Development of an adaptive method for early fire detection based on the dynamics of carbon fires in the gas environment for prevention of fire in premises

The object of research is the processes of formation and change of carbon monoxide concentration signals in the gas environment at
the early stages of ignition of various combustible materials. The problem that was solved is to develop an adaptive method for detecting
fires based on the use of current features of statistical instability of gas environment regimes preceding a fire. A method for early detection
of fires by a carbon monoxide concentration signal is proposed, based on adaptive statistical processing using exponential smoothing
and accumulation of instability energy. Unlike traditional approaches, the proposed method analyzes structural changes in the signal,
which are manifested in an increase in variability and deviation from the current adaptive average. The method is implemented by
forming a variational-squared indicator of fires, which is the product of the coefficient of variation and the deviation of the signal from
its exponential average, squared, with subsequent exponential accumulation and normalization. It is shown that the obtained normal
ized statistics allow to maximize the probability of correct fire detection for a given false alarm probability. In the course of the research,
expressions for the probabilistic characteristics of early detection were obtained and the problem of optimizing the exponential memory
parameter was solved, which provides a compromise between sensitivity and noise resistance. Experimental validation showed that the
proposed method provides earlier fire detection, a higher probability of detection at a given false alarm level and resistance to background
signal drift. The practical significance of the method is determined by the possibility of its implementation in real-time early fire detection
systems based on controllers with limited computing resources.
Keywords: fire detection, carbon monoxide, coefficient of variation, exponential filtering, statistical detection, signal processing