137091 research outputs found
Sort by
Structural Topology Analysis and Optimization Methods for Large-scale Problems and High-order Discretizations
No full textEstablished in 1988, structural topology optimization has become a promising method for designing high-performance structures with unconventional and sophisticated patterns that are not straightforward for human engineers to comprehend and develop. On the other hand, topology optimization faces challenges to be adopted for wider engineering applications, such as the demanding development and computational cost of the numerical simulation, and the occasional lack of robustness of solving the underlying mathematical optimization problems.
This dissertation aims to contribute to both mathematical programming algorithms specific to topology optimization, and high-order numerical schemes and discretizations.
For the first part, a benchmarking study is performed that identifies some performance deficiency of the nonlinear optimizers on certain types of structural topology optimization problems compared to the method of moving asymptotes. Then a quasi-Newton correction technique is proposed that is compatible with general-purpose second-order optimization methods that employ a quasi-Newton approximation scheme. Superiority are demonstrated based on linear compliance and natural frequency problems. Suitability for large-scale problems of the proposed optimization method is also demonstrated.
For the second part, an Galerkin method-oriented algorithmic differentiation (AD) approach is developed to speed up the code development of complex physical models for both the forward analysis and the sensitivity evaluation, and then demonstrated using a high-order finite element framework. Next, a high-order cut-element continuous Galerkin difference (cut-CGD) method is developed that employs various stencil construction techniques and treatments of the void domain to improve robustness for topology optimization. Finally, with the developed AD method and the cut-CGD method, a topology optimization framework based on the level-set method is developed that is capable of performing high-order analyses for both the bulk physics within the analysis domain, and also physics on the cut interface of the domains.Ph.D.Aerospace Engineerin
Observations of wildfire volatile organic compound emissions and urban ozone precursors
No full textAnthropogenic emissions have and continue to degrade air quality and alter chemical regimes in the atmosphere. Emissions of greenhouse gases have increased global temperatures, contributed to increasingly arid conditions, and exacerbated wildfire seasonal lengths and intensities. Wildfires emit a wide array of carcinogenic, highly reactive volatile organic compounds (VOCs). Urban emissions of oxides of nitrogen leads to the formation of secondary pollutants like ozone and organic aerosols. Prediction and mitigation of pollution events requires the understanding of complex VOC oxidation pathways. This work employs ground observations to further constrain VOC emissions from wildfires, assess the accuracy of ground-monitoring network HCHO measurements and determine temporal changes in urban HCHO concentrations, and evaluate the accuracy of satellite products when monitoring NO2 over airports.
We use measurements from a proton transfer reaction time-of-flight mass spectrometer to develop emission factors (EFs) for 15 species for an understudied temperate forest biome in New South Wales, Australia. We conclude that EFs are applicable across geographically different but analogous biomes for modeling purposes. We then evaluate the precision and accuracy of three fast measurement, in-situ HCHO monitors and compare with observations from the TO-11A methodology at a photochemical assessment monitoring station in Atlanta. We find that TO-11A measurements are biased low, and that decreases in HCHO concentrations since 1999 are small. Finally, we use NO2 vertical profile measurements from a Pandora spectrometer and profile estimates from chemistry transport model to recalculate satellite retrievals. We find the recalculated quantities are biased low relative to independent ground observations.Ph.D.Environmental Engineerin
Development of Coupled Machine Learning and Optimization Framework for Comprehensive Molten Salt Reactor Design
No full textThe continued growth of the world population and push for clean energy has created a political impetus to modernize the existing nuclear reactor fleet. Such efforts have culminated in the creation of the Generation IV International Forum, a multinational coalition formed with the purpose of researching and testing six advanced reactor prototypes. Of these designs, the molten salt reactor (MSR) has received notable attention due to its attractive characteristics of high thermodynamic efficiency, inherent safety features, and improved proliferation resistance. Despite these benefits, significant work remains to be done before MSRs can be deployed to the commercial grid. The variation in designs poses a significant challenge towards down selection and necessitates optimization methods that can reliably provide information about variables associated with different design metrics. This work addresses this challenge through a comprehensive modeling framework developed for both moderated and unmoderated liquid-fueled molten salt reactors. It does so through an extensive thermal hydraulic/neutronic coupled, time-dependent database of reactor designs that is dependent on multiple input variables associated with geometry, materials, and fuel salts. From this database, a predictive machine learning model is trained to determine output parameters given any set of input reactor design variables. This model is then used in a genetic algorithm optimization sequence. The optimization sequence allows for flexibility in the optimization process, where different metrics can be chosen for different design goals. For example, one case could involve multiplication factor constraints coupled with minimization of transuranics production, whereas another could focus on maximizing breeding ratio instead. The end result of the framework provides a potential core designer with the required input metrics for building a molten salt reactor based on user-defined performance metrics.Ph.D.Nuclear and Radiological Engineerin
Impacts of Humidity and Degradation on Adsorbents for CO2 Capture
No full textCarbon capture technologies are critical for mitigating the global temperature increase and its associated consequences. Advancements in solid adsorbents can reduce the cost of carbon capture technologies, particularly in three key areas: degradation, interactions with humidity, and integration into structured contactors. First, economic models were developed to assess the impact of sorbent degradation on the cost of direct air capture (DAC) and bioenergy with carbon capture and storage (BECCS). The results emphasize that increasing the lifetime of adsorbents should be prioritized. As a preliminary investigation into which adsorbents will be stable in BECCS, adsorbents were directly exposed to biomass combustion flue gas. Pre- and post-exposure characterization highlighted the range of possible degradation mechanisms due to structural differences in adsorbents. The second focus was on developing carbon capture materials that can adsorb appreciable amounts of carbon dioxide from humid feed streams. The addition of guanidine groups to the backbone of polymer of intrinsic microporosity (PIM)-1 was shown to mitigate amine impregnation issues and further enhance carbon dioxide chemisorption in the presence of humidity. Then, an optimum feed humidity range was identified for a diamine-appended metal-organic framework (MOF), in which a humidity-promoted sorption mechanism leads to a significant enhancement in capacity. Finally, fabrication methods were developed to translate the powder materials into scalable contactors. Fiber sorbents and monoliths containing the diamine-appended MOF exhibited the unique stepped adsorption behavior of the MOF. All-polymer hollow fibers containing PIM-guanidine were also fabricated, addressing the energy and productivity consequences of inactive support materials in contactors. By lowering the cost of carbon capture technologies, the insights facilitate the rapid and widescale deployment that is necessary for limiting the global temperature increase. The results also provide a foundation for further advancements in understanding, controlling, and utilizing the impacts of humidity and degradation on carbon dioxide adsorption processes, which is crucial for their success.Ph.D.Chemical and Biomolecular Engineerin
In the Blink of an Eye: The Truth of Epilepsy
No full textThis short documentary was created as a course requirement in HTS 3086 – Sociology of Medicine and Health under the supervision of Dr. Jennifer Singh.Runtime: 07:36 minutesThis documentary explores the illness experience of having epilepsy from the perspective of both the person living with epilepsy and their caregiver (parent)
Trustworthy and Robust Early Sepsis Prediction for Intensive Care Unit Patients using Reinforcement Learning and Conformal Prediction
No full textThis dissertation develops advanced machine learning frameworks to improve early sepsis prediction in ICU patients. Three novel approaches are introduced: OnAI-Comp, a multi-armed bandit framework that selects the best-performing model for each patient; Sepsyn-OLCP, a reinforcement learning algorithm with conformal prediction for reliable outcomes; and NeuroSep-CP-LCB, a neural network-based contextual bandit integrating conformal prediction for calibrated, data-driven decisions. These methods prioritize accuracy and trustworthiness, addressing critical needs in predictive healthcare and advancing sepsis prediction in critical care environments.Ph.D.Electrical and Computer Engineerin
Manual Vs. Automated Window Blinds: Analysis of Energy Use Based on Climate Scenarios
No full textWindows in buildings impact energy usage for temperature control through solar heat gain and enable natural light to reduce reliance on artificial lighting. Balancing solar heat gain and daylight utilization is a challenge, which can be addressed by employing automated or manual blind systems to manage daylight and enhance user comfort and energy efficiency. Additionally, accurate weather forecasts are essential for predicting energy-efficient strategies through individual building energy simulations, as weather conditions synergistically interact with occupant behavior to influence energy consumption patterns. This research aims to assess the energy consumption associated with manual and automated internal window blinds in medium-sized office buildings situated within two distinct and significant climate zones in the United States, considering both present and future climate change scenarios based on the IPCC report (RCP 4.5 - 8.5). Employing a simulation-based methodology, the study unveiled varying effectiveness levels of diverse window blind configurations contingent on the specific climate zones (e.g., 4A Mixed-Humid, 2B Hot-Dry). In different climate zones, on-site energy consumption alterations for heating and cooling become evident as temperatures escalate in the forthcoming years. Using simulation as the method and comparing RCP 4.5 and 8.5 scenarios reveals that automated blinds—a more efficient choice than manual blinds—significantly reduces cooling energy consumption, particularly under RCP 8.5 in a 4A Mixed-Humid zone. Rising temperatures in Climate Zone 2B Hot-Dry are a factor in increased energy requirements for cooling and decreased energy requirements for heating due to climate change. This study provides enlightening insights into the potential benefits of diverse window-covering strategies concerning energy conservation within varied climatic contexts for the future
Design Space Exploration Methodology including Aircraft Design, Industrial System, and Economic Considerations
No full textAIAA AVIATION FORUM AND ASCEND 2025During high-level decisions for complex aerospace system designs, Top-Level Aircraft Requirements need to be implemented in a wide range of performance, industrial, and economic environments and under the same range of constraints. The trade-offs between the different range of environments are poorly reflected in traditional sequential design methods due to their nonlinear and interdisciplinary nature. In this paper, a Design Space Exploration methodology is discussed with an integration of Model-Based Systems Engineering and Multi-Disciplinary Analysis and Optimization. The goal is to identify, capture, and quantify key trade-offs across the constraint sets. The proposed framework allows rapid iteration in multiple design spaces with low- and medium-fidelity disciplinary models for aircraft performance, logistics modeling, and manufacturer economics. Surrogate modeling ensures traceability while supporting fast evaluation, and the methodology is validated on a derivative single-aisle commercial aircraft where multiple constraints in performance, logistics, and economics are analyzed. The approach demonstrates feasible design spaces, supports informed decision-making across stakeholders, and analyzes system-level feasibility early in the conceptual design phase using an interactive dashboard embedding metrics from performance, industrial, and economic domains
DFA-based Synthesized Assembly Dataset for “Deep Unsupervised Learning-Based Supplier Selection and Ranking for Assembly Manufacturing”
No full textThe dataset provides all assembly-level and component-level information used to evaluate the DU-ASM framework presented in the manuscript “Deep Unsupervised Learning-Based Supplier Selection and Ranking for Assembly Manufacturing.” Each assembly is constructed based on DFA (Design for Assembly) rules, and all constituent components are represented as voxel grids (.binvox format). The dataset also includes graph-based assembly connection information and quantitative manufacturing capability metrics (tolerance, time, cost, quantity) assigned for the various case studies in the paper. These data enable reproducible experiments for supplier selection, multi-label capability matching, and preference-based supplier ranking. The dataset is fully synthetic and designed to support research in assembly representation learning, manufacturing capability modeling, and supplier selection in cyber manufacturing systems.This dataset contains synthesized assembly data used in the study “Deep Unsupervised Learning-Based Supplier Selection and Ranking for Assembly Manufacturing.” It includes voxelized component geometry (.binvox), assembly connectivity information, and manufacturing capability metrics assigned for each case study. The dataset supports the development and evaluation of DU-ASM, an unsupervised framework for capability-aware supplier selection and ranking.National Science Foundation Future Manufacturing Research Grant (Award # 2229260
