NIST Time|NIST Home|About NIST|Contact NIST

HomeAll Years:AuthorKeywordTitle2005-2010:AuthorKeywordTitle

Review of Algorithms for Fast and Reliable Fire Detection.

pdf icon Review of Algorithms for Fast and Reliable Fire Detection. (2203 K)
Jones, W. W.

NISTIR 7060; 26 p. October 2003.


fire detection; algorithms; sensors


The purpose of detecting fires early is to provide an alarm when there is an environment which is deemed to be a threat to people or a building. High reliability detection is based on the supposition that it is possible to utilize a sufficient number of sensors to ascertain unequivocally that there is a growing threat either to people or to a building and provide an estimation of the seriousness of the threat. The current generation of fire detection systems is designed to respond to smoke, heat, gaseous emission or electromagnetic radiation generated during smoldering and flaming combustion. Smoke is sensed either by light scattering or changes in conductive properties of the air, heat by thermocouples and thermistors, the electromagnetic spectrum by photodiodes, and gas concentrations by chemical cells. There is much additional work in progress to use solid-state and electrochemical sensors for oxygen, hydrogen, water vapor, carbon dioxide, chlorine, hydrogen sulfide. The full gamut of fire detection is possible utilizing currently available sensor technology. This includes very early detection as well as fIfe following. It has been shown to be possible to detect fires early and reliably using the analog signal of the current generation of fire detectors. The best combination for early detection has been shown to be the complement of ionization, photoelectric, carbon monoxide and temperature. This is "best" in the sense that it is possible, using current day sensors, to see characteristic signatures very early, as well as to deduce quantitative information beyond the normal tenability limits. This paper will demonstrate that low level sensing can achieve the goal of producing early detection, while improving reliability. The example we use is a neural network trained with a model of fire growth and smoke spread. This allows us to reduce the time to detection as well as reduce the error rate for both false alarms as well as missing fires.