| dc.description.abstract | Drilling fast and safe has and is a key value for operator companies. Maintaining a high Rate of Penetration (ROP) during drilling without causing incidents enables drilling project costs to be reduced.
In 1964 Bingham created the first mathematical model for calculating the ROP based upon the parameters Weight on Bit (WOB) and Rotations per Minute (RPM). Many researchers have improved Bingham‘s model since it‘s creation and developed other models with other parameters as well. Some of the parameters relevant in ROP models are Equivalent Circulating Density (ECD), Confined Compressive Strength (CCS) and drill bit properties to name a few. Slowly and steadily the accuracy of ROP calculations have improved since 1964 and in recent years, as a part of the ongoing digitalization in the industry, a new method for modeling ROP has been implemented, machine learning (ML).
Using mathematical ROP models, or physics based ROP models to calculate ROP can be inaccurate due to the complexity of geology, bit geometry and forces in the well. ML models handles the complexity by monitoring a vast amount of drilling data. Using ML models for calculating ROP can be expensive and time consuming due to big data being monitored, prepared, modelled, optimized and analyzed.
In contrast to ML models, using all available well data for calculating ROP from Bingham‘s physic-based model
does not provide optimal results. However, by developing a new method of data preparation by selecting data by Bingham‘s model‘s principle, the well data set can be reduced from 16063 rows of data to only 53 rows of data while improving the accuracy of ROP for Bingham’s model. Implementing a method of selecting data will also be applied for the ML models, as well as a new physics-based model, modified Bingham model. | |