Numerical simulation of scalar dispersion downstream of a square obstacle

The analysis of scalar dispersion in turbulent flows is relevant for a broad range of applications, including investigation of hazardous releases, bio-terrorism and control of air quality. In the last 50 years, experimental and numerical investigations of canonical flows, such as turbulent boundary layers (Fackrell & Robins 1982) or grid-generated turbulence (Livescu et al. 2000), have enhanced theoretical understanding and subsequently enabled the development of simplified, semi-empirical models. Although these models have been successfully employed in simplified flow configurations (Sykes et al. 1984), prediction of scalar dispersion over complex and realistic geometries remains challenging, especially because of large-scale unsteady effects which cannot be properly accounted for in a simple phenomenological framework. On the other hand, it is generally accepted that detailed flow simulations (e.g., Large-Eddy Simulation, LES) provide accurate predictions of the turbulence dynamics and the scalar mixing rates, thus promising to enhance our ability to study turbulent dispersion in realistic environments. This, in turn, could lead to the development of better reduced order models. A recent example where LES has been applied to scalar dispersion in complex geometries is the work of Walton & Cheng (2002), but the comparison with experimental measurements was limited to mean scalar concentration. In order to establish if detailed flow simulations are able to provide accurate predictions of scalar dispersion in complex geometries, in this preliminary study the experiment of Vinçont et al. (2000) is numerically reproduced. To the best of the authors knowledge, this is the first experimental setup where detailed measurements of turbulent scalar fluxes in a non-trivial geometry have been carried out. Moreover, the flow is characterized by low Reynolds numbers, thus allowing Direct Numerical Simulations (DNS) to be performed in the limit of large Schmidt numbers. In the first part of this research brief, we present a comprehensive RANS-based analysis of the experimental setup. We compare the predictions obtained using various one-point statistical models, namely k − ǫ, k − ω and Reynolds Stress Transport, to the experimental measurements. The simulation results appear to give only a limited agreement with the experiments, both in terms of low-order (mean concentration) and high-order statistics (turbulent flux). Furthermore, no closure was able to reproduce the streamwise component of turbulent scalar fluxes. Although the resulting effect on the scalar transport equation is found to be not significant, this limitation of eddy-diffusivity-based models is more severe in the analysis of scalar dispersion over complex boundaries (i.e., strongly spatially developing flows). As a second step, we performed DNS of the same experimental setup aiming at providing a detailed analysis of the interaction between turbulence structures and scalar dispersion. Note that several examples exist in the literature where DNS have been applied to the