Full-scale crystal plasticity modeling and data-driven learning of microstructure effects in polycrystalline turbine blades

Key Points

• This research aims to model the effects of microstructure on the performance of polycrystalline turbine blades through advanced simulations and machine learning techniques.
• Full-scale crystal plasticity simulations performed on a turbine blade with 40,000 grains.
• Dislocation-density based crystal plasticity model incorporating geometrically necessary dislocations (GNDs) applied under finite thermomechanical loads.
• Graph neural networks trained on morphological descriptors to predict grain-averaged fields.
• The interface-aware meshing strategy effectively embeds realistic microstructures into complex geometries.
• Grain-wise parallel domain decomposition was successfully solved using Adaptive Multipreconditioned FETI (AMPFETI).
• Graph neural networks accurately predicted grain-averaged fields, enhancing microstructural analysis.