Accelerated First-Principles Calculations Based on Machine Learning for Interfacial Modification Element Screening of SiCp/Al Composites

Abstract

SiCp/Al composites offer the advantages of lightweight construction, high strength, and corrosion resistance, rendering them extensively applicable across various domains such as aerospace and precision instrumentation. Nonetheless, the interfacial reaction between SiC and Al under high temperatures leads to degradation in material properties. In this study, the interface segregation energy and interface binding energy subsequent to the inclusion of alloying elements were computed through a first-principle methodology, serving as a dataset for machine learning. Feature descriptors for machine learning undergo refinement via feature engineering. Leveraging the theory of machine-learning-accelerated first-principle computation, six machine learning models—RBF, SVM, BPNN, ENS, ANN, and RF—were developed to train the dataset, with the ANN model selected based on R2 and MSE metrics. Through this model, the accelerated computation of interface segregation energy and interface binding energy was achieved for 89 elements. The results indicate that elements including B, Si, Fe, Co, Ni, Cu, Zn, Ga, and Ge exhibit dual functionality, inhibiting interfacial reactions while bolstering interfacial binding. Furthermore, the atomic-scale mechanism elucidates the interfacial modulation of these elements. This investigation furnishes a theoretical framework for the compositional design of SiCp/Al composites.