An LNN Model for the Dynamic Response of MOS Gas Sensors

With the expanding applications of electronic noses in areas such as agriculture, petrochemicals, and environmental monitoring, improving their classification accuracy and gas concentration detection precision is essential. Electronic noses often employ neural networks to process their data, and these neural networks require a substantial number of test samples for training. Therefore, it was necessary to obtain a large number of training samples. Based on the sensor's competitive adsorption and desorption properties of the mixed gas, the chemical reaction between the different gases, the chemical reaction between mixed gases and metal oxides, and the transport characteristics of flow carriers, 14 state variables were determined to be used to construct the sensor dynamic response model of a MOS (metal oxide semiconductor) gas sensor through the liquid neural network. Simulations and experiments demonstrate that the model effectively produces large training datasets from small amounts of test data and achieves higher concentration prediction accuracy compared to existing models.