DYNAMIC MODELING OF COMPETITIVE ELUTION OF ACTIVATED CARBON IN COLUMNS USING NEURAL NETWORKS

In previous papers the mechanism and dynamics of the elution of gold c yanide from activated carbon have been investigated in detail. Sub-pro cesses such as the pre-soaking step, the degradation of cyanide, the e lution of the spectator cations, the associated shift in the equilibri um of adsorption or desorption as a result of the removal of cations, the reactivation of the carbon surface, and the elution of gold cyanid e have been explained quantitatively to some extent, although further work is evidently required. Previous work has also shown that equilibr ium conditions may be assumed when adsorption is weak, hence when aggr essive pre-soaking conditions have been used. However, these studies h ave not taken the competitive effect of base metals into account, alth ough this is known to have an adverse effect on the efficiency of gold elution. The present study has shown quantitatively that copper has a significant effect on the recovery of gold. Nickel and silver also ha ve a detrimental effect, but only if they are present as high loadings . In contrast, the elution of the base metals is to a large degree una ffected by the elution of gold. It is shown in this paper that the mul ti-component equilibrium relationship between the spectator cations an d the various metal cyanides can be very complex, and perhaps ill-defi ned. In such circumstances it is preferable to use a non-parametric te chnique such as a back-propagation neural network to represent such an equilibrium relationship. Owing to the difficulty of estimating the f inal conditions of the pre-soaking step, it is not always possible to predict the exact level of equilibrium. Therefore, it could be necessa ry in practice to adjust the equilibrium predicted by a neural net by a factor which is dependent on the conditions of pre-soaking.