Machine Learning Leads Cost Effective Intelligent Fluid Design: Fluid Engineering Perspective

Abstract

Abstract What is the catch for intelligent fluid design? Several hours of laboratory testing (some successful, some unsuccessful), dealing with various complex chemistries, listening to multiple people, as long as they are around, some of whom educated with the science called Chemistry rationalizing and trying to match the theories with the laboratory results? Getting superstitious about the physics of the problem due to lack of data? Dealing with contradictory data, and convincing ourselves, engineers, that this is the way to go? Seeing laboratory personnel taking pride on the number of the tests that have conducted while we still need to conduct more testing to determine the next additive concentration for the new well with different downhole condition? Facing with enormous non-productive time and dealing with the fear of failing to deliver to the job and lose the tender, on the service side, or choose the most cost effective scenario for your well, on the operator side? Or shall we go with what Joe proposed and reduce the risk? He has been there for years. How much more testing is enough for a known system, area or formation? Chemists are creating better and novel formulations, and laboratory personnel are doing their job following the API, ideally around the world, but fluid engineers are still testing or manage testing the fluids by trial and error whether in fracturing, cementing or drilling fluids. While designing fluids by trial and error in each fluid research group might sound unrelated to the other, all have one in common because of the traditional approach in fluid design. Trial and error testing lacks the vision for sustainable fluid knowledge management and intelligent design. The purpose of this paper is to challenge the status quo in fluid design, discuss what is missing in stepping towards intelligent fluid design and identify the challenges and limitations to execute the proposed method. Disadvantages of the traditional fluid designs, lack of knowledge/data management or sharing and the cost incurred are elaborated. Machine learning algorithms lead as the prime solution for intelligent fluid design, saving thousands of hours of laboratory testing, efficiently integrating sizable experimental database and optimizing the operational cost and delivery.