48440 research outputs found
Sort by
Continual Learning For Cardiovascular Modeling
No full textAccurate cardiovascular dynamics prediction is vital for clinical decision-making in congenital heart disease, yet traditional CFD is too slow for real-time use. This work introduces an integrated scientific machine learning framework that combines multiple advanced methodologies. We use neural networks with domain decomposition and adaptive spatial refinement. To achieve real-time predictions, we adopt operator learning where we emulate reduced-order modeling and apply continual learning to improve temporal prediction. Applied to a T-shaped vascular model, the method achieves substantial gains, with continual learning reducing errors by more than 80%. The framework provides a foundation for real-time, patient-specific hemodynamic simulation with strong potential for clinical translation
Hydrodynamic Flexible Spindle Polishing of Complex Channels
No full textComplex channels, featuring small diameters, extended lengths, tortuosity, internal branching, thin walls, non-uniform cross-sections, and/or high length to diameter (L:Ø), are designed and fabricated for optimized thermal and fluid transfer efficiency in aerospace, energy, and tooling industries. Polishing these channels to improve internal surface finishing is critical to fatigue life, dimensional integrity, corrosion resistance, and fluid flow efficiency. Hydrodynamic flexible spindle (HydroFlex) polishing uses a high-speed fixed-abrasive grinding wheel driven by a flexible spindle to navigate through complex internal channels for fast, uniform, and controllable material removal and surface improvement. Critical to HydroFlex polishing is the presence and consistency of consistent grinding wheel orbital motion around the internal contour of the channel for controllable, consistent, and efficient surface improvement. The wheel orbital motion is a result of a dynamic equilibrium among the hydrodynamic force from the surrounding coolant, the grinding force for material removal, and the elastic force introduced by the flexible spindle. The influence of HydroFlex polishing parameters, including fluid viscosity, wheel rotational speed, and grinding wheel and spindle properties on these forces and the presence and consistency of the wheel orbital motion is challenging to understand due to complicated underlying physical phenomena. This dissertation establishes the fundamental physical relationships between wheel motion, hydrodynamic force generation, and polishing performance in HydroFlex polishing of both conventionally and additively manufactured metallic channels. First, a Taguchi-based sensitivity analysis was conducted to quantify the effects of wheel geometry, fluid viscosity, rotational speed, and workpiece material on orbital frequency and polishing outcome. The results showed that wheel orbit frequency and consistency are influenced by grinding speeds, channel material, and fluid media with down-grinding motion and stable orbital frequencies up to 588 Hz. Performance in deal conditions validated HydroFlex, producing rough surface reduction from 13.4 μm (as-built) to below 1.2 μm. Second, a computational fluid dynamics (CFD) model was developed to elucidate the hydrodynamic forces acting on the wheel under varying viscosity and rotational speed. The model revealed that rotationally induced asymmetric pressure gradients around the grinding zone are the dominant source of the tangential hydrodynamic force responsible for orbital motion in high viscosity fluids. A process map revealed viscosity and rotational speed regions, above which, this hydrodynamic component exceeds the grinding contact force, inverting the orbital direction to upgrinding. Simulated and experimental force comparisons confirmed that the transition between up- and down-grinding can be controlled through direct modification of grinding zone forces. Finally, HydroFlex was applied to curved and tapered channels representative of turbine cooling and fuel delivery systems. X-ray computed tomography and optical profilometry quantified the geometric and topographical evolution during polishing. For 10 mm and 25 mm radius channels, circularity error was reduced by over 60%, and surface roughness decreased from 10–12 μm to below 1.5 μm. In tapered geometries, HydroFlex maintained the originally specified dimension within ±1%, demonstrating its capacity for dimensional control even in non-uniform internal profiles. These findings collectively establish a model of the hydrodynamic and grinding forces governing orbital motion in HydroFlex polishing. By linking flow-field behavior, force dynamics, and material removal characteristics, this work provides a unified framework for predictive HydroFlex process control. The outcomes demonstrate that HydroFlex can achieve controllable, uniform surface finishing in complex channels while preserving dimensional fidelity, advancing its applicability for post-processing of additively manufactured and conventionally fabricated aerospace and energy components
Webjam: A team-based software competition platform
No full textAs the landscape of software engineering is rapidly transforming—driven by AI adoption and a changing industry—junior developers are increasingly seeking higher qualifications and more experience. Academic programs for training software engineers lag, leaving a gap between academia and the industry. Shifts in the software industry and the existing gaps in the educational process are explored. To address these issues, a game-jam-inspired web platform was designed that enables users to collaborate on challenging programming projects using contemporary technologies. This webjam platform gives candidates opportunities to learn the architectural and soft skills required to be successful in a modern-day software engineering career, giving a competitive edge in the job market
Biological Applications of a Zinc Photocage
No full textThe project employs Human Neonatal Dermal Fibroblasts (CRL-2097), an adherent cell line isolated from the skin of a normal male neonate. Standardized cell culture protocols maintained high cell viability and identified the optimal method for adherence. These fibroblasts have served and will continue to serve as a critical biological system for validating new zinc photocages, such as DAPdeCage, and investigating dynamic intracellular Zn2+ signaling using the ZTRS and Zinpyr-1 fluorescent probes and associated photocage systems. The optimized culture of these adherent cells was essential for conducting in vivo imaging assays
Neuroprosthetic Device for Computer Assistance
No full textThis project worked to design and prototype a novel human interface device targeted at helping individuals with single-sided upper limb amputations or congenital abnormalities to control a computer mouse. Recognizing the current limitations of conventional prostheses in helping these individuals interact with computers, the project evaluates and compares different biomedical signal acquisition methods and decides that surface electromyography (sEMG) and eye tracking (gaze estimation) were the most viable options for this device. sEMG signals are captured by electrodes and translated into mouse clicks by an Arduino system, while a gaze estimation deep learning algorithm plots the coordinates of where the user looks using a camera onto pixel coordinates on the screen, taking the place of cursor control. This project intends to address a gap in assistive technology by offering a replacement human interface device (HID) that enhances accessibility in the technological field
Development and Prototyping of a Series-Elastic-Actuator-Driven Lower Body Exoskeleton
No full textThis thesis describes the design, fitment, manufacturing, and testing of a lower-body exoskeleton driven by series elastic actuators. The exoskeleton has 5 degrees of freedom (2 powered, 3 unpowered) and encloses the leg from the foot to the hip. The control laws of the exoskeleton are described. These control laws are designed to enable "human-in-command" control of the exoskeleton, allowing for operation of the exoskeleton without manual input via joysticks, buttons, or other forms of human interface device. In testing, the exoskeleton proved capable of providing assistance to motion of the lower body. Finally, areas of possible improvement of the exoskeleton design are discussed
AI Generated Video for Cooking Instructions
No full textEncouraging plant-forward cooking is central to lowering carbon footprints and promoting sustainable living. Klimatopf, a Swiss cookbook project, has expanded its educational reach through a digital app offering climate-friendly recipes. This study explores the potential of AI-generated video in culinary education by producing nine short instructional clips demonstrating key kitchen techniques. Through iterative prompt design, expert feedback, and student evaluation, the research assessed video clarity, quality, and pedagogical flow. Findings highlight the feasibility of integrating generative AI into Klimatopf's digital resources to advance sustainable food education
Creating a Virtual Tour for the Mount Washington Observatory
No full textIn collaboration with the Mount Washington Observatory (MWOBS), we developed a virtual tour (VT) of the Mount Washington Summit to increase accessibility and educational impact. We analyzed eight VTs, conducted participant observation during four summit visits, interviewed three experts in VT design, and facilitated five usability tests with eleven participants. We found we could foster an educational and engaging experience using intuitive navigation, multi-sensory content, balancing the volume of informational points, and pairing careful wording with external sources. We recommend MWOBS updates the VT as the summit evolves, integrates the VT into their educational programming, and adds the ability to experience various seasonal and weather conditions
Developing an AI-Powered Query System to Enhance the NLDL
No full textThe Natasha Luzhkova Digital Library (NL-Digital Library) is an online database designed through a collaboration of various organizations to guide international users to sustainable trail information. The goal of this project was to develop a technical approach to enhance the usability of the NL-Digital Library by streamlining user contributions and querying processes. Using data from six interviews and seven usability tests, the team iterated and implemented improved search and contribution processes for the NL-Digital Library. This database can help build a vast pool of knowledge for trails communities worldwide, allowing them to foster deeper global connections
Mapping Needs and Tools for Responsible AI Development
No full textThis project pursued four objectives to operationalize responsible AI (RAI): (1) catalog contemporary RAI tools; (2) build a curated RAG database for an RAI Copilot; (3) interview stakeholders to identify priorities, pain points, and improvements; and (4) evaluate the Copilot with a structured test. We produced a tool catalog, an open-access evidence base with metadata, and stakeholder insights stressing citation quality, usability, and confidence signaling. Using 60 persona-based prompts graded against Gold Evidence, results show unreliable retrieval but acceptable performance in other metrics. Deliverables include the catalog, database, interview summary, and evaluation protocol, guiding future Copilot development