HYBRID MACHINE LEARNING MODEL APPLIED TO PHASE INVERSION PREDICTION IN LIQUID-LIQUID PIPE FLOW

Abstract

One of the current challenges in two-phase flow is the characterization of phase inversion in the oil and gas industry. Empirical and semi-empirical models have been developed by several researchers, allowing limited predictions through correlations. Recently, models obtained with application of artificial intelligence techniques, such as artificial neural networks, have become a promising alternative to identify flow patterns and their transition boundaries. This work's aim is to develop a hybrid model that identifies the phase inversion transition from oil-in-water to water-in-oil flow in vertical pipes. It is based on recent models found in the literature and logistic regression models based on artificial neural networks, for which information was obtained from the literature. The proposed hybrid model achieved an RMSE ≈ 0.0834, thus being an efficient contribution to the identification of phase inversion in oil-water two-phase flow.