On the uncertainty of flashover correlations and its effect on risk assessment

Flashover is a critical event in compartment fires, marking the transition to a fully developed fire. Its occurrence is commonly estimated using empirical correlations linking the compartment ventilation factor to a threshold heat release rate (HRR). Correlations such as those by Babrauskas and two McCaffrey-Quintiere-Harkleroad (MQH) variants are widely used, yet their uncertainties are rarely quantified. This study evaluates the uncertainty inherent in these three correlations based on their original datasets and models the resulting variability in the threshold HRR. A methodology is then presented to propagate this uncertainty into fire-risk assessments using stochastic event tree analysis, incorporating distributions of peak fuel-package HRR and ventilation factors to calculate flashover probabilities. The approach is demonstrated through a case study that estimates the loss exceedance function for fire-induced structural failure of an office building. Results show that different flashover correlations can produce expected and tail losses differing by up to 98% and 61%, demonstrating that neglecting intra-model uncertainty and inter-model variability can significantly affect estimates of fire consequences.