Cable Fault Recognition Based On Partial Discharge Analysis and Machine Learning

Chung, Jie Yi (2018) Cable Fault Recognition Based On Partial Discharge Analysis and Machine Learning. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Chung Jie Yi.pdf Restricted to Registered users only Download (1MB)

Abstract

Cable fault can be defined as defects which might affect resistance within the cable that will cause electrical breakdown if the problem is not resolved. Partial discharge (PD) analysis is being widely used as an insulation method due to the correlation between PD pattern and the nature of the fault. Partial discharge data is collected from different cable defects are provided for analysis. Conventionally statistical features are extracted from PD data to train the classifier. This approach has good classification accuracy under noise free condition but suffers from severe degradation when noise contamination is present. Principal Component Analysis (PCA) is applied for features extraction. PCA was performed on different arrangement of Phase-resolved PD data sets to determine which combination gives the good accuracy. T-sne was performed based on the statistical data to train the classifier. Noise contamination has been added to observe the noise tolerance of PCA features, statistical features and T-distributed stochastic neighbor embedding. The machine learning based classifiers such as Artificial Neural Network (ANN) and Support Vector Machine (SVM) are implemented in Matlab. The performance based on both of the classifiers and method of feature extraction was evaluated. As a result, the SVM has the higher accuracy compared to the ANN for the classification part. Yet, the t-sne based on statistical feature extraction performed well compared to PCA and statistical feature.