During the summer of 2019, several employees of the SNCF denounced a major health problem in some of the freight cars of the railway company. Indeed, these employees discovered asbestos in a dozen freight maintenance workshops, an insulating and resistant material used in many constructions until it was banned in 1997 because it was toxic to humans. When the SNCF became aware of the situation, it decided to tackle the problem by calling on Datategy, a data science company, to design a customised tool to detect asbestos.

A major health problem linked to the presence of asbestos in some SNCF facilities

In June 2019, following the protest of several SNCF employees about asbestos in a dozen maintenance workshops for freight cars, the company decided to call on Datategy to design a custom AI solution capable of detecting the presence of asbestos. Patrick Munsch, Head of Wagon Maintenance & Engineering at SNCF Voyageurs, explained how asbestos was previously detected and why he turned to AI:

“During our maintenance and engineering operations, we automatically perform an asbestos diagnosis to check whether the wagons are sound or not. This process can take up to three days, which delays maintenance times. To reduce these delays, our teams regularly asked us for a tool to identify the presence of asbestos in equipment. However, such a solution did not exist on the market. So we wondered if we couldn’t create it.”

In September 2020, while the model was being designed and trained, 183 railway workers at the SNCF technicentre in Oullins in the Rhône region had filed a complaint against the SNCF for exposure to asbestos during their careers.

A machine learning tool that immediately recognizes asbestos in rail cars

In order to train the tool for the SNCF, Datategy needed 15 months, many photographs and numerous statistical results on asbestos to achieve an error rate of less than 1%. To train the algorithm, Datategy separated the project into several phases. First, the company collected field data, thanks in particular to the photo library created by the SNCF. Once the model was stabilized, the team then added a diagnostic function to the application. Patrick Munsch provides more explanation as to how the AI solution works:

“Within seconds, we have the results. The interface is very simple, we train the employees (welders, boilermakers, industrialists, etc.) in less than two hours. We have about 100 users on a daily basis. [Of course, it is important to reduce the downtime of wagons during maintenance, but that is not the only benefit of the solution. It meets the demands of staff who have a reliable, easy-to-use tool that incorporates their recommendations.”

At present, thanks to this platform, 2,500 parts are diagnosed per year, but the objective remains to manage to scan the entire fleet of wagons in order to be sure that none of them have asbestos…

Translated from Afin de lutter contre la présence d’amiante dans ses wagons, la SNCF fait appel au machine learning