2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA)

Characterization of Phase Resolved Partial Discharge

Waveforms from Instrument Transformer using
Statistical Signal Processing Technique
Yasmin Hanum Md Thayoob #1, Syed Khaleel Ahmed #2 , Chua Chun Piau #3 , Chong Yu Ping #4 Yogendra
Balasubramaniam*5
#
College of Engineering, University Tenaga Nasional
Jalan IKRAM-UNITEN, 43000 Kajang, Selangor, Malaysia
1
yasmin@uniten.edu.my
2
syedkhaleel@uniten.edu.my
3
piau89_uniten@hotmail.com
4
cyping89@hotmail.com

*
TNB Research Sdn. Bhd.
Lorong Ayer Itam, 43000 Kajang, Selangor
5
yogendra@tnbr.com.my

Abstract—Partial Discharge (PD) detection is one of the IT. The failures due to PD may result into violent explosions
important methods used to evaluate the insulation condition of that cause personal injuries and damage to nearby equipment.
in-service high voltage (HV) electrical equipment. On-site PD This may then require high equipment replacement costs and
detection methods are preferred compared to other methods as increases unexpected expenditure [3, 4]. Therefore, condition
non-intrusive sensors are easier to install and have no
assessment of in-service high voltage instrument transformer
interruptions on the operation. An electrical PD testing utilizing
High Frequency Current Transformer (HFCT) installed at the should be carried out to assess the condition of the IT in order
ground terminal of the in-service instrument transformer was to avoid unexpected failures.
used to detect partial discharge activities. However, the limitation Electrical partial detection (PD) detection is one of the
of this method is the measurement often lacks of accuracy due to important methods used to monitor and evaluate the condition
interference in the measured PD waveforms. Therefore, such PD of in-service high voltage electrical equipment. On-line partial
signals are hard to be distinguished from the stochastic noise discharge testing is performed during real operating condition
pulses. Hence, the characteristics of PD signals and interference with normal operating voltage. This is a non-destructive test;
signals are studied in the Phase Resolved Partial Discharge therefore non-intrusive sensors which are easy to install and
(PRPD) pattern in order to identify the extent of PD occurrence
require no interruptions on the operation of the high voltage
in the signals. The discharge magnitude could increase
significantly because of the noise signals superposed with PD (HV) equipment are preferred for on-site PD detection. Hence,
signals. Therefore a method that can remove high magnitude in this research, electrical PD test utilizing High Frequency
noise will be proposed by using statistical signal processing Current Transformer (HFCT) installed at the ground terminal
technique. of the instrument transformer was used to detect partial
Keywords—partial discharge; instrument transformer; discharge activities [5].
inteferences; phase resolved partial discharge pattern; statistical As defined by IEC60270, ‘Partial discharge is a localized
signal processing technique electrical discharge that only partially bridges the insulation
between conductors and which can or cannot occur adjacent to
I. INTRODUCTION a conductor [6]. PD usually started within voids and cracks.
Instrument transformers (IT) are used to transform voltage Since PDs are limited to only a portion of the insulation, the
or current from high to low values that can be utilized by discharges only partially bridge the distance between
electrical equipment for monitoring or metering purposes in electrodes. PD can also occur along the boundary between
transmission or distribution systems. It is also designed to different insulating materials. The electric field across the void
safely isolate the utilization current or voltage for secondary is significantly higher than that across an equivalent distance of
end device from the primary supply voltage [1, 2]. IT plays an dielectric due to the dielectric constant of the void is
important role in electrical system; however, IT in services are considerably less than the surrounding dielectric. If the voltage
exposed to stresses, such as electrical, thermal and mechanical stress across the void is increased above the corona inception
stresses that can cause degradation or deterioration of the voltage (CIV) for the gas within the void, PD activity will start
insulating system in IT. Those stresses and environmental within the void [7].
condition will affect the performances of IT and lead to failure. In this research, electrical PD detection is used to monitor
Partial discharge (PD) is one of the cause of severe failure in the status of IT. However, PD measurement performed on-site

978-1-4799-8996-6/15/$31.00 ©2015 IEEE

355
 2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA)

and under noisy condition is a challenging task. The this characteristics can also be used for further analysis of the
measurement often lacks of accuracy due to interference in PD PRPD pattern obtained from HFCTs.
signals [8]. PD signals are often superposed with stochastic
noise pulses or even multiple PD sources, which leads to a
complex phase resolved PD patterns that are not easy to
analyze and distinguish. Therefore, characteristics of PD and
noise in the PRPD patterns are studied in order to differentiate
between PD signals and noise.
According to IEEE standard (IEEE Std C57.113-2010),
noise that could be encountered from on-site PD measurement
are harmonics, radio frequency (RF) noise, corona and
sparking which are generated by several different kinds of
sources [9]. Analysis of possible interference patterns is
essential to distinguish PD from interferences in the PRPD
pattern. The discharge magnitude could be increased
significantly because of the noise signals superposed with PD
signals. Therefore, a technique that can remove high magnitude
noise using statistical signal processing will be proposed.
However, in some cases, the noise signals cannot be removed
completely and analysis on the filtered signals will be essential.
Fig. 1. Sequence of cavity breakdown under alternating voltages: (a) voltage
II. PHASE RESOLVED PARTIAL DISCHARGE (PRPD) waveforms and (b) PD Pulses [13]

In PRPD, the phase of occurrence, apparent charge and

number of PDs which has the same phase and magnitude can
be recorded. For data recording purpose, usually an m × n
matrix is used where m is the number of phase channels and n
is the number of charge levels [10]. Each element of this
matrix shows the number of PDs with a particular magnitude
and phase. This method can provide different patterns such as
the phase of occurrence (ij) versus the number of PD (n), the
phase of occurrence (ij) versus the maximum apparent
charge(qm), the phase of occurrence (ij) versus the average
apparent charge (qa) and the apparent charge (q) versus the
number of discharge (n).
Fig. 2. Typical PD pattern in Phase Resolved PD pattern [3]
A. PD Pattern Characteristics
There are some common phenomena or characteristics of B. PD Parameters and the Extent of PD activity
partial discharge in PRPD diagram. PD shall occur only in both
first and third quadrants in the PRPD diagram, PD pulses shall In PRPD analysis, PD parameters such as PD magnitude
increase abruptly at the beginning of first and third quadrants and PD count are very important to determine the extent of
followed by rapid decrease and number of PD pulses in first partial discharge activity. Apparent charge (q), repetition rate
and third quadrants shall not have big difference [11]. Also, it and number of pulses (n) with respect to the phase angle (ș)
is not possible for PD to take place in either first or third were recorded when measuring the PD activity. PD magnitude
quadrants only as PD shall take place in both first and third usually measured in pico Couloumb (pC) while PD count is the
quadrants simultaneously [1-2,7]. Therefore these number of occurrence of PD in one second duration. There are
characteristics can be applied on partial discharge recognition no standard for PD magnitude and PD count to determine the
analysis. extent of PD. However, there are concerns for the safety of the
insulation system if the PD magnitude and PD count are high.
In Fig. 1, PD occurs when voltage across the void, V1 Nevertheless, the high values of the PD parameters can also be
reaches the breakdown voltage, Vi. Voltage V1 reduces until contributed by the influence of signals from noise and
the discharge extinguishes. Then, rises again as the applied interference.
voltage, V rises until new breakdown occurs again. This According to a study on PD conducted for IT in [15], the
situation repeats again and again and eventually causes the apparent charge above 300 pC were considered as abnormal
breakdown of insulation system [12]. and over 1000 pC was considered as extremely high and
Furthermore, PD activity increases at the beginning of immediate actions were recommended. Therefore, in this study,
each half cycle followed by decreases like a bell-shaped curve the values of 300 pC and 1000 pC for the measured apparent
as shown in the PRPD diagram Fig. 2. This pattern is recorded charge would be considered as the threshold values for further
in many published papers about PD in [8,13,14]. Therefore, analysis of the measured data from PRPD pattern to determine
the extends of the partial discharge occurrence in the IT.

356
2015 IEEE International Conference on Signal annd Image Processing Applications (ICSIPA)

III. NOISE IN ELECTRICAL PD DETECTION

E A. PD Parameters
During on-site, PD measurement is alw ways susceptible to Generally, the measured PD P waveforms are displayed in
interference from its surrounding and the meaasured data will be phase-resolved Partial Dischaarge (PRPD) pattern and the
affected by noise signal. The accuracy of measurement reduces common parameters used to determine the extent of PD
because the noise signals may superpose witth or dominate the activity are PD magnitude and number of PD pulses. However,
PD signals. Therefore, the magnitude of PD D will be increased there is no standard reference too determine the extent of PD. In
significantly by the noise signal. The noisse signals can be this research work, the PD waaveforms to be processed were
classified into two main groups, which are im
mpulsive and non- obtained from PD measurem ment and the PD parameters
impulsive interference. acquired are apparent charge (qq), phase angle (ij) and time (t).
The q is usually expressed in piicocoulombs (pC) [16].
A. Impulsive Interference
Impulsive interference includes repetitivee pulse and random B. Statistical Parameters
pulse. Impulsive interference is more diffficult to remove In this study, statistical parameters were applied to
because of its similarity with PD signal in some features. eliminate disturbance or noise in the measured PD waveforms.
Repetitive pulses are usually caused by Corona and switching Mean discharge magnitude (μ)) and standard deviation (ı) will
devices such as an AC/DC converter. The reppetitive pulses can be calculated from the measuured PD waveform and used to
be easily identified because pulses from the same source eliminate discharges using the criteria of two standard
usually have the same features. The pulsess tend to group at deviation with respect to norm mal distribution curve. According
equally spaced and roughly with same maagnitude. Random to the normal distribution curvee in Fig. 3, 68.2% of distribution
pulses are usually generated from swittching operations, lies within 1ı of the mean, 95.44% of the distribution lies within
welding, sparking and so on. Unlike repetitiive pulses, random 2ı of the mean and 99.8% of the t distribution lies within 3ı of
pulses are not necessarily to have pulses from
m same source with the mean [18]. The dischargges, q that are not within the
same different time moments. Howeverr, the frequency
selected tolerance interval will be eliminated. Hence, q smaller
distribution of random pulses that come from m the same source
than the value of μ - 2ı and q greater
g than the value of μ + 2ı,
must be highly similar and it is different from
m other pulses that
come from other sources [8, 9]. which is about 4.6% (100% % - 68.2% - 27.2%) of the
distribution of the q will be rem
moved and considered as outliers.
B. Non-impulsive Interference
Non-impulsive interference consists of white
w noise, which
is generated from any electrical equipment such as amplifier
and oscilloscope and harmonic signal, whichh is generated from
communication, wireless and electronics syystems. Impulsive
interferences are easier to remove thaan non-impulsive
interferences. Commonly, white noise and haarmonic signal can
be discarded by frequency-dependent threshoolding [9].
According to IEEE standard (IEEE Std C57.113-2010),
C the
possible noise that could be encounteered during PD
measurement are harmonics, radio frequeency (RF) noise,
corona and sparking [9]. The standard also shows the
characteristics of noise signatures for some of these noise and
interference. These signatures can be usedd to assist in the Fig. 3. Normal Distribution Curve [18]
analysis of the PD signals measured in PRP PD to identify the
presence of noise in the PD measurement. C. Analysis using Statistical Tolerance
To Intervals
After eliminating the disscharge values, q which are
IV. STATISTICAL SIGNAL PROCESSING TECHNIQUE considered as outliers or disturrbance, tolerance intervals based
In the proposed statistical signal processsing technique, the on normal distribution will be b applied for analysis of the
statistical parameters were computed from the measured PD results, in order to determine the
t coverage of q for the whole
waveforms, in order to eliminate the noise or o disturbance. PD set of PD waveform. Statisticaal analysis is done on the results
signals from the on-site PD measurement cann be assumed to be before and after elimination of o the disturbance. In order to
random signals and the apparent charge recoorded are normally show the effectiveness of thhe statistical signal processing
distributed [18]. Hence, discharges whichh are out of the technique applied in this researrch work, the results before and
tolerance intervals based on normal distribuution curve will be after the elimination of q weree compared. In Fig. 3, 31.8% of
removed. Statistical signal processing technnique is performed distribution is higher than the values of μ ± ı while μ ± 2ı
after the elimination of these outlier discharges.
d Then, indicates that 4.6% of distributtion is higher than the values of
statistical analysis is carried out on the PD P parameters to μ ± 2ı. Furthermore, for μ ± 3ı,3 0.3% of the distribution has
analyze the results before and after thhe elimination of values higher than μ ± 3ı.
disturbance or outliers.
Computation of μ and ı off the discharge for the apparent
charge, q are performed forr the average and worst case
conditions. Average case referss to the mean value of discharge

357
 2015 IEEE International Conference on Signal and Imagee Processing Applications (ICSIPA)

mean magnitude from each cycle and thhe mean value of

normalized standard deviation from each cyycle whereas worst
case refers to the maximum value of thee discharge mean
magnitude from each cycle and the maaximum value of
normalized standard deviation from each cyccle.
In this paper, the triggered values of q to determine the
extent of PD are 300 pC for the verage casee and 1000 pC for
the worst case [16]. If the q computed for avverage case is less
than 300 pC and worst case is less than 1000 pC, the IT is
suspected to have only very low PD and not harmful.
h However,
when q is higher than 1000 pC, the IT needs to
t be tested offline
using published method such as dissolved gaas analysis (DGA)
test as recommended in [3]. The results obtained
o from the
4.004 ms 8.009 ms 12.01 ms 16.02 ms 20.02 ms
statistical analysis of the PRPD measuremeent for the on-site
measurement will be compared with the DGA A results. (b)

V. RESULTS AND DISCUSSSION

PRPD measurement to be analyzed consiist of measurement
from four known experimental setup (nam med as exp154523,
exp132332, exp150246 and exp153818) andd three known on-
site PD measurement (named as KL4405, KL136 and
Perak205). The PRPD waveforms were recorrded for about four
to six minutes duration. Experiments were conducted in the
controlled laboratory condition. Void in thhe insulation layer
was created in the laboratory and externall interference was 250pC
4.004 ms 8.009 ms 12.01 ms 16.02 ms 20.02 ms

generated to create noise. Four sets of experimental 219pC

measurement were recorded under differrent situations as (c)
described in Table I. The three known on-sitee PD measurement 188pC
were also tested with DGA test.
156pC

TABLE I. EXPERIMENTAL SETUP UNDER CONTTROLLED SITUATION

125pC

Experimental Data Siituation 94pC

High PD activity without
Exp154523 63pC
noise
32pC
Exp132332 High PD activity with noise
500fC
800pC 4.004 ms 8.009 ms 12.01 ms 16.02 ms 20.02 ms
Low PD activity without
Exp150246 700pC (d)
noise
600pC
Exp153818 Low PD acctivity with noise
500pC

400pC

Fig. 4 shows the PRPD pattern meeasured from the

300pC
experimental setup. It can be seen that thhe PRPD patterns
obtained were in accordance with thee theory as the 200pC
measurement were conducted under conntrolled laboratory
100pC
conditions.
500fC
x-axis: time (ms) y
y-axis: discharge magnitude (pC)

Fig. 4. PRPD pattern from the exxperimental setup: (a) exp132332 (b)
exp154523 (c) exp150246 (d) exp1538818

358
2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA)

Fig. 5 illustrates the possible noise identified from the maximum value for ı per cycle. If q falls in the range, then a
PRPD measurement obtained from KL North Bay 405 which ‘low’ descriptor will be produced, otherwise a ‘high’. The
are in accordance with the possible noise that can exist in an results are summarized in Table II and Table III.
on-site PD measurement as stipulated by the IEEE Standard
(IEEE Std C57.113-2010). Hence, it is necessary to eliminate TABLE II. PARAMETERS FOR THE APPARENT CHARGE FROM THE
EXPERIMENTAL DATA
the noise in the on-site PD measurement data before any
further diagnosis can be made on the severity of the PD Before Noise Elimination After Noise Elimination

occurrence and determination of the PD level. As such, there DATA

Tolerance Average low(-) Worst low(-) Average low(-) Worst low(-)
Interval Case or Case or Case or Case or
Actual
Status
are insufficient evidences to determine there are PD activities q (pC) high(x) q (pC) high(x) q (pC) high(x) q (pC) high(x)
ȝ±ı 2695.01 x 6555.27 x 1024.86 x 2094.08 x high PD,
from the PRPD waveforms. Thus, proposed statistical signal exp154523 ȝ ± 2ı 4671.34 x 11957.65 x 1641.64 x 3441.14 x without

processing technique is recommended to be applied to ȝ ± 3ı

ȝ±ı
6647.66
3419.52
x
x
17360.03
6890.00
x
x
2258.42
1214.80
x
x
4788.21
3062.95
x
x
Noise
high PD,
eliminate the noise signals before the extend of PD activity can exp132332 ȝ ± 2ı 5968.44 x 12502.89 x 1980.07 x 5136.04 x
with Noise
ȝ ± 3ı 8517.37 x 18115.79 x 2745.34 x 7209.13 x
be determined. ȝ±ı 57.34 - 103.71 - 47.06 - 84.38 - low PD,
exp150246 ȝ ± 2ı 86.74 - 166.82 - 70.00 - 129.82 - without
ȝ ± 3ı 116.13 - 229.92 - 92.94 - 175.26 - Noise
ȝ±ı 79.47 - 715.71 - 46.30 - 93.21 -
low PD, with
exp153818 ȝ ± 2ı 128.45 - 1350.90 x 64.94 - 141.24 -
Noise
ȝ ± 3ı 177.44 - 1986.09 x 83.58 - 189.26 -

TABLE III. PARAMETERS FOR THE APPARENT CHARGE FROM THE ON-
SITE DATA

Before Noise Elimination After Noise Elimination

Tolerance Average low(-) Worst low(-) Average low(-) Worst low(-)
DATA DGA Result
Interval Case or Case or Case or Case or
q (pC) high(x) q (pC) high(x) q (pC) high(x) q (pC) high(x)
ȝ±ı 190.28 - 799.21 - 68.82 - 235.79 -
Satisfactory
KL405 ȝ ± 2ı 322.38 x 1387.92 x 102.15 - 364.97 -
Condition
ȝ ± 3ı 454.48 x 1976.63 x 135.48 - 494.15 -
ȝ±ı 32.28 - 183.62 - 16.21 - 30.84 -
Satisfactory
KL136 ȝ ± 2ı 51.20 - 335.66 - 21.35 - 43.12 -
Condition
ȝ ± 3ı 70.12 - 487.71 - 26.48 - 55.40 -
ȝ±ı 23.15 - 62.50 - 13.61 - 24.92 -
Satisfactory
Perak205 ȝ ± 2ı 36.56 - 110.04 - 19.42 - 36.63 -
Condition
ȝ ± 3ı 49.96 - 157.58 - 25.23 - 48.35 -

Table II shows the statistical analysis done on the experimental

data before and after the application of signal processing
technique for noise elimination. As expected, an ‘x’ is detected
for high PD activity with and without noise before and after
elimination, which shows that about 4.6% of distribution
eliminated by the proposed statistical signal processing
Fig. 5. Noise identification from the on-site PD measurement for KL North
Bay 405
technique only removed the outliers or noise. A fault alarm is
detected at exp153818 before noise elimination as it shows
A. Statistical Signal Processing Technique high PD activity which differs from the expected results. This
is because, the q is affected by disturbance and statistical signal
PRPD pattern recorded for all the three on-site processing technique is therefore, applied to eliminate
measurement have clearly showed that the measured PD disturbance pulses. After noise elimination, it shows the
signals are interfered with noise signals. Hence, statistical discharge values are within the range with low PD activity and
signal processing technique were applied to the experimental this is the same as expected. The results indicate that the
data and on-site data for elimination of the interference. The proposed statistical signal processing technique is able to
apparent charge, q that are not within the range of two standard eliminate high magnitude disturbance pulses.
deviation of the mean were removed. Hence, about 4.6% of the
discharge distribution were removed and considered as outliers Analysis of on-site data is summarized in Table III. On-site
or noise. New sets of data for q were obtained, which were the data from KL405 shows high PD activity before noise
remaining 95.4% of the distribution. elimination. However, DGA test shows satisfactory condition.
Since the measurement was recorded on-site, it is possible that
B. Analysis using Statistical Tolerance Intervals the waveform is affected by disturbances from environment.
After elimination of the outlier discharges using the After noise elimination, the high magnitude disturbance pulses
proposed statistical signal processing technique, statistical are eliminated and low PD activity was obtained which is same
analysis were applied to analyze the new set of data and the as the DGA result. Site data KL136 and Perak205 show low
results were compared to the data before noise or outliers PD activity before and after noise elimination, which are the
elimination The μ and ı per cycle will then be computed again same as DGA test results.
from new set of data for statistical tolerance intervals based on
normal distribution. Average case refers to the mean value of μ The summary of results is shown in Table IV. The results
per cycle and the mean value of ı per cycle, whereas worst from statistical analysis after noise elimination are compared
case refers to the maximum value of μ per cycle and the with the actual status for experimental and site data. Hence, the
results obtained have indicated the capability of the proposed

359
 2015 IEEE International Conference on Signal and Image Processing Applications (ICSIPA)

statistical signal processing technique for eliminating applied to on-site data to test the practicality of the technique
disturbances that exist on site during PD measurement. and the results obtained are the same as expected. Therefore,
the proposed statistical signal processing technique is capable
TABLE IV. SUMMARY OF RESULTS FROM STATISTICAL ANALYSIS and appropriate to be used for a more reliable diagnosis of PD
Statistical for on-site measurement of ITs.
Analysis
DATA Actual Status ACKNOWLEDGEMENT
after
Filtration This project is funded by TNB Research SDN. BHD. The
exp132332 High PD High PD
research work is carried out in collaboration with UNITEN
exp154523 Low PD Low PD
Experimental
exp150246 High PD High PD R&D SDN. BHD.
exp153818 Low PD Low PD
Satisfactory REFERENCES
KL405 Low PD
Condition [1] GE Digital Energy ITI, Instrument Transformer Basic Technical
Satisfactory Information and Application, pp. 3.
Site KL136 Low PD
Condition [2] ABB, Instrument Transformers Technical Information and Application
Satisfactory Guide, pp. 4.
Perak205 Low PD [3] V. R. Garcia-Colon, R. Linan-Garcia, M. A. Jacobo, “On-line condition
Condition
assessment of high voltage current transformers,” Electrical Insulation,
2004. Conference Record of the 2004 IEEE International Symposium
on, vol., no., pp. 182, 185, 19-22 Sept. 2004.
VI. CONCLUSION [4] T. R. Blackburn, B. T. Phung, Z. Liu, R. E. James, “On-line partial
In conclusion, partial discharge (PD) detection is an discharge measurement on instrument transformers,” Electrical
important method to evaluate the insulation condition of an Insulating Materials, 1998. Proceedings of 1998 International
Symposium on, vol., no., pp. 497, 500, 27-30 Sept 1998.
instrument transformer (IT). Electrical partial discharge [5] Guomin Luo, Daming Zhang, “Study on Performance of HFCT and
measurement using high frequency current transformer (HFCT) UHF Sensors in Partial Discharge Detection,” 2010. IPEC 2010 on
is an alternative technique for PD detection as PD pp.630.
measurement can be done on in-service IT and non-intrusive, [6] IEC 60720, “Partial Discharge Measurements”, Third Edition, 2000-12.
[7] A. Mehta, R. N. Sharma, S. Chauhan, “Partial discharge study by
without the IT to be offline. However, the limitation of monitoring key gases of power transformers,” Electronics Computer
electrical PD measurement is that it is highly susceptible to Technology (ICECT), 2011 3rd International Conference on, vol. 4, no.,
electromagnetic interference and noise found in any substations pp. 183, 186, 8-10 April 2011.
which may superposed with PD signals. Hence, this could also [8] A. Kraetge, S. Hoek, K. Rethmeier, M. Kruger, P. Winter, “Advanced
be the reason for high PD activities detected during on-site noise suppression during PD measurement by real-time pulse-waveform
analysis of PD pulses and pulse-shaped disturbances,” Electrical
measurement of the instrument transformers. DGA test carried Insulation (ISEI), Conference Record of the 2010 IEEE International
out on three of the instrument transformers showed that the Symposium on, vol., no., pp. 1, 6, 6-9 June 2010.
instrument transformers were in good or acceptable condition [9] IEEE Standard C57.113, “IEEE Recommended Pratice for Partial
although the PRPD measurement showed high values of PD. Discharge Measurement in Liquid-Filled Power Transformers and Shunt
As such in this research, a study has been carried out to Reactors,” 2010.
[10] Mohamad Ghaffarian Niasar, “Partial Discharge Signatures of Defects in
study the characteristics of PD to identify noise from on-site Insulation Systems Consisting of Oil and Oil-impregnated Pape,” KTH
electrical PD measurement data obtained from MPD 600 PD Electrical Engineering, pp. 11-20, 27, 28, 2012.
measurement system. Initially, the electrical PD measurement [11] Gabe Paolettie, P.E. and Alex Golubev, PhD, “Partial Discharge Theory
which was in the form of PRPD pattern were visualized and and Applications to Electrical Systems. Pulp and Paper,” 1999. Industry
Technical Conference Record of 1999 Annual, pp.128, 138, 21-25 June
evaluated. Then, the types of noise that could interfere with the 1999.
measured PD signals were identified. [12] M. S. Naidu, V. Kamaraju, High Voltage Engineering, Fourth Edition,
Furthermore, a statistical signal processing technique is The McGraw-Hill Companies, pp. 99-101.
proposed for eliminating disturbances using statistical [13] E. Kuffel, W. S. Zaengl, High Voltage Engineering Fundamentals,
parameters. The technique is designed to eliminate outlier Pergamon Press, pp. 409.
[14] Y. R. Chaudhari, N. R. Bhosale, P. M. Kothoke, “Composite analysis of
discharges based on criteria of two standard deviation (μ ± 2ı). phase resolved partial discharge patterns using statistical techniques,”
According to normal distribution curve, criteria for two International Journal of Modern Engineering Research on, vol. 3, pp.
standard deviation eliminates only about 4.6% of the 1947-1457, Jul-Aug 2013.
distribution and it will not affect the PD signal, as it has been [15] F. H. Kreuger, E. Gulski, A. Krivda, “Classification of partial
proven by laboratory experiments. The statistical analysis is discharges,” Electrical Insulation, IEEE Transactions on, vol. 28, no. 6,
pp. 917, 931, Dec 1993.
applied before and after the elimination of the PD data, in order [16] C. G. Azcarraga, V.R. Garcia-Colon, and A. Nava G, “On-site Testing
to give a clearer perspective on the measured data and analyze of Instrument Transformers,” 2006 Annual Report Conference on
the differences. Electrical Insulation and Dielectric Phenomena, pp. 39, 42, 2006.
Statistical analysis on experimental results after the [17] British Standard BS EN 60270:2001, IEC 60270:2000. “High-voltage
test techniques – Partial discharge measurements,” British Standard
application of the proposed technique has clearly proven the Institution (BSI), London, 2001.
effectiveness of this technique to eliminate noise pulses from [18] Open Stax College, A Single Population Mean using Normal
PRPD measurement. Moreover, this technique has been Distribution, Nov 2014.

360