Keywords: fire modelling, CFD, LES, SGS, computational fluid dynamics, large eddy simulation, validation.

The increasing use of computational fluid dynamics-based computer programs in the fire engineering industry requires a corresponding increase in validation work on these models. The Natural Fire Safety Concept compartment fire tests carried out at Cardington offered large-scale fire test data which could be utilised in this sense. The tests were modelled with the Fire Dynamics Simulator (FDS V4.0) using the mass loss data as the basis for the rate of heat release input, and comparing the predicted compartment gas temperatures across the domain against the measured temperature profiles from the thermocouple data. The sub-grid-scale (SGS) Smagorinsky model constant, C_s, was varied in order to firstly investigate the effect of such a variation on the output temperature predictions, and secondly to suggest an optimum value for this constant when modelling large-scale compartment fires. The results suggest a stronger dependence on C_s for coarser grids, but for finer grids the output temperatures show less sensitivity to the varying value of C_s.

Two further SGS constants, the Prandtl number, Pr, and the Schmidt number, Sc, were varied together with C_s. A matrix of 200 simulations was carried out to investigate the combination factor of these three SGS parameters. This sensitivity study analysed the effect of varying the SGS constants, C_s, Pr, and Sc, on large-scale compartment temperature data, where the main proportion of the energy is carried by the large eddies reveals some important distinctions. The results suggest that for any given scenario different combinations of these factors can result in markedly different output predictions.

When modelling large-scale compartment fires using CFD techniques there are several important factors which are critical in supporting the validity of the output of the modelling. The most important of these is how to couple the level of accuracy of the test measurements - used as the benchmark for the CFD results - with the conclusions drawn from the output from the modelling procedure.

The lack of reproducibility in fire phenomena, owing primarily to the sensitive dependence on initial conditions, requires that the output data of large-scale fire tests must be treated in the 'gross' rather than 'local' sense. This methodology is supported by the formulation of the model itself: the equations representing and linking the physical parameters are themselves extremely complex and the interaction of these equations can 'magnify' the effect of inherent uncertainty. Thus, it is unreliable to afford too much influence to the results of specific spatial and temporal analyses rather than looking at the 'global' indication to characterise behaviour.

Regarding the computational methodology, the relatively limited capacity of computational resources available forces an approximation of the SGS processes. The data from the Cardington NFSC post-flashover compartment fire tests were used to investigate the importance of the values of three SGS parameters: the Smagorinsky number, and the Prandtl and Schmidt numbers. This study reviewed the effect of these SGS modelling parameters using practicable computational techniques. It is concluded that for the two grid resolutions used in this study, the combination of higher C_s values coupled with higher Sc and Pr values produces more robust predictions of global compartment temperature profiles when modelling the NFSC post-flashover compartment fires, compared to using the current default values.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £135 +P&P)