Detecting bit balling events using time-series clustering and anomaly detection
Department of Petroleum Engineering, University of Port Harcourt, Port Harcourt, Rivers State, Nigeria.
International Journal of Scholarly Research in Engineering and Technology, 2026, 07(01), 019-021.
Article DOI: 10.56781/ijsret.2026.7.1.0013
Publication history:
Received on 11 February 2026; revised on 19 March 2026; accepted on 21 March 2026
Abstract:
Bit balling is an issue that severely affects the performance of drilling, and even though it is difficult to spot the problem at its early stages using the traditional methods, the initial indicators are subtle. This research paper introduces a machine learning architecture that is a hybrid of K-Means time-series clustering and later, the Isolation Forest model was trained on normal clusters to distinguish the abnormalities of the bit balling onset. It included 250+ hours of 1 Hz-sampled Rate of Penetration (ROP), Torque, Rotary Speed (RPM) and Pump Flow Rate of three horizontal sandstone wells. The lagged features of the 3-5 minute time patterns were designed, normalized, clustered to categorize the normal drilling conditions. Isolation Forest was subsequently trained on normal clusters to discriminate aberrations of bit balling onset. The model identified 85% of expert annotated bit balling cases with an average lead time of 8 minutes with a precision of 0.79, recall of 0.85 and an F1-score of 0.82. Optimal clusters with five clusters of drilling behaviors that represent different drilling behaviors were verified by the Minitab analysis, with balling events mostly noted in low-ROP and high-torque and were coincident with poor hole cleaning signatures. This combined system can provide a very strong early-warning system that can optimize real-time drilling and minimize non-productive time as a result of bit dysfunction.
Keywords:
Bit Balling; Time-Series Clustering; Anomaly Detection; Isolation Forest; K-Means; Drilling Data; Real-Time Monitoring
Full text article in PDF:
Copyright information:
Copyright © 2026 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0