A fast control-integrated and multiphysics-based multi-objective design optimization of switched reluctance machines

This paper proposes a comprehensive approach for the multi-objective design optimization of switched reluctance machines (SRMs). An analytical model that predicts the electromagnetic (EM) behaviors of SRMs with arbitrary geometries, materials or current profiles is first developed and validated by its finite-element analysis (FEA) counterpart in terms of various performance indices. Then, a hybrid thermal model combining a 2-dimensional (2D) finite-difference (FD) formulation and thermal equivalent circuits is used to estimate the temperature distribution within an SRM based on the loss distribution calculated by the analytical EM model. The multi-physics model can evaluate the performances of an SRM much faster than the prevalent finite-element analysis (FEA). In addition, to further reduce the computational cost, artificial neural networks (ANNs) are constructed based on the maximin Latin hypercube design (MLHD) to identify the relationship between the SRM performance indices and the selected design variables within the entire predefined design space. Then, incorporating the pre-trained ANN models, a particle swarm optimization (PSO) based multi-objective optimization approach is utilized to identify the Pareto optimal design candidates in terms of maximizing the efficiency and torque density, and minimizing the torque ripple. The thermal constraints as well as the effects of current profiles are incorporated in the design process. Simulation results demonstrate that the proposed method can effectively generate the Pareto front within about half an hour with the evaluation of over 60,000 design candidates.