基于变分模态分解和卷积神经网络融合的滚动轴承故障诊断方法  

A VMD and CNN Combined Fault Diagnosis Method for Rolling Bearings

作　　者：李魁 隋新[1] 刘春阳[1] 李济顺[1,2] 徐彦伟[1] 杨芳[1] Li Kui;Sui Xin;Liu Chunyang;Li Jishun;Xu Yanwei;Yang Fang(School of Mechanical Engineering,Henan University of Science and Technology,Luoyang 471003,China;Henan Key Laboratory for Machinery Design and Transmission System,Luoyang 471003,China)

机构地区：[1]河南科技大学机电工程学院,河南洛阳471003 [2]河南省机械设计及传动系统重点实验室,河南洛阳471003

出　　处：《机械传动》2022年第11期134-140,共7页Journal of Mechanical Transmission

基　　金：郑洛新国家自创区创新引领型产业集群专项(201200210400);国家重点研发计划(2020YFB2009602);河南省高等学校重点科研项目计划(21B460004)。

摘　　要：针对在强烈背景噪声影响下的滚动轴承故障特征提取困难,提出了一种基于变分模态分解与卷积神经网络融合的滚动轴承故障诊断方法。将原始振动信号分解为多个模态分量,结合皮尔逊相关系数作为自动分解终止阈值和最优模态分量选取指标;针对轴承故障特征构建卷积神经网络,将最优模态分量作为输入以提取、分类故障类型。试验结果表明,所提方法能够精确诊断滚动轴承故障,为强噪声影响下的滚动轴承故障识别提供了新的思路。Aiming at the difficulty of extracting fault features of rolling bearings under the influence of strong background noise,a rolling bearing fault diagnosis method based on the fusion of variational mode decom⁃position(VMD)and convolutional neural network(CNN)is proposed.After decomposing the original variation signal into multiple components,the proposed method employs the Pearson correlation coefficient as the auto⁃matic decomposition termination threshold and the optimal modal component selection index;a convolutional neural network is constructed according to bearing fault features and the optimal modal component is used as the input to extract and classify the fault types.The experiments validate that the proposed method can accurately di⁃agnose the rolling bearing faults,which is validated as a new method for rolling bearing fault diagnosis regarding strong background noise.

关 键 词：滚动轴承 故障诊断 强背景噪声 变分模态分解 卷积神经网络 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TH133.33[自动化与计算机技术—控制科学与工程]