Cuttings transport characteristics and lifting efficiency prediction in offshore oil and gas well sidetracking: A coupled CFD-DEM and machine learning approach

Offshore casing-window sidetracking is an important means of developing remaining oil and gas resources, and efficient cuttings transport is essential for safe and continuous sidetracking operations. However, cuttings transport efficiency is governed by nonlinear coupling among multiple factors, making lifting efficiency under different conditions difficult to predict rapidly and accurately. To address this issue, this study developed a coupled CFD-DEM and machine learning approach for cuttings transport analysis and lifting efficiency prediction. A CFD-DEM model was established and validated through cuttings transport experiments to analyze annular flow and cuttings migration. The results show that spiral channels enhance local flow and fluid-particle momentum exchange near the tool. Within the investigated range, flow rate and viscosity have substantially greater effects on lifting efficiency than fluid density and channel count. Based on CFD-DEM data, six prediction models, BPNN, SVR, RF, GPR, XGBoost, and LightGBM, were developed, with XGBoost achieving the best overall performance and a test-set R 2 of 0.97. SHAP and feature ablation analyses indicate that flow rate and viscosity are the dominant factors controlling lifting efficiency, with contributions of 33.7%-42.5% and 32.9%-37.6%, respectively. Latin hypercube sampling was further used to generate unseen parameter combinations for independent experimental validation, and XGBoost accurately characterized lifting-efficiency variations under different conditions. The proposed approach enables rapid evaluation of cuttings transport characteristics under offshore sidetracking conditions and supports field parameter adjustment.