Machine learning-assisted mechanical property prediction and descriptor-property correlation analysis of high-entropy ceramics

High-entropy ceramics have attracted extensive attention due to their unique properties. However, the development of novel ceramics has been hindered by extensive trial-and-error strategies, along with insufficient knowledge and computational power. In this work, we develop machine learning (ML) models based on the chemical attributes of constituent elements and metal carbides of high-entropy carbides (HECs) for predicting their related Young's modulus, hardness and wear resistance values. Our models demonstrate low mean absolute errors (15.3 GPa for modulus and 1.1 GPa for hardness), high R 2 scores (0.969 and 0.963) and excellent agreement with experimental measurements, indicating high model robustness. We further establish a database of 230,230 HECs and analyse the correlations between chemical descriptors and their properties, especially for those containing transition metals from Groups IV, V and VI. Our models can rapidly explore the mechanical properties of HECs and help guide descriptor-property correlation analysis in a low-cost and reliable manner, which provides an efficient method for accelerating the design of novel high-entropy materials with desired performance characteristics.