Advanced simulations of air distributions in buildings
Real-time or faster-than-real-time flow simulation is crucial for studying the airflow in buildings or other enclosed environments, such as building emergency management, public health protection, sustainable building design, and building energy performance evaluation. The simulation should be informative by providing airflow motion, temperature distribution, and species concentration. Meanwhile, it is also important for the simulation to compute quickly since the allowed time is limited. However, none of the current techniques for flow modeling and computing hardware can satisfy such requirements.^ Nodal models for flow simulations are simple and fast, but not informative. Computational Fluid Dynamics (CFD) is accurate, but it is too time-consuming. To obtain a quick and informative solution, this study has proposed a Fast Fluid Dynamics (FFD) method, which is an intermediate approach between the nodal model and the CFD. This investigation used the FFD method with and without turbulence treatments to systematically study four basic flows in buildings, and compared the numerical results with the corresponding CFD results and the data from the literature. The comparison showed that the FFD could offer more complete flow information than the nodal model, but less accurate results than the CFD. At the same time, the FFD was about 50 times faster than the CFD. Because of a significant numerical viscosity in the FFD model, the FFD with the laminar model had the best overall performance in terms of reasonable accuracy and simulation time.^ The FFD simulation was improved by advancing its numerical schemes and optimizing the implementation. By modifying the time-splitting method and optimizing the implementation, this work successfully saved the computing time by 50%. Meanwhile, a finite volume discretization scheme was applied to enhance the mass conservation. In addition, a mass correction function was proposed to provide a simple practical solution for mass conservation in any specific domain. Furthermore, a hybrid scheme was developed to minimize the numerical viscosity caused by the linear interpolation in the semi-Lagrangian method.^ The improved FFD program was further validated by simulating particle dispersion in a pipe. The prediction had a good agreement with the experimental data. However, this was only a simple case. The FFD needs more validations for contaminant transport.^ It is also possible to further enhance the computing speed by performing the computation in parallel. Multi-processor supercomputers are widely used for parallelized flow simulations. However, they are very expensive and not portable. Instead, this study conducted the FFD simulation in parallel on a Graphics Processing Unit (GPU). The implementation used a NVIDIA GTX 8800 GPU and a Compute Unified Device Architecture (CUDA) language that is compatible with C language. The FFD code on the GPU was then applied to simulate four basic indoor flows. It turned out that the FFD code on the GPU could produce the same result as the one on a Central Processing Unit (CPU). In addition, the FFD simulations on a GPU were 10 to 30 times faster than the simulations on a CPU, depending on the grid resolution.^ In the future, the FFD scheme can to be improved to achieve a better performance for turbulent flow. Meanwhile, the FFD prediction for species concentration and three-dimensional FFD code needs to be further validated. It is also worthwhile to extend the capability of the FFD program, such as studying the flow in a complex geometry and coupling it with an energy simulation tool.^
Qingyan Chen, Purdue University.
Engineering, Architectural|Engineering, Civil|Engineering, Mechanical
Off-Campus Purdue Users:
To access this dissertation, please log in to our