Electrical machine optimization using a kriging predictor
Jean-Laurent Duchaud, Sami Hlioui, François Louf, Mohamed Gabsi · 2014
This paper presents an optimization on a surface response created by kriging. It focuses on the predictor creation, its refinement and on the advantage of such a method over a direct optimization. In order to reduce the number of evaluations needed to construct the predictor, an adaptive method for the trial site selection is introduced as an improvement for the Latin Hypercubic Sample (LHS) algorithm. The method is applied to the optimization of a flux switching synchronous machine and the results are illustrated for two and four parameters.