Modelling of the mass transfer in gas-sparged electrolysers with neural nets

Abstract

This paper illustrates the application of a neural-net model to the modelling of mass transfer in gas-sparged electrolysers. The modelling of gas-sparged electrolysers was found to be very difficult using traditional methods, e.g. by semi-empirical fundamental mass transfer correlations or empirical regression equations. It will be shown that the proposed neural-net model is superior to the mentioned traditional methods. The numerous fundamental approaches to modelling the mass transfer in gas-sparged electrolysers are discussed initially, pointing to their non-generalized nature. Data, which were generated in electrolytic airlift and bubble column reactors for the purpose of this investigation, were fitted to these data. It was found that the fundamental models available in the literature were incapable of modelling the data of both the reactors simultaneously. As a consequence, another theoretical basis had to be found, which neural nets presented. It is shown that the neural net trained with data from the work described in this paper could predict data of other authors as well, a generality which is not inherent in any of the mass transfer correlations proposed for mass transfer in two-phase systems to date. A sensitivity analysis using the trained neural net revealed a non-linear relationship between log ( k), the mass transfer coefficient, and d, the bubble diameter, a result that fundamental as well as empirical approaches could not reveal.