基于精细复合多元多尺度加权排列熵与流形学习的滚动轴承故障诊断

Fault Diagnosis for Rolling Bearings Based on RCMMWPE and Manifold Learning

为充分利用多通道振动信息,将表征同步多通道数据多变量复杂度的多元多尺度熵理论引入滚动轴承故障诊断。同时,针对排列熵未考虑时间序列的振幅信息以及多尺度过程中粗粒化方式存在不足等缺陷,提出了精细复合多元多尺度加权排列熵并将其用于提取滚动轴承的故障特征。随后,利用t-SNE对特征进行二次提取,寻找相关性较大的特征组成候选样本,结合随机森林分类器进行故障分类。2种滚动轴承试验数据的分析表明,该方法能够提取高质量的滚动轴承故障特征,获得较高的分类准确率。

In order to make full use of multi-channel vibration information,the multivariate multi-scale entropy theory,which characterizes the multivariate complexity of synchronized multi-channel data,is introduced into fault diagnosis for rolling bearings.At the same time,in view of defects of permutation entropy(PE),such as not considering the amplitude information of time series and insufficiency of coarse grained method during multi-scale analysis process,a refined composite multivariate multi-scale weighted permutation entropy is proposed and used to extract the fault features of rolling bearings.Then,the t-SNE is used to extract the features twice,and the features with greater correlation are found to compose of candidate samples.The random forest classifier is used to classify the faults.The analysis of two kinds of rolling bearing test data shows that the method can extract high quality rolling bearing fault features and obtain high classification accuracy.