optimal shaping of non-conventional permanent magnet geometries for synchronous motors via surrogate modeling and multi-objective optimization approach

A methodology is proposed for optimal shaping of permanent magnets with nonconventional and complex geometries, used in synchronous motors. the algorithm includes artificial neural networkbased surrogate model and multiobjective search based optimization method that will lead to pareto front solutions. an interior permanent magnet topology with crescentshaped magnets is also introduced as the case study, on which the proposed optimal shaping methodology is applied. produced torque per magnets mass and percentage torque ripple are considered as the objectives, in order to take both performance and cost into account. multilayer perceptron architecture used to create the approximated model is trained to fit the samples collected via timestepping finite element simulations. the methodology can be easily generalized to offer a fast and accurate method to optimally define arbitrary permanent magnet shape parameters in various synchronous motors.