This thesis presents a new method for controlling the look of an ocean water simulation for the purpose of creating cartoon-styled fluid animations. Two popular techniques to simulate fluid, a statistical height field method via the Fast Fourier Transform and the Stable Fluid method for dynamic effects, are connected smoothly via a blend domain, thus allowing a height field to drive a physical simulation. In addition, the height field can be stylized by utilizing a keyframing technique on wave amplitudes defined in the Fourier domain, allowing for creative control of the fluid’s surface. Such stylized height fields therefore can be simulated to exhibit natural fluid motion as well as to produce dynamic effects such as breaking waves that were previously unattainable in common fluid pipelines.

<p>The fluid simulation in computer generates realistic animations of fluids by solving Navier-Stokes equation. The methods of simulation are divided into two types. The grid-based methods and particle-based methods. The former one is wildly used for scientific computation because of its precision of simulation while the latter one is used in visual effects, games and other areas requiring real-time simulation because of the less computation time it has. </p> <p> </p><p>The indoor airflow simulations with HVAC system in construction design is one specific application in scientific computation and uses grid-based simulation as the general-purpose simulation does. This study addresses the problem that this kind of airflow simulations in construction design using grid-based methods are very time consuming and always need designers to do pretreatment of the building model, which takes time, money, and effort. On the other hand, the particle-based methods would have less computation time with an acceptable accuracy in indoor airflow simulations because this kind of simulation does not require very high precision.</p> <p><br></p><p>Then this study conducts a detailed and practical comparison of different fluid simulation algorithms in both grid-based methods and particle-based ones. This study's deliverable is a comparison between particle-based and grid-based methods in indoor airflow simulations with HVAC system.</p> <p><br></p><p>The overall methodology used to arrive at the deliverables of this study will need two parts of work. The benchmark data is gathered from a CFD software simulation using FVM with a decent grid resolution. The particle-based data will be generated by simulation algorithms over the same set of room and furniture models implemented by OpenGL and CUDA. After the benchmark FVM simulation being conducted in a CFD software, the temperature field of airflow will be measured. After simulation, the temperature field are gained on each one of 4 particle-based simulation. A comparison standard is set and data will be analyzed to get the conclusion. The result shows that in a short simulation time period, after finding a proper number of particles, the particle-based method will achieve acceptable accuracy of temperature and velocity field while using much less time.</p><p></p>

Les suspensions colloïdales ont été étudiées par simulations numériques à partir de deux modèles : la dynamique Brownienne (BD) et la SRD-MD (Stochastic Rotation Dynamics - Molecular Dynamics). Ces études ont consisté à reprendre des travaux existants pour les porter sur GPU, tout en cherchant différentes optimisations possibles adaptées à ces simulations. Une amélioration de la recherche de voisinage de la littérature a pu être utilisée pour toutes ces simulations de type BD. Une simulation de SRD-MD avec couplage de force qui n'avait pas encore été parallélisée sur GPU dans la littérature, a été implémentée en utilisant un nouveau schéma de décomposition adapté à cette simulation, améliorant considérablement les performances. Ces simulations ont pu donner lieu par la suite à des études sur des suspensions colloïdales plus complexes : une hétéroagrégation entre deux suspensions avec des particules de même taille, une hétéroagrégation entre deux populations de colloïdes de tailles très différentes, et en dehors des suspensions colloïdales, une simulation de nanoalliages. Enfin, le modèle de SRD a été adapté afin d'être utilisé dans le cadre d'animation physique de fluide réaliste dans le contexte de l'informatique graphique. Des adaptations du modèle pour y incorporer des notions comme la gestion de la compressibilité, de la tension de surface ont dues être apportées. Des premiers résultats ont pu permettre de réaliser quelques simulations, dont une chute d'eau dans une verre. / Colloïdal suspensions have been studied by means of numerical simulation, using two physical models : Brownian dynamics and Stochastic Rotation Dynamics - Molecular Dynamics. These studies consist in parallizing colloïdal simulations from previous studies on GPU, and find some new optimisations for these specific simulations. An improvement of the neigborhood search has been implemented in all our BD type simulations. A SRD-MD with force coupling have been implemented for the first time in the literature, using a new decomposition scheme, which improves significantly its performances. Then, theses simulations have been adapted to study more complex colloidal suspensions : an interfacial heteroaggregation of colloidal suspensions, a heteroaggregation between two types of particles with a large size ratio, and outside this context, a nanoalloy simulation. Finally, the SRD model has been adapted to realistic fluid animtion from computer science context. Theses adaptations require to add to SRD model, the notion of compressibility and surface tension. First results have been released, like a pouring water into a glass simulation.

Particle simulations in real-time become reality only a few years before, when in computer science occured the idea of GPGPU. This new technology allows use the massive force of graphics card for general purposes. Today, the trend is to accelerate existing algorithms by rewriting into parallel form. On this priciple operate the particle systems too. An interesting area of particle systems are fluid simulations. The simulations are based on the theory of Navier-Stokes equations and their numerical solutions with SPH (Smoothed particle hydrodynamics). Liquids are part of everyday life, and therefore it is important to render them realistically. They are used in modern computer games and different visualizations that run in real time, therefore they must be quickly displayed.