Improving the regulatory acceptance and numerical performance of CFD based fire-modelling software

Angus Joseph Grandison

2003

The research of this thesis was concerned with practical aspects of Computational Fluid Dynamics (CFD) based fire modelling software, specifically its application and performance. Initially a novel CFD based fire suppression model was developed (FIREDASS). The FIREDASS (FIRE Detection And Suppression Simulation) programme was concerned with the development of water misting systems as a possible replacement for halon based fire suppression systems currently used in aircraft cargo holds and ship engine rooms. As part of this programme of work, a computational model was developed to assist engineers in optimising the design of water mist suppression systems. The model comprised of the following components: fire model; mist model; two-phase radiation model; suppression model; detector/activation model. The fire model uses prescribed release rates for heat and gaseous combustion products to represent the fire load. Typical release rates for heat and combustion products have been determined through experimentation. The radiation model is a six-flux model coupled to the gas (and mist) phase. The mist model is based on Lagrangian particle tracking. Only the fire and suppression model will be described in detail in this thesis as this constituted the author’s contribution to FIREDASS. This work highlighted a number of issues associated with the application of CFD fire modelling software used in design of fire safety systems. The first issue was the reliability of CFD based fire predictions while the second was a practical issue associated with the amount of time required to run CFD fire models in a practical design environment. The remainder of the thesis is concerned with addressing these issues.

To address the first issue a set of procedures was developed to test the applicability of CFD fire modelling software. This methodology was demonstrated on three CFD products that can be used for fire modelling purposes. The proposed procedure involved two phases. Phase 1 allowed comparison between different computer codes without the bias of the user or specialist features that may exist in one code and not another by rigidly defining the case set-up. Phase 2 allowed the software developer to perform the test using the best modelling features available in the code to best represent the scenario being modelled. In this way it was hoped to demonstrate that in addition to achieving a common minimum standard of performance, the software products were also capable of achieving improved agreement with the experimental or theoretical results. A significant conclusion drawn from this work suggests that an engineer using the basic capabilities of any of the products tested would be likely to draw the same conclusions from the results irrespective of which product was used. From a regulators view, this is an important result as it suggests that the quality of the predictions produced are likely to be independent of the tool used – at least in situations where the basic capabilities of the software were used.

The second issue raised from FIREDASS was addressed by utilising Parallel Processing techniques on office based computer equipment. Parallel Processing has been used for many years in the field of computational modelling including fire modelling. Parallel processing distributes the computational task over a number of processors and therefore allows computational problems to be solved in a shorter timeframe essentially by utilising more computational power. The majority of this work has focussed on the use of specialised proprietary hardware generally based around the UNIX operating system. The majority of engineering firms that would benefit from the reduced timeframes offered by parallel processing rarely have access to such specialised systems. However, in recent years with the increasing power of individual office PCs and the improved performance of Local Area Networks (LAN) it has now come to the point where parallel processing can be usefully utilised in a typical office environment where many such PCs maybe connected to a LAN. Harnessing this power for fire modelling has great promise. Modern low cost supercomputers are now typically constructed from commodity PC motherboards connected via a dedicated high-speed network. However, virtually no work has been published on using office based PCs connected via a LAN in a parallel manner on real applications. The SMARTFIRE fire field model was modified to utilise multiple PCs on a typical office based LAN. It was found that good speedup could be achieved on homogeneous PCs, for example for a problem composed of ~100,000 cells would run on a network of 12 PCs with a speedup of 9.3 over a single PC. A dynamic load balancing scheme was devised to allow the effective use of the software on heterogeneous PC networks. This scheme also ensured that the impact of the parallel processing on other computer users was minimised. This scheme also minimised the impact of other computer users on the parallel processing performed by the FSE.