‘Use machine learning to predict track maintenance’
Machine learning is an important asset to predict where track maintenance will be needed, Dirk Bothof of Asset Insight says. Machine learning is an artificial intelligence technique that gives computers the ability to learn. Asset Insight has developed a method with which track geometry can be predicted so that timely maintenance can be carried out.
In short, track geometry is how and where the track has been laid. It also involves looking at constructions such as tunnels and bridges, as well as where stations are and what objects are in the immediate vicinity of the tracks. When trains ride on the tracks, they sometimes want to ‘move’ if the ballast does too. That means, for example, that the tracks can end up slightly closer to a platform than intended, so that trains come too close to the edge of the platform, or conversely the tracks can move too far away.
If an infrastructure manager knows in advance how the tracks are behaving across the whole country, and most importantly knows the positions of the tracks in the near future, better decisions can be taken about where in the rail network needs extra attention, and above all when this attention is required.
As there are more and more trains on the network, this is important knowledge. “The toll on the tracks is increasing,” says Bothof. “For each account, railway companies have less time for maintenance. The tracks must be in the right position. If you want to plan maintenance properly, how the tracks are at the moment is less important than what problems there will be with them in the future.” If you know about the future state of the tracks, it benefits the rail sector. “It ensures that you carry out maintenance where it is needed, meaning that the maintenance is more efficient, which in turn leads to greater availability of the network.”
Asset Insight, a company that engages in measurement and inspection services for the rail sector among others, has made a model that uses machine learning. The program teaches itself how the tracks move. Until recently, machine learning was not used much in the rail sector, but now its use is booming, says Bothof. “It is very helpful, as it allows you to carry out more targeted maintenance, which prevents rising costs.”
By looking at data from the past, a self-learning model can be built which can determine where the tracks will be in the future. This is then confirmed manually by using a measuring train, to check the program’s accuracy. “That means we can assess how good the program is at making predictions. So we can see very specifically what it’s good at, and what it’s not so good at.”
The tracks must be within the limits for all parameters. But then the question is: when do you carry out maintenance? The program can help with this, too. “Imagine, the limit is a deviation of 0.4, and the current deviation is 0.39. Then you might not want to do maintenance straight away, as the tracks may not move for the next six months.” The model gives insight into this. “Without the model, you might have already done the maintenance, and that would have been for nothing. It would simply be throwing money away.”
Computer and man
That is not to say that everything completely depends on a computer. “During the development of the program, we have worked together with our client and have taken many intermediate steps. We have looked at how our algorithm would plan maintenance and how people would do it. In the majority of cases, this is the same, except that different priorities are given. In addition, the program can calculate the correct time interval more accurately.”
And that’s where the advantage of the model lies: not in where you must carry out maintenance, but what you must prioritise. “It helps to determine where a rail infrastructure manager must focus its attention.”
- ‘Railway sector must face challenge of loss of expertise with digitalisation’
- TU Graz is using acoustic waves to detect track failures
- ‘Combine data with observations and physics for track maintenance’
Manager Advanced Analytics Dirk Bothof will speak during the Intelligent Rail Summit 2018 on 27-29 November in Malmö, Sweden. The programme for the Intelligent Rail Summit 2018 is complete. Visit the website to discover the full programme.
Railway operators examine autonomous shuttles for last mile
Automatic Train Operation, more common as ATO, is becoming more and more familiar to the railway companies across Europe. Just in the Netherlands, two railway operators have already tested the technology and the third one plans to do this by the year’s end. Not only the autonomous trains attract the railway undertakings, they are also… Read more ›
SCNF, DB Cargo and KLM complete programme of Intelligent Rail Summit
With the addition of DB Cargo, SNCF and KLM/Air France, the programme of the Intelligent Rail Summit 2019 has been completed. During the rail conference that takes place from 19 to 21 November at the UIC headquarters in Paris, various experts will share their views on Automatic Train Operation (ATO) and the Digital Railway. Earlier… Read more ›
Digital rail will be top of mind during Intelligent Rail Summit 2019 in Paris
In November, Paris will be immersed in the world of digital rail when the three-day Intelligent Rail Summit 2019 takes place at the UIC headquarters. This high-level conference that will be held on 19, 20 and 21 November will address the topics Automatic Train Operation and the Digital Railway. The conference will kick off on… Read more ›