A Generative AI Model for irregular wave conditions that induce parametric roll

Parametric roll is a nonlinear and intermittent dynamical phenomenon that causes damage to or loss of cargo, and poses significant risks for passenger safety and comfort. The current engineering practice to evaluate susceptibility or likelihood of parametric roll requires hours of nonlinear simulation or model tests in many wave conditions from the long-term operation wave scatter diagram. In this paper a novel generative AI method is proposed to learn the specific conditions that lead to parametric roll without the need to perform exhaustive and time consuming Monte Carlo simulations of all wave conditions. The method uses a low fidelity hydrodynamics model to generate the required large set of training data. A distillation process is employed that produces a series of models that can generate new wave conditions that lead to increasingly severe parametric roll behavior. Then the exceedance probabilities derived from the generated wave conditions are shown to closely match those from Monte Carlo simulations, while requiring significantly fewer computations. Finally, the generated wave conditions are verified by performing simulation with a weakly-nonlinear IRF method, confirming the ability of the method to find conditions that lead to pronounced parametric roll responses at a small fraction of the cost relative to the current practice of using Monte Carlo simulation.