Pollutant dispersion in a large indoor space. Part 2: Computational fluid dynamics predictions and comparison with a scale model experiment for isothermal flow.

This paper reports on an investigation of the adequacy of computational fluid dynamics (CFD), using a standard Reynolds Averaged Navier-Stokes (RANS) model, for predicting dispersion of neutrally buoyant gas in a large indoor space. We used CFD to predict pollutant (dye) concentration distribution in a water-filled scale model of an atrium with a continuous pollutant source in the absence of furniture and occupants. Predictions from the RANS formulation are comparable with an ensemble average of independent identical experiments. Model results were compared with pollutant concentration data in a horizontal plane from experiments in a scale model atrium. Predictions were made for steady-state (fully developed) and transient (developing) pollutant concentrations. Agreement between CFD predictions and ensemble averaged experimental measurements is quantified using the ratios of CFD-predicted and experimentally measured dye concentration at a large number of points in the measurement plane. Agreement is considered good if these ratios fall between 0.5 and 2.0 at all points in the plane. The standard k-epsilon two-equation turbulence model obtains this level of agreement and predicts pollutant arrival time to the measurement plane within a few seconds. These results suggest that this modeling approach is adequate for predicting isothermal pollutant transport in a large room with simple geometry. CFD modeling of pollutant transport is becoming increasingly common but high quality comparisons between CFD and experiment remain rare. Our results provide such a comparison. We demonstrate that the standard k-epsilon model provides good predictions for both transient and fully developed pollutant concentrations for an isothermal large space where furnishings are unimportant. This model is less computationally intensive than a large eddy simulation or low Reynolds number k-epsilon model.