Masters Theses

Permanent URI for this collection

https://hdl.handle.net/10919/9291

Browse

Now showing 1 - 20 of 23729

A PCA-Based Framework Leveraging frequency-dependent Piezoelectric Impedance for Macro-Scale PUFs
Shah, Manav Bhavesh (Virginia Tech, 2026-04-16)
Piezoelectric sensors are widely employed in structural health monitoring of safety-critical systems including aerospace, industrial IoT, and military applications, because of their ability to transduce mechanical activity into electrical signals. These sensors relay sensitive information about potential damages to critical infrastructure, thereby motivating the need to embed the root of trust into the sensors themselves. This work proposes a secure ID generation framework leveraging the frequency-dependent impedance response of commercial off-the-shelf (COTS) piezoelectric sensors to create Physical Unclonable Functions (PUF). Manufacturing imperfections introduce stable, device-specific perturbations in the impedance characteristics. When analyzed across a sensor population using Principal Component Analysis (PCA) yield discriminative 64-bit IDs. To support experimental validation, a 70-sensor dataset was assembled spanning multiple temperature conditions and a multi-month data collection period, capturing realistic temporal drift and environmental variability. Experimental results demonstrate near-ideal uniqueness (50.32%) and a significant improvement in reliability compared to existing methods. This framework successfully demonstrates how widely deployed piezoelectric sensors can serve as practical, low-overhead security primitives without excessive hardware modification.
Development of a Benchmark Problem and Implementation of the DRF Methodology for Reactor Dosimetry in the Extended Beltine Region
Friedman, Cole Nathan (Virginia Tech, 2026-04-15)
Recent interest in extending the operating life of nuclear power reactor throughout the United States beyond 60 years of operation is prompting innovation in dosimeter response solving methods to aid in understanding material embrittlement of nuclear reactor pressure vessels. This innovation is possible due to improved computational hardware and simulation techniques which allow analysis of 3-D response calculations for geometry beyond what is possible for conventional methods used for lifetime extension of younger reactors. There is a need to create a benchmark problem with a high-fidelity reference solution to act as a point of comparison for the new dosimeter response methods. The Tennessee Valley Authority (TVA) Watts Bar Unit 1 Reactor Pressure Vessel Fluence Benchmark is being developed to fulfill this need. This thesis presents the development of the TVA Watts Bar Unit 1 Reactor Pressure Vessel Fluence Benchmark. Further, the Detector Response Function (DRF) Method is used to solve part of the problem presented in the benchmark. This solution is treated as a test to lay the groundwork for utilizing the DRF method to find a full solution for the TVA Watts Bar Unit 1 Reactor Pressure Vessel Fluence Benchmark problem in the future. Data is extracted throughout the presented methodology and analyzed for physical accuracy. The DRF method is chosen for this work as a method with low computation times, and the results from are compared to a reference calculation. Improving the reaction rate calculation methods will ultimately reduce the computation time required for safety analysis of nuclear power plant lifetime extensions, saving institutions and companies both time and money.
The Influence of Weather on Reproductive Behavior and Population Trends of Four-toed Salamanders (Hemidactylium scutatum)
Ferguson, Kalin Jo (Virginia Tech, 2026-04-10)
Environmental conditions influence amphibian reproduction, behavior, and development. Understanding these relationships is critical for developing effective management strategies to conserve populations and their habitats. However, knowledge gaps persist for many species, limiting our ability to predict responses to increasingly variable seasonal conditions driven by climate change. The first objective of this study was to further explain nest site fidelity in female four-toed salamanders (Hemidactylium scutatum) using photo-identification software (Hotspotter) to analyze individual ventral spot patterns. The second objective was to evaluate how seasonal weather patterns influenced the species' population abundance and trends over time at one site in Tennessee. I hypothesized that Hotspotter would accurately identify returning females and that females would exhibit nest site fidelity by returning to the same nesting area across multiple breeding seasons. To assess site fidelity, I used a Wilcoxon rank-sum test to compare re-nesting distances among individuals. To evaluate the effects of climate on reproduction, I examined the influence of daily maximum temperature, daily precipitation, and daily relative humidity on average clutch size and total annual nest abundance. Immediate and lagged climate effects were analyzed using a sliding window approach (1-48 months prior to nesting) within the R package climwin. I hypothesized that females would produce smaller clutches following unfavorable pre-nesting conditions and that the population would exhibit a recovery period of approximately two years. Results supported the site fidelity hypothesis, demonstrating that female-four-toed salamanders consistently rerun to the same or adjacent moss clump for oviposition. Climate analyses revealed that mid-fall to early-winter precipitation and early spring temperature in the year prior to nesting were the primary drivers of variation in clutch size, while only precipitation significantly influenced total annual nests. Our findings highlight the importance of long-term data for addressing knowledge gaps in nesting behavior and emphasize the need to protect critical habitat while continuing to collect data to better understand how four-toed salamanders may response to future environmental changes.
Characterization of Southern Apple Varieties for Cider Production
Vowell, Olivia (Virginia Tech, 2026-04-06)
Interest in regional ciders is growing in the Southern United States, with local apple cultivars playing a critical role in shaping the chemical composition and sensory characteristics of these styles of cider, yet limited research has systematically characterized Southern adapted apple varieties for cider production. This study evaluated the juice chemistry, cider chemistry, and sensory properties of single varietal ciders produced from fifteen apple cultivars grown in Virginia. Apples were harvested at commercial maturity, processed under standardized conditions, and fermented to dryness using a single yeast strain in a temperature-controlled environment. Juice samples were analyzed for key parameters relevant to cider production, including titratable acidity (TA), pH, yeast assimilable nitrogen (YAN), soluble solids, and total polyphenols. Finished ciders were analyzed for chemical composition and evaluated using descriptive sensory analysis by a trained panel. Significant cultivar dependent differences were observed across juice and cider chemistry. Polyphenol concentration exhibited the greatest variability among cultivars, with cider and crab apple cultivars generally showing higher phenolic content than dessert cultivars. Titratable acidity also varied significantly among cultivars and remained largely consistent from juice to cider, contributing to differences in perceived acidity and sharpness. Although alcohol concentrations differed among cultivars, alcohol related sensory attributes were not strongly discriminative. Multivariate analyses integrating chemical and sensory data indicated that polyphenols and organic acids were primary drivers of sensory differentiation, particularly for astringency, bitterness, sourness, and sharpness. Collectively, these results demonstrate that apple cultivar can be a driver of both chemical and sensory variation in cider. This work provides a framework for understanding how Southern adapted apple varieties influence cider quality and supports their potential use in producing regionally distinctive ciders.
Jasmine Center: Home Away From Home Home As Memory, Identity, And Light In The Heart Of Washington, DC
Fatima, Anosha (Virginia Tech, 2026-04-02)
HOME AWAY FROM HOME Home as Memory, Identity, and Light in the Heart of Washington, DC Anosha Fatima The thesis examines the concept of home not merely as a physical structure, but as a deeply felt experience. Home is conceived as something that goes beyond the boundaries of the actual building and is influenced by elements such as memory, emotion, and culture. The thesis further explores the role of architecture as an element that defines the experience of home. Architecture is seen here as a platform through which identity, tradition, and community can be made visible and accessible. Through the integration of theatre as a place of cultural performance, a restaurant as a location of hospitality, and open courtyards as gathering places, the thesis showcases how architectural design can be used to encourage interaction between different groups of people. Through the integration of these programmatic elements, the thesis aims to create a space that not only encourages interaction but also a space that allows a community to be seen and represented.
Exhibiting the Immersive: Museum Architecture with Digital Augmentation
Peng, Yuanting (Virginia Tech, 2026-03-31)
Through several examples in the recent 5 years, more and more museum seek to cooperate their experience of viewing with what's popular through social media: digital arts and moving image. As the designers starting to utilize more and more moving image and 3-d projecting into installation, even augmented reality and virtual reality, one thing seems to be neglected, the architectural space itself didn't match up with the desire of evolve in Museum. Museum designing coherently with digital augmentation can show its spatial qualities as both an independent architecture, and an immersive imagery experience in the present age. An age that calls and relies on digital media and screen more than ever before. This proposal aimed to explore the possibilities how architecture space shift with the participation of digital techniques, especially in the culture architectural space of museum. Through the design of a proposed National Women Museum in the National Mall in DC, this thesis project focusing on the study of how space is transformed and used in different situations while both with digital images and without. The study also go through different historical application of immersive design and their architectural character. By utilizing precedence experience with latest architecture technology, the thesis explored on the spatial aspect of how exhibition architectural reacting towards digital immersive technology. In the design project, the aim is to purpose a sequence of experience for the visitor with sanctuary-liked spatial quality and coherent structure system supporting the technical façade and architecture itself. Considering both the historical context of the National Mall as the site and the rubric for architecture, the design tries to develop a narrative through the argument of what an museum architecture should be like.
Pigs' BCS Estimation using Computer Vision and Deep Learning Approaches
Desai, Zeel Amitkumar (Virginia Tech, 2026-03-27)
This research presents an end-to-end automated system for Body Condition Score (BCS) assessment in pigs using multi-modal RGB-D computer vision and deep learning. Manual BCS evaluation is subjective, labor-intensive, and inconsistent across large-scale farming op erations. To address these limitations, we developed a hybrid ensemble deep learning pipeline combining ResNet-50 and DenseNet-121 architectures with integrated depth information from Intel RealSense D435 cameras. The system was trained and validated on a dataset of 268 pigs across six pens collected from Virginia State University farms, with video streams captured as .bag files and converted to PNG images for analysis. Experimental results demonstrate that the multi-modal RGB-D approach achieves a 13.68% accuracy improvement over traditional RGB-only methods when evaluated using the en semble model. The hybrid ensemble achieves 84.18% accuracy using multi-image temporal aggregation across five architectures: ResNet-50, DenseNet-121, EfficientNetV2-S , Vision Transformer, and the proposed hybrid ensemble . Overall, the system achieves 84.18% multi-image classification accuracy. The proposed automated pipeline demonstrates the feasibility of objective and scalable livestock health monitoring, with potential productivity gains through improved nutri tional management. Future work will focus on expanding the dataset through multi-farm validation and integrating behavioral monitoring systems to enable more comprehensive an imal welfare assessment.
Subverting the Script: Female Playwrights and the Fluidity of Gender and Class in 18th-Century France
Stacks, Kerrie Maria (Virginia Tech, 2026-03-27)
This thesis explores how gender and marriage are depicted in 18th-century pre-revolutionary French theaters written by women. While conventional tropes of arranged marriage and forbidden love permeate the period's theater, this study highlights a focus on gender inversion and female autonomy within the selected corpus. By analyzing the works of Barbier, Graffigny, and Benoist, the research demonstrates how Enlightenment values were interpreted through a gendered lens, resulting in fluid portrayals of behavior, emotion, and kinship systems. Furthermore, by expanding the scope beyond public stages to include private theatrical spheres, this thesis reveals how female dramatists utilized the public nature of theater to challenge social norms. Ultimately, these works facilitate a critical dialogue on the evolution of gendered identity and the subversive potential of early modern female authorship.
Methods for Improving Comparability of Propeller Acoustic Experiments Conducted at Different Facilities and Scales
Duong, ThanhLong James (Virginia Tech, 2026-03-26)
Development of urban air mobility (UAM) vehicles are on the rise, and their noisy operation near many people can cause negative health effects, so it is important to quickly understand their acoustic behaviors. This is done through many conduits, including computational fluid dynamics (CFD), outdoor field tests, and controlled indoor experiments. Outdoor field tests allow for full-scale vehicle testing, but subjects the test article to uncontrollable, unpredictable variables, whereas indoor experiments are much more controllable and predictable, but such indoor experiments are often constrained by space and are unable to test the full-scale vehicle. Research should develop methods to improve comparability between indoor and outdoor acoustic experiments. The presented work aims to provide such methods with three experiments: one investigating ground board acoustic behavior in an anechoic chamber, one investigating propeller noise in an outdoor facility, and one investigating propeller noise in an indoor facility. Ground boards are thin rigid boards, typically in the shape of a 1m diameter circle a few millimeters thick, and are standardized for use [5] [10] [11] [13] in outdoor acoustic experiments to mount microphones and provide a consistent reflective surface. An experiment was conducted in an anechoic chamber to observe the behavior of a 1m diameter, 6mm thick plywood ground board with the noise source at various incidence angles and with the microphone lain at the center and at three-quarters the radius of the ground board. The configuration of interest is a 23° incidence angle with the microphone at the center since this is the configuration used in the outdoor field test. With this configuration, the behavior of the ground board is similar to that of a perfect reflector (6dB increase in measured sound pressure level (SPL)) within about 200Hz to 10000Hz. Below 2000Hz, the shallower source angles (20°, 23°, and 45°) have a lower SPL magnification; in the 2000Hz to 8000Hz frequency range, there appears to be little dependence of ∆P SD on the source angle; above 8000Hz, the higher source angles (65° and 80°) have a lower SPL magnification. With the microphone lain at three-quarters the radius of the ground board, a similar trend was observed, but its sensitivity to the source angle was lower. A Techsburg Inc. propeller was tested in an outdoor facility at two scales: 0.9144m and 0.4572m diameters. The propeller was mounted with its center 1.905m above the ground, and 13 microphones were placed on top of ground boards in a 4.572m semicircle around the propeller, ranging from upstream to downstream of the propeller. Wind measurements were simultaneously collected from an anemometer a few meters away from the microphone arc. The background noise at this facility was significant up to about 800Hz, and motor noise was significant around 5000Hz to 9000Hz. Wind velocity data was collected simultaneously to acoustics, and a relationship was found between broadband noise and inflow velocity. The total propeller noise was decomposed into broadband and tonal components through phase averaging methods, and it was found that the overwhelming majority of the noise was dominated by the broadband component. Generally, when the inflow velocity was about −1m/s to 0.5m/s, the broadband noise within 2000Hz to 5000Hz was about 1dB to 2dB louder than when the inflow velocity was about 0.5m/s to 3m/s. Furthermore, the spectra calculated from when the inflow velocity was between −1m/s and 0.5m/s was much more comparable to the Gill-Lee Spectrum Model (GLSM), a trailing edge noise prediction model trained on hundreds of propeller noise datasets primarily at a hover condition. To compare the two scales, the tip Mach number was held constant, so the 0.9144m diameter propeller rotated half as fast as the 0.4572m diameter propeller, resulting in half the blade passage frequency (BPF) and twice the harmonics. The broadband noise of the larger propeller peaked around 2000Hz, whereas the smaller propeller's broadband curve peaked around 4500Hz. This relative peak behavior is reflected in the GLSM, although the GLSM generally overpredicts both scales. The expected difference in tones between the two scales was calculated from the thrust squared to be approximately 11dB, and the observed difference in tones was observed to be approximately 7dB at most. However, these thrust measurements were averaged over the entire sampling period, so the absolute magnitude of this calculation should be considered with reasonable skepticism. The reduction in pitch angle resulted in quieter broadband noise, which is expected since this would reduce thrust and therefore reduce noise. The addition of turbulence trips did not result in the expected consistent increase in broadband noise, but this may be due to the inherent difficulty of outdoor acoustic experiments: the environmental conditions in field tests are uncontrollable, and sequestration was unable to isolate period of similar wind velocities, so the two datasets were not entirely comparable. A similar experiment was performed in an anechoic wind tunnel, the Virginia Tech Subsonic Modular Anechoic Research Tunnel (VTSMART), to observe the effects of test facility. Five microphones were mounted at the same height as the propeller 1.016m away from the propeller, evenly spaced by 10° between each, starting from 10° upstream to 30° downstream. A sixth microphone was mounted further downstream closer to the axis of rotation. Due to size constraints of the wind tunnel, only the 0.4572m diameter propeller was tested at this facility. Background noise was much quieter and more consistent than the outdoor facility, and the SNR was positive throughout 100Hz to 20000Hz. Spectrograms were generated for both background and propeller noise measurements, and both were found to be relatively constant in time, so the background noise was directly subtracted from the propeller noise measurements. Again, the total propeller noise was broken into broadband and tonal components, and the majority of the noise was still composed of the broadband component. The tunnel was operated at two conditions: one without flow and one with 4m/s flow. Between the two tunnel conditions, the broad spectral features did not change significantly, and the largest differences were observed in the tonal features. With 4m/s flow, the tones at the BPF were quieter than without flow by approximately 5dB. The GLSM generally captured the broad spectral features at the indoor facility better than at the outdoor facility, typically within about 3dB of the experimental data. The data from both experiments were compared to each other after a correction to account for the difference in distance. The outdoor broadband noise with an inflow velocity of about 0.5m/s to 3m/s was similar to the indoor broadband noise with 4m/s flow in the 2000Hz to 6000Hz range within about 2dB, and a similar result is seen when comparing the outdoor broadband noise with an inflow velocity of −1m/s and 0.5m/s to the indoor broadband noise with no flow. All configurations were again compared to each other at this facility, but much smaller differences were observed. This is likely due to the confinement of the wind tunnel facility and the inability of the wake to be convected far enough downstream of the propeller, causing these broadband noise sources to dominate over changes in pitch and turbulence trips. In each experiment, there are many limitations and possible improvements with further research. For the ground board experiment, more source incidence angles, microphone orientations, and surrounding substrates could be tested, which would improve the understanding of ground board response. Such research has been done in the past [2] [4] [14] [22]. For the outdoor Drone Park experiment, the ground boards could have been calibrated before collecting propeller acoustics so that measurements could be properly corrected for the effects of the ground board in the field. Wind anemometer data could also be fully synchronized to the propeller acoustics such that direct correlations would be more meaningful. Similarly, synchronized performance data should be collected as well, specifically thrust and torque, which would allow for more accurate GLSM calculations, since this model takes thrust into consideration. For the indoor VTSMART experiment, particle imaging velocimetry (PIV) could be conducted to observe the physical phenomena and confirm if the confinement of the wind tunnel is indeed causing the wake to loiter near the propeller.
Flotation of Extreme Particle Sizes
Leland, Jack Hambleton (Virginia Tech, 2026-03-23)
Flotation is effective over a narrow particle size range. As particle sizes increase above ~150 µm, body detachment forces overcome surface attachment forces, while, at the same time, mineral liberation decreases, reducing attachment force increasing the detachment probability. As particle sizes decrease below ~20 µm, the collision probabilities between bubbles and particles decrease because fine particles follow streamlines around bubbles due to their low inertia. In this work, Jig Flotation and Two Liquid Flotation (TLF) are developed to extend the maximum and minimum particle size of flotation, respectively. The Jig Flotation process reduces detachment probabilities using an intermittent fluidized bed instead of a mechanically agitated flotation cell for bubble-particle collision. The pulsion and suction cycles creating the intermittent fluidization allow for the formation of a froth phase, allowing Jig Flotation to be selective for both fine and coarse particles. The jigging mechanism allows particles to be separated by density independent of particle size. This mechanism is enabled by the attachment of bubbles to selectively hydrophobized minerals, reducing their apparent density and enhancing their density difference with gangue minerals. In the present work, several Jig Flotation cells have been designed and tested, and encouraging results achieved on chalcopyrite ore. The TLF process uses recyclable oils instead of air bubbles to recover ultrafine coal particles. Oil, having higher contact angles than air bubbles, enhances the recovery of ultrafine particles. The TLF process has produced coal with below 5% ash and below 2% moisture with recoveries greater than 80% from fine coal wastes.
High-resolution Analysis of Demographic and Socioeconomic Characteristics of Households with Private Wells in the USA
Sear, Caroline Benson (Virginia Tech, 2026-03-19)
Private domestic wells supply drinking water to roughly 15% of the United States population, yet remain largely unregulated, and poorly characterized in terms of the populations they serve. While prior national studies describe broad patterns of domestic well use, they are generally limited to coarse regional resolutions or smaller geographic extents. Here, we present a national, high-resolution analysis of the demographic and socioeconomic profiles of households with private wells in the United States. We integrate geospatial data from a national database of groundwater wells with block- and block group- level demographic information from the 2020 U.S. Census and American Community Survey. Domestic wells are identified and spatially linked to population characteristics including race, educational attainment, and household income, with results aggregated to the county level. We further stratify findings by urban-rural classification to examine patterns across population density gradients. Our analysis revealed substantial spatial variation in the demographic characteristics of well users as compared to nonwell users. Overall, well users are more likely than their public water counterparts to identify as White and to reside in households with higher incomes than county medians. Differences in educational attainment are more variable across states and counties, but are generally modest in magnitude. Across counties, increases in educational attainment and in the proportion of White-identifying well users are associated with higher household incomes. These findings provide a demographic context for domestic well reliance and may support the development of more targeted policies and resources for domestic testing, treatment, and maintenance for populations dependent on private wells.
Distribution and Characterization of Herbicide-Resistant Italian ryegrass and Palmer amaranth in Virginia
Viric, Milos (Virginia Tech, 2026-03-11)
Weed infestation is the major reason for economic losses in agriculture. Italian ryegrass and Palmer amaranth are some of the most troublesome weed species in Virginia. These species are strong competitors with crops for growth resources which eventually leads to significant yield losses in absence of adequate control. One of the challenges for the control of these species is development of herbicide-resistant populations. There is a limited knowledge about the distribution of resistant populations of Italian ryegrass and Palmer amaranth in Virginia. Palmer amaranth resistance to glyphosate was confirmed in 2011 and Italian ryegrass resistance to diclofop was confirmed in 1993. These are the only two confirmed cases of herbicide resistance in Virginia but based on control failure reports, resistance to these species is suspected to be more widespread in Virginia. To investigate the distribution and levels of resistance in populations from Virginia there is a necessity for more updated surveys. A total of 32 populations of Italian ryegrass were collected. Plants were grown in the greenhouse to test for sensitivity to herbicides commonly used for burndown or in-crop control of Italian ryegrass: pinoxaden, diclofop, glyphosate, mesosulfuron, pyroxsulam, and pyroxasulfone. At 21 days after the herbicide treatments, visible injury ratings were recorded on a scale 0 to 100%, where 0 indicates no control and 100 represents complete plant necrosis. Populations exhibiting ≤49% control were suspected to be resistant. Based on this criteria, 10, 27, 0, 14, 0, and 7 populations were found to be resistant to pinoxaden, diclofop, glyphosate, mesosulfuron, pyroxasulfone, and pyroxsulam, respectively. Following the initial screening, dose-response assays with pinoxaden, diclofop, mesosulfuron and pyroxsulam were conducted. Resistance indices (R/S ratios), calculated based on GR50 (herbicide dose that reduced biomass by 50%) values for resistant and susceptible populations, were 20 for pinoxaden, 87 for mesosulfuron, and 161 for pyroxsulam. The R/S value for diclofop could not be determined because even the highest tested dose could not achieve 50% growth reduction in the resistant population. Cross and multiple resistance was observed in this study and 6% of populations were found resistant to pinoxaden, diclofop-methyl, mesosulfuron, and pyroxsulam. A total of 68 Palmer amaranth populations were collected from corn, soybean and cotton fields across Virginia. Palmer amaranth seedlings grown in the greenhouse were treated with: trifloxysulfuron, 2,4-D, fomesafen, atrazine, mesotrione, glyphosate, glufosinate and dicamba. Visible control ratings were recorded on a 0 to 100% scale, where populations with up to 49% injury were considered resistant. Upon testing the populations, resistance was found in 46, 1, 3, 7, 3, 50, 0 and 0 populations to trifloxysulfuron, 2,4-D, fomesafen, atrazine, mesotrione, glyphosate, glufosinate and dicamba, respectively. Dose-response assay for glyphosate revealed that GR50 value for resistant population was 1,238 g ae ha-1, however R/S value could not be calculated as susceptible population was not available. The R/S values for trifloxysulfuron, fomesafen and atrazine were 47, 14 and 18, respectively. Approximately 69% of the populations showed multiple resistance to two or more herbicide sites of action. Overall, findings from these statewide surveys provide critical insights into the current herbicide resistance status for both Italian ryegrass and Palmer amaranth in Virginia. This information will help growers better understand the effectiveness of commonly used herbicides and make more informed management decisions.
Family and Friend Support, Strain, and Loneliness Among Dementia Caregivers in Rural Appalachia
Stanfill-Carrillo, Brenda Liana (Virginia Tech, 2026-02-04)
Background: Loneliness, defined as a perceived deficit in the quantity or quality of social relationships, is associated with a range of adverse physical and mental health outcomes. Understanding contributors to loneliness among caregivers of people living with dementia (PLwD) is therefore an important public health concern. Family caregivers, particularly spouses or adult children, report higher levels of loneliness compared to non-caregivers and some other caregiver groups. This vulnerability may stem from time constraints on social activities, misunderstandings within existing relationships regarding care management, and losses in shared experiences as dementia progresses. Across the life course, family and friends serve as important sources of social support and relational connection, both of which may be associated with lower loneliness. Research Questions and hypotheses: Guided by Perlman and Peplau’s definition of loneliness and the Stress Process Model, this thesis examined whether perceived support and strain from caregivers’ family and friends are associated with loneliness among spousal and adult child caregivers of PLwD living in rural Appalachia, above and beyond caregiving demands. Caregiver relationships with the PLwD (wife, husband, daughter, son) were examined categorically to account for differences in relational roles and expectations. It was hypothesized that (1) caregiver relationship type would be associated with loneliness, with wives reporting the highest levels, and (2) greater family strain would be associated with higher loneliness, whereas greater family and friend support would be associated with lower loneliness. Methodology: Data were drawn from the Families in Appalachia Caring for Elders with Alzheimer’s Disease (FACES) study (N = 141). A three-step hierarchical multiple linear regression analysis was conducted to examine associations between caregiver relationship type, perceived family and friend support, perceived family and friend strain, and loneliness, controlling for unsupervised time and assistance with personal activities of daily living. Results: Wives reported higher levels of loneliness than husbands, daughters, and sons. After accounting for caregiving demands and relationship type, higher perceived family strain was associated with higher loneliness, whereas greater perceived friend support was associated with lower loneliness. Family support and friend strain were not significantly associated with loneliness in the final model. Implications: Findings suggest that perceived relationship quality, rather than the mere presence or number of social ties, is central to understanding loneliness among caregivers of PLwD in rural Appalachia. While friend support was associated with lower loneliness, addressing familial strain in caregiver interventions and psychosocial programming may be particularly relevant for efforts aimed at reducing caregiver loneliness.
A Qualitative Exploration of School-Based Intervention Needs Among Rural Appalachian Youth
Winograd, Dayna Gael (Virginia Tech, 2025-12-05)
Disordered eating is prevalent in the United States, with over 20% of children and adolescents reporting some form of disordered eating. Eating disorders are associated with detrimental physical effects and co-occurring mental health difficulties. One population that appears to be at high risk for developing disordered eating symptoms is rural youth. Unfortunately, rural youth often do not receive treatment for their disordered eating symptoms due to myriad care barriers, including geographical restrictions and financial constraints. School-based interventions offer promise to address such barriers and increase access to treatment among this vulnerable group. This study represents a first step at identifying rural youth needs and formatting preferences for a school-based intervention. Participants were 11 rural adolescents (Mage = 15.09) from Appalachia. Participants reported their demographic characteristics in surveys and completed a semi-structured interview assessing their needs and formatting preferences for a school-based intervention for disordered eating. Data were analyzed using inductive thematic analysis; the following themes emerged. Rural youth reported that an intervention should promote healthy and balanced eating, teach social media literacy, and discuss external factors and overlapping mental health difficulties, and de-emphasize the value of weight and shape. Rural youth also suggested that the intervention design take into consideration logistical and cultural factors of rural communities. These data suggest that rural youth’s treatment preferences align with existing school-based interventions for disordered eating. However, modifications may be needed to address logistical and cultural factors that may impact acceptability and feasibility of school-based eating disorder interventions in rural communities
Likelihood-Free Bayesian Inference with Efficient Uncertainty Quantification
Nouri, Arash (Virginia Tech, 2026-02-09)
Uncertainty quantification (UQ) in inverse problems is essential for reliable parameter estimation in scientific and engineering applications. This thesis presents a study on two frameworks that separately quantifies two fundamental types of uncertainty: aleatoric uncertainty, arising from inherent measurement noise and non-identifiability in the inverse mapping, and epistemic uncertainty, stemming from limited training data and model inadequacy. For aleatoric uncertainty quantification, a conditional Wasserstein Generative Adversarial Network with Full Gradient Penalty (cWGAN-GP) is employed to approximate the posterior distribution over parameters given observations. The trained generator enables efficient posterior sampling through a single forward pass, providing credible intervals and capturing potential multimodality in the solution space. A physics-informed extension, SGML-cWGAN, incorporates domain knowledge through physics-based loss terms to improve estimation accuracy. For epistemic uncertainty quantification, Prediction with Neural Network Corrections (PNC) is utilized, leveraging Neural Tangent Kernel theory to provide theoretically grounded uncertainty estimates. Bootstrap and stacking resampling methods generate multiple model instances, with prediction variance across instances serving as the epistemic uncertainty measure. The framework is evaluated on two benchmark problems: the FitzHugh-Nagumo (FHN) dynamical system and the Pacejka tire model. Results demonstrate that PNC achieves excellent performance on clean and structured noisy datasets, while cWGAN scales efficiently to large datasets containing up to 864,000 samples. The physics informed SGML-cWGAN achieves up to 33% improvement in mean squared error over the baseline cWGAN on the Pacejka dataset. However, a fundamental trade-off emerges: PNC faces computational constraints limiting applicability to datasets smaller than approximately 7,000 samples, while cWGAN requires a minimum of 8,000 samples for reliable performance. This incompatibility highlights the need for scalable epistemic uncertainty methods that complement data-hungry generative models. The findings demonstrate the viability of neural network-based approaches for uncertainty quantification in inverse problems, while identifying key limitations and directions for future research, including alternative simulation-based inference methods and improved posterior evaluation metrics
Communication in a Fractional World: MIMO MC-OTFS Precoder Prediction
Allen, Evan James (Virginia Tech, 2025-04-23)
As 6G technologies advance, international bodies and regulatory agencies are intensifying efforts to extend seamless connectivity especially for high-mobility scenarios such as Mobile Ad-Hoc Networks (MANETs) types such as Vehicular Ad-Hoc Networks (VANETs) and Flying Ad-Hoc Networks (FANETs). For these environments to be considered for long term adoption and use they must support Multiple-Input-Multiple- (MIMO) technology, rapidly fluctuating channel conditions in these environments place a heavy burden on traditional time-frequency CSI feedback schemes required for MIMO precoding. This motivates a shift toward delay-Doppler representations like those employed by Orthogonal Time-Frequency Space(OTFS) modulation, which offers greater stability under mobility. We derive an expression for the variation over time in the OTFS I/O relationship. We then use this to create a physics informed complex exponential basis expansion model prediction framework that maximizes the usefulness of outdated Channel State Information (CSI) in the presence of integer and fractional delay-Doppler channels and facilitates high mobility MIMO communication.
Arabinoxylan: Derivative Synthesis, Film Formation and Degradation
Mensah, Enock Dugbatey (Virginia Tech, 2025-08-06)
Polysaccharides are sustainable resources that can be processed into biofuel and biomaterials. However, a good understanding of their degradation kinetics and moisture content, which are critical factors in processing, is lacking and this merits investigation. In this work, the moisture content and chelator-mediated Fenton (CMF) degradation rate of regenerated arabinoxylan (RAX) films were studied and compared with regenerated cellulose (RC) films using a quartz crystal microbalance with dissipation monitoring (QCM-D) and atomic force microscopy (AFM). Results from the study revealed that RAX films had more water than RC films of comparable thickness. RAX films were more susceptible to CMF treatment than RC films and their degradation was maximum at pH 5 and 35 ℃. The regenerated arabinoxylan films described here would allow for in situ studies of biomacromolecule and small molecule interactions with hemicellulose films.
How Well Do Multimodal Large Language Models Score on Visual Personnel Assessments
Liu, Siyi (Virginia Tech, 2025-12-18)
Multimodal Large Language Models (MLLMs) pose new threats to the validity of visual personnel assessments in high stakes selection contexts, as their emergent visual perception and understanding capabilities may facilitate applicant cheating. The study investigated the performance of three popular MLLMs on one visual cognitive ability test bundle and one visual forced-choice personality test, across three prompt approaches and three temperature settings. It was found that MLLMs only achieved median-level scores (the 50th percentile) on the visual cognitive ability test, not as competitive as high-performing human test takers. However, they exhibited top-tier performance (over the 98th percentile) on Conscientiousness on the visual personality test, while exhibiting high scores on Agreeableness and Emotional Stability by nudging temperatures or prompts. Given MLLMs’ potential to enable applicant cheating in unproctored pre-employment assessments, the study urged test vendors and employers to implement anti-cheating measures and offered related recommendations.
Episodic Detail Production and Semantic Coherence in Down Syndrome and Fragile X Syndrome: Longitudinal Findings from Expressive Language Sampling
Van Vorce, Hailey (Virginia Tech, 2025-12-09)
Autobiographical memory requires the integration of episodic and semantic information and is closely tied to expressive language abilities. This study examined episodic detail production and narrative coherence in children and adolescents with Down syndrome (DS) and Fragile X syndrome (FXS) using conversational samples from the Expressive Language Sampling (ELS) Conversation task (Abbeduto et al., 2020, 2023). Participants (N = 50) contributed one matched autobiographical topic at two visits approximately 18 months apart. Episodic and semantic details were coded using the Autobiographical Interview (AI) framework (Levine et al., 2002), and narrative coherence was assessed using Semantic Distance (SemDis), a computational measure of conceptual relatedness (Beaty & Johnson, 2021). Multilevel models evaluated whether diagnostic group, expressive language, narrative length, and time predicted autobiographical memory performance. Across aims, children showed substantial variability in narrative output, with greater within-group than between-group differences. Diagnostic group did not significantly predict episodic detail production, and episodic content showed minimal change across time. Word count was the only significant predictor, indicating that children who produced more language provided more episodic content. No demographic or language variables uniquely predicted episodic detail production once narrative length was controlled. Semantic coherence was also stable across visits and did not differ by diagnostic group or narrative length. The only significant effect was a diagnostic group × CELF-FS interaction: higher expressive syntax predicted more coherent narratives among children with DS, whereas children with FXS showed a slight decrease in coherence as expressive syntax increased. Overall, findings indicate that expressive output, rather than diagnostic status, is the primary driver of autobiographical narrative performance in DS and FXS.
Towards Accurate and Reliable Industrial Intrusion Detection Systems Using Shadow Replicas
Nwodo, Kenechukwu Anthony (Virginia Tech, 2023-05-10)
Supervisory Control and Data Acquisition (SCADA) systems manage the operations of a plethora of safety-critical industrial control systems. Due to their sensitive nature, SCADA systems have been the target of adversaries employing a wide range of attacks. This thesis proposes an approach to protect SCADA systems against attacks that evade detection because of the lack of a comprehensive view of both application and network-layer responses. Specifically, we leverage multiple open-source Network Intrusion Detection Systems (NIDSs) paired with a SCADA shadow replica to provide both network and application threat detection. The shadow replica is augmented with a Finite State Machine (FSM) to compute the anticipated states of both the SCADA system and connected devices. Isolated from the operational network, it is protected from direct front-end attacks. When the SCADA system becomes compromised, even without an IDS alert, the replica can expose the attack and offer an operational failover. We implement a prototype of our system and evaluate it against locally executed attacks on commercial out-of-the-box devices and public IoT datasets. Results indicate that incorporating the shadow replica alongside NIDSs can enhance detection coverage in our evaluations.

Browse

Recent Submissions