Hurst, K.A.Conradie, P.D.F.2025-10-232025-10-232025http://hdl.handle.net/2263/104935Papers presented virtually at the 43rd International Southern African Transport Conference on 07 - 10 July 2025.This paper explores the development of a real-time railway defect detection system using machine learning and sensor integration, tested on an HO scale model train. The project builds upon previous research, aiming to enhance defect identification and blockage detection while the train is in motion. A custom YOLO-based object detection model was developed, alongside a LiDAR-based obstacle REAL-TIME TRACK detection system, both integrated into the onboard train technology. The system stops the train upon detecting obstacles or defects, using a custom-designed relay circuit. The project faced challenges such as optimizing the YOLO model, overcoming hardware limitations, and developing a solution for rotating the camera and LiDAR sensors to accommodate track curves. Technical innovations include the creation of a customizable defect detection system that is adaptable to various real-world conditions. The study demonstrates the potential for real-world application, showcasing the ability to improve railway maintenance practices and operational safety. Future commercial applications are discussed, highlighting the potential for implementation in large-scale railway systems to reduce unplanned maintenance and operational costs.11 pagesPDFSouthern African Transport Conference 2025Machine learningLiDARArticle