An analysis of the sensitivity of the rate of buoyancy-driven flame spread on a solid material to uncertainties in the pyrolysis and combustion properties. Is accurate prediction possible?

In this study, upward flame spread over a combustible solid in a vertical corner configuration was simulated using two widely used computational tools - ThermaKin2Ds and the Fire Dynamics Simulator (FDS) - and predicted sensitivity of fire growth dynamics to variations in material properties was examined to determine which ones carry measurement uncertainties that translate into the largest errors in predicted behavior. The nominal value of each material property was calculated as the average of measured values of a balanced set of 26 materials that represent the combustible polymeric solids widely used in industrial, transportation, and construction applications. The upper and lower boundaries of each property were defined based on an uncertainty quantification of each specific measurement. These average material properties and related uncertainties were determined based on an extensive literature review presented in this work. A measure of average flame spread velocity was defined (based on the time derivative of pyrolysis front location) as the model output to characterize fire growth dynamics. Overall, ThermaKin2Ds and FDS generally demonstrate similar sensitivity to model inputs. For both simulation tools, uncertainties in reaction kinetics (Ea/ ln(A)) and heat capacity of the virgin material, had the most significant effect on calculated average flame spread velocity, resulting in changes in this output (on average, for both simulation tools) of approximately 20%. FDS simulations also demonstrated a strong sensitivity to variations in heat of combustion when it was adjusted from its mean value to its lower or upper uncertainty bounds (17% change in average flame spread velocity; 2, 3 times as sensitive as ThermaKin2Ds results). In ThermaKin2Ds, an approximate 10% sensitivity in predicted average flame spread velocity was observed to variations in thermal conductivity of the virgin material, while in FDS this output demonstrated limited sensitivity to this parameter change. Uncertainties in the density of the virgin material and its char yield also showed a notable impact on predicted average flame spread velocity in both simulation tools. All other material properties considered in this work were found to demonstrate negligible influence on average flame spread velocity, which indicates that these parameters are measured with sufficient accuracy to deliver reasonably accurate predictions of flame spread dynamics.