Effects of Different Partial Discharge Noise Type on Insulation Fault Classification

Law, Ivan Kai Wen (2021) Effects of Different Partial Discharge Noise Type on Insulation Fault Classification. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Ivan Law Kai Wen.pdf Restricted to Registered users only Download (5MB)

Abstract

Partial discharge occurs in the insulation when there is a partial current leakage with a very small amount. Partial discharge patterns are PRPD data produced by researchers which are phase (degree), charge magnitude (pC), and pulse count (N). When the noise-contaminated in the Partial discharge pattern, the accuracy eventually fell gradually as the duration of the noise increased and the magnitude of the charge increased. The objective is to simulate different noise types to observe the effect on PD classification accuracy. And to propose ways that can improve the classification accuracy under noise-contamination conditions. The software required for this project would be MATLAB and Python. MATLAB will use to generate PRPD patterns, while Python is used to do image classification. The methodology is involved research design flowchart, PD Data Preparation, Convolutional Neural Network (CNN) and Image Classification. By using CNN, the results showed that the highest classification achieved 97% while the lowest classification achieved 29%. The average of the highest accuracy for all results are 90%. The best performance under noise-contamination duration between 0 -10 seconds because it has achieved more than 80%.