Nadja Jverberg
Stockholm 2007
Thesis work
The Royal Institute of Technology
School of Electrical Engineering
Division of Electromagnetic Engineering
This report accounts for the thesis work On simulation of surface discharges at variable
voltage frequency performed between January and July 2007 under supervision of Hans
Edin. It was performed during 20 weeks as a part of Master of Science education in
Electrical Engineering, major in Electrotechnical Design at the division of
Electromagnetic Engineering, The Royal Institute of Technology.
I would like to express my gratitude to my supervisor, Hans Edin, for proposing the
project, for guidance and motivation.
I thank Cecilia Forssn and Nathaniel Taylor, Ph. D. students at the division of
Electromagnetic Engineering, for valuable advices and support.
And last but not the least I would like thank my husband Magnus, who without the first
knowledge of what a capacitor is, managed to read through this rapport and come up with
most valuable comments.
Nadja Jverberg
Stockholm 2007
II
Abstract
Insulation diagnostics is a very important tool in optimization of electric installations
maintenance. One of the possible measures of insulation deterioration that can be used for
diagnostic purposes are partial discharges. This thesis work describes an attempt to model
a resistive-capacitive network for simulating partial surface discharges in Matlab.
Unfortunately this attempt proved to be a failure due to an unexpectedly considerable
dependency of high voltage capacitances on surface resistivity. Another attempt
described here was performed in COMSOL Multiphysics, a finite-element based program
for simulation of physical processes. The main drawback with COMSOL Multiphysics
model is long simulation times. It proved to be possible to simulate discharges in
COMSOL Multiphysics. Here surface resistance was modeled with the help of a resistive
layer. Discharges were simulated by changing conductivity of the mentioned layer. Here
another problem was discovered: very long simulation times when using non-linear,
electric field dependent expressions for conductivity.
All the simulations, both in Matlab and COMSOL Multiphysics, were performed on a
computer with Intel dual-core processor: 2.13 GHz, 0.99 GB of RAM.
Sammanfattning
Isolationsdiagnostik r ett redskap som r av stor betydelse fr underhllsoptimering av
elektriska anlggningar. Ett av de mjliga mtt p isolationsfrsmring som kan anvndas
i diagnosticeringssyfte r partiella urladdningar. Det hr examensarbetet beskriver ett
modelleringsfrsk av ett resistivt-kapacitivt ntverk fr simulering av partiella
yturladdningar i Matlab. Tyvrr blev frsket misslyckat p grund av ett ovntat stort
beroende av hgspnningskapacitanser p ytresistiviteten. Ytterligare ett frsk
genomfrdes i COMSOL Multiphysics, ett program baserat p finita elementmetoden
mnat fr simuleringar av fysikaliska processer. Den huvudsakliga nackdelen med
COMSOL Multiphysics modellen r lnga simuleringstider. Det visade sig vara mjligt
att simulera urladdningar i COMSOL Multiphysics. Hr modellerades ytresistansen med
hjlp av ett resistivt skikt. Yturladdningar simulerades genom att ndra det resistiva
skiktets konduktivitet. Hr upptcks ytterligare ett problem: mycket lnga
simuleringstider vid anvndandet av olinjra konduktivitetsuttryck som beror p det
elektriska fltet.
Alla simuleringar, bde i Matlab och COMSOL Multiphysics, utfrdes p en dator med
Intel dual-core processor: 2.13 GHz, 0.99 GB of RAM.
III
List of Symbols
U potential with respect to the earthed electrode
Ui potential in the ith node with respect to the earthed electrode
Ch capacitance to the high voltage side
Cp disc capacitance
Ck gas-gas capacitance along the disc surface
r radius
r - resistance
rdisc radius of the dielectric disc
ddisc thickness of the dielectric disc
ltot distance between the high voltage electrode and the outer disc edge
permittivity
0 permittivity of free space
r relative permittivity or dielectric constant
rel electrode radius
hel electrode height
rgnd radius of the earthed electrode
N number of nodes
lcyl cylinder height
lk thickness of the resistive layer
resistivity
Rk surface resistance
RD discharge resistance
Z impedance
t time
T period
conductivity
angular frequency
Je external current density
Qj current source
V voltage
E electric field strength
D electric displacement
IV
Contents
Foreword & Acknowledgements ......................................................................................... I
Abstract ............................................................................................................................. III
Sammanfattning ................................................................................................................ III
List of Symbols ................................................................................................................. IV
1. Introduction..................................................................................................................... 1
1.1 Background ............................................................................................................... 1
1.2 Aim ........................................................................................................................... 1
1.3 Method ...................................................................................................................... 2
2 Theory .............................................................................................................................. 3
2.1 Gaseous ionization mechanisms ............................................................................... 3
2.2 Surface discharge modeling with equivalent circuits ............................................... 5
3 Experimental background data ........................................................................................ 8
4 Modeling of surface discharges ..................................................................................... 12
4.1 Experimental set-up used in this work.................................................................... 12
4.2 Attempt 1: Resistive-capacitive network ................................................................ 14
4.2.1 Dimensioning the capacitive part of the resistive-capacitive network ............ 14
4.2.2 Dimensioning the resistive part of the resistive capacitive network............. 20
4.2.3 Testing the capacitive-resistive model in frequency domain........................... 21
4.2.4 Testing the capacitive-resistive model in time domain.................................... 23
4.2.5 Conclusions on Attempt 1 (calculations in frequency domain)....................... 27
4.2.6 Conclusions on Attempt 1 (calculations in time domain)................................ 27
4.2.7 Conclusions on Attempt 1................................................................................ 27
4.3 Attempt 2: COMSOL Multiphysics model............................................................. 28
4.3.1 General information on the model ................................................................... 28
4.3.2 Modeling a discharge in COMSOL Multiphysics model ................................ 30
5 Conclusions.................................................................................................................... 38
6 Recommendations for future work ................................................................................ 39
References......................................................................................................................... 40
Bibliography ..................................................................................................................... 40
APPENDIX................................................................................................................... 41
Appendix: Calculations in frequency domain for dimensioning of capacitances from
high voltage side ........................................................................................................... 43
Appendix: Calculations in time domain ....................................................................... 44
Appendix: Matlab Algorithm........................................................................................ 46
Appendix: Matlab codes ............................................................................................... 47
1. Introduction
1.1 Background
One of the most important parts of any electrical installation is electric insulating
material. Its main function is to isolate two different electric potentials. A good electric
insulation material has to be able to withstand
1. electrical stress (i.e. appropriate value of dielectric permittivity, low dielectric
loss, high breakdown strength)
2. thermal stress (i.e. high fire point, low thermal expansion coefficient, etc.)
3. mechanical stress
4. ambient conditions (i.e. insulation material cannot be a health hazard, has to be
resistant to weather conditions, etc.)
Normally insulation material is exposed to all the above mentioned stresses which in
general depend on one another, e.g. both electrical and mechanical properties depend on
temperature. This results in ageing which as a rule worsens insulation properties. As
many electrical installations, such as generators, transformers, power cables, etc., are
designed to last many years it is imperative to be able to predict lifetime for insulation
materials. A lot of research has been performed on the insulation diagnostic methods as it
is a valuable tool in optimization of electric installations maintenance.
Partial discharges (PD) can be studied as a measure of insulation deterioration. By
definition partial discharge is a localized electrical discharge that only partially
bridges the insulation between conductors and which may or may not occur adjacent to a
conductor. 1 A special case of partial discharges is surface discharges that appear at the
boundary of different insulation materials. Surface discharges can be explained with the
help of Paschens law if the electric field can be assumed uniform. It is possible to
enhance the probability of surface discharges to occur with the help of ultraviolet ray
irradiation. Surface discharges depend on a variety of factors, such as temperature,
humidity, setup geometry, materials used for insulation, etc. More on surface discharges
can be found in the theoretical section.
1.2 Aim
The aim with this thesis work is to simulate partial surface discharges at variable
frequency of the applied voltage and make a comparison to experimental results.
Definition according to IEC Standard 60270 (3rd ed., 2000) Partial Discharge Measurements. International
Electrotechnical Commision (IEC) , Geneva, Switzerland
1.3 Method
This simulation is to be performed with the help of a resistive-capacitive network in
Matlab. Using a capacitive-resistive lumped circuit model for describing surface
discharges showed great promise in earlier research [1], [2]. This attempt along with a
model using simple equivalent circuit [3] is described in short in the theoretical section.
The ambition of this work is to perform simulations for variable frequency of the applied
voltage numerically in Matlab adding high voltage side capacitances and finally compare
Matlab solutions to experimental results.
2 Theory
2.1 Gaseous ionization mechanisms
The aim with this section is to provide some information on discharges that take place in
air. Air is a very good insulator at normal temperature and pressure with a conductivity of
10-16 10-17 A/cm2 at low fields. This level of conduction originates from radiation, both
cosmic and radioactive. [4] A discharge implies electrical breakdown of air which can be
described by a change of the gas from insulator to a conductor. For an electrical
breakdown to take place there have to be electrons available. This is achieved through
ionization processes. There is a number of ionization processes in gases, here follows a
short recollection of them.
Ionization processes:
Ionization due to electron impact:
Consider a free electron in a medium consisting of atoms. It will accelerate in the applied
electric field until it collides with an atom in its path. If the electron energy is larger than
the atom ionization energy the atom may become ionized. One of the parameters that can
be used for quantitative description of this phenomenon is the Townsend primary
ionization coefficient, . It is defined as the number of ionization collisions made by an
electron in moving a unit distance along the direction of the electric field. A stepwise
ionization is also possible when the first collision only excites an atom and the next one
ionizes it.
Photoionization
Ionization of an atom caused by photon absorption. Photon energy has to be larger or
equal to the ionization energy of the atom.
Thermal ionization
Rising temperature results in the velocity increase of atoms in the gas, leading to higher
kinetic energies. Eventually the temperature will become high enough to allow ionization
through collisions. But this is not the only mechanism that takes place. Photoionization is
also possible due to thermal emissions. Both processes, ionization through collisions and
photoionization, produce electrons. This allows ionization by electron impact.
Penning ionization
Ionization caused by metastable excited atoms due to collision processes.
Chamber
Rx
Cx
AC
Electrode
Rx
Rx
Cx
Cx
R1
b)
Dielectric disc
AC
2a
C1/x
C1/x
C1/x
R1
Electrode
Rx
AC
2x0
Rx
Cx
Rx
Cx
Cx
a)
c)
Figure 1 a) Experimental set-up for studying surface discharges; b) simple equivalent circuit; c)
generalized simple equivalent circuit
R' =
C=
w ( x + x0 )
disc ( x + x0 )
(1)
Here x is a length element along the radius of the dielectric disc, is an angle. R is
surface resistivity, C is capacitance per unit length of the dielectric disc of half width
( d disc = a ) and is the specific resistivity of the gas in the test chamber.
During the experiments the length of the first discharge in a cycle is measured. These
measurements together with data from spectral analysis reported by Brzosko et al in 1979
[3] confirm the above mentioned assumption about electrical conduction. This
assumption is valid both before and during the discharge. The resistivity of the gas layer
next to the dielectric disc should of course change drastically depending whether a
discharge occurs or not.
The simple equivalent circuit is then generalized, as seen in Figure 1c, by adding a
longitudinal capacitance C1. It is modeled as follows:
C1 = gas w ( x + x0 )
(2)
Authors of the article conclude that the simple equivalent circuit is useful for
investigation of surface discharges at the inception voltage. It is important to note that
calculations concerning the simple equivalent circuit were performed analytically. An
attempt is also made for the problem to be solved numerically with the help of Fourier
expansion. With conclusion that such analysis should be treated cautiously with regard to
the resulting physical parameters. [3]
A capacitive-resistive lumped circuit model for the surface discharge
Another suggested circuit is described in articles [1] and [2]. Experimental set-up in this
case looks as presented on Figure 2a; it consists of two asymmetrically placed parallel
electrodes with a dielectric disc in between. The corresponding circuit used is depicted in
Figure 2b. Here the authors split the circuit in two parts: one for discharging part and one
for non-discharging part. The following assumptions are made:
1) dielectric plate is infinitely long
2) discharge advances only in x direction
3) cross-sectional area of the streamer stem is constant and does not depend on time
All the calculations are performed analytically; conditions for discharge advancing,
stopping and transition from polbschel 2 to gleitbschel are formulated as follows:
1) the electric field at the streamer tip remains constant as long as the streamer is
advancing (advancing condition)
2) the streamer stops advancing if the tip current is lower than a given critical value,
this implies disappearance of the conductive stem. (stopping condition)
3) transition from polbschel to gleitbschel occurs when the total amount of energy
dissipated in the whole stem reaches a critical value. (transition condition)
One sometimes talks about polbschel and gleitbschel instead of streamer and leader
respectively when surface discharges are concerned. [1]
= r ' i1
x
for discharging part
i1
V1
=C
x
t
V
1
2 =
i2 dt
x Cs
=C
x
t
(3)
Here C is the capacitance per unit length between the streamer and electrode B; r is the
resistance per unit length of the streamer stem and Cs is the series capacitance per unit
length. It is also important to note that these parameters are constant and independent of
both coordinate x and time t.
Boundary and initial conditions are assumed as follows:
1)
2)
3)
4)
5)
The authors conclude that this model explains the tendency of the experimental data
qualitatively very well.
For this work a number of PD patterns were recorded for a certain range in frequencies,
namely between 0.1 Hz and 100 Hz, over a number of periods. Figure 4 depicts a PD
pattern taken at the frequency 10 Hz, the measurement was performed at 4 kV sinusoidal
voltage over 500 periods.
SDPC0104.DAT 10 Hz q
max
9
6000
8.5
7.9
4000
7.4
6.8
6.2
Charge (pC)
2000
5.7
5.1
4.5
4
3.4
2.9
-2000
2.3
-4000
1.7
1.2
-6000
0.6
0
0
90
180
Phase (deg)
270
360
A lot of information can be extruded from a single PD pattern; here some examples will
be mentioned:
1. Total charge per cycle with respect to the phase, Figure 5
2. Total number of partial discharges with respect to the phase, Figure 6
3. A distribution of the partial discharge amplitude (for positive and negative PDs),
Figure 7
It is also possible to get frequency dependence if one has a number of PD patterns, as for
instance total number of PDs per cycle with respect to frequency, Figure 8.
Total charge (hqs) 10 Hz SDPC01
200
150
100
50
0
-50
-100
-150
-200
0
90
180
270
360
Phase (deg)
0.1
0.05
-0.05
-0.1
90
180
270
Phase (deg)
10
360
0.06
0.04
0.02
1000
2000
3000
4000
5000
6000
7000
8000
7000
8000
Charge (pC)
Negative PDs (Nqneg) 10 Hz SDPC01
PDs per cycle
0.4
0.3
0.2
0.1
0
1000
2000
3000
4000
5000
6000
Charge (pC)
SDPC01
0
-1
10
10
10
Frequency (Hz)
11
10
z
High potential
High potential
E
l
e
c
t
r
o
d
e
E
l
e
c
t
r
o
d
e
Ch
Ck
Small wedge
Ck
1
Dielectric disc
Earthed electrode
Cp
Earthed electrode
O
Ch
Ck
N
Cp
Ch
Ck
lcyl
N+1
Cp
r
Dielectric
disc
Applying alternating voltage to the high voltage electrode produces streamers on the
surface of the dielectric disc. These streamers occur according to the same mechanism as
described above in electrical breakdown in gases. The main difference is the criterion for
the avalanche to streamer discharge transition. The critical value for breakdown electric
field is much lower in this case as the electric field is non-uniform. The discharges are
more likely to start in the small wedge between the high voltage electrode and the
dielectric disc as the electric field is strongest there. A radial discharge pattern,
symmetrical about the high voltage electrode, is expected due to rotational symmetry.
There may be an exception though in case if a gleitbschel or leader channel is formed.
Applying voltage to the high voltage electrode will result in deposition of surface charges
on the dielectric disc. This causes memory consequences that affect discharges coming
after the first one, assuming that the dielectric disc is uncharged prior to the first
discharge. Back discharges are a typical result of these memory effects. They occur due
to the residual surface charge distribution on the dielectric disc when the applied
alternating voltage is decreasing to zero after passing the maximum amplitude. Back
discharges neutralize part of the surface charges deposited on the dielectric disc.
12
13
The second boundary condition introduces a systematic error into the calculations as will
be seen in Figure 13 and Figure 14.
Easiest to model are the disc capacitances, here a following distribution is assumed:
C p ,tot =
2
rgnd
d disc
2 r dr
Cp =
d disc
14
(4)
The gas - gas capacitances along the disc surface are modeled as the capacitances
between coaxial cylinders of equal height lcyl spaced at equal distances dr as shown in
Figure 9b.
Ck ,n +1 =
2 0 lcyl
r + ( n + 1) dr
ln el
rel + n dr
here n = 0 :1: N
(5)
Thus there is only one degree of freedom left for determining the potential distribution on
the disc surface, namely capacitances to the high voltage side, Ch. Keeping in mind that
this model should produce the same result as any other model, the potential distribution
obtained with the COMSOL Multiphysics model 3 was used in order to calculate
capacitances Ch. Some COMSOL Multiphysics results for U = 1 kV and f = 10 Hz are
shown in Figure 11 and Figure 12. Results from Figure 12 are used for comparison. This
is performed by using Kirchhoffs laws which can be rewritten for calculating the
capacitances in the following way (this can be achieved by rewriting equation system
found in Appendix: Calculations in frequency domain for dimensioning of capacitances
from high voltage side):
Ch (1) =
Ch ( i ) =
( C (1) + C ( 2) + C (1) ) U
k
Ck ( 2 ) U 2 Ck (1) U h
U h U1
( C ( i ) + C ( i + 1) + C ( i ) ) U
Ch ( N ) =
Ck ( i + 1) U i +1 Ck ( i ) U i 1
U h Ui
( C ( N ) + C ( N + 1) + C ( N ) ) U
k
(6)
Ck ( N ) U N 1
Uh UN
This gives the distribution of high voltage side capacitances over the distance between
the electrode and the outer disc radius. It can be approximated with a rather satisfactory
result (see Figure 13, Figure 14) with help of the following expression:
Ch ,appr
140 1019
=
+ 64 1017
5
r + 2.5 10
(7)
As seen from Figure 14 the approximation error does not go above 15% except for the
outer most edge of the disc. This can be readily explained as a fault caused by
approximation formula (7): Ch,appr never goes to zero at the outermost disc edge while the
distribution of high voltage side capacitances obtained from COMSOL Multiphysics does
so. Using formula (7) naturally leads to a systematic error, implying that all the results
obtained by Matlab model at the outermost disc edge will not be trustable. It has however
15
Figure 12 Potential on the disc surface with respect to distance from the electrode over one period
(T = 100 ms)
16
-13
1.6
x 10
Approximated distribution
Calculated distribution, C
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
Figure 13 Capacitance distribution, Ch, from high voltage electrode to disc surface
Approximation error in %
20
10
0
-10
-20
-30
-40
-50
-60
-70
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
Next step is using all the above mentioned capacitances for calculating potential
distribution on the disc surface. This is done by putting up the following equation system
using Kirchhoffs current law in the frequency domain (for details please refer to
17
(8)
Here it can be noted that the cylinder height lcyl is chosen in such a way as to produce
minimal condition number for matrix A of the equation system (8) in order to provide
reasonable values for capacitances. Here it is chosen to be lcyl = 1107 ( m ) giving
condition number of approximately 13.
Equation system (8) is solvable as it consists of N equations and N unknowns. Solving it
gives very good correspondence between capacitive network and COMSOL Multiphysics
models see Figure 15 and Figure 16.
Potential distribution on the disc surface
1000
Approximated (capacitive network)
Comsol
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
18
0.05
0.06
Approximation error in %
70
60
(Ucalc - UComsol)/Ucalc
50
40
30
20
10
0
-10
-20
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
Figure 16 Difference in % between the capacitive network and COMSOL Multiphysics models
19
Rk =
l
A
dr
2 r lk
(9)
These resistances are connected in parallel with the Ck gas-gas capacitances along the
disc surface in the resistive-capacitive network, see Figure 17.
In analogy with equation system (8) potential distribution on the disc surface in the
frequency domain is calculated as follows:
1
1
1
1
1
1
1
+
+
+
U +
U =
+
Node 1:
U h
Z (1) Z ( 2 ) Z (1) Z (1) 1 Z ( 2 ) 2
k
h
p
k
Z k (1) Z h (1)
k
Node i :
1
1
1
1
1
1
(10)
U i 1
+
+
+
U +
U =
U h
Z ( i ) Z ( i + 1) Z (1) Z ( i ) i Z ( i + 1) i +1
Zk ( i )
Z
i
k
h
p
k
h( )
k
Node N :
1
1
1
1
1
1
U N 1
+
+
+
U h
U N =
Zk ( N )
Zh ( N )
Z k ( N ) Z k ( N + 1) Z h ( N ) Z p ( N )
Zk =
Here:
Rk
1 + j Rk Ck
1
Zh =
j Ch
1
Zp =
j Cp
20
(11)
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
Figure 18 Comparison of the Matlab and COMSOL Multiphysics models for = 1*108 *m
Potential distribution on the disc surface
1000
Approximated (cap-res network with Ch)
Approximated (cap-res network w/o Ch)
Comsol
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
21
0.05
0.06
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
22
0.05
0.06
900
800
700
600
500
400
300
200
100
0
0.01
0.02
0.03
0.04
Distance from the electrode [m]
0.05
0.06
t
t
+
( C k 1 + C k 2 + C h1 + C p 1 ) U 1 ( t ) + C k 2
t ( C k 1 + C h1 ) (U h ( t + t ) U h ( t ) )
U 2 (t )
R
R
R
Rk 1
k2
k2
k1
Node i :
C ki U i 1 ( t + t ) ( C ki + C ki +1 + C hi + C pi ) U i ( t + t ) + C ki +1 U i +1 ( t + t )
t
C ki
Rki
t
t
+
( C ki + C ki +1 + C hi + C pi ) U i ( t ) + C ki +1
U i 1 ( t ) +
U i +1 ( t ) C hi (U h ( t + t ) U h ( t ) )
Rki +1
Rki R ki +1
Node N :
C kN U N 1 ( t + t ) ( C kN + C kN +1 + C hN + C pN ) U N ( t + t )
t
C kN
RkN
t
+
( C kN + C kN +1 + C hN + C pN ) U N ( t ) C hN (U h ( t + t ) U h ( t ) )
U N 1 ( t ) +
RkN RkN +1
23
(12)
800
600
400
200
40
20
-20
0.01
0.02
0.03
Distance [m]
0.04
0.05
-40
0.06
0.01
0.02
0.03
Distance [m]
0.04
0.05
0.06
-200
Potential distribution [V]
10
-300
-400
-500
-600
-700
0
-10
-20
-30
-800
-40
-900
-1000
0.01
0.02
0.03
Distance [m]
0.04
0.05
-50
0.06
0.01
0.02
0.03
Distance [m]
0.04
0.05
0.06
Figure 23 Potential distribution at different times (T/4, T/2, 3T/4, T) for = 1*108 *m
Potential distribution with respect to distance at t = 50 [ms], = 1e+7 [Ohm*m]
200
Comsol
TD model
150
Potential distribution [V]
800
600
400
100
50
200
0.01
0.02
0.03
Distance [m]
0.04
0.05
-50
0.06
0.01
0.02
0.03
Distance [m]
0.04
0.05
0.06
-20
-400
-600
-800
-40
-60
-80
-100
-120
-1000
-140
-1200
0.01
0.02
0.03
Distance [m]
0.04
0.05
0.06
-160
0.01
0.02
0.03
Distance [m]
0.04
0.05
0.06
Figure 24 Potential distribution at different times (T/4, T/2, 3T/4, T) for = 1*107 *m, T = 100 ms
24
150
50
TD model
Comsol
100
30
Potential distribution [V]
TD model
Comsol
40
50
-50
20
10
0
-10
-20
-30
-100
-40
-150
0.02
0.04
0.06
Time [s]
0.08
0.1
-50
0.12
0.02
0.04
0.06
Time [s]
0.08
0.1
0.12
25
20
TD model
Comsol
20
TD model
Comsol
15
15
Potential distribution [V]
10
10
5
0
-5
-10
5
0
-5
-10
-15
-15
-20
-25
0.02
0.04
0.06
Time [s]
0.08
0.1
-20
0.12
0.04
0.06
Time [s]
0.08
0.1
0.12
15
5
TD model
Comsol
TD model
Comsol
10
3
Potential distribution [V]
0.02
-5
2
1
0
-1
-2
-3
-10
-4
-15
0.02
0.04
0.06
Time [s]
0.08
0.1
-5
0.12
0.02
0.04
0.06
Time [s]
0.08
0.1
0.12
Figure 25 Potential distribution as a function of time at a distance r = {2, 4, 6, 8, 10, 20} mm away
from the eletrode for = 1*108 *m
25
200
50
TD model
Comsol
30
Potential distribution [V]
TD model
Comsol
40
150
100
50
20
10
0
-10
-20
-30
-50
-40
-100
0.02
0.04
0.06
Time [s]
0.08
0.1
-50
0.12
0.02
0.04
0.06
Time [s]
0.08
0.1
0.12
30
20
TD model
Comsol
TD model
Comsol
15
20
10
10
-10
5
0
-5
-10
-20
-15
-30
0.02
0.04
0.06
Time [s]
0.08
0.1
-20
0.12
0.04
0.06
Time [s]
0.08
0.1
0.12
15
5
TD model
Comsol
TD model
Comsol
10
3
Potential distribution [V]
0.02
-5
2
1
0
-1
-2
-3
-10
-4
-15
0.02
0.04
0.06
Time [s]
0.08
0.1
-5
0.12
0.02
0.04
0.06
Time [s]
0.08
0.1
0.12
Figure 26 Potential distribution as a function of time at a distance r = {2, 4, 6, 8, 10, 20} mm away
from the eletrode for = 1*107 *m
26
27
The experimental set-up (see Figure 9a) is drawn in COMSOL Multiphysics as shown in
Figure 27. The resistive layer has 0.1 mm thickness and thus is hard to see, an area close
to the electrode is zoomed in. Chosen module for analysis is 2D axial symmetry,
AC/DC Meridional Electric Currents (emqvw). All distances are in meters.
Solver Parameters:
For comparison with Matlab model solutions in both frequency and time domain were
obtained.
For solution in frequency domain Time-harmonic, Electric Currents/Stationary analysis
was used. Here the governing equation solved in COMSOL Multiphysics is:
( ( + j 0 r ) V J e ) = Q j
28
For solution in time domain Transient Electric Currents/Time dependent analysis was
performed. Here the governing equation solved in COMSOL Multiphysics is:
( 0 r V )
(V J e ) = Q j
t
Boundary settings are chosen in accordance with the necessary analysis while subdomain
settings stay the same.
Here it is important to note that all analysis was performed for isotropic materials
implying the constitutive equation: D = 0 r E , there are no current sources ( Q j = 0 ) and
external current density equals to zero ( J e = 0 ).
Subdomain Settings:
The symmetry line is set as axial symmetry r = 0 . The electrode is set as electric potential
either as V = V0 sin ( t ) for time dependent analysis or in phasor notations as V = V0 for
frequency domain analysis. Surfaces on the interface between air/resistive film and
resistive film/dielectric disc are set to continuity n ( J1 J 2 ) = 0 . The rest of the
boundaries are set to ground V = 0 .
Here all the simulations were performed for U = 5 kV and f = 10 Hz.
29
(13)
Here it is important to note that expression (13) does not depend on time. This is of
course impossible situation in practice, but it is an important step in simulations. Setting
conductivity to expression (13) and running simulations over a quarter of the period
(25 ms) with lcr = 0.02 m and time step dt = 10-4 s, the obtained results are presented in
Figure 28 through Figure 32. Simulation time: 11.172 s.
30
Figure 30 Electric potential with respect to distance from the electrode for various times (from 0 to
25 ms)
31
Figure 31Electric field strength, r - component, with respect to distance from the electrode
32
As stated earlier that is an impossible scenario. First discharge will occur only after some
certain time t0 after voltage is applied, when the electric field increases above a critical
value. Trying to add a time dependency to the conductivity leads to the following
expression (14):
(14)
Conductivity in this case is low, 10-9 S/m, in entire resistive layer until time t0 = 20 ms; it
then goes drastically up to a high value of 10-2 S/m over a certain length lcr in the vicinity
of high voltage electrode, remaining low over the rest of the layer. The results of this
simulation can be seen in Figure 33 through Figure 38. One can see on the plots
behaviour characteristical for a discharge. The electric potential start propagating to the
head of the discharge as soon as conductivity goes up; there is no voltage drop over the
part of the resistive layer with high conductivity, please refer to Figure 35 and Figure 36.
The electric field collapses after reaching the critical value, Figure 37. The current should
first increase and then, after the electric field collapses, decrease. Unfortunately due to a
numerical problem the current has a peak.
Simulation time: 68.359 s
33
Figure 35 Electric potential with respect to time at r = 2 mm away from the electrode
34
Figure 37 Electric field, r - component, with respect to time at r = 2 mm away from the electrode
35
Figure 38 Current, r - component, with respect to time at r = 2 mm away from the electrode
However there is a big minus with this approach, namely the point in time when
conductivity changes has to be chosen beforehand. Instead it has to depend on the electric
field. A naive approach to correct this oversight would be to try the following expression
(15):
= 109 ( Er < Ecr ) + 102 ( Er > Ecr )
(15)
Expression (15) depends directly on the electric field; it sets conductivity to a low value,
10-9 S/m, when r component of the electric field is below the critical value, and to a
high value, 10-2 S/m, otherwise. Unfortunately simulation time increases immensely, it
becomes far too long: 2 % of the simulation of 25 ms takes approximately 97000 s. This
is far too long for a thesis work. Instead a more sophisticated expression for conductivity,
expression (16), is tested:
Er
Ecr
= 109 exp
(16)
This simulation runs much faster, but it still takes over 10 hours. Trying out different
values for Ecr takes far too long time as the time step has to be reduced drastically (10-7 s
in this case).
Modifying expression (16) provides expression (17) that looks promising. It does
maintain conductivity high even when the electric field collapses for as long as the
current stays high enough. Unfortunately implementing it in COMSOL Multiphysics
proves to be tricky as expression (17) is recursive, i.e. conductivity depends on itself
36
(thanks to Ohms law: J = E ). Not to mention very long simulation times that will
doubtless occur.
Er J r
+
E
cr J cr
= 109 exp
(17)
The main problem with COMSOL Multiphysics simulation is time consumption. In order
to reduce simulation time two attempts were made in order to exchange the thin resistive
film for a boundary described by 1) PDE and 2) weak form. They are not presented here
as they did not give good correspondence with the original model described in this work.
The most likely reason is the lack of connection to the environment.
This is regrettably as far as I had time to investigate considering the time limit on the
thesis work and the fact that the lion part of the time was spent on the Matlab model as
well as trying to optimize the COMSOL Multiphysics model.
37
5 Conclusions
1. Regrettably resistive-capacitive network, suggested in this work, is not fit for
simulation of discharges. The most likely reason for deviation between Matlab
and COMSOL Multiphysics models is the high voltage side capacitances
changing with respect to resistivity.
2. The gas - gas capacitances along the disc surface can in fact be neglected as
claimed in [3]. All the simulation graphs in this work are obtained with Ck:s
present however.
3. It is possible to simulate first discharge in COMSOL Multiphysics; unfortunately
dependency on electric field is hard to implement. It does work to simulate by
changing conductivity in time over a certain length of the resistive layer.
4. Two unsuccessful attempts on optimization of the COMSOL Multiphysics model
were made in order to minimize simulation times: using PDE and weak form for
describing the resistive layer.
5. Simulation in COMSOL Multiphysics did not provide with sufficient statistics for
comparing to the experimental results as simulations could only be performed
over a quarter of a period (25 ms) within manageable time limits.
38
39
References
[1] A. Kanematsu, G. Sawa, M. Ieda (1983) An Analytical Approach to Empirical Law
on the Transition Voltage from Polbschel to Gleitbschel using a Transmission Line
Model. Japanese Journal of Applied Physics, 22(8), pp. 1271-1276
[2] A. Kanematsu, G. Sawa, M. Ieda (1983) An Analysis of Various Properties of Surface
Discharge Using a Transmission-Line Model. Japanese Journal of Applied Physics,
22(12), pp. 1906-1909
[3] J. Ashkenazi et al. (1982) Potential field solutions and dielectric surface discharges.
J. Phys. D: Appl. Phys., 15(1982), pp. 1849 1871.
[4] E. Kuffel, W.S. Zaengl, J. Kuffel (2000) High Voltage Engineering: Fundamentals.
2nd ed., Oxford: Butterworth Heinemann, ISBN 0 7506 3634 3
[5] V. Cooray (ed.) (2003) The Lightning Flash. London: Institution of Electrical
Engineers, IEE Power and Energy Series, ISBN 0 8529 6780 2.
[6] J.M. Meek, J.D. Craggs (1978) Electrical Breakdown of Gases. Chichester: John
Wiley & Sons, Ltd, ISBN 0 4719 9553 3.
[7] H. Edin (2001) Partial Discharges Studied with Variable Frequency of the Applied
Voltage. Ph. D. Thesis, Stockholm, Royal University of Technology, Department of
Electrical Engineering. TRITA EEK 0102, ISSN 1100 1593.
Bibliography
[1] G. Engdahl et al. (2005) Elektroteknisk Modellering. Stockholm: Kungliga Tekniska
Hgskolan, Elektrotekniska System, A ETS/EEK 0502.
[2] G. Engdahl et al. (2005) Elektroteknisk Modellering. Stockholm: Kungliga Tekniska
Hgskolan, Electrical Engineering, A ETS/EEK 0507.
40
APPENDIX
41
42
.
.
.
.
.
0
.
.
.
.
.
.
.
.
.
A=
0
0
. Ck ( i ) ( Ck ( i ) + Ck ( i + 1) + Ch ( i ) + C p ( i ) ) Ck ( i + 1) 0
.
0
.
.
.
.
.
.
.
.
.
0
0
0
0
0
0
0 Ck ( N ) ( Ck ( N ) + Ck ( N + 1) + Ch ( N ) + C p ( N ) )
b = ( Ck (1) + Ch (1) ) Ch ( 2 ) . Ch ( i ) . Ch ( N )
43
Rk1
Rk2
Ck1
AC
Ch2
Ck2
Cp1
ChN
RkN+1
N+1
CkN+1
Cp2
CpN
dU i U i ( t + t ) U i ( t )
dt
t
Node 1 :
U h ( t ) U1 ( t )
dU ( t ) U ( t ) U 2 ( t )
d
d
d
+ Ck 1 (U h ( t ) U1 ( t ) ) + Ch1 (U h ( t ) U1 ( t ) ) = C p1 1 + 1
+ Ck 2 (U1 ( t ) U 2 ( t ) )
Rk1
dt
dt
dt
Rk 2
dt
Ck1
dU ( t )
U (t ) U (t ) U (t ) U (t )
d
(U h ( t ) U1 ( t ) ) + Ch1 dtd (U h ( t ) U1 ( t ) ) C p1 dt1 Ck 2 dtd (U1 ( t ) U 2 ( t ) ) = 1 R 2 h R 1
dt
k2
k1
dU ( t ) 1
U 2 (t ) U h (t )
dU ( t )
dU1 ( t )
1
+ Ck 2 2
=
+
( Ck1 + Ch1 ) h
U1 ( t )
dt
dt
R
R
R
R
dt
k2
k2
k1
k1
U ( t + t ) U1 ( t )
U ( t + t ) U 2 ( t )
( Ck1 + Ck 2 + Ch1 + C p1 ) 1
+ Ck 2 2
t
t
U 2 (t ) U h (t )
U ( t + t ) U h ( t )
1
1
( Ck1 + Ch1 ) h
U1 ( t )
R
R
R
R
t
k2
k2
k1
k1
( Ck1 + Ck 2 + Ch1 + C p1 )
( Ck1 + Ck 2 + Ch1 + C p1 ) U1 ( t + t ) + Ck 2 U 2 ( t + t )
U h (t )
t t
+
( Ck 1 + Ck 2 + Ch1 + C p1 ) U1 ( t ) + Ck 2
t ( Ck1 + Ch1 ) (U h ( t + t ) U h ( t ) )
U 2 ( t )
R
R
R
Rk 1
k2
k2
k1
Node i :
U i 1 ( t ) U i ( t )
d
d
+ Cki (U i 1 ( t ) U i ( t ) ) + Chi (U h ( t ) U i ( t ) ) =
Rki
dt
dt
= C pi
dU i ( t ) U i ( t ) U i +1 ( t )
d
+
+ Cki +1 (U i ( t ) U i +1 ( t ) )
dt
Rki +1
dt
in analogy to Node 1 :
Cki U i 1 ( t + t ) ( Cki + Cki +1 + Chi + C pi ) U i ( t + t ) + Cki +1 U i +1 ( t + t )
t
Cki
R
ki
t
t
+
( Cki + Cki +1 + Chi + C pi ) U i ( t ) + Cki +1
U i 1 ( t ) +
U i +1 ( t ) Chi (U h ( t + t ) U h ( t ) )
R
R
R
ki +1
ki +1
ki
44
Node N :
U N 1 ( t ) U N ( t )
dU N ( t ) U N ( t )
dU N ( t )
d
d
+ CkN (U N 1 ( t ) U N ( t ) ) + ChN (U h ( t ) U N ( t ) ) = C pN
+
+ CkN +1
RkN
dt
dt
dt
RkN +1
dt
in analogy to Node 1 :
CkN U N 1 ( t + t ) ( CkN + CkN +1 + ChN + C pN ) U N ( t + t )
t
t
t
CkN
+
( CkN + CkN +1 + ChN + C pN ) U N ( t ) ChN (U h ( t + t ) U h ( t ) )
U N 1 ( t ) +
R
R
R
kN
kN +1
kN
45
46
47
48
Program code 2 for comparison between COMSOL Multiphysics and Matlab models for
various resistivity values (Frequency domain):
clear all; close all;
%Loading data from COMSOL Multiphysics for resistivity = 1e+8 Ohm*m
load realVe8.txt % Potential distribution on the disc surface for f = 10 Hz
load imagVe8.txt % Potential distribution on the disc surface for f = 10 Hz
%Complex potential distribution
U = realVe8(2:end,2) + j*imagVe8(2:end,2);
Uh = abs(realVe8(1,2) + j*imagVe8(1,2)); Ul = abs(realVe8(end,2) + j*imagVe8(end,2));
ra = realVe8(:,1);
figure
Eps0 = 8.85e-12; Epsr = 3.3; r_jord = 5.5e-2; r_sk = r_jord; r_el = 3e-3; d_sk = 1e-3; ltot = r_sk - r_el;
N = round(length(realVe8)/20) - 1; dr = ltot/(N+1); r1 = dr:dr:ltot-dr;
% Taking each 20th data point:
Ua = [];
for g = 1:N
Ua = [Ua U(20*g)]; %Potentials in nodes 1 through N
end
Ua = [Ua U(end)];
%Defining disc capacitances:
Cp = Eps0*Epsr*2*pi*(r_el + r1)*dr/d_sk;
%Defining gas-gas capacitances along the disc surface:
lcyl = 1e-7; Ck = zeros(1, N + 1);
for k = 0:N
Ck(k + 1) = 2*pi*Eps0*lcyl/log((r_el + (k+1)*dr)/(r_el + k*dr));
end
%Defining high voltage side capacitances with the help of approximating function obtained in the program code 1.
b = 64e-17; n = 1; r0 = -2.5e-5; a = 140e-19;
ya = @(r) a./(r - r0).^n + b; Ch = ya(r1);
%Defining surface resistances:
ro = 1e+8; lrk = 1e-5; f = 10; omega = 2*pi*f;
Rk = ro*dr./(2*pi*[r1 ltot]'*lrk);
%Calculating impedances:
Zp = 1./(j*omega*Cp); Zk = (Rk./(1 + j*omega.*Rk.*Ck'))'; Zh = 1./(j*omega*Ch);
%Calculating potential distribution with all capacitances and resistances present:
A = diag(-(1./Zh + 1./Zp + 1./Zk(1:N) + 1./Zk(2:N+1)))+ diag(1./Zk(2:end-1),1)+ diag(1./Zk(2:end-1),-1);
b = -[(1/Zk(1)+ 1/Zh(1))*Uh, 1./Zh(2:N)*Uh];
Uanp = [Uh; A\b'; 0];
%Calculating potential distribution, here high voltage side capacitances are neglected:
A2 = diag(-(1./Zp + 1./Zk(1:N) + 1./Zk(2:N+1)))+ diag(1./Zk(2:end-1),1)+ diag(1./Zk(2:end-1),-1);
b2 = -[Uh/Zk(1), zeros(1, N-1)]';
Uanp2 = [Uh; A2\b2; 0];
hold on
Uorig = [Uh; Ua'];
plot([0 r1 ltot],abs(Uanp),[0 r1 ltot],abs(Uanp2),'-.', [0 r1 ltot], abs(Uorig), '--')
title('Potential distribution on the disc surface')
xlabel('Distance from the electrode [m]')
ylabel('Potential distribution [V]')
legend('Approximated (cap-res network with Ch)','Approximated (cap-res network w/o Ch)', 'Comsol')
grid on
figure
Ures = abs(Uanp(1:end-1)) - [Uh abs(Ua(1:end-1))]';
plot([0, r1], Ures)
49
title('Approximation error')
xlabel('Distance from the electrode [m]')
ylabel('U_c_a_l_c - U_C_o_m_s_o_l [V]')
grid on
figure
Uresproc = Ures./[Uh abs(Ua(1:end-1))]'*1e+2;
plot([0, r1], Uresproc)
title('Approximation error in %')
xlabel('Distance from the electrode [m]')
ylabel('U_c_a_l_c - U_C_o_m_s_o_l')
grid on
50
Program code 3 for comparison between COMSOL Multiphysics and Matlab models for
various resistivity values (Time domain):
clear all; close all;
tic %starting the timer
%Loading data from COMSOL Multiphysics: Uh = 1 kV, f = 10 Hz
load Ve8r_t100.txt %Potential distribution with respect to distance from the electrode at time t = 25 ms
load Ve8t_r2.txt %Potential distribution with respect to time at r = 2 mm away from the electrode
load Ve8t_r4.txt %Potential distribution with respect to time at r = 4 mm away from the electrode
load Ve8t_r6.txt %Potential distribution with respect to time at r = 6 mm away from the electrode
load Ve8t_r8.txt %Potential distribution with respect to time at r = 8 mm away from the electrode
load Ve8t_r10.txt %Potential distribution with respect to time at r = 10 mm away from the electrode
load Ve8t_r20.txt %Potential distribution with respect to time at r = 20 mm away from the electrode
Eps0 = 8.85e-12; Epsr = 3.3; Uh = 1e+3;
r_jord = 5.5e-2; r_sk = r_jord; r_el = 3e-3;
d_sk = 1e-3; ltot = r_sk - r_el;
N = round(length(Ve8r_t100)/20) - 1;
dr = ltot/(N+1); r1 = dr:dr:ltot-dr;
%Defining disc capacitances:
Cp = Eps0*Epsr*2*pi*(r_el + r1)*dr/d_sk;
%Defining gas-gas capacitances along the disc surface:
lcyl = 1e-7; Ck = zeros(1, N + 1);
for k = 0:N
Ck(k + 1) = 2*pi*Eps0*lcyl/log((r_el + (k+1)*dr)/(r_el + k*dr));
end
%Defining high voltage side capacitances with the help of approximating function obtained in the program code 1.
b = 64e-17; n = 1; r0 = -2.5e-5; a = 140e-19;
ya = @(r) a./(r - r0).^n + b; Ch = ya(r1);
%Defining surface resistances:
ro = 1e+8; lrk = 1e-5; f = 10; omega = 2*pi*f;
Rk = ro*dr./(2*pi*[r1 ltot]'*lrk);
dt = 8e-4; %time step
tstop = 0.1; %defining simulation time
t = 0; M = floor(tstop/dt/10) + 1; Ut = zeros(N,1); Ares = zeros(N, M);
%Calculating potential distribution in time domain:
At = diag(-(Ck(1:N) + Ck(2:N+1) + Ch + Cp)) + diag(Ck(2:N), 1) + diag(Ck(2:N), -1);
Adt = diag(-(Ck(1:N) + Ck(2:N+1) + Ch + Cp - dt./Rk(1:N)' - dt./Rk(2:N+1)')) + diag(Ck(2:N) - dt./Rk(2:N)', 1) +
diag(Ck(2:N) - dt./Rk(2:N)', -1);
bdt1 = [Ch(1) + Ck(1); Ch(2:N)']; bdt2 = [dt/Rk(1); zeros(N-1, 1)];
h = 1;
for t = 0:dt:tstop
Uhf = Uh*sin(omega*t); Uhfdt = Uh*sin(omega*(t + dt));
bt = Adt*Ut - bdt1*(Uhfdt - Uhf) - bdt2*Uhf;
Ut = At\bt; %Ut at t = t + dt
Ares(:,h) = Ut; %Storing solutions for later comparison
h = h + 1;
end
plot(Ve8r_t100(:,1), Ve8r_t100(:,2), [0 r1 ltot], [Uh*sin(omega*tstop); Ut; 0], '--')
legend('Comsol', 'TD model')
title('Potential distribution with respect to distance at t = 25 [ms], \rho = 1e+8 [Ohm*m]')
xlabel('Distance [m]')
ylabel('Potential distribution [V]')
grid on
figure
plot(0:dt:tstop+dt, [0 Ares(29,:)], Ve8t_r2(:,1), Ve8t_r2(:,2), '--')
legend('TD model', 'Comsol')
51
52