Key points are not available for this paper at this time.
Slag entrainment in the continuous casting mold has always been the focus of surface quality control of the slab for the steel strips. However, the invisible and nonlinear molten steel flow that determines the defect formation make online measurement and evaluation particularly difficult. We developed a fast prediction method visualising the characteristics of the melt flow and estimating the slag entrainment online by combining numerical simulation and machine learning. The data-driven surrogate model was constructed using the proper orthogonal decomposition method and a fully connected neural network based on the numerical simulation data. The model showed a high hit rate of over 91 % with an absolute error of less than 0.05, and exhibited millisecond-scale real-time responsiveness. With the ultra-high computational efficiency, a high-resolution parameter-defect index map was established to clarify the effect of the operating parameters at multiple dimensions, and to locate the low-risk range, the argon flow rate within 3–8 L/min, the casting speed within 1.0–1.7 m/min, and the nozzle immersion depth within 180–220 mm. This approach provides a promising technical route for the design of an efficient online slag entrainment monitoring system.
Meng et al. (Fri,) studied this question.