PD characterization using short duration Fourier transform of acoustic emission signals

The frequency spectrum of acoustic emission signals obtained using short duration Fourier transform as the artificial neural network input parameter has been investigated. The acoustic spectrum is not significantly influenced by applied electric stresses compared with (/spl phi/-q-n patterns or statistical operators such as skewness and kurtosis. A feed forward neural network using the back propagation algorithm was initially applied to characterize acoustic emission signals produced from different shapes of void within a polyethylene dielectric. It was then used to discriminate between the background noise and acoustic emission signals produced by void discharges or treeing discharges. The obtained identification results for the different experimental arrangements are encouraging.