Abstract: High-entropy alloys (HEAs) eliminate the traditional distinction between “principal” and “alloying” elements, leading to the development of high-entropy superalloys (HESAs) with coherent γ-γ' dual-phase structures. However, the inclusion of multiple principal elements in HESAs significantly broadens the compositional design space, posing challenges for conventional trial-and-error approaches. To address this, we propose an integrated strategy that combines first-principles calculations with both unsupervised and supervised machine learning (ML) to accelerate the discovery of L1₂-strengthened Co-Ni-Cr-Al-Cu-Ti HESAs. In this framework, ML is first used to identify suitable strengthening elements, followed by thermodynamic calculations to refine candidate compositions. The mechanical properties and microstructural features of the optimized alloys are then validated experimentally. Notably, Cu plays a key role in stabilizing the L1₂ phase in HESAs; however, excessive Cu content (above 5 at. %) can depress both the melting point and mechanical strength. Using this approach, we successfully designed Co38Cr23Ni22Al6Ti6Cu5, which exhibits a compressive yield strength of 633 MPa at 700 ℃, an L1₂ solvus temperature of 1030 ℃, and a melting point of 1220 ℃. These results provide valuable insights into the accelerated design of HESAs with coherent γ-γ' dual-phase microstructures.

Key words: High-entropy superalloy, Machine learning, First-principle calculations, CALPHAD, Mechanical property