Onboard Estimation and Classification of a Railroad Curvature

When performing tests on a rail vehicle, it is often necessary to categorize the target data according to the characteristics of the railroad plane geometry. In this paper, a method to classify and identify the railroad plane geometry is considered, employing railroad curvature readings formed by onboard sensor data. The aim is to extract the characteristics of the railroad track to identify and categorize different segments of the track. The railroad curvature is modeled as a first-order piecewise linear polynomial representing sections of straight tracks, transition curves, and circular curves along the railroad. The sensor data are preprocessed in a Global Positioning System-aided dead reckoning navigation system to debias the curvature readings. Subsequently, the noise in the curvature readings is suppressed by nonlinear filtering techniques. Furthermore, the observed curvatures are processed with a linear filter by minimizing a discounted least-squares criterion, yielding the final estimate of the railroad curvature and its rate of change, which further on are utilized to form a detector where the position of a trend change in curvature measurements is located. The result from the presented method has been compared against database values on plane geometry from Banverkets Information System-a system belonging to the Swedish Rail Administration office. The reported accuracy of detecting a change in the railroad curvature has varied in the range of ??7 m in difference between the position entries in the curvature database and the positions found with the developed method.