Multi-objective surrogate optimization of mixed-variable IPM motor design problem with ADMM-based approach

作     者：Asanuma, Tatsuya Kanno, Yoshihiro 

作者机构：Univ Tokyo Grad Sch Informat Sci & Technol Dept Math Informat Hongo 7-3-1 Tokyo 1138656 Japan JMAG Business Co JSOL Corp Kudan Kaikan TerraceKudan Minami 1-6-5 Tokyo 1020074 Japan 

出 版 物：《OPTIMIZATION AND ENGINEERING》 (Optim. Eng.)

年 卷 期：2025年

页      面：1-26页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 08[工学] 0701[理学-数学] 

基　　金：Japan Society for the Promotion of Science JSOL Corporation 

主　　题：Electric motor Multi-objective optimization Mixed-integer nonlinear programming Sequential approximation optimization Surrogate-based optimization Radial basis function network 

摘      要：Designing electric motors boils down to solving multi-objective optimization problems where we look for design variables satisfying multiple requirements on, e.g., torque and electromagnetic losses. To design high-performance motors without restriction by existing design ideas, it is necessary to take the design space as wide as possible at the initial design stage. When dealing with the extensive design space, it is required to treat the basic structure parameters of motors and the detailed dimensions as design variables. For an interior permanent magnet (IPM) motor, which is one of the typical motor types, we need to handle discrete integer design variables such as the number of slots and the coil pitch in addition to the continuous ones. However, it is difficult to obtain the discrete design variable values without rounding off to the nearest allowed values in the framework of the existing continuous surrogate optimization such as sequential approximation optimization (SAO) with the response surface created from the electromagnetic finite element analysis results. We solve this mixed-variable optimization problem by the heuristic approach based on the alternating direction method of multipliers (ADMM) and show with example design problems that the better results are obtained more efficiently in terms of the number of expensive objective function evaluations compared to taking rounding-off measures.