A method for Localization of Single source Partial Discharges using Acoustic Sensors and Random Forest Algorithm

Partial Discharge localization is of vital importance in the context of condition monitoring of power transform-ers and other High Voltage equipment. Acoustic sensors offer quite straightforward method for localizing PD sources in such insulation systems through measuring the time difference of arrival (TDOA) between signals captured at multiple acoustic sensors placed on the walls of the transformer tank. However, in such cases, simultaneous measurement across multiple sensors is required which increases the complexity as well as the cost of the overall PD detection system. In this work, an alternate scheme is proposed for localizing PD sources through application of Random Forest Algorithm, a well-known machine learning technique. An experimental set-up has been developed emulating a real-life transformer tank with PD sources present within it and three acoustic sensors were placed at three walls of the tank. The location of the PD was varied and corresponding to each location, the generated acoustic signals measured by each sensor at a time connected with an oscilloscope. Suitable features were extracted from this recorded data corresponding to each location. These features were used for training and testing of a random forest based machine learning model. From the testing results, it was observed that the developed model can locate single source PD with an accuracy of over 90 %.