A new approach to analyzing gas mixtures

A new approach is presented for analyzing gas mixtures by transforming the problem into a pattern classification one to reduce the effect of the poor repeatability of sensor response on the prediction of gas concentration. Th e aim of numerical simulation is to determine how successfully the approach using the combination of artificial neural networks with multi-sensor arra ys can analyze multi-component gas mixtures. The results indicate that the new approach is realistic for gas mixture analysis, and numerical simulatio n is a powerful tool to determine the architecture of a network. (C) 2001 E lsevier Science B.V. All rights reserved.