深層学習を用いた高圧気液二相流遷移領域における流動様式識別

抄録

Currently, the MH21-S R&D consortium (MH21-S), supported by the Ministry of Economy, Trade, and Industry (METI), is developing the commercial production process of methane gas from methane hydrate. In order to develop a gas production method, the objective flow regime identification technique of gas-liquid two-phase flow under high-pressure conditions would be useful. Therefore, in the present research, a new method for identifying the flow regime by a convolutional neural network (CNN) using high-speed images obtained at a narrow view was proposed, and its performance was evaluated. Specifically, the upward gas-liquid two-phase flow images were recorded using a high-speed camera, and they were merged into a single image containing information from entire frames. The CNN was created and trained with these newly merged images, and it was utilized to identify flow regimes of unlabeled test images. As a result, the proposed method using the CNN model was capable of identifying the flow regime with high accuracy. Thus, the present study has shown that the CNN model with the proposed image conversion method can properly identify the upward gas-liquid two-phase flow regime. In addition, by utilizing the developed network, the bubbly-slug transition region, which has only been studied through visual observation or void fraction signals in the past, was successfully evaluated quantitatively.