Network Rail, the owner and manager of most of Britain’s railway infrastructure.
Ensuring that any trains can pass tunnels, platforms and other trackside infrastructure without risking a collision is vital for both the safety of passengers and service performance. The measurement of such assets in relation to the track is known as ‘structure gauging’.
But monitoring the placement of assets – which can move as ground shifts or new structures are built – over 20,000 miles of rail is a time-consuming and costly challenge. Sending engineering teams out with laser measuring devices to manually check clearance adds an element of risk for humans, too.
Faculty and AECOM responded to a Network Rail challenge to automate, for the first time, aspects of these rail asset surveys, using AI to provide accurate, safety-critical information from across its network.
We built a model that analyses point cloud data from train-mounted optical sensors, identifying assets and structures and comparing data to the existing Network Rail rail infrastructure database.
The model identifies structures like platforms, signals, tunnels and vegetation, measures track geometry and precise location, and uses these inputs to evaluate whether there is sufficient clearance. With this information, Network Rail can quickly identify which elements of its infrastructure are safe and in line with industry standards, and which assets are in need of attention.
Our tool takes a ‘human in the loop’ approach that prompts a human to periodically validate the outputs of the model. This approach allows us to combine the best of machine learning with human decision-making to maintain the highest possible accuracy standards and improve accuracy over time – crucial to both the safety of the rail network and Network Rail’s trust in the tool.
Our solution demonstrates for the first time that it is possible to deliver these safety-critical assessments with millimetre accuracy. This technology therefore offers a more efficient, safer method of delivering structure gauging assessment – potentially preventing a wide range of injuries, saving time, and reducing the cost of assessment to tax payers.