A Probabilistic Inferential Algorithm to Determine Fire Source Location Based on Inversion of Multidimensional Fire Parameters

Abstract

A probabilistic inferential framework is proposed to utilize transient temperature data measured at the ceiling of the compartment to determine location of fire source, as well as size of fire, based on Bayesian inferential theory. This approach treats the problem as one of parameter estimation, expressed as a function of posterior probability distribution based on the qualitative of agreement between predicted temperatures and observed temperatures at the sensor locations. A comparison of the measured temperature from full-scale burn tests with predicted temperatures from in verse problem solution algorithm indicate the error to be less than 5% when fires are small, but the error increases to more than 10% for large size fires. The accuracy of the inverse problem solution algorithm can be improved by utilizing data from sensitivity studies carried out on fire source location errors and heat release rate errors.