Torque-on-bit (TOB) prediction and optimization using machine learning algorithms

Abstract

Drilling optimization can have direct and indirect cost savings implications on drilling a well. As a result, over the years, attention has been given to developing analytical and data-driven models that effectively predict and optimize the rate of penetration (ROP). This paper focuses on another aspect of improving drilling performance - predicting and optimizing torque-on-bit (TOB). TOB is a critical component in estimating the energy expanded during drilling called mechanical specific energy (MSE). TOB was modeled as a function of rotary speed (RPM), weight-on-bit (WOB), flow rate, pump pressure, and unconfined compressive strength (UCS). Five regression-based machine learning algorithms - linear regression, ridge regression, support vector machines, random forest, and boosted trees were used to build TOB models. The performance of the five algorithms was compared using the root mean square error (RMSE) and results showed that boosted trees was the best performing across all the formations. Three direct search optimization algorithms - Nelder Mead, differential evolution, particle swarm optimization (PSO) were used to optimize TOB and results showed that PSO consistently produced minimized TOB values in all the 12 formations. Finally, hypothesis testing was used to statistically test if there was a significant difference between the measured and optimized TOB values. The p-value obtained for each formation was less than the significance level of 5%, indicating the minimized TOB values were significant.