- Author
-
Hart, S. J.
|
Hammond, M. H.
|
Rose-Pehrsson, S. L.
|
Shaffer, R. E.
|
Gottuk, D. T.
|
Wright, M. T.
|
Wong, J. T.
|
Street, T. T.
|
Tatem, P. A.
|
Williams, F. W.
- Title
- Real-Time Probabilistic Neural Network Performance and Optimization for Fire Detection and Nuisance Alarm Rejection: Test Series 1 Results. Memorandum. February 1, 2000-May 3, 2000.
- Coporate
- Naval Research Laboratory, Washington, DC
Hughes Associates, Inc., Baltimore, MD
- Sponsor
- Office of Naval Research, Arlington, VA
- Report
-
NRL/MR/6110-00-8480,
August 31, 2000,
37 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
|
false alarms
|
sensors
|
smoke detectors
|
data analysis
|
warning systems
|
ships
|
neural networks
|
pattern recognition
- Identifiers
- multi-sensor; sensor arrays; multivariate analysis; Probabilistic Neural Network (PNN); sensor data fusion; Early Warning Fire Detection (EWFD) system
- Abstract
- The U.S. Navy program Damage Control-Automation for Reduced Manning (DC-RM), sponsored by the Office of Naval Research, PE0603508N, is focused on enhancing automation of ship fire and damage control systems. A key element to this objective is the improvement in situational awareness by improving the current fire detection systems. As in many applications, it is desirable to increase detection sensitivity, decrease the detection time and increase the reliability of the detection system through improved nuisance alarm immunity. Improved reliability is needed such that fire detection systems can provide quick, remote and automatic fire suppression capability. The use of multi-criteria based detection technology offers the most promising means to achieve both improved sensitivity to real fires and reduced susceptibility to nuisance alarm sources. A multi-year effort to develop an early warning fire detection system is currently underway. The system being developed uses the output from sensors that measure different parameters of a developing fire or from analyzing multiple aspects of a given sensor output (e.g., rate of change as well as absolute value) and a neural network for fire recognition. A series of tests were conducted on the ex-USS SHADWELL from February 7-18, 2000 to evaluate candidate prototypes of the early warning fire detection system (EWFD). Improved fire recognition and low false alarm rates were observed using data from full-scale laboratory tests generated in a chamber located at Hughes Associates, Inc. Several different sensor combinations were identified for use with a probabilistic neural network (PNN). Full-scale shipboard tests were conducted on the ex-USS SHADWELL to further develop detection algorithms and to expand the fire/nuisance source database. Using these two data sets, two candidate suites of sensors were identified for prototype development. Test Series 1 tested the real-time responses of the prototypes. The algorithm development for the prototypes, the results of Test Series 1 shipboard testing, and the subsequent optimizations are described in this report. The two data sets (laboratory and shipboard tests) served as the basis for a comprehensive PNN training data set used for the subsequent real-time applications. The classification of fire and nuisance events and the speed of the probabilistic neural network (PNN) were used to determine the performance of the multi-criteria fire detection system in Test Series 1. The EWFD system with the PNN developed for real-time detection, demonstrated faster response times to fires compared to commercial smoke detectors, while the overall classification performance was comparable to the commercial detectors. Some problems with the real-time implementation of the algorithm were identified and have been addressed. Using a variety of methods for speed and classification improvements, the PNN has been extensively tested and modified accordingly. As a result of the optimization efforts, significant improvements in performance have been recognized. A detailed examination of PNN failures during fire testing has been undertaken and the initial resultsinlcuded. Using real data and simulated data, a variety of scenarios (taken from our recent field experiences) have been used or recreated for the purpose of understanding the behaviors and failure modesof the PNN in this application.