Quantitative determination of volatile compounds in a mixture using a single thermally modulated metal oxide semiconductor gas sensor and convolutional neural networks

Cross-sensitivity of metal oxide semiconductor (MOS) gas sensors has often hindered their application in analyzing volatiles compounds found in mixtures. Selectivity of these detectors can be enhanced through its thermal modulation, as due to their sensing mechanism, the operating temperature affects the interaction between gas molecules and the surface of the semiconductor. Thermal modulation of the sensor makes it possible to obtain response patterns containing information concerning the mixture composition; however, their sophisticated nature requires advanced interpretation techniques. The study proposes a new methodology for qualitative and quantitative determination of volatile components (acetone and isopropanol), co-occurring in mixtures, based on a single thermally modulated MOS gas sensor combined with Convolutional Neural Networks (CNNs). The conducted analysis revealed high usefulness of 2-Dimensional CNNs in extracting essential information from MOS gas sensor output signals recorded for mixtures containing various concentrations of the analyzed compounds. The developed 2D-CNN model demonstrates high accuracy and predictive capabilities (R2 = 0.9967 – 0.9996, RMSE = 0.0066 – 0.0161, MAE = 0.0053 – 0.0114) across a wide range of analyzed compound concentrations (31 – 1000 ppm). These results indicate high potential of the developed methodology for precise qualitative identification and quantitative analysis of mixtures containing volatile compounds, extending beyond the confines of this study and opening avenues for further exploration and application in various measuring systems utilizing MOS gas sensors.