A Review of Strip Shape Prediction Methods Transitioning From Mechanistic Models to Interpretable Deep Learning

Strip shape prediction is a core task in rolling process control, and its accuracy directly impacts product quality and production efficiency. However, complex multivariable nonlinear couplings, dynamic operating conditions, and strong environmental disturbances severely challenge the accuracy and adaptability of traditional mechanistic models. In recent years, machine learning (ML) has provided new pathways to overcome these limitations. Although research achievements are increasingly abundant, the field still lacks a systematic review that integrates technological evolution with industrial application experience. This paper systematically reviews the research progress of ML in strip shape prediction, tracing the evolution from mechanistic modeling to data‐driven approaches. By detailing the core modeling stages and using strip crown, flatness, and key influencing factors as specific contexts, we systematically categorize the functional positioning and applicable scenarios of various ML algorithms. Crucially, we outline their structural limitations in practical deployment, such as “black‐box” unreliability and a lack of physical consistency. Based on this synthesis, the paper explores current industrial bottlenecks and future development directions, specifically highlighting interpretable deep learning, mechanism‐data fusion, and causal inference. Ultimately, this review aims to provide valuable references and insights for academia and industry professionals striving to advance intelligent steel production and enhance process efficiency.