Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Using ZIF-8 to Protect Biomaterials and Develop Patient-friendly Modalities of Vaccine Delivery
No full textVaccines are a two-hundred-year-old technology that, despite new preparations and novel
formulations, still require stable low-temperature storage conditions and delivery via needle and
syringe. Aversion to needles is one of the most frequently cited issues for poor compliance with
vaccines. Yet, developing alternatives is complicated because vaccines are made of delicate
biomolecules, including the recently popularized lipid-based nanoparticles. Finding a way to make
these materials robust enough for alternative forms of delivery has numerous advantages. Needle-
free vaccines are a promising field of exploration owing to ease of administration; needle-based
vaccines require technical skills that cannot be transferred to the general population or volunteers
quickly. This is especially important in highly populated areas, where there are abysmally low
numbers of healthcare personnel per capita population. Disposable needles result in biohazardous
waste, and their accidental/intentional reuse spreads blood-borne pathogens worldwide.
Attempting to mitigate these challenges makes investigating alternate vaccine administration
methods an enticing research prospect.
One way to make these vaccine-based biomaterials more robust is to protect them inside an
“armor” made of material that can withstand the stresses associated with transportation and
delivery. Zeolitic imidazolate framework-8 (ZIF-8) is one of the most well-studied metal-organic
frameworks and has lately gained popularity in the field of nanomedicine. In this body of work , I
have investigated how we can use ZIF-8 as a coating for biomaterial protection, providing a slow-
release platform, and making these vaccine formulations shelf stable. I particularly focus on how
these ZIF-8-protected biomaterials can be exploited for needle-free and patient-compliant forms
of vaccine delivery—specifically intranasal and ballistic delivery. We have explored protecting
lipid-based nanoparticles inside ZIF-8, and demonstrated how they are mechanically protected
through ballistic delivery. We have also studied the biocompatibility of ZIF-8 when administered
intranasally and applied that knowledge to deliver novel tuberculosis antigens using ZIF-8
intranasally as a vaccine. To improve biocompatibility and enhance adjuvanting potential, we have
also developed a manganese-doped ZIF-8 system, which can synergistically work with
tuberculosis antigens to provide protection against tuberculosis
Digital Spill: Black Data and Meaning-making in New Social Media
No full textDigital practitioners from Black Twitter are trying new social media platforms in the wake of
Twitter’s platform collapse. In this thesis, I examine one of those platforms, SPILL, to
understand the interface, practices, and beliefs and perceptions of the app as it functions in its
introductory period. Toward that end, I utilize critical technocultural discourse analysis in this
thesis. I argue that SPILL has inconsistent messaging and non-user perceptions that mark it as
undesirable, which could spell its future demise. On the other hand, I argue SPILL is redeemable
in its aims for a cultivated, safe experience and a popular live audio-visual option. However, I
argue it is inconclusive whether SPILL can supplant the vital discursive publics of Black Twitter
and the SPILL team should be careful of its many pitfalls
Experimental Study of Riblet Treatments on Aerodynamic Performance and Energy Production of Wind Turbines
No full textWind energy continues to be a rapidly growing source of power generation among renewable
energy sources. To enhance the capacity of current wind farms, implementing various
modifications can significantly boost energy production. Riblets, a type of flow control
device, have been shown to effectively reduce drag in turbulent flows and have the potential
for improving turbine blade aerodynamic performance. This study presents a series of
measurements for load, flow dynamics and power generation, on the tested structure surfaces
with riblet treatment.
This first study explores the effect riblets play in the control of turbulent flow over an airfoil’s
surface. Understanding this phenomenon is crucial for increasing the flight efficiency of air
vehicles and enhancing the performance of wind turbines. Over recent decades, extensive
research has shown that engineered surfaces, such as riblets, can effectively alter the structure
of near-wall turbulence and reduce skin friction on a flat surface. This study aims to quantify
the effects of riblets on improving the aerodynamic performance of a DU-91-W2-250 airfoil
with a chord length of 0.4 m and a span of 0.8 m. By optimizing the riblet design based on
the tested Reynolds number and flow statistics over the suction side of the airfoil, the results
demonstrate that riblet surface treatment can effectively reduce drag across a wide range of
angles of attack and enhance the lift coefficient. Additional measurements of flow statistics
over the riblet surface using particle image velocimetry revealed that these microstructure
can reduce flow momentum loss near the trailing edge of the airfoil.
Quantifying the direct effects of riblets on wind turbines remains an area needing further
study, as few academic resources currently address this topic. This study aims to quantify the
impact of riblets on power generation when a turbine rotor is treated with a riblet coating.
A film treated with riblets was applied to the rotor of the G06 turbine, and power generation
measurements were taken using the turbine’s torque sensors. The riblets were applied to the
outer 50% of the rotor blades, as this section produces 75% of the torque acting on the rotor
hub. Two applications were tested: riblets on both sides of the rotor blade and riblets on
the suction side only. Tip speed ratio and airspeed velocity at the hub were the variables of
interest to quantify power generation. The study found that applying riblets on both sides
of the rotor blade resulted in up to a 2.32% increase in power generation, with a constant
hub speed velocity. When riblets were applied only to the suction side, a 2.05% improvement
was observed. This study highlights the positive impact of riblets on improving wind turbine
power generation and demonstrates an effective technique for testing riblets on scale wind
turbines.
The third study in this thesis takes a closer investigation at the boundary layer of the riblet
coating. Understanding the underlying mechanisms that govern the usefulness of riblets
leads to helpful insights as to the nature of their efficacy. This study seeks to investigate
the influence bladed riblet geometry play on the turbulent boundary layer’s large-scale flows
and near wall flows. The turbulent boundary layer dynamics and skin friction reduction over
the blade riblet surface was systematically examined in a water channel across various non-
dimensional riblet spacings. Shear velocity was deduced using the streamwise velocity gradient
from the logarithmic layer via planar Particle Image Velocimetry (PIV) measurements, while
the near-wall flow characteristics were quantified by a Micro Particle Image Velocimetry
(micro-PIV) system. The results indicated that the riblet surface exhibited improved drag
reduction performance at low non-dimensional riblet spacings. Inspection of near-wall flow
statistics demonstrated that at the optimal non-dimensional riblet spacing, the blade riblet
can effectively mitigate both viscous and turbulent shear stress terms, and therefore reduce
the total shear stress and friction drag. This experimental result demonstrated that the
modification of riblets onto a surface within a turbulent regime have the potential to reduce
drag at a specific flow speed range
Investigation of Source Extension Methods, the Discrepancy Algorithm, and Noise Estimation to Overcome Cycle Skipping in Full Waveform Inversion
No full textFull waveform inversion (FWI) is a geophysical technique used to create highly detailed mod-
els of the Earth’s subsurface which can be used to explore for hydrocarbons and to predict
natural hazards such as earthquakes. Seismic waves are generated from controlled sources
such as vibroseis trucks on land and airguns in marine environments. These waves propagate
through the Earth’s subsurface materials, reflecting off of interfaces underground. The for-
ward problem in FWI predicts the data based on a given model of the subsurface, while the
inverse problem estimates subsurface parameters, such as sound velocity, by minimizing the
difference between recorded data and data we predict from solving our mathematical model
(the wave equation). Solving this minimization problem is computationally prohibitive, so
we rely on local gradient-based optimization methods. The success of such methods depends
on the accuracy of the initial guess. Otherwise, FWI tends to get stuck at a suboptimal
solution, a problem known as cycle-skipping.
This dissertation explores several source extension methods to overcome the cycle-skipping
problem in FWI for transmitted data. These methods add additional degrees of freedom to
the objective function, expanding the solution space to include models which may or may not
be physical. This updated objective function is convex with appropriate penalty parameters,
allowing local gradient-based optimization methods to find the correct geological model from
a wider range of initial models. When close to the correct model, the physical constraints
are reimposed in the problem by the penalty term. For a simple homogeneous medium
experiment with single trace acoustic data, we illustrate how extended source inversion
(ESI) avoids cycle skipping by relaxing the requirement that the source must be compactly
supported and by adding a soft penalty to control the extent of the source. The discrepancy
algorithm dynamically adjusts the penalty weight to maintain data error within a specified
range, ensuring accurate model estimates. The update of the penalty parameter relies on
having an accurate estimate of the noise level in the data which is generally unknown a priori.
Numerical examples show that the extended method successfully overcomes cycle-skipping
without the need for a good initial model, that the algorithm can dynamically update the
noise level in the data (and hence the penalty parameter), leading to a reliable and accurate
solution to the inverse problem.
Additionally, this dissertation investigates the matched source waveform inversion (MSWI)
method which extends the solution space by assuming that each data trace is a function of
both receiver and source location. For single arrival data, MSWI is closely related to travel-
time inversion. The MSWI objective function includes a data misfit term and a penalty term
to keep an adaptive filter close to the Dirac delta function. MSWI is equivalent to the source
extension method described above when the penalty parameter in MSWI approaches zero.
Experiments demonstrate that for more complex heterogeneous media experiments MSWI
successfully reduces cycle-skipping in single-arrival transmission data (even when moderate
amounts of noise are present in the data), while FWI often fails due to cycle skipping. The
inverted model resulting from MSWI can, therefore, provide a good starting model for FWI.
However, the results also demonstrate that MSWI applied to multi arrival transmission data
fails
The Work of New Motherhood: Technocapitalism and Postpartum Labor
No full textThis dissertation examines how technocapitalist discourses and neoliberal ideology manifest ideas
of a contemporary prototypical motherhood that aims to turn mother into workers of a certain kind
inside and outside the family. I trace the circulation of these discourses in dominant narrative of
CEOs, political leaders, innovators, venture capital investors, users of digital networks, and
promotional materials of tech startups that offer technological solutions for new mothers. In this
work, prototypical motherhood refers to a standardized motherhood ideal that demarcates moral
limitations, corporate practices, and political actions of what is acceptable, possible and imagine
for mothers in relation to work. This research is centered around postpartum. The technologies
explored here are directed to women at this stage of motherhood. I undertake this analysis trough
discourse analysis of emergent technologies directed to new mothers, such as Work & Mother and
HearthDisplay, major virtual clinics that offer maternal health for new mothers via employer’s
benefits, like Maven Clinic, and discourse analysis of online exchanges in online networks that
exchange breastmilk in different forms (donation, trade, or sell), such as Facebook Groups, the site
OnlyTheBreast.com, and Facebook Market. Alongside the analysis on prototypical motherhood, I
examine neoliberal logics behind this prototype as it relates to labor, both productive and
reproductive. In addition to interrogating how motherhood, labor, technologies and postpartum
work conventionally through technocapitalist discourse and neoliberalist ideology, this work also
examines examples that critique or provide alternatives to different versions of prototypical
motherhood
Novel Materials and Methods to Enhance Vaccine Potency and Pain-free Delivery
No full textVaccines employed over two centuries in combatting infectious diseases face challenges like cold
chain and needle administration, hindering patient compliance. Research actively investigates
ways to create thermally stable vaccine formulations and eliminate booster doses. Herein, to
overcome these challenges regarding global vaccinations, porous reticular frameworks,
particularly metal-organic frameworks (MOFs) and covalent-organic frameworks (COFs), have
emerged. These high thermally stable scaffolds protect delicate vaccine antigens within their
porous structure, shielding against high temperatures, enzymatic degradation, and other
mechanical stresses. Administered subcutaneously, these bio-composites slowly release antigens
into the body, creating a vaccine depot. This has been shown to improve the overall
immunogenicity of subunit antigens and whole-cell vaccines. Biolistic vaccine methods,
promising for easy administration and circumventing needle-related issues, are crucial in densely
populated areas with limited healthcare resources. Disposable needles contribute to biohazardous
waste, while their reuse poses a global threat. MOF-encapsulated vaccines offer a biolistic delivery
approach with improved and controlled release kinetics, addressing these concerns
Physical Sensing and Physics-based Machine Learning for Actionable Environmental Insights
No full textCurrent methods for the evaluation of water quality are limited by sparse reference measure-
ments and infrequent satellite observations. Meanwhile, air quality standards are assessed at
annual and 24-hour averages which neglect the impact of short-term spikes on local pollution
exposure. This dissertation develops physics-based machine learning methods to fill these
gaps. To significantly accelerate water quality assessment, we design an autonomous robotic
team combining drone-based hyperspectral imaging with collocated, in situ data collection by
an autonomous boat. Models are trained to map observed reflectance spectra into 13 physical,
chemical, ionic, and biochemical water quality parameters. These models are then deployed
to map the small-scale spatial variability of water quality across a North Texas pond. For
scenarios in which specific contaminants are not known in advance, we utilize unsupervised
machine learning to visualize the distribution of water-leaving reflectance spectra and identify
spectral signatures corresponding to unique sources. As a key innovation, we extend this
approach by introducing a novel machine learning method called Generative Simplex Mapping
for nonlinear spectral unmixing. Using real data from a rhodamine tracer dye release, we
demonstrate the ability of this model to successfully identify localized contaminant sources.
Finally, we leverage data from a distributed network of low-cost air quality monitors to
construct time series models for real time particulate matter measurements. The approach
extends the Hankel Alternative View of Koopman framework to identify acute pollution
spikes and enable multi-step forecasts
Speech Emotion Recognition in the Presence of Background Noise
No full textDeploying a speech emotion recognition (SER) system on real-world applications can be an
instrumental tool that benefits several areas, such as entertainment, health care, and human-
computer interaction. Although SER approaches have been recently advanced, using a state-
of-the-art SER system in a real-world environment is still challenging due to unconstrained
background noises. This dissertation aims to increase the noise-robustness of an SER system
by focusing on two aspects: 1) front-end processing and 2) domain adaptation.
For front-end processing, we investigate a set of noise-robust features and frames that do not
need an enhancement to increase SER performance under noisy conditions. We first define
the noise-robust features for SER, increasing the SER performance by using only the noise-
robust features. We expand this framework to selective feature enhancement, which only
applies enhancement to noise-sensitive features. We apply those feature selection and selec-
tive enhancement frameworks in the SER system based on an SSL representation by defining
noise-robust frames. From those studies, we demonstrate that keeping the noise-robust in-
formation of the input speech rather than enhancing all signals or features is important.
For the domain adaptation approach, we investigate the appropriate ways to adapt the
pre-trained SER system to noisy environments that can avoid the influence of the noisy
adaptation set. We propose the decoupled ladder network, which separates the emotion
and reconstruction embeddings of the ladder network to minimize the influence of noise
while adapting an SER model to noisy environments. We also propose the contrastive
teacher-student learning of SSL representation for adapting a pre-trained SER model to the
target noisy condition, preventing a catastrophic forgetting of the pre-trained knowledge.
We also investigate using skip connection adapters to adapt a transformer-based SER model
to multiple noisy environments without requiring too much adaptation time and stored
parameters. Lastly, we explore the environment representation extracted from the pre-
trained text encoder to deal with unseen noisy environments for SER. Our approaches show
better SER performance than simply training an SER model with noisy speech, implying the
importance of keeping pre-trained knowledge and fusing the target environment information
obtained with the large language model (LLM)
Exploring the Impact of the Agile Mindset on High-performance Organizations in Local Government Operations: a Qualitative Case Study
No full textIn this rapidly evolving post-digital era, local governments face increasing pressure to transform
their operational strategies and governance models. This dissertation explores innovative
paradigms, including High-Performance Organizations (HPO) frameworks and Agile
methodologies, to enhance community welfare and governmental responsiveness. Drawing
inspiration from the Agile Manifesto's principles and guided by a commitment to diversity,
equity, and social responsibility, this research seeks to redefine the essence of high performance
in modern governance. The study examines the successful components of high-performance
municipal government organizations, analyzing how they differ from traditional bureaucratic
structures. It also investigates the potential benefits of transitioning to Agile methodologies
within high-performance local government bodies. Utilizing qualitative case study methodology,
the research provides pragmatic guidance by analytically exploring the efficacy of contemporary
organizational models like HPOs and Agile in the local government setting. The significance of
this study lies in its recognition of local governments as nerve centers for community well-being,
where the delivery of public services profoundly influences citizens' quality of life. By
integrating emerging organizational frameworks, such as HPO and Agile, with inclusive and
purpose-driven initiatives, this dissertation offers local governments an evidence-based playbook
for achieving sustained excellence in an increasingly demanding societal landscape. The research
aspires to contribute a nuanced voice to the discourse on high performance in organizations,
addressing evolving expectations and challenges municipal governments face.
Keywords: Leadership characteristics, inclusive leadership, transparency, effective
communication, employee collaboration, teamwork, ethics, agile, and high-performance
organizations
Deep Neural Network Based Robust Speech Recognition for Sustained Diverse Non-native/native Speakers
No full textAutomatic speech recognition (ASR) systems have shown remarkable performance in recognizing
human speech, but often struggle to accurately recognize speech across various accents,
dialects, and certainly languages. This challenge is particularly prominent in English, which
comprises numerous alternate accents and dialects. Additionally, the rise in global communication
has led to an increased number of bilingual speakers with varying degrees of English
proficiency, resulting in accents that further complicate English ASR performance. The diminished
performance in accented speech recognition can be attributed to several factors: 1)
Sustained performance with a single multi-accent model: Training an ASR model with data
pooled from multiple accents often leads to an overall suboptimal performance, particularly
for low-resource accents. 2) Limited availability of accented training data: Accents, particularly
non-native ones, often lack sufficient transcribed speech data for training ASR models
since collecting enough data can be impractical due to the vast array of accents and limited
speaker populations within accents/dialects. 3) The continual incoming of train/test data:
As new accented train/test data becomes available, storing large-scale datasets and retraining
ASR models can be computationally inefficient and expensive. This dissertation aims to
develop advanced accented speech recognition systems, coupled with automatic accentedness
assessment by examining these challenges and proposing novel approaches.
First, by considering foreign-accented speech as an interpolation between the native language
(L1) and English (L2), we propose a multi-task learning (MTL) framework in which the
primary task is trained with native English, and the secondary task is trained with potential
seen L1 languages. The proposed MTL approach yields improved performance with +11.95%
and +17.55% relative character error rate (CER) gains over the baseline for Hispanic and
Indian accents, respectively. Next, we propose a novel student-teacher learning approach for
an advanced multi-accent ASR solution. In this approach, we utilize an ensemble of accentspecific
teacher models to guide the training of a single multi-accent model by constraining
the student model to mimic the output of the accent-specific models for their corresponding
training data. The resulting model achieves a +20.1% relative CER reduction compared to
a baseline model trained without teacher knowledge. We also advance continual learningbased
techniques for an effective domain expansion solution in ASR for two scenarios: 1)
where only new data is available, and 2) where prior and new data are both available. We
propose Soft KL-Divergence (SKLD)-Model Averaging (MA) and Gradient Averaging (GA)
for these scenarios, respectively. Experiments demonstrate improved recognition accuracy
and reduced computational costs compared to several state-of-the-art continual learning
approaches.
Finally, to enhance accent identification (AID) and assess non-native accentedness, we propose
a multi-embedding approach that utilizes pre-trained language identification (LID) and
speaker identification (SID) models, which effectively encode accent and dialect information.
We also explore techniques to leverage ASR and AID systems for accentedness estimation.
Evaluation results demonstrate a strong correlation between objective scores estimated by
the two systems. Additionally, a robust correlation between the objective accentedness scores
and subjective scores based on human perception is demonstrated, providing evidence for
the reliability and validity of AID-based and ASR-based accentedness assessment systems
in non-native speech. Taken collectively, these advancements contribute to new algorithmic
solutions that improve speech recognition for non-native speakers, while maintaining performance
for native speakers. In addition, advancements made for improved and sustained
ASR performance can also provide insight into accent assessment simultaneously