TL;DR: Washington State University researchers used artificial intelligence to identify six workable 3D-printing settings for the NASA-developed GRCop-42 alloy from more than 100 million possibilities, including the first successful print at 500 watts, within 40 trials over three months.

Washington State University researchers have shown that a copper-chromium-niobium alloy used in liquid rocket engine combustion chambers can be 3D-printed at lower laser power than previously achieved, using artificial intelligence to select the process settings. According to the Washington State University Office of Research, the team found six successful configurations across different laser power levels within a total budget of 40 experiments conducted over three months, including a first successful print of GRCop-42 at 500 watts. The work was carried out jointly by the School of Electrical Engineering and Computer Science and the School of Mechanical and Materials Engineering, with a collaborator from the University of Minnesota, and was published in the Proceedings of the AAAI Conference on Artificial Intelligence, where it received the Innovative Deployed Application Award at the group's annual conference.

Six settings found in 40 tries

The central result is a sharp reduction in experimental effort. The search space contained more than 100 million possible printing configurations, of which only a very small fraction was expected to work, according to the Washington State University Office of Research.

The team started from results for 37 unsuccessful configurations previously tested in the School of Mechanical and Materials Engineering. It then developed a model that estimated the likelihood that an untested configuration would produce a successful print, and used that model to choose small batches of new configurations for physical testing.

The selection approach balanced two goals at once. It favored settings that looked promising based on results so far, while also sampling uncertain areas that could improve the model for later rounds. That balance between testing likely winners and learning about unknown regions is central to the reported efficiency gain.

After each batch, the engineering team printed GRCop-42 with the AI-selected settings and evaluated the resulting samples. Doctoral student and first author Azza Fadhel received those outcomes for updating the model, and valued both successful and failed prints because each result supplied information for the next choice.

Within three months and 40 total experiments, that loop produced six successful configurations at different laser power levels. According to Aerospace Testing International, one of those configurations produced the first successful print of GRCop-42 at 500 watts of laser power. The outcome matters because earlier attempts to print the alloy at the lower wattages used by more common commercial machines had not succeeded.

Why GRCop-42 matters for aerospace

GRCop-42 is an alloy of three metals: copper, chromium, and niobium. It was developed by NASA, according to the Washington State University Office of Research. Its combination of properties explains its aerospace role.

The alloy offers high thermal conductivity and retains strength under extreme heat. Those characteristics suit components that must move heat efficiently while remaining structurally sound at very high temperatures, such as liquid rocket engine combustion chambers. Aerospace Testing International describes the same pairing as both high-temperature strength and efficient heat transfer, suited to combustion-chamber environments.

The Washington State University Office of Research describes the alloy as common in the aerospace industry but notes many other potential applications. The published report does not enumerate those additional uses, presenting them as potential rather than demonstrated in this study. The near-term significance in the study is therefore tied to aerospace hardware that already uses the material, plus the possibility that easier printing could support wider examination of other uses over time.

The French-language trade publication account of the same work similarly identifies GRCop-42 as a NASA-developed copper-chromium-niobium material used for combustion-related components, underscoring that the composition and high-temperature role are consistently reported across coverage.

Why the alloy was hard to print

The difficulty was not the alloy's performance once formed, but the conditions needed to form it by additive manufacturing. GRCop-42 is expensive and energy-intensive to print, requiring a significant amount of laser power, according to the Washington State University Office of Research. Aerospace Testing International likewise reports that the process typically requires substantial laser power and energy.

That requirement limited who could work with the material. Research leader Jana Doppa, Huie-Rogers Endowed Chair Professor of Computer Science and Berry Distinguished Professor in Engineering, said that 90% of commercial printers cannot print this metal alloy. Finding feasible process parameters for those commercial machines, in Doppa's description, would essentially democratize printing of the alloy.

Previous efforts to print at lower wattages and laser powers used by more common commercial printers had been unsuccessful. Exhaustive searching was not practical. A single printing run can cost hundreds of dollars, while detailed post-print quality analysis can take days, according to the Washington State University Office of Research. Aerospace Testing International reports the same cost and time constraints: hundreds of dollars per print and several days to analyze a finished sample.

Physical failure modes added to the expense. In some cases, a selected configuration produced a part that simply melted and was not really printable, as described by Fadhel. Without guidance, researchers faced the prospect of testing large numbers of costly builds with little information about which combinations deserved priority.

How the AI narrowed 100 million options

The project began with failure data, not success data. The 37 unsuccessful configurations from earlier work in the School of Mechanical and Materials Engineering provided the starting point for estimating success probability elsewhere in the large search space.

The computer-science team built a predictive approach that scored untested configurations by their estimated chance of producing a successful print. The model then recommended small batches, rather than one setting at a time, for laboratory testing. Batch selection allowed the team to use laboratory time efficiently while still gathering enough feedback to refine predictions.

The method explicitly managed exploration and exploitation. Promising options received attention because they might work immediately. Uncertain options also received attention because testing them could improve the AI model and lead to better choices in later batches. Both publications describing the study emphasize that dual objective.

The laboratory work was conducted with Nathaniel Zuckschwerdt, Susmita Bose, and Amit Bandyopadhyay in the School of Mechanical and Materials Engineering. Aryan Deshwal of the University of Minnesota also collaborated on the project. The engineers printed samples using the AI-selected process configurations and evaluated the results, then returned outcomes to the modeling team.

Fadhel, the first author and a doctoral student in computer science, said she welcomed all returned results, including failures, because every result improved the model. That feedback loop is important to understanding why 37 initial failures plus 40 new experiments could be sufficient: negative outcomes were not discarded, but used to update estimates about neighboring regions of the configuration space.

Doppa characterized the learning problem as unusually difficult for AI. Each attempt produced essentially a binary success-or-failure signal, and the goal was to minimize the number of attempts needed to locate rare successful settings quickly. Aerospace Testing International quotes the same assessment, framing the task as finding successful needles very quickly while limiting the number of attempts.

The researchers also entered the project with the expectation that success would be rare. They knew that only a very small fraction of the more than 100 million configurations would succeed. They additionally faced uncertainty associated with deploying a method where real people, materials, and physical costs were involved, and did not know in advance whether the approach would succeed.

What lower-power printing would change

Printing at lower laser power is reported to offer several practical effects. Lower power could reduce energy use, equipment wear, and post-processing costs, according to the Washington State University Office of Research. Aerospace Testing International reports the same three categories: reduced energy consumption and wear on printing equipment, plus lower cost for processing samples after printing.

Access is the other reported effect. Lower-power settings could make GRCop-42 accessible to universities, small laboratories, and companies that do not own specialized high-power equipment, according to the Washington State University Office of Research. Aerospace Testing International similarly states that the alloy would be within reach of universities, smaller laboratories, and companies without access to high-power printing systems.

That access point connects directly to Doppa's statement about commercial printers. If settings compatible with widely available commercial equipment prove robust beyond the laboratory demonstration, organizations without specialized high-power systems could experiment with an alloy previously confined to facilities with greater laser capacity and budgets. The study itself demonstrates six configurations at different power levels, rather than a full qualification of the alloy for every commercial machine.

The findings also suggest a change in development cost structure for difficult alloys. Because a single run costs hundreds of dollars and analysis takes days, any method that reaches workable settings in dozens rather than thousands of trials reduces material, equipment-time, and labor demands. In this case, the reported total was 40 experiments over three months, starting from 37 prior failures.

Award, publication, and broader method

The research team published its work in the Proceedings of the AAAI Conference on Artificial Intelligence and received the Innovative Deployed Application Award at the group's annual conference, according to the Washington State University Office of Research. Aerospace Testing International likewise identifies the publication venue and the Innovative Deployed Application Award.

The award name reflects the study's emphasis on deployment under physical constraints, not simulation alone. The AI-selected configurations were physically printed and evaluated, with materials, equipment time, and analysis time at stake in each round. Doppa said there was always uncertainty when deploying something involving real people, materials, and physical costs, and expressed surprise at how well the effort performed despite that risk.

The researchers believe the same AI-guided framework could be adapted to discover processing conditions for other metal alloys and additive-manufacturing systems, according to the Washington State University Office of Research. More broadly, the team points to scientific discovery problems in which successful outcomes are rare and experiments are too costly to test every possibility. Aerospace Testing International describes the same potential extension to other alloys and additive-manufacturing systems, and to scientific problems where successful outcomes are rare and each experiment carries high material, financial, or time costs.

The Washington State University Office of Research also notes that the AI techniques used could be applied in a variety of fields, including drug discovery. That statement positions the contribution as a search strategy for rare successes under tight experimental budgets, rather than a printing recipe limited to one alloy. The reported evidence for that broader claim in these sources is the GRCop-42 demonstration itself, not a separate drug-discovery experiment described in the same announcement.

For readers tracking additive manufacturing, the concrete takeaway is narrow and documented: from 37 prior failures, plus 40 new AI-guided experiments over three months, the collaboration identified six successful GRCop-42 configurations at different laser power levels, including a first at 500 watts. For researchers in other fields facing large search spaces, binary feedback, and costly tests, the team offers the selection framework as a candidate approach for future adaptation.

Sources referenced