首页 >

压气机整机超大规模非定常模拟关键技术研究

王子维1,范召林1,李彬2,曹杰1,邓亮3,王年华1,江雄1   

  1. 1. 中国空气动力研究与发展中心
    2. 中国空气动力研究与发展中心计算空气动力研究所
    3. 中国空气动力研究与发展中心 计算空气动力研究所
  • 收稿日期:2023-11-13 修回日期:2024-03-08 出版日期:2024-03-14 发布日期:2024-03-14
  • 通讯作者: 王子维

Research on Key Technologies for Massive Unsteady Simulations of the Whole Compressor

  • Received:2023-11-13 Revised:2024-03-08 Online:2024-03-14 Published:2024-03-14

摘要: 航发压气机整机往往包含多个涵道以及多达十几个压气机级,全环非定常模拟技术是提高其内流场模拟保真度的手段之一。压气机整机数十亿量级的全环网格量导致非定常模拟计算量巨大,开发具备高可扩展度的用于超大规模压气机非定常CFD模拟的求解器十分必要。基于团队自主发展的轴流压气机数值模拟软件ASPAC,给出了基于分区后单通道网格定常流场生成分区后全环网格初场的通用解决方案;对国际上基于单通道壁面距离生成全环壁面距离的方法进行了分析和测试,给出该方法导致非定常模拟结果产生的偏差分布区域和量级;发展了考虑网格畸变的叶排交界面网格重叠定位策略;完成了压气机非定常模拟的MPI/OpenMP混合并行改造,并通过优化求解流程,解决了OpenMP并行模拟时的数据竞争问题。结果表明,发展的方法成功应用于双涵道13级压气机61.16亿网格的10万核级全环非定常模拟,MPI/OpenMP混合并行模式基本解决了叶排交界面动态插值处理导致的负载不平衡问题,10万核心相比万核心的并行效率达到84.7%,远高于MPI并行模式46.7%的并行效率。

关键词: 双涵道, 13级压气机, 全环非定常, 壁面距离, MPI/OpenMP混合并行

Abstract: The compressor of aero-engine often contains multiple ducts and up to a dozen compressor stages, and the full-annulus unsteady simulation technology is one of the means to improve the fidelity of its internal flow field simulation. The billions of grids of the compressor result in a huge amount of unsteady simulation computation, it is necessary to develop a highly scalable solver for the massive unsteady CFD simulation of the compressor. Based on the in-house software ASPAC, a universal solution for generating the initial field of the full-annulus grid after partitioning based on the single channel grid steady flow field is provided; analysis and testing were conducted on the international method of generating full-annulus wall distance based on single channel wall distance, and the deviation distribution area and magnitude caused by this method in unsteady simulation results were given; developed a grid overlap localization strategy for blade row interface considering grid distortion; the MPI/OpenMP hybrid parallel transformation of compressor unsteady simulation is conducted, and the problem of data competition during OpenMP parallel simulation is solved by optimizing the solution process. The results show that the developed method has been successfully applied to the full-annulus unsteady simulation of a twin-spool 13-stage compressor with 6.116 billion grids using 102,400 CPU cores. The MPI/OpenMP hybrid parallel mode has basically solved the load imbalance problem caused by the dynamic processing of the blade row interface interpolation. The parallel efficiency of the 102,400 cores compared to the 10,240 cores is 84.7%, far higher than the parallel efficiency of the MPI parallel mode of 46.7%.

Key words: twin-spool, 13-stage compressor, full-annulus unsteady, wall distance, MPI/OpenMP hybrid parallel

中图分类号: