A Computationally Efficient Algorithm for Rotor Design Optimization of Synchronous Reluctance Machines

A generalizable algorithm is proposed for the design optimization of synchronous reluctance machine rotors. Single-barrier models are considered to reduce the algorithm's computational complexity and provide a relative comparison for rotors with different slots-per-pole combinations. Two objective values per sampled design (average and ripple torques) are computed using 2-D finite-element analysis simulations. Non-linear regression or surrogate models are trained for the two objectives through a Bayesian regularization backpropagation neural network. A multi-objective genetic algorithm is used to find the validated Pareto front solutions. An analytical ellipse constraint is then suggested to encapsulate optimal solutions. Compared with a direct sampling approach, this restriction captures an optimal region within the double-barrier space for further torque ripple reduction.