Intelligent design of mechanical metamaterials: a GCNN-based structural genome database approach.

Area of Science:

• Materials Science
• Computational Mechanics
• Artificial Intelligence

Background:

• Intelligent material design requires understanding the geometry-property relationship in unit cells.
• Classical methods for this mapping are computationally intensive and limited.
• Advanced materials with specific mechanical properties are in high demand.

Purpose of the Study:

• To develop an efficient and accurate method for mapping unit cell geometry to material properties.
• To introduce a structural genome database (SGD) approach for inverse material design.
• To demonstrate the capability of the SGD for designing novel mechanical metamaterials.

Main Methods:

• Utilized a graph convolutional neural network (GCNN) for geometry-property mapping.
• Developed a structural genome database (SGD) integrating GCNN predictions.
• Employed transfer learning for predicting non-local behaviors and optimizing designs.

Main Results:

• Achieved 3-4 orders of magnitude improvement in design efficiency.
• Successfully designed metamaterials with target properties: Hashin-Shtrikman upper bound, auxetic behavior, and unimodal properties.
• Demonstrated enhanced critical buckling strength (nearly 200%) and significant relative bandgap width (51%).

Conclusions:

• The proposed SGD approach enables rapid and accurate inverse design of mechanical metamaterials.
• Experimental validation confirms the auxetic behavior and superior buckling resistance of designed metamaterials.
• The SGD approach shows significant potential for multi-scale and multi-physics system design.