Sensor-linked fire simulation using a Monte-Carlo approach
This study is aimed at developing a predictive capability for uncontrolled compartment fires which can be “steered” by real-time measurements. This capability is an essential step towards facilitating emergency response via systems such as FireGrid, which seek to provide fire and rescue services with information on the possible evolution of fire incidents on the scene. The strategy proposed to achieve this is a novel coupled simulation tool, based on the Monte-Carlo-based fire model, CRISP, with scenario selection achieved via comparison with (pseudo) sensor inputs. Here, some key aspects of such a system are illustrated and discussed in the context of the detailed measurements obtained in the full-scale fire test undertaken in a furnished apartment at Dalmarnock. The capability of CRISP in reproducing the fire conditions – given knowledge of the approximate heat release rate in the fire – was first verified. It is then shown that continuous selection from amongst a multiplicity of scenarios generated in Monte-Carlo fashion can be achieved, so that the predictions evolve in a way that closely follows the real fire conditions. Whilst the benefits of sensor-steering are already clearly apparent, further improvements will be possible by establishing an appropriate feedback loop between the results assessment and the parametric space in which new fires are generated, perhaps using Bayesian methods. Nevertheless, true predictive capability remains crucially dependent on the sufficient representation in the model of the mechanisms of fire growth, and this must be the focus in achieving better forecasting ability.
The following license files are associated with this item: