- +10 data sources coming from SNCF and Vinci
- 15% better incidents detection
- 1st delivery: 2 iterations of 1 month each
Vinci, through its subsidiaries Mesea-Lisea, has obtained in 2010 the concession for the LGV line operation (High Speed Line between Tours and Bordeaux, France). The achievement of contractual commitments with railway operators (SNCF today, other potential actors in the future) requires fine infrastructure management and the processing of large volumes of data. This high-stake service level concession comes with high penalties, and therefore, the incidents management and their number have a big impact on the railway operations’ profitability.
As a long-time maintenance and material management expert, it was also critical for the company to accelerate its digital transformation and reinvent the management of its digital assets to be able to focus on its core business: railway infrastructure.
Aggregating various data sources
Stakes are multiple: counting and identifying trains, infrastructures availability, responsibilities sharing when an issue arise (rail incident or disabled train), maintenance plan optimization by detecting weak early signals.
To be as simple and efficient as possible, the solution had to integrate more than 20 data sources, including:
- Data coming from communication between SNCF control center and trains
- IoT data
- Meteorological sensors
- Data from the maintenance train controlling various elements of the infrastructure
Operationalizing data processing
Although all this information already existed, it was completely fragmented between the different departments. This limited approach did not help to fully leverage and power data or to go further by developing new services based on Artificial Intelligence.
After a thorough evaluation of different solutions, Vinci chose ForePaaS for several reasons:
- Access to all modules required for a data project in one place: collection, storage, exposition, training and run of AI models
- A platform with the capability to connect and process multiple data sources and formats, including high-frequency IoT, passive and active data
- Agile collaboration with fast iteration cycles
A first demonstrator has been deployed in less than 6 months to better understand the origin of unavailabilities, with a focus on train switches for which data are harder to process. A second phase was launched in the spring of 2019, focusing now on data extraction and predictive models development to move towards more predictive maintenance.