An Accurate and Efficient Two-Level Optimization for YASA Machine From the Perspective of Analytical and Surrogate Models

Abstract

In quasi-direct-driving joints, a significant challenge lies in achieving high-torque-density machines within the compact dimensions. Therefore, this paper proposes an efficient two-level optimization strategy for the yokeless and segmented armature (YASA) machines, from the perspective of analytical and surrogate models. This strategy guarantees the modeling accuracy while significantly enhancing the optimization efficiency. In the initial design stage, the subdomain method considering the bilateral air-gap and yokeless stator structure is developed, which provides the guideline for selecting key design variables. Following the sensitivity analysis, the multi-objective optimization is divided into low-sensitive and high-sensitive levels. The two-level surrogate models are established based on the response surface and back-propagation neural network methods, respectively, which exhibit high accuracy in predicting design objectives. The effectiveness of the proposed strategy is verified by comparing the electromagnetic performance of the initial and optimized YASA machines. Finally, the YASA machine is prototyped, and experimental tests are performed to further validate the finite element analysis results.