Determining Characteristics of Two-Phase Oil-Water Flows by the Convolutional Neural Network

To date, various mechanistic models and empirical correlations have been developed to characterize and model two phase oil-water flow systems. However, in the most of these proposed models and correlations, simplified assumptions with the iterative solutions approach have been utilized, which do not have enough accuracy to estimate the flow characteristics. The aim of this study is to overcome this problem by developing a convolutional neural network through the deep learning. For this purpose, 270 flow tests including dispersed water-in-oil, dual continuous and dispersed oil-in-water flow tests have been conducted in the both horizontal and inclined (30o) states. The neural network was trained on 70% of the achieved laboratory data. It is necessary to explain that two-dimensional flow pattern images were used as the input data and flow patterns and liquid holdup fraction values were applied as the output data. The results of this study revealed that the applied flow convolutional neural network model is able to predict the flow regimes with 91% and 96% accuracies in the horizontal and inclined flows, respectively. This model is also able to predict the liquid holdup fraction with a reasonable error of 1.22% and 0.98% in horizontal and inclined flows, respectively. Therefore, it can be concluded that the proposed approach is able to automatically and accurately predict the flow regimes and liquid holdup fractions through flow images in the both horizontal and inclined states.