Abstract
Driven by recent advances in machine learning, steel rolling is transitioning toward intelligent, data-centric operation. This study presents a unified machine learning framework for predictive, closed-loop quality control of steel strips across hot and cold rolling processes. The model explicitly quantifies the complex, nonlinear effects of key operational parameters, such as rolling force and gap settings, on final product quality. The proposed framework enables real-time monitoring and dynamic compensation of dimensional deviations and shape defects, thereby improving dimensional consistency and process stability. Additionally, a multimodal perception-based system is introduced for early anomaly detection and coordinated parameter optimization, facilitating adaptive setpoint adjustment and proactive defect mitigation. Collectively, these machine learning-driven approaches enhance product uniformity and rolling efficiency while offering a scalable pathway toward more autonomous, resource-efficient, and sustainable rolling operations, aligning with the paradigm of AI-driven sustainable manufacturing.