A Machine Learning Approach for Accelerated Fission Product Speciation and Material Property Evolution in Nuclear Fuel

Nuclear fuel performance relies on understanding the evolution of fuel properties under operational conditions. This is a complex task; during fission, the fuel undergoes chemical changes as well as severe radiation damage, making the evolution of properties a complex function of many parameters. Historically, property evolution has been determined via empirical data from irradiated materials. These empirical correlations have limited applicability beyond the conditions under which the underlying measurements were obtained. Extrapolation to broader operating windows, fuel compositions, and geometries requires repeating the irradiation experiments and post-irradaition measurements to derive a new function.
In the present work, a new approach is explored to address this problem. First, machine learning is used to analyze large datasets from quantum materials databases to develop a model for predicting formation enthalpy. This allows for a quick evaluation of fission product speciation in diverse fuel compounds. The same model provides trends for each fission product regarding its relevance in shifting fuel properties compared to pure compounds. The obtained trends can be corroborated by designed separate effects samples in accordance with Accelerated Fuel Qualification (AFQ) approach. Finally, the correlations obtained, i.e., elemental fraction, elemental statistics, structural, and ionic configuration, can be used in the process of cross-property transfer learning to derive shifts in engineering material properties such as thermal conductivity.