Application of Neural Network to Predict Slug Liquid Holdup of Two Phase Flow in Horizontal Pipes

An artificial neural network (ANN) model for calculation of slug two-phase flow liquid hold-up in a horizontal pipe is developed based on 120 experimental data sets with a three-layer backpropagation network. The data sets are gathered from four compatible separate sources. Superficial gas and liquid velocities, pipe diameter, liquid density and viscosity are used as model inputs with liquid holdup as output. Data were divided into three portions of training, validation, and testing with 84 experimental data points presented to the network in the training phase, 18 used as testing data and the rest left for the validation phase. The model results correlates well with the experimental data at the testing phase with a root mean square error (RMSE) of 0.012 and a correlation coefficient (R) of 0.9993. The model with an overall RMSE of 0.019 and R of 0.996 is more accurate than other empirical and mechanistic model results thus far reported. The validated model outputs are compared to four other correlations and mechanistic models with results demonstrating the more accuracy and predictive power of the presented model. The effect of some network parameters including the type of transfer function, the percentage of data allocated to the training phase and the number of hidden nodes, on the network performance is also studied.