Abstract
Time savings and bit longevity is one of the major challenges in Coal Seam Gas unconventional fields Onshore Queensland. Maximize rate of penetration based on best drilling parameter was the key target to tackle these issues. The dataset was accumulated from more than 80 bits run. A supervised machine-learning algorithm was used to classify drilling parameters that increase ROP and bit endurance. To achieve this goal, our focus was on a) optimizing PDC bit design, drilling hydraulics and b) developing a Drillers Roadmap to ensure optimal parameters were run in the field.
Operator and Service Company worked closely to select two PDC frames and benchmark performance between them. A novel approach was then taken to capture the rig sensors data. These steps were as follows: Capture big data using sensors integrated on the rig to produce drilling parameters data sets. An algorithm to clean large data sets with the use of Python Pandas libraries was created. Using supervised machine learning and clustering the data to identify which drilling parameters gave the best results. Process the data set using visualization tools to develop heat maps vs depth for each of the drilling parameters (WOB, RPM, flow rates and Torque). Create a Drillers Roadmap identify drilling parameter trends by depth and provide a "sweet spot" based on machine learning for fine tuning operational window.
A formal process for optimizing performance was developed. After 6 wells, optimal operating ranges were identified. This process leads the driller to respond to changes in formation and deliver more consistent drilling performance with excellent results. ROP increased from 50m/hr to 150m/hr, driven by improved ROP in the lower section of the well. Time saving of more than 150hrs for the drilling campiagn. Refine PDC bit designs based on optimized parameters, therefore improving bit life through cutter and hydraulic enhancements.
The innovation of this methodology is to gather drilling information, cleaning the data with Python using Pandas library thus making this process more efficient. After the data files were clean and QAQC, a supervised machine Learning algorithm was used to optimize ROP's by taking drilling datasets and fine-tuning the parameters (WOB, RPM, Flow rate) using drilling dynamics (vibration) and MSE models. Finally, innovative visualizations (heat maps) were produced in order to give a clear insight of the drilling information gathered by the rig sensors to the company men and drillers.
