Please use this identifier to cite or link to this item: http://bura.brunel.ac.uk/handle/2438/30389
Full metadata record
DC FieldValueLanguage
dc.contributor.authorWang, C-
dc.contributor.authorWang, Z-
dc.contributor.authorLiu, H-
dc.contributor.authorDong, H-
dc.contributor.authorLu, G-
dc.date.accessioned2024-12-27T12:49:17Z-
dc.date.available2024-12-27T12:49:17Z-
dc.date.issued2024-05-01-
dc.identifierORCiD: Chuang Wang https://orcid.org/0000-0001-8938-9312-
dc.identifierORCiD: Zidong Wang https://orcid.org/0000-0002-9576-7401-
dc.identifierORCiD: Hongjian Liu https://orcid.org/0000-0001-6471-5089-
dc.identifierORCiD: Hongli Dong https://orcid.org/0000-0001-8531-6757-
dc.identifierORCiD: Guoping Lu https://orcid.org/0000-0002-6815-4554-
dc.identifier.citationWang, C. et al. (2024) 'An Optimal Unsupervised Domain Adaptation Approach With Applications to Pipeline Fault Diagnosis: Balancing Invariance and Variance', IEEE Transactions on Industrial Informatics, 20 (8), pp. 10019 - 10030. doi: 10.1109/TII.2024.3385533.en_US
dc.identifier.issn1551-3203-
dc.identifier.urihttps://bura.brunel.ac.uk/handle/2438/30389-
dc.description.abstractA practical yet challenging scenario in transfer learning is unsupervised domain adaptation (UDA), where knowledge is transferred from a labeled source domain to unlabeled target domains. The crucially important role of domain-variant characteristics is often neglected by most existing UDA methods, which can deteriorate adaptation performance and result in negative transfer. In this article, an optimal unsupervised domain adaptation (OUDA) algorithm is proposed in order to address this issue, which balances the invariance of domain-sharing features and the variance of domain-specific features. In the proposed approach, a gradient adversarial adaptation (GAA) method is introduced to align the gradient directions of source and target features within the same category, thereby facilitating knowledge transfer. In addition, a local manifold embedding (LME) technique is proposed to preserve the intrinsic geometric structure of the original feature space while implementing distribution alignment, providing distinguishable features for UDA. To stabilize the process of knowledge transfer, an evolutionary control strategy is developed to adaptively control the tradeoff between the GAA and LME by employing the particle swarm optimization algorithm. Extensive experiments are conducted on cross-domain natural gas pipeline fault diagnosis, and the results on nine cross-domain classification tasks indicate that our OUDA algorithm outperforms the existing state-of-the-art UDA methods. Moreover, the performance analysis in terms of accuracy, loss, and domain divergence demonstrates the superior stability of the proposed OUDA algorithm in dealing with unsupervised knowledge transfer.en_US
dc.description.sponsorship10.13039/501100001809-National Natural Science Foundation of China (Grant Number: 61933007, U21A2019, 62273005 and 62073180); Hainan Province Science and Technology Special Fund of China (Grant Number: ZDYF2022SHFZ105); AHPU High-End-Equipment Intelligent Control Innovation Team (Grant Number: 2021CXTD005); Alexander Von Humboldt Foundation of Germany.en_US
dc.format.extent10019 - 10030-
dc.format.mediumPrint-Electronic-
dc.languageEnglish-
dc.language.isoen_USen_US
dc.publisherInstitute of Electrical and Electronics Engineers (IEEE)en_US
dc.rightsCopyright © 2024 Institute of Electrical and Electronics Engineers (IEEE). Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. See: https://journals.ieeeauthorcenter.ieee.org/become-an-ieee-journal-author/publishing-ethics/guidelines-and-policies/post-publication-policies/-
dc.rights.urihttps://journals.ieeeauthorcenter.ieee.org/become-an-ieee-journal-author/publishing-ethics/guidelines-and-policies/post-publication-policies/-
dc.subjectevolutionary computationen_US
dc.subjectinvariance and varianceen_US
dc.subjectmanifold learningen_US
dc.subjectunsupervised domain adaptation (UDA)en_US
dc.titleAn Optimal Unsupervised Domain Adaptation Approach With Applications to Pipeline Fault Diagnosis: Balancing Invariance and Varianceen_US
dc.typeArticleen_US
dc.date.dateAccepted2024-03-17-
dc.identifier.doihttps://doi.org/10.1109/TII.2024.3385533-
dc.relation.isPartOfIEEE Transactions on Industrial Informatics-
pubs.issue8-
pubs.publication-statusPublished-
pubs.volume20-
dc.identifier.eissn1941-0050-
dc.rights.holderInstitute of Electrical and Electronics Engineers (IEEE)-
Appears in Collections:Dept of Computer Science Research Papers

Files in This Item:
File Description SizeFormat 
FullText.pdfCopyright © 2024 Institute of Electrical and Electronics Engineers (IEEE). Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. See: https://journals.ieeeauthorcenter.ieee.org/become-an-ieee-journal-author/publishing-ethics/guidelines-and-policies/post-publication-policies/4.44 MBAdobe PDFView/Open


Items in BURA are protected by copyright, with all rights reserved, unless otherwise indicated.