# An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials

**Authors:** Mohammad S. Khorrami, Jaber R. Mianroodi, Nima H. Siboni, Pawan Goyal, Bob Svendsen, Peter Benner, Dierk Raabe  
**arXiv:** [2208.13490](https://arxiv.org/abs/2208.13490)  
**Date:** 2022-08-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2208.13490v1/

## Key points

- Purpose: Develop an ANN for modeling the mechanical response of polycrystalline materials
- Utilize a trained CNN to account for history dependence of material behavior
- Training data consists of numerical simulation results for von Mises stress field under quasi-static tensile loading
- Trained CNN (tCNN) accurately reproduces average response and local von Mises stress field
- tCNN calculates stress field approximately 500 times faster than traditional methods relying on numerical solutions with spectral solvers
- tCNN can be applied to different microstructure morphologies and loading levels not present in training dataset, indicating versatility and generalizability
- Study presents a novel ANN-based approach for surrogate modeling of stress fields in viscoplastic polycrystalline materials
- tCNN offers significant computational advantages while maintaining high accuracy at macroscopic and microscopic scales.

## Summary

The purpose of this work is to develop an  for  of the mechanical response of  in polycrystalline materials. The researchers utilize a , trained to account for the history dependence of the material behavior. The training data consists of numerical simulation results for the von Mises stress field under <kd> quasi-static tensile loading. </ kd>The trained CNN, referred to as tCNN, demonstrates high accuracy in reproducing both the average response and the local von Mises stress field. Notably, the tCNN is capable of calculating the von Mises stress field for grain microstructures that were not included in the training dataset approximately 500 times faster than traditional methods relying on numerical solutions with spectral solvers for corresponding initial-boundary-value problems. Furthermore, the researchers successfully apply the tCNN to other types of microstructure morphologies, such as matrix-inclusion type topologies, and loading levels that were not present in the training dataset. This indicates the versatility and generalizability of their approach. Overall, this study presents a novel ANN-based approach for surrogate modeling of stress fields in viscoplastic polycrystalline materials. The developed tCNN offers significant computational advantages compared to traditional methods while maintaining high accuracy in predicting mechanical responses at both macroscopic and microscopic scales.