Prediction of LBB leakage for various conditions by genetic neural network and genetic algorithms

Abstract

In this study, three-layer Back Propagation Network (BPN) and Genetic Neural Network (GNN) were applied to predict the leakage of Leak Before Break (LBB) for various conditions. The inputs include six dimensionless variables, and the Reynolds number is set as the output. The GNN (with relative error of 22.7%) shows a higher accuracy than the BPN (with relative error of 26.1%), the existing models and commercial software. Influences of thermal–hydraulic properties and crack morphologies on LBB leakage were discussed based on the trained GNN: the LBB leakage is proportional to the Crack Opening Displacement (COD), crack length, subcooling degree, stagnation pressure and the area ratio of inlet to outlet, while it is inversely proportional to crack depth and local roughness. Moreover, mechanism-based correlations for LBB leakage were proposed by genetic algorithm in this study. The flow resistance due to phase transition and area variation was considered, and the entrance resistance coefficient, friction resistance factor and plugging were presented. The presented correlations provide higher precision than the existing correlation, with average error of 35.9%. The proposed correlations are meaningful for LBB leakage estimation.