Predicting gas volume fraction and flow regime identification in two-phase water–air flow: A deep learning solution for resource preservation in oil and gas industries

Abstract

This study investigates the integration of gamma-ray attenuation techniques and artificial neural networks for the precise identification of flow regimes and prediction of gas volume fractions in two-phase flow. Experimental procedures were conducted within a carefully designed two-phase flow loop, which integrated gamma-ray attenuation techniques with sophisticated data acquisition systems. An in-depth analysis was undertaken utilizing convolutional neural networks, a subtype of artificial neural networks, to analyze detailed patterns in the data and predict gas volume fraction alongside identifying flow regimes. The convolutional neural network model was precisely trained and optimized to handle the complexities inherent in multiphase flow dynamics. The results demonstrated the stability and efficacy of the convolutional neural network model in accurately predicting gas volume fraction and evaluating flow regimes, while also exploring the relationship between radiation measurement techniques and advanced machine learning methods. This comprehensive approach not only advances the current understanding of multiphase flow dynamics but also offers practical solutions for enhancing measurement accuracy and efficiency in industrial applications.