Analyse non stationnaire des signaux vibratoires dans la surveillance des machines et la prevention des defaillances

Abstract

The purpose of this thesis is the early diagnosis of dysfunctions coming from the mechanical defects such as the defects of bearings and gears. To achieve this goal, several methods based on EMD and wavelets were applied to the analysis of the simulated vibratory signals of bearings and operational and defective gears. The results of this research showed that the individual use of these methods is not effective. Consequently, it was proposed to improve vibratory predictive maintenance to use hybrid methods. These methods are based on the EEMD and the synchronous average or the wavelet transform for denoising as well as Kurtosis for the final decision of the diagnosis. The suggested methods were validated in experiments using real signals resulting from three test bench (for the gears defects, another for bearing defects and another for the separation of the signatures of gears and bearings).