采用多通道样本和深度卷积神经网络的轴承故障诊断方法  

Bearing Fault Diagnosis with Multi-Channel Sample and Deep Convolutional Neural Network

作　　者：张弘斌 袁奇[1,2] 赵柄锡 牛广硕 ZHANG Hongbin;YUAN Qi;ZHAO Bingxi;NIU Guangshuo(School of Energy and Power Engineering, Xi’an Jiaotong University, Xi’an 710049, China;Shaanxi Engineering Laboratory of Turbomachinery and Power Equipment, Xi’an Jiaotong University, Xi’an 710049, China)

机构地区：[1]西安交通大学能源与动力工程学院,西安710049 [2]陕西省叶轮机械及动力装备工程实验室,西安710049

出　　处：《西安交通大学学报》2020年第8期58-66,共9页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(11872289,11902237);西安交通大学自然科学基金重点资助项目(ZRZD2017025)。

摘　　要：提出了一种新的多通道样本构造方法,结合深度卷积神经网络来提高轴承故障诊断的效果。首先采用连续小波变换,分别提取了转子两端轴承振动信号的时频域特征,基于所得结果分别构造了针对两端轴承的单通道二维图形样本,并取上述两类单通道样本的均值构造了第3类单通道样本;将得到的3类单通道样本融合,得到用于故障诊断的三通道样本;建立不同结构的深度卷积神经网络,分别采用单通道样本和三通道样本对滚动轴承故障类型和故障严重程度进行诊断,并将所得结果进行对比分析。结果表明:在多种不同网络结构下,基于三通道样本的轴承故障诊断准确率均明显优于单通道样本,证明了提出的多通道样本构造方法在轴承故障诊断中有着更好的效果,可以为轴承故障诊断方法和样本构建提供参考。A new multi-channel sample construction method combing with deep convolutional neural network is proposed to improve the effectiveness of bearing fault diagnosis.The time-frequency domain features of vibration signals from bearings at both ends of the rotor are extracted by continuous wavelet transform,and two types of signal-channel samples are constructed based on the obtained results.Then the third type of single-channel sample is produced by performing average operation on the two former types of single-channel samples.Furthermore,the obtained three types of single-channel samples are combined to obtain the three-channel sample adopted for fault diagnosis.The deep convolutional neural networks of different structures are established with the three types of single-channel samples and the three-channel samples respectively,the diagnosis is thus carried out on the rolling bearing with respect to different fault types and severity levels.A comparison between the diagnosis performances of single-channel samples and multi-channel samples indicates that the diagnosis accuracy of three-channel samples is obviously better than that of single-channel samples with different network structures.This approach provides a reference for bearing fault diagnosis in terms of sample construction.

关 键 词：轴承故障诊断 三通道样本 深度卷积神经网络 连续小波变换 

分 类 号：TH17[机械工程—机械制造及自动化] TH133