26 / 2024-05-31 15:44:11

Simulation of MCCI phenomenon using a neural network model trained on advanced particle method

MCCI,MPS,Artificial Neural Network,Response model

 As the accident progresses, the failure of the Reactor Pressure Vessel (RPV) may lead to the occurrence of a Molten Corium-Concrete Interaction (MCCI) phenomenon. The accident characteristics involve multiphase states, large deformations of free surfaces, changes in fluid viscosity, and thermal chemical reactions. These complexities necessitate the traditional lumped parameter method to rely on a large number of empirical formulae, resulting in relatively rough calculations. However, the mesh-based method has inherent limitations in capturing phase interfaces and free surfaces. Recently, some researchers have simulated the MCCI phenomenon using meshless particle methods. This paper carries out simulation research of MCCI accidents based on the Moving Particle Semi-Implicit (MPS) method. Under the framework of Monte Carlo theory, a large number of calculation conditions are sampled for input parameters with uncertainty, obtaining the computation results of the MPS method as training data. Response model training based on Artificial Neural Networks (ANN) is conducted with key features like ablation depth as the output layer, and uncertain input parameters and simulation time as the input layer. The ANN response model for MCCI accident results has been successfully trained, and the model's computational accuracy is verified through experimental results. This research provides a new strategy for rapid analysis of MCCI accident phenomena in nuclear reactors.