Voxelization of Moving Deformable Geometries on GPU

Voxelization is a standard technique to represent arbitrary shaped geometries on a Cartesian grid. It is often utilized in pre-processing stage of any computational fluid dynamics (CFD) simulation for distinguishing the fluid and solid domain. In addition, identification of boundary fluid nodes in the immediate vicinity of the solid body is extremely crucial for proper imposition of boundary conditions and force evaluation. These nodes are therefore tagged separately and is often termed as surface voxelization. However, this procedure becomes non-trivial and computationally expensive as the complexity of geometry increases, especially if it is deformable and moving. Here voxelization needs to be performed in the solid volume as well, as the nodes keep switching from solid to fluid and vice versa at every iteration. For fluid-structure interaction problems, the analysis of flow behaviour requires an additional operation where the point of intersection of the lattice links connecting the fluid boundary nodes and solid bound nodes need to be further calculated. This ensures that deformation of geometry is properly captured and the correct boundary velocity is enforced onto the fluid (no slip). In this work we present techniques for GPU acceleration of voxelization for moving deformable geometries intended for CFD solvers based on the lattice Boltzmann method (LBM). The proposed techniques show speed-ups of up to 5.1x over equivalent parallel implementations.