NONLINEAR SIGNAL EVALUATION WITH LINEAR-REGRESSION TECHNIQUES FOR REDUNDANT SIGNALS

In this presentation it will be shown that it is possible to achieve sufficient prediction accuracy of gas concentrations with non-linear sensor characteristics and a linear regression technique for a 'redundant' sensor system. Redundancy means that the number of independent signals of a sensor array is higher than the number of gases or vapours in a mixture. Experiments have been done with mixtures of acetone and methanol vapours and an array of interdigital sensors with different types of coatings. The relative change of conductance and capacitance of three different coated sensors and one uncoated sensor has been measured so that altogether eight dimensions are available. The evaluation is done with the PLS (partial least squares) regression algorithm.