Evaluation of Derived Controllable Variables for Predicting Rop Using Artificial Intelligence in Autonomous Downhole Rotary Drilling System

Abstract

Abstract Fossil fuel energy dominate the world energy mix and plays a fundamental role in our economy and lifestyle. Drilling of wellbore is the only proven method to extract the hydrocarbon reserves, an operation which is both highly hazardous and capital intensive. To optimize the drilling operations, developing a high fidelity autonomous downhole drilling system that is self-optimizing using real-time drilling parameters and able to precisely predict the optimal rate of penetration is essential. Optimizing the input parameters; surface weight on bit (WOB), and rotary speed (RPM) which in turns improves drilling performance and reduces well delivery cost is not trivial due to the complexity of the non-linear bit-rock interactions and changing formation characteristics. However, application of derived variables shows potential to predict rate of penetration and determine the most influential parameters in a drilling process. In this study the use of derived controllable variables calculated from the drilling inputs parameters were evaluated for potential applicability in predicting penetration rate in autonomous downhole drilling system using the artificial neutral network and compared with predictions of actual input drilling parameters; (WOB, RPM). First, a detailed analysis of actual rock drilling data was performed and applied in understanding the relationship between these derived variables and penetration rate enabling the identification of patterns which predicts the occurrence of phenomena that affects the drilling process. Second, the physical law of conservation of energy using drilling mechanical specific energy (DMSE) defined as energy required to remove a unit volume of rock was applied to measure the efficiency of input energy in the drilling system, in combination with penetration rate per unit revolution and penetration rate per unit weight applied (feed thrust) are used to effective predict optimum penetration rate, enabling an adaptive strategize which optimize drilling rate whilst suppressing stick-slip. The derived controllable variable included mechanical specific energy, depth of cut and feed thrust are calculated from the real- time drilling parameters. Artificial Neutral Networks (ANNs) was used to predict ROP using both input drilling parameters (WOB, RPM) and derived controllable variables (MSE, FET) using same network functionality and model results compared. Results showed that derived controllable variable gave higher prediction accuracy when compared with the model performance assessment criteria commonly used in engineering analysis including the correlation coefficient (R2) and root mean square error (RMSE). The key contribution of this study when compared to the previous researches is that it introduced the concept of derived controllable variables with established relationship with both ROP and stick-slip which has an advantage of optimizing the drilling parameters by predicting optimal penetration rate at reduced stick-slip which is essential in achieving an autonomous drilling system. :