Advanced Spatial Temperature Monitoring of Power Modules via Fourier Neural Operator-Based Thermal Model

Accurate monitoring of power module spatial temperature (PMST) remains a critical challenge in power electronics. This letter proposes a Fourier Neural Operator based Thermal Model (FNO-TM) to enable efficient and precise prediction of PMST for the first time. Built on a data-driven paradigm, FNO-TM extracts spatial-frequency features, facilitating faster convergence and reduced training costs compared to conventional neural networks. To further enhance generalization, a Trunk-and-Branch architecture is employed to adapt to varying boundary conditions. Additionally, an electrothermal coupling strategy tailored for FNO-TM is developed and validated on a practical inverter platform. Experimental results show that FNO-TM maintains high accuracy across different current levels and cooling conditions, with maximum errors below 1.8°C. Furthermore, the PMST prediction is completed within 0.1 s, significantly outperforming traditional methods in computational efficiency. The prediction accuracy can be further enhanced through fine-tuning with measured data. These features make FNO-TM a promising solution for thermal monitoring and digital twin applications in power electronics.