Conference Record of the 2 "
IEEE International Symposium on Electrical Insulation, Indianapolis, IN USA, 19-22 September 2004
PD Pattern Recognition in Transformer by Using UHF Technology
C.R. Li, Wei Wang, Z.G Tang, Y.S. Ding
High Voltage & EMC Laboratory School of Electrical Engineering North China Electric Power University Dewai, Zhuxinzhuang, Beijing, 102206, China crli&ublic. fhnet.cn.net
We have designed 5 kinds of partial discharge (PD) models to simulate typical PD faults in transformers. We have measured these PD models in our laboratory by using both our UHF measuring system and a conventional PD measuring system. The results detected by the UHF-system indicated, that PD patterns of a specific PDdefects could be identified clearly from their power spectrum. The statistical distribution of PD-signals like Qm.,-i+?.the n q obtained from UHF transient signal envelops show different distributions for different PD faults. These statistical distributions are consistent with the well-known results obtained from the conventional PD measuring system by other scientists.
rate of a single Channel is 4 GHz. A conventional PD detector (DST-4 Model, filter bandwidth 40 kHz-800 kHz) was used in our experiments simultaneously with the UHF monitoring system for the comparison of the results. As a voltage source for the test cell with PD-defect models the trial transformer (5kVNSOkV) was used: The data detected by the UHF PD system are transferred via a GPIB.card to a computer. With the software, both the frequency spectrum of the UHF-PD-signals, and the statistical distribution of the demodulated UHF-PD-signals (envelopes) in dependence on the phase position of the test voltage can be presented. The frequency spectrum was obtained through FFT.
Some PD pattem recognition using UHF technology has been done for transformers in a laboratory [I]. There was a spectmm analyzer used for that study. The PD detecting system with a spectrum analyzer cannot capture the transient PD signals and is difficult to operate on-site for a long term monitoring. In order to solve these problems, we developed an UHF PD monitoring system without the spectrum analyzer. Our UHF monitoring system can capture both, the waveforms of the UHF-PD-signals and the envelops of the UHF-PD-signals by using a demodulation detector. The investigation described in this paper focuses on PD pattern recognition in transformers using our UHF PD detecting system with the aim to find a technical solution for on-site UHF-PD-signal monitoring system.
I P"k& I
Figure 1 - Setup of PD testing system
Based on analysis of the typical transformer faults in China, we designed 5 kinds of partial discharge models to simulate the PD faults in transformers as follows [Z]: Oil corona: needle-plate electrode configuration with pressboard between the electrodes (Fig 3a). Surface discharge: pressboard between two parallel electrodes (Fig. 3b). Cavitv-discharge: three pressboard layers with implemented 4 cavities between two parallel electrodes (Fig. 3c). Oil wedge discharge: hemispherical electrode ( @ =20mm) and plate electrode with pressboard between the electrodes (Fig. 3d). Floating conducting uarticle: screw is fixed on pressboard close to the HV electrodes (Fig. 3e) The PD models were placed in a test tank filled with oil. (See Fig. 4).
CONFIGURATION OF MONITORING SYSTEM
In Figure 1 the setup of our PD testing system is presented. The UHF sensor used in our study is an Archimedes spiral antenna with bandwidth from SOOMHZ to 1500MHz. The standing wave ratio (SWR) of the UHF sensor is shown in 'Figure 2. An UHF amplifier is connected to the antenna (bandwidth 300MHz to 1500MHz. SO& gain). The amplifier has two output channels: 1) RF channel for UHF-PD-signals, 2) demodulator channel for registration of the envelope of the UHF-PD-signals. A digital oscilloscope (LeCroy LC574A) was used to capture and record U H F PD signals. The effective bandwidth of the oscilloscope is lGHz and the highest physical sampling
The UHF sensor was installed into the oil tank and connected to the amplifier, which is positioned outside &e tank. Additionally the measuring impedance (TUC-circuit) of the conventional PD detector (DST-4) was Connected to the test object (oil tank with the PD-defect models) via coupling capacitor.
Figure 2 -The S W R of the UHF sensor
Figure 4 - 50kV test t n ak
We found two phenomena that are very helpful for PD defect recognition by analyzing the frequency spectra: I) The spectra of the same kind of discharges are quite similar to each other. Reproducibility of the frequency spectra is very good. (See Figure 5 ) 2) Different kinds of discharges generate different frequency spectra. Typical frequency spectra for each PD-defect model are shown in Firmre 6.
Figure 3 - Models of defects: (a) oil corona, (b)surface discharge, (c) cavity discharge, (d) oil wedge, (e) floating conducting particle
RECOGNITION BY FREQUENCY SPECTRA
The typical UHF-PD signals, generated by the PD-tests on five above described PD-defect models, were captured by our monitoring systr". Bv amlication of the Fast Fourier _ . _. Transformation (FFT) to the recorded PD-sipals in time domain, the spectra of these signals were obtained.
Figure 5 - Superimposed frequency specea of cavity discharge x: frequency/(OGH~-2GHz), y: number of spectra/(l-50). z: power densityi(O-0.18dB)
From the research mentioned above, we can draw a conclusion that PD-defect can be recognized by its spectrum. Recognition criterions are shown in Table I. Table I PD-defect Recognition Criterion by Spectrum
PD type Oil corona
Recognition criterion The power spectrum of oil corona discharge is distributed between S O O M H Z to IlOOMHz and peaks between 500MHz to 600MHz are relatively high.
The power spectrum of surface discharge Surface mainly concentrates around 500MHz and discharge there is a characteristic peak. The main spectrum of cavity discharges is Cavity mainly located between 300MHz to 600MHz discharge and there are multiple peaks with approximately same magnitude. The power spectrum of oil wedge discharge is mainly distributed between S O O M H z to Oil 900MHz, and peaks are close to each other. wedge The density of the peaks is higher than the cavity discharge. The power spectrum of floating conduction Floating particle is distributed extensively between discharge 300MHz to 1500MHz and the peak at 300MHz is very steady.
(c) Surface discharge
(d) Cavity discharge
(e) Oil wedge discharge
( f ) Floating discharge
Figure 6 - Frequency spectra of different types of PDdefects
RECOGNITION BY STATISTICAL DISTRIBUTIONS
The demodulated UHF-PD signals (envelopes) obtained from the second channel of the preamplifier during the tests on PD-defect models can be digitalized hy a lower sampling rate . Both results the maximal pulse height distribution as a function of the phase angle Q M - 9 and pulse counts distribution as a function of the phase angle n-cp are shown in Figure7.
Table n PD-defect Recognition Criterion by Statistical Distributions PD type Oil corona Recognition criterion Oil corona discharges occur in the positive power cycle.
The PD pulses from surface discharges occur Surface near the peak value of the power cycle, discharge mostly occurring from 20' to 120" and from
200" to 300".
The shape of the phase-resolved pattem of Cavity cavity discharge is .like a cone, most of them discharge occur before and close to the peak value of the power cycle. Oil wedge The PD pulses from oil wedge discharges mostly occnr close to zero region of power cycle.
rm Comparison of the statistical distributions obtained f o the conventional PD measuring system with the UHF PD statistical distributions shows consistent results. An example is presented in Figure 8.
These results have demonstrated, that it is possible to identify PD faults by Q M U 9 and n q statistical distributions drawn from the demodulated UHF-PD signals. Recognition criterions are summarized in table 1 . 1
Floating The PD from the floating discharges almost discharge occurs in the whole power cycle.
phase-resolved PD fingerprints, which are well known from the conventional phase, resolved PD-measurements .
(a) Oil corona discharge
(h) Surface discharge
.................................... J 1. . . . . . . . . . .
Figure 8 - Comparison of statistical distnbutious of the floating discharge from UHF and conventional method (top: UHF, bottom:
The type of PD can he recognized by both the UHF PD signal kequency spectrum and the statistical distributions QMa.-(p,n-cp of the demodulated UHF PD signals. There was a good reproducibility of the specific frequency spectrum for each PD-defect model (see Figure 5). The frequency spectra of each PD model are different (see Figure 6).
(c) Cavity discharge
(d) Oil wedge discharge
The statistical distributions of the demodulated UHF-PD-signal (enveloRes) are very good comparable with the statistical distributions of PD-signals detected by the conventional system (see Figure 8).
1. Raja K, Devaux F and Lelaidier S, “Recognition of discharge sources using UHF PD signamres”, IEEE Electrical Insulation -Magazine, Vol. 18, NOS, SeptemherlOctober 2002, pp. 8-14 2. Wang G.L., Hao Y.P., etc. “Study on pulse current of typical PD models in power transformer”, High voltage engheering,Vol. 27,No. 2,April2001, pp. 5-8. 3. Wang W, Li C.R., etc. “Simulation on the application of detection techniaue to detecting UHF PD signals”. High Voltage Engineering, Vol. 30, No.2, February 2004, pp. 32-33 4. Bartnikas R, “Partial discharge, their mechanism, detection and measurement”, IEEE Transactions on dielectrics and electrical insulation, Vo1.9, No.5, October 2002, pp763-808
(e) Floating discharge
Figure 7 - Statistical distributions of different PD models
COMPARISON OF THE TWO PD-DEFECT RECOGNITION METHOD
Based on the results obtained in our laboratory, we believe that it is possible in the laboratory to recognize the PD sources using analysis of the frequency spectrum, hut it needs very high sampling rate equipment. For on-line UHF monitoring, only the UHF-PD signal envelopes (demodulated PD-signals) could be used to get similar 49