Accident diagnosis of a PWR fuel pin during unprotected loss of flow accident with support vector machine

Abstract

In this study, we conducted various flow rate change simulations for a fuel pin during unprotected LOFA using MARS. The obtained transient outlet temperature profiles were used to establish a relationship with peak fuel temperatures and flow rate changes, using Support Vector Machine (SVM). Unless the number of training data is scarce, the SVM trained with the core outlet temperature gives an accurate prediction (R2 > 0.9) for peak cladding surface temperature, and mass flow rate changes in the early phase of LOFA transience (~0.5 s). It illuminates that key accident characteristics are well reflected in the early response of reactor core behavior (i.e., core outlet temperature). This implies that the possibility of (1) realizing an accident diagnosis framework different from today's practice which relies on the accumulated response of reactor behavior over an extended accident progression, and (2) providing an effective guideline for accident mitigation strategies in the early phase of accident progression. The high predictability (i.e., R2 > 0.9) presented in the early phase of unprotected LOFA indicates core outlet temperature is strongly correlated to both flow rate change, and peak cladding surface temperature during the entire transience. With these strong correlations between different physical parameters, the traditional boundaries of physical locations and physical quantities in detecting accident response and progression may be reduced, allowing the possibility of interdependent detector systems.