TU Darmstadt / ULB / TUbiblio

Using Machine Learning for Data-Based Assessing of the Aircraft Fuel Economy

Baumann, Sebastian (2019):
Using Machine Learning for Data-Based Assessing of the Aircraft Fuel Economy.
In: 2019 IEEE Aerospace Conference, Big Sky, Montana, USA, March 2-9, 2019, [Conference or Workshop Item]

Abstract

Weight, lift, drag and thrust are the key forces during flight influencing the consumption characteristics of an aircraft. Monitoring the aircraft’s fuel consumption allows statements to be made about the aircraft performance and flight efficiency. Here, valuations with regard to a propulsive or an aerodynamic degrading as well as procedural benefits are of interest for aircraft operators and airlines. Conventional data analyses and calculation methods (ex post), e. g. via physical models or statistical metrics like median or mean, have the disadvantage of not being able to quantify influencing parameters on the fuel economy accurately enough in many cases. A more precise method, which allows statements and assessments (diagnoses based on historical flight data) and estimations (prognoses based on estimated model inputs) on the aircraft fuel consumption, would be desirable to detect aircraft fuel and flight (in)efficiencies under real flight operational and environmental conditions. In this respect, a reliable trade off would be desirable in order to avoid uneconomical modeling effort and reduce expenses concerning data procurement. This paper is in the scope of an applicationoriented machine learning method development. For this reason, the present work deals with algorithms of machine learning, which can exceed current methods, especially for complex tasks with multivariate interdependencies. Ensembles of bagged decision trees for regression and classification purposes, as well as hierarchical cluster analyses applied on full flight data records from flight operations. Thus, the paper addresses two objectives and provides first approaches for: modeling of route, environment and aircraft (tail sign) specific characteristics for feasible fuel flow predictions on the one hand (diagnosis and prognosis in the field of data science); data-based analyses and quantification of influencing factors on the fuel economy to identify routes and anomalous flights on the other hand (data mining, pattern recognition). The latter, thus, aims at an improvement of reporting and the derivation of recommendations for actions for future flight profiles. Finally, the analysis capability of the concept is shown with first results for two analysis questions with three examples, evaluated based on an analysis with sample flight data provided by NASA Dashlink. To proof and evaluate the applicability of this method, a benchmark is also presented by using an available (physical) performance model and statistical metrics.

Item Type: Conference or Workshop Item
Erschienen: 2019
Creators: Baumann, Sebastian
Title: Using Machine Learning for Data-Based Assessing of the Aircraft Fuel Economy
Language: German
Abstract:

Weight, lift, drag and thrust are the key forces during flight influencing the consumption characteristics of an aircraft. Monitoring the aircraft’s fuel consumption allows statements to be made about the aircraft performance and flight efficiency. Here, valuations with regard to a propulsive or an aerodynamic degrading as well as procedural benefits are of interest for aircraft operators and airlines. Conventional data analyses and calculation methods (ex post), e. g. via physical models or statistical metrics like median or mean, have the disadvantage of not being able to quantify influencing parameters on the fuel economy accurately enough in many cases. A more precise method, which allows statements and assessments (diagnoses based on historical flight data) and estimations (prognoses based on estimated model inputs) on the aircraft fuel consumption, would be desirable to detect aircraft fuel and flight (in)efficiencies under real flight operational and environmental conditions. In this respect, a reliable trade off would be desirable in order to avoid uneconomical modeling effort and reduce expenses concerning data procurement. This paper is in the scope of an applicationoriented machine learning method development. For this reason, the present work deals with algorithms of machine learning, which can exceed current methods, especially for complex tasks with multivariate interdependencies. Ensembles of bagged decision trees for regression and classification purposes, as well as hierarchical cluster analyses applied on full flight data records from flight operations. Thus, the paper addresses two objectives and provides first approaches for: modeling of route, environment and aircraft (tail sign) specific characteristics for feasible fuel flow predictions on the one hand (diagnosis and prognosis in the field of data science); data-based analyses and quantification of influencing factors on the fuel economy to identify routes and anomalous flights on the other hand (data mining, pattern recognition). The latter, thus, aims at an improvement of reporting and the derivation of recommendations for actions for future flight profiles. Finally, the analysis capability of the concept is shown with first results for two analysis questions with three examples, evaluated based on an analysis with sample flight data provided by NASA Dashlink. To proof and evaluate the applicability of this method, a benchmark is also presented by using an available (physical) performance model and statistical metrics.

Journal or Publication Title: Proceedings of the 2019 IEEE Aerospace Conference
Divisions: 16 Department of Mechanical Engineering
16 Department of Mechanical Engineering > Institute of Flight Systems and Automatic Control (FSR)
Event Title: 2019 IEEE Aerospace Conference
Event Location: Big Sky, Montana, USA
Event Dates: March 2-9, 2019
Date Deposited: 24 Jan 2019 10:05
Export:

Optionen (nur für Redakteure)

View Item View Item