Radial basis neural networks for identification of volatile organic compounds

In this paper, radial basis neural network (RB-NN) was proposed for the identification of volatile organic compounds (VOCs). The measuring system with four 20 MHz quartz crystal microbalances (QCMs) as sensors was used in the experiments. The four sensors were modified with SnCl2 and PdCl2 to change the response characteristics. A flow-through type system was used to measure the VOC samples including ethyl alcohol, acetone, chloroform, and de-ionized water. Rise-time, peak, and fall-time data from the response characteristic curves were used as information for training the neural networks. It was found that the RB-NNs could be learned faster and better than the conventional back-propagation neural networks (BP-NNs). The samples were clearly separated and recognized with the RB-NNs, which could not be done with the BP-NNs.