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

ABSTRACT Laser Powder Bed Fusion of high performance nickel‐based superalloys is a transformative technology hindered by acute hot‐cracking susceptibility and prohibitively expensive process optimization cycles. Traditional data‐driven AI models often suffer from “black‐box” limitations and severe data scarcity. Here, we propose a knowledge‐informed hybrid AI framework that integrates Large Language Models (LLMs) as reasoning agents to bridge metallurgical expertise with autonomous discovery. By operationalizing empirical heuristics (e.g., Al/Ti ratio correlations), the LLM facilitated rapid heuristic pruning, compressing the initial search space of 76 800 candidates by >95%. Subsequently, LLM‐distilled process priors were injected into a reinforcement learning (RL) agent, enabling a “warm‐start” optimization that achieved safety‐constrained exploration in physically risky regimes. The resulting AMN01 alloy achieved crack‐free printability with a breakthrough yield strength exceeding 1.5 GPa and an ultimate tensile strength near 1.8 GPa in the direct‐aged state. Following solution‐aging treatment, the alloy maintained a UTS over 1.6 GPa with an exceptional elongation of over 15%. Multi‐scale characterization revealed a multi‐tier strengthening architecture involving nanoscale cellular dislocation networks, γ' precipitation, and deformation‐induced Lomer–Cottrell locks. This framework establishes a generalizable paradigm for accelerated material discovery in high‐cost, data‐scarce engineering environments.