Intelligent prediction and oriented design of high-hardness high-entropy ceramics

High-entropy ceramics are considered vital candidate materials for applications in rail transportation and advanced manufacturing due to their exceptional hardness and wear resistance. However, the intricate relationship between their composition and mechanical properties presents challenges for on-demand material design. This work utilizes machine learning and heuristic optimization algorithms to achieve accurate predictions of bulk high-entropy ceramics hardness (with validation set errors <10 %) and the oriented design of high-entropy ceramics with a hardness of 25 GPa (with an average error of 2.6 %). This achievement is attributed to three key innovations: the construction of the feature space based on the Pearson correlation coefficient and genetic algorithm, along with algorithm selection and optimization through hyperparameter tuning; the novel combination of reduced-dimensionality component compositions with atomic/precursor descriptors, achieving a model R 2 value of up to 0.898; the optimization of constituent elements using genetic algorithm and principal component analysis, providing direct guidance for the design of high-hardness high-entropy ceramics. Through these methodologies, new high-entropy ceramics compositions, typically represented by (Ti 0·25 Zr 0·25 Nb 0.25 Hf 0.25 )C, were successfully designed, achieving hardness of up to 23.9–25.2 GPa. This oriented design methodology holds promise for accelerating the on-demand design of high-entropy ceramics.