Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
The Influence of AR Head-Mounted Displays on Spatial Perception and Worker Response in Construction Training
No full textConstruction job sites present significant risks that extend beyond physical hazards to include psychophysiological factors that shape workers' perceptions, attention, and decision-making. The mobility and influence of Augmented Reality Head-Mounted Displays (AR-HMDs) directly interact with these factors by altering how workers perceive their environment, process information, and manage sensory input. While AR-HMDs offer new opportunities for immersive, adaptive training in real-world construction scenarios, they also introduce cognitive and sensory risks that have been insufficiently explored in construction research. Now that AR-HMDs are deployed on construction sites, it's crucial to understand not only their technical performance but also the human-centered impacts on safety, perception, and decision-making. Existing studies have largely overlooked the psychophysiological constraints that influence safety outcomes; yet, understanding these responses is essential for both immediate task performance and long-term learning and risk-taking behaviors. To address these gaps, we develop theoretical constructs and framework variables to evaluate and predict spatiotemporal psychophysiological responses associated with AR-HMD use in construction training, specifically asking: What are the risks associated with AR-HMDs' spatial influence on perception in construction environments?
The framework is developed and tested through four objectives, progressing from a detailed scoping literature review and evaluation of AR-HMD impact on humans to psychophysiological prediction and median severity-of-impact classification using EEG location matrices. Objective 1 develops a conceptual model that classifies AR-HMD and Human-Computer Interaction (HCI) risks and standardizes the domain language for evaluating these technologies, highlighting underexplored cognitive, sensory, and physical human-factor risks. Additionally, objective 1 identifies and classifies spatial perception variables; develops a mental model of how workers perceive and spatially analyze immersive AR-HMD environments; examines embodiment, presence, and spatial presence; and finalizes the theoretical framework for empirical testing. Objective 2 tests the framework in a controlled environment using a full-scale passive and active haptic frame. A within-subjects design captures both psychological (survey-based) and physiological (EEG, heart rate) responses, which are analyzed using Power Spectral Density (PSD), Independent Component Analysis, and regression modeling to identify hemispheric differences and misalignments between perceived safety and actual psychophysiological responses. Objective 3 advances the framework into predictive modeling, using the same haptic-frame environment, deep learning models, including 2D CNN-LSTM sequence modeling and 3D CNN-LSTM architectures that are applied to predict temporal and cognitive state changes from 4D EEG input (frequency, amplitude, time, channels), extending the framework from measurement to prediction. Together, these three objectives show how conceptual modeling, spatial perception analysis, experimental validation, and predictive analytics can be systematically connected to evaluate AR-HMD situational risks. The outcomes reveal the extent of spatial and behavioral influences across key variables, supporting the development of likelihood and severity matrices for academia and industry, as outlined in objective 4.Doctor of PhilosophyConstruction is one of the most hazardous industries, where risks are shaped not only by physical dangers but also by how workers perceive and respond to their environment. Augmented Reality (AR) Head-Mounted Displays (HMDs) are now being introduced on construction sites as mobile tools for training and task support. These devices provide immersive, hands-free instructions directly in the field, creating new opportunities for safer and more efficient learning. At the same time, they may misalign perception with reality, introducing new cognitive, sensory, and decision-making risks that remain insufficiently understood.
This dissertation examines these effects by focusing on how the left and right hemispheres of the brain respond differently during AR-HMD use. The left hemisphere is often associated with analytical, sequential, and rule-based processing, while the right hemisphere supports spatial awareness, attention, and holistic integration. In immersive AR training, workers must draw on both left- and right-hemisphere processing while navigating 3D spaces. If AR-HMD design disproportionately loads one hemisphere, it may create safety risks: workers focused too narrowly on a task may overlook hazards. At the same time, those immersed in spatial imagery may misjudge sequences or measurements.
To address these gaps, we develop theoretical constructs and framework variables to evaluate and predict spatiotemporal responses associated with AR-HMD use in construction training, specifically asking: What situational risks are associated with AR-HMDs' spatial influence on perception in construction environments? To investigate, this research integrates surveys with moment-to-moment recordings of brain activity and heart rate, capturing the contrast between what workers believe happened (subjective perception) and what their body reveals actually happened (physiological response). Guided by theories such as Risk Homeostasis and Human Factors, a framework is developed to explore under-examined issues, including spatial presence, situational awareness, and sensory misalignment. Finally, advanced 2D and 3D deep learning models are applied to classify and predict cognitive states with 67%-90% accuracy, demonstrating that worker responses can be measured and predicted.
The results highlight both the promise and the risks of AR-HMDs in construction. On the one hand, immersive and haptic-based training can enhance engagement, skill development, and attentional focus; on the other hand, hemispheric cognitive differences can reveal vulnerabilities that, if overlooked, may compromise safety. By developing a holistic framework that integrates human perception, psychophysiological responses, and predictive modeling, this research lays the foundation for safer AR-HMD systems, more effective training methods, and enhanced protections for workers in an industry where risks remain high
AI
No full textReservoir computing (RC) has emerged as an energy-efficient paradigm for temporal information processing, offering reduced training complexity by fixing recurrent dynamics and training only a simple readout layer. Among RC models, Echo State Networks (ESNs) and Liquid State Machines (LSMs) represent two distinct approaches based on continuous-valued and spiking neural dynamics, respectively. In this work, we present a comparative evaluation of ESNs and LSMs on the Mackey–Glass chaotic time-series prediction task, with emphasis on scalability, overfitting behavior, and robustness to reduced numerical error precision. Experimental results show that ESNs achieve lower prediction error with relatively small reservoirs but exhibit early performance saturation and signs of overfitting as reservoir size increases. In contrast, LSMs demonstrate more consistent generalization with increasing reservoir size and maintain stable performance under aggressive reservoir quantization. These findings highlight fundamental trade-offs between accuracy and hardware efficiency, and suggest that spiking RC models are well suited for energy-constrained and neuromorphic computing applications.Published versio
Determining the mechanism of protein transit through the peptidoglycan layer in Gram-positive bacteria and identifying phage-docking sites on Clostridium perfringens
No full textClostridium perfringens is a Gram-positive (Gram+) anaerobic spore-forming bacterium that can act as a lethal human pathogen. Depending on the pathway of entry and strain, infection can lead to gastrointestinal disease or gas gangrene. Disease, as with nearly all bacterial pathogens, is primarily the result of secreted proteins, better known as exotoxins. These toxins can aid in the breakdown of host tissue leading to increased nutrient availability for the pathogen. Toxins of particular interest include collagenase, perfringolysin O, and phospholipase C. Most toxins, assuming they are not directly injected into host cells, need to be secreted. This is no trivial task, as the Gram+ cell wall (the outermost structure) is a very thick, dense, mesh-like screen comprised primarily of peptidoglycan (PG) and teichoic acid side chains. Due to this barrier, which ordinarily benefits the bacterium in both defense and resisting osmotic stress, secretion can be an issue. Globular proteins smaller than 25 kDa have been found to passively move through the PG, but most toxins and larger macromolecules exceed this size threshold. A solution to this issue could be either a temporary opening in the PG layer or a designated protein-lined channel for this event to occur. In order to determine this mechanism, the aforementioned toxins come into play as they are all between 45-120 kilodaltons (kDa) in size. Each is passed through the cytoplasmic membrane via the general secretion system (Sec) in a linear fashion. From there it is highly supported that proper toxin folding occurs in the periplasmic-like space. Following this step, the cell wall secretion mechanism has yet to be identified. Such a mechanism could be a target for antimicrobials rendering the cell avirulent, but viable.
The strain used in this project, HN13, is both non-sporulating and only has two characterized secretion pathways (Sec and pilus-dependent). Random transposon mutagenesis and several negative selection rounds were used to construct multiple mariner mutant libraries. These libraries were screened on specific agar to visualize respective toxin secretion with approximately 200,000 mutant colonies screened. From the initial secretion screen mutants undergo an additional secondary screen through which only 186 mutants of interest were found (<0.1% identification rate). Through the use of various genetic techniques, 28 genes have been identified that lead to the lack-of-secretion phenotype. These genes encode various proteins including transcription regulators, sporulation factors, two-component regulatory systems, and potentially a protein which may play a role in secretion during cell division. Through genetic complementation/knockout trials the last major unknown physiological trait of Gram+ bacteria, protein secretion, has been studied here.
C. perfringens is the causative agent of gastrointestinal disease and gas gangrene. Gastrointestinal infection is typically self-limiting but may be treated with antibiotics which can lead to unintended clearing of normal healthy microflora. As for gangrenous infection, if left untreated it is 100% fatal. Even with swift treatment, amputation may still be required due to the rapid spreading of bacteria in tissue (cm/hour). Antibiotic treatment is nearly ineffective due to induced blood clotting. This leads to an anaerobic environment where most antibiotics will either not be able to function or even reach the cells. For both routes of infection, bacteriophages could supplement or entirely replace antibiotics. This could help greatly reduce the amount of collateral damage to the microbiome in gastrointestinal infection due to high phage specificity. A more speculatory approach is administering phages to tissue surrounding gangrene infection to diminish bacterial spread and density. Phages used as an alternative treatment could also help slow the observance of antibiotic resistance seen throughout the medical industry. In order to develop an effective phage therapy, phage docking sites need to be identified. Docking sites would allow for implementation of phage engineering and aid in the determination of untested phages that would be predicted to bind. Phage therapy may be a promising alternative treatment method to better treat both types of C. perfringens infections.Master of ScienceClostridium perfringens is a bacterium that infects humans resulting in either lethal gas gangrene or more commonly gastrointestinal disease. Most bacteria are classified by their Gram reaction being either positive or negative. Gram-positive bacteria, such as C. perfringens have a characteristically thick outer layer known as the cell wall. This structure is a dense mesh-like formation that aids the organism in resisting osmotic stress and providing overall rigidity. Due to its thickness, secretion of material through the cell wall has remained a mystery. This is important as many pathogenic bacteria secrete toxins that damage their host causing disease. If this mechanism could be identified it could be targeted by antimicrobial drugs rendering the bacterium essentially non-pathogenic. In an attempt, to determine this secretion mechanism bacteria that were unable to secrete toxins were sequenced to identify genes relevant in the secretion process. By identifying the genes, the inferred proteins they encode can then be investigated. This should ultimately lead to the identification of a protein structure(s) that allows for the secretion process to occur. Thus far, genes are mainly relevant to toxin synthesis, but one important for cellular division may be key. In order to understand the role of this gene it has been removed from a "normal" bacterium to observe the affects. From there, assays can be used to directly quantify the amount of toxin that is secreted. The hope is to identify and target this mechanism to better treat C. perfringens infections and hopefully other Gram-positive pathogenic bacteria.
Additionally, there has been investigation into the predators of bacteria known as bacteriophages. These are viruses that infect, replicate, and destroy their bacterial host. In order for this event to occur phages must bind somewhere on the bacterial surface. This binding occurs on a very specific place on the bacterium. It is not well understood where phages bind on some Gram-positive bacteria including C. perfringens. The thought is phages could serve as an alternative treatment method to antibiotics for C. perfringens and perhaps other Gram-positive bacteria. This has become relevant as of late as many bacteria are showing signs of antibiotic resistance, which phages can circumvent. The method is similar to the one mentioned previously, mariner transposon mutants that are able to survive in an environment with phages must not have the binding site present and therefore do not have the proper functional gene. These bacteria can then be sequenced and the genes that encode for binding sites of phage could be identified by determining when genes have the transposon insertion. Phages could then be modified to have these binding site receptors present through tail fiber modification or novel phages could be predicted to help treat C. perfringens infection
Journal of Cybersecurity and Privacy
No full textCybersquatting is a collection of methods commonly used by malicious actors to mislead or trick internet users into accessing fraudulent or malicious content. Much of the current research has concentrated on the specific techniques used by attackers in this domain, such as typosquatting, combosquatting, and sound squatting. Some research has explored the financial and time impacts of cybersquatting; however, an understanding of user privacy impacts is limited. Prior research into privacy implications has primarily relied on passive techniques such as analyzing DNS records, HTML content, and domain registrations. These passive approaches limit the ability to interact with these domains and track the downstream impact of sharing personally identifiable information (PII). This research develops an active open-source intelligence (OSINT) collection system capable of rapidly collecting and analyzing squatting domains through both passive and active techniques, with a particular emphasis on identifying those that solicit user information. Synthetic identities are then registered with these domains, and their associated communications are collected and analyzed to identify privacy-related risks and determine whether shared PII propagates.Published versio
IEEE Robotics and Automation Letters
No full textModel predictive control (MPC) with reduced-order template models has proven effective for dynamic legged locomotion, but loco-manipulation introduces additional complexity requiring efficient algorithms for high-degree-of-freedom (DoF) systems. This letter presents a computationally efficient nonlinear MPC (NMPC) framework tailored for loco-manipulation tasks of quadrupedal robots equipped with robotic manipulators whose dynamics are non-negligible relative to those of the quadruped. The proposed framework adopts a decomposition strategy that couples locomotion template models—such as the single rigid body model—with a full-order dynamic model of the robotic manipulator for torque-level control. This decomposition enables efficient real-time solution of the NMPC problem in a receding horizon fashion. The optimal state and input trajectories generated by the NMPC for locomotion are tracked by a low-level nonlinear whole-body controller, while the optimal torque commands for the manipulator are directly applied. The layered control architecture is validated through extensive numerical simulations and hardware experiments on a 15-kg Go2 quadrupedal robot augmented with a 4.4-kg 4-DoF Kinova arm. Given that the Kinova arm dynamics are non-negligible relative to the Go2 base, the proposed NMPC framework demonstrates robust stability in performing diverse loco-manipulation tasks, effectively handling external disturbances, payload variations, and uneven terrain.Accepted versio
BMC Infectious Diseases
No full textBackground: Sample pooling is a critical strategy to meet increased testing demand and conserve resources in surveillance testing. Much of its effectiveness depends on how well optimized the pool size is to the prevalence of infection in the sampled population, which can be difficult to anticipate in many circumstances. Multiple methods exist to better optimize pooling, with unique trade-offs.
Methods: Pooling optimization methods were simulated to examine trade-offs between surveillance priorities and operational characteristics using SARS-CoV-2 surveillance data and workflows generated by the Virginia Tech Molecular Diagnostics Laboratory under varying capacity conditions. All in-house validation procedures were designed and established exclusively under CLIA to ensure full control of the analytical framework and to accurately reflect true capacity constraints. We used binary surveillance data to run Monte Carlo simulations (MCS) comparing conservative and large fixed pools, historical prevalence optimization (HPO), prevalence estimation testing (PET), truly optimized pooling, and individual testing. Median test counts from the MCS fed a discrete-event simulation (DES) that assessed processing time at different lab capacities under surveillance and outbreak conditions. We then used the combined performance results to build a classification tree to guide method selection under different testing priorities and constraints.
Results: MCS results indicated that small pools (4 samples), HPO, and PET resulted in test counts that were not statistically different from truly optimized pooling (p > 0.05). The DES showed that pooling methods generally performed comparably to individual testing in processing time at low laboratory capacity, but individual testing became faster as capacity increased. Across capacity conditions, individual testing processed fewer than 500 daily samples more quickly, yet it demanded more hands-on time than pooling. Large-scale surveillance favored pooled methods, which were quicker under most conditions, while outbreak scenarios often favored individual testing when capacity wasn’t highly limited. Machine learning analysis highlighted surveillance priorities and sample intake as key determinants in selecting the best pooling optimization method for the given circumstance.
Conclusion: This study demonstrates the importance of maintaining multiple pooling optimization approaches and adapting strategies to match evolving demands and potential constraints in surveillance testing.Published versio
Journal of Marine Science Engineering
No full textRim-driven propellers (RDPs) have attracted renewed attention as an efficient propulsion concept for integrated electric propulsion systems, yet their structural configuration inherently limits duct geometry modification, and viscous losses associated with boundary layer separation near the duct trailing edge remain a key performance constraint. In this study, a vortex generator-based flow control strategy is proposed as a practical means of improving RDP performance without altering the duct geometry. Reynolds-averaged Navier–Stokes (RANS) simulations were conducted to examine the effects of vortex generators installed on the outer surface of the duct, with numerical reliability ensured through a grid convergence index (GCI) analysis. A steady-state multiple reference frame (MRF) approach was employed, and the resulting flow characteristics were analyzed using velocity profiles, line integral convolution (LIC) visualization, pressure field analysis, and distribution of the flow field in the wake. The results show that the vortex generators effectively delay boundary layer separation near the duct trailing edge by re-energizing the near-wall flow, thereby enhancing flow attachment and pressure recovery. Consequently, consistent improvements in thrust coefficient and propulsive efficiency are achieved over the entire range of advance ratios, while the increase in torque coefficient remains negligible. These findings demonstrate that vortex generator-based flow control offers a practical and effective approach for enhancing the open-water performance of rim-driven propellers under structural constraints.Published versio
Nature Water
No full textA major barrier to preventing per- and polyfluoroalkyl substances (PFAS) from entering drinking water supplies is identifying and quantifying their upstream sources, particularly in One Water systems that integrate diverse water inputs. Here we combine high-frequency measurements with mass-balance analysis to quantify PFAS and major-ion loading to the Occoquan Reservoir, a drinkingwater supply serving one million people in Northern Virginia, USA. Mass-balance analysis at the confluence of watershed inflows and treated wastewater inputs reveals seasonally varying contributions from domestic wastewater, watershed runoff, and a single significant industrial user (SIU) of the sanitary sewer system. A one-month, system-scale diversion of SIU effluent confirms this source attribution for several short-chain PFAS and major ions, with concentration deficits closely matching withheld mass. These results demonstrate that traditional mass-balance approaches can inform collaborative management of PFAS contamination in One Water systems.Submitted versio
Nature Communications
No full textSelf-propelled droplet jumping has widespread applications in surface cleaning, condensation heat transfer, hydrogen production, and triboelectric nanogenerator due to the passive yet effective cross-interface transfer of mass, momentum, energy and charge, whose rates generally increase with droplet size. However, as droplet size increases, gravity inevitably impedes droplet’s mobility, imposing a capillary length constraint of 2.7mm for water droplet, beyond which self-propelled jumping remains a persistent challenge. Here, we report passive jumping of water puddle in the unprecedented centimeter scale from a superhydrophobic surface through the burst of an entrained bubble, breaking the capillary length limitation for droplet jumping. By virtue of direct and localized impact at droplet base, the bubble-burst-induced capillary waves play a paradigm-shifting role in shortening the impact duration, depressing droplet spreading, and facilitating momentum transfer. With >90% conversion to droplet jumping momentum, the impacting momentum of capillary waves scales linearly while droplet jumping height scales quadratically with bubble radius. Through studying the synergistic interplay between bubble bursting, fluidic jetting and droplet jumping, this work reveals a previously unexplored mechanism of capillary wave impact in fluid-structure interactions and offers a promising strategy for droplet actuations and the directional printing of particles in additive manufacturing.Published versio
Challenges and Opportunities in Secure Real-Time Digital Twin Systems for AR/VR-Enabled Smart Factories
No full textModern high-efficiency factories face challenges in workforce training, real-time remote assistance, and safe, efficient coordination of automated guided vehicles (AGVs). However, realizing such tightly coupled human–machine collaboration introduces a number of challenges in real-time coordination, localization, network determinism, cybersecurity and cross-domain data consistency across cyber–physical layers. This paper discusses the challenges and opportunities in developing an integrated real-time digital twin control architecture that unifies augmented reality-enabled and virtual reality-enabled operator support with safety-aware orchestration of AGV fleets. Achieving centimeter-level localization and stable AR/VR overlays requires a deterministic timing architecture that combines precise clock alignment via the Precision Time Protocol (PTP) and IEEE 802.1AS with resilient data fusion across programmable logic controllers (PLCs), real-time locating systems (RTLS), and AGV telemetry. In addition, such highly networked digital twin ecosystems require end-to-end cybersecurity mechanisms that ensure confidentiality, integrity, and availability from edge devices to cloud services. Secure time synchronization, authenticated data exchange, and runtime anomaly detection become crucial to prevent cascading effects from cyber intrusions to physical operations. On the opportunity side, the convergence of edge computing, time-sensitive networking, and semantic digital twin opens new possibilities for resilient, trustworthy industrial automation system. Emerging research directions include adaptive virtual fencing, dynamic path replanning, security-aware scheduling, and AR/VR feedback loops that continuously align the physical and virtual worlds. Finally, we outline a vision for human-centric, secure, and real-time digital twin ecosystems that jointly optimize AR/VR experience quality, motion-to-photon latency, and industrial safety, thereby paving the way for resilient and data-driven factories of the future.Published versio