Neural net analysis of the liberation of gold using diagnostic leaching data

Abstract

The interrelationship between mineral liberation and leaching behaviour of a gold ore is ill defined, mainly due to the complexity of both leaching and mineral liberation. A better understanding of this relationship could result in lower operating costs on gold extraction plants, since an increase in the efficiency of gold dissolution and a decrease in costs related to the crushing and grinding operations could be expected. In this investigation artificial neural nets were used to analyse diagnostic leaching data of gold ores obtained from South African gold mines. A self-organising neural net with a Kohonen layer was used to generate order preserving topological maps of the characteristics of both the unmilled and milled ores. The arrangement and shapes of these clusters could then be used to develop simple neural net models which were capable of predicting the degree of liberation more accurately than previously proposed models. Moreover, the neural net models were also capable of providing direct estimates of the reliability of their predictions by comparing new inputs with the data in their training bases.