AI Found a Way to 3D Print NASA's GRCop-42 Rocket Alloy at Just 500 Watts

A NASA alloy that most printers cannot touch

GRCop-42 was developed by NASA for the kind of places where materials regularly fail. The alloy, a mix of copper, chromium, and niobium, conducts heat beautifully and stays strong under temperatures that would wreck ordinary metals. That is why it turns up in liquid rocket engine combustion chambers. It is also why it is a pain to 3D print. The process normally demands so much laser power that around 90 percent of commercial metal printers are ruled out before the first layer is laid down.

Turning 100 million options into 40 experiments

Researchers at Washington State University decided to stop guessing. Led by computer scientist Jana Doppa and PhD student Azza Fadhel, the team built an AI model that could predict which untested combinations of print settings might actually work. They fed it results from 37 earlier failed configurations. The model then recommended small batches of new settings, balancing obvious candidates against riskier ones that would teach it more about the problem.

Over three months, the team ran just 40 experiments. Out of more than 100 million possible configurations, the AI identified six that produced successful prints. One of those worked at 500 watts of laser power, the first time GRCop-42 has been printed at that low a level.

Why lower power matters

Printing at 500 watts is not just a number for a research paper. Lower laser power means less energy, less wear on expensive optics, and lower post-processing costs. It also opens the door for universities, small labs, and companies that own ordinary commercial metal printers but cannot justify a high-power industrial system. If the alloy can move out of specialized aerospace shops and into more common workshops, its applications could spread well beyond rocket nozzles.

The method could reach beyond metal printing

The WSU team, including materials scientists Nathaniel Zuckschwerdt, Susmita Bose, and Amit Bandyopadhyay, published the work in the Proceedings of the AAAI Conference on Artificial Intelligence and took home the Innovative Deployed Application Award. Collaborator Aryan Deshwal from the University of Minnesota also contributed. The same Bayesian experimental design approach could, in principle, be pointed at drug discovery or any other field where testing every option is too slow and too expensive.

What comes next

For now, the headline is that a notoriously difficult aerospace alloy just became printable on machines that were never supposed to handle it. That does not mean every workshop will be printing rocket parts next year. Certification, repeatability, and scale still have to follow. But the result suggests that some of additive manufacturing's hardest problems are better solved by smarter experimentation than by brute force.