Experimental study and feature extraction on UHF partial discharge detection for typical model in air

In this paper, in order to represent three kinds of typical partial discharge respectively happened in dielectric, out of dielectric and surface of dielectric, five kinds of typical partial discharge models, which included needle to plate, needle to needle, plate to plate, surface and air void, were designed and manufactured with aluminum. Based on the designed models, an experiment system of ultra high frequency partial discharge detection was comprised with non-partial discharge transformer, 2.5 GHz bandwidth oscilloscope, broad band amplifier and the developed double helical antenna. Afterwards, partial discharge experiments were carried out in air, and the method recommended in IEC60270 was referenced to measure the apparent quantity of partial discharge in pC. The experimental waveforms were given out. At last, the characteristics of the ultra high frequency signals excited by partial discharge were analyzed in time domain, frequency domain and wavelet coefficients. Using kernel principal component analysis, the feature extraction was realized. Then, several characteristics, which included deflection, protruding, frequency distributing, decay time, wavelet coefficients and so on, were proposed to describe the ultra high frequency signals. The conclusion educed in this paper is significant to research the feature extraction and pattern recognition of ultra high frequency partial discharge detection for high voltage apparatus.