Skip to Main Content
There are many applications in which it is desirable to be able to predict the concentrations of gases in a gas mixture using simple and relatively cheap non-selective sensors. The outputs of the sensors are functions of the concentrations both of the gases and of the different gas mixtures. It is not possible to uniquely determine the concentrations of the gases in a mixture from the output of a single sensor. However this might be achieved if a sufficient number of different and suitably chosen sensors were used and the outputs from this array of sensors could be analysed. The purpose of the work described was to find the means of determining the concentrations of the gases in a mixture of known gases given the outputs from an array of different tin oxide sensors placed in the mixture, and further to furnish, for each individual mixture of gases, estimates for the errors in the predicted concentrations. Committees of multilayer perceptron artificial neural networks were trained using artificial data to predict the gas concentrations and the errors in these predictions. In a second procedure the sensor outputs were modelled mathematically in terms of the gas concentrations, based upon some simple physical assumptions. The unknown coefficients of the model were determined mathematically using the experimental data.