With the growing demand of compact radio transmitters, so do the need for silent and efficient thermal management systems. Given the restricted form-factor, ensuring robustness of a unit with a detachable fan tray imposes a challenge. In this thesis the authors were tasked with creating a new design solution where the robustness, thermal and noise performance of the cooling system is revised. The project was initiated with an investigation of the baseline product, the AIR5322, to identify areas of in need of improvement. The initial goal was to design an integrated fan tray that would not require in-field maintenance. A literature study was performed to evaluate common integrated cooling solutions in compact electronic devices, which was then followed by concept generation study. These were then ranked based on Pugh’s selection method. An integrated solution was deemed not feasible due to the average lifespan of common fans used in this type of product. A CAD model of said design was created, for which a prototype was manufactured by Ericsson. The final design consisted of a set of 80mm Noctua (NF-A8) fans in a new fan tray, retrofitted onto a modified stock AIR5322 heatsink. Experimental tests were conducted where heat dissipation, noise generation and mechanical robustness was tested. The results concluded that the prototype design managed to reduce the temperature of the entire unit, up to 17°C cooler and lower the sound levels by up to 18 dB. The prototype, despite halving the number of fans, retained the single fan failure redundancy, as running just one fan allowed for performance that was equal to or better than the reference solution at full capacity. The prototype was deemed robust enough to withstand the drop test, which was performed at −18°C. Additional work would include to incorporate fans with an appropriate IP rating and to fully complete the testing using a climate chamber and performing impact testing. / Med den växande efterfrågan på kompakta radiosändare ökar också behovet av tysta och effektiva kylsystem. Givet den begränsade formfaktorn innebär det en utmaning att säkerställa robustheten hos en enhet med en löstagbar fläktlåda. I denna avhandling fick författarna i uppdrag att skapa en ny designlösning där kylsystemets robusthet, termisk prestada och ljudprestanda revideras. Projektet inleddes med en undersökning av basprodukten, AIR5322, för att identifiera eventuella förbättringsområden. Det ursprungliga målet var att designa en integrerad fläktlåda som inte skulle kräva fältetunderhåll. En litteraturstudie genomfördes för att utvärdera vanliga integrerade kyllösningar i kompakta elektroniska enheter, som sedan följdes av konceptgenerering. Dessa utvärderades sedan och rangordnades utifrån Pughs urvalsmetod. En integrerad lösning ansågs inte möjlig på grund av den genomsnittliga livslängden för typiska fläktar som används i liknande produkter. En CAD-modell av denna design skapades, för vilken en prototyp tillverkades av Ericsson. Den slutliga designen bestod av en uppsättning 80 mm Noctua (NF-A8) fläktar i en ny fläktlåda som monterades i en modifierad kylfläns för AIR5322. Experimentella tester där värmeavledning, bullergenerering och mekanisk robusthet testades. Med resultaten kunde följande slutsats dras: prototypdesignen lyckades sänka temperaturen på hela enheten med upp till 17°C och sänka ljudnivåerna med upp till 18 dB. Prototypen visade sig ha en fläkt vara termiskt redundant, eftersom en fläkt igång resulterade i en prestanda som var lika med eller bättre än referenslösningen. Prototypen ansågs vara robust nog att motstå falltestet, som utfördes med prototypen nedkyld til −18°C. Ytterligare arbete som kan vara till nytta för detta projekt skulle vara att införliva fläktar med en lämplig IP-klassning och att fullständigt testa prototypen med klimattester och slagtester.

Power device robustness refers to the device capability of withstanding abnormal events in power electronics applications, which is one of the key device capabilities that are desired in numerous applications. While the current robustness test methods and qualification standards are developed across the 70 years of Silicon (Si) device history, their applicability to the recent wide bandgap (WBG) power devices is questionable. While the market of WBG power devices has exceeded $1 billion and is fast growing, there are many knowledge gaps regarding their robustness, including the failure or degradation physics, testing methods, and lifetime extraction. This dissertation work studies the robustness of Gallium Nitride (GaN) power device. The structures of many GaN power devices are fundamentally different from Si or Silicon Carbide (SiC) power devices, leading to numerous open questions on GaN power device robustness. Based on the device structure, this dissertation is divided into two parts: The first half discusses the robustness of lateral GaN high electron mobility transistor (HEMT), which recently sees rapid adoption among wide range of applications such as the power adapter and chargers, data center, and photovoltaic panels. The absence of p-n junction between the source and drain of GaN HEMT results in the lack of avalanche mechanism. This raises a concern on the device capability of withstanding surge-energy or overvoltage stress, which hinders the penetration of GaN HEMTs in broader applications. To address this concern, the study begins with conducting the single-event unclamped inductive switching (UIS) test on two mainstream commercial p-gate GaN HEMTs with the Ohmic- and Schottky-type gate contacts, where the GaN HEMT is found to withstand surge energy through a resonant energy transfer between the device capacitance and the loop inductance. The failure mechanism is identified to be a pure electrical breakdown determined by device transient breakdown voltage (BV). The BV of GaN HEMT is further found to be "dynamic" from the switching tests with various pulse widths and frequencies, which is further explained by the time-dependent buffer trapping. This dynamic BV (BVDYN) phenomenon indicates that the static or single-pulse test may not reveal the true BV of GaN HEMT in high frequency switching applications. To address this gap, a novel testbed based on a zero-voltage-switching converter with an active clamping circuit is developed to enable the stable switching with kilovolt overvoltage and megahertz frequency. The overvoltage failure boundaries and failure mechanisms of four commercial p-gate GaN HEMTs from multiple vendors are explored. In addition to the frequency-dependent BVDYN, two new failure mechanisms are observed in some devices, which are attributable to the serious carrier trapping in GaN HEMTs under the high-frequency overvoltage switching. At last, based on the findings in the high frequency overvoltage test (HFOT), a physics-based lifetime model for commercial GaN HEMTs utilizing the device on resistance (RON) shift is established and validated by experimental results. Overall, the switching-based test methodology and experimental results provide critical references for the overvoltage protection and qualification of GaN power HEMTs. The second half of the dissertation discusses the robustness of the vertical GaN fin-channel junction field effect transistor (Fin-JFET), a promising pre-commercialized GaN power device with the p-n junction embedded between the gate and drain which enables the avalanche breakdown. The robustness study on GaN JFET follows similar test approaches as Si metal-oxide-semiconductor field-effect transistor (MOSFET) with two key interests: the avalanche and short circuit capabilities. The avalanche breakdown is first explored via the single-event and repetitive UIS tests and under various gate drivers, from which an interesting "avalanche-through-fin-channel" mechanism is discovered. By leveraging this avalanche path, the electro-thermal stress migrates from the main blocking p-n junction to the n-GaN fin channel, resulting in a very favorable failure-to-open-circuit signature. The single-pulse critical avalanche energy density (EAVA) of vertical GaN Fin-JFET is measured to be as high as 10 J/cm2, which is much higher than the Si MOSFET and comparable to the SiC MOSFET. The short circuit capability is explored utilizing the hard-switching fault on the 650-V rated GaN Fin-JFET, with a gate driving circuit identical to the switching application to best mimic device operation in converters. The short circuit withstanding time is measured to be 30.5 µs at an input voltage of 400 V, 17.0 µs at 600 V, and 11.6 µs at 800 V, all among the longest reported for 600-700 V normally-off transistors. In addition, the failure-to-open-circuit signature is also shown in the single-event and repetitive short circuit tests; all devices retain the avalanche breakdown after failure, which is highly desirable for system applications. These results suggest that, while GaN HEMT is already available in market, vertical GaN Fin-JFET shows superior avalanche and short-circuit robustness and thereby can unlock great potential of GaN devices for applications like automotive powertrains, motor drives, and grids. / Doctor of Philosophy / In recent years, many power electronics applications such as data centers and electric vehicles have witnessed a rapid increase in the adoption of wide bandgap (WBG) power devices. The Gallium Nitride (GaN) device is one of the most attractive candidates in WBG devices, owing to its good tradeoff between breakdown voltage and on resistance, as well as the small gate charge that enables high frequency switching. For power devices, their robustness against overvoltage and overcurrent stresses is as important as their performance under normal operations. However, the new material, new device structure, and new device physics in GaN power devices brought up many open knowledge gaps in their robustness study, particularly under the dynamic operation in switching circuits. This dissertation presents the work in exploring the robustness of GaN power devices. Based on the device structure, the discussion is divided in two parts: The first half of the dissertation focuses on the overvoltage robustness of the lateral GaN High Electron Mobility Transistor (HEMT), the commercially available device covering 30 to 900 V voltage classes. A key feature of this device is the lack of p-n junction between source and drain, leading to an absence of avalanche capability. The study is conducted on mainstream, commercial p-gate GaN HEMTs, with a combination of circuit testing, microscale failure analysis, and physics-based device simulation. The main contribution is on three aspects: identifying the single-event and high-frequency repetitive overvoltage boundaries of GaN HEMT, unveiling the failure and degradation mechanisms under transient overvoltage conditions, and providing guidelines to GaN HEMT device users with proper robustness test methodology for device qualification and screening. The second half of the dissertation focuses on the robustness of vertical GaN fin-channel junction field effect transistor (Fin-JFET), a promising pre-commercial GaN power device with the p-n junction implemented between the source and drain. The robustness tests follow the classic approaches deployed for Silicon power devices, where both the avalanche and short circuit capabilities are investigated. From the single-event and repetitive test results, the GaN JFET shows excellent avalanche robustness with a desirable failure-to-open-circuit behavior, as well as a critical avalanche energy (EAVA) of 10 J/cm2 that is higher than the Silicon metal-oxide-semiconductor field-effect transistor (MOSFET) and comparable to the Silicon Carbide MOSFET. For a 650-V rated GaN Fin-JFET, a record high 30.5 μs short circuit time is demonstrated under the hard-switching fault condition at 400 V input voltage. Overall, the results show great potential of GaN power devices for the power electronics applications that involve more stressful operation conditions for devices.

Heat flux and heat conductance at the metal mold interface plays a key role in controlling the final metal casting strength. It is difficult to obtain these parameters through direct measurement because of the required placement of sensors, however they can be obtained through inverse heat conduction calculations. Existing inverse heat conduction methods are analyzed and classified into three categories, i.e., direct inverse methods, observer-based methods and optimization methods. The solution of the direct inverse methods is based on the linear relationship between heat flux and temperature (either in the time domain or in the frequency domain) and is calculated in batch mode. The observer-based method consists on the application of observer theory to the inverse heat conduction problem. The prominent characteristic in this category is online estimation, but the methods in this category show weak robustness. Transforming estimation problems into optimization problems forms the methods in the third category. The methods in third category show very good robustness property and can be easily extended to multidimensional and nonlinear problems. The unknown parameters in some inverse heat conduction methods can be obtained by a proposed calibration procedure. A two-index property evaluation (accuracy and robustness) is also proposed to evaluate inverse heat conduction methods and thus determine which method is suitable for a given situation. The thermocouple dynamics effect on inverse calculation is also analyzed. If the thermocouple dynamics is omitted in the inverse calculation, the time constant of thermocouple should be as small as possible. Finally, a simple model is provided simulating the temperature measurement using a thermocouple. FEA (Finite Element Analysis) is employed to simulate temperature measurement.

Artificial intelligence (AI) and machine learning (ML) present some application opportunities and challenges that can be framed as learning problems. The performance of machine learning models depends on algorithms and the data. Moreover, learning algorithms create a model of reality through learning and testing with data processes, and their performance shows an agreement degree of their assumed model with reality. ML algorithms have been successfully used in numerous classification problems. With the developing popularity of using ML models for many purposes in different domains, the validation of such predictive models is currently required more formally. Traditionally, there are many studies related to model evaluation, robustness, reliability, and the quality of the data and the data-driven models. However, those studies do not consider the concept of the applicability domain (AD) yet. The issue is that the AD is not often well defined, or it is not defined at all in many fields. This work investigates the robustness of ML classification models from the applicability domain perspective. A standard definition of applicability domain regards the spaces in which the model provides results with specific reliability. The main aim of this study is to investigate the connection between the applicability domain approach and the classification model performance. We are examining the usefulness of assessing the AD for the classification model, i.e. reliability, reuse, robustness of classifiers. The work is implemented using three approaches, and these approaches are conducted in three various attempts: firstly, assessing the applicability domain for the classification model; secondly, investigating the robustness of the classification model based on the applicability domain approach; thirdly, selecting an optimal model using Pareto optimality. The experiments in this work are illustrated by considering different machine learning algorithms for binary and multi-class classifications for healthcare datasets from public benchmark data repositories. In the first approach, the decision trees algorithm (DT) is used for the classification of data in the classification stage. The feature selection method is applied to choose features for classification. The obtained classifiers are used in the third approach for selection of models using Pareto optimality. The second approach is implemented using three steps; namely, building classification model; generating synthetic data; and evaluating the obtained results. The results obtained from the study provide an understanding of how the proposed approach can help to define the model’s robustness and the applicability domain, for providing reliable outputs. These approaches open opportunities for classification data and model management. The proposed algorithms are implemented through a set of experiments on classification accuracy of instances, which fall in the domain of the model. For the first approach, by considering all the features, the highest accuracy obtained is 0.98, with thresholds average of 0.34 for Breast cancer dataset. After applying recursive feature elimination (RFE) method, the accuracy is 0.96% with 0.27 thresholds average. For the robustness of the classification model based on the applicability domain approach, the minimum accuracy is 0.62% for Indian Liver Patient data at r=0.10, and the maximum accuracy is 0.99% for Thyroid dataset at r=0.10. For the selection of an optimal model using Pareto optimality, the optimally selected classifier gives the accuracy of 0.94% with 0.35 thresholds average. This research investigates critical aspects of the applicability domain as related to the robustness of classification ML algorithms. However, the performance of machine learning techniques depends on the degree of reliable predictions of the model. In the literature, the robustness of the ML model can be defined as the ability of the model to provide the testing error close to the training error. Moreover, the properties can describe the stability of the model performance when being tested on the new datasets. Concluding, this thesis introduced the concept of applicability domain for classifiers and tested the use of this concept with some case studies on health-related public benchmark datasets. / Ministry of Higher Education in Libya

No description available.

No description available.

The timetable is an essential part for the operations of railway traffic, and its quality is considered to have large impact on capacity utilization and reliability of the transport mode. The process of generating a timetable is most often a manual task with limited computer aid, and is known to be a complex planning problem due to inter-train dependencies. These inter-train dependencies makes it hard to manually generate feasible timetables, and also makes it hard to improve a given timetable as new conflicts and surprising effects easily can occur. As the demand for railway traffic is expected to continue grow, higher frequencies and more saturated timetables are required. However, in many European countries there is also an on-going public debate on the punctuality of the railway, which may worsen by increased capacity utilization. It is therefore also a need to increase the robustness of the services. This calls for increased precision of both the planning and the operation, which can be achieved with a higher degree of automation. The research in this thesis is aimed at improving the robustness of railway timetables by combining micro-simulation with mathematical optimization, two methods that today are used frequently by practitioners and researchers but rarely in combination. In this research a sequential approach based on simulating a given timetable and re-optimizing it to reduce the weighted sum of scheduled travel time and predicted average delay is proposed. The approach has generated promising results in simulation studies, in which it has been possible to substantially improve the punctuality and reduce the average delays by only increasing the advertised travel times slightly. Further, the results have also indicated a positive socio-economic benefit. This demonstrates the methods potential usefulness and motivates further research. / För järnvägen har tidtabellen en central roll, och dess kvalité har stor betydelse för kapacitet och tillförlitlighet. Processen att konstruera en tidtabell är ofta en uppgift som utförs manuellt med begränsat datorstöd och på grund av beroenden mellan enskilda tåg är det ofta ett tidskrävande och svårt arbete. Dessa tågberoenden gör det svårt att manuellt konstruera konfliktfria tidtabeller samtidigt som det också är svårt att manuellt förbättra en given tidtabell, vilket beror på att de är svårt att förutsäga vad effekten av en given ändring blir. Eftersom efterfrågan på järnväg fortsatt förväntas öka, finns det ett behov av att kunna köra fler tåg. Samtidigt pågår det redan i många europeiska länder en offentlig debatt om järnvägen punktlighet, vilken riskeras att försämras vid högre kapacitetsanvändning. Därför finns det även ett behov av att förbättra tidtabellernas robusthet, där robusthet syftar till en tidtabells möjlighet att stå emot och återhämta mindre förseningar. För att hantera denna målkonflikt kommer det behövas ökad precision vid både planering och drift, vilket kan uppnås med en högre grad av automation. Forskningen i denna avhandling syftar till att förbättra robustheten för tågtidtabeller genom att kombinera mikro-simulering med matematisk optimering, två metoder som redan används i hög grad av både yrkesverksamma trafikplanerare och forskare men som sällan kombineras. I den här avhandlingen förslås en sekventiell metod baserad på att simulera en given tidtabell och optimera den för att minska den viktade summan av planerad restid och predikterad medelförsening. Metoden har visat på lovande resultat i simuleringsstudier, där det har varit möjligt att uppnå en väsentligt bättre punktlighet och minskad medelförsening, genom att endast förlänga de planerade restiderna marginellt. Även förbättrad samhällsekonomisk nytta har observerats av att tillämpa den föreslagna metoden. Sammantaget visar detta metodens potentiella nytta och motiverar även fortsatt forskning. / <p>QC 20191112</p>

This dissertation explores the modular learning in artificial neural networks that mainly driven by the inspiration from the neurobiological basis of the human learning. The presented modularization approaches to the neural network design and learning are inspired by the engineering, complexity, psychological and neurobiological aspects. The main theme of this dissertation is to explore the organization and functioning of the brain to discover new structural and learning inspirations that can be subsequently utilized to design artificial neural network. The artificial neural networks are touted to be a neurobiologicaly inspired paradigm that emulate the functioning of the vertebrate brain. The brain is a highly structured entity with localized regions of neurons specialized in performing specific tasks. On the other hand, the mainstream monolithic feed-forward neural networks are generally unstructured black boxes which is their major performance limiting characteristic. The non explicit structure and monolithic nature of the current mainstream artificial neural networks results in lack of the capability of systematic incorporation of functional or task-specific a priori knowledge in the artificial neural network design process. The problem caused by these limitations are discussed in detail in this dissertation and remedial solutions are presented that are driven by the functioning of the brain and its structural organization. Also, this dissertation presents an in depth study of the currently available modular neural network architectures along with highlighting their shortcomings and investigates new modular artificial neural network models in order to overcome pointed out shortcomings. The resulting proposed modular neural network models have greater accuracy, generalization, comprehensible simplified neural structure, ease of training and more user confidence. These benefits are readily obvious for certain problems, depending upon availability and usage of available a priori knowledge about the problems. The modular neural network models presented in this dissertation exploit the capabilities of the principle of divide and conquer in the design and learning of the modular artificial neural networks. The strategy of divide and conquer solves a complex computational problem by dividing it into simpler sub-problems and then combining the individual solutions to the sub-problems into a solution to the original problem. The divisions of a task considered in this dissertation are the automatic decomposition of the mappings to be learned, decompositions of the artificial neural networks to minimize harmful interaction during the learning process, and explicit decomposition of the application task into sub-tasks that are learned separately. The versatility and capabilities of the new proposed modular neural networks are demonstrated by the experimental results. A comparison of the current modular neural network design techniques with the ones introduced in this dissertation, is also presented for reference. The results presented in this dissertation lay a solid foundation for design and learning of the artificial neural networks that have sound neurobiological basis that leads to superior design techniques. Areas of the future research are also presented. / Ph. D.

The thesis aims to introduce an efficient resource development strategy in electric utility long term planning under uncertainty considerations. In recent years, electric utilities have recognized the concepts of robustness, flexibility, and risk exposure, to be considered in their resource development strategy. The concept of robustness means to develop resource plans that can perform well for most, if not all futures, while flexibility is to allow inexpensive changes to be made if the future conditions deviate from the base assumptions. A risk exposure concept is used to quantify the risk hazards in planning alternatives for different kinds of future conditions. This study focuses on two technical issues identified to be important to the process of efficient resource development: decision-making analysis considering robustness and flexibility, and decision-making analysis considering risk exposure. The technique combines probabilistic methods and tradeoff analysis, thereby producing a decision set analysis concept to determine robustness that includes flexibility measures. In addition, risk impact analysis is incorporated to identify the risk exposure in planning alternatives. Contributions of the work are summarized as follows. First, an efficient resource development framework for planning under uncertainty is developed that combines features of utility function, tradeoff analysis, and the analytical hierarchy process, incorporating a performance evaluation approach. Second, the multi-attribute risk-impact analysis method is investigated to handle the risk hazards exposed in power system resource planning. Third, the penetration levels of wind and photovoltaic generation technologies into the total generation system mix, with their constraints, are determined using the decision-making model. The results from two case studies show the benefits of the proposed framework by offering the decision makers various options for lower cost, lower emission, better reliability, and higher efficiency plans. / Ph. D.

The rapid advancement of cutting-edge techniques has propelled state-of-the-art (SOTA) language models to new heights. Despite their impressive capabilities across a variety of downstream tasks, large language models still face many challenges such as hallucination and bias. The thesis focuses on two key objectives: first, it measures the robustness of T0_3B and investigates feasible methodologies to enhance the model’s robustness. Second, it targets on the explainability of large language models, aiming to make the intrinsic working mechanism more transparent and, consequently enhance model’s steerability. Motivated by the importance of mitigating non-robust behavior in language models, the thesis initially measures model’s robustness on handling minor perturbation. After that, I proposed and verified an approach to enhance robustness by making input more contextualized, a method that does not require the step of fine-tuning. Moreover, to understand the complex working mechanism of large language models, I designed and introduced two novel visualization tools: ’Logit Lens’ and ’Hidden States Plot in Spherical Coordinate System’. These tools, combined with additional experimental analysis, revealed a noticeable differentiation of the predicted processes between the first predicted token and subsequent tokens. The contributions of the thesis are mainly in the two following aspects: it provides feasible methodologies to enhance the robustness of language models without the need of fine-tuning, and it contributes to the field of explainable AI through the development of two visualization tools that shed light on the understanding of the working mechanism. / Den snabba utvecklingen av banbrytande tekniker har drivit språkmodeller till nya höjder. Trots deras imponerande prestanda över diverse språkrelaterade uppgifter, trots detta har dessa modeller fortfarande problem som hallucinationer och bias. Avhandlingen är centrerad kring två huvudmål: för det första undersöker den robustheten hos T0_3B och undersöker framtida strategier för att förbättra dess robusthet. För det andra utforskar den språkmodellernas ”förklaringsbarhet” (dvs hur väl vi förstår deras beteende), i syfte att göra dem mer transparenta och följaktligen förbättra modellens styrbarhet. Det första vi gör är att visa experiment som vi har satt upp för att mäta modellens robusthet mot mindre störningar. Som svar föreslår och underbygger vi ett tillvägagångssätt för att öka robustheten genom att ge modellen mer kontext när en fråga ställs, en metod som inte kräver vidare träning av modellen. Dessutom, för att förstå den komplexiteten hos språkmodeller, introducerar jag två nya visualiseringsverktyg: Logit Lens och Hidden States Plot i sfäriskt koordinatsystem. Dessa verktyg, i kombination med ytterligare experimentell analys, avslöjar ett diskting mönstr för den första förutspådda ordet jämfört med efterföljande ord. Bidragen från avhandlingen är huvudsakligen i de två följande aspekterna: den ger praktiska åtgärder för att förbättra robustheten hos språkmodeller utan behov av vidare träning, och den bidrar till området för förklarabar AI genom utvecklingen av två visualiseringsverktyg som ökar våran förståelse för hur dessa modeller fungerar.