Enhanced Temperature Prediction in Ladle Furnace Steel Refining: Hybrid Process Modeling Based on Computational Thermodynamics and Statistical Learning Methods

The ladle furnace (LF) is the central refining aggregate in modern secondary steelmaking, enabling the admixture of alloying elements, chemical homogenization of the melt, and removal of non-metallic inclusions (NMIs). In addition to metallurgical tasks, precise temperature control and superheating are central to the subsequent casting processes, with optimized temperature management significantly reducing energy consumption and the associated CO2 footprint. The work’s first part deals with a comprehensive literature study of existing mechanistic (“white-box”), data-driven (“black-box”), and hybrid (“gray-box”) temperature models for secondary steelmaking, providing a detailed overview of the model’s scope, methodology, used boundary conditions, and approach to validate the predictions. Based on the literature study, an evaluation highlights the strengths and weaknesses of the different models, focusing on their steel temperature predictability and the process variables they incorporate. The work’s second part outlines the methodology and scope of the developed hybrid LF process model. It is based on the effective equilibrium reaction zone (EERZ) method, coupling fundamental computational thermodynamics with the ongoing kinetics within the steel ladle. Consequently, the heat effects of chemical reactions and mixing enthalpies of alloying materials can be quantified. Statistical learning techniques are applied to LF process data of an annual production at voestalpine Stahl GmbH. This data were extensively cleaned and analyzed using descriptive statistics. Finally, the influences of relevant process parameters on the temperature, e.g., heating power via electrodes, cooling in the ladle due to radiation and convection, and purging flow rate via bottom plugs, were quantified by applying multiple linear regression. In the work's final part, the findings from the data-driven analysis are presented, and the hybrid temperature model is thoroughly tested using several case studies. With a deviation of less than 5 °C for 94 pct of the predicted temperatures compared to the actual measured ones, the model can be considered precise and reliable. The developed ladle furnace model can be employed to simulate various process sequences, identify potential optimization measures, and make appropriate adaptations during the steel refinement.