Real-Time AI-Based Prediction of Stuck Pipe Events to Minimize Non-Productive Time in Drilling Operations

Abstract

Abstract Stuck pipe remains a major drilling challenge, causing significant non-productive time (NPT), increased costs, and operational risks. This paper presents an artificial intelligence (AI)-based framework for early prediction of stuck pipe events using real-time drilling data to enable proactive risk mitigation. The solution utilizes a Long Short-Term Memory (LSTM) network trained on refined WITSML data from multiple wells. Key drilling parameters, including hookload, torque, standpipe pressure, flow rate, rotary speed, and rate of penetration, were combined with engineered features to identify precursor patterns associated with stuck pipe incidents. Real-time inference workflows were developed to support field deployment. The model achieved 96.1% accuracy, 0.991 ROC-AUC, 83.3% precision, 87.5% recall, and an F1-score of 85.4%. It successfully predicted 7 of 8 confirmed stuck pipe events while maintaining a false alarm rate of only 2.1%. An average Early Warning Time of 2.9 minutes allowed drilling crews to implement preventive actions before conditions became irreversible. Deployment analysis indicated the potential to reduce stuck pipe-related NPT from 142.5 hours to 31hours, representing a 78.2% reduction. The results demonstrate that AI-driven prediction can improve drilling efficiency, reduce operational costs, and support digital transformation through reliable, actionable early-warning alerts.