Smoothing Particle Hydrodynamics parallel implementation for numerical modelling of solid-fluid interactions

Abstract

In this work is presented an effcient parallel implementation of a Lagrangian mesh-free method based on the Smoothing Particle Hydrodynamics technique to analyse the behaviour of solids and fluids. The proposed solution makes use of different techniques to improve the computational complexity of the simulation. To improve the performance, a parallel solution of the SPH numerical solution is implemented on the GPU using the NVIDIA CUDA architecture. Both the serial and the parallel implementation of SPH used benchmark experiments to compare effectiveness under different conditions. Results are provided with a set of discriminants that allow the program to select which of the implementation (serial or parallel) is best suited to solve the model studied. The SPH modelling implementation is used to solve fluid flows simulation in a soil core's pore structure. The study of this problem is important to understand how heterogeneous soil structure affects water and solute transfer. This helps to obtain accurate predictions of contaminants residence time (average amount of time that a contaminant particle spends inside the pore network) and the quantity leached in the zone represented by the structures. In order to have a method to quantify the accuracy of the results provided by the SPH implementation the commercial software COMSOL was used to obtain the numerical results on the same models solved by the SPH implementation. Finally both results are compared against the expected theoretical results. The second application of the developed computational implementation is the study of the deformation of solid objects under stress deformation. The SPH model in conjunction with a stereo technique provides a method capable of defining the solid material variables. The obtained variables can be used in solid deformation modelling using SPH.