Harness AI and machine learning in de-emulsifier chemical selection

Abstract

The most common method of preventing the formation of emulsions in the petroleum industry is by the application of a de-emulsifier chemical. The standard approach of selecting an appropriate de-emulsifying agent is by scanning different chemistries with changing properties to identify the emulsion breaking region. However, the disadvantage is that these tests are time-consuming. This work presents a faster alternative for choosing de-emulsifier chemicals by using machine learning. For data to train and test machine learning models, several bottle tests are analyzed at different combination of essential parameters. For both non-EOR/normal and EOR-induced emulsion, the models are validated with real crude oil to output a list of de-emulsifiers that work in breaking the emulsion and rank them for success based on probability. An application workflow of this de-emulsifier prediction tool is also created for deploying the model to provide guideline for quick de-emulsifier chemical selection.