Machine Learning Approach to Predict Pressure Drop of Multi-Phase Flow in Horizontal Pipe and Influence of Fluid Properties

Abstract

Multi-phase flow is very common in different applications and industries. In the petroleum industry, multi-phase flow can be observed in different parts of production systems such as tubing of vertical or horizontal wells, flowlines, and surface facilities as well as in the pipeline for exports& transportation of the oil and gas to the refineries. The prediction of the pressure drop is imperative for designing as well as the operation and maintenance of the production system. There are several experimental, theoretical modeling and numerical analyses were carried out to predict the pressure drop of multi-phase flow. The complex interactions of the different phases lead to different flow regimes which are essential for developing the computational model of the pressure drop. Machine learning is a promising approach that can address such complex problems. The objective of this study is to build an Artificial Intelligence (AI) model using dimensionless parameters to estimate the pressure drop of two-phase flow in a horizontal pipe and the influence of fluid properties. To achieve the objective of this study, a large set of experimental data was collected which was used to develop the AI model to predict the pressure drop of multi-phase flow in a horizontal pipe. The effect of fluid properties was investigated by changing the liquid properties (density, viscosity, and surface tension). The data was collected by flowing the two-phase air/liquid system on the flow loop with a pipe diameter of 1 inch (2.54 cm) and a length of 30 ft (9.15m). The surface tension was varied using the surfactant solution, viscosity was varied with the aid of glycerin, and density was varied with the aid of calcium bromide. The superficial velocity of the liquid ranges from 0 to 3.048 m/s (0–10 ft/s) and the superficial gas velocity ranges from 0 to 18.288 m/s (0–60 ft/s) respectively. Machine learning was utilized to develop models that can identify the pressure drop of multi-phase flow in a horizontal pipe with the effect of fluid properties. Results showed that different AI methods can be used to predict the pressure drop of multi-phase in horizontal pipes with high accuracy with few inputs. The wide range of data was processed by applying a machine learning technique for predicting the pressure drop of multi-phase flow. The models were built using dimensionless parameters to extend their validity for various design and operational conditions. The accuracy was improved by introducing the additional dimensionless parameter for all the models. The development in the computational methods emerges a new area of numerical and computational fluid dynamics and presently investigators are exploring the application of AI in resolving complex phenomena such as multi-phase flow. The complex interactions of the different phases lead to different flow patterns, which are essential elements during the development of the computational model of pressure drop.