Recognition and quantification of different VOCs by using impedance-spectroscopy-based gas sensors

Abstract

Pristine SnO₂ and WO₃ nanostructures decorated with gold and palladium were fabricated to detect volatile organic compounds (VOCs), including acetone, benzene, ethanol, formaldehyde, toluene, and xylene, under dry and humid conditions. Direct and alternating current signals were collected, and multilayer perceptron (MLP), convolutional neural network (CNN), and long short-term memory (LSTM) models were employed to differentiate among the VOCs. The transient variations in resistance was used as a fingerprint to identify VOCs. Normalized resistance signals were used as inputs to train the neural-network-based models to circumvent feature extraction and response calculation. The real and imaginary contributions of AC impedance was analyzed to determine the sensing properties. The relaxation behaviour of SnO₂ and WO₃ facilitated calculation of tunable responses for different frequencies. Different technique was used to compare multiple instantaneous $Z^{″}$ values at different frequencies to quantify the responses. The methods were validated using different sensors and VOCs in dry and humid atmospheres. Impedance spectra from different sensors in various VOC atmospheres were converted to pixelated images and CNN and LSTM models were adopted to discriminate different VOCs. A discrimination efficiency of 100% was achieved in cross-validated training and testing procedures for all collected data.