Analysis and Classification of Simulated Partial Discharge Signals Using DWT and SVM

Based on the importance of early detection of structural changes and insulation damage of high voltage (HV) equipment by measuring partial discharge (PD), in order to perform optimal maintenance that has the effect of both increasing the safety of non-operation and increasing the lifetime of HV equipment, this paper presents a method for automatic classification of theoretically simulated PD pulse signals based on support vector machines (SVM). A time and frequency characterisation of the main PD sources using the signal processing tool PD flex and the Matlab environment will be presented. Also, the paper presents the patterns for the main types of PD sources, and for the main types of signals that can be assimilated with the corresponding PD pulses are analytically described in the form of four categories of signals, their theoretical waveforms, their main characteristics in the time and frequency domain that will be used for training and signal recognition using SVM along with de-noising of these signals using Discrete Wavelet Transform (DWT). The main working assumptions needed in future approaches to achieve automatic classification of even PD source types according to the characteristics of PD pulse signals and the type of equipment under test are presented.