Machine learning sintering density prediction model for MOX fuel pellet

Kato, Masato   ; Nakamichi, Shinya  ; Hirooka, Shun ; Watanabe, Masashi  ; Murakami, Tatsutoshi ; Ishii, Katsunori 

Uranium and Plutonium mixed oxide (MOX) pellets used as fast reactor fuels have been produced from several raw materials by mechanical blending method through processes of ball milling, additive blending, granulation, pressing, sintering and so on. It is essential to control the pellet density which is one of the important fuel specifications, but it is difficult to understand relationships among many parameters in the production. Database for MOX production was prepared from production results in Japan, and input data of eighteen types were chosen from production process and made a data set. Machine learning model to predict sintered density of MOX pellet was derived by gradient boosting regressor, and represented the measured sintered density with coefficient of determination of R=0.996