This study used an artificial neural network (ANN) regression model in wire-wrapped fuel assemblies to estimate the transition-to-turbulence flow regime boundary (RebT) and friction factor. The ANN models were trained and validated using existing experimental datasets. The bundle dataset comprised several design parameters, such as the number of rods, rod diameter, wire diameter, lattice pitch, edge pitch, and wire helical pitch. The log–log scale Reynolds number and linearity characteristics of the friction coefficient were used to over-sample the friction factor in the laminar and turbulent regimes for resolving the data imbalance. Three-quarters of the entire dataset was used for training, while the remainder was used for validation. The Levenberg–Marquardt approach with the Gauss–Newton approximation for the Hessian of the training cost function was used for training the model. The number of hidden layers for RebT was selected based on the minimum validation error. The pin number effect was additionally considered for the friction factor while selecting the number of hidden layers. The ANN model predicted using the oversampled data set had a 50% reduction in the root mean square error (RMSE) than the model predicted using the original data set. Compared to previous correlations, the prediction of ANN models for the friction factor demonstrated significantly low errors (0.10% mean error and 7.36% RMSE of 142 bundle data).
Read full abstract