Anomaly detection for aircraft engine fault prediction

Warning

This publication doesn't include Faculty of Sports Studies. It includes Faculty of Informatics. Official publication website can be found on muni.cz.
Authors

RUDOLECKÝ Tomáš

Year of publication 2016
Type Article in Proceedings
Conference Proceedings in Informatics and Information Technologies. Bratislava: WIKT & DaZ
MU Faculty or unit

Faculty of Informatics

Citation
Field Informatics
Keywords fault prediction; aircraft engine; support vector machine; group anomaly detection
Description Aircraft engine failures can be expensive and an obvious security threat. When we are able to predict a potential failure of an engine in advance, then we can send the aircraft for maintenance. Sensor data is collected during engine starts, takeoffs, cruise or special events. Aim of this research is to create a model of standard behavior of so called healthy engines and based on that, detect serious change which can predicts a failure. Furthermore, we want to distinguish among particular failure types. The model don’t have just to be able to successfully pass data tests but also should have some physical explanation. Sometimes the resultant model shows big dependences on attributes which should be at most auxiliary, or it shows physically improbable relations among attributes. We present the first results obtained with One-class Support Vector Machine, which show significant increase of the anomaly factor of two out of four faulted engines when they were approaching the failure. We also made experiments with group anomaly detection.
Related projects:

You are running an old browser version. We recommend updating your browser to its latest version.

More info