Comparison of Efficient Global Optimization and Output Space Mapping on the Biobjective Optimization of a Safety Isolating Transformer

The optimal design of electromagnetic devices needs to address two particular aspects: 1) the use of accurate tools and 2) the limited time budget affected to this task. In order to respond to these issues, two low evaluation budget multiobjective optimization techniques-efficient global optimization (EGO) and output space mapping (OSM)-are investigated in this paper with regard to a biobjective optimization benchmark. The device to be optimally sized is a low-voltage safety isolating transformer, represented through two levels of modeling: 1) coarse (analytical model) and 2) fine (numerical 3-D FEM). The biobjective character of the problem is accounted for by OSM through an epsilon-constraint technique. A transformation technique is set up with EGO by considering the constraints as an additional objective function in order to handle the constraints of the initial optimization problem. The biobjective comparison of the two techniques on the transformer benchmark is discussed.