Strip Centerline‐Offset Prediction and Decision Support Based on Mechanism Guidance and SHAP Interpretability

Strip centerline‐offset is a critical defect that severely impacts product quality and production safety in hot‐rolled processes. Currently, the adjustment of this issue primarily relies on the subjective experience of operators, lacking an intelligent and reliable decision‐support tool. To address this challenge, this article proposes a hybrid intelligent model that integrates mechanistic analysis and data‐driven modeling for strip centerline‐offset prediction and human–machine collaborative decision. First, through mechanistic analysis of the rolling process, key influencing factors are identified, providing a physically meaningful foundation for feature selection in the data model. Subsequently, a data‐driven model is developed, which includes a fault detection module and a fault prediction module. Crucially, the interpretation module employs explainable artificial intelligence to elucidate the decision‐making process of the predictive model, visualizing the contribution trends of key features. This provides operators with intuitive and quantifiable adjustment guidance, forming an effective human–machine collaboration closed loop. The practical application results demonstrate that the average proportion of the strip centerline‐offset within the qualified range at the finishing mill exit reaches 98.33%, representing an increase of more than 20% compared to the previous level, significantly enhancing production stability and product quality.