Autonomous Rail Surface Defect Detection
Rail surface defects are a major type of rail defect which progressively propagate with the accumulation of tonnage. It is reported that around 90% of railway accidents have rail surface defects, as either a direct or indirect factor. Historically, railway tracks are inspected by trained personnel. However, manual inspection has low efficiency and low accuracy because it is heavily dependent on the experience of the inspectors. Many automatic track inspection systems have been developed over time, usually mounted on an inspection car or a hi-rail vehicle with various types of sensors. Those systems are mainly based on laser, acoustic emission, LiDAR, ultrasonic wave, and ground penetration radar (GPR) technologies, which are effective in identifying rail internal defects, hollow timber ties, fouled ballast, and drainage problems. However, those systems have limited effectiveness in detecting and quantifying the rail surface defects. Visual inspection systems using the image processing algorithms and deep learning-based object detection methods have been introduced for the rail surface inspection over the past decade, which typically use images taken by cameras mounted on rail inspection vehicles. The models are usually developed based on images taken with a consistent angle and good contrast between the rail surface and the track. However, these approaches cannot handle arbitrary-oriented rail surface in images. Recently, using unmanned aerial vehicle (UAV)-based cameras has drawn great attention due to its convenience. More importantly, using UAVs to acquire rail surface images does not require track time and does not disturb normal train operations. The hardware cost is also much lower compared to the previous vehicle mounted systems. However, issues associated with consistency of image quality of track due to environmental and drone operations are the current barriers to taking full advantage of the UAV benefits. The objective of the proposed research is to develop an automatic rail surface defect detection system based on machine learning and convolution networks that is suitable for UAV implementation. This research addresses questions pertaining to the consistency and quality of images acquired by UAV-based cameras during flight and will focus on mitigating effects of: (1) vibrations and other operating/environmental conditions, (2) variations of track width, location, and orientation, and (3) shadows, rail reflectivity, and sunlight intensity. Furthermore, the proposed system will be integrated to the proposed Intelligent Aerial Drones for Traversability Assessment of Railroad Tracks project.
Language
- English
Project
- Status: Active
- Funding: $78551
-
Contract Numbers:
69A3552348340
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590University of South Carolina, Columbia
502 Byrnes Building
Columbia, SC United States 29208 -
Managing Organizations:
University of South Carolina, Columbia
502 Byrnes Building
Columbia, SC United States 29208 -
Project Managers:
Stearns, Amy
-
Performing Organizations:
University of South Carolina, Columbia
502 Byrnes Building
Columbia, SC United States 29208 -
Principal Investigators:
Qian, Yu
Vitzilaios, Nikolaos
- Start Date: 20230601
- Expected Completion Date: 20240831
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers Program
Subject/Index Terms
- TRT Terms: Drones; Flaw detection; Image analysis; Machine learning; Railroad rails; Railroad safety; Railroad tracks
- Subject Areas: Maintenance and Preservation; Railroads;
Filing Info
- Accession Number: 01897742
- Record Type: Research project
- Source Agency: University Transportation Center for Railway Safety
- Contract Numbers: 69A3552348340
- Files: UTC, RIP
- Created Date: Oct 28 2023 8:03PM