Automated Structure Gauging

Using machine learning to improve trackside gauging. Trackside gauging is the process of determining the clearance between railway structures and passing trains - it is a critical component of maintaining the safety of Network Rail infrastructure and understanding the suitability of rolling stock for different train routes. Gauging has historically been a slow and costly operation requiring line closures, large amounts of manual work and sometimes years to return results. Faculty - a leading UK-based applied AI company - is partnering with the world’s leading infrastructure firm AECOM to research and develop machine learning capabilities to intelligently detect and inspect railside assets and conditions to automate the gauging process for Network Rail. Working from LIDAR point-cloud data, Faculty will train supervised machine learning models using state-of-the-art research methods to detect and identify lineside assets as well as measure the cant, curvature and clearance along the line. These models will demonstrate the feasibility of automating the time-consuming process of creating survey files which are critical to provide Network Rail an understanding of track conditions and clearance. This project will build on cutting-edge machine learning and computer vision approaches that Faculty and AECOM have already deployed to automate the detection and inspection railside and roadside assets with optical video cameras. This includes current, in-flight research with Network Rail in which forward-facing-video is being tested for the highly-adjacent task of foliage encroachment detection. The new technology developed within this project has the potential to bring significant benefits to the UK rail sector, improving customer safety, reducing delays and costs. More accurate and timely understanding of the data will facilitate the introduction of new rolling stock and automating the survey process will reduce the amount of staff time spent on track.