Identification of dangerous states in chemical batch reactors using neutral networks

This paper presents a neural network approach to the identification and diagnosis of dangerous states in chemical batch reactors. The efficiency of this approach has been proven when monitoring an exothermic chemical process, i.e. the esterification between acetic anhydride and methanol. For training and for testing the state classifiers, data sets delivered both from a process simulator and measurements in a laboratory reactor were used. The classification behaviour of neural networks is compared with fuzzy pattern classification. Results show that perceptron networks might be successfully applied as an additional supervision method to support the operator in making decisions under critical situations.