Large Language Model-Informed Dual-Track AI Framework for the Synergistic Design of Crack-Free and High-Strength...

Area of Science:

• Materials Science and Engineering
• Artificial Intelligence in Manufacturing
• Additive Manufacturing

Background:

• Laser Powder Bed Fusion (LPBF) of nickel-based superalloys is promising but limited by hot-cracking and costly optimization.
• Traditional AI models face "black-box" issues and data scarcity, hindering efficient material discovery.

Purpose of the Study:

• To develop a knowledge-informed hybrid AI framework integrating Large Language Models (LLMs) and reinforcement learning (RL) for accelerated superalloy discovery.
• To overcome limitations of traditional AI, such as data scarcity and lack of interpretability, in LPBF process optimization.

Main Methods:

• Implemented a hybrid AI framework combining LLMs for heuristic pruning and RL for process optimization.
• Utilized LLMs to operationalize empirical metallurgical heuristics, reducing the candidate search space by over 95%.
• Injected LLM-distilled process priors into an RL agent for "warm-start" optimization with safety constraints.

Main Results:

• Developed the AMN01 alloy with crack-free printability using the AI framework.
• Achieved a breakthrough yield strength >1.5 GPa and ultimate tensile strength (UTS) ~1.8 GPa in the direct-aged state.
• The alloy maintained UTS >1.6 GPa with >15% elongation after solution-aging, exhibiting a multi-tier strengthening architecture.

Conclusions:

• The knowledge-informed hybrid AI framework enables accelerated material discovery in data-scarce, high-cost environments.
• This approach successfully bridges metallurgical expertise with autonomous AI-driven discovery for advanced alloys.
• The developed framework offers a generalizable paradigm for optimizing complex material systems via additive manufacturing.