Calibration of nonlinear solid-state sensor arrays using multivariate regression techniques

Abstract

The application of an array of eight Taguchi gas sensors to analyze two- and three-component mixture sets of toluene, benzene, acetone, and trichlorethylene is presented. The calibration of each mixture set is performed by two linear-based parametric modeling techniques, partial least-squares and nonlinear partial least-squares, and two nonparametric modeling methods, multivariate adaptive regression splines and projection pursuit regression. The overall ability of nonparametric techniques to calibrate arrays of nonlinear responding sensors and predict future samples is much better than that of linear parametric models. The average prediction error for the nonparametric techniques is approximately half of that for the linear methods.