Sparse triangular solves (SpTRSV) form the latency–critical inner loop of many direct and iterative solvers, but strong data dependencies limit thread–level parallelism and make the kernel dominated by memory–latency.
This thesis explores whether redundant computation can be exchanged for improved data locality to accelerate SpTRSV on cache-based multi-core CPUs.

The proposed two-phase algorithm first uses the Recursive Algebraic Colouring Engine (RACE) to permute an arbitrary sparse triangular factor into a block-lower–bidiagonal form whose diagonal blocks fit into a private L2 cache.
Each block is then solved twice: an independent provisional pass followed by a lightweight correction that re-uses the still-resident data.
The resulting task graph halves the critical path, exposes ample parallelism, and leaves total memory traffic unchanged.

An OpenMP5.0 implementation employs the new affinity clause so that producer–consumer task pairs are likely to run on the same core.
Performance is compared against IntelMKL’s sparse triangular routine and the Kokkos-Kernels sptrsv, while LIKWID counters validate cache behaviour.

On a 24-core Cascade Lake node of the DelftBlue supercomputer the task graph achieves a strong-scaling speed-up between 1.25 and 1.45 that saturates after roughly six threads; Kokkos attains similar absolute time only beyond sixteen threads.
With 24 threads the solver is up to an order of magnitude faster than single-threaded MKL, and the affinity hint alone accelerates execution by 1.1–2 times, confirming the importance of cache reuse.
MKL remains preferable for small or highly structured SPD matrices, but on irregular, memory-bound factors the new solver rivals state-of-the-art libraries despite its simple prototype status.

This research investigates efficient matrix-vector multiplication and storage techniques for a Fortran implementation of the Conjugate Gradient (CG) method on GPUs. The CG method is a widely used iterative algorithm for solving large, sparse linear systems.

Because its performance is heavily influenced by the efficiency of matrix operations it is important to make use of parallel computing to run the computations on a GPU. Three types of matrices were considered as A, a dense symmetric positive-definite matrix and two sparse matrices, defined by the discretization of a 1D Laplace equation and a 2D Poisson equation respectively. For sparse matrices, the Compressed Diagonal Storage (CDS) format was implemented to reduce memory usage and computational cost. For parallel execution, three implementations were benchmarked: Standard Fortran, OpenACC, and CUDA Fortran. The results show that CDS significantly improves both memory efficiency and runtime performance for sparse matrices, while CUDA outperforms OpenACC and Standard Fortran in terms of speed, especially for large matrices. Standard Fortran remains competitive for small matrices due to low overhead, but it scales poorly. The convergence behavior of the CG method was also found to be highly dependent on the condition number of the matrix, with the 2D Poisson matrix exhibiting much faster convergence than the 1D Laplace matrix.

This study concludes that efficient storage and GPU-based matrix operations, particularly using CUDA, are critical for scalable performance in solving large linear systems with the CG method. Future work could explore combining the portability and ease of use of Standard Fortran with GPU acceleration to achieve both maintainability and high performance.

This thesis addresses the challenge of segmenting ultra-high-resolution images. Limitations of current approaches to segment these are that either detailed spatial contextual information is lost or many redundant computations are necessary. To overcome these issues, we propose a novel approach combining the U-Net architecture with domain decomposition strategies to balance incorporating spatial context and maintaining computational efficiency. Our proposed method partitions input images into non-overlapping patches, each processed independently on a separate device. A communication network facilitates the exchange of information between patches, enhancing the model's understanding of the spatial context.

Through theoretical analysis and practical experimentation on device memory usage during training, we demonstrate that our approach incurs minimal additional memory overhead for inter-device communication. Evaluation on synthetic and realistic datasets, including the Inria Aerial Image and DeepGlobe Satellite Segmentation datasets, demonstrates the effectiveness of our approach. Our model achieves competitive performance compared to the baseline U-Net model, with consistent class predictions around boundaries. Visualization of feature maps highlights the role of the communication network in transferring contextual information. Furthermore, it is shown that our approach remains scalable even when trained on limited subdomains.

In conclusion, our proposed model offers an intuitive solution for segmenting ultra-high-resolution images by effectively incorporating spatial context. Future research could explore variations of our model, such as overlapping subdomains or communication on different levels of the U-Net, to further enhance boundary consistency and information transfer.

Extreme Ultraviolet lithography is a vital step in the production of cutting edge computer chips that drive emergent technologies like AI, VR and the internet of things [1]. Careful control of Ultraviolet light beams by EUV machine modules such as the Unicom, Uniformity correction module, enable extremely precise printing of nano-scale structures on these chips. This thesis focuses on the construction of a model that predicts the impact of the Unicom on EUV illumination in ASML's lithography machines. Such a model could be used in a predictive maintenance scheme to prevent a fraction of unscheduled machine downtime, which can be estimated to cost ASML's customers around 38 million dollars per machine per year [2-4]. Similar problems have been tackled before [5, 6], but both existing models disregard multiple properties of the EUV machine and are incompatible with the type of measurements obtained by ASML. Therefore, the question remains how an accurate Unicom model can be constructed.
In this thesis, a physics based Unicom model was developed that can be fine-tuned to machine specific measurements. Significant reductions of up to 90.3% of prediction errors were obtained by using the model. Overall, making use of the model provided better or equivalent predictions when compared to not using the model for all but one of the investigated indicators of prediction quality. For the latter indicator errors remained within the desired bound, but further investigation is needed to discover why the Unicom model adversely affected this indicator. With an average execution time of 31.4 s, the created Unicom model in general enables swift and substantial accuracy gains.

In this study, a model is developed in COMSOL to investigate the nature of energy losses in ion exchange membranes used in electrochemical energy conversion. By use of electrochemical impedance spectroscopy, the nature of resistance and reactance, and therefore the nature of Ohmic resistances and charge transfer resistances, are evaluated. This is done by first obtaining a model of the setup and the resulting impedance spectrum and then making adjustments in the model to observe the impact on the impedance spectrum as well as the resulting ion concentration profiles. In the used system, the diffusion effects were dominant compared to the double layer effects. Therefore, the impedance spectrum is mainly shaped by the diffusion boundary layer, especially at low frequencies. The ion concentration, diffusion coefficients, and membrane properties were found to have a significant influence on the impedance spectrum. Insights from this study can be used in the optimization of ion exchange membranes and the efficiency of energy conversion systems.

In this thesis, the acoustic noise emitted by a small electric outrunner motor, such as the ones commonly used in unmanned areal vehicles, is analyzed using a multiphysics simulation based on the finite element method. The simulation covers the electromagnetic, mechanical and acoustic domain. The electromagnetic simulation is performed in the 2D time domain, the mechanical simulation in the 3D time domain and frequency domain and the acoustic simulation in the 3D frequency domain. The difference between results of the time and frequency domain mechanical study is investigated, as well as the difference between the obtained results and those obtained in other available literature. Furthermore, the results are compared to acoustic noise measurements performed in the laboratory. Based on the obtained results, noise reduction methods, specifically focussed on small electric outrunner PMSMs, are proposed and verified using the developed simulation.

Almost everyone is familiar with games such as poker and checkers, but games can also be found in non-entertaining settings, such as competing companies in a market or conflict resolutions between countries. However, what happens when players want to avoid working together? This thesis provides two game-theoretical models of boycotts based on existing literature and is complemented with new insights. We discuss the dynamic process of a consumer boycott, where the effectiveness can be determined by the predefined consumers’ and firm’s thresholds or by the maximum boycott duration. Besides, we consider a static model using the Shapley value to analyze the impacts of boycotts, which are balanced for 2-player boycotts but not when three players are involved. We extended the existing models by finding supermodularity as a sufficient requirement for the boycott to never be profitable for the involved players, by introducing a direct formula for the coalition value in a boycott game (the boycott-contraction), and by analyzing real-world data, where we observe the effect of boycotts on trading activities with uninvolved players.

In this work residual error estimates are constructed using Neural Networks for Finite Element Method. These can be used to do adaptive mesh refinement. Two neural networks are developed the Multilayer Perceptron and the Transformer model. The error estimates are made for 1d poisson equations but the idea will generalise to higher dimensions as well.

An increasing part of European electricity is provided by renewable sources, whose output varies considerably. In order to offset these variations, gas can be used as an auxiliary source of energy for power production, but extra power can also be used to produce hydrogen or methane for storage. As a consequence, gas and power networks are growing more and more interdependent and thus need to be modelled together in order to make meaningful predictions. A growing body of literature deals with the modelling and simulation of such coupled networks, but the main focus of most sources is accuracy. On the other hand, Gasunie and TenneT’s Infrastructure Outlook report aimed to study hourly snapshots of a coupled network over one year, in a variety of different scenarios. Therefore, a simplified and computationally inexpensive model was needed. Literature dealing with such models is lacking, so a linear model consisting of transport load minimization was constructed for this purpose. It was used for both gas and power simulations, but it is inadequate for the latter.
In this thesis, the transport load model is combined with the standard DC model of power flow in order to obtain more accurate results. The resulting combined linear model is tested on small sample networks, showing that it is functional given a set of modelling assumptions. In addition, the transport load model is compared to the DC model in a sample set of Dutch power networks. A large difference is observed between the two, confirming that there is no reason to use transport load as a model for electricity. Further research is needed to evaluate the accuracy of the transport load model for gas.

In this thesis work, we strive to describe and apply the modelling methodology for simulating a turbulent jet diffusion flame stabilized behind a bluff body by applying flamelet generated manifold (FGM) model and steady diffusion flamelet model for predicting the pollutant emissions from the flame. The flame under consideration is a CH4/H2 (1:1) bluff-body stabilized flame known as HM1. The numerical calculations of the flow physics has been completed by using a commercial CFD code namely Ansys Fluent 19-R3, where the computational domain is assumed to be two dimensional and axis-symmetric in nature due to the cylindrical symmetry of the burner with an structured grid for meshing. The turbulence is modelled by using a two equation Reynolds averaged Navier Stokes model, namely the standard k – ϵ model with a modification in the Cϵ1 from 1.44 to 1.60. The chemical mechanism used in the project was the GRI 2.11 mechanism developed by the Gas Research Institute, USA. Next, postprocessing tools like the Reactor Network Model (RNM) and Ansys NOx post-processor were used as an alternative, possibly a better way to obtain the NO species
concentrations. The acquired results from the simulation are thoroughly analysed and were compared to earlier results on the HM1 flame in the literature and validated by using the experimental data documented by the University of Sydney in collaboration with the TNF Workshop. The results showed that the steady diffusion flamelet performed better than the FGM model and was in acceptable agreement with the experimental data, although some under-prediction and overpredictions were reported for NO and OH species. In regard with the post-processing tools the RNM model performed better than the Ansys NOx post processor.

With the aim to reduce the CO2 emissions in the transportation sector, there have been significant developments in the recent few years where novel ground transport systems operating at airline speeds using all-electric propulsion have emerged to provide a sustainable alternative to classical methods of
transport such as air travel. These novel systems however use sub-systems for suspension and propulsion that produce significant power loss when compared to traditional rail on wheels. Hence, the aim of this project is to investigate a new system proposed in [1] which aims to use the same concept of Hyperloop
involving pods traveling at high speeds in vacuum tubes to minimize drag and thus reduce the energy requirements as well. Unlike Hyperloop, the system aims to use wheels for suspension and a multi-mode linear motor for charging and propulsion where the power supply is integrated into the overall system.

First, an overview of the system is presented before proceeding to the main content of this thesis where the Linear Doubly Fed Induction Machine (LDFIM), the machine intended to perform the charging operation for the vactrain system, is initially designed using an analytical model based on theoretical equations and formulae which is then validated using the Finite Element Method (FEM). Afterwards, the machine is optimized using the Genetic Algorithm (GA) method to enhance its performance. The optimized model is also validated in FEM and all of the results and design parameters for this specific machine are delivered in this report. It is deduced that the LDFIM is capable of performing according to the required performance standards and a sufficiently high efficiency is achieved.

After completing the design of the LDFIM, an outlook on the system’s propulsion is discussed before conducting a benchmarking study where the investigated system is compared to Hyperloop by applying the two concepts in a scenario setting and assessing their performance in a variety of criteria. The results
demonstrate the capability of the investigated system in reducing the energy and material requirements for ultra-high speed travel making it a promising candidate for a future transport system.

Over the past few decades, there has been a steep increase in efforts to generate electricity from clean and renewable energy sources so as to reduce consumption of fossil fuels. A considerable amount of research is in progress to explore the scope of various sustainable power sources; among which tidal energy is estimated to have a huge potential. With advancements in technology, it is predicted that electricity from tidal energy would increase in the coming years. It is necessary to have highly efficient generators that can convert tidal energy to electricity. Propeller type generation system using Permanent Magnet Synchronous Machines (PMSMs) are found to be suitable for the application of tidal generators. For the best utilisation of resources, it is essential to improve the performance of these machines in several aspects. And hence, multi-objective optimisation of electrical machines is emerging to be of high significance. The key goal of this thesis is to design, model and optimise a PMSM generator so as to get the desired power while considering other design objectives, including the efficiency, total volume of the machine and weight of magnets. The thesis is divided into two parts. The first part consists of design, modelling and analysis of the PMSM machine. Initially, all the required machine constraints are identified and a topology is selected. An analytical model is built after which the machine is modelled using the Finite Element Method (FEM). The model is analysed where various key characteristics of the machine such as flux linkages, induced EMF, inductance and torque are calculated. This is validated by ensuring that results from FEM model and analytical model match. In order to optimize the machine; certain geometric parameters are varied to locate the dominant geometrical parameters. The second part of the machine consists of optimisation. A multi-objective Genetic Algorithm (GA) optimization is carried out to generate data to search for an optimal design. The three objectives for optimisation are higher efficiency, lower total volume and lesser weight of magnets. The analytical model is used in the optimisation algorithm. Since this is a multi-objective optimisation, a set of solutions (Pareto set) is generated from which a single solution is chosen. This selected solution is applied to the FEM model and thus finally an optimized model of the machine is developed. An analysis is carried out on the optimised model. Finally, the optimised model and the initial model are compared and the results are explained.

The work presented in this report investigates the effect of replacing natural gas fuel in a radiant tube burner in an industrial furnace with hydrogen fuel. The flue gas is fully contained inside the radiant tube burner and does not come into contact with the iron products inside the furnace. The burner interacts with the process side of the furnace only by heat transfer. Therefore, the investigation could be divided into a study of two separate systems: the burner and the furnace systems. In the first part, the burner system was evaluated. The impact of changing the fuel of the burners from natural gas to hydrogen on heat-, flow- and radiation of the existing burner was studied. Moreover, the NOx generation has been considered. The investigations were performed using Reynolds Average Navier Stokes (RANS), Computational Fluid Dynamics (CFD) approach using detailed chemistry. Natural gas combustion was simulated as the basis for comparison. The result shows that the CFD model could estimate the NOx emission of the base case well. Next, the combustion with hydrogen was compared. The result suggests that the thermal efficiency of the burner is slightly higher with hydrogen fuel. On the other hand, the NOx emission is significantly higher. A modified burner including internal flue gas recycle was proposed and studied. The result shows that the modified burner can be operated at the same power and the same level of NOx emission as the existing system. Furthermore, different burner power could be applied in order to comply with the NOx emission limit or to increase processing speed. In the second part, a heat transfer analysis of the furnace has been performed. The model agrees well with measured data of furnace atmosphere temperature. The result suggests that having a higher burner power could increase the processing speed and reduce the energy requirement per ton of iron products.

The aim of this graduation thesis is to present a bipolar stepper motor semi-analytical model and compare its simulated motor dynamics. Stepper motor modeling has been successfully used to simulate motor performance and dynamics, yet, each modeling method holds advantages and disadvantages toward one another. In this thesis, the analytical modeling method is implemented for its faster simulation time. A semi-analytical model is constructed with motor non-linear entities on top of the motor analytical equations.

This thesis first examines and validates a finite element analysis model to extract such non-linear entities into lookup tables. A generalized electrical model and a flux-based model are constructed separately based on the analytical equation and lookup table. In a second stage, the motor dynamics from these models are compared and adapted to the semi-analytical model motor dynamics through non-linear parameters. Two methods to obtain the non-linear parameters, with and without motor measurements, are applied. Finally, Various motors with different parameters are simulated with the semi-analytical model. Comparison, accuracy, and limitations of the semi-analytical model is discussed. At the end of the thesis, conclusions are drawn from the simulated dynamics that a semi-analytical model can indeed be implemented to better simulate the motor dynamics; however, because of the estimation and measurements needed for the non-linear parameters, the semi-analytical model holds several limitations that made it less robust than the model with pure analytical equations.

Flameless combustion, or MILD combustion is a technology that enables stable combustion with high efficiency while maintaining very low NOx and soot emissions. It is characterized by uniform heat flux and can be achieved when fuel and preheated oxidant jets are mixed into a strong recirculating flow. It is a complex process which combines multiple physical phenomena such as, heat transfer and reactions in a turbulent flow and is difficult to capture numerically via simple Computational Fluid Dynamics (CFD) simulations. Therefore, appropriate turbulence, chemistry, radiation and turbulence-chemistry interaction models are required to address this combustion regime via computational modeling. The present work tries to tackle this issue by conducting a comparative study of different radiation, turbulence and chemistry models for the simulation of a labscale MILD combustion furnace using RANS CFD simulations. This assessment was conducted in two steps. Firstly, the importance of radiative heat transfer in furnaces was investigated, by simulating an axisymmetric oxycombustion furnace using advanced radiation models. The SLW, Domain-based WSGG and Cell-based WSGG models were evaluated using simple chemistry and turbulence models. The results were validated based on the work of Webb et al and experimental results. It was concluded that detailed simulation of the radiative gas properties is important and directly influences the resulting temperature field, especially in the bulk flow region. Secondly, a detailed evaluation of a Dutch natural gas fired Delft lab-scale furnace that is operating in MILD combustion was conducted. For the turbulence modeling, the Standard, the Modified Standard, the Realizable k − ε and the k − ω SST Models were compared firstly qualitatively, based on velocity and reaction field contours, and secondly, quantitatively, based on available LDA velocity measurements. The influence of the chemistry modeling was evaluated in two parts. Firstly, the qualitative effect of changing the Cξ constant of the EDC model on temperature and reaction zones of the GRI-Mech 3.0, the DRM-19, the KEE58 and Chen mechanisms, was investigated. And secondly, the quantitative effect on the use of the GRI-Mech 3.0, the KEE58, Chen and Lu-30, based on CARS temperature measurements was researched. Lastly, the influence of the radiation model was investigated qualitatively and quantitatively using the standard 1-RTE domain based WSGG, a 4-RTE WSGG, a cell-based WSGG and the SLW model.
Out of the mechanisms coupled with the EDC, the GRI-Mech 3.0 performs well. Of the radiative properties models the domain based WSGG performs worse compared to the other models. The cell based WSGG performs similarly to the more advanced spectral and non-grey (4-RTE) models and has lower computational cost. For these reasons, the best overall combination for the studied flameless furnace is found to be the combination of Standard k−ε model, standard EDC with GRI-Mech 3.0 and cell based WSGG.

The current economic-development system based on fossil fuels can no longer sustain itself and is incapable of satisfying the needs of the global population. Nearly one third are unable to meet even their most basic needs that involve harnessing energy for everyday essential tasks such as maintaining livable shelter and food preparation. Models of development have exacerbated the problem by promoting and subsidizing fossil fuel-based solutions that require those struggling to survive to do so at the expense of the environment and other species. This is a self-defeating trap, whereby the urgent needs of the most vulnerable are pitted against climate change and the mass extinction of complex life forms. The only way to break this contradiction is through the creation of a new development system that removes incentives for short-term, destructive profit and puts people and other forms of life at its center. Healthy, sustainable development must be rooted in the fair and democratic allocation of resources. Prioritization of truly life-focused progress would allow for human beings to shape their own destinies through the use of locally available, renewable energy resources that do not contribute to mass extinction or climate catastrophe. While the downward spiral towards extinction may seem unstoppable, in reality, the biggest obstacle is simply coming to accept that the old view of what constitutes positive development – an extractive, hierarchical philosophy powered by fossil fuels - must be replaced with the new paradigm based on sustainability and cultural realism. Once that understanding has been reached, it is a matter of targeting key areas for change that will make the most difference in the shortest amount of time. The project outlined herein is a contribution to that new development system. My carefully selected target will address the energy needs for clean cooking and heating for the one third of the world population that relies in low-efficiency biomass combustion technology. Although such technology may at first seem too modest to make a difference, the impacts of making this simple change would have profound, world-wide implications. If that vulnerable one third of the world’s population could access clean domestic technology, there would be enormous improvements in health and environmental conditions. The Kulkan reactor is a biomass gasifier coupled to a burner that optimizes the combustion process and heat transfer by means of heat recovery using an air pre-heater and partially premixed laminar combustion of the syngas. The novel design was founded on three hypotheses that addressed the optimization of efficiency, pollution minimization by flame quenching prevention, and reduction of ultra-fine alkali particle emissions. The first two were investigated using an algebraic model (0-D), and computational fluid dynamics of reacting flow for 1-D, 2-D and 3-D. The algebraic model predicts an efficiency of [60 − 65%] which is 4-8 times higher than traditional cooking technology and 25% higher than the best improved cookstove ever reported. The 1-D and 2-D simulations show reduction feasibility of laminar premixed combustion and a reduction of flame length of up to 50% compared to non-premixed, supporting the idea of more compact combustion chambers and prevention of flame quenching. The 3-D simulations show essential information on the importance of the pressure drop calculations necessary for the design process of a Kulkan reactor, particularly for the burner. Furthermore, arguments for the third hypothesis are presented which invite further investigation to reduce the ultra-fine particles containing highly toxic alkali metals, which are produced during high efficiency biomass combustion.

Danieli Corus supplies tailor-made solutions for the global steel industry. Cameras are used in tuyeres of blast furnaces to detect physical aspects and irregularities. No operator could monitor those videos all day, which is why video analyses are needed. A program is written to analyse a test video and give information about the temperature, the injection of pulverized coal and the movement of rocks. The mathematical background of these calculations is presented, which depend on the transport of light from the blast furnace to the camera through the tuyere. Also the implementation of the calculations can be found. The results are interpreted and discussed and further improvements are suggested.

The recent trend for monitoring and control of the distribution networks requires methods capable to calculate the network’s operating conditions. One way to monitor the distribution network is by state estimation. Although previously this method was established only for the transmission networks, its use has begun to also make sense for the distribution networks, due to the increased availability of smart meters. Typically, state estimation is a non-linear iterative procedure. Research aims to overcome this problem, by creating a non-iterative procedure. To achieve this, linearity of measurement function is a necessity. Although linearity is always desired, the measurement inputs or the procedure renders the state estimation problem difficult to formulate it in a linear fashion. The scope of this thesis is to create a linear state estimation algorithm for the distribution network. To achieve linearity, a reformulation of the states for the state estimation was used at this thesis. This reformulation created a linear optimization problem, which is easier and faster to be solved (Weighted Least Squared method). Since the measurements are not enough to provide full observability, based on these methods, additional assumptions were made. As an alternative to the primary method used for the state estimation, another optimization method was used, which is called Least Absolute Value. This method will be based on the same inputs as the previous method, but different optimization goal. The goal was to compare the two methods and find each one’s limitations. Measurements used for this thesis were provided tracked and provided by DEPsys S.A., by their main module, called GridEye. GridEye devices are synchronized SCADA-type devices, installed on cabinets of the distribution grid, based on the DSO needs. All measurements provided were from real networks within Switzerland. The smart meter device data were provided by collaborative projects of DEPsys S.A.. Smart meter devices were installed in different low-voltage consumers. Both methods were implemented correctly and were compared. Based on the results, it was obvious that the Weighted Least Square outperformed the Least Absolute Value method, in the absence of bad data. Moreover, the effect of the number of GridEye metering devices on the State Estimation was compared. While numerically and visually the outputs are comparable, the effect of having multiple devices is still important for different scenarios. Finally, bad data cases were examined for both methods under different types of errors for the smart meter devices. The smart meter devices were manually introduced to bad data, to understand the algorithm’s capabilities and limitations. An additional algorithm, in addition to the state estimation one, was implemented to not only detect and identify the location of the bad data occurrence but to correct the bad data inserted.

TenneT’s vision of the North Sea Infrastructure (NSI) creates a basis for a joint European approach up to 2050 which focuses specifically on developing the North Sea as a source and distribution hub for Europe’s energy transition. High wind speed, central location and shallow water qualify the Dogger Bank as the location of the central hub.
For further development of the NSI, more detailed research is needed.
The research gap is recognized that there is not sufficient research on proposing new HVDC grids in the North Sea considering optimization (e.g. following a cost-related objective) of different topology structures within a scope of six surrounding countries (BE, DE, DK, GB, NL and NO), with sensitivity tests regarding uncertainties (e.g. in meteorological condition, load, or industry development).
Therefore, the main research objective of this thesis project is to evaluate CAPEX, overall OPEX of participating countries and other operational performances (e.g. energy mix, nodal price, EENS) of possible topologies of HVDC network in the North Sea, including NSI (with a central North Sea Wind Power Hub at the Dogger Bank) and other two competitive topologies, considering uncertainties in green energy technology development, European coordination, load and meteorological condition (e.g. wind speed, solar radiation and hydrology).
Three specific research questions were studied in order to achieve the aforementioned research objective: How to define 3 topologies of the North Sea HVDC network with different feasible structures? What are the criteria to optimize each topology and to compare the topologies? What are the implications of each topology, when evaluated against a wide range of uncertainties, on the overall CAPEX and OPEX of countries involved?
The simulation considers the scenario in the year 2030. Software PowerGAMA and PowerGIM were used for operation simulation and topology optimization. When calculating OPEX throughout the lifetime of the equipment (assuming 30 years), the year 2030 is taken as a representative year and the 30-year OPEX is obtained by multiplying OPEX in 2030 by an annuity factor.
In short, 3 topologies of the North Sea HVDC grid, with hub-and-spoke structure (for the NSI), point-to-point structure and meshed (without central energy hub) structure, respectively, are defined. They are then optimized towards and compared for the lowest overall cost (i.e. the sum of CAPEX and OPEX including CO2 prices) throughout the lifetime.
Simulating under different uncertainties/selected critical scenarios (4 Visions from ENTSO-E’s Ten Year Network Development Plan which reflect different RES share target and European coordination level, and extreme RES inflow and load conditions), the optimized NSI design stays most socio-economically preferable (with lowest overall cost) topology.
It is also recognized that NSI is able to realize its expected functions, namely transmission of renewable energy, enhancement of system security and price convergence. On the other hand, launching of NSI brings in challenges such as grid congestion and benefit asymmetry.
Main contributions of this thesis include:
• Creation of the baseline model/dataset for European power system in 2030 as a background/environment for the North Sea HVDC grid planning;
• Design and optimization of the North Sea HVDC grid topologies in three different feasible structures;
• Verification of NSI’s advantage in cost saving, compared to two competitors, in 4 Visions reflecting different green energy transition and European coordination level, and under extreme RES inflow and load conditions;
• Verification of NSI’s function in improving energy sustainability, affordability and security;
• Realization of Non-Homogeneous Markov Chain algorithm in Excel to generate wind power inflow time series.

The efficient and clean combustion of liquid fuels is a fundamental requirement in the design of future energy systems. Simulation plays a more and more important role in the design of such burners. In this work the spray combustion simulation approach introduced by Ma (2016) is improved, and validated against the CORIA Jet Spray Flame database (Verdier et al., 2017). The database presents droplet temperatures measured by global rainbow refractometry technique, which gives a unique insight in the flame structure. The two phase flow is treated with an Eulerian-Lagrangian approach. Flamelet Generated Manifold (FGM) is used to model the gas phase combustion. The RANS equations are solved using final volume method, with standard k − ε turbulence modelling. The turbulence-chemistry interaction is addressed with assumed probability density function method. The spray cloud is modelled with the Lagrangian transport of droplets including heat and mass transfer. Ma (2016) developed a solver based on the OpenFOAM 2.3.x libraries. His development is complemented in this work with a novel spray model. The improved spray modelling allows the treatment of droplet evaporation in the context of FGM without limiting the complexity of the chemical mechanism. This improvement is crucial for the modelling of complex fuels and the correct prediction of emissions. The modelling concept is rather light-weight considering the RANS approach. Despite the low computational expenses, most of the results agree fairly well with the measurement data. However the correct prediction of droplet temperature remains an an unresolved problem.

Ma, L. (2016). Computational modelling of turbulent spray combustion. PhD Thesis, Delft University of Technology.

Verdier, A., Santiago, J. M., Vandel, A., Saengkaew, S., Cabot, G., Grehan, G., and Renou, B. (2017). Experimental study of local flame structures and fuel droplet properties of a spray jet flame. Proceedings of the Combustion Institute , 36(2):2595-2602.