Study of microscale heat transfer in UHTCMCs based on deep learning and finite element analysis

Abstract

The microstructure of ultra-high temperature ceramic matrix composites (UHTCMCs) is extremely complex, making it particularly challenging to conduct precise heat transfer analysis that accurately reflects the material's true structural features due to their heterogeneous multiphase characteristics. In this study, we propose a heat transfer model that combines deep learning with finite element analysis, which is used for feature extraction and heat transfer analysis of the microstructure of ultra-high temperature ceramic matrix composites, allowing for the calculation of effective thermal conductivity to predict the material's macroscopic thermal conductivity. Microstructural feature recognition is achieved by segmenting the phase structure of ultra-high temperature ceramic matrix composites using the BSE/EDS images, through the construction of the Unet deep learning model. Additionally, the structural mapping mesh method is employed to convert the actual structural information of the reactive material into a finite element mesh model for a detailed analysis of its micro-scale heat transfer characteristics. The macroscopic thermal conductivities of the materials are obtained by statistically calculating the thermal conductivities of the microscopic sections, showing consistency with the thermal conductivities from the theoretical model and the experiment. This study effectively reveals the heat transfer characteristics from the complex microstructure of UHTCMCs.