Development of a multi-dimensional time-based track safety and quality index (TSQI)
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
With the increasing demand for railway transportation, there is a need for infrastructure managers to develop plans and strategies to meet the new requirements and ensure high service quality and capacity. Reliable transportation networks rely on effective maintenance activities, and railways have been using analytic and autonomous track geometry assessments to improve safety and maintenance decision-making. How- ever, traditional techniques based on discrete threshold processing and track quality indices have limitations in capturing the multifaceted aspects of track safety and quality and applying advanced analytics to their fullest potential. Therefore, there is a need to enhance these techniques to leverage better the vast amount of data collected by inspection cars and make more informed decisions regarding safety and maintenance. ☐ This research focuses on developing a 3D track quality index for assessing and monitoring the condition of railway tracks. By incorporating machine learning techniques, and a multivariable normal distribution, the index captures the temporal dependencies and uncertainties associated with track parameters. ☐ The inclusion of a normal distribution layer allows for modeling the uncertainty in track parameters, providing insights into the range of potential values and associated uncertainties. Separate models are trained for each track parameter channel, accommodating their distinct characteristics and variability. This enables a more accurate distribution prediction for each specific channel, leading to a comprehensive understanding of the track’s condition. ☐ Utilizing multiple models and their corresponding output distributions offers valuable insights into the variation and uncertainty associated with track parameters. These distributions are combined into a multivariable normal distribution, capturing the joint behavior and interdependencies of the track quality parameters. Statistical measures such as confidence intervals, percentiles, and probabilities can be derived from this distribution, allowing for a nuanced assessment of track quality. ☐ The developed 3D track quality index incorporates probabilities of exceeding predefined safety limits for the mid-cord offset of the track geometry data set by the Federal Railroad Administration (FRA). A comprehensive assessment of the track's condition is obtained by calculating the probability of exceedance for each measured parameter based on the multivariable normal distribution. The index provides an overall measure of the likelihood of the track quality exceeding specified thresholds for multiple parameters simultaneously. ☐ The effectiveness of the 3D track quality index is demonstrated through its application to eight segments of railway tracks. Visual representations, such as 3D plots, facilitate the identification of segments with varying levels of irregularity and aid in prioritizing maintenance efforts. The index offers a customizable approach by assigning weights to each Parameter, allowing for the prioritization of specific parameters based on shareholders' needs. ☐ In conclusion, developing the 3D track quality index provides a valuable tool for assessing and monitoring the condition of railway tracks. It integrates machine learning techniques, probability analysis, and customization options to offer a comprehensive and customizable approach to track evaluation. The index enhances safety, optimizes maintenance practices, and contributes to the reliable operation of railway infrastructure.
Description
Keywords
Railway transportation, Machine learning techniques, Track parameters, Railway tracks, 3D track quality index