基于SWT和改进CNN的滚动轴承故障诊断方法  

Rolling Bearing Fault Diagnosis Method Based on SWT and Improved CNN

作　　者：宋存利[1] 袁晓萌 张雪松 SONG Cunli;YUAN Xiaomeng;ZHANG Xuesong(School of Railway Intelligence Engineering,Dalian Jiaotong University,Dalian 116028,China)

机构地区：[1]大连交通大学轨道智能工程学院,辽宁大连116028

出　　处：《大连交通大学学报》2025年第2期52-58,共7页Journal of Dalian Jiaotong University

基　　金：国家自然科学基金项目(62276042);辽宁省教育厅科学研究计划项目(LJKZ0486)。

摘　　要：针对滚动轴承故障分类识别精度低、变负载模型泛化能力差的问题,提出一种基于同步压缩小波变换(SWT)和密集连接混合空洞卷积神经网络(DMCNN)的故障诊断方法。SWT将一维振动信号转换成二维信号时频图;DMCNN在CNN基础上引入密集连接结构和混合空洞卷积来进行特征提取和故障分类,利用密集连接结构来提高特征利用率,并通过混合空洞卷积来减少训练时长。试验与对比分析结果表明,该方法有较高识别精度,在变负载情况下也有更好的泛化能力。A fault diagnosis method based on synchrosqueezed wavelet transform(SWT)and densely connected mixed void convolutional neural network(DMCNN)is proposed to solve the problems of low accuracy in fault classification and recognition of rolling bearings and poor generalization ability of variable load models.The SWT converts one-dimensional vibration signals into two-dimensional signal time-frequency maps.The DMCNN introduces dense connection structures and mixed void convolutions on the basis of CNN for performing feature extraction and fault classification,using dense connection structures to improve feature utilization and mixed void convolutions to reduce training time.The experiment results and comparative analysis show that this method has high recognition accuracy and better generalization ability under variable load conditions.

关 键 词：故障诊断 同步压缩小波变换 卷积神经网络 混合空洞卷积 密集连接 

分 类 号：TH133.33[机械工程—机械制造及自动化]