doi:10.3850/978-981-08-7724-8_15-02

ABSTRACT

The predictive capability of computer fire models depends on several factors, including the numerical quality of the differential equation solver, the quality of the spatial and temporal resolution, and the fidelity of the sub-models used to represent unresolved physical and chemical phenomena. Current pyrolysis models proposed to describe the thermal degradation of solid fuel sources and the associated production of flammable vapors represent one of the major bottlenecks in fire modelling. Pyrolysis models typically include a large number of unknown coefficients (i.e., material properties and parameters of the chemical reactions) and require a careful calibration phase. During the calibration phase, the pyrolysis model coefficients are determined by comparisons with experimental data, typically coming from thermo-gravimetric (TGA) and/or cone calorimeter experiments, and by error minimisation algorithms based on advanced optimisation techniques.

The present study examines the limitations of current approaches to pyrolysis modeling. The study considers: a global one-step Arrhenius-type pyrolysis reaction for charring materials (one of the pyrolysis models incorporated into NIST's Fire Dynamics Simulator (FDS) [1]); cone calorimeter data corresponding to a carpet material used in commercial aircraft applications; two optimisation techniques for parameter estimation (a stochastic hill-climber algorithm and a genetic algorithm); and three parameter estimation methodologies (corresponding to unrestricted or restricted searches of the parameter space). Results suggest that while parameter estimation techniques are successful at providing calibrated models that reproduce experimental cone calorimeter data, the values of the model parameters cannot be considered as representative of physical or chemical properties. Applications of the pyrolysis models outside the immediate calibration range may consequently lead to significant errors.

Keywords: Fire modeling, Pyrolysis modeling, Parameter estimation, Optimisation.

© 2011 CPD Unit, University of Leeds