Numerical simulation of large-scale combustion processes on distributed memory parallel computers using MPI

Abstract

The dimensions of full-scale utility boilers and the typical time and length scales of the most important physical phenomena pose additional difficulties. To use the numerical simulation for the design or improvements of large-scale furnaces, for example to enhance the combustion efficiency and to reduce pollutant emissions, an adequate level of accuracy and reasonable execution times are necessary, making parallel computing vital. For the radiation, which is the predominant mechanism of heat transfer in utility boilers, the discrete ordinates method is applied within the parallel simulations, too. Although this model has high additional memory requirements and uses a recursive algorithm, it is preferred over the simpler multiflux model. Additionally, the numerical efficiency is an important quantity that cannot be found in speedup curves. All results are carried out for a fixed number of outer iterations and, unless stated otherwise, the parallel computations converged to the same level of accuracy without loss in the numerical efficiency. Nevertheless, for the refined meshes, a noticeable increase in the number of outer iteration has to be expected.