Hot strip mill process optimization with machine learning: systematic review and methodical prediction framework based on open-source data

Abstract Although hot strip rolling is a fundamental process that has been established over the last decades, industry and science are continuously working on optimizing the rolling parameters and their influence on the final product. A major problem is that geometric errors in hot-rolled coils, which are subsequently processed in downstream processes such as cold rolling, can only be corrected to a limited extent afterwards. Consequently, data-driven methods, especially machine learning (ML), have become increasingly important for predicting key process and quality variables like strip thickness, width and the strip shape in hot strip mills (HSM). To provide a structured overview of the current state of research, this article presents a systematic literature review of ML-based prediction approaches addressing relevant quality and process parameters in HSM. The studies examined are categorized according to the predicted variables, plant sections, and applied modeling approaches, with a particular focus on their relevance to be deployed for potential process optimization. In a second step, representative prediction models are implemented for the main focus areas of the literature search, using publicly available HSM datasets. These are methodologically based on the essential part of the published literature, use common ML algorithms, and support the discussion on dataset dependencies, methodical limitations, and challenges in comparing reported prediction results across different studies. By combining literature insights with reference predictions, this work clarifies the role of ML-based prediction in hot strip rolling, outlines potentials and limitations in model deployment for process optimization and identifies key research gaps for future research.