推进技术 ›› 2017, Vol. 38 ›› Issue (5): 1147-1154.

• 控制 测量 故障诊断 • 上一篇    下一篇

基于监督流形学习的航空发动机振动故障诊断方法

张 赟1,杨 栋2,斯彦刚2,方旭萌2   

  1. 海军航空工程学院 飞行器工程系,山东 烟台 264001,中国人民解放军 92074部队,浙江 宁波 315000,中国人民解放军 92074部队,浙江 宁波 315000,中国人民解放军 92074部队,浙江 宁波 315000
  • 发布日期:2021-08-15
  • 作者简介:张 赟,男,博士,讲师,研究领域为航空发动机测试与故障诊断。
  • 基金资助:
    国家自然科学基金(51505492);山东省自然科学基金(ZR2013EEQ001);“泰山学者”建设工程专项经费资助。

Aero-Engine Vibration Fault Diagnosis Based on Supervised Manifold Learning

  1. Department of Airborne Vehicle Engineering,Naval Aeronautical Engineering Institute,Yantai 264001,China,92074 unit of PLA,Ningbo 315000,China,92074 unit of PLA,Ningbo 315000,China and 92074 unit of PLA,Ningbo 315000,China
  • Published:2021-08-15

摘要: 航空发动机故障诊断中一个有挑战性的难题就是如何处理具有高维数、非线性化特点的故障数据,传统模式识别方法很难发现这类数据集的真实结构,导致故障诊断准确性不高。针对这一问题,将一种新兴的非线性维数约简技术——流形学习引入航空发动机振动故障诊断,提出基于监督流形学习理论的航空发动机特征提取与识别方法。该方法首先采用最近兴起的监督局部线性嵌入流形学习算法对蕴含在高维振动故障数据中不同故障的流形特征进行学习,映射到低维嵌入空间以实现故障的特征提取,在降维后的流形特征空间中构造分类器实现故障识别。利用航空发动机转子故障数据对方法的有效性进行了验证,结果表明,该方法显著提高了故障诊断性能,克服了传统的模式识别方法PCA和LDA的不足,并且在训练样本数为每类100的条件下,该方法的平均故障诊断正确率比PCA和LDA分别高出2.93%和7.20%。

关键词: 航空发动机;振动信号;监督流形学习;故障诊断

Abstract: How to deal with the high-dimensional and nonlinear data is a challenging problem for aero-engine fault diagnosis. The conventional pattern recognition methods usually fail to discover the underlying structure of such data sets,leading to low accuracy of fault diagnosis. Thus by introducing the new nonlinear dimensionality reduction technique into aero-engine vibration fault diagnosis,an aero-engine fault feature extraction and recognition approach based on manifold learning is proposed. The approach firstly performs the recently proposed manifold learning algorithm supervised locally linear embedding on the high-dimensional fault signal samples to learn the intrinsic embedded multiple manifold features for achieving fault feature extraction. And the classifier is constructed for fault recognition in the reduced manifold feature space. The fault rotor data of aero-engine is employed to validate the proposed approach. The experiment results show that the proposed approach obviously improves the fault classification performance and outperform the other traditional approaches such as Principal Component Analysis(PCA)and Linear Discriminate Method(LDA). Comparing to PCA and LDA,the average diagnosis accuracies of the approach increase by 2.93% and 7.20% respectively as the training sample size is 100 per class.

Key words: Aeroengine;Vibration signal;Supervised manifold learning;Fault diagnosis