Professional Documents
Culture Documents
3, June 1990
453
ABSTRACT
An attempt is made to elucidate the mechanism of partial discharge (PD) occurring in the CIGRE Method I1 (CM-11) electrode system, which is a representative closed-void model system. We developed a computer-aided PD measuring system.
This allows us to obtain phase information of all PD pulses,
together with their amplitudes, so that a statistical analysis of
these data can be discussed. Measurements of PD are made
for the CM-I1 electrode system. Effects of the pressure and
gas inside the void on the PD are examined. Taking into account the experimental results, we propose a model for the PD
mechanism. This model assumes that the statistical time lag
of discharge depends on the overvoltage and that the residual
voltage depends on the PD magnitude. A Monte Carlo simulation of the PD distribution is made on the basis of this model.
The computed results agree well with the experimental data
and the appearance of swarming pulsive micro discharges. The
physics of the model also are discussed.
INTRODUCTION
the mechanism of P D degradation, a deeper understanding of the fundamental properties of P D is essential [2].
variety of organic polymers have been used as electrical insulating materials for electric power apparatus
and cables. A large amount of work has been done on
degradation of the insulating materials caused by partial discharges (PD) occurring at various defects in the
polymer insulator itself and at the interface between electrodes and the insulating materials [l]. P D degradation
is one of the most significant factors deciding the life of
the insulation system. Much remains unknown about the
mechanisms of the P D degradation. In order to clarify
For the study of PD phenomena, various types of measurement have been used such as observation of waveform
of PD pulse, high-speed photographic measurements of
discharge column, and analysis of the discharge spectrum.
Very few attempts, however, have been made to use statistical analysis of a large volume of data on PD pulses
[31.
. ...-..
454
Sphere Electrode
Stainless Steel Ball (@Set)
Copper Pipe (75mm,@4)
EXPERl M ENTAL
ELECTRODE S Y S T E M
1 illustrates the CIGRE Method I1 electrode
system, which is an improved type of CIGRE Method
I (CM-I) electrode and is expected to be accepted by CIGRE SC15 as a standard test electrode for internal PD
measurements [5]. The electrode system used had a cylindrical void of 125 pn in thickness and 3 cm in diameter.
The sample used was a board made of 1 mm thick epoxy
resin. The void was filled with dry air at atmospheric
pressure for the standard measurements. The PD inception voltage was about 4 kV of rms value. Ac voltage of
7 kV at 60 H a was applied to the test electrode through
this experiment. P D pulse amplitude distribution was
measured at a given aging time for 60 s, corresponding
to 3600 cycles, by using the developed system. In order
to examine the effects of pressure and gas content inside
a void, further measurements were made for different gas
pressures and for different ratios of oxygen and nitrogen
content in the gaseous mixture.
IGURE
Kapton (OB125nnt)
\ Plane Electrode
Stainless Steel (@to)
Figure 1.
Structure of CIGRE Method I1 electrode system.
measurement of all P D pulse amplitudes, as well as the
phase angle information, i.e., at which phase angle each
P D pulse takes place. The P D pulse amplitude distribution obtained by this method is expected to include
significant information on the P D phenomena. The measurements and the statistical analysis of the correlation
between P D pulse-height distribution and the phase angle of the applied voltage is a good prospect for a new
and useful means of PD study [3].
From this point of view, we have made measurements of
P D pulses using the CIGRE Method I1 (CM-11) electrode
system, developed a computer system, and examined the
statistical properties of partial discharges. Okamoto et
al. [4] made a simulation of P D characteristics including
the phase angle of the applied voltage. This reference
probably has been the only one concerning such P D simulation. Their model assumed Whiteheads equivalent
circuit and introduced a probability distribution of the
discharge time lag. The authors [4] approximated a sinusoidal voltage waveform as a triangular waveform because
it simplified the calculation.
In this paper, we propose a model for a P D mechanism which assumes that the statistical time lag of the
discharge is dependent on overvoltage of a void gap and
the residual voltage depends upon the charge of preceding PD. With Monte Carlo simulation of PD characteristics, the model can explain, to some extent, the P D distribution and occurrence of swarming pulsive microdischarges (SPMD) measured for CM-I1 electrode system.
There seems no other report except the present one that
has made Monte Carlo simulation of the P D distribution
with the phase angle information of a n applied voltage,
On the other hand, the CM-I1 electrode system is designed so that the wall does not influence the PD characteristics. This is one of the main reasons why CM-I1
electrode system is recognized as a standard test electrode for the evaluation of a sample against internal PD.
It has been reported that for the case of relatively low
voltage application no tree-like surface discharges took
place on the sample surface [6], and that P D occurred
over the sample surface within a radius of 2.4 to 3 mm
right below the upper electrode [7]. The electrode system
is thus considered appropriate for the fundamental study
of internal PD.
For a usual small-void system such as CM-I electrodes,
the wall surrounding a void strongly affects the internal
PD, which makes the understanding of the mechanism
more difficult.
MEASUREMENTS SYSTEM
The details of the developed computer-aided P D measuring system (CAPDMS) to accomplish the statistical
analysis of P D distribution are given below. The measuring system can record the number of P D pulses and the
SYMBOL LIST
the capacitance of a void [F]
the capacitance seen from the discharge gap C ,
PI
the capacitances of the insulator in parallel and
in series with the void [F]
gap distance or sample thickness [m]
frequency of applied voltage [H,]
the current flowing through the resistance z of the
discharge column [A]
PD pulse amplitude distribution
=xi
n($,,q j ) , the relation between the magnitude of charge and the number of PD pulses ( q - n
characteristics)
= ( l / T ) E i n ( & , q , ) , the phase angle dependence
of average discharge occurrence density ($ - fi
characteristics) [1 /SI
N / f [1 /cycle1
= d Q / d 6 the probability density function of Q ( 6 )
pressure in a void [Pa]
apparent charge [C]
= ( l / T ) C j qjn($i,q,), the phase angle dependence of average discharge current ($ characteristics) [ C / s ]
phase angle dependence of the average magnitude
of PD pulse ($ - i j / i i characteristics) [C/number]
the probability that the discharge starts at phase
6 = 6 after the gap voltage reaches v s m a t $ = 60
= ( l / T ) E i , j q,n($i,qj), the average discharge
current [C/s]
period of PD measurements
normalized gap instantaneous voltage
the peak value of the gap voltage [VI and the peak
value of the applied voltage [VI, respectively
residual voltage [VI and the minimum residual
voltage IV]
discharge inception voltage of a gap [VI
the normalized critical gap voltage at which
discharge eventually occurs at infinite time
normalized gap overvolt age
permittivities of the sample and air, respectively
[F/mI
phase angle of the applied alternating voltage
[degl
phase angle of the applied voltage at which PD
occurs [deg]
phase angle of the applied voltage for positive PD
and for negative PD at which SPMD starts to
occur, respectively
number of occurrence of discharge per unit phase
angle [l/deg]
a proportional constant relating X and A V
455
gas
d
6
456
FACOM
Results
M-382
T
NAGOYA UNIVERSITY
V o 1 t m et e r
0
MELCOM
Cd ' =200pF
Software
Differential
Attamtor
Q)
c?
3
.-C
-8
2
>
.d
.-V l mc0
wz
IGURE
- 7
!i
0
Phase angle of applied voltage # ( d e g )
Figure 3.
Concept of PD data acquisition to the computer memory.
for our CAPDMS, appears t o be detected a t 312 h after the initiation of voltage application. This is because
the P D pulses turned into so-called swarming pulsive microdischarnes (SPMD'I which are characterized bv much
"
(I
.IO1
[PCI
0 [PCI
110'
45 7
0 (PC)
500
500
s =21.3/cyc ,le
100
400
QC=74.5 n A
.IO1
=11.8/cycle
=24.8 n A
Q = 6 7 . 0 nA
300
100
200
200
IO0
IO0
200
- 0
-100
-100
-200
-200
-300
-100
-400
-400
-400
-500
-500
-500
(I
[PCI
,101
300
300
200
200
I00
100
500
500
100
400
0 (PCI
*IO'
=o.o4/cycle
ij =0.04nA
- 0
L
I
-100
-200
-200
-300
-300
-400
-400
-500
-500
Figure 4.
Experimental results on temporal change of q5 - q - n a n d q5 - q characteristics of PD for CIGRE
Method I1 electrode system for applying 7 kV a t 60 Hz.
is the average number of PD pulses per
cycle, a n d
is the average charge per pulse. D a t a acquisition time T is 60 s for each figure. L is the
number of divisions of time window in one cycle.
x,
458
a i r : Jbu'I'orr
-90
100
IO0
60
'
40
60
20
0
80
le
60
'
40
20
40
BO
'
-3
IO0
80
80
-40
40
20
0
:-20
-40
-60
-60
-BO
-80
-60
-80
-100 J
-100
Figure 5.
r$ - q - n and r$ - q characteristics for dry air at 100 kPa (left),
N500/
,IO,,.
air: 150Torr
N2:610Torr
air: 60'rorr
N 2 :700Torr
( 0 2 4:% 1
( o2 : 1.7 % )
-80
-63
-c
0 tPC1
40
-90
80
10
-3
4c
-80
80
IO0
IO0
80
EO
60
-0
40
20
1,
60
40
40
20
20
-"
-20
-40
e
-40
-40
.lo1
20
-20
:740Torr
( 0 2 :0.6%)
N2
air-: Lu'l'orr
le
le
60
0
-45
0 [PCI
.IO1
IO0
80
459
-"
-20
-40
-60
-60
-60
-80
-80
-80
-100
- io0
- IO0
Figure 6.
Effects of partial pressure ratios of oxygen a n d nitrogen inside a void on 4 - q - n and 4 - q characteristics. Ratios of dry air a n d nitrogen gas are 150:610 (left), 60:700 (center), a n d 20:740 (right),
respectively.
resemble in shape with those a t 96 h as shown in Figure 4, in which SPMD took place, consistent with Izeki's
suggestion.
When the partial pressure po of oxygen is high, there
seems almost no change in the P D characteristics. The
shape of distribution begins to change when the partial
pressure of oxygen goes down to a critical value. It was
also observed that further decrease of po brought the ordinary P D pulses into SPMD. Note that the P D distribution pattern similar to that for 25 h aging time shown in
Figure 4 were observed over a relatively long period from
10 to 60 h after the initiation of voltage application. On
the other hand, no matter what the compound ratio of
dry air and nitrogen gas was, any P D distribution similar
to that a t 25 h aging time could not be observed. The
reason for that is that the degradation of the sample surface and/or the resultant residual materials affect the P D
characteristics for long time aging.
SIMULATION OF PARTIAL
DISCHARGE DISTRIBUTION
N the previous Section, we have shown P D distribution
for the CM-I1 electrode system. This Section will de-
Figure 7.
Whitehead's equivalent circuit for PD. C, is the
capacitance of a void, a n d c
, a n d c b are, respectively, the capacitances of insulation in parallel
and in series with a void.
460
100
h
50
5 -50
2 -100
Figure 8.
Relation between sparking voltage 'U. and residual
voltage ' u ~
of a gap.
-150
Figure 10.
Experimental results of various phase characteristics of SPMD at aging time of 650 h. Detection
sensitivity is 0.3 pC for measurements of SPMD.
.(3)
The FACOM Fortran SSL 11, a scientific subroutine library, is used to generate random numbers.
Figure 9.
Throughout the calculation, we normalize the gap instantaneous voltage w when no discharge occurs as
w = sin4
(1)
Note that all voltages used in the simulation are normalized by Equation (1). The apparent charge q of discharge
is known, and given by
= C,(% - 0,)
where c, is the capacitance of a void, cmand c b are,
respectively, the capacitances of the insulator parallel and
461
1
w
m
J
3
J
3
e o
90
-5
180
270
360
PHRSE ( O E G I
(VI
4-9-"
a0
-0
yl
>
\
9- n
>
-80
2
I
>
- 0
40
3
-I
-2
-3
-4
-5
g a p voltage
Figure 11.
Simulation of PD distributions during a period of 3600 cycles of the applied voltage for XU = 5 . 6 lO-'/deg.
~
(4)
462
4 by which no discharge
(5)
(6)
cos
cos
eo - (e
+ (e - Bo)3(sine,)
6
(13)
--
02(cose,) - In( 1 - U )
2
XO
(15)
>
:eom
-.
-AV
11
03(sin80) - 02(cos6,)
1- ~ X P [ M
2
Q(4,eo)
( n o
2
>
e o . ,
gap voltage
Figure 12.
Simulation of PD distributions for Xo = 5.6 x 103/deg.
c
t
160
463
- -
- 320-
m0
$-q/n
(16)
v, = 0 . 3 7 ~ ~
(18)
Next, let vp be the peak value of a voltage applied between
the electrodes, one obtains from Figure 7
464
large overvoltage
AV 1
small overvoltage
AV2
Figure 16.
Predicted current vs. voltage characteristics for
discharge column of PD.
+++
Figure 14.
Conceptual illustration of growth of electron avalanche for different overvoltages AV, a n d Avz,
where AV, > A V Z .
t
Figure 17.
Predicted current i a n d voltage vfp of the resistance 2 as a function of time.
Figure 15.
Simplified equivalent circuit for discharges of a
void gap. C,, is the capacitance seen from the
discharge gap C,, Z is the resistance of the discharge column a n d v , is the voltage across the
gap.
vp =
If we express
c b
and
up = 0.29Vp
c b
(C, + Cb)
(20)
= 3.3~,S/l(mm)
(21)
C, = ~,S/125(pm)
(22)
where the relative permittivity of epoxy resin is 3.3. Using Equations (19)' (21) and (22)' one can obtain
(23)
465
this assumption cannot explain the obtained experimental results of the P D distribution where the ordinary pulsive PD and SPMD are observed simultaneously in one
alternating cycle. This discrepancy suggests that Xu is
not constant but varies with phase angle of the alternating voltage cycle. Physically, there is a possibility that
the mixture proportion of 0 2 present in the void decreases
due to electron attachment to oxygen molecule after the
first PD pulse takes place a t the beginning of each alternating half cycle. The modified simulation, with A0
varying, is under investigation.
b
Figure 18.
A CONSIDERATION ON
PHYSICS OF PARTIAL
DISCHARGE MECHANISM
Assuming that for the CM-I1 electrode system the discharge extends up to a state corresponding to maximum
normal glow discharge, the normal glow maintaining voltage is taken as v,.,in. Because the maintaining voltage
is close to the minimum sparking voltage, which is about
300 V for nitrogen and air, we take v,,;~ = 300 V. Normalization of the voltage gives v,.~;,, = 0.1.
RESULT OF SIMULATION
IGURES 11 and 12 show typical results of the computed PD distributions during a period of 3600 cycles of the applied voltage for two different values of Ao.
Note that in these Figures as the magnitude of charge
V ( V )is used instead of q ( C ) . This is because in the simulation the voltage waveform is normalized as v = s i n 4
(Equation (l)),and thus q is expressed by q = U' , - v,.
(Equation (2)). On the other hand, for comparison, Figure 13 shows the measured PD occurrence distributions
a t the beginning stage of aging. Comparison of Figure 11
(A, = 5 . 6 lO-l/deg)
~
with Figure 13 indicates that our
simulation gives similar PD distribution to that obtained
by the experiments. The results of the simulation also indicate that when the parameter A 0 deciding the number
of occurrence of discharge per unit phase angle increases,
as shown in Figure 12 (A" = 5.6 x 103/deg), the various
kinds of phase characteristics computed become to resemble those of the experimental results for SPMD shown in
Figure 10.
As shown above, it is concluded that the simulation based
upon the described PD model gives a reasonable agreement with the PD characteristics a t early aging time and
also in the case of SPMD for a dry-air void system. The
simulation assumes that the coefficient A0 is constant in
one alternating voltage cycle. It is noticed, however, that
HERE
466
RESIDUAL VOLTAGE
It is generally recognized that P D stops when a reverse
field is formed by the accumulation of charges on an insulation surface and thus discharge cannot be sustained.
However, in our model, only this consideration does not
give a satisfactory explanation to the residual voltage. AIthough including several assumptions, three attempts are
made to qualitatively explain how the residual voltage is
determined. The first two (1 and 2) consider the growth
of discharge and the third (3) considers the division of
discharge area.
(1) E X A M I N A T I O N O F v, FROM THE
GROWTH O F DISCHARGE
Let us illustrate the discharge in a void by a more simplified equivalent circuit as shown in Figure 15. In this
Figure, C,,, the capacitance seen from the discharge gap
C,, is given by
Cgr
= Cg
+ CmCb/(Cm + Cb)
(24)
and z is the resistance of discharge column and is expected to show a negative resistance characteristics as
shown in Figure 16. Well-known current vs. voltage (i-v)
characteristics of discharge channels a t low pressures for
steady state do not hold directly for temporary discharges
such as P D a t atmospheric pressure. There is however,
a possibility that the i - v g characteristics correspond
to the transition state from the former glow state to the
normal glow state. When the switch S gets closed (Figure 15), which means the onset of discharge, the current
i(t) flowing through z is expressed as
i(t)= -C,,(dv,(t)/dt)
(25)
CONCLUSION
of P D were made for the CM-I1 electrode system, a representative closed-void model electrode system by using the developed computer system.
Measuring temporal change of the PD pulse distribution,
the statistical properties including the phase information
of PD characteristics were examined. Effects of the pressure and gas inside the void on the PD properties were
also examined.
Results obtained are summarized as follows:
1.As aging time increased, the PD pulses occurring a t
the slower phase and those with the greater magnitude of charges began to disappear earlier. Finally, the
PD pulses turned into the so-called swarming pulsive
microdischarges (SPMD).
2. The P D distribution began to change when the partial
pressure of oxygen present in the void went down to a
critical value. Further decrease of partial pressure of
oxygen brought the P D pulses into SPMD.
Taking into account the experimental results, a model for
the mechanism of P D was proposed. The model assumes
that the statistical time lag of discharge depends linearly
on the overvoltage and that the residual voltage depends
on the charge of PD. With Monte Carlo simulation of the
PD distribution, it was shown that the computed results
well explains the experimental data on the P D distribution and the appearance of SPMD.
The physics of the model was also discussed. Consequently, it was suggested that the electron attachment
in oxygen gas greatly influences the time lag of the discharge, and that the size and area of the discharge give
significant contributions to the residual voltage.
EASUREMENTS
467
ACKNOWLEDGMENTS
HE authors are grateful to Mr. Yutaka Higashimura
of Hitachi Research Laboratory, Hitachi Ltd. for providing CIGRE Method I1 electrode system and to Japan
Ciba-Geigy Co. for providing epoxy resin samples. This
work was supported in part of a Grant-in Aid for Scientific Research from the Ministry of Education, Science
and Culture.
REFERENCES
[l] J . H. Mason, Di~charges,IEEE Trans. on Elec.
Insul., Vol. 13, pp. 211-238, 1978.
468