Prediction of Vickers hardness of amorphous alloys based on interpretable machine learning

Amorphous alloys are formed by liquid alloys under rapid cooling. Because there is no grain boundary, dislocation, and other defects, it shows excellent mechanical properties. Vickers hardness (HV) is one of its excellent properties. In this work, four machine learning models were applied for HV modeling. The input features are atomic fraction, structural features and load, respectively. The determination coefficients R2 of the Light Gradient Boosting Machine (LightGBM) model in the test set are 0.981 and 0.979, respectively, which is far superior to the other three models. The LightGBM model has the best generalization ability. In addition, the shapley additive explanations (SHAP) theory is introduced to improve the interpretability of model. An important discovery of SHAP theory is that XP1 and Tm1 are two most important features, each of which has a critical value. If XP1 and Tm1 of the alloy are in the correct region, which can improve HV.