航空学报 > 2024, Vol. 45 Issue (8): 128941-128941   doi: 10.7527/S1000-6893.2023.28941

GPU加速高阶谱差分方法在风扇噪声中的应用

张东飞, 高军辉()   

  1. 北京航空航天大学 能源与动力工程学院,北京 100191
  • 收稿日期:2023-04-28 修回日期:2023-05-17 接受日期:2023-06-16 出版日期:2024-04-25 发布日期:2023-06-27
  • 通讯作者: 高军辉 E-mail:gaojhui@buaa.edu.cn
  • 基金资助:
    国家重点研发计划(2018YFA0703300);国家科技重大专项(J2019-II-0006-0026);国家自然科学基金(NSFC-51876003)

Application of GPU⁃accelerated high⁃order spectral difference method in fan noise

Dongfei ZHANG, Junhui GAO()   

  1. School of Energy and Power Engineering,Beihang University,Beijing 100191,China
  • Received:2023-04-28 Revised:2023-05-17 Accepted:2023-06-16 Online:2024-04-25 Published:2023-06-27
  • Contact: Junhui GAO E-mail:gaojhui@buaa.edu.cn
  • Supported by:
    National Key Research and Development Project(2018YFA0703300);National Science and Technology Major Project(J2019-II-0006-0026);National Natural Science Foundation of China(NSFC-51876003)

摘要:

风扇是大涵道比涡扇发动机的主要噪声源之一,准确预测风扇噪声对声学设计、噪声机制分析具有重要意义。高精度计算气动声学方法是风扇噪声计算的一个重要途径,但是由于巨大的计算资源需求,限制了其在风扇噪声计算上的应用。将图形处理器(GPU)加速计算方法应用到高阶谱差分计算气动声学求解器中,对不同的GPU和中央处理器(CPU)异构计算模式进行了测试和优化,并分析了影响异构计算效率的瓶颈。在A100 GPU卡上测试的结果表明,相比于28核的CPU计算节点,采用静止网格、运动网格的GPU计算加速比分别为20.4、14.0。将GPU加速的计算气动声学求解器应用到低速风扇噪声数值模拟中,准确预测了管道内前两阶叶片通过频率(BPF)的主导模态,与实验测量结果相比,模态声功率级误差在5 dB以内。

关键词: GPU, 异构计算, 风扇噪声, 谱差分方法, 叶轮机械

Abstract:

The fan is one of the main noise sources of the large bypass turbofan engine, and accurate prediction of fan noise is of great significance for acoustic design and noise mechanism analysis. High-precision computational aeroacoustic methods are an important approach to fan noise calculation, where, however, their application is limited due to the enormous computational resource requirements. This paper applies GPU-accelerated computing methods to a high-order spectral difference computational aeroacoustic solver, tests and optimizes different GPU/CPU heterogeneous computing modes, and analyzes the bottlenecks affecting heterogeneous computing efficiency. Test results on a single A100 GPU card show that the GPU computational speed-up ratios are 20.4 and 14.0 for stationary and rotating grids, respectively, compared with a dual-CPU 28-core computing node. Finally, this paper applies the GPU-accelerated computational aeroacoustic solver to the numerical simulation of low-speed fan noise, accurately predicting the dominant modes of the first two orders of Blade Passage Frequency (BPF) in the duct. Compared with the experimental results, the error in the modal sound power level is within 5 dB.

Key words: GPU, heterogeneous computing, fan noise, spectral difference method, turbomachinery

中图分类号: