ANN-Based Model to Predict the Viscosity of Molten Blast Furnace Slag at High Temperatures of > 1600 K

Accurate viscosity of molten blast furnace (BF) slag is indispensable for the development of centrifugal-granulation-assisted thermal energy recovery (CGATER) to decarbonize the iron and steel industry. Yet, direct experimental measurement of this quantity remains a formidable task due to the high temperature of 1600 Kelvin or above. Empirical models, if with high fidelity, are ideal to enable a fast and accurate prediction of the high-temperature viscosity of molten blast furnace slag. In this communication, we embark on a new effort to develop an artificial neural network (ANN)-based model to provide accurate viscosity prediction of molten blast furnace slag. This model was established based on a grid search method with approximately 4000 experimental measurements collected as the train and validation datasets. The viscosities of three types of molten blast furnace slag were measured above 1600 K as cross validation for the ANN model. Our ANN model agrees well with the experimental measurement with a small uncertainty of < 6%. Finally, an open-source artificial neural network code with a graphical user interface (GUI) was developed to provide a user-friendly portal for high-fidelity viscosity prediction. The present study not only enables a definitive, unified viscosity determination but also provides a flexible tool for the database establishment of the thermophysical properties of molten BF slag. Graphical Abstract We developed an ANN-based model to enable fast and accurate viscosity prediction of molten blast furnace slags at high temperatures and provided a user-interfaced open-source program, i.e. X-slag: Thermophysical Property.