Data-driven surrogate modeling for frequency-domain boundary-element analysis

Boundary element methods (BEM) based on linear potential-flow theory are widely used for analyzing frequency-domain hydrodynamics in offshore and marine energy systems. However, the need for repeated BEM evaluations during geometry-driven design exploration and sensitivity analysis can result in substantial computational costs. This paper presents a surrogate-based modeling framework for rapid prediction of hydrodynamic coefficients directly from geometric parameters. The proposed approach combines dimensionality reduction with supervised machine learning to establish a compact and physically consistent mapping between geometric design variables and frequency-dependent hydrodynamic quantities, including added mass, radiation damping, and complex-valued wave excitation forces. The resulting surrogate model predicts the full set of frequency-domain hydrodynamic coefficients and wave-excitation forces for a new geometry in less than one second while maintaining high predictive accuracy ( R 2 > 0.94 for all predicted quantities), thereby facilitating efficient design exploration and sensitivity analysis for floating offshore structures and marine energy devices.