Deep-Learning Prediction of Dislocation Presence and Position at Grain Boundaries in FCC materials
Abstract
Predicting dislocation distributions at grain boundaries (GBs) is central to polycrystal plasticity, yet atomistic or crystal-plasticity simulation remain too costly for exploring the misorientation space. We train a convolutional neural network (CNN), with parallel classification and regression heads, on synthetic elementary Nye dislocation density tensor maps generated by translating a single relaxed copper bicrystal reference. Without retraining, the network is evaluated via slidingwindow inference and clustering on simulated copper and aluminium bicrystals with symmetric tilt boundaries, spanning twelve and eleven misorientation angles. Predicted dislocation positions reproduce the atomistic reference, and the predicted spacing follows the Frank-Bilby trend. Even at high angles where the Frank-Bilby model no longer holds, the network still detects a strong signal. CNN surrogates can thus act as fast, physically consistent proxies for atomistic dislocation detection at grain boundaries.
Related
- Misfit-dislocation hierarchy governs sliding of asymmetric non-CSL grain boundaries
- Examples of Gribov copies in the presence of planar boundaries, and an emergent boundary gauge invariance in Yang-Mills theory
- Indirect prediction and screening of frustrated magnetic materials using machine learning
- Learning the Chronology of Materials Discovery for Synthesizability Prediction
- Thermodynamically Consistent Machine Learning for Fracture Prediction in Heterogeneous Materials
BibTeX
@misc{woguem2026deeplearning,
title = {{Deep-Learning Prediction of Dislocation Presence and Position at Grain Boundaries in FCC materials}},
author = {Yen Fred Woguem and Vincent Taupin and Julien Guénolé},
howpublished = {HAL},
year = {2026},
month = oct,
url = {https://hal.univ-lorraine.fr/hal-05795530v1},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Deep-Learning Prediction of Dislocation Presence and Position at Grain Boundaries in FCC materials
AU - Yen Fred Woguem
AU - Vincent Taupin
AU - Julien Guénolé
PY - 2026
DA - 2026/10/07
PB - HAL
UR - https://hal.univ-lorraine.fr/hal-05795530v1
AB - Predicting dislocation distributions at grain boundaries (GBs) is central to polycrystal plasticity, yet atomistic or crystal-plasticity simulation remain too costly for exploring the misorientation space. We train a convolutional neural network (CNN), with parallel classification and regression heads, on synthetic elementary Nye dislocation density tensor maps generated by translating a single relaxed copper bicrystal reference. Without retraining, the network is evaluated via slidingwindow inference and clustering on simulated copper and aluminium bicrystals with symmetric tilt boundaries, spanning twelve and eleven misorientation angles. Predicted dislocation positions reproduce the atomistic reference, and the predicted spacing follows the Frank-Bilby trend. Even at high angles where the Frank-Bilby model no longer holds, the network still detects a strong signal. CNN surrogates can thus act as fast, physically consistent proxies for atomistic dislocation detection at grain boundaries.
ER -