Impedance based detection of HMDSO poisoning in metal oxide gas sensors

Abstract

Abstract A commercial SnO2-based metal oxide gas sensor (UST GGS 1330) operated at a constant temperature of 170 ℃ was evaluated in a frequency range from 40 Hz to 110 MHz using an Agilent 4294A high precision impedance analyzer. The sensor was exposed to carbon monoxide and ethanol at three concentrations each (1, 2 and 5 ppm) and at humidities of 40%rh and 60%rh. After application of the test gas profile, the sensor was repeatedly poisoned with 9.3 ppm of HMDSO for 20 min, and the gas test was repeated up to an overall poisoning dose of 930 ppm min (i.e. five times). Impedance data exhibit characteristic features for the different test gases as well as for sensor poisoning. Using Haar-Wavelet transformation and Adaptive Linear Approximation for feature extraction followed by feature selection with Recursive Feature Elimination Support Vector Machines a total classification rate well above 98% was achieved for test gas type and concentration as well as sensor poisoning state with linear discriminant analysis and a Mahalonobis distance classifier.