Search

Nonnegative matrix factorization (NMF) consists in finding two nonnegative matrices whose product is equal to (or approximates) a nonnegative data matrix. NMF is a powerful data analysis tool used to extract meaningful information when the observed data is nonnegative by nature; for example, in text mining, in hyperspectral unmixing, in audio source separation, and in many more applications. The minimum inner dimension of such a factorization is called the nonnegative rank, and computing this quantity is NP-hard in general. Nevertheless, it has several interesting applications; for example, in combinatorial optimization, the extension complexity of a polytope is equal to the nonnegative rank of its slack matrix. Other uses appear in statistics, in information theory, etc. Therefore, there has been interest in determining strong lower and upper bounds to approximate it as closely as possible. In this thesis, we introduce new lower bounds on the nonnegative rank based on some properties of geometric representations of the matrix. First, we introduce a lower bound for slack matrices of convex polytopes, based on a purely geometrical description of the polytope gathering its numbers of faces in each dimension, and on how it changes under projections. In the second part, we introduce a lower bound based on a representation of the matrix as a pair of nested polytopes. Then, we define a framework, based on a series of mixed integer formulations, in which lower bounds can be computed, and we introduce some dynamic methods to get stronger bounds. We compute these bounds on a collection of matrices and show how they can outperform other approaches from the literature., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The growing concerns about global warming combined with fossil fuel shortage call for developing green energy technologies. Even though solar light is one of the most promising renewable energy sources, harvesting it efficiently has its own inherent limitations. Two of the most critical components of solar cells in terms of efficiency are the transparent-conducting (TC) and absorber layers. This is particularly true in the case of thin-film cells that represent a potential low-cost path to scalable photovoltaic (PV) technology. Indeed, on the one hand, not all incident light passes through the TC layer and reaches the absorber layer, and, on the other hand, not all the generated electron-hole pairs are extracted and collected from the latter. In this thesis, we describe the impact of point defects on the electronic and optical properties of the materials used in PV applications. We discuss how ab initio calculations using hybrid functionals can help to explain the observed signatures from experimental studies. We demonstrate the importance of including such first-principles point-defect computations in the search for new efficient TC materials and PV absorbers. In addition, we examine the viability of high-throughput approaches for calculating point-defect properties based on semi-local DFT approaches. Finally, we extend the amount of performed computations from a small set of configurations to a dataset of thousands of compounds., (FSA - Sciences de l'ingénieur) -- UCL, 2022

In the coming years preventive medicine and medical treatment will strongly rely on wearable and implantable biomedical devices. Advances in manufacturing, energy harvesting systems, and information technologies are contributing to a transformation in the mobile health era in terms of healthcare management and real-time, personalised tracking of the health status. Thus, the demand of wearable devices for monitoring and treatment of chronic diseases such as diabetes in remote places is expected to drastically increase. Specifically, the design and development of next-generation wearable devices for the diabetes treatment involves multi-disciplinary fields such as transdermal drug delivery, radio-frequency energy harvesting, and embedded, low-power electronics. In this thesis, we present the integration of various concepts from these areas relevant for wearable healthcare technologies to progress towards wearable devices for monitoring and treatment of diabetic patients in remote places. First, our efforts have been focused on the design, simulation, and fabrication of planar, small and flexible antennas for both radio-frequency energy harvesting and wireless communication purposes, focusing on the unlicensed band of 2.45 GHz. Next, a novel painless patch design is presented as a transdermal insulin delivery system. The patch is based on the integration of nano-engineered heating elements - gold nanomesh thin layers - on polyimide substrates for transdermal insulin therapy by wireless, electrothermal actuation. A thin film of reduced graphene oxide integrated onto the nanoheater provides the drug reservoir. Numerical computations to analyse the Joule heating of the patch in contact with the human skin were performed. Finally, the architecture and the design of the wearable electronic system to control the electrothermal patch with a smart phone are described, considering that the transdermal patch is powered via batteries and radio-frequency energy harvesting., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Modern cryptography relies on the Kerckhoff’s principle which states that everything about a crypto-system should be public, except for the secret keys. In this thesis, I apply these principles to cryptography implementations by providing the evaluator full knowledge of the implementation, in a worst-case manner. As a result, it allows to estimate their long-term security against side-channel attacks. The first part is dedicated to the verification of the two underlying assumptions in masking proofs. More precisely, I contribute to proof-based evaluations by putting forward that i) the verification of independence thanks to leakage detection benefits from multivariate statistics ii) the noise can be estimated in a quantitative manner thanks to formal bounds on the mutual information. The second part of the thesis is dedicated to efficient methodologies to perform worst-case attacks which are a useful shortcut for evaluators since they allow to estimate, with a reduce profiling data complexity, the online complexity other attacks strategies. Concretely, I propose worst-case attacks against a recent open-source AES implementation by the French ANSSI and masked higher-order bitslice implementations. From the lessons learned from the side-channel evaluation of current proposals, I propose new designs to obtain strong side-channel protection. More precisely, the noise versus security trade-off can be improved thanks to a sound combination of masking and shuffling. Concretely, I apply masking and shuffling to ISW multiplications in order to exponentially amplify with masking the effect of shuffling., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The prior knowledge (a.k.a. priors) integrated into the design of a machine learning system strongly influences its generalization abilities. In the specific context of deep learning, some of these priors are poorly understood as they implicitly emerge from the successful heuristics and tentative approximations of biological brains involved in deep learning design. Through the lens of supervised image classification problems, this thesis investigates the implicit integration of a natural clustering prior composed of three statements: (i) natural images exhibit a rich clustered structure, (ii) image classes are composed of multiple clusters and (iii) each cluster contains examples from a single class. The decomposition of classes into multiple clusters implies that supervised deep learning systems could benefit from unsupervised clustering to define appropriate decision boundaries. Hence, this thesis attempts to identify implicit clustering abilities, mechanisms and hyperparameters in deep learning systems and evaluate their relevance for explaining the generalization abilities of these systems. We do so through an extensive empirical study of the training dynamics as well as the neuron- and layer-level representations of deep neural networks. The resulting collection of experiments provides preliminary evidence for the relevance of the natural clustering prior for understanding deep learning., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The commensal skin bacterium Staphylococcus epidermidis provides a range of benefits to human hosts, contributing to skin equilibrium and good health. However, it can also act as a pathogen when entering the host body. Encapsulation is thus considered as a way to maintain the benefits of bacteria and control bacteria dispersion. Among various materials, porous polycarbonate (PC) membranes modified by layer-by-layer (LbL) assembled multilayers attract an increasing interest due to their biocompatible fabrication process and tunable properties. In this thesis, two membrane-based methods have been studied to entrap bacteria. In the first method, a microtube mat is fabricated to encapsulate S. epidermidis. A bicompartmentalized LbL microtube, which contains an erasable external compartment, is designed to release the inner tube without having to dissolve the membrane template in non-biocompatible organic solvents. A hydrogen-bonded poly(N-vinyl caprolactam) (PVCL)/poly(methacrylic acid) (PMAA) system is chosen to build this erasable compartment. We investigate the growth and disaggregation of PVCL/PMAA LbL films, depending on pH, NaCl concentration, and PVCL molar mass. The use of PVCL/PMAA multilayers as a sacrificial compartment is proved to provide an efficient release of nano/micro-structures in biocompatible close-to-neutral aqueous conditions. S. epidermidis can be entrapped in LbL microtube under optimized loading conditions. After tube liberation and collection, the resulting microtube mat is shown to maintain the viability of S. epidermidis. However, the bacterial metabolic activity remains limited. Therefore, a second method of entrapping S. epidermidis in a membrane-in-gel patch is studied. S. epidermidis is loaded in LbL-modified membrane, followed by coating the membrane with a thick layer of agarose gel. The membrane-in-gel maintains the metabolic activity of bacteria. LbL multilayers comprising antibacterial polycations are then deposited over the patch., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Flows involving the transport of a dispersed solid phase by a carrier fluid are encountered in many environmental phenomena and industrial applications. Fluidized beds are now used in a wide variety of processes in the petrochemical and pharmaceutical sectors, in biomass conversion or in the food industry, among others. Those systems are characterized by a very large range of length and time scales. At the smallest scale, particle-resolved direct numerical simulation (PR-DNS) is a first-principles based approach, providing the complete details of the flow. With growing computational resources, PR-DNS has become a tool of choice to develop closures for the mass, momentum and energy transfer terms that need to be modeled in the large-scale Euler-Lagrange and Euler-Euler approaches. In this thesis, we first develop and validate a direct-forcing immersed boundary method to perform particle-resolved simulations of two- and three-dimensional fluid-particle flows. The method is then used to quantify the interfacial momentum and heat transfer terms in the flow past random particle arrays. While current drag laws only provide the mean value of the fluid-particle force, we propose a microstructure-based model to estimate the distribution of the force using a few quantities characterizing the local organization of the solid phase. The inhomogeneity of the thermal problem and the impact of thermal saturation on the distribution of the fluid-particle heat flux are also discussed. Finally, we perform a detailed comparison of the dynamics of small-scale fluidized beds in different regimes. In liquid fluidization, a homogeneous suspension is observed, owing to the small inertia of the particles. In gas fluidization, the higher inertia of the solid particles leads to the emergence of heterogeneities in the bed (bubbles and clusters) and to strong oscillatory motions. The impact of particle agitation on the fluid-particle force and fluid-particle heat flux is studied, as well as the, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Most of our interactions with the world occur through reaching movements. These movements were extensively studied in the past decades and revealed rich mechanisms both during planning and execution. Despite these numerous years of research, very little is known about the entanglement between the planning and execution facets of reaching movements. This thesis explores the interactions between the planning and execution steps of these reaching movements through a combination of experimental and modeling approaches. First, we demonstrate that the control policies selected to execute movements are not immutable and can be adjusted online to accommodate changes in task demands. These online adjustments suggest the existence of a mechanism whereby changes in the task could elicit replanning of movement. Elaborating on this first finding, we then report that this adjustment can also handle dynamical changes in the task. This finding reinforces the first one as it demonstrates that this adjustment mechanism consists in a continuous feedback loop monitoring task demands to optimally adjust the control policy online. Then, we reveal that selecting more robust control policy, intended at rejecting potential disturbances, reduces flexibility during movement. This last result demonstrates a competition between movement vigor and flexibility. This work adds to the understanding of the control policies underlying reaching movements by providing new insights into the interactions between movement planning and execution. These results, combined with the modeling work, highlight the fact that movement planning and execution are, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Flexible PCB technology allows to develop new types of windings to improve the performance of electrical motors. By replacing the usual wire winding with a flexible printed circuit board, it becomes possible to design copper tracks with complex shapes which was not the case in the past. Therefore, these additional degrees of freedom can be exploited for motor optimization. Specifically, this work focuses on the case of slotless micromotors with perma- nent magnets, for which flexible PCB are particularly well suited. The choice to use magnets is motivated by the developments of rare-earth materials which allow to obtain a high power density. The choice of the slotless topology is driven by the fact that, on the one hand, this topology leaves room for a completely free arrangement of the conductors in the airgap, and on the other hand, the absence of ferromagnetic teeth presents numerous advantages, for example, reduction of iron losses, cogging torque, vibrations or even of the noise. Although the PCB winding outperforms the classical wire winding at low speeds, it is more prone to eddy currents as the motor speed increases. Indeed, for the same conductor cross-section, the printed copper tracks are wider than classical wires and therefore larger conductive areas are exposed to a perpendicular magnetic flux varying over time. The objective of the thesis is therefore to model the eddy current losses in order to be able to estimate and reduce their impact during motor design and optimization process. The first part of this thesis deals with the modeling of the eddy currents. A refer- ence model using 3D finite elements and based on a vector potential formulation of magnetodynamics is primary explained. Starting from this model, it is shown that, knowing the airgap permanent magnets magnetic field, it is possible to estimate the eddy current losses using a 2D FEM model which is much faster and requires much less computational efforts. This model, in addition to being f, La technologie des PCB flexibles permet de développer de nouveaux types de bobinages qui améliorent les performances des moteurs électriques. On peut en effet remplacer le bobinage filaire habituel par un circuit imprimé, sur lequel il est possible de dessiner des pistes de cuivre aux formes très variées, ce qui amène de nouveaux degrés de liberté exploitables pour l’optimisation des moteurs. Ce travail se concentre plus précisément sur le cas des petits moteurs synchrones à aimants permanents avec entrefer lisse, pour lesquels les PCB flexibles sont particulièrement bien adaptés. Le choix d’utiliser des aimants se justifie par le développement des terres rares qui permettent maintenant d’obtenir une grande densité d’énergie. Le choix d’une topologie avec un entrefer lisse procure de nombreux avantages et est guidé par le fait que, d’une part elle laisse la place à un arrangement des conducteurs complètement libre, et que d’autre part elle permet la réduction des pertes fer, du couple de dentures, des vibrations ou encore du bruit, grâce à l’absence de dents ferromagnétiques. Bien que les bobinages sur PCB aient démontré leur efficacité à faible vitesse, on y observe rapidement l’apparition de courants induits lorsque la vitesse du moteur augmente. Car à même section de conducteur, les pistes de cuivre imprimées sont plus larges que le fil que l’on retrouve dans des bobinages standards et par con- séquent de plus grandes surfaces conductrices sont exposées perpendiculairement au champ magnétique des aimants en rotation. L’objectif principal de la thèse est donc de modéliser les pertes par courants in- duits pour les estimer et réduire leur impact lors de la conception et de l’optimisation des moteurs. La première partie de la thèse concerne la modélisation de ces pertes. Un modèle de départ utilisant des éléments finis 3D et, s’appuyant sur une formulation de la magnétodynamique en potentiel vecteur, est présenté. Partant de ce modèle de référence, dès lors que l’on, (FSA - Sciences de l'ingénieur) -- UCL, 2022

This thesis aims at curvilinear mesh generation and adaptation, we start from summary ideally the research project as a simple yet fundamental question: assume a unit square and a smooth function defined on the square, and consider a mesh made of P2 triangles that exactly covers the square, how can we compute the mesh that minimizes the interpolation error. We state the problem as to build a unit curvilinear mesh, i.e. build a mesh with unit edge lengths that are possibly curvilinear. We solve the problem in three stages of increasing complexity: 1) input a unit square and a metric field, to build a unit curvilinear mesh that is possibly anisotropic; 2) input a function, to build an anisotropic curvilinear mesh that minimizes the approximation error; 3) input a high order finite element solution, to build an anisotropic curvilinear adapted mesh. We propose a new framework of curvilinear mesh generation and adaptation: metric field construction, generation of points (point sampling on the boundary and point sampling in the domain), straight-sided mesh generation and adaptation (triangulation and straight-sided edges swap), curvilinear mesh generation, and adaptation (straight-sided edges curving, curvilinear edges swap, and Curvilinear Small Polygon Reconnection). A unit curvilinear mesh containing only valid ``Geodesic Delaunay triangles'' is obtained this way. In this approach, the curvature is not only used to match curved boundaries but also to capture features of the interpolated solutions, and it results in meshes that would not have been achievable by simply curving a posteriori a straight-sided mesh. A number of application examples are presented in order to demonstrate the capabilities of the mesh adaptation procedure., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Over the past decades, continued improvement in transistors density, performance and energy efficiency have been sustained by technological innovations. This exponential development predicted by Moore in the 60s transformed the first crude computers from the 70s into the ubiquitous and high-performance sophisticated machines of nowadays. This trend is however coming to end as the size of transistors is reaching the quantum limit. In the last decade, two-dimensional (2D) crystals have emerged as a disruptive materials platform. Those layered crystals are excellent candidates for next-generation technologies due to their ultimate thinness and dangling-bonds-free surface. In addition, they offer better scaling than traditional 2D electron gas based on III-V compounds. In spite of the aforementioned advantages, some technological obstacles still hinder the industrial integration of 2D materials in electronics. The most important ones are the growth and the post-processing of wafer-scale high-quality crystals, robust passivation schemes and device-related aspects such as the contact resistance or doping. In this thesis, first-principles simulations are performed on systems composed of graphene or transition metal dichalcogenides (TMDs). Their transport properties are computed using the OpenMX code based on the Green's functions formalism. First, metal-graphene interfaces are studied in the top-contacted geometry. Charge injection from metal to graphene is understood through the lens of a semi-empirical transport model so that the transport mechanisms impeding the contact resistance are identified. Afterwards, quantum transport is investigated inside the graphene channel. The objective is to benchmark the transport characteristics of polycrystalline samples resembling the ones produced by the industry. Finally, the operation of a TMDs-based atomically thin FET is investigated. Stain-induced current modulation is predicted through nano-scale quantum engineering, a mechanis, (FSA - Sciences de l'ingénieur) -- UCL, 2022

User interfaces (UI) are one of the main media of interaction with the devices we use on a daily basis. Evaluation of UIs is and will remain a topical research domain as long as existing evaluation methods are refined, improved and new methods appear to evaluate new types of UIs. However, not all the existing evaluation methods include explicit, formal, calculable and interpretable quality models that are a necessary condition to ensure their replicability. The large variety of evaluation methods and studies coupled to other various factors such as the hardly scalable and error-prone process of dataset creation led to a replicability crisis in the field of Human-Computer Interactions. In order to address the aformentioned challenges and concerns, there is a need to provide the field of experimental studies on GUI visual design with methodologies and tools that make the process of dataset creation efficient and realistic to foster the replicability of experimental studies. The purpose of this thesis is to provide a way of evaluating the quality of GUIs in an explicit, formal, calculable and thus reproducible way. This thesis introduces, motivates, defines, implements and validates a methodology, based on a conceptual model and supported by a software tool, for producing explicit, formal, calculable and interpretable quality models for evaluation of GUI visual design., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Essential to space missions involving an atmospheric entry, the thermal protection system (TPS) shields the spacecraft and its payload from the severe aerothermal loads. Low-density carbon/phenolic composite materials have gained renewed interest to serve as ablative thermal protection materials (TPMs). These materials can accommodate the high heating rates and heat loads encountered during the atmospheric entry, at hypersonic velocities, by absorbing part of the incoming heat through physico-chemical transformations. One of the main endothermic processes is the pyrolysis of the resin compound, whereby volatile products are released, leaving a carbonaceous residue is left on the fibers. Whereas new experimental data have already been published to characterize the decomposition of these low-density carbon/phenolic materials, they are yet to be exploited for the inversion of physico-chemical models. In addition, the issue of uncertainty quantification, required to assess the reliability of the numerical model and the physico-chemical models, is yet to be addressed. Therefore, the overarching objective of this thesis is to contribute to the development of an uncertainty-quantified numerical modeling of the ablation of new porous composite materials and to the analysis of the impact of uncertainty on TPS design. To that aim, we first address the development and the uncertainty characterization of physico-chemical models for resin pyrolysis on the basis of new experimental data relevant to the pyrolytic decomposition of the phenolic resin used in carbon/phenolic composite TPMs. Then, we analyze the impact of the uncertainty in the physico-chemical models on the numerical modeling of ablation of TPS by means of non-intrusive stochastic methods. The central contribution of this thesis is to infer from these new experimental data an uncertainty-quantified pyrolysis model. We adopt a Bayesian probabilistic approach to account for uncertainties in the model identification. We, (FSA - Sciences de l'ingénieur) -- UCL, 2022

In graphene, charge carriers behave as massless (Dirac) fermions, yielding unprecedented electrical properties. In this thesis, the electronic transport of Dirac fermions is studied in graphene quantum point contacts thanks to low-temperature scanning gate microscopy (SGM), a technique consisting in recording the sample resistance while changing locally the charge carrier density with a biased metallic tip. The focus has been put on backscattering mechanisms and three main directions are explored. First, real space signatures of Klein tunneling, preventing the backscattering of Dirac fermions, are investigated by creating a movable pn junction with the SGM movable top gate. The experimental results are discussed in the light of simulations, revealing that the SGM conductance maps yield an image of the current density around and through the lens, with a direct evidence of Klein tunneling. Second, SGM characterizations are reported under a large magnetic field, in the quantum Hall regime where charge carriers flow in topologically protected quantum Hall edge channels (QHECs). The recorded SGM signatures indicate that counterpropagating QHECs, separated by a few hundreds of nanometers, exist along the same edges in graphene and that charge carriers backscattering is achieved by coupling these QHECs through the localized states of antidots located between them. Third, these antidots are shown to act as nano-sized Fabry-Perot interferometers. A simple model is used to reproduce the experimental results, showing that the signatures associated to two distinct regimes, namely the Aharonov-Bohm and Coulomb dominated regimes, can be explained in a single framework ignoring the Coulomb interactions., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Sensitivity to process, voltage, and temperature (PVT) variations constitutes a serious obstacle in ultralow-voltage/ultralow-power (ULV/ULP) circuits and systems. In this thesis, we address this challenge in the domains of clock generation and frequency synthesis by exploiting the forward back-biasing (FBB) feature of 28-nm FDSOI CMOS technology. We tackle three questions: - How do FDSOI-enabled back-bias-controlled oscillators (BBCOs) compare to conventional current-starved ring oscillators (CSROs)? - How to leverage FBB in BBCO-based phase-locked loops (PLLs)? - How to exploit FBB in adaptive regulation loops for unified clock generation and PVT compensation in ULV/ULP circuits? We first show that BBCOs can intrinsically reach lower power and slightly better power/noise compromises compared to CSROs. We then identify several design challenges linked to the practical use of BBCOs in feedback loops and illustrate how to tackle them by proposing three silicon prototypes. Embedded within a 2.5 Gb/s ultra-wideband radio, our first demonstrator is a BBCO-based dual-PLL clock recovery unit. Thanks to FBB, it reaches a competitive energy level and an excellent power/jitter trade-off. Our last two prototypes are custom ULV/ULP microcontroller units (MCUs) which benefit from a unified frequency and back-bias regulation scheme. This new adaptive technique has limited power/area overheads while providing many advantages at the MCU level: PVT compensation, robustness against voltage droops, fast transitions between active and sleep modes, and energy efficiency., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Industrial Control Systems (ICS) are computer systems used for monitoring and controlling industrial facilities such as water treatment facilities, power plants, manufacturing factories. . . . Historically those systems were isolated but for business and management purposes, they got connected to other net- works. This connection brought the vulnerabilities of other networks to ICS which resulted in an increasing number of attacks against those systems. Consequently, the security of ICS has become an active research field. To im- prove the resilience of ICS against cyber-attacks, researchers are focusing on designing protection mechanisms such as Intrusion Detection Systems (IDS). In this thesis, we present two IDS solutions for ICS to consider both the cyber aspect and the physical aspect of ICS. The first solution focuses on the detec- tion of attacks targeting the network infrastructure of the ICS and leverages Software-defined Networking (SDN). SDN is a network paradigm that pro- vides a high-level of flexibility in term of network management. We use this property to improve a technique called Flow-Whitelisting. The second solution focuses on process-oriented attacks, meaning attacks that target physical processes. The IDS learns the temporal properties of a physical process to detect the disruptions caused by an attack. It analyzes the variables defining the physical process and monitors their behavior over time. Any deviation from the learned temporal properties triggers an alert. Each IDS is evaluated with several scenarios and gives interesting results. In addition, we provide a network dataset from a real system but also a tool to generate new datasets that can be used for the evaluation of network IDS. To show the usefulness of the dataset, we evaluate several IDS in the literature., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The control and characterization of the fracture behavior of thin films and 2D materials, in the era of continuous device miniaturization, are of paramount importance for the design and integrity assessment of many coatings and microelectronics devices towards more reliable devices. However, the available test methods suffer from many limitations mainly when the film rests on a substrate due to different constraints that are difficult to deconvolute and or with the introduction of an initial sharp precrack in freestanding configurations. In this context, the objective of this research is to develop a robust on-chip fracture technique for freestanding thin films and 2D materials. This crack-on-chip method consists of actuator beams produced by lithography undergoing large internal stress attached to a notched specimen beam. Both actuators and specimens are deposited on top of a Si substrate. The etching of a part of this substrate induces the release of the test structure, with the actuators then contracting and pulling on the test specimen. If the applied displacement is large enough, a crack initiates at the notch root and propagates. The design has been selected such as to involve a decrease of the crack driving force after some amount of crack growth leading to crack arrest, avoiding the problem of notch opening of the precrack. The technique can also be used to investigate the environmentally-assisted cracking in thin-film materials. The fracture toughness of SiO2 (142 nm), SiN (50 nm), Al2O3 (190 nm), and CVD-grown monolayer graphene was extracted. It was found that Al2O3 films are more prone to subcritical aging compared to SiN, while simultaneously exhibiting larger resistance to environmental cracking than thermally-grown SiO2 films. For single-layer graphene, fracture strain, Young’s modulus, strength, and Weibull modulus were determined with, for the first time, statistically representative data., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Thermoplastic elastomers (TPEs) are an extremely versatile class of engineering plastics that are suitable for many applications thanks to their high extensibility, reversible elasticity, and tunable hardness. They are widely used as fibers, cable insulators, medical devices, shoe soles, and many other applications. Additionally, like many other thermoplastic polymers, they are recyclable and can be easily processed. However, regardless of their exact chemical structure, many TPEs show a decline in their mechanical properties, such as toughness and ductility, at high temperatures. This behavior is in sharp contrast with standard thermoplastics and limits the range of applications where TPEs can be used as a replacement for thermoset rubbers. Understanding the different dynamics behind this behavior and finding out the key parameters that influence the high-temperature mechanical properties is critical to designing new TPEs and targeting new applications where traditional rubbers can be replaced. In this thesis, we explore the high-temperature and high-deformation mechanical properties of a model series of industrially relevant soft TPEs. Through a series of mechanical and morphological tests, we investigate the fundamental relationships between microstructure and mechanical properties. The ultimate aim is to draw a unifying picture that can describe which dynamics influence the deformation behavior and how they interplay with each other., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The problem of estimating a quantity that evolves with time from noisy measurements can be found in many fields, such as engineering (e.g., the position of a drone), epidemiology (e.g., the number of infected people), or medicine (e.g., the position of a lung tumor moving due to breathing). In some cases, the number of measurements must be reduced because the acquisition of measurements is expensive, energy-consuming, time-consuming or dangerous. In this thesis, we assume that the amount of measurements is limited, and we study the problem of the optimal choice of measurement times to estimate the state of a stochastic dynamical system over a finite time horizon. We first focus on the case of linear and Gaussian dynamical systems. We propose the optimal intermittent Kalman filter for which the measurement times are obtained by solving a combinatorial optimization problem. Then, we generalize this approach to non-linear and non-Gaussian dynamical systems. This leads to the formulation of three variants of optimal intermittent particle filters. The first one, for which the measurements are time-triggered, is based on combinatorial optimization. The other two are self-triggered. One requires a combinatorial program to be solved for each measurement, and the other requires to solve a stochastic program to choose when to perform a measurement. Robust estimation of the state of a linear dynamical system subject to bounded disturbances is also studied. Using the polytope containment method and Youla-parameterization, we show that the measurement times that minimize the worst-case estimation error are obtained by solving a mixed integer linear program. Numerical and empirical experiments are presented, with a particular focus on the problem of tumor tracking and its use in radiation therapy., (FSA - Sciences de l'ingénieur) -- UCL, 2022

In recent years and in the scope of Fifth Generation (5G) and Sixth Generation (6G) technologies, the joint deployment of Massive MIMO (mMIMO) and millimeter waves (mmWave) technologies has shown very promising results in the quest to meet the very high consumer data rate demands. To design real mmWave mMIMO wireless systems and predict their performances under certain conditions, it is imperative to have accurate mmWave mMIMO wireless channel models that capture the key characteristics for different scenarios. Very high gain and directive Transmitting Reconfigurable Intelligent Surface (T-RIS)s with electronic beamscanning capabilities are also expected to be ubiquitous in 5G and 6G wireless communication systems and hence the need to investigate the channel and T-RIS joint effect. This thesis presents different channel characterizations and models for indoor mmWave mMIMO radio propagation channels based on stochastic approaches. Different channel measurement campaigns were carried out in Indoor Laboratory (InLab), Indoor Open Office (InOpOffice), Indoor Mixed Office (InMixOffice), corridors and Indoor Factory (InF) environments. Massive virtual arrays and T-RISs were employed in the channel measurements. Line of Sight (LOS), Obstructed Line of Sight (OLOS) and Non Line of Sight (NLOS) propagation conditions were investigated in these environments. For the large virtual arrays’ measurements, an investigation on the impact of the array size and geometry on the mmWave mMIMO channel characteristics is performed. Spatial channel correlation over the massive array is investigated and Multipath Components (MPCs) are estimated by means of a high resolution detection algorithm. The estimated specular MPCs parameters are then used for further analysis. Dense Multipath Components (DMCs) are quantified. Common MPCs between different sub-arrays of a massive array are identified following a joint temporal, power and angular threshold criterion and their distributions are then a, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Modern cryptography has been widely deployed in the last decades, allowing any computing device to secure its communications. Facing the strength of the cryptographic algorithms, attackers turned to their implementations. In particular, side-channel attacks have been shown to threaten the security of cryptographic embedded devices by exploiting information leaked through physical characteristics such as power consumption or electromagnetic radiation. In order to mitigate side-channel attacks, implementations use countermeasures, among which masking is the most investigated choice. In this thesis, we design and analyze masking schemes with the goal of having efficient masking schemes with strong and well-understood security. Efficiency is an important criterion for masking schemes, since they sometimes have orders of magnitude performance overheads over an implementation without countermeasures. Regarding security, our approach is to prove the security of the masking scheme as comprehensively as possible by analyzing complete masked algorithms in leakage models which are well-aligned with the characteristics of real-world leakage. We first work in the threshold probing model, which is the simplest and most widely used side-channel leakage model. Our focus is the problem of composability, that is, the possibility to build complex algorithms from small building blocks while preserving the security. Concretely, we introduce a new composable security definition, design efficient constructions that satisfy it, and use them to build more complex circuits. Next, we adapt our results from the threshold probing model to the robust probing model. This model extends the threshold probing model by taking into account physical phenomena named glitches and transitions. These phenomena that appear in concrete hardware implementations violate the independence assumption of the threshold probing model. Finally, the threshold and robust probing models fix a bound on the amount of leak, (FSA - Sciences de l'ingénieur) -- UCL, 2022

In the recent years, the saturation of the smartphone market penetration fostered the development of new applications to connect human beings with smart objects. However, the explosive growth in the number of connected devices brings some concerns about the highly growing traffic and the environmental effect of batteries production and disposal. Simultaneous wireless information and power transfer (SWIPT) technique has been emerged as a solution to mitigate with these concerns. However, SWIPT also comes with some obstacles such as the limitations on the energy transfer efficiency due to the propagation losses and the constraints on the minimum received power imposed by the current radio frequency (RF) harvesting technologies. This restricts the range of application of wirelessly powered devices by limiting them to the execution of very simple computational tasks. Following this line of thought, the integration of SWIPT with mobile edge computing (MEC) has appeared as an enticing emerging research direction to increase the computational capabilities of SWIPT enabled systems. This dissertation focuses on studying the performance of SWIPT enabled MEC system in the presence of spatially correlated shadowing. The first step taken towards this objective is the analysis of the performance of SWIPT enabled systems with special emphasis on the impact of spatially correlated shadowing on the achievable rate-energy trade-off. Stochastic geometry is used to model the random location of the network nodes and the geometry of the propagation environment. Capitalizing on the analytical tool developed in the first part of the dissertation, the analysis of the achievable performance trade-offs has been extended to SWIPT-enabled computing systems. The last part of the dissertation addresses the study of SWIPT-MEC systems and provides an analytical tool to compute the probability of successfully execute a task as a function of the network parameters. The last chapter is devoted to a co, (FSA - Sciences de l'ingénieur) -- UCL, 2022

With the increase in the adoption of Cloud computing and big data processing systems, a common way to deploy analytics workloads is to acquire on-demand resources in a Cloud environment whenever the workloads need to execute. Public Cloud providers offer numerous infrastructure choices to Cloud users. For example, the number of different Virtual Machine (VM) instances offered by the Cloud providers has been steadily increasing, now numbering in dozens. This enables Cloud users to have access to the type of instance that fits their use case. However, this blessing can also be a curse since selecting the right resources is often not straightforward. When executing analytics workloads in the Cloud, ensuring the allocation of the right resources is paramount to achieve cost efficiency and satisfy strict service-level objectives (SLOs) on job completion times. This requires users to decide the number of instances (VMs) and type of instances for their deployments (together forming a Cloud configuration). In addition, the software platforms for big data processing have system parameters that need to be appropriately set. The choice of these cloud configurations and system parameter values dictate the performance and cost of the whole deployment. This thesis aims to reduce the decision burden on the users when deploying workloads in the Cloud. Our primary focus is to allow users to concentrate on defining the performance objectives rather than determining the resource allocation for their Cloud workloads. Specifically, we investigate ways to automatically determine configurations for Cloud workloads given user-specified performance and cost objectives. These workloads include jobs running on batch and stream processing systems and serverless functions. The configurations under consideration in this thesis include cloud configurations and system parameters. More concretely, in this thesis, we discuss automatic Cloud configuration optimization in two contexts. First, for dist, (FSA - Sciences de l'ingénieur) -- UCL, 2022

High entropy alloys (HEAs) have attracted a lot of attention in recent years, with the promise of new alloy systems with outstanding (combinations of) structural (and functional) properties. However, in-depth investigations of HEAs mechanical properties in direct comparison to conventional alloys have been seldom reported. The goal of this thesis was to characterize two so-called Cantor-based HEAs, CoCrFeMnNi and CoCrNi alloys, and to compare them to reference conventional alloys in the framework of structural materials for cryogenic applications, namely stainless steels (type 304L and 316L), iron-nickel alloys (Invar and Fe9Ni), and a high Mn TWIP steel (Fe-22Mn-0.5C). To this end, the selected alloys were tested at room and cryogenic temperatures, and thoroughly characterized in terms of microstructure, hardening and fracture properties. The essential work of fracture (EWF) method is well suited to quantify the fracture resistance of ductile thin sheets and was extensively used in this work. One of the outcomes of the thesis is an improved EWF methodology that minimizes the required material while keeping the same statistical error on the extracted EWF value. Among the seven investigated materials, HEAs showed excellent tensile and fracture properties both at 293 K and 77 K. However, stainless steels keep exhibiting even better performances in terms of strength, ductility, and fracture toughness., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The high-quality two-dimensional (2D) crystals possess remarkable mechanical behaviour which offers exciting opportunities for new material selection and design, like the combination of extremely high in-plane stiffness and bending flexibility. For this reason, nanomechanical testing of 2D layers and their 3D heterostructures is vital in order to prove that they can be practical and useful in different applications. In this work, different nanomechanical testing including membrane deflection, nanoindentation and tensile tests have been performed with various techniques such as atomic force microscopy, nanoindentation and lab-on-chip developed in UCLouvain, respectively. The elastic and plastic behaviour of graphene as well as its effect on the mechanical response of copper thin film have been investigated by using these methods. For each technique, some challenges are required to be faced and some uncertainties needed to be considered in order to extract the mechanical response precisely. Membrane deflection on single and bilayer graphene has been performed in order to extract Young’s modulus and fracture strength of these layers by considering the uncertainties and errors that can occur regarding the device calibration and experimental procedure. The effect of CVD-graphene on the plastic behaviour of Cu thin film in two different configurations has been investigated by nanoindentation in two different modes which are complementary to each other limitations. In this configuration, graphene acted as a perfect barrier against the movement and propagation of dislocations underneath the indent. In the last part, lab-on-chip (LOC) technique has been used to evaluate the effect of graphene on the fracture and creep behaviour of thin film of copper. Although convincing and reliable experimental data could not be generated, a well-defined, adapted process for Cu/Gr specimens with Ni actuators has been proposed and the solutions for most of the challenges in each step have b, (FSA - Sciences de l'ingénieur) -- UCL, 2022

The use of microbial fuel cells (MFCs) as a bioelectrochemical system (BES) for the treatment of textile wastewater has attracted attention due to the ability to produce bioelectricity from wastewater treatment. However, textile wastewater is far from being an ideal electrolyte for an electrochemical process. It can generate biofouling on electrodes and separators (membranes), affecting greatly the efficiency in such systems. In recent years, different types of membranes have been explored as separators in MFCs, although to our knowledge no nanofiltration membranes have ever been used in such systems. Nanofiltration membranes share advantages with porous and non-porous membranes. In addition, the wide variety of nanofiltration membranes, high oxygen retention, and lower electrical resistance compared to ion-selective membranes normally used in BES, made these types of membranes potential separators. In this work, an integrated process is proposed, including three technologies, i.e., membrane technology, an advanced oxidation process, and microbial fuel cells. Antifouling and self-cleaning membranes are developed as efficient separators for a new generation of osmotic systems (nanofiltration microbial fuel cells) in textile wastewater treatment. A new method of modification of bio-inspired membranes with commercial catalysts (TiO2 and ZnO) showed that the modified membranes acquired antibacterial, self-cleaning, and antifouling properties, improved when the catalysts were combined. The application of this method on low-cost membranes exhibited a significant improvement in COD and color removal, as well as an improvement in bioelectricity production in the MFCs. In order to further improve the activation efficiency of the photocatalysts under visible light and the antifouling properties of the membranes, a new catalyst (ZnO Er/Al) was synthesized, which demonstrated high photocatalytic and antibacterial activity. The new membranes also demonstrated efficient resistanc, (FSA - Sciences de l'ingénieur) -- UCL, 2022

From the child playing with its toys to the surgeon holding its scalpel, object manipulation in everyday life is omnipresent. This skill relies on collecting information on the state of the manipulated object using our senses, with touch being of first importance. Using the information provided by touch, our central nervous system sends commands to our muscles to perform movements that take into account how slippery an object is, its shape, the characteristics of the desired movement, and more. In this thesis, we studied how finger skin deformations influence object manipulation, as touch and skin deformations are closely linked. First, we present a novel device that allows for the measurement of fingerpad skin deformations during active manipulation and present the link between the grip force and skin deformations during a simple oscillation task. Then, we show how localized fingerpad strains allow to quickly adapt the grip force to the friction when lifting objects. Finally, we show how the grip force is adjusted so that the upper bound of the amount of partial slip and finger skin deformation remain similar across friction conditions. This work contributes to the understanding of the role of touch during object manipulation and provides useful information for the design of friction sensors and somatosensory prostheses., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Large gas power plants are—and will be in the years to come—a major element of power system operations. On one hand, for the flexibility they offer, which is most valuable for the massive integration of renewable energy sources in the energy mix and the associated uncertainty linked to the production of such sources. On the other hand, they can serve as a replacement for nuclear power plants, following the policy of some European countries like Germany and Belgium. These large gas power plants often obey the valve point effect. This physical effect is due to the increase of throttling losses when operating a unit off a valve point, that is, just after opening one of the several fuel admission valves. We show how the consideration of this physical effect makes the dispatch a nonconvex and nonsmooth optimization problem and propose algorithms that aim at efficiently finding a solution as close as possible to the global optimum, along with some guarantees. In the first part of the thesis, we present a three-step algorithm. The first step solves a relaxation of the problem to obtain an infeasible solution that we expect to be close to the feasible set, along with a lower bound to the global optimum. Then, the second step projects this infeasible solution onto the feasible set. Finally, the last step amounts to locally improving the feasible solution that is obtained in the second step via a Riemannian subgradient descent scheme. In the second part of the thesis, we further analyze the second step. The problem is posed abstractly as the projection of a given point onto the intersection of a quadratic hypersurface—or quadric—and a box. We show how to compute the exact projection onto a central non-cylindrical quadric and use splitting methods for the projection onto the full set., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Cognitive radio (CR) is the solution to the spectrum scarcity issue faced in wireless communications. It allows unlicensed secondary users (SUs) to opportunistically access the spectrum assigned to licensed primary users (PUs), but unused, on a non-interfering basis. Spectrum sensing is the process by which the CR node becomes aware of the spectrum occupancy, in order to decide which frequency bands to use. In the first part of this thesis, we consider the Bayesian changepoint detection theory to propose a new spectrum sensing algorithm that exploits the mobility of the SU and works for practical scenarios where the PU’s signal power is unknown to the SU. Simulation results show that the derived algorithm is a good choice when the SU could be required to detect very low signal-to-noise ratio (SNR) signals. In the second part, we introduce a new framework for joint transmission and sensing in mobile CR, by using changepoint detection for spectrum sensing. The optimal system parameters (sensing time, transmission time and detection threshold) that maximize the spectrum utilization are numerically computed and analyzed. In the last part, a new spectrum sensing algorithm, that exploits geolocation information about existing PUs by using the Bayesian block-based detection theory, is introduced. An approximate closed-form expression, based on Taylor series expansion, is provided. Numerical results show that the derived algorithm gives the minimum error probability when compared with an algorithm that does not use geolocation information., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Coral populations have declined dramatically worldwide under the combined effects of climate change and local anthropogenic stressors. In addition to ocean warming and acidification, threats to coral reefs in the Caribbean include frequent and virulent disease outbreaks, and intensifying hurricanes. In particular, Florida's Coral Reef (FCR) is currently facing a multi-year outbreak of the stony coral tissue loss disease (SCTLD). First observed in 2014 during the monitoring of the PortMiami Deep Dredge Project (PMDDP), the disease has now spread through the entire FCR and has been reported in numerous territories of the Caribbean. However, the propagation of the disease through FCR seemed to slow down when it reached its southwestern end, between the Marquesas and the Dry Tortugas (DRTO). Although the causative agent of the disease remains unknown, studies showed that its transmission was likely waterborne and that sediments could act as SCTLD vectors. The hydrodynamics should therefore be highly explanatory of its spread. Furthermore, Florida is a prime landfall target for hurricanes during the reproduction period of corals. In addition to causing wholesale destruction of the reefs, hurricanes might thus also impact larval dispersal through wind-wave-induced currents. Here, we use the high resolution coastal ocean model SLIM to capture transport processes in the FCR, and study the impact of SCTLD and hurricanes to coral reefs. First, a coupled epidemio-hydrodynamic model is developed to reproduce and understand the observed spread of SCTLD in Florida. A sediment transport model is then used to evaluate the impact of the PMDDP on the observed onset of the disease in 2014. Finally, a coupled wave-current model is implemented to study the impact of Hurricane Irma (2017) on transport processes over reefs in the Florida Keys. Assuming the dispersal of disease agents within neutrally buoyant material, our model successfully reproduced the observed spread of SCTLD through, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Semi-crystalline polymers exhibit performances that are highly dependent on their micro-structure as induced by the thermo-mechanical processes they are subjected to. In this thesis we developed a multiscale modeling and simulation framework able to predict the thermo-mechanical response of semi-crystalline polymers including crystallization and porosity evolution, with an application to additive manufacturing by selective laser sintering. Firstly, an enhanced phase field model was developed for the numerical simulation of crystallization in semi-crystalline polymers. The model is based on coupling the heat equation with the Allen-Cahn equation, which is derived from the Gibbs-Thomson solid-liquid interface equation. Starting from the nucleation of spherulites, existing phase field models can simulate their evolution in a surrounding liquid and separate the amorphous and crystalline phases. However, the predictions of the morphological characteristics of the spherulites remain qualitative only. Moreover, the predicted spherulite evolution as a function of crystallization temperature is not consistent with experimental results. In our work, existing phase field models were enhanced in order to obtain experimentally consistent results. We used spherulite growth, crystal morphology, and crystallinity degree in spherulite, as measures to determine the model accuracy. The model is numerically implemented using the finite difference method so that 2D and 3D simulation results are presented and compared to experimental data, illustrating the quantitative adequacy of the predictions with experimental evidence. Secondly, full-field micromechanical simulations were conducted on the micro-structures generated by the enhanced phase field model in order to predict the effective mechanical properties. Care is taken to obtain Representative Volume Elements (RVEs) by computing the number of subcells in each spherulite and the number of spherulite nucleations in each RVE. An FFT sol, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Machine learning algorithms are used to learn models capable of predicting on unseen data. In recent years, we have seen a rise in the use of black box (complex) models. They have become popular due to their performance and are used in almost all contexts. Unfortunately, it is not possible to understand the reasons behind their prediction. It is therefore risky to use these models in critical application cases. In contrast to these models, some interpretable models provide an explanation for their prediction. Decision trees (DTs) are an example of such interpretable models. A DT hierarchically divides a dataset into small groups using successive conditions on feature values. Then, it associates predictions with these groups. The conditions leading to a group correspond to the explanation of the associated prediction. Unfortunately, in practice, DTs are neglected in favor of black-box models because the well-known DTs learning algorithms are heuristic and sometimes produce non-accurate trees. Optimal decision trees (ODTs), on the other hand, offer better performance but are very difficult to learn. Existing approaches prior to this thesis spend many hours to find an ODT of depth 2. In this thesis, we have proposed a specialized algorithm to learn ODTs in a reasonable amount of time to promote their use in everyday applications. Our algorithm is based on dynamic programming (DP) and branch-and-bound techniques. In practice, it is very time-efficient but can lead to important memory consumption due to the caching operations performed in DP. We have then proposed a technique to bound the memory consumption of our algorithm. Finally, we proposed ways to use ODTs to learn optimal forests and solve several well-known machine learning tasks., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Context-oriented programming enables dynamic software evolution by supporting the creation of software systems that dynamically adapt their behaviour depending on the context of their surrounding environment. Upon sensing a new particular situation in the surrounding environment, which is reified as a context, the system activates this context and then continues by selecting and activating fine-grained features corresponding to that context. These features, representing functionalities specific to that context, are then installed in the system to refine its behaviour at runtime. Conceiving such systems is complex due to their high dynamicity and the combinatorial explosion of possible contexts and corresponding features that could be active. To address this complexity, we propose a feature-based context-oriented software development approach to design and implement context-oriented applications. This approach unifies context-oriented programming, feature modelling and dynamic software product lines into a single paradigm. In this novel paradigm we separate clearly and explicitly contexts and features that we model in terms of a context model, a feature model and the mapping between them. We also design an architecture, implement a programming framework, and develop a supporting development methodology and two visualisation tools to help designers and programmers in their modelling, development and debugging tasks. Furthermore we also develop a user interface library in our approach to create applications with user interfaces that are adaptive. To validate our feature-based context-oriented software development approach, we designed five case studies and implemented three of them. Then we discussed the design qualities to evaluate our implementation of the programming framework. We also assessed the usability of the programming framework from our own perspective based on the cognitive dimensions of notations framework. Finally we also conducted four user studies with, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Quasi-Newton methods are very popular in Optimization. They have a long, rich history, and perform extremely well for solving real-life problems. However, almost nothing is known about theoretical efficiency guarantees for these methods. The goal of this work is the advancement of the theory of quasi-Newton methods. This includes both obtaining new convergence estimates for the already existing algorithms and developing new methods with provable efficiency guarantees. In this thesis, we present our results in several directions. First, we provide a new theoretical analysis of local superlinear convergence of classical quasi-Newton methods and establish explicit and non-asymptotic bounds on their rate of convergence. Then, we develop and analyze new quasi-Newton methods which have some advanced features. Specifically, we propose a new family of greedy quasi-Newton methods for which, apart from local superlinear convergence, it is also possible to guarantee the convergence of Hessian approximations. Finally, we study one algorithm which is related to classical quasi-Newton methods, namely, the Ellipsoid Method, and develop a new variant of this method, which has a better dependency on the dimensionality of the problem than the standard one., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Proton therapy is a type of radiation therapy that uses a beam of protons to irradiate cancerous tissues. In principle, it offers a physical advantage over conventional radiotherapy due to the very localized dose deposition of protons in the body, which can be exploited to decrease the dose received by healthy tissues, leading to fewer complications. However, this unique characteristic comes at the cost of high vulnerability to uncertainties, requiring extremely precise machinery from beam production to treatment delivery. Moreover, accelerating protons requires a heavier infrastructure than conventional radiotherapy, which increases the cost of proton therapy over photon therapy. This thesis aims at improving the accessibility, cost-effectiveness, and treatment quality of proton therapy by using data-driven approaches wherever they can bring added value. In the first part of this work, with the aim of improving equipment maintenance, we develop predictive maintenance solutions based on machine learning to detect incoming failures and decrease the overall system downtime. In the second part of the thesis, we concentrate on reducing the time needed to install a new proton therapy system by developing an automatic procedure based on mathematical optimization to speed up the calibration of a proton therapy beamline. The third and final part of the thesis is devoted to improving the treatment quality of mobile tumors in proton therapy. To achieve this goal, we develop a method based on a library of treatment plans that uses real-time information about the patient's anatomy via the acquisition of images to guide the treatment delivery., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Nonnegative Matrix Factorization (NMF) is a popular data analysis tool for nonnegative data, able to extract meaningful features from a dataset, compress it and filter its noise. To do so, this method factorizes an input matrix Y as the product of two factors, the low-rank nonnegative matrices A and X. Recently, several interesting works have analyzed the possibility of imposing some additional structure to the NMF model, by requiring factors A and X to be based on nonnegative parametrizable functions, such as nonnegative splines or polynomials. When the input data is structured, for example for smooth continuous signals, this leads to factorizations that are less sensitive to noise and better able to find the characteristic elements of the dataset. This thesis is a continuation of that work. We first study NMF using polynomials and splines. We focus on the Hierarchical Alternating Least Squares (HALS) algorithm, which has the advantage of quickly finding a good solution, and converging to a stationary point under mild conditions. We generalize this algorithm and observe that it obtains good results and is competitive with existing approaches. We also seek to generalize the NMF model in a more formal way, so that any parameterizable function could be used in the factors. This leads to the definition of the H-NMF (NMF on Hilbert spaces), allowing to factorize a two-variable function (such as a matrix) as the sum of a small number of products of one-variable functions (the characteristic elements). This generalization is proved to have characteristics and theoretical guarantees similar to those of the standard NMF problem. Our generalization of HALS uses projections onto the functions under consideration (e.g. nonnegative polynomial or splines). As these projections can be costly and difficult to compute, we consider to approximate them. We are able to identify several cases where the algorithm keeps good convergence properties, and use these results to describe heuri, (FSA - Sciences de l'ingénieur) -- UCL, 2022

The exponential growth of technology opened the door for a new generation of locomotion assistive devices (e.g. prostheses and exoskeletons) to emerge. The latest generation is lighter weight due to the emergence of smart and composite materials, and advanced actuators and sensors. Controlling these devices to compensate for limb loss or walking disorder requires the generation of various gait patterns for different locomotion tasks like stairs ascending, descending, and level ground walking. However, this requires that these devices be able to detect the actions and intentions of their human user when traversing from one type of terrain to another. Such transition requires switching to the new task in a timely manner; otherwise, the user is more vulnerable to falls and injuries. Typical solutions to detect a new locomotion task are based either on sensors that measure the user’s kinematics or on a voluntary action of the user, such has tilting their chest or pushing a button. While the former suffers from a delay, the latter provides an unnatural way of behaving during locomotion. Recently, alternative methods have been proposed that uses wearable ranging sensors or camera. However, the implemented approaches for terrain identification with such sensors have been only validated in single indoor and/or outdoor path(s). Also, the system detection ability in a path being partially occluded by other walkers has not been addressed yet. Last but not least, the detection results are obtained based on a pre-specified region of interest in front of the user, without applying extra processing to predict the terrain information for several upcoming steps. In this thesis, we introduce a computer vision-based terrain detection algorithm that relies on a depth camera attached to the user’s chest. The camera, therefore, could capture the scene in front of the user in a point cloud data format, while the algorithm applies several steps on the captured data in order to achieve envi, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Dynamic Programming (DP) is a popular tool to solve combinatorial problems. This paradigm is ubiquitous and at the heart of many of the classic algorithms taught to every computer science student around the world. For instance, Dijkstra's shortest path algorithm is a case of DP. Still, when DP is used to solve NP-hard problems, it faces one major pitfall: the algorithm might not only require an exponential time but an exponential amount of memory as well -- which severely hampers its applicability. In 2016, a novel method combining DP with Decision Diagrams (DD) in a branch-and-bound framework was proposed. The strength of that approach stems from the ease of modeling leveraged from DP and the memory efficiency of the DD data structure. The latter is a graphical model providing an efficient encoding for exponentially sized sets of solutions to a given problem and allows the identification of an optimal solution in linear time. This thesis investigates the implementation and the trade-offs to make to benefit the most from this new technology. It also proposes algorithmic improvements to boost the performance of DD-based optimization solvers. It presents techniques to strengthen the bounds derived from each compiled DD and investigates ways to blend them with other resolution techniques such as Large Neighborhood Search (LNS) or approximate DP to swiftly find good solutions to a given problem., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Multi-agent systems are a powerful tool used to model a wide variety of problems involving interconnected independent entities collaborating to achieve some objective. The overwhelming majority of studies based on this model assume the composition of the systems remains constant, and yet at the same time aim for robustness and scalability. In this thesis, we question this assumption by considering open multi-agent systems, subject to arrivals and departures of agents which cannot be neglected. In the first part, we highlight and study general properties of such systems by analyzing the well-known consensus problem and the Gossip algorithm. On the one hand, we propose an appropriate way to represent open systems whose size changes with time, applied to this type of problems. On the other hand, we introduce tools and methods to characterize the performance of algorithms in such systems, whose convergence is prevented by nature. In the second part, we study the behavior of decentralized optimization algorithms solving the resource allocation problem subject to replacements of agents. Using two distinct types of metrics, we analyze the evolution of the error achieved by such algorithms in expectation. In particular, we characterize the noxious influence of replacements on their efficiency and behavior., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The MOS transistors of minimal gate length, universally favoured for the design of digital integrated circuits, are very sensitive to process variability and intrinsic noise. Time-domain characterization of the noise in individual tran- sistors is the starting point of this thesis. In 28 nm FD-SOI, we have observed random telegraph noise (RTN) whose relative current variation can reach 30% in weak inversion. Accurate characterization and compact modelling of such noise, as well as efficient and insightful simulation methodology that would rely on a rigorous and solid mathematical foundation, are therefore required toward assessment and prediction of circuit reliability. We have extensively studied the impact of the process variability and the intrinsic noise on the functionality of ultra-low power (ULP) SRAM bitcells in retention mode. We propose a non- Monte-Carlo methodology for fast and accurate calculation of the subthreshold SRAM hold static stability failure rate with a closed-form formula, which we illustrate on a 28 nm FD-SOI 6T SRAM design operating at ultra-low sup- ply voltage. We combine computationally intensive transient noise simulations and mathematical analyses to observe and predict transient bit flips due to the combined effects of the RTN and of the flicker noise and thermal noise sources. We conclude that, while process variability remains the major concern in 28 nm technology and beyond, experimentally measured RTN significantly reduce the noise margins and can severely endanger the dynamic stability of the latch., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Hydrogen and syngas production from natural gas is the fundamental process for ammonia production, methanol synthesis, and the petrochemical industry. However, the most widely deployed technology for hydrogen production technology, i.e., steam methane reforming (SMR), has several drawbacks including low energy efficiency and CO2 emission from its furnace that requires CO2 capture under the carbon-constrained scenario. Autothermal chemical looping reforming (aCLR) is proposed to address these two problems. The aCLR has been successfully tested from bench-scale semi-continuously operated fluidized beds to 140kWth continuously operated dual fluidized bed facilities. However, environmental concerns arising from the use of Ni-based oxygen carriers (OC). On the other hand, other investigated OCs, e.g., Cu- and Fe-based materials have lower reactivity towards methane reforming, which leads to low CH4 conversion. Modeling and simulation accounting for detailed reaction mechanisms and reactor hydrodynamics plays important role in designing, understanding, upscaling, and eventually optimizing the aCLR. Limited by the computational power, scale-bridging strategies are required to efficiently evaluate the reactor performance while understanding the relative importance of different phenomena, e.g., intrinsic reaction kinetics, gas-solid mass heat, mass transfer, and the residence time of solids in the bubbling fluidized bed (BFB). This thesis is aimed at investigating the feasibility of a more environmentally compatible aCLR. This process considers using the environmentally friendly Fe-based OC for transferring oxygen between the two reactors and PGM-based SMR catalyst for enhanced methane conversion, which is called catalyst-enhanced autothermal chemical looping reforming (CE-aCLR). CE-aCLR is studied by experimentally investigating the reaction kinetics of the potential Fe-based oxygen carrier and the PGM-based SMR catalysts, and performing modeling and simulation of the aCLR, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Birds have always been a source of inspiration for humans. From the record distances crossed by the bar-tailed godwit during its migrations, to swifts staying aloft continuously for up to ten months, they are capable of achievements that artificial devices are still far from reproducing. We still have a lot to learn from the flight of birds, and the unveiling of the mechanisms behind their efficiency could open the way to new levels of performances for our man-made air vehicles. While many works have relied on observations and experimental measurements conducted on birds and have led us to a certain knowledge of their flight mechanisms, the present work builds upon the idea that reproducing bird flight in simulation can shed light on what makes them so efficient. Indeed, plain mimicry of the natural flight of birds does little toward the better understanding of its underlying principles whereas the numerical reproduction of the broad range of phenomena involved allows us to test various hypotheses. In this thesis, we report on the development of a model for bird flapping flight. This model includes a representation of the wing skeleton and the main flight feathers, controllers aimed at the stabilization of flight and a faithful modeling of the wake behind a bird. The model is applied to the study of the dynamics of a bird in controlled flight, the analysis of its wake and of the interactions between two birds flying in formation. In the results, we present predictions regarding the distribution of aerodynamic forces and the contributions to the flight power. We also analyze the structure of a bird wake, verifying hypotheses concerning the positioning of a bird flying behind another, and we simulate the formation flight of two birds, exploring the control strategies that stabilize the follower, and how its position affects the reduction in its flight power., Les oiseaux ont toujours été une source d’inspiration pour l’humanité. De la distance record parcourue par la barge rousse durant ses migrations annuelles, aux hirondelles capables de rester en vol ininterrompu pendant dix mois, ils sont capables de performances que nous sommes encore loin de pouvoir reproduire artificiellement. Nous avons encore beaucoup à apprendre du vol des oiseaux, et découvrir les mécanismes à l’origine de leur efficacité peut mener à de nouveaux niveaux de performances pour nos appareils volants. Malgré de nombreux travaux basés sur l’observation et des mesures expérimentales du vol des oiseaux, qui nous ont amené une certaine connaissance des mécanismes de leur vol, ce travail se base sur l’idée que la reproduction de ce vol en simulation peut nous éclairer sur ce qui le rend si efficace. En effet, une simple reproduction du vol des oiseaux tel qu’observé dans la nature ne suffit pas pour comprendre ses principes fondamentaux, tandis que la reproduction numérique de la large gamme de phénomènes impliqués dans le vol des oiseaux nous permet de tester des hypothèses. Dans cette thèse, nous présentons le développement d’un modèle du vol de l’oiseau à ailes battantes. Ce modèle comprend une représentation du squelette de l’aile et de ses principales plumes, des contrôleurs pour la stabilisation du vol, et un modèle fidèle du sillage derrière un oiseau. Le modèle est utilisé pour l’étude de la dynamique d’un oiseau en vol contrôlé, l’analyse de son sillage et les interaction entre deux oiseaux volant en formation. Dans les résultats nous présentons des prédictions concernant la distribution des forces aérodynamiques pendant le vol, et les différentes contributions à la consommation d’énergie. Nous analysons aussi la structure du sillage et vérifions des hypothèses concernant le positionnement d’un oiseau volant derrière un autre. Finalement, nous simulons le vol en formation de deux oiseaux, explorant ainsi les stratégies de contrôle disponibles, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Every year, the sky comes to life with migratory birds that line up and draw impressive V-shapes. The energetic efficiency of this flying technique is explained by the existence of a wake in a region behind every individual and that each can perceive. Just like birds, airplanes also leave a wake consisting of two parallel vortices made visible in the sky by the condensation trails of the engines. Are airplanes also able to take advantage of the wake of an aircraft in front of them? This thesis is dedicated to bringing some answers to that question by focusing on the concept of extended formation flight, and by studying its impact on the air traffic management. The swirling movement of the air around the leader’s wake encountered by the following aircraft explains the benefits of the formation flight technique. However, this movement can lead to hazardous dynamic effects such as strong rolling moments. If properly controlled, those effects are also disseminated information that can be exploited by the follower in order to detect the position of the leader’s wake. In large eddy simulations, we aim to reproduce all those effects, from the aerodynamics of the aircraft to their six-degrees-of-freedom dynamics. Then, we exploit control and data assimilation tools designed to predict the characteristics of the leader’s wake without requiring direct visual information. We show that the tracking of the wake can be performed using controllers built either from model-based Ensemble Kalman Filters or from model-free Reinforcement Learning techniques. We highlight observability issues related to the prediction of the wake characteristics and we use an appropriate mathematical framework to quantify that observability in order to optimize the wake tracking controllers. Finally, thanks to the development of a methodology that combines both space-developing and time-developing simulations, we study the impact of formation flight on the long-term dynamics of the wake resulting from t, (FSA - Sciences de l'ingénieur) -- UCL, 2022

Electronic sensor systems, most often designed in CMOS technology, are key components of biomedical wearable devices, which are part of the emerging mobile health paradigm. These devices sense physiological signals from the patient and process the acquired data to provide clinical information. Besides the potential medical and cost benefits of wearable systems, the environmental sustainability of these devices is a critical factor given their expected widespread deployment. This thesis aims at optimizing mixed-signal sensor systems for low power consumption to limit the need for battery capacity and the associated environmental impacts. In addition, we pursue low-complexity yet programmable sensor systems to limit the environmental footprint of their production and the risk of obsolescence. In the first part, we demonstrate the circuit-level design of a low-power signal acquisition and processing system on a custom ultra-low-power microcontroller in the use case of electrocardiogram arrhythmia classification. System-level optimization of this cardiac monitoring system is performed in the second part, to further minimize the total power consumption while maintaining a high classification accuracy. Finally, we take a step back at the application level and show that, in addition to the direct environmental impacts of wearable devices related to their life cycle, more attention should be devoted to their indirect environmental impacts such as care optimization, induced medicalization, or structural rebound effects., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The energy transition reveals a number of challenges for the planning and operation of electric power systems, many of which are the subject of active research. Among the open issues that relate to market design, one has been the subject of particularly intense debate in Europe since the liberalization: the market-based allocation of transmission capacity. Currently, the European electricity market is organized around the principle of zonal pricing, which is characterized by the delimitation of regions on the power network, that are called zones, inside which the electricity price is the same in the wholesale market. This principle is opposed to nodal pricing, in which the price is differentiated between every location. In this thesis, we present detailed models and algorithms for analyzing the efficiency of zonal pricing. In the first part, we focus on the short-term efficiency and, in particular, on the impacts of the optimization of the grid topology through transmission line switching for improving the efficiency of zonal pricing. The second part of the dissertation investigates the efficiency of zonal pricing in the long run, when investment decisions in generating capacity are accounted for. First, we propose a model of the long-run equilibrium of zonal pricing with flow-based market coupling, which is the current methodology used in practice in Europe for allocating transmission capacity. We show that this methodology introduces new inefficiencies in the context of capacity expansion. Then, we investigate whether additional locational instruments, such as capacity and energy-based signals or market-based re-dispatch, can restore the efficiency of investments in zonal pricing. For each of the three main contributions, we present simulation results of our models on a large-scale instance that represents the Central Western European network and we discuss the implications of our results for the performance of zonal electricity pricing., (FSA - Sciences de l'ingénieur) -- UCL, 2022

Feature selection aims to find the most informative input variables for a prediction task. It has become compulsory for a wide variety of applications due to the appearance of high-dimensional data sets. Feature selection is said to be unstable when small changes in data result in large changes in the selected feature sets. The selected variables are often analyzed by domain experts to gain further insights and selection instability strongly limits their sound interpretation. This is a recurrent concern in bio-medical studies where high-dimensional and highly correlated data sets are common and often cause strong instability. This thesis tackles this instability issue. In particular, selection stability and predictive accuracy are optimized jointly in a bi-objective framework. Pareto-optimal trajectories are derived, from which domain experts can choose a trade-off based on their personal preferences for both objectives. The thesis also illustrates that current stability measures are unsatisfactory. This is first shown in the context of stability optimization. A novel and more robust stability measure is proposed and incorporates the importance of the selected features in predictive models. Stability estimation is notoriously difficult when the feature space contains strongly correlated features. The thesis also introduces a novel measure in that setting, where computing stability is shown to require solving a constrained optimization problem. Both measures improve decision-making in their respective context. By combining them with new interactive visualization software, this thesis provides domain experts with an enhanced toolbox for domain analysis., (FSA - Sciences de l'ingénieur) -- UCL, 2022

In order to achieve large-area photodetection over a few cm2 with higher sensitivity, higher opto-electronic transfer efficiency and lower cost, two p-type thin-film semiconductors, CIGS and CuO, are investigated in this thesis to realize high-performing large-area opto-electronic devices based on the investigations of novel fabrication processes, device designs, characterization and simulation results. CIGS photodiodes are optimized firstly by reducing the CIGS thickness below 1 um, introducing industry-compatible flexible steel substrates, alkali doping and multiple buffer layers (CdS and Cd-free In2S3). Dual-mode photodetection of CIGS photodiodes is fully characterized for the first time both in the photovoltaic and photoconductive modes. In addition, a corrected photoconductive gain mechanism is firstly investigated to accurately explain the operation mechanisms of CIGS photodiodes. As a more sustainable semiconductor material than CIGS, full fabrication and characterization of large-area CuO thin films and related devices with excellent performances require further investigation. Targeting the best-performing p-type CuO with a bandgap around 1.5 to 1.7 eV, a carrier concentration around 10e15 cm-3, an absorption coefficient over 10e5 cm-1, and a maximum carrier mobility, magnetron sputtering is introduced as a reliable technique in the large-area fabrication. To investigate the interfacial properties of CuO with metal and dielectric, the Au/CuO transmission line model (TLM) structures and a CuO metal-oxide-semiconductor (MOS) capacitor are studied. Finally, CuO TFTs with HfO2 gate dielectric are fabricated, characterized, and analyzed on soda lime glass., (FSA - Sciences de l'ingénieur) -- UCL, 2022

For more than 3000 years, windmills have been used to harvest energy from the wind. Over the past 30 years, electrical generators and advanced aerodynamics have turned them into the wind turbines we know. From hereon, how can the coming 30 years see them transition from outsider to key player of the energy system to help reach carbon neutrality? Increasing rotor diameters is a possible answer, clustering turbines into wind farms is another. Yet, with both options come new challenges, as the sensitivity of components to fatigue increases with the size of rotors and the wake phenomenon is responsible for power losses in wind farms. This thesis falls within the context of using control strategies to tackle these challenges. More specifically, this work relies on high fidelity simulations to investigate control approaches numerically. When it comes to reducing fatigue loads, individually controlling the pitch of each blade has proven to be efficient. This thesis introduces a novel controller architecture that relies on a neural network trained with reinforcement learning. This work demonstrates that a neural network can learn how to alleviate loads in simple wind conditions and that it is also capable of transferring that knowledge to realistic ones, such as turbulence and wakes. Regarding the question of wake mitigation, dynamically controlling wind turbines is gaining interest. While some strategies enhance the lateral displacement of the wake, others periodically modify its intensity. This work provides some insights into the mechanisms relating dynamic actuation of the blades to power gains in the wake. To do so, the effects of dynamic flow control on the wake destabilization and recovery processes are investigated. Attention is also paid to quantifying the impacts of such strategies on both power production and loads at the scale of a pair of turbines., (FSA - Sciences de l'ingénieur) -- UCL, 2022

The minimization of nitrogen oxides (NOx) emissions is among the major concerns of combustion-related processes, together with the adverse effects related to the emission of greenhouse gases. NOx emission control can become very challenging for those combustion processes which require high temperatures, and/or involve the exploitation of non-conventional fuel mixtures. This PhD thesis work aims to shed light on NOx formation during combustion of Coke Oven Gas (COG) and particularly to understand the role that aromatic chemical compounds could play during NOx formation. COG is an attractive energy source derived from the coal carbonization, due to its high H2 content and calorific value. It has been widely investigated as a fuel directly burnt in-situ to sustain the energy requirement of the integrated steel production processes, or as an important source of valuable compounds (i.e., H2, CH4, syngas, methanol, etc). The residual presence of aromatic compounds in COG, downstream the raw COG purification line, can potentially affect both its combustion efficiency and pollutant emissions. Therefore, a fundamental understanding of the interactions between NOx formation and aromatic species oxidation is required to both valorize the use of COG as energy carrier, and to highlight promising strategies for the minimization of NOx emissions during the combustion of COG in coke making plants. Benzene (C6H6) has been considered as the representative aromatic compound. The effect of its addition on NO formation chemistry, during combustion of COG surrogate fuel blends has been investigated, under different stoichiometric conditions. A hierarchical approach was used in this thesis work to achieve this objective. Indeed, NOx formation during combustion of a COG surrogate mixture (H2/CH4/CO) has been investigated at i) lab-scale, in low-pressure flat premixed laminar flames through a joint experimental and kinetic modeling study, and at ii) semi-industrial scale, in a 20-kW semi-in, (FSA - Sciences de l'ingénieur) -- UCL, 2022

With the exponential deployment of Internet-of-Things (IoT) connected devices, massive networks relying on first-generation low-power wide-area network (LPWAN) protocols begin to reach their limits. Although more scalable LPWAN protocols are expected to be commercialized, LPWAN radios are typically implemented in hardware and tied to a single protocol. Upgrading an IoT network to a newer protocol hence requires the replacement of functional devices, which implies a high environmental cost. We study in this thesis two approaches to prevent the disposal of existing IoT devices that could arise from technological obsolescence. The first part of this thesis tackles the challenge of designing LPWAN software-defined radios (SDRs) on ultra low-power microcontrollers. In particular, we target SDR implementations of the LoRa and Sigfox protocols. We initially design, thanks to an accurate analytical model of unsynchronized LoRa signals, a Nyquist-rate LoRa receiver chain with a low-complexity iterative synchronization procedure. We then conduct a hardware/software co-design of a microcontroller architecture for LPWAN SDRs on IoT devices. This architecture includes a general-purpose low-power processor for the protocol-specific baseband computations and a hardware digital front-end for the generic signal processing. The LoRa and Sigfox SDRs are evaluated with an ultra low-power prototype of the microcontroller architecture and attain a sub-mW power consumption for the baseband processing. The second part of this thesis studies the design of multi-user receivers for LoRa gateways. Since LoRa devices are not synchronized in time, collisions between uplink frames limit the scalability of LoRaWAN networks. As a first attempt, we derive from the maximum-likelihood criterion a two-user detector capable of decoding two colliding frames. We demonstrate the practicality of this approach thanks to an SDR implementation of the proposed detector, along with an interference-robust synchro, (FSA - Sciences de l'ingénieur) -- UCL, 2022