NIST Time|NIST Home|About NIST|Contact NIST

HomeAll Years:AuthorKeywordTitle2005-2010:AuthorKeywordTitle

Distributed Sensor Fire Detection.

pdf icon Distributed Sensor Fire Detection. (324 K)
Cleary, T. G.; Notarianni, K. A.

NIST SP 965; February 2001.

International Conference on Automatic Fire Detection "AUBE '01", 12th. Proceedings. National Institute of Standards and Technology. March 25-28, 2001, Gaithersburg, MD, Beall, K.; Grosshandler, W. L.; Luck, H., Editor(s)(s), 139-150 pp, 2001.


fire detection; predictive models; fire detection systems; sensors; fire models


This paper details a case study that utilized model simulations to assess the relative performance benefits of distributed sensing over single-station, single-sensor smoke detection and co-located multi-sensor detection. 500 individual CFAST computer fire model simulations, performed for a separate project at NIST, were used as the data set to verify the hypothesis that distributed sensing can improve detection time over single sensor or co-located multi-sensor detection. The modeled space configuration consisted of seven rooms representing a single-floor apartment residence. The 500 simulations encompass a range of fire sizes, locations, initial and boundary conditions deemed important from fire death statistics and sensitivity analysis of various parameters. Model outputs included smoke, CO and temperature levels as a function of time in the upper layer of each room, thus smoke and CO concentration along with temperature were chosen as the sensor outputs. It was assumed that the detector instantaneously sees the computed upper-layer value of smoke, CO, or temperature. Four sensor configurations were examined along with four different rules governing the alarm state. The base configuration was a smoke detector located in the entrance. Another configuration had the smoke, CO and temperature sensors co-located in the entrance, while the other two configurations had the CO and temperature sensors moved to other separate rooms. The rules consisted of smoke concentrations with threshold adjustments if CO or temperature reached a certain value, and a temperature threshold criterion. The results suggest that distributed sensing can improve detection in many cases over a single multi-sensor detector. While more work needs to be done to test the distributed sensing concept, a natural end product would be an adaptive artificial neural network that is trained by fire model outputs, adjusts automatically to system changes due to sensor failure or location changes and incorporates building environment conditions.