Dealing with the effects of turbulence is the main challenge for the numerical study of flows relevant to engineering applications, due to the tendency of turbulent fluctuations to span large ranges of time and length scales, which makes them costly to reproduce in terms of computational resources because large ranges of scales require fine discretization resolutions. In a research context, where the accuracy of the solution is usually prioritized, it is not uncommon for this demand to be abided by using a direct numerical simulation (DNS) approach, leading to an ideally exact reproduction of the flow. However, in an industrial context it would be impractical to assume such costs, and so alternative approaches with more moderate resolutions are adopted by avoiding the explicit representation of some turbulent features, instead modelling their effects on the resolved part of the flow.
RANS models are the dominant solution in industrial applications, as they allow for the least expensive -and thus quickest- simulations, by returning solely the behaviour of the mean flow, which is typically adequate for most purposes. Large eddy simulation (LES) approaches also include large scale turbulent fluctuations in the solution and they offer better accuracy overall, but they have been mostly relegated to niche applications in the past due to the associated increase in computational cost with respect to RANS models. However, LES is becoming increasingly relevant due to the rapid development of the state of the art in high performance computing and of LES models themselves.
The purpose of the present work is to test a new LES model with the simulation of the flow around a rectangular cylinder with a chord-thickness ratio of c/D = 5, commonly known as BARC -benchmark on the aerodynamics of a rectangular cylinder-. The diversity of turbulent phenomena that characterize this case of study, notwithstanding the simple geometry of the solid body, provides multiple aspects on which to assess the performance of the new model. The flow is simulated for the well-established dynamic Smagorinsky scheme, for the new model -which introduces a new tensorial, non-linear approach to eddy viscosity-, and for the implicit LES approach -forgoing the direct modeling of turbulence-. All simulations are otherwise performed under the same configuration, which is designed to suit the characteristics of the LES approach. The software employed for the simulations is T-Flows, an open-source code based on the finite volume method. Turbulent statistics are gathered from the solutions to observe how the different models reproduce the main turbulent features of the flow, in order to draw comparison among themselves and with previously existing high-accuracy DNS data.