Issuu

Fred DeMatteis School of Engineering and Applied Science

ASPiRe

Advanced Summer Program in Research

� • HOFSTRA ii

i: E 3

FRED DEMATTEIS SCHOOL OF ENGINEERING AND APPLIED SCIENCE

� 7

�

·r ".· , . • � A"

�

7Annual Symposium Hofstra University 208 Adams Hall

August 28, 2023 9:30 a.m. - 3:30 p.m.

ACKNOWLEDGMENTS

This is the seventh annual Advanced Summer Program in Research (ASPiRe) Symposium to be hosted by the Fred DeMatteis School of Engineering and Applied Science and I am pleased that twenty-one students have participated. A key component in its success and growth is the most generous financial support it receives from distinguished benefactors of the DeMatteis School. This year special recognition is due to: John D. Cameron, Jr., P.E. James K. Donaghy '89 Ronald Piervincenzi, '93; PhD Sina Y. Rabbany, PhD Thomas J. Sanzone '82 Paul R. Saueracker '78 Casimir S. Skrzypczak, '70 Their support of these exceptional researchers is of immeasurable value to the students themselves and adds luster to all our educational endeavors.

Sina Y. Rabbany, PhD Dean Fred DeMatteis School of Engineering and Applied Science

TABLE OF CONTENTS

Presentation Schedule ...................................................................................................................3 1. Investigations of Tightly Coupled Drivers for Multiport Antennas................................................4 2. Large Language Models in the Service of Cybersecurity ............................................................4 3. vProbing: Detecting hidden vCPU information from within a VM ................................................5 4. Enhancing Group Brainstorming through Large Language Models ............................................5 5. Evaluating Few-Shot Learning and Fine-Tuning in Large Language Models with Graph xx xxxNeural Networks for Legal Sentence and Document Classification ...........................................6 6. Breaking AI: The Nature of Adversarial Examples in Deep Learning Classifiers in Computer xxxVision........................................................................................................................................6 7. Robotics in Mechatronics ...........................................................................................................7 8. GPU-based HPC Clusters ..........................................................................................................7 9. IZPR: Instant Zero-knowledge Proof of Reserve ........................................................................8 10. Electric Bass Amplifier Analysis and Assembly ........................................................................8 11. Rapidly Deployable Rope Bridges ............................................................................................9 12. Quantitative Texture Analysis in Twin Preterm and Term Fetal Lung Ultrasound Images .........9 13. Preconditioning of Plant Based Vascular Grafts ...................................................................10 14. Ultrasonic Assessment of Plant-Derived Vascular Grafts .......................................................10 15. Culture and Differentiation of Mesenchymal Stem Cells with Hair Protein Biomaterials ..........11 16. Design of Real-Time Pressure Monitoring System and Application in Vascularized X xxxxxMicrofluidic Organoids .........................................................................................................11 17. Laminar Flow Studies in PDMS-Based Constructs to Optimize Network Stability ...................12 18. The Effect of Freestream Turbulence Level on the Drag and Lift Coefficients of Rotating xxxxBaseballs ..............................................................................................................................12 19. Construction of Upwelling Experiment for Soil Mechanics Laboratory ....................................13

PRESENTATION SCHEDULE TIME STUDENT/ADVISOR 1

3 4

6 7

PROJECT TITLE

9:30 Dean Rabbany Welcome Vincent G. Capone 9:45 Investigations of T ightly Coupled Drivers for Multiport Antennas Dr. D. Elliott Williamstyer Desray Desir 10:00 Mofoluwake Adesanya Large Language Models in the Service of Cybersecurity Dr. Oren Segal Edward Guo 10:18 vProbing: Detecting hidden vCPU information from within a VM Dr. Jianchen Shan Mason Malone 10:33 Enhancing Group Brainstorming through Large Language Models Dr. Simona Doboli Justine Prasad Evaluating Few-Shot Learning and Fine-T uning in Large Language Models with Graph Neural 10:48 Dr. Krishnan Pillaipakkmnatt Networks for Legal Sentence and Document Classification Dr. John E. Ortega Luke Wyszynski 11:03 Breaking AI: T he Nature of Adversarial Examples in Deep Learning Classifiers in Computer Vision Dr. Gerda Kamberova Greg Seoylemezian 11:18 Robotics in Mechatronics Dr. Kevin C. Craig Mani T ofigh 11:33 Dr. Edward Currie GPU-based HPC Clusters Mr. Alexander Rosenberg T revor Conley 11:48 IZPR: Instant Zero-knowledge Proof of Reserve Dr. Xiang Fu 12:03 Lunch Break

12:45

1:03

1:21

1:36

1:51

2:06

2:21

2:36

2:51

3:06 3:25

Katie Cameron Esther Visingardi Dr. Edward Currie Dr. Mauro Caputi Aurisha Rahman Sarah Samanez Dr. Edward M. Segal Christina Karras Dr. Sleiman Ghorayeb Nicole Gorbenko Dr. Nicholas Merna Christina Karras Dr. Sleiman Ghorayeb Andrew T arabokija Dr. Roche de Guzman Gayathri Suresh Dr. Sina Y. Rabbany Ashley Marquez-Villanueva Dr. Sina Y. Rabbany Matthew Hanson Dr. John C. Vacarro Dr. David M. Rooney Yahdim Lewin Mr. Robert Cerro Dr. David M. Rooney ASPiRe Group Photo

Electric Bass Amplifier Analysis and Assembly

Rapidly Deployable Rope Bridges

Quantitative T exture Analysis in T win Preterm and T erm Fetal Lung Ultrasound Images Preconditioning of Plant Based Vascular Grafts Ultrasonic Assessment of Plant-Derived Vascular Grafts Culture and Differentiation of Mesenchymal Stem Cells with Hair Protein Biomaterials Design of Real-T ime Pressure Monitoring System and Application in Vascularized Microfluidic Organoids Laminar Flow Studies in PDMS-Based Constructs to Optimize Network Stability T he Effect of Freestream T urbulence Level on the Drag and Lift Coefficients of Rotating Baseballs

Construction of Upwelling Experiment for Soil Mechanics Laboratory

Investigations of Tightly Coupled Drivers for Multiport Antennas Researcher: Vincent G. Capone Advisor: Dr. D. Elliott Williamstyer In our densely populated world, wireless communication has become standard. It now serves in many military, industrial, and personal use scenarios. Thus it is important to improve modern wireless communication by making it more efficient and secure. This can be achieved by designing antennas capable of sending signals in specific directions to achieve high gain. This “directing” of signals is known as beam steering. Beam steering is common when using an array of antennas but is less commonly seen with a single antenna setup, unless it is being physically turned. A multi-port antenna takes advantage of multiple signal driving power amplifiers to steer the direction of the radiation pattern as it is being created. However, the strong coupling between power amplifiers in a multi-port antenna can lead to instability, decreased field control, and lower efficiency. The purpose of this project was to design and test power amplifiers in order to study their behavior when connected to a multi-port antenna. The design and testing of power amplifiers was done using LTSpice software. If we get to build an actual prototype in the future, we will most likely use breadboards. We would also have to test the prototype to calculate real world efficiency, beam forming precision, and gain.

Large Language Models in the Service of Cybersecurity Research Team: Desray Desir, Mofoluwake Adesanya Advisor: Dr. Oren Segal As the world becomes increasingly digital and connected, securing information and systems is paramount. In this research project, we investigated the possibility of employing a large language model (LLM) in the service of cybersecurity. By harnessing the capabilities of OpenAI's Application Programming Interface (API), we have designed a Security Analysis Agent capable of detecting software vulnerabilities and providing recommendations on writing safer code. We delved extensively into the subject of neural networks and explored multiple Natural Language Processing (NLP) models to comprehend these foundational technologies. To select an optimal model, we measured the strengths and limitations of various leading models based on the Large Model Systems Organization (LMSYS) Leaderboard. Our ideal model would excel at several NLP tasks such as text analysis and completion, and be extensively robust and dependable. OpenAI’s state-of-the-art language model, GPT-3.5-turbo16k, achieved these objectives and was selected as the underlying model for the Agent. Using Flask, a Python microframework, we designed a web application that accepts user input - lines of code - and provides recommendations for safer code practices. HyperText Markup Language (HTML) and Cascading Style Sheets (CSS) were leveraged to improve user experience. Prompt engineering of the model was used to produce completions that are specific, accurate and visually palatable. The product of this research can be of potential use to cybersecurity professionals, aspiring learners, and software developers. It can serve as a support tool for a Human in the Loop (HITL) System in detecting and mitigating security risks, thereby contributing to the advancement of cybersecurity practices in this digital age. 4

vProbing: Detecting hidden vCPU information from within a VM Researcher: Edward Guo Advisor: Dr. Jianchen Shan

The majority of the modern internet runs on the cloud, with endless server rows provided and maintained by Amazon, Microsoft, etc. These servers are essentially massive computers, and these computers use "virtual machines" to pick apart these massive computers piecemeal and sell the pieces to customers. It would be no exaggeration to say the majority of compute around the world is spent within virtual machines. Our research has identified a notable inefficiency in the way Linux handles virtual machines. Specifically, Linux lacks insight into three critical areas of CPUrelated information—Capacity (indicating the strength of a “virtual” core, often referred to in virtual machines as a "VCPU"), Topology (identifying the location of each core), and Activity (determining how regularly a core is active or inactive). This lack of awareness impedes the task scheduler’s ability to make informed decisions, consequently affecting efficiency and performance. We point out these problems, and have created vCapacity and vTopology, custom software packages that use probing to find this information and communicate with vSched, a task scheduler for our custom operating system. Through this, we have laid the framework for a system that improves Virtual Machine performance significantly.

Enhancing Group Brainstorming through Large Language Models Researcher: Mason Malone Advisor: Dr. Simona Doboli

This study explores the use of large language models to improve creative idea generation in small groups. We investigated models such as ChatGPT, LLAMA, ALPACA, and VICUNA in their ability to generate ideas that could enhance group creativity. Through diverse scenarios and data from past group experiments, the research involves generating new ideas, elaborating upon existing ideas, and combining two or more ideas. We investigated different prompt engineering techniques to optimize the effectiveness of the model's responses. As context of the conversation is crucial for humans, we examined how context affects language models. We experimented with prompts with no conversational history to prompts that include human and/or bot-generated messages. Moving forward, our research aims to develop a real-time chatbot tool for improving creativity in group chats, and to observe its impact on collaborative ideation within the group chat dynamics.

Evaluating Few-Shot Learning and Fine-Tuning in Large Language Models with Graph Neural Networks for Legal Sentence and Document Classification Researcher: Justine Prasad Advisors: Dr. Krishnan Pillaipakkamnatt, Dr. John E. Ortega The research began with the exploration of legal document analysis, focusing on the classification of sentences in decision documents into various classes such as "finding sentences," and "reasoning sentences," among others. Recent advancements in large language models (LLMs) have demonstrated significant promise. The first phase involved experimentation with several BERT-based classification techniques, tailored for the U.S. Supreme Court decisions or Supreme Court Database (SCDB) documents. These techniques were specifically assessed for two classification tasks: a broad classification with 15 categories and a fine-grained task with 279 categories. This phase was not only about obtaining state-ofthe-art results but also about gaining hands-on experience with fine-tuning a pre-trained model. Moving forward, the project shifted focus to Graph Neural Networks (GNNs), diving into node classification on the Cora dataset. The focus on GNNs is built off prior research done for better classification of SCDB documents. Experimentation with different optimization methods fueled progress towards the ultimate goal: using a GNN in which information about surrounding nodes is aggregated to provide a correct classification of the document. This approach aims to reach a high degree of accuracy using a graph structure. By linking SCDB documents with other cited documents in a graph structure, and by incorporating classification results of these references weighted by the graph structure, the research aims to generate more nuanced and accurate classifications of SCDB documents. Throughout the summer, we worked hands-on with different technologies that are important to our project. This initial phase helped us get a feel for the tools and methods we'll be using. Further research is planned to build on what we've learned and take our understanding to the next level.

Breaking AI: The Nature of Adversarial Examples in Deep Learning Classifiers in Computer Vision Researcher: Luke Wyszynski Advisor: Dr. Gerda Kamberova

Deep learning artificial intelligence has been an emerging force that has penetrated numerous industries with never-before-seen methods of optimizing work. While deep learning models are very effective, we aim to find the inconsistencies in the models and how they can be exploited from the perspective of a malicious attacker. In the field of computer vision, this is done by finding or generating images with specific properties that mislead the CNN, a deep learning model that can identify and classify objects within an image. We use methods such as blending and ‘one-pixel attacks’ to generate Sorites’ chains that connect two pre-existing images, one of which is correctly classified by a network and the other is misclassified. The goal is to identify pairs of neighboring images in the chain constituting a pair of a correctly classified image and an image that is misclassified by the CNN with high confidence level, and yet is indistinguishable by humans from the correctly classified image. The purpose is two-fold: (i) automatically create more pernicious adversarial examples; (ii) demonstrate that such attacks are unavoidable. If these techniques can be used to consistently generate adversarial images, malicious attackers can more consistently mislead the neural network, leading to undesirable results. 6

Robotics in Mechatronics Researcher: Greg Seoylemezian Advisor: Dr. Kevin C. Craig Most engineering systems today are an integration of mechanical, electronic, control, and computer subsystems, i.e., they are Mechatronic. Robotics is a subset of Mechatronics. The two most familiar robotic systems in today’s world are the quadcopter (surveillance, package delivery, warfare) and the three-link serial robot (automation and human assistance). While these two fundamental systems can be purchased as “black boxes” and operated with a joystick, and typically are, engineers must know how they are built from scratch. Professor K. Craig and senior mechanical engineering student Greg Seoylemezian demonstrated this (see photographs). These two robotic systems will be used in the sequence of courses taught by Professor K. Craig in Mechatronics for mechanical engineering juniors and seniors. They will also be used in lower-level courses to illustrate the essential integration among mechanical, electronic, control, and computer elements from the start of the design process in a modern mechatronic system (e.g., SpaceX, Tesla). These two systems are available for viewing in the Mechatronics Studio, Adams Hall 113.

GPU-based HPC Clusters Researcher: Mani Tofigh Advisors: Dr. Edward H. Currie, Mr. Alexander Rosenberg I researched on High Performance Computing which are computing environments designed to accelerate the processing of complex, computationally-intense workloads, and how it can be implemented, used, and managed at Hofstra university at the new SIC building. HPC clusters consist of multiple interconnected compute nodes, often equipped with other hardware resources such as Graphics Processing Units (GPUs). They are characterized by their capacity to deliver a high level of computational throughput. HPC systems distribute the computational load over several nodes to enable faster processing and provide increased computational capacity for solving complex problems in applications such as scientific research, engineering, and data analysis, that would be difficult, if not impossible, to solve using traditional computing methods. HPC systems are leveraged to model complex systems, simulate physical and biological phenomena, analyze large datasets, and perform other computationally intensive tasks. HPC clusters are typically purpose-built and incorporate advanced hardware, such as NVIDIA's A100 GPUs, for accelerate computing. The A100 GPU, based on NVIDIA's Ampere architecture, is engineered for AI, Data Analytics, and HPC. It provides a large amount of memory and computational resources capable of executing a large number of concurrent operations, making it suitable for workloads with demanding parallel processing needs.

IZPR: Instant Zero-knowledge Proof of Reserve Researcher: Trevor Conley Advisor: Dr. Xiang Fu We present two innovative zero-knowledge protocols designed to instantly verify the solvency of financial organizations with high efficiency and throughput. These protocols are made possible by a recent breakthrough in lookup arguments, allowing the prover's cost to be independent of the lookup table size for subsequent queries after an initial preprocessing step. The Non-Intrusive Protocol for Public Cryptocurrencies is tailored for cryptocurrencies like Bitcoin, where transaction details are public. It boasts a prover complexity of O(n log(n)), where n is the number of transactions in a subset of a larger table and a constant verifier complexity and proof size of O(1). This ensures a scalable and efficient verification process without intruding into sensitive details. The Protocol for Privacy-Preserving Platforms is designed for platforms that prioritize privacy and conceal transaction details, this protocol achieves constant complexity for both the prover and verifier, O(1). While sacrificing non-intrusiveness, it offers a robust solution for privacy-focused financial systems. The foundation of these protocols lies in the extension of the cq protocol, leading to the development of an aggregated non-membership proof for zero-knowledge sets. This forms the basis for the two instant proof-of-reserve protocols, each catering to different requirements and platforms. In summary, these protocols represent a significant advancement in the field of cryptography, offering instant and efficient solutions to assert the solvency of financial organizations. Whether applied to public cryptocurrencies or privacy-preserving platforms, they provide a flexible and secure means to ensure financial integrity without compromising sensitive information. This project is ongoing and will continue into next semester.

10.

Electric Bass Amplifier Analysis and Assembly Research Team: Katie Cameron and Esther Visingardi Advisors: Dr. Edward Currie, Dr. Mauro Caputi Our ASPiRe project was designed to allow us the opportunity to reverse engineer a Fender BXR 25 bass amplifier. The project also provided exposure to various electrical engineering disciplines. As we are both rising Juniors in mechanical engineering, we completed a condensed version of the ENGG 34 lab during the first week. This provided sufficient theory to understand frequency response graphs and familiarize ourselves with the using the electrical lab equipment. The first stage of reverse engineering the bass amplifier was analyzing the schematic. This included circuit analysis and simplification, finding transfer functions and matrix multiplication. After sufficient analysis, we simulated our results with MATLAB code and replicated the circuit in Capture and PSpice. The final stage of our project was to fabricate a physical model of the circuit. This included replicating the amplifier front panel using CorelDraw and laser cutting, specifying, sourcing and purchasing the appropriate circuit components, soldering wires to the front panel jacks and potentiometers, and assembling/hardwiring the pre-amplifier circuit in a bread board. We encountered a couple issues when testing the circuit which consisted of some incorrect wire placements and burned out op-amps. Once all the issues were corrected, we tested the pre-amp to compare the control functions and sound with the Fender BXR amp. Since we did not reproduce the power amp stages, we used two 9 volt batteries to power our pre-amp, which worked well and removed any noise and grounding issues. We used two audio test files to compare the two amps: 1) white noise, 2) professionally produced unaccompanied electric bass (jazz walking bass lines). The volume control on both amps raised and lowered the volume sufficiently well. The three equalizer tone controls (bass, midrange, and treble) also changed the tone of the input signal in a very similar way. The project results proved to be very successful and was a major learning experience for us in several practical ways, far more than can be learned from in a typical circuit analysis theory course. Many of these practical experiences will be applied to future theoretical and hands-on laboratory work.

11.

Rapidly Deployable Rope Bridges Research Team: Aurisha Rahman, Sarah Samanez Advisor: Dr. Edward M. Segal Single-rope bridges have been used for ages in the military as a way for soldiers to cross bodies of water and other perilous environments. However, individuals need a lot of upper body strength to pull themselves across these kinds of bridges. These systems aren’t practical for civilians trying to get across treacherous terrains during an emergency. An adaptable, secure, and easily assembled method of rescue is vital in these situations. This project focuses on enhancing an existing rapidly deployable footbridge design that allows individuals to walk across rather than pull themselves across, but is limited in its potential to be used at different sites. This bridge includes a net that serves as both a walking surface and a guardrail. To enhance the existing design, we researched ways to make the bridge adjustable in size so that a single bridge can be used for sites requiring different span lengths. Our prototype includes a polyester net with a border consisting of a series of Tenex-Tec rope segments. The segments have an eye splice on each end and the segments are connected to one another with carabiners. The span length can be adjusted by connecting ropes at the end anchors to these carabiners that run the length of the net. In configurations, where the full length of the net is not required, these connections enable the net to hang down freely without impeding a pedestrian’s walking path. The next phase of this project is to develop a method of attaching the rope bridge to different end anchors with a drone.

12.

Quantitative Texture Analysis in Twin Preterm and Term Fetal Lung Ultrasound Images Researcher: Christina Karras Advisor: Dr. Sleiman Ghorayeb Lung immaturity remains the most common cause of morbidity and mortality in preterm and early term neonates. Although gestational age is the best predictor of lung maturity, respiratory distress syndrome (RDS) and Bronchopulmonary Dysplasia (BPD) are not restricted to preterm (<34 weeks of gestation) twin births. Late preterm (34 0/7-36 6/7 weeks of gestation) and early term (37 0/7-38 6/7 weeks of gestation) infants have significantly higher rates of these complications than infants born at or beyond 39 weeks of gestation. Thus, identification of fetuses at risk for respiratory morbidity after birth due to lung prematurity or hypoplasia remains very important in the practice of obstetrics. The decision to proceed with or delay delivery is often dependent on the ability to properly assess fetal lung maturity and the presence or absence of lung hypoplasia. Preterm (<37 weeks of gestation) from term (≥37 weeks of gestation) fetal lungs of twin pregnancies were differentiated by quantitative texture analysis of ultrasound images by determining the Heterogeneity Index (HI) associated with lung maturity, and the results were compared to previous findings for singleton pregnancies. The heterogeneity index was obtained through a MATLAB program which implements the Floyd-Steinberg dithering technique and applies it to ultrasound images. This study included 120 patients with twin pregnancies, and a total of 887 grayscale ultrasound images. Initial results indicate that there is a statistically significant divergence in mean HI beginning at 35 weeks gestational age between singleton and twin fetuses such that HI sharply increases in singleton fetuses. 9

13.

Preconditioning of Plant Based Vascular Grafts Researcher: Nicole Gorbenko Advisor: Dr. Nicholas Merna Throughout the United States, the leading cause of death is coronary artery disease. Coronary artery disease creates a partial or complete blockage in the blood vessel which prevents proper blood flow to the heart. The research project focuses on creating vascular grafts to be implanted to aid the damaged vessel. Current vascular grafts are developed from synthetic materials, but this 3D graft is developed from a natural material known as Leatherleaf Viburnum. The Leatherleaf is decellularized which removes any foreign DNA through the use of detergents and clearing solution. The leaf is then rolled into a tubular structure (3D graft) to mimic vessel properties. 3D grafts are seeded with endothelial cells to undergo endothelialization which prevents blood clotting and rejection. Following cell seeding, 3D grafts need to be preconditioned with a bioreactor to prevent the cells from coming off which can occur during implantation. A bioreactor is a physical system which replicates the functions present in the body (blood flow). The bioreactor system has been designed with a feedback system to simulate physiological pressure and fluid flow. Existing bioreactors are expensive and incompatible with small diameter vascular grafts whereas this bioreactor is inexpensive and designed to fit any size vessel. The most unique aspect of this bioreactor is its ability to be adjusted to any type of host, ranging from humans to rats since it can be modified to replicate target blood pressure and heart rate. The feedback system is designed using a circuit with a solenoid valve, pressure transducer, myRIO and LabVIEW. The myRIO is connected to the previously listed components which can be regulated with LabVIEW. The pressure transducer constantly records the pressure of the system and sends a signal to the solenoid valve, closing it at 80mmHg and opening at 120mmHg at 6.75 Hz (rat heart rate). Grafts have been tested using the bioreactor with and without the feedback system and results have shown successful endothelialization.

14.

Ultrasonic Assessment of Plant-Derived Vascular Grafts Researcher: Christina Karras Advisor: Dr. Sleiman Ghorayeb A successful tissue engineered vascular graft requires an appropriate combination of a scaffold, cell source, and cell seeding strategy to maintain long-term patency and minimize thrombosis in vivo. The goal of this project was to determine the graft dimensions, integrity, patency, and the velocity and direction of flow of a plant-derived vascular graft. The plantderived vascular grafts were analyzed through non-invasive ultrasound methods in vitro. A series of custom-built phantoms were tested using the GE LOGIQ e ultrasound system and the GE L8-18i multi-frequency ultrasound probe. Throughout the duration of the program, there were a total of 12 phantoms tested. The plant-derived vascular graft phantom design has been finalized after testing several differing phantom configurations. The volumetric flow rates tested on each phantom ranged from 0.2 mL/min to 5.0 mL/min. The velocity of the Doppler fluid through the plant-derived vascular graft in the phantom was measured through Doppler imaging. The experimental results obtained varied slightly from the expected velocity values due to the occurrence of turbulence. The differences seen in the velocity are predicted to be caused by an unevenness of the inner diameter throughout the length of each plant-derived vascular graft tested.

15.

Culture and Differentiation of Mesenchymal Stem Cells with Hair Protein Biomaterials Researcher: Andrew Tarabokija Advisor: Dr. Roche de Guzman Mesenchymal stem cells (MSCs) are multipotent adult stem cells of connective tissues, vital in repairing these tissues (such as cartilage, fat, and bone) after damage. Due to multiple pathways, their respective differentiation into specialized cells is dictated by different stimuli, Fig. 1: (Left) An MSC on hair. (Right) Adhesion after 24 h. including the interaction with surrounding protein molecules and substrates. Human hair extracts, the majority as keratin and keratin associated proteins (KAPs), offer biocompatibility with potential applications in tissue engineering and in vitro cell culture. Such proteins, either as higher-order structures like macrofibrils or individual, can have unique cellular influences. In our study, MSCs were seeded and cultured on or with different hair protein biomaterials to examine their adhesion, proliferation, and differentiation capacities. MSCs grown on intact reduced and delipidized hair strands (Fig. 1) showed relatively low adhesion, but still higher compared to agarose-only control. When cultured on tissue culture plastic (TCP), MSC treated with DIII + KAPs differentiated into adipocytes with significantly more fat droplets (Fig. 2, middle right). More importantly, even without DIII but with KTN (which contains keratins and KAPs), cells were still differentiated into fat-like cells, but with different morphology (Fig. 2, bottom left). Different responses upon treatment of MSCs with hair proteins indicate bioactivity influencing MSC adhesion, proliferation, and differentiation. We are currently investigating the bone differentiation pathway and mechanistic molecular Fig. 2: MSC culture on TCP without signal transduction events that will allow us better control of MSCs behavior. (-) or with treatments for 2 w.

16.

Design of Real-Time Pressure Monitoring System and Application in Vascularized Microfluidic Organoids Researcher: Gayathri Suresh Advisor: Dr. Sina Y. Rabbany Understanding pressure-flow relations is crucial in finding conditions to create the most durable and stable homogenous vasculature of ETV2 activated endothelial cells within microfluidic organoid devices, achieving a more faithful emulation of native vasculature. At the heart of every microfluidic process lies the critical parameter of pressure, profoundly influencing fluid flow behavior, system performance, and the overall success of microfluidic applications. To achieve more insight into vessel stability, we implement a novel and continuous pressure monitoring system, enabling real-time observations of pressure dynamics during vessel formation. Incorporating pressure sensors within the microfluidic device allows to directly measure and analyze pressure variations at critical stages of vascular development. A pressure sensor was positioned at the inlet and outlet of the microfluidic device to accurately measure the incoming fluid pressure. Initially, the sensor was installed within the fluid input and meticulously calibrated, scaling the measured pressure to match the estimated pressure calculated using a combination of the flow rate and Bernoulli's principle. Once an experiment of ETV2 cells are seeded into the device, variations in the actual pressure are continuously monitored in relation to the height of media flowing into the commercial microfluidic devices. An Arduino microcontroller and multiplexor setup are employed to monitor and regulate the pressure data within the microfluidic device both in person and remotely. The Arduino, along with the integrated development environment (IDE) programming environment, are utilized to read pressure values and graph data to observe air bubbles, failed vascularization, and any leaks within the system. By creating a continuous pressure monitoring system for vessels within the device, we can optimize vessel formation by analyzing differences between high and low-pressure conditions. Through prolonged culture studies and continuous pressure monitoring, we seek to gain insights into the long-term effects of varying pressure conditions on vessel maturation, functionality, and overall tissue viability.

17.

Laminar Flow Studies in PDMS-Based Constructs to Optimize Network Stability Researcher: Ashley Marquez-Villanueva Advisor: Sina Y. Rabbany Recent strides in regenerative medicine have been able to recreate self-assembling blood vessels in-vitro utilizing microfluidic devices but, they have not been able to recapitulate the complex microcirculation network. The vessels constructed are not durable and stable and do not have the same dimensions as normal capillaries. To achieve in-vivo like structures, we plan to investigate the formation of capillary-like structures in 3D using organoid structures and examine the role of different pressure and flow regimens on their stability and durability. In the vascular system, it is known that wall shear stress contributes to the formation of vessels through mechanical and biological cues using the Hippo Signaling Pathway. Using a peristaltic pump and various microfluidic device designs, we examine the role of gravity flow versus pump flow to dissect the role of shear stress and hydrostatic pressure on development of hydrodynamically stable vascular network constructs. We also see how adding multiple cell types, such as pericytes, might help provide more stability to the vessel network. Initial results demonstrate the crucial role of gravity flow in initiating an angiogenic network. However, we are still encountering challenges in maintaining system stability when utilizing pump flow. Attempts to establish stable networks through the co-culture of human endothelial cells with pericytes have proven unsuccessful as well. These discoveries hold the potential to significantly enhance the development of disease models based on microfluidics.

18.

The Effect of Freestream Turbulence Level on the Drag and Lift Coefficients of Rotating Baseballs Researcher: Matthew Hanson Advisors: Dr. John C. Vaccaro, Dr. David M. Rooney A wind tunnel study was undertaken to ascertain whether the level of background turbulence has any effect on the forces acting on a baseball which can be rotated to simulate the spins imparted to a pitched ball. The baseball was mounted on a thin shaft which could rotate at spin rates of up to 2600 rpm, a value which typically can be produced by a pitcher in a major league game. A load cell isolated from the tunnel structure was mounted below the test section, and recorded both the streamwise (drag) force and the side (lift) force produced by the interaction of the upstream tunnel velocity and the baseball situated at the center of the cross-section of the test area, and the velocity was varied from 10 m/sec up to 40 m/sec in multiple increments. Two specially designed frames could be placed at the intake section of the tunnel to produce freestream turbulence levels of 3.5% and 5%, significantly above the 0.3% value obtained in the tunnel in the absence of either frame. Data were taken at many different upstream velocities for all three turbulence levels and a range of spin rates, and were analyzed. The overall trend found is that increasing the freestream turbulence intensity decreases both the lift and the drag coefficients at almost all spin rates over the range of Reynolds numbers tested.

19.

Construction of Upwelling Experiment for Soil Mechanics Laboratory Researcher: Yahdim Lewin Advisors: Mr. Robert Cerro, Dr. David M. Rooney A laboratory experimental apparatus was constructed to demonstrate the principles associated with upwelling in non-cohesive soils, as well as to demonstrate how the bearing capacity of a square footing on a non-cohesive soil is a function of the distance of a stationary water table below the base of the footing. The rectangular apparatus is made from ¾ in. Thick Plexiglas, and is capable of supporting over 300 pounds of saturated soil and stone It has a 2 ft x 1 ft cross-sectional area, and has a 1.25 ft deep layer of sand through which water can filter upward at a variable flow rate up to a maximum of 3 gpm. A set of manometer tubes measures the hydrostatic pressure at four equidistant vertical locations in the sand bed, allowing a calculation of the permeability coefficient of the soil. In addition, after setting the water level at specific heights, the footings are loaded with incremental weights until the base of the footing sinks below the surface of the sand on which it rests. The theory holds that as the water table approaches the base of the footing, the bearing capacity will decrease. Students can plot the bearing capacity of the footing versus water depth to validate the Terzaghi Bearing Capacity Equation.