On the prediction of turbulent flows around full-scale buildings

Data from tests on a full-scale, single-span high eaves commercial glasshouse are used to quantify the uncertainties associated with the use of computational fluid dynamics to obtain wind load predictions for full-scale structures. It is demonstrated that the widely used assumption of two-dimensional flow field in the mid-span leads to serious overestimation of the suction pressures over the roof and on the leeward wall. It is further shown that the use of a Reynolds-stress closure enables the capture of flow reversal downstream of the windward eaves. In contrast, the industry-standard k–ε model is found to predict no flow separation, contrary to the experimental observation. Finally, guidelines are suggested for suitable mesh distributions and for the efficient sizing of the computational domain relative to the buildingʹs dimensions.