Sensitivity and fragility analysis of a railway temperature prediction model

Abstract

Rail temperature prediction is essential for railway safety, as extreme temperatures can lead to track buckling and structural instability. This study presents a comprehensive sensitivity analysis and probabilistic assessment of a validated physical rail temperature model implemented in MATLAB and integrated with the UQLab framework. Two sensitivity analysis approaches were employed: a global analysis treating all parameters as random variables, and a data-informed analysis using field measurements for weather variables. Results identify the convection coefficient as the most influential parameter when all variables are uncertain, while the solar absorption coefficient and area become critical when weather parameters are measured. Material properties such as density, specific heat, and volume exhibit negligible influence on predictions. Monte Carlo simulations comprising over 3.3 million scenarios were conducted for reliability assessment to calculate the probabilities of rail temperatures exceeding critical thresholds based on ambient air temperatures. Two approaches were employed: a hybrid approach using measured weather data and a synthetic approach with statistically generated weather scenarios. The synthetic approach enables probability estimation for extreme conditions beyond the measured dataset. Literature indicates that rail thermal buckling may occur at temperatures around 55 ∘C to 60 ∘C under certain track conditions. At 40 ∘C ambient temperature, the synthetic approach estimates 10 % and 0.6 % probabilities for rail temperatures reaching or exceeding 55 ∘C and 60 ∘C, respectively, while the hybrid approach yields 51 % and approximately 0.6 % for the same thresholds. This discrepancy highlights the importance of comprehensive scenario coverage for robust probability estimation.