Transient Reactive Grain Model with Uncertainty Quantifications and SensiTivity (TRUST) for Hydrogen Direct Reduction of Iron Ore

Global direct reduction of iron (DRI) for steelmaking has marked production increase of over 25% over the past five years. As production increases, the need for robust computational models predicting the DRI process becomes critical. However, existing modeling approaches are deterministic, which do not account for the probability distribution of parameters. So, when these models are calibrated to specific experimental data under fixed conditions, they need extrapolations for other operating conditions. Here we present a Transient Reactive Grain Model (TRGM) with Uncertainty Quantifications and Sensitivity (TRUST) by integrating probabilistic uncertainty quantification, surrogate-based Monte Carlo propagation, and global sensitivity analysis with the TRGM. Gaussian process regression (GPR) proved effective in capturing the nonlinear model dependencies and enabling uncertainty-aware predictions. Monte Carlo simulations show that activation energies of conversion chemical kinetics in iron ore reduction dominate conversion uncertainty through the entire conversion transient, while operational variables have secondary influence, and temperature effects are mostly local. Early-stage conversion exhibits high variability, whereas later stages converge predictably, suggesting streamlined monitoring strategies. Correlation analysis further indicates redundancy in late-stage measurements, and global sensitivity analysis identifies activation energies as critical targets for calibration. TRUST provides a validated, probabilistic framework for risk-informed process optimization and experimental design, and can be generalizable to other high-temperature metallurgical and chemical systems, supporting uncertainty-aware decision-making in hydrogenbased steelmaking.