Modeling of direct reduction, current experiences in literature, critical issues, suggestions
The modeling of direct reduction (DR) of iron oxides is a key process to produce Direct Reduced Iron (DRI). This approach, an alternative to traditional blast furnaces, allows metallic iron to be obtained without melting, using reducing gases. As a matter of fact, direct reduction is particularly relevant in the context of energy transition and decarbonization, thanks to the use of hydrogen as a reducing agent which causes reduced (also significantly, depending on other actions) CO2 footprint. As the steelmaking route Direct Reduction + Electric Arc Furnace is considered one of the most promising for steel carbon neutral production, it was chosen for a reduced scale investigation in the Hydra project.
In this view, DR modelling has been considered as support to drive properly experimental activity. Theoretical models are shown together with practical challenges, and emerging technologies, with particular attention to the integration of artificial intelligence (AI) in modeling and optimizing processes.
The modeling review is divided into two main parts:
1. single pellet reduction (micro and mesoscale), analyzing various mathematical approaches:
-
The unreacted shrinking core model, which describes superficial reactions as they progress inward through the pellet.
-
The grain model, which describes the pellet as a collection of microscopic grains and each grain undergoes topochemical reactions.
-
The random pore model, which incorporates the evolution of porosity, essential for predicting reaction kinetics.
-
The phase-field model, which allows detailed simulations of microstructural transformations.
2. industrial reactors. The focus is on shaft furnaces reactors, which represent the most diffused technology for DRI production. In this frame, several physical phenomena involved are discussed, such as:
-
Multiphase flow of particle and gas
-
Heat and mass transfer
-
Chemical reaction
-
Particle-particle interaction.
Such phenomena can be simulated by several modeling approaches:
-
CFD-DEM models (3D).
-
Eulerian two-phase approach, porous medium approach (1D-2D-3D).
-
REDUCTOR (2D).
-
Plug flow models (1D).
Although no single model description can catch all multiscale interactions, combinations of techniques enable accurate and practical simulations for industrial applications.
An innovative element is the integration of AI to enhance the efficiency of direct reduction processes. The use of techniques such as neural networks (MLPNN, RBFNN), and machine learning models (Random Forest) allows for:
-
predicting reactor performance by integrating experimental data and theoretical models for faster and more efficient simulations.
-
optimizing operational parameters, such as temperature, pressure, and the composition of the reducing gas, to maximize process efficiency and minimize emissions.
The application of AI represents a significant step forward in improving the sustainability and efficiency of DRI production processes, reducing costs and environmental impacts. More in general, the combination of advanced microstructural models and AI-driven simulations can improve operational efficiency and meet the growing demands for sustainability in the sector. The adoption of hybrid technologies integrating physical modeling, numerical simulations, and AI will be crucial to optimize the production of DRI, while ensuring competitiveness and compliance with environmental regulations.