Synergizing First-Principles and Machine Learning: Predicting Steel Flatness in the Era of Digital Twins and Physics-Informed Intelligence

Flatness in steel strips and sheets is a paramount quality metric in steel manufacturing, vital for ensuring product integrity and safety during production. Contemporary rolling processes adeptly maintain local flatness, but challenges arise particularly with thinner and high-strength steel grades. These complexities manifest as pronounced flatness deviations during subsequent processing and post-production phases. Merging traditional first-principle models with advanced machine learning algorithms, this new initiative achieves a comprehensive holistic grasp of the origins of flatness discrepancies. Taking into account the physical nature of the forming process. This synergy offers plausible prediction of cross-process flatness deviations while ensuring transparency and interpretability. Key advancements of this method are the development of predictive models, real-time flatness monitoring tools and a state-of-the-art digital twin structure connected to a network of software agents. Through sensitivity analysis, critical process parameters influencing flatness are identified. These outcomes not only elevate flatness control and product quality but also amplify plant efficiency. Ultimately, the research holds the potential for significant cost savings for steel producers, augmented product standards, and a dedicated focus on sustainable production with minimized environmental repercussions.