Personalized learning algorithm for diagnosis: Application to aircraft

Pierre Bect (1) , Zineb Simeu-Abazi (1) , Pierre-Loïc Maisonneuve (2) , Marc Pero (2) , Bruno Demerliac (2)

1 G-SCOP_GCSP - Gestion et Conduite des Systèmes de Production
2 Eurocopter France

The diagnosis of avionic equipments on aircrafts is based on recorded messages during the flight. These messages are downloaded on a ground station which is able to analyze them and provide a diagnosis to the specialized maintenance team. For a diagnosis establishment, a correlation between all messages is performed, thanks to specific rules. These rules are currently defined by an expert who uses his knowledge of the system, the avionics architecture and the feedback from customers. These rules are only based on the messages recorded during the flight. They are able to use specific characteristics of one failure code or of a grouping of failure code but they cannot take into account the environmental information. As a result, this implies the construction of more complex rules and more adapted to the system environment. This paper introduces our personalized learning process based on two learning loops. A new automatic learning algorithm uses customer feedback and takes into consideration contextual parameters relative to system's environment is proposed. This algorithm allows to design specific rules for every aircraft functioning modes.

Mots clés: filtering rules; learning loop; localization rules; maintenance; adaptive algorithm; diagnostic; expert system

Domaines: Automatique / Robotique

Pierre Bect, Zineb Simeu-Abazi, Pierre-Loïc Maisonneuve, Marc Pero, Bruno Demerliac. Personalized learning algorithm for diagnosis: Application to aircraft. 24th International Conference on Condition Monitoring And Diagnostic Engineering Management (COMADEM2011), May 2011, Stavanger, Norway. pp.1352-1359. ⟨hal-00634898⟩