Knowledge‐Constrained Multi‐Objective Reinforcement Learning for Simultaneous Optimization of Printability and Mechanical Properties in Additively Manufactured Superalloys

Multi‐component superalloys are prone to cracking under the extreme nonequilibrium solidification in additive manufacturing (AM), causing printability‐performance mismatches. Herein, we propose a knowledge‐constrained multi‐objective reinforcement learning (MORL) framework for rational design of crack‐free, high‐strength, high‐ductility and creep‐resistant superalloys for AM. Unlike conventional Pareto‐front approaches, our framework embeds a hard constraint on cracking susceptibility directly into a reward function, enabling goal‐directed navigation of materials satisfying both printability and balanced mechanical properties. The MORL efficiently explores a composition space of 44,880,000 candidates, designing two alloys in a single iteration that are crack‐free with yield strength over 800 MPa and elongation above 9% at 800°C. Analysis of the MORL‐guided trajectory reveals an unanticipated design strategy: elevating Cr, Ti, Mo, and W to promote discrete M 23 C 6 and MC carbides at grain boundaries, which refine grains via pinning and accommodate high‐temperature deformation, simultaneously enhancing strength and ductility. Notably, both alloys lie within composition domains predicted to be crack‐prone by established AM cracking criteria yet remain crack‐free, overcoming these limitations and demonstrating that data‐driven models capture complex interactions missed by semi‐empirical indicators. This work establishes MORL as an efficient design tool for discovering metallurgical principles and provides a transferable paradigm for developing high‐performance multicomponent alloys for AM.