Learning microstructure–property relationships in materials with robust features from vision transformers
Date:
Machine learning of microstructure–property relation- ships from data is an emerging approach in computational materials science. Most existing machine learning efforts focus on the development of task-specific models for each microstructure–property relationship. We propose utilizing a pre-trained foundational vision model for the extraction of task-agnostic microstructure features and subsequent light- weight machine learning. We demonstrate our approach with a pre-trained DinoV2 model on unsupervised repre- sentation of an ensemble of two-phase microstructures and modeling of their overall elastic stiffness. Our results show the potential of foundational vision models for robust mi- crostructure representation and efficient machine learning of microstructure–property relationships without the need for expensive task-specific training or fine-tuning.
Check out my slides from the talk at CVPR!