Crystal plasticity-informed machine learning prediction of stress hotspots in GH4169 superalloy

Local stress concentration is an early microstructural indicator of damage initiation in nickel-based superalloys, but its origin cannot be attributed to a single descriptor such as grain size, Schmid factor or grain-boundary distance. In polycrystalline GH4169, stress hotspots emerge from the coupled effects of crystallographic orientation, texture, intergranular compatibility and neighbourhood constraint. Here, crystal plasticity finite element (CPFE) simulations are integrated with grain-aware machine learning to predict stress hotspots across representative FCC texture states and EBSD-informed three-dimensional microstructures. The datasets include Brass, Copper, Cube, Goss, S and Uniform textures, a microstructure generated from experimental GH4169 Euler angles, and a representative three-dimensional EBSD reconstruction. Voxels within the top 10% of the simulated von Mises stress distribution are labelled as hotspots, and physically interpretable descriptors are extracted for crystallographic orientation, Schmid factor, grain geometry, local misorientation, grain-boundary proximity and slip-transfer compatibility. To avoid inflated performance from spatially correlated voxels within grains, models are evaluated using a Grain-ID-based grouped split. Random Forest achieves the best balance between predictive performance, robustness and interpretability, with ROC-AUC values of approximately 0.763 to 0.950 across texture states and about 0.87 for the reconstructed microstructures. Robustness tests preserve the main performance trends and feature rankings. SHAP analysis identifies orientation-distance descriptors, particularly r 001 , r 101 and r 111 , as dominant predictors, while Schmid factor, local misorientation and slip-transfer descriptors further modulate hotspot intensity and location. These results show that stress hotspots in GH4169 are texture-sensitive and physically interpretable local responses, providing a route for rapid hotspot screening and texture-guided microstructure design.