Graph Neural Networks for grain tracking in crystal plasticity simulations
This repository focuses on grain-scale materials modelling using graph representations, where grains are connected by neighbourhood links to predict responses across strain increments in both representative volume elements and experimental datasets. It describes workflows that construct graph data from grain connectivity, normalise physical attributes (shear, orientation, size, phase), and train and evaluate models for strain-path or single-step prediction tasks. It also includes supporting datasets, precomputed edge features, and results for multiple loading scenarios to study how microstructural interactions influence predicted grain responses under deformation.
