深層学習を用いたトポロジー最適化の効率化

抄録

This paper presents a new topology optimization of rotating machines using deep learning. The classifiers of the relations between the motor shape and its torque characteristics are made by training of a convolutional neural network (CNN). The training data is obtained during a topology optimization of torque characteristics based on the finite element method (FEM). Then the minimization of the iron loss with maintaining the torque characteristics can be processed effectively using the classifiers.