A significant challenge in the implementation of health monitoring systems for estimating the health state of devices is the lack of accurate information about design details. This challenge is particularly prominent in the field of power electronics, where both IC designers and converter designers are often hesitant to share information about their designs. Addressing this issue, this paper introduces a novel AI-driven digital twin modeling methodology that enables the detection and classification of failures in power semiconductors, particularly Wide Band Gap semiconductors. By employing AI-based system identification techniques, this method offers a noninvasive approach to health monitoring of power switches with high resolution, even while operating under real conditions. The proposed method has been validated by simulating wire bond failure in a SiC power MOSFET using MATLAB SIMULINK, and the results demonstrate its accuracy.

The introduction of silicon carbide(SiC) has reduced the superiority of traditional silicon-based power module pack-aging strategies. As packaging strategies become increasingly complex, classical thermal modelling tools often prove inadequate in balancing efficiency with accuracy. Integrating these tools with machine learning (ML) can significantly enhance their application potential. This discussion commences by addressing the pressing issues in thermal modelling of SiC modules, specifically the challenges associated with multiple heat sources and heat spreading. During the design stage, ML models can swiftly simulate the thermal response of various packaging strategies, aiding engineers in eliminating ineffective options. In the monitoring phase, the employment of a digital twin enables a deeper investigation into degradation phenomena. This article reviews the current status and explores the potential applications of ML in thermal modelling of SiC power modules.