Leveraging Generative AI for Stress Prediction and Design of Architected Graphene Structures

Generative artificial intelligence (AI) is a powerful tool for learning from existing data and generating new information, but its application in quantitative mechanical research remains underexplored. This study focuses on the structure-mechanics relationship in architected graphene, specifically graphene with patterned void defects. We use Molecular Dynamics (MD) to simulate uniaxial tension on these structures, extracting the von Mises stress field during mechanical loading. A fine-tuned generative AI model is then trained using Low-Rank Adaptation, enabling the design of architected graphene structures and the prediction of their stress fields under tension based on simple descriptive inputs. Our model, trained with just a few images, accurately predicts stress fields for various defect geometries and distributions not included in the training data, validated by MD simulations. Additionally, the AI can generate stress fields from schematic drawings of the graphene, highlighting its potential for end-to-end nanomaterial design and characterization without complex numerical modeling.