Identifying Dangerous States in Chemical Plants Using Neural Networks

This paper describes the application of three-layer perceptron networks to the identification and diagnosis of dangerous states in strongly exothermic semibatch reactions. To assess the potential danger of different faults, separate perceptron networks were used for danger assessment and for fault isolation. Results are presented to illustrate the performance of the neural-network approach on real process data obtained from typical faults of a catalytic esterification process which were simulated in a laboratory reactor.