Ballast particle behavior under varying conditions using tri-axial inertial sensors
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Maintaining ballast condition is an important aspect of maintaining a healthy track infrastructure system. Cyclic train loading over sections of track naturally degrades the ballast which must maintain a proper layer thickness and ballast particle gradation to optimally function. Over time, repeated axle loadings deform the ballast layer and degrade the ballast particles changing the effective ballast particle gradation and thus the strength of the track substructure. This behavior is historically handled by maintenance procedures aimed at reinforcing the track structure by strengthening the ballast. ☐ Maintenance engineers benefit from detailed information about specific degradation behavior which allows them to perform timely and effective maintenance strategies such as tamping, undercutting and ballast cleaning. Historically, ideal maintenance scheduling has relied heavily on past experience, manual inspection, and field observations. Delaying maintenance past when it is required can cause degradation of track geometry, which can result in safety concerns and substantial track damage. Conversely, performing maintenance too early increases costs. Since track deterioration results in larger financial consequences, railroads tend to err on the side of caution. ☐ The research conducted herein focuses on developing a specialized ballast condition evaluation hardware and associated software to quantify ballast condition and failure criteria. SmartRocks are battery-powered wireless devices that resemble a ballast particle that are placed in track within the existing ballast layer and are capable of measuring acceleration and orientation in real time using tri-axial accelerometers and tri-axial gyroscopes. SmartRocks are installed in chosen sections of ballast, operating remotely, and transmitting recorded data to receivers installed on the field side of the track. Using SmartRock data, analysis was conducted to investigate SmartRock behavior, and thus ballast behavior, under various track conditions in an effort to correlate their behavior with known installation site information. ☐ To compare individual SmartRock measurements, the data was first subject to a basis change such that all data was in a uniform track coordinate basis. Following, extensive Exploratory Data Analysis (EDA) was conducted to identify and compare SmartRock measurement characteristics using descriptive statistics and unsupervised machine learning methods which are fully explained. Using the available SmartRock data, analysis showed that they behave differently under varying conditions of design and installation. For example, the data clearly shows that SmartRock response from passing trains differed for SmartRocks installed under-tie pads vs. those installed without under-tie pads. It is hypothesized that response data from SmartRocks allows for determinations of ballast behavior which in turn can lead to better understanding of ballast degradation and failure mechanisms. SmartRock response data is analyzed and explained, with examples showing how the data was used to make determinations about the two track sections in which SmartRocks were installed.
Description
Keywords
Data science, Machine Learning, Railroad, SmartRock, Ballast particle, Tri-axial inertial sensors, Maintenance