- Author
-
Milke, J. A.
- Title
- Application of Neural Networks for Discriminating Fire Detectors.
- Coporate
- Maryland Univ., College Park
- Sponsor
- National Institute of Standards and Technology, Gaithersburg, MD
- Book or Conf
- University of Duisburg. International Conference on Automatic Fire Detection "AUBE '95", 10th. April 4-6, 1995,
Duisburg, Germany,
Luck, H., Editors,
213-222 p.,
1995
- Keywords
-
fire detection
|
fire detectors
|
experiments
|
small scale fire tests
|
large scale fire tests
|
smoke
|
odors
|
expert systems
|
smoldering
|
neural networks
|
light obscuration
- Identifiers
- neural networks
- Abstract
- Research is being conducted to describe the characteristics of an improved fire detector which promptly reacts to smoke while discriminating between smoke and odors from fire and non-fire sources. This study is investigating signature patterns associated with fire and environmental sources via small- and large-scale tests toward the development of an improved fire detector. On the tests, smoke and odors are produced from a variety of conditions: flaming, pyrolyzing and heated samples, and nuisance sources, such as aerosols, household products and cooked food. Measurements include light obscuration, temperature, mass loss, CO, CO₂, O₂ and oxidizable gas concentrations. The feasibility of an elementary expert system to classify the source of the signatures from small-scale experiments was demonstrated in the first phase. In the recently completed second phase, a similar expert system correctly classified the source of the signatures in large-scale experiments in 85% of the cases. Neural networks have been applied to both sets of data from the small- and large-scale tests providing an even greater successful classification rate.