Similar to oil-and-gas wells, drilling risks associated with the occurrence of stuck pipe can be significant also in geothermal well construction, particularly for wells drilled at greater depths in harder rock formations. However, as shown here, such risks can be effectively mitigated by leveraging data acquired from past wells, specifically to evaluate borehole conditions in real time and anticipate potential incidents on new wells. This work presents a multi-agent sticking prediction model, primarily constructed using data from the Utah FORGE geothermal wells, which can be applied for real-time sticking prevention in future wells. Our proposed model consists of six agents that detect sticking signatures in real time. Each agent operates within a specific variable space designed to clearly differentiate between anomalous data, indicative of stuck pipe signatures, and normal data. These variables combine statistical characteristics, such as the kurtosis and skewness of sensor signals, with physics-based modeling, for example, of drill string friction and cuttings transport. By combining these variables, the agents capture early signs of the primary mechanisms leading to sticking. Ultimately, the agents’ predictions are integrated to inform the user, in real time, about the presence, severity, and potential causes of sticking risks. Such information, when presented in a timely fashion, then allows effective mitigation action to be taken The proposed multi-agent model incorporates four agents related to sticking events that occurred in Utah FORGE wells. These events were primarily attributed to geometric incompatibility between the bottomhole assembly and the borehole, with two instances where insufficient hole cleaning was also a contributing factor. Additionally, the model includes two agents from deep offshore oil-and-gas wells, where similar stuck pipe incidents occurred at greater depths than those in Utah FORGE. We tested this model on the most recently drilled Utah FORGE well, simulating real-time processing of drilling data. Results showed that the model successfully anticipated the sticking event and accurately identified the sticking mechanism. This anticipation was represented by a gradual increase in the sticking risk level within a 25-minute window. This indicates that, if implemented during actual operations, this model could provide the drilling team with timely information about the existence, severity, and causes of sticking risks. This, in turn, enables better-informed decisions and preventative measures, potentially resulting in the avoidance of stuck pipe incidents. While this study focused on stuck pipe prevention, the methodology can be applied to other—equally relevant—drilling risks, such as lost circulation events, wellbore breathing occurrences, well control incidents, etc. This paper proposes the first multi-agent sticking prediction model that combines statistically derived and model-based variables to provide a timely, comprehensive, and interpretable assessment of sticking risks. We demonstrate that, although geothermal drilling operations are often exposed to significant risks, the proper collection and processing of geothermal data can substantially reduce these risks, ultimately lowering the cost of geothermal well construction.

Press the Back button in your browser, or search again.

Copyright 2025, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.