137091 research outputs found
Sort by
Eliciting Design Insights for Everyday Device Supporting Informal Mindfulness Practices
No full textAs smart devices increasingly dominate our daily lives, our attention has become increasingly fragmented. While informal mindfulness practices offer a way to restore present-moment awareness naturally integrated into daily activities, most technological solutions, such as meditation apps, paradoxically contribute to smart device distractions. This creates a need to explore alternative approaches that support informal mindfulness practices without relying on screen-based interactions. In this thesis, I examined ways to harmonize technology with focused attention by reconsidering time's role in mindfulness through a series of participatory study sessions. I developed two initial prototypes aimed at supporting mindful moments embedded in everyday living contexts. To gain deeper insights from mindfulness practitioners' embodied experiences and explore specific design considerations, I conducted a participatory design study with ten experienced mindfulness practitioners, inviting them to reimagine their relationship with time and everyday objects beyond screen-based interactions. One selected concept was developed into a working prototype through close collaboration with its original creator. The findings reveal time as a crucial element in mindfulness object design, with participants favoring ambient timing mechanisms over disruptive alerts and suggesting unobtrusive operation based on the nature of informal mindfulness activities.M.S.Industrial Desig
Opsin Engineering Advances Utilizing Automated Intracellular Electrophysiology and Language Models
No full textOptogenetic tools, particularly channelrhodopsins (ChRs), enable precise control of neuronal activity with light, but designing opsins with desired properties—high photocurrent, fast kinetics, or specific spectral sensitivity—remains challenging due to the vast sequence space and complex sequence-function relationships. Traditional approaches are labor-intensive and intuition-driven, limiting discovery of novel variants. This thesis advances opsin engineering by developing experimental and computational strategies that link sequence to function and enable systematic exploration. An automated patch clamp platform was constructed that measures photocurrent amplitude, spectral response, and channel kinetics while collecting single cells for downstream sequencing. This integrated approach enables, for the first time, direct correlation of functional electrophysiological data with opsin genetic sequence on a single cell basis. This platform was used to analyze more than 100 cells and multiple populations of opsins. Protein language models were used as a tool to guide opsin design. Trained on evolutionary sequence datasets, these models provide zero-shot predictions of plausible amino acid substitutions, allowing prioritization of candidates from a vast mutational landscape. Using single-cell electrophysiology to validate predictions, we show that substitutions such as E300P and E300G in ChrimsonR enhance light sensitivity and sustained photocurrent amplitude by a magnitude. Critically, these mutations would be ignored in conventional protein engineering campaigns as they would be expected to render the opsin non-functional. Together, these studies establish a framework for data-driven opsin engineering. By utilizing automated functional measurements and machine learning–guided predictions, this thesis demonstrates a path for rational, high-throughput exploration of sequence-function relationships and the rapid design of next-generation optogenetic tools
Global optimization of space mission concept of operations and systems design via mixed-integer nonlinear programming
No full textSimultaneously optimizing space mission Concepts of Operations (ConOps) and spacecraft systems design involves discrete choices and nonconvex systems design models. This paper formulates the problem as a mixed-integer nonlinear program (MINLP) and presents both a heuristic approach based on augmented Lagrangian decomposition and a global optimization scheme that exploits concave structures commonly found in space systems. Specifically, we consider a crewed Mars mission case study developed in collaboration with NASA’s Advanced Concepts Office. To efficiently find feasible solutions to this MINLP, a decomposition framework from our prior work is modified and applied, where a series of convex mixed-integer quadratic programs and nonconvex nonlinear programs are iteratively solved. To identify globally optimal solutions, the proposed scheme incorporates the decomposition method as an upper bounding heuristic, constructs tight relaxations via piecewise linear underestimators, and applies variable domain reduction to close the optimality gap. The case study results demonstrate high computational efficiency and robust convergence behavior for both approaches. In particular, the proposed global optimization scheme converges to the same solution found by a state-of-the-art MINLP solver with substantially reduced computational cost, often by a few orders of magnitude. These results indicate the potential of the proposed framework as a practical and scalable alternative for solving nonconvex MINLPs in space mission and ConOps optimization coupled with systems design
Temporal evolution of aviation safety topics: comparing Sructural Topic Modeling and BERTopic on FAA Service Difficulty Reports
No full textThe aviation industry has achieved remarkable safety improvements despite increasing flight volumes and aircraft in service. With fatal accidents now extremely rare, commercial aviation stands as one of the safest transportation modes. However, the projected growth in air traffic presents an ongoing safety challenge. Indeed, even with current low incident rates, the sheer increase in flight numbers could lead to more incidents.
To maintain and further improve these safety standards, the industry increasingly relies on proactive risk analysis, which identifies and mitigates potential hazards before incidents occur. This approach leverages extensive textual incident data, including Federal Aviation Administration (FAA) Service Difficulty Reports (SDR), to learn from past events and implement preventive measures. Yet the manual analysis of millions of incident reports presents a significant challenge for safety analysts.
Natural Language Processing (NLP) has emerged as a powerful solution to this challenge, enabling automated processing and understanding of vast amounts of safety-related text data. Within the realm of NLP applications, topic modeling techniques, particularly Bidirectional Encoder Representations from Transformers (BERT) for topic modeling (BERTopic) and Structural Topic Model (STM), offer promising approaches for analyzing aviation safety data. More specifically, these methods can identify emerging safety trends and potential risks by tracking how incident report topics evolve over time. Building on these capabilities, the key objective of this research is to evaluate the effectiveness of STM and BERTopic in identifying safety-related topics within the SDR dataset and monitoring their temporal evolution. This analysis aims to enhance our understanding of emerging aviation safety patterns and potential risk factors.M.S.Electrical and Computer Engineerin
Learning Vision and Language Cues for Video Understanding in Egocentric and Instructional Videos
No full textWe perceive the world through a combination of senses: such as sound, smell, and vi-
sion, to learn from and interact with our surroundings.Among these, vision and hearing are
the primary sources of information gathering, especially through reading and listening. Ef-
fectively utilizing and combining these senses is key to developing intelligent systems that
can operate in and understand complex environments. A critical challenge hindering effec-
tive vision-language learning is an understanding of why and how to effectively integrate
language for improved video understanding.
In this dissertation, we leverage the language modality to learn effective video repre-
sentations across a range of tasks, including action recognition, forecasting, and summa-
rization. The key ideas developed in this thesis are (i) Vision-Language supervision for
action understanding, and (ii) Leveraging language for video summarization.
In Vision-Language supervision for action understanding, we generate rich action de-
scriptions and leverage information from multiple modalities to recognize and anticipate
future actions in videos. We also discover the extent to which language contributes in un-
derstanding actions in videos, through effective cross-modal supervision between the vision
and language modalities.
Finally in Leveraging language for video summarization, we generate text outputs for
every input modality, and evaluate the performance of foundational models on video sum-
marization task. By using text as the primary mode of input, we evaluate how the text
representations perform on video summarization. Building on this, we propose a hierarchi-
cal framework that incorporates multi-granular language cues and evaluate its effectiveness
for video summarization.Ph.D.Electrical and Computer Engineerin
Predicting Aeroelastic Limit-Cycle Oscillations Due to Freeplay Nonlinearity Using Pre-Critical Output Data
No full textThis thesis introduces a novel output-based approach for predicting limit-cycle oscillations caused by freeplay, a stiffness nonlinearity resulting in no or highly reduced restoring elastic effects within specific motion ranges of actuated structures. This nonlinearity can affect moving parts of aerospace vehicles, such as control surfaces or tilting propellers and rotors, as well as other engineering systems that feature components in relative motion. The resulting self-excited, bounded periodic oscillations can degrade system performance and induce structural damage, fatigue, or failure. These issues require computationally efficient and accurate methods to predict the onsets and amplitudes of freeplay-induced limit-cycle oscillations in the design phase.
The proposed approach uses output data from free-decay time histories at pre-critical operating conditions, before limit-cycle oscillations develop, to estimate the recovery rate to equilibrium as a function of a monitored response variable and a varying (control) parameter. Recovery rate data points are then fitted and extrapolated to predict the values of the response variable and control parameter associated with limit-cycle oscillation solutions, which correspond to a recovery rate of zero. This approach preserves the non-intrusive nature of direct time-marching simulations while offering higher computational efficiency and numerical robustness. In addition, its output-based nature enables application to both computational and experimental data.
Building on previous research on systems with geometrical or polynomial stiffness nonlinearities, this thesis investigates the effectiveness of this output-based approach for predicting limit-cycle oscillations due to freeplay for the first time. Novel theoretical developments to the original output-based formulation are introduced to address the unique complexities of freeplay nonlinearity, which introduces nearly amplitude-discontinuous stiffness properties into the system. These developments are demonstrated by predicting limit-cycle oscillations of an idealized, elastically mounted tilting propeller in airplane mode, with freeplay in the tilting mechanism. Predictions are based on output data from simulations of an analytical, two-degree-of-freedom model and are verified against reference limit-cycle oscillation solutions from direct time marching. The sensitivity of the predictions to parameters in the approach is explored to understand its accuracy and numerical robustness. The proposed approach effectively captures regions of the parameter-amplitude plane where limit-cycle oscillations develop using output data from as few as one time history at pre-critical operating conditions. Prediction accuracy regularly improves as one uses output data collected closer to the onset of limit-cycle oscillations. These results pave the way for investigating applications to more complex configurations in future research.
The novel contributions from this thesis have the potential to streamline limit-cycle oscillation calculations in the design phase of nonlinear aeroelastic systems, enabling higher performance and safety in shorter, more cost-effective cycles. For example, the proposed approach can help ensure acceptable limit-cycle oscillations in systems such as aircraft lifting surfaces and emerging air vehicle configurations for regional and urban air mobility. In addition, while this study focuses on computational applications, the output-based nature of the proposed approach makes it suitable for future experimental applications in wind-tunnel or flight testing. Lastly, the approach may be applied to other nonlinear dynamical systems with freeplay or physics that can be modeled using a similar mathematical form
Simulation of Coupled Conjugate Heat Transfer and Nonequilibrium Boundary Layer Dynamics in High-Speed Flow Environments
No full textIntense thermal loading due to the nature of compressible aerodynamics, turbulent boundary layers, and combustion puts high-speed, high-enthalpy vehicles at risk for thermal fatigue and failure. Accurately predicting and controlling the interaction of heat transfer and compressible turbulent boundary layers is critical to optimal design of these vehicles. This interaction has been an active area of research for decades and is extremely challenging to study analytically, experimentally, and computationally.
A prominent high-speed flow phenomenon is in the shockwave boundary layer interaction, or SBLI, in which a shock impinges on a boundary layer, leading to an unsteady local recirculation region known as the “separation bubble”. SBLI further complicates already challenging thermal and fatigue management due to inducing strong pressure oscillations and locally highly unsteady heat transfer oscillations. A primary objective in the study of SBLI lies in understanding the physical mechanisms driving the “breathing” motion of the separation bubble, which has been shown to oscillate at low frequencies in both shape, size, and position.
Mechanisms driving SBLI unsteadiness broadly fit into two categories. The first is the upstream mechanism, in which the unsteady fluctuations of the SBLI are thought to be driven by the large-scale coherent superstructures imparting low-frequency oscillations onto the separation bubble. The second is the downstream, or global, mechanism, where downstream sources of unsteadiness are linked to the SBLI, or broadly covers any mechanism which views the unsteady motion of the SBLI as intrinsic to the instability of the separation itself. Note that the “global” terminology is merely a nomenclature used in the literature, here, the SBLI appears to be a convectively unstable. This classification is reinforced by the results of the reduced-order model derivation, which is elaborated upon below. Large-scale SBLI tend to be primarily influenced by the downstream mechanism, whereas small or weakly separated flows generally favor the upstream mechanism. While heat transfer demonstrably affects the hypothesized mechanisms driving the breathing motion of the separation bubble, their link has not been thoroughly explored in the literature.
Numerical simulations of these elusive flow physics is an attractive alternative to direct mathematical analysis and experimentation, due to its potential flexibility and the ability to sample the entire flow field for data. It is typical for Computational Fluid Dynamics (CFD) simulations of all classes, i.e., Direct Numerical Simulation (DNS), Large Eddy Simulation (LES), and Reynolds Averaged Navier-Stokes (RANS) simulations to implement Dirichletor Neumann-type boundary conditions in most flow analysis problems. While these boundary conditions are simple and computationally cheap, they neglect physical complexities that involve dynamically coupled heat transfer between the solid body and the fluid flow. A higher fidelity heat transfer boundary condition is the Conjugate Heat Transfer (CHT) condition, in which thermal energy transport within the solid domain is simulated and coupled to the fluid domain.
Although CHT represents a more physically accurate boundary condition, it has not seen wide application in DNS or LES, and for most practical engineering purposes has been limited to RANS simulations. This is because of the extreme cost CHT introduces due to the addition of time scales in the solid domain, which are typically orders of magnitude larger than the fluid time scales. The disparity between these scales significantly extends required run times of fully-coupled simulations making them prohibitively expensive.
A variety of loosely coupled methods have been proposed in the literature to mitigate the cost of CHT. A summary of past work is provided in Chapter 1. Loosely coupled methods for CHT+LES have been adopted in the study of turbine blades, rocket cooling systems, and burner-injector systems with good success. However, they rely on assumptions whose implications on the results have not been thoroughly explored and quantified. Prominent methods rely on arbitrary user input and a posteriori, problem specific validation. Long term fluid averages, implicitly enforced through arbitrary user input, may be intuitively what the solid domain responds to, but the trade off between longer fluid averages and accuracy of the results has not been fully characterized. Additionally, improper solid domain grid resolution may result in poor prediction of solid-fluid interface heat flux due to poor resolution of the thermal fluctuations in the solid domain.
In this work, a novel loosely-coupled CHT methodology, dubbed the “Hacked Material Method,” (HMM) is developed and integrated into a wall-modeling LES (WMLES) approach. This novel methodology is then validated through a series of canonicalized supersonic boundary layer simulations. These simulations, in addition to validating the HMM and WMLES, also explored the cost-error tradeoff of neglecting solid domain fluctuations, analyzed the consequences of a poorly resolved thermal penetration depth, and validated strategies to mitigate errors that the novel methodology introduces. Ultimately, it was found that the solid domain fluctuations, well-resolved or poorly resolved, are unlikely to be consequential for realistic solid materials and typical supersonic boundary layer flows. The HMM is shown to be straightforward to implement, have easily mitigated errors, and result in orders of magnitude of numerical cost reduction over fully coupled CHT. It also found to result in mean field convergence times that are faster than an existing, widely-used methodology. Recommendations to future users of the HMM, and LC-CHT + WMLES in general, are made in Chapter 7.
As a final validation step, the HMM and WMLES methodology is applied to a Mach 2.5 thin steel panel SBLI. The primary performance metric was in the separation length, which was found to have excellent agreement to the experiment for the CHT case. The adiabatic case, simulated as a comparison point to the HMM and WMLES, was found to underpredict the separation length. Other performance metrics included mean wall pressure curves, wall pressure RMS, and broad solid-fluid interface temperature variation. All metrics showed decent agreement for the HMM + WMLES, with improvements over the adiabatic case.
The baseline HMM + WMLES SBLI is then extended into a warm-wall and cold-wall case. The unsteadiness mechanisms of the SBLI are compared across these three wall temperature conditions. The non-adiabatic SBLI are validated against DNS of non-adiabatic SBLI, and are found to show decent agreement. Broad characteristics of the SBLI across the different thermal conditions are compared. Fluctuations throughout the domain are low- and high-pass filtered to separate the effects of low-frequency unsteadiness and potential sources of excitation, through, e.g., coherent superstructures from the high-frequency portion of the turbulence. Correlation coefficients show that the incoming momentum fluctuations correlate strongly to the separation length fluctuations. The strongest correlations across the board appear over the low-frequency fluctuation components and are strongest in the warm-wall case(s). Warm wall pressure PSD also shows evidence for the warm(er) wall cases having increased dynamic activity relative to the comparatively more stable cold case.
Through a modification of an existing low-order model of the SBLI, which views the SBLI system as a low-pass filter, further insight into the non-adiabatic SBLI dynamics is obtained. This modification stems from the strong influence of quasi-freestream flowfield gradients which were neglected in the original model derivation. These outer boundary layer gradients are strongest in the cold wall case, which is attributed to the smaller, more compacted size of the cold wall SBLI. It is found that the primary influence of the nonadiabatic wall is on the mean characteristics and structure of the SBLI, which in turn affect the SBLI’s reactivity to forcing. The modification of the non-adiabatic wall on the dynamic system’s forcing, through, e.g., modification of the coherent superstructures is found to be a less important, though certainly a non-negligible effect.
Altogether, this dissertation contributes in two major ways to the research community. First, in its derivation and validation of a novel loosely coupled CHT method for LES and DNS of statistically stationary flows. The HMM is straightforward to implement and was shown to provide statistically similar, if not identical, flow fields for orders of magnitude less computational cost. Second, the dissertation work applies the HMM + WMLES methodology to study the unsteadiness mechanisms of the SBLI. Ultimately, the fluctuations of the solid wall were found to be of minimal consequence to the behavior of the xxiv turbulent flow and separation bubble, with the mean temperature variation a far more significant influence. The mean wall temperature broadly affected the structure of the SBLI, e.g., changing the interaction and separation length which in turn alters the peak frequency of the SBLI. While it was found that the forcing of the system was indeed affected by the non-adiabatic wall, the overall structure of the SBLI was a far more dominant effect on the SBLI dynamics. Future investigations should clarify the dynamics between the interaction region of the SBLI (e.g., the compression wave structure and its movements) and the separation bubble itself, and how the non-adiabatic wall affects these dynamics
Development of Data-Driven and Model-Based Tools for Spectroscopic Slurry Monitoring: Case Studies on Low-Activity and High-Level Nuclear Waste
No full textNuclear waste is challenging to store, handle, and process. The United States Department of Energy is currently (as of August 2025) building a waste treatment plant to vitrify waste at the Hanford site, and there has been increasing interest in the use of real-time and remote spectroscopic sensors for monitoring radioactive waste at this and other sites. However, there are significant technological hurdles preventing the immediate application of spectroscopic sensors to nuclear waste. The radioactive waste is a multicomponent, inhomogeneous, multiphase, and radioactive slurry with the potential for ongoing chemical/nuclear reactions and batch-varying particle morphology. Spectroscopic sensors have seen regular use in laboratory-scale reaction monitoring, not industrial nuclear-waste processing. There remain many unsolved questions and unconquered hurdles precluding the use of spectroscopic sensors for a safer, more efficient, and more robust nuclear-waste processing outlook.
This thesis introduces methods, shows data, and presents studies for monitoring multicomponent solutions, quantifying signal attenuation in slurries, and utilizing data for real-time fault detection in multicomponent slurries. Given the legacy of nuclear waste that has been left for current and future generations, this thesis moves the scientific literature forward to enable safe and efficient nuclear-waste processing using the most appropriate methods, whether those methods exist or require development.Ph.D.Chemical and Biomolecular Engineerin
Robust Adversarial Reinforcement Learning for Antineutrino-based Nuclear Reactor Safeguards
No full textAntineutrino-based nuclear safeguards have been proposed to address many nuclear reactor verification challenges. Theoretically, these systems can detect reactor on-off status, monitor thermal power levels, and verify the special nuclear material (SNM) within a core. The situational details of these proposed capabilities, however, dictate the plausibility of applying antineutrino detectors for nuclear safeguards. For the most complex proposed capability, verifying SNM inventory, system performance depends highly on both general reactor-detector parameters, such as the reactor design of interest and detector efficiency, as well as scenario unknowns, such as diverted assembly targets and replacement fuels. An object-oriented modeling and simulation tool was developed for researchers and decision makers to explore various system-scenario parameters for antineutrino-based safeguards development and assessment. This tool comprises five modules: adversarial agent, diversion simulation, spectra simulation, system sensitivity, and protagonist agent. By iterating over these modules, the adversarial agent learns to select the most threatening diversion scenario while the protagonist agent trains the most well-prepared diversion classifier. This iterative process, referred to as robust adversarial reinforcement learning, could result in a fully robust nuclear safeguard - equally ready for any diversion scenarios of interest. A case study demonstrated that models only became semi-robust to the simulated diversion scenarios. While the semi-robust machine learning models did not perform as well as statistical classifiers, diversion-targeted machine learning models indicate that there is still room for system improvement.Ph.D.Nuclear Engineerin
