基于CEEMDAN与改进一维多尺度卷积神经网络结合的滚动轴承故障诊断  

Fault diagnosis of rolling bearing based on CEEMDAN combined with improved one-dimensional multi-scale convolutional neural network

作　　者：马宁 赵荣珍[1,2] 郑玉巧 MA Ning;ZHAO Rong-zhen;ZHENG Yu-qiao(School of Mechanical and Electronical Engineering,Lanzhou University of Technology,Lanzhou 730050,China;College of Intelligent Manufacturing and Electrical Engineering,Guangzhou Institute of Science and Technology,Guangzhou 510540,China)

机构地区：[1]兰州理工大学机电工程学院,甘肃兰州730050 [2]广州理工学院智能制造与电气工程学院,广东广州510540

出　　处：《兰州理工大学学报》2025年第1期45-54,共10页Journal of Lanzhou University of Technology

基　　金：国家自然科学基金(52465012,52475134)。

摘　　要：针对滚动轴承信号微弱故障特征提取困难、故障诊断依靠大量专家经验和故障识别率低等问题,提出了融合自适应噪声完备集合经验模态分解与改进一维多尺度卷积神经网络的滚动轴承故障诊断方法.首先,采用自适应噪声完备集合经验模态分解对轴承信号进行消噪处理,并利用皮尔逊相关系数法对所得IMF分量进行信号重构;其次,在网络首层将大尺寸卷积核与空洞卷积结合,并引入金字塔场景解析网络提出改进的一维多尺度卷积神经网络,对故障特征信息进行提取,采用PSO算法对卷积核进行参数寻优;最后,融合多尺度特征信息完成网络学习,并输入Sofmax分类器,实现滚动轴承故障诊断.采用西储大学轴承数据集和HZXT-DS-001型双跨综合故障模拟实验台的滚动轴承故障数据进行了验证.结果表明,相比传统故障诊断方法该方法可以得到良好的诊断结果.To solve the weak feature extraction of the fault signals in rolling bearing and the low fault identification accuracy relying on expert’s experience,a fault Identification method combining complete ensemble empirical mode decomposition with adaptive denoise and improved one-dimensional convolutional neural network was proposed.First,the complete ensemble empirical mode decomposition is used to denoise the bearing signal,followed by signal reconstruction using Person correlation coefficient method on the obtained intrinsic mode functions(IMFs).Next,the first layer of the network adopts a method of combining large-scale kernels and atrous convolutions,the pyramid scene parsing network is introduced and leading to a proposed one-dimensional multi-scale convolutional network for extracting the fault features and the scale of convolution kernel is optimized by PSO algorithm.Finally,the multi-scale feature information is integrated to complete the network learning.And it is input into the Softmax classifier to achieve the fault category.The method is verified by the bearing data set of Western Reserve University and the rolling bearing faults data collected at the simulating faults on HZXT-DS-001 rig.The results show that to compare with the traditional method,the method proposed in this paper has good diagnosis results.

关 键 词：自适应噪声完备集合经验模态分解 一维卷积神经网络 多尺度特征提取 特征可视化 故障诊断 

分 类 号：TH165[机械工程—机械制造及自动化] TP18[自动化与计算机技术—控制理论与控制工程]