Ongoing research

Crack-Path Prediction in
Polycrystalline Hexagonal Boron Nitride

Developing a molecular-dynamics and machine-learning framework to study how grain morphology and crystallographic orientation influence fracture trajectories in monolayer hexagonal boron nitride.

Atomistic modeling

Pristine monolayer h-BN models were investigated under armchair and zigzag tension using LAMMPS. Their elastic moduli and tensile strengths were compared with published results to assess the preliminary atomistic response.

Polycrystalline geometries were constructed through Voronoi tessellation with varying grain sizes and crystallographic orientations. Molecular dynamics simulations of edge-cracked specimens are underway to generate fracture trajectories for machine-learning applications.

The resulting dataset is intended to support crack-path prediction in previously untested polycrystalline h-BN microstructures using a CNN–BiRNN framework.

Original atomistic rendering of a polycrystalline hBN sheet: randomly oriented Voronoi grains, a centered slit from the left edge, and opposing tensile-loading arrows. — Initial edge-cracked polycrystalline h-BN geometry prior to tensile loading.

CNN–BiRNN for spatial crack-path prediction

The proposed CNN–BiRNN framework aims to predict crack trajectories from the initial polycrystalline microstructure. Candidate input descriptors include lattice orientation, grain boundaries, initial crack geometry, and local boron–nitrogen environments.

Convolutional layers extract features from successive spatial sections, while a bidirectional recurrent neural network combines information from both directions to generate a predicted crack-path mask.

Reference crack trajectories derived from molecular dynamics simulations will provide the supervised learning targets. Crack extraction will involve identifying changes in atomic neighbor connectivity and tracing damage associated with the initial notch. Training, validation, and testing will use separate MD specimens to avoid data leakage between partitions.

Architecture schematic: initial microstructure, shared slice-wise CNN, separate forward and backward recurrent branches, concatenated states, dense crack-mask readout, and supervision by MD-derived crack targets. — Proposed CNN–BiRNN framework for spatial crack-path prediction from initial polycrystalline microstructures.

Methodological reference: M. S. R. Elapolu, Md. I. R. Shishir and A. Tabarraei. A novel approach for studying crack propagation in polycrystalline graphene using machine learning algorithms. Computational Materials Science 201 (2022), 110878. doi:10.1016/j.commatsci.2021.110878