Automatic Classification of Three-Phase Flow Patterns of Heavy Oil in a Horizontal Pipe Using Support Vector Machines

Abstract

The pipe flow of a viscous-oil-gas-water mixture such as that involved in heavy oil production is a rather complex thereto-fluid dynamical problem. Considering the complexity of three-phase flow, it is of fundamental importance the introduction of a flow pattern classification tool to obtain useful information about the flow structure. Flow patterns are important because they indicate the degree of mixing during flow and the spatial distribution of phases. In particular, the pressure drop and temperature evolution along the pipe is highly dependent on the spatial configuration of the phases. In this work we investigate the three-phase water-assisted flow patterns, i.e. those configurations where water is injected in order to reduce friction caused by the viscous oil. Phase flow rates and pressure drop data from previous laboratory experiments in a horizontal pipe are used for flow pattern identification by means of the 'support vector machine' technique (SVM).