Classification of defect types in XLPE cable joints using partial discharge measurement / Nurliana Abu Bakar

Nurliana , Abu Bakar (2018) Classification of defect types in XLPE cable joints using partial discharge measurement / Nurliana Abu Bakar. Masters thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (675Kb)
Preview	PDF (Thesis M.A) Download (2145Kb) | Preview

Abstract

Cross-linked polyethylene (XLPE) cables are broadly used in power industries due to their excellent mechanical and electrical properties. Cable joints are the weakest part in XLPE cables and prone to failures of the insulation. Breakdown in cable joint insulation can cause large losses to the power companies. Therefore, it is important to analyse the quality of insulation for the early detection of insulation failure. It is known that there is a relationship between partial discharge (PD) and the quality of the insulation. PD is one of the important phenomena that engineers should take care of in high voltage (HV) engineering. PD analysis is an important tool for evaluating the quality of insulation in cable joints. In this work, three XLPE cable joints with artificial created defects, which are commonly found on site, have been prepared. The input data from PD measurement results were used to train the artificial intelligence methods to classify each type of defect in the samples of cable joints. The feature extractions composed of statistical features and principle components analysis (PCA) after discrete Fourier transform (DFT), discrete wavelet transform (DWT) and wavelet packet transform (WPT) were applied on PD signals. Classifications were implemented using two different types of classifiers, support vector machine (SVM) and artificial neural network (ANN). The performance of each feature extraction method and classifier were evaluated. The proposed methods were compared with the existing methods to confirm the advantages of the proposed methods over the available methods. From the comparison of the results obtained, it was found that statistical features with DFT signals classified by ANN yield the highest accuracy among all of the methods tested.

Actions (For repository staff only : Login required)