A comparative experimental study on the use of three denoising methods for bearing defect detection

In the conditional maintenance of mechanical components by vibratory analysis, one distinguishes two types of analyses which are necessary for obtaining a reliable diagnosis. The first analysis lies in the detection of potential defects; there are currently various succeeded methods based on the treatment of the vibratory signals allowing the localization of a defect. One can quote among these methods the analysis of spectrum (with constant resolution (RC) or percentage of bands constant (PCB)), the analysis of envelope, the analysis cepstrale, the analysis time-frequencies or the analysis time-scales (wavelet). The second analysis is interested in the determination and the evaluation of severity of a defect detected to estimate the influence of this defect on the operation of a mechanism. The scalar indicators which make it possible to estimate the gravity of a defect are indicators known as total which are based on the statistical analysis of a temporal signal. However, the signals resulting from accelerometer sensors are the results of a mixture of sources of vibrations, sources being able to be allotted to one or more defects and are generally polluted by noise. This work presents the three principal methods of denoising and the study of their influence on the scalar parameters (kurtosis, factor peak, value rms) and this within the framework of the detection of defects of the chippings types of bearings.