Numerical simulation of electrostatic

spray painting
Matthias Maischak
I AM
f
Institute for Applied Mathematics
Hannover University, Germany
BMBF-Projekt 03STM1HV
3D Adaptive Multilevel FEM-BEM Kopplung f ur ein Elektrostatisches
Teilproblem der Nasslackierung mit Aussenraumauadung
Head: Prof. Stephan (Hannover)
Partners: Prof. Peukert, Dr. Schmid, Hr. Horn (Erlangen)
Hr. Sonnenschein, Hr. Poppner (DaimlerChrysler)
Overview
Physical/Engineering problem
Model problem
Variational Formulation
Method of Characteristics
Least-Squares formulation
Numerical examples
I AM
f
Overview 2
Electrostatic spray painting
I AM
f
Electrostatic spray painting 3
Target

Electrode
Outer boundary

T
E

A
u
u

A
0
1
I AM
f
Model conguration (3D) 4
Electric corona discharge conguration
Applied Voltage between Target and Electrode(s)
Electrical eld (Poisson equation)
Ion transport (Convection), Space charge density
Fluid (Moving Air, CFD) v
Air
Particle tracking, Particle charge
P
Corona discharge
Corona discharge: Locally high electric eld strength leads to
ionization of air, ions are transported along eld lines. Cloud of
space charge near electrode has inuence on local eld strength.
I AM
f
Electric corona discharge conguration 5
Distribution of scalar potential u

0
u = +
P
Electric eld
E = u
Density of ionic current
j
Ion
= bE + v
Air
Charge conservation law

t
+ div(j
Ion
) = S

Source term S

due to to charge transfer between and

P
.

0
: 8.8542 10
12
Electron mobility: 2 10
4 m
2
V s
I AM
f
Electric corona discharge conguration 6
Electrostatic Model problem
Poisson equation
u =

0
in (1)
Electrical eld
E = u (2)
Boundary conditions (applied voltage)
u = u
0
on
T
(Target) (3)
u = u
1
on
E
(Electrode) (4)
I AM
f
Electrostatic Model problem 7
Artical boundary conditions on
u
n
[

0 (Outow b.c.)
Su (innite domain)
(5)
or
u[

= 0 (Faraday) (6)
Using the Poincare-Steklov operator
S :=
1
2
(W + (I K

)V
1
(I K)) we can describe a charge free
exterior domain as a boundary condition for the interior domain.
The symmetric fem-bem coupling formulation approximates the
Poincare-Steklov operator by the Schur-complement of Galerkin
matrices.
I AM
f
Artical boundary conditions on 8
Linear transport equation Neglecting the air ow and the
particle charge we can describe the unipolar ion transport by the
linear transport equation
div(E) =0 in (7)
=
A
on
E
(Inow-b.c.) (8)
Non-linear transport equation Inserting the Poisson equation
in the linear transport equation we eliminate div E and we obtain
E +

2

0
=0 in (9)
=
A
on
E
(Inow-b.c.) (10)
I AM
f
Transport equation 9
Kaptzovs assumption
Let E
onset
be the eld strength where corona emission starts. With
Kaptzovs assumption we obtain on
E
the boundary condition:

A
0, E n E
onset
(11)

A
(E n + E
onset
) = 0 (12)
Peek-eld strength E
onset
(argument r in cm, result in V/m)
E
onset
=3.1 10
6
(1 +
0.308

r
) (Cylinder) (13)
E
onset
=3.1 10
6
(1 +
0.308

r/2
) (Spherical) (14)
I AM
f
Kaptzovs assumption 10
Solve initial Laplace equation
?
Adjust
A
on
E
?
Solve coupled dierential equations
?
6
Solve non-linear transport equation
?
-
-
Method of Characteristics or
Non-linear Least-Squares
?
Solve Newton-steps
6
Solve Poisson equation
6
I AM
f
Flowchart of electro static subproblem 11
Poisson equation
Find u H
1
D
:= v H
1
() : u[

T
= u
0
, u[

E
= u
1
, such that:
(u, v)
0
=
1

0
(, v)
0
v H
1
D,0
(15)
Solver: diagonal preconditioned Conjugate Gradient
Overlaping domain decomposition as preconditioner for CG
Mixed formulation
Find (p, u) H(div; ) L
2
() such that
(q, p)
0
+ (u, div q)
0
= u
0
, q n)

T
+ u
1
, q n)

E
q H(div; )
(v, div p)
0
=
1

0
(, v)
0
v L
2
() (16)
Solver: unpreconditioned GMRES
Overlapping domain decomposition as preconditioner for GMRES
I AM
f
Poisson equation 12
Symmetric FEM-BEM-coupling
Find (u, ) H
1
D
()

H
1/2
(), such that:
u =

0
in
=
u
n
on
2 = Wu + (I K

) on
0 = (I K)u + V on
Variational formulation:
Find (u, ) H
1
D
()

H
1/2
(), such that:
2(u, v)
0
+ Wu, v) + (K

I), v) = 2

0
v dx v H
1
D,0
()
(K I)u, ) V , ) = 0

H
1/2
()
I AM
f
Symmetric FEM-BEM-coupling 13
V (x) :=
1
2

1
[x y[
(y) ds
y
single layer potential
K(x) :=
1
2

n
y
1
[x y[
(y) ds
y
double layer potential
K

(x) :=
1
2

n
x
1
[x y[
(y) ds
y
adjoint double layer potential
W(x) :=
1
2

n
x

n
y
1
[x y[
(y) ds
y
hypersingular operator
I AM
f
Symmetric FEM-BEM-coupling 14
Smoothing of electrical eld
Let E

= u
h
Clement-Interpolation: Let x be a node of the mesh T
h
.
Then we dene
U(x) =

TT
h
,xT
T
i.e. U(x) is the union of all elements connected to the node x.
Then the value of E
h
in the node x is given by
E
h
(x) :=

U(x)
E

(x) dy

U(x)
1 dy
L
2
-Projection: The L
2
-Projection E
h
of E

is given by:
Find E
h
V
E
h
:= E
h
[H
1
()]
3
: E
h
[
T
ist linear , T T
h
, such
that:
(E
h
, v
h
)
L
2
()
= (E

, v
h
)
L
2
()
v
h
V
E
h
. (17)
I AM
f
Smoothing of electrical eld 15
Method of Characteristics
We compute (x) along a eld line, i.e., along a characteristic line.
We write (s) = (x(s)). x(s) is determined by b E(x(s)) =
dx(s)
ds
and x(0) = x
0
. Then there holds
0 = bE + b

0
=

(s) + b

2
(s)

0
(18)
For
0
= (x(0)) the solution reads
1

=
1

0
+
bs

0
(19)
resp.
(s) =
0

0
+ b
0
s
(20)
I AM
f
Method of Characteristics 16
The remaining problem is the determination
of the characteristic eld line:
Find x(s), such that bE(x(s)) =
dx(s)
ds
, i.e.
x(s) = x(0) +

s
0
bE(x(t)) dt
On the nite element mesh follow the direction
of the electrical eld. Ecient computation of
boundary crossing points is essential.
Computated values of space charge density are
averaged on each element. Resulting average
density is smoothed using Clement interpola-
tion.
I AM
f
Method of Characteristics 17
Least-Squares Formulation
Find the minimum of the non-linear least squares functional
J(; f,
D
,
N
) =|E +

2

0
S

|
2
L
2
(
(21)
for functions V

A
with
V

A
= v H
1
() : v =
A
on
E
(22)
The rst and second Frechet-derivative of (21) read:
J

(; ) =2

(E +

2

0
S

)(E +
2

0
) dx
J

(; , ) =2

(E +
2

0
)(E +
2

0
) dx
+ 2

(E +

2

0
S

)
2

0
dx
I AM
f
Least-Squares Formulation 18
The Newton algorithm applied to (21) reads:
Find
k+1
V

A
, such that
0 = J

(
k
; ) + J

(
k
; ,
k+1

k
) V
0
(23)
I AM
f
Least-Squares Formulation 19
Numerical Examples
Mesh Elements Nodes
Single electrode (tetrahedral,Horn) 347193 68647
Single electrode (tetrahedral,Sonnenschein) 1650233 281829
Six electrodes (tetrahedral,Sonnenschein) 3030803 545429
Using single electrode mesh created by Erlangen-group (Horn)
Applied Voltage: 26300 V
Electrode radius: 0.0075 cm
Distance Electrode-Target: 0.32 m
Measured Ion ow density (central target): 1.4mA/m
2
Computed E
onset
: 1.4 10
7 V
m
(Cylinder),
1.86 10
7 V
m
(Spherical)
Charge injection law: [

A
=
0
max(0, (E n E
onset
)/(E
max
E
onset
))
I AM
f
Numerical Examples 20
I AM
f
Numerical Examples 21
0.1000E11
0.3497E11
0.1223E10
0.4276E10
0.1495E09
0.5229E09
0.1829E08
0.6394E08
0.2236E07
0.7819E07
0.2734E06
0.9562E06
0.3344E05
0.1169E04
0.4089E04
0.1430E03
0.5000E03
0.1000E11
0.3497E11
0.1223E10
0.4276E10
0.1495E09
0.5229E09
0.1829E08
0.6394E08
0.2236E07
0.7819E07
0.2734E06
0.9562E06
0.3344E05
0.1169E04
0.4089E04
0.1430E03
0.5000E03
Space charge density (MOC), rst and 7th iteration
I AM
f
Numerical Examples 22
0.1000E11
0.4663E11
0.2175E10
0.1014E09
0.4729E09
0.2205E08
0.1028E07
0.4795E07
0.2236E06
0.1043E05
0.4862E05
0.2267E04
0.1057E03
0.4931E03
0.2299E02
0.1072E01
0.5000E01
0.1000E11
0.4663E11
0.2175E10
0.1014E09
0.4729E09
0.2205E08
0.1028E07
0.4795E07
0.2236E06
0.1043E05
0.4862E05
0.2267E04
0.1057E03
0.4931E03
0.2299E02
0.1072E01
0.5000E01
Space charge density (MOC), full injection law (E
onset
= 0)
Piecewise linear E-Field (left), piecewise constant u (right)
I AM
f
Numerical Examples 23
0.7883E+02
0.1893E+03
0.4547E+03
0.1092E+04
0.2622E+04
0.6298E+04
0.1512E+05
0.3632E+05
0.8723E+05
0.2095E+06
0.5031E+06
0.1208E+07
0.2902E+07
0.6969E+07
0.1674E+08
0.4019E+08
0.9653E+08
absolute value
0.8214E+02
0.1938E+03
0.4574E+03
0.1080E+04
0.2548E+04
0.6012E+04
0.1419E+05
0.3348E+05
0.7901E+05
0.1865E+06
0.4400E+06
0.1038E+07
0.2451E+07
0.5783E+07
0.1365E+08
0.3221E+08
0.7601E+08
absolute value
Electric eld strength (MOC), rst and 7th iteration
I AM
f
Numerical Examples 24
0.1000E11
0.4663E11
0.2175E10
0.1014E09
0.4729E09
0.2205E08
0.1028E07
0.4795E07
0.2236E06
0.1043E05
0.4862E05
0.2267E04
0.1057E03
0.4931E03
0.2299E02
0.1072E01
0.5000E01
absolute value
0.1000E11
0.4663E11
0.2175E10
0.1014E09
0.4729E09
0.2205E08
0.1028E07
0.4795E07
0.2236E06
0.1043E05
0.4862E05
0.2267E04
0.1057E03
0.4931E03
0.2299E02
0.1072E01
0.5000E01
absolute value
Space charge density near tip (MOC), rst and 7th iteration
0.1915E+06
0.2825E+06
0.4168E+06
0.6149E+06
0.9072E+06
0.1339E+07
0.1975E+07
0.2914E+07
0.4299E+07
0.6343E+07
0.9358E+07
0.1381E+08
0.2037E+08
0.3006E+08
0.4434E+08
0.6543E+08
0.9653E+08
absolute value
0.1975E+06
0.2865E+06
0.4156E+06
0.6030E+06
0.8747E+06
0.1269E+07
0.1841E+07
0.2671E+07
0.3874E+07
0.5621E+07
0.8154E+07
0.1183E+08
0.1716E+08
0.2490E+08
0.3612E+08
0.5239E+08
0.7601E+08
absolute value
Electric eld strength near tip, rst and 7th iteration
I AM
f
Numerical Examples 25
0.0000E+00
0.1058E+03
0.2117E+03
0.3175E+03
0.4233E+03
0.5292E+03
0.6350E+03
0.7408E+03
0.8467E+03
0.9525E+03
0.1058E+04
0.1164E+04
0.1270E+04
0.1376E+04
0.1482E+04
0.1588E+04
0.1693E+04
0.0000E+00
0.8859E+02
0.1772E+03
0.2658E+03
0.3544E+03
0.4429E+03
0.5315E+03
0.6201E+03
0.7087E+03
0.7973E+03
0.8859E+03
0.9745E+03
0.1063E+04
0.1152E+04
0.1240E+04
0.1329E+04
0.1417E+04
n |E |
L
2
()
|
2
/
0
|
L
2
()
|E +
2
/
0
|
L
2
()
E
max
1 1.2160E+04 1.2068E+03 1.1657E+04 1.0297E+08
2 1.0299E+04 1.1297E+03 9.8192E+03 8.2916E+07
3 1.0067E+04 1.1132E+03 9.5929E+03 8.0682E+07
4 1.0117E+04 1.1163E+03 9.6416E+03 8.1190E+07
5 1.0095E+04 1.1141E+03 9.6202E+03 8.0963E+07
I AM
f
Least-Squares error indicator 26
Conclusion/ to do
Method of Characteristics gives space charge distribution
Need better E-eld (with mixed formulation?)
Initial mesh should be better adjusted to eld lines
Use Least squares functional of transport problem for adaptive
scheme
I AM
f
Conclusion/ to do 27