Automatic Indexation of Turbofan Data to Identify Anomalous Behaviors

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Sep 22, 2019
Jerome Lacaille Cynthia Faure Madalina Olteanu Marie Cottrell

Abstract

Flight data are now flowing in our databases. The fact is that we cannot analyze every single observation and we need a tool to automatically alert in case of unusual behavior and another tool to find similarities between parts of real aircraft flights. Our proposal is to systematically index the databases replacing each multivariate numerical signal acquired by the aircraft and the engine sensors by a sequence of labels. Each label should characterize a specific part of the signal such as a stationary phase or a transient phase. Stationary phases are summarized by snapshots made of statistics of the distribution parameters the multivariate signal and are easy to characterize. Transient phases are a more complex in a multivariate environment. This work apply a specific fast change detection algorithm to identify transient phases and an adaptive classification neural network to label each temporal behavior. However, as it seems natural to automatically separate standard flight phases like engine start, taxi, take-off, climb, etc. our goal is to identify different behaviors among those main classes. For example, we detect engines with slow thermal stabilization during the take-off and separate them from engines with fast thermal stabilization. We also separate hot engines during the climb phase to cold ones. The same sort of analysis is done on mechanical transfer as we may identify fast or slow crossing of specific vibration modes, etc. At the end of this segmentation and classification process, each multivariate signal is replaced by sequences of classes corresponding to context, rotation speed for example, and any endogenous observation like temperature or vibration. Then working on discrete data, it becomes easier to query the database for rare behaviors, usual behaviors, or to search some similarity with a specific engine observation. For example, looking at a specific temporal interval during a given flight it becomes possible to ask for flights and engines with similar behavior in the historic database.

How to Cite

Lacaille, J., Faure, C., Olteanu, M., & Cottrell, M. (2019). Automatic Indexation of Turbofan Data to Identify Anomalous Behaviors. Annual Conference of the PHM Society, 11(1). https://doi.org/10.36001/phmconf.2019.v11i1.772
Abstract 175 | PDF Downloads 209

##plugins.themes.bootstrap3.article.details##

Keywords

fickle flight, turbofan, PHM, aircraft, engine, SOM, synchronization

Section
Technical Papers