Digital transformation of rotary hearth furnace for process automation and CO2 reduction: Hybrid model under limited operational data utilization

Abstract

A hybrid model of the rotary hearth furnace was developed by integrating a limited number of operational data and first-principle knowledge of the process. The digital transformation was implemented in two sequential phases. Phase I involved obtaining sufficient data using a mathematical model governed by mass and energy balance. The first-principles model bridged the gap in the search space of the original operational data, thereby increasing the accuracy of the machine-learning model. In Phase II, extreme-gradient boosting was used to predict the degrees of metallization and dezincification. The training scores (R2 > 0.90 for both) and test scores (R2 > 0.96 for both) met the required criteria. An explainable feature of machine learning revealed that carbon content was the most important factor. Subsequently, the carbon mass of the feed composition was controlled based on different scenarios, demonstrating the effectiveness of the hybrid model. The optimization results indicated the presence of the ‘carbon window’, enabling CO2 mitigation under specific conditions. Among the scenarios, one achieved a reduction of 0.9 MtCO2 y−1, comparable to other CO2 mitigation strategies employed in blast furnace-based ironmaking processes. The hybrid approach can also be applied to offer operational guidance for manufacturing systems with known phenomenology but limited datasets.