USING UNSUPERVISED MACHINE LEARNING ALGORITHM TO PREVENT THE STICKING OF DRILLING AND CASING STRINGS

Abstract

The relevance. Drilling failures and accidents will continue to attract attention in drilling for oil and gas as more complex wells are being drilled across depleted zones to reach deeper reservoir targets. Stuck pipe incident continue to be a major contributor to non-productive time in drilling operations for oil and gas. When a stuck pipe incident occurs, costly corrective actions may include fishing operations, sidetracking the hole, or completely having to drill a new well. Stuck pipe warning signs are often undetected early enough for the deployment of effective mitigation strategies due to human mistakes and crew changes during drilling operations. The unsupervised machine learning algorithm is programmed to automatically detect abnormalities in real-time drilling parameter trends and predict potential stuck pipes, communicate observations in the form of alerts to engineers in advance to allow proactive corrective actions. Early detection of a stuck pipe and mitigating the incident in real time not only help to prevent its occurrence, but also help in making informed decisions to the appropriate freeing mechanism to adopt if it occurs. The main aim: create the stuck pipe detection model which predicts failure probability during the well drilling using mud logging service data. Objects: multivariate-sensing time-series data of mud logging service. Methods: analysis of current achievments in the field of anomaly detection techniques using machine learning; developing the stuck pipe detection model with open-source Python frameworks. Results. The authors have developed stuck pipe detection model with HTM algorithm, evaluated performance with test dataset. Promising areas of further research were identified.

Keywords

• drilling operation
• stuck pipe
• machine learning
• multivariate-sensing time-series
• anomaly detection
• hierarchical temporal memory