Session: Track 1-5B: Track Safety

Paper Number: 136485

136485 - Using AI and Machine Learning for Rail Lidar Processing at Scale 

The incorporation of mobile mapping LiDAR technology into rail infrastructure assessment has transformed the way we perform inspection of ballast, clearances, vegetation, and assets. For over a decade, LiDAR has outperformed conventional manual inspection methods in terms of cost-effectiveness and time efficiency. Despite these advantages, the processing of LiDAR point clouds remains a time-consuming task, particularly when dealing with extensive rail networks spanning thousands of miles of data. By the time the final extracted deliverables are submitted, the results could be already out of date. Furthermore, the LiDAR scanning process can be disruptive to train schedules, and full scan campaigns are done at sporadic intervals with many years in between or as a one-time occurrence. 

 

To facilitate a seamless transition from LiDAR scanning to actionable insights, this study considers a two-part approach to LiDAR scans on Railways; the first part examines a mostly hands-off approach ( or “set it and forget it”) of installing an automated LiDAR capture system capable of collecting continuous LiDAR mounted on passenger rail or revenue-generating cart with menial human intervention. The second part looks at the data side of LiDAR; leveraging automated workflows, AI and Machine Learning techniques to automate the processing, classification and extraction of features from the LiDAR data. 

 

The goal is to continuously update scans on railways on a regular basis to detect deficiencies in railway clearances and overall asset changes within terabytes of cloud-stored LiDAR data while not disrupting train schedules. The proposed rail data platform has significantly expedited deliverable timelines for numerous rail operators, including US passenger transit and UK and Australian railways. The acceleration is notable, increasing from processing under 50 miles of data to over 1000 miles per week. Importantly, this acceleration does not compromise data accuracy or integrity.

 

Furthermore, the rapid processing of LiDAR data facilitates the prompt identification and tagging of problematic areas or regions of concern. This approach enhances geolocation and tagging, allowing for early awareness compared to traditional point cloud processing methods across extensive distances. The streamlined methodology significantly improves the efficiency of response times and enhances overall rail network maintenance and safety protocols.

 

In essence, the integration of AI and Machine Learning into the analysis of LiDAR data not only overcomes the challenges associated with laborious processing but also revolutionizes the speed and effectiveness of rail infrastructure evaluation. The presented rail data platform stands as a testament to the transformative impact of advanced technologies in the realm of transportation infrastructure management.

 

Presenting Author: Justin Theriaut Cordel.ai

Presenting Author Biography: Justin Theriault has recently joined Cordel as a Sales Engineer,
He has 12 years of experience in the Geospatial industry working with mobile LiDAR, GPS
and GIS. Justin holds a degree in Geomatics Engineering from the University of New
Brunswick in Canada. Justin currently works and lives in Colorado.

Authors:

Justin Theriaut Cordel.ai
Negin Shafie Zadeh Cordel.ai