基于最优特征集和马氏距离KNN分类的机械故障分类方法研究  

A Method of Mechanical Fault Classification Based on Optimal Feature Subset and K-Nearest Neighbor Using Mahalanobis Distance

作　　者：孟亚辉[1] MENG Ya-hui(Guangdong University of Petrochemical Technology, Guangdong Maoming 525000, Chin)

机构地区：[1]广东石油化工学院,广东茂名525000

出　　处：《机械设计与制造》2017年第7期98-102,共5页Machinery Design & Manufacture

基　　金：茂名市科技计划项目(2012B01066)

摘　　要：针对传统K近邻(K-Nearest Neighbor,KNN)算法在进行机械故障信号识别的过程中,无法挖掘特征参数之间关联性,提出一种基于最优特征集的马氏距离KNN分类方法,根据机械故障信号的非线性特点,使用小波分解获得时频域故障特征,通过局部嵌入算法(Locally Linear Embedding,LLE)来进行二次故障特征提取,从而获得多相关特征集并对其进行主成分分析得到最优特征集,最后通过数值仿真信号和齿轮故障数据的分析了方法的有效性。结果表明该方法能够有效挖掘特征参数之间关联性,增加不同故障之间区分度,从而提高故障识别精度。The traditional K-Nearest Neighbor （KNN） arithmetic can＇t find the correlation between characteristic parameters in the process of mechanical failure signals identification. Therefore, the KNN classification method based onMahalanobis distance and optimal feature subset was proposed in this paper. Due to the nonlinear characteristic of the mechanical failure signals, signals were decomposed to obtain the fault feature in time-frequency domain by wavelet transform. Then, Locally Linear Embedding （LLE） algorithm was utilized to obtain multi-related feature subset aimed to secondary extracting fault feature. And the aptimal feature subset was obtained by principal component analysis （ PCA ） for multi-related feature subset. Finally, simulation and experiment analysis results indicate that this method can find the correlation between characteristic parameters effectively and improve the identification accuracy.

关 键 词：故障诊断 KNN算法 马氏距离 局部嵌入算法 主成分分析 

分 类 号：TH16[机械工程—机械制造及自动化] TH17