NEURAL METHODS OF CALIBRATION OF SENSORS FOR GAS MEASUREMENTS AND AROMA IDENTIFICATION SYSTEM

The article presents the neural methods of calibration of gas sensors for use in an artificial electronic nose for gas measurements. Different neural network solutions will be presented and compared. They include the classical multilayer perceptron, neuro-fuzzy networks and support vector machines. The other aspect illustrated in the article is the introductory preprocessing of the measured sensor signals in order to attain the highest possible efficiency of the gas measuring system. The theoretical considerations will be supported by the numerical experiments concerning the application of the electronic nose. The first practical aspect is concerned with the application of the developed system for classification problems and will be illustrated in the examples of the recognition of the biocomponents in the gasoline and the recognition of smells of cosmetic cream at the aging process. The second one belongs to the estimation problem and is concerned with the determination of the concentration of the particular gas components in the mixture of gases.

PRACTICAL APPLICATIONS

The results presented in the article may find practical application for calibration of the electronic nose in gas measurements. The electronic nose is widely used for smell recognition. It may find practical application in the petroleum, cosmetics or food industry for the assessment of the quality of their products. Military application is also of great interest.