A novel approach for large-scale hybrid mesh generation on distributed systems

This study presents an approach for the generation of a large-scale viscous mesh using a parallel advancing layer method based on distributed systems with multi-cores. A triangular surface mesh that conforms with the original geometrical model is prepared as input and decomposed into multiple subdomains. A coarser background mesh with high-quality viscous elements from the same model is employed for parallel computation. Instead of using a conventional approach to generate layer-wise prismatic elements, a normal calculation through linear interpolation is implemented to overcome the boundary problems arising in distributed systems. A novel strategy to deal with cross-processor transition elements is proposed to reduce frequent communications during viscous mesh generation. To improve the parallel efficiency, a fast domain decomposition method is introduced to speed up the parallel meshing procedure of tetrahedral elements. Numerical experiments are performed on high-performance computing systems with billions of mesh elements to demonstrate the effectiveness and efficiency of the proposed approach and numerical simulation results represent a promising applicability.