基于经验小波分解和卷积神经网络的液压泵故障诊断  

Fault Diagnosis of Hydraulic Pump Based on Empirical Wavelet Transform and Convolutional Neural Network

作　　者：杜名喆 王宝中 DU Ming-zhe;WANG Bao-zhong(School of Mechanical Engineering,North China University of Science and Technology,Tangshan,Hebei 063210)

机构地区：[1]华北理工大学机械工程学院

出　　处：《液压与气动》2020年第1期163-170,共8页Chinese Hydraulics & Pneumatics

摘　　要：液压泵振动信号具有非平稳性,需要一定的先验知识和专家经验从而实现故障诊断。为了实现对液压泵的智能故障诊断,提出了一种基于经验小波变换(Empirical Wavelet Transform,EWT)和卷积神经网络(Convolutional Neural Network,CNN)的智能故障诊断方法。首先对振动加速度信号进行经验小波变换得到若干不同模态的内禀模态函数(Intrinsic Mode Function,IMF);其次,基于峭度的评价指标,筛选出故障特征丰富的6个IMF分量作为诊断用的数据源;然后利用CNN融合IMF分量,并自动提取相关特征;最后,将特征应用于分类器识别,从而实现液压泵故障的自动故障诊断。试验结果表明:该方法能够准确、有效的对液压泵的工作状态和故障类型进行分类。As the vibration signal of the hydraulic pump is usually nonstationary,a significant level of prior knowledge and Expert experiences is required for fault diagnosis.In order to achieve intelligent diagnosis of the hydraulic pump,an intelligent fault diagnosis method based on empirical wavelet transform(EWT)and convolutional neural networks(CNN)is proposed.Firstly,EWT is used to decompose the vibration acceleration signal to obtain intrinsic mode function(IMF)components.Secondly,with a fusion evaluation method based on kurtosis,the six IMFs which are rich in fault features are selected as data source,and then the IMFs are fused through CNN and features are extracted automatically.Finally,the learned features serve as the input parameters of classifier to classify working condition,and the atomization of the hydraulic pump fault diagnosis can be implemented.The experimental results show that the method can classify the working state and fault type of the hydraulic pump accurately and effectively.

关 键 词：故障诊断 经验小波变换 卷积神经网络 液压泵 

分 类 号：TH137[机械工程—机械制造及自动化] TH17