Abstract
Partial Discharge is the breakdown confined to the localized regions of the insulating medium. The analysis and classification of Partial Discharge based on its site of occurrence is a complex process notwithstanding the intricacies involved in the digital computer acquisition system, since partial discharge is inherently a stochastic and a non-Makovian process in which there can be significant statistical variability. Since each defect has a particular deterioration mechanism, it is imperative to discern the correlation between the discharge patterns and the nature of defect in order to ascertain the quality of insulation. Neural Network which is a non-parametric method, has been a recent innovative technique suitable for partial discharge pattern classification. The Back Propagation Algorithm Neural Network is verified and validated for the various inputs based on established validation techniques. As against the conventional technique of using mathematical descriptors adopted by, previous researchers, unique characteristic input feature vectors have been devised in this work in order to provide a possible solution to a simple yet an effective preprocessing methodology. The efficacy of the classification of the standard partial discharge defect sources has been evaluated and analyzed.