PD pattern recognition of noise-buried acoustic signals from statistical indexes

An acoustic sensor system was used to detect three types of PD: point-point, point-plane and surface discharges. A wavelet-based de-noise method was carried out in signal extraction. The result of the extraction was provided with its high signal-to-noise ratio and excellent preservation of the energy and pulse amplitude. A group of frequency-domain indexes, which were generated through a spectral analysis describing the de-noised AE pulses of the three PD types, were defined and compared. The selected PD types were associated with AE pulses of characteristic shapes of frequency spectra and characteristic values of the associated descriptors. Then various types of PD can be identified by comparative analysis of the associated frequency spectra and bandwidths of dominant frequencies. Finally, using mean values and maximum values for AE pulses can be compared for different PD types when the lengths of the analyzed time intervals were same. For a specific PD type, the above values was depend on the time intervals.