Palese, Joseph W.2022-09-262022-09-262019https://udspace.udel.edu/handle/19716/31413Steel rail is one of the railways most vital track assets and is critical for safe and efficient operations. Rail fails due to loss of section (through various wear mechanisms) as well as fatigue. Maintaining the structural integrity of the rail is paramount and railways have implemented sophisticated inspection vehicles to monitor track components, including measuring the transverse profile of the rail. The resulting data is extensive and proliferate for many railways. Traditionally, railways have used simple threshold analyses of this extensive data to make maintenance decisions. ☐ Rail wear has been the subject of extensive research for the past century. Past research has focused on laboratory testing (along with field-testing) to evaluate metallurgical performance, benefits of lubrication, and basic wear relationships as a function of contact (creep, sliding, etc.). This has resulted in wear coefficients that have been employed in a linear relationship with load. The results have been deterministic/mechanistic and empirical models that provide valuable information with respect to one or more factors that affect wear. The past work has included several simulation modeling approaches, with only limited work performed with respect to modeling rail wear for field conditions. This is due to the fact that many factors affect rail wear in track, and many of these are not, or cannot, be measured. Thus, modeling attempts have resulted in only modest levels of accuracy and practical implementation. ☐ The research conducted herein focused on a stochastic approach to rail wear modeling, using only the data that was readily available. In this manner a probabilistic forecast of rail wear was developed resulting in a range of outputs that are conditionally dependent upon the known inputs. Specifically, Auto Regressive Integrated Moving Average (ARIMA) and Mixture Density Networks (MDNs) were used with a Laplace distribution to understand rail wear relationships, and predict rail wear rates to forecast rail wear. In addition, a stochastic classification scheme was developed, taking advantage of the Laplace cumulative probability function, to assess the wear performance of a rail segment, given the inputs and historic wear relationships for the given dataset. ☐ An extensive Exploratory Data Analysis (EDA) was performed, resulting in a preprocessing phase of the data to determine wear rate, which was then subjected to a secondary EDA. The stochastic methodologies developed and enhanced are fully explained, along with the framework that was developed to bundle the methodologies into a comprehensive model for rail wear analysis. ☐ An application of the model was performed for the dataset provided, 277 miles of railway with multiple inspections over 6 years (more than 2 million rail profiles and corresponding wear measurements), and the results discussed in detail, particularly from a practical implementation perspective. The research resulted in a framework that allows for structural health monitoring of the rail component of the track structure, with regard to rail wear.Mixture density networksSteel railAuto regressive integrated moving averageThesis1345671533https://doi.org/10.58088/c1b6-9w872022-08-11en