叶轮机械大规模CFD并行计算方法应用验证

Application verification of parallel computing method for large-scale CFD numerical simulation of turbomachinery

针对叶轮机械全环模型在大规模数值模拟中的计算需求,结合叶片通道模型的周期性特征,提出了一种创新的分层几何区域分解方法。这种方法通过有效地分解计算任务,实现了在不同计算核心间的负载均衡,优化了并行计算效率。在具备300PFlops(每秒3×10^(17)次浮点计算)和200PFlops(每秒2×10^(17)次浮点计算)计算能力的高性能计算平台上,本方法能够在1~6 h内完成1.8亿网格规模的多级压气机全环求解。通过对多个典型叶轮机械的算例分析,全面验证了所提出并行方法的性能和计算精度。结果表明,各模型的计算结果与实验数据高度一致,计算误差不超过6%,确保了计算准确。在万核规模的计算任务中,所采用的并行计算方法能够实现线性加速,并行效率在90%以上,证明了其在大规模计算中的高效性。此外,本方法在国产高性能计算平台上同样表现出良好的可移植性和适应性,充分展示了其在不同硬件架构上的广泛适用性及有效性。

In response to the computational demands of full-annulus models of turbomachinery in largescale numerical simulations,combined with the periodic nature of blade passage models,this study introduces an innovative method of hierarchical geometric domain decomposition.This method effectively decomposes com⁃putational tasks,achieving load balancing across different computing cores and optimizing parallel computing ef⁃ficiency.On high-performance computing platforms with capabilities of 300PFlops and 200PFlops,this method can complete the solution of full-annulus multi-stage compressor models with up to 180 million grid scales within 1~6 hours.Through the analysis of several typical turbomachinery cases,this research comprehensively validates the performance and computational accuracy of the proposed parallel method.The results demonstrate that the computational outcomes of each model are highly consistent with experimental data,and the calculation error does not exceed 6%,which ensures the accuracy of the calculation.In computational tasks of the scale of ten thousand cores,the employed parallel computing method achieves linear acceleration,and the parallel efficiency is more than 90%,proving its efficiency in large-scale computations.Additionally,this method also exhibits good portability and adaptability on domestic high-performance computing platforms,showing its broad applica⁃bility across different hardware architectures and the effectiveness.