high performance computing on graphics processing units: hgpu.org

This work describes and demonstrates a novel numerical framework suitable for simulating the behaviour of freely falling liquid droplets. The specific case studied is designed such that the properties of the system are similar to those of raindrops falling through air. The study of raindrops is interesting from both an engineering standpoint and from a standpoint of pure curiosity. As a natural phenomenon, rainfall is something which is experienced by everybody, yet its properties are often misunderstood. The primary engineering application is in improving the ability of radar to determine the characteristics of rainfall for meteorological purposes. The significant original contributions to knowledge within this work come from several areas. The numerical methods used are a unique combination of a high order incompressible implicit large eddy simulation method, a conservative level set method, and a pressure projection method. These methods have all been implemented on a highly parallel GPU architecture, with a resulting performance increase of approximately ten times when a single GPU was compared to a single CPU core. The water droplets were simulated in a regime not previously studied by three dimensional methods. The results of these simulations confirmed the validity of the numerical model by reproducing several important experimental results. New insight was then gained regarding the behaviour of droplet wakes, an area with little previous research. The results of the test simulations show great promise for future use of the numerical framework developed. While the simulations todate have been of air-water interactions, there is little reason the model should be constrained to such a system. In theory almost any low speed isothermal interaction of immiscible Newtonain fluids, with length scales of greater than 1mm, could be modeled accurately by these methods.