A statistical evaluation of multiple regression models for contact dynamics in rail vehicles using roller rig data

A statistical analysis of a large amount of data from experiments conducted on the Virginia Tech-Federal Railroad Administration (VT-FRA) roller rig under various field-emulated conditions is performed to develop multiple regression models for longitudinal and lateral tractions. The experiment-based models are intended to be an alternative to the classical wheel-rail contact models that have been available for decades. The VT-FRA roller rig data is used to develop parametric regression models that efficiently capture the relationship between traction and the combined effects of the influential variables. Single regression models for representing the individual effect of wheel load, creepage, and angle of attack on longitudinal and lateral traction were investigated by the authors in an earlier study. This study extends single regression models to multiple regression models and assesses the interaction among the variables using model selection approaches. The multiple-regression models are then compared with CONTACT, a well-known modelling tool for contact dynamics, in terms of prediction accuracy. The predictions made by both CONTACT and multiple regression models for longitudinal and lateral tractions are in close agreement with the measured data on the VT-FRA roller rig. The multiple regression model, however, offers an algebraic expression that can be solved far more efficiently than a simulation run in CONTACT for a new dynamic condition. The results of the study further indicate that the established multiple regression models are an effective means for studying the effect of multiple parameters such as wheel load, creepage, and angle of attack on longitudinal and lateral tractions. Such data-driven parametric models provide an essential analysis and engineering tool in contact dynamics, just as they have in many other areas of science and engineering.