Development of drilling optimization models for autonomous rotary drilling systems.

The growing global energy demand and strict environmental policies motivate the use of technology and performance improvement techniques in drilling operations. Traditional drilling methods depend on the effectiveness of the human-driller in the management of operating parameters to improve system performance. Although existing work has identified the need to upscale from manual drilling to autonomous drilling systems with results based on manipulation of surface drilling parameters to construct responses for rate of penetration (ROP) so as to identify local maxima, this approach has limited applicability and is prone to intense operational redundancy. This thesis presents predictive optimization models that use machine learning (ML) data analytics with actual field drilling data and experimental studies to develop predictive models for rock unconfined compressive strength (UCS) and ROP, enabling optimal decision-making protocols. To evaluate optimized operating procedure, this thesis presents a comparative study of surface operating parameters using weight on bit (WOB), and rotary speed (RPM) versus drilling mechanical specific energy (DMSE), and feed thrust (FET). The study used a data-driven approach, whereby offset drilling data was combined with the machine learning model to find a pair of input operating variables that serve as the best tuning parameters for the topdrive and drawwork system. The results illustrate that derived variables (DMSE, FET) gave higher prediction accuracy, with correlation coefficient (R2) of 0.985, root mean square error (RMSE) of 7.6 and average absolute percentage error (AAPE) of 34; whereas using the surface operating parameters (WOB, RPM) delivered an R2, RMSE and AAPE of 0.74, 28 and 106 respectively. Additionally, the ML predictive model for rock UCS using basic drilling parameters showed that Artificial Neutral Network (ANN) and CATBoost gave acceptable qualitative instantaneous UCS prediction whilst drilling. The work further showed that continuous drill-off testing can be formulated as a Markov Decision Process (MDP), which intermittently analyzes a batch of real-time data using a Q-value algorithm to select the pair of surface operating parameters. The findings showed that application of these models could improve drilling performance by 30-60% compared to best offset well. Moreover, it will enhance operational health and safety and provides an engineered approach to improve the efficiency of the drilling process in terms of cost and time.

Name dc:type.qualificationname

PhD

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

Robert Gordon University

Year dc:date.issued

2023

Author dc:creator

• Amadi, Kingsley Williams

Advisor dc:contributor.advisor

• I. Iyalla and P. Radhakrishna

• Drilling
• Machine learning
• Prediction

Language dc:language

en

Identifier

oai:rgu-repository.worktribe.com:2445718
https://doi.org/10.48526/rgu-wt-2445718

OAI identifier oai:identifier

oai:rgu-repository.worktribe.com:2445718