Numerical simulations of developing flow and vortex street in a rectangular channel with a cylindrical core

Three-dimensional, unsteady simulations of developing turbulent flows in a rectangular channel containing a cylindrical rod have been performed to investigate their sensitivity to the choices of boundary conditions and turbulence models. Among all methods, large eddy simulations, employed in a downstream sub-domain of the channel as part of the segregated hybrid model, reproduced most accurately the experimental results. However, unsteady Reynolds-averaged Navier-Stokes (URANS) simulations with a Reynolds stress model appear to be also an acceptable choice for approximate rod bundle analyses, making fairly accurate predictions at a much lower computational cost. In agreement with previous findings, steady RANS simulations are not recommended as a low-cost substitute of URANS for flows in tightly packed rod bundles. The URANS simulations were found to be insensitive to inlet turbulence specification and to be more accurate when a uniform inlet velocity was specified rather than a fully developed inlet velocity distribution. Developing flow simulations were found to be preferable to simulations with a streamwise-periodic boundary condition. Finally, unsteady inviscid (Euler) simulations with a fully developed initial velocity distribution predicted the onset of gap instability, but were otherwise found to be unsuitable for practical analysis of rod bundle flows.