A 6LoWPAN-Based Thermal Measurement, and Gas Leak for Early Fire Detection Using Artificial Neural Network

Abstract

Fire is a reoccurring problem in the Philippines, and is costing the government billions of pesos in property damage along with several hundred fatalities every year. Existing fire alarm systems are prone to false alarms because it relies solely on detecting smoke. Unmonitored heat and gas leakages were two of the top causes of fire incidents in previous years. Incorporating the MQ-5 and MQ-2 gas sensors with the AMG8833 thermal imaging camera will allow for an accurate fire monitoring system that is less prone to false alarms. Using the two gas sensors will allow for the monitoring of LPG, Butane, and CH4. Also method of interpolating the display of the AMG8833 from 8x8 pixels to 70x70 was developed and sensor data was sent wirelessly to ThingSpeak. The thermal camera and the gas sensors both correlated to accurately assess fire hazards. A wireless communication with the user was used on the system to shorten the time of response of the fire fighter when fire alarm is triggered. The sensors are connected wirelessly over 6LoWPAN and uses ANN (artificial neural network) for forecasting possible future sensor reading and identification. A best validation performance of 65.3892 at epoch 72 was achieved running the Matlab Neural Network Toolbox using the Scaled Conjugate Gradient Algorithm.