A FIELD-PORTABLE GAS ANALYZER WITH AN ARRAY OF 6 SEMICONDUCTOR SENSORS - PART 2 - IDENTIFICATION OF BEER SAMPLES USING ARTIFICIAL NEURAL NETWORKS

A method is described based on a new portable, multisensor gas analyze r applying an artificial neural network to discriminate between six be er brands. The gas analyzer as previously reported employed six differ ent tin-oxide semiconductor sensors, and the detection method was base d on headspace analysis of the vapor above beer samples. The artificia l neural network (ANN) used in this study was a three-layer network, s tandard back-propagation algorithm, The network was trained with the u se of 553 cycles in 7.11 min at a learning rate of 1.0 and training to lerance of 0.1, A Macintosh PowerBook 1400cs with PowerPC(TM) 603e at 133-MHz clock frequency and 128-kilobyte Level Two write-through cache memory on a processor system bus was used to train the ANN, This stud y indicates that the portable, multisensor analyzer is able to discrim inate among beers and thus may be used to monitor beer product quality in industrial processes, having the advantage of portability and low cost for use in sites remote from chemical laboratories. Further appli cations in food technology are in the testing of foods and beverages f or quality and shelf life. (C) 1998 John Wiley & Sons, Inc.