Gas sensors integrated with machine learning algorithms have aroused keen interest in pattern recognition, which ameliorates the drawback of poor selectivity on a sensor. Among various kinds of gas sensors, the yttria-stabilized zirconia (YSZ)-based mixed potential-type sensor possesses advantages of low cost, simple structure, high sensitivity, and superior stability. However, as the number of sensors increases, the increased power consumption and more complicated integration technology may impede their extensive application. Herein, we focus on the development of a single YSZ-based mixed potential sensor from sensing material to machine learning for effective detection and discrimination of unary, binary, and ternary gas mixtures. The sensor that is sensitive to isoprene, n-propanol, and acetone is manufactured with the MgSb₂O₆sensing electrode prepared by a simple sol-gel method. Unique response patterns for specific gas mixtures could be generated with temperature regulation. We chose seven algorithm models to be separately trained for discrimination. In order to realize more accurate discrimination, we further discuss the selection of suitable feature parameters and its reasons. With temperature regulation coefficients which are easily available as feature input to model, a single sensor is verified to achieve elevated accuracy rates of 95 and 99% for the discrimination of seven gases (three unary gases, three binary gas mixtures, and one ternary gas mixture) and redefined six gas mixtures. This article provides a potential new approach via a mixed potential sensor instead of a sensor array that could provide a wide application prospect in the field of electronic nose and artificial olfaction.

原文链接：https://pubs.acs.org/doi/10.1021/acssensors.3c01698