Metallurgy‐Guided Process Window Optimization and Stability Analysis for Customized Hot‐Rolled Steel Production

To meet the stringent quality demands for hot‐rolled medium‐thick plates used in safety‐critical structural applications, this study proposes a Residual Hybrid Prediction and Optimization (RHPO) framework to address dual challenges of performance stability and process customization. On the prediction side, a Residual Hybrid Metallurgy‐Informed Learning (RHML) model is developed, which integrates mechanism‐aligned feature construction with residual correction. This approach combines microstructural evolution knowledge with industrial data to enhance both the accuracy and interpretability of property predictions, including yield strength (YS), tensile strength, and elongation. To guide robust process decision‐making, a Groupwise Particle Beam Optimization (G‐PBO) method is introduced. G‐PBO extends conventional particle swarm optimization by incorporating sensitivity‐guided weighted control and groupwise search, thereby transforming single‐point parameter tuning into a robust process window design. Validation using a real industrial dataset of 4,547 samples from a Chinese steel enterprise demonstrates that RHML significantly improves prediction accuracy. Furthermore, G‐PBO effectively constrains performance fluctuations, achieving YS stability within ±20 MPa with a 100% qualification rate, surpassing the 82.17% benchmark. Application studies confirm that the framework delivers feasible process window optimization under strict equipment constraints. These results highlight RHPO as a unified, reliable paradigm for flexible and high‐quality hot rolling production.