Session: 05-04 Fault Detection, Optimization & Uncertainty

Paper Number: 124026

124026 - Fault Detection on Short-Haul or Highly Dynamic Flights Using Transient Flight Segments 

Low latency between the emergence of an engine fault and its detection is a prerequisite for the optimization of engine operation and overhaul. Existing approaches can detect an engine fault after a single flight, assuming that it contains sufficient periods of steady state operation. Many short-haul commercial flights lack the long cruise segments necessary to acquire enough steady state data. The same is true for the typical missions of helicopters and light aircraft in general, as they need a highly dynamic pilot input in order to cope with environmental influences like wind currents or turbulence. In such cases, an approach for fault detection is needed, which doesn’t rely on data recorded during steady-state flight phases.

A machine learning based approach is presented that utilizes in-flight measurements of the transient engine operation to detect persistent engine faults after a single flight. Using a single shaft turboshaft engine as an example, four standard gas path sensors are considered, so that no additional sensors need to be installed. The time series of the residuals between the measured data and the data resulting from performance synthesis is evaluated using moving windows containing at least one transient segment. Continuous wavelet transformation and a pretrained convolutional neural network are utilized on the residuals for feature extraction. The fault detection is carried out via a one-class support vector machine, trained exclusively on nominal engine operation data. Therefore, the approach requires no a-priory knowledge of the effects of engine faults on the in-flight measurements and it can be used on brand new engines or engine types. Under the assumption of persistent faults, all windows of a single flight which contain at least one transient segment are considered in order to improve the reliability of the fault detection.

This approach is demonstrated using measurement data of a small helicopter engine that replicates the dynamic flight of the corresponding helicopter on a ground test bed. Consequently, step changes of the shaft speed and the shaft power outtake as well as complex variations of both are considered. The one-class support vector machine is trained on nominal measurement data and is then used successfully to detect two types of total pressure sensor anomalies. Assuming a typical number of transient segments for an average short haul flight, it turns out that persistent faults can be detected successfully within one flight.

Presenting Author: Tihomir Varchev Institute of Aircraft Propulsion Systems (ILA), University of Stuttgart

Presenting Author Biography: Tihomir Varchev is a Ph.D. student and an academic employee at the Institute of Aircraft Propulsion Systems at the University of Stuttgart. He graduated both his Bachelor's and his Master's studies at the University of Stuttgart. During his Master's studies his main focus areas were aircraft propulsion systems and experimental and numerical methods for aerospace engineering. His current research focuses on the utilization of machine learning for engine condition monitoring using transient flight data.

Authors:

Tihomir Varchev Institute of Aircraft Propulsion Systems (ILA), University of Stuttgart
Jürgen Mathes MTU Aero Engines AG
Christian Koch Institute of Aircraft Propulsion Systems (ILA), University of Stuttgart
Stephan Staudacher Institute of Aircraft Propulsion Systems (ILA), University of Stuttgart