时频特征降维和多层次聚类相结合的轴承故障诊断新方法  

A New Method of Bearing Fault Diagnosis Based on the Combination of Time-frequency Feature Reduction and Multi-level Clustering

作　　者：柳霞 蒋淑霞[1] 张长伟 何泽江 刘文 LIU Xia;JIANG Shuxia;ZHANG Changwei;HE Zejiang;LIU Wen(School of Mechanical and Electrical Engineering,Central South University of Forestry Technology,Changsha 410004,China;SAGW,Shanghai 201800,China)

机构地区：[1]中南林业科技大学机电工程学院,长沙410004 [2]上海汽车变速器有限公司,上海201800

出　　处：《噪声与振动控制》2023年第6期123-128,共6页Noise and Vibration Control

基　　金：湖南省自然科学基金资助项目(2019JJ60076)。

摘　　要：从含噪信号中判断滚动轴承是否发生故障,同时确定故障发生位置和缺陷程度,针对这一轴承工作状态监测的核心问题,提出一种结合小波频带剖分、主成分分析、多层次聚类的滚动轴承故障诊断方法。首先,利用小波最优分解层数将获取的原始信号进行小波包分解得到小波能量谱;然后另取18个时域频域特征指标共同构成特征参数集,再经主成分分析处理,将参数集降到合适的维数;最后利用最佳类间距、样本间距组合和聚类有效性评价建立多层次聚类挖掘系统。实验和应用案例表明:该方法能够准确有效地诊断滚动轴承的不同故障类型和损伤程度,准确率达99.3%,可为轴承状态监测与智能故障诊断提供有效的理论参考。Judging whether the rolling bearing has fault according to the noisy signal,and determining the fault location and defect degree are the key problem of machine condition monitoring.Aiming at this problem,a rolling bearing fault identification method based on the combination of wavelet frequency band subdivision,principal component analysis and multi-level clustering is proposed.Firstly,the wavelet energy spectrum is obtained by wavelet packet decomposition of the original signal based on the optimal number of wavelet decomposition layers.Then,another 18 time-domain and frequency-domain characteristic indexes are taken to form the characteristic parameter set.And the parameter set is reduced to an appropriate dimension by principal component analysis.Finally,using the optimal class spacing,sample spacing combination and clustering effectiveness evaluation,a multi-level clustering mining system is established.Experiments and application cases show that this method can accurately and effectively identify different types of failures and damage levels of rolling bearings with an accuracy rate of 99.3%,which provides an effective theoretical reference for bearing condition monitoring and intelligent fault diagnosis.

关 键 词：故障诊断 小波频带剖分 多层次聚类系统 聚类有效性 状态评估 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置] TH133.33[自动化与计算机技术—控制科学与工程]