System identification and prediction models in ship dynamics
Doctoral thesis, 2025
This thesis investigates the enhancement of ship manoeuvring models through the
integration of prior knowledge embedded in parametric model structures and semi-
empirical formulas. The research is driven by the question: How can prior knowledge
be used to enhance the generalization of ship manoeuvring models?
The study begins with a prestudy focusing on one degree of freedom in ship
roll motion, aiming to develop parameter identification techniques and propose
a parametric model structure with good generalization. This knowledge is then
extended to the manoeuvring problem, with objectives including the development
of parameter identification techniques for ship manoeuvring models, proposing a
generalizable parametric model structure, mitigating multicollinearity, and identifying
added masses.
Methodologically, the research employs various parametric model structures for
roll motion and manoeuvring, investigated through free running model tests and
virtual captive tests (VCT). A novel parameter identification method combining
inverse dynamics with an extended Kalman filter (EKF) is proposed. Additionally, a
deterministic semi-empirical rudder model is introduced to address multicollinearity
issues.
Key findings indicate that inverse dynamics regression is an efficient method
for parameter identification in parametric models. The proposed quadratic model
structure for roll motion demonstrates good generalization, and the new parameter
identification method identifies models that accurately predict standard maneuvers.
However, challenges with multicollinearity and the need for more informative data are
highlighted. The study concludes that semi-empirical formulas can guide identification
towards more physically correct models, and VCT can provide the necessary data for
accurate model identification.
The implications of this research suggest that integrating semi-empirical rudder
models and utilizing VCT can enhance the accuracy and generalization of ship ma-
noeuvring models, contributing to more reliable and physically accurate manoeuvring
simulations.
integration of prior knowledge embedded in parametric model structures and semi-
empirical formulas. The research is driven by the question: How can prior knowledge
be used to enhance the generalization of ship manoeuvring models?
The study begins with a prestudy focusing on one degree of freedom in ship
roll motion, aiming to develop parameter identification techniques and propose
a parametric model structure with good generalization. This knowledge is then
extended to the manoeuvring problem, with objectives including the development
of parameter identification techniques for ship manoeuvring models, proposing a
generalizable parametric model structure, mitigating multicollinearity, and identifying
added masses.
Methodologically, the research employs various parametric model structures for
roll motion and manoeuvring, investigated through free running model tests and
virtual captive tests (VCT). A novel parameter identification method combining
inverse dynamics with an extended Kalman filter (EKF) is proposed. Additionally, a
deterministic semi-empirical rudder model is introduced to address multicollinearity
issues.
Key findings indicate that inverse dynamics regression is an efficient method
for parameter identification in parametric models. The proposed quadratic model
structure for roll motion demonstrates good generalization, and the new parameter
identification method identifies models that accurately predict standard maneuvers.
However, challenges with multicollinearity and the need for more informative data are
highlighted. The study concludes that semi-empirical formulas can guide identification
towards more physically correct models, and VCT can provide the necessary data for
accurate model identification.
The implications of this research suggest that integrating semi-empirical rudder
models and utilizing VCT can enhance the accuracy and generalization of ship ma-
noeuvring models, contributing to more reliable and physically accurate manoeuvring
simulations.
Manoeuvring
System identification
Inverse dynamics
Roll damping
Extended Kalman filter
Multicollinearity
HA2, Hörsalsvägen 4
Opponent: Prof. Ould El Moctar, University of Duisburg-Essen, Germany
Author
Martin Alexandersson
Chalmers, Mechanics and Maritime Sciences (M2), Marine Technology
System identification of a physics-informed ship model for better predictions in wind conditions
Ocean Engineering,;Vol. 310(2024)
Journal article
System identification of Vessel Manoeuvring Models
Ocean Engineering,;Vol. 266(2022)p. 1-17
Journal article
Analysis of roll damping model scale data
Ships and Offshore Structures,;Vol. 16(2021)p. 85-92
Journal article
Prediction of roll motion using fully nonlinear potential flow and Ikeda’s method
Proceedings of the International Offshore and Polar Engineering Conference,;(2021)p. 1670-1679
Paper in proceeding
Alexandersson, M., Mao, W., Ringsberg, J. W., and Kjellberg, M. (2025). Identification of manoeuvring models for wind-assisted ships with large rudders using virtual captive tests.
During the Apollo 13 mission, an oxygen tank explosion set the astronauts on a perilous trajectory, far from Earth. The famous quote “Houston, we have a problem” was actually “I believe we’ve had a problem here” – a huge understatement. Fortunately, NASA's "digital twin" of the spacecraft—a virtual replica—allowed engineers to simulate scenarios and solve the problem in real-time. This example demonstrates how prediction models can be invaluable, including in shipping, where they optimize efficiency and safety.
This thesis explores methods to enhance the system identification of ship manoeuvring models by integrating prior knowledge and semi-empirical formulas to improve their generalization. Generalization means that the model performs accurately not only on the data it was trained on but also on new, unseen data. This knowledge is applied to ship manoeuvring with goals including developing parameter identification techniques, proposing a generalizable model structure, addressing multicollinearity, and identifying added masses. Added mass refers to the mass of water that moves with the ship, affecting its manoeuvrability. Multicollinearity occurs when variables in a model are too closely related, which is often the case during standard manoeuvres, making it difficult to determine their individual effects and thus decreasing the model's generalization.
This thesis proposes using prior knowledge embedded in model structures, semi-empirical formulas, and additional data from virtual captive tests (VCT) to mitigate multicollinearity. The study concludes that semi-empirical formulas and VCT enhance the accuracy and generalization of ship manoeuvring models, resulting in more reliable simulations.
This thesis explores methods to enhance the system identification of ship manoeuvring models by integrating prior knowledge and semi-empirical formulas to improve their generalization. Generalization means that the model performs accurately not only on the data it was trained on but also on new, unseen data. This knowledge is applied to ship manoeuvring with goals including developing parameter identification techniques, proposing a generalizable model structure, addressing multicollinearity, and identifying added masses. Added mass refers to the mass of water that moves with the ship, affecting its manoeuvrability. Multicollinearity occurs when variables in a model are too closely related, which is often the case during standard manoeuvres, making it difficult to determine their individual effects and thus decreasing the model's generalization.
This thesis proposes using prior knowledge embedded in model structures, semi-empirical formulas, and additional data from virtual captive tests (VCT) to mitigate multicollinearity. The study concludes that semi-empirical formulas and VCT enhance the accuracy and generalization of ship manoeuvring models, resulting in more reliable simulations.
DEMOPS - Machine learning based speed-power performance modelling to reduce fuel cost and emissions from shipping
Swedish Transport Administration, 2020-01-01 -- 2022-12-31.
Swedish Transport Administration, 2020-01-01 -- 2024-12-31.
Lighthouse, 2020-01-01 -- 2022-12-31.
Driving Forces
Sustainable development
Areas of Advance
Transport
Roots
Basic sciences
Subject Categories (SSIF 2025)
Vehicle and Aerospace Engineering
Learning and teaching
Pedagogical work
ISBN
978-91-8103-186-7
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 5644
Publisher
Chalmers