Artificial neural networks and gas sensor arrays: quantification of individual components in a gas mixture

Abstract

A very promising way of increasing the selectivity and sensitivity of gas sensors is to treat the signals from a number of different gas sensors with pattern recognition (PARC) methods. A gas sensor array with six metal-oxide-semiconductor field-effect-transistors (MOSFETs) operating at elevated temperatures was exposed to two types of multiple-component gas mixture, one containing 5-65 ppm of hydrogen, ammonia, ethanol and ethylene in air and the other containing hydrogen and acetone in air. The signals from the sensors were analysed with both conventional multivariate analysis, partial least-squares (PLS), and artificial neural network (ANN) models. The results show that both hydrogen and ammonia concentrations can be predicted with PLS models; the predictions were even better with ANN models. The predictions for ethanol and ethylene concentrations were, however, poor for both types of model. Hydrogen and acetone, from the two-component mixture, were best predicted from an ANN model.