Paper presented at the Kuwait Oil & Gas Show, Kuwait City, Kuwait, February 2026.
Paper Number: SPE-230861-MS.

Preventing stuck pipe incidents during well construction depends on accurately identifying their underlying causes. The conventional method for identifying the root cause of stuck pipe incidents relies on a worksheet completed manually by the rig crew, based on their assessment of restrictions in pipe axial movement, rotation, and circulation before and after the pipe becomes stuck. This manual process is error-prone, can lead to delays, and may hinder effective recovery. The goal of this paper is to automate the diagnosis of stuck pipe root causes using a machine learning model.

The approach begins by analyzing hook load and surface torque signatures to detect time periods with elevated stuck pipe risk. When the risk exceeds a pre-defined threshold, data from the preceding hours/days are evaluated in real time to quantify the frequency and severity of events such as pack-offs, tight spots, and high breakover torque and drag. This analysis is integrated with pattern recognition of pipe axial movement, rotation, and circulation restrictions / actions, both before and after the high-risk event, using a Bayesian network to enable a more accurate determination of the root cause.

The algorithm was validated using datasets from over hundreds of stuck pipe incidents across a wide range of geographical locations, including both offshore and land drilling operations. It accurately identified the root cause, whether due to pack-off/bridging, differential sticking, or wellbore geometry, in over 95% of the cases. The use of a Bayesian network improves the model’s explainability, fostering greater user trust and adoption. Additionally, Bayesian networks offer the flexibility to incorporate uncertainty, leverage prior knowledge, and support the integration of new features over time.

The algorithm has been integrated into a real-time advisory system and successfully deployed at the rig site, on over 50 rigs, enabling timely corrective actions to prevent stuck pipe incidents. In several cases, the algorithm alerted a high stuck pipe risk more than 12 hours in advance of the stuck pipe event, allowing for proactive measures to avoid it altogether. To support recovery efforts, process charts were also provided to the rig crew, guiding response actions based on the root cause diagnosis provided by the advisory system.