Deep Learning for Peak-Stress Prediction in Polycrystalline Materials
Introduction
This project presents a deep learning approach to predict peak-stress clusters in heterogeneous polycrystalline materials. Unlike prior work that focused on overall stress fields, this method targets localized peak-stress regions critical to failure. Using a convolutional encoder–decoder network trained on synthetic microstructures and linear elasticity simulations, the model predicts stress fields and identifies peak-stress clusters. Evaluation using cosine similarity and geometric comparisons shows high accuracy, especially for higher normalized stress values.
Refer paper a (Shrivastava et al., 2022) for details .
For code refer GitHub Repository.
Caption: Overview of the approach.
