Washington State University researchers report an artificial intelligence breakthrough for GRCop-42 3D printing, finding a faster and less expensive path to produce the high-performance alloy without manually testing more than 100 million potential settings.
The team says the advance could allow the NASA-developed alloy, already used in aerospace components and potentially useful across other sectors, to be built on widely available commercial printers. They add that the AI strategy may be applied to other scientific domains that face massive experimental search spaces, including drug discovery.
According to the university, investigators from the School of Electrical Engineering and Computer Science and the School of Mechanical and Materials Engineering detailed the work in the Proceedings of the AAAI Conference on Artificial Intelligence, where it also received the Innovative Deployed Application Award.
“Ninety percent of commercial printers cannot print this metal alloy,” said Jana Doppa, Huie-Rogers Endowed Chair Professor of Computer Science and Berry Distinguished Professor in Engineering, who led the study. Doppa said the new process parameters could enable those printers to produce the alloy, effectively “democratizing” access.
GRCop-42 3D printing for extreme heat applications
GRCop-42 is a copper, chromium, and niobium alloy created by NASA for high-heat environments requiring both strength and efficient heat transfer. Its combination of thermal conductivity and high-temperature strength makes it suitable for aerospace systems such as liquid rocket engine combustion chambers.
Despite those attributes, the alloy is difficult and costly to print, typically demanding substantial laser power and energy.
Past efforts to manufacture GRCop-42 on lower-wattage, off-the-shelf machines have fallen short. Systematically testing the enormous range of possible settings is impractical because each print consumes expensive material, specialized equipment, and extensive labor.
A single build can cost hundreds of dollars and take days to analyze.
“Sometimes they printed a certain configuration, and the product just melted,” said first author and computer science PhD student Azza Fadhel. “They wouldn’t be able to try all 100 million options. We applied AI to efficiently choose candidates from this very large search space.”
AI narrows more than 100 million configurations
The project began with data from 37 failed printing configurations produced in prior mechanical and materials engineering experiments. Using those outcomes, the researchers built a model to estimate the likelihood that untested settings would succeed, then had the AI suggest small batches of new configurations to evaluate.
The selection balanced two goals. Some trials targeted the most promising conditions, while others probed uncertain regions that could yield new information and improve the model’s accuracy.
Mechanical and materials engineering researchers Nathaniel Zuckschwerdt, Susmita Bose, and Amit Bandyopadhyay used the AI-recommended parameters to print and assess GRCop-42 samples. Aryan Deshwal of the University of Minnesota also contributed to the work.
“Every result improved our AI model,” Fadhel said, noting that even failed prints helped refine predictions.
Lower power broadens access
Identifying settings that work with lower laser power could reduce energy use, lessen equipment wear, and cut post-processing costs. It could also open access to GRCop-42 for universities, small labs, and companies without specialized high-power systems.
The challenge was the scarcity of successful outcomes within more than 100 million possibilities. “It’s a very challenging case for AI,” Doppa said. “You get a binary success or failure signal and want to minimize the number of tries to find the needles quickly.”
Within three months, and limiting the effort to 40 total experiments, the team reported six successful parameter sets at multiple power levels. They also achieved a first, successfully printing GRCop-42 using 500 watts of laser power.
GRCop-42 3D printing as a general tool for discovery
The researchers say their AI-guided method can be adapted to other metal alloys and additive manufacturing platforms. More broadly, the approach could accelerate progress in fields where viable outcomes are rare, the search space is vast, and each experiment is costly in time, materials, or funding.
“There’s always uncertainty when deploying something with real materials and costs,” Doppa said. “We didn’t know whether we would succeed, and there are real stakes. I was very surprised we were able to do this so well.”
The team’s focus on AI-driven experimentation echoes broader efforts to apply artificial intelligence to scientific and engineering challenges, such as the use of agentic AI in semiconductor root cause analysis.
Washington State University, which led the research, is among the institutions expanding their work in advanced materials and AI, alongside peers such as the University of Minnesota.