Bi-objective optimization of compressive strength and thermal conductivity for UN-U₃Si₂ composite fuel based on AI techniques

The UN-U₃Si₂ composite fuel has been developed as a promising accident-tolerant fuel (ATF) due to its superior thermal conductivity and higher uranium density compared to conventional UO₂. To further reduce accident risks, it is highly desirable to optimize the physical properties of ATF fuel, such as compressive strength (CS) and thermal conductivity (TC). Tailoring the microstructure offers an effective approach to achieving multi-objective optimization of the composite fuel, ensuring a balanced trade-off between key properties. In this study, a relationship between the microstructure of UN-U₃Si₂ composite fuel and the associated CS and TC was established via a convolutional neural network (CNN). To address the challenge of data insufficiency, a dataset of 15,000 microstructure-property pairs was generated through the high-throughput finite element method (FEM). Using reconstructed metallographic images as input, the CNN models achieved a prediction of the CS or TC within a relative error of 3 %. Moreover, critical features strongly correlated with the CS and TC of composites were identified through the saliency map method analysis and Pearson correlation coefficient (PCC) evaluation. Finally, a bi-objective optimization strategy was employed to design microstructures for UN-U₃Si₂ composite fuel pellets that effectively balance CS and TC properties. This work not only provides practical guidelines for designing advanced ATF fuels with improved performance but also introduces a robust workflow for the multi-objective optimization of composite materials with superior physical properties.