Detection of Local Gear Tooth Defect by a multi resolution analysis : DWT and EEMD

Some mechanical defects represented by transient signals of non-stationary character characterized by complex variations of the spectrum. These mechanical defects cannot detected in the early stages of their appearance by using Fourier transform and classical filtering methods but they can be detected by applying advanced signal processing methods based on time scale analysis to deliver the necessary information on the variation of signal frequencies with time. Time-frequency domain signals processing methods are available to detect and eliminate defects in a rotating machine. The analysis of the signal spectrum can detect the occurrence of a fault while the decomposition of this signal in time can provide us with the nature and position of this fault. The denoising of these signals by wavelet transform technique provides high accuracy of indicators and increases the precision of the search. Our approach is to perform a wavelet decomposition with multiple resolution according to the complexity of the gear failure, and then the EEMD method adopted for the fast search of the failure. The method performed on the vibratory wave signals of the test bench in the CETIM France during 12 days. These tests have shown the start of the deformation phenomenon of the teeth of the mesh in the 5th day and its progress in the 8th day. These last mentioned problems are in conformity with the results of the expertise realized on the type of gearing.