Bi-directional prediction of structural characteristics and effective thermal conductivities of composite fuels through learning from finite element simulation results

Uranium dioxide (UO2) is widely used in nuclear reactors. This fuel has a low thermal conductivity (TC). Increasing its TC can effectively enhance the safety of reactors and fuel efficiencies. A prevalent approach to increasing the TC of UO2 is to inject a second phase material with a high TC into a UO2 matrix. Due to operational difficulties in the fabrication, deployment, and testing of such composite fuels, measurement data regarding effective thermal conductivity (ETC) of these composite fuels are rarely available, which hinders the development of these composites. To overcome such a barrier, finite element method is utilized to generate massive simulated measurements over the concerned composites. Subsequently, a novel algorithmic method is developed that automatically learns from gathered simulation results to accurately and reliably: 1) predict the ETC of a composite fuel according to its given structural characteristics, and 2) reversely infer the structural characteristics of a composite fuel from its expected ETC. The relative error of forward prediction and inverse design is