Deep Learning–Enabled Multisource Partial Discharge Diagnostics for Online Monitoring of High Voltage Rotating Machine Insulation

In this paper, a deep learning–based approach is proposed to identify single and multiple source partial discharge (PD) activities in high-voltage rotating machine insulation employing high-frequency current transformer (HFCT) measurements. To this end, HFCT-acquired PD signals are converted to phase-resolved partial discharge (PRPD) patterns. A comprehensive PRPD database is developed through controlled laboratory experiments involving different insulation materials and discharge mechanisms. Thereafter, a lightweight custom ConvNeXt model is designed for automated feature extraction and classification of the PRPD pattern. The proposed custom ConvNeXt model achieves an overall classification accuracy of 98.3%. Comparative studies with other deep learning models further validate the effectiveness and computational efficiency of the proposed approach. Hence, the proposed method is potentially worthy for online condition monitoring of high-voltage rotating machines.