Modeling the Thermal Properties of Hydrogen‐Based Direct Reduced Iron Using Computational Thermodynamics and Machine Learning

Hydrogen‐based direct reduced iron (H‐DRI) is a promising low‐carbon feedstock for steelmaking using electric‐arc‐furnace (EAF) or electric smelter (ESF). So far, the information on the thermal properties of H‐DRI, which is crucial for predicting the energy need and melting time, is limited. In this study, we investigated H‐DRI pellets with varying metallization levels (68%, 86%, 90%, and 95%) and measured thermal response using differential scanning calorimetry (DSC) up to 1600°C to determine the solidus and liquidus temperatures as well as the heat capacity and enthalpy of melting. The composition of the pellets with respect to bulk oxides (Fe, SiO 2 , Al 2 O 3 , CaO, FeO, MnO) was measured by X‐ray fluorescence (XRF) and used in FactSage 8.3 (FactPS, FToxid, FTmisc) to compute reference values for specific heat capacity, enthalpy, and phase transition temperatures. The melting points calculated using FactSage matched DSC results within about 15°C; calculated enthalpies were within ≈5%, and specific heat capacities within about 3%. To account for variability, we generated 1000 synthetic chemistries covering industrial ranges (55–97 wt% Fe, 2–10 wt% SiO 2 , 1–8 wt% CaO, 0–25 wt% FeO, 0–5 wt% Al 2 O 3 , 0–5 wt% MnO), split into 80/20 for training and validation. Simple linear regressions predicting solidus and liquidus temperatures from composition explained over 96% of the variance on held‐out data, while specific heat capacity and enthalpy were calculated via Shomate mass‐fraction mixing and validated against DSC and FactSage. Independent tests on four H‐DRI samples fell within the experimental uncertainty. These relations enable routine XRF analyses to determine melting temperatures and thermophysical properties quickly and reliably, simplifying heat‐balance calculations in the modeling and operation of the EAF.