Development of train parameters estimation algorithms to improve the performance of predictive maintenance systems in railways

The proposed work is aimed at developing accurate train parameters estimation approaches, to get more informations on the vehicle running on the rail, spacing from data about the vertical load (Weigh in Motion systems) to ones on the moving train. More focused, the provided vehicle data concern its speed, vertical axle loads, direction of travel, distance between axles and timetable of the crossing axles to ensure functionalities of dynamical estimations of running loads and aims of train detection. This study allows the formulation of approaches to estimate the quantities above mentioned, starting from the same measurement layout. The developed approaches are flexible against sensors of different typology as load cell, strain gauge or the more efficient Fiber Bragg grating sensors. An analysis of robustness has been involved, concerning the estimation accuracy as a function of the performance of the measurement/acquisition chain, in terms on noise affecting the measure. The performance have been verified in a wide range of speed and vehicle mass showing good results. A comparison between two solutions that involve functionalities of train detection has been done, both on aspects of performance and on the required computational times. The approaches have been stressed, reproducing operating conditions worse than those of a real measurement scenario. A focus has been done on the developed Weigh in Motion algorithm, able to estimate dynamical loads and vehicle centre of mass, for purpose of unbalance loads detection. In order to test the algorithms in the most operating conditions concerning both the vehicle and the measurement chain, in absence of experimental data, simulated track inputs are available thanks a physical model of the infrastructure, composed by vehicle, track and a global contact model that manages their interaction.