Grain boundary segregation significantly impacts the physical properties of structural materials such as iron and ferritic steels, with critical implications for applications in energy, transportation, and infrastructure. Solute segregation produces regions with distinct compositions and properties, which can have important and often detrimental effects on the overall performance of the material. However, understanding segregation remains challenging, as it involves a complex data space with a wide range of possible atomic configurations. Managing this complexity is difficult because the data is both high-dimensional and computationally expensive to generate. Therefore, ensuring a sufficient quantity and quality of training data is crucial, alongside strategies for efficient reuse. To address this, we apply semantic annotation techniques to build knowledge graphs that capture structure–property relationships. These graphs facilitate better organization, sharing, and reuse of simulation results, and support multi-modal learning approaches.

A key example is hydrogen embrittlement, where segregation is highly dependent on the local atomic environment at the grain boundary. To scan a broader range of grain boundaries, we train machine learning models to predict hydrogen segregation energies based on local atomic environment descriptors. We calculate solution energies for varying hydrogen concentrations across 20 Σ (≤34) grain boundaries in α-iron, using a combination of density functional theory and molecular statics simulations, with the latter employing a neural network interatomic potential developed for Fe–H systems. We evaluate multiple feature sets, including Voronoi volume, Steinhardt bond-order parameters, and Smooth Overlap of Atomic Positions (SOAP) descriptors. This enables us to develop a surrogate model for predicting segregation behavior across different grain boundaries. Altogether, this work moves us toward an integrated approach where computational insights and data-driven tools can bring a fundamental understanding of grain boundary segregation mechanisms.