Deep learning based design of porous graphene for enhanced mechanical resilience

Fracture behaviors of brittle materials are a crucial problem when it comes to reliability, especially for nanoscale devices and systems such as those built based on graphene. This study aims to use a new deep learning model, successfully incorporating data from molecular-level modeling, to predict the fracture path of graphene under the presence of various defects. In the process to build the model we first perform tensile test simulations on various graphene systems using molecular dynamics. The results are then transferred into image-based data for processing in the deep learning model. Based on this dataset we then construct multiple ConvLSTM-based models to learn the spatial-temporal information about crack propagation for each graphene system, respectively. The results show that our ConvLSTM-based models can predict the fracture path of graphene with 99 percent accuracy on a system of different crystallinity and 98 percent accuracy on different sets of defects, demonstrating excellent generalizability and transferability. These models demonstrate the power of exploiting deep learning for nanoengineering, and to specifically confer desired properties of materials based on defect engineering, which has great potential for next-generation materials by design.