The accurate prediction and further optimization of thermal conductivity for 3D fully ceramic microencapsulated fuel via graph convolutional neural network

Fully ceramic microencapsulated (FCM) is a new type of composite fuel in the nuclear energy field; it has emerged as a promising candidate for accident-tolerant fuel (ATF), owing to its good structural stability and retention capability of fission products. However, the inherently low thermal conductivity of FCM fuel poses a challenge and increases the risk of accidents in reactors. It is well-accepted that the microstructures of the fuel affect their physical properties. Thus, establishing the correlation between the 3D microstructures and corresponding properties of the fuel is crucial for improving their properties by optimizing the microstructure. In this study, we investigate the relationship between the 3D microstructures of FCM fuel and their corresponding effective thermal conductivity (ETC) by using the graph convolutional network (GCN) model. A dataset comprising 100,000 3D microstructures, generated through high-throughput finite element method simulations, was employed in the study. By transforming the 3D microstructure into graph data, the GCN model exhibited superior performance in predicting the ETC of 3D FCM fuel, with an error of 0.2 %. Furthermore, a microstructural feature influencing the ETC of the FCM fuel was identified using a two-point correlation function. This new microstructural feature contributes to the design of microstructures with enhanced ETC values and the improved performance of the GCN model in the present task, indicating its essential role in determining the ETC of FCM fuel. This work demonstrates the excellent performance of the GCN model in establishing microstructure–property correlations for 3D composites.