116964 research outputs found
Sort by
On the boundaries of STEM makerspaces
Full text linkMakerspaces are an increasingly popular venue for informal, opt-in, STEM educational experiences, and many have lauded their potential to increase student engagement within STEM (e.g., Martin, 2015; Roldan et al., 2018). However, STEM is a domain in which non-dominant populations have been repeatedly denied equitable experiences, and makerspaces may be another STEM space that recreates and reinforces this marginalizing environment. Scholars have critiqued the modern movement as a white, male, middle-class pursuit and have warned against an uncritical adoption of the narrow, STEM-oriented, techno-centric framing of making activities (e.g., Vossoughi et al., 2016; Worsley & Bar-El, 2020). Following these critiques, I investigate perceptions of making and makerspaces amongst undergraduate students who do not visit the space. Specifically, I explore why some STEM students choose not to visit makerspaces by asking: 1. How do students describe the visual design of a STEM makerspace in relation to their interest in visiting the space? 2. What repertoires of practice do students see as valid within a STEM makerspace? 3. How do students decide whether to cross the boundary into a STEM makerspace? To answer these questions, I interviewed students on the boundaries of a STEM makerspace about their perceptions of making, view of the makerspace, and their interest in visiting the facility. I found evidence supporting the critiques of the modern making movement throughout my dataset. Most students have a narrow definition of making that is restricted to creating something physical. Students differentiate between practices they see as “more technical” and more “artsy craftsy,” positioning the latter as lesser than or not appropriate within the bounds of a STEM makerspaces. This perception that certain practices are not validated in the makerspace acts as a barrier towards participation for many students, regardless of if they have prior experiences with making practices. Even if students do engage with the makerspace, they can face marginalizing experiences there that influence their interest in further participation and mediated the forms in which they were willing to engage with other makerspace participants. Implications for expanding and validating forms of making practices not currently valued in STEM spaces are discussed.Science, Technology, Engineering, and Mathematics Educatio
Advanced control of roll-to-roll mechanical dry transfer
Full text linkRoll-to-roll (R2R) manufacturing enables continuous, high-throughput processing of flexible substrates used in technologies such as printed electronics, energy storage devices, and two-dimensional (2D) materials. Within this framework, R2R mechanical dry transfer is an efficient, eco-friendly transfer method, where materials are continuously peeled and transferred between webs without the need for solvents or other liquids. Despite its advantages, this process presents substantial control challenges due to the strongly nonlinear dynamics at the peeling interface, sudden variations in adhesion energy, and abrupt transitions in material properties—especially when dealing with patterned materials or devices. Achieving uniform and defect-free transfer requires precise regulation of web tension and peeling geometry, as even small deviations can significantly degrade transfer quality.
Existing control approaches for R2R mechanical dry transfer rely on simplified linearized models that overlook critical nonlinear effects, parameter uncertainties, and actuator constraints. Consequently, these methods often fail to maintain robust performance under varying operating conditions or when processing patterned substrates, where adhesion and tension dynamics can change abruptly. The lack of control frameworks capable of systematically handling these complexities limits both the precision and scalability of R2R mechanical dry transfer, representing a major obstacle to its adoption in high-yield industrial manufacturing environments that demand consistent material quality and process reliability.
This dissertation develops a unified control framework to address the nonlinear, uncertain, and constrained nature of the R2R mechanical dry transfer process. The proposed methods include: 1) Full-state feedback control using linear differential inclusion (LDI) modeling to manage nonlinearities and uncertainties, 2) switched-system control for patterned materials using an uncertain almost periodic piecewise linear system (APPLS) framework to capture abrupt changes in adhesion and system dynamics, 3) a nonlinear model predictive control (MPC) strategy that incorporates input constraints and peeling pattern predictions for real-time control, 4) a repetitive learning MPC (RLMPC) strategy that leverages the repetitive nature of some peeling processes to match the performance of the full nonlinear MPC over successive cycles while substantially reducing computational cost, and 5) an iterative learning control (ILC) strategy for sequential transfers of identical samples, achieving MPC-level performance with minimal online computation. Collectively, these methods establish a scalable foundation for high-precision control of R2R mechanical dry transfer, substantially enhancing process stability, repeatability, and material quality.Mechanical Engineerin
Improvement and applications of field seismic SASW testing and investigation of the influence of water content changes on small-strain dynamic rock properties using laboratory free-free testing
Full text linkThe importance of seismic field and dynamic laboratory measurements in geotechnical engineering is undeniable. The dynamic parameters determined from these testing methods are used as key parameters in earthquake analyses and as essential data for characterizing the stiffnesses of geotechnical sites. Therefore, to enhance our understanding of geotechnical sites, improving the measurement techniques and increasing the applications in both seismic field testing and dynamic laboratory testing are very important. The Spectral-Analysis-of-Surface-Waves (SASW) method is a field testing method involving surface waves. This method is cost effective compared to other methods due to the nonintrusive and nondestructive characteristics of the method. Traditionally, SASW testing has been performed in only one direction to determine Vs profiles at sites. However, the measurement can be improved by performing an additional set of tests in the reverse direction at the same site with only a small increase in cost. The first part of this research introduces and discusses the benefits of the improved SASW method, using a case study performed on the crest of a dam. The SASW testing has accomplished very powerfully also cost-effective works at various geotechnical sites. For example, the SASW testing was performed to identify a vulnerable location for dam repair, estimate the burial depth in landfills, provide basic parameters for earthquake modeling. Furthermore, in this second part of the research, the SASW testing was performed on the differently compacted embankment as one of the application examples. The SASW testing was performed to evaluate the compacted embankment and to understand the natural ground of the site, and testing results at this site are presented and discussed. The last part of this research focuses on the influence of changes in water content on small-strain dynamic rock properties using free-free testing, which has not been effectively considered in the previous testing. Changes in the material damping ratios and Poisson’s ratio values due to the changes in the water content are mainly discussed. The water influence is a significant factor in small-strain dynamic rock properties measurements. The influence of the water was studied using free-free testing on the 23 different rock samples.Civil, Architectural, and Environmental Engineerin
Chemical solutions to concrete durability problems
Full text linkGiven recent changes in energy production that have significantly reduced the availability of fly ash, there is an increasing need for alternative materials that can provide similar levels of durability performance as fly ash. The objective of the project was to evaluate the use of commercially available chemical admixture products in improving concrete durability. Numerous products were tested across the major durability issues including corrosion of reinforcing steel, classical sulfate attack, delayed ettringite formation, and alkali-silica reaction. Testing parameters investigated included early-age hydration, compressive strength, expansion, electrical resistivity, corrosion potential, chloride diffusivity, and surface sorptivity. Lab samples included accelerated forms of testing to assess performance more quickly, while corresponding field samples were also cast to provide valuable long-term data and to correlate results with lab samples. This study found that some of the chemical admixtures evaluated improved the durability of concrete, while others were found to have insignificant impacts -- none of the products tested were able to achieve all the benefits imparted by Class F fly ash.Civil, Architectural, and Environmental Engineerin
Tuning electromigration forces on adsorbates on graphene
Full text linkElectromigration, the movement of atoms induced by electrical currents, is a phenomenon of significant importance in various nanoscale applications. By altering the electronic structure of current carrying membranes, fine-tuning and optimization of electromigration forces on adsorbates becomes possible, facilitating migration of adsorbates on membrane surfaces. This property finds diverse applications in nanoscale systems, including mass transport and ionic current generation. This study explores the adsorption characteristics of NH₃ and H₂S molecules on graphene surfaces and the accompanying electromigration forces. Using Density Functional Theory (DFT) and Non-equilibrium Green’s function (NEGF) calculations, we demonstrate that under applied gate voltages, a distinct threshold behavior in electromigration forces emerges, dominated by electron wind forces. The findings elucidate the critical role of orbital energy alignment and electron scattering forces in governing electromigration, offering insights into the design and optimization of nanoelectronic devices.Mechanical Engineerin
Towards human-centered generative design : cross-modal synthesis for three-dimensional design concept generation
No full textThere has been a fast-growing interest in generative design (GD) — an artificial intelligence (AI) — based approach for early-stage design that automates the creation and optimization of various design concepts. GD can quickly generate numerous design variants, allowing for efficient exploration of design space, and thus can shorten the product design cycle and reduce development costs. Despite advances in GD technologies, current GD approaches place AI at the center of the design process and lack direct involvement of human preference and judgment in the concept generation process. This raises significant concerns, as current GD methods lack mechanisms that incorporate human factors, such as aesthetic preferences, and the safety and comfort of designs. This omission could lead to the creation of design concepts that fail to meet human needs effectively. In contrast, human designers bring their expertise and knowledge to generate human-centered designs, playing an irreplaceable role in the design process. This is especially true in the early stages, where human domain knowledge and preferences are crucial in determining the potential to produce user-centered and user-friendly products. These elements critically influence the success of product design and development. To fill this research gap, the overarching goal is to realize human-centered generational design. Towards that goal, the research objective of my dissertation is to develop a human-supervised data-driven generative design framework that can actively keep humans in the loop, as well as in charge of the generation and evaluation of AI-assisted design concepts in early-stage design. In particular, my research is motivated to answer the following central research question: In what ways and to what extent can human designers’ intent and preferences be incorporated as input to actively interact and guide the GD process to improve the quality and relevance of the design outcomes? The central research hypothesis is that human designers' intent and preferences can be incorporated as input to guide the data-driven design generation using cross-modal synthesis. Cross-modal synthesis is a machine learning technique that can generate data in one modality (such as images) based on another input modality. Our rationale is that human designers can use different design modalities (such as sketches of car models) or a combination of them to represent their intent and preferences (e.g., certain curvatures in a car's exterior design), and cross-modal synthesis methods can automatically transform them to desired design modalities, such as 3D CAD models. However, the development of cross-modal synthesis methods for engineering design needs to address several fundamental challenges. These challenges include the scarcity of design data, complexities in 3D design representations, large semantic gaps between different modalities (challenging to maintain design integrity and intent embedded in the input design modality), and vectorized design representations for AI training. To test the central hypothesis, we aim to answer three Research Questions (RQs). (1) How feasible and to what extent can cross-modal synthesis methods with unimodal input incorporate human designers' intent and preferences as input to guide the data-driven design generation? (2) How feasible and to what extent can cross-modal synthesis methods with multimodal inputs incorporate human designers' intent and preferences as input to guide the data-driven design generation? (3) What are the effects of different representations of the generated designs on the data-driven design evaluation? To address these RQs, we explore various methodologies. For RQ 1, we conduct a systematic review of deep learning methods for cross-modal tasks, from which we identify technology on how to develop cross-modal synthesis methods for engineering design. We then develop a novel neural network architecture, target embedding variational autoencoders, based on which we create two cross-modal synthesis methods with unimodal input for the task of (1) silhouette contour sketch to 3D mesh and (2) image to CAD sequence. In response to RQ 2, we propose a multimodal CAD dataset to enable and evaluate large language models' ability to generate CAD models from multimodal inputs (i.e., textual descriptions, sketches, and images). Lastly, RQ 3 is answered by developing a data-driven structure-aware generative design and evaluation approach and examining vectorized design representations to improve the assessment of generated designs. From the results, we conclude that: (1) Cross-modal synthesis methods, capable of processing either unimodal or multimodal inputs, exhibit significant potential in capturing and integrating human designers' intent and preferences to guide the generation of early-stage design concepts in 3D representations. Although textual descriptions, sketches, and images may not fully encapsulate the designers' envisioned ideas—a challenge also prevalent in traditional design practices—our cross-modal synthesis approaches can still discern and interpret the underlying design preferences from these varied input modalities. Consequently, these methods can generate 3D designs that are closely aligned with the specified design requirements embedded in the input design modalities, thus bridging the gap between conceptual intent and tangible design artifacts. (2) The choice of mathematical design representations (e.g., vectors of design features) significantly influences the evaluation of generated designs in design performance prediction. Such influences are particularly significant when product geometries become more intricate and when dealing with systems design generation where complex interdependent relations between components exist. Although latent spaces are commonly utilized for these vectorized representations to accelerate AI-assisted design evaluation and optimization, such latent vectors may prove unsuitable if any information irrelevant to the engineering performance of interest is encapsulated during the formulation of design representations, consciously or unconsciously. This observation underscores the imperative for designers to consider the suitability of vectorized design representations for evaluation purposes from the beginning of developing DGD methodologies. In summary, this dissertation represents a critical step forward in human-centered generative design. It advances the design field by addressing a crucial gap in existing GD methodologies, facilitating a human-centered GD approach. Specifically, we introduce a novel Human-Supervised Data-Driven Generative Design Framework with cross-modal synthesis methods for design generation and AI-assisted design evaluation methods, which enhance human control and interaction within the GD process. Notably, this work develops an innovative neural network architecture tailored for cross-modal synthesis in engineering design. This architecture is particularly adept at incorporating human intent and preferences into the GD of 3D design concepts. The proposed methodologies can significantly accelerate design ideation, enhance the exploration of design spaces, and incorporate downstream considerations into early-stage decision-making. The methodologies are domain-independent and can be employed across different products in industries to expedite the product development cycle and decrease associated costs. Furthermore, the methods have the potential to be translated into pedagogical tools in design education, preparing next-generation engineers for future careers in an evolving landscape that increasingly values human-AI collaboration in engineering design.Mechanical Engineerin
Convex optimization for covariance steering and density control
No full textIn many real-world applications of control theory and trajectory optimization, one must design control algorithms which can account for the presence of uncertainties due to random disturbances acting upon the system (e.g., winds), modelling errors (higher order dynamics which are hard to model) and/or unknown parameters.
Accounting for the effect, or even controlling, these uncertainties is crucial to ensure safety and meet performance specifications in many applications.
One class of the stochastic control problems known as the `covariance steering/control (CS) problem', which is studied in this dissertation, specifically deals with the control of uncertainty.
In particular, the main objective of the covariance steering problem is to find a causal control policy for a given uncertain dynamical system which will steer the first two moments of the given initial state distribution to their desired values in infinite/finite time while minimizing a relevant cost function.
This dissertation aims to develop efficient techniques for solving optimal covariance steering (CS) problems using convex optimization and to extend CS theory to address more general density control problems.
The CS problem with soft terminal constraints and squared Wasserstein distance is tackled by formulating it as a difference of convex (DC) functions program.
A sufficient condition for the convexity of this problem is also provided.
The CS problem with Wasserstein terminal cost is formulated as a convex semi-definite program (SDP) by employing randomized state feedback policies.
It is demonstrated that the optimal policy for this problem corresponds to a deterministic state feedback policy.
Additionally, affine disturbance feedback policies are used to formulate the CS problems as convex optimization problems.
For computational efficiency, truncated policies are introduced, and theoretical bounds on their sub-optimality are provided.
The dissertation then focuses on CS problems for linear systems subjected to state and input-dependent noise processes. The terminal covariance assignment constraints are initially relaxed into semi-definite constraints.
The solution to the relaxed problem is subsequently used to address the exact problem, demonstrating that this two-step procedure provides a feasible solution as long as the relaxed problem remains feasible.
Lastly, the CS techniques are applied to solve more general density steering problems using Gaussian Mixture Models (GMM).
By utilizing the closed-form expression for the covariance steering cost between the Gaussian marginals, the GMM density steering problem is reduced to a linear program for the hard-constrained case and a bilinear program for the soft-constrained case.
The propagation of errors due to GMM approximation of general densities is analyzed, and the boundedness of the terminal state distribution approximation error is also demonstrated.Aerospace Engineerin
Doctoral thesis recital (piano (chamber))
No full text4 unidentified works.MusicName of supervisor not provided
Discovery, characterization, and application of novel MELK inhibitors to investigate the role of MELK in triple negative breast cancer
Full text linkMaternal embryonic leucine zipper kinase (MELK) is a redox and cell-cycle regulated kinase, evolutionarily conserved in eukaryotes from nematodes to humans. It is a member of the adenosine monophosphate kinase-related kinase (AMPK-RK) family, which mediate cell survival under stressful metabolic conditions. While MELK is expressed in the early stages of murine embryonic development, it is mainly restricted to non-vital organs, such as the testes, ovary, and thymus in adult tissues. MELK has received attention because its mRNA and protein levels are elevated in a wide array of cancer cell types and clinical tumor samples, and its expression correlates with poor prognosis in aggressive cancers, such as glioblastoma multiforme (GBM) and triple negative breast cancer (TNBC). MELK has also been functionally linked to multiple oncologically relevant cellular processes, including migration and invasion, the DNA damage response (DDR), maintenance of cancer stem cell multipotency, and resistance to radiation. Consequently, MELK has become an attractive target for drug development in cancers that have historically lacked molecularly targeted therapy options. However, the precise molecular mechanism behind MELK’s role in cancer remains to be delineated. In this work, we first review what is known regarding the structure, regulation, and biological function of MELK, and survey reported efforts to develop MELK inhibitors and their potential applications in cancers. We then describe our own MELK inhibitor development project, beginning with the identification of an indolinone core molecule, nintedanib, as a lead compound through a kinase inhibitor cross-screening. Further elaboration of the scaffold produces derivatives with subnanomolar potency towards MELK and preliminary selectivity against original tyrosine kinase targets of nintedanib. We then investigate the kinetic mechanism behind derivative potency, revealing that binding to MELK is limited by a slow conformational change that increases drug residence time and requires the ATP binding pocket residue Glu-93. Finally, we identify candidate MELK substrates in the context of transforming growth factor-β (TGF-β) signaling. Given the lack of sufficiently specific experimental tools available for validation, we develop and utilize the analog-sensitive (AS) kinase approach, wherein we engineer a gatekeeper mutant of MELK capable of accepting bulky ATP-analog inhibitors. Using the AS system, we show that expression of the candidate substrate semaphorin 4B is downstream of MELK in MDA-MB-231 cells, but is not phosphorylated by MELK directly in vitro. Taken together, the small molecules and AS MELK experimental system developed in this work will facilitate further understanding of MELK’s role within the context of both normal and malignant mammalian cells.Pharmaceutical Science
The comparative effects of narrative and factual information on children’s generosity
Full text linkIn the real world and the laboratory, narrative has been shown to have an extraordinary power to influence beliefs and behavior. Though stories have been shown to inspire children to act prosocially in certain circumstances, little is known about how stories affect children’s desire to give away resources to others. In the present study, 58 ten-year-olds from the Austin area were presented with either a narrative or factual information about Nepalese orphans, and were given the opportunity to donate compensation money to an organization that helps Nepalese orphans. Results showed that, although there were no significant differences in donation amount between condition, participants who were more transported into the world of the orphan (whether described through facts or narrative) donated significantly more than others, though the effect of transportation on donation was marginally stronger for those who heard only facts. Implications and directions for further research are discussed.Psycholog