48440 research outputs found
Sort by
Lighting the Way to Darkness: Combatting Light Pollution in Acadia National Park
No full textAcadia National Park is among the last accessible places in the northeastern United States where a pristine night sky remains visible. However, artificial lighting from nearby urban centers and park facilities disrupts wildlife behaviors and the clarity of the night sky. Building on eleven years of prior WPI dark sky research, this work aimed to preserve Acadia National Park’s dark sky conditions. The project followed three objectives: identifying key sources of light pollution, understanding visitor experience in terms of park lighting, and developing an actionable plan for increasing Acadia’s compliance with National Park Service dark sky standards
Design, Build, Fly of a Radio-Controlled Aircraft to Demonstrate Urban Air Mobility
No full textUnder the theme of urban mobility, a radio-controlled aircraft was created with the following configuration: single front-mounted motor, high wing, taildragger, and conventional tail. The aircraft had to fit within a 2.5 foot wide parking area; thus, the aircraft was designed to be able to rotate its 4.9 foot wide wing. Inserts and restraints were designed to carry payloads for three payloads. This included a crew-only mission; a crew, patient, and medical supply cabinet mission; and a crew and passenger mission. Overall, the aircraft was able to transport 2 pounds of payload during the second mission and 10 passengers for the third mission. During the competition, the team placed 61st out of 107 participants, successfully completing the ground mission and the first flight mission
Conservation Treatment Record Database at the Postal Museum
No full textThe Postal Museum (TPM) is a cultural heritage institution responsible for the preservation and sharing of British postal history. The former is accomplished in large part through conservation treatments, where accurate, consistent record keeping is essential; however, TPM’s current conservation treatment record (CTR) management system has inconsistent formatting and is difficult to use. To address this issue, we created ConservationHub, a bespoke database application that upholds conservation and CTR best practices, determined via interviews with conservators and CTR experts, while integrating efficiently into conservators’ workflow. Using Microsoft PowerApps with Power Automate and Sharepoint, we developed features such as searchability and record creation, storage, display, and sharing
Propelling Change: Implementing a Zero-Emission Zone in the Canals of Copenhagen
No full textWe partnered with Miljøpunkt Indre By & Christianshavn to collect data and create materials to support a zero-emission zone for tour boats and seaplanes in Copenhagen’s harbor. We measured air pollution and noise levels from canal tour boats and seaplanes at high-traffic locations. Our data demonstrates that air pollution and noise levels exceed safety thresholds set by the NIH and Danish EPA. We featured this information in multiple resources, including a website, informational flyers, email list, and articles. We presented our recommendations to Miljøpunkt and the local committees for Indre By and for Christianshavn, and the issue was featured by local media. Our recommendations outline steps for stakeholders to create the zero-emission zone through a citizen’s initiative
Canals of Venice
No full textThis project worked with Serenissima Development and Preservation through Technology (SerenDPT) to update and improve upon outdated or unorganized information regarding the canals of Venice, with a focus on boat traffic. We conducted surveys at 8 locations throughout Venice’s Historic City to record the movements of boat traffic in the canal network. The compiled information was used with similar data from past years to analyze how traffic was impacted by location, time of day, and year. The team concluded that the traffic situation in Venice could be improved by encouraging use of public transport, reducing congestion around Piazzale Roma, planning boat travel around the early-day surge of cargo boats, and continuing traffic counts in future years
From Sparse Records to Spatial Insights: Clustering and GIS Techniques in Analysis of Named Erratics
No full textGlacial erratics—boulders transported by ice sheets and deposited across North America—represent remarkable landscape features that exist at the intersection of geological science and cultural meaning-making. Despite their significance as both natural phenomena and cultural landmarks, information about named erratics remains scattered across diverse sources, creating barriers to public engagement and scholarly research. This paper presents the development and implementation of an interdisciplinary digital heritage platform that consolidates this fragmented knowledge into a unified, publicly accessible web application serving heritage tourists, educators, researchers, and local communities. The platform employs modern web technologies—React.js frontend, Node.js/Express backend, and PostgreSQL with PostGIS spatial extensions—to provide real-time filtering capabilities, interactive mapping, and route optimization functionality that transforms static heritage information into practical tourism planning tools. Through detailed examination of nine diverse case studies ranging from contested heritage sites like Plymouth Rock to Indigenous sacred places like Okotoks Big Rock, we demonstrate how standardized digital systems can accommodate complex cultural objects while maintaining respectful representation and scholarly accuracy. The platform's Traveling Salesman Problem route optimization, geodesically accurate distance calculations, and comprehensive filtering system enable users to discover and visit erratics based on geological characteristics, cultural significance, and accessibility requirements. This research contributes methodological insights to digital heritage practice, establishes precedents for heritage tourism applications, and validates interdisciplinary approaches that bridge geological science with cultural heritage preservation through accessible public interfaces
Robust Bayesian Models for Data Integration and Small Area Estimation
No full textWe demonstrate how to apply the Bayesian approach to achieve robust and efficient inference for population studies in real-world applications, addressing both continuous and binary study variables. First, we address the challenge of combining a large non-probability sample with a small, expensive probability sample to make inferences about the population. Second, the work extends this methodology by developing a link function that incorporates the variability of propensity scores into both the logit and population models, addressing limitations of using posterior mean propensity scores. Third, the research focuses on stabilizing small area estimation for binary outcome problems by introducing a stick-breaking process prior for area effects, enhancing robustness and reliability in domains with limited data. Throughout, we use an illustrative example on Body Mass Index data. When making inferences about a finite population, it is essential to first establish verifiable assumptions for the population model, as these assumptions are used to construct the sampling model. Building Bayesian models that are both flexible and robust enough to handle real-world data is difficult. In modern statistics, combining multiple data sources and dealing with uncertainties in modeling have become especially important when high-quality data are costly to collect. It is essential to minimize bias and cost with acceptable variability. Bayesian predictive inference of the finite population quantities is obtained using surrogate sampling from the population model. This framework also applies to problems like small area estimation, where small sample sizes make traditional methods unreliable. Together, these contributions advance the application of Bayesian methods to real-world data problems, providing tools that balance efficiency and robustness. We list three problems that we have studied. First, we tackle the challenge of combining a large non-probability sample with a small probability sample to make population inferences. While probability samples provide accurate inferences, they are often prohibitively expensive. In contrast, non-probability samples are cost-effective but yield biased results. We show how to supplement a large non-probability sample (NPS) with a small probability sample (PS) to achieve robust inference. A common practical scenario arises when the NPS includes auxiliary variables and the study variable but lacks survey weights. In contrast, the PS includes known weights and auxiliary variables but not the study variable. Both samples are drawn from the same population, with shared variables between the NPS and PS. We use a Bayesian approach to estimate propensity scores for the non-probability sample, which are then applied to model the study variable using a two-component mixture model. Second, we account for the uncertainty of the propensity scores by introducing a novel link function. In Bayesian approach, we usually draw samples from the posterior distribution of the parameter. This process eliminates the variability of the propensity scores. We introduce the link function to unify the logit and population models into a single uncertainty model, which can solve this problem. To improve the efficiency of the sampling from the posterior distributions of the uncertainty model, we use the new sampling approach which we call the product approach. In our third problem, we study obesity (binary data). We pay particular attention to data integration and small area estimation for non-probability samples, and posterior inference is needed about the finite population proportions. First, we estimate the propensity scores, and we use them to infer the population proportion. Second, accounting for the uncertainty in the propensity scores, posterior inference about the finite population proportion is made using the link function. The awkward computation in the standard logistic regression model is overcome using Polya-Gamma latent variables. In addition, a robust mixture model is used with stick-breaking random weights, where the number of components is also random. When the mixture model and logistic regression model are combined, there are increased difficulties in the computation and the data analysis. The use of Pólya-Gamma latent variables and the product sampling approach helps address these problems efficiently
A Probabilistic Method to Model Progressive Metatarsal Fatigue Failure and Microdamage Accumulation
No full textTo better understand the mechanisms of bone stress injuries (BSI) in metatarsals, we developed a program that modifies finite element (FE) models of metatarsals to mimic progressive fatigue damage through microdamage accumulation. Twenty-two human metatarsals were imaged using computed tomography (CT) and then cyclically loaded in uniaxial compression until failure. CT images were used to generate specimen-specific FE models and a custom program was developed to iteratively simulate cyclic loading and microdamage accumulation associated with mechanical fatigue. Probability was incorporated into microdamage accumulation through a Weibull distribution Fatigue life estimates were significantly affected by 1) the Weibull scatter variable, m, and 2) if damage occurred before or after yielding. Our findings show that the stiffness and displacement curves of the progressive damage model accurately represent those curves of the experimental data. This research is significant because it helps us better understand fatigue life and damage accumulation of bones in response to physical activity and contributes to prediction of BSI in humans
Using Machine Learning to Predict Human Metatarsal Fatigue Failure
No full textBone Stress Injuries (BSIs) are overuse injuries caused by repetitive mechanical loading without adequate recovery. Although clinical assessments incorporate training history, demographic factors, and biological characteristics to estimate BSI risk, there is currently no reliable or reproducible method to define individual-specific training thresholds. Fatigue life, defined as the number of loading cycles a bone can sustain before failure, serves as a quantitative surrogate for injury risk in cadaveric models. Finite Element (FE) methods have been validated for estimating bone strain and predicting fatigue life; however, their utility is constrained by the time, expertise, and computational resources required for model construction and analysis. This dissertation explores the potential of machine learning (ML) as a more scalable and efficient alternative for predicting bone fatigue life directly from medical imaging data, focusing specifically on the metatarsals, which account for approximately 20% of clinically observed BSIs. The first phase involved subjecting cadaveric metatarsals to physiologically relevant cyclic loading. Many of the bones failed and relationships between FE-derived strain metrics and experimentally measured fatigue life were analyzed to establish a baseline model. In the second phase, ML models were trained using CT images, demographic data, and selected FE-derived features to predict fatigue life. A final model was developed that achieved accurate predictions, ultimately without requiring explicit FE analysis. In the third phase, this model was applied to a dataset of runners to estimate metatarsal fatigue life under varying bone mineral density (BMD) conditions, simulating physiological adaptations to disuse and overuse. This work advances scientific understanding of how bone structure and material properties influence mechanical durability. It also establishes a foundation for in-vivo fatigue life estimation, providing a critical step toward personalized training recommendations aimed at mitigating BSI risk in high-load populations
From Design to Collaboration: A Journey Towards Integrity in Reverse Supply Chains
No full textThis dissertation addresses key challenges in reverse supply chains (RSC), including the variability in the quality of End-of-Use (EoU) products, complexities in acquiring technology for remanufacturing, risks of inevitable technology leaks, and the rising threat of counterfeiting in remanufactured product markets. To tackle these issues, a reverse supply chain network model was developed, incorporating minimum quality thresholds to account for input variability. The model also evaluates the impact of technology acquisition from OEMs and introduces coping strategies—penalizing and incentivizing—to manage information leakage, analyzing their trade-offs between technology access and profitability. The research then turns to the role of blockchain technology (BCT) in combating counterfeiting by enhancing transparency and traceability, recognizing that long-term stakeholder collaboration is crucial for successful implementation. To explore this, BCT co-investment dynamics are modeled using an evolutionary game framework and a Stackelberg structure, with OEMs and platforms as leaders and the uncertified market as a follower. Finally, an empirical event study assesses the financial impact of BCT adoption in firms with physical product supply chains, demonstrating its long-term influence on financial performance. Overall, the study highlights the strategic importance of integrating advanced technologies in RSCs, balancing strategies to manage technology leakage, sustaining collaboration for tech implementation, and leveraging blockchain to enhance both supply chain security and financial outcomes