EUROPEAN STEEL TECHNOLOGY AND APPLICATION DAYS 2017 - ESTAD 2017

There is one challenge known to all metallurgists: how much of each ferroalloy to add during steelmaking process in order to ensure the required chemistry of the steel at the lowest possible cost. As ferroalloys participate in numerous chemical reactions, the final absorption of the added elements depends on many factors, which measurements are rough or even unknown. The decision on the exact amount of ferroalloys to be added often relies on a combination of knowledge-based models and expert judgement of the operator, leading to many suboptimal results. Machine learning is a novel approach that allows to consistently increase the quality of decision-making, resulting in a decrease in the overall ferroalloy use without deterioration in the quality of the resulting product. The proposed solution consists of two parts: (1) the ferroalloy absorption model. Trained on available historical data on previous smeltings at a given plant, this model takes all available parameters – the mass of scrap and crude iron, results of chemical analyses, amounts of the added ferroalloys, and predicts the expected chemical composition of steel. Use of machine learning techniques allows forecasting the deviations of traditional physics-based models, and significantly increases the accuracy of prognosis. (2) optimization module. The module recommends the amounts of ferroalloys required to produce a specific steel grade at the lowest possible cost, while maximizing the confidence of meeting chemical composition requirements. The pilot tests in a production environment have demonstrated that the use of machine learning-based recommender system allows decreasing the use of ferroalloys by up to 5%, while confidently meeting quality requirements for a specific steel grade.