Here is a number that should make every process engineer wince: more than 100 million. That is how many combinations of laser power, scan speed, powder feed, gas flow, and layer height a team at Washington State University was staring at when they tried to 3D print GRCop-42, the copper alloy NASA developed for rocket engine combustion chambers. Nobody can test 100 million settings. A single print run costs hundreds of dollars and the quality analysis afterward can take days. So the team let an AI pick which experiments were worth running. Forty prints later, they had six working recipes, including one at 500 watts, a laser power the material was not supposed to tolerate.
The paper, published in the Proceedings of the AAAI Conference on Artificial Intelligence, won the conference's Innovative Deployed Application Award, and it got fresh attention this week in 3DPrint.com's September 10 news briefs. We think it matters far beyond rocket engines.
What GRCop-42 is, and why it fights the laser
GRCop-42 is a mix of copper, chromium, and niobium. Copper gives it very high thermal conductivity, which is what you want lining a combustion chamber. That same property is the problem for a laser printer: copper reflects much of the infrared light a fiber laser puts out, and the energy that does get absorbed drains away from the melt pool almost immediately. The usual fix is brute force, running the laser well above 900 watts. According to Jana Doppa, the WSU computer science professor who led the project, roughly ninety percent of commercial printers cannot print this alloy at all. Closing that gap would let universities, small labs, and small manufacturers with lower-power directed energy deposition (DED) systems work with it directly.
My prints keep failing and I don't know which setting to change. How do I find working parameters without burning a spool?
Start with your failures, change only a few variables at a time, and let each result narrow the next guess; that is exactly what WSU's AI did, just faster and with better bookkeeping. The method, called BEAM (Bayesian Experimental design for Additive Manufacturing), started with 37 earlier print attempts that had all failed. Instead of treating those as wasted, the model used them to estimate how likely any untested combination was to succeed. It then picked small batches of new settings that balanced two goals: try the most promising options, and probe the uncertain corners where the model might learn the most. Every result, pass or fail, went back into the model. First author Azza Fadhel, a PhD student, put it plainly: she liked every result, even the failures, because every one improved the AI.
The search covered powder feed rate, shield gas flow, the thickness of the Inconel 718 base, scan speed, and layer height, with laser power fixed at 950, 700, 600, and 500 watts and a budget of ten experiments per level. The system found at least one workable setting at every level, three of them at 600 watts. Failed samples looked like thin pillars or blobs. Successful ones were solid test blocks, later scaled to cylinders that machined clean with minimal internal defects. One insight is worth writing on the shop wall: the successful configurations generally paired fast scan speeds with thin layers. At 700 watts, a working recipe ran 1,600 mm/min at 0.11 mm layer height.
I need a part in a material my printer supposedly can't handle. Is there a way around that?
Often yes, but the answer is usually a process window, not a new machine, and finding that window is real engineering work that costs test prints. The WSU result is a good example. The alloy did not change. The printer did not change. What changed was the combination of feed rate, speed, and layer height that let a 500 watt laser keep a copper melt pool stable long enough to build a solid wall. The same logic applies on the polymer side. When a customer brings us a part that has to survive an autoclave or a hot engine bay, the conversation is not about whether PEEK or ULTEM can be printed. It is about chamber temperature, orientation, layer height, and dry material, all dialed in for that specific geometry. That is what our PEEK and ULTEM printing service is built around: we run the test prints so you do not have to.
One honest caveat the researchers make themselves: finding a settings window that produces a solid sample is not the same as qualifying a part for flight. The paper establishes feasibility from 500 to 950 watts. It does not certify anything for aerospace service. That distinction matters for anyone buying parts, too. A print that looks good is the start of a conversation about testing, not the end of it.
What this means for makers and engineering buyers
For makers, the takeaway is method. Most of us tune a new filament by changing five things at once, printing a cube, and hoping. BEAM shows the value of the opposite: change fewer things, record everything including the failures, and let the pattern emerge. A spreadsheet of what you tried and what happened is a crude version of the model WSU built. For buyers, the takeaway is capability. Materials once locked behind expensive hardware get more accessible as process windows widen, and we see the same trend in high-temperature polymers and carbon fiber composites. If you have a legacy part in a tough material, we can scan it, rebuild the model, and print it here in Rockville, Maryland, and we will tell you up front which materials we have a proven window for.
The WSU team says the same framework could be adapted to other alloys and other printers. If you are a materials or equipment brand with something new you want put through its paces on a working production floor, email info@dcadditivepros.com and tell us what you are building.
Frequently asked questions
What is GRCop-42 and why is it hard to 3D print?
GRCop-42 is a NASA-developed copper, chromium, and niobium alloy used in rocket engine combustion chambers. Copper reflects much of a fiber laser's light and pulls heat away from the melt pool very quickly, so it usually needs high laser power to print at all.
How did WSU use AI to find 3D printing settings?
Their BEAM method started with 37 failed prints, used them to predict which untested settings were most likely to work, then ran small batches of experiments and fed every result back into the model. In 40 experiments over three months it found six working configurations, one at 500 watts.
Can I use this kind of AI-guided tuning on my own 3D printer?
The research code is not a consumer product yet, but the approach is: change a few variables at a time, log every result including failures, and use the pattern to choose your next test. A careful spreadsheet gets you a surprising share of the benefit.
Does a successful test print mean the part is qualified for aerospace use?
No. The WSU study established that the alloy can be printed at lower laser power. It did not qualify parts for flight, and any load-bearing or high-temperature part still needs its own testing and certification.
Sources: Washington State University news release, August 24, 2026; 3D Printing Industry, September 2, 2026; 3DPrint.com 3D Printing News Briefs, September 10, 2026.