Computation and measurements of flows in rooms
This thesis contributes to the numerical modelling of flows in ventilated rooms. A range of advanced turbulence models (non-linear low Reynolds number Reynolds Averaged Navier-Stokes (RANS), Large Eddy Simulation (LES) and hybrid LES/RANS models) are used to model the flow in four ventilated rooms. These describe the flow in a more physically consistent manner than the commonly used linear RANS models. The performances of Explicit Algebraic Stress Model (EASM) and, cubic eddyviscosity RANS model are first analysed on four benchmark flow configurations. They show significant accuracy improvements when compared to their linear equivalents. Flows in ventilated rooms are complex. Their numerical modelling required an accurate definition of the various boundary conditions. This is often lacking in the literature and hence, as part of this work, measurements in a controlled ventilated office (optimised for Computational Fluid Dynamics (CFD) modelling) have been done. The measurements comprise airflow velocities, temperatures, concentration decay and, a careful description of the room's boundary conditions under six ventilation settings. This room data is thus seen as ideal for validating of CFD codes when applied to room ventilation problems. The numerical investigations show that the predictions with zero- or, one-equation (k - 1) RANS models (commonly used in room ventilation modelling) are less accurate than those using two-equation k-e models. The study shows that the accuracy improvements of the EASM and cubic models are of lesser magnitude when applied to room ventilation modelling than when applied to the benchmark flow configurations. The cubic model in particular, besides being more numerically unstable than the other RANS models, does not always improve flow predictions when compared with its linear equivalent. The EASM, about 20 to 30% more computationally demanding than its linear equivalent, improves solution accuracy for most flow considered in this work. LES predictions have highest level of agreement with measurements. LES is however too computationally expensive to be used for practical engineering applications. The novel hybrid RANS/LES model presented appears promising. It has similar accuracy to LES at lower computational costs.