Thèse de Nassia Daouayry

Sujet :
Développement de méthodes de prévision fiable et robuste de l’état futur du système complexe basées sur l’analyse d’historiques d’utilisation. Application à l’hélicoptère

Résumé :

Failure anticipation is an important and growing topic in the aeronautic domain, especially for the Main Gear Box (MGB). MGB is known to be the central component of a helicopter. Classical Health and Usage Monitoring System (HUMS) mostly consist of counters monitoring quantities related to the usage of the machine, with relatively simple algorithms for on-board alarming due to limited embedded hardware processing capabilities.
Using the huge amount of in-service flight data collected, we propose in this paper to build a "virtual sensor" for MGB oil pressure. This virtual sensor is based on machine learning algorithms to predict for each flight the oil pressure from other known parameters, not including the existing oil pressure sensor. It will raise an alert as early as possible when a significant deviation occurs between values measured by the MGB oil pressure physical sensor and those provided by the virtual sensor. We have integrated domain knowledge as early as possibleinto the virtual sensor’s construction, to get domain expert as involved and confident as possible in its industrialization process.

Encadrant : Jean-Marc Petit
Co-encadrant : Vasile-Marian Scuturici