Washington State University researchers used AI to find successful 3D printing configurations for a high-performance NASA alloy, avoiding millions of manual tests.
Key facts
- •The AI model searched through more than 100 million possible printing configurations to find successful settings.
- •GRCop-42 is a NASA-developed alloy made of copper, chromium, and niobium used in liquid rocket engine combustion chambers.
- •The research team successfully printed the alloy using 500 watts of laser power, a level previously considered unfeasible for common commercial printers.
- •The project was conducted by researchers from Washington State University's engineering schools and received the Innovative Deployed Application Award.
- •The AI-guided method could be applied to other scientific fields, such as drug discovery, where testing every possible option is prohibitively expensive.
Researchers at Washington State University have developed an artificial intelligence strategy to identify efficient 3D printing parameters for GRCop-42, a high-performance alloy used by NASA. By using AI to navigate over 100 million potential printing configurations, the team successfully printed the material using lower laser power settings. This approach could allow common commercial printers to produce the alloy, which is typically restricted to specialized, high-power equipment.
By the numbers
Overcoming Printing Challenges
GRCop-42 is a copper, chromium, and niobium alloy valued for its heat resistance and thermal conductivity in aerospace applications, such as rocket engine combustion chambers. However, the material is difficult to print because it usually requires high laser power. Previous manual testing was impractical due to the high costs of materials and the time required for analysis, with many attempts resulting in failed, melted products.
AI-Driven Optimization
The research team utilized data from 37 previously failed experiments to train an AI model capable of estimating the success probability of untested settings. The model selected small groups of configurations to test, balancing promising options with exploratory ones to improve its predictive accuracy. Over three months, the team limited their work to 40 experiments and successfully identified six viable configurations, including the first successful print using 500 watts of laser power.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by ScienceDaily.
