48440 research outputs found
Sort by
Theoretical Investigation of CO2 Reduction on Cobalt Carbon Nitrides
No full textConverting CO2 into valuable feedstocks and achieving a carbon-neutral cycle, in which the CO2 is continuously recycled, is crucial for environment and economy. However, the high stability of CO2 makes the reduction challenging. One promising strategy is to use photocatalyst to drive CO2 reduction. Cobalt single-atom catalysts (SAC) on carbon nitrides (Co/carbon nitride) are attractive due to their abundance, non-toxicity, tunable structures, and visible-light activity. Experiments show that Co/carbon nitrides can convert CO2 to CO with high selectivity. However, the exact of structures of Co/carbon nitrides are uncertain, especially when the degree of polymerization of the carbon nitride is unknown. X-ray absorption spectroscopy (XAS) alone is insufficient to identify the structures at the atomic scale. As a result, the structure-reactivity relationship of the Co/carbon nitrides and their CO2 reduction mechanisms have not been thoroughly investigated. In this study, we used density function theory (DFT) to model Co SACs on various carbon nitride structures and to identify their coordination environments. We examined Co bound to molecular carbon nitrides (melem and melem dimer) and to polymeric carbon nitrides (melon and graphitic carbon nitride), and related their coordination to CO2 activation ability. In addition, we simulated Co K-edge X-ray absorption near edge structure (XANES) spectra and compared them with experimental spectra to narrow down the most likely Co/carbon nitride structures. To further understand the catalytic performance of these Co/carbon ni- trides, we also investigated two electron CO2 reduction mechanisms, along with the hydrogen evolution reaction, a major competing process. We identified Co/2-melem-dimer as an optimal CO2 reduction photocatalyst, and carbon nitride structures strongly affected catalytic performance. Overall, our work provides atomic-level insight into Co/carbon nitrides and highlights their potential as efficient photocatalysts for CO2 reduction
Spot Ignition of Structural Fuels by Firebrand Accumulations
No full textFirebrand spotting is known as an important wildfire spread mechanism in both wildland and wildland-urban interface (WUI) fires over the past few decades. During this process, firebrands are lofted by fire plumes and transported downwind, where they may ignite receptive fuels (wildland fuels or structures) both near and far away from the flame front. Although previous experimental studies have demonstrated the enhanced ignition abilities of compact firebrand piles, there is less clear scientific understanding of the physical processes controlling the ignition of structural fuels by accumulated firebrands. The firebrand phenomenon is typically studied in three sequential sub-processes: generation, transportation by fire plumes and wind, and ignition of recipient fuels. The research presented in this dissertation focuses on characterizing airborne firebrands prior to the accumulation process and analyzing the capabilities of a group of firebrands to cause spot ignition of structural materials. In Chapter 2, the generation of firebrand showers was better understood by utilizing an imaging measurement methodology to characterize the temperature and velocity of flying firebrands produced from burning vegetation. Then in Chapter 3, the importance of cooperative spot ignition effect due to close proximity of two idealized firebrands was elucidated through experiments and numerical modeling. Lastly, in Chapter 4, the ability of accumulated firebrands (more than two) to cause flaming ignition of an engineered wood material was experimentally studied under varied air flow conditions. Models were developed to account for the burning and spotting behaviors of firebrands within well-characterized accumulations. The data and results form this dissertation provides valuable insights for the development of predictive spot fire models by providing input firebrand properties and improving the understanding of fundamental heat transfer processes from firebrands to receptive fuels
High-Resolution Three-Dimensional Ultrasound/Photoacoustic Imaging and Image-Guided Intervention Leveraging Acoustic Reflector
No full textAcoustic reflectors have been used in ultrasound (US) and photoacoustic (PA) imaging devices to change the acoustic wave’s orientation. Previous developments in acoustic reflectors have enabled US and PA microscopic imaging, as well as handheld co-axial PA imaging based on clinically available linear array transducers. Building on these developments, the author proposed the feasibility of using an acoustic reflector for US image-guided intervention, in which the needle path and the US image plane are inherently aligned. The author further investigated the clinical translation of this US image-guided needle intervention mechanism by providing solutions for device encapsulation and minimization. In parallel, the author explored the potential of the acoustic reflector for generating high-resolution three-dimensional (3-D) US images. Even though a synthetic aperture focusing method was used to improve elevation resolution, the elevation resolution remained limited by the acoustic lens. To overcome this limitation, the author developed a deconvolution-based resolution enhancement method using a simulated point-spread function, which further improved elevation resolution in 3-D PA imaging. Finally, the author demonstrated several scenarios in which the acoustic reflector facilitates imaging device design. A magnetic resonance imaging-compatible US/PA transrectal probe was designed based on the acoustic reflector to enable prostate cancer imaging and was used for in vivo mouse tumor model imaging. In addition, a wearable forearm US device was designed using an acoustic reflector to improve wearability and stability, and a US computer tomography imaging device was proposed to monitor tissue-engineered blood vessels inside bioreactors
Developing a User-Friendly Tool to Evaluate Parks
No full textLandscape architecture is the development of public green spaces, like urban parks. Existing evaluation tools are often time-consuming and require specialized knowledge. This paper details the adaptation of a park evaluation tool to improve accessibility for a general audience, guided by feedback from non-specialist users and input from industry experts. We recommend presenting evaluation results through story maps. Guided by proof of concepts from Boston and Worcester, MA, story maps proved engaging and effective in highlighting the often-overlooked environmental benefits of urban parks. These findings support advocacy for landscape architecture and public education on its role in urban ecology
Developing A Platform for Indigenous Cultures and Tourism
No full textOur project developed a digital platform with RedTuri, a Panamanian network of indigenous tour guides that connects tourists with indigenous communities. Additionally, This platform also features a repository documenting the cultural and historical heritage of Guna Yala, one of Panama’s indigenous comarcas. Through the use of archival research, semi-structured interviews, and ethnographies, we gathered accounts from tour guides, artisans, leaders, and residents to understand how indigenous-led tourism and digital tools foster economic growth, social empowerment, and most importantly bridges the digital divide by connecting communities and promoting these voices on the internet
Bridging Educational Gaps in Panama: AI Coach for WRO
No full textLimited funding, technology, and teacher training in Panamanian public schools hinder access to hands-on STEAM education. Supported by FUNDESTEAM, a nonprofit expanding access to robotics education, we developed a low-cost, scalable AI coach to guide students and teachers preparing for the World Robot Olympiad. Grounded in Panama’s educational landscape through archival research, interviews, and ethnographic observation, our proof-of-concept AI provides structured guidance and problem-solving support through adaptive, class-based responses. The modular design allows new features with minimal coding, while knowledge and behavior expansion require none. We recommend future teams enhance the AI’s functionality, deepen its knowledge base, integrate learning science, and test in classrooms
Stock Market Simulation
No full textA four-week stock market simulation was run to compare the profitability of two selected trading strategies: The Buy and Hold strategy and the Trade the News strategy. Foundational research was conducted to better understand the stock market as well as the overall economy. Information was gathered on the major US stock exchanges, common stock indexes, capital markets and stock trading methods. Interim returns for each portfolio were assessed and reported weekly. The results for the simulation were: Buy and Hold +3.36%, and Trade the News +3.54%. This project provided valuable experience researching and simulating stock market trading, as well as to become more financially and economically informed by following the financial news and economic reports available to the public
Stock Market Simulation
No full textThis project undertakes a controlled six-week stock market simulation to compare the risk-adjusted performance of a discretionary manual strategy against a fully automated quantitative strategy. Operating with identical $100,000 starting capital and a fixed universe of ten equities in a paper trading environment with no transaction cost or margin interest, the manual approach employed swing-trading principles while the automated system integrated a Z-score-based mean reversion model with an AI-driven (FinBERT) news sentiment filter for trade confirmation and a Risk Parity algorithm for position sizing. The experience learned will be helpful for future investments in the stock market
Graph Neural Networks for High-Throughput MXene Material Property Prediction
No full textMXenes are an important class of 2-D solid materials with novel properties and myriad applications, including efficient energy conversion in batteries and solar cells, environmen- tal and water treatment, supercapacitors, electromagnetic interference shielding, and catalysts. Unfortunately, MXenes are challenging to synthesize in the laboratory, thus slowing MXenes research and limiting their potential. MXene experimentalists can optimize their research by accurately simulating MXene properties, informing them of which MXenes are most useful to synthesize. However, current methods, such as density functional theory (DFT), remain com- putationally expensive. There is potential for machine learning (ML) models to improve this process by rapidly predicting large sets of MXene properties given only their chemical formula. Multiple prior works have attempted to create these models. However, these prior ML models do not take into account the graphical structure of MXene molecules, and may require com- putationally intensive inputs such as DFT or properties derived from DFT simulations. This increases the training runtime and limits the generalizability of prior machine learning model predictions beyond specialized subclasses of MXenes. In this thesis, we present a pipeline to perform high-throughput MXene property predictions with greater accuracy and computa- tionally cheaper inputs than current ML methods. This accuracy is due to 1) accounting for the graphical structure of the MXene’s chemical bonds and 2) training model on a much larger dataset than previous methods. Our pipeline consists of two algorithms. The first is a novel rules-based generative model that takes as input a subset of chemical elements, average bond lengths between these chemical elements, and possible MXene structures, and generates a large set of unique crystal graph structures. The second is a Deep Learning model that takes as input the predicted molecular graph structure and outputs predicted properties for the given MX- ene. This pipeline allows us to perform high-throughput MXene property prediction, enabling accurate property predictions across a variety of target properties for a large number of MX- enes. We use this pipeline to predict a number of electrical, chemical, mechanical, and thermal MXene target properties for more than 13 million unique MXenes, offering insight into the cor- relations between their chemical formulae and properties. To our knowledge, this is the largest publicly-available dataset of MXenes properties to date. We train our Graph Neural Network models, obtaining best test set mean absolute errors for models trained ten times using different random seeds for each train/test splitting: Band Gap 0.062±0.031 (eV), Bulk Modulus 22.0±2.1 (N/m), Density of States (at Fermi Level) 0.78±0.12, Heat of Formation 0.052±0.007 (eV/atom), Binding Energy 0.295±0.048 (eV) and Work Function 0.196±0.018 (eV). For d-Band Center, we obtain an RMSE of 0.156±0.031 eV. For classification target properties, we obtain the following F1 scores: Dynamically Stable (T/F) 0.877, and Magnetic (T/F) 0.859. Compared with previous works, our predictions improve or match the prediction accuracy for these specified MXene target properties by providing a models with computationally cheaper inputs. The techniques developed in this thesis may be of interest to machine learning for materials science property prediction of more general classes of molecules with a crystal structure satisfying rigid or dis- crete geometric constraints similar to those satisfied by MXenes
Role and Structural Study of Phospholipase Cβ1 in Neuronal Differentiation and RNA Interference
No full textPhospholipase Cβ1 (PLCβ1) is the predominant species of inositide-specific phospholipase C in neuronal cells. PLCβ1 is found on plasma membrane, where it plays important roles in G protein Ca2+ signaling, second messenger generation, as well as Protein Kinase C (PKC) activation. However, the functions of PLCβ1 are not limited on membrane. Studies showed that downregulation of PLCβ1 can return differentiated neuronal cells to their stem-cell-like state by a mechanism independent of G protein indicating that PLCβ1 may play additional roles in the cells. This thesis explores two novel functional roles of PLCβ1 that are independent of its classic calcium-signaling role through Gαq activation. The first focuses on its ability to regulation the state of differentiation of neuronal cells, and here we show that ability is due to PLCβ1’s regulation of the cytosolic state of the inducible transcription factor early growth response-1 protein (Egr-1). Specifically, we show that PLCβ1 regulates the accessibility of Egr-1 to its cytosolic binding partner TRBP thus regulating Egr-1’s nuclear localization of Egr-1 and the transcription of genes associated with differentiation. The second function of PLCβ1 studied here involves its binding to a protein that regulates RNA-induced silencing. This protein, translin-associated protein X (TRAX), forms a complex with the protein translin to produce Complex 3 Promoter of RISC (C3PO) which enhances RNA-induced silencing. Our studies also found that PLCβ1 promotes interaction between C3PO and Argonaut (Ago) 2, the main nuclease component of the RNA-induced silencing complex. Formation of the PLCβ1-C3PO/Ago2 complex blocks the ability of the complex to hydrolyze mRNAs. Taken together, PLCβ1 is a fine-tuner of RNAi silencing. Our future studies will try to learn how Egr-1 induce neuronal differentiation in nucleus, and how fine-tuner PLCβ1 interacts with RNAi’s participants in detail by mutagenesis, protein-protein and protein-nucleotide interactions