This study investigates the impact of rotor structure, material selection, and cooling methods on ultra-high-speed motor performance, revealing performance variation laws under multi-physical field coupling. Considering mechanical boundary constraints, we propose an optimization design method based on a multi-physical field coupling model. Using a MaxPro experimental design, initial samples are obtained and fitted using a Kriging surrogate model. The NSGA-2 algorithm is then applied for optimization, with Relative Maximum Absolute Error (RMAE) and Relative Average Absolute Error (RAAE) employed for accuracy evaluation. The Kriging model is iteratively updated based on evaluation results until the optimal design is achieved. This method enhances motor performance, ensures mechanical boundary conditions, and reduces computational load. Experimental results show significant improvements in efficiency and power density. This study provides theoretical support and technical guidance for ultra-high-speed motor design and offers new ideas for related motor research and development. Future work will explore more efficient and intelligent optimization algorithms to continuously advance ultra-high-speed motor technology.
LiDAR-based place recognition is critical for long-term autonomous navigation in Global Navigation Satellite System (GNSS)-denied environments, yet existing methods struggle to balance accuracy and efficiency under substantial yaw rotations. This paper proposes a robust framework based on a multi-channel polar bird’s-eye-view (BEV) representation. Under yaw-dominated revisits, the polar BEV image transforms yaw rotation into cyclic column shifts, providing a useful structural prior for rotation-equivariant feature extraction. Raw point clouds are projected onto polar BEV grids encoding density, height, and intensity. A rotation-equivariant feature extractor comprising a Radial Compression Module and a rotation-equivariant Transformer module captures long-range azimuthal dependencies via Conditional Positional Encoding and Circular Relative-Position Bias. The equivariant features are aggregated by NetVLAD into a compact global descriptor, trained end-to-end with a hard-example mining triplet loss. Extensive experiments on the public KITTI and NCLT datasets, as well as our self-constructed LiDAR Place Recognition Revisit (LPRR) dataset, demonstrate competitive performance on KITTI and superior performance on NCLT and LPRR among the compared methods. The proposed framework achieves a favorable trade-off between performance and computational cost, and shows promising cross-dataset generalization on the evaluated NCLT and LPRR datasets without fine-tuning.