Meta-Models for Torque Optimization of Spoke Type Permanent Magnet Synchronous Machines

The design of internal permanent magnet synchronous machines (IPMSM) is a complex task that often includes multiple objectives and constraints. Lately, a lot of research has been focused on the reduction or elimination of rare-earth elements (REE). To achieve that, the use of ferrite permanent magnets (PM) and the enhancement of reluctance torque are the most common solutions. In this paper, three different meta-models are developed and used in the optimization process to maximize the torque of a Spoke Type PMSM (Spoke) with ferrite PM taking into consideration both reluctance and magnetic torque. These meta-models are based on 1-dimensional convolutional neural networks (1DCNN), gaussian process regression (GPR) and polynomial chaos expansion (PCE). This study shows that the GPR-based meta-models generally present the lowest absolute error. Although, regardless of the different performance, all three meta-models achieve similar optimized solutions.