Detection of stuck pipe problems during drilling tripping operations using artificial intelligence approach

Abstract

Abstract Stuck pipe remains a critical challenge in the petroleum industry, often leading to significant non-productive time (NPT) and operational losses. Traditional prevention methods are frequently ineffective due to the high sensitivity to parameter changes and the presence of noisy or incomplete data. This study aims to apply machine learning algorithms to classify and predict stuck pipe incidents using real operational data, enabling early warnings to minimize NPT and reduce operational costs through models trained on both stuck and non-stuck conditions. This study tackled the stuck pipe issue through a two-phase machine learning approach. In the first phase, classification model Multi-Layer Perceptron (MLP) was employed to classify stuck pipe incidents across drilling and tripping operations in Middle Eastern wells. The second phase focused on forecasting hook load values using a Long Short-Term Memory (LSTM) model, aimed at predicting stuck pipe events in advance. Sequential hook load readings served as inputs to the LSTM, allowing the generation of early warnings during tripping operations. The results of this study illustrate that classification model demonstrated high performance, with accuracy levels ranging from 89% to 97.83%. Their evaluation relied on metrics such as Precision, Recall, F1-Score, Accuracy, Decision Boundaries and Principal Component Analysis (PCA). Likewise, the LSTM model showed strong predictive capabilities, achieving 90–97% accuracy based on R², Mean Absolute Error (MAE), and Mean Squared Error (MSE) values. These results confirm the effectiveness of machine learning in detecting and predicting stuck pipe incidents, enhancing safety and reducing NPT. The developed machine learning algorithms helped to classify the different suck pipe problems, identify, their reasons and ultimately can early predict stuck pipe problems during drilling operations. The proposed approach presents substantial potential for real-time monitoring and proactive drilling risk mitigation.