Detecting Emulsion Using Surface Temperature, Pressure, and the Application of Artificial Intelligence

Innovation in the analysis of oil well surface measurements has led to the discovery of an instantaneous and straightforward emulsion detection calculation. When applied in the Bahrain Field, this led to the treatment of emulsion in over 100 wells, resulting in a cumulative production gain of over 500,000 barrels to date at negligible cost. Artificial Intelligence (AI) was then employed to identify and understand factors related to emulsion and optimisation treatment programs. Once the wells were treated and the method was confirmed to prove emulsion existence, a focused approach was carried out to understand it further. Wells were categorised based on their production response to standard demulsifier bullheading. In addition to a variety of well parameters, this data was used to build a machine learning model that helped identify patterns with regards to problematic zones, properties of wells with emulsion, and the best treatment method for each well. The results of the study were rather substantial and resulted in numerous new insights. Firstly, a model was built to predict the sustainability and economics of expected bullheading job treatments. This is currently being used to rank the priority of wells for either bullheading treatment or continuous chemical injection. Once the wells were classified into basic sub groups and sorted by zones, geographic analysis was carried out to identify wells with emulsion being formed as a result of waterflooding. This led to further insight into the nature of emulsion blocks, where in some cases, although it was found that these blocks exist downhole, traces of emulsion will flow to the surface and can have a unique signature. This paper discusses in further detail insights into emulsion and the different types of AI algorithms used to answer questions raised as a result of the discovery. The necessity of using machine learning cannot be overstated enough and the observations made in the paper could not have been found if it were purely by observed by the naked eye. The topic of emulsion is highly understudied, and the concept of using the emulsion detection calculation was not published before. In addition to highlighting this discovery, this paper can influence other operators to test their findings and have a real world application of machine learning in their fields.