Experimental Investigation and Prediction of Droplet Entrainment Fraction in Horizontal Annular Flow

Droplet entrainment fraction significantly affects pressure drop and heat transfer in annular flow. In a 50 mm horizontal pipe, the entrained droplet parameters, including droplet fraction, velocity, size, and mass flux, are measured by an optical fiber probe, revealing the characteristic distribution and flow mechanism of droplets under gravity. Next, the entrainment fraction is derived by integrating the exponential distribution of local droplet mass flux. Then, a database of entrainment fractions in horizontal annular flow is established to evaluate four recognized entrainment models. These models exhibit significant scatter and uncertainties in their predictions and are often limited to specific flow regimes. In order to improve the accuracy and applicability of the model, a neural network is established to predict the droplet entrainment fraction, and the optimal hyperparameters of the network are determined by the Bayes optimization method. Compared to existing models, the new model significantly improves the prediction accuracy of droplet entrainment fraction under different conditions. The determination coefficient $R^{2}$ reaches 0.954, with over 82.5% of prediction errors within ±30%, showing satisfactory agreement with experiments.