USA: Norfolk Southern has begun deploying digital train inspection portals which use machine vision to identify faults on moving wagons.
The portals have been developed under a joint project between the railroad’s Data Science/Artificial Intelligence and Mechanical teams and the Georgia Tech Research Institute.
An array of 24 megapixel cameras captures an average of 1 000 ultra-high-resolution, 360° images of each vehicle as it passes at up to 110 km/h. The cameras are placed at angles where they can observe things that are difficult to detect with the human eye, and by capturing images while the train is moving they enable the identification of defects that cannot be found when it is stationary.
The Data Science/AI team has developed 38 deep learning algorithms for analysing the images, achieving high accuracy levels with low false-positives.
Information from the portals is transmitted to Norfolk Southern’s Network Operations Centre, where it can be reviewed by subject-matter experts to identify and address issues. Any critical defects are flagged for immediate handling.
The first portal has been installed at Leetonia in Ohio, where trains typically every hour. More than 12 are expected to be deployed across the NS network by the end of 2024.
