Drilling fluid (mud) serves various purposes in drilling operations, the most important being the primary well control barrier to prevent kicks and blowouts. Other duties include, but not limited to, maintaining wellbore stability, removing and transporting formation cuttings to the surface, cooling and lubricating downhole tools, and transmitting hydraulic energy to the drill bit. Mud quality is therefore related to most of the problems in drilling operations either directly or indirectly.
The physics-based models used in the industry with drilling fluid information (i.e., cuttings transport, well hydraulics, event detection) are computationally expensive, difficult to integrate for real-time analysis, and not always applicable for all drilling conditions. For this reason, researchers have shown extensive interest in machine learning (ML) approaches to alleviate their fluid-related problems.
In this study, a comprehensive review of the abundant literature on the various applications of ML in oil and gas operations, concentrating mainly on drilling fluids, is presented. It was shown that leveraging state-of-the-art supervised and unsupervised ML methods can help predict or eliminate most fluid-related issues in drilling. The review discusses various ML methods, their theory, applications, limitations, and achievements.