POems (INRIA) Patrick Joly Patrick Joly (INRIA) joint work with Sébastien Impériale (CEA-INRIA) Mathematical and numerical modeling of piezoelectric transducers e EPI POEMS (UMR CNRS-ENSTA-INRIA) Collaboration CEA (LIST)-INRIA Conference NUMPDE - Heraklion - September 2011
39
Embed
Mathematical and numerical modeling of piezoelectric ... · Mathematical and numerical modeling of piezoelectric transducers e EPI POEMS (UMR CNRS-ENSTA-INRIA) Collaboration CEA (LIST)-INRIA
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
POems (INRIA) Patrick Joly
Patrick Joly (INRIA)
joint work with Sébastien Impériale (CEA-INRIA)
Mathematical and numerical modeling of piezoelectric transducers
eEPI POEMS (UMR CNRS-ENSTA-INRIA)
Collaboration CEA (LIST)-INRIA
Conference NUMPDE - Heraklion - September 2011
Motivation: Non Destructive Testing
2
We want to model the emission and the reception of the signal by the piezoelectric transducer/sensor.
multiscale phenomenon, very hard to handle numerically
ρr∂
2
∂t2u−Div Cre(u) = −Div
d√C0ε0
E + f
µr∂
∂tH +
1√
µ0ε0
ρ0
C0∇× E = 0
εr∂
∂tE − 1
√µ0ε0
ρ0
C0∇×H = − ∂
∂t
dT
√C0ε0
e(u)
POems (INRIA) Patrick Joly
ρr∂
2
∂t2u
δ −Div Cre(uδ) = −Div drEδ + f
µr∂
∂tH
δ +1δ∇× E
δ = 0
εr∂
∂tE
δ − 1δ∇×H
δ = − ∂
∂tdT
r e(uδ)
We are going to use an approximate model for smallknown as the quasi-static model
that can be justified rigorously by a limit process
δ
δ → 0
POems (INRIA) 10
The quasi static approach
Patrick Joly
|∂tB| << |∇× E|
−∂tB = ∇× E
The quasi static approach
Fundamentals of Piezoelectricity.
Takuro Ikeda 1984 !!!
POems (INRIA) Patrick Joly
coupled elliptic - hyperbolic system
ρr∂2
∂t2u−Div Cr e(u) = Div dr ∇V + f in ΩS
∇ · εr∇V = ∇ · dTr e(u) in R3
ρr∂
2
∂t2u
δ −Div Cre(uδ) = −Div drEδ + f
µr∂
∂tH
δ +1δ∇× E
δ = 0
εr∂
∂tE
δ − 1δ∇×H
δ = − ∂
∂tdT
r e(uδ)
12
The quasi static approach
∇× E = 0 =⇒ E = −∇V
∇×H = 0 H = 0=⇒
At the limit , we expect thatδ → 0
div(µrH) = 0
The quasi-static model
We take the divergence
electric potential
and integrate in time
POems (INRIA) Patrick Joly
div
ρr∂
2
∂t2u
δ −Div Cre(uδ) = −Div drEδ + f
µr∂
∂tH
δ +1δ∇× E
δ = 0
εr∂
∂tE
δ − 1δ∇×H
δ = − ∂
∂tdT
r e(uδ)
ρ∂2
∂t2u−Div C e(u) = Div d∇V + f in ΩS
∇ · ε∇V = ∇ · dT e(u) in R3
13
The quasi static approach
E = −∇V H = 0
Eδ =
ΩS
ρr|∂tuδ|2 +Cr e(u
δ) : e(uδ)+
R3
εr|Eδ|2 + µr|Hδ|2
E =
ΩS
ρr|∂tu|2 +Cr e(u) : e(u)
+
R3
εr |∇V |2
POems (INRIA) Patrick Joly
||uδ − u||L∞
0,T ;H1(ΩS)
+ ||Eδ − E||L∞
0,T ;L2(ΩS)
+ ||Hδ||L∞
0,T ;L2(ΩS)
≤ C T δ
Theorem : Provided that , with f ∈ C3(0, T, L2(ΩS)) E = −∇V
ΩP
14
The quasi static approach
Ingredients of the proof
div(εrAδ) = 0Eδ = −∇V δ +AδHemholtz decomposition with
Energy estimates + Gronwall’s lemma
POems (INRIA) Patrick Joly
||uδ − u||L∞
0,T ;H1(ΩS)
+ ||Eδ − E||L∞
0,T ;L2(ΩS)
+ ||Hδ||L∞
0,T ;L2(ΩS)
≤ C T δ
Theorem : Provided that , with f ∈ C3(0, T, L2(ΩS)) E = −∇V
(uδ − u, V δ − V )
with as a source termAδ(uδ − u, V δ − V ) solve quasi-static equations with support inside the piezoelectric domain
≤ C AδL2(ΩP )
Using + stability estimates∇×Aδ ≡ ∇× E
δ = µr δ ∂tHδ
≤ C δα AδL2(ΩP ) ≤
15
The quasi static approach
Ingredients of the proof
Generalized Hardy-Poincaré-Friedrichs inequality
Aδ
1 + |x|
L2(R3)
≤ C∇ ×AδL2(R3) + div(εr Aδ)L2(R3)
POems (INRIA) Patrick Joly
||uδ − u||L∞
0,T ;H1(ΩS)
+ ||Eδ − E||L∞
0,T ;L2(ΩS)
+ ||Hδ||L∞
0,T ;L2(ΩS)
≤ C T δ
Theorem : Provided that , with f ∈ C3(0, T, L2(ΩS)) E = −∇V
div(εrAδ) = 0Eδ = −∇V δ +AδHemholtz decomposition with
∇ ×AδL2(R3) ≤ C δ
16
A second difficulty
V
POems (INRIA) Patrick Joly
ρ∂2
∂t2u−Div C e(u) = Div d∇V + f in ΩS
∇ · ε∇V = ∇ · dT e(u) in R3
d(x) = 0 ⇔ x ∈ ΩP
V
V
u
u
Difficulty : is defined in an unbounded domain (the whole space).
However, we need only inside .
V
V ΩP
εPr = εr|ΩP εextr = εr|R3\ΩP
∂V P
∂n− 1
εPrdTr e(u) · n =
εextr
εPr
∂V ext
∂n
17
A second difficulty
From we deduce that ∇ · εr∇V = ∇ · drT e(u)
maxx∈R3\ΩP
εextr (x) << minx∈ΩP
εPr (x)
A good approximate boundary condition is
∂ΩP \ Γelεr∂V
∂n+ dT
r e(u) · n = 0 on
ΩP
n
ΩS
<< 1εPr∂V P
∂n− dT
r e(u) · n = εextr∂V ext
∂n
POems (INRIA) Patrick Joly
It is possible to reduce the numerical calculations for to domain without using complex (nonlocal) boundary conditionsthanks to the following observation :
ΩPV
18
Piezoelectric sensors: emission / reception
POems (INRIA) Patrick Joly
ΩP =N
=1
ΩP
connectedcomponents
RElectrods : perfectly
conducting thin screens
V is constantalong each of them
V = 0
Γ+el
Γ−el
19
Ground electrod : the potential will be imposed to 0 .
Boundary conditions
Upper : we apply Ohm’s law.
R
V g = 0Reception
V g = 0Emission
Piezoelectric sensors: emission / reception
V = V g −Rd
dt
Γ+el
(εr∇V − dTr e(u)) · n
POems (INRIA) Patrick Joly
Boundary conditions : summary
Electric boundary conditions
V = 0
Elastic boundary conditions
(Cre(u) + dr∇V ) · n = 0 on ∂ΩS
V = V g −Rd
dt
Γ+el
(εr∇V − dTr e(u)) · n
V is constantΓ+elalong each
εr∂V
∂n− dT
r e(u) · n = 0
POems (INRIA) Patrick Joly
εr ∂nV − dTr e(u) · n = 0
The retained global model
V = 0
(Cre(u) + dr∇V ) · n = 0 on ∂ΩS
V = V g −Rd
dt
Γ+el
(εr∇V − dTr e(u)) · n
POems (INRIA) Patrick Joly
ρ∂2
∂t2u−Div C e(u) = Div d∇V + f in ΩS
∇ · ε∇V = ∇ · dT e(u) in R3
u ∈ VS := [H1(ΩS)]3
V0P =
W ∈ H
1(ΩP ) / W = 0 on Γel
VstatP = span
V 1, V 2, ..., V N
supp V = ΩP div(εr∇V ) = 0, (Ω
P )
V = 0 on Γ−,el, V = 1 on Γ+
,el,
∂nV = 0 on ∂ΩP \ (Γ−
,el ∪ Γ+,el),
Variational formulation and discretization
22
V = V0 +N
l=1
λV
V ∈ VP := V0P ⊕ Vstat
P
POems (INRIA) Patrick Joly
λ : electrod potentials
(V0,λ)−→
u ∈ VS := [H1(ΩS)]3 V ∈ VP := V0
P ⊕ VstatP
V0P =
W ∈ H
1(ΩP ) / W = 0 on Γel
VstatP = span
V 1, V 2, ..., V N
Variational formulation and discretization
23
V 1 V 2 V 3 V 41
0
1
0
1
0
1
0
POems (INRIA) Patrick Joly
λ : electrod potentials
(V0,λ)−→V = V0 +
N
l=1
λV
( if is constant )εr
ρ∂2
∂t2u−Div Cr e(u) = Div dr ∇V in ΩS
Variational formulation and discretization
24POems (INRIA) Patrick Joly
+ (Neumann like) elastic boundary conditions
∀ v ∈ VSd2
dt2
ΩS
ρu · v +
ΩS
Cr e(u) : e(v) = −
ΩP
dr ∇(V 0 +
i
λiV i) : e(v)
× test functions in VSv
∇ · εr∇V = ∇ · dTr e(u) in ΩP
+ electric boundary conditions
test functions in 1 V0PW0×
∀ W0 ∈ V0P
ΩP
εr ∇V0 +
i
λiV i
·∇W0 =
ΩP
dTr e(u) ·∇W0
∀ v ∈ VSd2
dt2
ΩS
ρu · v +
ΩS
Cr e(u) : e(v) = −
ΩP
dr ∇(V 0 +
i
λiV i) : e(v)
∀ W0 ∈ V0P
ρ∂2
∂t2u−Div Cr e(u) = Div dr ∇V in ΩS
×
1
ΩP
εr ∇V0 ·∇W0 =
ΩP
dTr e(u) ·∇W0
V0P
Variational formulation and discretization
25POems (INRIA) Patrick Joly
+ (Neumann like) elastic boundary conditions
test functions in VSv
∇ · εr∇V = ∇ · dTr e(u) in ΩP
+ electric boundary conditions
× test functions in W0
test functions
∀ v ∈ VSd2
dt2
ΩS
ρu · v +
ΩS
Cr e(u) : e(v) = −
ΩP
dr ∇(V 0 +
i
λiV i) : e(v)
ρ∂2
∂t2u−Div Cr e(u) = Div dr ∇V in ΩS
×
2 V
Variational formulation and discretization
26POems (INRIA) Patrick Joly
+ (Neumann like) elastic boundary conditions
test functions in VSv
∇ · εr∇V = ∇ · dTr e(u) in ΩP
+ electric boundary conditions
×
1 ≤ ≤ Nd
dt
ΩP
εr∇(V0 +
i
λiV i) ·∇V =d
dt
ΩP
dTr e(u) · V −
1
Rλ +
1
RV
g
test functions
∀ v ∈ VSd2
dt2
ΩS
ρu · v +
ΩS
Cr e(u) : e(v) = −
ΩP
dr ∇(V 0 +
i
λiV i) : e(v)
ρ∂2
∂t2u−Div Cr e(u) = Div dr ∇V in ΩS
×
2 V
− 1
Rλ +
1
RV
g
Variational formulation and discretization
27POems (INRIA) Patrick Joly
+ (Neumann like) elastic boundary conditions
test functions in VSv
∇ · εr∇V = ∇ · dTr e(u) in ΩP
+ electric boundary conditions
×
ΩP
εr∇V ·∇V
dλ
dt=
d
dt
ΩP
dTr e(u) · V 1 ≤ ≤ N
Variational formulation and discretization
28POems (INRIA) Patrick Joly
∀ v ∈ VSd2
dt2
ΩS
ρu · v +
ΩS
Cr e(u) : e(v) = −
ΩP
dr ∇(V 0 +
i
λiV i) : e(v)
ΩP
εr∇V ·∇V
dλ
dt+
1
Rλ =
d
dt
ΩP
dTr e(u) · V +
1
RV
g 1 ≤ ≤ N
ΩP
εr ∇V0 ·∇W0 =
ΩP
dTr e(u) ·∇W0 ∀ W0 ∈ V0
P
Md2u
dt2+Ku = −B V0 −Bstat Λ
AV0 = BTu
MstatdΛ
dt+R−1
stat Λ = BTstat
du
dt+R−1
statVg
After space semi-discretization with finite elements, we get
Variational formulation and discretization
29POems (INRIA) Patrick Joly
Md2u
dt2+Ku = −B V0 −Bstat Λ
AV0 = BTu
MstatdΛ
dt+R−1
stat Λ = BTstat
du
dt+R−1
statVg
K ←→ −DivCre(·)
B ←→ −Divdr∇
BT ←→ ∇ ·dTr ∇
A ←→ −∇ ·εTr ∇
Mstat,Rstat
diagonal
−→ M
Mass lumping process
diagonal
It is useful to eliminate via to getV0 = A−1BTuV0
Variational formulation and discretization
30POems (INRIA) Patrick Joly
Md2u
dt2+K∗ u = −Bstat Λ
MstatdΛ
dt+R−1
stat Λ = BTstat
du
dt+R−1
statVg
Md2u
dt2+Ku = −B V0 −Bstat Λ
AV0 = BTu
MstatdΛ
dt+R−1
stat Λ = BTstat
du
dt+R−1
statVg
K∗ = K+BA−1BT augmented stiffness matrix
Variational formulation and discretization
Sébastien Imperiale 31POems (INRIA)
Md2u
dt2+K∗ u = −Bstat Λ
MstatdΛ
dt+R−1
stat Λ = BTstat
du
dt+R−1
statVg
K∗ = K+BA−1BT augmented stiffness matrix
Time discretization by central finite differences
Mun+1 − 2un + un−1
∆t2+K∗ un = −B∗
statΛn+1 +Λn−1
2
MstatΛn+1 −Λn−1
2∆t+R−1
statΛn+1 +Λn−1
2= BT
statun+1 − un−1
2∆t+R−1
statVng
Variational formulation and discretization
Sébastien Imperiale 32POems (INRIA)
Mun+1 − 2un + un−1
∆t2+K∗ un = −B∗
statΛn+1 +Λn−1
2
MstatΛn+1 −Λn−1
2∆t+R−1
statΛn+1 +Λn−1
2= BT
statun+1 − un−1
2∆t+R−1
statVng
The advantage of the implicit discretization is that the stability condition of the global scheme is not affected by the electric supply process.
K∗unThe discretization in is explicit but the evaluation of requires the solution of a (small, block diagonal) linear system (laplacians in )Ω
p
u
K∗ = K+BA−1BT augmented stiffness matrix
The discretization in is implicit but its calculation is explicit since thematrices are diagonal.