Ladle furnace slag characterization through hyperspectral reflectance regression model for secondary metallurgy process optimization

In steelmaking process, close control of slag evolution is as important as control of steel composition. However, there are no industrially consolidated techniques that allow in-situ analysis of the slag chemical composition, as in the case of steel with OES-spectrometers. In this work, a method to analyze spectral reflectance of ladle furnace slag samples to estimate their composition is proposed. This method does not require sample preprocessing and is based on a regression algorithm that mathematically maps the spectral reflectance of the slag with its actual composition with errors lower than 10%. Specifically designed normalization and calibration steps have been proposed to allow a global model training with data from different locations. This allows real-time monitoring of the thermodynamical state of the steel process by feeding a thermodynamic equilibrium optimization model. The system has been validated on several ArcelorMittal locations achieving process savings of 0.71 Euro per liquid steel tons.