Real-Time Simulation of Indoor Air Flow using the Lattice Boltzmann Method on Graphics Processing Unit

This thesis investigates the usability of the lattice Boltzmann method (LBM) for the simulation of indoor air flows in real-time. It describes the work undertaken during the three years of a Ph.D. study in the School of Mechanical Engineering at the University of Leeds, England. Real-time fluid simulation, i.e. the ability to simulate a virtual system as fast as the real system would evolve, can benefit to many engineering application such as the optimisation of the ventilation system design in data centres or the simulation of pollutant transport in hospitals. And although real-time fluid simulation is an active field of research in computer graphics, these are generally focused on creating visually appealing animation rather than aiming for physical accuracy. The approach taken for this thesis is different as it starts from a physics based model, the lattice Boltzmann method, and takes advantage of the computational power of a graphics processing unit (GPU) to achieve real-time compute capability while maintaining good physical accuracy. The lattice Boltzmann method is reviewed and detailed references are given a variety of models. Particular attention is given to turbulence modelling using the Smagorinsky model in LBM for the simulation of high Reynolds number flow and the coupling of two LBM simulations to simulate thermal flows under the Boussinesq approximation. A detailed analysis of the implementation of the LBM on GPU is conducted. A special attention is given to the optimisation of the algorithm, and the program kernel is shown to achieve a performance of up to 1.5 billion lattice node updates per second, which is found to be sufficient for coarse real-time simulations. Additionally, a review of the real-time visualisation integrated within the program is presented and some of the techniques for automated code generation are introduced. The resulting software is validated against benchmark flows, using their analytical solutions whenever possible, or against other simulation results obtained using accepted method from classical computational fluid dynamics (CFD) either as published in the literature or simulated in-house. The LBM is shown to resolve the flow with similar accuracy and in less time.