The monitoring of track irregularities is essential to ensure safety, passenger comfort, and operational efficiency. These geometric irregularities arise from repeated train loads and environmental effects such as temperature-induced buckling and substructure settlement. Currently, track geometry is measured using dedicated inspection trains equipped with complex systems and specialized personnel. As a complementary approach, accelerometers installed on commercial trains offer a promising solution, enabling more frequent and cost-effective monitoring and supporting maintenance planning.

This thesis investigates railway track geometry monitoring using onboard accelerometers mounted on commercial trains. The first research axis focuses on quantifying uncertainties in the identification of vertical offset and cross-level irregularities through double integration of vertical axle-box accelerations. As the assumptions underlying this method are not fully understood, a geometric analysis and numerical simulations using railway dynamics software are conducted. A rigorous mathematical derivation of parasitic quasi-static terms interfering with the identification is presented, demonstrating their importance for the accurate assessment of long-wavelength irregularities.

The second research axis addresses lateral irregularities, whose reconstruction from onboard accelerometers is significantly more complex. In particular, the lateral alignment irregularity identification is studied. To improve understanding of this problem, simplified surrogate models incorporating key nonlinearities are developed and inverted to characterize the relationships between measurements and track geometry. Iterative estimation methods are then applied and extended to more complex railway dynamics models. Finally, a parametric study and a simulation-based refinement technique, treating the software as a black box, are shown to provide satisfactory results even under degraded identification conditions.