No Cover Image

Journal article 109 views 17 downloads

An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts

Yi Wang Orcid Logo, Xiaopei Cai, Bin Cui, Xueyang Tang, Clare Wood Orcid Logo, Yue Hou Orcid Logo

Computer-Aided Civil and Infrastructure Engineering, Volume: 49, Start page: 100098

Swansea University Authors: Clare Wood Orcid Logo, Yue Hou Orcid Logo

  • 71989.VoR.pdf

    PDF | Version of Record

    © 2026 The Authors. This is an open access article under the CC BY license.

    Download (7.3MB)

Check full text

DOI (Published version): 10.1016/j.cacaie.2026.100098

Abstract

High-speed railway (HSR) turnouts are among the most mechanically demanding components in the railway infrastructure, yet current operation and maintenance (O&M) practices remain largely reactive, experience-dependent, and disconnected from automated decision support. This paper presents a large...

Full description

Published in: Computer-Aided Civil and Infrastructure Engineering
ISSN: 1093-9687
Published: Elsevier BV 2026
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa71989
first_indexed 2026-05-28T13:27:16Z
last_indexed 2026-06-09T08:55:06Z
id cronfa71989
recordtype SURis
fullrecord <?xml version="1.0"?><rfc1807><datestamp>2026-06-08T14:25:09.7945032</datestamp><bib-version>v2</bib-version><id>71989</id><entry>2026-05-28</entry><title>An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts</title><swanseaauthors><author><sid>97bede20cc14db118af8abfbb687e895</sid><ORCID>0000-0003-0001-0121</ORCID><firstname>Clare</firstname><surname>Wood</surname><name>Clare Wood</name><active>true</active><ethesisStudent>false</ethesisStudent></author><author><sid>92bf566c65343cb3ee04ad963eacf31b</sid><ORCID>0000-0002-4334-2620</ORCID><firstname>Yue</firstname><surname>Hou</surname><name>Yue Hou</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2026-05-28</date><deptcode>ACEM</deptcode><abstract>High-speed railway (HSR) turnouts are among the most mechanically demanding components in the railway infrastructure, yet current operation and maintenance (O&amp;M) practices remain largely reactive, experience-dependent, and disconnected from automated decision support. This paper presents a large language model (LLM)-driven expert system that bridges the gap between raw sensor data and actionable maintenance decisions for turnout defect diagnosis. Three core contributions are made. First, a data textualization strategy is developed to convert train body acceleration signals into structured text sequences comprehensible to LLMs, enabling domain-specific diagnosis without architectural modification of the base model. Second, an enhanced instruction fine-tuning scheme is proposed, incorporating a contrastive loss function that tightens intra-class feature clusters and widens inter-class margins, alongside a hierarchical evaluation method that reliably extracts categorical intent from free-form model outputs. Third, a retrieval-augmented generation (RAG) module is integrated with the fine-tuned model, enabling the system to generate standards-compliant maintenance recommendations directly from diagnostic results. Controlled experiments across four pre-trained models and 26 experimental groups demonstrate that the proposed system reaches a peak diagnostic accuracy of 89.6%, while preserving the natural language generation capabilities essential for report production. The framework is evaluated on a physically representative dataset generated by a validated stochastic vehicle&#x2013;turnout dynamics model. The resulting integrated pipeline, from extracted signal features to maintenance decision output, offers a practical and scalable solution for intelligent O&amp;M of complex railway turnout infrastructure and beyond.</abstract><type>Journal Article</type><journal>Computer-Aided Civil and Infrastructure Engineering</journal><volume>49</volume><journalNumber/><paginationStart>100098</paginationStart><paginationEnd/><publisher>Elsevier BV</publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint>1093-9687</issnPrint><issnElectronic/><keywords>High-speed railway; Turnout; Large language model; Defect diagnosis; Intelligent operation and maintenance</keywords><publishedDay>1</publishedDay><publishedMonth>9</publishedMonth><publishedYear>2026</publishedYear><publishedDate>2026-09-01</publishedDate><doi>10.1016/j.cacaie.2026.100098</doi><url/><notes/><college>COLLEGE NANME</college><department>Aerospace Civil Electrical and Mechanical Engineering</department><CollegeCode>COLLEGE CODE</CollegeCode><DepartmentCode>ACEM</DepartmentCode><institution>Swansea University</institution><apcterm>Other</apcterm><funders>The work was supported by the State Key Laboratory of Advanced Rail Autonomous Operation (RAO2025ZT002), Beijing Jiaotong University; Tianjin Key R&amp;D Programme for Beijing&#x2013;Tianjin&#x2013;Hebei Collaborative Innovation (25YFXTHZ00260); the Natural Science Foundation of Beijing, China (L251029); the Fundamental Research Funds for the Central Universities (2025QYBS007); and the Key Research Project of China Railway Design Corporation (2024A0253805).</funders><projectreference/><lastEdited>2026-06-08T14:25:09.7945032</lastEdited><Created>2026-05-28T14:21:20.9476820</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Aerospace, Civil, Electrical, General and Mechanical Engineering - Civil Engineering</level></path><authors><author><firstname>Yi</firstname><surname>Wang</surname><orcid>0009-0005-0816-8152</orcid><order>1</order></author><author><firstname>Xiaopei</firstname><surname>Cai</surname><order>2</order></author><author><firstname>Bin</firstname><surname>Cui</surname><order>3</order></author><author><firstname>Xueyang</firstname><surname>Tang</surname><order>4</order></author><author><firstname>Clare</firstname><surname>Wood</surname><orcid>0000-0003-0001-0121</orcid><order>5</order></author><author><firstname>Yue</firstname><surname>Hou</surname><orcid>0000-0002-4334-2620</orcid><order>6</order></author></authors><documents><document><filename>71989__36882__8865bc780ad84c6b8e84e94caad18d5d.pdf</filename><originalFilename>71989.VoR.pdf</originalFilename><uploaded>2026-06-08T14:22:49.7983796</uploaded><type>Output</type><contentLength>7659037</contentLength><contentType>application/pdf</contentType><version>Version of Record</version><cronfaStatus>true</cronfaStatus><documentNotes>&#xA9; 2026 The Authors. This is an open access article under the CC BY license.</documentNotes><copyrightCorrect>true</copyrightCorrect><language>eng</language><licence>http://creativecommons.org/licenses/by/4.0/</licence></document></documents><OutputDurs/></rfc1807>
spelling 2026-06-08T14:25:09.7945032 v2 71989 2026-05-28 An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts 97bede20cc14db118af8abfbb687e895 0000-0003-0001-0121 Clare Wood Clare Wood true false 92bf566c65343cb3ee04ad963eacf31b 0000-0002-4334-2620 Yue Hou Yue Hou true false 2026-05-28 ACEM High-speed railway (HSR) turnouts are among the most mechanically demanding components in the railway infrastructure, yet current operation and maintenance (O&M) practices remain largely reactive, experience-dependent, and disconnected from automated decision support. This paper presents a large language model (LLM)-driven expert system that bridges the gap between raw sensor data and actionable maintenance decisions for turnout defect diagnosis. Three core contributions are made. First, a data textualization strategy is developed to convert train body acceleration signals into structured text sequences comprehensible to LLMs, enabling domain-specific diagnosis without architectural modification of the base model. Second, an enhanced instruction fine-tuning scheme is proposed, incorporating a contrastive loss function that tightens intra-class feature clusters and widens inter-class margins, alongside a hierarchical evaluation method that reliably extracts categorical intent from free-form model outputs. Third, a retrieval-augmented generation (RAG) module is integrated with the fine-tuned model, enabling the system to generate standards-compliant maintenance recommendations directly from diagnostic results. Controlled experiments across four pre-trained models and 26 experimental groups demonstrate that the proposed system reaches a peak diagnostic accuracy of 89.6%, while preserving the natural language generation capabilities essential for report production. The framework is evaluated on a physically representative dataset generated by a validated stochastic vehicle–turnout dynamics model. The resulting integrated pipeline, from extracted signal features to maintenance decision output, offers a practical and scalable solution for intelligent O&M of complex railway turnout infrastructure and beyond. Journal Article Computer-Aided Civil and Infrastructure Engineering 49 100098 Elsevier BV 1093-9687 High-speed railway; Turnout; Large language model; Defect diagnosis; Intelligent operation and maintenance 1 9 2026 2026-09-01 10.1016/j.cacaie.2026.100098 COLLEGE NANME Aerospace Civil Electrical and Mechanical Engineering COLLEGE CODE ACEM Swansea University Other The work was supported by the State Key Laboratory of Advanced Rail Autonomous Operation (RAO2025ZT002), Beijing Jiaotong University; Tianjin Key R&D Programme for Beijing–Tianjin–Hebei Collaborative Innovation (25YFXTHZ00260); the Natural Science Foundation of Beijing, China (L251029); the Fundamental Research Funds for the Central Universities (2025QYBS007); and the Key Research Project of China Railway Design Corporation (2024A0253805). 2026-06-08T14:25:09.7945032 2026-05-28T14:21:20.9476820 Faculty of Science and Engineering School of Aerospace, Civil, Electrical, General and Mechanical Engineering - Civil Engineering Yi Wang 0009-0005-0816-8152 1 Xiaopei Cai 2 Bin Cui 3 Xueyang Tang 4 Clare Wood 0000-0003-0001-0121 5 Yue Hou 0000-0002-4334-2620 6 71989__36882__8865bc780ad84c6b8e84e94caad18d5d.pdf 71989.VoR.pdf 2026-06-08T14:22:49.7983796 Output 7659037 application/pdf Version of Record true © 2026 The Authors. This is an open access article under the CC BY license. true eng http://creativecommons.org/licenses/by/4.0/
title An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
spellingShingle An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
Clare Wood
Yue Hou
title_short An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
title_full An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
title_fullStr An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
title_full_unstemmed An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
title_sort An instruction fine-tuning and retrieval-augmented generation framework for intelligent defect diagnosis and maintenance decision support in high-speed railway turnouts
author_id_str_mv 97bede20cc14db118af8abfbb687e895
92bf566c65343cb3ee04ad963eacf31b
author_id_fullname_str_mv 97bede20cc14db118af8abfbb687e895_***_Clare Wood
92bf566c65343cb3ee04ad963eacf31b_***_Yue Hou
author Clare Wood
Yue Hou
author2 Yi Wang
Xiaopei Cai
Bin Cui
Xueyang Tang
Clare Wood
Yue Hou
format Journal article
container_title Computer-Aided Civil and Infrastructure Engineering
container_volume 49
container_start_page 100098
publishDate 2026
institution Swansea University
issn 1093-9687
doi_str_mv 10.1016/j.cacaie.2026.100098
publisher Elsevier BV
college_str Faculty of Science and Engineering
hierarchytype
hierarchy_top_id facultyofscienceandengineering
hierarchy_top_title Faculty of Science and Engineering
hierarchy_parent_id facultyofscienceandengineering
hierarchy_parent_title Faculty of Science and Engineering
department_str School of Aerospace, Civil, Electrical, General and Mechanical Engineering - Civil Engineering{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Aerospace, Civil, Electrical, General and Mechanical Engineering - Civil Engineering
document_store_str 1
active_str 0
description High-speed railway (HSR) turnouts are among the most mechanically demanding components in the railway infrastructure, yet current operation and maintenance (O&M) practices remain largely reactive, experience-dependent, and disconnected from automated decision support. This paper presents a large language model (LLM)-driven expert system that bridges the gap between raw sensor data and actionable maintenance decisions for turnout defect diagnosis. Three core contributions are made. First, a data textualization strategy is developed to convert train body acceleration signals into structured text sequences comprehensible to LLMs, enabling domain-specific diagnosis without architectural modification of the base model. Second, an enhanced instruction fine-tuning scheme is proposed, incorporating a contrastive loss function that tightens intra-class feature clusters and widens inter-class margins, alongside a hierarchical evaluation method that reliably extracts categorical intent from free-form model outputs. Third, a retrieval-augmented generation (RAG) module is integrated with the fine-tuned model, enabling the system to generate standards-compliant maintenance recommendations directly from diagnostic results. Controlled experiments across four pre-trained models and 26 experimental groups demonstrate that the proposed system reaches a peak diagnostic accuracy of 89.6%, while preserving the natural language generation capabilities essential for report production. The framework is evaluated on a physically representative dataset generated by a validated stochastic vehicle–turnout dynamics model. The resulting integrated pipeline, from extracted signal features to maintenance decision output, offers a practical and scalable solution for intelligent O&M of complex railway turnout infrastructure and beyond.
published_date 2026-09-01T06:02:44Z
_version_ 1868490882258305024
score 11.109323

Similar Items