Partial Discharge Characteristics Under Repetitive Pulsed Voltage Excitation Using BiLSTM-Based Networks

Abstract

This article presents a novel approach for detecting partial discharges (PDs) in motor winding insulation systems subjected to repetitive pulsed voltage excitation, a scenario that is fundamentally different from conventional sinusoidal excitation and has been rarely addressed in existing research. Traditional PD detection methods, effective under 50/60 Hz sine wave excitation, face significant challenges in this context due to strong interference from the switching transients of power electronic devices. Most existing methods rely on antennas with ultrahigh frequency high-pass filters (typically around 500 MHz) to suppress switching noise, but this often leads to the loss of lower frequency PD information and results in lower detection sensitivity. Moreover, antenna-based detection requires removing the equipment enclosure to avoid electromagnetic shielding, making it suitable only for laboratory studies and impractical for industrial monitoring. In contrast, the proposed method detects PD signals through the electrical measurement circuit and applies a bidirectional long short-term memory network to filter out switching disturbances, enabling PD detection in the 100–500 MHz range without invasive procedures or the need for ultrahigh frequency filters. Experimental analysis reveals that PD events predominantly occur at the switching edges and are highly sensitive to excitation waveform transients. Shorter rise times lead to fewer but higher magnitude PD events, while longer rise times result in more frequent and lower magnitude discharges. The duty cycle also affects both PD distribution and discharge power. Space charge theory is used to explain the experimental results. The proposed approach can be further refined for future online or offline PD monitoring.