Washington State University researchers have used artificial intelligence to slash the search for viable 3D-printing settings for a hard-to-print NASA rocket alloy from more than 100 million possible configurations down to just 40 test prints, according to a paper the team presented at the AAAI Conference on Artificial Intelligence, where it won an Innovative Deployed Application Award.
The alloy, GRCop-42, is a copper-chromium-niobium blend developed by NASA for parts that must survive extreme heat, such as combustion chambers in liquid-propellant rocket engines. It conducts heat well and holds its strength under intense thermal stress, but printing it reliably has required specialized, high-power equipment: according to the researchers, roughly 90% of commercial 3D printers cannot produce usable parts from it.
Turning 37 failures into a working recipe
Led by computer science professor Jana Doppa and PhD student Azza Fadhel, the team built an AI system that started from 37 previously failed print attempts and used them to estimate which of the alloy’s many untested printer settings — laser power, speed, and other parameters — were likely to succeed. Rather than testing at random, the model picked small batches of experiments designed to both chase promising settings and probe the configurations it was least certain about, refining its estimates after each print.
Within a three-month, 40-experiment budget, the approach surfaced six workable configurations, including one that printed successfully at a record-low 500 watts — well within range of standard commercial machines. Each print run costs hundreds of dollars and takes days to evaluate for quality, the researchers said, which is what made an efficient search strategy essential.
“Ninety percent of commercial printers cannot print this metal alloy, so given that we were able to find these feasible process parameters, it allows us to use those commercial printers,” Doppa said. “We are essentially democratizing the printing of this alloy.”
Beyond rocket parts
The team, which also included researchers from WSU’s School of Mechanical and Materials Engineering and the University of Minnesota, says the same approach could be adapted to other alloys and additive-manufacturing systems, and more broadly to scientific problems where success is rare and every experiment is costly — a growing focus in AI for science aimed at speeding up lab work rather than replacing it. For more on how AI is being put to work in labs, see our guide on what AI can do for researchers — though a separate recent study found current models still struggle to recover the reasoning behind published research ideas. The findings could lower costs for universities, small labs, and aerospace suppliers working with NASA-developed materials once locked to specialized facilities.