Are maintenance practices for railroad tracks effective?

• 成果类型: Article
• 署名作者: Merrick, JRW; Soyer, R; Mazzuchi, TA
• 署名单位: Virginia Commonwealth University; George Washington University; George Washington University
• 刊物名称: JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
• ISSN/ISSBN: 0162-1459
• DOI: 10.1198/016214504000002104
• 发表日期: 2005
• 页码: 17-25
• 关键词: dirichlet; inference; mixtures; models

摘要

The Association of American Railroads wished to determine the effect of a maintenance practice known as grinding on the occurrence of rail fatigue defects and on the subsequent total traffic usage before a track must be replaced. Because a designed experiment was not practical, an analysis of historical data from the Canadian Northern Railroad is presented. In the analysis, certain covariate data are available, specifically the amount of grinding and some physical characteristics of the rail; other important covariate data are not available, however. A model for the number of defects as a function of traffic usage is developed based on a modulated Poisson point process. The model incorporates the effect of the available covariates and a mixture of Dirichlet processes set-up for the scale parameters of the individual rail sections that allows an assessment of the overall effect of the unavailable covariates. The model is then used to determine an optimal replacement period for a whole rail track. The analysis demonstrates that grinding reduces the expected number of defects and increases the optimal replacement interval.