Signal conditioning for semiconductor gas sensors being used as detectors in gas-chromatographs and similar applications

The ability of feedforward neural networks to approximate the dynamic behav ior of linear and non-linear sensors was investigated. The networks were tr ained with data from simulated and real systems to compute the time varying sensor input from the measurable sensor output. Characteristic curves desc ribing the static and dynamic network properties were calculated. The netwo rk performance was tested and visualized in time and frequency domain (Bode diagram). It is shown that feedforward neural networks are an appropriate tool to approximate the dynamic behavior of non-linear gas sensors. (C) 199 8 Elsevier Science S.A. All rights reserved.