This work presents Large Eddy Simulations of the flow and heat transfer characteristics in a matrix of surface mounted cubes, studying the effect of the grid resolution and the sub-grid scale modeling. Three sub-grid scale models, implemented in an unstructured, finite volume, commercial solver, are compared on four different grids in terms of first and second order statistical quantities. A classical Dynamic Smagorinsky model is compared with a no-model, Implicit Large Eddy Simulation, approach and a recently developed, two parameters, dynamic-mixed model. A general lack of sensitivity to the sub-grid scale model is evidenced for the flow quantities at all the resolutions, but the grid design emerges as the most determining factor for this kind of flows, showing that accurate results are possible with very coarse resolutions. In contrast, heat transfer characteristics show a strong dependence on both the grid and the sub-grid scale model with a lack of clear convergence in the investigated range of scales. The dynamic-mixed model, which for the first time is tested in a heat-transfer application, is found stable on strongly stretched grids and cheaper than the classical Dynamic Smagorinsky model due to its specific finite volume formulation, showing its suitability for more complex applications.
