Search

The objective of this study is to assess the predictive capabilities of low and high-fidelity turbulence models for low-Prflows. For this purpose, predictions by two-equation (k-ε) Reynolds-Averaged Navier Stokes (RANS), partially-averaged Navier Stokes (PANS) hybrid RANS/Large Eddy Simulation (LES), and DSM, WALE, filtered LES models are compared for four different test cases, namely vertical channel flow, vertical backward facing step, flow over a rod bundle and heat transfer in ascending and descending flow through a pipe with a constant wall heat flux. The test cases involve a range of complex flow conditions including separation/reattachment and aiding and opposing buoyant forcing (Re ranging from 640 to 40K; Ri ranging from −0.65 to 0.65) for water (Pr = 0.71) and liquid metals (Pr = 0.00585 to 0.025) flows. The validation study demonstrates that turbulence models are 4% more accurate for higher Prflows that for low-Pr flows; 6% more accurate for forced convective conditions than for flows involving mixed convective conditions; and predict aiding buoyant flow conditions better than the opposing buoyant flow conditions. Overall, LES performed the best and provided averaged error of 6% followed by 10% by PANS and RANS showed the largest error of 14%.

The predictive capability of RANS, hybrid RANS/LES and LES turbulence models are assessed for low-Pr flows involving complex flow features expected in engineering applications, such as forced convection, buoyancy, shear, reattachment, and recirculation. The models are tested for three test case vertical channel, vertical backward facing step with heated wall and rod bundle lattice cases for Pr ranging from 0.002 to 2.0. k-ε based model is used for RANS, partially averaged Navier Stokes (PANS) for hybrid RANS/LES and dynamic Smagorinsky model for LES. The channel and backward facing step cases show that LES performs the best followed by PANS and RANS. The RANS model performs reasonably well for flows with strong buoyant forces, which augments the mean flow more than the turbulent features. But for moderate buoyant flow condition, which augments turbulence characteristics and associated diffusion, RANS models do not perform well. Considering both the computational cost and accuracy, the PANS provides a good compromise between RANS and LES.