Doctoral Dissertations

Permanent URI for this collection

https://hdl.handle.net/10919/11041

Browse

Now showing 1 - 20 of 18619

Computational Approaches to Answering Evolutionary Biology Questions
Sarkar, Chandra (Virginia Tech, 2026-10-15)
Advances in computational biology have expanded the scale at which evolutionary questions can be addressed, allowing patterns of molecular evolution, genome evolution, and host-symbiont diversification to be examined using large comparative datasets. Additionally, the repurposing of publicly available datasets provides opportunities to address new questions beyond those for which the data were originally generated, thus avoiding contributing to already exploding amounts of data in the current data-driven world. In this dissertation, I reuse and integrate existing next-generation sequencing datasets through computational pipelines to investigate contradictions to an already established molecular evolution paradigm, survey the occurrence of possible symbiotic bacteria and elucidate host-symbiont co-evolution. My first chapter provides insight into the current state of bioinformatics and challenges in evolutionary biology and provides an overview of the aims of my dissertation. My second chapter examines how cellular environment and spatial organization of proteins influences protein evolutionary rates. Highly expressed proteins evolve slowly, a pattern known as the expression level evolutionary rate anticorrelation (E-R anticorrelation). Using transmembrane proteins, I found that this anticorrelation is stronger in intracellular domains of the proteins as compared to their extracellular counterparts. These results support the hypothesis that protein evolution is constrained to selection against misfolding and deleterious molecular interactions. My third chapter provides a literature review of lice-Wolbachia endosymbiosis, an understudied ectoparasitic-endosymbiont system, that provides the background for my work in the following chapters. My fourth chapter uses publicly available whole-genome sequencing data from avian chewing lice to investigate their associations with the known arthropod-associated Alphaproteobacteria, Wolbachia. Screening and genome reconstruction from a large dataset of 1025 avian chewing lice shotgun sequences show a substantial association between the two organisms. I also report on the classification of these associated Wolbachia into supergroups and their substantial variation of genomic characteristics including genome size, GC content, and coding sequence counts among other metrics. My fifth and last chapter delves into understanding the possible coevolution events of lice and their associated Wolbachia. I use phylogenomics and comparative genomic analyses to understand their shared evolutionary history and indicate possible candidates for mutualism using computational validation. Together, these studies illustrate how existing biological datasets can be reanalyzed through computational pipelines to address new evolutionary questions and aims. Across molecular, genomic, and host-symbiont scales, this dissertation examines how we can understand layers of biological processes using computational prowess.
In-Situ Perception-Guided Autonomy in Robotic Additive and Hybrid Manufacturing
Kosmal, Tadeusz Adlai Perry (Virginia Tech, 2026-10-14)
Additive Manufacturing (AM) provides geometric flexibility, but its autonomy remains constrained by stochastic process outcomes and reliance on skilled operators. Layer-wise deposition can introduce variations unique to each part, yet most AM systems execute predetermined toolpaths without observing or responding to the evolving as-built state. Inspection, corrective processing, and assembly of printed parts consequently require human interpretation and coordination across disconnected manufacturing systems. This dissertation develops a framework for reducing this burden by converging in-situ layer-wise sensing, probabilistic geometric state estimation, adaptive additive and subtractive toolpath generation, and robotic fabrication. First, modular arrays of stereo structured-light scanners are embedded within AM environments and localized to a common machine coordinate frame. Measurements from multiple viewpoints are combined and evaluated against target geometry to identify a component's as-printed error state. The perception architecture is extended by calibrating long-wave infrared cameras to the stereo scanners, producing spatially registered thermal point clouds. The measured geometric state is then converted into corrective action. A prior-guided stochastic surface-reconstruction method combines noisy and sparse scan data with geometric priors derived from expected geometry. The resulting posterior implicit field provides a closed estimate of the as-built surface for subtractive toolpath generation. A complementary additive repair framework identifies damaged regions, generates multi-axis deposition paths, and optimizes tool orientations for collision avoidance. Iterative application reduces a component's geometric deficit while remaining compatible with layer-wise replanning. Finally, a robotic workcell integrating material extrusion, component placement, electrical connection, embedding, and part removal extends this autonomy to functional mechatronic fabrication by constructing flight-capable UASs without human involvement during robotic printing or assembly. Collectively, these contributions advance AM from open-loop execution toward adaptive manufacturing systems that observe, interpret, and act on the physical state of a printed component.
The Effects of Anti-DEI Legislation on Equity-Minded Engineering Faculty and Staff
Bigby Jr, Paul C. (Virginia Tech, 2026-10-14)
Investigating the Enzootic Transmission and Cross-protective Immunity of Mosquito-borne Orthoflaviviruses
Persinger, Rachel (Virginia Tech, 2026-10-13)
Orthoflaviviruses are widely dispersed globally and account for millions of infections annually. Usutu virus (USUV) is an orthoflavivirus that, like West Nile virus (WNV) and St. Louis encephalitis virus (SLEV), is maintained in an enzootic transmission cycle between passerine birds and ornithophilic mosquitoes (primarily Culex spp.). USUV belongs to the Japanese encephalitis virus (JEV) serocomplex, sharing antigenic similarity with WNV and SLEV. WNV and SLEV co-circulate in the United States, while USUV and WNV co-circulate in Europe and Africa. Since USUV emerged in Europe in 2001, it has caused mass mortality in Eurasian blackbirds (Turdus merula), closely resembling WNV emergence and avian mortality in the United States. In humans, severe USUV infection can cause neuroinvasive disease. This dissertation sought to address several aims regarding enzootic transmission dynamics for orthoflaviviruses. First, we established several viremic passerine bird models of infection for USUV, using both wild and domestic species. We found that American robins, domestic canaries, and house finches were susceptible to USUV, with 100% of birds becoming viremic. Disease was most severe in American robins followed by domestic canaries and house finches. Zebra finches did not become infected with USUV. Furthermore, we assessed the vector competence of North American Culex spp. mosquitoes that are important vectors in WNV transmission and determined that Cx. pipiens were competent vectors and Cx. tarsalis had limited evidence of transmission of USUV. We found that extending the extrinsic incubation period for Cx. pipiens resulted in increased transmission and that the dose required for infection of this species is approximately 5-6 log10 PFU/mL of USUV. Lastly, because these viruses co-circulate, we sought to understand how prior heterologous orthoflavivirus immunity impacts secondary WNV infection in domestic canaries. We determined that prior infection with USUV and SLEV incompletely reduces WNV viremia and tissue titers, as well as transmission potential. However, whereas USUV is protective against WNV mortality, SLEV is not protective against WNV mortality. Together, the following chapters of this dissertation provide foundational knowledge on USUV infection and transmission in North American passerine birds and Culex spp. mosquitoes, as well as insight into the effects of cross-protective immunity for USUV and SLEV on WNV infection in passerine birds.
Assessing Cognitive Workload using Electroencephalogram (EEG) in Technology-Mediated Work Environments
Jamshid Nezhad Zahabi, Saman (Virginia Tech, 2026-10-13)
Human–machine interaction has expanded rapidly with the growing integration of automation, robotics, and intelligent systems into industrial and training environments. Emerging technologies such as Augmented Reality (AR) and Virtual Reality (VR) are designed to enhance these interactions by providing immersive training, real-time task guidance, and enriched visual feedback. While these technologies have the potential to improve efficiency, safety, and learning outcomes, they also impose additional cognitive demands on operators who must process complex visual and spatial information while maintaining attention and situational awareness. In complex collaborative and immersive work environments, operators are often required to interact simultaneously with humans, machines, and digital interfaces. As human information-processing capacity is inherently limited, assessing cognitive workload is critical for understanding how operators respond to technology-mediated systems and for designing systems that better align with human capabilities and limitations. This research examines changes in operators' cognitive workload within three rapidly evolving technology-mediated environments: VR-based training, triadic human–robot-human collaboration, and AR-assisted teleoperation. Rather than relying solely on subjective self-reports, this research employs neurophysiological measures, specifically electroencephalography (EEG), to assess workload during task performance. To examine cognitive workload across these technology-mediated environments, three separate studies were conducted. Study 1 investigated cognitive workload among novice forklift drivers in a VR-based training environment. In a within-subject experiment involving 20 novice drivers, workload variations were analyzed across different task difficulty levels and repeated training sessions. Results indicated that both subjective assessments (NASA-TLX) and objective measures (EEG) effectively distinguished workload across varying levels of task difficulty. However, an additional training session did not substantially reduce the workload experienced by novice operators. Study 2 evaluated cognitive workload during triadic human-robot collaboration involving a teleoperator and an onsite worker performing a collaborative assembly task with and without robotic assistance. Thirty-two participants were recruited and assigned to either teleoperator or onsite roles. We also evaluated whether manipulating the task demands of onsite participants indirectly affected teleoperator workload. Findings demonstrated that EEG effectively detected workload differences across collaboration conditions and showed that robotic assistance reduced cognitive workload. However, no evidence of workload transfer from the onsite participant to the teleoperator was observed when the onsite task demands were manipulated. Study 3 assessed the effects of AR-based visual augmentation and material visibility on teleoperator workload. We developed an AR interface that overlays the real-time posture and position of the onsite participant onto the teleoperator's view. In a within-subject study involving 17 participants, workload was evaluated using NASA-TLX, EEG, and eye tracking measures under AR guidance and traditional multi-camera conditions, as well as under varying levels of material occlusion. Results showed that AR guidance reduced workload, while Occlusion level influenced EEG patterns and gaze behavior, suggesting adaptive visual information-processing strategies under limited visibility conditions. Overall, these studies demonstrate the value of multimodal approaches for assessing cognitive workload in technology-mediated environments. In particular, the findings highlight the sensitivity of EEG in capturing subtle variations in operator workload, with different EEG measures responding to changes in task demands and system design across VR training, human–robot collaboration, and AR-assisted teleoperation. While increased task difficulty consistently elevated workload, supportive technologies such as robotic assistance and AR guidance effectively reduced cognitive demands. These findings underscore the importance of integrating cognitive factors into the design of technology-driven systems to enhance human–machine interaction, improve operator performance, and support safer and more effective work environments.
From Vibration to Action: User Interpretation, Mapping Compatibility, and Transitions in Directional Vibrotactile Warnings
Yang, Xiang (Virginia Tech, 2026-10-13)
Struck-by incidents pose a risk of serious injury and death to workers in highway work zones, where noise, obstructed sightlines, and competing task demands can limit the effectiveness of conventional visual and auditory warnings. Wearable vibrotactile warnings, delivered through protective equipment such as vests and helmets, offer a complementary warning channel, but workers must interpret a directional vibration and translate it into timely evasive action. The same bodily cue can carry different meanings to different users. This dissertation examined how individual interpretation, signal design, and changes in cue meaning shape responses to directional vibrotactile warnings. The first study analyzed Smart Vest data from 15 participants to examine untrained interpretations of directional cues and preferences among three mappings. Interpretations and preferences varied, with no significant participant-level association between them. The second study examined responses to pulse rate and cue location in 40 participants using a Smart Helmet during a working-memory task in an immersive virtual work zone. During initial use, mean reaction time to head-movement onset was approximately 0.9 s, and reaction times were about 50 ms longer for front than right cues. No overall differences between the two tested pulse rates were detected in reaction time, workload, or usability during initial use, or in urgency and annoyance rated after the experiment. The third study examined repeated use and mapping reversal in the same helmet experiment. Participants used Safe Path, indicating where to move, and Hazard Direction, indicating where a hazard was located. Usability increased with repeated use, but no corresponding reduction in reaction time was detected. Usability and reaction-time changes across reversal were less favorable from Safe Path to Hazard Direction than in the reverse order. Across the experiment, 95.5% of observable responses broadly followed the instructed direction. After both mappings, 27 of 40 participants preferred Safe Path. Collectively, these findings indicate that cue meaning and prior experience should be considered alongside signal design when directional vibrotactile warnings are evaluated for deployment. Whether users benefit from a cue-to-information mapping that matches their initial interpretation, and whether these warnings improve safety in real work zones, remain to be tested.
Understanding angler heterogeneity in recreational fisheries science and management in the Great Lakes
Badhon, Mahatub Khan (Virginia Tech, 2026-10-13)
Recreational fisheries research and management treat anglers as a relatively homogeneous population that primarily participates for sports or fun. This approach can obscure differences in why people fish, how they use fisheries resources, what they catch, and how they evaluate management institutions. In this dissertation, I examined angler heterogeneity and its implications for fisheries management, monitoring, and governance among shore-based fishers in urban communities in the lower Great Lakes. In the first study (i.e. Chapter 2), I used survey data from 443 shore anglers across 10 cities around Lakes Erie and Ontario to identify the principal dimensions of angler diversity. Using latent class analysis, I identified three distinct angler groups along a food provisioning–leisure/sport continuum: provisioning/food-oriented anglers, cultural/community-oriented anglers, and leisure/sport-oriented anglers. These findings demonstrate that recreational and non-recreational values coexist within urban shore fisheries, and that food provisioning and cultural practices are important components of angler diversity. Building on these angler groups, I examined whether this orientation-based angler heterogeneity was associated with differences in catch and species use in Chapter 3. Using directed acyclic graph-informed models, hierarchical catch models, and hierarchical Bernoulli models, I found that provisioning/food-oriented anglers had the highest model-standardized expected catch (535 lb per angler), approximately 1.8 times that of leisure/sport-oriented anglers (302 lb per angler). The expected catch for cultural/community-oriented anglers was intermediate (389 lb per angler). The groups also diverged in target species as provisioning/food-oriented anglers were most associated with carp and catfish/bullhead, cultural/community anglers with walleye and perch, and leisure/sport anglers with bass. Since current monitoring relies on aggregate catch metrics and has no specific estimates for provisioning/food-oriented and cultural/community-oriented angler groups, my findings suggest that current estimates can conceal systematic differences in catch intensity and fishery use among anglers. Therefore, current estimates may underestimate fishing intensity exerted by high-catch subgroups and overlook species-specific impacts. In Chapter 4, I investigated how multidimensional trust influences fishers' willingness to support fisheries authorities. Using confirmatory factor analysis and structural equation modeling with survey data from 385 recreational fishers in Southern Ontario, I found that affinitive, rational, and procedural trust were highly intercorrelated and best represented as a higher-order actionable trust construct. Actionable trust exerted a strong positive effect on support intention, whereas dispositional trust operated only indirectly. A hybrid deductive–inductive thematic analysis of qualitative responses further revealed that fishers' trust-building expectations span multiple trust dimensions simultaneously. My dissertation demonstrates that recreational fisheries in the Great Lakes are socially heterogeneous systems in which distinct orientation-based angler groups differ in their potential ecological footprints. Findings indicate that management and monitoring frameworks must account for orientation-specific patterns across anglers, and that fisheries management agencies can strengthen voluntary cooperation by deliberately cultivating multidimensional trust. These insights contribute to a more nuanced understanding of recreational fisheries in the lower Great Lakes.
Effects of Back-Support Exoskeletons on Balance Recovery From Overground Slips and Trips
Freidouny, Mohammadreza (Virginia Tech, 2026-10-13)
Back-support exoskeletons (BSEs) are widely used in industry to reduce risk factors for work-related musculoskeletal disorders during manual material handling tasks. They may, however, increase fall risk by adversely affecting balance recovery after slips and trips. Prior work on BSEs and their effects on balance recovery has mainly relied on treadmill or lean-release perturbations to simulate slips and trips, which may elicit different recovery responses than overground slips and trips. This dissertation comprises three studies that investigated the effects of passive BSEs on fall rate and balance recovery after overground slips and trips. The purpose of the first study was to investigate the effects of BSEs on fall rate and balance recovery after lab-induced overground trips. Two commercial passive BSEs with distinct designs were selected: the HeroWear, which has soft elements and elastic bands, and the BackX, which has a rigid frame and mechanical torque generators. Participants were exposed to two overground trips, one with and one without a BSE. The results did not show a difference in fall rate after trips between wearing and not wearing a BSE, or between BackX and HeroWear. However, BSEs elicited changes in balance recovery kinematics, including shorter recovery step time and shorter recovery step length with HeroWear and, when participants used an elevating recovery strategy, greater hip drop with BackX. The purpose of the second study was to investigate the effects of BSEs on fall rate and balance recovery after lab-induced overground slips. Participants were exposed to a single overground slip with or without a BSE. The results did not show a difference in fall rate after slips between wearing and not wearing a BSE, or between BackX and HeroWear. However, BSEs elicited changes in balance recovery kinematics, including smaller peak hip drop with HeroWear and greater slip distance with BackX. The purpose of the third study was to investigate the effects of anterior load carriage and BackX on fall rate and balance recovery after lab-induced overground trips. The results did not show an increase in fall rate after trips with anterior load carriage. However, anterior load carriage elicited changes in balance recovery kinematics, including longer recovery step time, longer recovery step length, greater hip drop in females, and smaller torso flexion angle. Additionally, the results did not show an increase in fall rate after trips with BackX during anterior load carriage. However, BackX elicited changes in balance recovery kinematics, including a shorter recovery step time and smaller hip drop. Taken together, these studies provide insight into how passive BSEs affect balance recovery after overground slips and trips. The findings do not support the concern that these BSEs increase fall rate after slips or trips. However, BSE effects on balance recovery kinematics were device-specific and varied from beneficial to adverse, which can inform BSE use in workplaces where slips and trips are common.
Flow and Flame Structure in Solid Fuel Combustion for Scramjets
Prokesch, Youri (Virginia Tech, 2026-10-12)
Scramjet technology is necessary to enable sustained hypersonic flight on an airbreathing engine architecture. Without moving mechanical parts within the engine, the scramjet already benefits from simplicity. With the use of solid fuels, a further reduction in system complexity can be obtained in addition to increased energy density, safer storage and handling, and added thermal management for combustor walls. To realize the advantages of a solid fuel scramjet (SFSJ), however, numerous challenges need to be addressed. Chief among these, the residence time of flow within an SFSJ may approach the time scale of combustion reactions, making gas phase flameholding challenging after contending with solid fuel decomposition. In line with the prior point, solid fuel combustion is inherently a complex, highly coupled process that lacks simplified crossflow experiments tailored to model validation. To address these challenges, this dissertation presents a series of experiments, beginning with an investigation of solid fuel diffusion flames in a small, low-speed crossflow burner, followed by two studies in optically accessible SFSJ combustors. The crossflow burner, in addition to serving as a useful experiment for optical diagnostic development and assessment, used O2/inert diluent (He, N2, Ar) oxidizer mixtures to explore the influence of thermodynamic gas properties on solid fuel regression rates. Results suggest interesting possibilities for future fuel design based on properties of pyrolysis products generated near the fuel surface. In a direct-connect facility, a lab-scale SFSJ combustor was used to explore flameholding stability under various flameholding cavity rear wall angles, along with a new fuel grain design that utilizes an upstream section of fuel flush with the inlet ("flush fuel grain"). The results demonstrated good agreement with mechanisms describing non-reacting supersonic flows, an increase in flame anchoring fluctuation with a steeper cavity rear wall, and that the flush fuel grain geometry can substantially improve flameholding ability and sustained combustion at lower inlet temperatures than prior SFSJ fuel geometries. The improved flammability by the flush fuel grain enabled the development of a three-sided optically accessible SFSJ by demonstrating sustained combustion of a single fuel grain on the bottom of the combustor. Laser diagnostics were applied to the three-sided SFSJ combustor for the first time and provided insight on flame structure and condensed phase behavior throughout the flow field. The work within this dissertation has improved the understanding of solid fuel combustion and expanded the operability limits of SFSJs to provide insight for future design and development efforts.
Development and Evaluation of Vision-Language Models for Ergonomic Physical Exposure Assessment in Occupational Manual Material Handling
Rajabi, Mohammad Sadra (Virginia Tech, 2026-10-12)
Work-related musculoskeletal disorders remain a leading cause of lost workdays and substantial costs for workers and employers, and manual material handling (MMH) tasks are a major contributor to the physical exposures that drive this risk. Existing physical exposure assessment methods, including self-reports, systematic observation, and direct measurement, are often labor-intensive, costly, or impractical for use in real work environments. Vision-language models (VLMs), which combine visual encoders with large language models to jointly process video and text, offer a potentially non-invasive and scalable alternative for quantifying physical exposures from regular RGB video. This dissertation developed and evaluated VLM-based methods for physical exposure assessment during MMH tasks using three sequential studies. The first study evaluated the use of VLMs to classify eight MMH tasks and specific task conditions (hand configuration and lift origin), and to temporally segment task start and end times. The work achieved task classification accuracies of ~82–85%, lift-origin classification accuracies of ~94–98%, and errors in mean absolute start and end time of <0.5 seconds. The second study assessed the feasibility of using two VLM-based pipelines to estimate horizontal (H) and vertical (V) hand distances required by the Revised NIOSH Lifting Equation (RNLE). Using a segmentation-based, multi-view pipeline consistently yielding the smallest errors (mean absolute errors of ~6–8 cm for H and ~5–8 cm for V). The third study evaluated a VLM-based pipeline for estimating dynamic, triaxial, bilateral external hand forces during MMH tasks. Root mean square errors of ~4.7–5.6 N were obtained for the horizontal and mediolateral force components and ~10.6–11.0 N for the vertical component. Collectively, these findings demonstrate the feasibility of VLM-based methods as objective, scalable, and minimally intrusive tools for quantifying physical exposures during MMH tasks, supporting their continued development toward practical, video-based occupational physical exposure assessment.
Where Automation Ends: Roadway Infrastructure and the Deactivation of Driving Automation Systems
Hong, Yubin (Virginia Tech, 2026-10-12)
Driving automation systems operate on roadway infrastructure designed for human perception and control, yet automation safety is often treated as a property of the vehicle alone. This dissertation reframes automation deactivation, the moment at which automation ends and control returns to the human, as a systems-level boundary event that emerges from the interaction among roadway infrastructure, automation capability, and human supervision. Deactivation is not necessarily a failure, but it is safety-relevant because it redistributes control, and both its frequency and its form depend on the roadway environment in which it occurs. Three studies examine where and how automation reaches its operational boundary. Study 1 analyzes seven months of naturalistic data from a low-speed automated shuttle (Level 4) deployed on public roads, identifying the roadway conditions under which operation degraded or required operator compensation, including four-way stop-sign-controlled intersections, three lane roadways with shared center turning lanes, segments with unclear pavement markings, and a mismatch between vehicle and surrounding traffic speed. Study 2 uses Bayesian multilevel logistic regression on 771 video-coded Level 2 event windows and finds that signalized intersection influence, left-curved alignment, more than two contiguous travel lanes, and ramp context were associated with higher odds of L2 deactivation, whereas not divided traffic flow was associated with lower odds. Study 3 applies a Bayesian multilevel multinomial model to 201 deactivation events and finds road type differences in deactivation subtype composition; a prior sensitivity analysis separated findings by strength of evidence. With freeway and system-initiated deactivation as the reference categories, the reduced odds of driver override on arterial roads held under every prior tested, whereas the other two patterns (increased driver-initiated deactivation on arterials and reduced driver override on collector/residential streets) were credible under only some priors and are interpreted more cautiously. These patterns are offered as hypotheses about how drivers calibrate their reliance on automation in different roadway contexts, not as direct measurements of trust. The findings are synthesized through two linked Haddon matrices that organize contributing factors and evidence separately from candidate countermeasures, across driver/operator, automation system, roadway/infrastructure, and policy/organizational domains and across prevention, pre-deactivation, deactivation, and post-deactivation phases. The dissertation contributes naturalistic field evidence linking specific roadway features to deactivation likelihood and type, a conceptual framework treating deactivation as an emergent systems-level event, and a methodological template for analyzing sparse, clustered naturalistic driving data.
Heteroskedastic Gaussian processes for ecological forecasting
Patil, Parul Vijay (Virginia Tech, 2026-10-09)
Modern stochastic simulators and ecological time series share a common challenge: they exhibit input dependent noise. In such situations, accurate predictions into the future along with principled uncertainty are essential. Heteroskedastic Gaussian processes (hetGPs) are a flexible and nonparametric modeling technique wherein, two processes are inferred simul- taneously – the mean response and the latent noise levels corresponding to input locations. In this dissertation, I show how hetGPs are versatile and can be used in both small and large scale regimes. First, I show how hetGPs outperform other models in data-driven forecasting of sparse and irregularly sampled tick densities across several different locations. Then, I develop a Bayesian and scalable hetGP (bhetGP) to handle the scale of modern stochastic simulation campaigns without undercutting UQ. Next, I use this novel bhetGP for cali- bration of a computer model designed to forecast lake temperatures. Finally, I provide an extension to hetGPs where the noise can only be modeled for a subset of dimensions re- ducing computational cost and complexity. An open-source implementation for the bhetGP method and the extension is provided as the bhetGP R package on CRAN. Together these contributions establish hetGPs as a principled, scalable, and broadly applicable framework for ecological forecasting – one where uncertainty is a core deliverable.
Evaluating Brewery and Ethanol Industry Byproducts as Functional Ingredients in Aquaculture Diets
Pough II, Jason D.'wayne (Virginia Tech, 2026-10-09)
From Detection to Reliability: Supervised Contrastive Learning and Perception Evaluation in Dynamic Conditions for Autonomous Driving
Jiang, Boyu (Virginia Tech, 2026-10-08)
Safe autonomous driving requires both reliably detecting rare safety-critical events (SCEs) — crashes and near-crashes — and understanding when the perception systems responsible for detection can be trusted. This dissertation addresses this progression from detection to reliability. On the detection side, this work develops representation learning frameworks grounded in Supervised Contrastive Learning (SCL). Applied to the Second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study (NDS) video data, SCL enhances intra-class cohesion and inter-class separation in the learned representation space, improving crash/near-crash/baseline classification over supervised and self-supervised contrastive benchmarks. Extending this principle to vehicle kinematics, the proposed supervised contrastive variational autoencoder (scVAE) uses a dual-encoder architecture to disentangle a label-regulated salient latent space from shared driving patterns, achieving state-of-the-art clustering and enabling scenario generation, denoising, and event prediction. On the reliability side, this dissertation introduces Perception Characteristics Distance (PCD), a metric that quantifies the maximum distance at which a perception system remains reliable under a given decision rule. PCD models the mean and variance of perception performance as a function of distance via spline regression and variance change-point detection. Evaluated on the newly introduced SensorRainFall dataset across weather and lighting conditions, PCD reveals distance-dependent reliability characteristics that conventional static metrics overlook. Together, these contributions provide a unified statistical paradigm for both detecting SCEs and assessing perception reliability in dynamic conditions, supporting the development and validation of safe automated driving systems.
Computing with Spikes: Legendre-based SNNs and Deep Local Learning toward Neuromorphic Intelligence
Gaurav, Ramashish (Virginia Tech, 2026-10-08)
The pursuit of Green AI has brought Neuromorphic Computing with Spiking Neural Networks (SNNs) to the forefront as an energy-efficient alternative to conventional Artificial Neural Networks (ANNs). This dissertation advances Neuromorphic Computing along four interconnected directions: the algorithmic design of reservoir-based SNNs, their deployment on Intel's Loihi-2 neuromorphic chip, a biologically-inspired local-learning method to train deep SNNs, and real-world neuromorphic applications spanning wireless domain and event-based vision. We begin by proposing a family of Legendre Delay Network (LDN) based SNNs for Time Series Classification (TSC): the Spiking Reservoir Computing (SRC) model, the Legendre-SNN (LSNN), and the Deep Legendre-SNN (DeepLSNN). These models use the LDN as a static reservoir to extract temporal features, followed by a trainable spiking network. Herein, we present the first spiking-TSC benchmark by evaluating the DeepLSNN on 102 TSC datasets. We then address the non-trivial challenge of deploying these heterogeneous models entirely on Loihi-2 - by implementing the quantized LDN on the chip's embedded Lakemont (LMT) cores and the spiking network on its Neuro-Cores; thus, this work contributes scarce technical documentation to program the LMT cores. We next propose "DALTON" – a Three-Factor-Rule-based local-learning method that effectively leverages the depth of Convolutional SNNs. Finally, we present two real-world neuromorphic applications on Loihi-2: 5G jamming detection and drone-recognition using our newly introduced related datasets. Across our works, we find that our neuromorphic methods are not only highly energy-efficient and low-latency, but also help close the performance gap between SNNs and ANNs, thereby paving the future of neuromorphic intelligence.
Computational Approaches to Support Environmental Surveillance of Antimicrobial Resistance
Manthapuri, Vineeth (Virginia Tech, 2026-10-07)
Antimicrobial resistance (AMR) is a major threat to global health. Antibiotic resistance genes (ARGs) are released from wastewater treatment plants and livestock farms, and many are carried on mobile genetic elements (MGEs) that can move between bacteria. AMR in these settings has mostly been studied with culture-based methods, which are slow and capture one organism at a time, or with quantitative PCR, which measures only a few pre-selected genes. Non-target chemical analysis and shotgun metagenomic sequencing measure many more chemicals and genes in each sample but require computational methods for interpretation. This dissertation develops and applies computational approaches to three related problems: predicting the removal of pharmaceuticals and personal care products (PPCPs) during wastewater treatment, characterizing airborne ARGs at livestock farms, and improving the reference database used to identify MGE-associated proteins. Using a two-step machine-learning framework that combined unsupervised clustering and supervised classification, this work predicted the removal patterns of 149 PPCPs across two full-scale wastewater and water-reuse treatment trains from chemical descriptors. Classification accuracy ranged from 42.5% to 65.2%, depending on the facility and clustering approach. PPCP clusters based on measured removal patterns overlapped by 58–75% with clusters based on physicochemical properties, including Abraham descriptors and logKow. These results show that chemical descriptors can help estimate PPCP removal patterns and prioritize compounds for direct monitoring. Shotgun metagenomics was applied to air and potential source samples collected over four seasons at a dairy farm and a swine farm with contrasting ventilation designs. Clinically relevant ARGs, including putatively plasmid-associated ARGs, were detected in farm-associated air. Total ARG abundance, Rank I ARG abundance, and human health resistome risk were significantly higher at the mechanically ventilated swine farm than at the naturally ventilated dairy farm. Source-tracking analysis identified pen manure as the dominant inferred contributor to airborne ARGs at the dairy farm (64–81%) and dust as the largest inferred contributor among the sources sampled at the swine farm (53–67%). These findings suggest that manure and dust may be important sources of airborne ARGs at livestock farms. Because MGE annotation affects interpretation of ARG mobility, this work also addressed limitations of mobileOG-db, a widely used database of MGE-associated proteins developed at Virginia Tech. In the first version of the database, proteins inherited element-class labels from source databases, which could result in conflicting labels, and the database was composed primarily of sequences from cultured bacteria. mobileOG-db was reorganized into five non-overlapping element classes. A protein large language model classifier, mobileOG-ESM2, was then developed by fine-tuning ESM-2 on these classes. The classifier achieved Matthews correlation coefficients of 0.66 and 0.74 on test sets with maximum sequence identities of 40% and 80%, respectively. Applying the classifier to approximately 288 million proteins from public plasmid and viral databases expanded mobileOG-db 2.0 to 67.15 million protein entries. Together, these studies show how broad chemical analysis and shotgun metagenomics, combined with machine learning and improved reference databases, can provide more information from environmental samples for AMR surveillance. The computational approaches developed in this dissertation complement direct measurement and laboratory validation.
Nitrogen Management in Cotton: Evaluating Early-Season Fertility Practices and Developing an NDVI-Based Nitrogen Recommendation Algorithm for the Southeastern Virginia Coastal Plain
Tate, Cody Brandt (Virginia Tech, 2026-10-07)
Nitrogen (N) management in cotton (Gossypium hirsutum L.) must supply the crop while avoiding excessive vegetative growth, delayed maturity, unnecessary fertilizer expense, and N loss. This dissertation evaluated early-season fertilizer practices and developed and evaluated PredictN, a normalized difference vegetation index (NDVI)-based sidedress N recommendation algorithm for cotton in the southeastern Virginia Coastal Plain. Three studies examined preplant N rate and source, small-plot algorithm performance, and field-scale implementation of algorithm-directed N management. The first study evaluated four preplant N rates (11, 22, 45, and 90 kg N ha⁻¹), two fertilizer sources (fluid and granular), and an unfertilized control across nine site-years from 2021 through 2023. Outcomes were stand establishment, early-season NDVI, and relative lint yield. Higher granular N rates reduced stand establishment, whereas fluid fertilizer generally maintained stands. Preplant N rate did not affect relative lint yield when fertilized treatments received sidedress N to reach a common seasonal total. Applications of 11–22 kg N ha⁻¹ supported relative yields comparable to higher preplant rates. Twenty of 72 NDVI regressions had nominal P ≤ 0.10, confirming that early-season NDVI did not consistently separate preplant N treatments. The second study evaluated PredictN across 12 site-years from 2020 through 2023 using handheld and unmanned aerial vehicle NDVI measurements. Of 24 site-year × N-source combinations, 14 responded to N and 10 were nonresponsive. Thirteen responsive combinations had estimable plateau-onset total N rates of 40–160 kg N ha⁻¹. Stored final-algorithm recommendations retained a modeled 96.1% of plateau lint yield on average. In nonresponsive environments, final total N recommendations ranged from 56 to 105 kg N ha⁻¹ and averaged 55 kg N ha⁻¹ below the Virginia comparison standard of 134.5 kg N ha⁻¹. The final algorithm therefore placed recommendations near fitted yield plateaus where cotton responded to N and reduced recommendations where a positive response was not detected. The third study compared PredictN with grower-standard management across nine field-years in Virginia and northeastern North Carolina from 2022 through 2025. Seven paired lint-yield differences were positive, one was zero, and one was negative; the median gain was 93 kg ha⁻¹ (exact sign-test P = 0.070). PredictN increased N in five field-years and decreased it in four, with a median paired difference of +4.3 kg N ha⁻¹. Median partial return above N fertilizer cost was +$129 ha⁻¹. These results show that PredictN functioned as a flexible sidedress-management tool rather than a fertilizer-reduction program. Together, the studies support 11–22 kg N ha⁻¹ at planting within a split-N program and crop-informed sidedress decisions. PredictN adjusted N in both directions, retained high modeled plateau yield in responsive environments, reduced recommendations in nonresponsive environments, and produced positive observed yield differences in seven of nine on-farm comparisons. Replicated field trials are the next step for measuring performance across a broader range of farms and seasons.
Seamless Integration of Inverter-Based Resources: Cybersecurity, Protection, and Control
Beikbabaei, Milad (Virginia Tech, 2026-10-07)
The number of installed inverter-based resources (IBR) in the U.S. is rising, introducing challenges for their cybersecurity, protection, and control. This dissertation examines the impact of successful false data injection and denial-of-service cyberattacks on IBRs through multiple cyber-physical testbeds and develops methods to detect the attacks and mitigate their adverse effects. This work utilizes deep learning, reinforcement learning, and a physics-informed neural network-based approach to design cyber-resilient grid-forming and grid-following IBRs. Implementing such cyber-resilient control will facilitate the reliable integration of IBRs. Beyond cybersecurity, high IBR penetration requires new protection schemes for several faults, including short-circuit faults, high-impedance faults, and broken-conductor faults. This dissertation develops various protection schemes, including an adaptive short-circuit fault detection scheme for a fully IBR-based grid, a machine learning-based approach that detects the fault, identifies its type, and locates it, a broken conductor detection scheme for the transmission system, and a high-impedance fault detection scheme to prevent wildfires. Existing IBRs may perform poorly as grid conditions change over time, so this work develops a black-box control scheme that operates outside the IBR without modifying internal control settings, improving IBR transient performance. The black-box IBR control is also used for battery energy storage systems for AI data center load-smoothing. These contributions enable reliable integration of IBRs into the grid.
An Examination of Recruitment and Retention of Pregnant Individuals for Neuroimaging and Developmental Studies of Early Life
Chinaka, Oziomachukwu (Virginia Tech, 2026-10-06)
Birthing individuals have long been underrepresented in research, despite initiatives like the National Institutes of Health Revitalization Act advocating for increased inclusion of women and minorities. As research is being conducted in biomedical fields on maternal and infant health to attain equitable health outcomes for all individuals who bear children, studies need to prioritize inclusion of diverse groups from various ethnicities, ages, races, genders, and socioeconomic backgrounds. Current literature in various fields has given recommendations on how to recruit underrepresented participants and insights to barriers that are faced when recruiting and retaining these participants, from the perspectives of underrepresented communities; however, the perspectives of researchers have been understudied and limited in current research. The proposed study aims to conduct a descriptive evaluation examining RandR within the HEALthy Brain and Child Development (HBCD) Study national consortium and various data collection site-level characteristics and procedures that influence RandR from the perspectives of research personnel. The sample size of this study was n=41 HBCD RandR personnel. We conducted semi-structured interviews as the data collection method. We used inductive coding and thematic analysis for the data analysis utilizing Braun and Clake multidirectional model to guide the analysis. For our first study, we identified strategies for recruitment into the HBCD Study. W found 6 emergent themes which were 1) in-person recruitment, 2) outreach, 3) distributing recruitment materials, 4) developing partnerships, 5) word of mouth/referrals, and 6) media recruitment. Peer navigators or support staff were also noted in this study as a key strategy for recruiting pregnant people with a history of substance use. These findings suggest alignment with previous research studies but provide specific context to where these recruitment strategies are occurring and which regions are doing which strategies. The second study looks at barriers and facilitators to RandR. Barriers identified to RandR included 1) time constraints and commitment, 2) distance and transportation, 3) staffing, 4) Procedures and procedural changes, 5) approvals, 6) misalignment with partnerships, 7) HBCD applications not working correctly 8) life circumstances, 9) No show / no communication, , 10) social influences. Facilitators to RandR noted in this study included 1) being a trusted establishment, 2) providing resources and incentives, 3) team characteristics, 4) HBCD RandR personnel characteristics, 5) soft touches, 6) extensions for completing visits, 7) participants feeling that they were contributing to science. These two studies provide information that aids in understanding the context of what is needed for the RandR for pregnant people from diverse backgrounds. The findings in this study also add to the literature as it identifies structural and institutional factors that also impact the RandR process. These factors could not be identified without receiving the perspectives of the HBCD RandR personnel.
Program Comprehension for Human and LLM-Based Software Maintenance
Haroon, Sabaat (Virginia Tech, 2026-10-06)
Modern software systems increasingly rely on abstractions, transformations, and automated tools that make it difficult to understand how source-level programs relate to their actual behavior. Effective debugging and testing therefore depend on accurate program comprehension by both human and automated agents that perform these tasks. This thesis investigates how program comprehension can be improved for developers and systematically assessed for Large Language Models (LLMs) used in software maintenance. The first part focuses on improving human program comprehension in distributed systems. This thesis develops an interactive debugging approach that connects developer-written SQL to its transformed distributed execution by mapping intermediate execution data back to corresponding parts of the original query. This allows developers to understand complex execution behavior using the same abstractions in which the program was originally expressed. As LLMs are increasingly used for coding tasks, program comprehension shifts from human developers to automated agents. The second part investigates LLM program comprehension in automated debugging. The evaluation examines fault localization under semantics-preserving code changes to determine whether LLMs consistently understand program behavior or are influenced by superficial properties of how code is written. The final part extends this investigation to LLM-based test generation under software evolution, evaluating whether generated tests remain reliable when program functionality is preserved and whether they appropriately reflect changes when program behavior is modified. Together, these contributions examine program comprehension from two complementary perspectives: improving developers' understanding of program behavior and evaluating whether LLMs possess the comprehension required for reliable automated debugging and testing.

Browse

Recent Submissions