Classification and regression models in Copper refinery

Abstract

The Chuquicamata Copper refinery has an annual production of 480, 000 Tons of copper cathode (A grade). The electrochemical process has a duration of 10 days with 300 A/m² current density. In this global context, there are a lot of process variables for the process control, like impurities, electrolyte flux in the cells, additive addition, short-cuts and electrical current efficiency. In the present work, classification and regression models are used for having a global process control. The classification models like SVM, Decision Trees, GLMNET, LDA, KNN and Logistic regression show an easy way to see the different effect of the process variables over the quality of the final product. The regression models show the future behaviour of process variables in different scenarios and how this result have a huge impact in the cost of the electrochemical process. In other line, the classification models are easy tool for the operation team. They can see the effect of process variables day by day in the electrochemical cell. The fusion of both models has a strong impact in the global process control for take future decision and minimising the process cost.