A surrogate for predicting heave added mass in cylindrical point absorbers using a sequential geometry-to-frequency regression framework

Abstract

Estimation of frequency-dependent added mass is essential for resonance tuning in point absorber-type wave energy converters (WECs). Boundary element method (BEM) solvers remain the standard approach for this task, offering high fidelity but at significant computational cost, which limit their practicality for extensive parametric analyses. This study presents a computationally efficient two-stage surrogate modeling framework for estimating the heave added mass of cylindrical point absorbers in deep water. The model spans a design space defined by diameters of 3-12 m, drafts of 0.6-27 m constrained by draft-to-diameter ratios between 0.2 and 2.25, and wave periods of 6-12 s. The underlying data were obtained from hydrodynamic simulations in Ansys AQWA. In Stage 1, polynomial regression models were constructed at discrete wave frequencies using an I-optimal experimental design, with a quartic polynomial selected based on statistical and physical validation criteria. Stage 2 generalized the Stage 1 coefficients as quadratic functions of coded frequency, resulting in a unified closed-form surrogate model across the full input space. The surrogate accurately captured nonlinear geometric effects and frequency-dependent interactions without overfitting. Validation against 666 independent prediction cases, covering unseen spatial configurations and refined spectral points, demonstrated consistent predictive accuracy, with maximum errors within +/- 2 % and normalized RMS errors consistently below +/- 1 %. The model's two-tiered architecture also provided strong interpretability, revealing physically meaningful patterns in both geometric and spectral contributions. Beyond the current application, the validated framework provides a generalizable foundation for extending surrogate modeling to other geometries, operating conditions, and hydrodynamic responses.