830 IEEE Transactions on Dielectrics and Electrical Insulation Vol. 5 No. 6, December 1998
Pattern Analysis
of
Partial Discharges
in
SFG
CIS
S.
Meijer, E.
Gulski and J.J.
Smit Delft University of Technology, Delft, The NetherlandsABSTRACT
The measurement of partial discharge (PD) of several faults in gas-insulated system (GIS) is
discussed. Phase-resolved PD patterns have been measured using three different PD detection
measuring systems: according to the IEC 270 recommendations, a VHF /UHF measuring system
with narrow band filtering, and the UHF measuring system with wide band filtering. PD pat-
terns are compared using computer-based discrimination tools. The influence of the selected
center frequency on the PD patterns is discussed for the narrow band VHF /UHF measuring sys-
tem. The influence of the number and type of GIS components between the discharging defect
and the capacitive coupler on the shape of the PD patterns is analyzed. For several GIS compo-
nents the signal reduction is studied. It was found that the shape of PD patterns is independent
on the used PD detection circuit and the propagation path of the PD signals. As a result, dis-
crimination and classification of PD distributions of several studied defects are possible using
digital tools.
1
INTRODUCTION
F6 gas-insulated system (GIS) have proven to be very reliable. However, it is known that faults cannot completely be excluded and that most faults which might lead to failure in a GIs show PD activ- ity [l, 21. Two kinds of electrical measuring systems are in use to detect PD in GIs: Conventional measuring systems based on the IEC 270 rec-
ommendation; most commercially available narrow or wide band PD
detectors are measuring PD signals using frequency ranges of -10 to
-500 kHz and the measurement in [pC] of actual discharge levels is
based on the IEC 270 recommendations [3] and modern VHF /UHF mea-
suring systems which are using narrow or wide band filters: detection
in a frequency range to -3 GHz which has been frequently used for the
last years during on-site tests [4-61.
1
.I THE
IEC 270MEASURING
SYSTEM
One of the main advantages of the standardized IEC 270 system is a
very broad scale of experience and recently also the presence of digital
tools related to phase-resolved PD recognition that support the evalu-
ation of PD measurements [7-101. Unfortunately, due to external noise
which is frequently present in the field (on-site/on-line), the use of the conventional test circuit as recommended by the IEC 270 [3] is quite dif- ficult. To solve this problem and to achieve a sensitive measuring circuit for on-line testing of GIs, a system called VHF /UHF detection has been introduced [5]. It has been shown that this detection system has about the same sensitivity as the IEC 270 system [ll, 121.
1.2
THE NARROW BAND VHF/UHF
SYSTEM
The main advantage of the narrow band VHF /UHF system is the pos-
sibility to suppress external noise and disturbances by selecting a fre- quency range with a sufficient signal to-noise ratio and lowest influence of disturbances [5,15]. In contrast to Figure l(a) which shows a nar- row band measurement tuned in a proper frequency range, Figure I@) shows a measurement tuned in a wrong frequency range.
Fast on-site inspection is possible if VHF /UHF capacitive couplers are permanently installed in the GIS installation. The occurrence of PD in
GIS caused by moving particles, particles on insulators and protrusions
on the conductor can be very dangerous to the insulation conditions of the GIS installation. This means that the detection and recognition of these faults at an early stage during on-site tests are important tools for condition-based maintenance of GIS. In particular, the digital tools re- lated to recognition and classification of phase-resolved PD patterns also
have been used successfully for narrow band vHF /UHF measurements
MI.
1.3
THE WIDE BAND UHF SYSTEM
A wide band UHF system has been introduced also [17]. This sys-
tem measures the time domain signal in a frequency range from 500 to
1500 MHz. Using wide band amplification of the signal, an increase of
the signal-to-noise ratio can be expected but on the other hand more in- fluence of disturbances cannot be excluded, see Figure l(c).
In practice, during an acceptance test, the evaluation of PD in GIS is
restricted to measuring the inception voltage (kV) and the largest dis-
charge magnitude (pC) and comparing these to the test specifications 1070-9878/98/ $3.00 0 1998 IEEE
5 5
...Hqmax(4)
-
:
...4 4
...Hqn(4)
:
...[nC]
o:...-
.... ...46
/. ... ...28
I...-
... ...Ilh
... ... ...U
OO"
180"
@
+360°
344
180"
+360°
Figure 1. PD patterns measured for a protrusion on the HV conductor (a) with the narrow band UHF system properly tuned at a measuring frequency of 1223 MHz; (b) with the narrow band UHF system not properly tuned at a measuring frequency of 1025 MHz; (c) with the wide band UHF system. However, if the maximum allowable discharge level is exceeded, it is important to know the cause of the discharge. The main goal of evalu- ation in GIS is to find one of the following defects.Protrusion on the HV conductor; protrusion on the enclosure; a parti-
cle on an insulator (spacer); free moving particles inside the enclosure; internal defects in moving parts like circuit breakers and disconnectors; and electrically floating parts in the installation.
For this purpose different discharge related parameters and quanti-
ties have been introduced in the past and are still in use [3]. Due to the
increasing automation of PD measurements, the use of computer-aided
evaluation has become popular. The use of a computer-aided system offers the opportunity to store sequences of the discharge pulses and to
Figure 2. Four parts are of importance for VHF /UHF PD measurements.
(1) Discharging defect; (2) Excitation of traveling waves; (3) Transfer func- tion of the sensor; and (4) Data processing.
post process these in the course of time or as a function of the power fre- quency cycle [U]. In this way a basis is created for further evaluation and diagnosis of PD measurements in G I s [17,29].
Figure 2 shows the most important stages of PD signal evaluation:
discharging defect; electromagnetic wave excitation by the discharge and influence of the GIs on the transmission of the electromagnetic waves; coupling and data acquisition; and data processing.
It is known that the discharge currents in SF6 are very short in time:
<
1 ns [17]. These fast pulses excite electromagnetic waves in the GIs enclosure, see (2) in Figure 2. Various components such as T-junctions, insulators, circuit breakers etc. are used to build a GIS. These compo- nents influence the transmission of the electromagnetic waves due to reflection, attenuation and dissipation. To pick up the PD signals, ca- pacitive couplers can be used, see (3) in Figure 2. Finally the measuringdata can be processed by a spectrum analyzer or a PD analyzer.
To evaluate different PD sources, digital post-processing provides a series of phase-resolved PD quantities. Sometimes it is not possible to characterize a fault by one single shape of the distribution. On the con-
trary, sometimes a PD source can be characterized by different PD pat-
terns, for example in the case of a protrusion on the conductor, see Fig- ure 3(a) to (d). In this case evaluation of the patterns becomes more dif- ficult. Moreover, several parameters such as the geometry of the GIs test setup, the insulating materials involved in the PD process, the electric field strength and the electrode material may influence the distributions. In particular, one must distinguish between signals emitted from fixed and moving particles. Fixed particles can produce corona discharges, whereas moving particles can produce both corona discharges during flight and contact discharges when striking a surface [20].
In addition to physically related conditions, several parameters like the measuring frequency in the case of the narrow band vHF / U H F sys- tem, the choice of PD quantities as well as the statistical tools for dis- crimination are of importance. To contribute to the discussion on the possibilities of digital PD monitoring of GIS and statistical discrimina- tion and classification of PD patterns, the following aspects are studied in this paper.
IEC 270, narrow band VHF /UHF and wide band UHF PD patterns are
compared for a sharp protrusion on the conductor in the G I s installation; the influence of the VHF /UHF measuring frequency on the PD patterns is
832
0 180" 360'
Figure 3. Examples of PD patterns made by 3D Hn($, q ) intensity dis- tribution of the discharge magnitude q and phase-angle 4 observed for dif- ferent typical defects in GIS [16,19]: (a) to (d) protrusion on the conductor, (e) a free moving particle, (f) a particle fixed to an insulator and (8) a pro- trusion on the enclosure. The PD-magnitudes are not calibrated.
analyzed; the influence of the distance and the number of insulators be- tween the fault and the coupler on the PD patterns is studied; and a com- parison of two different discrimination tools is made: the tree method and fractal analysis.
Based on these investigations it was found that the shape of PD pat-
terns is independent of the PD detection circuit used and of the prop-
agation path of the PD signals. As a result, discrimination and classi- fication of PD distributions of several studied defects is possible using digital tools.
2 PD MEASURING SYSTEMS:
PRACTICAL DETAl LS
The PD measurements discussed in this report were obtained using
two test setups. Figures 4 and 5 show commercially available 245 and
Meijer et al.: PD Paffern Analysis in G I s
Coupling capacitor
I IEC270 I detectio
1
.-detection
,
circuit rTE-571 II ccircuit I _
__
I - -L . _ - .Figure 4. 245 kV GIS measuring setup as has been used for IEC 270
and VHF /UHF PD measurements (courtesy of University of Stuttgart, Ger- many).
420 kV GIS. Exact information of the dimensions of both GIS is not nec-
essary for the understanding of the discussed results in this paper. In this Section a more detailed description of the three measuring sys- tems is given. In all systems a digital PD detector (TE-571) is used to
digitize, record and store the PD signals measured using a measuring
system according to the IEC 270 recommendations or a tunable narrow
band VHF /UHF filter or a wide band UHF filter. After a PD measurement,
different graphical representations of the measured PD signals can be
created to further analyze the PD source, see Section 3. Only in the case
of the measurements obtained with the IEC 270 measuring system, the
amplitudes of PD pulses are measured in pC. In both other systems no
calibration is possible and the PD pulses are uncalibrated.
2.1
IEC
270PD DETECTION
The IEC 270 PD detection circuit is composed of a coupling capac-
itor (60 pF), a coupling device (AKV-572) and a PD detector (TE-571). The PD signals measured using this circuit were correlated to the 50 Hz wave form of the applied voltage. In this way phase-resolved PD pat- terns were obtained and used to discriminate between different faults. To automate this process, several digital tools are in use to analyze the
0
T
I
c
Background noise spectm
Figure 5. 420 kV GIS measuring setup as has been used for on-site nar- row band VHF /UHF and wide band UHF PD measurements.
phase-resolved patterns: statistical analysis and cluster analysis (in par-
ticular, fractal analysis and tree analysis) [18,21-231. the amount and type of random noise pulses (Figure 6) a choice has to Depending on the frequency spectrum of a particular PD event and be made between methods 2 and 3.
2.2
NARROW BAND VHF/UHF
PDDETECTION
2.4 METHOD2
The VHF/UHF detection circuit is composed of a capacitive coupler
(in the case of the 245 kV setup a cone-shaped coupler with a capacitance
C = 1.058 pF [24] and in the case of the 420 kV setup disc couplers),
a spectrum analyzer (SA) (Tektronix 2711) and a modified PD detector (TE-571). A capacitive coupler was used to pick up the electromagnetic
waves as produced inside the GIS enclosure by each PD pulse, see Fig-
ure 2.
The first step in finding a proper measuring frequency for a VHF / U H F
PD measurement is to measure the full frequency span (ranging from 0 to
1.8 GHz in the case of a Tektronix 2711 spectrum analyzer). The SA can be seen as a tunable narrow band bandpass filter and the center frequency of the tunable filter is swept across the desired frequency range [25]. As a result, the frequency spectrum can be measured.
Using a SA, there are several methods to obtain a frequency spectrum.
This can best be used to detect PD pulses with a low repetition rate, Figure 7(a). The disadvantage of this method is that random noise pulses can strongly influence the measured frequency spectrum of the
PD event. Another disadvantage is that nothing can be concluded about
the intensity of the detected pulses at peaks in the frequency spectrum. Therefore, no conclusion is possible whether the detected pulses were
from continuous PD pulses or random noise pulses.
2.5 METHOD3
This method can best be used in situations with a high PD repetition
rate, Figure 7(b). Moreover, the method can suppress randomly occur- ring noise pulses because the measured signals at each frequency are av-
eraged over all x sweeps. Therefore the amplitude of a noise pulse can
be reduced to l/x of its original size. Obviously, for the same reason, 1. The spectrum can be measured during one sweep of the SA,
2. The spectrum can be measured during several sweeps of the SA and the
this method is not suitable
6
measure PD pulses with a low repetitionrate.
v
It has been shown that using both methods 2 and 3 at least 10 sweeps
frequency spectrum is composed of the maximum amplitudes at each fre- 3. The spectrum can be measured during several sweeps of the SA and the
quency, and
Of are required to Obtain a frequency 'pectrum [19]*
frequkcy spectrum is composed of the averaged amplitudes at each fre- quency
"
The SA measures a frequency spectrum (solid line in Figure 6) which
834 Meijer et al.: PD Pattern Analysis in GIS -55
,
I -95-
/
-(a) 0 300 600 900 1200 1500 1800Frequency
[MHz] m 0,$j
-75a
3 U-
-85 I -95J
(b) 0 300 600 900 1200 1500 1800Figure 7 Frequency spectra obtained 10 sweeps of 5 s using (a) hold maximum amplitudes and (b) averaging of a free moving oarticle of 10 mm
Frequency
[MHz]in the GIS enciosure. The applied v;t& was 260 kV a;;d the gas pressure was 420 kPa.
and continuous noise, for example from radio transmitters. To make
a distinction between noise and PD signals the measured frequency
spectrum can be compared to a frequency spectrum without PD signals
(dashed line in Figure 6). In particular this method can be used in a lab- oratory with a test setup. As an example, see Figure 8. First the back- ground noise level was measured, see Figure 8(a). Then voltage was ap-
plied to the GIS test setup with a protrusion fixed on the conductor. The
resulting frequency spectrum is shown in Figure 8(b). Visual compar- ison of both spectra shows no clear difference and therefore it was not
possible to select measuring frequencies for a VHF /UHF PD measure-
ment based on these spectra. Therefore, the difference between both spectra was processed, resulting in the frequency spectrum shown in Figure Y(a).
For further analysis of the processed frequency spectrum of Fig- ure 9(a), a frequency scan (FS) can be made. During a FS the SA operates
in the zero spanmode. In the case of Figure 9(a) five measuring frequen-
cies are further analyzed. During measuring time
T
the SA is tuned toeach selected measuring frequency At each frequency the maximum PD
magnitude is obtained during t, see Figure Y(b).
To determine the most sensitive and suitable measuring frequencies
for a narrow-band VHF /UHF PD measurement, the obtained PD pulses
are processed and shown as a point-on-wave (POW). Therefore analysis
of the changes in the measured POW for different measuring frequencies
can give more information about the frequency behavior of the fault and
measuring setup, see Figure 9(c). Although the PD event has the highest
amplitude in the frequency spectrum and frequency scan at frequency
f i , the POW in Figure 9(c) shows that this frequency also is strongly sub-
-60
-
-70 E m 9 -80s
-90 100 110 a, U 3e
I----
(a) O 300 600 900 1200 1500 1800Frequency
[MHz]-
-70E
m 23-80s
-90 a, -0 SE
%oo
II I -110I'
I
(b)
0 300 600 900 1200 1500 1800Figure 8. Frequency spectrum analysis to select measuring frequencies
for narrow band VHF /UHF PD detection. (a) Background noise spectrum,
@) spectrum with protrusion on the conductor.
jected to noise signals. Therefore, this frequency is not suitable as mea-
suring frequency for narrow band VHF /UHF PD measurements.
From the measured frequency spectrum and the frequency scan, a frequency well above the noise level can be selected for a narrow band
VHF /UHF PD measurement. For further analysis of the signals at this se-
lected measuring frequency, a SA can be used to demodulate the signals
into narrower range.
The center frequency f c of the SA is set to the selected frequency; the
measured span is set to zero: the zero span mode of the SA, and the
sweep time is set to 20 ms to obtain phase-resolved patterns related to the 50 Hz power cycle.
In zero span mode the measured frequencyband is determined by the resolution bandwidth (RBW). In fact, the VHF /UHF signals are passed to
a bandpass filter with a bandwidth equal to the RBW and the frequencies
between f c - 0.5b, and f c
+
0.56, are further processed. Therefore,the SA can be seen as a bandpass filter with a tunable center frequency
f c . In this study an RBW b, = 5 MHz was used. Frequency [MHz]
2.6
WIDE BAND
UHFPARTIAL
DISCHARGE DETECTION
The wide band UHF detection circuit is composed of a capacitive disc
coupler, a signal conditioning unit
(scu)
controlled by aPC
and a mod-Figure 9. Analysis of a measured frequency spectrum: (a) from the fre- quency spectrum 5 different measuring frequencies were selected; (b) at these measuring frequencies the maximum PD level was measured for t
(a frequency scan) and (c) the phase-resolved patterns can be further ana- lyzed; in this case HqmaZ
( 4 )
distribution of the maximum values of dis- charges.electromagnetic (EM) waves. The SCU measures the signals between 500
and 1500 MHz, see Figure 6. The envelope of the measured UHF sig-
nals is passed to a peak-detect circuit. Although the maximum value of these signals depends on the frequency characteristics of the GIS and the coupler, it is taken as indication for the PD magnitude. The data can be
transferred between the
scu
and a PC. The same signals from the SCUcan also be used as input signals for the PD detector. Using this config- uration, the digital tools used to analyze PD patterns can be used also
for PD patterns detected with the wide band UHF system.
3
PD DISCRIMINATION TOOLS
It is known that a strong relationship exists between the shape of
phase-resolved PD patterns and the originating discharge source (type
of defect) [18]. Each discharge source with its geometry, location, dielec- tric properties and applied field, is characterized by a specific sequence of PD. Analysis of these sequences is thus a good means of discrimina-
tion between different discharge sources. The PD pulses are grouped
with respect to their intensity and their phase-angle. Different phase-
resolved PD distributions or patterns can be processed, e.g.: the max-
imum pulse height Ifq,,,
(4);
the mean pulse height Hqn(4);
andthe pulse count Hn
( 4 ) .
After a PD measurement has been finished, 29 statistical operators are processed to describe the properties of the PD measurement: a 'finger-
print' [23]. When several fingerprints are available, it is possible to de- velop a collection containing user specific data: the PD data bank. In this way an unknown discharge measurement can be compared to a collec- tion of known situations, represented by their fingerprints. A data bank is judged to be well designed if it produces a high similarity for the cor- rect defect and low or nil for all the others. If no recognition is possible, the result should be low for all defects.
Of course, the result of such a recognition process strongly depends on the test conditions at which the reference data are obtained, the num- ber of measurements used to represent a defect and on the way the data bank is organized.
I
03 50z 100%
Figure IO. Example of cluster analysis of fingerprints using the tree
method as obtained from different defects. 1: Electrically floating part; 2 Free moving particles.
For this purpose, several discrimination techniques can be used [23].
The main goal of all these methods is the recognition of clusters within the group of measurements without a priori knowledge. This means that no indication of the membership of an individual measurement to a particular cluster is present beforehand. To analyze the measurements discussed in this report, the group average analysis technique was used [23]. The result of the analysis is a tree structure, which illustrates the re- lationships between individual fingerprints, Figure 10. The percentage scale in the lower part of this Figure shows the dissimilarity between fin- gerprints that were fused together. In the example shown in Figure 10, the last two groups were measurements of an electrically floating part, each fingerprint indicated by A, and measurements of free moving par- ticles, each fingerprint indicated by B. It follows that similar fingerprints will be connected at relatively low dissimilarity levels (e.g. all A levels in Figure lo), while different fingerprints will be connected at relatively high dissimilarity levels (1 and 2 in Figure 10). By cutting such a tree structure at a certain level, the data can be divided into different clus- ters.
Three-dimensional pictures can also be analyzed with fractal features [22]. The fractal method processes two parameters.
The fractal dimension corresponds to the roughness of the surface. A smooth surface has a low fractal dimension. A flat surface has the lower limit 2, and a very rough surface has a high fractal dimension with upper limit 3.
The lacunarity corresponds to the density of the surface. An empty surface has a low lacunarity (lower limit
O),
a dense surface has a high lacunarity (upper limit 1).Figure 11 shows the fractal method applied to the same defects as in Figure 10. Again, two different clusters with similar 3 D-patterns are formed.
4
STUDIED FAULTS
In this paper the following faults are studied. Besides the protrusion
on the enclosure which was only studied in the 245 kV GIs setup, all
836 -55 --65 -. E
a,
2a
-75 . 3 3Meijer et al.: PD Pattern Analysis in G I s
0 0507
0
oo261
0220 010 I
2 510 2 547 2581 2 621 2 6 5 8 2 695 2 7 3 2 2 169 2806 2 843 2 880 Fractal DimenSlOn
Figure 1 1 . Example of fractal analysis as obtained from different de- fects. l: Electrically floating part; 2: Free moving particles.
Figure 12. Photograph of the HV conductor with a fixed protrusion of 10 mm length and 0.5 mm diameter.
A metallic protrusion was fixed to the conductor, resulting in a long
sharp point protruding into the gas of 10 mm length and 0.5 mm in di-
ameter, see Figure 12.
Figure 13. Photograph of a free moving particle inside the GIs installa- tion.
To simulate a free moving particle inside the GIs-enclosure, particles as shown in Figure 13 were used. The length of the metallic particle was
15 mm and the diameter 0.4 mm.
A free moving particle in the GIS enclosure can stick to an insulator if the insulator is not perfectly clean. After some time of voltage applica- tion the particle can produce PD and even initiate flash over. To simulate this situation, a metallic particle was located on the insulator of 10 mm
5 COMPARISON
OF
MEASURING SYSTEMS
To compare the PD patterns, as measured using an IEC 270 detection
circuit and a VHF /UHF detection circuit in a GIs setup, a protrusion on the conductor was used. Figure 15 shows the frequency spectrum that
was produced by a protrusion on the conductor present in the 245 kV
setup. In this particular case the applied test voltage was 172
kV
In thisSection three groups of measurements are compared: IEC 270, narrow
band VHF /UHF and wide band UHF measurements. Within the frame
of this study measurements at several VHF and UHF measuring frequen-
cies have been performed, at different voltage levels and in both GIS
setups. From frequency spectra and frequency scans several measur- ing frequencies were selected: in the case of the 245 kV setup 170,294,
460 and 617 MHz and in the case of the 420 kV setup 361,829 MHz and
1.2232 GHz. Concerning the 245
kV
setup voltage levels of 128,172 and216 kV were applied and concerning the 420
kV
setup the voltage levelsnmow hand 1
Figure 16. Phase-resolved distributions from a protrusion on the con-
ductor as measured using three different measuring systems: IEC 270, VHF and UHF narrowband. The PD magnitudes are not calibrated.
Descriptions:
I t1F nnrrnnhand 100
I t C 2 7 0 iiii~tliod 100
\ I I C n;irrnvh:ind 108
Figure 17. Statistical analysis applied to a fingerprint of a PD pattern of
a protrusion on the conductor measured with the UHF detechon urcuit.
Comparison of the measurements was performed in two steps, com- parison of IEC 270
vs.
narrow band VHF /UHF detection and comparisonof narrow band VHF /UHF
vs.
wide band UHF detection.Unfortunately it was not possible to use the different detection cir- cuits simultaneously, but all measurements were done under exactly the
same measuring conditions. Typical patterns as measured with the IEC
270 system, VHF and UHF narrowband system are gathered in Figure 16.
Due to the fact that a subjective judgment was not possible on the base of visual interpretation of the measurements, numerical discrimination was applied, see Figure 17.
Mea.Wl"
s w m 3D distnbiitions
Figure 18. Phase-resolved distributions from a protrusion on the con- ductor as measured using narrow band and wide band UHF PD detection. The PD magnitudes are not calibrated.
Descriptians: % 0
25
50 75 100EEEm
UHF Narrow band 100
UHF Wide band 98
Figure 19. Statistical analysis applied to a fingerprint of a PD pattern of a protrusion on the conductor measured with the narrow band UHF method
For comparison of the narrow band and wide band UHF detection
system, a protrusion was fixed on the conductor of the 420 kV GIS setup. Typical measuring results as obtained with bothnarrow and wide
band systems are shown in Figure 18. As can be seen, the measure-
ment obtained with a wide band UHF system has several peaks in the
H,,,, ($) and Hqn
(4)
distributions. However, these signals disap- pear in the intensity distribution H n ( 4 ) . Therefore it can be concluded that these signals are single disturbances. Although the wide band sys- tem is more sensitive for disturbances, it can be concluded that, leaving these disturbances out of consideration, there are no significant differ-ences between the measuring results. To check this conclusion, statisti-
cal analysis was applied for further analysis of the PD patterns [7-91. Sta- tistical recognition confirms the small differences between narrow band
VHF, narrow band UHF and IEC 270 measurements, see Figure 17, and
between narrow and wide band UHF, see Figure 19. Figure 17 show the
result of statistical comparison of a narrow band VHF PD measurement
and a PD data bank that consists of narrow band VHF, narrow band UHF
and IEC 270 PD measurements. As a reisult the VHF PD measurement is recognized for 100% as a VHF, a UHF and an IEC 270 PD measurement. These experimental results show that in this particular case no differ-
ences between the PD patterns as measured with the three different sys-
tems can be concluded. To generalize this conclusion, more careful in- vestigation is necessary.
6 INFLUENCE OF THE
COUPLER LOCATION ON THE
FREQUENCY SPECTRA AND
THE PD PATTERNS
6.1
INFLUENCE
ON THEMEASURABLE
FREQUENCYSPECTRA
Using a free moving particle as shown in Figure 13, the influence of the coupler location on the frequency :;pectra was studied.
The frequency spectra from a free moving particle shown in Figure 20 are spectra measured at three different coupler locations as indicated in Figure 4. To obtain the frequency :spectra, the data measured dur- ing 10 sweeps of 5 s were averaged. It follows from Figure 20 that the frequency spectrum measured at coupler 52 shows no spectral content between 900 and 1100 MHz; no explanation can be given for the fact that coupler S2 shows a different behavior in this frequency range compared to S1 and S3. Moreover, for the mean value of the frequency spectra mea- sured at different coupler locations it can be concluded that the mean value of S1> S2
>
S3. However, the differences are very small so that it is not possible to draw any specific conclusions regarding the influence on the measured frequency responses of the location of the capacitive coupler or the number of insulators between the fault and the coupler. To obtain the frequency scan (Fs), the same free moving particle in thetest vessel was used as injection source. In particular FS was measured
at three different coupler locations, see Figure 21, which shows the PD
magnitude (not calibrated) as function of different frequencies. From the results of the FS can be concluded that in general the highest PD sig-
nals are measured at coupler location S1, that different couplers show
838 m
E
-75a
cb,
! 0 300 600 900 1200 1500 1800(a)
Frequency
[MHzJ
0 300 600 900 1200 1500 1800(b)
Frequency [MHz]
-
-75E
a
a,
a, -85-.
0 300 600 900 1200 1500 1800(c)
Frequency [MHz]
Figure 20. Frequency spectra measured at (a) coupler location S1 with 1 insulator, (b) coupler location S2 with 2 insulators, and (c) coupler location S3 with 4 insulators between the fault and the coupler.
from 1200 to 1400 MHz, an overlap in the frequency scans of all three
couplers is observed, and that the insulators between the PD source and
the different couplers attenuate the electromagnetic waves, especially at the lower frequencies.
6.2
INFLUENCE
OFTHE COUPLER
LOCATION
ONTHE SHAPE
OFTHE PD PATTERNS
Using a protrusion on the conductor as shown in Figure 12, the influ-
ence of the coupler location on the shape of the PD patterns was stud-
ied. It has been shown by frequency spectra and frequency scans that insulators introduce signal reduction. Furthermore, the signal reduction depends on the frequency: lower frequencies are more subjected to sig- nal reduction than higher frequencies. To compare the influence of the
coupler location on the shape of phase-resolved patterns PD measure-
ments were done at the three different coupler locations using the VHF
Meijer et al.: PD Pattern Analysis in G I s
100
1
1
80
a
60E
40 E CI U I3 c.-
a
20 0 I 0 0500
900 1300 7700 Frequency [MHz]Figure 21. Frequency behavior of the discharge magnitude as a function of measuring frequency of a free moving particle at (a) coupler location SI,
(b) coupler location 52 and (e) coupler location S3.
Figure 22. Phase-resolved distributions of a protrusion on the conduc- tor as measured at three different coupler locations SI, S2 and S3 and at two different measuring frequencies in the VHF and UHF range. The PD magni- tudes are not calibrated.
Table 1. Overview of the G I s components between the defect and the capacitive coupler for UHF PD measurements in 420 kV GIS setup.
/UHF narrow band system, see Figure 5. An overview of the GIS compo-
nents between the defect in the defect vessel and the capacitive coupler
used for UHF PD detection is given in Table 1. As an example PD mea-
surements as obtained at a measuring frequency in both VHF and UHF
ranges are shown in Figure 22.
From visual comparison no differences in the phase-resolved patterns
measured in the VHF range can be concluded. In the UHF range, the PD
patterns measured at S1 differ from the PD patterns measured at S2 and
S3. Due to the fact that S1 is located inside the test vessel, no signal re- duction between the defect and the coupler is introduced.
Therefore, a higher measuring sensitivity can be obtained at coupler location S1 compared to S2 and S3. This explains the possibility to detect
Descriptions: UHF couplcr S3 UHF coupler S2 UHF coupler S 1 VHF coupler 53 VHF coupler SZ VHF coupler SI
Figure 23. Statistical recognition applied to a fingerprint of a PD pattern measured at coupler S1 and a measuring frequency in the VHF range
Table 2. Signal reduction R (dB) between couplers S1 and S2 resp. S3
low magnitude PD at S1. However, this difference has no influence on
the result of statistical recognition, see Figure 23.
From the same measurements, the signal reduction between the cou- plers could be calculated. Table 2 shows the signal reductions compared to coupler location S1 as measured at the two other coupler locations S2 and S3. From comparison of the signal reductions a signal reduction of
0.4 dB between coupler location
S1
and S2 can be concluded. The onlyexplanation of this decrease can be the presence of two insulators and
6 m busbar between the defect and coupler S2.
Between coupler locations S1 and S3, a signal reduction of 4.4 dB was
measured. Two effects can explain this decrease.
The presence of four insulators and 12 m busbar between
S1
and 53 will introduce twice the signal reduction of 0.4 dB as was measured be- tween coupler location S1 and S2. Therefore, a signal reduction of 0.8 dB will be introduced by these components.At the T-junction at coupler location S2 the energy of the PD pulses will be divided into two parts and therefore a signal reduction of at least
3 dB can be expected.
From this analysis can be concluded that a T-junction introduces a sig- nal reduction of at least 3.8 dB on the signal path straight on. However it has been shown that the signal reduction strongly depends on the G I s configuration and values ranging from 3 to 12 dB have been measured P61.
7 ANALYSIS
OF
PD PATTERNS
It has been shown in the previous Sections that the PD patterns ob-
tained from measurements using three different detection systems and different test voltages show no differences for a particular fault. As a re- sult it is possible to define a typical phase-resolved PD pattern for each single fault which is independent of the PD detection system used and of
other parameters such as measuring frequency. The PD patterns as mea-
sured from the four studied faults are compared and cluster analysis is applied to investigate the possibility to discriminate between different faults.
7.1 VISUAL ANALYSIS
Figure 24 shows examples of PD measurements with typical phase-
resolved PD patterns of the four studied defects. In Figure 24(a) to (d)
180" 6 ,360"
Figure 24. Phase-resolved distributions of four typical defects in GIs.
(a) to (d) protrusion on the conductor, (e) a free moving particle, (f) a par- ticle fixed to an insulator, and (8) a protrusion on the enclosure. The PD magnitudes are not calibrated.
four typical phase-resolved PD patterns 'of a protrusion on the conductor
are shown. From visual comparison of the PD patterns, four different
groups could be formed. The differences in the patterns can be found in the asymmetry of the H ,
( 4 )
distributions. In group 1, the PD pattern issymmetric in the positive and negative half of the sine wave. For group
2 the number of PD in the positive half is larger than in the negative
half, in group 3 PD occurs only in the positive half of the sine wave, and for group 4, the number of PD in the negative half is larger than in the positive one.
The different types of PD patterns were investigated as function of
the GIS setup configuration, the gas pressure, and the test voltage. All
PD measurements classified in groups 1 to 3 were performed during one
day with the same protrusion sample, and no changes in the measuring
circuit or in the 245 kV GIS setup were -made during the measurements.
Therefore all measurements were done under exactly the same condi- tions and it can be concluded that neither the GIS nor the pressure have
840 Meijer et al.: P D Patteem Analysis in G I s
Table 3. Overview of the membership of a specific PD measurement to
one of the three (visually formed) groups.
been the reason for the three different groups.
Table 3 indicates the relation of the test voltage, thc measuring fre-
quency and the percentage of PD measurements that are part of one of
the indicated groups. The Table shows that the membership of d PD mea-
surement to one of the three groups is independent on the applied test
voltage. The PD measurements in group 4 were measured using a pro-
trusion on the HV conductor of the 420
kV
GIS setup. Therefore it can be concluded that the four PD patterns as shown in Figure 24(a) to (d) are typical PD patterns for a protrusion on the HV conductor.H w " [PCI (9) I
Figure 25. Typical phase-resolved distributions for the four studied faults: a protrusion on the conductor, a free moving particle, a protrusion on the enclosure and a particle fixed to an insulator. Protrusion on conduc- tor -, free moving particle . . . , protrusion on enclosure - - -, -
-
-.Typical phase-resolved PD patterns of the three other studied faults are shown in Figure 24(e) to (g). Visual comparison of all PD patterns results in the following.
The phase-resolved distributions of a protrusion on the conductor show four different typical patterns where the PD are concentrated
mainly around the positive peak and differences in the number of PD can
be found around the negative peak or the PD are concentrated mainly
around the negative peak. The phase-resolved distributions of a pro- trusion on the enclosure and a particle fixed to an insulator are similar
and the PD distributions are concentrated around the negative top of the
applied test voltage, while the phase-resolved patterns of a free mov-
ing particle are very typical: the patterns follow the sine wave of the applied test voltage Figure 25 summarizes these observations.
From visual comparison of PD patterns it is possible to recognize a free moving particle and a protrusion on the conductor. However it is not possible to distinguish between a protrusion on the enclosure and a particle fixed to an insulator. To study the possibility to use computer-
based discrimination tools two methods have been applied to the PD
measurements, fractal analysis and the tree method
0.072 0.056 0 O o Z 4 I 070 0 008 2 4 1 2.40 2 5 1 2 5 6 2 6 1 2 6 6 2 7 1 2 7 8 2.81 2.86 2 9 1 Fractal Dimension
Figure 26. Fractal analysis as applied to all PD measurements: 1: Pro-
trusion on the conductor; 2: Free moving particles; 3: Particle on insulator; 4: Protrusion on the enclosure.
R -
I
l ) A Panicle fixed to an insulator 2)C: Protrusion on the encIo8ure 3) 0: Free moving particle
F: Protrusion on the mndudor (gmup 3) 4)G: Protrusion on the tondudor (gmup 4)
5)IJ Pmtrusion on the conductor (gmup 1)
E Protrusion on the wnducior (group 2)
7 F h I I D D D E E E E D 5 0% 50% lob;!
Figure 27. Cluster analysis using the tree method applied to all PD mea- surements.
7.2
FRACTAL ANALYSIS
The result from the fractal analysis [29] is shown in Figure 26. Fractal
analysis shows that two clusters can be formed: A and B. From analy-
sis of both clusters it is found that cluster A contains only PD measure-
ments as obtained from a free moving particle. As explained before, the
PD patterns of a free moving particle are very typical whilst pattcrns of the other three faults show somc ovcrlapping parts. The second cluster consists of measurements of a protrusion on the conductor (marker l), a particle fixed to an insulator (marker 3) and a protrusion 011 the en- closure (marker 4). Because fractal analysis analyzes the surface of the 3D distributions, it is not sensitive to asymmetry in the patterns, As can be observed in Figure 24, the only difference between a protrusion on the conductor on one side and a protrusion on the enclosure and a par-
ticle fixed to an insulator on the other side is the asymmetry in the PD
patterns. Therefore, these three faults are clustered in one and the same cluster.
Table 4. Comparison of both discrimination tools to discriminate be- tween the four studied faults.
Free part. Prot.enc1.
tree
7.3
TREE ANALYSISThe result from the tree analysis [18] is shown in Figure 27. With the tree method five different clusters can be observed. Although the phase- resolved PD patterns of a particle fixed to an insulator (A) and a protru-
sion on the enclosure (C) show similarities (see Figure 24) both faults
can be discriminated using the tree method.
Cluster 3 combines measurements of a free moving particle (B) and group 3 of a protrusion on the conductor (F). Cluster 4 consists of mea- surements of group 4 of a protrusion on the conductor. Cluster 5 com-
bines measurements of group 1 and 2 of a protrusion on the conductor.
7.4
COMPARISON OF FRACTAL
AND TREE ANALYSIS
Within each cluster no influence of different parameters such as the
detection system and the measuring frequency could be found [28]. As
a result, it can be concluded from fractal analysis and the tree method that discrimination is possible between different faults. However, it is shown that some faults are better discriminated with the tree method
than with the fractal method. Table 4 summarizes which discrimination
method can be used to discriminate a particular fault from other faults.
8
CONCLUSIONS
ASED on the experiments discussed in this paper we found that no
B
differences between PD patterns were observed, using an IEC 270PD detection circuit, a narrow band VHF /UHF PD detection circuit and
a wide band UHF PD detection circuit. This observation can be of im-
portance in the future through PD pattern data exchange for evaluation
purposes.
It has been shown that depending on the GIS components where the
PD signals propagate, such as insulators and T-junctions, different sig-
nal reductions were observed. However, no changes in the shape of PD
patterns were observed.
This study shows that both discrimination methods, the tree method and fractal analysis, are complementary, In the case that one of the meth- ods fails, the other one can be used for discrimination. For example, no discrimination is possible between a protrusion on the conductor and a particle fixed to an insulator using fractal analysis. Using the tree method, a clear distinction between these faults is shown.
Of course, to generalize these conclusions, more systematic and care- ful investigations and convincing demonstrations are necessary.
ACKNOWLEDGMENT
The authors acknowledge Prof. Dr. Ing. K. Feser of the University of Stuttgart for experimental support in performing PD measurements.
REFERENCES
[1] D. Kopejtkova, T. Molony, S. Kobayashi, I. M. Welch, "A twenty-five year review
of experience with SF6 gas insulated substations (GIs)", CIGR, Vol. I, paper 23-101, 1992.
[2] B. E Hampton, J. S. Pearson, C. J. Jones, T. Irwin, I. M. Welch and B. M. Pryor, "Ex- perience and progress with UHF diagnostics in GIS", CIGR, Vol. I, paper 15/23-03, 1992.
[3] IEC, Partial discharge measurements, 2nd edition, IEc Publication 270,1981. [4] H.-D. Schlemper, R. Kurrer and K. Feser, "Sensitivity of on-site partial discharge de-
tection in GIS", Proc. 8th Int. Symp. on HV Engineering, Yokohama, Vol. 3, pp. 157- 160,1993.
[5] B. F. Hampton, R. J. Meats, "Diagnostic measurements at U H F in gas insulated sub- stations", IEE Proceedings, Vol. 135, Pt. C, No. 2, pp. 137-144,1988.
[6] J. S. Pearson, B. E Hampton, A. G. Sellars, "A continuous UHF monitor for gash- sulated substations", IEEE Transactions on Electrical Insulation, Vol. 26, No. 3, pp. [7] E. Gulski, Computer-aided Recognition of PD Using Statistical Tools, Delft University
Press, 1991.
[8] E. Gulski, "Computer-aided Measurement of Partial Discharges in HV Equipment", IEEE Trans. on Elec. Insulation, Vol. 28, No. 6, pp. 969-983,1993.
[9] E. Gulski, P. Seitz, "Computer-aided Registration and Analysis of Partial Discharges in H V Equipments", Proc. 8th Int. Symp. on HV Engineering, Vol. 3, pp. 13-16, Yoko-
hama, 1993.
[IO] W. Ziomek, H.-D. Schlemper, K. Feser, "Computer Aided Recognition of Defects in
GIS", IEEE International Symp. On Electr. Insulation, pp. 91-94, Montreal, 1996. [Ill R. Kurrer, K. Klunzinger, K. Feser, N. de Kock, D. Sologuren, "Sensitivity of the UHF-
method for Defects in G I s with Regard to On-line Partial Discharge Detection", IEEE International Symp. On Electr. Insulation, pp. 95-98, Montreal, 1996.
1121 N. de Kock, B. Coric, R. Pietsch, "UHF PD detection in gas-insulated switchgear - suit- ability and sensitivity of the UHF method in comparison with the IEC 270 method, IEEE Electrical Insulation Magazine, Vol. 12, No. 6, pp. 20-26,1996.
[13] A. Bargigia, W. Koltunowicz, A. Pigini, "Detection of Partial Discharges in Gas In- sulated Substations", IEEE Trans. On Power Delivery, Vol. 7, No. 3, pp. 1239- 1249, 1992.
[14] E. Colombo, W. Koltunowicz, A. Pigini, "Sensitivity of electrical and acoustical methods for G I s diagnostics with particular reference to on-site testing", CIGRE Symp. Berlin, paper 130-13,1993,
[15] A. Petit, "Field Experience of Partial Discharge Monitoring with the UHF method, 9th Int. Symp. on HV Engineering, paper 5596, Graz, 1995.
[16] E. Gulski, S. Meijer, W. R. Rutgers, R. Brooks, "Recognition of PD in SF6 insulation using digital data processing", Conference on Electrical Insulation and Dielectric Phenomi?na, pp. 577-580,1996.
[17] J. S. Pearson, 0. Farish, B. F. Hampton, M. D. Judd, D. Templeton, B. M. Pryor, I. M. Welch, "Partial Discharge Diagnostics for Gas Insulated Substations", IEEE Trans. on Dielectrics and Elec. Insulation, Vol. 2, No. 5, pp. 893-905,1995.
[18] E. Gulski, "Digital analysis of partial discharges", IEEE Trans. on Dielectrics and
Elec. Insulation, Vol. 2, No. 5, pp. 822-837,1995.
[19] S. Meijer, W. R. Rutgers and J. J. Smit, "Acquisition of partial discharges in SFs insu- lation", IConference on Electrical Insulation and Dielectric Phenomena, pp. 581-584, 1996.
[20] Cigre working group 15.03, "Effects of Particles on G I s insulation and the Evaluation of Relevant Diagnostic Tools", CIGRE, Vol. I, paper 15-103,1994.
[21] E. Gulski, H. l? Burger, A. Zielonka, R. Brooks, "Classification of defects in HV com- ponents by fractal analysis of PD measurements", Conference on Electrical Insula- tion and Dielectric Phenomena, pp. 484-487,1996.
[22] A. Krivda, E. Gulski, L. Satish, W. S. Zaengl, "The Use of Fractal Features for Recog- nition of 3-D Discharge Pattems", IEEE Trans. on Dielectrics and Elec. Insulation, 1231 A. Krivda, "Automated Recognition of Partial Discharges", IEEE Trans. on Di-
electricE, and Elec. Insulation, Vol. 2, No. 5, pp. 796-821,1995.
[24] R. Kurrer, K. Feser, T. Krau, "Antenna Theory of Flat Sensors for Parial Discharge De- tection at Ultra-High-Frequencies in GIs", 9th Int. Symp. on H V Engineering, Graz, paper 5615,1995.
469-478,1991.
842 Meijer et al.: PD Patfem Analysis in C I S
[25] Leon W. Couch 11, Digital und Analog communication systems, Macmillan Publishing Company, New York, 1993.
[26] R. Kurrer, K. Feser, "Attenuation Measurements of Ultra-High-Frequency Partial Discharge Signals in Gas-Insulated Substations", 10th hi. Symp. on HV Engineer- ing, Montreal, Vol. 2, pp. 161-164,1997.
[27] L. Satish, W. S. Zaengl, "Can Fractal Features be Used for Recognizing 3-d Partial Discharge Patterns?", IEEE Trans. On Dielectrics and Elec. Insulation, Vol. 2, No. 3, [28] S. Meijer, E. Gulski, J, J. Smit, R. Brooks, "Comparison of Conventional and VHF /
UHF Partial Discharge Detection Methods for SFe Gas Insulated Systems, 10th Int. Symp. on HV Engineering, Montreal, Vol. 4, pp. 157-190,1997,
[29] S. Meijer, E. Gulski, W. R. Rutgers, "Evaluation of Partial Discharge Measurements in SF6 Gas Insulated Systems", 10th Int. Symp. on HV Engineering, Montreal, Vol. 4, pp. 469473,1997.
Manuscr@f was received On 8Augusf 199z in final forln I 5 h e 19%
