Fire Safety Science Digital Archive

Many fire alarm systems are being developed based on digital processing of signals from fire sensors. As background research for development of algorithms and sensors for such a system, a data logging exercise was undertaken using analogue fire sensors based on components from existing optical chamber smoke, ionisation chamber smoke and heat detectors. The exercise was carried out using different areas within a factory which included off ices, a laboratory, assembly shop, machine shop, warehouse and a kitchen. Data logging was also carried out on other sites which included a record of a small real fire incident in a hospital. Results are presented in the form of graphs of sensor output against time over time scales of approximately 1 week and 435 days. They are also presented as frequency plots of variations from a mean, on logarithm/linear axes and, on natural logarithm against variation squared axes in order to test for a normal distribution. Sensor noise can be considered to be composed of several components, but that which causes most false alarms is dominated by signals which are artifacts of sensor and transmission faults. Algorithms should therefore be designed to reduce the effects of such faults, and also to diagnose incorrectly sited sensors.