CMU Uses AI to Make 3D-Printable Tungsten Alloys That Don't Crack

The problem with 3D printing tungsten

Tungsten has the highest melting point of any pure metal, making it ideal for rocket engines, fusion reactors, and hypersonic vehicles. But it cracks like glass when it cools. That brittleness has kept tungsten out of most additive manufacturing workflows.

Researchers at Carnegie Mellon University set out to change that. Instead of testing thousands of alloy combinations by hand, they used machine learning to narrow the field to a handful of promising recipes.

How the AI helped

The team started with machine learning interatomic potentials, or MLIPs. These models predict atomic behavior 10,000 times faster than traditional physics simulations. The MLIPs scanned the tungsten-tantalum-niobium system and flagged the compositions most likely to be ductile.

From thousands of candidates, the MLIPs reduced the list to six alloys. The researchers then ran density functional theory simulations on those six, checking elastic properties at the atomic level.

The winning recipes

Two specific ratios stood out. One alloy uses 20% tungsten, 70% tantalum, and 10% niobium. The other swaps in 30% tungsten, 60% tantalum, and 10% niobium. When the team 3D printed both, they came out with no intergranular cracking.

Tantalum turned out to be the key ingredient. Adding it changed how electrons bond across the alloy, making the material roughly 50-60% more ductile. Niobium helped, but tantalum did the heavy lifting.

Why it matters

Crack-free tungsten opens the door to complex, high-temperature parts that additive manufacturing can finally produce reliably. The findings were published in Computational Materials Science and Physical Review Materials, and the team is already working on follow-up studies to test mechanical properties under real operating conditions.