深層学習を用いた電動機の性能予測

抄録

When designing electric motors, many types of performances (electrical and mechanical characteristics) must be predicted with good accuracy. In general, these performances are determined based on complex theoretical calculations, but theoretical calculations include various assumptions. Therefore, it is difficult to eliminate prediction errors when predicting performance, and it is necessary to improve accuracy by referring actual test data. Recently, with the digitalization of the manufacturing process, a large amount of actual data has been converted into a database, and it is expected to be put to effective use. Here, a neural network that predicts various performances of electric motors using a large amount of actual data as a training dataset, is constructed to achieve uniform and high-precision performance prediction via deep learning. Its practical use for actual design work is verified in this study.