The development of rail defects due to the presence of geometry defects in Class 1 Railroads

The purpose of this thesis is to develop a relationship between the development of a rail defect and the presence of a geometry defect prior to the occurrence of the rail defect. This was done with the use of a Class 1 Railroad's geometry and rail defect data for their whole 22,000 mile system over a five year period. Regression analyses were performed using multilinear regression and multivariate adaptive regression splines (MARS) that showed that the occurrence of a geometry exception or defect resulted in approximately 30% loss in rail life. A Bayesian Network was then developed with the use of NETICA. This network showed that the presence of a single geometry defect increased the likelihood of a rail defect by approximately seven, with multiple geometry defects increasing this likelihood even more. These findings show that geometry defects do have an impact on the development of a rail defect.