Road map to develop an artificial neural network to predict two-phase flow regime in inclined pipes

Abstract

In this research, the utilization of data-driven approaches to predict the flow pattern of a two-phase flow in a horizontal, inclined, and vertical pipe (−90 to 90°) is investigated. Although multiphase flows have been modeled using artificial neural network models, there is still a long way to draw a road map to optimize these models. To fill this gap, an artificial neural network (ANN) was applied using 8766 experimental samples where 70% of the data was used to train the models, 15% was used for cross-validation, and 15% was used for testing. The generated neural network considers the flow pattern as the output that is predicted using 10 inputs which are flow pressure, liquid surface tension, inclination angle, density, viscosity, and superficial velocity of both gas and liquid streams plus the pipe diameter. Also, superficial liquid and gas Reynold's numbers, mixture Froud number, and Weber number are used to replace the dimensional inputs. It is to be noted that the neural network performance increased when the model is considered a classification problem. The results indicate that using a multi-layer perceptron artificial neural network has high accuracy in predicting the flow regime, demonstrating a high accuracy of 97.3% in predicting the flow pattern. Moreover, the dimensionless-based model's accuracy was demonstrated to be inferior in forecasting the flow regime compared to the dimensional-based model in terms of accuracy. Furthermore, the accuracy of the generated neural network was also validated against Barnea's model which showed an incorrect classification percentage of 1.7%.