Direct Simulation Monte Carlo Modelling using FPGAs
Direct Simulation Monte Carlo Modelling using FPGAs The Direct Simulation Monte Carlo (DSMC) method is the most commonly used numerical method for computing rarefied flows, such as those experienced upon planetary entry of a spacecraft, or in very small-scale flows. Invented by Prof. Graham Bird, the method uses statistical modelling of collisions between computed particles, based upon their relative velocities. Because collisions between atoms and molecules are directly simulated, the technique is particularly well suited to modelling thermal nonequilibrium in hypersonic flows and other collisionally determined flow behaviours. Although DSMC is capable of computing a wide variety of flow fields, it places extremely large computational requirements on the solution of three-dimensional flow problems. Even using large computational or GPU clusters, it is challenging and very power-intensive to produce solutions for problems of current engineering interest. Field-programmable gate arrays (FPGAs) are a means of directly programming hardware to achieve very fast computations for certain problems. In particular, FPGAs have the potential for producing very energy-efficient computations that are of comparable speed to the best solutions currently available. This PhD project involves developing an FPGA implementation of the DSMC algorithm, and combining this implementation with implementations for GPUs and CPUs in a computational cluster environment. The algorithm will be tested against existing codes for simulation of planetary entry flows. This project is part of an international collaboration combining high-quality computational codes and state-of-the art experiment to provide a better understanding of hypersonic separated flows.