An artificial neural net based rise time reduction for chemical sensors

During the last years the application of Artificial Neural networks (ANN) has proved to be a model independent instrument for signal evaluation systems. ANNs improve commonly the performance of single sensors and sensor arrays. Usually sensor signals near the equilibrium were used to train and test ANNs. All chemical sensors have certain time constants to reach their equilibrium, which range from a few seconds up to minutes. Therefore, an accurate classification and prediction of gas concentrations by ANNs are even possible minutes after a sudden gas concentration change. For many applications of gas sensing devices a fast classification of gases with the pattern recognition system is necessary. This work systematically investigates the properties of ANNs handling with time dependant sensor signals, which have not reached the equilibrium.