The increasing concerns regarding users‘ data privacy leads to the infeasibility of distributed Machine Learning applications, which are usually data-hungry. Federated Learning has emerged as a privacy-preserving distributed machine learning paradigm, in which the client dataset is kept locally, and only the local model parameters are transmitted to the central server. However, adoption of the Federated Learning paradigm leads to new edge computing challenges, since it assumes computationally intensive tasks can be executed locally by each device. The diverse hardware resources in a population of edge devices (e.g., smartphone models) can negatively impact the performance of Federated Learning, at both the global and local levels. This thesis contributes to this context with the implementation of a hardware-aware Federated Learning platform, which provides comprehensive support regarding the impacts of hardware heterogeneity on Federated Learning performance metrics by modeling the costs associated with training tasks on aspects of computation and communication.

Thesis (MScEng (Electrical and Electronic Engineering))--University of Stellenbosch, 2010. / ENGLISH ABSTRACT: The successful development of a three degree of freedom motion simulation platform, capable of simulating a vessel’s flight deck at sea, is presented. The motion simulation platform was developed to practically simulate and test an unmanned aerial vehicle’s capability of landing on a moving vessel, before practically being demonstrated on an actual vessel. All aspects of the motion simulation platform’s development are considered, from the conceptual design to its practical implementation. The mechanical design and construction of a pneumatic motion simulation platform, as well as the electronics and software to enable the operation of this motion simulation platform, are presented. Mathematical models of the pneumatic process and platform orientation are developed. A controller architecture capable of regulating the pneumatic process, resulted in the successful control of the motion simulation platform. Practical motion simulation results of one of the South African Navy Patrol Corvettes, demonstrate the motion simulation platform’s success. The successful development of the motion simulation platform can largely be attributed to extensive research, planning and evaluation of the different development phases. / AFRIKAANSE OPSOMMING: In hierdie studie word die suksesvolle ontwikkeling van ’n drie-grade-van-vryheid bewegingsimulasieplatform, wat in staat is daartoe om ’n skip se vliegdek ter see te simuleer, aangebied. Die bewegingsimulasieplatform is ontwikkel om ’n onbemande lugvaartuig se vermoë om op ’n bewegende skip te land, te simuleer en te toets, voor dit op ’n werklike skip gedemonstreer word. Alle aspekte van die ontwikkeling van die bewegingsimulasieplatform word in ag geneem – van die konsepontwerp tot die praktiese implementering daarvan. Die meganiese ontwerp en konstruksie van ’n pneumatiese bewegingsimulasieplatform word bespreek, sowel as die elektronika en programmatuur wat die werking van hierdie bewegingsimulasieplatform bemoontlik. Wiskundige modelle van die pneumatiese proses en platformoriëntering word ontwikkel. ’n Beheerderargitektuur wat in staat is daartoe om die pneumatiese proses te reguleer, lei tot die suksesvolle beheer van die bewegingsimulasieplatform. Praktiese resultate van die bewegingsimulering van een van die Suid-Afrikaanse Vloot se patrolliekorvette wys daarop dat die bewegingsimulasieplatform wel suksesvol is. Die geslaagde ontwikkeling van die bewegingsimulasieplatform kan grootliks toegeskryf word aan omvangryke navorsing, beplanning en evaluering van die onderskeie ontwikkelingsfases.

The goal of this Master's thesis was to solve resource allocation problems in wireless networks through the implementation of a lightweight simulation platform. The spectrum and power resources of wireless networks have to be efficiently used to accommodate the growing number of wireless terminals and the massive increase of data transferred by their applications. The major problem that needs to be tackled is interference, which significantly limits the performance of wireless systems. In this thesis, the resource allocation of interest was the joint problem of scheduling and power control with Quality of Service (QoS) constraints. The Signal-to-Interference-plus-Noise Ratio (SINR) was used to quantify QoS. This thesis studied the recently proposed mixed-integer linear programming (MILP) formulation of the problem. Due to the scheduling component, the problem is inherently combinatorial and NP-hard, therefore computationally expensive and difficult to solve in tractable time. A simulation platform was implemented in order to automate and facilitate the solving process.As a starting point, wireless channels and channel modeling issues were studied. Then, the platform was implemented to simulate random instances of multi-cellular wireless networks, with several mobile stations per cell, and generate the corresponding channels. Finally, the platform was extended to use the GNU Linear Programming Kit (GLPK) API in order to optimally solve the aforementioned formulated problem for various inputs of generated channels.Tests of the simulation platform were performed to check the consistency of the results. Indeed, the output results satisfied the initial expectations regarding the SINR constraints and the formulation. Moreover, they were produced in reasonable time. An analysis of the output results was presented.This thesis resulted in a configurable and lightweight simulation platform which is able to solve the MILP-formulated resource allocation problem. The simulation platform is basic and does not cover all the aspects of multi-cellular wireless networks and wireless channels. Due to its modularity, it can be extended in a future project.

Off-label use of traditional Oropharyngeal Airway (OPA) as a bite-block, and the subsequential procedure of force exertion of the device by physician has caused many cases of patient’s teeth damage and monetary loss, as the patient’s incisors were damaged while clenching on the OPA during an adverse scenario called “Emergency Clenching”. To remedy this harmful situation, Bardo OPA was developed by Dr. Theodore Burdumy. The Bardo airway has unique design to transfer the clenching force from incisor to the molar. However, the Bardo OPA is one-sized, and cannot fit most of the patients like the commonly-used OPAs, such as the Berman and Gudel OPA, which have a spectrum of sizes to ensure fit. In this project, a Computer Assisted Design (CAD) simulation platform was developed to simulate the scenario where OPA is placed in a patient’s oral cavity. CAD – related technique and tools, such as 3D scanner (ScanStudio HD), RapidWorks, SolidWorks and Mimics were utilized to create the models used to construct the platform. The purpose of this platform is to generate data to support the development of additional sizes and other modification to improve the current design of the Bardo OPA. MRI sets of nine (9) patients were obtained and converted into STL mesh models. Berman and Guedel OPA were used as the standard for comparison against the Bardo OPA. It was found that the Bardo OPA was able to fit into all sample patients’ models, while these models were fitted with Berman and Guedel OPA of 70-90mm (Small to medium adult) sizes. It can only be concluded that the Bardo is compatible with these OPA sizes and there was not enough evidence to show the need for additional sizes. Nevertheless, some functional features of the Bardo OPA were found potentially harmful to the patients or ineffective. Three approaches were suggested to improve the design of the Bardo to achieve better safety and efficacy.

The vestibuloocular reflex can be manually elicited by tilting or rotating the head. Manual techniques serve their purpose well and is the golden standard in the clinical work, but they lack control of velocity and movement pattern. However, motion simulation platforms enable automatic control of both velocity and movement pattern. One motion simulation platform, named BIRGIT, has been built at the Department of Clinical Neurosciences at Karolinska Institutet but has not yet been in service and require a performance evaluation. The objectives with this thesis is to evaluate the accuracy and precision of BIRGIT and evaluate how bodyweight and movement direction impact the performance. The thesis also evaluate whether it is possible to stabilize the head of the patient during the acceleration phase. Repeated measurements of acceleration with different loads, desired ac- celeration, direction and motion type (rotational and translational), are per- formed. Dummies are used to simulate bodyweight in the performance study and real persons are used in the head stabilization study. Analysis of variance (ANOVA) is the main statistical tool. The results suggest that the platform does not perform equally at dif- ferent load or directions and that there is a bias between desired and true acceleration. The main problems are an inclination of the rails, upon which the chair is mounted, that causes differences between directions and an un- desirable performance characteristic for rotational motions. The stabilization study suggest that the head can be stabilized. / Den vestibulo-okul ̈ara reflexen kan framkallas genom att manuellt rotera eller rycka p ̊a huvudet. Att manuellt framkalla reflexen fungerar i de flesta sam- manhang va ̈l och a ̈r standard i m ̊anga underso ̈kningar. Dock g ̊ar det inte att till fullo kontrollera vare sig hastighet eller ro ̈relsebana perfekt. Emellertid g ̊ar detta att kontrollera genom att anva ̈nda sig av en s ̊a kallad ro ̈relsesimu- leringsplatform. En s ̊adan platform, d ̈opt till BIRGIT, har byggts vid Institutionen f ̈or kliniska neurovetenskaper vid Karolinska Institutet. Denna har ej bo ̈rjat anva ̈ndas ̈annu d ̊a dess prestanda fo ̈rst beho ̈ver utv ̈arderas. Syftet med denna uppsats ̈ar att utva ̈rdera precision och noggrannhet hos BIRGIT. Dessutom, att utv ̈ardera hur kroppsvikt och ro ̈relseriktning inverkar p ̊a prestandan. I arbetet ing ̊ar ̈aven att testa om det a ̈r m ̈ojligt att stabilisera huvudet under accelerationsfasen. Repeterade ma ̈tningar av sann acceleration med olika last, riktning, bo ̈rac- celeration och r ̈orelsetyp (rotation eller sidledes) genomfo ̈rdes. Testdockor anva ̈ndes fo ̈r att simulera lasten i prestandatesterna och riktiga testpersoner anva ̈ndes i huvudstabiliseringsdelen. Variansanalys (ANOVA) var det hu- vudsakliga statistiska verktyget. Resultatet antyder att last och ro ̈relseriktning inverkar p ̊a acceleratio- nen och prestandan och att den sanna accelerationen alltigenom a ̈r la ̈gre a ̈n bo ̈raccelerationen. Det finns tv ̊a sto ̈rre problem hos plattformen, det fo ̈rsta a ̈r att uppha ̈ngningen till stolen lutar och detta resulterar i en skillnad mellan riktningarna (det g ̊ar fortare nedf ̈ors). Det andra problemet a ̈r ett cykliskt uppfo ̈rande na ̈r flera rotationsr ̈orelser skall fo ̈lja p ̊a varandra. Stabiliseringsstudien visade att det g ̊ar att stabilisera huvudet.

Food, water and energy are three essential resources for human well-being, poverty reduction and sustainable development. These resources are very much linked to one another, meaning that the actions in any one particular area often can have effects in one or both of the other areas. At the same time, an economy's shift towards climate neutrality requires a massive expansion of energy production from renewable sources. Among these ground-mounted photovoltaic (PV) and biomass will be expanded massively to meet the clean energy generation goal, simultaneously influence regional water and food availability and supply security. It is crucial to understand Food-Water-Energy Nexus (FWE) nexus during the energy transition. However, current studies have limitation both methodically (qualitative assessments) and spatially (aggregated data on a national level is more available). Firstly, a consistent share input data set in geographical format was created with the resolution of building/field. An energy simulation platform (SimStadt) was then extended with new workflows on biomass potential, ground-mounted PV potential, food demand/potential, and urban water demand. Combining with existing workflows on urban building heating/electricity demand and roof PV potential, the dissertation created a complete simulation environmental covering most-relating FWE topics in energy transition with consistent input and output structures at a fine resolution. Secondly, the most representative inter-linkage between ground-mounted PV and biomass on hinterland is investigated in details with the new tools. The output data of each field from ground-mounted PV and biomass workflows are linked and ranked according to the scenarios emphasizing PV yield, feasibility, profit, or biomass. The assessment and scenarios are applied at three representative German counties with distinguished land-use structures and geometries as case studies. Results show that current policies does not guarantee the technically efficient allocation of fields. The optimal technical strategy is to follow the individual market profit drive, which is very likely, at the same time for the social good, to achieve high PV yields with limited biomass losses and more significant crop water-saving effects. The local food, water, and energy demands are also included as a metric for resource allocation on the potential side. Besides focusing on the biomass-PV tradeoff simulation and analysis, pioneer works have also been done to test the transferability of the method in cases outside Germany, and the complement of urban solid waste to agricultural biomass is explored to achieve energy autarky.

Dans le contexte automobile actuel de réduction des émissions de CO2, une réponse semble être apportée par le véhicule électrique. De nombreuses questions se posent alors, notamment concernant l’autonomie, d’autant plus que le nombre de consommateurs électriques dans les véhicules est en constante augmentation. Il est nécessaire de concevoir un groupe-motopropulseur électrique alliant autonomie et performances en considérant les contraintes de coût auxquelles est confronté tout constructeur.Concernant les outils de conception, très nombreux sont les constructeurs qui s’orientent vers d’autres solutions que la réalisation de prototypes en faisant appel à la simulation numérique dès le début du cycle en V pour optimiser leur prédimensionnement et assurer un gain non négligeable sur les délais et les coûts. Les travaux de thèse s’articulent ainsi autour de deux axes. Une première étape consiste à modéliser le réseau électrique automobile (chaîne de traction, système de confort thermique et réseau 14V) afin de simuler son fonctionnement en régime dynamique. L’aspect dynamique est important : des cartographies de pertes ne peuvent suffire si nous nous intéressons aux performances du véhicule en termes d’accélérations. D’autre part, l’autonomie est impactée de façon non négligeable par cet aspect dynamique.Dans une seconde étape, nous procéderons à l’optimisation du système de puissance. Les critères qui nous intéressent, autonomie et performances, sont antagonistes, ce qui donne lieu à la recherche de « meilleurs » compromis. Nous distinguons l’optimisation des lois de pilotage des organes de l’optimisation de l’architecture, les deux étant menées séquentiellement. / In the current automobile trend of global CO2 emissions reduction, the electric vehicle seems to be one of the most effective solutions. It is a new technology and, consequently, many questions appear, in particular concerning the range impacted by the constant increase of the number of electric loads embedded in vehicles. It is necessary to design an electric powertain allying sufficient range and good performances and respecting cost constraints, which is a determining criterion for every car manufacturer. Concerning conception tools, many car manufacturers turn to other solutions that only physical prototypes realization by using numerical simulation from the beginning of the V-cycle in order to optimize powertrain presizing and ensure a significant costs and delays reduction.The thesis works are based on two axes.The first part consists in modeling the power system of electric vehicle (powertrain, air-conditioning, heating system and 14V network) in order to simulate the global system behavior in transient state. The dynamic aspect is important: for example, losses maps cannot be used if we are interested in vehicle performances in terms of accelerations. Moreover, as that will be shown, the vehicle range is impacted in a significant way by the dynamic phenomena. The models are validated by comparison with tests results.In a second part, we proceed to electric power system optimization. The criteria that interest us namely range and performances are conflicting, what gives rise in search of "best" compromises. It is necessary to distinguish control laws optimization from architecture optimization, both being led sequentially.

Companies realize to stay competitive they have to introduce quality improvement programs. Many companies are challenged today with the sustainability of these quality improvement programs. Generally the understanding of the dynamic behaviour of quality improvement programs is poorly understood with soft issues as factors of the system. System dynamics may solve this problem. This research was focussed on operational management of operations. The organising framework of this research has been qualitative research where the research design was a polar type research design where the case studies focussed on initiatives that were dramatic success or failures, with the expectation that their comparison would help identify those processes that prevent competence enhancing change. A dynamic hypothesis has been constructed from archival data, semi-structured interviews and direct observations, gathered during these case studies. A system dynamics model for quality improvement programs in an automotive environment has been tested and expanded to be applicable for a heavy engineering manufacturing environment. The structure of the system dynamics model has been expanded to include a sustainability feedback loop which also included a management support model. The management support model included soft factors such as management support, management pressure and managerial effectiveness. The complete quality improvement program system dynamics simulation model with sustainability has been tested and validated against a real system data, for a heavy engineering manufacturing environment, gathered during the case studies. The model parameters were determined from a calibration algorithm, by using the Vensim® simulation platform, that fitted the real system behaviour the best. A sensitivity analysis has been done on the model parameters determining the information cues for the management decision policies. From the system dynamics model of the complete quality improvement program, including the sustainability feedback loop, proposed management decision policies have been studied that could lead to sustainable quality improvement programs for a heavy engineering manufacturing environment. From these simulation studies several management policies have been proposed. / Thesis (PhD)--University of Pretoria, 2013. / gm2013 / Graduate School of Technology Management / Unrestricted

The move from Standard Definition (SD) to High Definition (HD) represents a six times increases in data, which needs to be processed. With expanding resolutions and evolving compression, there is a need for high performance with flexible architectures to allow for quick upgrade ability. The technology advances in image display resolutions, advanced compression techniques, and video intelligence. Software implementation of these systems can attain accuracy with tradeoffs among processing performance (to achieve specified frame rates, working on large image data sets), power and cost constraints. There is a need for new architectures to be in pace with the fast innovations in video and imaging. It contains dedicated hardware implementation of the pixel and frame rate processes on Field Programmable Gate Array (FPGA) to achieve the real-time performance. The following outlines the contributions of the dissertation. (1) We develop a target detection system by applying a novel running average mean threshold (RAMT) approach to globalize the threshold required for background subtraction. This approach adapts the threshold automatically to different environments (indoor and outdoor) and different targets (humans and vehicles). For low power consumption and better performance, we design the complete system on FPGA. (2) We introduce a safe distance factor and develop an algorithm for occlusion occurrence detection during target tracking. A novel mean-threshold is calculated by motion-position analysis. (3) A new strategy for gesture recognition is developed using Combinational Neural Networks (CNN) based on a tree structure. Analysis of the method is done on American Sign Language (ASL) gestures. We introduce novel point of interests approach to reduce the feature vector size and gradient threshold approach for accurate classification. (4) We design a gesture recognition system using a hardware/ software co-simulation neural network for high speed and low memory storage requirements provided by the FPGA. We develop an innovative maximum distant algorithm which uses only 0.39% of the image as the feature vector to train and test the system design. Database set gestures involved in different applications may vary. Therefore, it is highly essential to keep the feature vector as low as possible while maintaining the same accuracy and performance

The vestibuloocular reflex can be manually elicited by tilting or rotating the head. Manual techniques serve their purpose well and is the golden standard in the clinical work, but they lack control of velocity and movement pattern. However, motion simulation platforms enable automatic control of both velocity and movement pattern. One motion simulation platform, named BIRGIT, has been built at the Department of Clinical Neurosciences at Karolinska Institutet but has not yet been in service and require a performance evaluation. The objectives with this thesis is to evaluate the accuracy and precision of BIRGIT and evaluate how bodyweight and movement direction impact the performance. The thesis also evaluate whether it is possible to stabilize the head of the patient during the acceleration phase. Repeated measurements of acceleration with different loads, desired ac- celeration, direction and motion type (rotational and translational), are per- formed. Dummies are used to simulate bodyweight in the performance study and real persons are used in the head stabilization study. Analysis of variance (ANOVA) is the main statistical tool. The results suggest that the platform does not perform equally at dif- ferent load or directions and that there is a bias between desired and true acceleration. The main problems are an inclination of the rails, upon which the chair is mounted, that causes differences between directions and an un- desirable performance characteristic for rotational motions. The stabilization study suggest that the head can be stabilized. / Den vestibulo-okul ̈ara reflexen kan framkallas genom att manuellt rotera eller rycka p ̊a huvudet. Att manuellt framkalla reflexen fungerar i de flesta sam- manhang va ̈l och a ̈r standard i m ̊anga underso ̈kningar. Dock g ̊ar det inte att till fullo kontrollera vare sig hastighet eller ro ̈relsebana perfekt. Emellertid g ̊ar detta att kontrollera genom att anva ̈nda sig av en s ̊a kallad ro ̈relsesimu- leringsplatform. En s ̊adan platform, d ̈opt till BIRGIT, har byggts vid Institutionen f ̈or kliniska neurovetenskaper vid Karolinska Institutet. Denna har ej bo ̈rjat anva ̈ndas ̈annu d ̊a dess prestanda fo ̈rst beho ̈ver utv ̈arderas. Syftet med denna uppsats ̈ar att utva ̈rdera precision och noggrannhet hos BIRGIT. Dessutom, att utv ̈ardera hur kroppsvikt och ro ̈relseriktning inverkar p ̊a prestandan. I arbetet ing ̊ar ̈aven att testa om det a ̈r m ̈ojligt att stabilisera huvudet under accelerationsfasen. Repeterade ma ̈tningar av sann acceleration med olika last, riktning, bo ̈rac- celeration och r ̈orelsetyp (rotation eller sidledes) genomfo ̈rdes. Testdockor anva ̈ndes fo ̈r att simulera lasten i prestandatesterna och riktiga testpersoner anva ̈ndes i huvudstabiliseringsdelen. Variansanalys (ANOVA) var det hu- vudsakliga statistiska verktyget. Resultatet antyder att last och ro ̈relseriktning inverkar p ̊a acceleratio- nen och prestandan och att den sanna accelerationen alltigenom a ̈r la ̈gre a ̈n bo ̈raccelerationen. Det finns tv ̊a sto ̈rre problem hos plattformen, det fo ̈rsta a ̈r att uppha ̈ngningen till stolen lutar och detta resulterar i en skillnad mellan riktningarna (det g ̊ar fortare nedf ̈ors). Det andra problemet a ̈r ett cykliskt uppfo ̈rande na ̈r flera rotationsr ̈orelser skall fo ̈lja p ̊a varandra. Stabiliseringsstudien visade att det g ̊ar att stabilisera huvudet.