IEEE Transactions on Dielectrics and Electrical Insulation Vol. 12, No.

6; December 2005 1249

Modeling Metal Oxide Surge Arrester for the Modern

Polarization Based Diagnostics
T. K. Saha
School of Information Technology and Electrical Engineering,
University of Queensland, Australia

and K. P. Mardira
Energy Australia, Newcastle, Australia

ABSTRACT
Recently a number of new non-destructive diagnostic techniques have been investi-
gated for the reliable condition assessment of the ageing of metal oxide surge ar-
rester (MOSA). Among them polarizationr rdepolarization current and return volt-
age measurement techniques showed very promising results. This paper presents
an insulation polarization model for investigating polarization-based diagnostics
for metal oxide surge arrester. The simulations of the proposed model are vali-
dated and verified by comparing the data from the polarizationr rdepolarization
current measurements. The simulation of return voltage for MOSA is also per-
formed to validate the insulation polarization model. Finally, the usefulness of this
model for MOSA diagnostics has been highlighted in this paper.
Index Terms — Metal oxide surge arrester, ageing and diagnostics, polariza-
tionrrdepolarization current, return voltage.

1 INTRODUCTION number of publications w1-4x. Results from these diagnos-

tics cannot be interpreted confidently unless the exact re-
G APLESS Metal Oxide Surge Arresters ŽMOSA. pri-
mary function is to protect all equipment in the sys-
tem against various electrical overstress. It is known that
lationship between the measured dielectric parameters
and the fundamental dielectric processes of insulations is
current-voltage characteristics of metal oxide varistors be- clearly understood.
come degraded due to the continuous application of ac or Surge protection of transmission and distribution sys-
due to transients with currents larger than the varistors tems are commonly examined based on the results of their
ratings. To test for any such degradation manufacturers network simulation of their responses subject to different
normally apply single pulse 8r20 ␮ s lightning transients at electrical overstress. Correct simulation needs an accurate
the rated current of the device. However, it is now known model of the network elements and this necessitates an
that most lightning ground flashes consist of multiple re- accurate model for Metal Oxide Surge Arrester ŽMOSA..
turn strokes. The simulation of the dielectric processes gives the possi-
bility to explain some exact relationships that may help to
Many investigations of field aged cables and transform-
interpret the test results correctly. An effort has been
ers have reported meaningful interpretations of the condi-
made in this paper to understand the dielectric response
tion of the insulation by using new diagnostic techniques
of MOSA as related to its fundamental structure and its
such as- return voltage measurement and polarizationrde-
behavior to external electric field. A model of the insula-
polarization current measurement. Since a MOSA is an
tion structure of MOSA based on the principles of linear
insulator at normal conditions Žbelow its rated voltage.,
dielectric response has been derived. The model pre-
the effect of polarization and depolarization of its dipole
sented here allows quantitative analysis of the dielectric
within the insulation can be monitored by any of these
response of a MOSA.
polarization-based diagnostics. These new diagnostic tech-
niques have been successfully applied to a number of arti- More specifically, the aim here is to determine the insu-
ficially degraded MOSA’s and have been reported in a lation equivalent model of MOSA, which could be useful
to predict the degree of degradation. The simulations of
the proposed models are validated using MATLAB based
Paper recei®ed on 23 No®ember 2004, in final form 11 March 2005. software. Effectiveness and accuracy of the proposed

r$20.00 䊚 2005 IEEE

r05r
1070-9878r
1250 Saha and Mardira: Modeling Metal Oxide Surge Arrester for the Modern Polarization Based Diagnostics

model are verified by comparing the data from the polar-

izationrdepolarization current measurement with the re-
sults from polarization model. The simulations of return
voltage on the polarization models are also performed us-
ing MATLAB to validate the insulation polarization
model. The effect of lightning current impulses on MOSA’s
polarization model is also discussed in this paper in detail.

2 THEORY
2.1 POLARIZATION/DEPOLARIZATION
CURRENT [5]
Assuming a homogeneous electric field EŽt. is applied
to the dielectric material, the current density through the Figure 1. Polarizationrdepolarization current curves.
surface of the material can be written as
dE Ž t . d t
Jp Ž t . s ␴ E Ž t . q ␧ 0 ␧r q␧0 H0 f Ž t ᎐ ␶ . . E Ž ␶ . d␶ Ž 1. The polarization Žcharging. current through the object can
dt dt thus be expressed as
where ␴ is the dc conductivity, ␧ o is the vacuum permit- ␴
tivity, ␧ r is the relative permittivity of the insulation at i p Ž t . sC g U0 q f Žt. Ž 5.
␧0
power frequency and f(t) is the response function of the
material. For a homogeneous material, the field strength Once the step voltage is replaced by a short circuit, a
E(t) can be considered as generated by an external volt- depolarization current builds up. The depolarization cur-
age UŽt., where EŽt. s UŽt.rmeterŽm.. To convert cur- rent is expressed as in equation Ž6.. The process of polar-
rent density to a current equation, EŽt. in equation Ž1. is izationrdepolarization is described in Figure 1.
replaced by UŽt.rm and equation Ž1. is multiplied by the
geometric factor of the insulation structure. If the capaci- i d Ž t . s ᎐ C g Uo f Ž t . ᎐ f Ž t q tp . Ž 6.
tor configuration consists of parallel plate electrodes of
It was found that the response function f(t) for the
area A and spacing d sandwiching a dielectric of permit-
MOSA could be expressed in a general expression follow-
tivity ␧ , the capacitance of the test object is given by equa-
ing the universal relaxation law as in equation Ž7., which
tion Ž2.
is also found in many other experimental observations w6x,
C s Ž ␧ Ard . s Ž ␧ r ␧ o Ard . s ␧ r Ž ␧ o Ard . s ␧ r C g Ž 2 . where, m and n are constants.
where C g s geometric capacitance s Cr␧ r s Ž ␧ o Ard. f Ž t . s mt ᎐ n Ž 7.
s ␧ o ŽGeometric Factor.; and geometric Factor s Ard
s C g r␧ o Žunit m.. In order to estimate the dielectric response function f(t)
from a depolarization current measurement it is assumed
Then the polarization current is given by equation Ž3.
that the dielectric response function is a continuously de-
␴ dU Ž t . d t creasing function in time, then if the polarization period is
i p Ž t . sC g U Ž t . q ␧r q H0 f Ž t ᎐ ␶ . .U Ž ␶ . d␶ sufficiently long, so that f(t q t p ) ( 0, the dielectric
␧0 dt dt
response function f(t) is proportional to the depolariza-
Ž 3. tion current. Thus from equation Ž6., the dielectric re-
sponse function f(t) can be approximated as in equation
If the applied voltage is a step voltage as given in equa- Ž8..
tion Ž4. ᎐ idŽ t .
f Žt.f Ž 8.
°0 t-0 C g Uo
U Ž t . s~U 0 0 F t F tc Ž 4.
¢0 t ) tc
From the measurements of polarization and depolariza-
tion currents, it is possible to estimate the dc conductivity
This will give zero current for times before t s 0, and ␴ of a test object. If the test object is charged for a suffi-
so-called polarization currents for times 0 F t F t c . Due ciently long time so that f(t q t p ) ( 0. Equations Ž5.
to a constant step voltage, the displacement component of and Ž6. can be combined to express the dc conductivity of
the current is zero. Hence, the polarization current is built the dielectric as
up in two parts. One part is related to the conductivity of ␧0
the test object and the other is related to the activation of ␴f Ž i Ž t . ᎐ idŽ t . . Ž 9.
the different polarization processes within the test object. C g U0 p
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 12, No. 6; December 2005 1251

be numerically solved provided f(t) and the conductivity

␴ of the dielectric material are known.

d
j Ž t . s ␴ Er Ž t . q ␧ 0 ␧ r Er Ž t .
dt
d t
q␧0
dt ½H t2
5
f Ž t ᎐ ␶ . Er Ž ␶ . d␶ s 0 Ž 11 .

If f(t) can be parametrised from the curve fitting of de-

Figure 2. Return voltage phenomena. polarisation current then conductivity can be estimated
from equation Ž9.. However, equation Ž11. can not be
solved analytically and some numerical integration tech-
2.2 RETURN OR RECOVERY VOLTAGE nique can be used for the return voltage estimation pro-
vided appropriate relative permittivity value of the metal
The return voltage is based on the polarization and
oxide is accurately known. In reality, relative permittivity
subsequent depolarization of dipoles within the insulating
of the metal oxide can be only accurately estimated by
material as well as on the charging and discharging of grain
frequency domain dielectric loss factor measurement.
boundaries and space charge effect w2x. Figure 2 shows the
Hence, an alternate electrical circuit based model has been
typical steps of a return voltage measurement.
proposed to explain the polarization characteristics of
The return voltage measurement comprises three steps: transformerrcable dielectric materials by a number of re-
䢇 Charge the tested object for a pre-selected time Žt .
c searchers. This circuit based model has been applied in
with a dc voltage ŽUo ., which is much lower than the rated this paper to explain the MOSA characteristics due to the
voltage. application of an external electric field.
䢇 Discharge the tested object for a short period Žnor-

mally half of the charging time, t d s 0.5 t c ..

3 POLARIZATION EQUIVALENT
MODEL FOR METAL OXIDE SURGE
䢇 Measure the open circuit voltage built up across the
ARRESTERS-DEBYE APPROACH
object ŽUr ..
In the presence of an electric field, E, there are two
Three important parameters may characterize the con- main leakage current components that exist in a dielec-
dition of insulation. They are maximum return voltage tric. They are conduction and polarization currents. The
ŽUrmax ., time to reach maximum return voltage Žcentral conduction and polarization currents both will be affected
time constant ŽCTC- t m . and the initial slope Žslope of by any changes to the microstructure. The polarization
return voltage curve for first few seconds S r s durdt. w3, current is due to the tendency of dipoles to align in the
4x. direction of the field.
The polarization processes, which are not completely The polarization process is described as a process of
relaxed during the grounding period, will relax and give energy storage and only exists in the presence of an exter-
rise to a recovery voltage across the electrodes of the insu- nal field. When the field is removed, the dipoles relax and
lation. The current density during the return voltage mea- then return to the original state after a long time w12x. An
surement is zero and equation Ž10. gives the expression of insulation polarization model can be represented by an
current density. Where E r Žt. is the electric field resulting equivalent model comprising the insulation resistance and
from the return voltage build up Ur across the open cir- geometric capacitance with a number of additional R-C
cuited dielectric. branches w8, 12x, as shown in Figure 3. The parallel R-C
branches represent the individual polarization processes
d d t with their time constants R pn C pn and can be treated in-
j Ž t . s ␴ Er Ž t . q ␧ 0 ␧ r Er Ž t . q ␧ 0 ½H Ž f t ᎐ ␶ . Er Ž ␶ . d␶ 5 dependent to each other. The representation of the polar-
dt dt 0
ization processes by parallel R-C branches allows follow-
Ž 10.
ing the theory of distributed relaxation times to predict
the response of the insulation, which hardly follows ideal
Debye behavior w6x. The number of branches in most
For t 2 - t - ⬁, i.e. during the return voltage measure- practical modeling purposes varies from four to ten de-
ment period, the current is zero Žopen circuit condition of pending upon the nature of the depolarization current.
insulation . iŽt s t 2 . s 0. Also the current density jŽt. The range of the values of the individual time constants,
being zero during the return voltage measurement. Equa- R pn and C pn are found to be dependent upon the insula-
tion Ž10. can be re-written as equation Ž11. and E r Žt. can tion condition.
1252 Saha and Mardira: Modeling Metal Oxide Surge Arrester for the Modern Polarization Based Diagnostics

Figure 4. Polarization model for an insulation network.

Figure 3. Equivalent circuit of insulation for polarization model.

final value given by equation Ž12.
VC pk Ž t c . s V Ž 1᎐exp ᎐ t crR pk C pk . Ž 12.
The presence of the polarization introduces a permittiv-
ity ␧ r in the field equation D s ␧ o ␧ r E. The polarization The switch is then closed to position 2 at t )10000
is not observable but we can study its time development seconds with pre-selected depolarization time Žt d . of
by measuring the current. The current density is the sum 10,000 s. The corresponding depolarization branch cur-
of the conduction and displacement currents. The con- rent ŽI dk . is given by
duction current in the insulation is due to a leakage or ᎐ Ž t ᎐ tc .
internal resistance R g while the displacement current is
due to the geometric capacitance C g , internal resistance Idk Ž t . s
VC pk exp
ž ␶k / Ž 13.
R g and the pairs ŽR pn , C pn . with the associated time con- R pk
stant given by ␶n s R pnC pn .
where ␶ k s R pk C pk is the individual elements ŽR pk ᎐C pk .
In this paper, an attempt is made to identify the polar- corresponding time constant. The total depolarisation
ization model of MOSA by means of polarizationrde- current comprises of summation of various relaxation
polarization current measurement. The aim is to investi- mechanisms that appears at different locations within the
gate the MOSA polarization model parameters, which insulation. Various parts of the insulation have their
could be useful to determine the condition of its insula- unique relaxation characteristics depending upon the age-
tion after various electrical stresses particularly lightning ing condition. The relaxation or depolarization current can
impulse current pulses. The calculation of model parame- thus be modeled as the sum of exponentials of the various
ters was performed using MATLAB. relaxation mechanisms.
n
IdT Ž t . s Ý Idk Ž t . Ž 14 .
4 IDENTIFICATION OF THE k s1
EQUIVALENT CIRCUIT USING - The individual elements R pk ᎐C pk with the correspond-
POLARIZATION CURRENTS AND ing time constants ␶ k s R pk C pk can then be determined
DEPOLARIZATION CURRENTS by fitting the depolarization current with the following
The polarizationrdepolarization currents are based equations
on an application of a dc voltage to an insulator to estab- n ᎐ Žt ᎐ tc .
.
lish polarization, and then the removal of the charging
voltage to relax the dipoles Ždepolarization current .. The
IdT Ž t . s Ý
k s1
ž A k eŽ ␶k / Ž 15.

equivalent polarization model can be identified by calcu- where

lating the slope of depolarization current on each R-C ᎐ tc
.
branch and comparing it to the slope of the total depolar- V 1᎐ e Ž
ž ␶k /
Ak s Ž 16.
ization current w9x. Figure 4 shows the polarizationrde- R pk
polarization current measurement set up on an insulation
network.
The modeling process starts with the largest time con-
During the application of dc voltage V across the di- stant branch. The depolarization current at longer time
electric for a pre-selected polarization time Žt c ., in this can be assumed due only to the largest time constant
case for up to 10,000 s with the switch closed in position 1 branch, with the influences of the rest of the smaller time
Žas in Figure 4., the k-th capacitor C pk is charged with a constant branches dying down well before that time.
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 12, No. 6; December 2005 1253

Therefore, the final part of the depolarization current is

used to find out the values of ␶ k corresponding to the
largest time constant branch using an exponential curve
fitting technique w9x. Once the exponential component with
the largest time constant is identified, it is then subtracted
from the depolarization current to go to the next level. In
this level, the final part of the resultant current curve is
influenced by the second largest time constant only, with
the next smaller time constant branches practically going
to zero well before that time. Following the same curve
fitting technique the value of ␶ k corresponding to the sec-
ond largest time constant branch is identified. The identi-
fication of the next R-C branches can be achieved by fol-
lowing the same procedure. Once the values of ␶ k corre-
sponding to different time constant branches are identi-
fied, the values of R pk and C pk can be easily separated
and the equivalent model can be constructed.

5 EXPERIMENTAL SET-UP
Figure 6. RVM circuit diagram.
New distribution class metal oxide surge arresters of
10 kA ratings used for this study were commercial devices
produced by one particular manufacturer. There were two and 2 were performed. This was done to eliminate the
different types of arresters- Ži. a double block type labeled previous polarization effects Žcommonly known as mem-
D and Žii. a single block type labeled S. The arrester D ory effects ., which normally affect the accuracy of the
had rated voltage of 12.5 kV while arrester S consisted measurement.
only one block and had rated voltage of 6.3 kV. The Ex-
perimental procedure is summarized as follows: 5.2 ARTIFICIAL DEGRADATION
PROCESSES
5.1 BEFORE DIAGNOSTICS All MOSA’s were degraded by lightning current im-
1. Polarization and depolarization currents were mea- pulses. Arresters D and S were subjected to 5 groups of
sured with 1000 Vdc and polarization and depolarization multi-pulse current at 2 p.u. of the rated current with small
time was 10 4 s. The measurement circuit diagram is shown time intervals required to charge the system. The multi-
in Figure 5. pulse current test consisted of quintuple Ž5. 8r20 ␮ s light-
ning current impulses with a difference of 20-40 ms be-
2. Return voltage measurement ŽRVM. was performed tween each pulse w10x.
on all MOSA varistors with 1000 Vdc , 200 s charging time
and 100 s discharging time. The RVM circuit diagram is 5.3 AFTER DIAGNOSTICS
shown in Figure 6.
Steps 1 and 2 were repeated on the artificially de-
Two terminals of the arresters were short circuited to graded arresters to see the changes and will be referred as
ground for at least 24 h before the measurement steps 1 ‘‘after diagnostic’’ measurements.

6 POLARIZATION MODEL
VALIDATION
The available values from the measured polarization
current are the total polarization and depolarization cur-
rents with respect to time, insulation resistance R g Žfrom
the difference between polarization and depolarization at
larger values of time.. The capacitance, C, was measured
using a Schering bridge at the power frequency. The rela-
tive permittivity, Ž ⑀ r , at power frequency is typically 10 for
ZnO varistor w11x and it was used to calculate geometric
capacitance Cg by equation Ž2.
During the depolarization process the capacitance C pk
is discharging through the resistance R pk . Each branch
Figure 5. PDC measurement. current is modeled by an exponential function as de-
1254 Saha and Mardira: Modeling Metal Oxide Surge Arrester for the Modern Polarization Based Diagnostics

Figure 9. Polarization current measurements on arrester D for be-

Figure 7. Polarization current measurements on arrester S for be- fore and after application of lightning current impulses.
fore and after application of lightning current impulses.

scribed in equation Ž13.. The software starts calculation at

the largest time constant. In Figure 7, x axis is time in
logarithmic scale and y axis is in linear scale. A straight
line is fitted with the last few points of the measured de-
polarization current data. The slope of the line is calcu-
lated from the fitted straight line. Once the slope is calcu-
lated from equations Ž17. and Ž18., equations Ž16. and Ž18.
can be used to calculate R pk and C pk .
The slope of depolarization current between last two
selected points t d s t 1 and t d s t 2 , can be calculated by
equations Ž17. and Ž18.
ln IdT Ž t 2 . ᎐ ln IdT Ž t 1 .
n n
s ln Ý Idk Ž t 2 . ᎐1n Ý Id k Ž t 1 .
Figure 10. Depolarization current measurements on arrester D for
k s1 k s1
before and after application of lightning current impulses.
᎐ t1 ᎐ t2

( ln
Vc pk exp
ž /
␶k
᎐ ln
Vc pk exp
ž /
␶k
Ž 17. t 2 ᎐ t1
Rk Rk ⬖ slope( R pk C pk ( Ž 18.
ln IdT Ž t 2 . ᎐1n IdT Ž t 1 .

where 0 - t 2 - t 1 - 10000 s.
First the values of R-C’s are estimated from the simula-
tion methodology as described in this section and then
polarizationrdepolarization current was simulated based
on these simulated circuit parameters. Finally, the valida-
tion was performed by comparing the measurement re-
sults of the polarizationrdepolarization current curves and
the simulation results of the polarization model.
Polarizationrdepolarization currents of S and D ar-
resters are plotted in Figures 7, 8 and 9 and 10, respec-
tively for before and after application of lightning current
impulses. The polarization models Žbefore and after the
application of lightning current impulses. showed good
accuracies compared to the measurement results at the
long constant polarization and depolarization stages. Both
Figure 8. Depolarization currents on arrester S for before and after Figures 11 and 12 show that the arrester S polarization
application of lightning current impulses. models before and after the application of lightning im-
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 12, No. 6; December 2005 1255

Table 1. Polarization model parameters for arrester S before and

after application of lightning current impulses.
Before After
Cg 1.59 = 10y1 1 F 1.62 = 10y11 F
Rg 3.09 = 10 11 ⍀ 2.9 = 10 11 ⍀
R p1 7.45 = 10 11 ⍀ 6.05 = 10 11 ⍀s
R p2 3.03 = 10 11 ⍀ 1.32 = 10 11 ⍀
R p3 8.02 = 10 10 ⍀ 1.73 = 10 10 ⍀
R p4 1.29 = 10 11 ⍀s 2.01 = 10 11 ⍀s
C p1 1.25 = 10y10 F 1.33 = 10y11 F
C p2 3.71 = 10y11 F 7.27 = 10y11 F
C p3 3.66 = 10y11 F 7.28 = 10y11 1 F
C p4 2.75 = 10y12 F 3.02 = 10y11 F

impulses. The comparisons of simulation results to the ac-

tual polarizationrdepolarization current curves on ar-
rester D are shown in Figures 13 and 14.
Figure 11. Comparison of measured polarizationrdepolarization Both Figures 13 and 14 showed that the simulation re-
current on arrester S before the application of lightning current im- sults showed good comparisons with the measured polar-
pulses.
izationrdepolarization current on arrester D. The ar-
rester D polarization models before and after the applica-
tion of lightning impulses consist of 4 polarization
branches with each branch having a different time con-
stant. Arrester D model parameters before and after the
application of lightning current impulses are given in Table
2.
The arrester D polarization model parameters showed
that the geometric capacitance and branch capacitances
ŽC p1 to C p4 . increased after the application of lightning
current impulses. Meanwhile, the insulation resistance
ŽR g . and branch resistances ŽR p1 to R p4 . are all reduced.
This might indicate that arrester D has degraded after the
application of lightning current impulses by lowering its
insulation resistance.

Figure 12. Comparison of measured polarizationrdepolarization

current on arrester S after the application of lightning current im-
pulses.

pulses; consist of 4 polarization branches with each branch

having different time constants. Table 1 compares the
model parameters before and after the application of
lightning current impulses.
The comparisons of the model parameters showed that
the insulation resistance ŽR g . and branch resistances ŽR p1
᎐ R p3 . reduced while the geometric capacitance ŽC g . and
branch capacitances ŽC p1 ᎐ C p4 . increased after the appli-
cation of lightning current impulses. These results suggest
that the insulation condition of arrester S have degraded
by lowering its insulation resistance and increasing its ca-
pacitances. These would increase the leakage current dur-
ing MOSA’s normal operation.
Figure 13. Comparison of measured polarizationrdepolarization
The polarization model was also simulated on arrester current on arrester D before the Application of Lightning Current
D before and after the application of lightning current Impulses.
1256 Saha and Mardira: Modeling Metal Oxide Surge Arrester for the Modern Polarization Based Diagnostics

VCpn is achieved when t c s ␶n ln 3, where ␶n s R pn C pn .

This value is obtained by differentiating equation Ž20. and
setting the result equal to zero.
4. As soon as the short circuit is removed, the RV ob-
served will be the resultant of all remaining voltage in all
k capacitors. The measured RV corresponds to the volt-
age across the dielectric given by

Vr Ž t . s Aˆ1 Ž t . Ž 1᎐exp w ᎐ t r␶ 1 x . Ž exp w ᎐ t r2␶ 1 x . q...

qVAˆn Ž t . Ž 1᎐exp Ž ᎐ t r␶n . exp w ᎐ t r2␶n x . Ž 21 .

where:

A i Ž t p . s A i ,1exp Ž t p1 . q...q A i , k exp t p nq1 .

Figure 14. Comparison of measured polarizationrdepolarization
current on arrester D after the application of lightning current im-
pulses. k i ⌸ Aˆ1Ž pj ᎐ z i , j .
Ai, j s
pj ⌸ Aˆk / j Ž pj ᎐ pk .
Table 2. The polarization model parameters for arrester D before
and after application of lightning current impulses.
j,k s1,...,nq1 i,1s1,...,n
Before After
Cg 7.25 = 10y1 2 F 80.6 = 10y12 F In addition, n is the number of polarization branches.
Rg 5.75 = 10 11 ⍀ 4.03 = 10 11 ⍀ In this investigation n s 4 is used since the polarization
R p1 1.13 = 10 12 ⍀ 9.87 = 10 11 ⍀ models for arresters S and D were derived previously
R p2 6.67 = 10 11 ⍀ 3.67 = 10 11 ⍀
R p3 1.55 = 10 11 ⍀ 1.05 = 10 11 ⍀ which consist of only four polarization branches.
R p4 2.52 = 10 11 ⍀ 1.91 = 10 11 ⍀ The return voltage simulation of the polarization model
C p1 1.41 = 10y10 F 1.61 = 10y10 F
C p2 2.63 = 10y11 F 2.83 = 10y11 F was performed using MATLAB by expanding equation
C p3 3.82 = 10y11 F 4.02 = 10y11 F Ž21.. The charging voltage is 1000 Vdc , t c s 200 s and t d
C p4 3.81 = 10y12 F 4.32 = 10y12 F s 100 s. Figures 15, 16, 17 and 18 show the comparison
between the simulated results and measured RV data on
arresters S and D before and after the application of cur-
7 POLARIZATION MODEL rent impulses respectively.
VALIDATION BY RETURN VOLTAGE The return voltage simulation results on Figures 15 to
To validate the polarization model for arresters S 18 show good comparable results to the return voltage re-
and D before and after the application of current im- sults obtained from the measurements. The simulation re-
pulses, the return voltage of each arrester was simulated sults are within 6.5% of the measured data. This proves
and compared to the measured data. The return voltage that arresters S and D polarization models derived previ-
can be described mathematically as follows w12x ously could be used to represent the insulation polariza-
tion parameters of arresters S and D.
1. After the application of dc voltage V across the dielec-
tric for pre-selected time t c Žin s., the n-th capacitor C pn
is charged to a voltage Vc with final value of

2. Vc Ž t c . sV 1yexp y t crŽ R pn C pn .
Ž . Ž 19.

3. The dielectric is short-circuited for t d seconds. The ra-

tio of t d is set to 1r2 of t c in this study. The remaining
voltage in C pn is given by

VC p nŽ t c ,t d . sV 1yexp Ž y t crR p p C pn .
Ž .=
exp Ž y t drR pnC pn . Ž 20 .

Figure 15. Comparison of return voltage simulation on arrester S

For each pair Žt c ,t d ., the voltage in the n-th capacitor polarization model with the measured RV data before the applica-
VCpn achieves a different value. The maximum value of tion of current impulses.
IEEE Transactions on Dielectrics and Electrical Insulation Vol. 12, No. 6; December 2005 1257

additional R-C branches. The calculation of model pa-

rameters is performed and simulated using MATLAB.
The calculated models on arresters S and D before and
after the application of lightning current impulses showed
a comparable good accuracy to the actual measurement
curves Žpolarizationrdepolarization currents and return
voltage.. The comparisons of model parameters indicate
that the application of lightning current impulses degrade
the arresters by lowering the insulation and branch resis-
tances as well as increasing the geometric and branch ca-
pacitances. These would increase the leakage current dur-
ing arrester’s normal operating condition. Therefore,
Figure 16. Comparison of return voltage simulation on arrester S these findings could be useful for the study of the degree
polarization model with the measured RV data after the application of degradation of MOSA after the application of electri-
of current impulses.
cal stresses.
This study will be very useful while comparing the cir-
cuit parameters of suspected arresters collected from the
field with the new ones. Very often field arresters do show
clear-cut evidence of passrfail condition by measuring 1
mA leakage current or lightning impulse residual voltage.
However, these tests do not show any evidence of ZnO
material degradation under lightning currents. We have
previously tested many such arresters with these newer di-
agnostics and found very important results. A comprehen-
sive mathematical tool like this would provide more infor-
mation to predict the condition of field arresters in detail.
This is under investigation in our current project and will
be reported in a future paper.
Figure 17. Comparison of return voltage simulation on arrester D
polarization model with the measured RV data before the applica-
tion of current impulses. REFERENCES
w1x K. P. Mardira, T. K. Saha, and R. A. Sutton, ‘‘Investigation of
Modern Diagnostic Techniques for Metal Oxide Surge Ar-
resters’’, IEEE Trans. Dielectr. Electr. Insul., Vol. 12, pp. 50᎐59,
2005.
w2x C. Heinrich and W. Kalkner ‘‘Return Voltage Measurement on
Metal Oxide Surge Arrester’’, 10 th Intern. Sympos. High Voltage
Eng., Vol. 5, pp. 137᎐40, Montreal, Canada, 1997.
w3x T. K. Saha and T. Dinh, ‘‘Return Voltage Measurements on
Metal Oxide Surge Arresters’’, Intern. Sympos. High Voltage
Eng., London, UK, Vol. 2, pp. 313᎐316, 1999.
w4x K. P. Mardira, M. Darveniza and T. K. Saha, ‘‘Search for New
Diagnostics for Metal Oxide Surge Arrester’’, IEEE 6 th Intern.
Conf. Properties and Applications of Dielectric Materials, Xian,
China, pp. 947᎐50, Vol. 2, 2000.
w5x T. K. Saha and P. Purkait, ‘‘Investigation of Polarization and
Depolarisation Current Measurements for the Assessment of
Oil-paper Insulation of Aged Transformers’’, IEEE Trans. Di-
Figure 18. Comparison of return voltage simulation on arrester D electr. Electr. Insul., Vol. 11, pp. 144᎐154, 2004.
polarization model with the measured RV data after the application w6x A. K. Jonscher, Dielectric Relaxation in Solids, Chelsea Dielectric
of current. Press, London, UK, 1983.
w7x E. Nemeth, ‘‘Measuring Voltage Response: a Non Destructive
Diagnostic Test Method of High Voltage Insulation’’, IEE Proc.
Sci. Measur. Techn., Vol. 146, pp. 249᎐252, 1999.
8 CONCLUSIONS
w8x E. Nemeth and T. Horvath, ‘‘Fundamental of the Simulation of
In this study, an attempt is made to identify the insu- Dielectric Processes of Insulation’’, 8 th Intern. Sympos. High
lation polarization model of MOSA by means of polariza- Voltage Eng., Budapest, Hungary, pp. 177᎐180, 1993.
tionrdepolarization current measurement. The model can ¨
w9x T. K. Saha, P. Purkait and F. Muller, ‘‘Deriving an Equivalent
Circuit of Transformers Insulation for Understanding the Di-
be represented by an equivalent model comprising a cir- electric Response Measurements’’. IEEE Transactions on Power
cuit used to describe dielectric materials with a number of Deli®ery, Vol. 20, pp. 149᎐157, 2005.
1258 Saha and Mardira: Modeling Metal Oxide Surge Arrester for the Modern Polarization Based Diagnostics

w10x M. Darveniza, C. J. Andrews, D. R. Mercer and T. M. Parnell,

‘‘A Multiple Lightning Impulse Generator’’, Int. Sympos. High
Voltage Eng., New Orleans, paper 47.07, 1989.
w11x L. M. Levinson, and H. R. Philipp, ‘‘Application and Character-
isation of ZnO Varistors’’, in Ceramic Materials for Electronics -
Processing, Properties and Applications ŽDuchanan, R. C. ed,
Marcel Dekker Inc, New York. pp. 375᎐402, 1986.
w12x P. R. S. Jota, S. M. Islam, and F. G. Jota, ‘‘Modeling the Polar-
ization Spectrum in Composite OilrPaper Insulation Systems’’,
IEEE Trans. Dielectr. Electr. Insul., Vol. 6, pp. 146᎐151, 1999.