MODELING OF RARE-EARTH SOLVENT-EXTRACTION WITH ARTIFICIAL NEURAL NETS

The design and operation of mass transfer units such as rare earth sol vent extraction systems require accurate models of the mass transfer p henomena that occur in these systems, The modelling of rare earth solv ent extraction systems from first principles is severely constrained b y the physico-chemical similarities of the lanthanides and the strong interactions that can occur between the components of these systems, A rtificial neural networks are widely recognized as one of the fastest expanding computer technologies for the modelling of complex or ill-de fined systems that are difficult to model otherwise. In this paper, it is shown that the general mass transfer of rare earth solvent extract ion in various systems can be modelled significantly more accurately b y means of artificial neural network models as compared to conventiona l models. It is shown that the crystal radius of the lanthanide elemen ts can be used to generalize the behaviour of rare earths in solvent e xtraction systems, In cases where large numeric variations in data occ ur, the accuracy of the neural network models can be significantly aff ected by the logarithmic scaling of data. The incorporation of a self- organizing (Kohonen) layer in a neural network can give an improved pe rformance in systems with clustered data.