LAUSR

Abstract In order to identify and quantify reactor coolant leaks in nuclear power plants, it is important to understand and model the leak phenomena of the fluid. When reactor coolant leaks from a high pressure and temperature reactor coolant system, two-phase mixture or steam is jetted out by the flashing phenomenon and has critical flow characteristics. Therefore, critical flow analysis is needed to understand and quantify the leakage phenomena. In this study, simplified cascade fuzzy neural networks (SCFNNs) based on the artificial intelligence methods are developed to predict critical flow. The SCFNNs consist of syllogistic fuzzy reasoning and are the simplified model of the existing cascade fuzzy neural network (CFNN) model in terms of not using the outputs of all the preceding fuzzy neural network (FNN) modules but only the output of the very preceding FNN module. The SCFNN model is developed to predict the critical flow for the critical mass flux and critical pressure while repeating a stepwise fuzzy neural network. The proposed SCFNN model uses the data obtained from the Henry–Fauske model to understand the fluid properties and predict the critical flow. The SCFNN method is approximately 17 times more accurate than a previous correlation method and over 20 times more accurate than the FNN and support vector regression models. It is expected that effective information will be provided to quantify the reactor coolant leak from the SCFNN model. Also, the developed SCFNN model can be used as a universal standalone program to predict critical flow, which runs faster without iterative calculations, and without a need for steam tables.