During deep-water drilling, the pressure window between pore pressure and leakage pressure is narrow, leading to frequent gas kick incidents. Loss of control can lead to severe gas kick or well blowout, resulting in incalculable losses. Therefore, the early detection of gas kick to allow for efficient well control strategies has been a focus of research in recent years. This paper presents a machine learning-based approach and research framework for early warning of gas kick during deep-water drilling. This approach is also applicable to handle other complex downhole incidents.


Safety is the lifeline of deep-water drilling, and blowout is the greatest threat to safety. Gas kick is a precursor to blowout and may lead to disasters like blowout explosion in the Gulf of Mexico (Pranesh et al., 2017). This paper aims to develop early warning systems for gas kick risks in deep-water drilling operations, which is a valuable research endeavor in preventing potential accidents. It serves as an essential barrier to safeguard deep-water drilling operations. It has significant implications for enhancing emergency management and on-site emergency response in deep-water drilling.

Drilling is an essential aspect of deepwater oil and gas exploration and development. However, the narrow pressure window available during deep-water drilling increases the risk of gas kick and blowout incidents. As natural gas circulates from the well to the surface through the drilling fluid, the pressure in the wellbore drops, and the volume of natural gas expands, causing the amount of drilling fluid discharged to increase as it approaches the surface. If gas kick is not discovered and controlled promptly, it can lead to serious safety and environmental risks, resulting in significant economic losses and even blowout. The Deepwater Horizon blowout was a shocking reminder of the importance of early detection and control of gas kick in the narrow pressure window of deep-water drilling. Currently, two significant challenges are faced:

Challenge 1: Traditional deep-water drilling gas kick monitoring is prone to lag and inaccuracies, resulting in poor quality data sample sets. For example, in the case of LW22-1-1, the deepest ultra-deepwater well in the Pacific with a water depth of 2616.30m, the use of 21-inch (outside diameter) risers results in a volume of 530.28m3. This leads to prolonged cycle delays of up to 90 minutes, causing significant monitoring lag. Furthermore, strong wind (25.70m/s), waves (4.50m), and currents (1.03m/s) lead to fluctuations in semi-submersible platform elevation. These fluctuations also cause sedimentation in the outlet pipeline/mud pit. As a result, the monitoring data is inaccurate, and the sample set quality is poor. The lag in monitoring and lack of early warning leads to frequent deepwater gas kick incidents, resulting in wellbore failure.

This content is only available via PDF.
You can access this article if you purchase or spend a download.