Fire Banner

Wednesday, December 08, 2010

FireGrid: An e-infrastructure for next-generation emergency response support

by Dr Sung-Han Koo

A recent journal paper titled "FireGrid: An e-infrastructure for next-generation emergency response support" has been published in the Journal of Parallel and Distributed Computing. Its content is presented here.

The costs of fire are great, commonly estimated in the range of 1-2% of GDP. Despite this, emergency service intervention at fires is often reliant upon very basic information (i.e. fire alarm panel information) or simple “gut instinct” of experienced fire officers. This need not be the case in the modern era, when a range of technologies are available which, if effectively harnessed, could transform the way in which fire emergencies are tackled, thereby significantly impacting the costs associated with failures. Here we describe development and demonstration of a novel concept which integrates sensor technologies, fire simulation, High Performance Computing (HPC) and knowledge-based reasoning, to provide an “intelligent” emergency response system known as FireGrid.

The heart of the system is the sensor-linked fire model (described in more detail in reference 17). While fire simulation has found wide application historically for design purposes, the uncertainties of fire development defeat any attempt to provide a true predictive capability of hazard evolution, generally precluding real-time use. We bypass these uncertainties by continually updating our model with a flow of sensor-derived information regarding conditions in the building. The modelling strategy exploits Monte-Carlo techniques in combination with Bayesian inference for “steering”; being “embarrassingly parallel” in nature it is ideal for implementation on multiprocessor HPC systems. The output contains embedded probabilistic information about the likelihoods of various future hazard conditions, encompassing both threat to humans (i.e. escaping occupants, and incoming fire and rescue personnel) and to the building itself (in terms of structural weaknesses, or collapse potential). The interpreted information is conveyed rapidly to the end user, i.e. the “incident commander”, to provide decision support information that can effectively assist their intervention strategies.

Initial application of a system such as FireGrid would be most relevant to high-risk and critical infrastructures, including tall buildings. It is readily apparent that better information to incident commanders could be vital in avoiding scenarios comparable to the World Trade Center tragedies, where emergency responders continued intervention operations totally oblivious to the impending
collapse of the towers. FireGrid is an ambitious vision, and its success also depends upon an effective partnership and engagement with potential end users. Our initial project was undertaken in conjunction with various members of the UK fire and rescue services, culminating in a live fullscale demonstration test attended by a broad audience including a senior fire officer. The complex evolution of the fire, with unexpected behaviours and ultimate transition to “flashover”, was an ideal test of the sensor-linked model running on the grid, and the system capabilities were effectively demonstrated. Further development of such systems extends a genuine hope that some of the chronic and long-standing problems associated with accidental fires might be eventually be overcome, with wide–ranging benefits to all relevant stakeholders.

Editor note: A related paper is discussed in "Towards the forecast of fire dynamics to assist the emergency response"

No comments: