Developpement d'un nez electronique applique a l'odeur de biogaz

In the present project, the quantification of the odor of biogas is being studied. Biogas is the principal release and the first cause of concern of the residents near wastetreatment works.; Olfactometry is currently the technique used worldwide for odor measurement. It is a time-consuming and expensive method relying on human expertise. Measurement of odors were tried with analytical instruments, like GC/MS or GC/FID, without much success. The relationships between the odor concentration and those of the compounds measured by the instruments are difficult to realize.; A new technique employed in the quantification of odors is the electronic nose. The term electronic nose is used when a network of nonspecific chemical sensors is twinned with a data-processing treatment unit with an aim of recognizing or measuring the concentration of a gas or an odor. The electronic nose makes it possible to get rid of the human subjectivity of the olfactometry and the limitations of the analytical instruments. The relationship between the signals coming from the chemical sensors of an electronic nose and the odor concentration of the measured gas mixture is highly non-linear. This non-linearity implies the use of regression techniques able to circumvent the curse of dimensionality inherent in the approximation of this type of functions. The neural machines, used as universal approximators, coupled to an algorithm of structural risk minimization, make it possible to effectively construct non-linear relations at several variables and to minimize the generalization error of the model tested.; The structural risk minimization algorithm permits the determination of the optimal model of neural machine to be used with the odor studied and the electronic nose employed. (Abstract shortened by UMI.)...