136879 research outputs found
Sort by
Impact of Organic Management Practices on Soil Greenhouse Gas Emissions from Cotton-Winter Cover Crop Systems in East-Central Texas
No full textConventional cotton production practices demand extensive management involving high pesticides, fertilizers, and tillage inputs, contributing significantly to environmental impacts. To address these concerns, we explored the potential of organic cotton systems employing diverse cover crops, manure, and biochar to mitigate soil greenhouse gas (GHG) emissions while preserving soil carbon and nitrogen. Our field experiment, conducted in the humid subtropical climate of East-Central Texas, assessed various cover crops such as oats (Avena sativa), Austrian winter pea (Pisum sativum), purple top turnip (Brassica rapa subsp. rapa), and mixed species cover crops compared against a control (no cover crop) over three consecutive years. We monitored soil GHG emissions, moisture, and temperature dynamics throughout the cover crop and cotton seasons.
Simultaneously, we conducted multiple laboratory incubations to assess the carbon and nitrogen mineralization of these cover crops in combination with poultry litter manure, examining associated GHG emissions. Our laboratory simulations also considered the impact of tillage practices on residue mineralization. Additionally, we investigated the potential of cotton residue biochar to mitigate GHG emissions during cover crop and manure decomposition.
Our findings revealed that cover crops with a lower C:N ratio, especially legume and mixed species, exhibited higher GHG emissions. However, incorporating biochar alongside cover crops demonstrated significant emission reduction. Furthermore, we observed that cover crops led to reduced soil moisture during their growth phase but contributed to increased water retention during cotton seasons
Transparency, Accuracy, & Uncertainty in Human-AI Collaborative Decision-Making for Spacecraft Anomaly Diagnosis
No full textAI agents are becoming increasingly ubiquitous in a variety of domains, from safety-critical environments to day-to-day activities. These days they are being considered more as a virtual peer rather than a decision-making tool. In the coming years, AI agents will have a key role to play in spaceflight missions that will voyage beyond low earth orbit, where communication delays with the ground control will become longer and more frequent. On-board AI agents can help the crewmembers detect, diagnose, and treat spacecraft anomalies faster, giving them more autonomy and allowing them to respond faster to emergencies, or to focus on other critical aspects of their mission.
In order for any technology to be accepted and used by the operators, a sufficient amount of trust needs to be established first. Trust in automation is a key factor that determines willingness of a human operator to rely on an AI agent. Previous research on trust in an AI agent highlights some key elements that influence its development, such as its transparency, accuracy, and reliability. However, having perfectly accurate and reliable agents may not be possible or even enough to establish trust, especially in scenarios where there is significant uncertainty in the agent���s recommendations. In light of this fact, the link between trust, accuracy, and uncertainty merits further examination. This dissertation aims to elucidate this potential link in an agent that provides explanations for its recommendations compared to one that does not explain its decisions to the user.
This thesis presents the development and use of an AI-agent, Daphne, for detecting, diagnosing, and treating spacecraft anomalies related to the Environment Control and Life Support Systems (ECLSS). We present an experiment where human operators rely on Daphne���s recommendations induced with various levels of inaccuracies and uncertainties to detect and diagnose ECLSS-based anomalies. Human performance (number of anomalies correctly diagnosed and time to diagnosis), trust, situational awareness, cognitive workload, satisfaction, and confidence in their response were measured using both objective and subjective techniques.
Our results show that the effects of automation transparency can influence operator task performance, trust, situational awareness, workload, user confidence, satisfaction, and appropriate reliance positively. Results also suggested that agent accuracy improved task performance, appropriate reliance, and partially improved user confidence, while trust, workload, and SA were not significantly affected. Results also showed that uncertainty in agent���s recommendations reduces task performance, trust, situational awareness, user confidence, satisfaction, and appropriate reliance, and increases mental workload.
Overall, this work sheds light on under-investigated issues in Human-AI Collaboration by providing insights on factors that are most likely to effect the human-AI relationship during long duration exploration missions for spacecraft anomaly diagnosis. Further, this work provides recommendations and guidelines for designers and developers of XAI systems for developing transparent AI agents to support operators in time- and safety-critical tasks and environments, such as crew members during long-duration exploration missions
Effect of Bone Segementation Data of the Maxilla and Mandible on the Accuracy of Bone-Supported Surgical Guides
No full textPurpose: The purpose of this study was to identify whether the quantity of Cone Beam Computed Tomography Scan (CBCT) image slices included in the jaw segmentation process has a significant effect on the accuracy of bone-supported guides and whether there is a minimum number of slices that can be identified as being needed to produce a surface model of the mandible and the maxilla that will yield an accurate bone-supported guide.
Materials and Methods: Dried human edentulous mandibles and maxillae, three of each, were selected. Each bone model was digitized using an intraoral scanner, and each scan was saved as a Standard Tessellation Language (STL). A bone-supported surgical guide was created on the surface scan of each jaw using a 3D implant planning software program, each serving as a reference guide for each specimen. CBCT images were then acquired once for each bone model, which were then imported into an implant planning software in digital imaging in communication in medicine (DICOM) format. A CBCT image segmentation procedure was then performed for each jaw to create a surface model of each bone with varying numbers of CBCT slices included for each segmentation: 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100 slices. A surgical guide was then created on each surface model rendered from CBCT segmentations and saved as STL. Then the STL of each experimental guide and its corresponding reference guide were imported into an image analysis program and Root Mean Square (RMS) values and heat maps generated to study the internal fit deviation of each guide.
Results: There were statistically significant differences (P<0.001) between the guides created from different numbers of CBCT slices segmented. Additionally, there were statistically significant results between the mandibular and maxillary models, with the mandibular groups having a higher mean deviation (238 ��m) than the maxillary guides (230 ��m). For the mandibles combined, groups below 70 slices produced significant results, while for the maxillae combined, anything below 90 slices was significant.
Conclusions: The internal fit of bone-supported guides is affected by the number of CBCT slices used in the jaw segmentations process for rendering the bone model on which the guide is created. There is a difference in the minimum number of slices needed to create an accurate bone supported guide on the maxilla and mandible. The threshold number of slices is higher for the maxilla compared to the mandible. Although differences between the different guides were observed, the clinical implications are not as clear and should be further studied
Ultrasonic Spot Welding of FFF Printed Samples as a Means of Improving Interlayer Adhesion
No full textAdditive manufacturing (AM) is a technology that has improved manufacturing capabilities in a huge variety of industries, as well as increasing rapid prototyping capabilities. Fused filament fabrication (FFF) is one of these technologies, that has seen use in both industry and hobbyist settings to print thermoplastics and polymer matrix composites that have a thermoplastic matrix. However, it suffers from a major flaw in that the strength of printed parts is anisotropic and uneven, with tensile strength in the direction perpendicular to the printed layers, the interlayer adhesion, being anywhere from 50-75% lower than in the other orthogonal directions for fully dense parts made from standard glassy thermoplastics. This study attempts to improve this weakness by introducing an ultrasonic welding process to the normal FFF 3D printing method. Two welding treatments are attempted following different patterns, one where there are overlapping welds to cover as much surface area as possible, and another with no overlapping welds to prevent the risk of damaging or over-welding the part. The samples, as tested by a modified version of ASTM D5528, showed drastically increased maximum fracture load, suggesting a much stronger level of layer adhesion. Nylon samples showed a change from an average of 85.94 N for the control samples and 191.2 N for the best performing welding group average. ABS samples showed an average of 68.3 N for the control group and 102.58 N for the best performing welding group average. More testing will be required to produce a procedure that can be reliably applied to printed parts, but the procedure used in this study proved effective for both ABS and Nylon double cantilever beam samples
Developing the Texas A&M Smart and Connected Homes Testbed
No full textThe Texas A&M Smart and Connected Homes Testbed has been developed to support residential HVAC research. The flexible testbed enables the windows and walls to be replaced, the floorplan to be reconfigured, provides two separate duct networks, and incorporates on-site renewable energy. Heavy instrumentation is done at the testbed to capture information on local weather conditions, building envelope performance, occupant comfort, HVAC equipment performance, and the power consumption of all household end-uses. A smart thermostat is also incorporated to provide HVAC control capabilities. Occupants are simulated inside the home to mimic actual operation and internal loads. A Modelica model has been created for the building envelope and split system HVAC at the testbed. Using data from the experimental testbed, the model is tuned to ensure accurate implementation. Researchers can use these models to bridge the gap between simulation-based studies and their real-world application
An Evaluation of Alternatives for Updating Base Acres in the 2024 Farm Bill
No full textGiven that the income-support provisions in the farm bill have been decoupled from production for more than two decades, base acres are no longer reflective of planted acres in the United States. Three alternatives are currently being discussed to better align base acres with current plantings in the next farm bill: 1) a reallocation of bases to current plantings, 2) a forced update to current plantings, and 3) a rolling average of the plantings from the previous two years. There undoubtedly will be winners and losers across individual operations, crops, and regions. Knowing who will be impacted would enable Congress to make more informed decisions regarding base updates. To comprehensively evaluate the alternatives, this study undertook a farm-level and national analysis.
The farm-level analysis was implemented by collecting data from four representative farms maintained by the Agricultural and Food Policy Center at Texas A&M University. For each farm, base acres were calculated for the three alternatives, and stochastic simulation was used to determine a five-year forecast of average government payments, ending cash, and ending real net worth. The national analysis was conducted by collecting public data from the Department of Agriculture���s Farm Service Agency for nine covered commodities and calculating new base acres per county to determine absolute gains and losses. The absolute change per crop per county was totaled to determine the deviations between the current baseline and each scenario being analyzed.
The results from the farm-level and national analysis indicate a wide variety of impacts on individual farms, crops, and regions. While maintaining current base acres might be the optimal solution for one location or commodity, it is apparent that the same scenario would be less preferred by others. One universal option (e.g., to reallocate base acres) might not be the best solution; in fact, allowing producers to choose amongst several options might be the best route for baseline modification in the next farm bill, recognizing that approach will also cost the most
Investigation of Co-resident Attacks in Serverless Cloud Environment
No full textIn the rapidly evolving landscape of cloud computing, serverless environments have gained prominence for their scalability and efficiency. However, the security of such environments re-mains a critical concern given how ubiquitous their implementation is in both academic and industrial settings. This study delves into the realm of co-resident attacks within serverless cloud con-texts, aiming to exploit the weak isolation within software infrastructure implementation. Specifically, this research investigates the potential flaws in the Function as a Service (FaaS) framework and sheds light on the vulnerabilities that arise when multiple tenants share the same underlying hardware. By identifying these vulnerabilities, this research aims to improve the security of serverless cloud domain and explores avenues to protect against co-resident attacks. This work specifically investigates the feasibility of cache covert channel attacks within serverless environments deployed on commercial cloud platforms. In this study, the open-source Apache OpenWhisk function-as-a-service (FaaS) is deployed on top of a Kubernetes (K8s) cluster. The cluster is pro-visioned and managed using Google Kubernetes Engine (GKE) on the Google Cloud Platform (GCP) to emulate a real-world scenario. With this, the potential of malicious actors establishing covert communication channel across co-resident functions is investigated. The research is conducted in a shared host environment within GCP, emphasizing on virtual CPU (vCPU) resource sharing. Dedicated hosts are intentionally avoided to simulate real-world cloud deployment scenarios. Our findings indicate that commercial-grade clouds like GCP inherently provide a degree of security obfuscation, making covert channel establishment more challenging. While attacks may be more feasible on dedicated single-tenant machines, the shared nature of commercial cloud environments adds a layer of complexity. Serverless frameworks, while introducing new abstractions, ultimately rely on the same underlying infrastructure; therefore, they introduce additional noise rather than fundamentally altering the attack surface. While effective defense mechanisms can be implemented, they invariably introduce performance overheads
Common Operating Picture Enhancement for Cyber-Physcial Data and Under Contingencies
No full textThe power grid is the foundation of contemporary society's infrastructure, so it is crucial to take strong cybersecurity precautions to guard against potential disruptions that could have a significant impact on the stability and functionality of society. The power grid is particularly vulnerable to new cyber threats that pose a growing threat to its resilience.
This research explores the crucial role of common operating pictures in bolstering the cybersecurity of critical infrastructure with easily understandable data. The primary approach to depicting the common operating picture involves the utilization of Graphical User Interfaces (GUIs), which serve as digital platforms enabling operators to engage with physical equipment via visual graphics and graphical icons. The study investigates the Cyber-Physical Resilient Energy Systems Energy Management System (CYPRES EMS) and places a significant emphasis on Cyber-Physical Situational Awareness (CyPSA).
Furthermore, the study illuminates the advantages of integrating a risk matrix graph within and environment called CYPSA live environment. This integration utilizes data on common vulnerabilities and exposures sourced from the National Vulnerabilities Database (NVD). The inclusion of the risk matrix graph provides heightened clarity on the intricacies of the risk landscape, equipping users with discerning options for informed decision-making.
This study builds upon established analytical methodologies, such as attack tree graphs and a method called Failure Modes, Effects, and Criticality Analysis (FMECA), to comprehensively identify and address potential risks and threats to the system.
In conclusion, the results underscore the potential insights derived from combining these technologies, offering a more effective means to protect critical infrastructure from evolving cyber threats and vulnerabilities. The findings contribute to the growing body of knowledge aimed at bolstering cybersecurity measures for vital systems
Human-Centered AI for Precision Medicine: Methods and Applications
No full textAI has emerged as a powerful tool in the healthcare and biomedical domains. In the field of medicine, AI must demonstrate strong performance while adhering to human ethics. Throughout my Ph.D., I focused on designing and developing human-centered AI tools for precision medicine, with a special emphasis on addressing ethical concerns in medical AI. To bridge the gap between AI and medicine, I delved into cutting-edge AI methods, including knowledge distillation, reinforcement learning, multi-task learning, multi-modality learning and contrastive learning. To make our contribution focused, I specialized in the phenotyping, disease diagnosis, organ transplant, and health event prediction scenario which are essential medical tasks. Four-folds challenges of designing human-centered AI framework towards precision medicine are summarized as (1).Trade-off between performance and fairness, (2).AI integration in clinical workflow, (3).Multi-task prediction on related medical indicators. (4).Multi-modality EHR data. To comprehensively investigate the fairness issue in the clinical prediction algorithm, I conduct extensive experiments on the disease diagnosis to benchmark the performance and bias in the electronic phenotyping. I design a twostep debiasing strategy with unbiased knowledge distillation to predict the graft failure after liver transplant fairly and precisely. In order to support the doctor���s clinical decision, FairAlloc framework is proposed to directly generate accurate and unbiased patient prioritization decisions with reinforcement learning. To simultaneously predict highly related medical indicators, CoD-MTL is designed to take advantage of multiple highly related tasks to predict multiple cause-of-death after the liver transplant. When it comes to multimodal EHR data, I design a cross-modality knowledge distillation framework to distill the knowledge from LLM into the predictive model on structured EHR. My research efforts paved a way to design powerful and trustworthy AI frameworks to support precision medicine with human-centered AI principles
John Bickham field notebook: AK25001-AK25500.pdf
No full textBound book, each page corresponds to a karyotype slide data.Data pages for AK25501-AK26000 corresponding to unique identifiers of specimens/samples examined for biological research. Specimens are primarily housed at Texas A&M University; Biodiverstiy Research and Teaching Collection