ENHANCING THE EFFECTIVENESS OF TOXIC GAS IDENTIFICATION USING A MOS MULTI-SENSOR AND MACHINE LEARNING MODELS
Abstract
An electronic nose is defined as a smart device for the detection and analysis of gases. It typically consists of two main components: an array of sensors (olfactory system) and an intelligent processing unit (brain). This study presents the design of a gas-sensing device utilizing a multi-sensor chip based on metal oxide semiconductors (MOS). Surveys measuring the response to various concentrations of harmful gases, such as NH3, CO, and NO2, were conducted. The measurement data indicate that the employed multi-sensor chip with three MOS sensors exhibits excellent selectivity for each gas. This outcome also demonstrates that using a sensor array allows for easier identification of gases compared to using a single sensor. Additionally, several typical machine learning models in artificial intelligence (AI), including PCA, LDA, SVM, DT, and RF, were employed to analyze gas response data. The performance of these models was evaluated based on the accuracy rate of gas sample identification. The results reveal that the utilization of machine learning models has enhanced the efficiency of gas classification, particularly with models such as DT and RF. This research may provide valuable contributions to the design of electronic noses for the analysis of multiple gases in various environmental settings.