Using machine learning to predict wall shear stress considering the presence of sediment in wellbores

Sediment transport is a fundamental topic that has been extensively investigated in hydraulic engineering and fluid mechanics. The shear stress acting on the particle surface is a primary variable governing sediment transport in closed conduits. In this study, the parameters influencing the wall shear stress are determined by a dimensional analysis applying the Buckingham Pi Theorem. According to dimensional analysis, the resulting wall shear stress is a function of the Reynolds number, Froude number, sediment concentration, hole inclination, and pipe rotation. The dimensionless wall shear stress coefficient is estimated using five experimentally measured inputs. The difficulty and cost of experimental studies are increasing the importance of computational approaches, including CFD and data-driven methods. Therefore, machine learning captures complex patterns without a fixed formula, provides fast alternatives for quick design checks, and quantifies uncertainty for risk-aware decisions. The features are standardized within a 5-fold cross-validation, and Ridge, Bayesian Ridge, Support Vector Regression (SVR-RBF), XGBoost, and Gaussian Process Regression (GPR) are compared. The results show that XGBoost outperforms the other models with highest accuracy (R 2 ≈0.994, RMSE ≈2.3×10 -3 ), AAPE≈4.55%). SVR-RBF and GPR followed with ≈18.8% and ≈19.4%, respectively, while linear baselines (Ridge and Bayesian Ridge) exhibited higher errors (≥23%). These results indicate that boosted trees provide the most accurate percentage-error performance.