Naval Postgraduate School
Calhoun, Institutional Archive of the Naval Postgraduate SchoolNot a member yet
71232 research outputs found
Sort by
ANOMALY DETECTION IN A MICROGRID USING MACHINE LEARNING METHODS TO ENHANCE POWER SECURITY
No full textMicrogrids are a critical tool for the U.S. Navy to keep ships powered and forward deployed for the duration of their operations. The microgrid has the advantage of a localized system disconnected from outside entities while in operation. However, like any complex electrical system, microgrids are susceptible to internal faults, component breakdown, and system inefficiencies. The early detection of these issues helps avoid unexpected failures that compromise mission capability and system integrity. This thesis aims to advance predictive maintenance strategies and enhance power security by detecting power anomalies in a fully developed microgrid. This research uses an established microgrid testbed with three modes of operation as a testing ground to create several datasets. Power data is collected using a non-intrusive power monitoring probe and analyzed through supervised machine learning algorithms to detect abnormal patterns indicative of operational deviations. Various power anomalies—including power failure, sag, surge, undervoltage, and overvoltage— are deliberately introduced into the system. To improve anomaly detection, a range of power features beyond standard voltage and current measurements are incorporated, enhancing the training and predictive accuracy of the machine learning models. By implementing this detection method, this work supports the increased reliability of the microgrid and enables proactive maintenance decisions.Distribution Statement A. Approved for public release: Distribution is unlimited.Lieutenant, United States NavyDeputy Assistant Secretary of the Navy-Operational Energy Offic
EVALUATING THE RELIABILITY OF AI DETECTORS
No full textSynthetic media poses a challenge for the Marine Corps, specifically artificial intelligence (AI) generated text that adversaries may use to mass produce Marine Corps communications to spread inauthentic messages. The detection of AI-generated text is important for ensuring the authenticity of communications against adversarial threats. Current detection tools focus on general text, such as academic papers, and are not tailored to military content. Research regarding the reliability of AI-generated text detectors has a narrow focus and even fewer detectors. This thesis helps close this gap by evaluating an expanded set of AI text detectors using datasets of press releases that are human and AI-authored. Converting detector outputs into binary classification to evaluate the performance using confusion matrices, sensitivity analysis, and area under the curve (AUC). The results show that GPTZero achieves the highest accuracy in terms of sensitivity and specificity. Scribbr performs moderately well, specifically with Claude-authored press releases. In contrast, ZeroGPT and Sapling performed less accurately with more false negatives and false positives. An ensemble model, specifically with soft voting outperforms the majority of the individual detectors, achieving an AUC of 0.98. These findings demonstrate that combining detection tools provides robustness to support defense from AI-generated misinformation imitating military press releases.Distribution Statement A. Approved for public release: Distribution is unlimited.Captain, United States Marine Corp
AN EXTENDED STOCHASTIC SALVO MODEL IMPLEMENTING MULTI-ROUND LAYERED DEFENSE
No full textHughes developed the first salvo combat model, which mathematically models the coordinated firing of multiple missiles simultaneously between warships. Armstrong developed a stochastic version of Hughes’s model to incorporate probability and statistics into the model. Like Hughes’s model, Armstrong’s model keeps several simplifying assumptions: it simulates only a single round of engagement, assumes all missiles are launched in a single round, does not account for layered defenses, and does not track missile inventory. This thesis extends Armstrong’s model by implementing a series of improvements in Matrix Laboratory (MATLAB) to increase fidelity and realism. Experimental results suggest that offensive power has a greater influence on win probability compared to defensive power. Gaining a size advantage early has a significant effect on the outcome. Experimental results also showed that the model can be used to perform a tradeoff analysis with cost constraints of different warship attributes such as offensive power, defensive power, and staying power.Distribution Statement A. Approved for public release: Distribution is unlimited.Ensign, United States Nav
Faces of NPS: Senator Mark Kelly
No full textFaces of NPS features interviews spotlighting the students, faculty, staff and alumni of our Nation's premier defense education and research institution
Faces of NPS: Capt. Madison Tikalsky, USAF
No full textFaces of NPS features interviews spotlighting the students, faculty, staff and alumni of our Nation's premier defense education and research institution
Faces of NPS: Karen Hargrove
Get PDFFaces of NPS features Interviews spotlighting the students, faculty, staff and alumni of our Nation’s premier defense education and research institution.Link to video can be found at: https://youtu.be/GY7khOxLhpw?si=4nrfY2gGDTPKp9E
Faces of NPS: Lt. Col. Pedro Ortiz, USA
No full textFaces of NPS features interviews spotlighting the students, faculty, staff and alumni of our Nation's premier defense education and research institution
MITIGATING THE LOSS OF INSTITUTIONAL KNOWLEDGE: ANALYZING KNOWLEDGE RISK MANAGEMENT STRATEGIES FOR THE OFFICE OF NAVAL RESEARCH'S ACQUISITION WORKFORCE
No full textThe United States federal workforce is grappling with a significant issue: the gradual loss of institutional knowledge. Over decades, accumulated expertise, practical experience, and historical insights have formed a vital backbone that supports government efficiency, continuity, and adaptability. If this knowledge isn’t properly preserved and passed on, we risk undermining the very foundation of government operations. Consequently, it is crucial for government agencies to implement knowledge risk management (KRM) strategies that prioritize not only the retention but also the active transfer of this essential resource.This capstone investigates the potential risks associated with institutional knowledge loss within the Office of Naval Research’s (ONR’s) acquisition workforce. A comprehensive literature review, coupled with a survey administered to ONR’s acquisition personnel, revealed critical weaknesses in areas such as the current knowledge management system (KMS), offboarding procedures, and the practice of regular knowledge audits. In response, the study puts forth a series of KRM recommendations to bolster these vulnerable areas while also incorporating feedback from the survey – suggesting enhanced informal knowledge-sharing practices and a greater reliance on cutting-edge technologies like artificial intelligence.Distribution Statement A. Approved for public release: Distribution is unlimited.Civilian, Department of the Nav
WHAT DO PALAUANS SEE? AMERICA'S STRUGGLE TO REMAIN THE TOP PARTNER IN OCEANIA
No full textThis study explores the strategic competition between the United States and China in Palau, a small Pacific island nation whose geopolitical importance has grown due to its continued diplomatic recognition of Taiwan and its strategic location. As both powers seek to expand their influence, Palau has become a focal point in the evolving contest for regional dominance, particularly within the information environment. China’s increasing use of soft power and influence operations poses a challenge to U.S. interests, prompting Washington to reinforce its engagement with Palau. This research assesses how China’s information activities influence Palau’s public perception, contrasts these efforts with U.S. strategies, and considers the broader implications for regional stability and international competition.Distribution Statement A. Approved for public release: Distribution is unlimited.Lieutenant Colonel, United States Marine Corp
EVALUATING THE SUITABILITY OF SGP4 FOR HIGH PRECISION SATELLITE PROPAGATION IN LOW EARTH ORBIT
No full textIncludes Supplementary MaterialSupplemental 2:2 Classified. Forthcoming.The propagation accuracy of satellites in low Earth orbit (LEO) depends heavily on the propagation model used. SGP4 is one such model that is widely used due to the frequency and availability of two-line element (TLE) updates from the U.S. Space Force. Since LEO satellites are significantly impacted by stochastic atmospheric conditions, this thesis seeks to assess SGP4’s viability for high precision applications. To quantify viability, this thesis constructed probability distribution functions using historical propagation and atmospheric data to determine trends in propagation errors over time. The effects on propagation accuracy are separated by axis in the radial, in-track and cross-track frame. Analysis showed that in-track errors were greatest due to the unpredictability of atmospheric density, with lower satellites experiencing greater impacts. Further analysis indicates that propagation accuracy is unaffected by density variations during the TLE generation process. Additionally, the impact of station-keeping maneuvers was examined, revealing reduced accuracy for TLEs generated within three days of a maneuver. These findings clarify the conditions under which SGP4 maintains high accuracy and highlight its limitations in precise applications.Distribution Statement A. Approved for public release: Distribution is unlimited.Major, United States Marine Corp