PARALLEL GENERATION OF HEXAHEDRAL-DOMINANT HIGH ORDER CURVILINEAR MESH

Higher-order methods play a significant role in numerical simulations of turbulence, combustion, and aeroacoustics. When applied to higher-order methods, straight meshes may introduce non-physical solutions, whereas employing higher-order curved meshes can effectively reduce numerical errors and improve simulation accuracy. Based on initial coarse straight meshes, hexahedral-dominant(hex-dominant) higher-order curved meshes are generated by first subdividing elements into hex-dominant meshes, followed by order elevation via point insertion, geometry-preserving projection, and deformation optimization. During the mesh generation process, higher-order points are added to the straight mesh using a concurrent hash table data structure. High-order points on boundary surfaces are projected onto the geometry using a multi-threaded parallel approach. Additionally, a localized radial basis function method based on the R-Tree data structure is employed to deform interior points, untangling potential inverted elements. Large-scale higher-order curved mesh generation for the DLR-F6 wing-body configuration demonstrates that, starting from 16.55 million elements, elevating to P2-order curved meshes using 16 threads takes 15.5 minutes, achieving a parallel efficiency of 54%. Finally, numerical verification and validations using the GPU-accelerated PyFR higher-order solver are conducted. The accuracy test results for Taylor-Couette flow show that the expected order of accuracy is achieved on the high-order meshes presented in this paper. Other three validation cases—flow over a sphere, an ellipsoid, and a SD7003 wing—at different Reynolds numbers confirm that the generated higher-order curved meshes meet the computational requirements of high-order numerical schemes.