Early detection of stuck pipe events is pivotal due to their role as major causes of non-productive time. Although researchers have adopted data-driven approaches, such as machine learning, challenging issues persist, such as limited datasets, limited stuck pipe events, and various causes of stuck pipe. This study proposes a novel machine learning approach that incorporates physical knowledge, explaining universal facts, to overcome the aforementioned challenges.

This study creates a prediction model that focuses on the hook load to predict occurrences of stuck pipes. To create the model, we adopted a novel idea that incorporates physical knowledge such as torque and drag models. Because the measured hook load contains uncertainties and is affected by operating conditions, the hook load is expressed using physical knowledge with unknown parameters, and the parameters are determined using data science technologies, including machine learning, based on adjacent historical hook load data. The prediction model is created using only the data in the normal conditions, and the model can express the hook load in normal condition. When the measured hook load exceeds that predicted via the model, it can be assumed that a larger a friction (drag) force is exerted, resulting in a high stuck risk. The deviation, calculated from the predicted and measured hook loads, is acquired, and finally, if the deviation exceeds the normal range determined through data science techniques, including machine learning, the risk of blockage is outputted.

This study initially introduces the stuck pipe predictions using both supervised and unsupervised machine learning approaches. Subsequently, we present machine learning integrating physical knowledge and demonstrate early stuck pipe detection using field data containing stuck pipe events. This novel machine learning approach incorporating physical insights may contribute to significant reduction of nonproductive time in drilling operations, potentially preventing well abandonment.

You can access this article if you purchase or spend a download.