Unifying Composition and Process Design: A Heterogeneous Graph Neural Network for Discovering High-Performance Cu Alloys.

The properties of copper alloys critical for sustainable development depend on their processing pathways. However, the complex and variable character of these pathways is incompatible with conventional machine learning models that require fixed-input structures, creating a barrier to AI-driven materials design. To address this challenge, we introduce a heterogeneous graph neural network to model the material system including both elemental composition and process steps as a unified graph. The graph integrates element and process nodes, interconnected by learnable edges representing the intricate relationships between composition and processing steps. This approach inherently accommodates variable-length process pathways and bypasses manual feature engineering by directly utilizing native elemental properties. It effectively captures the complex coupling between composition and processing without suffering from data sparsity or dimensional explosion. Applying this framework, we designed and experimentally validated a novel copper alloy (Cu-Cr-Zr-Y-La-Mg-Zn) and tailored process, achieving an exceptional combination of 710 MPa in yield strength, 726 MPa in tensile strength, and 75% IACS (International Annealed Copper Standard) in electrical conductivity. This work presents a new approach for materials discovery, offering a scalable solution to co-optimize composition and processing for materials defined by complex manufacturing histories.