Document Type : Original Research Paper

Authors

• Akbar Esmaeilzadeh ¹
• Shahrokh Khosravimanesh ²
• Reza Mikaeil ¹
• Sasan Ghorbani ³

¹ Department of Mining Engineering, Environment Faculty, Urmia University of Technology (UUT), Urmia, Iran

² Department of Mining Engineering, Isfahan University of Technology (IUT), Isfahan, Iran

³ Department of Mining Engineering, Hamedan University of Technology, Hamedan, Iran

Abstract

The rate of penetration (ROP) is a key indicator in drilling operations, directly influencing cost estimation and project scheduling. Direct measurement of ROP is typically costly and limited in applicability; therefore, developing accurate data-driven predictive models is of great importance. In this study, a dataset of 492 laboratory measurements was collected, incorporating four independent variables-Schimazek’s Abrasivity Index (SFa), drilling fluid electrical conductivity (EC), weight on bit (WOB), and bit rotation speed (BRS)-while ROP was considered as the dependent variable. Four supervised machine learning algorithms, namely Huber regression, Elastic Net, Ensemble, and Decision Tree, were employed for ROP prediction. The predictive performance of the models was evaluated using MAPE, R2, RMSE, and EVS metrics. Additionally, a composite objective function integrating these indices was defined to provide an overall ranking of the models. The results revealed that the Decision Tree model achieved the best performance with an objective function value of 0.2975, followed by the Ensemble, Elastic Net, and Huber regression models with values of 0.5655, 1.4471, and 1.2702, respectively. These findings demonstrate that the Decision Tree algorithm offers the most accurate and reliable predictions of ROP and can serve as an effective tool for optimizing drilling time and reducing operational costs.

Keywords

• Rate of penetration
• Electrical conductivity
• Huber regression
• Elastic network
• Decision tree

Main Subjects

• Rock Mechanics