Uncertainty quantification of the standard k-ε turbulence model closure coefficients in predicting aerodynamics of high-speed train

• Published in: Engineering applications of computational fluid mechanics 2024-12, Vol.18 (1)
• Main Authors: Liu, Hongkang, Kong, Zhiqiang, Li, Guoshun, Chen, Can, Zhao, Yatian, Zhang, Shishang
• Format: Article
• Language: English
• Subjects: aerodynamic; Aerodynamic forces; Aerodynamics; Drag coefficients; High speed rail; High-speed train; K-epsilon turbulence model; K-ε model; Modelling; Polynomials; Pressure head; Regression models; Reynolds averaged Navier-Stokes method; Sensitivity analysis; Spoilers; Turbulence models; Turbulent flow; Uncertainty analysis; uncertainty quantification; Undercarriages
• ISSN: 1994-2060; 1997-003X
• DOI: 10.1080/19942060.2024.2430658

Turbulence modelling is crucial in predicting train aerodynamics while its performance is closely related to the model closure coefficients. To reveal the influence, uncertainty quantification and sensitivity analysis are conducted on the closure coefficients of the turbulence model for high-speed train. The standard k-ϵ model for Reynolds-averaged Navier-Stokes simulation is utilized. Five main closure coefficients are analyzed using the Non-Intrusive Polynomial Chaos method to assess their uncertainty and sensitivity with respect to the aerodynamic force and pressure field. The results indicate that the drag exhibits higher sensitivity to the model coefficients with an uncertainty interval as high as 53%, especially for the tail car. The pressure exhibits significant uncertainty in the separation region around the head car cab and bogie, as well as in the wake flow region of the tail car. With sensitivity analysis, the drag coefficient C d is highly uncertain due to the prominent contribution of C ϵ2 , which accounts for 68.76% of the uncertainty. Additionally, C ϵ2 and σ k are key factors affecting pressure uncertainty, each dominating distinct areas. Finally, the linear regression identifies a strong positive correlation between C ϵ 2 and the drag, which could provide guidance for improving the modelling performance by optimizing the turbulence model coefficients.