Extraction d’indicateurs robustes pour le diagnostic des défauts mécaniques : Comparaison de L’EMD et des ondelettes (WT)

Abstract

This thesis focuses on the preventive maintenance of rotating machinery using vibration analysis. Vibration analysis has become a powerful tool for monitoring a machine's operating condition in real time, utilizing specific statistical and spectral indicators calculated from the vibration signals generated by the machinery. To enhance the sensitivity of these indicators, modern methods designed for non-stationary and non-linear signals—such as wavelets and Empirical Mode Decomposition (EMD)—have been developed. This work presents these methods, outlining their properties and implementation techniques. We applied them to signals from defective gears and bearings subjected to varying levels of noise. Finally, to validate the results, we applied the aforementioned methods to real-world signals acquired using purpose-built test rigs.