Pattern vector parameters optimization: application to the automatic vibratory diagnosis of helicopter gear defects box

The rotating machines deterioration processes, which can be accidental (brittle fracture of a component) or resulting from normal wear of the machine elements, are in general very complex. However, the deterioration appearance, results in a modification of the machine behavior and in particular an evolution of the vibratory levels. So there are several procedures of defects diagnosis. The choice of an approach is related to the knowledge which we want obtain on the system and also with the complexity of this system. The methods of monitoring by pattern recognition are more elaborated compared to the simple statistical tests and are able to detect and to diagnose the failures in an automated way, it is sufficient to extract a pattern vector (PV) on each measurement taken on the machine. The rule of decision used makes it possible to classify the observations described by the pattern vector compared to the various operating modes known, with or without defect. The work presented in this article concerns the selection of a reduced number of relevant parameters to represent the vibratory signatures of the American navy helicopter CH-46 gear box defects. We could show, that the pattern recognition (PR) methods performances are closely related to the relevance of the defects indicators.