The Canada Line is a 19km urban rail system in Vancouver, British Columbia, which opened in 2009. It uses fully-automated trains and is reckoned to be the most reliable railway in North America, with around 99.9% availability and punctuality. The operations and maintenance are run by SNC-Lavalin Group, which has the contract until 2040.
The digital twin is a representation of every asset in the system, recording its attributes, parameters and performance, often accompanied by a virtual 3D model. Assets and systems can be monitored, examined and assessed virtually, so that failures can be predicted, maintenance scheduled and improvements modelled.
Most digital twins come out of ‘greenfield’ projects: the digital twin is constructed alongside or in advance of the physical project itself, with every asset designed and built with this in mind. But the Canada Line was a ‘brownfield’ project, where the digital twin had to be constructed on the basis of existing assets. Ryan Versteeg-Biln, director of asset management at SNC-Lavalin operations & maintenance, pictured below, explains: “We started the journey in early 2021, and we really wanted it to be user-driven, not technology- or developer-driven. So the first few months was really just workshops with myself and my colleagues – user meetings with some of the information collection and data analytics specialists to identify the issues that we were having on the Canada Line, regardless of whether or not the technology was there.
“We ended up with a bunch of potential process improvements and issues, and then specialists would work with the graphical and technical design teams to identify those that we know a digital twin is not going to assist with, or those that would require a substantial investment in additional sensors; we put them to the side.”
“One of the benefits of brownfield implementation is you don’t have the harsh timelines [of] engineering construction,” says Versteeg-Biln, “so we could afford to take our time and make sure we got it right. We wanted to identify what we were trying to improve, and what were the metrics associated with those process improvements.”
PROJECT PLANS
“We started technical development and delivery in August 2021. It was in a beta state by about March 2022”. Since then, he adds, “we have been refining, developing and adding additional functionality”. He says it is in ‘developer-assisted use’ at the moment, and “we’re working on getting it ingrained into the base processes of the Canada line.”
Versteeg-Biln describes it as “an implemented asset register with a graphical interface and data connected to it”. The system currently covers trackwork and ‘guideway’ (the support structures), but not all the maintainable assets of the railway: “We’ve got a development pipeline for rolling stock next. We want to get that wheel-rail interface built up. Power supply and distribution will probably be after that, although we don’t have the sensor suites in place to really capitalise on that” – this is where a ‘greenfield’ setup would have had an advantage.
Versteeg-Biln demonstrates the live system, bringing up the line superimposed on a map, then a schematic trackplan, then details of specific items of track such as a set of points (or switch). There is also a full 3D scan of the trackwork and its surroundings. This ‘point cloud’ was generated using a Pegasus 2 LiDAR scanner system, and the impressive graphics show everything down to switches, platforms and even the foliage beside the track.
“You can see the immense value for planning maintenance work, especially in our tunnel environment.” There is only a two-hour window during the night when the regular trains are not running, so this model can be used to plan a procedure in detail.
“The other aspect of the digital twin is the maintenance dashboard. The backbone of the digital twin is built in Azure, Microsoft’s cloud computing framework, [and] the interface is React JS – it’s a completely in-house developed design. The dashboard is the web-enabled visualisation tool for the data hosting and analytics side.”
One of the main issues to look at was switch machine failure: “If they fail, they shut down the system. So we wanted to look at an AI-type approach or statistical numerical analysis of a failure that up until now had been unpredictable.
“For switch machines, hydraulic actuators move the track from one side to the other. We measure the motor current and the time the switch takes to swing from one position to another. We don’t have hydraulic fluid sensors yet, but that is one of the next steps.”
The switch failures had varying root causes: “Sometimes hydraulic leaks, sometimes electrical failures, but nothing really associated with age or any wear parameter that we could determine. We couldn’t find any correlation to predict these failures. So we took all the data available, and looked at a couple of years of data and associated failures, and said, what can an AI make of this?”
The most useful metric they found was not the swing time itself, nor the trend (longer or shorter) but its ‘volatility’: “If it is two seconds and then four and then one-and-a-half seconds and then five. If it started showing volatility, you had a much higher likelihood of imminent failure.
“So within the dashboard, we looked at the volatility in real time and assigned a score. And you can see right now that switch 302 and switch 326 are at high risk, and that will trigger a special inspection. And possibly if it is a critical one, just replace the switch machine preemptively, do a rebuild and reassessment on the bench. Because it takes half an hour to change a switch machine, and you’d much rather do that in the middle of the night than during rush hour.”
His enthusiasm for the dashboard system is obvious: “I love it. It really lets us move from a schedule basis to predictive maintenance. Condition-based, assessment-based, anticipating failures.”
Importantly, the lessons learnt have applications not just for this and other rail projects, but for other industries: “The basis behind our switch predictive maintenance module can be used for any sort of statistical data analysis,” says Versteeg-Biln. He mentions valves as an example, and adds, “anything you have information on and want to draw correlations, we’ve got the backbone created.”
BOX: A SUITE OF SENSORS
While the initial mapping of the Canada Line used a LiDAR scan, for ongoing track assessment the firm has designed a range of sensor heads which mount to the bogies of passenger trains going out during regular service hours.
“For the running rail itself we have a non-contact wear and track geometry measurement system,” says Versteeg-Biln. “It measures the rail profile every meter, looks for twist faults and different issues with the rail – that brings back a ream of data that is used for a track condition assessment.”
Other sensor heads include a six-axis gyroscopic accelerometer and “a camera-based system that uses AI image processing to detect surface flaws, corrugation, as well as certain other conditions like broken rail clips”. These sensors can also be mounted to an unmanned ‘drone’ for close-ups.