Analysis of aeronautical engines based on machine learning
Professor Advisor
dc.contributor.advisor
López Droguett, Enrique
Author
dc.contributor.author
Soto Espinosa, Alejandro Flavio
Associate professor
dc.contributor.other
Meruane Naranjo, Viviana
Associate professor
dc.contributor.other
Bravo Márquez, Felipe
Admission date
dc.date.accessioned
2021-06-08T15:54:07Z
Available date
dc.date.available
2021-06-08T15:54:07Z
Publication date
dc.date.issued
2021
Identifier
dc.identifier.uri
https://repositorio.uchile.cl/handle/2250/180036
General note
dc.description
Memoria para optar al título de Ingeniero Civil Mecánico
es_ES
Abstract
dc.description.abstract
The aviation industry depends heavily on the availability of planes and their mechanical
components, specially their engines. In addition, the fuel consumption of the engine is the
highest operating expense for airlines.
In the last decade Machine Learning (ML) has had a big impact in Prognostics and Health
Management (PHM), hence making it a suitable option for creating models that deliver interesting outputs related to the maintenance of machinery.
The main objective of this thesis is to create a model able to identify when the features
of the engine are similar to the ones obtained when the engine needs a maintenance. Data
is analyzed during the take off of the plane, being that during this window of time the consumption rate of fuel reaches its maximum. This model is conceived using machine learning
techniques.
The specific objectives accomplished in this thesis are:
The identification of sections of interest from the original data set. In this sections an
estimation of the fuel consumption of the engine is performed. Finally an anomaly detector
that identifies differences between predicted and true fuel consumption is presented. This
anomaly detector labels as anomalies the bigger differences between predicted and true fuel
consumption.
Results show a clear relation between degradation over time and fuel consumption in the
engines. For the prediction of fuel consumption 4 models were used: polynomial regression,
decision trees, gradient boosting and long short term memory. The best results were obtained
by gradient boosting, with an error below 0.7%.
Evaluation metrics show accuracies of 92 % for engines AIY 2 and AHD 1 and 83 % for
engine AHD 2.