Uncertainty quantification in SST k-ω and standard k-ϵ turbulence models for aerodynamic simulations of high-Speed trains under crosswind conditions

Accurate prediction of the aerodynamic loads on high-speed trains (HSTs) under crosswind conditions is essential for reliable safety assessment. However, the empirical closure coefficients embedded in Reynolds-averaged Navier-Stokes (RANS) turbulence models introduce parametric uncertainty that may propagate substantially into aerodynamic predictions. This study presents a controlled comparative analysis of closure-coefficient uncertainty in the Standard k-ϵ and Shear Stress Transport (SST) k-ω models under an identical numerical framework. Latin hypercube sampling and non-intrusive polynomial chaos are employed for uncertainty propagation, together with total-order Sobol indices and correlation analysis to characterise global and spatial parameter sensitivities. The results show that the lateral force coefficient (Cy) is particularly sensitive to closure-coefficient variations, with UQ2 values of 43.95% for the Standard k-ϵ model and 50.78% for the SST k-ω model. Although the SST k-ω model exhibits more localised pressure uncertainty in several separated-flow regions, it does not consistently yield lower uncertainty in the integrated aerodynamic loads. The two models also display fundamentally different sensitivity patterns. The Standard k-ϵ model is strongly dominated by Cε2, which contributes a total-order Sobol index of 0.9835 to Cy, whereas the SST k-ω model exhibits distributed multi-parameter sensitivity, with σω1 providing the largest contribution of 0.5100. These parameters are also strongly correlated with Cy, with Pearson correlation coefficients of −0.7768 and −0.7089, respectively. The results demonstrate that turbulence-model calibration for HST crosswind aerodynamics should account for model-dependent parameter sensitivity and provide a quantitative basis for uncertainty-aware model selection and calibration.