Data-driven approaches to predicting six-degree-of-freedom ship motion under irregular sea states

Abstract Accurate prediction of six-degree-of-freedom ship motions in irregular seas is vital for seakeeping, stability, and operational safety. This study introduces a hybrid Physics-Informed Neural Network with Long Short-Term Memory (PINN-LSTM) model integrating hydrodynamic laws with temporal deep learning for robust, physically consistent predictions. Experiments in wave basins under varying periods, headings (0°, 45°, 90°), and wind forcing (30 m/s) achieved high accuracy ( R 2 > 0.92). Pitch accuracy peaked in head seas ( R 2 = 0.9349), while roll dominated beam seas ( R 2 = 0.9332). Errors were minimal (RMSE: 0.0220–0.0545). Findings confirm PINN-LSTM’s applicability for real-time navigation, design optimization, and offshore operations.