Generative machine learning algorithm for lattice structures with superior mechanical properties.

Area of Science:

• Mechanical Engineering
• Materials Science
• Computational Design

Background:

• Lattice structures offer effective material distribution but face a density-mechanical property trade-off.
• Uniform beam cross-sections in current designs limit optimization and weight-to-performance ratios.

Purpose of the Study:

• To investigate optimized beam element shapes in lattice structures using advanced computational methods.
• To explore an augmented design space for superior lattice structures with improved weight-to-performance ratios.

Main Methods:

• Utilized a deep learning approach with high-order Bézier curves to define beam element shapes.
• Employed a hybrid neural network and genetic optimization (NN-GO) adaptive method for lattice generation.
• Fabricated optimized designs via additive manufacturing and validated through compression testing.

Main Results:

• Optimized designs shifted material towards joints, enhancing modulus and strength.
• Achieved a balance between axial and bending deformation modes for efficient load bearing and energy absorption.
• Demonstrated superior performance over benchmark designs, with good correlation between simulations and experiments.

Conclusions:

• High-order Bézier curves enable smoother transitions, reducing stress concentration in lattice structures.
• The deep learning and NN-GO approach successfully generates high-performance lattice designs.
• Optimized lattice structures show significant improvements in mechanical properties and efficiency.