Machine learning-enabled constrained multi-objective design of architected materials.

Area of Science:

• Materials Science
• Biomedical Engineering
• Computational Mechanics

Background:

• Architected materials offer tunable properties but their design is complex and relies on expert knowledge.
• Current design processes for architected materials are often labor-intensive and lack efficiency.
• Optimizing materials for specific applications like orthopedic implants requires balancing multiple properties.

Purpose of the Study:

• To develop a data-efficient machine learning (ML) method for optimizing 3D-printed architected materials.
• To apply this ML method to the design of orthopedic implants with enhanced mechanical properties.
• To demonstrate the capability of ML-guided design for adapting to complex biological structures.

Main Methods:

• Utilized a machine learning (ML) cycle integrating finite element method (FEM) simulations and 3D neural networks.
• Developed an experience-free, high-dimensional multi-property optimization framework for architected materials.
• Implemented a machine-human synergy approach to adapt ML-designed architectures for specific defect geometries.

Main Results:

• Designed microscale heterogeneous architectures with a biocompatible elastic modulus and increased strength compared to uniform designs.
• Achieved a 20% higher experimental load-bearing capacity in adapted implant designs for animal bone defects.
• Demonstrated the ML method's ability to generate novel material architectures without prior expert knowledge.

Conclusions:

• The presented data-efficient ML method enables fast and intelligent design of architected materials.
• This approach facilitates the creation of materials with tailored mechanical, physical, and chemical properties.
• The findings offer a new paradigm for designing advanced materials, particularly in biomedical applications like orthopedic implants.