Flow regime identification of air-water two-phase counter-current flow in vertical annulus and eccentricity effect analysis：a machine learning approach

Gas-liquid two-phase counter-current flow has been involved in many industrial fields, due to the existence of the inner pipe and the variation of the eccentricity of the inner and outer pipes, the flow regimes are difficult to be visual identification. To solve this problem, a vertical annulus with adjustable eccentricity was designed. Experiments were carried out under different combination of gas-liquid flowrate and eccentricity. The feature vector was formed by PDF and principal components of PSD of two differential pressure signals. The clustering algorithm CFDP based on local density was used to cluster the experimental data, results show that this method can realize the identification of flow regimes without prior information, and the flow regime transition boundary agrees with the experimental results. The influence of pipe eccentricity on the flow regime transition is analyzed. The results show that, in most cases, pipe eccentricity has no obvious influence on the flow regimes. However, in the flow regime transition region, such as the bubbly flow to slug flow transition region, when the eccentricity is at a medium value, that is, e = 0.25,0.5,0.75, the flow is more likely to transit to slug flow, however, when e = 0,1. The flow tends to remain bubbly flow.