Award Date

5-1-2022

Degree Type

Thesis

Degree Name

Master of Science in Engineering (MSE)

Department

Electrical and Computer Engineering

First Committee Member

Yingtao Jiang

Second Committee Member

Mei Yang

Third Committee Member

Ming Zhu

Fourth Committee Member

Hualiang Teng

Number of Pages

73

Abstract

Rail tracks need to be consistently monitored and inspected for problems associated with rust, deformation, and cracks that, at their worst, can cause catastrophic train derailments. Many non-destructive testing approaches have been explored and extensively utilized to help inspect rails’ health, but most of them require intensive human power and/or heavy sensor systems (e.g. total stations, manual/car-mounted trolly, etc.) that are not efficient or convenient to cover a long range of rails and may interfere with the normal operation of trains.In light of the rapid development of unmanned aerial systems/vehicles (UAS’s/UAVs) and high definition photographic and optical distance measuring sensors, this paper proposes a novel UAV-based rail track irregularity monitoring and measuring platform that can remotely inspect the geometry irregularity of tracks at various angles and cover a long distance by only a few personnel. By mounting a light distance and range (LiDAR) scanning sensor and a data acquisition system on the UAV, we can continuously collect 3D point cloud data (PCD) frames that reflect the surfaces of tracks, ground, and other objects. Data points in these PCD frames are manually annotated into two classes: rail tracks and background. Then, annotated PCD frames are pre-processed and fed to train a state-of-the-art machine-learning-based 3D point cloud semantic segmentation network, RandLA-Net, to assign each point into one of the two aforementioned classes, so that point clusters that represent rail tracks can be extracted. The trained model can be deployed for real-time distinction between rails and background. Then, principal component analysis (PCA) and multiple regressions are conducted to identify the top and inner surface of the rails. In the end, various geometry measurement of rails, such as gauge, cross level, etc. can be performed to inspect any irregularities. The geometry measurement obtained by the proposed UAV-LiDAR-based framework is compared against standard official value of each geometry. The evaluation results have confirmed the similar or the more advanced performance of the proposed platform with more terrain flexibilities.

Controlled Subject

Artificial intelligence;Geometry;Optical radar;

Disciplines

Electrical and Computer Engineering

File Format

pdf

File Size

4100 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/


COinS