A Review and Machine Learning Work for Downward Liquid-Gas Flow in Vertical Pipes

This study aims to better understand liquid-gas flow in downward vertical tubulars, a less studied topic with diverse applications in various industries, including Carbon Capture, Utilization and Storage (CCUS). The methodology includes a comprehensive literature review, data analysis, and leveraging machine learning to predict void fraction and flow pattern. A comprehensive dataset is compiled including 1500 points from 11 studies. The models are developed and evaluated for two cases, one using dimensional features, and then using dimensionless parameters. LGBM and CatBoost models perform best for both flow pattern and void fraction predictions. In addition, dimensionless parameters improve the model performances for both cases.