Engine-Fault Detection with Full-Flight Data
Buch in deiner Nähe kaufen
Beschreibung
To remain competitive under increasing cost pressure, airlines increasingly rely on condition-based maintenance of aircraft engines. This requires reliable engine-condition monitoring, which to date is largely based on a few discrete measurement points (snapshots) recorded per flight. Because of this limited data availability, faults can often only be distinguished from statistical noise after several flights. However, an increasing number of aircraft now record continuous flight data (full-flight data), providing a statistically relevant amount of data per flight and thus potentially enabling fault detection after a single flight.
This thesis develops two complementary machine-learning-based fault-detection methods based on full-flight data. First, a data filter for identifying steady-state operating conditions is presented; the resulting steady-state data points are analyzed for anomalous behavior using a combination of principal component analysis and a one-class support vector machine. Second, a complementary method is developed that analyzes the entire flight trajectory, including transient flight phases, using artificial neural networks; to reduce false alarms, aleatoric and epistemic uncertainties are explicitly quantified.
Both methods were tested on twelve benchmark fault cases and achieve high detection rates at low false-positive rates. After validation with synthetic data, both approaches were also confirmed using operational flight data. The results demonstrate that continuous flight data can substantially improve the speed and reliability of aircraft engine-condition monitoring.
Autor*in
Matthias Christian Weiss
Themen in »Engine-Fault Detection with Full-Flight Data«
Flugzeugtriebwerk Zustandsüberwachung Maschinelles Lernen