Domain-knowledge-guided active design and transfer learning of high-performance bimetallic composites

Integrating dissimilar metals into a single load-bearing architecture offers a route to lightweight structural materials with high damage tolerance. Yet, the reliability of reactive bimetallic is constrained by brittle intermetallic compounds (IMCs), triggering premature interfacial fracture. Herein, we develop a domain-knowledge-guided active design framework that integrates mixing-enthalpy and phase-structure descriptors with source-domain generation and target-domain transfer learning to accelerate filler-alloy discovery. This framework identified Cr 1.1 Cu 24.4 Ni 20.3 Ti 3.5 Co 21.8 Fe 20.3 Al 8.6 at.% and Cu 17.6 Ni 17.7 Co 18.0 Cr 18.3 Mo 10.5 Fe 17.9 at.% filler alloys, enabling Ti-Ni and Ti-Fe composites with fracture strengths of 475.44 and 340.07 MPa and stable friction properties. Ti-Ni composites formed a Ti 2 (Ni, Co, Fe)/Ti(Ni, Co, Fe) dual-phase with deformation discrepancies. Ti-Fe composites developed a Ti 2 (Ni, Co, Fe)/Ti(Ni, Co, Fe)/(Mo, Cr) ss multiphase, revealing incoherent interfaces with localized strain. Despite distinct failure pathways, both composites performed outstandingly, attributed to the incorporation of chemically complex IMCs within the proposed framework. This work established a transferable, physics-guided filler design paradigm for bimetallic composites.