Probabilistic Modelling of Sensitivity in Fire Simulations

Abstract

The objective of this thesis is to apply probabilistic sensitivity analyses to the emerging field of toxic hazard simulation in fire science. Fire simulation based on computational fluid dynamics (CFD) plays an important role in performance-based fire design. However, the thermal-physical process of material decomposition in fires and the chemical reactions of fire effluents are not well enough understood to make useful predictions about their burning behaviour. Input parameters are subject to uncertainties which can cause deviations in the results. Conventional engineering approaches are not suitable for reliable prediction of effects of input uncertainties on the results. In this work, probabilistic sensitivity analysis based on random sampling is used to determine the most important input variables and how uncertainties in their value influences the output.
The theoretical basis for hazard analyses is summarised and uncertainties are described which typically influence the results of fire tests and numerical fire simulations. The simulations are run with the Fire Dynamics Simulator (FDS), version 6. The proposed method was used for both mixing-controlled and finite-rate combustion with Large-Eddy- and Direct Numerical Simulation. Results of three different sensitivity analyses, each based on up to 500 samples are presented. The input parameter sets were systematically generated using advanced Latin Hypercube Sampling. The results of the sensitivity analyses were evaluated using the Metamodel of Optimal Prognosis which provides a measure of the predictability of the simulation that can be applied as an indicator for model quality. The results allow conclusions to be made, which quality of prognosis can be achieved using current fire simulation technology. The most influential parameters have been identified. Based on these simulation results a recommendation is made as to how the technology of probabilistic fire simulation and sensitivity analysis can be developed further in order to allow material parameter identification as a basis for the prediction of toxic effluents from solid fuels.