Detecting lateral track irregularities by onboard measurements of lateral acceleration and displacements and Machine Learning algorithms // Rilievo delle irregolarità laterali del binario attraverso misure di accelerazioni laterali e spostamenti da bordo treno e algoritmi di Machine Learning

Understanding the development of track defects, which can have a significant impact on the safety of train operations, is one of the most crucial responsibilities for railway engineers. Analysing track geometry (TG) parameters is necessary to establish an efficient track condition monitoring approach. Research has recently focused on monitoring track geometry irregularities using data collected from trains in service. Most of the previous research has focused on vertical irregularity of track, since the lateral irregularity is considerably more challenging because of the complex relative wheel-rail motion. This research work focuses on finding the relationship between the lateral acceleration and lateral irregularities from the Lateral Displacement of the Wheel relative to the Rail (LDWR). A supervised Machine Learning (ML) model is trained and tested with a dataset composed of numerical simulation on a diverse set of operational conditions. Among the tested algorithms, Random Forest regression shows the best result and polynomial regression is on the second place. Sufficient results are achieved which shows that it is possible to detect the lateral irregularities with on board computer vision sensor which can detect the LDWR. Also, a sensitivity analysis is performed which shows that lateral displacement is the most important parameter in determining the lateral irregularity from vehicle monitoring. The result of this work could be a step toward the predictive maintenance as well.