FRA report confirms utility of AI-based railway inspection system

An AI-based inspection system is able to consistently identify railway changes (such as unsafe weather conditions) warranting immediate action, which could help the Federal Railroad Administration (FRA) address such defects before they affect rail service, according to an April FRA report on extended field trials using the system.

The Laser Rail Inspection System (LRAIL), developed by Quebec-based Pavemetrics, is a 3-dimensional laser inspection system that combines high-speed laser triangulation sensors and artificial intelligence to automatically inspect the condition of, and detect safety-related changes in, railway infrastructure. Whereas conventional rail inspection systems typically focus on a single task, like imaging, LRAIL's combination of high-resolution imaging and 3D scanning is designed to deliver greater return on investment.

For the FRA project, LRAIL technology was installed on a specialized hi-rail inspection trailer and data were captured on Amtrak lines over a four-month period in three separate deployments across an eight-mile test loop near the Delaware-Pennsylvania border. This stretch of railway was selected because it is active, receives regular traffic, and both natural deterioration and restorative maintenance were expected over the test period.

Repeat runs in the same direction and reverse runs were captured during each deployment to model repeatability, mean, and standard deviation of measurements. The noise floor for change detection (the practical limit for the magnitude of change that can be detected) was also modeled through analysis of repeat runs and the detection of known/deliberate changes.

LRAIL change measurements were determined to be highly repeatable, with an overall average of 99.28% agreement between runs. Most importantly, the ability to detect organic changes above the noise floor was validated with numerous examples, including ballast level, ballast fouling, tie condition, joint bar bolting, and joint gap being detected.

Read more: https://railroads.dot.gov/elibrary/extended-field-trials-lrail-automated...

Read the report: https://railroads.dot.gov/sites/fra.dot.gov/files/2020-04/Pavemetrics%20...