基于邻域粗糙集概念的一种滚动轴承特征提取方法

A character extraction method feature of rolling-element bearing based on concept of neighborhood rough set

针对滚动轴承早期微弱故障识别率偏低的问题,提出一种基于EEMD、邻域粗糙集和模糊C均值聚类(FCM)算法相结合的滚动轴承特征提取方法.该方法将滚动轴承的原始信号进行EEMD分解得到若干个IMF分量,通过均方差和欧氏距离两个评价指标选取出敏感特征分量,构造原始特征数据集,对处理后的原始特征集属性进行NRS约简,剔除冗余属信息,最后将剩余属性的特征数据集作为模糊C均值聚类的输入,实现滚动轴承故障识别.为了对比本文方法对于滚动轴承的故障识别效果,分别添加了FCM、NRS-FCM和EEMD-FCM三种方法进行故障辨识,利用划分系数(PC)和划分熵(CE)对聚类结果进行评价与对比.通过实验表明:邻域粗糙集对于改进滚动轴承的故障识别效果十分明显,具有良好的应用前景.

Aimed at the problem in rolling-element bearing that the identification rate of its early weak fault is lower,a method for character extraction of rolling-element bearing is proposed based on a combined algorithm of ensemble empirical mode decomposition(EEMD),neighborhood rough set(NRS)and fuzzy C-means clustering(FCMC).In this method,the original signals of the bearing are decomposed by means of EEMD and several IMF components are obtained.By analyzing the both assessment indices of mean square deviation and Euclidean distance,the sensitive characteristic components are selected and an original characteristic data set is constructed,then the NRS reduction and simplification is carried out for the attribute of the processed data set to eliminate the redundant information,and finally,the remainder of the attribute characteristic data set is put into the fuzzy C-means clustering to identify the fault in the bearing.In order to compare the fault identification result,a simple FCM,EEMD-FCM and EEMD-NRS-FCM are added else here,and a partition coefficient(PC)and a compartmentalization entropy(CE)are used to evaluate the clustering result.It is shown by experiment that the fault identification result with neighborhood rough set will be very obvious and have a good application prospect.