Advanced Fluid Influx Detection in Drilling Wells Using Real-Time Data and Probabilistic Uncertainty Analysis

Abstract This study presents a real-time method for detecting fluid influx during well drilling. Fluid influx, caused by intrusion of formation fluid into the wellbore, builds pressure differences between the formation and the wellbore. Wellbore experiences "kick" as a consequence of such influx. Not all kicks are necessarily dangerous, however, in extreme cases these can lead to catastrophic failures if not identified and addressed promptly. The goal is to improve wellbore safety by identifying fluid influx events and providing accurate early assessments in real time. The proposed method analyzes pressure data collected from distributed sensors network along the wellbore and transmitted to the surface by fast telemetry such as wired drill pipe (WDP). It utilizes both physics and data science techniques to preprocess, formulate features, analyze the data, and identify potential fluid influx events. A critical step in the process is uncertainty quantification, which estimates the probability of true fluid influx events by considering wellbore configurations, including sensor placements and flow rate of well fluids. This probabilistic approach refines the detection process, enabling more accurate and reliable real-time identification of fluid influx events. The effectiveness of the proposed method was confirmed using actual wellbore data where multiple influx incidents were simulated. The data were concurrently gathered through eleven pressure sensors placed along the pipe string, arranged sequentially from 100 ft to 10,000 ft away from the drill bit, to monitor the pressure changes over time. In the results, the proposed method identified all pressure changes, considered the most probable fluid influx, as analyzed and suggested by human wellbore experts. The probability of a fluid influx event is dynamically updated by analyzing the expected propagation sequence, event-time intervals between sensors, and other sources of uncertainty during the operation. Suspicious pressure transmission events could be detected before they reach 5,000 feet downhole, and reported to the operator, providing enough time to initiate mitigation and well control plans. The combination of probabilistic modeling and real-time monitoring provides a significant improvement over existing methods, enhancing wellbore safety and reducing the risk of catastrophic failures.