Abstract—Quite a few methods have been proposed to detect when frequency is between 300 MHz and 1.5 GHz, the S11
partial discharge in power transformers, such as dissolved gas curve of the antennas is shown in Fig. 2) amounted on the
analysis (DGA), pulse current method, ultrasonic wave method walls of the transformer and analyzed by several methods. The
and ultra-high frequency (UHF) method. UHF method, is widely sampling frequency was 2.5 GHz throughout the experiment.
used in condition monitoring recently for the sake of its wider
spectrum and higher sensitivity. This paper presents a PD
detection experiment inside a true size transformer which was II. WAVELET ANALYSIS , DE-NOISING , THRESHOLDS & A
rebuilt to install 9 UHF sensors. Four types of PD sources were NOVEL METHOD
placed into in the transformer. The pattern and intensity of PD
can be changed by adjusting PD voltage and insulation distance. A. Continuous Wavelet Transform, Discrete Wavelet
Signals detected by the sensors were decomposed and Transform & Multi-Resolution Analysis
reconstructed by methods of Continuous Wavelet Transform For pattern recognition, proper, rich and significant
(CWT) and Multi-Resolution Analysis (MRA) based on different characteristics should be extracted from original PD signals
wavelets. The time-frequency characteristic and de-noising and be input into neural networks. Therefore a UHF signal
performance of different transforms applying to PD signals were
should be processed and analyzed by mathematical tools to get
compared and the result shows that different transforms reflect
advantages in different respects. A new method is proposed to
extract PD signals from noises.
1012
2013 IEEE International Conference on Solid Dielectrics, Bologna, Italy, June 30 – July 4, 2013
1013
2013 IEEE International Conference on Solid Dielectrics, Bologna, Italy, June 30 – July 4, 2013
1014
2013 IEEE International Conference on Solid Dielectrics, Bologna, Italy, June 30 – July 4, 2013
IV. CONCLUSIONS
• Morlet wavelet, with a small effective support and more
vanishing moments, was a good wavelet basis to analyze
PD signals acquired from an oil-insulating transformer.
• To decompose and de-noise the PD signal, Daubechies
family was desirable unless the defect was complicated
like a point-plain in oil defect where symlets performance
better.
• With a communication noise whose frequency was much
higher than PD UHF signals, a high decomposition level
of MRA brought no sacrifice when dealing with
reconstruction issues.
• The novel method using both CWT and MRA can help us
recognize and extract the noises in the field. It behaved
well when dealing with signals acquired from the true size
oil-insulating transformer.
ACKNOWLEDGMENT
The authors wish to express their thanks to Jiangsu Electric
Power Company Research Institute for financial support and
Jiangsu Huapeng Group for technical support.
REFERENCES
1015