Nihon Kikai Gakkai ronbunshu (Jan 2022)
Unsupervised learning for classification of flow regime in high-pressure gas-liquid two-phase flow including transition region
Abstract
The MH21-S R&D consortium (MH21-S), supported by the Ministry of Economy, Trade, and Industry (METI), is currently developing the commercial production process of methane gas from methane hydrate. In order to develop the gas production method through thermal-hydraulic models, an objective identification method of gas-liquid two-phase flow regime under high-pressure conditions is necessary. Furthermore, in identifying the flow regime, it is necessary to distinguish the transition states and improve the calculation accuracy. Therefore, flow regime identification has been conducted in the present research by classifying the high-speed images using clustering algorithms, namely the principal component analysis (PCA) and the k-means method. Specifically, the sequence images of the upward gas-liquid two-phase flow under high pressure taken with a high-speed camera are merged into a single image by the time-strip method, and these single images were then processed with PCA and classified by the k-means method. Furthermore, the PCA and the Gaussian Mixture Model (GMM) were also applied to quantify the flow regime of the transition region. As a result, PCA has shown that the merged images of bubbly flow and slug flow occupy different regions, and the bubbly flow and slug flow are classified with high recall values. The flow regime map obtained from the classification by the PCA and GMM mostly showed similar trend compared with the transition models studied in the past. Thus, this study has shown that the flow regime identification can be performed using the single images of upward gas-liquid two-phase flow merged by the time-strip method, with clustering algorithms. Moreover, it was shown that the unsupervised machine learning method is capable of clustering the flow regime at transition regions.
Keywords
