Practical application-focused online V-I curve monitoring system for power semiconductor devices

• We developed a practical, non-intrusive, online V-I curve monitoring system for power semiconductor devices. • Compact PCB Rogowski coil with a two-stage integrator accurately measured conduction current. • V CE(SAT) -I C of IGBT and V F -I F of power diode are acquired in real-time during inverter operation. • Bond wire lift-off and solder layer degradation are detected based on statistical methods. • Compared residual-based fit model and multivariate T² outlier methods for monitoring degradations. As reliability is growing critical in power electronics systems, preventive maintenance for power semiconductor devices is crucial for extending system lifespan and improving robustness. However, conventional time-based maintenance often increases operational costs and unexpected downtime due to significant variation and uncertainty in the power semiconductor device's lifetime. In contrast, condition-based maintenance offers an optimized supervision scheme based on real-time health monitoring, reducing unnecessary costs and increasing overall system reliability. This paper presents a practical, non-intrusive, online condition monitoring system for detecting degradations in power semiconductor devices. The developed V-I curve monitoring system accurately measures V CE(SAT) - I C curves for IGBT and V F - I F curves for power diode during the normal operation of a full-bridge inverter. A novel conduction current ( I C ) sensing method based on a low-cost, tiny PCB sensor and a two-stage integrator was proposed. On-state voltage ( V ON ) is measured by a diode-based circuit, and case temperature ( T C ) is sensed by a thermocouple. Real-time V ON - I C - T C samples are captured simultaneously with 16-bit ADCs of the microcontroller and stored in the SD card. Acquired V-I curve data is analyzed using two statistical approaches to detect bond wire and solder layer degradations. Both the residual-based fit model and the multivariate T 2 outlier methods successfully detected abnormal conditions, and comparisons are made for these two approaches.