Study On Realistic Simulation And Acceleration Techniques Of Incompressbile SPH Fluids

Smoothed Particle Hydrodynamics(SPH) is a common physics-based fluid simulation method, which has gained increasing interest of computer graphics researchers over the last decade. It has been extensively used to simulate many kinds of complex phenomena including water, fire and smoke in such application areas as feature films, commercials, video games and military simulations due to its ability to naturally conserve the mass and accurately handle free surfaces, solid boundaries as well as splashes and foams.However, existing SPH approaches still feature some disadvantageous properties when simulating incompressible fluids. First, it is expensive to enforce incompressibility. Second, it is not efficient to use fixed spatial and temporal resolution. Third, non-uniformly sampled boundary particles cause stability issues. Finally, multiple specialized solvers are needed for two-way fluid-rigid interactions.To address these issues, this thesis focuses on the following aspects of incompressible SPH fluid simulation methods: Graphic Processing Unit(GPU) parallelization, adaptive acceleration, complex boundary handling and fluid-rigid coupling. The main contributions in this thesis are summarized as follows.1. A novel parallel framework for simulating incompressible fluids on the GPU is presented. To take advantage of the high degree of data parallelism of incompressible SPH methods, an efficient streaming pipeline to map the entire computational task onto the GPU is proposed, fully exploiting the massive computational power of state-of-theart GPUs. A GPU-friendly particle data representation for fluid attributes and a parallel sorting technique for fast neighbor search are also introduced. The presented simulation framework is highly scalable and efficient, making it very suitable for parallelizing various SPH fluid simulation algorithms such as adaptive acceleration strategies, complex boundary handling algorithms and fluid-rigid coupling techniques. The performance results illustrate that the method can efficiently simulate realistic incompressible fluids and can achieve 34 fps for 60 k particles on a high-end GPU.2. A new adaptive SPH method based on sleepy strategy to accelerate fluid simulations is presented. Traditional SPH-based methods simulate all fluid particles in every time step. This guarantees the numerical accuracy, but it is not usually efficient since in many scenarios some particles appearing to be at rest can be safely ignored without notably affecting the fluid flow behavior. To reduce the computational cost in the fixed resolution setting, a novel sleepy strategy is introduced which partitions the whole fluid particles into three different types, i.e., non-sleeping particles, intermediate particles and sleeping particles. By utilizing this strategy, only a portion of the fluid particles requires computational resources, thus an obvious performance gain can be achieved.Furthermore, a new adaptive time step technique is incorporated to further improve the performance. It is demonstrated that this approach can improve the efficiency without sacrificing visual realism compared to existing methods.3. An adaptive boundary particle sampling technique for realistically simulating the interaction of incompressible fluids with complex solid boundaries is proposed. Current SPH approaches sample solid surfaces with particles, and prevent particle deficiencies and density discontinuities at interfaces between fluids and solids by taking the solid particles into account when accumulating fluid density. But these methods suffer from density estimation errors and even stability issues in non-uniformly sampled solid boundary regions. The proposed method resolves these issues by generating boundary particles as two layers at the solid surface which have different influence factors. This approach can more robustly handle complex boundaries without the side effects of existing techniques,and produces more realistic simulation results.4. A novel parallel simulation method for two-way fluid-rigid coupling is presented. Traditionally, fluids and rigid bodies are simulated with different solvers as well as different geometrical representations, which adds more complexity to software projects.To address this issue, the GPU-based unified particle framework is extended to support two-way coupling of incompressible fluids and rigid bodies. Specifically, the Implicit Incompressible SPH(IISPH) method is parallelized using the GPU framework, and then a new rigid body simulator based on the two-layer particle sampling technique is incorporated. With the unified particle framework, this approach takes full advantage of the data parallelism of particle-based fluid-rigid coupling and the massively parallel computing capability of modern GPU hardware. The achieved results show that the method can efficiently simulate two-way coupling of incompressible fluids with rigid bodies.