HVDC Defect Identification: Partial Discharges Signal Acquisition and Machine Learning Classification

The aim of this work is to achieve effective features to enable machine learning algorithms to identify defects and incipient faults in the case of insulating systems used for high-voltage direct current (HVDC) applications. Partial discharge (PD) testing and analysis has become a key tool in all aspects of the design, commissioning, maintenance and monitoring of AC and DC insulating systems. In this study, PD measurements were conducted on a medium voltage (MV) cable segment with artificial defects. These measurements were performed under a negative HVDC power supply using various sensors, including a High Frequency Current Transformer (HFCT), a Rogowski Coil, and a VHF Antenna. The acquired signals were post-processed using the Normalized Auto-correlation Function (NACF). The 3D NACF space plots have been obtained by selecting the three larger dispersion components obtained by the comparison of the NACF of the acquired signals. The objective was to analyze the differences in the measured PD signal data obtained with the three different sensors using machine learning algorithms. This analysis provides insights into the effectiveness of the developed features and their potential when different kind of sensors are used for the detection of faults in case of HVDC applications.