Organic vapour sensing using a coated piezoelectric quartz crystal sensor array

The pattern of responses from a four sensor array have been used for the classification of methanol, propanol, butanol, hexane, heptane and toluene using artificial intelligence (AI) based pattern recognition methods. A feedforward forward network with backpropagation was trained using sensor array data with approximately 300 training vectors and 100 test cases and covering a period of four months. The network consisting of four input nodes, six output nodes, learning rate of 0.1 and momentum of 0 was built using a commercial package (NeuroShell). A classification success rate of 75% was achieved. The bulk of the mis-classifications arose from propanol being classified as butanol and hexane being classified as heptane. These mis-classifications are rational since the respective compounds are very similar in nature. A fuzzy logic algorithm where class membership functions are developed using the mean frequency change and standard deviation of individual sensors was developed for classification of the vapours. In this particular case, classification using the developed fuzzy logic gaussian algorithm was not as good as the feedforward network with backpropagation, but the guassian membership function offers a more rational approach than the previously published trapezoidal membership function.