Parallel simulation of multidisperse granular flows using GPUs

In this paper, we present an application of GPU-based parallel computation for the simulation of multidisperse granular flows. We also show an application for the case of the Corona Electrostatic Separation (CES) process used in the waste management industry, where a strong electric field is used to separate plastic from metal particles in dense multidisperse granular flow of oddly-shaped fragments. The two major bottlenecks of the simulation are the collision detection and the solution of a complementarity problem at each time step; this limits the number of particles that can be simulated in reasonable time frames on the CPU, so we ported our simulation software to a parallel computing architecture. A custom collision detection has been used, where both broad-phase and narrow-phase collision stages have been designed in order to exploit parallel computation; such an algorithm is able to deal with particles of different shape and size, as needed in multidisperse granular flow. Also, a custom solver has been developed for solving the complementarity problem on parallel hardware. Such a solver requires multiple kernels and complex computational primitives because the complementarity problem does not fit in the perfectly-parallel computational paradigm, moreover, special care must be used to exploit data coalescence as much as possible. Finally, external force fields have been introduced, to simulate and reproduce the physics of electrostatic forces in the CES separation process.