Landslide hazard assessment framework for cut-slopes along railroad rights-of-way using statistical analysis of images
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Transportation corridors created using cuts and fills often introduce steep slopes, or geohazards, along rights-of-way that require frequent monitoring to guarantee their stability. Slope failure events, in which material is wasted onto rights-of-way, have historically caused infrastructure damage, operational suspensions, train accidents, and in the worst cases, loss of life. Numerous techniques, both statistical and deterministic in nature, have been developed to spatially and temporally assess unstable slope hazards and failure risk. However, these methods are reliant on the scope and accuracy of geotechnical, maintenance, and remotely-sensed datasets. ☐ The railway industry has benefited from the rise in "Big Data" analysis techniques over the last several decades using routinely collected inspection data such as right-of-way videos and track geometry measurements. Using computer science and engineering methods, the introduction of large and more varied datasets has allowed new types of analyses to be developed to optimize railroad construction, maintenance, and operational activities. This dissertation introduces a hazard assessment framework for cut-slopes in railroad rights-of-way using quantitative assessments of videos recorded during track geometry inspection runs. Using developed computer vision models, candidate hazardous sections and unstable slope features were identified to statistically analyze slope condition along a railroad section using a novel hazard assessment framework. Application of the framework indicated that right-of-way inspection videos could be used as an additional remotely-sensed data source to enhance statistical slope hazard assessments. ☐ This research aimed to expand upon previous investigations into the quantitative analysis of right-of-way videos to bolster inspection data collected during routine track geometry car runs, specifically to assess cut-slope hazards. It was hypothesized that combining slope hazard analysis results with remotely-sensed geospatial data sources will allow the study area to be assessed both locally and regionally. The hazard assessment framework aimed to utilize both data sources as inputs to spatially analyze locations along the studied track section to statistically determine those most susceptible to a failure event. Importantly, the proposed framework was dependent on only remotely-sensed data and thus the produced hazard scores do not require sending engineers to take geotechnical measurements and instead could be used to prioritize locations of interest for follow-up site investigations.
Description
Keywords
Data science, Digital image processing, Landslide hazard assessment, Machine learning, Predictive maintenance, Railroad engineering