GRCop-42 is NASA’s copper-chromium-niobium alloy for rocket combustion components: high thermal conductivity, good fatigue strength, excellent creep resistance, and notoriously hard to print. It normally needs high laser power and specialized machines, which keeps it out of reach for smaller labs and companies.
A Washington State University team led by Jana Doppa attacked the parameter search with AI instead of brute-force trial and error. They started with 37 failed print attempts already on record from WSU’s School of Mechanical and Materials Engineering, used those as training data, and built a model to predict which laser and process settings would actually work before running them on the machine. Every new attempt, success or failure, fed back into the model. As Doppa put it, the signal per try is “basically binary, success or failure,” and the point of the model is cutting the number of tries needed to hit a working setting.
Over three months the team ran 40 experiments and came out with six settings that print GRCop-42 at different laser power levels. The headline result: one of those six settings printed the alloy successfully at 500 watts, the first time GRCop-42 has been printed at that power level. Lower laser power means less energy draw and, per the researchers, the possibility of skipping the heavy, expensive, specialized equipment the alloy has needed until now.
What’s not in the source: the laser power levels used in the original 37 failed attempts, the machine or printer model used in testing, the specific print settings (scan speed, layer thickness, hatch spacing) behind the six identified parameter sets, and any mechanical test data (fatigue, tensile, porosity) on the parts printed at 500 W. Cost figures for the AI-guided approach versus conventional trial and error are also not given, beyond the general claim of being “more affordable.”