Abstract Various Reynolds stress turbulence model (RSM) formulations have been widely employed in computational fluid dynamics (CFD) simulations of gas cyclone separators for more than two decades, primarily because they do not rely on the Boussinesq hypothesis (isotropic eddy-viscosity assumption) and can better represent anisotropic turbulence and strongly swirling flows, which simpler eddy-viscosity turbulence models (EVMs) often struggle to capture accurately. But are RSMs truly the only Reynolds-averaged Navier-Stokes (RANS) turbulence models capable of simulating the mean flow in gas cyclone separators? Does their continued popularity result from their actual performance or from modeling bias arising from the Boussinesq (isotropic eddy-viscosity) assumption and the limitations in the discretization approaches (including grid generation and numerical schemes) adopted in previous CFD studies of EVMs? To address these questions, this work compares predictions from several EVMs (Spalart-Allmaras, standard k-epsilon, renormalization group k-epsilon, realizable k-epsilon, standard k-omega, and shear-stress transport k-omega) with and without curvature correction, against those from linear and quadratic pressure-strain RSMs and reference experimental data for a 0.29-m-diameter Stairmand gas cyclone. The results show that appropriate discretization approach enhances the capability of some non-modified EVMs to predict mean velocities in weak-curvature-effect circumstance (e.g., inside the cyclone barrel). Curvature correction improves the mean flow prediction accuracy of EVMs. In addition, caution is advised when applying RANS-predicted root-mean-square velocity fluctuations to describe cyclone flow phenomena. Finally, this study demonstrates that, for the study conditions, the realizable k-epsilon model with curvature correction can provide accuracy comparable to linear and quadratic pressure-strain RSMs in predicting the mean flow in gas cyclone separators while requiring shorter computational time, thereby challenging the long-held preference that RSMs are the only suitable RANS turbulence models for such applications.
Thongnoi et al. (Fri,) studied this question.