Partial Discharge Patterns Recognition with Deep Convolutional Neural Networks

Abstract

Artificial neural networks have been widely used in the field of partial discharge This study uses the 2D-convolution neural network of deep learning architecture to extract features and classify them to achieve diagnosis The main purpose of this study is to identify a partial discharge failure mode using a diagnostic system This thesis extracts the signals of four different partial discharge modes of motor from IEC60034 such as internal voids PD in stator winding insulation system internal delamination PD in the main insulation delamination PD between conductor and insulation in the main insulation and surface PD in slot Analytical classification was performed using a 2D-convolution neural network From the experimental results the CNN can effectively diagnose four different failure modes of partial discharge The best recognition rate and loss rate are 98 30% and1 41% and the model has high precision and recall which about 99% and 98%

Date of Award	2019
Original language	English
Supervisor	Jiann-Fuh Chen (Supervisor)