Multi-scale heat transfer path optimization in magnetic-thermal coupling simulation of winding conductors using graph neural networks

This study proposes a GNN-based multi-scale heat transfer path optimization method for magnetic-thermal coupling simulation of wound conductors. Key parameters like magnetic vector potential, flux density, and temperature distribution are identified. An adaptive graph network integrates these parameters to build a 3D conductor model. A multi-scale spatio-temporal graph convolution module captures heat transfer path characteristics, while the GraphSAGE algorithm aggregates thermal resistance and electromagnetic loss data from adjacent nodes to train the GNN.The trained GNN outputs optimized multi-scale heat transfer path results, including temperature distribution and magnetic field loss. Experiments show the method effectively simulates magnetic-thermal coupling, with ohmic losses in low-voltage and high-voltage windings at ∼600 W and ∼300 W, respectively, and peak eddy current losses reaching ∼1600 W and ∼1700 W. Temperatures mainly range between 320–340 K (low-voltage) and 300–320 K (high-voltage). The method’s optimization reduces magnetic losses and material usage.