正交小波变换k-中心点聚类算法在故障诊断中的应用  

Orthogonal wavelet transform KCA in fault diagnosis

作　　者：李卫鹏 曹岩[1,2] 李丽娟 LI Weipeng;CAO Yan;LI Lijuan(School of Mechanical and Electrical Engineering,Xi’an Technological University,Xi’an 710600,China;School of Intelligent Manufacturing,Nanyang Institute of Technology,Nanyang 473004,China)

机构地区：[1]西安工业大学机电工程学院,西安710600 [2]南阳理工学院智能制造学院,河南南阳473004

出　　处：《振动与冲击》2021年第7期291-296,共6页Journal of Vibration and Shock

基　　金：陕西省创新能力支撑计划项目(2018TD-036);陕西省重点研发计划项目(2019GY-125)。

摘　　要：k-中心点聚类算法(k-medoids cluster algorithm,KCA)是改进的机器学习聚类算法,该方法通过初始聚类中心选取和聚类中心更新,对无标记训练样本的学习揭示数据的内在性质及规律,从而区分出机器的运行状态。提出了一种正交小波变换k-中心点聚类算法(orthogonal wavelet transform k-medoids clustering algorithm,OWTKCA)诊断方法,利用正交小波变换(orthogonal wavelet transformation,OWT)方法提取各细节信号作为训练样本,用KCA方法进行分类。通过滚动轴承的试验数据分类结果显示,该方法相对于没有提取特征值的KCA能有效处理复杂机械振动信号,明显提高了故障数据聚类效果,缩短了聚类时间,提高了智能诊断效率。k-medoids cluster algorithm(KCA)is an improved machine learning clustering algorithm.This method reveals the inherent properties and laws of the data,through selecting the initial clustering center,updating the clustering center and learning the unmarked training samples,so as to distinguish the running state of the machine.In this paper,an orthogonal wavelet transform k-medoids clustering algorithm(OWTKCA)was proposed for diagnosis,which uses the orthogonal wavelet transform(OWT)method to extract the detailed signals as training samples,and uses the KCA method to classify them.The results of test data classification of rolling bearing show that this method can deal with complex mechanical vibration signals more effectively than KCA without extracting characteristic values,it obviously improves the clustering effect of fault data,shortens the clustering time and improves the efficiency of intelligent diagnosis.

关 键 词：k-中心点聚类算法(KCA) 机器学习 故障诊断 正交小波变换(OWT) 

分 类 号：TH165.3[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]