Predictive maintenance for railway signalling, before it had a buzzword

How Bombardier Transportation and Diadrom built PREGO, a predictive maintenance module for railway signalling inside Bombardier’s Maintenance and Diagnostics Center.

Diadrom and Bombardier started PREGO (Predictive Goal) together in 2015: a web application, part of Bombardier’s MDC product line, that analyses historical and real-time data from signalling equipment, finds the behaviour that comes before a fault and raises alarms by rules the operator defines. It was delivered in January 2016 and taken into operation by Bombardier.

From the calendar to the condition of the equipment

Railway signalling is maintained the way most infrastructure is: on fixed intervals, and by repair when something breaks. Both are expensive. A fixed interval services equipment that did not need it; a failure in an interlocking, a point machine or a train-detection system stops traffic until it is fixed. The equipment already reports what it is doing. The step that changes maintenance is reading those reports in time to act before the fault, and planning the work, the technicians and the spares around what the data shows.

A global rail and signalling supplier

Bombardier Transportation was, at the time, one of the world’s leading suppliers of rail vehicles and signalling, with operations in 26 countries, about 40,000 employees and more than 100,000 vehicles installed worldwide. Its Maintenance and Diagnostics Center (MDC) is the product line that collects and presents data from installed signalling systems. PREGO was built to sit inside it.

The data was there. The foresight was not.

MDC held years of event data from signalling installations and received a live stream from the field: reports, alarms, messages and performance parameters from interlockings, point machines and detection systems. What Bombardier needed was a way to use that data before a failure, not after:

• detect incipient faults before they become failures, in point machines and train-detection systems above all;
• see performance degrade over time, as a trend rather than a single reading;
• correlate data across subsystems, and get an overview of the maintenance situation;
• prioritise maintenance work, and avoid work and spares that were not needed.

PREGO: a predictive module inside MDC

Diadrom and Bombardier ran PREGO as a joint development project. The brief was a predictive module, in the form of a web application, that can be implemented as a part of the MDC product line.

The module brings historical event data and real-time data together and analyses them with R scripts to identify deviating behaviour and anticipate faults. Operators see the result as dashboards, KPIs, trends and reports, and the module raises alarms by rules the users define themselves. It runs in a distributed environment across four servers, so it scales with the number of installations it watches.

A team of twelve did the work: a project manager, ten developers and two testers, in Gothenburg and in Poland. Base development ran from May 2015; the development phase began on 27 August 2015 and ran in sprints, each ending in a demonstration to Bombardier, until 15 January 2016. Final delivery was on 29 January 2016, and Bombardier took full ownership of the module in the second quarter of 2016.

“The progress was tremendous.”

— Bombardier Transportation, after the final sprint demonstration, January 2016

For the engineers in the room

• System: PREGO (Predictive Goal), a predictive maintenance module for Bombardier’s Maintenance and Diagnostics Center (MDC)
• Domain: Railway signalling: interlockings, point machines, train-detection systems; reports, alarms, messages and performance parameters
• Back end: Java EE, R for the analyses, MongoDB
• Front end: AngularJS, D3.js and Crossfilter, for interactive dashboards, KPIs and trends
• Data: Historical event data and real-time data, analysed together; alarms from user-defined rules
• Scale: Distributed environment across four servers
• Verification: End-to-end, functional and performance testing, with a demonstration to Bombardier at the end of each sprint
• Team and time: Twelve people in Sweden and Poland; development from August 2015, final delivery 29 January 2016

What it changed

PREGO gave Bombardier’s MDC a way to look ahead: to see in the data which equipment was heading for a fault, while there was still time to plan the work, the people and the parts. That was the point of the project: higher availability through early warning, maintenance done when it is needed rather than when the calendar says so, and a clearer picture for Bombardier’s product organisation of how its installed systems actually behave. Bombardier took the module over and ran it from 2016.

“Most organisations already have the data. The hard part is knowing what is about to happen before it affects operations, and building that into the tools people use every day.”

— Viktor Eliasson, CEO, Diadrom

Advice to peers

PREGO was built in 2015, before “predictive maintenance” and “AI” were on every slide, and the principles have not changed: find problems early, understand how systems affect each other, prioritise the right action, and base the decision on the data rather than on the interval. They apply just as directly to defence fleets, as in the Archer fleet-management case, and to software-defined vehicles today.

Details drawn from Diadrom’s project material on PREGO, 2015–2016. Company facts as Bombardier Transportation stated them at the time; Bombardier Transportation has been part of Alstom since 2021.