基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法  

Fault diagnosis method for rolling bearings based on short-time Fourier transform and convolution neural network

作　　者：李恒 张氢[1] 秦仙蓉[1] 孙远韬[1] LI Heng;ZHANG Qin;QIN Xianrong;SUN Yuantao(School of Mechanical Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学机械与能源工程学院,上海201804

出　　处：《振动与冲击》2018年第19期124-131,共8页Journal of Vibration and Shock

基　　金：国家科技支撑计划项目(2014BAF08B05; 2015BAF06B05)

摘　　要：针对具有较强非平稳性和易被强烈背景噪声干扰特点的滚动轴承振动信号,提出了基于短时傅里叶变换和卷积神经网络的故障诊断方法,实现了端到端的故障模式识别。首先,对滚动轴承振动信号进行短时傅里叶变换,得到时频谱样本,分为训练集和测试集;然后将训练集输入卷积神经网络中进行学习,不断更新网络参数;最后,将学习好参数的卷积神经网络模型应用于测试集,输出故障识别结果;通过滚动轴承故障模拟试验,进行可行性和有效性的验证。结果表明提出的方法对不同类型故障有着很高的识别精度,并可以通过增加故障数据种类和数量的方式来提高此方法的鲁棒性,是一种适应于处理"大数据"的故障诊断方法。Aiming at fault vibration signals of rolling bearings with features of stronger non-stationarity and easy to be disturbed by strong background noise,a fault diagnosis method based on the short-time Fourier transform(STFT)and the convolution neural network(CNN)was proposed to realize the end-to-end fault pattern recognition.Firstly,STFT was performed for vibration signals of rolling bearings to get their time-frequency spectrum samples divided into a training set and a test one.Then,the training set was input into CNN to do learning and update parameters of CNN.Finally,the CNN model with updated parameters was applied in the test set to output the fault recognition results.Through simulation tests of rolling bearing faults,the feasibility and effectiveness of the proposed method were verified.The results indicated that the proposed method has higher recognition accuracies for different types of faults;the robustness of this method can be improved with increase in amount and type of fault data;it is a fault diagnosis method suitable for dealing with big data.

关 键 词：滚动轴承 故障诊断 短时傅里叶变换 卷积神经网络 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3