PhD Thesis - staffweb - University of Greenwich

PhD Thesis by John Ewer.SOR solvers processing all cells equally. It should be noted that much of this saving is due tothe "de-coupled" void group which, as shown in Case 2, saves 26.1% of the processing. Theremaining 11.2% saving is due to the optimisation of processing within the extended regionwhich targets less solver processing in cells with relatively low velocity flow. The fact that thissaving is comparatively less than for the "de-coupled" region is also anticipated. This can beexplained by considering the work performed in the "de-coupled" and dynamic groups. In the"de-coupled" group, it was not necessary to build the system matrix coefficients for the membercells whereas any cell in a solved group that performs one (or more) iterations must build thesystem matrix coefficients in order to perform any calculation. Building the system matrixcoefficients is relatively costly compared to solving the matrix.The results indicate that there are large potential savings to be gained in the simulation of firemodelling scenarios by the targeting and optimisation of processing effort in fully de-coupled,suitably stratified or geometrically related flows. Furthermore, these savings need not result incompromised accuracy of the final solution. The techniques developed and presented hereresulted in considerable run-time savings of up to 37% of processing time. It is anticipated thatthis figure can be improved significantly when a better understanding of the balancing requiredbetween groups and variables is achieved.As group solvers are a new concept, there was little or no expertise to guide in the optimalselection of number of groups to use, the choice of group membership conditions and therelative amounts of processing used in each group. Furthermore there are a number of remaininggroup solver control options which were not varied during the test simulations.It is anticipated that in large scale simulations, which may involve whole buildings, there arelikely to be much greater savings possible with intelligent use of group solvers that can targetthe processing only on the active flow and fire regions until the solution characteristics in otherregions become significant.Current research efforts are directed at gaining a better understanding of when it is appropriate7-118