Concentration Estimator of Mixed VOC Gases Using Sensor Array With Neural Networks and Decision Tree Learning

Abstract

This paper aims to estimate the concentrations of volatile organic compounds (VOCs), such as ethanol, acetone, formaldehyde, and toluene, in a quaternary mixture. In order to develop an electronic nose for practical applications, the concentrations of VOCs and their combinations in the mixture are randomly assigned. The sensor array consists of five metal oxide sensors, which are produced in a laboratory. The algorithm for the estimation of concentration is realized using a back-propagation neural network (BPNN) with two hidden layers and decision tree learning. First, the data set of the VOC mixture is divided into four subclasses based on the concentration using classification and regression trees. Second, every subclass is classified and regressed using a corresponding BPNN; furthermore, its four output nodes provide a continuous prediction of the concentration of each VOC in the mixture. A single BPNN with two hidden layers is also constructed and evaluated for the purpose of comparison. The maximum error in the concentration estimation of each VOC using the proposed method is approximately 2 ppm, and the accuracy is better than the result obtained using the single BPNN. Moreover, the relative error is less than 5% when the predicted concentration is higher than 20 ppm. This paper reports some aspects of the potential of using neural networks for quantitatively analyzing the concentrations of a VOC mixture.