Use of artificial neural networks in estimation of Hydrocyclone parameters with unusual input variables

The accuracy of the estimation of the Hydrocyclone parameter, d50/sub c/, can substantially be improved by application of an Artificial Neural Network (ANN). With an ANN, many non-conventional Hydrocyclone variables, such as water and solid split ratios, overflow and underflow densities, apex and spigot flowrates, can easily be incorporated into the prediction of d50/sub c/. Selection of training parameters is also shown to affect the accuracy.