Deep-neural-network-based Electromagnetic Analysis and Optimal Design of Fractional-slot Brushless DC Motor for High Torque Robot Joints

Fractional-slot brushless DC motors (FS-BLDCMs) have the advantages of high torque density and low cogging torque for robot joints. However, finite element analysis (FEA) of the FS-BLDCMs causes time consumption, which obstructs the progress on finding optimal electromagnetic characteristics of the FS-BLDCMs. This paper presents a novel design method to improve the FS-BLDCM motor characteristics and improve the computation efficiency by the deep neural network (DNN). The FS-BLDCM motor performance is optimized by the genetic algorithm and validated by finite element analysis. The computation time between the finite element analysis (FEA) and the deep neural network (DNN) is compared. The result shows the efficiency of the deep neural network.