データ拡張による構造に生じる最大応力の予測精度の向上

抄録

In recent years, the field of machine learning has been rapidly developing and deep learning has been applied to a lot of engineering problems. On the other hand, it is not possible to prepare a huge amount of data for sufficient learning. In the field of computer recognition, data augmentation is a common technique for a better result of machine learning. Training data such as stress values, there are a few data augmentation methods for physical phenomenon prediction, however the accuracy is low where frequency data is low. In this study, we’d like to improve the accuracy with limited number of data sets using a new data augmentation method and evaluated it with the aim of verifying whether the method is effective.