Two-phase Droplet Simulation in Pressure-Driven Channel Flow with Physics-Informed Neural Networks

Purpose Using artificial intelligence to model two-phase flow could significantly impact the field of computational science. In this study, neural networks are used to simulate two-phase flow. A new neural network architecture is proposed for two-phase flow simulations, and the evolution of a droplet within a channel under pressure-driven flow is investigated. The neural networks are trained solely using the governing equations along with the initial and boundary conditions. The purpose of this study is to simulate a droplet dynamics in channel using the neural network. Design/methodology/approach The physics-informed neural networks (PINNs) method is used to simulate droplet evolution in channel. Computational fluid dynamics (CFD) is also used to simulate droplet evolution for comparison with the PINNs results. Two connected neural networks are used to simulate the two-phase flow. Different Reynolds numbers are investigated. The velocity and pressure contours obtained from PINNs are compared against CFD, and the interface evolution is analyzed. Findings The PINNs results show good agreement with the CFD study. The findings demonstrate that PINNs are capable of simulating two-dimensional, two-phase flow in a channel. The results for intermediate Reynolds numbers closely match the CFD predictions. The PINNs method satisfies mass conservation and accurately captures the interface dynamics. Both pressure and velocity fields are predicted with acceptable accuracy. Originality/value This study highlights the potential of the PINNs method for simulating two-phase flow phenomena. It represents an initial step in the development of artificial intelligence–based modeling approaches. The findings demonstrate that the PINNs framework can provide a new perspective and methodology for more complicated two-phase flow simulation.