Study On Pseudo-potential Multiphase Flow Model Based On Lattice Boltzmann Method And Its GPU Parallel Acceleration

Multiphase flow is a widely observed fluid phenomenon in both natural and engineering domains.Studying multiphase flow is important for understanding complex phenomena in nature and enhancing productivity in engineering.The lattice Boltzmann method,known for its clear physical concepts,ease of handling complex boundaries,and inherent parallelism,has become a mainstream numerical simulation approach for multiphase flow.Among them,the pseudo-potential model has emerged as an important model for simulating multiphase flow due to its simplicity,high computational efficiency,and ease of implementation.However,the original pseudo-potential model has some limitations,such as low density ratios,the inability to independently adjust surface tension,and large spurious current,which limit its practical application in engineering.To address these limitations,this thesis proposes an improved pseudo-potential lattice Boltzmann multiphase flow model that achieves extremely large density ratios and independently adjustable surface tension.Additionally,when using the lattice Boltzmann model to solve problems,discretizing the flow field into a large number of grids results in significant computational load.This thesis introduces GPU parallel computing techniques and designs and implements two parallel algorithms to accelerate the simulation process of the model.The main work is as follows:（1）A multi-relaxation pseudo-potential lattice Boltzmann model was developed,capable of simulating multiphase flows with large density ratios and independently adjustable surface tension.The model employed a decoupling approach between the moment space and the grid space to widen the gas-liquid transition interface,and a high-order accuracy finite difference method was utilized to improve effective density gradient calculation precision for achieving large density ratios.Moreover,a modified pressure tensor scheme was adopted to enable independent adjustment of surface tension.Numerical experiments were conducted to validate the performance of the proposed model,including thermodynamic consistency,Laplace’s law,spurious velocities,oscillation of elliptical droplets,and droplet impact on liquid films.The experimental results show that the model still meets thermodynamic consistency at very low temperatures with gas-liquid density ratios exceeding 10¹⁰.The model can adjust surface tension over a wide range without being affected by the density ratio,and the maximum spurious current does not exceed10^-4.When simulating complex fluid problems involving extremely large density ratios,adjustable surface tension,and high Reynolds numbers,this model also maintains good numerical stability.（2）Two GPU parallel algorithms were designed and implemented.Firstly,based on the Structure of Arrays data layout and utilizing the Push streaming scheme,a dual-grid GPU parallel algorithm was developed.Comprehensive optimization was achieved through thread organization and register utilization schemes.Experimental results demonstrated that compared to traditional CPU serial algorithms,this algorithm achieved an acceleration of approximately 778 times.However,this algorithm required double the memory space,which was limited by the capacity of the GPU memory and thereby restricted the scale of the flow field under study.To address this limitation,a single-grid GPU parallel algorithm was designed based on the ET memory access pattern.Similarly,thread organization and register utilization schemes were applied for comprehensive optimization.Experimental results showed that the algorithm achieved an acceleration of approximately 723 times and a reduction of nearly 50%in memory consumption.