Benchmarking of CFD modelling of a reactor vessel auxiliary cooling system (RVACS)

Three-dimensional direct numerical simulations (DNS) and two-dimensional Reynolds-averaged Navier–Stokes (RANS) simulations using a range of modelling approaches were carried out for forced, mixed and natural convection flows in a Reactor Vessel Auxiliary Cooling System (RVACS), a key facility for passive decay heat removal in nuclear reactor systems. The forced and mixed convection cases both have a fully-developed turbulent inflow at R e τ = 180 . The mixed convection case has Grashof numbers of 1.41 × 1 0 7 and 4.44 × 1 0 6 at the cold-side and hot-side channels respectively. The natural convection case has Rayleigh numbers of 6.72 × 1 0 6 and 2.32 × 1 0 6 at the cold-side and hot-side channels respectively. The study aims to (i) reveal the fundamental flow and heat transfer physics in the RVACS, (ii) assess the performance of various RANS modelling approaches against a high-fidelity reference, and (iii) provide benchmark data and practical guidelines for future numerical studies of similar engineering applications. The DNS results identify a number of complex flow and heat transfer features, including flow separation and redevelopment around the baffle and at the sharp corner, the modification of turbulence in buoyancy-aiding and buoyancy-opposing flows, the development of buoyancy-driven boundary layers, and the formation of natural convection recirculation loops, all of which pose significant challenges for turbulence modelling in RANS simulations. The comparison between DNS and RANS simulations shows that, for the forced convection case, the best agreement with DNS is obtained by the simulations using the k − ɛ − v 2 ¯ / k model without wall functions, together with the Launder–Sharma k − ɛ simulation using the FEAT wall function and the standard k − ɛ simulation with the enhanced wall treatment (EWT). All of these well capture the response of turbulence and heat transfer to flow separation and reattachment. In the mixed convection case, with the impact of buoyancy introduced, the performance of the k − ɛ − v 2 ¯ / k without wall functions deteriorate. In the natural convection case, a clear difference in performance is observed in general between the simulations that resolve the near-wall region and those using wall functions, indicating that accurate resolution of the near-wall flow, in particular of the buoyancy-driven boundary layer, is of critical importance for capturing the overall flow and heat transfer behaviour in this type of flow.