Spike detection using advanced analytics and data analysis

Abstract

Anode spike crises have a deleterious effect on current efficiency to the point of jeopardizing smelter operation. All the time spent with a spike under an anode is a period where current is lost, and the longer the spike remains present, the more serious are the mid-term consequences for the reduction process. It is therefore most important to detect these spiky anodes as early as possible and remove them from the pots. A new tool based on anode current measurements, combined with machine learning, has been developed and tested. It is an effective way of detecting many of these spikes, usually a few days before they become obvious. This article describes the development of the tool and the first results obtained on industrial cells.