Design of LEU Fuel Assembly Using Artificial Neural Network at Kyoto University Critical Assembly

For the effective core experiments, the combination of the fuels and moderator plates should be optimized for each experimental purpose as well as reaching the criticality. Generally, these cores have been designed by human experiences; therefore, it requires huge human resources and times for design each core. Also, it is considerably difficult to deduct some innovative designs based on the human knowledges because there are lots of variables in the core design. As a first step for developing an automatic design method of the reactor core, in this study, a program based on ANN for designing the fuel assembly is developed to obtain highest multiplication factor with using small number of the fuel plates. In this study, an automatic design method of the fuel assembly at KUCA using ANN was proposed for the U10Mo LEU fuel. To efficiently conduct the machine learning of ANN without previous big data, a method for conducting the machine learning with automatically generating and updating the big data was developed. With the methods based on ANN, the design of fuel assemblies was performed to obtain maximum multiplication factor. The fuel assemblies designed by the proposed method considerably showed high performance for increasing the multiplication factor comparing to the fuel assemblies used in previous studies.