A systematic review of machine learning applications in gas-liquid two-phase flow: From physical modeling to data-driven insights

Abstract

Gas–liquid two-phase flow has long been recognized as a difficult subject in the energy and process industries, mainly because of its highly complex fluid dynamics that make reliable modeling and prediction challenging. Over the years, a wide range of methods have been employed, including experimental studies, semi-empirical correlations, and numerical simulations. With the recent progress in machine learning (ML), data-driven modeling has opened new opportunities for analyzing and predicting two-phase flow behavior. This review summarizes research efforts on several representative problems—phase interface tracking, flow pattern recognition, pressure drop estimation, and critical heat flux (CHF) prediction. For each topic, we first examine conventional experimental and numerical techniques, then discuss emerging ML-based approaches, emphasizing their advantages, limitations, and practical scope. By bringing these methods together, the paper provides an integrated overview of the field and suggests future directions for advancing both fundamental research and industrial applications of two-phase flow.