Evaluation of Turbulence Models for Accurate Prediction of Airflow Structure in a Subject-Specific Upper Airway Geometry

Abstract

Accurate prediction of complex airflow fields in the human upper airway is crucial for precise quantifications of aerosol transport and the optimization of inhalation therapy. While large eddy simulation (LES) has been widely validated for capturing more detailed turbulence features, its high computational cost limits widespread applications. Reynolds-averaged Navier-Stokes (RANS) models are more computationally efficient but cannot resolve key flow features (i.e., the random fluctuation terms) in anatomically complex airway geometries. Recently, hybrid models such as stress-blended eddy simulation (SBES) have emerged as promising alternatives, yet their performance in airways remains underexplored. This study is a research effort to address the knowledge gap by evaluating the computational accuracy and efficiency of LES, RANS, and SBES in simulating the airflow field through a subject-specific mouth-to-trachea geometry. The experimentally validated in-house LES results served as the benchmark for assessing the RANS and SBES models in predicting velocity distributions, secondary vortex structures, and turbulent kinetic energy (TKE). Compared with LES results, the realizable k-ε model significantly underpredicted secondary flow structures and TKE due to excessive numerical diffusion and inherent model limitations, although it completed simulations in less than 1% of the time required by LES. The SBES model demonstrated excellent agreement with LES in capturing localized flow features. Computational efficiency-wise, although SBES required roughly 25% more overall CPU time than LES, SBES achieved a reduction in average CPU time per iteration of approximately 7% compared to LES.

Keywords: Reynolds-averaged Navier–Stokes (RANS); computational fluid dynamics (CFD); large eddy simulation (LES); mouth-to-trachea; pulmonary airflow; stress-blended eddy simulation (SBES).