Torino, 2013
Sergio De Rosa
PASTA-Lab
Laboratory for Promoting experiences
in Aeronautical STructures and Acoustics
Dipartimento di Ingegneria Industriale
Sezione Aerospaziale
Universit degli Studi di Napoli "Federico II"
Via Claudio 21, 80125 Napoli, Italia
sergio.derosa@unina.it
Indice
1) Inquadramento Generale (Interazione FluidoStruttura).
2) Onde, Guide donda e Modi Propri di un Sistema
Lineare.
3) Sistema Tubo e Pistone.
4) Introduzione allAeroelasticit.
5) Instabilit Supersonica di un Pannello.
6) Aeroelasticit Dinamica di un Profilo.
7) Aeroelasticit Statica di Modellli Alari 2D e 3D.
8) Cenno alle Forze Aerodinamiche Instazionarie
9) Il Problema del Flutter (e la Risposta Dinamica).
10) Buffett, Gallopping, Vortex shedding, Stall flutter.
11) Metodi (Statistici) Energetici per Sistemi Lineari.
12) Approccio modale per interpretazioni
energetiche (EDA).
13) Similitudini
14) Introduzione al FEM spettrale.
15) Cenni di Acustica (e AcustoElasticit) Interna dei
Velivoli
16) Potenza Acustica Radiata da Componenti
Strutturali Piani.
17) Un Modello Aero(Idro)Acusto-Elastico Completo:
risposta di pannelli sotto lazione dello strato
limite turbolento
18) Rappresentazioni Universali dei Dati AeroAcusto-Elastici.
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
84
R. L
ohner et al.
CFD
DNS
Biomedical
Applications
LES
RANS
Advanced
Aero/Hydroelasticity
Euler
Conjugate
Heat
Transfer
Full
Potential
Classic
Aero/Hydroelasticity
Potential/
Acoustics
No Fluid
ShockStructure
Interaction
CSD
Rigid
Walls
Ridid Body
(6 DOF)
Modal
Analysis
Linear
FEM
NonLinear
FEM
Rupture/
Tearing
Prescribed
Flux/Temperature
Network Models
Thermal Stress
Fatigue
Linear FEM
Nonlinear FEM
CTD
CFD
DNS
Arterial Flows
Tents
Parachutes
Airbags
LES
RANS
Chip Cooling
Engine Cooling
Underhood Flows
Flutter
Buzz Whipping
Euler
Full
Potential
Aerodynamics
Galloping Noise
Flutter
Potential/
Acoustics
No Fluid
Extrusion
Material Forming
BlastStructure
Weapon Fragmentation
Rigid
Walls
Ridid Body
(6 DOF)
Modal
Analysis
Linear
FEM
CSD
NonLinear
FEM
Prescribed
Flux/Temperature
Network Models
Linear FEM
Nonlinear FEM
CTD
Rupture/
Tearing
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1
La risposta di un oscillatore semplice e di un
tubo
2
1 p(x,t)
c2
t2
(1.1)
dx x=0
La (1.2) relaziona il gradiente di pressione con laccelerazione del pistone alla
sezione x = 0: se il pistone accelera verso destra (
x(t) > 0), a x = 0 il fluido
dp
espande ( dx
|x=0 < 0).
Sullaltra estremita del condotto (x = L), supponiamo che il tubo e chiuso,
e quindi la seconda condizione al contorno e
dp
= 0.
dx x=L
(1.3)
x(t) = Xejt
(1.5)
2 m + k = 0.
(1.6)
k
0 =
(1.7)
m
e quindi la frequenza propria delloscillatore
1
k
f0 =
.
(1.8)
2
m
Analisi modale complessa: oscillatore con smorzamento viscoso
Se e presente uno smorzamento viscoso (vedi Capitolo 4), lequazione
omogenea associata diventa
m
x(t) + bx + kx(t) = 0
(1.9)
(1.10)
p=
b2
2m
4km
(1.11)
Si definisce per comodita uno smorzamento critico bcr , che permette di discriminare le varie possibilita di moto. Esso e funzione dei parametri strutturali,
al pari della pulsazione naturale.
bcr = 2 km
(1.12)
b
b
=
bcr
2 km
(1.13)
b
= 20
m
(1.14)
p1,2 = 0
1 .
(1.15)
(1.16)
Prima radice
0 + j 1 2
Seconda radice
0 j 1 2
subcritico
critico
( = 1)
reali coincidenti
0 [ ]
0 [ ]
supercritico
reali distinte
( < 1)
( > 1)
Forma risolutiva
2
x(t) = e( 0 t) X1 e(j0 t 1 ) + X2 e(
Y1 e( 0 t) cos j0 t 1 2 +
x(t) = Xe(
j0 t
2)
0 t)
x(t) = e( 0 t) X1 e(0 t
e( 0 t) X1 cosh 0 t 2
2 1)
+ X 2 e( 0 t
=
1 + X2 sinh 0 t 2
1)
(1.17)
(1.18)
mp2 + [1 + j]k = 0
(1.19)
da cui si ricava
p2 =
k
[1 + j]
m
p2 = 02 [1 + j].
(1.20)
(1.21)
(1.22)
(1.23)
1
F
2
m ( + 02 )
x(t) =
F
m
2 +
02 )
ejt .
(1.24)
La funzione di trasferimento e
H() =
1
.
+ 02 )
(1.25)
F
m
2
0 ) +
2j0
ejt .
(1.26)
( 2 +
F
m
02 ) +
2j0
(1.27)
10
x(t) =
F
m
02 )
+ j02
ejt
(1.28)
e la funzione di trasferimento e
x(t) =
1
.
+ 02 ) + j02
(1.29)
(1.30)
(1.31)
=
=
.
FDV (t)
20
2 0
(1.32)
2 0
0
.
2
(1.33)
.
2
(1.34)
11
Figura 1.5: analisi del rapporto delle forze dissipative in funzione della
frequenza.
p(x, t) =
1 2 p(x, t)
c2 t2
(1.35)
dove:
(1.37)
1
P (x)T (t)
c2
P (x)
1 T (t)
= 2
P (x)
c T (t)
(1.38)
12
P (x)
2
= 2.
P (x)
c
(1.40)
x +P2 cos x = P1 sin(kx)+P2 cos(kx) = P1 e
c
c
jkx
+P2 ejkx
(1.41)
dove k = c e il numero donda. Le costanti P1 e P2 sono determinate dalle
condizioni al contorno. In particolare, le condizioni al contorno possono essere:
P (0)
= P0 , ovvero tubo aperto in x = 0;
dP
= Q0 , ovvero tubo chiuso in x = 0;
dx
x=0
P (L)
= PL , ovvero tubo aperto in x = L;
dP
= QL , ovvero tubo chiuso in x = L.
dx
x=0
dx x=0
dP
=0
dx x=L
P (x) = kP1 cos(kx) kP2 sin(kx)
(1.42)
(1.43)
(1.44)
kP2 sin(k0) = 0
(1.45)
kP2 sin(kL) = 0
(1.46)
P (x) =
13
(1.51)
i=0
fi =
(i + 1) (i)
c
c=
2L
2L
(1.52)
2L
c
(1.53)
mediante la quale e possibile calcolare in modo statistico quanti modi risuonano in un dato intervallo in frequenza f .
N=
f n =
2L
c
(1.54)
(1.55)
P (L) = 0
(1.56)
(1.57)
(1.58)
Quindi
P (x) =
P1,i sin(ki x).
(1.59)
i=1
14
kP2 sin(kL) = 0
(1.60)
(1.61)
c 1
+i
L 2
(1.63)
c 1
+i
(1.64)
2L 2
{0, 1, 2, . . . N }. La risposta del sistema e pertanto ancora esprimibile
P (x) =
P1,i sin(ki x).
(1.65)
fi =
con i
come
i=0
Dal confronto degli autovalori e delle frequenze proprie del tubo apertochiuso con quelli del tubo aperto-aperto e chiuso-chiuso, risulta che essi sono
differenti. Calcolando la distanza modale tra due modi successivi anche per
il sistema tubo aperto-chiuso, risulta che esso resta invariato rispetto ai casi
precedentemente analizzati.
f = fi+1
fi =
c 1
+i+1
2L 2
c 1
c
+i =
2L 2
2L
(1.66)
k = 2L
. Riepilogando le frequenze naturali per i casi analizzati:
ic
tubo chiuso-chiuso fi = 2L
, con i {0, 1, 2, . . . N };
ic
tubo aperto-aperto fi = 2L
, con i {1, 2, . . . N };
tubo aperto-chiuso fi =
c
2L
1
2
+ i , con i
{0, 1, 2, . . . N }.
In tutti i casi analizzati, sono costanti sia lintervallo nello spazio dei numeri
c
donda k = 2L
che il passo in frequenza f = 2L
.
15
(1.68)
(1.69)
Conseguentemente
kP2 sin(kx)
(1.70)
A
m
P (0)
.
2 + 02
(1.71)
(0) A
kP2 sin(k0) = 2P+
2
0 m
kP2 sin(kL) = 0
(1.72)
P2 A
kP1 +
=0
2 +02 m
kP1 cos(kL) kP2 sin(kL) = 0
2 P2 A
k
P1
0
2
2
+0 m
=
.
0
P2
cos(kL) sin(kL)
(1.73)
(1.74)
cos(kL)
A
2
=0
2
2
+ 0 m
tan
L
c A
= 2
.
c
02 m
(1.75)
16
17
2
0
m
c2 L
0.0
18.7
137.3
272.6
408.4
544.3
680.3
816.2
136
= 0.045
272
408
544
680
816
13.7
0.0
0.0
127.6
146.3
272.9
408.5
544.3
680.3
816.2
136
= 4.588
272
408
544
680
816
137.6
0.0
0.0
135.4
238.0
274.7
408.7
544.4
680.3
816.2
136
= 13.95
272
408
544
680
816
240.0
A c2A
S c2S
(1.77)
c2
S S
ar c2ar
(1.78)
18
ac =
c2
S S
ac c2ac
(1.79)
ac c2ac
.
ar c2ar
(1.80)
jx
(1.82)
jkP2 ex .
(1.83)
A
jkP1 jkP2 =
(P1 + P2 )
2 +02 m
jkL
jkL
P1 e
P2 e
=0
Introducendo la variabile
=
2
A
2 + 02 m
jk +
jk +
P1
P2
ejkL
e jkL
(1.84)
(1.85)
0
0
(1.86)
jkL
( + jk) + ejkL (
( + jk) cos(kL)
19
jk) = 0
( + jk)j sin(kL) + (
jk) cos(kL) + (
jk)j sin(kL) = 0
(1.87)
A questo punto, e necessario annullare contemporaneamente e separatamente
la parte reale e quella immaginaria:
cos(kL) + k sin(kL) + cos(kL) + k sin(kL) = 0
jk cos(kL) j sin(kL) jk cos(kL) + j sin(kL) = 0
2 cos(kL) + 2k sin(kL) = 0
0=0
(1.88)
tan(kL) =
c
A
2 + 02 m
(1.89)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
Introduzione allAeroelasticit
Sergio De Rosa
Dipartimento di Ingegneria Industriale
Sezione Aerospaziale
Universit degli Studi di Napoli Federico II
Carico
(Input)
Struttura
(Sistema)
Esempi:
Risposta Statica
{ F} = [ K]{ x}
Spostamento
(Output)
Risposta Aeroelastica
Carichi
Sistema
Risposta
U
"
"
m"
"
Angolo d Attacco
"
Profilo
Elastico
Profilo
Rigido
Rigidezza
Torsionale
Portanza
Comportamento
Dinamico delle
Strutture
Comportamento
Dinamico delle
Strutture
Comportamento
Dinamico delle
Strutture
10
Comportamento
Dinamico delle
Strutture
11
Il Concetto di Operatore
Operatore
Aeroelastico
12
+ cx + kx
F = mx
13
14
Ramo A-E
!D Divergenza
!EC Efficacia dei Comandi
!IC Inversione dei Comandi
!DCE Distribuzione del Carico sul Velivolo
Elastico
!SSE Stabilit Statica del Velivolo Elastico
15
16
Ramo I-E
17
!F Flutter
!B Buffetting
!RD Risposta Dinamica
!S DE Stabilit Dinamica del Velivolo
Elastico
18
cit
sti
ela
vo
ser
ae
roe
las
tic
it
aeroservodinamica
19
s
cu
a
tic
r oa
ae
ae
roe
las
tic
it
acustoelasticit
20
A
H
21
22
(A+B+C)q=A(q)q+QD
ed esplicitando le funzionalit col tempo:
d 2q
dq
A 2 +B +Cq = A( q )q + Q D
dt
dt
Si pu anche esplicitare analogamente loperatore aerodinamico:
d 2q
dq
( A A) 2 + ( B B) + ( C C)q = Q D
dt
dt
23
([ A ] [ A]){q}
+ ([ B] [ B]){q }
+ ([ C] [ C ]){q}
= {Q D }
24
Flutter
([ C] [ C]){q} = {Q D }
Aeroelasticit Statica
Divergenza
([ C] [ C]){q} = 0
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
2
Linstabilit
a dinamica (flutter supersonico) di
un pannello
22
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
h @ 4
i
@4
@4
@ 2 w(x, y, t)
+
+
w(x,
y,
t)
+ s h
+ p(x, y, t) = 0 (2.1)
4
2
2
4
@x
@x @y
@y
@t2
o equivalentemente
Dr4 w(x, y, t) + m
@ 2 w(x, y, t)
+ p(x, y, t) = 0
@t2
(2.2)
dove:
1 X
1
X
p=1 r=1
sin
px
a
sin
ry
b
qpr (t)
(2.3)
o equivalentemente
w(x, y, t) =
1 X
1
X
(2.4)
p=1 r=1
pr (x, y) sono gli autovettori, che sono noti (avendo imposto le condizioni
al contorno);
qpr (t) e la variabile lagrangiana modale, dipendente solo dal tempo.
pr (x, y)
2
!pr
m
pr (x, y)
=0
(2.5)
(2.6)
23
con p, r 2 {1, 2, . . . , N }.
Si osservi che per denotare un singolo autovettore pr (x, y) occorre una
coppia di interi (p, r), che definisce in modo univoco un modo proprio di
vibrare del pannello e la corrispondente pulsazione propria. In particolare la
coppia dinteri (p, r) rappresenta il numero di semionde flessionali sviluppate
secondo gli assi x e y, rispettivamente, come mostrato nella Figura 2.2.
p=1,r=3 (x, y)
Lespansione modale qui presentata e applicabile anche a casi in cui le condizioni al contorno del pannello siano diverse da quelle del semplice appoggio.
In tal caso, lintero procedimento e approssimato (sviluppo di Rayleigh - Ritz)
in quanto sia lespansione dei modi spaziali sia le autosoluzioni temporali sono
grandezze approssimate.
Si noti inoltre che spesso e arduo classificare i modi propri mediante il conteggio delle semionde in due direzioni ortogonali: basti pensare che il pannello,
in generale, potrebbe non essere dotato di regolarita geometrica.
Pertanto, lespansione modale esatta presentata in questo capitolo e possibile solo in pochi casi.
2.1.2 Soluzione dellequazione completa
Nel paragrafo precedente, si e delineata una procedura per modellare loperatore strutturale, trovando una soluzione per lequazione omogenea associata. Si cercano ora soluzioni modali anche per lequazione completa, in cui e
presente un generico carico di pressione p(x, y, t). I passi da eseguire sono, in
sequenza:
1.
2.
3.
4.
24
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
Si noti che al secondo membro della (2.7) compare la capacita che la generica distribuzione di pressione p(x, y, t) ha di compiere lavoro sul modo proprio
mn (x, y). Al primo membro troviamo i termini modali, in particolare:
= mpr = s h
Z aZ
0
mn (x, y) dx dy
1
[ pr (x, y)] dx dy = s hab.
4
0
(2.8)
1
4
2
kpr = mpr !pr
.
(2.9)
(2.11)
Risolvendo, si ottiene
2
mpr qpr (t) + mpr !pr
qpr (t) =
f (t) sin
px
F
sin
ry
F
(2.12)
j!t
j!t
(2.13)
(2.14)
p
facile dimostrare che la soluzione dellequazione completa e
con j =
1. E
data dallespansione
w(x, y, t) = e
j!t
1 X
1
X
Apr sin
p=1 r=1
= Fe
j!t
1 X
1 sin
X
p=1 r=1
px
px
a
sin
25
ry
=
b
(2.15)
pxF
ryF
sin ry
sin
sin
b
a
b
.
2
mpr !pr
mpr ! 2
1 X
1 sin px sin ry sin pxF sin ryF
X
a
b
a
b
4F
w(x, y, t) =
e j!t
.
2
s hab
!pr
!2
p=1 r=1
(2.16)
Per regolare il comportamento in condizioni di risonanza (! = !pr ), e
necessario introdurre uno smorzamento; il metodo pi
u semplice per introdurre
lo smorzamento del pannello nel modello analitico e quello di considerare un
modulo di elasticit
a complesso. Nel caso in esame, si definisce una rigidezza
flessionale complessa D(1 + ), dove il coefficiente di smorzamento.
In tal caso, la risposta del pannello in presenza di smorzamento e pari a
1 X
1 sin px sin ry sin pxF sin ryF
X
a
b
a
b
4F
e j!t
.
w(x, y, t) =
2
2 ) + j! 2
s hab
(!
!
pr
pr
p=1 r=1
(2.17)
isolato;
forzato, con un carico puntuale in (xF , yF );
smorzato (che e lipotesi pi
u forte per il modello di smorzamento assunto).
26
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
e il potenziale di velocita;
U e la velocit
a della corrente asintotica;
C e la velocit
a della corrente asintotica.
Le condizioni al contorno per il potenziale sono:
|z!1 = 0
(
@
|z!0 =
@z
condizione asintotica
@w
@t
+U
@w
@x ,
0,
sul pannello
fuori dallarea del pannello
(2.19)
(2.20)
@w
@t ,
@t
+U
@
@x
(2.21)
p=
p=
@w
+U
@t
@x
@w
f C U
@x
f C
@w
27
(2.22)
(2.23)
Il vantaggio dellutilizzo della Piston Theory e che, mediante alcuni passaggi matematici, si giunge a scrivere la parte aerodinamica associata alla vibrazione del pannello, cioe la distribuzione di pressione, nelle stesse coordinate
modali utilizzate per lequazione del pannello.
Utilizzando le (2.3),(2.7),(2.22), e possibile ottenere:
2
mpr qpr (t) + mpr !pr
qpr (t) + f U 2 Qpr (t) = 0
dove
Qpr (t) =
Z aZ
0
b
0
p(x, y, t)
f U 2
pr (x, y) dx dy.
(2.24)
(2.25)
La matrice Qpr e una matrice modale, e rappresenta il lavoro che la dis una matrice
tribuzione di pressione p(x, y, t) svolge per il modo proprio pr . E
piena, nella quale:
Pertanto, per descrivere ogni termine della matrice Qpr sono necessari 4
indici, poiche il termine Qprmn esprime il lavoro compiuto dalle forze aerodinamiche dovute al modo pr-simo del pannello per gli spostamenti associati
al modo mn-esimo. La forza generalizzata relativa al modo pr-esimo del pannello Qpr e data dalla somma dei contributi dei lavori compiuti dalle forze
aerodinamiche di tutti i modi mn-esimi del pannello, ovvero:
Qpr =
M X
N
X
Qprmn
(2.26)
m=1 n=1
Grazie alla Piston Theory, la matrice Qpr puo essere scomposta in una
parte statica ed una dinamica, ottenendo
Qpr =
M X
N h
X
m=1 n=1
M X
N h
i X
i
(d)
Q(s)
qmn (t)Sprmn + qmn (t)Dprmn .
prmn + Qprmn =
m=1 n=1
(2.27)
Nella (2.27), la matrice Spr e la matrice statica, legata alleetto aerodinamico
stazionario, ed e rappresentativa della rigidezza introdotta dalla parte aerodinamica (si parla di matrice di rigidezza aerodinamica generalizzata); la matrice Dpr e la matrice dinamica, legata alleetto aerodinamico instazionario,
e tiene conto dello smorzamento introdotto dalla parte aerodinamica (matrice
di smorzamento aerodinamico generalizzata).
28
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
Sprmn =
1
M1
1
=
M1
Dprmn =
Z aZ
"Z0
sin
0
1
(U )(M1 )
1
=
(U )(M1 )
pr (x, y)
@x
ry
b
Z aZ
"Z0
ny
b
dy
#"
p
a
cos
0
Dprmn =
sin
mx
a
dx
(2.28)
pr (x, y) mn (x, y) dx dy
0
sin
ry
b
sin
ny
b
=
#"Z
dy
sin
px
sin
mn (x, y) dx dy
(m p)(m+p)
px
a
p)] 2m
sin
mx
a
dx
(2.29)
se r 6= n o p = m
se r = n o p 6= m
(2.30)
0
ab
4(U )(M1 )
Dpr
se r 6= n o p 6= m
se r = n o p = m
(2.31)
La novit
a dellapproccio eettuato sta nellutilizzare la base modale per
sviluppare sia loperatore strutturale che quello aerodinamico. Infatti, grazie
allipotesi di lavoro della teoria del pistone, le equazioni si presentano tutte in
funzione degli spostamenti del pannello, espressi dallunica variabile modale
q(t).
M X
N h
X
m=1 n=1
i
qmn (t)Sprmn + qmn (t)Dprmn = 0
(2.32)
29
30
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
f U ab
q1,1 (t) = 0.
M1 4
(2.34)
Dato che S1,1 1,1 = 0, dividendo primo e secondo membro per ab
4 e
ponendo per semplicit
a q1,1 q e !1,1 , la (2.34) diventa:
1,1 q1,1 (t)
s h
q (t) + s h 2 q(t) + f C q(t)
=0
(2.35)
1 f C
2 s h
C 2
f
s h
4 2 .
(2.37)
Dallesame
(2.37)si puo dedurre che si ottengono autovalori comp della
f C 2
lessi solo se s h < 4 2 . In tal caso, gli autovalori sono necessariamente
complessi coniugati a parte reale negativa.
Allo stesso tempo, e possibile constatare che nel caso di autovalori puramente reali, essi sono certamente minori di zero (al massimo nulli).
Si ricordi a questo punto che per come si e sviluppata la procedura di
estrazione degli autovalori, la condizione necessaria affinche si abbia instabilita
aeroelastica (panel flutter ) e che la parte reale dellautovalore deve essere
positiva, ovvero Re( ) > 0.
Sulla base delle precedenti considerazioni, e possibile aermare che:
31
1
1
!
1 0 q1,1 (t)
q1,1 (t)
abs h
+ abs h 1,1 2
=
0 1 q2,1 (t)
0 !2,1 q2,1 (t)
4
4
f U ab 1 0 q1,1 (t)
S
S
q1,1 (t)
f U 2 1,1 1,1 1,1 2,1
. (2.38)
S2,1 1,1 S2,1 2,1 q2,1 (t)
M1 4 0 1 q2,1 (t)
Dato che S1,1 1,1 = S2,1
ponendo per semplicit
a
2,1
q1,1 q1 ;
q2,1 q2 ;
!1,1 1 ;
!2,1 2 ;
la (2.38) diventa
f C
q1 (t)
+
q2 (t)
s h
q1 (t)
q2 (t)
f U 2 4
12 0
0
+
0 22
S
s h ab 2,1
1,1
S1,1
0
2,1
q1 (t)
q2 (t)
(2.39)
Si cercano ancora soluzioni del tipo q(t) = q0 e t . Lequazione secolare e la
seguente
f C
+ 12
s h
f C
+ 22
s h
U 2 4 2
f
S2,1
s h ab
f C
+ 12
s h
2
f C
+ 12
s h
S2,1
11
= 0.
(2.40)
Sostituen-
U 2 4 2 2 b 2
f C
f
+ 22 +
= 0 (2.41)
s h
s h ab
3 M1
2
U C 8 2
f C
f
+ 22 +
=0
s h
s ha 3
(2.42)
0
.
0
32
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
Figura 2.3: frequenze naturali del sistema instazionario al variare della velocita
a Mach fissato (M1 = 2).
33
1
1
!1,1
1 0 q1,1 (t)
q1,1 (t)
abs h
+ abs h
=
2
0 1 q2,1 (t)
0 !2,1
q2,1 (t)
4
4
q1,1 (t)
2 S1,1 1,1 S1,1 2,1
= f U
S2,1 1,1 S2,1 2,1 q2,1 (t)
(2.43)
34
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
Figura 2.5: frequenze naturali del sistema stazionario al variare della velocita
a Mach fissato (M = 21 ).
q1 (t)
q2 (t)
f U 2 4
12 0
0
+
0 22
s h ab S2,1
1,1
S1,1
0
2,1
q1 (t)
q2 (t)
0
. (2.44)
0
+ +12
U C 8 2
f
+ +22 +
= 0.
s ha 3
(2.45)
2.6 Generalizzazione
35
2.6 Generalizzazione
In forma matriciale, la relazione generale per un numero T = N M di
modi presenti nel sistema e
mg [I]{
q (t)} + mg []{q(t)} + f U 2 [S]{q(t)} + f U 2 [D]{q(t)}
= {0} (2.46)
dove con [I] si e indicata la matrice identica, con [] la matrice diagonale
delle pulsazioni naturali al quadrato. Ordinando i termini, e possibile scrivere
[I]{
q (t)} +
U2
f U 2
f
[D]{q(t)}
+
[S] + [] {q(t)} = {0}.
mg
mg
(2.47)
(2.48)
36
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
Riportiamo lesame degli autovalori. Cio e possibile riscrivendo interamente il problema, passando pero ad un problema a 2T gradi di liberta nello
spazio degli stati (raddoppia lordine del sistema di equazioni dierenziali, ma
ci si riconduce ad un problema di estrazione degli autovalori perfettamente
simmetrico).
[0] [I]
{
q (t)}
[I] [0]
{q(t)}
0
+
=
(2.49)
[I] [B(U )] {q(t)}
{q(t)}
Ponendo {z(t)} =
= {Z}e t , ci si riconduce al problema agli
{q(t)}
autovalori (2.50).
!
[0] [I]
[I] [0]
0
+
{Z} =
[I] [B(U )]
[0] [C(U )]
0
[0] [I]
[I] [B(U )]
[0] [I]
[0] [I]
+
[I] [B(U )]
[I] [B(U )]
[I] [0]
=
[0] [I]
[I] [0]
=
[0] [C(U )]
[B(U )] [ C(U )]
=
[I]
[0]
0
0
0
0
(2.50)
In particolare, si ha che
01
1b
b1
1 0
37
38
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
smorzamento. Infatti, mentre nel modello stazionario alle basse velocita gli
smorzamenti del sistema aeroelastico restano costanti e pari allo smorzamento strutturale, nel caso di modello instazionario essi deviano dallo smorzamento strutturale per eetto delle forze aerodinamiche che hanno un eetto
smorzante. Anche in questo caso, comunque, in corrispondenza della velocita
di flutter uno degli smorzamenti diventa positivo, e quindi si passa in una
condizione di instabilit
a aeroelastica.
39
40
2 Linstabilit
a dinamica (flutter supersonico) di un pannello
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1. Introduzione
In queste note simposter lanalisi di uno dei pi semplici sistemi aeroelastici, in cui vi la
possibilit di uninstabilit dinamica.
1.1 Equazioni del Moto
Consideriamo unala infinitamente lunga, in cui le uniche rigidezze siano quelle flessionale
(nel piano xz) e torsionale (intorno allasse y). Tali rigidezze siano rappresentate da elementi noti
e concentrati nellasse elastico, e denotate dai simboli kh e k , rispettivamente.
z,w
x,r
O
b
b
ba h
bx
CM
AE
AC
h(t)
(t)
u ( x, y, z, t ) = u ( x, y, z, q1 , q 2 ,, q N )
v( x, y, z, t ) = v( x, y, z, q1 , q 2 ,, q N ) .
w ( x, y, z, t ) = w ( x, y, z, q1 , q 2 ,, q N )
(1)
w ( x, t ) = h ( t ) r tan(( t )) .
(2)
Lultima relazione, nel caso di piccoli disturbi (ovvero piccoli spostamenti e angoli
dattacco) diverr:
u ( x, t ) = 0
v( x , t ) = 0
(3)
w ( x, t ) = h ( t ) r( t )
Per completezza si riscritto linsieme completo degli spostamenti alari. Lo spostamento
flessionale h, stato assunto positivo verso il basso, mentre la torsione , positiva oraria. Una
possibile scelta (tra le infinite) delle coordinate generalizzate o lagrangiane data proprio da h
ed . Scriviamo le energia potenziale e cinetica del sistema in esame:
U( t ) =
1
1
k h h( t ) 2 + k ( t ) 2 ,
2
2
T( t ) =
2
1
1
2
w
(
x
,
t
)
dm
=
h
(
t
)
+
r
(
t
)
dm .
2 b
2 b
(4)
(5)
2
1 2
1
h ( t ) dm = m h ( t ) ;
2 b
2
[ ]
[ ]
1
[r ( t )]2 dm = 1 [ ( t )]2 r 2 dm = 1 I [ ( t )]2 ;
2 b
2
2
b
(6)
r (t )h (t )dm = (t )h (t )S
dove
I il momento dinerzia di massa del segmento alare intorno allasse elastico, e S il
momento statico del segmento alare rispetto allasse elastico.
Si ricordi che essendo m la massa del segmento alare:
S = mx b x =
S
,
mb
(7)
e quindi x (positivo per centro di massa dietro allasse elastico) la distanza in semicorde
tra centro di massa ed asse elastico.
Applicando le equazioni di LAGRANGE1, otterremo le volute equazioni del moto, per le due
coordinate generalizzate h ed :
-m
+S
,
S * &h( t ) # -k h
%
"+
I () $
( t )! +, 0
0 * & h ( t ) # &Q h ( t ) #
%
"=%
".
k () $( t )! $Q ( t )!
(8)
Per brevit di scrittura, salvo dove strettamente necessario, si ometter dora in poi di
indicare la dipendenza temporale. Notiamo che le matrici dei coefficienti non sono consistenti
come dimensioni2, infatti:
[m]=[ML-1], [S ]=[M], [I ]=[ML], [h]=[L], [kh]=[FL-2], [k ]=[F], [Qh]=[FL-1],
[Q ]=[F].
1.2 Adimensionalizzazione
Prima di esplicitare gli operatori aerodinamici e di disturbo, conviene passare ad una
opportuna adimensionalizzazione delle eq.(9). E ora conveniente esprimere le rigidezze
mediante le pulsazioni naturali del sistema strutturale disaccoppiato (S =0):
2h =
k
kh
; 2 = .
m
I
(9)
A questo punto dividendo la prima delle eq.(8) per il termine b2b e la seconda per
b2b2, si otterr:
0 m
. b 2
.
. S
./ b 3
0 m 2h
S
'h $ .
b 3 + ! ! . b 2
+
+
I +& b # .
!
!
"
%
. 0
b 4 +,
/
' Q $
0 +' h $ ! h 3 !
+ ! ! ! b !# . (10)
2 &b# = &
I + ! ! ! Q !
% "
!% b 4 !"
b 4 +,
Le coordinate generalizzate sono ora entrambe adimensionali cos come i coefficienti delle
matrici e le forze generalizzate. Introduciamo i gruppi adimensionali:
m
=
b 2
S
x mb
= 3 = x
3
b
b
I
mb 2 r2
I
=
= r2 ; r2 = 2
4
4
b
b
mb
(11)
Si noti che: la densit del fluido alla quota considerata ed []=[ML-3]; r il raggio
dinerzia della sezione alare misurata in semicorde; il rapporto di massa della sezione alare.
Si pu allora riscrivere lintero sistema come segue:
01
.
/x
x - '! h $!
02h
& # + .
r2 +, ! b !
/0
% "
' Qh $
h
'
$
0 - ! ! !! b 3 !!
#.
+& b # = &
r2 2 , ! ! ! Q !
% "
!% b 4 !"
(12)
Imponiamo che la forma temporale delle soluzioni e dei termini noti siano assegnati:
( T
&
t &' q i
% U W
## +
=
con i {1,2 N}
q
i
i
$
Ci pu non essere un problema sostanziale, ma lo sicuramente dal punto di vista formale. Inoltre, lutilit
dei termini matriciali proprio quella di poter comparare tra loro i singoli coefficienti, cosa che possibile solo se
essi vengono resi tutti omogenei.
h( t ) = h 0 e pt
( t ) = 0 e pt
Q h ( t ) = Q h 0 e pt
,
Q ( t ) = Q 0 e pt
(13)
con pC e [p]=[T-1].
In tal modo si avr la forma definitiva delle equazioni del moto per il problema in esame:
0 26 1
. p 4
.
5x
/
2
h
x 3
6
+
4
r2 12
50
' Qh0 $
h
'
$
0 3 - ! 0 ! !! b 3 !!
#.
1 +& b # = &
r2 2 2 +, ! ! ! Q 0 !
% 0" !
4
% b !"
(14)
2. Operatore Strutturale
2.1 Analisi delle Autosoluzioni
Loperatore strutturale bidimensionale in oggetto molto semplice e permette delle facili
considerazioni. Consideriamo leq. omogenea associata al problema completo retto dalleq.(15):
x 1
42h
+ 2
r2 /0
30
. 24 1
, p 2
,
3x
-
(15)
Per lassenza di termini dissipativi di qualsiasi natura, le autosoluzioni non potranno che
essere che di tipo armonico, tali quindi da verificare p=j. Le soluzioni esisteranno, com ben
noto, se e solo se il determinante della matrice dei coefficienti nullo:
x #
&2h
+ $
r2 !"
%0
&1
$
%x
2
0 #
! = 0.
r 2 "
2
(16)
12, 2
2h
- 2
1 + ++ 2
, h
*
((
)
& - 2 *#
- 2
$1 + ++ 2 ((! 4++ 2
$% , h )!"
, h
& x 2 #
2 $1 2 !
% r "
* & x 2 #
(( $1 2 !
) % r "
(17)
Lannullarsi del momento statico (x =0), il braccio sussistente tra centro di massa ed asse
elastico (centro di rotazione della sezione considerata), disaccoppia completamente i due gradi di
libert strutturali, come daltronde era gi ovvio dato che x compare solo nei termini fuori della
diagonale della matrice di massa; le forze non producono alcuna rotazione elastica e viceversa, i
momenti non provocano flessione. Le pulsazioni naturali sono allora date da:
12, 2
2h
=
x =0
1 2
1 + // 2
0 h
.
,,
-
+ 1 2
)1 // 2
)* 0 h
2
.(
,,&
-&'
$ 1
! 2
= # .
!" 2h
(18)
Per la ricerca degli autovalori, indeterminati a meno di una costante, si utilizza la prima
delle eq.(16) da cui 0 b / h 0 = ( 1 / x )(1 2h / 2 ) ; imponendo arbitrariamente che i due
autovettori abbiano h0/b=1 si ha che:
&
= $ 1
$
% x
E facile altres dimostrare che: x
2
h
2
1
,
**1
+
=0
)
''
(
#
1 , )! .
*1
'!
x *+ '("
&1 0#
=$
!.
%0 1 "
2
h
2
2
(19)
;
h
x
= .
r
s 1( , )
2 2
2
4. . 1
2
.2
0.11 , 0.15 .. 9
s 2( , )
2 2
100
100
10
10
s 1( , 0.3)
2
4. . 1
2
.2
s 1( , 0 )
1
s 2( , 0.3)
s 2( , 0 )
0.1
0.01
0.1
0.1
0.01
10
0.1
15
0.9
s 1( 1 , ) 10
s 2( 1 , )
0.8
s 1( 3 , )
0.7
0.6
10
s 2( 3 , )
0.5
0.5
0.5
0.5
40
0.2
rad
10.
sec
ORIGIN
0.6222
n
rad
100.
sec
Matrice di Massa
1
M
sec
eig
2
2
r .
1
M .K
rad
Matrice di Rigidezza
x r
0.
eigenvals( D)
1 .. 2
eig =
<i >
9.93
107.628
eigenvec D , eigi
sec
0.035 0.977
1.014 0
0
0.999 0.215
T
. K. =
0.556
10.065 0
0
59.799
sec
10.
0.2
rad
b
F
T1
0.75. m
F. 0.5. m
Mo
F
3
2
..M
Dyn( )
. . b
. K. ( 1
( 1).
1
M
Mo
T2
. . b
res( )
rad
100.
sec
x r
kg
1.225.
3
m
0.01
Matrice di Rigidezza
2
j. )
rad
0.
sec
Matrice di Massa
newton
1.
m
Dyn( )
0.6222
sec
ORIGIN
2
2
r .
h h
,
.. . 2
8 7
0.1
0.01
res( ) 1
res( ) 2
0.001
1 10
1 10
1 10
6
0
0.5
10
1.5
T1
1.
0.2
rad
10.
sec
newton
0.75. m
1.225.
Matrice di Massa
1
M .K
0.01
Mo
T2
M
4
. . b
eig
eigenvals( D)
<1 >
x r
rad
0.
sec
Matrice di Rigidezza
2
eigenvec D , eig1
<2 >
2
2
r .
eigenvec D , eig2
kg
m
F
. . b
rad
100.
sec
F. 0.5. m
Mo
0.6222
ORIGIN
T.
Fg
M.
T
.( 1
x( k , )
. 2
Mg
k, k
Kg
k, k
T
.T
j. ) . K.
h h
,
.. . 2
8 7
Coordinate Modali
10
1
x( 1 , )
x( 2 , )
0.1
0.01
0.001
0.5
11
1.5
k=
b
.
U
(20)
Per il sistema in esame, le forze siano considerate positive se verso lalto e i momenti
positivi se antiorari. Al secondo membro delle equazioni del moto si ha, quindi:
(21)
M M AE = M AC + L( 1 2 + a h )b .
(22)
L = L( NC) + L(C) ;
M = M ( NC) + M ( C) .
(23)
;
L( NC ) = b 2 U + h a h b
L( C ) = 2UbC( k )[ U + h + (1 2 a h )b ] ; (24)
*
1 2 Ub b 2 '%
M ( NC) = b 2 (a h b U + h a h b
8 & . (25)
)
M ( C) = 2Ub 2 ( 1 2 + a )C(k ) U + h + ( 1 2 a )b
12
h ( t ) = h 0 e j t
( t ) = 0 e jt ,
(27)
cos da ottenere:
'- j
$
h0
*
!+ k 0 b + a h 0 ( +
!
)
L
!
2 !,
= &
#
3
b
!2C( k ) - 1 + j h 0 1 + 42 1 a 1/ j a * ! ;
h
h 0(
+ k2 0
!%
b k 32
0k
,
) !"
'- j 0
$
h0
j 0 0 *
2
! +a h k a h b + a h 0 2 k + 8 ( +
!
)
M
!
2 !,
= &
#;
4
b
h
j
*
1
j
1
4
1
4
1
!2C(k )2 + a / 0 + 0 + 2 a / 0 !
h + 2
h
(
!%
b k 32
32
0, k
0 k ) !"
(28)
(29)
e quindi:
.
L
1
= 2 , 1 + 2C( k )
3
b
k
-
'h $
j
. 1 41
1 j ++ ! 0 !
+ a h + 2C( k) , 2 + 2 a h / a h )) & b # ; (30)
k
32
0k
-k
** !% 0 !"
.
,
M
41
1 j
2
= , a h + 2C( k )2 + a h /
4
b
,
32
0k
,
-
j
j 1
+
2
+ ah
+ +
) '! h 0 $!
k
2k 8
)& b #
41
1. 1 4 1
1; j + ) ! (31)
2C( k )2 + a h / , 2 + 2 a h / ) % 0 !"
32
0- k
32
0 k * )*
ah
In termini matriciali:
13
' L $
!! b 3 !!
C lh
2.
& M #= ,
-C mh
!
!
4
!% b !"
C l + '! h 0 $!
& b #.
C m )* ! !
% 0"
(32)
&
-1
* #
L( QS) = 2Ub $ U + h + + a h ( b! ;
,2
) "
%
(33)
-1
*&
-1
* #
M ( QS) = 2Ub 2 + + a h ($U + h + + a h ( b! .
,2
)%
,2
) "
(34)
L( QS )
&
#
$
-1
* !
= 2Ub $ U
+
h
+ + a h ( b ! , ed analogamente per il momento.
) !
$Stazionario Quasi StazionarioI ,2
Quasi
Stazionari
oII "
%
h
h
2Ub
2Ub 2 ( 1 2 + a h )
2U 2 b
2Ub 2 [( 1 2 a h )]
2U 2 b 2 ( 1 2 + a h )
2Ub 3 1 4 a 2h
. L( QS ) +
(( b 3 ((
2
(QS)
- ( QS ) * = A
(M
(
(, b 4 ()
[A ]
(QS)
&
$
=$
$
$%
. h0 +
( (
-b*
(, 0 ()
2j
k
2( 1 2 + a h )
j
k
2
j
#
+ (1 2a h )
2
!
k
k
!
j #!
&1
1
(1 + 2a h )$ 2 + ( 2 a h ) !
k "!"
%k
. (35)
L(S) =
1
1
V 2 (c)c l = V 2 (2b)c l = V 2 bc l .
2
2
(36)
L( QS) = b 3
2 2
2
= b 3 2 V 2 = 2bV 2 ,
2
k
b
(37)
= 2 ), si avr pertanto:
L(S) = V 2 bc l = 2V 2 b = L( QS) ,
(38)
come ci si aspettava.
15
(39)
( t ) =
1 dh ( t )
1
=
Re{h 0 j exp( jt )}.
U dt
U
(40)
d = L h dh = L h
dh
1
dt =
dt
T
T=
) dh ( t ) &
Re{L h ( t )}Re (
%dt . (41)
' dt $
stata considerata come unica forza aerodinamica, la portanza. Possiamo esprimerla come
rapporto, r, della portanza istantanea rispetto al quella che si otterrebbe per un moto del profilo a
velocit costante, h = h 0 ; inoltre si deve tenere presente che laerodinamica introduce uno
smorzamento e ci sar quindi un ritardo (o un anticipo) nella portanza prodotta dalla data
variazione dangolo dattacco:
& h #
1
L h ( t ) = L0 ( t )r exp( j), L0 ( t ) = U 2Sc l $$ 0 !! .
2
% U "
(42)
= U Sc l h 02 r cos .
2
(43)
Il segno del lavoro ci dice se loscillazione estrae energia dal fluido o viceversa: affinch si
manifesti il flutter necessario che sia >0. I valori di r e sono, in generale, funzioni di
REYNOLDS e MACH.
16
L
h
bU 0
b
2
(k) =
L( QS)
= k 2 + 2C( k ) jk ;
L
( QS )
r exp( j) =
h
bU 0
b
2
= 2 jk
(44)
jk
jk
+ C( k ) =
+ F( k ) + jG( k ) ; (45)
2
2
r
(
)
Im ( kk ( i) )
0.5
1.5
(
)
Re ( kk ( i) )
L
0
T=
h dt +
h dt .
(46)
Gli altri contributi della portanza sono trascurabili. Si noti che il secondo integrale
nellespressione del lavoro sempre negativo (cfr. paragrafo precedente).
Il primo termine nullo se h ed sono in fase, ed maggiore di 0 se h ed hanno una fase
di /2. Ci significa che solo se i due modi del profilo sono sfasati di un dato angolo sussister la
17
possibilit dellinsorgere del flutter. Le figure seguenti riportano in modo semplificato lo schema
fisico.
Fig. 3: Fase Nulla.
h = sin ( t ), = sin ( t )
/4
/2
3/4
5/4
Velocit
3/2
7/4
Portanza
h = cos(t ), = sin (t )
/4
/2
3/4
Velocit
5/4
3/2
7/4
Portanza
18
, m S ) & h( t ) # ,k h
*S
' % " + * 0
( t)! +
+ I ( $
dove il simbolo q denota la pressione dinamica per unit di apertura, [FL-3]. Quindi:
-mp 2 + k h
+
2
, S p
* & h # &0#
S p 2 + qScl
(% " = % " ,
I p 2 + k 2qSebcl ) $ ! $0!
(48)
a 4 p4 + a 2 p2 + a 0 = 0
con
a4 =
(49)
a2 =
a0 =
Discutiamo le possibili radici in base al segno dei coefficienti:
a4
a0
a2
Soluzioni
Tipo di Moto
>0
>0
<0
>0
>0
pI,1 = +j1
pI,2 = -j1
pI,2 = +j2
pI,2 = -j2
Armonico
Tipo di Stabilit
Neutrale
Controlla la tabella sulla divergenza
<0
<0
pI,1 = +1
pI,2 = -1
pI,2 = +2
pI,2 = -2
Aperiodico
Divergenza
19
pI,1 = +j
pI,2 = -j
pI,2 = +j
pI,2 = -j
Aperiodico
Armonico
Divergenza
pI,1 = +j1
pI,2 = -j1
pI,2 = +j2
pI,2 = -j2
Oscillatorio
1
U 2 [A( M , k )]{q}
2
2 [ ] [M ][ ]{ } + [ ] [K ][ ]{ } =
,
1
U 2 [ ]T [A( M , k )][ ]{ }
2
(50)
&
&
(51)
1
71
/ 2 5
/
6x
0
2
8
6 2 1h
/
6
00
7
2
h
1
+/
00
x 4
7
+ (1 + jg) 5
2 2
r 3
60
1
0 . 11
2 2, /
r - 0 x
' L $
h
'
$
0
0 4 . ! ! !! b 2 !!'! h 0 $!
#& b # ;
2 ,& b # = &
r2 2 3 ,- ! ! ! M !! !
% 0" !
4 % 0"
% b !"
1
x .
12h
+
(
1
+
jg
)
/
r2 ,00
0 . 12h
, /
r2 2 - 0 0
& L #
0 . '' b 2 '' &' h 0 #' &0#
"% b " = % "
, %
r2 2 - ' M ' ' ' $0!
'$ b 4 '! $ 0 !
(52)
0 . 53 &' h 0 #'
, %b"
r2 2 - 3 ' '
4$ 0 !
; (53)
20
/
- 2
- 8 h
- 67 0
.
0 5 81
3 6
r 2 4 7 x
2
2
h
x 5 / 1 , 8
+*6
r2 34 -. 2 *+ 7 0
1
/ 52
0 2 5 1 x 2
- h
+
- 34 0 r2 2 01 34 x r2 01
1
- / 1 , 5 2
2
0
5 C lh
-- -- 2 ** 3 h
2 20 3
. . + 4 0 r 1 4 C mh
' L $,
*
0 5 !! b 2 !! * '! h 0 $! (1 + jg) '! h 0 $!
# & b #=
& b # ; (54)
3 &
2 ! !
r2 2 4 ! M ! * ! !
% 0"
4 *% 0 "
%! b "! +
1
,
*
* '! h 0 $! (1 + jg) '! h 0 $!
*& b # =
& b #;
2 ! !
C l 2 * !% 0 !"
% 0"
*
C m 01 *+
(55)
(56)
p=
' k$
k
U
U
' k$
% U " + jU = % U " + j k = (k + jk )
& b#
b
b
b
& b#
&p 2 + 2h
$
% x
&p 2 + 2h
$
% x
C lh ( k ) C l ( k ) # 0* h 0 -*
# 0* h 0 -*
x
2&
/b,
!/ b , = $
C mh ( k ) C m ( k )!" * *
r2 p 2 + r2 2 " * *
%
. 0+
. 0+
C lh ( k ) C l ( k ) #
#
x
2&
!
$ C ( k ) C ( k )! = 0
r2 p 2 + r2 2 "
m
% mh
"
&p 2 + 2h
$
% x
C ( k) Cl ( k) #
#
2 & lh
!
$ C ( k ) C ( k )! = 0 ;
r2 p 2 + r2 2 "
m
% mh
"
x
21
(57)
(58)
Appendici
22
A. Funzione di Theodorsen
Funzioni di Hankel del Secondo Tipo
H( n , k )
Jn( n , k )
H0( k )
j. Yn( n , k )
1 .. 50
Funzione di Theodorsen
j. Y0( k )
J0( k )
ki
10
10
H( 1 , k )
H( 1 , k ) j. H0( k )
C( k )
1.1
F( k )
Re( C( k ) )
G( k )
Im( C( k ) )
0.05
Gk
i
0.1
0.15
0.2
0.4
0.5
0.6
0.7
0.8
0.9
F k
i
1
0.9
F k
i
0.8
0.7
0.6
0.5
0.001
0.01
0.1
10
k
i
0
Gk
i
0.1
0.2
0.001
0.01
0.1
k
i
23
10
J1( k )
j. Y1( k )
H0( k )
j. Y0( k )
J0( k )
H1( k )
H1( k )
H0( k ) . j
Distanza dell'asse elastico dall'origine del sistema di
riferimento in semicorde: . . .h.a
C L k , a h
2
k .a h
j. k
2. C( k ) . 1
i
1
2
1. , 1.15 .. 4
a h . j. k
k( i)
C Lh k , a h
10
2. C( k ) . j. k
1.4
4
Re C L ( k( i ) , 0.5)
Im C L ( k( i ) , 0.5)
Re C L ( k( i ) , 0.2)
Im C L ( k( i ) , 0.2)
0.01
0.1
10
k( i )
2
Re C Lh( k( i) , 0.5)
Im C Lh( k( i) , 0.5)
Re C Lh( k( i) , 0.2)
Im C Lh( k( i) , 0.2)
0.01
0.1
1
k( i)
24
10
H1( k )
H1( k )
H0( k ) . j
C M k , a h
1. 2
k
8
C Mh k , a h
2. C( k ) .
2
2
k .a h
1
2
a h . j. k
i
a h . j. k
1
2. C( k ) .
2
ah . 1
1
2
a h . j. k
2
k .a h
1. , 1.15 .. 4
k( i)
10
1.4
2
Re C M ( k( i) , 0.5)
Im C M ( k( i) , 0.5)
Re C M ( k( i) , 0.2)
Im C M ( k( i) , 0.2)
2
0.01
0.1
10
10
k( i )
3
Re C Mh ( k( i) , 0.5)
Im C Mh ( k( i) , 0.5)
Re C Mh ( k( i) , 0.2)
Im C Mh ( k( i) , 0.2)
1
0.01
0.1
k( i )
25
C M k , a h
1
2.
2
ah . 1
1
2
1. , 1.15 .. 4
k( i)
a h . j. k
10
C Mh k , a h
1.4
1
2.
2
a h . j. k
1.5
Re C M ( k( i) , 0.2)
Im C M ( k( i) , 0.2)
0.5
0.01
0.1
10
10
k( i)
Im C Mh ( k( i) , 0.2)
0.01
0.1
k( i )
26
2. 1
1. , 1.15 .. 4
k( i)
10
C Lh k , a h
a h . j. k
1.4
2. j. k
5.02377
Re C L ( k( i) , 0.5) 4
Im C L ( k( i) , 0.5)
Re C L ( k( i) , 0.2)
Im C L ( k( i) , 0.2)
0.00150713
0.01
0.1
0.01
1
k( i )
10
2.51189
4
Im C Lh( k( i ) , 0.5)
Im C Lh( k( i ) , 0.2)
0.01
0.1
1
k( i )
27
10
C. Funzione di Sears
j :=
H0 ( k) := J0 ( k) j Y0 ( k)
C ( k) :=
H1 ( k)
H1 ( k) + j H0 ( k)
H1 ( k) := J1 ( k) j Y1 ( k)
( k) := ( J0 ( k) j J1 ( k) ) C ( k) + j J1 ( k)
k := 0.01 , 0.02 .. 10
Funzione di Sears
0.2
0.1
Im ( ( k) )
0.1
0.2
0.4
0.2
0.2
0.4
Re ( ( k) )
28
0.6
0.8
$
2
2
& h
&
x 2
&%
$
&
&
+ 2 &
&
&
%
x 2
r2 2 2
'* h .
), 0 ,
)+ b / +
)(, ,
- 0 0
$ 1 11
4 j'
2& 2 + 3 ah 6 )
5k(
%k 2 2
11
4$ 1 1 1
4 j'
2 3 + ah 6& 2 + 3 ah 6 )
22
5% k 2 2
5k(
j
2
k
11
4j
2 3 + ah 6
22
5k
'
)*
.
), h0 , *, 0 .,
/
)+ b / = +
,
,0
0
), , 0
0
)
(
$
&
&
&
&
&
+ 2 &
&
&
&
&
&%
x 2
r2 2 2
1+ 2C(k)
'* h .
), 0 ,
)+ b / +
)(, ,
- 0 0
j
k
11
4j
ah + 2C(k) 3 + ah 6
22
5k
$ 1 11
4 j'
j
+ ah + 2C(k) & 2 + 3 ah 6 )
k
5k(
%k 2 2
j
j 1
ah ah2 + +
k
2k 8
11
4$ 1 1 1
4 j'
+2C(k) 3 + ah 6& 2 + 3 ah 6 )
22
5% k 2 2
5k(
29
'
)
)
)
)* h .
), 0 , *, 0 .,
/
)+ b / = +
,
,0
0
,
,
)
0
0
))
)
)(
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
Sommario
! Generalit sui fenomeni e sulle grandezze associate alla
diminuzione della lunghezza d onda caratteristica.
! Cenni riassuntivi sulla metodologia SEA e della sua
formulazione inversa.
! Tecniche e problematiche sperimentali:
"Sistema di Acquisizione
"Sistema di Eccitazione
"Media Spaziale
"Media sulle Condizioni di Eccitazione
"Media in Frequenza
40
60
logv ,
Lx Ly
,
2.1 3.1
logv ,
Lx Ly
,
.9 .87
logv ,
Lx Ly
,
.3 .7
80
CA
LO
LI
L
BA
LO
EG
! Densit Modale, n:
100
A
OB
GL
LI
120
! Smorzamento,
140
10
100
1000
1 10
2.
. . . rileggiamo il grafico !
40
Metodologie
Numeriche e Sperimentali
60
Lx Ly
logv ,
,
2.1 3.1
logv ,
Lx Ly
,
.9 .87
logv ,
Lx Ly
,
.3 .7
m
80
10
100
120
0.1
140
10
100
1000
1 10
Metodi Energetici
m>>1
Metodi Modali
m<<1
Densit Modale
Modi Propri
Frequenze Proprie
Energie Medie
Smorzamento Modale
Potenze Trasmesse
Risposta Modale
Energie Scambiate
2.
Cosa il SEA ?
S.E.A.
am
! Il sottosistema un dominio spaziale
orz
sm
nza
caratterizzato da una singola modalit di
ede
p
propagazione del disturbo strutturale (o
im
acustico):
nza
e
t
$onde longitudinali
po
F.E.M.
$onde flessionali
$onde di taglio
$onde di torsione
e
dal
mo
t
i
s
den
Anni
9
10
,1 + 12
*
+ 12
) & E1 # 1 & P1 #
%
"= % "
+ '
$P2 !
2
21 ( $E 2 !
N
$
& 1 + 1 j
j= 2
&
&
& 12
&
&
&
&
&
1N
&%
21
n112 = n 2 21
$( 1 + 12 ) n1
& simm.
%
21 n 2 ' * e1 - 1 * P1 - * e1 - * E 1 / n1 + . = + .; + . = +
.
( 2 + 21 ) n 2 )( ,e 2 / , P2 / ,e 2 / , E 2 / n 2 /
21
N
2 +
j=1; j 2
2j
i +
2 N
j=1; j i
'
)
)
* P1 .
) * E1 .
N 2 ) , ,
, ,
,E , 1 ,P ,
)+ 2 / = + 2 /
), , , ,
) ,E ,
,- PN ,0
)- N 0
N 1
N + Nj )
)(
j=1
N1
ij
n i ij = n j ji
12
Sistemi Classici
! Vari sistemi possono essere studiati con la formulazione
P2
P1
P12
Sottosistema 2:
E2
Sottosistema 1:
E1
P21
P1, DISSIPATA
) & E1 # 1 & P1 #
%
"= % "
+ '
$P2 !
2
21 ( $E 2 !
,1 + 12
*
+ 12
P2, DISSIPATA
21
13
14
E 11 + E 2 2
#
E & #
E &
= %1 + 2 ( %1 + 2 2 (
E1 + E 2
1 $
E1 ' $
E 11 '
E
+ 2
( a ): 1 = 1 = 1
E2
2
j=1
N
E
( b ): 1 = 0 = 2
E2
E
( c): 2 = 0 = 1
E1
j =1
15
16
$/- 1 + 12
!--
!. 12
#/
!-- 1 + 12
!"-. 12
,
21 *
*
+ *
2
21 +
$
'
$ '
! E (1)1 !
1 !P !
& = # 1&
#
! (1) ! ! 0 !
" %
"E 2 %
21 *
*
+ *
2
21 +
$
'
$ '
! E ( 2 )1 !
1 !0 !
& = # &
#
! ( 2) ! !P !
" 2%
"E 2 %
- E(1)1
+ 0
+ ( 2)
+E 1
+, 0
E (1)1
E (1) 2
E (1)1
E (1) 2
E ( 2 )1
E( 2)2
E ( 2 )1
E( 2)2
{} = [E
17
* ' 1 $ ' P $
1
(! !
E (1) 2 !12 ! 1 ! 0 !
(& # = & #
0
0 ( 21
! ! !P !
E( 2)2 (
) !% 2 !" % 2 "
0
(i)
( j)
] {P }
(i)
( j)
18
Sistema di Eccitazione
P=
1
Re{F v SISTEMA }
2
20
Sistema di Acquisizione
Media Spaziale
Acquisizione
Eccitazione
21
22
Acquisizione
Eccitazione
Eccitazioni
H
Acquisizioni
Piastre Irregolari
1 2 3 .. .. .. .. .. .. M
1
2
3
..
..
N
Generalmente la Matrice H,
per ogni sottosistema quadrata
ed generalmente 3<N<7
23
Medie Spaziali
Medie in Frequenza
! v2 $ *
10 1 N M '
)log # jk & ,
N M j=1 k =1 )( 10 " v 2REF % ,+
Media Logaritmica
Media Aritmetica
! 1
1
dB( vel) = 10 log10 #
2
" N * M v REF
v
j=1 k =1
2
jk
$
&
%
Z=
Powe r Input
Velocit
del Punto di Eccitazione
Forza
del Punto di Eccitazione
F
"1%
; n Re # &
v
$Z'
= P ( E )
Z=
F
; n = n 1 ++ n G
v
Powe r Input
Velocit
del Punto di Eccitazione
Forza
del Punto di Eccitazione
{} = [E
Ve loc it Me die
Velocit
per N punti di Acquisizione
26
(i)
( j)
] {P }
(i)
( j)
Ve loc it Me die
Velocit
per M punti di Eccitazione
Velocit
per N punti di Acquisizione
Velocit
per M punti di Eccitazione
27
28
Il CPP sempre
applicabile
Il CPP sempre
applicabile
n i ( i ) ij = n j( i ) ji
n i ( j) ij = n j( j) ji
n i ( i ) ij = n j( j) ji
29
ij non pu
essere negativo !
a
nz
rie
pe
Es
a
a
nz
nz
rie
rie
pe
pe
Es
Es
a
nz
rie
a
pe
nz
Es
rie
pe
Es
a
nz
rie
pe Strumento
Es
! E s t a t o d i m o s t r a t o c h e i
coefficienti di perdita per
accoppiamento possono diventare
negativi.
! Esso il risultato dell analisi di
una configurazione in cui il SEA
non pu a rigori essere applicato.
! Se un coefficiente ij diventa
negativo, ci significa che i due
sottosistemi interessati si trovano
in una instabilit energetica (il
sottosistema con maggiore energia
ne cede a quello che ne ha di
meno): questa situazione non
prevedibile dal SEA, ma pu essere
evidenziata sperimentalmente.
Numerico
a
nz
rie
a
pe
nz
Es
rie
pe
s
E
Strumento
Analitico
a
nz
rie
pe
Es
SEA
za
en
eri
p
s
Strumento E
Sperimentale
a
nz
rie
pe
s
E
31
32
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
6
I coefficienti di influenza energetici (dai modi
allenergia)
94
95
In particolare:
nr
Es
ns
(6.2)
!sr Es .
(6.3)
(6.4)
(6.5)
96
12 + 12 + . . .
12
6
12
12
12
6
+! 6
12
12
4
..
..
.
.
12
...
12 + 12 + . . .
..
.
...
7
7
. . .7
5
..
.
(6.6)
dove diag( ) indica una matrice diagonale. La somma degli elementi appartenti alle colonne della matrice dei fattori di perdita per accoppiamento e pari
a zero (la potenza netta scambiata in tutto il sistema e nulla). Si osservi che
le equazioni del SEA sono valide in senso stretto nel senso di medie dinsieme,
ovvero quando potenze ed energie sono mediate su un insieme di sistemi simili,
ma leggermente dierenti.
L = ! diag j
12
Se sono soddisfatte le condizioni necessarie, allora e valido il CPP e la potenza scambiata tra due sottosistemi accoppiati dipende solo dalle loro energie
modali e dai fattori di perdita per accoppiamento.
Inoltre, ci sono un certo numero di condizioni desiderabili, che permettono
di classificare il tipo di modello SEA realizzato. La prima condizione desiderabile, e la pi
u importante, riguarda i fattori di perdita per accoppiamento
relativi a due sottosistemi non fisicamente collegati (fattori di perdita per accoppiamento indiretto, o indirect coupling loss factor ), che dovrebbero essere
nulli. Le altre condizioni desiderabili riguardano, pi
u in generale, i fattori di
perdita per accoppiamento, che dovrebbero essere:
positivi;
indipendenti dai fattori di perdita per smorzamento (almeno nellapproccio
classico al SEA).
6.2 Modello ED
97
6.2 Modello ED
In un modello ED, la relazione che lega le energie e le potenze in ingresso
e
E = A Pin
(6.7)
Ci
o e in contrasto con un modello ED, in quanto i coefficienti di influenza
energetici dipendono implicitamente dalle propriet
a dellintera struttura.
98
dove
j (x)
!j2
1
!2 + i
j!
(6.9)
e la recettanza modale del modo j-esimo (!j e la frequenza naturale del modo
j-esimo). Come predetto, nella (6.9) si e assunto uno smorzamento viscoso,
ovvero
(6.10)
j = 2j !j = j !j .
In particolare, secondo il metodo dellhalf-power bandwidth (vedi Capitolo
4), si assume il fattore di smorzamento j proporzionale alla larghezza di
banda j in cui esso agisce (dimezzando la potenza), sebbene vari da modo
a modo e quindi dipenda dalla frequenza. Alternativamente, come mostrato
dalla (6.10), e possibile fare lipotesi di smorzamento strutturale con fattore di
perdita , ma ci
o comporta delle approssimazioni sulle equazioni che seguono
(che sono esatte nelle ipotesi di smorzamento viscoso).
Si osservi che i modi del sistema sono normalizzati rispetto alla massa
generalizzata, e per la proprieta di ortogonalita dei modi propri risulta essere
valida la (6.11)
Z
(x)
j (x) k (x)
jk
(6.11)
dove ( x) e la densit
a di massa e jk e il delta di Kronecker (che pero, in
questo caso, deve essere dotato di opportune dimensioni).
6.2.2 Densit
a di energia cinetica, potenza in ingresso e
conservazione dellenergia
Definita la risposta del sistema, e possibile esprimere il proncipio di
conservazione dellenergia in funzione delle caratteristiche modali del sistema.
6.2 Modello ED
99
La densit
a di energia cinetica mediata nel tempo in x = x2 e pari a
o
1 n1
DT (!, x1 , x2 ) = Re (x2 )! 2 W W =
2
2
X X 1
2
=
! 2 jk (!)
(x
)
(x
)|F
|
(x
)
(x
)
(x
)
j
1 k 2
2 j
2 k 2
4
j
k
(6.12)
dove
jk (!)
(6.13)
X1
1
Re{i!W (!, x1 , x1 )F } =
!2
2
2
j
dove
jj
=h
2
2
j jj (!) j (x1 )|F |
1
!j2
2
!2
j!
(6.14)
(6.15)
Integrando la densit
a di energia cinetica sullintero sistema, e possibile
calcolare lenergia cinetica totale. Dalla proprieta di ortogonalita (6.11), segue
che per ogni modo j-esimo vale
Pin,j (!) = 2
j Tj
(6.16)
100
1 2
|F | = Sf (!)(x) !,
2
(6.17)
dove la densit
a spettrale Sf (!) potrebbe essere una funzione della frequenza,
ma nel corso della trattazione e considerata costante per convenienza.
Fatte le dovute considerazioni sul tipo di eccitazione, e possibile calcolare
lenergia cinetica, mediata nel tempo e in frequenza, e la potenza in ingresso,
mediata nel tempo.
In particolare, la media in frequenza dellenergia cinetica del sottosistema r
per una potenza in ingresso nel sottosistema s e ottenuta integrando la (6.12)
sul sistema sorgente s e sul sistema ricevente r (in quanto stiamo parlando
di una densit
a di energia), oltre che sulla banda in frequenza considerata ,
ovvero
Z
1
T (r) =
DT (!, x1 , x2 )d! dx1 dx2 .
(6.18)
!2;x1 2s;x2 2r
Nel caso di eccitazione rain-on-the-roof, e possibile scrivere energia cinetica media nel sottosistema r e potenza in ingresso media nel sottosistema s
rispettivamente come
XX
(s) (r)
T (r) = 2Sf
(6.19)
jk jk jk
j
(r)
Pin = 2Sf
(s)
jj
(6.20)
jk (!) d!
(6.21)
jj
dove:
jk
1
!2 4
!2
Z
(r)
(x) j (x) k (x) dx
(6.22)
jk =
x2r
Lintroduzione dei termini jk e jk permette di suddividere la caratterizzazione spaziale da quella in frequenza. Nei paragrafi successivi tali termini
sono trattati pi
u approfonditamente.
6.2 Modello ED
101
Dalla (6.20) si evince come la potenza in ingresso sia la somma delle potenze in ingresso di ogni modo, mentre dalla (6.19) si evince come lenergia cinetica sia la somma dei termini modali incrociati che coinvolgono una coppia
di modi alla volta.
A questo punto, e possibile ricavare i coefficienti di influenza energetici.
Infatti, come gi
a scritto allinizio di questo paragrafo, il generico elemento Ars
e lenergia (mediata nel tempo e in frequenza) nel sottosistema r per unit
a
di potenza (mediata nel tempo e in frequenza) in ingresso nel sottosistema s.
Pertanto, e possibile scrivere
Ars =
E (r)
(s)
Pin
2T (r)
(s)
Pin
P P
j
(s) (r)
jk jk jk
.
(s)
j jj jj
(6.23)
Si osservi che i termini Ars sono esatti, perche non e stata fatta alcun tipo
di ipotesi oltre alla linearit
a del sistema e alleccitazione non correlata.
Integrali di frequenza
jk
jj
!k |
(6.24)
102
costante, quando i modi sono risonanti gli integrali di frequenza possono essere
approssimati come
1
(6.25)
jj =
8 j
jk
(!j + !k )2 ( j + k )
1
h
i
16 ! 2 + ! 2 2 + ( ! +
2
j j
k !k )
j
k
(6.26)
2 = jj
2
jk =
8
! !
! !
1+ j k
1+ j k
dove tutti i termini
essere riscritta come
jj
dove
jk =
jj jk
(6.28)
2
Mjk
1
=
2
2
1 + Sjk
1 + Mjk
Sjk =
Mjk =
|!j
!k |
1
=
Sjk
|!j
(6.29)
(6.30)
!k |
(6.31)
(r)
jk
(r)
jj
Il termine jk (Cross-mode partecipation factor ) indica la correlazione della coppia di modi (jk) allinterno del sottosistema r. Per la (6.11), il principio
di ortogonalit
a dimostra che i modi globali j (x) e k (x) non si accoppiano
(non scambiano energia) sullintero sistema essendo ortogonali, e quindi ogni
modo compie lavoro solo su se stesso. Diversamente, se si considera un singolo
sottosistema, i due modi riescono a scambiarsi energia localmente, ottenendo
6.2 Modello ED
103
(r)
potenza Sf dal j-esimo modo. In particolare, il termine jj fornisce unindicazione dellenergia cinetica immagazzinata nel sottosistema s dal modo j, e
pertanto permette di classificare se il modo j e globale o locale:
(s)
(6.35)
104
Ars =
(6.36)
P P
j
(s)
jk jk
P (s)
k
(r)
jk
(6.37)
jj
1
(I
!
(6.38)
dove:
I e la matrice unitaria;
lo smorzamento, rispettando le convenzioni del modello SEA classico, e
espresso in termini di fattore di perdita;
105
P P
j
rs
(r)
jk jk
P (s)
k
(s)
jk
(6.39)
jj
C = (I
(6.40)
= ( + 2 + 3 + . . . ).
(6.41)
Poiche e nulla la somma delle colonne della matrice , allora e nulla anche la
somma delle colonne della matrice C. Inoltre, riscrivendo la matrice come
!
1
= diag P (s)
(6.42)
j
dove
jj
(r) (s)
jk jk
(6.43)
segue che 2 , 3 , . . . e quindi C nella (6.41) sono ottenuti dal prodotto di una
matrice simmetrica per una matrice diagonale, come si evince dalla (6.42).
A questo punto e possibile osservare che la matrice X trovata e costituita
dalla somma di due matrici, in accordo con la formulazione SEA (6.6); in
particolare, la matrice !I (diagonale) e la matrice dei coefficienti di perdita
diretta legati alla dissipazione interna; la matrice C rappresenta la matrice dei
coefficienti di perdita per accoppiamento, con elementi fuori dalla diagonale
Crs = rs . Ovviamente, affinche la trattazione sia corretta, e necessario che
gli elementi di C soddisfino la relazione di consistenza, ovvero
nr Csr = ns Csr .
(6.44)
P (r)
La (6.44) e verificata se il rapporto j jj /nr e costante per tutti i sottosistemi, il che significa che i modi nella banda in frequenza scelta sono tali
che
(r)
(r)
= r
(6.45)
jj = E
jj
(r)
(r)
106
107
Area
Lunghezza
1
2
3
4
1
0.4
0.9
0.6
0.5171
0.8885
0.7072
1.0288
0.1646
0.2828
0.2251
0.3275
(r)
frazione di densit
a modale r . Dalle figure si nota che i jj sono pi
u o meno
uniformemente distribuiti a seconda del sottosistema considerato, ma in ogni
(r)
caso si percepisce che il valor medio jj coincide proprio con r . Tale convergenza non e verificata in assoluto, ma dipende dal numero di modi considerato,
come mostrato in Figura 6.5.
La Figura 6.5 mostra che la media in frequenza dei fattori di parteci(r)
108
alcuni fattori di perdita per accoppiamento indiretto sono negativi, e tendono asintoticamente a zero per elevati valori dei fattore di sovrapposizione
modale;
per tutti i coefficienti di accoppiamento (sia diretto che indiretto) vale
luguaglianza !nr sr = !ns rs , ovvero vale il principio di consistenza
richiesto da un approccio SEA.
(r)
Figura 6.4: fattori di partecipazione jj (+) per i primi 100 modi propri e (- -) frazione di densit
a modale r per le quattro aste.
109
Da quanto detto, si pu
o evincere che e possibile passare da un modello
EDA ad un modello SEA quando:
Pertanto, i risultati statistici ottenuti da un approccio SEA sono rappresentativi della risposta dellintero sistema strutturale solo se sono valide le
condizioni suddette.
110
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
article info
abstract
Keywords:
Structural similitude
Energy distribution approach
Scaling laws
A structural similitude is proposed for the analysis of the dynamic response of plates or
assemblies of plates. The similitude is defined by invoking the energy distribution
approach which allows the representation of all the fundamental parameters. Then, the
similitude laws are defined by looking for equalities in the structural responses. Two
test cases are herein discussed: the first involves a single plate response and the second
is related to an assembly of two plates. Only the bending waves are taken into account.
If the original damping values are kept, a complete similitude is defined in both the
cases which allows to enlarge or reduce independently the plate surfaces and the
thickness. An approximate similitude is defined if the damping is modified: in this case
only a mean response can be predicted in similitude.
& 2010 Elsevier Ltd. All rights reserved.
1. Introduction
It is well-known that the possibility to transport the same engineering problem in different scales can offer several
advantages. Thinking only to the geometry, a very small object could be investigated in a scale larger than the original one,
so making easy the location of a given set of sensors. On the contrary, very large structural components can be analysed in
a standard laboratory by using a concentrated set of excitation and acquisition instrumentation and with small surfaces to
be controlled.
The theory of the models and the analysis of the possible similitudes and analogies is a very large branch of
the engineering literature and cannot be replicated here. An interesting summary of the similarity conditions between
a full-scale model and a scaled one, by using the modal approach, is given in [1], with specific reference to dynamic
response. A more general view of the problem is in [2], even if the main textbooks about this subject are the work
in [3,4].
The approach of the similitudes between models is largely used in the aeroelastic tests where, during the wind tunnel
measurements, the aircraft component is designed to represent the same natural frequencies or flutter speed and/or the
wind-tunnel data have to be correlated to the flight-test ones, [5].
The relationships among mode shapes, natural frequencies, damping loss factors and energies are in [6] where the
energy distribution approach (EDA) is introduced. EDA was specifically used in order to predict the original and scaled
responses of linear dynamic systems, [79]. In detail, some investigations for a simple plate are in [7], and the extension to
two plates is in [8]; the scaling between structural components with different modal density (a beam coupled with a plate)
is introduced in [9]. The configurations were very simple but the results were very attractive when comparing them to
those obtainable with large finite element models.
In the present work, the idea is to enlarge the number of parameters in order to generate a complete similitude and to
set-up similitude laws for some coefficients able to represent some of the items to be investigated.
The adopted approach would formally invoke the process of full dimensional analysis, [3], but here it is preferred to
work with the energy distribution approach (EDA) that allows the investigation of a dynamic system via the modes and
natural frequencies in order to determine the power input and the energy associated to each subsystem.
The work, after these introductive remarks, presents in Section 2 a summary of EDA. This approach is then used to
define a similitude with a parent model tailored for representing some specific items under observation. Section 3 contains
the numerical investigations concerning the dynamic response of a simple bending plate and an assembly of two bending
plates.
0888-3270/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2010.10.004
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
Despite their simplicity, the results show promising applications of the proposed similitude laws.
List of symbols
AEIC
A
AEIC
rs
cB
E
E
f(P, t)
h
iu
mgen
n
NM
NP
NS
Pinput
P(r)
in
Sf
T(r)
t
x
Gjk
d
Z
m
n
r
fj
cjk
O
o
oj
with
AEIC
rs
PP
j
s r
k Gjk cjk cjk
s
j j j Gjj cjj
Zo
with
Eq. (1) is directly evaluated by using a modal expansion theorem. Therefore, it is an exact relationship representing an
energy balance. The use of an analytical relationship strengths the successive mathematical derivation instead of an energy
balance equation based on the SEA formulation.
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
The energy response is defined in terms of some global parameters which depend on the modal properties of the
system. The spatial coupling parameter for the generic rth subsystem is the following:
Z
r
cjk
This term depends on the global mode shapes acting within the rth subsystem: specifically, the cjk term measures the
generalised work inside the rth subsystem due to interaction between the jth and kth modes. The modes can be considered
r
lagrangian coordinates and then the cjk term expresses the work of the generalised force (represented by the jth mode) for
the generalised displacement (represented by the kth mode).
The global mode shapes are considered mass-normalised and then the cjk terms are dimensionless. Further, for a single
system, the only non-zero terms are the cjj thanks to the well-known orthogonality property.
The frequency dependent members are here recalled:
"
!
!#
Z
1
1
1
2
Gjk
o Re
4
d o:
4O o2O
o2j $o2 iuZj o2j
o2k $o2 iuZk o2k
The frequency response operators can be approximated, [6]:
Gjj ffi
Gjk ffi
1 p
,
8O Zj oj
oj ok 2 Zj oj Zk ok
16O o2 $o2 2 Zj o2 Zk o2 2
j
The term O represents a generic frequency band in which the system response is interrogated for a given excitation.
Two approximations can be invoked: they are named large and small terms. The first represents the situation in which
two modes well-overlap o2j ffi o2k ; the second is associated to the case in which two modes do not welloverlap,o2j $o2k 2 bZj o2j Zk o2k 2 . In the next section, both will be explicitly expressed when dealing with the similitude.
It has to be further noted that in EDA, (i) the loading was assumed to be proportional to mass density with zero crossspectral density Sf, [6], and then (ii) being Sf represented by the dimensional product of a squared force by a velocity
F2 LT$1 .
The expressions of the input power to the sth subsystem and the kinetic energy for the rth subsystem complete the
dynamic set for the analysis:
X
s
s
PIN
2Sf
2Zj oj Gjj cjj ,
7
j
T r 2Sf
XX
r
Gjk cs
jk cjk :
'
'
the distribution of the natural frequencies. For a generic assembly there is the possibility, in principle, of changing the
material of each plate. This choice enlarges the possibility of getting the same results with different similar
configurations, but it has been preferred to neglect the possibility to introduce this parameter, too, in order to keep the
analysis as simple as possible. It is also easy to consider that in an experimental laboratory, the variation of the
dimensions is a step easier than to change the material.
The plates are excited by concentrated harmonic forces, f P,t FodPe$iuot .
The structural damping is such that the system response can be obtained by using the real mode shapes and undamped
natural frequencies: more complicated models, based on the complex mode shapes, do not add further contributions to
the present developments and results.
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
The field of investigation is restricted to the analysis of the values of area, A, thickness, h, damping, Z and forces, F, since
it is further assumed that Fo F. The overline will always denote the scaled items. Then, a similitude is searched with
this set:
rA
A
,
A
rh
h
,
h
rF
F
,
F
rZ
Z
:
Z
It is assumed that the global mode shapes, after imposing the similitude, remain unaffected:
fj f j :
10
As consequence, one gets that the scaled spatial coupling parameters are equal:
Z
Z
r
r
c jk
r xf j xf k x dx
rxfj xfk x dx cjk
:
x2r
x2r
11
Having assumed to work with bending plates, the natural frequencies are modified with the parameter ro , being
ro rh =rA .
The scaled auto and cross-modal frequency response operators are here defined:
G jj ffi
G jk ffi
8OZ j o j
8OZj rZ oj ro
Gjj
1
,
rZ ro
3
ro
rZ oj ok 2 Zj oj Zk ok
4 o2 &o2 2 r 4 r 2 Z o2 Z o2 2
16O ro
o Z j
k
j
12
13
These cross terms cannot be directly posed in similitude: the quantities involved are different functions of the scaling
parameters defined for the natural frequencies and for the modal damping. The approximations before introduced are here
used: they are named large and small terms.
The situation in which two modes well-overlap o2j ffi o2k is as follows:
large
G jk
ffi
p oj ok 2 Zj oj Zk ok 1
,
rZ ro
16O
Zj o2j Zk o2k 2
14
whereas if the two modes do not well-overlap, o2j &o2k 2 bZj o2j Zk o2k 2 :
small
G jk
ffi
p oj ok 2 Zj oj Zk ok rZ
:
16O
ro
o2j &o2k 2
15
Then, the scaled auto and cross modal large and small terms are, respectively:
G jj
1
G ,
rZ ro jj
large
G jk
small
G jk
16
1
Glarge ,
rZ ro jk
17
rZ small
G
:
ro jk
18
It is evident the effect of a given modification of the damping between the original and scaled models. The role of the
cross-modal terms is not exactly reproduced if rZ a1, even for the same distribution of the natural frequencies. This was
already studied and discussed for a number of cases in which scaled finite element models were obtained just increasing
the original damping value, [8], in order to reduce the computational efforts.
According to the choice of scaling parameters, the excitation can be scaled, [6]:
S f Sf
rF2
rmass
19
being
rmass rA rh :
20
For the sake of completeness, it is useful to derive all the remaining parameters of the EDA with the scaling procedure:
!
"
2
r2 X 1
s
s rF
P IN 2Sf F
Gjj cs
PIN
,
21
jj
rmass j ro rZ
rmass
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
ffi 2Sf
!
"
rF2 XX 1
rF2
r
Gjk cs
cjk T r
:
jk
rmass j k ro rZ
rmass ro rZ
22
Ars
:
ro rZ
23
Rather than discuss the overall development, it is here preferred to apply the scaling procedure to simple test cases in
which the effect of the several parameters can be easily analysed and discussed.
fj Sfj R
iuoFo X
:
2 %o2 iuZo2
mgen
o
j
j
j
24
The symbol mgen denotes the plate generalised mass. The natural radian frequencies can be obtained from the relation:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi "
! "2 ! "2 #
2
pjy
Eh
pjx
oj
,
25
Lx
Ly
12r1%n2
and the mode shapes are given by
!
" !
"
x
y
fj S sin jx p S sin jy p S :
Lx
Ly
26
The jth mode is mapped into wavenumber space with the two associated integers jx and jy. It is very simple to check
that the scaled response can be written as follows:
vo; S,R
fj Sfj R
rF iuoFo X
;
2
2
2r
rmass mgen
oj ro %o2 iuZo2j ro
Z
j
27
Table 1
Derived parameters for single plates or assemblies of plates.
Natural frequencies
later defined
Bending wavespeed
cB ph1=2
Mass
Modal density
Modal overlap factor
mpAh
A
np
h
mpnZ
Frequency terms
Eq. (16)
Eq. (17)
Eq. (18)
Spectral density
Eq. (19)
Energy
Eq. (22)
Square velocity
T
v2 p
m
ro rh rA%1
1=2
rc rh
rmass rh rA
rn = rh% 1 rA
rm rh%1 rA rZ
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
rA
12r1#n2
Lx
Ly
28
From Eqs. (27) and (28), it descends that only for rh = rA the scaled plate will work with the same eigensolutions as the
original ones. To get exactly the same local response, it has to be verified that rZ 1 and rF rmass rh rA : this choice is
unique, if only the area and the thickness are modified.
Figs. 1 and 2 illustrate the local behaviour of the dynamic response (modules of the local velocity) for two scaled
models. In Fig. 1, the plate is larger and thicker than the original one; in Fig. 2, the similitude procedure leads to a reduction
of both plate surface and thickness. On the right side of each figure, the complete set of parameters is reported: this is also
reported in Table 2.
Fig. 1. Single plate response (I-A) for a generic couple of sourcereceiver points: velocity (m/s) vs. frequency (Hz).
Fig. 2. Single plate response (I-B) for a generic couple of sourcereceiver points: velocity (m/s) vs. frequency (Hz).
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
Table 2
Configurations for the single plate response.
#
rZ
rA
rh
rF
ro
rmass
I-A
I-B
I-C
I-D
I-E
1
1
2
1
2
4
0.4
0.4
0.7
0.5
4
0.4
0.4
0.5
1.0
16
0.16
0.226
0.296
1
1
1
1
0.714
2
16
0.16
0.16
0.35
0.5
Fig. 3. Single plate response (I-C) for a generic couple of sourcereceiver points: velocity (m/s) vs. frequency (Hz).
The variation of the damping changes the response amplitude, Fig. 3, as expected; there, the variations of the rh and rA
are still such that the natural frequencies are kept: this is true in all the frequency range. The real poles of the system are
kept while the amplitudes are reduced in this case of augmented damping.
Even keeping the same material, a plate larger or smaller than the original one can be used for numerical modelling
and/or for experimental purposes with the obvious advantages that each configuration can allow.
Variations of the thickness and the area such that rh arA , lead to a modification of the distribution of the natural
frequencies, Figs. 4 and 5. In both figures, the top parts present the results of the simple numerical interrogations of the
original and scaled models, while in the bottom ones, the re-modulated responses are shown.
Specifically, the curves referring to the similitude responses are plotted as follows:
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
Fig. 4. Single plate response (I-D) for a generic couple of sourcereceiver points: velocity (m/s) vs. frequency (Hz); (a) direct response, (b) similitude
response.
h1
h2
rh
h1
h2
A1
A2
rA
A1
A2
Z1
Z2
rZ
Z1
Z2 ,
F1
F2
rF
F1
F2
29
977
Fig. 5. Single plate response (I-E) for a generic couple of sourcereceiver points: velocity (m/s) vs. frequency (Hz); (a) direct response, (b) similitude
response.
The similitude involves all the parameters in the same way; clearly, it is possible to study also the case in which each of
the parameters, belonging to the given plate, is scaled independently, but this falls outside the aims of the present work.
The terms in Eq. (13) are the problem of any scaling choice since the Gjk do not scale as the Gjj and further, by invoking
the approximations it was shown that the terms Gsmall
scale with different combination of ro and rZ if compared with
jk
Glarge
.
jk
Again, a complete similitude can be obtained only by using rZ 1.
Here, the mean square out-of-plane velocities, W1 and W2, of each plate are investigated as results of an average over
three mechanical driving points on the first plate.
The results of a complete similitude are given in Figs. 7 and 8, whereas Fig. 9 presents those coming from an
approximate similitude: the latter approach was named asymptotic scaled modal analysis, [8]. Again, all the results are
presented by using NP points for each graph and the frequency range of analysis is the half of the natural frequency of the
last resonating mode. Table 3 presents a summary of the analysed configurations.
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
Fig. 7. Two plates assembly (II-A): mean square velocity (m2/s2) vs. frequency (Hz).
4. Concluding remarks
The structural similitude herein presented and discussed with analytical models, allows reproducing the local response
of plates for any given choice of area, rA, and thickness, rh. This leads to the definition of similar plates in which the only
parameter to be evaluated is the excitation level, rF. It is always possible to switch from the original model to the parent
one and to perform the reversal step, if the similitude is complete.
Specifically, in the present work the distribution of the altered natural frequencies is represented by ro rh =rA , while
"1
the cost of the simulation of the bending wavelength is ro
rA =rh . The scaled model presents only in principle a reduction
of the computational costs since the re-modulation of the frequency axis requires the evaluation of an increased number of
modes resonating in a frequency range wider than the original one.
The only way to reduce the computational costs is to work by accepting the modification of distribution of the natural
frequencies, and then a variation of the damping can be accordingly defined in order to get the mean square response. This
can be done in the frequency range dominated by the high values of the modal overlap factors. In this case, one will work
with an approximated similitude: it was named ASMA, asymptotic scaled modal analysis, [8].
979
Fig. 8. Two plates assembly (II-B): mean square velocity (m2/s2) vs. frequency (Hz).
Fig. 9. Two plates assembly (II-C): mean square velocity (m2/s2) vs. frequency (Hz).
From an experimental point of view, the defined similitude offers numerous advantages since, in case of the complete
one, all the dynamic items (natural frequencies, mode shapes and modal response) can be easily reproduced on the
desired scale.
Letter to the Editor / Mechanical Systems and Signal Processing 25 (2011) 969980
Table 3
Configurations for the assembly of two plates NP= 150, AT = (0.198, 0.162), FT = (1,0).
#
NM
rZ
rA
rh
rF
II-A
II-B
II-C
135, 108
570, 456
400, 320
0.03, 0.03
0.05, 0.05
0.05, 0.01
0.001, 0.003
0.001, 0.005
0.005, 0.001
1
1
2
0.2
0.5
0.5
0.2
3
1
0.04
3.674
1
Further steps will involve the adoption of finite element models and experimental measurements in order to face with
more realistic assemblies.
References
[1] J. Wu, Prediction of the dynamic characteristics of an elastically supported full-size plate from those of its complete-similitude model, Computers
and Structures 84 (2006) 102114.
[2] P. Kroes, Structural analogies between physical systems, British Journal for the Philosophy of Science 40 (1989) 145154.
[3] E. Szucs, Similitude and Modelling, Elsevier Science Ltd, 1980, ISBN:0444997806.
[4] S.J. Kline, Similitude and Approximation Theory, Springer-Verlag, 1986, ISBN:0387165185.
[5] C.H. Wolowicz, J.S. Bowman, Jr., W.P. Gilbert, Similitude requirements and scaling relationships as applied to model testing, NASA Technical Paper
1435, 1976.
[6] B.R. Mace, Statistical energy analysis, energy distribution models and system modes, Journal of Sound and Vibration 264 (2003) 391409.
[7] S. De Rosa, F. Franco, A scaling procedure for the response of an isolated system with high modal overlap factor, Mechanical Systems and Signal
Processing 22 (2008) 15491565.
[8] S. De Rosa, F. Franco, On the use of the asymptotic scaled modal analysis for time-harmonic structural analysis and for the prediction of coupling loss
factors for similar systems, Mechanical Systems and Signal Processing 24 (2010) 455480.
[9] S. De Rosa, F. Franco, T. Polito, Analysis of the short and long wavelength coupling through original and scaled models, in: NOVEM 2009, Noise and
Vibration: Emerging Methods, Keble College, Oxford (UK), April 2009, paper No. 28.
[10] R.H. Lyon, R.G. De Jong, Theory and application of statistical energy analysis. Butterworth-Heinemann, ISBN: 0750691115, 1995.
12 April 2010
Available online 14 October 2010
! Corresponding author.
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
Abstract
This work presents the extension of an existing procedure for evaluating the waveguides and the dispersion curves of a laminate made up of
thin orthotropic composite plate arbitrarily oriented. The adopted approach is based on one-dimensional finite-element mesh throughout the
thickness. Stiness and mass matrices available in literature for isotropic
material, are here reported in full expanded form for the selected problem. The aim of the work is the development of a tool for the simulation
of the most common composite materials. The knowledge of the wave
characteristics in a plate allows the correct sizing of the numerical mesh
for the frequency dependent analysis. The development of new stiness
matrices and the analysis for dierent heading angles are here detailed
for taking into account the general anisotropic nature of the composite.
The procedure concerns a standard polynomial eigenvalue problem in the
Corresponding
Author: e.barbieri@bath.ac.uk
The wide and increasing use of innovative materials in the transportation engineering is one of the most fascinating challenges of the material science. In the
aerospace field, this challenge can be defined as always open since the manufacturers are continuously looking for stier, robust, long-life and lighter structural
components. The need of such material performances has driven for a long time
the attention to the fiber-reinforced composite materials and now they are a
standard in several fields of transportation engineering design. One of the main
problem of these innovative materials is their vibroacoustic behavior, since sometimes the lightness requirement is in conflict with the acoustic target. Hence,
in the design phase there is the need of simulating the dynamic behavior of
innovative materials. This target can be obtained by using the deterministic
methods, (Cook et al., 1989), and energy ones, (Lyon, 1975). The deterministic
techniques, such as the Finite Element Analysis (FEA), work by discretising the
given wavelength. In principle, they can be used for evaluating the response at
any excitation frequency, but the computational cost (CPU time) can become
easily unacceptable. In absence of any analytical development, the deterministic techniques model directly the wavelength, by assigning at least five solution
points (four elements) for each complete wave (ref.(Cook et al., 1989) and (Cremer et al., 2005)). For increasing excitation frequency, the response becomes
global, that is any quadratic mean can represent the overall behavior, and the
possibility to discriminate the dynamic response among dierent points in space
is lost. The best technique working under these conditions is the well-known
Statistical Energy Analysis (SEA), ref.(Lyon, 1975). In (Ghinet et al., 2005) for
example, a SEA method for the transmission loss of sandwich shells is illustrated,
in which the dispersion properties of laminates are exploited. (Finnveden, 2004)
studied the waveguides in thin-walled structures with a finite element formulation in order to compute group velocity and modal density as input to a SEA
model. Wave propagation in laminated composite plates and rods (Baz, 2000),
has been treated in depth by numerous authors in the past. (Datta et al., 1988)
have applied a stiness method for a laminate made of transversely isotropic
laminae, while in (Nayfeh, 1991) dispersion curves for anisotropic laminates are
analitically extracted by the means of the transfer matrix method. A similiar
method can be found in (L. Wang, 2007) for the evaluation of the group velocities
and their applications in nondestructive techniques. (Chitnis et al., 2001) used
a higher-order theory displacement based formulation. In-plane elastic waves
properties. Very recently two authors stressed this point with great attention
(Shorter, 2004),(Mace et al., 2005). (Shorter, 2004) proposed an efficient approach for simulating an infinite flat plate, in which a low-cost one-dimensional
finite element was used for simulating the propagating waves inside the materials. Further, (Mace et al., 2005) proposed to use directly a finite element
model assembled for evaluating the modal response for getting the dispersion
curves. Two dimensional approaches are also present in literature, (Johnson
and Kienholz, 1982), and the work by (Heron, 2002) has to be considered as
one of the first addressing the problem of the dynamic response of a generic
laminate. An extensive analysis of the related literature is reported in (Shorter,
2004). Nevertheless, it is worth mentioning the applications of spectral finite
elements in wave propagation for laminates, for example (Roy Mahapatra and
Gopalakrishnan, 2003),(Mahapatra et al., 2006),(Chakraborty and Gopalakrishnan, 2006) in the frequency domain, whereas a time-domain based spectral
element with high order shape functions are used in (Kudela et al., 2007). The
present work is a straight extension of the approach proposed by (Shorter, 2004).
A one dimensional finite element model is developed for a generic plate made
by a laminate composite. The approach in (Shorter, 2004) assumes an isotropic
stress-strain relationship through the use of two independent variables for each
material. In the present work the proper stiness and mass matrices are developed for each lamina introducing a local reference system, and the whole
laminate is assembled in a global reference system. At that point a Spectral
Finite Element Approach is used (SFEA), (Shorter, 2004),(Mace et al., 2005).
The spectral term refers to the fact that the method is based on the wavenumber at a given excitation frequency rather than the classical analysis of natural
frequencies. The SFEA used here is the same presented in (Shorter, 2004): a
full three-dimensional displacement field within the laminate with 1D elements.
, for a generic
composite plate this is not true because of its anisotropic nature. Indeed, in an
2D orthotropic material for example, named x and y the principal directions of
elasticity, being Ex 6= Ey , the wavenumber at a fixed frequency changes with
the angle from x to y according to an elliptic pattern. Then, if an isotropic
plate is considered, that pattern is circular. In section 3 an analytic formulation
for a thin composite plate is derived, following the assumptions of the classical
laminate plate theory (CLPT) (Jones, 1999). A displacements 3D free wave is
then imposed to the thin plate leading to a polynomial eigenvalue problem in
k where k is the wavenumber. Bending, shear and longitudinal waves can be
easily identified from the resulting equations. Further, a spectral finite element
method (SFEM) will be used for getting the dispersion curves for a generic
composite plate and then compared to the analytical approach. These results
are showed in section 6 in form of polar patterns at fixed frequency and spectra
at fixed heading angle. The section 4 is devoted to the overall development of
the required SFEM matrices. The section 5 is centered on the solution of the
characteristic equation, and the results are presented in section 6. These results
are for both an homogeneous and a composite plates. The work has been concluded in section 7 where some considerations are given about the validity of the
CLPT. It is there demonstrated how the present numerical method can work
with any configuration, overcoming the limitations of the standard theoretical
models.
(1
2 ) 2
!
E
ks2 =
2(1 + ) 2
!
E
kb4 =
12(1 2 ) 2
!
Eh2
(1)
E
(1 2 ) ,
ing stiness D =
Eh
12(1 2 ) .
E
2(1+) ,
@!
;
@kg
Cg
f
(2)
and the modal density for a plate of area A (ref. (Lyon, 1975) )
ng (!) =
A k(!)
2 2 cg (!)
(3)
analytical approach.
According to the assumptions of the classical thin plate theory, in a plane stress
problem, the following equations of equilibrium can be written (Timoshenko
and Goodier, 1970)
@Nx
@Nxy
+
@x
@y
s u
=0
(4)
@Nxy
@Ny
+
@x
@y
s v = 0
(5)
@ 2 Mx
@ 2 My
@ 2 Mxy
+
+2
2
2
@x
@y
@x@y
s w
=0
(6)
where Nx , Ny , Nxy are shearing forces per unit length and Mx My bending
moments per unit length and Mxy twisting moments per unit length, while s
is the surface mass density. For a composite plate, the following relationships
between forces - moments and strain - curvatures can be estabilished (Jones,
1999)
N = A0 + B
(7)
M = B0 + D
(8)
6 x
6
0 = 6
6 y
4
xy
@u
@x
6
6
7 6
7 6
@v
7=6
7 6
@y
5 6
6
4 @u @v
+
@y
@x
7
3
7
7
7
7
7
7
7
7
5
(9)
6 x
6
=6
6 y
4
xy
6
6
7 6
7 6
7=6
7 6
5 6
6
4
@2w
@x2
7
7
7
2
@ w 7
7
@y 2 7
7
7
2
@ w 5
2
@x@y
(10)
3 2
3
u(x,
y,
t)
U
6
7 6
7
6
7 6
7
6 v(x, y, t) 7 = 6 V 7 ej[k(cos(
6
7 6
7
4
5 4
5
w(x, y, t)
W
where
(11)
is the heading angle of the wave, u,v and w are respectively the dis-
placement fields in the x,y and z axis, whereas U ,V and W are the respective
magnitudes of the propagating wave. Substituting equations (7), (8), (10), (9)
in (4), (5) and (6) and with the assumed displacements field (11), the following
polynomial eigenproblem in k is obtained
2
6
6
6
6
4
32
3
2
3
jk 3 LT BP 7 6 U 7
U
6
7
76
7
6
7
7 6 V 7 + s ! 2 I 6 V 7 = 0
76
7
6
7
54
5
4
5
k 4 PT DP
W
W
k 2 LT AL
jk 3 PT BL
where
PT =
6 cos( )
6
L=6
0
6
4
sin( )
7
7
sin( ) 7
7
5
cos( )
(12)
(13)
(14)
(15)
k 4 PT DP W + s ! 2 W = 0
(16)
U 7
6 U 7
2 6
k 2 LT AL 4
5 + s ! I 4
5=0
V
V
It can be seen from the equation (15) that, being coupled shear-longitudinal
p
waves, k / ! while from equation(16) for the flexural wave follows that k / !.
Indeed, from equation (16) directly derives that the flexural wavenumber is
kf ( , !) =
r
4
s p
!
T
P DP
(17)
For the equation (15), if l and s are the two eigenvalues of the matrix LT AL,
then
ks ( , !) =
s
!
s ( )
ke ( , !) =
s
!
e ( )
(18)
Cf = 2
4
4.1
s
4
PT DP p
!
s
Cs =
s ( )
s
Ce =
e ( )
s
(19)
Assembly of Matrices
Variational Formulation
With the intent of estabilish a finite element model, the Hamiltons principle is
used
(T + U )dt = 0
(20)
where T is the kinetic energy and U is the strain energy. If a time average over
a period is considered, then
1
T =
2
d H d d
(21)
where d is the displacements field, is the mass density and the superscript H
means hermitian
=1
U
2
1
s d =
2
eH Cx
(22)
where e is the vectorized strain tensor and s is the vectorized stress tensor in
the laminate reference system, Cx
the laminate reference, is the space domain made of a single element of length
L and an arbitrary rectangular domain in the plane xy.
4.2
3
u(z)
6
7
6
7 j[!t
6
d(x, y, z, t) = 6 v(z) 7
7e
4
5
w(z)
(23)
The equation (23) resemble the equation (11), with the only exception that
in equation (11), the displacements refer to a plane stress problem, while in
equation (23) a full 3D stress-strain problem is considered. The displacements
at x = 0 for the single element of length L can be approximated by the following
10
Figure 1: 1D mesh
one-dimensional interpolating function
2
3 2
u(z)
0
0
6
7 6 Ni (z)
6
7 6
6 v(z) 7 = 6 0
Ni (z)
0
6
7 6
4
5 4
w(z)
0
0
Ni (z)
32
3
q
76 u 7
76
7
7 6 q 7 = N(z) q0
76 v 7
54
5
qw
(24)
with N(z) the matrix of shape functions (size 36) and q0 the vector of complex
amplitudes of nodal displacements (Cook et al., 1989)
Ni (z) =
11
z
L
z
L
(25)
6 qu
6
6 q
6 v
4
qw
4.3
3
u
6 1 7
6
7
3 6 u 7
6 2 7
6
7
7 6
7
7 6 v1 7
7=6
7
7 6
7
5 6 v2 7
6
7
6
7
6 w 7
6 1 7
4
5
w2
(26)
6 xx
6
6
6 yy
6
6
6 zz
e(x, y, z, t) = 6
6
6 yz
6
6
6
6 xz
4
xy
6
6
6
7 6
7 6
7 6
7 6
7 6
7 6
7 6
7=6
7 6
7 6
7 6
7 6
7 6
7 6
5 6
6
6
4
3
@
@x
0
0
@
@y
@
@z
@
@y
@
@z
@
@x
@
@z
@
@x
@
@y
7
7
7
7
7
7
7
7
7
7
7d
7
7
7
7
7
7
7
7
7
5
(27)
12
(28)
6
6
6
6
6
6
6
6
6
6
F(k, , z) = 6
6
6
6
6
6
6
6
6
4
7
7
7
0
7
7
7
@Ni
7
7
7
@z
7
7 (29)
jk sin( )Ni (z) 7
7
7
7
jk cos( )Ni (z) 7
7
7
7
5
0
@Ni
@z
@Ni
@z
6
6
6
6
6
6
6
6
6
SL = 6
6
6
6
6
6
6
6
6
4
CL = SL 1
(30)
1
E11
21
E22
31
E33
12
E11
1
E22
32
E33
13
E11
23
E22
1
E33
1
G23
1
G31
1
G12
i, j = 1, 2, 3
3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5
(31)
(32)
= T()
CL RT()R
yL axis and x
(33)
y axis (the z
axis is the
6
6
6
6
6
6
6
6
6
T() = 6
6
6
6
6
6
6
6
6
4
cos2 ()
sin2 ()
sin2 ()
cos2 ()
cos()
sin()
2 sin() cos()
7
7
2 sin() cos() 7
7
7
7
0
0 7
7
7
7
7
sin()
0 7
7
7
7
cos()
0 7
7
7
5
2
2
0 cos () sin ()
0
(34)
14
and
6
6
6
6
6
6
6
R=6
6
6
6
6
6
6
4
3
0 7
7
0 7
7
7
7
0 7
7
7
0 7
7
7
0 7
7
5
2
(35)
Then, the time averaged kinetic and the strain energies can be evaluated as
1 2 H
! q0
2
T =
= 1 qH
U
2 0
N(z)T N(z)d q0
(36)
F(k, , z)H Cx
y F(k,
, z)d q0
(37)
Substituting equation (36) and equation (37) in equation (20), it can be obtained
!2 T
1
q0 Mq0 + qT0 K(k, )q0 = 0
2
2
(38)
(39)
N(z)T N(z)dz
K(k, ) =
FH (k, , z)Cx
y F(k,
, z)dz
(40)
where L is the length of the element. Furthermore, the equation (40) is consistent, i.e. they return for = 0 ,E11 = E22 = E33 = E, 12 = 31 = 23 = and
G12 = G31 = G23 =
E
2(1+)
15
(41)
A single element has 6 degrees of freedom. The number of the eigenvalues that
could be extracted from such an element is then 6, which is the exact number of
the propagating waves (3 for both directions of travel). Moreover, even only one
element can give good results for the dispersive relations in a homogeneous plate,
as it will showed later. Then a rule of thumb for discretizing the thickness could
be one or two elements per layer of the laminate. Being the proposed mesh a 1D
discretization, all these matrices can be easily assembled in a straightforward
way. Indeed, given the elastic properties of each layer, the matrices K0 , K1 ,
K2 and M are readily calculated. Afterwards, these matrices are expanded to
the size of the total number of nodes, allocated in the right position and then
all summed in order to give the assembled matrices of the whole thickness.
8 q0
(42)
! 2 M]q0 = 0
(43)
value problem for k and the dispersion curves result in varying the frequency,
while at fixed ! the polar pattern is derived when the heading angle range be16
tween 0 and 360 . This kind of problem can be easily transformed in a linear
one by the positions
q1 = kq0
C2 = K2
C1 = K1
(44)
C0 = K0
!2 M
(45)
so that
2
6
4
0
C2
I
1
C0
C2
C1
32
3
q
76 0 7
54
5
q1
6 I
k4
0
32
3
0 7 6 q0 7
54
5=0
I
q1
(46)
axis.
17
Results
6.1
6.1.1
Homogeneous Plate
In this section the consistency with the isotropic case is shown together with the
comparison with the analytical formulas (1) for an aluminium plate of thickness
0.0012m, density 2700kgm
10
Wavenumber [m]
(a)
1
10
(b)
10
(c)
10
100
200
300
400
500
600
700
800
900 1000
Frequency [Hz]
Figure 3: Homogeneous Plate: continuous line, analytical result; dotted line, SFEM
results; (a) bending wave, (b) shear wave, (c) longitudinal wave
from Figures (3) and (4), there is a perfect agreement with the analytical results
for both spectra and polar patterns, even if only two linear elements are used;
the FE results converge using simply one element, too. It must be pointed out
that the patterns are circular, due to the isotropic nature of the material, so the
dispersion curves in Figure (3) are the same for every heading angle. This is
18
not the case for the composite plate, where a heading angle must be specified.
Figure 4: Polar Patterns for the Homogeneous Plate: Continuos Line: Analytical; Diamond Line: SFEM
6.1.2
Composite Plate
The engineering properties on the plate are resumed in table 1. The transverse
properties were set in order to satisfy the assumptions of plane stress on which
the formulas used in section 3 are based. One element per layer was used in
the analysis. These results agreed with the analytical ones, again with spectra
as in Figures (5) and patterns as in Figures (6). As it can seen from these
figures, the wavenumbers are strongly dependent from the heading angle, due
19
E11
E22
G12
12
Thickness
Lay-up Sequence
Material A
65e9 Pa
65e9 Pa
3.86e9 Pa
0.05
1467 kgm 3
2.37 mm
[0A, 45B, 90B, 45B, 0B, 90B]S
Material B
145e9 Pa
7.79e9 Pa
4e9 Pa
0.34
1550 kgm 3
2.37 mm
20
=0
= 45
= 90
Figure 5: Dispersion Curves for the Composite Plate: Continuos Line: Analytical; Symbols Line: SFEM. (f) flexural, (s) shear, (l) longitudinal
21
Figure 6: Polar Patterns for the Composite Plate: Continuos Line: Analytical;
Diamond Line: SFEM
22
Figure 7: Eigenvector Plots for the Composite Plate at = 0 . Axis scales are
equal, the thickness is much smaller than the wavelength
23
In this section the limits of the CLPT are illustrated. At this purpose, a characteristic adimensional number could be introduced
=
(47)
is the smallest wavelength (usually
the flexural one). When << 1 the plate could be considered thin, since the
h is much smaller than the characteristic length of the problem, which is the
wavelength of the propagating wave. Since the
2
h
(48)
New waveforms appear when ! >> !lim as it can seen from the dispersion plot
in Figure 8 and from Figures 9. The bending wave in Figure 9(a) is slightly
dierent from the one in Figure 7(a) and its dispersion curve diers from the
analytical theory. The wave 9(b) cannot be predicted from the CLPT since
according to the eigenproblem (12), only 6 eigenvalues (3 for both directions
of travel) are obtainable, whereas 8 eigenvalues come from the SFEM above
17 kHz. Thus, when the frequency increases, a SFEM approach is useful to
overcome the limits of the CLPT.
24
Figure 8: Dispersion Curves for the Composite Plate: Continuos Line: SFEM;
Dash-dotted Line: Analytical. (s) shear, (l) longitudinal, (a) and (b) new waveforms
25
(a)
(b)
26
Conclusions
An extension to anisotropic laminates of an existing FEA based waveguide procedure was illustrated. The laminate is made up of orthotropic layers arbitrarily
oriented. Using Hamiltons variational principle, mass and stiness matrices
were developed. The followed approach is based on a one dimensional finite
element discretization throughout the thickness,then a full 3D stress-strain relationship is required. An orthotropic constitutive relation is used so 9 independent variables are needed. A reference transformation between lamina and
laminate was also necessary in order to introduce the lamina orientation. The
resulting equation is expressed in terms of a quadratic eigenproblem in the
wavenumber variable at a fixed frequency and heading angle. This problem
was transformed into a linear one and easily solved with standard numerical
routines. Dispersion curves and polar patterns are obtained for the case of an
uniform isotropic and for a laminate composite plates. Comparison with analytical formulas available from the classical plate laminate theory showed good
agreement with the FE results. Nevertheless, the analytical theory fails with
the increasing of the frequency. In fact, CLPT cant predict the appearance of
new waveforms, whereas SFEM can identify new structural waveguides. Thus
SFEM can overcome the limits of the CLPT, allowing a more satisfactory analysis of the dispersion properties at higher frequencies. The procedure can be
easily completed by including the evaluation of the resulting loss factor, given
the individual loss factor of each lamina and the group velocity curves of the
propagating waves, as well as their modal densities. All the present evaluations
and results have been obtained by using Matlab: the m-files can be requested
to the first author.
27
References
A. Baz. Spectral finite-element modeling of the longitudinal wave propagation
in rods treated with active constrained layer damping. Smart Mater. Struct,
9:372377, 2000.
F. Birgersson, S. Finnveden, and G. Robert. Modelling turbulence-induced
vibration of pipes with a spectral finite element method. Journal of Sound
and Vibration, 278(4-5):749772, 2004.
A. Chakraborty and S. Gopalakrishnan. A Spectral Finite Element Model for
Wave Propagation Analysis in Laminated Composite Plate. Journal of Vibration and Acoustics, 128:477, 2006.
M. Chitnis, Y. Desai, and T. Kant. Wave propagation in laminated composite
plates using higher order theory. Journal of applied mechanics, 68(3):503505,
2001.
R. Cook, D. Malkus, M. Plesha, and R. Witt. Concepts and Applications of
Finite Element Analysis. John Wiley & Sons, 1989.
L. Cremer, B. Petersson, and M. Heckl. Structure-Borne Sound: structural
vibrations and sound radiation at audio frequencies. Springer, 2005.
S. Datta, A. Shah, R. Bratton, and T. Chakraborty. Wave propagation in
laminated composite plates. The Journal of the Acoustical Society of America,
83:2020, 1988.
S. Finnveden. Evaluation of modal density and group velocity by a finite element
method. Journal of Sound and Vibration, 273(1):5175, 2004.
S. Ghinet, N. Atalla, and H. Osman. The transmission loss of curved laminates
and sandwich composite panels. The Journal of the Acoustical Society of
America, 118:774, 2005.
28
P. Kudela, A. Zak,
M. Krawczuk, and W. Ostachowicz. Modelling of wave propagation in composite plates using the time domain spectral element method.
Journal of Sound and Vibration, 302(4-5):728745, 2007.
F. Y. L. Wang. Group velocity and characteristic wave curves of Lamb waves in
composites: Modeling and experiments. Composites Science and Technology,
67:13701384, 2007.
A. Leissa. Vibration of Plates. Published for the Acoustical Society of America
through the American Institute of Physics, 1993.
R. Lyon. Statistical energy analysis of dynamical systems: theory and applications. MIT Press, 1975.
B. Mace, D. Duhamel, M. Brennan, and L. Hinke. Finite element prediction of
wave motion in structural waveguides. The Journal of the Acoustical Society
of America, 117:2835, 2005.
D. Mahapatra, A. Singhal, and S. Gopalakrishnan. A higher-order finite waveguide model for spectral analysis of composite structures. Computer Methods
in Applied Mechanics and Engineering, 195(9&12):11161135, 2006.
A. Nayfeh. The general problem of elastic wave propagation in multilayered
anisotropic media. The Journal of the Acoustical Society of America, 89:
1521, 1991.
29
A. Zak,
M. Krawczuk, and W. Ostachowicz. Propagation of in-plane elastic
waves in a composite panel. Finite Elements in Analysis and Design, 43(2):
145154, 2006.
30
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
Sommario
Indicatori del rumore interno
Generazione del rumore
Sorgenti di rumore:
Il propulsore, lo strato limite turbolento, le
sorgenti interne.
Noise Index
Esistono numerosi indici di misura
del rumore, e tra essi i pi
utilizzati per fornire un immagine
del
comfort
acustico
di
un
velivolosono:
OASPL
dBA
SIL3, SIL4, PSIL
Noise Index
LOASPL e loverall sound pressure
level calcolato sullo spettro (010KHz)
ed
normalmente
utilizzato per quelle applicazioni in
cui dominano le alte frequenze
(velivoli a getto).
Noise Index
Il dBA e loverall calcolato sullo
spettro (0-12.5 KHz) cui
applicato il filtro A (effetto orecchio
umano)
ed
normalmente
utilizzato per quelle applicazioni in
cui dominano le basse frequenze
(velivoli a elica).
Aircraft Interior Noise
A.S. 1/6/04
SPL [dB]
60
Pesato A
50
Lineare
40
30
20
10
0
10
100
1000
10000
Frequency
[Hz]
Aircraft
Interior
Noise
A.S. 1/6/04
Noise Index
Il SIL (Speech Interference Level)
una misura del disturbo acustico
nelle frequenze del parlato:
SIL3 media aritmetica del SPL nelle
bande di Ottave 500,1K, 2K Hz
SIL4 media aritmetica del SPL nelle
bande di Ottave 500, 1K, 2K, 4K Hz
turbofan
turboprop
Aircraft Interior Noise
A.S. 1/6/04
Il motore turbofan
Il motore turboprop
BPF
RPM
xNum.Pale
60
10 dB
SPL [dB]
290
240
190
140
90
broadband noise
Frequency [Hz]
1
fn =
2p
K
M
12000.00
11500.00
11500.00
11000.00
11000.00
10500.00
10500.00
10000.00
10000.00
9500.00
9500.00
9000.00
9000.00
8500.00
8500.00
-2000.00
-1000.00
0.00
1000.00
2000.00
96.00
94.00
92.00
90.00
88.00
86.00
84.00
82.00
80.00
78.00
76.00
74.00
72.00
70.00
68.00
66.00
64.00
62.00
60.00
-2000.00
-1000.00
0.00
1000.00
2000.00
A.S. 1/6/04
Airconditioning
ON
70
65
60
55
Airconditioning
OFF
50
A.S. 1/6/04
12500
8000
10000
6300
5000
4000
3150
2500
2000
1600
1250
800
1000
630
500
400
315
250
200
160
125
100
80
63
50
40
32
25
45
20
SPLi [dB]
75
Metodi Deterministici
I metodi deterministici FEM/BEM
sono indicati per lo studio della
trasmissione del rumore a basse
frequenze.
Tale approccio tanto pi accurato
quanto pi fitta la mesh del
modello numerico.
Aircraft Interior Noise
A.S. 1/6/04
1.01
0.972
0.933
250
0.895
0.857
0.818
0.78
0.742
200
z
0.703
t
0.665
0.626
0.588
0.55
150
0.511
Hz
0.473
Output X
Set: Mode
Y 4 16.34406 Hz
Deformed(1.01): Total Translation
Contour: Total Translation
0.434
0.396
100
50
0
A2
S2
A3
S3
A4
S4
A5
S5
A6
S6
modo
450.00
400.00
350.00
EMA
300.00
R2 = 0.9844
250.00
200.00
150.00
100.00
50.00
0.00
0.00
Correlazione
Ideale
Lineare (Ideale)
Lineare (Correlazione)
[0] u&& [K s ] [] u F
+
= ;
M A &p& [0] [K A ] p 0
V1
4.941
C1
4.633
4.324
130
4.015
3.706
3.088
110
2.78
2.471
100
2.162
1.853
90
1.544
1.235
80
0.9265
Y
X
0.6177
0.3088
0.
SPL [dB]
Valore sperimentale
Val. numerico 1 Watt
Val. Numerico 10 Watt
120
3.397
70
60
50
40
30
20
10
0
-10
0
100
200
300
400
500
600
700
800
900
Frequenza [Hz]
1000
Z
X
Eccitazione acustica
sulla fusoliera
Metodi energetici
I metodi energetici si basano sul
bilancio di energia interna ed
esterna e dellenergia che si
scambiano i diversi volumi che
rappresentano la nostra cavit.
Metodi energetici
Metodi energetici
Eass
2
1 pavg
= t a =
Sa
4 ro c
Wtrai
2
TLwalli
1 pavg
=
Si 10 10
4 roc
WBLi
2
TL
BLi
1 pavgBL
=
Sexti10 10
4 r oc
WENi
2
TL
ENi
1 pavgEN
=
Sexti10 10
4 roc
Wp = I P Si = I p S
i
Metodi energetici
Wref
10
SPLint = 10logSexti10 10 +10 10 + 410
+ Sinti 10 10
i
TLwalli
10logSi ai +10 10
i
PILOT
PILOT
118
CO-PILOT
116
114
110
108
106
104
102
65
63
61
59
57
55
53
51
49
47
45
43
41
39
37
35
33
31
29
27
25
23
21
19
17
15
13
11
100
1
oaspl
112
VOLUMI
Sommario
Indicatori del rumore interno
Generazione del rumore
Sorgenti di rumore:
Il propulsore, lo strato limite turbolento, le
sorgenti interne.
Noise Index
Esistono numerosi indici di misura
del rumore, e tra essi i pi
utilizzati per fornire un immagine
del
comfort
acustico
di
un
velivolosono:
OASPL
dBA
SIL3, SIL4, PSIL
Noise Index
LOASPL e loverall sound pressure
level calcolato sullo spettro (010KHz)
ed
normalmente
utilizzato per quelle applicazioni in
cui dominano le alte frequenze
(velivoli a getto).
Noise Index
Il dBA e loverall calcolato sullo
spettro (0-12.5 KHz) cui
applicato il filtro A (effetto orecchio
umano)
ed
normalmente
utilizzato per quelle applicazioni in
cui dominano le basse frequenze
(velivoli a elica).
Aircraft Interior Noise
A.S. 1/6/04
SPL [dB]
60
Pesato A
50
Lineare
40
30
20
10
0
10
100
1000
10000
Frequency
[Hz]
Aircraft
Interior
Noise
A.S. 1/6/04
Noise Index
Il SIL (Speech Interference Level)
una misura del disturbo acustico
nelle frequenze del parlato:
SIL3 media aritmetica del SPL nelle
bande di Ottave 500,1K, 2K Hz
SIL4 media aritmetica del SPL nelle
bande di Ottave 500, 1K, 2K, 4K Hz
turbofan
turboprop
Aircraft Interior Noise
A.S. 1/6/04
Il motore turbofan
Il motore turboprop
BPF
RPM
xNum.Pale
60
10 dB
SPL [dB]
290
240
190
140
90
broadband noise
Frequency [Hz]
1
fn =
2p
K
M
12000.00
11500.00
11500.00
11000.00
11000.00
10500.00
10500.00
10000.00
10000.00
9500.00
9500.00
9000.00
9000.00
8500.00
8500.00
-2000.00
-1000.00
0.00
1000.00
2000.00
96.00
94.00
92.00
90.00
88.00
86.00
84.00
82.00
80.00
78.00
76.00
74.00
72.00
70.00
68.00
66.00
64.00
62.00
60.00
-2000.00
-1000.00
0.00
1000.00
2000.00
A.S. 1/6/04
Airconditioning
ON
70
65
60
55
Airconditioning
OFF
50
A.S. 1/6/04
12500
8000
10000
6300
5000
4000
3150
2500
2000
1600
1250
800
1000
630
500
400
315
250
200
160
125
100
80
63
50
40
32
25
45
20
SPLi [dB]
75
Metodi Deterministici
I metodi deterministici FEM/BEM
sono indicati per lo studio della
trasmissione del rumore a basse
frequenze.
Tale approccio tanto pi accurato
quanto pi fitta la mesh del
modello numerico.
Aircraft Interior Noise
A.S. 1/6/04
1.01
0.972
0.933
250
0.895
0.857
0.818
0.78
0.742
200
z
0.703
t
0.665
0.626
0.588
0.55
150
0.511
Hz
0.473
Output X
Set: Mode
Y 4 16.34406 Hz
Deformed(1.01): Total Translation
Contour: Total Translation
0.434
0.396
100
50
0
A2
S2
A3
S3
A4
S4
A5
S5
A6
S6
modo
450.00
400.00
350.00
EMA
300.00
R2 = 0.9844
250.00
200.00
150.00
100.00
50.00
0.00
0.00
Correlazione
Ideale
Lineare (Ideale)
Lineare (Correlazione)
[0] u&& [K s ] [] u F
+
= ;
M A &p& [0] [K A ] p 0
V1
4.941
C1
4.633
4.324
130
4.015
3.706
3.088
110
2.78
2.471
100
2.162
1.853
90
1.544
1.235
80
0.9265
Y
X
0.6177
0.3088
0.
SPL [dB]
Valore sperimentale
Val. numerico 1 Watt
Val. Numerico 10 Watt
120
3.397
70
60
50
40
30
20
10
0
-10
0
100
200
300
400
500
600
700
800
900
Frequenza [Hz]
1000
Z
X
Eccitazione acustica
sulla fusoliera
Metodi energetici
I metodi energetici si basano sul
bilancio di energia interna ed
esterna e dellenergia che si
scambiano i diversi volumi che
rappresentano la nostra cavit.
Metodi energetici
Metodi energetici
Eass
2
1 pavg
= t a =
Sa
4 ro c
Wtrai
2
TLwalli
1 pavg
=
Si 10 10
4 roc
WBLi
2
TL
BLi
1 pavgBL
=
Sexti10 10
4 r oc
WENi
2
TL
ENi
1 pavgEN
=
Sexti10 10
4 roc
Wp = I P Si = I p S
i
Metodi energetici
Wref
10
SPLint = 10logSexti10 10 +10 10 + 410
+ Sinti 10 10
i
TLwalli
10logSi ai +10 10
i
PILOT
PILOT
118
CO-PILOT
116
114
110
108
106
104
102
65
63
61
59
57
55
53
51
49
47
45
43
41
39
37
35
33
31
29
27
25
23
21
19
17
15
13
11
100
1
oaspl
112
VOLUMI
P2 _reference
P8_stringerTT
39.00
36.00
33.00
30.00
27.00
24.00
21.00
18.00
15.00
12.00
100
1000
10000
Transmission Loss
Dal valore ideale alla realt
12
2 dBA
90
88
10
86
8
82
80
78
76
74
72
70
68
0
1
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Window location
Flight Direction
dBA reduction
84
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
5
La potenza acustica radiata da un pannello
piano
(5.2)
74
i (!) =
(5.3)
dove 0 e la densit
a del mezzo in cui si propaga il disturbo.
j! 2 0
2c0
e jk|P Q|
v(!, Q) dS
Q|
S k|P
(5.5)
rad (!) =
R(!, P, Q) =
e jk(!)|P Q|
.
k(!)|P Q|
(5.7)
75
lim
Q|!0
sin(k|P Q|)
= 1.
k|P Q|
(5.8)
cv 2 S
(5.11)
76
definizione di , pertanto, sono presenti sia la geometria della superficie considerata che le caratteristiche del mezzo in cui si propaga il suono. In alcuni
testi, lefficienza di radiazione di un componente strutturale e definita come
il rapporto tra la potenza acustica radiata dallo stesso e la potenza acustica
radiata teoricamente da un pistone infinitamente rigido, avente stessa area e
stessa distribuzione di velocita superficiale, ovvero:
=
rad
,
piston
(5.12)
dove piston = cv 2 S. Tale definizione trae origine dal fatto che un pistone
rigido infinito e un radiatore perfetto. Se infatti consideriamo il movimento
di un pistone rigido infinito, questo forza le particelle di fluido a muoversi su
linee parallele normali al piano del pistone: non ce divergenza, cioe non ce
alcuna forza di reazione inerziale in quanto non ci sono bordi, e quindi la forza
di reazione per unit
a di area (ovvero la pressione) e dovuta esclusivamente ad
eetti di compressione.
Il rapporto
(5.13)
R= 2
v
e detto resistenza di radiazione.
I parametri introdotti sinora sono globali, ovvero riferiti allintera superficie vibrante, e sono indicatori dellaccoppiamento del campo acustico con
lintero pannello.
Risulta conveniente calcolare lefficienza di radiazione per ogni singolo modo naturale del pannello, in modo tale da identificare quale modo e causa di
una maggiore radiazione di potenza acustica: in questo caso, si parla di efficienza di radiazione modale (Ref. [10]). La resistenza di radiazione del modo
(m, n) del pannello e espressa come
Rm,n =
m,n
|v(!)|2
(5.14)
Rm,n
=
cab
cab |vm |2
(5.15)
77
(5.18)
1]
)2
sin d d . (5.19)
se m e un intero dispari, sin 2
se invece m e pari; analogamente, si utilizza cos 2 se n e un intero dispari,
sin 2 se invece n e pari. Nelle Figura 5.2 e mostrato landamento dellefficienza di radiazione di diversi modi propri di vibrare di un pannello le cui
caratteristiche sono descritte in Tabella 5.1.
Nella (5.19), e necessario utilizzare cos
1m
0.5 m
0.005 m
densit
a
modulo di Young E
modulo di Poisson
2700 Kg/m3
71 GP a
0.33
78
ij (f )
79
(5.20)
(5.21)
T L indica, in deciBel, labbattimento che il suono subisce passando attraverso una parete, e quindi permette di quantificare la potenza persa nella
trasmissione del suono tra due ambienti separati da un divisorio.
Figura 5.4: diminuzione della potenza sonora attraverso una parete avente un
potere fonoisolante di 45dB.
80
81
uguale alla frequenza di uno dei modi propri: in questo caso, infatti, si verifica
che la deformata modale ha una lunghezza donda uguale a quella dellonda
incidente, per cui il modo proprio sollecitato e trasparente al suono, e determina una riduzione complessiva del potere fonoisolante del pannello. Al primo
modo normale e associata la pi
u importante frequenza di risonanza, in corrispondenza della quale il T L raggiunge un minimo. Si noti come lampiezza
delle valli di isolamento dipenda dal valore del fattore di smorzamento : pi
u
e elevato lo smorzamento, maggiore e lenergia meccanica vibratoria dissipata
in calore. In pratica, se lo smorzamento e elevato, il pannello vibra meno, e
quindi e meno trasparente al suono.
Per valori della frequenza inferiori alla frequenza di risonanza naturale,
la trasmissione sonora dipende essenzialmente dalla rigidezza (o, equivalentemente, dallelasticit
a) della struttura, e si ha una diminuzione di T L di 6 dB
ogni raddoppio della frequenza. Si parla di zona della rigidezza.
Per valori della frequenza compresi tra la zona di risonanza e una certa
frequenza critica, il potere fonoisolante e governato dalla cosiddetta legge della
massa: in questa zona, si ha un aumento del T L di 6 dB per raddoppio della
frequenza e della massa, e il comportamento della struttura dipende esclusivamente dalla sua massa. In questa zona, e possibile utilizzare la seguente
formula di previsione:
T LM L, = 10 log10
h f cos() 2 i
s
[dB]
0 c 0
(5.23)
42.5[dB].
(5.24)
T LM L,dif f = 10 log10
ln
"
0.978 s0f
c0
2
s f
0 c0
0.208 s f
0 c0
1+
1+
#.
(5.25)
La validit
a della legge della massa e limitata superiormente dal fenomeno
della coincidenza, conseguenza del fatto che la velocita del suono nellaria
82
e costante al variare della frequenza mentre la velocita delle onde flessionali nelle strutture varia con la frequenza. Le onde flessionali sono quelle che
hanno maggiore importanza nella radiazione acustica delle strutture: cio e
dovuto principalmente al fatto che la deflessione laterale degli elementi su cui
si propagano le onde (cioe normale al piano delle onde) e rilevante rispetto
alla lunghezza donda, e pertanto e in grado di perturbare il fluido adiacente.
Nel caso di un pannello sottile, cioe di spessore molto inferiore alla lunghezze
donda del suono alle frequenze in analisi, su cui si abbia propagazione in entrambe le direzioni di giacitura del pannello, la velocita di propagazione delle
onde flessionali e:
v
s
u
u
E
cb = t2f s
(5.26)
12(1 2 )
Figura 5.6: variazione delle velocita di propagazione del suono in aria c e delle
onde flessionali cb al variare della frequenza.
Le onde sonore piane che incidono con un certo angolo su una parete sottile
infinita originano nella parete unonda flessionale forzata avente lunghezza
donda tr , pari alla lunghezza donda di traccia dellonda sonora incidente
i
sin() , ovvero:
tr
sin()
(5.27)
dove i e la lunghezza donda dellonda incidente. Il fenomeno della coincidenza si verifica quando, per un determinato angolo di incidenza, la lunghezza
donda di traccia dellonda sonora eguaglia la lunghezza donda b dellonda
flessionale libera, ossia:
b
sin()
tr .
(5.28)
83
12000
[Hz]
s
(5.30)
84
delle propriet
a acustiche del pannello, si riprendono brevemente i concetti alla
base dellanalisi della risposta strutturale.
5.4.1 Riepilogo della risposta strutturale in vacuo
Si consideri un generico operatore strutturale piano rappresentato in N G
coordinate discrete, ognuna delle quali rappresenta lo spostamento w(x, y)
fuori dal piano in quella posizione del pannello. Si trascurano gli spostamenti
u(x, y) e v(x, y) nel piano del pannello.
(5.31)
j!t
(5.32)
{f (t)} = {F (!)}e
j!t
(5.33)
il sistema pu
o essere riscritto come segue:
h
[M ]{
q (t)} + [K]{q(t)} = {f (t)} ) [K]
i
! 2 [M ] {w(!)} = {F (!)} (5.34)
85
Se presente, lo smorzamento puo essere introdotto direttamente negli operatori modali, o semplicemente considerando la presenza di una matrice viscosa
j![B]{w(!)}
oppure introducendo una matrice di rigidezza complessa
1 + j(!) [K]{w(!)}.
Una volta caratterizzato il sistema descritto dalla (5.31) in modi e frequenze proprie, e possibile passare in coordinate modali {(!)} mediante la
matrice modale [ ], ovvero
{w(!)} = [ ]{(!)},
(5.35)
(5.36)
<i> T
{F (!)}.
(5.37)
(5.38)
mi
!2
1
.
+ ki + ji (!)ki
(5.40)
(5.41)
86
A1
60
6
A=6 .
4 ..
0
A2
..
.
...
...
..
.
0
0
..
.
0 0 . . . AN G
7
7
7
5
(5.42)
(Nx
a b
1)(Ny
1)
(5.43)
! 4 0
{w(!)}H [A][R(!)][A]{w(!)},
4c0
(5.44)
in cui ri,j e la distanza tra i due punti i,j. Usando le coordinate modali, e
possibile esprimere il vettore degli spostamenti normali {w(!)} come:
{w(!)} = [ ][H(!)][ ]T {F (!)}.
(5.46)
! 4 0
[ ][H(!)][ ]T {F (!)}
4c0
(5.47)
[(!)] = [A][R(!)][A]
87
(5.48)
[Rrad (!)] =
! 4 0 T
[ ] [(!)][ ],
4c0
(5.50)
(5.51)
1
Re {p(!)H [A]{v(!)} ,
2
(5.52)
per cui, una volta nota la pressione agente e la matrice delle aree equivalenti
nodali, ed essendo {v(!)} = j!{w(!)}, e possibile scrivere
inc (!) =
1
Re j!{F (!)H {w(!)} .
2
(5.53)
! 4 0
H
4c0 {w(!)} [(!)]{w(!)}
2
! 0 c0
{w(!)}H [A]{w(!)}
2
! 2 {w(!)}H [(!)]{w(!)}
. (5.54)
2c20 {w(!)}H [A]{w(!)}
T
T
k(!)2 {F (!) [ ][H(!)] [ ] [(!)] [ ][H(!)][ ] {F (!)
h
i h
i . (5.55)
(!) =
2
{F (!)H [ ][H(!)] [ ]T [A] [ ][H(!)][ ]T {F (!)
88
(a) = 0.04
(b) = 0.08
(c) = 0.1
(d) = 0.2
89
(!)i,j =
(5.56)
Nella Figura 5.10 e riportato il confronto tra le curve dellandamento dellefficienza di radiazione calcolate secondo il modello analitico di Wallace (vedi
Figura 5.2) e i punti ottenuti mediante lapproccio discreto. Si considera il
pannello di Tabella 5.1 senza alcun tipo di smorzamento. La mesh utilizzata
e quella illustrata in Figura 5.8. In questo caso, in analogia con Wallace, si
considera un campo sonoro perfettamente diuso.
Figura 5.10: confronto tra lefficienza di radiazione modale calcolata analiticamente () e quella mediante approccio discreto () per diversi modi al
variare della frequenza.
90
Figura 5.11: efficienze di radiazione auto e mutua per il primo modo del
pannello in funzione della frequenza.
! 4 0
4c0
91
(5.58)
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
ARTICLE IN PRESS
Abstract
Recent studies have demonstrated that the characterisation of wall-pressure fluctuations for surface ships is of great
interest not only for military applications but also for civil marine vehicles. A ship model towed in a towing tank is used
to perform pressure and structural measurements at high Reynolds numbers. This facility provides ideal flow conditions
because background turbulence and noise are almost absent. Free surface effects are naturally included in the analysis,
although in the particular section chosen for the present study do not have significant consequences on pressure spectra.
Scaling laws for the power spectral density are identified providing the possibility to estimate pressure spectra for
different flow conditions and in particular for full-scale applications. The range of validity of some theoretical models
for the cross-spectral density representation is analysed by direct comparison with experimental data of wall-pressure
fluctuations measured in streamwise and spanwise direction. In a second phase, an indirect validation is performed by
comparing the measured vibrational response of an elastic plate inserted in the catamaran hull with that obtained
numerically using, as a forcing function, the modelled pressure load. In general, marine structures are able to accept
energy mainly from the sub-convective components of the pressure field because the typical bending wavenumber values
are usually lower than the convective one; thus, a model that gives an accurate description of the phenomenon at low
wavenumbers is needed. In this work, it is shown that the use of the Chase model for the description of the pressure field
provides a satisfactory agreement between the numerical and the experimental response of the hull plate. These
experimental data, although acquired at model scale, represent a significant test case also for the real ship problem.
r 2008 Elsevier Ltd. All rights reserved.
Keywords: Wall-pressure fluctuations; High Reynolds number flow; High-speed vessels; Theoretical models; Vibrational response
1. Introduction
Vibrations of elastic structures excited by the turbulent boundary layer (TBL) are of interest for interior and exterior
noise emission problems in aeronautical, automotive and marine applications. In particular, new requirements in terms
of comfort on board high-speed ships for passenger transportation have addressed the attention of the scientific
community to the identification and to the characterisation of noise sources including those of hydrodynamic nature.
Recent studies performed in the framework of the European RTD project NORMA (Noise Reduction for Marine
!Corresponding author: Tel.: +39 081 7683581; fax: +39 081 624609.
ARTICLE IN PRESS
322
Nomenclature
a
Areai
b
c
cB
D
d
d0
d+
Fr
g
H
H
h
i
j
k
kc
kB
Lpp
m, n
me
mn
N
NG
NM
ReW
Ret
Rpp
Saa
SW
SF
SFF
u+
ut
U
Uc
x
y
y+
convection velocity
streamwise reference axis
spanwise reference axis
wall unit, y+ yut/n
Greek symbols
g1
g3
G
d
d*
Dx
Dy
Z
Zp
W
y
n
x
rs
r
t
tw
Fpp
Fpp0
U
o
oj
omn
mn
o
Applications G3RD-2001-0393) demonstrated that, at least for new concept design fast ships, flow noise sources, e.g.,
the TBL, play an important role above 30 knots.
The typical way to characterise wall-pressure fluctuations (WPF) is via experimental tests performed in suitable
facilities like wind or water tunnels. In fact, direct numerical simulations (DNS) or large eddy simulations (LES) are
often not applicable in the case of complex geometries and realistic flow conditions (high Reynolds numbers) due to the
limitation of computational resources. DNS of WPF were performed by Choi and Moin (1990), analysing the channel
flow problem for ReW UW/n 287. Furthermore, Chang et al. (1999) analysed the influence of the different TBL
velocity components on the wavenumber pressure spectra in a channel flow for a Reynolds number, based on the
channel half width, equal to 3200. Recently, Lee et al. (2005a) proposed a new methodology to calculate numerically
wall-pressure spectra. The method uses the predicted mean flow field obtained from RANS calculations and a spectral
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
323
correlation model, and integrates across the TBL. The method was validated both for an equilibrium flow at
ReW 3582 and for a non-equilibrium flow resulting from flow over a backward-facing step. Using the same
methodology, Lee et al. (2005b) characterised wall-pressure spectra for a surface ship model including the effects of hull
curvature and of the free surface. The comparison of the scaled spectra, obtained while varying the axial location, and
the distance from the free surface with spectra, obtained for an equilibrium flow, showed that, in several locations, the
former deviate from the canonical case.
On the other hand, there are many experimental works related to WPF, most of them devoted to the identification of
the appropriate scaling laws for the auto-spectral density (ASD) for zero pressure gradient flow. The pressure ASD
frequency range is subdivided according to the boundary layer regions that give contributions to wall-pressure spectra
where different scaling variables hold. In particular, Farabee and Casarella (1991) identified four frequency ranges in
their data: the low-frequency and the mid-frequency range where outer variables hold, the high-frequency range where
inner variables hold, and an overlap scale-independent region proportional to o"1, whose extent depends on the
Reynolds number. With respect to this point, Keith et al. (1992) presented the most extensive comparison among many
available experimental data obtained in fully developed and developing channel flow, in fully developed pipe flow and in
wind tunnel, over a wide range of Reynolds numbers, with the aim of identifying the best choice for the scaling
parameters in the different frequency regions. Goody (1999) performed an experimental campaign in a two-dimensional
boundary layer for ReW values ranging from 7800 to 23 400, investigating different combinations of scaling parameters.
Finally, a detailed review of the state-of-the-art on this subject can be found in Bull (1996).
The spatial characterisation of WPFs was first analysed by Corcos (1963) on the basis of measurements performed by
Willmarth and Wooldridge (1962). Assuming the validity of separation of variables in the streamwise and spanwise
directions, Corcos stated an exponential decay for the cross-spectral density (CSD) as a function of the similarity
variables ox/Uc and oZ/Uc, where Uc is the convection velocity, and x and Z are the streamwise and spanwise spatial
separation, respectively. Several authors have performed comparisons between measured CSD data and Corcos model
(Blake, 1986; Bull, 1967); in particular, Farabee and Casarella (1991) from the analysis of their experimental data
provided, at least in a certain nondimensional frequency range, a confirmation of this pressure behaviour for a wide
series of spatial separations in streamwise direction and for different flow velocities or local Reynolds number values.
The success of the Corcos model lies in its simplicity and in its predictive character since the model parameters are
substantially case-independent. Nevertheless, it is generally stated that Corcos model gives a correct representation of
the WPF behaviour in the convective domain, i.e. when the wavenumbers are close to the convective wavenumber
kc o/Uc. On the contrary, in the sub-convective domain the white Corcos spectrum largely overpredicts the real
amplitude. Since for several applications and in particular in the case of underwater and surface marine vehicles, the
convective wavenumber is greater than the bending wavenumber kB o/cB, it is of primary importance to evaluate
correctly the sub-convective domain of pressure spectra that corresponds to the high-sensitivity region for the structure.
Several new models, some directly derived by the Corcos one (Efimtsov, 1982; Ffowcs Williams, 1982), others
overcoming the Corcos multiplicative approach such as those by Chase (1980) and Smolyakov and Tkachenko (1991),
were developed to improve the estimation of pressure spectra in this region. A comparison between the predictions of
the radiated acoustic power by rectangular plates was carried out numerically by Graham (1997); it was performed for
different test conditions and applying the above models. It was there concluded that the use of sophisticated models
such as the Chase one is needed only for structures that do not exhibit coincidence, but that for aircraft the best model is
the one which provides an accurate description of the convective peak, thus suggesting the use of the Efimtsov model.
Nonetheless, no experimental evidence supporting these conclusions was reported in Grahams work. However, the
spatial domain comparison between pressure experimental spectra and theoretical models cannot definitively indicate
the best in describing the different wavenumber regions. It is usually possible to find a set of parameters for each model
able to provide a good data fit. It is clear that most of the energy of WPF is concentrated around the convective peak
and then any correlation data is mainly the representation of the convective character of the TBL. Unfortunately, only
few experimental data concerning direct measurements of the wavenumber-frequency spectrum are available
(Abraham, 1998; Choi and Moin, 1990; Panton and Robert, 1994; Farabee and Geib, 1991; Manoha, 1996) and,
among them, a big spread of the spectra magnitude at low wavenumbers is present as reported for example by Hwang
and Maidanik (1990).
In order to overcome the limitations of flow measurements, an indirect approach to estimate the validity of different
models for WPF representation, based on the analysis of the response of simple elastic structures to the TBL load, is
proposed here. The same idea was recently applied by Finnveden et al. (2005), who compared the measured response of
a flat plate with those obtained numerically using modelled pressure loads. This work presented the first and, to the
authors best knowledge, the only correlation between aerodynamic and structural data measured in the same facility
and with the same set-up. They suggested a modified version of the Corcos model by introducing a frequency and flow
speed dependence on the parameters and of the Chase model by introducing two new parameters to better fit the
ARTICLE IN PRESS
324
spanwise coherence to measurements. Despite the modifications made, the conclusion was that, above the aerodynamic
coincidence (kc kB), only the Chase model, that does not make use of the multiplicative approach, provides a fair
agreement with experimental data. In this work, the lower kB/kc ratio was 0.4 and the average difference between the
modified Chase model predictions and the experimental data was 5 dB. Furthermore, Hambric et al. (2004), although
retaining the multiplicative approach, proposed a modification of Corcos streamwise coherence to better represent the
low wavenumber domain. The model was compared with the experimental response of an elastic plate measured by Han
et al. (1999). The ratio between the structural wavenumber in flow direction and the convective wavenumber was
between 0.3 and 0.8 and the agreement with the experimental data was quite good. On the other hand, Han et al. (1999)
chose Smolyakov and Tkachenko to model the surface pressure field. The comparison with measurements was
performed using the energy flow analysis method to predict the numerical plate response thus, direct information about
the validity of the pressure model are difficult to extract from their data.
The aim of this work is to develop a general procedure based on the identification of the scaling laws and on the use of
predictive models for the surface pressure field suitable for application to full-scale problems. In particular, the capabilities of
Corcos and Chase models to predict the response of an elastic plate inserted in the hull of a ship model were investigated on
the basis of hydrodynamic and vibration data acquired, at high Reynolds numbers, in a towing tank. In a first step, pressure
data were analysed to provide their spectral characteristics. This analysis is fundamental to identify the scaling laws for the
ASD and the free parameters contained in the CSD wall-pressure fluctuation models. The high Reynolds number achieved
with this set-up provides an interesting extension to the previous analyses. In a second phase, a comparison between the
numerical response of the plate obtained using the two models and the experimental response is provided. Since in the
present problem the convection velocity is very low, the ratio between the bending and the convective wavenumber is
sensibly lower than those previously analysed in the technical literature. This fact is fundamental for real size marine
applications for which hydrodynamic coincidence appears, even for high-speed vehicles, at very low frequency.
This first section is aimed to frame the work in the proper existing literature. Section 2 presents the experimental setup and all the data concerning the acquisition instrumentation. The treatment of the pressure data is the specific
argument of Section 3. Section 4 is fully devoted to the analysis of the structural response and the final comparison
between predictive and measured data. Section 5 presents the concluding remarks with some foreseen activities. For the
sake of completeness, a graphic workflow has been also added in Chart 1.
2. Experimental set-up
2.1. Pressure measurements
The experiments were performed on a 1:15 scale model ofpthe
ffiffiffiffiffiffiffifast
ffiffi catamaran Jumbo CAT (Fig. 1). The scale of the
model was chosen according to Froude similarity: Fr U= gLpp where Lpp is the length between perpendiculars, i.e.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
325
the length of the vessel along the waterline between the forward and aft perpendiculars, as depicted in Fig. 1. The
maximum model width is 1.467 m, Lpp is 4.635 m and its draft in calm water conditions is 0.2 m.
The experiments were carried out in the INSEAN towing tank no. 2 which is 220 m long, 9 m wide and 3.5 m deep
and is equipped with a carriage that can reach a maximum speed of 8 m/s. The use of this kind of facility creates ideal
flow conditions because background turbulence and noise are avoided. The measuring section was chosen in the stern
part of the ship bottom where the hull surface is almost flat. To perform pressure measurements a 2 cm thick rigid
plexiglas plate was inserted in the hull bottom where pressure transducers were positioned.
The basic set-up is presented in Fig. 2 and consisted in an array of nine transducers in streamwise direction and five
transducers in the spanwise direction flush-mounted with the plate at constant distance of 1 cm between each other.
Additional tests were performed with 13 transducers mounted in streamwise direction within a maximum distance of
40 cm. Thus, the first pressure sensor was located at x/Lpp 0.88 while the last at x/Lpp 0.97. The minimum distance
between transducers was constrained by the transducers maximum external size while the maximum distance was
chosen according to the fact that for x/d*420 the longitudinal correlation is almost zero as demonstrated by previous
measurements (Bull, 1967; Blake, 1986). Pressure signals were acquired in calm water conditions with fixed trim and
sink and for two different ship model velocities: 3.31 m/s (25 knots) and 5.3 m/s (40 knots) corresponding to Fr 0.49
and 0.78, respectively. The measurement error in the carriage velocity was within 1% of the nominal mean velocity.
Differential piezoresistive pressure transducers Endevco 8510-B, characterised by a maximum range of 2 psig and by
a certified flat response until 14 kHz were used to measure pressure fluctuations. The transducers were statically
calibrated in water using known water level heights. All the transducers showed a linear trend; however, the deviation
around the regression line of the data points used for the sensitivity estimate was evaluated. The standard error of
estimate was very low for all the transducers, of the order of 1%. Moreover, the total error due to thermal sensitivity,
nonlinearity and pressure hysteresis, as reported in the data sheet, is around 1%. The rectangular sensing element has
an area of 1 " 0.3 mm2, hence the effect of the finite size of the transducers surface can be expressed in term of the
nondimensional parameters d0 d/d* and d+ dut/n, where d is the bigger sensor dimension.
Pressure signals were acquired and amplified by the 16 channels acquisition system PROSIG; the sampling frequency
was 12.5 kHz, the acquisition length was 15 s. Several repetitions of the test (typically 1215) under nominally the same
conditions were performed. The data record began a few seconds after the achievement of steady conditions. The
Fig. 2. Set-up for pressure measurements (left) and top view of the installation of the plexiglas plate (right).
ARTICLE IN PRESS
326
Table 1
Mean flow velocity parameters: numerical estimation
U (knots)
U (m/s)
d (m)
d* (m)
ut (m/s)
ReW UW/n
Ret dut/n
d0 d/d*
d+ dut/n
25
40
3.31
5.31
0.12
0.113
0.0142
0.0137
1.27
1.3
0.11
0.1626
29 535
42 807
10 153
14 133
0.07
0.073
84
125
reaching of a stationary random process was verified by comparing the ensemble average value, autocorrelation and
cross-correlation of the WPFs of different runs.
Concerning the flow velocity field, the TBL mean parameters (as used in the data analysis and shown in Table 1) were
obtained by available RANS simulations performed in the past over the whole model. This solution was, in this case,
preferred because the experimental evaluation of the boundary layer velocity profiles in a towing tank, although
possible, is a time-consuming process. In fact, it is clear that the acquisition time is limited for each carriage run,
especially for the higher velocities, and that the time needed for the re-establishment of calm water conditions between
two consecutive runs is at least 10 min. A detailed description of the numerical code is provided in Ciappi and
Magionesi (2005) and in the references cited there. The numerical errors can be predictable in an uncertainty of about
4% in the estimation of the TBL parameters from the velocity profiles.
2.2. Vibration measurements
Vibration measurements were performed replacing the rigid plate with a flexible one. The panel, made of plexiglas, is
0.58 m long, 0.2 m wide and 0.003 m thick, it was fixed to the hull model with some mastic in order to provide
impermeable conditions and to reduce the transmission of model vibrations.
A preliminary series of numerical analyses have been performed to exclude the presence of significant plate
deformations due to static and dynamic pressure loads. In fact, for all the flow speeds under consideration, the
maximum displacement was predicted to be 1% of the longitudinal plate dimension.
The acceleration responses were acquired in eight different points (one for each carriage run) randomly chosen on the
plate surface. A Bruel & Kjaer piezoelectric accelerometer type 4393 characterised by a sensitivity of 4.19 mV/g and a
weight of 2.2 g was used for the acquisition. Its mass was negligible with respect to the plate mass in the frequency range
of interest. The accelerometer signal was amplified by a Bruel & Kjaer amplifier type 2635 and acquired with a sampling
frequency of 12.5 kHz by a National Instruments PXI 6052E acquisition system. Preliminary dry and wet calm water
tests were performed with the same set-up and instrumentation to evaluate the plates natural frequencies, hence the
added fluid mass and the modal damping factors. Two additional accelerometers were mounted on the ships hull and
on the connecting system to acquire the spurious vibrations transmitted by the carriage structure.
3. Pressure analysis
In the following sections, the results of the experimental programme devoted to the characterisation of wall-pressure
spectra are presented. The purpose of this analysis was to verify the pressure scaling laws and to provide a general
model for its spatial behaviour. To this aim, ASDs, streamwise and spanwise coherences and convection velocities were
extracted from measurements. Although free surface effects were naturally present, pressure gradient values calculated
on the basis of numerical simulations can be considered negligible in the measuring section. In Fig. 3, the velocity
profiles obtained numerically, used to extract the mean TBL parameter values of Table 1, are shown in wall units y+,
u+. From preliminary analysis it was decided to consider only the Corcos and Chase models as antagonists in this
analysis. In fact, the Efimtsov model has the same trend as Corcos in the low wavenumber domain but, this last is to be
preferred because describes the wall pressure by a simpler expression containing less empirical parameters. The Ffowcs
Williams model was built to extend Corcos model to the acoustic domain, which is beyond the purpose of this analysis;
finally, the Smolyakov and Tkachenko model does not fit well the present hydrodynamic data.
3.1. Power spectral density: scaling laws
The analysis of the scaling laws for the ASD is essential to understand the contribution of the different boundary
layer regions to WPFs. Moreover, due to the particular section chosen to perform pressure measurements and the
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
327
relatively high speed of this vessel, local Reynolds numbers were sensibly high (ReW 29 535 for 3.31 m/s and
ReW 42 807 for 5.31 m/s), providing an interesting extension of the validity of the scaling laws for high Reynolds
number. Following the frequency range division proposed by Farabee and Casarella (1991), three different spectral
regions can be identified: at very low frequency, the spectra collapse using the classical outer flow variables d* and U,
showing a o2 behaviour, advising that sources are associated with the large-scale structures. In the low-mid-frequency
range, the pressure auto spectral densities collapse into a single curve when scaled with the friction velocity ut, the wall
shear stress tw and d, implying that the mid-frequency structures are related with turbulence activity in the outer region
of the boundary layer. In this interval, pressure spectra exhibit their maxima for od/ut 50. Finally, at high-frequency
inner variable scale, which employs ut, tw and n, allows the collapse of the data independently of the Reynolds number,
suggesting that sources are associated with the buffer region of the wall layer. Moreover, between mid- and highfrequency an overlap region, characterised by an o"1 decay, exists where both outer and inner variable scales hold. This
region is related to the turbulence activity in the logarithmic part of the boundary layer and its extension depends on the
Reynolds number value. Recently, Ciappi and Magionesi (2005), considering the frequency division stated by Farabee
and Casarella provided another confirmation of the proposed scaling laws. ASDs were determined using 700 spectral
averages for each signal and a Hamming window function is used to reduce bandwidth leakage. According to classical
pffiffiffiffiffi
theory of random data (Bendat and Piersol, 1991), the statistical convergence error was defined as !r 1= nd , where nd
is the number of spectral averages. In the present analysis, the data random error was equal to 73.8%; thus, the
uncertainty in the calculated pressure spectra, obtained by considering the above and all the previously defined
experimental sources of error (see Section 2), was within the range of 71 dB.
In Fig. 4, a typical ASD signal is displayed showing high peaks in the frequency region between 8 and 20 Hz due to
structural vibrations of the carriage and of the connecting system. The peaks were eliminated, for the ASD analysis,
using suitable relations based on the coherence function (Bendat and Piersol, 1991) between two pressure sensors
located sufficiently far from each other to be correlated only by structural vibrations. The result of this cleaning
procedure is shown in the same figure. Fig. 5 shows the cleaned ASD for the two different test velocities, scaled using
outer flow variables: od=ut ; Fpp ou2t =t2w d; they are shown and compared with the results of Farabee and Casarella
(1991) obtained for ReW 6050 and with the results of Blake (1970) [extracted from Lee et al. (2005a)] obtained for
ReW 8210. From the inspection of the figure, it is evident that there is an excellent agreement of the present
experimental curves in the low-mid-frequency range, i.e. for 20ood/uto1760. Moreover, for od/uto800, they are in a
very good agreement with the Farabee and Casarella curve and in fair agreement with the Blake data. Finally, the
scaled spectra achieved the maximum value for od/utE63. Low-frequency behaviour (od/uto5) is not analysed, since
frequency resolution is too poor to obtain a realistic trend in this region. Fig. 6 shows the present wall-pressure spectra
and the results of Bull and Thomas (1976), Farabee and Casarella (1991) and Blake (1970) scaled on inner flow
ARTICLE IN PRESS
328
variables: on=u2t ; Fpp ou2t =t2w n. According to Blake (1986), attenuation in the spectra should occur approximately for
od/U41.2 that implies on=u2t 40:3 or 0:42, depending on flow velocity. A collapse of the two sets of measurements
occurs at high frequency, i.e. for 0:033oon=u2t o0:3. The Bull and Thomas and Blake data are in excellent agreement in
the same frequency range, although the Blake curve is higher for higher nondimensional frequencies. On the contrary,
the Farabee and Casarella curve shows quite a different trend, characterised by slower high-frequency decay.
Differences can be due to spectra attenuation caused by the finite sensor dimensions and, when considering similar d+
values, to the use of different pressure transducers. In particular, the Farabee and Casarella and Blake data were
obtained using open pinhole microphones with d+ 33 and 68, respectively, while the Bull and Thomas data were
obtained using both filled pinhole microphones and piezoelectric transducers for d+ 44. For the present data,
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
329
obtained with piezoresistive pressure transducers, d+ was equal to 84 for the lower velocity and to 125 for the higher.
Bull and Thomas (1976) showed that the use of an open pinhole microphone leads to a higher amplitude of pressure
spectra for on=u2t 40:1. This fact can explain the mismatch between different sets of measurements. Finally, due to the
high Reynolds numbers, the overlap region had a considerable extension. From the analysis of the range of validity of
the outer and inner scales or from the direct inspection of the range of validity of the o!1 law, included in both Figs. 5
and 6, it can be concluded that the overlap region extends in the range 0:033 Ret 335ood=ut o1760 or
0:033oon=u2t o1760=Ret 0:174 where the lowest value of Ret was used.
3.2. Cross-spectral density
The spatial characterisation of WPFs is now analysed extracting from the experimental data the streamwise
Fpp0 x; 0; o and the spanwise Fpp0 0; Z; o CSDs. Since the CSD is a complex quantity, as usual, the coherence function
.qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
F1
Gx; Z; o jFpp0 x; Z; oj
pp o % Fpp o, where in the square root appear the ASDs of the two pressure signals, is
used to display the results and to discuss their comparison with the Corcos and Chase theoretical models. Coherence
spectra were obtained for quite a large number of streamwise spacings (0:09px=dp1:44), while only few spanwise
separations were considered 0:09pZ=dp0:36, as the coherence decay of the pressure field is very fast in this direction.
(1)
ARTICLE IN PRESS
330
occurs because as far as ox/Uc-0 the coherences do not tend to unity as Corcos model predicts. Unit coherence would
imply that the low-frequency components should be correlated for all spatial separations: this is physically unrealistic.
From the observation of the figure it can also be noted that all curves exhibit a maximum that is indicated by Farabee
and Casarella (1991) as the limit value below which the similarity variables do not hold anymore. The same
considerations can be drawn for 3.3 m/s; in particular, the best fit of the experimental data is found for the same
value of g1.
Less experimental data concerning spanwise coherence are available in the literature: the value usually suggested for
the decay coefficient g3 is 0.7 (Corcos, 1963; Blake, 1986; Bull, 1967). Fig. 8 shows curves of spanwise coherence relative
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
331
to the lower velocity for various nondimensional separations Z/d as a function of the similarity variable oZ/Uc. The
thick solid line represents the Corcos model with g3 equal to 0.7. The exponential function seems to correctly estimate
the measured coherences, above their maxima, for value of the nondimensional frequency oZ/Uc40.8. Also in this case,
velocity variations seem not to have an influence on the decay coefficient value.
3.2.2. The Chase model
A descriptive model of the wavenumber-frequency spectrum of turbulent wall-pressure was proposed by Chase
(1980) with the intention of overcoming the limitations of models built to capture the characteristics of the convective
domain only. Starting from the properties of the fluctuating velocity spectrum and considering its relation with the
fluctuating pressure, Chase proposed a model able to correctly describe the pressure field in the convective and subconvective domains. The inverse Fourier transform of the Chase (1980) expression is here proposed in its complete form
as determined by Josserand and Lauchle (1989) because, as it will be clear in the following, the assumptions at the basis
of some simplifications made by Chase are not necessarily fulfilled. Thus, the complete Chase model in the spacefrequency domain is given by
Fpp0 o; x; Z Fpp oC m f m e$zm e$iox=U m C t f t e$zt eiox=U t ,
"
#
2
Um
$3
2 2 1 $ zm1
f m qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi am 1 zm am mm
2iam mm zm1 ,
zm
U 2c h2m u2t
"
"
#
#
Ut
z2t3 m2t z2t1
$3
2
2
f t qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi at 1 zt at 1 mt $
2iat mt zt1 ,
zt
U 2c h2t u2t
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
am mm ox
at mt ox
am oZ
at oZ
; zt1
; zm3
; zt3
; zm z2m1 z2m3 ,
zm1
U
Ut
Um
Ut
qffiffiffiffiffiffiffiffimffiffiffiffiffiffiffiffi
zt z2t1 z2t3 ,
rm
rt
; Ct
; rm 1 $ rt ,
rt f t0 rm f m0
rt f t0 rm f m0
Um
Ut
2 2
2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
f t0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
m 1 am mm ;
t 1 at 1 mt ,
2 2
2 2
2
2
U c hm ut
U c ht ut
Cm
f m0
Um
Ut
2 2
2 4 $1
2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
Fpp o rm a r2 u4t o$1 a$3
m qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 mm am rt a r ut o
t 1 at 1 mt ,
2 2
2 2
2
2
U c hm ut
U c ht ut
#2
$
% "
U cm
Uc
1
Ui
; a2i
; i m; t.
hi qffiffiffiffiffiffiiffiffiffiffiffiffiffi ; U i
2
2
bm do
1 $ mi
1 $ mi
ut 1 $ m2i
(2)
It can be noted that six free parameters have to be determined by comparison with experimental data. The values
suggested by Chase, based on the comparison between the experimental data of Bull (1967) and limiting values of
Eq. (2), are the following: bm 0.756, bt 0.378, mm 0.176, rt 0.389, and a+ 0.766. As pointed out by Chase,
these values are not supposed to be universal, as already shown for example by the measurements of Finnveden et al.
(2005). In fact, the attempt to fit the experimental data with the Chase model using these values for the free parameters
gave unsatisfactory results. Thus, the first five parameters were evaluated using a nonlinear least-square formulation
based on a trust-region approach. The best fit of the experimental streamwise and spanwise coherences for both velocity
conditions is found for: bm 0.51, bt 0.35, mm 0.13, mt 0.4, and rt 0.3. Fig. 9 presents the comparison among
the experimental streamwise coherence for U 5.3 m/s, the Chase model, using the identified parameters, and the
Corcos model for different values of the ratio x/d. In Fig. 10, the same comparison is shown for the spanwise coherence
relative to U 3.3 m/s. At this time, some considerations must be made: the coefficient bm gives the position of the ASD
maximum that, according to the
measurements occurs for od/utE60; on the other hand, using Chase relations,
pffipresent
ffi
the maximum is given by o 2U c =bm d, thus bm 0.51 is required if Uc 0.65U is assumed. In fact, the value 0.756
suggested by Chase represented an average between the value 0.53 needed to fit the maximum in the Bull spectrum, thus
very close to that already found, and the value 0.9 needed to fit the measured spatial correlation. Preliminary
comparisons between the experimental CSD and the simplified Chase model gave a value for the coefficient mt sensibly
higher than that suggested by Chase. Thus, it was evident that the hypothesis mm, mt51 at the basis of the simplifications
ARTICLE IN PRESS
332
made by Chase was, in this case, not valid. For this reason, the complete form as in Eq. (2) is used to perform the
analysis. It can be concluded that, except for mt and somewhat for mm, the values of the identified parameters are not so
far to those found by Chase analysing pressure experimental data acquired in completely different flow conditions.
Finally, a+ is obtained by a direct comparison with the measured ASD (see Fig. 11) when the other coefficients are
fixed. This parameter determines the amplitude of the ASD spectrum and in particular of its maximum, the best
agreement with experimental data is found for a+ 0.8. On the other hand, if the interest is not in the maximum but in
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
333
the higher frequency scale-independent region, the value of a+ should be 1.5. However, to perform the structural
analysis described in Section 4.3, the measured ASD was used.
ARTICLE IN PRESS
334
convection velocities divided by the free-stream velocity U, obtained for fixed spatial separations x/d, are plotted as a
function of the dimensionless frequency od/ut. Different figures refer, as usual, to the two different free-stream
velocities. Farabee and Casarella (1991) have observed a peak value of convection velocity for od/ut 50,
independently of the x/d values. This peak corresponds exactly to the maximum observed in the spectra and to the value
that separates the low and the high-frequency behaviour in the coherence function. This fact demonstrated that not only
the lowest wavenumber components experienced a decay, but that they are also convected at lower overall velocity.
In the present case, CSD analysis was performed without using noise cancellation technique to avoid phase alteration.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
335
On the other hand, convection velocity obtained by a division for the CSD phase is very sensitive to small disturbances
leading to unreasonable value of the convection velocity. For this reason, frequencies below 25 Hz were cancelled out
from the graph.
By inspection of Figs. 12 and 13, an increase of the convection velocity with the growing of the spatial
separation is observed; this trend of Uc is related to an increasing dominance of large-scale events to the two-point
correlation as the separation increases. For small separations, the correlation data and hence the convection velocity are
dominated by the small-scale eddies close to the wall, which move with a lower velocity than the large-scale events.
However, as far as spatial separation increases, the curves tend to collapse in a unique curve indicating that, even if the
Taylor frozen flow hypothesis does not strictly hold for the smallest spatial separations, the convection velocity can be
represented as a function of the single variable od/ut. Moreover, with increasing frequency, the convection velocities
assume a flat trend.
The dependence of the convection velocity on the spatial separation can be better highlighted from the analysis of the
space-time correlation functions of the pressure signals. The convection velocity is obtained as the ratio x/t at which
the cross-correlation Rpp(x,t) has a maximum. In Fig. 14, the convection velocities normalised with respect to the
free-stream velocity for 3.31 and 5.31 m/s are depicted as a function of the nondimensional parameter x/d*. The
experimental data are also compared with those obtained experimentally by Bull (1967) and numerically by Na and
Moin (1996). The values of Uc range from 0.6 for the smaller separations associated with the small-scale structures, to
0.73 at higher separations, related to the larger ones. From the above considerations, as proposed by several authors,
the convection velocity can be modelled with a constant average value between small- and large-scale convection
velocities, in this case, equal to 0.65 U.
where ks k2m k2n is the primary effective wavenumber component of the vibration. Hence, the numerical wet
natural frequencies were computed by the following expression:
!
"$1=2
r
mn ffi omn 1
o
,
(5)
rs hks
where omn are the numerical dry natural frequencies computed by the aforementioned FE model. The primary
wavenumber used in Eq. (5) corresponds to simply supported boundary conditions, i.e. km mp/a and kn np/b this
last being the only one analytically known. It is clear that this assumption is valid as far as the frequency increases. In
Fig. 15, the experimental and the theoretical added mass curves, mef o and mnf o, respectively, for the first 16 modes,
are displayed showing a difference of about 23% for the first mode that decreases, as expected, with increasing mode
order. Moreover, although natural modes of plates surrounded by unbounded fluids are not mathematically
orthogonal, their shapes remain almost unchanged (Fahy, 1985). This result was partially verified by the experimental
ARTICLE IN PRESS
336
analysis. In fact, a modal identification was performed using a number of points, sufficient to identify the mode order
and the position of nodes and maxima along some selected lines, but not to represent the whole mode shapes.
In conclusion, to calculate the structural response with the numerical procedure described in the following section: (i)
the dry modes provided by the FE analysis are used, (ii) the first 16 wet natural frequencies were obtained
experimentally, and (iii) the remaining ones were estimated by using the dry natural frequencies provided by the FE
analysis and Eq. (5).
The structural modal damping coefficient, Z, was evaluated from wet hammer impact tests: it decreases from 0.034 for
the first modes to a quite constant value around 0.018 for the higher modes.
4.2. Remarks on the prediction of structural response
A procedure based on the finite element approach was presented and discussed in the recent literature to solve the
response of a plate under a TBL excitation (De Rosa and Franco, 2007). Specifically, the Corcos model was used for
comparing the numerical response with the exact one, and in order to define a general methodology, able to work for
any TBL model at acceptable computational costs. For the sake of clarity, some details are herein briefly recalled.
The cited finite element procedure is assembled by using the following equation suitable for all the methods working
with discrete coordinates (Elishakoff, 1983); the CSD matrix of displacements of a structural operator represented by
using NG degrees of freedom and NM mode shapes is given by
SW o UHoSU oHo$ UT ,
(6)
SU o UT SFF oU,
(7)
with
where U is the structural modal matrix (each column is an eigenvector sampled at the NG selected points), [NG % NM]
and the generic term of H(o) is H j o o2j ' o iZo2j )'1 , [NM % NM].
The translation of the distributed random loads to the set of NG points, in other words the way of representing the
SFF, can be solved in the framework of the finite element method by using consistent approach, that is by using the
shape function vector, N, belonging to each element:
Z Z Z Z
E
k q q
k
q
k
q
SCFF
NT Fpp0 xk
(8)
1 ; x2 ; x1 ; x2 ; oNdx1 dx1 dx2 dx2 ,
k;q
xk
1
xk
2
xq
1
xq
2
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
337
where the double integers k and q indicate here two generic finite elements and the integration is related to the area of
each of them. The vector N can be interpreted as the interpolating function basis selected by the analyst according to the
specific problem and boundary conditions (Cook, 1981). Thus, Eq. (8) serves to evaluate a generic kqth member of the
NE ! NE load matrix, where NE is the number of elements. A simplified approach could refer to each grid point rather
than each finite element. This means that the load acting on the ith grid point will be the resultant of the distributed load
working on the equivalent nodal area, say Areai, belonging to it. This area vector can be evaluated easily by using a
static deterministic unit pressure load (De Rosa et al., 1994; Hambric et al., 2004). Accordingly, one gets the generic ijth
member of the NG ! NG matrix:
Z xi Dx=2 Z xj Dx=2 Z yi Dy=2 Z yj Dy=2
G
SCFF i;j
Fpp0 xi ; xj ; yi ; yj ; odyj dyi dxj dxi .
(9)
xi &Dx=2
xj &Dx=2
yi &Dy=2
yj &Dy=2
An area DxDy is assigned to both points P(xi, yi) and Q(xj, yj) and the double space integration refers to these finite
domains. A further approximation could also be introduced, considering that the wall-pressure distribution due to the
TBL in the low-frequency range does not fluctuate very quickly. In this case, the last integral could be approximated as
follows:
G
2
SLFF
i;j Fpp0 xi ; xj ; yi ; yj ; oDxDy( .
(10)
Obviously, the approximations represented by Eqs. (8)(10) are associated with decreasing computational cost.
The problem of the plate response under a convective random load, expressed in discrete form as described by
Eqs. (6) and (7), can be accurately approached only when adequately resolving both the spatial distributions of the
response function and of the forcing function; in particular, since in this case Uc5cB, the discretisation length, is ruled
by the hydrodynamic load.
In this work, Eq. (9) was used to calculate the SFF matrix because Eq. (10) was not adequate for the present
simulations. In fact, in the frequency range of interest, Eq. (9) allows the avoidance of the numerical divergence of the
structural response due to the incorrect representation of the pressure load, as approximated by Eq. (10). Some further
details of the numerical simulations are given in the next paragraph. The finite element approach was used to generate
the modal base, while the responses were calculated by a specific Fortran code.
4.3. Experimental analysis and comparisons
The ASDs of the acceleration signals, experimentally measured in eight different points over the plate, were
computed and the results averaged and compared with the results obtained by the numerical procedure exposed in the
previous section. The numerical results were obtained applying Eqs. (6), (7) and (9) for both the Corcos and Chase
models.
The FE analysis was performed using 61 ! 21 grid points corresponding to a spatial discretisation of 1 cm in both
directions. The integral in Eq. (9) was calculated using the trapezoidal rule. The hydrodynamic parameters inserted in
Eq. (9) were those identified in Sections 3.13.3; in particular, the experimental ASD was used and the convection
velocity was assumed constant over the whole frequency range and equal to 0.65U for both ship speeds. After having
performed the convergence analysis, each integration domain was finally subdivided, for both velocity conditions, in
eight intervals when the Corcos model was used and in 24 intervals when the Chase model was used instead. The Corcos
numerical solution of integral in Eq. (9) was compared and validated by an analogous analytical solution (De Rosa and
Franco, 2007). In both cases, the number of retained natural modes was 100.
The response of the plate was computed for a frequency range between 1 and 1000 Hz (the step was 4.5 Hz) for the
higher velocity and only between 1 and 600 Hz for the lower one, because above this frequency a refined mesh
ffiffiffiffiffiffiffiffipbe
pffiffiffiffiffimust
ffiffiffi
used to obtain convergence. In these frequency ranges, the ratio between the bending wavenumber kB 4 rs h=D o
and the convective wavenumber kc (see Fig. 16) varied between 0.045 and 0.21 for the lowest velocity and between 0.057
and 0.35 for the highest velocity.
Figs. 17 and 18 present the experimental and numerical averaged ASDs of the plates acceleration for 3.3 and 5.3 m/s,
respectively.
Unsurprisingly, an overestimation of the plate response is evident in the whole frequency range if the Corcos model is
used; on the other hand, the numerical response obtained using the Chase model is undoubtedly in better agreement
with experimental data. However, below 2530 Hz both pressure spectra (although cleaned) and structural response are
contaminated by the carriage vibrations transmitted to the model; thus, any comparison is meaningless. Above these
frequencies, the agreement is really satisfactory until 420 Hz for the lower velocity case and until 650 Hz for the higher
one. In the high-frequency part, the Chase model tends to slightly underestimate the experimental curve; this fact, more
ARTICLE IN PRESS
338
visible for 3.3 m/s, can be partly due to the poor spatial resolution of pressure transducers that attenuate the highfrequency part of the ASD spectrum.
Finally, an evident mismatch between the experimental and the numerical curve, generated by the presence of high
peaks in the experimental data, can be observed in Fig. 18 around 800 Hz probably due to local flow disturbances. To
better quantify the difference between model and experimental results, the previous curves are plotted in Figs. 19 and 20
in third-octave bands. It can be seen that the root mean square of the difference between the response obtained applying
the Chase model and the experimental data is 5.2 dB for the lower velocity and 4.1 for the higher one. The response
obtained by using the Corcos model to represent the surface pressure field, overpredicts in both cases the experimental
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
339
data resulting in an average difference of 18 dB. The small gap between the numerical predictions obtained by applying
the Chase model and the experimental plate response that is less or, at least, of the same order of that found by
Finnveden et al. (2005) demonstrated the validity of the developed procedure and the capability of the Chase model to
represent the surface pressure field on a ship hull. However, a more careful determination of the added water mass in
the whole frequency range can improve the numerical estimation, as well as a deep uncertainty analysis can better
indicate the confidence interval of the present results.
ARTICLE IN PRESS
340
5. Concluding remarks
In this paper, a complete analysis of the coupled structural-fluid problem concerning the response of an elastic plate
inserted in the bottom of a catamaran hull excited by the TBL WPFs, has been carried out.
The hydrodynamic analysis, performed on a rigid plate at high Reynolds number and in equilibrium flow conditions
allowed the determination of the appropriate scaling laws for the ASDs in the different frequency ranges. Moreover, the
analysis of the cross-spectral densities in longitudinal and lateral flow directions was used to fit the theoretical models,
available for the pressure representation, to the experimental data. Two theoretical models are analysed: the Corcos and
Chase models. It was demonstrated that it is possible to find for both a complete set of free parameters that provide a
fair agreement with experimental CSD data.
Since spatial resolution was too poor to analyse the CSD behaviour at high frequency and since with this type of
experimental analysis it was not possible to isolate the longest wavelengths, the studied pressure behaviour mainly
concerned the characteristics of the convective domain. Thus, an indirect comparison based on the vibrational response
of a plate was performed; in particular, the numerical structural responses obtained using the two models were
compared with experimental measurements. The conclusion of this analysis is that, although the Chase model is
complex and dependent on several empirical parameters, it provides a very good agreement with experimental data at
low wavenumbers. The performed analysis can give interesting information also for the full-scale problem; in fact,
considering realistic values of the hydrodynamic and of the structural parameters, coincidence conditions usually
appear at very low frequency both for underwater and surface marine vehicles.
The main disadvantage in using Chase model lies in its non-predictive character. In fact, it was shown that the
original Chase parameters do not fit the experimental data; thus a new set of parameters have been determined and the
complete version of the model has been used.
It is clear that the aim of any numerical procedure is to produce robust predictive tools to be used at the design stage.
Ongoing comparisons between pressure measurements performed on different ship models and for various flow
conditions in terms of Reynolds number values are aimed to analyse the range of variability of the parameters and their
dependence on the particular flow conditions.
Acknowledgement
The research was supported by the Ministero dei Trasporti in the frame of Programma Ricerche Luglio2006Dicembre 2007.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
341
References
Abraham, B.M., 1998. Direct measurements of the turbulent boundary layer wall pressure wavenumber-frequency spectra. ASME
Journal of Fluids Engineering 120, 2939.
Bendat, J.S., Piersol, A.G., 1991. Random Data: Analysis and Measurements Procedure. Wiley, New York.
Blake, W.K., 1970. Turbulent boundary-layer wall-pressure fluctuations on smooth and rough walls. Journal of Fluid Mechanics 44,
637660.
Blake, W.K., 1986. Mechanics of Flow Induced Sound and Vibration. Academic Press, Orlando.
Blevins, R.D., 1987. Formulas for Natural Frequency and Mode Shape. Robert E. Krieger Publishing Company, Melbourne, FL.
Bull, K.M., 1967. Wall pressure fluctuations associated with subsonic turbulent boundary layer flow. Journal of Fluid Mechanics 28,
719754.
Bull, M.K., 1996. Wall pressure fluctuations beneath turbulent boundary layers: some reflections on forty years of research. Journal of
Sound and Vibration 190, 299315.
Bull, M.K., Thomas, S.W., 1976. High frequency wall pressure fluctuations in turbulent boundary layers. Physics of Fluids 19,
597599.
Chang, P.A., Piomelli, U., Blake, W.K., 1999. Relationship between wall pressure and velocity-field sources. Physics of Fluids A 11,
34343448.
Chase, D.M., 1980. Modelling the wavevector-frequency spectrum of turbulent boundary layer wall pressure. Journal of Sound and
Vibration 70 (1), 2967.
Choi, H., Moin, P., 1990. On the space-time characteristics of wall pressure fluctuations. Physics of Fluids A 2, 14501460.
Ciappi, E., Magionesi, F., 2005. Characteristics of the turbulent boundary layer pressure spectra for high speed vessels. Journal of
Fluids and Structures 21, 321333.
Cook, R.D., 1981. Concept and Application of Finite Element Analysis. Wiley, New York.
Corcos, G.M., 1963. Resolution of pressure in turbulence. Journal of the Acoustical Society of America 35, 192199.
De Rosa, S., Franco, F., 2007. Exact and numerical responses of a plate under a turbulent boundary layer excitation. Journal of Fluids
and Structures 24, 212230.
De Rosa, S., Pezzullo, G., Lecce, L., Marulo, F., 1994. Structural acoustic calculations in the low frequency range. AIAA Journal of
Aircraft 31 (6), 13871394.
Efimtsov, B.M., 1982. Characteristics of the field of turbulent wall pressure fluctuations at large Reynolds numbers. Soviet PhysicsAcoustics 28 (4), 289292.
Elishakoff, I., 1983. Probabilistic Methods in the Theory of Structures. Wiley, New York.
Fahy, F., 1985. Sound and Structural Vibration. Academic Press, Orlando.
Farabee, T.M., Casarella, M.J., 1991. Spectral features of wall pressure fluctuations beneath turbulent boundary layers. Physics of
Fluids A 3, 24102420.
Farabee, T.M., Geib, F.E., 1991. Measurements of boundary layer pressure fluctuations at low wavenumbers on smooth and rough
walls. In: ASME Symposium on Flow Noise Modelling, Measurements and Control, NCA-vol. 11, FED-vol. 130, pp. 5568.
Ffowcs Williams, J.E., 1982. Boundary layer pressures and the Corcos model: a development to incorporate low wavenumber
constraints. Journal of Fluid Mechanics 125, 925.
Finnveden, S., Birgersson, F., Ross, U., Kremer, T., 2005. A model of wall pressure correlation for prediction of turbulence-induced
vibration. Journal of Fluids and Structures 20, 11271143.
Goody, M., 1999. An experimental investigation of pressure fluctuations in three-dimensional turbulent boundary layers. Ph.D. Thesis,
Department of Aerospace and Ocean Engineering, Virginia Tech, USA.
Graham, W.R., 1997. A comparison of models for the wavenumber-frequency spectrum of turbulent boundary layer pressures. Journal
of Sound and Vibration 206 (4), 541565.
Hambric, S.A., Hwang, Y.F., Bonness, W.K., 2004. Vibrations of plates with clamped and free edges excited by low-speed turbulent
boundary layer flow. Journal of Fluids and Structures 19, 93110.
Han, F., Bernhard, R.J., Mongeau, L.G., 1999. Prediction of flow-induced structural vibration and sound radiation using energy flow
analysis. Journal of Sound and Vibration 227 (4), 685709.
Hwang, Y.F., Maidanik, G., 1990. A wavenumber analysis of the coupling of a structural mode and flow turbulence. Journal of Sound
and Vibration 142, 135152.
Keith, W.L., Hurdis, D.A., Abraham, B.M., 1992. A comparison of turbulent boundary layer wall-pressure spectra. ASME Journal of
Fluids Engineering 114, 338347.
Josserand, M.A., Lauchle, G.C., 1989. ERRATA: D.M. Chase JSV 1980. Journal of Sound and Vibration 128, 519523.
Lee, Y.T., Blake, W.K., Farabee, T.M., 2005a. Modeling of wall pressure fluctuations based on time mean flow field. ASME Journal of
Fluids Engineering 127, 233240.
Lee, Y.T., Miller, R., Gorski, J., Farabee, T., 2005b. Predictions of hull pressure fluctuations for a ship model. In: Proceedings of
International Conference on Marine Research and Transportation (ICMRT05), Ischia, Italy.
Manoha, E., 1996. The wavenumber-frequency spectrum of the wall pressure fluctuations beneath a turbulent boundary layer. In:
Proceedings of the AIAA Aeroacoustics Conference, AIAA paper 96-1758.
Na, Y., Moin, P., 1996. Direct numerical simulation of turbulent boundary layers with adverse pressure gradients and separation.
Report No. TF-68. Thermosciences Division, Department of Mechanical Engineering, Stanford University.
ARTICLE IN PRESS
342
Panton, R.L., Robert, G., 1994. The wavenumber-phase velocity representation for the turbulent wall-pressure spectrum. ASME
Journal of Fluids Engineering 116, 477483.
Smolyakov, A.V., Tkachenko, V.M., 1991. Model of pseudosonic turbulent wall pressures and experimental data. Soviet PhysicsAcoustics 37 (6), 627631.
Willmarth, W.W., Wooldridge, C.E., 1962. Measurements of the fluctuating pressure at the wall beneath a thick turbulent boundary
layer. Journal of Fluid Mechanics 14, 187210.
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
18)
a r t i c l e in f o
abstract
Article history:
Received 3 November 2010
Received in revised form
27 October 2011
Accepted 2 November 2011
Available online 5 December 2011
The high computational costs, associated to the numerical solution of the fluctuating
pressure field generated at the wall by the turbulent boundary layer and of the induced
structural response, push for the exploration of alternative methodologies of analysis.
Wall pressure fluctuations spectra are often modeled using semi-empirical expressions
based on the experimental evidence and on the identification of universal scaling laws.
In this work the possibility to adopt a dimensionless representation, able to provide a
universal expression for the structural response of plates under turbulent boundary
layer excitations, is investigated with the help of pressure fluctuations and acceleration
experimental data sets. The test article is a plane thin plate wetted by a fluid over one
face, the boundary layer is fully developed and pressure gradient effects are negligible.
The attention is devoted to the investigation and the definition of a normalization of the
required axes: the excitation frequency and the power spectral density of the structural
response. The analysis is initially based on analytical models for the structural response
under turbulent boundary layer excitations. The proposed scaling laws are successively
and successfully applied to four data sets measured in different conditions both in wind
tunnels and in a towing tank.
& 2011 Elsevier Ltd. All rights reserved.
Keywords:
Wall pressure fluctuations
Plate response
Dimensional analysis
Scaling laws
1. Introduction
Turbulent boundary layer (TBL), inducing vibrations of elastic structures, is one of the major noise sources in naval,
aerospace, and automotive engineering.
It is well known that the numerical solution of this fluid structure interaction problem can be so computationally
demanding as to be impractical for real application. In fact, direct numerical simulations (DNS) of the NavierStokes
equations are generally limited to problems in which the local Reynolds number, based on the momentum thickness, is in
the order of 300 (Choi and Moin, 1990). A significant reduction of the computational time can be certainly obtained using
RANS (Reynolds Average Navier Stokes) simulations (Lee et al., 2005; Peltier and Hambric, 2007). In particular Peltier and
Hambric proposed an original stochastic model for the representation of the space-time wall pressure spectrum that used
statistical data obtained from RANS calculations. The values of ReW are in this case between 1400 and 8000.
To overcome the limitations of actual CFD capabilities, the attention of the research community is mainly directed to
the analytical characterization of the pressure field by the definition of scaling laws for the power spectral density (PSD)
(Bull, 1996; Ciappi et al., 2009; Goody, 2004; Keith et al., 1992) and of predictive models (Corcos, 1964; Chase, 1980;
Corresponding author. Tel.: 39 065 0299 268; fax: 39 065 070 619.
E-mail address: e.ciappi@insean.it (E. Ciappi).
0889-9746/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfluidstructs.2011.11.003
ARTICLE IN PRESS
Abstract
Recent studies have demonstrated that the characterisation of wall-pressure fluctuations for surface ships is of great
interest not only for military applications but also for civil marine vehicles. A ship model towed in a towing tank is used
to perform pressure and structural measurements at high Reynolds numbers. This facility provides ideal flow conditions
because background turbulence and noise are almost absent. Free surface effects are naturally included in the analysis,
although in the particular section chosen for the present study do not have significant consequences on pressure spectra.
Scaling laws for the power spectral density are identified providing the possibility to estimate pressure spectra for
different flow conditions and in particular for full-scale applications. The range of validity of some theoretical models
for the cross-spectral density representation is analysed by direct comparison with experimental data of wall-pressure
fluctuations measured in streamwise and spanwise direction. In a second phase, an indirect validation is performed by
comparing the measured vibrational response of an elastic plate inserted in the catamaran hull with that obtained
numerically using, as a forcing function, the modelled pressure load. In general, marine structures are able to accept
energy mainly from the sub-convective components of the pressure field because the typical bending wavenumber values
are usually lower than the convective one; thus, a model that gives an accurate description of the phenomenon at low
wavenumbers is needed. In this work, it is shown that the use of the Chase model for the description of the pressure field
provides a satisfactory agreement between the numerical and the experimental response of the hull plate. These
experimental data, although acquired at model scale, represent a significant test case also for the real ship problem.
r 2008 Elsevier Ltd. All rights reserved.
Keywords: Wall-pressure fluctuations; High Reynolds number flow; High-speed vessels; Theoretical models; Vibrational response
1. Introduction
Vibrations of elastic structures excited by the turbulent boundary layer (TBL) are of interest for interior and exterior
noise emission problems in aeronautical, automotive and marine applications. In particular, new requirements in terms
of comfort on board high-speed ships for passenger transportation have addressed the attention of the scientific
community to the identification and to the characterisation of noise sources including those of hydrodynamic nature.
Recent studies performed in the framework of the European RTD project NORMA (Noise Reduction for Marine
!Corresponding author: Tel.: +39 081 7683581; fax: +39 081 624609.
ARTICLE IN PRESS
322
Nomenclature
a
Areai
b
c
cB
D
d
d0
d+
Fr
g
H
H
h
i
j
k
kc
kB
Lpp
m, n
me
mn
N
NG
NM
ReW
Ret
Rpp
Saa
SW
SF
SFF
u+
ut
U
Uc
x
y
y+
convection velocity
streamwise reference axis
spanwise reference axis
wall unit, y+ yut/n
Greek symbols
g1
g3
G
d
d*
Dx
Dy
Z
Zp
W
y
n
x
rs
r
t
tw
Fpp
Fpp0
U
o
oj
omn
mn
o
Applications G3RD-2001-0393) demonstrated that, at least for new concept design fast ships, flow noise sources, e.g.,
the TBL, play an important role above 30 knots.
The typical way to characterise wall-pressure fluctuations (WPF) is via experimental tests performed in suitable
facilities like wind or water tunnels. In fact, direct numerical simulations (DNS) or large eddy simulations (LES) are
often not applicable in the case of complex geometries and realistic flow conditions (high Reynolds numbers) due to the
limitation of computational resources. DNS of WPF were performed by Choi and Moin (1990), analysing the channel
flow problem for ReW UW/n 287. Furthermore, Chang et al. (1999) analysed the influence of the different TBL
velocity components on the wavenumber pressure spectra in a channel flow for a Reynolds number, based on the
channel half width, equal to 3200. Recently, Lee et al. (2005a) proposed a new methodology to calculate numerically
wall-pressure spectra. The method uses the predicted mean flow field obtained from RANS calculations and a spectral
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
323
correlation model, and integrates across the TBL. The method was validated both for an equilibrium flow at
ReW 3582 and for a non-equilibrium flow resulting from flow over a backward-facing step. Using the same
methodology, Lee et al. (2005b) characterised wall-pressure spectra for a surface ship model including the effects of hull
curvature and of the free surface. The comparison of the scaled spectra, obtained while varying the axial location, and
the distance from the free surface with spectra, obtained for an equilibrium flow, showed that, in several locations, the
former deviate from the canonical case.
On the other hand, there are many experimental works related to WPF, most of them devoted to the identification of
the appropriate scaling laws for the auto-spectral density (ASD) for zero pressure gradient flow. The pressure ASD
frequency range is subdivided according to the boundary layer regions that give contributions to wall-pressure spectra
where different scaling variables hold. In particular, Farabee and Casarella (1991) identified four frequency ranges in
their data: the low-frequency and the mid-frequency range where outer variables hold, the high-frequency range where
inner variables hold, and an overlap scale-independent region proportional to o"1, whose extent depends on the
Reynolds number. With respect to this point, Keith et al. (1992) presented the most extensive comparison among many
available experimental data obtained in fully developed and developing channel flow, in fully developed pipe flow and in
wind tunnel, over a wide range of Reynolds numbers, with the aim of identifying the best choice for the scaling
parameters in the different frequency regions. Goody (1999) performed an experimental campaign in a two-dimensional
boundary layer for ReW values ranging from 7800 to 23 400, investigating different combinations of scaling parameters.
Finally, a detailed review of the state-of-the-art on this subject can be found in Bull (1996).
The spatial characterisation of WPFs was first analysed by Corcos (1963) on the basis of measurements performed by
Willmarth and Wooldridge (1962). Assuming the validity of separation of variables in the streamwise and spanwise
directions, Corcos stated an exponential decay for the cross-spectral density (CSD) as a function of the similarity
variables ox/Uc and oZ/Uc, where Uc is the convection velocity, and x and Z are the streamwise and spanwise spatial
separation, respectively. Several authors have performed comparisons between measured CSD data and Corcos model
(Blake, 1986; Bull, 1967); in particular, Farabee and Casarella (1991) from the analysis of their experimental data
provided, at least in a certain nondimensional frequency range, a confirmation of this pressure behaviour for a wide
series of spatial separations in streamwise direction and for different flow velocities or local Reynolds number values.
The success of the Corcos model lies in its simplicity and in its predictive character since the model parameters are
substantially case-independent. Nevertheless, it is generally stated that Corcos model gives a correct representation of
the WPF behaviour in the convective domain, i.e. when the wavenumbers are close to the convective wavenumber
kc o/Uc. On the contrary, in the sub-convective domain the white Corcos spectrum largely overpredicts the real
amplitude. Since for several applications and in particular in the case of underwater and surface marine vehicles, the
convective wavenumber is greater than the bending wavenumber kB o/cB, it is of primary importance to evaluate
correctly the sub-convective domain of pressure spectra that corresponds to the high-sensitivity region for the structure.
Several new models, some directly derived by the Corcos one (Efimtsov, 1982; Ffowcs Williams, 1982), others
overcoming the Corcos multiplicative approach such as those by Chase (1980) and Smolyakov and Tkachenko (1991),
were developed to improve the estimation of pressure spectra in this region. A comparison between the predictions of
the radiated acoustic power by rectangular plates was carried out numerically by Graham (1997); it was performed for
different test conditions and applying the above models. It was there concluded that the use of sophisticated models
such as the Chase one is needed only for structures that do not exhibit coincidence, but that for aircraft the best model is
the one which provides an accurate description of the convective peak, thus suggesting the use of the Efimtsov model.
Nonetheless, no experimental evidence supporting these conclusions was reported in Grahams work. However, the
spatial domain comparison between pressure experimental spectra and theoretical models cannot definitively indicate
the best in describing the different wavenumber regions. It is usually possible to find a set of parameters for each model
able to provide a good data fit. It is clear that most of the energy of WPF is concentrated around the convective peak
and then any correlation data is mainly the representation of the convective character of the TBL. Unfortunately, only
few experimental data concerning direct measurements of the wavenumber-frequency spectrum are available
(Abraham, 1998; Choi and Moin, 1990; Panton and Robert, 1994; Farabee and Geib, 1991; Manoha, 1996) and,
among them, a big spread of the spectra magnitude at low wavenumbers is present as reported for example by Hwang
and Maidanik (1990).
In order to overcome the limitations of flow measurements, an indirect approach to estimate the validity of different
models for WPF representation, based on the analysis of the response of simple elastic structures to the TBL load, is
proposed here. The same idea was recently applied by Finnveden et al. (2005), who compared the measured response of
a flat plate with those obtained numerically using modelled pressure loads. This work presented the first and, to the
authors best knowledge, the only correlation between aerodynamic and structural data measured in the same facility
and with the same set-up. They suggested a modified version of the Corcos model by introducing a frequency and flow
speed dependence on the parameters and of the Chase model by introducing two new parameters to better fit the
ARTICLE IN PRESS
324
spanwise coherence to measurements. Despite the modifications made, the conclusion was that, above the aerodynamic
coincidence (kc kB), only the Chase model, that does not make use of the multiplicative approach, provides a fair
agreement with experimental data. In this work, the lower kB/kc ratio was 0.4 and the average difference between the
modified Chase model predictions and the experimental data was 5 dB. Furthermore, Hambric et al. (2004), although
retaining the multiplicative approach, proposed a modification of Corcos streamwise coherence to better represent the
low wavenumber domain. The model was compared with the experimental response of an elastic plate measured by Han
et al. (1999). The ratio between the structural wavenumber in flow direction and the convective wavenumber was
between 0.3 and 0.8 and the agreement with the experimental data was quite good. On the other hand, Han et al. (1999)
chose Smolyakov and Tkachenko to model the surface pressure field. The comparison with measurements was
performed using the energy flow analysis method to predict the numerical plate response thus, direct information about
the validity of the pressure model are difficult to extract from their data.
The aim of this work is to develop a general procedure based on the identification of the scaling laws and on the use of
predictive models for the surface pressure field suitable for application to full-scale problems. In particular, the capabilities of
Corcos and Chase models to predict the response of an elastic plate inserted in the hull of a ship model were investigated on
the basis of hydrodynamic and vibration data acquired, at high Reynolds numbers, in a towing tank. In a first step, pressure
data were analysed to provide their spectral characteristics. This analysis is fundamental to identify the scaling laws for the
ASD and the free parameters contained in the CSD wall-pressure fluctuation models. The high Reynolds number achieved
with this set-up provides an interesting extension to the previous analyses. In a second phase, a comparison between the
numerical response of the plate obtained using the two models and the experimental response is provided. Since in the
present problem the convection velocity is very low, the ratio between the bending and the convective wavenumber is
sensibly lower than those previously analysed in the technical literature. This fact is fundamental for real size marine
applications for which hydrodynamic coincidence appears, even for high-speed vehicles, at very low frequency.
This first section is aimed to frame the work in the proper existing literature. Section 2 presents the experimental setup and all the data concerning the acquisition instrumentation. The treatment of the pressure data is the specific
argument of Section 3. Section 4 is fully devoted to the analysis of the structural response and the final comparison
between predictive and measured data. Section 5 presents the concluding remarks with some foreseen activities. For the
sake of completeness, a graphic workflow has been also added in Chart 1.
2. Experimental set-up
2.1. Pressure measurements
The experiments were performed on a 1:15 scale model ofpthe
ffiffiffiffiffiffiffifast
ffiffi catamaran Jumbo CAT (Fig. 1). The scale of the
model was chosen according to Froude similarity: Fr U= gLpp where Lpp is the length between perpendiculars, i.e.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
325
the length of the vessel along the waterline between the forward and aft perpendiculars, as depicted in Fig. 1. The
maximum model width is 1.467 m, Lpp is 4.635 m and its draft in calm water conditions is 0.2 m.
The experiments were carried out in the INSEAN towing tank no. 2 which is 220 m long, 9 m wide and 3.5 m deep
and is equipped with a carriage that can reach a maximum speed of 8 m/s. The use of this kind of facility creates ideal
flow conditions because background turbulence and noise are avoided. The measuring section was chosen in the stern
part of the ship bottom where the hull surface is almost flat. To perform pressure measurements a 2 cm thick rigid
plexiglas plate was inserted in the hull bottom where pressure transducers were positioned.
The basic set-up is presented in Fig. 2 and consisted in an array of nine transducers in streamwise direction and five
transducers in the spanwise direction flush-mounted with the plate at constant distance of 1 cm between each other.
Additional tests were performed with 13 transducers mounted in streamwise direction within a maximum distance of
40 cm. Thus, the first pressure sensor was located at x/Lpp 0.88 while the last at x/Lpp 0.97. The minimum distance
between transducers was constrained by the transducers maximum external size while the maximum distance was
chosen according to the fact that for x/d*420 the longitudinal correlation is almost zero as demonstrated by previous
measurements (Bull, 1967; Blake, 1986). Pressure signals were acquired in calm water conditions with fixed trim and
sink and for two different ship model velocities: 3.31 m/s (25 knots) and 5.3 m/s (40 knots) corresponding to Fr 0.49
and 0.78, respectively. The measurement error in the carriage velocity was within 1% of the nominal mean velocity.
Differential piezoresistive pressure transducers Endevco 8510-B, characterised by a maximum range of 2 psig and by
a certified flat response until 14 kHz were used to measure pressure fluctuations. The transducers were statically
calibrated in water using known water level heights. All the transducers showed a linear trend; however, the deviation
around the regression line of the data points used for the sensitivity estimate was evaluated. The standard error of
estimate was very low for all the transducers, of the order of 1%. Moreover, the total error due to thermal sensitivity,
nonlinearity and pressure hysteresis, as reported in the data sheet, is around 1%. The rectangular sensing element has
an area of 1 " 0.3 mm2, hence the effect of the finite size of the transducers surface can be expressed in term of the
nondimensional parameters d0 d/d* and d+ dut/n, where d is the bigger sensor dimension.
Pressure signals were acquired and amplified by the 16 channels acquisition system PROSIG; the sampling frequency
was 12.5 kHz, the acquisition length was 15 s. Several repetitions of the test (typically 1215) under nominally the same
conditions were performed. The data record began a few seconds after the achievement of steady conditions. The
Fig. 2. Set-up for pressure measurements (left) and top view of the installation of the plexiglas plate (right).
ARTICLE IN PRESS
326
Table 1
Mean flow velocity parameters: numerical estimation
U (knots)
U (m/s)
d (m)
d* (m)
ut (m/s)
ReW UW/n
Ret dut/n
d0 d/d*
d+ dut/n
25
40
3.31
5.31
0.12
0.113
0.0142
0.0137
1.27
1.3
0.11
0.1626
29 535
42 807
10 153
14 133
0.07
0.073
84
125
reaching of a stationary random process was verified by comparing the ensemble average value, autocorrelation and
cross-correlation of the WPFs of different runs.
Concerning the flow velocity field, the TBL mean parameters (as used in the data analysis and shown in Table 1) were
obtained by available RANS simulations performed in the past over the whole model. This solution was, in this case,
preferred because the experimental evaluation of the boundary layer velocity profiles in a towing tank, although
possible, is a time-consuming process. In fact, it is clear that the acquisition time is limited for each carriage run,
especially for the higher velocities, and that the time needed for the re-establishment of calm water conditions between
two consecutive runs is at least 10 min. A detailed description of the numerical code is provided in Ciappi and
Magionesi (2005) and in the references cited there. The numerical errors can be predictable in an uncertainty of about
4% in the estimation of the TBL parameters from the velocity profiles.
2.2. Vibration measurements
Vibration measurements were performed replacing the rigid plate with a flexible one. The panel, made of plexiglas, is
0.58 m long, 0.2 m wide and 0.003 m thick, it was fixed to the hull model with some mastic in order to provide
impermeable conditions and to reduce the transmission of model vibrations.
A preliminary series of numerical analyses have been performed to exclude the presence of significant plate
deformations due to static and dynamic pressure loads. In fact, for all the flow speeds under consideration, the
maximum displacement was predicted to be 1% of the longitudinal plate dimension.
The acceleration responses were acquired in eight different points (one for each carriage run) randomly chosen on the
plate surface. A Bruel & Kjaer piezoelectric accelerometer type 4393 characterised by a sensitivity of 4.19 mV/g and a
weight of 2.2 g was used for the acquisition. Its mass was negligible with respect to the plate mass in the frequency range
of interest. The accelerometer signal was amplified by a Bruel & Kjaer amplifier type 2635 and acquired with a sampling
frequency of 12.5 kHz by a National Instruments PXI 6052E acquisition system. Preliminary dry and wet calm water
tests were performed with the same set-up and instrumentation to evaluate the plates natural frequencies, hence the
added fluid mass and the modal damping factors. Two additional accelerometers were mounted on the ships hull and
on the connecting system to acquire the spurious vibrations transmitted by the carriage structure.
3. Pressure analysis
In the following sections, the results of the experimental programme devoted to the characterisation of wall-pressure
spectra are presented. The purpose of this analysis was to verify the pressure scaling laws and to provide a general
model for its spatial behaviour. To this aim, ASDs, streamwise and spanwise coherences and convection velocities were
extracted from measurements. Although free surface effects were naturally present, pressure gradient values calculated
on the basis of numerical simulations can be considered negligible in the measuring section. In Fig. 3, the velocity
profiles obtained numerically, used to extract the mean TBL parameter values of Table 1, are shown in wall units y+,
u+. From preliminary analysis it was decided to consider only the Corcos and Chase models as antagonists in this
analysis. In fact, the Efimtsov model has the same trend as Corcos in the low wavenumber domain but, this last is to be
preferred because describes the wall pressure by a simpler expression containing less empirical parameters. The Ffowcs
Williams model was built to extend Corcos model to the acoustic domain, which is beyond the purpose of this analysis;
finally, the Smolyakov and Tkachenko model does not fit well the present hydrodynamic data.
3.1. Power spectral density: scaling laws
The analysis of the scaling laws for the ASD is essential to understand the contribution of the different boundary
layer regions to WPFs. Moreover, due to the particular section chosen to perform pressure measurements and the
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
327
relatively high speed of this vessel, local Reynolds numbers were sensibly high (ReW 29 535 for 3.31 m/s and
ReW 42 807 for 5.31 m/s), providing an interesting extension of the validity of the scaling laws for high Reynolds
number. Following the frequency range division proposed by Farabee and Casarella (1991), three different spectral
regions can be identified: at very low frequency, the spectra collapse using the classical outer flow variables d* and U,
showing a o2 behaviour, advising that sources are associated with the large-scale structures. In the low-mid-frequency
range, the pressure auto spectral densities collapse into a single curve when scaled with the friction velocity ut, the wall
shear stress tw and d, implying that the mid-frequency structures are related with turbulence activity in the outer region
of the boundary layer. In this interval, pressure spectra exhibit their maxima for od/ut 50. Finally, at high-frequency
inner variable scale, which employs ut, tw and n, allows the collapse of the data independently of the Reynolds number,
suggesting that sources are associated with the buffer region of the wall layer. Moreover, between mid- and highfrequency an overlap region, characterised by an o"1 decay, exists where both outer and inner variable scales hold. This
region is related to the turbulence activity in the logarithmic part of the boundary layer and its extension depends on the
Reynolds number value. Recently, Ciappi and Magionesi (2005), considering the frequency division stated by Farabee
and Casarella provided another confirmation of the proposed scaling laws. ASDs were determined using 700 spectral
averages for each signal and a Hamming window function is used to reduce bandwidth leakage. According to classical
pffiffiffiffiffi
theory of random data (Bendat and Piersol, 1991), the statistical convergence error was defined as !r 1= nd , where nd
is the number of spectral averages. In the present analysis, the data random error was equal to 73.8%; thus, the
uncertainty in the calculated pressure spectra, obtained by considering the above and all the previously defined
experimental sources of error (see Section 2), was within the range of 71 dB.
In Fig. 4, a typical ASD signal is displayed showing high peaks in the frequency region between 8 and 20 Hz due to
structural vibrations of the carriage and of the connecting system. The peaks were eliminated, for the ASD analysis,
using suitable relations based on the coherence function (Bendat and Piersol, 1991) between two pressure sensors
located sufficiently far from each other to be correlated only by structural vibrations. The result of this cleaning
procedure is shown in the same figure. Fig. 5 shows the cleaned ASD for the two different test velocities, scaled using
outer flow variables: od=ut ; Fpp ou2t =t2w d; they are shown and compared with the results of Farabee and Casarella
(1991) obtained for ReW 6050 and with the results of Blake (1970) [extracted from Lee et al. (2005a)] obtained for
ReW 8210. From the inspection of the figure, it is evident that there is an excellent agreement of the present
experimental curves in the low-mid-frequency range, i.e. for 20ood/uto1760. Moreover, for od/uto800, they are in a
very good agreement with the Farabee and Casarella curve and in fair agreement with the Blake data. Finally, the
scaled spectra achieved the maximum value for od/utE63. Low-frequency behaviour (od/uto5) is not analysed, since
frequency resolution is too poor to obtain a realistic trend in this region. Fig. 6 shows the present wall-pressure spectra
and the results of Bull and Thomas (1976), Farabee and Casarella (1991) and Blake (1970) scaled on inner flow
ARTICLE IN PRESS
328
variables: on=u2t ; Fpp ou2t =t2w n. According to Blake (1986), attenuation in the spectra should occur approximately for
od/U41.2 that implies on=u2t 40:3 or 0:42, depending on flow velocity. A collapse of the two sets of measurements
occurs at high frequency, i.e. for 0:033oon=u2t o0:3. The Bull and Thomas and Blake data are in excellent agreement in
the same frequency range, although the Blake curve is higher for higher nondimensional frequencies. On the contrary,
the Farabee and Casarella curve shows quite a different trend, characterised by slower high-frequency decay.
Differences can be due to spectra attenuation caused by the finite sensor dimensions and, when considering similar d+
values, to the use of different pressure transducers. In particular, the Farabee and Casarella and Blake data were
obtained using open pinhole microphones with d+ 33 and 68, respectively, while the Bull and Thomas data were
obtained using both filled pinhole microphones and piezoelectric transducers for d+ 44. For the present data,
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
329
obtained with piezoresistive pressure transducers, d+ was equal to 84 for the lower velocity and to 125 for the higher.
Bull and Thomas (1976) showed that the use of an open pinhole microphone leads to a higher amplitude of pressure
spectra for on=u2t 40:1. This fact can explain the mismatch between different sets of measurements. Finally, due to the
high Reynolds numbers, the overlap region had a considerable extension. From the analysis of the range of validity of
the outer and inner scales or from the direct inspection of the range of validity of the o!1 law, included in both Figs. 5
and 6, it can be concluded that the overlap region extends in the range 0:033 Ret 335ood=ut o1760 or
0:033oon=u2t o1760=Ret 0:174 where the lowest value of Ret was used.
3.2. Cross-spectral density
The spatial characterisation of WPFs is now analysed extracting from the experimental data the streamwise
Fpp0 x; 0; o and the spanwise Fpp0 0; Z; o CSDs. Since the CSD is a complex quantity, as usual, the coherence function
.qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
F1
Gx; Z; o jFpp0 x; Z; oj
pp o % Fpp o, where in the square root appear the ASDs of the two pressure signals, is
used to display the results and to discuss their comparison with the Corcos and Chase theoretical models. Coherence
spectra were obtained for quite a large number of streamwise spacings (0:09px=dp1:44), while only few spanwise
separations were considered 0:09pZ=dp0:36, as the coherence decay of the pressure field is very fast in this direction.
(1)
ARTICLE IN PRESS
330
occurs because as far as ox/Uc-0 the coherences do not tend to unity as Corcos model predicts. Unit coherence would
imply that the low-frequency components should be correlated for all spatial separations: this is physically unrealistic.
From the observation of the figure it can also be noted that all curves exhibit a maximum that is indicated by Farabee
and Casarella (1991) as the limit value below which the similarity variables do not hold anymore. The same
considerations can be drawn for 3.3 m/s; in particular, the best fit of the experimental data is found for the same
value of g1.
Less experimental data concerning spanwise coherence are available in the literature: the value usually suggested for
the decay coefficient g3 is 0.7 (Corcos, 1963; Blake, 1986; Bull, 1967). Fig. 8 shows curves of spanwise coherence relative
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
331
to the lower velocity for various nondimensional separations Z/d as a function of the similarity variable oZ/Uc. The
thick solid line represents the Corcos model with g3 equal to 0.7. The exponential function seems to correctly estimate
the measured coherences, above their maxima, for value of the nondimensional frequency oZ/Uc40.8. Also in this case,
velocity variations seem not to have an influence on the decay coefficient value.
3.2.2. The Chase model
A descriptive model of the wavenumber-frequency spectrum of turbulent wall-pressure was proposed by Chase
(1980) with the intention of overcoming the limitations of models built to capture the characteristics of the convective
domain only. Starting from the properties of the fluctuating velocity spectrum and considering its relation with the
fluctuating pressure, Chase proposed a model able to correctly describe the pressure field in the convective and subconvective domains. The inverse Fourier transform of the Chase (1980) expression is here proposed in its complete form
as determined by Josserand and Lauchle (1989) because, as it will be clear in the following, the assumptions at the basis
of some simplifications made by Chase are not necessarily fulfilled. Thus, the complete Chase model in the spacefrequency domain is given by
Fpp0 o; x; Z Fpp oC m f m e$zm e$iox=U m C t f t e$zt eiox=U t ,
"
#
2
Um
$3
2 2 1 $ zm1
f m qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi am 1 zm am mm
2iam mm zm1 ,
zm
U 2c h2m u2t
"
"
#
#
Ut
z2t3 m2t z2t1
$3
2
2
f t qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi at 1 zt at 1 mt $
2iat mt zt1 ,
zt
U 2c h2t u2t
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
am mm ox
at mt ox
am oZ
at oZ
; zt1
; zm3
; zt3
; zm z2m1 z2m3 ,
zm1
U
Ut
Um
Ut
qffiffiffiffiffiffiffiffimffiffiffiffiffiffiffiffi
zt z2t1 z2t3 ,
rm
rt
; Ct
; rm 1 $ rt ,
rt f t0 rm f m0
rt f t0 rm f m0
Um
Ut
2 2
2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
f t0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
m 1 am mm ;
t 1 at 1 mt ,
2 2
2 2
2
2
U c hm ut
U c ht ut
Cm
f m0
Um
Ut
2 2
2 4 $1
2
2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a$3
Fpp o rm a r2 u4t o$1 a$3
m qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 mm am rt a r ut o
t 1 at 1 mt ,
2 2
2 2
2
2
U c hm ut
U c ht ut
#2
$
% "
U cm
Uc
1
Ui
; a2i
; i m; t.
hi qffiffiffiffiffiffiiffiffiffiffiffiffiffi ; U i
2
2
bm do
1 $ mi
1 $ mi
ut 1 $ m2i
(2)
It can be noted that six free parameters have to be determined by comparison with experimental data. The values
suggested by Chase, based on the comparison between the experimental data of Bull (1967) and limiting values of
Eq. (2), are the following: bm 0.756, bt 0.378, mm 0.176, rt 0.389, and a+ 0.766. As pointed out by Chase,
these values are not supposed to be universal, as already shown for example by the measurements of Finnveden et al.
(2005). In fact, the attempt to fit the experimental data with the Chase model using these values for the free parameters
gave unsatisfactory results. Thus, the first five parameters were evaluated using a nonlinear least-square formulation
based on a trust-region approach. The best fit of the experimental streamwise and spanwise coherences for both velocity
conditions is found for: bm 0.51, bt 0.35, mm 0.13, mt 0.4, and rt 0.3. Fig. 9 presents the comparison among
the experimental streamwise coherence for U 5.3 m/s, the Chase model, using the identified parameters, and the
Corcos model for different values of the ratio x/d. In Fig. 10, the same comparison is shown for the spanwise coherence
relative to U 3.3 m/s. At this time, some considerations must be made: the coefficient bm gives the position of the ASD
maximum that, according to the
measurements occurs for od/utE60; on the other hand, using Chase relations,
pffipresent
ffi
the maximum is given by o 2U c =bm d, thus bm 0.51 is required if Uc 0.65U is assumed. In fact, the value 0.756
suggested by Chase represented an average between the value 0.53 needed to fit the maximum in the Bull spectrum, thus
very close to that already found, and the value 0.9 needed to fit the measured spatial correlation. Preliminary
comparisons between the experimental CSD and the simplified Chase model gave a value for the coefficient mt sensibly
higher than that suggested by Chase. Thus, it was evident that the hypothesis mm, mt51 at the basis of the simplifications
ARTICLE IN PRESS
332
made by Chase was, in this case, not valid. For this reason, the complete form as in Eq. (2) is used to perform the
analysis. It can be concluded that, except for mt and somewhat for mm, the values of the identified parameters are not so
far to those found by Chase analysing pressure experimental data acquired in completely different flow conditions.
Finally, a+ is obtained by a direct comparison with the measured ASD (see Fig. 11) when the other coefficients are
fixed. This parameter determines the amplitude of the ASD spectrum and in particular of its maximum, the best
agreement with experimental data is found for a+ 0.8. On the other hand, if the interest is not in the maximum but in
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
333
the higher frequency scale-independent region, the value of a+ should be 1.5. However, to perform the structural
analysis described in Section 4.3, the measured ASD was used.
ARTICLE IN PRESS
334
convection velocities divided by the free-stream velocity U, obtained for fixed spatial separations x/d, are plotted as a
function of the dimensionless frequency od/ut. Different figures refer, as usual, to the two different free-stream
velocities. Farabee and Casarella (1991) have observed a peak value of convection velocity for od/ut 50,
independently of the x/d values. This peak corresponds exactly to the maximum observed in the spectra and to the value
that separates the low and the high-frequency behaviour in the coherence function. This fact demonstrated that not only
the lowest wavenumber components experienced a decay, but that they are also convected at lower overall velocity.
In the present case, CSD analysis was performed without using noise cancellation technique to avoid phase alteration.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
335
On the other hand, convection velocity obtained by a division for the CSD phase is very sensitive to small disturbances
leading to unreasonable value of the convection velocity. For this reason, frequencies below 25 Hz were cancelled out
from the graph.
By inspection of Figs. 12 and 13, an increase of the convection velocity with the growing of the spatial
separation is observed; this trend of Uc is related to an increasing dominance of large-scale events to the two-point
correlation as the separation increases. For small separations, the correlation data and hence the convection velocity are
dominated by the small-scale eddies close to the wall, which move with a lower velocity than the large-scale events.
However, as far as spatial separation increases, the curves tend to collapse in a unique curve indicating that, even if the
Taylor frozen flow hypothesis does not strictly hold for the smallest spatial separations, the convection velocity can be
represented as a function of the single variable od/ut. Moreover, with increasing frequency, the convection velocities
assume a flat trend.
The dependence of the convection velocity on the spatial separation can be better highlighted from the analysis of the
space-time correlation functions of the pressure signals. The convection velocity is obtained as the ratio x/t at which
the cross-correlation Rpp(x,t) has a maximum. In Fig. 14, the convection velocities normalised with respect to the
free-stream velocity for 3.31 and 5.31 m/s are depicted as a function of the nondimensional parameter x/d*. The
experimental data are also compared with those obtained experimentally by Bull (1967) and numerically by Na and
Moin (1996). The values of Uc range from 0.6 for the smaller separations associated with the small-scale structures, to
0.73 at higher separations, related to the larger ones. From the above considerations, as proposed by several authors,
the convection velocity can be modelled with a constant average value between small- and large-scale convection
velocities, in this case, equal to 0.65 U.
where ks k2m k2n is the primary effective wavenumber component of the vibration. Hence, the numerical wet
natural frequencies were computed by the following expression:
!
"$1=2
r
mn ffi omn 1
o
,
(5)
rs hks
where omn are the numerical dry natural frequencies computed by the aforementioned FE model. The primary
wavenumber used in Eq. (5) corresponds to simply supported boundary conditions, i.e. km mp/a and kn np/b this
last being the only one analytically known. It is clear that this assumption is valid as far as the frequency increases. In
Fig. 15, the experimental and the theoretical added mass curves, mef o and mnf o, respectively, for the first 16 modes,
are displayed showing a difference of about 23% for the first mode that decreases, as expected, with increasing mode
order. Moreover, although natural modes of plates surrounded by unbounded fluids are not mathematically
orthogonal, their shapes remain almost unchanged (Fahy, 1985). This result was partially verified by the experimental
ARTICLE IN PRESS
336
analysis. In fact, a modal identification was performed using a number of points, sufficient to identify the mode order
and the position of nodes and maxima along some selected lines, but not to represent the whole mode shapes.
In conclusion, to calculate the structural response with the numerical procedure described in the following section: (i)
the dry modes provided by the FE analysis are used, (ii) the first 16 wet natural frequencies were obtained
experimentally, and (iii) the remaining ones were estimated by using the dry natural frequencies provided by the FE
analysis and Eq. (5).
The structural modal damping coefficient, Z, was evaluated from wet hammer impact tests: it decreases from 0.034 for
the first modes to a quite constant value around 0.018 for the higher modes.
4.2. Remarks on the prediction of structural response
A procedure based on the finite element approach was presented and discussed in the recent literature to solve the
response of a plate under a TBL excitation (De Rosa and Franco, 2007). Specifically, the Corcos model was used for
comparing the numerical response with the exact one, and in order to define a general methodology, able to work for
any TBL model at acceptable computational costs. For the sake of clarity, some details are herein briefly recalled.
The cited finite element procedure is assembled by using the following equation suitable for all the methods working
with discrete coordinates (Elishakoff, 1983); the CSD matrix of displacements of a structural operator represented by
using NG degrees of freedom and NM mode shapes is given by
SW o UHoSU oHo$ UT ,
(6)
SU o UT SFF oU,
(7)
with
where U is the structural modal matrix (each column is an eigenvector sampled at the NG selected points), [NG % NM]
and the generic term of H(o) is H j o o2j ' o iZo2j )'1 , [NM % NM].
The translation of the distributed random loads to the set of NG points, in other words the way of representing the
SFF, can be solved in the framework of the finite element method by using consistent approach, that is by using the
shape function vector, N, belonging to each element:
Z Z Z Z
E
k q q
k
q
k
q
SCFF
NT Fpp0 xk
(8)
1 ; x2 ; x1 ; x2 ; oNdx1 dx1 dx2 dx2 ,
k;q
xk
1
xk
2
xq
1
xq
2
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
337
where the double integers k and q indicate here two generic finite elements and the integration is related to the area of
each of them. The vector N can be interpreted as the interpolating function basis selected by the analyst according to the
specific problem and boundary conditions (Cook, 1981). Thus, Eq. (8) serves to evaluate a generic kqth member of the
NE ! NE load matrix, where NE is the number of elements. A simplified approach could refer to each grid point rather
than each finite element. This means that the load acting on the ith grid point will be the resultant of the distributed load
working on the equivalent nodal area, say Areai, belonging to it. This area vector can be evaluated easily by using a
static deterministic unit pressure load (De Rosa et al., 1994; Hambric et al., 2004). Accordingly, one gets the generic ijth
member of the NG ! NG matrix:
Z xi Dx=2 Z xj Dx=2 Z yi Dy=2 Z yj Dy=2
G
SCFF i;j
Fpp0 xi ; xj ; yi ; yj ; odyj dyi dxj dxi .
(9)
xi &Dx=2
xj &Dx=2
yi &Dy=2
yj &Dy=2
An area DxDy is assigned to both points P(xi, yi) and Q(xj, yj) and the double space integration refers to these finite
domains. A further approximation could also be introduced, considering that the wall-pressure distribution due to the
TBL in the low-frequency range does not fluctuate very quickly. In this case, the last integral could be approximated as
follows:
G
2
SLFF
i;j Fpp0 xi ; xj ; yi ; yj ; oDxDy( .
(10)
Obviously, the approximations represented by Eqs. (8)(10) are associated with decreasing computational cost.
The problem of the plate response under a convective random load, expressed in discrete form as described by
Eqs. (6) and (7), can be accurately approached only when adequately resolving both the spatial distributions of the
response function and of the forcing function; in particular, since in this case Uc5cB, the discretisation length, is ruled
by the hydrodynamic load.
In this work, Eq. (9) was used to calculate the SFF matrix because Eq. (10) was not adequate for the present
simulations. In fact, in the frequency range of interest, Eq. (9) allows the avoidance of the numerical divergence of the
structural response due to the incorrect representation of the pressure load, as approximated by Eq. (10). Some further
details of the numerical simulations are given in the next paragraph. The finite element approach was used to generate
the modal base, while the responses were calculated by a specific Fortran code.
4.3. Experimental analysis and comparisons
The ASDs of the acceleration signals, experimentally measured in eight different points over the plate, were
computed and the results averaged and compared with the results obtained by the numerical procedure exposed in the
previous section. The numerical results were obtained applying Eqs. (6), (7) and (9) for both the Corcos and Chase
models.
The FE analysis was performed using 61 ! 21 grid points corresponding to a spatial discretisation of 1 cm in both
directions. The integral in Eq. (9) was calculated using the trapezoidal rule. The hydrodynamic parameters inserted in
Eq. (9) were those identified in Sections 3.13.3; in particular, the experimental ASD was used and the convection
velocity was assumed constant over the whole frequency range and equal to 0.65U for both ship speeds. After having
performed the convergence analysis, each integration domain was finally subdivided, for both velocity conditions, in
eight intervals when the Corcos model was used and in 24 intervals when the Chase model was used instead. The Corcos
numerical solution of integral in Eq. (9) was compared and validated by an analogous analytical solution (De Rosa and
Franco, 2007). In both cases, the number of retained natural modes was 100.
The response of the plate was computed for a frequency range between 1 and 1000 Hz (the step was 4.5 Hz) for the
higher velocity and only between 1 and 600 Hz for the lower one, because above this frequency a refined mesh
ffiffiffiffiffiffiffiffipbe
pffiffiffiffiffimust
ffiffiffi
used to obtain convergence. In these frequency ranges, the ratio between the bending wavenumber kB 4 rs h=D o
and the convective wavenumber kc (see Fig. 16) varied between 0.045 and 0.21 for the lowest velocity and between 0.057
and 0.35 for the highest velocity.
Figs. 17 and 18 present the experimental and numerical averaged ASDs of the plates acceleration for 3.3 and 5.3 m/s,
respectively.
Unsurprisingly, an overestimation of the plate response is evident in the whole frequency range if the Corcos model is
used; on the other hand, the numerical response obtained using the Chase model is undoubtedly in better agreement
with experimental data. However, below 2530 Hz both pressure spectra (although cleaned) and structural response are
contaminated by the carriage vibrations transmitted to the model; thus, any comparison is meaningless. Above these
frequencies, the agreement is really satisfactory until 420 Hz for the lower velocity case and until 650 Hz for the higher
one. In the high-frequency part, the Chase model tends to slightly underestimate the experimental curve; this fact, more
ARTICLE IN PRESS
338
visible for 3.3 m/s, can be partly due to the poor spatial resolution of pressure transducers that attenuate the highfrequency part of the ASD spectrum.
Finally, an evident mismatch between the experimental and the numerical curve, generated by the presence of high
peaks in the experimental data, can be observed in Fig. 18 around 800 Hz probably due to local flow disturbances. To
better quantify the difference between model and experimental results, the previous curves are plotted in Figs. 19 and 20
in third-octave bands. It can be seen that the root mean square of the difference between the response obtained applying
the Chase model and the experimental data is 5.2 dB for the lower velocity and 4.1 for the higher one. The response
obtained by using the Corcos model to represent the surface pressure field, overpredicts in both cases the experimental
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
339
data resulting in an average difference of 18 dB. The small gap between the numerical predictions obtained by applying
the Chase model and the experimental plate response that is less or, at least, of the same order of that found by
Finnveden et al. (2005) demonstrated the validity of the developed procedure and the capability of the Chase model to
represent the surface pressure field on a ship hull. However, a more careful determination of the added water mass in
the whole frequency range can improve the numerical estimation, as well as a deep uncertainty analysis can better
indicate the confidence interval of the present results.
ARTICLE IN PRESS
340
5. Concluding remarks
In this paper, a complete analysis of the coupled structural-fluid problem concerning the response of an elastic plate
inserted in the bottom of a catamaran hull excited by the TBL WPFs, has been carried out.
The hydrodynamic analysis, performed on a rigid plate at high Reynolds number and in equilibrium flow conditions
allowed the determination of the appropriate scaling laws for the ASDs in the different frequency ranges. Moreover, the
analysis of the cross-spectral densities in longitudinal and lateral flow directions was used to fit the theoretical models,
available for the pressure representation, to the experimental data. Two theoretical models are analysed: the Corcos and
Chase models. It was demonstrated that it is possible to find for both a complete set of free parameters that provide a
fair agreement with experimental CSD data.
Since spatial resolution was too poor to analyse the CSD behaviour at high frequency and since with this type of
experimental analysis it was not possible to isolate the longest wavelengths, the studied pressure behaviour mainly
concerned the characteristics of the convective domain. Thus, an indirect comparison based on the vibrational response
of a plate was performed; in particular, the numerical structural responses obtained using the two models were
compared with experimental measurements. The conclusion of this analysis is that, although the Chase model is
complex and dependent on several empirical parameters, it provides a very good agreement with experimental data at
low wavenumbers. The performed analysis can give interesting information also for the full-scale problem; in fact,
considering realistic values of the hydrodynamic and of the structural parameters, coincidence conditions usually
appear at very low frequency both for underwater and surface marine vehicles.
The main disadvantage in using Chase model lies in its non-predictive character. In fact, it was shown that the
original Chase parameters do not fit the experimental data; thus a new set of parameters have been determined and the
complete version of the model has been used.
It is clear that the aim of any numerical procedure is to produce robust predictive tools to be used at the design stage.
Ongoing comparisons between pressure measurements performed on different ship models and for various flow
conditions in terms of Reynolds number values are aimed to analyse the range of variability of the parameters and their
dependence on the particular flow conditions.
Acknowledgement
The research was supported by the Ministero dei Trasporti in the frame of Programma Ricerche Luglio2006Dicembre 2007.
ARTICLE IN PRESS
E. Ciappi et al. / Journal of Fluids and Structures 25 (2009) 321342
341
References
Abraham, B.M., 1998. Direct measurements of the turbulent boundary layer wall pressure wavenumber-frequency spectra. ASME
Journal of Fluids Engineering 120, 2939.
Bendat, J.S., Piersol, A.G., 1991. Random Data: Analysis and Measurements Procedure. Wiley, New York.
Blake, W.K., 1970. Turbulent boundary-layer wall-pressure fluctuations on smooth and rough walls. Journal of Fluid Mechanics 44,
637660.
Blake, W.K., 1986. Mechanics of Flow Induced Sound and Vibration. Academic Press, Orlando.
Blevins, R.D., 1987. Formulas for Natural Frequency and Mode Shape. Robert E. Krieger Publishing Company, Melbourne, FL.
Bull, K.M., 1967. Wall pressure fluctuations associated with subsonic turbulent boundary layer flow. Journal of Fluid Mechanics 28,
719754.
Bull, M.K., 1996. Wall pressure fluctuations beneath turbulent boundary layers: some reflections on forty years of research. Journal of
Sound and Vibration 190, 299315.
Bull, M.K., Thomas, S.W., 1976. High frequency wall pressure fluctuations in turbulent boundary layers. Physics of Fluids 19,
597599.
Chang, P.A., Piomelli, U., Blake, W.K., 1999. Relationship between wall pressure and velocity-field sources. Physics of Fluids A 11,
34343448.
Chase, D.M., 1980. Modelling the wavevector-frequency spectrum of turbulent boundary layer wall pressure. Journal of Sound and
Vibration 70 (1), 2967.
Choi, H., Moin, P., 1990. On the space-time characteristics of wall pressure fluctuations. Physics of Fluids A 2, 14501460.
Ciappi, E., Magionesi, F., 2005. Characteristics of the turbulent boundary layer pressure spectra for high speed vessels. Journal of
Fluids and Structures 21, 321333.
Cook, R.D., 1981. Concept and Application of Finite Element Analysis. Wiley, New York.
Corcos, G.M., 1963. Resolution of pressure in turbulence. Journal of the Acoustical Society of America 35, 192199.
De Rosa, S., Franco, F., 2007. Exact and numerical responses of a plate under a turbulent boundary layer excitation. Journal of Fluids
and Structures 24, 212230.
De Rosa, S., Pezzullo, G., Lecce, L., Marulo, F., 1994. Structural acoustic calculations in the low frequency range. AIAA Journal of
Aircraft 31 (6), 13871394.
Efimtsov, B.M., 1982. Characteristics of the field of turbulent wall pressure fluctuations at large Reynolds numbers. Soviet PhysicsAcoustics 28 (4), 289292.
Elishakoff, I., 1983. Probabilistic Methods in the Theory of Structures. Wiley, New York.
Fahy, F., 1985. Sound and Structural Vibration. Academic Press, Orlando.
Farabee, T.M., Casarella, M.J., 1991. Spectral features of wall pressure fluctuations beneath turbulent boundary layers. Physics of
Fluids A 3, 24102420.
Farabee, T.M., Geib, F.E., 1991. Measurements of boundary layer pressure fluctuations at low wavenumbers on smooth and rough
walls. In: ASME Symposium on Flow Noise Modelling, Measurements and Control, NCA-vol. 11, FED-vol. 130, pp. 5568.
Ffowcs Williams, J.E., 1982. Boundary layer pressures and the Corcos model: a development to incorporate low wavenumber
constraints. Journal of Fluid Mechanics 125, 925.
Finnveden, S., Birgersson, F., Ross, U., Kremer, T., 2005. A model of wall pressure correlation for prediction of turbulence-induced
vibration. Journal of Fluids and Structures 20, 11271143.
Goody, M., 1999. An experimental investigation of pressure fluctuations in three-dimensional turbulent boundary layers. Ph.D. Thesis,
Department of Aerospace and Ocean Engineering, Virginia Tech, USA.
Graham, W.R., 1997. A comparison of models for the wavenumber-frequency spectrum of turbulent boundary layer pressures. Journal
of Sound and Vibration 206 (4), 541565.
Hambric, S.A., Hwang, Y.F., Bonness, W.K., 2004. Vibrations of plates with clamped and free edges excited by low-speed turbulent
boundary layer flow. Journal of Fluids and Structures 19, 93110.
Han, F., Bernhard, R.J., Mongeau, L.G., 1999. Prediction of flow-induced structural vibration and sound radiation using energy flow
analysis. Journal of Sound and Vibration 227 (4), 685709.
Hwang, Y.F., Maidanik, G., 1990. A wavenumber analysis of the coupling of a structural mode and flow turbulence. Journal of Sound
and Vibration 142, 135152.
Keith, W.L., Hurdis, D.A., Abraham, B.M., 1992. A comparison of turbulent boundary layer wall-pressure spectra. ASME Journal of
Fluids Engineering 114, 338347.
Josserand, M.A., Lauchle, G.C., 1989. ERRATA: D.M. Chase JSV 1980. Journal of Sound and Vibration 128, 519523.
Lee, Y.T., Blake, W.K., Farabee, T.M., 2005a. Modeling of wall pressure fluctuations based on time mean flow field. ASME Journal of
Fluids Engineering 127, 233240.
Lee, Y.T., Miller, R., Gorski, J., Farabee, T., 2005b. Predictions of hull pressure fluctuations for a ship model. In: Proceedings of
International Conference on Marine Research and Transportation (ICMRT05), Ischia, Italy.
Manoha, E., 1996. The wavenumber-frequency spectrum of the wall pressure fluctuations beneath a turbulent boundary layer. In:
Proceedings of the AIAA Aeroacoustics Conference, AIAA paper 96-1758.
Na, Y., Moin, P., 1996. Direct numerical simulation of turbulent boundary layers with adverse pressure gradients and separation.
Report No. TF-68. Thermosciences Division, Department of Mechanical Engineering, Stanford University.
ARTICLE IN PRESS
342
Panton, R.L., Robert, G., 1994. The wavenumber-phase velocity representation for the turbulent wall-pressure spectrum. ASME
Journal of Fluids Engineering 116, 477483.
Smolyakov, A.V., Tkachenko, V.M., 1991. Model of pseudosonic turbulent wall pressures and experimental data. Soviet PhysicsAcoustics 37 (6), 627631.
Willmarth, W.W., Wooldridge, C.E., 1962. Measurements of the fluctuating pressure at the wall beneath a thick turbulent boundary
layer. Journal of Fluid Mechanics 14, 187210.
Nomenclature
Roman Symbols
a
AQ j Q k
b
cB
cL
D
E
f
h
Hi
i
kc
Lj
Lx
Ly
p
PIN
r
R
Rey
Sp
Sw
Sw,
U
mean
Uc
ut
v
w
x
Xpp
y
91
convective velocity
friction velocity
cinematic viscosity
out of plane displacement of the plate
streamwise reference axis
cross spectral density of the wall pressure distribution due to the turbulent boundary layer
crosswise reference axis
Greek symbols
ax
ay
G
gj
d
d*
Z
u
n
xx
xy
l
rf
rs
F
ci
o
oj
Ciappi et al., 2009; Efimtsov, 1982) for the cross spectral density (CSD) of wall pressure fluctuations (WPF). These
quantities represent the input for the structural analysis. The identification of the scaling laws permits to make pressure
spectra independent of the test conditions (i.e. of Reynolds number) and extrapolate data at full scale from low Reynolds
number laboratory experiments. The CSD models are supposed to be valid only under the hypotheses of two-dimensional
fully developed turbulent boundary layer with zero pressure gradient acting on a flat plate and contain empirical
parameters usually obtained from experimental data. Unfortunately, these parameters are not always universal but
necessitate dedicated experimental campaigns to be defined. A comparison between the most popular CSD models can be
found in Graham (1997) and Hambric et al. (2004).
Besides being time demanding and expensive, a model scale experiment for this kind of measurements would only
provide a partial information about the physics of the problem. For example, standard experimental set-up and data
analyses for WPF characterization, that make use of a spatial domain comparison between measured CSD spectra and
theoretical models, are inadequate to indicate which one is the best for the description of WPF in the different
wavenumber regions. In fact, most of the energy of WPF is concentrated around the convective peak and any correlation
data are mainly the representation of the convective character of the TBL. Only direct measurements of the wavenumberfrequency spectra could give significant insight, but only few experimental data have been made by using this approach
(Abraham, 1998), due to the complexity of the experimental set-up and of the data analysis.
For these reasons, the definition of a predictive model able to represent correctly the spatial variation of the wall
pressure field, in all the frequency and wave-number ranges, is still an open issue.
On the other hand, the numerical solution of the structural equations, especially when dealing with complex structures
such as an aircraft fuselage or a ship hull, deserves some attention. When the structural wavelengths are small compared
to the typical dimension of the problem, i.e. at high frequency, the number of degrees of freedom, necessary to calculate
accurately the structural response, increases rapidly. In this frame, energy methods such as the Statistical Energy Analysis
(Lyon and De Jong, 1995) can be invoked at high excitation frequencies. However, the definition of the input power,
starting from a general model for the pressure cross spectral density not using the separation of variables, can be very
92
complicated and time consuming. Interesting advances are in Totaro et al. (2008) and in Totaro (2004), and those findings
are here fruitfully used; other research works, which have addressed the problem of the computational cost versus the
frequency bandwidth of interest (De Rosa and Franco, 2008; Hong and Shin, 2010; Ichchou et al. 2009), have to cited too.
A chance to drastically reduce the computational time can be the identification of scaling laws for the structural
response able to determine, at least for a certain class of problems, a unique representation of the response independent of
the particular flow conditions or structural properties.
In this work a dimensional analysis is used to recover the dimensionless parameters for the definition of the scaling
laws for the required axes: the excitation frequency and the power spectral density of the response. The structure is a thin,
flat and elastic plate with no pre-stresses wetted over one face by a stationary turbulent boundary layer in an
incompressible flow with zero pressure gradient.
The obtained dimensionless groups are used to report and compare on dimensionless scales three analytical responses
obtained with the modal approach and, more important, four data sets measured in different conditions both in wind
tunnels and in a towing tank. The functional relations between the physical parameters are discussed and an analytical
expression for the dimensionless plate response is also derived. It is shown that the functional group in Totaro et al. (2008)
is here found by invoking a dimensional analysis. It includes the main dimensional variables for the plate response and the
fluid velocity, as expected. The collapse of the different experimental plate responses, as consequence of the choice of the
dimensionless functional groups, is very encouraging. The proposed scaling laws are successively and successfully applied
to four data sets measured in different conditions both in wind tunnels and in a towing tank.
The work is structured as follows. After this introduction, Section 2 summarizes all the theoretical backgrounds for the
evaluation of the modal and energy response of a plate under a convective and turbulent pressure load; further, it presents
the dimensional analysis and the derivation of the dimensionless parameters governing the problem. In Section 3, all the
experimental set-up and data are presented, together with the resulting dimensionless forms of the structural responses.
Finally, Section 4 remarks the main achievements of the present work.
#
"
# "
1 X
1
X
cj xA ,yA ck xB ,yB
Sp oab2
AQ j Q k o,
U
gj gk
Lnj oLk o
j1k1
with
AQ j Q k o
gj
a
0
b
0
b
0
"
X pp x,y,x0 ,y0 , o
Sp oab2
The symbol ci denotes the ith mode shape and oi the ith natural radian frequency. The integrals defined by the symbol
AQjQk are well known also as the acceptances: joint acceptance for j k, or cross acceptance for jak. The formulation
contained in the Eqs. (1) and (2) can be applied to any structural operator once assumed its modal base.
From the analysis of Eq. (2), it is evident that the quality of the predictions is strictly related to the spatial
characterization of wall pressure fluctuations expressed in terms of its cross-spectral density function X pp .
93
Both terms of the right-hand side of Eq. (4) are in general complex functions of flow parameters. In particular, the single
point spectral density depends on the boundary layer characteristic lengths d, d*, on the characteristic velocities U and ut
and on the flow properties namely rf and v. According to the scaling laws provided by several authors (Ciappi et al., 2009;
Keith et al., 1992) and on the corresponding analytical formulation (Goody, 2004), the role of d and d* is equivalent thus,
only d is retained for the dimensional analysis.
The semi-empirical models available in the scientific literature to represent the function G suggest different functional
dependences on flow parameters. The most popular of them is the Corcos (1964) model, which considers the dependence
on the convection velocity Uc, only, whereas more sophisticated models (Chase, 1980; Efimtsov, 1982) include also the
dependence on d and ut. Thus, Eq. (4) can be rewritten in the more general form as
X pp Sp o, d,U,ut , rf , nGxx , xy , o, d,U c ,ut :
In view of a dimensional analysis related to the evaluation of the structural response, it is convenient to consider
directly as one of the dimensional parameter the power spectral density of wall pressure fluctuations. In this case the
dependence on rf and v, appearing only in the first term of Eq. (6), is not considered explicitly. The same considerations
hold for U.
Under the above assumptions, the plate response to the pressure field induced by a turbulent boundary layer, according
to Eq. (1), can be represented as a generic function f depending on both the following dimensional fluid dynamic and
structural variables:
f Sw ,Sp , o, d,U c ,ut , rs ,E, Z,h,a,b, xx , xy 0,
where rs is the material density, E the Young modulus, Z is the total damping coefficient (sum of the material and of the
aero/fluid dynamic damping) and h, a, b are the plate thickness, length and width, respectively. In Eq. (6) there are 14
dimensional parameters thus, according to the Buckingham theorem (Buckingham, 1914), there are 11 dimensionless
parameters governing the problem. The identification of these last is not unique but one set is given by
sffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffi
Sw U c
Uc 3 h
h oh ut a b d xx xy
;
r
;
;
; ; ; ; ; ; Z:
7
;E
s
3
Sp
Sp U c U c U c h h h h h
h
Then, the power spectral density of the plate displacement can be rewritten in the following form:
0 sffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffi
1
3
h @
U 3c h
h oh ut a b d xx xy A
g rs
,
, , , , , ,Z :
,
,E
Sw
Sp
Sp U c U c U c h h h h h
Uc
on
oh
Uc
Eq. (9) can be also expressed by introducing the flow convective wavenumber kc o=U c , the structural bending
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffi 4
2
wavenumber kB o 12rS 1&n2 =Eh , the flexural and longitudinal structural wave speed cB o=kB and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cL E=rs 1&n2 , respectively
on p
kc cB
,
kB cL
10
thus highlighting that the dimensionless frequency relates the convective and the structural wavenumbers.
At this stage a critical analysis of the identified parameters is fundamental to recognize which of them are the most
significant for the plate response.
Dimensionless ratios a/h and b/h are strictly related to the structural model used to describe the plate motion. In the
present analysis only thin plates (i.e. governed by the Kirchoff equation) (Graff, 1991) are considered and then the value of
these two parameters can be assumed definitely large.
The dimensionless parameter d=h indicates the fluid-structure degree of coupling for conventional plates; it is excluded
the analysis of the cases of thick and flexible plates. More specifically, it can be interpreted as a measure of the influence of
the structural deformation on the flow field. Consistent numerical values of d and h for the present test cases indicate that
94
d=hb 1 moreover, since the order of magnitude of the displacement response is of the same order of magnitude than the
plate thickness, it does not affect the fluid domain, indicating a one-way coupling between the structure and the fluid.
Furthermore, if it is assumed that the major contribution to the plate response is due to diagonal terms of the crossspectral density matrix, xx/h and xy/h can be neglected too.
Finally, although the ratio ut =U c for an equilibrium boundary layer over a flat plate is a function of the Reynolds
number (Schlichting, 1978), its variation is small and can be considered of minor importance.
Damping coefficient can be introduced by defining a complex longitudinal wave speed cnL 2 pcL 2 1&Z2 =4, (Cremer
and Heckl, 1987); however, for the considered plates damping coefficient values are approximately the same thus, in what
follows Z is neglected.
The only two dimensionless parameters that seem important for the present problem are those involving the pressure
power spectral density.
Without any further considerations on the physics of the problem, the plate response can be dependent on one of them
or on a combination of the two. Nevertheless, in order to make the plate response independent of the input it seems
convenient to consider the ratio Sw/Sp as done already in Finnveden et al. (2005).
Thus, it is possible to define three possible functional dependences for the spectral density functions related to the plate
displacement, by evidencing also a generic function, g, of the dimensionless frequency, o*
sffiffiffiffiffiffiffiffiffiffi!&2
3
h
h
Sw
E
gon ,
11
Sp U c
Uc
0
sffiffiffiffiffiffiffiffiffiffisffiffiffiffiffiffiffiffi 1&1
3
h @
h
U 3c hA
Sw
rs E
gon ,
Sp U c
Sp
Uc
12
0 sffiffiffiffiffiffiffiffi 1&2
3
h @
U 3c hA
Sw
rs
gon :
Sp
Uc
13
" #
" #2
Sw U c 2
S
U
rs E w c r2s c2L g II on ,
Sp h
Sp h
15
" # " #
Sw $rs %2 4 Sw E 2 U c 4
Uc
g III on :
Sp h
Sp h
cL
16
Accordingly, it is possible to define three dimensionless displacement functions of the dimensionless frequency o* only
" #
Sw E 2 Sw $rs %2 4
cL g I on ,
14
Sp h
Sp h
The capability of Eqs.(14)(16) to produce a data collapse and the functional dependence on the dimensionless
frequency is investigated in Section 3.2 from the direct analysis of the experimental data presented in Section 3.1.
2.3. Energy response
The analysis of an energy model for the response of the plate can be difficult since a correct representation of the power
input is needed. This aim is addressed in the work by Totaro (2004) and Totaro et al. (2008), using, as fundamental relationship,
the modal response expressed in Eq. (1). Specifically, the main TBL models, due to Corcos, Corcos with Davies approximation
(Davies, 1971), Efimstov and Blake, are investigated. In Totaro et al. (2008), the following representation is adopted:
P IN o Zo3 abrS hSw,mean ;
P IN opSp o
ab U 2c
2rS h o2
rffiffiffiffiffiffiffiffiffi
rh
UFa,b,U c , o, rS ,cL :
D
17
18
The dimensionless function F represents the average value of the joint acceptance integral in wavenumber space; this
term depends on the selected model for the TBL excitation. PIN and Sw,mean denote the input power and the mean response.
It is straightforward to get that
"
! #&1 " #
2
Sw,mean o
h
Uc 4
1
p
UFa,b,U c , o, r2S ,cL ;
Z
3
Sp o
oh
r2S cL U c
!
19
Sw,mean o Zr2S cL U 3c
n
o
:
g
IV
2
Sp o
h
95
Fig. 1. Analytical responses for three different plates excited by a Corcos TBL source with Davies approximation.
Table 1
Parameters for the configuration used in the analytical responses.
Material
Density (kg/m3)
Youngs modulus
Plate thickness (mm)
Plate length (m)
Plate width (m)
Fluid density (kg/m3)
Speed of sound (m/s)
Asymptotic speed (m/s)
Damping loss factor
Conf 1
Conf 2
Conf 3
Aluminum
2700
7.1E 10
0.5
0.4
0.3
1000
1000
5
0.02
Steel
7500
21.1E 10
2.0
0.7
0.2
1.225
340
100
0.04
PVC
1400
4.5E 9
1.0
0.372
0.22
1.225
340
72
0.05
Therefore, a full dimensionless response is again obtained and it evidences the same group highlighted in the previous
section with a classical dimensional analysis, Eqs. (14)(16). According to the present result, the dependence is with the
fourth power of the reciprocal of the dimensionless frequency. All the remaining frequency dependence is left in the F
function in which they can be attributed specifically to the difference among the TBL models.
2.4. Results
The approximate solution of the Eq. (1), when the TBL source model is due to Corcos with the Davies approximation, is
in Fig. 1; the main parameters for the tested plates are summarized in Table 1. It is evident how the functional groups
associated in with the gIV function works well in unifying the responses.
It has been also performed a full analysis of the problem by invoking a scheme in which the joint acceptance is solved
analytically, Eq. (2); the results are in Fig. 2. The main difference among these analytical results is in the low frequency
range, which is dominated by the first resonating structural modes.
Rather than discussing all the possible effects of the adoption of any of the g formulations, moreover with reference to
all possible TBL model, it has been decided to apply directly the derived g functions on the experimental data.
3. Experimental measurements
3.1. Description of the set-up
The four data sets, considered for this analysis, regard incompressible and steady turbulent boundary layers in almost
zero pressure gradient flow acting on thin and flat plates.
The boundary conditions of the plates were neither measured nor estimated since this information is relevant only for
the tuning of the predictive modal models: this falls outside the present aims.
96
Fig. 2. Analytical responses for three different plates excited by a Corcos TBL source.
Among the large amount of data available in the literature on wall pressure and induced structural vibrations, these
experimental setups are the few ones able to provide, for each test case, information on both pressure fluctuations and
structural deformations acquired in the same facility and in the same nominal conditions. As it is specified in the next
section this aspect is fundamental for the validation of the proposed scaling procedure.
The first two sets of data are extracted from a database containing measurements of wall pressure fluctuations and
structural response acquired in the INSEAN towing tank. The first one is relative to an experimental campaign performed
on a 1:15 model of a catamaran hull (Fig. 3). A thin PVC plate is inserted in the bottom of the hull in correspondence of the
stern region; pressure and acceleration measurements are performed for model speeds of 3.3 m/s and 5.3 m/s,
respectively. A complete description of this experimental campaign can be found in Ciappi et al. (2009).
The second set of data belongs to an experimental setup designed to measure wall pressure fluctuations and the
response of elastic portions of a 1:8 scaled bulbous model of a surface ship (Fig. 4). The considered data regard the stern
measuring section where the flow reaches quasi stationary conditions and where pressure gradient effects, due both to
water surface deformation and to structural curvature, can be neglected, (Magionesi and Ciappi, 2010a, 2010b). The elastic
element inserted in the model is an PVC thin plate, the model velocity in this case ranges between 3.636 m/s and 6.36 m/s.
A complete description of this experimental campaign can be found in Magionesi and Ciappi (2010a, 2010b).
The last two sets of data are obtained from measurements performed in aerodynamic tunnels. The former is due to
Finnveden et al. (2005) in the frame of ENABLE project: it consists of an aluminum plate exposed to flow velocities of 80,
100 and 120 m/s, respectively.
The latter is part of the experimental campaign performed by Totaro et al. (2008) on four different plates of different
geometries and materials. The data considered for this analysis regard the PVC plate for flow velocities equal to 35 and
50 m/s, respectively. Table 2 presents the principal characteristics of the four plates. Table 3 lists the principal mean flow
TBL parameters of the 10 experimental test cases.
The chosen four data sets have the aim to analyze complete different flow conditions and structural configurations.
The power spectral densities of the plate responses are represented by their mean response over 8 points for plate 1, 3
points for plate 2 and 5 points for plate 3. The reference power spectral densities of wall pressure fluctuations are
represented by their average over 10 points for plates 1 and 2 and 19 points for plate 3. Data for plate 4 are directly derived
from Fig. 18 of the paper by Totaro et al. (2008) in terms of metric response R defined as
R
o4 Sw rs h2
Sp
20
In the same work it is stated that the plate velocity response acquired with a laser vibrometer is averaged over 75 points.
Values for the convection velocity are usually obtained from time domain cross-correlation analyses or from the phase
of the cross spectral densities. The values of Uc used for the present analysis come from the results given in Ciappi et al.
(2009), Finnveden et al. (2005), Magionesi and Ciappi (2010a, 2010b) and Totaro et al. (2008) and are listed in Table 3.
It is important to remark that Eq. (20) represents the first attempt to make the power spectral density of the plate
response dimensionless and independent of the power spectral density of WPF as done in Finnveden et al. (2005), De Rosa
and Franco (2008) and Ciappi et al. (2009). However, the spatial characteristics of the fluidstructure interaction are not
taken into account neither in the response axis nor in the frequency one. In the next section, it is shown that this
representation does not lead to a collapse of the spectra. In fact, for the selected test cases, the fluidstructure coupling is
97
Test section
Table 2
Dimensions of the plates and material properties.
Material
Density (kg/m3)
Youngs modulus
Thickness (mm)
Length (m)
Width (m)
Plate 1
Plate 2
Plate 3
Plate 4
PVC
1190
3.2E 9
3.0
0.6
0.2
PVC
1190
3.2E9
3.3
0.242
0.144
Aluminum
2700
7.1E 10
1.6
0.768
0.328
PVC
1400
4.5E 9
1.0
0.6
0.3
98
of different kind in the frequency ranges of interest (sub-convective, convective and super-convective bandwidths).
Furthermore, these frequency ranges can be completely different among all possible experimental configurations. As a
consequence, the only normalization of the frequency axis is not sufficient, too.
Table 3
TBL mean flow parameters.
Plate
Fluid
1 and 2
Water
Air
d (mm)
ut (m/s)
Uc (m/s)
3.30
5.30
3.64
5.45
6.36
120.0
113.0
51.0
49.7
48.0
0.110
0.163
0.102
0.147
0.171
0.70 U
80.00
100.00
120.00
35.00
50.00
50.0
2.600
3.100
3.700
1.4
1.960
0.75 U
U (m/s)
53.0
55.0
85.0
0.65 U
0.62 U
99
between different curves of the same data set either. As already said this dimensionless form for the structural response
does not contain any information about the spatial characteristics of the fluidstructure interaction.
In Fig. 7 the same experimental curves are represented in terms of the ratio between the power spectral densities of the
plate displacement and of the pressure. The considerations done for the contents of Fig. 5 hold in this case as well.
In order to investigate the validity of the dimensionless functional relations provided by the dimensional analysis
performed in Section 2.2, the plate spectra are scaled according to Eqs. (14)(16) and the corresponding results are
reported in Figs. 810, respectively.
The first non-dimensional quantity, defined by Eq. (14), is represented in Fig. 8 on the y-axis to check its attitude to
provide a universal scaling of the four data sets. The results clearly indicate an excellent collapse of the three curves
relative to Plate 3; moreover, the curves relative to Plate 4 show the same trend in a quite similar non-dimensional
frequency range and a complete superposition with the previous ones. Curves relative to Plate 1 and 2 exhibit a good
collapse with the other ones mostly for the higher values of the dimensionless frequency well above the convective range.
On the other hand, from the inspection of Figs. 9 and 10, it appears that Eqs. (15) and (16) do not lead to a collapse of
the experimental data. The main difference between Eq. (14) and Eqs. (15) and (16) is the explicit presence of the flow
velocity as one of the scaling parameters for the response axis (it always appears in defining, together with the plate
thickness, the reference time scale of the dimensionless frequency axis). In particular, focusing on the curves relative to the
100
same plate, it is evident that the introduction of Uc to scale Sw =Sp seems to prevent the data to collapse. Additionally, it is
worthwhile to highlight that, passing from Eqs. (15) to (16), that is, with a higher power of Uc in the dimensionless
functional representation, the gap between the different curves increases considerably, confirming that the proposed
dependence on the convective speed, Uc, is incorrect.
With reference to Fig. 8, it is then possible to find the functional relation between the dimensionless response and the
dimensionless frequency. Three distinct regions can be identified. Different slopes of the linear regression curves
characterize each of them (in logarithmic scale); their expressions are well approximated by the simple relations
8
>
>
4108 on &4
>
>
>
>
>
<
8
&8
Snw 110 on
>
>
>
>
>
>
7
&3
>
: 210 on
oh
Uc
0:07 o
oh
Uc
o0:07;
oh
Uc
o0:21;
21
4 0:21;
and they are drawn for comparison in Fig. 8, along with the experimental measurements.
Although these frequency regions cannot be recast exactly in terms of the ratio between the bending and the convective
wavenumbers, some general considerations can be drawn. In particular, for low values of o*, the first relations of Eq. (21)
101
represent the amplitude of the response for frequencies that are around the coincidence frequency i.e., for values of
kB =kc ( 1. The second and the third relations represent the amplitude of plate response in correspondence of frequencies
for which the ratio kB =kc ( 1.
It is straightforward now to introduce what already presented in Fig. 2, derived with reference to Eq. (19). This
relationship is based on the balance between the input power and the energy content of the structural domain. In fact, it is
the only dimensionless form directly involving the structural damping.
Eq. (19) is shown in Fig. 11 when applied to the four sets of experimental data set.
The experimental results, in this dimensionless form, are perfectly in agreement with those analytical presented in
Fig. 2 and show a good collapse. It is worthwhile to recall that the analytical results are based on the Corcos TBL model and
on structural and flow characteristics different from the experimental ones. The dimensionless curves of the experimental
data separate in the low frequency bandwidth as already shown in the analytical results.
It is possible to check that, in logarithmic scale, the experimental measurements collapse on a simple curve
Snw 3:310&1 on &4:8
oh
Uc
40:04:
22
102
Some remarks are needed about this regression: (i) the power of the dimensionless frequency is not equal to four
because the g function includes the average value of the joint acceptance integral, which is a function of frequency, too; (ii)
the law has been obtained by using the analytical responses presented in Fig. 2.
At this time, some considerations on the usefulness and practical relevance of dimensionless relations can be done. As
already underlined in this work, scaling laws for the power spectral density of WPF exist and are well established.
Analytical expressions for pressure PSD make use of some basic TBL mean flow parameters such as the boundary layer
thickness and the friction velocity. Thus, starting from the desired test conditions in terms of material properties,
geometrical configurations and undisturbed flow velocity and, assuming that the convection velocity can be represented as
a constant fraction of U, it is possible, from the knowledge of d and ut only, to give a quick estimate of the amplitude of the
structural response spectra in the whole frequency range.
It is not worthwhile to recall that Eq. (21) are obtained by invoking a dimensional analysis; on the contrary, Eq. (19)
comes from an energy-based formulation. For all the results presented herein, the structural response has been considered
a mean one, obtained through an average over the acquired locations.
4. Concluding remarks
In this work the scaling laws for the response of an elastic thin plate excited by a steady turbulent boundary layer are
derived. These laws are based on both a pure dimensional analysis and an energy response formulation. Simple analytical
expressions for the dimensionless curves are provided and more important, the proposed scaling laws are used for
representing a quite large amount of experimental data.
These experimental data set involves four types of plates in air and in water flow, excited by a turbulent boundary layer,
thus representing a severe test data set.
With this data, it is shown, that all the presented experimental measurements collapse very close one to each other
according with the dimensionless functions selected for the power spectral densities of the structural displacements. Then,
the proposed representations could be usefully utilized to perform preliminary predictive steps. In fact, it has been
demonstrated that it is possible to obtain an estimate of the power spectral densities of the displacement response in the
whole frequency range, starting from the desired test conditions i.e. material properties, geometrical characteristics and
flow velocities, from the knowledge of the boundary layer thickness and of the friction velocity necessary to define the
pressure power spectral density.
In this regard, it has been shown that both the relations obtained using dimensional analysis (Eq. (14)) as well as energy
response formulation (Eq. (19)) produce a quite good collapse of all data sets. On the other hand, Eq. (19) describes the
dependence between non-dimensional frequency and non-dimensional acceleration response with one simple expression
valid in the whole frequency range providing a very quick estimate of the structural response. Notwithstanding, the use of
Eq. (14) should be preferred for all cases where damping value is not easy to be identified for example when the
considered panel is a section of a real and complex structure (ship, airplane, etc.).
At this stage, the main aim has been to test the possibility to get a universal representation of the response able to
include the fluid-structure coupling in all frequency and wavenumber ranges. It is evident that many other points remain
to be investigated in order to widen the significance of the representation and thus to extend the proposed dimensionless
forms to a more complicated structural components.
References
Abraham, B.M., 1998. Direct measurements of the turbulent boundary layer wall pressure wavenumber-frequency spectra. Journal of Fluids Engineering
120, 2939.
Buckingham, E., 1914. On physically similar systems; illustrations of the use of dimensional equations. Physical Review 4, 345376.
Bull, M.K., 1996. Wall pressure fluctuations beneath turbulent boundary layers: some reflections on forty years of research. Journal of Sound and
Vibration 190, 299315.
Chase, D.M., 1980. Modelling the wavevector-frequency spectrum of turbulent boundary layer wall pressure. Journal of Sound and Vibration 70, 2967.
Choi, H., Moin, P., 1990. On the space-time characteristics of wall pressure fluctuations. Physics of Fluids A 2, 14501460.
Ciappi, E., Magionesi, F., De Rosa, S., Franco, F., 2009. Hydrodynamic and hydroelastic analyses of a plate excited by the turbulent boundary layer. Journal
of Fluids and Structures 25, 321342.
Corcos, G.M., 1964. The structure of the turbulent pressure field in boundary-layer flows. Journal of Fluid Mechanics 18, 353379.
Cremer, L., Heckl, M., 1987. Structure borne sound, 2nd Ed. Springer-Verlag, ISBN: 0-387-18241-1.
Davies, H.G., 1971. Sound from turbulent-boundary-layer-excited panels. Journal of the Acoustical Society of America 49, 878889.
De Rosa, S., Franco, F., 2008. Exact and numerical responses of a plate under a turbulent boundary layer excitation. Journal of Fluids and Structures 24,
212230.
Elishakoff, I., 1983. Probabilistic Methods in the Theory of Structures. John Wiley & Sons, New York.
Efimtsov, B.M., 1982. Characteristics of the field of turbulent wall pressure fluctuations at large Reynolds numbers. Soviet PhisicsAcoustics 28, 289292.
Finnveden, S., Birgersson, F., Ross, U., Kremer, T., 2005. A model for wall pressure correlation for prediction of turbulence-induced vibration. Journal of
Fluids and Structures 20, 11271143.
Goody, M., 2004. Empirical spectral model of surface pressure fluctuations. AIAA Journal 42, 17881794.
Graff, K.F., 1991. Wave motion in elastic solids. Dover Publications, New York.
Graham, W.R., 1997. A comparison of models for the wavenumber frequency spectrum of turbulent boundary layer pressures. Journal of Sound and
Vibration 206, 541565.
Hambric, S.A., Hwang, Y.F., Bonness, W.K., 2004. Vibrations of plates with clamped and free edges excited by low-speed turbulent boundary layer flow.
Journal of Fluids and Structures 19, 93110.
103
Hong, C., Shin, K.K., 2010. Modelling of wall pressure fluctuations for finite element structural analysis. Journal of Sound and Vibration 329, 16731685.
Ichchou, M.N., Hiverniau, B., Troclet, B., 2009. Equivalent rain on the roof loads for random spatially correlated excitations in the mid frequency range.
Journal of Sound and Vibration 322, 926940.
Keith, W.L., Hurdis, D.A., Abraham, B.M., 1992. A comparison of turbulent boundary layer wall-pressure spectra. Journal of Fluids Engineering 114,
338347.
Lee, Y.T., Blake, W.K., Farabee, T.M., 2005. Modeling of wall pressure fluctuations based on time mean flow field. Journal of Fluids Engineering 127,
233240.
Lyon, R.H., De Jong, R.G., 1995. Theory and Application of Statistical Energy Analysis. Butterworth-Heinemann, London.
Magionesi, F., Ciappi, E., 2010a. Characterisation of the response of a curved elastic shell to turbulent boundary layer. In: Proceedings of the Seventh
International Symposium on FluidStructure Interactions, FlowSound Interactions, and Flow-Induced Vibration & Noise, Montreal, Canada.
Magionesi, F., Ciappi, E., 2010b. Experimental investigation of turbulent boundary layer excitation acting on the sonar dome. In: Proceedings Internoise
2010, Lisbon, Portugal.
Peltier, L.J., Hambric, S.A., 2007. Estimating turbulent-boundary-layer wall-pressure spectra from CFD RANS solutions. Journal of Fluids and Structures 23,
920937.
Schlichting, H., 1978. Boundary Layer Theory. McGraw-Hill, New York.
Totaro, N., Robert, G., Guyader, J.L., 2008. Frequency averaged injected power under boundary layer excitation: an experimental validation. Acta Acustica
94, 534547.
Totaro, N., 2004. Caracterisation de Sources Aerodynamiques et Sous-structuration Pour la Method SEA. Ph.D. Thesis, INSA, Lyon, France, 04 ISAL 010.