改进经验小波变换和改进字典学习在轴承故障识别中的应用  

Application of Enhanced EWT and Enhanced Dictionary Learning in Bearing Faults Identification

作　　者：武彩霞[1] 李帆 刘育博 Wu Caixia;Li Fan;Liu Yubo(School of Intelligent Manufacturing,Wuxi Vocational College of Science and Technology,Wuxi 214028,China;School of Mechanical Engineering,Xi´an Jiaotong University,Xi´an 710049,China;Joint Transmission and Bearing Technology Research Center,Shizuishan 753000,China)

机构地区：[1]无锡科技职业学院智能制造学院,江苏无锡214028 [2]西安交通大学机械工程学院,陕西西安710049 [3]联合传动及轴承技术研究中心,宁夏石嘴山753000

出　　处：《机械传动》2023年第1期138-146,共9页Journal of Mechanical Transmission

基　　金：国家自然科学基金(51575421)江苏省教育厅高等教育教改研究课题(2021JSJG411)。

摘　　要：通过深度学习进行滚动轴承故障识别时,存在因信号噪声导致故障识别率较低和深层网络收敛速度慢的问题。针对上述问题,提出了一种改进经验小波变换(EEWT)和改进字典学习(EDL)的轴承故障识别方法。首先,将轴承振动信号进行包络谱变换,通过包络谱的极值点与自适应阈值的关系进行包络谱边界自动划分,进而利用经验小波变换(EWT)将信号自动分解为调幅-调频(AM-FM)分量;其次,提出一种新的AM-FM分量筛选指标,利用筛选指标选取合适的AM-FM分量进行重构,进而对信号进行有效降噪;最后,利用稀疏性约束逐层学习降噪后轴承故障样本中的典型结构特征,并构造深层故障字典(DFD),将故障样本输入DFD中,根据样本的重建误差确定故障类别。试验结果表明,该方法对噪声的鲁棒性高,故障识别能力优于其他模型,而且该方法可利用驱动字典自动提取轴承振动信号样本中的故障特征;同时,EDL结构使所提取的故障特征具有较好的层次性,符合人对故障的直观认识,可用于滚动轴承故障识别工程中。When realizing rolling bearing fault identification through deep learning, there is a low recognition rate and convergence rate due to ambient noise. Aiming at the above problem, a fault identification model based on enhanced empirical wavelet transform(EEWT) and enhanced dictionary learning(EDL) is proposed.Firstly, the vibration signals of rolling bearing are transformed by envelope spectrum, and envelope spectrum adaptive segmentation is implemented through the relationship between the envelope point and the adaptive threshold, and signals are decomposed into several amplitude modulation-frequency modulation(AM-FM) components. Secondly, a new component screening index is proposed, and then the appropriate AM-FM components are reconstructed to effectively reduce the noise of signals. Finally, the sparsity constraint is used to learn the typical structural characteristics in the bearing fault sample layer by layer, and the deep fault dictionary(DFD) is constructed. Then the fault samples are fed into the DFD to determine the fault category according to the reconstruction error of the samples. The test results show that the proposed method is robust to noise and has better fault recognition ability than other models. And the proposed method utilizes the sparse constraint driving dictionary to automatically extract the fault features in the vibration signal samples, while the EDL structure makes the extracted fault features have better hierarchical and physical meaning, which is in line with people′s intuitive understanding of the fault and can be used in the rolling bearing fault identification engineering.

关 键 词：滚动轴承 故障识别 经验小波变换 字典学习 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP18[自动化与计算机技术—控制理论与控制工程]