Machine-Learning Supported Localization of Partial Discharges in Real-Size Transformer Winding Using Noninvasive Capacitively- Coupled Pulse Injection

Abstract

Partial discharges (PDs) in transformer windings can lead to failures, causing power grid instability and outages. Therefore, early detection and accurate localization of PD incidents are crucial. This article introduces an electrical approach for PD localization in transformer windings. The approach involves the noninvasive injection of pulses (resembling PD) into the winding and recording the propagated signals at the winding terminals (i.e., line bushing and neutral ends). Noninvasive PD injection is performed using a capacitive coupling due to the absence of electrical access to transformer winding conductors. Next, features are extracted from the recorded signals to correlate them with the location of the injection point. This study investigates the effect of the PD pulse injection point on its response waveform recorded at the terminals to identify features corresponding to the PD source using a single measurement. Certain features of captured signals are highly correlated with PD location. The proposed approach establishes an accurate disk-to-disk localization model, particularly for large transformers. It employs a supervised classifier for PD localization. The performance of the proposed approach is verified using experimental measurements.