Acquiring insights into the effects of variations in operational parameters and physical characteristics on the internal dynamics of blast furnaces with rapidity and precision holds substantial importance for enhancing energy efficiency and diminishing operational expenditures. This research introduces a data-driven strategy for the prompt prediction of complex combustion characteristics within the furnace, crucial to the operational and material parameters of BFs. The strategy relies on a carefully constructed database employing the Computational Fluid Dynamics-Discrete Element Method (CFD-DEM). Formulated through a comprehensive full-factorial approach, this database incorporates 40 simulation scenarios encompassing key variables such as flow field, temperature distribution, gas concentration, and cohesive zone height. To investigate these phenomena, the study employs a machine learning (ML) model that effectively combines random forest regression and artificial neural networks, selected for their high predictive accuracy. The results demonstrate the model exhibits high accuracy in predicting key furnace phenomena, including temperature profiles, gas species concentrations, and the extent of the cohesive zone. The efficacy of ML predictions is further demonstrated by successful extrapolation and comparison of furnace phenomena in five new scenarios beyond the original database, achieving comprehensive virtualization. Impressively, the method's response time is approximately 864,000 times faster than that of conventional CFD-DEM simulations, while maintaining comparable accuracy. This predictive model represents a significant advancement in optimizing furnace responses to operational and material parameter variations efficiently, both in terms of time and cost, Providing a new perspective on blast furnace prediction and control.
 

Blast furnace process; CFD-DEM; Machine learning; Hybrid data-driven; Random Forest