Photo credit (banner): Alex Profit

Pioneering predictive maintenance at SNCF

Innovative predictive maintenance solutions are delivering enhanced reliability and on-time performance for passengers. They’re also revolutionizing our day-to-day work. Get the inside story from Cyril Verdun, Director of Maintenance Engineering in our Rolling Stock Division.

Published on | Updated on

“Predictive maintenance has reduced breakdowns by nearly two-thirds.”

What is predictive maintenance?

Do you visit the doctor just to find out whether your knee hurts? Of course not. Our knees have a web of built-in sensors that constantly send out signals. When the brain receives them, it acts like an algorithm, interpreting the messages and telling us when the pain is constant or gets too sharp.

And tells us it’s time to see the doctor?

Exactly. That’s predictive maintenance: we receive data from our trains, put it to work and then interpret it with algorithms we’ve developed in-house. Today we can assess the condition of our trains remotely and in real time, which means that we can eliminate routine preventive maintenance tasks. The fact is that 90% of preventive maintenance involves checking something—a level, a value, a condition. Now that we already have that data, we can skip all the checking and focus on the remaining 10%.

But what’s predictive about it?

In my opinion, “predictive maintenance” is a misnomer. I prefer “analytic maintenance” or “data maintenance”. Instead of preventive maintenance based on averages—like 20,000-km servicing on a car—we eliminate periodic inspections altogether. In addition to predicting and anticipating failures, we can tailor maintenance schedules to the actual condition of rolling stock. To stick with the medical metaphor: your knee isn’t exactly the same as your co-worker’s knee or your neighbour’s. No two trains are identical either. So we adapt our periodic checks based on actual use and the condition of each train.

What parts of the train do you check?

We analyse data from the doors, the mobile steps, the air conditioning and heating systems, the toilets, the compressors (which use air to release the brakes and raise the pantograph1), the batteries, the lighting, the engines, the onboard passenger information system, the video surveillance system and the brakes. We can also control the pressure applied by the pantographs without climbing onto the train’s roof. Batteries are another example. Thanks to some new settings, we know exactly what condition they’re in. On the ground, that means that we no longer need technicians for annual check-ups on heating and air conditioning. Similarly, we track the water level in the toilets so we can optimize filling. And we know which cooling and heating units are working and what kinds of problems to expect: a leak, a clogged filter, a breakdown in an air-conditioning compressor, and so on.

How do you collect and analyse all the data? 

On older trainsets, we install sensors connected to the Internet of Things2 (IOT) and use them for remote diagnostics. Output from the sensors is limited to select data of our choosing. Once we’ve analysed that, we know when to expect a breakdown.

What about new trains?

There’s no need to install sensors on more recent models, such as the Francilien, the Régiolis and the Regio 2N3, the future RER NG trainsets, the TGV M and the AMLD medium and long-distance multiple units for INTERCITÉS. These trains are natively designed with a network and 4G SIM cards that transmit thousands of data points for every train every day. We have 8,000 variables for each train, and analyse 2,000 in real time. If a user on one of these trains opens a door, the door receives a command, opens and tells the network: “I opened in X seconds and consumed Y amperes of current.” We simply capture this dialogue between the command, the door and the IT network. The same is true for the air conditioning, heating, compressors, lighting and so on.

So there are two levels of predictive maintenance?

Absolutely. The first is condition-based maintenance (CBM). At this level, which is extremely advanced, we receive innumerable data points covering all of the train’s variables. At the second level, we use remote diagnostics to collect targeted data, which is often binary. Obviously, this type of maintenance is less precise, because native networks can transmit about 1,000 times more data than the sensors we install for remote diagnostics. Right now 1,000 SNCF trainsets are equipped with the most recent technology, which lets us do CBM, and 2,000 are set up for remote diagnostics. We’re the only ones in the world doing predictive maintenance on this scale.

Which lines are the most technologically advanced?

We’ve developed these tools over the past 8 years. Originally, we focused on Transilien lines H, K, J, L, R and P, but in the past 3 years we’ve deployed them on Régiolis and Regio 2N trains on TER lines. The next step will be the RER NG, set for rollout on line E in 2022 as part of the Eole project. It’s manufactured by Alstom, and thanks to our specifications, it will have 100% predictive maintenance.

We’re also doing a lot of work with SNCF Voyages to equip existing TGV trainsets with IOT technology. We want to expand our use of remote diagnostics and CBM if possible, especially for the batteries, the heating and cooling systems, the toilets, and so on. But we’ll really take a big step forward when our new TGV M goes into operation.

How does this benefit your passengers?

I’m tempted to tell them, ‘Relax. We’re “spying” on your train.’ If anything goes wrong, we know about it in real time, and we build in a lot of redundancy. For example, if the air conditioning stops working in one trainset, the other coaches can compensate. With Francilien, Regio 2N and Régiolis trainsets, we can look ahead and know with 95% certainty whether a failure will happen in a week, in two days or in three. Over the past 5 years, we’ve cut breakdowns by more than half in trainsets with remote diagnostics, and by nearly two-thirds on lines with predictive maintenance.

And that improves on-time performance? 

Unquestionably. Here’s a case in point: we have nearly 100 Francilien trainsets at Paris Nord station, and in the past, 9 of them were always offline for maintenance. Today it’s only 6 or 7. That’s a significant difference, because we can make rail traffic more robust—by adding an extra train at rush hour, for example.

20 %

reduction in train maintenance costs

Predictive maintenance has also revolutionized the day-to-day work of rolling stock employees, hasn’t it?

Digital technology is changing the workplace for all of us, from front-line operations staff to the heads of our specialized maintenance centres. Technicians are transitioning from a world where trains didn’t talk at all to trainsets that are literally in dialogue with them—like a veterinarian who changes careers and becomes a general practitioner for humans. Now patients can tell us what their symptoms are, so we can make a faster, much more accurate diagnosis.

How does that work on the ground?

In the past, we scheduled maintenance around deadlines and a calendar. We’d bring rolling stock in for repairs without realizing we’d need certain tools, or that we’d have to move the trainset to a special track to access the roof. Predictive maintenance is completely different: now we schedule work based on the variables for each train. Routine servicing is a thing of the past, and technicians are no longer flying blind. When they carry out maintenance, they know exactly what and where the problem is, they know the history of the part they’re working on, and thanks to data signatures, they know how much time the repair will take. Bottom line: we’ve cut maintenance costs by 20% and reduced shunting and maintenance centre visits by 30%. And those aren’t estimated gains—we’ve actually measured them.

That know-how has to be an advantage for SNCF as the market opens to competition.

I’m convinced of it. Tomorrow’s maintenance market will belong to the provider that is best able to know what’s happening with the train in real time. Our know-how is so good that we’re even better at predictive maintenance and remote diagnostics than the train and equipment manufacturers that design and build our rolling stock. We didn’t reach this level of excellence by accident. We were convinced that data analysis would pay dividends, so my teams and I got in very early. In the spring of 2013, we hired a data scientist, and we’re exploring our data more and more every day. We’re looking for new practices and new ways to make everyday life better for our passengers and our employees.

 

Découvrez en vidéo le télédiagnostic et la maintenance prédictive chez SNCF

How remote diagnostics work

Diagram detailing the remote diagnostics chain. Step 1: Train communicates independently with IOT as it passes over instruments in the ground. Step 2: Data is transmitted to data storage servers. Step 3: Scientists and rail experts analyse the data. Step 4: Once the data has been analysed, it is retrieved for use by computerized maintenance management software.

1 A hinged device mounted on an electric train to collect power through contact with the catenary.

2 The “Internet of Things” (IOT) is the network of physical objects that are connected to the Internet.

3 Built by Bombardier, a Canadian train manufacturer.