A data-driven model for nonlinear marine dynamics

The design and engineering of ships platforms that operate in the ocean environment requires understanding a nonlinear dynamical system responds according to complex interaction with wide range sea wind conditions. Time domain observation marine dynamics either experiments or high-fidelity numerical simulation tools is costly due random nature full environmental loading conditions are experienced lifetime ship platform. In this paper, data-driven method presented predict input–output relationship typical systems. A Long Short-Term Memory neural net used learn wave propagation roll section beam seas. Training data generated second-order theory volume-of-fluid computational fluid dynamics, although directly applicable by other means such as potential flow experimental measurements. cost amount apply estimated measured. results compared unseen demonstrate accuracy feasibility. • Nonlinear heave can be predicted using recurrent net. small for high model two problems. Model training negligible GPU acceleration.