A Hybrid Data-driven Model of Ship Roll

Abstract

A hybrid data-driven method, which combines low-fidelity physics with machine learning (ML) to model nonlinear forces and moments at a reduced computational cost, is applied to predict the roll motions of an appended ONR Tumblehome (ONRT) hull in waves. The method is trained using CFD data of unforced roll decay time series--a common data set used in parameter identification for ship roll damping and restoring moments. The trained model is then used to predict wave excited roll responses in a range of wave frequencies and the results are compared to CFD validation data. The predictions show that the method improves predictions of roll responses, especially near the natural frequency.