Degradation Estimation Analysis of an Aeronautical Pneumatic Valve Using Machine Learning

This work aims to propose a methodology to evaluate machine learning models' capability to estimate the degradation level of an aircraft pneumatic system valve (PRSOV - pressure regulating and shutoff valve) in open-loop. A more accurate PRSOV degradation estimation may reduce operational costs...

Full description

Saved in:
Bibliographic Details
Main Authors: de Assis Silva, Lucas, Sano, Humberto Hayashi, Junior, Cairo Lucio Nascimento
Format: Conference Proceeding
Language:English
Subjects:
aeronautical system
Analytical models
Atmospheric modeling
data science
Degradation
diagnostics
failure analysis
Machine learning
Maximum likelihood estimation
PHM
pneumatic actuator
Predictive models
Uncertainty
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This work aims to propose a methodology to evaluate machine learning models' capability to estimate the degradation level of an aircraft pneumatic system valve (PRSOV - pressure regulating and shutoff valve) in open-loop. A more accurate PRSOV degradation estimation may reduce operational costs and raise aircraft availability by: supporting the logistics and maintenance scheduling, enabling controller reconfiguration to keep the desired performance or extend the valve end-of-life, and reducing the corrective maintenance costs through faster fault identification. Therefore, the data used in this study were provided by a computational model that considers multiple degradation types simultaneously and adjusted to improve its accordance with the reality by regarding sensor characteristics and operating conditions. Compared with the previous works, the results show that the new input features developed by this work can improve the models' performance, mainly when the measurement uncertainties are regarded, reaching more than 85% of R 2 using only features that do not need additional sensors for the aircraft. Furthermore, it was highlighted the importance of properly choosing the excitation vector, whether fast or slow, to stimulate a particular phenomenon. Regarding the applied algorithms, XGBoost generally provides the best results. The main contributions of this work to the PRSOV degradation estimation state-of-the-art are: regard the sensor uncertainty and sampling time, study fast excitation profiles to reduce the turnaround time, and propose new features to enhance the machine learning models' performance.
ISSN:2472-9647
DOI:10.1109/SysCon53536.2022.9773917