叶轮机械全环非定常大规模并行模拟程序设计

Massive parallel program design for unsteady full annulus turbomachine simulation

  • 摘要: 为了更加精确地模拟叶轮机械内部复杂的三维非定常流动,开展了全环网格模拟方法探索,提出了针对滑移面的守恒变量插值和高效的大规模并行计算解决方案。基于千万亿次高性能计算平台,自主开发了非定常数值模拟程序,采用单级压气机、1.5级涡轮和3.5级高负荷压气机三个算例进行了程序的验证与确认,以及并行效率测试工作。模拟采用的最大网格数为2亿,计算核心数为4096个。结果表明:程序计算得到的整体气动性能与试验数据吻合良好;程序所采用的全环非定常计算模型能保证交界面处物理量的连续,削弱非物理熵增,得到更加可信的流场解;所采用的METIS的网格分区方法以及MPI并行策略使程序具有良好的负载均衡和并行效率;同时程序对超大规模的复杂问题具备良好的可扩展性和适应性,能够满足工程实际问题的流动分析和叶轮机械的精细流动研究要求。

     

    Abstract: In order to get an accurate simulation of the complex three-dimensional unsteady flows in turbomachine, an in-housed unsteady turbomachinery simulation program is developed based on a petaflops high performance computing platform. The full annulus grid simulation method has been developed. The conservative variable interpolation method for sliding interfaces and the high efficient massive parallel computation scheme are discussed in detail. A single stage compressor, a 1.5 stage turbine, and a 3.5 stage high-loaded compressor were simulated to verify and validate the program. The parallel efficiency of the program was also tested. The maximum grid number of the testcases is approximately 200 million, and the number of CPU cores is up to 4096. The results show that aerodynamic performance computed by this program agrees well with experimental data. The full unsteady computation method adopted in this paper guarantees variables' continuity at the sliding interface, weakens nonphysical entropy generation, and yields a reasonable flowfield. The grid partition methods by METIS and the parallel program scheme of MPI contribute to the program with a good load balance and a high parallel efficiency. This program has good expansibility and adaptability to the complex problem of large scale and satisfies the need of complex application flow analysis and exquisite turbomachiney flow filed study.

     

/

返回文章
返回