Partial discharge defects classification using neuro-fuzzy inference system

Partial discharge measurement is among the most important diagnostics methods of insulation systems in high voltage equipments, which makes it convenient to assess the insulation status and its prospective condition. Partial discharge activities may stem from various defects, and correspondingly behave differently. Since the origins of the PD activities are of major concern in insulation diagnostics, a large number of recognition methodologies have been proposed and used for this purpose. Among them, Phase Resolve Partial Discharge (PRPD) analysis has gained more attractions. However, sometimes the complexity of the patterns is of much more sophistication where utilization of intelligent based or expert system is inevitable. In this paper an Adaptive Neuro-Fuzzy Inference System (ANFIS) based classification method has been considered, where statistical features are extracted from the measured PRPD data first, then Fuzzy IF-THEN classification rules which are obtained from experts are employed in training procedures of the ANFIS model, and finally the classification of the PD defects is done automatically by the trained system, which makes it ready to be interpreted for decision making purposes.