Title

COIL QUALITY PREDICTION BASED ON STEELMAKING SHOP VARIABLES USING MACHINE LEARNING

COIL QUALITY PREDICTION BASED ON STEELMAKING SHOP VARIABLES USING MACHINE LEARNING

Authorship

Alves, Estelita Simoes Ribeiro; Ney, Vitor Bogaci; Soares, Danilo Nunes

Alves, Estelita Simoes Ribeiro

I am this author

Ney, Vitor Bogaci

I am this author

Soares, Danilo Nunes

I am this author

DOI

10.5151/2594-5297-39639

Downloads

Abstract

The sliver defect leads to the rework, downgrading, or scrapping of products, which results financial losses for the company and might impact order weight targets that directly affect production planning. The later the problem is detected in the process, the greater the impact. Therefore, this work aims to detect, in the steelmaking shop, the probability of a slab presenting sliver defect in consecutive rolling processes. For this purpose, a Machine Learning model was developed, which takes as input process variables such as secondary refining and continuous casting and data that characterizes the material, such as chemical composition. An optimizer that defines probability ranges for each decision was developed to support the use of probabilities in decision-making with better cost-benefit. The ranges defined by the optimizer and the probabilities generated by the model are integrated into the current judgment system of ArcelorMittal Tubarão. The developed solution can potentially prevent 30.7% of sliver defects through predictive detection and preventive treatment. In addition, it provides an opportunity for slab recovery in a scenario where it is indicated as disqualified by the current judgment system and has a very low defect probability according to the model.

The sliver defect leads to the rework, downgrading, or scrapping of products, which results financial losses for the company and might impact order weight targets that directly affect production planning. The later the problem is detected in the process, the greater the impact. Therefore, this work aims to detect, in the steelmaking shop, the probability of a slab presenting sliver defect in consecutive rolling processes. For this purpose, a Machine Learning model was developed, which takes as input process variables such as secondary refining and continuous casting and data that characterizes the material, such as chemical composition. An optimizer that defines probability ranges for each decision was developed to support the use of probabilities in decision-making with better cost-benefit. The ranges defined by the optimizer and the probabilities generated by the model are integrated into the current judgment system of ArcelorMittal Tubarão. The developed solution can potentially prevent 30.7% of sliver defects through predictive detection and preventive treatment. In addition, it provides an opportunity for slab recovery in a scenario where it is indicated as disqualified by the current judgment system and has a very low defect probability according to the model.

Keywords

Machine Learning; Sliver; Continuous Casting; Quality

Machine Learning; Sliver; Continuous Casting; Quality