Robust and Rapid Detection of Mixed Volatile Organic Compounds in Flow Through Air by a Low Cost Electronic Nose

This work aims to detect volatile organic compounds (VOC), i.e., acetone, ethanol and isopropyl alcohol (IPA) and their binary and ternary mixtures in a simulated indoor ventilation system. Four metal-oxide-semiconductor (MOS) gas sensors were chosen to form an electronic nose and it was used in a flow-through system. To speed up the detection process, transient signals were used to extracted features, as opposed to commonly used steady-state signals, which would require long time stabilization of testing parameters. Five parameters were extracted including three in phase space and two in time space. Classifier and regression models based on backpropagation neural network (BPNN) were used for the qualitative and quantitative detection of VOC mixtures. The VOCs were mixed at different ratios; ethanol and isopropyl alcohol had similar physical and chemical properties, both being challenging in terms of obtaining quantitative results. To estimate the amounts of VOC in the mixtures, the Levenberg-Marquardt algorithm was chosen in network training. When compared with the multivariate linear regression method, the BPNN-based model offered better performance on differentiating ethanol and IPA. The test accuracy of the classification was 82.6%. The concept used in this work could be readily translated for detecting closely related chemicals.