Calcium-Treated Steel Cleanliness Prediction Using High-Dimensional Steelmaking Process Data

Abstract Control of calcium treatment in steel is challenging due to the reactivity Ca and difficulty measuring total oxygen in-process make actionable decisions. In this work, a method combining statistics process engineering are developed using partial least squares regression (PLS) predict non-metallic inclusion content (oxides CaS) composition at end ladle tundish extensive data SEM/EDS-based analysis. Total can be predicted an accuracy 7 ppm, Mg/(Mg+Al) ratio inclusions 3at% providing enough recommend addition based on thermodynamic calculation for liquid window. Alternatively, model fraction 20 ppm accuracy, accurately average CaS Ca/Al tundish. Model interpretability hindered by high dimensionality multicollinearity data. Non-metallic compositions correspond expected onset formation composition.