基于全矢-CNN的轴承故障诊断研究  

Bearing Fault Diagnosis Based on Full Vector-CNN

作　　者：郝旺身[1] 陈耀[1] 孙浩[1] 付耀琨 李伟[1] HAO Wangshen;CHEN Yao;SUN Hao;FU Yaokun;LI Wei(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China;Navigation Bureau of Henan Communications and Transportation Department,Zhengzhou 450016,China)

机构地区：[1]郑州大学机械与动力工程学院,河南郑州450001 [2]河南省交通运输厅航务局,河南郑州450016

出　　处：《郑州大学学报（工学版）》2020年第5期92-96,共5页Journal of Zhengzhou University（Engineering Science）

基　　金：国家重点研发计划项目(2016YFF0203100)。

摘　　要：针对传统智能故障诊断系统需要大量先验知识,以及模型复杂度高和单通道信号不完整造成信息遗漏的问题,将全矢谱技术与卷积神经网络(CNN)结合,提出一种新的滚动轴承的故障诊断模型。该方法将全矢谱技术与深度卷积神经网络结合,相比于单通道数据建立的模型而言,具有特征信息完整、模型适应性强等优点。首先利用全矢谱技术对采集的双通道信号进行信息融合,得到融合后的主振矢数据。然后结合主振矢数据与深度学习算法构建全矢深度卷积神经网络,模型能够自适应地提取故障特征,利用反向传播算法调节优化模型参数。实验结果表明:该方法能够提取更加完整的轴承故障信息,该模型具有更高的准确率和更好的稳定性。Aimed to improved at the traditional intelligent fault diagnosis system,which required a large amount of prior knowledge,and had the complexity of its model and the information loss caused by the incompleteness of single-channel signal,a full-vector deep convolutional neural network diagnosis model of rolling bearing was proposed.The full-vector technique was used to fuse the acquired two-channel signals to obtain the fused main vibration vector data,which contained more complete information than the single-channel data.Combining the main vibration vector and deep learning algorithm to construct the full vector depth convolutional neural network,the model could adaptively extract the fault features and use the back propagation algorithm to adjust the model parameters.The experimental results showed that the method could extract more complete fault information,and the model had higher accuracy and better stability.

关 键 词：故障诊断 全矢谱 深度学习 卷积神经网络 滚动轴承 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3