AI Makes NASA’s Aerospace Alloy Easier to 3D Print

Metal 3D printing can create rocket engine parts with internal cooling channels and geometries that conventional manufacturing struggles to produce. However, the ability to draw an advanced component on a computer does not mean a manufacturer knows how to print it successfully.

Every combination of laser power, scanning speed, powder delivery, gas flow, and layer height can change the result. A poor combination may create pores, cracking, or weak bonding. Discovering the right settings often requires months of costly trial and error.

Researchers from Washington State University and the University of Minnesota have demonstrated¹ that artificial intelligence can make this search substantially more efficient. Their system identified multiple ways to print GRCop-42, a difficult NASA-developed copper alloy, using lower-power and more widely available metal 3D printers.

The achievement is important beyond one alloy. It suggests AI could help manufacturers bring new materials into production faster, lower the cost of process development, and expand the number of facilities capable of producing specialized aerospace components.

Why GRCop-42 Is Valuable For Aerospace Manufacturing

GRCop-42 is a copper, chromium, and niobium alloy developed for environments where components must survive extreme heat. It combines high thermal conductivity with strength, oxidation resistance, creep resistance, and resistance to thermal fatigue.

Those properties make it especially useful in rocket engine combustion chambers, heat exchangers, and other high-heat-flux systems. Copper conducts heat well, but conventional copper alloys can lose strength at elevated temperatures. GRCop-42 was engineered to balance both requirements.

The same characteristics that make the alloy valuable also make it difficult to print. Copper reflects much of the infrared energy produced by common fiber lasers. It also moves absorbed heat away from the melt pool very quickly. If the printer cannot create and maintain a stable melt pool, the deposited material may not bond properly.

Manufacturers can compensate by using more powerful equipment, but that raises the cost of entry. The researchers therefore asked whether GRCop-42 could be deposited successfully below 900 watts, with particular interest in the 500 to 700-watt range supported by many more machines.

How AI Replaced An Expensive Trial-And-Error Process

The team developed a framework called Bayesian Experimental Design for Additive Manufacturing, or BEAM. It does not operate the printer autonomously or invent a part. Instead, it decides which printing configurations are most useful to test next.

BEAM begins with earlier experiments, including failed prints. It builds a probabilistic model connecting printer settings with the likelihood of producing an acceptable structure, then selects another small batch for physical testing. Those results return to the model and the cycle repeats.

This is a practical use of AI because it keeps physical experimentation in the loop. Other researchers are similarly applying AI to improve the precision of laser-based metal processing. The model does not need to simulate every metallurgical interaction perfectly. It only needs to make each expensive experiment more informative than a largely manual guess.

The economic incentive is substantial. According to the paper, a single directed energy deposition run can cost between $500 and $1,000. Post-print analysis can require another $500 per sample and take one to four weeks. Exploring a new alloy’s process window can consequently exceed $100,000, even before accounting for months of specialist labor.

What The Researchers Tested

The researchers used a five-axis directed energy deposition system equipped with a 1,000-watt infrared fiber laser. Directed energy deposition feeds metal powder into a laser-generated melt pool, gradually building material onto a surface. In this case, the system deposited GRCop-42 onto Inconel 718, another alloy widely used in high-temperature applications.

The potential combinations of feed rate, gas flow, scanning speed, layer height, and laser power created roughly one million candidate configurations at each power level. Rather than exploring this enormous space manually, BEAM received a budget of only 10 experiments for each selected laser power.

Within three months, the framework found at least one feasible configuration at every power level. The successful samples included solid interiors with minimal defects and a smooth transition between the GRCop-42 and Inconel 718 materials. The team also discovered that unusually high scanning speeds and thinner-than-normal layers were important to producing solid structures.

Lower-Power Printing Could Expand Access To Advanced Alloys

Printing GRCop-42 at 500 watts does not merely reduce the electricity consumed by one laser. It changes which equipment can participate. The paper estimates that more than 90 percent of relevant printers operate between 500 and 1,000 watts. Finding reliable settings in that range could allow universities, smaller manufacturers, and specialized research facilities to work with an alloy that might otherwise require more expensive systems.

Lower-power operation may also reduce distortion, improve surface quality, limit wear on laser optics, and decrease post-processing requirements. These benefits would need to be measured for a specific production process, but they show why parameter discovery affects the economics of manufacturing as much as the technical feasibility.

Potential advantages include:

• Fewer failed builds during material development
• Lower equipment and post-processing requirements
• Faster qualification of new materials and components
• More distributed production of specialized metal parts

This last point is particularly relevant to aerospace and defense. These industries need low volumes of specialized parts and often face supply chains with few qualified suppliers. A process transferable to more machines could make limited production runs less dependent on a single facility.

AI Could Become A Materials Commercialization Tool

The larger opportunity is not confined to GRCop-42. Metal additive manufacturing supports many potentially valuable alloys, but each new material and machine combination requires its own process window. A setting that works on one printer may not transfer perfectly to another because of differences in lasers, powder, atmosphere, geometry, and thermal behavior.

This creates a hidden bottleneck. Materials scientists can formulate an alloy with impressive properties, while manufacturers may still lack an affordable and repeatable method of turning it into finished components. Similar AI-assisted methods are already being used to develop stronger and more corrosion-resistant 3D-printable steel. AI-guided experimental design can help bridge the remaining manufacturing gap by treating failed experiments as useful data and directing resources toward the parts of the search space most likely to produce viable results.

As companies gather data across machines and materials, models may eventually begin projects with knowledge learned from similar alloys. That would turn process-development history into a reusable manufacturing asset.

However, the present study should not be interpreted as proof that flight-ready components can now be printed at 500 watts. The researchers produced laboratory samples, not certified rocket engines. Aerospace adoption would still require repeatability studies, mechanical testing, component-level validation, and approval under demanding quality systems. BEAM accelerated the discovery of feasible settings, which is an early but expensive stage of a much longer commercialization process.

Investing In AI-Enabled Metal Additive Manufacturing

For investors interested in companies working where aerospace, AI, and additive manufacturing converge, Velo3D (VELO )offers a particularly relevant example. The company develops the Sapphire family of metal additive manufacturing systems for mission-critical aerospace and defense applications.

Velo3D has already qualified GRCop-42 for its platform. Its technology uses laser powder bed fusion rather than the directed energy deposition process tested in the paper, so BEAM’s configurations cannot simply be copied to a Sapphire printer. The broader principle is still relevant: AI-guided discovery could help printer manufacturers qualify difficult materials and reduce development costs.

That provides a natural investment thesis around the accumulation of process knowledge. The long-term competitive advantage in metal printing may not come solely from laser power or build volume. It may also depend on which company can reliably convert the widest range of advanced materials into repeatable, certifiable parts.

Velo3D remains a speculative small-cap company exposed to execution, financing, competition, and customer-concentration risks. Nevertheless, its aerospace focus and existing GRCop-42 capability make it a relevant publicly traded company to watch as AI becomes more deeply integrated into industrial process development.

AI Is Changing How Physical Technologies Are Developed

The most consequential feature of BEAM is that it applies AI to a physical industrial constraint. It does not generate a concept that still needs to be tested. It chooses which real-world test should happen next, learns from the outcome, and makes the following experiment more informed.

That closed loop could become a defining pattern for AI in manufacturing. As materials and components become more specialized, exhaustive experimentation becomes too slow and expensive. Systems that combine machine learning with expert knowledge can narrow the search without removing scientists or engineers from the process.

For aerospace manufacturers, the immediate result is a more accessible way to print a valuable NASA alloy. The deeper implication is that AI may shorten the path between discovering an advanced material and producing useful parts from it. That could expand additive manufacturing beyond geometric freedom and make it a faster platform for commercializing the materials themselves.

References:

^{1. Fadhel, A., Zuckschwerdt, N. W., Deshwal, A., Bose, S., Bandyopadhyay, A., & Doppa, J. (2026). Discovery of feasible 3D printing configurations for metal alloys via AI-driven adaptive experimental design. Proceedings of the AAAI Conference on Artificial Intelligence, 40(47), 39939-39948. https://doi.org/10.1609/aaai.v40i47.41428}