AN INNOVATIVE DATA-DRIVEN HYBRID METHODOLOGY FOR RELIABLE PREDICTION OF RAIL TRACK DEFORMATION

Paper in proceeding, 2026

Accurate prediction of accumulated permanent track deformation (settlement) under cyclic loading is vital for effective
track geometry maintenance and cost-efficient railway infrastructure management. This study introduces a novel approach to
forecasting long-term settlement with reduced computational demands, combining advanced numerical and data-driven methodologies.
Leveraging a comprehensive parameter study, the approach evaluates both short- and long-term track performance, integrating machine
learning techniques to develop predictive models. A machine learning algorithm was studied, with model validation performed using
field data from a ballasted track demonstrator from a transition zone. Performance metrics are used to assess model accuracy. The
random forest algorithm was selected and exhibits strong agreement with measured settlement data, demonstrating its robustness and
predictive reliability. By bridging the gap between numerical simulations and empirical observations, this hybrid methodology
enhances the prediction of both resilient and permanent track deformation, both critical parameters in design and operation. The
findings offer a cost-effective tool for railway managers, paving the way for its integration into computational decision support systems
for maintenance and renewal planning. This innovative hybrid methodology not only enhances infrastructure management practices
but also supports the sustainable and dependable operation of railway networks, ensuring both short- and long-term efficiency and
resilience.