Abstract Printed circuit board (PCB) based power semiconductors are essential elements in modern electrical systems by their fast and reliable capabilities within compact miniaturized sizes. However, due to the recent trends of increased operating voltage and frequency ranges, the dielectric strength of insulation systems in PCBs has been faced a lot of challenges regarding degradation and acceleration of insulation properties, initiating partial discharges (PD). This paper proposes a PCB based microstrip patch-type PD sensor to detect high frequency PD signals generated from insulation defects. Four typical types of PCB defect models were fabricated to simulate the insulation defects within the power semiconductor. Phase-resolved partial discharge (PRPD) patterns were obtained at 120% of each partial discharge inception voltage (PDIV) level, and various statistical PD features were extracted to establish PD datasets. Four representative machine learning (ML) algorithms were comparatively analyzed, and the random forest (RF) model achieved the highest performance with an accuracy of 96.9%.
Lee et al. (2026) studied this question.