Abstract

Copper electrorefining is a rather complex process involving multiple operational variables that affect its performance. Even though modern tankhouses have incorporated mechanization and automation in material handling systems, process control is still modest. The vast availability of sensing equipment enables using a data-driven approach for generating an automatic control strategy capable of improving the performance of the electrorefining process. In this context, machine learning techniques gain relevance. This article analyses the opportunities and limitations of the use of machine learning in electrorefining for process control purposes. As a concrete example, a recent study developed in a major electrorefining tank house is described, where machine learning techniques were used on real industrial data for developing operational data-driven models capable of predicting production quality and process variables. The models were then used for generating a real-time Decision Support module, which is the first step towards developing a modern automatic control strategy for electrorefining.