Selective methane detection under varying moisture conditions using static and dynamic sensor signals

This paper reports the development of a methane analyser, which meets some of the requirements of a reliable domestic methane detector. The equipment is based on a five-element array that includes four tin oxide gas sensors and a humidity sensor. The pattern recognition engine is based on multilayer perceptron (MLP) neural networks trained with the back-propagation algorithm. Both static and dynamic signals from the sensors are fed into the neural networks to identify and quantify methane and ethanol. The system identifies and quantifies unknown samples of these contaminants, even when the ambient relative humidity varies broadly (from 25% to 80%). As a result, the equipment detects methane, rejects false alarms caused by ethanol and complies with some aspects of the European directive for domestic methane detectors. (C) 1999 Elsevier Science S.A. All rights reserved.