基于主成分分析的多域特征融合轴承故障诊断  

Bearing Fault Diagnosis Based on Principal Component Analysis and Multi-domain Feature Fusion

作　　者：周凌孟 邓飞其[1] 张清华 孙国玺[2] 苏乃权[2] 朱冠华 ZHOU Lingmeng;DENG Feiqi;ZHANG Qinghua;SUN Guoxi;SU Naiquan;ZHU Guanhua(School of Automation Science and Engineering,South China University of Technology,Guangzhou Guangdong 510000,China;Guangdong Provincial Key Laboratory of Petrochemical Equipment Fault Diagnosis,Guangdong University of Petrochemical Technology,Maoming Guangdong 525000,China)

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510000 [2]广东石油化工学院,广东省石化装备故障诊断重点实验室,广东茂名525000

出　　处：《机床与液压》2024年第6期167-176,共10页Machine Tool & Hydraulics

基　　金：国家自然科学基金重点项目(6193000428);广东省自然科学基金面上项目(2022A1515010599);茂名市科技计划(170607111706145)。

摘　　要：针对复杂工况下难以区分轴承故障状态的问题,提出一种基于主成分分析的多域特征融合轴承故障诊断方法。采集轴承振动加速度信号,提取轴承时域新量纲一化特征、频域幅值谱特征和时频域经验模态分解特征共13维特征用于完整表征轴承状态;利用主成分分析方法对所提取特征融合与降维,降低诊断模型复杂度与数据分析难度;最后,选择合适的卷积神经网络进行分类,通过石化机组故障诊断实验平台进行验证。结果表明:多域融合特征相对于单域特征诊断效果更好,卷积神经网络分类模型相对于其他经典分类模型诊断准确率更高,融合诊断分类方法整体诊断准确率达到86%。To solve the problem that it is difficult to distinguish bearing fault state under complex working conditions,a bearing fault diagnosis method of multi-domain feature fusion based on principal component analysis was proposed.The vibration acceleration signals was collected,and the new dimensionless features in time domain,amplitude spectrum features in frequency domain and empirical mode decomposition features in time frequency domain were extracted to fully described the bearing state.The extracted features were fused and reduced in dimension by the principal component analysis method,it can effectively reduce the complexity of diagnostic models and the difficulty of data analysis.Finally,a suitable convolutional neural network was selected to classify,the verification was performed by the petrochemical unit fault diagnosis experimental platform.The results show that the multi-domain fusion feature diagnosis is better than the single domain feature diagnosis,the convolutional neural network classification model has higher diagnostic accuracy than other classical classification models,the diagnostic accuracy of the fusion diagnosis classification method reaches 86%.

关 键 词：轴承 特征融合 主成分分析方法 卷积神经网络 故障诊断 

分 类 号：TH133.3[机械工程—机械制造及自动化]