Parallel Algorithm Research Of Lattice Boltzmann Method Based On Multi-layer Grid

Large-scale numerical simulation of complex flows has been an important research area in scientific and engineering computing.Computational fluid dynamics(CFD)uses computer science and numerical simulation for fluid flow.Lattice Boltzmann Method(LBM)has become a novel research direction in computational fluid dynamics as a mesoscopic approach to solve flow field problems.The main features of LBM are clear physical background,simple algorithm construction,and good parallelism,so it is suitable for parallel computation on large clusters.However,the standard LBM uses a standard Cartesian grid as the computational grid.Therefore,in order to accurately simulate complex flows,the resolution of the full area grid must be increased,which will lead to computational inefficiencies.In the context of grid resolution affecting the efficiency of numerical simulation runs,non-uniform grid design methods are beginning to be applied to simulations for practical applications,especially multi-layer grids.The multi-layer grid refinement strategy can handle different levels of computational complexity through multi-scale grids,and can be used to solve the complex flow fields of non-uniform grid LBMs without destroying the parallelism of standard LBM.For the numerical simulation of complex flows,the application of mesh refinement techniques to LBMs can effectively address the problems of accuracy,efficiency,and mesh dependence.High-performance computing techniques provide efficient,promising,and reliable solvers for handling large-scale computational problems in LBMs.Many current frameworks exhibit good scalability while providing support for multi-layer local grid refinement.Existing research focuses on parallel implementation of multi-layer grid LBMs based on traditional meshing techniques,and the interpolation steps between different resolution grids are also usually implemented by complex time and space interpolation methods.By analyzing the existing studies and schemes,this paper finds that the load-balancing based parallel potential of multilayer grid LBM needs to be further optimized and enhanced,so there is a need to design implementation methods that can fully utilize the parallel performance of multi-layer grid LBM.In this paper,a lattice Boltzmann method applicable to the multi-layer grid condition is proposed based on three aspects of multi-layer grid technology,LBM and high performance computing,and a multilayer grid generation based on the lattice Boltzmann method is also given.The potential parallelism is analyzed in detail for multilayer grid generation and multi-layer grid lattice Boltzmann method,and the corresponding parallel strategies and algorithms are given from the perspectives of buffer optimization and distributed storage,respectively.The main research contents and results of the paper are as follows:(1)By using buffering technique,a multi-layer grid LBM with only spatial interpolation calculation is proposed in this paper.based on the computational characteristics of LBM,a buffer grid is proposed for the first time at the interface of different levels of grids for eliminating the time interpolation calculation and simplifying the spatial interpolation calculation.For two-dimensional and three-dimensional flow field calculation problems,the corresponding multi-layer grid LBM algorithm is proposed.For the two-dimensional problem of cylindrical winding flow,the simulation results show that the multi-layer grid LBM proposed in this paper can accurately obtain the aerodynamic coefficients and Strouhal numbers at different Reynolds numbers,and its total number of grids is about 1/9 of that of the single-layer grid LBM,which is 6.76 times faster.For the three-dimensional flow calculation problem,numerical experiments of flow through a sphere are simulated to verify the numerical accuracy of the proposed method for Reynolds numbers of 100,200,250,300 and 1000.Using flow and velocity contours,it is demonstrated that the multilayer mesh LBM can be computed accurately even at interfaces with different size meshes.(2)The MPI cross-node message-passing based parallel algorithm for multilayer grid LBM is proposed.Taking a one-dimensional grid parallel partitioning scheme as an example,this paper presents the implementation details of the proposed parallel algorithm.Simply by expanding the distribution functions of non-uniform meshes in different parallel domains and the necessary data transfer of macroscopic quantities,our approach can be used for numerical simulations of flow field problems with complex geometries.Finally,the good scalability and parallel efficiency of the parallel algorithm is verified by numerical experiments simulating 2D and 3D flow scenarios.In this paper,a specific fluid simulation scenario is selected for comparison experiments,and the weak scalability performance can be as high as 88.90% in a 16-thread environment,while the numerical simulation still has 77.4% parallel efficiency when the parallel domain is extended to 16 threads.