On the development of a predictive rail meintenance planning methodology utilizing parametric Weibull prediction methods
Date
2020
Authors
Cronin, John
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The railroad industry has used for the past 50 years the 2-Parameter Weibull equation to determine the rate of rail fatigue defect occurrences and to forecast the fatigue life of railroad rail. With the advent of more powerful computers, more frequent data collection and new techniques to analyze data, a new field of data analysis has been created, Data Analytics, sometimes referred to as “Big Data”. This thesis makes use of this new area of Data Analytics to develop and implement an improved rail defect forecasting approach building upon the traditional Weibull equation to overcome many of its limitations and problems. ☐ Because of the serious nature of broken rail defects and the approximately 100 broken rail derailments that occur in the US each year, railroads continue to improve rail inspection techniques, rail maintenance techniques and its rail defect data collection process. The railways industry currently collects data on the occurrence of defects, rail inspection results, rail maintenance techniques such as rail grinding and a broad range of rail statistics, which have the potential to provide increased insight into the rate of occurrence of rail defects. The Weibull Equation, while giving a basic forecast capability, does not explicitly account for many of the key operating and maintenance variables that affect the development of rail defects. As such, using traditional Weibull analysis techniques, it is not possible to predict what the effects of differing maintenance operating or material parameters would be on the rate of defects development. Thus, while the current use of the 2-Parameter Weibull equation is adequate for its current limited use in rail life forecasting, it appears to be possible to improve on the this rail life forecasting and prediction of defects through the use of new Data Analytic techniques which make more aggressive use of the extensive rail defect data available. These improvements can lead to a more accurate prediction method with practical implications in rail life forecasting, maintenance management and replacement planning. ☐ This thesis presents such a series of Data Analytic applications to include: Machine Learning methods such as K-Nearest Neighbors, and Logistic Regression, as well as methods used to deal with unbalanced data1, such as bootstrapping and over-/under-sampling, and a novel method developed from Parametric Bootstrapping. This latter approach is designed to provide for an application of the Parametric Bootstrapping modified Weibull forecasting to rail segments with insufficient numbers of defects to allow for the traditional Weibull forecasting analysis. Thus, the Bootstrapping method provides reasonable estimates of the rate of defects for track segments that have little or no prior defect data, which allows far more track to be analyzed, and to be accounted for in maintenance planning efforts. In addition, there is a range of values used in the prediction, allowing for an estimate of “best case” and “worst case” scenario. This approach results in an ability to forecast the probability of rail defect occurrence as a function of cumulative tonnage experienced by the rail as well as other key track and traffic parameters that affect the development of fatigue defects. ☐ The results presented here show that the Parametric Bootstrapping Weibull Analysis approach offers a more accurate and effective approach to determining the probability of developing future defects with an overall benefit to the railroads in their maintenance of an expensive rail asset.
Description
Keywords
Parametric bootstrapped Weibull, Predictive rail maintenance, Railroad maintenance planning, Weibull prediction methods