Statisctical Analysis

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

F E A T U R E A R T I C L E

Statistical Analysis of Online


Ultrahigh-Frequency Partial-Discharge
Measurement of Power Transformers
Key words: power transformer partial discharge, ultrahigh-frequency partial-discharge monitoring,
phase-resolved partial-discharge pattern

Introduction Michael Beltle, Andreas Müller,


The measurement of partial discharge (PD) has been estab-
lished as a common method for transformer testing using elec-
and Stefan Tenbohlen
trical measurements according to IEC 60270 [1]. However, in Institute of Power Transmission and High
this article, an UHF approach is presented. The UHF method Voltage Technology, University of Stuttgart,
was first introduced on gas-insulated switchgear but is also ap-
plicable to power transformers [2]. In the technique, an antenna Pfaffenwaldring 47, 70569 Stuttgart,
is inserted into the transformer tank through an oil valve. The Germany
method detects PD by measuring electromagnetic emissions
originating from transient currents of PD [3], [4]. Because the
tank is grounded, it represents a Faraday cage shielding external
noise, which makes the UHF method less sensitive to external
PD such as corona.
The article discusses the ultrahigh-
Compared to electrical PD measurements, the UHF method frequency partial-discharge (PD)
has to be treated differently. Unlike electrical PD measurements, monitoring of a 50-year-old genera-
UHF measurements cannot be calibrated in terms of apparent
charge in picocoulombs because the strength of the electromag- tor step-up transformer with the aim
netic field is measured. However, like electrical measurements, of being able to discriminate various
it is possible to correlate PD to the phase angle of the applied
voltage using the phase-resolved PD (PRPD) pattern.
PD sources from the phase-resolved
This article presents an overview of UHF PD measurements PD pattern.
of a generator step-up transformer over a three-year period. Sev-
eral PD sources are active in the transformer, and fast changes
in the PD pattern and levels are observed. An approach for the
evaluation of measurement data is introduced, and the long-term Since the beginning of the monitoring process in the summer
development of PD is discussed in the article. The aim is to dis- of 2009, the generator step-up transformer has been in service
criminate different PD sources from the PRPD pattern and clas- for 80 days. An UHF probe (Figure 1) was installed in the DN
sify them in terms of their level and frequency of occurrence on 80 oil flange valve of the transformer [5], [6] for measurement
a long-term basis. Thus, a classification of different sources and of UHF PD.
their severity is attempted. The UHF PD monitoring of the transformer was done with a
Doble Lemke LDS-6/UHF (Doble Lemke diagnostics, Dresden,
Experimental Setup Germany) measuring system. The UHF PD signals were mea-
The UHF PD monitoring was performed on a 50-year-old sured with approximately 35-dB amplification and a bandwidth
generator step-up transformer with a rated voltage of 110/10 of 9 MHz at a center frequency of 505 MHz. Phase L1 was used
kV and a rated power of 120 MVA. The generating unit mainly for phase correlation, and Matlab (software by Mathworks in
operates in the winter season compensating peak-load days. Nattick, MA) was used for data analysis.

November/December — Vol. 28, No. 6 0883-7554/12/$31/©2012/IEEE 17


ity. No regularity or trend of PD events over time can be seen.
The mean value of all considered PDs was at 180,000 events
per hour, which corresponds to one event per cycle, whereas
the standard deviation was at 330,000 events per hour or 1.8
events per cycle (Table 1). The average of the UHF amplitude
was 15.38 mV.
The high volatility of PD events can be considered to make
Figure 1. The UHF probe with transformer DN 80 flange con- a valid condition assessment of the transformer demanding. A
nector for the inside-tank measurement of UHF partial dis- single electrical or UHF measurement, if performed at times
charge. with high or low activity, could result in making a misleading
condition assessment of the transformer. So, continuous moni-
toring delivers considerable gain in information compared with
Because the 110-kV paper–oil transformer bushing does not
a single test measurement and represents the main advantage of
have measuring taps, no electrical PD monitoring was done.
the technique.
However, before the transformer was put into service, an electri-
cal PD measurement with external HV coupling capacitors was Discrimination of PD Using PRPD Pattern
done. It revealed different PD sources on each phase with PD Prior electrical PD measurements according to IEC 60270 in-
between 100 and 500 pC. Also, unsteady PD up to 1 nC was oc- dicated different PD sources in the transformer, and to discrimi-
casionally observed [7]. nate the PD sources, the PRPD pattern method was used. Partial-
discharge sources are phase stable, and PRPD patterns can be
Long-Term Analysis of PD synthetically created by using one phase of the transformer as a
reference. In this study, phase L1 was used.
Number of PD Events The UHF PD measurements showed a fast-changing PRPD
The first approach to evaluate the long-term PD of the trans- pattern over time. However, some patterns occurred regularly,
former considered all UHF PD events. Figure 2 shows the num- whereas others were only found a few times. Regularly occur-
ber of recorded events and the corresponding maximum am- ring patterns are considered to be typical for this transformer.
plitude in millivolts over a three-year monitoring period. The Figure 3 shows PRPD patterns of two different typical types.
UHF-measured PD over one hour was accumulated and repre- The PDs were accumulated for one hour to generate the pre-
sents one bar in the plot. For a better survey, a compressed view sented PRPD patterns, and the color indicates the number of PD
was used that neglects the days when the generating unit was events per hour.
not in operation. At times the measurement system was over- A third typical pattern is shown in Figure 4. The PD source
driven; a constant UHF amplitude is shown in Figure 2. The am- is characterized by the area and the shape it covers in the PRPD
plitudes of the overdriven PD pulses are not known; however, pattern. The entire area of the PRPD pattern was defined by a
their events were counted. Only PD above a noise level of 2 mV 2-dimensional matrix given by the phase angle from 0 to 360° in
up to the maximum at 25 mV was considered. The noise level is 5° steps and the amplitude from 2 to 25 mV in 1-mV steps. Fig-
defined by the highest antenna voltage level that is reached by ure 5 shows the selected areas for source 1 in red color, covering
non-phase-correlated signals. The dark blue line in Figure 2 is a 13% of the entire PRPD diagram area. All recorded PDs that lay
moving average illustrating the high volatility of the PD activ- within the selected area were assigned to this PD source. Only
these PDs were selected from the entire UHF data and evaluated.
To be selective, the areas should be as small as possible. Large
areas covering the entire PRPD pattern do not provide any other
information.
This method was used for all introduced typical sources. Fig-
ure 6 shows the selected areas of the three different sources that
are determined in the following. Investigations are explained us-
ing source 1.

Table 1. Statistical Analysis of Long-Term UHF Partial-Discharge Measure-


ment.

Maximum number within one hour  4,500,000/hour = 25 per cycle

Minimum number within one hour  <200/hour = 0.001 per cycle

Mean value  180,000/hour = 1 per cycle


Figure 2. Entire partial-discharge (PD) events per hour and
Standard deviation  330,000/hour = 1.8 per cycle
maximum amplitude in millivolts over a three-year period.

18 IEEE Electrical Insulation Magazine


Figure 5. Area of the phase-resolved partial-discharge (PD)
pattern assigned to source 1.

Analysis of Selected PD Sources


In Figure 7 all PD events inside the area of source 1 are accu-
mulated for one hour and plotted over time. As explained earlier,
a compressed data setup was used that neglects the times the
transformer is off-line. In this particular case, the source was
responsible for about 72,000 PD events per hour or 0.4 events
per cycle (mean value). The dark-blue line is a moving average
illustrating the high volatility of source 1.
Comparing the PD events of source 1 with the entire PD
events per hour from Figure 2 shows a rise of both curves at
the same time at hour 230. Hence, it can be concluded that this
selected source 1 influenced the entire number of PD. To evalu-
ate the contribution of a source on the entire number of PD, its
Figure 3. Two typical UHF phase-resolved partial-discharge relative fraction was used.
(PD) patterns present in the step-up transformer. Figure 8 shows the relative fraction of source 1. Therefore,
the amount of PD assigned to source 1 was divided by the entire
number of PD. The dark-blue line is a moving average illustrat-

Figure 4. Phase-resolved partial-discharge (PD) pattern of Figure 6. Selected pattern areas of sources 2 (green), 3 (dark
source 1. blue), and 1 (light blue) for reference.

November/December — Vol. 28, No. 6 19


Figure 7. Partial-discharge (PD) events of selected area of
Figure 9. Average number of partial-discharge (PD) events of
source 1 per hour versus time. Moving average is plotted in
sources 1, 2, and 3. Only days the transformer was in service
dark blue.
are considered.

ing the changes of source 1 over time. On average, 48% of the


came dominant. Source 3 had a low relative fraction (on average
entire number of PD occurred in the area of source 1. Thus, on
13%; see Table 2) but showed a high volatility with times when
average nearly half of all PDs can be assigned to source 1.
the source is dominant providing about 52% of all PDs (see Fig-
Both the number of PDs and the relative fraction were used
ure 10 at an operation time of 700 hours). Source 2 had a very
to compare different sources and to evaluate their influence on
low relative fraction, with a mean value of 6% (see Table 2). At
the entire PD activity. The number of PD events per hour of all
one time, source 2 was dominant with a relative fraction of near-
three sources defined in Figure 6 is plotted in Figure 9. Source
ly 40%. Thus, the standard deviation for source 2 was high com-
1 showed the highest number of PD per hour and only small
pared with the mean value (see Table 2). On average, 62% of all
variations over time compared with sources 2 and 3. Therefore,
PDs were originated by the considered three sources. Therefore,
source 1 can be considered as permanently active, whereas the
the selected sources are suited to describe the PD activity of the
changes and the absolute number of PD of sources 2 and 3 il-
transformer. By evaluating the sum of all three sources, it can
lustrate their smaller contribution to the entire PD activity of the
also be seen that there were rare times when they provided only
transformer. The relative influence of different sources on the
10 to 11% of the entire number of PD. These times have to be
entire PD activity is shown in Figure 10.
investigated separately. The reasons for a low relative fraction
On average, source 1 had the highest influence on the entire
could be other sources dominating PRPD pattern, external noise,
PD activity. Nevertheless, at some times, the other sources be-
and so on.
To perform an automated evaluation of PD, statistical meth-
ods were used. The presented approach combines the absolute
number of PDs of a source and its relative fraction with respect
to the entire number of PD. Figure 11 shows the relative fraction
on the x-axis and the corresponding absolute number of PDs
on the logarithmical-scaled y-axis of source 1; time dependency
is not included here. Again, every dot represents the accumu-
lated data of one hour. To qualify and quantify a pattern and thus
judge the relevance of a defined source, the mean value on the
x-axis, representing the relative fraction, and the mean value on
the y-axis, representing the average number of PD, can be used
to characterize the statistic behavior of the source.
Sources 2 and 3 were determined using the same methods.
Source 2 (green) covered 8.5% of the area of the PRPD pat-
tern. Source 3 (dark blue) covered 8.7% of the pattern. Figure 12
shows the absolute and the relative fractions of all three sources.
Source 2 (green) showed low relative fraction (mean value is
2%) and low numbers of PDs (4,000 events per hour or 0.02
Figure 8. Relative activity of source 1 versus time. Moving aver- per cycle on average). It represents an example for a source that
age is plotted in dark blue. PD = partial discharge. had a low influence on the entire PD activity of the transformer.

20 IEEE Electrical Insulation Magazine


Table 2. Statistical Parameters of the Three Partial-Discharge (PD) Sources.

Relative fraction per hour (%) Absolute number of PD

Source In percentage Standard deviation Number of events Standard deviation

Source 1 48 24 72,000/hour, 0.4/cycle 98,384

Source 2 2 6 4,000/hour, 0.02/ cycle 12,059

Source 3 13 14 30,000/hour, 0.16/cycle 76,906

Source 3 (dark blue) had a relative fraction of 13% and 30,000 sented. Ahead of the UHF PD monitoring, electrical PD mea-
events per hour or 0.17 per cycle on average. Thus, the source surements according to IEC 60270 revealed at least 3 different
is of higher interest for long-term analysis. A direct comparison PD sources inside the transformer.
with source 1 illustrates the differences between sources with Using UHF PD measurement, the overall number of PDs
strong long-term activity and sources with alternating activity. during a certain period of time showed high volatility. A trend
In comparison to Figure 9 and Figure 10, which illustrate the could not be recognized. To distinguish between different PD
time dependency of the different sources, the combined plot of sources, the PRPD pattern was used. Similar to the electrical
Figure 12 delivers a quick overview of different sources. Large measurement, the UHF PRPD pattern showed different PD
mean values of the relative fraction of the source indicate a sources. Long-time surveillance revealed the activity of these
strong influence on the entire PD activity of a transformer. The sources. Most of them showed repetitive behavior but were not
absolute number on the y-axis gives an impression on the actual permanently active. To manage the amount of available UHF
activity of a source. The development over time of the sources data, semiautomated evaluation was used. For each source, the
can be evaluated by dividing the available UHF data in subse- relevant data were extracted. The changes of a source could be
quent chronological parts and comparing the mean values of determined over the entire observation time. The source could
these parts. be described by the number of PD correlated to the source and
Table 2 summarizes the statistical data of all three sources its relative fraction with respect to the entire number of PD. Dif-
regarding absolute number of PD and its relative fraction. The ferent sources can be compared with each other, and their contri-
relative fraction of source 1 showed a small standard deviation bution on the total PD activity can be used as an indicator for the
(half of the mean value), indicating a stable source. In compari- relevance of a source. Hence, long-term surveillance of different
son to this, the other sources had a standard deviation in the PD sources becomes possible.
range of the mean value (source 3) or even higher (source 2). In comparison to PD monitoring, single measurements could
be performed at times with high or low activity and result in mis-
Conclusion leading interpretations of the PD status of transformers. Thus,
In this article, the evaluation of three years of UHF PD moni- the approach demonstrates the abilities of continuous UHF PD
toring data on a 120-MVA generator step-up transformer is pre- monitoring and the gain of information it can generate in com-

Figure 10. Relative fraction of all sources with respect to entire Figure 11. Absolute number of partial-discharge (PD) events
number of partial discharges (PD) over time. versus relative fraction of source 1.

November/December — Vol. 28, No. 6 21


Michael Beltle received the Dipl.-Ing.
degree in electrical engineering from the
University of Stuttgart, Germany, in 2009.
After his graduation he started working at
the Institute of Power Transmission and
High Voltage Technology, University of
Stuttgart. He operates in the field of power
transformer diagnostics and determines the
long-term development of partial discharg-
es and investigates the mechanical vibrations of the active parts
of transformers. He is member of CIGRE and the German Power
Engineering Society VDE-ETG.

Andreas Müller received the Dipl.-Ing.


degree in electrical engineering from the
University of Stuttgart, Germany, in 2008.
Figure 12. Absolute number of partial-discharge (PD) events In 2009 he started working at the Institute
versus relative occurrence of assumed sources 2 and 3 from of Power Transmission and High Voltage
Figure 6 (green and dark blue) and main source 1 as reference Technology, University of Stuttgart, in the
(light blue). field of power transformer diagnostics. His
focus lies in the application of dissolved
gas analysis (DGA) and long-term moni-
parison with conventional measurement that provides only snap-
toring of partial discharges. He is a member of the German Pow-
shot information. Therefore, PD monitoring is considered as a
er Engineering Society VDE-ETG.
useful extension for transformer assessment. Further investiga-
tions can optimize the presented method using automated pat-
Stefan Tenbohlen (M ’07) received his
tern recognition and evaluation.
Diploma and Dr.-Ing. degrees from the
Technical University of Aachen, Germany,
References in 1992 and 1997, respectively. In 1997
[1] High-Voltage Test Techniques—Partial Discharge Measurements, IEC
Standard 60270, 3rd ed., International Electrotechnical Commission,
he joined AREVA Schorch Transforma-
Geneva, Switzerland, 2000. toren GmbH, Monchengladbach, Ger-
[2] M. Hikita, S. Ohtsuka, and S. Matsumoto, “Recent trend of the partial many, where he was responsible for basic
discharge measurement technique using the UHF electromagnetic wave research and product development, and
detection method,” IEEJ Trans., no. 2, pp. 504–509, 2007.
[3] S. Tenbohlen, S. Hoek, D. Denissov, and S. M. Markalous, “Partial
with this function, he worked in the field
discharge measurement in the ultra high frequency (UHF) range,” IEEE of online monitoring of power transformers. From 2002 to 2004
Trans. Dielectr. Electr. Insul., vol. 15, no. 6, pp. 1544–1552, 2008. he was the head of the electrical and mechanical design depart-
[4] M. D. Judd, G. P. Cleary, and C. J. Bennoch, “Applying UHF partial ment. In 2004, he was appointed as the professor and head of
discharge detection to power transformers,” IEEE Power Eng. Rev., pp.
57–58, Aug. 2002.
the Institute of Power Transmission and High Voltage Technol-
[5] S. Coenen, S. Tenbohlen, T. Strehl, and S. Markalous, “Fundamental ogy of the University of Stuttgart, Germany. In this position, his
characteristics of UHF PD probes and the radiation behaviour of PD main research fields are diagnostics of equipment of electrical
sources in power transformers,” International Symposium on High Volt- networks, development of high voltage measurement technique,
age Engineering, 2009, Cape Town, South Africa, pp. 666–671.
[6] S. Coenen, S. Tenbohlen, S. M. Markalous, and T. Strehl, “Sensitivity of
behavior of gas insulated insulation systems, and different as-
UHF PD measurements in power transformers,” IEEE Trans. Dielectr. pects of electromagnetic compatibility (EMC). Tenbohlen holds
Electr. Insul., vol. 15, no. 6, pp. 1553–1558, 2008. several patents and published more than 130 papers. He is a
[7] S. Tenbohlen, A. Pfeffer, and S. Coenen, “On-site experiences with multi- member of the German committees of CIGRE A2 (Power Trans-
terminal IEC PD measurements, UHF PD measurements and acoustic
PD localisation,” Proceedings of the IEEE International Symposium on
formers), D1 (Emerging Technologies), C4 (System Technical
Electrical Insulation, 2010, paper no. 095. Performance), and several international working groups. Since
2008, he has been a member of the board of the German Power
Engineering Society VDE-ETG.

22 IEEE Electrical Insulation Magazine

You might also like