Optimization of thermal conductivity of UO2–Mo composite with continuous Mo channel based on finite element method and machine learning

Abstract Uranium dioxide (UO2) is widely used in nuclear reactors. Low thermal conductivity (TC) is one of its greatest disadvantages. Improving the TC of the UO2 pellet has great significance to the safety and economy of the reactors. Introducing Mo with a continuous distribution into the UO2 matrix can obviously improve the TC of the UO2 pellet. However, considering uranium density, burn-up and postprocessing of the composite pellet, which determine the economy of the nuclear fuel, it is anticipated that the addition of the second phase is as little as possible for the UO2 composite, which is key to further improving the TC of the UO2-Mo composite with a low and constant content of Mo. This target can be achieved by optimizing the microstructure of the composite. In this work, a new method similar to ‘PixelMapPaint’ was introduced for modelling the UO2-Mo composite with a complex microstructure. The effects of various structural characteristics on the TC of the UO2-Mo composite were quickly analysed via the finite element method and machine learning methods. Guided by the analysis results, the actual UO2-Mo pellet with 2 vol% Mo was fabricated with measured TC higher by approximately 20% than that of pure UO2. Additionally, the machine learning model was further developed from isotropic to anisotropic composites, with less than 10% relative error between predicted and simulated TCs in different directions.