Please use this identifier to cite or link to this item:
https://dair.nps.edu/handle/123456789/4989
Title: | Use of Machine-Learning Techniques Based on Python Language Code to Classify Failure Data from the Brazilian Air Force Database |
Authors: | Ygor de Almeida |
Keywords: | Machine Learning Natural Language Processing Failure Data |
Issue Date: | 17-Oct-2023 |
Publisher: | Acquisition Research Program |
Citation: | Published--Unlimited Distribution |
Series/Report no.: | Program Management;NPS-PM-23-215 |
Abstract: | Like other major flight operators, the Brazilian Air Force (BrAF) must effectively manage multiple aircraft fleets, which are costly assets. The failure data collected from these assets is essential for decision-makers to assess the systems’ reliability, availability, and maintainability. Obtaining accurate and reliable information depends on the quality of failure data collected. BrAF engineers typically preprocess the data by classifying it as failure or non-failure for analysis, but this task is repetitive and time-consuming. Therefore, this study aims to develop and evaluate a machine-learning model capable of automatically performing this classification task. Of the six machine-learning techniques assessed, the Support Vector Classifier (SVC) model performed best in the F1-score metric. The results suggest that the SVC model has the potential to classify failure data from the BrAF database accurately, saving a significant amount of time. Additionally, the model could aid maintainers during the failure recording process, preventing them from inserting non-useful data in the database, and for inventory management of specific workshop repairs, thus providing more accurate information about the number of failures. |
Description: | Program Management / Graduate Student Research |
URI: | https://dair.nps.edu/handle/123456789/4989 |
Appears in Collections: | NPS Graduate Student Theses & Reports |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
NPS-PM-23-215.pdf | Student Thesis | 24.66 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.