The Use of Pattern Recognition Techniques applied to Signals Generated by a Multi-Element Gas Sensor Array as a Means of Compensating for Poor Individual Element Response

Abstract

It has been hypothesised that the lack of specificity associated with organic semiconductor gas sensors can largely be overcome with the adoption of a multi-element sensor array, thereby allowing the elimination of unwanted sensitivities through suitable signal processing. The use of a pattern recognition strategy in the realisation of such an “intelligent” chemical sensor is implicit in the assumption that individual gas species can be identified against a background of various interfering gases. It is, however, infeasible to consider the characterisation of sensor responses to individual gases and the building of a set of pattern recognition rules other than by automatic means. Thus the approach reported here was to enable the intelligent chemical sensor to learn the various response patterns associated with particular analytes and hence build a knowledge base from which future inferences may be drawn. This paper describes how a multi-element array of gas sensitive metal phthalocyanine films, constructed on a single thick-film substrate, was used as the sensing element in an intelligent chemical sensor. Since the individual sensor sites may be coated with different phthalocyanines showing varying degrees of gas sensitivity, the individual responses of each to any particular analyte will give rise to a characteristic change in the output template comprised of each of the sensor resistances. By monitoring the change in this template on exposure to specific gases of predetermined concentration and employing a suitable feature extraction algorithm, the characteristic responses to particular analytes were learnt. The success of suitable signal processing techniques to accommodate the inherent cross-sensitivities exhibited by metal-substituted phthalocyanine film gas sensors is demonstrated. The results clearly show the viability of pattern recognition methods to analyse gas mixtures through the mathematical evaluation of data from gas sensor arrays as a means of improving the selectivity and specificity.