Leveraging Computer Vision and LLM for Smart Railway Fastener Defect Detection and Management

カテゴリ: 研究会(論文単位)

論文No: CMN25030

グループ名: 【C】電子・情報・システム部門 通信研究会

発行日: 2025/02/28

タイトル(英語): Leveraging Computer Vision and LLM for Smart Railway Fastener Defect Detection and Management

著者名: Biswas Soumen(Hitachi India Pvt. Ltd.),Mitra Abhsihek(Hitachi India Pvt. Ltd.),Ganesh Ananth(Hitachi India Pvt. Ltd.),Chankravortty Nilanjan(Hitachi India Pvt. Ltd.)

著者名(英語): Soumen Biswas(Hitachi India Pvt. Ltd.),Abhsihek Mitra(Hitachi India Pvt. Ltd.),Ananth Ganesh(Hitachi India Pvt. Ltd.),Nilanjan Chankravortty(Hitachi India Pvt. Ltd.)

キーワード: Railway Fastener|Defect Detection|Object Detection|Computer Vision Application|Large Language Model|Railway Fastener|Defect Detection|Object Detection|Computer Vision Application|Large Language Model

要約(日本語): The identification and systematic documentation of various defects in railway track fasteners are critical concerns, as these defects pose significant safety risks and could potentially lead to rail accidents if not promptly addressed. The present paper i

要約(英語): The identification and systematic documentation of various defects in railway track fasteners are critical concerns, as these defects pose significant safety risks and could potentially lead to rail accidents if not promptly addressed. The present paper introduces a methodology that facilitate user interaction and expedite data analysis to detect the railway fastener defects accurately with cutting-edge Vision-LLM powered chatbot. The proposed methodology incorporates several key components: 1) a geotagging camera mounted on a train inspection vehicle, 2) a YOLO-based model for accurate identification of fastener defects, and 3) a chatbot system for facilitating user interaction. A geotagging camera installed on a train inspection vehicle captures image frames of railway track fasteners, which are then analysed for defects. A YOLO-based object detection model is employed to accurately identify various types of track defects in these frames, creating a comprehensive database that includes fastener defects information, their confidence score to measure criticalness of the defects and their precise GPS locations . To enhance accessibility and understanding of the defects within this database, a large language model (LLM)-powered chatbot is integrated. This chatbot allows users to query the database and receive relevant, detailed responses in a clear and conversational manner, simplifying the process of defect visualization and information retrieval. The proposed solution demonstrates improved accuracy in detecting fastener defects, while also providing a smart method for retrieving detailed information about the detected defects. This assists the track maintenance team in prioritizing high-risk defects that may require immediate attention.

本誌: 2025年3月3日-2025年3月4日通信研究会

本誌掲載ページ: 47-51 p

原稿種別: 英語

PDFファイルサイズ: 2,055 Kバイト