|
|
<TITLEGases
There are various scenarios in which it is desirable to determine the concentrations of gases in a gas mixture. These include the escape of vapours from stores of hazardous chemicals, the gases in car exhaust fumes, the combustion products from fires, atmospheric emissions from chimneys and landfill sites, and in the food and drinks industries. The interest here lies in using the outputs from an array of semiconductor (or other) gas sensors located within the gas mixture to determine the individual gas concentrations.
Sensor arrays contain different types of sensors and are used because single sensors are usually sensitive to a number of different gases so that their outputs are the combined effects of all the gases in the mixture. This means the individual concentrations cannot be determined. By analysing the outputs of the array the individual gas concentrations may be found
There is a dearth of information concerning the effects of gas mixtures upon the sensor outputs. In fact one interest is in trying to better understand this behaviour and some theoretical progress has been made. More pragmatically, sensor output data for a large number of different gas mixtures has been generated artificially starting from the best available experimental database located worldwide. In one method this data has been used to train committees of multilayer perceptron artificial neural networks by Bayesian techniques and also by back-propagation to predict the gas concentrations and to attempt to estimate the errors in the predictions. The trained networks were then successfully tested using test data. Further developments should improve the accuracy and simplify the methods. In a second method a macromodel of the gas concentrations was made in terms of the sensor outputs. After determining the coefficients of the model, the concentrations of gases in the mixtures was satisfactorily computed from the equations.
Mail to:
Dr. B.W. Jervis, b.w.jervis@shu.ac.uk
Back To: