Research On Multi-process Mesh Subdivision Projection Algorithm For E-level Computing

After long-term development,the numerical simulation method represented by finite element has gradually become a key tool to support academic research and industrial design.However,computational bottlenecks in mesh generation and other aspects still restrict this tool to play a greater role.Using the powerful computing power of supercomputer clusters to break through the time performance bottleneck encountered in the grid generation and optimization process has become a solution.However,the current research on grid parallel generation is still immature.This article first introduces the advantages and disadvantages of different parallel protocols and briefly introduces the content of parallel grid generation,and points out the possibility of improvement in the program framework of traditional parallel grid survival.Then,this paper developed a grid projection algorithm based on one-dimensional search algorithm.In order to run the grid optimization steps in parallel on distributed computers,this paper obtains a local optimization function based on an improved global optimization function.The main work of this paper is as follows:(1)In the research of this thesis,a parallel gird generation technique is proposed to generate large-scale hybrid meshes.Our method is to process the initial coarse mesh with tetrahedral,triangular prism,pyramidal and other mesh elements as the original mesh.The original mesh is divided into multiple sub-regions by the METIS library,and the sub-regions are allocated.For different CPU cores,coordinate work is performed between different CPUs through the communication mechanism in MPI,and the mesh generation operation is synchronized.This paper provides several subdivision methods for common mesh cell types,and proposes a solution to the boundary problem of different mesh partitions when retrieving mesh data in parallel mesh generation.(2)In this paper,a mixed coarse grid of tetrahedrons,triangular prisms,pyramids and other grid units will be used as the original grid,and the corresponding data structure will be designed to complete the grid subdivision function.On this basis,this paper uses the synchronization mechanism of local index and global index to solve the subsequent recycling problem of different partition grids.In the process of grid projection,this paper introduces a mapping table to speed up the search speed of grid point projection.In this paper,the surface projection problem is transformed into a one-dimensional search optimization problem,and the one-dimensional search method is used to solve the grid points on the curve or surface to improve the calculation speed of the projection algorithm.This paper improves the local grid interpolation algorithm based on the global grid interpolation algorithm.By modifying the weight parameters in the global grid interpolation algorithm and increasing the damping function,the local grid interpolation algorithm retains the flexibility of the global grid interpolation algorithm and improves the calculation time performance.In this paper,the grid space is divided into a number of boxes,and the digital alternating tree is used for storage,so that the local grid interpolation algorithm does not need to calculate most of the non-influencing grid points.Finally,on this basis,this paper presents a step-by-step grid optimization algorithm for optimizing the displacement of surface grid points to solve the problem of grid cell quality after grid projection.This paper proves through experiments that the grid structure obtained by the grid subdivision scheme used is accurate,and the combined output of the grids recovered from each partition is consistent with the expected;in the comparison experiment,the robustness of the mapping table method and the grid The grid projection function has a certain optimization effect on the calculation efficiency.The optimization algorithm based on the one-dimensional search method has a better performance in the optimization of the calculation efficiency compared with the traditional algorithm;the comparison between the global interpolation algorithm and the local interpolation algorithm In the experiment,the computational efficiency advantage of the local optimization algorithm and the advantage of the mesh generation quality near the displacement point are demonstrated.