Design and Optimization of Thermal Conductivity of UO₂ Composite Fuel Based on Finite Element Simulation and Machine Learning

Biaojie Yan^1*, Pengcheng Zhang¹, Xiangdong Ding², Cheng Liang¹, Bingqing Li¹

¹ Institute of Materials, China Academy of Engineering Physics, Jiangyou, Sichuan 621700, China；

² Xi'an Jiaotong University, Xi'an 710049, China

ABSTRACT: The research and development (R&D) of UO₂ composite fuel has the characteristics of long cycle, high cost, toxic and radioactive, in which the traditional trial-and-error method applied is inefficient. Machine learning (ML) can greatly shorten and reduce the cycle and cost of the R&D of a new material. In this work, a method combining finite element method (FEM), ML and experiment was established to realize rapid R&D of the new UO₂ composite fuel, whose details are as follow. The variation ranges of FEM input parameters are determined according to the existing experimental and theoretical data. A certain number of parameter combinations are obtained by proper sampling methods within setting range, which are as the input of FEM modeling and analysis. Then the set of FEM samples including predictive features and target feature is generated. The features appropriate to predict target feature are selected by feature selection methods, the mapping from predictive features to target feature is constructed using proper machine learning methods, which are verified by FEM and experimental samples. Combined with the proper optimization methods, the reverse design from the target property to the predictive features is realized. The main results obtained are as follow. Firstly, two predictive models for predicting TC of composite fuels with dispersed and continuous second-phase were established and verified respectively. Based on the models, the bi-directional prediction of structural characteristics and TC of UO₂-SiC composite was realized with the relative error less than 5%; meantime, and a microcrack-free UO₂-2Mo (vol%) composite pellet was successfully optimized and prepared, whose TC is about 20% higher than that of pure UO₂. Lastly, the predictive models of TC and thermal stress of UO₂-Mo-Nb multi-component composite were established and verified. The optimized composition of UO₂-3Mo-1Nb (vol%) was obtained by multi-objective optimization, and the balance between TC and thermal stress of UO₂-Mo-Nb composite fuel was realized.

Keywords: UO₂ composite fuel; Machine learning; Finite element method; Thermal conductivity