Application of data analytics and machine learning: A steel manufacturing facilities

Abstract

Statistically, global steel production data, especially regarding specialty steel, is crucial in high-quality steel manufacturing. This study focused on estimating the scrap generated during the production of round steel material in the rolling mill of a high-quality steel manufacturing facility. The impact of element ratios in the steel on scrap quantity was analyzed. A dataset was created using Oracle PL/SQL and Python, which was then cleansed of outliers. The analysis identified specific quality and dimensions linked to the highest scrap quantities, as well as the quality most responsible for scrap based on production input. The dimensions and quality associated with the highest production volumes were also determined. The element ratios within the material were examined to ascertain which element significantly influenced the scrap quantity. Additionally, it was analyzed which quality and size consumed more energy, impacting material pricing. Seven machine learning algorithms were developed, including four regression and three classification algorithms. These algorithms were evaluated using performance metrics. Among the regression algorithms, the Random Forest algorithm showed the best overall performance. For the classification algorithms, the K-Nearest Neighbors algorithm exhibited the best overall performance. In addition, an application was developed to display the results of model performance metrics based on the input parametric values.