Crystal structure guided machine learning for the discovery and design of intrinsically hard materials

In this work, a machine learning (ML) model was created to predict intrinsic hardness of various compounds using their crystal chemistry. For this purpose, an initial dataset, containing the hardness values of 270 compounds and counterpart applied loads, was employed in the learning process. Based on various features generated using crystal information, an ML model, with a high accuracy (R2 = 0.942), was built using extreme gradient boosting (XGB) algorithm. Experimental validations conducted by hardness measurements of various compounds, including MSi2 (M = Nb, Ce, V, and Ta), Al2O3, and FeB4, showed that the XGB model was able to reproduce load-dependent hardness behaviors of these compounds. In addition, this model was also used to predict the behavior based on prototype crystal structures that are randomly substituted with elements.