基于IMF奇异值熵和t-SNE的滚动轴承故障识别  

Rolling bearing fault identification based on IMF singular value entropy and t-SNE

作　　者：段萍[1] 王旭[1] 丁承君[1] 冯玉伯 秦越 DUAN Ping;WANG Xu;DING Chengjun;FENG Yubo;QIN Yue(College of Mechanical Engineering,Hebei University of Technology,Tianjin 300130,China)

机构地区：[1]河北工业大学机械工程学院,天津300130

出　　处：《传感器与微系统》2021年第3期134-137,共4页Transducer and Microsystem Technologies

基　　金：河北省科技计划资助项目(14214902D)。

摘　　要：针对滚动轴承振动信号非线性、非平稳性以及故障难以识别的问题,提出了一种经验小波变换(EWT)、奇异值熵和t分布随机领域嵌入(t-SNE)相结合的滚动轴承故障识别方法。对原始振动信号进行EWT分解得到若干固有模态分量(IMF),对IMF进行奇异值分解求取奇异值熵。利用t-SNE算法对奇异值熵组成的特征矩阵进行降维,所提取的低维特征能够有效反映故障信息。最后,将低维特征输入到Kmeans分类器中进行模式识别。将该方法应用到滚动轴承实验中并与EMD+奇异值熵+t-SNE、EWT+奇异值熵+PCA方法进行对比,结果表明:所提方法能够更有效地提取滚动轴承的故障特征,提高了故障识别的精度。Aiming at the problems of nonlinearity,non-stationarity and difficult identification of rolling bearing vibration signals,a fault identification method for rolling bearings based on empirical wavelet transform(EWT),singular value entropy and t-distributed stochastic neighbor embedding(t-SNE)is proposed.The method performs EWT decomposition on the original vibration signal to obtain several intrinsic modal function(IMF),and performs singular value decomposition on the IMF to obtain singular value entropy.The t-SNE algorithm is used to reduce the feature matrix composed of singular value entropy,and the extracted low-dimensional features can effectively reflect the fault information.Finally,low-dimensional features are input into the K-means classifier for pattern recognition.The method is applied to rolling bearing experiments and compared with EMD+singular value entropy+t-SNE,EWT+singular value entropy+PCA method.The results show that the method can extract the fault characteristics of rolling bearings more effectively and improve the accuracy of fault identification.

关 键 词：经验小波变换 奇异值熵 t分布随机领域嵌入 故障识别 

分 类 号：TH133.33[机械工程—机械制造及自动化]