Physics-guided ML to build digital twin for wind-assisted propulsion ships

Research Project, 2025 – 2026

Wind-assisted propulsion (WAP) technologies, which use renewable wind energy for ship propulsion, can reduce more than 20% emissions from shipping. Thanks to many successful R&D projects dedicated on developing physical/generic models to optimize WAPs design and evaluate their payback time, the installation of WAPs is increasing dramatically over the past years. However, installation of WAP technologies onboard ships can significantly change a ship’s operational characteristics, such as the manoeuvrability, energy performance and even seafarers’ ability and attitude to navigate these ships. Furthermore, the physical models developed for WAPs design/evaluation are not capable to be utilized to assist WAPs operations, which require accurate ship-WAP coupling performance prediction for real-time individual sailing conditions, but physical models are static and contain large uncertainties due to large simplification/assumptions.

This project aims at developing a physics-guided machine learning architecture (PML), which can combine benefits of both physical ship performance models and operational ship monitoring data, leading to more accurate prediction of ship-WAP dynamic coupling performance in real-time. The PML model will be further utilized to develop optimal control mechanism to increase automation of WAP operations. Finally, various sailing scenarios based on the PML model will be developed to train and study the seafarers’ behavior to deploy WAP technologies during the navigation of ships.