深層学習モデルによる混相流れ場特徴量抽出手法の開発

抄録

Gas-liquid two-phase flow is observed in various engineering disciplines including heatexchanging devices, aeration system, chemical reactors and so on. Knowledge of the geometrical and topological features of two-phase flow structure is often needed to develop two-phase flow models. In the present study, convolutional neural networks (CNN) based object detection technique was utilized to develop robust and instantaneous bubbly flow feature extraction model. In order to overcome the issue of the lack of a large high-quality labeled dataset for the network training, bubble generative adversarial networks was utilized in the present study. Our results show the quite promising potential for the fully automated and versatile bubbly flow feature extraction.