Classification of Insulation Defect Type in Power Apparatus Based on Partial Discharge Signals by CNN

Partial Discharge (PD) is important method to diagnose electric power equipment and to find hidden dangers in it. In this phenomenon, a strong correlation exists between PD sources and PD signals. These characteristics have been used in order to identify PD source. In this study, we report on the results of classification of PD source which occurs in several electric power equipment using Convolutional Neural Network (CNN). Thirty-six type of PD models with different in discharge factor and insulator were fabricated in total, and 1000 PD signals were obtained from each model. The Characteristics based on phase angle φ, discharge magnitude q, and frequency n were extracted from PD signals and entered CNN. The result shows it can be possible to identify the PD source in several electric power equipment.