Machine learning technique for the identification of two-phase (oilwater) flow patterns through pipelines

Daniel Yesid Uribe-Tarazona¹Carlos Mauricio Ruiz-Diaz²Octavio Andrés González-Estrada^1*

• ¹Industrial University of Santander, Bucaramanga, Colombia
• ²University of São Paulo, São Carlos, São Carlos, São Paulo, Brazil

Accurate prediction of flow patterns in oil-water two-phase flow within horizontal pipelines is essential for ensuring operational efficiency, integrity, and cost-effective design in the oil and gas industry. This study develops a robust machine learning model based on artificial neural networks to classify six flow patterns: stratified, stratified with mixture, slug, annular, oil-in-water dispersion, and water-in-oil dispersion. A comprehensive database comprising 1,846 experimental data points was assembled from the literature, encompassing a wide range of operating conditions, including fluid properties, superficial velocities, and pipe diameters. After evaluating 104 network configurations, the optimal model achieved an overall accuracy of 95.4%, with training, validation, and testing accuracies of 97.1%, 92.8%, and 90.3%, respectively, and a cross-entropy error of 0.024. The model demonstrates rapid convergence with a training time of only two seconds, offering a reliable and computationally efficient tool for flow pattern recognition. The outcomes of this study provide significant value for improving pipeline design, optimizing flow assurance strategies, enhancing corrosion control, and supporting real-time operational decision-making in multiphase transport systems in the oil and gas sector.