基于独立特征选择与流形学习的故障诊断  

Fault diagnosis based on individual feature selection and manifold learning

作　　者：杜伟 房立清 齐子元 DU Wei;FANG Liqing;QI Ziyuan(Department of Artillery engineering,The Army Engineering University of PLA,Shijiazhuang 050003,China)

机构地区：[1]陆军工程大学火炮工程系,石家庄050003

出　　处：《振动与冲击》2018年第16期77-82,123,共7页Journal of Vibration and Shock

基　　金：河北省自然科学基金项目(E2016506003)

摘　　要：为了有效利用特征集所包含的敏感特征进行故障诊断,提出基于独立特征选择(Individual Feature Selection,IFS)与流形学习的故障诊断方法。从多个角度提取振动信号的混合特征,构建原始高维特征集。采用一种改进的核Fisher特征选择方法为每两类故障状态独立选择敏感特征集,并通过线性局部切空间排列(Linear Local Tangent Space Alignment,LLTSA)算法挖掘出可区分度更高的特征子集。采用"一对一"的方法训练多个二类分类支持向量机(Supported Vector Machine,SVM),将得到的低维特征输入多分类故障诊断模型进行识别。液压泵故障诊断实验表明,所提方法具备较高的诊断准确率。In order to diagnose fault effectively by using sensitive features contained in a feature set,a fault diagnosis method based on individual feature selection(IFS)and manifold learning was proposed.Firstly,the mixed feature of the vibration signal was extracted from multiple domains,and the original high-dimensional feature set was constructed.Then,an improved kernel Fisher feature selection method was proposed and used to select individual sensitive feature subset for each pair of classes,and the mining performance of the feature subset with higher distinguishability was further implemented by using Linear local tangent space alignment(LLTSA).Finally,a one-against-one approach was applied to train several SVM binary classifiers,and low-dimensional feature was input into the multi-class fault diagnosis model for recognizing the fault types.The experimental results of hydraulic pump indicate that the proposed method is of high diagnostic accuracy.

关 键 词：故障诊断 独立特征选择 核FISHER判别分析 流形学习 

分 类 号：TH165.3[机械工程—机械制造及自动化] TN911.7[电子电信—通信与信息系统]