Mavris, Dimitri N.

Associated Organization(s)
ArchiveSpace Name Record

Publication Search Results

Now showing 1 - 10 of 270
  • Item
    Air Traffic Flow Identification and Recognition in Terminal Airspace through Machine Learning Approaches
    (Georgia Institute of Technology, 2024-01) Zhang, Wenxin ; Payan, Alexia P. ; Mavris, Dimitri N.
    In modern aviation, a significant amount of data is generated during routine operations and collected using technologies like Automatic Dependent Surveillance-Broadcast (ADS-B). The abundance of such data presents great potential for utilizing emerging data analysis techniques like machine learning to enhance the future of aviation. This paper presents a methodology that leverages clustering and classification models for offline identification and online recognition of air traffic flows. This research utilizes real trajectories in the terminal area of Zurich Airport to train and assess various machine learning models. To prepare the raw trajectory data for analysis, we apply a preprocessing step to clean and resample the data. Clustering is performed using the Ordering Points to Identify the Clustering Structure (OPTICS) algorithm, and its performance is compared to Density-Based Spatial Clustering of Applications with Noise (DBSCAN). For classification of the data, we employ two ensemble methods, Random Forest and Extreme Gradient Boosting (XGBoost), and compare their outcomes with those of Long Short-term Memory (LSTM). Our results demonstrate the superior reliability of OPTICS compared to the baseline method for clustering, and the ensemble models perform as effectively as the deep learning model, but with shorter training times due to their relative simplicity. The proposed methodology enhances the understanding of air traffic flows at specific airports and facilitates subsequent trajectory-centric tasks such as anomaly detection, trajectory prediction, and conflict detection, ultimately contributing to the improvement of safety in the terminal airspace.
  • Item
    Aviation-BERT: A Preliminary Aviation-Specific Natural Language Model
    (Georgia Institute of Technology, 2023-06) Chandra, Chetan ; Jing, Xiao ; Bendarkar, Mayank ; Sawant, Kshitij ; Elias, Lidya R. ; Kirby, Michelle ; Mavris, Dimitri N.
    Data-driven methods form the frontier of reactive aviation safety analysis. While analysis of quantitative data from flight operations is common, text narratives of accidents and incidents have not been sufficiently mined. Among the many use cases of aviation text-data mining, automatically extracting safety concepts is probably the most important. Bidirectional EncoderRepresentations from Transformers (BERT) is a transformer-based large language model that is openly available and has been adapted to numerous domain-specific tasks. The present work provides a comprehensive methodology to develop domain-specific BERT model starting from the base model. A preliminary aviation domain-specific BERT model is developed in this work. This Aviation-BERT model is pre-trained from the BERT-Base model using accident and incident text narratives from the National Transportation Safety Board (NTSB) and AviationSafety Reporting System (ASRS) using mixed-domain pre-training. Aviation-BERT is shown to outperform BERT when it comes to text-mining tasks on aviation text datasets. It is also expected to be of tremendous value in numerous downstream tasks in the analysis of aviation text corpora.
  • Item
    BERT for Aviation Text Classification
    (Georgia Institute of Technology, 2023-06) Jing, Xiao ; Chennakesavan, Akul ; Chandra, Chetan ; Bendarkar, Mayank ; Kirby, Michelle ; Mavris, Dimitri N.
    The advent of transformer-based models pre-trained on large-scale text corpora has revolutionized Natural Language Processing (NLP) in recent years. Models such as BERT (Bidirectional Encoder Representations from Transformers) offer powerful tools for understanding contextual information and have achieved impressive results in numerous language understanding tasks. However, their application in the aviation domain remains relatively unexplored. This study discusses the challenges of applying multi-label classification problems on aviation text data. A custom aviation domain specific BERT model (Aviation-BERT) is compared against BERT-base-uncased for anomaly event classification in the Aviation Safety Reporting System (ASRS) data. Aviation-BERT is shown to have superior performance based on multiple metrics. By focusing on the potential of NLP in advancing complex aviation safety report analysis, the present work offers a comprehensive evaluation of BERT on aviation domain datasets and discusses its strengths and weaknesses. This research highlights the significance of domain-specific NLP models in improving the accuracy and efficiency of safety report classification and analysis in the aviation industry.
  • Item
    Development of a Parametric Structural Analysis Environment to Support the Design, Manufacturing, and Production of a Composite UAV Wing
    (AIAA, 2023-01) dos Santos, Marcos ; Cox, Adam ; Fischer, Olivia J. Pinion ; Mavris, Dimitri N.
    As digital and physical systems become more complex, collect voluminous quantities of data, and move towards greater integration over time, the concept of digital engineering has become of great interest in the engineering world. The integration of digital methods with traditional engineering approaches in product lifecycle management has posed challenges on how techniques such as digital twins can be best used during the design and manufacturing phases of the product lifecycle. To address this need, this research supports the integration of design, manufacturing, and production by assessing the structural integrity of various designs of a parametric UAV wing built with a composite material. A systematic and efficient environment is developed to modify the wing design parameters, develop and analyze the finite element model, obtain structural data, and identify feasible design regions for decision making. The sharing of models, data, and analyses with the manufacturing and production segments of the lifecycle permits integration of the various disciplines in early design phases to allow greater design freedom and avoid great costs during the design of the product. The results indicate that (1) the need for a trade-off analysis between key disciplinary considerations in UAV wing design decision making can be addressed and that (2) the developed capability enables decision makers to choose the configurations to be studied in later design stages after the structural integrity and weight considerations are assessed for multiple wing designs.
  • Item
    Development of an Open Rotor Propulsion System Model and Power Management Strategy
    (Georgia Institute of Technology, 2023-01) Clark, Robert A. ; Perron, Christian ; Tai, Jimmy C. M. ; Airdo, Benjamin ; Mavris, Dimitri N.
    The development of an open rotor propulsion system architecture model and fuel burn-minimizing power management strategy is investigated. The open rotor architecture consists of a single-rotor open rotor (SROR) connected to the low speed shaft of a traditional turbojet engine in a puller configuration. The proposed architecture is modeled in the Numerical Propulsion System Simulation (NPSS) tool, and performance is evaluated across a complete flight envelope typical for a narrow body commercial airliner. Rotor performance maps are generated using a custom blade element momentum theory (BEMT) code, while compressor performance maps are created using CMPGEN. The performance of the overall propulsion system is detailed in the context of a notional 150 passenger aircraft mission, and a method for scheduling rotor power across the flight envelope is developed in order to minimize aircraft mission fuel burn. It is demonstrated that the power absorbed by the rotor can be optimized by scheduling rotor blade pitch angle versus fan speed. A power management technique using the optimal blade pitch angle at only six points in the flight envelope was shown to provide significant computational benefits without sacrificing any fuel burn when compared to a method using a schedule generated from data across the complete flight envelope.
  • Item
    Uncertainty Quantification of a Parallel Hybrid-Electric Propulsion EPFD Vehicle
    (AIAA, 2023-01) Uzodinma, Jaylon ; Zaidi, Turab ; Walter, Miguel ; Gautier, Raphael ; Mavris, Dimitri N.
    The NASA Electrified Powertrain Flight Demonstration (EPFD) program is a collaboration between industry and academia to accelerate the development and implementation of megawatt-class power systems in commercial aviation. Technology development programs are often associated with cost, performance, and schedule risks, which can result from technical uncertainty. To assess and offer insight to effective mitigation of risks associated with the NASA EPFD program, an uncertainty quantification analysis for future hybrid-electric commercial aircraft is addressed. An uncertainty analysis is presented for the electrified aircraft propulsion systems of a 150-passenger hybrid-electric aircraft model. Uncertainty at the component-level of the powertrain system is considered and its effect is propagated to vehicle-level metrics. The primary focus is identifying and assessing the key uncertain technological inputs driving the variability of the vehicle’s performance responses.
  • Item
    Decision-Making and Optimization Framework for the Design of Emerging Satellite Constellations
    (AIAA, 2023-01) Koerschner, Marc A. ; Krishnan, Kavya Navaneetha ; Payan, Alexia P. ; Mavris, Dimitri N.
    With the parallel increase in global orbital debris due to passive object collisions, as well as in the number of proposed low earth orbit mega-constellations, in anti-satellite missile tests, and the fielding of new satellites, there is an inherent need for a framework to optimize the design of Low Earth Orbit (LEO) mega-constellations to avoid collisions while maintaining the functionality of the constellation. In this paper, we aim to provide a framework that unifies these considerations in the conceptual design phase of mega-constellations. We start with a discussion of metrics of importance for the design of mega-constellations, namely coverage, collision risk, collision avoidance, and station-keeping costs. With these metrics defined, we utilize the first principles of orbital mechanics and statistical models to analyze potential alternative mega-constellation designs. These designs are then optimized using Non-denominated Sorting Genetic Algorithm 2 (NSGA2) with our own defined objective function to create a repository of Pareto optimal configurations. We then showcase how a multi-criteria decision-making methodology can be utilized by a variety of unique stakeholders and subject-matter experts to select an optimal constellation design for a given scenario. A Pareto Frontier collection with optimal solutions of 10 constellations was produced by the framework. Radar plots to assess the significance of the weighted metric of the framework shows several trading options for conceptual designs of the constellations. We finally discuss the scope, limitations, applications, and future work for various scenarios.
  • Item
    A Feasibility Study for the Development of Air Mobility Operations within an Airport City (Aerotropolis)
    (AIAA, 2023-01) Wang, Xi ; Balchanos, Michael G. ; Mavris, Dimitri N.
    This study aims to create a simulation environment to study the feasibility of an Advanced Air Mobility (AAM) system in an airport-centric infrastructure or aerotropolis area. The environment and the study are to confirm the effectiveness of the AAM system in terms of reducing traffic congestion for road networks and the reduction in carbon emissions for transportation methods. The traffic simulation will run a baseline simulation with the currently available mobility methods and an alternative simulation with a proposed small network with close distance flights AAM system of 9 vertiports. The traffic modeling utilizes Agent-Based Modeling (ABM) to accurately models the two cases and compare trip times of the two cases. The emission modeling models the emission of carbon per mile of travel for different mobility methods and use the miles traveled from the traffic simulation to calculate the emission. The conclusion was drawn based on the two comparisons of the change in travel time and the change in emission. A small AAM system servicing a small area with short flight legs is found to be effective in both decreasing trip times and decreasing emissions and is significantly more effective when the ground mobility network is congested and not accessible.
  • Item
    Surrogate Modeling of Orbital Decay of Lunar Orbits
    (AIAA, 2023-01) Varoqui, Maxime ; Steffens, Michael J. ; Mavris, Dimitri N.
    Operations in cislunar space are expected to greatly increase over the next decade, which will place a heightened demand on satellites operating in cislunar space. The orbit selection of the satellites is a key parameter of the mission. Orbital decay can present significant challenges for some lunar orbits due to gravitational perturbations. This study focuses on developing a fast method to assess the decay of lunar orbits. The method is based on modeling lunar orbits propagation in the presence of these perturbations to quantify orbital decay as a function of orbital parameters, then using the model to generate data and fit surrogate models. Results from this effort will enable decision makers to trade performance and station-keeping costs associated with relevant lunar orbits.
  • Item
    A Bibliometric Approach to Characterizing Technology Readiness Levels Using Machine Learning
    (Georgia Institute of Technology, 2023-01) Dastoor, Jehan ; Zhang, Heying ; Balchanos, Michael G. ; Mavris, Dimitri N.
    As cislunar space becomes more accessible to national space agencies and commercial entities, there is a constant need to improve the way in which space missions are planned, and development progress is tracked. A technologies stage of development, which is related to mission budget and schedule, is typically quantified using technology readiness levels (TRL). The process of determining TRL is often long and laborious, and requires the use of subject matter experts. As a part of the Georgia Institute of Technology Cislunar Architecture Initiative, this work serves to develop the early stages of an environmental scanning approach to maturity assessment that allows for the automatic determination of a technologies TRL using machine learning ordinal regression techniques with bibliometric factors. The bibliometric factors considered were: scientific publications, National Science Foundation awards, patents, and NASA Spinoff articles. Annual data on these factors was collected for 31 technologies between 1995-2015 using public APIs, and S-curves fit to the data to estimate each technologies point in the development cycle. The final model performed with an R² of 0.817, 0.812, and 0.567 on the training, validation, and test data, respectively. Additionally, a better performing model to classify a technologies technology life cycle phase was created and drawbacks to this approach discussed.