Theses and Dissertations
Permanent URI for this collection
Browse
Search Collection
1 - 5 of 33845
-
ThesisDevelopment of a precast UHPC seat-type abutment for seismic applications( 2024-12)About 50% of bridges in the USA are more than 50 years or older and a significant number (47000 bridges) are deemed structurally deficient. Compounded by continuously increasing traffic demands, these bridges face accelerated deterioration, necessitating more frequent maintenance, partial rehabilitation, or even complete replacement. However, the partial or total closure required for such activities poses significant disruptions to the traveling public and impacts local economies, given the critical role bridges play in daily societal operations. This situation underscores the necessity for Accelerated Bridge Construction (ABC) methods that seek innovative ways to hasten onsite construction. Various ABC methodologies, such as the use of structural precast elements and modular bridge systems, have been developed and implemented across the United States. Focusing largely on superstructures, these efforts aim to streamline onsite construction processes, thereby enhancing overall transportation efficiency. For substructures, the ABC methods are primarily in the experimental stage, with most projects presenting one innovative idea after another but lacking in standardization. The significant challenge within ABC in substructures lies in handling heavy precast elements, introducing transportation challenges, and requiring large cranes for on-site assembly. Thus, this thesis proposes innovative solutions for precast elements in the foundation by addressing the lack of standardization and the need for reducing the number of robust connections among components transported in pieces. To reduce costs and improve transportability, the concept of hollow precast substructures was proposed in this study, focusing on investigating the ABC design of seat-type bridge abutments frequently used in seismic regions utilizing ultra-high-performance concrete (UHPC). By employing UHPC, it becomes feasible to reduce the sectional dimensions, enabling the transportation of the main abutment component as a single, sizable unit that would require a reduced number of connections in the field. The proposed abutment stem adopts a hollow section with a 6-inch thickness, incorporating a hammer head at the top to meet the 30-inch seat width requirement outlined in California Department of Transportation (Caltrans) Seismic Design Criteria (SDC). The precast abutment cap features corrugated steel pipe sockets filled with UHPC cast-in-place closure pours to ensure rigid connections with steel H piles. The performance of the new system and its components will be validated by a half-scale experimental test unit. It is anticipated that Caltrans will implement this detail in constructing standard ordinary bridges with two or three spans. This study also explores another hybrid approach to implementing ABC in substructures through the experimental use of thin, open-hollow precast UHPC shell members. These lightweight elements are designed to improve handling and mobility both before and during assembly. Serving as stay-in-place (SIP) formwork, these shells eliminate the need for onsite formwork installation—a process that is time-consuming and dependent on labor availability at the bridge site. The experimental arrangement included straight and battered steel piles for the foundation, with an open hollow precast thin shell unit used for the pile cap, the primary substructure component tested. A top-loading block shell was also incorporated to apply loads to the system, along with rebar cages for each component. Another lateral load block was constructed directly at the top of the pile cap. The test unit was subjected to combined vertical and lateral loads that simulated both normal operating conditions and extreme events for two-lane bridges in Iowa and California. Despite these demanding conditions, the battered and vertical pile-pile cap system demonstrated exceptional resilience, maintaining its integrity and full elasticity even under unrealistic severe loads. The employment of UHPC SIP formwork played a critical role in preventing microcrack development and widening within the regular concrete, thereby enhancing the system’s elasticity and structural integrity. Additionally, the connections between the piles and the pile cap showed excellent structural stability, with no evidence of pullout, pile yielding or buckling. The observed load-displacement behavior was mostly linear, and the slight nonlinearity was primarily due to gap formation in the surrounding soil. This paper provides an extensive overview of the outdoor test setup and its pivotal findings, highlighting the effectiveness battered steel piles in the foundation serving to perform much adequately than anticipated under extreme seismic loading and the role of UHPC SIP formwork in improving the performance and durability of bridge foundations.
-
DissertationFrom pathway dissection to developing genetic tools: A focus on Fyn driven neurodegeneration, STAT3-dependent neuroinflammation, and caspase conditional alleles( 2024-12)Neurodegenerative diseases such as Alzheimer’s and Parkinson’s are characterized by progressive loss of specific neuronal populations and the induction of inflammatory responses, resulting in severe cognitive and motor impairments. Understanding the cellular and molecular mechanisms underlying these diseases remains a critical challenge. My dissertation focuses on investigating the role of Fyn kinase in neurodegeneration and developing innovative zebrafish models and genetic tools to study the complex interactions between inflammation, mitochondrial dysfunction, and gene regulation in these diseases. In Chapter 2, a zebrafish model was developed to explore the in vivo effects of constitutively active FynY531F signaling, a known risk factor in neurodegeneration. The model utilized neural-specific Gal4 expression of FynY531F, allowing targeted and sustained activation of Fyn in zebrafish neurons. Live imaging in zebrafish larvae revealed significant dopaminergic neuron loss, along with notable mitochondrial aggregation in the brain. Concomitant with neuronal loss, microglial activation was observed, which correlated with a significant upregulation of pro-inflammatory cytokines, including tnfα, IL-1β, and IL-12α. These findings indicate that FynY531F signaling drives dopaminergic neurodegeneration, likely mediated by mitochondrial dysfunction and an inflammatory response. Importantly, chemical inhibition experiments showed that this neurodegenerative process was dependent on the activation of the NF-κB pathway, highlighting NF-κB as a critical effector in Fyn-induced neurotoxicity. Building upon these findings, Chapter 3 delves deeper into the downstream signaling pathways involved in Fyn-mediated neurodegeneration. Transcriptomic analysis of the FynY531F zebrafish model identified STAT3 as a key downstream target of Fyn signaling. Experimental inhibition confirmed that activation of both STAT3 and NF-κB pathways was necessary for Fyn-induced neurodegeneration and inflammation. Moreover, dual inhibition experiments revealed a synergistic relationship between STAT3 and NF-κB, suggesting that these pathways collaboratively promote dopaminergic neuron loss. This chapter uncovers STAT3 as a novel effector of Fyn signaling, shedding light on its potential as a therapeutic target for modulating neuroinflammatory and neurodegenerative pathways. The focus of Chapter 4 shifts to the development of innovative genetic tools for precise spatial and temporal control of gene activity. Although the Cre/lox recombinase system is widely used in zebrafish for conditional gene manipulation, challenges remain in isolating stable and efficient Cre/lox-regulated alleles. To address these limitations, the GeneWeld CRISPR-Cas9 integration strategy was employed to generate robust floxed alleles for conditional gene inactivation. A universal targeting vector, UFlip, containing short homology arms flanking a floxed 2A-mRFP gene trap, was integrated into the caspa locus in zebrafish. The resulting caspaoff/+ allele demonstrated strong mRFP expression and accurately recapitulated loss-of-function phenotypes. This approach provides a streamlined and versatile toolkit for generating Cre/lox-responsive zebrafish alleles, enabling more precise studies of gene function in various developmental and disease contexts. Overall, this dissertation integrates advanced zebrafish models and genetic tools to unravel the complex mechanisms underlying neurodegeneration. By leveraging the strengths of conditional gene manipulation and transcriptomic analysis, the findings presented herein contribute significant insights into the role of Fyn kinase in driving neuroinflammatory and neurodegenerative processes. The work not only advances fundamental knowledge of these diseases but also offers promising avenues for future therapeutic interventions aimed at modulating key signaling pathways implicated in neurodegeneration.
-
DissertationScalable and resource-efficient federated learning: Techniques for resource-constrained heterogeneous systems( 2024-12)Federated Learning (FL) has emerged as a promising paradigm for decentralized model training, enabling collaborative learning across distributed edge devices while preserving user data privacy. This dissertation presents a series of contributions that address key challenges in FL, including resource constraints, communication efficiency, model optimization, and data heterogeneity. First, we propose SPATL, a salient parameter aggregation and transfer learning approach, which reduces communication overhead and accelerates model inference in heterogeneous FL environments. SPATL achieves up to 86.45% communication reduction and improves local inference efficiency through a shared encoder and local predictor architecture. Next, we extend this work by proposing Resource-Aware Federated Learning (RaFL), a framework combining Neural Architecture Search (NAS) with FL to optimize model performance for resource-constrained edge devices. RaFL enables heterogeneous model deployment across clients, enhancing resource utilization while preserving the integrity of distributed learning. Finally, we explore the integration of Foundation Models (FMs) into FL, introducing the Federated Foundation Models (FFMs) paradigm. This approach preserves data privacy during the training and fine-tuning of large pre-trained models such as BERT and GPT, while facilitating scalable, privacy-preserving learning across heterogeneous clients. Together, these contributions advance the state-of-the-art in FL by addressing critical challenges related to communication efficiency, model optimization, and resource heterogeneity, paving the way for scalable and privacy-preserving collaborative learning across distributed systems.
-
DissertationAssessing gait activities and balance control with the use of wearable sensors and deep learning model( 2024-12)ABSTRACT This dissertation investigates the application of inertial measurement units (IMUs) for assessing gait balance control and explores the integration of IMUs and machine learning to classify gait activities. Wearable sensor technology, particularly IMUs, has gained attention due to its portability, cost-effectiveness, and ability to capture precise motion data in real-world settings. With the growing need to prevent falls and maintain mobility in aging populations, the role of IMUs in gait analysis is promising. The research presented in this dissertation explores several aspects of IMU-based gait assessment, comparing IMU performance with traditional motion capture systems, applying machine learning models for gait activity classification and recognition, and improving conventional clinical tests to better evaluate fall risk. A narrative review sets the foundation by summarizing existing work on using IMUs to assess balance control. The dissertation continues with experimental studies comparing IMU data to gold-standard methods, applying deep learning to advanced gait data, and enhancing the Timed-Up-and-Go (TUG) test using IMU data. These studies demonstrate the potential of IMUs in quantifying balance control and classifying gait tasks with a deep learning model. The results highlight the strong correlations between IMU-based and traditional measurements of biomechanical data, suggesting that advanced machine learning models can recognize various gait tasks and can segment the subtask to reveal the potential issue underlying the inferior performance. The findings emphasize the benefits of wearable sensors in both clinical and everyday environments and emphasize the need for further research to address the accuracy and usability. The dissertation concludes with recommendations for future research, including exploring the integration of artificial intelligence, digital health platforms, and smart home technologies with IMUs to enable continuous, real-time monitoring of balance control and mobility. These advancements could lead to more effective interventions and prevention strategies for individuals at risk of falls.
-
ThesisThe essential role of soil maps to support precision agriculture( 2024-12)Soil mapping is an essential tool for precision agriculture users. Although many studies emphasize modeling the spatial variation of different soil properties, the temporal variation is usually underestimated. Therefore, this study focuses on the spatial and temporal variation of soil fertility properties at the field scale and evaluates different methods to improve soil maps and their implementation in precision agriculture. Soil samples were collected monthly from a field in Central Iowa over two consecutive growing seasons following a maize (Zea mays) – soybean (Glycine max) rotation. Significant temporal changes were observed for soil nitrate, phosphorus, potassium, organic matter, and pH. Spatial variability was described across hillslope positions and the data’s spatial autocorrelation was evaluated with geostatistics. Soil maps were then created with ordinary kriging, and a time sequence was constructed for each soil property. Alternatively, machine learning was used to create soil fertility maps using remote sensing spectral data and topographic attributes as predictor variables (covariates). Finally, a method to delineate and validate management zones for precision agriculture was evaluated utilizing unsupervised classification algorithms, soil maps, and spectral data. The results showed that spectral data can describe sources of temporal variability that topography alone cannot explain, such as the impact of management practices. However, there was not an absolute best combination of covariates to construct spatial models to create soil fertility maps or to delineate management zones.