Professor Ni Attoh-Okine and Ahmed Lasisi, University of Delaware
Track geometry data is often combined into a single parameter index referred to as a Track Quality Index or TQI. TQIs exhibit classical big data attributes: value, volume, velocity, veracity and variety and are used to obtain average-based assessment of track segments and schedule track maintenance. Using track geometry data from a sample mile track, this activity examines how to combine track geometry parameters into a low dimensional form (TQI) that simplifies the track properties without losing much variability in the data. This led to a principal component analysis approach, with a two-phase approach used to validate the use of principal components. First phase was to identify a classic machine learning technique that works well with track geometry data. The second step was to train the identified machine learning technique on the sample mile-track data using combined TQIs and principal components as defect predictors. The performance of the predictors was compared using true and false positive rates. The results show that three principal components were better at predicting defects and revealing salient characteristics in track geometry data than combined TQIs even though there were some correlations that are potentially useful for track maintenance.
Implementation of Research Outcomes
This report shows the potential benefits of using a principal component analysis approach to analyze railroad infrastructure data collected over time and to develop a single parameter TQI based on this approach. This research thus allows for a new approach to track geometry analysis modeling which will be implemented as part of ongoing UTC program activity.
Impacts/Benefits of Implementation
This approach has not been implemented to date in the form of a maintenance planning model but it offers significant potential in improved track geometry maintenance modeling, planning and implementation.