Prediction and Feature Analysis of Entrapped Slag Defect on Casting Slab Based on Optimized XGBoost Model

Machine learning (ML) algorithms have been proven to be effective methods in steel design and production processes. In this study, the feature analysis and prediction of slag entrapment in continuous casting process were carried out based on XGBoost model. Results show that the area under the curve (AUC) and accuracy of BO-XGBoost are 0.811 and 0.756, respectively, which are higher than PSO-XGBoost and GA-XGBoost. Feature analysis shows that mold flux, oscillation frequency, serial number, inner arc, casting speed (CS) and slab width are the key features which affect the entrapped slag defect, based on Gini coefficient and SHAP values. In addition, partial dependence plots of SHAP value suggest that it is conducive to product the qualified slabs with less entrapped slag defect, when the oscillation frequency is 60–140 cpm, inner arc is 7.5–9.5 m, CS is 0.35–1.2 m/min, serial number is less than 5, slab width is less than 1300 mm and 2# mold flux.