Flow regimes identification of air water counter current flow in vertical annulus using differential pressure signals and machine learning

Abstract

AbstractThis study investigates the gas–liquid two-phase counter-current flow through a vertical annulus, a phenomenon prevalent across numerous industrial fields. The presence of an inner pipe and varying degrees of eccentricity between the inner and outer pipes often blur the clear demarcation of flow regime boundaries. To address this, we designed a vertical annulus with adjustable eccentricity (outer and inner diameters of 125 mm and 75 mm, respectively). We conducted gas–liquid counter-current flow experiments under specific conditions: gas superficial velocity ranging from 0.06 to 5.04 m/s, liquid superficial velocity from 0.01 to 0.25 m/s, and five levels of eccentricity (e = 0, 0.25, 0.5, 0.75, 1). We collected differential pressure data at two distinct height distances (DP1: 50 mm and DP2: 1000 mm). We used vectors, composed of both the probability density functions (PDFs) of the differential pressure signals and the power spectral density (PSD) reduced via Principal Component Analysis, as features. Using the CFDP clustering algorithm—based on local density—we clustered the flow regimes of the experimental data, thereby achieving an objective and consistent identification of the flow regime of gas–liquid two-phase counter-current flow in a vertical annulus. Our analysis reveals that for DP1, the main differences in the PSD of various flow regimes occur within the 0.5–1 Hz range. Among the three flow regimes involved, the slug flow exhibits the highest power intensity, followed by the bubbly flow, with the churn flow having the least. In terms of differential pressure distribution, the bubbly and churn flows have a concentrated distribution, while the slug flow is more dispersed. For DP2, the PSD differences primarily exist within the 0.5–2 Hz range. The churn flow has the highest power intensity, followed by the slug flow, with the bubbly flow being the weakest. Here, the bubbly flow's differential pressure distribution is concentrated, while the slug and churn flows are more dispersed. Based on the results of the flow regime classification, we generated a flow regime map and analyzed the influence of annulus eccentricity on the flow regime. We found that in most cases, pipe eccentricity does not significantly affect the flow regime. However, in the transition region—such as the bubbly to slug flow transition zone—flows with medium eccentricity values (e = 0.5, 0.75) are more likely to transition to slug flow. We compared the visual recognition results of flow regimes with the clustering results. 4.04% of the total samples showed different results from visual recognition and clustering, primarily located in the flow regime transition area. Since visually distinguishing flow regimes in these areas is typically challenging, our methodology offers an objective classification approach for gas–liquid two-phase counter-current flow in a vertical annulus.