Calibration of microhotplate conductometric gas sensors by non-linear multivariate regression methods

Abstract

This paper presents a demonstration of quantitative multicomponent multivariate calibration of microhotplate (MHP) conductometric sensors for binary and tertiary mixtures of light gases in air. Four element microsensor arrays of TiO 2, SnO 2 with surface-dispersed gold, and two different grain structures SnO 2, were used to differentiate among the analytes in the mixtures. We illustrate results from isothermal operation of these varied sensors, as well as the value of high-information content operation in dynamic temperature programmed settings, where the rate response change is dependent on the kinetic response of each sensing layer to the gas. The conductometric sensors have a marked non-linear profile with change in concentration. Several non-linear multivariate regression methods have been investigated to best calibrate the resulting signals from the mixtures of analyte gases: locally weighted regression (LWR), alternating conditional expectation (ACE), and projection pursuit (PP). In the best scenario, these non-linear regression methods have predicted mixtures of methanol and hydrogen gas to within 10 μmol/mol air (10 ppm) when calibrated within a concentration range of 0–150 μmol/mol air (150 ppm).