Sequence Classification in HSM: Enhancing Processes with Artificial Intelligence

Resumo

The selection of rolling sequence types in the Hot Strip Mill (HSM) in ArcelorMittal Tubarão (Largest steel producer in Latin America and global market leader) is traditionally based on empirical evaluation by operational experts. However, the complexity arising from order variability, slab yard dynamics, and operational constraints limits the efficiency of manual decision-making. This study proposes a methodology to automate the classification of sequence types, based on feature analysis of approximately 360 empirically assessed sequences. Preprocessing and feature engineering techniques were employed to support supervised learning models trained to categorize four main sequence types. The results show consistent performance, with high levels of accuracy, sensitivity, and specificity, highlighting the potential of the approach to improve in 10% process efficiency and compliance with key production KPIs such as decarbonization and customer satisfaction.