A Novel Technique for Online Partial Discharge Pattern Recognition in Large Electrical Motors

In this paper, a fully automated system for source detection of the partial discharges (PD) as an online diagnosis test in rotating machineries is proposed. This technique uses a modified version of the Expectation Maximization-based (EM) clustering technique to separate the multi source Phase-Resolved Partial Discharge (PRPD) measurements into multiple single-source clusters. Afterwards, the fuzzy rule-based classifier determines the degree of membership of individual clusters to the possible PD origins based on the extracted features and exploiting expert knowledge. For the first time, the concept of cluster analysis is introduced for separation of PD data coming from different sources. Interestingly, the results demonstrate the robustness of the proposed technique in classifying multi-source data even in presence of strong noise in online measurements. Among 5 available datasets with multiple PD sources, the proposed technique were successful in correct classification of 90% of the sources.