Research On Tetrahedral Mesh Generation Method With Multilevel Parallelism

Finite element analysis uses numerical methods to simulate real physical systems,optimize physical prototype design and experiments,and is widely used in the research and application fields of computational electromagnetic.However,the reality model is becoming more and more complex,and the calculation accuracy is gradually improved,which puts forward higher requirements for the quality and scale of finite element mesh generation.As a common mesh generation technology,tetrahedral mesh has the advantages of high flexibility and good adaptability,which is the focus of mesh generation research at present.In this paper,the three major problems of tetrahedral mesh generation,such as slow generation speed,low quality and insufficient independence between sub regions,are studied.The main work includes:(1)For the problem of low quality of tetrahedral mesh,a multi-strategy parallel optimization algorithm is designed.Several groups of different optimization strategies are designed,and experiments are carried out one by one to verify the optimization effect.In order to improve the experimental speed,a multi-strategy parallel optimization algorithm is designed.Multiple groups of strategies are handed over to multiple processes to verify the optimization effect in parallel.The optimization strategy with the best effect is found through one program run,which greatly improves the experimental efficiency.When the four strategies are tested in parallel at the same time,the experimental time is greatly reduced and the acceleration of 2.13 times is realized.(2)For the slow speed of generating large-scale grid,a multi-level parallel grid generation algorithm is designed.The Delaunay fine-grained parallel point insertion algorithm is improved.A coarse-grained parallel mesh generation algorithm based on domain decomposition is designed and combined with the improved Delaunay algorithm to form a coarse-grained and fine-grained multi-level parallel mesh generation algorithm.Experiments show the feasibility and efficiency of the improved Delaunay algorithm and multi-level parallel algorithm.When the number of processes is set to 20 in the experimental environment,the speedup ratio of the improved Delaunay algorithm is 12.9 and the parallel efficiency is 0.6;The multi-level parallel algorithm achieves 2x acceleration in 3 processes and 9x acceleration in 10 processes.(3)For the problem of low independence between sub regions after domain decomposition,the concept of mapping region projection is proposed,and a complete multi-level parallel tetrahedral mesh generation algorithm with mapping region element projection is realized based on the numbering method of barycentric global ID.Two models with demonstration effect are tested to verify the feasibility of the algorithm and the correctness of the parallel effect,projection effect and global ID number achieved by the algorithm.