Revealing the Materials Genome of Superhard High-Entropy Diborides via the Hybrid Data-driven and Knowledge-enabled Model

Abstract Materials descriptors with multivariate, multiphase and multiscale of a complex system have been treated as the remarkable materials genome, addressing the composition-processing-structure-property-performance (CPSPP) relationships during the development of advanced materials. With the aid of high-performance computations, big data and artificial intelligent technologies, it is still a challenge to derive the explainable machine learned model to reveal the underlaying CPSPP relationship, especially, under the extreme conditions. Here, we propose a hybrid data-driven and knowledge-enabled model with two key descriptors to design the superhard high entropy boride ceramics (HEBs), which is not only in line with the common features from various machine learning algorithms but also integrate the solid-solution strengthening mechanisms. While five dominate features in terms of load, valence differences, electronegativity, electron work functions, and the differences among solutes in various column of periodical elementary table were screened out from 149 ones, the best optimal machine learning (ML) algorithm was addressed among decision tree, support vector regression, K-Nearest Neighbor, random forest, Adaboost, gradient enhanced regression tree, Bagging, ExtraTree, and XGBoost. The Shapley additive explanation the key influence trend for material hardness with the change of HEBs electronic properties. Correspondingly, the predicted 14 potential best superhard HEB candidates via ML are further validated by first-principles calculations via the aforementioned knowledge-based model. This work supports a smart strategy to derive the hybrid data-driven and knowledge-enable explainable model predicting the target properties of advanced HEBs and paves a path accelerating their development at cost-effective approach.