How Alstom is transforming rail with Artificial Intelligence

Nothing stands still in rail operations, as the state of trains and the networks they run on is constantly evolving. But they must stay in good health for safe, efficient and on time service. So, what happens when rail systems have everyday aches and pains or more serious concerns?

Like doctors, train maintenance experts analyse operational logs to understand the symptoms, diagnose problems and recommend “treatment.” This can take time, especially when a huge amount of information is involved.

Speeding up a painstaking task with AI
Our data scientists make it their business to work alongside engineers to develop a series of AI “super analysers,” that can halve the time taken to make diagnoses, while making them more accurate. “We are advancing fast in several applications of our AI diagnostics,” says Ossee Yiboe, Data Scientist at Alstom. “As part of everyday maintenance, data from trains and infrastructure is recorded chronologically in operation logs. When trains run smoothly, these logs are deleted automatically, but if something goes wrong, they are communicated in real time to drivers and troubleshooting experts.”

To avoid service disruption, solutions must be found quickly. Often several maintenance interventions are necessary before the right solution is found.

“Our goal is to use AI to create tools that help experts quickly find and fix system issues by analysing logs,” adds Ossee.

To make the process more efficient, AI models based on existing data sets recognise patterns, identify causes and root causes and suggest solutions to technicians on the ground. “In a particular example, our solution analyses around one thousand system log variables to identify the most likely causes of a problem and potential remedies, effectively narrowing the failure environment down to just a dozen probable causes. By leveraging this AI technology, organisations can significantly speed up troubleshooting by 8 times, thereby enhancing overall maintenance productivity.”

Making maintenance more efficient
AI-driven solutions can provide results with 90% accuracy when identifying the reasons of failure, a great support for less experienced maintainers. “At the end of the day we’re looking for cause and effect,” Ossee says. “Using interesting new techniques, we can learn from hundreds of variables and narrow them down to the one that has caused the failure. Each use case we develop will help diagnose future issues and make maintenance more efficient.”

www.alstom.com