Bewirb dich bis zum 15. Januar für dein deutschsprachiges Masterstudium oder das Studienvorbereitungssemester International an der TU Hamburg.

i³-Project: Predicting Ship Hydrodynamics to Enable Autonomous Shipping: Nonlinear Physics and Machine Learning

Autonomous ships are seen by the maritime industry as key to improve shipping efficiency and safety in the future. Fully autonomous shipping requires significant developments within ship design, navigation and control systems and operational modes combined with the ability to assess and verify the safety and performance in a credible manner. One key aspect on the development path to autonomous shipping is the very complex problem of autonomous manoeuvring in waves in order to ensure collision avoidance, ship safety, economical efficiency and general environmental impact. At fully automation, the ship must take over human decision making by translating environmental parameters such as ship speed, surrounding sea state and other ships into operational decisions in real-time. Moreover, structural consequences have to be taken into account to identify optimal speed and course.

This project addresses one aspect of the very complex overall problem of ship hydrodynamics in waves and the associated structural consequences based on near field information such as radar snapshots of surrounding waves. For fully autonomous shipping, radar information have to be translated into consequences for the ship or structure as well as operational conditions in real-time. At this, consequences of alternative operating decisions have to be evaluated in order to identify the optimum.

The physical complexity with respect to classical numerical methods represents hereby a huge technical barrier as the required highly accurate simulation results can only be obtained at the expense of computational time, hindering a real-time application. Machine learning (ML) can bridge the gap by enabling the efficient processing of complex, specific tasks without using explicit instructions or physical models. For this purpose, efficient and accurate numerical methods are required as only such methods are applicable for the generation of training data for end-to-end ML models to directly predict motions and loads based on radar snapshots in real-time.

Partners: M10 (Franz von Bock und Polach) and M14 (Norbert Hoffmann)

Duration: 2022-2024