TY - JOUR
T1 - Detecting downhole vibrations through drilling horizontal sections
T2 - machine learning study
AU - Saadeldin, Ramy
AU - Gamal, Hany
AU - Elkatatny, Salaheldin
N1 - Publisher Copyright:
© 2023, The Author(s).
PY - 2023/12
Y1 - 2023/12
N2 - During the drilling operations and because of the harsh downhole drilling environment, the drill string suffered from downhole vibrations that affect the drilling operation and equipment. This problem is greatly affecting the downhole tools (wear and tear), hole problems (wash-out), mechanical energy loss, and ineffective drilling performance. Extra non-productive time to address these complications during the operation, and hence, extra cost. Detecting the drillstring vibrations during drilling through the downhole sensors is costly due to the extra service and downhole sensors. Currently, the new-technology-based solutions are providing huge capabilities to deal intelligently with the data, and machine learning applications provide high computational competencies to learn and correlate the parameters for technical complex problems. This research presented a successful case study for developing machine learning models through a comprehensive methodology process for vibration detection using surface rig data through data collection, preprocessing, analytics, training and optimizing the models’ parameters, and evaluating the performance to have the best prediction results. Evaluating the models’ performance showed that obtained predictions have a great match with actual measurements for the different stages of training, testing, and even during models’ validation with unseen well data. Real-field horizontal drilling data was utilized to feed and train the models through different tools named radial basis function (RBF), support vector machines (SVMs), adaptive neuro-fuzzy inference system (ANFIS), and functional networks (FN) to auto-detect the three types of downhole vibrations (axial, torsional, and lateral). The study results showed a high correlation coefficient (higher than 0.9) and technically accepted average absolute percentage error (below 7.5%) between actual readings and predictions of the developed ML models. The study outcomes will add to the automation process of drilling operations to avoid many tools failure by comparing predicted vibrations versus downhole tools limits such as red zone and continuing drilling without interruption to the well total depth especially while drilling horizontal sections.
AB - During the drilling operations and because of the harsh downhole drilling environment, the drill string suffered from downhole vibrations that affect the drilling operation and equipment. This problem is greatly affecting the downhole tools (wear and tear), hole problems (wash-out), mechanical energy loss, and ineffective drilling performance. Extra non-productive time to address these complications during the operation, and hence, extra cost. Detecting the drillstring vibrations during drilling through the downhole sensors is costly due to the extra service and downhole sensors. Currently, the new-technology-based solutions are providing huge capabilities to deal intelligently with the data, and machine learning applications provide high computational competencies to learn and correlate the parameters for technical complex problems. This research presented a successful case study for developing machine learning models through a comprehensive methodology process for vibration detection using surface rig data through data collection, preprocessing, analytics, training and optimizing the models’ parameters, and evaluating the performance to have the best prediction results. Evaluating the models’ performance showed that obtained predictions have a great match with actual measurements for the different stages of training, testing, and even during models’ validation with unseen well data. Real-field horizontal drilling data was utilized to feed and train the models through different tools named radial basis function (RBF), support vector machines (SVMs), adaptive neuro-fuzzy inference system (ANFIS), and functional networks (FN) to auto-detect the three types of downhole vibrations (axial, torsional, and lateral). The study results showed a high correlation coefficient (higher than 0.9) and technically accepted average absolute percentage error (below 7.5%) between actual readings and predictions of the developed ML models. The study outcomes will add to the automation process of drilling operations to avoid many tools failure by comparing predicted vibrations versus downhole tools limits such as red zone and continuing drilling without interruption to the well total depth especially while drilling horizontal sections.
UR - http://www.scopus.com/inward/record.url?scp=85152652036&partnerID=8YFLogxK
U2 - 10.1038/s41598-023-33411-9
DO - 10.1038/s41598-023-33411-9
M3 - Article
C2 - 37069188
AN - SCOPUS:85152652036
SN - 2045-2322
VL - 13
JO - Scientific Reports
JF - Scientific Reports
IS - 1
M1 - 6204
ER -