Surrogate‐Assisted Vibroacoustic Optimisation of High‐Power‐Density PMSMs

ABSTRACT

High‐power‐density PMSMs with a wide operating speed range exhibit complex electromagnetic excitation characteristics. Performing vibration and noise optimisation over the full operating speed range using multi‐physics finite element (FE) methods incurs prohibitive computational costs. To address the limitations of conventional surrogate models in multi‐condition optimisation across wide speed ranges, which include excessive data volume requirements and limited iteration capability, this paper proposes a rapid electromagnetic vibration and noise optimisation framework for high‐power‐density PMSMs over the full speed range under representative no‐load and peak‐load conditions. This framework reduces the data sampling burden by leveraging operating condition characteristics, constructs an electromagnetic force prediction model based on deep neural networks (DNNs), and incorporates a hierarchical multi‐objective optimisation method to achieve noise reduction across the full speed range under these conditions. Following the proposed workflow, a parametric optimisation study is conducted on a high‐power‐density PMSM prototype, and its rapid iterative optimisation capability is verified by comparison with the conventional multi‐physics coupled iterative optimisation scheme. Finally, the absolute noise performance of the prototype is validated through no‐load speed ramp tests, and the optimisation effect is confirmed via comparative multi‐physics simulations under peak load conditions.