The discrimination and concentration estimation of inflammable gases with tree-structured neurofuzzy algorithm

Describes the discrimination and concentration estimation of inflammable gases using a tree-structured neurofuzzy algorithm. A neural network and three fuzzy neural networks are connected in series in the algorithm; the former is employed in discriminating the kind of inflammable gases and the latter is engaged for estimating the concentration of the discriminated gas. Output signals from a semiconductor gas sensor operated periodically were sampled for a period and Fourier-transformed to divide them into AC and DC components. In the first stage, the AC components, which can hardly be influenced by the change of gas concentration are introduced to the neural network to discriminate the kinds of gases. In the second stage the DC components, which contain information of gas concentration, are introduced to a fuzzy neural network to estimate concentration of the discriminated gas. The proposed algorithm is examined in discrimination and concentration estimation of three kinds of inflammable gases, that is, hydrogen, butane and methane, and good performance is achieved