Countries, cities and transport operators rely on Thales’ ground transportation solutions to adapt to rapid urbanization and meet new mobility demands – locally, between cities and across national frontiers.
Regarding the Rail systems, one of the solutions provided by Thales offers a Big Data, a cloud-hosted platform for visualizing and analyzing railway asset time-series data. The intelligence built into the product allows operators to identify and respond to failures before they cause major disruption to the rail network.
We collect large quantities of sensor data from various railway assets (points machines, communication logs, axle counters, track circuits, etc.). We provide a tool for visualizing and analyzing this data. As part of this product, we want to use machine learning techniques to automate the detection and diagnosis of faults.
However, traditional machine learning approaches require a large, labeled training set. In our use case, failures are rare and we generally have no or few labels for a particular fault type. We are interested in techniques that can handle this paradigm of very few (or no) labels, in real-time streaming deployed environment.
This challenge is global in scope and does not focus on a specific project.