Sequential approximate multiobjective optimisation of switched reluctance motor design using surrogate models and nongradient local search algorithm

The design optimisation of switched reluctance motors poses many challenges to the designer. Different and often conflicting design objectives are normally considered in the design process. High-fidelity analysis programs offer the best-design analysis tools. However, direct coupling of these tools with optimisation algorithms is difficult, if not impractical. To allow the application of these tools, surrogate models are used which can take the form of fitted surface models or low-fidelity models obtained directly from the analysis procedures. Fitted surface surrogates are used and a trust-region approach to manage the objective and constraint functions during the optimisation process. This implies decomposition of the optimisation process into optimisation subproblems specific to each trust region. In this method the surrogates are applied in a sequence of optimisation steps, where the original objective and constraint functions are used to update the surrogates during optimisation.