Abstract
The application of ceramic particle-reinforced metal matrix composites (CPRMMCs) in the nuclear power sector is primarily dependent on their mechanical and thermal properties. A comprehensive understanding of the structure-property (SP) linkages between microstructures and macroscopic properties is critical for optimizing material properties. However, traditional studies on SP linkages generally rely on experimental methods, theoretical analysis, and numerical simulations, which are often associated with high time and economic costs. To address this challenge, this study proposes a novel method based on Materials Informatics (MI), combining the finite element method (FEM), graph Fourier transform, principal component analysis (PCA), and machine learning models to establish the SP linkages between the microstructure and thermodynamic properties of CPRMMCs. Specifically, FEM is used to model the microstructures of CPRMMCs with varying particle volume fractions and sizes, and their elastic modulus, thermal conductivity, and coefficient of thermal expansion are computed. Next, the statistical features of the microstructure are captured using graph Fourier transform based on two-point spatial correlations, and PCA is applied to reduce dimensionality and extract key features. Finally, a polynomial kernel support vector regression (Poly-SVR) model optimized by Bayesian methods is employed to establish the nonlinear relationship between the microstructure and thermodynamic properties. The results show that this method can effectively predict FEM results using only 5-6 microstructure features, with the R(2) values exceeding 0.91 for the prediction of thermodynamic properties. This study provides a promising approach for accelerating the innovation and design optimization of CPRMMCs.