Computationally Efficient Optimization Approach of Synchronous Reluctance Machines Using Variance-Based Sensitivity Analysis

Optimal design of synchronous reluctance machines usually involves a large number of parameters and several objectives or constraints, making the optimization problem hard to solve. This study proposes a method to deal with optimization problems prone to local minima by partitioning it into a sequence of smaller optimization problems. The partitioning is performed by interpreting the results of a global variance-based sensitivity analysis, conducted prior the optimizations. The interest of the approach is demonstrated on the rotor optimization of a 4 poles synchronous reluctance machine, to maximize the mean torque while both limiting the torque ripple and the local Von Mises stresses. The parametrization, finite-element models and results of the sensitivity analysis are presented, from which a coherent partitioning of the optimization problem is proposed. Results show that this approach is computationally time efficient, with a total optimization time that is a fraction of the one necessary to perform a single optimization on all the parameters. The torque ripple is substantially lowered below 5% of the mean torque. Independent runs of the design approach show a good reproducibility in the results.