NEURAL-NETWORK SENSITIVITY ANALYSIS OF THE DETECTED SIGNAL FROM AN SO2 ELECTRODE

A novel SO, electrode was made by doping polyaniline onto Nafion to de tect SO, at ppm level with a linear correlation between the response c urrent and the SO, concentration in the range of 20 to 250 ppm. By app lying artificial neural network, it was possible to predict the SO, co ncentration so that a shorter response time (3 min compared with 6 min ) was achieved without the need of a noise filter. The operation condi tions, cycles for acid/base treatment and immersion time, considered a s the affecting factors of the polyaniline membrane treatment were use d for the study of sensitivity analysis. Both the results analyzed fro m the transfer functions of 2/(1 + e(-x)) and sgn(x).x(2)/(1 + x(2)) a re presented. Prediction of the neural network for sensitivity analysi s of the SO, electrode signal was tested first. Then, the two operatio n variables considered as important factors affecting the properties o f the membrane were studied in view of their influence toward the resp onse current. (C) 1998 Elsevier Science S.A. All rights reserved.