Person:
Mavris,
Dimitri N.
Mavris,
Dimitri N.
Permanent Link
Associated Organization(s)
Organizational Unit
Organizational Unit
ORCID
ArchiveSpace Name Record
Publication Search Results
Now showing
1 - 7 of 7
-
ItemMachine Learning Enabled Turbulence Prediction Using Flight Data for Safety Analysis(International Council of the Aeronautical Sciences (ICAS), 2021-09) Emara, Mariam ; dos Santos, Marcos ; Chartier, Noah ; Ackley, Jamey ; Puranik, Tejas G. ; Payan, Alexia P. ; Kirby, Michelle R. ; Pinon, Olivia J. ; Mavris, Dimitri N.The hazards posed by turbulence remain an important issue in commercial aviation safety analysis. Turbulence is among the leading cause of in-flight injury to passengers and flight attendants. Current methods of turbulence detection may suffer from sparse or inaccurate forecast data sets, low spatial and temporal resolution , and lack of in-situ reports. The increased availability of flight data records offers an opportunity to improve the state-of-the-art in turbulence detection. The Eddy Dissipation Rate (EDR) is consistently recognized as a reliable measure of turbulence and is widely used in the aviation industry. In this paper, both classification and regression supervised machine learning models are used in conjunction with flight operations quality assurance (FOQA) data collected from 6,000 routine flights to estimate the EDR (and thereby turbulence severity) in future time horizons. Data from routine airline operations that encountered different levels of turbulence is collected and analyzed for this purpose. Results indicate that the models are able to perform reasonably well in predicting the EDR and turbulence severity around 10 seconds prior to encountering a turbulence event. Continuous deployment of the model enables obtaining a near-continuous prediction of possible future turbulence events and builds the capability towards an early warning system for pilots and flight attendants.
-
ItemAircraft Performance Model Calibration and Validation for General Aviation Safety Analysis(Georgia Institute of Technology, 2020-03) Puranik, Tejas G. ; Harrison, Evan D. ; Chakraborty, Imon ; Mavris, Dimitri N.Performance models facilitate a wide range of safety analyses in aviation. In an ideal scenario, the performance models would show inherently good agreement with the true performance of the aircraft. However, in reality, this is rarely the case: either owing to underlying simplifications or due to the limited fidelity of applicable tools or data. In such cases, calibration is required to fine-tune the behavior of the performance models. For point-mass steady-state performance models, challenges arise due to the fact that there is no obvious, unique metric or flight condition at which to assess the accuracy of the model predictions, as well as because a large number of model parameters may potentially influence model accuracy. This work presents a two-level approach to aircraft performance model calibration. The first level consists of using manufacturer-developed performance manuals for calibration, whereas the second level provides additional refinement when flight data are available. The performance models considered in this work consist of aerodynamic and propulsion models (performance curves) that are capable of predicting the non-dimensional lift, drag, thrust, and torque at any given point in time. The framework is demonstrated on two representative general aviation aircraft. The demonstrated approach results in models that can predict critical energy-based safety metrics with improved accuracy for use in retrospective safety analyses.
-
ItemDeep Gaussian Process Enabled Surrogate Models for Aerodynamic Flows(Georgia Institute of Technology, 2020-01) Rajaram, Dushhyanth ; Puranik, Tejas G. ; Renganathan, Sudharshan Ashwin ; Sung, WoongJe ; Pinon-Fischer, Olivia ; Mavris, Dimitri N.Deep Gaussian process (DGP) models are multi-layered hierarchical generalizations of the well-known Gaussian process (GP) models widely used to construct surrogate models of aerodynamic quantities of interest. Combining the desirable features of GP models and deep neural networks (DNN), DGP models are known to perform well when training data is scarce and the behavior of the system response is highly non-stationary. In this paper, the performance of DGP models is evaluated against GP models. Detailed comparisons are made and conclusions are drawn in terms of training time, data requirements, predictive error, and robustness to choice of training design of experiments, among other metrics. Additionally, sensitivity and scalability analyses are conducted for the GP models to evaluate their usability. Finally, an adaptive construction of both models is presented, where the models are built sequentially by selecting points that maximize posterior variance. Several experiments are conducted with canonical test functions at varying input dimensions and a viscous transonic airfoil test case at 42 input dimensions. The experiments suggest that DGP models outperform traditional GP models in terms of accuracy but incur higher computational costs for training.
-
ItemApplication of Machine Learning to the Analysis and Prediction of the Coincidence of Ground Delay Programs and Ground Stop(Georgia Institute of Technology, 2020-01) Mangortey, Eugene ; Bleu-Laine, Marc-Henri ; Puranik, Tejas G. ; Pinon, Olivia J. ; Mavris, Dimitri N.Traffic Management Initiatives such as Ground Delay Programs and Ground Stops are implemented by traffic management personnel to control air traffic volume to constrained airports when traffic demand is projected to exceed the airports’ acceptance rate due to conditions such as inclement weather, volume constraints, etc. Ground Delay Programs are issued for lengthy periods of time and aircraft are assigned departure times later than scheduled. Ground Stops on the other hand, are issued for short periods of time and aircraft are not permitted to land at the constrained airport. Occasionally, Ground Stops are issued during an ongoing Ground Delay Program, and vice versa, which hinders the efficient planning and implementation of these Traffic Management Initiatives. This research proposes a methodology to help stakeholders better capture the impact of the coincidence of weather related Ground Delay Programs and Ground Stops, and potentially help reduce the number and duration of such coincidences. This is achieved by leveraging Machine Learning techniques to predict their coincidence at a given hour, predict which Traffic Management Initiative would precede the other during their coincidence, and identify key predictors that cause their coincidence. The Random Forests Machine Learning algorithm was identified as the best suited algorithm for predicting the coincidence of weather-related Ground Delay Programs and Ground Stops, as well as the Traffic Management Initiative that would precede the other during their coincidence.
-
ItemClassification, Analysis, and Prediction of the Daily Operations of Airports Using Machine Learning(Georgia Institute of Technology, 2020-01) Mangortey, Eugene ; Puranik, Tejas G. ; Pinon, Olivia J. ; Mavris, Dimitri N.The Federal Aviation Administration (FAA) is the regulatory body in the United States responsible for the advancement, safety, and regulation of civil aviation. The FAA also oversees the development of the air traffic control system in the U.S. Over the years, the FAA has made tremendous progress in modernizing the National Airspace System (NAS) by way of technological advancements and the introduction of procedures and policies that have maintained the safety of the United States airspace. However, as with any other system, there is a need to continuously address evolving challenges pertaining to the sustainment and resiliency of the NAS. One of these challenges involves efficiently analyzing and assessing the operations of airports. In particular, there is a need to assess the impact and effectiveness of the implementation of Traffic Management Initiatives (TMI) and other procedures on daily airport operations, as this will lead to the identification of trends and patterns to inform better decision making. The FAA currently manually classifies the daily operations of airports into three categories: “Good Days”, “Average Days”, and “Bad Days” as a means to assess their efficiency. However, this exercise is time-consuming and can be improved. In particular, Big Data Analytics can be leveraged to develop a systematic approach for classifying or clustering the daily operations of airports. This research presents a methodology for clustering the daily operations of Newark International Airport (EWR) using metrics such as the number of diversions, Ground Stops, departure delays, etc. Each of these categories/clusters is then analyzed to identify key characteristics, trends and patterns, which can then be used by airport operators, and FAA analysts and researchers to improve the operations at the airport. Finally, the Boosting Ensemble Machine Learning algorithm is used to predict the category of operations at the airport, hence enabling airport operators, FAA analysts and researchers to take appropriate actions.
-
ItemPrediction and Analysis of Ground Stops with Machine Learning(Georgia Institute of Technology, 2020-01) Mangortey, Eugene ; Puranik, Tejas G. ; Pinon, Olivia J. ; Mavris, Dimitri N.A flight is considered to be delayed when it arrives 15 or more minutes later than scheduled. Delays attributed to the National Airspace System are one of the most common type of delays. Such delays may be caused by Traffic Management Initiatives (TMI) such as Ground Stops (GS), issued at affected airports. Ground Stops are implemented to control air traffic volume to specific airports where the projected traffic demand is expected to exceed the airports’ acceptance rate over a short period of time due to conditions such as inclement weather, volume constraints, closed runways, etc. Ground Stops can be considered to be the strictest Traffic Management Initiative (TMI), particularly because all flights destined to affected airports are grounded until conditions improve. Efforts have been made over the years to reduce the impact of Traffic Management Initiatives on airports and flight operations. However, these efforts have largely focused on otherTraffic Management Initiatives such as Ground Delay Programs (GDP), due to their frequency and duration compared to Ground Stops. Limited work has also been carried out on Ground Stops because of the limited amount of time that traffic management personnel often have between planning and implementing Ground Stops and external factors that influence decisions of traffic management personnel. Consequently, this research primarily focuses on the prediction of weather-related Ground Stops at Newark Liberty International (EWR) and LaGuardia (LGA) airports, with the secondary goal of gaining insights into factors that influence their occurrence. It is expected that this research will provide stakeholders with further insights into factors that influence the occurrence of weather-related Ground Stops at both airports. This is achieved by benchmarking Machine Learning algorithms in order to identify the best suited algorithm(s) for the prediction models, and identifying and analyzing key factors that influence the occurrence of weather-related Ground Stops at both airports. This is achieved by 1) fusing data from the Traffic Flow Management System (TFMS) and Automated Surface Observing Systems (ASOS) datasets, and 2) leveraging supervised Machine Learning algorithms to predict the occurrence of weather-related Ground Stops. The performance of these algorithms is evaluated using balanced accuracy, and identifies the Boosting Ensemble algorithm as the best suited algorithm for predicting the occurrence of Ground Stops at EWR and LGA. Further analysis also revealed that model performance is significantly better when using balanced datasets compared to imbalanced datasets.
-
ItemApplication of Machine Learning Techniques to Parameter Selection for Flight Risk Identification(Georgia Institute of Technology, 2020-01) Mangortey, Eugene ; Monteiro, Dylan J. ; Ackley, Jamey ; Gao, Zhenyu ; Puranik, Tejas G. ; Kirby, Michelle ; Pinon, Olivia J. ; Mavris, Dimitri N.In recent years, the use of data mining and machine learning techniques for safety analysis, incident and accident investigation, and fault detection has gained traction among the aviation community. Flight data collected from recording devices contains a large number of heterogeneous parameters, sometimes reaching up to thousands on modern commercial aircraft. More data is being collected continuously which adds to the ever-increasing pool of data available for safety analysis. However, among the data collected, not all parameters are important from a risk and safety analysis perspective. Similarly, in order to be useful for modern analysis techniques such as machine learning, using thousands of parameters collected at a high frequency might not be computationally tractable. As such, an intelligent and repeatable methodology to select a reduced set of significant parameters is required to allow safety analysts to focus on the right parameters for risk identification. In this paper, a step-by-step methodology is proposed to down-select a reduced set of parameters that can be used for safety analysis. First, correlation analysis is conducted to remove highly correlated, duplicate, or redundant parameters from the data set. Second, a pre-processing step removes metadata and empty parameters. This step also considers requirements imposed by regulatory bodies such as the Federal Aviation Administration and subject matter experts to further trim the list of parameters. Third, a clustering algorithm is used to group similar flights and identify abnormal operations and anomalies. A retrospective analysis is conducted on the clusters to identify their characteristics and impact on flight safety. Finally, analysis of variance techniques are used to identify which parameters were significant in the formation of the clusters. Visualization dashboards were created to analyze the cluster characteristics and parameter significance. This methodology is employed on data from the approach phase of a representative single-aisle aircraft to demonstrate its application and robustness across heterogeneous data sets. It is envisioned that this methodology can be further extended to other phases of flight and aircraft.