MACHINE LEARNING-BASED MODEL FOR MULTIPHASE FLOW USING VOLUME OF FLUID METHOD: AIR INJECTION INTO A TANK FILLED WITH LIQUID

Multiphase flow (MF) refers to the simultaneous flow of materials in two or more thermodynamic phases and is encountered in diverse applications, including bioengineering, power generation, oil and gas transport, pharmaceuticals, combustion engines, chemical processes, and biological systems. Accurate modelling of MF is essential for understanding and optimizing such processes. Among the various numerical techniques, the Volume of Fluid (VOF) method is one of the most widely used, as it effectively captures free-surface phenomena such as liquid sloshing, bubble dynamics, and multiphase mixing. A representative case is air injection into a liquid-filled tank, a process relevant to wastewater treatment, fermentation, and aquaculture. In this work, a machine learning (ML)-based model is developed to predict the dynamics of air injection into a water-filled tank. The two-dimensional geometry consists of a 0.5m × 0.25m tank filled with water, with air injected through a centrally located 5mm nozzle at the bottom. The flow is assumed to be laminar. The methodology involves two steps: generation of training data using ANSYS Fluent, followed by training the ML model through joint learning of shared parameters between physics-uninformed and physics-informed neural networks, using mean squared error minimization. To the best of our knowledge, no ML-based framework for air-injection problem has been reported in the literature, highlighting the novelty of this study. Simulations for training data were performed until the injected air reached the tank surface. Results show that the ML model accurately predicts air volume fraction, velocity fields, and pressure differences across different time steps.