Machine-learning-guided design of a biomedical high-entropy alloy for additive manufacturing: cast-state benchmark and preliminary LPBF feasibility assessment

Additive manufacturing of biomedical high-entropy alloys (BioHEAs) demands a combination of low elastic modulus, high strength, and damage tolerance, yet composition discovery remains largely empirical. Here, we establish a machine-learning framework that couples virtual screening with physical prototyping to link composition, deformation mechanism, and properties. Ensemble models for strength, elongation, and modulus were applied to screen Ti–Zr–Nb–Ta–Mo-centered quinary-to-septenary spaces (∼15 million compositions), revealing discrete performance islands anchored by a Ti–Zr backbone. A Zr-rich BCC alloy (Zr₃₉.₃Ti₁₉.₅Nb₁₇.₉Ta₁₆.₈Mo₆.₅) was identified and validated. In the as-cast state, it delivers ∼1.0 GPa yield strength, 22.3% elongation, and an 88 GPa elastic modulus; ductility originates from dislocation-mediated kink-band plasticity. Preliminary laser powder bed fusion (LPBF) trials confirm printability (97% relative density, single BCC, 821 MPa ultimate tensile strength, 87.9 GPa modulus) but exhibit limited elongation (2.5%) due to residual porosity and stresses, indicating that defect suppression is needed to approach cast-state ductility. This digital-to-physical workflow offers a transferable strategy for prioritizing BioHEA compositions before costly AM optimization.