# Prediction of the evolution of the stress field of polycrystals undergoing elastic-plastic deformation with a hybrid neural network model

**Authors:** Ari Frankel, Kousuke Tachida, Reese Jones  
**arXiv:** [1910.03172](https://arxiv.org/abs/1910.03172)  
**Date:** 2019-10-08  
**Summary page:** https://www.summarizepaper.com/en/paper/1910.03172v1/

## Key points

- Paper explores the use of crystal plasticity theory in predicting mesoscopic states of polycrystalline metals
- Proposes a novel approach using a neural network with convolutional layers to address computational expense
- Objective is to predict stress field evolution based on initial microstructure and external loading
- Study accurately predicts spatial average and captures entire stress field without artifacts
- Predicted stress fields and rates match well with ground truth data from experiments
- Hybrid neural network model effectively captures complex material behavior during deformation
- Trained model useful for material characterization and optimization tasks
- Promising approach for predicting stress fields in polycrystalline metals using hybrid neural network model

## Summary

The paper titled "Prediction of the evolution of the stress field of polycrystals undergoing elastic-plastic deformation with a hybrid neural network model" by Ari Frankel, Kousuke Tachida, and Reese Jones explores the use of crystal plasticity theory in predicting the mesoscopic states of polycrystalline metals. The authors propose a novel approach using a neural network with convolutional layers that encode correlations in both time and space to address the computational expense of simulating crystal plasticity theory. The main objective is to predict the stress field evolution based on initial microstructure and external loading. Compared to their previous work, this study not only accurately predicts the spatial average but also captures the entire stress field itself without any visible artifacts. The predicted stress fields and rates show high fidelity with crystal plasticity data and match well with ground truth data obtained from held-out experiments. This demonstrates the effectiveness of their hybrid neural network model in capturing complex material behavior during deformation. The trained model is also shown to be useful for material characterization and optimization tasks, highlighting its potential for practical applications in understanding material behavior under different loading conditions and optimizing material properties for specific uses. Overall, this study presents a promising approach for predicting stress fields in polycrystalline metals using a hybrid neural network model that achieves both accuracy and efficiency.