To improve the reliability of coal mine safety monitoring systems we have analyzed the characteristics of a methane sensor, an important component of the monitoring system of production safety in a coal mine and studi...To improve the reliability of coal mine safety monitoring systems we have analyzed the characteristics of a methane sensor, an important component of the monitoring system of production safety in a coal mine and studied the main type and mode of faults when the sensor was used on-line. We introduced a new method based on artificial neural network to detect faults of methane sensors. In addition, using the output information of a single methane sensor, we established a sensor output model of a dynamic non-linear neural network for on-line fault detection. Finally, the fault of the heating wire of the sensor was simulated, indicating that, when the methane sensor had a fault, the predicted output of the neural network clearly deviated from the actual output, exceeding the pre-set threshold and showing that a fault had occurred in the methane sensor. The result shows that the model has good convergence and stability, and is quite capable of meeting the requirements for on-line fault detection of methane sensors.展开更多
基金Projects 50534080 supported by the National Natural Science Foundation of ChinaNCET-05-0602 by the Program for New Century Excellent Talents in Universities of China2006KJ019B by the National Natural Science Foundation of Anhui Province Education Office
文摘To improve the reliability of coal mine safety monitoring systems we have analyzed the characteristics of a methane sensor, an important component of the monitoring system of production safety in a coal mine and studied the main type and mode of faults when the sensor was used on-line. We introduced a new method based on artificial neural network to detect faults of methane sensors. In addition, using the output information of a single methane sensor, we established a sensor output model of a dynamic non-linear neural network for on-line fault detection. Finally, the fault of the heating wire of the sensor was simulated, indicating that, when the methane sensor had a fault, the predicted output of the neural network clearly deviated from the actual output, exceeding the pre-set threshold and showing that a fault had occurred in the methane sensor. The result shows that the model has good convergence and stability, and is quite capable of meeting the requirements for on-line fault detection of methane sensors.
