Prediction of tensile strength and composition optimization of AlCoCrFeNiMn high entropy alloy driven by machine learning

High-entropy alloys (HEAs) exhibit superior tensile strength due to the synergistic effects of multiple principal elements. However, the complex correlations between composition and tensile strength severely constrain the efficiency of traditional trial-and-error development strategies. Compared to quaternary and quinary systems, senary HEAs, represented by the AlCoCrFeNiMn system, feature more intricate elemental synergistic interactions, where minor compositional adjustments can induce significant variations in tensile strength, thereby dramatically increasing the difficulty of establishing reliable composition–property relationship. Here, a physics-informed machine-learning design strategy is proposed. By constructing a physical feature library and optimizing seven machine-learning algorithms, an optimal model is developed for high-throughput screening of 102 642 potential compositions. Experimental validation demonstrates that the optimized Al18Co15Cr16Fe20Ni30Mn1 HEA exhibits a tensile strength of 1050.22 MPa, with a deviation of only 1.4% from the extreme gradient boosting prediction. Furthermore, feature importance analysis reveals the dominant strengthening mechanisms: Al enhances solid-solution strengthening by inducing moderate lattice distortion through tuning the atomic size mismatch (δr), whereas Ni stabilizes the face-centered cubic structure and suppresses the precipitation of brittle phases. These effects synergistically interact with high configurational entropy (ΔS) and elastic modulus mismatch (η) to collectively govern the tensile strength of the alloy. This work systematically validates the prediction accuracy within the AlCoCrFeNiMn system and establishes a precise alloy-design paradigm characterized by a “data-driven prediction–physical mechanism coupling–experimental validation” closed loop, promoting the development of HEA tensile-strength design from empirical trial-and-error toward rational material design.