非结构混合网格自适应并行技术

计算流体力学（CFD）模拟实际工程问题所采用的网格规模可达千万量级,并行技术是减少计算时间的有效方法。耦合流场信息的网格自适应技术能有效动态优化计算网格,被NASA视为一项亟待发展的CFD关键技术。混合网格自适应系统包含网格分布优化、表面网格投影和空间网格匹配等关键技术。针对以上3项关键技术分别建立了高效的并行算法。首先,提出了先唯一后同一的两步法策略实现了网格单元分布优化过程的并行相容性;其次,基于局部曲面拟合思想,实现了曲面重构和新增物理网格点投影的完全并行;再次,提出了空间网格匹配技术的半并行算法,快速解决了网格单元交错问题。为了提高后续流场计算的并行效率,发展了基于并行重分区-网格数据迁移方法的动态负载平衡技术,并采用圆柱激波流场自适应模拟对动态负载平衡技术进行初步验证。最后,采用三角翼自适应加密测试了自适应系统的并行效率。结果表明,建立的混合网格自适应系统并行效率较高,且相比流场求解耗费总时间的比例低于1%。

Parallel algorithms for unstructured hybrid mesh adaptation

The mesh elements can reach ten-millions for real engineering cases to simulate their flow field with CFD. The parallel technique is an effective way to reduce the computing time. Mesh adaptation coupled with flow information can dynamically optimize computing mesh and has been considered by NASA as a key technique to be urgently developed. Elements distribution optimization, surface points projection, and inner points matching are three essential techniques for hybrid mesh adaptation. In this study, parallel algorithms for these three techniques are established. First, the two-stage strategy, which is called uniqueness-then-identity, is proposed to obtain the parallel consistence for the process of elements distribution optimization. Next, based on the local surface fitting method, the entire parallel algorithm is developed for surface reconstruction and new physical points projection. Then, the half-parallel algorithm for inner points movement is devised to efficiently eliminate the overlapping of mesh elements. To increase the parallel efficiency of the sequent flow field calculation, the dynamic load balance method, based on parallel partitioning and mesh transferring methods, is developed, and the flow around a cylinder is simulated to verify the dynamic load balance method. Finally, mesh refinement with adaptation for a delta-wing was adopted to test the parallel efficiency of the adaptation system. Numerical results show that the parallel efficiency of the hybrid mesh adaptation system is high enough and the total time cost is less than 1% of the flow solver.