基于多尺度样本熵与PCA-FCM的滚动轴承故障诊断  

Rolling Bearing Fault Diagnosis Method Based on Multiscale Sample Entropy and PCA-FCM

作　　者：许凡[1] 方彦军[1] 张荣[1] 

机构地区：[1]武汉大学自动化系,湖北武汉430072

出　　处：《青岛科技大学学报（自然科学版）》2017年第2期100-106,111,共8页Journal of Qingdao University of Science and Technology:Natural Science Edition

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多尺度样本熵特征向量维数高及其维度难以确定问题,提出了一种基于多尺度样本熵的主成分分析的模糊聚类故障识别模型。该模型首先使用多尺度样本熵方法提取滚动轴承正常、内圈故障、外圈故障、滚动体故障的振动信号特征。其次对多尺度样本熵特征向量使用主成分分析方法进行降维。然后通过累积贡献率来确定其特征向量的维度,并利用选定的特征向量属性作为模糊C均值聚类模型的输入并进行故障识别。最后通过分类系数和分类熵这两个聚类评价指标进行聚类效果的检验。实验结果表明该模型能较好的区分滚动轴承的正常与内圈故障、外圈故障、滚动体故障这4种信号。To solve the problems of multiscale sample entropy eigenvectors with high-dimension and difficult to determine its dimensions in rolling bearing fault diagnosis,a combination method based on multiscale sample entropy,principal component analysis and fuzzy c-means cluster for rolling bearing fault diagnosis was proposed.Firstly,the multiscale sample entropy was used to extract the features of rolling bearing vibration signals under normal,inner race fault,outter race fault and ball fault conditions.Secondly,the principal component analysis method was selected to reduce the dimension of multiscale sample entropy eigenvectors.Then the dimension of eigenvectors which determinded by the cumulative contribution rate were choosed as the input of fuzzy C-means cluster method for fault recognition.Finally,using the partition coefficient and classification entropy these two indexs to assess the effect of fuzzy c-means cluster model.The experimental results shows that proposed method can bet-ter distinguish the rolling bearing vibration signals which contains normal,inner race fault,outter race fault and ball fault these four signals.

关 键 词：多尺度样本熵 主成分分析 模糊C均值 滚动轴承 故障诊断 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]