Advanced Design Optimization Technique for Torque Profile Improvement in Six-Phase PMSM Using Supervised Machine Learning for Direct-Drive EV

Few of the challenges with development of a single onboard motor for directdrive electric vehicles include high torque density and low torque ripple. Therefore, in this paper, a 36slot, 34pole consequent pole sixphase permanent magnet synchronous machine (PMSM) has been optimized to address the aforementioned challenges for directdrive application. Existing literature on optimization processes that rely solely on finite element models are restricted to threephase machines only and also take longer computation time. Therefore, this paper proposes a novel optimization approach based on supervised machine learning for sixphase PMSM. In this approach, a nonconventional extended dual dqframe model that accounts for higher order space harmonics in inductances and flux linkages has been developed and used for accurate computation of average torque and torque ripple of sixphase PMSM. Using the performance characteristics obtained from the extended dual dqframe model for a set of initial design candidates, support vector regression algorithm is employed for supervised machine learning and increasing solutions in the design space. Furthermore, pareto front is used for selecting optimal machine models with maximum torque density and reduced torque ripple. Multiobjective tradeoffs and comparison of initial and optimized designs based on average torque, torque ripple, efficiency and cost are performed.