Online Detection of Anomalous Behavior in Drilling Fluids Using Intelligent Monitoring Techniques: A Comparative Analysis Between Local Outlier Factor and Principal Component Analysis Approaches

Summary Drilling fluids are essential for controlling formation pressures, transporting solids, and cooling and lubricating equipment while drilling oil and gas wells. Keeping fluid properties within defined operating limits is essential to ensure process efficiency. This study presents a real-time monitoring system using the local outlier factor (LOF), a machine learning technique that identifies outliers based on deviations in local density. The system was developed and tested in a fluid circulation unit equipped with online sensors to measure physicochemical properties of drilling fluids. Fluid disturbances have been introduced by the sudden addition of particulate matter to change the apparent viscosity, measured at 511 second-1, and the density of the fluid. The adaptive algorithm associated with LOF processed the data in real time, flagging abnormal points as possible deviations from the process, with an accuracy of over 90%. A comparison with the monitoring done via principal component analysis (PCA) allowed us to validate the approach in detecting and diagnosing failures. In addition, it was possible to observe the gain of the proposed approach for the sector because the proposed methodology combines detection, explainability, and adaptability, contributing significantly to the evolution of intelligent decision support systems in drilling operations. By supporting operators in diagnosing the causes of deviations, this integration of real-time modeling and analysis improves decision-making, reduces nonproductive time, and improves well integrity and drilling performance.