- Author
-
Rose-Pehrsson, S. L.
|
Hart, S. J.
|
Shaffer, R. E.
|
Gottuk, D. T.
|
Wong, J. T.
|
Tatem, P. A.
|
Williams, F. W.
- Title
- Analysis of Multi-Critical Fire Detection Data and Early Warning Fire Detection Prototype Selection. Final Report.
- Coporate
- Naval Research Laboratory, Washington, DC
Hughes Associates, Inc., Baltimore, MD
- Sponsor
- Office of Naval Research, Arlington, VA
- Report
-
NRL/MR/6110-00-8484,
September 18, 2000,
28 p.
- Distribution
- AVAILABLE FROM National Technical Information Service (NTIS), Technology Administration, U.S. Department of Commerce, Springfield, VA 22161. Telephone: 1-800-553-6847 or 703-605-6000; Fax: 703-605-6900; Rush Service (Telephone Orders Only) 800-553-6847; Website: http://www.ntis.gov
- Keywords
-
fire detection
|
warning systems
|
sensors
|
neural networks
|
pattern recognition
|
smoke detectors
|
fire detection systems
|
experiments
- Identifiers
- sensor data fusion; sensor array; multivariate analysis; Probabilistic Neural Network (PNN); multi-sensors
- Abstract
- A series of tests were conducted to evaluate sensors for an early warning fire detection system under development. The tests were conducted from August 30-September 3, 1999 onboard the ex-USS SHADWELL, the Naval Research laboratory's full scale fire research facility in Mobile, Alabama. The tests have been used to evaluate and improve the multivariate data analysis methods and candidate sensor suites. The objective of the program is to develop an improved early warning fire detection system that will provide early fire detection with a low false alarm rate. The system under development combines a multi-criteria (sensor array) approach with sophisticated data analysis methods. Together an array of sensors and a multivariate classification algorithm produce early fire detection with a low false alarm rate. Several sensors measuring different parameters of the environment produce a pattern or response fingerprint for an event. Multivariate data analysis methods are trained to recognize the pattern of an important event such as a fire. Multivariate methods are trained using data in a training set and the training set consists of sensor responses to events and nonevents under various conditions. The data sets used for sensor array evaluation require that the sensors be in close proximity so that it can be assumed that the sensors are observing the same test atmosphere. Multivariate classification methods rely on the comparison of events with nonevents. Variations in the response of sensors can be used to train an algorithm to recognize events when they occur. A key to the success of these methods is the appropriate design of sensor arrays and the training sets used to develop the algorithm. Although, the event is most important, it is critical that the algorithms recognize nonevents as well. Standard test procedures included a baseline response to establish initial sensor conditions, exposure test, and recovery back to baseline. For example, a typical test collected 10 minutes of ambient air, followed by an exposure to a fire for 20 minutes, then re-exposure to ambient conditions for 10 minutes. Chemical sensors are subject to noise and other fluctuations with time. Therefore, typical sensor performance or typical baseline responses over the test period were determined by exposing the sensors to ambient air for the entire test period. Baseline tests conducted at different times of the day can be used to determine environmental effects and how they influence the sensors.