High-Precision Constitutive Modeling of CMC Interphase Under Thermo-Chemo-Mechanical Conditions Based on Molecular Simulation and Machine Learning

Abstract

Ceramic matrix composite (CMC) is emerging as a leading candidate for next-generation aeronautical materials. While ceramics are brittle, CMCs demonstrate improved toughness thanks to the matrix-fiber interphase, which deflects crack propagation. To date, accurately predicting the mechanical behavior of the CMC interphase under complex thermo-chemo-mechanical conditions remains a major challenge. In this context, we introduce an AI-based generative framework that directly generates highly accurate strain–stress relations for the CMC interphase based on measurements of temperature, oxidation state, and strain rate. The model combines an unsupervised autoencoder, which learns the key features of the strain–stress relation, with a multilayer feed-forward neural network that maps loading conditions to these features. Pre-trained by extensive molecular dynamics simulations and calibrated with minimal experimental data, the model is thoroughly validated through push-in tests of single-fiber composites and tensile tests of unidirectional fiber-bundle composites, demonstrating satisfactory accuracy. The primary application of this AI-based method is to evaluate the mechanical performance of the CMC interphase directly from easily measurable loading conditions, bypassing the need for microstructure. This approach offers an efficient solution for load design and health monitoring of ceramic matrix composite structures.

Graphical Abstract