Trident Scholar Abstracts 2021
Daniel J. Butchko
Midshipman First Class
United States Navy
Cyber-Physical System Security of Surface Ships using Intelligent Constraints
Cyber-physical systems are vulnerable to attacks that can cause them to reach undesirable states. This work provides theoretical solutions for increasing the resiliency of control systems through the use of a high-authority supervisor that monitors and regulates control signals sent to the actuator. The supervisor aims to determine the control signal limits that provide maximum freedom of operation while protecting the system. Approaches for finding control signal limits include analytical, reachable set, and Monte Carlo methods. These methods are applied to linear models for autonomous surface vessels to determine the rudder constraints necessary to protect vessels from collisions in the plane of motion. Promising methods are evaluated through simulations that incorporate natural geography with coastline data and realistic traffic patterns produced from ships’ automatic identification system data.
FACULTY ADVISORS
Professor Kiriakos Kiriakidis
Weapons, Robotics, and Control Engineering Department
Assistant Professor Brien Croteau
Cyber Science Department
Lenning A. Davis
Midshipman First Class
United States Navy
Tracking Additive Manufacturing using Machine Vision
This project investigates multiple methods of verifying Additive Manufacturing (AM) products using computer vision, including feature-based visual odometry and image registration. With the rise of AM comes cyber-physical risks and the potential for part defects. Feature-based Visual Odometry (VO) is a field of computer vision research that uses information from images to calculate the relative motion of the camera through space. We developed a VO algorithm to reconstruct extruder position and motion from stereo images as a cost-effective means to monitor AM part construction. Preliminary results in simulation demonstrate feasibility of the proposed VO method and identify factors that may limit performance. In addition to VO, we seek to verify product quality using image registration and template matching techniques. Investigation with feature-based methods led to the development of a print bed tracking pattern, which enables the estimation of the extruder’s position and orientation. Combined with knowledge of the product's pose and the position of the camera module a render of the product can be created. This simulated image can be compared to images captured from the camera module using image registration algorithms to check for defects. Preliminary results yield successful image segmentation and detection of structural defects.
FACULTY ADVISORS
Associate Professor Michael D. Kutzer
Weapons, Robotics, and Control Engineering Department
Assistant Professor John Donnal
Weapons, Robotics, and Control Engineering Department
Alec J. Engl
Midshipman First Class
United States Navy
Photorealistic Image Generation for Satellite Pose Estimation Using Generative Adversarial Networks
In autonomous satellite servicing operations, pose estimation is an integral process to guide the servicing satellite for rendezvous and capture of the satellite to be serviced. Convolutional Neural Network (CNN)-based methods show promise in satellite pose estimation. In order to train CNNs for pose estimation, sufficient quantity and quality of real training imagery that is labelled with detailed pose data are required. Such images are either unavailable or very costly to produce, often forcing augmentation using computer-generated or ‘synthetic’ image datasets. The consequences of using a synthetic training set has been shown to be detrimental to real-world application, even in cases where care has been taken to make the synthetic data more photorealistic; this is due to reliance on human perception and not the CNN’s perception when defining what is ‘photorealistic.’
In order to enable CNN-based pose estimators to fulfill their robust and efficient potential, one may draw from the distribution-matching ability of the Generative Adversarial Network (GAN) to modify an existing training dataset of synthetic imagery based on the characteristics of markedly fewer real images. This research focuses on the Cycle-Consistent GAN (CycleGAN) architecture for its strength in such style transfer tasks. Both a geometrically simple ‘proof-of-concept’ object and the on-orbit images of a small satellite are employed for ‘photorealistic’ image generation using CycleGAN and training of a simple CNN pose estimator. Resulting improvement to real image pose estimation accuracy of this CNN when trained on such photorealistic imagery vice synthetic imagery provides valuable insight to future applications of the implementation of CycleGAN for such training data generation, and to fulfill the potential of CNN-based pose estimators, trained using monocular camera images, as initial pose acquisition devices for autonomous servicing satellites.
FACULTY ADVISORS
Associate Professor Tae Lim
Aerospace Engineering Department
Associate Professor Randy Broussard
Weapons, Robotics, and Control Engineering Department
Associate Professor Gavin Taylor
Computer Science Department
EXTERNAL COLLABORATOR
Dr. David Gaylor
NASA Goddard Space Flight Center
Ψ Harrison D. Foley
Midshipman First Class
United States Navy
As neural networks are deployed to solve a wide variety of problems, it becomes increasingly important to understand what can cause them to fail. The goal of our project is to cause neural networks to perform poorly via adversarial methods that are more destructive than previous state-of-the-art approaches. Specifically, we have drastically improved adversarial attacks on images of faces in order to avoid detection by facial recognition, and we have carried out the first successful data-poisoning attacks for reinforcement learning.
University of Maryland
Midshipman First Class
United States Navy
Modern embedded devices are under attack at an unprecedented rate. These devices exist in every facet of society from mobile phones to hard drives, and breaches in their security result in loss of capital and sensitive data. These devices use encryption to protect their data, but attackers may still be able to defeat these protection mechanisms. This is a result of vulnerabilities in the implementation of the encryption algorithm. Our contribution is a set of methods to detect the existence of a particular set of vulnerabilities in a chosen embedded device based on power analysis. This work focuses specifically on Solid State Drives (SSDs) that protect data with the Advanced Encryption Standard and a vulnerability within the SSD’s ATA Security Unlock command that has been exploited in previous work. This work analyzes three commonly implemented versions of the SSD’s ATA Security Unlock command: (1) a string comparison of the passwords, (2) a hash comparison of the passwords, and (3) a key derivation function that generates a decryption key based on the password. The first two methods are known to have weaknesses in authentication implementations, while the key derivation method is recognized as providing stronger protection of authentication credentials This work implements these three SSD unlock functions on an open source SSD board (Jasmine OpenSSD) and demonstrates the feasibility of detection and classification of these vulnerabilities through power analysis. This work also analyses and detects these vulnerabilities through firmware analysis. Applying these findings to proprietary devices, this work also demonstrates the ability to classify drives based on the specific SSD unlock function implemented in the Crucial family of SSDs.
FACULTY ADVISORS
Associate Professor Owens Walker
Electrical and Computer Engineering Department
Assistant Professor Dane Brown
Cyber Science Department
Elizabeth K. Gergal
Midshipman First Class
United States Navy
Drone Swarming Tactics using Reinforcement Learning and Policy Optimization
This project aims to develop defensive drone swarming tactics using reinforcement learning (RL). Swarming is a military tactic where many individually operated units maneuver as one mass to attack an enemy. Defensive swarm tactics are current topics of interest for the US military as other countries and non-state actors are gaining advantages over the US Military because swarm agents are usually simple, inexpensive, and easy to implement. Current work has already developed the means of flying (drones), communicating, and swarming. However, swarms do not yet have the ability to coordinate an attack against an enemy swarm. We simulated drone battles between two swarms of military fixed wings drones using pre-programmed tactics. Even when outnumbered by up to 100%, there were effective tactics that could overcome the difference in size. When used in defense of a ship, these programmed tactics, on average, allowed between 0 and .5 drones to pass the defense and hit the ship which outperforms the current defenses on an Arleigh Burke Class Destroyer and other researched drone swarm defenses. This research shows that it is possible to gain a tactical advantage over an enemy swarm using certain maneuvers and tactics. In order to develop even more effective tactics, we trained an "Agent" tactic using RL. RL is a branch of machine learning that allows an agent to learn an environment, train, and learn which actions that will result in success. The "Agent" tactic does not exhibit emergent behavior yet, but it does kill some enemy drones and outperform other researched RL trained drone swarm tactics. Continuing to implement RL into the development of swarm and counterswarm tactics will aid the US in maintaining our military advantage over our enemies and protecting America's national interests.
FACULTY ADVISOR
Professor Frederick Crabbe
Computer Science Department
Midshipman First Class
United States Navy
Freestream Deceleration Effects on Film Cooling Over a Flat Plate
Film cooling is a method used within gas turbines to cool the turbine blades by blowing air out of small holes in the blade surface of the blades. This allows the gas turbine to be run at higher internal temperatures, increasing the efficiency and power output of the engine. Regions of the curved turbine blade cause the freestream air flowing around it to accelerate and decelerate. This study seeks to determine the effects of this freestream deceleration on film cooling. The experiments accomplish this by using a flat plane with circular, angled holes in the surface to model a small portion of the gas turbine blade. Measurements were made for a zero acceleration case, and two decelerated cases at acceleration parameters (K), a non-dimensional value representing acceleration magnitude, of -0.50×10^-6 and -0.68×10^-6. Three different blowing ratios (M), ratios of the coolant jet velocity to the freestream airflow velocity, were studied in each case. Results show that freestream deceleration has little to no effect on film cooling effectiveness and Stanton number ratio. This information increases understanding of film cooling and allows models to be made more accurate.
FACULTY ADVISOR
Professor Ralph Volino
Mechanical and Nuclear Engineering Department
Elana P. Kozak
Midshipman First Class
United States Navy
Analyzing Behaviors of Artificial Intelligence Methods for a Search Game
Monte Carlo Tree Search (MCTS) is a branch of stochastic modeling that utilizes decision trees for optimization. So far, the method has largely been applied to artificial intelligence (AI) game players. This project imagines a “game" in which an AI player searches for a stationary target within a 2-D grid. We define specific constraints for this search problem and adapt the MCTS method to solve for an efficient path. We analyze its behavior with different target distributions and constraints, including the decision time and domain size. This work covers both a single searcher scenario and the multi-searcher case. The MCTS player is compared to a simple random walk, a nearly self-avoiding random walk, and the Levy Flight Search, a model for animal foraging behavior. We provide data from simulations and prove theoretical results regarding the convergence of the MCTS when computational constraints disappear. Overall, we conclude that a searcher (or multiple) using MCTS is effective against targets with delta-like distributions but quickly loses its strength when the a priori knowledge becomes more vague.
FACULTY ADVISOR
Assistant Professor Scott Hottovy
Mathematics Department
EXTERNAL COLLABORATOR
Dr. Ira Schwartz
Naval Research Laboratory
Sarah M. Nguyen
Midshipman First Class
United States Navy
Parametric Analysis and Optimization of an Elastocaloric Refrigeration Cycle
Elastocaloric heating/cooling takes advantage of the structural changes in shape memory alloys (SMAs) that release and absorb heat in response to an induced strain. Heating/cooling cycles that use SMAs offer a potential solution to issues associated with the use of ozone-depleting refrigerants found in common heat pump/refrigeration cycles. Several recent works have found that the Coefficient of Performance (COP) for elastocaloric cycles is comparable to those achieved with standard HVAC systems of the same scale, which exhibit COPs of 3 on average. Conversely, SMA wires in lab settings have shown to achieve COPs as high as 3.5, and with sufficient optimization are expected to achieve COPs greater than 10. This study is a continuation of the work performed by Sharar et al. (Army Research Lab), which established a first-of-its-kind solidstate continuous flow loop using nitinol (NiTi) wire, whereby the wire rotates continuously in a loop configuration, initiating heat release and absorption when the wire enters and exits a region of curvature [13], [14]. In this work, we explore the parametric space for optimization of this cycle using both computational and experimental methods. The cycle is modeled using COMSOL Multiphysics (v. 5.5). We perform a parametric study that varies the wire radius, disk radius, rotational speed, convection coefficient over the disk, and contact resistance between the wire and disk to optimize its performance and identify parameters that maximize COP, temperature change, and cooling power. Simultaneously, a physical experiment is constructed and paired with the model to validate any trends identified in the simulation. Such insight is critical to improving the performance of these systems in small-to-medium sized heating and cooling systems. This research contributes to the use of elastocalorics in microclimate heating and cooling, and will be a novel contribution to the field of solid-state refrigeration.
FACULTY ADVISORS
Associate Professor Ronald Warzoha
Mechanical and Nuclear Engineering Department
Professor Andrew Smith
Mechanical and Nuclear Engineering Department
Associate Professor Joshua Radice
Mechanical and Nuclear Engineering Department
Assistant Professor Brian Donovan
Physics Department
Zachary Nygaard
Midshipman First Class
United States Navy
Near-Body Velocity and Turbulence Measurements in the Lee of an Inclined 6:1 Prolate Spheroid
Despite dramatic advancements in computing power over the last several decades, computational fluid dynamics (CFD) models have not yet replaced experiments as the primary design and verification tool for the development of air- and water-borne vehicles. The benefits of advanced computational tools include the ability to iterate rapidly on a design, predict a wide range of parameters to include forces and moments, wake characteristics and acoustics, allow for design optimization, all relatively inexpensively. Experiments, on the other hand, are often expensive and time-consuming to conduct, and generally offer results over a narrow range of parameters. However, when conducted carefully at an appropriate model scale, the results provide a measure of certainty that is not yet offered by computational models. To improve CFD models, a more nuanced understanding is required of the flow physics that they describe. This, ironically, requires more experimental data.
To that end, experiments are to be conducted in the large recirculating water channel at the U.S. Naval Academy. The flow in the near-body region of a 6:1 prolate spheroid measuring 0.43 m (17 in.) in length will be examined for length-based Reynolds numbers of 1 to 3x10^6 at angles of inclination of 2.5°, 5°, 10°, and 20°. Boundary layer trip designs will be evaluated and results compared to those of a smooth, un-tripped body. Velocity measurements will be made using a stereo particle image velocimetry (SPIV) system, with sufficient measurements taken to calculate mean flow and turbulence quantities. Results from this experiment will be made available for CFD verification and validation studies. They will also be used to inform follow-on experiments.
FACULTY ADVISOR
LCDR Ethan Lust, USN
Mechanical and Nuclear Engineering Department
Natalie A. Schieuer
Midshipman First Class
United States Navy
Wireless Electromechanical Power Transfer Using Piezoelectric Materials
With the realities of a finite defense budget, efficient systems for power transfer are vitally important for a wide array of applications such as surface ships, submarines, and weapons systems. Piezoelectric materials are an excellent choice for electromechanical power transfer applications owing to their bidirectional conversion between electrical signal and mechanical response. Piezoelectric materials are a specific type of smart materials that are characterized by their ability to induce an electrical charge when subject to a mechanical strain. This phenomenon is bidirectional as it can be observed in reverse, as piezoelectric materials will also undergo mechanical strain when an electrical voltage is induced (Ramadan et al., 2014). In military applications, as well as civilian applications, the use of piezoelectric materials instead of wires allows for reduced mass in power transfer, which is especially applicable to systems designed for flight and/or for orbit. This research has focused on optimizing the location and size of a system of piezoelectric actuators used to transfer electrical power via transduction from electrical voltage to mechanical vibrations and back to electrical voltage.
This research project developed models of the system of interest using COMSOL Multiphysics to consider solid mechanics, viscoelasticity, piezoelectricity, electrostatics, electrical circuits and by introducing structural acoustic coupling. The accuracy of the computational model was validated by comparison with published experimental results for existing hardware. The COMSOL model was used in a computational parametric study of electrical transfer efficiency versus the mechanical and geometric parameters for a single piezoelectric transmitter/receiver pair, where the electrical transfer efficiency is defined as the ratio of the power output to the power input. Through the results of this parametric study, guidelines as to what configurations are responsive and unresponsive for a given excitation frequency were developed. These results led to the novel investigation of the single transmitter/multiple receiver array used to selectively excite a target receiver with a single transmitter.
FACULTY ADVISORS
Associate Professor Joshua Radice
Mechanical and Nuclear Engineering Department
Associate Professor John Burkhardt
Mechanical and Nuclear Engineering Department
EXTERNAL COLLABORATOR
Dr. Sarah Bedair
Army Research Laboratory
Philip B. Smith
Midshipman First Class
United States Navy
Predicting Naval Academy Performance Using Holistic Personality Analysis in Multidimensional Space
Psychological models of personality have used trait measures to index individuals with respect to specific traits or categorical types to bin individuals into defined personality types. Both of these approaches may be suboptimal for predicting performance. Categorical models rely on rough dichotomization of data and trait models may overfit variance to rigid traits and obscure trait interaction effects. This project compared predictions of midshipmen performance and outcomes at the United States Naval Academy, specifically comparing predictions derived from regression techniques with standard traits and type variables with predictions using machine learning techniques, such as k-nearest neighbors and boosted random forests. Using data from recent Naval Academy graduates (N = 725), we first applied traditional penalized regression techniques to predict performance, specifically academic and military order of merit at graduation, using standard personality traits, types, and values. These predictions served as the baseline for assessing the quality of prediction from the selected machine learning techniques. We next built, optimized, and analyzed machine learning models, finding that their accuracies were, at best, at the level of traditional penalized regression models. Finally, we examined the optimization process of the machine learning models to identify potential optimum dimensionalities for personality predictions, finding it matches currently accepted models of personality. While the machine learning models were more complex, computationally expensive, and less interpretable, we found they did not outperform regression models.
FACULTY ADVISORS
CAPT Kevin Mullaney, USN
Leadership, Ethics, and Law Department
Professor William Traves
Mathematics Department
Lilian N. Usadi
Midshipman First Class
United States Navy
Acoustic Manipulation of Microrobots Using Chladni Plates and Multimode Membrane Resonators
The advent of micro/nanorobotics promises to transform the physical, chemical, and biological domains by harnessing opportunities otherwise limited by size. Most notable is the biomedical field in which the ability to manipulate micro/nano particles has numerous applications in biophysics, drug delivery, tissue engineering, and microsurgery. Acoustics, the physics of vibrational waves through matter, offers a precise, accurate, and minimally invasive technique to manipulate microrobots or microparticles (stand-ins for microrobots). One example is through the use of flexural vibrations induced in resonant structures such as Chladni plates.
In this research, we developed a platform for precise two-dimensional microparticle manipulation via acoustic forces arising from Chladni figures - nodal patterns - and resonating microscale membranes. The project included two distinct phases: (1) macroscale manipulation with a Chladni plate in air and (2) microscale manipulation using microscale membranes in liquid. In the first phase (macroscale in air), we reproduced previous studies in order to gain a better understanding of the underlying physics and to develop control algorithms based on statistical modeling techniques. In the second phase (microscale in liquid), we developed and tested a new setup using custom microfabricated structures. The macroscale statistical modeling techniques were integrated with microscale autonomous control systems. It is shown that control methods developed on the macroscale can be implemented and used on the microscale with good precision and accuracy.
FACULTY ADVISORS
Professor Samara Firebaugh
Electrical and Computer Engineering Department
Associate Professor Hatem ElBidweihy
Electrical and Computer Engineering Department
LT Steven Yee
Electrical and Computer Engineering Department
Professor Murray Korman
Physics Department
Joesph B. Wiedemann
Midshipman First Class
United States Navy
Superradiance of Few Driven Two-Level Quantum Dot Emitters in the Bad Cavity Limit
The pursuit of an integrated quantum optics system requires the ability to determine the effect of design parameters on the quantum electrodynamics regime. We develop a simple master equation for few, driven, two-level emitters in the bad cavity regime. Comparing the resulting photonphoton correlation function in the steady state with experimental data from InGaAs quantum dots coupled to a photonic crystal waveguide, we validate our model parameters for coupling strength, spontaneous decay rate and incoherent driving rate. Building upon this model, we explore the superradiant regime for InGaAs quantum dots coherently driven within a waveguide, creating a model that captures the collective behavior of two-level emitters in the bad cavity limit.
FACULTY ADVISORS
Associate Professor Seth Rittenhouse
Physics Department
Assistant Professor Peter Brereton
Physics Department
EXTERNAL COLLABORATORS
Dr. Joel Grim
Dr. Samuel Carter
Naval Research Laboratory
Logan R. Williams
Midshipman First Class
United States Navy
Characterizing the Economic Abundance of Water Through a Theoretical and Empirical Framework
Water is a naturally heterogeneous good. It is valued, priced, and provided differently around the world. Physical abundance is generally represented by the absolute quantity of a resource available, typically relative to the size of a population or the current rate of consumption. Economic abundance is typically measured by “price” of some form and captures what the average person must give up to obtain a certain quantity of the resource, or more precisely, a certain benefit from the resource. This project examines empirical practices of water provision and characterizes economic abundance through a theoretical model of provision, identifying four primary mechanisms: (1) an unregulated, profit-maximizing private firm (“private provision”) (2) non-governmental organization (NGO) provision with water charges used to recover costs (“community-based provision”, (3) NGO provision without water charges (“NGO-aid provision”), and (4) a public provision system with a wide range of possible pricing mechanisms (“public provision”). We then consider a theoretical weighted price metric that would describe the affordability of water in an economy with these four mechanisms serving different portions of the population.
A comprehensive empirical analysis of the economic abundance of water requires understanding and incorporating three primary determinants of abundance: quality, affordability and accessibility. We take this tripartite approach to characterizing abundance through an individual case study of Ethiopia. With regards to trends in the economic abundance of water in Ethiopia, we have reason to be optimistic. First, the percentage of the population served by improved water sources has increased significantly in the last two decades (from 25% in 2000 to 69% in 2017) and continues to increase nationally. Second, water utility charges constitute a relatively small and decreasing portion of monthly household income over time. Lastly, while access to public water utilities is inconsistent in Ethiopia, access to both piped and non-piped water sources has increased significantly over the last two decades, with non-piped access outpacing piped access.
In summary, this project demonstrates that characterizing the economic abundance of water is much more complex than for other goods and natural resources. Such a characterization must necessarily incorporate widely varying provision mechanisms as well as water quality and accessibility in addition to affordability. Our analysis supports a qualified conclusion of increasing global economic abundance of water, and we discuss the implications of this finding in the context of population growth, climate change, technological advancement and increased investment in international economic development.
FACULTY ADVISORS
Professor Kurtis Swope
Economics Department
LT R.J. Armador, USN
English Department
