# Using Thermodynamics and Microstructure to Mitigate Overfitting in Pellet Reduction Models

Source: [Using Thermodynamics and Microstructure to Mitigate Overfitting in Pellet Reduction Models](<https://www.sota2.com/research/paper/68fa4b3c1e209131ce944699>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T08:13:56.241Z

Authors: \\"Omer K. B\\"uy\\"ukuslu, Fabrice Yang, Dierk Raabe, Moritz to Baben, Anna L. Ravensburg

Published: 2025-03-20T16:05:02.000Z

## Abstract

Direct reduction of iron using hydrogen-rich gas is rapidly emerging as a key strategy for green steel production. This process involves complex, multiscale phenomena, encompassing solid-state phase transformations and gas transport through pores, that must be accurately represented for predictive industrial implementation. Here, we present a thermodynamically sound pellet-scale model that describes these mechanisms and can serve as a foundation for improving the understanding of pellet reduction kinetics in H$\_2$/CO-containing atmospheres. The model assumes that the gas phase remains in thermodynamic equilibrium, meaning that the composition of the gas instantaneously adjusts to any changes in the system. This reduces the number of fitting parameters drastically compared to other existing models, while maintaining a strict thermodynamic upper bound estimate. A driving force term is included in the reaction rate equation based on the partial pressure of O$\_2$ in the equilibrated gas phase. This constrained equilibrium-based approach ensures that the three iron oxide reduction steps and the formation of graphite and cementite in carbon-containing gases occur only if they are thermodynamically possible. It is demonstrated that fitting kinetic parameters based on conversion degree data alone leads to overfitting. This is true both for existing models and the model introduced here, despite the fact that the latter contains fewer parameters. To overcome this overfitting problem, spatially resolved microstructural data at key reduction stages can be considered, as shown here for recently reported data for a pellet reduced in H$\_2$ atmosphere.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2503.16274>)

[PDF](<https://arxiv.org/pdf/2503.16274.pdf>)
