Based On Real-time Simulation Of Fluid And Solid Interaction Of Complex Shape Of CUDA

Fluid motion is a common phenomenon in nature. In the field of computer-aided design, real-time fluid simulation is desperately needed. To support ancillary device research and development for National Special Superfine Powder Engineering Research Center, a3D fluid simulation framework is needed, which can be used to simulate the motion of large-scale fluid interacting with complex geometrical rigid bodies. However, duo to the hight complexity of fluid motion, real-time fliud simulation is most concerned in the field. So, this paper studies real-time fliud simulation, and the main contents are summarized as follows.Firstly, studying Smoothed Particle Hydrodynamic (SPH) approach, a model based on physics. And we’ll analysis the data dependence and time complexity of the serial SPH algorithm.Secondly, studying high-performance parallel computing architecture (CUDA), and implementing a parallel SPH algorithm based on CUDA. The searching process of neighborhood particles greatly affect the performance of parallel SPH algorithm. So we take advantages of the new features of Kepler Architecture GPU, use Z-order to optimize spatial locality, so as to improve the existing parallel particle neighborhood search.Thirdly, we use triangle meshes to represent rigid body with complex geometry, propose a two-phase collision detection algorithm based on bounding volume hierarchy (BVH) to significantly reduce the time complexity and amount of calculation of the collision detection between particles and triangular meshes. As a result, we realized the real-time collision detection between fluid and dynamic rigid bodies with complex geometry.Finally, the scene of simulation is visualized by using OGRE image engine, and different techniques are adopted to render solids and fluid particles. It is a key point to effectively render large scales of particles since fluid particles are generally in very large amount. By binding particles’ position and color information to GPU vertex buffer and using CG hardware shading language to write vertex and fragment shaders, thus rendering particles into the pixels on GPU directly, the above aim can be achieved.