An improved Kriging surrogate model method with high robustness for electrical machine optimization

The uncertainties of electrical machines manufacturing decrease the prediction precision of traditional multi-objective optimization methods based on Kriging surrogate model. Existing robust optimization method requires a large amount of calculation time. In order to improve the accurateness and release the computational burden of the Kriging surrogate model method in the robust optimization, two genetic algorithm (GA)-based optimization methods with different sample principles are proposed and compared. The one is adding the final optimization result of GA as the samples into the surrogate model, while the other one is adding the samples from the GA process for the target surrogate model. Taking a 12-slot 14-pole interior permanent magnet (IPM) machine as a case study, the simulation results show that the latter one is more accurate than the former. Furthermore, the comparison between the deterministic optimization and robust optimization in the case study demonstrates the superior of the second GA method.