Data-driven analysis of dynamic amplification for ballast-less railway track: using a triple-beam model and XGBoost.

This paper presents a data-driven approach for modelling ballast-less railway tracks using a triple-beam model and delineating effective parameters influencing dynamic amplification factor (DAF). An analytical–numerical approach is employed to solve the triple-beam model subjected to a concentrated moving load with constant velocity. This approach represents rails, concrete slab and concrete base as beams, while also characterizing visco-elastic connections between beams and incorporating a visco-elastic model for foundation. Experimental tests are used to validate the established triple-beam model. The DAF values determined based on the extreme deflections are integrated with machine learning model, i.e. XGBoost, to identify influential parameters. In general, parametric variations of the visco-elastic connection layers, including rail fasteners, cement-emulsified asphalt mortar (CEAM) layer, and subgrade, indicate that an increase in spring stiffness results in a higher DAF. Moreover, considering the bending stiffness of the three specified beams, adopting different structural thicknesses for the concrete beams leads to negligible variations in the DAF when damping is taken into account. The results of the machine learning models indicate that the damping characteristics of the CEAM layer and subgrade are the most significant parameters influencing the DAF. Furthermore, it is demonstrated that the stiffness values of these layers provide an additional contribution to regulating the DAF.