Machine Learning-Assisted Gas-Specific Fingerprint Detection/Classification Strategy Based on Mutually Interactive Features of Semiconductor Gas Sensor Arrays

A high-performance machine learning-assisted gas sensor strategy based on the integration of supervised and unsupervised learning with a gas-sensitive semiconductor metal oxide (SMO) gas sensor array is introduced. A 4-SMO sensor array was chosen as a test sensor system for detecting carbon monoxide (CO) and ethyl alcohol (C2H5OH) mixtures using 15 different combinations. Gas sensing detection/classification was performed with different numbers of gas sensor and machine learning algorithms. K-Means clustering was successfully employed to rationally identify the similarity features of targeted gases among 4 different groups, i.e., matrix gas, two single-component gases, and one two-gas mixture, based on only unlabeled voltage-based gas sensing information. Detailed classification was performed through a multitude of supervised algorithms, i.e., 2-layer artificial neural networks (ANNs), 4-layer deep neural networks (DNNs), 1-dimensional convolutional neural networks (1D CNNs), and 2-dimensional CNNs (2D CNNs). The numerical-based DNNs and image-based CNNs are shown to be excellent approaches for gas detection and classification, as indicated by the highest accuracy and lowest loss indicators. Through the analysis of the influence of the number of sensors on the arrayed gas sensor system, the application of machine learning methodology to an arrayed gas sensor system demonstrates four unique features, i.e., a data augmentation methodology, machine learning approach of combining K-means clustering and neural networks, and a systematic approach to optimized sensor combinations, potentially leading to the practical sensor networks based on chemical sensors. Even two SMO sensor combinations are shown to be highly effective in gas discrimination against diverse gas environments assisted through numeric-based DNNs and image-based 1D CNNs, overcoming the simple clustering proposed through the unsupervised K-means clustering