Addressing the thermal management challenges in single-rotor dual-stator axial flux permanent magnet motors (AFPMM) used for humanoid robot hip joint drives

this paper proposes an improved three-dimensional lumped parameter thermal network (3D-LPTN) modeling method. This method enables accurate prediction of temperature rise across all components

validated through experiments. By calculating thermal resistances between adjacent nodes

partitioning heat sources

and employing parametric modeling

the theory for instantaneous temperatures at multi-order nodes is derived

establishing quantitative criteria for thermal capacitance

thermal resistance

and power loss. To address the significant axial thermal resistance gradients and complex transient operating conditions characteristic of AFPMMs

the model topology is optimized to reduce the node count of the 3D model. Crucially

axial thermal resistances are introduced within the composite stator-rotor structural region

rectifying the absence of axial heat transfer paths inherent in traditional 2D models. Experimental validation demonstrates: under intermittent loading conditions

the improved model achieves an average prediction error below 3℃ (a 23.1% improvement over the 2D model) with a maximum error not exceeding 6℃; during transient tests with periodic loading at low speed and small torque

the maximum tracking error is reduced to within 2℃. Six sets of variable-load tests confirmed the model's prediction accuracy and computational efficiency (single solution time: 3.2 s). This model provides a novel approach for the thermal reliability design of highly integrated AFPMM

offering significant practical value for engineering applications.