Machine Learning for the Steel Industry: Behind the Buzzword

In 2020, every major company’s annual report contained the word digitalization, A.I. or industry 4.0. It is easy to perceive these as buzzwords, aimed at investors, but the reality is more complex: companies are expected to transform now, driven by the fear of becoming obsolete. As researchers, this exciting transition creates significant opportunities: huge amounts of data are becoming readily available, while computing power and machine learning (ML) algorithms are more accessible than ever. However, this is also leading to disproportionate hopes and expectations regarding the actual capabilities of such methods, that only a working knowledge of ML combined with technical expertise in your field can rationalize. As R&D engineers, this critical view will be expected from you. Since technical expertise has already been the focus of your professional career, the effort should therefore be put on acquiring a practical knowledge of ML, that is, what problems can be solved and how to solve them? In this talk, some applications of ML to solve industrial issues (predictive modeling, visualization, combination with physical models...) will be discussed. Furthermore, practical aspects, such as data preparation, models implementation and maintenance will be reviewed, with the aim of providing actual insights on the root causes of successes and failures of ML applied to the steelmaking process.