Interpretable Molecular Dynamics–Machine Learning Framework for Hydrogen Uptake Mechanism in Magnesium

We present an interpretable molecular dynamics-machine learning (MD-ML) study that investigates the effects of surface accumulation, history-dependent kinetics, and dislocation structure in shaping the staged hydrogen uptake kinetics of magnesium. Using molecular dynamics simulation, Mg spheres with specified dislocation networks are subjected to external H injection under different programmed temperature schedules and initial external hydrogen pressure, yielding hydrogen uptake time series across various temperatures, external hydrogen pressure and dislocation structure conditions. The MD model predicts a 26% volumetric expansion of magnesium, which is consistent with experimental observation of approximate 30% volume expansion from Mg to MgH2. The simulation results reveal a three-stage hydrogen uptake mechanism of “surface accumulation—accelerated bulk diffusion—stabilized bulk diffusion” into Mg. The diffusion rate is found to depend on temperature, external hydrogen pressure, and dislocation structure. A transformer machine learning model is developed to predict the hydrogen uptake rate using 29 physical variables. The model achieves excellent prediction with R2 = 0.9925, RMSE = 1.985 × 10−2. Interpretability analyses identify that hydrogen uptake rate is history-dependent and external hydrogen pressure as the dominant predictive contributors, with screw-dislocation also exerting a significant influence on the diffusion rate. These results provide important and novel insights on the hydrogen uptake mechanism in Mg and furnish quantitative guidance for microstructure engineering for effective solid-state hydrogen storage.