Approche numérique et expérimentale d’aide à la détection des défauts dans le cadre d'une maintenance préventive conditionnelle par l’analyse vibratoire

Abstract

During the preparation of this thesis, we have developed two different methods for the diagnosis of rotating machine faults. The first method, presents a procedure for bearing degradation monitoring at an early stage. The analysis of variance (ANOVA) coupled with Tukey’s test is used to single out the suitable parameters to follow the fault size evolution ranging from 50 µm to 150µm. The Tukey's criterion is adopted in this case to study the ability of time and frequency indicators. The rotational speed, centrifugal load and fault size are considered as independent variables while the time and frequency indicators are taken as independent variables. The experiments are performed on bearings having a fault on outer race. Based on the results of this study, the Kurtosis and Skewness show a good ability to assess the evolution of degradation in the bearings at an early stage. The work discusses the weakness of the time and frequency indicator. For example, The Kurtosis is able to distinguish the difference between the first cases 50 µm and 100 µm, but it is not able to check the difference between 100 µm and 150 µm. The Skewness is able to distinguish between 100 µm and 150 µm but it is not able to check the difference between 50 µm and 100 µm. The second method aims to analyse the vibration response of damaged rolling element bearings experimentally and to assess their degrees of degradation by examining parameters extracted from the time and frequency domains. Empirical Modal decomposition (EMD) may be used for decomposing a time signal into Intrinsic Mode Functions (IMFs) presenting the filtered information from the highest frequency to the lowest one. However, EMD can present problems due to its sensitivity to noise and mode mixing. To avoid the mode mixing effect, the Ensemble Empirical Mode Decomposition (EEMD) may be used. The principle of the EEMD is to add white noise to the original signal and then average it, IMF by IMF. This method provides improvements over the original method EMD. However, the amplitude of the added noise stays empirical, which makes this method ineffective in some situations. In order to improve the diagnosis of bearing defects, a new method called Variational Mode Decomposition (VMD) is proposed. This research shows an experimental application for the detection of bearing defects at two stages of degradation: at the early stage and at an advanced stage. In this thesis, a comparative experimental study between vibratory and acoustic emission measurements, analysed by VMD is carried out. However, the application of this method requires determining the number of significant frequencies (K) and a balancing parameter (α). For this purpose, the Operational Modal Analysis (OMA) and a correlation study are used to determine the required parameters K and α. The results show that both measurements methods work well to detect the bearing frequencies at early stage and advanced stage, but the vibration measurement revealed more sensitive for an early detection.