ROP and TOB optimization using machine learning classification algorithms

Drilling Optimization has consistently generated research interest over the years because the cost saving beneifts associated to improve drilling efficiency. Rate of penetration (ROP) and torque-on-bit (TOB) predictions have become critical to the successful drilling optimization efforts. Several physics-based and data-driven models have been developed for ROP and TOB prediction and majority of the data-driven models use regression-based approaches. This paper introduces a new approach to ROP and TOB prediction by modeling them as a classification problem consisting of two regions (low and high ROP and TOB respectively) based on a user-defined threshold. ROP and TOB are modeled as a function of with weight-on-bit (WOB), flow rate, rotray speed (RPM) and unconfined compressive strength (UCS). Five different machine learning classifcation algorithms - logistic regression, linear discriminant analysis (LDA), quadratic driscriminant analysis (QDA), support vector machines (SVM) and random forest were implemented in this paper to develop the classification model. Using the Area Under Curve (AUC) as the classification performance metric, results from the simulations showed that the best classifier should be chosen for each formation. Also, for the practical application of this approach to ROP and TOB prediction, a probability gradient heat map of RPM and WOB was developed as a tool to help the driller make informed decisions on the combinations of RPM and WOB values that would yield the desired regions of ROP and TOB.