Abstract: Pyrolysis models used in Computational-Fluid-Dynamics-based fire models are typically semi-empirical, include a large number of unknown parameters (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 reference experimental data, for instance data taken from thermo-gravimetric and/or bench-scale experiments. The present study examines the predictive capability of pyrolysis models developed via a calibrated semi-empirical approach. The study first introduces six different semi-empirical models developed to simulate pyrolysis of polyvinyl chloride (PVC). All of the models are similar and use a global one-step Arrhenius-type pyrolysis reaction. They differ because of different modeling assumptions made that impact the number of unknown model parameters and/or because of the optimization technique used to determine the unknown parameters (a genetic algorithm or a stochastic hill-climber algorithm). The six models are calibrated and by design, provide similar results under conditions that are close to those of reference cone calorimeter experiments. The study then considers an evaluation of the predictive capability of the six pyrolysis models through a series of numerical experiments, including several cone calorimeter tests and one vertical upward flame spread problem; these configurations feature conditions that are significantly different from the reference conditions used in the model calibration phase. It is found that predictions from the PVC pyrolysis models start to diverge for conditions that lie outside of the calibration range. Most notably, the models lead to conflicting results when applied to the flame spread problem. These results suggest that the domain of validity of semi-empirical pyrolysis models is limited to the conditions that were used during model calibration and that extrapolation to non-calibrated conditions may result in a significant loss of accuracy.

Autoría

Descargar referencia