Partial Discharge Source Classification for Switchgears with Transient Earth Voltage Sensor Using Convolutional Neural Network

Abstract

Partial discharge (PD) signals are used for insulation diagnosis of switchgears. Transient earth voltage (TEV) sensors are studied to detect PD signals, since PD signals can be easily measured by attaching the sensor to a metal casing of the switchgear. As with sensing and denoising techniques, classifying techniques are also important to determine the types of defects causing PD for insulation diagnosis. In this paper, the Convolutional Neural Network (CNN) is introduced to classify the types of defects with the TEV sensor signals. Focusing on solid insulators, two types of artificial PD models were designed. The CNN classifier was trained and tested with data derived from these artificial PD models. In addition, that was also tested with data derived from an actual voltage transformer (VT) component to confirm capability of practical use. Also, the trained classifiers are investigated to confirm what features it obtains through the training by its partial derivatives.