An attempt to discriminate food aromas using a multisensor array was described by focused on the utilization of semiconductor gas sensors. Other similar attempts such as utilizing a quarts resonator array followed by subsequent pattern recognition analysis using neural networks were briefly reviewed.
Although response properties can be somewhat controlled by doping noble metals into metal oxides (SnO2), responses of semiconductor gas sensors are nonselective in nature. However, our olfaction system discriminate aroma differences based on the nonselective signals emitted from receptor cells locating at the olfactory epithelium by applying a kind of pattern recognition to the signals. Therefore, responses from a gas sensor array composed of six semiconductor gas sensors were integrated in order to discriminate food aromas using chemometric pattern recognition techniques. The problems to be answered before making such a system were as follows; the minimum number of gas sensors needed for discriminations, methods for standardization of gas sensing conditions and for removing excess amount of ethanol in fermented products, and appropriate pattern recognition techniques.
A semiautomatic headspace concentrator installing a Tenax TA trap was effective to standardize aroma introduction for sensing besides removing excess amount of ethanol from headspace vapors in fermented products. All six sensors in the array responded immediately when thermally desorbed aroma components were introduced into a flask installing the gas sensor array. Response patterns for different samples were similar but the slight differences were well reproduced in every measurement. Chemometric pattern recognition analyses such as factor analysis, cluster analysis and linear discriminant analysis (LDA) were applied to the resulting six dimensional data matrix. LDA suggested that only three gas sensors were sufficient enough to discriminate most sample aromas. Coffee samples, liquors and essential oils were reasonably classified and were discriminated into correct sample groups. Thus facile novel methodology utilizing semiconductor multisensor array and the subsequent pattern recognition analysis showed the capability for discriminating food aromas.
