抄録
The future generation sodium-cooled fast reactor has the design strategy to achieve in-vessel retention for Core Disruptive Accident (CDA). Discharged molten-fuel from the core during CDA becomes solidified particle debris with fuel and coolant interaction in the lower sodium plenum, and then the debris forms a bed on the core catcher located at the bottom of the reactor vessel. Thus, it is important to understand the coolable conditions of the debris bed for the studies of the CDA scenarios and the design options of the mitigation devices such as a core catcher design, etc. On the other hand, in order to understand the coolable conditions, many parametric calculations are needed which envelop uncertainty of the CDA scenarios and the design options of the mitigation devices. It takes an enormous amount of time. The purposes of this study are to accelerate coolability evaluation of debris beds and to enable us to instantaneously understand the coolable conditions. With these purposes, we have studied to apply machine learning to debris bed coolability evaluations and to create an empirical model which can evaluate debris bed coolability under various parameter conditions. This paper describes the empirical model created with machine learning on the basis of the training data consisting of many coolability calculations changing main parameters such as particle diameter and porosity of debris bed and coolant temperature around debris bed, etc. These parameter sets are based on Latin hypercube sampling. The coolability calculations were performed with the existing mathematical model. We have confirmed that the empirical model created with machine learning can evaluate the debris bed coolability with sufficient accuracy and the evaluation time is quite short in the applicable range. In addition, the empirical model enables us to instantaneously understand debris bed coolable conditions which change with the combination of multiple parameter value changes. Thus, the application of machine learning to debris bed coolability evaluations contributes the discussion on the balance between the CDA scenarios and the design options of the mitigation devices.
