Data-Driven Prediction and Process Parameter Analysis of Zinc Coating Thickness in Hot-Dip Galvanizing

Precise control of zinc coating thickness is essential for ensuring the quality and performance of hot-dip galvanized steel products. In this study, production data from a steel plant were utilized to develop XGBoost-based predictive models for coating thickness under three zinc bath conditions zinc-iron alloy coating (GA), zinc-aluminum-magnesium coating (ZM), and pure zinc coating (GI). Unlike earlier ML-based studies that typically focus on a single coating type or a limited operating window, this work establishes and compares three dedicated models across distinct bath chemistries using unified process variables, enabling a cross-condition assessment of model robustness and controllability. The models demonstrated high predictive accuracy, achieving R² values of 0.80, 0.98, and 0.967, with mean squared errors of 2.235, 5.637, and 4.350, respectively. Feature importance analysis identified air knife pressure, strip speed, and nozzle-to-strip distance as the primary factors affecting coating thickness. Quantitative analysis revealed that coating thickness decreases linearly with increasing air knife pressure, with slopes of approximately -0.018, -0.125, and -0.128 g/(m²·MPa) for GA, ZM, and GI, respectively, and increases with strip speed, with slopes of +0.052, +0.16, and +0.16 (g·min)/m³. Importantly, the slope magnitudes differ substantially across bath conditions, indicating that control sensitivity is coating-dependent. The influence of distance becomes pronounced beyond a critical threshold, exhibiting larger slopes relative to pressure and speed. These findings underscore the feasibility of employing machine learning for accurate prediction and control of coating thickness, and provide quantitative insights into the effects of air knife parameters to enhance product quality and process efficiency in hot-dip galvanizing.