48440 research outputs found
Sort by
Multiscale Modeling and Simulation of Drying Processes in a Spray Dryer
No full textThis dissertation presents a comprehensive multiscale modeling framework to simulate the drying behavior of bi-component slurry droplets in spray drying, a process widely used in the pharmaceutical and food industries for producing dry powders. The research integrates both droplet-level (microscale) and dryer-level (macroscale) physics to predict key outcomes such as final particle morphology, particle size distribution, and residual moisture content. The multiscale model consists of two main components: an Eulerian framework for macroscale dryer-level simulation and a Lagrangian framework for detailed microscale droplet modeling. The Eulerian framework solves the governing equations of fluid flow and heat transfer inside the spray dryer, capturing spatial variations in gas temperature, pressure, velocity, and humidity. This provides system-level insights into drying gas behavior inside the dryer. The Lagrangian framework, on the other hand, tracks individual droplets using a detailed microscale model. A one-dimensional finite-difference method with a moving grid captures the three major stages of droplet drying: initial evaporation with suspended solids, crust formation and vapor diffusion through the shell, and final particle heating. The model accounts for non-uniform internal temperature, porosity-dependent vapor transport, and the moving wet-core/crust interface. It is validated against experimental data for both dissolved and suspended solids, showing strong agreement in temperature and moisture content profiles. The study also examines the validity of key modeling assumptions such as the shape of nanoparticles suspended inside the droplets, initial crust thickness, activation energy models, and the continuum nature of vapor flow through the crust. Parametric studies evaluate how drying gas conditions (temperature, velocity, and humidity) and solids loading affect drying rates and final particle structure. A regime map is developed to distinguish solid versus hollow particle formation based on these factors. A key contribution of this work is the coupling of the Eulerian and Lagrangian frameworks using User-Defined Functions (UDFs). This enables dynamic information exchange between the gas phase and droplet phase, allowing the gas flow field to influence droplet drying behavior. The coupled model is validated against experimental data from several pilot-scale spray dryers operating with different materials. It accurately predicts dryer performance and particle properties under a range of operating conditions. The study also considers droplet-droplet interaction and particle stickiness, recognizing their importance in industrial spray drying. These effects are identified as important areas for future investigation
Adapting to Noise: Anomaly Detection Methods using Unsupervised Learning
No full textAnomaly detection in images is a critical task in computer vision, essential for various applications ranging from industrial inspection to medical diagnosis. This thesis presents a comparative analysis of state-of-the-art image anomaly detection methods under the challenging scenario of noisy training data, where both normal and anomalous instances are mixed without labels, following an unsupervised learning setting. Our comparative study covers a comprehensive selection of anomaly detection methods, including generative models for reconstruction-based detection, such as GANomaly, DDPM, Autoencoder and representation-based methods such as PatchCore, PaDiM, STFPM. Using benchmark datasets VisA, MVTec, and MNIST, we assess the robustness and effectiveness of these models. Our evaluation combines traditional quantitative metrics with qualitative analyses, which visually highlight localized anomaly regions to provide intuitive insights into model performance. This study is motivated by the reliance of current image anomaly detection methods on clean training data, which is often impractical in real-world scenarios where data cannot be guaranteed to be entirely clean. The findings offer significant insights into the strengths and limitations of each approach, providing valuable guidance for future research and practical implementations in noisy and unsupervised settings
The review of Lockdown Zone Wars: Culture, Art, Tech and Project Management
No full textLockdown Zone Wars is a 3D real-time tactics (RTT) game prototype developed independently using Unity. The project combines squad-based tactical combat with narrative themes of nationalism, identity, and corporate manipulation, set on a fictional volcanic island governed by a megacorporation and a declining international alliance. Players control a small gang composed of stylized clone soldiers, fighting against the enemy gang, and avoiding elite enforcer units deployed under high threat levels. The project began preparations in the Fall 2024 semester and involves the design and development of several aspects of content. This project demonstrates my ability to integrate narrative, game design, 3D art, and technical development into a cohesive prototype that reflects both creative vision and practical production skills
Neuromodulation of the C. elegans Functional Connectome
No full textOur brains are composed of cells that are in constant communication, forming intricate networks of connectivity. The physical connections between neurons define the connectome, while the way these connections are modulated to influence behavior is known as the functional connectome. The mechanisms that regulate and refine these neural pathways fall under the realm of neuromodulation. To explore how the functional connectome can be modulated, we utilized a pioneering model system with a well-mapped connectome: Caenorhabditis elegans. This transparent, microscopic worm allowed us to investigate how neuropeptides, infection, and internal states – such as hunger and trauma – shape neural connectivity. Beyond this, we extended our research to human learning and pedagogy, conducting a survey of how a student’s interest in science and confidence in pursuing scientific studies or careers were influenced by a neuroscience summer camp. Through these studies, we have identified novel mechanisms by which neuronal connections can be flexibly reshaped, enabling adaptive behaviors in response to an organism’s external environment
A First Principles Investigation into the Impact of Chemical Composition on the Properties of Two-Dimensional MXenes
No full textMXenes are a remarkable family of layered 2D materials useful for a wide variety of applications, including energy harvesting and storage, air and water purification, biomedical devices, and nanoelectronics. MXenes have the chemical formula Mn+1XnT2, where M is an early transition metal, X is either carbon or nitrogen, T is a surface termination, and n can range from 1-4. It has been shown that the structural, electronic, and mechanical properties of MXenes can be fine-tuned by choice of chemical composition, allowing the properties to be optimized for a given application. In this work, we used density functional theory (DFT) to model over 300 different carbide MXenes. We considered 9 metals, 10 terminations, and n = 1, 2, and 3. We first determined the energetically preferred structure for each MXene, and then investigated the impact of chemical composition on various structural, electronic, and mechanical properties of the most stable configurations. All together, this work includes the results of nearly 6,000 DFT calculations, providing the most comprehensive systematic investigation into MXene properties to date. Our work highlights how different properties of MXenes can be tuned based on the choice of metal, termination, and number of atomic layers, and thus motivates further work on these materials
Development of bioinformatic tools for metabolic engineering
No full textGenomics and transcriptomics have revolutionized our understanding and use of microbes. Yet, there is limited use of this genomic information in biology and biotechnology. Recently, genomics and transcriptomics have begun to be more fully integrated into biotechnology workflows. Here, I describe improvements to genomic analysis of strains and strain engineering and the development of machine learning approaches to better interpret transcriptomics data for strain engineering. We apply these tools to non-model organisms, both because they are potential hosts for biomanufacturing and because our tools can provide unique insights into their biology. Our lab developed Prymetime to assemble genomes and to identify engineering signatures throughout the engineering process (Collins et al., 2021). Prymetime acts as a quality control step in genetic engineering and can prevent unintended edits from being unintentionally released into our environment. Transcriptomics provides information on gene structure and function, expression regulation, and genome dynamics (Dong and Chen, 2013). For well-characterized organisms, this identifies genes of interest; but this analysis is limited when genes are not functionally annotated, which is characteristic of non-model organisms. We show that developing integrated genomic tools that resolve such obstacles enables genetic engineering to effectively tackle contemporary challenges like producing medicines, growing and protecting food crops, and producing these products sustainably. Specifically, we have developed genomic tools for non-model yeast organisms to characterize, annotate, and identify genes of interest from their genomes and transcriptomes. While there are existing genomic tools, they are tailored to a few specific common yeasts (Sherman, 2022; Subramanian et al., 2005). With the broadening capabilities of bioinformatics, new yeasts with favorable genotypes can be analyzed for potential favorable phenotypes. Yet, desirable genotypes discovered in non-model yeasts can only be leveraged through extensive genomic characterization. Therefore, this research establishes genomic tools for non-model organisms from raw whole-genome sequencing and RNA sequencing through assembly and downstream analysis
Digital Hardware Beamforming for Accurate MIMO System Testing
No full textThis thesis evaluates a proof-of-concept method for testing beamforming capabilities, like those in Multiple-Input-Multiple-Output (MIMO) systems, in a digital hardware emulated channel environment. We begin by deriving the theoretical gain performance of an array of antennas. We then describe a method for conversion of a standard communications channel to a geometry-based channel and show the accuracy of a plane-wave approximation. Third, we evaluate the proof-of-concept system at two corner cases to showcase drastic changes in received power for both a signal of interest and an unintentional interferer, based on physical location within the geometric channel. Finally, we evaluate the total performance of the system across all angles within the channel and compare this performance to the theoretical value derived at the start of the work. In this work, we will show that the emulated null locations are accurate to 2 degrees, and the gain is accurate within 2dB at the lobe peaks
Secure by Simulation: A Novel Timing-Driven Verification of Masked Hardware Designs
No full textMasking uses the principles of secret sharing as a countermeasure for side-channel leakage in digital logic circuits. It allows individual shares to leak independently without revealing the secret. In practice, however, masking shares often leak jointly due to implementation effects in the physical hardware. Circuit-level timing effects create glitches resulting from data dependent timing variations and are a source of side-channel leakage. Existing formal verification tools attempt to model hardware implementation effects as a means for leakage assessment. However, they are frequently based on the generally true properties of a circuit rather than the concrete implementation itself, leading to conservative masked implementations. This work presents a novel approach for masked circuit verification in the presence of glitches using low-level timing simulation. Timing simulation allows for concrete circuit implementation analysis while utilizing common-place chip design tools. Using detailed circuit timing models enables a highly accurate estimate of technology process variations’ impacts, including temperature and voltage, for robustly sized chip designs. The methodology presented in this work is able to quantify timing-based side-channel leakage for the circuit under test. Additionally, examples show the ability to distinguish between leaky and non-leaky masked designs that otherwise appear indistinguishable from traditional formal verification models. This work demonstrates that using circuit timing for masked hardware verification leads to a more efficient concrete circuit implementation and tighter leakage assertion
FaithfulPersona: Balancing Faithfulness and Personalization in Code Explanations through Self-CritiqueENTER YOUR ETD TITLE
No full textCode explanations are crucial in real-world life, from educating students to aligning technical projects with business goals. However, existing approaches face challenges balancing faithfulness to the original code and personalization for diverse user needs. This thesis addresses these challenges by introducing a novel benchmark and method for generating faithful personalized code explanations. Our benchmark, FaithfulPersonaCodeX, incorporates code samples and user profiles, employing various evaluation metrics to evaluate both faithfulness and personalization. We propose DISCO, a new method that uses a self-critique mechanism and two-stage optimization to balance faithfulness and personalization in code explanations, addressing the limitations of current large language model approaches. Our proposed model, DISCO, achieves a notable 3.7% improvement in Pass@5 compared to the strong baseline method, Self-Consistency, while maintaining high personalization with a 61.08% win rate in the LLM-as-a-Judge evaluation, effectively balancing faithfulness and user-specific needs in code explanations
Investigating Adversarial Examples as a Universal Property of Trainable Models from a Dynamical Systems Perspective
No full textThis thesis examines adversarial examples in trainable models through a dynamical systems perspective, investigating whether such vulnerabilities are universal or can be mitigated through design choices. Using our Degrees of Confusion framework, we evaluate targeted gradient-based attacks against multilayer perceptrons, denoising autoencoders for classification (DAE-Cs), and Hadamard models. Experiments across both MNIST and CIFAR-10 datasets demonstrate that architecture selection and iterative application impact adversarial robustness. DAE-C models with appropriate bottleneck dimensions showed improvement as iteration count increased, with some configurations requiring perturbations twice as large as baseline models. Across DAE-C model iterations, successful attack rates progressively declined, though all architectures retained some degree of vulnerability despite these improvements. These findings present a two-fold conclusion: vulnerability can be meaningfully mitigated through architectural choices, yet the persistence across diverse architectures suggests adversarial examples may be fundamental to trainable models