Comparison of partially averaged Navier-Stokes and large eddy simulation of the aerodynamic behaviors of a generic high-speed train

Abstract

This paper investigates the influence of numerical methods and mesh resolution on the prediction accuracy of the aerodynamic behaviors of a 1/20 scaled generic high-speed train (HST) model. A thorough comparison is made between partially averaged Navier-Stokes (PANS), large eddy simulation (LES), and wind tunnel experiments, covering aerodynamic forces, surface pressure, velocity distribution, and Reynolds stress and turbulent kinetic energy in the wake region. The Reynolds number for both simulations and experiments is set to 4.75×105. The results show that the PANS approach accurately predicts flow characteristics observed in experiments and fine LES calculations, even with a low-resolution grid. PANS exhibits a distinct advantage over LES when grid resolutions are insufficient for resolving near-wall flow structures around the HST, both in open-air conditions and crosswind environments. Additionally, grid refinement improves the predictive accuracy of the HST's aerodynamic performance, particularly in the presence of small yaw angle.

1. Introduction

With the rapid advancement of computer technology, computational fluid dynamics (CFD) has become widely employed due to its ability to efficiently compute and visualize the flow field around trains. In the numerical prediction of the flow around high-speed trains (HSTs), characterized by highly unsteady separated flow, traditional Reynolds-averaged Navier-Stokes (RANS) method is found to yield inaccurate results. This is attributed to its modeling of all flow scales with one-point closures. Large eddy simulation (LES) has proven to be effective in capturing a broad spectrum of turbulent scales, offering high accuracy in predicting turbulent flow around bluff bodies [1, 2]. Despite significant strides in computing resources, achieving precise LES predictions for detailed vehicles, especially at realistic or higher Reynolds numbers, remains challenging and expensive.

Considering these challenges, it becomes imperative to explore appropriate hybrid techniques to address different regions of bluff body flows: from the growth of the boundary layer to its separation, the formation of shear layers, and the development of wakes. The detached-eddy simulation (DES), initially proposed as a zonal approach by SPALART et al [3], represents a hybrid method that employs RANS to simulate the boundary layer of an HST and utilizes LES to capture the large-scale turbulent flow away from the HST. This approach has demonstrated success in numerically investigating various aspects of train aerodynamic behavior, including underbody flow simulation [4 −6], side flow (slipstream) assessment [7 −10], wake flow analysis [11 −13], safety evaluation in crosswind condition [14 −16] and aeroacoustics prediction [17−19]. Nevertheless, in the complex flow around a high-speed train, a grey area exists between RANS and LES approaches. This issue arises primarily from the grid-dependent nature of DES, where inadequate resolution in the transition region disrupts the intended shift from RANS to LES. Insufficient grid refinement may prevent the smooth resolution of large-scale turbulent structures, forcing excessive reliance on RANS modeling, while overly dense grids undermine the computational efficiency central to hybrid methods. Such unresolved interactions within the grey area compromise the accurate representation of turbulent energy transfer.