How Network Rail turns AI fault warnings into maintenance work

Network Rail’s insight platform brings measurement-train data, track images and remote monitoring into one place to help engineers anticipate faults. Its purpose is to show what needs attention, where it is and when intervention may be needed. The useful output is a repair that can be planned before deterioration makes the problem urgent.[1]

Reading the railway

The New Measurement Train gathers evidence while travelling over the track. Lasers measure the rail’s profile; motion sensors record how the train moves; cameras inspect visible components. Its image-analysis system compares what it sees with the expected appearance of the track, including clips that hold the rails to sleepers.[2]

The Office of Rail and Road’s 2024 assessment records a deployed deep-learning upgrade to visual inspection, intended to distinguish real defects more effectively from false alarms such as debris. This is one AI task within the wider process: improving which images deserve attention before engineers decide what those findings require.[4]

Keeping a fault’s history

In a January 2026 account, Network Rail engineers explain how insight aligns measurements from successive train runs, improving relative location accuracy. It can then recognise a defect of the same type at the same place and retain one identifier as its severity changes. This prevents each observation from being treated as a new fault.[3]

With that history, insight fits a trend through previous measurements using relatively simple linear regression models. It estimates when the defect will reach an intervention threshold. Repeated readings become an expected maintenance deadline, provided the observed pattern continues.[3]

Reaching the maintenance team

Integration with the Ellipse work-management system in April 2024 allowed insight to create work orders that reach maintenance teams’ tablets, with completed work feeding back into the record. A fault already covered by timely planned work leaves the actionable queue; worsening conditions can trigger another alert.[3]

Earlier detection matters because maintenance needs preparation: the appropriate people, equipment and access to the railway must come together. A warning’s value lies partly in that usable lead time. Showing the condition and location in one place helps the engineer turn a forecast into a specific job to arrange.[1]

Report an error or suggest a correction