机械故障的稀疏流形聚类与嵌入诊断方法  

The Method of Machinery Fault Detection using Sparse Manifold Clustering and Embedding

作　　者：王江萍[1] 段腾飞[1] 

机构地区：[1]西安石油大学机械工程学院,西安710065

出　　处：《机械科学与技术》2017年第10期1582-1588,共7页Mechanical Science and Technology for Aerospace Engineering

基　　金：西安石油大学全日制硕士研究生优秀学位论文培育项目(2015YP140407)资助

摘　　要：传统流形学习算法中邻域尺寸是固定的,在故障诊断中并不恰当。本文中提出了一种基于新型流形学习算法稀疏流形聚类与嵌入(SMCE)的机械故障诊断方法来解决这个问题。SMCE通过求解稀疏优化问题自动确定邻域的大小,将传统流形学习中邻域尺寸选择变为优化问题的惩罚系数选择,进而从高维非线性观测数据中提取流形结构。利用SMCE从轴承和齿轮振动信号中提取特征进行诊断,实验表明,所提方法可以较好的提取故障信号内在的几何结构,应用无监督的谱聚类和有监督的支持向量机进行诊断准确率均高于98%。The number of neighbors in traditional manifold learning is fixed,but it is unbefitting in fault detection. In order to solve this problem,a new fault detection method using sparse manifold clustering and embedding（ SMCE） is presented in this paper. SMCE extracts the low dimension manifold structure from observation space with high dimension and nonlinearity by solving sparse optimization problem and finding neighbors automatically. The parameter selection of neighbors is converted to select penalty coefficient of sparse optimization problem. By means of SMCE,the vibration feature of bearing and gear is extracted and supervised and unsupervised fault detection is achieved. The experiment illustrates that this method is better to extract the internal structure of fault signal and to detect mechanical fault. The detection accuracy with spectral clustering and SVM is higher than 98 %.

关 键 词：稀疏流形聚类与嵌入 流形学习 故障诊断 

分 类 号：TH17[机械工程—机械制造及自动化]