Sensor Fusion based Health Management Systems for Semiconductor Power Modules in Electric Vehicle Inverters

Power semiconductor devices in electric vehicle inverters face critical reliability challenges, with temperature-related aging issues accounting for 55% of converter failures. Accurate junction temperature monitoring is essential for managing aging mechanisms and preventing catastrophic failures. Insofar, existing single Temperature Sensitive Electrical Parameter (TSEP) approaches suffer from the limitation due to high measurement noise susceptibility, sensitivity to operating conditions, and inability to decouple temperature effects from device degradation. With such situation in mind, this thesis develops a novel multi-TSEP fusion framework for SiC power modules for the purpose of systematically analysing how working current, voltage, gate parameters, and wear-out conditions have effects on TSEP characteristics.Two methods based on the Principal Component Analysis-Multiple Linear Regression (PCA-MLR) are proposed to address aforementioned limitations. The first method tackles the noise susceptibility problem by extracting and fusing 16 statistical features from single gate current waveforms, using PCA to eliminate multicollinearity and enhance robustness against measurement disturbances. The second method addresses both operating condition sensitivity and temperature-degradation coupling by integrating multiple TSEPs across different measurement channels, leveraging complementary temperature-dependent characteristics from various electrical parameters such as on-state voltage drop, gate threshold voltage, and switching transients to achieve condition-independent estimation.Extensive validation demonstrates that both methods maintain temperature estimation errors within 4-5°C even under severe noise conditions with input errors up to 40-50°C—an order lower of magnitude improvement than conventional single-TSEP approaches. The multi-channel fusion approach particularly excels in decoupling temperature effects from device aging, enabling simultaneous health monitoring and thermal management. These results establish a practical pathway for robust temperature monitoring in the power device converter system for next generation, particularly for electric vehicle applications where reliability is paramount.