Analyse et surveillance vibratoire d'Une machine en rotation

Abstract

We show that our original approach for modal analysis and vibration monitoring — initially developped for mechanical structures — may also be applied to rotating machines. Our algorithms for identifying and monitoring the vibrating characteristics of the structure or the machine do not use any artificial excitation, and manage with nonstationary natural excitation.

The modal identification method is based upon ARMA modelling, and extracts eigenvalues and eigenvectors not only related to the vibrating modes of the machine but also to the harmonic components of the excitation.

Our monitoring techniques are based upon algorithms for detection and diagnosis of changes in the AR part of an ARMA model. They provide us with several types of informations:

• global alarm (something has changed in the vibrating characteristics — maybe only in the modal vectors and not in the frequencies);

• modal diagnosis (which modes have actually changed);

• physical diagnosis (which types of physical characteristics have changed and where in the structure).

These informations are obtained without using any finite elements model updating technique. Results of the identification and the monitoring algorithms are shown for a turbo-alternator.

This work was supported by EDF under grant EDF/INRIA n°P53/1D5467