An unsupervised Gaussian mixer model for detection and localization of partial discharge sources using RF sensors

This paper presents an automated system for condition monitoring of power distribution utilities through detection and localization of partial discharge (PD) sources. Among the various outcomes of PDs, emission of Electromagnetic Magnetic wave mainly in the radio frequency (RF) and ultra-high-frequency (UHF) band, is an eminent phenomena. In the present work, a Gaussian Mixer Model (GMM) has been applied in an unsupervised manner, primarily for detecting presence of PD sources inside power system utilities, and thereafter for classification of PD events. Finally, time-delay-of-arrival (TDOA) based algorithm has been applied for localization of the PD sources. The proposed system has been tested in laboratory using different types of artificial PD sources and in different power distribution sub-stations. The obtained results were found to be reasonable encouraging.