Development of an Accelerated Nuclear Fuel Qualification Framework

Nuclear fuel qualification is one of the major challenges towards deployment of advanced nuclear reactors. The conventional qualification approaches require testing using unique reactor and hot cell facilities, which cannot handle large volumes of testing of various potential fuel compositions. Hence this process is extremely time and cost sensitive, requiring prioritization criteria. Here we propose an accelerated fuel qualification process by combining machine learning techniques with advanced nuclear fuel characterization experiments. More specifically, doped variants of a potential nuclear fuel U(X)N are fabricated via a solution-gel (sol-gel) method. The solid fission product dopants (X) being considered are Zr, Hf, and Ti. Multiple concentration levels are generated ranging from X = 1at% to 2at% in a way that allows for correlations to be established for individual fission products for each fuel compound. The doped composites are subjected to standard characterization techniques, e.g. scanning electron microscopy (SEM). The SEM microstructure data is input to a machine learning framework, namely Generative Adversarial Networks (GANs). Starting from the experimental SEM images, GANs are used to generate the synthetic microstructure data. The image processing techniques are employed to determine the residual porosity and grain size change as a function of composition of elements (X). This image processing analysis provides connections between the composition of these elements and the diffusion in UN fuels. By harnessing machine learning, we are able to understand Post Irradiation Examination (pie) microstructures of fuels, which is a significant step towards nuclear fuel qualification processes for diverse reactor designs and nuclear fuel composites.