Ships can be operated more efficiently by utilizing intelligent decision support integrated with onboard data collection systems. In this study, a Bayesian optimization-based decision support system, which utilizes ship performance models built by machine learning methods, is proposed to help determine the operational set-points of two engines for double-ended ferries. By optimizing the ferries’ power allocation between the stern and bow engines, the Decision Support System (DSS) will simultaneously attempt to keep the ETA of the ferry fixed under a set of operational constraints using the Bayesian optimization. Its objective is to minimize fuel consumption along individual trips. Based on simulation environment, the DSS can reduce at maximum 40 % fuel consumption with no significant change of the ETA. Final full-scale experiments of a double-ended ferry demonstrated an average of 15 %, where at least half of this saving was achieved by the optimized power allocation between bow and stern engines. 

System identification offers ways to obtain proper models describing a ship’s dynamics in real operational conditions but poses significant challenges, such as the multicollinearity and generality of the identified model. This paper proposes a new physics-informed ship manoeuvring model, where a deterministic semi-empirical rudder model has been added, to guide the identification towards a physically correct hydrodynamic model. This is an essential building block to distinguish the hydrodynamic modelling uncertainties from wind, waves, and currents – in real sea conditions – which is particularly important for ships with wind-assisted propulsion. In the physics-informed manoeuvring modelling framework, a systematical procedure is developed to establish various force/motion components within the manoeuvring system by inverse dynamics regression. The novel test case wind-powered pure car carrier (wPCC) assesses the physical correctness. First, a reference model, assumed to resemble the physically correct kinetics, is established via parameter identification on virtual captive tests. Then, the model tests are used to build both the physics-informed model and a physics-uninformed mathematical model for comparison. All models predicted the zigzag tests with satisfactory agreement. Thus, they can indeed be considered as being mathematically correct. However, introducing a semi-empirical rudder model seems to have guided the identification towards a more physically correct calm water hydrodynamic model, having lower multicollinearity and better generalization.

It is essential to describe a ship’s manoeuvrability for various applications, e.g. optimal control of unmanned surface vehicles (USVs). In this study, the capability of two recognised manoeuvrability models to predict ships’ trajectories is investigated based on both simulation and open-water experiment test data. The parameters of these models are estimated by a statistical learning method. The goodness of the two estimated models for describing a merchant ship’s manoeuvrability is first studied using her manoeuvring simulation data. Then, experimental manoeuvring tests to use a USV in open water with wind and drifting effects are used to check the conventional model identification procedures. Finally, some modifications and adjustments are proposed to improve the conventional procedures. It shows that the proposed procedures can accurately derive the ship’s manoeuvrability based on experimental data. © 2022 The Author(s). 

Fuel is one of the highest cost items while operating a ship, and its combustion results in air emissions polluting environments. Finding ways to increase shipping operations efficiency without compromising the provided service quality is necessary for economic and environmental reasons. This study first used data analysis to find hidden information in one-year navigation data of a double-ended ferry operated along the Swedish coast. The case study ferry was operated using both bow and stern engines partly loaded. A new feature of the power ratio is defined to describe the influence of engine power allocation on total fuel consumption. Then, different machine learning methods are used to establish the ship’s total fuel consumption model due to influences of external factors such as wind and sea currents, etc., together with the power ratio. The established machine learning model is used to find the most efficient operation of allocating power to different engines. It shows that, in theory, up to 35% fuel savings can be achieved for the case study vessel. These findings can further aid with the operational planning for the scope of Eco-driving.Double-ended ferries are an alternative to bridges or tunnels for transporting passengers and cars over water. They are used for commuting in big cities like New York (Siferry 2022), London (TRL 2022), and connecting islands along the coast. Double-ended ferries can achieve this task on short routes where maneuverability may be difficult (Waterhouse, 2016), and relieve road congestion (Leung et al., 2017). The objective in maritime transport is to reduce CO2 emissions by at least 40% by 2030 and to pursue efforts to reach 70% reduction by 2050, compared to the 2008 setup by IMO (2018). Maritime authorities also push regional ferries to become more environmentally friendly and save fuel costs. It becomes important to investigate how to operate these vessels to reduce energy consumption. For this purpose, a fundamental problem is establishing a reliable performance model to describe those ferries’ fuel consumption in terms of their operational profiles, such as allocation of engine load, ship speed, etc. Different models have been researched in the maritime community to address this problem.

It is common today that operational data is recorded onboard ships within the Internet of Ships (IoS) paradigm. This enables the possibility to build ship digital twins as digital copies of the real ships. Predicting the ship’s motions with ship dynamics could be an important sub-component of these ship digital twins. A model for the ship’s dynamics can be identified based on observations of the ship’s motions. The identified model will have model uncertainty due to imperfections and idealizations made in physical model formulations as well as uncertainty from errors in the measurement data, which can be very pronounced when using full scale operational data. It is easier to develop accurate models with low model uncertainty using data obtained in a controlled laboratory environment where the measurement errors are much lower, especially in calm water conditions. The prediction model should be able to describe scenarios that a ship has never encountered before, which is possible if much of the underlying physics has been identified. Grey-box modelling is a technique which combines operational data with physical principles to achieve this. The objective of this thesis is to develop system identification methods for grey box models with good generalization of the model scale rigid body ship dynamics in calm waters. A model development procedure is proposed to handle the model uncertainty through the selection of candidate models based on a hold-out evaluation procedure. The measurement noise is handled by an iterative preprocessor, which uses an extended Kalman filter (EKF) and a Rauch Tung Striebel (RTS) smoother that uses an initially estimated predictor model from semi-empirical formulas. It is demonstrated that the ship’s roll motion with high accuracy can be described using a quadratic damping model. For the more complex manoeuvring models, multicollinearity is a large problem where the appropriate complexity needs to be selected with the bias-variance trade-off between underfitting or overfitting the data. Hold-out turning circle tests were predicted with high accuracy for the wPCC and KVLCC2 test case ships with models from the proposed development procedure and parameter estimation method. The proposed methods can produce prediction models with high generalization given that a suitable model structure has been selected from the candidate models and an appropriate split in the hold-out evaluation of the model development process has been applied.

Identifying the ship's maneuvering dynamics can build models for ship maneuverability predictions with a wide range of useful applications. A majority of the publications in this field are based on simulated data. In this paper model test data is used. The identification process can be decomposed into finding a suitable manoeuvring model for the hydrodynamic forces and to correctly handle errors from the measurement noise. A parameter estimation is proposed to identify the hydrodynamic derivatives. The most suitable manoeuvring model is found using the parameter estimation with cross-validation on a set of competing manoeuvring models. The parameter estimation uses inverse dynamics regression and Extended Kalman filter (EKF) with a Rauch Tung Striebel (RTS) smoother. Two case study vessels, wPCC and KVLCC2, with very different maneuverability characteristics are used to demonstrate and validate the proposed method. Turning circle predictions with the robust manoeuvring models, trained on zigzag model tests, show good agreement with the corresponding model test results for both ships. © 2022 The Author(s)

Having an accurate prediction of ship roll damping is crucial when analysing roll motions. In this paper, the simplified Ikeda method (SI-method) is compared with the original Ikeda method. The methods are compared using results from a database of roll decay tests carried out on modern merchant ships and a smaller set of predictions in which the original Ikeda method was used. It was found that most of the ships in the database had dimensions outside the limits of the SI-method. Thus, the SI-method showed poor agreement with model tests outside its limits but acceptable agreement for ships within limits. It was found that the deviations were caused by extrapolation errors of the wave-damping in the SI-method. Two ways to improve the accuracy of the SI-method were proposed based on regression, which gave about the same accuracy as the original Ikeda method. © 2021 The Author(s).

Getting the best possible accuracy with the lowest possible computational cost is an important factor in the early design stage of ships. Potential flow-based analysis presents such a solution for seakeeping analyses. The accuracy of roll motion in potential flow is however not so good, due to the large influence from vicsous roll damping, which is missing in these calculations. This paper proposes a hybrid method, as a solution to this problem, where the viscous roll damping from Ikeda’s semi-empirical method is injected into an existing 3D unsteady fully nonlinear potential flow (FNPF) method. The hybrid method is investigated using roll decay tests with the KVLCC2 test case. This investigation shows that the accuracy of simulated roll motions is significantly improved and also shows good agreement with the corresponding roll decay model tests.

When designing a hull that needs to account for sails, either rigid or flexible, it is necessary to consider the larger leeway and heel angles deriving from the sails side-forces compared to a traditional ship. It is therefore necessary to explore the possibility of adding appendages to the hull to balance those forces, achieving an optimum trade-off between hydrodynamic efficiency and manoeuvrability. The possibility of numerically simulating the manoeuvre coefficients at design stage will increase the chances of understanding the behaviour of a ship from an early stage in the design process. The current research is based on the evaluation of hydrodynamic efficiency and manoeuvre coefficients of a hull with rudders and shafts in a pure resistance, self-propulsion and in a wind-assisted mode. Having assessed the performances of the vessel with a Velocity Prediction Program (VPP), an in-depth research on suitable appendages was performed to reduce the experienced leeway angle, and ultimately increase the performances of a wind-assisted ship, especially when subject to wind angles ranging between 40 and 80. Many CFD simulations were initially performed to assess the hydrodynamic characteristics of the hull in a range of flow directions, rudder angles, ship speed and combinations. Those simulations encompass the whole range of datapoint needed to describe the forces and moments acting on a wind-assisted ship, simulating a towing-tank captive test, namely performing a Virtual Captive Test (VCT) [1]. Three types of possible appendages used to increase the generated side-force of the wind-assisted vessel are further investigated and the advantages and disadvantages are described. The findings of those preliminary simulations are then used as a basis for a structured model testing campaign.

Hybrid testing is an experimental technique that can be used to test ships and marine structures when both hydrodynamic and aerodynamic effects are important, for example for wind powered or wind assisted ships and sailing vessels. SSPA is currently developing an experimental method using hybrid testing involving fan forces added to ship decks to simulate sails to assess the course keeping, seakeeping and manoeuvring performance of a wind powered ship. For conventional motor ships there are well established test methods and knowledge on how to scale the results from model to full-scale. For a wind propelled ship however, the driving force is no longer located at the propeller shaft but high above deck and at another longitudinal position that could vary with true wind angle and speed. Moreover, there is a large side force coming from the aerodynamic forces of the wingsails that needs to be counteracted with lifting surfaces underwater. The side-force and yaw moment are much more prominent than in conventional vessels. The combination of those factors will influence the manoeuvrability and course keeping, especially in waves. Having built up the research tools for predicting and simulating the behaviour of a full-scale vessel, making the model sail in a similar way as predicted for the full-scale vessel remains a challenge because of the difference between Froude scaling and Reynolds scaling applicable for the hull and lifting surfaces respectively. Using Computational Fluid Dynamics (CFD) to understand the scale effects in model tests for a wind powered ship and developing a methodology for determining the fan parameters that correctly model the ships behaviour and performance are the key objectives of the research study. The art of model testing encompasses the need to learn from different techniques to ultimately achieve the best agreement between model tests and full-scale results in terms of accuracy, repeatability, cost, and speed. Learning from preliminary experimental tests, through studies on CFD and ultimately paving the way to new testing methodologies is the main aim of the current paper.