An Improved Kriging Surrogate Model Method With High Robustness for Electrical Machine Optimization
Hengliang Zhang, Guangchen Wang, Junli Zhang, Yuan Gao, Wei Hua, Yuchen Wang
Abstract
This article presents a highly robust optimization method for electrical machines, taking the uncertain tolerances of machine manufacturing into account. Different from the traditional multi-objective optimization methods based on Kriging surrogate model, two genetic algorithm (GA) models with disparate sampling principles are used here to release heavy computational burden and to improve prediction accuracy. One is adding the final optimization result of GA as the samples into the initial surrogate model, while the other one is adding the samples from the optimization process for the initial surrogate model. A 12-slot 14-pole interior permanent magnet synchronous machine (IPMSM) is used for the case study, and two GA models are compared. Furthermore, the proposed robust optimization method is compared with a deterministic optimization method to demonstrate its superiority, and its effectiveness is verified by prototype tests.