Using the wind as an additional propelling force for modern cargo ships is on the rise. This is due to the regulations about greenhouse gas emissions. Various wind units are available to create a propelling force for the ship. One of these wind units is the suction wing. A couple of ships are equipped with it, like the Amadeus Saffier. Some reasons for choosing a suction wing are the compactness, controllability from the bridge and the ability to create higher lift than traditional sails.
The suction wing in itself has already been researched. Next to that, the optimal position in terms of counteracting its forces with the ship’s hydrodynamic forces is investigated. In the middle of these two lies the gap of the actual forces and interactions between the ship and the suction wing at sea.
This research investigates the interaction effects between the ship, deckhouse and wind units and how they influence the forces. The goal is to get insight into the interaction effects and their consequences.
Based on that, the suction wings can be positioned on the ship with minimal interaction and on fewer disturbed regions. Moreover, knowing the suction wing forces, which are expected to be generated at sea, leads to better estimations of the engine power reduction and savings in emissions. Additionally, this knowledge can be applied to design the ship and suction wings more effectively for use at sea on a wind-assisted vessel. All of this helps to optimise the wind-assisted ships and reduce emissions.
Computational fluid dynamics (CFD) simulations with the RBVMS (residual-based variational multiscale) code were conducted to get insight into the flow. A simple, real-size model of a ship hull, deckhouse and suction wing was created for the CFD simulation. The ship and suction wing were simulated separately, but also in a combined case. Results were obtained for two wind conditions, respectively a uniform wind of 10 m/s and a logarithmic wind profile in an apparent wind angle of 30∘. Reliable results were achieved based on a mesh size study for the individual cases. Additional studies were done for the suction wing to get reliable results. These included a domain size study and the calculation in 2D and 3D. Moreover, a realistic suction pressure and blowing velocity were identified and matched for each wind condition. The lift and drag coefficients were validated with experiments.
The 2D and 3D simulations of the suction wing showed that the 3D case is not an extrapolation of the 2D case. In the 3D simulation, the lift-induced drag plays a strong role, which is produced by the tip vortices.
The interaction effects were investigated with vector plots, the lift and drag coefficient and frequency spectra. The frequency spectra were realised through a polynomial fit of the forces and the Discrete Fourier Transform. The results showed a strong interaction between the ship and the suction wing. Especially close to the ship deck, more turbulence and a change in wind speed can be seen. For the simulated cases, the deckhouse stayed unaffected.
Superposition was investigated to determine if it is possible to have a faster way of knowing the wind above the ship deck at different angles and ship configurations at sea. Linear superposition was applied to the single simulations to compare them with the full simulation. Linear superposition can not represent the full simulation. Nevertheless, it can be used to get a better approximation of the flow behaviour of the full simulation than using only one single simulation.

Turbulent flows can be found in many industrial applications, and have a profound impact on the drag resistance and heat transfer characteristics of engineering equipment. While the behavior of turbulence is well-documented for canonical flow cases, its behavior under complex conditions is largely unknown. This poses a significant challenge when designing engineering equipment, since it is essential to accurately predict the impact of variable property flows and surface roughness on pressure losses and heat transfer rates. Yet such practical cases are beyond the scope of canonical flow data. During recent years, the combination of advances in machine learning and GPU-accelerated flow solvers has yielded promising results in the fields of turbulence and scalar transport. Using a few modern GPUs and state-of-the-art algorithms, it is now possible to simulate highly complex turbulent flows in short periods of time. This has enabled further progress in the field of machine learning for fluid mechanics, since high-fidelity turbulent flow databases can be easily generated, and new models can be trained to predict increasingly complex flow effects accurately. In this project, multiple challenges are addressed, such as optimizing GPU-accelerated DNS solvers for extreme-scale simulations, creating data-augmented RANS turbulence models for flows with strong variations in their thermophysical properties, and developing machine learning models for turbulent flows past rough surfaces.

Prior to working in GPU-accelerated simulations, the sub-project about machine learning for variable property flows focused on building data-augmented RANS turbulence models using a technique known as FIML (Field Inversion Machine Learning). In this framework, non-linear optimization is first performed to obtain an ideal set of corrections, after which a neural network (or another model) is trained to predict the observed corrections. When working with variable-property flows, several challenges arise, such as not knowing beforehand the local fluid properties due to thermal changes. This makes it difficult, for instance, to calculate the input features for the neural network. The solution created is a feedback loop, where CFD predictions are used to re-calculate the local fluid properties and to update the neural network input features, etc. The results show that the data-augmented RANS model can accurately improve the predictions for unique flow cases, which need unusually high corrections not observed in the training set. Additionally, a weighted relaxation factor methodology is proposed, to ensure convergence of the RANS models after inserting neural network corrections. The final results show that, for the most challenging CFD case identified, our machine learning system is able to reduce the L-infinity error on the velocity profile from 23.4% to 4.0%.

To generate the required amounts of data for machine learning studies regarding complex geometries like roughness, it was necessary to develop a GPU-accelerated DNS solver. This work focuses on the implementation of a parallel tridiagonal solver for extreme-scale simulations, and the creation of a new cross-platform communication library for supercomputers with either AMD or NVIDIA GPUs. In general terms, turbulent flow simulations in GPUs can be highly efficient, since all operations can be mapped to different GPU threads. However, large-scale data transfers are the main performance bottleneck of GPU-based simulations. While halo exchanges between GPU sub-domains have a minimal impact, the large-scale transpose operations needed for 3D arrays inside Poisson/Helmhotz solvers occupy most of the total running time. Therefore, any transpose operation avoided will drastically improve the running times for the entire DNS solver. By using a parallel tridiagonal solver, it is possible to reduce the number of transposes (for full 3D arrays) by 50% for 2D pencil decompositions inside the Poisson/Helmhotz solvers, and to replace these avoided transposes by simplified operations with a computational cost resembling halo exchanges. Additionally, in this work, a new opportunity to speed up simulations was found, by re-deriving the coefficients of parallel tridiagonal solvers to substantially improve the efficiency and the GPU parallelization of the DNS solver for implicit 1-D diffusion equations. Based on these improvements, it is observed in the results that the efficiency of the DNS solver is substantially improved for extreme-scale simulations in the LUMI and Leonardo supercomputers. Moreover, we show that the entire DNS solver can operate using 2D pencil decompositions without performance degradation compared to most optimal 1D decompositions available for smaller systems. Therefore, the parallel tridiagonal solver enables high-performance in extreme-scale simulations, where 1D decompositions are not feasible.

To enable extreme-scale simulations in AMD GPUs, a new cross-platform communication library was created, named diezDecomp. This library is able to achieve high performance working with both CPUs and GPUs in NVIDIA or AMD-based supercomputer. The underlying algorithm corresponds to an advanced implementation that works by directly intersecting the x/y/z bounds of all MPI tasks and scheduling data transfer operations. This allows the implementation of any-to-any transpose operations between mismatched 2D pencil decompositions, with complex communication patterns beyond the scope of traditional all-to-all operations. In extreme-scale simulations, direct x-to-z transposes can improve efficiency while solving for implicit 1D diffusion, but they are not available in existing libraries. Thanks to the flexibility of the diezDecomp library, x-to-z transposes can be easily implemented, and the running times of implicit 1D diffusion solvers were improved up to 55% for extreme-scale simulations in the LUMI supercomputer with 1024 GCDs.

The benefits of machine learning to predict the thermal and hydrodynamic behavior of turbulent flows past rough surfaces are explored in Chapter 4. Due to the complexity of this task, a convolutional neural network was used to (independently) scan the input height maps of rough surfaces, and to generate detailed 2-D maps with the local skin friction factors and Nusselt numbers. The proposed neural network is optimized to have linear time complexity while creating 2D maps, instead of quadratic complexity as in naive approaches. The validation study using randomized surfaces with the Fourier spectrum of grit-blasted surfaces shows that machine learning can make accurate 2D predictions for both the local skin friction factors and Nusselt numbers of rough surfaces, with median deviations of 28.43% for the skin friction factors and 6.37% for the Nusselt numbers respectively. The averaged errors in the predictions for skin friction factors and Nusselt numbers were reduced from 24.9% and 13.5% using traditional correlations to only 8.1% and 2.9% thanks to machine learning.

Since the neural network predictions for the thermal behavior of rough surfaces were particularly promising, a further optimization study is presented in Chapter 5. Here, the objective is to combine the benefits of machine learning and GPU-accelerated simulations to improve convective heat transfer in dimpled-surfaces. Remarkably, it was found that machine learning can find a highly optimized dimpled-surface with a 53% higher Nusselt number using cross-aligned dimples after being trained with only random surfaces (displaying lower thermal performance). After the second iteration of the reinforcement learning loop, it was confirmed that the surface found by machine learning created a flow pattern with helical structures inside the dimples. All other tested parameters had lower thermal performance.

In summary, it is concluded that GPU-based DNS solvers can be optimized to enable extreme-scale simulations with minimal performance degradation. Creating highly flexible communication frameworks, such as the diezDecomp library, is also possible while keeping identical running times as traditional libraries. This research project also showcases how machine learning can be an effective tool in fluid mechanics, to predict the behavior of turbulent flows past complex geometries or to account for changes in flows with strong variations in their thermophysical properties.

This thesis explores fast simulation of time-periodic flows, characterized by behavior that repeats at regular intervals. These flows are present in both natural and engineered systems. Examples include flows past wind turbines, in the heart and arteries, and around propellers. Due to their long time domains, time-periodic flows pose challenges in both experimental and numerical studies.
Many engineering tasks, like optimization, require numerous model evaluations across a wide range of inputs. We need a fast and accurate model to directly interact with the model in the design process. To contribute to this goal, we first develop a high-fidelity method specifically designed for time-periodic flows. Using this model, we then create a time-periodic reduced-order model to enhance simulation efficiency.

In the high-fidelity model, we employ the isogeometric analysis framework to achieve higher-order smoothness in both space and time. The discretization is performed using residual-based variational multiscale modelling and weak boundary conditions are adopted to enhance the accuracy near the moving boundaries of the computational domain. We enforce the time-periodic boundary condition within the isogeometric discretization spaces, which converts the two-dimensional time-dependent problem into a three-dimensional boundary value problem. The motion is known a priori and we restrict ourselves to two spatial dimensions. Application of the computational setup to heaving and pitching hydrofoils displays very accurate and exactly periodic results for lift and drag.

We use the high-fidelity model to develop a POD-Galerkin reduced-order model, which retains inherits the features of the high-fidelity model while reducing the number of variables in both space and time. We evaluate the reduced-order model with numerical experiments on moving hydrofoils. Reduced-order model solutions agree well with those of the high-fidelity model. The errors over the entire time period of the computed forces are less than 0.2%. Our time-periodic reduced-order model offers speed-ups ranging from O(10^2) - O(10^3) compared to the full-order model.

The non-linear nature of the Navier-Stokes equations creates a computational bottleneck in the reduced-order model. We explore hyper-reduction techniques to mitigate these challenges. We focus on empirical interpolation methods, which have shown promise in reducing the complexity of non-linear operators. The model performed well for the experiment with steady flow, with force and solution errors ranging from O(0.01% − 10%), depending on the sampling method. The hyper-reduced model achieves a speed-up of O(10^5) compared to the full-order model, and O(10^3) compared to the reduced-order model. This enables real-time computations with direct interation of the user. However, the hyper-reduced model could not provide a solution for the time-periodic flow experiment.

To show the applicability of the reduced-order model to an industrial problem, we apply the the model to a vertical axis wind turbine. Vertical-axis wind turbines offer significant advantages for urban applications over conventional wind turbines due to their lower noise levels. However, their performance is highly sensitive to various factors requiring high-fidelity simulations to optimize its performance. The model was used to determine the optimal operating point of the turbine, maximizing energy production per cycle under the given conditions. It was observed that the turbine's energy output was negative, likely due to the low Reynolds number (Re = 1000) used in this study.

Future research can expand on this thesis in several ways. Extending the model to three spatial dimensions would require solving four-dimensional boundary value problems, potentially benefiting from mesh adaptivity techniques. Further exploration of hyper-reduction methods, such as empirical quadrature procedures or neural network-based models, could enhance the model's efficiency. Applying hyper-reduction directly within isogeometric discretizations may also offer significant advantages. Additionally, further verification for higher Reynolds numbers and adaptation to other industrial applications, like wind farm and ship propeller optimization, could bridge the current theoretical advances with practical use. Finally, using time-periodic data from standard models to develop time-periodic reduced-order models could expand their applicability.

Introduction - Hydrofoils are often used for high-performance craft. They enable the hull to rise out of the water and reduce drag. A recurring problem is ventilation, this is the presence of a cavity filled with ambient air on the lifting surface. Ventilation causes the lift of the hydrofoil to decrease drastically. A lot of research has been done on ventilation behaviour. Ventilation remains a stochastic phenomenon to this day. The same operational conditions do not yield the same flow behaviour, a correlation between the pressure on the lifting surface and ventilation behaviour might exist. This research is an intermediate step towards finding a correlation between pressure on the surface of a foil and ventilation, making ventilation repeatable. The main goal of this research is to find out if a full surface pressure reconstruction can aid the search for a correlation between pressure and ventilation.

Method - The test geometry is a NACA0012 surface-piercing hydrofoil with pressure measured using 2 rows of 15 pressure tappings. Extension pieces are used to obtain the span-wise pressure distribution while keeping the submerged span constant. The test program has three sets: all vertical runs (Set 1), high roll angle runs (Set 2), and runs with ventilation (Set 3). During the experiment, the loads are measured with force transducers on a frame that measures forces in 6 degrees of freedom. The calibration of the force transducers is performed prior to the experiment. The calibration of the pressure sensors is done first using a 9 meter water column before the experiment and during the experiment in the towing tank, extra re-calibration (stepped runs) data is collected. The collected force measurement data is processed to attain the lift and drag coefficients. The pressure measurements are translated to the pressure coefficient and placed in a matrix on the right location of the lifting surface.

Results - Based on repeated experimental runs the percentage differences are obtained on sensor and array levels. 30% of the sensors have a percentage difference of more than 10%, while the maximum difference for an entire array is 2.5%. A cross-check between the upper and lower arrays showed a percentage difference above 15% for 28% of the sensors, with a maximum difference of 1.2% at the array level. The pressure distributions at 0∘ roll angle, from Set 1, match the general pressure distribution characteristics. Comparing the 3D lift coefficient computed from the pressure measurements to that of the force data or an empirical method yields ambiguous results. The 3D lift coefficient based on the pressure measurement is within a 4% difference to Xfoil. The result of the high roll angle runs, Set 2, show the effect free-surface proximity has on the chord-wise pressure distribution. For the shallowest pressure reconstructions, close to the free-surface, the pressure distribution shows a minor peak near the leading edge after which the curve drops to near zero quickly. The operational conditions do not affect the quality of measurements. During two runs ventilation occurred, this is Set 3. The time-traces of the three forwardmost sensors show a dip prior to ventilation, with one sensor displaying an oscillating response. It is hypothesised that this is the position where ventilation is induced, but based on the results this can not be proven. All sensor response follows the same pattern when the ventilation bursts over the surface. The set-up is able to capture quick pressure changes.

Conclusion -The percentage difference is too high for too many sensors. On the array level, however, the difference is considered small enough. The results of set 1 are in good agreement with known data. The results obtained at 60∘ roll angle are concluded to be of the same quality as for 0∘ roll. The method is capable of providing results in a wide range of operational conditions. Based on the ventilation runs, no correlation to pressure reading has been found. The time-traces of the sensors near the leading edge of these runs do show fluctuations prior to ventilation. With more ventilation runs available, it is highly likely that a correlation can be established. Due to the slim data set, it is possible that the conclusions drawn will be refuted in the future when a larger data set is available. Currently, the goal has not been achieved, and it cannot be stated with certainty that the method used here to attain a full surface pressure reconstruction can be used to correlate pressure to ventilation.

This research thesis focuses on conducting experimental research into the hydrodynamic properties of a NACA 0012 T-foil in dynamic conditions. An aerodynamically smooth surface was created and used for all dynamic experiments. For exploratory research into the effect of surface finish on the performance of a T-foil in steady state conditions, a second, unfinished but structurally equal, T-foil was made using .15 mm 3D printing layers aligned with the flow.
The hydrodynamic properties were determined by conducting experiments in steady state conditions using conventional towing tank methods. Combining all steady state experiments, the lift and drag coefficient on the foil are expressed as a function of angle of attack, freestream velocity and submergence for a Reynolds number of 1.5·105. The surface finish of the second, raw printed, T-foil resulted in a lower critical Reynolds number of 3.3 · 105 compared to 4.5 · 105 for the aerodynamically smooth foil.
Dynamic flow conditions were simulated in experiments where the setup was subjected to either: sinusoidal pitch motion, sinusoidal heaving motion or waves. As a result of the imposed motions, the static forces on the setup change and inertial forces occur. To compensate for this, a function was made from all static measurements. The inertial forces are compensated by applying a mass/inertia matrix.
Using this new method of towing tank research, the steady state behaviour of a T-foil is described using dynamic experiments. The novel method to determine steady state characteristics includes two angle sweeps, one positive and one negative. The rate of change proves to be sufficiently slow at 0.8 ◦ s−1 for v∞ = 2 ms−1, producing more accurate results while towing tank time can be reduced by 78%.
From the steady state results a model is created. Because this model is limited to Re = 1.5 · 105, it is enriched with a prediction model based on X-foil which is compensated for induced drag, finite foil shape and free surface effects.
The prediction shows good correlation with the results in z-direction for both imposed sinusoidal heaving and pitching motions. During sinusoidal heaving, flow remains attached longer leading to higher and more stable lift and drag at higher absolute angles. In pitching a similar effect can be seen leading to increased fluctuations and slightly higher average lift compared to the prediction models. For both imposed motions, lower drag is measured with respect to both the predictions and the measurement in steady state conditions. An additional component to the drag occurs during a sinusoidal pitching motion. This additional component has a sinusoidal behaviour equal to the imposed motion which suggests that it is caused by induced drag.
When the freestream velocity was increased from 2 to 6 ms−1 in similar waves, it was found that the sinusoidal response of the hydrodynamic forces shifted by ± π rad. Similar to the imposed dynamic motions, predictions show larger values for drag. It was found that the wave height can be related to the drag. When wave height is increased, the drag is reduced.

Wind-assisted ships are installed with wind propulsors capable of generating thrust for aiding the propulsion of the ship, with the purpose of saving fuel by reducing the required propeller speed and thus engine power. Wind-assisted ships are becoming more common due to high fuel prices and rising environmental concerns. The subject of this thesis is wind-assisted ship maneuverability, which is a relatively new and unexplored topic. There are many factors which this study can involve, and so the goal of this thesis is to study the general behavior of wind-assisted ships during the Turning Circle and Zig-Zag standard maneuvers, and investigate in depth the effect of different control optimization schemes for the wind propulsors. This study requires building a 4 degree-of-freedom (DOF) time domain maneuvering model in which the equations of motions for a wind-assisted ship are solved and the required parameters are input in order to execute the standard maneuvers. The model includes calculation of all the major forces on the ship, including hydrodynamic forces common among maneuvering models, and with the addition of aerodynamic forces due to the wind propulsors. Once the model is built and shows sufficient accuracy, the maneuvers are simulated for a variety of wind conditions, sail configurations, and wind propulsor trim optimization schemes. The behavior of the ship due to these different conditions and the compliance with maneuvering standards are discussed.

The results of the simulations show that the maneuvering behavior of the ship is indeed significantly dependant on the wind, and also how the wind propulsors are controlled. Based on the results, in general, the ship shows the behavior to turn quickly into the wind and slowly while turning against the wind. The effect this has on the maneuvers is that the Turning Circle is generally tighter when the initial turn is turning toward the wind, and larger when the ship is initially turning away from the wind. For the Zig-Zag maneuvers, the ship is alternating in turning toward and away from the wind and so the behavior of turning faster toward the wind is evident in the measured quantities of the Zig-Zag maneuver. The standardized requirements on maneuvering do not yet have considerations for wind-assisted ships and thus it is unknown what will be allowed in the future in terms of wind propulsor control during the standardized tests. Additionally, the abilities of the ships and wind propulsors when it comes to active optimization and wind propulsor trimming are not well known. Thus in this thesis a variety of options for wind propulsor control are studied in order to provide some insight on what is possible, covering some more conservative and more aggressive approaches. The wind propulsor optimization shows to have a big effect on the ship maneuverability, and with thoughtfully constructed optimization schemes there is a general improvement in the measured quantities that result from the maneuvers.

Implementation of an Explicit Algebraic Reynolds Stress Model in Isogeometric Analysis for seperating flows using a two-equation k-omega model. The model is verified for simple cases such as homogeneous isotropic decaying turbulence, and the turbulent backward facing step

Delft University of Technology has access to a towing tank where a wave maker can generate waves. When generating waves, non-linear start-up effects occur at the start of a wave train which affect the rest of the signal.
The aim of this thesis is to develop a model which can take these effects into account. A numerical modelling approach is adopted, due to the necessity for generating a large space of wave profiles and the need to optimize towards a wave maker input given a wave profile.

A clear assessment of the state-of-the-art revealed a scope of present methods to model non-linear free surface waves. A potential flow assumption proved a clear balance between accuracy versus computational cost. Analysis of this model showed that the state is described by a balance between potential and kinetic energy and should, in principle, be conserved.
For numerical modelling, the Finite Element Method is adopted. The sparse matrices and ease of evaluation of the integrals allow for a better adaptation in present computer architectures. Additionally this method provides more rigorous mathematical tools to demonstrate properties such as stability and convergence. Lastly, the use of Isogeometric Analysis where the solution space is described by splines instead of polynomials could provide an advantage over conventional methods with respect to continuity, convergence and refinement strategies. Subsequently, literature revealed that the potential flow model in conjunction with FEM will result in model that is accurate, stable and fast.

A novel numerical model is presented where the spatial discretization is done using IgA and the resulting semi-discrete Ordinary Differential Equation is integrated in time with a separate method. Analysis shows that this spatial model inhibits the same energy conservation laws as the physical model.

Implementation is done in the DelFI, an in-house fixed domain Navier-Stokes solver with an Open Source back-end named MFEM which utilizes clever FEM abstractions and parallelization. To successfully implement the non-linear problem, first a linear problem is implemented to facilitate computation of variables that are defined on the domain only, in this work the free-surface elevation. This results in introducing an additional problem to compute the elevation on the interior which is dependent on the free-surface, but not vice-versa. This, to ensure the wave problem remains unaffected. Coincidentally, this definition of the free-surface elevation on the interior is used to deform the mesh required to capture non-linear effects in the time-dependent domain.

Results of the linear problem show to agree with literature. Conservation of energy is guaranteed, yet conservation of mass can be attained with sufficient mesh and time resolutions. This affirms the successful implementation of free-surface problems in DelFI. Extension to the non-linear case shows that energy is not conserved, yet analysis shows it should. The same holds for conservation of mass. Still, both quantities can again be contained with sufficient mesh and time resolution. Additionally results demonstrate that implementation of a stabilization scheme is needed. Finally a benchmark case demonstrates that with the current limitations, results agree with others.

This research demonstrated a novel mathematical framework to compute non-linear free surface waves with special emphasis on conservation laws and geometric compliance with the fluid through Isogeometric Analysis. A basis has been laid towards optimization of wave maker signals given a free-surface wave envelope.

Hydrofoil crafts with fully submerged foils can provide fast and economical waterway transport. However, their operation requires reliable onboard control systems to ensure the safety and comfort of their passengers, especially in rough sea conditions. This thesis project is focused on the dynamical modelling and the design of motion control systems for an experimental scale hydrofoil craft that is available at TU Delft, namely the Hydrofoil Education and Research Platform (HEARP).

The development of the dynamical model of HEARP is done by taking inspiration from the dynamics of marine crafts and aircraft and relying on different assumptions to obtain a simple and low-order model. The resulting model is a linearized state-space model with three degrees of freedom, namely heave, roll, and pitch, and includes the influence of regular waves. Because of inaccurate available data for the mass properties of HEARP, variations of the system parameters due to nonlinearities, and changes in the operating conditions, different uncertainties are assigned to most system parameters.

The use of multivariable feedback control methods for the motion control of hydrofoil crafts is limited, so this work is focused on exploiting such methods to improve the performance and robustness of such systems. The representation of the perturbed system using real parametric uncertainties is proved to be computationally expensive for the control design. Thus, the perturbed system is approximated by complex (dynamic) perturbations. A signal-based H-infinity optimal controller is designed using the nominal system, and a mu-synthesis optimal robust controller is designed using the approximated perturbed system.

The performance and robustness of the proposed controllers are evaluated in both frequency and time domains through simulations. From the results, it is concluded that both controllers offer high-performance system responses for both reference tracking and disturbance rejection of incident waves. Furthermore, by comparing the two controllers, it is observed that the mu-synthesis controller shows superior robustness for the modelled uncertainty. In contrast, the H-infinity controller has a slightly better performance when considering the perturbed systems with the real parametric uncertainty. The results of this thesis project can be used in the future to experimentally validate the accuracy of the proposed dynamical model and the performance of the designed controllers.

The diffuser stern is part of a design concept from Damen and the TU delft: the interceptor diffuser. This concept aims to improve the seakeeping behavior of fast ships in waves. With PIV measurements, a contribution to the development of the diffuser stern is made. The flow acceleration and detachment are studied, and a comparison with a CFD pre-study from Damen is made.

Floating structures have developed significantly in recent years. As land is becoming scarce, the use the water surface will contribute to ease this scarcity. Therefore in recent years, floating structures covering large areas have been developed, also called flexible floating structures.
In this project the focus is set on the mooring system very flexible floating structures (VFFS). At the TU Delft, two towing tanks can be used to investigate the mooring system of VFFS, however first a reliable measuring system is required that is able to examine a specific part of the mooring system. Conventional setups that measure the mooring forces consist of large instruments, as these instruments only have a small effect on their investigated structure (vessels). The response of VFFS is dominated by elastic deformations and differs from conventional rigid structures. For VFFS, these type of instruments will have a large effect on the structure motions and thus these conventional setups cannot be used. Therefore, a new measuring system is required to conduct small scale experiments with VFFS, and the following objective is formulated: Develop an instrumented mooring system for VFFS at model scale for the towing tank at the TU Delft and determine its accuracy.
A new concept is developed in this project. This concept resulted from an extensive concept development where all functions of the system were analyzed. With the use of a Morphological Chart and a Multi Criteria Analysis the best concept was selected. For this concept, it was determined that the focus should be on the sensor configuration and calibration procedure.
First, the optimal sensor configuration of the concept was specified by analysing the working principle of the concept. Second, the calibration procedure was further analyzed. From this analysis, three calibration procedures were developed: the single sensor calibration matrix, the full fixed calibration matrix and the full rotated calibration matrix. From literature and theory, it was not possible to determine in advance what calibration procedure should be selected, and therefore the performance of the procedures were verified with experiments. All calibration procedures were executed, whereafter the performance of the different procedures were compared. The two main considerations for the comparison were the accuracy and the usability of the procedures. After performing the comparison, the main conclusion was that the full upright calibration procedure is the optimal procedure.
To verify the concept under realistic conditions, an example application was performed in the towing tank No.1 at the TU Delft. By doing this, the concept has proven to be suitable to measure the mooring force and transform them into usable data.
In this project a new concept was developed into a working system. This system forms an excellent base for extensive research into the mooring system of VFFS, and is a good addition to the measurement instruments for the towing tank at the TU Delft. It is concluded that the system is able to measure the mooring forces and the direction. The accuracy of the system still has to be improved, and with additional research the working concept can be further developed.

An important step in determining the performance of a commercial ship is to determine the vessel fuel consumption in offdesign conditions. In previous studies, the vessel fuel consumption is obtained using machine learning algorithms which use navigational data and meteorological data to train the models. Due to the inaccuracy of the weather hindcasts, the accuracy of these machine learning techniques is not satisfactory for Compagnie Maritime Belge (CMB) which is the company for which this study is performed. Since the vessel motions are a direct result of the meteorological conditions an approach is investigated in this research to replace the meteorological data by motion data to train a machine learning algorithm and obtain accurate results. The vessel motions are captured by a motion sensor on board the vessel. Multiple machine learning techniques were tested in this research. The data used to train the best performing learning algorithm represent the motions by the first and second moment of the nonderivative motion and the second moment of the first derived motion. These moments are obtained from the motion distributions of a time window of 5 minutes. The obtained best parameters for representing the motions were obtained by testing if a spectrum must be provided, how many motion derivatives are required, which moments are needed and how large the time window must be. Although the accuracy of the developed learning algorithm is determined as satisfied no digital twin could be made and tested. Unfortunately, due to the limited amount of data available no separate algorithm could be made which uses the Extra Tree learning, as digital twin base, to determine the overconsumption of the vessel. Only data from one voyage was available. All this data was used to train the learning algorithm to have as much variance as possible in the data set such that a good learning algorithm could be developed.

The thesis proposes a methodology for a Surrogate Based Optimisation (SBO) study and its application in the Naval Architecture field using the open-source software DAKOTA. Surrogate Based Optimisation is a useful and powerful way of reducing time and costs in hull shape optimisation. It allows the employ of exploratory data analysis techniques and machine learning methods to get more insight, discover hidden patterns, and detect anomalies in the design space. However, an SBO consists of several interconnected steps, and each increases complexity and uncertainty to the overall process.

The main goal is the investigation of the effects of different surrogate models on ship geometry optimisation from a resistance point of view. Also, different sampling plans, infill techniques, and optimisation algorithms are analysed and compared, as elemental steps for an SBO. This is done by the use of test functions that emulate the problem of ship optimisation. Furthermore, the intent is to try and list general guidelines to follow when building up an SBO routine with CFD simulations involved, with a focus in the marine field.

This research led to the assembly of Halton sampling sequence, Kriging meta-model, Expected Improvement function, Genetic Algorithms, and Pattern Search tools to implement a working SBO routine. This routine is used to demonstrate the success of the SBO method for a Hull Vane and an aft ship optimisations. Moreover, it is used to validate the research outcomes of this thesis and to prove that this work can be safely used for every-day commercial work.

The two design applications provided a resistance reduction, with respect to a benchmark hull, of about 19% and 7%. Thus not only they were successful, but also they were excellent examples to show how a surrogate model and its correct visualisation can give the Naval Architect the right tools to critically analyse the results, gain more understanding on the problem, spot and correct anomalies, and provide creative solutions to a client.

Hydrofoil ships with fully submerged wings offer great speed at a relatively low fuel cost. They are also unstable, which is why they require an active control system. Current hydrofoil ships are more limited in their operation by waves. A passive suspension system could reduce the motions of a hydrofoil ship in waves but it must remain controllable. This becomes more difficult when the ship is sailing in waves because the wave orbital velocities affect the lift of the wings. In this thesis, a state-space model is developed that predicts the stability and the motions in waves of a hydrofoil ship with a suspension system between the hull and the wings. A foiling ship experiences forces of different origins. To segment the model development, three levels of equations are created; physical level, component level and system level. The physical level consists of expressions that describe the forces like the lift, drag or added mass. All forces acting on a component are then combined into the component level equations. The effects of all components are then combined to find the system dynamics. The model development is split into two parts. The first part is the development of a fixed-wing model. The predictions of this model are compared to experiments of the Demonstrator of MARIN. The Demonstrator is a 3.5m long model and has a mass of 150kg. A subset of these tests is used to compare to the predictions of the developed model. The motions are predicted well for the lowest wave frequency, but the motions for higher frequency waves are not predicted sufficiently accurate. The suspected cause for the inaccuracies is a lag in the control signal in the experiments, which plays a greater role in the higher frequencies. It is recommended to implement this delay effect into the model to improve the range of frequencies the model can be validated for. The model is also compared to higher wave amplitudes for the lowest wave frequency. The predictions of the model for the higher wave amplitudes match the measurements from the experiment reasonably well. The state-space model is then extended to allow for a suspension at the forward wing, aft wing or both wings. No experiments were available to compare to the model predictions, which means the model is only verified. The verification is done by making the increasing the stiffness to sufficiently high values, that the fixed-wing limit is reached. In the analysis of the results, it was found that the suspension system can reduce the motions of the ship, but only in very specific circumstances. In all other circumstances, the motions are either larger than the fixed-wing model or the system becomes unstable. It is therefore recommended for future research to try rotational springs.

At the moment, offshore wind is an indispensable source of energy. Compared to onshore wind, wind speeds offshore are considerably higher and more constant. Furthermore, offshore wind turbines reduce noise and visual impacts. In contrast to fixed bottom foundations, floating wind can be economically feasible in global offshore waters that exceed a depth of 60 meters. However, floating wind has its challenges. Research has shown that the interaction between floater motions and wind turbine controller can be significant, especially above rated wind speed. The interaction between the hydrodynamics and aerodynamics above rated wind speed is the main reason why fully coupled aero-hydro-servo-elastic simulations are needed to determine the motions and mooring loads of the system. The downside of fully coupled simulations is, next to the need of comprehensive technical background, the high computational time which makes these simulations unsuitable for concept design studies. The purpose of the research is to developed a fast and simplified model that can be used for design iterations in concept design stages of Floating Offshore Wind Turbines (FOWTs). The simplified model must be able to predict the general motions and mooring loads with acceptable accuracy. Moreover, the simplified model should be based upon available wind turbine input data in an early design phase. A sensitivity analysis is carried out to determine the contribution of various physical phenomena, present in fully coupled aero-hydro-servo-elastic simulations, on the motions and mooring loads of the GustoMSC Tri-Floater with a 5MW NREL wind turbine located at its center. The fully coupled simulations are performed with aNySIM-Phatas. By this analysis, the main physical contributors to the motions and mooring loads of the GustoMSC Tri-Floater are selected to include in the simplified model. It is confirmed that the interaction between the aerodynamics and hydrodynamics is significant, especially for platform pitch motions in the above rated conditions. Furthermore, the (controlled) aerodynamic thrust calculated from Blade Element Momentum (BEM) theory, as well as the wind drag loads are found to be the main driving forces on the system. The simplified model is constructed by coupling an external Python code to the multi-body time-domain software aNySIM. Whereas aNySIM is responsible for the hydrodynamic and mooring calculations, Python is used for the simplified aerodynamic calculations. The effect of the wind turbine controller on the aerodynamic thrust is included by the use of the steady state thrust curve, aerodynamic damping ratios and filter on the incoming turbulent wind. The simplified model is compared with fully coupled aNySIM-Phatas simulations. Compared to fully coupled simulations, obtained results suggest that in a free floating condition the differences in the mean are found to be within 5% and 10% for platform surge and platform pitch, respectively. The mean and maximum differences in the forward fairlead tension were found to be within 2% and 5%, respectively. Moreover, the results show that the simplified model is 5 times faster and much easier to construct compared to the fully coupled simulations. The results of this research are encouraging as the simplified model can serve as a tool to predict the general motions and mooring loads with acceptable accuracy. Also, as Python is not coupled to aNySIM before to the best of the authors knowledge, this coupling opens doors to other simplified modelling strategies.

The present research performed a design exploration of fast ferry catamarans with hard chined demihulls sailing in irregular head waves, to analyse the influence of the demihull shape (without appendages) on wet deck slamming, and identify the geometrical properties and their trends to reduce it. Wet deck slamming is a phenomenon known to cause passenger discomfort and seasickness, often a source of delay in the ferry's schedule and can even cause irreversible damages and endanger the crew. The research is carried out with numerical methods, where two main programs are used, namely the strip theory-based program for non-linear motions of high speed crafts FASTSHIP, and the design optimisation, parameter estimation and sensitivity analysis software, DAKOTA. To set up the experimental analysis, design and analysis of computer experiments (DACE) methods are used. The use of DACE involved the application of an in-house built parametric design tool for mass generation of hulls, coupled with a sampling-based method for the exploration of the design space. Because DACE methods require a large amount of data points (of the order of tens of thousands), it is deemed unfeasible to perform all the analysis with FASTSHIP. Therefore, a surrogate model is built instead, with the global approximation-based method known as Kriging. By considering the surrogate model in the study, mass production and evaluation of different hull forms were achieved, which led towards a brute force surrogate-based optimisation approach. Such approach enabled an in-depth exploration of both sub-optimal and Pareto-optimal regions of the multi-objective domain, thus not only showing the hull characteristics of the best hulls but also their trend of change from the sub-optimal to the Pareto-optimal region. Additionally, an alternative approach based on evolutionary algorithm is coupled to the surrogate model for a more direct and straightforward optimisation process, namely the multi-objective genetic algorithm (MOGA). Both approaches converged towards the same region, thus reinforcing the reached solution. The outcome of the research established a valid framework for performing hull design explorations on catamarans while addressing wet deck slamming. Furthermore, it showed that the demihulls should adopt bulkier forms. However, a trade-off existed between the various considered objective functions, hence the final decision should be based upon the user's criteria.

Recently the interest in hydrofoil vessels using a fully submerged hydrofoil increased. Hydrofoil vessels experience a resistance hump and possible instabilities during their takeoff of which both are influence by waves. In order to determine the required power during the takeoff to overcome the resistance hump and to investigate whether the hydrofoil vessel is stable during the takeoff numerical simulations or experiments have to be performed. Experiments covering the full takeoff require a long towing tank and high towing velocities and therefore these experiments are costly. The alternative, performing numerical simulations, is favorable but requires a suitable numerical code. As no validated code is available in which a takeoff can be performed in waves, in this research Panship, a potential flow code suitable for high speed craft and adapted for use with hydrofoil craft, is validated for the takeoff of a hydrofoil vessel. In this research the results of Panship are compared with results of experiments and other numerical codes in four different stage as little validation data is present for the takeoff of a hydrofoil vessel. The four different stages are: a foil without the influence of the free surface, a foil with the influence of the free surface, two foils with the influence of the free surface and the hull and foil system near during the takeoff. In these four stages the lift and drag determined in Panship is compared with available experimental results and numerical codes. As the simulations in Panship require a vast amount of input parameters, the influence of these parameters is studied first to ensure the reliability of the results. Comparing the results of Panship for a deeply submerged foil with another potential flow solver shows that the lift is predicted correctly, but that the induced drag is underpredicted. If viscous effects are required in the solution, the quality of the results of Panship is dependent on the type of hydrofoil used. As some foils have a large viscous effect on lift, the lack of viscosity on Panship, leads to a low quality of the results for lift and induced drag as they are overpredicted. The trends for interaction of a foil with the free surface are predicted correctly but the underprediction of induced drag influences the quality of the results. Panship is able to predict the effects of foil interaction for lift correctly, but the results for foil interaction for drag are poorer. During the takeoff Panship is able to predict the lift fairly correct but the total drag is underpredicted severely. Recommendations are given to improve the results of Panship and for future work.

Accurately predicting the ship motions, is essential knowledge when operating in water. To predict the ship motions, multiple ship characteristics have to be known. For the roll motion, one of these characteristics is the roll damping coefficient. Nowadays the industry standard to determine the roll damping coefficient of a ship is by performing model tests or using empirical formulas created in the late ’70s. In recent years computational power has increased significantly. This increasemakes it possible to start using computational fluid dynamics (CFD) for practical purposes instead of solely scientific research. The roll damping coefficient is a very interesting parameter to examine with viscous flow solvers as the roll motion is heavily influenced by the viscosity of the water. Multiple studies have been performed with 3D and 2D CFD simulations. By simplifying the simulation to 2D, calculation time is reduced drastically. This simplification is performed by only simulating one cross-section of the midship. Only simulating one cross-section causes a loss of accuracy in how well the roll damping coefficients can be determined as the damping effects of the bow and stern are not taken into account, or are roughly estimated. During this research, a method to accurately determine the roll damping coefficients of a ship, while maintaining low computation time, is investigated. This method is based on performing multiple 2D CFD simulations of different cross-sections and combining the results to capture the behaviour of the entire ship. With this approach, the high accuracy of 3D simulations and the low computation time of 2D simulations are combined. The accuracy of this 2D section method is determined by comparing the results with experimental data. The experimental data is obtained through free roll decay tests of a ship from Van Oord on model scale. A 3D CFD model, which simulates a free roll decay test, is created using OpenFOAM. The results of this simulation are compared to the experimental data to determine how well a 3D simulation can predict the roll damping coefficients. Next, a 2D CFD model is created using OpenFOAM which is validated using experimental data of a structure which has a uniform cross-section along the entire length. Lastly, the validated 2D CFD model is used for the 2D section method. Multiple cross-sections of the same ship from which the experimental data is obtained and with which the 3D CFD simulation is performed, are simulated. The results of the 2D section method are compared to the experimental results in order to determine how well this method can determine the roll damping coefficient of a ship. The simulation of the 3D free roll decay test underestimated the linear damping. The non-linear damping was within the margins of the experimental data. The 2D forced oscillation simulations showed an excellent agreement in linear and non-linear damping with both the experimental data as with a comparable CFD simulation. The 2D section method slightly overestimates both the linear and non-linear damping of an entire ship but overall this method shows promising results. However, more research should be done to optimise the method further.

The design of polar expedition cruise ship is not a trivial task to the naval architect due to the complex nature of the ship itself. Concept exploration are hereby required to improve the understanding of the design problem and identify its challenge as well as possible solutions. However, it is always difficult to generate the technically-feasible design solutions while still satisfying multiple and potentially conflicting performance criteria for the polar expedition cruise ship. There are currently two ways of optimizing the concept design: manual optimization and computer-optimization. The computer-based optimization, such as the Packing Approach developed by TU Delft, can semi-/automatically generate a set of design solutions. After the generation of these design solutions which are unknown in many performance behaviors, it is necessary to post-process them and increase the quality of those designs created from both optimization methods. Design rationale is an effective way to serve as qualitative metrics in post-processing the configuration layouts of the concept designs. However, the ambiguity and potential conflicting aspects within the design rationale are difficult to handle by the current post-processing method while fuzzy logic theory can be utilized to cope with the design problems involving subjective and ambiguity. Hence, it is proposed to develop a new methodology to better post-process the concept design of a polar expedition cruise ship.

Stolt tankers is the largest operator of deep-sea chemical parcel tankers. Parcel tankers differentiate form bulk chemical tankers by the additional features that the ships have, such as cargo cooling, cargo heating, many (small) cargo tanks, highest level of chemical resistant materials and notation to carry the most dangerous chemicals. These features make the parcel tanker versatile compared to other tanker types, however this results in a higher cost. In the last twenty years, a shift has taken place in the deep-sea chemical shipping industry, driven by changes in regulations, funding of ships and macroeconomic factors. These changes have made the playing field of deep-sea chemical tankers more equal and commoditised, resulting in more competition by entries of new, privately funded companies with simple chemical tankers. Currently a large part of the parcel tanker fleet is nearing their end of lifetime and are expected to be recycled in the next five years. This creates the need to determine operating needs and requirements for replacements of Stolt Tankers’ ageing parcel tanker fleet. The goal of this study is twofold, first, is to determine the need for parcel tanker features in the market and, secondly, to determine the value that these features add. The findings have been concluded in terms of a priority for the feature in a new parcel tanker specification according to the MoSCoW prioritising technique.Analysis of the features is based on D37 parcel tanker class of Stolt Tankers. This typical parcel tanker class is a candidate for recycling in the next 5 years, and therefore a ideal study object for a replacement study. Four features were identified as giving the highest differentiation of a parcel tanker and are analysed in more detail: cargo cooling, cargo heating with thermal oil heating medium, small cargo tanks, and ship type 1notation. These features are analysed both qualitatively and qualitatively on its operation in the Stolt fleet, market demand and supply, competition and alternatives. The net added value of the features over the life of the ship has been calculated by means of a discounted cash flow analysis for base-,best and worst case scenarios.The cargo cooling feature yields the best added value for a new parcel tanker. The thermal oil cargo heating feature has a negative added net present value for only cargoes with this specific requirement. The study of the ship type notation has a low investment cost, however the loss of revenue by the capacity reduction is far more than potential revenues from thetype 1 cargoes. The number of cargo tanks has a large impact of the capital cost of the ship.Reducing the number of cargo tanks, i.e. the average tank size, yields a relatively large savingfor parcel tanker ship types. Combined with a decrease in the number of small cargoes, makes this a logical development for future parcel tankers. However, more research is required since there is a large discrepancy between the number of small cargoes on different trade routes.Based on the findings from the market and operational review of the D37 features the following priorities are given to the features. The cargo cooling system has been given a’Should Have’priority, the thermal oil cargo heating system has a’Could Have’and the ship type 1 notation a’Won’t Have’. Further research is recommended for the requirement of features of new ships with respect to the total fleet of Stolt Tankers. Secondly, modelling of cargo tank layouts with the statistical trading data is recommended to determine the requirements for the number and sizes ranges of cargo tanks