基于局部与全局结构保持算法的滚动轴承故障诊断

发布时间：2018-09-01 09:29

【摘要】：为了精准稳定地提取滚动轴承故障特征,提出一种基于局部与全局结构保持算法的低维敏感特征提取方法,采用K近邻(K-nearest neighbor algorithm,KNN)分类算法进行故障识别。对滚动轴承振动信号的时域和频域特征进行特征提取,组成初始特征集,采用该方法对初始特征集进行特征提取,提取过程中综合考虑初始特征集的局部结构特征和全局结构特征以避免其结构信息的丢失,同时引入正交约束减小主特征分量间的信息冗余,提取出最优表征初始特征集特征的敏感特征矢量,并通过K近邻分类算法对其进行分类。将该方法应用于滚动轴承故障诊断,通过与其他几种典型特征提取方法对比,该方法能更有效地提取滚动轴承四种状态的敏感特征矢量,在故障诊断中表现出更好的分类性能,整体故障识别率保持为100%。因此,该方法能有效提取敏感故障特征,为实际滚动轴承智能故障诊断提供参考。
[Abstract]:In order to extract the fault feature of rolling bearing accurately and stably, a low-dimensional sensitive feature extraction method based on local and global structure preserving algorithm is proposed, and K-nearest neighbor (K-nearest neighbor algorithm,KNN) classification algorithm is used for fault identification. The feature extraction of rolling bearing vibration signal in time domain and frequency domain is carried out, and the initial feature set is formed. The method is used to extract the initial feature set. In order to avoid the loss of structural information, the local and global structural features of the initial feature set are considered synthetically in the extraction process, and orthogonal constraints are introduced to reduce the information redundancy among the principal feature components. The sensitive feature vectors which best characterize the initial feature set are extracted and classified by K-nearest neighbor classification algorithm. This method is applied to the fault diagnosis of rolling bearings. Compared with other typical feature extraction methods, the method can extract the sensitive feature vectors of the four states of rolling bearings more effectively, and show better classification performance in fault diagnosis. The overall fault recognition rate is maintained at 100. Therefore, this method can effectively extract sensitive fault features and provide a reference for intelligent fault diagnosis of rolling bearings.
【作者单位】：新疆大学电气工程学院;清华大学自动化系;
【基金】：国家自然科学基金资助项目(51575469)
【分类号】：TH133.33

【参考文献】