Accelerated Search and Design of Stretchable Graphene Kirigami Using Machine Learning

Area of Science:

• Materials Science
• Computational Materials Design
• Machine Learning Applications

Background:

• Kirigami-inspired cuts enable the design of stretchable materials with metamorphic 2D-to-3D shape-changing properties.
• Optimizing kirigami cutting patterns is challenging due to the exponential growth in design complexity with system size.

Purpose of the Study:

• To apply machine learning (ML) for approximating material properties as a function of kirigami cutting patterns.
• To enable rapid discovery of kirigami designs exhibiting extreme stretchability.

Main Methods:

• Utilized convolutional neural networks (CNNs) for regression analysis of kirigami cutting patterns and material properties.
• Employed molecular dynamics (MD) simulations for verification of predicted kirigami designs and their stretchability.
• Trained ML models using a dataset of 1000 samples within a design space of approximately 4x10^6 candidate designs.

Main Results:

• Achieved high accuracy in predicting material properties (yield stress, yield strain) using CNNs, closely matching MD simulation precision.
• Successfully identified optimal kirigami designs that maximize elastic stretchability.
• Demonstrated ML's efficiency in exploring a vast design space, requiring significantly fewer iterations than purely simulation-based or experimental approaches.

Conclusions:

• Machine learning provides a powerful and efficient tool for optimizing kirigami-based stretchable material design.
• This ML-driven approach significantly reduces the computational cost and time for discovering novel metamorphic material designs.
• Highlights the potential of ML in materials science for accelerating innovation where underlying physics are not fully known or easily modeled.