If the exact outcome of an upcoming ship maneuver can be predicted, both safety can be improved and fuel consumption reduced. Within the Swedish Transport Administration’s industry program Sustainable Shipping, a research project and a preliminary study have developed new technology to achieve this. And it is needed. The standardized tests used today to determine whether a ship is maneuverable according to IMO regulations are not sufficient.

The research field of hydrodynamics is crucial for ships to operate efficiently and be steered safely. In shipping, it is usually divided into three main areas: propulsion – concerning water resistance and propeller performance, seakeeping – how the ship handles high waves and harsh weather, and maneuvering – how the ship steers and responds to steering commands.

“I have chosen to focus on the latter, that is, the maneuvering of a ship. And it is an important aspect, especially when it comes to safety. Essentially, it is about reducing the risk of collisions between ships or groundings caused by poor maneuverability. For this, simulation models or prediction models are used. There are many advantages to that. If you have a prediction model that is good enough, you can, so to speak, see into the future and know what will happen to a ship after a certain maneuver,” says Martin Alexandersson, researcher at RISE, who led the work within the research project DEMOPS (Develop Machine learning methods for Operational Performance of Ships) and the preliminary study Physics informed grey box modelling of ship dynamics.

The models can be used in two ways – as a digital twin, where the model mirrors a ship already in operation, or as a virtual prototype, which makes it possible to evaluate the ship’s design before it is built.

“In the longer term, the technology is also highly relevant for the development of autonomous ships,” says Martin Alexandersson.

Digital twins are created using sensor data about a ship’s position, speed, maneuvers, and engine output. This means that the “twin” can easily describe what has happened, but predicting what may happen is much more difficult. Therefore, research often focuses on simpler conditions – calm weather and no waves – to first grasp the fundamental dynamics before adding more complex realities.

“It is, of course, also an advantage to do this in a laboratory environment, but the idea is that our models should be usable at full scale on real ships.”

A major challenge is obtaining enough data, both in quantity and quality. Informative, varied measurements are needed to show how the ship behaves in different situations. Otherwise, the model risks becoming overfitted to old patterns – and being helpless when something unexpected happens.

“It is when the storm comes, or the black swan appears, that we really get to see if the model holds up,” says Martin Alexandersson.

In digital twin research, there are different ways to build models. A common method is the so-called black box model – a black box where you feed in data and get an answer, without knowing how the model arrived at it or whether it even follows the laws of physics. Martin Alexandersson has therefore chosen to work with physics-based models, or so-called grey box models.

“A grey box model is significantly safer when something unexpected happens that it has not seen before. It does not violate the laws of physics, whereas a black box can make up just about anything.”

The research results from the two projects show that much of the earlier published research has been based on data that was not sufficiently informative. As a result, the models have not been accurate. Many previous studies, for example, have been based on a standardized test – the so-called zigzag test, which is used internationally to show that a ship is maneuverable according to IMO rules.

“It is used in perhaps 99 out of 100 scientific articles. But the data it provides is not sufficiently informative to predict new situations. You need to supplement it with other types of maneuvers – preferably more varied or random – if the model is to work in reality,” says Martin Alexandersson.

The reports DEMOPS (Develop Machine learning methods for Operational Performance of Ships) and Physics informed grey box modelling of ship dynamics were authored by Martin Alexandersson, RISE, and Wengang Mao, Chalmers.
In collaboration with: Joakim Möller, Molflow AB; Sofia Werner, RISE; and Gaute Storhaug, DNV.