Physics-informed machine learning and computational analysis of a gas-assisted water jet cleaning system for cylindrical offshore structures

This study presents a combined experimental, computational, and machine learning framework for analyzing a gas-assisted high-pressure water jet cleaning system designed for offshore cylindrical pile-leg maintenance. The system employs an annular air envelope surrounding a high-pressure water jet to reduce momentum losses in submerged conditions. Bubble formation dynamics were modeled using the Rayleigh–Plesset equation and validated through high-speed imaging, with stable bubble generation governed by an empirical pressure–depth relationship \\({P}_{a,\\text{m}\\text{i}\\text{n}}=0.20+0.03h\\) MPa. A three-factor orthogonal design varying water jet pressure (10–30 MPa), air pressure (0.3–0.9 MPa), and standoff distance (33–66 mm) revealed that water jet pressure dominates cleaning performance, followed by standoff distance and air injection pressure. The optimal standoff range was 18–25 nozzle diameters, and the gas-assisted configuration improved cleaning efficiency by approximately 40% relative to a fully submerged jet.A synthetic dataset of 800 samples was generated from coupled governing relations describing bubble dynamics, hydrostatic pressure effects, and jet decay behavior, and used to train multiple regression models. Among seven evaluated algorithms, Gradient Boosting achieved the highest predictive accuracy (R 2 = 0.9409, RMSE = 2.15 kPa), outperforming linear and kernel-based methods and confirming strong nonlinear dependencies in the system. Feature importance analysis indicated that water jet pressure contributes approximately 74% of output variance, while standoff distance contributes 17%, with air pressure exhibiting a saturation effect beyond the bubble formation threshold. Overall, the results demonstrate a robust physics-informed machine learning framework capable of accurately predicting cleaning performance and supporting extrapolation toward offshore operating conditions up to 50 m depth.