基于CNN并引入EMD预处理机制的滚动轴承故障诊断方法  

A rolling bearing fault diagnosis method based on CNN and introducing EMD preprocessing mechanism

作　　者：徐先峰[1] 王研 刘阿慧 郎彬 XU Xianfeng;WANG Yan;LIU Ahui;LANG Bin(School of Electronic & Control Engineering,Chang’an University, Xi’an 710064, China)

机构地区：[1]长安大学电子与控制工程学院,西安710064

出　　处：《工业仪表与自动化装置》2020年第4期7-11,共5页Industrial Instrumentation & Automation

基　　金：国家自然科学基金(61201407);陕西省自然科学基础研究计划(2016JQ5103)。

摘　　要：滚动轴承故障信号具有非线性、非平稳、强噪声特性,传统算法依赖于人工特征提取且缺乏自适应性。为此,该文利用经验模态分解(EMD)将原始特征集分解为一系列平稳的本征模态函数(IMF),结合相关系数遴选能突出更多局部特征的IMF构建特征向量,剔除部分噪声干扰信号。构造卷积神经网路(CNN)的多层特征提取网络,以遴选的特征向量为输入将其逐级变换为抽象的深层特征,最后完成特征域到故障类别域的映射。实验结果表明,该算法相比较其他方法具有更高的准确率、更好的鲁棒性。The signal of rolling bearing fault has non-linear,non-stationary and strong noise characteristics.The traditional algorithm depends on artificial feature extraction and lacks adaptability.In order to solve this problem,the original feature set was decomposed into a series of stationary intrinsic modal functions(IMF) by empirical mode decomposition(EMD),combined with correlation coefficient to select IMF which can highlight more local features constructing the feature vector,and eliminates some noise interference signals.A multi-layer feature extraction network of convolutional neural networks(CNN) is constructed,which is transformed into abstract deep features step by step with selected feature vectors as input,and finally the mapping between feature domain and fault category domain is completed.The experimental results show that the algorithm has higher accuracy and better robustness compared with other methods.

关 键 词：滚动轴承 EMD 相关系数 CNN 故障诊断 

分 类 号：TH113[机械工程—机械设计及理论]