Research On High Quality Unstructured Mesh Generation Algorithms With Multilevel Parallelism

Mesh generation is a very important pre-processing step in the numerical solution of partial differential equations,which usually takes up most of the time in the whole solution process.At the same time,as the complexity of solving the problem becomes higher and higher,and the precision of the result is higher and higher,the scale and complexity of the mesh required are larger and higher.Mesh generation is a very important pre-processing step in the numerical solution of partial differential equations,which usually takes up most of the entire solution process.In recent years,unstructured mesh have become a research hotspot because of their higher flexibility and better proximity to geometric surfaces than structural mesh.Aiming at the problems of low automation of unstructured mesh generation and long time-consuming of large-scale mesh generation,this paper studies unstructured mesh optimization,parallel algorithm of mesh generation and geometric adaptation.The main work and contributions are as follows:(1)In this paper,the whole process of constructing unstructured grid generators is deeply analyzed,and the technical difficulties that researchers should pay attention to when building grid generators that meet their own needs are pointed out.In order to solve the problem that many low-quality mesh cells are generated in the process of mesh automation generation,this paper constructs a composite mesh optimization algorithm,and puts forward its own composite optimization scheme through experiments.The experimental results show that after optimization,the number of mesh decreases by 32.9%,the number of low-quality mesh elements is reduced.There are almost no grid cells with a quality score below 0.5,and more than 85% of the grid cells have a quality score above 0.75.(2)Aiming at the time-consuming problem of large-scale mesh generation,a multi-level parallel unstructured mesh generation algorithm is implemented in this paper.Firstly,thread-level parallel optimization is performed on five functions with natural parallelism that are called many times and time-consuming in the unstructured mesh generator.Secondly,two different pre-processing parallel algorithms are designed,which are suitable for different scenarios.Finally,millions of mesh generation experiments on mechanical devices,three-dimensional wing and wing-body assembly with tail were carried out on Tianhe-2 supercomputer.The test results show that the average time-consuming reduction of the function optimized by OpenMP is 41.8%.The process-level parallel grid generation time is gradually reduced as the number of processes increases.The acceleration ratio of the 24 processes is 21.5,and a good parallel effect is achieved.Aiming at the complexity and time-consuming characteristics of using background mesh to realize geometric adaptive mesh generation,this paper proposes an algorithm based on Support Vector Machine(SVM)to construct geometric adaptive mesh.In the application of this algorithm,there is no need to construct background mesh and smooth the size field.By comparing the simple two-dimensional surface meshes,it is found that the generated meshes have better geometric adaptability.From the area with low mesh density to the area with high mesh density,the change trend of mesh size is gentle.