Comparison between single-model and multi-model optimization methods for multiphysical design

The purpose of this work is to compare two optimization approaches, used for the research of optimal configurations of electrical machines. The first one uses a single multiphysical model, called “reference model”, directly used by a classical optimization routine. The second approach proposes the use of a second surrogate model in association with the previous reference model. This second configuration corresponds to the Space Mapping (SM) methodology. In both cases, models involved are multiphysical descriptions of the main phenomena occurring in electrical machines. They consider electrical, magnetic, thermal and mechanical aspects. They use different modeling techniques (analytical relations, lumped parameter networks, Finite Element analyses), depending on the physics to be described. Appropriate coupling strategies between these different sub models are chosen to get the best results in the framework of SM methods. This work evaluates both approaches by a comparison of the corresponding computation costs; in this context, the number of evaluations of the fine (reference) model, during each optimization process is chosen as the comparison criterion. The paper shows that SM-like methods allow faster optimization processes and require a smaller number of evaluations (of the reference model) and lead to similar final machine designs.