Fire safety design based on calculations: Uncertainty analysis and safety verification

Evacuation life safety in a one-room public assembly building has been analysed with regard to uncertainty and risk. Limit state equations have been defined, using response surface approximations of output from computer programmes. A number of uncertainty analysis procedures have been employed and compared. the analytical first-order second-moment (FOSM) method, two numerical random sampling procedures (simple random sampling and Latin hypercube sampling) and standard PRA method. Eight scenarios have been analysed in isolation as well as aggregated into an event tree, with branches denoting functioning/failing protection system (alarm, sprinkler and emergency door). Input parameter distributions have been subjectively quantified and classified with respect to category: knowledge or stochastic uncertainty. Risk assessment results comprise probability of failure p(f), reliability index beta and CCDF (complementary cumulative distribution function) for evaluation time-margin deficit. Of special interest is the calculation of confidence intervals for the distribution of CCDFs obtained by the two-phase Monte Carlo sampling procedure, allowing a distinction between knowledge and stochastic uncertainty. The importance analysis carried out analytically gives data of fundamental significance for an understanding of the practical design problem. Partial coefficients have been treated only by calculating values implicit or inherent in a few existing sample design configurations. Future studies, preferrably using optimization procedures, are needed to produce generally valid values. (C) 1997 Elsevier Science Ltd.