Artificial Intelligence Approaches For The Ladle Predictive Maintenance In Electric Steel Plant

Abstract

This paper presents an Artificial Intelligence based tool for real-time estimation of ageing status of all the ladles operating within an electric steelworks. The developed system exploits real data collected and operators’ experience for tuning its core models and formulates the problem as a multi-class classification problem on a highly imbalanced dataset. Two classifiers are applied, based on the Decision Tree and Random Forest approaches. Both approaches provide satisfactory results, especially in the identification of the most critical situations, namely when the ladle is close to its end of life or needs an immediate maintenance intervention, but the Decision Tree shows slightly better performance. The system, therefore, effectively supports predictive maintenance approaches.