Real-Time, Data-Driven Cuttings Concentration Estimation Across a Wide Range of Drilling Conditions

Abstract This study explores the potential of machine learning models to predict cuttings concentration in various drilling operations, leveraging a comprehensive database from the Tulsa University Drilling Research Projects (TUDRP). The database encompasses over 50 years of experimental data, including more than 1,800 data points from large-scale flow loop experiments. Key parameters such as flow rate, rate of penetration (ROP), inclination angle, mud type, and pipe rotation were considered, with dimensional analysis and preprocessing techniques applied to ensure data quality. Advanced machine learning algorithms, including Artificial Neural Networks (ANN), XGBoost, Random Forest, and Support Vector Regressor, were trained on the dataset, incorporating both raw variables and dimensionless groups. Model evaluation using RMSE and R² metrics demonstrated high predictive accuracy, with R² values exceeding 0.89 and RMSE within 5.41. The results underscore the critical role of factors such as inclination, ROP, and fluid velocity in influencing cuttings concentration, aligning with prior studies. This novel, data-driven approach highlights the potential of machine learning to enhance drilling efficiency by providing reliable insights for wellbore cleaning strategies and decision-making. By bridging historical data with modern analytical techniques, this study offers a significant advancement in predicting and optimizing cuttings transport in drilling operations.