Due to the lack of visual information in conventional fire detection methods, false alarms and missing alarms are easy to occur. N-heptane and aviation kerosene were used as experimental fuels to carry out oil pool fire combustion experiment. Combining the convolutional neural network and the principle of image correlation to analyze the classification and recognition of combustion images and the frequency of flame oscillation, which were used as parameters to determine whether a fire occurred. The results show that the peak flame temperature and increasing rate of temperature of combustion is slightly lower by low pressure, which cause jitter of temperature and the trend is more obvious. The O2 decreases firstly and then increases and finally tends to be stable, while, the content of CO2 increases firstly and then decreases and finally comes to be stable at 96kPa. However, Under the equidistant combustion time series of n-heptane and aviation kerosene, the average flame oscillation frequency is 3.26 Hz and 3.06 Hz, respectively, and the error between its theoretical flame oscillation frequency is small, and the accuracy rate is as high as 90%. It can be seen that low pressure has a greater influence on the fire behavior of combustion, which could provide theoretical support for studying the key technologies of fire alarm under low pressure.
|