Physics-informed machine learning with high-throughput design module for evaluating rupture life and guiding design of oxide/oxide ceramic matrix composites

Abstract

In this paper, a unified physics-informed machine learning (PIML) model is proposed for the accurate prediction of the creep rupture life of oxide/oxide ceramic matrix composites (CMCs) by incorporating prior physical information and microstructure features. The collected dataset undergoes missing value imputation and standardization methods for processing. Key feature parameters that influence rupture life are identified through global sensitivity analysis. Furthermore, a high-throughput design module is developed to explore optimal composition under specific creep conditions, providing design guidelines for new materials in oxide/oxide CMCs. This study demonstrates that the predicted results consistently fall within the ±3 error bands. Improved life expectancy is attained with fiber modulus (220–290 GPa) and matrix modulus (>310 GPa). The transformation from expensive creep testing to low-cost exploration of modulus and proportional limit is achieved, offering theoretical support and analytical methods for the design of CMC components and the development of new materials for high-temperature applications.