Search

In recent studies, electrical resistance tomography (ERT) has been explored as a non-destructive testing imaging modality in conjunction with structural health monitoring (SHM). This imaging modality has been shown to be able to locate cracks in cement-based materials as well as reconstruct strain and stress distributions in nano-composite materials. However, due to the ill-conditioned nature of the ERT inverse problem, the computational cost of solving such problems can be high. In order to reduce the overall computational cost of solving the ERT inverse problem in practical applications, we propose using a deep learning approach to address this challenge. The deep-learned ERT frameworks have been successfully implemented and validated using simulation and experimental data for various materials relevant to SHM. The results indicate that the deep-learned ERT frameworks are feasible for implementation in SHM applications.

Machine learning is playing a pivotal role in medical image analysis. Many algorithms based on machine learning have been applied in medical imaging to solve classification, detection, and segmentation problems. Particularly, with the wide application of deep learning approaches, the performance of medical image analysis has been significantly improved. In this thesis, we investigate machine learning methods for two key challenges in medical image analysis: The first one is segmentation of medical images. The second one is learning with weak supervision in the context of medical imaging. The first main contribution of the thesis is a series of novel approaches for image segmentation. First, we propose a framework based on multi-scale image patches and random forests to segment small vessel disease (SVD) lesions on computed tomography (CT) images. This framework is validated in terms of spatial similarity, estimated lesion volumes, visual score ratings and was compared with human experts. The results showed that the proposed framework performs as well as human experts. Second, we propose a generic convolutional neural network (CNN) architecture called the DRINet for medical image segmentation. The DRINet approach is robust in three different types of segmentation tasks, which are multi-class cerebrospinal fluid (CSF) segmentation on brain CT images, multi-organ segmentation on abdomen CT images, and multi-class tumour segmentation on brain magnetic resonance (MR) images. Finally, we propose a CNN-based framework to segment acute ischemic lesions on diffusion weighted (DW)-MR images, where the lesions are highly variable in terms of position, shape, and size. Promising results were achieved on a large clinical dataset. The second main contribution of the thesis is two novel strategies for learning with weak supervision. First, we propose a novel strategy called context restoration to make use of the images without annotations. The context restoration strategy is a proxy learning process based on the CNN, which extracts semantic features from images without using annotations. It was validated on classification, localization, and segmentation problems and was superior to existing strategies. Second, we propose a patch-based framework using multi-instance learning to distinguish normal and abnormal SVD on CT images, where there are only coarse-grained labels available. Our framework was observed to work better than classic methods and clinical practice.

Pension schemes are facing more difficulties on matching their underlying liabilities with assets, mainly due to faster mortality improvements for their underlying populations, better environments and medical treatments and historically low interest rates. Given most of the pension schemes are relatively much smaller than the national population, modelling and forecasting the small populations' longevity risk become urgent tasks for both the industrial practitioners and academic researchers. This thesis starts with a systematic analysis on the influence of population size on the uncertainties of mortality estimates and forecasts with a stochastic mortality model, based on a parametric bootstrap methodology with England and Wales males as our benchmark population. The population size has significant effect on the uncertainty of mortality estimates and forecasts. The volatilities of small populations are over-estimated by the maximum likelihood estimators. A Bayesian model is developed to improve the estimation of the volatilities and the predictions of mortality rates for the small populations by employing the information of larger population with informative prior distributions. The new model is validated with the simulated small death scenarios. The Bayesian methodologies generate smoothed estimations for the mortality rates. Moreover, a methodology is introduced to use the information of large population for obtaining unbiased volatilities estimations given the underlying prior settings. At last, an empirical study is carried out based on the Scotland mortality dataset.

While the fuel resource scarcity and the environment crisis are becoming two of the major problems for the human society in the new century, in the context that the on-road transport is the largest energy-consuming society sector, electrical vehicles (EVs) serving resource-sustainable environmentally-clean transport attract increasing attention. Permanent magnet (PM) traction machines with high torque and power densities, and high energy efficiency gain great interest from the engineer and industry communities. Jointly with the chances, fresh challenges are brought by EVs to PM machine designers. In response to the vehicles' call for high speed, mighty acceleration, long range, safe and robust system, the engineers need to develop a powerful design platform that allows for multi-physic evaluations of machine designs over a large torque-speed range, especially for energy efficiency, PM health, and thermal-withstanding ability. More importantly, these evaluations must be against driving cycles rather than on a single rated operation point in catering to various real-world driving conditions. The complex and enormous computation efforts required necessitate new, effective and feasible design techniques. In this work a set of modelling techniques for PM machines are developed, in order to establish a computationally-efficient yet accurate design and optimisation method for EV traction PM machines. Through the method, comprehensive machine multi-physics assessments against driving cycles are enabled; electro-thermally coupled evaluation is achieved; 3-dimensional eddy-current loss of PMs are accurately monitored in the context of PM protection; lastly with the techniques integrated together, a fast and effective optimisation method for EV traction PM machines is acquired. To exploit the benefits of the proposed method, a design, optimisation and manufacture process of PM machines for a light-duty EV distributed traction system is formatted, which includes a quantity assessment of machine topologies, investigation of driving cycle influence on designs, optimisation of the selected 18-slot 8-pole interior PM machine against a series of EV machine design criteria, the subsequent prototype experiments, and optimisation of combinations and power split ratios of PM machines for the distributed traction system. Through the design process for the EV traction system and experiments, the effectiveness, computational efficiency, and accuracy of the proposed designing methods are exhibited and validated.

International business activities and foreign direct investment in particular involve an element of risk and uncertainty, sometimes to a great extent. Despite this almost axiomatic statement, less academic research has been conducted on this subject, compared to the considerations of return and cost. While the globalisation era introduces an integration of world economy, ever more diverse types of risks are looming on the horizon in relation to cross-border investment, ranging from political unrest to creeping appropriation, from natural disaster to terrorism, and from technological failure to industrial espionage, all of which tend to deter the free movement of capital and value-adding resources around the world. This thesis provides by far the most extensive review of the research on risk and uncertainty across the macro management fields of studies. Particular attention has been paid to construct clarity and how construct clarity can serve to categorise a vast body of knowledge and address previous inconsistent conclusions. To further the inquiry, this thesis focuses on one of the conceptualisations of risk that is most relevant to the context of foreign investment decision making. New insights are generated by proposing a microfoundational framework based on a key construct – risk propensity – as a necessary complement to the current research on risk in foreign direct investment. Conventional understanding of the capabilities paradigm is reformulated in this light. In order to test the efficacy of “risk propensity” and the static assumption of risk preference embedded in the conventional theories, this thesis draws upon quasi-experiment methods and models managerial heterogeneity based on stated preference data. In addition to individual level choice modelling, firm level analysis of location choice is conducted, yielding new insights into the role of experience in firms’ decision making. China provides the empirical setting for both analyses. It is found that a theoretical understanding of risk helps explain the varying effect of context and experience on risk-taking. Generalisations of this statement may be made to strategic decision making, organisational learning and behavioural strategy research.

A category CoLog of distributive laws is introduced to unify different approaches to modal logic for coalgebras, based merely on the presence of a contravariant functor P that maps a state space to its collection of predicates. We show that categorical constructions, including colimits, limits, and compositions of distributive laws as a tensor product, in CoLog generalise and extend existing constructions given for Set coalgebraic logics and that the framework does not depend on any particular propositional logic or state space. In the case that P establishes a dual adjunction with its dual functor S, we show that a canonically defined coalgebraic logic exists for any type of coalgebras. We further restrict our discussion to finitary algebraic logics and study equational coalgebraic logics. Objects of predicate liftings are used to characterise equational coalgebraic logics. The expressiveness problem is studied via the mate correspondence, which gives an isomorphism between CoLog and the comma category from the pre-composition to the post-composition with S. Then, the modularity of the expressiveness is studied in the comma category via the notion of factorisation system.

CVD diamond possesses certain attractive electrochemical properties inter alia low background current, broad potential window, chemical inertness and resistance to electrocorrosion and fouling. As a consequence its use in various areas of electrochemistry, such as electrochemical sensing, wastewater treatment and electrocatalysis is being explored. Unfortunately, alongside these attractive features, bare CVD diamond electrodes, in common with all other electrode materials, cannot be effectively applied in all electrochemical systems of interest, since for example it may not display useful electrochemical activity for the redox process of interest. In these circumstances it may be possible to modify the electrode by addition of other chemical species to the surface, to introduce the relevant activity. One of the main aims of this thesis was therefore to investigate the properties of certain chemical modifications to the diamond electrode surface. A second aim was also to explore for the first time the use of a practically useful form of single crystal diamond, so-called heteroepitaxial diamond, in electrochemistry. The diamond electrodes used were boron-doped material grown by chemical vapour deposition. A range of electrochemical methods, including especially cyclic voltammetry, square-wave voltammetry, impedance spectroscopy and scanning electrochemical microscopy, were used to characterise electrode properties. Other physical methods employed included scanning electron and atomic force microscopy, X-ray photoelectron spectroscopy and dynamic light scattering techniques. The electrochemical properties of heteroepitaxial single crystal diamond were explored and compared to polycrystalline counterparts. The single crystal diamond electrode was found to show superior properties in terms of wide potential window, low background current and homogeneous activity across the electrode surface, coupled with resistance to fouling. Heterogenous electron transfer rate constants were found to be lower than normally found on polycrystalline diamond; this was attributed to reduced density of states and absence of functional groups. An electrochemical route to the preparation of diamond electrodes, modified by PrOx@Pt core-shell particles was demonstrated. It was observed that these electrode modifiers were far less susceptible to poisoning than bare Pt nanoparticles when used in the electrochemical oxidation of methanol. It was also shown that diamond electrodes with these core-shell particles deposited on them, displayed useful activity for the electrochemical oxidation of nitric oxide. The presence of the PrOx layer was shown to impart useful selectivity against the oxidation of interfering compounds such as nitrite and ascorbic acid, without the loss of sensitivity which normally occurs if nafion coatings are used instead. Basic electrochemical characterisation of the PrOx coating showed that the layer was chemically active and did not serve as a simple blocking layer when deposited on the electrode. The activity of Pt modified diamond electrodes for the oxidation of nitrite species was also studied. It was also shown that the addition of carbon black to a diamond electrode resulted in much enhanced electrochemical properties in the detection of riboflavin.

Role-based access control (RBAC) has become today's dominant access control model, and many of its theoretical and practical aspects are well understood. However, certain aspects of more advanced RBAC models, such as the relationship between permission usage and role activation and the interaction between inheritance and constraints, remain poorly understood. Moreover, the computational complexity of some important problems in RBAC remains unknown. In this thesis we consider these issues, develop new RBAC models and answer a number of these questions. We develop an extended RBAC model that proposes an alternative way to distinguish between activation and usage hierarchies. Our extended RBAC model has well-defined semantics, derived from a graph-based interpretation of RBAC state. Pervasive computing environments have created a requirement for access control systems in which authorization is dependent on spatio-temporal constraints. We develop a family of simple, expressive and flexible spatio-temporal RBAC models, and extend these models to include activation and usage hierarchies. Unlike existing work, our models address the interaction between spatio-temporal constraints and inheritance in RBAC, and are consistent and compatible with the ANSI RBAC standard. A number of interesting problems have been defined and studied in the context of RBAC recently. We explore some variations on the set cover problem and use these variations to establish the computational complexity of these problems. Most importantly, we prove that the minimal cover problem -- a generalization of the set cover problem -- is NP-hard. The minimal cover problem is then used to determine the complexity of the inter-domain role mapping problem and the user authorization query problem in RBAC. We also design a number of efficient heuristic algorithms to answer the minimal cover problem, and conduct experiments to evaluate the quality of these algorithms.

This thesis presents and studies a timed computational model of parallelism, a Timed Calculus of Communicating Systems or Timed CCS for short. Timed CCS is an extension of Milner's CCS with time. We allow time to be discrete, such as the natural numbers, or dense, such as the non-negative rationals or the non-negative reals. We make no assumption of the Maximal Progress Principle, but the calculus is consistent with the principle. Time variables in the language allow us to express a notion of time dependency and the language is more expressive than those without time variables or infinite summation. We extend the well known notion of bisimulation to timed processes and study the abstract sensation of timed processes. We show that strong equivalence (the largest strong bisimulation) is decidable for finite processes, i.e. processes without recursion. The decidability is independent of the choice of time domain. We also present a simple proof system for strong equivalence and the proof system is again independent of the choice of time domain. We show that the proof system is sound and complete for finite processes over dense time domains, but only complete for a restricted language over discrete time domains. We discuss how to modify the definition of time expressions to get the restricted language. We also study behavioural abstraction in timed processes. The thesis also presents and studies a general model, Timed Synchronisation Trees, for timed calculi. Timed synchronisation trees are extensions of synchronisation trees with time. All constructions on timed synchronisation trees are continuous with respect to a natural complete partial order. We can interpret a wide range of real-time process algebras in timed synchronisation trees. As an example, we give a denotational semantics for Timed CCS based on timed synchronisation trees. We show that the denotational and operational semantics of Timed CCS coincide. CCS is a symbolic calculus in the sense that it treats solely the observation of events of a system. The relative time, location and duration of events are abstracted away from the consideration. If we postulate that every action has a non-zero constant duration, we can observe the usual notions of causality, concurrency and conflict relations of events of a system. By interpreting CCS in Timed CCS based on a postulation that for any two events which are causally related there is at least a non-zero constant delay between them, we get a timed semantics for CCS. The timed semantics of CCS is a partial order or true concurrency semantics. As a consequence, we develop a partial order or true concurrency semantics based on an interleaving approach.

As a familiar landscape of American neighborhoods, vacant urban land evokes negative images and is often conceived as a barrier to sustainable development. To enhance the knowledge of vacant land and facilitate evidence-based policymaking, this research examines the relationship between vacant urban land and critical neighborhood sustainability measures in the largest Midwestern city, Chicago, Illinois. In particular, this research, consisting of three essays, offers an in-depth observation and investigation of the multidimensionality, reciprocal causality, and spatial heterogeneity of the relationship. This research has three primary findings: (1) the associations between vacant land and neighborhood sustainability measures are multidimensional and differ by individual measures, (2) vacant land has solid reciprocal associations with crime and poverty at the neighborhood scale over time, and (3) the associations between vacant land and neighborhood sustainability measures are spatially non-stationary. These findings encourage further studies to expand investigations to all sustainability dimensions and prompt vacant land management policies to be long-term and spatially adaptive for different urban contexts.

Sizing and topology optimization are the two main structural optimization tools in a wide range of applications in aerospace, mechanical, and design. An iterative process solves the sizing optimization using classical gradient-based methods, usually carried out with an integrated process including a full-scale finite element analysis (FEA) to evaluate the design performance and a gradient search step at each iteration. With a complex real-world model, the optimization process is extremely cumbersome, time-consuming, and with no guarantee for an optimal solution or design. Alternatively, global population-based methods, such as genetic algorithm and particle swarm, can achieve the global optimal design with many simulations for every iteration to evaluate different designs for searching for the best candidates. This tremendous computational effort for simulations at each iteration prevents the global method from optimizing with complex physics simulation models. As for topology optimization, state-of-the-art methods, such as the Solid Isotropic Material with Penalty (SIMP) method, uses hand-coded gradient functions for optimization and must be run repeatedly for different boundary and loading conditions. Several practical and efficient machine-learning-based data-driven approaches have been proposed to optimize structures instantaneously using the generative adversarial network. Nevertheless, a complex machine learning model is costly because of the large amount of data and long training time.This dissertation presents several new, rapid, and accurate optimal design approaches for improving current structural sizing and topology optimization methods. First, for sizing optimization, to reduce the optimization time while preserving global optimality, a new optimization framework with response surface method and global sensitivity method is presented to approximate the simulation model with high accuracy while using a minimum number of simulations. The response surface approximates the original physics model using a mathematical basis, which reduces the time for function evaluation significantly. Global sensitivity analysis is conducted to reduce the model dimension and further reduce the number of function evaluations. A novel vehicle subframe compliance optimization framework is proposed using a response surface. The response surfaces approximate vehicle dynamic performance from multi-body dynamic simulation and subframe structure compliance from finite element analysis. It is shown that the mass of the subframe is reduced without violating the vehicle performance constraints. A physics-informed deep neural network for the structural topology optimization process is proposed in the second part of the dissertation. The deep neural network is trained with a built-in finite element model. The finite element model is made to be fully differentiable, and with the help of automatic differentiation, the gradient of the finite element model can be backpropagated to adjust the weights of the neural network. It is shown that the neural network can quickly learn the optimized structure without using much training data. Furthermore, no hand-code adjoint/gradient equations are required since the neural network can be trained to generate optimal structures based on the gradient information directly from the finite element model via automatic differentiation. The benefits of this physics-informed deep learning approach provide a data-efficient training process to construct a generator neural network to learn optimized topology under different design target conditions.

As US cities continue to accommodate more people and jobs, they have created increasing travel demands, especially on inter-city commute. Due to the spatial distance among cities and a long tradition of car-oriented lifestyle in the south, cars are the major mode for people traveling to different cities. With emerging papers and reports on building a regional framework for the US mega-region, a sustainable transportation network with various transport options has become a heated topic for state and local transportation agencies. Multiple National Household Travel Surveys (NHTS) have shown that private vehicles dominated intra- and inter-megaregion travel in the United States and such travel pattern will cause further congestion on regional highways and negatively impact passenger and commodity flows in mega-region. An efficient mobility supply for megaregions aims to achieve multi-modality that utilize different modes (automobile, rail, bus, and air) for mega-regional travel. This report utilizes the National Household Travel Survey and ACS commuting flow data to explore the travel patterns of Texans. A mode choice model from National Cooperative Rail Research Program (NCRRP) Report 4 was calibrated to investigate how mode share would change for travelers in the twin-megaregion area that includes the Texas Triangle and western Louisiana. The aggregated findings provide solutions for effective network performance, and the report further discusses the possibilities of modern railway service in the twin megaregion-area. This study starts two case studies where train transit can be effective solutions to transportation supply. Then it explores the traffic corridors in these two states with commuting survey and data. Last, trip generation and mode choice analysis confirm the abundance of rail riders in the future, and the report offers policy suggestions for policymakers to prepare for the potential changes

Modern research on gene regulation and disorder-related pathways utilize the tools such as microarray and RNA-Seq to analyze the changes in the expression levels of large sets of genes. In silico motif discovery was performed based on the gene expression profile data, which generated a large set of candidate motifs (usually hundreds or thousands of motifs). How to pick a set of biologically meaningful motifs from the candidate motif set is a challenging biological and computational problem. As a computational problem it can be modeled as motif selection problem (MSP). Building solutions for motif selection problem will give biologists direct help in finding transcription factors (TF) that are strongly related to specific pathways and gaining insights of the relationships between genes. This study implemented an algorithm based on simulated annealing (SA) optimization algorithm for the motif selection problem, and investigated the properties of the implemented algorithm with the real world datasets (ENCODE project data). The results of evaluation based on ENCODE datasets indicate that simulated annealing algorithm is good for solving motif selection problem. The performance of simulated annealing algorithm can be tuned based on some parameters to fit for special requirements. Future improvement may be achieved via extending algorithm model (adaptive simulated annealing) and applying high dimensional cost function.

The key transcription factor sterol regulatory element binding protein-1 (SREBP1) plays a central role in milk fat synthesis. SREBP1 stimulates the transcription of genes encoding lipogenic enzymes. The overall objective of these studies was to investigate the mechanisms of SREBP1 regulation by nutrients. In the first study, chromatin immunoprecipitation (ChIP) accompanied with deep-sequencing was employed to investigate the potential sterol regulatory elements (SRE) in the promoter of SREBP1-target genes. The SRE in three known SREBP1-target genes SREBP1, fatty acid synthase (FASN) and stearoyl-CoA desaturase (SCD) were first validated in a bovine mammary epithelial cell line (MacT) and in bovine mammary tissues. At least one or two SRE binding sites in 24 selected lipogenic genes were identified within 50,000 base pair to the 5'-transcription start site through ChIP-seq. The genes closest to the highest enriched peaks were involved in cell integrity, defense or signal transduction whereas lipogenic genes were not among the top enrichment leading to the questions about the success of the ChIP. The second study was conducted to determine the effect of t10, c12-conjugated linoleic acid (CLA) on insulin induced gene-1 (Insig1), an endoplasmic reticulum (ER) protein that anchors SREBP1 and prevents proteolytic activation of SREBP1. MacT cells were treated with increasing levels of t10, c12-CLA. High concentration of t10, c12-CLA inhibited Insig1 degradation therefore decreased SREBP1 maturation. Furthermore, immunoprecipitation (IP) confirmed that t10, c12-CLA reduced Insig1 proteasomal degradation by disrupting the interaction between Insig1 and UBX domain-containing protein 8 (Ubxd8), which is part of a degradation complex that removes Insig1 from the ER. In the third study, three potential regulators of SREBP1 activation and their pathways were investigated in insulin, t10, c12-CLA or glucose treated MacT cells. Insulin-induced mammalian target of rapamycin (mTOR) signaling stimulated lipogenesis via activation of SREBP1 and the stimulatory effect was based on the regulation on cAMP response element binding protein coactivator 2 (CRTC2) phosphorylation, Lipin1 translocation and glycogen synthase kinase-3 (GSK3)-dependent proteasomal degradation. t10, c12-CLA inhibited SREBP1 through AMP-activated protein kinase (AMPK) phosphorylation, a key protein kinase in energy homeostasis. Glucose stabilized the SREBP1 chaperone protein SCAP and facilitated SREBP1 activation. Overall, SREBP1 activation is under specific regulation of t10, c12-CLA and interacts with multiple major cellular signaling pathways in response to hormonal stimulation and nutrient availability.

An experimental investigation has been performed to verify heat flux measurements on a metallic film-cooled flat plate by using infrared thermography in a transient facility. Infrared thermography provides high-resolution temperature distributions and makes it easy to locate hot spots on the object of interest. Previous work has shown that infrared thermography can produce accurate measurement for an uncooled flat plate. The goal of this thesis is to show that infrared thermography can also measure heat flux for film-cooled components and lay the foundation for its use in full-scale rotating experiments. A stainless steel plate with rows of cooling holes was built for testing in a blowdown facility. During the experiment, the plate was exposed to hot main flow and supplied with low temperature air through the cooling holes. The heat flux on the plate surface was determined by performing a 3D ANSYS transient thermal analysis using the infrared temperature distribution as time dependent top surface boundary conditions. The challenge of this technique for the film-cooled plate is that the boundary conditions are not known for the cooling channel walls and backside walls that exposed to cooling air since it is not practical to obtain measurements in those regions. Although strong jet impingement and forced convection are taking place on these walls, they are treated as adiabatic for the numerical analysis and the analysis time window is kept short so that through-wall conduction will not affect the top surface during the run. It is shown that the through-wall conduction from the backside and cooling channel walls only impact the regions right upstream the cooling holes and only after a relatively long run time. The fidelity of this technique is verified by comparing the results calculated based on the infrared images to the results obtained from traditional heat-flux gauges. The infrared thermography and the heat-flux gauges measurements agree within 10% for the regions that are not affected by backside and cooling channel walls boundary conditions. When the boundary conditions and through-wall conduction do impact the results, the adiabatic wall assumption causes the infrared thermography to predict a lower heat flux than the heat-flux gauges. After the method had been clearly demonstrated, it was used to make comparisons between cooled and uncooled cases to clearly identify how the film cooling spreads and mixes from the rows of cooling holes.

Supplier development (SD) has been intensively and increasingly used in practice and studied in academia. Many studies find that SD can generate operational, capability-based, attitudinal, and financial performance measures for both the supplying firm (supplier) and the buying firm (buyer), but very few studies systematically explain why SD yields supplier’s performance improvements and, in turn, buyer’s performance improvements. Using a meta-analysis approach, this dissertation finds that SD does lead to positive outcomes, but SD is found to have very weak or even negative relationship with performance improvements in some cases. Such findings further support the importance of examining the main research question: why SD works. In order to answer the main research question, this dissertation adopts a multiphase triangulation approach: theoretical construction, conceptual examination, and empirical examination. Doing so, this dissertation constructs and validates a knowledge management (KM) view of SD. The purpose of theoretical construction (Chapter 3) is to develop a KM view of supplier development via a systematic view of previous studies. Presented in Chapter 4, conceptual examination reveals that all SD activities can be subsumed into KM activities, and further conceptually supports the feasibility of the KM view in SD. Empirical examination, including a survey of 39 SD scholars and a survey of 295 SD practitioners (156 complete responses), is presented in Chapters 5 and 6. Most hypotheses are strongly supported, demonstrating the importance of the knowledge-management view of SD. Overall, this dissertation has both theoretical contributions for KM and SD sides, and practical contributions for researchers, practitioners, and educators/students. First, it contributes by supporting the addition of KM variables to other theories when explaining why SD works, confirming the role of KM in SD, providing a complete KM view of SD, and revealing why SD works. Second, it contributes by implementing mixed research methods, integrating multiple disciplines, and exemplifying collecting data on LinkedIn. Third, it contributes by offering a catalog of SD activities and guidance for designing, implementation, and evaluation of SD initiatives. Fourth, it contributes by advancing a mental model to understand SD literature. Conclusions, limitations, and future research directions are also discussed.

Scene understanding, such as image classification and semantic image segmentation, has been a challenging problem in computer vision. The difficulties mainly come from the feature representation, i.e., how to find a good representation for images. Instead of improving over hand-crafted features such as SIFT or HoG, we focus on learning image representations by generative and discriminative methods.In this thesis, we explore three areas: (1) generative models, (2) graphical models, and (3) deep neural networks for learning image representations. In particular, we propose a dictionary of epitomes, a compact generative representation for explicitly modeling object co-relation within edge patches, and for explicitly modeling photometric and position variability of image patches. Subsequently, we exploit Conditional Random Fields (CRFs) to take into account the dependencies between outputs. Finally, we employ Deep Convolutional Neural Networks trained with large-scale datasets to learn feature representations. We further combine CRFs with deep networks to estimate complex representations. Specifically, We show that our proposed model can achieve state-of-art performance on challenging semantic image segmentation benchmarks.

Abstract: The formation of stable water-in-oil emulsions during bitumen extraction poses problems for water separation from diluted bitumen, which leads to equipment corrosion and catalyst fouling in downstream operations. Demulsifiers are used to break the stable emulsions and assist the separation of water from diluted bitumen. To study the factors influencing demulsification, the efficiency of four industrial demulsifiers in dewatering of water-in-toluene diluted bitumen emulsions was probed at various temperatures. The ability of the demulsifiers to compete for the interface was assessed by measuring interfacial tension of the toluene diluted bitumen-water interface in the presence and absence of demulsifiers. Demulsification tests were conducted using two different methods. A dynamic method was used to allow in-situ and real-time observation of the demulsification process. A static method was used to probe water removal efficiency of the four demulsifiers by gravity to assist the understanding of the demulsifiers’ performance. In order to understand the impact of temperature on demulsification, the effect of temperature on the viscosity of the diluted bitumen was investigated and correlated to demulsification efficiency.

Previous studies on hot biopsy forceps for colorectal polypectomy indicated that simulated ablation tests could provide useful information on hot biopsy forceps design and ablation parameter optimization that would help reducing the rate of potential polypectomy complications such as perforation. In this thesis, a novel ex vivo ablation test model for colonoscopic monopolar hot biopsy forceps was designed and implemented; simulated ablation tests had also been performed on this ablation test model for (a) verifying the model design and implementation, and (b) finding the roles of the major ablation parameters in tissue injury during the ablation process. In the design of this novel ex vivo ablation test model, a body equivalent block was introduced to simulate the properties of the human body. Porcine colon was used as tissue source for the simulated ablation tests. Major ablation parameters that were tested on this model covered (a) output power level, (b) firing duration, (c) contact area at the electrode tip, and (d) electrical impedance. Simulated ablation tests that were performed on this ablation test model covered the main combinations of targeted ablation parameters. During simulated ablation tests various ablation sites were generated under different parameter combinations, which indicated that this ablation test model can be used for hot biopsy forceps ablation studies. Some ablation parameters such as output power level and firing duration had dominating effects on the resulting ablation site geometry and tissue injury. The simulated test results showed the potential applications for testing hot biopsy forceps and optimizing ablation parameters.

IKKβ is a protein kinase critical for the transmission of inflammatory cytokine signals, such as TNFα, to activation of transcription factor NF-κB and regulation of gene expression. Genetic studies in mice, however, show that IKKβ is not only involved in inflammation, but also plays a pivotal role in tissue homeostasis and stress responses. Our previous studies show that IKKβ is essential for maintaining redox homeostasis, but the underlying mechanisms and biological consequences have not been understood. The central theme of this thesis is to investigate how IKKβ modulates homeostasis and what roles it plays in biological processes. The thesis is composed of three main components. First, mouse fibroblasts were used to investigate the molecular signatures and functions of IKKβ. I showed that genetic knockout of Ikkβ gene in fibroblasts reduced the expression of anti-oxidants thus increased the intracellular ROS level. The increased ROS in turn activated AP-1, leading to transcriptional activation of the TGFβ gene promoter and its expression, while increased TGFβ can act through NOX4 to further stimulate ROS production. Removal of IKKβ therefore triggered the autocrine amplification of the ROS-TGFβ loop that led to progressive ROS accumulation. The IKKβ-deficient fibroblasts ultimately displayed SMAD activation, leading to the overproduction of extracellular matrices and remodeling enzymes, accelerated cell motility, and enhanced markers for myofibroblast transformation and cellular senescence. These results suggested that IKKβ safeguarded the homeostasis of fibroblasts by repressing a ROS-AP-1-TGFβ autocrine regulatory loop. Second, the in vivo roles of IKKβ were assessed in keratocytes, the fibroblasts in corneal stoma of the eye. I generated mice with Ikkβ ablation specifically in corneal keratocytes. While the knockout mice had normal corneal appearance, suggesting that IKKβ was dispensable for corneal keratocytes development and maintenance, they exhibited aggravated infiltration of inflammatory cells, cellular senescence and severe scar formation compared to normal mice in response to alkali burn injury. The wounded corneas of the knockout mice also exhibited enhanced activation of stress response and fibrogenic pathways, including JNK/AP-1 and TGFβ/SMAD cascade. Hence, similar to fibroblasts, keratocytes in corneal stroma utilize IKKβ for maintaining homeostasis and repressing ROS and TGFβ during injury responsesFinally, the roles of IKKβ in epithelial cells of the cornea were investigated by generating and using the corneal epithelial-specific Ikkβ knockout mice. These mice had normal cornea development and maintenance, but significantly delayed healing of corneal epithelium wounds. Furthermore, delayed wound healing in the knockout cornea was not due to reduced cell proliferation or increased apoptosis, but was correlated to a slower epithelial cell migration. Taken together, results presented in this thesis reveal that IKKβ has highly cell type-specific functions apart from its well-recognized roles in inflammatory responses. While IKKβ promotes corneal epithelial cell migration at the surface of ocular tissues, it prevents senescence, fibrosis and scar tissue formation in corneal stromal fibroblasts in response to injuries. Correspondingly, IKKβ represses a ROS-TGFβ regulatory loop in fibroblasts, thereby preventing ROS, cell migration, senescence and fibrosis.

We are witnessing a growing number of applications that involve both structured data and unstructured data. A simple example is academic citations : while the citation's content is unstructured text, the citation is associated with structured data such as author list, categories and publication time. To query such hybrid data, a natural approach is to combine structured queries with keyword search. Two fundamental problems arise for this unique marriage : (1) How to evaluate hybrid queries efficiently? (2) How to model relevance ranking? The second problem is especially difficult, because all the foundations of relevance ranking in information retrieval are built on unstructured text and no structures are considered. We present context-sensitive ranking, a ranking framework that integrates structured queries and relevance ranking. The key insight is that structured queries provide expressive search contexts. The ranking model collects keyword statistics in the contexts and feeds them into conventional ranking formulas to compute ranking scores. The query evaluation challenge is the computation of keyword statistics at runtime, which involves expensive online aggregations. At the core of our solution to overcome the efficiency issue is an innovative reduction from computing keyword statistics to answering aggregation queries. Many statistics, such as document frequency, require aggregations over the data space returned by the structured query. This is analogous to analytical queries in OLAP applications, which involve a large number of aggregations. We leverage and extend the materialized view research in OLAP to deliver algorithms and data structures that evaluate context-sensitive ranking efficiently

Polymer membranes have been studied for several decades, and conventional utility of these membranes has been realized over a wide range of applications such as water purification, gas separation, and fuel cells. Separation performance of polymer membranes for these applications can be evaluated from different aspects including selectivity, material cost, and thermal and mechanical stabilities. This research focused on the design of crosslinked polymer membranes from reactive block copolymers, which could possess some attractive features such as controlled functionality and morphology. Novel block copolymers containing a chemically crosslinkable block and a chemically modifiable block were of particular interest. Metathesis reactions were employed to crosslink these copolymers with additional tough polymers such as polycyclooctene (PCOE) and polydicyclopentadiene (polyDCPD), to realize enhanced toughness in the resulting materials. First, a norbornene-functionalized polystyrene-polylactide (PNS-PLA) block copolymer was synthesized. A self-assembled blend containing this copolymer and DCPD was cured by metathesis reactions, and nanoporous monoliths with a cylindrical morphology were successfully produced after removing the PLA component. These monoliths exhibited pronounced mechanical and thermal stabilities superior to nanoporous polystyrene (PS). Second, polymerization induced phase separation (PIPS) during the ring-opening metathesis polymerization (ROMP) of DCPD in the presence of the PNS-PLA copolymer rendered continuous PLA nanodomains in a crosslinked PNS/polyDCPD matrix. Upon etching the PLA component, the resultant nanoporous membranes exhibited well-defined percolated nanopores and good thermal and mechanical stabilities. Preliminary diffusion measurements demonstrated potential utility of such membranes in ultrafiltration. Third, crosslinked polymer electrolyte membranes (PEMs) were fabricated from a PNS-poly(n-propyl-p-styrenesulfonate) (PSSP) block polymer and COE/DCPD via the PIPS scheme followed by deprotection of the PSSP block. These PEMs possessed a bicontinuous morphology and mechanical and thermal robustness. Select PEMs exhibited high proton conductivity similar to Nafion at high humidity and reduced methanol crossover. Tunable domain size and mechanical strength of the resulting PEMs are advantageous attributes of this preparation protocol. Additionally, the use of such PEMs for NH3 separation from gas mixtures was demonstrated. The appendix chapter represented our efforts to produce amine-functionalized polymer membranes from PNS-poly(dimethylaminoethylmethacrylate) (PNS-PDMAEMA) and COE by metathesis reactions, potentially useful for CO2 separation from gas mixtures.

There are two main objectives in my research work. The first objective is toinvestigate the adsorption behavior of various gases on single walled carbon nanotubes. This is accomplishedby using the classical molecular simulation methods. Our simulation work has providedmolecular level interpretation of some interesting phenomena observed experimentally by our collaborators.The second objective is to study the catalytic properties of metal/metal carbide surfaces and interfacialphenomena by using ab initio density functional theory.We have studied the adsorption of various gases on carbon nanotubes byusing classical molecular simulation and optimization techniques. We specifically haveinvestigated the displacement of adsorption on different adsorption sites.The systems investigated includeCO2 on SWNT, Xe/CF4 on SWNT and CO2/Xe on SWNT. Our simulations indicate that CO2 is easily replaced fromthe endohedral and interstitial sites of SWNT bundles by Xe,while the groove/external surface sites loose much less CO2. These calculations agreevery well with the experimental observations.We have also observed unique one dimensional behavior of gases adsorbed on carbon nanotubes by using optimization andparallel tempering Monte Carlo. The results show that CO2 molecules adsorbed in thegroove sites of single walled carbon nanotubes display behavior thatis quasi-1-dimensional. At finite coverages of CO2 in groovesclusters containing only odd numbers of molecules are formed at lowtemperatures. Even numbers of molecules form two clusters, eachcontaining an odd number of molecules.We have carried out density functional theory studies on the catalyticproperties of metal surfaces. We investigated adsorption of CO on the Ag(110)surface and CO adsorption, diffusion and dissociation on the W(111) surface.The CO molecule isfound to non-dissociatively adsorb in end-on configurations (alphastates) and dissociatively absorb in inclined configurations (betastates). The dissociation of beta state CO is found to have anactivation energy of about 0.8 eV, which is lower than the energyrequired to desorb CO molecularly from the surface.We have also studied the tungsten difussion mechanisms in cobalt. Our calculationsindicate that the diffusion is vacancy mediated. Therefore, we proposed the triangleand quadrangle mechanims, and examined the full diffusion pathways.

Increasingly, a number of applications rely on, or can potentially benefit from, analysis and monitoring of data streams. Moreover, many of these applications involve high volume data streams and require real-time and distributed processing of data arising from a distributed set of sources. We believe that a grid environment is well suited for exible and adaptive analysis of these streams. Thus, our research focuses on designing and evaluating a middleware to support the processing of these streams. Our system is referred to as GATES (Grid-based AdapTive Execution on Streams). It exibly achieves the best accuracy that is possible while maintaining the real-time constraint on the analysis. We have developed self-adaptation algorithms for this purpose. Further, we have addressed the problem of resource allocation in the GATES system. We design a static resource allocation algorithm to generate sub-optimal allocations and the experimental evaluation shows these allocations are very close to the optimal one. We also develop an infrastructure to support resource monitoring and dynamic resource allocation. Moreover, we implement ecient and low-cost dynamic migration for pipelined applications in a grid environment by using the notion of Light-weight Summary Structure (LSS). We have studied various data stream applications with support of GATES to intensively evaluate our middleware. The experimental results have demonstrated the expected advantages of utilizing GATES.

This dissertation addresses the use of driver model uncertainty for vehicle steering assist controller design, and uses road departure prevention as an application scenario. A robust serial steering assist controller is designed and tested to assess whether lane keeping performance is improved over the cases without a steering assist controller and with a nominal (non-robust) steering assist controller. A system identification approach to compute both the driver model and model uncertainty from driving simulator data is investigated. Both the structured and unstructured uncertainties are computed. The driver model uncertainty is shown to be useful for robust safety systems design. The uncertainty model can represent the uncertainty within one driver over time, or the uncertainty across different drivers, depending on what situation the system is designed for. Based on the computed driver structured uncertainty, a serial robust steering assist controller is designed. The goal of the controller is to avoid the problem associated with the degraded driver performance observed from the identification results. The serial controller is effective in avoiding the low performance situations, and robust stability is achieved. However, due to the large driver model uncertainty involved, robust performance cannot be guaranteed. An adaptive controller is presented as an approach to reduce uncertainty as has been validated in computer simulations. A PC-based driving simulator is used to conduct human-in-the-loop experiments to evaluate the proposed controllers. Two types of experiments are conducted. A long driving task with nominal configurations is designed to evaluate the change in driving behavior over time. A short driving task with artificially large lateral position error is designed so that the driver's regulation performance can be assessed. The experimental results indicate that statistically the robust controller introduces lane-keeping improvements during the short driving experiments. The improvements during the long driving experiments are less evident due to driver adaptation. However, the long driving tests do indicate benefits in terms of ease of driving. A nominal (non-robust) controller that is designed without consideration of driver model uncertainty is also tested. It is found that the nominal controller is characterized by high-gain and should be avoided. The importance of investigating driver uncertainty is therefore justified.

We have bombarded Cu and Al surfaces with O2+ ions and measured the relative yields of secondary ions in the energy range from 50 to 500 eV. We have determined both the relative yield as a function of incident ion energy and the kinetic energy distributions of the ejected ions for selected incident ion energies. In addition to looking at Cu+ ions from Cu and Al+ ions from Al, we have investigated ion signals from alkali impurities in the targets. For the Cu surface, ions ejected by Ar+ bombardment were examined both before and after sputter cleaning of the surface. Data on beam energy dependent secondary ion yields from the literature and this investigation have been fit by an exponential formula Acdotexp(-B/(C + E)), where A, B and C are fitting parameters. The results, with standard deviations for fitting parameters, are reported. A distinct plateau structure has been found for beam energy dependent yields of Na+ and K+ ions sputtered from untreated (long existing) Cu surfaces. A series of well controlled experiments indicate that this structure is caused by a surface excess of sodium and potassium at the surfaces of copper samples. The calculation of beam energy dependent energy depositi on shows that the energy deposited on the top surface of a polycrystal copper target by oxygen ions (O2+) does not further increase with the increase of beam energy above 250 eV. This result very well explains the beam energy dependence of the yield of secondary alkali ions contributed from the surface excess. Combined with the experimental data, thermodynamic calculations indicate that the positive surface excess of alkali metals near Cu surfaces is due to their segregation near the copper surface. Thus, the data for secondary ion yield vs. beam energy at low energy may provide a very surface sensitive probe with high spatial resolution (monolayer) for investigating segregation near solid material surfaces Compared with secondary ion signals produced from Cu surfaces by O2+ ion bombardment, the signals produced by Ar+ bombardment were found to be much weaker, and the secondary ion yield vs. beam energy curve had a slightly different shape