Theses and Dissertations
Permanent URI for this collection
Browse
Search Collection
1 - 5 of 33772
-
DissertationA machine learning study of wind-driven runback/flow-off multiphase flows pertinent to aircraft icing phenomena( 2024-12)In-flight icing and ground icing are widely recognized hazards that have a substantial impact on the safety of aircraft during the processes of takeoff, cruising, and landing in cold weather conditions. The unsteady runback behavior of wind-driven runback water film (WDRWF) flows over aircraft surfaces has a significant impact on the aircraft in-flight icing process. Conventional theoretical/numerical methods cannot accurately predict WDRWF flow due to the limited comprehension of the intricate multiphase interactions among freestream airflow, water film motion, and solid airframe surface. Machine learning methodologies can effectively capture intricate physics phenomena through data assimilation, rendering them a compelling substitute for conventional approaches. In the present study, a deep-learning framework ConvLSTM-AE is developed to forecast the intricate spatial-temporal progression of an experimental multiphase WDRWF flow on a flat plate, considering different water flow rates and wind speeds. To predict the WDRWF flow with a long-time evolution without depending on previous ground truth, a physics-guided Fourier neural operator is developed. Then the nonlinear dynamics of the experimental WDRWF flow are identified by utilizing an interpretable data-driven framework known as sparse identification of nonlinear dynamics. Moreover, to extract new physical properties from the experimental WDRWF flow dataset, a physics-informed neural network is utilized to inversely forecast the interfacial shear stress field from thickness field. The last section of the study shifts its attention to the deicing fluids flow-off process during aircraft take-off. Deicing fluids employed on aircraft surfaces serve the purpose of mitigating ice accumulation; however, their presence might have adverse effects on aerodynamic performance during the take-off phase. Consequently, a thorough experimental investigation is conducted to characterize the wind-driven flow-off process of Newtonian deicing fluid over a flat surface utilizing an innovative Digital Image Projection (DIP) technology. In summary, machine learning techniques are proved to be able to forecast complex multiphase flow features related to aircraft icing and gain new physical understanding from current experimental datasets. The experimental study on deicing fluid flow-off can improve ground icing strategy optimization and provide unique datasets for machine learning approaches to predict the flow-off process.
-
DissertationCalculation of two and three nucleon observables using chiral effective field theory and quantum computing( 2024-12)Chiral effective field theory offers a framework for obtaining internucleon interactions in a systematically improvable fashion from first principles while also providing for the derivation of consistent electroweak current operators. In this work, we use consistently derived interactions and currents to calculate nuclear observables for two- and three-nucleon systems, specifically the magnetic dipole moments of the A = 3 nuclei H-3 and He-3, and the electric dipole polarizability of the A = 2 system (deuteron). We focus on semilocal coordinate-space (SCS) regularized LENPIC interactions. Starting from the momentum-space form of the LENPIC chiral effective field theory vector current, we derive the SCS-regularized magnetic dipole operator up to N2LO. Using this operator, we perform no-core shell model calculations for the H-3 and He-3 nuclei with SCS LENPIC interactions at N2LO in chiral effective field theory and evaluate the magnetic dipole moments obtained using the consistently derived single-nucleon and two-nucleon electromagnetic currents. As anticipated by prior results with chiral effective field theory currents, the current corrections through N2LO provide improved but not complete agreement with the experiment for the H-3 and He-3 magnetic dipole moments. We extracted the electric dipole polarizability of the deuteron using SCS-regulated LENPIC nucleon-nucleon interactions up to N4LO with chirally-improved magnetic and electric dipole operators from augmented photo-absorption data and using the inverse-square energy-weighted photo-nuclear sum rule. Our calculations, consistent up to N2LO, were compared with results from the Reid Soft Core 68 nucleon-nucleon interaction and with old photo-absorption data, showing agreement and reduced uncertainty in our results. A direct theoretical calculation of the deuteron’s polarizability using LENPIC interactions also yielded predictions within the range of other nucleon-nucleon interactions, although most results, including ours, fall outside the experimental range reported in Ref. [1]. While including new data can reduce the uncertainty of deuteron polarizability extraction, an improved agreement between extracted and theoretical predictions using LENPIC interactions would require consistent incorporation of higher-order chiral corrections to the dipole operators. We investigated the application of quantum computing to calculate nuclear physics observables, with a particular focus on the evaluation of nuclear transitions. Specifically, we examined the Gamow-Teller beta decay transition from the ground state of a neutron-neutron system to the ground state of a neutron-proton system. We employed the SCS-regulated LENPIC two-nucleon N2LO interaction, with nucleons confined using a harmonic oscillator (HO) potential. The matrix representations of the initial and final Hamiltonians and the transition operator are constructed using a three-dimensional HO basis. We used the Okubo-Lee-Suzuki transformation method to unitarily transform the model problem onto a smaller model space before evaluating the transition amplitude on a quantum computer. We utilize compact encoding techniques to map Hamiltonians and transition operators onto the quantum computer and adopt hardware-efficient ansatz for constructing trial parameterized states. We determined the ground states of the initial and final Hamiltonians using the variational quantum eigensolver with both ideal statevector and shot-based simulators, employing various classical optimizers. We then calculated the transition amplitude on the quantum computer using a quantum circuit adapted from Ref. [2], again leveraging ideal statevector and shot-based simulators. Our results demonstrate that the transition amplitudes calculated using quantum computing can reproduce classical results with high accuracy. This work demonstrates the reliability and potential applicability of quantum computing for evaluating low-energy nuclear physics observables, particularly transition amplitudes. While it does not seek to demonstrate quantum advantage, the study provides a foundational understanding of how quantum computational techniques can be applied to nuclear transitions, laying the groundwork for future research in this domain.
-
DissertationInformation extraction with weak supervision( 2024-12)This dissertation explores the development and application of weak supervision techniques to address key challenges in three fundamental information extraction (IE) tasks: Named Entity Recognition (NER), Relation Extraction (RE), and Entity Linking (EL). Traditional supervised learning methods in these domains often require extensive human annotations, which are costly and time-consuming, limiting their scalability and applicability in real-world scenarios. To overcome these limitations, this research introduces innovative weakly supervised methodologies for each of these tasks, aiming to reduce reliance on manual labeling while maintaining high performance. The first part of the dissertation presents a novel framework, Confidence-Based Multi-Class Positive and Unlabeled (Conf-MPU) learning, designed to enhance the performance of distantly supervised NER. By incorporating confidence scores into a multi-class PU learning approach, Conf-MPU effectively handles incomplete labeling and varying false negative rates inherent in distantly supervised data. Experimental results on benchmark datasets demonstrate that Conf-MPU significantly outperforms existing state-of-the-art methods, advancing the field of distantly supervised NER. The second part focuses on improving Relation Extraction through the integration of indirect supervision. A novel approach, DSRE-NLI, is introduced, which leverages a Natural Language Inference (NLI) engine and a Semi-Automatic Relation Verbalization (SARV) mechanism to diagnose and mitigate label noise in distantly supervised RE tasks. This method enhances the semantic diversity of relation templates with minimal human input, resulting in a significant performance boost over traditional distantly supervised methods on real and simulated datasets. The third part of the dissertation addresses challenges in Zero-Shot Entity Linking (ZSEL) with a new re-ranking approach, GenDecider, which incorporates “None of the Candidates” (NoC) judgments into the re-ranking process. By formulating the task as a generative process using the Llama model, GenDecider effectively detects scenarios where the correct entity is not among the retrieved candidates. This approach significantly improves the accuracy and reliability of ZSEL systems, as evidenced by its performance on the benchmark ZESHEL dataset. Collectively, the contributions of this dissertation lie in advancing weak supervision techniques across three critical IE tasks, reducing the dependency on extensive manual annotations, and improving the robustness and scalability of information extraction systems. The findings have broad implications for the development of practical, scalable IE solutions in data-rich environments. Future research directions include refining noise-handling mechanisms, optimizing computational efficiency, and expanding the proposed methods to multilingual and low-resource settings.
-
ThesisAnalyzing redundancy in code-trained language models( 2024-12)Code-trained language models have proven to be highly effective for various code intelligence tasks. However, they can be challenging to train and deploy due to computational bottlenecks and memory constraints. Implementing effective strategies to address these issues requires a better understanding of these ’black box’ models. In this paper, I perform a neuron-level analysis of code-trained language models on three different software engineering and one high performance computing downstream task. I identify important neurons within latent representations by eliminating neurons that are highly similar or irrelevant to the given task. This approach helps us understand which neurons and layers can be eliminated (redundancy analysis) and where important code properties are located within the network (concept analysis). We find that over 95% of the neurons can be eliminated without significant loss in accuracy for our code intelligence tasks. We also discover several compositions of neurons that can make predictions with baseline accuracy. Additionally, I explore the traceability and distribution of human-recognizable concepts within latent representations. I also demonstrate the effectiveness of our redundancy approach by creating an efficient transfer learning pipeline.
-
DissertationIntegrated approaches for profitable shea nut value chain enhancement: A case study in rural Ghana( 2024-12)Shea nut production is a critical economic activity for rural communities in the Sahel Region of Sub-Saharan Africa, particularly Ghana, where it contributes significantly to rural household incomes. However, challenges in post-harvest management and value-added processing hinder the full economic potential of the shea value chain. This dissertation investigated three key aspects of shea nut production: post-harvest loss (PHL) assessment, storage technology evaluation, and economic viability of shea butter processing methods in rural Ghana. By addressing these areas, the research aimed to provide sustainable solutions for improving the quality of shea nuts, reducing post-harvest losses, and guiding investment decisions in processing methods. The first study documented post-harvest handling practices and assessed the extent of post-harvest loss in the Ullo Traditional Area, Upper West Region, Ghana. Surveys of 280 women involved in shea nut production revealed that more than 60% of respondents experienced post-harvest losses exceeding 12%, primarily due to inadequate drying methods and poor storage conditions. Most women relied on traditional sun-drying techniques that exposed nuts to adverse weather and pests. The study highlighted the need for low-cost, scalable post-harvest technologies, such as drying platforms and hermetic bags, to reduce losses and enhance the quality of shea nuts. This study also highlighted some of the on-going interventions we have initiated in the Ullo Traditional Area in collaboration with Self Help International. The second study evaluated the effectiveness of Hermetic Bag Storage Technology (HBST) in preserving the physical quality of shea nuts. Over a 30-week storage period, HBST was compared to traditional jute sacks and woven polypropylene bags. The results showed that HBST significantly reduced moisture variation and insect damage, maintaining insect infestation below 30%, compared to over 80% in the other two methods. The study demonstrated that HBST is a viable, cost-effective solution for preserving quality and quantity of shea nuts and recommended their adoption in rural production systems. The third study investigated the moisture sorption isotherm characteristics of shea nuts, identifying safe storage moisture content between 5-7% over a temperature range of 20-35°C. By evaluating equilibrium moisture content (EMC) models across that temperature range, this research expanded the application of a handheld commercially available low-cost moisture meter to monitor drying and storage conditions. Additionally, practical recommendations for improving post-harvest handling, drying and storage resulted that can be utilized by Extension workers and others to train smallholder farmers in best practices and technologies that prevent spoilage and improve quality of shea nuts. The fourth study assessed the financial profitability of four shea butter processing methods: fully manual (S1), semi-mechanized with process-outsourcing (S2A), semi-mechanized with group-owned equipment (S2B), and mechanized (S3). Using financial indicators such as Net Present Value (NPV), Internal Rate of Return (IRR), Benefit-Cost Ratio (BCR), and Payback Period (PBP), this study identified the mechanized process (S3) as the most financially attractive option, with an NPV of GHC 50,175, an IRR of 85%, and a PBP of 1.2 years, making it ideal for large-scale investors. However, for rural women’s groups, the semi-mechanized method with group-owned equipment (S2B) purchased through grant funds could be a viable path to profitability, allowing these groups to scale production and increase income with moderate investment in equipment and manageable financial risk. The study recommends a phased approach to mechanization, integrating technical training and grant funds from NGO partnerships to empower rural women and enable sustainable shea butter production adapted to local and regional markets. Overall, this dissertation comprehensively analyzed post-harvest management practices and technologies, and the economics of value-added processing in the shea value chain. By addressing these key areas, the research offers valuable insights for policymakers, rural cooperatives, and development organizations seeking to improve women's livelihoods in Sub-Saharan Africa by engaging in shea production and processing sustainably and profitably.