Identification of metal particles in transformer oil using partial discharge signals

Abstract

In this paper, the partial discharge current signals (pulses), along with pattern recognition methods for the assessment and diagnosis of PD (partial discharge) source of metal particles in the oil tank of the transformer is used. Three defect models, i.e.; a fixed metal particle, a free metal particle and a sharp metal particle is used for modeling all types of metal particles in the oil tank of a transformer. Also, corona discharge in air (as an unavoidable disturbance) is considered in a separate class of sources for partial discharge. After extracting the single PD pulse, FFT, DWT and PCA feature extraction methods are used to classify the various defects. SVM Classification, as a nonlinear and non-parametric methods of machine learning algorithms, is used for classification of PD single pulse signals recorded when the fault model is examined. By using FFT and PCA feature extraction methods with SVM classifier, more than 99% accuracy obtained in process of discovering the origin of the partial discharge pulses.