Moving from corrective maintenance (i.e. when a system fails or is under-performing) to condition-based maintenance (i.e. by monitoring condition indicators) has been one of the key improvements of the rail sector in the past decades. This shift was made possible by the democratization of IoT sensors which gave access to real-time data capture. It extended the lifecycle of the assets and increased the availability of the infrastructure. For the infrastructure Manager or the Rolling Stock Company, it meant more durability, less failures, less penalties ; for the passenger: less incidents with less delays.
Data science pushes this shift one significant step further. Through statistical analysis and modelling, it challenges state-of-the-art knowledge and allows for the application of predictive maintenance (i.e. predicting degradation to anticipate the need for maintenance) [1] [2].
Why is predictive maintenance so important for track management?
Regular maintenance of the track is crucial to ensure it constantly complies with the highest quality and safety standards. However, track maintenance operations such as tamping are partially destructive: each operation slightly deteriorates the geometrical properties of the ballast, the rocks that support the sleepers and the rails, thus leading to a slight increase in the degradation speed of the track. In a nutshell: the more you maintain, the more you will need to maintain and the sooner you will need to conduct investment-heavy renewal operations…
Hence, achieving “just-in-time” maintenance is key.
AI usage for predictive maintenance on the Tours-Bordeaux High Speed Line
LISEA (Concession owner) and MESEA (Maintenance Contractor) work jointly to optimize the level of performance and the lifecycle of the High Speed Line. Thanks to an in-house measurement wagon called “DRING”, they collect high-quality track geometry data on a weekly basis. In collaboration with the AI Program by Leonard, LISEA & MESEA were able to model the degradation of the track, the evolution of defects and the impact of a maintenance operation. As a result, maintenance operations could be optimized on a predictive basis. This led to the development of a decision supporting tool to assist experts in planning track geometry operations on both short term (few months) and mid-term (3-years).
Leveraging the data collected and machine learning techniques, LISEA & MESEA have already seen the first benefits of the solution.
- The understanding of track degradation and of the impact of a maintenance operation has improved, enabling the identification of some root factors that could benefit from further investigation.
- Time spent on monitoring the infrastructure and planning the maintenance operations has shrunk, thus allowing the experts to reallocate their time to more value-added tasks.
- And most importantly, the frequency of maintenance operations itself could be significantly reduced. This saves time, money and extends the global lifecycle of the assets while ensuring that the highest quality and safety standards are constantly met.
This decision-supporting tool has been particularly useful in the period of Covid19 where maintenance plans had to be revised in urgency. The plans proposed by the solution have all been proofed and validated successfully by a field expert.
This project demonstrates the relevance of applying machine learning techniques to maintenance in the rail sector and shows an example of the immediate and longer term benefits that can be derived from it.
Railway infrastructure is associated with massive construction costs and has been therefore traditionally state financed. In 2017 LISEA, a company partially owned by VINCI, Caisse des Dépôts et Consignations, Méridiam and Ardian, inaugurated the first public-private French partnership for a high-speed line between Tours and Bordeaux . The business model of this PPP (Public-Private Partnership) project is the following: the concession owner carries the financing, construction, maintenance, and renewal costs (DBOFM) and bears the performance and traffic risks. In a performance-based contract maintenance, investment and planning play a strategic role for a concession owner.
This article is part of a series involving the participants of the AI program by Leonard. The program has been specifically designed to accelerate the adoption of IA technologies within the group VINCI. It consists in a five-month incubation period where selected VINCI collaborators follow a learn by doing process where they codevelop an AI-based use case under the coaching and mentoring of the Leonard Team and Eleven consultants.
Sources
1. The rail sector’s changing maintenance game – Mc Kinsey – February 2018
2. Track geometry degradation and maintenance modeling: A review – July 2016
