加工误差对压气机叶栅气动性能及稳定性影响的数据挖掘

Data Mining on Effects of Manufacturing Error on Aerodynamic Performance and Stability of Compressor Cascade

关键词:  法向加工误差  数据挖掘  人工神经网络  气动性能  气动稳定性  压气机

Key words:  Normal manufacturing error  Data mining  Artificial Neural Network  Aerodynamic performance  Aerodynamic stability  Compressor

摘要:

为缓解加工误差影响评估过程中的“维数灾难”，结合展向平均假设和基于高斯过程的Karhunen-Loève展开，提出了一种由法向加工误差导致的三维叶片表面几何不确定性降阶模型。通过给定加工误差分布的标准差函数求解几何不确定性降阶模型，并运用伪蒙特卡洛方法随机生成样本，最终训练人工神经网络预测了加工误差对高负荷直叶栅气动性能和角区气动稳定性的影响。结果表明：加工误差的影响与所处的工况有关；当处于近失速工况时，相对于原型，平均气动性能和稳定性降低，总压损失系数增大的概率约为83.29%，静压系数减小的概率约为79.20%，失速因子增大的概率约为69.10%； 与设计工况相比，近失速工况下气动稳定性的不确定性增加，同时气动稳定性对于加工误差更加敏感。总压损失系数、静压系数和失速因子互相单调相关。前缘和吸力面型线的加工误差对损失影响较大，相应几何精度应加以重点关注。

Abstract:

To alleviate the ‘dimension disaster’ in the process of evaluating the impact of manufacturing errors， combining the spanwise average assumption and the Karhunen-Loève expansion based on the Gaussian process， a reduced order model of normal manufacturing errors on three-dimensional blade surface geometric variability was proposed. The reduced order model on geometric variability was solved by the standard deviation function of the given manufacturing error distribution， the Pseudo-Monte Carlo method was used to randomly generate samples， and the Artificial Neural Network was trained to predict the impact of manufacturing error on the aerodynamic performance and corner aerodynamic stability of a high-load linear cascade. The results show that the effects of manufacturing errors are related to the working conditions. Under the near stall conditions， compared with the nominal， the mean aerodynamic performance and stability decrease， the probability that the total pressure loss coefficient increases is about 83.29%， the probability that the static pressure coefficient decreases is about 79.20%， and the probability that the stall indicator increases is about 69.10%. Compared with the design conditions， the variability of aerodynamic stability under the near stall conditions increases， and the aerodynamic stability is more sensitive to manufacturing errors. The total pressure loss coefficient， static pressure coefficient and stall indicator are monotonously related to each other. The manufacturing error of the leading edge and suction surface profile has a greater impact on the loss， and the corresponding geometric accuracy should be further focused on.