Optimization of high alumina slag practice in blast furnace ironmaking: an industrial approach. Part 2: Data-driven aspects

Abstract

The prediction of transport and thermodynamic properties of the blast furnace slag is an experimentally tedious job to accomplish as seen in part-1 of this work. Literature studies have shown that the use of machine learning in the determination of the properties as a function of composition is an effective technique for glassy slags. However, the application of machine learning and data science techniques in the prediction of high alumina slag properties has not been studied extensively so far. In this paper, the use of Support Vector Machine (SVM) and ExtraTrees Regressor have been done to predict the viscosity, liquidus temperature, and other thermodynamic properties of blast furnace type slag in the high alumina regime, i.e. Al₂O₃ varying from 18 to 22 wt-%. The minimization of detrimental effects of high alumina slag has been studied by varying the MgO content and CaO/SiO₂ ratio in the range of 8–12 wt-% and 0.8–1.2, respectively. The accuracy of models has been tuned to be fairly high and the results of the prediction have been discussed with possible solutions to operate under the high alumina regime of blast furnace slag.