Cable Fault Recognition Using Image Analysis Of Phase Resolved Partial Discharge Pattern

Lee, Heng Soon (2018) Cable Fault Recognition Using Image Analysis Of Phase Resolved Partial Discharge Pattern. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Lee Heng Soon.pdf Restricted to Registered users only Download (2MB)

Abstract

Partial Discharge (PD) source classification is vital to diagnose the insulation quality to prevent any cable joint insulation breakdown that could cause major loss to power companies. There are many research works done on PD pattern recognition but it is usually done in noise free condition and through literature review, once noise is introduced the accuracy is not promising. Throughout literature review, several image analysis methods like fractal dimensions and lacunarity and wavelet shape decomposition and others algorithm were used but their work can still be improved as the accuracy when the PD data is contaminated dropped significantly to unusable levels. PD measured from five cross-linked polyethylene (XLPE) cables with 5 different defects including insulation incision defect, axial direction shift defect, semiconductor layer tip defect and air gap defect and metal particle on XLPE defect were used as input PD data in this research work. To replicate the real environment, noise was added into the PD data. Noise duration, noise amplitude and the combination of the previous two were the three type of noise condition used here. New feature extraction method using image analysis of phase resolved partial discharge pattern to extract features for PD classification was performed to test its noise tolerance rate. In this work, the images used as input for feature extraction were generated from the PD data which are the pulse count and pulse height distribution. The performance of conventional statistical features was compared with the new image features. The features generated were undergone classification using two different artificial intelligence classifiers, which are the Support Vector Machine (SVM) and Artificial Neural Network (ANN). The performance of ANN and SVM were compared within same feature type, then the best results were compared with the best results from another type of feature. It was found that among ANN and SVM, ANN provides better results than SVM in each type of feature with respective noise condition. As for the comparison between statistical and proposed image feature, the image features displayed a better performance.