Machine Learning Of LBM+FTM Coupling For Single Vacuole Ascent

Multiphase fluid flow at mesoscopic scale is a common phenomenon between macroscopic and microscopic,which contains complex fluid dynamics characteristics.It is widely used in the fields of biomedicine,aerospace,nano engineering and environmental engineering,and has extremely important research value.However,due to the limitations of the current experimental research and simulation research,the study of its fluid dynamics is not well understood.Currently,there are many methods applied to the numerical simulation of multiphase flows.One is the Front-tracking Method(FTM),which can achieve high precision capture of phase interface morphology in multiphase flows.The other method is lattice Boltzmann method(LBM),which can realize the efficient solution of complex flow field.Machine learning tends to establish internal connections between data.Based on the accuracy of Neural Network training,this paper chooses Bayesian Regularized Back Propagation Neural Network(BRBPNN)model.However,FTM method can accurately track the interface,capture the location of the interface and the interface shape of multiphase fluid,but its solution process is complicated.LBM method has a simple calculation process,which is suitable for solving the flow field in complex structures.However,its interface tracking accuracy is very low,and it cannot accurately calculate the phase interface problem in the dynamic process.In order to give consideration to interface capturing and efficient flow field solving in multiphase fluid flow,a FT-LB coupling model based on machine learning is proposed in this paper.In this paper,single vacuole rise is selected as the physical model,FTM numerical simulation is used to obtain the velocity and position of the flow field and phase interface markers,and machine learning input set and output set are made with it.The corresponding relationship between velocity and position of phase interface is established through BRBPNN machine learning model training.Then,the flow field data simulated by LBM is taken as the input,and the trained BRBPNN neural network is used to predict the position of phase interface in LBM flow field.In order to verify the accuracy of the predicted interface,it needs to be compared with the position of phase interface simulated by FTM.In order to ensure the applicability and accuracy of the FT-LB coupling model based on machine learning,a variety of working conditions are selected in this paper.The vacuole radius was 0.06,0.08,0.1,0.12 and 0.14,the surface tension coefficient was 0.01,0.015,0.02,0.025 and 0.03,and the two-phase dynamic viscosity ratio was 5,7.5,10,12.5 and 15,respectively.The gravity acceleration is-0.005,-0.0075,-0.01,-0.0125 and-0.15 respectively under 17 conditions.After calculation,the simulation results of FTM and LBM under various working conditions accord with the basic laws of fluid dynamics.The fitting degree coefficient R∈[0.9999,1] of the training results of BRPBNN machine learning model indicates that the training results of neural network are very good.The BRBPNN model trained by FTM simulation data continues to accurately predict the position of phase interface in LBM flow field.Through image comparison and regression analysis,the fitting correlation coefficient R∈[0.995,1] between the predicted interface and the actual interface is obtained,which shows a high degree of fitting.Successfully completed the development,debugging,optimization and verification of the FT-LB coupling model based on machine learning.