Risk Analysis of Fire Protection System Using Bayesian Networks. Milan Holický, CTU KI Šolínova 7, 166 08 Praha 6, Czech Republic, telephone + 420 2 24310208, fax + 420 2 243120736, e-mail holicky@vc.cvut; and Miloš Vorlíček, Czech Technical University in Prague, Klokner Institute, Czech Republic
Safety in case of fire is one of the essential requirements imposed on construction works by Council Directive 89/106/EEC. Experience and available data clearly indicate that depending on particular conditions and applied fire protection system, the probability of fire flashover (outbreak) may be expected within a broad range. It appears that Bayesian causal (belief) networks provide an effective tool to find a more accurate estimate for the probability of fire flashover and to propose an optimum fire safety system in a rational way.
Several conceivable networks including networks with active measures (sprinklers and fire brigade) are considered in the presented study. Typical nodes of a network consist of Fire start, Detection, Tampering, Sprinklers, Smoke Detection, Fire Brigade and Fire Flashover. All nodes are inter-connected by directional links corresponding to causal dependencies of relevant nodes. At present all the nodes are described by alternative random variables (having two states only). The input data consist of conditional probabilities of node states related to appropriate states of parent nodes (if there are any). These conditional probabilities can be assessed using a combination of theoretical insights, empirical studies, and various more or less subjective estimates.
The study shows that the resulting probability of flashover for a particular network is controlled by relevant links and input data for conditional probabilities. Bayesian belief network is then used for various parametric studies to analyse the significance of input data relating to the resulting probability of fire flashover. For example the resulting flashover probability is usually strongly dependent on the conditional probability of sprinklers being in the active state given the fire started. Moreover, an effect of new causal links corresponding to actual conditions is investigated and modified, more effective fire safety systems are proposed.
Go to . . .
1999 SRA-Europe Table of Contents
1999 SRA-Europe Author Index
Main Abstracts Menu Page
RiskWorld Home Page