Data-Driven Bi-Directional Lattice Property Customization and Optimization.

Area of Science:

• Materials Science
• Computational Engineering
• Machine Learning Applications

Background:

• Customizing and optimizing lattice materials presents significant design challenges.
• Existing methods often lack the flexibility and accuracy needed for specific performance requirements.

Purpose of the Study:

• To propose a data-driven generative method for customizing and optimizing lattice materials.
• To develop a machine learning framework for predicting elastic modulus and optimizing mechanical performance.

Main Methods:

• Utilized subdivision modeling for parametric description of lattice structures.
• Employed the homogenization method to analyze elastic moduli and generate a dataset.
• Developed a two-tiered machine learning (ML) framework for forward design (elastic modulus prediction).

Main Results:

• The two-tiered ML model achieved less than 10% error in elastic modulus prediction compared to finite element analysis.
• Optimization strategies resulted in up to a 25% improvement in mechanical performance under relative density constraints.
• Demonstrated practical value through case studies in customization and optimization.

Conclusions:

• The proposed data-driven generative method offers a flexible and accurate approach to tailoring lattice material properties.
• Combining generative design, ML, and genetic algorithms enhances material performance and navigates complex design spaces effectively.