Search NASA⌕ Search

SEARCH · Search NASA

Results for “traffic optimization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

501 records · Page 28
All sourcesNASA NTRSDOE OSTI
NASA NTRS2006

Technology Developments Integrating a Space Network Communications Testbed

As future manned and robotic space explorations missions involve more complex systems, it is essential to verify, validate, and optimize such systems through simulation and emulation in a low cost testbed environment. The goal of such a testbed is to perform detailed testing of advanced space and ground communications networks, technologies, and client applications that are essential for future space exploration missions. We describe the development of new technologies enhancing our Multi-mission Advanced Communications Hybrid Environment for Test and Evaluation (MACHETE) that enable its integration in a distributed space communications testbed. MACHETE combines orbital modeling, link analysis, and protocol and service modeling to quantify system performance based on comprehensive considerations of different aspects of space missions. It can simulate entire networks and can interface with external (testbed) systems. The key technology developments enabling the integration of MACHETE into a distributed testbed are the Monitor and Control module and the QualNet IP Network Emulator module. Specifically, the Monitor and Control module establishes a standard interface mechanism to centralize the management of each testbed component. The QualNet IP Network Emulator module allows externally generated network traffic to be passed through MACHETE to experience simulated network behaviors such as propagation delay, data loss, orbital effects and other communications characteristics, including entire network behaviors. We report a successful integration of MACHETE with a space communication testbed modeling a lunar exploration scenario. This document is the viewgraph slides of the presentation.

hybrid simulation↗
NASA NTRSDate not supplied

Traffic Flow Analysis for Package Delivery Drones using a Queueing Model

A key component of the small unmanned aircraft systems traffic management ecosystem is the design of scalable algorithms for strategic deconfliction of drones prior to takeoff. In this work, we focus on efficient flow management of drones on a network of intersecting edges subject to two kinds of spacing constraints: 1) between any two adjacent vehicles on an edge and 2) between any two vehicles on two different edges arriving one after the other at an intersection. The spacing is designed to enable non-intersection of operational volumes corresponding to two different vehicles thereby properly separating the vehicles inside each volume. For simplicity, we assume a constant ground speed for the drones and fixed dimensions for the operational volume blocks. The deconfliction is managed by adjusting the takeoff time of the drones, thereby regulating their arrival time at various crossing waypoints in the network. This framework allows us to study the maximum flow (throughput) of vehicles on a network of edges connecting depots to drop off sites subject to the temporal spacing constraints. The departure scheduling of individual drones results in a combinatorial optimization problem. To alleviate this, we solve a max-flow formulation and use queueing theory to simplify the analysis and provide upper bounds to the underlying optimization problem for individual drone departure scheduling. Our results indicate that throughput drops rapidly after the density of drones in the network passes the max-flow limits.

Alexey A Munishkin↗
DOE OSTI2026

Human Factors and Technologies Design to Improve User Acceptance of Pooled Rideshare for Increasing Transportation System Energy Efficiency

This multi-year project delivered a comprehensive, human-factors-driven framework to understand, model, and improve pooled rideshare (PR) adoption in the United States. Through three large-scale national survey studies involving more than 16,000 participants across multiple cities and demographic groups, the research established one of the most extensive datasets to date on user perceptions, behavioral barriers, and service expectations related to pooled rideshare. These data revealed key human factors barriers of user acceptance of PR and suggested potential actionable experience optimizations that could lead to increased PR usage. This foundational knowledge guided the development of novel human-factors models and behavioral choice models that quantify how psychological, demographic, and trip-level factors influence willingness to pool. Building on these empirical insights, the project developed advanced behavioral modeling tools, including mixed logit and integrated choice and latent variable models, to capture both observable and latent influences on PR adoption. These models significantly improved the ability to predict riders’ acceptance of pooled trips, explaining choice heterogeneity through latent constructs such as safety, service experience, privacy concerns, time sensitivity, and environmental attitudes. Together, these models provide a robust analytical foundation for designing PR systems that more effectively meet user needs. The project translated human-factors insights and behavioral models into actionable technology innovations by extending POLARIS—an agent-based, activity-based travel simulation platform—into a fully functional pooled rideshare simulation environment. New PR modules, acceptance models, and regional scenarios were implemented for Greenville, SC and Austin, TX, enabling high-fidelity validation of algorithmic strategies under realistic demand and traffic conditions. The simulation platform supported the development and evaluation of adaptive discount-based assignment algorithms, enhanced willingness-to-pay formulations, demographic-aware incentive mechanisms, and a proactive joint assignment and repositioning strategy. Simulation results demonstrated substantial gains in pooling uptake, average vehicle occupancy, energy efficiency, and fleet profitability. In Greenville, pooling adoption more than doubled, while reductions in vehicle-miles traveled and energy consumption were significant. In Austin, pooling improvements were achieved with minimal service-quality trade-offs, and profitability increased across all fleet sizes. Through this research, we developed a comprehensive understanding of the human factors barriers that limit user acceptance of pooled rideshare services. These insights enabled the design of human-factors-aware pooled rideshare technologies that more effectively address user concerns and improve adoption rates. By integrating these models into an advanced agent-based simulation framework, we demonstrated that higher adoption of pooled rideshare can lead to measurable improvements in energy efficiency and system performance. Together, these contributions establish a validated pathway from human-centered analysis to technology development and energy-saving outcomes, supporting national goals for more sustainable and efficient mobility systems.

Jia, Yunyi↗
NASA NTRS2016

Exact and Heuristic Algorithms for Runway Scheduling

This paper explores the Single Runway Scheduling (SRS) problem with arrivals, departures, and crossing aircraft on the airport surface. Constraints for wake vortex separations, departure area navigation separations and departure time window restrictions are explicitly considered. The main objective of this research is to develop exact and heuristic based algorithms that can be used in real-time decision support tools for Air Traffic Control Tower (ATCT) controllers. The paper provides a multi-objective dynamic programming (DP) based algorithm that finds the exact solution to the SRS problem, but may prove unusable for application in real-time environment due to large computation times for moderate sized problems. We next propose a second algorithm that uses heuristics to restrict the search space for the DP based algorithm. A third algorithm based on a combination of insertion and local search (ILS) heuristics is then presented. Simulation conducted for the east side of Dallas/Fort Worth International Airport allows comparison of the three proposed algorithms and indicates that the ILS algorithm performs favorably in its ability to find efficient solutions and its computation times.

safe & surface operations↗
NASA NTRS2014

Calibration and Validation of a Finite ELement Model of THor-K Anthropomorphic Test Device for Aerospace Safety Applications

The THOR anthropomorphic test device (ATD) has been developed and continuously improved by the National Highway Traffic Safety Administration to provide automotive manufacturers an advanced tool that can be used to assess the injury risk of vehicle occupants in crash tests. Recently, a series of modifications were completed to improve the biofidelity of THOR ATD [1]. The updated THOR Modification Kit (THOR-K) ATD was employed at Wright-Patterson Air Base in 22 impact tests in three configurations: vertical, lateral, and spinal [2]. Although a computational finite element (FE) model of the THOR had been previously developed [3], updates to the model were needed to incorporate the recent changes in the modification kit. The main goal of this study was to develop and validate a FE model of the THOR-K ATD. The CAD drawings of the THOR-K ATD were reviewed and FE models were developed for the updated parts. For example, the head-skin geometry was found to change significantly, so its model was re-meshed (Fig. 1a). A protocol was developed to calibrate each component identified as key to the kinematic and kinetic response of the THOR-K head/neck ATD FE model (Fig. 1b). The available ATD tests were divided in two groups: a) calibration tests where the unknown material parameters of deformable parts (e.g., head skin, pelvis foam) were optimized to match the data and b) validation tests where the model response was only compared with test data by calculating their score using CORrelation and Analysis (CORA) rating system. Finally, the whole ATD model was validated under horizontal-, vertical-, and lateral-loading conditions against data recorded in the Wright Patterson tests [2]. Overall, the final THOR-K ATD model developed in this study is shown to respond similarly to the ATD in all validation tests. This good performance indicates that the optimization performed during calibration by using the CORA score as objective function is not test specific. Therefore confidence is provided in the ATD model for uses in predicting response in test conditions not performed in this study such those observed in the spacecraft landing. Comparison studies with ATD and human models may also be performed to contribute to future changes in THOR ATD design in an effort to improve its biofidelity, which has been traditionally based on post-mortem human subject testing and designer experience.

Putnam, J. B.↗
DOE OSTI2026

Audi e-tron Green Light Optimized Speed Advisory On-Road Data

To aid researchers in studying the capabilities and benefits of vehicle-to-infrastructure communication, Argonne National Laboratory collected a robust set of on-road driving data of the Audi Green Light Optimized Speed Advisory (GLOSA) system implemented in the e-tron battery electric vehicle. This dataset includes 33 tests, each roughly 27 miles in length and roughly 45 to 75 minutes in duration. The team selected Kane County Highway Route 34 from Main Street in Batavia, Illinois to Middlecreek Lane in St. Charles, Illinois as the route do to its high density of GLOSA-active lights and the most opportunities to observe the system per hour of test time. The data include parameters from the following sources: GLOSA system driving the dash indicators, multiple powertrain parameters including real-time battery power/energy consumption, GPS, front radar gap, and rear radar gap. ![audio-e-tron image](audi-e-tron.jpg)

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2026

Open-Source Data for MAC-POSTS: Mobility Data Analytics Center - Prediction, Optimization, and Simulation Toolkit for Transportation Systems

MAC-POSTS (Mobility Data Analytics Center - Prediction, Optimization, and Simulation toolkit for Transportation Systems) is a toolkit for dynamic transportation network modeling. Developed by the Mobility Data Analytics Center (MAC) at Carnegie Mellon University, this package implements many classic dynamic transportation network models, as well as new models proposed by MAC members. It has served as one building block for many other models and research projects. As such, this package used to be treated as an internal research project of the MAC lab, and admittedly, the code base is messy, and the interface is hard to use. However, we are working hard to make it a generally usable and useful toolkit for dynamic transportation network modeling. We would really appreciate any feedback, comments, suggestions, or criticisms.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
NASA NTRS2010

Trajectory Optimization: OTIS 4

The latest release of the Optimal Trajectories by Implicit Simulation (OTIS4) allows users to simulate and optimize aerospace vehicle trajectories. With OTIS4, one can seamlessly generate optimal trajectories and parametric vehicle designs simultaneously. New features also allow OTIS4 to solve non-aerospace continuous time optimal control problems. The inputs and outputs of OTIS4 have been updated extensively from previous versions. Inputs now make use of objectoriented constructs, including one called a metastring. Metastrings use a greatly improved calculator and common nomenclature to reduce the user s workload. They allow for more flexibility in specifying vehicle physical models, boundary conditions, and path constraints. The OTIS4 calculator supports common mathematical functions, Boolean operations, and conditional statements. This allows users to define their own variables for use as outputs, constraints, or objective functions. The user-defined outputs can directly interface with other programs, such as spreadsheets, plotting packages, and visualization programs. Internally, OTIS4 has more explicit and implicit integration procedures, including high-order collocation methods, the pseudo-spectral method, and several variations of multiple shooting. Users may switch easily between the various methods. Several unique numerical techniques such as automated variable scaling and implicit integration grid refinement, support the integration methods. OTIS4 is also significantly more user friendly than previous versions. The installation process is nearly identical on various platforms, including Microsoft Windows, Apple OS X, and Linux operating systems. Cross-platform scripts also help make the execution of OTIS and post-processing of data easier. OTIS4 is supplied free by NASA and is subject to ITAR (International Traffic in Arms Regulations) restrictions. Users must have a Fortran compiler, and a Python interpreter is highly recommended.

Riehl, John P.↗
NASA NTRS2014

An Evaluation of a Flight Deck Interval Management Algorithm Including Delayed Target Trajectories

NASA's first Air Traffic Management (ATM) Technology Demonstration (ATD-1) was created to facilitate the transition of mature air traffic management technologies from the laboratory to operational use. The technologies selected for demonstration are the Traffic Management Advisor with Terminal Metering (TMA-TM), which provides precise timebased scheduling in the terminal airspace; Controller Managed Spacing (CMS), which provides controllers with decision support tools enabling precise schedule conformance; and Interval Management (IM), which consists of flight deck automation that enables aircraft to achieve or maintain precise in-trail spacing. During high demand operations, TMA-TM may produce a schedule and corresponding aircraft trajectories that include delay to ensure that a particular aircraft will be properly spaced from other aircraft at each schedule waypoint. These delayed trajectories are not communicated to the automation onboard the aircraft, forcing the IM aircraft to use the published speeds to estimate the target aircraft's estimated time of arrival. As a result, the aircraft performing IM operations may follow an aircraft whose TMA-TM generated trajectories have substantial speed deviations from the speeds expected by the spacing algorithm. Previous spacing algorithms were not designed to handle this magnitude of uncertainty. A simulation was conducted to examine a modified spacing algorithm with the ability to follow aircraft flying delayed trajectories. The simulation investigated the use of the new spacing algorithm with various delayed speed profiles and wind conditions, as well as several other variables designed to simulate real-life variability. The results and conclusions of this study indicate that the new spacing algorithm generally exhibits good performance; however, some types of target aircraft speed profiles can cause the spacing algorithm to command less than optimal speed control behavior.

Swieringa, Kurt A.↗
DOE OSTI2024

DS-GL: Advancing Graph Learning via Harnessing the Power of Nature within Dynamic Systems

With the rapid digitization of the world, an increasing number of real-world applications are turning to nonEuclidean data, modeled as graphs. Due to their intrinsic high complexity and irregularity, learning from graph data demands tremendous computational power. Recently, CMOS-compatible Ising machines, i.e., dynamic systems composed of CMOS components, have emerged as a new approach that harnesses the inherent power of natural annealing within dynamic systems to efficiently resolve binary optimization problems and have been adopted for traditional graph computation, such as max-cut. However, when performing complex Graph Learning (GL) tasks, Ising machines face significant hurdles: (i) they are inherently binary and thus ill-suited for real-valued problems; (ii) their expensive all-to-all coupling network that guarantees effective natural annealing poses daunting scalability concerns. To address these challenges, this paper proposes a nature-powered graph learning framework dubbed DS-GL, which is the first effort to transform the process of solving graph learning problems into the natural annealing process within a parameterized dynamic system embodied as a CMOS chip. To tackle the two major hurdles, DS-GL first augments the Ising machine architecture to modify the self-reaction term of its Hamiltonian function from linear to quadratic, effectively serving as an energy regulator. This adjustment maintains the system’s original physical interpretation while enabling it to process continuous, real-valued data. Second, to address the scaling issue, DS-GL further upgrades the real-valued dense Ising machine by decomposing it into a mesh-based multi-PE dynamic system that supports efficient distributed spatial-temporal co-annealing across different PEs through sparse interconnects. By exploiting the inherent sparsity and component structures in real-world graphs, DS-GL is able to map complex graph learning tasks onto the scalable dynamic system while maintaining high accuracy. Evaluations with three diverse GL applications across six real-world datasets, including traffic flow and COVID-19 prediction, show that DS-GL can deliver from 102× to 106× speedups and 500× energy reduction over Graph Neural Networks on GPUs, with 5% - 20% accuracy enhancement.

Song, Ruibing↗
NASA NTRS2000

Wind Prediction Accuracy for Air Traffic Management Decision Support Tools

The performance of Air Traffic Management and flight deck decision support tools depends in large part on the accuracy of the supporting 4D trajectory predictions. This is particularly relevant to conflict prediction and active advisories for the resolution of conflicts and the conformance with of traffic-flow management flow-rate constraints (e.g., arrival metering / required time of arrival). Flight test results have indicated that wind prediction errors may represent the largest source of trajectory prediction error. The tests also discovered relatively large errors (e.g., greater than 20 knots), existing in pockets of space and time critical to ATM DST performance (one or more sectors, greater than 20 minutes), are inadequately represented by the classic RMS aggregate prediction-accuracy studies of the past. To facilitate the identification and reduction of DST-critical wind-prediction errors, NASA has lead a collaborative research and development activity with MIT Lincoln Laboratories and the Forecast Systems Lab of the National Oceanographic and Atmospheric Administration (NOAA). This activity, begun in 1996, has focussed on the development of key metrics for ATM DST performance, assessment of wind-prediction skill for state of the art systems, and development/validation of system enhancements to improve skill. A 13 month study was conducted for the Denver Center airspace in 1997. Two complementary wind-prediction systems were analyzed and compared to the forecast performance of the then standard 60 km Rapid Update Cycle - version 1 (RUC-1). One system, developed by NOAA, was the prototype 40-km RUC-2 that became operational at NCEP in 1999. RUC-2 introduced a faster cycle (1 hr vs. 3 hr) and improved mesoscale physics. The second system, Augmented Winds (AW), is a prototype en route wind application developed by MITLL based on the Integrated Terminal Wind System (ITWS). AW is run at a local facility (Center) level, and updates RUC predictions based on an optimal interpolation of the latest ACARS reports since the RUC run. This paper presents an overview of the study's results including the identification and use of new large mor wind-prediction accuracy metrics that are key to ATM DST performance.

Cole, Rod↗
NASA NTRSDate not supplied

Aerial Vehicle Routing and Scheduling for UAS Traffic Management: A Monte Carlo Tree Search Approach

Numerous unmanned aircraft systems operating at low altitudes to deliver goods and services may one day become ubiquitous in our cities. In the Unmanned Aircraft Systems (UAS) Traffic Management (UTM) framework, such a concept is envisioned, where aerial vehicles operate beyond visual line of sight (BVLOS) within specifically reserved and time stamped “corridors” in the airspace. For example, these corridors or operational intent volumes can connect an aerial vehicle’s origin site to its destination site for package delivery operations. There may also be more than one corridor available for an aerial vehicle to choose from and often different corridors may intersect with one another. Thus, it is imperative to ensure flight trajectories belonging to different aerial vehicles are not in conflict. Per the UTM CONOPs, we assume that a vehicle almost always stays inside its corridor or operational volume. This work provides a framework for strategic deconfliction of UTM or package delivery drones, where we schedule the departure time of all vehicles subject to various temporal constraints (including the corridor deconfliction at the intersections). We present the “multi-route weighted package delivery problem” which serves as an exemplifying model for strategic deconfliction in UTM. In the multi-route weighted package delivery problem, a graph network is given which consists of a set of depots (source) and drop-off (destination) nodes, with multiple routes (defined as a sequence of waypoints) connecting the depots to drop-off nodes. In addition, routes are weighted by the associated ground risk and total travel distance for package delivery. The goal is for a known set of aerial vehicles to depart from the depots, choose a route and take off time, while avoiding conflicts with other aerial vehicles, and minimizing both risk and distance traveled. We provide a mixed integer linear programming (MILP) formulation of the problem, as well as a heuristic solution based on Monte Carlo Tree Search (MCTS) – a method used in game theory and artificial intelligence – to overcome limitations inherent to optimal solvers. Computational results show the advantages of using MCTS over the MILP formulation; the former can provide a sub-optimal solution quickly, and may sometimes even reach an optimal solution, whereas the latter may not even produce a solution in reasonable time. Furthermore, results from both the MILP formulation and MCTS methods were validated using a preliminary agent-based simulator implementing the UTM concept of operations. Thus, the MCTS method can be seen as a scalable solution to the complex multi-route weighted package delivery problem and may possibly be extended to similar complex optimization problems.

Kenny Chour↗
NASA NTRS2013

Optimal Integration of Departures and Arrivals in Terminal Airspace

Coordination of operations with spatially and temporally shared resources, such as route segments, fixes, and runways, improves the efficiency of terminal airspace management. Problems in this category are, in general, computationally difficult compared to conventional scheduling problems. This paper presents a fast time algorithm formulation using a non-dominated sorting genetic algorithm (NSGA). It was first applied to a test problem introduced in existing literature. An experiment with a test problem showed that new methods can solve the 20 aircraft problem in fast time with a 65% or 440 second delay reduction using shared departure fixes. In order to test its application in a more realistic and complicated problem, the NSGA algorithm was applied to a problem in LAX terminal airspace, where interactions between 28% of LAX arrivals and 10% of LAX departures are resolved by spatial separation in current operations, which may introduce unnecessary delays. In this work, three types of separations - spatial, temporal, and hybrid separations - were formulated using the new algorithm. The hybrid separation combines both temporal and spatial separations. Results showed that although temporal separation achieved less delay than spatial separation with a small uncertainty buffer, spatial separation outperformed temporal separation when the uncertainty buffer was increased. Hybrid separation introduced much less delay than both spatial and temporal approaches. For a total of 15 interacting departures and arrivals, when compared to spatial separation, the delay reduction of hybrid separation varied between 11% or 3.1 minutes and 64% or 10.7 minutes corresponding to an uncertainty buffer from 0 to 60 seconds. Furthermore, as a comparison with the NSGA algorithm, a First-Come-First-Serve based heuristic method was implemented for the hybrid separation. Experiments showed that the results from the NSGA algorithm have 9% to 42% less delay than the heuristic method with varied uncertainty buffer sizes.

terminal airspace↗
NASA NTRS2018

Mars Planetary Network for Human Exploration Era – Potential Challenges and Solutions

During 2016-2017, a study was conducted under the sponsorship of the NASA’s Space Communications and Navigation (SCaN) Program to investigate the deep space communications capacity taking into account the needs of all the present and envisioned future missions toward 2030s. It was soon recognized that planning for human exploration to Mars would impose certain unprecedent challenges, both fiscal and technical, to the current space communications paradigm. Targeting the assumed missions concepts, i.e., Crewed Mission to Phobos (CMTP) and Mars Short Stay Mission (MSSM), a Mars Planetary Network for the human exploration era has been formulated. The activity modeling and network traffic simulation/modeling we performed gave some insight into the technical challenges in space communications for the envisioned human Mars exploration era. Chief among the potential challenges are: (1) the high demand on the deep space network (DSN) assets for achieving the high-rate links, both return and forward, from Mars farthest/farther distance (up to 2.67 AU); (2) the need for resilient, persistent communication coverage for crewed vehicles, on surface and in orbits; (3) the significant period of outage for the Mars-Earth link due to superior solar conjunction; (4) the need for on-demand, simultaneous access to the proximity link by multiple vehicles and astronauts in the exploration zone; (5) the capability of determining precise, real-time, positions of surface vehicles and astronauts by the deep space habitat and/or other tele-operations entities. Solution space to each of the above challenges has been explored and analyzed in the context of the individual problem domain and, more importantly, in conjunction with that for the other challenges. This has led to an end-to-end definition of a Mars Planetary Network that would feature: (1) the fusion of deep space Ka-band and optical communications for achieving Mars-Earth high-rate links taking advantage of the optical/RF hybrid 8m/34m antennas in DSN; (2) the integrated application of the Multiple Spacecraft Per Antenna (MSPA) technique, for return link data acquisition, and the Multiple Uplink Per Antenna (MUPA) technique, for forward link data trasnmission, to reduce the number of 34m beam-wave guide (BWG) antennas needed for the era; (3) the integration of three, arrayed, 34m beam-wave guide (BWG) antennas, to provide a high G/T aperture, with the MSPA/MUPA techniques, and a dual “trunk link”approach to cut down the needed G/T -- hence, reducing the number of 34m antennas, relative to that in the single trunk approach by 50%; (4) the deployment of two areostationary/areosynchronous Mars relay orbiters; one of them could also function as (or be served by) a notional Deep Space Habitat (DSH); (5) the opportunistic deployment of a science orbiter in a Pioneer-6 type orbit, equidistant Mars/Earth, that could also serve as an intermediary relay during the Mars superior solar conjunction perod; (6) the existence of the multi-function Mars proximity link that provides the demand-assigned, multiple access (DAMA) capability; and (7) the provision of tracking observables, by leveraging on the planned Mars orbiting and surface infrastructure, to enable the in-situ navigation for surface and orbiting vehicles. This paper provides the description of the proposed Mars Planetary Network in the human exploration era, the trade-off analysis for the various alternative architectures, and the optimal solutions to the key challenges in defining this end-to-end network.

Cheung, Kar-Ming↗
NASA NTRSDate not supplied

Towards an Aviation Large Language Model by Fine-tuning and Evaluating Transformers

In the aviation domain, there are many applications for machine learning and artificial intelligence tools that utilize natural language. For example, there is a desire to know the commonalities in written safety reports such as voluntary post incidents reports or aerial wildfire operations reports to better understand the risks present. Another use-case is the possibility of extracting airspace procedures and constraints currently written in documents such as Letters of Agreement. These applications can benefit from the use of state-of-the-art natural language processing techniques when adapted to the language/phraseology specific to the aviation domain. This paper evaluates the viability of adaptation of NLP tools to the aviation domain by fine-tuning transformer based models using aviation data sets. In 2018, a novel language model based on neural units (also called transformers) was created and became known as “Bidirectional Encoder Representations from Transformers” or BERT. This architecture combined with large amounts of English training data and innovative semi-supervised training tasks set the standard for what would later emerge as Large Language Models. The performance of these models was further improved by hyperparameter tuning and refinement of the semi-supervised training task and resulted in “Robustly Optimized BERT Pre-training Approach through hyperparameter tuning” or RoBERTa models. These pre-trained Large Language Models proved to be useful for a wide variety of natural language processing tasks such as text classification and question answering through a process called fine-tuning. The transformer architecture with pre-trained weights served as the basis with the last few layers replaced with layers fine-tuned to perform a new task e.g., a layer that provides a label for the entire input text. This process of fine-tuning can also be used to adapt the models to new domains; e.g., BioBERT started with the pre-trained BERT model and was completed by additional fine-tuning and training on biomedical documents. Transformer-based architectures can also be used to create rich representations of text called embeddings which can serve as the input to other machine learning models. This allows simpler algorithms such as logistic regression to use context-rich representations of the text while still remaining quick to train and evaluate. In the world of aviation, there is a growing demand for natural language processing and understanding but the domain presents unique challenges. Due to the technical content (and specialized language) of most aviation documents, fine-tuning pre-trained Large Language Models to specific tasks has not met the benchmark on natural language processing tasks set by simpler models trained from scratch on the data. To address this deficiency, this paper evaluates the improvements from fine-tuning a Large Language Model on a large set of aviation documents using the original semi-supervised training tasks before performing specific natural language tasks. In fine-tuning, a domain-specific dataset is used on the original training task but with the pre-trained Large Language Model instead of starting from a random initialization. This approach allows the model to be adapted to the specific domain language without discarding the information gained from training on general English data. This paper utilized two major dataset types to train and assess the RoBERTa fine-tuning performance. The first are 7,057 Letters of Agreement which are Federal Aviation Administration (FAA) documents that formalize airspace operations across the national airspace system. They contain many examples of ‘aviation English’ using domain specific terminology and phrasing which serves as a representative basis to perform the semi-supervised fine-tuning. The second type is the 494 document classification labels to be used for evaluation. This down-stream evaluation aims to show the performance of the fine-tuned model, better understand how much data is needed for an effective fine-tuning, and how fine-tuning can be adapted for different applications in-the domain. After semi-supervised training, evaluation begins by encoding the documents for classification using the fine-tuned RoBERTa model. Then a logistic regression classifier is trained to label the document type and compared against our ground truth labels. This currently leads to a 82.8% accuracy on 10-fold cross validation showing improvement over baseline RoBERTa which achieved 81.0%. We plan to measure the improvements on additional tasks and it is expected that these improvements will lead to more robust models that can tackle the natural language processing challenges present in aviation datasets.

ATM↗