采用监督局部切空间排列算法的航空发动机磨损故障诊断  

Aero-Engine Wear Fault Diagnosis with Supervised Locally Tangent Space Alignment

作　　者：张赟 林学森 王琳 陈应付 李朋 ZHANG Yun;LIN Xuesen;WANG Lin;CHEN Yingfu;LI Peng(Aeronautical Basic Institute,Naval Aeronautical University,Yantai,Shangdong 264001,China;77120 Unit of PLA,Chengdu 610000,China)

机构地区：[1]海军航空大学航空基础学院,山东烟台264001 [2]中国人民解放军77120部队,成都610000

出　　处：《西安交通大学学报》2020年第4期179-185,共7页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(51505492);山东“泰山学者”建设工程专项经费资助项目。

摘　　要：为解决传统特征提取技术难以处理具有非线性结构的复杂故障数据、影响故障诊断准确性的问题,将非线性维数约简技术——局部切空间排列引入航空发动机滑油光谱数据特征提取中,提出了一种基于监督局部切空间排列的发动机磨损故障诊断方法。该方法对非线性分布故障流形数据的内在几何特征进行捕捉,并将数据向低维故障特征空间进行非线性映射,完成故障特征的提取,最后在故障特征空间里构造分类器,完成磨损故障的识别诊断。采用某型发动机磨损故障滑油光谱数据开展实验,结果表明:与传统主元分析、线性鉴别分析特征提取方法相比,该方法能够更有效地提取出嵌入于故障数据中的非线性特征,提高了故障分类的准确率,并且只需采用简单的线性分类器就能具有很好的故障诊断性能。To solve the problem that the traditional feature extraction technique is hard to deal with the complex fault data containing nonlinear structure,an aero-engine fault diagnosis method based on supervised locally tangent space alignment is proposed by introducing the non-linear dimensionality reduction method named locally tangent space alignment algorithm.The method learns the intrinsic geometric features of fault manifold data,and non-linearly maps them into a low-dimensional fault feature space to achieve fault feature extraction.The wear fault recognition and diagnosis are carried out in the fault feature space by constructing classifier.The oil spectra data of engine wear fault are used for experiment.Compared with the conventional feature extraction approaches such as PCA and LDA,the proposed approach can effectively extract the nonlinear features embedded in the fault data to improve the fault classification accuracy and provide good fault diagnosis performance with a simple linear classifier.

关 键 词：局部切空间排列 非线性特征提取 航空发动机 磨损故障诊断 

分 类 号：V233.7[航空宇航科学与技术—航空宇航推进理论与工程]