Perpustakaan Digital ITB

Accurate prediction of Rate of Penetration (ROP) and the identification of drilling efficiency remain critical challenges in optimizing drilling operations and reducing non-productive time. Traditional methods often struggle to capture the complex, time-dependent performance and bit-rock interactions inherent in diverse drilling zones. This study introduces an Artificial Neural Network (ANN) based approach to predict Rate of Penetration (ROP) and identify drilling efficiency zones. Utilizing forward and backward propagation with stochastic gradient descent, the model incorporates Mechanical Specific Energy (MSE) as a key input to capture the influence of critical variables such as Weight on Bit (WOB), RPM, and friction factors on bit-rock interaction. Following ROP prediction, drilling performance is evaluated across depth intervals to classify efficient and inefficient zones, enabling targeted parameter optimization to reduce non-productive time and costs. Model reliability was ensured through Holdout Method. In parallel, a drilling risk assessment module was developed to monitor loss, overflow, and stuck pipe conditions using trip tank sensors and torque/drag indicators. All predictive results were integrated into a web-based monitoring platform capable of real-time ROP prediction, drilling risk visualization, and Early Warning System classification. The model training achieved a Mean Squared Error of 0.6783 and an R² of 0.9989, with validation on an offset well yielding an R² of 0.8627. In addition, the Early Warning System demonstrated reliable classification performance, with F1-scores ranging from 0.67 to 0.89 for stuck, loss, and overflow risk prediction. System integration enabled real-time responses with a lag time of less than 30 minutes, significantly reducing rig time.