ABSTRACT
The characteristics of multi-dimensional sensor signals obtained from 10 sensors were analyzed using the PCA, and we developed an electronic nose system that can not only classify the kinds of the combustible gases but also provide the concentration values of the identified gas. Above all, we adapted a multi-layer neural network with an error-backpropagation learning algorithm for identification of the kinds of gas and for quantifying of gas concentration. Using the sensing signals of the arrays and neural networks, a gas pattern recognition was then implemented using a DSP board with the aim of real-time classifying and quantifying the combustible gases, including butane, propane, methane, and carbon monoxide below their explosion limit values (TLVs).
