Integrating Particle Method Simulations and Machine Learning for Predicting Core Melt Flow Pattern

In the event of severe accident of PWR nuclear power plants, the transient process of core heating and may result in the core melting process. Accurate prediction of core melt flow pattern is a key research content, because melt with different flow pattern has different flow heat transfer characteristics, and affect the subsequent melt migration process. In this paper, Incompressible Smoothed Particle Hydrodynamics (ISPH) method is used to investigate the flow pattern of core melt, which does not need to trace the free surface, and is especially suitable for simulating core melt flow. The surface tension has a great influence on the flow pattern of the melt. In order to accurately describe the surface tension, the Pairwise Force (PF) model is adopted in this paper, and the influence of the contact angle of the wall is considered. Based on this method, the effects of fracture size, melting depth and melting height on flow pattern of the core melt was studied. Then 160 numerical simulation cases were generated by Latin hypercube sampling method. The database is divided into training sets and test sets in a ratio of 4:1. Based on the flow pattern data obtained by numerical simulation, the Support Vector Machine (SVM) method is used to predict the core melt flow pattern. Results demonstrate that the proposed approach can achieve 97.5% correct prediction, machine learning could be used as an effective tool for automatic prediction of core melt flow pattern.