Data‐driven phase prediction for polymorphic high‐entropy rare‐earth disilicates

Abstract Determining the phase formation of high‐entropy rare‐earth disilicates (HEREDs) is the prerequisite to developing novel HEREDs for thermal environmental barrier coating. However, the polymorphism of HEREDs has made the exploitation challenging. Herein, a data‐driven strategy combined with high‐throughput experiments and multi‐class classification algorithms is proposed to explore the polymorphic phase formation of HEREDs. Specifically, 92 equimolar HERED samples with five polymorphic phases are well synthesized by the high‐throughput laser synthesis techniques, and 32 potential phase formation descriptors based on fundamental parameters of constituent precursors are collected simultaneously. By using feature engineering and multi‐class classifications, a combination of four descriptors within the SVC.ploy model is determined to achieve the highest validation accuracy of 98.9%. Further studies have verified the average ionic radius as the single criterion in distinguishing the formed HERED phases. The accuracy of our established model is further validated by reproducing previously reported 20 HEREDs with a high validation accuracy of 85%. Based on the trained model, high‐throughput screening in the equimolar HERED system has been conducted, where 4770 C‐type, 538 D‐type, 35717 E‐type, 319 F‐type, and 23495 G‐type HEREDs have been predicted. This work provides effective guidance to develop HEREDs in the desired phases.