Materials Science-Poland, Vol. 25, No. 2, 2007 Application of an artificial neural network in the processing of output signals from a gas sensor with sol-gel-derived TiO 2 film A. ŁUKOWIAK 1 , K. KOZŁOWSKA 1* , K. URBAŃSKI 2 , A. SZCZUREK 3 , K. DUDEK 4 , K. MARUSZEWSKI 1,5 1 Institute of Materials Science and Applied Mechanics, Wrocław University of Technology, ul. Smoluchowskiego 25, 50-370 Wrocław, Poland 2 Faculty of Microsystem Electronics and Photonics, Wrocław University of Technology, ul. Janiszewskiego 11/17, 50-372 Wrocław, Poland 3 Institute of Environmental Protection Engineering, Wrocław University of Technology, pl. Grunwaldzki 9, 50-370 Wrocław, Poland 4 Institute of Machine Design and Operation, Wrocław University of Technology, ul. Łukasiewicza 7/9, 50-371 Wrocław, Poland 5 Electrotechnical Institute, Skłodowskiej-Curie 55/61, 50-369 Wrocław, Poland TiO 2 thin film obtained by the sol-gel technique was used as the active layer in an electric sensor to distinguish the vapours of four volatile organic compounds: hexane, hexanol, cyclohexane and benzene. The measurements were performed at various temperatures of the sensing layer. Some of the output sig- nals obtained from the sensor were characterized by low reproducibility, even within the data series ob- tained for the same gas. With the current design of the gas sensor, it was sometimes impossible to obtain a reproducible and stable output signal. Therefore, a neural network was used to pre-process the data. A bipolar transfer function of neurons was used as it had the shortest learning time of the network and produced the most stable results. The best results were obtained for a 4-4-4 topology of the neural net- work, where the input data were the values of the current at 440 and 360 °C when the sensor was exposed to a flow of air with or without organic vapours, with a 4-neuron hidden layer, and BE, CH, HL, HX outputs, each one associated with specific substance (benzene, cyclohexane, hexanol and hexane). The neural network was configured as a classifier recognizing four specific gases. Key words: smart sensor; volatile organic compound; sol-gel; artificial neural network; signal condi- tioner __________ * Corresponding author, e-mail: katarzyna.kozlowska@pwr.wroc.pl