AI Searches 100 Million Options to Find Cheaper Way to 3D-Print NASA Rocket Alloy

Researchers at Washington State University have used artificial intelligence to find a cheaper, lower-power way to 3D-print a NASA-developed metal alloy, avoiding the need to manually test more than 100 million possible printing settings. The work, led by computer science professor Jana Doppa, was published in the Proceedings of the AAAI Conference on Artificial Intelligence and won the conference's Innovative Deployed Application Award.

The alloy, GRCop-42, is a mixture of copper, chromium and niobium that NASA developed for extreme-heat environments, including liquid rocket engine combustion chambers, where it must resist high temperatures while conducting heat efficiently. It is made with 3D printing, or additive manufacturing, which builds metal parts layer by layer instead of cutting or casting them. GRCop-42 normally needs high laser power to print, and about 90% of commercial printers cannot supply enough; earlier attempts at lower power had failed, sometimes melting the material.

Testing settings one by one was impractical, since each print can cost hundreds of dollars and take days to analyze, and there were more than 100 million possible printing configurations to check. Starting from data on 37 earlier failed attempts, the WSU team, working with Aryan Deshwal of the University of Minnesota, built an AI model that estimated how likely an untested setting was to succeed, then recommended small batches to try. Over three months and just 40 total experiments, the team found six successful configurations, including the first successful print of GRCop-42 using only 500 watts of laser power.

Lower-power printing could cut energy use, reduce wear on equipment, and lower the cost of processing finished samples, potentially making GRCop-42 usable on standard commercial printers at universities and smaller labs. Lead researcher Jana Doppa said the advance is "essentially democratizing the printing of this alloy." The researchers say the same AI-guided approach could be adapted to other metal alloys and to other fields with costly experiments, such as drug discovery.