IMPROVEMENT IN SIGNAL EVALUATION METHODS FOR SEMICONDUCTOR GAS SENSORS

Applied chemical sensor research focuses on sensor arrays and signal evaluation methods, to improve reliability, selectivity and other features of the single sensor. State-of-the-art is the use of self-adapting systems like artificial neural networks (ANNs), mostly used for classification purposes. Systems for the prediction of gas concentrations were seldom investigated, because one of the main problems for those signal processing systems is the enormous amount of training data and the time dependency of the sensor signal. This work uses an array of semiconductor sensors (RGTO method and commercial sensors) and a modified ANN method for signal processing. After a drift correction based on an empirical model, a feed forward network predicts gas concentrations more precisely. A new method, the dynamic test point distribution (DTPD) has been invented, which achieves a significant reduction in the calibration time, together with a high accuracy in calculating the gas concentration.