Classification of Partial Discharges in Insulation Materials via Support Vector Machine and Discrete Wavelet Transform

Abstract

Long term partial discharges (PDs) within an insulation material of high voltage equipment can cause equipment failure. Thus, it is important to detect PDs within the insulation material and classify the PD type with high accuracy so that repair and maintenance can be performed effectively. In this work, three different types of PD, which include internal, surface and corona discharges, are measured from insulation materials. To evaluate the effect of noise on the PD measurement data, different levels of Additive White Gaussian Noise were added to the signals. Then, feature extractions were performed from the PD signals using Discrete Wavelet Transform (DWT). Different types of DWT families were used for feature extraction. The extracted features were then fed into support vector machine (SVM) for training and testing purposes. The classification accuracy of each test was recorded and compared. It was found that classification of PD signals using SVM as a classifier and DWT as a feature extraction yields reasonable classification accuracy results under different noise levels, which is in the range of 90%-99%.