Analysis of signature patterns for discriminating fire detection with multiple sensors

Abstract

Incorporating intelligence into a fire detector so that it can recognize signature patterns is intended to permit prompt fire detection while allowing the detector to discriminate between signatures from fire and nonfire sources. The primary purpose of this preliminary study is to investigate the patterns of signatures associated with fire and environmental sources using small-scale experiments. We generated products from a wide range of conditions, from flaming or pyrolyzing samples, to heated samples and samples obtained with an atomizer. We also measured gas concentrations, light obscuration, and temperature to characterize the products. By analyzing the data, we identified trends from which an elementary expert system can be formulated to identify the source of the airborne products. Several patterns are evident. The maximum CO2 concentrations achieved during experiments with flaming fires are significantly greater than the maximum CO2 concentrations achieved during experiments with nonflaming fires (pyrolyzing fires, heated liquids, and environmental odors). The nonflaming sources can be identified based on the CO and metal oxide sensor peak measurements. Except for three experiments using pyrolyzing solids, the peak CO concentration is greater—though the Taguchi detector response is less—for nonflaming fires than for environmental sources. Subsequent application of a neural network properly classifies all except one pyrolyzing fire.