Applied Machine Learning to Predict Stress Hotspots in Materials

This work focuses on integrating crystal plasticity based deformation models and machine learning techniques to gain data driven insights about the microstructural properties
of polycrystalline metals. An inhomogeneous stress distribution in materials leads to the development of stress hotspots in polycrystalline metals under uniaxial tensile deformation. We simulate uniaxial tensile deformation in synthetic microstructures to get full field solutions for local micromechanical elds (stress and strain rates). After identifying stress hotspots by thresholding stress values, we characterize their neighborhoods using metrics that re
ect the local crystallography, geometry, and connectivity. This data is used to create input feature vectors to train a random forest learning algorithm, which predicts the grains that become stress hotspots. We are able to achieve an area under the receiving operating characteristic curve (ROC-AUC) of 0.82 for hexagonal close packed and 0.74 for face centered cubic materials. Inspired by the recent advances in the deep learning field, we also explore using these techniques to automatically extract long range microstructural descriptors. The results show the power and the limitations of the machine learning approach applied to the polycrystalline grain network.