NIST Time|NIST Home|About NIST|Contact NIST

HomeAll Years:AuthorKeywordTitle2005-2010:AuthorKeywordTitle

Development of a Fire Detection System Using FT-IR Spectroscopy and Artificial Neural Networks.


pdf icon Development of a Fire Detection System Using FT-IR Spectroscopy and Artificial Neural Networks. (1065 K)
Chen, Y.; Serio, M. A.; Sathyamoorthy, S.

Fire Safety Science. Proceedings. Sixth (6th) International Symposium. International Association for Fire Safety Science (IAFSS). July 5-9, 1999, Poitiers, France, Intl. Assoc. for Fire Safety Science, Boston, MA, Curtat, M., Editor(s), 791-802 pp, 2000.

Sponsor:

National Institute of Standards and Technology, Gaithersburg, MD

Keywords:

fire research; fire safety; fire science; fire detection systems; FT-IR; spectroscopy; neural networks; fire detection; false alarms; vapor phases; cables

Abstract:

Extensive measurements of flaming and smoldering fires and nuisance/environmental sources were performed with Fourier Transform Infrared (FT-IR) spectroscopy of gas phase products. A neural network model was formulated using the so-called Learning Vector Quantization (LVQ) network approach. The LVQ approach contains input and output layers with a hidden layer being a Kohenen layer. The hidden layer learns and performs classification. The inputs to the network are concentrations (from FT-IR measurements) of eighteen (18) gas species. The outputs of the network are classification of the input data as a flaming fire, smoldering fire, nuisance or environmental source. The network was trained and tested using the test data collected during this project. The results were very successful as, among the 248 cases tested, only 12 cases were misclassified, mostly due to the difficulties in classifying the modes of combustion during a transition from a smoldering to a flaming tire. Each case represents the gas phase concentration data at a time step from one of the validation fires, which were different types of fires from the training set. A first generation fire detection system us-ing FT-IR gas measurements and neural networks has been built and implemented.