Probabilistic Approach for Development of Track Geometry Defects as a Function of Ground Penetrating Radar Measurements

Recent research has shown a relationship between track geometry defects and track subsurface conditions as measured by Ground Penetrating Radar (GPR). This paper presents the results of a comprehensive study looking at the development of a probabilitic model for the prediction of track geometry defects as a function of key subgrade parameters as measured by GPR. Specifically, several Logistical Regression (LR) analyses were performed, to include conventional LR modeling as well as a hybrid LR modeling approach based on Hierarchical Clustering Analysis with Histogram Data. The result was a higher order polynomial Logistic Regression model for determination of the probability of a track geometry surface defect occurring at locations with measured ballast fouling and measured ballast thickness. The results showed that there was a statistically significant relationship between high rates of geometry degradation and poor subsurface condition as defined by the GPR parameters: Ballast Fouling Index (BFI) and Ballast Layer Thickness (BLT). Furthermore, a predictive model was developed to determine the probability of a high rate of geometry degradation as a function of these key GPR parameters.