RF-based monitoring and classification of partial discharge on wet silicone rubber surface

The main objective of this study is to investigate the feasibility of using an RF antenna to detect and classify partial discharges on polymer insulation surfaces. The detected discharges are identified based on their originating sources, employing a cascade of feature extraction, feature selection and the standard artificial neural network classifier. Both statistical and spectral analyses have been used for feature extraction. Feature extraction is followed by a feature selection stage with a novel implementation of stepwise regression method in order to derive representative feature vectors, yet with minimum dimensionality. Suppression of redundant features is therefore achieved, thereby enhancing the classification reliability. Finally, classification is performed using a standard feed forward neural network with back propagation training algorithm. The proposed method is found to be successful in classifying different types of partial discharge with recognition accuracy exceeding 96%. The proposed method can be an essential stage towards overhead line inspection to assess the status of outdoor polymer insulators, where partial discharges could be initiated from surface discharges due to pollution and/or from corona discharges from energized ends.