CFD-machine learning integrated global sensitivity analysis of horizontal axis tidal turbine under wave-current loading

This study presents an integrated framework, which combines high-fidelity computational fluid dynamics, machine learning and global sensitivity analysis, to quantify the unsteady hydrodynamic performance of horizontal axis tidal turbine under combined wave-current loading. A validated computational fluid dynamics model, benchmarked against flume tests (with errors of 4.1% for thrust and 7.2% for torque), generates a matrix of tidal turbine's responses across a four-dimensional space (wave height, wave period, mean tip speed ratio, and relative submergence) through Latin Hypercube Sampling, which enables computationally efficient global assessment of horizontal axis tidal turbine. Gaussian Process Regression surrogate models trained on the dataset, provides accurate mapping for the turbine. This surrogate enables a full variance-based global sensitivity analysis using Sobol analysis, revealing physical insights of each parameter, which are otherwise inaccessible due to computational constraints. The proposed computational fluid dynamics model captures the turbine performance under both steady and unsteady loads, and the trained surrogate reliably predicts the wave-induced fluctuations in thrust and power. The mean thrust and power are primarily governed by the mean tip speed ratio, whereas unsteady hydrodynamic fluctuations of turbines are primarily governed by surface wave properties, particularly wave period, followed by wave height, tip speed ratio, and submergence depth. The CFD model is made available at: https://github.com/wangshiyeye/HorizontalAxisTidalTurbine.