Real-Time Probabilistic Neural Network Performance and Optimization for Fire Detection and Nuisance Alarm Rejection: Test Series 2 Results
Memorandum rept. 1 Feb-31 May 2000
NAVAL RESEARCH LAB WASHINGTON DC
Pagination or Media Count:
A second series of tests was conducted to evaluate and improve the multivariate data analysis notebooks and candidate sensor suites used for the Early Warning Fire Detection EWFD system under development. The EWFD system is to provide reliable warning of actual fire conditions in less time with fewer nuisance alarms than commercially available smoke detection systems. Tests were conducted from 25 April to 5 May 2000 onboard the ex-USS SHADWELL. This report documents the performance of the probabilistic neural network achieved in real-time during this test series. Further optimization of the algorithm has yielded performance gains over the real-time results. Modifications have been made that improve the real-time data acquisition and the ion sensor calibration. Background subtraction was investigated and will be used in future tests. The best performance was provided by a four sensor array consisting of ionization, photoelectric carbon monoxide and carbon dioxide sensors.
- Statistics and Probability
- Safety Engineering