Fluid Flow Simulation from Geometry Data Based on Point Clouds

It is nowadays a real challenge to perform fluid flow simulation from human-acquired data, in particular when the geometries are reduced to a set of 3D points without any connectivity. Assuming that a sufficient set of points can represent the underlying geometries with a high level of details, we present in this paper a method to perform CFD computations directly from the point cloud raw data. Using an error estimator, a level-set function is built directly from the point cloud, which bypass the explicit surface reconstruction step. The level-set is then used in a mesh-adaptation procedure to optimize the representation of the distance field near the its zero value. Secondly, we use the immersion volume method to define the boundary conditions at nodes. Finally, we used a VMS finite element solver to perform the fluid flow calculation. We finally present computations on 3D point clouds self-acquired in urban environments.