Detection of Formaldehyde in Mixed VOCs Gases Using Sensor Array With Neural Networks

Abstract

A four-sensor array with neural networks was developed to identify formaldehyde in three possible interfering volatile organic vapors, such as acetone, ethanol, and toluene. The sensor array consisted of four metal oxide-based gas sensors: two of them are commercial SnO2 sensors, other two sensors are made in our laboratory. The responses of the sensors to each gas and to the mixture of two or all of them were tested and evaluated. It was found that every sensor has response to these four kinds of gases, and the response value of each sensor to the mixture gases was lower than the simple added value of the responses to each gas. This phenomenon is due to the properties of gas and the sensing materials. For recognizing formaldehyde in the background of ethanol, acetone, and toluene in air, 108 gas samples were tested taking into account of possible practical concentrations. Among these samples, 91 samples were used for training the pattern recognition methods and 17 samples for testing the robustness. Three neural networks were used in this report, including back propagation neural network support vector machines (SVM) and extreme learning machine (ELM) with principal component analysis (PCA). The PCA helps to improve the accuracy of the ELM by preprocessing the sensor data, while the SVM method achieves the best accuracy. The ELM method indicates a better way to train the sensor array and to identify the particular gas species with very less training time and good accuracy.