Fire Detection of Belt Conveyor Using Random Forest

This study investigates a fire detection method using random forest that is one of the machine learning algorithms, which is applied to catch signs from minute temperature changes indirectly measured by a temperature sensor deployed 011 an idler of a belt conveyor. Belt conveyors are often used outdoors and the ambient temperature changes, which makes it difficult to distinguish them from temperature changes due to fire, especially for a conventional simple threshold determination. Based 011 features of measured temperature with a trend, random forests classify whether the individual temperature is abnormal or not. It is described that even if the individual classification results include false detections, false classification can be reduced by rearranging the individual results in temporal order. To implement this temporal factor, a method of using the past classification result as a feature quantity is proposed. Simulations are conducted with the feature quantities added the preceding classification result, and it is shown that false detections are eliminated. © 2021 The Institute of Electrical Engineers of Japan.