Surrogate Modeling of Dynamic Thermal Field Performance via Recursive Multi-Resolution Reduced Order Model

Electric propulsion motors frequently operate under multi-condition and transient scenarios, where the coexistence of slow thermal inertia and fast local excitations creates multi-scale thermal behaviors. To address the inability of conventional methods to simultaneously achieve high spatial resolution and real-time performance, this paper proposes a multimodal dynamic reduced model (MDRM). Spatially, hierarchical proper orthogonal decomposition is employed to efficiently compress high-dimensional full-field data by sequentially reducing spatial and parametric dimensions. Temporally, a recursive multi-resolution dynamic mode decomposition strategy is proposed to capture the temporal evolution. It recursively decomposes system dynamics over hierarchical time windows, effectively decoupling slow global temperature rises from fast local transients. Furthermore, a parameter-adaptive reduced basis model is integrated to handle multiple operating conditions by establishing a nonlinear mapping between operating parameters and expansion coefficients. Experimental validation shows that the MDRM achieves an average relative error of 3.2% while requiring only 0.8 s per calculation, demonstrating significant potential for real-time digital twin applications. Additionally, the model is applied to the computationally intensive task of multi-condition electromagnetic-thermal coupled optimization, achieving efficient and rapid multi-query analysis.