{"title":"Multi-scale deep residual shrinkage networks with a hybrid attention mechanism for rolling bearing fault diagnosis","authors":"Xinliang Zhang, Yanqi Wang, Shengqiang Wei, Yitian Zhou, Lijie Jia","doi":"10.1088/1748-0221/19/05/p05015","DOIUrl":null,"url":null,"abstract":"\n The fault diagnosis of rolling bearings based on deep\n networks is hindered by the unexpected noise involved with\n accessible vibration signals and global information abatement in\n deepened networks. To combat the degradation, a multi-scale deep\n residual shrinkage network with a hybrid attention mechanism\n (MH-DRSN) is proposed in this paper. First, a spatial domain\n attention mechanism is introduced into the residual shrinkage module\n to represent the distance dependence of the feature maps. Then, a\n hybrid attention mechanism considering both the inner-channeled and\n cross-channeled characteristics is constructed. Through the\n comprehensive evaluation of the feature map, it provides a soft\n threshold for the activation function and realizes the feature-map\n selection adaptively. Second, the dilated convolution with different\n dilation rates is implemented for multi-scale context information\n extraction. Through the feature combination of the DRSN and the\n dilated convolution, the global information of the rolling bearing\n fault is strengthened and preserved as the fault diagnosis network\n is deepened. Finally, the performance of the proposed\n fault-diagnosis model is validated on the dataset from Case Western\n Reserve University (CWRU). The experimental results show that,\n compared with common convolution neural networks, the proposed\n neural diagnosis model provides a higher identification accuracy and\n better robustness under noise interference.","PeriodicalId":16184,"journal":{"name":"Journal of Instrumentation","volume":null,"pages":null},"PeriodicalIF":1.3000,"publicationDate":"2024-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Instrumentation","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.1088/1748-0221/19/05/p05015","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"INSTRUMENTS & INSTRUMENTATION","Score":null,"Total":0}
引用次数: 0
Abstract
The fault diagnosis of rolling bearings based on deep
networks is hindered by the unexpected noise involved with
accessible vibration signals and global information abatement in
deepened networks. To combat the degradation, a multi-scale deep
residual shrinkage network with a hybrid attention mechanism
(MH-DRSN) is proposed in this paper. First, a spatial domain
attention mechanism is introduced into the residual shrinkage module
to represent the distance dependence of the feature maps. Then, a
hybrid attention mechanism considering both the inner-channeled and
cross-channeled characteristics is constructed. Through the
comprehensive evaluation of the feature map, it provides a soft
threshold for the activation function and realizes the feature-map
selection adaptively. Second, the dilated convolution with different
dilation rates is implemented for multi-scale context information
extraction. Through the feature combination of the DRSN and the
dilated convolution, the global information of the rolling bearing
fault is strengthened and preserved as the fault diagnosis network
is deepened. Finally, the performance of the proposed
fault-diagnosis model is validated on the dataset from Case Western
Reserve University (CWRU). The experimental results show that,
compared with common convolution neural networks, the proposed
neural diagnosis model provides a higher identification accuracy and
better robustness under noise interference.
期刊介绍:
Journal of Instrumentation (JINST) covers major areas related to concepts and instrumentation in detector physics, accelerator science and associated experimental methods and techniques, theory, modelling and simulations. The main subject areas include.
-Accelerators: concepts, modelling, simulations and sources-
Instrumentation and hardware for accelerators: particles, synchrotron radiation, neutrons-
Detector physics: concepts, processes, methods, modelling and simulations-
Detectors, apparatus and methods for particle, astroparticle, nuclear, atomic, and molecular physics-
Instrumentation and methods for plasma research-
Methods and apparatus for astronomy and astrophysics-
Detectors, methods and apparatus for biomedical applications, life sciences and material research-
Instrumentation and techniques for medical imaging, diagnostics and therapy-
Instrumentation and techniques for dosimetry, monitoring and radiation damage-
Detectors, instrumentation and methods for non-destructive tests (NDT)-
Detector readout concepts, electronics and data acquisition methods-
Algorithms, software and data reduction methods-
Materials and associated technologies, etc.-
Engineering and technical issues.
JINST also includes a section dedicated to technical reports and instrumentation theses.