An Integrated CFD–Machine Learning Framework for Flow Assurance Risk Assessment During Hot Oil Commissioning of Subsea Jumpers

To mitigate flow assurance risks during the hot oil commissioning of deepwater jumpers, this study develops a transient displacement risk assessment framework integrating CFD–machine learning surrogate models. A 3D numerical model using VOF and conjugate heat transfer simulated hot crude displacing cold diesel. For three representative jumper configurations (M-shaped, U-shaped, and straight pipes), the effects of spatial geometry on mixing, heat transfer, and gelled oil length were investigated. The results show that jumper geometry significantly affects flow regime and thermal evolution. The M-shaped jumper enhances secondary-flow-induced mixing and heat transfer, yielding the shortest gelled oil length (0.5 m). In contrast, straight pipes show weak mixing and stratified flow, indicating higher local gelation risk. Sensitivity analysis shows that higher flow velocity and water cut reduce gelled oil length by decreasing residence time and increasing fluid thermal inertia. A machine learning surrogate trained on CFD data enables rapid prediction of key flow assurance parameters: maximum temperature drop, pressure drop, and gelled oil length. The multilayer perceptron outperforms the random forest model, with R2 > 0.94 for all three parameters. This study provides a theoretical basis and efficient tool for rapid assessment, parameter optimization, and gelation risk management in deepwater hot oil commissioning.