) imaginary eigenvalues with negative signs
corresponds with an undamped natural eigenfrequency ω c of the double-
panel partition, according to the relation (F.25). Each part { X} in the
corresponding eigenvector consists of the modal participation factors of the
uncoupled acoustic and uncoupled structural modes in the associated mode
shape.
F.2.3. Wave model
According to the discussion in chapter 4 of the application of the wave

based prediction technique for three-dimensional coupled vibro-acoustic
problems, an accurate prediction model for the considered double-panel
partition can be obtained by approximating the cavity pressure in terms of
the following acoustic wave function expansion,
p( x, y, z ) ≈
n a1 na 2
− jk z ,a1a2 z jk z ,a1a2 z
∑ ∑ cos( k xa1 x ) cos(k ya 2 y)( pa1 a 2 ,1. e + pa 1 a 2 , 2 . e ),
a1 = 0 a2 = 0
(F.33)
with
aπ a π
k xa 1 = 1 , k ya 2 = 2 , k z, a1 a 2 = k 2 − k xa
2
1
− k ya
2
2
. (F.34)
Lx Ly
The lower and upper panel normal displacements are approximated as
w1 ( x, y ) ≈ ws( 1) ( x, y ) + w F ( x, y ) + wa( 1) ( x, y ) ,
(F.35)
w 2 ( x, y ) ≈ w s( 2) ( x, y ) + w a( 2) ( x, y ) .
(F.36)
The structural wave function expansions ws(i) (x, y) (i = 1,2) are
(w )
n s1
w s(i ) ( x, y ) = ∑ s 1 ,1 .Ψ s 1 ,1 ( x , y ) + w s 1 , 2 .Ψ s 1 , 2 ( x , y )
(i ) (i )

s1 = 0
(F.37)
(w )
n s2
+ ∑ s 2 ,1 .Ψ s 2 ,1 ( x , y ) + w s 2 , 2 .Ψ s 2 , 2 ( x , y )
(i ) (i )
s2 = 0
where
− jk ys1 , 1 . y
Ψ s1 ,1 ( x, y ) = cos( k xs 1, 1 . x ). e ,
− jk ys1 , 2 . y
Ψ s1 , 2 ( x, y ) = cos( k xs1, 2 . x ). e ,
(F.38)
− jk xs2 , 1 . x
Ψ s 2 ,1 ( x , y ) = e .cos( k ys 2, 1. y ),
− jk xs2 , 2 . x
Ψ s 2 , 2 ( x, y ) = e .cos( k ys 2, 2 . y ),
with
2
sπ s π
( k xs 1 ,1 , k ys 1 ,1 ) = ( 1 ,± kb2 −  1  ),
Lx  Lx 
2
sπ s π
( k xs 1 , 2 , k ys 1 , 2 ) = ( 1 ,± j k b2 +  1  ),
Lx  Lx 
(F.39)
2
s π s π
( k xs 2 ,1 , k ys 2 ,1 ) = ( ± kb2 −  2  , 2 ),
 Ly  Ly
2
s π s π
( k xs 2 , 2 , k ys 2 , 2 ) = ( ± j k b2 +  2  , 2 ).
 Ly  L y
The particular solution function for the mechanical point force excitation is
jF
w F ( x, y ) = − [ H 0( 2) ( kb r ) − H 0( 2) ( − jk b r )]

(F.40)
8 kb2 D
with
r = ( x − x F ) 2 + ( y − yF ) 2 . (F.41)
The particular solution functions for the cavity pressure loading are
n a1 n a2
cos( k xa 1 x )cos( k ya 2 y )( pa 1 a 2 ,1 + pa 1 a 2 , 2 )
w a( 1) ( x, y ) = − ∑ ∑

, (F.42)
a1 = 0 a 2 = 0
2
D[( k xa 1
+ k ya
2
2
) 2 − kb4 ]
402 Appendix F
w a( 2) ( x , y ) =
− jk z ,a1a2 L z jk z ,a1a2 L z
n a1 n a2
cos( k xa 1 x ) cos( k ya 2 y )( pa 1 a 2 ,1. e + pa 1 a 2 , 2 . e )
∑ ∑ .
a1 = 0 a2 = 0
2
D[( k xa 1
+ k ya
2
2
) 2 − kb4 ]
(F.43)
The acoustic and structural wave function contributions in the expansions

(F.33) and (F.37) result from a weighted residual formulation of the
boundary conditions (F.5) and (F.7). Note that, besides the Helmholtz
equation (F.1), the proposed cavity pressure expansion (F.33) also satisfies a
priori the rigid wall boundary conditions (F.6).
Due to the frequency dependence of the acoustic and structural wave
functions, the resulting weighted residual wave model is frequency
dependent and the predictions for the undamped natural eigenfrequencies
and mode shapes of the double-panel partition cannot be obtained from a
standard eigenvalue problem. However, by calculating the forced response
of the undamped double-panel partition, the resulting resonance frequencies
and corresponding operational deflection shapes are very close to the exact
modal parameters, provided that the resonance frequencies are well
separated from each other.
F.3. Identification of the modal properties

Although the energy in each vibro-acoustic mode is distributed over both
the acoustic and structural components, in most coupled systems, this
energy is still predominantly stored in either the acoustic or structural
components (PAN and BIES (1990)). Therefore, the vibro-acoustic modes of
a double-panel partition may be classified into panel-controlled modes, for
which most of the energy is stored in the panel displacement fields, and
cavity-controlled modes, for which most of the energy is stored in the cavity
pressure field.
Several references can be found, which describe only the effect of a cavity
on the lowest panel-controlled mode in e.g. a cavity-backed panel (DOWELL
and VOSS (1963), SGARD and ATALLA (1997)) and a double-panel partition
(RAJALINGHAM et al. (1994)). This section gives a more general discussion
of the fluid-structure coupling effects, governing the low-frequency panel-
controlled and cavity-controlled modes of double-panel partitions. Some
modal properties are identified, based on a very small version of the
analytical model, described in section F.2.2. Since the simplifications, made

in this small analytical model, imply a loss of accuracy, the validity of the
conclusions, drawn from this simple model, is verified with the accurate
prediction results of the wave model, described in section F.2.3.
F.3.1. Panel-controlled modes
In each panel-controlled mode, the modal upper and lower panel

displacements have the same patterns, which are either out-of-phase or in-
phase, since the plane z=Lz/2 (see figure F.1) is a plane of symmetry of the
double-panel partition.
F.3.1.1. Modal properties of the out-of-phase modes

For each vibro-acoustic mode, all uncoupled panel and cavity modes, which
belong to the same group of mutually interacting modes as listed in table
F.1, contribute to the coupled modal panel displacements and cavity
pressure. To simplify the analytical model, it is assumed that, for the panel-
controlled modes, only one uncoupled panel mode with annotation (u,v) (see
(F.21)) and uncoupled eigenfrequency ωs contributes to the coupled modal
upper and lower panel displacement and that two uncoupled cavity modes
with annotations (l1,m1,n1) and (l2,m2,n2) (see (F.20)) and uncoupled
eigenfrequencies ωa1 and ωa2, which have a non-zero modal coupling factor
with the considered panel mode, contribute to the coupled modal cavity
pressure field. With these assumptions, the analytical model, as defined in
(F.15)-(F.17), becomes
(ω 2 − ω a21 ). Λ a 1. Pa1 = ρ 0 c 2ω 2 (Ws( 1) Ca( 1)s − Ws( 2) Ca( 2s) ) , (F.44)

1 1
(ω 2 − ω a2 2 ). Λ a 2 . Pa 2 = ρ 0 c 2ω 2 (Ws( 1) Ca( 1)s − Ws( 2) Ca( 2)s ) , (F.45)

2 2
ms (ω s2 − ω 2 ). Λ s . Ws( 1) = − Pa 1Ca( 1)s − Pa 2Ca( 1)s , (F.46)

1 2
ms (ω s2 − ω 2 ). Λ s . Ws( 2) = Pa 1Ca( 2s) + Pa 2Ca( 2)s . (F.47)

1 2
For the case of an out-of-phase panel-controlled mode, the contributions of

the considered panel mode to the coupled modal upper and lower panel
displacement are opposite ( Ws(1) = − Ws(2) = Ws ). Since the z-component of
404 Appendix F
1 ∂p
the modal fluid displacement field ( ) in an out-of-phase mode is
ρ0 ω 2 ∂z
anti-symmetric with respect to plane z=Lz/2, two contributing cavity modes
with even mode numbers n1 and n2 are used in the analytical model. As a
result, according to (F.22), the modal coupling factors between these cavity
modes and the upper and lower panel mode are the same
( Ca(1)s = Ca(2)s = C1 and Ca(1)s = Ca(2)s = C2 ). With these considerations and
1 1 2 2
with the substitution of expressions (F.44) and (F.45) for the cavity mode
contributions Pa1 and Pa2 into the equations (F.46) and (F.47), one can see
that the two latter equations become identical,
C2 ω2 C22 ω2
Ws .[ ms .(ω 2 − ω s2 ). Λ s ] = Ws .[ 2K ( 1 . 2 + . )] ,
Λ a1 ω − ω a21 Λ a 2 ω 2 − ω a2 2
(F.48)
where K= ρ0 c 2 is the bulk modulus of the fluid.
eigenfrequencies of out-of-phase panel-controlled modes

To get a non-trivial solution (Ws ≠ 0) , the expressions between square
brackets in both sides of (F.48) must be identical. The solution of this
characteristic equation yields the coupled eigenfrequency ω c of the out-of-
phase panel-controlled mode.
The characteristic equation (F.48) can be represented graphically. In figure
F.2, the left-hand side of this equation, denoted by [1], is plotted against the
right-hand side, denoted by [2]. The three intersections between both curves
yield the three coupled eigenfrequencies, associated with the simplified
analytical model. Two eigenfrequencies are associated with the two
uncoupled cavity modes in the model; the third one represents the coupled
eigenfrequency ωc of the out-of-phase panel-controlled mode, denoted by
. Based on the two uncoupled acoustic eigenfrequencies ωa1 and ωa2 and
frequency Ω, at which the right-hand side of (F.48) becomes zero, i.e.
C22 Λ a 1ω a21 + C12 Λ a 2ω a2 2

Ω2 = , (F.49)
C22 Λ a1 + C12 Λ a 2
four different cases may be distinguished:

Figure F.2 : characteristic equation for the eigenfrequency ωc

of a panel-controlled mode
case 1 : ω s < ω a1 < Ω < ω a2 (figure F.2a) :

both cavity modes have an added mass effect on the panel mode, resulting
in a decrease of the coupled eigenfrequency ωc in comparison with the
uncoupled panel eigenfrequency ωs,
case 2 : ω a1 < ω s < Ω < ω a2 (figure F.2b) :

the stiffening effect of the first cavity mode predominates the mass effect
of the second cavity mode, resulting in an increase of the coupled
eigenfrequency ωc in comparison with the uncoupled panel
eigenfrequency ωs,
case 3 : ω a1 < Ω < ω s < ω a2 (figure F.2c) :

the mass effect of the second cavity mode predominates the stiffness
effect of the first cavity mode, resulting in a decrease of the coupled
406 Appendix F
eigenfrequency ωc in comparison with the uncoupled panel

eigenfrequency ωs,
case 4 : ω a1 < Ω < ω a2 < ω s (figure F.2d) :

both cavity modes have a stiffening effect on the panel mode, resulting in
an increase of the coupled eigenfrequency ωc in comparison with the
uncoupled panel eigenfrequency ωs.
mode shapes of the out-of-phase panel-controlled modes

According to (F.48), the effect of the cavity on a panel mode is mainly
determined by the cavity modes, which have a large modal coupling factor
C with the considered panel mode and whose uncoupled acoustic
eigenfrequencies ωa are close to the uncoupled panel eigenfrequency ωs. It
follows from (F.22) that the magnitude of the modal coupling factor C with
a panel mode (u,v) is the largest for the cavity modes (l,m,n) with
l = u ± 1 and m = v ± 1 and that it is independent of the mode number n. The
acoustic eigenfrequency ωa of a cavity mode Φa (see (F.20)), is
K lπ 2 mπ 2 nπ 2
ωa = [( ) + ( ) + ( ) ], (l, m,n = 0,1,...) . (F.50)
ρ 0 Lx Ly Lz
Since the cavity thickness Lz in most double-panel partitions is substantially

smaller than the dimensions Lx and Ly and since the low-frequency structural
wavenumber kb is usually significantly larger than the low-frequency
acoustic wavenumber k, the eigenfrequencies ωa of the cavity modes with
large mode numbers n are too high, compared with the low-frequency panel
eigenfrequencies ωs, to significantly affect the low-frequency panel-
controlled modes. Since the out-of-phase panel-controlled modes result
from the interaction of the considered panel mode with the cavity modes
with an even mode number n, the cavity pressure field for the low-frequency
out-of-phase panel-controlled modes is dominated by the cavity modes with
the lowest possible even mode number, i.e. n=0. Hence, the pressure
variation in the thickness direction (z-direction) of the double-panel partition
is fairly small. Moreover, the frequency difference between the
eigenfrequency ωs of a certain panel mode (u,v) and the eigenfrequencies ωa
of the cavity modes (l,m,n)=(u±1,v±1,0), which have the largest modal
coupling factor with the panel mode, usually becomes high, especially for
the higher mode numbers u and/or v. As a result, lower-order cavity modes
(i.e. l and m smaller compared with u resp. v), which have a smaller modal
coupling factor, but whose eigenfrequencies ωa are much closer to ωS, can
have a substantial contribution to the coupled modal pressure field of the
low-frequency out-of-phase panel-controlled modes.
As stated by CRAGGS (1971a), the overall stiffening or added mass effect of
the cavity on the panels is not only revealed by the increase or decrease of
the coupled eigenfrequencies, compared with the uncoupled panel
eigenfrequencies, but also by the phase relationship between the modal
pressure and modal panel displacements along the panel-cavity interfaces,
or, in general, by the work of the cavity on the panels. If the cavity has a
pure stiffening effect on the panel (case 4), a positive modal pressure occurs
together with a modal panel displacement into the cavity, while a negative
modal pressure occurs together with a modal panel displacement away from
the cavity along the whole panel-cavity interfaces, indicating the restoring
force effect of the cavity on the panel. If the cavity has a pure added mass
effect on the panel (case 1), a positive modal pressure occurs together with a
panel displacement away from the cavity, while a negative modal pressure
occurs together with a modal panel displacement into the cavity along the
whole panel-cavity interfaces. When the coupling interaction is a
combination of both stiffness and inertial effects, the phase relations
between the modal panel displacements and modal cavity pressure along the
panel-cavity interfaces are such that the work of the cavity on the panels,
which is defined as
∫ p. w dΩ (F.51)
Ωs
with the modal panel displacement w, being positive in the direction away
from the cavity, is negative, when the stiffness effects predominate the
inertial effects (case 2), while a positive work indicates the domination of
the inertial effects (case 3).
F.3.1.2. Modal properties of the in-phase modes

Again, an analytical model is considered, which consists of one uncoupled
panel mode for the expansion of the upper and lower panel displacements
and two uncoupled cavity modes for the expansion of the cavity pressure
field.
For the case of an in-phase panel-controlled mode, the contributions of the
considered panel mode to the coupled modal upper and lower panel
408 Appendix F
displacement are the same ( Ws(1) = Ws(2) = Ws ). Since the z-component of

the modal fluid displacement field in an in-phase mode is symmetric with
respect to plane z=Lz/2, two contributing cavity modes with odd mode
numbers n1 and n2 are used in the analytical model. As a result, according to
(F.22), the modal coupling factors between these cavity modes and the
upper and lower panel mode are opposite ( Ca(1)s = −Ca(2)s = C1
1 1
and Ca(1)s = −Ca(2)s = C2 ). With these considerations, the resulting

2 2
characteristic equation for the coupled eigenfrequency ωc of an in-phase
panel-controlled mode is identical to the characteristic equation (F.48) for
the out-of-phase panel-controlled modes.
eigenfrequencies of the in-phase panel-controlled modes

Since the cavity mode numbers n1 and n2 are odd, the smallest possible
uncoupled eigenfrequency ωa is the eigenfrequency of the (0,0,1) mode.
Since the cavity thickness Lz in most double-panel partitions is substantially
smaller than the dimensions Lx and Ly and since the low-frequency structural
wavenumber kb is usually significantly larger than the low-frequency
acoustic wavenumber k, even the smallest possible eigenfrequency ωa is
already substantially larger than the eigenfrequencies ωs of the low-
frequency panel modes. Consequently, the characteristic equation (F.48)
only applies in its form, as shown in figure F.2a (case 1). All cavity modes
have an added mass effect on the considered panel mode, resulting in a
decrease of the coupled eigenfrequency ωc in comparison with the
uncoupled eigenfrequency ωs. Since the frequency differences ω a2 − ω s2 of
the dominating cavity modes are significantly larger for the in-phase modes
compared to the out-of-phase modes, the coupling effect (see (F.48)) and the
subsequent frequency shifts ω c2 − ω s2 are usually smaller for the low-
frequency in-phase panel-controlled modes.
mode shapes of the in-phase panel-controlled modes

Since the uncoupled eigenfrequencies ωa of all cavity modes are larger than
the uncoupled eigenfrequencies ωs of the low-frequency panel modes and
since the magnitude of the modal coupling factor C is independent of the
mode number n, the modal pressure field in the low-frequency in-phase
panel-controlled modes is dominated by the cavity modes with the lowest
possible odd mode number, i.e. n=1. In contrast with the out-of-phase
modes, the cavity modes which have the largest modal coupling factor with
the considered panel mode usually dominate the modal pressure field in the
in-phase modes, since the modal coupling factor varies strongly with mode
numbers l and m, while the corresponding frequency differences ω a2 − ω s2
vary only slightly. The latter is due to the fact that the eigenfrequencies ωa
(see (F.50)) of the cavity modes (l,m,1) are mainly determined by the small
cavity thickness Lz, so that a variation on the mode numbers l and m only
slightly affects the eigenfrequencies ωa. As a result, the modal pressure field
in the low-frequency in-phase panel-controlled modes is mainly determined
by the cavity modes (l,m,n)=(u±1, v±1,1).
The mass-loading effect of the cavity on the panels is not only revealed by
the decrease of the coupled eigenfrequencies, compared with the uncoupled
panel eigenfrequencies, but also by the modal fluid displacement patterns
and by the phase relationship between the modal pressure and modal panel
displacements along the panel-cavity interfaces. Since the modal pressure
field is dominated by the cavity modes with mode numbers n=1, the z-
component of the cavity fluid displacement has the same sign in most parts
of the cavity, so that the cavity fluid can be regarded as a lumped mass,
mainly moving in phase with the upper and lower panels. As mentioned in
section F.3.1.1., these inertial effects yield a positive modal pressure,
occurring together with a panel displacement away from the cavity, while a
negative modal pressure occurs together with a modal panel displacement
into the cavity along the whole panel-cavity interfaces.
F.3.2. Cavity-controlled modes
As for the case of the panel-controlled modes, a simplified analytical model

for the cavity-controlled modes is built. It is assumed that only one
uncoupled cavity mode with annotation (l,m,n) and eigenfrequency ωa
contributes to the coupled modal cavity pressure field and that two
uncoupled panel modes with annotations (u1,v1) and (u2,v2) and
eigenfrequencies ωs1 and ωs2, which have a non-zero modal coupling factor
with the considered cavity mode, contribute to the coupled modal upper and
lower panel displacements.
Due to the symmetry of the double-panel partition, the panels vibrate out-of-
phase or in-phase, depending on whether the cavity pressure field is
controlled by a cavity mode with an even or an odd mode number n. By
applying the same procedure as described in section F.3.1.1., the
410 Appendix F
characteristic equation for the coupled eigenfrequencies of both out-of-

phase and in-phase cavity-controlled modes becomes
Λ C12 ω2 C22 ω2
Pa .[(ω 2 − ω a2 ). a ] = Pa .[ 2( . 2 + . )] .
K ms Λ s 1 ω − ω s21 ms Λ s 2 ω 2 − ω s22
(F.52)
eigenfrequencies of the cavity-controlled modes
Again, the non-trivial solution (Pa ≠ 0) of the characteristic equation can be
represented graphically (see figure F.3).
Figure F.3 : characteristic equation for the eigenfrequency ωc

of a cavity-controlled mode
Based on the two uncoupled structural eigenfrequencies ωs1 and ωs2 and
frequency Ω, at which the right-hand side of (F.52) becomes zero, i.e.
C22 Λ s 1ω s21 + C12 Λ s 2ω s22

Ω2 = , (F.53)
C22 Λ s 1 + C12 Λ s 2
four different cases may be distinguished:
case 1 : ω a < ω s1 < Ω < ω s2 (figure F.3a) :

both panel modes have an added mass effect on the cavity mode, resulting
in a decrease of the coupled eigenfrequency ωc in comparison with the
uncoupled eigenfrequency ωa,
case 2 : ω s1 < ω a < Ω < ω s2 (figure F.3b) :

the stiffening effect of the first panel mode predominates the mass effect
of the second panel mode, resulting in an increase of the coupled
eigenfrequency ωc in comparison with the uncoupled eigenfrequency ωa,
case 3 : ω s1 < Ω < ω a < ω s2 (figure F.3c) :

the mass effect of the second panel mode predominates the stiffness effect
of the first panel mode, resulting in a decrease of the coupled
eigenfrequency ωc in comparison with the uncoupled eigenfrequency ωa,
case 4 : ω s1 < Ω < ω s2 < ω a (figure F.3d) :

both panel modes have a stiffening effect on the cavity mode, resulting in
an increase of the coupled eigenfrequency ωc in comparison with the
uncoupled eigenfrequency ωa.
mode shapes of the cavity-controlled modes

Due to the small cavity thickness Lz in most double-panel partitions, only
cavity modes with mode number n=0 occur at low frequencies, so that the
low-frequency cavity-controlled modes have an out-of-phase panel
displacement. The contributions of the uncoupled panel modes to the
coupled modal panel displacements depend on both the magnitude of their
modal coupling factor with the considered cavity mode and the frequency
difference between their uncoupled structural eigenfrequencies ωs and the
uncoupled acoustic eigenfrequency ωa. In contrast with the panel-controlled
modes, it is not possible anymore to indicate in general which uncoupled
panel modes have the largest contributions to the panel displacement of the
cavity-controlled modes.
412 Appendix F
F.3.3. Wave model validation
The above identified modal properties of double-panel partitions result from

an analytical model, in which only a very few uncoupled modes are used for
the expansion of the dynamic field variables. Since these simplifications
imply a loss of accuracy, the validity of the identified modal properties is
verified with the accurate prediction results of a wave model, described in
section F.2.3.
For these validation purposes, a wave model, consisting of 306 acoustic
wave functions (na1=16, na2=8) for the pressure expansion (F.33) and 56
structural wave functions (ns1=8, ns2=4) for the panel displacement
expansions (F.37), is built for a double-panel partition (Lx=1m, Ly=0.5m,
Lz=0.15m) with two aluminium (E=70.109 N/m2, ν=0.3, ρs=2790 kg/m )
3
panels of thickness t=2mm and an air-filled (c=340 m/s, ρ0=1.225 kg/m3)

cavity. The panels are assumed to be undamped, η=0. A point force F
(magnitude 1N) is applied at (xF,yF)=(0.25*Lx,0.25*Ly) on the lower panel.
As mentioned in section F.2.3., the eigenfrequencies and mode shapes are
extracted from the undamped forced response calculations.
Table F.2 lists the low-frequency (< 300 Hz) coupled eigenfrequencies of
the considered double-panel partition, obtained with the accurate wave
model. Table F.3 lists the corresponding coupled eigenfrequencies, obtained
from the analytical model. According to the characteristic equation (F.48),
the modal model for each panel-controlled mode consists of one uncoupled
panel mode and the two uncoupled cavity modes with the largest factors
C 2 / ( Λ a (ω a2 − ω s2 )) . In a similar way, the two uncoupled panel modes for
the cavity-controlled modes are selected based on the magnitude of
C 2 / ( Λ s (ω s2 − ω a2 )) . The resulting modal base of uncoupled panel and
cavity modes and their corresponding uncoupled eigenfrequencies are also
listed in table F.3 for each mode of the double-panel partition.
Table F.2 confirms the findings from the analytical model that the
eigenfrequencies of the out-of-phase panel-controlled modes may increase
or decrease with respect to the corresponding uncoupled panel
eigenfrequencies, while the in-phase panel-controlled modes have always a
frequency decrease, due to the added mass effect of the cavity.
The comparison of tables F.2 and F.3 reveals also that the analytical model
correctly indicates whether the fluid-structure coupling interaction causes a
frequency increase or decrease, based on the four distinguished cases (the
appropriate case for each coupled mode is listed in table F.3). Only for the
mode with the smallest frequency shift, i.e. the out-of-phase mode,
controlled by the (1,3) panel mode, the analytical model yields a small
frequency increase, while the actual coupled eigenfrequency is slightly
smaller than the uncoupled eigenfrequency of the (1,3) panel mode.
uncoupled coupled
panel mode (u,v) in-phase out-of-phase
(odd,odd) fs (Hz) fc (Hz) (fc-fs)/fs (%) fc (Hz) (fc-fs)/fs (%)
(1,1) 47.4 47.1 -0.6 71.8 +51.5
(3,1) 86.4 85.8 -0.7 94.4 +9.3
(5,1) 168.4 167.4 -0.6 167.8 -0.4
(1,3) 237.9 236.5 -0.6 237.6 -0.1
(3,3) 274.7 273.4 -0.5 271.9 -1.0
(7,1) 290.6 289.2 -0.5 288.1 -0.9
(even,odd) fs (Hz) fc (Hz) (fc-fs)/fs (%) fc (Hz) (fc-fs)/fs (%)
(2,1) 61.4 61.0 -0.6 57.2 -6.8
(4,1) 122.2 121.4 -0.6 118.2 -3.3
(6,1) 224.6 223.4 -0.5 225.3 +0.3
(2,3) 251.6 250.2 -0.5 252.8 +0.5
(odd,even) fs (Hz) fc (Hz) (fc-fs)/fs (%) fc (Hz) (fc-fs)/fs (%)
(1,2) 123.5 122.7 -0.6 120.6 -2.3
(3,2) 160.7 159.8 -0.6 158.6 -1.3
(5,2) 238.7 237.5 -0.5 237.8 -0.4
(even,even) fs (Hz) fc (Hz) (fc-fs)/fs (%) fc (Hz) (fc-fs)/fs (%)
(2,2) 137.2 136.3 -0.7 134.7 -1.8
(4,2) 194.5 193.4 -0.6 192.5 -1.0
(6,2) 293.2 291.8 -0.5 290.8 -0.8
cavity mode in-phase out-of-phase
(l,m,n) fa (Hz) fc (Hz) (fc-fa)/fa (%) fc (Hz) (fc-fa)/fa (%)
(1,0,0) 170.0 180.2 +6.0
Table F.2 : uncoupled and coupled eigenfrequencies (f=ω/2π) of the considered

double-panel partition
panel-controlled modes
unc. in-phase out-of-phase
(u,v) (l,m,n) (l,m,n) Ω (Hz) fc (Hz) (l,m,n) (l,m,n) Ω (Hz) fc (Hz)
fs (Hz) fa1(Hz) fa2(Hz) case fa1(Hz) fa2(Hz) case
(1,1) (0,0,1) (2,0,1) 1174.3 47.2 (0,0,0) (2,0,0) 307.5 88.2
47.4 1133.3 1183.2 case 1: ωs < ωa1 0 340 case 2:ωa1 < ωs < Ω
414 Appendix F
(3,1) (2,0,1) (4,0,1) 1276.6 86.0 (0,0,0) (2,0,0) 124.3 88.3

(5,1) (4,0,1) (6,0,1) 1446.7 167.9 (0,0,0) (4,0,0) 167.7 168.3
168.4 1321.7 1524.7 case 1: ωs < ωa1 0 680 case 3:Ω < ωs < ωa2
(1,3) (0,2,1) (0,4,1) 1632.6 237.2 (0,0,0) (0,2,0) 248.6 238.0
(3,3) (2,2,1) (4,2,1) 1446.4 274.1 (2,0,0) (2,2,0) 421.2 272.4
274.7 1364.7 1486.4 case 1: ωs < ωa1 340 760.3 case 1: ωs < ωa1
(7,1) (6,0,1) (8,0,1) 1669.4 289.9 (2,0,0) (6,0,0) 432.2 288.9
290.6 1524.7 1770.3 case 1: ωs < ωa1 340 1020 case 1: ωs < ωa1
(2,1) (1,0,1) (3,0,1) 1217.9 61.1 (1,0,0) (3,0,0) 446.0 56.0
(4,1) (3,0,1) (5,0,1) 1353.8 121.8 (1,0,0) (3,0,0) 265.0 118.2
(6,1) (5,0,1) (7,0,1) 1552.8 224.0 (1,0,0) (5,0,0) 302.1 225.7
(2,3) (1,2,1) (3,2,1) 1394.9 250.9 (1,0,0) (1,2,0) 301.2 252.6
251.6 1332.6 1416.7 case 1: ωs < ωa1 170 700.9 case 2:ωa1 < ωs < Ω
(1,2) (0,1,1) (0,3,1) 1442.2 123.0 (0,1,0) (2,1,0) 458.5 120.3
(3,2) (2,1,1) (4,1,1) 1321.1 160.2 (0,1,0) (2,1,0) 362.0 158.8
(5,2) (4,1,1) (6,1,1) 1486.2 238.1 (0,1,0) (4,1,0) 379.1 237.6
(2,2) (1,1,1) (3,1,1) 1264.5 136.6 (1,1,0) (3,1,0) 560.8 134.6
137.2 1195.4 1288.5 case 1: ωs < ωa1 380.1 612.9 case 1: ωs < ωa1
(4,2) (3,1,1) (5,1,1) 1395.8 193.9 (1,1,0) (3,1,0) 431.1 192.9
(6,2) (5,1,1) (7,1,1) 1589.6 292.5 (1,1,0) (5,1,0) 454.8 291.9
cavity-controlled modes
unc. in-phase out-of-phase
(l,m,n) (u,v) (u,v) Ω (Hz) fc (Hz) (u,v) (u,v) Ω (Hz) fc (Hz)
fa (Hz) fs1 (Hz) fs2 (Hz) case fs1 (Hz) fs2 (Hz) case
(1,0,0) (2,1) (4,1) 115.7 189.4
170 61.4 122.2 case 4: ωs2 < ωa
Table F.3 : coupled eigenfrequencies (f=ω/2π) from the analytical model

The findings from the analytical model concerning the contributions of the
different cavity modes to the modal pressure field in the low-frequency
panel-controlled modes are also confirmed by the wave model predictions.
The modal pressure fields in the out-of-phase and in-phase panel-controlled
modes consist mainly of the uncoupled cavity modes with mode numbers
n=0 and n=1, respectively. Figures F.4 and F.5 show, for instance, the wave
model prediction results for the forced responses at 71.8Hz and 47.1 Hz,
which are very close to the eigenfrequencies of, respectively, the out-of-
phase and in-phase modes, controlled by the (1,1) panel mode.
Figure F.4 : forced response at 71.8 Hz;

contour plots of the instantaneous pressure in planes y=yF (a) and x=xF (b) and
instantaneous structural and fluid displacement in planes y=yF (c) and x=xF (d)
416 Appendix F

These figures confirm that the modal pressure field of the (1,1) out-of-phase
and in-phase panel-controlled modes consist mainly of, respectively, the
(0,0,0)/(2,0,0) and (0,0,1)/(2,0,1) cavity modes, as indicated by the
analytical model (see table F.3). Note also that the z-component of the
modal fluid displacement is indeed anti-symmetric with respect to the mid-
plane of the cavity for the out-of-phase mode (figure F.4) and symmetric for
the in-phase mode (figure F.5). Note that the lines of symmetry in the
operational deflection shapes on these figures are not entirely preserved, due
to the non-symmetric location of the force excitation of the considered
double-panel partition.
The wave model results confirm also the phase relationship between the
modal pressure and modal panel displacements at the panel-cavity
interfaces. For the (1,1) out-of-phase panel-controlled mode, for instance,
the increased eigenfrequency results mainly from the stiffening effect of the

(0,0,0) cavity mode, which predominates the inertial effect of the (2,0,0)
cavity mode (see table F.3). Consequently, most parts of the panels - at least
those with the largest displacement amplitudes - move into the cavity at the
time the pressures along the panel-cavity interfaces are positive (see figure
F.4). On the contrary, the (1,1) in-phase panel-controlled mode is mainly
determined by the inertial effects of the (0,0,1) and (2,0,1) cavity modes.
Consequently, the upper and lower panel move into and away from the
cavity at the time the pressure along the corresponding interface is,
respectively, negative and positive (see figure F.5). Note also that the
418 Appendix F

coupling effect and subsequent frequency shift is substantially larger for the
(1,1) out-of-phase panel-controlled mode than for the corresponding in-
phase mode (see table F.2). This can be derived also from a comparison of
their modal patterns (figures F.4 and F.5). The modal pressure amplitudes
are substantially larger in the out-of-phase mode, while the amplitudes of
the modal panel displacements are similar. The fluid displacement patterns
in figure F.5 illustrate also that, for the in-phase panel-controlled modes, the
cavity fluid can be regarded as a lumped mass, mainly moving in-phase with
the upper and lower panels.
For the out-of-phase modes, controlled by panel modes with higher mode
numbers u and/or v, the cavity modes (l,m,n)=(u±1,v±1,0), which have the
largest modal coupling factors with a certain panel mode (u,v), are not
necessarily the most contributing modes; instead, the cavity modes with
lower mode numbers l and m are more dominant. On the contrary, the cavity
pressure field for the in-phase modes is always dominated by the cavity
modes with the largest modal coupling factors, i.e. (l,m,n)=(u±1,v±1,1).
These findings from the analytical model are confirmed by the wave model
results. Figures F.6 and F.7, showing the forced responses, corresponding
with, respectively, the (6,2) out-of-phase and in-phase panel-controlled
modes, confirm, for example, that the cavity pressure field consists mainly
of, respectively, the (1,1,0)/(5,1,0) and (5,1,1)/(7,1,1) modes, as indicated in
table F.3.

instantaneous structural and fluid displacement in planes y=yF (a) and x=xF (b)
((c) : enlarged view on the instantaneous structural displacement in plane y=yF)
The analytical model yields also a good characterisation of the cavity-

controlled modes. According to this model (see table F.3), the coupled
eigenfrequency of the (1,0,0) cavity-controlled mode is larger than the
eigenfrequency of the uncoupled (1,0,0) cavity mode, mainly due to the
stiffening effect of the (2,1) and (4,1) panel modes, which dominate the out-
of-phase panel displacements. Again, the corresponding forced response,
obtained from the wave model, confirms these findings (see figure F.8).
420 Appendix F
F.4. Modal sensitivity

Although the dynamic behaviour of double-panel partitions results from a
complicated fluid-structure coupling interaction between the cavity and both
panels, the simple analytical model, described in section F.2.2., enables a
qualitative sensitivity analysis of the low-frequency vibro-acoustic modes to
changes in some physical parameters of the double-panel partition.
F.4.1. Cavity thickness
A modification of the cavity thickness Lz influences the eigenfrequencies ωa

(see (F.50)) and the modal scale factors Λa (see (F.23)) of the uncoupled
cavity modes. As a result, changes occur only in the right-hand side of the
characteristic equation (F.48) and in the left-hand side of characteristic
equation (F.52), which determine the coupled eigenfrequencies ωc.
out-of-phase panel-controlled modes

Since the modal pressure field in a low-frequency out-of-phase panel-
controlled mode is dominated by the uncoupled cavity modes with mode
number n=0, the eigenfrequencies ωa of these cavity modes are independent
of the cavity thickness. Only the factors Λa are sensitive to modifications of
the cavity thickness. Since these factors are proportional to Lz, factor Λa of
each cavity mode changes with the same extent, while the modal coupling
factors C and eigenfrequencies ωa remain unchanged. Consequently, the
modal coupling interaction of each cavity mode with the considered panel
mode - which is determined by the modal coupling factor C, factor Λa and
the difference between the uncoupled acoustic and uncoupled structural
eigenfrequencies (see right-hand side of (F.48)) - changes also with the
same extent, so that always the same cavity modes dominate the modal
pressure field in a low-frequency out-of-phase panel-controlled mode,
irrespective of the cavity thickness1. As a result, the same cavity modes
determine the right-hand side of the characteristic equation (F.48); only the
coefficients change. Figure F.9 shows the effect of a cavity thickness
modification on the coupled eigenfrequency of an out-of-phase panel-
controlled mode for all four cases, mentioned in section F.3.1.1. This figure
indicates that a decrease of the cavity thickness (Lz2<Lz1) intensifies the
1
provided that the modified cavity thickness Lz is still smaller, compared to the
panel dimensions Lx and Ly
fluid-structure coupling interaction, so that the frequency shifts ω c2 − ω s2

increase for all four cases.
Figure F.9 : sensitivity of the eigenfrequency ωc of an out-of-phase panel-controlled

mode to the cavity thickness (Lz2<Lz1) (solid : Lz1, dotted: Lz2)
in-phase panel-controlled modes

In the in-phase panel-controlled modes, the cavity has an added mass effect
on the panels, in that the cavity fluid can be regarded as a lumped mass,
mainly moving in phase with the upper and lower panel. Since a cavity
thickness reduction implies a reduction of the total fluid mass, the added
mass effect is also reduced. Hence, the frequency decrease of the coupled
eigenfrequency of an in-phase panel-controlled mode, compared with the
uncoupled eigenfrequency of the considered panel mode, becomes smaller,
when the cavity thickness is reduced.
cavity-controlled modes
The modal pressure in a low-frequency cavity-controlled mode is dominated
by an uncoupled cavity mode with mode number n=0. Consequently, as it is
422 Appendix F
the case for the out-of-phase panel-controlled modes, only the factor Λa
changes and the same uncoupled panel modes dominate the coupled modal
panel displacements, so that only the left-hand side of the characteristic
equation (F.52) changes. Figure F.10 shows the effect of a cavity thickness
modification on the coupled eigenfrequency of a cavity-controlled mode for
all four cases, mentioned in section F.3.2. This figure indicates that a
decrease of the cavity thickness (Lz2<Lz1) intensifies the fluid-structure
coupling interaction, so that the frequency shifts ω c2 − ω a2 increase for all
four cases.
Figure F.10 : sensitivity of the eigenfrequency ωc of a cavity-controlled mode

to the cavity thickness (Lz2<Lz1) (solid : Lz1, dotted: Lz2)
F.4.2. Fluid density
A modification of the fluid density ρ0 (without modification of the fluid bulk

modulus K=ρ0c2) influences the eigenfrequencies ωa of the uncoupled cavity
modes (see (F.50)). As a result, the modal coupling interaction between the
uncoupled panel and uncoupled cavity modes is also influenced, since one
of its determining factors is the difference between their uncoupled
eigenfrequencies. In contrast with a modification of the cavity thickness, the
coupling interaction with a certain panel mode is not changing with the
same extent for every cavity mode. Consequently, it is not ensured that
always the same uncoupled cavity modes dominate the modal pressures in
the panel-controlled modes. Figure F.11 shows the effect of a fluid density
modification on the coupled eigenfrequency of a panel-controlled mode,
assuming that the dominating uncoupled cavity modes do remain the same.
Figure F.11 : sensitivity of the eigenfrequency ωc of a panel-controlled mode

to the fluid density (ρ0,2<ρ0,1) (solid : ρ0,1, dotted: ρ0,2)
Provided that the an uncoupled panel mode still belongs to the same case,
defined in section F.3.1.1., after decreasing the fluid density (ρ0,2< ρ0,1), the
fluid-structure coupling effect and subsequent frequency shifts ω c2 − ω s2
may increase or decrease, depending on the case. When the two dominant
cavity modes have an added mass effect on the considered panel mode (case
424 Appendix F
1 : figure F.11a), which is the case, for example, for all low-frequency in-
phase panel-controlled modes, or when the mass effect of the first cavity
mode predominates the stiffening effect of the second cavity mode (case 3 :
figure F.11c), the decrease of the coupled eigenfrequency ωc with respect to
the uncoupled panel eigenfrequency ωs becomes smaller for a decreased
fluid density. When the two dominant cavity modes have a stiffening effect
on the considered panel mode (case 4 : figure F.11d) or when the stiffening
effect of the first cavity mode predominates the mass effect of the second
cavity mode (case 2 : figure F.11b), a decrease of the fluid density yields a
larger increase of the coupled eigenfrequency.
F.4.3. Wave model validation
The validity of the above identified modal sensitivities is verified with the
wave model results for two cases. In the first case, the thickness of the
double-panel partition, considered in section F.3.3., is reduced from 0.15m
to 0.10m. In the second case, the cavity of the double-panel partition is filled
with helium (ρ0=0.166 kg/m , c=1017 m/s) instead of air.
2 3
Table F.4 lists the low-frequency coupled eigenfrequencies of the double-

panel partition with reduced cavity thickness. As identified with the
analytical model, the comparison of tables F.2 and F.4 confirms that a cavity
thickness reduction amplifies the coupling effects for the low-frequency out-
of-phase panel-controlled modes and cavity-controlled modes, so that the
subsequent frequency shifts (both increases and decreases) become larger.
Only for the (1,3) out-of-phase panel-controlled mode, a slight frequency
decrease is converted to a slight increase and for the (5,1) out-of-phase
panel controlled mode, the slight eigenfrequency decrease becomes smaller.
The comparison of tables F.2 and F.4 confirms also the reduced frequency
decreases of the eigenfrequencies of the in-phase panel-controlled modes,
due to the reduced mass of the cavity fluid.
Table F.5 lists the low-frequency coupled eigenfrequencies of the double-
panel partition, filled with helium instead of air. Helium has a substantially
smaller density than air, but the bulk moduli K are similar for both fluids.
According to the analytical model, a reduction of the fluid density (without
modification of the bulk modulus) amplifies the stiffness effect (see cases 2
and 4 in figures F.11b and F.11d) or reduces the inertial effect (see cases 1
2
the cavity of a double glazing window is sometimes filled with helium
and 3 in figures F.11a and F.11c) of the cavity on the out-of-phase panel-
controlled modes, provided that, on the one hand, the assumption is correct
uncoupled coupled
(1,1) 47.4 47.2 -0.4 74.6 +57.4
(3,1) 86.4 86.0 -0.5 102.6 +18.7
(5,1) 168.4 167.6 -0.5 168.0 -0.2
(1,3) 237.9 236.8 -0.5 238.2 +0.1
(3,3) 274.7 273.6 -0.4 271.1 -1.3
(7,1) 290.6 289.4 -0.4 287.6 -1.0
(2,1) 61.4 61.1 -0.5 55.4 -9.8
(4,1) 122.2 121.6 -0.5 116.8 -4.4
(6,1) 224.6 223.6 -0.4 226.2 +0.7
(2,3) 251.6 250.5 -0.4 254.3 +1.1
(1,2) 123.5 122.8 -0.6 119.4 -3.3
(3,2) 160.7 160.0 -0.4 157.8 -1.8
(5,2) 238.7 237.6 -0.4 236.0 -1.1
(2,2) 137.2 136.5 -0.5 133.8 -2.5
(4,2) 194.5 193.7 -0.4 191.9 -1.3
(6,2) 293.2 292.0 -0.4 290.1 -1.1
cavity mode in-phase out-of-phase
(l,m,n) fa (Hz) fc (Hz) (fc-fa)/fa (%) fc (Hz) (fc-fa)/fa (%)
(1,0,0) 170.0 184.0 +8.2
Table F.4 : uncoupled and coupled eigenfrequencies (f=ω/2π) of the double-panel

partition with reduced cavity thickness
that the modal pressure is still dominated by the same uncoupled cavity
modes and that, on the other hand, the considered uncoupled panel mode
still belongs to the same case. Both conditions are satisfied for all uncoupled
panel modes, marked with an asterisk in table F.5. A comparison of the
eigenfrequencies of the out-of-phase modes, controlled by one of these
panel modes (see tables F.2 and F.5), confirms that all frequency increases
and decreases become, respectively, larger and smaller, due to a fluid
density reduction.
426 Appendix F
For the (5,1), (6,1) and (2,3) out-of-phase modes, the same cavity modes are
still dominating the modal cavity pressure, but the relative position of the
uncoupled panel frequency with respect to the uncoupled acoustic
eigenfrequencies is modified, so that a frequency increase becomes a
frequency decrease and vice versa.
uncoupled coupled
(1,1) * 47.4 47.3 -0.1 75.1 +58.5
(3,1) * 86.4 86.3 -0.1 98.6 +14.1
(5,1) 168.4 168.3 -0.06 169.5 +0.65
(1,3) * 237.9 237.7 -0.08 240.0 +0.88
(3,3) 274.7 274.5 -0.06 274.8 +0.02
(7,1) 290.6 290.4 -0.07 290.7 +0.03
(2,1) * 61.4 61.3 -0.08 60.8 -0.9
(4,1) * 122.2 122.1 -0.08 121.8 -0.3
(6,1) 224.6 224.4 -0.09 224.1 -0.2
(2,3) 251.6 251.4 -0.08 250.8 -0.3
(1,2) * 123.5 123.4 -0.08 123.1 -0.3
(3,2) * 160.7 160.6 -0.06 160.4 -0.2
(5,2) 238.7 238.6 -0.04 238.4 -0.1
(2,2) * 137.2 137.1 -0.07 136.8 -0.3
(4,2) * 194.5 194.4 -0.05 194.3 -0.1
(6,2) * 293.2 293.1 -0.04 293.0 -0.07
Table F.5 : uncoupled and coupled eigenfrequencies (f=ω/2π)

of the helium-filled double-panel partition (* : same type of modal model)
For the (5,2) out-of-phase mode, the (2,1,0) and (4,1,0) cavity mode are
dominant, instead of the (0,1,0) and (4,1,0) cavity mode. Since both cavity
modes still have an added mass effect, the panel eigenfrequency still
decreases. For the (3,3) and (7,1) out-of-phase modes, one of the dominant
cavity modes has also changed: the (0,0,0) rather than the (2,0,0) cavity
mode becomes dominant. Due to the stiffening effect of the (0,0,0) mode,
instead of the added mass effect of the (2,0,0), the panel eigenfrequencies no
longer decrease but increase. All these modifications on the coupling effects
and frequency shifts, identified from the analytical model, are confirmed by
the wave model predictions (see table F.5).
The comparison of tables F.2 and F.5 confirms also the reduced frequency
decreases of the eigenfrequencies of the in-phase panel-controlled modes,
due to the reduced mass of the cavity fluid.
F.5. Some remarks on the sound transmission properties

In order to correctly model the sound transmission through a double-panel
partition, the incident and transmitted sound fields and their interactions
with the lower and upper panel displacements should be taken into account
in a vibro-acoustic prediction model. Although the wave model, described in
section F.2.3., doesn’t take into account these sound fields, the ratio of the
total kinetic energy of the mechanically excited lower panel to the total
kinetic energy of the upper panel still renders at least a qualitative indication
of the sound transmission properties: the frequency bands, in which the
energy of the upper panel becomes substantially larger than the energy of
the lower panel, usually correspond with the frequency bands of poor sound
insulation of the double-panel partition (CRAIK (1996)).
Figure F.12 shows the kinetic energy ratio for a unit point force excitation at
(xF,yF) = (0.25*Lx,0.25*Ly) on the lower panel of the double-panel partition,
considered in section F.3.3. The kinetic energy levels of both panels result
from a forced response calculation of the wave model between 30 Hz and
300 Hz with a frequency step of 1 Hz. Due to the symmetry of the double
panel partition, the energy levels of both panels become identical at each
coupled eigenfrequency. Note that the energy ratio curve, shown in figure
F.12a, is not crossing the line of unit energy ratio at every coupled
eigenfrequency due to the finite frequency resolution.
In the higher frequency region, the fluid-structure coupling interaction is
fairly weak, so that the eigenfrequencies of the out-of-phase and in-phase
modes, controlled by a given panel mode, are close to each other. In the
narrow frequency region between these two eigenfrequencies, the forced
panel displacements are mainly determined by the modal displacement
patterns of these two modes. Since the main difference between both modal
patterns is the phase relationship between the upper and lower panel
displacement, while the modal displacement shapes are very similar, the
forced lower panel displacement becomes very small, due to destructive
interference between both modes, while their constructive interference
yields a large upper panel displacement. Figure F.13 illustrates this effect
428 Appendix F
for the forced response at 135.5 Hz, which lies between the eigenfrequencies
of the (2,2) out-of-phase and (2,2) in-phase mode. As a result, the energy
ratio curve shows sharp dips in each narrow frequency region between the
eigenfrequencies of an out-of-phase and corresponding in-phase mode,
indicating a poor sound insulation of the double-panel partition. In the other
frequency regions, the kinetic energy of the lower panel becomes
substantially larger than the upper panel energy, indicating a good sound
insulation. Since the latter frequency regions are usually much wider, the
sharp energy ratio dips almost disappear in a commonly used band
representation, even in the narrow bands of a 1/12 octave band
representation, as shown in figure F.12b.
Figure F.12 : ratio of the kinetic energy of the lower panel to the kinetic energy of
the upper panel ((a) linear, (b) 1/12 octave bands)
In the very low frequency region, the fluid-structure coupling interaction

induces larger frequency shifts, so that the differences between the
eigenfrequencies of corresponding out-of-phase and in-phase modes are
larger and the subsequent regions of small (<1) kinetic energy ratios are
wider. As a result, the energy ratio dips in the frequency regions (57.2 Hz -
61.0 Hz) and (85.8 Hz - 94.4 Hz) between the eigenfrequencies of,
respectively, the (2,1) and (3,1) out-of-phase and in-phase modes are still
present in a banded representation (see figure F.12b). The energy ratio
reaches the unit level at the eigenfrequencies of the (1,1) in-phase and out-
of-phase modes (47.1 Hz and 71.8 Hz), but no additional dip, besides the dip
associated with the (2,1) modes, occurs between these frequencies.

instantaneous structural and fluid displacement in planes y=yF (a) and x=xF (b)
This is due to the fact that the destructive and constructive interferences of
the (1,1) in-phase and out-of-phase modes, which would cause an additional
dip, are disturbed by the (2,1) out-of-phase and in-phase modes, since their
eigenfrequencies (57.2 Hz and 61.0 Hz) are located in the considered
frequency region (47.1 Hz - 71.8 Hz). When the point force is applied in the
middle of the lower panel (i.e. (xF,yF) = (0.5*Lx,0.5*Ly) instead of
(0.25*Lx,0.25*Ly)), the destructive and constructive interferences of the (1,1)
in-phase and out-of-phase modes are no longer disturbed by the (2,1) in-
phase and out-of-phase modes, since the latter modes cannot be excited by
the applied force. As a result, a strong dip occurs between the
eigenfrequencies of the (1,1) in-phase and out-of-phase mode, as shown in
figure F.14. Note also that the smaller number of excited modes yields a
larger energy ratio in the higher frequency region.
From these considerations on the kinetic energy levels of both panels, it may
be stated that double-panel partitions have good sound insulation properties,
except in the very low frequency range. In this respect, the insulation
properties of bounded double-panel partitions are similar to those of
unbounded double-panel partitions. For the latter, the poor low-frequency
sound insulation occurs in the region of the so-called mass-air-mass
resonance frequency, where a strong out-of-phase displacement of the
unbounded panels is maintained by the stiffness of the enclosed fluid (FAHY
430 Appendix F
(1985)). For the bounded case, however, the poor sound insulation no longer
results from one, but from several resonance phenomena.
the upper panel for a point force excitation at (xF,yF)=(0.25*Lx,0.25*Ly) (solid) and at
(xF,yF)=(0. 5*Lx,0. 5*Ly) (dashed) on the lower panel
Figure F.15 shows the effect on the kinetic energy ratio of the double-panel
partition, when the cavity thickness is reduced from 0.15 m to 0.10 m and
when the cavity is filled with helium instead of air. This figure clearly
indicates that a cavity thickness reduction yields a decrease of the kinetic
energy ratio, due to the increased fluid-structure coupling effect, as
indicated in section F.4. Changing the cavity fluid to helium reduces the
fluid-structure coupling effect and yields an increased kinetic energy ratio.
These sensitivities of bounded double-panel partitions are similar to the
ones, that were identified earlier for unbounded double-panel partitions
(BREKKE (1981), GÖSELE et al. (1977)).
the upper panel for a point force excitation at (xF,yF)=(0.25*Lx,0.25*Ly) on the lower
panel of three different double-panel partitions : an air-filled cavity with thickness
Lz=0.15 m (solid), a helium-filled cavity with thickness Lz=0.15 m (dashed) and an
air-filled cavity with thickness Lz=0.10 m (dotted)
Another important issue regarding the sound transmission characteristics of

double-panel partitions is the shape of the panel deformations, since the
shape is a determining factor for the interaction with the incident and
transmitted sound fields, especially at low frequencies. In this respect, a
panel deformation shape is often quantified by its radiation efficiency. This
is defined as the ratio of the average acoustic power per unit area, radiated
by the panel shape, to the average acoustic power per unit area of a piston,
that is vibrating with the same average mean square velocity as the
considered panel shape (FAHY (1985)). The radiation efficiency of a panel
deformation shape is usually derived from the radiation efficiencies of the
modes that contribute to that shape. In this respect, the (odd,odd) panel
modes are identified as efficient radiators, since they produce a net volume
displacement of the fluid. As indicated in section F.3, the out-of-phase
modes, controlled by low-order (odd,odd) panel modes, result from a strong
fluid-structure coupling interaction and their modal panel displacements are
composed of several uncoupled panel modes. As a result, the radiation
efficiency of these coupled modes is strongly altered in comparison with
their corresponding uncoupled panel modes, as was identified already for
the case of a cavity-backed panel (PRETLOVE and CRAGGS (1970), CRAGGS
(1971b)).
Figure F.16 : normalised panel displacement in plane y=0.5*Ly for the (1,1) (a) and
(3,1) (b) out-of-phase panel-controlled modes
(air-filled cavity with thickness Lz=0.15m (solid), helium-filled cavity with thickness
Lz=0.15m (dashed) and air-filled cavity with thickness Lz=0.10m (dotted))
432 Appendix F
Figure F.16 shows the shape of the normalised panel displacement in the
plane y=0.5*Ly for the out-of-phase modes, controlled by the (1,1) (figure
F.16a) and the (3,1) (figure F.16b) uncoupled panel mode. In comparison
with the uncoupled (1,1) panel mode, the net fluid volume displacement and
subsequent radiation efficiency of the (1,1) panel-controlled mode decrease,
due to the significant contribution of the (3,1) panel mode to the coupled
modal panel displacement. On the other hand, the radiation efficiency of the
(3,1) panel-controlled mode increases since the contribution of the (1,1)
panel mode enhances the net fluid volume displacement. Figure F.16
illustrates also that these changes in radiation efficiency increase when the
fluid-structure coupling interaction becomes stronger, as it is the case for the
considered modes, when the cavity thickness is reduced or when the cavity
is filled with helium instead of air.
REFERENCES
Adeye, J.O., Bernal, M.J.M., and Pitman, K.E., “An improved boundary
integral equation method for Helmholtz equations”, Int. J. Num. Meth.
Eng. 21, 779-787 (1985).
Ali, A., Rajakumar, C., and Yunus, S.M., “Advances in acoustic eigenvalue
analysis using boundary element method”, Computers and Structures 56,
837-847 (1995).
Allard, J.F., Depollier, C., and Lauriks, W., “Measurement and prediction of
surface impedance at oblique incidence of a plastic foam of high flow
resistivity”, J. Sound Vib. 132, 51-60 (1989).
Allik, H., Dees, R.N., and Moore, S.W., “Efficient structural acoustic
analysis using finite and infinite elements”, Proceedings of the 1995
Design Engineering Technical Conferences (Boston, 1995), Vol. 3 - part
B ASME 1995, 87-95.
Assaad, J., Decarpigny, J.-N., Bruneel, C., Bossut, R., and Hamonic, B.,
“Application of the finite element method to two-dimensional radiation
problems”, J. Acoust. Soc. Am. 94, 562-573 (1993).
Assaad, J., Bruneel, C., Rouvaen, J.-M., and Bossut, R., “An extrapolation
method to compute far-field pressures from near-field pressures obtained
by finite element method” , Proceedings of the 1995 Design Engineering
Technical Conferences (Boston, 1995), Vol. 3 - part B ASME 1995, 185-
190.
Astley, R.J., “Wave envelope and infinite elements for acoustic radiation”,
Int. J. Num. Meth. Fluids 3, 507-526 (1983).
434 References
Astley, R.J., and Eversman, W., “Wave envelope elements for acoustical
radiation in inhomogeneous media”, Computers and Structures 30, 801-
810 (1988).
Astley, R.J., Macaulay, G.J., and Coyette, J.P., “Mapped wave envelope
elements for acoustical radiation and scattering”, J. Sound. Vib. 170, 97-
118 (1994).
Astley, R.J., Macaulay, G.J., Coyette, J.P., and Cremers, L., “Three-
dimensional wave-envelope elements of variable order for acoustic
radiation and scattering. Part I: formulation in the frequency domain”, J.
Acoust. Soc. Am. 103, 49-63 (1998).
Atalla, N., and Bernhard, R.J., “Review of numerical solutions for low-
frequency structural-acoustic problems”, Applied Acoustics 43, 271-294
(1994).
Atluri, S.N., Gallagher, R.H., and Zienkiewicz, O.C., Hybrid and mixed
finite element methods (Wiley, London, 1983).
Babuška, I., “Are high degree elements preferable? Some aspects of the h
and h-p version of the finite element method”, Proceedings of NUMETA
87 Vol. 1 (Swansea, 1987).
Banerjee, P.K., and Butterfield, R., Boundary element methods in
engineering science (McGraw-Hill, London, 1981).
Bathe, K.J., Finite Element Procedures in Engineering Analysis (Prentice
Hall, Englewood Cliffs, New Jersey, 1982).
Bathe, K.J., Nitikitpaiboon, C., and Wang, X., “A mixed displacement-
based finite element formulation for acoustic fluid-structure interaction”,
Computers and Structures 56, 225-237 (1995).
Bayliss, A., and Turkel, E., “Radiation boundary conditions for wave-like
equations”, Comm. Pure Appl. Math. 33, 707-725 (1980).
Bayliss, A., Gunzburger, M., and Turkel, E., “Boundary conditions for the
numerical solution of elliptic equations in exterior regions”, SIAM J.
Appl. Math. 42, 430-450 (1982).
Belytschko, T.B., “Fluid-structure interaction”, Computers and Structures
12, 459-469 (1980).
Beranek, L.L., and Work, G.A., “Sound transmission through multiple
structures containing flexible blankets”, J. Acoust. Soc. Am. 21, 419-428
(1949).
Berg, P.W., and McGregor, J.L., Elementary partial differential equations
(Holden-Day, San Francisco, 1964).
Bettess, P., Infinite elements (Penshaw Press, Sunderland, 1992).
References 435
Bettess, P., and Zienkiewicz, O.C., “Diffraction and refraction of surface

waves using finite and infinite elements”, Int. J. Num. Meth. Eng. 11,
1271-1290 (1977).
Bilodeau, B.A., and Doyle, J.F., “Spectral elements for acoustic wave
propagation through thin-walled complex structures”, Proceedings of the
6th International Conference on Recent Advances in Structural
Dynamics, edited by Ferguson, N.S., Wolfe, H.F. and Mei, C.,
(Southampton, 1997), 279-293.
Biot, M.A., “Theory of propagation of elastic waves in a fluid-saturated
porous solid. I. Low-frequency range”, J. Acoust. Soc. Am. 28, 168-191
(1956).
Bogomolny, A., “Fundamental solutions method for elliptic boundary value
problems”, SIAM J. Num. Anal. 22, 644-649 (1985).
Bolton, J.S., Shiau, N.-M., and Kang, Y.J., “Sound transmission through
multi-panel structures lined with elastic porous materials”, J. Sound. Vib.
191, 317-347 (1996).
Bossut, R., and Decarpigny, J.-N., “Finite element modeling of radiating
structures using dipolar damping elements”, J. Acoust. Soc. Am. 86,
1234-1244 (1989).
Brebbia, C.A., Telles, J.C.F., and Wrobel, L.C., Boundary element
techniques. Theory and applications in engineering (Springer-Verlag
Berlin, 1984).
Brekke, A., “Calculation methods for the transmission loss of single, double
and triple partitions”, Applied Acoustics 14, 225-240 (1981).
Burgess, G., and Mahajerin, E., “A comparison of the boundary element and
superposition methods”, Computers and Structures 19, 697-705 (1984).
Burnett, D.S., “A three dimensional acoustic infinite element based on a
prolate spheroidal multipole expansion”, J. Acoust. Soc. Am. 96, 2798-
2816 (1994).
Burton, A.J., and Miller, G.F., “The application of integral equation
methods to the numerical solution of some exterior boundary value
problems”, Proc. Royal Soc. London A 323, 201-210 (1971).
Charron, F., Panneton, R., Champoux, Y., Guérin, J.-M., and Boily, S.,
“Analytical, numerical and experimental study of the vibro-acoustic
behaviour of a stiffened shell structure”, Proceedings of the 1995 Design
Engineering Technical Conferences (Boston, 1995), Vol. 3 - part B
ASME 1995, 611-619.
436 References
Cheung, Y.K., Jin, W.G., and Zienkiewicz, O.C., “Direct solution procedure
for solution of harmonic problems using complete, non-singular, Trefftz
functions”, Comm. Appl. Num. Meth. 5, 159-169 (1989).
Cheung, Y.K., Jin, W.G., and Zienkiewicz, O.C., “Solution of Helmholtz
equation by Trefftz method”, Int. J. Num. Meth. Eng. 32, 63-78 (1991).
Cho, P.E. , Energy flow analysis of coupled structures (Ph.D. dissertation,
Purdue University, 1993).
Ciskowski, R.D., and Brebbia, C.A., Boundary element methods in acoustics
(Elsevier Applied Science, London, 1991).
Coyette, J.P., “Ritz vector synthesis versus modal synthesis for fluid-
structure interaction modeling”, Proceedings of International Congress
on Recent Developments in Air- and Structure-borne Sound and
Vibration, edited by Crocker, M.J. (Auburn, 1990), 115-126.
Coyette, J.P., “Validation of a new wave envelope formulation for handling
exterior acoustic and elasto-acoustic problems in the frequency domain”,
Proceedings of the DGLR/AIAA 14th Aeroacoustics Conference, edited
by DGLR, (Aachen, 1992), 421-427.
Coyette, J.P., Dejehet, E., Guisset, P., and Fyfe, K.R., “Numerical analysis
of acoustic and elasto-acoustic problems using combined finite element
and boundary element methods”, Proceedings of the 7th IMAC (Las
Vegas, 1989), 268-275.
Coyette, J.P., and Fyfe, K.R., “Solution of elasto-acoustic problems using a
variational finite element/boundary element method for fluid-structure
interaction problems”, International Symposium on Numerical
Techniques in Acoustic Radiation - ASME Annual Meeting, edited by
Bernhard, R.J., and Keltie, R.F. (San Francisco,1989), 15-25.
Coyette, J.P., and Fyfe, K.R., “An improved formulation for acoustic
eigenmode extraction from boundary element models”, Journal of
Vibration and Acoustics 112, 392-397 (1990).
Coyette, J.P., and Van de Peer, J., “Acoustic boundary elements”,
Proceedings ISAAC 4, edited by Sas, P. (Leuven, 1993), course note
VIII.
Coyette, J.P., and Dubois-Pélerin, Y., “An efficient coupling procedure for
handling large size interior structural-acoustic problems”, Proceedings
ISMA 19, edited by Sas, P. (Leuven, 1994), 729-738.
Craggs, A., “The transient response of a coupled plate-acoustic system using
plate and acoustic finite elements”, J. Sound. Vib. 15, 509-528 (1971).
References 437
Craggs, A., “Computation of the response of coupled plate-acoustic sustems

using plate finite elements and acoustic volume-displacement theory”, J.
Sound. Vib. 18, 235-245 (1971).
Craggs, A., “An acoustic finite element approach for studying boundary
flexibility and sound transmission between irregular enclosures”, J.
Sound. Vib. 30, 343-357 (1973).
Craig Jr., R.R., “A review of time-domain and frequency-domain
component-mode synthesis methods”, Int. J. Anal. & Exp. Modal
Analysis 2, 59-72 (1987).
Craig Jr., R.R., and Bampton, M.C.C., “Coupling of substructures for
dynamic analyses”, AIAA J. 6, 1313-1319 (1968).
Craik, R.J.M., and Wilson, R., “Sound transmission through masonry cavity
walls”, J. Sound Vib. 179, 79-96 (1995).
Craik, R.J.M., Sound transmission through buildings using statistical energy
analysis (Gower, Hampshire, 1996).
Cremers, L., Fyfe, K.R., and Coyette, J.P., “A variable order infinite
acoustic wave envelope element”, J. Sound. Vib. 171, 483-508 (1994).
Cura, F., Curti, G., and Scarpa, F., “State space methods in an eulerian
symmetrical formulation for vibroacoustics”, Modern Practice in Stress
and Vibration Analysis, edited by Gilchrist (Balkema, Rotterdam, 1997),
193-198.
Daniel, W.J.T., “Modal methods in finite element fluid-structure eigenvalue
problems”, Int. J. Num. Meth. Eng. 15, 1161-1175 (1980).
De Fonseca, P., Desmet, W., Cops, A., Cooper, J., and Sas, P., “Active
control of the sound transmission through a double-glazing window”,
Proceedings Inter-Noise 96, edited by Hill, F.A., and Lawrence, R.
(Liverpool, 1996), 1811-1816.
Degrande, G., and De Roeck, G., “A spectral element method for two-
dimensional wave propagation in horizontally layered satureated porous
media”, Computers and Structures 44, 717-728 (1992).
Degrande G., De Roeck, G., Van den Broeck, P., and Smeulders, D.,
“Application of a direct stiffness method to wave propagation in
multiphase poroelastic media”, Meccanica 32, 205-214 (1997).
Delanghe, K., High frequency vibrations: contributions to experimental and
computational SEA parameter identification techniques (Ph.D.
dissertation, K.U.Leuven, 1996).
de Mey, G., “Integral equations for potential problems with the source
function not located on the boundary”, Computers and Structures 8, 113-
115 (1978).
438 References
Desmet, W., and Sas, P., “Sound transmission of a finite double wall
structure with active and passive damping”, Proceedings 3de Nationaal
Congres over Theoretische en Toegepaste Mechanica (Luik, 1994) 35-
38.
Desmet, W., Sas, P., and De Weer, D., “Vibro-acoustic behaviour of finite
double wall structures”, Proceedings of the Third International Congress
on Air- and Structure-borne Sound and Vibration, edited by Crocker,
M.J. (Montreal, 1994), 75-85.
Desmet, W., and Sas, P., “The effect of sound absorbing materials, plate
stiffeners and vibration isolators on the vibro-acoustic behaviour of finite
double wall structures”, Proceedings ISMA 19, edited by Sas, P.
(Leuven, 1994), 101-123.
Desmet, W., and Sas, P., “Sound transmission of finite double-panel
partitions with sound absorbing material and panel stiffeners”,
Proceedings of the First Joint CEAS/AIAA Aeroacoustics Conference,
edited by DGLR (Munich, 1995), 311-320.
Desmet, W., and Sas, P., “Vibro-acoustic analysis of the low frequency
insertion loss of finite double-panel partitions”, in New Advances in
Modal Synthesis of Large Structures, edited by Jezequel, L. (A.A.
Balkema Publishers, Rotterdam, 1997), 383-398.
Dowell, E.H., and Voss, H.M., “The effect of a cavity on panel vibration”,
AIAA J. 1, 476 (1963).
Dowell, E.H., Gorman, G.F., and Smith, D.A., “Acoustoelasticity: general
theory, acoustic natural modes and forced response to sinusoidal
excitation, including comparisons with experiments”, J. Sound Vib. 52,
519-542 (1977).
Evans, G., Practical numerical integration (John Wiley and Sons,
Chichester et al., 1993).
Everstine, G.C., “A symmetric potential formulation for fluid-structure
interactions”, J. Sound. Vib. 79, 157-160 (1981).
Everstine, G.C., and Henderson, F.M., “Coupled finite element/boundary
element approach for fluid-structure interaction”, J. Acoust. Soc. Am. 87,
1938-1947 (1990).
Fahy, F., Sound and Structural Vibration - Radiation, Transmission and
Response (Academic Press, London, 1985).
Favre, B., “The need for effective measurement and analysis methods to
determine the acoustic performance of automotive products”, Procee-
dings Euro•Noise ’98, edited by Fastl, H., and Scheuren, J. (München,
1998), 15-27.
References 439
Feng, G.C., and Kiefling, L., “Fluid-structure finite element vibrational

analysis”, AIAA J. 14, 199-203 (1976).
Filippi, P.J.T., “Layer potentials and acoustic diffraction”, J. Sound Vib. 54,
473-500 (1977).
Finnveden, S., “Exact spectral finite element analysis of stationary
vibrations in a railway car structure”, Acta Acustica 2, 461-482 (1994).
Finnveden, S., “Spectral finite element analysis of stationary vibrations in a
beam-plate structure”, Acustica • Acta Acustica 82, 478-497 (1996).
Finnveden, S., “Spectral finite element analysis of the vibration of straight
fluid-filled pipes with flanges”, J. Sound Vib. 199, 125-154 (1997).
Gerald, C.F., and Wheatley, P.O., Applied numerical analysis (Addison-
Wesley, Massachusetts , 1989).
Giordano, J.A., and Koopmann, G.H., “State space boundary element-finite
element coupling for fluid-structure interaction analysis”, J. Acoust. Soc.
Am. 98, 363-372 (1995).
Givoli, D., Patlashenko, I., and Keller, J.B., “High-order boundary
conditions and finite elements for infinite domains”, Comput. Methods
Appl. Mech. Engng 143, 13-39 (1997).
Golub, G.H., and Van Loan, C.F., Matrix computations (Johns Hopkins
University Press, 1983).
Göransson, P., “On the representation of general damping properties in
modal synthesis solutions of fluid structure interaction problems”,
Proceedings of the DGLR/AIAA 14th Aeroacoustics Conference, edited
by DGLR, (Aachen, 1992), 679-686.
Gösele, K., Gösele, U., and Lakatos, B., “Einfluss einer Gasfüllung auf die
Schalldämmung von Isolierglasscheiben”, Acustica 38, 167-174 (1977).
Green, I.S., “Vibro-acoustic FE analysis of the SAAB 2000 aircraft”,
Proceedings of the 4th NASA/SAE/DLR Aircraft Interior Noise Workshop
- NASA Conference Publication 10103, edited by Stephens, D.G.
(Friedrichshafen, 1992), 47-69.
Greenberg, M.D., Advanced Engineering Mathematics (Prentice-Hall,
Upper Saddle River, New Jersey, 1998).
Gurgeon, H., and Herrera, I., “Boundary methods. C-complete systems for
the biharmonic equation”, Boundary element methods, edited by Brebbia,
C.A. (CML Publ., Springer-Verlag, Berlin, 1981).
Hamdi, M.A., Ousset, Y., and Verchery, G., “A displacement method for the
analysis of vibrations of coupled fluid-structure systems”, Int. J. Num.
Meth. Eng. 13, 139-150 (1978).
440 References
Harari, I., and Hughes, T.J.R., “Galerkin/least-squares finite element

methods for the reduced wave equation with non-reflecting boundary
conditions in unbounded domains”, Comput. Methods Appl. Mech.
Engng. 98, 411-454 (1992).
Harari, I., and Blejer, G., “Finite element methods for the interaction of
acoustic fluids with elastic solids”, Proceedings of the 1995 Design
Engineering Technical Conferences (Boston, 1995), Vol. 3 - part B
ASME 1995, 39-48.
Herrera, I., “Trefftz method”, Topics in Boundary Element Research, Vol. 1:
Basic Principles and Applications, edited by Brebbia, C.A. (Springer-
Verlag, Berlin, 1984), Chap. 10, 225-253.
Herrera, I., Boundary methods. An algebraic theory (Pitman Advanced
Publishing Program, Boston, 1984).
Herrera, I., “Trefftz-Herrera domain decomposition”, Advances in
Engineering Software 24, 43-56 (1995).
Herrera, I., and Gurgeon, H., “Boundary methods. C-complete systems for
Stokes problems”, Comput. Methods Appl. Mech. Engng 30, 225-241
(1982).
Heylen, W., Lammens, S., and Sas, P., Modal Analysis Theory and Testing
(K.U.Leuven, 1995).
Huang, S.C., and Shaw, R.P., “The Trefftz method as an integral equation”,
Advances in Engineering Software 24, 57-63 (1995).
Hughes, T.J.R., The Finite Element Method : Linear Static and Dynamic
Finite Element Analysis (Prentice Hall, Englewood Cliffs, New Jersey,
1987).
Hughes, T.J.R., Franca, L.P., and Hulbert, G.M., “A new finite element
formulation for computational fluid dynamics: VIII The Galerkin least
squares method for advective-diffusive equations”, Comput. Methods
Appl. Mech. Engrg. 73, 173-189 (1989).
Ichikawa, T., and Hagiwara, I., “Coupled acoustic fluid-structure interaction
analysis using component mode synthesis”, Proceedings InterNoise 94,
(Yokohama, 1994), 665-668.
Ihlenburg, F., and Babuška, I., “Finite element solution of the Helmholtz
equation with high wave number - part I: the h-version of the FEM”,
Comput. Math. Appl. 30, 9-37 (1995).
Ihlenburg, F., and Babuška, I., “Finite element solution of the Helmholtz
equation with high wave number - part II: the h-p version of the FEM”,
SIAM J. Numer. Anal. 34, 315-358 (1997).
References 441
Jeans, R.A., and Mathews, I.C., “Solution of fluid-structure interaction

problems using a coupled finite element and variational boundary
element technique”, J. Acoust. Soc. Am. 88, 2459-2466 (1990).
Jin, W.G., Cheung, Y.K., and Zienkiewicz, O.C., ”Application of the Trefftz
method in plane elasticity problems”, Int. J. Num. Meth. Eng. 30, 1147-
1161 (1990).
Jin, W.G., Cheung, Y.K., and Zienkiewicz, O.C., ”Trefftz method for
Kirchoff plate bending problems”, Int. J. Num. Meth. Eng. 36, 765-781
(1993).
Jirousek, J., “Basis for development of large finite elements locally
satisfying all field equations”, Comput. Methods Appl. Mech. Engng 14,
65-92 (1978).
Jirousek, J., “A new finite element model with special suitability for
adaptive solutions and local effect calculations”, in Finite element
methods for plate and shell structures - vol. 1, edited by Hughes, T.J.R.,
and Hinton, E., (Pineridge Press, Swansea, 1986).
Jirousek, J., “The hybrid-Trefftz model - a finite element model with special
suitability to adaptive solutions and local effect calculations”, Finite
element methods for plate and shell structures - vol. 2, edited by Hughes,
T.J.R., and Hinton, E. (Pineridge Press, Swansea, 1986).
Jirousek, J., “Hybrid-Trefftz plate bending elements with p-method
capabilities”, Int. J. Num. Meth. Eng. 24, 1367-1393 (1987).
Jirousek, J., and Leon, N., “A powerful finite element for plate bending”,
Comput. Methods Appl. Mech. Engng 12, 77-96 (1977).
Jirousek, J., and Guex, L., “The hybrid-Trefftz finite element model and its
application to plate bending”, Int. J. Num. Meth. Eng. 23, 651-693
(1986).
Jirousek, J., and Venkatesh, A., “A simple stress error estimator for hybrid-
Trefftz p-version elements”, Int. J. Num. Meth. Eng. 28, 211-236 (1989).
Jirousek, J., and Wroblewski, A., “Least-squares T-elements: equivalent FE
and BE forms of a substructure-oriented boundary solution approach”,
Comm. Num. Meth. Eng. 10, 21-32 (1994).
Johnson, D., “Plate bending by a boundary point method”, Computers and
Structures 26, 673-680 (1987).
Johnston, R.L., and Fairweather, G., “The method of fundamental solutions
for problems in potential flow”, Appl. Math. Modelling 8, 265-270
(1984).
Junger, M.C., and Feit, D., Sound, Structures and Their Interaction, (The
MIT Press, Cambridge, Massachusetts, London, 1972).
442 References
Karasalo, I., “Exact finite elements for wave propagation in range-

independent fluid-solid media”, J. Sound Vib. 172, 671-688 (1994).
Keller, J.B., and Givoli, D., “Exact non-reflecting boundary conditions”, J.
Comput. Phys. 82, 172-192 (1989).
Kita, E., and Kamiya, N., “Trefftz method: an overview”, Advances in
Engineering Software 24, 3-12 (1995).
Kita, E., Kamiya, N., and Ikeda, Y., “Boundary-type sensitivity analysis
scheme based on indirect Trefftz formulation”, Advances in Engineering
Software 24, 89-96 (1995).
Kolodziej, J.A., and Kleiber, M., “Boundary collocation method vs FEM for
some harmonic 2-D problems”, Computers and Structures 33, 155-168
(1989).
Kompella, M.S., and Bernhard, R.J., “Measurement of the statistical
variation of structural-acoustic characteristics of automotive vehicles”,
Proceedings of the 1993 Noise and Vibration Conference, SAE paper
931272 (Traverse City, 1993), 65-81.
Koopmann, G.H., and Benner, H., “Method for computing the sound power
of machines based on the Helmholtz integral”, J. Acoust. Soc. Am. 71,
78-89 (1982).
Kupradze, V.D. (ed.), Three-dimensional problems of the mathematical
theory of elasticity and thermoelasticity (North-Holland, Amsterdam,
1979), (Russian edition, Izd. Tbilis. Univ., Tbilisi, 1968).
Langley, R.S., “Analysis of power flow in beams and frameworks using the
direct-dynamic stiffness method”, J. Sound Vib. 136, 439-452 (1990).
Langley, R.S., “Some perspectives on wave-mode duality in SEA”, IUTAM
Symposium on Statistical Energy Analysis, (Southampton, 1997).
Lee, U., and Lee, J., “Dynamic analysis of one- and two-dimensional
structures using spectral element method”, Proceedings of the 6th
International Conference on Recent Advances in Structural Dynamics,
edited by Ferguson, N.S., Wolfe, H.F. and Mei, C., (Southampton, 1997),
263-277.
Leissa, A.W., Vibration of shells (Acoustical Society of America/ American
Institute of Physics, Woodbury, 1993).
Le Salver, R., “NVH optimization in car body design”, Proceedings
Euro•Noise ’98, edited by Fastl, H., and Scheuren, J. (München, 1998),
457-460.
Leuridan, J., “Software for acoustics ... Crossing the chasm”, Proceedings
Euro•Noise ’95, edited by Millot, P. (Lyon, 1995), 17-32.
References 443
Lim, T.C., and Steyer, G.C., “Hybrid experimental-analytical simulation of

structure-borne noise and vibration problems in automotive systems”,
SAE Technical paper series 920408 (Detroit, 1992), 24-28.
Luo, J., and Gea, H.C., “Modal sensitivity analysis of coupled acoustic-
structural systems”, Journal of Vibration and Acoustics 119, 545-550
(1997).
Lyon, R.H., and DeJong, R.G., Theory and application of statistical energy
analysis - second edition (Butterworth-Heinemann, Boston et al., 1995).
Manning, J.E., and Hebert, B.F., “SEA models for transmission loss of
double wall design”, Proceedings Inter-Noise 95, edited by Bernhard,
R.J., and Bolton, J.S. (Newport Beach, 1995), 1247-1252.
Mariem, J.B., and Hamdi, M.A., “A new boundary finite element method
for fluid-structure interaction problems”, Int. J. Num. Meth. Eng. 24,
1251-1267 (1987).
Masson, P., Redon, E., Priou, J.-P., and Gervais, Y., “The application of the
Trefftz method to acoustics”, Proceedings of the Third International
Congress on Air- and Structure-borne Sound and Vibration, edited by
Crocker, M.J. (Montreal, 1994), 1809-1816.
Mathews, I.C., “Numerical techniques for three-dimensional steady-state
fluid-structure interaction”, J. Acoust. Soc. Am. 79, 1317-1325 (1986).
Missaoui, J., and Cheng, L., “A combined integro-modal approach for
predicting acoustic properties of irregular-shaped cavities”, J. Acoust.
Soc. Am. 101, 3313-3321 (1997).
Mitchell, A., and Hazell, C., ”A simple frequency formula for clamped
rectangular plates”, J. Sound Vib. 118, 271-281 (1987).
Morand, H., and Ohayon, R., “Substructure variational analysis of the
vibrations of coupled fluid-structure systems. Finite element results”, Int.
J. Num. Meth. Eng. 14, 741-755 (1979).
Morand, H., and Ohayon, R., Fluid structure interaction - Applied
numerical methods (John Wiley & Sons, Chichester et al., 1995).
Morse, P.M., and Feshbach, H., Methods of Theoretical Physics - Part II
(McGraw-Hill, New York, 1953).
Mulholland, K.A., Parbrook, H.D., and Cummings, A., “The transmission
loss of double panels”, J. Sound Vib. 6, 324-334 (1967).
Nefske, D.J., and Sung, S.H., “Power flow finite element analysis of
dynamic systems”, Journal of Vibration, Acoustics, Stress and Reliability
in Design 111, 94-100 (1989).
Norris, A.N., and Rebinsky, D.A., “Acoustic reciprocity for fluid-structure
444 References
Ohayon, R., “Variational analysis of a slender fluid-structure system: the

elastic-acoustic beam”, Proceedings of NUMETA 85, Numerical methods
in engineering: theory and applications, edited by Middleton, J., and
Pande, G.N. (Swansea, 1985).
Olson, L.G., and Bathe, K.J., “A study of displacement-based finite
elements for calculating frequencies of fluid and fluid-structure
systems”, Nucl. Eng. Design 76, 137-151 (1983).
Olson, L.G., and Bathe, K.J., “Analysis of fluid-structure interactions. A
direct symmetric coupled formulation based on the fluid velocity
potential”, Computers and Structures 21, 21-32 (1985).
Pan, J., and Bies, D.A., “The effect of fluid-structure coupling on sound
waves in an enclosure - Theoretical part”, J. Acoust. Soc. Am. 87, 691-
707 (1990).
Patera, A.T., “A spectral element method for fluid dynamics: laminar flow
in a channel expansion”, J. Comput. Physics 54, 468-488 (1984).
Patterson, C., and Sheikh, M.A., ”A modified Trefftz method for fluid
flow”, Finite Elements Flow Analysis - Proceedings 4th Int. Symp. on
Finite Element Methods in Flow Problems, edited by Kawai, T. (Univ.
Tokyo Press, Tokyo ,1982) , 973-980.
Patterson, C., and Sheikh, M.A., “A modified Trefftz method for three-
dimensional elasticity”, Boundary Elements - Proceedings 5th Int. Conf.
on BEM (Hiroshima, 1983), edited by Brebbia, C.A., Futagami, T., and
Tanaka, M., (Comp. Mech. Pub./Springer-Verlag, Berlin, 1983), 427-
437.
Pierce, A.D., “The Helmholtz Kirchhoff integral relation as a framework for
developing algorithms for sound propagation through inhomogeneous
moving media”, Proceedings of the 2nd IMACS Symposium on
Computational Acoustics, edited by Lee, D., Carmak, A., and
Vichnevetsky, R. (Amsterdam Elsevier, 1990), 64.
Pierce, A.D., and Wu, X.-F., “Variational method for prediction of acoustic
radiation from vibrating bodies”, J. Acoust. Soc. Am. 74 Suppl.1, S107
(1983).
Piltner, R., “Special finite elements with holes and internal cracks”, Int. J.
Num. Meth. Eng. 21 1471-1485 (1985).
Pinsky, P.M., Thompson, L.L., and Abboud, N.N., “Local high-order
radiation boundary conditions for the two-dimensional time-dependent
structural acoustics problem”, J. Acoust. Soc. Am. 91, 1320-1335 (1992).
Pretlove, A.J., and Craggs, A., “A simple approach to coupled plate-cavity
vibrations”, J. Sound Vib. 11, 207-215 (1970).
References 445
Price, A.J., and Crocker, M.J., “Sound transmission through double panels
using Statistical Energy Analysis”, J. Acoust. Soc. Am. 47, 683-693
(1970).
Priolo, E., and Seriani, G., “A numerical investigation of Chebyshev
spectral element method for acoustic wave propagation”, Proceedings
13th IMACS Conf. on Comp. Appl. Math., edited by Vichnevetsky, R.
(Dublin, 1991), 551-556.
Priolo, E., and Seriani, G., “Spectral element method with substructuring: an
accurate and efficient high-order finite element approach to wave
modelling”, Environmental Acoustics: International Conference on
Theoretical and Computational Acoustics - Volume II, edited by Lee, D.,
and Schultz, M.H. (World Scientific Publishing Co. Pte. Ltd., Singapore
et al., 1994), 509-527.
Rajalingham, C., Bhat, R.B., and Xistris, G.D., “Interior noise in a
rectangular cavity subjected to uniform external excitation on two
opposite flexible walls”, Proceedings of the Third International
Congress on Air- and Structure-borne Sound and Vibration, edited by
Crocker, M.J. (Montreal, 1994), 353-366.
Rektorys, K., Variational Methods in Mathematics, Science and
Engineering (Reidel Publishing Company, Dordrecht, 1977).
Roozen, N.B., Quiet by design : Numerical acousto-elastic analysis of
aircraft structures (Ph.D. dissertation, Technical University Eindhoven,
1992).
Sandberg, G., and Göransson, P., “A symmetric finite element formulation
for acoustic fluid-structure interaction analysis”, J. Sound. Vib. 123, 507-
515 (1988).
Sas, P., Bao, C., Augusztinovicz, F., Van de Peer, J., and Desmet, W., “On
increase of the insertion loss of a double-panel by active noise control”,
Proceedings of the 2nd Conf. on Recent Advances in Active Control of
Sound and Vibration, edited by Burdisso, R.A. (Blacksburg, 1993), 98-
116.
Sas, P., Bao, C., Augusztinovicz, F., and Desmet, W., “Active control of
sound transmission through a double-panel partition”, J. Sound Vib. 180,
609-625 (1995).
Schenck, H.A., “Improved integral formulation for acoustic radiation
Schenck, H.A., and Benthien, G.W., “The application of a coupled finite
element/boundary element technique to large-scale structural acoustic
problems”, Proceedings of the 11th International Conference on
446 References
Boundary Element Methods - Volume 2: Field and Fluid Flow solutions,

(Cambridge, MA, 1989), 309-317.
Seriani, G., and Priolo, E., “High-order spectral element method for acoustic
wave modeling”, Proceedings of the 61st Annual International SEG
Meeting (Houston, 1991), 1561-1564.
Seriani, G., Priolo, E., Carcione, J., and Padovani, E., “High-order spectral
element method for elastic wave modeling”, Proceedings of the 62nd
Annual International SEG Meeting (New Orleans, 1992), 1285-1288.
Seriani, G., and Priolo, E., “Spectral element method for acoustic wave
simulation in heterogenous media”, Finite Elements in Analysis and
Design 16, 337-348 (1994).
Seriani, G., Priolo, E., and Pregarz, A., “Modelling waves in anisotropic
media by a spectral element method”, Proceedings of the Third
International Conference on Mathematical and Numerical Aspects of
Wave Propagation, edited by Cohen, G. (Mandelieu-La Napoule, 1995),
289-298.
Seybert, A.F., Soenarko, B., Rizzo, F.J., and Shippy, D.J., “An advanced
computational method for radiation and scattering of acoustic waves in
three dimensions”, J. Acoust. Soc. Am. 77, 362-368 (1985).
Seybert, A.F., Wu, T.W., and Li, W.L., “A coupled FEM/BEM for fluid-
structure interaction using Ritz vectors and eigenvectors”, Journal of
Vibration and Acoustics, Transactions of the ASME 115, 152-158 (1993).
Sgard, F., and Atalla, N., “Mean flow effects on a plate-backed cavity - Part
2: results”, Acustica • Acta Acustica 83, 252-259 (1997).
Sharp, B.H., “Prediction methods for the sound transmission of building
elements”, Noise Control Engineering 11, 53-63 (1978).
Sommerfeld, A., Partial Differential Equations in Physics (Academic Press,
New York, 1949).
Sorgatz, U., “Integration der numerischen Simulation in die Produkt-
entwicklung”, CIM Management 11, 26-31 (1995).
Sorgatz, U., “Is NVH still an obstacle in product development process
integration”, Proceedings ISMA 21, edited by Sas, P. (Leuven, 1996)
xxiii-xxxix.
Spiegel, M.R., Theory and problems of Fourier analysis with applications to
boundary value problems (Schaum’s outline series, McGraw-Hill, New
York, 1974).
Stewart, J.R., and Hughes, T.J.R., “Explicit residual-based a posteriori error
estimation for finite element discretizations of the Helmholtz equation:
References 447
computation of the constant and new measures of error estimator

quality”, Comput. Methods Appl. Mech. Engrg. 131, 335-363 (1996).
Szabo, B.A., “Mesh design for the p-version of the finite element method”,
Comput. Methods Appl. Mech. Engrg. 55, 181-197 (1986).
Szu, C., “Vibration analysis of structures using fixed-interface component
modes”, Shock and Vibration Bull. 46, 239-251 (1976).
Thompson, L.L., “A multi-field space-time finite element method for
structural acoustics”, Proceedings of the 1995 Design Engineering
Technical Conferences (Boston, 1995), Vol. 3 - part B ASME 1995, 49-
64.
Thompson, L.L., and Pinsky, P.M., “A Galerkin least-squares finite element
method for the two-dimensional Helmholtz equation”, Int. J. Num. Meth.
Eng. 38, 371-397 (1995).
Trefftz, E., “Ein Gegenstück zum Ritzschen Verfahren”, Proc. 2nd Int.
Cong. Appl. Mech. (Zurich, 1926), 131-137.
Trochidis, A., and Kalaroutis, A., “Sound transmission through double
partitions with cavity absorption”, J. Sound Vib. 107, 321-327 (1986).
Varah, J.M., “A practical examination of some numerical methods for linear
discrete ill-posed problems”, SIAM Rev. 21, 100-111 (1979).
Varah, J.M., “Pitfalls in the numerical solution of linear ill-posed
problems”, SIAM J. Sci. Stat. Comput. 4, 164-176 (1983).
Venkatesh, A., and Jirousek, J., “Accurate representation of local effects due
to concentrated and discontinuous loads in hybrid-Trefftz plate bending
elements”, Computers and Structures 57, 863-870 (1995).
Vinokur, R.Y., “Evaluating sound transmission effects in multi-layer
partitions”, Sound and Vibration 7/96, 24-28 (1996).
Wearing, J.L., and Sheikh, M.A., “A regular indirect BEM for thermal
analysis”, Int. J. Num. Meth. Eng. 25, 495-515 (1988).
White, R.H., and Powell, A., “Transmission of random sound and vibration
through a rectangular double wall”, J. Acoust. Soc. Am. 40, 821-832
(1965).
Wilton, D.T., “Acoustic radiation and scattering from elastic structures”, Int.
J. Num. Meth. Eng. 13, 123-138 (1978).
Wolf Jr., J.A., “Modal synthesis for combined structural-acoustic systems”,
AIAA J. 15, 743-745 (1977).
Wyckaert, K., Augusztinovicz, F., and Sas, P., “Vibro-acoustical modal
analysis: reciprocity, model symmetry and model validity”, J. Acoust.
Soc. Am. 100, 3172-3181 (1996).
448 References
Yano, H., Fukutani, S., Watanabe, T., Nakajima, N., and Kieda, A.,
“Application of discrete singularity method in engineering”, Boundary
Elements - Proceedings 5th Int. Conf. on BEM (Hiroshima, 1983), edited
by Brebbia, C.A., Futagami, T., and Tanaka, M., (Comp. Mech.
Pub./Springer-Verlag, Berlin, 1983).
Younes, T.A., and Hamdi, M.A., “Modal shapes reconstruction method for
vibro-acoustic subdomains”, Proceedings of the 15th IMAC-Japan,
edited by Okubo, N. (Tokyo, 1997), 288-294.
Zielinski, A.P., “On trial functions applied in the generalized Trefftz
method”, Advances in Engineering Software 24, 147-155 (1995).
Zielinski, A.P., and Zienkiewicz, O.C., “Generalized finite element analysis
with T-complete boundary solution functions”, Int. J. Num. Meth. Eng.
21, 509-528 (1985).
Zielinski, A.P., and Herrera, I., “Trefftz method: fitting boundary
conditions”, Int. J. Num. Meth. Eng. 24, 871-891 (1987).
Zienkiewicz, O.C., The Finite Element Method - Vol. 1: Basic formulation
and linear problems (McGraw-Hill, London, 1977).
Zienkiewicz, O.C., and Newton, R.E., “Coupled vibration of structures
submerged in a compressible fluid”, Int. Symp. Finite Element Tech.,
(Stuttgart, 1969).
Zienkiewicz, O.C., Kelly, D.W., and Bettess, P., “The coupling of the finite
element method and boundary solution procedures”, Int. J. Num. Meth.
Eng. 11, 355-375 (1977).
Zienkiewicz, O.C., and Bettess, P., “Fluid-structure dynamic interaction and
wave forces. An introduction to numerical treatment”, Int. J. Num. Meth.
Eng. 13, 1-16 (1978).
Zienkiewicz, O.C., and Zhu, J.Z., “A simple error estimator and adaptive
procedure for practical engineering analysis”, Int. J. Num. Meth. Eng. 24,
337-357 (1987).
Zienkiewicz, O.C., and Taylor, R.L., The Finite Element Method - Vol. 2:
Solid and fluid mechanics, Dynamics and Non-linearity (McGraw-Hill,
London, 1991).

Desmet PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Desmet PDF

Загружено:

Авторское право:

Доступные форматы

KATHOLIEKE UNIVERSITEIT LEUVEN

FACULTEIT TOEGEPASTE WETENSCHAPPEN

A WAVE BASED PREDICTION TECHNIQUE

Promotoren : Proefschrift voorgedragen tot

98D12 December 1998

A WAVE BASED PREDICTION TECHNIQUE

Jury : Proefschrift voorgedragen tot

UDC 534.1 December 1998

In de eerste plaats wil ik mijn beide promotoren oprecht bedanken voor de

Coyette, professor Lauriks en professor Degrande wil ik bedanken voor hun

Furthermore I would like to express my gratitude to professor Bolton and

Dit proefschrift stelt een nieuwe, golfgebaseerde voorspellingstechniek voor

1. Inleiding en doelstelling van het onderzoek

lichtgewicht composietmaterialen, zeker in de auto-, luchtvaart- en ruimte-

beperkingen van de huidige voorspellingstechnieken

akoestische druk en de structurele verplaatsing de dynamische

2. Methodiek van de voorspellingstechniek

De nieuwe voorspellingstechniek is gebaseerd op de indirecte Trefftz-

kleinere modellen dan met overeenkomstige eindige-elementenmodellen die

De basisprincipes van de voorspellingstechniek worden besproken aan de

Figuur 1 : tweedimensionaal gekoppeld vibro-akoestisch probleem

waarbij k(=ω/c) het akoestisch golfgetal is en waarbij δ een Dirac delta

waarbij het buigingsgolfgetal kb en de buigstijfheid D van de plaat als volgt

De akoestische randvoorwaarden voor het beschouwde gekoppelde vibro-

waarbij p , vn en Z respectievelijk de opgelegde druk-, normaalsnelheids-

2.2.2. Benadering voor de veldveranderlijken

Voor de caviteitsdruk p(x,y) wordt een benaderende oplossing p(x, y)

Elke akoestische golffunctie Φ a (x, y) is een homogene oplossing van de

De functie pq (x, y) is een particuliere oplossing voor de niet-homogene

De structurele golffuncties Ψ s (x’) zijn vier linear onafhankelijke, homo-

De functie w F (x’) is een particuliere oplossing voor de lijnkrachtterm in

2.2.3. Gekoppeld vibro-akoestisch golfmodel

Gezien de voorgestelde oplossingsbenaderingen (5) en (8) exact voldoen aan

Door elke akoestische golffunctie Φ a in (5) als gewichtsfunctie te

De parameters β p, β v, β Z en β s laten toe het relatief belang van de

gekoppeld vibro-akoestisch golfmodel

In de oplossingsbenaderingen (5) en (8) kan slechts een eindig aantal, na,

waarbij a1,a2=0,±1,±2,... en waarbij Lx en Ly de afmetingen zijn van een

2.3.1. Vergelijking met de eindige-elementenmethode

De golfgebaseerde voorspellingstechniek biedt een aantal belangrijke

• Gezien de benaderingen voor de veldveranderlijken exact voldoen aan de

De golfgebaseerde voorspellingstechniek heeft ook een aantal nadelige

• Gezien de golffuncties complexe functies zijn, die in het volledige of

2.3.2. Vergelijking met de randelementenmethode

Net zoals de randelementenmethode vereist de golfgebaseerde voorspel-

• In heel wat gevallen bevat een integraalformulering, ttz. een gewogen-

De golfgebaseerde voorspellingstechniek heeft ook een aantal nadelige

• De convergentie van een randelementenmodel wordt reeds bekomen met

2.4. Numeriek voorbeeld

Het volgende voorbeeld illustreert de nauwkeurigheid en de convergentie

c=340 m/s) en de aluminium plaat (E=70.10 N/m , ρs=2790 kg/m3, ν=0.3,

t=0.002 m, η=0) wordt geëxciteerd door een eenheidslijnkracht op positie

Figuur 2 : tweedimensionaal gekoppeld vibro-akoestisch systeem

Om een convergerend golfmodel te bekomen, wordt de concave caviteit

Figuur 3 : ogenblikkelijke normaalverplaatsing w van de plaat (volle lijn) en

Figuur 4 toont het resulterende drukveld in de caviteit. Het feit dat de

Figuur 4 : ogenblikkelijke caviteitsdruk bij 200 Hz

Gezien alle randvoorwaarden nauwkeurig voorgesteld zijn en gezien de

Figuur 5 : structurele en akoestische convergentiecurves bij 200 Hz

Figuur 5 toont de relatieve benaderingsfout ∆w op de normaalverplaatsing

3. Toepassing voor twee- en driedimensionale gekoppel-

Hoofdstukken 3 en 4 van dit proefschrift beschrijven hoe de golfgebaseerde

als de fluïdumfaze van een tweedimensionaal poreus medium benaderd met