EEMD排列熵与PCA-GK的滚动轴承聚类故障诊断  

Rolling Bearing Clustering Faults Diagnosis Based on EEMD,Permutation Entropy and PCA-GK

作　　者：黄友朋 赵山 许凡[2] 方彦军[2] 

机构地区：[1]广东电网有限责任公司电力科学研究院,广东广州510080 [2]武汉大学自动化系,湖北武汉430072

出　　处：《河南科技大学学报（自然科学版）》2017年第2期17-24,30,共9页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(61201168)

摘　　要：针对滚动轴承故障诊断中,用振动信号的总体经验模式分解(EEMD)方法分解后的熵特征向量维数高,且样本熵(SE)计算效率差等问题,提出了一种基于EEMD排列熵(PE)的主成分分析(PCA)-GK滚动轴承聚类故障诊断组合方法。首先,使用EEMD方法将信号分解为若干个固有模态函数(IMFs),使用PE/SE计算其IMFs熵值;然后,使用PCA对熵特征向量进行可视化降维,并作为模糊C均值(FCM)与GK聚类算法的输入,实现对滚动轴承的故障诊断。利用分类系数和平均模糊熵,对聚类结果进行了评价与对比,实验结果表明:本文模型(EEMD-PE-PCA-GK)的聚类效果比其他3种模型(EEMD-SE-PCA-FCM、EEMD-SE-PCA-GK和EEMDPE-PCA-FCM)更好,且PE比SE的计算效率更快。Aiming at the problem that the dimension of entropy eigenvectors decomposed by ensemble empirical mode decomposition( EEMD) of rolling bearing vibration signals was high,and the computational efficiency of sample entropy( SE) method was poor,a combination method based on EEMD,permutation entropy( PE),principal component analysis( PCA) and GK( Gustafson-Kessel) clustering algorithm was proposed for rolling bearings fault diagnosis. Firstly,the original signals were decomposed into a series intrinsic mode functions( IMFs) by using EEMD methods,then the PE and the SE methods were used to calculate the corresponding entropy values. Secondly,the dimension reduction and visualization of SE and PE eigenvectors which were regarded as the input of fuzzy C-mean( FCM) and GK for the rolling bearing fault recognition were achieved by PCA model. The partition coeffcient( PC) and classification entropy( CE) were used to verify and compare clustering method. The experiment results show that the method( EEMD-PE-PCA-GK) proposed in this paper is better than the other three models( EEMD-SE-PCA-FCM,EEMD-SE-PCA-GK,EEMD-PE-PCA-FCM),and the computational efficiency of PE is faster than SE method.

关 键 词：滚动轴承 故障诊断 总体经验模式分解 排列熵 GK聚类算法 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP206.3[自动化与计算机技术—检测技术与自动化装置]