Predicting how quickly a fire can grow over surrounding combustible materials is a critical aspect in assessing fire scenarios. A promising approach to this challenge lies in numerical flame spread simulations that are able to predict the heat release rate (HRR) and rate of spread (ROS), based on pyrolysis modelling. However, a major constraint in developing pyrolysis models is the difficulty in obtaining all the required material parameters. These parameters are commonly derived using inverse modelling and optimisation techniques, taking bench-scale experimental data as targets. This thesis investigates two fundamental aspects of such techniques that may affect the reliability of estimated material properties for modelling pyrolysis and predicting flame spread: (1) the assumption that material parameters estimated from bench-scale experiments (e.g., cone calorimeter) are transferable to flame spread scenarios, and (2) the level of accuracy required in fitting experimental data to ensure that uncertainties in the estimated parameters are sufficiently low. Using horizontal flame spread over poly(methyl methacrylate) (PMMA) as a case study, sensitivity analyses are applied to quantify how uncertainties in material parameters influence ROS and HRR predictions. Results show that parameter sensitivities differ considerably between a simplified cone calorimeter and flame spread simulations conducted using the Fire Dynamics Simulator (FDS), challenging the assumption of direct parameter transferability. Moreover, even small mass loss rate peaks observed in differential thermogravimetric analysis data can measurably affect ROS predictions, highlighting the importance of accurately characterising pyrolysis rates. Further, a validation study using data from a small-scale horizontal flame spread experiment revealed difficulties in simultaneously achieving accurate predictions of ROS and temperature profiles within the solid. This discrepancy suggests that further research is needed to assess current limitations related to heat flux calculations and the 1D heat conduction model in FDS, so that the overall predictive capabilities of flame spread simulations are improved.
Tássia Lins da Silva Quaresma (Wed,) studied this question.