深層学習による円管内乱流の脈動制御のための渦構造予測モデル

抄録

We performed direct numerical simulation of pulsating controlled flow for drag reduction and predicted time evolution of vortical structure distribution using deep learning. The datasets for the deep learning is the distribution of vortical structure defined by second invariant of velocity gradient tensor in circumferential cross sections. There are the correlation between the time variation of the friction drag and the ratio of the vortical structures occupying in the sections. We predicted the time variation of the ratio by predicting the time variation of the vortical structures using convolutional neural networks with convolutional long short-term memory (ConvLSTM). It was revealed that the model can predict the periodic variation of the ratio.