Prediction of steel nanohardness by using graph neural networks on surface polycrystallinity maps

Nanoscale hardness in polycrystalline metals is strongly dependent on microstructural features that are believed to be influenced from polycrystallinity - namely, grain orientations and neighboring grain properties. We train a graph neural networks (GNN) model, with grain centers as graph nodes, to assess the predictability of micromechanical responses of nano-indented 310S steel surfaces, based on surface polycrystallinity, captured by electron backscatter diffraction maps. The grain size distribution ranges between 1-100 mu m, with mean size at 18 mu m. The GNN model is trained on nanomechanical load-displacement curves to make predictions of nano -hardness, with sole input being the grain locations and orientations. We explore model performance and its dependence on various structural/topological grain-level descriptors (e.g. grain size and number of neighbors). Analogous GNN-based frameworks may be utilized for quick, inexpensive hardness estimates, for guidance to detailed nanoindentation experiments, akin to cartography tool developments in the world exploration era.

Automated summary

By training a graph neural networks (GNN) model with grain centers as graph nodes, we can predict the nanomechanical responses of nano-indented metal surfaces based on surface polycrystallinity. The GNN model uses grain locations and orientations as the sole input to make predictions of nano-hardness. The model's performance and dependence on various grain-level descriptors such as grain size and number of neighbors are explored.