Artificial Neural Network Approach to the Prediction of Drift Velocity of Elongated Bubble in Liquid in Pipe

Abstract

Abstract Accurate determination of drift velocity is required, as drift velocity is an important parameter in multiphase flow modelling. There are existing correlations for determining this parameter, but some of these models contain some degree of errors in their predictions. Recently, Machine Learning has been applied in the study of fluid flow both in porous media and pipelines. In this work, Artificial Neural Network (ANN) is applied in the development of drift velocity closure relationship for multiphase flow models. The model is trained using some input parameters: pipe internal diameter, pipe angle, liquid density, liquid viscosity, and surface tension; the output parameter is the Froude number, which is a dimensionless variable for the expression of drift velocity. The developed neural network model was compared with existing drift velocity correlations available in open literature. In general, The ANN model, with a larger set of the predicted data below the ±20% error bandwidth and a narrower percentage error spread, performed better than some of the correlations investigated in this study.