Generative AI for dangerous wave environments in marine dynamics
Abstract
Ships and offshore structures are designed to perform optimally in typical conditions, and to survive the harshest conditions that will occur in their lifetime. While there is a substantial corpus of research into design-event analysis, the arrival of generative artificial intelligence (AI) has opened a new portal into describing and distilling the complex ocean environment into something controllable for the engineer to create new realistic ocean environments that can be used for extreme event observation and calculation. In this paper a generative AI method, called GenWave, is presented that shares the same principle as large-language modeling. The method learns the phase relationships in filtered wave data sets so that new wave fields can be generated on demand. GenWaveis based on recurrent neural nets, and it is cast into a bootstrapping algorithm to generate focused wave fields of increasing amplitude. The method is used to generate large waves for both a Gaussian linear water wave, and a non-Gaussian second-order water wave. The comparison of the distributions of phases for the generative AI model to those from Monte-Carlo simulation demonstrate the effectiveness of the method to create new dangerous wave fields for time-domain simulation or laboratory experiment.