Rapid Detection of CO, H2S, and NO2 Mixtures Using an Integrated SnO2-Based Sensor Array Combined with Machine Learning.

The increasing interest in the precise detection of mixed gases in the industrial field has led to an unprecedented level of demand for gas sensing equipment. However, a single sensor with the classic single-output signal and cross-sensitivity can no longer meet the application requirements for detecting complex gas environments. Herein, we propose a feasible way to realize the low-cost fast detection of CO, H2S, and NO2 mixed gases based on an integrated sensor array. Sensor units with significant response distinguishability are designed. The rapidly switched heating signal is applied to the microheater, and then the time-domain and frequency-domain feature parameters are extracted to train the machine learning models. The results show that the proposed method significantly reduces the required data set size for rapidly detecting mixed gases, and successful classification and prediction can be achieved using only the response data of the first 20 s of the adsorption process. Eventually, an average accuracy of 96.30% and a determination coefficient (R2) of 0.97 have been achieved for the concentration classification and prediction toward the CO, H2S, and NO2 ternary mixtures. It is now possible to move an important step toward fully utilizing metal oxide semiconductor sensor arrays for rapid gas sensing applications.