Dynamic electromagnetic-thermal mapping in motor digital twins: A real-time multiphysics framework

Digital twin technology offers a promising approach to real-time monitoring and predictive analysis of electric motors, but achieving high-fidelity multiphysics simulation under varying operational conditions remains challenging. This paper proposes a real-time electromagnetic-thermal coupling framework for motor digital twins, integrating Proper Orthogonal Decomposition (POD)-based model order reduction and neural network (NN)-assisted surrogate modeling. First, a reduced-order model (ROM) is constructed by applying POD to finite element-derived electromagnetic and thermal field matrices, followed by NN training to map operational points to the reduced modal coefficients, enabling rapid field reconstruction across wide operating ranges. Second, an interactive electromagnetic-thermal digital twin framework is developed, where experimentally measured data are dynamically fed back to the ROM-NN model, allowing online calibration and adaptive updates. Experimental validation demonstrates that the proposed framework achieves real-time solving of coupled electromagnetic-thermal fields while maintaining high accuracy. The system also enables concurrent prediction of torque, losses, and temperature distributions, proving its potential for motor health monitoring and optimization.