
From data to predictive maintenance at ÖBB
“To get a complete and clear picture of the rail infrastructure, it is important to organise data so it speaks the same language”, says Erik Pinter, track analysis expert at Austrian Railways ӦBB. There might also be valuable data hidden in sources that are not always used.
An important element that ӦBB is looking into, is in the subgrade layers under the track, as well as the drainage systems. Understanding and monitoring these are vital for being able to improve maintenance and notice faults earlier to keep a good track alignment.
The Austrian rail network of ÖBB Infrastructure consists of 4,875 route networks, or 9,762 kilometres of operational tracks and 13,385 turnouts. Currently, a big data platform is being implemented at ÖBB to analyse, understand and model track condition and degradation from various data sources, called the Data Science Hub (DSH). Erik Pinter is Civil Engineer and Data Scientist, and has been working at ÖBB for around 15 years. Currenty, he is working on the data platform, heading a team for track analysis.
The overall goal of the data platform? To provide simple, specific solutions to the questions, says Pinter. “The great thing is that you can look at the data on different scales. You can look at the whole network, all 9800 kilometres, but also zoom in to see the details, up to 25 centimetres of track.”
Big data – overrated?
“I think a lot of people overestimate what big data, and all those buzzwords AI, deep learning can provide. First you have to think what the questions are that you want to answer. And then whether your data can actually provide answers to these questions. We have put a lot of emphasis in providing clean, meaningful data on the platform.”
The measurement vehicles have been running on the Austrian track for around 20 years, and all this data is stored. “Until now it was tedious work to look at this data, and it was far from being able to be used for things like predictive maintenance or prognosis. The way the data was ingested in the database was too complex for the kind of questions we want to ask. For the last one and a half years we thought of ways to transfer the data and save it in a meaningful way.”
For example, so that the parameters of the track measurement vehicles that are one year old can be compared to a value of 20 years ago. “This is critical, because if the data is garbage, no matter how intelligent the tools, it will still be garbage. We made sure all data speaks the same language.”
A good picture of the network
Data from measurement vehicles is just one part, there is also data from ground penetrating radar systems, Lidar scanners, and asset data. This is brought together in the Data Science Hub, to have one common understanding of which data belongs to what assets. “We now have a very good picture of our network, and that is the basis of comparing different data with each other.”
After organising the data, you can then also import completely new data. The next steps are to get data from, until now, unknown sources, says Pinter. For example, from regular trains. “We don’t start adding sensors to them. The vehicles already have an internal system where they keep track of changes in engine temperatures, and other operational parameters.”
These are normally only used for operational purposes, but some of these can also make sense for the tracks, says Pinter. “For example, seeing how much energy is put through wheels into the track. The vehicles measure this, and we are interested in using it on top of the network. I think with enough data we will be able to explain why we get certain failures on the track which we couldn’t explain until now”.
Text continues below picture

Underneath the tracks
One of the first questions ÖBB wants to answer is to find the spots on the network where there are defects that do not originate from the tracks, but from the subgrade. “This includes moisture in the ballast, so we can point people on the ground where to look, and on which spots it could be necessary to renew the drainage or the ballast.”
Also, an AI model was developed to be able to see small changes in signals from the measurement vehicles which seemed to correlate with failures in the subgrade layers. “That was something we couldn’t do until now, because if you want to have a ‘deeper’ look you always need additional measurements by ground penetrating radar to see what’s going on beneath the tracks.” With a deep learning model, it was studied if those effects can also be found in subtle changes of the signals of the measurement vehicles which routinely inspect the tracks. “It looks very promising, to detect those anomalies from a regular behaving track. We need to scale this up however to see what other benefits it may have.”
Predictive maintenance
Where it’s all headed to is predictive maintenance, to predict what will happen with the track. “We already have a lot of data, but to make a serious prognosis that is reliable, we need to import some additional data in the Data Science hub, which is planned in the next half year.” These are things like data on how many trains there are on a single track, the exact weights of the trains, and if there are irregularities in the wheels, he explains. But also the environmental surroundings. “A track which receives a lot of rain behaves very differently than in a dry area. Hydrological and geological properties are also important to import in the platform, otherwise it remains guesswork.”
One part is actually using data that is already there, says Pinter. “Data might be hidden somewhere, and you have to find those sources, such as in the geology. Then there are also still blind spots, which need additional solutions with extra sensors. We have a current project with sensors for drainage systems, to see what’s going on underneath the tracks.”
“It is quite unique that we have quite a lot of Lidar data gathered by maintenance vehicles, so we know a lot about the surroundings of the tracks. On top of that data we already offer some services, such as viewing how much vegetation there is on the sides of tracks, and identifying trees that might be dangerous.” According to Pinter, the next step is to combine Lidar with track alignment data and ground penetrating radar data to get a complete picture and monitor drainage systems and trenches next to the rail, which is essential for proper alignment of the track.
Open-source solutions
What is different about the ÖBB data platform than most, is that it is based on free and open-source solutions. Erik Pinter explains: “We didn’t want to use some proprietary format or be stuck to a certain company. Everything is made up of modules, so if there is a better solution for a certain part, it can be replaced.” It has some big advantages to use open- source software, he sees. “Some of the big companies that provide big data technology are not pushing their technology as much as people had hoped for. We hope open-source keeps developing itself and comes up with new things.”
Erik Pinter shares more insights at the Intelligent Rail Summit on 21-23 September. Take a look at the programme for more information. Download the free magazine about Intelligent Rail with more interviews here.