A Novel Artificial Intelligence-Enabled Junction Temperature Estimation Method for Multiple Chips in a SiC Power Module Based on an Inherent Built-in NTC Sensor

Junction temperature is a crucial parameter for monitoring SiC power modules. However, the existing non- invasive junction temperature estimation methods still exhibit certain deficiencies in accuracy and resolution limited by the handling of multi-chip thermal cross-coupling and actual operating conditions. Therefore, this paper proposes a novel artificial intelligence-enabled method to realize junction temperature estimation for multi-chip SiC power modules. The proposed approach is fully non-invasive and leverages the temperature profile of the built-in negative temperature coefficient (NTC) thermistor to decouple the effects of complex power losses and cooling conditions. A hybrid neural network model (HNNM) integrating both data-driven and physical information is developed to reconstruct the thermal field of the module. The mapping relationship between the NTC temperature profile and the module thermal field is analyzed, enabling the decoupling of operating condition parameters over a specified range. The HNNM with good generalization capability and noise robustness is constructed and tuned. The resolution and accuracy of the proposed estimation method are verified by finite element simulation and experiment. The contribution of this paper is to provide an accurate, non-invasive, easy-to-implement and generalized junction temperature monitoring scheme for multi-chip SiC power modules under complex operating conditions.