COMPLEX CHEMICAL SYNTHESIS : DATA ANALYSIS USING NEURAL NETWORKS

摘要

The results shown here come from the study of an industrial pilot plant, but, owing to confidentiality matters, neither the industrial partner nor the products involved in the process can be named. The target product is obtained by reacting a heavy organic compound with a multi-functional reagent, using solid catalyst. This reaction is followed by several unwanted ones, leading to 7 major by-products. Satisfying kinetic schemes are not available. The experimental installation uses a fixed catalytic bed, and operating conditions are chosen to maximise the reaction yield, like in industrial plants. Consequently, measurements of the by-product concentrations are inaccurate, and the available kinetic information is poor. After preliminary analysis of the experimental data, a neural network was fitted, able to determine the concentrations in the liquid and gas products of the reactor. The influence of operating parameter has been analysed. The results allowed to bring to the fore the appearance of an unexpected liquid phase in the reactor, and led to interesting informations about reaction kinetic. Neural networks allow to easily determine what the best operating parameter set is, to minimise any cost function. They allow to determine what the optimal working conditions are for a given plant, given the economic and technical conditions at given moment.