Deep Learning Co-Optimizes LPBF Titanium Gyroid Lattices — University of Birmingham Breakthrough

AI-Powered Lattice Design Breakthrough

Researchers at the University of Birmingham have published a groundbreaking study demonstrating how deep learning can co-optimize both Laser Powder Bed Fusion (LPBF) process parameters and titanium gyroid lattice geometry simultaneously.

The study, published in the Journal of Manufacturing and Materials Processing, addresses one of additive manufacturing's persistent challenges: mapping process settings and design parameters to final part performance.

How It Works

The research team developed data-driven models that link energy density and gyroid lattice geometry to key mechanical properties including Young's modulus and yield strength. Using stacked-autoencoder pre-training to stabilize small-data learning, the models enable an inverse-design workflow where engineers can specify desired mechanical properties and receive feasible combinations of LPBF settings and lattice parameters.

Why Gyroid Lattices Matter

Gyroid triply periodic minimal surface (TPMS) structures have attracted significant attention for their high specific strength, excellent energy absorption, and geometric adaptability. These properties make them ideal for aerospace, automotive, and biomedical applications.

However, optimizing both the manufacturing process and the lattice design simultaneously has traditionally required extensive trial-and-error experimentation.

Property-Driven Design

The new approach allows designers to work backwards from desired properties. Instead of designing a lattice and then hoping the printing process produces acceptable results, engineers can specify target stiffness and strength, and the AI system returns the optimal combination of laser power, scan speed, hatch spacing, and lattice density.

This represents a significant step toward truly digital, AI-assisted design and manufacturing of architected materials.