Publication

Abstract : In this present work, an innovative framework has been proposed to localize single and multiple partial discharge (PD) events using acoustic sensor. For this purpose, an experimental setup has been configured in laboratory where 47 single and multi-location PD events have been simulated within an oil-filled emulated cubical tank (resembling a HV equipment). For each PD event, PD signature has been recorded through 5 acoustic sensors. Thereafter, recorded PD signals are subjected to smoothed pseudo-Wigner Ville distribution in order to obtain time-frequency representation. Thereafter, image plot of time-frequency representation correspond to each sensor has been separately fed to pre-trained deep architecture for automated feature extraction. The extracted features are then subsequently fed to a bi-LSTM classifier for PD event classification, which returned high recognition accuracies in classifying different single and multi-location PD event. The proposed framework is also effective under strong noisy environment, which signifies it’s worthiness for real life PD localization within a HV equipment.