Minimizing Sensor Usage in TSEPs-based Junction Temperature Estimation of SiC Power Transistor through Time-Series Analysis

This paper presents a machine learning-based sensor fusion approach for non-invasive junction temperature estimation in SiC power modules, integrating multiple temperature-sensitive electrical parameters (TSEPs). By incorporating both current and previous TSEP values through Long Short-Term Memory (LSTM) networks, the junction temperature can be accurately predicted using a minimal number of sensors. Optimal TSEP combinations are explored to enhance accuracy and efficiency of the temperature estimation model and to consider the device degradation effect. Experimental results are presented on the dataset, acquired from two commercial SiC power modules under different operating conditions, with reference temperature measured by an optical sensor for verifying the proposed approach. Results demonstrate the method’s effectiveness, offering a sensor-efficient solution for an accurate and degradation-aware temperature estimation in SiC MOSFET devices, achieving a mean estimation error of less than 1 K and a maximum error below 8 K.