Swiss Federal Railways CIO relies more on AI than concrete

News Analysis

Mar 28, 20244 mins

Artificial IntelligenceCIOData Management

Jochen Decker is fully committed to AI for complex optimization projects to yield measurable cost benefits.

Railway construction couldn’t be more laborious than in Switzerland, as the country consists almost exclusively of mountains, most of which are now spanned with bridges and riddled with holes, like the famous local cheese.

The rail network is also the densest in Europe to the point where it can no longer be expanded because all the necessary areas are already fully utilized. “We can only optimize,” said Jochen Decker at the Hamburg IT Strategy Days in February. And this is urgently needed because Swiss Federal Railways (SBB) expects 30 to 40% more passengers in 2034 than today, putting that much additional strain on an unchanged route network. So Decker came to Hamburg to report on how it can be achieved, and show the central role that artificial intelligence will play.

Opportunities not realized before

SBB, unlike Deutsche Bahn, is an integrated group that brings together passenger and freight transport, infrastructure, and real estate under a single roof, which facilitates the planning and implementation of investments and innovations. The IT budget amounts to €850 million per year, which is about 7% of sales.

A few years ago, SBB prescribed three optimization programs that will cost around €1 billion by 2027. In terms of traffic management, the aim is to make better use of routes, in particular by reducing the distances between trains. Production planning wants to get more kilometers out of people and materials, ensuring that trains stand still as little as possible, and that train drivers spend as much of their working time as possible driving rather than on other things. The third part of the program, asset management, is intended to reduce material wear and tear, and make better use of the workshops.

Of the €1 billion allocated to the three programs, only €20 million is allocated to AI, however. “Nevertheless, this opens up opportunities for us that we didn’t have before,” says Decker, who’s been working in this space at SBB for five years.

AI enabling predictive maintenance

What fascinates Decker about AI is not only its possibilities, but its low costs, like in wheelset and track management. With the help of constant monitoring of wheel wear by cameras and sensors, and the evaluation of the results obtained in the process, data allows him to predict very precisely when a wheel needs to be replaced. If this forecast is then matched with the utilization data from the repair shop, it becomes real predictive maintenance since the wheel is replaced neither too early nor too late, and the repair shop has the time and capacity to make the change immediately. “The prerequisite for this is high-quality data,” said Decker, yet it doesn’t take a lot of money — at least for the AI. In this example, its use accounted for less than €300,000.

SBB takes a very similar approach to track maintenance. They are filmed with the help of a measuring vehicle that drives over them at 120 km/h and their condition is evaluated. “If a crack is found during a measurement run,” he said, “the question always arises: Is it the same one we found the day before, or a new one that is perhaps only five centimeters off?” AI helps to distinguish between the two cracks.

Another example of AI use at SBB is operations management, or optimizing train path utilization. After all, the answer to the question of which train to run and where is highly complex. If you want to roll across Switzerland, you can choose between innumerable routes. Of course, it costs much less to plan using AI than to build new tunnels and tracks, which is no longer a viable option.

Keep it simple

Decker is also convinced that the use of AI is much easier today than it was two or three years ago because popular applications such as ChatGPT have opened the doors for it, including those of corporate boards. However, the fascination with technology sometimes leads to overcomplication. “In some cases, data scientists invent problems that the customer doesn’t even have, simply because the data allows it,” he says.