Numerical Verification of Wave–Cylinder Diffraction and Machine-Learning Prediction of Free-Surface Evolution over Variable Bathymetry
Numerical Verification of Wave–Cylinder Diffraction and Machine-Learning Prediction of Free-Surface Evolution over Variable Bathymetry
Milad Baghban1
1) PhD Student in Marine Engineering (Hydromechanics specialization), Amirkabir University of Technology, Tehran, Iran
IOI Code
:
ACCA-ACAGCA
Publication :
4th International Congress on Science, engineering & New Technologies(secongress.com)
Abstract :
Numerical prediction of wave–cylinder diffraction and bathymetry-induced wave transformation requires computational models with clearly defined physical validity. This study investigates two complementary numerical approaches and evaluates machine-learning surrogates for free-surface evolution. First, a radial finite-volume formulation of linear potential-flow diffraction is implemented for a vertical cylinder under three regular-wave conditions corresponding to the DHI reference parameters R1–R3. The computed horizontal force amplitudes are 31.713, 63.425, and 45.671 N, with deviations of 0.781%, 0.781%, and 0.692% from the analytical diffraction solution, respectively. Radial-grid refinement from 100 to 1200 cells reduces the complex-force error from 12.884% to 0.237%. Second, a one-dimensional nonlinear shallow-water solver employing HLL numerical fluxes and hydrostatic reconstruction is used to simulate 27 combinations of wave conditions and bathymetric profiles. Numerical consistency is examined through lake-at-rest and spatial-refinement tests. Gaussian process regression and a temporal convolutional network are subsequently trained on these numerical solutions to predict maximum positive free-surface elevation and its time history, respectively. With 15 training, 3 validation, and 9 unseen-period test scenarios, the GP achieves a test MAE of 1.168 mm, while the TCN achieves a test RMSE of 3.428 mm. The findings support numerical verification of the implemented hydrodynamic models and demonstrate the feasibility of surrogate prediction within the shallow-water approximation, while highlighting reduced generalization accuracy under an unseen wave period. The study also establishes a computational basis for subsequent three-dimensional wave–structure investigations.
Keywords : wave cylinder diffraction variable bathymetry shallow water equations finite volume verification gaussian process temporal convolutional network