A Lattice Boltzmann Front-Tracking Interface Capturing Method based on Neural Network for Gas-Liquid Two-Phase Flow

Abstract

This paper presents a new method that accurately captures the interface of gas–liquid two-phase flow using a neural network-based lattice Boltzmann front-tracking interface capturing method. The motion of a single free-falling droplet is simulated using the Front Tracking Method (FTM), enabling the acquisition of information regarding the velocity field and interface points. The velocity field and interface points from the simulations are then utilised to generate input and output datasets for training the neural network (NN) models. Subsequently, the trained Bayesian regularised Back Propagation Neural Network (BRBPNN) model is integrated into the Lattice Boltzmann method (LBM), utilising the velocity field obtained from LBM simulation as input. The predicted LBM interface exhibits remarkable agreement with the FTM interface, as evidenced by a high correlation coefficient of 0.99945 for the ordinate value of the interface point in both methods. Therefore, the proposed method achieves precise positioning of the phase interface of LBM.