基于流动特征增强型Co-Kriging的航空压气机稳定裕度不确定性评估

  • 郭正涛 ,
  • 张国斌 ,
  • 李潮隆 ,
  • 王一 ,
  • 包磊 ,
  • 高显忠
展开
  • 1. 中国人民解放军国防科技大学
    2. 国防科技大学空天科学学院
    3. 国防科技大学试验训练基地

收稿日期: 2026-07-25

  修回日期: 2026-10-11

  网络出版日期: 2026-10-19

基金资助

国家自然科学基金;军队引进高层次创新人才科研启动项目;国防科技大学自主创新基金项目

Flow Feature-Enhanced Co-Kriging for Uncertainty Quantification of Stability Margin in Aero-Compressors

  • GUO Zheng-Tao ,
  • ZHANG Guo-Bin ,
  • LI Chao-Long ,
  • WANG Yi ,
  • BAO Lei ,
  • GAO Xian-Zhong
Expand

Received date: 2026-07-25

  Revised date: 2026-10-11

  Online published: 2026-10-19

摘要

轴流压气机向“高负荷、少级数、薄叶片”方向发展,叶片加工误差引发的气动稳定问题,已成为国产航空发动机试飞阶段性能分散大、非预期喘振频发的重要诱因。目前,稳定边界的数值判定依赖逐步抬升出口背压直至计算发散,单次计算成本高昂,导致基于大样本几何抽样的稳定裕度不确定性量化面临极高的计算成本。针对这一难点,本文提出一种流动特征增强的稳定裕度不确定性建模方法。该方法避开单纯数据驱动的路径,转而利用叶顶间隙泄漏流与主流的轴向动量比同压气机失稳的强相关性,首先通过稀疏多项式混沌构建加工误差与该微观流动特征的映射关系,再将其作为物理先验知识嵌入多可信度Co-Kriging框架:利用大批量低成本的流动特征数据捕捉稳定裕度变化趋势,仅依靠少量高精度数值逼喘数据修正局部偏差。研究表明,引入物理先验知识后的模型泛化能力显著提升,在同等几何训练样本规模下,稳定裕度的预测精度较传稀疏多项式混沌、Kriging及全连接神经网络统等纯数据驱动模型提高了一个数量级。在此基础上,以某实际八级压气机为研究对象,定量揭示了加工误差作用下稳定裕度的非高斯概率演化规律,辨识出前面级叶顶区域的安装角等关键几何敏感参数,并结合实装试验与流动机理分析验证了结果的有效性。研究成果为高负荷压气机由“被动容差”向“主动设计驾驭不确定性”的转变提供了理论与方法支撑。

本文引用格式

郭正涛 , 张国斌 , 李潮隆 , 王一 , 包磊 , 高显忠 . 基于流动特征增强型Co-Kriging的航空压气机稳定裕度不确定性评估[J]. 航空学报, 0 : 1 -0 . DOI: 10.7527/S1000-6893.2026.34297

Abstract

Axial compressors are evolving toward high-load, fewer-stage, and thin-blade configurations. However, aerodynamic stability issues induced by blade manufacturing errors have become a major contributor to significant performance scatter and frequent unanticipated surges during flight testing of domestic aero-engines. Currently, numerical determination of the stability boundary relies on incrementally raising the outlet back-pressure until computational divergence occurs, resulting in prohibitively high costs per simulation. This poses a severe computational challenge for uncertainty quantification (UQ) of stability margins based on large-sample Monte Carlo geometric sampling. To address this bottleneck, this paper proposes a stability margin uncertainty modeling method enhanced by flow physics. Diverging from purely data-driven approaches, the proposed method leverages the strong correlation between compressor instability and the axial momentum ratio of tip clearance leakage flow to the mainstream. First, a mapping between manufacturing errors and this microscopic flow feature is established using sparse polynomial chaos expansion (PCE). Subsequently, this mapping is embedded into a multi-fidelity Co-Kriging framework as physical prior knowledge: massive low-cost flow feature data are utilized to capture the global trend of the stability margin, while only a limited number of high-fidelity near-stall simulation data are required to correct local deviations. The results demonstrate that incorporating physical priors significantly enhances the model's generalization capability. Under the same geometric sample size, the prediction accuracy of the stability margin is improved by an order of magnitude compared to conventional purely data-driven models such as standalone PCE, Kriging, and fully connected neural networks. Furthermore, based on a practical eight-stage axial compressor, the non-Gaussian probabilistic evolution of the stability margin under manufacturing errors is quantitatively revealed. Key geometric sensitivity parameters, such as the stagger angle in the front-stage tip region, are identified. The validity of the findings is corroborated through full-annulus experiments and flow mechanism analysis. This research provides theoretical and methodological support for shifting high-load compressor design from "passive tolerance" to "active uncertainty management."
Options
文章导航

/

〈 〉