基于EMD与KPCA的滚动轴承故障特征提取及诊断方法研究  

Study on Feature Extraction and Diagnosis Method of Rolling Bearing Faults Based on EMD and KPCA

作　　者：徐卓飞[1] 张海燕[1] 王丹[1] 张明龙[1] 

机构地区：[1]西安理工大学,西安710048

出　　处：《机械科学与技术》2014年第10期1518-1524,共7页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(51275406;51305340);陕西省自然科学基础研究计划项目(2013JM7009);陕西省教育厅科学研究计划项目(2013JK1030)资助

摘　　要：针对滚动轴承故障种类识别与程度判断问题,提出了一种融合经验模式分解与核主元分析的故障诊断方法:首先,运用经验模式分解将滚动轴承故障信号分解成不同特征尺度下的本征模式分量,采用Hilbert-Huang变换对信号进行相应的时频分析,从本征模式分量函数和瞬时频率中分别提取时域和频域的统计特征集与无量纲特征集;其次,引入非线性核主元分析方法,对故障特征集进行处理,从而消除特征集中的冗余特征,并大幅度降低特征向量维数,得到能够反映故障本质的主元特征集;最后,构造支持向量机多类分类网络,实现了对不同故障模式与不同损伤程度滚动轴承的故障诊断。A new diagnosis mode decomposition （EMD） method for the fault classification and damage degree of rolling bearing based on empirical and kernel principal component analysis （KPCA） is proposed. Firstly, the fault signal of rolling bearing was decomposed into several intrinsic mode functions under different characteristic scales with EMD. Then Hilbert-Huang transform was employed to calculate the statistical features and dimensionless features in both time domain and frequency domain. Secondly, KPCA based on samples was introduced to eliminate redundant features and reduce the vector dimension greatly. Thus, the KPCA feature vector was acquired, which could reflect the failure. Furthermore, the classifier based on one-versus-one support vector machine was constructed. At last, the classification of three failure modes and different damage degree for rolling bearing was completed.

关 键 词：经验模式分解 核主元分析 轴承 故障诊断 损伤程度 

分 类 号：TH117[机械工程—机械设计及理论] TP183[自动化与计算机技术—控制理论与控制工程]