
AI improves Railway Maintenance and Operations
Pierre Dersin, who represents Alstom Digital & Integrated Systems, is a member of the board of AI Factory Railways, AIFR, at Luleå University of Technology.
“With digitalization, physical operations on hardware are, as much as possible, replaced with virtual operations relying on software.Therefore software is becoming increasingly crucial, which is why the AIFR initiative is very opportune”, he says.
The railway industry wants to minimize risks related to punctuality and availability. To achieve that goal, real-time management of maintenance and operations is key, and so are off-line simulations; in that context , trust in software and data is of paramount importance, according to him.
Pierre Dersin, who is also an adjunct professor in the Division of Operation and Maintenance Engineering at Luleå University of Technology, believes that digitalization is an important enabler in order to be able to simulate how different parts of the system interact with each other and how they are used.
“Objectives include predicting what can cause delays in the railway traffic. We can also analyze the need for maintenance before service-affecting failures occur, and organize maintenance scheduling accordingly ”, he says.
Predictive maintenance
Under this new philosophy, traditional time-based or mileage-based preventive maintenance gives way to condition-based, predictive maintenance, and rigid maintenance plans will be replaced with dynamic maintenance planning.
Alstom is a major player in the railway industry and sustainable mobility worldwide, in particular in Europe, in rolling stock, signaling, infrastructure, turnkey systems and services. When the company acquired Bombardier Transportation last January, Sweden became even more interesting for Alstom.
“ The AIFR project is well suited for us. Right now, the work is underway to share data among the participating stakeholders, including train operating companies, asset owners ,infrastructure managers, and original equipment manufacturers, and to feed the AI algorithms that transform such data into useful, actionable information.
The various stakeholders need each other, but at the same time it is understandable that they are careful about sharing data. Within AIFR, this aspect of data ownership is governed by legal agreements. Cybersecurity aspects must be dealt with as well.
There are business models in other industries that may be sources of inspiration”.
Transfer knowledge
It should be possible to extract value from the information that is useful to each participating entity without making confidential raw data available, according to Pierre Dersin.
An important consideration, Pierre Dersin adds, is that usually one tends to look at one individual asset, such as a train, or a train door, or a point machine, while the real challenge is to manage the maintenance of an entire fleet of trains or point machines. Different assets may perform similar functions, but with variations in context.
“They do not quite have the same mission profile. For instance, a train may travel along a more difficult route than another; or a point machine may perform far more maneuvers than another or may be subject to a heavier traffic flow. The question then arises, if we have learned about one of these machines then how do we transfer the knowledge to the other ones? ”
Advanced AI methods, such as domain-adaptive transfer learning, deal with those questions. That subject is addressed in one of the use cases of AIFR.
An innovative project
Another key issue is cybersecurity and data integrity. When data is contributed to the system, or shared with other stakeholders , it is important that it should not be possible to alter such data without the proper authorization. Another AIFR use case addresses the feasibility of using block chain as a data-securing mechanism.
“There are other use cases being investigated as part of AIFR; such as using revenue-service trains to observe and monitor the condition of the tracks to detect early damage and deterioration trends before they result in safety or operational hazards. The question then is exactly what physical variables we should monitor, how the data should be processed and how optimal maintenance decisions could be made on that basis.
Pierre Dersin thinks that the AIFR project is interesting and innovative for Alstom, as it complements the latter’s R&D.
“It has given new perspectives on our ongoing projects. This is something we can benefit from. I look forward to being able to share such a good academic work with the industry”, says he.
[https://www.uochd.se/article/view/803480/artificiell_intelligens_vid_jarnvagsunderhall]