四川理工学院学报（自然科学版）2018，Vol.31Issue(3)：54-59,6.DOI:10.11863/j.suse.2018.03.09
基于小波包与CNN的滚动轴承故障诊断
Roller Bearing Fault Diagnosis Based on Wavelet Packet and CNN
许理 1李戈 2余亮 1姚毅1
作者信息
- 1. 四川理工学院自动化与信息工程学院,,四川 自贡 643000
- 2. 四川理工学院物理与电子工程学院,四川 自贡 643000
- 折叠
摘要
Abstract
The vibration signal of rolling bearing has strong non-stationarity,and wavelet packet (WP)time-frequency analysis method can effectively extract the time-frequency characteristics of non-stationary signals and have fine time-frequency resolution.The convolutional neural network (CNN)has a strong feature learning ability which makes it better than the fault recognition rate of shallow network.In order to diagnose the running state of rolling bearing more accurately,a fault diagnosis method of rolling bearing based on wavelet packet and CNN is proposed.The time-frequency analysis of the collected bearing vibration signal iscarried out using wavelet packets to obtain the time-frequency characteristics of various types of signals.Fine-tune technology is used to fine-tuning the caffeNet model to solve the problem of training a small number of samples to train the CNN model.Finally,a CNN model that can be used for rolling bearing fault diagnosis is obtained.The experimental verification shows that the fault diagnosis is achieved using the combination of wavelet packet and CNN,and the fault recognition rate reaches 99. 1%,which is higher than that of continuous wavelet transform (CWT)and short-time Fou-rier transform (STFT)combined with CNN.The combination of principal component analysis (PCA)and support vector machine (SVM)has the lowest fault recognition rate,and the recognition effect of composite faults is obviously insufficient.关键词
滚动轴承/小波包/卷积神经网络/故障诊断/fine-tuning技术Key words
rolling bearing/wavelet packet/convolutional neural network/fault diagnosis/fine-tuning technology分类
信息技术与安全科学引用本文复制引用
许理,李戈,余亮,姚毅..基于小波包与CNN的滚动轴承故障诊断[J].四川理工学院学报（自然科学版）,2018,31(3):54-59,6.
