Topology Optimization of a Complete Reluctance Machine with no Initial Information on its Geometry

This paper addresses the topology optimization of an entire multiphase reluctance machine. Multiple materials are automatically and simultaneously distributed in the rotor and stator to maximize torque, minimizing the designer bias. As an extension of density-based techniques, this methodology relies on the multidimensional interpolation of the physical properties of the candidate materials. To handle a large number of design variables, projected gradient descent is used, and the sensitivities are computed efficiently using the adjoint variable method. A wide variety of optimized structures is obtained and discussed.