Low Capillary Elastic Flow Model Optimization Using the Lattice Boltzmann Method and Non-Dominated Sorting Genetic Algorithm

In simulations of elastic flow using the lattice Boltzmann method (LBM), the steady-state behavior of the flow at low capillary numbers is typically poor and prone to the formation of bubbles with inhomogeneous lengths. This phenomenon undermines the precise control of heat transfer, mass transfer, and reactions within microchannels and microreactors. This paper establishes an LBM multiphase flow model enhanced by machine learning. The hyperparameters of the machine learning model are optimized using the particle swarm algorithm. In contrast, the non-dominated sorting genetic algorithm (NSGA-II) is incorporated to optimize bubble lengths and stability. This results in a coupled multiphase flow numerical simulation model that integrates LBM, machine learning, and the particle swarm algorithm. Using this model, we investigate the influence of elastic flow parameters on bubble length and stability in a T-shaped microchannel. The simulation results demonstrate that the proposed LBM multiphase flow model can effectively predict bubble elongation rates under complex conditions. Furthermore, multi-objective optimization determines the optimal gas-liquid two-phase inlet flow rate relationship, significantly mitigating elastic flow instability at low capillary numbers. This approach enhances the controllability of the elastic flow process and improves the efficiency of mass and heat transfer.