High-throughput computational and machine-learning design of high-entropy alloys based on thermodynamic and empirical parameters: Al-Si-Cr-Fe-(Ni,Mn) systems

A coupled high-throughput computational and machine-learning design approach was developed to identify new body-centered-cubic-based alloys in Al-Si-Cr-Fe-(Ni,Mn) systems with high strength, high-thermal stability, low density and low cost. This study highlights the possibility of designing alloys using a series of thermodynamic and empirical models-based calculations for each alloy system. The CALPHAD method was used to generate a general map of stable phases from 873 K to the melting point for further screening of alloys with no undesirable intermetallic compounds. In addition, phase prediction of quinary, quaternary and senary alloys containing all elements (17,832 alloys) at 5 K below the solidus temperature was accomplished using tree-based ML models. High-throughput alloy screening was conducted on the predicted results of XGBoost, as the best performing model, to find alloys with no intermetallic compounds. The conducted screening criteria suggest single-phase HEAs over a wide temperature range and a structure-based strength calculation model estimates a high strength of >1000 MPa for these alloys. It is a step forward to address the potential of surrogate ML models for phase prediction across many alloys, followed by high-throughput screening in order to develop high-performance alloys. • A concept of coupling high-throughput computational and ML models was conducted in order to develop new BCC-structured HEAs in Al-Si-Cr-Fe-(Ni,Mn) systems. • Surrogate ML models for phase prediction were proposed as fast and accurate method instead of computationally expensive models. • High-throughput alloy screening from predicted results by XGBoost model suggested 234 HEAs with BCC/B2 phases among 17,832 HEAs in the Al-Si-Cr-Fe-Mn system.