Hybrid dimensionless-machine learning approach for predicting cuttings bed height in inclined wellbores

Effective cuttings removal in deviated and horizontal wells is critical for enhancing drilling efficiency and minimizing non-productive time (NPT) caused by inadequate hole cleaning. Although computational fluid dynamics (CFD) and mechanistic models have been widely used to simulate cuttings accumulation, their high computational cost, complexity, and reliance on detailed input parameters limit their applicability for real-time drilling operations. To address these challenges, this study introduces a hybrid dimensionless machine learning framework for predicting cuttings bed height in inclined wellbores. A novel set of dimensionless parameters was derived and used to train and evaluate multiple machine learning (ML) models, including linear regression (LR), deep neural networks (DNN), support vector regression (SVR), random forests (RF), and extreme gradient boosting (XGBoost). The models were developed using 1069 bed height measurements collected from diverse experimental flow loops. Among all candidates, the XGBoost model achieved the best performance, with a mean absolute percentage error (MAPE) of 13% and a root mean square error (RMSE) of 0.08 on unseen datasets. It also outperformed the established Duan and Ozbayoglu empirical models. Feature analyses using SHapley Additive Explanations (SHAP) and RF feature importance highlighted the Froude number, inlet feed concentration (correlated with the drilling rate of penetration), and drillpipe eccentricity as the most influential factors governing bed height ratio. The findings confirm that integrating dimensionless analysis with ML yields a robust, interpretable, and computationally efficient framework for real-time monitoring and prediction of cuttings bed accumulation, providing valuable insights for optimizing hole cleaning in inclined wellbores.