A Comparison of Sources of Uncertainty for Calculating Sprinkler Activation

Abstract

Treatment of uncertainty is an important consideration when performing a quantitative risk assessment (QRA) for the purpose of performance-based fire safety design. It has been noted in New Zealand that the outcome for fire QRAs is sensitive to uncertainty in the effectiveness of fire safety systems, including sprinkler systems. This paper considers the uncertainty in sprinkler activation time and the heat release rate (HRR) at the time of sprinkler activation for two scenarios; one where a set of sprinkler experiments in a room is modelled and the second which considers a more general design case for the same room. The relative importance of aleatoric (natural) and epistemic (model) uncertainties is shown to depend on the scenario being modelled. Twelve sources of uncertainty were quantified based on literature data and ranked for the scenarios considered. A new probabilistic-deterministic design tool for making risk-informed fire safety decisions is used to calculate output distributions for the two parameters considered and the sources of uncertainty are ranked in terms of sensitivity for output uncertainty. It is found that fire growth rate tends to be more important than the model uncertainty associated with the deterministic zone model engine for the scenarios considered, and uncertainty in the fire location becomes more important than the model uncertainty in design cases where there is greater uncertainty. Uncertainty in sprinkler response parameters and the ambient interior temperature provided the least contribution to the output uncertainties.

Life safety in single-story steel frame buildings, Part II - probabilistic design

“…the uncertainty in the modelling process and uncertainty regarding discrepancy between the model and reality (Elms and Brown, 2006). However, combining the models and using a conservative approach is assumed to present an efficient model with reasonable results on the safe side (Frank et al, 2011). Important parameters were assumed to be randomly distributed.…”