基于并行网络多尺度特征融合的轴承故障诊断  

Bearing fault diagnosis based on multi-scale feature fusion of parallel network

作　　者：姜山 封松林[1] 吴波[1] 王文瑞 鲁方林 袁晓兵[3] JIANG Shan;FENG Songlin;WU Bo;WANG Wenrui;LU Fanglin;YUAN Xiaobing(Shanghai Advanced Research Institute,Chinese Academy of Sciences,Shanghai 201210,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Wireless Sensor Network&Communication,Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China)

机构地区：[1]中国科学院,上海高等研究院,上海201210 [2]中国科学院大学,北京100049 [3]中国科学院,上海微系统与信息技术研究所,无线传感网与通信重点实验室,上海200050

出　　处：《传感器与微系统》2023年第10期121-125,共5页Transducer and Microsystem Technologies

基　　金：上海市自然科学基金资助项目(19ZR1463800);中国科学院上海微系统与信息技术研究所无线传感网与通信重点实验室开放项目(20190903)。

摘　　要：针对健康带标签的滚动轴承故障数据稀少的问题,提出了一种基于并行卷积神经网络(CNN)多尺度特征融合的轴承故障诊断方法。首先,采用不同的轴承振动数据,获取对轴承数据的不同视角表达,分别作为并行CNN的输入;其次,用ImageNet数据集对VGG16网络进行预训练,并用带标签的轴承故障数据对预训练后的VGG16网络进行微调,取微调后的VGG16网络作为特征提取器,分别提取不同视角故障数据中的中间特征;最后,设计特征融合模块,通过多尺度特征融合得到高层次故障特征,并用高层次故障特征训练分类模块,实现轴承的故障诊断。经过实验验证,所提出的算法可以对轴承故障数据进行更全面的挖掘,最终获得更高的诊断准确率。Aiming at the problem that health labeled rolling bearing fault data is rare,a bearing fault diagnosis method based on multi-scale feature fusion of parallel convolutional neural network(CNN)is proposed.Firstly,different vibration data of bearings are used to obtain the expression of bearing data from different perspectives,which are respectively used as inputs of the parallel CNN.Secondly,ImageNet dataset is used to pre-train the VGG16 network,and labeled bearing fault data is used to fine-tune the pre-trained VGG16 network,the fine-tuned VGG16 network is used as feature extractor,intermediate characteristics of fault data from different perspectives are extracted respectively.Finally,a feature fusion module is designed to obtain high-level fault features through multi-scale feature fusion,and the classification module is trained with high-level fault features to realize fault diagnosis of bearings.It is verified by experiment that the proposed algorithm can conduct more comprehensive mining of bearing fault data,and the higher diagnosis accuracy is obtained finally.

关 键 词：轴承 故障诊断 并行网络 迁移学习 特征提取 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置] TH133.3[自动化与计算机技术—控制科学与工程] TP391[机械工程—机械制造及自动化]