Session: 11-03-03 Data Science Applications in Drilling

Paper Number: 128569

128569 - Data Driven Approach for Gel-Breaking Time Predictions 

The main objective of this study is to develop a machine learning model for predicting gel breaking time in drilling operations. This predictive model  is capable of reducing nonproductive time, improving drilling efficiency and drilling  safely  on the field by utilizing a wide range of input parameters readily available by mud engineer such as fluid rheological properties, volume percent, mud weight, and oil-to-water ratio.

To achieve this goal, a selected data set, limited to drilling fluids used in the North Sea and Azerbaijan, was collected and tested in the laboratory using a high-precision rheometer. Stress overshoot tests using crosshatched cylindrical geometry were used to obtain the measurements. Later, the measurements were fitted by a double decay function and the times to reach the plateau or steady state shear stress were determined. A combination of supervised machine learning techniques, including regression and ensemble methods, was used to construct a predictive model of gel breaking time based on 20 independent variables. Feature design and selection played a critical role in isolating the most influential factors affecting gel break time.

Cross-validation and hyper-parameter tuning were used to ensure robustness and generalizability of the model for comparison with laboratory experimental data. Preliminary assessments within controlled environments demonstrated an acceptable degree of accuracy, with a mean absolute error of 4.7 seconds and a mean absolute percentage error of 25%. While the machine learning approach for gel breaking time prediction has shown promising results in controlled settings, it is essential to highlight that field testing has not yet been conducted. Successful field testing could potentially lead to reduced downtime, enhanced drilling efficiency, and improved safety measures. The adaptability of the model across various geographic location and drilling conditions holds promise, and its integration of multiple input parameters presents an opportunity for proactive decision-making in field-based drilling operations.

Presenting Author: Roger Aragall Baker Hughes Company

Presenting Author Biography: Roger Aragall is a fluid mechanics lead engineer at Baker Hughes. He graduated with a B. Eng. in Mechanical Engineering at the Polytechnic University of Catalonia, with a M. Sc. in Mechanical Engineering at the University of Girona, Spain, and with a PhD in Applied Mechanics at the Clausthal University of Technology, Germany. His field of research deals with multiphase flows, non-Newtonian fluid mechanics, flow instabilities and transient flows, where he develops simulation and experimental methods.

Authors:

Roger Aragall Baker Hughes Company
Syed Ehsanur Rahman Baker Hughes Company
Alexa Milstein Baker Hughes Company
Reza Ettehadi Osgouei Baker Hughes Company