Comprehensive Study of Gas-Liquid Flow Patterns in Concentric Annuli: Experimental and Deep Learning Perspectives

Summary Complex gas-liquid two-phase flow occurs in the wellbore after overflow during drilling operations. Accurate prediction of the wellbore flow pattern distribution and void fraction is crucial for real-time formation pressure inversion and efficient well killing. A large-scale vertical annular gas-liquid two-phase flow experimental system was used to measure the gas-liquid two-phase flow patterns distribution and void fraction at liquid viscosities ranging from 16 to 39 mPa·s. A research database was established using literature data on the flow patterns and void fraction of gas-liquid two-phase flow. A deep learning model based on long short-term memory (LSTM) networks, integrated with convolutional neural networks (CNN) and attention mechanisms, was developed to predict the flow patterns and void fraction of gas-liquid two-phase flow. The results show that increasing liquid viscosity shifts the transition boundary from bubbly flow to slug flow to lower superficial gas velocities, while the transition boundary from slug flow to churn flow shifts to higher superficial gas velocities. At a constant gas-liquid superficial velocity, the void fraction in the annulus increases as liquid viscosity rises. As the gas superficial velocity increases, the contribution of small bubbles to the section void fraction decreases in annular flow, reducing the influence of viscosity on void fraction. The deep learning prediction model demonstrates an excellent performance, achieving over 90% accuracy in predicting slug flow, churn flow, and annular flow void fractions, which surpasses the performance of the drift-flux model.