Discrimination of chemically similar organic vapours and vapour mixtures using the Kohonen network

A Kohonen network was employed to discriminate between a series of chemical ly similar alcohols and mixtures of organic solvents. The input data for th e Kohonen analysis was generated using an optimized eight-sensor array desi gned to sample the headspace of the solvents. Different sizes of output gri d were investigated to devise a network that gave optimum discrimination an d maintained relationships within the data set. When the output grid was la rge compared to the number of classes in the sample set, discrimination was shown to be enhanced compared to a small output grid. An advantage of the small output grid is that it was shown to maintain information within the o riginal data set. The Kohonen network generated easily distinguishable outp ut patterns, which could be used as an alternative to pattern recognition o r in conjunction with output grid maps.