Abstract
One of the biggest challenges in the design of fire detection systems is the discrimination between actual fire signatures and noise. This is especially true as the fire size at the time of detection decreases, as the signal to noise ratio decreases accordingly. Smoke detectors, for example, are simply particle detectors calibrated for specific particle size distributions, number densities and refractive indices. Smoke particles are electronically or optically sensed, and when a predetermined signal level is reached, an alarm condition is signalled.
Unfortunately, most currently available smoke detectors cannot discriminate between products of combustion, and non-combustion generated particulates. In addition, they lack the "intelli gence" to discriminate between products of combustion generated by a fire threat condition, and products of combustion generated by cigarette smoke, automobile exhaust or similar non- threatening combustion processes.
This is changing, however, with the introduction of artificial intelligence techniques into fire detection systems. Recent experimental work has shown that by sensing more than one fire signature, applying signal discrimination techniques, and making fire/non-fire decisions with fuzzy reasoning or neural networks, fire detection systems are able to accurately discriminate between fire and non-threatening or deceptive conditions. This paper looks at how these techniques are currently being employed and offers some ideas for further research and development.
Get full access to this article
View all access options for this article.